cmd/compile: intrinsics for math/bits.TrailingZerosX

Implement math/bits.TrailingZerosX using intrinsics.

Generally reorganize the intrinsic spec a bit.
The instrinsics data structure is now built at init time.
This will make doing the other functions in math/bits easier.

Update sys.CtzX to return int instead of uint{64,32} so it
matches math/bits.TrailingZerosX.

Improve the intrinsics a bit for amd64.  We don't need the CMOV
for <64 bit versions.

Update #18616

Change-Id: Ic1c5339c943f961d830ae56f12674d7b29d4ff39
Reviewed-on: https://go-review.googlesource.com/38155
Run-TryBot: Keith Randall <khr@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Robert Griesemer <gri@golang.org>

src/cmd/compile/internal/gc/asm_test.go

@@ -162,7 +162,7 @@ var allAsmTests = []*asmTests{
 	{
 		arch:    "amd64",
 		os:      "linux",
-		imports: []string{"encoding/binary"},
+		imports: []string{"encoding/binary", "math/bits"},
 		tests:   linuxAMD64Tests,
 	},
 	{
@@ -174,7 +174,7 @@ var allAsmTests = []*asmTests{
 	{
 		arch:    "s390x",
 		os:      "linux",
-		imports: []string{"encoding/binary"},
+		imports: []string{"encoding/binary", "math/bits"},
 		tests:   linuxS390XTests,
 	},
 	{
@@ -543,6 +543,39 @@ var linuxAMD64Tests = []*asmTest{
 		`,
 		[]string{"\tBTQ\t\\$60"},
 	},
+	// Intrinsic tests for math/bits
+	{
+		`
+		func f41(a uint64) int {
+			return bits.TrailingZeros64(a)
+		}
+		`,
+		[]string{"\tBSFQ\t", "\tMOVQ\t\\$64,", "\tCMOVQEQ\t"},
+	},
+	{
+		`
+		func f42(a uint32) int {
+			return bits.TrailingZeros32(a)
+		}
+		`,
+		[]string{"\tBSFQ\t", "\tORQ\t[^$]", "\tMOVQ\t\\$4294967296,"},
+	},
+	{
+		`
+		func f43(a uint16) int {
+			return bits.TrailingZeros16(a)
+		}
+		`,
+		[]string{"\tBSFQ\t", "\tORQ\t\\$65536,"},
+	},
+	{
+		`
+		func f44(a uint8) int {
+			return bits.TrailingZeros8(a)
+		}
+		`,
+		[]string{"\tBSFQ\t", "\tORQ\t\\$256,"},
+	},
 }
 
 var linux386Tests = []*asmTest{
@@ -710,6 +743,39 @@ var linuxS390XTests = []*asmTest{
 		`,
 		[]string{"\tFMSUBS\t"},
 	},
+	// Intrinsic tests for math/bits
+	{
+		`
+		func f18(a uint64) int {
+			return bits.TrailingZeros64(a)
+		}
+		`,
+		[]string{"\tFLOGR\t"},
+	},
+	{
+		`
+		func f19(a uint32) int {
+			return bits.TrailingZeros32(a)
+		}
+		`,
+		[]string{"\tFLOGR\t", "\tMOVWZ\t"},
+	},
+	{
+		`
+		func f20(a uint16) int {
+			return bits.TrailingZeros16(a)
+		}
+		`,
+		[]string{"\tFLOGR\t", "\tOR\t\\$65536,"},
+	},
+	{
+		`
+		func f21(a uint8) int {
+			return bits.TrailingZeros8(a)
+		}
+		`,
+		[]string{"\tFLOGR\t", "\tOR\t\\$256,"},
+	},
 }
 
 var linuxARMTests = []*asmTest{

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

@@ -2455,270 +2455,334 @@ const (
 	callGo
 )
 
-// TODO: make this a field of a configuration object instead of a global.
-var intrinsics *intrinsicInfo
-
-type intrinsicInfo struct {
-	std      map[intrinsicKey]intrinsicBuilder
-	intSized map[sizedIntrinsicKey]intrinsicBuilder
-	ptrSized map[sizedIntrinsicKey]intrinsicBuilder
-}
+var intrinsics map[intrinsicKey]intrinsicBuilder
 
 // An intrinsicBuilder converts a call node n into an ssa value that
 // implements that call as an intrinsic. args is a list of arguments to the func.
 type intrinsicBuilder func(s *state, n *Node, args []*ssa.Value) *ssa.Value
 
 type intrinsicKey struct {
-	pkg string
-	fn  string
-}
-
-type sizedIntrinsicKey struct {
+	arch *sys.Arch
 	pkg  string
 	fn   string
-	size int
 }
 
-// disableForInstrumenting returns nil when instrumenting, fn otherwise
-func disableForInstrumenting(fn intrinsicBuilder) intrinsicBuilder {
-	if instrumenting {
-		return nil
+func init() {
+	intrinsics = map[intrinsicKey]intrinsicBuilder{}
+
+	var all []*sys.Arch
+	var i4 []*sys.Arch
+	var i8 []*sys.Arch
+	var p4 []*sys.Arch
+	var p8 []*sys.Arch
+	for _, a := range sys.Archs {
+		all = append(all, a)
+		if a.IntSize == 4 {
+			i4 = append(i4, a)
+		} else {
+			i8 = append(i8, a)
+		}
+		if a.PtrSize == 4 {
+			p4 = append(p4, a)
+		} else {
+			p8 = append(p8, a)
+		}
 	}
-	return fn
-}
 
-// enableOnArch returns fn on given archs, nil otherwise
-func enableOnArch(fn intrinsicBuilder, archs ...sys.ArchFamily) intrinsicBuilder {
-	if Thearch.LinkArch.InFamily(archs...) {
-		return fn
+	// add adds the intrinsic b for pkg.fn for the given list of architectures.
+	add := func(pkg, fn string, b intrinsicBuilder, archs ...*sys.Arch) {
+		for _, a := range archs {
+			intrinsics[intrinsicKey{a, pkg, fn}] = b
+		}
+	}
+	// addF does the same as add but operates on architecture families.
+	addF := func(pkg, fn string, b intrinsicBuilder, archFamilies ...sys.ArchFamily) {
+		m := 0
+		for _, f := range archFamilies {
+			if f >= 32 {
+				panic("too many architecture families")
+			}
+			m |= 1 << uint(f)
+		}
+		for _, a := range all {
+			if m>>uint(a.Family)&1 != 0 {
+				intrinsics[intrinsicKey{a, pkg, fn}] = b
+			}
+		}
+	}
+	// alias defines pkg.fn = pkg2.fn2 for all architectures in archs for which pkg2.fn2 exists.
+	alias := func(pkg, fn, pkg2, fn2 string, archs ...*sys.Arch) {
+		for _, a := range archs {
+			if b, ok := intrinsics[intrinsicKey{a, pkg2, fn2}]; ok {
+				intrinsics[intrinsicKey{a, pkg, fn}] = b
+			}
+		}
 	}
-	return nil
-}
 
-func intrinsicInit() {
-	i := &intrinsicInfo{}
-	intrinsics = i
-
-	// initial set of intrinsics.
-	i.std = map[intrinsicKey]intrinsicBuilder{
-		/******** runtime ********/
-		intrinsicKey{"runtime", "slicebytetostringtmp"}: disableForInstrumenting(func(s *state, n *Node, args []*ssa.Value) *ssa.Value {
-			// Compiler frontend optimizations emit OARRAYBYTESTRTMP nodes
-			// for the backend instead of slicebytetostringtmp calls
-			// when not instrumenting.
-			slice := args[0]
-			ptr := s.newValue1(ssa.OpSlicePtr, ptrto(Types[TUINT8]), slice)
-			len := s.newValue1(ssa.OpSliceLen, Types[TINT], slice)
-			return s.newValue2(ssa.OpStringMake, n.Type, ptr, len)
-		}),
-		intrinsicKey{"runtime", "KeepAlive"}: func(s *state, n *Node, args []*ssa.Value) *ssa.Value {
+	/******** runtime ********/
+	if !instrumenting {
+		add("runtime", "slicebytetostringtmp",
+			func(s *state, n *Node, args []*ssa.Value) *ssa.Value {
+				// Compiler frontend optimizations emit OARRAYBYTESTRTMP nodes
+				// for the backend instead of slicebytetostringtmp calls
+				// when not instrumenting.
+				slice := args[0]
+				ptr := s.newValue1(ssa.OpSlicePtr, ptrto(Types[TUINT8]), slice)
+				len := s.newValue1(ssa.OpSliceLen, Types[TINT], slice)
+				return s.newValue2(ssa.OpStringMake, n.Type, ptr, len)
+			},
+			all...)
+	}
+	add("runtime", "KeepAlive",
+		func(s *state, n *Node, args []*ssa.Value) *ssa.Value {
 			data := s.newValue1(ssa.OpIData, ptrto(Types[TUINT8]), args[0])
 			s.vars[&memVar] = s.newValue2(ssa.OpKeepAlive, ssa.TypeMem, data, s.mem())
 			return nil
 		},
+		all...)
 
-		/******** runtime/internal/sys ********/
-		intrinsicKey{"runtime/internal/sys", "Ctz32"}: enableOnArch(func(s *state, n *Node, args []*ssa.Value) *ssa.Value {
-			return s.newValue1(ssa.OpCtz32, Types[TUINT32], args[0])
-		}, sys.AMD64, sys.ARM64, sys.ARM, sys.S390X, sys.MIPS),
-		intrinsicKey{"runtime/internal/sys", "Ctz64"}: enableOnArch(func(s *state, n *Node, args []*ssa.Value) *ssa.Value {
-			return s.newValue1(ssa.OpCtz64, Types[TUINT64], args[0])
-		}, sys.AMD64, sys.ARM64, sys.ARM, sys.S390X, sys.MIPS),
-		intrinsicKey{"runtime/internal/sys", "Bswap32"}: enableOnArch(func(s *state, n *Node, args []*ssa.Value) *ssa.Value {
+	/******** runtime/internal/sys ********/
+	addF("runtime/internal/sys", "Ctz32",
+		func(s *state, n *Node, args []*ssa.Value) *ssa.Value {
+			return s.newValue1(ssa.OpCtz32, Types[TINT], args[0])
+		},
+		sys.AMD64, sys.ARM64, sys.ARM, sys.S390X, sys.MIPS)
+	addF("runtime/internal/sys", "Ctz64",
+		func(s *state, n *Node, args []*ssa.Value) *ssa.Value {
+			return s.newValue1(ssa.OpCtz64, Types[TINT], args[0])
+		},
+		sys.AMD64, sys.ARM64, sys.ARM, sys.S390X, sys.MIPS)
+	addF("runtime/internal/sys", "Bswap32",
+		func(s *state, n *Node, args []*ssa.Value) *ssa.Value {
 			return s.newValue1(ssa.OpBswap32, Types[TUINT32], args[0])
-		}, sys.AMD64, sys.ARM64, sys.ARM, sys.S390X),
-		intrinsicKey{"runtime/internal/sys", "Bswap64"}: enableOnArch(func(s *state, n *Node, args []*ssa.Value) *ssa.Value {
+		},
+		sys.AMD64, sys.ARM64, sys.ARM, sys.S390X)
+	addF("runtime/internal/sys", "Bswap64",
+		func(s *state, n *Node, args []*ssa.Value) *ssa.Value {
 			return s.newValue1(ssa.OpBswap64, Types[TUINT64], args[0])
-		}, sys.AMD64, sys.ARM64, sys.ARM, sys.S390X),
+		},
+		sys.AMD64, sys.ARM64, sys.ARM, sys.S390X)
 
-		/******** runtime/internal/atomic ********/
-		intrinsicKey{"runtime/internal/atomic", "Load"}: enableOnArch(func(s *state, n *Node, args []*ssa.Value) *ssa.Value {
+	/******** runtime/internal/atomic ********/
+	addF("runtime/internal/atomic", "Load",
+		func(s *state, n *Node, args []*ssa.Value) *ssa.Value {
 			v := s.newValue2(ssa.OpAtomicLoad32, ssa.MakeTuple(Types[TUINT32], ssa.TypeMem), args[0], s.mem())
 			s.vars[&memVar] = s.newValue1(ssa.OpSelect1, ssa.TypeMem, v)
 			return s.newValue1(ssa.OpSelect0, Types[TUINT32], v)
-		}, sys.AMD64, sys.ARM64, sys.S390X, sys.MIPS, sys.PPC64),
-		intrinsicKey{"runtime/internal/atomic", "Load64"}: enableOnArch(func(s *state, n *Node, args []*ssa.Value) *ssa.Value {
+		},
+		sys.AMD64, sys.ARM64, sys.S390X, sys.MIPS, sys.PPC64)
+
+	addF("runtime/internal/atomic", "Load64",
+		func(s *state, n *Node, args []*ssa.Value) *ssa.Value {
 			v := s.newValue2(ssa.OpAtomicLoad64, ssa.MakeTuple(Types[TUINT64], ssa.TypeMem), args[0], s.mem())
 			s.vars[&memVar] = s.newValue1(ssa.OpSelect1, ssa.TypeMem, v)
 			return s.newValue1(ssa.OpSelect0, Types[TUINT64], v)
-		}, sys.AMD64, sys.ARM64, sys.S390X, sys.PPC64),
-		intrinsicKey{"runtime/internal/atomic", "Loadp"}: enableOnArch(func(s *state, n *Node, args []*ssa.Value) *ssa.Value {
+		},
+		sys.AMD64, sys.ARM64, sys.S390X, sys.PPC64)
+	addF("runtime/internal/atomic", "Loadp",
+		func(s *state, n *Node, args []*ssa.Value) *ssa.Value {
 			v := s.newValue2(ssa.OpAtomicLoadPtr, ssa.MakeTuple(ptrto(Types[TUINT8]), ssa.TypeMem), args[0], s.mem())
 			s.vars[&memVar] = s.newValue1(ssa.OpSelect1, ssa.TypeMem, v)
 			return s.newValue1(ssa.OpSelect0, ptrto(Types[TUINT8]), v)
-		}, sys.AMD64, sys.ARM64, sys.S390X, sys.MIPS, sys.PPC64),
+		},
+		sys.AMD64, sys.ARM64, sys.S390X, sys.MIPS, sys.PPC64)
 
-		intrinsicKey{"runtime/internal/atomic", "Store"}: enableOnArch(func(s *state, n *Node, args []*ssa.Value) *ssa.Value {
+	addF("runtime/internal/atomic", "Store",
+		func(s *state, n *Node, args []*ssa.Value) *ssa.Value {
 			s.vars[&memVar] = s.newValue3(ssa.OpAtomicStore32, ssa.TypeMem, args[0], args[1], s.mem())
 			return nil
-		}, sys.AMD64, sys.ARM64, sys.S390X, sys.MIPS, sys.PPC64),
-		intrinsicKey{"runtime/internal/atomic", "Store64"}: enableOnArch(func(s *state, n *Node, args []*ssa.Value) *ssa.Value {
+		},
+		sys.AMD64, sys.ARM64, sys.S390X, sys.MIPS, sys.PPC64)
+	addF("runtime/internal/atomic", "Store64",
+		func(s *state, n *Node, args []*ssa.Value) *ssa.Value {
 			s.vars[&memVar] = s.newValue3(ssa.OpAtomicStore64, ssa.TypeMem, args[0], args[1], s.mem())
 			return nil
-		}, sys.AMD64, sys.ARM64, sys.S390X, sys.PPC64),
-		intrinsicKey{"runtime/internal/atomic", "StorepNoWB"}: enableOnArch(func(s *state, n *Node, args []*ssa.Value) *ssa.Value {
+		},
+		sys.AMD64, sys.ARM64, sys.S390X, sys.PPC64)
+	addF("runtime/internal/atomic", "StorepNoWB",
+		func(s *state, n *Node, args []*ssa.Value) *ssa.Value {
 			s.vars[&memVar] = s.newValue3(ssa.OpAtomicStorePtrNoWB, ssa.TypeMem, args[0], args[1], s.mem())
 			return nil
-		}, sys.AMD64, sys.ARM64, sys.S390X, sys.MIPS),
+		},
+		sys.AMD64, sys.ARM64, sys.S390X, sys.MIPS)
 
