languages/go
languages/go
1
0
Fork 0
mirror of https://github.com/golang/go synced 2024-11-05 18:36:08 +00:00
Code Issues Projects Releases Packages Wiki Activity

cmd/compile: move SSA rotate instruction detection to arch-independent rules

Browse source 
Detect rotate instructions while still in architecture-independent form.
It's easier to do here, and we don't need to repeat it in each
architecture file.

Change-Id: I9396954b3f3b3bfb96c160d064a02002309935bb
Reviewed-on: https://go-review.googlesource.com/c/go/+/421195
TryBot-Result: Gopher Robot <gobot@golang.org>
Reviewed-by: Eric Fang <eric.fang@arm.com>
Reviewed-by: Cuong Manh Le <cuong.manhle.vn@gmail.com>
Reviewed-by: David Chase <drchase@google.com>
Reviewed-by: Joedian Reid <joedian@golang.org>
Reviewed-by: Ruinan Sun <Ruinan.Sun@arm.com>
Run-TryBot: Keith Randall <khr@golang.org>
This commit is contained in:
Keith Randall 2022-08-03 22:58:30 -07:00 committed by Keith Randall
parent a36a0c440e
commit 60ad3c48f5
23 changed files with 5885 additions and 4531 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

29
src/cmd/compile/internal/ssa/gen/386.rules
View file

@ -148,10 +148,14 @@
(Rsh16x64 x (Const64 [c])) && uint64(c) >= 16 => (SARWconst x [15])
(Rsh8x64 x (Const64 [c])) && uint64(c) >= 8 => (SARBconst x [7])
// rotates
(RotateLeft32 ...) => (ROLL ...)
(RotateLeft16 ...) => (ROLW ...)
(RotateLeft8 ...) => (ROLB ...)
// constant rotates
(RotateLeft32 x (MOVLconst [c])) => (ROLLconst [c&31] x)
(RotateLeft16 x (MOVLconst [c])) => (ROLWconst [int16(c&15)] x)
(RotateLeft8 x (MOVLconst [c])) => (ROLBconst [int8(c&7)] x)
(ROLL x (MOVLconst [c])) => (ROLLconst [c&31] x)
(ROLW x (MOVLconst [c])) => (ROLWconst [int16(c&15)] x)
(ROLB x (MOVLconst [c])) => (ROLBconst [int8(c&7)] x)
// Lowering comparisons
(Less32 x y) => (SETL (CMPL x y))
@ -425,29 +429,10 @@
// Rotate instructions
(ADDL (SHLLconst [c] x) (SHRLconst [d] x)) && d == 32-c => (ROLLconst [c] x)
( ORL (SHLLconst [c] x) (SHRLconst [d] x)) && d == 32-c => (ROLLconst [c] x)
(XORL (SHLLconst [c] x) (SHRLconst [d] x)) && d == 32-c => (ROLLconst [c] x)
(ADDL <t> (SHLLconst x [c]) (SHRWconst x [d])) && c < 16 && d == int16(16-c) && t.Size() == 2
=> (ROLWconst x [int16(c)])
( ORL <t> (SHLLconst x [c]) (SHRWconst x [d])) && c < 16 && d == int16(16-c) && t.Size() == 2
=> (ROLWconst x [int16(c)])
(XORL <t> (SHLLconst x [c]) (SHRWconst x [d])) && c < 16 && d == int16(16-c) && t.Size() == 2
=> (ROLWconst x [int16(c)])
(ADDL <t> (SHLLconst x [c]) (SHRBconst x [d])) && c < 8 && d == int8(8-c) && t.Size() == 1
=> (ROLBconst x [int8(c)])
( ORL <t> (SHLLconst x [c]) (SHRBconst x [d])) && c < 8 && d == int8(8-c) && t.Size() == 1
=> (ROLBconst x [int8(c)])
(XORL <t> (SHLLconst x [c]) (SHRBconst x [d])) && c < 8 && d == int8(8-c) && t.Size() == 1
=> (ROLBconst x [int8(c)])
(ROLLconst [c] (ROLLconst [d] x)) => (ROLLconst [(c+d)&31] x)
(ROLWconst [c] (ROLWconst [d] x)) => (ROLWconst [(c+d)&15] x)
(ROLBconst [c] (ROLBconst [d] x)) => (ROLBconst [(c+d)& 7] x)
// Constant shift simplifications
(SHLLconst x [0]) => x

3
src/cmd/compile/internal/ssa/gen/386Ops.go
View file

@ -275,6 +275,9 @@ func init() {
{name: "SARWconst", argLength: 1, reg: gp11, asm: "SARW", aux: "Int16", resultInArg0: true, clobberFlags: true}, // signed arg0 >> auxint, shift amount 0-15
{name: "SARBconst", argLength: 1, reg: gp11, asm: "SARB", aux: "Int8", resultInArg0: true, clobberFlags: true}, // signed arg0 >> auxint, shift amount 0-7
{name: "ROLL", argLength: 2, reg: gp21shift, asm: "ROLL", resultInArg0: true, clobberFlags: true}, // 32 bits of arg0 rotate left by arg1
{name: "ROLW", argLength: 2, reg: gp21shift, asm: "ROLW", resultInArg0: true, clobberFlags: true}, // low 16 bits of arg0 rotate left by arg1
{name: "ROLB", argLength: 2, reg: gp21shift, asm: "ROLB", resultInArg0: true, clobberFlags: true}, // low 8 bits of arg0 rotate left by arg1
{name: "ROLLconst", argLength: 1, reg: gp11, asm: "ROLL", aux: "Int32", resultInArg0: true, clobberFlags: true}, // arg0 rotate left auxint, rotate amount 0-31
{name: "ROLWconst", argLength: 1, reg: gp11, asm: "ROLW", aux: "Int16", resultInArg0: true, clobberFlags: true}, // arg0 rotate left auxint, rotate amount 0-15
{name: "ROLBconst", argLength: 1, reg: gp11, asm: "ROLB", aux: "Int8", resultInArg0: true, clobberFlags: true}, // arg0 rotate left auxint, rotate amount 0-7

75
src/cmd/compile/internal/ssa/gen/AMD64.rules
View file

@ -853,86 +853,11 @@
((SHLL|SHRL|SARL|SHLXL|SHRXL|SARXL) x (ANDLconst [c] y)) && c & 31 == 31 => ((SHLL|SHRL|SARL|SHLXL|SHRXL|SARXL) x y)
((SHLL|SHRL|SARL|SHLXL|SHRXL|SARXL) x (NEGL <t> (ANDLconst [c] y))) && c & 31 == 31 => ((SHLL|SHRL|SARL|SHLXL|SHRXL|SARXL) x (NEGL <t> y))
// Constant rotate instructions
((ADDQ|ORQ|XORQ) (SHLQconst x [c]) (SHRQconst x [d])) && d==64-c => (ROLQconst x [c])
((ADDL|ORL|XORL) (SHLLconst x [c]) (SHRLconst x [d])) && d==32-c => (ROLLconst x [c])
((ADDL|ORL|XORL) <t> (SHLLconst x [c]) (SHRWconst x [d])) && d==16-c && c < 16 && t.Size() == 2 => (ROLWconst x [c])
((ADDL|ORL|XORL) <t> (SHLLconst x [c]) (SHRBconst x [d])) && d==8-c && c < 8 && t.Size() == 1 => (ROLBconst x [c])
(ROLQconst [c] (ROLQconst [d] x)) => (ROLQconst [(c+d)&63] x)
(ROLLconst [c] (ROLLconst [d] x)) => (ROLLconst [(c+d)&31] x)
(ROLWconst [c] (ROLWconst [d] x)) => (ROLWconst [(c+d)&15] x)
(ROLBconst [c] (ROLBconst [d] x)) => (ROLBconst [(c+d)& 7] x)
(RotateLeft8 ...) => (ROLB ...)
(RotateLeft16 ...) => (ROLW ...)
(RotateLeft32 ...) => (ROLL ...)
(RotateLeft64 ...) => (ROLQ ...)
// Non-constant rotates.
// We want to issue a rotate when the Go source contains code like
// y &= 63
// x << y | x >> (64-y)
// The shift rules above convert << to SHLx and >> to SHRx.
// SHRx converts its shift argument from 64-y to -y.
// A tricky situation occurs when y==0. Then the original code would be:
// x << 0 | x >> 64
// But x >> 64 is 0, not x. So there's an additional mask that is ANDed in
// to force the second term to 0. We don't need that mask, but we must match
// it in order to strip it out.
(ORQ (SHLQ x y) (ANDQ (SHRQ x (NEG(Q|L) y)) (SBBQcarrymask (CMP(Q|L)const (NEG(Q|L) (ADD(Q|L)const (AND(Q|L)const y [63]) [-64])) [64])))) => (ROLQ x y)
(ORQ (SHRQ x y) (ANDQ (SHLQ x (NEG(Q|L) y)) (SBBQcarrymask (CMP(Q|L)const (NEG(Q|L) (ADD(Q|L)const (AND(Q|L)const y [63]) [-64])) [64])))) => (RORQ x y)
(ORQ (SHLXQ x y) (ANDQ (SHRXQ x (NEG(Q|L) y)) (SBBQcarrymask (CMP(Q|L)const (NEG(Q|L) (ADD(Q|L)const (AND(Q|L)const y [63]) [-64])) [64])))) => (ROLQ x y)
(ORQ (SHRXQ x y) (ANDQ (SHLXQ x (NEG(Q|L) y)) (SBBQcarrymask (CMP(Q|L)const (NEG(Q|L) (ADD(Q|L)const (AND(Q|L)const y [63]) [-64])) [64])))) => (RORQ x y)
(ORL (SHLL x y) (ANDL (SHRL x (NEG(Q|L) y)) (SBBLcarrymask (CMP(Q|L)const (NEG(Q|L) (ADD(Q|L)const (AND(Q|L)const y [31]) [-32])) [32])))) => (ROLL x y)
(ORL (SHRL x y) (ANDL (SHLL x (NEG(Q|L) y)) (SBBLcarrymask (CMP(Q|L)const (NEG(Q|L) (ADD(Q|L)const (AND(Q|L)const y [31]) [-32])) [32])))) => (RORL x y)
(ORL (SHLXL x y) (ANDL (SHRXL x (NEG(Q|L) y)) (SBBLcarrymask (CMP(Q|L)const (NEG(Q|L) (ADD(Q|L)const (AND(Q|L)const y [31]) [-32])) [32])))) => (ROLL x y)
(ORL (SHRXL x y) (ANDL (SHLXL x (NEG(Q|L) y)) (SBBLcarrymask (CMP(Q|L)const (NEG(Q|L) (ADD(Q|L)const (AND(Q|L)const y [31]) [-32])) [32])))) => (RORL x y)
// Help with rotate detection
(CMPQconst (NEGQ (ADDQconst [-16] (ANDQconst [15] _))) [32]) => (FlagLT_ULT)
(CMPQconst (NEGQ (ADDQconst [ -8] (ANDQconst [7] _))) [32]) => (FlagLT_ULT)
(ORL (SHLL x (AND(Q|L)const y [15]))
(ANDL (SHRW x (NEG(Q|L) (ADD(Q|L)const (AND(Q|L)const y [15]) [-16])))
(SBBLcarrymask (CMP(Q|L)const (NEG(Q|L) (ADD(Q|L)const (AND(Q|L)const y [15]) [-16])) [16]))))
&& v.Type.Size() == 2
=> (ROLW x y)
(ORL (SHRW x (AND(Q|L)const y [15]))
(SHLL x (NEG(Q|L) (ADD(Q|L)const (AND(Q|L)const y [15]) [-16]))))
&& v.Type.Size() == 2
=> (RORW x y)
(ORL (SHLXL x (AND(Q|L)const y [15]))
(ANDL (SHRW x (NEG(Q|L) (ADD(Q|L)const (AND(Q|L)const y [15]) [-16])))
(SBBLcarrymask (CMP(Q|L)const (NEG(Q|L) (ADD(Q|L)const (AND(Q|L)const y [15]) [-16])) [16]))))
&& v.Type.Size() == 2
=> (ROLW x y)
(ORL (SHRW x (AND(Q|L)const y [15]))
(SHLXL x (NEG(Q|L) (ADD(Q|L)const (AND(Q|L)const y [15]) [-16]))))
&& v.Type.Size() == 2
=> (RORW x y)
(ORL (SHLL x (AND(Q|L)const y [ 7]))
(ANDL (SHRB x (NEG(Q|L) (ADD(Q|L)const (AND(Q|L)const y [ 7]) [ -8])))
(SBBLcarrymask (CMP(Q|L)const (NEG(Q|L) (ADD(Q|L)const (AND(Q|L)const y [ 7]) [ -8])) [ 8]))))
&& v.Type.Size() == 1
=> (ROLB x y)
(ORL (SHRB x (AND(Q|L)const y [ 7]))
(SHLL x (NEG(Q|L) (ADD(Q|L)const (AND(Q|L)const y [ 7]) [ -8]))))
&& v.Type.Size() == 1
=> (RORB x y)
(ORL (SHLXL x (AND(Q|L)const y [ 7]))
(ANDL (SHRB x (NEG(Q|L) (ADD(Q|L)const (AND(Q|L)const y [ 7]) [ -8])))
(SBBLcarrymask (CMP(Q|L)const (NEG(Q|L) (ADD(Q|L)const (AND(Q|L)const y [ 7]) [ -8])) [ 8]))))
&& v.Type.Size() == 1
=> (ROLB x y)
(ORL (SHRB x (AND(Q|L)const y [ 7]))
(SHLXL x (NEG(Q|L) (ADD(Q|L)const (AND(Q|L)const y [ 7]) [ -8]))))
&& v.Type.Size() == 1
=> (RORB x y)
// rotate left negative = rotate right
(ROLQ x (NEG(Q|L) y)) => (RORQ x y)
(ROLL x (NEG(Q|L) y)) => (RORL x y)

