aboutsummaryrefslogtreecommitdiff
path: root/sysdeps/alpha
diff options
context:
space:
mode:
Diffstat (limited to 'sysdeps/alpha')
-rw-r--r--sysdeps/alpha/Makefile14
-rw-r--r--sysdeps/alpha/bzero.S113
-rw-r--r--sysdeps/alpha/memset.S130
-rw-r--r--sysdeps/alpha/stpcpy.S49
-rw-r--r--sysdeps/alpha/stpncpy.S103
-rw-r--r--sysdeps/alpha/strcat.S66
-rw-r--r--sysdeps/alpha/strchr.S88
-rw-r--r--sysdeps/alpha/strcpy.S36
-rw-r--r--sysdeps/alpha/strncat.S90
-rw-r--r--sysdeps/alpha/strncpy.S85
-rw-r--r--sysdeps/alpha/strrchr.S104
-rw-r--r--sysdeps/alpha/stxcpy.S307
-rw-r--r--sysdeps/alpha/stxncpy.S350
13 files changed, 1530 insertions, 5 deletions
diff --git a/sysdeps/alpha/Makefile b/sysdeps/alpha/Makefile
index 6aaedea..45babb6 100644
--- a/sysdeps/alpha/Makefile
+++ b/sysdeps/alpha/Makefile
@@ -21,17 +21,21 @@ sysdep_routines += _mcount
endif
ifeq ($(subdir),setjmp)
-sysdep_routines := $(sysdep_routines) setjmp_aux
+sysdep_routines += setjmp_aux
endif
ifeq ($(subdir),gnulib)
-routines = $(divrem)
-endif # gnulib
+sysdep_routines += $(divrem)
+endif
+
+ifeq ($(subdir),string)
+sysdep_routines += stxcpy stxncpy
+endif
ifeq ($(subdir),elf)
-# The ld.so code cannot use literals until it self-relocates.
+# The ld.so startup code cannot use literals until it self-relocates.
ifeq ($(elf),yes)
-CFLAGS-rtld.c = -mbuild-constants
+ CFLAGS-rtld.c = -mbuild-constants
endif
# The rest of ld.so shouldn't use FP regs for block moves so
# that the lazy link trampoline doesn't have to save them.
diff --git a/sysdeps/alpha/bzero.S b/sysdeps/alpha/bzero.S
new file mode 100644
index 0000000..fffa53d
--- /dev/null
+++ b/sysdeps/alpha/bzero.S
@@ -0,0 +1,113 @@
+/* Copyright (C) 1996 Free Software Foundation, Inc.
+ Contributed by Richard Henderson (rth@tamu.edu)
+
+This file is part of the GNU C Library.
+
+The GNU C Library is free software; you can redistribute it and/or
+modify it under the terms of the GNU Library General Public License as
+published by the Free Software Foundation; either version 2 of the
+License, or (at your option) any later version.
+
+The GNU C Library 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
+Library General Public License for more details.
+
+You should have received a copy of the GNU Library General Public
+License along with the GNU C Library; see the file COPYING.LIB. If
+not, write to the Free Software Foundation, Inc., 675 Mass Ave,
+Cambridge, MA 02139, USA. */
+
+/* Fill a block of memory with zeros. Optimized for the Alpha architecture:
+
+ - memory accessed as aligned quadwords only
+ - destination memory not read unless needed for good cache behaviour
+ - basic blocks arranged to optimize branch prediction for full-quadword
+ aligned memory blocks.
+ - partial head and tail quadwords constructed with byte-mask instructions
+
+ This is generally scheduled for the EV5 (got to look out for my own
+ interests :-), but with EV4 needs in mind. There *should* be no more
+ stalls for the EV4 than there are for the EV5.
+*/
+
+
+#include <sysdep.h>
+
+ .set noat
+ .set noreorder
+
+ .text
+
+/* There is a problem with either gdb (as of 4.16) or gas (as of 2.7) that
+ doesn't like putting the entry point for a procedure somewhere in the
+ middle of the procedure descriptor. Work around this by putting the main
+ loop in its own procedure descriptor. */
+
+ /* On entry to this basic block:
+ t3 == loop counter
+ t4 == bytes in partial final word
+ a0 == possibly misaligned destination pointer */
+
+ .ent bzero_loop
+ .align 3
+bzero_loop:
+ .frame sp, 0, ra, 0
+ .prologue 0
+
+ beq t3, $tail #
+ blbc t3, 0f # skip single store if count even
+
+ stq_u zero, 0(a0) # e0 : store one word
+ subq t3, 1, t3 # .. e1 :
+ addq a0, 8, a0 # e0 :
+ beq t3, $tail # .. e1 :
+
+0: stq_u zero, 0(a0) # e0 : store two words
+ subq t3, 2, t3 # .. e1 :
+ stq_u zero, 8(a0) # e0 :
+ addq a0, 16, a0 # .. e1 :
+ bne t3, 0b # e1 :
+
+$tail: bne t4, 1f # is there a tail to do?
+ ret # no
+
+1: ldq_u t0, 0(a0) # yes, load original data
+ mskqh t0, t4, t0 #
+ stq_u t0, 0(a0) #
+ ret #
+
+ .end bzero_loop
+
+ENTRY(bzero)
+ .prologue 0
+
+ mov a0, v0 # e0 : move return value in place
+ beq a1, $done # .. e1 : early exit for zero-length store
+ and a0, 7, t1 # e0 :
+ addq a1, t1, a1 # e1 : add dest misalignment to count
+ srl a1, 3, t3 # e0 : loop = count >> 3
+ and a1, 7, t4 # .. e1 : find number of bytes in tail
+ unop # :
+ beq t1, bzero_loop # e1 : aligned head, jump right in
+
+ ldq_u t0, 0(a0) # e0 : load original data to mask into
+ cmpult a1, 8, t2 # .. e1 : is this a sub-word set?
+ bne t2, $oneq # e1 :
+
+ mskql t0, a0, t0 # e0 : we span words. finish this partial
+ subq t3, 1, t3 # e0 :
+ addq a0, 8, a0 # .. e1 :
+ stq_u t0, -8(a0) # e0 :
+ br bzero_loop # .. e1 :
+
+ .align 3
+$oneq:
+ mskql t0, a0, t2 # e0 :
+ mskqh t0, a1, t3 # e0 :
+ or t2, t3, t0 # e1 :
+ stq_u t0, 0(a0) # e0 :
+
+$done: ret
+
+ END(bzero)
diff --git a/sysdeps/alpha/memset.S b/sysdeps/alpha/memset.S
new file mode 100644
index 0000000..55271f0
--- /dev/null
+++ b/sysdeps/alpha/memset.S
@@ -0,0 +1,130 @@
+/* Copyright (C) 1996 Free Software Foundation, Inc.
+ Contributed by Richard Henderson (rth@tamu.edu)
+
+This file is part of the GNU C Library.
+
+The GNU C Library is free software; you can redistribute it and/or
+modify it under the terms of the GNU Library General Public License as
+published by the Free Software Foundation; either version 2 of the
+License, or (at your option) any later version.
+
+The GNU C Library 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
+Library General Public License for more details.
+
+You should have received a copy of the GNU Library General Public
+License along with the GNU C Library; see the file COPYING.LIB. If
+not, write to the Free Software Foundation, Inc., 675 Mass Ave,
+Cambridge, MA 02139, USA. */
+
+/* Fill a block of memory with a character. Optimized for the Alpha
+ architecture:
+
+ - memory accessed as aligned quadwords only
+ - destination memory not read unless needed for good cache behaviour
+ - basic blocks arranged to optimize branch prediction for full-quadword
+ aligned memory blocks.
+ - partial head and tail quadwords constructed with byte-mask instructions
+
+ This is generally scheduled for the EV5 (got to look out for my own
+ interests :-), but with EV4 needs in mind. There *should* be no more
+ stalls for the EV4 than there are for the EV5.
