cmd/compile: wire up Bswap/ReverseBytes intrinsics for loong64

Updates golang#59120

Change-Id: I39c1edbd7363f454ad1e848a25abeced722b16ac

src/cmd/compile/internal/loong64/ssa.go

@@ -357,6 +357,9 @@ func ssaGenValue(s *ssagen.State, v *ssa.Value) {
 		ssa.OpLOONG64CTOV,
 		ssa.OpLOONG64CLOW,
 		ssa.OpLOONG64CLOV,
+		ssa.OpLOONG64REVB2H,
+		ssa.OpLOONG64REVB2W,
+		ssa.OpLOONG64REVBV,
 		ssa.OpLOONG64SQRTD,
 		ssa.OpLOONG64SQRTF:
 		p := s.Prog(v.Op.Asm())

src/cmd/compile/internal/ssa/_gen/LOONG64.rules

@@ -130,6 +130,7 @@
 (Ctz(8|16|32|64)NonZero ...) => (Ctz(8|16|32|64) ...)
 (BitLen64 <t> x) => (NEGV <t> (SUBVconst <t> [64] (CLZV <t> x)))
 (BitLen32 <t> x) => (NEGV <t> (SUBVconst <t> [32] (CLZW <t> x)))
+(Bswap(16|32|64) ...) => (REVB(2H|2W|V) ...)
 
 (Com(64|32|16|8) x) => (NOR (MOVVconst [0]) x)
 (NOR (MOVVconst [0]) (C(TZW|TZV|LZW|LZV) x)) => (C(TOW|TOV|LOW|LOV) x)

src/cmd/compile/internal/ssa/_gen/LOONG64Ops.go

@@ -187,19 +187,22 @@ func init() {
 		{name: "NOR", argLength: 2, reg: gp21, asm: "NOR", commutative: true},                // ^(arg0 | arg1)
 		{name: "NORconst", argLength: 1, reg: gp11, asm: "NOR", aux: "Int64"},                // ^(arg0 | auxInt)
 
-		{name: "NEGV", argLength: 1, reg: gp11},                // -arg0
-		{name: "NEGF", argLength: 1, reg: fp11, asm: "NEGF"},   // -arg0, float32
-		{name: "NEGD", argLength: 1, reg: fp11, asm: "NEGD"},   // -arg0, float64
-		{name: "SQRTD", argLength: 1, reg: fp11, asm: "SQRTD"}, // sqrt(arg0), float64
-		{name: "SQRTF", argLength: 1, reg: fp11, asm: "SQRTF"}, // sqrt(arg0), float32
-		{name: "CTZW", argLength: 1, reg: gp11, asm: "CTZW"},   // Count trailing (low order) zeroes (returns 0-32)
-		{name: "CTZV", argLength: 1, reg: gp11, asm: "CTZV"},   // Count trailing (low order) zeroes (returns 0-64)
-		{name: "CLZW", argLength: 1, reg: gp11, asm: "CLZW"},   // Count leading (high order) zeroes (returns 0-32)
-		{name: "CLZV", argLength: 1, reg: gp11, asm: "CLZV"},   // Count leading (high order) zeroes (returns 0-64)
-		{name: "CTOW", argLength: 1, reg: gp11, asm: "CTOW"},   // Count trailing (low order) ones (returns 0-32)
-		{name: "CTOV", argLength: 1, reg: gp11, asm: "CTOV"},   // Count trailing (low order) ones (returns 0-64)
-		{name: "CLOW", argLength: 1, reg: gp11, asm: "CLOW"},   // Count leading (high order) ones (returns 0-32)
-		{name: "CLOV", argLength: 1, reg: gp11, asm: "CLOV"},   // Count leading (high order) ones (returns 0-64)
+		{name: "NEGV", argLength: 1, reg: gp11},                  // -arg0
+		{name: "NEGF", argLength: 1, reg: fp11, asm: "NEGF"},     // -arg0, float32
+		{name: "NEGD", argLength: 1, reg: fp11, asm: "NEGD"},     // -arg0, float64
+		{name: "SQRTD", argLength: 1, reg: fp11, asm: "SQRTD"},   // sqrt(arg0), float64
+		{name: "SQRTF", argLength: 1, reg: fp11, asm: "SQRTF"},   // sqrt(arg0), float32
+		{name: "CTZW", argLength: 1, reg: gp11, asm: "CTZW"},     // Count trailing (low order) zeroes (returns 0-32)
+		{name: "CTZV", argLength: 1, reg: gp11, asm: "CTZV"},     // Count trailing (low order) zeroes (returns 0-64)
+		{name: "CLZW", argLength: 1, reg: gp11, asm: "CLZW"},     // Count leading (high order) zeroes (returns 0-32)
+		{name: "CLZV", argLength: 1, reg: gp11, asm: "CLZV"},     // Count leading (high order) zeroes (returns 0-64)
+		{name: "CTOW", argLength: 1, reg: gp11, asm: "CTOW"},     // Count trailing (low order) ones (returns 0-32)
+		{name: "CTOV", argLength: 1, reg: gp11, asm: "CTOV"},     // Count trailing (low order) ones (returns 0-64)
+		{name: "CLOW", argLength: 1, reg: gp11, asm: "CLOW"},     // Count leading (high order) ones (returns 0-32)
+		{name: "CLOV", argLength: 1, reg: gp11, asm: "CLOV"},     // Count leading (high order) ones (returns 0-64)
+		{name: "REVB2H", argLength: 1, reg: gp11, asm: "REVB2H"}, // Swap bytes: 0x11223344 -> 0x22114433 (sign extends to 64 bits)
+		{name: "REVB2W", argLength: 1, reg: gp11, asm: "REVB2W"}, // Swap bytes: 0x1122334455667788 -> 0x4433221188776655
+		{name: "REVBV", argLength: 1, reg: gp11, asm: "REVBV"},   // Swap bytes: 0x1122334455667788 -> 0x8877665544332211
 