8
src/cmd/compile/internal/ssa/gen/ARM.rules
View file

@ -1130,14 +1130,6 @@
(CMNshiftRLreg x y (MOVWconst [c])) && 0 <= c && c < 32 => (CMNshiftRL x y [c])
(CMNshiftRAreg x y (MOVWconst [c])) && 0 <= c && c < 32 => (CMNshiftRA x y [c])
// Generate rotates
(ADDshiftLL [c] (SRLconst x [32-c]) x) => (SRRconst [32-c] x)
( ORshiftLL [c] (SRLconst x [32-c]) x) => (SRRconst [32-c] x)
(XORshiftLL [c] (SRLconst x [32-c]) x) => (SRRconst [32-c] x)
(ADDshiftRL [c] (SLLconst x [32-c]) x) => (SRRconst [ c] x)
( ORshiftRL [c] (SLLconst x [32-c]) x) => (SRRconst [ c] x)
(XORshiftRL [c] (SLLconst x [32-c]) x) => (SRRconst [ c] x)
(RotateLeft16 <t> x (MOVWconst [c])) => (Or16 (Lsh16x32 <t> x (MOVWconst [c&15])) (Rsh16Ux32 <t> x (MOVWconst [-c&15])))
(RotateLeft8 <t> x (MOVWconst [c])) => (Or8 (Lsh8x32 <t> x (MOVWconst [c&7])) (Rsh8Ux32 <t> x (MOVWconst [-c&7])))
(RotateLeft32 x y) => (SRR x (RSBconst [0] <y.Type> y))

47
src/cmd/compile/internal/ssa/gen/ARM64.rules
View file

@ -1820,56 +1820,9 @@
(ORNshiftRA (SRAconst x [c]) x [c]) => (MOVDconst [-1])
(ORNshiftRO (RORconst x [c]) x [c]) => (MOVDconst [-1])
// Generate rotates with const shift
(ADDshiftLL [c] (SRLconst x [64-c]) x) => (RORconst [64-c] x)
( ORshiftLL [c] (SRLconst x [64-c]) x) => (RORconst [64-c] x)
(XORshiftLL [c] (SRLconst x [64-c]) x) => (RORconst [64-c] x)
(ADDshiftRL [c] (SLLconst x [64-c]) x) => (RORconst [ c] x)
( ORshiftRL [c] (SLLconst x [64-c]) x) => (RORconst [ c] x)
(XORshiftRL [c] (SLLconst x [64-c]) x) => (RORconst [ c] x)
(ADDshiftLL <t> [c] (UBFX [bfc] x) x) && c < 32 && t.Size() == 4 && bfc == armBFAuxInt(32-c, c)
=> (RORWconst [32-c] x)
( ORshiftLL <t> [c] (UBFX [bfc] x) x) && c < 32 && t.Size() == 4 && bfc == armBFAuxInt(32-c, c)
=> (RORWconst [32-c] x)
(XORshiftLL <t> [c] (UBFX [bfc] x) x) && c < 32 && t.Size() == 4 && bfc == armBFAuxInt(32-c, c)
=> (RORWconst [32-c] x)
(ADDshiftRL <t> [c] (SLLconst x [32-c]) (MOVWUreg x)) && c < 32 && t.Size() == 4 => (RORWconst [c] x)
( ORshiftRL <t> [c] (SLLconst x [32-c]) (MOVWUreg x)) && c < 32 && t.Size() == 4 => (RORWconst [c] x)
(XORshiftRL <t> [c] (SLLconst x [32-c]) (MOVWUreg x)) && c < 32 && t.Size() == 4 => (RORWconst [c] x)
(RORconst [c] (RORconst [d] x)) => (RORconst [(c+d)&63] x)
(RORWconst [c] (RORWconst [d] x)) => (RORWconst [(c+d)&31] x)
// Generate rotates with non-const shift.
// These rules match the Go source code like
// y &= 63
// x << y | x >> (64-y)
// "|" can also be "^" or "+".
// As arm64 does not have a ROL instruction, so ROL(x, y) is replaced by ROR(x, -y).
((ADD|OR|XOR) (SLL x (ANDconst <t> [63] y))
(CSEL0 <typ.UInt64> [cc] (SRL <typ.UInt64> x (SUB <t> (MOVDconst [64]) (ANDconst <t> [63] y)))
(CMPconst [64] (SUB <t> (MOVDconst [64]) (ANDconst <t> [63] y))))) && cc == OpARM64LessThanU
=> (ROR x (NEG <t> y))
((ADD|OR|XOR) (SRL <typ.UInt64> x (ANDconst <t> [63] y))
(CSEL0 <typ.UInt64> [cc] (SLL x (SUB <t> (MOVDconst [64]) (ANDconst <t> [63] y)))
(CMPconst [64] (SUB <t> (MOVDconst [64]) (ANDconst <t> [63] y))))) && cc == OpARM64LessThanU
=> (ROR x y)
// These rules match the Go source code like
// y &= 31
// x << y | x >> (32-y)
// "|" can also be "^" or "+".
// As arm64 does not have a ROLW instruction, so ROLW(x, y) is replaced by RORW(x, -y).
((ADD|OR|XOR) (SLL x (ANDconst <t> [31] y))
(CSEL0 <typ.UInt32> [cc] (SRL <typ.UInt32> (MOVWUreg x) (SUB <t> (MOVDconst [32]) (ANDconst <t> [31] y)))
(CMPconst [64] (SUB <t> (MOVDconst [32]) (ANDconst <t> [31] y))))) && cc == OpARM64LessThanU
=> (RORW x (NEG <t> y))
((ADD|OR|XOR) (SRL <typ.UInt32> (MOVWUreg x) (ANDconst <t> [31] y))
(CSEL0 <typ.UInt32> [cc] (SLL x (SUB <t> (MOVDconst [32]) (ANDconst <t> [31] y)))
(CMPconst [64] (SUB <t> (MOVDconst [32]) (ANDconst <t> [31] y))))) && cc == OpARM64LessThanU
=> (RORW x y)
// rev16w | rev16
// ((x>>8) | (x<<8)) => (REV16W x), the type of x is uint16, "|" can also be "^" or "+".
((ADDshiftLL|ORshiftLL|XORshiftLL) <typ.UInt16> [8] (UBFX <typ.UInt16> [armBFAuxInt(8, 8)] x) x) => (REV16W x)

35
src/cmd/compile/internal/ssa/gen/PPC64.rules
View file

@ -128,39 +128,8 @@
// Rotates
(RotateLeft8 <t> x (MOVDconst [c])) => (Or8 (Lsh8x64 <t> x (MOVDconst [c&7])) (Rsh8Ux64 <t> x (MOVDconst [-c&7])))
(RotateLeft16 <t> x (MOVDconst [c])) => (Or16 (Lsh16x64 <t> x (MOVDconst [c&15])) (Rsh16Ux64 <t> x (MOVDconst [-c&15])))
(RotateLeft32 x (MOVDconst [c])) => (ROTLWconst [c&31] x)
(RotateLeft64 x (MOVDconst [c])) => (ROTLconst [c&63] x)
// Rotate generation with const shift
(ADD (SLDconst x [c]) (SRDconst x [d])) && d == 64-c => (ROTLconst [c] x)
( OR (SLDconst x [c]) (SRDconst x [d])) && d == 64-c => (ROTLconst [c] x)
(XOR (SLDconst x [c]) (SRDconst x [d])) && d == 64-c => (ROTLconst [c] x)
(ADD (SLWconst x [c]) (SRWconst x [d])) && d == 32-c => (ROTLWconst [c] x)
( OR (SLWconst x [c]) (SRWconst x [d])) && d == 32-c => (ROTLWconst [c] x)
(XOR (SLWconst x [c]) (SRWconst x [d])) && d == 32-c => (ROTLWconst [c] x)
// Rotate generation with non-const shift
// these match patterns from math/bits/RotateLeft[32|64], but there could be others
(ADD (SLD x (Select0 (ANDCCconst [63] y))) (SRD x (SUB <typ.UInt> (MOVDconst [64]) (Select0 <typ.UInt> (ANDCCconst [63] y))))) => (ROTL x y)
(ADD (SLD x (Select0 (ANDCCconst [63] y))) (SRD x (SUBFCconst <typ.UInt> [64] (Select0 <typ.UInt> (ANDCCconst [63] y))))) => (ROTL x y)
( OR (SLD x (Select0 (ANDCCconst [63] y))) (SRD x (SUB <typ.UInt> (MOVDconst [64]) (Select0 <typ.UInt> (ANDCCconst [63] y)))))=> (ROTL x y)
( OR (SLD x (Select0 (ANDCCconst [63] y))) (SRD x (SUBFCconst <typ.UInt> [64] (Select0 <typ.UInt> (ANDCCconst [63] y))))) => (ROTL x y)
(XOR (SLD x (Select0 (ANDCCconst [63] y))) (SRD x (SUB <typ.UInt> (MOVDconst [64]) (Select0 <typ.UInt> (ANDCCconst [63] y))))) => (ROTL x y)
(XOR (SLD x (Select0 (ANDCCconst [63] y))) (SRD x (SUBFCconst <typ.UInt> [64] (Select0 <typ.UInt> (ANDCCconst [63] y))))) => (ROTL x y)
(ADD (SLW x (Select0 (ANDCCconst [31] y))) (SRW x (SUBFCconst <typ.UInt> [32] (Select0 <typ.UInt> (ANDCCconst [31] y))))) => (ROTLW x y)
(ADD (SLW x (Select0 (ANDCCconst [31] y))) (SRW x (SUB <typ.UInt> (MOVDconst [32]) (Select0 <typ.UInt> (ANDCCconst [31] y))))) => (ROTLW x y)
( OR (SLW x (Select0 (ANDCCconst [31] y))) (SRW x (SUBFCconst <typ.UInt> [32] (Select0 <typ.UInt> (ANDCCconst [31] y))))) => (ROTLW x y)
( OR (SLW x (Select0 (ANDCCconst [31] y))) (SRW x (SUB <typ.UInt> (MOVDconst [32]) (Select0 <typ.UInt> (ANDCCconst [31] y))))) => (ROTLW x y)
(XOR (SLW x (Select0 (ANDCCconst [31] y))) (SRW x (SUBFCconst <typ.UInt> [32] (Select0 <typ.UInt> (ANDCCconst [31] y))))) => (ROTLW x y)
(XOR (SLW x (Select0 (ANDCCconst [31] y))) (SRW x (SUB <typ.UInt> (MOVDconst [32]) (Select0 <typ.UInt> (ANDCCconst [31] y))))) => (ROTLW x y)
// Lowering rotates
(RotateLeft32 x y) => (ROTLW x y)
(RotateLeft64 x y) => (ROTL x y)
(RotateLeft32 ...) => (ROTLW ...)
(RotateLeft64 ...) => (ROTL ...)
// Constant rotate generation
(ROTLW x (MOVDconst [c])) => (ROTLWconst x [c&31])

4
src/cmd/compile/internal/ssa/gen/S390X.rules
View file

@ -691,10 +691,6 @@
(RLLG x (MOVDconst [c])) => (RISBGZ x {s390x.NewRotateParams(0, 63, uint8(c&63))})
(RLL x (MOVDconst [c])) => (RLLconst x [uint8(c&31)])
// Match rotate by constant pattern.
((ADD|OR|XOR) (SLDconst x [c]) (SRDconst x [64-c])) => (RISBGZ x {s390x.NewRotateParams(0, 63, c)})
((ADD|OR|XOR)W (SLWconst x [c]) (SRWconst x [32-c])) => (RLLconst x [c])
// Signed 64-bit comparison with immediate.
(CMP x (MOVDconst [c])) && is32Bit(c) => (CMPconst x [int32(c)])
(CMP (MOVDconst [c]) x) && is32Bit(c) => (InvertFlags (CMPconst x [int32(c)]))

5
src/cmd/compile/internal/ssa/gen/dec64.rules
View file

@ -217,6 +217,11 @@
(Rsh8x64 x y) => (Rsh8x32 x (Or32 <typ.UInt32> (Zeromask (Int64Hi y)) (Int64Lo y)))
(Rsh8Ux64 x y) => (Rsh8Ux32 x (Or32 <typ.UInt32> (Zeromask (Int64Hi y)) (Int64Lo y)))
(RotateLeft64 x (Int64Make hi lo)) => (RotateLeft64 x lo)
(RotateLeft32 x (Int64Make hi lo)) => (RotateLeft32 x lo)
(RotateLeft16 x (Int64Make hi lo)) => (RotateLeft16 x lo)
(RotateLeft8 x (Int64Make hi lo)) => (RotateLeft8 x lo)
// Clean up constants a little
(Or32 <typ.UInt32> (Zeromask (Const32 [c])) y) && c == 0 => y
(Or32 <typ.UInt32> (Zeromask (Const32 [c])) y) && c != 0 => (Const32 <typ.UInt32> [-1])