+*/
+
+
+#include <sysdep.h>
+
+ .set noat
+ .set noreorder
+
+ .text
+
+/* There is a problem with either gdb (as of 4.16) or gas (as of 2.7) that
+ doesn't like putting the entry point for a procedure somewhere in the
+ middle of the procedure descriptor. Work around this by putting the main
+ loop in its own procedure descriptor. */
+
+ /* On entry to this basic block:
+ t3 == loop counter
+ t4 == bytes in partial final word
+ a0 == possibly misaligned destination pointer
+ a1 == replicated source character */
+
+ .ent memset_loop
+ .align 3
+memset_loop:
+ .frame sp, 0, ra, 0
+ .prologue 0
+
+ beq t3, $tail
+ blbc t3, 0f # skip single store if count even
+
+ stq_u a1, 0(a0) # e0 : store one word
+ subq t3, 1, t3 # .. e1 :
+ addq a0, 8, a0 # e0 :
+ beq t3, $tail # .. e1 :
+
+0: stq_u a1, 0(a0) # e0 : store two words
+ subq t3, 2, t3 # .. e1 :
+ stq_u a1, 8(a0) # e0 :
+ addq a0, 16, a0 # .. e1 :
+ bne t3, 0b # e1 :
+
+$tail: bne t4, 1f # is there a tail to do?
+ ret # no
+
+ .align 3
+1: ldq_u t0, 0(a0) # e1 : yes, load original data
+ mskql a1, t4, t1 # .. e0 :
+ mskqh t0, t4, t0 # e0 :
+ or t0, t1, t0 # e1 (stall)
+ stq_u t0, 0(a0) # e0 :
+ ret # .. e1 :
+
+ .end memset_loop
+
+ENTRY(memset)
+ .prologue 0
+
+ zapnot a1, 1, a1 # e0 : zero extend input character
+ mov a0, v0 # .. e1 : move return value in place
+ sll a1, 8, t0 # e0 : begin replicating the char
+ beq a2, $done # .. e1 : early exit for zero-length store
+ or t0, a1, a1 # e0 :
+ and a0, 7, t1 # .. e1 : dest misalignment
+ sll a1, 16, t0 # e0 :
+ addq a2, t1, a2 # .. e1 : add dest misalignment to count
+ or t0, a1, a1 # e0 :
+ srl a2, 3, t3 # .. e1 : loop = count >> 3
+ sll a1, 32, t0 # e0 :
+ and a2, 7, t4 # .. e1 : find number of bytes in tail
+ or t0, a1, a1 # e0 : character replication done
+
+ beq t1, memset_loop # .. e1 : aligned head, jump right in
+
+ ldq_u t0, 0(a0) # e1 : load original data to mask into
+ mskqh a1, a0, t1 # .. e0 :
+
+ cmpult a2, 8, t2 # e0 : is this a sub-word set?
+ bne t2, $oneq # .. e1 (zdb)
+
+ mskql t0, a0, t0 # e0 : we span words. finish this partial
+ subq t3, 1, t3 # .. e1 :
+ addq a0, 8, a0 # e0 :
+ or t0, t1, t0 # .. e1 :
+ stq_u t0, -8(a0) # e0 :
+ br memset_loop # .. e1 :
+
+ .align 3
+$oneq:
+ mskql t1, a2, t1 # e0 : entire operation within one word
+ mskql t0, a0, t2 # e0 :
+ mskqh t0, a2, t3 # e0 :
+ or t1, t2, t0 # .. e1 :
+ or t0, t3, t0 # e1 :
+ stq_u t0, 0(a0) # e0 (stall)
+
+$done: ret
+
+ END(memset)
diff --git a/sysdeps/alpha/stpcpy.S b/sysdeps/alpha/stpcpy.S
new file mode 100644
index 0000000..0dc44d3
--- /dev/null
+++ b/sysdeps/alpha/stpcpy.S
@@ -0,0 +1,49 @@
+/* Copyright (C) 1996 Free Software Foundation, Inc.
+ Contributed by Richard Henderson (rth@tamu.edu)
+
+This file is part of the GNU C Library.
+
+The GNU C Library is free software; you can redistribute it and/or
+modify it under the terms of the GNU Library General Public License as
+published by the Free Software Foundation; either version 2 of the
+License, or (at your option) any later version.
+
+The GNU C Library 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
+Library General Public License for more details.
+
+You should have received a copy of the GNU Library General Public
+License along with the GNU C Library; see the file COPYING.LIB. If
+not, write to the Free Software Foundation, Inc., 675 Mass Ave,
+Cambridge, MA 02139, USA. */
+
+/* Copy a null-terminated string from SRC to DST. Return a pointer
+ to the null-terminator in the source. */
+
+#include <sysdep.h>
+
+ .text
+
+ENTRY(__stpcpy)
+ ldgp gp, 0(pv)
+ .prologue 1
+
+ jsr t9, __stxcpy # do the work of the copy
+
+ and t8, 0xf0, t2 # binary search for byte offset of the
+ and t8, 0xcc, t1 # last byte written.
+ and t8, 0xaa, t0
+ andnot a0, 7, a0
+ cmovne t2, 4, t2
+ cmovne t1, 2, t1
+ cmovne t0, 1, t0
+ addq a0, t2, v0
+ addq t0, t1, t0
+ addq v0, t0, v0
+
+ ret
+
+ END(__stpcpy)
+
+weak_alias (__stpcpy, stpcpy)
diff --git a/sysdeps/alpha/stpncpy.S b/sysdeps/alpha/stpncpy.S
new file mode 100644
index 0000000..50cda26
--- /dev/null
+++ b/sysdeps/alpha/stpncpy.S
@@ -0,0 +1,103 @@
+/* Copyright (C) 1996 Free Software Foundation, Inc.
+ Contributed by Richard Henderson (rth@tamu.edu)
+
+This file is part of the GNU C Library.
+
+The GNU C Library is free software; you can redistribute it and/or
+modify it under the terms of the GNU Library General Public License as
+published by the Free Software Foundation; either version 2 of the
+License, or (at your option) any later version.
+
+The GNU C Library 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
+Library General Public License for more details.
+
+You should have received a copy of the GNU Library General Public
+License along with the GNU C Library; see the file COPYING.LIB. If
+not, write to the Free Software Foundation, Inc., 675 Mass Ave,
+Cambridge, MA 02139, USA. */
+
+/* Copy no more than COUNT bytes of the null-terminated string from
+ SRC to DST. If SRC does not cover all of COUNT, the balance is
+ zeroed. Return the address of the terminating null in DEST, if
+ any, else DEST + COUNT. */
+
+#include <sysdep.h>
+
+ .set noat
+ .set noreorder
+
+ .text
+
+ENTRY(__stpncpy)
+ ldgp gp, 0(pv)
+ .prologue 1
+
+ beq a2, $zerocount
+ jsr t9, __stxncpy # do the work of the copy
+
+ and t8, 0xf0, t3 # binary search for byte offset of the
+ and t8, 0xcc, t2 # last byte written.
+ and t8, 0xaa, t1
+ andnot a0, 7, v0
+ cmovne t3, 4, t3
+ cmovne t2, 2, t2
+ cmovne t1, 1, t1
+ addq v0, t3, v0
+ addq t1, t2, t1
+ addq v0, t1, v0
+
+ bne a2, $multiword # do we have full words left?
+
+ .align 3
+ zapnot t0, t8, t4 # e0 : was last byte a null?
+ subq t8, 1, t2 # .. e1 :
+ addq v0, 1, t5 # e0 :
+ subq t10, 1, t3 # .. e1 :
+ or t2, t8, t2 # e0 : clear the bits between the last
+ or t3, t10, t3 # .. e1 : written byte and the last byte in
+ andnot t3, t2, t3 # e0 : COUNT
+ cmovne t4, t5, v0 # .. e1 : if last written wasnt null, inc v0
+ zap t0, t3, t0 # e0 :
+ stq_u t0, 0(a0) # e1 :
+ ret # .. e1 :
+
+ .align 3
+$multiword:
+ subq t8, 1, t7 # e0 : clear the final bits in the prev
+ or t7, t8, t7 # e1 : word
+ zapnot t0, t7, t0 # e0 :
+ subq a2, 1, a2 # .. e1 :
+ stq_u t0, 0(a0) # e0 :
+ addq a0, 8, a0 # .. e1 :
+
+ beq a2, 1f # e1 :
+ blbc a2, 0f # e1 :
+
+ stq_u zero, 0(a0) # e0 : zero one word
+ subq a2, 1, a2 # .. e1 :
+ addq a0, 8, a0 # e0 :
+ beq a2, 1f # .. e1 :
+
+0: stq_u zero, 0(a0) # e0 : zero two words
+ subq a2, 2, a2 # .. e1 :
+ stq_u zero, 8(a0) # e0 :
+ addq a0, 16, a0 # .. e1 :
+ bne a2, 0b # e1 :
+ unop
+
+1: ldq_u t0, 0(a0) # e0 : clear the leading bits in the final
+ subq t10, 1, t7 # .. e1 : word
+ or t7, t10, t7 # e0 :
+ zap t0, t7, t0 # e1 (stall)
+ stq_u t0, 0(a0) # e0 :
+ ret # .. e1 :
+
+$zerocount:
+ mov a0, v0
+ ret
+
+ END(__stpncpy)
+
+weak_alias (__stpncpy, stpncpy)
diff --git a/sysdeps/alpha/strcat.S b/sysdeps/alpha/strcat.S
new file mode 100644
index 0000000..d3afff3
--- /dev/null
+++ b/sysdeps/alpha/strcat.S
@@ -0,0 +1,66 @@
+/* Copyright (C) 1996 Free Software Foundation, Inc.
+ Contributed by Richard Henderson (rth@tamu.edu)
+
+This file is part of the GNU C Library.
+
+The GNU C Library is free software; you can redistribute it and/or
+modify it under the terms of the GNU Library General Public License as
+published by the Free Software Foundation; either version 2 of the
+License, or (at your option) any later version.
+
+The GNU C Library 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
+Library General Public License for more details.
+
+You should have received a copy of the GNU Library General Public
+License along with the GNU C Library; see the file COPYING.LIB. If
+not, write to the Free Software Foundation, Inc., 675 Mass Ave,
+Cambridge, MA 02139, USA. */
+
+/* Append a null-terminated string from SRC to DST. */
+
+#include <sysdep.h>
+
+ .text
+
+ENTRY(strcat)
+ ldgp gp, 0(pv)
+ .prologue 1
+
+ mov a0, v0 # set up return value
+
+ /* Find the end of the string. */
+
+ ldq_u t0, 0(a0) # load first quadword (a0 may be misaligned)
+ lda t1, -1(zero)
+ insqh t1, a0, t1
+ andnot a0, 7, a0
+ or t1, t0, t0
+ cmpbge zero, t0, t1 # t1 <- bitmask: bit i == 1 <==> i-th byte == 0
+ bne t1, $found
+
+$loop: ldq t0, 8(a0)
+ addq a0, 8, a0 # addr += 8
+ cmpbge zero, t0, t1
+ beq t1, $loop
+
+$found: negq t1, t2 # clear all but least set bit
+ and t1, t2, t1
+
+ and t1, 0xf0, t2 # binary search for that set bit
+ and t1, 0xcc, t3
+ and t1, 0xaa, t4
+ cmovne t2, 4, t2
+ cmovne t3, 2, t3
+ cmovne t4, 1, t4
+ addq t2, t3, t2
+ addq a0, t4, a0
+ addq a0, t2, a0
+
+ /* Now do the append. */
+
+ jsr t9, __stxcpy
+ ret
+
+ END(strcat)
diff --git a/sysdeps/alpha/strchr.S b/sysdeps/alpha/strchr.S
new file mode 100644
index 0000000..c26a843
--- /dev/null
+++ b/sysdeps/alpha/strchr.S
@@ -0,0 +1,88 @@
+/* Copyright (C) 1996 Free Software Foundation, Inc.
+ Contributed by Richard Henderson (rth@tamu.edu)
+
+This file is part of the GNU C Library.
+
+The GNU C Library is free software; you can redistribute it and/or
+modify it under the terms of the GNU Library General Public License as
+published by the Free Software Foundation; either version 2 of the
+License, or (at your option) any later version.
+
+The GNU C Library 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
+Library General Public License for more details.
+
+You should have received a copy of the GNU Library General Public
+License along with the GNU C Library; see the file COPYING.LIB. If
+not, write to the Free Software Foundation, Inc., 675 Mass Ave,
+Cambridge, MA 02139, USA. */
+
+/* Return the address of a given character within a null-terminated
+ string, or null if it is not found.
+
+ This is generally scheduled for the EV5 (got to look out for my own
+ interests :-), but with EV4 needs in mind. There *should* be no more
+ stalls for the EV4 than there are for the EV5.
+*/
+
+#include <sysdep.h>
+
+ .set noreorder
+ .set noat
+
+ENTRY(strchr)
+ .prologue 0
+
+ zapnot a1, 1, a1 # e0 : zero extend the search character
+ ldq_u t0, 0(a0) # .. e1 : load first quadword
+ sll a1, 8, t5 # e0 : replicate the search character
+ andnot a0, 7, v0 # .. e1 : align our loop pointer
+ or t5, a1, a1 # e0 :
+ lda t4, -1 # .. e1 : build garbage mask
+ sll a1, 16, t5 # e0 :
+ cmpbge zero, t0, t2 # .. e1 : bits set iff byte == zero
+ mskqh t4, a0, t4 # e0 :
+ or t5, a1, a1 # .. e1 :
+ sll a1, 32, t5 # e0 :
+ cmpbge zero, t4, t4 # .. e1 : bits set iff byte is garbage
+ or t5, a1, a1 # e0 :
+ xor t0, a1, t1 # .. e1 : make bytes == c zero
+ cmpbge zero, t1, t3 # e0 : bits set iff byte == c
+ or t2, t3, t0 # e1 : bits set iff char match or zero match
+ andnot t0, t4, t0 # e0 : clear garbage bits
+ bne t0, $found # .. e1 (zdb)
+
+$loop: ldq t0, 8(v0) # e0 :
+ addq v0, 8, v0 # .. e1 :
+ nop # e0 :
+ xor t0, a1, t1 # .. e1 (ev5 data stall)
+ cmpbge zero, t0, t2 # e0 : bits set iff byte == 0
+ cmpbge zero, t1, t3 # .. e1 : bits set iff byte == c
+ or t2, t3, t0 # e0 :
+ beq t0, $loop # .. e1 (zdb)
+
+$found: negq t0, t1 # e0 : clear all but least set bit
+ and t0, t1, t0 # e1 (stall)
+
+ and t0, t3, t1 # e0 : bit set iff byte was the char
+ beq t1, $retnull # .. e1 (zdb)
+
+ and t0, 0xf0, t2 # e0 : binary search for that set bit
+ and t0, 0xcc, t3 # .. e1 :
+ and t0, 0xaa, t4 # e0 :
+ cmovne t2, 4, t2 # .. e1 :
+ cmovne t3, 2, t3 # e0 :
+ cmovne t4, 1, t4 # .. e1 :
+ addq t2, t3, t2 # e0 :
+ addq v0, t4, v0 # .. e1 :
+ addq v0, t2, v0 # e0 :
+ ret # .. e1 :
+
+$retnull:
+ mov zero, v0 # e0 :
+ ret # .. e1 :
+
+ END(strchr)
+
+weak_alias (strchr, index)
diff --git a/sysdeps/alpha/strcpy.S b/sysdeps/alpha/strcpy.S
new file mode 100644
index 0000000..2975181
--- /dev/null
+++ b/sysdeps/alpha/strcpy.S
@@ -0,0 +1,36 @@
+/* Copyright (C) 1996 Free Software Foundation, Inc.