-		intrinsicKey{"runtime/internal/atomic", "Xchg"}: enableOnArch(func(s *state, n *Node, args []*ssa.Value) *ssa.Value {
+	addF("runtime/internal/atomic", "Xchg",
+		func(s *state, n *Node, args []*ssa.Value) *ssa.Value {
 			v := s.newValue3(ssa.OpAtomicExchange32, ssa.MakeTuple(Types[TUINT32], ssa.TypeMem), args[0], args[1], s.mem())
 			s.vars[&memVar] = s.newValue1(ssa.OpSelect1, ssa.TypeMem, v)
 			return s.newValue1(ssa.OpSelect0, Types[TUINT32], v)
-		}, sys.AMD64, sys.ARM64, sys.S390X, sys.MIPS, sys.PPC64),
-		intrinsicKey{"runtime/internal/atomic", "Xchg64"}: enableOnArch(func(s *state, n *Node, args []*ssa.Value) *ssa.Value {
+		},
+		sys.AMD64, sys.ARM64, sys.S390X, sys.MIPS, sys.PPC64)
+	addF("runtime/internal/atomic", "Xchg64",
+		func(s *state, n *Node, args []*ssa.Value) *ssa.Value {
 			v := s.newValue3(ssa.OpAtomicExchange64, ssa.MakeTuple(Types[TUINT64], ssa.TypeMem), args[0], args[1], s.mem())
 			s.vars[&memVar] = s.newValue1(ssa.OpSelect1, ssa.TypeMem, v)
 			return s.newValue1(ssa.OpSelect0, Types[TUINT64], v)
-		}, sys.AMD64, sys.ARM64, sys.S390X, sys.PPC64),
+		},
+		sys.AMD64, sys.ARM64, sys.S390X, sys.PPC64)
 
-		intrinsicKey{"runtime/internal/atomic", "Xadd"}: enableOnArch(func(s *state, n *Node, args []*ssa.Value) *ssa.Value {
+	addF("runtime/internal/atomic", "Xadd",
+		func(s *state, n *Node, args []*ssa.Value) *ssa.Value {
 			v := s.newValue3(ssa.OpAtomicAdd32, ssa.MakeTuple(Types[TUINT32], ssa.TypeMem), args[0], args[1], s.mem())
 			s.vars[&memVar] = s.newValue1(ssa.OpSelect1, ssa.TypeMem, v)
 			return s.newValue1(ssa.OpSelect0, Types[TUINT32], v)
-		}, sys.AMD64, sys.ARM64, sys.S390X, sys.MIPS, sys.PPC64),
-		intrinsicKey{"runtime/internal/atomic", "Xadd64"}: enableOnArch(func(s *state, n *Node, args []*ssa.Value) *ssa.Value {
+		},
+		sys.AMD64, sys.ARM64, sys.S390X, sys.MIPS, sys.PPC64)
+	addF("runtime/internal/atomic", "Xadd64",
+		func(s *state, n *Node, args []*ssa.Value) *ssa.Value {
 			v := s.newValue3(ssa.OpAtomicAdd64, ssa.MakeTuple(Types[TUINT64], ssa.TypeMem), args[0], args[1], s.mem())
 			s.vars[&memVar] = s.newValue1(ssa.OpSelect1, ssa.TypeMem, v)
 			return s.newValue1(ssa.OpSelect0, Types[TUINT64], v)
-		}, sys.AMD64, sys.ARM64, sys.S390X, sys.PPC64),
+		},
+		sys.AMD64, sys.ARM64, sys.S390X, sys.PPC64)
 
-		intrinsicKey{"runtime/internal/atomic", "Cas"}: enableOnArch(func(s *state, n *Node, args []*ssa.Value) *ssa.Value {
+	addF("runtime/internal/atomic", "Cas",
+		func(s *state, n *Node, args []*ssa.Value) *ssa.Value {
 			v := s.newValue4(ssa.OpAtomicCompareAndSwap32, ssa.MakeTuple(Types[TBOOL], ssa.TypeMem), args[0], args[1], args[2], s.mem())
 			s.vars[&memVar] = s.newValue1(ssa.OpSelect1, ssa.TypeMem, v)
 			return s.newValue1(ssa.OpSelect0, Types[TBOOL], v)
-		}, sys.AMD64, sys.ARM64, sys.S390X, sys.MIPS, sys.PPC64),
-		intrinsicKey{"runtime/internal/atomic", "Cas64"}: enableOnArch(func(s *state, n *Node, args []*ssa.Value) *ssa.Value {
+		},
+		sys.AMD64, sys.ARM64, sys.S390X, sys.MIPS, sys.PPC64)
+	addF("runtime/internal/atomic", "Cas64",
+		func(s *state, n *Node, args []*ssa.Value) *ssa.Value {
 			v := s.newValue4(ssa.OpAtomicCompareAndSwap64, ssa.MakeTuple(Types[TBOOL], ssa.TypeMem), args[0], args[1], args[2], s.mem())
 			s.vars[&memVar] = s.newValue1(ssa.OpSelect1, ssa.TypeMem, v)
 			return s.newValue1(ssa.OpSelect0, Types[TBOOL], v)
-		}, sys.AMD64, sys.ARM64, sys.S390X, sys.PPC64),
+		},
+		sys.AMD64, sys.ARM64, sys.S390X, sys.PPC64)
 
-		intrinsicKey{"runtime/internal/atomic", "And8"}: enableOnArch(func(s *state, n *Node, args []*ssa.Value) *ssa.Value {
+	addF("runtime/internal/atomic", "And8",
+		func(s *state, n *Node, args []*ssa.Value) *ssa.Value {
 			s.vars[&memVar] = s.newValue3(ssa.OpAtomicAnd8, ssa.TypeMem, args[0], args[1], s.mem())
 			return nil
-		}, sys.AMD64, sys.ARM64, sys.MIPS, sys.PPC64),
-		intrinsicKey{"runtime/internal/atomic", "Or8"}: enableOnArch(func(s *state, n *Node, args []*ssa.Value) *ssa.Value {
+		},
+		sys.AMD64, sys.ARM64, sys.MIPS, sys.PPC64)
+	addF("runtime/internal/atomic", "Or8",
+		func(s *state, n *Node, args []*ssa.Value) *ssa.Value {
 			s.vars[&memVar] = s.newValue3(ssa.OpAtomicOr8, ssa.TypeMem, args[0], args[1], s.mem())
 			return nil
-		}, sys.AMD64, sys.ARM64, sys.MIPS, sys.PPC64),
-
-		/******** math ********/
-		intrinsicKey{"math", "Sqrt"}: enableOnArch(func(s *state, n *Node, args []*ssa.Value) *ssa.Value {
+		},
+		sys.AMD64, sys.ARM64, sys.MIPS, sys.PPC64)
+
+	alias("runtime/internal/atomic", "Loadint64", "runtime/internal/atomic", "Load64", all...)
+	alias("runtime/internal/atomic", "Xaddint64", "runtime/internal/atomic", "Xadd64", all...)
+	alias("runtime/internal/atomic", "Loaduint", "runtime/internal/atomic", "Load", i4...)
+	alias("runtime/internal/atomic", "Loaduint", "runtime/internal/atomic", "Load64", i8...)
+	alias("runtime/internal/atomic", "Loaduintptr", "runtime/internal/atomic", "Load", p4...)
+	alias("runtime/internal/atomic", "Loaduintptr", "runtime/internal/atomic", "Load64", p8...)
+	alias("runtime/internal/atomic", "Storeuintptr", "runtime/internal/atomic", "Store", p4...)
+	alias("runtime/internal/atomic", "Storeuintptr", "runtime/internal/atomic", "Store64", p8...)
+	alias("runtime/internal/atomic", "Xchguintptr", "runtime/internal/atomic", "Xchg", p4...)
+	alias("runtime/internal/atomic", "Xchguintptr", "runtime/internal/atomic", "Xchg64", p8...)
+	alias("runtime/internal/atomic", "Xadduintptr", "runtime/internal/atomic", "Xadd", p4...)
+	alias("runtime/internal/atomic", "Xadduintptr", "runtime/internal/atomic", "Xadd64", p8...)
+	alias("runtime/internal/atomic", "Casuintptr", "runtime/internal/atomic", "Cas", p4...)
+	alias("runtime/internal/atomic", "Casuintptr", "runtime/internal/atomic", "Cas64", p8...)
+	alias("runtime/internal/atomic", "Casp1", "runtime/internal/atomic", "Cas", p4...)
+	alias("runtime/internal/atomic", "Casp1", "runtime/internal/atomic", "Cas64", p8...)
+
+	/******** math ********/
+	addF("math", "Sqrt",
+		func(s *state, n *Node, args []*ssa.Value) *ssa.Value {
 			return s.newValue1(ssa.OpSqrt, Types[TFLOAT64], args[0])
-		}, sys.AMD64, sys.ARM, sys.ARM64, sys.MIPS, sys.PPC64, sys.S390X),
-	}
-
-	// aliases internal to runtime/internal/atomic
-	i.std[intrinsicKey{"runtime/internal/atomic", "Loadint64"}] =
-		i.std[intrinsicKey{"runtime/internal/atomic", "Load64"}]
-	i.std[intrinsicKey{"runtime/internal/atomic", "Xaddint64"}] =
-		i.std[intrinsicKey{"runtime/internal/atomic", "Xadd64"}]
-
-	// intrinsics which vary depending on the size of int/ptr.
-	i.intSized = map[sizedIntrinsicKey]intrinsicBuilder{
-		sizedIntrinsicKey{"runtime/internal/atomic", "Loaduint", 4}: i.std[intrinsicKey{"runtime/internal/atomic", "Load"}],
-		sizedIntrinsicKey{"runtime/internal/atomic", "Loaduint", 8}: i.std[intrinsicKey{"runtime/internal/atomic", "Load64"}],
-	}
-	i.ptrSized = map[sizedIntrinsicKey]intrinsicBuilder{
-		sizedIntrinsicKey{"runtime/internal/atomic", "Loaduintptr", 4}:  i.std[intrinsicKey{"runtime/internal/atomic", "Load"}],
-		sizedIntrinsicKey{"runtime/internal/atomic", "Loaduintptr", 8}:  i.std[intrinsicKey{"runtime/internal/atomic", "Load64"}],
-		sizedIntrinsicKey{"runtime/internal/atomic", "Storeuintptr", 4}: i.std[intrinsicKey{"runtime/internal/atomic", "Store"}],
-		sizedIntrinsicKey{"runtime/internal/atomic", "Storeuintptr", 8}: i.std[intrinsicKey{"runtime/internal/atomic", "Store64"}],
-		sizedIntrinsicKey{"runtime/internal/atomic", "Xchguintptr", 4}:  i.std[intrinsicKey{"runtime/internal/atomic", "Xchg"}],
-		sizedIntrinsicKey{"runtime/internal/atomic", "Xchguintptr", 8}:  i.std[intrinsicKey{"runtime/internal/atomic", "Xchg64"}],
-		sizedIntrinsicKey{"runtime/internal/atomic", "Xadduintptr", 4}:  i.std[intrinsicKey{"runtime/internal/atomic", "Xadd"}],
-		sizedIntrinsicKey{"runtime/internal/atomic", "Xadduintptr", 8}:  i.std[intrinsicKey{"runtime/internal/atomic", "Xadd64"}],
-		sizedIntrinsicKey{"runtime/internal/atomic", "Casuintptr", 4}:   i.std[intrinsicKey{"runtime/internal/atomic", "Cas"}],
-		sizedIntrinsicKey{"runtime/internal/atomic", "Casuintptr", 8}:   i.std[intrinsicKey{"runtime/internal/atomic", "Cas64"}],
-		sizedIntrinsicKey{"runtime/internal/atomic", "Casp1", 4}:        i.std[intrinsicKey{"runtime/internal/atomic", "Cas"}],
-		sizedIntrinsicKey{"runtime/internal/atomic", "Casp1", 8}:        i.std[intrinsicKey{"runtime/internal/atomic", "Cas64"}],
-	}
+		},
+		sys.AMD64, sys.ARM, sys.ARM64, sys.MIPS, sys.PPC64, sys.S390X)
+
+	/******** math/bits ********/
+	addF("math/bits", "TrailingZeros64",
+		func(s *state, n *Node, args []*ssa.Value) *ssa.Value {
+			return s.newValue1(ssa.OpCtz64, Types[TINT], args[0])
+		},
+		sys.AMD64, sys.ARM64, sys.ARM, sys.S390X, sys.MIPS)
+	addF("math/bits", "TrailingZeros32",
+		func(s *state, n *Node, args []*ssa.Value) *ssa.Value {
+			return s.newValue1(ssa.OpCtz32, Types[TINT], args[0])
+		},
+		sys.AMD64, sys.ARM64, sys.ARM, sys.S390X, sys.MIPS)
+	addF("math/bits", "TrailingZeros16",
+		func(s *state, n *Node, args []*ssa.Value) *ssa.Value {
+			x := s.newValue1(ssa.OpZeroExt16to32, Types[TUINT32], args[0])
+			c := s.constInt32(Types[TUINT32], 1<<16)
+			y := s.newValue2(ssa.OpOr32, Types[TUINT32], x, c)
+			return s.newValue1(ssa.OpCtz32, Types[TINT], y)
+		},
+		sys.ARM, sys.MIPS)
+	addF("math/bits", "TrailingZeros16",
+		func(s *state, n *Node, args []*ssa.Value) *ssa.Value {
+			x := s.newValue1(ssa.OpZeroExt16to64, Types[TUINT64], args[0])
+			c := s.constInt64(Types[TUINT64], 1<<16)
+			y := s.newValue2(ssa.OpOr64, Types[TUINT64], x, c)
+			return s.newValue1(ssa.OpCtz64, Types[TINT], y)
+		},
+		sys.AMD64, sys.ARM64, sys.S390X)
+	addF("math/bits", "TrailingZeros8",
+		func(s *state, n *Node, args []*ssa.Value) *ssa.Value {
+			x := s.newValue1(ssa.OpZeroExt8to32, Types[TUINT32], args[0])
+			c := s.constInt32(Types[TUINT32], 1<<8)
+			y := s.newValue2(ssa.OpOr32, Types[TUINT32], x, c)
+			return s.newValue1(ssa.OpCtz32, Types[TINT], y)
+		},
+		sys.ARM, sys.MIPS)
+	addF("math/bits", "TrailingZeros8",
+		func(s *state, n *Node, args []*ssa.Value) *ssa.Value {
+			x := s.newValue1(ssa.OpZeroExt8to64, Types[TUINT64], args[0])
+			c := s.constInt64(Types[TUINT64], 1<<8)
+			y := s.newValue2(ssa.OpOr64, Types[TUINT64], x, c)
+			return s.newValue1(ssa.OpCtz64, Types[TINT], y)
+		},
+		sys.AMD64, sys.ARM64, sys.S390X)
 
