From: Carlos Eduardo Seo <cseo@linux.vnet.ibm.com>
Date: Fri, 22 Feb 2019 19:12:37 +0000 (-0300)
Subject: cmd/compile: make math/bits.RotateLeft{32,64} intrinsics on ppc64x
X-Git-Tag: go1.13beta1~1037
X-Git-Url: http://www.git.cypherpunks.su/?a=commitdiff_plain;h=f2a18b1456e9cbb83ada13776195c56d2a6fb951;p=gostls13.git

cmd/compile: make math/bits.RotateLeft{32,64} intrinsics on ppc64x

Extends CL 132435 to ppc64x. ppc64x has 32- and 64-bit variable
rotate left instructions.

name             old time/op  new time/op  delta
RotateLeft32-16  1.39ns ± 0%  1.37ns ± 0%  -1.44%  (p=0.008 n=5+5)
RotateLeft64-16  1.35ns ± 0%  1.32ns ± 0%  -2.22%  (p=0.008 n=5+5)

Updates #17566

Change-Id: I567f634ff90d0691db45df0a25c99fcdfe10ca00
Reviewed-on: https://go-review.googlesource.com/c/go/+/163760
Reviewed-by: Lynn Boger <laboger@linux.vnet.ibm.com>
---

diff --git a/src/cmd/compile/internal/gc/ssa.go b/src/cmd/compile/internal/gc/ssa.go
index ecc449114d..d61f463ccf 100644
--- a/src/cmd/compile/internal/gc/ssa.go
+++ b/src/cmd/compile/internal/gc/ssa.go
@@ -3427,12 +3427,12 @@ func init() {
 		func(s *state, n *Node, args []*ssa.Value) *ssa.Value {
 			return s.newValue2(ssa.OpRotateLeft32, types.Types[TUINT32], args[0], args[1])
 		},
-		sys.AMD64, sys.ARM64, sys.S390X)
+		sys.AMD64, sys.ARM64, sys.S390X, sys.PPC64)
 	addF("math/bits", "RotateLeft64",
 		func(s *state, n *Node, args []*ssa.Value) *ssa.Value {
 			return s.newValue2(ssa.OpRotateLeft64, types.Types[TUINT64], args[0], args[1])
 		},
-		sys.AMD64, sys.ARM64, sys.S390X)
+		sys.AMD64, sys.ARM64, sys.S390X, sys.PPC64)
 	alias("math/bits", "RotateLeft", "math/bits", "RotateLeft64", p8...)
 
 	makeOnesCountAMD64 := func(op64 ssa.Op, op32 ssa.Op) func(s *state, n *Node, args []*ssa.Value) *ssa.Value {
diff --git a/src/cmd/compile/internal/ssa/gen/PPC64.rules b/src/cmd/compile/internal/ssa/gen/PPC64.rules
index e5d5295908..8dee5a1cba 100644
--- a/src/cmd/compile/internal/ssa/gen/PPC64.rules
+++ b/src/cmd/compile/internal/ssa/gen/PPC64.rules
@@ -100,6 +100,14 @@
 ( OR (SLW x (ANDconst <typ.Int32> [31] y)) (SRW x (SUB <typ.UInt> (MOVDconst [32]) (ANDconst <typ.UInt> [31] y)))) -> (ROTLW x y)
 (XOR (SLW x (ANDconst <typ.Int32> [31] y)) (SRW x (SUB <typ.UInt> (MOVDconst [32]) (ANDconst <typ.UInt> [31] y)))) -> (ROTLW x y)
 
