From 80834af2063962bad87a56f0f45546cc18f30a9c Mon Sep 17 00:00:00 2001 From: Lynn Boger Date: Wed, 27 Sep 2023 12:15:04 -0500 Subject: [PATCH] cmd/compile: avoid ANDCCconst on PPC64 if condition not needed In the PPC64 ISA, the instruction to do an 'and' operation using an immediate constant is only available in the form that also sets CR0 (i.e. clobbers the condition register.) This means CR0 is being clobbered unnecessarily in many cases. That affects some decisions made during some compiler passes that check for it. In those cases when the constant used by the ANDCC is a right justified consecutive set of bits, a shift instruction can be used which has the same effect if CR0 does not need to be set. The rule to do that has been added to the late rules file after other rules using ANDCCconst have been processed in the main rules file. Some codegen tests had to be updated since ANDCC is no longer generated for some cases. A new test case was added to verify the ANDCC is present if the results for both the AND and CR0 are used. Change-Id: I304f607c039a458e2d67d25351dd00aea72ba542 Reviewed-on: https://go-review.googlesource.com/c/go/+/531435 Run-TryBot: Lynn Boger Reviewed-by: Paul Murphy LUCI-TryBot-Result: Go LUCI Reviewed-by: Carlos Amedee Reviewed-by: Jayanth Krishnamurthy TryBot-Result: Gopher Robot Reviewed-by: Dmitri Shuralyov --- src/cmd/compile/internal/ssa/_gen/PPC64.rules | 8 +- .../internal/ssa/_gen/PPC64latelower.rules | 3 + src/cmd/compile/internal/ssa/rewritePPC64.go | 280 ++++-------------- .../internal/ssa/rewritePPC64latelower.go | 24 ++ test/codegen/arithmetic.go | 12 +- test/codegen/bool.go | 24 +- test/codegen/shift.go | 18 +- 7 files changed, 120 insertions(+), 249 deletions(-) diff --git a/src/cmd/compile/internal/ssa/_gen/PPC64.rules b/src/cmd/compile/internal/ssa/_gen/PPC64.rules index c5309f16c6..77eb553aed 100644 --- a/src/cmd/compile/internal/ssa/_gen/PPC64.rules +++ b/src/cmd/compile/internal/ssa/_gen/PPC64.rules @@ -323,10 +323,6 @@ (NE (CMPWconst [0] (Select0 (ANDCCconst [1] ((Equal|NotEqual|LessThan|LessEqual|GreaterThan|GreaterEqual) cc)))) yes no) => ((EQ|NE|LT|LE|GT|GE) cc yes no) (NE (CMPWconst [0] (Select0 (ANDCCconst [1] ((FLessThan|FLessEqual|FGreaterThan|FGreaterEqual) cc)))) yes no) => ((FLT|FLE|FGT|FGE) cc yes no) -// Elide compares of bit tests -((EQ|NE) (CMPconst [0] (Select0 (ANDCCconst [c] x))) yes no) => ((EQ|NE) (Select1 (ANDCCconst [c] x)) yes no) -((EQ|NE) (CMPWconst [0] (Select0 (ANDCCconst [c] x))) yes no) => ((EQ|NE) (Select1 (ANDCCconst [c] x)) yes no) - // absorb flag constants into branches (EQ (FlagEQ) yes no) => (First yes no) (EQ (FlagLT) yes no) => (First no yes) @@ -407,8 +403,8 @@ // Elide compares of bit tests -((EQ|NE|LT|LE|GT|GE) (CMPconst [0] (Select0 (ANDCCconst [c] x))) yes no) => ((EQ|NE|LT|LE|GT|GE) (Select1 (ANDCCconst [c] x)) yes no) -((EQ|NE|LT|LE|GT|GE) (CMPWconst [0] (Select0 (ANDCCconst [c] x))) yes no) => ((EQ|NE|LT|LE|GT|GE) (Select1 (ANDCCconst [c] x)) yes no) +((EQ|NE|LT|LE|GT|GE) (CMPconst [0] (Select0 z:(ANDCCconst [c] x))) yes no) => ((EQ|NE|LT|LE|GT|GE) (Select1 z) yes no) +((EQ|NE|LT|LE|GT|GE) (CMPWconst [0] (Select0 z:(ANDCCconst [c] x))) yes no) => ((EQ|NE|LT|LE|GT|GE) (Select1 z) yes no) ((EQ|NE|LT|LE|GT|GE) (CMPconst [0] z:(AND x y)) yes no) && z.Uses == 1 => ((EQ|NE|LT|LE|GT|GE) (Select1 (ANDCC x y)) yes no) ((EQ|NE|LT|LE|GT|GE) (CMPconst [0] z:(OR x y)) yes no) && z.Uses == 1 => ((EQ|NE|LT|LE|GT|GE) (Select1 (ORCC x y)) yes no) ((EQ|NE|LT|LE|GT|GE) (CMPconst [0] z:(XOR x y)) yes no) && z.Uses == 1 => ((EQ|NE|LT|LE|GT|GE) (Select1 (XORCC x y)) yes no) diff --git a/src/cmd/compile/internal/ssa/_gen/PPC64latelower.rules b/src/cmd/compile/internal/ssa/_gen/PPC64latelower.rules index c4af55c328..5980fc922e 100644 --- a/src/cmd/compile/internal/ssa/_gen/PPC64latelower.rules +++ b/src/cmd/compile/internal/ssa/_gen/PPC64latelower.rules @@ -18,6 +18,9 @@ (SETBC [1] cmp) && buildcfg.GOPPC64 <= 9 => (ISELZ [1] (MOVDconst [1]) cmp) (SETBCR [1] cmp) && buildcfg.GOPPC64 <= 9 => (ISELZ [5] (MOVDconst [1]) cmp) +// Avoid using ANDCCconst if the value for CR0 is not needed, since ANDCCconst +// always sets it. +(Select0 z:(ANDCCconst [m] x)) && z.Uses == 1 && isPPC64ValidShiftMask(m) => (RLDICL [encodePPC64RotateMask(0,m,64)] x) // The upper bits of the smaller than register values is undefined. Take advantage of that. (AND x:(MOVDconst [m]) n) && t.Size() <= 2 => (Select0 (ANDCCconst [int64(int16(m))] n)) diff --git a/src/cmd/compile/internal/ssa/rewritePPC64.go b/src/cmd/compile/internal/ssa/rewritePPC64.go index 089e4e614c..1809b0a77b 100644 --- a/src/cmd/compile/internal/ssa/rewritePPC64.go +++ b/src/cmd/compile/internal/ssa/rewritePPC64.go @@ -15271,56 +15271,6 @@ func rewriteBlockPPC64(b *Block) bool { typ := &b.Func.Config.Types switch b.Kind { case BlockPPC64EQ: - // match: (EQ (CMPconst [0] (Select0 (ANDCCconst [c] x))) yes no) - // result: (EQ (Select1 (ANDCCconst [c] x)) yes no) - for b.Controls[0].Op == OpPPC64CMPconst { - v_0 := b.Controls[0] - if auxIntToInt64(v_0.AuxInt) != 0 { - break - } - v_0_0 := v_0.Args[0] - if v_0_0.Op != OpSelect0 { - break - } - v_0_0_0 := v_0_0.Args[0] - if v_0_0_0.Op != OpPPC64ANDCCconst { - break - } - c := auxIntToInt64(v_0_0_0.AuxInt) - x := v_0_0_0.Args[0] - v0 := b.NewValue0(v_0.Pos, OpSelect1, types.TypeFlags) - v1 := b.NewValue0(v_0.Pos, OpPPC64ANDCCconst, types.NewTuple(typ.Int, types.TypeFlags)) - v1.AuxInt = int64ToAuxInt(c) - v1.AddArg(x) - v0.AddArg(v1) - b.resetWithControl(BlockPPC64EQ, v0) - return true - } - // match: (EQ (CMPWconst [0] (Select0 (ANDCCconst [c] x))) yes no) - // result: (EQ (Select1 (ANDCCconst [c] x)) yes no) - for b.Controls[0].Op == OpPPC64CMPWconst { - v_0 := b.Controls[0] - if auxIntToInt32(v_0.AuxInt) != 0 { - break - } - v_0_0 := v_0.Args[0] - if v_0_0.Op != OpSelect0 { - break - } - v_0_0_0 := v_0_0.Args[0] - if v_0_0_0.Op != OpPPC64ANDCCconst { - break - } - c := auxIntToInt64(v_0_0_0.AuxInt) - x := v_0_0_0.Args[0] - v0 := b.NewValue0(v_0.Pos, OpSelect1, types.TypeFlags) - v1 := b.NewValue0(v_0.Pos, OpPPC64ANDCCconst, types.NewTuple(typ.Int, types.TypeFlags)) - v1.AuxInt = int64ToAuxInt(c) - v1.AddArg(x) - v0.AddArg(v1) - b.resetWithControl(BlockPPC64EQ, v0) - return true - } // match: (EQ (FlagEQ) yes no) // result: (First yes no) for b.Controls[0].Op == OpPPC64FlagEQ { @@ -15349,8 +15299,8 @@ func rewriteBlockPPC64(b *Block) bool { b.resetWithControl(BlockPPC64EQ, cmp) return true } - // match: (EQ (CMPconst [0] (Select0 (ANDCCconst [c] x))) yes no) - // result: (EQ (Select1 (ANDCCconst [c] x)) yes no) + // match: (EQ (CMPconst [0] (Select0 z:(ANDCCconst [c] x))) yes no) + // result: (EQ (Select1 z) yes no) for b.Controls[0].Op == OpPPC64CMPconst { v_0 := b.Controls[0] if auxIntToInt64(v_0.AuxInt) != 0 { @@ -15360,22 +15310,17 @@ func rewriteBlockPPC64(b *Block) bool { if v_0_0.Op != OpSelect0 { break } - v_0_0_0 := v_0_0.Args[0] - if v_0_0_0.Op != OpPPC64ANDCCconst { + z := v_0_0.Args[0] + if z.Op != OpPPC64ANDCCconst { break } - c := auxIntToInt64(v_0_0_0.AuxInt) - x := v_0_0_0.Args[0] v0 := b.NewValue0(v_0.Pos, OpSelect1, types.TypeFlags) - v1 := b.NewValue0(v_0.Pos, OpPPC64ANDCCconst, types.NewTuple(typ.Int, types.TypeFlags)) - v1.AuxInt = int64ToAuxInt(c) - v1.AddArg(x) - v0.AddArg(v1) + v0.AddArg(z) b.resetWithControl(BlockPPC64EQ, v0) return true } - // match: (EQ (CMPWconst [0] (Select0 (ANDCCconst [c] x))) yes no) - // result: (EQ (Select1 (ANDCCconst [c] x)) yes no) + // match: (EQ (CMPWconst [0] (Select0 z:(ANDCCconst [c] x))) yes no) + // result: (EQ (Select1 z) yes no) for b.Controls[0].Op == OpPPC64CMPWconst { v_0 := b.Controls[0] if auxIntToInt32(v_0.AuxInt) != 0 { @@ -15385,17 +15330,12 @@ func rewriteBlockPPC64(b *Block) bool { if v_0_0.Op != OpSelect0 { break } - v_0_0_0 := v_0_0.Args[0] - if v_0_0_0.Op != OpPPC64ANDCCconst { + z := v_0_0.Args[0] + if z.Op != OpPPC64ANDCCconst { break } - c := auxIntToInt64(v_0_0_0.AuxInt) - x := v_0_0_0.Args[0] v0 := b.NewValue0(v_0.Pos, OpSelect1, types.TypeFlags) - v1 := b.NewValue0(v_0.Pos, OpPPC64ANDCCconst, types.NewTuple(typ.Int, types.TypeFlags)) - v1.AuxInt = int64ToAuxInt(c) - v1.AddArg(x) - v0.AddArg(v1) + v0.AddArg(z) b.resetWithControl(BlockPPC64EQ, v0) return true } @@ -15517,8 +15457,8 @@ func rewriteBlockPPC64(b *Block) bool { b.resetWithControl(BlockPPC64LE, cmp) return true } - // match: (GE (CMPconst [0] (Select0 (ANDCCconst [c] x))) yes no) - // result: (GE (Select1 (ANDCCconst [c] x)) yes no) + // match: (GE (CMPconst [0] (Select0 z:(ANDCCconst [c] x))) yes no) + // result: (GE (Select1 z) yes no) for b.Controls[0].Op == OpPPC64CMPconst { v_0 := b.Controls[0] if auxIntToInt64(v_0.AuxInt) != 0 { @@ -15528,22 +15468,17 @@ func rewriteBlockPPC64(b *Block) bool { if v_0_0.Op != OpSelect0 { break } - v_0_0_0 := v_0_0.Args[0] - if v_0_0_0.Op != OpPPC64ANDCCconst { + z := v_0_0.Args[0] + if z.Op != OpPPC64ANDCCconst { break } - c := auxIntToInt64(v_0_0_0.AuxInt) - x := v_0_0_0.Args[0] v0 := b.NewValue0(v_0.Pos, OpSelect1, types.TypeFlags) - v1 := b.NewValue0(v_0.Pos, OpPPC64ANDCCconst, types.NewTuple(typ.Int, types.TypeFlags)) - v1.AuxInt = int64ToAuxInt(c) - v1.AddArg(x) - v0.AddArg(v1) + v0.AddArg(z) b.resetWithControl(BlockPPC64GE, v0) return true } - // match: (GE (CMPWconst [0] (Select0 (ANDCCconst [c] x))) yes no) - // result: (GE (Select1 (ANDCCconst [c] x)) yes no) + // match: (GE (CMPWconst [0] (Select0 z:(ANDCCconst [c] x))) yes no) + // result: (GE (Select1 z) yes no) for b.Controls[0].Op == OpPPC64CMPWconst { v_0 := b.Controls[0] if auxIntToInt32(v_0.AuxInt) != 0 { @@ -15553,17 +15488,12 @@ func rewriteBlockPPC64(b *Block) bool { if v_0_0.Op != OpSelect0 { break } - v_0_0_0 := v_0_0.Args[0] - if v_0_0_0.Op != OpPPC64ANDCCconst { + z := v_0_0.Args[0] + if z.Op != OpPPC64ANDCCconst { break } - c := auxIntToInt64(v_0_0_0.AuxInt) - x := v_0_0_0.Args[0] v0 := b.NewValue0(v_0.Pos, OpSelect1, types.TypeFlags) - v1 := b.NewValue0(v_0.Pos, OpPPC64ANDCCconst, types.NewTuple(typ.Int, types.TypeFlags)) - v1.AuxInt = int64ToAuxInt(c) - v1.AddArg(x) - v0.AddArg(v1) + v0.AddArg(z) b.resetWithControl(BlockPPC64GE, v0) return true } @@ -15686,8 +15616,8 @@ func rewriteBlockPPC64(b *Block) bool { b.resetWithControl(BlockPPC64LT, cmp) return true } - // match: (GT (CMPconst [0] (Select0 (ANDCCconst [c] x))) yes no) - // result: (GT (Select1 (ANDCCconst [c] x)) yes no) + // match: (GT (CMPconst [0] (Select0 z:(ANDCCconst [c] x))) yes no) + // result: (GT (Select1 z) yes no) for b.Controls[0].Op == OpPPC64CMPconst { v_0 := b.Controls[0] if auxIntToInt64(v_0.AuxInt) != 0 { @@ -15697,22 +15627,17 @@ func rewriteBlockPPC64(b *Block) bool { if v_0_0.Op != OpSelect0 { break } - v_0_0_0 := v_0_0.Args[0] - if v_0_0_0.Op != OpPPC64ANDCCconst { + z := v_0_0.Args[0] + if z.Op != OpPPC64ANDCCconst { break } - c := auxIntToInt64(v_0_0_0.AuxInt) - x := v_0_0_0.Args[0] v0 := b.NewValue0(v_0.Pos, OpSelect1, types.TypeFlags) - v1 := b.NewValue0(v_0.Pos, OpPPC64ANDCCconst, types.NewTuple(typ.Int, types.TypeFlags)) - v1.AuxInt = int64ToAuxInt(c) - v1.AddArg(x) - v0.AddArg(v1) + v0.AddArg(z) b.resetWithControl(BlockPPC64GT, v0) return true } - // match: (GT (CMPWconst [0] (Select0 (ANDCCconst [c] x))) yes no) - // result: (GT (Select1 (ANDCCconst [c] x)) yes no) + // match: (GT (CMPWconst [0] (Select0 z:(ANDCCconst [c] x))) yes no) + // result: (GT (Select1 z) yes no) for b.Controls[0].Op == OpPPC64CMPWconst { v_0 := b.Controls[0] if auxIntToInt32(v_0.AuxInt) != 0 { @@ -15722,17 +15647,12 @@ func rewriteBlockPPC64(b *Block) bool { if v_0_0.Op != OpSelect0 { break } - v_0_0_0 := v_0_0.Args[0] - if v_0_0_0.Op != OpPPC64ANDCCconst { + z := v_0_0.Args[0] + if z.Op != OpPPC64ANDCCconst { break } - c := auxIntToInt64(v_0_0_0.AuxInt) - x := v_0_0_0.Args[0] v0 := b.NewValue0(v_0.Pos, OpSelect1, types.TypeFlags) - v1 := b.NewValue0(v_0.Pos, OpPPC64ANDCCconst, types.NewTuple(typ.Int, types.TypeFlags)) - v1.AuxInt = int64ToAuxInt(c) - v1.AddArg(x) - v0.AddArg(v1) + v0.AddArg(z) b.resetWithControl(BlockPPC64GT, v0) return true } @@ -15950,8 +15870,8 @@ func rewriteBlockPPC64(b *Block) bool { b.resetWithControl(BlockPPC64GE, cmp) return true } - // match: (LE (CMPconst [0] (Select0 (ANDCCconst [c] x))) yes no) - // result: (LE (Select1 (ANDCCconst [c] x)) yes no) + // match: (LE (CMPconst [0] (Select0 z:(ANDCCconst [c] x))) yes no) + // result: (LE (Select1 z) yes no) for b.Controls[0].Op == OpPPC64CMPconst { v_0 := b.Controls[0] if auxIntToInt64(v_0.AuxInt) != 0 { @@ -15961,22 +15881,17 @@ func rewriteBlockPPC64(b *Block) bool { if v_0_0.Op != OpSelect0 { break } - v_0_0_0 := v_0_0.Args[0] - if v_0_0_0.Op != OpPPC64ANDCCconst { + z := v_0_0.Args[0] + if z.Op != OpPPC64ANDCCconst { break } - c := auxIntToInt64(v_0_0_0.AuxInt) - x := v_0_0_0.Args[0] v0 := b.NewValue0(v_0.Pos, OpSelect1, types.TypeFlags) - v1 := b.NewValue0(v_0.Pos, OpPPC64ANDCCconst, types.NewTuple(typ.Int, types.TypeFlags)) - v1.AuxInt = int64ToAuxInt(c) - v1.AddArg(x) - v0.AddArg(v1) + v0.AddArg(z) b.resetWithControl(BlockPPC64LE, v0) return true } - // match: (LE (CMPWconst [0] (Select0 (ANDCCconst [c] x))) yes no) - // result: (LE (Select1 (ANDCCconst [c] x)) yes no) + // match: (LE (CMPWconst [0] (Select0 z:(ANDCCconst [c] x))) yes