 	/******** sync/atomic ********/
-	if flag_race {
-		// The race detector needs to be able to intercept these calls.
-		// We can't intrinsify them.
-		return
-	}
-	// these are all aliases to runtime/internal/atomic implementations.
-	i.std[intrinsicKey{"sync/atomic", "LoadInt32"}] =
-		i.std[intrinsicKey{"runtime/internal/atomic", "Load"}]
-	i.std[intrinsicKey{"sync/atomic", "LoadInt64"}] =
-		i.std[intrinsicKey{"runtime/internal/atomic", "Load64"}]
-	i.std[intrinsicKey{"sync/atomic", "LoadPointer"}] =
-		i.std[intrinsicKey{"runtime/internal/atomic", "Loadp"}]
-	i.std[intrinsicKey{"sync/atomic", "LoadUint32"}] =
-		i.std[intrinsicKey{"runtime/internal/atomic", "Load"}]
-	i.std[intrinsicKey{"sync/atomic", "LoadUint64"}] =
-		i.std[intrinsicKey{"runtime/internal/atomic", "Load64"}]
-	i.ptrSized[sizedIntrinsicKey{"sync/atomic", "LoadUintptr", 4}] =
-		i.std[intrinsicKey{"runtime/internal/atomic", "Load"}]
-	i.ptrSized[sizedIntrinsicKey{"sync/atomic", "LoadUintptr", 8}] =
-		i.std[intrinsicKey{"runtime/internal/atomic", "Load64"}]
-
-	i.std[intrinsicKey{"sync/atomic", "StoreInt32"}] =
-		i.std[intrinsicKey{"runtime/internal/atomic", "Store"}]
-	i.std[intrinsicKey{"sync/atomic", "StoreInt64"}] =
-		i.std[intrinsicKey{"runtime/internal/atomic", "Store64"}]
+
+	// Note: these are disabled by flag_race in findIntrinsic below.
+	alias("sync/atomic", "LoadInt32", "runtime/internal/atomic", "Load", all...)
+	alias("sync/atomic", "LoadInt64", "runtime/internal/atomic", "Load64", all...)
+	alias("sync/atomic", "LoadPointer", "runtime/internal/atomic", "Loadp", all...)
+	alias("sync/atomic", "LoadUint32", "runtime/internal/atomic", "Load", all...)
+	alias("sync/atomic", "LoadUint64", "runtime/internal/atomic", "Load64", all...)
+	alias("sync/atomic", "LoadUintptr", "runtime/internal/atomic", "Load", p4...)
+	alias("sync/atomic", "LoadUintptr", "runtime/internal/atomic", "Load64", p8...)
+
+	alias("sync/atomic", "StoreInt32", "runtime/internal/atomic", "Store", all...)
+	alias("sync/atomic", "StoreInt64", "runtime/internal/atomic", "Store64", all...)
 	// Note: not StorePointer, that needs a write barrier.  Same below for {CompareAnd}Swap.
-	i.std[intrinsicKey{"sync/atomic", "StoreUint32"}] =
-		i.std[intrinsicKey{"runtime/internal/atomic", "Store"}]
-	i.std[intrinsicKey{"sync/atomic", "StoreUint64"}] =
-		i.std[intrinsicKey{"runtime/internal/atomic", "Store64"}]
-	i.ptrSized[sizedIntrinsicKey{"sync/atomic", "StoreUintptr", 4}] =
-		i.std[intrinsicKey{"runtime/internal/atomic", "Store"}]
-	i.ptrSized[sizedIntrinsicKey{"sync/atomic", "StoreUintptr", 8}] =
-		i.std[intrinsicKey{"runtime/internal/atomic", "Store64"}]
-
-	i.std[intrinsicKey{"sync/atomic", "SwapInt32"}] =
-		i.std[intrinsicKey{"runtime/internal/atomic", "Xchg"}]
-	i.std[intrinsicKey{"sync/atomic", "SwapInt64"}] =
-		i.std[intrinsicKey{"runtime/internal/atomic", "Xchg64"}]
-	i.std[intrinsicKey{"sync/atomic", "SwapUint32"}] =
-		i.std[intrinsicKey{"runtime/internal/atomic", "Xchg"}]
-	i.std[intrinsicKey{"sync/atomic", "SwapUint64"}] =
-		i.std[intrinsicKey{"runtime/internal/atomic", "Xchg64"}]
-	i.ptrSized[sizedIntrinsicKey{"sync/atomic", "SwapUintptr", 4}] =
-		i.std[intrinsicKey{"runtime/internal/atomic", "Xchg"}]
-	i.ptrSized[sizedIntrinsicKey{"sync/atomic", "SwapUintptr", 8}] =
-		i.std[intrinsicKey{"runtime/internal/atomic", "Xchg64"}]
-
-	i.std[intrinsicKey{"sync/atomic", "CompareAndSwapInt32"}] =
-		i.std[intrinsicKey{"runtime/internal/atomic", "Cas"}]
-	i.std[intrinsicKey{"sync/atomic", "CompareAndSwapInt64"}] =
-		i.std[intrinsicKey{"runtime/internal/atomic", "Cas64"}]
-	i.std[intrinsicKey{"sync/atomic", "CompareAndSwapUint32"}] =
-		i.std[intrinsicKey{"runtime/internal/atomic", "Cas"}]
-	i.std[intrinsicKey{"sync/atomic", "CompareAndSwapUint64"}] =
-		i.std[intrinsicKey{"runtime/internal/atomic", "Cas64"}]
-	i.ptrSized[sizedIntrinsicKey{"sync/atomic", "CompareAndSwapUintptr", 4}] =
-		i.std[intrinsicKey{"runtime/internal/atomic", "Cas"}]
-	i.ptrSized[sizedIntrinsicKey{"sync/atomic", "CompareAndSwapUintptr", 8}] =
-		i.std[intrinsicKey{"runtime/internal/atomic", "Cas64"}]
-
-	i.std[intrinsicKey{"sync/atomic", "AddInt32"}] =
-		i.std[intrinsicKey{"runtime/internal/atomic", "Xadd"}]
-	i.std[intrinsicKey{"sync/atomic", "AddInt64"}] =
-		i.std[intrinsicKey{"runtime/internal/atomic", "Xadd64"}]
-	i.std[intrinsicKey{"sync/atomic", "AddUint32"}] =
-		i.std[intrinsicKey{"runtime/internal/atomic", "Xadd"}]
-	i.std[intrinsicKey{"sync/atomic", "AddUint64"}] =
-		i.std[intrinsicKey{"runtime/internal/atomic", "Xadd64"}]
-	i.ptrSized[sizedIntrinsicKey{"sync/atomic", "AddUintptr", 4}] =
-		i.std[intrinsicKey{"runtime/internal/atomic", "Xadd"}]
-	i.ptrSized[sizedIntrinsicKey{"sync/atomic", "AddUintptr", 8}] =
-		i.std[intrinsicKey{"runtime/internal/atomic", "Xadd64"}]
+	alias("sync/atomic", "StoreUint32", "runtime/internal/atomic", "Store", all...)
+	alias("sync/atomic", "StoreUint64", "runtime/internal/atomic", "Store64", all...)
+	alias("sync/atomic", "StoreUintptr", "runtime/internal/atomic", "Store", p4...)
+	alias("sync/atomic", "StoreUintptr", "runtime/internal/atomic", "Store64", p8...)
+
+	alias("sync/atomic", "SwapInt32", "runtime/internal/atomic", "Xchg", all...)
+	alias("sync/atomic", "SwapInt64", "runtime/internal/atomic", "Xchg64", all...)
+	alias("sync/atomic", "SwapUint32", "runtime/internal/atomic", "Xchg", all...)
+	alias("sync/atomic", "SwapUint64", "runtime/internal/atomic", "Xchg64", all...)
+	alias("sync/atomic", "SwapUintptr", "runtime/internal/atomic", "Xchg", p4...)
+	alias("sync/atomic", "SwapUintptr", "runtime/internal/atomic", "Xchg64", p8...)
+
+	alias("sync/atomic", "CompareAndSwapInt32", "runtime/internal/atomic", "Cas", all...)
+	alias("sync/atomic", "CompareAndSwapInt64", "runtime/internal/atomic", "Cas64", all...)
+	alias("sync/atomic", "CompareAndSwapUint32", "runtime/internal/atomic", "Cas", all...)
+	alias("sync/atomic", "CompareAndSwapUint64", "runtime/internal/atomic", "Cas64", all...)
+	alias("sync/atomic", "CompareAndSwapUintptr", "runtime/internal/atomic", "Cas", p4...)
+	alias("sync/atomic", "CompareAndSwapUintptr", "runtime/internal/atomic", "Cas64", p8...)
+
+	alias("sync/atomic", "AddInt32", "runtime/internal/atomic", "Xadd", all...)
+	alias("sync/atomic", "AddInt64", "runtime/internal/atomic", "Xadd64", all...)
+	alias("sync/atomic", "AddUint32", "runtime/internal/atomic", "Xadd", all...)
+	alias("sync/atomic", "AddUint64", "runtime/internal/atomic", "Xadd64", all...)
+	alias("sync/atomic", "AddUintptr", "runtime/internal/atomic", "Xadd", p4...)
+	alias("sync/atomic", "AddUintptr", "runtime/internal/atomic", "Xadd64", p8...)
 
