cmd/compile: define high bits of AuxInt

Previously if we were only using the low bits of AuxInt,
the high bits were ignored and could be junk.  This CL
changes that behavior to define the high bits to be the
sign-extended version of the low bits for all cases.

There are 2 main benefits:
- Deterministic representation.  This helps with CSE.
  (Const8 [0x1]) and (Const8 [0x101]) used to be the same "value"
  but CSE couldn't see them as such.
- Testability.  We can check that all ops leave AuxInt in a state
  consistent with the new rule.  In the old scheme, it was hard
  to check whether a rule correctly used only the low-order bits.
Side benefits:
- ==0 and !=0 tests are easier.

Drawbacks:
- This differs from the runtime representation in registers,
  where it is important that we allow upper bits to be undefined
  (so we're not sign/zero-extending all the time).
- Ops that treat AuxInt as unsigned (shifts, mostly) need to be
  a bit more careful.

Change-Id: I9a685ff27e36dc03287c9ab1cecd6c0b4045c819
Reviewed-on: https://go-review.googlesource.com/21256Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>

src/cmd/compile/internal/amd64/ssa.go

@@ -420,7 +420,7 @@ func ssaGenValue(s *gc.SSAGenState, v *ssa.Value) {
 		r := gc.SSARegNum(v)
 		a := gc.SSARegNum(v.Args[0])
 		if r == a {
-			if v.AuxInt2Int64() == 1 {
+			if v.AuxInt == 1 {
 				var asm obj.As
 				switch v.Op {
 				// Software optimization manual recommends add $1,reg.
@@ -439,7 +439,7 @@ func ssaGenValue(s *gc.SSAGenState, v *ssa.Value) {
 				p.To.Type = obj.TYPE_REG
 				p.To.Reg = r
 				return
-			} else if v.AuxInt2Int64() == -1 {
+			} else if v.AuxInt == -1 {
 				var asm obj.As
 				switch v.Op {
 				case ssa.OpAMD64ADDQconst:
@@ -456,7 +456,7 @@ func ssaGenValue(s *gc.SSAGenState, v *ssa.Value) {
 			} else {
 				p := gc.Prog(v.Op.Asm())
 				p.From.Type = obj.TYPE_CONST
-				p.From.Offset = v.AuxInt2Int64()
+				p.From.Offset = v.AuxInt
 				p.To.Type = obj.TYPE_REG
 				p.To.Reg = r
 				return
@@ -474,7 +474,7 @@ func ssaGenValue(s *gc.SSAGenState, v *ssa.Value) {
 		p := gc.Prog(asm)
 		p.From.Type = obj.TYPE_MEM
 		p.From.Reg = a
-		p.From.Offset = v.AuxInt2Int64()
+		p.From.Offset = v.AuxInt
 		p.To.Type = obj.TYPE_REG
 		p.To.Reg = r
 
@@ -494,7 +494,7 @@ func ssaGenValue(s *gc.SSAGenState, v *ssa.Value) {
 		// Constant into AX, after arg0 movement in case arg0 is in AX
 		p := gc.Prog(moveByType(v.Type))
 		p.From.Type = obj.TYPE_CONST
-		p.From.Offset = v.AuxInt2Int64()
+		p.From.Offset = v.AuxInt
 		p.To.Type = obj.TYPE_REG
 		p.To.Reg = x86.REG_AX
 
@@ -516,7 +516,7 @@ func ssaGenValue(s *gc.SSAGenState, v *ssa.Value) {
 		}
 		p := gc.Prog(v.Op.Asm())
 		p.From.Type = obj.TYPE_CONST
-		p.From.Offset = v.AuxInt2Int64()
+		p.From.Offset = v.AuxInt
 		p.To.Type = obj.TYPE_REG
 		p.To.Reg = r
 		// TODO: Teach doasm to compile the three-address multiply imul $c, r1, r2
@@ -531,7 +531,7 @@ func ssaGenValue(s *gc.SSAGenState, v *ssa.Value) {
 		// a = b + (- const), saves us 1 instruction. We can't fit
 		// - (-1 << 31) into  4 bytes offset in lea.
 		// We handle 2-address just fine below.
-		if v.AuxInt2Int64() == -1<<31 || x == r {
+		if v.AuxInt == -1<<31 || x == r {
 			if x != r {
 				// This code compensates for the fact that the register allocator
 				// doesn't understand 2-address instructions yet. TODO: fix that.
@@ -543,10 +543,10 @@ func ssaGenValue(s *gc.SSAGenState, v *ssa.Value) {
 			}
 			p := gc.Prog(v.Op.Asm())
 			p.From.Type = obj.TYPE_CONST
-			p.From.Offset = v.AuxInt2Int64()
+			p.From.Offset = v.AuxInt
 			p.To.Type = obj.TYPE_REG
 			p.To.Reg = r
-		} else if x == r && v.AuxInt2Int64() == -1 {
+		} else if x == r && v.AuxInt == -1 {
 			var asm obj.As
 			// x = x - (-1) is the same as x++
 			// See OpAMD64ADDQconst comments about inc vs add $1,reg
@@ -561,7 +561,7 @@ func ssaGenValue(s *gc.SSAGenState, v *ssa.Value) {
 			p := gc.Prog(asm)
 			p.To.Type = obj.TYPE_REG
 			p.To.Reg = r
-		} else if x == r && v.AuxInt2Int64() == 1 {
+		} else if x == r && v.AuxInt == 1 {
 			var asm obj.As
 			switch v.Op {
 			case ssa.OpAMD64SUBQconst:
@@ -587,7 +587,7 @@ func ssaGenValue(s *gc.SSAGenState, v *ssa.Value) {
 			p := gc.Prog(asm)
 			p.From.Type = obj.TYPE_MEM
 			p.From.Reg = x
-			p.From.Offset = -v.AuxInt2Int64()
+			p.From.Offset = -v.AuxInt
 			p.To.Type = obj.TYPE_REG
 			p.To.Reg = r
 		}
@@ -614,7 +614,7 @@ func ssaGenValue(s *gc.SSAGenState, v *ssa.Value) {
 		}
 		p := gc.Prog(v.Op.Asm())
 		p.From.Type = obj.TYPE_CONST
-		p.From.Offset = v.AuxInt2Int64()
+		p.From.Offset = v.AuxInt
 		p.To.Type = obj.TYPE_REG
 		p.To.Reg = r
 	case ssa.OpAMD64SBBQcarrymask, ssa.OpAMD64SBBLcarrymask:
@@ -661,18 +661,18 @@ func ssaGenValue(s *gc.SSAGenState, v *ssa.Value) {
 		p.From.Type = obj.TYPE_REG
 		p.From.Reg = gc.SSARegNum(v.Args[0])
 		p.To.Type = obj.TYPE_CONST
-		p.To.Offset = v.AuxInt2Int64()
+		p.To.Offset = v.AuxInt
 	case ssa.OpAMD64TESTQconst, ssa.OpAMD64TESTLconst, ssa.OpAMD64TESTWconst, ssa.OpAMD64TESTBconst:
 		p := gc.Prog(v.Op.Asm())
 		p.From.Type = obj.TYPE_CONST
-		p.From.Offset = v.AuxInt2Int64()
+		p.From.Offset = v.AuxInt
 		p.To.Type = obj.TYPE_REG
 		p.To.Reg = gc.SSARegNum(v.Args[0])
 	case ssa.OpAMD64MOVBconst, ssa.OpAMD64MOVWconst, ssa.OpAMD64MOVLconst, ssa.OpAMD64MOVQconst:
 		x := gc.SSARegNum(v)
 		p := gc.Prog(v.Op.Asm())
 		p.From.Type = obj.TYPE_CONST
-		p.From.Offset = v.AuxInt2Int64()
+		p.From.Offset = v.AuxInt
 		p.To.Type = obj.TYPE_REG
 		p.To.Reg = x
 		// If flags are live at this instruction, suppress the
@@ -804,17 +804,7 @@ func ssaGenValue(s *gc.SSAGenState, v *ssa.Value) {
 		p := gc.Prog(v.Op.Asm())
 		p.From.Type = obj.TYPE_CONST
 		sc := v.AuxValAndOff()
-		i := sc.Val()
-		switch v.Op {
-		case ssa.OpAMD64MOVBstoreconst:
-			i = int64(int8(i))
-		case ssa.OpAMD64MOVWstoreconst:
-			i = int64(int16(i))
-		case ssa.OpAMD64MOVLstoreconst:
-			i = int64(int32(i))
-		case ssa.OpAMD64MOVQstoreconst:
-		}
-		p.From.Offset = i
+		p.From.Offset = sc.Val()
 		p.To.Type = obj.TYPE_MEM
 		p.To.Reg = gc.SSARegNum(v.Args[0])
 		gc.AddAux2(&p.To, v, sc.Off())
@@ -822,18 +812,15 @@ func ssaGenValue(s *gc.SSAGenState, v *ssa.Value) {
 		p := gc.Prog(v.Op.Asm())
 		p.From.Type = obj.TYPE_CONST
 		sc := v.AuxValAndOff()
+		p.From.Offset = sc.Val()
 		switch v.Op {
 		case ssa.OpAMD64MOVBstoreconstidx1:
-			p.From.Offset = int64(int8(sc.Val()))
 			p.To.Scale = 1
 		case ssa.OpAMD64MOVWstoreconstidx2:
-			p.From.Offset = int64(int16(sc.Val()))
 			p.To.Scale = 2
 		case ssa.OpAMD64MOVLstoreconstidx4:
-			p.From.Offset = int64(int32(sc.Val()))
 			p.To.Scale = 4
 		case ssa.OpAMD64MOVQstoreconstidx8:
-			p.From.Offset = sc.Val()
 			p.To.Scale = 8
 		}
 		p.To.Type = obj.TYPE_MEM

src/cmd/compile/internal/arm/ssa.go

@@ -85,7 +85,7 @@ func ssaGenValue(s *gc.SSAGenState, v *ssa.Value) {
 	case ssa.OpARMMOVWconst:
 		p := gc.Prog(v.Op.Asm())
 		p.From.Type = obj.TYPE_CONST
-		p.From.Offset = v.AuxInt2Int64()
+		p.From.Offset = v.AuxInt
 		p.To.Type = obj.TYPE_REG
 		p.To.Reg = gc.SSARegNum(v)
 	case ssa.OpARMCMP:

src/cmd/compile/internal/ssa/check.go

@@ -171,7 +171,27 @@ func checkFunc(f *Func) {
 			canHaveAuxInt := false
 			switch opcodeTable[v.Op].auxType {
 			case auxNone:
-			case auxBool, auxInt8, auxInt16, auxInt32, auxInt64, auxFloat64:
+			case auxBool:
+				if v.AuxInt < 0 || v.AuxInt > 1 {
+					f.Fatalf("bad bool AuxInt value for %v", v)
+				}
+				canHaveAuxInt = true
+			case auxInt8:
+				if v.AuxInt != int64(int8(v.AuxInt)) {
+					f.Fatalf("bad int8 AuxInt value for %v", v)
+				}
+				canHaveAuxInt = true
+			case auxInt16:
+				if v.AuxInt != int64(int16(v.AuxInt)) {
+					f.Fatalf("bad int16 AuxInt value for %v", v)
+				}
+				canHaveAuxInt = true
+			case auxInt32:
+				if v.AuxInt != int64(int32(v.AuxInt)) {
+					f.Fatalf("bad int32 AuxInt value for %v", v)
+				}
+				canHaveAuxInt = true
+			case auxInt64, auxFloat64:
 				canHaveAuxInt = true
 			case auxFloat32:
 				canHaveAuxInt = true

src/cmd/compile/internal/ssa/gen/AMD64.rules

@@ -2,13 +2,6 @@
 // Use of this source code is governed by a BSD-style
 // license that can be found in the LICENSE file.
 
-// x86 register conventions:
-//  - Integer types live in the low portion of registers. Upper portions are junk.
-//  - Boolean types use the low-order byte of a register. Upper bytes are junk.
-//  - We do not use AH,BH,CH,DH registers.
-//  - Floating-point types will live in the low natural slot of an sse2 register.
-//    Unused portions are junk.
-
 // Lowering arithmetic
 (Add64 x y) -> (ADDQ x y)
 (AddPtr x y) -> (ADDQ x y)
@@ -1162,33 +1155,33 @@
 // generic constant folding
 // TODO: more of this
 (ADDQconst [c] (MOVQconst [d])) -> (MOVQconst [c+d])
-(ADDLconst [c] (MOVLconst [d])) -> (MOVLconst [c+d])
-(ADDWconst [c] (MOVWconst [d])) -> (MOVWconst [c+d])
-(ADDBconst [c] (MOVBconst [d])) -> (MOVBconst [c+d])
+(ADDLconst [c] (MOVLconst [d])) -> (MOVLconst [int64(int32(c+d))])
+(ADDWconst [c] (MOVWconst [d])) -> (MOVWconst [int64(int16(c+d))])
+(ADDBconst [c] (MOVBconst [d])) -> (MOVBconst [int64(int8(c+d))])
 (ADDQconst [c] (ADDQconst [d] x)) && is32Bit(c+d) -> (ADDQconst [c+d] x)
-(ADDLconst [c] (ADDLconst [d] x)) -> (ADDLconst [c+d] x)
-(ADDWconst [c] (ADDWconst [d] x)) -> (ADDWconst [c+d] x)
-(ADDBconst [c] (ADDBconst [d] x)) -> (ADDBconst [c+d] x)
-(SUBQconst [c] (MOVQconst [d])) -> (MOVQconst [d-c])
-(SUBLconst [c] (MOVLconst [d])) -> (MOVLconst [d-c])
-(SUBWconst [c] (MOVWconst [d])) -> (MOVWconst [d-c])
-(SUBBconst [c] (MOVBconst [d])) -> (MOVBconst [d-c])
-(SUBQconst [c] (SUBQconst [d] x)) && is32Bit(-c-d) -> (ADDQconst [-c-d] x)
-(SUBLconst [c] (SUBLconst [d] x)) -> (ADDLconst [-c-d] x)
-(SUBWconst [c] (SUBWconst [d] x)) -> (ADDWconst [-c-d] x)
-(SUBBconst [c] (SUBBconst [d] x)) -> (ADDBconst [-c-d] x)
+(ADDLconst [c] (ADDLconst [d] x)) -> (ADDLconst [int64(int32(c+d))] x)
+(ADDWconst [c] (ADDWconst [d] x)) -> (ADDWconst [int64(int16(c+d))] x)
+(ADDBconst [c] (ADDBconst [d] x)) -> (ADDBconst [int64(int8(c+d))] x)
+(SUBQconst (MOVQconst [d]) [c]) -> (MOVQconst [d-c])
+(SUBLconst (MOVLconst [d]) [c]) -> (MOVLconst [int64(int32(d-c))])
+(SUBWconst (MOVWconst [d]) [c]) -> (MOVWconst [int64(int16(d-c))])
+(SUBBconst (MOVBconst [d]) [c]) -> (MOVBconst [int64(int8(d-c))])
+(SUBQconst (SUBQconst x [d]) [c]) && is32Bit(-c-d) -> (ADDQconst [-c-d] x)
+(SUBLconst (SUBLconst x [d]) [c]) -> (ADDLconst [int64(int32(-c-d))] x)
+(SUBWconst (SUBWconst x [d]) [c]) -> (ADDWconst [int64(int16(-c-d))] x)
+(SUBBconst (SUBBconst x [d]) [c]) -> (ADDBconst [int64(int8(-c-d))] x)
 (SARQconst [c] (MOVQconst [d])) -> (MOVQconst [d>>uint64(c)])
 (SARLconst [c] (MOVQconst [d])) -> (MOVQconst [d>>uint64(c)])
 (SARWconst [c] (MOVQconst [d])) -> (MOVQconst [d>>uint64(c)])
 (SARBconst [c] (MOVQconst [d])) -> (MOVQconst [d>>uint64(c)])
 (NEGQ (MOVQconst [c])) -> (MOVQconst [-c])
-(NEGL (MOVLconst [c])) -> (MOVLconst [-c])
-(NEGW (MOVWconst [c])) -> (MOVWconst [-c])
-(NEGB (MOVBconst [c])) -> (MOVBconst [-c])
+(NEGL (MOVLconst [c])) -> (MOVLconst [int64(int32(-c))])
+(NEGW (MOVWconst [c])) -> (MOVWconst [int64(int16(-c))])
+(NEGB (MOVBconst [c])) -> (MOVBconst [int64(int8(-c))])
 (MULQconst [c] (MOVQconst [d])) -> (MOVQconst [c*d])
-(MULLconst [c] (MOVLconst [d])) -> (MOVLconst [c*d])
-(MULWconst [c] (MOVWconst [d])) -> (MOVWconst [c*d])
-(MULBconst [c] (MOVBconst [d])) -> (MOVBconst [c*d])
+(MULLconst [c] (MOVLconst [d])) -> (MOVLconst [int64(int32(c*d))])
+(MULWconst [c] (MOVWconst [d])) -> (MOVWconst [int64(int16(c*d))])
+(MULBconst [c] (MOVBconst [d])) -> (MOVBconst [int64(int8(c*d))])
 (ANDQconst [c] (MOVQconst [d])) -> (MOVQconst [c&d])
 (ANDLconst [c] (MOVLconst [d])) -> (MOVLconst [c&d])
 (ANDWconst [c] (MOVWconst [d])) -> (MOVWconst [c&d])

src/cmd/compile/internal/ssa/gen/AMD64Ops.go

@@ -6,6 +6,25 @@ package main
 
 import "strings"
 
+// Notes:
+//  - Integer types live in the low portion of registers. Upper portions are junk.
+//  - Boolean types use the low-order byte of a register. 0=false, 1=true.
+//    Upper bytes are junk.
+//  - Floating-point types live in the low natural slot of an sse2 register.
+//    Unused portions are junk.
+//  - We do not use AH,BH,CH,DH registers.
+//  - When doing sub-register operations, we try to write the whole
+//    destination register to avoid a partial-register write.
+//  - Unused portions of AuxInt (or the Val portion of ValAndOff) are
+//    filled by sign-extending the used portion.  Users of AuxInt which interpret
+//    AuxInt as unsigned (e.g. shifts) must be careful.
+
+// Suffixes encode the bit width of various instructions.
+// Q (quad word) = 64 bit
+// L (long word) = 32 bit
+// W (word)      = 16 bit
+// B (byte)      = 8 bit
+
 // copied from ../../amd64/reg.go
 var regNamesAMD64 = []string{
 	"AX",
@@ -129,7 +148,6 @@ func init() {
 		gpfp     = regInfo{inputs: gponly, outputs: fponly}
 		fp11     = regInfo{inputs: fponly, outputs: fponly}
 		fp2flags = regInfo{inputs: []regMask{fp, fp}, outputs: flagsonly}
-		// fp1flags = regInfo{inputs: fponly, outputs: flagsonly}
 
 		fpload    = regInfo{inputs: []regMask{gpspsb, 0}, outputs: fponly}
 		fploadidx = regInfo{inputs: []regMask{gpspsb, gpsp, 0}, outputs: fponly}
@@ -137,12 +155,7 @@ func init() {
 		fpstore    = regInfo{inputs: []regMask{gpspsb, fp, 0}}
 		fpstoreidx = regInfo{inputs: []regMask{gpspsb, gpsp, fp, 0}}
 	)
-	// TODO: most ops clobber flags
-
-	// Suffixes encode the bit width of various instructions.
-	// Q = 64 bit, L = 32 bit, W = 16 bit, B = 8 bit
 
-	// TODO: 2-address instructions. Mark ops as needing matching input/output regs.
 	var AMD64ops = []opData{
 		// fp ops
 		{name: "ADDSS", argLength: 2, reg: fp21, asm: "ADDSS", commutative: true, resultInArg0: true}, // fp32 add

src/cmd/compile/internal/ssa/gen/genericOps.go

@@ -4,6 +4,19 @@
 
 package main
 
+// Generic opcodes typically specify a width. The inputs and outputs
+// of that op are the given number of bits wide. There is no notion of
+// "sign", so Add32 can be used both for signed and unsigned 32-bit
+// addition.
+
+// Signed/unsigned is explicit with the extension ops
+// (SignExt*/ZeroExt*) and implicit as the arg to some opcodes
+// (e.g. the second argument to shifts is unsigned). If not mentioned,
+// all args take signed inputs, or don't care whether their inputs
+// are signed or unsigned.
+
+// Unused portions of AuxInt are filled by sign-extending the used portion.
+// Users of AuxInt which interpret AuxInt as unsigned (e.g. shifts) must be careful.
 var genericOps = []opData{
 	// 2-input arithmetic
 	// Types must be consistent with Go typing. Add, for example, must take two values
@@ -15,7 +28,6 @@ var genericOps = []opData{
 	{name: "AddPtr", argLength: 2}, // For address calculations.  arg0 is a pointer and arg1 is an int.
 	{name: "Add32F", argLength: 2},
 	{name: "Add64F", argLength: 2},
-	// TODO: Add64C, Add128C
 
 	{name: "Sub8", argLength: 2}, // arg0 - arg1
 	{name: "Sub16", argLength: 2},
@@ -35,8 +47,8 @@ var genericOps = []opData{
 	{name: "Div32F", argLength: 2}, // arg0 / arg1
 	{name: "Div64F", argLength: 2},
 
-	{name: "Hmul8", argLength: 2}, // (arg0 * arg1) >> width
-	{name: "Hmul8u", argLength: 2},
+	{name: "Hmul8", argLength: 2},  // (arg0 * arg1) >> width, signed
+	{name: "Hmul8u", argLength: 2}, // (arg0 * arg1) >> width, unsigned
 	{name: "Hmul16", argLength: 2},
 	{name: "Hmul16u", argLength: 2},
 	{name: "Hmul32", argLength: 2},
@@ -47,8 +59,8 @@ var genericOps = []opData{
 	// Weird special instruction for strength reduction of divides.
 	{name: "Avg64u", argLength: 2}, // (uint64(arg0) + uint64(arg1)) / 2, correct to all 64 bits.
 
