cmd/compile: ppc64x intrinsics for math/bits

This adds math/bits intrinsics for OnesCount, Len, TrailingZeros on
ppc64x.

benchmark                       old ns/op     new ns/op     delta
BenchmarkLeadingZeros-16        4.26          1.71          -59.86%
BenchmarkLeadingZeros16-16      3.04          1.83          -39.80%
BenchmarkLeadingZeros32-16      3.31          1.82          -45.02%
BenchmarkLeadingZeros64-16      3.69          1.71          -53.66%
BenchmarkTrailingZeros-16       2.55          1.62          -36.47%
BenchmarkTrailingZeros32-16     2.55          1.77          -30.59%
BenchmarkTrailingZeros64-16     2.78          1.62          -41.73%
BenchmarkOnesCount-16           3.19          0.93          -70.85%
BenchmarkOnesCount32-16         2.55          1.18          -53.73%
BenchmarkOnesCount64-16         3.22          0.93          -71.12%

Update #18616

I also made a change to bits_test.go because when debugging some failures
the output was not quite providing the right argument information.

Change-Id: Ia58d31d1777cf4582a4505f85b11a1202ca07d3e
Reviewed-on: https://go-review.googlesource.com/41630
Run-TryBot: Lynn Boger <[email protected]>
TryBot-Result: Gobot Gobot <[email protected]>
Reviewed-by: Carlos Eduardo Seo <[email protected]>
Reviewed-by: Keith Randall <[email protected]>

Author: laboger

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

@@ -2730,12 +2730,12 @@ func init() {
 		func(s *state, n *Node, args []*ssa.Value) *ssa.Value {
 			return s.newValue1(ssa.OpCtz64, types.Types[TINT], args[0])
 		},
-		sys.AMD64, sys.ARM64, sys.ARM, sys.S390X, sys.MIPS)
+		sys.AMD64, sys.ARM64, sys.ARM, sys.S390X, sys.MIPS, sys.PPC64)
 	addF("math/bits", "TrailingZeros32",
 		func(s *state, n *Node, args []*ssa.Value) *ssa.Value {
 			return s.newValue1(ssa.OpCtz32, types.Types[TINT], args[0])
 		},
-		sys.AMD64, sys.ARM64, sys.ARM, sys.S390X, sys.MIPS)
+		sys.AMD64, sys.ARM64, sys.ARM, sys.S390X, sys.MIPS, sys.PPC64)
 	addF("math/bits", "TrailingZeros16",
 		func(s *state, n *Node, args []*ssa.Value) *ssa.Value {
 			x := s.newValue1(ssa.OpZeroExt16to32, types.Types[TUINT32], args[0])
@@ -2776,7 +2776,7 @@ func init() {
 		func(s *state, n *Node, args []*ssa.Value) *ssa.Value {
 			return s.newValue1(ssa.OpBitLen64, types.Types[TINT], args[0])
 		},
-		sys.AMD64, sys.ARM64, sys.ARM, sys.S390X, sys.MIPS)
+		sys.AMD64, sys.ARM64, sys.ARM, sys.S390X, sys.MIPS, sys.PPC64)
 	addF("math/bits", "Len32",
 		func(s *state, n *Node, args []*ssa.Value) *ssa.Value {
 			if s.config.PtrSize == 4 {
@@ -2785,7 +2785,7 @@ func init() {
 			x := s.newValue1(ssa.OpZeroExt32to64, types.Types[TUINT64], args[0])
 			return s.newValue1(ssa.OpBitLen64, types.Types[TINT], x)
 		},
-		sys.AMD64, sys.ARM64, sys.ARM, sys.S390X, sys.MIPS)
+		sys.AMD64, sys.ARM64, sys.ARM, sys.S390X, sys.MIPS, sys.PPC64)
 	addF("math/bits", "Len16",
 		func(s *state, n *Node, args []*ssa.Value) *ssa.Value {
 			if s.config.PtrSize == 4 {
@@ -2795,7 +2795,7 @@ func init() {
 			x := s.newValue1(ssa.OpZeroExt16to64, types.Types[TUINT64], args[0])
 			return s.newValue1(ssa.OpBitLen64, types.Types[TINT], x)
 		},
-		sys.AMD64, sys.ARM64, sys.ARM, sys.S390X, sys.MIPS)
+		sys.AMD64, sys.ARM64, sys.ARM, sys.S390X, sys.MIPS, sys.PPC64)
 	// Note: disabled on AMD64 because the Go code is faster!
 	addF("math/bits", "Len8",
 		func(s *state, n *Node, args []*ssa.Value) *ssa.Value {
@@ -2806,7 +2806,7 @@ func init() {
 			x := s.newValue1(ssa.OpZeroExt8to64, types.Types[TUINT64], args[0])
 			return s.newValue1(ssa.OpBitLen64, types.Types[TINT], x)
 		},
-		sys.ARM64, sys.ARM, sys.S390X, sys.MIPS)
+		sys.ARM64, sys.ARM, sys.S390X, sys.MIPS, sys.PPC64)
 