no) + // result: (LE (Select1 z) yes no) for b.Controls[0].Op == OpPPC64CMPWconst { v_0 := b.Controls[0] if auxIntToInt32(v_0.AuxInt) != 0 { @@ -15986,17 +15901,12 @@ func rewriteBlockPPC64(b *Block) bool { if v_0_0.Op != OpSelect0 { break } - v_0_0_0 := v_0_0.Args[0] - if v_0_0_0.Op != OpPPC64ANDCCconst { + z := v_0_0.Args[0] + if z.Op != OpPPC64ANDCCconst { break } - c := auxIntToInt64(v_0_0_0.AuxInt) - x := v_0_0_0.Args[0] v0 := b.NewValue0(v_0.Pos, OpSelect1, types.TypeFlags) - v1 := b.NewValue0(v_0.Pos, OpPPC64ANDCCconst, types.NewTuple(typ.Int, types.TypeFlags)) - v1.AuxInt = int64ToAuxInt(c) - v1.AddArg(x) - v0.AddArg(v1) + v0.AddArg(z) b.resetWithControl(BlockPPC64LE, v0) return true } @@ -16119,8 +16029,8 @@ func rewriteBlockPPC64(b *Block) bool { b.resetWithControl(BlockPPC64GT, cmp) return true } - // match: (LT (CMPconst [0] (Select0 (ANDCCconst [c] x))) yes no) - // result: (LT (Select1 (ANDCCconst [c] x)) yes no) + // match: (LT (CMPconst [0] (Select0 z:(ANDCCconst [c] x))) yes no) + // result: (LT (Select1 z) yes no) for b.Controls[0].Op == OpPPC64CMPconst { v_0 := b.Controls[0] if auxIntToInt64(v_0.AuxInt) != 0 { @@ -16130,22 +16040,17 @@ func rewriteBlockPPC64(b *Block) bool { if v_0_0.Op != OpSelect0 { break } - v_0_0_0 := v_0_0.Args[0] - if v_0_0_0.Op != OpPPC64ANDCCconst { + z := v_0_0.Args[0] + if z.Op != OpPPC64ANDCCconst { break } - c := auxIntToInt64(v_0_0_0.AuxInt) - x := v_0_0_0.Args[0] v0 := b.NewValue0(v_0.Pos, OpSelect1, types.TypeFlags) - v1 := b.NewValue0(v_0.Pos, OpPPC64ANDCCconst, types.NewTuple(typ.Int, types.TypeFlags)) - v1.AuxInt = int64ToAuxInt(c) - v1.AddArg(x) - v0.AddArg(v1) + v0.AddArg(z) b.resetWithControl(BlockPPC64LT, v0) return true } - // match: (LT (CMPWconst [0] (Select0 (ANDCCconst [c] x))) yes no) - // result: (LT (Select1 (ANDCCconst [c] x)) yes no) + // match: (LT (CMPWconst [0] (Select0 z:(ANDCCconst [c] x))) yes no) + // result: (LT (Select1 z) yes no) for b.Controls[0].Op == OpPPC64CMPWconst { v_0 := b.Controls[0] if auxIntToInt32(v_0.AuxInt) != 0 { @@ -16155,17 +16060,12 @@ func rewriteBlockPPC64(b *Block) bool { if v_0_0.Op != OpSelect0 { break } - v_0_0_0 := v_0_0.Args[0] - if v_0_0_0.Op != OpPPC64ANDCCconst { + z := v_0_0.Args[0] + if z.Op != OpPPC64ANDCCconst { break } - c := auxIntToInt64(v_0_0_0.AuxInt) - x := v_0_0_0.Args[0] v0 := b.NewValue0(v_0.Pos, OpSelect1, types.TypeFlags) - v1 := b.NewValue0(v_0.Pos, OpPPC64ANDCCconst, types.NewTuple(typ.Int, types.TypeFlags)) - v1.AuxInt = int64ToAuxInt(c) - v1.AddArg(x) - v0.AddArg(v1) + v0.AddArg(z) b.resetWithControl(BlockPPC64LT, v0) return true } @@ -16490,56 +16390,6 @@ func rewriteBlockPPC64(b *Block) bool { b.resetWithControl(BlockPPC64FGE, cc) return true } - // match: (NE (CMPconst [0] (Select0 (ANDCCconst [c] x))) yes no) - // result: (NE (Select1 (ANDCCconst [c] x)) yes no) - for b.Controls[0].Op == OpPPC64CMPconst { - v_0 := b.Controls[0] - if auxIntToInt64(v_0.AuxInt) != 0 { - break - } - v_0_0 := v_0.Args[0] - if v_0_0.Op != OpSelect0 { - break - } - v_0_0_0 := v_0_0.Args[0] - if v_0_0_0.Op != OpPPC64ANDCCconst { - break - } - c := auxIntToInt64(v_0_0_0.AuxInt) - x := v_0_0_0.Args[0] - v0 := b.NewValue0(v_0.Pos, OpSelect1, types.TypeFlags) - v1 := b.NewValue0(v_0.Pos, OpPPC64ANDCCconst, types.NewTuple(typ.Int, types.TypeFlags)) - v1.AuxInt = int64ToAuxInt(c) - v1.AddArg(x) - v0.AddArg(v1) - b.resetWithControl(BlockPPC64NE, v0) - return true - } - // match: (NE (CMPWconst [0] (Select0 (ANDCCconst [c] x))) yes no) - // result: (NE (Select1 (ANDCCconst [c] x)) yes no) - for b.Controls[0].Op == OpPPC64CMPWconst { - v_0 := b.Controls[0] - if auxIntToInt32(v_0.AuxInt) != 0 { - break - } - v_0_0 := v_0.Args[0] - if v_0_0.Op != OpSelect0 { - break - } - v_0_0_0 := v_0_0.Args[0] - if v_0_0_0.Op != OpPPC64ANDCCconst { - break - } - c := auxIntToInt64(v_0_0_0.AuxInt) - x := v_0_0_0.Args[0] - v0 := b.NewValue0(v_0.Pos, OpSelect1, types.TypeFlags) - v1 := b.NewValue0(v_0.Pos, OpPPC64ANDCCconst, types.NewTuple(typ.Int, types.TypeFlags)) - v1.AuxInt = int64ToAuxInt(c) - v1.AddArg(x) - v0.AddArg(v1) - b.resetWithControl(BlockPPC64NE, v0) - return true - } // match: (NE (FlagEQ) yes no) // result: (First no yes) for b.Controls[0].Op == OpPPC64FlagEQ { @@ -16567,8 +16417,8 @@ func rewriteBlockPPC64(b *Block) bool { b.resetWithControl(BlockPPC64NE, cmp) return true } - // match: (NE (CMPconst [0] (Select0 (ANDCCconst [c] x))) yes no) - // result: (NE (Select1 (ANDCCconst [c] x)) yes no) + // match: (NE (CMPconst [0] (Select0 z:(ANDCCconst [c] x))) yes no) + // result: (NE (Select1 z) yes no) for b.Controls[0].Op == OpPPC64CMPconst { v_0 := b.Controls[0] if auxIntToInt64(v_0.AuxInt) != 0 { @@ -16578,22 +16428,17 @@ func rewriteBlockPPC64(b *Block) bool { if v_0_0.Op != OpSelect0 { break } - v_0_0_0 := v_0_0.Args[0] - if v_0_0_0.Op != OpPPC64ANDCCconst { + z := v_0_0.Args[0] + if z.Op != OpPPC64ANDCCconst { break } - c := auxIntToInt64(v_0_0_0.AuxInt) - x := v_0_0_0.Args[0] v0 := b.NewValue0(v_0.Pos, OpSelect1, types.TypeFlags) - v1 := b.NewValue0(v_0.Pos, OpPPC64ANDCCconst, types.NewTuple(typ.Int, types.TypeFlags)) - v1.AuxInt = int64ToAuxInt(c) - v1.AddArg(x) - v0.AddArg(v1) + v0.AddArg(z) b.resetWithControl(BlockPPC64NE, v0) return true } - // match: (NE (CMPWconst [0] (Select0 (ANDCCconst [c] x))) yes no) - // result: (NE (Select1 (ANDCCconst [c] x)) yes no) + // match: (NE (CMPWconst [0] (Select0 z:(ANDCCconst [c] x))) yes no) + // result: (NE (Select1 z) yes no) for b.Controls[0].Op == OpPPC64CMPWconst { v_0 := b.Controls[0] if auxIntToInt32(v_0.AuxInt) != 0 { @@ -16603,17 +16448,12 @@ func rewriteBlockPPC64(b *Block) bool { if v_0_0.Op != OpSelect0 { break } - v_0_0_0 := v_0_0.Args[0] - if v_0_0_0.Op != OpPPC64ANDCCconst { + z := v_0_0.Args[0] + if z.Op != OpPPC64ANDCCconst { break } - c := auxIntToInt64(v_0_0_0.AuxInt) - x := v_0_0_0.Args[0] v0 := b.NewValue0(v_0.Pos, OpSelect1, types.TypeFlags) - v1 := b.NewValue0(v_0.Pos, OpPPC64ANDCCconst, types.NewTuple(typ.Int, types.TypeFlags)) - v1.AuxInt = int64ToAuxInt(c) - v1.AddArg(x) - v0.AddArg(v1) + v0.AddArg(z) b.resetWithControl(BlockPPC64NE, v0) return true } diff --git a/src/cmd/compile/internal/ssa/rewritePPC64latelower.go b/src/cmd/compile/internal/ssa/rewritePPC64latelower.go index d17e695e5a..8b22a7d02f 100644 --- a/src/cmd/compile/internal/ssa/rewritePPC64latelower.go +++ b/src/cmd/compile/internal/ssa/rewritePPC64latelower.go @@ -17,6 +17,8 @@ func rewriteValuePPC64latelower(v *Value) bool { return rewriteValuePPC64latelower_OpPPC64SETBC(v) case OpPPC64SETBCR: return rewriteValuePPC64latelower_OpPPC64SETBCR(v) + case OpSelect0: + return rewriteValuePPC64latelower_OpSelect0(v) } return false } @@ -292,6 +294,28 @@ func rewriteValuePPC64latelower_OpPPC64SETBCR(v *Value) bool { } return false } +func rewriteValuePPC64latelower_OpSelect0(v *Value) bool { + v_0 := v.Args[0] + // match: (Select0 z:(ANDCCconst [m] x)) + // cond: z.Uses == 1 && isPPC64ValidShiftMask(m) + // result: (RLDICL [encodePPC64RotateMask(0,m,64)] x) + for { + z := v_0 + if z.Op != OpPPC64ANDCCconst { + break + } + m := auxIntToInt64(z.AuxInt) + x := z.Args[0] + if !(z.Uses == 1 && isPPC64ValidShiftMask(m)) { + break + } + v.reset(OpPPC64RLDICL) + v.AuxInt = int64ToAuxInt(encodePPC64RotateMask(0, m, 64)) + v.AddArg(x) + return true + } + return false +} func rewriteBlockPPC64latelower(b *Block) bool { return false } diff --git a/test/codegen/arithmetic.go b/test/codegen/arithmetic.go index f381b34ade..b91a904be9 100644 --- a/test/codegen/arithmetic.go +++ b/test/codegen/arithmetic.go @@ -260,7 +260,7 @@ func Pow2Mods(n1 uint, n2 int) (uint, int) { // amd64:"ANDL\t[$]31",-"DIVQ" // arm:"AND\t[$]31",-".*udiv" // arm64:"AND\t[$]31",-"UDIV" - // ppc64x:"ANDCC\t[$]31" + // ppc64x:"RLDICL" a := n1 % 32 // unsigned // 386:"SHRL",-"IDIVL" @@ -279,14 +279,14 @@ func Pow2DivisibleSigned(n1, n2 int) (bool, bool) { // amd64:"TESTQ\t[$]63",-"DIVQ",-"SHRQ" // arm:"AND\t[$]63",-".*udiv",-"SRA" // arm64:"TST\t[$]63",-"UDIV",-"ASR",-"AND" - // ppc64x:"ANDCC\t[$]63",-"SRAD" + // ppc64x:"RLDICL",-"SRAD" a := n1%64 == 0 // signed divisible // 386:"TESTL\t[$]63",-"DIVL",-"SHRL" // amd64:"TESTQ\t[$]63",-"DIVQ",-"SHRQ" // arm:"AND\t[$]63",-".*udiv",-"SRA" // arm64:"TST\t[$]63",-"UDIV",-"ASR",-"AND" - // ppc64x:"ANDCC\t[$]63",-"SRAD" + // ppc64x:"RLDICL",-"SRAD" b := n2%64 != 0 // signed indivisible return a, b @@ -464,7 +464,7 @@ func LenMod1(a []int) int { // arm64:"AND\t[$]1023",-"SDIV" // arm/6:"AND",-".*udiv" // arm/7:"BFC",-".*udiv",-"AND" - // ppc64x:"ANDCC\t[$]1023" + // ppc64x:"RLDICL" return len(a) % 1024 } @@ -474,7 +474,7 @@ func LenMod2(s string) int { // arm64:"AND\t[$]2047",-"SDIV" // arm/6:"AND",-".*udiv" // arm/7:"BFC",-".*udiv",-"AND" - // ppc64x:"ANDCC\t[$]2047" + // ppc64x:"RLDICL" return len(s) % (4097 >> 1) } @@ -493,7 +493,7 @@ func CapMod(a []int) int { // arm64:"AND\t[$]4095",-"SDIV" // arm/6:"AND",-".*udiv" // arm/7:"BFC",-".*udiv",-"AND" - // ppc64x:"ANDCC\t[$]4095" + // ppc64x:"RLDICL" return cap(a) % ((1 << 11) + 2048) } diff --git a/test/codegen/bool.go b/test/codegen/bool.go index faf7033a2a..109c3aa0cd 100644 --- a/test/codegen/bool.go +++ b/test/codegen/bool.go @@ -10,49 +10,49 @@ package codegen func convertNeq0B(x uint8, c bool) bool { // amd64:"ANDL\t[$]1",-"SETNE" - // ppc64x:"ANDCC",-"CMPW",-"ISEL" + // ppc64x:"RLDICL",-"CMPW",-"ISEL" b := x&1 != 0 return c && b } func convertNeq0W(x uint16, c bool) bool { // amd64:"ANDL\t[$]1",-"SETNE" - // ppc64x:"ANDCC",-"CMPW",-"ISEL" + // ppc64x:"RLDICL",-"CMPW",-"ISEL" b := x&1 != 0 return c && b } func convertNeq0L(x uint32, c bool) bool { // amd64:"ANDL\t[$]1",-"SETB" - // ppc64x:"ANDCC",-"CMPW",-"ISEL" + // ppc64x:"RLDICL",-"CMPW",-"ISEL" b := x&1 != 0 return c && b } func convertNeq0Q(x uint64, c bool) bool { // amd64:"ANDL\t[$]1",-"SETB" - // ppc64x:"ANDCC",-"CMP",-"ISEL" + // ppc64x:"RLDICL",-"CMP",-"ISEL" b := x&1 != 0 return c && b } func convertNeqBool32(x