-	{name: "Div8", argLength: 2}, // arg0 / arg1
-	{name: "Div8u", argLength: 2},
+	{name: "Div8", argLength: 2},  // arg0 / arg1, signed
+	{name: "Div8u", argLength: 2}, // arg0 / arg1, unsigned
 	{name: "Div16", argLength: 2},
 	{name: "Div16u", argLength: 2},
 	{name: "Div32", argLength: 2},
@@ -56,8 +68,8 @@ var genericOps = []opData{
 	{name: "Div64", argLength: 2},
 	{name: "Div64u", argLength: 2},
 
-	{name: "Mod8", argLength: 2}, // arg0 % arg1
-	{name: "Mod8u", argLength: 2},
+	{name: "Mod8", argLength: 2},  // arg0 % arg1, signed
+	{name: "Mod8u", argLength: 2}, // arg0 % arg1, unsigned
 	{name: "Mod16", argLength: 2},
 	{name: "Mod16u", argLength: 2},
 	{name: "Mod32", argLength: 2},
@@ -81,6 +93,7 @@ var genericOps = []opData{
 	{name: "Xor64", argLength: 2, commutative: true},
 
 	// For shifts, AxB means the shifted value has A bits and the shift amount has B bits.
+	// Shift amounts are considered unsigned.
 	{name: "Lsh8x8", argLength: 2}, // arg0 << arg1
 	{name: "Lsh8x16", argLength: 2},
 	{name: "Lsh8x32", argLength: 2},
@@ -178,8 +191,8 @@ var genericOps = []opData{
 	{name: "Neq32F", argLength: 2},
 	{name: "Neq64F", argLength: 2},
 
-	{name: "Less8", argLength: 2}, // arg0 < arg1
-	{name: "Less8U", argLength: 2},
+	{name: "Less8", argLength: 2},  // arg0 < arg1, signed
+	{name: "Less8U", argLength: 2}, // arg0 < arg1, unsigned
 	{name: "Less16", argLength: 2},
 	{name: "Less16U", argLength: 2},
 	{name: "Less32", argLength: 2},
@@ -189,8 +202,8 @@ var genericOps = []opData{
 	{name: "Less32F", argLength: 2},
 	{name: "Less64F", argLength: 2},
 
-	{name: "Leq8", argLength: 2}, // arg0 <= arg1
-	{name: "Leq8U", argLength: 2},
+	{name: "Leq8", argLength: 2},  // arg0 <= arg1, signed
+	{name: "Leq8U", argLength: 2}, // arg0 <= arg1, unsigned
 	{name: "Leq16", argLength: 2},
 	{name: "Leq16U", argLength: 2},
 	{name: "Leq32", argLength: 2},
@@ -200,8 +213,8 @@ var genericOps = []opData{
 	{name: "Leq32F", argLength: 2},
 	{name: "Leq64F", argLength: 2},
 
-	{name: "Greater8", argLength: 2}, // arg0 > arg1
-	{name: "Greater8U", argLength: 2},
+	{name: "Greater8", argLength: 2},  // arg0 > arg1, signed
+	{name: "Greater8U", argLength: 2}, // arg0 > arg1, unsigned
 	{name: "Greater16", argLength: 2},
 	{name: "Greater16U", argLength: 2},
 	{name: "Greater32", argLength: 2},
@@ -211,8 +224,8 @@ var genericOps = []opData{
 	{name: "Greater32F", argLength: 2},
 	{name: "Greater64F", argLength: 2},
 
-	{name: "Geq8", argLength: 2}, // arg0 <= arg1
-	{name: "Geq8U", argLength: 2},
+	{name: "Geq8", argLength: 2},  // arg0 <= arg1, signed
+	{name: "Geq8U", argLength: 2}, // arg0 <= arg1, unsigned
 	{name: "Geq16", argLength: 2},
 	{name: "Geq16U", argLength: 2},
 	{name: "Geq32", argLength: 2},
@@ -223,7 +236,7 @@ var genericOps = []opData{
 	{name: "Geq64F", argLength: 2},
 
 	// 1-input ops
-	{name: "Not", argLength: 1}, // !arg0
+	{name: "Not", argLength: 1}, // !arg0, boolean
 
 	{name: "Neg8", argLength: 1}, // -arg0
 	{name: "Neg16", argLength: 1},
@@ -266,9 +279,9 @@ var genericOps = []opData{
 	{name: "ConstBool", aux: "Bool"},     // auxint is 0 for false and 1 for true
 	{name: "ConstString", aux: "String"}, // value is aux.(string)
 	{name: "ConstNil", typ: "BytePtr"},   // nil pointer
-	{name: "Const8", aux: "Int8"},        // value is low 8 bits of auxint
-	{name: "Const16", aux: "Int16"},      // value is low 16 bits of auxint
-	{name: "Const32", aux: "Int32"},      // value is low 32 bits of auxint
+	{name: "Const8", aux: "Int8"},        // auxint is sign-extended 8 bits
+	{name: "Const16", aux: "Int16"},      // auxint is sign-extended 16 bits
+	{name: "Const32", aux: "Int32"},      // auxint is sign-extended 32 bits
 	{name: "Const64", aux: "Int64"},      // value is auxint
 	{name: "Const32F", aux: "Float32"},   // value is math.Float64frombits(uint64(auxint)) and is exactly prepresentable as float 32
 	{name: "Const64F", aux: "Float64"},   // value is math.Float64frombits(uint64(auxint))
@@ -337,8 +350,8 @@ var genericOps = []opData{
 
 	// Automatically inserted safety checks
 	{name: "IsNonNil", argLength: 1, typ: "Bool"},        // arg0 != nil
-	{name: "IsInBounds", argLength: 2, typ: "Bool"},      // 0 <= arg0 < arg1
-	{name: "IsSliceInBounds", argLength: 2, typ: "Bool"}, // 0 <= arg0 <= arg1
+	{name: "IsInBounds", argLength: 2, typ: "Bool"},      // 0 <= arg0 < arg1. arg1 is guaranteed >= 0.
+	{name: "IsSliceInBounds", argLength: 2, typ: "Bool"}, // 0 <= arg0 <= arg1. arg1 is guaranteed >= 0.
 	{name: "NilCheck", argLength: 2, typ: "Void"},        // arg0=ptr, arg1=mem. Panics if arg0 is nil, returns void.
 
