cmd/compile: intrinsify math/bits.ReverseBytes

Update #18616

Change-Id: I0c2d643cbbeb131b4c9b12194697afa4af48e1d2
Reviewed-on: https://go-review.googlesource.com/38166
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

@@ -185,6 +185,7 @@ var allAsmTests = []*asmTests{
 	{
 		arch:    "arm64",
 		os:      "linux",
+		imports: []string{"math/bits"},
 		tests:   linuxARM64Tests,
 	},
 }
@@ -576,6 +577,30 @@ var linuxAMD64Tests = []*asmTest{
 		`,
 		[]string{"\tBSFQ\t", "\tORQ\t\\$256,"},
 	},
+	{
+		`
+		func f45(a uint64) uint64 {
+			return bits.ReverseBytes64(a)
+		}
+		`,
+		[]string{"\tBSWAPQ\t"},
+	},
+	{
+		`
+		func f46(a uint32) uint32 {
+			return bits.ReverseBytes32(a)
+		}
+		`,
+		[]string{"\tBSWAPL\t"},
+	},
+	{
+		`
+		func f47(a uint16) uint16 {
+			return bits.ReverseBytes16(a)
+		}
+		`,
+		[]string{"\tROLW\t\\$8,"},
+	},
 }
 
 var linux386Tests = []*asmTest{
@@ -776,6 +801,23 @@ var linuxS390XTests = []*asmTest{
 		`,
 		[]string{"\tFLOGR\t", "\tOR\t\\$256,"},
 	},
+	// Intrinsic tests for math/bits
+	{
+		`
+		func f22(a uint64) uint64 {
+			return bits.ReverseBytes64(a)
+		}
+		`,
+		[]string{"\tMOVDBR\t"},
+	},
+	{
+		`
+		func f23(a uint32) uint32 {
+			return bits.ReverseBytes32(a)
+		}
+		`,
+		[]string{"\tMOVWBR\t"},
+	},
 }
 
 var linuxARMTests = []*asmTest{
@@ -854,6 +896,22 @@ var linuxARM64Tests = []*asmTest{
 		`,
 		[]string{"\tRORW\t[$]25,"},
 	},
+	{
+		`
+		func f22(a uint64) uint64 {
+			return bits.ReverseBytes64(a)
+		}
+		`,
+		[]string{"\tREV\t"},
+	},
+	{
+		`
+		func f23(a uint32) uint32 {
+			return bits.ReverseBytes32(a)
+		}
+		`,
+		[]string{"\tREVW\t"},
+	},
 }
 
 // TestLineNumber checks to make sure the generated assembly has line numbers

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

@@ -2702,6 +2702,10 @@ func init() {
 			return s.newValue1(ssa.OpCtz64, Types[TINT], y)
 		},
 		sys.AMD64, sys.ARM64, sys.S390X)
+	alias("math/bits", "ReverseBytes64", "runtime/internal/sys", "Bswap64", all...)
+	alias("math/bits", "ReverseBytes32", "runtime/internal/sys", "Bswap32", all...)
+	// ReverseBytes inlines correctly, no need to intrinsify it.
+	// ReverseBytes16 lowers to a rotate, no need for anything special here.
 
 	/******** sync/atomic ********/