From d292f77e95fd9afdbfcfa12c9552d5926cdde8b1 Mon Sep 17 00:00:00 2001
From: Josh Bleecher Snyder <josharian@gmail.com>
Date: Mon, 23 Apr 2018 13:49:51 -0700
Subject: [PATCH] cmd/compile: rewrite 2*x+c into LEAx1 on amd64
Rewrite x<<1+c into x+x+c, which can be expressed as a single LEAQ/LEAL.
Bit of a special case, but the single-instruction
LEA is both shorter and faster than SHL then ADD.
Triggers 293 times during make.bash.
Change-Id: I3f09c8e9a8f3859d1eeed336f095fc3ada79c2c1
Reviewed-on: https://go-review.googlesource.com/108938
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
---
src/cmd/compile/internal/ssa/gen/AMD64.rules | 1 +
src/cmd/compile/internal/ssa/rewriteAMD64.go | 43 +++++++++++++++++++-
test/codegen/arithmetic.go | 5 +++
3 files changed, 48 insertions(+), 1 deletion(-)
diff --git a/src/cmd/compile/internal/ssa/gen/AMD64.rules b/src/cmd/compile/internal/ssa/gen/AMD64.rules
index 482a1558dcc..cab0f660799 100644
--- a/src/cmd/compile/internal/ssa/gen/AMD64.rules
+++ b/src/cmd/compile/internal/ssa/gen/AMD64.rules
@@ -925,6 +925,7 @@
// combine ADDQ/ADDQconst into LEAQ1
(ADDQconst [c] (ADDQ x y)) -> (LEAQ1 [c] x y)
(ADDQ (ADDQconst [c] x) y) -> (LEAQ1 [c] x y)
+(ADD(Q|L)const [c] (SHL(Q|L)const [1] x)) -> (LEA(Q|L)1 [c] x x)
// fold ADDQ into LEAQ
(ADDQconst [c] (LEAQ [d] {s} x)) && is32Bit(c+d) -> (LEAQ [c+d] {s} x)
diff --git a/src/cmd/compile/internal/ssa/rewriteAMD64.go b/src/cmd/compile/internal/ssa/rewriteAMD64.go
index e51a25527c6..313b6bef9c3 100644
--- a/src/cmd/compile/internal/ssa/rewriteAMD64.go
+++ b/src/cmd/compile/internal/ssa/rewriteAMD64.go
@@ -26,7 +26,7 @@ func rewriteValueAMD64(v *Value) bool {
case OpAMD64ADDQ:
return rewriteValueAMD64_OpAMD64ADDQ_0(v) || rewriteValueAMD64_OpAMD64ADDQ_10(v) || rewriteValueAMD64_OpAMD64ADDQ_20(v)
case OpAMD64ADDQconst:
- return rewriteValueAMD64_OpAMD64ADDQconst_0(v)
+ return rewriteValueAMD64_OpAMD64ADDQconst_0(v) || rewriteValueAMD64_OpAMD64ADDQconst_10(v)
case OpAMD64ADDQconstmem:
return rewriteValueAMD64_OpAMD64ADDQconstmem_0(v)
case OpAMD64ADDQmem:
@@ -1313,6 +1313,25 @@ func rewriteValueAMD64_OpAMD64ADDL_10(v *Value) bool {
return false
}
func rewriteValueAMD64_OpAMD64ADDLconst_0(v *Value) bool {
+ // match: (ADDLconst [c] (SHLLconst [1] x))
+ // cond:
+ // result: (LEAL1 [c] x x)
+ for {
+ c := v.AuxInt
+ v_0 := v.Args[0]
+ if v_0.Op != OpAMD64SHLLconst {
+ break
+ }
+ if v_0.AuxInt != 1 {
+ break
+ }
+ x := v_0.Args[0]
+ v.reset(OpAMD64LEAL1)
+ v.AuxInt = c
+ v.AddArg(x)
+ v.AddArg(x)
+ return true
+ }
// match: (ADDLconst [c] x)
// cond: int32(c)==0
// result: x
@@ -2077,6 +2096,25 @@ func rewriteValueAMD64_OpAMD64ADDQconst_0(v *Value) bool {
v.AddArg(y)
return true
}
+ // match: (ADDQconst [c] (SHLQconst [1] x))
+ // cond:
+ // result: (LEAQ1 [c] x x)
+ for {
+ c := v.AuxInt
+ v_0 := v.Args[0]
+ if v_0.Op != OpAMD64SHLQconst {
+ break
+ }
+ if v_0.AuxInt != 1 {
+ break
+ }
+ x := v_0.Args[0]
+ v.reset(OpAMD64LEAQ1)
+ v.AuxInt = c
+ v.AddArg(x)
+ v.AddArg(x)
+ return true
+ }
// match: (ADDQconst [c] (LEAQ [d] {s} x))
// cond: is32Bit(c+d)
// result: (LEAQ [c+d] {s} x)
@@ -2240,6 +2278,9 @@ func rewriteValueAMD64_OpAMD64ADDQconst_0(v *Value) bool {
v.AddArg(x)
return true
}
+ return false
+}
+func rewriteValueAMD64_OpAMD64ADDQconst_10(v *Value) bool {
// match: (ADDQconst [off] x:(SP))
// cond:
// result: (LEAQ [off] x)
diff --git a/test/codegen/arithmetic.go b/test/codegen/arithmetic.go
index ce7a7c27f4a..3948ef301d6 100644
--- a/test/codegen/arithmetic.go
+++ b/test/codegen/arithmetic.go
@@ -173,3 +173,8 @@ func CapMod(a []int) int {
// amd64:"ANDQ\t[$]4095"
return cap(a) % ((1 << 11) + 2048)
}
+
+func AddMul(x int) int {
+ // amd64:"LEAQ\t1"
+ return 2*x + 1
+}
--
GitLab