1
0
mirror of https://github.com/golang/go synced 2024-11-23 17:50:06 -07:00

cmd/compile: switch to typed aux for 386 lowering rules

Convert all the 386 lowering rules to the typed aux form.

Passes

  GOARCH=386 gotip build -toolexec 'toolstash -cmp' -a std

Change-Id: I15256f20bc4442391755e6fffb8206dcaab94830
Reviewed-on: https://go-review.googlesource.com/c/go/+/228818
Reviewed-by: Keith Randall <khr@golang.org>
This commit is contained in:
Alberto Donizetti 2020-04-17 23:00:35 +02:00
parent af9ab6b2e8
commit bbaae9c43d
2 changed files with 483 additions and 429 deletions

View File

@ -3,243 +3,243 @@
// license that can be found in the LICENSE file.
// Lowering arithmetic
(Add(Ptr|32|16|8) ...) -> (ADDL ...)
(Add(32|64)F ...) -> (ADDS(S|D) ...)
(Add32carry ...) -> (ADDLcarry ...)
(Add32withcarry ...) -> (ADCL ...)
(Add(Ptr|32|16|8) ...) => (ADDL ...)
(Add(32|64)F ...) => (ADDS(S|D) ...)
(Add32carry ...) => (ADDLcarry ...)
(Add32withcarry ...) => (ADCL ...)
(Sub(Ptr|32|16|8) ...) -> (SUBL ...)
(Sub(32|64)F ...) -> (SUBS(S|D) ...)
(Sub32carry ...) -> (SUBLcarry ...)
(Sub32withcarry ...) -> (SBBL ...)
(Sub(Ptr|32|16|8) ...) => (SUBL ...)
(Sub(32|64)F ...) => (SUBS(S|D) ...)
(Sub32carry ...) => (SUBLcarry ...)
(Sub32withcarry ...) => (SBBL ...)
(Mul(32|16|8) ...) -> (MULL ...)
(Mul(32|64)F ...) -> (MULS(S|D) ...)
(Mul32uhilo ...) -> (MULLQU ...)
(Mul(32|16|8) ...) => (MULL ...)
(Mul(32|64)F ...) => (MULS(S|D) ...)
(Mul32uhilo ...) => (MULLQU ...)
(Select0 (Mul32uover x y)) -> (Select0 <typ.UInt32> (MULLU x y))
(Select1 (Mul32uover x y)) -> (SETO (Select1 <types.TypeFlags> (MULLU x y)))
(Select0 (Mul32uover x y)) => (Select0 <typ.UInt32> (MULLU x y))
(Select1 (Mul32uover x y)) => (SETO (Select1 <types.TypeFlags> (MULLU x y)))
(Avg32u ...) -> (AVGLU ...)
(Avg32u ...) => (AVGLU ...)
(Div(32|64)F ...) -> (DIVS(S|D) ...)
(Div(32|32u|16|16u) ...) -> (DIV(L|LU|W|WU) ...)
(Div8 x y) -> (DIVW (SignExt8to16 x) (SignExt8to16 y))
(Div8u x y) -> (DIVWU (ZeroExt8to16 x) (ZeroExt8to16 y))
(Div(32|64)F ...) => (DIVS(S|D) ...)
(Div(32|32u|16|16u) ...) => (DIV(L|LU|W|WU) ...)
(Div8 x y) => (DIVW (SignExt8to16 x) (SignExt8to16 y))
(Div8u x y) => (DIVWU (ZeroExt8to16 x) (ZeroExt8to16 y))
(Hmul(32|32u) ...) -> (HMUL(L|LU) ...)
(Hmul(32|32u) ...) => (HMUL(L|LU) ...)
(Mod(32|32u|16|16u) ...) -> (MOD(L|LU|W|WU) ...)
(Mod8 x y) -> (MODW (SignExt8to16 x) (SignExt8to16 y))
(Mod8u x y) -> (MODWU (ZeroExt8to16 x) (ZeroExt8to16 y))
(Mod(32|32u|16|16u) ...) => (MOD(L|LU|W|WU) ...)
(Mod8 x y) => (MODW (SignExt8to16 x) (SignExt8to16 y))
(Mod8u x y) => (MODWU (ZeroExt8to16 x) (ZeroExt8to16 y))
(And(32|16|8) ...) -> (ANDL ...)
(Or(32|16|8) ...) -> (ORL ...)
(Xor(32|16|8) ...) -> (XORL ...)
(And(32|16|8) ...) => (ANDL ...)
(Or(32|16|8) ...) => (ORL ...)
(Xor(32|16|8) ...) => (XORL ...)
(Neg(32|16|8) ...) -> (NEGL ...)
(Neg32F x) && !config.use387 -> (PXOR x (MOVSSconst <typ.Float32> [auxFrom32F(float32(math.Copysign(0, -1)))]))
(Neg64F x) && !config.use387 -> (PXOR x (MOVSDconst <typ.Float64> [auxFrom64F(math.Copysign(0, -1))]))
(Neg32F x) && config.use387 -> (FCHS x)
(Neg64F x) && config.use387 -> (FCHS x)
(Neg(32|16|8) ...) => (NEGL ...)
(Neg32F x) && !config.use387 => (PXOR x (MOVSSconst <typ.Float32> [float32(math.Copysign(0, -1))]))
(Neg64F x) && !config.use387 => (PXOR x (MOVSDconst <typ.Float64> [math.Copysign(0, -1)]))
(Neg32F x) && config.use387 => (FCHS x)
(Neg64F x) && config.use387 => (FCHS x)
(Com(32|16|8) ...) -> (NOTL ...)
(Com(32|16|8) ...) => (NOTL ...)
// Lowering boolean ops
(AndB ...) -> (ANDL ...)
(OrB ...) -> (ORL ...)
(Not x) -> (XORLconst [1] x)
(AndB ...) => (ANDL ...)
(OrB ...) => (ORL ...)
(Not x) => (XORLconst [1] x)
// Lowering pointer arithmetic
(OffPtr ...) -> (ADDLconst ...)
(OffPtr [off] ptr) => (ADDLconst [int32(off)] ptr)
(Bswap32 ...) -> (BSWAPL ...)
(Bswap32 ...) => (BSWAPL ...)
(Sqrt ...) -> (SQRTSD ...)
(Sqrt ...) => (SQRTSD ...)
(Ctz16 x) -> (BSFL (ORLconst <typ.UInt32> [0x10000] x))
(Ctz16NonZero ...) -> (BSFL ...)
(Ctz16 x) => (BSFL (ORLconst <typ.UInt32> [0x10000] x))
(Ctz16NonZero ...) => (BSFL ...)
// Lowering extension
(SignExt8to16 ...) -> (MOVBLSX ...)
(SignExt8to32 ...) -> (MOVBLSX ...)
(SignExt16to32 ...) -> (MOVWLSX ...)
(SignExt8to16 ...) => (MOVBLSX ...)
(SignExt8to32 ...) => (MOVBLSX ...)
(SignExt16to32 ...) => (MOVWLSX ...)
(ZeroExt8to16 ...) -> (MOVBLZX ...)
(ZeroExt8to32 ...) -> (MOVBLZX ...)
(ZeroExt16to32 ...) -> (MOVWLZX ...)
(ZeroExt8to16 ...) => (MOVBLZX ...)
(ZeroExt8to32 ...) => (MOVBLZX ...)
(ZeroExt16to32 ...) => (MOVWLZX ...)
(Signmask x) -> (SARLconst x [31])
(Zeromask <t> x) -> (XORLconst [-1] (SBBLcarrymask <t> (CMPLconst x [1])))
(Slicemask <t> x) -> (SARLconst (NEGL <t> x) [31])
(Signmask x) => (SARLconst x [31])
(Zeromask <t> x) => (XORLconst [-1] (SBBLcarrymask <t> (CMPLconst x [1])))
(Slicemask <t> x) => (SARLconst (NEGL <t> x) [31])
// Lowering truncation
// Because we ignore high parts of registers, truncates are just copies.
(Trunc16to8 ...) -> (Copy ...)
(Trunc32to8 ...) -> (Copy ...)
(Trunc32to16 ...) -> (Copy ...)
(Trunc16to8 ...) => (Copy ...)
(Trunc32to8 ...) => (Copy ...)
(Trunc32to16 ...) => (Copy ...)
// Lowering float <-> int
(Cvt32to32F ...) -> (CVTSL2SS ...)
(Cvt32to64F ...) -> (CVTSL2SD ...)
(Cvt32to32F ...) => (CVTSL2SS ...)
(Cvt32to64F ...) => (CVTSL2SD ...)
(Cvt32Fto32 ...) -> (CVTTSS2SL ...)
(Cvt64Fto32 ...) -> (CVTTSD2SL ...)
(Cvt32Fto32 ...) => (CVTTSS2SL ...)
(Cvt64Fto32 ...) => (CVTTSD2SL ...)
(Cvt32Fto64F ...) -> (CVTSS2SD ...)
(Cvt64Fto32F ...) -> (CVTSD2SS ...)
(Cvt32Fto64F ...) => (CVTSS2SD ...)
(Cvt64Fto32F ...) => (CVTSD2SS ...)
(Round32F ...) -> (Copy ...)
(Round64F ...) -> (Copy ...)
(Round32F ...) => (Copy ...)
(Round64F ...) => (Copy ...)
(CvtBoolToUint8 ...) -> (Copy ...)
(CvtBoolToUint8 ...) => (Copy ...)
// Lowering shifts
// Unsigned shifts need to return 0 if shift amount is >= width of shifted value.