80
src/cmd/compile/internal/ssa/gen/generic.rules
View file

@ -2551,3 +2551,83 @@
// Elide self-moves. This only happens rarely (e.g test/fixedbugs/bug277.go).
// However, this rule is needed to prevent the previous rule from looping forever in such cases.
(Move dst src mem) && isSamePtr(dst, src) => mem
// Constant rotate detection.
((Add64|Or64|Xor64) (Lsh64x64 x z:(Const64 <t> [c])) (Rsh64Ux64 x (Const64 [d]))) && c < 64 && d == 64-c && canRotate(config, 64) => (RotateLeft64 x z)
((Add32|Or32|Xor32) (Lsh32x64 x z:(Const64 <t> [c])) (Rsh32Ux64 x (Const64 [d]))) && c < 32 && d == 32-c && canRotate(config, 32) => (RotateLeft32 x z)
((Add16|Or16|Xor16) (Lsh16x64 x z:(Const64 <t> [c])) (Rsh16Ux64 x (Const64 [d]))) && c < 16 && d == 16-c && canRotate(config, 16) => (RotateLeft16 x z)
((Add8|Or8|Xor8) (Lsh8x64 x z:(Const64 <t> [c])) (Rsh8Ux64 x (Const64 [d]))) && c < 8 && d == 8-c && canRotate(config, 8) => (RotateLeft8 x z)
// Non-constant rotate detection.
// We use shiftIsBounded to make sure that neither of the shifts are >64.
// Note: these rules are subtle when the shift amounts are 0/64, as Go shifts
// are different from most native shifts. But it works out.
((Add64|Or64|Xor64) left:(Lsh64x64 x y) right:(Rsh64Ux64 x (Sub64 (Const64 [64]) y))) && (shiftIsBounded(left) || shiftIsBounded(right)) && canRotate(config, 64) => (RotateLeft64 x y)
((Add64|Or64|Xor64) left:(Lsh64x32 x y) right:(Rsh64Ux32 x (Sub32 (Const32 [64]) y))) && (shiftIsBounded(left) || shiftIsBounded(right)) && canRotate(config, 64) => (RotateLeft64 x y)
((Add64|Or64|Xor64) left:(Lsh64x16 x y) right:(Rsh64Ux16 x (Sub16 (Const16 [64]) y))) && (shiftIsBounded(left) || shiftIsBounded(right)) && canRotate(config, 64) => (RotateLeft64 x y)
((Add64|Or64|Xor64) left:(Lsh64x8 x y) right:(Rsh64Ux8 x (Sub8 (Const8 [64]) y))) && (shiftIsBounded(left) || shiftIsBounded(right)) && canRotate(config, 64) => (RotateLeft64 x y)
((Add64|Or64|Xor64) right:(Rsh64Ux64 x y) left:(Lsh64x64 x z:(Sub64 (Const64 [64]) y))) && (shiftIsBounded(left) || shiftIsBounded(right)) && canRotate(config, 64) => (RotateLeft64 x z)
((Add64|Or64|Xor64) right:(Rsh64Ux32 x y) left:(Lsh64x32 x z:(Sub32 (Const32 [64]) y))) && (shiftIsBounded(left) || shiftIsBounded(right)) && canRotate(config, 64) => (RotateLeft64 x z)
((Add64|Or64|Xor64) right:(Rsh64Ux16 x y) left:(Lsh64x16 x z:(Sub16 (Const16 [64]) y))) && (shiftIsBounded(left) || shiftIsBounded(right)) && canRotate(config, 64) => (RotateLeft64 x z)
((Add64|Or64|Xor64) right:(Rsh64Ux8 x y) left:(Lsh64x8 x z:(Sub8 (Const8 [64]) y))) && (shiftIsBounded(left) || shiftIsBounded(right)) && canRotate(config, 64) => (RotateLeft64 x z)
((Add32|Or32|Xor32) left:(Lsh32x64 x y) right:(Rsh32Ux64 x (Sub64 (Const64 [32]) y))) && (shiftIsBounded(left) || shiftIsBounded(right)) && canRotate(config, 32) => (RotateLeft32 x y)
((Add32|Or32|Xor32) left:(Lsh32x32 x y) right:(Rsh32Ux32 x (Sub32 (Const32 [32]) y))) && (shiftIsBounded(left) || shiftIsBounded(right)) && canRotate(config, 32) => (RotateLeft32 x y)
((Add32|Or32|Xor32) left:(Lsh32x16 x y) right:(Rsh32Ux16 x (Sub16 (Const16 [32]) y))) && (shiftIsBounded(left) || shiftIsBounded(right)) && canRotate(config, 32) => (RotateLeft32 x y)
((Add32|Or32|Xor32) left:(Lsh32x8 x y) right:(Rsh32Ux8 x (Sub8 (Const8 [32]) y))) && (shiftIsBounded(left) || shiftIsBounded(right)) && canRotate(config, 32) => (RotateLeft32 x y)
((Add32|Or32|Xor32) right:(Rsh32Ux64 x y) left:(Lsh32x64 x z:(Sub64 (Const64 [32]) y))) && (shiftIsBounded(left) || shiftIsBounded(right)) && canRotate(config, 32) => (RotateLeft32 x z)
((Add32|Or32|Xor32) right:(Rsh32Ux32 x y) left:(Lsh32x32 x z:(Sub32 (Const32 [32]) y))) && (shiftIsBounded(left) || shiftIsBounded(right)) && canRotate(config, 32) => (RotateLeft32 x z)
((Add32|Or32|Xor32) right:(Rsh32Ux16 x y) left:(Lsh32x16 x z:(Sub16 (Const16 [32]) y))) && (shiftIsBounded(left) || shiftIsBounded(right)) && canRotate(config, 32) => (RotateLeft32 x z)
((Add32|Or32|Xor32) right:(Rsh32Ux8 x y) left:(Lsh32x8 x z:(Sub8 (Const8 [32]) y))) && (shiftIsBounded(left) || shiftIsBounded(right)) && canRotate(config, 32) => (RotateLeft32 x z)
((Add16|Or16|Xor16) left:(Lsh16x64 x y) right:(Rsh16Ux64 x (Sub64 (Const64 [16]) y))) && (shiftIsBounded(left) || shiftIsBounded(right)) && canRotate(config, 16) => (RotateLeft16 x y)
((Add16|Or16|Xor16) left:(Lsh16x32 x y) right:(Rsh16Ux32 x (Sub32 (Const32 [16]) y))) && (shiftIsBounded(left) || shiftIsBounded(right)) && canRotate(config, 16) => (RotateLeft16 x y)
((Add16|Or16|Xor16) left:(Lsh16x16 x y) right:(Rsh16Ux16 x (Sub16 (Const16 [16]) y))) && (shiftIsBounded(left) || shiftIsBounded(right)) && canRotate(config, 16) => (RotateLeft16 x y)
((Add16|Or16|Xor16) left:(Lsh16x8 x y) right:(Rsh16Ux8 x (Sub8 (Const8 [16]) y))) && (shiftIsBounded(left) || shiftIsBounded(right)) && canRotate(config, 16) => (RotateLeft16 x y)
((Add16|Or16|Xor16) right:(Rsh16Ux64 x y) left:(Lsh16x64 x z:(Sub64 (Const64 [16]) y))) && (shiftIsBounded(left) || shiftIsBounded(right)) && canRotate(config, 16) => (RotateLeft16 x z)
((Add16|Or16|Xor16) right:(Rsh16Ux32 x y) left:(Lsh16x32 x z:(Sub32 (Const32 [16]) y))) && (shiftIsBounded(left) || shiftIsBounded(right)) && canRotate(config, 16) => (RotateLeft16 x z)
((Add16|Or16|Xor16) right:(Rsh16Ux16 x y) left:(Lsh16x16 x z:(Sub16 (Const16 [16]) y))) && (shiftIsBounded(left) || shiftIsBounded(right)) && canRotate(config, 16) => (RotateLeft16 x z)
((Add16|Or16|Xor16) right:(Rsh16Ux8 x y) left:(Lsh16x8 x z:(Sub8 (Const8 [16]) y))) && (shiftIsBounded(left) || shiftIsBounded(right)) && canRotate(config, 16) => (RotateLeft16 x z)
((Add8|Or8|Xor8) left:(Lsh8x64 x y) right:(Rsh8Ux64 x (Sub64 (Const64 [8]) y))) && (shiftIsBounded(left) || shiftIsBounded(right)) && canRotate(config, 8) => (RotateLeft8 x y)
((Add8|Or8|Xor8) left:(Lsh8x32 x y) right:(Rsh8Ux32 x (Sub32 (Const32 [8]) y))) && (shiftIsBounded(left) || shiftIsBounded(right)) && canRotate(config, 8) => (RotateLeft8 x y)
((Add8|Or8|Xor8) left:(Lsh8x16 x y) right:(Rsh8Ux16 x (Sub16 (Const16 [8]) y))) && (shiftIsBounded(left) || shiftIsBounded(right)) && canRotate(config, 8) => (RotateLeft8 x y)
((Add8|Or8|Xor8) left:(Lsh8x8 x y) right:(Rsh8Ux8 x (Sub8 (Const8 [8]) y))) && (shiftIsBounded(left) || shiftIsBounded(right)) && canRotate(config, 8) => (RotateLeft8 x y)
((Add8|Or8|Xor8) right:(Rsh8Ux64 x y) left:(Lsh8x64 x z:(Sub64 (Const64 [8]) y))) && (shiftIsBounded(left) || shiftIsBounded(right)) && canRotate(config, 8) => (RotateLeft8 x z)
((Add8|Or8|Xor8) right:(Rsh8Ux32 x y) left:(Lsh8x32 x z:(Sub32 (Const32 [8]) y))) && (shiftIsBounded(left) || shiftIsBounded(right)) && canRotate(config, 8) => (RotateLeft8 x z)
((Add8|Or8|Xor8) right:(Rsh8Ux16 x y) left:(Lsh8x16 x z:(Sub16 (Const16 [8]) y))) && (shiftIsBounded(left) || shiftIsBounded(right)) && canRotate(config, 8) => (RotateLeft8 x z)
((Add8|Or8|Xor8) right:(Rsh8Ux8 x y) left:(Lsh8x8 x z:(Sub8 (Const8 [8]) y))) && (shiftIsBounded(left) || shiftIsBounded(right)) && canRotate(config, 8) => (RotateLeft8 x z)
// Rotating by y&c, with c a mask that doesn't change the bottom bits, is the same as rotating by y.
(RotateLeft64 x (And(64|32|16|8) y (Const(64|32|16|8) [c]))) && c&63 == 63 => (RotateLeft64 x y)
(RotateLeft32 x (And(64|32|16|8) y (Const(64|32|16|8) [c]))) && c&31 == 31 => (RotateLeft32 x y)
(RotateLeft16 x (And(64|32|16|8) y (Const(64|32|16|8) [c]))) && c&15 == 15 => (RotateLeft16 x y)
(RotateLeft8 x (And(64|32|16|8) y (Const(64|32|16|8) [c]))) && c&7 == 7 => (RotateLeft8 x y)
// Rotating by -(y&c), with c a mask that doesn't change the bottom bits, is the same as rotating by -y.
(RotateLeft64 x (Neg(64|32|16|8) (And(64|32|16|8) y (Const(64|32|16|8) [c])))) && c&63 == 63 => (RotateLeft64 x (Neg(64|32|16|8) <y.Type> y))
(RotateLeft32 x (Neg(64|32|16|8) (And(64|32|16|8) y (Const(64|32|16|8) [c])))) && c&31 == 31 => (RotateLeft32 x (Neg(64|32|16|8) <y.Type> y))
(RotateLeft16 x (Neg(64|32|16|8) (And(64|32|16|8) y (Const(64|32|16|8) [c])))) && c&15 == 15 => (RotateLeft16 x (Neg(64|32|16|8) <y.Type> y))
(RotateLeft8 x (Neg(64|32|16|8) (And(64|32|16|8) y (Const(64|32|16|8) [c])))) && c&7 == 7 => (RotateLeft8 x (Neg(64|32|16|8) <y.Type> y))
// Rotating by y+c, with c a multiple of the value width, is the same as rotating by y.
(RotateLeft64 x (Add(64|32|16|8) y (Const(64|32|16|8) [c]))) && c&63 == 0 => (RotateLeft64 x y)
(RotateLeft32 x (Add(64|32|16|8) y (Const(64|32|16|8) [c]))) && c&31 == 0 => (RotateLeft32 x y)
(RotateLeft16 x (Add(64|32|16|8) y (Const(64|32|16|8) [c]))) && c&15 == 0 => (RotateLeft16 x y)
(RotateLeft8 x (Add(64|32|16|8) y (Const(64|32|16|8) [c]))) && c&7 == 0 => (RotateLeft8 x y)
// Rotating by c-y, with c a multiple of the value width, is the same as rotating by -y.
(RotateLeft64 x (Sub(64|32|16|8) (Const(64|32|16|8) [c]) y)) && c&63 == 0 => (RotateLeft64 x (Neg(64|32|16|8) <y.Type> y))
(RotateLeft32 x (Sub(64|32|16|8) (Const(64|32|16|8) [c]) y)) && c&31 == 0 => (RotateLeft32 x (Neg(64|32|16|8) <y.Type> y))
(RotateLeft16 x (Sub(64|32|16|8) (Const(64|32|16|8) [c]) y)) && c&15 == 0 => (RotateLeft16 x (Neg(64|32|16|8) <y.Type> y))
(RotateLeft8 x (Sub(64|32|16|8) (Const(64|32|16|8) [c]) y)) && c&7 == 0 => (RotateLeft8 x (Neg(64|32|16|8) <y.Type> y))
// Ensure we don't do Const64 rotates in a 32-bit system.
(RotateLeft64 x (Const64 <t> [c])) && config.PtrSize == 4 => (RotateLeft64 x (Const32 <t> [int32(c)]))
(RotateLeft32 x (Const64 <t> [c])) && config.PtrSize == 4 => (RotateLeft32 x (Const32 <t> [int32(c)]))
(RotateLeft16 x (Const64 <t> [c])) && config.PtrSize == 4 => (RotateLeft16 x (Const32 <t> [int32(c)]))
(RotateLeft8 x (Const64 <t> [c])) && config.PtrSize == 4 => (RotateLeft8 x (Const32 <t> [int32(c)]))