+ Contributed by Richard Henderson (rth@tamu.edu)
+
+This file is part of the GNU C Library.
+
+The GNU C Library is free software; you can redistribute it and/or
+modify it under the terms of the GNU Library General Public License as
+published by the Free Software Foundation; either version 2 of the
+License, or (at your option) any later version.
+
+The GNU C Library 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
+Library General Public License for more details.
+
+You should have received a copy of the GNU Library General Public
+License along with the GNU C Library; see the file COPYING.LIB. If
+not, write to the Free Software Foundation, Inc., 675 Mass Ave,
+Cambridge, MA 02139, USA. */
+
+/* Copy a null-terminated string from SRC to DST. Return a pointer
+ to the null-terminator in the source. */
+
+#include <sysdep.h>
+
+ .text
+
+ENTRY(strcpy)
+ ldgp gp, 0(pv)
+ .prologue 1
+
+ mov a0, v0 # set up return value
+ jsr t9, __stxcpy # do the copy
+ ret
+
+ END(strcpy)
diff --git a/sysdeps/alpha/strncat.S b/sysdeps/alpha/strncat.S
new file mode 100644
index 0000000..d502037
--- /dev/null
+++ b/sysdeps/alpha/strncat.S
@@ -0,0 +1,90 @@
+/* Copyright (C) 1996 Free Software Foundation, Inc.
+ Contributed by Richard Henderson (rth@tamu.edu)
+
+This file is part of the GNU C Library.
+
+The GNU C Library is free software; you can redistribute it and/or
+modify it under the terms of the GNU Library General Public License as
+published by the Free Software Foundation; either version 2 of the
+License, or (at your option) any later version.
+
+The GNU C Library 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
+Library General Public License for more details.
+
+You should have received a copy of the GNU Library General Public
+License along with the GNU C Library; see the file COPYING.LIB. If
+not, write to the Free Software Foundation, Inc., 675 Mass Ave,
+Cambridge, MA 02139, USA. */
+
+/* Append no more than COUNT characters from the null-terminated string SRC
+ to the null-terminated string DST. Always null-terminate the new DST. */
+
+#include <sysdep.h>
+
+ .text
+
+ENTRY(strncat)
+ ldgp gp, 0(pv)
+ .prologue 1
+
+ mov a0, v0 # set up return value
+ beq a2, $zerocount
+
+ /* Find the end of the string. */
+
+ ldq_u t0, 0(a0) # load first quadword (a0 may be misaligned)
+ lda t1, -1(zero)
+ insqh t1, a0, t1
+ andnot a0, 7, a0
+ or t1, t0, t0
+ cmpbge zero, t0, t1 # t1 <- bitmask: bit i == 1 <==> i-th byte == 0
+ bne t1, $found
+
+$loop: ldq t0, 8(a0)
+ addq a0, 8, a0 # addr += 8
+ cmpbge zero, t0, t1
+ beq t1, $loop
+
+$found: negq t1, t2 # clear all but least set bit
+ and t1, t2, t1
+
+ and t1, 0xf0, t2 # binary search for that set bit
+ and t1, 0xcc, t3
+ and t1, 0xaa, t4
+ cmovne t2, 4, t2
+ cmovne t3, 2, t3
+ cmovne t4, 1, t4
+ addq t2, t3, t2
+ addq a0, t4, a0
+ addq a0, t2, a0
+
+ /* Now do the append. */
+
+ jsr t9, __stxncpy
+
+ /* Worry about the null termination. */
+
+ zapnot t0, t8, t1 # was last byte a null?
+ bne t1, 0f
+ ret
+
+0: and t10, 0x80, t1
+ bne t1, 1f
+
+ /* Here there are bytes left in the current word. Clear one. */
+ addq t10, t10, t10 # end-of-count bit <<= 1
+ zap t0, t10, t0
+ stq_u t0, 0(a0)
+ ret
+
+1: /* Here we must read the next DST word and clear the first byte. */
+ ldq_u t0, 8(a0)
+ zap t0, 1, t0
+ stq_u t0, 8(a0)
+
+$zerocount:
+ ret
+
+ END(strncat)
diff --git a/sysdeps/alpha/strncpy.S b/sysdeps/alpha/strncpy.S
new file mode 100644
index 0000000..e13769c
--- /dev/null
+++ b/sysdeps/alpha/strncpy.S
@@ -0,0 +1,85 @@
+/* Copyright (C) 1996 Free Software Foundation, Inc.
+ Contributed by Richard Henderson (rth@tamu.edu)
+
+This file is part of the GNU C Library.
+
+The GNU C Library is free software; you can redistribute it and/or
+modify it under the terms of the GNU Library General Public License as
+published by the Free Software Foundation; either version 2 of the
+License, or (at your option) any later version.
+
+The GNU C Library 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
+Library General Public License for more details.
+
+You should have received a copy of the GNU Library General Public
+License along with the GNU C Library; see the file COPYING.LIB. If
+not, write to the Free Software Foundation, Inc., 675 Mass Ave,
+Cambridge, MA 02139, USA. */
+
+/* Copy no more than COUNT bytes of the null-terminated string from
+ SRC to DST. If SRC does not cover all of COUNT, the balance is
+ zeroed. */
+
+#include <sysdep.h>
+
+ .set noat
+ .set noreorder
+
+ .text
+
+ENTRY(strncpy)
+ ldgp gp, 0(pv)
+ .prologue 1
+
+ mov a0, v0 # set return value now
+ beq a2, $zerocount
+ jsr t9, __stxncpy # do the work of the copy
+
+ bne a2, $multiword # do we have full words left?
+
+ .align 3
+ subq t8, 1, t2 # e0 : guess not
+ subq t10, 1, t3 # .. e1 :
+ or t2, t8, t2 # e0 : clear the bits between the last
+ or t3, t10, t3 # .. e1 : written byte and the last byte in
+ andnot t3, t2, t3 # e0 : COUNT
+ zap t0, t3, t0 # e1 :
+ stq_u t0, 0(a0) # e0 :
+ ret # .. e1 :
+
+$multiword:
+
+ subq t8, 1, t7 # e0 : clear the final bits in the prev
+ or t7, t8, t7 # e1 : word
+ zapnot t0, t7, t0 # e0 :
+ subq a2, 1, a2 # .. e1 :
+ stq_u t0, 0(a0) # e0 :
+ addq a0, 8, a0 # .. e1 :
+
+ beq a2, 1f # e1 :
+ blbc a2, 0f # e1 :
+
+ stq_u zero, 0(a0) # e0 : zero one word
+ subq a2, 1, a2 # .. e1 :
+ addq a0, 8, a0 # e0 :
+ beq a2, 1f # .. e1 :
+
+0: stq_u zero, 0(a0) # e0 : zero two words
+ subq a2, 2, a2 # .. e1 :
+ stq_u zero, 8(a0) # e0 :
+ addq a0, 16, a0 # .. e1 :
+ bne a2, 0b # e1 :
+ unop
+
+1: ldq_u t0, 0(a0) # e0 : clear the leading bits in the final
+ subq t10, 1, t7 # .. e1 : word
+ or t7, t10, t7 # e0 :
+ zap t0, t7, t0 # e1 (stall)
+ stq_u t0, 0(a0) # e0 :
+
+$zerocount:
+ ret # .. e1 :
+
+ END(strncpy)
diff --git a/sysdeps/alpha/strrchr.S b/sysdeps/alpha/strrchr.S
new file mode 100644
index 0000000..464f754
--- /dev/null
+++ b/sysdeps/alpha/strrchr.S
@@ -0,0 +1,104 @@
+/* Copyright (C) 1996 Free Software Foundation, Inc.