// result = (arg << shift) & (shift >= argbits ? 0 : 0xffffffffffffffff)
(Lsh32x(32|16|8) <t> x y) && !shiftIsBounded(v) -> (ANDL (SHLL <t> x y) (SBBLcarrymask <t> (CMP(L|W|B)const y [32])))
(Lsh16x(32|16|8) <t> x y) && !shiftIsBounded(v) -> (ANDL (SHLL <t> x y) (SBBLcarrymask <t> (CMP(L|W|B)const y [32])))
(Lsh8x(32|16|8) <t> x y) && !shiftIsBounded(v) -> (ANDL (SHLL <t> x y) (SBBLcarrymask <t> (CMP(L|W|B)const y [32])))
(Lsh32x(32|16|8) <t> x y) && !shiftIsBounded(v) => (ANDL (SHLL <t> x y) (SBBLcarrymask <t> (CMP(L|W|B)const y [32])))
(Lsh16x(32|16|8) <t> x y) && !shiftIsBounded(v) => (ANDL (SHLL <t> x y) (SBBLcarrymask <t> (CMP(L|W|B)const y [32])))
(Lsh8x(32|16|8) <t> x y) && !shiftIsBounded(v) => (ANDL (SHLL <t> x y) (SBBLcarrymask <t> (CMP(L|W|B)const y [32])))
(Lsh32x(32|16|8) <t> x y) && shiftIsBounded(v) -> (SHLL <t> x y)
(Lsh16x(32|16|8) <t> x y) && shiftIsBounded(v) -> (SHLL <t> x y)
(Lsh8x(32|16|8) <t> x y) && shiftIsBounded(v) -> (SHLL <t> x y)
(Lsh32x(32|16|8) <t> x y) && shiftIsBounded(v) => (SHLL <t> x y)
(Lsh16x(32|16|8) <t> x y) && shiftIsBounded(v) => (SHLL <t> x y)
(Lsh8x(32|16|8) <t> x y) && shiftIsBounded(v) => (SHLL <t> x y)
(Rsh32Ux(32|16|8) <t> x y) && !shiftIsBounded(v) -> (ANDL (SHRL <t> x y) (SBBLcarrymask <t> (CMP(L|W|B)const y [32])))
(Rsh16Ux(32|16|8) <t> x y) && !shiftIsBounded(v) -> (ANDL (SHRW <t> x y) (SBBLcarrymask <t> (CMP(L|W|B)const y [16])))
(Rsh8Ux(32|16|8) <t> x y) && !shiftIsBounded(v) -> (ANDL (SHRB <t> x y) (SBBLcarrymask <t> (CMP(L|W|B)const y [8])))
(Rsh32Ux(32|16|8) <t> x y) && !shiftIsBounded(v) => (ANDL (SHRL <t> x y) (SBBLcarrymask <t> (CMP(L|W|B)const y [32])))
(Rsh16Ux(32|16|8) <t> x y) && !shiftIsBounded(v) => (ANDL (SHRW <t> x y) (SBBLcarrymask <t> (CMP(L|W|B)const y [16])))
(Rsh8Ux(32|16|8) <t> x y) && !shiftIsBounded(v) => (ANDL (SHRB <t> x y) (SBBLcarrymask <t> (CMP(L|W|B)const y [8])))
(Rsh32Ux(32|16|8) <t> x y) && shiftIsBounded(v) -> (SHRL <t> x y)
(Rsh16Ux(32|16|8) <t> x y) && shiftIsBounded(v) -> (SHRW <t> x y)
(Rsh8Ux(32|16|8) <t> x y) && shiftIsBounded(v) -> (SHRB <t> x y)
(Rsh32Ux(32|16|8) <t> x y) && shiftIsBounded(v) => (SHRL <t> x y)
(Rsh16Ux(32|16|8) <t> x y) && shiftIsBounded(v) => (SHRW <t> x y)
(Rsh8Ux(32|16|8) <t> x y) && shiftIsBounded(v) => (SHRB <t> x y)
// Signed right shift needs to return 0/-1 if shift amount is >= width of shifted value.
// We implement this by setting the shift value to -1 (all ones) if the shift value is >= width.
(Rsh32x(32|16|8) <t> x y) && !shiftIsBounded(v) -> (SARL <t> x (ORL <y.Type> y (NOTL <y.Type> (SBBLcarrymask <y.Type> (CMP(L|W|B)const y [32])))))
(Rsh16x(32|16|8) <t> x y) && !shiftIsBounded(v) -> (SARW <t> x (ORL <y.Type> y (NOTL <y.Type> (SBBLcarrymask <y.Type> (CMP(L|W|B)const y [16])))))
(Rsh8x(32|16|8) <t> x y) && !shiftIsBounded(v) -> (SARB <t> x (ORL <y.Type> y (NOTL <y.Type> (SBBLcarrymask <y.Type> (CMP(L|W|B)const y [8])))))
(Rsh32x(32|16|8) <t> x y) && !shiftIsBounded(v) => (SARL <t> x (ORL <y.Type> y (NOTL <y.Type> (SBBLcarrymask <y.Type> (CMP(L|W|B)const y [32])))))
(Rsh16x(32|16|8) <t> x y) && !shiftIsBounded(v) => (SARW <t> x (ORL <y.Type> y (NOTL <y.Type> (SBBLcarrymask <y.Type> (CMP(L|W|B)const y [16])))))
