diff options
author | Bernd Schmidt <bernds@codesourcery.com> | 2011-07-15 09:36:40 +0000 |
---|---|---|
committer | Bernd Schmidt <bernds@gcc.gnu.org> | 2011-07-15 09:36:40 +0000 |
commit | bcead286bfebf2312981135b9c7d72eadc6d396c (patch) | |
tree | c4bd2120a6d9ab956422bfe82ffb27dcc91bc73b /gcc/config | |
parent | 9e36aa2367507a6700c4edbd693b530cb11c64fa (diff) | |
download | gcc-bcead286bfebf2312981135b9c7d72eadc6d396c.zip gcc-bcead286bfebf2312981135b9c7d72eadc6d396c.tar.gz gcc-bcead286bfebf2312981135b9c7d72eadc6d396c.tar.bz2 |
invoke.texi (C6X Options): New section.
gcc/
* doc/invoke.texi (C6X Options): New section.
* doc/md.texi (TI C6X family): New section.
* config.gcc: Handle tic6x, in particular tic6x-*-elf and
tic6x-*-uclinux.
* longlong.h (add_ssaaaa, __umulsidi3, umul_ppmm,
count_leading_zeros, count_trailing_zeros, UMUL_TIME, UDIV_TIME):
Provide C6X definitions.
* config/c6x/c6x.md: New file.
* config/c6x/constraints.md: New file.
* config/c6x/predicates.md: New file.
* config/c6x/c6x-sched.md.in: New file.
* config/c6x/c6x-sched.md: New file.
* config/c6x/gensched.sh: New file.
* config/c6x/c6x-mult.md.in: New file.
* config/c6x/genmult.sh: New file.
* config/c6x/c6x-mult.md: New file.
* config/c6x/sync.md: New file.
* config/c6x/c6x-protos.h: New file.
* config/c6x/sfp-machine.h: New file.
* config/c6x/c6x.c: New file.
* config/c6x/c6x.h: New file.
* config/c6x/crti.s: New file.
* config/c6x/crtn.s: New file.
* config/c6x/lib1funcs.asm: New file.
* config/c6x/c6x-modes.def: New file.
* config/c6x/genopt.sh: New file.
* config/c6x/c6x.opt: New file.
* config/c6x/c6x-tables.opt: New file.
* config/c6x/c6x-opts.h: New file.
* config/c6x/c6x-isas.def: New file.
* config/c6x/elf.h: New file.
* config/c6x/elf-common.h: New file.
* config/c6x/uclinux-elf.h: New file.
* config/c6x/t-c6x: New file.
* config/c6x/t-c6x-elf: New file.
* config/c6x/t-c6x-uclinux: New file.
* config/c6x/t-c6x-softfp: New file.
* config/c6x/gtd.c: New file.
* config/c6x/gtf.c: New file.
* config/c6x/ltd.c: New file.
* config/c6x/ltf.c: New file.
* config/c6x/ged.c: New file.
* config/c6x/gef.c: New file.
* config/c6x/led.c: New file.
* config/c6x/lef.c: New file.
* config/c6x/eqd.c: New file.
* config/c6x/eqf.c: New file.
* config/c6x/libgcc-c6xeabi.ver: New file.
contrib/
* gcc_update: Add C6X generated files.
* contrib/config-list.mk: Add c6x-elf and c6x-uclinux.
libgcc/
* config.host: Handle tic6x-*-*.
* config/c6x/c6x-abi.h: New file.
From-SVN: r176308
Diffstat (limited to 'gcc/config')
42 files changed, 14362 insertions, 0 deletions
diff --git a/gcc/config/c6x/c6x-isas.def b/gcc/config/c6x/c6x-isas.def new file mode 100644 index 0000000..8469763 --- /dev/null +++ b/gcc/config/c6x/c6x-isas.def @@ -0,0 +1,38 @@ +/* C6X ISA names. + Copyright (C) 2011 + Free Software Foundation, Inc. + + This file is part of GCC. + + GCC is free software; you can redistribute it and/or modify it + under the terms of the GNU General Public License as published by + the Free Software Foundation; either version 3, or (at your option) + any later version. + + GCC is distributed in the hope that it will be useful, but + WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + General Public License for more details. + + You should have received a copy of the GNU General Public License + along with GCC; see the file COPYING3. If not see + <http://www.gnu.org/licenses/>. */ + +/* Define ISAs for the -march option, used both in C6X.c and to + generate c6x-tables.opt. Before including this file, define a + macro: + + C6X_ISA (NAME, ENUM_VALUE, FLAGS) + + where NAME is the name for use with -march=, ENUM_VALUE is an enum + corresponding to this arch, and FLAGS is a combination of flags + that together specify the available instructions. */ + +C6X_ISA("c62x", C6X_CPU_C62X, C6X_INSNS_C62X) +C6X_ISA("c64x", C6X_CPU_C64X, C6X_INSNS_C62X | C6X_INSNS_C64X) +C6X_ISA("c64x+", C6X_CPU_C64XP, C6X_INSNS_C62X | C6X_INSNS_C64X | C6X_INSNS_C64XP) +C6X_ISA("c67x", C6X_CPU_C67X, C6X_INSNS_C62X | C6X_INSNS_C67X) +C6X_ISA("c67x+", C6X_CPU_C67XP, C6X_INSNS_C62X | C6X_INSNS_C67X | C6X_INSNS_C67XP) +C6X_ISA("c674x", C6X_CPU_C674X, + (C6X_INSNS_C62X | C6X_INSNS_C64X | C6X_INSNS_C64XP | C6X_INSNS_C67X + | C6X_INSNS_C67XP | C6X_INSNS_C674X)) diff --git a/gcc/config/c6x/c6x-modes.def b/gcc/config/c6x/c6x-modes.def new file mode 100644 index 0000000..a5af026 --- /dev/null +++ b/gcc/config/c6x/c6x-modes.def @@ -0,0 +1,24 @@ +/* Definitions of target machine for GNU compiler, for TI C6x. + Copyright (C) 2010, 2011 Free Software Foundation, Inc. + + This file is part of GCC. + + GCC is free software; you can redistribute it and/or modify it + under the terms of the GNU General Public License as published + by the Free Software Foundation; either version 3, or (at your + option) any later version. + + GCC is distributed in the hope that it will be useful, but WITHOUT + ANY WARRANTY; without even the implied warranty of MERCHANTABILITY + or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public + License for more details. + + You should have received a copy of the GNU General Public License + along with GCC; see the file COPYING3. If not see + <http://www.gnu.org/licenses/>. */ + +VECTOR_MODES (INT, 4); /* V4QI V2HI */ +VECTOR_MODES (INT, 8); /* V8QI V4HI V2SI */ + +VECTOR_MODE (FRACT, SQ, 2); /* V2SQ. */ +VECTOR_MODE (FRACT, HQ, 2); /* V2HQ. */ diff --git a/gcc/config/c6x/c6x-mult.md b/gcc/config/c6x/c6x-mult.md new file mode 100644 index 0000000..ec18ba3 --- /dev/null +++ b/gcc/config/c6x/c6x-mult.md @@ -0,0 +1,840 @@ +;; -*- buffer-read-only: t -*- +;; Generated automatically from c6x-mult.md.in by genmult.sh +;; Multiplication patterns for TI C6X. +;; This file is processed by genmult.sh to produce two variants of each +;; pattern, a normal one and a real_mult variant for modulo scheduling. +;; Copyright (C) 2010, 2011 Free Software Foundation, Inc. +;; Contributed by Bernd Schmidt <bernds@codesourcery.com> +;; Contributed by CodeSourcery. +;; +;; This file is part of GCC. +;; +;; GCC is free software; you can redistribute it and/or modify +;; it under the terms of the GNU General Public License as published by +;; the Free Software Foundation; either version 3, or (at your option) +;; any later version. +;; +;; GCC is distributed in the hope that it will be useful, +;; but WITHOUT ANY WARRANTY; without even the implied warranty of +;; MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the +;; GNU General Public License for more details. +;; +;; You should have received a copy of the GNU General Public License +;; along with GCC; see the file COPYING3. If not see +;; <http://www.gnu.org/licenses/>. + +;; ------------------------------------------------------------------------- +;; Miscellaneous insns that execute on the M units +;; ------------------------------------------------------------------------- + +(define_insn "rotlsi3" + [(set (match_operand:SI 0 "register_operand" "=a,b,a,b") + (rotate:SI (match_operand:SI 1 "register_operand" "a,b,?b,?a") + (match_operand:SI 2 "reg_or_ucst5_operand" "aIu5,bIu5,aIu5,bIu5")))] + "TARGET_INSNS_64" + "%|%.\\trotl\\t%$\\t%1, %2, %0" + [(set_attr "units" "m") + (set_attr "type" "mpy2") + (set_attr "cross" "n,n,y,y")]) + +(define_insn "bitrevsi2" + [(set (match_operand:SI 0 "register_operand" "=a,a,b,b") + (unspec:SI [(match_operand:SI 1 "register_operand" "a,?b,b,?a")] + UNSPEC_BITREV))] + "TARGET_INSNS_64" + "%|%.\\tbitr\\t%$\\t%1, %0" + [(set_attr "units" "m") + (set_attr "type" "mpy2") + (set_attr "cross" "n,y,n,y")]) + +;; Vector average. + +(define_insn "avgv2hi3" + [(set (match_operand:V2HI 0 "register_operand" "=a,b,a,b") + (unspec:V2HI [(match_operand:V2HI 1 "register_operand" "a,b,?b,?a") + (match_operand:V2HI 2 "register_operand" "a,b,a,b")] UNSPEC_AVG))] + "TARGET_INSNS_64" + "%|%.\\tavg2\\t%$\\t%1, %2, %0" + [(set_attr "units" "m") + (set_attr "type" "mpy2") + (set_attr "cross" "n,n,y,y")]) + +(define_insn "uavgv4qi3" + [(set (match_operand:V4QI 0 "register_operand" "=a,b,a,b") + (unspec:V4QI [(match_operand:V4QI 1 "register_operand" "a,b,?b,?a") + (match_operand:V4QI 2 "register_operand" "a,b,a,b")] UNSPEC_AVG))] + "TARGET_INSNS_64" + "%|%.\\tavgu4\\t%$\\t%1, %2, %0" + [(set_attr "units" "m") + (set_attr "type" "mpy2") + (set_attr "cross" "n,n,y,y")]) + +;; ------------------------------------------------------------------------- +;; Multiplication +;; ------------------------------------------------------------------------- + +(define_insn "mulhi3" + [(set (match_operand:HI 0 "register_operand" "=a,b,a,b") + (mult:HI (match_operand:HI 1 "register_operand" "a,b,?b,?a") + (match_operand:HI 2 "reg_or_scst5_operand" "aIs5,bIs5,aIs5,bIs5")))] + "" + "%|%.\\tmpy\\t%$\\t%2, %1, %0" + [(set_attr "type" "mpy2") + (set_attr "units" "m") + (set_attr "cross" "n,n,y,y")]) + +(define_insn "mulhisi3_const" + [(set (match_operand:SI 0 "register_operand" "=a,b,ab") + (mult:SI (sign_extend:SI + (match_operand:HI 1 "register_operand" "a,b,?ab")) + (match_operand:HI 2 "scst5_operand" "Is5,Is5,Is5")))] + "" + "%|%.\\tmpy\\t%$\\t%2, %1, %0" + [(set_attr "type" "mpy2") + (set_attr "units" "m") + (set_attr "cross" "n,n,y")]) + +(define_insn "*mulhisi3_insn" + [(set (match_operand:SI 0 "register_operand" "=a,b,a,b") + (mult:SI (sign_extend:SI + (match_operand:HI 1 "register_operand" "%a,b,?a,?b")) + (sign_extend:SI + (match_operand:HI 2 "reg_or_scst5_operand" "a,b,b,a"))))] + "" + "%|%.\\tmpy\\t%$\\t%1, %2, %0" + [(set_attr "type" "mpy2") + (set_attr "units" "m") + (set_attr "cross" "n,n,y,y")]) + +(define_insn "mulhisi3_lh" + [(set (match_operand:SI 0 "register_operand" "=a,b,a,b") + (mult:SI (sign_extend:SI + (match_operand:HI 1 "register_operand" "a,b,?a,?b")) + (ashiftrt:SI + (match_operand:SI 2 "register_operand" "a,b,b,a") + (const_int 16))))] + "" + "%|%.\\tmpylh\\t%$\\t%1, %2, %0" + [(set_attr "type" "mpy2") + (set_attr "units" "m") + (set_attr "cross" "n,n,y,y")]) + +(define_insn "mulhisi3_hl" + [(set (match_operand:SI 0 "register_operand" "=a,b,a,b") + (mult:SI (ashiftrt:SI + (match_operand:SI 1 "register_operand" "a,b,?a,?b") + (const_int 16)) + (sign_extend:SI + (match_operand:HI 2 "register_operand" "a,b,b,a"))))] + "" + "%|%.\\tmpyhl\\t%$\\t%1, %2, %0" + [(set_attr "type" "mpy2") + (set_attr "units" "m") + (set_attr "cross" "n,n,y,y")]) + +(define_insn "mulhisi3_hh" + [(set (match_operand:SI 0 "register_operand" "=a,b,a,b") + (mult:SI (ashiftrt:SI + (match_operand:SI 1 "register_operand" "%a,b,?a,?b") + (const_int 16)) + (ashiftrt:SI + (match_operand:SI 2 "register_operand" "a,b,b,a") + (const_int 16))))] + "" + "%|%.\\tmpyh\\t%$\\t%1, %2, %0" + [(set_attr "type" "mpy2") + (set_attr "units" "m") + (set_attr "cross" "n,n,y,y")]) + +(define_insn "umulhisi3" + [(set (match_operand:SI 0 "register_operand" "=a,b,a,b") + (mult:SI (zero_extend:SI + (match_operand:HI 1 "register_operand" "%a,b,?a,?b")) + (zero_extend:SI + (match_operand:HI 2 "register_operand" "a,b,b,a"))))] + "" + "%|%.\\tmpyu\\t%$\\t%1, %2, %0" + [(set_attr "type" "mpy2") + (set_attr "units" "m") + (set_attr "cross" "n,n,y,y")]) + +(define_insn "umulhisi3_lh" + [(set (match_operand:SI 0 "register_operand" "=a,b,a,b") + (mult:SI (zero_extend:SI + (match_operand:HI 1 "register_operand" "a,b,?a,?b")) + (lshiftrt:SI + (match_operand:SI 2 "register_operand" "a,b,b,a") + (const_int 16))))] + "" + "%|%.\\tmpylhu\\t%$\\t%1, %2, %0" + [(set_attr "type" "mpy2") + (set_attr "units" "m") + (set_attr "cross" "n,n,y,y")]) + +(define_insn "umulhisi3_hl" + [(set (match_operand:SI 0 "register_operand" "=a,b,a,b") + (mult:SI (lshiftrt:SI + (match_operand:SI 1 "register_operand" "a,b,?a,?b") + (const_int 16)) + (zero_extend:SI + (match_operand:HI 2 "register_operand" "a,b,b,a"))))] + "" + "%|%.\\tmpyhlu\\t%$\\t%1, %2, %0" + [(set_attr "type" "mpy2") + (set_attr "units" "m") + (set_attr "cross" "n,n,y,y")]) + +(define_insn "umulhisi3_hh" + [(set (match_operand:SI 0 "register_operand" "=a,b,a,b") + (mult:SI (lshiftrt:SI + (match_operand:SI 1 "register_operand" "%a,b,?a,?b") + (const_int 16)) + (lshiftrt:SI + (match_operand:SI 2 "register_operand" "a,b,b,a") + (const_int 16))))] + "" + "%|%.\\tmpyhu\\t%$\\t%1, %2, %0" + [(set_attr "type" "mpy2") + (set_attr "units" "m") + (set_attr "cross" "n,n,y,y")]) + +(define_insn "usmulhisi3_const" + [(set (match_operand:SI 0 "register_operand" "=a,b,ab") + (mult:SI (zero_extend:SI + (match_operand:HI 1 "register_operand" "a,b,?ab")) + (match_operand:SI 2 "scst5_operand" "Is5,Is5,Is5")))] + "" + "%|%.\\tmpysu\\t%$\\t%2, %1, %0" + [(set_attr "type" "mpy2") + (set_attr "units" "m") + (set_attr "cross" "n,n,y")]) + +(define_insn "*usmulhisi3_insn" + [(set (match_operand:SI 0 "register_operand" "=a,b,a,b") + (mult:SI (zero_extend:SI + (match_operand:HI 1 "register_operand" "a,b,?a,?b")) + (sign_extend:SI + (match_operand:HI 2 "reg_or_scst5_operand" "aIs5,bIs5,bIs5,aIs5"))))] + "" + "%|%.\\tmpyus\\t%$\\t%1, %2, %0" + [(set_attr "type" "mpy2") + (set_attr "units" "m") + (set_attr "cross" "n,n,y,y")]) + +(define_insn "usmulhisi3_lh" + [(set (match_operand:SI 0 "register_operand" "=a,b,a,b") + (mult:SI (zero_extend:SI + (match_operand:HI 1 "register_operand" "a,b,?a,?b")) + (ashiftrt:SI + (match_operand:SI 2 "register_operand" "a,b,b,a") + (const_int 16))))] + "" + "%|%.\\tmpyluhs\\t%$\\t%1, %2, %0" + [(set_attr "type" "mpy2") + (set_attr "units" "m") + (set_attr "cross" "n,n,y,y")]) + +(define_insn "usmulhisi3_hl" + [(set (match_operand:SI 0 "register_operand" "=a,b,a,b") + (mult:SI (lshiftrt:SI + (match_operand:SI 1 "register_operand" "a,b,?a,?b") + (const_int 16)) + (sign_extend:SI + (match_operand:HI 2 "register_operand" "a,b,b,a"))))] + "" + "%|%.\\tmpyhuls\\t%$\\t%1, %2, %0" + [(set_attr "type" "mpy2") + (set_attr "units" "m") + (set_attr "cross" "n,n,y,y")]) + +(define_insn "usmulhisi3_hh" + [(set (match_operand:SI 0 "register_operand" "=a,b,a,b") + (mult:SI (lshiftrt:SI + (match_operand:SI 1 "register_operand" "a,b,?a,?b") + (const_int 16)) + (ashiftrt:SI + (match_operand:SI 2 "register_operand" "a,b,b,a") + (const_int 16))))] + "" + "%|%.\\tmpyhus\\t%$\\t%1, %2, %0" + [(set_attr "type" "mpy2") + (set_attr "units" "m") + (set_attr "cross" "n,n,y,y")]) + +(define_insn "mulsi3_insn" + [(set (match_operand:SI 0 "register_operand" "=a,b,a,b") + (mult:SI (match_operand:SI 1 "register_operand" "%a,b,?a,?b") + (match_operand:SI 2 "register_operand" "a,b,b,a")))] + "TARGET_MPY32" + "%|%.\\tmpy32\\t%$\\t%1, %2, %0" + [(set_attr "type" "mpy4") + (set_attr "units" "m") + (set_attr "cross" "n,n,y,y")]) + +(define_insn "<u>mulsidi3" + [(set (match_operand:DI 0 "register_operand" "=a,b,a,b") + (mult:DI (any_ext:DI + (match_operand:SI 1 "register_operand" "%a,b,?a,?b")) + (any_ext:DI + (match_operand:SI 2 "register_operand" "a,b,b,a"))))] + "TARGET_MPY32" + "%|%.\\tmpy32<u>\\t%$\\t%1, %2, %0" + [(set_attr "type" "mpy4") + (set_attr "units" "m") + (set_attr "cross" "n,n,y,y")]) + +(define_insn "usmulsidi3" + [(set (match_operand:DI 0 "register_operand" "=a,b,a,b") + (mult:DI (zero_extend:DI + (match_operand:SI 1 "register_operand" "a,b,?a,?b")) + (sign_extend:DI + (match_operand:SI 2 "register_operand" "a,b,b,a"))))] + "TARGET_MPY32" + "%|%.\\tmpy32us\\t%$\\t%1, %2, %0" + [(set_attr "type" "mpy4") + (set_attr "units" "m") + (set_attr "cross" "n,n,y,y")]) + +;; Widening vector multiply and dot product + +(define_insn "mulv2hiv2si3" + [(set (match_operand:V2SI 0 "register_operand" "=a,b,a,b") + (mult:V2SI + (sign_extend:V2SI (match_operand:V2HI 1 "register_operand" "a,b,a,b")) + (sign_extend:V2SI (match_operand:V2HI 2 "register_operand" "a,b,?b,?a"))))] + "TARGET_INSNS_64" + "%|%.\\tmpy2\\t%$\\t%1, %2, %0" + [(set_attr "type" "mpy4") + (set_attr "units" "m") + (set_attr "cross" "n,n,y,y")]) + +(define_insn "umulv4qiv4hi3" + [(set (match_operand:V4HI 0 "register_operand" "=a,b,a,b") + (mult:V4HI + (zero_extend:V4HI (match_operand:V4QI 1 "register_operand" "a,b,a,b")) + (zero_extend:V4HI (match_operand:V4QI 2 "register_operand" "a,b,?b,?a"))))] + "TARGET_INSNS_64" + "%|%.\\tmpyu4\\t%$\\t%1, %2, %0" + [(set_attr "type" "mpy4") + (set_attr "units" "m") + (set_attr "cross" "n,n,y,y")]) + +(define_insn "usmulv4qiv4hi3" + [(set (match_operand:V4HI 0 "register_operand" "=a,b,a,b") + (mult:V4HI + (zero_extend:V4HI (match_operand:V4QI 1 "register_operand" "a,b,?b,?a")) + (sign_extend:V4HI (match_operand:V4QI 2 "register_operand" "a,b,a,b"))))] + "TARGET_INSNS_64" + "%|%.\\tmpyus4\\t%$\\t%1, %2, %0" + [(set_attr "type" "mpy4") + (set_attr "units" "m") + (set_attr "cross" "n,n,y,y")]) + +(define_insn "dotv2hi" + [(set (match_operand:SI 0 "register_operand" "=a,b,a,b") + (plus:SI + (mult:SI + (sign_extend:SI + (vec_select:HI + (match_operand:V2HI 1 "register_operand" "a,b,a,b") + (parallel [(const_int 0)]))) + (sign_extend:SI + (vec_select:HI + (match_operand:V2HI 2 "register_operand" "a,b,?b,?a") + (parallel [(const_int 0)])))) + (mult:SI + (sign_extend:SI + (vec_select:HI (match_dup 1) (parallel [(const_int 1)]))) + (sign_extend:SI + (vec_select:HI (match_dup 2) (parallel [(const_int 1)]))))))] + "TARGET_INSNS_64" + "%|%.\\tdotp2\\t%$\\t%1, %2, %0" + [(set_attr "type" "mpy4") + (set_attr "units" "m") + (set_attr "cross" "n,n,y,y")]) + +;; Fractional multiply + +(define_insn "mulv2hqv2sq3" + [(set (match_operand:V2SQ 0 "register_operand" "=a,b,a,b") + (ss_mult:V2SQ + (fract_convert:V2SQ + (match_operand:V2HQ 1 "register_operand" "%a,b,?a,?b")) + (fract_convert:V2SQ + (match_operand:V2HQ 2 "register_operand" "a,b,b,a"))))] + "" + "%|%.\\tsmpy2\\t%$\\t%1, %2, %0" + [(set_attr "type" "mpy4") + (set_attr "units" "m") + (set_attr "cross" "n,n,y,y")]) + +(define_insn "mulhqsq3" + [(set (match_operand:SQ 0 "register_operand" "=a,b,a,b") + (ss_mult:SQ + (fract_convert:SQ + (match_operand:HQ 1 "register_operand" "%a,b,?a,?b")) + (fract_convert:SQ + (match_operand:HQ 2 "register_operand" "a,b,b,a"))))] + "" + "%|%.\\tsmpy\\t%$\\t%1, %2, %0" + [(set_attr "type" "mpy2") + (set_attr "units" "m") + (set_attr "cross" "n,n,y,y")]) + +(define_insn "mulhqsq3_lh" + [(set (match_operand:SQ 0 "register_operand" "=a,b,a,b") + (ss_mult:SQ + (fract_convert:SQ + (match_operand:HQ 1 "register_operand" "a,b,?a,?b")) + (fract_convert:SQ + (truncate:HQ (match_operand:SQ 2 "register_operand" "a,b,b,a")))))] + "" + "%|%.\\tsmpylh\\t%$\\t%1, %2, %0" + [(set_attr "type" "mpy2") + (set_attr "units" "m") + (set_attr "cross" "n,n,y,y")]) + +(define_insn "mulhqsq3_hl" + [(set (match_operand:SQ 0 "register_operand" "=a,b,a,b") + (ss_mult:SQ + (fract_convert:SQ + (truncate:HQ (match_operand:SQ 1 "register_operand" "a,b,b,a"))) + (fract_convert:SQ + (match_operand:HQ 2 "register_operand" "a,b,b,a"))))] + "" + "%|%.\\tsmpyhl\\t%$\\t%1, %2, %0" + [(set_attr "type" "mpy2") + (set_attr "units" "m") + (set_attr "cross" "n,n,y,y")]) + +(define_insn "mulhqsq3_hh" + [(set (match_operand:SQ 0 "register_operand" "=a,b,a,b") + (ss_mult:SQ + (fract_convert:SQ + (truncate:HQ (match_operand:SQ 1 "register_operand" "a,b,b,a"))) + (fract_convert:SQ + (truncate:HQ (match_operand:SQ 2 "register_operand" "a,b,b,a")))))] + "" + "%|%.\\tsmpyh\\t%$\\t%1, %2, %0" + [(set_attr "type" "mpy2") + (set_attr "units" "m") + (set_attr "cross" "n,n,y,y")]) +;; Multiplication patterns for TI C6X. +;; This file is processed by genmult.sh to produce two variants of each +;; pattern, a normal one and a real_mult variant for modulo scheduling. +;; Copyright (C) 2010, 2011 Free Software Foundation, Inc. +;; Contributed by Bernd Schmidt <bernds@codesourcery.com> +;; Contributed by CodeSourcery. +;; +;; This file is part of GCC. +;; +;; GCC is free software; you can redistribute it and/or modify +;; it under the terms of the GNU General Public License as published by +;; the Free Software Foundation; either version 3, or (at your option) +;; any later version. +;; +;; GCC is distributed in the hope that it will be useful, +;; but WITHOUT ANY WARRANTY; without even the implied warranty of +;; MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the +;; GNU General Public License for more details. +;; +;; You should have received a copy of the GNU General Public License +;; along with GCC; see the file COPYING3. If not see +;; <http://www.gnu.org/licenses/>. + +;; ------------------------------------------------------------------------- +;; Miscellaneous insns that execute on the M units +;; ------------------------------------------------------------------------- + +(define_insn "rotlsi3_real" + [(unspec [(match_operand:SI 0 "const_int_operand" "=JA,JB,JA,JB") + (rotate:SI (match_operand:SI 1 "register_operand" "a,b,?b,?a") + (match_operand:SI 2 "reg_or_ucst5_operand" "aIu5,bIu5,aIu5,bIu5"))] UNSPEC_REAL_MULT)] + "TARGET_INSNS_64" + "%|%.\\trotl\\t%$\\t%1, %2, %k0" + [(set_attr "units" "m") + (set_attr "type" "mpy2") + (set_attr "cross" "n,n,y,y")]) + +(define_insn "bitrevsi2_real" + [(unspec [(match_operand:SI 0 "const_int_operand" "=JA,JA,JB,JB") + (unspec:SI [(match_operand:SI 1 "register_operand" "a,?b,b,?a")] + UNSPEC_BITREV)] UNSPEC_REAL_MULT)] + "TARGET_INSNS_64" + "%|%.\\tbitr\\t%$\\t%1, %k0" + [(set_attr "units" "m") + (set_attr "type" "mpy2") + (set_attr "cross" "n,y,n,y")]) + +;; Vector average. + +(define_insn "avgv2hi3_real" + [(unspec [(match_operand:SI 0 "const_int_operand" "=JA,JB,JA,JB") + (unspec:V2HI [(match_operand:V2HI 1 "register_operand" "a,b,?b,?a") + (match_operand:V2HI 2 "register_operand" "a,b,a,b")] UNSPEC_AVG)] UNSPEC_REAL_MULT)] + "TARGET_INSNS_64" + "%|%.\\tavg2\\t%$\\t%1, %2, %k0" + [(set_attr "units" "m") + (set_attr "type" "mpy2") + (set_attr "cross" "n,n,y,y")]) + +(define_insn "uavgv4qi3_real" + [(unspec [(match_operand:SI 0 "const_int_operand" "=JA,JB,JA,JB") + (unspec:V4QI [(match_operand:V4QI 1 "register_operand" "a,b,?b,?a") + (match_operand:V4QI 2 "register_operand" "a,b,a,b")] UNSPEC_AVG)] UNSPEC_REAL_MULT)] + "TARGET_INSNS_64" + "%|%.\\tavgu4\\t%$\\t%1, %2, %k0" + [(set_attr "units" "m") + (set_attr "type" "mpy2") + (set_attr "cross" "n,n,y,y")]) + +;; ------------------------------------------------------------------------- +;; Multiplication +;; ------------------------------------------------------------------------- + +(define_insn "mulhi3_real" + [(unspec [(match_operand:HI 0 "const_int_operand" "=JA,JB,JA,JB") + (mult:HI (match_operand:HI 1 "register_operand" "a,b,?b,?a") + (match_operand:HI 2 "reg_or_scst5_operand" "aIs5,bIs5,aIs5,bIs5"))] UNSPEC_REAL_MULT)] + "" + "%|%.\\tmpy\\t%$\\t%2, %1, %k0" + [(set_attr "type" "mpy2") + (set_attr "units" "m") + (set_attr "cross" "n,n,y,y")]) + +(define_insn "mulhisi3_const_real" + [(unspec [(match_operand:SI 0 "const_int_operand" "=JA,JB,JAJB") + (mult:SI (sign_extend:SI + (match_operand:HI 1 "register_operand" "a,b,?ab")) + (match_operand:HI 2 "scst5_operand" "Is5,Is5,Is5"))] UNSPEC_REAL_MULT)] + "" + "%|%.\\tmpy\\t%$\\t%2, %1, %k0" + [(set_attr "type" "mpy2") + (set_attr "units" "m") + (set_attr "cross" "n,n,y")]) + +(define_insn "*mulhisi3_insn_real" + [(unspec [(match_operand:SI 0 "const_int_operand" "=JA,JB,JA,JB") + (mult:SI (sign_extend:SI + (match_operand:HI 1 "register_operand" "%a,b,?a,?b")) + (sign_extend:SI + (match_operand:HI 2 "reg_or_scst5_operand" "a,b,b,a")))] UNSPEC_REAL_MULT)] + "" + "%|%.\\tmpy\\t%$\\t%1, %2, %k0" + [(set_attr "type" "mpy2") + (set_attr "units" "m") + (set_attr "cross" "n,n,y,y")]) + +(define_insn "mulhisi3_lh_real" + [(unspec [(match_operand:SI 0 "const_int_operand" "=JA,JB,JA,JB") + (mult:SI (sign_extend:SI + (match_operand:HI 1 "register_operand" "a,b,?a,?b")) + (ashiftrt:SI + (match_operand:SI 2 "register_operand" "a,b,b,a") + (const_int 16)))] UNSPEC_REAL_MULT)] + "" + "%|%.\\tmpylh\\t%$\\t%1, %2, %k0" + [(set_attr "type" "mpy2") + (set_attr "units" "m") + (set_attr "cross" "n,n,y,y")]) + +(define_insn "mulhisi3_hl_real" + [(unspec [(match_operand:SI 0 "const_int_operand" "=JA,JB,JA,JB") + (mult:SI (ashiftrt:SI + (match_operand:SI 1 "register_operand" "a,b,?a,?b") + (const_int 16)) + (sign_extend:SI + (match_operand:HI 2 "register_operand" "a,b,b,a")))] UNSPEC_REAL_MULT)] + "" + "%|%.\\tmpyhl\\t%$\\t%1, %2, %k0" + [(set_attr "type" "mpy2") + (set_attr "units" "m") + (set_attr "cross" "n,n,y,y")]) + +(define_insn "mulhisi3_hh_real" + [(unspec [(match_operand:SI 0 "const_int_operand" "=JA,JB,JA,JB") + (mult:SI (ashiftrt:SI + (match_operand:SI 1 "register_operand" "%a,b,?a,?b") + (const_int 16)) + (ashiftrt:SI + (match_operand:SI 2 "register_operand" "a,b,b,a") + (const_int 16)))] UNSPEC_REAL_MULT)] + "" + "%|%.\\tmpyh\\t%$\\t%1, %2, %k0" + [(set_attr "type" "mpy2") + (set_attr "units" "m") + (set_attr "cross" "n,n,y,y")]) + +(define_insn "umulhisi3_real" + [(unspec [(match_operand:SI 0 "const_int_operand" "=JA,JB,JA,JB") + (mult:SI (zero_extend:SI + (match_operand:HI 1 "register_operand" "%a,b,?a,?b")) + (zero_extend:SI + (match_operand:HI 2 "register_operand" "a,b,b,a")))] UNSPEC_REAL_MULT)] + "" + "%|%.\\tmpyu\\t%$\\t%1, %2, %k0" + [(set_attr "type" "mpy2") + (set_attr "units" "m") + (set_attr "cross" "n,n,y,y")]) + +(define_insn "umulhisi3_lh_real" + [(unspec [(match_operand:SI 0 "const_int_operand" "=JA,JB,JA,JB") + (mult:SI (zero_extend:SI + (match_operand:HI 1 "register_operand" "a,b,?a,?b")) + (lshiftrt:SI + (match_operand:SI 2 "register_operand" "a,b,b,a") + (const_int 16)))] UNSPEC_REAL_MULT)] + "" + "%|%.\\tmpylhu\\t%$\\t%1, %2, %k0" + [(set_attr "type" "mpy2") + (set_attr "units" "m") + (set_attr "cross" "n,n,y,y")]) + +(define_insn "umulhisi3_hl_real" + [(unspec [(match_operand:SI 0 "const_int_operand" "=JA,JB,JA,JB") + (mult:SI (lshiftrt:SI + (match_operand:SI 1 "register_operand" "a,b,?a,?b") + (const_int 16)) + (zero_extend:SI + (match_operand:HI 2 "register_operand" "a,b,b,a")))] UNSPEC_REAL_MULT)] + "" + "%|%.\\tmpyhlu\\t%$\\t%1, %2, %k0" + [(set_attr "type" "mpy2") + (set_attr "units" "m") + (set_attr "cross" "n,n,y,y")]) + +(define_insn "umulhisi3_hh_real" + [(unspec [(match_operand:SI 0 "const_int_operand" "=JA,JB,JA,JB") + (mult:SI (lshiftrt:SI + (match_operand:SI 1 "register_operand" "%a,b,?a,?b") + (const_int 16)) + (lshiftrt:SI + (match_operand:SI 2 "register_operand" "a,b,b,a") + (const_int 16)))] UNSPEC_REAL_MULT)] + "" + "%|%.\\tmpyhu\\t%$\\t%1, %2, %k0" + [(set_attr "type" "mpy2") + (set_attr "units" "m") + (set_attr "cross" "n,n,y,y")]) + +(define_insn "usmulhisi3_const_real" + [(unspec [(match_operand:SI 0 "const_int_operand" "=JA,JB,JAJB") + (mult:SI (zero_extend:SI + (match_operand:HI 1 "register_operand" "a,b,?ab")) + (match_operand:SI 2 "scst5_operand" "Is5,Is5,Is5"))] UNSPEC_REAL_MULT)] + "" + "%|%.\\tmpysu\\t%$\\t%2, %1, %k0" + [(set_attr "type" "mpy2") + (set_attr "units" "m") + (set_attr "cross" "n,n,y")]) + +(define_insn "*usmulhisi3_insn_real" + [(unspec [(match_operand:SI 0 "const_int_operand" "=JA,JB,JA,JB") + (mult:SI (zero_extend:SI + (match_operand:HI 1 "register_operand" "a,b,?a,?b")) + (sign_extend:SI + (match_operand:HI 2 "reg_or_scst5_operand" "aIs5,bIs5,bIs5,aIs5")))] UNSPEC_REAL_MULT)] + "" + "%|%.\\tmpyus\\t%$\\t%1, %2, %k0" + [(set_attr "type" "mpy2") + (set_attr "units" "m") + (set_attr "cross" "n,n,y,y")]) + +(define_insn "usmulhisi3_lh_real" + [(unspec [(match_operand:SI 0 "const_int_operand" "=JA,JB,JA,JB") + (mult:SI (zero_extend:SI + (match_operand:HI 1 "register_operand" "a,b,?a,?b")) + (ashiftrt:SI + (match_operand:SI 2 "register_operand" "a,b,b,a") + (const_int 16)))] UNSPEC_REAL_MULT)] + "" + "%|%.\\tmpyluhs\\t%$\\t%1, %2, %k0" + [(set_attr "type" "mpy2") + (set_attr "units" "m") + (set_attr "cross" "n,n,y,y")]) + +(define_insn "usmulhisi3_hl_real" + [(unspec [(match_operand:SI 0 "const_int_operand" "=JA,JB,JA,JB") + (mult:SI (lshiftrt:SI + (match_operand:SI 1 "register_operand" "a,b,?a,?b") + (const_int 16)) + (sign_extend:SI + (match_operand:HI 2 "register_operand" "a,b,b,a")))] UNSPEC_REAL_MULT)] + "" + "%|%.\\tmpyhuls\\t%$\\t%1, %2, %k0" + [(set_attr "type" "mpy2") + (set_attr "units" "m") + (set_attr "cross" "n,n,y,y")]) + +(define_insn "usmulhisi3_hh_real" + [(unspec [(match_operand:SI 0 "const_int_operand" "=JA,JB,JA,JB") + (mult:SI (lshiftrt:SI + (match_operand:SI 1 "register_operand" "a,b,?a,?b") + (const_int 16)) + (ashiftrt:SI + (match_operand:SI 2 "register_operand" "a,b,b,a") + (const_int 16)))] UNSPEC_REAL_MULT)] + "" + "%|%.\\tmpyhus\\t%$\\t%1, %2, %k0" + [(set_attr "type" "mpy2") + (set_attr "units" "m") + (set_attr "cross" "n,n,y,y")]) + +(define_insn "mulsi3_insn_real" + [(unspec [(match_operand:SI 0 "const_int_operand" "=JA,JB,JA,JB") + (mult:SI (match_operand:SI 1 "register_operand" "%a,b,?a,?b") + (match_operand:SI 2 "register_operand" "a,b,b,a"))] UNSPEC_REAL_MULT)] + "TARGET_MPY32" + "%|%.\\tmpy32\\t%$\\t%1, %2, %k0" + [(set_attr "type" "mpy4") + (set_attr "units" "m") + (set_attr "cross" "n,n,y,y")]) + +(define_insn "<u>mulsidi3_real" + [(unspec [(match_operand:DI 0 "const_int_operand" "=JA,JB,JA,JB") + (mult:DI (any_ext:DI + (match_operand:SI 1 "register_operand" "%a,b,?a,?b")) + (any_ext:DI + (match_operand:SI 2 "register_operand" "a,b,b,a")))] UNSPEC_REAL_MULT)] + "TARGET_MPY32" + "%|%.\\tmpy32<u>\\t%$\\t%1, %2, %K0" + [(set_attr "type" "mpy4") + (set_attr "units" "m") + (set_attr "cross" "n,n,y,y")]) + +(define_insn "usmulsidi3_real" + [(unspec [(match_operand:DI 0 "const_int_operand" "=JA,JB,JA,JB") + (mult:DI (zero_extend:DI + (match_operand:SI 1 "register_operand" "a,b,?a,?b")) + (sign_extend:DI + (match_operand:SI 2 "register_operand" "a,b,b,a")))] UNSPEC_REAL_MULT)] + "TARGET_MPY32" + "%|%.\\tmpy32us\\t%$\\t%1, %2, %K0" + [(set_attr "type" "mpy4") + (set_attr "units" "m") + (set_attr "cross" "n,n,y,y")]) + +;; Widening vector multiply and dot product + +(define_insn "mulv2hiv2si3_real" + [(unspec [(match_operand:V2SI 0 "const_int_operand" "=JA,JB,JA,JB") + (mult:V2SI + (sign_extend:V2SI (match_operand:V2HI 1 "register_operand" "a,b,a,b")) + (sign_extend:V2SI (match_operand:V2HI 2 "register_operand" "a,b,?b,?a")))] UNSPEC_REAL_MULT)] + "TARGET_INSNS_64" + "%|%.\\tmpy2\\t%$\\t%1, %2, %k0" + [(set_attr "type" "mpy4") + (set_attr "units" "m") + (set_attr "cross" "n,n,y,y")]) + +(define_insn "umulv4qiv4hi3_real" + [(unspec [(match_operand:V4HI 0 "const_int_operand" "=JA,JB,JA,JB") + (mult:V4HI + (zero_extend:V4HI (match_operand:V4QI 1 "register_operand" "a,b,a,b")) + (zero_extend:V4HI (match_operand:V4QI 2 "register_operand" "a,b,?b,?a")))] UNSPEC_REAL_MULT)] + "TARGET_INSNS_64" + "%|%.\\tmpyu4\\t%$\\t%1, %2, %k0" + [(set_attr "type" "mpy4") + (set_attr "units" "m") + (set_attr "cross" "n,n,y,y")]) + +(define_insn "usmulv4qiv4hi3_real" + [(unspec [(match_operand:V4HI 0 "const_int_operand" "=JA,JB,JA,JB") + (mult:V4HI + (zero_extend:V4HI (match_operand:V4QI 1 "register_operand" "a,b,?b,?a")) + (sign_extend:V4HI (match_operand:V4QI 2 "register_operand" "a,b,a,b")))] UNSPEC_REAL_MULT)] + "TARGET_INSNS_64" + "%|%.\\tmpyus4\\t%$\\t%1, %2, %k0" + [(set_attr "type" "mpy4") + (set_attr "units" "m") + (set_attr "cross" "n,n,y,y")]) + +(define_insn "dotv2hi_real" + [(unspec [(match_operand:SI 0 "const_int_operand" "=JA,JB,JA,JB") + (plus:SI + (mult:SI + (sign_extend:SI + (vec_select:HI + (match_operand:V2HI 1 "register_operand" "a,b,a,b") + (parallel [(const_int 0)]))) + (sign_extend:SI + (vec_select:HI + (match_operand:V2HI 2 "register_operand" "a,b,?b,?a") + (parallel [(const_int 0)])))) + (mult:SI + (sign_extend:SI + (vec_select:HI (match_dup 1) (parallel [(const_int 1)]))) + (sign_extend:SI + (vec_select:HI (match_dup 2) (parallel [(const_int 1)])))))] UNSPEC_REAL_MULT)] + "TARGET_INSNS_64" + "%|%.\\tdotp2\\t%$\\t%1, %2, %k0" + [(set_attr "type" "mpy4") + (set_attr "units" "m") + (set_attr "cross" "n,n,y,y")]) + +;; Fractional multiply + +(define_insn "mulv2hqv2sq3_real" + [(unspec [(match_operand:SI 0 "const_int_operand" "=JA,JB,JA,JB") + (ss_mult:V2SQ + (fract_convert:V2SQ + (match_operand:V2HQ 1 "register_operand" "%a,b,?a,?b")) + (fract_convert:V2SQ + (match_operand:V2HQ 2 "register_operand" "a,b,b,a")))] UNSPEC_REAL_MULT)] + "" + "%|%.\\tsmpy2\\t%$\\t%1, %2, %k0" + [(set_attr "type" "mpy4") + (set_attr "units" "m") + (set_attr "cross" "n,n,y,y")]) + +(define_insn "mulhqsq3_real" + [(unspec [(match_operand:SI 0 "const_int_operand" "=JA,JB,JA,JB") + (ss_mult:SQ + (fract_convert:SQ + (match_operand:HQ 1 "register_operand" "%a,b,?a,?b")) + (fract_convert:SQ + (match_operand:HQ 2 "register_operand" "a,b,b,a")))] UNSPEC_REAL_MULT)] + "" + "%|%.\\tsmpy\\t%$\\t%1, %2, %k0" + [(set_attr "type" "mpy2") + (set_attr "units" "m") + (set_attr "cross" "n,n,y,y")]) + +(define_insn "mulhqsq3_lh_real" + [(unspec [(match_operand:SI 0 "const_int_operand" "=JA,JB,JA,JB") + (ss_mult:SQ + (fract_convert:SQ + (match_operand:HQ 1 "register_operand" "a,b,?a,?b")) + (fract_convert:SQ + (truncate:HQ (match_operand:SQ 2 "register_operand" "a,b,b,a"))))] UNSPEC_REAL_MULT)] + "" + "%|%.\\tsmpylh\\t%$\\t%1, %2, %k0" + [(set_attr "type" "mpy2") + (set_attr "units" "m") + (set_attr "cross" "n,n,y,y")]) + +(define_insn "mulhqsq3_hl_real" + [(unspec [(match_operand:SI 0 "const_int_operand" "=JA,JB,JA,JB") + (ss_mult:SQ + (fract_convert:SQ + (truncate:HQ (match_operand:SQ 1 "register_operand" "a,b,b,a"))) + (fract_convert:SQ + (match_operand:HQ 2 "register_operand" "a,b,b,a")))] UNSPEC_REAL_MULT)] + "" + "%|%.\\tsmpyhl\\t%$\\t%1, %2, %k0" + [(set_attr "type" "mpy2") + (set_attr "units" "m") + (set_attr "cross" "n,n,y,y")]) + +(define_insn "mulhqsq3_hh_real" + [(unspec [(match_operand:SI 0 "const_int_operand" "=JA,JB,JA,JB") + (ss_mult:SQ + (fract_convert:SQ + (truncate:HQ (match_operand:SQ 1 "register_operand" "a,b,b,a"))) + (fract_convert:SQ + (truncate:HQ (match_operand:SQ 2 "register_operand" "a,b,b,a"))))] UNSPEC_REAL_MULT)] + "" + "%|%.\\tsmpyh\\t%$\\t%1, %2, %k0" + [(set_attr "type" "mpy2") + (set_attr "units" "m") + (set_attr "cross" "n,n,y,y")]) diff --git a/gcc/config/c6x/c6x-mult.md.in b/gcc/config/c6x/c6x-mult.md.in new file mode 100644 index 0000000..96de44f --- /dev/null +++ b/gcc/config/c6x/c6x-mult.md.in @@ -0,0 +1,419 @@ +;; Multiplication patterns for TI C6X. +;; This file is processed by genmult.sh to produce two variants of each +;; pattern, a normal one and a real_mult variant for modulo scheduling. +;; Copyright (C) 2010, 2011 Free Software Foundation, Inc. +;; Contributed by Bernd Schmidt <bernds@codesourcery.com> +;; Contributed by CodeSourcery. +;; +;; This file is part of GCC. +;; +;; GCC is free software; you can redistribute it and/or modify +;; it under the terms of the GNU General Public License as published by +;; the Free Software Foundation; either version 3, or (at your option) +;; any later version. +;; +;; GCC is distributed in the hope that it will be useful, +;; but WITHOUT ANY WARRANTY; without even the implied warranty of +;; MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the +;; GNU General Public License for more details. +;; +;; You should have received a copy of the GNU General Public License +;; along with GCC; see the file COPYING3. If not see +;; <http://www.gnu.org/licenses/>. + +;; ------------------------------------------------------------------------- +;; Miscellaneous insns that execute on the M units +;; ------------------------------------------------------------------------- + +(define_insn "rotlsi3_VARIANT_" + [(_SET_ _OBRK_(match_operand:SI 0 "_DESTOPERAND_" "=_A_,_B_,_A_,_B_") + (rotate:SI (match_operand:SI 1 "register_operand" "a,b,?b,?a") + (match_operand:SI 2 "reg_or_ucst5_operand" "aIu5,bIu5,aIu5,bIu5"))_CBRK_)] + "TARGET_INSNS_64" + "%|%.\\trotl\\t%$\\t%1, %2, %_MODk_0" + [(set_attr "units" "m") + (set_attr "type" "mpy2") + (set_attr "cross" "n,n,y,y")]) + +(define_insn "bitrevsi2_VARIANT_" + [(_SET_ _OBRK_(match_operand:SI 0 "_DESTOPERAND_" "=_A_,_A_,_B_,_B_") + (unspec:SI [(match_operand:SI 1 "register_operand" "a,?b,b,?a")] + UNSPEC_BITREV)_CBRK_)] + "TARGET_INSNS_64" + "%|%.\\tbitr\\t%$\\t%1, %_MODk_0" + [(set_attr "units" "m") + (set_attr "type" "mpy2") + (set_attr "cross" "n,y,n,y")]) + +;; Vector average. + +(define_insn "avgv2hi3_VARIANT_" + [(_SET_ _OBRK_(match_operand:_MV2HI 0 "_DESTOPERAND_" "=_A_,_B_,_A_,_B_") + (unspec:V2HI [(match_operand:V2HI 1 "register_operand" "a,b,?b,?a") + (match_operand:V2HI 2 "register_operand" "a,b,a,b")] UNSPEC_AVG)_CBRK_)] + "TARGET_INSNS_64" + "%|%.\\tavg2\\t%$\\t%1, %2, %_MODk_0" + [(set_attr "units" "m") + (set_attr "type" "mpy2") + (set_attr "cross" "n,n,y,y")]) + +(define_insn "uavgv4qi3_VARIANT_" + [(_SET_ _OBRK_(match_operand:_MV4QI 0 "_DESTOPERAND_" "=_A_,_B_,_A_,_B_") + (unspec:V4QI [(match_operand:V4QI 1 "register_operand" "a,b,?b,?a") + (match_operand:V4QI 2 "register_operand" "a,b,a,b")] UNSPEC_AVG)_CBRK_)] + "TARGET_INSNS_64" + "%|%.\\tavgu4\\t%$\\t%1, %2, %_MODk_0" + [(set_attr "units" "m") + (set_attr "type" "mpy2") + (set_attr "cross" "n,n,y,y")]) + +;; ------------------------------------------------------------------------- +;; Multiplication +;; ------------------------------------------------------------------------- + +(define_insn "mulhi3_VARIANT_" + [(_SET_ _OBRK_(match_operand:HI 0 "_DESTOPERAND_" "=_A_,_B_,_A_,_B_") + (mult:HI (match_operand:HI 1 "register_operand" "a,b,?b,?a") + (match_operand:HI 2 "reg_or_scst5_operand" "aIs5,bIs5,aIs5,bIs5"))_CBRK_)] + "" + "%|%.\\tmpy\\t%$\\t%2, %1, %_MODk_0" + [(set_attr "type" "mpy2") + (set_attr "units" "m") + (set_attr "cross" "n,n,y,y")]) + +(define_insn "mulhisi3_const_VARIANT_" + [(_SET_ _OBRK_(match_operand:SI 0 "_DESTOPERAND_" "=_A_,_B_,_A__B_") + (mult:SI (sign_extend:SI + (match_operand:HI 1 "register_operand" "a,b,?ab")) + (match_operand:HI 2 "scst5_operand" "Is5,Is5,Is5"))_CBRK_)] + "" + "%|%.\\tmpy\\t%$\\t%2, %1, %_MODk_0" + [(set_attr "type" "mpy2") + (set_attr "units" "m") + (set_attr "cross" "n,n,y")]) + +(define_insn "*mulhisi3_insn_VARIANT_" + [(_SET_ _OBRK_(match_operand:SI 0 "_DESTOPERAND_" "=_A_,_B_,_A_,_B_") + (mult:SI (sign_extend:SI + (match_operand:HI 1 "register_operand" "%a,b,?a,?b")) + (sign_extend:SI + (match_operand:HI 2 "reg_or_scst5_operand" "a,b,b,a")))_CBRK_)] + "" + "%|%.\\tmpy\\t%$\\t%1, %2, %_MODk_0" + [(set_attr "type" "mpy2") + (set_attr "units" "m") + (set_attr "cross" "n,n,y,y")]) + +(define_insn "mulhisi3_lh_VARIANT_" + [(_SET_ _OBRK_(match_operand:SI 0 "_DESTOPERAND_" "=_A_,_B_,_A_,_B_") + (mult:SI (sign_extend:SI + (match_operand:HI 1 "register_operand" "a,b,?a,?b")) + (ashiftrt:SI + (match_operand:SI 2 "register_operand" "a,b,b,a") + (const_int 16)))_CBRK_)] + "" + "%|%.\\tmpylh\\t%$\\t%1, %2, %_MODk_0" + [(set_attr "type" "mpy2") + (set_attr "units" "m") + (set_attr "cross" "n,n,y,y")]) + +(define_insn "mulhisi3_hl_VARIANT_" + [(_SET_ _OBRK_(match_operand:SI 0 "_DESTOPERAND_" "=_A_,_B_,_A_,_B_") + (mult:SI (ashiftrt:SI + (match_operand:SI 1 "register_operand" "a,b,?a,?b") + (const_int 16)) + (sign_extend:SI + (match_operand:HI 2 "register_operand" "a,b,b,a")))_CBRK_)] + "" + "%|%.\\tmpyhl\\t%$\\t%1, %2, %_MODk_0" + [(set_attr "type" "mpy2") + (set_attr "units" "m") + (set_attr "cross" "n,n,y,y")]) + +(define_insn "mulhisi3_hh_VARIANT_" + [(_SET_ _OBRK_(match_operand:SI 0 "_DESTOPERAND_" "=_A_,_B_,_A_,_B_") + (mult:SI (ashiftrt:SI + (match_operand:SI 1 "register_operand" "%a,b,?a,?b") + (const_int 16)) + (ashiftrt:SI + (match_operand:SI 2 "register_operand" "a,b,b,a") + (const_int 16)))_CBRK_)] + "" + "%|%.\\tmpyh\\t%$\\t%1, %2, %_MODk_0" + [(set_attr "type" "mpy2") + (set_attr "units" "m") + (set_attr "cross" "n,n,y,y")]) + +(define_insn "umulhisi3_VARIANT_" + [(_SET_ _OBRK_(match_operand:SI 0 "_DESTOPERAND_" "=_A_,_B_,_A_,_B_") + (mult:SI (zero_extend:SI + (match_operand:HI 1 "register_operand" "%a,b,?a,?b")) + (zero_extend:SI + (match_operand:HI 2 "register_operand" "a,b,b,a")))_CBRK_)] + "" + "%|%.\\tmpyu\\t%$\\t%1, %2, %_MODk_0" + [(set_attr "type" "mpy2") + (set_attr "units" "m") + (set_attr "cross" "n,n,y,y")]) + +(define_insn "umulhisi3_lh_VARIANT_" + [(_SET_ _OBRK_(match_operand:SI 0 "_DESTOPERAND_" "=_A_,_B_,_A_,_B_") + (mult:SI (zero_extend:SI + (match_operand:HI 1 "register_operand" "a,b,?a,?b")) + (lshiftrt:SI + (match_operand:SI 2 "register_operand" "a,b,b,a") + (const_int 16)))_CBRK_)] + "" + "%|%.\\tmpylhu\\t%$\\t%1, %2, %_MODk_0" + [(set_attr "type" "mpy2") + (set_attr "units" "m") + (set_attr "cross" "n,n,y,y")]) + +(define_insn "umulhisi3_hl_VARIANT_" + [(_SET_ _OBRK_(match_operand:SI 0 "_DESTOPERAND_" "=_A_,_B_,_A_,_B_") + (mult:SI (lshiftrt:SI + (match_operand:SI 1 "register_operand" "a,b,?a,?b") + (const_int 16)) + (zero_extend:SI + (match_operand:HI 2 "register_operand" "a,b,b,a")))_CBRK_)] + "" + "%|%.\\tmpyhlu\\t%$\\t%1, %2, %_MODk_0" + [(set_attr "type" "mpy2") + (set_attr "units" "m") + (set_attr "cross" "n,n,y,y")]) + +(define_insn "umulhisi3_hh_VARIANT_" + [(_SET_ _OBRK_(match_operand:SI 0 "_DESTOPERAND_" "=_A_,_B_,_A_,_B_") + (mult:SI (lshiftrt:SI + (match_operand:SI 1 "register_operand" "%a,b,?a,?b") + (const_int 16)) + (lshiftrt:SI + (match_operand:SI 2 "register_operand" "a,b,b,a") + (const_int 16)))_CBRK_)] + "" + "%|%.\\tmpyhu\\t%$\\t%1, %2, %_MODk_0" + [(set_attr "type" "mpy2") + (set_attr "units" "m") + (set_attr "cross" "n,n,y,y")]) + +(define_insn "usmulhisi3_const_VARIANT_" + [(_SET_ _OBRK_(match_operand:SI 0 "_DESTOPERAND_" "=_A_,_B_,_A__B_") + (mult:SI (zero_extend:SI + (match_operand:HI 1 "register_operand" "a,b,?ab")) + (match_operand:SI 2 "scst5_operand" "Is5,Is5,Is5"))_CBRK_)] + "" + "%|%.\\tmpysu\\t%$\\t%2, %1, %_MODk_0" + [(set_attr "type" "mpy2") + (set_attr "units" "m") + (set_attr "cross" "n,n,y")]) + +(define_insn "*usmulhisi3_insn_VARIANT_" + [(_SET_ _OBRK_(match_operand:SI 0 "_DESTOPERAND_" "=_A_,_B_,_A_,_B_") + (mult:SI (zero_extend:SI + (match_operand:HI 1 "register_operand" "a,b,?a,?b")) + (sign_extend:SI + (match_operand:HI 2 "reg_or_scst5_operand" "aIs5,bIs5,bIs5,aIs5")))_CBRK_)] + "" + "%|%.\\tmpyus\\t%$\\t%1, %2, %_MODk_0" + [(set_attr "type" "mpy2") + (set_attr "units" "m") + (set_attr "cross" "n,n,y,y")]) + +(define_insn "usmulhisi3_lh_VARIANT_" + [(_SET_ _OBRK_(match_operand:SI 0 "_DESTOPERAND_" "=_A_,_B_,_A_,_B_") + (mult:SI (zero_extend:SI + (match_operand:HI 1 "register_operand" "a,b,?a,?b")) + (ashiftrt:SI + (match_operand:SI 2 "register_operand" "a,b,b,a") + (const_int 16)))_CBRK_)] + "" + "%|%.\\tmpyluhs\\t%$\\t%1, %2, %_MODk_0" + [(set_attr "type" "mpy2") + (set_attr "units" "m") + (set_attr "cross" "n,n,y,y")]) + +(define_insn "usmulhisi3_hl_VARIANT_" + [(_SET_ _OBRK_(match_operand:SI 0 "_DESTOPERAND_" "=_A_,_B_,_A_,_B_") + (mult:SI (lshiftrt:SI + (match_operand:SI 1 "register_operand" "a,b,?a,?b") + (const_int 16)) + (sign_extend:SI + (match_operand:HI 2 "register_operand" "a,b,b,a")))_CBRK_)] + "" + "%|%.\\tmpyhuls\\t%$\\t%1, %2, %_MODk_0" + [(set_attr "type" "mpy2") + (set_attr "units" "m") + (set_attr "cross" "n,n,y,y")]) + +(define_insn "usmulhisi3_hh_VARIANT_" + [(_SET_ _OBRK_(match_operand:SI 0 "_DESTOPERAND_" "=_A_,_B_,_A_,_B_") + (mult:SI (lshiftrt:SI + (match_operand:SI 1 "register_operand" "a,b,?a,?b") + (const_int 16)) + (ashiftrt:SI + (match_operand:SI 2 "register_operand" "a,b,b,a") + (const_int 16)))_CBRK_)] + "" + "%|%.\\tmpyhus\\t%$\\t%1, %2, %_MODk_0" + [(set_attr "type" "mpy2") + (set_attr "units" "m") + (set_attr "cross" "n,n,y,y")]) + +(define_insn "mulsi3_insn_VARIANT_" + [(_SET_ _OBRK_(match_operand:SI 0 "_DESTOPERAND_" "=_A_,_B_,_A_,_B_") + (mult:SI (match_operand:SI 1 "register_operand" "%a,b,?a,?b") + (match_operand:SI 2 "register_operand" "a,b,b,a"))_CBRK_)] + "TARGET_MPY32" + "%|%.\\tmpy32\\t%$\\t%1, %2, %_MODk_0" + [(set_attr "type" "mpy4") + (set_attr "units" "m") + (set_attr "cross" "n,n,y,y")]) + +(define_insn "<u>mulsidi3_VARIANT_" + [(_SET_ _OBRK_(match_operand:DI 0 "_DESTOPERAND_" "=_A_,_B_,_A_,_B_") + (mult:DI (any_ext:DI + (match_operand:SI 1 "register_operand" "%a,b,?a,?b")) + (any_ext:DI + (match_operand:SI 2 "register_operand" "a,b,b,a")))_CBRK_)] + "TARGET_MPY32" + "%|%.\\tmpy32<u>\\t%$\\t%1, %2, %_MODK_0" + [(set_attr "type" "mpy4") + (set_attr "units" "m") + (set_attr "cross" "n,n,y,y")]) + +(define_insn "usmulsidi3_VARIANT_" + [(_SET_ _OBRK_(match_operand:DI 0 "_DESTOPERAND_" "=_A_,_B_,_A_,_B_") + (mult:DI (zero_extend:DI + (match_operand:SI 1 "register_operand" "a,b,?a,?b")) + (sign_extend:DI + (match_operand:SI 2 "register_operand" "a,b,b,a")))_CBRK_)] + "TARGET_MPY32" + "%|%.\\tmpy32us\\t%$\\t%1, %2, %_MODK_0" + [(set_attr "type" "mpy4") + (set_attr "units" "m") + (set_attr "cross" "n,n,y,y")]) + +;; Widening vector multiply and dot product + +(define_insn "mulv2hiv2si3_VARIANT_" + [(_SET_ _OBRK_(match_operand:V2SI 0 "_DESTOPERAND_" "=_A_,_B_,_A_,_B_") + (mult:V2SI + (sign_extend:V2SI (match_operand:V2HI 1 "register_operand" "a,b,a,b")) + (sign_extend:V2SI (match_operand:V2HI 2 "register_operand" "a,b,?b,?a")))_CBRK_)] + "TARGET_INSNS_64" + "%|%.\\tmpy2\\t%$\\t%1, %2, %_MODk_0" + [(set_attr "type" "mpy4") + (set_attr "units" "m") + (set_attr "cross" "n,n,y,y")]) + +(define_insn "umulv4qiv4hi3_VARIANT_" + [(_SET_ _OBRK_(match_operand:V4HI 0 "_DESTOPERAND_" "=_A_,_B_,_A_,_B_") + (mult:V4HI + (zero_extend:V4HI (match_operand:V4QI 1 "register_operand" "a,b,a,b")) + (zero_extend:V4HI (match_operand:V4QI 2 "register_operand" "a,b,?b,?a")))_CBRK_)] + "TARGET_INSNS_64" + "%|%.\\tmpyu4\\t%$\\t%1, %2, %_MODk_0" + [(set_attr "type" "mpy4") + (set_attr "units" "m") + (set_attr "cross" "n,n,y,y")]) + +(define_insn "usmulv4qiv4hi3_VARIANT_" + [(_SET_ _OBRK_(match_operand:V4HI 0 "_DESTOPERAND_" "=_A_,_B_,_A_,_B_") + (mult:V4HI + (zero_extend:V4HI (match_operand:V4QI 1 "register_operand" "a,b,?b,?a")) + (sign_extend:V4HI (match_operand:V4QI 2 "register_operand" "a,b,a,b")))_CBRK_)] + "TARGET_INSNS_64" + "%|%.\\tmpyus4\\t%$\\t%1, %2, %_MODk_0" + [(set_attr "type" "mpy4") + (set_attr "units" "m") + (set_attr "cross" "n,n,y,y")]) + +(define_insn "dotv2hi_VARIANT_" + [(_SET_ _OBRK_(match_operand:SI 0 "_DESTOPERAND_" "=_A_,_B_,_A_,_B_") + (plus:SI + (mult:SI + (sign_extend:SI + (vec_select:HI + (match_operand:V2HI 1 "register_operand" "a,b,a,b") + (parallel [(const_int 0)]))) + (sign_extend:SI + (vec_select:HI + (match_operand:V2HI 2 "register_operand" "a,b,?b,?a") + (parallel [(const_int 0)])))) + (mult:SI + (sign_extend:SI + (vec_select:HI (match_dup 1) (parallel [(const_int 1)]))) + (sign_extend:SI + (vec_select:HI (match_dup 2) (parallel [(const_int 1)])))))_CBRK_)] + "TARGET_INSNS_64" + "%|%.\\tdotp2\\t%$\\t%1, %2, %_MODk_0" + [(set_attr "type" "mpy4") + (set_attr "units" "m") + (set_attr "cross" "n,n,y,y")]) + +;; Fractional multiply + +(define_insn "mulv2hqv2sq3_VARIANT_" + [(_SET_ _OBRK_(match_operand:_MV2SQ 0 "_DESTOPERAND_" "=_A_,_B_,_A_,_B_") + (ss_mult:V2SQ + (fract_convert:V2SQ + (match_operand:V2HQ 1 "register_operand" "%a,b,?a,?b")) + (fract_convert:V2SQ + (match_operand:V2HQ 2 "register_operand" "a,b,b,a")))_CBRK_)] + "" + "%|%.\\tsmpy2\\t%$\\t%1, %2, %_MODk_0" + [(set_attr "type" "mpy4") + (set_attr "units" "m") + (set_attr "cross" "n,n,y,y")]) + +(define_insn "mulhqsq3_VARIANT_" + [(_SET_ _OBRK_(match_operand:_MSQ 0 "_DESTOPERAND_" "=_A_,_B_,_A_,_B_") + (ss_mult:SQ + (fract_convert:SQ + (match_operand:HQ 1 "register_operand" "%a,b,?a,?b")) + (fract_convert:SQ + (match_operand:HQ 2 "register_operand" "a,b,b,a")))_CBRK_)] + "" + "%|%.\\tsmpy\\t%$\\t%1, %2, %_MODk_0" + [(set_attr "type" "mpy2") + (set_attr "units" "m") + (set_attr "cross" "n,n,y,y")]) + +(define_insn "mulhqsq3_lh_VARIANT_" + [(_SET_ _OBRK_(match_operand:_MSQ 0 "_DESTOPERAND_" "=_A_,_B_,_A_,_B_") + (ss_mult:SQ + (fract_convert:SQ + (match_operand:HQ 1 "register_operand" "a,b,?a,?b")) + (fract_convert:SQ + (truncate:HQ (match_operand:SQ 2 "register_operand" "a,b,b,a"))))_CBRK_)] + "" + "%|%.\\tsmpylh\\t%$\\t%1, %2, %_MODk_0" + [(set_attr "type" "mpy2") + (set_attr "units" "m") + (set_attr "cross" "n,n,y,y")]) + +(define_insn "mulhqsq3_hl_VARIANT_" + [(_SET_ _OBRK_(match_operand:_MSQ 0 "_DESTOPERAND_" "=_A_,_B_,_A_,_B_") + (ss_mult:SQ + (fract_convert:SQ + (truncate:HQ (match_operand:SQ 1 "register_operand" "a,b,b,a"))) + (fract_convert:SQ + (match_operand:HQ 2 "register_operand" "a,b,b,a")))_CBRK_)] + "" + "%|%.\\tsmpyhl\\t%$\\t%1, %2, %_MODk_0" + [(set_attr "type" "mpy2") + (set_attr "units" "m") + (set_attr "cross" "n,n,y,y")]) + +(define_insn "mulhqsq3_hh_VARIANT_" + [(_SET_ _OBRK_(match_operand:_MSQ 0 "_DESTOPERAND_" "=_A_,_B_,_A_,_B_") + (ss_mult:SQ + (fract_convert:SQ + (truncate:HQ (match_operand:SQ 1 "register_operand" "a,b,b,a"))) + (fract_convert:SQ + (truncate:HQ (match_operand:SQ 2 "register_operand" "a,b,b,a"))))_CBRK_)] + "" + "%|%.