 		{name: "MASKEQZ", argLength: 2, reg: gp21, asm: "MASKEQZ"}, // returns 0 if arg1 == 0, otherwise returns arg0
 		{name: "MASKNEZ", argLength: 2, reg: gp21, asm: "MASKNEZ"}, // returns 0 if arg1 != 0, otherwise returns arg0

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

@@ -4000,7 +4000,7 @@ func InitTables() {
 		},
 		sys.ARM64, sys.PPC64)
 
-	brev_arch := []sys.ArchFamily{sys.AMD64, sys.I386, sys.ARM64, sys.ARM, sys.S390X}
+	brev_arch := []sys.ArchFamily{sys.AMD64, sys.I386, sys.ARM64, sys.ARM, sys.Loong64, sys.S390X}
 	if buildcfg.GOPPC64 >= 10 {
 		// Use only on Power10 as the new byte reverse instructions that Power10 provide
 		// make it worthwhile as an intrinsic
@@ -4542,6 +4542,11 @@ func InitTables() {
 		sys.Loong64, sys.S390X)
 	alias("math/bits", "ReverseBytes64", "runtime/internal/sys", "Bswap64", all...)
 	alias("math/bits", "ReverseBytes32", "runtime/internal/sys", "Bswap32", all...)
+	addF("math/bits", "ReverseBytes16",
+		func(s *state, n *ir.CallExpr, args []*ssa.Value) *ssa.Value {
+			return s.newValue1(ssa.OpBswap16, types.Types[types.TUINT16], args[0])
+		},
+		sys.Loong64)
 	// ReverseBytes inlines correctly, no need to intrinsify it.
 	// Nothing special is needed for targets where ReverseBytes16 lowers to a rotate
 	// On Power10, 16-bit rotate is not available so use BRH instruction

test/codegen/mathbits.go

@@ -203,6 +203,7 @@ func ReverseBytes(n uint) uint {
 	// 386:"BSWAPL"
 	// s390x:"MOVDBR"
 	// arm64:"REV"
+	// loong64:"REVBV"
 	return bits.ReverseBytes(n)
 }
 
@@ -211,6 +212,7 @@ func ReverseBytes64(n uint64) uint64 {
 	// 386:"BSWAPL"
 	// s390x:"MOVDBR"
 	// arm64:"REV"
+	// loong64:"REVBV"
 	// ppc64x/power10: "BRD"
 	return bits.ReverseBytes64(n)
 }
@@ -220,6 +222,7 @@ func ReverseBytes32(n uint32) uint32 {
 	// 386:"BSWAPL"
 	// s390x:"MOVWBR"
 	// arm64:"REVW"
+	// loong64:"REVB2W"
 	// ppc64x/power10: "BRW"
 	return bits.ReverseBytes32(n)
 }
@@ -230,6 +233,7 @@ func ReverseBytes16(n uint16) uint16 {
 	// arm/5:"SLL","SRL","ORR"
 	// arm/6:"REV16"
 	// arm/7:"REV16"
+	// loong64:"REVB2H"
 	// ppc64x/power10: "BRH"
 	return bits.ReverseBytes16(n)
 }