(Rsh8x(32|16|8) <t> x y) && !shiftIsBounded(v) => (SARB <t> x (ORL <y.Type> y (NOTL <y.Type> (SBBLcarrymask <y.Type> (CMP(L|W|B)const y [8])))))
(Rsh32x(32|16|8) <t> x y) && shiftIsBounded(v) -> (SARL x y)
(Rsh16x(32|16|8) <t> x y) && shiftIsBounded(v) -> (SARW x y)
(Rsh8x(32|16|8) <t> x y) && shiftIsBounded(v) -> (SARB x y)
(Rsh32x(32|16|8) <t> x y) && shiftIsBounded(v) => (SARL x y)
(Rsh16x(32|16|8) <t> x y) && shiftIsBounded(v) => (SARW x y)
(Rsh8x(32|16|8) <t> x y) && shiftIsBounded(v) => (SARB x y)
// constant shifts
// generic opt rewrites all constant shifts to shift by Const64
(Lsh32x64 x (Const64 [c])) && uint64(c) < 32 -> (SHLLconst x [c])
(Rsh32x64 x (Const64 [c])) && uint64(c) < 32 -> (SARLconst x [c])
(Rsh32Ux64 x (Const64 [c])) && uint64(c) < 32 -> (SHRLconst x [c])
(Lsh16x64 x (Const64 [c])) && uint64(c) < 16 -> (SHLLconst x [c])
(Rsh16x64 x (Const64 [c])) && uint64(c) < 16 -> (SARWconst x [c])
(Rsh16Ux64 x (Const64 [c])) && uint64(c) < 16 -> (SHRWconst x [c])
(Lsh8x64 x (Const64 [c])) && uint64(c) < 8 -> (SHLLconst x [c])
(Rsh8x64 x (Const64 [c])) && uint64(c) < 8 -> (SARBconst x [c])
(Rsh8Ux64 x (Const64 [c])) && uint64(c) < 8 -> (SHRBconst x [c])
(Lsh32x64 x (Const64 [c])) && uint64(c) < 32 => (SHLLconst x [int32(c)])
(Rsh32x64 x (Const64 [c])) && uint64(c) < 32 => (SARLconst x [int32(c)])
(Rsh32Ux64 x (Const64 [c])) && uint64(c) < 32 => (SHRLconst x [int32(c)])
(Lsh16x64 x (Const64 [c])) && uint64(c) < 16 => (SHLLconst x [int32(c)])
(Rsh16x64 x (Const64 [c])) && uint64(c) < 16 => (SARWconst x [int16(c)])
(Rsh16Ux64 x (Const64 [c])) && uint64(c) < 16 => (SHRWconst x [int16(c)])
(Lsh8x64 x (Const64 [c])) && uint64(c) < 8 => (SHLLconst x [int32(c)])
(Rsh8x64 x (Const64 [c])) && uint64(c) < 8 => (SARBconst x [int8(c)])
(Rsh8Ux64 x (Const64 [c])) && uint64(c) < 8 => (SHRBconst x [int8(c)])
// large constant shifts
(Lsh32x64 _ (Const64 [c])) && uint64(c) >= 32 -> (Const32 [0])
(Rsh32Ux64 _ (Const64 [c])) && uint64(c) >= 32 -> (Const32 [0])
(Lsh16x64 _ (Const64 [c])) && uint64(c) >= 16 -> (Const16 [0])
(Rsh16Ux64 _ (Const64 [c])) && uint64(c) >= 16 -> (Const16 [0])
(Lsh8x64 _ (Const64 [c])) && uint64(c) >= 8 -> (Const8 [0])
(Rsh8Ux64 _ (Const64 [c])) && uint64(c) >= 8 -> (Const8 [0])
(Lsh32x64 _ (Const64 [c])) && uint64(c) >= 32 => (Const32 [0])
(Rsh32Ux64 _ (Const64 [c])) && uint64(c) >= 32 => (Const32 [0])
(Lsh16x64 _ (Const64 [c])) && uint64(c) >= 16 => (Const16 [0])
(Rsh16Ux64 _ (Const64 [c])) && uint64(c) >= 16 => (Const16 [0])
(Lsh8x64 _ (Const64 [c])) && uint64(c) >= 8 => (Const8 [0])
(Rsh8Ux64 _ (Const64 [c])) && uint64(c) >= 8 => (Const8 [0])
// large constant signed right shift, we leave the sign bit
(Rsh32x64 x (Const64 [c])) && uint64(c) >= 32 -> (SARLconst x [31])
(Rsh16x64 x (Const64 [c])) && uint64(c) >= 16 -> (SARWconst x [15])
(Rsh8x64 x (Const64 [c])) && uint64(c) >= 8 -> (SARBconst x [7])
(Rsh32x64 x (Const64 [c])) && uint64(c) >= 32 => (SARLconst x [31])
(Rsh16x64 x (Const64 [c])) && uint64(c) >= 16 => (SARWconst x [15])
(Rsh8x64 x (Const64 [c])) && uint64(c) >= 8 => (SARBconst x [7])
// constant rotates
(RotateLeft32 x (MOVLconst [c])) -> (ROLLconst [c&31] x)
(RotateLeft16 x (MOVLconst [c])) -> (ROLWconst [c&15] x)
(RotateLeft8 x (MOVLconst [c])) -> (ROLBconst [c&7] x)
(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)
// Lowering comparisons
(Less32 x y) -> (SETL (CMPL x y))
(Less16 x y) -> (SETL (CMPW x y))
(Less8 x y) -> (SETL (CMPB x y))
(Less32U x y) -> (SETB (CMPL x y))
(Less16U x y) -> (SETB (CMPW x y))
(Less8U x y) -> (SETB (CMPB x y))
(Less32 x y) => (SETL (CMPL x y))
(Less16 x y) => (SETL (CMPW x y))
(Less8 x y) => (SETL (CMPB x y))
(Less32U x y) => (SETB (CMPL x y))
(Less16U x y) => (SETB (CMPW x y))
(Less8U x y) => (SETB (CMPB x y))
// Use SETGF with reversed operands to dodge NaN case
(Less64F x y) -> (SETGF (UCOMISD y x))
(Less32F x y) -> (SETGF (UCOMISS y x))
(Less64F x y) => (SETGF (UCOMISD y x))
(Less32F x y) => (SETGF (UCOMISS y x))
(Leq32 x y) -> (SETLE (CMPL x y))
(Leq16 x y) -> (SETLE (CMPW x y))
(Leq8 x y) -> (SETLE (CMPB x y))
(Leq32U x y) -> (SETBE (CMPL x y))
(Leq16U x y) -> (SETBE (CMPW x y))
(Leq8U x y) -> (SETBE (CMPB x y))
(Leq32 x y) => (SETLE (CMPL x y))
(Leq16 x y) => (SETLE (CMPW x y))
(Leq8 x y) => (SETLE (CMPB x y))
(Leq32U x y) => (SETBE (CMPL x y))
(Leq16U x y) => (SETBE (CMPW x y))
(Leq8U x y) => (SETBE (CMPB x y))
// Use SETGEF with reversed operands to dodge NaN case
(Leq64F x y) -> (SETGEF (UCOMISD y x))
(Leq32F x y) -> (SETGEF (UCOMISS y x))
(Leq64F x y) => (SETGEF (UCOMISD y x))
(Leq32F x y) => (SETGEF (UCOMISS y x))
(Eq32 x y) -> (SETEQ (CMPL x y))
(Eq16 x y) -> (SETEQ (CMPW x y))
(Eq8 x y) -> (SETEQ (CMPB x y))
(EqB x y) -> (SETEQ (CMPB x y))
(EqPtr x y) -> (SETEQ (CMPL x y))
(Eq64F x y) -> (SETEQF (UCOMISD x y))
(Eq32F x y) -> (SETEQF (UCOMISS x y))
(Eq32 x y) => (SETEQ (CMPL x y))
(Eq16 x y) => (SETEQ (CMPW x y))
(Eq8 x y) => (SETEQ (CMPB x y))
(EqB x y) => (SETEQ (CMPB x y))
(EqPtr x y) => (SETEQ (CMPL x y))
(Eq64F x y) => (SETEQF (UCOMISD x y))
(Eq32F x y) => (SETEQF (UCOMISS x y))
(Neq32 x y) -> (SETNE (CMPL x y))
(Neq16 x y) -> (SETNE (CMPW x y))
(Neq8 x y) -> (SETNE (CMPB x y))
(NeqB x y) -> (SETNE (CMPB x y))
(NeqPtr x y) -> (SETNE (CMPL x y))
(Neq64F x y) -> (SETNEF (UCOMISD x y))
(Neq32F x y) -> (SETNEF (UCOMISS x y))
(Neq32 x y) => (SETNE (CMPL x y))
(Neq16 x y) => (SETNE (CMPW x y))
(Neq8 x y) => (SETNE (CMPB x y))
(NeqB x y) => (SETNE (CMPB x y))
(NeqPtr x y) => (SETNE (CMPL x y))
(Neq64F x y) => (SETNEF (UCOMISD x y))
(Neq32F x y) => (SETNEF (UCOMISS x y))
// Lowering loads
(Load <t> ptr mem) && (is32BitInt(t) || isPtr(t)) -> (MOVLload ptr mem)
(Load <t> ptr mem) && is16BitInt(t) -> (MOVWload ptr mem)
(Load <t> ptr mem) && (t.IsBoolean() || is8BitInt(t)) -> (MOVBload ptr mem)
(Load <t> ptr mem) && is32BitFloat(t) -> (MOVSSload ptr mem)
(Load <t> ptr mem) && is64BitFloat(t) -> (MOVSDload ptr mem)
(Load <t> ptr mem) && (is32BitInt(t) || isPtr(t)) => (MOVLload ptr mem)
(Load <t> ptr mem) && is16BitInt(t) => (MOVWload ptr mem)
(Load <t> ptr mem) && (t.IsBoolean() || is8BitInt(t)) => (MOVBload ptr mem)
(Load <t> ptr mem) && is32BitFloat(t) => (MOVSSload ptr mem)
(Load <t> ptr mem) && is64BitFloat(t) => (MOVSDload ptr mem)
// Lowering stores
// These more-specific FP versions of Store pattern should come first.