\\tsmpyh\\t%$\\t%1, %2, %_MODk_0" + [(set_attr "type" "mpy2") + (set_attr "units" "m") + (set_attr "cross" "n,n,y,y")]) diff --git a/gcc/config/c6x/c6x-opts.h b/gcc/config/c6x/c6x-opts.h new file mode 100644 index 0000000..7c00813 --- /dev/null +++ b/gcc/config/c6x/c6x-opts.h @@ -0,0 +1,36 @@ +/* Definitions for option handling for TI C6X. + Copyright (C) 2011 + Free Software Foundation, Inc. + +This file is part of GCC. + +GCC is free software; you can redistribute it and/or modify +it under the terms of the GNU General Public License as published by +the Free Software Foundation; either version 3, or (at your option) +any later version. + +GCC is distributed in the hope that it will be useful, +but WITHOUT ANY WARRANTY; without even the implied warranty of +MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the +GNU General Public License for more details. + +You should have received a copy of the GNU General Public License +along with GCC; see the file COPYING3. If not see +<http://www.gnu.org/licenses/>. */ + +#ifndef C6X_OPTS_H +#define C6X_OPTS_H + +/* An enumeration of all supported target devices. */ +typedef enum c6x_cpu_type +{ +#define C6X_ISA(NAME,ENUM_VALUE,FLAGS) \ + ENUM_VALUE, +#include "c6x-isas.def" +#undef C6X_ISA + unk_isa +} c6x_cpu_t; + +enum c6x_sdata { C6X_SDATA_NONE, C6X_SDATA_DEFAULT, C6X_SDATA_ALL }; + +#endif diff --git a/gcc/config/c6x/c6x-protos.h b/gcc/config/c6x/c6x-protos.h new file mode 100644 index 0000000..b835ffc --- /dev/null +++ b/gcc/config/c6x/c6x-protos.h @@ -0,0 +1,66 @@ +/* Prototypes for exported functions defined in c6x.c. + Copyright (C) 2010, 2011 + Free Software Foundation, Inc. + Contributed by CodeSourcery. + + This file is part of GCC. + + GCC is free software; you can redistribute it and/or modify it + under the terms of the GNU General Public License as published + by the Free Software Foundation; either version 3, or (at your + option) any later version. + + GCC is distributed in the hope that it will be useful, but WITHOUT + ANY WARRANTY; without even the implied warranty of MERCHANTABILITY + or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public + License for more details. + + You should have received a copy of the GNU General Public License + along with GCC; see the file COPYING3. If not see + <http://www.gnu.org/licenses/>. */ + +#ifndef GCC_C6X_PROTOS_H +#define GCC_C6X_PROTOS_H + +/* Functions defined in c6x.c. */ + +#ifdef RTX_CODE +extern void c6x_init_cumulative_args (CUMULATIVE_ARGS *, const_tree, rtx, int); +extern bool c6x_block_reg_pad_upward (enum machine_mode, const_tree, bool); + +extern bool c6x_legitimate_address_p_1 (enum machine_mode, rtx, bool, bool); +extern bool c6x_mem_operand (rtx, enum reg_class, bool); +extern bool expand_move (rtx *, enum machine_mode); + +extern bool c6x_long_call_p (rtx); +extern void c6x_expand_call (rtx, rtx, bool); +extern rtx c6x_expand_compare (rtx, enum machine_mode); +extern bool c6x_force_op_for_comparison_p (enum rtx_code, rtx); +extern bool c6x_expand_movmem (rtx, rtx, rtx, rtx, rtx, rtx); + +extern rtx c6x_subword (rtx, bool); +extern void split_di (rtx *, int, rtx *, rtx *); +extern bool c6x_valid_mask_p (HOST_WIDE_INT); + +extern char c6x_get_unit_specifier (rtx); + +extern void c6x_final_prescan_insn(rtx insn, rtx *opvec, int noperands); + +extern int c6x_nsaved_regs (void); +extern HOST_WIDE_INT c6x_initial_elimination_offset (int, int); +extern void c6x_expand_prologue (void); +extern void c6x_expand_epilogue (bool); + +extern rtx c6x_return_addr_rtx (int); + +extern void c6x_set_return_address (rtx, rtx); +#endif + +extern void c6x_override_options (void); +extern void c6x_optimization_options (int, int); + +extern void c6x_output_file_unwind (FILE *); + +extern void c6x_function_end (FILE *, const char *); + +#endif /* GCC_C6X_PROTOS_H */ diff --git a/gcc/config/c6x/c6x-sched.md b/gcc/config/c6x/c6x-sched.md new file mode 100644 index 0000000..6cb4b66 --- /dev/null +++ b/gcc/config/c6x/c6x-sched.md @@ -0,0 +1,934 @@ +;; -*- buffer-read-only: t -*- +;; Generated automatically from c6x-sched.md.in by gensched.sh + +;; Definitions for side 1, cross n + +;; Scheduling description for TI C6X. +;; Copyright (C) 2010, 2011 Free Software Foundation, Inc. +;; Contributed by Bernd Schmidt <bernds@codesourcery.com> +;; Contributed by CodeSourcery. +;; +;; This file is part of GCC. +;; +;; GCC is free software; you can redistribute it and/or modify +;; it under the terms of the GNU General Public License as published by +;; the Free Software Foundation; either version 3, or (at your option) +;; any later version. +;; +;; GCC is distributed in the hope that it will be useful, +;; but WITHOUT ANY WARRANTY; without even the implied warranty of +;; MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the +;; GNU General Public License for more details. +;; +;; You should have received a copy of the GNU General Public License +;; along with GCC; see the file COPYING3. If not see +;; <http://www.gnu.org/licenses/>. + +;; Input file for gensched.sh We process this file multiple times, +;; replacing 1 with either 1 or 2 for each of the sides of the +;; machine, and a correspondingly with "a" or "b". n and +;; are replaced with yes/no and the appropriate reservation. + +(define_insn_reservation "load_d1n" 5 + (and (eq_attr "type" "load") + (and (eq_attr "cross" "n") + (and (eq_attr "units" "d_addr") + (eq_attr "addr_regfile" "a")))) + "d1+t1") + +(define_insn_reservation "store_d1n" 1 + (and (eq_attr "type" "store") + (and (eq_attr "cross" "n") + (and (eq_attr "units" "d_addr") + (eq_attr "addr_regfile" "a")))) + "d1+t1") + +(define_insn_reservation "loadn_d1n" 5 + (and (eq_attr "type" "loadn") + (and (eq_attr "cross" "n") + (and (eq_attr "units" "d_addr") + (eq_attr "addr_regfile" "a")))) + "d1+t1+t2") + +(define_insn_reservation "storen_d1n" 1 + (and (eq_attr "type" "storen") + (and (eq_attr "cross" "n") + (and (eq_attr "units" "d_addr") + (eq_attr "addr_regfile" "a")))) + "d1+t1+t2") + +(define_insn_reservation "single_d1n" 1 + (and (eq_attr "type" "single") + (and (eq_attr "cross" "n") + (and (eq_attr "units" "d") + (eq_attr "dest_regfile" "a")))) + "d1") + +(define_insn_reservation "single_l1n" 1 + (and (eq_attr "type" "single") + (and (eq_attr "cross" "n") + (and (eq_attr "units" "l") + (eq_attr "dest_regfile" "a")))) + "l1+l1w") + +(define_insn_reservation "fp4_l1n" 4 + (and (eq_attr "type" "fp4") + (and (eq_attr "cross" "n") + (and (eq_attr "units" "l") + (eq_attr "dest_regfile" "a")))) + "l1,nothing*2,l1w") + +(define_insn_reservation "intdp_l1n" 5 + (and (eq_attr "type" "intdp") + (and (eq_attr "cross" "n") + (and (eq_attr "units" "l") + (eq_attr "dest_regfile" "a")))) + "l1,nothing*2,l1w*2") + +(define_insn_reservation "adddp_l1n" 7 + (and (eq_attr "type" "adddp") + (and (eq_attr "cross" "n") + (and (eq_attr "units" "l") + (eq_attr "dest_regfile" "a")))) + "(l1)*2,nothing*3,l1w*2") + +(define_insn_reservation "branch_s1n" 6 + (and (eq_attr "type" "branch") + (and (eq_attr "cross" "n") + (and (eq_attr "units" "s") + (eq_attr "dest_regfile" "a")))) + "(s1+s1w)+br1") + +(define_insn_reservation "call_addkpc_s1n" 6 + (and (eq_attr "type" "call") + (and (ne (symbol_ref "TARGET_INSNS_64") (const_int 0)) + (and (eq_attr "cross" "n") + (and (eq_attr "units" "s") + (eq_attr "dest_regfile" "a"))))) + "(s1+s1w)+br1,s2+br0+br1") + +(define_insn_reservation "call_mvk_s1n" 6 + (and (eq_attr "type" "call") + (and (eq (symbol_ref "TARGET_INSNS_64") (const_int 0)) + (and (eq_attr "cross" "n") + (and (eq_attr "units" "s") + (eq_attr "dest_regfile" "a"))))) + "(s1+s1w)+br1,s2,s2") + +(define_insn_reservation "single_s1n" 1 + (and (eq_attr "type" "single") + (and (eq_attr "cross" "n") + (and (eq_attr "units" "s") + (eq_attr "dest_regfile" "a")))) + "(s1+s1w)") + +(define_insn_reservation "cmpdp_s1n" 2 + (and (eq_attr "type" "cmpdp") + (and (eq_attr "cross" "n") + (and (eq_attr "units" "s") + (eq_attr "dest_regfile" "a")))) + "s1,(s1)+s1w") + +(define_insn_reservation "dp2_s1n" 2 + (and (eq_attr "type" "dp2") + (and (eq_attr "cross" "n") + (and (eq_attr "units" "s") + (eq_attr "dest_regfile" "a")))) + "s1+s1w,s1w") + +(define_insn_reservation "fp4_s1n" 4 + (and (eq_attr "type" "fp4") + (and (eq_attr "cross" "n") + (and (eq_attr "units" "s") + (eq_attr "dest_regfile" "a")))) + "s1,nothing*2,s1w") + +(define_insn_reservation "mvilc4_s1n" 4 + (and (eq_attr "type" "mvilc") + (and (eq_attr "cross" "n") + (and (eq_attr "units" "s") + (eq_attr "dest_regfile" "a")))) + "(s1+s1w)") + +(define_insn_reservation "single_dl1n" 1 + (and (eq_attr "type" "single") + (and (eq_attr "cross" "n") + (and (eq_attr "units" "dl") + (eq_attr "dest_regfile" "a")))) + "(d1|(l1+l1w))") + +(define_insn_reservation "single_ds1n" 1 + (and (eq_attr "type" "single") + (and (eq_attr "cross" "n") + (and (eq_attr "units" "ds") + (eq_attr "dest_regfile" "a")))) + "(d1|(s1+s1w))") + +(define_insn_reservation "single_ls1n" 1 + (and (eq_attr "type" "single") + (and (eq_attr "cross" "n") + (and (eq_attr "units" "ls") + (eq_attr "dest_regfile" "a")))) + "((l1+l1w)|(s1+s1w))") + +(define_insn_reservation "dp2_l1n" 2 + (and (eq_attr "type" "dp2") + (and (eq_attr "cross" "n") + (and (eq_attr "units" "l") + (eq_attr "dest_regfile" "a")))) + "l1+l1w,l1w") + +(define_insn_reservation "fp4_ls1n" 4 + (and (eq_attr "type" "fp4") + (and (eq_attr "cross" "n") + (and (eq_attr "units" "ls") + (eq_attr "dest_regfile" "a")))) + "(s1,nothing*2,s1w)|(l1,nothing*2,l1w)") + +(define_insn_reservation "adddp_ls1n" 7 + (and (eq_attr "type" "adddp") + (and (eq_attr "cross" "n") + (and (eq_attr "units" "ls") + (eq_attr "dest_regfile" "a")))) + "((s1)*2,nothing*3,s1w*2)|((l1)*2,nothing*3,l1w*2)") + +(define_insn_reservation "single_dls1n" 1 + (and (eq_attr "type" "single") + (and (eq_attr "cross" "n") + (and (eq_attr "units" "dls") + (eq_attr "dest_regfile" "a")))) + "(d1|(l1+l1w)|(s1+s1w))") + +(define_insn_reservation "mpy2_m1n" 2 + (and (eq_attr "type" "mpy2") + (and (eq_attr "cross" "n") + (and (eq_attr "units" "m") + (eq_attr "dest_regfile" "a")))) + "m1,m1w") + +(define_insn_reservation "mpy4_m1n" 4 + (and (eq_attr "type" "mpy4") + (and (eq_attr "cross" "n") + (and (eq_attr "units" "m") + (eq_attr "dest_regfile" "a")))) + "m1,nothing,nothing,m1w") + +(define_insn_reservation "mpydp_m1n" 10 + (and (eq_attr "type" "mpydp") + (and (eq_attr "cross" "n") + (and (eq_attr "units" "m") + (eq_attr "dest_regfile" "a")))) + "(m1)*4,nothing*4,m1w*2") + +(define_insn_reservation "mpyspdp_m1n" 7 + (and (eq_attr "type" "mpyspdp") + (and (eq_attr "cross" "n") + (and (eq_attr "units" "m") + (eq_attr "dest_regfile" "a")))) + "(m1)*2,nothing*3,m1w*2") + +(define_insn_reservation "mpysp2dp_m1n" 5 + (and (eq_attr "type" "mpysp2dp") + (and (eq_attr "cross" "n") + (and (eq_attr "units" "m") + (eq_attr "dest_regfile" "a")))) + "m1,nothing*2,m1w*2") + +;; Definitions for side 2, cross n + +;; Scheduling description for TI C6X. +;; Copyright (C) 2010, 2011 Free Software Foundation, Inc. +;; Contributed by Bernd Schmidt <bernds@codesourcery.com> +;; Contributed by CodeSourcery. +;; +;; This file is part of GCC. +;; +;; GCC is free software; you can redistribute it and/or modify +;; it under the terms of the GNU General Public License as published by +;; the Free Software Foundation; either version 3, or (at your option) +;; any later version. +;; +;; GCC is distributed in the hope that it will be useful, +;; but WITHOUT ANY WARRANTY; without even the implied warranty of +;; MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the +;; GNU General Public License for more details. +;; +;; You should have received a copy of the GNU General Public License +;; along with GCC; see the file COPYING3. If not see +;; <http://www.gnu.org/licenses/>. + +;; Input file for gensched.sh We process this file multiple times, +;; replacing 2 with either 1 or 2 for each of the sides of the +;; machine, and b correspondingly with "a" or "b". n and +;; are replaced with yes/no and the appropriate reservation. + +(define_insn_reservation "load_d2n" 5 + (and (eq_attr "type" "load") + (and (eq_attr "cross" "n") + (and (eq_attr "units" "d_addr") + (eq_attr "addr_regfile" "b")))) + "d2+t2") + +(define_insn_reservation "store_d2n" 1 + (and (eq_attr "type" "store") + (and (eq_attr "cross" "n") + (and (eq_attr "units" "d_addr") + (eq_attr "addr_regfile" "b")))) + "d2+t2") + +(define_insn_reservation "loadn_d2n" 5 + (and (eq_attr "type" "loadn") + (and (eq_attr "cross" "n") + (and (eq_attr "units" "d_addr") + (eq_attr "addr_regfile" "b")))) + "d2+t1+t2") + +(define_insn_reservation "storen_d2n" 1 + (and (eq_attr "type" "storen") + (and (eq_attr "cross" "n") + (and (eq_attr "units" "d_addr") + (eq_attr "addr_regfile" "b")))) + "d2+t1+t2") + +(define_insn_reservation "single_d2n" 1 + (and (eq_attr "type" "single") + (and (eq_attr "cross" "n") + (and (eq_attr "units" "d") + (eq_attr "dest_regfile" "b")))) + "d2") + +(define_insn_reservation "single_l2n" 1 + (and (eq_attr "type" "single") + (and (eq_attr "cross" "n") + (and (eq_attr "units" "l") + (eq_attr "dest_regfile" "b")))) + "l2+l2w") + +(define_insn_reservation "fp4_l2n" 4 + (and (eq_attr "type" "fp4") + (and (eq_attr "cross" "n") + (and (eq_attr "units" "l") + (eq_attr "dest_regfile" "b")))) + "l2,nothing*2,l2w") + +(define_insn_reservation "intdp_l2n" 5 + (and (eq_attr "type" "intdp") + (and (eq_attr "cross" "n") + (and (eq_attr "units" "l") + (eq_attr "dest_regfile" "b")))) + "l2,nothing*2,l2w*2") + +(define_insn_reservation "adddp_l2n" 7 + (and (eq_attr "type" "adddp") + (and (eq_attr "cross" "n") + (and (eq_attr "units" "l") + (eq_attr "dest_regfile" "b")))) + "(l2)*2,nothing*3,l2w*2") + +(define_insn_reservation "branch_s2n" 6 + (and (eq_attr "type" "branch") + (and (eq_attr "cross" "n") + (and (eq_attr "units" "s") + (eq_attr "dest_regfile" "b")))) + "(s2+s2w)+br1") + +(define_insn_reservation "call_addkpc_s2n" 6 + (and (eq_attr "type" "call") + (and (ne (symbol_ref "TARGET_INSNS_64") (const_int 0)) + (and (eq_attr "cross" "n") + (and (eq_attr "units" "s") + (eq_attr "dest_regfile" "b"))))) + "(s2+s2w)+br1,s2+br0+br1") + +(define_insn_reservation "call_mvk_s2n" 6 + (and (eq_attr "type" "call") + (and (eq (symbol_ref "TARGET_INSNS_64") (const_int 0)) + (and (eq_attr "cross" "n") + (and (eq_attr "units" "s") + (eq_attr "dest_regfile" "b"))))) + "(s2+s2w)+br1,s2,s2") + +(define_insn_reservation "single_s2n" 1 + (and (eq_attr "type" "single") + (and (eq_attr "cross" "n") + (and (eq_attr "units" "s") + (eq_attr "dest_regfile" "b")))) + "(s2+s2w)") + +(define_insn_reservation "cmpdp_s2n" 2 + (and (eq_attr "type" "cmpdp") + (and (eq_attr "cross" "n") + (and (eq_attr "units" "s") + (eq_attr "dest_regfile" "b")))) + "s2,(s2)+s2w") + +(define_insn_reservation "dp2_s2n" 2 + (and (eq_attr "type" "dp2") + (and (eq_attr "cross" "n") + (and (eq_attr "units" "s") + (eq_attr "dest_regfile" "b")))) + "s2+s2w,s2w") + +(define_insn_reservation "fp4_s2n" 4 + (and (eq_attr "type" "fp4") + (and (eq_attr "cross" "n") + (and (eq_attr "units" "s") + (eq_attr "dest_regfile" "b")))) + "s2,nothing*2,s2w") + +(define_insn_reservation "mvilc4_s2n" 4 + (and (eq_attr "type" "mvilc") + (and (eq_attr "cross" "n") + (and (eq_attr "units" "s") + (eq_attr "dest_regfile" "b")))) + "(s2+s2w)") + +(define_insn_reservation "single_dl2n" 1 + (and (eq_attr "type" "single") + (and (eq_attr "cross" "n") + (and (eq_attr "units" "dl") + (eq_attr "dest_regfile" "b")))) + "(d2|(l2+l2w))") + +(define_insn_reservation "single_ds2n" 1 + (and (eq_attr "type" "single") + (and (eq_attr "cross" "n") + (and (eq_attr "units" "ds") + (eq_attr "dest_regfile" "b")))) + "(d2|(s2+s2w))") + +(define_insn_reservation "single_ls2n" 1 + (and (eq_attr "type" "single") + (and (eq_attr "cross" "n") + (and (eq_attr "units" "ls") + (eq_attr "dest_regfile" "b")))) + "((l2+l2w)|(s2+s2w))") + +(define_insn_reservation "dp2_l2n" 2 + (and (eq_attr "type" "dp2") + (and (eq_attr "cross" "n") + (and (eq_attr "units" "l") + (eq_attr "dest_regfile" "b")))) + "l2+l2w,l2w") + +(define_insn_reservation "fp4_ls2n" 4 + (and (eq_attr "type" "fp4") + (and (eq_attr "cross" "n") + (and (eq_attr "units" "ls") + (eq_attr "dest_regfile" "b")))) + "(s2,nothing*2,s2w)|(l2,nothing*2,l2w)") + +(define_insn_reservation "adddp_ls2n" 7 + (and (eq_attr "type" "adddp") + (and (eq_attr "cross" "n") + (and (eq_attr "units" "ls") + (eq_attr "dest_regfile" "b")))) + "((s2)*2,nothing*3,s2w*2)|((l2)*2,nothing*3,l2w*2)") + +(define_insn_reservation "single_dls2n" 1 + (and (eq_attr "type" "single") + (and (eq_attr "cross" "n") + (and (eq_attr "units" "dls") + (eq_attr "dest_regfile" "b")))) + "(d2|(l2+l2w)|(s2+s2w))") + +(define_insn_reservation "mpy2_m2n" 2 + (and (eq_attr "type" "mpy2") + (and (eq_attr "cross" "n") + (and (eq_attr "units" "m") + (eq_attr "dest_regfile" "b")))) + "m2,m2w") + +(define_insn_reservation "mpy4_m2n" 4 + (and (eq_attr "type" "mpy4") + (and (eq_attr "cross" "n") + (and (eq_attr "units" "m") + (eq_attr "dest_regfile" "b")))) + "m2,nothing,nothing,m2w") + +(define_insn_reservation "mpydp_m2n" 10 + (and (eq_attr "type" "mpydp") + (and (eq_attr "cross" "n") + (and (eq_attr "units" "m") + (eq_attr "dest_regfile" "b")))) + "(m2)*4,nothing*4,m2w*2") + +(define_insn_reservation "mpyspdp_m2n" 7 + (and (eq_attr "type" "mpyspdp") + (and (eq_attr "cross" "n") + (and (eq_attr "units" "m") + (eq_attr "dest_regfile" "b")))) + "(m2)*2,nothing*3,m2w*2") + +(define_insn_reservation "mpysp2dp_m2n" 5 + (and (eq_attr "type" "mpysp2dp") + (and (eq_attr "cross" "n") + (and (eq_attr "units" "m") + (eq_attr "dest_regfile" "b")))) + "m2,nothing*2,m2w*2") + +;; Definitions for side 1, cross y + +;; Scheduling description for TI C6X. +;; Copyright (C) 2010, 2011 Free Software Foundation, Inc. +;; Contributed by Bernd Schmidt <bernds@codesourcery.com> +;; Contributed by CodeSourcery. +;; +;; This file is part of GCC. +;; +;; GCC is free software; you can redistribute it and/or modify +;; it under the terms of the GNU General Public License as published by +;; the Free Software Foundation; either version 3, or (at your option) +;; any later version. +;; +;; GCC is distributed in the hope that it will be useful, +;; but WITHOUT ANY WARRANTY; without even the implied warranty of +;; MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the +;; GNU General Public License for more details. +;; +;; You should have received a copy of the GNU General Public License +;; along with GCC; see the file COPYING3. If not see +;; <http://www.gnu.org/licenses/>. + +;; Input file for gensched.sh We process this file multiple times, +;; replacing 1 with either 1 or 2 for each of the sides of the +;; machine, and a correspondingly with "a" or "b". y and +;; +x1 are replaced with yes/no and the appropriate reservation. + +(define_insn_reservation "load_d1y" 5 + (and (eq_attr "type" "load") + (and (eq_attr "cross" "y") + (and (eq_attr "units" "d_addr") + (eq_attr "addr_regfile" "a")))) + "d1+t2") + +(define_insn_reservation "store_d1y" 1 + (and (eq_attr "type" "store") + (and (eq_attr "cross" "y") + (and (eq_attr "units" "d_addr") + (eq_attr "addr_regfile" "a")))) + "d1+t2") + +(define_insn_reservation "loadn_d1y" 5 + (and (eq_attr "type" "loadn") + (and (eq_attr "cross" "y") + (and (eq_attr "units" "d_addr") + (eq_attr "addr_regfile" "a")))) + "d1+t1+t2") + +(define_insn_reservation "storen_d1y" 1 + (and (eq_attr "type" "storen") + (and (eq_attr "cross" "y") + (and (eq_attr "units" "d_addr") + (eq_attr "addr_regfile" "a")))) + "d1+t1+t2") + +(define_insn_reservation "single_d1y" 1 + (and (eq_attr "type" "single") + (and (eq_attr "cross" "y") + (and (eq_attr "units" "d") + (eq_attr "dest_regfile" "a")))) + "d1+x1") + +(define_insn_reservation "single_l1y" 1 + (and (eq_attr "type" "single") + (and (eq_attr "cross" "y") + (and (eq_attr "units" "l") + (eq_attr "dest_regfile" "a")))) + "l1+l1w+x1") + +(define_insn_reservation "fp4_l1y" 4 + (and (eq_attr "type" "fp4") + (and (eq_attr "cross" "y") + (and (eq_attr "units" "l") + (eq_attr "dest_regfile" "a")))) + "l1+x1,nothing*2,l1w") + +(define_insn_reservation "intdp_l1y" 5 + (and (eq_attr "type" "intdp") + (and (eq_attr "cross" "y") + (and (eq_attr "units" "l") + (eq_attr "dest_regfile" "a")))) + "l1+x1,nothing*2,l1w*2") + +(define_insn_reservation "adddp_l1y" 7 + (and (eq_attr "type" "adddp") + (and (eq_attr "cross" "y") + (and (eq_attr "units" "l") + (eq_attr "dest_regfile" "a")))) + "(l1+x1)*2,nothing*3,l1w*2") + +(define_insn_reservation "branch_s1y" 6 + (and (eq_attr "type" "branch") + (and (eq_attr "cross" "y") + (and (eq_attr "units" "s") + (eq_attr "dest_regfile" "a")))) + "(s1+s1w)+x1+br1") + +(define_insn_reservation "call_addkpc_s1y" 6 + (and (eq_attr "type" "call") + (and (ne (symbol_ref "TARGET_INSNS_64") (const_int 0)) + (and (eq_attr "cross" "y") + (and (eq_attr "units" "s") + (eq_attr "dest_regfile" "a"))))) + "(s1+s1w)+x1+br1,s2+br0+br1") + +(define_insn_reservation "call_mvk_s1y" 6 + (and (eq_attr "type" "call") + (and (eq (symbol_ref "TARGET_INSNS_64") (const_int 0)) + (and (eq_attr "cross" "y") + (and (eq_attr "units" "s") + (eq_attr "dest_regfile" "a"))))) + "(s1+s1w)+x1+br1,s2,s2") + +(define_insn_reservation "single_s1y" 1 + (and (eq_attr "type" "single") + (and (eq_attr "cross" "y") + (and (eq_attr "units" "s") + (eq_attr "dest_regfile" "a")))) + "(s1+s1w)+x1") + +(define_insn_reservation "cmpdp_s1y" 2 + (and (eq_attr "type" "cmpdp") + (and (eq_attr "cross" "y") + (and (eq_attr "units" "s") + (eq_attr "dest_regfile" "a")))) + "s1+x1,(s1+x1)+s1w") + +(define_insn_reservation "dp2_s1y" 2 + (and (eq_attr "type" "dp2") + (and (eq_attr "cross" "y") + (and (eq_attr "units" "s") + (eq_attr "dest_regfile" "a")))) + "s1+s1w+x1,s1w") + +(define_insn_reservation "fp4_s1y" 4 + (and (eq_attr "type" "fp4") + (and (eq_attr "cross" "y") + (and (eq_attr "units" "s") + (eq_attr "dest_regfile" "a")))) + "s1+x1,nothing*2,s1w") + +(define_insn_reservation "mvilc4_s1y" 4 + (and (eq_attr "type" "mvilc") + (and (eq_attr "cross" "y") + (and (eq_attr "units" "s") + (eq_attr "dest_regfile" "a")))) + "(s1+s1w)+x1") + +(define_insn_reservation "single_dl1y" 1 + (and (eq_attr "type" "single") + (and (eq_attr "cross" "y") + (and (eq_attr "units" "dl") + (eq_attr "dest_regfile" "a")))) + "(d1|(l1+l1w))+x1") + +(define_insn_reservation "single_ds1y" 1 + (and (eq_attr "type" "single") + (and (eq_attr "cross" "y") + (and (eq_attr "units" "ds") + (eq_attr "dest_regfile" "a")))) + "(d1|(s1+s1w))+x1") + +(define_insn_reservation "single_ls1y" 1 + (and (eq_attr "type" "single") + (and (eq_attr "cross" "y") + (and (eq_attr "units" "ls") + (eq_attr "dest_regfile" "a")))) + "((l1+l1w)|(s1+s1w))+x1") + +(define_insn_reservation "dp2_l1y" 2 + (and (eq_attr "type" "dp2") + (and (eq_attr "cross" "y") + (and (eq_attr "units" "l") + (eq_attr "dest_regfile" "a")))) + "l1+l1w+x1,l1w") + +(define_insn_reservation "fp4_ls1y" 4 + (and (eq_attr "type" "fp4") + (and (eq_attr "cross" "y") + (and (eq_attr "units" "ls") + (eq_attr "dest_regfile" "a")))) + "(s1+x1,nothing*2,s1w)|(l1+x1,nothing*2,l1w)") + +(define_insn_reservation "adddp_ls1y" 7 + (and (eq_attr "type" "adddp") + (and (eq_attr "cross" "y") + (and (eq_attr "units" "ls") + (eq_attr "dest_regfile" "a")))) + "((s1+x1)*2,nothing*3,s1w*2)|((l1+x1)*2,nothing*3,l1w*2)") + +(define_insn_reservation "single_dls1y" 1 + (and (eq_attr "type" "single") + (and (eq_attr "cross" "y") + (and (eq_attr "units" "dls") + (eq_attr "dest_regfile" "a")))) + "(d1|(l1+l1w)|(s1+s1w))+x1") + +(define_insn_reservation "mpy2_m1y" 2 + (and (eq_attr "type" "mpy2") + (and (eq_attr "cross" "y") + (and (eq_attr "units" "m") + (eq_attr "dest_regfile" "a")))) + "m1+x1,m1w") + +(define_insn_reservation "mpy4_m1y" 4 + (and (eq_attr "type" "mpy4") + (and (eq_attr "cross" "y") + (and (eq_attr "units" "m") + (eq_attr "dest_regfile" "a")))) + "m1+x1,nothing,nothing,m1w") + +(define_insn_reservation "mpydp_m1y" 10 + (and (eq_attr "type" "mpydp") + (and (eq_attr "cross" "y") + (and (eq_attr "units" "m") + (eq_attr "dest_regfile" "a")))) + "(m1+x1)*4,nothing*4,m1w*2") + +(define_insn_reservation "mpyspdp_m1y" 7 + (and (eq_attr "type" "mpyspdp") + (and (eq_attr "cross" "y") + (and (eq_attr "units" "m") + (eq_attr "dest_regfile" "a")))) + "(m1+x1)*2,nothing*3,m1w*2") + +(define_insn_reservation "mpysp2dp_m1y" 5 + (and (eq_attr "type" "mpysp2dp") + (and (eq_attr "cross" "y") + (and (eq_attr "units" "m") + (eq_attr "dest_regfile" "a")))) + "m1+x1,nothing*2,m1w*2") + +;; Definitions for side 2, cross y + +;; Scheduling description for TI C6X. +;; Copyright (C) 2010, 2011 Free Software Foundation, Inc. +;; Contributed by Bernd Schmidt <bernds@codesourcery.com> +;; Contributed by CodeSourcery. +;; +;; This file is part of GCC. +;; +;; GCC is free software; you can redistribute it and/or modify +;; it under the terms of the GNU General Public License as published by +;; the Free Software Foundation; either version 3, or (at your option) +;; any later version. +;; +;; GCC is distributed in the hope that it will be useful, +;; but WITHOUT ANY WARRANTY; without even the implied warranty of +;; MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the +;; GNU General Public License for more details. +;; +;; You should have received a copy of the GNU General Public License +;; along with GCC; see the file COPYING3. If not see +;; <http://www.gnu.org/licenses/>. + +;; Input file for gensched.sh We process this file multiple times, +;; replacing 2 with either 1 or 2 for each of the sides of the +;; machine, and b correspondingly with "a" or "b". y and +;; +x2 are replaced with yes/no and the appropriate reservation. + +(define_insn_reservation "load_d2y" 5 + (and (eq_attr "type" "load") + (and (eq_attr "cross" "y") + (and (eq_attr "units" "d_addr") + (eq_attr "addr_regfile" "b")))) + "d2+t1") + +(define_insn_reservation "store_d2y" 1 + (and (eq_attr "type" "store") + (and (eq_attr "cross" "y") + (and (eq_attr "units" "d_addr") + (eq_attr "addr_regfile" "b")))) + "d2+t1") + +(define_insn_reservation "loadn_d2y" 5 + (and (eq_attr "type" "loadn") + (and (eq_attr "cross" "y") + (and (eq_attr "units" "d_addr") + (eq_attr "addr_regfile" "b")))) + "d2+t1+t2") + +(define_insn_reservation "storen_d2y" 1 + (and (eq_attr "type" "storen") + (and (eq_attr "cross" "y") + (and (eq_attr "units" "d_addr") + (eq_attr "addr_regfile" "b")))) + "d2+t1+t2") + +(define_insn_reservation "single_d2y" 1 + (and (eq_attr "type" "single") + (and (eq_attr "cross" "y") + (and (eq_attr "units" "d") + (eq_attr "dest_regfile" "b")))) + "d2+x2") + +(define_insn_reservation "single_l2y" 1 + (and (eq_attr "type" "single") + (and (eq_attr "cross" "y") + (and (eq_attr "units" "l") + (eq_attr "dest_regfile" "b")))) + "l2+l2w+x2") + +(define_insn_reservation "fp4_l2y" 4 + (and (eq_attr "type" "fp4") + (and (eq_attr "cross" "y") + (and (eq_attr "units" "l") + (eq_attr "dest_regfile" "b")))) + "l2+x2,nothing*2,l2w") + +(define_insn_reservation "intdp_l2y" 5 + (and (eq_attr "type" "intdp") + (and (eq_attr "cross" "y") + (and (eq_attr "units" "l") + (eq_attr "dest_regfile" "b")))) + "l2+x2,nothing*2,l2w*2") + +(define_insn_reservation "adddp_l2y" 7 + (and (eq_attr "type" "adddp") + (and (eq_attr "cross" "y") + (and (eq_attr "units" "l") + (eq_attr "dest_regfile" "b")))) + "(l2+x2)*2,nothing*3,l2w*2") + +(define_insn_reservation "branch_s2y" 6 + (and (eq_attr "type" "branch") + (and (eq_attr "cross" "y") + (and (eq_attr "units" "s") + (eq_attr "dest_regfile" "b")))) + "(s2+s2w)+x2+br1") + +(define_insn_reservation "call_addkpc_s2y" 6 + (and (eq_attr "type" "call") + (and (ne (symbol_ref "TARGET_INSNS_64") (const_int 0)) + (and (eq_attr "cross" "y") + (and (eq_attr "units" "s") + (eq_attr "dest_regfile" "b"))))) + "(s2+s2w)+x2+br1,s2+br0+br1") + +(define_insn_reservation "call_mvk_s2y" 6 + (and (eq_attr "type" "call") + (and (eq (symbol_ref "TARGET_INSNS_64") (const_int 0)) + (and (eq_attr "cross" "y") + (and (eq_attr "units" "s") + (eq_attr "dest_regfile" "b"))))) + "(s2+s2w)+x2+br1,s2,s2") + +(define_insn_reservation "single_s2y" 1 + (and (eq_attr "type" "single") + (and (eq_attr "cross" "y") + (and (eq_attr "units" "s") + (eq_attr "dest_regfile" "b")))) + "(s2+s2w)+x2") + +(define_insn_reservation "cmpdp_s2y" 2 + (and (eq_attr "type" "cmpdp") + (and (eq_attr "cross" "y") + (and (eq_attr "units" "s") + (eq_attr "dest_regfile" "b")))) + "s2+x2,(s2+x2)+s2w") + +(define_insn_reservation "dp2_s2y" 2 + (and (eq_attr "type" "dp2") + (and (eq_attr "cross" "y") + (and (eq_attr "units" "s") + (eq_attr "dest_regfile" "b")))) + "s2+s2w+x2,s2w") + +(define_insn_reservation "fp4_s2y" 4 + (and (eq_attr "type" "fp4") + (and (eq_attr "cross" "y") + (and (eq_attr "units" "s") + (eq_attr "dest_regfile" "b")))) + "s2+x2,nothing*2,s2w") + +(define_insn_reservation "mvilc4_s2y" 4 + (and (eq_attr "type" "mvilc") + (and (eq_attr "cross" "y") + (and (eq_attr "units" "s") + (eq_attr "dest_regfile" "b")))) + "(s2+s2w)+x2") + +(define_insn_reservation "single_dl2y" 1 + (and (eq_attr "type" "single") + (and (eq_attr "cross" "y") + (and (eq_attr "units" "dl") + (eq_attr "dest_regfile" "b")))) + "(d2|(l2+l2w))+x2") + +(define_insn_reservation "single_ds2y" 1 + (and (eq_attr "type" "single") + (and (eq_attr "cross" "y") + (and (eq_attr "units" "ds") + (eq_attr "dest_regfile" "b")))) + "(d2|(s2+s2w))+x2") + +(define_insn_reservation "single_ls2y" 1 + (and (eq_attr "type" "single") + (and (eq_attr "cross" "y") + (and (eq_attr "units" "ls") + (eq_attr "dest_regfile" "b")))) + "((l2+l2w)|(s2+s2w))+x2") + +(define_insn_reservation "dp2_l2y" 2 + (and (eq_attr "type" "dp2") + (and (eq_attr "cross" "y") + (and (eq_attr "units" "l") + (eq_attr "dest_regfile" "b")))) + "l2+l2w+x2,l2w") + +(define_insn_reservation "fp4_ls2y" 4 + (and (eq_attr "type" "fp4") + (and (eq_attr "cross" "y") + (and (eq_attr "units" "ls") + (eq_attr "dest_regfile" "b")))) + "(s2+x2,nothing*2,s2w)|(l2+x2,nothing*2,l2w)") + +(define_insn_reservation "adddp_ls2y" 7 + (and (eq_attr "type" "adddp") + (and (eq_attr "cross" "y") + (and (eq_attr "units" "ls") + (eq_attr "dest_regfile" "b")))) + "((s2+x2)*2,nothing*3,s2w*2)|((l2+x2)*2,nothing*3,l2w*2)") + +(define_insn_reservation "single_dls2y" 1 + (and (eq_attr "type" "single") + (and (eq_attr "cross" "y") + (and (eq_attr "units" "dls") + (eq_attr "dest_regfile" "b")))) + "(d2|(l2+l2w)|(s2+s2w))+x2") + +(define_insn_reservation "mpy2_m2y" 2 + (and (eq_attr "type" "mpy2") + (and (eq_attr "cross" "y") + (and (eq_attr "units" "m") + (eq_attr "dest_regfile" "b")))) + "m2+x2,m2w") + +(define_insn_reservation "mpy4_m2y" 4 + (and (eq_attr "type" "mpy4") + (and (eq_attr "cross" "y") + (and (eq_attr "units" "m") + (eq_attr "dest_regfile" "b")))) + "m2+x2,nothing,nothing,m2w") + +(define_insn_reservation "mpydp_m2y" 10 + (and (eq_attr "type" "mpydp") + (and (eq_attr "cross" "y") + (and (eq_attr "units" "m") + (eq_attr "dest_regfile" "b")))) + "(m2+x2)*4,nothing*4,m2w*2") + +(define_insn_reservation "mpyspdp_m2y" 7 + (and (eq_attr "type" "mpyspdp") + (and (eq_attr "cross" "y") + (and (eq_attr "units" "m") + (eq_attr "dest_regfile" "b")))) + "(m2+x2)*2,nothing*3,m2w*2") + +(define_insn_reservation "mpysp2dp_m2y" 5 + (and (eq_attr "type" "mpysp2dp") + (and (eq_attr "cross" "y") + (and (eq_attr "units" "m") + (eq_attr "dest_regfile" "b")))) + "m2+x2,nothing*2,m2w*2") diff --git a/gcc/config/c6x/c6x-sched.md.in b/gcc/config/c6x/c6x-sched.md.in new file mode 100644 index 0000000..271109b --- /dev/null +++ b/gcc/config/c6x/c6x-sched.md.in @@ -0,0 +1,230 @@ +;; Scheduling description for TI C6X. +;; Copyright (C) 2010, 2011 Free Software Foundation, Inc. +;; Contributed by Bernd Schmidt <bernds@codesourcery.com> +;; Contributed by CodeSourcery. +;; +;; This file is part of GCC. +;; +;; GCC is free software; you can redistribute it and/or modify +;; it under the terms of the GNU General Public License as published by +;; the Free Software Foundation; either version 3, or (at your option) +;; any later version. +;; +;; GCC is distributed in the hope that it will be useful, +;; but WITHOUT ANY WARRANTY; without even the implied warranty of +;; MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the +;; GNU General Public License for more details. +;; +;; You should have received a copy of the GNU General Public License +;; along with GCC; see the file COPYING3. If not see +;; <http://www.gnu.org/licenses/>. + +;; Input file for gensched.sh We process this file multiple times, +;; replacing _N_ with either 1 or 2 for each of the sides of the +;; machine, and _RF_ correspondingly with "a" or "b". _CROSS_ and +;; _CUNIT_ are replaced with yes/no and the appropriate reservation. + +(define_insn_reservation "load_d_N__CROSS_" 5 + (and (eq_attr "type" "load") + (and (eq_attr "cross" "_CROSS_") + (and (eq_attr "units" "d_addr") + (eq_attr "addr_regfile" "_RF_")))) + "d_N_+t_NX_") + +(define_insn_reservation "store_d_N__CROSS_" 1 + (and (eq_attr "type" "store") + (and (eq_attr "cross" "_CROSS_") + (and (eq_attr "units" "d_addr") + (eq_attr "addr_regfile" "_RF_")))) + "d_N_+t_NX_") + +(define_insn_reservation "loadn_d_N__CROSS_" 5 + (and (eq_attr "type" "loadn") + (and (eq_attr "cross" "_CROSS_") + (and (eq_attr "units" "d_addr") + (eq_attr "addr_regfile" "_RF_")))) + "d_N_+t1+t2") + +(define_insn_reservation "storen_d_N__CROSS_" 1 + (and (eq_attr "type" "storen") + (and (eq_attr "cross" "_CROSS_") + (and (eq_attr "units" "d_addr") + (eq_attr "addr_regfile" "_RF_")))) + "d_N_+t1+t2") + +(define_insn_reservation "single_d_N__CROSS_" 1 + (and (eq_attr "type" "single") + (and (eq_attr "cross" "_CROSS_") + (and (eq_attr "units" "d") + (eq_attr "dest_regfile" "_RF_")))) + "d_N__CUNIT_") + +(define_insn_reservation "single_l_N__CROSS_" 1 + (and (eq_attr "type" "single") + (and (eq_attr "cross" "_CROSS_") + (and (eq_attr "units" "l") + (eq_attr "dest_regfile" "_RF_")))) + "l_N_+l_N_w_CUNIT_") + +(define_insn_reservation "fp4_l_N__CROSS_" 4 + (and (eq_attr "type" "fp4") + (and (eq_attr "cross" "_CROSS_") + (and (eq_attr "units" "l") + (eq_attr "dest_regfile" "_RF_")))) + "l_N__CUNIT_,nothing*2,l_N_w") + +(define_insn_reservation "intdp_l_N__CROSS_" 5 + (and (eq_attr "type" "intdp") + (and (eq_attr "cross" "_CROSS_") + (and (eq_attr "units" "l") + (eq_attr "dest_regfile" "_RF_")))) + "l_N__CUNIT_,nothing*2,l_N_w*2") + +(define_insn_reservation "adddp_l_N__CROSS_" 7 + (and (eq_attr "type" "adddp") + (and (eq_attr "cross" "_CROSS_") + (and (eq_attr "units" "l") + (eq_attr "dest_regfile" "_RF_")))) + "(l_N__CUNIT_)*2,nothing*3,l_N_w*2") + +(define_insn_reservation "branch_s_N__CROSS_" 6 + (and (eq_attr "type" "branch") + (and (eq_attr "cross" "_CROSS_") + (and (eq_attr "units" "s") + (eq_attr "dest_regfile" "_RF_")))) + "(s_N_+s_N_w)_CUNIT_+br1") + +(define_insn_reservation "call_addkpc_s_N__CROSS_" 6 + (and (eq_attr "type" "call") + (and (ne (symbol_ref "TARGET_INSNS_64") (const_int 0)) + (and (eq_attr "cross" "_CROSS_") + (and (eq_attr "units" "s") + (eq_attr "dest_regfile" "_RF_"))))) + "(s_N_+s_N_w)_CUNIT_+br1,s2+br0+br1") + +(define_insn_reservation "call_mvk_s_N__CROSS_" 6 + (and (eq_attr "type" "call") + (and (eq (symbol_ref "TARGET_INSNS_64") (const_int 0)) + (and (eq_attr "cross" "_CROSS_") + (and (eq_attr "units" "s") + (eq_attr "dest_regfile" "_RF_"))))) + "(s_N_+s_N_w)_CUNIT_+br1,s2,s2") + +(define_insn_reservation "single_s_N__CROSS_" 1 + (and (eq_attr "type" "single") + (and (eq_attr "cross" "_CROSS_") + (and (eq_attr "units" "s") + (eq_attr "dest_regfile" "_RF_")))) + "(s_N_+s_N_w)_CUNIT_") + +(define_insn_reservation "cmpdp_s_N__CROSS_" 2 + (and (eq_attr "type" "cmpdp") + (and (eq_attr "cross" "_CROSS_") + (and (eq_attr "units" "s") + (eq_attr "dest_regfile" "_RF_")))) + "s_N__CUNIT_,(s_N__CUNIT_)+s_N_w") + +(define_insn_reservation "dp2_s_N__CROSS_" 2 + (and (eq_attr "type" "dp2") + (and (eq_attr "cross" "_CROSS_") + (and (eq_attr "units" "s") + (eq_attr "dest_regfile" "_RF_")))) + "s_N_+s_N_w_CUNIT_,s_N_w") + +(define_insn_reservation "fp4_s_N__CROSS_" 4 + (and (eq_attr "type" "fp4") + (and (eq_attr "cross" "_CROSS_") + (and (eq_attr "units" "s") + (eq_attr "dest_regfile" "_RF_")))) + "s_N__CUNIT_,nothing*2,s_N_w") + +(define_insn_reservation "mvilc4_s_N__CROSS_" 4 + (and (eq_attr "type" "mvilc") + (and (eq_attr "cross" "_CROSS_") + (and (eq_attr "units" "s") + (eq_attr "dest_regfile" "_RF_")))) + "(s_N_+s_N_w)_CUNIT_") + +(define_insn_reservation "single_dl_N__CROSS_" 1 + (and (eq_attr "type" "single") + (and (eq_attr "cross" "_CROSS_") + (and (eq_attr "units" "dl") + (eq_attr "dest_regfile" "_RF_")))) + "(d_N_|(l_N_+l_N_w))_CUNIT_") + +(define_insn_reservation "single_ds_N__CROSS_" 1 + (and (eq_attr "type" "single") + (and (eq_attr "cross" "_CROSS_") + (and (eq_attr "units" "ds") + (eq_attr "dest_regfile" "_RF_")))) + "(d_N_|(s_N_+s_N_w))_CUNIT_") + +(define_insn_reservation "single_ls_N__CROSS_" 1 + (and (eq_attr "type" "single") + (and (eq_attr "cross" "_CROSS_") + (and (eq_attr "units" "ls") + (eq_attr "dest_regfile" "_RF_")))) + "((l_N_+l_N_w)|(s_N_+s_N_w))_CUNIT_") + +(define_insn_reservation "dp2_l_N__CROSS_" 2 + (and (eq_attr "type" "dp2") + (and (eq_attr "cross" "_CROSS_") + (and (eq_attr "units" "l") + (eq_attr "dest_regfile" "_RF_")))) + "l_N_+l_N_w_CUNIT_,l_N_w") + +(define_insn_reservation "fp4_ls_N__CROSS_" 4 + (and (eq_attr "type" "fp4") + (and (eq_attr "cross" "_CROSS_") + (and (eq_attr "units" "ls") + (eq_attr "dest_regfile" "_RF_")))) + "(s_N__CUNIT_,nothing*2,s_N_w)|(l_N__CUNIT_,nothing*2,l_N_w)") + +(define_insn_reservation "adddp_ls_N__CROSS_" 7 + (and (eq_attr "type" "adddp") + (and (eq_attr "cross" "_CROSS_") + (and (eq_attr "units" "ls") + (eq_attr "dest_regfile" "_RF_")))) + "((s_N__CUNIT_)*2,nothing*3,s_N_w*2)|((l_N__CUNIT_)*2,nothing*3,l_N_w*2)") + +(define_insn_reservation "single_dls_N__CROSS_" 1 + (and (eq_attr "type" "single") + (and (eq_attr "cross" "_CROSS_") + (and (eq_attr "units" "dls") + (eq_attr "dest_regfile" "_RF_")))) + "(d_N_|(l_N_+l_N_w)|(s_N_+s_N_w))_CUNIT_") + +(define_insn_reservation "mpy2_m_N__CROSS_" 2 + (and (eq_attr "type" "mpy2") + (and (eq_attr "cross" "_CROSS_") + (and (eq_attr "units" "m") + (eq_attr "dest_regfile" "_RF_")))) + "m_N__CUNIT_,m_N_w") + +(define_insn_reservation "mpy4_m_N__CROSS_" 4 + (and (eq_attr "type" "mpy4") + (and (eq_attr "cross" "_CROSS_") + (and (eq_attr "units" "m") + (eq_attr "dest_regfile" "_RF_")))) + "m_N__CUNIT_,nothing,nothing,m_N_w") + +(define_insn_reservation "mpydp_m_N__CROSS_" 10 + (and (eq_attr "type" "mpydp") + (and (eq_attr "cross" "_CROSS_") + (and (eq_attr "units" "m") + (eq_attr "dest_regfile" "_RF_")))) + "(m_N__CUNIT_)*4,nothing*4,m_N_w*2") + +(define_insn_reservation "mpyspdp_m_N__CROSS_" 7 + (and (eq_attr "type" "mpyspdp") + (and (eq_attr "cross" "_CROSS_") + (and (eq_attr "units" "m") + (eq_attr "dest_regfile" "_RF_")))) + "(m_N__CUNIT_)*2,nothing*3,m_N_w*2") + +(define_insn_reservation "mpysp2dp_m_N__CROSS_" 5 + (and (eq_attr "type" "mpysp2dp") + (and (eq_attr "cross" "_CROSS_") + (and (eq_attr "units" "m") + (eq_attr "dest_regfile" "_RF_")))) + "m_N__CUNIT_,nothing*2,m_N_w*2") diff --git a/gcc/config/c6x/c6x-tables.opt b/gcc/config/c6x/c6x-tables.opt new file mode 100644 index 0000000..e012e9e --- /dev/null +++ b/gcc/config/c6x/c6x-tables.opt @@ -0,0 +1,43 @@ +; -*- buffer-read-only: t -*- +; Generated automatically by genopt.sh from c6x-isas.def. +; +; Copyright (C) 2011 Free Software Foundation, Inc. +; +; This file is part of GCC. +; +; GCC is free software; you can redistribute it and/or modify it under +; the terms of the GNU General Public License as published by the Free +; Software Foundation; either version 3, or (at your option) any later +; version. +; +; GCC is distributed in the hope that it will be useful, but WITHOUT ANY +; WARRANTY; without even the implied warranty of MERCHANTABILITY or +; FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License +; for more details. +; +; You should have received a copy of the GNU General Public License +; along with GCC; see the file COPYING3. If not see +; <http://www.gnu.org/licenses/>. + +Enum +Name(c6x_isa) Type(int) +Known C6X ISAs (for use with the -march= option): + +EnumValue +Enum(c6x_isa) String(c62x) Value(0) + +EnumValue +Enum(c6x_isa) String(c64x) Value(1) + +EnumValue +Enum(c6x_isa) String(c64x+) Value(2) + +EnumValue +Enum(c6x_isa) String(c67x) Value(3) + +EnumValue +Enum(c6x_isa) String(c67x+) Value(4) + +EnumValue +Enum(c6x_isa) String(c674x) Value(5) + diff --git a/gcc/config/c6x/c6x.c b/gcc/config/c6x/c6x.c new file mode 100644 index 0000000..deb2f53 --- /dev/null +++ b/gcc/config/c6x/c6x.c @@ -0,0 +1,5572 @@ +/* Target Code for TI C6X + Copyright (C) 2010, 2011 Free Software Foundation, Inc. + Contributed by Andrew Jenner <andrew@codesourcery.com> + Contributed by Bernd Schmidt <bernds@codesourcery.com> + + This file is part of GCC. + + GCC is free software; you can redistribute it and/or modify it + under the terms of the GNU General Public License as published + by the Free Software Foundation; either version 3, or (at your + option) any later version. + + GCC is distributed in the hope that it will be useful, but WITHOUT + ANY WARRANTY; without even the implied warranty of MERCHANTABILITY + or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public + License for more details. + + You should have received a copy of the GNU General Public License + along with GCC; see the file COPYING3. If not see + <http://www.gnu.org/licenses/>. */ + +#include "config.h" +#include "system.h" +#include "coretypes.h" +#include "tm.h" +#include "rtl.h" +#include "tree.h" +#include "insn-flags.h" +#include "output.h" +#include "insn-attr.h" +#include "insn-codes.h" +#include "expr.h" +#include "regs.h" +#include "optabs.h" +#include "recog.h" +#include "ggc.h" +#include "sched-int.h" +#include "timevar.h" +#include "tm_p.h" +#include "tm-preds.h" +#include "tm-constrs.h" +#include "df.h" +#include "integrate.h" +#include "diagnostic-core.h" +#include "cgraph.h" +#include "cfglayout.h" +#include "langhooks.h" +#include "target.h" +#include "target-def.h" +#include "sel-sched.h" +#include "debug.h" +#include "opts.h" + +/* Table of supported architecture variants. */ +typedef struct +{ + const char *arch; + enum c6x_cpu_type type; + unsigned short features; +} c6x_arch_table; + +/* A list of all ISAs, mapping each one to a representative device. + Used for -march selection. */ +static const c6x_arch_table all_isas[] = +{ +#define C6X_ISA(NAME,DEVICE,FLAGS) \ + { NAME, DEVICE, FLAGS }, +#include "c6x-isas.def" +#undef C6X_ISA + { NULL, C6X_CPU_C62X, 0 } +}; + +/* This is the parsed result of the "-march=" option, if given. */ +enum c6x_cpu_type c6x_arch = C6X_DEFAULT_ARCH; + +/* A mask of insn types that are allowed by the architecture selected by + the -march option. */ +unsigned long c6x_insn_mask = C6X_DEFAULT_INSN_MASK; + +/* The instruction that is being output (as obtained from FINAL_PRESCAN_INSN). + */ +static rtx c6x_current_insn = NULL_RTX; + +/* A decl we build to access __c6xabi_DSBT_base. */ +static GTY(()) tree dsbt_decl; + +/* Determines whether we run our final scheduling pass or not. We always + avoid the normal second scheduling pass. */ +static int c6x_flag_schedule_insns2; + +/* Determines whether we run variable tracking in machine dependent + reorganization. */ +static int c6x_flag_var_tracking; + +/* Determines whether we use modulo scheduling. */ +static int c6x_flag_modulo_sched; + +/* Record the state of flag_pic before we set it to 1 for DSBT. */ +int c6x_initial_flag_pic; + +typedef struct +{ + /* We record the clock cycle for every insn during scheduling. */ + int clock; + /* After scheduling, we run assign_reservations to choose unit + reservations for all insns. These are recorded here. */ + int reservation; + /* Records the new condition for insns which must be made + conditional after scheduling. An entry of NULL_RTX means no such + change is necessary. */ + rtx new_cond; + /* True for the first insn that was scheduled in an ebb. */ + bool ebb_start; +} c6x_sched_insn_info; + +DEF_VEC_O(c6x_sched_insn_info); +DEF_VEC_ALLOC_O(c6x_sched_insn_info, heap); + +/* Record a c6x_sched_insn_info structure for every insn in the function. */ +static VEC(c6x_sched_insn_info, heap) *insn_info; + +#define INSN_INFO_LENGTH (VEC_length (c6x_sched_insn_info, insn_info)) +#define INSN_INFO_ENTRY(N) (*VEC_index (c6x_sched_insn_info, insn_info, (N))) + +static bool done_cfi_sections; + +/* The DFA names of the units, in packet order. */ +static const char *const c6x_unit_names[] = +{ + "d1", "l1", "s1", "m1", + "d2", "l2", "s2", "m2", +}; + +#define RESERVATION_FLAG_D 1 +#define RESERVATION_FLAG_L 2 +#define RESERVATION_FLAG_S 4 +#define RESERVATION_FLAG_M 8 +#define RESERVATION_FLAG_DL (RESERVATION_FLAG_D | RESERVATION_FLAG_L) +#define RESERVATION_FLAG_DS (RESERVATION_FLAG_D | RESERVATION_FLAG_S) +#define RESERVATION_FLAG_LS (RESERVATION_FLAG_L | RESERVATION_FLAG_S) +#define RESERVATION_FLAG_DLS (RESERVATION_FLAG_D | RESERVATION_FLAG_LS) + +#define RESERVATION_S1 2 +#define RESERVATION_S2 6 + +/* Register map for debugging. */ +int const dbx_register_map[FIRST_PSEUDO_REGISTER] = +{ + 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, /* A0 - A15. */ + 37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, 49, /* A16 - A32. */ + 50, 51, 52, + 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, /* B0 - B15. */ + 29, 30, 31, + 53, 54, 55, 56, 57, 58, 59, 60, 61, 62, 63, 64, 65, /* B16 - B32. */ + 66, 67, 68, + -1, -1, -1 /* FP, ARGP, ILC. */ +}; + +/* Allocate a new, cleared machine_function structure. */ + +static struct machine_function * +c6x_init_machine_status (void) +{ + return ggc_alloc_cleared_machine_function (); +} + +/* Implement TARGET_OPTION_OVERRIDE. */ + +static void +c6x_option_override (void) +{ + if (global_options_set.x_c6x_arch_option) + { + c6x_arch = all_isas[c6x_arch_option].type; + c6x_insn_mask &= ~C6X_INSNS_ALL_CPU_BITS; + c6x_insn_mask |= all_isas[c6x_arch_option].features; + } + + c6x_flag_schedule_insns2 = flag_schedule_insns_after_reload; + flag_schedule_insns_after_reload = 0; + + c6x_flag_modulo_sched = flag_modulo_sched; + flag_modulo_sched = 0; + + init_machine_status = c6x_init_machine_status; + + if (flag_pic && !TARGET_DSBT) + { + error ("-fpic and -fPIC not supported without -mdsbt on this target"); + flag_pic = 0; + } + c6x_initial_flag_pic = flag_pic; + if (TARGET_DSBT && !flag_pic) + flag_pic = 1; +} + + +/* Implement the TARGET_CONDITIONAL_REGISTER_USAGE hook. */ + +static void +c6x_conditional_register_usage (void) +{ + int i; + if (c6x_arch == C6X_CPU_C62X || c6x_arch == C6X_CPU_C67X) + for (i = 16; i < 32; i++) + { + fixed_regs[i] = 1; + fixed_regs[32 + i] = 1; + } + if (TARGET_INSNS_64) + { + SET_HARD_REG_BIT (reg_class_contents[(int)PREDICATE_A_REGS], + REG_A0); + SET_HARD_REG_BIT (reg_class_contents[(int)PREDICATE_REGS], + REG_A0); + CLEAR_HARD_REG_BIT (reg_class_contents[(int)NONPREDICATE_A_REGS], + REG_A0); + CLEAR_HARD_REG_BIT (reg_class_contents[(int)NONPREDICATE_REGS], + REG_A0); + } +} + +static GTY(()) rtx eqdf_libfunc; +static GTY(()) rtx nedf_libfunc; +static GTY(()) rtx ledf_libfunc; +static GTY(()) rtx ltdf_libfunc; +static GTY(()) rtx gedf_libfunc; +static GTY(()) rtx gtdf_libfunc; +static GTY(()) rtx eqsf_libfunc; +static GTY(()) rtx nesf_libfunc; +static GTY(()) rtx lesf_libfunc; +static GTY(()) rtx ltsf_libfunc; +static GTY(()) rtx gesf_libfunc; +static GTY(()) rtx gtsf_libfunc; +static GTY(()) rtx strasgi_libfunc; +static GTY(()) rtx strasgi64p_libfunc; + +/* Implement the TARGET_INIT_LIBFUNCS macro. We use this to rename library + functions to match the C6x ABI. */ + +static void +c6x_init_libfuncs (void) +{ + /* Double-precision floating-point arithmetic. */ + set_optab_libfunc (add_optab, DFmode, "__c6xabi_addd"); + set_optab_libfunc (sdiv_optab, DFmode, "__c6xabi_divd"); + set_optab_libfunc (smul_optab, DFmode, "__c6xabi_mpyd"); + set_optab_libfunc (neg_optab, DFmode, "__c6xabi_negd"); + set_optab_libfunc (sub_optab, DFmode, "__c6xabi_subd"); + + /* Single-precision floating-point arithmetic. */ + set_optab_libfunc (add_optab, SFmode, "__c6xabi_addf"); + set_optab_libfunc (sdiv_optab, SFmode, "__c6xabi_divf"); + set_optab_libfunc (smul_optab, SFmode, "__c6xabi_mpyf"); + set_optab_libfunc (neg_optab, SFmode, "__c6xabi_negf"); + set_optab_libfunc (sub_optab, SFmode, "__c6xabi_subf"); + + /* Floating-point comparisons. */ + eqsf_libfunc = init_one_libfunc ("__c6xabi_eqf"); + nesf_libfunc = init_one_libfunc ("__c6xabi_neqf"); + lesf_libfunc = init_one_libfunc ("__c6xabi_lef"); + ltsf_libfunc = init_one_libfunc ("__c6xabi_ltf"); + gesf_libfunc = init_one_libfunc ("__c6xabi_gef"); + gtsf_libfunc = init_one_libfunc ("__c6xabi_gtf"); + eqdf_libfunc = init_one_libfunc ("__c6xabi_eqd"); + nedf_libfunc = init_one_libfunc ("__c6xabi_neqd"); + ledf_libfunc = init_one_libfunc ("__c6xabi_led"); + ltdf_libfunc = init_one_libfunc ("__c6xabi_ltd"); + gedf_libfunc = init_one_libfunc ("__c6xabi_ged"); + gtdf_libfunc = init_one_libfunc ("__c6xabi_gtd"); + + set_optab_libfunc (eq_optab, SFmode, NULL); + set_optab_libfunc (ne_optab, SFmode, "__c6xabi_neqf"); + set_optab_libfunc (gt_optab, SFmode, NULL); + set_optab_libfunc (ge_optab, SFmode, NULL); + set_optab_libfunc (lt_optab, SFmode, NULL); + set_optab_libfunc (le_optab, SFmode, NULL); + set_optab_libfunc (unord_optab, SFmode, "__c6xabi_unordf"); + set_optab_libfunc (eq_optab, DFmode, NULL); + set_optab_libfunc (ne_optab, DFmode, "__c6xabi_neqd"); + set_optab_libfunc (gt_optab, DFmode, NULL); + set_optab_libfunc (ge_optab, DFmode, NULL); + set_optab_libfunc (lt_optab, DFmode, NULL); + set_optab_libfunc (le_optab, DFmode, NULL); + set_optab_libfunc (unord_optab, DFmode, "__c6xabi_unordd"); + + /* Floating-point to integer conversions. */ + set_conv_libfunc (sfix_optab, SImode, DFmode, "__c6xabi_fixdi"); + set_conv_libfunc (ufix_optab, SImode, DFmode, "__c6xabi_fixdu"); + set_conv_libfunc (sfix_optab, DImode, DFmode, "__c6xabi_fixdlli"); + set_conv_libfunc (ufix_optab, DImode, DFmode, "__c6xabi_fixdull"); + set_conv_libfunc (sfix_optab, SImode, SFmode, "__c6xabi_fixfi"); + set_conv_libfunc (ufix_optab, SImode, SFmode, "__c6xabi_fixfu"); + set_conv_libfunc (sfix_optab, DImode, SFmode, "__c6xabi_fixflli"); + set_conv_libfunc (ufix_optab, DImode, SFmode, "__c6xabi_fixfull"); + + /* Conversions between floating types. */ + set_conv_libfunc (trunc_optab, SFmode, DFmode, "__c6xabi_cvtdf"); + set_conv_libfunc (sext_optab, DFmode, SFmode, "__c6xabi_cvtfd"); + + /* Integer to floating-point conversions. */ + set_conv_libfunc (sfloat_optab, DFmode, SImode, "__c6xabi_fltid"); + set_conv_libfunc (ufloat_optab, DFmode, SImode, "__c6xabi_fltud"); + set_conv_libfunc (sfloat_optab, DFmode, DImode, "__c6xabi_fltllid"); + set_conv_libfunc (ufloat_optab, DFmode, DImode, "__c6xabi_fltulld"); + set_conv_libfunc (sfloat_optab, SFmode, SImode, "__c6xabi_fltif"); + set_conv_libfunc (ufloat_optab, SFmode, SImode, "__c6xabi_fltuf"); + set_conv_libfunc (sfloat_optab, SFmode, DImode, "__c6xabi_fltllif"); + set_conv_libfunc (ufloat_optab, SFmode, DImode, "__c6xabi_fltullf"); + + /* Long long. */ + set_optab_libfunc (smul_optab, DImode, "__c6xabi_mpyll"); + set_optab_libfunc (ashl_optab, DImode, "__c6xabi_llshl"); + set_optab_libfunc (lshr_optab, DImode, "__c6xabi_llshru"); + set_optab_libfunc (ashr_optab, DImode, "__c6xabi_llshr"); + + set_optab_libfunc (sdiv_optab, SImode, "__c6xabi_divi"); + set_optab_libfunc (udiv_optab, SImode, "__c6xabi_divu"); + set_optab_libfunc (smod_optab, SImode, "__c6xabi_remi"); + set_optab_libfunc (umod_optab, SImode, "__c6xabi_remu"); + set_optab_libfunc (sdivmod_optab, SImode, "__c6xabi_divremi"); + set_optab_libfunc (udivmod_optab, SImode, "__c6xabi_divremu"); + set_optab_libfunc (sdiv_optab, DImode, "__c6xabi_divlli"); + set_optab_libfunc (udiv_optab, DImode, "__c6xabi_divull"); + set_optab_libfunc (smod_optab, DImode, "__c6xabi_remlli"); + set_optab_libfunc (umod_optab, DImode, "__c6xabi_remull"); + set_optab_libfunc (udivmod_optab, DImode, "__c6xabi_divremull"); + + /* Block move. */ + strasgi_libfunc = init_one_libfunc ("__c6xabi_strasgi"); + strasgi64p_libfunc = init_one_libfunc ("__c6xabi_strasgi_64plus"); +} + +/* Begin the assembly file. */ + +static void +c6x_file_start (void) +{ + /* Variable tracking should be run after all optimizations which change order + of insns. It also needs a valid CFG. This can't be done in + c6x_override_options, because flag_var_tracking is finalized after + that. */ + c6x_flag_var_tracking = flag_var_tracking; + flag_var_tracking = 0; + + done_cfi_sections = false; + default_file_start (); + + /* Arrays are aligned to 8-byte boundaries. */ + asm_fprintf (asm_out_file, + "\t.c6xabi_attribute Tag_ABI_array_object_alignment, 0\n"); + asm_fprintf (asm_out_file, + "\t.c6xabi_attribute Tag_ABI_array_object_align_expected, 0\n"); + + /* Stack alignment is 8 bytes. */ + asm_fprintf (asm_out_file, + "\t.c6xabi_attribute Tag_ABI_stack_align_needed, 0\n"); + asm_fprintf (asm_out_file, + "\t.c6xabi_attribute Tag_ABI_stack_align_preserved, 0\n"); + +#if 0 /* FIXME: Reenable when TI's tools are fixed. */ + /* ??? Ideally we'd check flag_short_wchar somehow. */ + asm_fprintf (asm_out_file, "\t.c6xabi_attribute Tag_ABI_wchar_t, %d\n", 2); +#endif + + /* We conform to version 1.0 of the ABI. */ + asm_fprintf (asm_out_file, + "\t.c6xabi_attribute Tag_ABI_conformance, \"1.0\"\n"); + +} + +/* The LTO frontend only enables exceptions when it sees a function that + uses it. This changes the return value of dwarf2out_do_frame, so we + have to check before every function. */ + +void +c6x_output_file_unwind (FILE * f) +{ + if (done_cfi_sections) + return; + + /* Output a .cfi_sections directive if we aren't + already doing so for debug info. */ + if (write_symbols != DWARF2_DEBUG && write_symbols != VMS_AND_DWARF2_DEBUG + && dwarf2out_do_frame ()) + { + asm_fprintf (f, "\t.cfi_sections .c6xabi.exidx\n"); + done_cfi_sections = true; + } +} + +/* Output unwind directives at the end of a function. */ + +static void +c6x_output_fn_unwind (FILE * f) +{ + /* Return immediately if we are not generating unwinding tables. */ + if (! (flag_unwind_tables || flag_exceptions)) + return; + + /* If this function will never be unwound, then mark it as such. */ + if (!