21
src/cmd/compile/internal/ssa/gen/genericOps.go
View file

@ -249,14 +249,19 @@ var genericOps = []opData{
{name: "BitRev32", argLength: 1}, // Reverse the bits in arg[0]
{name: "BitRev64", argLength: 1}, // Reverse the bits in arg[0]
{name: "PopCount8", argLength: 1}, // Count bits in arg[0]
{name: "PopCount16", argLength: 1}, // Count bits in arg[0]
{name: "PopCount32", argLength: 1}, // Count bits in arg[0]
{name: "PopCount64", argLength: 1}, // Count bits in arg[0]
{name: "RotateLeft8", argLength: 2}, // Rotate bits in arg[0] left by arg[1]
{name: "RotateLeft16", argLength: 2}, // Rotate bits in arg[0] left by arg[1]
{name: "RotateLeft32", argLength: 2}, // Rotate bits in arg[0] left by arg[1]
{name: "RotateLeft64", argLength: 2}, // Rotate bits in arg[0] left by arg[1]
{name: "PopCount8", argLength: 1}, // Count bits in arg[0]
{name: "PopCount16", argLength: 1}, // Count bits in arg[0]
{name: "PopCount32", argLength: 1}, // Count bits in arg[0]
{name: "PopCount64", argLength: 1}, // Count bits in arg[0]
// RotateLeftX instructions rotate the X bits of arg[0] to the left
// by the low lg_2(X) bits of arg[1], interpreted as an unsigned value.
// Note that this works out regardless of the bit width or signedness of
// arg[1]. In particular, RotateLeft by x is the same as RotateRight by -x.
{name: "RotateLeft64", argLength: 2},
{name: "RotateLeft32", argLength: 2},
{name: "RotateLeft16", argLength: 2},
{name: "RotateLeft8", argLength: 2},
// Square root.
// Special cases:

71
src/cmd/compile/internal/ssa/opGen.go
View file

@ -434,6 +434,9 @@ const (
Op386SARLconst
Op386SARWconst
Op386SARBconst
Op386ROLL
Op386ROLW
Op386ROLB
Op386ROLLconst
Op386ROLWconst
Op386ROLBconst
@ -2963,10 +2966,10 @@ const (
OpPopCount16
OpPopCount32
OpPopCount64
OpRotateLeft8
OpRotateLeft16
OpRotateLeft32
OpRotateLeft64
OpRotateLeft32
OpRotateLeft16
OpRotateLeft8
OpSqrt
OpSqrt32
OpFloor
@ -4636,6 +4639,54 @@ var opcodeTable = [...]opInfo{
},
},
},
{
name: "ROLL",
argLen: 2,
resultInArg0: true,
clobberFlags: true,
asm: x86.AROLL,
reg: regInfo{
inputs: []inputInfo{
{1, 2}, // CX
{0, 239}, // AX CX DX BX BP SI DI
},
outputs: []outputInfo{
{0, 239}, // AX CX DX BX BP SI DI
},
},
},
{
name: "ROLW",
argLen: 2,
resultInArg0: true,
clobberFlags: true,
asm: x86.AROLW,
reg: regInfo{
inputs: []inputInfo{
{1, 2}, // CX
{0, 239}, // AX CX DX BX BP SI DI
},
outputs: []outputInfo{
{0, 239}, // AX CX DX BX BP SI DI
},
},
},
{
name: "ROLB",
argLen: 2,
resultInArg0: true,
clobberFlags: true,
asm: x86.AROLB,
reg: regInfo{
inputs: []inputInfo{
{1, 2}, // CX
{0, 239}, // AX CX DX BX BP SI DI
},
outputs: []outputInfo{
{0, 239}, // AX CX DX BX BP SI DI
},
},
},
{
name: "ROLLconst",
auxType: auxInt32,
@ -38457,12 +38508,7 @@ var opcodeTable = [...]opInfo{
generic: true,
},
{
name: "RotateLeft8",
argLen: 2,
generic: true,
},
{
name: "RotateLeft16",
name: "RotateLeft64",
argLen: 2,
generic: true,
},
@ -38472,7 +38518,12 @@ var opcodeTable = [...]opInfo{
generic: true,
},
{
name: "RotateLeft64",
name: "RotateLeft16",
argLen: 2,
generic: true,
},
{
name: "RotateLeft8",
argLen: 2,
generic: true,
},

17
src/cmd/compile/internal/ssa/rewrite.go
View file

@ -1977,3 +1977,20 @@ func makeJumpTableSym(b *Block) *obj.LSym {
s.Set(obj.AttrLocal, true)
return s
}
// canRotate reports whether the architecture supports
// rotates of integer registers with the given number of bits.
func canRotate(c *Config, bits int64) bool {
if bits > c.PtrSize*8 {
// Don't rewrite to rotates bigger than the machine word.
return false
}
switch c.arch {
case "386", "amd64":
return true
case "arm", "arm64", "s390x", "ppc64", "ppc64le", "wasm":
return bits >= 32
default:
return false
}
}

345
src/cmd/compile/internal/ssa/rewrite386.go
View file

@ -146,10 +146,16 @@ func rewriteValue386(v *Value) bool {
return rewriteValue386_Op386ORLload(v)
case Op386ORLmodify:
return rewriteValue386_Op386ORLmodify(v)
case Op386ROLB:
return rewriteValue386_Op386ROLB(v)
case Op386ROLBconst:
return rewriteValue386_Op386ROLBconst(v)
case Op386ROLL:
return rewriteValue386_Op386ROLL(v)
case Op386ROLLconst:
return rewriteValue386_Op386ROLLconst(v)
case Op386ROLW:
return rewriteValue386_Op386ROLW(v)
case Op386ROLWconst:
return rewriteValue386_Op386ROLWconst(v)
case Op386SARB:
@ -541,11 +547,14 @@ func rewriteValue386(v *Value) bool {
case OpPanicExtend:
return rewriteValue386_OpPanicExtend(v)
case OpRotateLeft16:
return rewriteValue386_OpRotateLeft16(v)
v.Op = Op386ROLW
return true
case OpRotateLeft32:
return rewriteValue386_OpRotateLeft32(v)
v.Op = Op386ROLL
return true
case OpRotateLeft8:
return rewriteValue386_OpRotateLeft8(v)
v.Op = Op386ROLB
return true
case OpRound32F:
v.Op = OpCopy
return true
@ -734,80 +743,6 @@ func rewriteValue386_Op386ADDL(v *Value) bool {
}
break
}
// match: (ADDL (SHLLconst [c] x) (SHRLconst [d] x))
// cond: d == 32-c
// result: (ROLLconst [c] x)
for {
for _i0 := 0; _i0 <= 1; _i0, v_0, v_1 = _i0+1, v_1, v_0 {
if v_0.Op != Op386SHLLconst {
continue
}
c := auxIntToInt32(v_0.AuxInt)
x := v_0.Args[0]
if v_1.Op != Op386SHRLconst {
continue
}
d := auxIntToInt32(v_1.AuxInt)
if x != v_1.Args[0] || !(d == 32-c) {
continue
}
v.reset(Op386ROLLconst)
v.AuxInt = int32ToAuxInt(c)
v.AddArg(x)
return true
}
break
}
// match: (ADDL <t> (SHLLconst x [c]) (SHRWconst x [d]))
// cond: c < 16 && d == int16(16-c) && t.Size() == 2
// result: (ROLWconst x [int16(c)])
for {
t := v.Type
for _i0 := 0; _i0 <= 1; _i0, v_0, v_1 = _i0+1, v_1, v_0 {
if v_0.Op != Op386SHLLconst {
continue
}
c := auxIntToInt32(v_0.AuxInt)
x := v_0.Args[0]
if v_1.Op != Op386SHRWconst {
continue
}
d := auxIntToInt16(v_1.AuxInt)
if x != v_1.Args[0] || !(c < 16 && d == int16(16-c) && t.Size() == 2) {
continue
}
v.reset(Op386ROLWconst)
v.AuxInt = int16ToAuxInt(int16(c))
v.AddArg(x)
return true
}
break
}
// match: (ADDL <t> (SHLLconst x [c]) (SHRBconst x [d]))
// cond: c < 8 && d == int8(8-c) && t.Size() == 1
// result: (ROLBconst x [int8(c)])
for {
t := v.Type
for _i0 := 0; _i0 <= 1; _i0, v_0, v_1 = _i0+1, v_1, v_0 {
if v_0.Op != Op386SHLLconst {
continue
}
c := auxIntToInt32(v_0.AuxInt)
x := v_0.Args[0]
if v_1.Op != Op386SHRBconst {
continue
}
d := auxIntToInt8(v_1.AuxInt)
if x != v_1.Args[0] || !(c < 8 && d == int8(8-c) && t.Size() == 1) {
continue
}
v.reset(Op386ROLBconst)
v.AuxInt = int8ToAuxInt(int8(c))
v.AddArg(x)
return true
}
break
}
// match: (ADDL x (SHLLconst [3] y))
// result: (LEAL8 x y)
for {
@ -6305,80 +6240,6 @@ func rewriteValue386_Op386ORL(v *Value) bool {
}
break
}
// match: ( ORL (SHLLconst [c] x) (SHRLconst [d] x))
// cond: d == 32-c
// result: (ROLLconst [c] x)
for {
for _i0 := 0; _i0 <= 1; _i0, v_0, v_1 = _i0+1, v_1, v_0 {
if v_0.Op != Op386SHLLconst {
continue
}
c := auxIntToInt32(v_0.AuxInt)
x := v_0.Args[0]
if v_1.Op != Op386SHRLconst {
continue
}
d := auxIntToInt32(v_1.AuxInt)
if x != v_1.Args[0] || !(d == 32-c) {
continue
}
v.reset(Op386ROLLconst)
v.AuxInt = int32ToAuxInt(c)
v.AddArg(x)
return true
}
break
}
// match: ( ORL <t> (SHLLconst x [c]) (SHRWconst x [d]))
// cond: c < 16 && d == int16(16-c) && t.Size() == 2
// result: (ROLWconst x [int16(c)])
for {
t := v.Type
for _i0 := 0; _i0 <= 1; _i0, v_0, v_1 = _i0+1, v_1, v_0 {
if v_0.Op != Op386SHLLconst {
continue
}
c := auxIntToInt32(v_0.AuxInt)
x := v_0.Args[0]
if v_1.Op != Op386SHRWconst {
continue
}
d := auxIntToInt16(v_1.AuxInt)
if x != v_1.Args[0] || !(c < 16 && d == int16(16-c) && t.Size() == 2) {
continue
}
v.reset(Op386ROLWconst)
v.AuxInt = int16ToAuxInt(int16(c))
v.AddArg(x)
return true
}
break
}
// match: ( ORL <t> (SHLLconst x [c]) (SHRBconst x [d]))
// cond: c < 8 && d == int8(8-c) && t.Size() == 1
// result: (ROLBconst x [int8(c)])
for {
t := v.Type
for _i0 := 0; _i0 <= 1; _i0, v_0, v_1 = _i0+1, v_1, v_0 {
if v_0.Op != Op386SHLLconst {
continue
}
c := auxIntToInt32(v_0.AuxInt)
x := v_0.Args[0]
if v_1.Op != Op386SHRBconst {
continue
}
d := auxIntToInt8(v_1.AuxInt)
if x != v_1.Args[0] || !(c < 8 && d == int8(8-c) && t.Size() == 1) {
continue
}
v.reset(Op386ROLBconst)
v.AuxInt = int8ToAuxInt(int8(c))
v.AddArg(x)
return true
}
break
}
// match: (ORL x l:(MOVLload [off] {sym} ptr mem))
// cond: canMergeLoadClobber(v, l, x) && clobber(l)
// result: (ORLload x [off] {sym} ptr mem)
@ -6809,6 +6670,24 @@ func rewriteValue386_Op386ORLmodify(v *Value) bool {
}
return false
}
func rewriteValue386_Op386ROLB(v *Value) bool {
v_1 := v.Args[1]
v_0 := v.Args[0]
// match: (ROLB x (MOVLconst [c]))
// result: (ROLBconst [int8(c&7)] x)
for {
x := v_0
if v_1.Op != Op386MOVLconst {
break
}
c := auxIntToInt32(v_1.AuxInt)
v.reset(Op386ROLBconst)
v.AuxInt = int8ToAuxInt(int8(c & 7))
v.AddArg(x)
return true
}
return false
}
func rewriteValue386_Op386ROLBconst(v *Value) bool {
v_0 := v.Args[0]
// match: (ROLBconst [c] (ROLBconst [d] x))
@ -6837,6 +6716,24 @@ func rewriteValue386_Op386ROLBconst(v *Value) bool {
}
return false
}
func rewriteValue386_Op386ROLL(v *Value) bool {
v_1 := v.Args[1]
v_0 := v.Args[0]
// match: (ROLL x (MOVLconst [c]))
// result: (ROLLconst [c&31] x)
for {
x := v_0
if v_1.Op != Op386MOVLconst {
break
}
c := auxIntToInt32(v_1.AuxInt)
v.reset(Op386ROLLconst)
v.AuxInt = int32ToAuxInt(c & 31)
v.AddArg(x)
return true
}
return false
}
func rewriteValue386_Op386ROLLconst(v *Value) bool {
v_0 := v.Args[0]
// match: (ROLLconst [c] (ROLLconst [d] x))
@ -6865,6 +6762,24 @@ func rewriteValue386_Op386ROLLconst(v *Value) bool {
}
return false
}
func rewriteValue386_Op386ROLW(v *Value) bool {
v_1 := v.Args[1]
v_0 := v.Args[0]
// match: (ROLW x (MOVLconst [c]))
// result: (ROLWconst [int16(c&15)] x)
for {
x := v_0
if v_1.Op != Op386MOVLconst {
break
}
c := auxIntToInt32(v_1.AuxInt)
v.reset(Op386ROLWconst)
v.AuxInt = int16ToAuxInt(int16(c & 15))
v.AddArg(x)
return true
}
return false
}
func rewriteValue386_Op386ROLWconst(v *Value) bool {
v_0 := v.Args[0]
// match: (ROLWconst [c] (ROLWconst [d] x))
@ -8346,80 +8261,6 @@ func rewriteValue386_Op386XORL(v *Value) bool {
}
break
}
// match: (XORL (SHLLconst [c] x) (SHRLconst [d] x))
// cond: d == 32-c
// result: (ROLLconst [c] x)
for {
for _i0 := 0; _i0 <= 1; _i0, v_0, v_1 = _i0+1, v_1, v_0 {
if v_0.Op != Op386SHLLconst {
continue
}
c := auxIntToInt32(v_0.AuxInt)
x := v_0.Args[0]
if v_1.Op != Op386SHRLconst {
continue
}
d := auxIntToInt32(v_1.AuxInt)
if x != v_1.Args[0] || !(d == 32-c) {
continue
}
v.reset(Op386ROLLconst)
v.AuxInt = int32ToAuxInt(c)
v.AddArg(x)
return true
}
break
}
// match: (XORL <t> (SHLLconst x [c]) (SHRWconst x [d]))
// cond: c < 16 && d == int16(16-c) && t.Size() == 2
// result: (ROLWconst x [int16(c)])
for {
t := v.Type
for _i0 := 0; _i0 <= 1; _i0, v_0, v_1 = _i0+1, v_1, v_0 {
if v_0.Op != Op386SHLLconst {
continue
}
c := auxIntToInt32(v_0.AuxInt)
x := v_0.Args[0]
if v_1.Op != Op386SHRWconst {
continue
}
d := auxIntToInt16(v_1.AuxInt)
if x != v_1.Args[0] || !(c < 16 && d == int16(16-c) && t.Size() == 2) {
continue
}
v.reset(Op386ROLWconst)
v.AuxInt = int16ToAuxInt(int16(c))
v.AddArg(x)
return true
}
break
}
// match: (XORL <t> (SHLLconst x [c]) (SHRBconst x [d]))
// cond: c < 8 && d == int8(8-c) && t.Size() == 1
// result: (ROLBconst x [int8(c)])
for {
t := v.Type
for _i0 := 0; _i0 <= 1; _i0, v_0, v_1 = _i0+1, v_1, v_0 {
if v_0.Op != Op386SHLLconst {
continue
}
c := auxIntToInt32(v_0.AuxInt)
x := v_0.Args[0]
if v_1.Op != Op386SHRBconst {
continue
}
d := auxIntToInt8(v_1.AuxInt)
if x != v_1.Args[0] || !(c < 8 && d == int8(8-c) && t.Size() == 1) {
continue
}
v.reset(Op386ROLBconst)
v.AuxInt = int8ToAuxInt(int8(c))
v.AddArg(x)
return true
}
break
}
// match: (XORL x l:(MOVLload [off] {sym} ptr mem))
// cond: canMergeLoadClobber(v, l, x) && clobber(l)
// result: (XORLload x [off] {sym} ptr mem)
@ -10298,60 +10139,6 @@ func rewriteValue386_OpPanicExtend(v *Value) bool {
}
return false
}
func rewriteValue386_OpRotateLeft16(v *Value) bool {
v_1 := v.Args[1]
v_0 := v.Args[0]
// match: (RotateLeft16 x (MOVLconst [c]))
// result: (ROLWconst [int16(c&15)] x)
for {
x := v_0
if v_1.Op != Op386MOVLconst {
break
}
c := auxIntToInt32(v_1.AuxInt)
v.reset(Op386ROLWconst)
v.AuxInt = int16ToAuxInt(int16(c & 15))
v.AddArg(x)
return true
}
return false
}
func rewriteValue386_OpRotateLeft32(v *Value) bool {
v_1 := v.Args[1]
v_0 := v.Args[0]
// match: (RotateLeft32 x (MOVLconst [c]))
// result: (ROLLconst [c&31] x)
for {
x := v_0
if v_1.Op != Op386MOVLconst {
break
}
c := auxIntToInt32(v_1.AuxInt)
v.reset(Op386ROLLconst)
v.AuxInt = int32ToAuxInt(c & 31)
v.AddArg(x)
return true
}
return false
}
func rewriteValue386_OpRotateLeft8(v *Value) bool {
v_1 := v.Args[1]
v_0 := v.Args[0]
// match: (RotateLeft8 x (MOVLconst [c]))
// result: (ROLBconst [int8(c&7)] x)
for {
x := v_0
if v_1.Op != Op386MOVLconst {
break
}
c := auxIntToInt32(v_1.AuxInt)
v.reset(Op386ROLBconst)
v.AuxInt = int8ToAuxInt(int8(c & 7))
v.AddArg(x)
return true
}
return false
}
func rewriteValue386_OpRsh16Ux16(v *Value) bool {
v_1 := v.Args[1]
v_0 := v.Args[0]