(Store {t} ptr val mem) && t.(*types.Type).Size() == 8 && is64BitFloat(val.Type) -> (MOVSDstore ptr val mem)
(Store {t} ptr val mem) && t.(*types.Type).Size() == 4 && is32BitFloat(val.Type) -> (MOVSSstore ptr val mem)
(Store {t} ptr val mem) && t.Size() == 8 && is64BitFloat(val.Type) => (MOVSDstore ptr val mem)
(Store {t} ptr val mem) && t.Size() == 4 && is32BitFloat(val.Type) => (MOVSSstore ptr val mem)
(Store {t} ptr val mem) && t.(*types.Type).Size() == 4 -> (MOVLstore ptr val mem)
(Store {t} ptr val mem) && t.(*types.Type).Size() == 2 -> (MOVWstore ptr val mem)
(Store {t} ptr val mem) && t.(*types.Type).Size() == 1 -> (MOVBstore ptr val mem)
(Store {t} ptr val mem) && t.Size() == 4 => (MOVLstore ptr val mem)
(Store {t} ptr val mem) && t.Size() == 2 => (MOVWstore ptr val mem)
(Store {t} ptr val mem) && t.Size() == 1 => (MOVBstore ptr val mem)
// Lowering moves
(Move [0] _ _ mem) -> mem
(Move [1] dst src mem) -> (MOVBstore dst (MOVBload src mem) mem)
(Move [2] dst src mem) -> (MOVWstore dst (MOVWload src mem) mem)
(Move [4] dst src mem) -> (MOVLstore dst (MOVLload src mem) mem)
(Move [3] dst src mem) ->
(Move [0] _ _ mem) => mem
(Move [1] dst src mem) => (MOVBstore dst (MOVBload src mem) mem)
(Move [2] dst src mem) => (MOVWstore dst (MOVWload src mem) mem)
(Move [4] dst src mem) => (MOVLstore dst (MOVLload src mem) mem)
(Move [3] dst src mem) =>
(MOVBstore [2] dst (MOVBload [2] src mem)
(MOVWstore dst (MOVWload src mem) mem))
(Move [5] dst src mem) ->
(Move [5] dst src mem) =>
(MOVBstore [4] dst (MOVBload [4] src mem)
(MOVLstore dst (MOVLload src mem) mem))
(Move [6] dst src mem) ->
(Move [6] dst src mem) =>
(MOVWstore [4] dst (MOVWload [4] src mem)
(MOVLstore dst (MOVLload src mem) mem))
(Move [7] dst src mem) ->
(Move [7] dst src mem) =>
(MOVLstore [3] dst (MOVLload [3] src mem)
(MOVLstore dst (MOVLload src mem) mem))
(Move [8] dst src mem) ->
(Move [8] dst src mem) =>
(MOVLstore [4] dst (MOVLload [4] src mem)
(MOVLstore dst (MOVLload src mem) mem))
// Adjust moves to be a multiple of 4 bytes.
(Move [s] dst src mem)
&& s > 8 && s%4 != 0 ->
&& s > 8 && s%4 != 0 =>
(Move [s-s%4]
(ADDLconst <dst.Type> dst [s%4])
(ADDLconst <src.Type> src [s%4])
(ADDLconst <dst.Type> dst [int32(s%4)])
(ADDLconst <src.Type> src [int32(s%4)])
(MOVLstore dst (MOVLload src mem) mem))
// Medium copying uses a duff device.
(Move [s] dst src mem)
&& s > 8 && s <= 4*128 && s%4 == 0
&& !config.noDuffDevice && logLargeCopy(v, s) ->
&& !config.noDuffDevice && logLargeCopy(v, s) =>
(DUFFCOPY [10*(128-s/4)] dst src mem)
// 10 and 128 are magic constants. 10 is the number of bytes to encode:
// MOVL (SI), CX
@ -249,27 +249,27 @@
// and 128 is the number of such blocks. See src/runtime/duff_386.s:duffcopy.
// Large copying uses REP MOVSL.
(Move [s] dst src mem) && (s > 4*128 || config.noDuffDevice) && s%4 == 0 && logLargeCopy(v, s) ->
(REPMOVSL dst src (MOVLconst [s/4]) mem)
(Move [s] dst src mem) && (s > 4*128 || config.noDuffDevice) && s%4 == 0 && logLargeCopy(v, s) =>
(REPMOVSL dst src (MOVLconst [int32(s/4)]) mem)
// Lowering Zero instructions
(Zero [0] _ mem) -> mem
(Zero [1] destptr mem) -> (MOVBstoreconst [0] destptr mem)
(Zero [2] destptr mem) -> (MOVWstoreconst [0] destptr mem)
(Zero [4] destptr mem) -> (MOVLstoreconst [0] destptr mem)
(Zero [0] _ mem) => mem
(Zero [1] destptr mem) => (MOVBstoreconst [0] destptr mem)
(Zero [2] destptr mem) => (MOVWstoreconst [0] destptr mem)
(Zero [4] destptr mem) => (MOVLstoreconst [0] destptr mem)
(Zero [3] destptr mem) ->
(MOVBstoreconst [makeValAndOff(0,2)] destptr
(MOVWstoreconst [0] destptr mem))
(Zero [5] destptr mem) ->
(MOVBstoreconst [makeValAndOff(0,4)] destptr
(MOVLstoreconst [0] destptr mem))
(Zero [6] destptr mem) ->
(MOVWstoreconst [makeValAndOff(0,4)] destptr
(MOVLstoreconst [0] destptr mem))
(Zero [7] destptr mem) ->
(MOVLstoreconst [makeValAndOff(0,3)] destptr
(MOVLstoreconst [0] destptr mem))
(Zero [3] destptr mem) =>
(MOVBstoreconst [makeValAndOff32(0,2)] destptr
(MOVWstoreconst [makeValAndOff32(0,0)] destptr mem))
(Zero [5] destptr mem) =>
(MOVBstoreconst [makeValAndOff32(0,4)] destptr
(MOVLstoreconst [makeValAndOff32(0,0)] destptr mem))
(Zero [6] destptr mem) =>
(MOVWstoreconst [makeValAndOff32(0,4)] destptr
(MOVLstoreconst [makeValAndOff32(0,0)] destptr mem))
(Zero [7] destptr mem) =>
(MOVLstoreconst [makeValAndOff32(0,3)] destptr
(MOVLstoreconst [makeValAndOff32(0,0)] destptr mem))
// Strip off any fractional word zeroing.