 	// Pseudo-ops

src/cmd/compile/internal/ssa/rewrite.go

@@ -111,13 +111,6 @@ func canMergeSym(x, y interface{}) bool {
 	return x == nil || y == nil
 }
 
-func inBounds8(idx, len int64) bool       { return int8(idx) >= 0 && int8(idx) < int8(len) }
-func inBounds16(idx, len int64) bool      { return int16(idx) >= 0 && int16(idx) < int16(len) }
-func inBounds32(idx, len int64) bool      { return int32(idx) >= 0 && int32(idx) < int32(len) }
-func inBounds64(idx, len int64) bool      { return idx >= 0 && idx < len }
-func sliceInBounds32(idx, len int64) bool { return int32(idx) >= 0 && int32(idx) <= int32(len) }
-func sliceInBounds64(idx, len int64) bool { return idx >= 0 && idx <= len }
-
 // nlz returns the number of leading zeros.
 func nlz(x int64) int64 {
 	// log2(0) == 1, so nlz(0) == 64

src/cmd/compile/internal/ssa/rewriteAMD64.go

@@ -840,7 +840,7 @@ func rewriteValueAMD64_OpAMD64ADDBconst(v *Value, config *Config) bool {
 	}
 	// match: (ADDBconst [c] (MOVBconst [d]))
 	// cond:
-	// result: (MOVBconst [c+d])
+	// result: (MOVBconst [int64(int8(c+d))])
 	for {
 		c := v.AuxInt
 		v_0 := v.Args[0]
@@ -849,12 +849,12 @@ func rewriteValueAMD64_OpAMD64ADDBconst(v *Value, config *Config) bool {
 		}
 		d := v_0.AuxInt
 		v.reset(OpAMD64MOVBconst)
-		v.AuxInt = c + d
+		v.AuxInt = int64(int8(c + d))
 		return true
 	}
 	// match: (ADDBconst [c] (ADDBconst [d] x))
 	// cond:
-	// result: (ADDBconst [c+d] x)
+	// result: (ADDBconst [int64(int8(c+d))] x)
 	for {
 		c := v.AuxInt
 		v_0 := v.Args[0]
@@ -864,7 +864,7 @@ func rewriteValueAMD64_OpAMD64ADDBconst(v *Value, config *Config) bool {
 		d := v_0.AuxInt
 		x := v_0.Args[0]
 		v.reset(OpAMD64ADDBconst)
-		v.AuxInt = c + d
+		v.AuxInt = int64(int8(c + d))
 		v.AddArg(x)
 		return true
 	}
@@ -939,7 +939,7 @@ func rewriteValueAMD64_OpAMD64ADDLconst(v *Value, config *Config) bool {
 	}
 	// match: (ADDLconst [c] (MOVLconst [d]))
 	// cond:
-	// result: (MOVLconst [c+d])
+	// result: (MOVLconst [int64(int32(c+d))])
 	for {
 		c := v.AuxInt
 		v_0 := v.Args[0]
@@ -948,12 +948,12 @@ func rewriteValueAMD64_OpAMD64ADDLconst(v *Value, config *Config) bool {
 		}
 		d := v_0.AuxInt
 		v.reset(OpAMD64MOVLconst)
-		v.AuxInt = c + d
+		v.AuxInt = int64(int32(c + d))
 		return true
 	}
 	// match: (ADDLconst [c] (ADDLconst [d] x))
 	// cond:
-	// result: (ADDLconst [c+d] x)
+	// result: (ADDLconst [int64(int32(c+d))] x)
 	for {
 		c := v.AuxInt
 		v_0 := v.Args[0]
@@ -963,7 +963,7 @@ func rewriteValueAMD64_OpAMD64ADDLconst(v *Value, config *Config) bool {
 		d := v_0.AuxInt
 		x := v_0.Args[0]
 		v.reset(OpAMD64ADDLconst)
-		v.AuxInt = c + d
+		v.AuxInt = int64(int32(c + d))
 		v.AddArg(x)
 		return true
 	}
@@ -1461,7 +1461,7 @@ func rewriteValueAMD64_OpAMD64ADDWconst(v *Value, config *Config) bool {
 	}
 	// match: (ADDWconst [c] (MOVWconst [d]))
 	// cond:
-	// result: (MOVWconst [c+d])
+	// result: (MOVWconst [int64(int16(c+d))])
 	for {
 		c := v.AuxInt
 		v_0 := v.Args[0]
@@ -1470,12 +1470,12 @@ func rewriteValueAMD64_OpAMD64ADDWconst(v *Value, config *Config) bool {
 		}
 		d := v_0.AuxInt
 		v.reset(OpAMD64MOVWconst)
-		v.AuxInt = c + d
+		v.AuxInt = int64(int16(c + d))
 		return true
 	}
 	// match: (ADDWconst [c] (ADDWconst [d] x))
 	// cond:
-	// result: (ADDWconst [c+d] x)
+	// result: (ADDWconst [int64(int16(c+d))] x)
 	for {
 		c := v.AuxInt
 		v_0 := v.Args[0]
@@ -1485,7 +1485,7 @@ func rewriteValueAMD64_OpAMD64ADDWconst(v *Value, config *Config) bool {
 		d := v_0.AuxInt
 		x := v_0.Args[0]
 		v.reset(OpAMD64ADDWconst)
-		v.AuxInt = c + d
+		v.AuxInt = int64(int16(c + d))
 		v.AddArg(x)
 		return true
 	}
@@ -9218,7 +9218,7 @@ func rewriteValueAMD64_OpAMD64MULBconst(v *Value, config *Config) bool {
 	_ = b
 	// match: (MULBconst [c] (MOVBconst [d]))
 	// cond:
-	// result: (MOVBconst [c*d])
+	// result: (MOVBconst [int64(int8(c*d))])
 	for {
 		c := v.AuxInt
 		v_0 := v.Args[0]
@@ -9227,7 +9227,7 @@ func rewriteValueAMD64_OpAMD64MULBconst(v *Value, config *Config) bool {
 		}
 		d := v_0.AuxInt
 		v.reset(OpAMD64MOVBconst)
-		v.AuxInt = c * d
+		v.AuxInt = int64(int8(c * d))
 		return true
 	}
 	return false
@@ -9272,7 +9272,7 @@ func rewriteValueAMD64_OpAMD64MULLconst(v *Value, config *Config) bool {
 	_ = b
 	// match: (MULLconst [c] (MOVLconst [d]))
 	// cond:
-	// result: (MOVLconst [c*d])
+	// result: (MOVLconst [int64(int32(c*d))])
 	for {
 		c := v.AuxInt
 		v_0 := v.Args[0]
@@ -9281,7 +9281,7 @@ func rewriteValueAMD64_OpAMD64MULLconst(v *Value, config *Config) bool {
 		}
 		d := v_0.AuxInt
 		v.reset(OpAMD64MOVLconst)
-		v.AuxInt = c * d
+		v.AuxInt = int64(int32(c * d))
 		return true
 	}
 	return false
@@ -9491,7 +9491,7 @@ func rewriteValueAMD64_OpAMD64MULWconst(v *Value, config *Config) bool {
 	_ = b
 	// match: (MULWconst [c] (MOVWconst [d]))
 	// cond:
-	// result: (MOVWconst [c*d])
+	// result: (MOVWconst [int64(int16(c*d))])
 	for {
 		c := v.AuxInt
 		v_0 := v.Args[0]
@@ -9500,7 +9500,7 @@ func rewriteValueAMD64_OpAMD64MULWconst(v *Value, config *Config) bool {
 		}
 		d := v_0.AuxInt
 		v.reset(OpAMD64MOVWconst)
-		v.AuxInt = c * d
+		v.AuxInt = int64(int16(c * d))
 		return true
 	}
 	return false
@@ -10096,7 +10096,7 @@ func rewriteValueAMD64_OpAMD64NEGB(v *Value, config *Config) bool {
 	_ = b
 	// match: (NEGB (MOVBconst [c]))
 	// cond:
-	// result: (MOVBconst [-c])
+	// result: (MOVBconst [int64(int8(-c))])
 	for {
 		v_0 := v.Args[0]
 		if v_0.Op != OpAMD64MOVBconst {
@@ -10104,7 +10104,7 @@ func rewriteValueAMD64_OpAMD64NEGB(v *Value, config *Config) bool {
 		}
 		c := v_0.AuxInt
 		v.reset(OpAMD64MOVBconst)
-		v.AuxInt = -c
+		v.AuxInt = int64(int8(-c))
 		return true
 	}
 	return false
@@ -10114,7 +10114,7 @@ func rewriteValueAMD64_OpAMD64NEGL(v *Value, config *Config) bool {
 	_ = b
 	// match: (NEGL (MOVLconst [c]))
 	// cond:
-	// result: (MOVLconst [-c])
+	// result: (MOVLconst [int64(int32(-c))])
 	for {
 		v_0 := v.Args[0]
 		if v_0.Op != OpAMD64MOVLconst {
@@ -10122,7 +10122,7 @@ func rewriteValueAMD64_OpAMD64NEGL(v *Value, config *Config) bool {
 		}
 		c := v_0.AuxInt
 		v.reset(OpAMD64MOVLconst)
-		v.AuxInt = -c
+		v.AuxInt = int64(int32(-c))
 		return true
 	}
 	return false
@@ -10150,7 +10150,7 @@ func rewriteValueAMD64_OpAMD64NEGW(v *Value, config *Config) bool {
 	_ = b
 	// match: (NEGW (MOVWconst [c]))
 	// cond:
-	// result: (MOVWconst [-c])
+	// result: (MOVWconst [int64(int16(-c))])
 	for {
 		v_0 := v.Args[0]
 		if v_0.Op != OpAMD64MOVWconst {
@@ -10158,7 +10158,7 @@ func rewriteValueAMD64_OpAMD64NEGW(v *Value, config *Config) bool {
 		}
 		c := v_0.AuxInt
 		v.reset(OpAMD64MOVWconst)
-		v.AuxInt = -c
+		v.AuxInt = int64(int16(-c))
 		return true
 	}
 	return false
@@ -14591,33 +14591,33 @@ func rewriteValueAMD64_OpAMD64SUBBconst(v *Value, config *Config) bool {
 		v.AddArg(x)
 		return true
 	}
-	// match: (SUBBconst [c] (MOVBconst [d]))
+	// match: (SUBBconst (MOVBconst [d]) [c])
 	// cond:
-	// result: (MOVBconst [d-c])
+	// result: (MOVBconst [int64(int8(d-c))])
 	for {
-		c := v.AuxInt
 		v_0 := v.Args[0]
 		if v_0.Op != OpAMD64MOVBconst {
 			break
 		}
 		d := v_0.AuxInt
+		c := v.AuxInt
 		v.reset(OpAMD64MOVBconst)
-		v.AuxInt = d - c
+		v.AuxInt = int64(int8(d - c))
 		return true
 	}
-	// match: (SUBBconst [c] (SUBBconst [d] x))
+	// match: (SUBBconst (SUBBconst x [d]) [c])
 	// cond:
-	// result: (ADDBconst [-c-d] x)
+	// result: (ADDBconst [int64(int8(-c-d))] x)
 	for {
-		c := v.AuxInt
 		v_0 := v.Args[0]
 		if v_0.Op != OpAMD64SUBBconst {
 			break
 		}
-		d := v_0.AuxInt
 		x := v_0.Args[0]
+		d := v_0.AuxInt