2016
src/cmd/compile/internal/ssa/rewriteAMD64.go
View file

File diff suppressed because it is too large Load diff

96
src/cmd/compile/internal/ssa/rewriteARM.go
View file

@ -2080,22 +2080,6 @@ func rewriteValueARM_OpARMADDshiftLL(v *Value) bool {
v.AddArg(x)
return true
}
// match: (ADDshiftLL [c] (SRLconst x [32-c]) x)
// result: (SRRconst [32-c] x)
for {
c := auxIntToInt32(v.AuxInt)
if v_0.Op != OpARMSRLconst || auxIntToInt32(v_0.AuxInt) != 32-c {
break
}
x := v_0.Args[0]
if x != v_1 {
break
}
v.reset(OpARMSRRconst)
v.AuxInt = int32ToAuxInt(32 - c)
v.AddArg(x)
return true
}
// match: (ADDshiftLL <typ.UInt16> [8] (BFXU <typ.UInt16> [int32(armBFAuxInt(8, 8))] x) x)
// result: (REV16 x)
for {
@ -2285,22 +2269,6 @@ func rewriteValueARM_OpARMADDshiftRL(v *Value) bool {
v.AddArg(x)
return true
}
// match: (ADDshiftRL [c] (SLLconst x [32-c]) x)
// result: (SRRconst [ c] x)
for {
c := auxIntToInt32(v.AuxInt)
if v_0.Op != OpARMSLLconst || auxIntToInt32(v_0.AuxInt) != 32-c {
break
}
x := v_0.Args[0]
if x != v_1 {
break
}
v.reset(OpARMSRRconst)
v.AuxInt = int32ToAuxInt(c)
v.AddArg(x)
return true
}
return false
}
func rewriteValueARM_OpARMADDshiftRLreg(v *Value) bool {
@ -8596,22 +8564,6 @@ func rewriteValueARM_OpARMORshiftLL(v *Value) bool {
v.AddArg(x)
return true
}
// match: ( ORshiftLL [c] (SRLconst x [32-c]) x)
// result: (SRRconst [32-c] x)
for {
c := auxIntToInt32(v.AuxInt)
if v_0.Op != OpARMSRLconst || auxIntToInt32(v_0.AuxInt) != 32-c {
break
}
x := v_0.Args[0]
if x != v_1 {
break
}
v.reset(OpARMSRRconst)
v.AuxInt = int32ToAuxInt(32 - c)
v.AddArg(x)
return true
}
// match: (ORshiftLL <typ.UInt16> [8] (BFXU <typ.UInt16> [int32(armBFAuxInt(8, 8))] x) x)
// result: (REV16 x)
for {
@ -8831,22 +8783,6 @@ func rewriteValueARM_OpARMORshiftRL(v *Value) bool {
v.AddArg(x)
return true
}
// match: ( ORshiftRL [c] (SLLconst x [32-c]) x)
// result: (SRRconst [ c] x)
for {
c := auxIntToInt32(v.AuxInt)
if v_0.Op != OpARMSLLconst || auxIntToInt32(v_0.AuxInt) != 32-c {
break
}
x := v_0.Args[0]
if x != v_1 {
break
}
v.reset(OpARMSRRconst)
v.AuxInt = int32ToAuxInt(c)
v.AddArg(x)
return true
}
// match: (ORshiftRL y:(SRLconst x [c]) x [c])
// result: y
for {
@ -12755,22 +12691,6 @@ func rewriteValueARM_OpARMXORshiftLL(v *Value) bool {
v.AddArg(x)
return true
}
// match: (XORshiftLL [c] (SRLconst x [32-c]) x)
// result: (SRRconst [32-c] x)
for {
c := auxIntToInt32(v.AuxInt)
if v_0.Op != OpARMSRLconst || auxIntToInt32(v_0.AuxInt) != 32-c {
break
}
x := v_0.Args[0]
if x != v_1 {
break
}
v.reset(OpARMSRRconst)
v.AuxInt = int32ToAuxInt(32 - c)
v.AddArg(x)
return true
}
// match: (XORshiftLL <typ.UInt16> [8] (BFXU <typ.UInt16> [int32(armBFAuxInt(8, 8))] x) x)
// result: (REV16 x)
for {
@ -12990,22 +12910,6 @@ func rewriteValueARM_OpARMXORshiftRL(v *Value) bool {
v.AddArg(x)
return true
}
// match: (XORshiftRL [c] (SLLconst x [32-c]) x)
// result: (SRRconst [ c] x)
for {
c := auxIntToInt32(v.AuxInt)
if v_0.Op != OpARMSLLconst || auxIntToInt32(v_0.AuxInt) != 32-c {
break
}
x := v_0.Args[0]
if x != v_1 {
break
}
v.reset(OpARMSRRconst)
v.AuxInt = int32ToAuxInt(c)
v.AddArg(x)
return true
}
// match: (XORshiftRL (SRLconst x [c]) x [c])
// result: (MOVWconst [0])
for {