(flag_unwind_tables || crtl->uses_eh_lsda) + && (TREE_NOTHROW (current_function_decl) + || crtl->all_throwers_are_sibcalls)) + fputs("\t.cantunwind\n", f); + + fputs ("\t.endp\n", f); +} + + +/* Stack and Calling. */ + +int argument_registers[10] = +{ + REG_A4, REG_B4, + REG_A6, REG_B6, + REG_A8, REG_B8, + REG_A10, REG_B10, + REG_A12, REG_B12 +}; + +/* Implements the macro INIT_CUMULATIVE_ARGS defined in c6x.h. */ + +void +c6x_init_cumulative_args (CUMULATIVE_ARGS *cum, const_tree fntype, rtx libname, + int n_named_args ATTRIBUTE_UNUSED) +{ + cum->count = 0; + cum->nregs = 10; + if (!libname && fntype) + { + /* We need to find out the number of named arguments. Unfortunately, + for incoming arguments, N_NAMED_ARGS is set to -1. */ + if (stdarg_p (fntype)) + cum->nregs = type_num_arguments (fntype) - 1; + if (cum->nregs > 10) + cum->nregs = 10; + } +} + +/* Implements the macro FUNCTION_ARG defined in c6x.h. */ + +static rtx +c6x_function_arg (cumulative_args_t cum_v, enum machine_mode mode, + const_tree type, bool named ATTRIBUTE_UNUSED) +{ + CUMULATIVE_ARGS *cum = get_cumulative_args (cum_v); + if (cum->count >= cum->nregs) + return NULL_RTX; + if (type) + { + HOST_WIDE_INT size = int_size_in_bytes (type); + if (TARGET_BIG_ENDIAN && AGGREGATE_TYPE_P (type)) + { + if (size > 4) + { + rtx reg1 = gen_rtx_REG (SImode, argument_registers[cum->count] + 1); + rtx reg2 = gen_rtx_REG (SImode, argument_registers[cum->count]); + rtvec vec = gen_rtvec (2, gen_rtx_EXPR_LIST (VOIDmode, reg1, const0_rtx), + gen_rtx_EXPR_LIST (VOIDmode, reg2, GEN_INT (4))); + return gen_rtx_PARALLEL (mode, vec); + } + } + } + return gen_rtx_REG (mode, argument_registers[cum->count]); +} + +static void +c6x_function_arg_advance (cumulative_args_t cum_v, + enum machine_mode mode ATTRIBUTE_UNUSED, + const_tree type ATTRIBUTE_UNUSED, + bool named ATTRIBUTE_UNUSED) +{ + CUMULATIVE_ARGS *cum = get_cumulative_args (cum_v); + cum->count++; +} + + +/* Return true if BLOCK_REG_PADDING (MODE, TYPE, FIRST) should return + upward rather than downward. */ + +bool +c6x_block_reg_pad_upward (enum machine_mode mode ATTRIBUTE_UNUSED, + const_tree type, bool first) +{ + HOST_WIDE_INT size; + + if (!TARGET_BIG_ENDIAN) + return true; + if (!first) + return true; + if (!type) + return true; + size = int_size_in_bytes (type); + return size == 3; +} + +/* Implement TARGET_FUNCTION_ARG_BOUNDARY. */ + +static unsigned int +c6x_function_arg_boundary (enum machine_mode mode, const_tree type) +{ + unsigned int boundary = type ? TYPE_ALIGN (type) : GET_MODE_BITSIZE (mode); + + if (boundary > BITS_PER_WORD) + return 2 * BITS_PER_WORD; + + if (mode == BLKmode) + { + HOST_WIDE_INT size = int_size_in_bytes (type); + if (size > 4) + return 2 * BITS_PER_WORD; + if (boundary < BITS_PER_WORD) + { + if (size >= 3) + return BITS_PER_WORD; + if (size >= 2) + return 2 * BITS_PER_UNIT; + } + } + return boundary; +} + +/* Implement TARGET_FUNCTION_ARG_ROUND_BOUNDARY. */ +static unsigned int +c6x_function_arg_round_boundary (enum machine_mode mode, const_tree type) +{ + return c6x_function_arg_boundary (mode, type); +} + +/* TARGET_FUNCTION_VALUE implementation. Returns an RTX representing the place + where function FUNC returns or receives a value of data type TYPE. */ + +static rtx +c6x_function_value (const_tree type, const_tree func ATTRIBUTE_UNUSED, + bool outgoing ATTRIBUTE_UNUSED) +{ + /* Functions return values in register A4. When returning aggregates, we may + have to adjust for endianness. */ + if (TARGET_BIG_ENDIAN && type && AGGREGATE_TYPE_P (type)) + { + HOST_WIDE_INT size = int_size_in_bytes (type); + if (size > 4) + { + + rtx reg1 = gen_rtx_REG (SImode, REG_A4 + 1); + rtx reg2 = gen_rtx_REG (SImode, REG_A4); + rtvec vec = gen_rtvec (2, gen_rtx_EXPR_LIST (VOIDmode, reg1, const0_rtx), + gen_rtx_EXPR_LIST (VOIDmode, reg2, GEN_INT (4))); + return gen_rtx_PARALLEL (TYPE_MODE (type), vec); + } + } + return gen_rtx_REG (TYPE_MODE (type), REG_A4); +} + +/* Implement TARGET_LIBCALL_VALUE. */ + +static rtx +c6x_libcall_value (enum machine_mode mode, const_rtx fun ATTRIBUTE_UNUSED) +{ + return gen_rtx_REG (mode, REG_A4); +} + +/* TARGET_STRUCT_VALUE_RTX implementation. */ + +static rtx +c6x_struct_value_rtx (tree type ATTRIBUTE_UNUSED, int incoming ATTRIBUTE_UNUSED) +{ + return gen_rtx_REG (Pmode, REG_A3); +} + +/* Implement TARGET_FUNCTION_VALUE_REGNO_P. */ + +static bool +c6x_function_value_regno_p (const unsigned int regno) +{ + return regno == REG_A4; +} + +/* Types larger than 64 bit, and variable sized types, are passed by + reference. The callee must copy them; see c6x_callee_copies. */ + +static bool +c6x_pass_by_reference (cumulative_args_t cum_v ATTRIBUTE_UNUSED, + enum machine_mode mode, const_tree type, + bool named ATTRIBUTE_UNUSED) +{ + int size = -1; + if (type) + size = int_size_in_bytes (type); + else if (mode != VOIDmode) + size = GET_MODE_SIZE (mode); + return size > 2 * UNITS_PER_WORD || size == -1; +} + +/* Decide whether a type should be returned in memory (true) + or in a register (false). This is called by the macro + TARGET_RETURN_IN_MEMORY. */ + +static bool +c6x_return_in_memory (const_tree type, const_tree fntype ATTRIBUTE_UNUSED) +{ + int size = int_size_in_bytes (type); + return size > 2 * UNITS_PER_WORD || size == -1; +} + +/* Values which must be returned in the most-significant end of the return + register. */ + +static bool +c6x_return_in_msb (const_tree valtype) +{ + HOST_WIDE_INT size = int_size_in_bytes (valtype); + return TARGET_BIG_ENDIAN && AGGREGATE_TYPE_P (valtype) && size == 3; +} + +/* Implement TARGET_CALLEE_COPIES. */ + +static bool +c6x_callee_copies (cumulative_args_t cum_v ATTRIBUTE_UNUSED, + enum machine_mode mode ATTRIBUTE_UNUSED, + const_tree type ATTRIBUTE_UNUSED, + bool named ATTRIBUTE_UNUSED) +{ + return true; +} + +/* Return the type to use as __builtin_va_list. */ +static tree +c6x_build_builtin_va_list (void) +{ + return build_pointer_type (char_type_node); +} + +static void +c6x_asm_trampoline_template (FILE *f) +{ + fprintf (f, "\t.long\t0x0000002b\n"); /* mvkl .s2 fnlow,B0 */ + fprintf (f, "\t.long\t0x01000028\n"); /* || mvkl .s1 sclow,A2 */ + fprintf (f, "\t.long\t0x0000006b\n"); /* mvkh .s2 fnhigh,B0 */ + fprintf (f, "\t.long\t0x01000068\n"); /* || mvkh .s1 schigh,A2 */ + fprintf (f, "\t.long\t0x00000362\n"); /* b .s2 B0 */ + fprintf (f, "\t.long\t0x00008000\n"); /* nop 5 */ + fprintf (f, "\t.long\t0x00000000\n"); /* nop */ + fprintf (f, "\t.long\t0x00000000\n"); /* nop */ +} + +/* Emit RTL insns to initialize the variable parts of a trampoline at + TRAMP. FNADDR is an RTX for the address of the function's pure + code. CXT is an RTX for the static chain value for the function. */ + +static void +c6x_initialize_trampoline (rtx tramp, tree fndecl, rtx cxt) +{ + rtx fnaddr = XEXP (DECL_RTL (fndecl), 0); + rtx t1 = copy_to_reg (fnaddr); + rtx t2 = copy_to_reg (cxt); + rtx mask = gen_reg_rtx (SImode); + int i; + + emit_block_move (tramp, assemble_trampoline_template (), + GEN_INT (TRAMPOLINE_SIZE), BLOCK_OP_NORMAL); + + emit_move_insn (mask, GEN_INT (0xffff << 7)); + + for (i = 0; i < 4; i++) + { + rtx mem = adjust_address (tramp, SImode, i * 4); + rtx t = (i & 1) ? t2 : t1; + rtx v1 = gen_reg_rtx (SImode); + rtx v2 = gen_reg_rtx (SImode); + emit_move_insn (v1, mem); + if (i < 2) + emit_insn (gen_ashlsi3 (v2, t, GEN_INT (7))); + else + emit_insn (gen_lshrsi3 (v2, t, GEN_INT (9))); + emit_insn (gen_andsi3 (v2, v2, mask)); + emit_insn (gen_iorsi3 (v2, v2, v1)); + emit_move_insn (mem, v2); + } +#ifdef CLEAR_INSN_CACHE + tramp = XEXP (tramp, 0); + emit_library_call (gen_rtx_SYMBOL_REF (Pmode, "__gnu_clear_cache"), + LCT_NORMAL, VOIDmode, 2, tramp, Pmode, + plus_constant (tramp, TRAMPOLINE_SIZE), Pmode); +#endif +} + +/* Determine whether c6x_output_mi_thunk can succeed. */ + +static bool +c6x_can_output_mi_thunk (const_tree thunk ATTRIBUTE_UNUSED, + HOST_WIDE_INT delta ATTRIBUTE_UNUSED, + HOST_WIDE_INT vcall_offset ATTRIBUTE_UNUSED, + const_tree function ATTRIBUTE_UNUSED) +{ + return !TARGET_LONG_CALLS; +} + +/* Output the assembler code for a thunk function. THUNK is the + declaration for the thunk function itself, FUNCTION is the decl for + the target function. DELTA is an immediate constant offset to be + added to THIS. If VCALL_OFFSET is nonzero, the word at + *(*this + vcall_offset) should be added to THIS. */ + +static void +c6x_output_mi_thunk (FILE *file ATTRIBUTE_UNUSED, + tree thunk ATTRIBUTE_UNUSED, HOST_WIDE_INT delta, + HOST_WIDE_INT vcall_offset, tree function) +{ + rtx xops[5]; + /* The this parameter is passed as the first argument. */ + rtx this_rtx = gen_rtx_REG (Pmode, REG_A4); + + c6x_current_insn = NULL_RTX; + + xops[4] = XEXP (DECL_RTL (function), 0); + if (!vcall_offset) + { + output_asm_insn ("b .s2 \t%4", xops); + if (!delta) + output_asm_insn ("nop 5", xops); + } + + /* Adjust the this parameter by a fixed constant. */ + if (delta) + { + xops[0] = GEN_INT (delta); + xops[1] = this_rtx; + if (delta >= -16 && delta <= 15) + { + output_asm_insn ("add .s1 %0, %1, %1", xops); + if (!vcall_offset) + output_asm_insn ("nop 4", xops); + } + else if (delta >= 16 && delta < 32) + { + output_asm_insn ("add .d1 %0, %1, %1", xops); + if (!vcall_offset) + output_asm_insn ("nop 4", xops); + } + else if (delta >= -32768 && delta < 32768) + { + output_asm_insn ("mvk .s1 %0, A0", xops); + output_asm_insn ("add .d1 %1, A0, %1", xops); + if (!vcall_offset) + output_asm_insn ("nop 3", xops); + } + else + { + output_asm_insn ("mvkl .s1 %0, A0", xops); + output_asm_insn ("mvkh .s1 %0, A0", xops); + output_asm_insn ("add .d1 %1, A0, %1", xops); + if (!vcall_offset) + output_asm_insn ("nop 3", xops); + } + } + + /* Adjust the this parameter by a value stored in the vtable. */ + if (vcall_offset) + { + rtx a0tmp = gen_rtx_REG (Pmode, REG_A0); + rtx a3tmp = gen_rtx_REG (Pmode, REG_A3); + + xops[1] = a3tmp; + xops[2] = a0tmp; + xops[3] = gen_rtx_MEM (Pmode, a0tmp); + output_asm_insn ("mv .s1 a4, %2", xops); + output_asm_insn ("ldw .d1t1 %3, %2", xops); + + /* Adjust the this parameter. */ + xops[0] = gen_rtx_MEM (Pmode, plus_constant (a0tmp, vcall_offset)); + if (!memory_operand (xops[0], Pmode)) + { + rtx tmp2 = gen_rtx_REG (Pmode, REG_A1); + xops[0] = GEN_INT (vcall_offset); + xops[1] = tmp2; + output_asm_insn ("mvkl .s1 %0, %1", xops); + output_asm_insn ("mvkh .s1 %0, %1", xops); + output_asm_insn ("nop 2", xops); + output_asm_insn ("add .d1 %2, %1, %2", xops); + xops[0] = gen_rtx_MEM (Pmode, a0tmp); + } + else + output_asm_insn ("nop 4", xops); + xops[2] = this_rtx; + output_asm_insn ("ldw .d1t1 %0, %1", xops); + output_asm_insn ("|| b .s2 \t%4", xops); + output_asm_insn ("nop 4", xops); + output_asm_insn ("add .d1 %2, %1, %2", xops); + } +} + +/* Return true if EXP goes in small data/bss. */ + +static bool +c6x_in_small_data_p (const_tree exp) +{ + /* We want to merge strings, so we never consider them small data. */ + if (TREE_CODE (exp) == STRING_CST) + return false; + + /* Functions are never small data. */ + if (TREE_CODE (exp) == FUNCTION_DECL) + return false; + + if (TREE_CODE (exp) == VAR_DECL && DECL_WEAK (exp)) + return false; + + if (TREE_CODE (exp) == VAR_DECL && DECL_SECTION_NAME (exp)) + { + const char *section = TREE_STRING_POINTER (DECL_SECTION_NAME (exp)); + + if (strcmp (section, ".neardata") == 0 + || strncmp (section, ".neardata.", 10) == 0 + || strncmp (section, ".gnu.linkonce.s.", 16) == 0 + || strcmp (section, ".bss") == 0 + || strncmp (section, ".bss.", 5) == 0 + || strncmp (section, ".gnu.linkonce.sb.", 17) == 0 + || strcmp (section, ".rodata") == 0 + || strncmp (section, ".rodata.", 8) == 0 + || strncmp (section, ".gnu.linkonce.s2.", 17) == 0) + return true; + } + else + return PLACE_IN_SDATA_P (exp); + + return false; +} + +/* Return a section for X. The only special thing we do here is to + honor small data. We don't have a tree type, so we can't use the + PLACE_IN_SDATA_P macro we use everywhere else; we choose to place + everything sized 8 bytes or smaller into small data. */ + +static section * +c6x_select_rtx_section (enum machine_mode mode, rtx x, + unsigned HOST_WIDE_INT align) +{ + if (c6x_sdata_mode == C6X_SDATA_ALL + || (c6x_sdata_mode != C6X_SDATA_NONE && GET_MODE_SIZE (mode) <= 8)) + /* ??? Consider using mergeable sdata sections. */ + return sdata_section; + else + return default_elf_select_rtx_section (mode, x, align); +} + +static section * +c6x_elf_select_section (tree decl, int reloc, + unsigned HOST_WIDE_INT align) +{ + const char *sname = NULL; + unsigned int flags = SECTION_WRITE; + if (c6x_in_small_data_p (decl)) + { + switch (categorize_decl_for_section (decl, reloc)) + { + case SECCAT_SRODATA: + sname = ".rodata"; + flags = 0; + break; + case SECCAT_SDATA: + sname = ".neardata"; + break; + case SECCAT_SBSS: + sname = ".bss"; + flags |= SECTION_BSS; + default: + break; + } + } + else + { + switch (categorize_decl_for_section (decl, reloc)) + { + case SECCAT_DATA: + sname = ".fardata"; + break; + case SECCAT_DATA_REL: + sname = ".fardata.rel"; + break; + case SECCAT_DATA_REL_LOCAL: + sname = ".fardata.rel.local"; + break; + case SECCAT_DATA_REL_RO: + sname = ".fardata.rel.ro"; + break; + case SECCAT_DATA_REL_RO_LOCAL: + sname = ".fardata.rel.ro.local"; + break; + case SECCAT_BSS: + sname = ".far"; + flags |= SECTION_BSS; + break; + case SECCAT_RODATA: + sname = ".const"; + flags = 0; + break; + case SECCAT_SRODATA: + case SECCAT_SDATA: + case SECCAT_SBSS: + gcc_unreachable (); + default: + break; + } + } + if (sname) + { + /* We might get called with string constants, but get_named_section + doesn't like them as they are not DECLs. Also, we need to set + flags in that case. */ + if (!DECL_P (decl)) + return get_section (sname, flags, NULL); + return get_named_section (decl, sname, reloc); + } + + return default_elf_select_section (decl, reloc, align); +} + +/* Build up a unique section name, expressed as a + STRING_CST node, and assign it to DECL_SECTION_NAME (decl). + RELOC indicates whether the initial value of EXP requires + link-time relocations. */ + +static void ATTRIBUTE_UNUSED +c6x_elf_unique_section (tree decl, int reloc) +{ + const char *prefix = NULL; + /* We only need to use .gnu.linkonce if we don't have COMDAT groups. */ + bool one_only = DECL_ONE_ONLY (decl) && !HAVE_COMDAT_GROUP; + + if (c6x_in_small_data_p (decl)) + { + switch (categorize_decl_for_section (decl, reloc)) + { + case SECCAT_SDATA: + prefix = one_only ? ".s" : ".neardata"; + break; + case SECCAT_SBSS: + prefix = one_only ? ".sb" : ".bss"; + break; + case SECCAT_SRODATA: + prefix = one_only ? ".s2" : ".rodata"; + break; + case SECCAT_RODATA_MERGE_STR: + case SECCAT_RODATA_MERGE_STR_INIT: + case SECCAT_RODATA_MERGE_CONST: + case SECCAT_RODATA: + case SECCAT_DATA: + case SECCAT_DATA_REL: + case SECCAT_DATA_REL_LOCAL: + case SECCAT_DATA_REL_RO: + case SECCAT_DATA_REL_RO_LOCAL: + gcc_unreachable (); + default: + /* Everything else we place into default sections and hope for the + best. */ + break; + } + } + else + { + switch (categorize_decl_for_section (decl, reloc)) + { + case SECCAT_DATA: + case SECCAT_DATA_REL: + case SECCAT_DATA_REL_LOCAL: + case SECCAT_DATA_REL_RO: + case SECCAT_DATA_REL_RO_LOCAL: + prefix = one_only ? ".fd" : ".fardata"; + break; + case SECCAT_BSS: + prefix = one_only ? ".fb" : ".far"; + break; + case SECCAT_RODATA: + case SECCAT_RODATA_MERGE_STR: + case SECCAT_RODATA_MERGE_STR_INIT: + case SECCAT_RODATA_MERGE_CONST: + prefix = one_only ? ".fr" : ".const"; + break; + case SECCAT_SRODATA: + case SECCAT_SDATA: + case SECCAT_SBSS: + gcc_unreachable (); + default: + break; + } + } + + if (prefix) + { + const char *name, *linkonce; + char *string; + + name = IDENTIFIER_POINTER (DECL_ASSEMBLER_NAME (decl)); + name = targetm.strip_name_encoding (name); + + /* If we're using one_only, then there needs to be a .gnu.linkonce + prefix to the section name. */ + linkonce = one_only ? ".gnu.linkonce" : ""; + + string = ACONCAT ((linkonce, prefix, ".", name, NULL)); + + DECL_SECTION_NAME (decl) = build_string (strlen (string), string); + return; + } + default_unique_section (decl, reloc); +} + +static unsigned int +c6x_section_type_flags (tree decl, const char *name, int reloc) +{ + unsigned int flags = 0; + + if (strcmp (name, ".far") == 0 + || strncmp (name, ".far.", 5) == 0) + flags |= SECTION_BSS; + + flags |= default_section_type_flags (decl, name, reloc); + + return flags; +} + +/* Checks whether the given CALL_EXPR would use a caller saved + register. This is used to decide whether sibling call optimization + could be performed on the respective function call. */ + +static bool +c6x_call_saved_register_used (tree call_expr) +{ + CUMULATIVE_ARGS cum_v; + cumulative_args_t cum; + HARD_REG_SET call_saved_regset; + tree parameter; + enum machine_mode mode; + tree type; + rtx parm_rtx; + int i; + + INIT_CUMULATIVE_ARGS (cum_v, NULL, NULL, 0, 0); + cum = pack_cumulative_args (&cum_v); + + COMPL_HARD_REG_SET (call_saved_regset, call_used_reg_set); + for (i = 0; i < call_expr_nargs (call_expr); i++) + { + parameter = CALL_EXPR_ARG (call_expr, i); + gcc_assert (parameter); + + /* For an undeclared variable passed as parameter we will get + an ERROR_MARK node here. */ + if (TREE_CODE (parameter) == ERROR_MARK) + return true; + + type = TREE_TYPE (parameter); + gcc_assert (type); + + mode = TYPE_MODE (type); + gcc_assert (mode); + + if (pass_by_reference (&cum_v, mode, type, true)) + { + mode = Pmode; + type = build_pointer_type (type); + } + + parm_rtx = c6x_function_arg (cum, mode, type, 0); + + c6x_function_arg_advance (cum, mode, type, 0); + + if (!parm_rtx) + continue; + + if (REG_P (parm_rtx) + && overlaps_hard_reg_set_p (call_saved_regset, GET_MODE (parm_rtx), + REGNO (parm_rtx))) + return true; + if (GET_CODE (parm_rtx) == PARALLEL) + { + int n = XVECLEN (parm_rtx, 0); + while (n-- > 0) + { + rtx x = XEXP (XVECEXP (parm_rtx, 0, n), 0); + if (REG_P (x) + && overlaps_hard_reg_set_p (call_saved_regset, + GET_MODE (x), REGNO (x))) + return true; + } + } + } + return false; +} + +/* Decide whether we can make a sibling call to a function. DECL is the + declaration of the function being targeted by the call and EXP is the + CALL_EXPR representing the call. */ + +static bool +c6x_function_ok_for_sibcall (tree decl, tree exp) +{ + /* Registers A10, A12, B10 and B12 are available as arguments + register but unfortunately caller saved. This makes functions + needing these registers for arguments not suitable for + sibcalls. */ + if (c6x_call_saved_register_used (exp)) + return false; + + if (!flag_pic) + return true; + + if (TARGET_DSBT) + { + /* When compiling for DSBT, the calling function must be local, + so that when we reload B14 in the sibcall epilogue, it will + not change its value. */ + struct cgraph_local_info *this_func; + + if (!decl) + /* Not enough information. */ + return false; + + this_func = cgraph_local_info (current_function_decl); + return this_func->local; + } + + return true; +} + +/* Return true if DECL is known to be linked into section SECTION. */ + +static bool +c6x_function_in_section_p (tree decl, section *section) +{ + /* We can only be certain about functions defined in the same + compilation unit. */ + if (!TREE_STATIC (decl)) + return false; + + /* Make sure that SYMBOL always binds to the definition in this + compilation unit. */ + if (!targetm.binds_local_p (decl)) + return false; + + /* If DECL_SECTION_NAME is set, assume it is trustworthy. */ + if (!DECL_SECTION_NAME (decl)) + { + /* Make sure that we will not create a unique section for DECL. */ + if (flag_function_sections || DECL_ONE_ONLY (decl)) + return false; + } + + return function_section (decl) == section; +} + +/* Return true if a call to OP, which is a SYMBOL_REF, must be expanded + as a long call. */ +bool +c6x_long_call_p (rtx op) +{ + tree decl; + + if (!TARGET_LONG_CALLS) + return false; + + decl = SYMBOL_REF_DECL (op); + + /* Try to determine whether the symbol is in the same section as the current + function. Be conservative, and only cater for cases in which the + whole of the current function is placed in the same section. */ + if (decl != NULL_TREE + && !flag_reorder_blocks_and_partition + && TREE_CODE (decl) == FUNCTION_DECL + && c6x_function_in_section_p (decl, current_function_section ())) + return false; + + return true; +} + +/* Emit the sequence for a call. */ +void +c6x_expand_call (rtx retval, rtx address, bool sibcall) +{ + rtx callee = XEXP (address, 0); + rtx call_insn; + + if (!c6x_call_operand (callee, Pmode)) + { + callee = force_reg (Pmode, callee); + address = change_address (address, Pmode, callee); + } + call_insn = gen_rtx_CALL (VOIDmode, address, const0_rtx); + if (sibcall) + { + call_insn = emit_call_insn (call_insn); + use_reg (&CALL_INSN_FUNCTION_USAGE (call_insn), + gen_rtx_REG (Pmode, REG_B3)); + } + else + { + if (retval == NULL_RTX) + call_insn = emit_call_insn (call_insn); + else + call_insn = emit_call_insn (gen_rtx_SET (GET_MODE (retval), retval, + call_insn)); + } + if (flag_pic) + use_reg (&CALL_INSN_FUNCTION_USAGE (call_insn), pic_offset_table_rtx); +} + +/* Legitimize PIC addresses. If the address is already position-independent, + we return ORIG. Newly generated position-independent addresses go into a + reg. This is REG if nonzero, otherwise we allocate register(s) as + necessary. PICREG is the register holding the pointer to the PIC offset + table. */ + +static rtx +legitimize_pic_address (rtx orig, rtx reg, rtx picreg) +{ + rtx addr = orig; + rtx new_rtx = orig; + + if (GET_CODE (addr) == SYMBOL_REF || GET_CODE (addr) == LABEL_REF) + { + int unspec = UNSPEC_LOAD_GOT; + rtx tmp; + + if (reg == 0) + { + gcc_assert (can_create_pseudo_p ()); + reg = gen_reg_rtx (Pmode); + } + if (flag_pic == 2) + { + if (can_create_pseudo_p ()) + tmp = gen_reg_rtx (Pmode); + else + tmp = reg; + emit_insn (gen_movsi_gotoff_high (tmp, addr)); + emit_insn (gen_movsi_gotoff_lo_sum (tmp, tmp, addr)); + emit_insn (gen_load_got_gotoff (reg, picreg, tmp)); + } + else + { + tmp = gen_rtx_UNSPEC (Pmode, gen_rtvec (1, addr), unspec); + new_rtx = gen_const_mem (Pmode, gen_rtx_PLUS (Pmode, picreg, tmp)); + + emit_move_insn (reg, new_rtx); + } + if (picreg == pic_offset_table_rtx) + crtl->uses_pic_offset_table = 1; + return reg; + } + + else if (GET_CODE (addr) == CONST || GET_CODE (addr) == PLUS) + { + rtx base; + + if (GET_CODE (addr) == CONST) + { + addr = XEXP (addr, 0); + gcc_assert (GET_CODE (addr) == PLUS); + } + + if (XEXP (addr, 0) == picreg) + return orig; + + if (reg == 0) + { + gcc_assert (can_create_pseudo_p ()); + reg = gen_reg_rtx (Pmode); + } + + base = legitimize_pic_address (XEXP (addr, 0), reg, picreg); + addr = legitimize_pic_address (XEXP (addr, 1), + base == reg ? NULL_RTX : reg, + picreg); + + if (GET_CODE (addr) == CONST_INT) + { + gcc_assert (! reload_in_progress && ! reload_completed); + addr = force_reg (Pmode, addr); + } + + if (GET_CODE (addr) == PLUS && CONSTANT_P (XEXP (addr, 1))) + { + base = gen_rtx_PLUS (Pmode, base, XEXP (addr, 0)); + addr = XEXP (addr, 1); + } + + return gen_rtx_PLUS (Pmode, base, addr); + } + + return new_rtx; +} + +/* Expand a move operation in mode MODE. The operands are in OPERANDS. + Returns true if no further code must be generated, false if the caller + should generate an insn to move OPERANDS[1] to OPERANDS[0]. */ + +bool +expand_move (rtx *operands, enum machine_mode mode) +{ + rtx dest = operands[0]; + rtx op = operands[1]; + + if ((reload_in_progress | reload_completed) == 0 + && GET_CODE (dest) == MEM && GET_CODE (op) != REG) + operands[1] = force_reg (mode, op); + else if (mode == SImode && symbolic_operand (op, SImode)) + { + if (flag_pic) + { + if (sdata_symbolic_operand (op, SImode)) + { + emit_insn (gen_load_sdata_pic (dest, pic_offset_table_rtx, op)); + crtl->uses_pic_offset_table = 1; + return true; + } + else + { + rtx temp = (reload_completed || reload_in_progress + ? dest : gen_reg_rtx (Pmode)); + + operands[1] = legitimize_pic_address (op, temp, + pic_offset_table_rtx); + } + } + else if (reload_completed + && !sdata_symbolic_operand (op, SImode)) + { + emit_insn (gen_movsi_high (dest, op)); + emit_insn (gen_movsi_lo_sum (dest, dest, op)); + return true; + } + } + return false; +} + +/* This function is called when we're about to expand an integer compare + operation which performs COMPARISON. It examines the second operand, + and if it is an integer constant that cannot be used directly on the + current machine in a comparison insn, it returns true. */ +bool +c6x_force_op_for_comparison_p (enum rtx_code code, rtx op) +{ + if (!CONST_INT_P (op) || satisfies_constraint_Iu4 (op)) + return false; + + if ((code == EQ || code == LT || code == GT) + && !satisfies_constraint_Is5 (op)) + return true; + if ((code == GTU || code == LTU) + && (!TARGET_INSNS_64 || !satisfies_constraint_Iu5 (op))) + return true; + + return false; +} + +/* Emit comparison instruction if necessary, returning the expression + that holds the compare result in the proper mode. Return the comparison + that should be used in the jump insn. */ + +rtx +c6x_expand_compare (rtx comparison, enum machine_mode mode) +{ + enum rtx_code code = GET_CODE (comparison); + rtx op0 = XEXP (comparison, 0); + rtx op1 = XEXP (comparison, 1); + rtx cmp; + enum rtx_code jump_code = code; + enum machine_mode op_mode = GET_MODE (op0); + + if (op_mode == DImode && (code == NE || code == EQ) && op1 == const0_rtx) + { + rtx t = gen_reg_rtx (SImode); + emit_insn (gen_iorsi3 (t, gen_lowpart (SImode, op0), + gen_highpart (SImode, op0))); + op_mode = SImode; + cmp = t; + } + else if (op_mode == DImode) + { + rtx lo[2], high[2]; + rtx cmp1, cmp2; + + if (code == NE || code == GEU || code == LEU || code == GE || code == LE) + { + code = reverse_condition (code); + jump_code = EQ; + } + else + jump_code = NE; + + split_di (&op0, 1, lo, high); + split_di (&op1, 1, lo + 1, high + 1); + + if (c6x_force_op_for_comparison_p (code, high[1]) + || c6x_force_op_for_comparison_p (EQ, high[1])) + high[1] = force_reg (SImode, high[1]); + + cmp1 = gen_reg_rtx (SImode); + cmp2 = gen_reg_rtx (SImode); + emit_insn (gen_rtx_SET (VOIDmode, cmp1, + gen_rtx_fmt_ee (code, SImode, high[0], high[1]))); + if (code == EQ) + { + if (c6x_force_op_for_comparison_p (code, lo[1])) + lo[1] = force_reg (SImode, lo[1]); + emit_insn (gen_rtx_SET (VOIDmode, cmp2, + gen_rtx_fmt_ee (code, SImode, lo[0], lo[1]))); + emit_insn (gen_andsi3 (cmp1, cmp1, cmp2)); + } + else + { + emit_insn (gen_rtx_SET (VOIDmode, cmp2, + gen_rtx_EQ (SImode, high[0], high[1]))); + if (code == GT) + code = GTU; + else if (code == LT) + code = LTU; + if (c6x_force_op_for_comparison_p (code, lo[1])) + lo[1] = force_reg (SImode, lo[1]); + emit_insn (gen_cmpsi_and (cmp2, gen_rtx_fmt_ee (code, SImode, + lo[0], lo[1]), + lo[0], lo[1], cmp2)); + emit_insn (gen_iorsi3 (cmp1, cmp1, cmp2)); + } + cmp = cmp1; + } + else if (TARGET_FP && !flag_finite_math_only + && (op_mode == DFmode || op_mode == SFmode) + && code != EQ && code != NE && code != LT && code != GT + && code != UNLE && code != UNGE) + { + enum rtx_code code1, code2, code3; + rtx (*fn) (rtx, rtx, rtx, rtx, rtx); + + jump_code = NE; + code3 = UNKNOWN; + switch (code) + { + case UNLT: + case UNGT: + jump_code = EQ; + /* fall through */ + case LE: + case GE: + code1 = code == LE || code == UNGT ? LT : GT; + code2 = EQ; + break; + + case UNORDERED: + jump_code = EQ; + /* fall through */ + case ORDERED: + code3 = EQ; + /* fall through */ + case LTGT: + code1 = LT; + code2 = GT; + break; + + case UNEQ: + code1 = LT; + code2 = GT; + jump_code = EQ; + break; + + default: + gcc_unreachable (); + } + + cmp = gen_reg_rtx (SImode); + emit_insn (gen_rtx_SET (VOIDmode, cmp, + gen_rtx_fmt_ee (code1, SImode, op0, op1))); + fn = op_mode == DFmode ? gen_cmpdf_ior : gen_cmpsf_ior; + emit_insn (fn (cmp, gen_rtx_fmt_ee (code2, SImode, op0, op1), + op0, op1, cmp)); + if (code3 != UNKNOWN) + emit_insn (fn (cmp, gen_rtx_fmt_ee (code3, SImode, op0, op1), + op0, op1, cmp)); + } + else if (op_mode == SImode && (code == NE || code == EQ) && op1 == const0_rtx) + cmp = op0; + else + { + bool is_fp_libfunc; + is_fp_libfunc = !TARGET_FP && (op_mode == DFmode || op_mode == SFmode); + + if ((code == NE || code == GEU || code == LEU || code == GE || code == LE) + && !is_fp_libfunc) + { + code = reverse_condition (code); + jump_code = EQ; + } + else if (code == UNGE) + { + code = LT; + jump_code = EQ; + } + else if (code == UNLE) + { + code = GT; + jump_code = EQ; + } + else + jump_code = NE; + + if (is_fp_libfunc) + { + rtx insns; + rtx libfunc; + switch (code) + { + case EQ: + libfunc = op_mode == DFmode ? eqdf_libfunc : eqsf_libfunc; + break; + case NE: + libfunc = op_mode == DFmode ? nedf_libfunc : nesf_libfunc; + break; + case GT: + libfunc = op_mode == DFmode ? gtdf_libfunc : gtsf_libfunc; + break; + case GE: + libfunc = op_mode == DFmode ? gedf_libfunc : gesf_libfunc; + break; + case LT: + libfunc = op_mode == DFmode ? ltdf_libfunc : ltsf_libfunc; + break; + case LE: + libfunc = op_mode == DFmode ? ledf_libfunc : lesf_libfunc; + break; + default: + gcc_unreachable (); + } + start_sequence (); + + cmp = emit_library_call_value (libfunc, 0, LCT_CONST, SImode, 2, + op0, op_mode, op1, op_mode); + insns = get_insns (); + end_sequence (); + + emit_libcall_block (insns, cmp, cmp, + gen_rtx_fmt_ee (code, SImode, op0, op1)); + } + else + { + cmp = gen_reg_rtx (SImode); + if (c6x_force_op_for_comparison_p (code, op1)) + op1 = force_reg (SImode, op1); + emit_insn (gen_rtx_SET (VOIDmode, cmp, + gen_rtx_fmt_ee (code, SImode, op0, op1))); + } + } + + return gen_rtx_fmt_ee (jump_code, mode, cmp, const0_rtx); +} + +/* Return one word of double-word value OP. HIGH_P is true to select the + high part, false to select the low part. When encountering auto-increment + addressing, we make the assumption that the low part is going to be accessed + first. */ + +rtx +c6x_subword (rtx op, bool high_p) +{ + unsigned int byte; + enum machine_mode mode; + + mode = GET_MODE (op); + if (mode == VOIDmode) + mode = DImode; + + if (TARGET_BIG_ENDIAN ? !high_p : high_p) + byte = UNITS_PER_WORD; + else + byte = 0; + + if (MEM_P (op)) + { + rtx addr = XEXP (op, 0); + if (GET_CODE (addr) == PLUS || REG_P (addr)) + return adjust_address (op, word_mode, byte); + /* FIXME: should really support autoincrement addressing for + multi-word modes. */ + gcc_unreachable (); + } + + return simplify_gen_subreg (word_mode, op, mode, byte); +} + +/* Split one or more DImode RTL references into pairs of SImode + references. The RTL can be REG, offsettable MEM, integer constant, or + CONST_DOUBLE. "operands" is a pointer to an array of DImode RTL to + split and "num" is its length. lo_half and hi_half are output arrays + that parallel "operands". */ + +void +split_di (rtx operands[], int num, rtx lo_half[], rtx hi_half[]) +{ + while (num--) + { + rtx op = operands[num]; + + lo_half[num] = c6x_subword (op, false); + hi_half[num] = c6x_subword (op, true); + } +} + +/* Return true if VAL is a mask valid for a clr instruction. */ +bool +c6x_valid_mask_p (HOST_WIDE_INT val) +{ + int i; + for (i = 0; i < 32; i++) + if (!(val & ((unsigned HOST_WIDE_INT)1 << i))) + break; + for (; i < 32; i++) + if (val & ((unsigned HOST_WIDE_INT)1 << i)) + break; + for (; i < 32; i++) + if (!(val & ((unsigned HOST_WIDE_INT)1 << i))) + return false; + return true; +} + +/* Expand a block move for a movmemM pattern. */ + +bool +c6x_expand_movmem (rtx dst, rtx src, rtx count_exp, rtx align_exp, + rtx expected_align_exp ATTRIBUTE_UNUSED, + rtx expected_size_exp ATTRIBUTE_UNUSED) +{ + unsigned HOST_WIDE_INT align = 1; + unsigned HOST_WIDE_INT src_mem_align, dst_mem_align, min_mem_align; + unsigned HOST_WIDE_INT count = 0, offset = 0; + unsigned int biggest_move = TARGET_STDW ? 8 : 4; + + if (CONST_INT_P (align_exp)) + align = INTVAL (align_exp); + + src_mem_align = MEM_ALIGN (src) / BITS_PER_UNIT; + dst_mem_align = MEM_ALIGN (dst) / BITS_PER_UNIT; + min_mem_align = MIN (src_mem_align, dst_mem_align); + + if (min_mem_align > align) + align = min_mem_align / BITS_PER_UNIT; + if (src_mem_align < align) + src_mem_align = align; + if (dst_mem_align < align) + dst_mem_align = align; + + if (CONST_INT_P (count_exp)) + count = INTVAL (count_exp); + else + return false; + + /* Make sure we don't need to care about overflow later on. */ + if (count > ((unsigned HOST_WIDE_INT) 1 << 30)) + return false; + + if (count >= 28 && (count & 3) == 0 && align >= 4) + { + tree dst_expr = MEM_EXPR (dst); + tree src_expr = MEM_EXPR (src); + rtx fn = TARGET_INSNS_64PLUS ? strasgi64p_libfunc : strasgi_libfunc; + rtx srcreg = force_reg (Pmode, XEXP (src, 0)); + rtx dstreg = force_reg (Pmode, XEXP (dst, 0)); + + if (src_expr) + mark_addressable (src_expr); + if (dst_expr) + mark_addressable (dst_expr); + emit_library_call (fn, LCT_NORMAL, VOIDmode, 3, + dstreg, Pmode, srcreg, Pmode, count_exp, SImode); + return true; + } + + if (biggest_move > align && !TARGET_INSNS_64) + biggest_move = align; + + if (count / biggest_move > 7) + return false; + + while (count > 0) + { + rtx reg, reg_lowpart; + enum machine_mode srcmode, dstmode; + unsigned HOST_WIDE_INT src_size, dst_size, src_left; + int shift; + rtx srcmem, dstmem; + + while (biggest_move > count) + biggest_move /= 2; + + src_size = dst_size = biggest_move; + if (src_size > src_mem_align && src_size == 2) + src_size = 1; + if (dst_size > dst_mem_align && dst_size == 2) + dst_size = 1; + + if (dst_size > src_size) + dst_size = src_size; + + srcmode = mode_for_size (src_size * BITS_PER_UNIT, MODE_INT, 0); + dstmode = mode_for_size (dst_size * BITS_PER_UNIT, MODE_INT, 0); + if (src_size >= 4) + reg_lowpart = reg = gen_reg_rtx (srcmode); + else + { + reg = gen_reg_rtx (SImode); + reg_lowpart = gen_lowpart (srcmode, reg); + } + + srcmem = adjust_address (copy_rtx (src), srcmode, offset); + + if (src_size > src_mem_align) + { + enum insn_code icode = (srcmode == SImode ? CODE_FOR_movmisalignsi + : CODE_FOR_movmisaligndi); + emit_insn (GEN_FCN (icode) (reg_lowpart, srcmem)); + } + else + emit_move_insn (reg_lowpart, srcmem); + + src_left = src_size; + shift = TARGET_BIG_ENDIAN ? (src_size - dst_size) * BITS_PER_UNIT : 0; + while (src_left > 0) + { + rtx dstreg = reg_lowpart; + + if (src_size > dst_size) + { + rtx srcword = reg; + int shift_amount = shift & (BITS_PER_WORD - 1); + if (src_size > 4) + srcword = operand_subword_force (srcword, src_left >= 4 ? 0 : 4, + SImode); + if (shift_amount > 0) + { + dstreg = gen_reg_rtx (SImode); + emit_insn (gen_lshrsi3 (dstreg, srcword, + GEN_INT (shift_amount))); + } + else + dstreg = srcword; + dstreg = gen_lowpart (dstmode, dstreg); + } + + dstmem = adjust_address (copy_rtx (dst), dstmode, offset); + if (dst_size > dst_mem_align) + { + enum insn_code icode = (dstmode == SImode ? CODE_FOR_movmisalignsi + : CODE_FOR_movmisaligndi); + emit_insn (GEN_FCN (icode) (dstmem, dstreg)); + } + else + emit_move_insn (dstmem, dstreg); + + if (TARGET_BIG_ENDIAN) + shift -= dst_size * BITS_PER_UNIT; + else + shift += dst_size * BITS_PER_UNIT; + offset += dst_size; + src_left -= dst_size; + } + count -= src_size; + } + return true; +} + +/* Subroutine of print_address_operand, print a single address offset OFF for + a memory access of mode MEM_MODE, choosing between normal form and scaled + form depending on the type of the insn. Misaligned memory references must + use the scaled form. */ + +static void +print_address_offset (FILE *file, rtx off, enum machine_mode mem_mode) +{ + rtx pat; + + if (c6x_current_insn != NULL_RTX) + { + pat = PATTERN (c6x_current_insn); + if (GET_CODE (pat) == COND_EXEC) + pat = COND_EXEC_CODE (pat); + if (GET_CODE (pat) == PARALLEL) + pat = XVECEXP (pat, 0, 0); + + if (GET_CODE (pat) == SET + && GET_CODE (SET_SRC (pat)) == UNSPEC + && XINT (SET_SRC (pat), 1) == UNSPEC_MISALIGNED_ACCESS) + { + gcc_assert (CONST_INT_P (off) + && (INTVAL (off) & (GET_MODE_SIZE (mem_mode) - 1)) == 0); + fprintf (file, "[" HOST_WIDE_INT_PRINT_DEC "]", + INTVAL (off) / GET_MODE_SIZE (mem_mode)); + return; + } + } + fputs ("(", file); + output_address (off); + fputs (")", file); +} + +static bool +c6x_print_operand_punct_valid_p (unsigned char c) +{ + return c == '$' || c == '.' || c == '|'; +} + +static void c6x_print_operand (FILE *, rtx, int); + +/* Subroutine of c6x_print_operand; used to print a memory reference X to FILE. */ + +static void +c6x_print_address_operand (FILE *file, rtx x, enum machine_mode mem_mode) +{ + rtx off; + switch (GET_CODE (x)) + { + case PRE_MODIFY: + case POST_MODIFY: + if (GET_CODE (x) == POST_MODIFY) + output_address (XEXP (x, 0)); + off = XEXP (XEXP (x, 1), 1); + if (XEXP (x, 0) == stack_pointer_rtx) + { + if (GET_CODE (x) == PRE_MODIFY) + gcc_assert (INTVAL (off) > 0); + else + gcc_assert (INTVAL (off) < 0); + } + if (CONST_INT_P (off) && INTVAL (off) < 0) + { + fprintf (file, "--"); + off = GEN_INT (-INTVAL (off)); + } + else + fprintf (file, "++"); + if (GET_CODE (x) == PRE_MODIFY) + output_address (XEXP (x, 0)); + print_address_offset (file, off, mem_mode); + break; + + case PLUS: + off = XEXP (x, 1); + if (CONST_INT_P (off) && INTVAL (off) < 0) + { + fprintf (file, "-"); + off = GEN_INT (-INTVAL (off)); + } + else + fprintf (file, "+"); + output_address (XEXP (x, 0)); + print_address_offset (file, off, mem_mode); + break; + + case PRE_DEC: + gcc_assert (XEXP (x, 0) != stack_pointer_rtx); + fprintf (file, "--"); + output_address (XEXP (x, 0)); + fprintf (file, "[1]"); + break; + case PRE_INC: + fprintf (file, "++"); + output_address (XEXP (x, 0)); + fprintf (file, "[1]"); + break; + case POST_INC: + gcc_assert (XEXP (x, 0) != stack_pointer_rtx); + output_address (XEXP (x, 0)); + fprintf (file, "++[1]"); + break; + case POST_DEC: + output_address (XEXP (x, 0)); + fprintf (file, "--[1]"); + break; + + case SYMBOL_REF: + case CONST: + case LABEL_REF: + gcc_assert (sdata_symbolic_operand (x, Pmode)); + fprintf (file, "+B14("); + output_addr_const (file, x); + fprintf (file, ")"); + break; + + case UNSPEC: + switch (XINT (x, 1)) + { + case UNSPEC_LOAD_GOT: + fputs ("$GOT(", file); + output_addr_const (file, XVECEXP (x, 0, 0)); + fputs (")", file); + break; + case UNSPEC_LOAD_SDATA: + output_addr_const (file, XVECEXP (x, 0, 0)); + break; + default: + gcc_unreachable (); + } + break; + + default: + gcc_assert (GET_CODE (x) != MEM); + c6x_print_operand (file, x, 0); + break; + } +} + +/* Return a single character, which is either 'l', 's', 'd' or 'm', which + specifies the functional unit used by INSN. */ + +char +c6x_get_unit_specifier (rtx insn) +{ + enum attr_units units; + + if (insn_info) + { + int unit = INSN_INFO_ENTRY (INSN_UID (insn)).reservation; + return c6x_unit_names[unit][0]; + } + + units = get_attr_units (insn); + switch (units) + { + case UNITS_D: + case UNITS_DL: + case UNITS_DS: + case UNITS_DLS: + case UNITS_D_ADDR: + return 'd'; + break; + case UNITS_L: + case UNITS_LS: + return 'l'; + break; + case UNITS_S: + return 's'; + break; + case UNITS_M: + return 'm'; + break; + default: + gcc_unreachable (); + } +} + +/* Prints the unit specifier field. */ +static void +c6x_print_unit_specifier_field (FILE *file, rtx insn) +{ + enum attr_units units = get_attr_units (insn); + enum attr_cross cross = get_attr_cross (insn); + enum attr_dest_regfile rf = get_attr_dest_regfile (insn); + int half; + char unitspec; + + if (units == UNITS_D_ADDR) + { + enum attr_addr_regfile arf = get_attr_addr_regfile (insn); + int t_half; + gcc_assert (arf != ADDR_REGFILE_UNKNOWN); + half = arf == ADDR_REGFILE_A ? 1 : 2; + t_half = rf == DEST_REGFILE_A ? 1 : 2; + fprintf (file, ".d%dt%d", half, t_half); + return; + } + + if (insn_info) + { + int unit = INSN_INFO_ENTRY (INSN_UID (insn)).reservation; + fputs (".", file); + fputs (c6x_unit_names[unit], file); + if (cross == CROSS_Y) + fputs ("x", file); + return; + } + + gcc_assert (rf != DEST_REGFILE_UNKNOWN); + unitspec = c6x_get_unit_specifier (insn); + half = rf == DEST_REGFILE_A ? 1 : 2; + fprintf (file, ".%c%d%s", unitspec, half, cross == CROSS_Y ? "x" : ""); +} + +/* Output assembly language output for the address ADDR to FILE. */ +static void +c6x_print_operand_address (FILE *file, rtx addr) +{ + c6x_print_address_operand (file, addr, VOIDmode); +} + +/* Print an operand, X, to FILE, with an optional modifier in CODE. + + Meaning of CODE: + $ -- print the unit specifier field for the instruction. + . -- print the predicate for the instruction or an emptry string for an + unconditional one. + | -- print "||" if the insn should be issued in parallel with the previous + one. + + C -- print an opcode suffix for a reversed condition + d -- H, W or D as a suffix for ADDA, based on the factor given by the + operand + D -- print either B, H, W or D as a suffix for ADDA, based on the size of + the operand + J -- print a predicate + j -- like J, but use reverse predicate + k -- treat a CONST_INT as a register number and print it as a register + k -- like k, but print out a doubleword register + n -- print an integer operand, negated + p -- print the low part of a DImode register + P -- print the high part of a DImode register + r -- print the absolute value of an integer operand, shifted right by 1 + R -- print the absolute value of an integer operand, shifted right by 2 + f -- the first clear bit in an integer operand assumed to be a mask for + a clr instruction + F -- the last clear bit in such a mask + s -- the first set bit in an integer operand assumed to be a mask for + a set instruction + S -- the last set bit in such a mask + U -- print either 1 or 2, depending on the side of the machine used by + the operand */ + +static void +c6x_print_operand (FILE *file, rtx x, int code) +{ + int i; + HOST_WIDE_INT v; + tree t; + enum machine_mode mode; + + if (code == '|') + { + if (GET_MODE (c6x_current_insn) != TImode) + fputs ("||", file); + return; + } + if (code == '$') + { + c6x_print_unit_specifier_field (file, c6x_current_insn); + return; + } + + if (code == '.') + { + x = current_insn_predicate; + if (x) + { + unsigned int regno = REGNO (XEXP (x, 0)); + fputs ("[", file); + if (GET_CODE (x) == EQ) + fputs ("!", file); + fputs (reg_names [regno], file); + fputs ("]", file); + } + return; + } + + mode = GET_MODE (x); + + switch (code) + { + case 'C': + case 'c': + { + enum rtx_code c = GET_CODE (x); + if (code == 'C') + c = swap_condition (c); + fputs (GET_RTX_NAME (c), file); + } + return; + + case 'J': + case 'j': + { + unsigned int regno = REGNO (XEXP (x, 0)); + if ((GET_CODE (x) == EQ) == (code == 'J')) + fputs ("!", file); + fputs (reg_names [regno], file); + } + return; + + case 'k': + gcc_assert (GET_CODE (x) == CONST_INT); + v = INTVAL (x); + fprintf (file, "%s", reg_names[v]); + return; + case 'K': + gcc_assert (GET_CODE (x) == CONST_INT); + v = INTVAL (x); + gcc_assert ((v & 1) == 0); + fprintf (file, "%s:%s", reg_names[v + 1], reg_names[v]); + return; + + case 's': + case 'S': + case 'f': + case 'F': + gcc_assert (GET_CODE (x) == CONST_INT); + v = INTVAL (x); + for (i = 0; i < 32; i++) + { + HOST_WIDE_INT tst = v & 1; + if (((code == 'f' || code == 'F') && !tst) + || ((code == 's' || code == 'S') && tst)) + break; + v >>= 1; + } + if (code == 'f' || code == 's') + { + fprintf (file, "%d", i); + return; + } + for (;i < 32; i++) + { + HOST_WIDE_INT tst = v & 1; + if ((code == 'F' && tst) || (code == 'S' && !tst)) + break; + v >>= 1; + } + fprintf (file, "%d", i - 1); + return; + + case 'n': + gcc_assert (GET_CODE (x) == CONST_INT); + output_addr_const (file, GEN_INT (-INTVAL (x))); + return; + + case 'r': + gcc_assert (GET_CODE (x) == CONST_INT); + v = INTVAL (x); + if (v < 0) + v = -v; + output_addr_const (file, GEN_INT (v >> 1)); + return; + + case 'R': + gcc_assert (GET_CODE (x) == CONST_INT); + v = INTVAL (x); + if (v < 0) + v = -v; + output_addr_const (file, GEN_INT (v >> 2)); + return; + + case 'd': + gcc_assert (GET_CODE (x) == CONST_INT); + v = INTVAL (x); + fputs (v == 2 ? "h" : v == 4 ? "w" : "d", file); + return; + + case 'p': + case 'P': + gcc_assert (GET_CODE (x) == REG); + v = REGNO (x); + if (code == 'P') + v++; + fputs (reg_names[v], file); + return; + + case 'D': + v = 0; + if (GET_CODE (x) == CONST) + { + x = XEXP (x, 0); + gcc_assert (GET_CODE (x) == PLUS); + gcc_assert (GET_CODE (XEXP (x, 1)) == CONST_INT); + v = INTVAL (XEXP (x, 1)); + x = XEXP (x, 0); + + } + gcc_assert (GET_CODE (x) == SYMBOL_REF); + + t = SYMBOL_REF_DECL (x); + if (DECL_P (t)) + v |= DECL_ALIGN_UNIT (t); + else + v |= TYPE_ALIGN_UNIT (TREE_TYPE (t)); + if (v & 1) + fputs ("b", file); + else if (v & 2) + fputs ("h", file); + else + fputs ("w", file); + return; + + case 'U': + if (MEM_P (x)) + { + x = XEXP (x, 0); + if (GET_CODE (x) == PLUS + || GET_RTX_CLASS (GET_CODE (x)) == RTX_AUTOINC) + x = XEXP (x, 0); + if (GET_CODE (x) == CONST || GET_CODE (x) == SYMBOL_REF) + { + gcc_assert (sdata_symbolic_operand (x, Pmode)); + fputs ("2", file); + return; + } + } + gcc_assert (REG_P (x)); + if (A_REGNO_P (REGNO (x))) + fputs ("1", file); + if (B_REGNO_P (REGNO (x))) + fputs ("2", file); + return; + + default: + switch (GET_CODE (x)) + { + case REG: + if (GET_MODE_SIZE (mode) == 8) + fprintf (file, "%s:%s", reg_names[REGNO (x) + 1], + reg_names[REGNO (x)]); + else + fprintf (file, "%s", reg_names[REGNO (x)]); + break; + + case MEM: + fputc ('*', file); + gcc_assert (XEXP (x, 0) != stack_pointer_rtx); + c6x_print_address_operand (file, XEXP (x, 0), GET_MODE (x)); + break; + + case SYMBOL_REF: + fputc ('(', file); + output_addr_const (file, x); + fputc (')', file); + break; + + case CONST_INT: + output_addr_const (file, x); + break; + + case CONST_DOUBLE: + output_operand_lossage ("invalid const_double operand"); + break; + + default: + output_addr_const (file, x); + } + } +} + +/* Return TRUE if OP is a valid memory address with a base register of + class C. If SMALL_OFFSET is true, we disallow memory references which would + require a long offset with B14/B15. */ + +bool +c6x_mem_operand (rtx op, enum reg_class c, bool small_offset) +{ + enum machine_mode mode = GET_MODE (op); + rtx base = XEXP (op, 0); + switch (GET_CODE (base)) + { + case REG: + break; + case PLUS: + if (small_offset + && (XEXP (base, 0) == stack_pointer_rtx + || XEXP (base, 0) == pic_offset_table_rtx)) + { + if (!c6x_legitimate_address_p_1 (mode, base, true, true)) + return false; + } + + /* fall through */ + case PRE_INC: + case PRE_DEC: + case PRE_MODIFY: + case POST_INC: + case POST_DEC: + case POST_MODIFY: + base = XEXP (base, 0); + break; + + case CONST: + case LABEL_REF: + case SYMBOL_REF: + gcc_assert (sdata_symbolic_operand (base, Pmode)); + return !small_offset && c == B_REGS; + + default: + return false; + } + return TEST_HARD_REG_BIT (reg_class_contents[ (int) (c)], REGNO (base)); +} + +/* Returns true if X is a valid address for use in a memory reference + of mode MODE. If STRICT is true, we do not allow pseudo registers + in the address. NO_LARGE_OFFSET is true if we are examining an + address for use in a load or store misaligned instruction, or + recursively examining an operand inside a PRE/POST_MODIFY. */ + +bool +c6x_legitimate_address_p_1 (enum machine_mode mode, rtx x, bool strict, + bool no_large_offset) +{ + int size, size1; + HOST_WIDE_INT off; + enum rtx_code code = GET_CODE (x); + + switch (code) + { + case PRE_MODIFY: + case POST_MODIFY: + /* We can't split these into word-sized pieces yet. */ + if (!TARGET_STDW && GET_MODE_SIZE (mode) > UNITS_PER_WORD) + return false; + if (GET_CODE (XEXP (x, 1)) != PLUS) + return false; + if (!c6x_legitimate_address_p_1 (mode, XEXP (x, 1), strict, true)) + return false; + if (!rtx_equal_p (XEXP (x, 0), XEXP (XEXP (x, 1), 0))) + return false; + + /* fall through */ + case PRE_INC: + case PRE_DEC: + case POST_INC: + case POST_DEC: + /* We can't split these into word-sized pieces yet. */ + if (!TARGET_STDW && GET_MODE_SIZE (mode) > UNITS_PER_WORD) + return false; + x = XEXP (x, 0); + if (!REG_P (x)) + return false; + + /* fall through */ + case REG: + if (strict) + return REGNO_OK_FOR_BASE_STRICT_P (REGNO (x)); + else + return REGNO_OK_FOR_BASE_NONSTRICT_P (REGNO (x)); + + case PLUS: + if (!REG_P (XEXP (x, 0)) + || !c6x_legitimate_address_p_1 (mode, XEXP (x, 0), strict, false)) + return false; + /* We cannot ensure currently that both registers end up in the + same register file. */ + if (REG_P (XEXP (x, 1))) + return false; + + if (mode == BLKmode) + size = 4; + else if (mode == VOIDmode) + /* ??? This can happen during ivopts. */ + size = 1; + else + size = GET_MODE_SIZE (mode); + + if (flag_pic + && GET_CODE (XEXP (x, 1)) == UNSPEC + && XINT (XEXP (x, 1), 1) == UNSPEC_LOAD_SDATA + && XEXP (x, 0) == pic_offset_table_rtx + && sdata_symbolic_operand (XVECEXP (XEXP (x, 1), 0, 0), SImode)) + return !no_large_offset && size <= 4; + if (flag_pic == 1 + && mode == Pmode + && GET_CODE (XEXP (x, 1)) == UNSPEC + && XINT (XEXP (x, 1), 1) == UNSPEC_LOAD_GOT + && XEXP (x, 0) == pic_offset_table_rtx + && (GET_CODE (XVECEXP (XEXP (x, 1), 0, 0)) == SYMBOL_REF + || GET_CODE (XVECEXP (XEXP (x, 1), 0, 0)) == LABEL_REF)) + return !no_large_offset; + if (GET_CODE (XEXP (x, 1)) != CONST_INT) + return false; + + off = INTVAL (XEXP (x, 1)); + + /* If the machine does not have doubleword load/stores, we'll use + word size accesses. */ + size1 = size; + if (size == 2 * UNITS_PER_WORD && !TARGET_STDW) + size = UNITS_PER_WORD; + + if (((HOST_WIDE_INT)size1 - 1) & off) + return false; + off /= size; + if (off > -32 && off < (size1 == size ? 32 : 28)) + return true; + if (no_large_offset || code != PLUS || XEXP (x, 0) != stack_pointer_rtx + || size1 > UNITS_PER_WORD) + return false; + return off >= 0 && off < 32768; + + case CONST: + case SYMBOL_REF: + case LABEL_REF: + return (!no_large_offset + /* With -fpic, we must wrap it in an unspec to show the B14 + dependency. */ + && !flag_pic + && GET_MODE_SIZE (mode) <= UNITS_PER_WORD + && sdata_symbolic_operand (x, Pmode)); + + default: + return false; + } +} + +static bool +c6x_legitimate_address_p (enum machine_mode mode, rtx x, bool strict) +{ + return c6x_legitimate_address_p_1 (mode, x, strict, false); +} + +static bool +c6x_legitimate_constant_p (enum machine_mode mode ATTRIBUTE_UNUSED, + rtx x ATTRIBUTE_UNUSED) +{ + return true; +} + +/* Implements TARGET_PREFERRED_RENAME_CLASS. */ +static reg_class_t +c6x_preferred_rename_class (reg_class_t cl) +{ + if (cl == A_REGS) + return NONPREDICATE_A_REGS; + if (cl == B_REGS) + return NONPREDICATE_B_REGS; + if (cl == ALL_REGS || cl == GENERAL_REGS) + return NONPREDICATE_REGS; + return NO_REGS; +} + +/* Implements FINAL_PRESCAN_INSN. */ +void +c6x_final_prescan_insn (rtx insn, rtx *opvec ATTRIBUTE_UNUSED, + int noperands ATTRIBUTE_UNUSED) +{ + c6x_current_insn = insn; +} + +/* A structure to describe the stack layout of a function. The layout is + as follows: + + [saved frame pointer (or possibly padding0)] + --> incoming stack pointer, new hard frame pointer + [saved call-used regs] + [optional padding1] + --> soft frame pointer + [frame] + [outgoing arguments] + [optional padding2] + + The structure members are laid out in this order. */ + +struct c6x_frame +{ + int padding0; + /* Number of registers to save. */ + int nregs; + int padding1; + HOST_WIDE_INT frame; + int outgoing_arguments_size; + int padding2; + + HOST_WIDE_INT to_allocate; + /* The offsets relative to the incoming stack pointer (which + becomes HARD_FRAME_POINTER). */ + HOST_WIDE_INT frame_pointer_offset; + HOST_WIDE_INT b3_offset; + + /* True if we should call push_rts/pop_rts to save and restore + registers. */ + bool push_rts; +}; + +/* Return true if we need to save and modify the PIC register in the + prologue. */ + +static bool +must_reload_pic_reg_p (void) +{ + struct cgraph_local_info *i = NULL; + + if (!TARGET_DSBT) + return false; + + i = cgraph_local_info (current_function_decl); + + if ((crtl->uses_pic_offset_table || !current_function_is_leaf) && !i->local) + return true; + return false; +} + +/* Return 1 if we need to save REGNO. */ +static int +c6x_save_reg (unsigned int regno) +{ + return ((df_regs_ever_live_p (regno) + && !call_used_regs[regno] + && !fixed_regs[regno]) + || (regno == RETURN_ADDR_REGNO + && (df_regs_ever_live_p (regno) + || !current_function_is_leaf)) + || (regno == PIC_OFFSET_TABLE_REGNUM && must_reload_pic_reg_p ())); +} + +/* Examine the number of regs NREGS we've determined we must save. + Return true if we should use __c6xabi_push_rts/__c6xabi_pop_rts for + prologue and epilogue. */ + +static bool +use_push_rts_p (int nregs) +{ + if (TARGET_INSNS_64PLUS && optimize_function_for_size_p (cfun) + && !cfun->machine->contains_sibcall + && !cfun->returns_struct + && !TARGET_LONG_CALLS + && nregs >= 6 && !frame_pointer_needed) + return true; + return false; +} + +/* Return number of saved general prupose registers. */ + +int +c6x_nsaved_regs (void) +{ + int nregs = 0; + int regno; + + for (regno = 0; regno < FIRST_PSEUDO_REGISTER; regno++) + if (c6x_save_reg (regno)) + nregs++; + return nregs; +} + +/* The safe debug order mandated by the ABI. */ +static unsigned reg_save_order[] = +{ + REG_A10, REG_A11, REG_A12, REG_A13, + REG_A14, REG_B3, + REG_B10, REG_B11, REG_B12, REG_B13, + REG_B14, REG_A15 +}; + +#define N_SAVE_ORDER (sizeof reg_save_order / sizeof *reg_save_order) + +/* Compute the layout of the stack frame and store it in FRAME. */ + +static void +c6x_compute_frame_layout (struct c6x_frame *frame) +{ + HOST_WIDE_INT size = get_frame_size (); + HOST_WIDE_INT offset; + int nregs; + + /* We use the four bytes which are technically inside the caller's frame, + usually to save the frame pointer. */ + offset = -4; + frame->padding0 = 0; + nregs = c6x_nsaved_regs (); + frame->push_rts = false; + frame->b3_offset = 0; + if (use_push_rts_p (nregs)) + { + frame->push_rts = true; + frame->b3_offset = (TARGET_BIG_ENDIAN ? -12 : -13) * 4; + nregs = 14; + } + else if (c6x_save_reg (REG_B3)) + { + int idx; + for (idx = N_SAVE_ORDER - 1; reg_save_order[idx] != REG_B3; idx--) + { + if (c6x_save_reg (reg_save_order[idx])) + frame->b3_offset -= 4; + } + } + frame->nregs = nregs; + + if (size == 0 && nregs == 0) + { + frame->padding0 = 4; + frame->padding1 = frame->padding2 = 0; + frame->frame_pointer_offset = frame->to_allocate = 0; + frame->outgoing_arguments_size = 0; + return; + } + + if (!frame->push_rts) + offset += frame->nregs * 4; + + if (offset == 0 && size == 0 && crtl->outgoing_args_size == 0 + && !current_function_is_leaf) + /* Don't use the bottom of the caller's frame if we have no + allocation of our own and call other functions. */ + frame->padding0 = frame->padding1 = 4; + else if (offset & 4) + frame->padding1 = 4; + else + frame->padding1 = 0; + + offset += frame->padding0 + frame->padding1; + frame->frame_pointer_offset = offset; + offset += size; + + frame->outgoing_arguments_size = crtl->outgoing_args_size; + offset += frame->outgoing_arguments_size; + + if ((offset & 4) == 0) + frame->padding2 = 8; + else + frame->padding2 = 4; + frame->to_allocate = offset + frame->padding2; +} + +/* Return the offset between two registers, one to be eliminated, and the other + its replacement, at the start of a routine. */ + +HOST_WIDE_INT +c6x_initial_elimination_offset (int from, int to) +{ + struct c6x_frame frame; + c6x_compute_frame_layout (&frame); + + if (from == ARG_POINTER_REGNUM && to == HARD_FRAME_POINTER_REGNUM) + return 0; + else if (from == FRAME_POINTER_REGNUM + && to == HARD_FRAME_POINTER_REGNUM) + return -frame.frame_pointer_offset; + else + { + gcc_assert (to == STACK_POINTER_REGNUM); + + if (from == ARG_POINTER_REGNUM) + return frame.to_allocate + (frame.push_rts ? 