cmd/compile: don't emit unnecessary amd64 extension checks

In case of amd64 the compiler issues checks if extensions are
available on a platform. With GOAMD64 microarchitecture levels
provided, some of the checks could be eliminated.

Change-Id: If15c178bcae273b2ce7d3673415cb8849292e087
Reviewed-on: https://go-review.googlesource.com/c/go/+/352010
Reviewed-by: Keith Randall <khr@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
Run-TryBot: Keith Randall <khr@golang.org>
TryBot-Result: Go Bot <gobot@golang.org>

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

@@ -4245,6 +4245,11 @@ func InitTables() {
 				s.vars[n] = s.callResult(n, callNormal) // types.Types[TFLOAT64]
 				return s.variable(n, types.Types[types.TFLOAT64])
 			}
+
+			if buildcfg.GOAMD64 >= 3 {
+				return s.newValue3(ssa.OpFMA, types.Types[types.TFLOAT64], args[0], args[1], args[2])
+			}
+
 			v := s.entryNewValue0A(ssa.OpHasCPUFeature, types.Types[types.TBOOL], ir.Syms.X86HasFMA)
 			b := s.endBlock()
 			b.Kind = ssa.BlockIf
@@ -4307,6 +4312,10 @@ func InitTables() {
 
 	makeRoundAMD64 := func(op ssa.Op) func(s *state, n *ir.CallExpr, args []*ssa.Value) *ssa.Value {
 		return func(s *state, n *ir.CallExpr, args []*ssa.Value) *ssa.Value {
+			if buildcfg.GOAMD64 >= 2 {
+				return s.newValue1(op, types.Types[types.TFLOAT64], args[0])
+			}
+
 			v := s.entryNewValue0A(ssa.OpHasCPUFeature, types.Types[types.TBOOL], ir.Syms.X86HasSSE41)
 			b := s.endBlock()
 			b.Kind = ssa.BlockIf
@@ -4513,6 +4522,10 @@ func InitTables() {
 
 	makeOnesCountAMD64 := func(op ssa.Op) func(s *state, n *ir.CallExpr, args []*ssa.Value) *ssa.Value {
 		return func(s *state, n *ir.CallExpr, args []*ssa.Value) *ssa.Value {
+			if buildcfg.GOAMD64 >= 2 {
+				return s.newValue1(op, types.Types[types.TINT], args[0])
+			}
+
 			v := s.entryNewValue0A(ssa.OpHasCPUFeature, types.Types[types.TBOOL], ir.Syms.X86HasPOPCNT)
 			b := s.endBlock()
 			b.Kind = ssa.BlockIf

test/codegen/math.go

@@ -11,6 +11,8 @@ import "math"
 var sink64 [8]float64
 
 func approx(x float64) {
+	// amd64/v2:-".*x86HasSSE41" amd64/v3:-".*x86HasSSE41"
+	// amd64:"ROUNDSD\t[$]2"
 	// s390x:"FIDBR\t[$]6"
 	// arm64:"FRINTPD"
 	// ppc64:"FRIP"
@@ -18,6 +20,8 @@ func approx(x float64) {
 	// wasm:"F64Ceil"
 	sink64[0] = math.Ceil(x)
 
+	// amd64/v2:-".*x86HasSSE41" amd64/v3:-".*x86HasSSE41"
+	// amd64:"ROUNDSD\t[$]1"
 	// s390x:"FIDBR\t[$]7"
 	// arm64:"FRINTMD"
 	// ppc64:"FRIM"
@@ -31,6 +35,8 @@ func approx(x float64) {
 	// ppc64le:"FRIN"
 	sink64[2] = math.Round(x)
 
+	// amd64/v2:-".*x86HasSSE41" amd64/v3:-".*x86HasSSE41"
+	// amd64:"ROUNDSD\t[$]3"
 	// s390x:"FIDBR\t[$]5"
 	// arm64:"FRINTZD"
 	// ppc64:"FRIZ"
@@ -38,6 +44,8 @@ func approx(x float64) {
 	// wasm:"F64Trunc"
 	sink64[3] = math.Trunc(x)
 
+	// amd64/v2:-".*x86HasSSE41" amd64/v3:-".*x86HasSSE41"
+	// amd64:"ROUNDSD\t[$]0"
 	// s390x:"FIDBR\t[$]4"
 	// arm64:"FRINTND"
 	// wasm:"F64Nearest"
@@ -123,6 +131,7 @@ func copysign(a, b, c float64) {
 }
 
 func fma(x, y, z float64) float64 {
+	// amd64/v3:-".*x86HasFMA"
 	// amd64:"VFMADD231SD"
 	// arm/6:"FMULAD"
 	// arm64:"FMADDD"

test/codegen/mathbits.go

@@ -118,8 +118,9 @@ func Len8(n uint8) int {
 //    bits.OnesCount    //
 // -------------------- //
 
-// TODO(register args) Restore a m d 6 4 :.*x86HasPOPCNT when only one ABI is tested.
+// TODO(register args) Restore a m d 6 4 / v 1 :.*x86HasPOPCNT when only one ABI is tested.
 func OnesCount(n uint) int {
+	// amd64/v2:-".*x86HasPOPCNT" amd64/v3:-".*x86HasPOPCNT"
 	// amd64:"POPCNTQ"
 	// arm64:"VCNT","VUADDLV"
 	// s390x:"POPCNT"
@@ -130,6 +131,7 @@ func OnesCount(n uint) int {
 }
 
 func OnesCount64(n uint64) int {
+	// amd64/v2:-".*x86HasPOPCNT" amd64/v3:-".*x86HasPOPCNT"
 	// amd64:"POPCNTQ"
 	// arm64:"VCNT","VUADDLV"
 	// s390x:"POPCNT"
@@ -140,6 +142,7 @@ func OnesCount64(n uint64) int {
 }
 
 func OnesCount32(n uint32) int {
+	// amd64/v2:-".*x86HasPOPCNT" amd64/v3:-".*x86HasPOPCNT"
 	// amd64:"POPCNTL"
 	// arm64:"VCNT","VUADDLV"
 	// s390x:"POPCNT"
@@ -150,6 +153,7 @@ func OnesCount32(n uint32) int {
 }
 
 func OnesCount16(n uint16) int {
+	// amd64/v2:-".*x86HasPOPCNT" amd64/v3:-".*x86HasPOPCNT"
 	// amd64:"POPCNTL"
 	// arm64:"VCNT","VUADDLV"
 	// s390x:"POPCNT"