+// Lowering rotates
+(RotateLeft32 x y) -> (ROTLW x y)
+(RotateLeft64 x y) -> (ROTL x y)
+
+// Constant rotate generation
+(ROTLW  x (MOVDconst [c])) -> (ROTLWconst  x [c&31])
+(ROTL   x (MOVDconst [c])) -> (ROTLconst   x [c&63])
+
 (Lsh64x64  x (Const64 [c])) && uint64(c) < 64 -> (SLDconst x [c])
 (Rsh64x64  x (Const64 [c])) && uint64(c) < 64 -> (SRADconst x [c])
 (Rsh64Ux64 x (Const64 [c])) && uint64(c) < 64 -> (SRDconst x [c])
diff --git a/src/cmd/compile/internal/ssa/rewritePPC64.go b/src/cmd/compile/internal/ssa/rewritePPC64.go
index 302b785f61..9245f403b8 100644
--- a/src/cmd/compile/internal/ssa/rewritePPC64.go
+++ b/src/cmd/compile/internal/ssa/rewritePPC64.go
@@ -537,6 +537,10 @@ func rewriteValuePPC64(v *Value) bool {
 		return rewriteValuePPC64_OpPPC64ORN_0(v)
 	case OpPPC64ORconst:
 		return rewriteValuePPC64_OpPPC64ORconst_0(v)
+	case OpPPC64ROTL:
+		return rewriteValuePPC64_OpPPC64ROTL_0(v)
+	case OpPPC64ROTLW:
+		return rewriteValuePPC64_OpPPC64ROTLW_0(v)
 	case OpPPC64SUB:
 		return rewriteValuePPC64_OpPPC64SUB_0(v)
 	case OpPPC64XOR:
@@ -551,6 +555,10 @@ func rewriteValuePPC64(v *Value) bool {
 		return rewriteValuePPC64_OpPopCount64_0(v)
 	case OpPopCount8:
 		return rewriteValuePPC64_OpPopCount8_0(v)
+	case OpRotateLeft32:
+		return rewriteValuePPC64_OpRotateLeft32_0(v)
+	case OpRotateLeft64:
+		return rewriteValuePPC64_OpRotateLeft64_0(v)
 	case OpRound:
 		return rewriteValuePPC64_OpRound_0(v)
 	case OpRound32F:
@@ -25533,6 +25541,44 @@ func rewriteValuePPC64_OpPPC64ORconst_0(v *Value) bool {
 	}
 	return false
 }
+func rewriteValuePPC64_OpPPC64ROTL_0(v *Value) bool {
+	// match: (ROTL x (MOVDconst [c]))
+	// cond:
+	// result: (ROTLconst x [c&63])
+	for {
+		_ = v.Args[1]
+		x := v.Args[0]
+		v_1 := v.Args[1]
+		if v_1.Op != OpPPC64MOVDconst {
+			break
+		}
+		c := v_1.AuxInt
+		v.reset(OpPPC64ROTLconst)
+		v.AuxInt = c & 63
+		v.AddArg(x)
+		return true
+	}
+	return false
+}
+func rewriteValuePPC64_OpPPC64ROTLW_0(v *Value) bool {
+	// match: (ROTLW x (MOVDconst [c]))
+	// cond:
+	// result: (ROTLWconst x [c&31])
+	for {
+		_ = v.Args[1]
+		x := v.Args[0]
+		v_1 := v.Args[1]
+		if v_1.Op != OpPPC64MOVDconst {
+			break
+		}
+		c := v_1.AuxInt
+		v.reset(OpPPC64ROTLWconst)
+		v.AuxInt = c & 31
+		v.AddArg(x)
+		return true
+	}
+	return false
+}
 func rewriteValuePPC64_OpPPC64SUB_0(v *Value) bool {
 	// match: (SUB x (MOVDconst [c]))
 	// cond: is32Bit(-c)
@@ -26086,6 +26132,34 @@ func rewriteValuePPC64_OpPopCount8_0(v *Value) bool {
 		return true
 	}
 }
+func rewriteValuePPC64_OpRotateLeft32_0(v *Value) bool {
+	// match: (RotateLeft32 x y)
+	// cond:
+	// result: (ROTLW x y)
+	for {
+		_ = v.Args[1]
+		x := v.Args[0]
+		y := v.Args[1]
+		v.reset(OpPPC64ROTLW)
+		v.AddArg(x)
+		v.AddArg(y)
+		return true
+	}
+}
+func rewriteValuePPC64_OpRotateLeft64_0(v *Value) bool {
+	// match: (RotateLeft64 x y)
+	// cond:
+	// result: (ROTL x y)
+	for {
+		_ = v.Args[1]
+		x := v.Args[0]
+		y := v.Args[1]
+		v.reset(OpPPC64ROTL)
+		v.AddArg(x)
+		v.AddArg(y)
+		return true
+	}
+}
 func rewriteValuePPC64_OpRound_0(v *Value) bool {
 	// match: (Round x)
 	// cond: