cmd/compile: omit redundant sign/unsign extension on arm64

On Arm64, all 32-bit instructions will ignore the upper 32 bits and
clear them to zero for the result. No need to do an unsign extend before
a 32 bit op.

This CL removes the redundant unsign extension only for the existing
32-bit opcodes, and also omits the sign extension when the upper bit of
the result can be predicted.

Fixes #42162

Change-Id: I61e6670bfb8982572430e67a4fa61134a3ea240a
CustomizedGitHooks: yes
Reviewed-on: https://go-review.googlesource.com/c/go/+/427454
Reviewed-by: Keith Randall <khr@google.com>
Auto-Submit: Eric Fang <eric.fang@arm.com>
Reviewed-by: Keith Randall <khr@golang.org>
Run-TryBot: Eric Fang <eric.fang@arm.com>
Reviewed-by: Cherry Mui <cherryyz@google.com>
TryBot-Result: Gopher Robot <gobot@golang.org>

5 files changed, 177 insertions, 0 deletions

diff --git a/src/cmd/compile/internal/ssa/_gen/ARM64.rules b/src/cmd/compile/internal/ssa/_gen/ARM64.rules
index cf43542615..34ee907e3a 100644
--- a/src/cmd/compile/internal/ssa/_gen/ARM64.rules
+++ b/src/cmd/compile/internal/ssa/_gen/ARM64.rules
@@ -1665,6 +1665,16 @@
 // zero upper bit of the register; no need to zero-extend
 (MOVBUreg x:((Equal|NotEqual|LessThan|LessThanU|LessThanF|LessEqual|LessEqualU|LessEqualF|GreaterThan|GreaterThanU|GreaterThanF|GreaterEqual|GreaterEqualU|GreaterEqualF) _)) => (MOVDreg x)
 
+// omit unsign extension
+
+(MOVWUreg x) && zeroUpper32Bits(x, 3) => x
+
+// omit sign extension
+
+(MOVWreg <t> (ANDconst x [c])) && uint64(c) & uint64(0xffffffff80000000) == 0 => (ANDconst <t> x [c])
+(MOVHreg <t> (ANDconst x [c])) && uint64(c) & uint64(0xffffffffffff8000) == 0 => (ANDconst <t> x [c])
+(MOVBreg <t> (ANDconst x [c])) && uint64(c) & uint64(0xffffffffffffff80) == 0 => (ANDconst <t> x [c])
+
 // absorb flag constants into conditional instructions
 (CSEL [cc] x _ flag) && ccARM64Eval(cc, flag) > 0 => x
 (CSEL [cc] _ y flag) && ccARM64Eval(cc, flag) < 0 => y
diff --git a/src/cmd/compile/internal/ssa/_gen/ARM64Ops.go b/src/cmd/compile/internal/ssa/_gen/ARM64Ops.go
index 2a9c2ae486..ca5b929ad7 100644
--- a/src/cmd/compile/internal/ssa/_gen/ARM64Ops.go
+++ b/src/cmd/compile/internal/ssa/_gen/ARM64Ops.go
@@ -13,6 +13,7 @@ import "strings"
 //  - *const instructions may use a constant larger than the instruction can encode.
 //    In this case the assembler expands to multiple instructions and uses tmp
 //    register (R27).
+//  - All 32-bit Ops will zero the upper 32 bits of the destination register.
 