56 : 0); + + gcc_assert (from == FRAME_POINTER_REGNUM); + return frame.to_allocate - frame.frame_pointer_offset; + } +} + +/* Given FROM and TO register numbers, say whether this elimination is + allowed. Frame pointer elimination is automatically handled. */ + +static bool +c6x_can_eliminate (const int from ATTRIBUTE_UNUSED, const int to) +{ + if (to == STACK_POINTER_REGNUM) + return !frame_pointer_needed; + return true; +} + +/* Emit insns to increment the stack pointer by OFFSET. If + FRAME_RELATED_P, set the RTX_FRAME_RELATED_P flag on the insns. + Does nothing if the offset is zero. */ + +static void +emit_add_sp_const (HOST_WIDE_INT offset, bool frame_related_p) +{ + rtx to_add = GEN_INT (offset); + rtx orig_to_add = to_add; + rtx insn; + + if (offset == 0) + return; + + if (offset < -32768 || offset > 32767) + { + rtx reg = gen_rtx_REG (SImode, REG_A0); + rtx low = GEN_INT (trunc_int_for_mode (offset, HImode)); + + insn = emit_insn (gen_movsi_high (reg, low)); + if (frame_related_p) + RTX_FRAME_RELATED_P (insn) = 1; + insn = emit_insn (gen_movsi_lo_sum (reg, reg, to_add)); + if (frame_related_p) + RTX_FRAME_RELATED_P (insn) = 1; + to_add = reg; + } + insn = emit_insn (gen_addsi3 (stack_pointer_rtx, stack_pointer_rtx, + to_add)); + if (frame_related_p) + { + if (REG_P (to_add)) + add_reg_note (insn, REG_FRAME_RELATED_EXPR, + gen_rtx_SET (VOIDmode, stack_pointer_rtx, + gen_rtx_PLUS (Pmode, stack_pointer_rtx, + orig_to_add))); + + RTX_FRAME_RELATED_P (insn) = 1; + } +} + +/* Prologue and epilogue. */ +void +c6x_expand_prologue (void) +{ + struct c6x_frame frame; + rtx insn, mem; + int nsaved = 0; + HOST_WIDE_INT initial_offset, off, added_already; + + c6x_compute_frame_layout (&frame); + + if (flag_stack_usage_info) + current_function_static_stack_size = frame.to_allocate; + + initial_offset = -frame.to_allocate; + if (frame.push_rts) + { + emit_insn (gen_push_rts ()); + nsaved = frame.nregs; + } + + /* If the offsets would be too large for the memory references we will + create to save registers, do the stack allocation in two parts. + Ensure by subtracting 8 that we don't store to the word pointed to + by the stack pointer. */ + if (initial_offset < -32768) + initial_offset = -frame.frame_pointer_offset - 8; + + if (frame.to_allocate > 0) + gcc_assert (initial_offset != 0); + + off = -initial_offset + 4 - frame.padding0; + + mem = gen_frame_mem (Pmode, stack_pointer_rtx); + + added_already = 0; + if (frame_pointer_needed) + { + rtx fp_reg = gen_rtx_REG (SImode, REG_A15); + /* We go through some contortions here to both follow the ABI's + recommendation that FP == incoming SP, and to avoid writing or + reading the word pointed to by the stack pointer. */ + rtx addr = gen_rtx_POST_MODIFY (Pmode, stack_pointer_rtx, + gen_rtx_PLUS (Pmode, stack_pointer_rtx, + GEN_INT (-8))); + insn = emit_move_insn (gen_frame_mem (Pmode, addr), fp_reg); + RTX_FRAME_RELATED_P (insn) = 1; + nsaved++; + insn = emit_insn (gen_addsi3 (hard_frame_pointer_rtx, stack_pointer_rtx, + GEN_INT (8))); + RTX_FRAME_RELATED_P (insn) = 1; + off -= 4; + added_already = -8; + } + + emit_add_sp_const (initial_offset - added_already, true); + + if (nsaved < frame.nregs) + { + unsigned i; + + for (i = 0; i < N_SAVE_ORDER; i++) + { + int idx = N_SAVE_ORDER - i - 1; + unsigned regno = reg_save_order[idx]; + rtx reg; + enum machine_mode save_mode = SImode; + + if (regno == REG_A15 && frame_pointer_needed) + /* Already saved. */ + continue; + if (!c6x_save_reg (regno)) + continue; + + if (TARGET_STDW && (off & 4) == 0 && off <= 256 + && (regno & 1) == 1 + && i + 1 < N_SAVE_ORDER + && reg_save_order[idx - 1] == regno - 1 + && c6x_save_reg (regno - 1)) + { + save_mode = DImode; + regno--; + i++; + } + reg = gen_rtx_REG (save_mode, regno); + off -= GET_MODE_SIZE (save_mode); + + insn = emit_move_insn (adjust_address (mem, save_mode, off), + reg); + RTX_FRAME_RELATED_P (insn) = 1; + + nsaved += HARD_REGNO_NREGS (regno, save_mode); + } + } + gcc_assert (nsaved == frame.nregs); + emit_add_sp_const (-frame.to_allocate - initial_offset, true); + if (must_reload_pic_reg_p ()) + { + if (dsbt_decl == NULL) + { + tree t; + + t = build_index_type (integer_one_node); + t = build_array_type (integer_type_node, t); + t = build_decl (BUILTINS_LOCATION, VAR_DECL, + get_identifier ("__c6xabi_DSBT_BASE"), t); + DECL_ARTIFICIAL (t) = 1; + DECL_IGNORED_P (t) = 1; + DECL_EXTERNAL (t) = 1; + TREE_STATIC (t) = 1; + TREE_PUBLIC (t) = 1; + TREE_USED (t) = 1; + + dsbt_decl = t; + } + emit_insn (gen_setup_dsbt (pic_offset_table_rtx, + XEXP (DECL_RTL (dsbt_decl), 0))); + } +} + +void +c6x_expand_epilogue (bool sibcall) +{ + unsigned i; + struct c6x_frame frame; + rtx mem; + HOST_WIDE_INT off; + int nsaved = 0; + + c6x_compute_frame_layout (&frame); + + mem = gen_frame_mem (Pmode, stack_pointer_rtx); + + /* Insert a dummy set/use of the stack pointer. This creates a + scheduler barrier between the prologue saves and epilogue restores. */ + emit_insn (gen_epilogue_barrier (stack_pointer_rtx, stack_pointer_rtx)); + + /* If the offsets would be too large for the memory references we will + create to restore registers, do a preliminary stack adjustment here. */ + off = frame.to_allocate - frame.frame_pointer_offset + frame.padding1; + if (frame.push_rts) + { + nsaved = frame.nregs; + } + else + { + if (frame.to_allocate > 32768) + { + /* Don't add the entire offset so that we leave an unused word + above the stack pointer. */ + emit_add_sp_const ((off - 16) & ~7, false); + off &= 7; + off += 16; + } + for (i = 0; i < N_SAVE_ORDER; i++) + { + unsigned regno = reg_save_order[i]; + rtx reg; + enum machine_mode save_mode = SImode; + + if (!c6x_save_reg (regno)) + continue; + if (regno == REG_A15 && frame_pointer_needed) + continue; + + if (TARGET_STDW && (off & 4) == 0 && off < 256 + && (regno & 1) == 0 + && i + 1 < N_SAVE_ORDER + && reg_save_order[i + 1] == regno + 1 + && c6x_save_reg (regno + 1)) + { + save_mode = DImode; + i++; + } + reg = gen_rtx_REG (save_mode, regno); + + emit_move_insn (reg, adjust_address (mem, save_mode, off)); + + off += GET_MODE_SIZE (save_mode); + nsaved += HARD_REGNO_NREGS (regno, save_mode); + } + } + if (!frame_pointer_needed) + emit_add_sp_const (off + frame.padding0 - 4, false); + else + { + rtx fp_reg = gen_rtx_REG (SImode, REG_A15); + rtx addr = gen_rtx_PRE_MODIFY (Pmode, stack_pointer_rtx, + gen_rtx_PLUS (Pmode, stack_pointer_rtx, + GEN_INT (8))); + emit_insn (gen_addsi3 (stack_pointer_rtx, hard_frame_pointer_rtx, + GEN_INT (-8))); + emit_move_insn (fp_reg, gen_frame_mem (Pmode, addr)); + nsaved++; + } + gcc_assert (nsaved == frame.nregs); + if (!sibcall) + { + if (frame.push_rts) + emit_jump_insn (gen_pop_rts ()); + else + emit_jump_insn (gen_return_internal (gen_rtx_REG (SImode, + RETURN_ADDR_REGNO))); + } +} + +/* Return the value of the return address for the frame COUNT steps up + from the current frame, after the prologue. + We punt for everything but the current frame by returning const0_rtx. */ + +rtx +c6x_return_addr_rtx (int count) +{ + if (count != 0) + return const0_rtx; + + return get_hard_reg_initial_val (Pmode, RETURN_ADDR_REGNO); +} + +/* Return true iff TYPE is one of the shadow types. */ +static bool +shadow_type_p (enum attr_type type) +{ + return (type == TYPE_SHADOW || type == TYPE_LOAD_SHADOW + || type == TYPE_MULT_SHADOW); +} + +/* Return true iff INSN is a shadow pattern. */ +static bool +shadow_p (rtx insn) +{ + if (!NONDEBUG_INSN_P (insn) || recog_memoized (insn) < 0) + return false; + return shadow_type_p (get_attr_type (insn)); +} + +/* Return true iff INSN is a shadow or blockage pattern. */ +static bool +shadow_or_blockage_p (rtx insn) +{ + enum attr_type type; + if (!NONDEBUG_INSN_P (insn) || recog_memoized (insn) < 0) + return false; + type = get_attr_type (insn); + return shadow_type_p (type) || type == TYPE_BLOCKAGE; +} + +/* Translate UNITS into a bitmask of units we can reserve for this + insn. */ +static int +get_reservation_flags (enum attr_units units) +{ + switch (units) + { + case UNITS_D: + case UNITS_D_ADDR: + return RESERVATION_FLAG_D; + case UNITS_L: + return RESERVATION_FLAG_L; + case UNITS_S: + return RESERVATION_FLAG_S; + case UNITS_M: + return RESERVATION_FLAG_M; + case UNITS_LS: + return RESERVATION_FLAG_LS; + case UNITS_DL: + return RESERVATION_FLAG_DL; + case UNITS_DS: + return RESERVATION_FLAG_DS; + case UNITS_DLS: + return RESERVATION_FLAG_DLS; + default: + return 0; + } +} + +/* Compute the side of the machine used by INSN, which reserves UNITS. + This must match the reservations in the scheduling description. */ +static int +get_insn_side (rtx insn, enum attr_units units) +{ + if (units == UNITS_D_ADDR) + return (get_attr_addr_regfile (insn) == ADDR_REGFILE_A ? 0 : 1); + else + { + enum attr_dest_regfile rf = get_attr_dest_regfile (insn); + if (rf == DEST_REGFILE_ANY) + return get_attr_type (insn) == TYPE_BRANCH ? 0 : 1; + else + return rf == DEST_REGFILE_A ? 0 : 1; + } +} + +/* After scheduling, walk the insns between HEAD and END and assign unit + reservations. */ +static void +assign_reservations (rtx head, rtx end) +{ + rtx insn; + for (insn = head; insn != NEXT_INSN (end); insn = NEXT_INSN (insn)) + { + rtx within; + int pass; + int rsrv[2]; + int rsrv_count[2][4]; + + if (GET_MODE (insn) != TImode) + continue; + + rsrv[0] = rsrv[1] = 0; + memset (rsrv_count, 0, sizeof rsrv_count); + + /* Walk through the insns that occur in the same cycle. We use multiple + passes to assign units, assigning for insns with the most specific + requirements first. */ + for (pass = 0; pass < 4; pass++) + for (within = insn; + (within != NEXT_INSN (end) + && (within == insn || GET_MODE (within) != TImode)); + within = NEXT_INSN (within)) + { + int this_rsrv, side; + int icode; + enum attr_units units; + int j; + + if (!NONDEBUG_INSN_P (within)) + continue; + icode = recog_memoized (within); + if (icode < 0) + continue; + units = get_attr_units (within); + this_rsrv = get_reservation_flags (units); + if (this_rsrv == 0) + continue; + side = get_insn_side (within, units); + + if ((this_rsrv & (this_rsrv - 1)) == 0) + { + int t = exact_log2 (this_rsrv) + side * 4; + rsrv[side] |= this_rsrv; + INSN_INFO_ENTRY (INSN_UID (within)).reservation = t; + continue; + } + + if (pass == 1) + { + for (j = 0; j < 4; j++) + if (this_rsrv & (1 << j)) + rsrv_count[side][j]++; + continue; + } + if ((pass == 2 && this_rsrv != RESERVATION_FLAG_DLS) + || (pass == 3 && this_rsrv == RESERVATION_FLAG_DLS)) + { + int best = -1, best_cost = INT_MAX; + for (j = 0; j < 4; j++) + if ((this_rsrv & (1 << j)) + && !(rsrv[side] & (1 << j)) + && rsrv_count[side][j] < best_cost) + { + best_cost = rsrv_count[side][j]; + best = j; + } + gcc_assert (best != -1); + rsrv[side] |= 1 << best; + for (j = 0; j < 4; j++) + if ((this_rsrv & (1 << j)) && j != best) + rsrv_count[side][j]--; + + INSN_INFO_ENTRY (INSN_UID (within)).reservation + = best + side * 4; + } + } + } +} + +/* Backend scheduling state. */ +typedef struct c6x_sched_context +{ + /* The current scheduler clock, saved in the sched_reorder hook. */ + int curr_sched_clock; + + /* Number of insns issued so far in this cycle. */ + int issued_this_cycle; + + /* We record the time at which each jump occurs in JUMP_CYCLES. The + theoretical maximum for number of jumps in flight is 12: 2 every + cycle, with a latency of 6 cycles each. This is a circular + buffer; JUMP_CYCLE_INDEX is the pointer to the start. Earlier + jumps have a higher index. This array should be accessed through + the jump_cycle function. */ + int jump_cycles[12]; + int jump_cycle_index; + + /* In parallel with jump_cycles, this array records the opposite of + the condition used in each pending jump. This is used to + predicate insns that are scheduled in the jump's delay slots. If + this is NULL_RTX no such predication happens. */ + rtx jump_cond[12]; + + /* Similar to the jump_cycles mechanism, but here we take into + account all insns with delay slots, to avoid scheduling asms into + the delay slots. */ + int delays_finished_at; + + /* The following variable value is the last issued insn. */ + rtx last_scheduled_insn; + + int reg_n_accesses[FIRST_PSEUDO_REGISTER]; + int reg_n_xaccesses[FIRST_PSEUDO_REGISTER]; + int reg_set_in_cycle[FIRST_PSEUDO_REGISTER]; + + int tmp_reg_n_accesses[FIRST_PSEUDO_REGISTER]; + int tmp_reg_n_xaccesses[FIRST_PSEUDO_REGISTER]; +} *c6x_sched_context_t; + +/* The current scheduling state. */ +static struct c6x_sched_context ss; + +/* Set when we discover while processing an insn that it would lead to too + many accesses of the same register. */ +static bool reg_access_stall; + +/* Look up the jump cycle with index N. For an out-of-bounds N, we return 0, + so the caller does not specifically have to test for it. */ +static int +get_jump_cycle (int n) +{ + if (n >= 12) + return 0; + n += ss.jump_cycle_index; + if (n >= 12) + n -= 12; + return ss.jump_cycles[n]; +} + +/* Look up the jump condition with index N. */ +static rtx +get_jump_cond (int n) +{ + if (n >= 12) + return NULL_RTX; + n += ss.jump_cycle_index; + if (n >= 12) + n -= 12; + return ss.jump_cond[n]; +} + +/* Return the index of the first jump that occurs after CLOCK_VAR. If no jump + has delay slots beyond CLOCK_VAR, return -1. */ +static int +first_jump_index (int clock_var) +{ + int retval = -1; + int n = 0; + for (;;) + { + int t = get_jump_cycle (n); + if (t <= clock_var) + break; + retval = n; + n++; + } + return retval; +} + +/* Add a new entry in our scheduling state for a jump that occurs in CYCLE + and has the opposite condition of COND. */ +static void +record_jump (int cycle, rtx cond) +{ + if (ss.jump_cycle_index == 0) + ss.jump_cycle_index = 11; + else + ss.jump_cycle_index--; + ss.jump_cycles[ss.jump_cycle_index] = cycle; + ss.jump_cond[ss.jump_cycle_index] = cond; +} + +/* Set the clock cycle of INSN to CYCLE. Also clears the insn's entry in + new_conditions. */ +static void +insn_set_clock (rtx insn, int cycle) +{ + unsigned uid = INSN_UID (insn); + + if (uid >= INSN_INFO_LENGTH) + VEC_safe_grow (c6x_sched_insn_info, heap, insn_info, uid * 5 / 4 + 10); + + INSN_INFO_ENTRY (uid).clock = cycle; + INSN_INFO_ENTRY (uid).new_cond = NULL; + INSN_INFO_ENTRY (uid).ebb_start = false; +} + +/* Return the clock cycle we set for the insn with uid UID. */ +static int +insn_uid_get_clock (int uid) +{ + return INSN_INFO_ENTRY (uid).clock; +} + +/* Return the clock cycle we set for INSN. */ +static int +insn_get_clock (rtx insn) +{ + return insn_uid_get_clock (INSN_UID (insn)); +} + +/* Examine INSN, and if it is a conditional jump of any kind, return + the opposite of the condition in which it branches. Otherwise, + return NULL_RTX. */ +static rtx +condjump_opposite_condition (rtx insn) +{ + rtx pat = PATTERN (insn); + int icode = INSN_CODE (insn); + rtx x = NULL; + + if (icode == CODE_FOR_br_true || icode == CODE_FOR_br_false) + { + x = XEXP (SET_SRC (pat), 0); + if (icode == CODE_FOR_br_false) + return x; + } + if (GET_CODE (pat) == COND_EXEC) + { + rtx t = COND_EXEC_CODE (pat); + if ((GET_CODE (t) == PARALLEL + && GET_CODE (XVECEXP (t, 0, 0)) == RETURN) + || (GET_CODE (t) == UNSPEC && XINT (t, 1) == UNSPEC_REAL_JUMP) + || (GET_CODE (t) == SET && SET_DEST (t) == pc_rtx)) + x = COND_EXEC_TEST (pat); + } + + if (x != NULL_RTX) + { + enum rtx_code code = GET_CODE (x); + x = gen_rtx_fmt_ee (code == EQ ? NE : EQ, + GET_MODE (x), XEXP (x, 0), + XEXP (x, 1)); + } + return x; +} + +/* Return true iff COND1 and COND2 are exactly opposite conditions + one of them NE and the other EQ. */ +static bool +conditions_opposite_p (rtx cond1, rtx cond2) +{ + return (rtx_equal_p (XEXP (cond1, 0), XEXP (cond2, 0)) + && rtx_equal_p (XEXP (cond1, 1), XEXP (cond2, 1)) + && GET_CODE (cond1) == reverse_condition (GET_CODE (cond2))); +} + +/* Return true if we can add a predicate COND to INSN, or if INSN + already has that predicate. If DOIT is true, also perform the + modification. */ +static bool +predicate_insn (rtx insn, rtx cond, bool doit) +{ + int icode; + if (cond == NULL_RTX) + { + gcc_assert (!doit); + return false; + } + + if (get_attr_predicable (insn) == PREDICABLE_YES + && GET_CODE (PATTERN (insn)) != COND_EXEC) + { + if (doit) + { + rtx newpat = gen_rtx_COND_EXEC (VOIDmode, cond, PATTERN (insn)); + PATTERN (insn) = newpat; + INSN_CODE (insn) = -1; + } + return true; + } + if (GET_CODE (PATTERN (insn)) == COND_EXEC + && rtx_equal_p (COND_EXEC_TEST (PATTERN (insn)), cond)) + return true; + icode = INSN_CODE (insn); + if (icode == CODE_FOR_real_jump + || icode == CODE_FOR_jump + || icode == CODE_FOR_indirect_jump) + { + rtx pat = PATTERN (insn); + rtx dest = (icode == CODE_FOR_real_jump ? XVECEXP (pat, 0, 0) + : icode == CODE_FOR_jump ? XEXP (SET_SRC (pat), 0) + : SET_SRC (pat)); + if (doit) + { + rtx newpat; + if (REG_P (dest)) + newpat = gen_rtx_COND_EXEC (VOIDmode, cond, PATTERN (insn)); + else + newpat = gen_br_true (cond, XEXP (cond, 0), dest); + PATTERN (insn) = newpat; + INSN_CODE (insn) = -1; + } + return true; + } + if (INSN_CODE (insn) == CODE_FOR_br_true) + { + rtx br_cond = XEXP (SET_SRC (PATTERN (insn)), 0); + return rtx_equal_p (br_cond, cond); + } + if (INSN_CODE (insn) == CODE_FOR_br_false) + { + rtx br_cond = XEXP (SET_SRC (PATTERN (insn)), 0); + return conditions_opposite_p (br_cond, cond); + } + return false; +} + +/* Initialize SC. Used by c6x_init_sched_context and c6x_sched_init. */ +static void +init_sched_state (c6x_sched_context_t sc) +{ + sc->last_scheduled_insn = NULL_RTX; + sc->issued_this_cycle = 0; + memset (sc->jump_cycles, 0, sizeof sc->jump_cycles); + memset (sc->jump_cond, 0, sizeof sc->jump_cond); + sc->jump_cycle_index = 0; + sc->delays_finished_at = 0; + sc->curr_sched_clock = 0; + + memset (sc->reg_n_accesses, 0, sizeof sc->reg_n_accesses); + memset (sc->reg_n_xaccesses, 0, sizeof sc->reg_n_xaccesses); + memset (sc->reg_set_in_cycle, 0, sizeof sc->reg_set_in_cycle); +} + +/* Allocate store for new scheduling context. */ +static void * +c6x_alloc_sched_context (void) +{ + return xmalloc (sizeof (struct c6x_sched_context)); +} + +/* If CLEAN_P is true then initializes _SC with clean data, + and from the global context otherwise. */ +static void +c6x_init_sched_context (void *_sc, bool clean_p) +{ + c6x_sched_context_t sc = (c6x_sched_context_t) _sc; + + if (clean_p) + { + init_sched_state (sc); + } + else + *sc = ss; +} + +/* Sets the global scheduling context to the one pointed to by _SC. */ +static void +c6x_set_sched_context (void *_sc) +{ + c6x_sched_context_t sc = (c6x_sched_context_t) _sc; + + gcc_assert (sc != NULL); + ss = *sc; +} + +/* Free _SC. */ +static void +c6x_free_sched_context (void *_sc) +{ + free (_sc); +} + +/* Provide information about speculation capabilities, and set the + DO_BACKTRACKING flag. */ +static void +c6x_set_sched_flags (spec_info_t spec_info) +{ + unsigned int *flags = &(current_sched_info->flags); + + if (*flags & SCHED_EBB) + { + *flags |= DO_BACKTRACKING; + } + + spec_info->mask = 0; +} + +/* Implement the TARGET_SCHED_ISSUE_RATE hook. */ + +static int +c6x_issue_rate (void) +{ + return 8; +} + +/* We're beginning a new block. Initialize data structures as necessary. */ + +static void +c6x_sched_init (FILE *dump ATTRIBUTE_UNUSED, + int sched_verbose ATTRIBUTE_UNUSED, + int max_ready ATTRIBUTE_UNUSED) +{ + init_sched_state (&ss); +} + +static void +c6x_mark_regno_read (int regno, bool cross) +{ + int t = ++ss.tmp_reg_n_accesses[regno]; + + if (t > 4) + reg_access_stall = true; + + if (cross) + { + int set_cycle = ss.reg_set_in_cycle[regno]; + /* This must be done in this way rather than by tweaking things in + adjust_cost, since the stall occurs even for insns with opposite + predicates, and the scheduler may not even see a dependency. */ + if (set_cycle > 0 && set_cycle == ss.curr_sched_clock) + reg_access_stall = true; + /* This doesn't quite do anything yet as we're only modeling one + x unit. */ + ++ss.tmp_reg_n_xaccesses[regno]; + } +} + +/* Note that REG is read in the insn being examined. If CROSS, it + means the access is through a cross path. Update the temporary reg + access arrays, and set REG_ACCESS_STALL if the insn can't be issued + in the current cycle. */ + +static void +c6x_mark_reg_read (rtx reg, bool cross) +{ + unsigned regno = REGNO (reg); + unsigned nregs = hard_regno_nregs[regno][GET_MODE (reg)]; + + while (nregs-- > 0) + c6x_mark_regno_read (regno + nregs, cross); +} + +/* Note that register REG is written in cycle CYCLES. */ + +static void +c6x_mark_reg_written (rtx reg, int cycles) +{ + unsigned regno = REGNO (reg); + unsigned nregs = hard_regno_nregs[regno][GET_MODE (reg)]; + + while (nregs-- > 0) + ss.reg_set_in_cycle[regno + nregs] = cycles; +} + +/* Update the register state information for an instruction whose + body is X. Return true if the instruction has to be delayed until the + next cycle. */ + +static bool +c6x_registers_update (rtx insn) +{ + enum attr_cross cross; + enum attr_dest_regfile destrf; + int i, nops; + rtx x; + + if (!reload_completed || recog_memoized (insn) < 0) + return false; + + reg_access_stall = false; + memcpy (ss.tmp_reg_n_accesses, ss.reg_n_accesses, + sizeof ss.tmp_reg_n_accesses); + memcpy (ss.tmp_reg_n_xaccesses, ss.reg_n_xaccesses, + sizeof ss.tmp_reg_n_xaccesses); + + extract_insn (insn); + + cross = get_attr_cross (insn); + destrf = get_attr_dest_regfile (insn); + + nops = recog_data.n_operands; + x = PATTERN (insn); + if (GET_CODE (x) == COND_EXEC) + { + c6x_mark_reg_read (XEXP (XEXP (x, 0), 0), false); + nops -= 2; + } + + for (i = 0; i < nops; i++) + { + rtx op = recog_data.operand[i]; + if (recog_data.operand_type[i] == OP_OUT) + continue; + if (REG_P (op)) + { + bool this_cross = cross; + if (destrf == DEST_REGFILE_A && A_REGNO_P (REGNO (op))) + this_cross = false; + if (destrf == DEST_REGFILE_B && B_REGNO_P (REGNO (op))) + this_cross = false; + c6x_mark_reg_read (op, this_cross); + } + else if (MEM_P (op)) + { + op = XEXP (op, 0); + switch (GET_CODE (op)) + { + case POST_INC: + case PRE_INC: + case POST_DEC: + case PRE_DEC: + op = XEXP (op, 0); + /* fall through */ + case REG: + c6x_mark_reg_read (op, false); + break; + case POST_MODIFY: + case PRE_MODIFY: + op = XEXP (op, 1); + gcc_assert (GET_CODE (op) == PLUS); + /* fall through */ + case PLUS: + c6x_mark_reg_read (XEXP (op, 0), false); + if (REG_P (XEXP (op, 1))) + c6x_mark_reg_read (XEXP (op, 1), false); + break; + case SYMBOL_REF: + case LABEL_REF: + case CONST: + c6x_mark_regno_read (REG_B14, false); + break; + default: + gcc_unreachable (); + } + } + else if (!CONSTANT_P (op) && strlen (recog_data.constraints[i]) > 0) + gcc_unreachable (); + } + return reg_access_stall; +} + +/* Helper function for the TARGET_SCHED_REORDER and + TARGET_SCHED_REORDER2 hooks. If scheduling an insn would be unsafe + in the current cycle, move it down in the ready list and return the + number of non-unsafe insns. */ + +static int +c6x_sched_reorder_1 (rtx *ready, int *pn_ready, int clock_var) +{ + int n_ready = *pn_ready; + rtx *e_ready = ready + n_ready; + rtx *insnp; + int first_jump; + + /* Keep track of conflicts due to a limit number of register accesses, + and due to stalls incurred by too early accesses of registers using + cross paths. */ + + for (insnp = ready; insnp < e_ready; insnp++) + { + rtx insn = *insnp; + int icode = recog_memoized (insn); + bool is_asm = (icode < 0 + && (GET_CODE (PATTERN (insn)) == ASM_INPUT + || asm_noperands (PATTERN (insn)) >= 0)); + bool no_parallel = (is_asm + || (icode >= 0 + && get_attr_type (insn) == TYPE_ATOMIC)); + + /* We delay asm insns until all delay slots are exhausted. We can't + accurately tell how many cycles an asm takes, and the main scheduling + code always assumes at least 1 cycle, which may be wrong. */ + if ((no_parallel + && (ss.issued_this_cycle > 0 || clock_var < ss.delays_finished_at)) + || c6x_registers_update (insn)) + { + memmove (ready + 1, ready, (insnp - ready) * sizeof (rtx)); + *ready = insn; + n_ready--; + ready++; + } + else if (shadow_p (insn)) + { + memmove (ready + 1, ready, (insnp - ready) * sizeof (rtx)); + *ready = insn; + } + } + + /* Ensure that no other jump is scheduled in jump delay slots, since + it would put the machine into the wrong state. Also, we must + avoid scheduling insns that have a latency longer than the + remaining jump delay slots, as the code at the jump destination + won't be prepared for it. + + However, we can relax this condition somewhat. The rest of the + scheduler will automatically avoid scheduling an insn on which + the jump shadow depends so late that its side effect happens + after the jump. This means that if we see an insn with a longer + latency here, it can safely be scheduled if we can ensure that it + has a predicate opposite of the previous jump: the side effect + will happen in what we think of as the same basic block. In + c6x_variable_issue, we will record the necessary predicate in + new_conditions, and after scheduling is finished, we will modify + the insn. + + Special care must be taken whenever there is more than one jump + in flight. */ + + first_jump = first_jump_index (clock_var); + if (first_jump != -1) + { + int first_cycle = get_jump_cycle (first_jump); + rtx first_cond = get_jump_cond (first_jump); + int second_cycle = 0; + + if (first_jump > 0) + second_cycle = get_jump_cycle (first_jump - 1); + + for (insnp = ready; insnp < e_ready; insnp++) + { + rtx insn = *insnp; + int icode = recog_memoized (insn); + bool is_asm = (icode < 0 + && (GET_CODE (PATTERN (insn)) == ASM_INPUT + || asm_noperands (PATTERN (insn)) >= 0)); + int this_cycles; + enum attr_type type; + + gcc_assert (!is_asm); + if (icode < 0) + continue; + this_cycles = get_attr_cycles (insn); + type = get_attr_type (insn); + /* Treat branches specially; there is also a hazard if two jumps + end at the same cycle. */ + if (type == TYPE_BRANCH || type == TYPE_CALL) + this_cycles++; + if (clock_var + this_cycles <= first_cycle) + continue; + if ((first_jump > 0 && clock_var + this_cycles > second_cycle) + || !predicate_insn (insn, first_cond, false)) + { + memmove (ready + 1, ready, (insnp - ready) * sizeof (rtx)); + *ready = insn; + n_ready--; + ready++; + } + } + } + + return n_ready; +} + +/* Implement the TARGET_SCHED_REORDER hook. We save the current clock + for later and clear the register access information for the new + cycle. We also move asm statements out of the way if they would be + scheduled in a delay slot. */ + +static int +c6x_sched_reorder (FILE *dump ATTRIBUTE_UNUSED, + int sched_verbose ATTRIBUTE_UNUSED, + rtx *ready ATTRIBUTE_UNUSED, + int *pn_ready ATTRIBUTE_UNUSED, int clock_var) +{ + ss.curr_sched_clock = clock_var; + ss.issued_this_cycle = 0; + memset (ss.reg_n_accesses, 0, sizeof ss.reg_n_accesses); + memset (ss.reg_n_xaccesses, 0, sizeof ss.reg_n_xaccesses); + + if (ready == NULL) + return 0; + + return c6x_sched_reorder_1 (ready, pn_ready, clock_var); +} + +/* Implement the TARGET_SCHED_REORDER2 hook. We use this to record the clock + cycle for every insn. */ + +static int +c6x_sched_reorder2 (FILE *dump ATTRIBUTE_UNUSED, + int sched_verbose ATTRIBUTE_UNUSED, + rtx *ready ATTRIBUTE_UNUSED, + int *pn_ready ATTRIBUTE_UNUSED, int clock_var) +{ + /* FIXME: the assembler rejects labels inside an execute packet. + This can occur if prologue insns are scheduled in parallel with + others, so we avoid this here. Also make sure that nothing is + scheduled in parallel with a TYPE_ATOMIC insn or after a jump. */ + if (RTX_FRAME_RELATED_P (ss.last_scheduled_insn) + || JUMP_P (ss.last_scheduled_insn) + || (recog_memoized (ss.last_scheduled_insn) >= 0 + && get_attr_type (ss.last_scheduled_insn) == TYPE_ATOMIC)) + { + int n_ready = *pn_ready; + rtx *e_ready = ready + n_ready; + rtx *insnp; + + for (insnp = ready; insnp < e_ready; insnp++) + { + rtx insn = *insnp; + if (!shadow_p (insn)) + { + memmove (ready + 1, ready, (insnp - ready) * sizeof (rtx)); + *ready = insn; + n_ready--; + ready++; + } + } + return n_ready; + } + + return c6x_sched_reorder_1 (ready, pn_ready, clock_var); +} + +/* Subroutine of maybe_clobber_cond, called through note_stores. */ + +static void +clobber_cond_1 (rtx x, const_rtx pat ATTRIBUTE_UNUSED, void *data1) +{ + rtx *cond = (rtx *)data1; + if (*cond != NULL_RTX && reg_overlap_mentioned_p (x, *cond)) + *cond = NULL_RTX; +} + +/* Examine INSN, and if it destroys the conditions have recorded for + any of the jumps in flight, clear that condition so that we don't + predicate any more insns. CLOCK_VAR helps us limit the search to + only those jumps which are still in flight. */ + +static void +maybe_clobber_cond (rtx insn, int clock_var) +{ + int n, idx; + idx = ss.jump_cycle_index; + for (n = 0; n < 12; n++, idx++) + { + rtx cond, link; + int cycle; + + if (idx >= 12) + idx -= 12; + cycle = ss.jump_cycles[idx]; + if (cycle <= clock_var) + return; + + cond = ss.jump_cond[idx]; + if (cond == NULL_RTX) + continue; + + if (CALL_P (insn)) + { + ss.jump_cond[idx] = NULL_RTX; + continue; + } + + note_stores (PATTERN (insn), clobber_cond_1, ss.jump_cond + idx); + for (link = REG_NOTES (insn); link; link = XEXP (link, 1)) + if (REG_NOTE_KIND (link) == REG_INC) + clobber_cond_1 (XEXP (link, 0), NULL_RTX, ss.jump_cond + idx); + } +} + +/* Implement the TARGET_SCHED_VARIABLE_ISSUE hook. We are about to + issue INSN. Return the number of insns left on the ready queue + that can be issued this cycle. + We use this hook to record clock cycles and reservations for every insn. */ + +static int +c6x_variable_issue (FILE *dump ATTRIBUTE_UNUSED, + int sched_verbose ATTRIBUTE_UNUSED, + rtx insn, int can_issue_more ATTRIBUTE_UNUSED) +{ + ss.last_scheduled_insn = insn; + if (GET_CODE (PATTERN (insn)) != USE && GET_CODE (PATTERN (insn)) != CLOBBER) + ss.issued_this_cycle++; + if (insn_info) + { + int curr_clock = ss.curr_sched_clock; + int uid = INSN_UID (insn); + int icode = recog_memoized (insn); + rtx first_cond; + int first, first_cycle; + + insn_set_clock (insn, curr_clock); + INSN_INFO_ENTRY (uid).ebb_start + = curr_clock == 0 && ss.issued_this_cycle == 1; + + first = first_jump_index (ss.curr_sched_clock); + if (first == -1) + { + first_cycle = 0; + first_cond = NULL_RTX; + } + else + { + first_cycle = get_jump_cycle (first); + first_cond = get_jump_cond (first); + } + if (icode >= 0 + && first_cycle > curr_clock + && first_cond != NULL_RTX + && (curr_clock + get_attr_cycles (insn) > first_cycle + || get_attr_type (insn) == TYPE_BRANCH + || get_attr_type (insn) == TYPE_CALL)) + INSN_INFO_ENTRY (uid).new_cond = first_cond; + + maybe_clobber_cond (insn, curr_clock); + + if (icode >= 0) + { + int i, cycles; + + c6x_registers_update (insn); + memcpy (ss.reg_n_accesses, ss.tmp_reg_n_accesses, + sizeof ss.reg_n_accesses); + memcpy (ss.reg_n_xaccesses, ss.tmp_reg_n_accesses, + sizeof ss.reg_n_xaccesses); + + cycles = get_attr_cycles (insn); + if (ss.delays_finished_at < ss.curr_sched_clock + cycles) + ss.delays_finished_at = ss.curr_sched_clock + cycles; + if (get_attr_type (insn) == TYPE_BRANCH + || get_attr_type (insn) == TYPE_CALL) + { + rtx opposite = condjump_opposite_condition (insn); + record_jump (ss.curr_sched_clock + cycles, opposite); + } + + /* Mark the cycles in which the destination registers are written. + This is used for calculating stalls when using cross units. */ + extract_insn (insn); + /* Cross-path stalls don't apply to results of load insns. */ + if (get_attr_type (insn) == TYPE_LOAD + || get_attr_type (insn) == TYPE_LOADN + || get_attr_type (insn) == TYPE_LOAD_SHADOW) + cycles--; + for (i = 0; i < recog_data.n_operands; i++) + { + rtx op = recog_data.operand[i]; + if (MEM_P (op)) + { + rtx addr = XEXP (op, 0); + if (GET_RTX_CLASS (GET_CODE (addr)) == RTX_AUTOINC) + c6x_mark_reg_written (XEXP (addr, 0), + insn_uid_get_clock (uid) + 1); + } + if (recog_data.operand_type[i] != OP_IN + && REG_P (op)) + { + c6x_mark_reg_written (op, + insn_uid_get_clock (uid) + cycles); + } + } + } + } + return can_issue_more; +} + +/* Implement the TARGET_SCHED_ADJUST_COST hook. We need special handling for + anti- and output dependencies. */ + +static int +c6x_adjust_cost (rtx insn, rtx link, rtx dep_insn, int cost) +{ + enum attr_type insn_type = TYPE_UNKNOWN, dep_insn_type = TYPE_UNKNOWN; + int dep_insn_code_number, insn_code_number; + int shadow_bonus = 0; + enum reg_note kind; + dep_insn_code_number = recog_memoized (dep_insn); + insn_code_number = recog_memoized (insn); + + if (dep_insn_code_number >= 0) + dep_insn_type = get_attr_type (dep_insn); + + if (insn_code_number >= 0) + insn_type = get_attr_type (insn); + + kind = REG_NOTE_KIND (link); + if (kind == 0) + { + /* If we have a dependency on a load, and it's not for the result of + the load, it must be for an autoincrement. Reduce the cost in that + case. */ + if (dep_insn_type == TYPE_LOAD) + { + rtx set = PATTERN (dep_insn); + if (GET_CODE (set) == COND_EXEC) + set = COND_EXEC_CODE (set); + if (GET_CODE (set) == UNSPEC) + cost = 1; + else + { + gcc_assert (GET_CODE (set) == SET); + if (!reg_overlap_mentioned_p (SET_DEST (set), PATTERN (insn))) + cost = 1; + } + } + } + + /* A jump shadow needs to have its latency decreased by one. Conceptually, + it occurs in between two cycles, but we schedule it at the end of the + first cycle. */ + if (shadow_type_p (insn_type)) + shadow_bonus = 1; + + /* Anti and output dependencies usually have zero cost, but we want + to insert a stall after a jump, and after certain floating point + insns that take more than one cycle to read their inputs. In the + future, we should try to find a better algorithm for scheduling + jumps. */ + if (kind != 0) + { + /* We can get anti-dependencies against shadow insns. Treat these + like output dependencies, so that the insn is entirely finished + before the branch takes place. */ + if (kind == REG_DEP_ANTI && insn_type == TYPE_SHADOW) + kind = REG_DEP_OUTPUT; + switch (dep_insn_type) + { + case TYPE_CALLP: + return 1; + case TYPE_BRANCH: + case TYPE_CALL: + if (get_attr_has_shadow (dep_insn) == HAS_SHADOW_Y) + /* This is a real_jump/real_call insn. These don't have + outputs, and ensuring the validity of scheduling things + in the delay slot is the job of + c6x_sched_reorder_1. */ + return 0; + /* Unsplit calls can happen - e.g. for divide insns. */ + return 6; + case TYPE_LOAD: + case TYPE_LOADN: + case TYPE_INTDP: + if (kind == REG_DEP_OUTPUT) + return 5 - shadow_bonus; + return 0; + case TYPE_MPY4: + case TYPE_FP4: + if (kind == REG_DEP_OUTPUT) + return 4 - shadow_bonus; + return 0; + case TYPE_MPY2: + if (kind == REG_DEP_OUTPUT) + return 2 - shadow_bonus; + return 0; + case TYPE_CMPDP: + if (kind == REG_DEP_OUTPUT) + return 2 - shadow_bonus; + return 2; + case TYPE_ADDDP: + case TYPE_MPYSPDP: + if (kind == REG_DEP_OUTPUT) + return 7 - shadow_bonus; + return 2; + case TYPE_MPYSP2DP: + if (kind == REG_DEP_OUTPUT) + return 5 - shadow_bonus; + return 2; + case TYPE_MPYI: + if (kind == REG_DEP_OUTPUT) + return 9 - shadow_bonus; + return 4; + case TYPE_MPYID: + case TYPE_MPYDP: + if (kind == REG_DEP_OUTPUT) + return 10 - shadow_bonus; + return 4; + + default: + if (insn_type == TYPE_SPKERNEL) + return 0; + if (kind == REG_DEP_OUTPUT) + return 1 - shadow_bonus; + + return 0; + } + } + + return cost - shadow_bonus; +} + +/* Create a SEQUENCE rtx to replace the instructions in SLOT, of which there + are N_FILLED. REAL_FIRST identifies the slot if the insn that appears + first in the original stream. */ + +static void +gen_one_bundle (rtx *slot, int n_filled, int real_first) +{ + rtx bundle; + rtx t; + int i; + + bundle = gen_rtx_SEQUENCE (VOIDmode, gen_rtvec_v (n_filled, slot)); + bundle = make_insn_raw (bundle); + BLOCK_FOR_INSN (bundle) = BLOCK_FOR_INSN (slot[0]); + INSN_LOCATOR (bundle) = INSN_LOCATOR (slot[0]); + PREV_INSN (bundle) = PREV_INSN (slot[real_first]); + + t = NULL_RTX; + + for (i = 0; i < n_filled; i++) + { + rtx insn = slot[i]; + remove_insn (insn); + PREV_INSN (insn) = t ? t : PREV_INSN (bundle); + if (t != NULL_RTX) + NEXT_INSN (t) = insn; + t = insn; + if (i > 0) + INSN_LOCATOR (slot[i]) = INSN_LOCATOR (bundle); + } + + NEXT_INSN (bundle) = NEXT_INSN (PREV_INSN (bundle)); + NEXT_INSN (t) = NEXT_INSN (bundle); + NEXT_INSN (PREV_INSN (bundle)) = bundle; + PREV_INSN (NEXT_INSN (bundle)) = bundle; +} + +/* Move all parallel instructions into SEQUENCEs, so that no subsequent passes + try to insert labels in the middle. */ + +static void +c6x_gen_bundles (void) +{ + basic_block bb; + rtx insn, next, last_call; + + FOR_EACH_BB (bb) + { + rtx insn, next; + /* The machine is eight insns wide. We can have up to six shadow + insns, plus an extra slot for merging the jump shadow. */ + rtx slot[15]; + int n_filled = 0; + int first_slot = 0; + + for (insn = BB_HEAD (bb);; insn = next) + { + int at_end; + rtx delete_this = NULL_RTX; + + if (NONDEBUG_INSN_P (insn)) + { + /* Put calls at the start of the sequence. */ + if (CALL_P (insn)) + { + first_slot++; + if (n_filled) + { + memmove (&slot[1], &slot[0], + n_filled * sizeof (slot[0])); + } + if (!shadow_p (insn)) + { + PUT_MODE (insn, TImode); + if (n_filled) + PUT_MODE (slot[1], VOIDmode); + } + n_filled++; + slot[0] = insn; + } + else + { + slot[n_filled++] = insn; + } + } + + next = NEXT_INSN (insn); + while (next && insn != BB_END (bb) + && !(NONDEBUG_INSN_P (next) + && GET_CODE (PATTERN (next)) != USE + && GET_CODE (PATTERN (next)) != CLOBBER)) + { + insn = next; + next = NEXT_INSN (insn); + } + + at_end = insn == BB_END (bb); + if (delete_this == NULL_RTX + && (at_end || (GET_MODE (next) == TImode + && !(shadow_p (next) && CALL_P (next))))) + { + if (n_filled >= 2) + gen_one_bundle (slot, n_filled, first_slot); + + n_filled = 0; + first_slot = 0; + } + if (at_end) + break; + } + } + /* Bundling, and emitting nops, can separate + NOTE_INSN_CALL_ARG_LOCATION from the corresponding calls. Fix + that up here. */ + last_call = NULL_RTX; + for (insn = get_insns (); insn; insn = next) + { + next = NEXT_INSN (insn); + if (CALL_P (insn) + || (INSN_P (insn) && GET_CODE (PATTERN (insn)) == SEQUENCE + && CALL_P (XVECEXP (PATTERN (insn), 0, 0)))) + last_call = insn; + if (!NOTE_P (insn) || NOTE_KIND (insn) != NOTE_INSN_CALL_ARG_LOCATION) + continue; + if (NEXT_INSN (last_call) == insn) + continue; + NEXT_INSN (PREV_INSN (insn)) = NEXT_INSN (insn); + PREV_INSN (NEXT_INSN (insn)) = PREV_INSN (insn); + PREV_INSN (insn) = last_call; + NEXT_INSN (insn) = NEXT_INSN (last_call); + PREV_INSN (NEXT_INSN (insn)) = insn; + NEXT_INSN (PREV_INSN (insn)) = insn; + last_call = insn; + } +} + +/* Emit a NOP instruction for CYCLES cycles after insn AFTER. Return it. */ + +static rtx +emit_nop_after (int cycles, rtx after) +{ + rtx insn; + + /* mpydp has 9 delay slots, and we may schedule a stall for a cross-path + operation. We don't need the extra NOP since in this case, the hardware + will automatically insert the required stall. */ + if (cycles == 10) + cycles--; + + gcc_assert (cycles < 10); + + insn = emit_insn_after (gen_nop_count (GEN_INT (cycles)), after); + PUT_MODE (insn, TImode); + + return insn; +} + +/* Determine whether INSN is a call that needs to have a return label + placed. */ + +static bool +returning_call_p (rtx insn) +{ + if (CALL_P (insn)) + return (!SIBLING_CALL_P (insn) + && get_attr_type (insn) != TYPE_CALLP + && get_attr_type (insn) != TYPE_SHADOW); + if (recog_memoized (insn) < 0) + return false; + if (get_attr_type (insn) == TYPE_CALL) + return true; + return false; +} + +/* Determine whether INSN's pattern can be converted to use callp. */ +static bool +can_use_callp (rtx insn) +{ + int icode = recog_memoized (insn); + if (!TARGET_INSNS_64PLUS + || icode < 0 + || GET_CODE (PATTERN (insn)) == COND_EXEC) + return false; + + return ((icode == CODE_FOR_real_call + || icode == CODE_FOR_call_internal + || icode == CODE_FOR_call_value_internal) + && get_attr_dest_regfile (insn) == DEST_REGFILE_ANY); +} + +/* Convert the pattern of INSN, which must be a CALL_INSN, into a callp. */ +static void +convert_to_callp (rtx insn) +{ + rtx lab; + extract_insn (insn); + if (GET_CODE (PATTERN (insn)) == SET) + { + rtx dest = recog_data.operand[0]; + lab = recog_data.operand[1]; + PATTERN (insn) = gen_callp_value (dest, lab); + INSN_CODE (insn) = CODE_FOR_callp_value; + } + else + { + lab = recog_data.operand[0]; + PATTERN (insn) = gen_callp (lab); + INSN_CODE (insn) = CODE_FOR_callp; + } +} + +/* Scan forwards from INSN until we find the next insn that has mode TImode + (indicating it starts a new cycle), and occurs in cycle CLOCK. + Return it if we find such an insn, NULL_RTX otherwise. */ +static rtx +find_next_cycle_insn (rtx insn, int clock) +{ + rtx t = insn; + if (GET_MODE (t) == TImode) + t = next_real_insn (t); + while (t && GET_MODE (t) != TImode) + t = next_real_insn (t); + + if (t && insn_get_clock (t) == clock) + return t; + return NULL_RTX; +} + +/* If COND_INSN has a COND_EXEC condition, wrap the same condition + around PAT. Return PAT either unchanged or modified in this + way. */ +static rtx +duplicate_cond (rtx pat, rtx cond_insn) +{ + rtx cond_pat = PATTERN (cond_insn); + if (GET_CODE (cond_pat) == COND_EXEC) + pat = gen_rtx_COND_EXEC (VOIDmode, copy_rtx (COND_EXEC_TEST (cond_pat)), + pat); + return pat; +} + +/* Walk forward from INSN to find the last insn that issues in the same clock + cycle. */ +static rtx +find_last_same_clock (rtx insn) +{ + rtx retval = insn; + rtx t = next_real_insn (insn); + + while (t && GET_MODE (t) != TImode) + { + if (!DEBUG_INSN_P (t) && recog_memoized (t) >= 0) + retval = t; + t = next_real_insn (t); + } + return retval; +} + +/* For every call insn in the function, emit code to load the return + address. For each call we create a return label and store it in + CALL_LABELS. If are not scheduling, we emit the labels here, + otherwise the caller will do it later. + This function is called after final insn scheduling, but before creating + the SEQUENCEs that represent execute packets. */ + +static void +reorg_split_calls (rtx *call_labels) +{ + unsigned int reservation_mask = 0; + rtx insn = get_insns (); + gcc_assert (GET_CODE (insn) == NOTE); + insn = next_real_insn (insn); + while (insn) + { + int uid; + rtx next = next_real_insn (insn); + + if (DEBUG_INSN_P (insn)) + goto done; + + if (GET_MODE (insn) == TImode) + reservation_mask = 0; + uid = INSN_UID (insn); + if (c6x_flag_schedule_insns2 && recog_memoized (insn) >= 0) + reservation_mask |= 1 << INSN_INFO_ENTRY (uid).reservation; + + if (returning_call_p (insn)) + { + rtx label = gen_label_rtx (); + rtx labelref = gen_rtx_LABEL_REF (Pmode, label); + rtx reg = gen_rtx_REG (SImode, RETURN_ADDR_REGNO); + + LABEL_NUSES (label) = 2; + if (!c6x_flag_schedule_insns2) + { + if (can_use_callp (insn)) + convert_to_callp (insn); + else + { + rtx t; + rtx slot[4]; + emit_label_after (label, insn); + + /* Bundle the call and its delay slots into a single + SEQUENCE. While these do not issue in parallel + we need to group them into a single EH region. */ + slot[0] = insn; + PUT_MODE (insn, TImode); + if (TARGET_INSNS_64) + { + t = gen_addkpc (reg, labelref, GEN_INT (4)); + slot[1] = emit_insn_after (duplicate_cond (t, insn), + insn); + PUT_MODE (slot[1], TImode); + gen_one_bundle (slot, 2, 0); + } + else + { + slot[3] = emit_insn_after (gen_nop_count (GEN_INT (3)), + insn); + PUT_MODE (slot[3], TImode); + t = gen_movsi_lo_sum (reg, reg, labelref); + slot[2] = emit_insn_after (duplicate_cond (t, insn), + insn); + PUT_MODE (slot[2], TImode); + t = gen_movsi_high (reg, labelref); + slot[1] = emit_insn_after (duplicate_cond (t, insn), + insn); + PUT_MODE (slot[1], TImode); + gen_one_bundle (slot, 4, 0); + } + } + } + else + { + /* If we scheduled, we reserved the .S2 unit for one or two + cycles after the call. Emit the insns in these slots, + unless it's possible to create a CALLP insn. + Note that this works because the dependencies ensure that + no insn setting/using B3 is scheduled in the delay slots of + a call. */ + int this_clock = insn_get_clock (insn); + rtx last_same_clock; + rtx after1; + + call_labels[INSN_UID (insn)] = label; + + last_same_clock = find_last_same_clock (insn); + + if (can_use_callp (insn)) + { + /* Find the first insn of the next execute packet. If it + is outside the branch delay slots of this call, we may + use a CALLP insn. */ + rtx next_cycle_start = next_nonnote_nondebug_insn (last_same_clock); + + if (CALL_P (next_cycle_start) + && (insn_get_clock (next_cycle_start) == this_clock + 5)) + { + convert_to_callp (next_cycle_start); + insn_set_clock (next_cycle_start, this_clock); + if (GET_MODE (insn) == TImode) + { + rtx new_cycle_first = NEXT_INSN (insn); + while (!NONDEBUG_INSN_P (new_cycle_first) + || GET_CODE (PATTERN (new_cycle_first)) == USE + || GET_CODE (PATTERN (new_cycle_first)) == CLOBBER) + new_cycle_first = NEXT_INSN (new_cycle_first); + PUT_MODE (new_cycle_first, TImode); + if (new_cycle_first != next_cycle_start) + PUT_MODE (next_cycle_start, VOIDmode); + INSN_INFO_ENTRY (INSN_UID (new_cycle_first)).ebb_start + = INSN_INFO_ENTRY (INSN_UID (insn)).ebb_start; + } + else + PUT_MODE (next_cycle_start, VOIDmode); + delete_insn (insn); + goto done; + } + } + after1 = find_next_cycle_insn (last_same_clock, this_clock + 1); + if (after1 == NULL_RTX) + after1 = last_same_clock; + else + after1 = find_last_same_clock (after1); + if (TARGET_INSNS_64) + { + rtx x1 = gen_addkpc (reg, labelref, const0_rtx); + x1 = emit_insn_after (duplicate_cond (x1, insn), after1); + insn_set_clock (x1, this_clock + 1); + INSN_INFO_ENTRY (INSN_UID (x1)).reservation = RESERVATION_S2; + if (after1 == last_same_clock) + PUT_MODE (x1, TImode); + } + else + { + rtx x1, x2; + rtx after2 = find_next_cycle_insn (after1, this_clock + 2); + if (after2 == NULL_RTX) + after2 = after1; + x2 = gen_movsi_lo_sum (reg, reg, labelref); + x2 = emit_insn_after (duplicate_cond (x2, insn), after2); + x1 = gen_movsi_high (reg, labelref); + x1 = emit_insn_after (duplicate_cond (x1, insn), after1); + insn_set_clock (x1, this_clock + 1); + insn_set_clock (x2, this_clock + 2); + INSN_INFO_ENTRY (INSN_UID (x1)).reservation = RESERVATION_S2; + INSN_INFO_ENTRY (INSN_UID (x2)).reservation = RESERVATION_S2; + if (after1 == last_same_clock) + PUT_MODE (x1, TImode); + if (after1 == after2) + PUT_MODE (x2, TImode); + } + } + } + done: + insn = next; + } +} + +/* Called as part of c6x_reorg. This function emits multi-cycle NOP + insns as required for correctness. CALL_LABELS is the array that + holds the return labels for call insns; we emit these here if + scheduling was run earlier. */ + +static void +reorg_emit_nops (rtx *call_labels) +{ + bool first; + rtx prev, last_call; + int prev_clock, earliest_bb_end; + int prev_implicit_nops; + rtx insn = get_insns (); + + /* We look at one insn (or bundle inside a sequence) in each iteration, storing + its issue time in PREV_CLOCK for the next iteration. If there is a gap in + clocks, we must insert a NOP. + EARLIEST_BB_END tracks in which cycle all insns that have been issued in the + current basic block will finish. We must not allow the next basic block to + begin before this cycle. + PREV_IMPLICIT_NOPS tells us whether we've seen an insn that implicitly contains + a multi-cycle nop. The code is scheduled such that subsequent insns will + show the cycle gap, but we needn't insert a real NOP instruction. */ + insn = next_real_insn (insn); + last_call = prev = NULL_RTX; + prev_clock = -1; + earliest_bb_end = 0; + prev_implicit_nops = 0; + first = true; + while (insn) + { + int this_clock = -1; + rtx next; + int max_cycles = 0; + + next = next_real_insn (insn); + + if (DEBUG_INSN_P (insn) + || GET_CODE (PATTERN (insn)) == USE + || GET_CODE (PATTERN (insn)) == CLOBBER + || shadow_or_blockage_p (insn) + || (JUMP_P (insn) + && (GET_CODE (PATTERN (insn)) == ADDR_DIFF_VEC + || GET_CODE (PATTERN (insn)) == ADDR_VEC))) + goto next_insn; + + if (!c6x_flag_schedule_insns2) + /* No scheduling; ensure that no parallel issue happens. */ + PUT_MODE (insn, TImode); + else + { + int cycles; + + this_clock = insn_get_clock (insn); + if (this_clock != prev_clock) + { + PUT_MODE (insn, TImode); + + if (!first) + { + cycles = this_clock - prev_clock; + + cycles -= prev_implicit_nops; + if (cycles > 1) + { + rtx nop = emit_nop_after (cycles - 1, prev); + insn_set_clock (nop, prev_clock + prev_implicit_nops + 1); + } + } + prev_clock = this_clock; + + if (last_call + && insn_get_clock (last_call) + 6 <= this_clock) + { + emit_label_before (call_labels[INSN_UID (last_call)], insn); + last_call = NULL_RTX; + } + prev_implicit_nops = 0; + } + } + + /* Examine how many cycles the current insn takes, and adjust + LAST_CALL, EARLIEST_BB_END and PREV_IMPLICIT_NOPS. */ + if (recog_memoized (insn) >= 0 + /* If not scheduling, we've emitted NOPs after calls already. */ + && (c6x_flag_schedule_insns2 || !returning_call_p (insn))) + { + max_cycles = get_attr_cycles (insn); + if (get_attr_type (insn) == TYPE_CALLP) + prev_implicit_nops = 5; + } + else + max_cycles = 1; + if (returning_call_p (insn)) + last_call = insn; + + if (c6x_flag_schedule_insns2) + { + gcc_assert (this_clock >= 0); + if (earliest_bb_end < this_clock + max_cycles) + earliest_bb_end = this_clock + max_cycles; + } + else if (max_cycles > 1) + emit_nop_after (max_cycles - 1, insn); + + prev = insn; + first = false; + + next_insn: + if (c6x_flag_schedule_insns2 + && (next == NULL_RTX + || (GET_MODE (next) == TImode + && INSN_INFO_ENTRY (INSN_UID (next)).ebb_start)) + && earliest_bb_end > 0) + { + int cycles = earliest_bb_end - prev_clock; + if (cycles > 1) + { + prev = emit_nop_after (cycles - 1, prev); + insn_set_clock (prev, prev_clock + prev_implicit_nops + 1); + } + earliest_bb_end = 0; + prev_clock = -1; + first = true; + + if (last_call) + emit_label_after (call_labels[INSN_UID (last_call)], prev); + last_call = NULL_RTX; + } + insn = next; + } +} + +/* If possible, split INSN, which we know is either a jump or a call, into a real + insn and its shadow. */ +static void +split_delayed_branch (rtx insn) +{ + int code = recog_memoized (insn); + rtx i1, newpat; + rtx pat = PATTERN (insn); + + if (GET_CODE (pat) == COND_EXEC) + pat = COND_EXEC_CODE (pat); + + if (CALL_P (insn)) + { + rtx src = pat, dest = NULL_RTX; + rtx callee; + if (GET_CODE (pat) == SET) + { + dest = SET_DEST (pat); + src = SET_SRC (pat); + } + callee = XEXP (XEXP (src, 0), 0); + if (SIBLING_CALL_P (insn)) + { + if (REG_P (callee)) + newpat = gen_indirect_sibcall_shadow (); + else + newpat = gen_sibcall_shadow (callee); + pat = gen_real_jump (callee); + } + else if (dest != NULL_RTX) + { + if (REG_P (callee)) + newpat = gen_indirect_call_value_shadow (dest); + else + newpat = gen_call_value_shadow (dest, callee); + pat = gen_real_call (callee); + } + else + { + if (REG_P (callee)) + newpat = gen_indirect_call_shadow (); + else + newpat = gen_call_shadow (callee); + pat = gen_real_call (callee); + } + pat = duplicate_cond (pat, insn); + newpat = duplicate_cond (newpat, insn); + } + else + { + rtx src, op; + if (GET_CODE (pat) == PARALLEL + && GET_CODE (XVECEXP (pat, 0, 0)) == RETURN) + { + newpat = gen_return_shadow (); + pat = gen_real_ret (XEXP (XVECEXP (pat, 0, 1), 0)); + newpat = duplicate_cond (newpat, insn); + } + else + switch (code) + { + case CODE_FOR_br_true: + case CODE_FOR_br_false: + src = SET_SRC (pat); + op = XEXP (src, code == CODE_FOR_br_true ? 1 : 2); + newpat = gen_condjump_shadow (op); + pat = gen_real_jump (op); + if (code == CODE_FOR_br_true) + pat = gen_rtx_COND_EXEC (VOIDmode, XEXP (src, 0), pat); + else + pat = gen_rtx_COND_EXEC (VOIDmode, + reversed_comparison (XEXP (src, 0), + VOIDmode), + pat); + break; + + case CODE_FOR_jump: + op = SET_SRC (pat); + newpat = gen_jump_shadow (op); + break; + + case CODE_FOR_indirect_jump: + newpat = gen_indirect_jump_shadow (); + break; + + case CODE_FOR_return_internal: + newpat = gen_return_shadow (); + pat = gen_real_ret (XEXP (XVECEXP (pat, 0, 1), 0)); + break; + + default: + return; + } + } + i1 = emit_insn_before (pat, insn); + PATTERN (insn) = newpat; + INSN_CODE (insn) = -1; + record_delay_slot_pair (i1, insn, 5); +} + +/* Split every insn (i.e. jumps and calls) which can have delay slots into + two parts: the first one is scheduled normally and emits the instruction, + while the second one is a shadow insn which shows the side effect taking + place. The second one is placed in the right cycle by the scheduler, but + not emitted as an assembly instruction. */ + +static void +split_delayed_insns (void) +{ + rtx insn; + for (insn = get_insns (); insn; insn = NEXT_INSN (insn)) + { + if (JUMP_P (insn) || CALL_P (insn)) + split_delayed_branch (insn); + } +} + +/* For every insn that has an entry in the new_conditions vector, give it + the appropriate predicate. */ +static void +conditionalize_after_sched (void) +{ + basic_block bb; + rtx insn; + FOR_EACH_BB (bb) + FOR_BB_INSNS (bb, insn) + { + unsigned uid = INSN_UID (insn); + rtx cond; + if (!NONDEBUG_INSN_P (insn) || uid >= INSN_INFO_LENGTH) + continue; + cond = INSN_INFO_ENTRY (uid).new_cond; + if (cond == NULL_RTX) + continue; + if (dump_file) + fprintf (dump_file, "Conditionalizing insn %d\n", uid); + predicate_insn (insn, cond, true); + } +} + +/* Implement the TARGET_MACHINE_DEPENDENT_REORG pass. We split call insns here + into a sequence that loads the return register and performs the call, + and emit the return label. + If scheduling after reload is requested, it happens here. */ + +static void +c6x_reorg (void) +{ + basic_block bb; + rtx *call_labels; + bool do_selsched = (c6x_flag_schedule_insns2 && flag_selective_scheduling2 + && !maybe_skip_selective_scheduling ()); + + /* We are freeing block_for_insn in the toplev to keep compatibility + with old MDEP_REORGS that are not CFG based. Recompute it now. */ + compute_bb_for_insn (); + + df_clear_flags (DF_LR_RUN_DCE); + + /* If optimizing, we'll have split before scheduling. */ + if (optimize == 0) + split_all_insns (); + + if (c6x_flag_schedule_insns2) + { + int sz = get_max_uid () * 3 / 2 + 1; + + insn_info = VEC_alloc (c6x_sched_insn_info, heap, sz); + + /* Make sure the real-jump insns we create are not deleted. */ + sched_no_dce = true; + + split_delayed_insns (); + timevar_push (TV_SCHED2); + if (do_selsched) + run_selective_scheduling (); + else + schedule_ebbs (); + conditionalize_after_sched (); + timevar_pop (TV_SCHED2); + + free_delay_pairs (); + sched_no_dce = false; + } + + call_labels = XCNEWVEC (rtx, get_max_uid () + 1); + + reorg_split_calls (call_labels); + + if (c6x_flag_schedule_insns2) + { + FOR_EACH_BB (bb) + if ((bb->flags & BB_DISABLE_SCHEDULE) == 0) + assign_reservations (BB_HEAD (bb), BB_END (bb)); + } + + if (c6x_flag_var_tracking) + { + timevar_push (TV_VAR_TRACKING); + variable_tracking_main (); + timevar_pop (TV_VAR_TRACKING); + } + + reorg_emit_nops (call_labels); + + /* Post-process the schedule to move parallel insns into SEQUENCEs. */ + if (c6x_flag_schedule_insns2) + { + free_delay_pairs (); + c6x_gen_bundles (); + } + + df_finish_pass (false); +} + +/* Called when a function has been assembled. It should perform all the + tasks of ASM_DECLARE_FUNCTION_SIZE in elfos.h, plus target-specific + tasks. + We free the reservation (and other scheduling) information here now that + all insns have been output. */ +void +c6x_function_end (FILE *file, const char *fname) +{ + c6x_output_fn_unwind (file); + + if (insn_info) + VEC_free (c6x_sched_insn_info, heap, insn_info); + insn_info = NULL; + + if (!flag_inhibit_size_directive) + ASM_OUTPUT_MEASURED_SIZE (file, fname); +} + +/* Determine whether X is a shift with code CODE and an integer amount + AMOUNT. */ +static bool +shift_p (rtx x, enum rtx_code code, int amount) +{ + return (GET_CODE (x) == code && GET_CODE (XEXP (x, 1)) == CONST_INT + && INTVAL (XEXP (x, 1)) == amount); +} + +/* Compute a (partial) cost for rtx X. Return true if the complete + cost has been computed, and false if subexpressions should be + scanned. In either case, *TOTAL contains the cost result. */ + +static bool +c6x_rtx_costs (rtx x, int code, int outer_code, int *total, bool speed) +{ + int cost2 = COSTS_N_INSNS (1); + rtx op0, op1; + + switch (code) + { + case CONST_INT: + if (outer_code == SET || outer_code == PLUS) + *total = satisfies_constraint_IsB (x) ? 0 : cost2; + else if (outer_code == AND || outer_code == IOR || outer_code == XOR + || outer_code == MINUS) + *total = satisfies_constraint_Is5 (x) ? 0 : cost2; + else if (GET_RTX_CLASS (outer_code) == RTX_COMPARE + || GET_RTX_CLASS (outer_code) == RTX_COMM_COMPARE) + *total = satisfies_constraint_Iu4 (x) ? 0 : cost2; + else if (outer_code == ASHIFT || outer_code == ASHIFTRT + || outer_code == LSHIFTRT) + *total = satisfies_constraint_Iu5 (x) ? 0 : cost2; + else + *total = cost2; + return true; + + case CONST: + case LABEL_REF: + case SYMBOL_REF: + case CONST_DOUBLE: + *total = COSTS_N_INSNS (2); + return true; + + case TRUNCATE: + /* Recognize a mult_highpart operation. */ + if ((GET_MODE (x) == HImode || GET_MODE (x) == SImode) + && GET_CODE (XEXP (x, 0)) == LSHIFTRT + && GET_MODE (XEXP (x, 0)) == GET_MODE_2XWIDER_MODE (GET_MODE (x)) + && GET_CODE (XEXP (XEXP (x, 0), 0)) == MULT + && GET_CODE (XEXP (XEXP (x, 0), 1)) == CONST_INT + && INTVAL (XEXP (XEXP (x, 0), 1)) == GET_MODE_BITSIZE (GET_MODE (x))) + { + rtx mul = XEXP (XEXP (x, 0), 0); + rtx op0 = XEXP (mul, 0); + rtx op1 = XEXP (mul, 1); + enum rtx_code code0 = GET_CODE (op0); + enum rtx_code code1 = GET_CODE (op1); + + if ((code0 == code1 + && (code0 == SIGN_EXTEND || code0 == ZERO_EXTEND)) + || (GET_MODE (x) == HImode + && code0 == ZERO_EXTEND && code1 == SIGN_EXTEND)) + { + if (GET_MODE (x) == HImode) + *total = COSTS_N_INSNS (2); + else + *total = COSTS_N_INSNS (12); + *total += rtx_cost (XEXP (op0, 0), code0, speed); + *total += rtx_cost (XEXP (op1, 0), code1, speed); + return true; + } + } + return false; + + case ASHIFT: + case ASHIFTRT: + case LSHIFTRT: + if (GET_MODE (x) == DImode) + *total = COSTS_N_INSNS (CONSTANT_P (XEXP (x, 1)) ? 