 	/******** math/big ********/
-	i.intSized[sizedIntrinsicKey{"math/big", "mulWW", 8}] =
-		enableOnArch(func(s *state, n *Node, args []*ssa.Value) *ssa.Value {
+	add("math/big", "mulWW",
+		func(s *state, n *Node, args []*ssa.Value) *ssa.Value {
 			return s.newValue2(ssa.OpMul64uhilo, ssa.MakeTuple(Types[TUINT64], Types[TUINT64]), args[0], args[1])
-		}, sys.AMD64)
-	i.intSized[sizedIntrinsicKey{"math/big", "divWW", 8}] =
-		enableOnArch(func(s *state, n *Node, args []*ssa.Value) *ssa.Value {
+		},
+		sys.ArchAMD64)
+	add("math/big", "divWW",
+		func(s *state, n *Node, args []*ssa.Value) *ssa.Value {
 			return s.newValue3(ssa.OpDiv128u, ssa.MakeTuple(Types[TUINT64], Types[TUINT64]), args[0], args[1], args[2])
-		}, sys.AMD64)
+		},
+		sys.ArchAMD64)
 }
 
 // findIntrinsic returns a function which builds the SSA equivalent of the
@@ -2730,23 +2794,17 @@ func findIntrinsic(sym *Sym) intrinsicBuilder {
 	if sym == nil || sym.Pkg == nil {
 		return nil
 	}
-	if intrinsics == nil {
-		intrinsicInit()
-	}
 	pkg := sym.Pkg.Path
 	if sym.Pkg == localpkg {
 		pkg = myimportpath
 	}
-	fn := sym.Name
-	f := intrinsics.std[intrinsicKey{pkg, fn}]
-	if f != nil {
-		return f
-	}
-	f = intrinsics.intSized[sizedIntrinsicKey{pkg, fn, Widthint}]
-	if f != nil {
-		return f
+	if flag_race && pkg == "sync/atomic" {
+		// The race detector needs to be able to intercept these calls.
+		// We can't intrinsify them.
+		return nil
 	}
-	return intrinsics.ptrSized[sizedIntrinsicKey{pkg, fn, Widthptr}]
+	fn := sym.Name
+	return intrinsics[intrinsicKey{Thearch.LinkArch.Arch, pkg, fn}]
 }
 
 func isIntrinsicCall(n *Node) bool {

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

@@ -98,7 +98,7 @@
 
 // Lowering other arithmetic
 (Ctz64 <t> x) -> (CMOVQEQ (Select0 <t> (BSFQ x)) (MOVQconst <t> [64]) (Select1 <TypeFlags> (BSFQ x)))
-(Ctz32 <t> x) -> (CMOVLEQ (Select0 <t> (BSFL x)) (MOVLconst <t> [32]) (Select1 <TypeFlags> (BSFL x)))
+(Ctz32 x) -> (Select0 (BSFQ (ORQ <config.Frontend().TypeUInt64()> (MOVQconst [1<<32]) x)))
 
 (Bswap64 x) -> (BSWAPQ x)
 (Bswap32 x) -> (BSWAPL x)
@@ -2083,3 +2083,9 @@
 	(CMPXCHGQlock [off1+off2] {sym} ptr old new_ mem)
 (CMPXCHGLlock [off1] {sym} (ADDQconst [off2] ptr) old new_ mem) && is32Bit(off1+off2) ->
 	(CMPXCHGLlock [off1+off2] {sym} ptr old new_ mem)
+
+// We don't need the conditional move if we know the arg of BSF is not zero.
+(CMOVQEQ x _ (Select1 (BSFQ (ORQconst [c] _)))) && c != 0 -> x
+// Extension is unnecessary for trailing zeros.
+(BSFQ (ORQconst <t> [1<<8] (MOVBQZX x))) -> (BSFQ (ORQconst <t> [1<<8] x))
+(BSFQ (ORQconst <t> [1<<16] (MOVWQZX x))) -> (BSFQ (ORQconst <t> [1<<16] x))

src/cmd/compile/internal/ssa/gen/dec64.rules

@@ -108,13 +108,11 @@
 		(Com32 <config.fe.TypeUInt32()> (Int64Lo x)))
 
 (Ctz64 x) ->
-	(Int64Make
-		(Const32 <config.fe.TypeUInt32()> [0])
-		(Add32 <config.fe.TypeUInt32()>
-			(Ctz32 <config.fe.TypeUInt32()> (Int64Lo x))
-			(And32 <config.fe.TypeUInt32()>
-				(Com32 <config.fe.TypeUInt32()> (Zeromask (Int64Lo x)))
-				(Ctz32 <config.fe.TypeUInt32()> (Int64Hi x)))))
+	(Add32 <config.fe.TypeUInt32()>
+		(Ctz32 <config.fe.TypeUInt32()> (Int64Lo x))
+		(And32 <config.fe.TypeUInt32()>
+			(Com32 <config.fe.TypeUInt32()> (Zeromask (Int64Lo x)))
+			(Ctz32 <config.fe.TypeUInt32()> (Int64Hi x))))
 
 (Bswap64 x) ->
 	(Int64Make

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

@@ -236,7 +236,7 @@ var genericOps = []opData{
 	{name: "Com32", argLength: 1},
 	{name: "Com64", argLength: 1},
 
-	{name: "Ctz32", argLength: 1}, // Count trailing (low  order) zeroes (returns 0-32)
+	{name: "Ctz32", argLength: 1}, // Count trailing (low order) zeroes (returns 0-32)
 	{name: "Ctz64", argLength: 1}, // Count trailing zeroes (returns 0-64)
 