+
+This file is part of the GNU C Library.
+
+The GNU C Library is free software; you can redistribute it and/or
+modify it under the terms of the GNU Library General Public License as
+published by the Free Software Foundation; either version 2 of the
+License, or (at your option) any later version.
+
+The GNU C Library 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
+Library General Public License for more details.
+
+You should have received a copy of the GNU Library General Public
+License along with the GNU C Library; see the file COPYING.LIB. If
+not, write to the Free Software Foundation, Inc., 675 Mass Ave,
+Cambridge, MA 02139, USA. */
+
+/* Return the address of the last occurrance of a given character
+ within a null-terminated string, or null if it is not found.
+
+ This is generally scheduled for the EV5 (got to look out for my own
+ interests :-), but with EV4 needs in mind. There are, in fact, fewer
+ stalls on the EV4 than there are on the EV5.
+*/
+
+#include <sysdep.h>
+
+ .set noreorder
+ .set noat
+
+ENTRY(strrchr)
+ .prologue 0
+
+ zapnot a1, 1, a1 # e0 : zero extend our test character
+ mov zero, t6 # .. e1 : t6 is last match aligned addr
+ sll a1, 8, t5 # e0 : replicate our test character
+ mov zero, t7 # .. e1 : t7 is last match byte compare mask
+ or t5, a1, a1 # e0 :
+ ldq_u t0, 0(a0) # .. e1 : load first quadword
+ sll a1, 16, t5 # e0 :
+ andnot a0, 7, v0 # .. e1 : align source addr
+ or t5, a1, a1 # e0 :
+ lda t4, -1 # .. e1 : build garbage mask
+ sll a1, 32, t5 # e0 :
+ cmpbge zero, t0, t1 # .. e1 : bits set iff byte == zero
+ mskqh t4, a0, t4 # e0 :
+ or t5, a1, a1 # .. e1 : character replication complete
+ xor t0, a1, t2 # e0 : make bytes == c zero
+ cmpbge zero, t4, t4 # .. e1 : bits set iff byte is garbage
+ cmpbge zero, t2, t3 # e0 : bits set iff byte == c
+ andnot t1, t4, t1 # .. e1 : clear garbage from null test
+ andnot t3, t4, t3 # e0 : clear garbage from char test
+ bne t1, $eos # .. e1 : did we already hit the terminator?
+
+ /* Character search main loop */
+$loop:
+ ldq t0, 8(v0) # e0 : load next quadword
+ cmovne t3, v0, t6 # .. e1 : save previous comparisons match
+ cmovne t3, t3, t7 # e0 :
+ addq v0, 8, v0 # .. e1 :
+ xor t0, a1, t2 # e0 :
+ cmpbge zero, t0, t1 # .. e1 : bits set iff byte == zero
+ cmpbge zero, t2, t3 # e0 : bits set iff byte == c
+ beq t1, $loop # .. e1 : if we havnt seen a null, loop
+
+ /* Mask out character matches after terminator */
+$eos:
+ negq t1, t4 # e0 : isolate first null byte match
+ and t1, t4, t4 # e1 :
+ subq t4, 1, t5 # e0 : build a mask of the bytes upto...
+ or t4, t5, t4 # e1 : ... and including the null
+
+ and t3, t4, t3 # e0 : mask out char matches after null
+ cmovne t3, t3, t7 # .. e1 : save it, if match found
+ cmovne t3, v0, t6 # e0 :
+
+ /* Locate the address of the last matched character */
+
+ /* Retain the early exit for the ev4 -- the ev5 mispredict penalty
+ is 5 cycles -- the same as just falling through. */
+ beq t7, $retnull # .. e1 :
+
+ and t7, 0xf0, t2 # e0 : binary search for the high bit set
+ cmovne t2, t2, t7 # .. e1 (zdb)
+ cmovne t2, 4, t2 # e0 :
+ and t7, 0xcc, t1 # .. e1 :
+ cmovne t1, t1, t7 # e0 :
+ cmovne t1, 2, t1 # .. e1 :
+ and t7, 0xaa, t0 # e0 :
+ cmovne t0, 1, t0 # .. e1 (zdb)
+ addq t2, t1, t1 # e0 :
+ addq t6, t0, v0 # .. e1 : add our aligned base ptr to the mix
+ addq v0, t1, v0 # e0 :
+ ret # .. e1 :
+
+$retnull:
+ mov zero, v0 # e0 :
+ ret # .. e1 :
+
+ END(strrchr)
+
+weak_alias (strrchr, rindex)
diff --git a/sysdeps/alpha/stxcpy.S b/sysdeps/alpha/stxcpy.S
new file mode 100644
index 0000000..e381b70
--- /dev/null
+++ b/sysdeps/alpha/stxcpy.S
@@ -0,0 +1,307 @@
+/* Copyright (C) 1996 Free Software Foundation, Inc.
+ Contributed by Richard Henderson (rth@tamu.edu)
+
+This file is part of the GNU C Library.
+
+The GNU C Library is free software; you can redistribute it and/or
+modify it under the terms of the GNU Library General Public License as
+published by the Free Software Foundation; either version 2 of the
+License, or (at your option) any later version.
+
+The GNU C Library 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
+Library General Public License for more details.
+
+You should have received a copy of the GNU Library General Public
+License along with the GNU C Library; see the file COPYING.LIB. If
+not, write to the Free Software Foundation, Inc., 675 Mass Ave,
+Cambridge, MA 02139, USA. */
+
+/* Copy a null-terminated string from SRC to DST.
+
+ This is an internal routine used by strcpy, stpcpy, and strcat.
+ As such, it uses special linkage conventions to make implementation
+ of these public functions more efficient.
+
+ On input:
+ t9 = return address
+ a0 = DST
+ a1 = SRC
+
+ On output:
+ t8 = bitmask (with one bit set) indicating the last byte written
+ a0 = unaligned address of the last *word* written
+
+ Furthermore, v0, a3-a5, t11, and t12 are untouched.
+*/
+
+/* This is generally scheduled for the EV5, but should still be pretty
+ good for the EV4 too. */
+
+#include <sysdep.h>
+
+ .set noat
+ .set noreorder
+
+ .text
+
+/* There is a problem with either gdb (as of 4.16) or gas (as of 2.7) that
+ doesn't like putting the entry point for a procedure somewhere in the
+ middle of the procedure descriptor. Work around this by putting the
+ aligned copy in its own procedure descriptor */
+
+ .ent stxcpy_aligned
+ .align 3
+stxcpy_aligned:
+ .frame sp, 0, t9
+ .prologue 0
+
+ /* On entry to this basic block:
+ t0 == the first destination word for masking back in
+ t1 == the first source word. */
+
+ /* Create the 1st output word and detect 0's in the 1st input word. */
+ lda t2, -1 # e1 : build a mask against false zero
+ mskqh t2, a1, t2 # e0 : detection in the src word
+ mskqh t1, a1, t3 # e0 :
+ ornot t1, t2, t2 # .. e1 :
+ mskql t0, a1, t0 # e0 : assemble the first output word
+ cmpbge zero, t2, t7 # .. e1 : bits set iff null found
+ or t0, t3, t1 # e0 :
+ bne t7, $a_eos # .. e1 :
+
+ /* On entry to this basic block:
+ t0 == the first destination word for masking back in
+ t1 == a source word not containing a null. */
+
+$a_loop:
+ stq_u t1, 0(a0) # e0 :
+ addq a0, 8, a0 # .. e1 :
+ ldq_u t1, 0(a1) # e0 :
+ addq a1, 8, a1 # .. e1 :
+ cmpbge zero, t1, t7 # e0 (stall)
+ beq t7, $a_loop # .. e1 (zdb)
+
+ /* Take care of the final (partial) word store.