1055
src/cmd/compile/internal/ssa/rewriteARM64.go
View file

File diff suppressed because it is too large Load diff

752
src/cmd/compile/internal/ssa/rewritePPC64.go
View file

@ -648,9 +648,11 @@ func rewriteValuePPC64(v *Value) bool {
case OpRotateLeft16:
return rewriteValuePPC64_OpRotateLeft16(v)
case OpRotateLeft32:
return rewriteValuePPC64_OpRotateLeft32(v)
v.Op = OpPPC64ROTLW
return true
case OpRotateLeft64:
return rewriteValuePPC64_OpRotateLeft64(v)
v.Op = OpPPC64ROTL
return true
case OpRotateLeft8:
return rewriteValuePPC64_OpRotateLeft8(v)
case OpRound:
@ -3898,8 +3900,6 @@ func rewriteValuePPC64_OpOffPtr(v *Value) bool {
func rewriteValuePPC64_OpPPC64ADD(v *Value) bool {
v_1 := v.Args[1]
v_0 := v.Args[0]
b := v.Block
typ := &b.Func.Config.Types
// match: (ADD l:(MULLD x y) z)
// cond: buildcfg.GOPPC64 >= 9 && l.Uses == 1 && clobber(l)
// result: (MADDLD x y z)
@ -3921,236 +3921,6 @@ func rewriteValuePPC64_OpPPC64ADD(v *Value) bool {
}
break
}
// match: (ADD (SLDconst x [c]) (SRDconst x [d]))
// cond: d == 64-c
// result: (ROTLconst [c] x)
for {
for _i0 := 0; _i0 <= 1; _i0, v_0, v_1 = _i0+1, v_1, v_0 {
if v_0.Op != OpPPC64SLDconst {
continue
}
c := auxIntToInt64(v_0.AuxInt)
x := v_0.Args[0]
if v_1.Op != OpPPC64SRDconst {
continue
}
d := auxIntToInt64(v_1.AuxInt)
if x != v_1.Args[0] || !(d == 64-c) {
continue
}
v.reset(OpPPC64ROTLconst)
v.AuxInt = int64ToAuxInt(c)
v.AddArg(x)
return true
}
break
}
// match: (ADD (SLWconst x [c]) (SRWconst x [d]))
// cond: d == 32-c
// result: (ROTLWconst [c] x)
for {
for _i0 := 0; _i0 <= 1; _i0, v_0, v_1 = _i0+1, v_1, v_0 {
if v_0.Op != OpPPC64SLWconst {
continue
}
c := auxIntToInt64(v_0.AuxInt)
x := v_0.Args[0]
if v_1.Op != OpPPC64SRWconst {
continue
}
d := auxIntToInt64(v_1.AuxInt)
if x != v_1.Args[0] || !(d == 32-c) {
continue
}
v.reset(OpPPC64ROTLWconst)
v.AuxInt = int64ToAuxInt(c)
v.AddArg(x)
return true
}
break
}
// match: (ADD (SLD x (Select0 (ANDCCconst [63] y))) (SRD x (SUB <typ.UInt> (MOVDconst [64]) (Select0 <typ.UInt> (ANDCCconst [63] y)))))
// result: (ROTL x y)
for {
for _i0 := 0; _i0 <= 1; _i0, v_0, v_1 = _i0+1, v_1, v_0 {
if v_0.Op != OpPPC64SLD {
continue
}
_ = v_0.Args[1]
x := v_0.Args[0]
v_0_1 := v_0.Args[1]
if v_0_1.Op != OpSelect0 {
continue
}
v_0_1_0 := v_0_1.Args[0]
if v_0_1_0.Op != OpPPC64ANDCCconst || auxIntToInt64(v_0_1_0.AuxInt) != 63 {
continue
}
y := v_0_1_0.Args[0]
if v_1.Op != OpPPC64SRD {
continue
}
_ = v_1.Args[1]
if x != v_1.Args[0] {
continue
}
v_1_1 := v_1.Args[1]
if v_1_1.Op != OpPPC64SUB || v_1_1.Type != typ.UInt {
continue
}
_ = v_1_1.Args[1]
v_1_1_0 := v_1_1.Args[0]
if v_1_1_0.Op != OpPPC64MOVDconst || auxIntToInt64(v_1_1_0.AuxInt) != 64 {
continue
}
v_1_1_1 := v_1_1.Args[1]
if v_1_1_1.Op != OpSelect0 || v_1_1_1.Type != typ.UInt {
continue
}
v_1_1_1_0 := v_1_1_1.Args[0]
if v_1_1_1_0.Op != OpPPC64ANDCCconst || auxIntToInt64(v_1_1_1_0.AuxInt) != 63 || y != v_1_1_1_0.Args[0] {
continue
}
v.reset(OpPPC64ROTL)
v.AddArg2(x, y)
return true
}
break
}
// match: (ADD (SLD x (Select0 (ANDCCconst [63] y))) (SRD x (SUBFCconst <typ.UInt> [64] (Select0 <typ.UInt> (ANDCCconst [63] y)))))
// result: (ROTL x y)
for {
for _i0 := 0; _i0 <= 1; _i0, v_0, v_1 = _i0+1, v_1, v_0 {
if v_0.Op != OpPPC64SLD {
continue
}
_ = v_0.Args[1]
x := v_0.Args[0]
v_0_1 := v_0.Args[1]
if v_0_1.Op != OpSelect0 {
continue
}
v_0_1_0 := v_0_1.Args[0]
if v_0_1_0.Op != OpPPC64ANDCCconst || auxIntToInt64(v_0_1_0.AuxInt) != 63 {
continue
}
y := v_0_1_0.Args[0]
if v_1.Op != OpPPC64SRD {
continue
}
_ = v_1.Args[1]
if x != v_1.Args[0] {
continue
}
v_1_1 := v_1.Args[1]
if v_1_1.Op != OpPPC64SUBFCconst || v_1_1.Type != typ.UInt || auxIntToInt64(v_1_1.AuxInt) != 64 {
continue
}
v_1_1_0 := v_1_1.Args[0]
if v_1_1_0.Op != OpSelect0 || v_1_1_0.Type != typ.UInt {
continue
}
v_1_1_0_0 := v_1_1_0.Args[0]
if v_1_1_0_0.Op != OpPPC64ANDCCconst || auxIntToInt64(v_1_1_0_0.AuxInt) != 63 || y != v_1_1_0_0.Args[0] {
continue
}
v.reset(OpPPC64ROTL)
v.AddArg2(x, y)
return true
}
break
}
// match: (ADD (SLW x (Select0 (ANDCCconst [31] y))) (SRW x (SUBFCconst <typ.UInt> [32] (Select0 <typ.UInt> (ANDCCconst [31] y)))))
// result: (ROTLW x y)
for {
for _i0 := 0; _i0 <= 1; _i0, v_0, v_1 = _i0+1, v_1, v_0 {
if v_0.Op != OpPPC64SLW {
continue
}
_ = v_0.Args[1]
x := v_0.Args[0]
v_0_1 := v_0.Args[1]
if v_0_1.Op != OpSelect0 {
continue
}
v_0_1_0 := v_0_1.Args[0]
if v_0_1_0.Op != OpPPC64ANDCCconst || auxIntToInt64(v_0_1_0.AuxInt) != 31 {
continue
}
y := v_0_1_0.Args[0]
if v_1.Op != OpPPC64SRW {
continue
}
_ = v_1.Args[1]
if x != v_1.Args[0] {
continue
}
v_1_1 := v_1.Args[1]
if v_1_1.Op != OpPPC64SUBFCconst || v_1_1.Type != typ.UInt || auxIntToInt64(v_1_1.AuxInt) != 32 {
continue
}
v_1_1_0 := v_1_1.Args[0]
if v_1_1_0.Op != OpSelect0 || v_1_1_0.Type != typ.UInt {
continue
}
v_1_1_0_0 := v_1_1_0.Args[0]
if v_1_1_0_0.Op != OpPPC64ANDCCconst || auxIntToInt64(v_1_1_0_0.AuxInt) != 31 || y != v_1_1_0_0.Args[0] {
continue
}
v.reset(OpPPC64ROTLW)
v.AddArg2(x, y)
return true
}
break
}
// match: (ADD (SLW x (Select0 (ANDCCconst [31] y))) (SRW x (SUB <typ.UInt> (MOVDconst [32]) (Select0 <typ.UInt> (ANDCCconst [31] y)))))
// result: (ROTLW x y)
for {
for _i0 := 0; _i0 <= 1; _i0, v_0, v_1 = _i0+1, v_1, v_0 {
if v_0.Op != OpPPC64SLW {
continue
}
_ = v_0.Args[1]
x := v_0.Args[0]
v_0_1 := v_0.Args[1]
if v_0_1.Op != OpSelect0 {
continue
}
v_0_1_0 := v_0_1.Args[0]
if v_0_1_0.Op != OpPPC64ANDCCconst || auxIntToInt64(v_0_1_0.AuxInt) != 31 {
continue
}
y := v_0_1_0.Args[0]
if v_1.Op != OpPPC64SRW {
continue
}
_ = v_1.Args[1]
if x != v_1.Args[0] {
continue
}
v_1_1 := v_1.Args[1]
if v_1_1.Op != OpPPC64SUB || v_1_1.Type != typ.UInt {
continue
}
_ = v_1_1.Args[1]
v_1_1_0 := v_1_1.Args[0]
if v_1_1_0.Op != OpPPC64MOVDconst || auxIntToInt64(v_1_1_0.AuxInt) != 32 {
continue
}
v_1_1_1 := v_1_1.Args[1]
if v_1_1_1.Op != OpSelect0 || v_1_1_1.Type != typ.UInt {
continue
}
v_1_1_1_0 := v_1_1_1.Args[0]
if v_1_1_1_0.Op != OpPPC64ANDCCconst || auxIntToInt64(v_1_1_1_0.AuxInt) != 31 || y != v_1_1_1_0.Args[0] {
continue
}
v.reset(OpPPC64ROTLW)
v.AddArg2(x, y)
return true
}
break
}
// match: (ADD x (MOVDconst [c]))
// cond: is32Bit(c)
// result: (ADDconst [c] x)
@ -11642,236 +11412,6 @@ func rewriteValuePPC64_OpPPC64OR(v *Value) bool {
b := v.Block
config := b.Func.Config
typ := &b.Func.Config.Types
// match: ( OR (SLDconst x [c]) (SRDconst x [d]))
// cond: d == 64-c
// result: (ROTLconst [c] x)
for {
for _i0 := 0; _i0 <= 1; _i0, v_0, v_1 = _i0+1, v_1, v_0 {
if v_0.Op != OpPPC64SLDconst {
continue
}
c := auxIntToInt64(v_0.AuxInt)
x := v_0.Args[0]
if v_1.Op != OpPPC64SRDconst {
continue
}
d := auxIntToInt64(v_1.AuxInt)
if x != v_1.Args[0] || !(d == 64-c) {
continue
}
v.reset(OpPPC64ROTLconst)
v.AuxInt = int64ToAuxInt(c)
v.AddArg(x)
return true
}
break
}
// match: ( OR (SLWconst x [c]) (SRWconst x [d]))
// cond: d == 32-c
// result: (ROTLWconst [c] x)
for {
for _i0 := 0; _i0 <= 1; _i0, v_0, v_1 = _i0+1, v_1, v_0 {
if v_0.Op != OpPPC64SLWconst {
continue
}
c := auxIntToInt64(v_0.AuxInt)
x := v_0.Args[0]
if v_1.Op != OpPPC64SRWconst {
continue
}
d := auxIntToInt64(v_1.AuxInt)
if x != v_1.Args[0] || !(d == 32-c) {
continue
}
v.reset(OpPPC64ROTLWconst)
v.AuxInt = int64ToAuxInt(c)
v.AddArg(x)
return true
}
break
}
// match: ( OR (SLD x (Select0 (ANDCCconst [63] y))) (SRD x (SUB <typ.UInt> (MOVDconst [64]) (Select0 <typ.UInt> (ANDCCconst [63] y)))))
// result: (ROTL x y)
for {
for _i0 := 0; _i0 <= 1; _i0, v_0, v_1 = _i0+1, v_1, v_0 {
if v_0.Op != OpPPC64SLD {
continue
}
_ = v_0.Args[1]
x := v_0.Args[0]
v_0_1 := v_0.Args[1]
if v_0_1.Op != OpSelect0 {
continue
}
v_0_1_0 := v_0_1.Args[0]
if v_0_1_0.Op != OpPPC64ANDCCconst || auxIntToInt64(v_0_1_0.AuxInt) != 63 {
continue
}
y := v_0_1_0.Args[0]
if v_1.Op != OpPPC64SRD {
continue
}
_ = v_1.Args[1]
if x != v_1.Args[0] {
continue
}
v_1_1 := v_1.Args[1]
if v_1_1.Op != OpPPC64SUB || v_1_1.Type != typ.UInt {
continue
}
_ = v_1_1.Args[1]
v_1_1_0 := v_1_1.Args[0]
if v_1_1_0.Op != OpPPC64MOVDconst || auxIntToInt64(v_1_1_0.AuxInt) != 64 {
continue
}
v_1_1_1 := v_1_1.Args[1]
if v_1_1_1.Op != OpSelect0 || v_1_1_1.Type != typ.UInt {
continue
}
v_1_1_1_0 := v_1_1_1.Args[0]
if v_1_1_1_0.Op != OpPPC64ANDCCconst || auxIntToInt64(v_1_1_1_0.AuxInt) != 63 || y != v_1_1_1_0.Args[0] {
continue
}
v.reset(OpPPC64ROTL)
v.AddArg2(x, y)
return true
}
break
}
// match: ( OR (SLD x (Select0 (ANDCCconst [63] y))) (SRD x (SUBFCconst <typ.UInt> [64] (Select0 <typ.UInt> (ANDCCconst [63] y)))))
// result: (ROTL x y)
for {
for _i0 := 0; _i0 <= 1; _i0, v_0, v_1 = _i0+1, v_1, v_0 {
if v_0.Op != OpPPC64SLD {
continue
}
_ = v_0.Args[1]
x := v_0.Args[0]
v_0_1 := v_0.Args[1]
if v_0_1.Op != OpSelect0 {
continue
}
v_0_1_0 := v_0_1.Args[0]
if v_0_1_0.Op != OpPPC64ANDCCconst || auxIntToInt64(v_0_1_0.AuxInt) != 63 {
continue
}
y := v_0_1_0.Args[0]
if v_1.Op != OpPPC64SRD {
continue
}
_ = v_1.Args[1]
if x != v_1.Args[0] {
continue
}
v_1_1 := v_1.Args[1]
if v_1_1.Op != OpPPC64SUBFCconst || v_1_1.Type != typ.UInt || auxIntToInt64(v_1_1.AuxInt) != 64 {
continue
}
v_1_1_0 := v_1_1.Args[0]
if v_1_1_0.Op != OpSelect0 || v_1_1_0.Type != typ.UInt {
continue
}
v_1_1_0_0 := v_1_1_0.Args[0]
if v_1_1_0_0.Op != OpPPC64ANDCCconst || auxIntToInt64(v_1_1_0_0.AuxInt) != 63 || y != v_1_1_0_0.Args[0] {
continue
}
v.reset(OpPPC64ROTL)
v.AddArg2(x, y)
return true
}
break
}
// match: ( OR (SLW x (Select0 (ANDCCconst [31] y))) (SRW x (SUBFCconst <typ.UInt> [32] (Select0 <typ.UInt> (ANDCCconst [31] y)))))
// result: (ROTLW x y)
for {
for _i0 := 0; _i0 <= 1; _i0, v_0, v_1 = _i0+1, v_1, v_0 {
if v_0.Op != OpPPC64SLW {
continue
}
_ = v_0.Args[1]
x := v_0.Args[0]
v_0_1 := v_0.Args[1]
if v_0_1.Op != OpSelect0 {
continue
}
v_0_1_0 := v_0_1.Args[0]
if v_0_1_0.Op != OpPPC64ANDCCconst || auxIntToInt64(v_0_1_0.AuxInt) != 31 {
continue
}
y := v_0_1_0.Args[0]
if v_1.Op != OpPPC64SRW {
continue
}
_ = v_1.Args[1]
if x != v_1.Args[0] {
continue
}
v_1_1 := v_1.Args[1]
if v_1_1.Op != OpPPC64SUBFCconst || v_1_1.Type != typ.UInt || auxIntToInt64(v_1_1.AuxInt) != 32 {
continue
}
v_1_1_0 := v_1_1.Args[0]
if v_1_1_0.Op != OpSelect0 || v_1_1_0.Type != typ.UInt {
continue
}
v_1_1_0_0 := v_1_1_0.Args[0]
if v_1_1_0_0.Op != OpPPC64ANDCCconst || auxIntToInt64(v_1_1_0_0.AuxInt) != 31 || y != v_1_1_0_0.Args[0] {
continue
}
v.reset(OpPPC64ROTLW)
v.AddArg2(x, y)
return true
}
break
}
// match: ( OR (SLW x (Select0 (ANDCCconst [31] y))) (SRW x (SUB <typ.UInt> (MOVDconst [32]) (Select0 <typ.UInt> (ANDCCconst [31] y)))))
// result: (ROTLW x y)
for {
for _i0 := 0; _i0 <= 1; _i0, v_0, v_1 = _i0+1, v_1, v_0 {
if v_0.Op != OpPPC64SLW {
continue
}
_ = v_0.Args[1]
x := v_0.Args[0]
v_0_1 := v_0.Args[1]
if v_0_1.Op != OpSelect0 {
continue
}
v_0_1_0 := v_0_1.Args[0]
if v_0_1_0.Op != OpPPC64ANDCCconst || auxIntToInt64(v_0_1_0.AuxInt) != 31 {
continue
}
y := v_0_1_0.Args[0]
if v_1.Op != OpPPC64SRW {
continue
}
_ = v_1.Args[1]
if x != v_1.Args[0] {
continue
}
v_1_1 := v_1.Args[1]
if v_1_1.Op != OpPPC64SUB || v_1_1.Type != typ.UInt {
continue
}
_ = v_1_1.Args[1]
v_1_1_0 := v_1_1.Args[0]
if v_1_1_0.Op != OpPPC64MOVDconst || auxIntToInt64(v_1_1_0.AuxInt) != 32 {
continue
}
v_1_1_1 := v_1_1.Args[1]
if v_1_1_1.Op != OpSelect0 || v_1_1_1.Type != typ.UInt {
continue
}
v_1_1_1_0 := v_1_1_1.Args[0]
if v_1_1_1_0.Op != OpPPC64ANDCCconst || auxIntToInt64(v_1_1_1_0.AuxInt) != 31 || y != v_1_1_1_0.Args[0] {
continue
}
v.reset(OpPPC64ROTLW)
v.AddArg2(x, y)
return true
}
break
}
// match: (OR x (NOR y y))
// result: (ORN x y)
for {
@ -13950,238 +13490,6 @@ func rewriteValuePPC64_OpPPC64SUBFCconst(v *Value) bool {
func rewriteValuePPC64_OpPPC64XOR(v *Value) bool {
v_1 := v.Args[1]
v_0 := v.Args[0]
b := v.Block
typ := &b.Func.Config.Types
// match: (XOR (SLDconst x [c]) (SRDconst x [d]))
// cond: d == 64-c
// result: (ROTLconst [c] x)
for {
for _i0 := 0; _i0 <= 1; _i0, v_0, v_1 = _i0+1, v_1, v_0 {
if v_0.Op != OpPPC64SLDconst {
continue
}
c := auxIntToInt64(v_0.AuxInt)
x := v_0.Args[0]
if v_1.Op != OpPPC64SRDconst {
continue
}
d := auxIntToInt64(v_1.AuxInt)
if x != v_1.Args[0] || !(d == 64-c) {
continue
}
v.reset(OpPPC64ROTLconst)
v.AuxInt = int64ToAuxInt(c)
v.AddArg(x)
return true
}
break
}
// match: (XOR (SLWconst x [c]) (SRWconst x [d]))
// cond: d == 32-c
// result: (ROTLWconst [c] x)
for {
for _i0 := 0; _i0 <= 1; _i0, v_0, v_1 = _i0+1, v_1, v_0 {
if v_0.Op != OpPPC64SLWconst {
continue
}
c := auxIntToInt64(v_0.AuxInt)
x := v_0.Args[0]
if v_1.Op != OpPPC64SRWconst {
continue
}
d := auxIntToInt64(v_1.AuxInt)
if x != v_1.Args[0] || !(d == 32-c) {
continue
}
v.reset(OpPPC64ROTLWconst)
v.AuxInt = int64ToAuxInt(c)
v.AddArg(x)
return true
}
break
}
// match: (XOR (SLD x (Select0 (ANDCCconst [63] y))) (SRD x (SUB <typ.UInt> (MOVDconst [64]) (Select0 <typ.UInt> (ANDCCconst [63] y)))))
// result: (ROTL x y)
for {
for _i0 := 0; _i0 <= 1; _i0, v_0, v_1 = _i0+1, v_1, v_0 {
if v_0.Op != OpPPC64SLD {
continue
}
_ = v_0.Args[1]
x := v_0.Args[0]
v_0_1 := v_0.Args[1]
if v_0_1.Op != OpSelect0 {
continue
}
v_0_1_0 := v_0_1.Args[0]
if v_0_1_0.Op != OpPPC64ANDCCconst || auxIntToInt64(v_0_1_0.AuxInt) != 63 {
continue
}
y := v_0_1_0.Args[0]
if v_1.Op != OpPPC64SRD {
continue
}
_ = v_1.Args[1]
if x != v_1.Args[0] {
continue
}
v_1_1 := v_1.Args[1]
if v_1_1.Op != OpPPC64SUB || v_1_1.Type != typ.UInt {
continue
}
_ = v_1_1.Args[1]
v_1_1_0 := v_1_1.Args[0]
if v_1_1_0.Op != OpPPC64MOVDconst || auxIntToInt64(v_1_1_0.AuxInt) != 64 {
continue
}
v_1_1_1 := v_1_1.Args[1]
if v_1_1_1.Op != OpSelect0 || v_1_1_1.Type != typ.UInt {
continue
}
v_1_1_1_0 := v_1_1_1.Args[0]
if v_1_1_1_0.Op != OpPPC64ANDCCconst || auxIntToInt64(v_1_1_1_0.AuxInt) != 63 || y != v_1_1_1_0.Args[0] {
continue
}
v.reset(OpPPC64ROTL)
v.AddArg2(x, y)
return true
}
break
}
// match: (XOR (SLD x (Select0 (ANDCCconst [63] y))) (SRD x (SUBFCconst <typ.UInt> [64] (Select0 <typ.UInt> (ANDCCconst [63] y)))))
// result: (ROTL x y)
for {
for _i0 := 0; _i0 <= 1; _i0, v_0, v_1 = _i0+1, v_1, v_0 {
if v_0.Op != OpPPC64SLD {
continue
}
_ = v_0.Args[1]
x := v_0.Args[0]
v_0_1 := v_0.Args[1]
if v_0_1.Op != OpSelect0 {
continue
}
v_0_1_0 := v_0_1.Args[0]
if v_0_1_0.Op != OpPPC64ANDCCconst || auxIntToInt64(v_0_1_0.AuxInt) != 63 {
continue
}
y := v_0_1_0.Args[0]
if v_1.Op != OpPPC64SRD {
continue
}
_ = v_1.Args[1]
if x != v_1.Args[0] {
continue
}
v_1_1 := v_1.Args[1]
if v_1_1.Op != OpPPC64SUBFCconst || v_1_1.Type != typ.UInt || auxIntToInt64(v_1_1.AuxInt) != 64 {
continue
}
v_1_1_0 := v_1_1.Args[0]
if v_1_1_0.Op != OpSelect0 || v_1_1_0.Type != typ.UInt {
continue
}
v_1_1_0_0 := v_1_1_0.Args[0]
if v_1_1_0_0.Op != OpPPC64ANDCCconst || auxIntToInt64(v_1_1_0_0.AuxInt) != 63 || y != v_1_1_0_0.Args[0] {
continue
}
v.reset(OpPPC64ROTL)
v.AddArg2(x, y)
return true
}
break
}
// match: (XOR (SLW x (Select0 (ANDCCconst [31] y))) (SRW x (SUBFCconst <typ.UInt> [32] (Select0 <typ.UInt> (ANDCCconst [31] y)))))
// result: (ROTLW x y)
for {
for _i0 := 0; _i0 <= 1; _i0, v_0, v_1 = _i0+1, v_1, v_0 {
if v_0.Op != OpPPC64SLW {
continue
}
_ = v_0.Args[1]
x := v_0.Args[0]
v_0_1 := v_0.Args[1]
if v_0_1.Op != OpSelect0 {
continue
}
v_0_1_0 := v_0_1.Args[0]
if v_0_1_0.Op != OpPPC64ANDCCconst || auxIntToInt64(v_0_1_0.AuxInt) != 31 {
continue
}
y := v_0_1_0.Args[0]
if v_1.Op != OpPPC64SRW {
continue
}
_ = v_1.Args[1]
if x != v_1.Args[0] {
continue
}
v_1_1 := v_1.Args[1]
if v_1_1.Op != OpPPC64SUBFCconst || v_1_1.Type != typ.UInt || auxIntToInt64(v_1_1.AuxInt) != 32 {
continue
}
v_1_1_0 := v_1_1.Args[0]
if v_1_1_0.Op != OpSelect0 || v_1_1_0.Type != typ.UInt {
continue
}
v_1_1_0_0 := v_1_1_0.Args[0]
if v_1_1_0_0.Op != OpPPC64ANDCCconst || auxIntToInt64(v_1_1_0_0.AuxInt) != 31 || y != v_1_1_0_0.Args[0] {
continue
}
v.reset(OpPPC64ROTLW)
v.AddArg2(x, y)
return true
}
break
}
// match: (XOR (SLW x (Select0 (ANDCCconst [31] y))) (SRW x (SUB <typ.UInt> (MOVDconst [32]) (Select0 <typ.UInt> (ANDCCconst [31] y)))))
// result: (ROTLW x y)
for {
for _i0 := 0; _i0 <= 1; _i0, v_0, v_1 = _i0+1, v_1, v_0 {
if v_0.Op != OpPPC64SLW {
continue
}
_ = v_0.Args[1]
x := v_0.Args[0]
v_0_1 := v_0.Args[1]
if v_0_1.Op != OpSelect0 {
continue
}
v_0_1_0 := v_0_1.Args[0]
if v_0_1_0.Op != OpPPC64ANDCCconst || auxIntToInt64(v_0_1_0.AuxInt) != 31 {
continue
}
y := v_0_1_0.Args[0]
if v_1.Op != OpPPC64SRW {
continue
}
_ = v_1.Args[1]
if x != v_1.Args[0] {
continue
}
v_1_1 := v_1.Args[1]
if v_1_1.Op != OpPPC64SUB || v_1_1.Type != typ.UInt {
continue
}
_ = v_1_1.Args[1]
v_1_1_0 := v_1_1.Args[0]
if v_1_1_0.Op != OpPPC64MOVDconst || auxIntToInt64(v_1_1_0.AuxInt) != 32 {
continue
}
v_1_1_1 := v_1_1.Args[1]
if v_1_1_1.Op != OpSelect0 || v_1_1_1.Type != typ.UInt {
continue
}
v_1_1_1_0 := v_1_1_1.Args[0]
if v_1_1_1_0.Op != OpPPC64ANDCCconst || auxIntToInt64(v_1_1_1_0.AuxInt) != 31 || y != v_1_1_1_0.Args[0] {
continue
}
v.reset(OpPPC64ROTLW)
v.AddArg2(x, y)
return true
}
break
}
// match: (XOR (MOVDconst [c]) (MOVDconst [d]))
// result: (MOVDconst [c^d])
for {
@ -14461,58 +13769,6 @@ func rewriteValuePPC64_OpRotateLeft16(v *Value) bool {
}
return false
}
func rewriteValuePPC64_OpRotateLeft32(v *Value) bool {
v_1 := v.Args[1]
v_0 := v.Args[0]
// match: (RotateLeft32 x (MOVDconst [c]))
// result: (ROTLWconst [c&31] x)
for {
x := v_0
if v_1.Op != OpPPC64MOVDconst {
break
}
c := auxIntToInt64(v_1.AuxInt)
v.reset(OpPPC64ROTLWconst)
v.AuxInt = int64ToAuxInt(c & 31)
v.AddArg(x)
return true
}
// match: (RotateLeft32 x y)
// result: (ROTLW x y)
for {
x := v_0
y := v_1
v.reset(OpPPC64ROTLW)
v.AddArg2(x, y)
return true
}
}
func rewriteValuePPC64_OpRotateLeft64(v *Value) bool {
v_1 := v.Args[1]
v_0 := v.Args[0]
// match: (RotateLeft64 x (MOVDconst [c]))
// result: (ROTLconst [c&63] x)
for {
x := v_0
if v_1.Op != OpPPC64MOVDconst {
break
}
c := auxIntToInt64(v_1.AuxInt)
v.reset(OpPPC64ROTLconst)
v.AuxInt = int64ToAuxInt(c & 63)
v.AddArg(x)
return true
}
// match: (RotateLeft64 x y)
// result: (ROTL x y)
for {
x := v_0
y := v_1
v.reset(OpPPC64ROTL)
v.AddArg2(x, y)
return true
}
}
func rewriteValuePPC64_OpRotateLeft8(v *Value) bool {
v_1 := v.Args[1]
v_0 := v.Args[0]

