Revert "compile: prefer an AND instead of SHR+SHL instructions"

This reverts commit 9ec7074a946b7c2812a1a044e84276a36f46d14d.

Reason for revert: broke s390x (copysign, abs) and arm64 (bitfield) tests.

Change-Id: I16c1b389c062e8c4aa5de079f1d46c9b25b0db52
Reviewed-on: https://go-review.googlesource.com/c/go/+/193850
Run-TryBot: Martin Möhrmann <moehrmann@google.com>
Reviewed-by: Agniva De Sarker <agniva.quicksilver@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

3 files changed, 121 insertions, 6 deletions

diff --git a/src/cmd/compile/internal/ssa/gen/ARM64.rules b/src/cmd/compile/internal/ssa/gen/ARM64.rules
index 6539a0ce7b..d4b47bfb0b 100644
--- a/src/cmd/compile/internal/ssa/gen/ARM64.rules
+++ b/src/cmd/compile/internal/ssa/gen/ARM64.rules
@@ -1863,8 +1863,9 @@
 (XORshiftLL <t> [c] (UBFX [bfc] x) x2) && c < 32 && t.Size() == 4 && bfc == armBFAuxInt(32-c, c)
 	-> (EXTRWconst [32-c] x2 x)
 
-// Rewrite special pairs of shifts to AND.
-// On ARM64 the bitmask can fit into an instruction.
+// Generic rules rewrite certain AND to a pair of shifts.
+// However, on ARM64 the bitmask can fit into an instruction.
+// Rewrite it back to AND.
 (SRLconst [c] (SLLconst [c] x)) && 0 < c && c < 64 -> (ANDconst [1<<uint(64-c)-1] x) // mask out high bits
 (SLLconst [c] (SRLconst [c] x)) && 0 < c && c < 64 -> (ANDconst [^(1<<uint(c)-1)] x) // mask out low bits
 
diff --git a/src/cmd/compile/internal/ssa/gen/generic.rules b/src/cmd/compile/internal/ssa/gen/generic.rules
index 8696464a70..ef5d7a63ff 100644
--- a/src/cmd/compile/internal/ssa/gen/generic.rules
+++ b/src/cmd/compile/internal/ssa/gen/generic.rules
@@ -542,6 +542,14 @@
 (Slicemask (Const64 [x])) && x > 0 -> (Const64 [-1])
 (Slicemask (Const64 [0]))          -> (Const64 [0])
 