 	addF("math/bits", "Len",
 		func(s *state, n *Node, args []*ssa.Value) *ssa.Value {
@@ -2815,7 +2815,7 @@ func init() {
 			}
 			return s.newValue1(ssa.OpBitLen64, types.Types[TINT], args[0])
 		},
-		sys.AMD64, sys.ARM64, sys.ARM, sys.S390X, sys.MIPS)
+		sys.AMD64, sys.ARM64, sys.ARM, sys.S390X, sys.MIPS, sys.PPC64)
 	// LeadingZeros is handled because it trivially calls Len.
 	addF("math/bits", "Reverse64",
 		func(s *state, n *Node, args []*ssa.Value) *ssa.Value {
@@ -2845,7 +2845,7 @@ func init() {
 			return s.newValue1(ssa.OpBitRev64, types.Types[TINT], args[0])
 		},
 		sys.ARM64)
-	makeOnesCount := func(op64 ssa.Op, op32 ssa.Op) func(s *state, n *Node, args []*ssa.Value) *ssa.Value {
+	makeOnesCountAMD64 := func(op64 ssa.Op, op32 ssa.Op) func(s *state, n *Node, args []*ssa.Value) *ssa.Value {
 		return func(s *state, n *Node, args []*ssa.Value) *ssa.Value {
 			aux := s.lookupSymbol(n, &ssa.ExternSymbol{Sym: syslook("support_popcnt").Sym.Linksym()})
 			addr := s.entryNewValue1A(ssa.OpAddr, types.Types[TBOOL].PtrTo(), aux, s.sb)
@@ -2881,17 +2881,27 @@ func init() {
 		}
 	}
 	addF("math/bits", "OnesCount64",
-		makeOnesCount(ssa.OpPopCount64, ssa.OpPopCount64),
+		makeOnesCountAMD64(ssa.OpPopCount64, ssa.OpPopCount64),
 		sys.AMD64)
+	addF("math/bits", "OnesCount64",
+		func(s *state, n *Node, args []*ssa.Value) *ssa.Value {
+			return s.newValue1(ssa.OpPopCount64, types.Types[TINT], args[0])
+		},
+		sys.PPC64)
 	addF("math/bits", "OnesCount32",
-		makeOnesCount(ssa.OpPopCount32, ssa.OpPopCount32),
+		makeOnesCountAMD64(ssa.OpPopCount32, ssa.OpPopCount32),
 		sys.AMD64)
+	addF("math/bits", "OnesCount32",
+		func(s *state, n *Node, args []*ssa.Value) *ssa.Value {
+			return s.newValue1(ssa.OpPopCount32, types.Types[TINT], args[0])
+		},
+		sys.PPC64)
 	addF("math/bits", "OnesCount16",
-		makeOnesCount(ssa.OpPopCount16, ssa.OpPopCount16),
+		makeOnesCountAMD64(ssa.OpPopCount16, ssa.OpPopCount16),
 		sys.AMD64)
 	// Note: no OnesCount8, the Go implementation is faster - just a table load.
 	addF("math/bits", "OnesCount",
-		makeOnesCount(ssa.OpPopCount64, ssa.OpPopCount32),
+		makeOnesCountAMD64(ssa.OpPopCount64, ssa.OpPopCount32),
 		sys.AMD64)
 
 	/******** sync/atomic ********/

src/cmd/compile/internal/ppc64/ssa.go

@@ -596,7 +596,7 @@ func ssaGenValue(s *gc.SSAGenState, v *ssa.Value) {
 		p.To.Type = obj.TYPE_REG
 		p.To.Reg = ppc64.REGTMP // Ignored; this is for the carry effect.
 