(Zero [s] destptr mem) && s%4 != 0 && s > 4 ->
@ -277,23 +277,23 @@
(MOVLstoreconst [0] destptr mem))
// Zero small numbers of words directly.
(Zero [8] destptr mem) ->
(MOVLstoreconst [makeValAndOff(0,4)] destptr
(MOVLstoreconst [0] destptr mem))
(Zero [12] destptr mem) ->
(MOVLstoreconst [makeValAndOff(0,8)] destptr
(MOVLstoreconst [makeValAndOff(0,4)] destptr
(MOVLstoreconst [0] destptr mem)))
(Zero [16] destptr mem) ->
(MOVLstoreconst [makeValAndOff(0,12)] destptr
(MOVLstoreconst [makeValAndOff(0,8)] destptr
(MOVLstoreconst [makeValAndOff(0,4)] destptr
(MOVLstoreconst [0] destptr mem))))
(Zero [8] destptr mem) =>
(MOVLstoreconst [makeValAndOff32(0,4)] destptr
(MOVLstoreconst [makeValAndOff32(0,0)] destptr mem))
(Zero [12] destptr mem) =>
(MOVLstoreconst [makeValAndOff32(0,8)] destptr
(MOVLstoreconst [makeValAndOff32(0,4)] destptr
(MOVLstoreconst [makeValAndOff32(0,0)] destptr mem)))
(Zero [16] destptr mem) =>
(MOVLstoreconst [makeValAndOff32(0,12)] destptr
(MOVLstoreconst [makeValAndOff32(0,8)] destptr
(MOVLstoreconst [makeValAndOff32(0,4)] destptr
(MOVLstoreconst [makeValAndOff32(0,0)] destptr mem))))
// Medium zeroing uses a duff device.
(Zero [s] destptr mem)
&& s > 16 && s <= 4*128 && s%4 == 0
&& !config.noDuffDevice ->
&& !config.noDuffDevice =>
(DUFFZERO [1*(128-s/4)] destptr (MOVLconst [0]) mem)
// 1 and 128 are magic constants. 1 is the number of bytes to encode STOSL.
// 128 is the number of STOSL instructions in duffzero.
@ -302,63 +302,66 @@
// Large zeroing uses REP STOSQ.
(Zero [s] destptr mem)
&& (s > 4*128 || (config.noDuffDevice && s > 16))
&& s%4 == 0 ->
(REPSTOSL destptr (MOVLconst [s/4]) (MOVLconst [0]) mem)
&& s%4 == 0 =>
(REPSTOSL destptr (MOVLconst [int32(s/4)]) (MOVLconst [0]) mem)
// Lowering constants
(Const(8|16|32) ...) -> (MOVLconst ...)
(Const(32|64)F ...) -> (MOVS(S|D)const ...)
(ConstNil) -> (MOVLconst [0])
(ConstBool ...) -> (MOVLconst ...)
(Const8 [c]) => (MOVLconst [int32(c)])
(Const16 [c]) => (MOVLconst [int32(c)])
(Const32 ...) => (MOVLconst ...)
(Const(32|64)F ...) => (MOVS(S|D)const ...)
(ConstNil) => (MOVLconst [0])
(ConstBool [c]) => (MOVLconst [int32(b2i(c))])
// Lowering calls
(StaticCall ...) -> (CALLstatic ...)
(ClosureCall ...) -> (CALLclosure ...)
(InterCall ...) -> (CALLinter ...)
(StaticCall ...) => (CALLstatic ...)
(ClosureCall ...) => (CALLclosure ...)
(InterCall ...) => (CALLinter ...)
// Miscellaneous
(IsNonNil p) -> (SETNE (TESTL p p))
(IsInBounds idx len) -> (SETB (CMPL idx len))
(IsSliceInBounds idx len) -> (SETBE (CMPL idx len))
(NilCheck ...) -> (LoweredNilCheck ...)
(GetG ...) -> (LoweredGetG ...)
(GetClosurePtr ...) -> (LoweredGetClosurePtr ...)