4 : 15); + else + *total = COSTS_N_INSNS (1); + return false; + + case PLUS: + case MINUS: + *total = COSTS_N_INSNS (1); + op0 = code == PLUS ? XEXP (x, 0) : XEXP (x, 1); + op1 = code == PLUS ? XEXP (x, 1) : XEXP (x, 0); + if (GET_MODE_SIZE (GET_MODE (x)) <= UNITS_PER_WORD + && INTEGRAL_MODE_P (GET_MODE (x)) + && GET_CODE (op0) == MULT + && GET_CODE (XEXP (op0, 1)) == CONST_INT + && (INTVAL (XEXP (op0, 1)) == 2 + || INTVAL (XEXP (op0, 1)) == 4 + || (code == PLUS && INTVAL (XEXP (op0, 1)) == 8))) + { + *total += rtx_cost (XEXP (op0, 0), ASHIFT, speed); + *total += rtx_cost (op1, (enum rtx_code)code, speed); + return true; + } + return false; + + case MULT: + op0 = XEXP (x, 0); + op1 = XEXP (x, 1); + if (GET_MODE (x) == DFmode) + { + if (TARGET_FP) + *total = COSTS_N_INSNS (speed ? 10 : 1); + else + *total = COSTS_N_INSNS (speed ? 200 : 4); + } + else if (GET_MODE (x) == SFmode) + { + if (TARGET_FP) + *total = COSTS_N_INSNS (speed ? 4 : 1); + else + *total = COSTS_N_INSNS (speed ? 100 : 4); + } + else if (GET_MODE (x) == DImode) + { + if (TARGET_MPY32 + && GET_CODE (op0) == GET_CODE (op1) + && (GET_CODE (op0) == ZERO_EXTEND + || GET_CODE (op0) == SIGN_EXTEND)) + { + *total = COSTS_N_INSNS (speed ? 2 : 1); + op0 = XEXP (op0, 0); + op1 = XEXP (op1, 0); + } + else + /* Maybe improve this laster. */ + *total = COSTS_N_INSNS (20); + } + else if (GET_MODE (x) == SImode) + { + if (((GET_CODE (op0) == ZERO_EXTEND + || GET_CODE (op0) == SIGN_EXTEND + || shift_p (op0, LSHIFTRT, 16)) + && (GET_CODE (op1) == SIGN_EXTEND + || GET_CODE (op1) == ZERO_EXTEND + || scst5_operand (op1, SImode) + || shift_p (op1, ASHIFTRT, 16) + || shift_p (op1, LSHIFTRT, 16))) + || (shift_p (op0, ASHIFTRT, 16) + && (GET_CODE (op1) == SIGN_EXTEND + || shift_p (op1, ASHIFTRT, 16)))) + { + *total = COSTS_N_INSNS (speed ? 2 : 1); + op0 = XEXP (op0, 0); + if (scst5_operand (op1, SImode)) + op1 = NULL_RTX; + else + op1 = XEXP (op1, 0); + } + else if (!speed) + *total = COSTS_N_INSNS (1); + else if (TARGET_MPY32) + *total = COSTS_N_INSNS (4); + else + *total = COSTS_N_INSNS (6); + } + else if (GET_MODE (x) == HImode) + *total = COSTS_N_INSNS (speed ? 2 : 1); + + if (GET_CODE (op0) != REG + && (GET_CODE (op0) != SUBREG || GET_CODE (SUBREG_REG (op0)) != REG)) + *total += rtx_cost (op0, MULT, speed); + if (op1 && GET_CODE (op1) != REG + && (GET_CODE (op1) != SUBREG || GET_CODE (SUBREG_REG (op1)) != REG)) + *total += rtx_cost (op1, MULT, speed); + return true; + + case UDIV: + case DIV: + /* This is a bit random; assuming on average there'll be 16 leading + zeros. FIXME: estimate better for constant dividends. */ + *total = COSTS_N_INSNS (6 + 3 * 16); + return false; + + case IF_THEN_ELSE: + /* Recognize the cmp_and/ior patterns. */ + op0 = XEXP (x, 0); + if ((GET_CODE (op0) == EQ || GET_CODE (op0) == NE) + && REG_P (XEXP (op0, 0)) + && XEXP (op0, 1) == const0_rtx + && rtx_equal_p (XEXP (x, 1), XEXP (op0, 0))) + { + *total = rtx_cost (XEXP (x, 1), (enum rtx_code)outer_code, speed); + return false; + } + return false; + + default: + return false; + } +} + +/* Implements target hook vector_mode_supported_p. */ + +static bool +c6x_vector_mode_supported_p (enum machine_mode mode) +{ + switch (mode) + { + case V2HImode: + case V4QImode: + case V2SImode: + case V4HImode: + case V8QImode: + return true; + default: + return false; + } +} + +/* Implements TARGET_VECTORIZE_PREFERRED_SIMD_MODE. */ +static enum machine_mode +c6x_preferred_simd_mode (enum machine_mode mode) +{ + switch (mode) + { + case HImode: + return V2HImode; + case QImode: + return V4QImode; + + default: + return word_mode; + } +} + +/* Implement TARGET_SCALAR_MODE_SUPPORTED_P. */ + +static bool +c6x_scalar_mode_supported_p (enum machine_mode mode) +{ + if (ALL_FIXED_POINT_MODE_P (mode) + && GET_MODE_PRECISION (mode) <= 2 * BITS_PER_WORD) + return true; + + return default_scalar_mode_supported_p (mode); +} + +/* Output a reference from a function exception table to the type_info + object X. Output these via a special assembly directive. */ + +static bool +c6x_output_ttype (rtx x) +{ + /* Use special relocations for symbol references. */ + if (GET_CODE (x) != CONST_INT) + fputs ("\t.ehtype\t", asm_out_file); + else + fputs ("\t.word\t", asm_out_file); + output_addr_const (asm_out_file, x); + fputc ('\n', asm_out_file); + + return TRUE; +} + +/* Modify the return address of the current function. */ + +void +c6x_set_return_address (rtx source, rtx scratch) +{ + struct c6x_frame frame; + rtx addr; + HOST_WIDE_INT offset; + + c6x_compute_frame_layout (&frame); + if (! c6x_save_reg (RETURN_ADDR_REGNO)) + emit_move_insn (gen_rtx_REG (Pmode, RETURN_ADDR_REGNO), source); + else + { + + if (frame_pointer_needed) + { + addr = hard_frame_pointer_rtx; + offset = frame.b3_offset; + } + else + { + addr = stack_pointer_rtx; + offset = frame.to_allocate - frame.b3_offset; + } + + /* TODO: Use base+offset loads where possible. */ + if (offset) + { + HOST_WIDE_INT low = trunc_int_for_mode (offset, HImode); + + emit_insn (gen_movsi_high (scratch, GEN_INT (low))); + if (low != offset) + emit_insn (gen_movsi_lo_sum (scratch, scratch, GEN_INT(offset))); + emit_insn (gen_addsi3 (scratch, addr, scratch)); + addr = scratch; + } + + emit_move_insn (gen_frame_mem (Pmode, addr), source); + } +} + +/* We save pairs of registers using a DImode store. Describe the component + registers for DWARF generation code. */ + +static rtx +c6x_dwarf_register_span (rtx rtl) +{ + unsigned regno; + unsigned real_regno; + int nregs; + int i; + rtx p; + + regno = REGNO (rtl); + nregs = HARD_REGNO_NREGS (regno, GET_MODE (rtl)); + if (nregs == 1) + return NULL_RTX; + + p = gen_rtx_PARALLEL (VOIDmode, rtvec_alloc(nregs)); + for (i = 0; i < nregs; i++) + { + if (TARGET_BIG_ENDIAN) + real_regno = regno + nregs - (i + 1); + else + real_regno = regno + i; + + XVECEXP (p, 0, i) = gen_rtx_REG (SImode, real_regno); + } + + return p; +} + +/* Codes for all the C6X builtins. */ +enum c6x_builtins +{ + C6X_BUILTIN_SADD, + C6X_BUILTIN_SSUB, + C6X_BUILTIN_ADD2, + C6X_BUILTIN_SUB2, + C6X_BUILTIN_ADD4, + C6X_BUILTIN_SUB4, + C6X_BUILTIN_SADD2, + C6X_BUILTIN_SSUB2, + C6X_BUILTIN_SADDU4, + + C6X_BUILTIN_SMPY, + C6X_BUILTIN_SMPYH, + C6X_BUILTIN_SMPYHL, + C6X_BUILTIN_SMPYLH, + C6X_BUILTIN_MPY2, + C6X_BUILTIN_SMPY2, + + C6X_BUILTIN_CLRR, + C6X_BUILTIN_EXTR, + C6X_BUILTIN_EXTRU, + + C6X_BUILTIN_SSHL, + C6X_BUILTIN_SUBC, + C6X_BUILTIN_ABS, + C6X_BUILTIN_ABS2, + C6X_BUILTIN_AVG2, + C6X_BUILTIN_AVGU4, + + C6X_BUILTIN_MAX +}; + + +static GTY(()) tree c6x_builtin_decls[C6X_BUILTIN_MAX]; + +/* Return the C6X builtin for CODE. */ +static tree +c6x_builtin_decl (unsigned code, bool initialize_p ATTRIBUTE_UNUSED) +{ + if (code >= C6X_BUILTIN_MAX) + return error_mark_node; + + return c6x_builtin_decls[code]; +} + +#define def_builtin(NAME, TYPE, CODE) \ +do { \ + tree bdecl; \ + bdecl = add_builtin_function ((NAME), (TYPE), (CODE), BUILT_IN_MD, \ + NULL, NULL_TREE); \ + c6x_builtin_decls[CODE] = bdecl; \ +} while (0) + +/* Set up all builtin functions for this target. */ +static void +c6x_init_builtins (void) +{ + tree V4QI_type_node = build_vector_type (unsigned_intQI_type_node, 4); + tree V2HI_type_node = build_vector_type (intHI_type_node, 2); + tree V2SI_type_node = build_vector_type (intSI_type_node, 2); + tree int_ftype_int + = build_function_type_list (integer_type_node, integer_type_node, + NULL_TREE); + tree int_ftype_int_int + = build_function_type_list (integer_type_node, integer_type_node, + integer_type_node, NULL_TREE); + tree v2hi_ftype_v2hi + = build_function_type_list (V2HI_type_node, V2HI_type_node, NULL_TREE); + tree v4qi_ftype_v4qi_v4qi + = build_function_type_list (V4QI_type_node, V4QI_type_node, + V4QI_type_node, NULL_TREE); + tree v2hi_ftype_v2hi_v2hi + = build_function_type_list (V2HI_type_node, V2HI_type_node, + V2HI_type_node, NULL_TREE); + tree v2si_ftype_v2hi_v2hi + = build_function_type_list (V2SI_type_node, V2HI_type_node, + V2HI_type_node, NULL_TREE); + + def_builtin ("__builtin_c6x_sadd", int_ftype_int_int, + C6X_BUILTIN_SADD); + def_builtin ("__builtin_c6x_ssub", int_ftype_int_int, + C6X_BUILTIN_SSUB); + def_builtin ("__builtin_c6x_add2", v2hi_ftype_v2hi_v2hi, + C6X_BUILTIN_ADD2); + def_builtin ("__builtin_c6x_sub2", v2hi_ftype_v2hi_v2hi, + C6X_BUILTIN_SUB2); + def_builtin ("__builtin_c6x_add4", v4qi_ftype_v4qi_v4qi, + C6X_BUILTIN_ADD4); + def_builtin ("__builtin_c6x_sub4", v4qi_ftype_v4qi_v4qi, + C6X_BUILTIN_SUB4); + def_builtin ("__builtin_c6x_mpy2", v2si_ftype_v2hi_v2hi, + C6X_BUILTIN_MPY2); + def_builtin ("__builtin_c6x_sadd2", v2hi_ftype_v2hi_v2hi, + C6X_BUILTIN_SADD2); + def_builtin ("__builtin_c6x_ssub2", v2hi_ftype_v2hi_v2hi, + C6X_BUILTIN_SSUB2); + def_builtin ("__builtin_c6x_saddu4", v4qi_ftype_v4qi_v4qi, + C6X_BUILTIN_SADDU4); + def_builtin ("__builtin_c6x_smpy2", v2si_ftype_v2hi_v2hi, + C6X_BUILTIN_SMPY2); + + def_builtin ("__builtin_c6x_smpy", int_ftype_int_int, + C6X_BUILTIN_SMPY); + def_builtin ("__builtin_c6x_smpyh", int_ftype_int_int, + C6X_BUILTIN_SMPYH); + def_builtin ("__builtin_c6x_smpyhl", int_ftype_int_int, + C6X_BUILTIN_SMPYHL); + def_builtin ("__builtin_c6x_smpylh", int_ftype_int_int, + C6X_BUILTIN_SMPYLH); + + def_builtin ("__builtin_c6x_sshl", int_ftype_int_int, + C6X_BUILTIN_SSHL); + def_builtin ("__builtin_c6x_subc", int_ftype_int_int, + C6X_BUILTIN_SUBC); + + def_builtin ("__builtin_c6x_avg2", v2hi_ftype_v2hi_v2hi, + C6X_BUILTIN_AVG2); + def_builtin ("__builtin_c6x_avgu4", v4qi_ftype_v4qi_v4qi, + C6X_BUILTIN_AVGU4); + + def_builtin ("__builtin_c6x_clrr", int_ftype_int_int, + C6X_BUILTIN_CLRR); + def_builtin ("__builtin_c6x_extr", int_ftype_int_int, + C6X_BUILTIN_EXTR); + def_builtin ("__builtin_c6x_extru", int_ftype_int_int, + C6X_BUILTIN_EXTRU); + + def_builtin ("__builtin_c6x_abs", int_ftype_int, C6X_BUILTIN_ABS); + def_builtin ("__builtin_c6x_abs2", v2hi_ftype_v2hi, C6X_BUILTIN_ABS2); +} + + +struct builtin_description +{ + const enum insn_code icode; + const char *const name; + const enum c6x_builtins code; +}; + +static const struct builtin_description bdesc_2arg[] = +{ + { CODE_FOR_saddsi3, "__builtin_c6x_sadd", C6X_BUILTIN_SADD }, + { CODE_FOR_ssubsi3, "__builtin_c6x_ssub", C6X_BUILTIN_SSUB }, + { CODE_FOR_addv2hi3, "__builtin_c6x_add2", C6X_BUILTIN_ADD2 }, + { CODE_FOR_subv2hi3, "__builtin_c6x_sub2", C6X_BUILTIN_SUB2 }, + { CODE_FOR_addv4qi3, "__builtin_c6x_add4", C6X_BUILTIN_ADD4 }, + { CODE_FOR_subv4qi3, "__builtin_c6x_sub4", C6X_BUILTIN_SUB4 }, + { CODE_FOR_ss_addv2hi3, "__builtin_c6x_sadd2", C6X_BUILTIN_SADD2 }, + { CODE_FOR_ss_subv2hi3, "__builtin_c6x_ssub2", C6X_BUILTIN_SSUB2 }, + { CODE_FOR_us_addv4qi3, "__builtin_c6x_saddu4", C6X_BUILTIN_SADDU4 }, + + { CODE_FOR_subcsi3, "__builtin_c6x_subc", C6X_BUILTIN_SUBC }, + { CODE_FOR_ss_ashlsi3, "__builtin_c6x_sshl", C6X_BUILTIN_SSHL }, + + { CODE_FOR_avgv2hi3, "__builtin_c6x_avg2", C6X_BUILTIN_AVG2 }, + { CODE_FOR_uavgv4qi3, "__builtin_c6x_avgu4", C6X_BUILTIN_AVGU4 }, + + { CODE_FOR_mulhqsq3, "__builtin_c6x_smpy", C6X_BUILTIN_SMPY }, + { CODE_FOR_mulhqsq3_hh, "__builtin_c6x_smpyh", C6X_BUILTIN_SMPYH }, + { CODE_FOR_mulhqsq3_lh, "__builtin_c6x_smpylh", C6X_BUILTIN_SMPYLH }, + { CODE_FOR_mulhqsq3_hl, "__builtin_c6x_smpyhl", C6X_BUILTIN_SMPYHL }, + + { CODE_FOR_mulv2hqv2sq3, "__builtin_c6x_smpy2", C6X_BUILTIN_SMPY2 }, + + { CODE_FOR_clrr, "__builtin_c6x_clrr", C6X_BUILTIN_CLRR }, + { CODE_FOR_extr, "__builtin_c6x_extr", C6X_BUILTIN_EXTR }, + { CODE_FOR_extru, "__builtin_c6x_extru", C6X_BUILTIN_EXTRU } +}; + +static const struct builtin_description bdesc_1arg[] = +{ + { CODE_FOR_ssabssi2, "__builtin_c6x_abs", C6X_BUILTIN_ABS }, + { CODE_FOR_ssabsv2hi2, "__builtin_c6x_abs2", C6X_BUILTIN_ABS2 } +}; + +/* Errors in the source file can cause expand_expr to return const0_rtx + where we expect a vector. To avoid crashing, use one of the vector + clear instructions. */ +static rtx +safe_vector_operand (rtx x, enum machine_mode mode) +{ + if (x != const0_rtx) + return x; + x = gen_reg_rtx (SImode); + + emit_insn (gen_movsi (x, CONST0_RTX (SImode))); + return gen_lowpart (mode, x); +} + +/* Subroutine of c6x_expand_builtin to take care of binop insns. MACFLAG is -1 + if this is a normal binary op, or one of the MACFLAG_xxx constants. */ + +static rtx +c6x_expand_binop_builtin (enum insn_code icode, tree exp, rtx target, + bool match_op) +{ + int offs = match_op ? 1 : 0; + rtx pat; + tree arg0 = CALL_EXPR_ARG (exp, 0); + tree arg1 = CALL_EXPR_ARG (exp, 1); + rtx op0 = expand_expr (arg0, NULL_RTX, VOIDmode, EXPAND_NORMAL); + rtx op1 = expand_expr (arg1, NULL_RTX, VOIDmode, EXPAND_NORMAL); + enum machine_mode op0mode = GET_MODE (op0); + enum machine_mode op1mode = GET_MODE (op1); + enum machine_mode tmode = insn_data[icode].operand[0].mode; + enum machine_mode mode0 = insn_data[icode].operand[1 + offs].mode; + enum machine_mode mode1 = insn_data[icode].operand[2 + offs].mode; + rtx ret = target; + + if (VECTOR_MODE_P (mode0)) + op0 = safe_vector_operand (op0, mode0); + if (VECTOR_MODE_P (mode1)) + op1 = safe_vector_operand (op1, mode1); + + if (! target + || GET_MODE (target) != tmode + || ! (*insn_data[icode].operand[0].predicate) (target, tmode)) + { + if (tmode == SQmode || tmode == V2SQmode) + { + ret = gen_reg_rtx (tmode == SQmode ? SImode : V2SImode); + target = gen_lowpart (tmode, ret); + } + else + target = gen_reg_rtx (tmode); + } + + if ((op0mode == V2HImode || op0mode == SImode || op0mode == VOIDmode) + && (mode0 == V2HQmode || mode0 == HQmode || mode0 == SQmode)) + { + op0mode = mode0; + op0 = gen_lowpart (mode0, op0); + } + if ((op1mode == V2HImode || op1mode == SImode || op1mode == VOIDmode) + && (mode1 == V2HQmode || mode1 == HQmode || mode1 == SQmode)) + { + op1mode = mode1; + op1 = gen_lowpart (mode1, op1); + } + /* In case the insn wants input operands in modes different from + the result, abort. */ + gcc_assert ((op0mode == mode0 || op0mode == VOIDmode) + && (op1mode == mode1 || op1mode == VOIDmode)); + + if (! (*insn_data[icode].operand[1 + offs].predicate) (op0, mode0)) + op0 = copy_to_mode_reg (mode0, op0); + if (! (*insn_data[icode].operand[2 + offs].predicate) (op1, mode1)) + op1 = copy_to_mode_reg (mode1, op1); + + if (match_op) + pat = GEN_FCN (icode) (target, target, op0, op1); + else + pat = GEN_FCN (icode) (target, op0, op1); + + if (! pat) + return 0; + + emit_insn (pat); + + return ret; +} + +/* Subroutine of c6x_expand_builtin to take care of unop insns. */ + +static rtx +c6x_expand_unop_builtin (enum insn_code icode, tree exp, + rtx target) +{ + rtx pat; + tree arg0 = CALL_EXPR_ARG (exp, 0); + rtx op0 = expand_expr (arg0, NULL_RTX, VOIDmode, EXPAND_NORMAL); + enum machine_mode op0mode = GET_MODE (op0); + enum machine_mode tmode = insn_data[icode].operand[0].mode; + enum machine_mode mode0 = insn_data[icode].operand[1].mode; + + if (! target + || GET_MODE (target) != tmode + || ! (*insn_data[icode].operand[0].predicate) (target, tmode)) + target = gen_reg_rtx (tmode); + + if (VECTOR_MODE_P (mode0)) + op0 = safe_vector_operand (op0, mode0); + + if (op0mode == SImode && mode0 == HImode) + { + op0mode = HImode; + op0 = gen_lowpart (HImode, op0); + } + gcc_assert (op0mode == mode0 || op0mode == VOIDmode); + + if (! (*insn_data[icode].operand[1].predicate) (op0, mode0)) + op0 = copy_to_mode_reg (mode0, op0); + + pat = GEN_FCN (icode) (target, op0); + if (! pat) + return 0; + emit_insn (pat); + return target; +} + +/* Expand an expression EXP that calls a built-in function, + with result going to TARGET if that's convenient + (and in mode MODE if that's convenient). + SUBTARGET may be used as the target for computing one of EXP's operands. + IGNORE is nonzero if the value is to be ignored. */ + +static rtx +c6x_expand_builtin (tree exp, rtx target ATTRIBUTE_UNUSED, + rtx subtarget ATTRIBUTE_UNUSED, + enum machine_mode mode ATTRIBUTE_UNUSED, + int ignore ATTRIBUTE_UNUSED) +{ + size_t i; + const struct builtin_description *d; + tree fndecl = TREE_OPERAND (CALL_EXPR_FN (exp), 0); + unsigned int fcode = DECL_FUNCTION_CODE (fndecl); + + for (i = 0, d = bdesc_2arg; i < ARRAY_SIZE (bdesc_2arg); i++, d++) + if (d->code == fcode) + return c6x_expand_binop_builtin (d->icode, exp, target, + fcode == C6X_BUILTIN_CLRR); + + for (i = 0, d = bdesc_1arg; i < ARRAY_SIZE (bdesc_1arg); i++, d++) + if (d->code == fcode) + return c6x_expand_unop_builtin (d->icode, exp, target); + + gcc_unreachable (); +} + +/* Target Structure. */ + +/* Initialize the GCC target structure. */ +#undef TARGET_FUNCTION_ARG +#define TARGET_FUNCTION_ARG c6x_function_arg +#undef TARGET_FUNCTION_ARG_ADVANCE +#define TARGET_FUNCTION_ARG_ADVANCE c6x_function_arg_advance +#undef TARGET_FUNCTION_ARG_BOUNDARY +#define TARGET_FUNCTION_ARG_BOUNDARY c6x_function_arg_boundary +#undef TARGET_FUNCTION_ARG_ROUND_BOUNDARY +#define TARGET_FUNCTION_ARG_ROUND_BOUNDARY \ + c6x_function_arg_round_boundary +#undef TARGET_FUNCTION_VALUE_REGNO_P +#define TARGET_FUNCTION_VALUE_REGNO_P c6x_function_value_regno_p +#undef TARGET_FUNCTION_VALUE +#define TARGET_FUNCTION_VALUE c6x_function_value +#undef TARGET_LIBCALL_VALUE +#define TARGET_LIBCALL_VALUE c6x_libcall_value +#undef TARGET_RETURN_IN_MEMORY +#define TARGET_RETURN_IN_MEMORY c6x_return_in_memory +#undef TARGET_RETURN_IN_MSB +#define TARGET_RETURN_IN_MSB c6x_return_in_msb +#undef TARGET_PASS_BY_REFERENCE +#define TARGET_PASS_BY_REFERENCE c6x_pass_by_reference +#undef TARGET_CALLEE_COPIES +#define TARGET_CALLEE_COPIES c6x_callee_copies +#undef TARGET_STRUCT_VALUE_RTX +#define TARGET_STRUCT_VALUE_RTX c6x_struct_value_rtx +#undef TARGET_FUNCTION_OK_FOR_SIBCALL +#define TARGET_FUNCTION_OK_FOR_SIBCALL c6x_function_ok_for_sibcall + +#undef TARGET_ASM_OUTPUT_MI_THUNK +#define TARGET_ASM_OUTPUT_MI_THUNK c6x_output_mi_thunk +#undef TARGET_ASM_CAN_OUTPUT_MI_THUNK +#define TARGET_ASM_CAN_OUTPUT_MI_THUNK c6x_can_output_mi_thunk + +#undef TARGET_BUILD_BUILTIN_VA_LIST +#define TARGET_BUILD_BUILTIN_VA_LIST c6x_build_builtin_va_list + +#undef TARGET_ASM_TRAMPOLINE_TEMPLATE +#define TARGET_ASM_TRAMPOLINE_TEMPLATE c6x_asm_trampoline_template +#undef TARGET_TRAMPOLINE_INIT +#define TARGET_TRAMPOLINE_INIT c6x_initialize_trampoline + +#undef TARGET_LEGITIMATE_CONSTANT_P +#define TARGET_LEGITIMATE_CONSTANT_P c6x_legitimate_constant_p +#undef TARGET_LEGITIMATE_ADDRESS_P +#define TARGET_LEGITIMATE_ADDRESS_P c6x_legitimate_address_p + +#undef TARGET_IN_SMALL_DATA_P +#define TARGET_IN_SMALL_DATA_P c6x_in_small_data_p +#undef TARGET_ASM_SELECT_RTX_SECTION +#define TARGET_ASM_SELECT_RTX_SECTION c6x_select_rtx_section +#undef TARGET_ASM_SELECT_SECTION +#define TARGET_ASM_SELECT_SECTION c6x_elf_select_section +#undef TARGET_ASM_UNIQUE_SECTION +#define TARGET_ASM_UNIQUE_SECTION c6x_elf_unique_section +#undef TARGET_SECTION_TYPE_FLAGS +#define TARGET_SECTION_TYPE_FLAGS c6x_section_type_flags +#undef TARGET_HAVE_SRODATA_SECTION +#define TARGET_HAVE_SRODATA_SECTION true +#undef TARGET_ASM_MERGEABLE_RODATA_PREFIX +#define TARGET_ASM_MERGEABLE_RODATA_PREFIX ".const" + +#undef TARGET_OPTION_OVERRIDE +#define TARGET_OPTION_OVERRIDE c6x_option_override +#undef TARGET_CONDITIONAL_REGISTER_USAGE +#define TARGET_CONDITIONAL_REGISTER_USAGE c6x_conditional_register_usage + +#undef TARGET_INIT_LIBFUNCS +#define TARGET_INIT_LIBFUNCS c6x_init_libfuncs +#undef TARGET_LIBFUNC_GNU_PREFIX +#define TARGET_LIBFUNC_GNU_PREFIX true + +#undef TARGET_SCALAR_MODE_SUPPORTED_P +#define TARGET_SCALAR_MODE_SUPPORTED_P c6x_scalar_mode_supported_p +#undef TARGET_VECTOR_MODE_SUPPORTED_P +#define TARGET_VECTOR_MODE_SUPPORTED_P c6x_vector_mode_supported_p +#undef TARGET_VECTORIZE_PREFERRED_SIMD_MODE +#define TARGET_VECTORIZE_PREFERRED_SIMD_MODE c6x_preferred_simd_mode + +#undef TARGET_RTX_COSTS +#define TARGET_RTX_COSTS c6x_rtx_costs + +#undef TARGET_SCHED_INIT +#define TARGET_SCHED_INIT c6x_sched_init +#undef TARGET_SCHED_SET_SCHED_FLAGS +#define TARGET_SCHED_SET_SCHED_FLAGS c6x_set_sched_flags +#undef TARGET_SCHED_ADJUST_COST +#define TARGET_SCHED_ADJUST_COST c6x_adjust_cost +#undef TARGET_SCHED_ISSUE_RATE +#define TARGET_SCHED_ISSUE_RATE c6x_issue_rate +#undef TARGET_SCHED_VARIABLE_ISSUE +#define TARGET_SCHED_VARIABLE_ISSUE c6x_variable_issue +#undef TARGET_SCHED_REORDER +#define TARGET_SCHED_REORDER c6x_sched_reorder +#undef TARGET_SCHED_REORDER2 +#define TARGET_SCHED_REORDER2 c6x_sched_reorder2 +#undef TARGET_SCHED_EXPOSED_PIPELINE +#define TARGET_SCHED_EXPOSED_PIPELINE true + +#undef TARGET_SCHED_ALLOC_SCHED_CONTEXT +#define TARGET_SCHED_ALLOC_SCHED_CONTEXT c6x_alloc_sched_context +#undef TARGET_SCHED_INIT_SCHED_CONTEXT +#define TARGET_SCHED_INIT_SCHED_CONTEXT c6x_init_sched_context +#undef TARGET_SCHED_SET_SCHED_CONTEXT +#define TARGET_SCHED_SET_SCHED_CONTEXT c6x_set_sched_context +#undef TARGET_SCHED_FREE_SCHED_CONTEXT +#define TARGET_SCHED_FREE_SCHED_CONTEXT c6x_free_sched_context + +#undef TARGET_CAN_ELIMINATE +#define TARGET_CAN_ELIMINATE c6x_can_eliminate + +#undef TARGET_PREFERRED_RENAME_CLASS +#define TARGET_PREFERRED_RENAME_CLASS c6x_preferred_rename_class + +#undef TARGET_MACHINE_DEPENDENT_REORG +#define TARGET_MACHINE_DEPENDENT_REORG c6x_reorg + +#undef TARGET_ASM_FILE_START +#define TARGET_ASM_FILE_START c6x_file_start + +#undef TARGET_PRINT_OPERAND +#define TARGET_PRINT_OPERAND c6x_print_operand +#undef TARGET_PRINT_OPERAND_ADDRESS +#define TARGET_PRINT_OPERAND_ADDRESS c6x_print_operand_address +#undef TARGET_PRINT_OPERAND_PUNCT_VALID_P +#define TARGET_PRINT_OPERAND_PUNCT_VALID_P c6x_print_operand_punct_valid_p + +/* C6x unwinding tables use a different format for the typeinfo tables. */ +#undef TARGET_ASM_TTYPE +#define TARGET_ASM_TTYPE c6x_output_ttype + +#undef TARGET_DWARF_REGISTER_SPAN +#define TARGET_DWARF_REGISTER_SPAN c6x_dwarf_register_span + +#undef TARGET_INIT_BUILTINS +#define TARGET_INIT_BUILTINS c6x_init_builtins +#undef TARGET_EXPAND_BUILTIN +#define TARGET_EXPAND_BUILTIN c6x_expand_builtin +#undef TARGET_BUILTIN_DECL +#define TARGET_BUILTIN_DECL c6x_builtin_decl + +struct gcc_target targetm = TARGET_INITIALIZER; + +#include "gt-c6x.h" diff --git a/gcc/config/c6x/c6x.h b/gcc/config/c6x/c6x.h new file mode 100644 index 0000000..5d34d59 --- /dev/null +++ b/gcc/config/c6x/c6x.h @@ -0,0 +1,617 @@ +/* Target Definitions for TI C6X. + Copyright (C) 2010, 2011 Free Software Foundation, Inc. + Contributed by Andrew Jenner <andrew@codesourcery.com> + Contributed by Bernd Schmidt <bernds@codesourcery.com> + + This file is part of GCC. + + GCC is free software; you can redistribute it and/or modify it + under the terms of the GNU General Public License as published + by the Free Software Foundation; either version 3, or (at your + option) any later version. + + GCC is distributed in the hope that it will be useful, but WITHOUT + ANY WARRANTY; without even the implied warranty of MERCHANTABILITY + or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public + License for more details. + + You should have received a copy of the GNU General Public License + along with GCC; see the file COPYING3. If not see + <http://www.gnu.org/licenses/>. */ + +#ifndef GCC_C6X_H +#define GCC_C6X_H + +/* Feature bit definitions that enable specific insns. */ +#define C6X_INSNS_C62X 1 +#define C6X_INSNS_C64X 2 +#define C6X_INSNS_C64XP 4 +#define C6X_INSNS_C67X 8 +#define C6X_INSNS_C67XP 16 +#define C6X_INSNS_C674X 32 +#define C6X_INSNS_ATOMIC 64 +#define C6X_INSNS_ALL_CPU_BITS 127 + +#define C6X_DEFAULT_INSN_MASK \ + (C6X_INSNS_C62X | C6X_INSNS_C64X | C6X_INSNS_C64XP) + +/* A mask of allowed insn types, as defined above. */ +extern unsigned long c6x_insn_mask; + +/* Value of -march= */ +extern c6x_cpu_t c6x_arch; +#define C6X_DEFAULT_ARCH C6X_CPU_C64XP + +/* True if the target has C64x instructions. */ +#define TARGET_INSNS_64 ((c6x_insn_mask & C6X_INSNS_C64X) != 0) +/* True if the target has C64x+ instructions. */ +#define TARGET_INSNS_64PLUS ((c6x_insn_mask & C6X_INSNS_C64XP) != 0) +/* True if the target has C67x instructions. */ +#define TARGET_INSNS_67 ((c6x_insn_mask & C6X_INSNS_C67X) != 0) +/* True if the target has C67x+ instructions. */ +#define TARGET_INSNS_67PLUS ((c6x_insn_mask & C6X_INSNS_C67XP) != 0) + +/* True if the target supports doubleword loads. */ +#define TARGET_LDDW (TARGET_INSNS_64 || TARGET_INSNS_67) +/* True if the target supports doubleword loads. */ +#define TARGET_STDW TARGET_INSNS_64 +/* True if the target supports the MPY32 family of instructions. */ +#define TARGET_MPY32 TARGET_INSNS_64PLUS +/* True if the target has floating point hardware. */ +#define TARGET_FP TARGET_INSNS_67 +/* True if the target has C67x+ floating point extensions. */ +#define TARGET_FP_EXT TARGET_INSNS_67PLUS + +#define TARGET_DEFAULT 0 + +/* Run-time Target. */ + +#define TARGET_CPU_CPP_BUILTINS() \ + do \ + { \ + builtin_assert ("machine=tic6x"); \ + builtin_assert ("cpu=tic6x"); \ + builtin_define ("__TMS320C6X__"); \ + builtin_define ("_TMS320C6X"); \ + \ + if (TARGET_DSBT) \ + builtin_define ("__DSBT__"); \ + \ + if (TARGET_BIG_ENDIAN) \ + builtin_define ("_BIG_ENDIAN"); \ + else \ + builtin_define ("_LITTLE_ENDIAN"); \ + \ + switch (c6x_arch) \ + { \ + case C6X_CPU_C62X: \ + builtin_define ("_TMS320C6200"); \ + break; \ + \ + case C6X_CPU_C64XP: \ + builtin_define ("_TMS320C6400_PLUS"); \ + /* ... fall through ... */ \ + case C6X_CPU_C64X: \ + builtin_define ("_TMS320C6400"); \ + break; \ + \ + case C6X_CPU_C67XP: \ + builtin_define ("_TMS320C6700_PLUS"); \ + /* ... fall through ... */ \ + case C6X_CPU_C67X: \ + builtin_define ("_TMS320C6700"); \ + break; \ + \ + case C6X_CPU_C674X: \ + builtin_define ("_TMS320C6740"); \ + builtin_define ("_TMS320C6700_PLUS"); \ + builtin_define ("_TMS320C6700"); \ + builtin_define ("_TMS320C6400_PLUS"); \ + builtin_define ("_TMS320C6400"); \ + break; \ + } \ + } while (0) + +#define OPTION_DEFAULT_SPECS \ + {"arch", "%{!march=*:-march=%(VALUE)}" } + +/* Storage Layout. */ + +#define BITS_BIG_ENDIAN 0 +#define BYTES_BIG_ENDIAN (TARGET_BIG_ENDIAN != 0) +#define WORDS_BIG_ENDIAN (TARGET_BIG_ENDIAN != 0) + +#define REG_WORDS_BIG_ENDIAN 0 + +#define UNITS_PER_WORD 4 +#define PARM_BOUNDARY 8 +#define STACK_BOUNDARY 64 +#define FUNCTION_BOUNDARY 32 +#define BIGGEST_ALIGNMENT 64 +#define STRICT_ALIGNMENT 1 + +/* The ABI requires static arrays must be at least 8 byte aligned. + Really only externally visible arrays must be aligned this way, as + only those are directly visible from another compilation unit. But + we don't have that information available here. */ +#define DATA_ALIGNMENT(TYPE, ALIGN) \ + (((ALIGN) < BITS_PER_UNIT * 8 && TREE_CODE (TYPE) == ARRAY_TYPE) \ + ? BITS_PER_UNIT * 8 : (ALIGN)) + +/* Type Layout. */ + +#define DEFAULT_SIGNED_CHAR 1 + +#undef SIZE_TYPE +#define SIZE_TYPE "unsigned int" +#undef PTRDIFF_TYPE +#define PTRDIFF_TYPE "int" + +/* Registers. */ + +#define FIRST_PSEUDO_REGISTER 67 +#define FIXED_REGISTERS \ + { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, \ + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, \ + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, \ + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, \ + 1, 1, 1} +#define CALL_USED_REGISTERS \ + { 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, \ + 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, \ + 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 1, 1, \ + 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, \ + 1, 1, 1} + +/* This lists call-used non-predicate registers first, followed by call-used + registers, followed by predicate registers. We want to avoid allocating + the predicate registers for other uses as much as possible. */ +#define REG_ALLOC_ORDER \ + { \ + REG_A0, REG_A3, REG_A4, REG_A5, REG_A6, REG_A7, REG_A8, REG_A9, \ + REG_A16, REG_A17, REG_A18, REG_A19, REG_A20, REG_A21, REG_A22, REG_A23, \ + REG_A24, REG_A25, REG_A26, REG_A27, REG_A28, REG_A29, REG_A30, REG_A31, \ + REG_B4, REG_B5, REG_B6, REG_B7, REG_B8, REG_B9, REG_B16, \ + REG_B17, REG_B18, REG_B19, REG_B20, REG_B21, REG_B22, REG_B23, REG_B24, \ + REG_B25, REG_B26, REG_B27, REG_B28, REG_B29, REG_B30, REG_B31, \ + REG_A10, REG_A11, REG_A12, REG_A13, REG_A14, REG_A15, \ + REG_B3, REG_B10, REG_B11, REG_B12, REG_B13, REG_B14, REG_B15, \ + REG_A1, REG_A2, REG_B0, REG_B1, REG_B2, REG_ILC \ + } + +#define HARD_REGNO_NREGS(regno, mode) \ + ((GET_MODE_SIZE (mode) + UNITS_PER_WORD - 1) \ + / UNITS_PER_WORD) + +#define HARD_REGNO_MODE_OK(reg, mode) (GET_MODE_SIZE (mode) <= UNITS_PER_WORD \ + ? 1 : ((reg) & 1) == 0) + +#define MODES_TIEABLE_P(mode1, mode2) \ + ((mode1) == (mode2) || \ + (GET_MODE_SIZE (mode1) <= UNITS_PER_WORD && \ + GET_MODE_SIZE (mode2) <= UNITS_PER_WORD)) + + +/* Register Classes. */ + +enum reg_class + { + NO_REGS, + PREDICATE_A_REGS, + PREDICATE_B_REGS, + PREDICATE_REGS, + PICREG, + SPREG, + CALL_USED_B_REGS, + NONPREDICATE_A_REGS, + NONPREDICATE_B_REGS, + NONPREDICATE_REGS, + A_REGS, + B_REGS, + GENERAL_REGS, + ALL_REGS, + LIM_REG_CLASSES + }; + +#define N_REG_CLASSES (int) LIM_REG_CLASSES + +#define REG_CLASS_NAMES { \ + "NO_REGS", \ + "PREDICATE_A_REGS", \ + "PREDICATE_B_REGS", \ + "PREDICATE_REGS", \ + "PICREG", \ + "SPREG", \ + "CALL_USED_B_REGS", \ + "NONPREDICATE_A_REGS", \ + "NONPREDICATE_B_REGS", \ + "NONPREDICATE_REGS", \ + "A_REGS", \ + "B_REGS", \ + "GENERAL_REGS", \ + "ALL_REGS" } + +#define REG_CLASS_CONTENTS \ +{ \ + /* NO_REGS. */ \ + { 0x00000000, 0x00000000, 0 }, \ + /* PREDICATE_A_REGS. */ \ + { 0x00000006, 0x00000000, 0 }, \ + /* PREDICATE_B_REGS. */ \ + { 0x00000000, 0x00000007, 0 }, \ + /* PREDICATE_REGS. */ \ + { 0x00000006, 0x00000007, 0 }, \ + /* PICREG. */ \ + { 0x00000000, 0x00004000, 0 }, \ + /* SPREG. */ \ + { 0x00000000, 0x00008000, 0 }, \ + /* CALL_USED_B_REGS. */ \ + { 0x00000000, 0xFFFF03FF, 0 }, \ + /* NONPREDICATE_A_REGS. */ \ + { 0xFFFFFFF9, 0x00000000, 0 }, \ + /* NONPREDICATE_B_REGS. */ \ + { 0x00000000, 0xFFFFFFF8, 0 }, \ + /* NONPREDICATE_REGS. */ \ + { 0xFFFFFFF9, 0xFFFFFFF8, 0 }, \ + /* A_REGS. */ \ + { 0xFFFFFFFF, 0x00000000, 3 }, \ + /* B_REGS. */ \ + { 0x00000000, 0xFFFFFFFF, 3 }, \ + /* GENERAL_REGS. */ \ + { 0xFFFFFFFF, 0xFFFFFFFF, 3 }, \ + /* ALL_REGS. */ \ + { 0xFFFFFFFF, 0xFFFFFFFF, 7 }, \ +} + +#define A_REGNO_P(N) ((N) <= REG_A31) +#define B_REGNO_P(N) ((N) >= REG_B0 && (N) <= REG_B31) + +#define A_REG_P(X) (REG_P (X) && A_REGNO_P (REGNO (X))) +#define CROSS_OPERANDS(X0,X1) \ + (A_REG_P (X0) == A_REG_P (X1) ? CROSS_N : CROSS_Y) + +#define REGNO_REG_CLASS(reg) \ + ((reg) >= REG_A1 && (reg) <= REG_A2 ? PREDICATE_A_REGS \ + : (reg) == REG_A0 && TARGET_INSNS_64 ? PREDICATE_A_REGS \ + : (reg) >= REG_B0 && (reg) <= REG_B2 ? PREDICATE_B_REGS \ + : A_REGNO_P (reg) ? NONPREDICATE_A_REGS \ + : call_used_regs[reg] ? CALL_USED_B_REGS : B_REGS) + +#define BASE_REG_CLASS ALL_REGS +#define INDEX_REG_CLASS ALL_REGS + +#define REGNO_OK_FOR_BASE_STRICT_P(X) \ + ((X) < FIRST_PSEUDO_REGISTER \ + || (reg_renumber[X] >= 0 && reg_renumber[X] < FIRST_PSEUDO_REGISTER)) +#define REGNO_OK_FOR_BASE_NONSTRICT_P(X) 1 + +#define REGNO_OK_FOR_INDEX_STRICT_P(X) \ + ((X) < FIRST_PSEUDO_REGISTER \ + || (reg_renumber[X] >= 0 && reg_renumber[X] < FIRST_PSEUDO_REGISTER)) +#define REGNO_OK_FOR_INDEX_NONSTRICT_P(X) 1 + +#ifdef REG_OK_STRICT +#define REGNO_OK_FOR_BASE_P(X) REGNO_OK_FOR_BASE_STRICT_P (X) +#define REGNO_OK_FOR_INDEX_P(X) REGNO_OK_FOR_INDEX_STRICT_P (X) +#else +#define REGNO_OK_FOR_BASE_P(X) REGNO_OK_FOR_BASE_NONSTRICT_P (X) +#define REGNO_OK_FOR_INDEX_P(X) REGNO_OK_FOR_INDEX_NONSTRICT_P (X) +#endif + +#define CLASS_MAX_NREGS(class, mode) \ + ((GET_MODE_SIZE (mode) + UNITS_PER_WORD - 1) / UNITS_PER_WORD) + +#define REGNO_OK_FOR_INDIRECT_JUMP_P(REGNO, MODE) B_REGNO_P (REGNO) + +/* Stack and Calling. */ + +/* SP points to 4 bytes below the first word of the frame. */ +#define STACK_POINTER_OFFSET 4 +/* Likewise for AP (which is the incoming stack pointer). */ +#define FIRST_PARM_OFFSET(fundecl) 4 +#define STARTING_FRAME_OFFSET 0 +#define FRAME_GROWS_DOWNWARD 1 +#define STACK_GROWS_DOWNWARD + +#define STACK_POINTER_REGNUM REG_B15 +#define HARD_FRAME_POINTER_REGNUM REG_A15 +/* These two always get eliminated in favour of the stack pointer + or the hard frame pointer. */ +#define FRAME_POINTER_REGNUM REG_FRAME +#define ARG_POINTER_REGNUM REG_ARGP + +#define PIC_OFFSET_TABLE_REGNUM REG_B14 + +/* We keep the stack pointer constant rather than using push/pop + instructions. */ +#define ACCUMULATE_OUTGOING_ARGS 1 + +/* Before the prologue, the return address is in the B3 register. */ +#define RETURN_ADDR_REGNO REG_B3 +#define INCOMING_RETURN_ADDR_RTX gen_rtx_REG (Pmode, RETURN_ADDR_REGNO) + +#define RETURN_ADDR_RTX(COUNT, FRAME) c6x_return_addr_rtx (COUNT) + +#define INCOMING_FRAME_SP_OFFSET 0 +#define ARG_POINTER_CFA_OFFSET(fundecl) 0 + +#define STATIC_CHAIN_REGNUM REG_A2 + +struct c6x_args { + /* Number of arguments to pass in registers. */ + int nregs; + /* Number of arguments passed in registers so far. */ + int count; +}; + +#define CUMULATIVE_ARGS struct c6x_args + +#define INIT_CUMULATIVE_ARGS(cum, fntype, libname, fndecl, n_named_args) \ + c6x_init_cumulative_args (&cum, fntype, libname, n_named_args) + +#define BLOCK_REG_PADDING(MODE, TYPE, FIRST) \ + (c6x_block_reg_pad_upward (MODE, TYPE, FIRST) ? upward : downward) + +#define FUNCTION_ARG_REGNO_P(r) \ + (((r) >= REG_A4 && (r) <= REG_A13) || ((r) >= REG_B4 && (r) <= REG_B13)) + +#define DEFAULT_PCC_STRUCT_RETURN 0 + +#define FUNCTION_PROFILER(file, labelno) \ + fatal_error ("profiling is not yet implemented for this architecture") + + +/* Trampolines. */ +#define TRAMPOLINE_SIZE 32 +#define TRAMPOLINE_ALIGNMENT 256 + +#define ELIMINABLE_REGS \ +{{ ARG_POINTER_REGNUM, STACK_POINTER_REGNUM}, \ + { ARG_POINTER_REGNUM, HARD_FRAME_POINTER_REGNUM}, \ + { FRAME_POINTER_REGNUM, STACK_POINTER_REGNUM}, \ + { FRAME_POINTER_REGNUM, HARD_FRAME_POINTER_REGNUM}} \ + +/* Define the offset between two registers, one to be eliminated, and the other + its replacement, at the start of a routine. */ + +#define INITIAL_ELIMINATION_OFFSET(FROM, TO, OFFSET) \ + ((OFFSET) = c6x_initial_elimination_offset ((FROM), (TO))) + +/* Addressing Modes. */ + +#define CONSTANT_ADDRESS_P(x) (CONSTANT_P(x) && GET_CODE(x) != CONST_DOUBLE) +#define MAX_REGS_PER_ADDRESS 2 + +#define HAVE_PRE_DECREMENT 1 +#define HAVE_POST_DECREMENT 1 +#define HAVE_PRE_INCREMENT 1 +#define HAVE_POST_INCREMENT 1 + +/* Register forms are available, but due to scaling we currently don't + support them. */ +#define HAVE_PRE_MODIFY_DISP 1 +#define HAVE_POST_MODIFY_DISP 1 + +#define LEGITIMATE_PIC_OPERAND_P(X) \ + (!symbolic_operand (X, SImode)) + +struct GTY(()) machine_function +{ + /* True if we expanded a sibling call. */ + int contains_sibcall; +}; + +/* Costs. */ +#define NO_FUNCTION_CSE 1 + +#define SLOW_BYTE_ACCESS 0 + +#define BRANCH_COST(speed_p, predictable_p) 6 + + +/* Model costs for the vectorizer. */ + +/* Cost of conditional branch. */ +#ifndef TARG_COND_BRANCH_COST +#define TARG_COND_BRANCH_COST 6 +#endif + +/* Cost of any scalar operation, excluding load and store. */ +#ifndef TARG_SCALAR_STMT_COST +#define TARG_SCALAR_STMT_COST 1 +#endif + +/* Cost of scalar load. */ +#undef TARG_SCALAR_LOAD_COST +#define TARG_SCALAR_LOAD_COST 2 /* load + rotate */ + +/* Cost of scalar store. */ +#undef TARG_SCALAR_STORE_COST +#define TARG_SCALAR_STORE_COST 10 + +/* Cost of any vector operation, excluding load, store, + or vector to scalar operation. */ +#undef TARG_VEC_STMT_COST +#define TARG_VEC_STMT_COST 1 + +/* Cost of vector to scalar operation. */ +#undef TARG_VEC_TO_SCALAR_COST +#define TARG_VEC_TO_SCALAR_COST 1 + +/* Cost of scalar to vector operation. */ +#undef TARG_SCALAR_TO_VEC_COST +#define TARG_SCALAR_TO_VEC_COST 1 + +/* Cost of aligned vector load. */ +#undef TARG_VEC_LOAD_COST +#define TARG_VEC_LOAD_COST 1 + +/* Cost of misaligned vector load. */ +#undef TARG_VEC_UNALIGNED_LOAD_COST +#define TARG_VEC_UNALIGNED_LOAD_COST 2 + +/* Cost of vector store. */ +#undef TARG_VEC_STORE_COST +#define TARG_VEC_STORE_COST 1 + +/* Cost of vector permutation. */ +#ifndef TARG_VEC_PERMUTE_COST +#define TARG_VEC_PERMUTE_COST 1 +#endif + +/* Exception handling. */ +#define TARGET_EXTRA_CFI_SECTION(unwind) ((unwind) ? ".c6xabi.exidx" : NULL) +/* ttype entries (the only interesting data references used) are + sb-relative got-indirect (aka .ehtype). */ +#define ASM_PREFERRED_EH_DATA_FORMAT(code, data) \ + (((code) == 0 && (data) == 1) ? (DW_EH_PE_datarel | DW_EH_PE_indirect) \ + : DW_EH_PE_absptr) + +/* This should be the same as the definition in elfos.h, plus the call + to output special unwinding directives. */ +#undef ASM_DECLARE_FUNCTION_NAME +#define ASM_DECLARE_FUNCTION_NAME(FILE, NAME, DECL) \ + do \ + { \ + c6x_output_file_unwind (FILE); \ + ASM_OUTPUT_TYPE_DIRECTIVE (FILE, NAME, "function"); \ + ASM_DECLARE_RESULT (FILE, DECL_RESULT (DECL)); \ + ASM_OUTPUT_LABEL (FILE, NAME); \ + } \ + while (0) + +/* This should be the same as the definition in elfos.h, plus the call + to output special unwinding directives. */ +#undef ASM_DECLARE_FUNCTION_SIZE +#define ASM_DECLARE_FUNCTION_SIZE(STREAM, NAME, DECL) \ + c6x_function_end (STREAM, NAME) + +/* Arbitrarily choose A4/A5. */ +#define EH_RETURN_DATA_REGNO(N) (((N) < 2) ? (N) + 4 : INVALID_REGNUM) + +/* The register that holds the return address in exception handlers. */ +#define C6X_EH_STACKADJ_REGNUM 3 +#define EH_RETURN_STACKADJ_RTX gen_rtx_REG (SImode, C6X_EH_STACKADJ_REGNUM) + + +/* Assembler Format. */ + +#define DWARF2_ASM_LINE_DEBUG_INFO 1 + +#undef ASM_APP_ON +#define ASM_APP_ON "\t; #APP \n" +#undef ASM_APP_OFF +#define ASM_APP_OFF "\t; #NO_APP \n" + +#define ASM_OUTPUT_COMMON(stream, name, size, rounded) +#define ASM_OUTPUT_LOCAL(stream, name, size, rounded) + +#define GLOBAL_ASM_OP "\t.global\t" + +#define REGISTER_NAMES \ + { \ + "A0", "A1", "A2", "A3", "A4", "A5", "A6", "A7", \ + "A8", "A9", "A10", "A11", "A12", "A13", "A14", "A15", \ + "A16", "A17", "A18", "A19", "A20", "A21", "A22", "A23", \ + "A24", "A25", "A26", "A27", "A28", "A29", "A30", "A31", \ + "B0", "B1", "B2", "B3", "B4", "B5", "B6", "B7", \ + "B8", "B9", "B10", "B11", "B12", "B13", "B14", "B15", \ + "B16", "B17", "B18", "B19", "B20", "B21", "B22", "B23", \ + "B24", "B25", "B26", "B27", "B28", "B29", "B30", "B31", \ + "FP", "ARGP", "ILC" } + +#define DBX_REGISTER_NUMBER(N) (dbx_register_map[(N)]) + +extern int const dbx_register_map[FIRST_PSEUDO_REGISTER]; + +#define FINAL_PRESCAN_INSN c6x_final_prescan_insn + +#define TEXT_SECTION_ASM_OP ".text;" +#define DATA_SECTION_ASM_OP ".data;" + +#define ASM_OUTPUT_ALIGN(stream, power) \ + do \ + { \ + if (power) \ + fprintf ((stream), "\t.align\t%d\n", power); \ + } \ + while (0) + +#define ASM_OUTPUT_ADDR_VEC_ELT(FILE, VALUE) \ +do { char __buf[256]; \ + fprintf (FILE, "\t.long\t"); \ + ASM_GENERATE_INTERNAL_LABEL (__buf, "L", VALUE); \ + assemble_name (FILE, __buf); \ + fputc ('\n', FILE); \ + } while (0) + +/* Determine whether to place EXP (an expression or a decl) should be + placed into one of the small data sections. */ +#define PLACE_IN_SDATA_P(EXP) \ + (c6x_sdata_mode == C6X_SDATA_NONE ? false \ + : c6x_sdata_mode == C6X_SDATA_ALL ? true \ + : !AGGREGATE_TYPE_P (TREE_TYPE (EXP))) + +#define SCOMMON_ASM_OP "\t.scomm\t" + +#undef ASM_OUTPUT_ALIGNED_DECL_COMMON +#define ASM_OUTPUT_ALIGNED_DECL_COMMON(FILE, DECL, NAME, SIZE, ALIGN) \ + do \ + { \ + if (DECL != NULL && PLACE_IN_SDATA_P (DECL)) \ + fprintf ((FILE), "%s", SCOMMON_ASM_OP); \ + else \ + fprintf ((FILE), "%s", COMMON_ASM_OP); \ + assemble_name ((FILE), (NAME)); \ + fprintf ((FILE), ",%u,%u\n", (int)(SIZE), (ALIGN) / BITS_PER_UNIT);\ + } \ + while (0) + +/* This says how to output assembler code to declare an + uninitialized internal linkage data object. */ + +#undef ASM_OUTPUT_ALIGNED_DECL_LOCAL +#define ASM_OUTPUT_ALIGNED_DECL_LOCAL(FILE, DECL, NAME, SIZE, ALIGN) \ +do { \ + if (PLACE_IN_SDATA_P (DECL)) \ + switch_to_section (sbss_section); \ + else \ + switch_to_section (bss_section); \ + ASM_OUTPUT_TYPE_DIRECTIVE (FILE, NAME, "object"); \ + if (!flag_inhibit_size_directive) \ + ASM_OUTPUT_SIZE_DIRECTIVE (FILE, NAME, SIZE); \ + ASM_OUTPUT_ALIGN ((FILE), exact_log2((ALIGN) / BITS_PER_UNIT)); \ + ASM_OUTPUT_LABEL(FILE, NAME); \ + ASM_OUTPUT_SKIP((FILE), (SIZE) ? (SIZE) : 1); \ +} while (0) + +#define CASE_VECTOR_PC_RELATIVE flag_pic +#define JUMP_TABLES_IN_TEXT_SECTION flag_pic + +#define ADDR_VEC_ALIGN(VEC) (JUMP_TABLES_IN_TEXT_SECTION ? 5 : 2) + +/* This is how to output an element of a case-vector that is relative. */ +#define ASM_OUTPUT_ADDR_DIFF_ELT(FILE, BODY, VALUE, REL) \ + do { char buf[100]; \ + fputs ("\t.long ", FILE); \ + ASM_GENERATE_INTERNAL_LABEL (buf, "L", VALUE); \ + assemble_name (FILE, buf); \ + putc ('-', FILE); \ + ASM_GENERATE_INTERNAL_LABEL (buf, "L", REL); \ + assemble_name (FILE, buf); \ + putc ('\n', FILE); \ + } while (0) + +/* Misc. */ + +#define CASE_VECTOR_MODE SImode +#define MOVE_MAX 4 +#define MOVE_RATIO(SPEED) 4 +#define TRULY_NOOP_TRUNCATION(outprec, inprec) 1 +#define CLZ_DEFINED_VALUE_AT_ZERO(MODE, VALUE) ((VALUE) = 32, 1) +#define Pmode SImode +#define FUNCTION_MODE QImode + +extern int c6x_initial_flag_pic; + +#endif /* GCC_C6X_H */ diff --git a/gcc/config/c6x/c6x.md b/gcc/config/c6x/c6x.md new file mode 100644 index 0000000..7935a81 --- /dev/null +++ b/gcc/config/c6x/c6x.md @@ -0,0 +1,3004 @@ +;; Machine description for TI C6X. +;; Copyright (C) 2010, 2011 Free Software Foundation, Inc. +;; Contributed by Andrew Jenner <andrew@codesourcery.com> +;; Contributed by Bernd Schmidt <bernds@codesourcery.com> +;; Contributed by CodeSourcery. +;; +;; This file is part of GCC. +;; +;; GCC is free software; you can redistribute it and/or modify +;; it under the terms of the GNU General Public License as published by +;; the Free Software Foundation; either version 3, or (at your option) +;; any later version. +;; +;; GCC is distributed in the hope that it will be useful, +;; but WITHOUT ANY WARRANTY; without even the implied warranty of +;; MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the +;; GNU General Public License for more details. +;; +;; You should have received a copy of the GNU General Public License +;; along with GCC; see the file COPYING3. If not see +;; <http://www.gnu.org/licenses/>. + + +;; Register names + +(define_constants + [(REG_A0 0) + (REG_A1 1) + (REG_A2 2) + (REG_A3 3) + (REG_A4 4) + (REG_A5 5) + (REG_A6 6) + (REG_A7 7) + (REG_A8 8) + (REG_A9 9) + (REG_A10 10) + (REG_A11 11) + (REG_A12 12) + (REG_A13 13) + (REG_A14 14) + (REG_A15 15) + (REG_A16 16) + (REG_A17 17) + (REG_A18 18) + (REG_A19 19) + (REG_A20 20) + (REG_A21 21) + (REG_A22 22) + (REG_A23 23) + (REG_A24 24) + (REG_A25 25) + (REG_A26 26) + (REG_A27 27) + (REG_A28 28) + (REG_A29 29) + (REG_A30 30) + (REG_A31 31) + (REG_B0 32) + (REG_B1 33) + (REG_B2 34) + (REG_B3 35) + (REG_B4 36) + (REG_B5 37) + (REG_B6 38) + (REG_B7 39) + (REG_B8 40) + (REG_B9 41) + (REG_B10 42) + (REG_B11 43) + (REG_B12 44) + (REG_B13 45) + (REG_B14 46) + (REG_SP 47) + (REG_B15 47) + (REG_B16 48) + (REG_B17 49) + (REG_B18 50) + (REG_B19 51) + (REG_B20 52) + (REG_B21 53) + (REG_B22 54) + (REG_B23 55) + (REG_B24 56) + (REG_B25 57) + (REG_B26 58) + (REG_B27 59) + (REG_B28 60) + (REG_B29 61) + (REG_B30 62) + (REG_B31 63) + (REG_FRAME 64) + (REG_ARGP 65) + (REG_ILC 66)]) + +(define_c_enum "unspec" [ + UNSPEC_NOP + UNSPEC_RCP + UNSPEC_MISALIGNED_ACCESS + UNSPEC_ADDKPC + UNSPEC_SETUP_DSBT + UNSPEC_LOAD_GOT + UNSPEC_LOAD_SDATA + UNSPEC_BITREV + UNSPEC_GOTOFF + UNSPEC_MVILC + UNSPEC_REAL_JUMP + UNSPEC_REAL_LOAD + UNSPEC_REAL_MULT + UNSPEC_JUMP_SHADOW + UNSPEC_LOAD_SHADOW + UNSPEC_MULT_SHADOW + UNSPEC_EPILOGUE_BARRIER + UNSPEC_ATOMIC + UNSPEC_CLR + UNSPEC_EXT + UNSPEC_EXTU + UNSPEC_SUBC + UNSPEC_AVG +]) + +(define_c_enum "unspecv" [ + UNSPECV_BLOCKAGE + UNSPECV_SPLOOP + UNSPECV_SPKERNEL + UNSPECV_EH_RETURN + UNSPECV_CAS +]) + +;; ------------------------------------------------------------------------- +;; Instruction attributes +;; ------------------------------------------------------------------------- + +(define_attr "cpu" + "c62x,c64x,c64xp,c67x,c67xp,c674x" + (const (symbol_ref "(enum attr_cpu)c6x_arch"))) + +;; Define a type for each insn which is used in the scheduling description. +;; These correspond to the types defined in chapter 4 of the C674x manual. +(define_attr "type" + "unknown,single,mpy2,store,storen,mpy4,load,loadn,branch,call,callp,dp2,fp4, + intdp,cmpdp,adddp,mpy,mpyi,mpyid,mpydp,mpyspdp,mpysp2dp,spkernel,sploop, + mvilc,blockage,shadow,load_shadow,mult_shadow,atomic" + (const_string "single")) + +;; The register file used by an instruction's destination register. +;; The function destreg_file computes this; instructions can override the +;; attribute if they aren't a single_set. +(define_attr "dest_regfile" + "unknown,any,a,b" + (cond [(eq_attr "type" "single,load,mpy2,mpy4,dp2,fp4,intdp,cmpdp,adddp,mpy,mpyi,mpyid,mpydp,mpyspdp,mpysp2dp") + (cond [(match_operand 0 "a_register" "") (const_string "a") + (match_operand 0 "b_register" "") (const_string "b")] + (const_string "unknown")) + (eq_attr "type" "store") + (cond [(match_operand 1 "a_register" "") (const_string "a") + (match_operand 1 "b_register" "") (const_string "b")] + (const_string "unknown"))] + (const_string "unknown"))) + +(define_attr "addr_regfile" + "unknown,a,b" + (const_string "unknown")) + +(define_attr "cross" + "n,y" + (const_string "n")) + +(define_attr "has_shadow" + "n,y" + (const_string "n")) + +;; The number of cycles the instruction takes to finish. Any cycles above +;; the first are delay slots. +(define_attr "cycles" "" + (cond [(eq_attr "type" "branch,call") (const_int 6) + (eq_attr "type" "load,loadn") (const_int 5) + (eq_attr "type" "dp2") (const_int 2) + (eq_attr "type" "mpy2") (const_int 2) + (eq_attr "type" "mpy4") (const_int 4) + (eq_attr "type" "fp4") (const_int 4) + (eq_attr "type" "mvilc") (const_int 4) + (eq_attr "type" "cmpdp") (const_int 2) + (eq_attr "type" "intdp") (const_int 5) + (eq_attr "type" "adddp") (const_int 7) + (eq_attr "type" "mpydp") (const_int 10) + (eq_attr "type" "mpyi") (const_int 9) + (eq_attr "type" "mpyid") (const_int 10) + (eq_attr "type" "mpyspdp") (const_int 7) + (eq_attr "type" "mpysp2dp") (const_int 5)] + (const_int 1))) + +(define_attr "predicable" "no,yes" + (const_string "yes")) + +(define_attr "enabled" "no,yes" + (const_string "yes")) + +;; Specify which units can be used by a given instruction. Normally, +;; dest_regfile is used to select between the two halves of the machine. +;; D_ADDR is for load/store instructions; they use the D unit and use +;; addr_regfile to choose between D1 and D2. + +(define_attr "units62" + "unknown,d,d_addr,l,m,s,dl,ds,dls,ls" + (const_string "unknown")) + +(define_attr "units64" + "unknown,d,d_addr,l,m,s,dl,ds,dls,ls" + (const_string "unknown")) + +(define_attr "units64p" + "unknown,d,d_addr,l,m,s,dl,ds,dls,ls" + (attr "units64")) + +(define_attr "units67" + "unknown,d,d_addr,l,m,s,dl,ds,dls,ls" + (attr "units62")) + +(define_attr "units67p" + "unknown,d,d_addr,l,m,s,dl,ds,dls,ls" + (attr "units67")) + +(define_attr "units674" + "unknown,d,d_addr,l,m,s,dl,ds,dls,ls" + (attr "units64")) + +(define_attr "units" + "unknown,d,d_addr,l,m,s,dl,ds,dls,ls" + (cond [(eq_attr "cpu" "c62x") + (attr "units62") + (eq_attr "cpu" "c67x") + (attr "units67") + (eq_attr "cpu" "c67xp") + (attr "units67p") + (eq_attr "cpu" "c64x") + (attr "units64") + (eq_attr "cpu" "c64xp") + (attr "units64p") + (eq_attr "cpu" "c674x") + (attr "units674") + ] + (const_string "unknown"))) + +(define_automaton "c6x_1,c6x_w1,c6x_2,c6x_w2,c6x_m1,c6x_m2,c6x_t1,c6x_t2,c6x_branch") +(automata_option "ndfa") + +(define_cpu_unit "d1,l1,s1" "c6x_1") +(define_cpu_unit "x1" "c6x_1") +(define_cpu_unit "l1w,s1w" "c6x_w1") +(define_cpu_unit "m1" "c6x_m1") +(define_cpu_unit "m1w" "c6x_m1") +(define_cpu_unit "t1" "c6x_t1") +(define_cpu_unit "d2,l2,s2" "c6x_2") +(define_cpu_unit "x2" "c6x_2") +(define_cpu_unit "l2w,s2w" "c6x_w2") +(define_cpu_unit "m2" "c6x_m2") +(define_cpu_unit "m2w" "c6x_m2") +(define_cpu_unit "t2" "c6x_t2") + +;; There can be up to two branches in one cycle (on the .s1 and .s2 +;; units), but some instructions must not be scheduled in parallel +;; with a branch. We model this by reserving either br0 or br1 for a +;; normal branch, and both of them for an insn such as callp. +;; Another constraint is that two branches may only execute in parallel +;; if one uses an offset, and the other a register. We can distinguish +;; these by the dest_regfile attribute; it is "any" iff the branch uses +;; an offset. br0 is reserved for these, while br1 is reserved for +;; branches using a register. +(define_cpu_unit "br0,br1" "c6x_branch") + +(include "c6x-sched.md") + +;; Some reservations which aren't generated from c6x-sched.md.in + +(define_insn_reservation "branch_s1any" 6 + (and (eq_attr "type" "branch") + (and (eq_attr "cross" "n") + (and (eq_attr "units" "s") + (eq_attr "dest_regfile" "any")))) + "s1+s1w+br0") + +;; For calls, we also reserve the units needed in the following cycles +;; to load the return address. There are two options; using addkpc or +;; mvkh/mvkl. The code in c6x_reorg knows whether to use one of these +;; or whether to use callp. The actual insns are emitted only after +;; the final scheduling pass is complete. +;; We always reserve S2 for PC-relative call insns, since that allows +;; us to turn them into callp insns later on. +(define_insn_reservation "call_addkpc_s1any" 6 + (and (eq_attr "type" "call") + (and (ne (symbol_ref "TARGET_INSNS_64") (const_int 0)) + (and (eq_attr "cross" "n") + (and (eq_attr "units" "s") + (eq_attr "dest_regfile" "any"))))) + "s2+s2w+br0,s2+s2w+br0+br1") + +(define_insn_reservation "call_mvk_s1any" 6 + (and (eq_attr "type" "call") + (and (eq (symbol_ref "TARGET_INSNS_64") (const_int 0)) + (and (eq_attr "cross" "n") + (and (eq_attr "units" "s") + (eq_attr "dest_regfile" "any"))))) + "s2+s2w+br0,s2+s2w,s2+s2w") + +(define_reservation "all" "s1+s2+d1+d2+l1+l2+m1+m2") + +(define_insn_reservation "callp_s1" 1 + (and (eq_attr "type" "callp") (eq_attr "dest_regfile" "a")) + "s1+s1w,all*5") + +(define_insn_reservation "callp_s2" 1 + (and (eq_attr "type" "callp") (eq_attr "dest_regfile" "b")) + "s2+s2w,all*5") + +;; Constraints + +(include "constraints.md") + +;; Predicates + +(include "predicates.md") + +;; General predication pattern. + +(define_cond_exec + [(match_operator 0 "eqne_operator" + [(match_operand 1 "predicate_register" "AB") + (const_int 0)])] + "" + "") + +;; ------------------------------------------------------------------------- +;; NOP instruction +;; ------------------------------------------------------------------------- + +(define_insn "nop" + [(const_int 0)] + "" + "nop") + +(define_insn "nop_count" + [(unspec [(match_operand 0 "const_int_operand" "n")] UNSPEC_NOP)] + "" + "%|%.\\tnop\\t%0") + +;; ------------------------------------------------------------------------- +;; Move instructions +;; ------------------------------------------------------------------------- + +(define_mode_iterator QIHIM [QI HI]) +(define_mode_iterator SIDIM [SI DI]) +(define_mode_iterator SIDIVM [SI DI V2HI V4QI]) +(define_mode_iterator VEC4M [V2HI V4QI]) +(define_mode_iterator VEC8M [V2SI V4HI V8QI]) +(define_mode_iterator SISFVM [SI SF V2HI V4QI]) +(define_mode_iterator DIDFM [DI DF]) +(define_mode_iterator DIDFVM [DI DF V2SI V4HI V8QI]) +(define_mode_iterator SFDFM [SF DF]) +(define_mode_iterator M32 [QI HI SI SF V2HI V4QI]) + +;; The C6X LO_SUM and HIGH are backwards - HIGH sets the low bits, and +;; LO_SUM adds in the high bits. Fortunately these are opaque operations +;; so this does not matter. +(define_insn "movsi_lo_sum" + [(set (match_operand:SI 0 "register_operand" "=ab") + (lo_sum:SI (match_operand:SI 1 "register_operand" "0") + (match_operand:SI 2 "const_int_or_symbolic_operand" "i")))] + "reload_completed" + "%|%.\\tmvkh\\t%$\\t%2, %0" + [(set_attr "units" "s")]) + +(define_insn "movsi_high" + [(set (match_operand:SI 0 "register_operand" "=ab") + (high:SI (match_operand:SI 1 "const_int_or_symbolic_operand" "i")))] + "reload_completed" + "%|%.\\tmvkl\\t%$\\t%1, %0" + [(set_attr "units" "s")]) + +(define_insn "movsi_gotoff_lo_sum" + [(set (match_operand:SI 0 "register_operand" "=ab") + (lo_sum:SI (match_operand:SI 1 "register_operand" "0") + (unspec:SI [(match_operand:SI 2 "symbolic_operand" "S2")] + UNSPEC_GOTOFF)))] + "flag_pic == 2" + "%|%.\\tmvkh\\t%$\\t$dpr_got%2, %0" + [(set_attr "units" "s")]) + +(define_insn "movsi_gotoff_high" + [(set (match_operand:SI 0 "register_operand" "=ab") + (high:SI (unspec:SI [(match_operand:SI 1 "symbolic_operand" "S2")] + UNSPEC_GOTOFF)))] + "flag_pic == 2" + "%|%.\\tmvkl\\t%$\\t$dpr_got%1, %0" + [(set_attr "units" "s")]) + +;; Normally we'd represent this as a normal load insn, but we can't currently +;; represent the addressing mode. +(define_insn "load_got_gotoff" + [(set (match_operand:SI 0 "register_operand" "=a,b") + (unspec:SI [(match_operand:SI 1 "register_operand" "Z,Z") + (match_operand:SI 2 "register_operand" "b,b")] + UNSPEC_GOTOFF))] + "flag_pic == 2" + "%|%.\\tldw\\t%$\\t*+%1[%2], %0" + [(set_attr "type" "load") + (set_attr "units" "d_addr") + (set_attr "dest_regfile" "a,b") + (set_attr "addr_regfile" "b")]) + +(define_insn "*movstricthi_high" + [(set (match_operand:SI 0 "register_operand" "+ab") + (ior:SI (and:SI (match_dup 0) (const_int 65535)) + (ashift:SI (match_operand:SI 1 "const_int_operand" "IuB") + (const_int 16))))] + "reload_completed" + "%|%.