+// Rewrite AND of consts as shifts if possible, slightly faster for 64 bit operands
+// leading zeros can be shifted left, then right
+(And64 <t> (Const64 [y]) x) && nlz(y) + nto(y) == 64 && nto(y) >= 32
+  -> (Rsh64Ux64 (Lsh64x64 <t> x (Const64 <t> [nlz(y)])) (Const64 <t> [nlz(y)]))
+// trailing zeros can be shifted right, then left
+(And64 <t> (Const64 [y]) x) && nlo(y) + ntz(y) == 64 && ntz(y) >= 32
+  -> (Lsh64x64 (Rsh64Ux64 <t> x (Const64 <t> [ntz(y)])) (Const64 <t> [ntz(y)]))
+
 // simplifications often used for lengths.  e.g. len(s[i:i+5])==5
 (Sub(64|32|16|8) (Add(64|32|16|8) x y) x) -> y
 (Sub(64|32|16|8) (Add(64|32|16|8) x y) y) -> x
diff --git a/src/cmd/compile/internal/ssa/rewritegeneric.go b/src/cmd/compile/internal/ssa/rewritegeneric.go
index a2d091d3d6..8aa07d20db 100644
--- a/src/cmd/compile/internal/ssa/rewritegeneric.go
+++ b/src/cmd/compile/internal/ssa/rewritegeneric.go
@@ -5735,6 +5735,112 @@ func rewriteValuegeneric_OpAnd64_10(v *Value) bool {
 		v.AddArg(y)
 		return true
 	}
+	// match: (And64 <t> (Const64 [y]) x)
+	// cond: nlz(y) + nto(y) == 64 && nto(y) >= 32
+	// result: (Rsh64Ux64 (Lsh64x64 <t> x (Const64 <t> [nlz(y)])) (Const64 <t> [nlz(y)]))
+	for {
+		t := v.Type
+		x := v.Args[1]
+		v_0 := v.Args[0]
+		if v_0.Op != OpConst64 {
+			break
+		}
+		y := v_0.AuxInt
+		if !(nlz(y)+nto(y) == 64 && nto(y) >= 32) {
+			break
+		}
+		v.reset(OpRsh64Ux64)
+		v0 := b.NewValue0(v.Pos, OpLsh64x64, t)
+		v0.AddArg(x)
+		v1 := b.NewValue0(v.Pos, OpConst64, t)
+		v1.AuxInt = nlz(y)
+		v0.AddArg(v1)
+		v.AddArg(v0)
+		v2 := b.NewValue0(v.Pos, OpConst64, t)
+		v2.AuxInt = nlz(y)
+		v.AddArg(v2)
+		return true
+	}
+	// match: (And64 <t> x (Const64 [y]))
+	// cond: nlz(y) + nto(y) == 64 && nto(y) >= 32
+	// result: (Rsh64Ux64 (Lsh64x64 <t> x (Const64 <t> [nlz(y)])) (Const64 <t> [nlz(y)]))
+	for {
+		t := v.Type
+		_ = v.Args[1]
+		x := v.Args[0]
+		v_1 := v.Args[1]
+		if v_1.Op != OpConst64 {
+			break
+		}
+		y := v_1.AuxInt
+		if !(nlz(y)+nto(y) == 64 && nto(y) >= 32) {
+			break
+		}
+		v.reset(OpRsh64Ux64)
+		v0 := b.NewValue0(v.Pos, OpLsh64x64, t)
+		v0.AddArg(x)
+		v1 := b.NewValue0(v.Pos, OpConst64, t)
+		v1.AuxInt = nlz(y)
+		v0.AddArg(v1)
+		v.AddArg(v0)
+		v2 := b.NewValue0(v.Pos, OpConst64, t)
+		v2.AuxInt = nlz(y)
+		v.AddArg(v2)
+		return true
+	}
+	// match: (And64 <t> (Const64 [y]) x)
+	// cond: nlo(y) + ntz(y) == 64 && ntz(y) >= 32
+	// result: (Lsh64x64 (Rsh64Ux64 <t> x (Const64 <t> [ntz(y)])) (Const64 <t> [ntz(y)]))
+	for {
+		t := v.Type
+		x := v.Args[1]
+		v_0 := v.Args[0]
+		if v_0.Op != OpConst64 {
+			break
+		}
+		y := v_0.AuxInt
+		if !(nlo(y)+ntz(y) == 64 && ntz(y) >= 32) {
+			break
+		}
+		v.reset(OpLsh64x64)
+		v0 := b.NewValue0(v.Pos, OpRsh64Ux64, t)
+		v0.AddArg(x)
+		v1 := b.NewValue0(v.Pos, OpConst64, t)
+		v1.AuxInt = ntz(y)
+		v0.AddArg(v1)
+		v.AddArg(v0)
+		v2 := b.NewValue0(v.Pos, OpConst64, t)
+		v2.AuxInt = ntz(y)
+		v.AddArg(v2)
+		return true
+	}
+	// match: (And64 <t> x (Const64 [y]))
+	// cond: nlo(y) + ntz(y) == 64 && ntz(y) >= 32
+	// result: (Lsh64x64 (Rsh64Ux64 <t> x (Const64 <t> [ntz(y)])) (Const64 <t> [ntz(y)]))
+	for {
+		t := v.Type
+		_ = v.Args[1]
+		x := v.Args[0]
+		v_1 := v.Args[1]
+		if v_1.Op != OpConst64 {
+			break
+		}
+		y := v_1.AuxInt
+		if !(nlo(y)+ntz(y) == 64 && ntz(y) >= 32) {
+			break
+		}
+		v.reset(OpLsh64x64)
+		v0 := b.NewValue0(v.Pos, OpRsh64Ux64, t)
+		v0.AddArg(x)
+		v1 := b.NewValue0(v.Pos, OpConst64, t)
+		v1.AuxInt = ntz(y)
+		v0.AddArg(v1)
+		v.AddArg(v0)
+		v2 := b.NewValue0(v.Pos, OpConst64, t)
+		v2.AuxInt = ntz(y)
+		v.AddArg(v2)
+		return true
+	}
 	// match: (And64 (And64 i:(Const64 <t>) z) x)
 	// cond: (z.Op != OpConst64 && x.Op != OpConst64)
 	// result: (And64 i (And64 <t> z x))
@@ -5761,6 +5867,10 @@ func rewriteValuegeneric_OpAnd64_10(v *Value) bool {
 		v.AddArg(v0)
 		return true
 	}
+	return false
+}
+func rewriteValuegeneric_OpAnd64_20(v *Value) bool {
+	b := v.Block
 	// match: (And64 (And64 z i:(Const64 <t>)) x)
 	// cond: (z.Op != OpConst64 && x.Op != OpConst64)
 	// result: (And64 i (And64 <t> z x))
@@ -5874,10 +5984,6 @@ func rewriteValuegeneric_OpAnd64_10(v *Value) bool {
 		v.AddArg(x)
 		return true
 	}
-	return false
-}
-func rewriteValuegeneric_OpAnd64_20(v *Value) bool {
-	b := v.Block
 	// match: (And64 (Const64 <t> [c]) (And64 x (Const64 <t> [d])))
 	// cond:
 	// result: (And64 (Const64 <t> [c&d]) x)