 	{name: "Bswap32", argLength: 1}, // Swap bytes

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

@@ -28,8 +28,12 @@ func rewriteValueAMD64(v *Value, config *Config) bool {
 		return rewriteValueAMD64_OpAMD64ANDQ(v, config)
 	case OpAMD64ANDQconst:
 		return rewriteValueAMD64_OpAMD64ANDQconst(v, config)
+	case OpAMD64BSFQ:
+		return rewriteValueAMD64_OpAMD64BSFQ(v, config)
 	case OpAMD64BTQconst:
 		return rewriteValueAMD64_OpAMD64BTQconst(v, config)
+	case OpAMD64CMOVQEQ:
+		return rewriteValueAMD64_OpAMD64CMOVQEQ(v, config)
 	case OpAMD64CMPB:
 		return rewriteValueAMD64_OpAMD64CMPB(v, config)
 	case OpAMD64CMPBconst:
@@ -2158,6 +2162,59 @@ func rewriteValueAMD64_OpAMD64ANDQconst(v *Value, config *Config) bool {
 	}
 	return false
 }
+func rewriteValueAMD64_OpAMD64BSFQ(v *Value, config *Config) bool {
+	b := v.Block
+	_ = b
+	// match: (BSFQ (ORQconst <t> [1<<8] (MOVBQZX x)))
+	// cond:
+	// result: (BSFQ (ORQconst <t> [1<<8] x))
+	for {
+		v_0 := v.Args[0]
+		if v_0.Op != OpAMD64ORQconst {
+			break
+		}
+		t := v_0.Type
+		if v_0.AuxInt != 1<<8 {
+			break
+		}
+		v_0_0 := v_0.Args[0]
+		if v_0_0.Op != OpAMD64MOVBQZX {
+			break
+		}
+		x := v_0_0.Args[0]
+		v.reset(OpAMD64BSFQ)
+		v0 := b.NewValue0(v.Pos, OpAMD64ORQconst, t)
+		v0.AuxInt = 1 << 8
+		v0.AddArg(x)
+		v.AddArg(v0)
+		return true
+	}
+	// match: (BSFQ (ORQconst <t> [1<<16] (MOVWQZX x)))
+	// cond:
+	// result: (BSFQ (ORQconst <t> [1<<16] x))
+	for {
+		v_0 := v.Args[0]
+		if v_0.Op != OpAMD64ORQconst {
+			break
+		}
+		t := v_0.Type
+		if v_0.AuxInt != 1<<16 {
+			break
+		}
+		v_0_0 := v_0.Args[0]
+		if v_0_0.Op != OpAMD64MOVWQZX {
+			break
+		}
+		x := v_0_0.Args[0]
+		v.reset(OpAMD64BSFQ)
+		v0 := b.NewValue0(v.Pos, OpAMD64ORQconst, t)
+		v0.AuxInt = 1 << 16
+		v0.AddArg(x)
+		v.AddArg(v0)
+		return true
+	}
+	return false
+}
 func rewriteValueAMD64_OpAMD64BTQconst(v *Value, config *Config) bool {
 	b := v.Block
 	_ = b
@@ -2177,6 +2234,37 @@ func rewriteValueAMD64_OpAMD64BTQconst(v *Value, config *Config) bool {
 	}
 	return false
 }
+func rewriteValueAMD64_OpAMD64CMOVQEQ(v *Value, config *Config) bool {
+	b := v.Block
+	_ = b
+	// match: (CMOVQEQ x _ (Select1 (BSFQ (ORQconst [c] _))))
+	// cond: c != 0
+	// result: x
+	for {
+		x := v.Args[0]
+		v_2 := v.Args[2]
+		if v_2.Op != OpSelect1 {
+			break
+		}
+		v_2_0 := v_2.Args[0]
+		if v_2_0.Op != OpAMD64BSFQ {
+			break
+		}
+		v_2_0_0 := v_2_0.Args[0]
+		if v_2_0_0.Op != OpAMD64ORQconst {
+			break
+		}
+		c := v_2_0_0.AuxInt
+		if !(c != 0) {
+			break
+		}
+		v.reset(OpCopy)
+		v.Type = x.Type
+		v.AddArg(x)
+		return true
+	}
+	return false
+}
 func rewriteValueAMD64_OpAMD64CMPB(v *Value, config *Config) bool {
 	b := v.Block
 	_ = b
@@ -17902,26 +17990,20 @@ func rewriteValueAMD64_OpConvert(v *Value, config *Config) bool {
 func rewriteValueAMD64_OpCtz32(v *Value, config *Config) bool {
 	b := v.Block
 	_ = b
-	// match: (Ctz32 <t> x)
+	// match: (Ctz32 x)
 	// cond:
-	// result: (CMOVLEQ (Select0 <t> (BSFL x)) (MOVLconst <t> [32]) (Select1 <TypeFlags> (BSFL x)))
+	// result: (Select0 (BSFQ (ORQ <config.Frontend().TypeUInt64()> (MOVQconst [1<<32]) x)))
 	for {
-		t := v.Type
 		x := v.Args[0]
-		v.reset(OpAMD64CMOVLEQ)
-		v0 := b.NewValue0(v.Pos, OpSelect0, t)
-		v1 := b.NewValue0(v.Pos, OpAMD64BSFL, MakeTuple(config.fe.TypeUInt32(), TypeFlags))
+		v.reset(OpSelect0)
+		v0 := b.NewValue0(v.Pos, OpAMD64BSFQ, MakeTuple(config.fe.TypeUInt64(), TypeFlags))
+		v1 := b.NewValue0(v.Pos, OpAMD64ORQ, config.Frontend().TypeUInt64())
+		v2 := b.NewValue0(v.Pos, OpAMD64MOVQconst, config.fe.TypeUInt64())
+		v2.AuxInt = 1 << 32
+		v1.AddArg(v2)
 		v1.AddArg(x)
 		v0.AddArg(v1)
 		v.AddArg(v0)
-		v2 := b.NewValue0(v.Pos, OpAMD64MOVLconst, t)
-		v2.AuxInt = 32
-		v.AddArg(v2)
-		v3 := b.NewValue0(v.Pos, OpSelect1, TypeFlags)
-		v4 := b.NewValue0(v.Pos, OpAMD64BSFL, MakeTuple(config.fe.TypeUInt32(), TypeFlags))
-		v4.AddArg(x)
-		v3.AddArg(v4)
-		v.AddArg(v3)
 		return true
 	}
 }

src/cmd/compile/internal/ssa/rewritedec64.go

@@ -368,34 +368,30 @@ func rewriteValuedec64_OpCtz64(v *Value, config *Config) bool {
 	_ = b
 	// match: (Ctz64 x)
 	// cond:
-	// result: (Int64Make 		(Const32 <config.fe.TypeUInt32()> [0]) 		(Add32 <config.fe.TypeUInt32()> 			(Ctz32 <config.fe.TypeUInt32()> (Int64Lo x)) 			(And32 <config.fe.TypeUInt32()> 				(Com32 <config.fe.TypeUInt32()> (Zeromask (Int64Lo x))) 				(Ctz32 <config.fe.TypeUInt32()> (Int64Hi x)))))
+	// result: (Add32 <config.fe.TypeUInt32()> 		(Ctz32 <config.fe.TypeUInt32()> (Int64Lo x)) 		(And32 <config.fe.TypeUInt32()> 			(Com32 <config.fe.TypeUInt32()> (Zeromask (Int64Lo x))) 			(Ctz32 <config.fe.TypeUInt32()> (Int64Hi x))))
 	for {
 		x := v.Args[0]
-		v.reset(OpInt64Make)
-		v0 := b.NewValue0(v.Pos, OpConst32, config.fe.TypeUInt32())
-		v0.AuxInt = 0
+		v.reset(OpAdd32)
+		v.Type = config.fe.TypeUInt32()
+		v0 := b.NewValue0(v.Pos, OpCtz32, config.fe.TypeUInt32())
+		v1 := b.NewValue0(v.Pos, OpInt64Lo, config.fe.TypeUInt32())
+		v1.AddArg(x)
+		v0.AddArg(v1)
 		v.AddArg(v0)
-		v1 := b.NewValue0(v.Pos, OpAdd32, config.fe.TypeUInt32())
-		v2 := b.NewValue0(v.Pos, OpCtz32, config.fe.TypeUInt32())
-		v3 := b.NewValue0(v.Pos, OpInt64Lo, config.fe.TypeUInt32())
-		v3.AddArg(x)
+		v2 := b.NewValue0(v.Pos, OpAnd32, config.fe.TypeUInt32())
+		v3 := b.NewValue0(v.Pos, OpCom32, config.fe.TypeUInt32())
+		v4 := b.NewValue0(v.Pos, OpZeromask, config.fe.TypeUInt32())
+		v5 := b.NewValue0(v.Pos, OpInt64Lo, config.fe.TypeUInt32())
+		v5.AddArg(x)
+		v4.AddArg(v5)
+		v3.AddArg(v4)
 		v2.AddArg(v3)
-		v1.AddArg(v2)
-		v4 := b.NewValue0(v.Pos, OpAnd32, config.fe.TypeUInt32())
-		v5 := b.NewValue0(v.Pos, OpCom32, config.fe.TypeUInt32())
-		v6 := b.NewValue0(v.Pos, OpZeromask, config.fe.TypeUInt32())
-		v7 := b.NewValue0(v.Pos, OpInt64Lo, config.fe.TypeUInt32())
+		v6 := b.NewValue0(v.Pos, OpCtz32, config.fe.TypeUInt32())
+		v7 := b.NewValue0(v.Pos, OpInt64Hi, config.fe.TypeUInt32())
 		v7.AddArg(x)
 		v6.AddArg(v7)
-		v5.AddArg(v6)
-		v4.AddArg(v5)
-		v8 := b.NewValue0(v.Pos, OpCtz32, config.fe.TypeUInt32())
-		v9 := b.NewValue0(v.Pos, OpInt64Hi, config.fe.TypeUInt32())
-		v9.AddArg(x)
-		v8.AddArg(v9)
-		v4.AddArg(v8)
-		v1.AddArg(v4)
-		v.AddArg(v1)
+		v2.AddArg(v6)
+		v.AddArg(v2)
 		return true
 	}
 }