(GetCallerPC ...) -> (LoweredGetCallerPC ...)
(GetCallerSP ...) -> (LoweredGetCallerSP ...)
(Addr ...) -> (LEAL ...)
(LocalAddr {sym} base _) -> (LEAL {sym} base)
(IsNonNil p) => (SETNE (TESTL p p))
(IsInBounds idx len) => (SETB (CMPL idx len))
(IsSliceInBounds idx len) => (SETBE (CMPL idx len))
(NilCheck ...) => (LoweredNilCheck ...)
(GetG ...) => (LoweredGetG ...)
(GetClosurePtr ...) => (LoweredGetClosurePtr ...)
(GetCallerPC ...) => (LoweredGetCallerPC ...)
(GetCallerSP ...) => (LoweredGetCallerSP ...)
(Addr {sym} base) => (LEAL {sym} base)
(LocalAddr {sym} base _) => (LEAL {sym} base)
// block rewrites
(If (SETL cmp) yes no) -> (LT cmp yes no)
(If (SETLE cmp) yes no) -> (LE cmp yes no)
(If (SETG cmp) yes no) -> (GT cmp yes no)
(If (SETGE cmp) yes no) -> (GE cmp yes no)
(If (SETEQ cmp) yes no) -> (EQ cmp yes no)
(If (SETNE cmp) yes no) -> (NE cmp yes no)
(If (SETB cmp) yes no) -> (ULT cmp yes no)
(If (SETBE cmp) yes no) -> (ULE cmp yes no)
(If (SETA cmp) yes no) -> (UGT cmp yes no)
(If (SETAE cmp) yes no) -> (UGE cmp yes no)
(If (SETO cmp) yes no) -> (OS cmp yes no)
(If (SETL cmp) yes no) => (LT cmp yes no)
(If (SETLE cmp) yes no) => (LE cmp yes no)
(If (SETG cmp) yes no) => (GT cmp yes no)
(If (SETGE cmp) yes no) => (GE cmp yes no)
(If (SETEQ cmp) yes no) => (EQ cmp yes no)
(If (SETNE cmp) yes no) => (NE cmp yes no)
(If (SETB cmp) yes no) => (ULT cmp yes no)
(If (SETBE cmp) yes no) => (ULE cmp yes no)
(If (SETA cmp) yes no) => (UGT cmp yes no)
(If (SETAE cmp) yes no) => (UGE cmp yes no)
(If (SETO cmp) yes no) => (OS cmp yes no)
// Special case for floating point - LF/LEF not generated
(If (SETGF cmp) yes no) -> (UGT cmp yes no)
(If (SETGEF cmp) yes no) -> (UGE cmp yes no)
(If (SETEQF cmp) yes no) -> (EQF cmp yes no)
(If (SETNEF cmp) yes no) -> (NEF cmp yes no)
(If (SETGF cmp) yes no) => (UGT cmp yes no)
(If (SETGEF cmp) yes no) => (UGE cmp yes no)
(If (SETEQF cmp) yes no) => (EQF cmp yes no)
(If (SETNEF cmp) yes no) => (NEF cmp yes no)
(If cond yes no) -> (NE (TESTB cond cond) yes no)
(If cond yes no) => (NE (TESTB cond cond) yes no)
// Write barrier.
(WB ...) -> (LoweredWB ...)
(WB ...) => (LoweredWB ...)
(PanicBounds [kind] x y mem) && boundsABI(kind) == 0 -> (LoweredPanicBoundsA [kind] x y mem)
(PanicBounds [kind] x y mem) && boundsABI(kind) == 1 -> (LoweredPanicBoundsB [kind] x y mem)
(PanicBounds [kind] x y mem) && boundsABI(kind) == 2 -> (LoweredPanicBoundsC [kind] x y mem)
(PanicBounds [kind] x y mem) && boundsABI(kind) == 0 => (LoweredPanicBoundsA [kind] x y mem)
(PanicBounds [kind] x y mem) && boundsABI(kind) == 1 => (LoweredPanicBoundsB [kind] x y mem)
(PanicBounds [kind] x y mem) && boundsABI(kind) == 2 => (LoweredPanicBoundsC [kind] x y mem)
(PanicExtend [kind] hi lo y mem) && boundsABI(kind) == 0 -> (LoweredPanicExtendA [kind] hi lo y mem)
(PanicExtend [kind] hi lo y mem) && boundsABI(kind) == 1 -> (LoweredPanicExtendB [kind] hi lo y mem)
(PanicExtend [kind] hi lo y mem) && boundsABI(kind) == 2 -> (LoweredPanicExtendC [kind] hi lo y mem)
(PanicExtend [kind] hi lo y mem) && boundsABI(kind) == 0 => (LoweredPanicExtendA [kind] hi lo y mem)
(PanicExtend [kind] hi lo y mem) && boundsABI(kind) == 1 => (LoweredPanicExtendB [kind] hi lo y mem)
(PanicExtend [kind] hi lo y mem) && boundsABI(kind) == 2 => (LoweredPanicExtendC [kind] hi lo y mem)
// ***************************
// Above: lowering rules

File diff suppressed because it is too large Load Diff