-	case ssa.OpPPC64NEG, ssa.OpPPC64FNEG, ssa.OpPPC64FSQRT, ssa.OpPPC64FSQRTS, ssa.OpPPC64FCTIDZ, ssa.OpPPC64FCTIWZ, ssa.OpPPC64FCFID, ssa.OpPPC64FRSP:
+	case ssa.OpPPC64NEG, ssa.OpPPC64FNEG, ssa.OpPPC64FSQRT, ssa.OpPPC64FSQRTS, ssa.OpPPC64FCTIDZ, ssa.OpPPC64FCTIWZ, ssa.OpPPC64FCFID, ssa.OpPPC64FRSP, ssa.OpPPC64CNTLZD, ssa.OpPPC64CNTLZW, ssa.OpPPC64POPCNTD, ssa.OpPPC64POPCNTW, ssa.OpPPC64POPCNTB:
 		r := v.Reg()
 		p := s.Prog(v.Op.Asm())
 		p.To.Type = obj.TYPE_REG

src/cmd/compile/internal/ssa/gen/PPC64.rules

@@ -244,6 +244,17 @@
 // (Addr {sym} base) -> (ADDconst {sym} base)
 (OffPtr [off] ptr) -> (ADD (MOVDconst <typ.Int64> [off]) ptr)
 
+(Ctz64 x) -> (POPCNTD (ANDN <types.Int64> (ADDconst <types.Int64> [-1] x) x))
+(Ctz32 x) -> (POPCNTW (MOVWZreg (ANDN <types.Int> (ADDconst <types.Int> [-1] x) x)))
+
+(BitLen64 x) -> (SUB (MOVDconst [64]) (CNTLZD <types.Int> x))
+(BitLen32 x) -> (SUB (MOVDconst [32]) (CNTLZW <types.Int> x))
+
+(PopCount64 x) -> (POPCNTD x)
+(PopCount32 x) -> (POPCNTW (MOVWZreg x))
+(PopCount16 x) -> (POPCNTW (MOVHZreg x))
+(PopCount8 x) -> (POPCNTB (MOVBreg x))
+
 (And64 x y) -> (AND x y)
 (And32 x y) -> (AND x y)
 (And16 x y) -> (AND x y)

src/cmd/compile/internal/ssa/gen/PPC64Ops.go

@@ -198,6 +198,13 @@ func init() {
 		{name: "ROTLconst", argLength: 1, reg: gp11, asm: "ROTL", aux: "Int64"},   // arg0 rotate left by auxInt bits
 		{name: "ROTLWconst", argLength: 1, reg: gp11, asm: "ROTLW", aux: "Int64"}, // uint32(arg0) rotate left by auxInt bits
 
+		{name: "CNTLZD", argLength: 1, reg: gp11, asm: "CNTLZD", clobberFlags: true}, // count leading zeros
+		{name: "CNTLZW", argLength: 1, reg: gp11, asm: "CNTLZW", clobberFlags: true}, // count leading zeros (32 bit)
+
+		{name: "POPCNTD", argLength: 1, reg: gp11, asm: "POPCNTD"}, // number of set bits in arg0
+		{name: "POPCNTW", argLength: 1, reg: gp11, asm: "POPCNTW"}, // number of set bits in each word of arg0 placed in corresponding word
+		{name: "POPCNTB", argLength: 1, reg: gp11, asm: "POPCNTB"}, // number of set bits in each byte of arg0 placed in corresonding byte
+
 		{name: "FDIV", argLength: 2, reg: fp21, asm: "FDIV"},   // arg0/arg1
 		{name: "FDIVS", argLength: 2, reg: fp21, asm: "FDIVS"}, // arg0/arg1