src/runtime/internal/sys/intrinsics.go

@@ -32,22 +32,22 @@ var deBruijnIdx32 = [32]byte{
 
 // Ctz64 counts trailing (low-order) zeroes,
 // and if all are zero, then 64.
-func Ctz64(x uint64) uint64 {
+func Ctz64(x uint64) int {
 	x &= -x                      // isolate low-order bit
 	y := x * deBruijn64 >> 58    // extract part of deBruijn sequence
-	y = uint64(deBruijnIdx64[y]) // convert to bit index
-	z := (x - 1) >> 57 & 64      // adjustment if zero
-	return y + z
+	i := int(deBruijnIdx64[y])   // convert to bit index
+	z := int((x - 1) >> 57 & 64) // adjustment if zero
+	return i + z
 }
 
 // Ctz32 counts trailing (low-order) zeroes,
 // and if all are zero, then 32.
-func Ctz32(x uint32) uint32 {
+func Ctz32(x uint32) int {
 	x &= -x                      // isolate low-order bit
 	y := x * deBruijn32 >> 27    // extract part of deBruijn sequence
-	y = uint32(deBruijnIdx32[y]) // convert to bit index
-	z := (x - 1) >> 26 & 32      // adjustment if zero
-	return y + z
+	i := int(deBruijnIdx32[y])   // convert to bit index
+	z := int((x - 1) >> 26 & 32) // adjustment if zero
+	return i + z
 }
 
 // Bswap64 returns its input with byte order reversed

src/runtime/internal/sys/intrinsics_386.s

@@ -4,14 +4,12 @@
 
 #include "textflag.h"
 
-TEXT runtime∕internal∕sys·Ctz64(SB), NOSPLIT, $0-16
-	MOVL	$0, ret_hi+12(FP)
-
+TEXT runtime∕internal∕sys·Ctz64(SB), NOSPLIT, $0-12
 	// Try low 32 bits.
 	MOVL	x_lo+0(FP), AX
 	BSFL	AX, AX
 	JZ	tryhigh
-	MOVL	AX, ret_lo+8(FP)
+	MOVL	AX, ret+8(FP)
 	RET
 
 tryhigh:
@@ -20,12 +18,12 @@ tryhigh:
 	BSFL	AX, AX
 	JZ	none
 	ADDL	$32, AX
-	MOVL	AX, ret_lo+8(FP)
+	MOVL	AX, ret+8(FP)
 	RET
 
 none:
 	// No bits are set.
-	MOVL	$64, ret_lo+8(FP)
+	MOVL	$64, ret+8(FP)
 	RET
 
 TEXT runtime∕internal∕sys·Ctz32(SB), NOSPLIT, $0-8

src/runtime/internal/sys/intrinsics_stubs.go

@@ -6,7 +6,7 @@
 
 package sys
 
-func Ctz64(x uint64) uint64
-func Ctz32(x uint32) uint32
+func Ctz64(x uint64) int
+func Ctz32(x uint32) int
 func Bswap64(x uint64) uint64
 func Bswap32(x uint32) uint32

src/runtime/internal/sys/intrinsics_test.go

@@ -6,17 +6,17 @@ import (
 )
 
 func TestCtz64(t *testing.T) {
-	for i := uint(0); i <= 64; i++ {
-		x := uint64(5) << i
-		if got := sys.Ctz64(x); got != uint64(i) {
+	for i := 0; i <= 64; i++ {
+		x := uint64(5) << uint(i)
+		if got := sys.Ctz64(x); got != i {
 			t.Errorf("Ctz64(%d)=%d, want %d", x, got, i)
 		}
 	}
 }
 func TestCtz32(t *testing.T) {
-	for i := uint(0); i <= 32; i++ {
-		x := uint32(5) << i
-		if got := sys.Ctz32(x); got != uint32(i) {
+	for i := 0; i <= 32; i++ {
+		x := uint32(5) << uint(i)
+		if got := sys.Ctz32(x); got != i {
 			t.Errorf("Ctz32(%d)=%d, want %d", x, got, i)
 		}
 	}

src/runtime/malloc.go

@@ -491,7 +491,7 @@ func nextFreeFast(s *mspan) gclinkptr {
 			if freeidx%64 == 0 && freeidx != s.nelems {
 				return 0
 			}
-			s.allocCache >>= (theBit + 1)
+			s.allocCache >>= uint(theBit + 1)
 			s.freeindex = freeidx
 			v := gclinkptr(result*s.elemsize + s.base())
 			s.allocCount++

src/runtime/mbitmap.go

@@ -248,7 +248,7 @@ func (s *mspan) nextFreeIndex() uintptr {
 		return snelems
 	}
 
-	s.allocCache >>= (bitIndex + 1)
+	s.allocCache >>= uint(bitIndex + 1)
 	sfreeindex = result + 1
 
 	if sfreeindex%64 == 0 && sfreeindex != snelems {

test/intrinsic.dir/main.go

@@ -22,7 +22,7 @@ func logf(f string, args ...interface{}) {
 	}
 }
 
-func test(i, x uint64) {
+func test(i int, x uint64) {
 	t := T.Ctz64(x) // ERROR "intrinsic substitution for Ctz64"
 	if i != t {
 		logf("Ctz64(0x%x) expected %d but got %d\n", x, i, t)
@@ -36,12 +36,12 @@ func test(i, x uint64) {
 	if i <= 32 {
 		x32 := uint32(x)
 		t32 := T.Ctz32(x32) // ERROR "intrinsic substitution for Ctz32"
-		if uint32(i) != t32 {
+		if i != t32 {
 			logf("Ctz32(0x%x) expected %d but got %d\n", x32, i, t32)
 		}
 		x32 = -x32
 		t32 = T.Ctz32(x32) // ERROR "intrinsic substitution for Ctz32"
-		if uint32(i) != t32 {
+		if i != t32 {
 			logf("Ctz32(0x%x) expected %d but got %d\n", x32, i, t32)
 		}
 	}
@@ -83,10 +83,10 @@ func main() {
 		logf("ctz64(0) != 64")
 	}
 
-	for i := uint64(0); i <= 64; i++ {
+	for i := 0; i <= 64; i++ {
 		for j := uint64(1); j <= 255; j += 2 {
 			for k := uint64(1); k <= 65537; k += 128 {
-				x := (j * k) << i
+				x := (j * k) << uint(i)
 				test(i, x)
 			}
 		}