114
src/cmd/compile/internal/ssa/rewriteS390X.go
View file

@ -5280,25 +5280,6 @@ func rewriteValueS390X_OpS390XADD(v *Value) bool {
}
break
}
// match: (ADD (SLDconst x [c]) (SRDconst x [64-c]))
// result: (RISBGZ x {s390x.NewRotateParams(0, 63, c)})
for {
for _i0 := 0; _i0 <= 1; _i0, v_0, v_1 = _i0+1, v_1, v_0 {
if v_0.Op != OpS390XSLDconst {
continue
}
c := auxIntToUint8(v_0.AuxInt)
x := v_0.Args[0]
if v_1.Op != OpS390XSRDconst || auxIntToUint8(v_1.AuxInt) != 64-c || x != v_1.Args[0] {
continue
}
v.reset(OpS390XRISBGZ)
v.Aux = s390xRotateParamsToAux(s390x.NewRotateParams(0, 63, c))
v.AddArg(x)
return true
}
break
}
// match: (ADD idx (MOVDaddr [c] {s} ptr))
// cond: ptr.Op != OpSB
// result: (MOVDaddridx [c] {s} ptr idx)
@ -5473,25 +5454,6 @@ func rewriteValueS390X_OpS390XADDW(v *Value) bool {
}
break
}
// match: (ADDW (SLWconst x [c]) (SRWconst x [32-c]))
// result: (RLLconst x [c])
for {
for _i0 := 0; _i0 <= 1; _i0, v_0, v_1 = _i0+1, v_1, v_0 {
if v_0.Op != OpS390XSLWconst {
continue
}
c := auxIntToUint8(v_0.AuxInt)
x := v_0.Args[0]
if v_1.Op != OpS390XSRWconst || auxIntToUint8(v_1.AuxInt) != 32-c || x != v_1.Args[0] {
continue
}
v.reset(OpS390XRLLconst)
v.AuxInt = uint8ToAuxInt(c)
v.AddArg(x)
return true
}
break
}
// match: (ADDW x (NEGW y))
// result: (SUBW x y)
for {
@ -11689,25 +11651,6 @@ func rewriteValueS390X_OpS390XOR(v *Value) bool {
}
break
}
// match: (OR (SLDconst x [c]) (SRDconst x [64-c]))
// result: (RISBGZ x {s390x.NewRotateParams(0, 63, c)})
for {
for _i0 := 0; _i0 <= 1; _i0, v_0, v_1 = _i0+1, v_1, v_0 {
if v_0.Op != OpS390XSLDconst {
continue
}
c := auxIntToUint8(v_0.AuxInt)
x := v_0.Args[0]
if v_1.Op != OpS390XSRDconst || auxIntToUint8(v_1.AuxInt) != 64-c || x != v_1.Args[0] {
continue
}
v.reset(OpS390XRISBGZ)
v.Aux = s390xRotateParamsToAux(s390x.NewRotateParams(0, 63, c))
v.AddArg(x)
return true
}
break
}
// match: (OR (MOVDconst [-1<<63]) (LGDR <t> x))
// result: (LGDR <t> (LNDFR <x.Type> x))
for {
@ -12387,25 +12330,6 @@ func rewriteValueS390X_OpS390XORW(v *Value) bool {
}
break
}
// match: (ORW (SLWconst x [c]) (SRWconst x [32-c]))
// result: (RLLconst x [c])
for {
for _i0 := 0; _i0 <= 1; _i0, v_0, v_1 = _i0+1, v_1, v_0 {
if v_0.Op != OpS390XSLWconst {
continue
}
c := auxIntToUint8(v_0.AuxInt)
x := v_0.Args[0]
if v_1.Op != OpS390XSRWconst || auxIntToUint8(v_1.AuxInt) != 32-c || x != v_1.Args[0] {
continue
}
v.reset(OpS390XRLLconst)
v.AuxInt = uint8ToAuxInt(c)
v.AddArg(x)
return true
}
break
}
// match: (ORW x x)
// result: x
for {
@ -14972,25 +14896,6 @@ func rewriteValueS390X_OpS390XXOR(v *Value) bool {
}
break
}
// match: (XOR (SLDconst x [c]) (SRDconst x [64-c]))
// result: (RISBGZ x {s390x.NewRotateParams(0, 63, c)})
for {
for _i0 := 0; _i0 <= 1; _i0, v_0, v_1 = _i0+1, v_1, v_0 {
if v_0.Op != OpS390XSLDconst {
continue
}
c := auxIntToUint8(v_0.AuxInt)
x := v_0.Args[0]
if v_1.Op != OpS390XSRDconst || auxIntToUint8(v_1.AuxInt) != 64-c || x != v_1.Args[0] {
continue
}
v.reset(OpS390XRISBGZ)
v.Aux = s390xRotateParamsToAux(s390x.NewRotateParams(0, 63, c))
v.AddArg(x)
return true
}
break
}
// match: (XOR (MOVDconst [c]) (MOVDconst [d]))
// result: (MOVDconst [c^d])
for {
@ -15068,25 +14973,6 @@ func rewriteValueS390X_OpS390XXORW(v *Value) bool {
}
break
}
// match: (XORW (SLWconst x [c]) (SRWconst x [32-c]))
// result: (RLLconst x [c])
for {
for _i0 := 0; _i0 <= 1; _i0, v_0, v_1 = _i0+1, v_1, v_0 {
if v_0.Op != OpS390XSLWconst {
continue
}
c := auxIntToUint8(v_0.AuxInt)
x := v_0.Args[0]
if v_1.Op != OpS390XSRWconst || auxIntToUint8(v_1.AuxInt) != 32-c || x != v_1.Args[0] {
continue
}
v.reset(OpS390XRLLconst)
v.AuxInt = uint8ToAuxInt(c)
v.AddArg(x)
return true
}
break
}
// match: (XORW x x)
// result: (MOVDconst [0])
for {

76
src/cmd/compile/internal/ssa/rewritedec64.go
View file

@ -66,6 +66,14 @@ func rewriteValuedec64(v *Value) bool {
return rewriteValuedec64_OpOr32(v)
case OpOr64:
return rewriteValuedec64_OpOr64(v)
case OpRotateLeft16:
return rewriteValuedec64_OpRotateLeft16(v)
case OpRotateLeft32:
return rewriteValuedec64_OpRotateLeft32(v)
case OpRotateLeft64:
return rewriteValuedec64_OpRotateLeft64(v)
case OpRotateLeft8:
return rewriteValuedec64_OpRotateLeft8(v)
case OpRsh16Ux64:
return rewriteValuedec64_OpRsh16Ux64(v)
case OpRsh16x64:
@ -1266,6 +1274,74 @@ func rewriteValuedec64_OpOr64(v *Value) bool {
return true
}
}
func rewriteValuedec64_OpRotateLeft16(v *Value) bool {
v_1 := v.Args[1]
v_0 := v.Args[0]
// match: (RotateLeft16 x (Int64Make hi lo))
// result: (RotateLeft16 x lo)
for {
x := v_0
if v_1.Op != OpInt64Make {
break
}
lo := v_1.Args[1]
v.reset(OpRotateLeft16)
v.AddArg2(x, lo)
return true
}
return false
}
func rewriteValuedec64_OpRotateLeft32(v *Value) bool {
v_1 := v.Args[1]
v_0 := v.Args[0]
// match: (RotateLeft32 x (Int64Make hi lo))
// result: (RotateLeft32 x lo)
for {
x := v_0
if v_1.Op != OpInt64Make {
break
}
lo := v_1.Args[1]
v.reset(OpRotateLeft32)
v.AddArg2(x, lo)
return true
}
return false
}
func rewriteValuedec64_OpRotateLeft64(v *Value) bool {
v_1 := v.Args[1]
v_0 := v.Args[0]
// match: (RotateLeft64 x (Int64Make hi lo))
// result: (RotateLeft64 x lo)
for {
x := v_0
if v_1.Op != OpInt64Make {
break
}
lo := v_1.Args[1]
v.reset(OpRotateLeft64)
v.AddArg2(x, lo)
return true
}
return false
}
func rewriteValuedec64_OpRotateLeft8(v *Value) bool {
v_1 := v.Args[1]
v_0 := v.Args[0]
// match: (RotateLeft8 x (Int64Make hi lo))
// result: (RotateLeft8 x lo)
for {
x := v_0
if v_1.Op != OpInt64Make {
break
}
lo := v_1.Args[1]
v.reset(OpRotateLeft8)
v.AddArg2(x, lo)
return true
}
return false
}
func rewriteValuedec64_OpRsh16Ux64(v *Value) bool {
v_1 := v.Args[1]
v_0 := v.Args[0]

5504
src/cmd/compile/internal/ssa/rewritegeneric.go
View file

File diff suppressed because it is too large Load diff

22
src/cmd/compile/internal/test/shift_test.go
View file

@ -1039,3 +1039,25 @@ func BenchmarkShiftArithmeticRight(b *testing.B) {
}
shiftSink64 = x
}
//go:noinline
func incorrectRotate1(x, c uint64) uint64 {
// This should not compile to a rotate instruction.
return x<<c | x>>(64-c)
}
//go:noinline
func incorrectRotate2(x uint64) uint64 {
var c uint64 = 66
// This should not compile to a rotate instruction.
return x<<c | x>>(64-c)
}
func TestIncorrectRotate(t *testing.T) {
if got := incorrectRotate1(1, 66); got != 0 {
t.Errorf("got %x want 0", got)
}
if got := incorrectRotate2(1); got != 0 {
t.Errorf("got %x want 0", got)
}
}

1
src/cmd/compile/internal/x86/ssa.go
View file

@ -158,6 +158,7 @@ func ssaGenValue(s *ssagen.State, v *ssa.Value) {
ssa.Op386SHLL,
ssa.Op386SHRL, ssa.Op386SHRW, ssa.Op386SHRB,
ssa.Op386SARL, ssa.Op386SARW, ssa.Op386SARB,
ssa.Op386ROLL, ssa.Op386ROLW, ssa.Op386ROLB,
ssa.Op386ADDSS, ssa.Op386ADDSD, ssa.Op386SUBSS, ssa.Op386SUBSD,
ssa.Op386MULSS, ssa.Op386MULSD, ssa.Op386DIVSS, ssa.Op386DIVSD,
ssa.Op386PXOR,

40
test/codegen/rotate.go
View file

@ -39,8 +39,8 @@ func rot64(x uint64) uint64 {
// s390x:"RISBGZ\t[$]0, [$]63, [$]10, "
// ppc64:"ROTL\t[$]10"
// ppc64le:"ROTL\t[$]10"
// arm64:"ROR\t[$]57" // TODO this is not great line numbering, but then again, the instruction did appear
// s390x:"RISBGZ\t[$]0, [$]63, [$]7, " // TODO ditto
// arm64:"ROR\t[$]54"
// s390x:"RISBGZ\t[$]0, [$]63, [$]10, "
a += bits.RotateLeft64(x, 10)
return a
@ -77,8 +77,8 @@ func rot32(x uint32) uint32 {
// s390x:"RLL\t[$]10"
// ppc64:"ROTLW\t[$]10"
// ppc64le:"ROTLW\t[$]10"
// arm64:"RORW\t[$]25" // TODO this is not great line numbering, but then again, the instruction did appear
// s390x:"RLL\t[$]7" // TODO ditto
// arm64:"RORW\t[$]22"
// s390x:"RLL\t[$]10"
a += bits.RotateLeft32(x, 10)
return a
@ -123,12 +123,16 @@ func rot64nc(x uint64, z uint) uint64 {
z &= 63
// amd64:"ROLQ"
// ppc64:"ROTL"
// ppc64le:"ROTL"
// amd64:"ROLQ",-"AND"
// arm64:"ROR","NEG",-"AND"
// ppc64:"ROTL",-"NEG",-"AND"
// ppc64le:"ROTL",-"NEG",-"AND"
a += x<<z | x>>(64-z)
// amd64:"RORQ"
// amd64:"RORQ",-"AND"
// arm64:"ROR",-"NEG",-"AND"
// ppc64:"ROTL","NEG",-"AND"
// ppc64le:"ROTL","NEG",-"AND"
a += x>>z | x<<(64-z)
return a
@ -139,12 +143,16 @@ func rot32nc(x uint32, z uint) uint32 {
z &= 31
// amd64:"ROLL"
// ppc64:"ROTLW"
// ppc64le:"ROTLW"
// amd64:"ROLL",-"AND"
// arm64:"ROR","NEG",-"AND"
// ppc64:"ROTLW",-"NEG",-"AND"
// ppc64le:"ROTLW",-"NEG",-"AND"
a += x<<z | x>>(32-z)
// amd64:"RORL"
// amd64:"RORL",-"AND"
// arm64:"ROR",-"NEG",-"AND"
// ppc64:"ROTLW","NEG",-"AND"
// ppc64le:"ROTLW","NEG",-"AND"
a += x>>z | x<<(32-z)
return a
@ -155,10 +163,10 @@ func rot16nc(x uint16, z uint) uint16 {
z &= 15
// amd64:"ROLW"
// amd64:"ROLW",-"ANDQ"
a += x<<z | x>>(16-z)
// amd64:"RORW"
// amd64:"RORW",-"ANDQ"
a += x>>z | x<<(16-z)
return a
@ -169,10 +177,10 @@ func rot8nc(x uint8, z uint) uint8 {
z &= 7
// amd64:"ROLB"
// amd64:"ROLB",-"ANDQ"
a += x<<z | x>>(8-z)
// amd64:"RORB"
// amd64:"RORB",-"ANDQ"
a += x>>z | x<<(8-z)
return a