+ On entry to this basic block we have:
+ t1 == the source word containing the null
+ t7 == the cmpbge mask that found it. */
+$a_eos:
+ negq t7, t6 # e0 : find low bit set
+ and t7, t6, t8 # e1 (stall)
+
+ /* For the sake of the cache, don't read a destination word
+ if we're not going to need it. */
+ and t8, 0x80, t6 # e0 :
+ bne t6, 1f # .. e1 (zdb)
+
+ /* We're doing a partial word store and so need to combine
+ our source and original destination words. */
+ ldq_u t0, 0(a0) # e0 :
+ subq t8, 1, t6 # .. e1 :
+ zapnot t1, t6, t1 # e0 : clear src bytes >= null
+ or t8, t6, t7 # .. e1 :
+ zap t0, t7, t0 # e0 : clear dst bytes <= null
+ or t0, t1, t1 # e1 :
+
+1: stq_u t1, 0(a0) # e0 :
+ ret (t9) # .. e1 :
+
+ .end stxcpy_aligned
+
+ .align 3
+ .ent __stxcpy
+ .globl __stxcpy
+__stxcpy:
+ .frame sp, 0, t9
+ .prologue 0
+
+ /* Are source and destination co-aligned? */
+ xor a0, a1, t0 # e0 :
+ unop # :
+ and t0, 7, t0 # e0 :
+ bne t0, $unaligned # .. e1 :
+
+ /* We are co-aligned; take care of a partial first word. */
+ ldq_u t1, 0(a1) # e0 : load first src word
+ and a0, 7, t0 # .. e1 : take care not to load a word ...
+ addq a1, 8, a1 # e0 :
+ beq t0, stxcpy_aligned # .. e1 : ... if we wont need it
+ ldq_u t0, 0(a0) # e0 :
+ br stxcpy_aligned # .. e1 :
+
+
+/* The source and destination are not co-aligned. Align the destination
+ and cope. We have to be very careful about not reading too much and
+ causing a SEGV. */
+
+ .align 3
+$u_head:
+ /* We know just enough now to be able to assemble the first
+ full source word. We can still find a zero at the end of it
+ that prevents us from outputting the whole thing.
+
+ On entry to this basic block:
+ t0 == the first dest word, for masking back in, if needed else 0
+ t1 == the low bits of the first source word
+ t6 == bytemask that is -1 in dest word bytes */
+
+ ldq_u t2, 8(a1) # e0 :
+ addq a1, 8, a1 # .. e1 :
+
+ extql t1, a1, t1 # e0 :
+ extqh t2, a1, t4 # e0 :
+ mskql t0, a0, t0 # e0 :
+ or t1, t4, t1 # .. e1 :
+ mskqh t1, a0, t1 # e0 :
+ or t0, t1, t1 # e1 :
+
+ or t1, t6, t6 # e0 :
+ cmpbge zero, t6, t7 # .. e1 :
+ lda t6, -1 # e0 : for masking just below
+ bne t7, $u_final # .. e1 :
+
+ mskql t6, a1, t6 # e0 : mask out the bits we have
+ or t6, t2, t2 # e1 : already extracted before
+ cmpbge zero, t2, t7 # e0 : testing eos
+ bne t7, $u_late_head_exit # .. e1 (zdb)
+
+ /* Finally, we've got all the stupid leading edge cases taken care
+ of and we can set up to enter the main loop. */
+
+ stq_u t1, 0(a0) # e0 : store first output word
+ addq a0, 8, a0 # .. e1 :
+ extql t2, a1, t0 # e0 : position ho-bits of lo word
+ ldq_u t2, 8(a1) # .. e1 : read next high-order source word
+ addq a1, 8, a1 # e0 :
+ cmpbge zero, t2, t7 # .. e1 :
+ nop # e0 :
+ bne t7, $u_eos # .. e1 :
+
+ /* Unaligned copy main loop. In order to avoid reading too much,
+ the loop is structured to detect zeros in aligned source words.
+ This has, unfortunately, effectively pulled half of a loop
+ iteration out into the head and half into the tail, but it does
+ prevent nastiness from accumulating in the very thing we want
+ to run as fast as possible.
+
+ On entry to this basic block:
+ t0 == the shifted high-order bits from the previous source word
+ t2 == the unshifted current source word
+
+ We further know that t2 does not contain a null terminator. */
+
+ .align 3
+$u_loop:
+ extqh t2, a1, t1 # e0 : extract high bits for current word
+ addq a1, 8, a1 # .. e1 :
+ extql t2, a1, t3 # e0 : extract low bits for next time
+ addq a0, 8, a0 # .. e1 :
+ or t0, t1, t1 # e0 : current dst word now complete
+ ldq_u t2, 0(a1) # .. e1 : load high word for next time
+ stq_u t1, -8(a0) # e0 : save the current word
+ mov t3, t0 # .. e1 :
+ cmpbge zero, t2, t7 # e0 : test new word for eos
+ beq t7, $u_loop # .. e1 :
+
+ /* We've found a zero somewhere in the source word we just read.
+ If it resides in the lower half, we have one (probably partial)
+ word to write out, and if it resides in the upper half, we
+ have one full and one partial word left to write out.
+
+ On entry to this basic block:
+ t0 == the shifted high-order bits from the previous source word
+ t2 == the unshifted current source word. */
+$u_eos:
+ extqh t2, a1, t1 # e0 :
+ or t0, t1, t1 # e1 : first (partial) source word complete
+
+ cmpbge zero, t1, t7 # e0 : is the null in this first bit?
+ bne t7, $u_final # .. e1 (zdb)
+
+$u_late_head_exit:
+ stq_u t1, 0(a0) # e0 : the null was in the high-order bits
+ addq a0, 8, a0 # .. e1 :
+ extql t2, a1, t1 # e0 :
+ cmpbge zero, t1, t7 # .. e1 :
+
+ /* Take care of a final (probably partial) result word.
+ On entry to this basic block:
+ t1 == assembled source word
+ t7 == cmpbge mask that found the null. */
+$u_final:
+ negq t7, t6 # e0 : isolate low bit set
+ and t6, t7, t8 # e1 :
+
+ and t8, 0x80, t6 # e0 : avoid dest word load if we can
+ bne t6, 1f # .. e1 (zdb)
+
+ ldq_u t0, 0(a0) # e0 :
+ subq t8, 1, t6 # .. e1 :
+ or t6, t8, t7 # e0 :
+ zapnot t1, t6, t1 # .. e1 : kill source bytes >= null
+ zap t0, t7, t0 # e0 : kill dest bytes <= null
+ or t0, t1, t1 # e1 :
+
+1: stq_u t1, 0(a0) # e0 :
+ ret (t9) # .. e1 :
+
+ /* Unaligned copy entry point. */
+ .align 3
+$unaligned:
+
+ ldq_u t1, 0(a1) # e0 : load first source word
+
+ and a0, 7, t4 # .. e1 : find dest misalignment
+ and a1, 7, t5 # e0 : find src misalignment
+
+ /* Conditionally load the first destination word and a bytemask
+ with 0xff indicating that the destination byte is sacrosanct. */
+
+ mov zero, t0 # .. e1 :
+ mov zero, t6 # e0 :
+ beq t4, 1f # .. e1 :
+ ldq_u t0, 0(a0) # e0 :
+ lda t6, -1 # .. e1 :
+ mskql t6, a0, t6 # e0 :
+1:
+ subq a1, t4, a1 # .. e1 : sub dest misalignment from src addr
+
+ /* If source misalignment is larger than dest misalignment, we need
+ extra startup checks to avoid SEGV. */
+
+ cmplt t4, t5, t8 # e0 :
+ beq t8, $u_head # .. e1 (zdb)
+
+ lda t2, -1 # e1 : mask out leading garbage in source
+ mskqh t2, t5, t2 # e0 :
+ nop # e0 :
+ ornot t1, t2, t3 # .. e1 :
+ cmpbge zero, t3, t7 # e0 : is there a zero?