\\tmvklh\\t%$\\t%1, %0" + [(set_attr "units" "s")]) + +;; Break up SImode loads of immediate operands. + +(define_split + [(set (match_operand:SI 0 "register_operand" "") + (match_operand:SI 1 "const_int_operand" ""))] + "reload_completed + && !satisfies_constraint_IsB (operands[1])" + [(set (match_dup 0) (match_dup 2)) + (set (match_dup 0) (ior:SI (and:SI (match_dup 0) (const_int 65535)) + (ashift:SI (match_dup 3) (const_int 16))))] +{ + HOST_WIDE_INT val = INTVAL (operands[1]); + operands[2] = GEN_INT (trunc_int_for_mode (val, HImode)); + operands[3] = GEN_INT ((val >> 16) & 65535); +}) + +(define_split + [(set (match_operand:VEC4M 0 "register_operand" "") + (match_operand:VEC4M 1 "const_vector_operand" ""))] + "reload_completed" + [(set (match_dup 2) (match_dup 3))] +{ + unsigned HOST_WIDE_INT mask, val; + enum machine_mode inner_mode = GET_MODE_INNER (<MODE>mode); + int i; + + val = 0; + mask = GET_MODE_MASK (inner_mode); + if (TARGET_BIG_ENDIAN) + { + for (i = 0; i < GET_MODE_NUNITS (<MODE>mode); i++) + { + val <<= GET_MODE_BITSIZE (inner_mode); + val |= INTVAL (CONST_VECTOR_ELT (operands[1], i)) & mask; + } + } + else + { + i = GET_MODE_NUNITS (<MODE>mode); + while (i-- > 0) + { + val <<= GET_MODE_BITSIZE (inner_mode); + val |= INTVAL (CONST_VECTOR_ELT (operands[1], i)) & mask; + } + } + operands[2] = gen_rtx_REG (SImode, REGNO (operands[0])); + operands[3] = GEN_INT (trunc_int_for_mode (val, SImode)); +}) + +(define_split + [(set (match_operand:VEC8M 0 "register_operand" "") + (match_operand:VEC8M 1 "const_vector_operand" ""))] + "reload_completed" + [(set (match_dup 2) (match_dup 3)) + (set (match_dup 4) (match_dup 5))] +{ + unsigned HOST_WIDE_INT mask; + unsigned HOST_WIDE_INT val[2]; + rtx lo_half, hi_half; + enum machine_mode inner_mode = GET_MODE_INNER (<MODE>mode); + int i, j; + + split_di (operands, 1, &lo_half, &hi_half); + + val[0] = val[1] = 0; + mask = GET_MODE_MASK (inner_mode); + if (TARGET_BIG_ENDIAN) + { + for (i = 0, j = 1; i < GET_MODE_NUNITS (<MODE>mode); i++) + { + if (i * 2 == GET_MODE_NUNITS (<MODE>mode)) + j--; + val[j] <<= GET_MODE_BITSIZE (inner_mode); + val[j] |= INTVAL (CONST_VECTOR_ELT (operands[1], i)) & mask; + } + } + else + { + i = GET_MODE_NUNITS (<MODE>mode); + j = 1; + while (i-- > 0) + { + val[j] <<= GET_MODE_BITSIZE (inner_mode); + val[j] |= INTVAL (CONST_VECTOR_ELT (operands[1], i)) & mask; + if (i * 2 == GET_MODE_NUNITS (<MODE>mode)) + j--; + } + } + operands[2] = lo_half; + operands[3] = GEN_INT (trunc_int_for_mode (val[0], SImode)); + operands[4] = hi_half; + operands[5] = GEN_INT (trunc_int_for_mode (val[1], SImode)); +}) + +(define_split + [(set (match_operand:SF 0 "register_operand" "") + (match_operand:SF 1 "immediate_operand" ""))] + "reload_completed" + [(set (match_dup 2) (match_dup 3)) + (set (match_dup 2) (ior:SI (and:SI (match_dup 2) (const_int 65535)) + (ashift:SI (match_dup 4) (const_int 16))))] +{ + long values; + REAL_VALUE_TYPE value; + + gcc_assert (GET_CODE (operands[1]) == CONST_DOUBLE); + + REAL_VALUE_FROM_CONST_DOUBLE (value, operands[1]); + REAL_VALUE_TO_TARGET_SINGLE (value, values); + + operands[2] = gen_rtx_REG (SImode, true_regnum (operands[0])); + operands[3] = GEN_INT (trunc_int_for_mode (values, HImode)); + if (values >= -32768 && values < 32768) + { + emit_move_insn (operands[2], operands[3]); + DONE; + } + operands[4] = GEN_INT ((values >> 16) & 65535); +}) + +(define_split + [(set (match_operand:SI 0 "register_operand" "") + (match_operand:SI 1 "symbolic_operand" ""))] + "reload_completed + && (!TARGET_INSNS_64PLUS + || !sdata_symbolic_operand (operands[1], SImode))" + [(set (match_dup 0) (high:SI (match_dup 1))) + (set (match_dup 0) (lo_sum:SI (match_dup 0) (match_dup 1)))] + "") + +;; Normally, we represent the load of an sdata address as a normal +;; move of a SYMBOL_REF. In DSBT mode, B14 is not constant, so we +;; should show the dependency. +(define_insn "load_sdata_pic" + [(set (match_operand:SI 0 "register_operand" "=a,b") + (plus:SI (match_operand:SI 1 "pic_register_operand" "Z,Z") + (unspec:SI [(match_operand:SI 2 "sdata_symbolic_operand" "S0,S0")] + UNSPEC_LOAD_SDATA)))] + "flag_pic" + "@ + %|%.\\tadda%D2\\t%$\\t%1, %2, %0 + %|%.\\tadda%D2\\t%$\\t%1, %2, %0" + [(set_attr "units" "d") + (set_attr "cross" "y,n") + (set_attr "predicable" "no")]) + +;; Move instruction patterns + +(define_mode_attr LDST_SUFFIX [(QI "b") (HI "h") + (SI "w") (SF "w") (V2HI "w") (V4QI "w") + (DI "dw") (V2SI "dw") (V4HI "dw") (V8QI "dw")]) + +(define_insn "mov<mode>_insn" + [(set (match_operand:QIHIM 0 "nonimmediate_operand" + "=a,b, a, b, ab, ab,a,?a, b,?b, Q, R, R, Q") + (match_operand:QIHIM 1 "general_operand" + "a,b,?b,?a,Is5,IsB,Q, R, R, Q, a,?a, b,?b"))] + "GET_CODE (operands[0]) != MEM || GET_CODE (operands[1]) == REG" + "@ + %|%.\\tmv\\t%$\\t%1, %0 + %|%.\\tmv\\t%$\\t%1, %0 + %|%.\\tmv\\t%$\\t%1, %0 + %|%.\\tmv\\t%$\\t%1, %0 + %|%.\\tmvk\\t%$\\t%1, %0 + %|%.\\tmvk\\t%$\\t%1, %0 + %|%.\\tld<LDST_SUFFIX>\\t%$\\t%1, %0 + %|%.\\tld<LDST_SUFFIX>\\t%$\\t%1, %0 + %|%.\\tld<LDST_SUFFIX>\\t%$\\t%1, %0 + %|%.\\tld<LDST_SUFFIX>\\t%$\\t%1, %0 + %|%.\\tst<LDST_SUFFIX>\\t%$\\t%1, %0 + %|%.\\tst<LDST_SUFFIX>\\t%$\\t%1, %0 + %|%.\\tst<LDST_SUFFIX>\\t%$\\t%1, %0 + %|%.\\tst<LDST_SUFFIX>\\t%$\\t%1, %0" + [(set_attr "type" "*,*,*,*,*,*,load,load,load,load,store,store,store,store") + (set_attr "units62" "dls,dls,ls,ls,s,s,d_addr,d_addr,d_addr,d_addr,d_addr,d_addr,d_addr,d_addr") + (set_attr "units64" "dls,dls,ls,ls,dl,s,d_addr,d_addr,d_addr,d_addr,d_addr,d_addr,d_addr,d_addr") + (set_attr "addr_regfile" "*,*,*,*,*,*,a,b,b,a,a,b,b,a") + (set_attr "dest_regfile" "*,*,*,*,*,*,a,a,b,b,a,a,b,b") + (set_attr "cross" "n,n,y,y,n,n,n,y,n,y,n,y,n,y")]) + +(define_insn "mov<mode>_insn" + [(set (match_operand:SISFVM 0 "nonimmediate_operand" + "=a,b, a, b, ab, ab,a,b,ab,a,?a, b,?b, Q, R, R, Q") + (match_operand:SISFVM 1 "general_operand" + "a,b,?b,?a,Is5,IsB,S0,S0,Si,Q, R, R, Q, a,?a, b,?b"))] + "(GET_CODE (operands[0]) != MEM || GET_CODE (operands[1]) == REG + || (GET_CODE (operands[1]) == SUBREG && REG_P (SUBREG_REG (operands[1]))))" + "@ + %|%.\\tmv\\t%$\\t%1, %0 + %|%.\\tmv\\t%$\\t%1, %0 + %|%.\\tmv\\t%$\\t%1, %0 + %|%.\\tmv\\t%$\\t%1, %0 + %|%.\\tmvk\\t%$\\t%1, %0 + %|%.\\tmvk\\t%$\\t%1, %0 + %|%.\\tadda%D1\\t%$\\tB14, %1, %0 + %|%.\\tadda%D1\\t%$\\tB14, %1, %0 + # + %|%.\\tldw\\t%$\\t%1, %0 + %|%.\\tldw\\t%$\\t%1, %0 + %|%.\\tldw\\t%$\\t%1, %0 + %|%.\\tldw\\t%$\\t%1, %0 + %|%.\\tstw\\t%$\\t%1, %0 + %|%.\\tstw\\t%$\\t%1, %0 + %|%.\\tstw\\t%$\\t%1, %0 + %|%.\\tstw\\t%$\\t%1, %0" + [(set_attr "type" "*,*,*,*,*,*,*,*,*,load,load,load,load,store,store,store,store") + (set_attr "units62" "dls,dls,ls,ls,s,s,d,d,*,d_addr,d_addr,d_addr,d_addr,d_addr,d_addr,d_addr,d_addr") + (set_attr "units64" "dls,dls,ls,ls,dl,s,d,d,*,d_addr,d_addr,d_addr,d_addr,d_addr,d_addr,d_addr,d_addr") + (set_attr "addr_regfile" "*,*,*,*,*,*,*,*,*,a,b,b,a,a,b,b,a") + (set_attr "dest_regfile" "*,*,*,*,*,*,*,*,*,a,a,b,b,a,a,b,b") + (set_attr "cross" "n,n,y,y,n,n,y,n,*,n,y,n,y,n,y,n,y") + (set_attr "predicable" "yes,yes,yes,yes,yes,yes,no,no,yes,yes,yes,yes,yes,yes,yes,yes,yes")]) + +(define_insn "*mov<mode>_insn" + [(set (match_operand:DIDFVM 0 "nonimmediate_operand" + "=a,b, a, b,ab,a,?a, b,?b, Q, R, R, Q") + (match_operand:DIDFVM 1 "general_operand" + "a,b,?b,?a,iF,Q, R, R, Q, a,?a, b,?b"))] + "(!MEM_P (operands[0]) || REG_P (operands[1]) + || (GET_CODE (operands[1]) == SUBREG && REG_P (SUBREG_REG (operands[1]))))" +{ + if (MEM_P (operands[1]) && TARGET_LDDW) + return "%|%.\\tlddw\\t%$\\t%1, %0"; + if (MEM_P (operands[0]) && TARGET_STDW) + return "%|%.\\tstdw\\t%$\\t%1, %0"; + if (TARGET_INSNS_64PLUS && REG_P (operands[0]) && REG_P (operands[1]) + && A_REGNO_P (REGNO (operands[0])) == A_REGNO_P (REGNO (operands[1]))) + return "%|%.\\tdmv\\t%$\\t%P1, %p1, %0"; + return "#"; +} + [(set_attr "units" "s,s,*,*,*,d_addr,d_addr,d_addr,d_addr,d_addr,d_addr,d_addr,d_addr") + (set_attr "addr_regfile" "*,*,*,*,*,a,b,b,a,a,b,b,a") + (set_attr "dest_regfile" "*,*,*,*,*,a,a,b,b,a,a,b,b") + (set_attr "type" "*,*,*,*,*,load,load,load,load,store,store,store,store") + (set_attr "cross" "n,n,y,y,*,n,y,n,y,n,y,n,y")]) + +(define_split + [(set (match_operand:DIDFVM 0 "nonimmediate_operand" "") + (match_operand:DIDFVM 1 "general_operand" ""))] + "reload_completed + && !((MEM_P (operands[0]) && TARGET_STDW) + || (MEM_P (operands[1]) && TARGET_LDDW)) + && !const_vector_operand (operands[1], <MODE>mode) + && !(TARGET_INSNS_64PLUS && REG_P (operands[0]) && REG_P (operands[1]) + && A_REGNO_P (REGNO (operands[0])) == A_REGNO_P (REGNO (operands[1])))" + [(set (match_dup 2) (match_dup 3)) + (set (match_dup 4) (match_dup 5))] +{ + rtx lo_half[2], hi_half[2]; + split_di (operands, 2, lo_half, hi_half); + + /* We can't have overlap for a register-register move, but if + memory is involved, we have to make sure we don't clobber the + address. */ + if (reg_overlap_mentioned_p (lo_half[0], hi_half[1])) + { + operands[2] = hi_half[0]; + operands[3] = hi_half[1]; + operands[4] = lo_half[0]; + operands[5] = lo_half[1]; + } + else + { + operands[2] = lo_half[0]; + operands[3] = lo_half[1]; + operands[4] = hi_half[0]; + operands[5] = hi_half[1]; + } +}) + +(define_insn "real_load<mode>" + [(unspec [(match_operand 0 "const_int_operand" "JA,JA,JB,JB") + (match_operand:M32 1 "memory_operand" "Q,R,R,Q")] + UNSPEC_REAL_LOAD)] + "" + "%|%.\\tld<LDST_SUFFIX>\\t%$\\t%1, %k0" + [(set_attr "type" "load") + (set_attr "units" "d_addr") + (set_attr "addr_regfile" "a,b,b,a") + (set_attr "dest_regfile" "a,a,b,b") + (set_attr "cross" "n,y,n,y")]) + +(define_insn "real_load<mode>" + [(unspec [(match_operand 0 "const_int_operand" "JA,JA,JB,JB") + (match_operand:DIDFVM 1 "memory_operand" "Q,R,R,Q")] + UNSPEC_REAL_LOAD)] + "TARGET_LDDW" + "%|%.\\tlddw\\t%$\\t%1, %K0" + [(set_attr "type" "load") + (set_attr "units" "d_addr") + (set_attr "addr_regfile" "a,b,b,a") + (set_attr "dest_regfile" "a,a,b,b") + (set_attr "cross" "n,y,n,y")]) + +(define_insn "load_shadow" + [(set (match_operand 0 "register_operand" "=ab") + (unspec [(pc)] UNSPEC_LOAD_SHADOW))] + "" + ";; load to %0 occurs" + [(set_attr "type" "load_shadow")]) + +(define_insn "mult_shadow" + [(set (match_operand 0 "register_operand" "=ab") + (unspec [(pc)] UNSPEC_MULT_SHADOW))] + "" + ";; multiplication occurs and stores to %0" + [(set_attr "type" "mult_shadow")]) + + +(define_mode_iterator MOV [QI HI SI SF DI DF V2HI V4QI V2SI V4HI V8QI]) + +(define_expand "mov<mode>" + [(set (match_operand:MOV 0 "nonimmediate_operand" "") + (match_operand:MOV 1 "general_operand" ""))] + "" +{ + if (expand_move (operands, <MODE>mode)) + DONE; +}) + +(define_expand "movmisalign<mode>" + [(set (match_operand:SIDIVM 0 "nonimmediate_operand" "") + (unspec:SIDIVM [(match_operand:SIDIVM 1 "nonimmediate_operand" "")] + UNSPEC_MISALIGNED_ACCESS))] + "TARGET_INSNS_64" +{ + if (memory_operand (operands[0], <MODE>mode)) + { + emit_insn (gen_movmisalign<mode>_store (operands[0], operands[1])); + DONE; + } +}) + +(define_insn_and_split "movmisalign<mode>_store" + [(set (match_operand:SIDIVM 0 "memory_operand" "=W,Q,T,Q,T") + (unspec:SIDIVM [(match_operand:SIDIVM 1 "register_operand" "r,a,b,b,a")] + UNSPEC_MISALIGNED_ACCESS)) + (clobber (match_scratch:SI 2 "=r,X,X,X,X"))] + "TARGET_INSNS_64" + "@ + # + %|%.\\tstn<LDST_SUFFIX>\\t%$\\t%1, %0 + %|%.\\tstn<LDST_SUFFIX>\\t%$\\t%1, %0 + %|%.\\tstn<LDST_SUFFIX>\\t%$\\t%1, %0 + %|%.\\tstn<LDST_SUFFIX>\\t%$\\t%1, %0" + "&& reload_completed && satisfies_constraint_W (operands[0])" + [(parallel + [(set (match_dup 3) (unspec:SIDIVM [(match_dup 1)] UNSPEC_MISALIGNED_ACCESS)) + (clobber (match_dup 4))])] +{ + rtx addr = XEXP (operands[0], 0); + rtx tmpreg = operands[2]; + + if (GET_CODE (addr) == PLUS && XEXP (addr, 0) == stack_pointer_rtx + && GET_CODE (XEXP (addr, 1)) == CONST_INT) + { + unsigned HOST_WIDE_INT val = INTVAL (XEXP (addr, 1)); + val &= GET_MODE_SIZE (<MODE>mode) - 1; + if (val == 0) + { + emit_move_insn (operands[0], operands[1]); + DONE; + } + } + operands[3] = change_address (operands[0], <MODE>mode, tmpreg); + emit_move_insn (tmpreg, addr); + operands[4] = gen_rtx_SCRATCH (SImode); +} + [(set_attr "type" "storen") + (set_attr "units" "d_addr") + (set_attr "addr_regfile" "*,a,b,a,b") + (set_attr "dest_regfile" "*,a,b,b,a") + (set_attr "cross" "*,n,n,y,y")]) + +(define_insn_and_split "movmisalign<mode>_load" + [(set (match_operand:SIDIVM 0 "register_operand" "=ab,a,b,b,a") + (unspec:SIDIVM [(match_operand:SIDIVM 1 "memory_operand" "W,Q,T,Q,T")] + UNSPEC_MISALIGNED_ACCESS))] + "TARGET_INSNS_64" + "@ + # + %|%.\\tldn<LDST_SUFFIX>\\t%$\\t%1, %0 + %|%.\\tldn<LDST_SUFFIX>\\t%$\\t%1, %0 + %|%.\\tldn<LDST_SUFFIX>\\t%$\\t%1, %0 + %|%.\\tldn<LDST_SUFFIX>\\t%$\\t%1, %0" + "&& reload_completed && satisfies_constraint_W (operands[1])" + [(set (match_dup 0) (unspec:SIDIVM [(match_dup 2)] UNSPEC_MISALIGNED_ACCESS))] +{ + rtx addr = XEXP (operands[1], 0); + rtx tmpreg = (GET_MODE (operands[0]) == SImode ? operands[0] + : operand_subword_force (operands[0], 0, DImode)); + + if (GET_CODE (addr) == PLUS && XEXP (addr, 0) == stack_pointer_rtx + && GET_CODE (XEXP (addr, 1)) == CONST_INT) + { + unsigned HOST_WIDE_INT val = INTVAL (XEXP (addr, 1)); + val &= GET_MODE_SIZE (<MODE>mode) - 1; + if (val == 0) + { + emit_move_insn (operands[0], operands[1]); + DONE; + } + } + operands[2] = change_address (operands[1], <MODE>mode, tmpreg); + emit_move_insn (tmpreg, addr); +} + [(set_attr "type" "loadn") + (set_attr "units" "d_addr") + (set_attr "addr_regfile" "*,a,b,a,b") + (set_attr "dest_regfile" "*,a,b,b,a") + (set_attr "cross" "*,n,n,y,y")]) + +;; + +;; ------------------------------------------------------------------------- +;; Extensions/extractions +;; ------------------------------------------------------------------------- + +(define_code_iterator any_extract [zero_extract sign_extract]) +(define_code_iterator any_ext [zero_extend sign_extend]) + +(define_code_attr ext_name [(zero_extend "zero_extend") (sign_extend "sign_extend")]) + +(define_code_attr u [(zero_extend "u") (sign_extend "")]) + +(define_code_attr z [(zero_extract "z") (sign_extract "")]) +(define_code_attr zu [(zero_extract "u") (sign_extract "")]) + +(define_mode_attr ext_shift [(QI "24") (HI "16")]) + +(define_insn "<ext_name><mode>si2" + [(set (match_operand:SI 0 "register_operand" "=a,b,a,?a, b,?b") + (any_ext: SI (match_operand:QIHIM 1 "nonimmediate_operand" "a,b,Q, R, R, Q")))] + "" + "@ + %|%.\\text<u>\\t%$\\t%1, <ext_shift>, <ext_shift>, %0 + %|%.\\text<u>\\t%$\\t%1, <ext_shift>, <ext_shift>, %0 + %|%.\\tld<LDST_SUFFIX><u>\\t%$\\t%1, %0 + %|%.\\tld<LDST_SUFFIX><u>\\t%$\\t%1, %0 + %|%.\\tld<LDST_SUFFIX><u>\\t%$\\t%1, %0 + %|%.\\tld<LDST_SUFFIX><u>\\t%$\\t%1, %0" + [(set_attr "type" "*,*,load,load,load,load") + (set_attr "units" "s,s,d_addr,d_addr,d_addr,d_addr") + (set_attr "addr_regfile" "*,*,a,b,b,a") + (set_attr "dest_regfile" "*,*,a,a,b,b") + (set_attr "cross" "n,n,n,y,n,y")]) + +(define_insn "*ext<z>v_const" + [(set (match_operand:SI 0 "nonimmediate_operand" "=a,b") + (any_extract:SI (match_operand:SI 1 "register_operand" "a,b") + (match_operand:SI 2 "const_int_operand" "n,n") + (match_operand:SI 3 "const_int_operand" "n,n")))] + "INTVAL (operands[3]) >= 0 + && INTVAL (operands[2]) + INTVAL (operands[3]) <= 32" +{ + int pos = INTVAL (operands[3]); + int len = INTVAL (operands[2]); + rtx xop[4]; + xop[0] = operands[0]; + xop[1] = operands[1]; + xop[2] = GEN_INT (32 - pos - len); + xop[3] = GEN_INT (32 - len); + + output_asm_insn ("%|%.\\text<zu>\\t%$\\t%1, %2, %3, %0", xop); + return ""; +} + [(set_attr "units" "s") + (set_attr "cross" "n")]) + +(define_expand "ext<z>v" + [(set (match_operand:SI 0 "register_operand" "") + (any_extract:SI (match_operand:SI 1 "register_operand" "") + (match_operand:SI 2 "const_int_operand" "") + (match_operand:SI 3 "const_int_operand" "")))] + "" +{ + if (INTVAL (operands[2]) < 0 + || INTVAL (operands[2]) + INTVAL (operands[3]) > 32) + FAIL; +}) + +(define_insn "real_<ext_name><mode>" + [(unspec [(match_operand 0 "const_int_operand" "JA,JA,JB,JB") + (any_ext:SI (match_operand:QIHIM 1 "memory_operand" "Q,R,R,Q"))] + UNSPEC_REAL_LOAD)] + "" + "%|%.\\tld<LDST_SUFFIX><u>\\t%$\\t%1, %k0" + [(set_attr "type" "load") + (set_attr "units" "d_addr") + (set_attr "addr_regfile" "a,b,b,a") + (set_attr "dest_regfile" "a,a,b,b") + (set_attr "cross" "n,y,n,y")]) + +(define_insn "clrr" + [(set (match_operand:SI 0 "register_operand" "=a,b,a,b") + (unspec:SI [(match_operand:SI 1 "register_operand" "0,0,0,0") + (match_operand:SI 2 "register_operand" "a,b,?b,?a") + (match_operand:SI 3 "reg_or_const_int_operand" "ai,bi,a,b")] + UNSPEC_CLR))] + "" +{ + if (CONST_INT_P (operands[2])) + { + rtx xops[4]; + int v1 = INTVAL (operands[2]); + int v2 = (v1 >> 5) & 0x1f; + v1 &= 0x1f; + xops[0] = operands[0]; + xops[1] = operands[1]; + xops[2] = GEN_INT (v1); + xops[3] = GEN_INT (v2); + output_asm_insn ("%|%.\\tclr\\t%$\\t%1, %3, %2, %0", xops); + return ""; + } + return "%|%.\\tclr\\t%$\\t%2, %3, %0"; +} + [(set_attr "units" "s") + (set_attr "cross" "n,n,y,y")]) + +(define_insn "extr" + [(set (match_operand:SI 0 "register_operand" "=a,b,a,b") + (unspec:SI [(match_operand:SI 1 "register_operand" "a,b,?b,?a") + (match_operand:SI 2 "reg_or_const_int_operand" "ai,bi,a,b")] + UNSPEC_EXT))] + "" +{ + if (CONST_INT_P (operands[2])) + { + rtx xops[4]; + int v1 = INTVAL (operands[2]); + int v2 = (v1 >> 5) & 0x1f; + v1 &= 0x1f; + xops[0] = operands[0]; + xops[1] = operands[1]; + xops[2] = GEN_INT (v1); + xops[3] = GEN_INT (v2); + output_asm_insn ("%|%.\\text\\t%$\\t%1, %3, %2, %0", xops); + return ""; + } + return "%|%.\\text\\t%$\\t%1, %2, %0"; +} + [(set_attr "units" "s") + (set_attr "cross" "n,n,y,y")]) + +(define_insn "extru" + [(set (match_operand:SI 0 "register_operand" "=a,b,a,b") + (unspec:SI [(match_operand:SI 1 "register_operand" "a,b,?b,?a") + (match_operand:SI 2 "reg_or_const_int_operand" "ai,bi,a,b")] + UNSPEC_EXTU))] + "" +{ + if (CONST_INT_P (operands[2])) + { + rtx xops[4]; + int v1 = INTVAL (operands[2]); + int v2 = (v1 >> 5) & 0x1f; + v1 &= 0x1f; + xops[0] = operands[0]; + xops[1] = operands[1]; + xops[2] = GEN_INT (v1); + xops[3] = GEN_INT (v2); + output_asm_insn ("%|%.\\textu\\t%$\\t%1, %3, %2, %0", xops); + return ""; + } + return "%|%.\\textu\\t%$\\t%1, %2, %0"; +} + [(set_attr "units" "s") + (set_attr "cross" "n,y,n,y")]) + +;; ------------------------------------------------------------------------- +;; Compare instructions +;; ------------------------------------------------------------------------- + +(define_insn "scmpsi_insn" + [(set (match_operand:SI 0 "register_operand" "=ab,a,b,a,b") + (match_operator:SI 1 "eqltgt_operator" + [(match_operand:SI 2 "register_operand" "ab,a,b,?b,?a") + (match_operand:SI 3 "reg_or_scst5_operand" "Is5,aIs5,bIs5,aIs5,bIs5")]))] + "" + "%|%.\\tcmp%C1\\t%$\\t%3, %2, %0" + [(set_attr "units" "l") + (set (attr "cross") + (symbol_ref "CROSS_OPERANDS (operands[0], operands[2])"))]) + +(define_insn "*ucmpsi_insn_64" + [(set (match_operand:SI 0 "register_operand" "=ab,a,b,a,b") + (match_operator:SI 1 "ltugtu_operator" + [(match_operand:SI 2 "register_operand" "ab,a,b,?b,?a") + (match_operand:SI 3 "reg_or_ucst5_operand" "Iu5,aIu5,bIu5,aIu5,bIu5")]))] + "TARGET_INSNS_64" + "%|%.\\tcmp%C1\\t%$\\t%3, %2, %0" + [(set_attr "units" "l") + (set (attr "cross") + (symbol_ref "CROSS_OPERANDS (operands[0], operands[2])"))]) + +(define_insn "*ucmpsi_insn" + [(set (match_operand:SI 0 "register_operand" "=ab,a,b,a,b") + (match_operator:SI 1 "ltugtu_operator" + [(match_operand:SI 2 "register_operand" "ab,a,b,?b,?a") + (match_operand:SI 3 "reg_or_ucst4_operand" "Iu4,aIu4,bIu4,aIu4,bIu4")]))] + "!TARGET_INSNS_64" + "%|%.\\tcmp%C1\\t%$\\t%3, %2, %0" + [(set_attr "units" "l") + (set (attr "cross") + (symbol_ref "CROSS_OPERANDS (operands[0], operands[2])"))]) + +(define_code_iterator andior_eqne [eq ne]) +(define_code_attr andior_name [(eq "and") (ne "ior")]) +(define_code_attr andior_condmod [(eq "") (ne "!")]) + +(define_insn "*scmpsi_<andior_name>_insn" + [(set (match_operand:SI 0 "register_operand" "=A,B,A,B") + (if_then_else:SI + (andior_eqne:SI (match_operand:SI 4 "register_operand" "0,0,0,0") + (const_int 0)) + (match_dup 4) + (match_operator:SI 1 "eqltgt_operator" + [(match_operand:SI 2 "register_operand" "a,b,?b,?a") + (match_operand:SI 3 "reg_or_scst5_operand" "aIs5,bIs5,aIs5,bIs5")])))] + "" + "%|[<andior_condmod>%4]\\tcmp%C1\\t%$\\t%3, %2, %0" + [(set_attr "units" "l") + (set_attr "cross" "n,n,y,y") + (set_attr "predicable" "no")]) + +(define_insn "*ucmpsi_<andior_name>_insn_64" + [(set (match_operand:SI 0 "register_operand" "=A,B,A,B") + (if_then_else:SI + (andior_eqne:SI (match_operand:SI 4 "register_operand" "0,0,0,0") + (const_int 0)) + (match_dup 4) + (match_operator:SI 1 "ltugtu_operator" + [(match_operand:SI 2 "register_operand" "a,b,?b,?a") + (match_operand:SI 3 "reg_or_ucst5_operand" "aIu5,bIu5,aIu5,bIu5")])))] + "TARGET_INSNS_64" + "%|[<andior_condmod>%4]\\tcmp%C1\\t%$\\t%3, %2, %0" + [(set_attr "units" "l") + (set_attr "cross" "n,n,y,y") + (set_attr "predicable" "no")]) + +(define_insn "*ucmpsi_<andior_name>_insn" + [(set (match_operand:SI 0 "register_operand" "=A,B,A,B") + (if_then_else:SI + (andior_eqne:SI (match_operand:SI 4 "register_operand" "0,0,0,0") + (const_int 0)) + (match_dup 4) + (match_operator:SI 1 "ltugtu_operator" + [(match_operand:SI 2 "register_operand" "a,b,?b,?a") + (match_operand:SI 3 "reg_or_ucst4_operand" "aIu4,bIu4,aIu4,bIu4")])))] + "!TARGET_INSNS_64" + "%|[<andior_condmod>%4]\\tcmp%C1\\t%$\\t%3, %2, %0" + [(set_attr "units" "l") + (set_attr "cross" "n,n,y,y") + (set_attr "predicable" "no")]) + +(define_expand "cmpsi_<andior_name>" + [(set (match_operand:SI 0 "register_operand" "") + (if_then_else:SI + (andior_eqne:SI (match_operand:SI 4 "register_operand" "0,0,0,0") + (const_int 0)) + (match_dup 4) + (match_operator:SI 1 "c6x_comparison_operator" + [(match_operand:SI 2 "register_operand" "") + (match_operand:SI 3 "reg_or_const_int_operand" "")])))] + "" +{ + if (c6x_force_op_for_comparison_p (GET_CODE (operands[1]), operands[3])) + operands[3] = force_reg (SImode, operands[3]); +}) + +(define_insn "*cmpsf_insn" + [(set (match_operand:SI 0 "register_operand" "=a,b,a,b") + (match_operator:SI 1 "eqltgt_operator" + [(match_operand:SF 2 "register_operand" "a,b,a,b") + (match_operand:SF 3 "register_operand" "a,b,?b,?a")]))] + "TARGET_FP" + "%|%.\\tcmp%c1sp\\t%$\\t%2, %3, %0" + [(set_attr "units" "s") + (set_attr "cross" "n,n,y,y")]) + +(define_insn "*cmpdf_insn" + [(set (match_operand:SI 0 "register_operand" "=a,b,a,b") + (match_operator:SI 1 "eqltgt_operator" + [(match_operand:DF 2 "register_operand" "a,b,a,b") + (match_operand:DF 3 "register_operand" "a,b,?b,?a")]))] + "TARGET_FP" + "%|%.\\tcmp%c1dp\\t%$\\t%2, %3, %0" + [(set_attr "type" "cmpdp") + (set_attr "units" "s") + (set_attr "cross" "n,n,y,y")]) + +(define_expand "cmp<mode>_<andior_name>" + [(set (match_operand:SI 0 "register_operand" "") + (if_then_else:SI + (andior_eqne:SI (match_operand:SI 4 "register_operand" "0,0,0,0") + (const_int 0)) + (match_dup 4) + (match_operator:SI 1 "eqltgt_operator" + [(match_operand:SFDFM 2 "register_operand" "") + (match_operand:SFDFM 3 "register_operand" "")])))] + "TARGET_FP") + +(define_insn "*cmpsf_<andior_name>_insn" + [(set (match_operand:SI 0 "register_operand" "=A,B,A,B") + (if_then_else:SI + (andior_eqne:SI (match_operand:SI 4 "register_operand" "0,0,0,0") + (const_int 0)) + (match_dup 4) + (match_operator:SI 1 "eqltgt_operator" + [(match_operand:SF 2 "register_operand" "a,b,a,b") + (match_operand:SF 3 "register_operand" "a,b,?b,?a")])))] + "TARGET_FP" + "%|[<andior_condmod>%4]\\tcmp%c1sp\\t%$\\t%2, %3, %0" + [(set_attr "units" "s") + (set_attr "cross" "n,n,y,y") + (set_attr "predicable" "no")]) + +;; reload_reg_class_lower will ensure that two-word reloads are allocated first, +;; which could exhaust the predicate registers if we used just "a" and "b" +;; constraints on operands 2 and 3. +(define_insn "*cmpdf_<andior_name>_insn" + [(set (match_operand:SI 0 "register_operand" "=A,B,A,B") + (if_then_else:SI + (andior_eqne:SI (match_operand:SI 4 "register_operand" "0,0,0,0") + (const_int 0)) + (match_dup 4) + (match_operator:SI 1 "eqltgt_operator" + [(match_operand:DF 2 "register_operand" "Da,Db,Da,Db") + (match_operand:DF 3 "register_operand" "Da,Db,?Db,?Da")])))] + "TARGET_FP" + "%|[<andior_condmod>%4]\\tcmp%c1dp\\t%$\\t%2, %3, %0" + [(set_attr "type" "cmpdp") + (set_attr "units" "s") + (set_attr "cross" "n,n,y,y") + (set_attr "predicable" "no")]) + +(define_split + [(set (match_operand:SI 0 "register_operand" "") + (ior:SI (match_operand 1 "c6x_any_comparison_operand" "") + (match_operand 2 "c6x_any_comparison_operand" "")))] + "!reg_overlap_mentioned_p (operands[0], operands[2])" + [(set (match_dup 0) (match_dup 1)) + (set (match_dup 0) + (if_then_else:SI (ne:SI (match_dup 0) (const_int 0)) + (match_dup 0) + (match_dup 2)))]) + +(define_split + [(set (match_operand:SI 0 "register_operand" "") + (and:SI (match_operand 1 "c6x_any_comparison_operand" "") + (match_operand 2 "c6x_any_comparison_operand" "")))] + "!reg_overlap_mentioned_p (operands[0], operands[2])" + [(set (match_dup 0) (match_dup 1)) + (set (match_dup 0) + (if_then_else:SI (eq:SI (match_dup 0) (const_int 0)) + (match_dup 0) + (match_dup 2)))]) + + +;; ------------------------------------------------------------------------- +;; setcc instructions +;; ------------------------------------------------------------------------- + +(define_expand "cstoresi4" + [(set (match_operand:SI 0 "register_operand" "") + (match_operator:SI 1 "comparison_operator" + [(match_operand:SI 2 "register_operand" "") + (match_operand:SI 3 "reg_or_ucst4_operand" "")]))] + "" +{ + if (!c6x_comparison_operator (operands[1], SImode)) + { + rtx tmpreg = gen_reg_rtx (SImode); + rtx t = gen_rtx_fmt_ee (reverse_condition (GET_CODE (operands[1])), + SImode, operands[2], operands[3]); + emit_insn (gen_rtx_SET (VOIDmode, tmpreg, t)); + emit_insn (gen_scmpsi_insn (operands[0], + gen_rtx_fmt_ee (EQ, SImode, tmpreg, const0_rtx), + tmpreg, const0_rtx)); + DONE; + } +}) + +;; ------------------------------------------------------------------------- +;; Jump instructions +;; ------------------------------------------------------------------------- + +(define_insn "indirect_jump" + [(set (pc) (match_operand:SI 0 "register_operand" "a,b"))] + "" + "%|%.\\tb\\t%$\\t%0" + [(set_attr "type" "branch") + (set_attr "units" "s") + (set_attr "cross" "y,n") + (set_attr "dest_regfile" "b")]) + +(define_insn "jump" + [(set (pc) + (label_ref (match_operand 0 "" "")))] + "" + "%|%.\\tb\\t%$\\t%l0" + [(set_attr "type" "branch") + (set_attr "units" "s") + (set_attr "dest_regfile" "any")]) + +(define_expand "tablejump" + [(parallel [(set (pc) (match_operand:SI 0 "register_operand" "")) + (use (label_ref (match_operand 1 "" "")))])] + "!flag_pic || !TARGET_INSNS_64" +{ +}) + +(define_insn "*tablejump_internal" + [(set (pc) (match_operand:SI 0 "register_operand" "b")) + (use (label_ref (match_operand 1 "" "")))] + "!flag_pic || !TARGET_INSNS_64" + "%|\\tb\\t%$\\t%0" + [(set_attr "type" "branch") + (set_attr "predicable" "no") + (set_attr "units" "s") + (set_attr "dest_regfile" "b")]) + +;; Implement switch statements when generating PIC code. Switches are +;; implemented by `tablejump' when not using -fpic. + +;; Emit code here to do the range checking and make the index zero based. +;; operand 0 is the index +;; operand 1 is the lower bound +;; operand 2 is the range of indices (highest - lowest + 1) +;; operand 3 is the label that precedes the table itself +;; operand 4 is the fall through label + +(define_expand "casesi" + [(use (match_operand:SI 0 "register_operand" "")) + (use (match_operand:SI 1 "const_int_operand" "")) + (use (match_operand:SI 2 "const_int_operand" "")) + (use (match_operand 3 "" "")) + (use (match_operand 4 "" ""))] + "flag_pic && TARGET_INSNS_64" +{ + rtx indx; + rtx low = operands[1]; + rtx range = operands[2]; + rtx table = operands[3]; + rtx fail = operands[4]; + + gcc_assert (GET_CODE (operands[1]) == CONST_INT); + gcc_assert (GET_CODE (operands[2]) == CONST_INT); + + if (!reg_or_ucst4_operand (range, SImode)) + range = force_reg (SImode, range); + + /* If low bound is 0, we don't have to subtract it. */ + if (INTVAL (operands[1]) == 0) + indx = operands[0]; + else + { + rtx offset = GEN_INT (-INTVAL (low)); + indx = gen_reg_rtx (SImode); + if (!addsi_operand (offset, SImode)) + offset = force_reg (SImode, offset); + emit_insn (gen_addsi3 (indx, operands[0], offset)); + } + emit_cmp_and_jump_insns (indx, range, GTU, NULL_RTX, SImode, 1, fail); + + emit_jump_insn (gen_casesi_internal (indx, table)); + DONE; +}) + +;; This is the only instance in this file where a pattern emits more than +;; one instruction. The concern here is that the addkpc insn could otherwise +;; be scheduled too far away from the label. A tablejump always ends an +;; extended basic block, so it shouldn't happen that the scheduler places +;; something in the delay slots. +(define_insn "casesi_internal" + [(set (pc) + (mem:SI (plus:SI (mult:SI (match_operand:SI 0 "register_operand" "b") + (const_int 4)) + (label_ref (match_operand 1 "" ""))))) + (clobber (match_scratch:SI 2 "=&b")) + (clobber (match_scratch:SI 3 "=b"))] + "flag_pic && TARGET_INSNS_64" + "addkpc\t.s2\t%l1,%2, 0\n\t\tldw\t.d2t2\t*+%2[%0], %3\n\t\tnop\t\t4\n\t\tadd\t.l2\t%2, %3, %3\n\t\tb\t.s2\t%3" + [(set_attr "type" "branch") + (set_attr "predicable" "no") + (set_attr "dest_regfile" "b")]) + +(define_expand "cbranch<mode>4" + [(set (pc) + (if_then_else (match_operator 0 "comparison_operator" + [(match_operand:SIDIM 1 "register_operand" "") + (match_operand:SIDIM 2 "reg_or_const_int_operand" "")]) + (label_ref (match_operand 3 "" "")) + (pc)))] + "" +{ + rtx t = c6x_expand_compare (operands[0], VOIDmode); + operands[0] = t; + operands[1] = XEXP (t, 0); + operands[2] = XEXP (t, 1); +}) + +(define_expand "cbranch<mode>4" + [(set (pc) + (if_then_else (match_operator 0 "c6x_fp_comparison_operator" + [(match_operand:SFDFM 1 "register_operand" "") + (match_operand:SFDFM 2 "register_operand" "")]) + (label_ref (match_operand 3 "" "")) + (pc)))] + "" +{ + rtx t = c6x_expand_compare (operands[0], VOIDmode); + operands[0] = t; + operands[1] = XEXP (t, 0); + operands[2] = XEXP (t, 1); +}) + +(define_insn "br_true" + [(set (pc) + (if_then_else (match_operator 0 "predicate_operator" + [(match_operand:SI 1 "register_operand" "AB") + (const_int 0)]) + (label_ref (match_operand 2 "" "")) + (pc)))] + "" + "%|[%J0]\\tb\\t%$\\t%l2" + [(set_attr "type" "branch") + (set_attr "predicable" "no") + (set_attr "units" "s") + (set_attr "dest_regfile" "any")]) + +(define_insn "br_false" + [(set (pc) + (if_then_else (match_operator 0 "predicate_operator" + [(match_operand:SI 1 "register_operand" "AB") + (const_int 0)]) + (pc) + (label_ref (match_operand 2 "" ""))))] + "" + "%|[%j0]\\tb\\t%$\\t%l2" + [(set_attr "type" "branch") + (set_attr "predicable" "no") + (set_attr "units" "s") + (set_attr "dest_regfile" "any")]) + +(define_expand "return" + [(parallel + [(return) + (use (reg:SI REG_B3))])] + "reload_completed && get_frame_size () == 0 && c6x_nsaved_regs () == 0") + +;; We can't expand this before we know where the link register is stored. +(define_insn_and_split "eh_return" + [(unspec_volatile [(match_operand:SI 0 "register_operand" "ab")] + UNSPECV_EH_RETURN) + (clobber (match_scratch:SI 1 "=&ab"))] + "" + "#" + "&& reload_completed" + [(const_int 0)] + " + { + c6x_set_return_address (operands[0], operands[1]); + DONE; + }" +) + +;; ------------------------------------------------------------------------- +;; Delayed-branch real jumps and shadows +;; ------------------------------------------------------------------------- + +(define_insn "real_jump" + [(unspec [(match_operand 0 "c6x_jump_operand" "a,b,s") (const_int 0)] + UNSPEC_REAL_JUMP)] + "" +{ + if (GET_CODE (operands[0]) == LABEL_REF) + return "%|%.\\tb\\t%$\\t%l0"; + return "%|%.\\tb\\t%$\\t%0"; +} + [(set_attr "type" "branch") + (set_attr "has_shadow" "y") + (set_attr "units" "s") + (set_attr "cross" "y,n,n") + (set_attr "dest_regfile" "b,b,any")]) + +(define_insn "real_call" + [(unspec [(match_operand 0 "c6x_call_operand" "a,b,S1") (const_int 1)] + UNSPEC_REAL_JUMP) + (clobber (reg:SI REG_B3))] + "" + "%|%.\\tcall\\t%$\\t%0" + [(set_attr "type" "call") + (set_attr "has_shadow" "y") + (set_attr "predicable" "no") + (set_attr "units" "s") + (set_attr "cross" "y,n,n") + (set_attr "dest_regfile" "b,b,any")]) + +(define_insn "real_ret" + [(unspec [(match_operand 0 "register_operand" "a,b") (const_int 2)] + UNSPEC_REAL_JUMP)] + "" + "%|%.\\tret\\t%$\\t%0" + [(set_attr "type" "branch") + (set_attr "has_shadow" "y") + (set_attr "units" "s") + (set_attr "cross" "y,n") + (set_attr "dest_regfile" "b")]) + +(define_insn "indirect_jump_shadow" + [(set (pc) (unspec [(pc)] UNSPEC_JUMP_SHADOW))] + "" + ";; indirect jump occurs" + [(set_attr "type" "shadow")]) + +;; Operand 0 may be a PARALLEL which isn't handled by output_operand, so +;; we don't try to print it. +(define_insn "indirect_call_value_shadow" + [(set (match_operand 0 "" "") + (call (unspec [(pc)] UNSPEC_JUMP_SHADOW) + (const_int 0)))] + "" + ";; indirect call occurs, with return value" + [(set_attr "type" "shadow")]) + +(define_insn "indirect_sibcall_shadow" + [(call (unspec [(pc)] UNSPEC_JUMP_SHADOW) + (const_int 0))] + "SIBLING_CALL_P (insn)" + ";; indirect sibcall occurs" + [(set_attr "type" "shadow")]) + +(define_insn "indirect_call_shadow" + [(call (unspec [(pc)] UNSPEC_JUMP_SHADOW) + (const_int 0))] + "" + ";; indirect call occurs" + [(set_attr "type" "shadow")]) + +(define_insn "call_value_shadow" + [(set (match_operand 0 "" "") + (call (unspec [(match_operand 1 "" "")] UNSPEC_JUMP_SHADOW) + (const_int 0)))] + "" + ";; call to %1 occurs, with return value" + [(set_attr "type" "shadow")]) + +(define_insn "call_shadow" + [(call (unspec [(match_operand 0 "" "")] UNSPEC_JUMP_SHADOW) + (const_int 0))] + "!SIBLING_CALL_P (insn)" + ";; call to %0 occurs" + [(set_attr "type" "shadow")]) + +(define_insn "sibcall_shadow" + [(call (unspec [(match_operand 0 "" "")] UNSPEC_JUMP_SHADOW) + (const_int 0))] + "SIBLING_CALL_P (insn)" + ";; sibcall to %0 occurs" + [(set_attr "type" "shadow")]) + +(define_insn "jump_shadow" + [(set (pc) (unspec [(match_operand 0 "" "")] UNSPEC_JUMP_SHADOW))] + "" + ";; jump to %0 occurs" + [(set_attr "type" "shadow")]) + +(define_insn "condjump_shadow" + [(set (pc) + (if_then_else (eq (unspec [(const_int 0)] UNSPEC_JUMP_SHADOW) + (const_int 0)) + (match_operand 0 "" "") + (pc)))] + "" + ";; condjump to %0 occurs" + [(set_attr "type" "shadow")]) + +(define_insn "return_shadow" + [(unspec [(const_int 0)] UNSPEC_JUMP_SHADOW) + (return)] + "" + ";; return occurs" + [(set_attr "type" "shadow")]) + +;; ------------------------------------------------------------------------- +;; Add instructions +;; ------------------------------------------------------------------------- + +(define_insn "addsi3" + [(set (match_operand:SI 0 "register_operand" + "=a ,b , a, b, a, b, a, b, ab, a, b, a, b,ab") + (plus:SI (match_operand:SI 1 "register_operand" + "%a ,b , a, b, b, a, b, a, 0, a, b, z, z,0") + (match_operand:SI 2 "addsi_operand" + "aIs5,bIs5,?b,?a,?a,?b,?aIs5,?bIs5,I5x,I5x,I5x,Iux,Iux,IsB")))] + "" +{ + if (CONSTANT_P (operands[2])) + { + HOST_WIDE_INT val = INTVAL (operands[2]); + + if (c6x_get_unit_specifier (insn) == 'd') + { + bool issp = (TARGET_INSNS_64PLUS + && operands[1] == stack_pointer_rtx + && GET_CODE (PATTERN (insn)) != COND_EXEC); + + if (get_attr_cross (insn) == CROSS_N) + { + if (satisfies_constraint_Iu5 (operands[2])) + return "%|%.\\tadd\\t%$\\t%1, %2, %0"; + else if (satisfies_constraint_In5 (operands[2])) + return "%|%.\\tsub\\t%$\\t%1, %n2, %0"; + } + + if (issp && val > 0 && val < 32768) + { + return "%|%.\\taddab\\t%$\\t%1, %2, %0"; + } + if ((val & 1) == 0 && ((val >= -62 && val <= 62) + || (issp && val > 0 && val < 65536))) + { + if (val < 0) + return "%|%.\\tsubah\\t%$\\t%1, %r2, %0"; + else + return "%|%.\\taddah\\t%$\\t%1, %r2, %0"; + } + else if ((val & 3) == 0 && ((val >= -124 && val <= 124) + || (issp && val > 0 && val < 131072))) + { + if (val < 0) + return "%|%.\\tsubaw\\t%$\\t%1, %R2, %0"; + else + return "%|%.\\taddaw\\t%$\\t%1, %R2, %0"; + } + else if ((val & 7) == 0 && val > 0 && val <= 248) + { + rtx xop[3]; + xop[0] = operands[0]; + xop[1] = operands[1]; + xop[2] = GEN_INT (val >> 3); + output_asm_insn ("%|%.\\taddad\\t%$\\t%1, %2, %0", xop); + return ""; + } + } + else + { + if (satisfies_constraint_Is5 (operands[2])) + return "%|%.\\tadd\\t%$\\t%2, %1, %0"; + } + gcc_assert (rtx_equal_p (operands[0], operands[1])); + return "%|%.\\taddk\\t%$\\t%2, %0"; + } + if (which_alternative == 4 || which_alternative == 5) + return "%|%.\\tadd\\t%$\\t%2, %1, %0"; + else + return "%|%.\\tadd\\t%$\\t%1, %2, %0"; +} + [(set_attr "units62" "dls,dls,ls,ls,ls,ls,ls,ls,s,d,d,*,*,s") + (set_attr "units67" "dls,dls,ls,ls,ls,ls,ls,ls,ds,d,d,*,*,s") + (set_attr "units64" "dls,dls,dls,dls,dls,dls,ls,ls,ds,d,d,d,d,s") + (set_attr "cross" "n,n,y,y,y,y,y,y,n,n,n,y,n,n") + (set_attr "predicable" "yes,yes,yes,yes,yes,yes,yes,yes,yes,yes,yes,no,no,yes")]) + +(define_insn "subsi3" + [(set (match_operand:SI 0 "register_operand" "=a,b,a,b,a,b") + (minus:SI (match_operand:SI 1 "reg_or_scst5_operand" "a,b,aIs5,bIs5,bIs5,aIs5") + (match_operand:SI 2 "register_operand" "a,b,a,b,?a,?b")))] + "" + "%|%.\\tsub\\t%$\\t%1, %2, %0" + [(set_attr "units62" "dls,dls,ls,ls,l,l") + (set_attr "units64" "dls,dls,ls,ls,ls,ls") + (set_attr "cross" "n,n,n,n,y,y")]) + +(define_insn "*addshiftsi" + [(set (match_operand:SI 0 "register_operand" "=a,b") + (plus:SI (mult:SI (match_operand:SI 2 "register_operand" "a,b") + (match_operand:SI 3 "adda_scale_operand" "n,n")) + (match_operand:SI 1 "register_operand" "a,b")))] + "" + "%|%.\\tadda%d3\\t%$\\t%1, %2, %0" + [(set_attr "units" "d")]) + +(define_insn "*subshiftsi" + [(set (match_operand:SI 0 "register_operand" "=a,b") + (minus:SI (match_operand:SI 1 "register_operand" "a,b") + (mult:SI (match_operand:SI 2 "register_operand" "a,b") + (match_operand:SI 3 "suba_scale_operand" "n,n"))))] + "" + "%|%.\\tsuba%d3\\t%$\\t%1, %2, %0" + [(set_attr "units" "d")]) + +(define_insn "addsidi3_widen" + [(set (match_operand:DI 0 "register_operand" "=a,b,a,b") + (plus:DI (zero_extend:DI (match_operand:SI 1 "register_operand" "%a,b,a,b")) + (zero_extend:DI (match_operand:SI 2 "register_operand" "a,b,?b,?a"))))] + "" + "%|%.\\taddu\\t%$\\t%1, %2, %0" + [(set_attr "units" "l") + (set_attr "cross" "n,n,y,y")]) + +(define_expand "adddi3" + [(set (match_operand:DI 0 "register_operand" "") + (plus:DI (match_operand:DI 1 "register_operand" "") + (match_operand:DI 2 "register_operand" "")))] + "" +{ + rtx tmp; + rtx lo_half[3], hi_half[3]; + split_di (operands + 1, 2, lo_half + 1, hi_half + 1); + if (reg_overlap_mentioned_p (operands[0], hi_half[1]) + || reg_overlap_mentioned_p (operands[0], hi_half[2])) + tmp = gen_reg_rtx (DImode); + else + tmp = operands[0]; + split_di (&tmp, 1, lo_half, hi_half); + emit_insn (gen_addsidi3_widen (tmp, lo_half[1], lo_half[2])); + emit_insn (gen_addsi3 (hi_half[0], copy_rtx (hi_half[0]), hi_half[1])); + emit_insn (gen_addsi3 (copy_rtx (hi_half[0]), + copy_rtx (hi_half[0]), hi_half[2])); + if (tmp != operands[0]) + emit_move_insn (operands[0], tmp); + DONE; +}) + +(define_insn "addsf3" + [(set (match_operand:SF 0 "register_operand" "=a,b,a,b") + (plus:SF (match_operand:SF 1 "register_operand" "%a,b,a,b") + (match_operand:SF 2 "register_operand" "a,b,?b,?a")))] + "TARGET_FP" + "%|%.\\taddsp\\t%$\\t%1, %2, %0" + [(set_attr "type" "fp4") + (set_attr "units67" "l") + (set_attr "units67p" "ls") + (set_attr "units674" "ls") + (set_attr "cross" "n,n,y,y")]) + +(define_insn "adddf3" + [(set (match_operand:DF 0 "register_operand" "=a,b,a,b") + (plus:DF (match_operand:DF 1 "register_operand" "%a,b,a,b") + (match_operand:DF 2 "register_operand" "a,b,?b,?a")))] + "TARGET_FP" + "%|%.\\tadddp\\t%$\\t%1, %2, %0" + [(set_attr "type" "adddp") + (set_attr "units67" "l") + (set_attr "units67p" "ls") + (set_attr "units674" "ls") + (set_attr "cross" "n,n,y,y")]) + +(define_insn "subsf3" + [(set (match_operand:SF 0 "register_operand" "=a,b, a, b, a, b") + (minus:SF (match_operand:SF 1 "register_operand" "a,b, b, a, a, b") + (match_operand:SF 2 "register_operand" "a,b,?a,?b,?b,?a")))] + "TARGET_FP" + "%|%.\\tsubsp\\t%$\\t%1, %2, %0" + [(set_attr "type" "fp4") + (set_attr "units67" "l") + (set_attr "units67p" "ls") + (set_attr "units674" "ls") + (set_attr "cross" "n,n,y,y,y,y")]) + +(define_insn "subdf3" + [(set (match_operand:DF 0 "register_operand" "=a,b, a, b, a, b") + (minus:DF (match_operand:DF 1 "register_operand" "a,b, b, a, a, b") + (match_operand:DF 2 "register_operand" "a,b,?a,?b,?b,?a")))] + "TARGET_FP" + "%|%.\\tsubdp\\t%$\\t%1, %2, %0" + [(set_attr "type" "adddp") + (set_attr "units67" "l") + (set_attr "units67p" "ls") + (set_attr "units674" "ls") + (set_attr "cross" "n,n,y,y,y,y")]) + +;; ------------------------------------------------------------------------- +;; Logical instructions +;; ------------------------------------------------------------------------- + +(define_insn "andsi3" + [(set (match_operand:SI 0 "register_operand" "=a,b,a,b,a,b") + (and:SI (match_operand:SI 1 "register_operand" "%a,b,b,a,a,b") + (match_operand:SI 2 "andsi_operand" "aIs5,bIs5,?aIs5,?bIs5,aJc,bJc")))] + "" +{ + if (which_alternative < 4) + return "%|%.\\tand\\t%$\\t%2, %1, %0"; + else + return "%|%.\\tclr\\t%$\\t%1, %f2, %F2, %0"; +} + [(set_attr "units62" "ls,ls,ls,ls,s,s") + (set_attr "units64" "dls,dls,dls,dls,s,s") + (set_attr "cross" "n,n,y,y,n,n")]) + +(define_insn "iorsi3" + [(set (match_operand:SI 0 "register_operand" "=a,b,a,b,a,b") + (ior:SI (match_operand:SI 1 "register_operand" "%a,b,b,a,a,b") + (match_operand:SI 2 "iorsi_operand" "aIs5,bIs5,?aIs5,?bIs5,aJs,bJs")))] + "" +{ + if (which_alternative < 4) + return "%|%.\\tor\\t%$\\t%2, %1, %0"; + else + return "%|%.\\tset\\t%$\\t%1, %s2, %S2, %0"; +} + [(set_attr "units62" "ls,ls,ls,ls,s,s") + (set_attr "units64" "dls,dls,dls,dls,s,s") + (set_attr "cross" "n,n,y,y,n,n")]) + +(define_insn "xorsi3" + [(set (match_operand:SI 0 "register_operand" "=a,b,a,b") + (xor:SI (match_operand:SI 1 "register_operand" "%a,b,b,a") + (match_operand:SI 2 "reg_or_scst5_operand" "aIs5,bIs5,?aIs5,?bIs5")))] + "" + "%|%.\\txor\\t%$\\t%2, %1, %0" + [(set_attr "units62" "ls") + (set_attr "units64" "dls") + (set_attr "cross" "n,n,y,y")]) + +;; ------------------------------------------------------------------------- +;; Conversions +;; ------------------------------------------------------------------------- + +(define_insn "extendsfdf2" + [(set (match_operand:DF 0 "register_operand" "=a,b,a,b") + (float_extend:DF (match_operand:SF 1 "register_operand" "a,b,?b,?a")))] + "TARGET_FP" + "%|%.\\tspdp\\t%$\\t%1,%0" + [(set_attr "type" "dp2") + (set_attr "units" "s") + (set_attr "cross" "n,n,y,y")]) + +(define_insn "truncdfsf2" + [(set (match_operand:SF 0 "register_operand" "=a,b") + (float_truncate:SF (match_operand:DF 1 "register_operand" "a,b")))] + "TARGET_FP" + "%|%.\\tdpsp\\t%$\\t%1,%0" + [(set_attr "type" "fp4") + (set_attr "units" "l") + (set_attr "cross" "n")]) + +;;;; Convert between signed integer types and floating point. +(define_insn "floatsisf2" + [(set (match_operand:SF 0 "register_operand" "=a,b,a,b") + (float:SF (match_operand:SI 1 "register_operand" "a,b,?b,?a")))] + "TARGET_FP" + "%|%.\\tintsp\\t%$\\t%1,%0" + [(set_attr "type" "fp4") + (set_attr "units" "l") + (set_attr "cross" "n,n,y,y")]) + +(define_insn "floatunssisf2" + [(set (match_operand:SF 0 "register_operand" "=a,b,a,b") + (unsigned_float:SF (match_operand:SI 1 "register_operand" "a,b,?b,?a")))] + "TARGET_FP" + "%|%.\\tintspu\\t%$\\t%1,%0" + [(set_attr "type" "fp4") + (set_attr "units" "l") + (set_attr "cross" "n,n,y,y")]) + +(define_insn "floatsidf2" + [(set (match_operand:DF 0 "register_operand" "=a,b,a,b") + (float:DF (match_operand:SI 1 "register_operand" "a,b,?b,?a")))] + "TARGET_FP" + "%|%.\\tintdp\\t%$\\t%1,%0" + [(set_attr "type" "intdp") + (set_attr "units" "l") + (set_attr "cross" "n,n,y,y")]) + +(define_insn "floatunssidf2" + [(set (match_operand:DF 0 "register_operand" "=a,b,a,b") + (unsigned_float:DF (match_operand:SI 1 "register_operand" "a,b,?b,?a")))] + "TARGET_FP" + "%|%.\\tintdpu\\t%$\\t%1,%0" + [(set_attr "type" "intdp") + (set_attr "units" "l") + (set_attr "cross" "n,n,y,y")]) + +(define_insn "fix_truncsfsi2" + [(set (match_operand:SI 0 "register_operand" "=a,b,a,b") + (fix:SI (match_operand:SF 1 "register_operand" "a,b,?b,?a")))] + "TARGET_FP" + "%|%.\\tsptrunc\\t%$\\t%1,%0" + [(set_attr "type" "fp4") + (set_attr "units" "l") + (set_attr "cross" "n,n,y,y")]) + +(define_insn "fix_truncdfsi2" + [(set (match_operand:SI 0 "register_operand" "=a,b") + (fix:SI (match_operand:DF 1 "register_operand" "a,b")))] + "TARGET_FP" + "%|%.\\tdptrunc\\t%$\\t%1,%0" + [(set_attr "type" "fp4") + (set_attr "units" "l") + (set_attr "cross" "n")]) + +;; ------------------------------------------------------------------------- +;; Saturating arithmetic +;; ------------------------------------------------------------------------- + +(define_insn "saddsi3" + [(set (match_operand:SI 0 "register_operand" "=a,b,a,b,a,b,a,b") + (ss_plus:SI (match_operand:SI 1 "register_operand" "a,b,?b,?a,a,b,?b,?a") + (match_operand:SI 2 "reg_or_const_int_operand" "a,b,a,b,aIs5,bIs5,aIs5,bIs5")))] + "" + "%|%.\\tsadd\\t%$\\t%2, %1, %0" + [(set_attr "units" "ls,ls,ls,ls,l,l,l,l") + (set_attr "cross" "n,n,y,y,n,n,y,y")]) + +(define_insn "ssubsi3" + [(set (match_operand:SI 0 "register_operand" "=a,b,a,b") + (ss_minus:SI (match_operand:SI 1 "reg_or_scst5_operand" "aIs5,bIs5,?bIs5,?aIs5") + (match_operand:SI 2 "register_operand" "a,b,a,b")))] + "" + "%|%.\\tssub\\t%$\\t%1, %2, %0" + [(set_attr "units" "l") + (set_attr "cross" "n,n,y,y")]) + +(define_insn "subcsi3" + [(set (match_operand:SI 0 "register_operand" "=a,b,a,b") + (unspec:SI + [(match_operand:SI 1 "register_operand" "a,b,a,b") + (match_operand:SI 2 "register_operand" "a,b,?b,?a")] + UNSPEC_SUBC))] + "" + "%|%.\\tsubc\\t%$\\t%1, %2, %0" + [(set_attr "units" "l") + (set_attr "cross" "n,n,y,y")]) + +;; ------------------------------------------------------------------------- +;; Call instructions +;; ------------------------------------------------------------------------- + +(define_expand "call" + [(match_operand 0 "" "")] + "" +{ + c6x_expand_call (NULL_RTX, operands[0], false); + DONE; +}) + +(define_expand "call_value" + [(match_operand 0 "" "") + (match_operand 1 "" "")] + "" +{ + c6x_expand_call (operands[0], operands[1], false); + DONE; +}) + +(define_expand "sibcall" + [(match_operand 0 "" "")] + "" +{ + c6x_expand_call (NULL_RTX, operands[0], true); + cfun->machine->contains_sibcall = true; + DONE; +}) + +(define_expand "sibcall_value" + [(match_operand 0 "" "") + (match_operand 1 "" "")] + "" +{ + c6x_expand_call (operands[0], operands[1], true); + cfun->machine->contains_sibcall = true; + DONE; +}) + +(define_insn "call_internal" + [(call (mem (match_operand:SI 0 "c6x_call_operand" "S1,a,b")) + (const_int 0))] + "!SIBLING_CALL_P (insn)" + "%|%.\\tcall\\t%$\\t%0" + [(set_attr "type" "call") + (set_attr "predicable" "no") + (set_attr "units" "s") + (set_attr "dest_regfile" "any,b,b") + (set_attr "cross" "n,y,n")]) + +(define_insn "call_value_internal" + [(set (match_operand 0 "" "") + (call (mem (match_operand:SI 1 "c6x_call_operand" "S1,a,b")) + (const_int 0)))] + "" + "%|%.\\tcall\\t%$\\t%1" + [(set_attr "type" "call") + (set_attr "predicable" "no") + (set_attr "units" "s") + (set_attr "dest_regfile" "any,b,b") + (set_attr "cross" "n,y,n")]) + +(define_insn "sibcall_internal" + [(call (mem (match_operand:SI 0 "c6x_call_operand" "S1,C")) + (const_int 0))] + "SIBLING_CALL_P (insn)" + "%|%.\\tb\\t%$\\t%0" + [(set_attr "type" "branch") + (set_attr "predicable" "no") + (set_attr "units" "s") + (set_attr "dest_regfile" "any,b")]) + +(define_insn "callp" + [(call (mem (match_operand:SI 0 "c6x_call_operand" "S1")) + (const_int 0)) + (unspec [(const_int 6)] UNSPEC_NOP)] + "!SIBLING_CALL_P (insn)" + "%|%.\\tcallp\\t%$\\t%0, B3" + [(set_attr "type" "callp") + (set_attr "predicable" "no") + (set_attr "units" "s") + (set_attr "dest_regfile" "b") + (set_attr "cross" "n")]) + +(define_insn "callp_value" + [(set (match_operand:SI 0 "register_operand" "") + (call (mem (match_operand:SI 1 "c6x_call_operand" "S1")) + (const_int 0))) + (unspec [(const_int 6)] UNSPEC_NOP)] + "!SIBLING_CALL_P (insn)" + "%|%.\\tcallp\\t%$\\t%1, B3" + [(set_attr "type" "callp") + (set_attr "predicable" "no") + (set_attr "units" "s") + (set_attr "dest_regfile" "b") + (set_attr "cross" "n")]) + +(define_insn "return_internal" + [(return) + (use (match_operand:SI 0 "register_operand" "b"))] + "reload_completed" + "%|%.\\tret\\t%$\\t%0" + [(set_attr "type" "branch") + (set_attr "units" "s") + (set_attr "dest_regfile" "b")]) + +(define_insn "addkpc" + [(set (match_operand:SI 0 "register_operand" "=b") + (unspec:SI [(match_operand 1 "" "")] UNSPEC_ADDKPC)) + (unspec [(match_operand 2 "const_int_operand" "n")] UNSPEC_NOP)] + "TARGET_INSNS_64" + "%|%.\\taddkpc\\t%$\\t%l1, %0, %2" + [(set_attr "units" "s") + (set_attr "dest_regfile" "b")]) + +;; ------------------------------------------------------------------------- +;; Unary operations +;; ------------------------------------------------------------------------- + +(define_insn "negsi2" + [(set (match_operand:SI 0 "register_operand" "=a, a, b, b") + (neg:SI (match_operand:SI 1 "register_operand" "a,?b, b,?a")))] + "" + "%|%.\\tneg\\t%$\\t%1, %0" + [(set_attr "units" "ls") + (set_attr "cross" "n,y,n,y")]) + +(define_insn "one_cmplsi2" + [(set (match_operand:SI 0 "register_operand" "=a, a, b, b") + (not:SI (match_operand:SI 1 "register_operand" "a,?b, b,?a")))] + "" + "%|%.\\tnot\\t%$\\t%1, %0" + [(set_attr "units" "ls") + (set_attr "cross" "n,y,n,y")]) + +(define_insn "clrsbsi2" + [(set (match_operand:SI 0 "register_operand" "=a, a, b, b") + (clrsb:SI (match_operand:SI 1 "register_operand" "a,?b, b,?a")))] + "" + "%|%.\\tnorm\\t%$\\t%1, %0" + [(set_attr "units" "l") + (set_attr "cross" "n,y,n,y")]) + +(define_insn "clzsi2" + [(set (match_operand:SI 0 "register_operand" "=a, a, b, b") + (clz:SI (match_operand:SI 1 "register_operand" "a,?b, b,?a")))] + "" + "%|%.\\tlmbd\\t%$\\t1, %1, %0" + [(set_attr "units" "l") + (set_attr "cross" "n,y,n,y")]) + +;; bitrevsi2 is defined in c6x-mult.md.in. + +(define_expand "ctzsi2" + [(set (match_operand:SI 0 "register_operand" "") + (ctz:SI (match_operand:SI 1 "register_operand" "")))] + "TARGET_INSNS_64" +{ + rtx tmpreg = gen_reg_rtx (SImode); + emit_insn (gen_bitrevsi2 (tmpreg, operands[1])); + emit_insn (gen_clzsi2 (operands[0], tmpreg)); + DONE; +}) + +(define_expand "ctzdi2" + [(set (match_operand:DI 0 "register_operand" "") + (ctz:SI (match_operand:DI 1 "register_operand" "")))] + "TARGET_INSNS_64" +{ + rtx tmpreg = gen_reg_rtx (DImode); + rtx out; + emit_insn (gen_bitrevsi2 (gen_highpart (SImode, tmpreg), + gen_lowpart (SImode, operands[1]))); + emit_insn (gen_bitrevsi2 (gen_lowpart (SImode, tmpreg), + gen_highpart (SImode, operands[1]))); + out = expand_unop (DImode, clz_optab, tmpreg, operands[0], 1); + if (!rtx_equal_p (out, operands[0])) + emit_move_insn (operands[0], out); + DONE; +}) + +(define_insn "ssabssi2" + [(set (match_operand:SI 0 "register_operand" "=a, a, b, b") + (ss_abs:SI (match_operand:SI 1 "register_operand" "a,?b, b,?a")))] + "" + "%|%.\\tabs\\t%$\\t%1, %0" + [(set_attr "units" "l") + (set_attr "cross" "n,y,n,y")]) + +;; ------------------------------------------------------------------------- +;; Shift instructions +;; ------------------------------------------------------------------------- + +(define_code_iterator any_shift [ss_ashift ashift ashiftrt lshiftrt]) +(define_code_iterator any_rshift [ashiftrt lshiftrt]) +(define_code_attr shift_code [(ss_ashift "ss_ashl") (ashift "ashl") + (ashiftrt "ashr") (lshiftrt "lshr")]) +(define_code_attr shift_insn [(ss_ashift "sshl") (ashift "shl") + (ashiftrt "shr") (lshiftrt "shru")]) + +(define_insn "<shift_code>si3" + [(set (match_operand:SI 0 "register_operand" "=a,b,a,b") + (any_shift:SI (match_operand:SI 1 "register_operand" "a,b,?b,?a") + (match_operand:SI 2 "reg_or_ucst5_operand" "aIu5,bIu5,aIu5,bIu5")))] + "" + "%|%.\\t<shift_insn>\\t%$\\t%1, %2, %0" + [(set_attr "units" "s") + (set_attr "cross" "n,n,y,y")]) + +;; See c6x-mult.md.in for the rotlsi3 pattern. + +(define_insn "rotrdi3_16" + [(set (match_operand:DI 0 "register_operand" "=a,b") + (rotatert:DI (match_operand:DI 1 "register_operand" "a,b") + (const_int 16)))] + "TARGET_INSNS_64PLUS" + "%|%.\\tdpackx2\\t%$\\t%P1, %p1, %0" + [(set_attr "units" "l") + (set_attr "cross" "n")]) + +(define_insn "shlmbsi3" + [(set (match_operand:SI 0 "register_operand" "=a,b,a,b") + (ior:SI (ashift:SI (match_operand:SI 1 "register_operand" "a,b,?b,?a") + (const_int 8)) + (lshiftrt:SI (match_operand:SI 2 "register_operand" "a,b,a,b") + (const_int 24))))] + "TARGET_INSNS_64" + "%|%.\\tshlmb\\t%$\\t%2, %1, %0" + [(set_attr "units" "ls") + (set_attr "cross" "n,n,y,y")]) + +(define_expand "ashldi3" + [(set (match_operand:DI 0 "register_operand" "") + (ashift:DI (match_operand:DI 1 "register_operand" "") + (match_operand:SI 2 "const_int_operand" "")))] + "TARGET_INSNS_64" +{ + if (CONST_INT_P (operands[2]) && INTVAL (operands[2]) == 8) + { + rtx lo0, lo1, hi0, hi1, tmp; + lo0 = gen_lowpart (SImode, operands[0]); + hi0 = gen_highpart (SImode, operands[0]); + lo1 = gen_lowpart (SImode, operands[1]); + hi1 = gen_highpart (SImode, operands[1]); + if (reg_overlap_mentioned_p (hi0, lo1)) + tmp = gen_reg_rtx (SImode); + else + tmp = hi0; + emit_insn (gen_shlmbsi3 (tmp, hi1, lo1)); + emit_insn (gen_ashlsi3 (lo0, lo1, operands[2])); + if (tmp != hi0) + emit_move_insn (hi0, tmp); + DONE; + } + FAIL; +}) + +(define_expand "rotrdi3" + [(set (match_operand:DI 0 "register_operand" "") + (rotatert:DI (match_operand:DI 1 "register_operand" "") + (match_operand:SI 2 "const_int_operand" "")))] + "TARGET_INSNS_64PLUS" +{ + if (CONST_INT_P (operands[2]) && INTVAL (operands[2]) == 16) + { + emit_insn (gen_rotrdi3_16 (operands[0], operands[1])); + DONE; + } + FAIL; +}) + +(define_insn "bswapv2hi2" + [(set (match_operand:V2HI 0 "register_operand" "=a,b,a,b") + (bswap:V2HI (match_operand:V2HI 1 "register_operand" "a,b,?b,?a")))] + "TARGET_INSNS_64" + "%|%.