uint32) bool { - // ppc64x:"ANDCC",-"CMPW",-"ISEL" + // ppc64x:"RLDICL",-"CMPW",-"ISEL" return x&1 != 0 } func convertEqBool32(x uint32) bool { - // ppc64x:"ANDCC",-"CMPW","XOR",-"ISEL" + // ppc64x:"RLDICL",-"CMPW","XOR",-"ISEL" return x&1 == 0 } func convertNeqBool64(x uint64) bool { - // ppc64x:"ANDCC",-"CMP",-"ISEL" + // ppc64x:"RLDICL",-"CMP",-"ISEL" return x&1 != 0 } func convertEqBool64(x uint64) bool { - // ppc64x:"ANDCC","XOR",-"CMP",-"ISEL" + // ppc64x:"RLDICL","XOR",-"CMP",-"ISEL" return x&1 == 0 } @@ -211,3 +211,11 @@ func TestSetInvGeFp64(x float64, y float64) bool { b := !(x >= y) return b } +func TestAndCompareZero(x uint64, y uint64) uint64 { + // ppc64x:"ANDCC" + b := x&3 + if b!=0 { + return b + } + return b+8 +} diff --git a/test/codegen/shift.go b/test/codegen/shift.go index 302560d5b0..bf8b633905 100644 --- a/test/codegen/shift.go +++ b/test/codegen/shift.go @@ -70,7 +70,7 @@ func rshConst64x32(v int64) int64 { func lshMask64x64(v int64, s uint64) int64 { // arm64:"LSL",-"AND" - // ppc64x:"ANDCC",-"ORN",-"ISEL" + // ppc64x:"RLDICL",-"ORN",-"ISEL" // riscv64:"SLL",-"AND\t",-"SLTIU" // s390x:-"RISBGZ",-"AND",-"LOCGR" return v << (s & 63) @@ -78,7 +78,7 @@ func lshMask64x64(v int64, s uint64) int64 { func rshMask64Ux64(v uint64, s uint64) uint64 { // arm64:"LSR",-"AND",-"CSEL" - // ppc64x:"ANDCC",-"ORN",-"ISEL" + // ppc64x:"RLDICL",-"ORN",-"ISEL" // riscv64:"SRL\t",-"AND\t",-"SLTIU" // s390x:-"RISBGZ",-"AND",-"LOCGR" return v >> (s & 63) @@ -86,7 +86,7 @@ func rshMask64Ux64(v uint64, s uint64) uint64 { func rshMask64x64(v int64, s uint64) int64 { // arm64:"ASR",-"AND",-"CSEL" - // ppc64x:"ANDCC",-"ORN",-"ISEL" + // ppc64x:"RLDICL",-"ORN",-"ISEL" // riscv64:"SRA",-"OR",-"SLTIU" // s390x:-"RISBGZ",-"AND",-"LOCGR" return v >> (s & 63) @@ -123,7 +123,7 @@ func rshMask32x64(v int32, s uint64) int32 { func lshMask64x32(v int64, s uint32) int64 { // arm64:"LSL",-"AND" - // ppc64x:"ANDCC",-"ORN" + // ppc64x:"RLDICL",-"ORN" // riscv64:"SLL",-"AND\t",-"SLTIU" // s390x:-"RISBGZ",-"AND",-"LOCGR" return v << (s & 63) @@ -131,7 +131,7 @@ func lshMask64x32(v int64, s uint32) int64 { func rshMask64Ux32(v uint64, s uint32) uint64 { // arm64:"LSR",-"AND",-"CSEL" - // ppc64x:"ANDCC",-"ORN" + // ppc64x:"RLDICL",-"ORN" // riscv64:"SRL\t",-"AND\t",-"SLTIU" // s390x:-"RISBGZ",-"AND",-"LOCGR" return v >> (s & 63) @@ -139,28 +139,28 @@ func rshMask64Ux32(v uint64, s uint32) uint64 { func rshMask64x32(v int64, s uint32) int64 { // arm64:"ASR",-"AND",-"CSEL" - // ppc64x:"ANDCC",-"ORN",-"ISEL" + // ppc64x:"RLDICL",-"ORN",-"ISEL" // riscv64:"SRA",-"OR",-"SLTIU" // s390x:-"RISBGZ",-"AND",-"LOCGR" return v >> (s & 63) } func lshMask64x32Ext(v int64, s int32) int64 { - // ppc64x:"ANDCC",-"ORN",-"ISEL" + // ppc64x:"RLDICL",-"ORN",-"ISEL" // riscv64:"SLL",-"AND\t",-"SLTIU" // s390x:-"RISBGZ",-"AND",-"LOCGR" return v << uint(s&63) } func rshMask64Ux32Ext(v uint64, s int32) uint64 { - // ppc64x:"ANDCC",-"ORN",-"ISEL" + // ppc64x:"RLDICL",-"ORN",-"ISEL" // riscv64:"SRL\t",-"AND\t",-"SLTIU" // s390x:-"RISBGZ",-"AND",-"LOCGR" return v >> uint(s&63) } func rshMask64x32Ext(v int64, s int32) int64 { - // ppc64x:"ANDCC",-"ORN",-"ISEL" + // ppc64x:"RLDICL",-"ORN",-"ISEL" // riscv64:"SRA",-"OR",-"SLTIU" // s390x:-"RISBGZ",-"AND",-"LOCGR" return v >> uint(s&63) -- 2.44.0