 // Suffixes encode the bit width of various instructions.
 // D (double word) = 64 bit
diff --git a/src/cmd/compile/internal/ssa/rewrite.go b/src/cmd/compile/internal/ssa/rewrite.go
index c56447d336..54ea2d3f4f 100644
--- a/src/cmd/compile/internal/ssa/rewrite.go
+++ b/src/cmd/compile/internal/ssa/rewrite.go
@@ -1298,6 +1298,10 @@ func zeroUpper32Bits(x *Value, depth int) bool {
 		OpAMD64SHRL, OpAMD64SHRLconst, OpAMD64SARL, OpAMD64SARLconst,
 		OpAMD64SHLL, OpAMD64SHLLconst:
 		return true
+	case OpARM64REV16W, OpARM64REVW, OpARM64RBITW, OpARM64CLZW, OpARM64EXTRWconst,
+		OpARM64MULW, OpARM64MNEGW, OpARM64UDIVW, OpARM64DIVW, OpARM64UMODW,
+		OpARM64MADDW, OpARM64MSUBW, OpARM64RORW, OpARM64RORWconst:
+		return true
 	case OpArg:
 		return x.Type.Size() == 4
 	case OpPhi, OpSelect0, OpSelect1:
diff --git a/src/cmd/compile/internal/ssa/rewriteARM64.go b/src/cmd/compile/internal/ssa/rewriteARM64.go
index 7cc7a2a424..a43a366c67 100644
--- a/src/cmd/compile/internal/ssa/rewriteARM64.go
+++ b/src/cmd/compile/internal/ssa/rewriteARM64.go
@@ -8271,6 +8271,25 @@ func rewriteValueARM64_OpARM64MOVBreg(v *Value) bool {
 		v.AuxInt = int64ToAuxInt(int64(int8(c)))
 		return true
 	}
+	// match: (MOVBreg <t> (ANDconst x [c]))
+	// cond: uint64(c) & uint64(0xffffffffffffff80) == 0
+	// result: (ANDconst <t> x [c])
+	for {
+		t := v.Type
+		if v_0.Op != OpARM64ANDconst {
+			break
+		}
+		c := auxIntToInt64(v_0.AuxInt)
+		x := v_0.Args[0]
+		if !(uint64(c)&uint64(0xffffffffffffff80) == 0) {
+			break
+		}
+		v.reset(OpARM64ANDconst)
+		v.Type = t
+		v.AuxInt = int64ToAuxInt(c)
+		v.AddArg(x)
+		return true
+	}
 	// match: (MOVBreg (SLLconst [lc] x))
 	// cond: lc < 8
 	// result: (SBFIZ [armBFAuxInt(lc, 8-lc)] x)
@@ -11991,6 +12010,25 @@ func rewriteValueARM64_OpARM64MOVHreg(v *Value) bool {
 		v.AuxInt = int64ToAuxInt(int64(int16(c)))
 		return true
 	}
+	// match: (MOVHreg <t> (ANDconst x [c]))
+	// cond: uint64(c) & uint64(0xffffffffffff8000) == 0
+	// result: (ANDconst <t> x [c])
+	for {
+		t := v.Type
+		if v_0.Op != OpARM64ANDconst {
+			break
+		}
+		c := auxIntToInt64(v_0.AuxInt)
+		x := v_0.Args[0]
+		if !(uint64(c)&uint64(0xffffffffffff8000) == 0) {
+			break
+		}
+		v.reset(OpARM64ANDconst)
+		v.Type = t
+		v.AuxInt = int64ToAuxInt(c)
+		v.AddArg(x)
+		return true
+	}
 	// match: (MOVHreg (SLLconst [lc] x))
 	// cond: lc < 16
 	// result: (SBFIZ [armBFAuxInt(lc, 16-lc)] x)
@@ -13687,6 +13725,17 @@ func rewriteValueARM64_OpARM64MOVWUreg(v *Value) bool {
 		v.AuxInt = int64ToAuxInt(int64(uint32(c)))
 		return true
 	}
+	// match: (MOVWUreg x)
+	// cond: zeroUpper32Bits(x, 3)
+	// result: x
+	for {
+		x := v_0
+		if !(zeroUpper32Bits(x, 3)) {
+			break
+		}
+		v.copyOf(x)
+		return true
+	}
 	// match: (MOVWUreg (SLLconst [lc] x))
 	// cond: lc >= 32
 	// result: (MOVDconst [0])
@@ -14189,6 +14238,25 @@ func rewriteValueARM64_OpARM64MOVWreg(v *Value) bool {
 		v.AuxInt = int64ToAuxInt(int64(int32(c)))
 		return true
 	}
+	// match: (MOVWreg <t> (ANDconst x [c]))
+	// cond: uint64(c) & uint64(0xffffffff80000000) == 0
+	// result: (ANDconst <t> x [c])
+	for {
+		t := v.Type
+		if v_0.Op != OpARM64ANDconst {
+			break
+		}
+		c := auxIntToInt64(v_0.AuxInt)
+		x := v_0.Args[0]
+		if !(uint64(c)&uint64(0xffffffff80000000) == 0) {
+			break
+		}
+		v.reset(OpARM64ANDconst)
+		v.Type = t
+		v.AuxInt = int64ToAuxInt(c)
+		v.AddArg(x)
+		return true
+	}
 	// match: (MOVWreg (SLLconst [lc] x))
 	// cond: lc < 32
 	// result: (SBFIZ [armBFAuxInt(lc, 32-lc)] x)
diff --git a/test/codegen/noextend.go b/test/codegen/noextend.go
index e4081e3915..61c98a796a 100644
--- a/test/codegen/noextend.go
+++ b/test/codegen/noextend.go
@@ -6,6 +6,8 @@
 
 package codegen
 
+import "math/bits"
+
 var sval64 [8]int64
 var sval32 [8]int32
 var sval16 [8]int16
@@ -185,3 +187,95 @@ func cmp64(u8 *uint8, x16 *int16, u16 *uint16, x32 *int32, u32 *uint32) bool {
 	}
 	return false
 }
+
+// no unsign extension following 32 bits ops
+
+func noUnsignEXT(t1, t2, t3, t4 uint32, k int64) uint64 {
+	var ret uint64
+
+	// arm64:"RORW",-"MOVWU"
+	ret += uint64(bits.RotateLeft32(t1, 7))
+
+	// arm64:"MULW",-"MOVWU"
+	ret *= uint64(t1 * t2)
+
+	// arm64:"MNEGW",-"MOVWU"
+	ret += uint64(-t1 * t3)
+
+	// arm64:"UDIVW",-"MOVWU"
+	ret += uint64(t1 / t4)
+
+	// arm64:-"MOVWU"
+	ret += uint64(t2 % t3)
+
+	// arm64:"MSUBW",-"MOVWU"
+	ret += uint64(t1 - t2*t3)
+
+	// arm64:"MADDW",-"MOVWU"
+	ret += uint64(t3*t4 + t2)
+
+	// arm64:"REVW",-"MOVWU"
+	ret += uint64(bits.ReverseBytes32(t1))
+
+	// arm64:"RBITW",-"MOVWU"
+	ret += uint64(bits.Reverse32(t1))
+
+	// arm64:"CLZW",-"MOVWU"
+	ret += uint64(bits.LeadingZeros32(t1))
+
+	// arm64:"REV16W",-"MOVWU"
+	ret += uint64(((t1 & 0xff00ff00) >> 8) | ((t1 & 0x00ff00ff) << 8))
+
+	// arm64:"EXTRW",-"MOVWU"
+	ret += uint64((t1 << 25) | (t2 >> 7))
+
+	return ret
+}
+
+// no sign extension when the upper bits of the result are zero
+
+func noSignEXT(x int) int64 {
+	t1 := int32(x)
+
+	var ret int64
+
+	// arm64:-"MOVW"
+	ret += int64(t1 & 1)
+
+	// arm64:-"MOVW"
+	ret += int64(int32(x & 0x7fffffff))
+
+	// arm64:-"MOVH"
+	ret += int64(int16(x & 0x7fff))
+
+	// arm64:-"MOVB"
+	ret += int64(int8(x & 0x7f))
+
+	return ret
+}
+
+// corner cases that sign extension must not be omitted
+
+func shouldSignEXT(x int) int64 {
+	t1 := int32(x)
+
+	var ret int64
+
+	// arm64:"MOVW"
+	ret += int64(t1 & (-1))
+
+	// arm64:"MOVW"
+	ret += int64(int32(x & 0x80000000))
+
+	// arm64:"MOVW"
+	ret += int64(int32(x & 0x1100000011111111))
+
+	// arm64:"MOVH"
+	ret += int64(int16(x & 0x1100000000001111))
+
+	// arm64:"MOVB"
+	ret += int64(int8(x & 0x1100000000000011))
+
+	return ret
+
+}