diff options
author | Andrew Cagney <cagney@redhat.com> | 2004-02-20 16:26:45 +0000 |
---|---|---|
committer | Andrew Cagney <cagney@redhat.com> | 2004-02-20 16:26:45 +0000 |
commit | e866a2571a7e3fa835ee3915ff723b0482e4c2d6 (patch) | |
tree | 319a4c80e21266028f45332f9daf761b9eb51001 /cpu/m32r.cpu | |
parent | 299d901ce6a983dfecbbe5232c4fe33522b79d85 (diff) | |
download | gdb-e866a2571a7e3fa835ee3915ff723b0482e4c2d6.zip gdb-e866a2571a7e3fa835ee3915ff723b0482e4c2d6.tar.gz gdb-e866a2571a7e3fa835ee3915ff723b0482e4c2d6.tar.bz2 |
2004-02-20 Andrew Cagney <cagney@redhat.com>
* m32r.opc, m32r.cpu: New files. Written by , Doug Evans, Nick
Clifton, Ben Elliston, Matthew Green, and Andrew Haley.
Diffstat (limited to 'cpu/m32r.cpu')
-rw-r--r-- | cpu/m32r.cpu | 2127 |
1 files changed, 2127 insertions, 0 deletions
diff --git a/cpu/m32r.cpu b/cpu/m32r.cpu new file mode 100644 index 0000000..a2f94a8 --- /dev/null +++ b/cpu/m32r.cpu @@ -0,0 +1,2127 @@ +; Mitsubishi M32R CPU description. -*- Scheme -*- +; +; Copyright 1998, 1999, 2000, 2001, 2003 Free Software Foundation, Inc. +; +; Contributed by Red Hat Inc; developed under contract from Mitsubishi +; Electric Corporation. +; +; This file is part of the GNU Binutils. +; +; This program 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 2 of the License, or +; (at your option) any later version. +; +; This program 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 this program; if not, write to the Free Software +; Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA. + +(include "simplify.inc") + +; FIXME: Delete sign extension of accumulator results. +; Sign extension is done when accumulator is read. + +; define-arch must appear first + +(define-arch + (name m32r) ; name of cpu family + (comment "Mitsubishi M32R") + (default-alignment aligned) + (insn-lsb0? #f) + (machs m32r m32rx) + (isas m32r) +) + +; Attributes. + +; An attribute to describe which pipeline an insn runs in. + +(define-attr + (for insn) + (type enum) + (name PIPE) + (comment "parallel execution pipeline selection") + (values NONE O S OS) +) + +; A derived attribute that says which insns can be executed in parallel +; with others. This is a required attribute for architectures with +; parallel execution. + +(define-attr + (for insn) + (type enum) + (name PARALLEL) + (attrs META) ; do not define in any generated file for now + (values NO YES) + (default (if (eq-attr (current-insn) PIPE NONE) (symbol NO) (symbol YES))) +) + +; Instruction set parameters. + +(define-isa + (name m32r) + + ; This is 32 because 16 bit insns always appear as pairs. + ; ??? See if this can go away. It's only used by the disassembler (right?) + ; to decide how long an unknown insn is. One value isn't sufficient (e.g. if + ; on a 16 bit (and not 32 bit) boundary, will only want to advance pc by 16.) + (default-insn-bitsize 32) + + ; Number of bytes of insn we can initially fetch. + ; The M32R is tricky in that insns are either two 16-bit insns + ; (executed sequentially or in parallel) or one 32-bit insn. + ; So on one hand the base insn size is 16 bits, but on another it's 32. + ; 32 is chosen because: + ; - if the chip were ever bi-endian it is believed that the byte order would + ; be based on 32 bit quantities + ; - 32 bit insns are always aligned on 32 bit boundaries + ; - the pc will never stop on a 16 bit (and not 32 bit) boundary + ; [well actually it can, but there are no branches to such places] + (base-insn-bitsize 32) + + ; Used in computing bit numbers. + (default-insn-word-bitsize 32) + + ; The m32r fetches 2 insns at a time. + (liw-insns 2) + + ; While the m32r can execute insns in parallel, the base mach can't + ; (other than nop). The base mach is greatly handicapped by this, but + ; we still need to cleanly handle it. + (parallel-insns 2) + + ; Initial bitnumbers to decode insns by. + (decode-assist (0 1 2 3 8 9 10 11)) + + ; Classification of instructions that fit in the various frames. + ; wip, not currently used + (insn-types (long ; name + 31 ; length + (eq-attr (current-insn) LENGTH 31) ; matching insns + (0 1 2 7 8 9 10) ; decode-assist + ) + (short + 15 + (eq-attr (current-insn) LENGTH 15) ; matching insns + (0 1 2 7 8 9 10) + ) + ) + + ; Instruction framing. + ; Each m32r insn is either one 32 bit insn, two 16 bit insns executed + ; serially (left->right), or two 16 bit insns executed parallelly. + ; wip, not currently used + (frame long32 ; name + ((long)) ; list of insns in frame, plus constraint + "$0" ; assembler + (+ (1 1) (31 $0)) ; value + (sequence () (execute $0)) ; action + ) + (frame serial2x16 + ((short) + (short)) + "$0 -> $1" + (+ (1 0) (15 $0) (1 0) (15 $1)) + (sequence () + (execute $0) + (execute $1)) + ) + (frame parallel2x16 + ((short (eq-attr (current-insn) PIPE "O,BOTH")) + (short (eq-attr (current-insn) PIPE "S,BOTH"))) + "$0 || $1" + (+ (1 0) (15 $0) (1 1) (15 $1)) + (parallel () + (execute $0) + (execute $1)) + ) +) + +; Cpu family definitions. + +; ??? define-cpu-family [and in general "cpu-family"] might be clearer than +; define-cpu. +; ??? Have define-arch provide defaults for architecture that define-cpu can +; then override [reduces duplication in define-cpu]. +; ??? Another way to go is to delete cpu-families entirely and have one mach +; able to inherit things from another mach (would also need the ability to +; not only override specific inherited things but also disable some, +; e.g. if an insn wasn't supported). + +(define-cpu + ; cpu names must be distinct from the architecture name and machine names. + ; The "b" suffix stands for "base" and is the convention. + ; The "f" suffix stands for "family" and is the convention. + (name m32rbf) + (comment "Mitsubishi M32R base family") + (endian big) + (word-bitsize 32) + ; Override isa spec (??? keeps things simpler, though it was more true + ; in the early days and not so much now). + (parallel-insns 1) +) + +(define-cpu + (name m32rxf) + (comment "Mitsubishi M32Rx family") + (endian big) + (word-bitsize 32) + ; Generated files have an "x" suffix. + (file-transform "x") +) + +(define-mach + (name m32r) + (comment "Generic M32R cpu") + (cpu m32rbf) +) + +(define-mach + (name m32rx) + (comment "M32RX cpu") + (cpu m32rxf) +) + +; Model descriptions. + +; The meaning of this value is wip but at the moment it's intended to describe +; the implementation (i.e. what -mtune=foo does in sparc gcc). +; +; Notes while wip: +; - format of pipeline entry: +; (pipeline name (stage1-name ...) (stage2-name ...) ...) +; The contents of a stage description is wip. +; - each mach must have at least one model +; - the default model must be the first one +;- maybe have `retire' support update total cycle count to handle current +; parallel insn cycle counting problems + +(define-model + (name m32r/d) (comment "m32r/d") (attrs) + (mach m32r) + + ;(prefetch) + ;(retire) + + (pipeline p-non-mem "" () ((fetch) (decode) (execute) (writeback))) + (pipeline p-mem "" () ((fetch) (decode) (execute) (memory) (writeback))) + + ; `state' is a list of variables for recording model state + (state + ; bit mask of h-gr registers, =1 means value being loaded from memory + (h-gr UINT) + ) + + (unit u-exec "Execution Unit" () + 1 1 ; issue done + () ; state + ((sr INT -1) (dr INT -1)) ; inputs + ((dr INT -1)) ; outputs + () ; profile action (default) + ) + (unit u-cmp "Compare Unit" () + 1 1 ; issue done + () ; state + ((src1 INT -1) (src2 INT -1)) ; inputs + () ; outputs + () ; profile action (default) + ) + (unit u-mac "Multiply/Accumulate Unit" () + 1 1 ; issue done + () ; state + ((src1 INT -1) (src2 INT -1)) ; inputs + () ; outputs + () ; profile action (default) + ) + (unit u-cti "Branch Unit" () + 1 1 ; issue done + () ; state + ((sr INT -1)) ; inputs + ((pc)) ; outputs + () ; profile action (default) + ) + (unit u-load "Memory Load Unit" () + 1 1 ; issue done + () ; state + ((sr INT) + ;(ld-mem AI) + ) ; inputs + ((dr INT)) ; outputs + () ; profile action (default) + ) + (unit u-store "Memory Store Unit" () + 1 1 ; issue done + () ; state + ((src1 INT) (src2 INT)) ; inputs + () ; ((st-mem AI)) ; outputs + () ; profile action (default) + ) +) + +(define-model + (name test) (comment "test") (attrs) + (mach m32r) + (pipeline all "" () ((fetch) (decode) (execute) (writeback))) + (unit u-exec "Execution Unit" () + 1 1 ; issue done + () () () ()) +) + +; Each mach must have at least one model. + +(define-model + (name m32rx) (comment "m32rx") (attrs) + (mach m32rx) + + ; ??? It's 6 stages but I forget the details right now. + (pipeline p-o "" () ((fetch) (decode) (execute) (writeback))) + (pipeline p-s "" () ((fetch) (decode) (execute) (writeback))) + (pipeline p-o-mem "" () ((fetch) (decode) (execute) (memory) (writeback))) + + (unit u-exec "Execution Unit" () + 1 1 ; issue done + () ; state + ((sr INT -1) (dr INT -1)) ; inputs + ((dr INT -1)) ; outputs + () ; profile action (default) + ) + (unit u-cmp "Compare Unit" () + 1 1 ; issue done + () ; state + ((src1 INT -1) (src2 INT -1)) ; inputs + () ; outputs + () ; profile action (default) + ) + (unit u-mac "Multiply/Accumulate Unit" () + 1 1 ; issue done + () ; state + ((src1 INT -1) (src2 INT -1)) ; inputs + () ; outputs + () ; profile action (default) + ) + (unit u-cti "Branch Unit" () + 1 1 ; issue done + () ; state + ((sr INT -1)) ; inputs + ((pc)) ; outputs + () ; profile action (default) + ) + (unit u-load "Memory Load Unit" () + 1 1 ; issue done + () ; state + ((sr INT)) ; inputs + ((dr INT)) ; outputs + () ; profile action (default) + ) + (unit u-store "Memory Store Unit" () + 1 1 ; issue done + () ; state + ((src1 INT) (src2 INT)) ; inputs + () ; outputs + () ; profile action (default) + ) +) + +; The instruction fetch/execute cycle. +; This is split into two parts as sometimes more than one instruction is +; decoded at once. +; The `const SI' argument to decode/execute is used to distinguish +; multiple instructions processed at the same time (e.g. m32r). +; +; ??? This is wip, and not currently used. +; ??? Needs to be moved to define-isa. + +; This is how to fetch and decode an instruction. + +;(define-extract +; (sequence VOID +; (if VOID (ne AI (and AI pc (const AI 3)) (const AI 0)) +; (sequence VOID +; (set-quiet USI (scratch UHI insn1) (ifetch UHI pc)) +; (decode VOID pc (and UHI insn1 (const UHI #x7fff)) +; (const SI 0))) +; (sequence VOID +; (set-quiet USI (scratch USI insn) (ifetch USI pc)) +; (if VOID (ne USI (and USI insn (const USI #x80000000)) +; (const USI 0)) +; (decode VOID pc (srl USI insn (const WI 16)) (const SI 0)) +; (sequence VOID +; ; ??? parallel support +; (decode VOID pc (srl USI insn (const WI 16)) +; (const SI 0)) +; (decode VOID (add AI pc (const AI 2)) +; (and USI insn (const WI #x7fff)) +; (const SI 1)))))) +; ) +;) + +; This is how to execute a decoded instruction. + +;(define-execute +; (sequence VOID () ; () is empty option list +; ((AI new_pc)) +; (set AI new_pc (execute: AI (const 0)) #:quiet) +; (set AI pc new_pc #:direct) +; ) +;) + +; FIXME: It might simplify things to separate the execute process from the +; one that updates the PC. + +; Instruction fields. +; +; Attributes: +; PCREL-ADDR: pc relative value (for reloc and disassembly purposes) +; ABS-ADDR: absolute address (for reloc and disassembly purposes?) +; RESERVED: bits are not used to decode insn, must be all 0 +; RELOC: there is a relocation associated with this field (experiment) + +(define-attr + (for ifield operand) + (type boolean) + (name RELOC) + (comment "there is a reloc associated with this field (experiment)") +) + +(dnf f-op1 "op1" () 0 4) +(dnf f-op2 "op2" () 8 4) +(dnf f-cond "cond" () 4 4) +(dnf f-r1 "r1" () 4 4) +(dnf f-r2 "r2" () 12 4) +(df f-simm8 "simm8" () 8 8 INT #f #f) +(df f-simm16 "simm16" () 16 16 INT #f #f) +(dnf f-shift-op2 "shift op2" () 8 3) +(dnf f-uimm4 "uimm4" () 12 4) +(dnf f-uimm5 "uimm5" () 11 5) +(dnf f-uimm16 "uimm16" () 16 16) +(dnf f-uimm24 "uimm24" (ABS-ADDR RELOC) 8 24) +(dnf f-hi16 "high 16 bits" (SIGN-OPT) 16 16) +(df f-disp8 "disp8, slot unknown" (PCREL-ADDR RELOC) 8 8 INT + ((value pc) (sra WI (sub WI value (and WI pc (const -4))) (const 2))) + ((value pc) (add WI (sll WI value (const 2)) (and WI pc (const -4))))) +(df f-disp16 "disp16" (PCREL-ADDR RELOC) 16 16 INT + ((value pc) (sra WI (sub WI value pc) (const 2))) + ((value pc) (add WI (sll WI value (const 2)) pc))) +(df f-disp24 "disp24" (PCREL-ADDR RELOC) 8 24 INT + ((value pc) (sra WI (sub WI value pc) (const 2))) + ((value pc) (add WI (sll WI value (const 2)) pc))) + +(dnf f-op23 "op2.3" () 9 3) +(dnf f-op3 "op3" () 14 2) +(dnf f-acc "acc" () 8 1) +(dnf f-accs "accs" () 12 2) +(dnf f-accd "accd" () 4 2) +(dnf f-bits67 "bits67" () 6 2) +(dnf f-bit14 "bit14" () 14 1) + +(define-ifield (name f-imm1) (comment "1 bit immediate, 0->1 1->2") + (attrs) + (start 15) (length 1) + (encode (value pc) (sub WI value (const WI 1))) + (decode (value pc) (add WI value (const WI 1))) +) + +; Enums. + +; insn-op1: bits 0-3 +; FIXME: should use die macro or some such +(define-normal-insn-enum insn-op1 "insn format enums" () OP1_ f-op1 + ("0" "1" "2" "3" "4" "5" "6" "7" + "8" "9" "10" "11" "12" "13" "14" "15") +) + +; insn-op2: bits 8-11 +; FIXME: should use die macro or some such +(define-normal-insn-enum insn-op2 "op2 enums" () OP2_ f-op2 + ("0" "1" "2" "3" "4" "5" "6" "7" + "8" "9" "10" "11" "12" "13" "14" "15") +) + +; Hardware pieces. +; These entries list the elements of the raw hardware. +; They're also used to provide tables and other elements of the assembly +; language. + +(dnh h-pc "program counter" (PC PROFILE) (pc) () () ()) + +(dnh h-hi16 "high 16 bits" () + (immediate (UINT 16)) + () () () +) + +; These two aren't technically needed. +; They're here for illustration sake mostly. +; Plus they cause the value to be stored in the extraction buffers to only +; be 16 bits wide (vs 32 or 64). Whoopie ding. But it's fun. +(dnh h-slo16 "signed low 16 bits" () + (immediate (INT 16)) + () () () +) +(dnh h-ulo16 "unsigned low 16 bits" () + (immediate (UINT 16)) + () () () +) + +(define-keyword + (name gr-names) + (print-name h-gr) + (prefix "") + (values (fp 13) (lr 14) (sp 15) + (r0 0) (r1 1) (r2 2) (r3 3) (r4 4) (r5 5) (r6 6) (r7 7) + (r8 8) (r9 9) (r10 10) (r11 11) (r12 12) (r13 13) (r14 14) (r15 15)) +) + +(define-hardware + (name h-gr) + (comment "general registers") + (attrs PROFILE CACHE-ADDR) + (type register WI (16)) + (indices extern-keyword gr-names) +) + +(define-keyword + (name cr-names) + (print-name h-cr) + (prefix "") + (values (psw 0) (cbr 1) (spi 2) (spu 3) + (bpc 6) (bbpsw 8) (bbpc 14) + (cr0 0) (cr1 1) (cr2 2) (cr3 3) + (cr4 4) (cr5 5) (cr6 6) (cr7 7) + (cr8 8) (cr9 9) (cr10 10) (cr11 11) + (cr12 12) (cr13 13) (cr14 14) (cr15 15)) +) + +(define-hardware + (name h-cr) + (comment "control registers") + (type register UWI (16)) + (indices extern-keyword cr-names) + (get (index) (c-call UWI "@cpu@_h_cr_get_handler" index)) + (set (index newval) (c-call VOID "@cpu@_h_cr_set_handler" index newval)) +) + +; The actual accumulator is only 56 bits. +; The top 8 bits are sign extended from bit 8 (when counting msb = bit 0). +; To simplify the accumulator instructions, no attempt is made to keep the +; top 8 bits properly sign extended (currently there's no point since they +; all ignore them). When the value is read it is properly sign extended +; [in the `get' handler]. +(define-hardware + (name h-accum) + (comment "accumulator") + (type register DI) + (get () (c-call DI "@cpu@_h_accum_get_handler")) + (set (newval) (c-call VOID "@cpu@_h_accum_set_handler" newval)) +) + +; FIXME: Revisit after sanitization can be removed. Remove h-accum. +(define-hardware + (name h-accums) + (comment "accumulators") + (attrs (MACH m32rx)) + (type register DI (2)) + (indices keyword "" ((a0 0) (a1 1))) + ; get/set so a0 accesses are redirected to h-accum. + ; They're also so reads can properly sign extend the value. + ; FIXME: Needn't be a function call. + (get (index) (c-call DI "@cpu@_h_accums_get_handler" index)) + (set (index newval) (c-call VOID "@cpu@_h_accums_set_handler" index newval)) +) + +; For condbit operand. FIXME: Need to allow spec of get/set of operands. +; Having this separate from h-psw keeps the parts that use it simpler +; [since they greatly outnumber those that use h-psw]. +(dsh h-cond "condition bit" () (register BI)) + +; The actual values of psw,bpsw,bbpsw are recorded here to allow access +; to them as a unit. +(define-hardware + (name h-psw) + (comment "psw part of psw") + (type register UQI) + ; get/set to handle cond bit. + ; FIXME: missing: use's and clobber's + ; FIXME: remove c-call? + (get () (c-call UQI "@cpu@_h_psw_get_handler")) + (set (newval) (c-call VOID "@cpu@_h_psw_set_handler" newval)) +) +(dsh h-bpsw "backup psw" () (register UQI)) +(dsh h-bbpsw "backup bpsw" () (register UQI)) + +; FIXME: Later make add get/set specs and support SMP. +(dsh h-lock "lock" () (register BI)) + +; Instruction Operands. +; These entries provide a layer between the assembler and the raw hardware +; description, and are used to refer to hardware elements in the semantic +; code. Usually there's a bit of over-specification, but in more complicated +; instruction sets there isn't. + +; M32R specific operand attributes: + +(define-attr + (for operand) + (type boolean) + (name HASH-PREFIX) + (comment "immediates have an optional '#' prefix") +) + +; ??? Convention says this should be o-sr, but then the insn definitions +; should refer to o-sr which is clumsy. The "o-" could be implicit, but +; then it should be implicit for all the symbols here, but then there would +; be confusion between (f-)simm8 and (h-)simm8. +; So for now the rule is exactly as it appears here. + +(dnop sr "source register" () h-gr f-r2) +(dnop dr "destination register" () h-gr f-r1) +;; The assembler relies upon the fact that dr and src1 are the same field. +;; FIXME: Revisit. +(dnop src1 "source register 1" () h-gr f-r1) +(dnop src2 "source register 2" () h-gr f-r2) +(dnop scr "source control register" () h-cr f-r2) +(dnop dcr "destination control register" () h-cr f-r1) + +(dnop simm8 "8 bit signed immediate" (HASH-PREFIX) h-sint f-simm8) +(dnop simm16 "16 bit signed immediate" (HASH-PREFIX) h-sint f-simm16) +(dnop uimm4 "4 bit trap number" (HASH-PREFIX) h-uint f-uimm4) +(dnop uimm5 "5 bit shift count" (HASH-PREFIX) h-uint f-uimm5) +(dnop uimm16 "16 bit unsigned immediate" (HASH-PREFIX) h-uint f-uimm16) + +(dnop imm1 "1 bit immediate" ((MACH m32rx) HASH-PREFIX) h-uint f-imm1) +(dnop accd "accumulator destination register" ((MACH m32rx)) h-accums f-accd) +(dnop accs "accumulator source register" ((MACH m32rx)) h-accums f-accs) +(dnop acc "accumulator reg (d)" ((MACH m32rx)) h-accums f-acc) + +; slo16,ulo16 are used in both with-hash-prefix/no-hash-prefix cases. +; e.g. add3 r3,r3,#1 and ld r3,@(4,r4). We could use HASH-PREFIX. +; Instead we create a fake operand `hash'. The m32r is an illustration port, +; so we often try out various ways of doing things. + +(define-operand (name hash) (comment "# prefix") (attrs) + (type h-sint) ; doesn't really matter + (index f-nil) + (handlers (parse "hash") (print "hash")) +) + +; For high(foo),shigh(foo). +(define-operand + (name hi16) + (comment "high 16 bit immediate, sign optional") + (attrs) + (type h-hi16) + (index f-hi16) + (handlers (parse "hi16")) +) + +; For low(foo),sda(foo). +(define-operand + (name slo16) + (comment "16 bit signed immediate, for low()") + (attrs) + (type h-slo16) + (index f-simm16) + (handlers (parse "slo16")) +) + +; For low(foo). +(define-operand + (name ulo16) + (comment "16 bit unsigned immediate, for low()") + (attrs) + (type h-ulo16) + (index f-uimm16) + (handlers (parse "ulo16")) +) + +(dnop uimm24 "24 bit address" (HASH-PREFIX) h-addr f-uimm24) + +(define-operand + (name disp8) + (comment "8 bit displacement") + (attrs RELAX) + (type h-iaddr) + (index f-disp8) + ; ??? Early experiments had insert/extract fields here. + ; Moving these to f-disp8 made things cleaner, but may wish to re-introduce + ; fields here to handle more complicated cases. +) + +(dnop disp16 "16 bit displacement" () h-iaddr f-disp16) +(dnop disp24 "24 bit displacement" (RELAX) h-iaddr f-disp24) + +; These hardware elements are refered to frequently. + +(dnop condbit "condition bit" (SEM-ONLY) h-cond f-nil) +(dnop accum "accumulator" (SEM-ONLY) h-accum f-nil) + +; Instruction definitions. +; +; Notes while wip: +; - dni is a cover macro to the real "this is an instruction" keyword. +; The syntax of the real one is yet to be determined. +; At the lowest level (i.e. the "real" one) it will probably take a variable +; list of arguments where each argument [perhaps after the standard three of +; name, comment, attrs] is "(keyword arg-to-keyword)". This syntax is simple +; and yet completely upward extensible. And given the macro facility, one +; needn't code at that low a level so even though it'll be more verbose than +; necessary it won't matter. This same reasoning can be applied to most +; types of entries in this file. + +; M32R specific instruction attributes: + +; FILL-SLOT: Need next insn to begin on 32 bit boundary. +; (A "slot" as used here is a 32 bit quantity that can either be filled with +; one 32 bit insn or two 16 bit insns which go in the "left bin" and "right +; bin" where the left bin is the one with a lower address). + +(define-attr + (for insn) + (type boolean) + (name FILL-SLOT) + (comment "fill right bin with `nop' if insn is in left bin") +) + +(define-attr + (for insn) + (type boolean) + (name SPECIAL) + (comment "non-public m32rx insn") +) + +; IDOC attribute for instruction documentation. + +(define-attr + (for insn) + (type enum) + (name IDOC) + (comment "insn kind for documentation") + (attrs META) + (values + (MEM - () "Memory") + (ALU - () "ALU") + (BR - () "Branch") + (ACCUM - () "Accumulator") + (MAC - () "Multiply/Accumulate") + (MISC - () "Miscellaneous") + ) +) + +(define-pmacro (bin-op mnemonic op2-op sem-op imm-prefix imm) + (begin + (dni mnemonic + (.str mnemonic " reg/reg") + ((PIPE OS) (IDOC ALU)) + (.str mnemonic " $dr,$sr") + (+ OP1_0 op2-op dr sr) + (set dr (sem-op dr sr)) + () + ) + (dni (.sym mnemonic "3") + (.str mnemonic " reg/" imm) + ((IDOC ALU)) + (.str mnemonic "3 $dr,$sr," imm-prefix "$" imm) + (+ OP1_8 op2-op dr sr imm) + (set dr (sem-op sr imm)) + () + ) + ) +) +(bin-op add OP2_10 add "$hash" slo16) +; sub isn't present because sub3 doesn't exist. +(bin-op and OP2_12 and "" uimm16) +(bin-op or OP2_14 or "$hash" ulo16) +(bin-op xor OP2_13 xor "" uimm16) + +(dni addi "addi" + ((PIPE OS) (IDOC ALU)) + ;#.(string-append "addi " "$dr,$simm8") ; #. experiment + "addi $dr,$simm8" + (+ OP1_4 dr simm8) + (set dr (add dr simm8)) + ((m32r/d (unit u-exec)) + (m32rx (unit u-exec))) +) + +(dni addv "addv" + ((PIPE OS) (IDOC ALU)) + "addv $dr,$sr" + (+ OP1_0 OP2_8 dr sr) + (parallel () + (set dr (add dr sr)) + (set condbit (add-oflag dr sr (const 0)))) + () +) + +(dni addv3 "addv3" + ((IDOC ALU)) + "addv3 $dr,$sr,$simm16" + (+ OP1_8 OP2_8 dr sr simm16) + (parallel () + (set dr (add sr simm16)) + (set condbit (add-oflag sr simm16 (const 0)))) + () +) + +(dni addx "addx" + ((PIPE OS) (IDOC ALU)) + "addx $dr,$sr" + (+ OP1_0 OP2_9 dr sr) + (parallel () + (set dr (addc dr sr condbit)) + (set condbit (add-cflag dr sr condbit))) + () +) + +(dni bc8 "bc with 8 bit displacement" + (COND-CTI (PIPE O) (IDOC BR)) + "bc.s $disp8" + (+ OP1_7 (f-r1 12) disp8) + (if condbit (set pc disp8)) + ((m32r/d (unit u-cti)) + (m32rx (unit u-cti))) +) + +(dnmi bc8r "relaxable bc8" + (COND-CTI RELAXABLE (PIPE O) (IDOC BR)) + "bc $disp8" + (emit bc8 disp8) +) + +(dni bc24 "bc with 24 bit displacement" + (COND-CTI (IDOC BR)) + "bc.l $disp24" + (+ OP1_15 (f-r1 12) disp24) + (if condbit (set pc disp24)) + ((m32r/d (unit u-cti)) + (m32rx (unit u-cti))) +) + +(dnmi bc24r "relaxable bc24" + (COND-CTI RELAXED (IDOC BR)) + "bc $disp24" + (emit bc24 disp24) +) + +(dni beq "beq" + (COND-CTI (IDOC BR)) + "beq $src1,$src2,$disp16" + (+ OP1_11 OP2_0 src1 src2 disp16) + (if (eq src1 src2) (set pc disp16)) + ((m32r/d (unit u-cti) (unit u-cmp (cycles 0))) + (m32rx (unit u-cti) (unit u-cmp (cycles 0)))) +) + +(define-pmacro (cbranch sym comment op2-op comp-op) + (dni sym comment (COND-CTI (IDOC BR)) + (.str sym " $src2,$disp16") + (+ OP1_11 op2-op (f-r1 0) src2 disp16) + (if (comp-op src2 (const WI 0)) (set pc disp16)) + ((m32r/d (unit u-cti) (unit u-cmp (cycles 0))) + (m32rx (unit u-cti) (unit u-cmp (cycles 0)))) + ) +) +(cbranch beqz "beqz" OP2_8 eq) +(cbranch bgez "bgez" OP2_11 ge) +(cbranch bgtz "bgtz" OP2_13 gt) +(cbranch blez "blez" OP2_12 le) +(cbranch bltz "bltz" OP2_10 lt) +(cbranch bnez "bnez" OP2_9 ne) + +(dni bl8 "bl with 8 bit displacement" + (UNCOND-CTI FILL-SLOT (PIPE O) (IDOC BR)) + "bl.s $disp8" + (+ OP1_7 (f-r1 14) disp8) + (sequence () + (set (reg h-gr 14) + (add (and pc (const -4)) (const 4))) + (set pc disp8)) + ((m32r/d (unit u-cti)) + (m32rx (unit u-cti))) +) + +(dnmi bl8r "relaxable bl8" + (UNCOND-CTI FILL-SLOT RELAXABLE (PIPE O) (IDOC BR)) + "bl $disp8" + (emit bl8 disp8) +) + +(dni bl24 "bl with 24 bit displacement" + (UNCOND-CTI (IDOC BR)) + "bl.l $disp24" + (+ OP1_15 (f-r1 14) disp24) + (sequence () + (set (reg h-gr 14) (add pc (const 4))) + (set pc disp24)) + ((m32r/d (unit u-cti)) + (m32rx (unit u-cti))) +) + +(dnmi bl24r "relaxable bl24" + (UNCOND-CTI RELAXED (IDOC BR)) + "bl $disp24" + (emit bl24 disp24) +) + +(dni bcl8 "bcl with 8 bit displacement" + (COND-CTI FILL-SLOT (MACH m32rx) (PIPE O) (IDOC BR)) + "bcl.s $disp8" + (+ OP1_7 (f-r1 8) disp8) + (if condbit + (sequence () + (set (reg h-gr 14) + (add (and pc (const -4)) + (const 4))) + (set pc disp8))) + ((m32rx (unit u-cti))) +) + +(dnmi bcl8r "relaxable bcl8" + (COND-CTI FILL-SLOT (MACH m32rx) (PIPE O) RELAXABLE (IDOC BR)) + "bcl $disp8" + (emit bcl8 disp8) +) + +(dni bcl24 "bcl with 24 bit displacement" + (COND-CTI (MACH m32rx) (IDOC BR)) + "bcl.l $disp24" + (+ OP1_15 (f-r1 8) disp24) + (if condbit + (sequence () + (set (reg h-gr 14) (add pc (const 4))) + (set pc disp24))) + ((m32rx (unit u-cti))) +) + +(dnmi bcl24r "relaxable bcl24" + (COND-CTI (MACH m32rx) RELAXED (IDOC BR)) + "bcl $disp24" + (emit bcl24 disp24) +) + +(dni bnc8 "bnc with 8 bit displacement" + (COND-CTI (PIPE O) (IDOC BR)) + "bnc.s $disp8" + (+ OP1_7 (f-r1 13) disp8) + (if (not condbit) (set pc disp8)) + ((m32r/d (unit u-cti)) + (m32rx (unit u-cti))) +) + +(dnmi bnc8r "relaxable bnc8" + (COND-CTI RELAXABLE (PIPE O) (IDOC BR)) + "bnc $disp8" + (emit bnc8 disp8) +) + +(dni bnc24 "bnc with 24 bit displacement" + (COND-CTI (IDOC BR)) + "bnc.l $disp24" + (+ OP1_15 (f-r1 13) disp24) + (if (not condbit) (set pc disp24)) + ((m32r/d (unit u-cti)) + (m32rx (unit u-cti))) +) + +(dnmi bnc24r "relaxable bnc24" + (COND-CTI RELAXED (IDOC BR)) + "bnc $disp24" + (emit bnc24 disp24) +) + +(dni bne "bne" + (COND-CTI (IDOC BR)) + "bne $src1,$src2,$disp16" + (+ OP1_11 OP2_1 src1 src2 disp16) + (if (ne src1 src2) (set pc disp16)) + ((m32r/d (unit u-cti) (unit u-cmp (cycles 0))) + (m32rx (unit u-cti) (unit u-cmp (cycles 0)))) +) + +(dni bra8 "bra with 8 bit displacement" + (UNCOND-CTI FILL-SLOT (PIPE O) (IDOC BR)) + "bra.s $disp8" + (+ OP1_7 (f-r1 15) disp8) + (set pc disp8) + ((m32r/d (unit u-cti)) + (m32rx (unit u-cti))) +) + +(dnmi bra8r "relaxable bra8" + (UNCOND-CTI FILL-SLOT RELAXABLE (PIPE O) (IDOC BR)) + "bra $disp8" + (emit bra8 disp8) +) + +(dni bra24 "bra with 24 displacement" + (UNCOND-CTI (IDOC BR)) + "bra.l $disp24" + (+ OP1_15 (f-r1 15) disp24) + (set pc disp24) + ((m32r/d (unit u-cti)) + (m32rx (unit u-cti))) +) + +(dnmi bra24r "relaxable bra24" + (UNCOND-CTI RELAXED (IDOC BR)) + "bra $disp24" + (emit bra24 disp24) +) + +(dni bncl8 "bncl with 8 bit displacement" + (COND-CTI FILL-SLOT (MACH m32rx) (PIPE O) (IDOC BR)) + "bncl.s $disp8" + (+ OP1_7 (f-r1 9) disp8) + (if (not condbit) + (sequence () + (set (reg h-gr 14) + (add (and pc (const -4)) + (const 4))) + (set pc disp8))) + ((m32rx (unit u-cti))) +) + +(dnmi bncl8r "relaxable bncl8" + (COND-CTI FILL-SLOT (MACH m32rx) (PIPE O) RELAXABLE (IDOC BR)) + "bncl $disp8" + (emit bncl8 disp8) +) + +(dni bncl24 "bncl with 24 bit displacement" + (COND-CTI (MACH m32rx) (IDOC BR)) + "bncl.l $disp24" + (+ OP1_15 (f-r1 9) disp24) + (if (not condbit) + (sequence () + (set (reg h-gr 14) (add pc (const 4))) + (set pc disp24))) + ((m32rx (unit u-cti))) +) + +(dnmi bncl24r "relaxable bncl24" + (COND-CTI (MACH m32rx) RELAXED (IDOC BR)) + "bncl $disp24" + (emit bncl24 disp24) +) + +(dni cmp "cmp" + ((PIPE OS) (IDOC ALU)) + "cmp $src1,$src2" + (+ OP1_0 OP2_4 src1 src2) + (set condbit (lt src1 src2)) + ((m32r/d (unit u-cmp)) + (m32rx (unit u-cmp))) +) + +(dni cmpi "cmpi" + ((IDOC ALU)) + "cmpi $src2,$simm16" + (+ OP1_8 (f-r1 0) OP2_4 src2 simm16) + (set condbit (lt src2 simm16)) + ((m32r/d (unit u-cmp)) + (m32rx (unit u-cmp))) +) + +(dni cmpu "cmpu" + ((PIPE OS) (IDOC ALU)) + "cmpu $src1,$src2" + (+ OP1_0 OP2_5 src1 src2) + (set condbit (ltu src1 src2)) + ((m32r/d (unit u-cmp)) + (m32rx (unit u-cmp))) +) + +(dni cmpui "cmpui" + ((IDOC ALU)) + "cmpui $src2,$simm16" + (+ OP1_8 (f-r1 0) OP2_5 src2 simm16) + (set condbit (ltu src2 simm16)) + ((m32r/d (unit u-cmp)) + (m32rx (unit u-cmp))) +) + +(dni cmpeq "cmpeq" + ((MACH m32rx) (PIPE OS) (IDOC ALU)) + "cmpeq $src1,$src2" + (+ OP1_0 OP2_6 src1 src2) + (set condbit (eq src1 src2)) + ((m32rx (unit u-cmp))) +) + +(dni cmpz "cmpz" + ((MACH m32rx) (PIPE OS) (IDOC ALU)) + "cmpz $src2" + (+ OP1_0 OP2_7 (f-r1 0) src2) + (set condbit (eq src2 (const 0))) + ((m32rx (unit u-cmp))) +) + +(dni div "div" + ((IDOC ALU)) + "div $dr,$sr" + (+ OP1_9 OP2_0 dr sr (f-simm16 0)) + (if (ne sr (const 0)) (set dr (div dr sr))) + ((m32r/d (unit u-exec (cycles 37))) + (m32rx (unit u-exec (cycles 37)))) +) + +(dni divu "divu" + ((IDOC ALU)) + "divu $dr,$sr" + (+ OP1_9 OP2_1 dr sr (f-simm16 0)) + (if (ne sr (const 0)) (set dr (udiv dr sr))) + ((m32r/d (unit u-exec (cycles 37))) + (m32rx (unit u-exec (cycles 37)))) +) + +(dni rem "rem" + ((IDOC ALU)) + "rem $dr,$sr" + (+ OP1_9 OP2_2 dr sr (f-simm16 0)) + ; FIXME: Check rounding direction. + (if (ne sr (const 0)) (set dr (mod dr sr))) + ((m32r/d (unit u-exec (cycles 37))) + (m32rx (unit u-exec (cycles 37)))) +) + +(dni remu "remu" + ((IDOC ALU)) + "remu $dr,$sr" + (+ OP1_9 OP2_3 dr sr (f-simm16 0)) + ; FIXME: Check rounding direction. + (if (ne sr (const 0)) (set dr (umod dr sr))) + ((m32r/d (unit u-exec (cycles 37))) + (m32rx (unit u-exec (cycles 37)))) +) + +(dni divh "divh" + ((MACH m32rx) (IDOC ALU)) + "divh $dr,$sr" + (+ OP1_9 OP2_0 dr sr (f-simm16 #x10)) + (if (ne sr (const 0)) (set dr (div (ext WI (trunc HI dr)) sr))) + ((m32rx (unit u-exec (cycles 21)))) +) + +(dni jc "jc" + (COND-CTI (MACH m32rx) (PIPE O) SPECIAL (IDOC BR)) + "jc $sr" + (+ OP1_1 (f-r1 12) OP2_12 sr) + (if condbit (set pc (and sr (const -4)))) + ((m32rx (unit u-cti))) +) + +(dni jnc "jnc" + (COND-CTI (MACH m32rx) (PIPE O) SPECIAL (IDOC BR)) + "jnc $sr" + (+ OP1_1 (f-r1 13) OP2_12 sr) + (if (not condbit) (set pc (and sr (const -4)))) + ((m32rx (unit u-cti))) +) + +(dni jl "jl" + (UNCOND-CTI FILL-SLOT (PIPE O) (IDOC BR)) + "jl $sr" + (+ OP1_1 (f-r1 14) OP2_12 sr) + (parallel () + (set (reg h-gr 14) + (add (and pc (const -4)) (const 4))) + (set pc (and sr (const -4)))) + ((m32r/d (unit u-cti)) + (m32rx (unit u-cti))) +) + +(dni jmp "jmp" + (UNCOND-CTI (PIPE O) (IDOC BR)) + "jmp $sr" + (+ OP1_1 (f-r1 15) OP2_12 sr) + (set pc (and sr (const -4))) + ; The above works now so this kludge has been commented out. + ; It's kept around because the f-r1 reference in the semantic part + ; should work. + ; FIXME: kludge, instruction decoding not finished. + ; But this should work, so that's another FIXME. + ;(sequence VOID (if VOID (eq SI f-r1 (const SI 14)) + ; FIXME: abuf->insn should be a macro of some sort. + ;(sequence VOID + ; (if VOID (eq SI (c-code SI "((abuf->insn >> 8) & 15)") + ; (const SI 14)) + ; (set WI (reg WI h-gr 14) + ; (add WI (and WI pc (const WI -4)) (const WI 4)))) + ; (set WI pc sr)) + ((m32r/d (unit u-cti)) + (m32rx (unit u-cti))) +) + +(define-pmacro (no-ext-expr mode expr) expr) +(define-pmacro (ext-expr mode expr) (ext mode expr)) +(define-pmacro (zext-expr mode expr) (zext mode expr)) + +(define-pmacro (load-op suffix op2-op mode ext-op) + (begin + (dni (.sym ld suffix) (.str "ld" suffix) + ((PIPE O) (IDOC MEM)) + (.str "ld" suffix " $dr,@$sr") + (+ OP1_2 op2-op dr sr) + (set dr (ext-op WI (mem mode sr))) + ((m32r/d (unit u-load)) + (m32rx (unit u-load))) + ) + (dnmi (.sym ld suffix "-2") (.str "ld" suffix "-2") + (NO-DIS (PIPE O) (IDOC MEM)) + (.str "ld" suffix " $dr,@($sr)") + (emit (.sym ld suffix) dr sr)) + (dni (.sym ld suffix -d) (.str "ld" suffix "-d") + ((IDOC MEM)) + (.str "ld" suffix " $dr,@($slo16,$sr)") + (+ OP1_10 op2-op dr sr slo16) + (set dr (ext-op WI (mem mode (add sr slo16)))) + ((m32r/d (unit u-load (cycles 2))) + (m32rx (unit u-load (cycles 2)))) + ) + (dnmi (.sym ld suffix -d2) (.str "ld" suffix "-d2") + (NO-DIS (IDOC MEM)) + (.str "ld" suffix " $dr,@($sr,$slo16)") + (emit (.sym ld suffix -d) dr sr slo16)) + ) +) +(load-op "" OP2_12 WI no-ext-expr) +(load-op b OP2_8 QI ext-expr) +(load-op h OP2_10 HI ext-expr) +(load-op ub OP2_9 QI zext-expr) +(load-op uh OP2_11 HI zext-expr) + +(dni ld-plus "ld+" + ((PIPE O) (IDOC MEM)) + "ld $dr,@$sr+" + (+ OP1_2 dr OP2_14 sr) + (parallel () + ; wip: memory addresses in profiling support + ;(set dr (name ld-mem (mem WI sr))) + (set dr (mem WI sr)) + (set sr (add sr (const 4)))) + ; Note: `pred' is the constraint. Also useful here is (ref name) + ; and returns true if operand <name> was referenced + ; (where "referenced" means _read_ if input operand and _written_ if + ; output operand). + ; args to unit are "unit-name (name1 value1) ..." + ; - cycles(done),issue,pred are also specified this way + ; - if unspecified, default is used + ; - for ins/outs, extra arg is passed that says what was specified + ; - this is AND'd with `written' for outs + ((m32r/d (unit u-load (pred (const 1))) + (unit u-exec (in sr #f) (in dr sr) (out dr sr) (cycles 0) (pred (const 1)))) + (m32rx (unit u-load) + (unit u-exec (in sr #f) (in dr sr) (out dr sr) (cycles 0) (pred (const 1)))) + ) +) + +(dnmi pop "pop" + ((IDOC MEM)) + "pop $dr" + (emit ld-plus dr (sr 15)) ; "ld %0,@sp+" +) + +(dni ld24 "ld24" + ((IDOC MEM)) + "ld24 $dr,$uimm24" + (+ OP1_14 dr uimm24) + (set dr uimm24) + () +) + +; ldi8 appears before ldi16 so we try the shorter version first + +(dni ldi8 "ldi8" + ((PIPE OS) (IDOC ALU)) + "ldi8 $dr,$simm8" + (+ OP1_6 dr simm8) + (set dr simm8) + () +) + +(dnmi ldi8a "ldi8 alias" + ((PIPE OS) (IDOC ALU)) + "ldi $dr,$simm8" + (emit ldi8 dr simm8) +) + +(dni ldi16 "ldi16" + ((IDOC ALU)) + "ldi16 $dr,$hash$slo16" + (+ OP1_9 OP2_15 (f-r2 0) dr slo16) + (set dr slo16) + () +) + +(dnmi ldi16a "ldi16 alias" + ((IDOC ALU)) + "ldi $dr,$hash$slo16" + (emit ldi16 dr slo16) +) + +(dni lock "lock" + ((PIPE O) (IDOC MISC)) + "lock $dr,@$sr" + (+ OP1_2 OP2_13 dr sr) + (sequence () + (set (reg h-lock) (const BI 1)) + (set dr (mem WI sr))) + ((m32r/d (unit u-load)) + (m32rx (unit u-load))) +) + +(dni machi "machi" + ( + ; (MACH m32r) is a temporary hack. This insn collides with machi-a + ; in the simulator so disable it for m32rx. + (MACH m32r) (PIPE S) (IDOC MAC) + ) + "machi $src1,$src2" + (+ OP1_3 OP2_4 src1 src2) + ; FIXME: TRACE_RESULT will print the wrong thing since we + ; alter one of the arguments. + (set accum + (sra DI + (sll DI + (add DI + accum + (mul DI + (ext DI (and WI src1 (const #xffff0000))) + (ext DI (trunc HI (sra WI src2 (const 16)))))) + (const 8)) + (const 8))) + ((m32r/d (unit u-mac))) +) + +(dni machi-a "machi-a" + ((MACH m32rx) (PIPE S) (IDOC MAC)) + "machi $src1,$src2,$acc" + (+ OP1_3 src1 acc (f-op23 4) src2) + (set acc + (sra DI + (sll DI + (add DI + acc + (mul DI + (ext DI (and WI src1 (const #xffff0000))) + (ext DI (trunc HI (sra WI src2 (const 16)))))) + (const 8)) + (const 8))) + ((m32rx (unit u-mac))) +) + +(dni maclo "maclo" + ((MACH m32r) (PIPE S) (IDOC MAC)) + "maclo $src1,$src2" + (+ OP1_3 OP2_5 src1 src2) + (set accum + (sra DI + (sll DI + (add DI + accum + (mul DI + (ext DI (sll WI src1 (const 16))) + (ext DI (trunc HI src2)))) + (const 8)) + (const 8))) + ((m32r/d (unit u-mac))) +) + +(dni maclo-a "maclo-a" + ((MACH m32rx) (PIPE S) (IDOC MAC)) + "maclo $src1,$src2,$acc" + (+ OP1_3 src1 acc (f-op23 5) src2) + (set acc + (sra DI + (sll DI + (add DI + acc + (mul DI + (ext DI (sll WI src1 (const 16))) + (ext DI (trunc HI src2)))) + (const 8)) + (const 8))) + ((m32rx (unit u-mac))) +) + +(dni macwhi "macwhi" + ((MACH m32r) (PIPE S) (IDOC MAC)) + "macwhi $src1,$src2" + (+ OP1_3 OP2_6 src1 src2) + (set accum + (sra DI + (sll DI + (add DI + accum + (mul DI + (ext DI src1) + (ext DI (trunc HI (sra WI src2 (const 16)))))) + (const 8)) + (const 8))) + ((m32r/d (unit u-mac))) +) + +(dni macwhi-a "macwhi-a" + ((MACH m32rx) (PIPE S) SPECIAL (IDOC MAC)) + "macwhi $src1,$src2,$acc" + (+ OP1_3 src1 acc (f-op23 6) src2) + ; Note that this doesn't do the sign extension, which is correct. + (set acc + (add acc + (mul (ext DI src1) + (ext DI (trunc HI (sra src2 (const 16))))))) + ((m32rx (unit u-mac))) +) + +(dni macwlo "macwlo" + ((MACH m32r) (PIPE S) (IDOC MAC)) + "macwlo $src1,$src2" + (+ OP1_3 OP2_7 src1 src2) + (set accum + (sra DI + (sll DI + (add DI + accum + (mul DI + (ext DI src1) + (ext DI (trunc HI src2)))) + (const 8)) + (const 8))) + ((m32r/d (unit u-mac))) +) + +(dni macwlo-a "macwlo-a" + ((MACH m32rx) (PIPE S) SPECIAL (IDOC MAC)) + "macwlo $src1,$src2,$acc" + (+ OP1_3 src1 acc (f-op23 7) src2) + ; Note that this doesn't do the sign extension, which is correct. + (set acc + (add acc + (mul (ext DI src1) + (ext DI (trunc HI src2))))) + ((m32rx (unit u-mac))) +) + +(dni mul "mul" + ((PIPE S) (IDOC ALU)) + "mul $dr,$sr" + (+ OP1_1 OP2_6 dr sr) + (set dr (mul dr sr)) + ((m32r/d (unit u-exec (cycles 4))) + (m32rx (unit u-exec (cycles 4)))) +) + +(dni mulhi "mulhi" + ((MACH m32r) (PIPE S) (IDOC ACCUM)) + "mulhi $src1,$src2" + (+ OP1_3 OP2_0 src1 src2) + (set accum + (sra DI + (sll DI + (mul DI + (ext DI (and WI src1 (const #xffff0000))) + (ext DI (trunc HI (sra WI src2 (const 16))))) + (const 16)) + (const 16))) + ((m32r/d (unit u-mac))) +) + +(dni mulhi-a "mulhi-a" + ((MACH m32rx) (PIPE S) (IDOC ACCUM)) + "mulhi $src1,$src2,$acc" + (+ OP1_3 (f-op23 0) src1 acc src2) + (set acc + (sra DI + (sll DI + (mul DI + (ext DI (and WI src1 (const #xffff0000))) + (ext DI (trunc HI (sra WI src2 (const 16))))) + (const 16)) + (const 16))) + ((m32rx (unit u-mac))) +) + +(dni mullo "mullo" + ((MACH m32r) (PIPE S) (IDOC ACCUM)) + "mullo $src1,$src2" + (+ OP1_3 OP2_1 src1 src2) + (set accum + (sra DI + (sll DI + (mul DI + (ext DI (sll WI src1 (const 16))) + (ext DI (trunc HI src2))) + (const 16)) + (const 16))) + ((m32r/d (unit u-mac))) +) + +(dni mullo-a "mullo-a" + ((MACH m32rx) (PIPE S) (IDOC ACCUM)) + "mullo $src1,$src2,$acc" + (+ OP1_3 src1 acc (f-op23 1) src2) + (set acc + (sra DI + (sll DI + (mul DI + (ext DI (sll WI src1 (const 16))) + (ext DI (trunc HI src2))) + (const 16)) + (const 16))) + ((m32rx (unit u-mac))) +) + +(dni mulwhi "mulwhi" + ((MACH m32r) (PIPE S) (IDOC ACCUM)) + "mulwhi $src1,$src2" + (+ OP1_3 OP2_2 src1 src2) + (set accum + (sra DI + (sll DI + (mul DI + (ext DI src1) + (ext DI (trunc HI (sra WI src2 (const 16))))) + (const 8)) + (const 8))) + ((m32r/d (unit u-mac))) +) + +(dni mulwhi-a "mulwhi-a" + ((MACH m32rx) (PIPE S) SPECIAL (IDOC ACCUM)) + "mulwhi $src1,$src2,$acc" + (+ OP1_3 src1 acc (f-op23 2) src2) + ; Note that this doesn't do the sign extension, which is correct. + (set acc + (mul (ext DI src1) + (ext DI (trunc HI (sra src2 (const 16)))))) + ((m32rx (unit u-mac))) +) + +(dni mulwlo "mulwlo" + ((MACH m32r) (PIPE S) (IDOC ACCUM)) + "mulwlo $src1,$src2" + (+ OP1_3 OP2_3 src1 src2) + (set accum + (sra DI + (sll DI + (mul DI + (ext DI src1) + (ext DI (trunc HI src2))) + (const 8)) + (const 8))) + ((m32r/d (unit u-mac))) +) + +(dni mulwlo-a "mulwlo-a" + ((MACH m32rx) (PIPE S) SPECIAL (IDOC ACCUM)) + "mulwlo $src1,$src2,$acc" + (+ OP1_3 src1 acc (f-op23 3) src2) + ; Note that this doesn't do the sign extension, which is correct. + (set acc + (mul (ext DI src1) + (ext DI (trunc HI src2)))) + ((m32rx (unit u-mac))) +) + +(dni mv "mv" + ((PIPE OS) (IDOC ALU)) + "mv $dr,$sr" + (+ OP1_1 OP2_8 dr sr) + (set dr sr) + () +) + +(dni mvfachi "mvfachi" + ((MACH m32r) (PIPE S) (IDOC ACCUM)) + "mvfachi $dr" + (+ OP1_5 OP2_15 (f-r2 0) dr) + (set dr (trunc WI (sra DI accum (const 32)))) + ((m32r/d (unit u-exec (cycles 2)))) +) + +(dni mvfachi-a "mvfachi-a" + ((MACH m32rx) (PIPE S) (IDOC ACCUM)) + "mvfachi $dr,$accs" + (+ OP1_5 dr OP2_15 accs (f-op3 0)) + (set dr (trunc WI (sra DI accs (const 32)))) + ((m32rx (unit u-exec (cycles 2)))) +) + +(dni mvfaclo "mvfaclo" + ((MACH m32r) (PIPE S) (IDOC ACCUM)) + "mvfaclo $dr" + (+ OP1_5 OP2_15 (f-r2 1) dr) + (set dr (trunc WI accum)) + ((m32r/d (unit u-exec (cycles 2)))) +) + +(dni mvfaclo-a "mvfaclo-a" + ((MACH m32rx) (PIPE S) (IDOC ACCUM)) + "mvfaclo $dr,$accs" + (+ OP1_5 dr OP2_15 accs (f-op3 1)) + (set dr (trunc WI accs)) + ((m32rx (unit u-exec (cycles 2)))) +) + +(dni mvfacmi "mvfacmi" + ((MACH m32r) (PIPE S) (IDOC ACCUM)) + "mvfacmi $dr" + (+ OP1_5 OP2_15 (f-r2 2) dr) + (set dr (trunc WI (sra DI accum (const 16)))) + ((m32r/d (unit u-exec (cycles 2)))) +) + +(dni mvfacmi-a "mvfacmi-a" + ((MACH m32rx) (PIPE S) (IDOC ACCUM)) + "mvfacmi $dr,$accs" + (+ OP1_5 dr OP2_15 accs (f-op3 2)) + (set dr (trunc WI (sra DI accs (const 16)))) + ((m32rx (unit u-exec (cycles 2)))) +) + +(dni mvfc "mvfc" + ((PIPE O) (IDOC MISC)) + "mvfc $dr,$scr" + (+ OP1_1 OP2_9 dr scr) + (set dr scr) + () +) + +(dni mvtachi "mvtachi" + ((MACH m32r) (PIPE S) (IDOC ACCUM)) + "mvtachi $src1" + (+ OP1_5 OP2_7 (f-r2 0) src1) + (set accum + (or DI + (and DI accum (const DI #xffffffff)) + (sll DI (ext DI src1) (const 32)))) + ((m32r/d (unit u-exec (in sr src1)))) +) + +(dni mvtachi-a "mvtachi-a" + ((MACH m32rx) (PIPE S) (IDOC ACCUM)) + "mvtachi $src1,$accs" + (+ OP1_5 src1 OP2_7 accs (f-op3 0)) + (set accs + (or DI + (and DI accs (const DI #xffffffff)) + (sll DI (ext DI src1) (const 32)))) + ((m32rx (unit u-exec (in sr src1)))) +) + +(dni mvtaclo "mvtaclo" + ((MACH m32r) (PIPE S) (IDOC ACCUM)) + "mvtaclo $src1" + (+ OP1_5 OP2_7 (f-r2 1) src1) + (set accum + (or DI + (and DI accum (const DI #xffffffff00000000)) + (zext DI src1))) + ((m32r/d (unit u-exec (in sr src1)))) +) + +(dni mvtaclo-a "mvtaclo-a" + ((MACH m32rx) (PIPE S) (IDOC ACCUM)) + "mvtaclo $src1,$accs" + (+ OP1_5 src1 OP2_7 accs (f-op3 1)) + (set accs + (or DI + (and DI accs (const DI #xffffffff00000000)) + (zext DI src1))) + ((m32rx (unit u-exec (in sr src1)))) +) + +(dni mvtc "mvtc" + ((PIPE O) (IDOC MISC)) + "mvtc $sr,$dcr" + (+ OP1_1 OP2_10 dcr sr) + (set dcr sr) + () +) + +(dni neg "neg" + ((PIPE OS) (IDOC ALU)) + "neg $dr,$sr" + (+ OP1_0 OP2_3 dr sr) + (set dr (neg sr)) + () +) + +(dni nop "nop" + ((PIPE OS) (IDOC MISC)) + "nop" + (+ OP1_7 OP2_0 (f-r1 0) (f-r2 0)) + (c-code VOID "PROFILE_COUNT_FILLNOPS (current_cpu, abuf->addr);\n") + ; FIXME: quick hack: parallel nops don't contribute to cycle count. + ; Other kinds of nops do however (which we currently ignore). + ((m32r/d (unit u-exec (cycles 0))) + (m32rx (unit u-exec (cycles 0)))) +) + +(dni not "not" + ((PIPE OS) (IDOC ALU)) + "not $dr,$sr" + (+ OP1_0 OP2_11 dr sr) + (set dr (inv sr)) + () +) + +(dni rac "rac" + ((MACH m32r) (PIPE S) (IDOC MAC)) + "rac" + (+ OP1_5 OP2_9 (f-r1 0) (f-r2 0)) + (sequence ((DI tmp1)) + (set tmp1 (sll DI accum (const 1))) + (set tmp1 (add DI tmp1 (const DI #x8000))) + (set accum + (cond DI + ((gt tmp1 (const DI #x00007fffffff0000)) + (const DI #x00007fffffff0000)) + ((lt tmp1 (const DI #xffff800000000000)) + (const DI #xffff800000000000)) + (else (and tmp1 (const DI #xffffffffffff0000))))) + ) + ((m32r/d (unit u-mac))) +) + +(dni rac-dsi "rac-dsi" + ((MACH m32rx) (PIPE S) (IDOC MAC)) + "rac $accd,$accs,$imm1" + (+ OP1_5 accd (f-bits67 0) OP2_9 accs (f-bit14 0) imm1) + (sequence ((DI tmp1)) + (set tmp1 (sll accs imm1)) + (set tmp1 (add tmp1 (const DI #x8000))) + (set accd + (cond DI + ((gt tmp1 (const DI #x00007fffffff0000)) + (const DI #x00007fffffff0000)) + ((lt tmp1 (const DI #xffff800000000000)) + (const DI #xffff800000000000)) + (else (and tmp1 (const DI #xffffffffffff0000))))) + ) + ((m32rx (unit u-mac))) +) + +(dnmi rac-d "rac-d" + ((MACH m32rx) (PIPE S) (IDOC MAC)) + "rac $accd" + (emit rac-dsi accd (f-accs 0) (f-imm1 0)) +) + +(dnmi rac-ds "rac-ds" + ((MACH m32rx) (PIPE S) (IDOC MAC)) + "rac $accd,$accs" + (emit rac-dsi accd accs (f-imm1 0)) +) + + +(dni rach "rach" + ((MACH m32r) (PIPE S) (IDOC MAC)) + "rach" + (+ OP1_5 OP2_8 (f-r1 0) (f-r2 0)) + (sequence ((DI tmp1)) + ; Lop off top 8 bits. + ; The sign bit we want to use is bit 55 so the 64 bit value + ; isn't properly signed which we deal with in the if's below. + (set tmp1 (and accum (const DI #xffffffffffffff))) + (if (andif (ge tmp1 (const DI #x003fff80000000)) + (le tmp1 (const DI #x7fffffffffffff))) + (set tmp1 (const DI #x003fff80000000)) + ; else part + (if (andif (ge tmp1 (const DI #x80000000000000)) + (le tmp1 (const DI #xffc00000000000))) + (set tmp1 (const DI #xffc00000000000)) + (set tmp1 (and (add accum (const DI #x40000000)) + (const DI #xffffffff80000000))))) + (set tmp1 (sll tmp1 (const 1))) + ; Sign extend top 8 bits. + (set accum + ; FIXME: 7? + (sra DI (sll DI tmp1 (const 7)) (const 7))) + ) + ((m32r/d (unit u-mac))) +) + +(dni rach-dsi "rach-dsi" + ((MACH m32rx) (PIPE S) (IDOC MAC)) + "rach $accd,$accs,$imm1" + (+ OP1_5 accd (f-bits67 0) OP2_8 accs (f-bit14 0) imm1) + (sequence ((DI tmp1)) + (set tmp1 (sll accs imm1)) + (set tmp1 (add tmp1 (const DI #x80000000))) + (set accd + (cond DI + ((gt tmp1 (const DI #x00007fff00000000)) + (const DI #x00007fff00000000)) + ((lt tmp1 (const DI #xffff800000000000)) + (const DI #xffff800000000000)) + (else (and tmp1 (const DI #xffffffff00000000))))) + ) + ((m32rx (unit u-mac))) +) + +(dnmi rach-d "rach-d" + ((MACH m32rx) (PIPE S) (IDOC MAC)) + "rach $accd" + (emit rach-dsi accd (f-accs 0) (f-imm1 0)) +) + +(dnmi rach-ds "rach-ds" + ((MACH m32rx) (PIPE S) (IDOC MAC)) + "rach $accd,$accs" + (emit rach-dsi accd accs (f-imm1 0)) +) + +(dni rte "rte" + (UNCOND-CTI (PIPE O) (IDOC BR)) + "rte" + (+ OP1_1 OP2_13 (f-r1 0) (f-r2 6)) + (sequence () + ; pc = bpc & -4 + (set pc (and (reg h-cr 6) (const -4))) + ; bpc = bbpc + (set (reg h-cr 6) (reg h-cr 14)) + ; psw = bpsw + (set (reg h-psw) (reg h-bpsw)) + ; bpsw = bbpsw + (set (reg h-bpsw) (reg h-bbpsw)) + ) + () +) + +(dni seth "seth" + ((IDOC ALU)) + "seth $dr,$hash$hi16" + (+ OP1_13 OP2_12 dr (f-r2 0) hi16) + (set dr (sll WI hi16 (const 16))) + () +) + +(define-pmacro (shift-op sym op2-r-op op2-3-op op2-i-op sem-op) + (begin + (dni sym sym ((PIPE O) (IDOC ALU)) + (.str sym " $dr,$sr") + (+ OP1_1 op2-r-op dr sr) + (set dr (sem-op dr (and sr (const 31)))) + () + ) + (dni (.sym sym "3") sym ((IDOC ALU)) + (.str sym "3 $dr,$sr,$simm16") + (+ OP1_9 op2-3-op dr sr simm16) + (set dr (sem-op sr (and WI simm16 (const 31)))) + () + ) + (dni (.sym sym "i") sym ((PIPE O) (IDOC ALU)) + (.str sym "i $dr,$uimm5") + (+ OP1_5 (f-shift-op2 op2-i-op) dr uimm5) + (set dr (sem-op dr uimm5)) + () + ) + ) +) +(shift-op sll OP2_4 OP2_12 2 sll) +(shift-op sra OP2_2 OP2_10 1 sra) +(shift-op srl OP2_0 OP2_8 0 srl) + +(define-pmacro (store-op suffix op2-op mode) + (begin + (dni (.sym st suffix) (.str "st" suffix) + ((PIPE O) (IDOC MEM)) + (.str "st" suffix " $src1,@$src2") + (+ OP1_2 op2-op src1 src2) + (set mode (mem mode src2) src1) + ((m32r/d (unit u-store (cycles 1))) + (m32rx (unit u-store (cycles 1)))) + ) + (dnmi (.sym st suffix "-2") (.str "st" suffix "-2") + (NO-DIS (PIPE O) (IDOC MEM)) + (.str "st" suffix " $src1,@($src2)") + (emit (.sym st suffix) src1 src2)) + (dni (.sym st suffix -d) (.str "st" suffix "-d") + ((IDOC MEM)) + (.str "st" suffix " $src1,@($slo16,$src2)") + (+ OP1_10 op2-op src1 src2 slo16) + (set mode (mem mode (add src2 slo16)) src1) + ((m32r/d (unit u-store (cycles 2))) + (m32rx (unit u-store (cycles 2)))) + ) + (dnmi (.sym st suffix -d2) (.str "st" suffix "-d2") + (NO-DIS (IDOC MEM)) + (.str "st" suffix " $src1,@($src2,$slo16)") + (emit (.sym st suffix -d) src1 src2 slo16)) + ) +) +(store-op "" OP2_4 WI) +(store-op b OP2_0 QI) +(store-op h OP2_2 HI) + +(dni st-plus "st+" + ((PIPE O) (IDOC MEM)) + "st $src1,@+$src2" + (+ OP1_2 OP2_6 src1 src2) + ; This has to be coded carefully to avoid an "earlyclobber" of src2. + (sequence ((WI new-src2)) + (set new-src2 (add WI src2 (const WI 4))) + (set (mem WI new-src2) src1) + (set src2 new-src2)) + ((m32r/d (unit u-store) + (unit u-exec (in dr src2) (out dr src2) (cycles 0))) + (m32rx (unit u-store) + (unit u-exec (in dr src2) (out dr src2) (cycles 0))) + ) +) + +(dni st-minus "st-" + ((PIPE O) (IDOC MEM)) + "st $src1,@-$src2" + (+ OP1_2 OP2_7 src1 src2) + ; This is the original way. It doesn't work for parallel execution + ; because of the earlyclobber of src2. + ;(sequence () + ; (set src2 (sub src2 (const 4))) + ; (set (mem WI src2) src1)) + (sequence ((WI new-src2)) + (set new-src2 (sub src2 (const 4))) + (set (mem WI new-src2) src1) + (set src2 new-src2)) + ((m32r/d (unit u-store) + (unit u-exec (in dr src2) (out dr src2) (cycles 0))) + (m32rx (unit u-store) + (unit u-exec (in dr src2) (out dr src2) (cycles 0))) + ) +) + +(dnmi push "push" ((IDOC MEM)) + "push $src1" + (emit st-minus src1 (src2 15)) ; "st %0,@-sp" +) + +(dni sub "sub" + ((PIPE OS) (IDOC ALU)) + "sub $dr,$sr" + (+ OP1_0 OP2_2 dr sr) + (set dr (sub dr sr)) + () +) + +(dni subv "sub:rv" + ((PIPE OS) (IDOC ALU)) + "subv $dr,$sr" + (+ OP1_0 OP2_0 dr sr) + (parallel () + (set dr (sub dr sr)) + (set condbit (sub-oflag dr sr (const 0)))) + () +) + +(dni subx "sub:rx" + ((PIPE OS) (IDOC ALU)) + "subx $dr,$sr" + (+ OP1_0 OP2_1 dr sr) + (parallel () + (set dr (subc dr sr condbit)) + (set condbit (sub-cflag dr sr condbit))) + () +) + +(dni trap "trap" + (UNCOND-CTI FILL-SLOT (PIPE O) (IDOC MISC)) + "trap $uimm4" + (+ OP1_1 OP2_15 (f-r1 0) uimm4) + (sequence () + ; bbpc = bpc + (set (reg h-cr 14) (reg h-cr 6)) + ; Set bpc to the return address. Actually it's not quite the + ; return address as RTE rounds the address down to a word + ; boundary. + (set (reg h-cr 6) (add pc (const 4))) + ; bbpsw = bpsw + (set (reg h-bbpsw) (reg h-bpsw)) + ; bpsw = psw + (set (reg h-bpsw) (reg h-psw)) + ; sm is unchanged, ie,c are set to zero. + (set (reg h-psw) (and (reg h-psw) (const #x80))) + ; m32r_trap handles operating vs user mode + (set WI pc (c-call WI "m32r_trap" pc uimm4)) + ) + () +) + +(dni unlock "unlock" + ((PIPE O) (IDOC MISC)) + "unlock $src1,@$src2" + (+ OP1_2 OP2_5 src1 src2) + (sequence () + (if (reg h-lock) + (set (mem WI src2) src1)) + (set (reg h-lock) (const BI 0))) + ((m32r/d (unit u-load)) + (m32rx (unit u-load))) +) + +; Saturate into byte. +(dni satb "satb" + ((MACH m32rx) (IDOC ALU)) + "satb $dr,$sr" + (+ OP1_8 dr OP2_6 sr (f-uimm16 #x0300)) + (set dr + ; FIXME: min/max would simplify this nicely of course. + (cond WI + ((ge sr (const 127)) (const 127)) + ((le sr (const -128)) (const -128)) + (else sr))) + () +) + +; Saturate into half word. +(dni sath "sath" + ((MACH m32rx) (IDOC ALU)) + "sath $dr,$sr" + (+ OP1_8 dr OP2_6 sr (f-uimm16 #x0200)) + (set dr + (cond WI + ((ge sr (const 32767)) (const 32767)) + ((le sr (const -32768)) (const -32768)) + (else sr))) + () +) + +; Saturate word. +(dni sat "sat" + ((MACH m32rx) SPECIAL (IDOC ALU)) + "sat $dr,$sr" + (+ OP1_8 dr OP2_6 sr (f-uimm16 0)) + (set dr + (if WI condbit + (if WI (lt sr (const 0)) + (const #x7fffffff) + (const #x80000000)) + sr)) + () +) + +; Parallel compare byte zeros. +; Set C bit in condition register if any byte in source register is zero. +(dni pcmpbz "pcmpbz" + ((MACH m32rx) (PIPE OS) SPECIAL (IDOC ALU)) + "pcmpbz $src2" + (+ OP1_0 (f-r1 3) OP2_7 src2) + (set condbit + (cond BI + ((eq (and src2 (const #xff)) (const 0)) (const BI 1)) + ((eq (and src2 (const #xff00)) (const 0)) (const BI 1)) + ((eq (and src2 (const #xff0000)) (const 0)) (const BI 1)) + ((eq (and src2 (const #xff000000)) (const 0)) (const BI 1)) + (else (const BI 0)))) + ((m32rx (unit u-cmp))) +) + +; Add accumulators +(dni sadd "sadd" + ((MACH m32rx) (PIPE S) (IDOC ACCUM)) + "sadd" + (+ OP1_5 (f-r1 0) OP2_14 (f-r2 4)) + (set (reg h-accums 0) + (add (sra (reg h-accums 1) (const 16)) + (reg h-accums 0))) + ((m32rx (unit u-mac))) +) + +; Multiply and add into accumulator 1 +(dni macwu1 "macwu1" + ((MACH m32rx) (PIPE S) (IDOC MAC)) + "macwu1 $src1,$src2" + (+ OP1_5 src1 OP2_11 src2) + (set (reg h-accums 1) + (sra DI + (sll DI + (add DI + (reg h-accums 1) + (mul DI + (ext DI src1) + (ext DI (and src2 (const #xffff))))) + (const 8)) + (const 8))) + ((m32rx (unit u-mac))) +) + +; Multiply and subtract from accumulator 0 +(dni msblo "msblo" + ((MACH m32rx) (PIPE S) (IDOC MAC)) + "msblo $src1,$src2" + (+ OP1_5 src1 OP2_13 src2) + (set accum + (sra DI + (sll DI + (sub accum + (sra DI + (sll DI + (mul DI + (ext DI (trunc HI src1)) + (ext DI (trunc HI src2))) + (const 32)) + (const 16))) + (const 8)) + (const 8))) + ((m32rx (unit u-mac))) +) + +; Multiply into accumulator 1 +(dni mulwu1 "mulwu1" + ((MACH m32rx) (PIPE S) (IDOC MAC)) + "mulwu1 $src1,$src2" + (+ OP1_5 src1 OP2_10 src2) + (set (reg h-accums 1) + (sra DI + (sll DI + (mul DI + (ext DI src1) + (ext DI (and src2 (const #xffff)))) + (const 16)) + (const 16))) + ((m32rx (unit u-mac))) +) + +; Multiply and add into accumulator 1 +(dni maclh1 "maclh1" + ((MACH m32rx) (PIPE S) (IDOC MAC)) + "maclh1 $src1,$src2" + (+ OP1_5 src1 OP2_12 src2) + (set (reg h-accums 1) + (sra DI + (sll DI + (add DI + (reg h-accums 1) + (sll DI + (ext DI + (mul SI + (ext SI (trunc HI src1)) + (sra SI src2 (const SI 16)))) + (const 16))) + (const 8)) + (const 8))) + ((m32rx (unit u-mac))) +) + +; skip instruction if C +(dni sc "sc" + ((MACH m32rx) (PIPE O) SPECIAL (IDOC BR)) + "sc" + (+ OP1_7 (f-r1 4) OP2_0 (f-r2 1)) + (skip (zext INT condbit)) + () +) + +; skip instruction if not C +(dni snc "snc" + ((MACH m32rx) (PIPE O) SPECIAL (IDOC BR)) + "snc" + (+ OP1_7 (f-r1 5) OP2_0 (f-r2 1)) + (skip (zext INT (not condbit))) + () +) |