+ beq t7, $u_head # .. e1 (zdb)
+
+ /* At this point we've found a zero in the first partial word of
+ the source. We need to isolate the valid source data and mask
+ it into the original destination data. (Incidentally, we know
+ that we'll need at least one byte of that original dest word.) */
+
+ ldq_u t0, 0(a0) # e0 :
+
+ negq t7, t6 # .. e1 : build bitmask of bytes <= zero
+ and t6, t7, t8 # e0 :
+ nop # .. e1 :
+ subq t8, 1, t6 # e0 :
+ or t6, t8, t7 # e1 :
+
+ zapnot t2, t7, t2 # e0 : prepare source word; mirror changes
+ and t1, t2, t1 # e1 : to source validity mask
+ extql t2, a1, t2 # e0 :
+ extql t1, a1, t1 # e0 :
+
+ andnot t0, t2, t0 # e0 : zero place for source to reside
+ or t0, t1, t1 # e1 : and put it there
+ stq_u t1, 0(a0) # e0 :
+ ret (t9) # .. e1 :
+
+ .end __stxcpy
diff --git a/sysdeps/alpha/stxncpy.S b/sysdeps/alpha/stxncpy.S
new file mode 100644
index 0000000..8c5fcf6
--- /dev/null
+++ b/sysdeps/alpha/stxncpy.S
@@ -0,0 +1,350 @@
+/* Copyright (C) 1996 Free Software Foundation, Inc.
+ Contributed by Richard Henderson (rth@tamu.edu)
+
+This file is part of the GNU C Library.
+
+The GNU C Library is free software; you can redistribute it and/or
+modify it under the terms of the GNU Library General Public License as
+published by the Free Software Foundation; either version 2 of the
+License, or (at your option) any later version.
+
+The GNU C Library 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
+Library General Public License for more details.
+
+You should have received a copy of the GNU Library General Public
+License along with the GNU C Library; see the file COPYING.LIB. If
+not, write to the Free Software Foundation, Inc., 675 Mass Ave,
+Cambridge, MA 02139, USA. */
+
+/* Copy no more than COUNT bytes of the null-terminated string from
+ SRC to DST.
+
+ This is an internal routine used by strncpy, stpncpy, and strncat.
+ As such, it uses special linkage conventions to make implementation
+ of these public functions more efficient.
+
+ On input:
+ t9 = return address
+ a0 = DST
+ a1 = SRC
+ a2 = COUNT
+
+ Furthermore, COUNT may not be zero.
+
+ On output:
+ t0 = last word written
+ t8 = bitmask (with one bit set) indicating the last byte written
+ t10 = bitmask (with one bit set) indicating the byte position of
+ the end of the range specified by COUNT
+ a0 = unaligned address of the last *word* written
+ a2 = the number of full words left in COUNT
+
+ Furthermore, v0, a3-a5, t11, and t12 are untouched.
+*/
+
+
+/* This is generally scheduled for the EV5, but should still be pretty
+ good for the EV4 too. */
+
+#include <sysdep.h>
+
+ .set noat
+ .set noreorder
+
+ .text
+
+/* There is a problem with either gdb (as of 4.16) or gas (as of 2.7) that
+ doesn't like putting the entry point for a procedure somewhere in the
+ middle of the procedure descriptor. Work around this by putting the
+ aligned copy in its own procedure descriptor */
+
+ .ent stxncpy_aligned
+ .align 3
+stxncpy_aligned:
+ .frame sp, 0, t9, 0
+ .prologue 0
+
+ /* On entry to this basic block:
+ t0 == the first destination word for masking back in
+ t1 == the first source word. */
+
+ /* Create the 1st output word and detect 0's in the 1st input word. */
+ lda t2, -1 # e1 : build a mask against false zero
+ mskqh t2, a1, t2 # e0 : detection in the src word
+ mskqh t1, a1, t3 # e0 :
+ ornot t1, t2, t2 # .. e1 :
+ mskql t0, a1, t0 # e0 : assemble the first output word
+ cmpbge zero, t2, t7 # .. e1 : bits set iff null found
+ or t0, t3, t0 # e0 :
+ beq a2, $a_eoc # .. e1 :
+ bne t7, $a_eos # .. e1 :
+
+ /* On entry to this basic block:
+ t0 == a source word not containing a null. */
+
+$a_loop:
+ stq_u t0, 0(a0) # e0 :
+ addq a0, 8, a0 # .. e1 :
+ ldq_u t0, 0(a1) # e0 :
+ addq a1, 8, a1 # .. e1 :
+ subq a2, 1, a2 # e0 :
+ cmpbge zero, t0, t7 # .. e1 (stall)
+ beq a2, $a_eoc # e1 :
+ beq t7, $a_loop # e1 :
+
+ /* Take care of the final (partial) word store. At this point
+ the end-of-count bit is set in t7 iff it applies.
+
+ On entry to this basic block we have:
+ t0 == the source word containing the null
+ t7 == the cmpbge mask that found it. */
+
+$a_eos:
+ negq t7, t8 # e0 : find low bit set
+ and t7, t8, t8 # e1 (stall)
+
+ /* For the sake of the cache, don't read a destination word
+ if we're not going to need it. */
+ and t8, 0x80, t6 # e0 :
+ bne t6, 1f # .. e1 (zdb)
+
+ /* We're doing a partial word store and so need to combine
+ our source and original destination words. */
+ ldq_u t1, 0(a0) # e0 :
+ subq t8, 1, t6 # .. e1 :
+ or t8, t6, t7 # e0 :
+ unop #
+ zapnot t0, t7, t0 # e0 : clear src bytes > null
+ zap t1, t7, t1 # .. e1 : clear dst bytes <= null
+ or t0, t1, t0 # e1 :
+
+1: stq_u t0, 0(a0) # e0 :
+ ret (t9) # e1 :
+
+ /* Add the end-of-count bit to the eos detection bitmask. */
+$a_eoc:
+ or t10, t7, t7
+ br $a_eos
+
+ .end stxncpy_aligned
+
+ .align 3
+ .ent __stxncpy
+ .globl __stxncpy
+__stxncpy:
+ .frame sp, 0, t9, 0
+ .prologue 0
+
+ /* Are source and destination co-aligned? */
+ xor a0, a1, t1 # e0 :
+ and a0, 7, t0 # .. e1 : find dest misalignment
+ and t1, 7, t1 # e0 :
+ addq a2, t0, a2 # .. e1 : bias count by dest misalignment
+ subq a2, 1, a2 # e0 :
+ and a2, 7, t2 # e1 :
+ srl a2, 3, a2 # e0 : a2 = loop counter = (count - 1)/8
+ addq zero, 1, t10 # .. e1 :
+ sll t10, t2, t10 # e0 : t10 = bitmask of last count byte
+ bne t1, $unaligned # .. e1 :
+
+ /* We are co-aligned; take care of a partial first word. */
+
+ ldq_u t1, 0(a1) # e0 : load first src word
+ addq a1, 8, a1 # .. e1 :
+
+ beq t0, stxncpy_aligned # avoid loading dest word if not needed
+ ldq_u t0, 0(a0) # e0 :
+ br stxncpy_aligned # .. e1 :
+
+
+/* The source and destination are not co-aligned. Align the destination
+ and cope. We have to be very careful about not reading too much and
+ causing a SEGV. */
+
+ .align 3
+$u_head:
+ /* We know just enough now to be able to assemble the first
+ full source word. We can still find a zero at the end of it
+ that prevents us from outputting the whole thing.
+
+ On entry to this basic block:
+ t0 == the first dest word, unmasked
+ t1 == the shifted low bits of the first source word
+ t6 == bytemask that is -1 in dest word bytes */
+
+ ldq_u t2, 8(a1) # e0 : load second src word
+ addq a1, 8, a1 # .. e1 :
+ mskql t0, a0, t0 # e0 : mask trailing garbage in dst
+ extqh t2, a1, t4 # e0 :
+ or t1, t4, t1 # e1 : first aligned src word complete
+ mskqh t1, a0, t1 # e0 : mask leading garbage in src
+ or t0, t1, t0 # e0 : first output word complete
+ or t0, t6, t6 # e1 : mask original data for zero test
+ cmpbge zero, t6, t7 # e0 :
+ beq a2, $u_eocfin # .. e1 :
+ bne t7, $u_final # e1 :
+
+ lda t6, -1 # e1 : mask out the bits we have
+ mskql t6, a1, t6 # e0 : already seen
+ stq_u t0, 0(a0) # e0 : store first output word
+ or t6, t2, t2 # .. e1 :
+ cmpbge zero, t2, t7 # e0 : find nulls in second partial
+ addq a0, 8, a0 # .. e1 :
+ subq a2, 1, a2 # e0 :
+ bne t7, $u_late_head_exit # .. e1 :
+
+ /* Finally, we've got all the stupid leading edge cases taken care
+ of and we can set up to enter the main loop. */
+
+ extql t2, a1, t1 # e0 : position hi-bits of lo word
+ ldq_u t2, 8(a1) # .. e1 : read next high-order source word
+ addq a1, 8, a1 # e0 :
+ cmpbge zero, t2, t7 # e1 (stall)
+ beq a2, $u_eoc # e1 :
+ bne t7, $u_eos # e1 :
+
+ /* Unaligned copy main loop. In order to avoid reading too much,
+ the loop is structured to detect zeros in aligned source words.
+ This has, unfortunately, effectively pulled half of a loop
+ iteration out into the head and half into the tail, but it does
+ prevent nastiness from accumulating in the very thing we want
+ to run as fast as possible.
+
+ On entry to this basic block:
+ t1 == the shifted high-order bits from the previous source word
+ t2 == the unshifted current source word
+
+ We further know that t2 does not contain a null terminator. */
+
+ .align 3
+$u_loop:
+ extqh t2, a1, t0 # e0 : extract high bits for current word
+ addq a1, 8, a1 # .. e1 :
+ extql t2, a1, t3 # e0 : extract low bits for next time
+ addq a0, 8, a0 # .. e1 :
+ or t0, t1, t0 # e0 : current dst word now complete
+ ldq_u t2, 0(a1) # .. e1 : load high word for next time
+ stq_u t0, -8(a0) # e0 : save the current word
+ mov t3, t1 # .. e1 :
+ subq a2, 1, a2 # e0 :
+ cmpbge zero, t2, t7 # .. e1 : test new word for eos
+ beq a2, $u_eoc # e1 :
+ beq t7, $u_loop # e1 :
+
+ /* We've found a zero somewhere in the source word we just read.
+ If it resides in the lower half, we have one (probably partial)
+ word to write out, and if it resides in the upper half, we
+ have one full and one partial word left to write out.
+
+ On entry to this basic block:
+ t1 == the shifted high-order bits from the previous source word
+ t2 == the unshifted current source word. */
+$u_eos:
+ extqh t2, a1, t0 # e0 :
+ or t0, t1, t0 # e1 : first (partial) source word complete
+
+ cmpbge zero, t0, t7 # e0 : is the null in this first bit?
+ bne t7, $u_final # .. e1 (zdb)
+
+ stq_u t0, 0(a0) # e0 : the null was in the high-order bits
+ addq a0, 8, a0 # .. e1 :
+ subq a2, 1, a2 # e1 :
+
+$u_late_head_exit:
+ extql t2, a1, t0 # .. e0 :
+ cmpbge zero, t0, t7 # e0 :
+ or t7, t10, t6 # e1 :
+ cmoveq a2, t6, t7 # e0 :
+ nop # .. e1 :
+
+ /* Take care of a final (probably partial) result word.
+ On entry to this basic block:
+ t0 == assembled source word
+ t7 == cmpbge mask that found the null. */
+$u_final:
+ negq t7, t6 # e0 : isolate low bit set
+ and t6, t7, t8 # e1 :
+
+ and t8, 0x80, t6 # e0 : avoid dest word load if we can
+ bne t6, 1f # .. e1 (zdb)
+
+ ldq_u t1, 0(a0) # e0 :
+ subq t8, 1, t6 # .. e1 :
+ or t6, t8, t7 # e0 :
+ zapnot t0, t7, t0 # .. e1 : kill source bytes > null
+ zap t1, t7, t1 # e0 : kill dest bytes <= null
+ or t0, t1, t0 # e1 :
+
+1: stq_u t0, 0(a0) # e0 :
+ ret (t9) # .. e1 :
+
+$u_eoc: # end-of-count
+ extqh t2, a1, t0
+ or t0, t1, t0
+ cmpbge zero, t0, t7
+
+$u_eocfin: # end-of-count, final word
+ or t10, t7, t7
+ br $u_final
+
+ /* Unaligned copy entry point. */
+ .align 3
+$unaligned:
+
+ ldq_u t1, 0(a1) # e0 : load first source word
+
+ and a0, 7, t4 # .. e1 : find dest misalignment
+ and a1, 7, t5 # e0 : find src misalignment
+
+ /* Conditionally load the first destination word and a bytemask
+ with 0xff indicating that the destination byte is sacrosanct. */
+
+ mov zero, t0 # .. e1 :
+ mov zero, t6 # e0 :
+ beq t4, 1f # .. e1 :
+ ldq_u t0, 0(a0) # e0 :
+ lda t6, -1 # .. e1 :
+ mskql t6, a0, t6 # e0 :
+1:
+ subq a1, t4, a1 # .. e1 : sub dest misalignment from src addr
+
+ /* If source misalignment is larger than dest misalignment, we need
+ extra startup checks to avoid SEGV. */
+
+ cmplt t4, t5, t8 # e1 :
+ extql t1, a1, t1 # .. e0 : shift src into place
+ lda t2, -1 # e0 : for creating masks later
+ beq t8, $u_head # e1 :
+
+ mskqh t2, t5, t2 # e0 : begin src byte validity mask
+ cmpbge zero, t1, t7 # .. e1 : is there a zero?
+ extql t2, a1, t2 # e0 :
+ or t7, t10, t6 # .. e1 : test for end-of-count too
+ cmpbge zero, t2, t3 # e0 :
+ cmoveq a2, t6, t7 # .. e1 :
+ andnot t7, t3, t7 # e0 :
+ beq t7, $u_head # .. e1 (zdb)
+
+ /* At this point we've found a zero in the first partial word of
+ the source. We need to isolate the valid source data and mask
+ it into the original destination data. (Incidentally, we know
+ that we'll need at least one byte of that original dest word.) */
+
+ ldq_u t0, 0(a0) # e0 :
+ negq t7, t6 # .. e1 : build bitmask of bytes <= zero
+ mskqh t1, t4, t1 # e0 :
+ and t6, t7, t8 # .. e1 :
+ subq t8, 1, t6 # e0 :
+ or t6, t8, t7 # e1 :
+
+ zapnot t2, t7, t2 # e0 : prepare source word; mirror changes
+ zapnot t1, t7, t1 # .. e1 : to source validity mask
+
+ andnot t0, t2, t0 # e0 : zero place for source to reside
+ or t0, t1, t0 # e1 : and put it there
+ stq_u t0, 0(a0) # e0 :
+ ret (t9) # .. e1 :
+
+ .end __stxncpy