+		c := v.AuxInt
 		v.reset(OpAMD64ADDBconst)
-		v.AuxInt = -c - d
+		v.AuxInt = int64(int8(-c - d))
 		v.AddArg(x)
 		return true
 	}
@@ -14689,33 +14689,33 @@ func rewriteValueAMD64_OpAMD64SUBLconst(v *Value, config *Config) bool {
 		v.AddArg(x)
 		return true
 	}
-	// match: (SUBLconst [c] (MOVLconst [d]))
+	// match: (SUBLconst (MOVLconst [d]) [c])
 	// cond:
-	// result: (MOVLconst [d-c])
+	// result: (MOVLconst [int64(int32(d-c))])
 	for {
-		c := v.AuxInt
 		v_0 := v.Args[0]
 		if v_0.Op != OpAMD64MOVLconst {
 			break
 		}
 		d := v_0.AuxInt
+		c := v.AuxInt
 		v.reset(OpAMD64MOVLconst)
-		v.AuxInt = d - c
+		v.AuxInt = int64(int32(d - c))
 		return true
 	}
-	// match: (SUBLconst [c] (SUBLconst [d] x))
+	// match: (SUBLconst (SUBLconst x [d]) [c])
 	// cond:
-	// result: (ADDLconst [-c-d] x)
+	// result: (ADDLconst [int64(int32(-c-d))] x)
 	for {
-		c := v.AuxInt
 		v_0 := v.Args[0]
 		if v_0.Op != OpAMD64SUBLconst {
 			break
 		}
-		d := v_0.AuxInt
 		x := v_0.Args[0]
+		d := v_0.AuxInt
+		c := v.AuxInt
 		v.reset(OpAMD64ADDLconst)
-		v.AuxInt = -c - d
+		v.AuxInt = int64(int32(-c - d))
 		v.AddArg(x)
 		return true
 	}
@@ -14792,31 +14792,31 @@ func rewriteValueAMD64_OpAMD64SUBQconst(v *Value, config *Config) bool {
 		v.AddArg(x)
 		return true
 	}
-	// match: (SUBQconst [c] (MOVQconst [d]))
+	// match: (SUBQconst (MOVQconst [d]) [c])
 	// cond:
 	// result: (MOVQconst [d-c])
 	for {
-		c := v.AuxInt
 		v_0 := v.Args[0]
 		if v_0.Op != OpAMD64MOVQconst {
 			break
 		}
 		d := v_0.AuxInt
+		c := v.AuxInt
 		v.reset(OpAMD64MOVQconst)
 		v.AuxInt = d - c
 		return true
 	}
-	// match: (SUBQconst [c] (SUBQconst [d] x))
+	// match: (SUBQconst (SUBQconst x [d]) [c])
 	// cond: is32Bit(-c-d)
 	// result: (ADDQconst [-c-d] x)
 	for {
-		c := v.AuxInt
 		v_0 := v.Args[0]
 		if v_0.Op != OpAMD64SUBQconst {
 			break
 		}
-		d := v_0.AuxInt
 		x := v_0.Args[0]
+		d := v_0.AuxInt
+		c := v.AuxInt
 		if !(is32Bit(-c - d)) {
 			break
 		}
@@ -14893,33 +14893,33 @@ func rewriteValueAMD64_OpAMD64SUBWconst(v *Value, config *Config) bool {
 		v.AddArg(x)
 		return true
 	}
-	// match: (SUBWconst [c] (MOVWconst [d]))
+	// match: (SUBWconst (MOVWconst [d]) [c])
 	// cond:
-	// result: (MOVWconst [d-c])
+	// result: (MOVWconst [int64(int16(d-c))])
 	for {
-		c := v.AuxInt
 		v_0 := v.Args[0]
 		if v_0.Op != OpAMD64MOVWconst {
 			break
 		}
 		d := v_0.AuxInt
+		c := v.AuxInt
 		v.reset(OpAMD64MOVWconst)
-		v.AuxInt = d - c
+		v.AuxInt = int64(int16(d - c))
 		return true
 	}
-	// match: (SUBWconst [c] (SUBWconst [d] x))
+	// match: (SUBWconst (SUBWconst x [d]) [c])
 	// cond:
-	// result: (ADDWconst [-c-d] x)
+	// result: (ADDWconst [int64(int16(-c-d))] x)
 	for {
-		c := v.AuxInt
 		v_0 := v.Args[0]
 		if v_0.Op != OpAMD64SUBWconst {
 			break
 		}
-		d := v_0.AuxInt
 		x := v_0.Args[0]
+		d := v_0.AuxInt
+		c := v.AuxInt
 		v.reset(OpAMD64ADDWconst)
-		v.AuxInt = -c - d
+		v.AuxInt = int64(int16(-c - d))
 		v.AddArg(x)
 		return true
 	}

src/cmd/compile/internal/ssa/value.go

@@ -81,24 +81,6 @@ func (v *Value) AuxInt32() int32 {
 	return int32(v.AuxInt)
 }
 
-// AuxInt2Int64 is used to sign extend the lower bits of AuxInt according to
-// the size of AuxInt specified in the opcode table.
-func (v *Value) AuxInt2Int64() int64 {
-	switch opcodeTable[v.Op].auxType {
-	case auxInt64:
-		return v.AuxInt
-	case auxInt32:
-		return int64(int32(v.AuxInt))
-	case auxInt16:
-		return int64(int16(v.AuxInt))
-	case auxInt8:
-		return int64(int8(v.AuxInt))
-	default:
-		v.Fatalf("op %s doesn't have an aux int field", v.Op)
-		return -1
-	}
-}
-
 func (v *Value) AuxFloat() float64 {
 	if opcodeTable[v.Op].auxType != auxFloat32 && opcodeTable[v.Op].auxType != auxFloat64 {
 		v.Fatalf("op %s doesn't have a float aux field", v.Op)