\\tswap4\\t%$\\t%1, %0" + [(set_attr "units" "l") + (set_attr "cross" "n,n,y,y")]) + +(define_expand "bswapsi2" + [(set (match_operand:SI 0 "register_operand" "") + (bswap:SI (match_operand:SI 1 "register_operand" "")))] + "TARGET_INSNS_64" +{ + rtx tmpreg = gen_reg_rtx (SImode); + rtx tmpv2 = gen_lowpart (V2HImode, tmpreg); + rtx op0v2 = gen_lowpart (V2HImode, operands[0]); + emit_insn (gen_rotlsi3 (tmpreg, operands[1], GEN_INT (16))); + emit_insn (gen_bswapv2hi2 (op0v2, tmpv2)); + DONE; +}) + +;; ------------------------------------------------------------------------- +;; Division +;; ------------------------------------------------------------------------- + +(define_insn "divsi3_insn" + [(set (reg:SI REG_A4) (div:SI (reg:SI REG_A4) (reg:SI REG_B4))) + (clobber (reg:SI REG_A0)) + (clobber (reg:SI REG_A1)) + (clobber (reg:SI REG_A2)) + (clobber (reg:SI REG_A6)) + (clobber (reg:SI REG_B0)) + (clobber (reg:SI REG_B1)) + (clobber (reg:SI REG_B2)) + (clobber (reg:SI REG_B3)) + (clobber (reg:SI REG_B4)) + (clobber (reg:SI REG_B5)) + (clobber (reg:SI REG_B30)) + (clobber (reg:SI REG_B31))] + "" + "%|%.\\tcall\\t%$\\t__c6xabi_divi" + [(set_attr "type" "call") + (set_attr "dest_regfile" "any") + (set_attr "units" "s") + (set_attr "cross" "n")]) + +(define_insn "divsi3_insn_indcall" + [(set (reg:SI REG_A4) (div:SI (reg:SI REG_A4) (reg:SI REG_B4))) + (use (match_operand:SI 0 "register_operand" "b")) + (clobber (reg:SI REG_A0)) + (clobber (reg:SI REG_A1)) + (clobber (reg:SI REG_A2)) + (clobber (reg:SI REG_A6)) + (clobber (reg:SI REG_B0)) + (clobber (reg:SI REG_B1)) + (clobber (reg:SI REG_B2)) + (clobber (reg:SI REG_B3)) + (clobber (reg:SI REG_B4)) + (clobber (reg:SI REG_B5)) + (clobber (reg:SI REG_B30)) + (clobber (reg:SI REG_B31))] + "" + "%|%.\\tcall\\t%$\\t%0" + [(set_attr "type" "call") + (set_attr "dest_regfile" "any") + (set_attr "units" "s") + (set_attr "cross" "n")]) + +(define_insn "udivsi3_insn" + [(set (reg:SI REG_A4) (udiv:SI (reg:SI REG_A4) (reg:SI REG_B4))) + (clobber (reg:SI REG_A0)) + (clobber (reg:SI REG_A1)) + (clobber (reg:SI REG_A2)) + (clobber (reg:SI REG_A6)) + (clobber (reg:SI REG_B0)) + (clobber (reg:SI REG_B1)) + (clobber (reg:SI REG_B2)) + (clobber (reg:SI REG_B3)) + (clobber (reg:SI REG_B4)) + (clobber (reg:SI REG_B30)) + (clobber (reg:SI REG_B31))] + "" + "%|%.\\tcall\\t%$\\t__c6xabi_divu" + [(set_attr "type" "call") + (set_attr "dest_regfile" "any") + (set_attr "units" "s") + (set_attr "cross" "n")]) + +(define_insn "udivsi3_insn_indcall" + [(set (reg:SI REG_A4) (udiv:SI (reg:SI REG_A4) (reg:SI REG_B4))) + (use (match_operand:SI 0 "register_operand" "b")) + (clobber (reg:SI REG_A0)) + (clobber (reg:SI REG_A1)) + (clobber (reg:SI REG_A2)) + (clobber (reg:SI REG_A6)) + (clobber (reg:SI REG_B0)) + (clobber (reg:SI REG_B1)) + (clobber (reg:SI REG_B2)) + (clobber (reg:SI REG_B3)) + (clobber (reg:SI REG_B4)) + (clobber (reg:SI REG_B30)) + (clobber (reg:SI REG_B31))] + "" + "%|%.\\tcall\\t%$\\t%0" + [(set_attr "type" "call") + (set_attr "dest_regfile" "any") + (set_attr "units" "s") + (set_attr "cross" "n")]) + +(define_insn "modsi3_insn" + [(set (reg:SI REG_A4) (mod:SI (reg:SI REG_A4) (reg:SI REG_B4))) + (clobber (reg:SI REG_A1)) + (clobber (reg:SI REG_A2)) + (clobber (reg:SI REG_A5)) + (clobber (reg:SI REG_A6)) + (clobber (reg:SI REG_B0)) + (clobber (reg:SI REG_B1)) + (clobber (reg:SI REG_B2)) + (clobber (reg:SI REG_B3)) + (clobber (reg:SI REG_B4)) + (clobber (reg:SI REG_B30)) + (clobber (reg:SI REG_B31))] + "" + "%|%.\\tcall\\t%$\\t__c6xabi_remi" + [(set_attr "type" "call") + (set_attr "dest_regfile" "any") + (set_attr "units" "s") + (set_attr "cross" "n")]) + +(define_insn "modsi3_insn_indcall" + [(set (reg:SI REG_A4) (mod:SI (reg:SI REG_A4) (reg:SI REG_B4))) + (use (match_operand:SI 0 "register_operand" "b")) + (clobber (reg:SI REG_A1)) + (clobber (reg:SI REG_A2)) + (clobber (reg:SI REG_A5)) + (clobber (reg:SI REG_A6)) + (clobber (reg:SI REG_B0)) + (clobber (reg:SI REG_B1)) + (clobber (reg:SI REG_B2)) + (clobber (reg:SI REG_B3)) + (clobber (reg:SI REG_B4)) + (clobber (reg:SI REG_B30)) + (clobber (reg:SI REG_B31))] + "" + "%|%.\\tcall\\t%$\\t%0" + [(set_attr "type" "call") + (set_attr "dest_regfile" "any") + (set_attr "units" "s") + (set_attr "cross" "n")]) + +(define_insn "divmodsi4_insn" + [(set (reg:SI REG_A4) (div:SI (reg:SI REG_A4) (reg:SI REG_B4))) + (set (reg:SI REG_A5) (mod:SI (reg:SI REG_A4) (reg:SI REG_B4))) + (clobber (reg:SI REG_A1)) + (clobber (reg:SI REG_A2)) + (clobber (reg:SI REG_A6)) + (clobber (reg:SI REG_B0)) + (clobber (reg:SI REG_B1)) + (clobber (reg:SI REG_B2)) + (clobber (reg:SI REG_B3)) + (clobber (reg:SI REG_B4)) + (clobber (reg:SI REG_B30)) + (clobber (reg:SI REG_B31))] + "" + "%|%.\\tcall\\t%$\\t__c6xabi_divremi" + [(set_attr "type" "call") + (set_attr "dest_regfile" "any") + (set_attr "units" "s") + (set_attr "cross" "n")]) + +(define_insn "divmodsi4_insn_indcall" + [(set (reg:SI REG_A4) (div:SI (reg:SI REG_A4) (reg:SI REG_B4))) + (set (reg:SI REG_A5) (mod:SI (reg:SI REG_A4) (reg:SI REG_B4))) + (use (match_operand:SI 0 "register_operand" "b")) + (clobber (reg:SI REG_A1)) + (clobber (reg:SI REG_A2)) + (clobber (reg:SI REG_A5)) + (clobber (reg:SI REG_A6)) + (clobber (reg:SI REG_B0)) + (clobber (reg:SI REG_B1)) + (clobber (reg:SI REG_B2)) + (clobber (reg:SI REG_B3)) + (clobber (reg:SI REG_B4)) + (clobber (reg:SI REG_B30)) + (clobber (reg:SI REG_B31))] + "" + "%|%.\\tcall\\t%$\\t%0" + [(set_attr "type" "call") + (set_attr "dest_regfile" "any") + (set_attr "units" "s") + (set_attr "cross" "n")]) + +(define_insn "umodsi3_insn" + [(set (reg:SI REG_A4) (umod:SI (reg:SI REG_A4) (reg:SI REG_B4))) + (clobber (reg:SI REG_A1)) + (clobber (reg:SI REG_A5)) + (clobber (reg:SI REG_A7)) + (clobber (reg:SI REG_B0)) + (clobber (reg:SI REG_B1)) + (clobber (reg:SI REG_B2)) + (clobber (reg:SI REG_B3)) + (clobber (reg:SI REG_B4)) + (clobber (reg:SI REG_B30)) + (clobber (reg:SI REG_B31))] + "" + "%|%.\\tcall\\t%$\\t__c6xabi_remu" + [(set_attr "type" "call") + (set_attr "dest_regfile" "any") + (set_attr "units" "s") + (set_attr "cross" "n")]) + +(define_insn "umodsi3_insn_indcall" + [(set (reg:SI REG_A4) (umod:SI (reg:SI REG_A4) (reg:SI REG_B4))) + (use (match_operand:SI 0 "register_operand" "b")) + (clobber (reg:SI REG_A1)) + (clobber (reg:SI REG_A5)) + (clobber (reg:SI REG_A7)) + (clobber (reg:SI REG_B0)) + (clobber (reg:SI REG_B1)) + (clobber (reg:SI REG_B2)) + (clobber (reg:SI REG_B3)) + (clobber (reg:SI REG_B4)) + (clobber (reg:SI REG_B30)) + (clobber (reg:SI REG_B31))] + "" + "%|%.\\tcall\\t%$\\t%0" + [(set_attr "type" "call") + (set_attr "dest_regfile" "any") + (set_attr "units" "s") + (set_attr "cross" "n")]) + +(define_insn "udivmodsi4_insn" + [(set (reg:SI REG_A4) (udiv:SI (reg:SI REG_A4) (reg:SI REG_B4))) + (set (reg:SI REG_A5) (umod:SI (reg:SI REG_A4) (reg:SI REG_B4))) + (clobber (reg:SI REG_A0)) + (clobber (reg:SI REG_A1)) + (clobber (reg:SI REG_A2)) + (clobber (reg:SI REG_A6)) + (clobber (reg:SI REG_B0)) + (clobber (reg:SI REG_B1)) + (clobber (reg:SI REG_B2)) + (clobber (reg:SI REG_B3)) + (clobber (reg:SI REG_B4)) + (clobber (reg:SI REG_B30)) + (clobber (reg:SI REG_B31))] + "" + "%|%.\\tcall\\t%$\\t__c6xabi_divremu" + [(set_attr "type" "call") + (set_attr "dest_regfile" "any") + (set_attr "units" "s") + (set_attr "cross" "n")]) + +(define_insn "udivmodsi4_insn_indcall" + [(set (reg:SI REG_A4) (udiv:SI (reg:SI REG_A4) (reg:SI REG_B4))) + (set (reg:SI REG_A5) (umod:SI (reg:SI REG_A4) (reg:SI REG_B4))) + (use (match_operand:SI 0 "register_operand" "b")) + (clobber (reg:SI REG_A0)) + (clobber (reg:SI REG_A1)) + (clobber (reg:SI REG_A2)) + (clobber (reg:SI REG_A6)) + (clobber (reg:SI REG_B0)) + (clobber (reg:SI REG_B1)) + (clobber (reg:SI REG_B2)) + (clobber (reg:SI REG_B3)) + (clobber (reg:SI REG_B4)) + (clobber (reg:SI REG_B30)) + (clobber (reg:SI REG_B31))] + "" + "%|%.\\tcall\\t%$\\t%0" + [(set_attr "type" "call") + (set_attr "dest_regfile" "any") + (set_attr "units" "s") + (set_attr "cross" "n")]) + +(define_insn_and_split "divmodsi4" + [(set (match_operand:SI 0 "register_operand" "") + (div:SI (match_operand:SI 1 "register_operand" "") + (match_operand:SI 2 "register_operand" ""))) + (set (match_operand:SI 3 "register_operand" "") + (mod:SI (match_dup 1) (match_dup 2))) + (clobber (reg:SI REG_A0)) + (clobber (reg:SI REG_A1)) + (clobber (reg:SI REG_A2)) + (clobber (reg:SI REG_A4)) + (clobber (reg:SI REG_A5)) + (clobber (reg:SI REG_A6)) + (clobber (reg:SI REG_B0)) + (clobber (reg:SI REG_B1)) + (clobber (reg:SI REG_B2)) + (clobber (reg:SI REG_B3)) + (clobber (reg:SI REG_B4)) + (clobber (reg:SI REG_B5)) + (clobber (reg:SI REG_B30)) + (clobber (reg:SI REG_B31))] + "" + "#" + "" + [(const_int 0)] +{ + rtx reg = NULL_RTX; + + if (TARGET_LONG_CALLS) + { + if (reload_completed) + reg = gen_rtx_REG (SImode, REG_A6); + else + reg = gen_reg_rtx (SImode); + } + emit_move_insn (gen_rtx_REG (SImode, REG_A4), operands[1]); + emit_move_insn (gen_rtx_REG (SImode, REG_B4), operands[2]); + if (find_reg_note (curr_insn, REG_UNUSED, operands[3])) + { + if (TARGET_LONG_CALLS) + { + emit_move_insn (reg, optab_libfunc (sdiv_optab, SImode)); + emit_insn (gen_divsi3_insn_indcall (reg)); + } + else + emit_insn (gen_divsi3_insn ()); + emit_move_insn (operands[0], gen_rtx_REG (SImode, REG_A4)); + } + else if (find_reg_note (curr_insn, REG_UNUSED, operands[0])) + { + if (TARGET_LONG_CALLS) + { + emit_move_insn (reg, optab_libfunc (smod_optab, SImode)); + emit_insn (gen_modsi3_insn_indcall (reg)); + } + else + emit_insn (gen_modsi3_insn ()); + emit_move_insn (operands[3], gen_rtx_REG (SImode, REG_A4)); + } + else + { + if (TARGET_LONG_CALLS) + { + emit_move_insn (reg, optab_libfunc (sdivmod_optab, SImode)); + emit_insn (gen_divmodsi4_insn_indcall (reg)); + } + else + emit_insn (gen_divmodsi4_insn ()); + emit_move_insn (operands[0], gen_rtx_REG (SImode, REG_A4)); + emit_move_insn (operands[3], gen_rtx_REG (SImode, REG_A5)); + } + DONE; +}) + +(define_insn_and_split "udivmodsi4" + [(set (match_operand:SI 0 "register_operand" "") + (udiv:SI (match_operand:SI 1 "register_operand" "") + (match_operand:SI 2 "register_operand" ""))) + (set (match_operand:SI 3 "register_operand" "") + (umod:SI (match_dup 1) (match_dup 2))) + (clobber (reg:SI REG_A0)) + (clobber (reg:SI REG_A1)) + (clobber (reg:SI REG_A2)) + (clobber (reg:SI REG_A4)) + (clobber (reg:SI REG_A5)) + (clobber (reg:SI REG_A6)) + (clobber (reg:SI REG_A7)) + (clobber (reg:SI REG_B0)) + (clobber (reg:SI REG_B1)) + (clobber (reg:SI REG_B2)) + (clobber (reg:SI REG_B3)) + (clobber (reg:SI REG_B4)) + (clobber (reg:SI REG_B30)) + (clobber (reg:SI REG_B31))] + "" + "#" + "" + [(const_int 0)] +{ + rtx reg = NULL_RTX; + + if (TARGET_LONG_CALLS) + { + if (reload_completed) + reg = gen_rtx_REG (SImode, REG_A6); + else + reg = gen_reg_rtx (SImode); + } + + emit_move_insn (gen_rtx_REG (SImode, REG_A4), operands[1]); + emit_move_insn (gen_rtx_REG (SImode, REG_B4), operands[2]); + if (find_reg_note (curr_insn, REG_UNUSED, operands[3])) + { + if (TARGET_LONG_CALLS) + { + emit_move_insn (reg, optab_libfunc (udiv_optab, SImode)); + emit_insn (gen_udivsi3_insn_indcall (reg)); + } + else + emit_insn (gen_udivsi3_insn ()); + emit_move_insn (operands[0], gen_rtx_REG (SImode, REG_A4)); + } + else if (find_reg_note (curr_insn, REG_UNUSED, operands[0])) + { + if (TARGET_LONG_CALLS) + { + emit_move_insn (reg, optab_libfunc (umod_optab, SImode)); + emit_insn (gen_umodsi3_insn_indcall (reg)); + } + else + emit_insn (gen_umodsi3_insn ()); + emit_move_insn (operands[3], gen_rtx_REG (SImode, REG_A4)); + } + else + { + if (TARGET_LONG_CALLS) + { + emit_move_insn (reg, optab_libfunc (udivmod_optab, SImode)); + emit_insn (gen_udivmodsi4_insn_indcall (reg)); + } + else + emit_insn (gen_udivmodsi4_insn ()); + emit_move_insn (operands[0], gen_rtx_REG (SImode, REG_A4)); + emit_move_insn (operands[3], gen_rtx_REG (SImode, REG_A5)); + } + DONE; +}) + +;; ------------------------------------------------------------------------- +;; Multiplication +;; See c6x-mult.md.in for define_insn patterns. +;; ------------------------------------------------------------------------- + +(define_expand "mulhisi3" + [(set (match_operand:SI 0 "register_operand" "") + (mult:SI (sign_extend:SI (match_operand:HI 1 "register_operand" "")) + (sign_extend:SI (match_operand:HI 2 "reg_or_scst5_operand" ""))))] + "" +{ + if (CONSTANT_P (operands[2])) + { + emit_insn (gen_mulhisi3_const (operands[0], operands[1], operands[2])); + DONE; + } +}) + +(define_expand "usmulhisi3" + [(set (match_operand:SI 0 "register_operand" "") + (mult:SI (zero_extend:SI (match_operand:HI 1 "register_operand" "")) + (sign_extend:SI (match_operand:HI 2 "reg_or_scst5_operand" ""))))] + "" +{ + if (CONSTANT_P (operands[2])) + { + emit_insn (gen_usmulhisi3_const (operands[0], operands[1], operands[2])); + DONE; + } +}) + +(define_expand "mulsi3" + [(set (match_operand:SI 0 "register_operand" "") + (mult:SI (match_operand:SI 1 "register_operand" "") + (match_operand:SI 2 "register_operand" "")))] + "" +{ + if (!TARGET_MPY32) + { + rtx lo1 = gen_lowpart (HImode, operands[1]); + rtx lo2 = gen_lowpart (HImode, operands[2]); + /* (N * AH + AL) * (N * BH + BL) + = N*(AH * BL + BH * AL) + AL*BL */ + rtx tmp1 = gen_reg_rtx (SImode); + rtx tmp2 = gen_reg_rtx (SImode); + rtx tmp3 = gen_reg_rtx (SImode); + emit_insn (gen_umulhisi3 (tmp1, lo1, lo2)); + emit_insn (gen_umulhisi3_lh (tmp2, lo1, operands[2])); + emit_insn (gen_umulhisi3_hl (tmp3, operands[1], lo2)); + emit_insn (gen_addsi3 (tmp2, tmp2, tmp3)); + emit_insn (gen_ashlsi3 (tmp2, tmp2, GEN_INT (16))); + emit_insn (gen_addsi3 (operands[0], tmp1, tmp2)); + DONE; + } +}) + +;; ------------------------------------------------------------------------- +;; Floating point multiplication +;; ------------------------------------------------------------------------- + +(define_insn "mulsf3" + [(set (match_operand:SF 0 "register_operand" "=a,b,a,b") + (mult:SF (match_operand:SF 1 "register_operand" "%a,b,?a,?b") + (match_operand:SF 2 "register_operand" "a,b,b,a")))] + "TARGET_FP" + "%|%.\\tmpysp\\t%$\\t%1, %2, %0" + [(set_attr "type" "mpy4") + (set_attr "units" "m") + (set_attr "cross" "n,n,y,y")]) + +(define_insn "muldf3" + [(set (match_operand:DF 0 "register_operand" "=a,b") + (mult:DF (match_operand:DF 1 "register_operand" "%a,b") + (match_operand:DF 2 "register_operand" "a,b")))] + "TARGET_FP" + "%|%.\\tmpydp\\t%$\\t%1, %2, %0" + [(set_attr "type" "mpydp") + (set_attr "units" "m") + (set_attr "cross" "n")]) + +;; Note that mpyspdp and mpysp2dp are available on C67x, despite what the +;; manual says. +(define_insn "*muldf_ext1" + [(set (match_operand:DF 0 "register_operand" "=a,b,a,b") + (mult:DF (float_extend:DF (match_operand:SF 1 "register_operand" "a,b,a,b")) + (match_operand:DF 2 "register_operand" "a,b,?b,?a")))] + "TARGET_FP_EXT" + "%|%.\\tmpyspdp\\t%$\\t%1, %2, %0" + [(set_attr "type" "mpyspdp") + (set_attr "units" "m") + (set_attr "cross" "n,n,y,y")]) + +(define_insn "*muldf_ext2" + [(set (match_operand:DF 0 "register_operand" "=a,b,a,b") + (mult:DF (float_extend:DF (match_operand:SF 1 "register_operand" "%a,b,a,b")) + (float_extend:DF (match_operand:SF 2 "register_operand" "a,b,?b,?a"))))] + "TARGET_FP_EXT" + "%|%.\\tmpysp2dp\\t%$\\t%1, %2, %0" + [(set_attr "type" "mpysp2dp") + (set_attr "units" "m") + (set_attr "cross" "n,n,y,y")]) + +;; ------------------------------------------------------------------------- +;; Floating point division +;; ------------------------------------------------------------------------- + +(define_insn "rcpsf2" + [(set (match_operand:SF 0 "register_operand" "=a,b,a,b") + (unspec:SF [(match_operand:SF 1 "register_operand" "a,b,?b,?a")] + UNSPEC_RCP))] + "TARGET_FP" + "%|%.\\trcpsp\\t%$\\t%1, %0" + [(set_attr "units" "s") + (set_attr "cross" "n,n,y,y")]) + +(define_insn "rcpdf2" + [(set (match_operand:DF 0 "register_operand" "=a,b") + (unspec:DF [(match_operand:DF 1 "register_operand" "a,b")] + UNSPEC_RCP))] + "TARGET_FP" + "%|%.\\trcpdp\\t%$\\t%1, %0" + [(set_attr "type" "dp2") + (set_attr "units" "s") + (set_attr "cross" "n")]) + +(define_expand "divsf3" + [(set (match_dup 4) + (unspec:SF [(match_operand:SF 2 "register_operand" "")] + UNSPEC_RCP)) + (set (match_dup 5) (mult:SF (match_dup 2) (match_dup 4))) + (set (match_dup 6) (minus:SF (match_dup 3) (match_dup 5))) + (set (match_dup 4) (mult:SF (match_dup 4) (match_dup 6))) + (set (match_dup 5) (mult:SF (match_dup 2) (match_dup 4))) + (set (match_dup 6) (minus:SF (match_dup 3) (match_dup 5))) + (set (match_dup 4) (mult:SF (match_dup 4) (match_dup 6))) + (set (match_operand:SF 0 "register_operand" "") + (mult:SF (match_operand:SF 1 "register_operand") + (match_dup 4)))] + "TARGET_FP && flag_reciprocal_math" +{ + operands[3] = force_reg (SFmode, + CONST_DOUBLE_FROM_REAL_VALUE (dconst2, SFmode)); + operands[4] = gen_reg_rtx (SFmode); + operands[5] = gen_reg_rtx (SFmode); + operands[6] = gen_reg_rtx (SFmode); +}) + +(define_expand "divdf3" + [(set (match_dup 4) + (unspec:DF [(match_operand:DF 2 "register_operand" "")] + UNSPEC_RCP)) + (set (match_dup 5) (mult:DF (match_dup 2) (match_dup 4))) + (set (match_dup 6) (minus:DF (match_dup 3) (match_dup 5))) + (set (match_dup 4) (mult:DF (match_dup 4) (match_dup 6))) + (set (match_dup 5) (mult:DF (match_dup 2) (match_dup 4))) + (set (match_dup 6) (minus:DF (match_dup 3) (match_dup 5))) + (set (match_dup 4) (mult:DF (match_dup 4) (match_dup 6))) + (set (match_dup 5) (mult:DF (match_dup 2) (match_dup 4))) + (set (match_dup 6) (minus:DF (match_dup 3) (match_dup 5))) + (set (match_dup 4) (mult:DF (match_dup 4) (match_dup 6))) + (set (match_operand:DF 0 "register_operand" "") + (mult:DF (match_operand:DF 1 "register_operand") + (match_dup 4)))] + "TARGET_FP && flag_reciprocal_math" +{ + operands[3] = force_reg (DFmode, + CONST_DOUBLE_FROM_REAL_VALUE (dconst2, DFmode)); + operands[4] = gen_reg_rtx (DFmode); + operands[5] = gen_reg_rtx (DFmode); + operands[6] = gen_reg_rtx (DFmode); +}) + +;; ------------------------------------------------------------------------- +;; Block moves +;; ------------------------------------------------------------------------- + +(define_expand "movmemsi" + [(use (match_operand:BLK 0 "memory_operand" "")) + (use (match_operand:BLK 1 "memory_operand" "")) + (use (match_operand:SI 2 "nonmemory_operand" "")) + (use (match_operand:SI 3 "const_int_operand" "")) + (use (match_operand:SI 4 "const_int_operand" "")) + (use (match_operand:SI 5 "const_int_operand" ""))] + "" +{ + if (c6x_expand_movmem (operands[0], operands[1], operands[2], operands[3], + operands[4], operands[5])) + DONE; + else + FAIL; +}) + +;; ------------------------------------------------------------------------- +;; Prologue and epilogue. +;; ------------------------------------------------------------------------- + +;; UNSPEC_VOLATILE is considered to use and clobber all hard registers and +;; all of memory. This blocks insns from being moved across this point. + +(define_insn "blockage" + [(unspec_volatile [(const_int 0)] UNSPECV_BLOCKAGE)] + "" + "" + [(set_attr "type" "blockage")]) + +(define_insn "push_rts" + [(set (mem:SI (reg:SI REG_SP)) (reg:SI REG_B14)) + (set (mem:DI (plus:SI (reg:SI REG_SP) (const_int -8))) (reg:DI REG_A14)) + (set (mem:DI (plus:SI (reg:SI REG_SP) (const_int -16))) (reg:DI REG_B12)) + (set (mem:DI (plus:SI (reg:SI REG_SP) (const_int -24))) (reg:DI REG_A12)) + (set (mem:DI (plus:SI (reg:SI REG_SP) (const_int -32))) (reg:DI REG_B10)) + (set (mem:DI (plus:SI (reg:SI REG_SP) (const_int -40))) (reg:DI REG_A10)) + (set (mem:DI (plus:SI (reg:SI REG_SP) (const_int -48))) (reg:DI REG_B2)) + (set (reg:SI REG_SP) (plus:SI (reg:SI REG_SP) (const_int -56))) + (unspec_volatile [(const_int 0)] UNSPECV_BLOCKAGE) + (clobber (reg:SI REG_A3))] + "TARGET_INSNS_64PLUS" + "%|%.\\tcallp\\t%$\\t__c6xabi_push_rts, a3" + [(set_attr "type" "callp") + (set_attr "dest_regfile" "a") + (set_attr "units" "s") + (set_attr "cross" "n")]) + +(define_insn "pop_rts" + [(set (reg:SI REG_B14) (mem:SI (plus:SI (reg:SI REG_SP) (const_int 56)))) + (set (reg:DI REG_A14) (mem:DI (plus:SI (reg:SI REG_SP) (const_int 48)))) + (set (reg:DI REG_B12) (mem:DI (plus:SI (reg:SI REG_SP) (const_int 40)))) + (set (reg:DI REG_A12) (mem:DI (plus:SI (reg:SI REG_SP) (const_int 32)))) + (set (reg:DI REG_B10) (mem:DI (plus:SI (reg:SI REG_SP) (const_int 24)))) + (set (reg:DI REG_A10) (mem:DI (plus:SI (reg:SI REG_SP) (const_int 16)))) + (set (reg:DI REG_B2) (mem:DI (plus:SI (reg:SI REG_SP) (const_int 8)))) + (set (reg:SI REG_SP) (plus:SI (reg:SI REG_SP) (const_int 56))) + (clobber (reg:SI REG_A3)) + (return)] + "TARGET_INSNS_64PLUS" + "%|%.\\tretp\\t%$\\t__c6xabi_pop_rts, a3" + [(set_attr "type" "callp") + (set_attr "dest_regfile" "a") + (set_attr "units" "s") + (set_attr "cross" "n")]) + +(define_expand "prologue" + [(const_int 1)] + "" + "c6x_expand_prologue (); DONE;") + +(define_expand "epilogue" + [(const_int 1)] + "" + "c6x_expand_epilogue (false); DONE;") + +(define_expand "sibcall_epilogue" + [(return)] + "" +{ + c6x_expand_epilogue (true); + DONE; +}) + +(define_insn "setup_dsbt" + [(set (match_operand:SI 0 "pic_register_operand" "+Z") + (unspec:SI [(match_dup 0) + (match_operand:SI 1 "symbolic_operand" "")] + UNSPEC_SETUP_DSBT))] + "TARGET_DSBT" + "%|%.\\tldw\\t%$\\t*+%0($DSBT_index%1), %0" + [(set_attr "type" "load") + (set_attr "units" "d_addr") + (set_attr "dest_regfile" "b") + (set_attr "addr_regfile" "b")]) + + +;; A dummy use/set to prevent prologue and epiloge overlapping. +;; This can be caused by sched-ebb in the presence of multiple +;; exit sequences, and causes the unwinding table generation to explode. +(define_insn "epilogue_barrier" + [(set (match_operand:SI 0 "register_operand" "") + (unspec:SI [(match_operand:SI 1 "register_operand" "")] + UNSPEC_EPILOGUE_BARRIER))] + "" + "" + [(set_attr "type" "blockage")]) + +;; ------------------------------------------------------------------------- +;; Vector insns +;; ------------------------------------------------------------------------- + +(define_code_iterator logical [and ior xor]) +(define_code_attr logical_insn [(and "and") (ior "ior") (xor "xor")]) +(define_code_attr logical_opcode [(and "and") (ior "or") (xor "xor")]) +(define_code_iterator plusminus [plus minus]) +(define_code_attr plusminus_insn [(plus "add") (minus "sub")]) +(define_code_iterator ss_plusminus [ss_plus ss_minus]) +(define_code_attr ss_plusminus_insn [(ss_plus "add") (ss_minus "sub")]) + +;; Vector logical insns + +(define_insn "<logical_insn><mode>3" + [(set (match_operand:VEC4M 0 "register_operand" "=a,b,a,b") + (logical:VEC4M (match_operand:VEC4M 1 "register_operand" "a,b,a,b") + (match_operand:VEC4M 2 "register_operand" "a,b,?b,?a")))] + "" + "%|%.\\t<logical_opcode>\\t%$\\t%1, %2, %0" + [(set_attr "units62" "ls") + (set_attr "units64" "dls") + (set_attr "cross" "n,n,y,y")]) + +;; Vector add/subtract + +(define_insn "<plusminus_insn>v2hi3" + [(set (match_operand:V2HI 0 "register_operand" "=a,b,a,b") + (plusminus:V2HI (match_operand:V2HI 1 "register_operand" "a,b,a,b") + (match_operand:V2HI 2 "register_operand" "a,b,?b,?a")))] + "" + "%|%.\\t<plusminus_insn>2\\t%$\\t%1, %2, %0" + [(set_attr "units62" "l") + (set_attr "units64" "dls") + (set_attr "cross" "n,n,y,y")]) + +(define_insn "<plusminus_insn>v4qi3" + [(set (match_operand:V4QI 0 "register_operand" "=a,b,a,b") + (plusminus:V4QI (match_operand:V4QI 1 "register_operand" "a,b,a,b") + (match_operand:V4QI 2 "register_operand" "a,b,?b,?a")))] + "TARGET_INSNS_64" + "%|%.\\t<plusminus_insn>4\\t%$\\t%1, %2, %0" + [(set_attr "units" "l") + (set_attr "cross" "n,n,y,y")]) + +(define_insn "ss_addv2hi3" + [(set (match_operand:V2HI 0 "register_operand" "=a,b,a,b") + (ss_plus:V2HI (match_operand:V2HI 1 "register_operand" "a,b,a,b") + (match_operand:V2HI 2 "register_operand" "a,b,?b,?a")))] + "TARGET_INSNS_64" + "%|%.\\tsadd2\\t%$\\t%1, %2, %0" + [(set_attr "units" "s") + (set_attr "cross" "n,n,y,y")]) + +(define_insn "ss_subv2hi3" + [(set (match_operand:V2HI 0 "register_operand" "=a,b,a,b") + (ss_minus:V2HI (match_operand:V2HI 1 "register_operand" "a,b,a,b") + (match_operand:V2HI 2 "register_operand" "a,b,?b,?a")))] + "TARGET_INSNS_64" + "%|%.\\tssub2\\t%$\\t%1, %2, %0" + [(set_attr "units" "l") + (set_attr "cross" "n,n,y,y")]) + +(define_insn "us_addv4qi3" + [(set (match_operand:V4QI 0 "register_operand" "=a,b,a,b") + (ss_plus:V4QI (match_operand:V4QI 1 "register_operand" "a,b,a,b") + (match_operand:V4QI 2 "register_operand" "a,b,?b,?a")))] + "TARGET_INSNS_64" + "%|%.\\tsaddu4\\t%$\\t%1, %2, %0" + [(set_attr "units" "s") + (set_attr "cross" "n,n,y,y")]) + +;; Vector/scalar min/max + +(define_mode_iterator SMINMAX [HI V2HI]) +(define_mode_iterator UMINMAX [QI V4QI]) + +(define_insn "smax<mode>3" + [(set (match_operand:SMINMAX 0 "register_operand" "=a,b,a,b") + (smax:SMINMAX (match_operand:SMINMAX 1 "register_operand" "a,b,a,b") + (match_operand:SMINMAX 2 "register_operand" "a,b,?b,?a")))] + "TARGET_INSNS_64" + "%|%.\\tmax2\\t%$\\t%1, %2, %0" + [(set_attr "units64" "l") + (set_attr "units64p" "ls") + (set_attr "cross" "n,n,y,y")]) + +(define_insn "smin<mode>3" + [(set (match_operand:SMINMAX 0 "register_operand" "=a,b,a,b") + (smin:SMINMAX (match_operand:SMINMAX 1 "register_operand" "a,b,a,b") + (match_operand:SMINMAX 2 "register_operand" "a,b,?b,?a")))] + "TARGET_INSNS_64" + "%|%.\\tmin2\\t%$\\t%1, %2, %0" + [(set_attr "units64" "l") + (set_attr "units64p" "ls") + (set_attr "cross" "n,n,y,y")]) + +(define_insn "umax<mode>3" + [(set (match_operand:UMINMAX 0 "register_operand" "=a,b,a,b") + (umax:UMINMAX (match_operand:UMINMAX 1 "register_operand" "a,b,a,b") + (match_operand:UMINMAX 2 "register_operand" "a,b,?b,?a")))] + "TARGET_INSNS_64" + "%|%.\\tmaxu4\\t%$\\t%1, %2, %0" + [(set_attr "units" "l") + (set_attr "cross" "n,n,y,y")]) + +(define_insn "umin<mode>3" + [(set (match_operand:UMINMAX 0 "register_operand" "=a,b,a,b") + (umin:UMINMAX (match_operand:UMINMAX 1 "register_operand" "a,b,a,b") + (match_operand:UMINMAX 2 "register_operand" "a,b,?b,?a")))] + "TARGET_INSNS_64" + "%|%.\\tminu4\\t%$\\t%1, %2, %0" + [(set_attr "units" "l") + (set_attr "cross" "n,n,y,y")]) + +;; Vector shifts + +(define_insn "<shift_code>v2hi3" + [(set (match_operand:V2HI 0 "register_operand" "=a,b,a,b") + (any_rshift:V2HI (match_operand:V2HI 1 "register_operand" "a,b,?b,?a") + (match_operand:SI 2 "reg_or_ucst5_operand" "aIu5,bIu5,aIu5,bIu5")))] + "TARGET_INSNS_64" + "%|%.\\t<shift_insn>2\\t%$\\t%1, %2, %0" + [(set_attr "units" "s") + (set_attr "cross" "n,n,y,y")]) + +;; See c6x-mult.md.in for avg2/avgu4 + +;; Widening vector multiply and dot product. +;; See c6x-mult.md.in for the define_insn patterns + +(define_expand "sdot_prodv2hi" + [(match_operand:SI 0 "register_operand" "") + (match_operand:V2HI 1 "register_operand" "") + (match_operand:V2HI 2 "register_operand" "") + (match_operand:SI 3 "register_operand" "")] + "TARGET_INSNS_64" +{ + rtx t = gen_reg_rtx (SImode); + emit_insn (gen_dotv2hi (t, operands[1], operands[2])); + emit_insn (gen_addsi3 (operands[0], operands[3], t)); + DONE; +}) + +;; Unary vector operations + +(define_insn "ssabsv2hi2" + [(set (match_operand:V2HI 0 "register_operand" "=a, a, b, b") + (ss_abs:V2HI (match_operand:V2HI 1 "register_operand" "a,?b, b,?a")))] + "TARGET_INSNS_64" + "%|%.\\tabs2\\t%$\\t%1, %0" + [(set_attr "units" "l") + (set_attr "cross" "n,y,n,y")]) + +;; Pack insns + +(define_insn "*packv2hi_insv" + [(set (zero_extract:SI (match_operand:SI 0 "register_operand" "+a,b,a,b,ab") + (const_int 16) + (const_int 16)) + (match_operand:SI 1 "nonmemory_operand" "a,b,?b,?a,n"))] + "TARGET_INSNS_64" + "@ + %|%.\\tpack2\\t%$\\t%1, %0, %0 + %|%.\\tpack2\\t%$\\t%1, %0, %0 + %|%.\\tpack2\\t%$\\t%1, %0, %0 + %|%.\\tpack2\\t%$\\t%1, %0, %0 + %|%.\\tmvklh\\t%$\\t%1, %0" + [(set_attr "units" "ls") + (set_attr "cross" "n,n,y,y,n")]) + +(define_insn "movstricthi" + [(set (strict_low_part (match_operand:HI 0 "register_operand" "+a,b,a,b")) + (match_operand:HI 1 "register_operand" "a,b,?b,?a"))] + "TARGET_INSNS_64" + "%|%.\\tpackhl2\\t%$\\t%0, %1, %0" + [(set_attr "units" "ls") + (set_attr "cross" "n,n,y,y")]) + +(include "c6x-mult.md") +(include "sync.md") diff --git a/gcc/config/c6x/c6x.opt b/gcc/config/c6x/c6x.opt new file mode 100644 index 0000000..14ad81b --- /dev/null +++ b/gcc/config/c6x/c6x.opt @@ -0,0 +1,67 @@ +; Option definitions for TI C6X. +; Copyright (C) 2010, 2011 Free Software Foundation, Inc. +; Contributed by Bernd Schmidt <bernds@codesourcery.com> +; Contributed by CodeSourcery. +; +; This file is part of GCC. +; +; GCC is free software; you can redistribute it and/or modify +; it under the terms of the GNU General Public License as published by +; the Free Software Foundation; either version 3, or (at your option) +; any later version. +; +; GCC is distributed in the hope that it will be useful, +; but WITHOUT ANY WARRANTY; without even the implied warranty of +; MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the +; GNU General Public License for more details. +; +; You should have received a copy of the GNU General Public License +; along with GCC; see the file COPYING3. If not see +; <http://www.gnu.org/licenses/>. + +HeaderInclude +config/c6x/c6x-opts.h + +SourceInclude +config/c6x/c6x-opts.h + +mbig-endian +Target Report RejectNegative Mask(BIG_ENDIAN) +Use big-endian byte order + +mlittle-endian +Target Report RejectNegative InverseMask(BIG_ENDIAN, LITTLE_ENDIAN) +Use little-endian byte order + +msim +Target RejectNegative +Use simulator runtime + +msdata= +Target RejectNegative Enum(c6x_sdata) Joined Var(c6x_sdata_mode) Init(C6X_SDATA_DEFAULT) +Select method for sdata handling + +Enum +Name(c6x_sdata) Type(enum c6x_sdata) +Valid arguments for the -msdata= option + +EnumValue +Enum(c6x_sdata) String(none) Value(C6X_SDATA_NONE) + +EnumValue +Enum(c6x_sdata) String(default) Value(C6X_SDATA_DEFAULT) + +EnumValue +Enum(c6x_sdata) String(all) Value(C6X_SDATA_ALL) + +mdsbt +Target Mask(DSBT) +Compile for the DSBT shared library ABI + +mlong-calls +Target Report Mask(LONG_CALLS) +Avoid generating pc-relative calls; use indirection + +march= +Target RejectNegative Joined Enum(c6x_isa) Var(c6x_arch_option) +Specify the name of the target architecture diff --git a/gcc/config/c6x/c6x_intrinsics.h b/gcc/config/c6x/c6x_intrinsics.h new file mode 100644 index 0000000..2ae7dd5 --- /dev/null +++ b/gcc/config/c6x/c6x_intrinsics.h @@ -0,0 +1,194 @@ +/* Intrinsics for TI C6X. + + Copyright (C) 2011 Free Software Foundation, Inc. + Contributed by CodeSourcery. + + This file is part of GCC. + + GCC is free software; you can redistribute it and/or modify it + under the terms of the GNU General Public License as published + by the Free Software Foundation; either version 3, or (at your + option) any later version. + + GCC is distributed in the hope that it will be useful, but WITHOUT + ANY WARRANTY; without even the implied warranty of MERCHANTABILITY + or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public + License for more details. + + Under Section 7 of GPL version 3, you are granted additional + permissions described in the GCC Runtime Library Exception, version + 3.1, as published by the Free Software Foundation. + + You should have received a copy of the GNU General Public License and + a copy of the GCC Runtime Library Exception along with this program; + see the files COPYING3 and COPYING.RUNTIME respectively. If not, see + <http://www.gnu.org/licenses/>. */ + +#ifndef _GCC_C6X_INTRINSICS_H +#define _GCC_C6X_INTRINSICS_H + +#if !defined(__TMS320C6X__) +# error "c6x_intrinsics.h is only supported for C6X targets" +#endif + +#ifdef __cplusplus +extern "C" { +#endif + +#include <stdint.h> + +/* Define vector types. */ +typedef uint8_t __uv4qi __attribute__((vector_size (4))); +typedef int16_t __v2hi __attribute__((vector_size (4))); +typedef int32_t __v2si __attribute__((vector_size (8))); + +__extension__ static __inline int __attribute__ ((__always_inline__)) +_abs (int src) +{ + return __builtin_c6x_abs (src); +} + +__extension__ static __inline int __attribute__ ((__always_inline__)) +_abs2 (int src) +{ + return (int)__builtin_c6x_abs2 ((__v2hi)src); +} + +__extension__ static __inline int __attribute__ ((__always_inline__)) +_sadd (int src1, int src2) +{ + return __builtin_c6x_sadd (src1, src2); +} + +__extension__ static __inline int __attribute__ ((__always_inline__)) +_ssub (int src1, int src2) +{ + return __builtin_c6x_ssub (src1, src2); +} + +__extension__ static __inline int __attribute__ ((__always_inline__)) +_add2 (int src1, int src2) +{ + return (int)__builtin_c6x_add2 ((__v2hi)src1, (__v2hi)src2); +} + +__extension__ static __inline int __attribute__ ((__always_inline__)) +_sub2 (int src1, int src2) +{ + return (int)__builtin_c6x_sub2 ((__v2hi)src1, (__v2hi)src2); +} + +__extension__ static __inline int __attribute__ ((__always_inline__)) +_add4 (int src1, int src2) +{ + return (int)__builtin_c6x_add4 ((__uv4qi)src1, (__uv4qi)src2); +} + +__extension__ static __inline int __attribute__ ((__always_inline__)) +_sub4 (int src1, int src2) +{ + return (int)__builtin_c6x_sub4 ((__uv4qi)src1, (__uv4qi)src2); +} + +__extension__ static __inline int __attribute__ ((__always_inline__)) +_sadd2 (int src1, int src2) +{ + return (int)__builtin_c6x_sadd2 ((__v2hi)src1, (__v2hi)src2); +} + +__extension__ static __inline int __attribute__ ((__always_inline__)) +_ssub2 (int src1, int src2) +{ + return (int)__builtin_c6x_ssub2 ((__v2hi)src1, (__v2hi)src2); +} + +__extension__ static __inline int __attribute__ ((__always_inline__)) +_saddu4 (int src1, int src2) +{ + return (int)__builtin_c6x_saddu4 ((__uv4qi)src1, (__uv4qi)src2); +} + +__extension__ static __inline int __attribute__ ((__always_inline__)) +_smpy (int src1, int src2) +{ + return __builtin_c6x_smpy (src1, src2); +} + +__extension__ static __inline int __attribute__ ((__always_inline__)) +_smpylh (int src1, int src2) +{ + return __builtin_c6x_smpylh (src1, src2); +} + +__extension__ static __inline int __attribute__ ((__always_inline__)) +_smpyhl (int src1, int src2) +{ + return __builtin_c6x_smpyhl (src1, src2); +} + +__extension__ static __inline int __attribute__ ((__always_inline__)) +_smpyh (int src1, int src2) +{ + return __builtin_c6x_smpyh (src1, src2); +} + +__extension__ static __inline long long __attribute__ ((__always_inline__)) +_smpy2ll (int src1, int src2) +{ + return (long long)__builtin_c6x_smpy2 ((__v2hi)src1, (__v2hi)src2); +} + +__extension__ static __inline long long __attribute__ ((__always_inline__)) +_mpy2ll (int src1, int src2) +{ + return (long long)__builtin_c6x_mpy2 ((__v2hi)src1, (__v2hi)src2); +} + +__extension__ static __inline int __attribute__ ((__always_inline__)) +_extr (int src1, int src2) +{ + return __builtin_c6x_extr (src1, src2); +} + +__extension__ static __inline int __attribute__ ((__always_inline__)) +_extru (int src1, int src2) +{ + return __builtin_c6x_extru (src1, src2); +} + +__extension__ static __inline int __attribute__ ((__always_inline__)) +_clrr (int src1, int src2) +{ + return __builtin_c6x_clrr (src1, src2); +} + +__extension__ static __inline int __attribute__ ((__always_inline__)) +_avg2 (int src1, int src2) +{ + return (int)__builtin_c6x_avg2 ((__v2hi)src1, (__v2hi)src2); +} + +__extension__ static __inline int __attribute__ ((__always_inline__)) +_avgu4 (int src1, int src2) +{ + return (int)__builtin_c6x_avgu4 ((__uv4qi)src1, (__uv4qi)src2); +} + +__extension__ static __inline int __attribute__ ((__always_inline__)) +_sshl (int src1, int src2) +{ + return __builtin_c6x_sshl (src1, src2); +} + +__extension__ static __inline int __attribute__ ((__always_inline__)) +_subc (int src1, int src2) +{ + return __builtin_c6x_subc (src1, src2); +} + + +#ifdef __cplusplus +} +#endif + +#endif diff --git a/gcc/config/c6x/constraints.md b/gcc/config/c6x/constraints.md new file mode 100644 index 0000000..1a6710b --- /dev/null +++ b/gcc/config/c6x/constraints.md @@ -0,0 +1,174 @@ +;; Constraint definitions for TI C6X. +;; Copyright (C) 2010, 2011 Free Software Foundation, Inc. +;; Contributed by Andrew Jenner <andrew@codesourcery.com> +;; Contributed by Bernd Schmidt <bernds@codesourcery.com> +;; Contributed by CodeSourcery. +;; +;; This file is part of GCC. +;; +;; GCC is free software; you can redistribute it and/or modify +;; it under the terms of the GNU General Public License as published by +;; the Free Software Foundation; either version 3, or (at your option) +;; any later version. +;; +;; GCC is distributed in the hope that it will be useful, +;; but WITHOUT ANY WARRANTY; without even the implied warranty of +;; MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the +;; GNU General Public License for more details. +;; +;; You should have received a copy of the GNU General Public License +;; along with GCC; see the file COPYING3. If not see +;; <http://www.gnu.org/licenses/>. + +(define_register_constraint "a" "A_REGS" + "Register file A (A0--A31).") + +(define_register_constraint "b" "B_REGS" + "Register file B (B0--B31).") + +(define_register_constraint "A" "PREDICATE_A_REGS" + "Predicate registers in register file A (A0--A2 on C64X and higher, + A1 and A2 otherwise).") + +(define_register_constraint "B" "PREDICATE_B_REGS" + "Predicate registers in register file B (B0--B2).") + +(define_register_constraint "C" "CALL_USED_B_REGS" + "A call-used register in register file B (B0--B9, B16--B31).") + +(define_register_constraint "Da" "NONPREDICATE_A_REGS" + "Register file A, excluding predicate registers (A3--A31, plus A0 if +not C64X or higher).") + +(define_register_constraint "Db" "NONPREDICATE_B_REGS" + "Register file B, excluding predicate registers (B3--B31).") + +(define_register_constraint "Z" "PICREG" + "Register B14 (aka DP).") + +(define_register_constraint "z" "SPREG" + "Register B15 (aka SP).") + +(define_constraint "Iu4" + "Integer constant in the range 0 @dots{} 15, aka ucst4." + (and (match_code "const_int") + (match_test "ival >= 0 && ival <= 15"))) + +(define_constraint "Iu5" + "Integer constant in the range 0 @dots{} 31, aka ucst5." + (and (match_code "const_int") + (match_test "ival >= 0 && ival <= 31"))) + +(define_constraint "In5" + "Integer constant in the range @minus{}31 @dots{} 0, negation of ucst5." + (and (match_code "const_int") + (match_test "ival >= -31 && ival <= 0"))) + +(define_constraint "Is5" + "Integer constant in the range @minus{}16 @dots{} 15, aka scst5." + (and (match_code "const_int") + (match_test "ival >= -16 && ival <= 15"))) + +(define_constraint "I5x" + "Integer constant that can be the operand of an ADDA or a SUBA insn." + (and (match_code "const_int") + (match_test "(ival >= -31 && ival <= 31) + || ((ival & 1) == 0 && ival >= -62 && ival <= 62) + || ((ival & 3) == 0 && ival >= -124 && ival <= 124) + || ((TARGET_INSNS_64 || TARGET_INSNS_67) + && (ival & 7) == 0 && ival > 0 && ival <= 248)"))) + +(define_constraint "Iux" + "Integer constant that can be the operand of a long ADDA or a SUBA insn, + i.e. one involving B14 or B15 as source operand." + (and (match_code "const_int") + (and (match_test "TARGET_INSNS_64PLUS") + (match_test "ival >= 0 + && (ival < 32768 + || ((ival & 1) == 0 && ival < 65536) + || ((ival & 3) == 0 && ival < 131072))")))) + +(define_constraint "IuB" + "Integer constant in the range 0 @dots{} 65535, aka ucst16." + (and (match_code "const_int") + (match_test "ival >= 0 && ival <= 65535"))) + +(define_constraint "IsB" + "Integer constant in the range @minus{}32768 @dots{} 32767." + (and (match_code "const_int") + (match_test "ival >= -32768 && ival <= 32767"))) + +(define_constraint "IsC" + "Integer constant in the range @math{-2^{20}} @dots{} @math{2^{20} - 1}." + (and (match_code "const_int") + (match_test "ival >= -0x100000 && ival <= 0xfffff"))) + +(define_constraint "JA" + "@internal + Integer constant in the range 0 @dots{} 31, corresponding to an A register + number." + (and (match_code "const_int") + (match_test "ival >= 0 && ival < 32"))) + +(define_constraint "JB" + "@internal + Integer constant in the range 32 @dots{} 63, corresponding to a B register + number." + (and (match_code "const_int") + (match_test "ival >= 32 && ival < 64"))) + +(define_constraint "Jc" + "Integer constant that is a valid mask for the clr instruction" + (and (match_code "const_int") + (match_test "c6x_valid_mask_p (ival)"))) + +(define_constraint "Js" + "Integer constant that is a valid mask for the set instruction" + (and (match_code "const_int") + (match_test "c6x_valid_mask_p (~ival)"))) + +(define_memory_constraint "Q" + "Memory location with A base register." + (and (match_code "mem") + (match_test "c6x_mem_operand (op, A_REGS, false)"))) + +(define_memory_constraint "R" + "Memory location with B base register." + (and (match_code "mem") + (match_test "c6x_mem_operand (op, B_REGS, false)"))) + +(define_memory_constraint "T" + "@internal + Memory location with B base register, but not using a long offset." + (and (match_code "mem") + (match_test "c6x_mem_operand (op, B_REGS, true)"))) + +(define_constraint "S0" + "@internal + On C64x+ targets, a GP-relative small data reference" + (and (match_test "TARGET_INSNS_64PLUS") + (match_operand 0 "sdata_symbolic_operand"))) + +(define_constraint "S1" + "@internal + Any kind of @code{SYMBOL_REF}, for use in a call address." + (and (match_code "symbol_ref") + (match_operand 0 "c6x_call_operand"))) + +(define_constraint "S2" + "@internal + Any SYMBOL_REF or LABEL_REF." + (ior (match_code "symbol_ref") (match_code "label_ref"))) + +(define_constraint "Si" + "@internal + Any immediate value, unless it matches the S0 constraint." + (and (match_operand 0 "immediate_operand") + (match_test "!satisfies_constraint_S0 (op)"))) + +(define_memory_constraint "W" + "@internal + A memory operand with an address that can't be used in an unaligned access." + (and (match_code "mem") + (match_test "!c6x_legitimate_address_p_1 (GET_MODE (op), XEXP (op, 0), + reload_completed, true)"))) diff --git a/gcc/config/c6x/crti.s b/gcc/config/c6x/crti.s new file mode 100644 index 0000000..8fe35c1 --- /dev/null +++ b/gcc/config/c6x/crti.s @@ -0,0 +1,39 @@ +/* Copyright 2010, 2011 Free Software Foundation, Inc. + Contributed by Bernd Schmidt <bernds@codesourcery.com>. + +This file is free software; you can redistribute it and/or modify it +under the terms of the GNU General Public License as published by the +Free Software Foundation; either version 3, or (at your option) any +later version. + +This file is distributed in the hope that it will be useful, but +WITHOUT ANY WARRANTY; without even the implied warranty of +MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU +General Public License for more details. + +Under Section 7 of GPL version 3, you are granted additional +permissions described in the GCC Runtime Library Exception, version +3.1, as published by the Free Software Foundation. + +You should have received a copy of the GNU General Public License and +a copy of the GCC Runtime Library Exception along with this program; +see the files COPYING3 and COPYING.RUNTIME respectively. If not, see +<http://www.gnu.org/licenses/>. */ + +/* + * This file just supplies function prologues for the .init and .fini + * sections. It is linked in before crtbegin.o. + */ + + .section .init + .globl _init + .type _init,@function +_init: + add .l2 -8, B15, B15 + stw .d2t2 B3,*+B15(4) + .section .fini + .globl _fini + .type _fini,@function +_fini: + add .l2 -8, B15, B15 + stw .d2t2 B3,*+B15(4) diff --git a/gcc/config/c6x/crtn.s b/gcc/config/c6x/crtn.s new file mode 100644 index 0000000..5900a4b --- /dev/null +++ b/gcc/config/c6x/crtn.s @@ -0,0 +1,41 @@ +/* Copyright 2010, 2011 Free Software Foundation, Inc. + Contributed by Bernd Schmidt <bernds@codesourcery.com>. + +This file is free software; you can redistribute it and/or modify it +under the terms of the GNU General Public License as published by the +Free Software Foundation; either version 3, or (at your option) any +later version. + +This file is distributed in the hope that it will be useful, but +WITHOUT ANY WARRANTY; without even the implied warranty of +MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU +General Public License for more details. + +Under Section 7 of GPL version 3, you are granted additional +permissions described in the GCC Runtime Library Exception, version +3.1, as published by the Free Software Foundation. + +You should have received a copy of the GNU General Public License and +a copy of the GCC Runtime Library Exception along with this program; +see the files COPYING3 and COPYING.RUNTIME respectively. If not, see +<http://www.gnu.org/licenses/>. */ + +/* + * This file supplies function epilogues for the .init and .fini sections. + * It is linked in after all other files. + */ + + .section .init + ldw .d2t2 *+B15(4), B3 + add .d2 B15, 8, B15 + nop 3 + ret .s2 B3 + nop 5 + + .section .fini + ldw .d2t2 *+B15(4), B3 + add .d2 B15, 8, B15 + nop 3 + ret .s2 B3 + nop 5 + diff --git a/gcc/config/c6x/elf-common.h b/gcc/config/c6x/elf-common.h new file mode 100644 index 0000000..50a3c3a --- /dev/null +++ b/gcc/config/c6x/elf-common.h @@ -0,0 +1,37 @@ +/* ELF definitions for TI C6X + Copyright (C) 2010 Free Software Foundation, Inc. + Contributed by Andrew Jenner <andrew@codesourcery.com> + Contributed by Bernd Schmidt <bernds@codesourcery.com> + + This file is part of GCC. + + GCC is free software; you can redistribute it and/or modify it + under the terms of the GNU General Public License as published + by the Free Software Foundation; either version 3, or (at your + option) any later version. + + GCC is distributed in the hope that it will be useful, but WITHOUT + ANY WARRANTY; without even the implied warranty of MERCHANTABILITY + or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public + License for more details. + + You should have received a copy of the GNU General Public License + along with GCC; see the file COPYING3. If not see + <http://www.gnu.org/licenses/>. */ + +/* Controlling the Compilation Driver. */ +#define ENDIAN_LINK_SPEC "%{mbig-endian:-EB} %{mlittle-endian:-EL} " + +#undef ASM_SPEC +#define ASM_SPEC "%{march=*:-march=%*} %{mbig-endian:-mbig-endian} \ + %{mdsbt:-mdsbt %{!fPIC:-mpid=near} %{fPIC:-mpid=far -mpic} %{fpic:-mpic}} \ + %{!mdsbt:%{fpic:-mpic -mpid=near} %{fPIC:-mpic -mpid=far}}" + +#undef DATA_SECTION_ASM_OP +#define DATA_SECTION_ASM_OP "\t.section\t\".fardata\",\"aw\"" +#undef READONLY_DATA_SECTION_ASM_OP +#define READONLY_DATA_SECTION_ASM_OP "\t.section\t\".const\",\"a\",@progbits" +#define BSS_SECTION_ASM_OP "\t.section\t\".far\",\"aw\",@nobits" +#define SDATA_SECTION_ASM_OP "\t.section\t\".neardata\",\"aw\"" +#define SBSS_SECTION_ASM_OP "\t.section\t\".bss\",\"aw\",@nobits" +#define TARGET_LIBGCC_SDATA_SECTION ".neardata" diff --git a/gcc/config/c6x/elf.h b/gcc/config/c6x/elf.h new file mode 100644 index 0000000..d89ed43 --- /dev/null +++ b/gcc/config/c6x/elf.h @@ -0,0 +1,35 @@ +/* ELF definitions for TI C6X + Copyright (C) 2010 Free Software Foundation, Inc. + Contributed by Andrew Jenner <andrew@codesourcery.com> + Contributed by Bernd Schmidt <bernds@codesourcery.com> + + This file is part of GCC. + + GCC is free software; you can redistribute it and/or modify it + under the terms of the GNU General Public License as published + by the Free Software Foundation; either version 3, or (at your + option) any later version. + + GCC is distributed in the hope that it will be useful, but WITHOUT + ANY WARRANTY; without even the implied warranty of MERCHANTABILITY + or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public + License for more details. + + You should have received a copy of the GNU General Public License + along with GCC; see the file COPYING3. If not see + <http://www.gnu.org/licenses/>. */ + +/* crt0.o should come from the linker script, but for compatibility, + we mention it here for -msim. */ +#undef STARTFILE_SPEC +#define STARTFILE_SPEC "%{msim:crt0%O%s} crti%O%s crtbegin%O%s" + +#undef ENDFILE_SPEC +#define ENDFILE_SPEC "crtend%O%s crtn%O%s" + +#undef LIB_SPEC +#define LIB_SPEC "%{msim:--start-group -lc -lsim --end-group;" \ + ":-lc}" + +#undef LINK_SPEC +#define LINK_SPEC ENDIAN_LINK_SPEC diff --git a/gcc/config/c6x/eqd.c b/gcc/config/c6x/eqd.c new file mode 100644 index 0000000..c19f886 --- /dev/null +++ b/gcc/config/c6x/eqd.c @@ -0,0 +1,47 @@ +/* Software floating-point emulation. + Return 1 iff a == b, 0 otherwise. + Copyright (C) 1997,1999,2006,2007,2011 Free Software Foundation, Inc. + Contributed by Richard Henderson (rth@cygnus.com) and + Jakub Jelinek (jj@ultra.linux.cz). + + This file is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + In addition to the permissions in the GNU Lesser General Public + License, the Free Software Foundation gives you unlimited + permission to link the compiled version of this file into + combinations with other programs, and to distribute those + combinations without any restriction coming from the use of this + file. (The Lesser General Public License restrictions do apply in + other respects; for example, they cover modification of the file, + and distribution when not linked into a combine executable.) + + This file is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with GCC; see the file COPYING.LIB. If not see + <http://www.gnu.org/licenses/>. */ + +#include "config/soft-fp/soft-fp.h" +#include "config/soft-fp/double.h" + +CMPtype __c6xabi_eqd(DFtype a, DFtype b) +{ + FP_DECL_EX; + FP_DECL_D(A); FP_DECL_D(B); + CMPtype r; + + FP_UNPACK_RAW_D(A, a); + FP_UNPACK_RAW_D(B, b); + FP_CMP_EQ_D(r, A, B); + if (r && (FP_ISSIGNAN_D(A) || FP_ISSIGNAN_D(B))) + FP_SET_EXCEPTION(FP_EX_INVALID); + FP_HANDLE_EXCEPTIONS; + + return !r; +} diff --git a/gcc/config/c6x/eqf.c b/gcc/config/c6x/eqf.c new file mode 100644 index 0000000..c87c92b --- /dev/null +++ b/gcc/config/c6x/eqf.c @@ -0,0 +1,47 @@ +/* Software floating-point emulation. + Return 1 iff a == b, 0 otherwise. + Copyright (C) 1997,1999,2006,2007,2011 Free Software Foundation, Inc. + Contributed by Richard Henderson (rth@cygnus.com) and + Jakub Jelinek (jj@ultra.linux.cz). + + This file is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + In addition to the permissions in the GNU Lesser General Public + License, the Free Software Foundation gives you unlimited + permission to link the compiled version of this file into + combinations with other programs, and to distribute those + combinations without any restriction coming from the use of this + file. (The Lesser General Public License restrictions do apply in + other respects; for example, they cover modification of the file, + and distribution when not linked into a combine executable.) + + This file is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with GCC; see the file COPYING.LIB. If not see + <http://www.gnu.org/licenses/>. */ + +#include "config/soft-fp/soft-fp.h" +#include "config/soft-fp/single.h" + +CMPtype __c6xabi_eqf(SFtype a, SFtype b) +{ + FP_DECL_EX; + FP_DECL_S(A); FP_DECL_S(B); + CMPtype r; + + FP_UNPACK_RAW_S(A, a); + FP_UNPACK_RAW_S(B, b); + FP_CMP_EQ_S(r, A, B); + if (r && (FP_ISSIGNAN_S(A) || FP_ISSIGNAN_S(B))) + FP_SET_EXCEPTION(FP_EX_INVALID); + FP_HANDLE_EXCEPTIONS; + + return !r; +} diff --git a/gcc/config/c6x/ged.c b/gcc/config/c6x/ged.c new file mode 100644 index 0000000..30590b2 --- /dev/null +++ b/gcc/config/c6x/ged.c @@ -0,0 +1,47 @@ +/* Software floating-point emulation. + Return 1 iff a >= b, 0 otherwise. + Copyright (C) 1997,1999,2006,2007,2011 Free Software Foundation, Inc. + Contributed by Richard Henderson (rth@cygnus.com) and + Jakub Jelinek (jj@ultra.linux.cz). + + This file is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + In addition to the permissions in the GNU Lesser General Public + License, the Free Software Foundation gives you unlimited + permission to link the compiled version of this file into + combinations with other programs, and to distribute those + combinations without any restriction coming from the use of this + file. (The Lesser General Public License restrictions do apply in + other respects; for example, they cover modification of the file, + and distribution when not linked into a combine executable.) + + This file is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with GCC; see the file COPYING.LIB. If not see + <http://www.gnu.org/licenses/>. */ + +#include "config/soft-fp/soft-fp.h" +#include "config/soft-fp/double.h" + +CMPtype __c6xabi_ged(DFtype a, DFtype b) +{ + FP_DECL_EX; + FP_DECL_D(A); FP_DECL_D(B); + CMPtype r; + + FP_UNPACK_RAW_D(A, a); + FP_UNPACK_RAW_D(B, b); + FP_CMP_D(r, A, B, -2); + if (r == -2 && (FP_ISSIGNAN_D(A) || FP_ISSIGNAN_D(B))) + FP_SET_EXCEPTION(FP_EX_INVALID); + FP_HANDLE_EXCEPTIONS; + + return r >= 0; +} diff --git a/gcc/config/c6x/gef.c b/gcc/config/c6x/gef.c new file mode 100644 index 0000000..0a22b13 --- /dev/null +++ b/gcc/config/c6x/gef.c @@ -0,0 +1,47 @@ +/* Software floating-point emulation. + Return 1 iff a >= b, 0 otherwise. + Copyright (C) 1997,1999,2006,2007 Free Software Foundation, Inc. + Contributed by Richard Henderson (rth@cygnus.com) and + Jakub Jelinek (jj@ultra.linux.cz). + + This file is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + In addition to the permissions in the GNU Lesser General Public + License, the Free Software Foundation gives you unlimited + permission to link the compiled version of this file into + combinations with other programs, and to distribute those + combinations without any restriction coming from the use of this + file. (The Lesser General Public License restrictions do apply in + other respects; for example, they cover modification of the file, + and distribution when not linked into a combine executable.) + + This file is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with GCC; see the file COPYING.LIB. If not see + <http://www.gnu.org/licenses/>. */ + +#include "config/soft-fp/soft-fp.h" +#include "config/soft-fp/single.h" + +CMPtype __c6xabi_gef(SFtype a, SFtype b) +{ + FP_DECL_EX; + FP_DECL_S(A); FP_DECL_S(B); + CMPtype r; + + FP_UNPACK_RAW_S(A, a); + FP_UNPACK_RAW_S(B, b); + FP_CMP_S(r, A, B, -2); + if (r == -2 && (FP_ISSIGNAN_S(A) || FP_ISSIGNAN_S(B))) + FP_SET_EXCEPTION(FP_EX_INVALID); + FP_HANDLE_EXCEPTIONS; + + return r >= 0; +} diff --git a/gcc/config/c6x/genmult.sh b/gcc/config/c6x/genmult.sh new file mode 100644 index 0000000..993c826 --- /dev/null +++ b/gcc/config/c6x/genmult.sh @@ -0,0 +1,33 @@ +#! /bin/sh +# Generate c6x-mult.md from c6x-mult.md.in +# The input file is passed as an argument. + +# Copyright 2011 Free Software Foundation, Inc. + +#This file is part of GCC. + +#GCC is free software; you can redistribute it and/or modify +#it under the terms of the GNU General Public License as published by +#the Free Software Foundation; either version 3, or (at your option) +#any later version. + +#GCC is distributed in the hope that it will be useful, +#but WITHOUT ANY WARRANTY; without even the implied warranty of +#MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the +#GNU General Public License for more details. + +#You should have received a copy of the GNU General Public License +#along with GCC; see the file COPYING3. If not see +#<http://www.gnu.org/licenses/>. + +echo ";; -*- buffer-read-only: t -*-" +echo ";; Generated automatically from c6x-mult.md.in by genmult.sh" + +sed -e "s,_VARIANT_,,g" -e "s,_SET_,set,g" -e "s,_.BRK_,,g" \ + -e "s,_A_,a,g" -e "s,_B_,b,g" -e "s,_DESTOPERAND_,register_operand,g" \ + -e "s,_MOD._,,g" -e "s,:_M,:,g" < $1 + +sed -e "s,_VARIANT_,_real,g" -e "s,_SET_,unspec,g" -e "s,_OBRK_,[,g" \ + -e "s,_CBRK_,] UNSPEC_REAL_MULT,g" -e "s,_A_,JA,g" -e "s,_B_,JB,g" \ + -e "s,_DESTOPERAND_,const_int_operand,g" -e "s,_MODk_,k,g" \ + -e "s,_MODK_,K,g" -e 's,:_MV..[IQ],:SI,g' -e "s,:_MSQ,:SI,g" < $1 diff --git a/gcc/config/c6x/genopt.sh b/gcc/config/c6x/genopt.sh new file mode 100644 index 0000000..5f323c9 --- /dev/null +++ b/gcc/config/c6x/genopt.sh @@ -0,0 +1,59 @@ +#!/bin/sh +# Generate c6x-tables.opt from the lists in *.def. +# Copyright (C) 2011 Free Software Foundation, Inc. +# +# This file is part of GCC. +# +# GCC is free software; you can redistribute it and/or modify +# it under the terms of the GNU General Public License as published by +# the Free Software Foundation; either version 3, or (at your option) +# any later version. +# +# GCC is distributed in the hope that it will be useful, +# but WITHOUT ANY WARRANTY; without even the implied warranty of +# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the +# GNU General Public License for more details. +# +# You should have received a copy of the GNU General Public License +# along with GCC; see the file COPYING3. If not see +# <http://www.gnu.org/licenses/>. + +cat <<EOF +; -*- buffer-read-only: t -*- +; Generated automatically by genopt.sh from c6x-isas.def. +; +; Copyright (C) 2011 Free Software Foundation, Inc. +; +; This file is part of GCC. +; +; GCC is free software; you can redistribute it and/or modify it under +; the terms of the GNU General Public License as published by the Free +; Software Foundation; either version 3, or (at your option) any later +; version. +; +; GCC is distributed in the hope that it will be useful, but WITHOUT ANY +; WARRANTY; without even the implied warranty of MERCHANTABILITY or +; FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License +; for more details. +; +; You should have received a copy of the GNU General Public License +; along with GCC; see the file COPYING3. If not see +; <http://www.gnu.org/licenses/>. + +Enum +Name(c6x_isa) Type(int) +Known C6X ISAs (for use with the -march= option): + +EOF + +awk -F'[(, ]+' 'BEGIN { + value = 0 +} +/^C6X_ISA/ { + name = $2 + gsub("\"", "", name) + print "EnumValue" + print "Enum(c6x_isa) String(" name ") Value(" value ")" + print "" + value++ +}' $1/c6x-isas.def diff --git a/gcc/config/c6x/gensched.sh b/gcc/config/c6x/gensched.sh new file mode 100644 index 0000000..61ea70d --- /dev/null +++ b/gcc/config/c6x/gensched.sh @@ -0,0 +1,44 @@ +#! /bin/sh +# Generate c6x-sched.md from c6x-sched.md.in +# The input file is passed as an argument. + +# Copyright 2010, 2011 Free Software Foundation, Inc. + +#This file is part of GCC. + +#GCC is free software; you can redistribute it and/or modify +#it under the terms of the GNU General Public License as published by +#the Free Software Foundation; either version 3, or (at your option) +#any later version. + +#GCC is distributed in the hope that it will be useful, +#but WITHOUT ANY WARRANTY; without even the implied warranty of +#MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the +#GNU General Public License for more details. + +#You should have received a copy of the GNU General Public License +#along with GCC; see the file COPYING3. If not see +#<http://www.gnu.org/licenses/>. + +echo ";; -*- buffer-read-only: t -*-" +echo ";; Generated automatically from c6x-sched.md.in by gensched.sh" + +for cross in n y; do + for side in 1 2; do + tside=$side + case $side in + 1) rf="a"; otherside=2 ;; + 2) rf="b"; otherside=1 ;; + esac + case $cross in + y) cunit="+x$side"; tside=$otherside;; + n) cunit="";; + esac + echo + echo ";; Definitions for side $side, cross $cross" + echo + sed -e "s,_CROSS_,$cross,g" -e "s,_CUNIT_,$cunit,g" \ + -e "s,_N_,$side,g" -e "s,_RF_,$rf,g" -e "s,_NX_,$tside,g" \ + < $1 + done +done diff --git a/gcc/config/c6x/gtd.c b/gcc/config/c6x/gtd.c new file mode 100644 index 0000000..042192b --- /dev/null +++ b/gcc/config/c6x/gtd.c @@ -0,0 +1,47 @@ +/* Software floating-point emulation. + Return 1 iff a > b, 0 otherwise. + Copyright (C) 1997,1999,2006,2007,2011 Free Software Foundation, Inc. + Contributed by Richard Henderson (rth@cygnus.com) and + Jakub Jelinek (jj@ultra.linux.cz). + + This file is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + In addition to the permissions in the GNU Lesser General Public + License, the Free Software Foundation gives you unlimited + permission to link the compiled version of this file into + combinations with other programs, and to distribute those + combinations without any restriction coming from the use of this + file. (The Lesser General Public License restrictions do apply in + other respects; for example, they cover modification of the file, + and distribution when not linked into a combine executable.) + + This file is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with GCC; see the file COPYING.LIB. If not see + <http://www.gnu.org/licenses/>. */ + +#include "config/soft-fp/soft-fp.h" +#include "config/soft-fp/double.h" + +CMPtype __c6xabi_gtd(DFtype a, DFtype b) +{ + FP_DECL_EX; + FP_DECL_D(A); FP_DECL_D(B); + CMPtype r; + + FP_UNPACK_RAW_D(A, a); + FP_UNPACK_RAW_D(B, b); + FP_CMP_D(r, A, B, -2); + if (r == -2 && (FP_ISSIGNAN_D(A) || FP_ISSIGNAN_D(B))) + FP_SET_EXCEPTION(FP_EX_INVALID); + FP_HANDLE_EXCEPTIONS; + + return r > 0; +} diff --git a/gcc/config/c6x/gtf.c b/gcc/config/c6x/gtf.c new file mode 100644 index 0000000..0d0e10a --- /dev/null +++ b/gcc/config/c6x/gtf.c @@ -0,0 +1,47 @@ +/* Software floating-point emulation. + Return 1 iff a > b, 0 otherwise. + Copyright (C) 1997,1999,2006,2007,2011 Free Software Foundation, Inc. + Contributed by Richard Henderson (rth@cygnus.com) and + Jakub Jelinek (jj@ultra.linux.cz). + + This file is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + In addition to the permissions in the GNU Lesser General Public + License, the Free Software Foundation gives you unlimited + permission to link the compiled version of this file into + combinations with other programs, and to distribute those + combinations without any restriction coming from the use of this + file. (The Lesser General Public License restrictions do apply in + other respects; for example, they cover modification of the file, + and distribution when not linked into a combine executable.) + + This file is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with GCC; see the file COPYING.LIB. If not see + <http://www.gnu.org/licenses/>. */ + +#include "config/soft-fp/soft-fp.h" +#include "config/soft-fp/single.h" + +CMPtype __c6xabi_gtf(SFtype a, SFtype b) +{ + FP_DECL_EX; + FP_DECL_S(A); FP_DECL_S(B); + CMPtype r; + + FP_UNPACK_RAW_S(A, a); + FP_UNPACK_RAW_S(B, b); + FP_CMP_S(r, A, B, -2); + if (r == -2 && (FP_ISSIGNAN_S(A) || FP_ISSIGNAN_S(B))) + FP_SET_EXCEPTION(FP_EX_INVALID); + FP_HANDLE_EXCEPTIONS; + + return r > 0; +} diff --git a/gcc/config/c6x/led.c b/gcc/config/c6x/led.c new file mode 100644 index 0000000..b97f3d1 --- /dev/null +++ b/gcc/config/c6x/led.c @@ -0,0 +1,47 @@ +/* Software floating-point emulation. + Return 1 iff a <= b, 0 otherwise. + Copyright (C) 1997,1999,2006,2007,2011 Free Software Foundation, Inc. + Contributed by Richard Henderson (rth@cygnus.com) and + Jakub Jelinek (jj@ultra.linux.cz). + + This file is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + In addition to the permissions in the GNU Lesser General Public + License, the Free Software Foundation gives you unlimited + permission to link the compiled version of this file into + combinations with other programs, and to distribute those + combinations without any restriction coming from the use of this + file. (The Lesser General Public License restrictions do apply in + other respects; for example, they cover modification of the file, + and distribution when not linked into a combine executable.) + + This file is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with GCC; see the file COPYING.LIB. If not see + <http://www.gnu.org/licenses/>. */ + +#include "config/soft-fp/soft-fp.h" +#include "config/soft-fp/double.h" + +CMPtype __c6xabi_led(DFtype a, DFtype b) +{ + FP_DECL_EX; + FP_DECL_D(A); FP_DECL_D(B); + CMPtype r; + + FP_UNPACK_RAW_D(A, a); + FP_UNPACK_RAW_D(B, b); + FP_CMP_D(r, A, B, 2); + if (r == 2 && (FP_ISSIGNAN_D(A) || FP_ISSIGNAN_D(B))) + FP_SET_EXCEPTION(FP_EX_INVALID); + FP_HANDLE_EXCEPTIONS; + + return r <= 0; +} diff --git a/gcc/config/c6x/lef.c b/gcc/config/c6x/lef.c new file mode 100644 index 0000000..d234a18 --- /dev/null +++ b/gcc/config/c6x/lef.c @@ -0,0 +1,47 @@ +/* Software floating-point emulation. + Return 1 iff a <= b, 0 otherwise. + Copyright (C) 1997,1999,2006,2007,2011 Free Software Foundation, Inc. + Contributed by Richard Henderson (rth@cygnus.com) and + Jakub Jelinek (jj@ultra.linux.cz). + + This file is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + In addition to the permissions in the GNU Lesser General Public + License, the Free Software Foundation gives you unlimited + permission to link the compiled version of this file into + combinations with other programs, and to distribute those + combinations without any restriction coming from the use of this + file. (The Lesser General Public License restrictions do apply in + other respects; for example, they cover modification of the file, + and distribution when not linked into a combine executable.) + + This file is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with GCC; see the file COPYING.LIB. If not see + <http://www.gnu.org/licenses/>. */ + +#include "config/soft-fp/soft-fp.h" +#include "config/soft-fp/single.h" + +CMPtype __c6xabi_lef(SFtype a, SFtype b) +{ + FP_DECL_EX; + FP_DECL_S(A); FP_DECL_S(B); + CMPtype r; + + FP_UNPACK_RAW_S(A, a); + FP_UNPACK_RAW_S(B, b); + FP_CMP_S(r, A, B, 2); + if (r == 2 && (FP_ISSIGNAN_S(A) || FP_ISSIGNAN_S(B))) + FP_SET_EXCEPTION(FP_EX_INVALID); + FP_HANDLE_EXCEPTIONS; + + return r <= 0; +} diff --git a/gcc/config/c6x/lib1funcs.asm b/gcc/config/c6x/lib1funcs.asm new file mode 100644 index 0000000..5bf3447 --- /dev/null +++ b/gcc/config/c6x/lib1funcs.asm @@ -0,0 +1,438 @@ +/* Copyright 2010, 2011 Free Software Foundation, Inc. + Contributed by Bernd Schmidt <bernds@codesourcery.com>. + +This file is free software; you can redistribute it and/or modify it +under the terms of the GNU General Public License as published by the +Free Software Foundation; either version 3, or (at your option) any +later version. + +This file is distributed in the hope that it will be useful, but +WITHOUT ANY WARRANTY; without even the implied warranty of +MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU +General Public License for more details. + +Under Section 7 of GPL version 3, you are granted additional +permissions described in the GCC Runtime Library Exception, version +3.1, as published by the Free Software Foundation. + +You should have received a copy of the GNU General Public License and +a copy of the GCC Runtime Library Exception along with this program; +see the files COPYING3 and COPYING.RUNTIME respectively. If not, see +<http://www.gnu.org/licenses/>. */ + + ;; ABI considerations for the divide functions + ;; The following registers are call-used: + ;; __c6xabi_divi A0,A1,A2,A4,A6,B0,B1,B2,B4,B5 + ;; __c6xabi_divu A0,A1,A2,A4,A6,B0,B1,B2,B4 + ;; __c6xabi_remi A1,A2,A4,A5,A6,B0,B1,B2,B4 + ;; __c6xabi_remu A1,A4,A5,A7,B0,B1,B2,B4 + ;; + ;; In our implementation, divu and remu are leaf functions, + ;; while both divi and remi call into divu. + ;; A0 is not clobbered by any of the functions. + ;; divu does not clobber B2 either, which is taken advantage of + ;; in remi. + ;; divi uses B5 to hold the original return address during + ;; the call to divu. + ;; remi uses B2 and A5 to hold the input values during the + ;; call to divu. It stores B3 in on the stack. + +#ifdef L_divsi3 +.text +.align 2 +.global __c6xabi_divi +.hidden __c6xabi_divi +.type __c6xabi_divi, STT_FUNC + +__c6xabi_divi: + call .s2 __c6xabi_divu +|| mv .d2 B3, B5 +|| cmpgt .l1 0, A4, A1 +|| cmpgt .l2 0, B4, B1 + + [A1] neg .l1 A4, A4 +|| [B1] neg .l2 B4, B4 +|| xor .s1x A1, B1, A1 + +#ifdef _TMS320C6400 + [A1] addkpc .s2 1f, B3, 4 +#else + [A1] mvkl .s2 1f, B3 + [A1] mvkh .s2 1f, B3 + nop 2 +#endif +1: + neg .l1 A4, A4 +|| mv .l2 B3,B5 +|| ret .s2 B5 + nop 5 +#endif + +#if defined L_modsi3 || defined L_divmodsi4 +.align 2 +#ifdef L_modsi3 +#define MOD_OUTPUT_REG A4 +.global __c6xabi_remi +.hidden __c6xabi_remi +.type __c6xabi_remi, STT_FUNC +#else +#define MOD_OUTPUT_REG A5 +.global __c6xabi_divremi +.hidden __c6xabi_divremi +.type __c6xabi_divremi, STT_FUNC +__c6xabi_divremi: +#endif + +__c6xabi_remi: + stw .d2t2 B3, *B15--[2] +|| cmpgt .l1 0, A4, A1 +|| cmpgt .l2 0, B4, B2 +|| mv .s1 A4, A5 +|| call .s2 __c6xabi_divu + + [A1] neg .l1 A4, A4 +|| [B2] neg .l2 B4, B4 +|| xor .s2x B2, A1, B0 +|| mv .d2 B4, B2 + +#ifdef _TMS320C6400 + [B0] addkpc .s2 1f, B3, 1 + [!B0] addkpc .s2 2f, B3, 1 + nop 2 +#else + [B0] mvkl .s2 1f,B3 + [!B0] mvkl .s2 2f,B3 + + [B0] mvkh .s2 1f,B3 + [!B0] mvkh .s2 2f,B3 +#endif +1: + neg .l1 A4, A4 +2: + ldw .d2t2 *++B15[2], B3 + +#ifdef _TMS320C6400_PLUS + mpy32 .m1x A4, B2, A6 + nop 3 + ret .s2 B3 + sub .l1 A5, A6, MOD_OUTPUT_REG + nop 4 +#else + mpyu .m1x A4, B2, A1 + nop 1 + mpylhu .m1x A4, B2, A6 +|| mpylhu .m2x B2, A4, B2 + nop 1 + add .l1x A6, B2, A6 +|| ret .s2 B3 + shl .s1 A6, 16, A6 + add .d1 A6, A1, A6 + sub .l1 A5, A6, MOD_OUTPUT_REG + nop 2 +#endif + +#endif + +#if defined L_udivsi3 || defined L_udivmodsi4 +.align 2 +#ifdef L_udivsi3 +.global __c6xabi_divu +.hidden __c6xabi_divu +.type __c6xabi_divu, STT_FUNC +__c6xabi_divu: +#else +.global __c6xabi_divremu +.hidden __c6xabi_divremu +.type __c6xabi_divremu, STT_FUNC +__c6xabi_divremu: +#endif + ;; We use a series of up to 31 subc instructions. First, we find + ;; out how many leading zero bits there are in the divisor. This + ;; gives us both a shift count for aligning (shifting) the divisor + ;; to the, and the number of times we have to execute subc. + + ;; At the end, we have both the remainder and most of the quotient + ;; in A4. The top bit of the quotient is computed first and is + ;; placed in A2. + + ;; Return immediately if the dividend is zero. Setting B4 to 1 + ;; is a trick to allow us to leave the following insns in the jump + ;; delay slot without affecting the result. + mv .s2x A4, B1 + +#ifndef _TMS320C6400 +[!b1] mvk .s2 1, B4 +#endif +[b1] lmbd .l2 1, B4, B1 +||[!b1] b .s2 B3 ; RETURN A +#ifdef _TMS320C6400 +||[!b1] mvk .d2 1, B4 +#endif +#ifdef L_udivmodsi4 +||[!b1] zero .s1 A5 +#endif + mv .l1x B1, A6 +|| shl .s2 B4, B1, B4 + + ;; The loop performs a maximum of 28 steps, so we do the + ;; first 3 here. + cmpltu .l1x A4, B4, A2 +[!A2] sub .l1x A4, B4, A4 +|| shru .s2 B4, 1, B4 +|| xor .s1 1, A2, A2 + + shl .s1 A2, 31, A2 +|| [b1] subc .l1x A4,B4,A4 +|| [b1] add .s2 -1, B1, B1 +[b1] subc .l1x A4,B4,A4 +|| [b1] add .s2 -1, B1, B1 + + ;; RETURN A may happen here (note: must happen before the next branch) +0: + cmpgt .l2 B1, 7, B0 +|| [b1] subc .l1x A4,B4,A4 +|| [b1] add .s2 -1, B1, B1 +[b1] subc .l1x A4,B4,A4 +|| [b1] add .s2 -1, B1, B1 +|| [b0] b .s1 0b +[b1] subc .l1x A4,B4,A4 +|| [b1] add .s2 -1, B1, B1 +[b1] subc .l1x A4,B4,A4 +|| [b1] add .s2 -1, B1, B1 +[b1] subc .l1x A4,B4,A4 +|| [b1] add .s2 -1, B1, B1 +[b1] subc .l1x A4,B4,A4 +|| [b1] add .s2 -1, B1, B1 +[b1] subc .l1x A4,B4,A4 +|| [b1] add .s2 -1, B1, B1 + ;; loop backwards branch happens here + + ret .s2 B3 +|| mvk .s1 32, A1 + sub .l1 A1, A6, A6 +#ifdef L_udivmodsi4 +|| extu .s1 A4, A6, A5 +#endif + shl .s1 A4, A6, A4 + shru .s1 A4, 1, A4 +|| sub .l1 A6, 1, A6 + or .l1 A2, A4, A4 + shru .s1 A4, A6, A4 + nop + +#endif + +#ifdef L_umodsi3 +.align 2 +.global __c6xabi_remu +.hidden __c6xabi_remu +.type __c6xabi_remu, STT_FUNC +__c6xabi_remu: + ;; The ABI seems designed to prevent these functions calling each other, + ;; so we duplicate most of the divsi3 code here. + mv .s2x A4, B1 +#ifndef _TMS320C6400 +[!b1] mvk .s2 1, B4 +#endif + lmbd .l2 1, B4, B1 +||[!b1] b .s2 B3 ; RETURN A +#ifdef _TMS320C6400 +||[!b1] mvk .d2 1, B4 +#endif + + mv .l1x B1, A7 +|| shl .s2 B4, B1, B4 + + cmpltu .l1x A4, B4, A1 +[!a1] sub .l1x A4, B4, A4 + shru .s2 B4, 1, B4 + +0: + cmpgt .l2 B1, 7, B0 +|| [b1] subc .l1x A4,B4,A4 +|| [b1] add .s2 -1, B1, B1 + ;; RETURN A may happen here (note: must happen before the next branch) +[b1] subc .l1x A4,B4,A4 +|| [b1] add .s2 -1, B1, B1 +|| [b0] b .s1 0b +[b1] subc .l1x A4,B4,A4 +|| [b1] add .s2 -1, B1, B1 +[b1] subc .l1x A4,B4,A4 +|| [b1] add .s2 -1, B1, B1 +[b1] subc .l1x A4,B4,A4 +|| [b1] add .s2 -1, B1, B1 +[b1] subc .l1x A4,B4,A4 +|| [b1] add .s2 -1, B1, B1 +[b1] subc .l1x A4,B4,A4 +|| [b1] add .s2 -1, B1, B1 + ;; loop backwards branch happens here + + ret .s2 B3 +[b1] subc .l1x A4,B4,A4 +|| [b1] add .s2 -1, B1, B1 +[b1] subc .l1x A4,B4,A4 + + extu .s1 A4, A7, A4 + nop 2 +#endif + +#if defined L_strasgi_64plus && defined _TMS320C6400_PLUS + +.align 2 +.global __c6xabi_strasgi_64plus +.hidden __c6xabi_strasgi_64plus +.type __c6xabi_strasgi_64plus, STT_FUNC +__c6xabi_strasgi_64plus: + shru .s2x a6, 2, b31 +|| mv .s1 a4, a30 +|| mv .d2 b4, b30 + + add .s2 -4, b31, b31 + + sploopd 1 +|| mvc .s2 b31, ilc + ldw .d2t2 *b30++, b31 + nop 4 + mv .s1x b31,a31 + spkernel 6, 0 +|| stw .d1t1 a31, *a30++ + + ret .s2 b3 + nop 5 +#endif + +#ifdef L_strasgi +.global __c6xabi_strasgi +.type __c6xabi_strasgi, STT_FUNC +__c6xabi_strasgi: + ;; This is essentially memcpy, with alignment known to be at least + ;; 4, and the size a multiple of 4 greater than or equal to 28. + ldw .d2t1 *B4++, A0 +|| mvk .s2 16, B1 + ldw .d2t1 *B4++, A1 +|| mvk .s2 20, B2 +|| sub .d1 A6, 24, A6 + ldw .d2t1 *B4++, A5 + ldw .d2t1 *B4++, A7 +|| mv .l2x A6, B7 + ldw .d2t1 *B4++, A8 + ldw .d2t1 *B4++, A9 +|| mv .s2x A0, B5 +|| cmpltu .l2 B2, B7, B0 + +0: + stw .d1t2 B5, *A4++ +||[b0] ldw .d2t1 *B4++, A0 +|| mv .s2x A1, B5 +|| mv .l2 B7, B6 + +[b0] sub .d2 B6, 24, B7 +||[b0] b .s2 0b +|| cmpltu .l2 B1, B6, B0 + +[b0] ldw .d2t1 *B4++, A1 +|| stw .d1t2 B5, *A4++ +|| mv .s2x A5, B5 +|| cmpltu .l2 12, B6, B0 + +[b0] ldw .d2t1 *B4++, A5 +|| stw .d1t2 B5, *A4++ +|| mv .s2x A7, B5 +|| cmpltu .l2 8, B6, B0 + +[b0] ldw .d2t1 *B4++, A7 +|| stw .d1t2 B5, *A4++ +|| mv .s2x A8, B5 +|| cmpltu .l2 4, B6, B0 + +[b0] ldw .d2t1 *B4++, A8 +|| stw .d1t2 B5, *A4++ +|| mv .s2x A9, B5 +|| cmpltu .l2 0, B6, B0 + +[b0] ldw .d2t1 *B4++, A9 +|| stw .d1t2 B5, *A4++ +|| mv .s2x A0, B5 +|| cmpltu .l2 B2, B7, B0 + + ;; loop back branch happens here + + cmpltu .l2 B1, B6, B0 +|| ret .s2 b3 + +[b0] stw .d1t1 A1, *A4++ +|| cmpltu .l2 12, B6, B0 +[b0] stw .d1t1 A5, *A4++ +|| cmpltu .l2 8, B6, B0 +[b0] stw .d1t1 A7, *A4++ +|| cmpltu .l2 4, B6, B0 +[b0] stw .d1t1 A8, *A4++ +|| cmpltu .l2 0, B6, B0 +[b0] stw .d1t1 A9, *A4++ + + ;; return happens here + +#endif + +#ifdef _TMS320C6400_PLUS +#ifdef L_push_rts +.align 2 +.global __c6xabi_push_rts +.hidden __c6xabi_push_rts +.type __c6xabi_push_rts, STT_FUNC +__c6xabi_push_rts: + stw .d2t2 B14, *B15--[2] + stdw .d2t1 A15:A14, *B15-- +|| b .s2x A3 + stdw .d2t2 B13:B12, *B15-- + stdw .d2t1 A13:A12, *B15-- + stdw .d2t2 B11:B10, *B15-- + stdw .d2t1 A11:A10, *B15-- + stdw .d2t2 B3:B2, *B15-- +#endif + +#ifdef L_pop_rts +.align 2 +.global __c6xabi_pop_rts +.hidden __c6xabi_pop_rts +.type __c6xabi_pop_rts, STT_FUNC +__c6xabi_pop_rts: + lddw .d2t2 *++B15, B3:B2 + lddw .d2t1 *++B15, A11:A10 + lddw .d2t2 *++B15, B11:B10 + lddw .d2t1 *++B15, A13:A12 + lddw .d2t2 *++B15, B13:B12 + lddw .d2t1 *++B15, A15:A14 +|| b .s2 B3 + ldw .d2t2 *++B15[2], B14 + nop 4 +#endif + +#ifdef L_call_stub +.align 2 +.global __c6xabi_call_stub +.type __c6xabi_call_stub, STT_FUNC +__c6xabi_call_stub: + stw .d2t1 A2, *B15--[2] + stdw .d2t1 A7:A6, *B15-- +|| call .s2 B31 + stdw .d2t1 A1:A0, *B15-- + stdw .d2t2 B7:B6, *B15-- + stdw .d2t2 B5:B4, *B15-- + stdw .d2t2 B1:B0, *B15-- + stdw .d2t2 B3:B2, *B15-- +|| addkpc .s2 1f, B3, 0 +1: + lddw .d2t2 *++B15, B3:B2 + lddw .d2t2 *++B15, B1:B0 + lddw .d2t2 *++B15, B5:B4 + lddw .d2t2 *++B15, B7:B6 + lddw .d2t1 *++B15, A1:A0 + lddw .d2t1 *++B15, A7:A6 +|| b .s2 B3 + ldw .d2t1 *++B15[2], A2 + nop 4 +#endif + +#endif + diff --git a/gcc/config/c6x/libgcc-c6xeabi.ver b/gcc/config/c6x/libgcc-c6xeabi.ver new file mode 100644 index 0000000..6bce556 --- /dev/null +++ b/gcc/config/c6x/libgcc-c6xeabi.ver @@ -0,0 +1,103 @@ +# Copyright (C) 2011 Free Software Foundation, Inc. +# +# This file is part of GCC. +# +# GCC is free software; you can redistribute it and/or modify +# it under the terms of the GNU General Public License as published by +# the Free Software Foundation; either version 3, or (at your option) +# any later version. +# +# GCC is distributed in the hope that it will be useful, +# but WITHOUT ANY WARRANTY; without even the implied warranty of +# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the +# GNU General Public License for more details. +# +# You should have received a copy of the GNU General Public License +# along with GCC; see the file COPYING3. If not see +# <http://www.gnu.org/licenses/>. + +GCC_4.7.0 { + __c6xabi_strasgi + __c6xabi_call_stub + __c6xabi_mpyll + __c6xabi_negll + __c6xabi_llshru + __c6xabi_llshl + __c6xabi_llshr + __c6xabi_fixfu + __c6xabi_fixdu + __c6xabi_fixflli + __c6xabi_fixdlli + __c6xabi_fixfull + __c6xabi_fixdull + __c6xabi_fltllif + __c6xabi_fltllid + __c6xabi_fltullf + __c6xabi_fltulld + __c6xabi_divlli + __c6xabi_remlli + __c6xabi_divull + __c6xabi_remull + __c6xabi_divremull + __c6xabi_gef + __c6xabi_gtf + __c6xabi_lef + __c6xabi_ltf + __c6xabi_eqf + __c6xabi_ged + __c6xabi_gtd + __c6xabi_led + __c6xabi_ltd + __c6xabi_eqd + __c6xabi_addf + __c6xabi_divf + __c6xabi_neqf + __c6xabi_cmpf + __c6xabi_mpyf + __c6xabi_negf + __c6xabi_subf + __c6xabi_unordf + __c6xabi_fixfi + __c6xabi_fltif + __c6xabi_fltuf + __c6xabi_addd + __c6xabi_divd + __c6xabi_neqd + __c6xabi_cmpd + __c6xabi_mpyd + __c6xabi_negd + __c6xabi_subd + __c6xabi_unordd + __c6xabi_fixdi + __c6xabi_fltid + __c6xabi_fltud + __c6xabi_cvtfd + __c6xabi_cvtdf + __c6xabi_mulcf + __c6xabi_mulcd + __c6xabi_divcf + __c6xabi_divcd + + __gnu_ltsf2 + __gnu_ltdf2 + __gnu_gesf2 + __gnu_gedf2 + __gnu_gtsf2 + __gnu_gtdf2 + __gnu_eqsf2 + __gnu_eqdf2 + + # Exception-Handling + _Unwind_Complete + _Unwind_VRS_Get + _Unwind_VRS_Set + _Unwind_VRS_Pop + __c6xabi_unwind_cpp_pr0 + __c6xabi_unwind_cpp_pr1 + __c6xabi_unwind_cpp_pr2 + __c6xabi_unwind_cpp_pr3 + __c6xabi_unwind_cpp_pr4 + # The libstdc++ exception-handling personality routine uses this + # GNU-specific entry point. + __gnu_unwind_frame +} diff --git a/gcc/config/c6x/ltd.c b/gcc/config/c6x/ltd.c new file mode 100644 index 0000000..5518583 --- /dev/null +++ b/gcc/config/c6x/ltd.c @@ -0,0 +1,47 @@ +/* Software floating-point emulation. + Return 1 iff a < b, 0 otherwise. + Copyright (C) 1997,1999,2006,2007,2011 Free Software Foundation, Inc. + Contributed by Richard Henderson (rth@cygnus.com) and + Jakub Jelinek (jj@ultra.linux.cz). + + This file is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + In addition to the permissions in the GNU Lesser General Public + License, the Free Software Foundation gives you unlimited + permission to link the compiled version of this file into + combinations with other programs, and to distribute those + combinations without any restriction coming from the use of this + file. (The Lesser General Public License restrictions do apply in + other respects; for example, they cover modification of the file, + and distribution when not linked into a combine executable.) + + This file is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with GCC; see the file COPYING.LIB. If not see + <http://www.gnu.org/licenses/>. */ + +#include "config/soft-fp/soft-fp.h" +#include "config/soft-fp/double.h" + +CMPtype __c6xabi_ltd(DFtype a, DFtype b) +{ + FP_DECL_EX; + FP_DECL_D(A); FP_DECL_D(B); + CMPtype r; + + FP_UNPACK_RAW_D(A, a); + FP_UNPACK_RAW_D(B, b); + FP_CMP_D(r, A, B, 2); + if (r == 2 && (FP_ISSIGNAN_D(A) || FP_ISSIGNAN_D(B))) + FP_SET_EXCEPTION(FP_EX_INVALID); + FP_HANDLE_EXCEPTIONS; + + return r < 0; +} diff --git a/gcc/config/c6x/ltf.c b/gcc/config/c6x/ltf.c new file mode 100644 index 0000000..7ffe9f8 --- /dev/null +++ b/gcc/config/c6x/ltf.c @@ -0,0 +1,47 @@ +/* Software floating-point emulation. + Return 1 iff a < b, 0 otherwise. + Copyright (C) 1997,1999,2006,2007,2011 Free Software Foundation, Inc. + Contributed by Richard Henderson (rth@cygnus.com) and + Jakub Jelinek (jj@ultra.linux.cz). + + This file is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + In addition to the permissions in the GNU Lesser General Public + License, the Free Software Foundation gives you unlimited + permission to link the compiled version of this file into + combinations with other programs, and to distribute those + combinations without any restriction coming from the use of this + file. (The Lesser General Public License restrictions do apply in + other respects; for example, they cover modification of the file, + and distribution when not linked into a combine executable.) + + This file is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with GCC; see the file COPYING.LIB. If not see + <http://www.gnu.org/licenses/>. */ + +#include "config/soft-fp/soft-fp.h" +#include "config/soft-fp/single.h" + +CMPtype __c6xabi_ltf(SFtype a, SFtype b) +{ + FP_DECL_EX; + FP_DECL_S(A); FP_DECL_S(B); + CMPtype r; + + FP_UNPACK_RAW_S(A, a); + FP_UNPACK_RAW_S(B, b); + FP_CMP_S(r, A, B, 2); + if (r == 2 && (FP_ISSIGNAN_S(A) || FP_ISSIGNAN_S(B))) + FP_SET_EXCEPTION(FP_EX_INVALID); + FP_HANDLE_EXCEPTIONS; + + return r < 0; +} diff --git a/gcc/config/c6x/predicates.md b/gcc/config/c6x/predicates.md new file mode 100644 index 0000000..c4ccf54 --- /dev/null +++ b/gcc/config/c6x/predicates.md @@ -0,0 +1,226 @@ +/* Predicates for TI C6X + Copyright (C) 2010, 2011 Free Software Foundation, Inc. + Contributed by Andrew Jenner <andrew@codesourcery.com> + Contributed by Bernd Schmidt <bernds@codesourcery.com> + + This file is part of GCC. + + GCC is free software; you can redistribute it and/or modify it + under the terms of the GNU General Public License as published + by the Free Software Foundation; either version 3, or (at your + option) any later version. + + GCC is distributed in the hope that it will be useful, but WITHOUT + ANY WARRANTY; without even the implied warranty of MERCHANTABILITY + or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public + License for more details. + + You should have received a copy of the GNU General Public License + along with GCC; see the file COPYING3. If not see + <http://www.gnu.org/licenses/>. */ + +(define_predicate "reg_or_const_int_operand" + (ior (match_operand 0 "register_operand") + (match_operand 0 "const_int_operand"))) + +(define_predicate "const_vector_operand" + (match_code "const_vector")) + +(define_predicate "scst5_operand" + (and (match_operand 0 "const_int_operand") + (match_test "satisfies_constraint_Is5 (op)"))) + +(define_predicate "reg_or_ucst4_operand" + (ior (match_operand 0 "register_operand") + (and (match_operand 0 "const_int_operand") + (match_test "satisfies_constraint_Iu4 (op)")))) + +(define_predicate "reg_or_scst5_operand" + (ior (match_operand 0 "register_operand") + (match_operand 0 "scst5_operand"))) + +(define_predicate "reg_or_ucst5_operand" + (ior (match_operand 0 "register_operand") + (and (match_operand 0 "const_int_operand") + (match_test "satisfies_constraint_Iu5 (op)")))) + +(define_predicate "addsi_operand" + (ior (match_operand 0 "register_operand") + (and (match_operand 0 "const_int_operand") + (match_test "satisfies_constraint_IsB (op)")))) + +(define_predicate "andsi_operand" + (ior (match_operand 0 "reg_or_scst5_operand") + (and (match_operand 0 "const_int_operand") + (match_test "satisfies_constraint_Jc (op)")))) + +(define_predicate "iorsi_operand" + (ior (match_operand 0 "reg_or_scst5_operand") + (and (match_operand 0 "const_int_operand") + (match_test "satisfies_constraint_Js (op)")))) + +(define_predicate "insv_operand" + (and (match_operand 0 "const_int_operand") + (match_test "INTVAL (op) == 0 || INTVAL (op) == -1"))) + +(define_predicate "c6x_jump_operand" + (match_code "label_ref,symbol_ref,reg")) + +(define_predicate "c6x_call_operand" + (ior (match_code "symbol_ref,reg") + (and (match_code "subreg") + (match_test "GET_CODE (XEXP (op, 0)) == REG"))) +{ + /* The linker transforms jumps to undefined weak symbols in a way that + is incompatible with our code generation. */ + return (GET_CODE (op) != SYMBOL_REF + || (!SYMBOL_REF_WEAK (op) + && !c6x_long_call_p (op))); +}) + +;; Returns 1 if OP is a symbolic operand, i.e. a symbol_ref or a label_ref, +;; possibly with an offset. +(define_predicate "symbolic_operand" + (ior (match_code "symbol_ref,label_ref") + (and (match_code "const") + (match_test "GET_CODE (XEXP (op,0)) == PLUS + && (GET_CODE (XEXP (XEXP (op, 0), 0)) == SYMBOL_REF + || GET_CODE (XEXP (XEXP (op, 0), 0)) == LABEL_REF) + && GET_CODE (XEXP (XEXP (op, 0), 1)) == CONST_INT")))) + +(define_predicate "const_int_or_symbolic_operand" + (ior (match_operand 0 "symbolic_operand") + (match_operand 0 "const_int_operand"))) + +;; Return nonzero iff OP is one of the integer constants 2, 4 or 8. +(define_predicate "adda_scale_operand" + (and (match_code "const_int") + (match_test "INTVAL (op) == 2 || INTVAL (op) == 4 + || ((TARGET_INSNS_64 || TARGET_INSNS_67) + && INTVAL (op) == 8)"))) + +;; Return nonzero iff OP is one of the integer constants 2 or 4. +(define_predicate "suba_scale_operand" + (and (match_code "const_int") + (match_test "INTVAL (op) == 2 || INTVAL (op) == 4"))) + +;; True if this operator is valid for predication. +(define_predicate "predicate_operator" + (match_code "eq,ne")) + +(define_predicate "c6x_comparison_operator" + (match_code "eq,ltu,gtu,lt,gt")) + +(define_predicate "non_c6x_comparison_operator" + (match_code "ne,leu,geu,le,ge")) + +;; FP Comparisons handled by c6x_expand_compare. +(define_predicate "c6x_fp_comparison_operator" + (ior (match_code "eq,lt,gt,le,ge") + (and (match_test "TARGET_FP") + (match_code "ltgt,uneq,unlt,ungt,unle,unge,ordered,unordered")))) + +(define_predicate "c6x_any_comparison_operand" + (match_code "eq,lt,gt,le,ge,ltu,gtu") +{ + rtx op0 = XEXP (op, 0); + rtx op1 = XEXP (op, 1); + if (ltugtu_operator (op, SImode) + && register_operand (op0, SImode) + && ((TARGET_INSNS_64 && reg_or_ucst5_operand (op1, SImode)) + || (!TARGET_INSNS_64 && reg_or_ucst4_operand (op1, SImode)))) + return true; + if (eqltgt_operator (op, SImode) + && register_operand (op0, SImode) + && reg_or_scst5_operand (op1, SImode)) + return true; + if (!TARGET_FP) + return false; + if (!eqltgt_operator (op, SFmode) && !eqltgt_operator (op, DFmode)) + return false; + if (register_operand (op0, GET_MODE (op)) + && register_operand (op1, GET_MODE (op))) + return true; + return false; +}) + +(define_predicate "ltugtu_operator" + (match_code "ltu,gtu")) + +(define_predicate "eqltgt_operator" + (match_code "eq,lt,gt")) + +(define_predicate "eqne_operator" + (match_code "eq,ne")) + +(define_predicate "predicate_register" + (and (match_code "reg") + (ior (match_test "REGNO_REG_CLASS (REGNO (op)) == PREDICATE_A_REGS") + (match_test "REGNO_REG_CLASS (REGNO (op)) == PREDICATE_B_REGS")))) + +;; Allow const_ints for things like the real_mult patterns. +(define_predicate "a_register" + (ior (and (match_code "reg") + (match_test "A_REGNO_P (REGNO (op))")) + (and (match_code "const_int") + (match_test "A_REGNO_P (INTVAL (op))")))) + +(define_predicate "b_register" + (ior (and (match_code "reg") + (match_test "B_REGNO_P (REGNO (op))")) + (and (match_code "const_int") + (match_test "B_REGNO_P (INTVAL (op))")))) + +(define_predicate "pic_register_operand" + (and (match_code "reg") + (match_test "op == pic_offset_table_rtx"))) + +;; True if OP refers to a symbol in the sdata section. +(define_predicate "sdata_symbolic_operand" + (match_code "symbol_ref,const") +{ + HOST_WIDE_INT offset = 0, size = 0; + tree t; + + switch (GET_CODE (op)) + { + case CONST: + op = XEXP (op, 0); + if (GET_CODE (op) != PLUS + || GET_CODE (XEXP (op, 0)) != SYMBOL_REF + || GET_CODE (XEXP (op, 1)) != CONST_INT) + return false; + offset = INTVAL (XEXP (op, 1)); + op = XEXP (op, 0); + /* FALLTHRU */ + + case SYMBOL_REF: + /* For shared libraries, only allow symbols we know are local. + For executables, the linker knows to create copy relocs if + necessary so we can use DP-relative addressing for all small + objects. */ + if ((c6x_initial_flag_pic && !SYMBOL_REF_LOCAL_P (op)) + || !SYMBOL_REF_SMALL_P (op)) + return false; + + /* Note that in addition to DECLs, we can get various forms + of constants here. */ + t = SYMBOL_REF_DECL (op); + if (DECL_P (t)) + t = DECL_SIZE_UNIT (t); + else + t = TYPE_SIZE_UNIT (TREE_TYPE (t)); + if (t && host_integerp (t, 0)) + { + size = tree_low_cst (t, 0); + if (size < 0) + size = 0; + } + + /* Don't allow addressing outside the object. */ + return (offset >= 0 && offset <= size); + + default: + gcc_unreachable (); + } +}) diff --git a/gcc/config/c6x/sfp-machine.h b/gcc/config/c6x/sfp-machine.h new file mode 100644 index 0000000..2c90e58 --- /dev/null +++ b/gcc/config/c6x/sfp-machine.h @@ -0,0 +1,120 @@ +/* Soft-FP definitions for TI C6X. + Copyright (C) 2010, 2011 Free Software Foundation, Inc. + + This files is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + In addition to the permissions in the GNU Lesser General Public + License, the Free Software Foundation gives you unlimited + permission to link the compiled version of this file into + combinations with other programs, and to distribute those + combinations without any restriction coming from the use of this + file. (The Lesser General Public License restrictions do apply in + other respects; for example, they cover modification of the file, + and distribution when not linked into a combine executable.) + + This file is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with GCC; see the file COPYING.LIB. If not see + <http://www.gnu.org/licenses/>. */ + +#define _FP_W_TYPE_SIZE 32 +#define _FP_W_TYPE unsigned long +#define _FP_WS_TYPE signed long +#define _FP_I_TYPE long + +#define _FP_MUL_MEAT_S(R,X,Y) \ + _FP_MUL_MEAT_1_wide(_FP_WFRACBITS_S,R,X,Y,umul_ppmm) +#define _FP_MUL_MEAT_D(R,X,Y) \ + _FP_MUL_MEAT_2_wide(_FP_WFRACBITS_D,R,X,Y,umul_ppmm) +#define _FP_MUL_MEAT_Q(R,X,Y) \ + _FP_MUL_MEAT_4_wide(_FP_WFRACBITS_Q,R,X,Y,umul_ppmm) + +#define _FP_DIV_MEAT_S(R,X,Y) _FP_DIV_MEAT_1_loop(S,R,X,Y) +#define _FP_DIV_MEAT_D(R,X,Y) _FP_DIV_MEAT_2_udiv(D,R,X,Y) +#define _FP_DIV_MEAT_Q(R,X,Y) _FP_DIV_MEAT_4_udiv(Q,R,X,Y) + +#define _FP_NANFRAC_H ((_FP_QNANBIT_H << 1) - 1) +#define _FP_NANFRAC_S ((_FP_QNANBIT_S << 1) - 1) +#define _FP_NANFRAC_D ((_FP_QNANBIT_D << 1) - 1), -1 +#define _FP_NANFRAC_Q ((_FP_QNANBIT_Q << 1) - 1), -1, -1, -1 +#define _FP_NANSIGN_H 0 +#define _FP_NANSIGN_S 0 +#define _FP_NANSIGN_D 0 +#define _FP_NANSIGN_Q 0 + +#define _FP_KEEPNANFRACP 1 + +/* Someone please check this. */ +#define _FP_CHOOSENAN(fs, wc, R, X, Y, OP) \ + do { \ + if ((_FP_FRAC_HIGH_RAW_##fs(X) & _FP_QNANBIT_##fs) \ + && !(_FP_FRAC_HIGH_RAW_##fs(Y) & _FP_QNANBIT_##fs)) \ + { \ + R##_s = Y##_s; \ + _FP_FRAC_COPY_##wc(R,Y); \ + } \ + else \ + { \ + R##_s = X##_s; \ + _FP_FRAC_COPY_##wc(R,X); \ + } \ + R##_c = FP_CLS_NAN; \ + } while (0) + +#define __LITTLE_ENDIAN 1234 +#define __BIG_ENDIAN 4321 + +#if defined _BIG_ENDIAN +# define __BYTE_ORDER __BIG_ENDIAN +#else +# define __BYTE_ORDER __LITTLE_ENDIAN +#endif + + +/* Define ALIASNAME as a strong alias for NAME. */ +# define strong_alias(name, aliasname) _strong_alias(name, aliasname) +# define _strong_alias(name, aliasname) \ + extern __typeof (name) aliasname __attribute__ ((alias (#name))); + +/* Rename helper functions to the names specified in the C6000 ELF ABI. */ +#define __fixdfsi __c6xabi_fixdi +#define __fixsfsi __c6xabi_fixfi +#define __floatsidf __c6xabi_fltid +#define __floatunsidf __c6xabi_fltud +#define __floatsisf __c6xabi_fltif +#define __floatunsisf __c6xabi_fltuf +#define __truncdfsf2 __c6xabi_cvtdf +#define __extendsfdf2 __c6xabi_cvtfd +#define __adddf3 __c6xabi_addd +#define __subdf3 __c6xabi_subd +#define __muldf3 __c6xabi_mpyd +#define __divdf3 __c6xabi_divd +#define __negdf2 __c6xabi_negd +#define __absdf2 __c6xabi_absd +#define __addsf3 __c6xabi_addf +#define __subsf3 __c6xabi_subf +#define __mulsf3 __c6xabi_mpyf +#define __divsf3 __c6xabi_divf +#define __negsf2 __c6xabi_negf +#define __abssf2 __c6xabi_absf +#define __lesf2 __c6xabi_cmpf +#define __ledf2 __c6xabi_cmpd +#define __ltsf2 __gnu_ltsf2 +#define __ltdf2 __gnu_ltdf2 +#define __gesf2 __gnu_gesf2 +#define __gedf2 __gnu_gedf2 +#define __gtsf2 __gnu_gtsf2 +#define __gtdf2 __gnu_gtdf2 +#define __eqsf2 __gnu_eqsf2 +#define __eqdf2 __gnu_eqdf2 +#define __nesf2 __c6xabi_neqf +#define __nedf2 __c6xabi_neqd +#define __unordsf2 __c6xabi_unordf +#define __unorddf2 __c6xabi_unordd diff --git a/gcc/config/c6x/sync.md b/gcc/config/c6x/sync.md new file mode 100644 index 0000000..05ab02a --- /dev/null +++ b/gcc/config/c6x/sync.md @@ -0,0 +1,270 @@ +;; GCC machine description for C6X synchronization instructions. +;; Copyright (C) 2011 Free Software Foundation, Inc. +;; +;; This file is part of GCC. +;; +;; GCC is free software; you can redistribute it and/or modify +;; it under the terms of the GNU General Public License as published by +;; the Free Software Foundation; either version 3, or (at your option) +;; any later version. +;; +;; GCC is distributed in the hope that it will be useful, +;; but WITHOUT ANY WARRANTY; without even the implied warranty of +;; MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the +;; GNU General Public License for more details. +;; +;; You should have received a copy of the GNU General Public License +;; along with GCC; see the file COPYING3. If not see +;; <http://www.gnu.org/licenses/>. + +;; C64X+ has atomic instructions, but they are not atomic on all +;; devices and have other problems. We use normal loads and stores, +;; and place them in overlapping branch shadows to ensure interrupts +;; are disabled during the sequence, which guarantees atomicity on all +;; single-core systems. + +(define_code_iterator FETCHOP [plus minus ior xor and]) +(define_code_attr fetchop_name + [(plus "add") (minus "sub") (ior "ior") (xor "xor") (and "and")]) +(define_code_attr fetchop_pred + [(plus "reg_or_scst5_operand") (minus "register_operand") + (ior "reg_or_scst5_operand") (xor "reg_or_scst5_operand") + (and "reg_or_scst5_operand")]) +(define_code_attr fetchop_constr + [(plus "bIs5") (minus "b") (ior "bIs5") (xor "bIs5") (and "bIs5")]) +(define_code_attr fetchop_opcode + [(plus "add") (minus "sub") (ior "or") (xor "xor") (and "and")]) +(define_code_attr fetchop_inops02 + [(plus "%2, %0") (minus "%0, %2") (ior "%2, %0") (xor "%2, %0") + (and "%2, %0")]) +(define_code_attr fetchop_inops21 + [(plus "%1, %2") (minus "%2, %1") (ior "%1, %2") (xor "%1, %2") + (and "%1, %2")]) + +(define_expand "sync_compare_and_swapsi" + [(parallel + [(set (match_operand:SI 0 "register_operand" "") + (match_operand:SI 1 "memory_operand" "")) + (set (match_dup 1) + (unspec_volatile:SI + [(match_operand:SI 2 "register_operand" "") + (match_operand:SI 3 "register_operand" "")] + UNSPECV_CAS)) + (clobber (match_scratch:SI 4 ""))])] + "" +{ +}) + +(define_expand "sync_<fetchop_name>si" + [(parallel + [(set (match_operand:SI 0 "memory_operand" "") + (unspec:SI + [(FETCHOP:SI (match_dup 0) + (match_operand:SI 1 "<fetchop_pred>" ""))] + UNSPEC_ATOMIC)) + (clobber (match_scratch:SI 2 ""))])] + "" +{ +}) + +(define_expand "sync_old_<fetchop_name>si" + [(parallel + [(set (match_operand:SI 0 "register_operand" "") + (match_operand:SI 1 "memory_operand" "")) + (set (match_dup 1) + (unspec:SI + [(FETCHOP:SI (match_dup 1) + (match_operand:SI 2 "<fetchop_pred>" ""))] + UNSPEC_ATOMIC)) + (clobber (match_scratch:SI 3 ""))])] + "" +{ +}) + +(define_expand "sync_new_<fetchop_name>si" + [(parallel + [(set (match_operand:SI 0 "register_operand" "") + (FETCHOP:SI (match_operand:SI 1 "memory_operand" "") + (match_operand:SI 2 "<fetchop_pred>" ""))) + (set (match_dup 1) + (unspec:SI [(FETCHOP:SI (match_dup 1) (match_dup 2))] + UNSPEC_ATOMIC)) + (clobber (match_scratch:SI 3 ""))])] + "" +{ +}) + +(define_expand "sync_nandsi" + [(parallel + [(set (match_operand:SI 0 "memory_operand" "") + (unspec:SI + [(not:SI (and:SI (match_dup 0) + (match_operand:SI 1 "reg_or_scst5_operand" "")))] + UNSPEC_ATOMIC)) + (clobber (match_scratch:SI 2 ""))])] + "" +{ +}) + +(define_expand "sync_old_nandsi" + [(parallel + [(set (match_operand:SI 0 "register_operand" "") + (match_operand:SI 1 "memory_operand" "")) + (set (match_dup 1) + (unspec:SI + [(not:SI (and:SI (match_dup 1) + (match_operand:SI 2 "reg_or_scst5_operand" "")))] + UNSPEC_ATOMIC)) + (clobber (match_scratch:SI 3 ""))])] + "" +{ +}) + +(define_expand "sync_new_nandsi" + [(parallel + [(set (match_operand:SI 0 "register_operand" "") + (not:SI (and:SI (match_operand:SI 1 "memory_operand" "") + (match_operand:SI 2 "reg_or_scst5_operand" "")))) + (set (match_dup 1) + (unspec:SI [(not:SI (and:SI (match_dup 1) (match_dup 2)))] + UNSPEC_ATOMIC)) + (clobber (match_scratch:SI 3 ""))])] + "" +{ +}) + +(define_insn "*sync_compare_and_swapsi" + [(set (match_operand:SI 0 "register_operand" "=&b") + (match_operand:SI 1 "memory_operand" "+m")) + (set (match_dup 1) + (unspec_volatile:SI + [(match_operand:SI 2 "register_operand" "B") + (match_operand:SI 3 "register_operand" "b")] + UNSPECV_CAS)) + (clobber (match_scratch:SI 4 "=&B"))] + "" + "0: b .s2 1f ; 0\n\\ + || ldw .d%U1t%U0 %1, %0\n\\ + nop 4\n\\ +|| b .s2 2f ; 1\n\\ + cmpeq .l2 %0, %2, %2 ; 5\n\\ +1: [%2] stw .d%U1t%U3 %3, %1 ; 6\n\\ +2:" + [(set_attr "type" "atomic")]) + +(define_insn "sync_<fetchop_name>si_insn" + [(set (match_operand:SI 0 "memory_operand" "+m") + (unspec:SI + [(FETCHOP:SI (match_dup 0) + (match_operand:SI 1 "<fetchop_pred>" "<fetchop_constr>"))] + UNSPEC_ATOMIC)) + (clobber (match_scratch:SI 2 "=&B"))] + "" + "0: b .s2 1f ; 0\n\\ +|| ldw .d%U0t%U2 %0, %2\n\\ + nop 4\n\\ +|| b .s2 2f ; 1\n\\ + <fetchop_opcode> .l2 <fetchop_inops21>, %2 ; 5\n\\ +1: stw .d%U0t%U2 %2, %0 ; 6\n\\ +2:" + [(set_attr "type" "atomic")]) + +(define_insn "sync_old_<fetchop_name>si_insn" + [(set (match_operand:SI 0 "register_operand" "=&b") + (match_operand:SI 1 "memory_operand" "+m")) + (set (match_dup 1) + (unspec:SI + [(FETCHOP:SI (match_dup 1) + (match_operand:SI 2 "<fetchop_pred>" "<fetchop_constr>"))] + UNSPEC_ATOMIC)) + (clobber (match_scratch:SI 3 "=&B"))] + "" + "0: b .s2 1f ; 0\n\\ +|| ldw .d%U1t%U0 %1, %0\n\\ + nop 4\n\\ +|| b .s2 2f ; 1\n\\ + <fetchop_opcode> .l2 <fetchop_inops02>, %3 ; 5\n\\ +1: stw .d%U1t%U3 %3, %1 ; 6\n\\ +2:" + [(set_attr "type" "atomic")]) + +(define_insn "sync_new_<fetchop_name>si_insn" + [(set (match_operand:SI 0 "register_operand" "=&b") + (FETCHOP:SI (match_operand:SI 1 "memory_operand" "+m") + (match_operand:SI 2 "<fetchop_pred>" "<fetchop_constr>"))) + (set (match_dup 1) + (unspec:SI + [(FETCHOP:SI (match_dup 1) + (match_dup 2))] + UNSPEC_ATOMIC)) + (clobber (match_scratch:SI 3 "=&B"))] + "" + "0: b .s2 1f ; 0\n\\ +|| ldw .d%U1t%U0 %1, %0\n\\ + nop 4\n\\ +|| b .s2 2f ; 1\n\\ + <fetchop_opcode> .l2 <fetchop_inops02>, %0 ; 5\n\\ +1: stw .d%U1t%U0 %0, %1 ; 6\n\\ +2:" + [(set_attr "type" "atomic")]) + +(define_insn "sync_nandsi_insn" + [(set (match_operand:SI 0 "memory_operand" "+m") + (unspec:SI + [(not:SI (and:SI (match_dup 0) + (match_operand:SI 1 "reg_or_scst5_operand" "bIs5")))] + UNSPEC_ATOMIC)) + (clobber (match_scratch:SI 2 "=&B"))] + "" + "0: b .s2 1f ; 0\n\\ +|| ldw .d%U0t%U2 %0, %2\n\\ + nop 1\n\\ + nop 3\n\\ +|| b .s2 2f ; 2\n\\ + and .l2 %1, %2, %2 ; 5\n\\ +1: not .l2 %2, %2 ; 6\n\\ + stw .d%U0t%U2 %2, %0 ; 7\n\\ +2:" + [(set_attr "type" "atomic")]) + +(define_insn "sync_old_nandsi_insn" + [(set (match_operand:SI 0 "register_operand" "=&b") + (match_operand:SI 1 "memory_operand" "+m")) + (set (match_dup 1) + (unspec:SI + [(not:SI (and:SI (match_dup 1) + (match_operand:SI 2 "reg_or_scst5_operand" "bIs5")))] + UNSPEC_ATOMIC)) + (clobber (match_scratch:SI 3 "=&B"))] + "" + "0: b .s2 1f ; 0\n\\ +|| ldw .d%U1t%U0 %1, %0\n\\ + nop 1\n\\ + nop 3\n\\ +|| b .s2 2f ; 2\n\\ + and .l2 %2, %0, %3 ; 5\n\\ +1: not .l2 %3, %3 ; 6\n\\ + stw .d%U1t%U3 %3, %1 ; 7\n\\ +2:" + [(set_attr "type" "atomic")]) + +(define_insn "sync_new_nandsi_insn" + [(set (match_operand:SI 0 "register_operand" "=&b") + (not:SI (and:SI (match_operand:SI 1 "memory_operand" "+m") + (match_operand:SI 2 "reg_or_scst5_operand" "bIs5")))) + (set (match_dup 1) + (unspec:SI + [(not:SI (and:SI (match_dup 1) (match_dup 2)))] + UNSPEC_ATOMIC)) + (clobber (match_scratch:SI 3 "=&B"))] + "" + "0: b .s2 1f ; 0\n\\ +|| ldw .d%U1t%U0 %1, %0\n\\ + nop 1\n\\ + nop 3\n\\ +|| b .s2 2f ; 2\n\\ + and .l2 %2, %0, %0 ; 5\n\\ +1: not .l2 %0, %0 ; 6\n\\ + stw .d%U1t%U0 %0, %1 ; 7\n\\ +2:" + [(set_attr "type" "atomic")]) diff --git a/gcc/config/c6x/t-c6x b/gcc/config/c6x/t-c6x new file mode 100644 index 0000000..6a7ced7 --- /dev/null +++ b/gcc/config/c6x/t-c6x @@ -0,0 +1,43 @@ +# Target Makefile Fragment for TI C6X. +# Copyright (C) 2010, 2011 +# Free Software Foundation, Inc. +# Contributed by CodeSourcery. +# +# This file is part of GCC. +# +# GCC is free software; you can redistribute it and/or modify it +# under the terms of the GNU General Public License as published +# by the Free Software Foundation; either version 3, or (at your +# option) any later version. +# +# GCC is distributed in the hope that it will be useful, but WITHOUT +# ANY WARRANTY; without even the implied warranty of MERCHANTABILITY +# or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public +# License for more details. +# +# You should have received a copy of the GNU General Public License +# along with GCC; see the file COPYING3. If not see +# <http://www.gnu.org/licenses/>. + +MD_INCLUDES= $(srcdir)/config/c6x/constraints.md \ + $(srcdir)/config/c6x/predicates.md \ + $(srcdir)/config/c6x/c6x-mult.md \ + $(srcdir)/config/c6x/c6x-sched.md + +s-config s-conditions s-flags s-codes s-constants s-emit s-recog s-preds \ + s-opinit s-extract s-peep s-attr s-attrtab s-output: $(MD_INCLUDES) + +$(srcdir)/config/c6x/c6x-sched.md: $(srcdir)/config/c6x/gensched.sh \ + $(srcdir)/config/c6x/c6x-sched.md.in + $(SHELL) $(srcdir)/config/c6x/gensched.sh \ + $(srcdir)/config/c6x/c6x-sched.md.in > $@ + +$(srcdir)/config/c6x/c6x-mult.md: $(srcdir)/config/c6x/genmult.sh \ + $(srcdir)/config/c6x/c6x-mult.md.in + $(SHELL) $(srcdir)/config/c6x/genmult.sh \ + $(srcdir)/config/c6x/c6x-mult.md.in > $@ + +$(srcdir)/config/c6x/c6x-tables.opt: $(srcdir)/config/c6x/genopt.sh \ + $(srcdir)/config/c6x/c6x-isas.def + $(SHELL) $(srcdir)/config/c6x/genopt.sh $(srcdir)/config/c6x > \ + $(srcdir)/config/c6x/c6x-tables.opt diff --git a/gcc/config/c6x/t-c6x-elf b/gcc/config/c6x/t-c6x-elf new file mode 100644 index 0000000..ed28642 --- /dev/null +++ b/gcc/config/c6x/t-c6x-elf @@ -0,0 +1,66 @@ +# Target Makefile Fragment for TI C6X using ELF. +# Copyright (C) 2010, 2011 Free Software Foundation, Inc. +# Contributed by CodeSourcery. +# +# This file is part of GCC. +# +# GCC is free software; you can redistribute it and/or modify it +# under the terms of the GNU General Public License as published +# by the Free Software Foundation; either version 3, or (at your +# option) any later version. +# +# GCC is distributed in the hope that it will be useful, but WITHOUT +# ANY WARRANTY; without even the implied warranty of MERCHANTABILITY +# or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public +# License for more details. +# +# You should have received a copy of the GNU General Public License +# along with GCC; see the file COPYING3. If not see +# <http://www.gnu.org/licenses/>. + +LIB1ASMSRC = c6x/lib1funcs.asm +LIB1ASMFUNCS = _divsi3 _udivsi3 _umodsi3 _modsi3 _udivmodsi4 _divmodsi4 +LIB1ASMFUNCS += _strasgi _strasgi_64plus _clzsi2 _clzdi2 _clz +LIB1ASMFUNCS += _push_rts _pop_rts _call_stub + +LIB2FUNCS_EXCLUDE = _cmpdi2 _ucmpdi2 _gcc_bcmp _eprintf _clzsi _clzdi + +LIB2FUNCS_EXTRA = $(srcdir)/config/c6x/gef.c \ + $(srcdir)/config/c6x/gtf.c \ + $(srcdir)/config/c6x/lef.c \ + $(srcdir)/config/c6x/ltf.c \ + $(srcdir)/config/c6x/eqf.c \ + $(srcdir)/config/c6x/ged.c \ + $(srcdir)/config/c6x/gtd.c \ + $(srcdir)/config/c6x/led.c \ + $(srcdir)/config/c6x/ltd.c \ + $(srcdir)/config/c6x/eqd.c + +# Use this variant for fully testing all CPU types +#MULTILIB_OPTIONS = mbig-endian march=c674x/march=c64x/march=c67x/march=c67x+/march=c62x +#MULTILIB_DIRNAMES = be c674x c64x c67x c67x+ c62x + +MULTILIB_OPTIONS = mbig-endian march=c674x +MULTILIB_DIRNAMES = be c674x +MULTILIB_EXCEPTIONS = +MULTILIB_MATCHES = + +# Assemble startup files. +$(T)crti.o: $(srcdir)/config/c6x/crti.s $(GCC_PASSES) + $(GCC_FOR_TARGET) $(GCC_CFLAGS) $(MULTILIB_CFLAGS) $(INCLUDES) \ + -c -o $(T)crti.o $(CRTSTUFF_T_CFLAGS) -x assembler-with-cpp \ + $(srcdir)/config/c6x/crti.s + +$(T)crtn.o: $(srcdir)/config/c6x/crtn.s $(GCC_PASSES) + $(GCC_FOR_TARGET) $(GCC_CFLAGS) $(MULTILIB_CFLAGS) $(INCLUDES) \ + -c -o $(T)crtn.o $(CRTSTUFF_T_CFLAGS) -x assembler-with-cpp \ + $(srcdir)/config/c6x/crtn.s + +EXTRA_MULTILIB_PARTS = crtbegin.o crtend.o crtbeginS.o crtendS.o crti.o crtn.o + +# Avoid failures when the user's GOT becomes too large. +CRTSTUFF_T_CFLAGS = -msdata=none +CRTSTUFF_T_CFLAGS_S = -msdata=none +TARGET_LIBGCC2_CFLAGS = -msdata=none + +SHLIB_MAPFILES += $(srcdir)/config/c6x/libgcc-c6xeabi.ver diff --git a/gcc/config/c6x/t-c6x-softfp b/gcc/config/c6x/t-c6x-softfp new file mode 100644 index 0000000..5df90eb --- /dev/null +++ b/gcc/config/c6x/t-c6x-softfp @@ -0,0 +1,9 @@ +softfp_float_modes := sf df +softfp_int_modes := si di +softfp_extensions := sfdf +softfp_truncations := dfsf +softfp_machine_header := c6x/sfp-machine.h +softfp_exclude_libgcc2 := y + +# softfp seems to be missing a whole bunch of prototypes. +TARGET_LIBGCC2_CFLAGS += -Wno-missing-prototypes diff --git a/gcc/config/c6x/t-c6x-uclinux b/gcc/config/c6x/t-c6x-uclinux new file mode 100644 index 0000000..fdc447a --- /dev/null +++ b/gcc/config/c6x/t-c6x-uclinux @@ -0,0 +1,7 @@ +MULTILIB_OSDIRNAMES = march.c674x=!c674x +MULTILIB_OSDIRNAMES += mbig-endian=!be +MULTILIB_OSDIRNAMES += mbig-endian/march.c674x=!be/c674x + +CRTSTUFF_T_CFLAGS = -fPIC -msdata=none +CRTSTUFF_T_CFLAGS_S = -fPIC -msdata=none +TARGET_LIBGCC2_CFLAGS = -fPIC -msdata=none diff --git a/gcc/config/c6x/uclinux-elf.h b/gcc/config/c6x/uclinux-elf.h new file mode 100644 index 0000000..f770e7e --- /dev/null +++ b/gcc/config/c6x/uclinux-elf.h @@ -0,0 +1,64 @@ +/* Definitions for TI C6X running ucLinux using ELF + Copyright (C) 2010, 2011 Free Software Foundation, Inc. + Contributed by Andrew Jenner <andrew@codesourcery.com> + Contributed by Bernd Schmidt <bernds@codesourcery.com> + + This file is part of GCC. + + GCC is free software; you can redistribute it and/or modify it + under the terms of the GNU General Public License as published + by the Free Software Foundation; either version 3, or (at your + option) any later version. + + GCC is distributed in the hope that it will be useful, but WITHOUT + ANY WARRANTY; without even the implied warranty of MERCHANTABILITY + or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public + License for more details. + + You should have received a copy of the GNU General Public License + along with GCC; see the file COPYING3. If not see + <http://www.gnu.org/licenses/>. */ + +#undef TARGET_OS_CPP_BUILTINS +#define TARGET_OS_CPP_BUILTINS() \ + do \ + { \ + builtin_define ("__uClinux__"); \ + builtin_define_std ("linux"); \ + builtin_define_std ("unix"); \ + builtin_assert ("system=linux"); \ + builtin_assert ("system=unix"); \ + builtin_assert ("system=posix"); \ + } \ + while (false) + +#undef STARTFILE_SPEC +#define STARTFILE_SPEC \ + "%{!shared:crt1%O%s} crti%O%s %{shared|pie:crtbeginS.o%s;:crtbegin.o%s}" + +#define UCLIBC_DYNAMIC_LINKER "/lib/ld-uClibc.so.0" + +#undef LINK_SPEC +#define LINK_SPEC ENDIAN_LINK_SPEC \ + "%{shared} %{fpie|fPIE:-pie} \ + %{!shared: %{!static: \ + %{rdynamic:-export-dynamic} \ + %{!dynamic-linker:-dynamic-linker " UCLIBC_DYNAMIC_LINKER "}} \ + %{static}}" + +#undef DRIVER_SELF_SPECS +#define DRIVER_SELF_SPECS "%{!mno-dsbt:-mdsbt}" + +/* Clear the instruction cache from `beg' to `end'. This makes an + inline system call to SYS_cacheflush. */ +#undef CLEAR_INSN_CACHE +#define CLEAR_INSN_CACHE(BEG, END) \ +{ \ + register unsigned long _beg __asm ("A4") = (unsigned long) (BEG); \ + register unsigned long _end __asm ("B4") = (unsigned long) (END); \ + register unsigned long _scno __asm ("B0") = 244; \ + __asm __volatile ("swe ; sys_cache_sync" \ + : "=a" (_beg) \ + : "0" (_beg), "b" (_end), "b" (_scno)); \ +} + |