cmd/compile: use MOVQ + MOVL for 11 byte moves

Instead of using two MOVQ, use a shorter MOVL for the remaining 3 bytes.

compilecmp (linux/amd64):

runtime/pprof
runtime/pprof.printCountProfile.func1 875 -> 869  (-0.69%)

fmt [cmd/compile]
fmt.Sprintf 5009 -> 4977  (-0.64%)

runtime/pprof [cmd/compile]
runtime/pprof.printCountProfile.func1 875 -> 869  (-0.69%)

cmd/vendor/github.com/ianlancetaylor/demangle
cmd/vendor/github.com/ianlancetaylor/demangle.(*GlobalCDtor).print 957 -> 938  (-1.99%)
cmd/vendor/github.com/ianlancetaylor/demangle.(*ImaginaryType).printInner 270 -> 261  (-3.33%)
cmd/vendor/github.com/ianlancetaylor/demangle.(*TemplateTemplateParam).print 876 -> 874  (-0.23%)

cmd/cgo/internal/test
cmd/cgo/internal/test.test18146.func1 73 -> 66  (-9.59%)

cmd/go/internal/work
cmd/go/internal/work.(*Builder).gccSupportsFlag 3469 -> 3461  (-0.23%)

file                                            before   after    Δ       %
runtime/pprof.s                                 79729    79723    -6      -0.008%
fmt [cmd/compile].s                             83679    83647    -32     -0.038%
runtime/pprof [cmd/compile].s                   79729    79723    -6      -0.008%
cmd/vendor/github.com/ianlancetaylor/demangle.s 254999   254969   -30     -0.012%
cmd/cgo/internal/test.s                         212272   212265   -7      -0.003%
cmd/go/internal/work.s                          287700   287692   -8      -0.003%
cmd/compile/internal/ssa.s                      3193233  3193553  +320    +0.010%
cmd/compile/internal/ssa [cmd/compile].s        3362126  3362446  +320    +0.010%
total                                           30810158 30810709 +551    +0.002%

Change-Id: Iad09a42fb247c85ca1c649c367aaf75a620dbbc5
Reviewed-on: https://go-review.googlesource.com/c/go/+/521475
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
Run-TryBot: Jakub Ciolek <jakub@ciolek.dev>
TryBot-Result: Gopher Robot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
Auto-Submit: Matthew Dempsky <mdempsky@google.com>
Reviewed-by: Keith Randall <khr@google.com>

diff --git a/src/cmd/compile/internal/ssa/_gen/AMD64.rules b/src/cmd/compile/internal/ssa/_gen/AMD64.rules
index 9ffb196880a1a361a721a4a4127bef8c4e49ed45..7840600ef6f44208a1bb68d9217cd4f4f177668a 100644 (file)
--- a/src/cmd/compile/internal/ssa/_gen/AMD64.rules
+++ b/src/cmd/compile/internal/ssa/_gen/AMD64.rules
@@ -303,10 +303,13 @@
 (Move [10] dst src mem) =>
        (MOVWstore [8] dst (MOVWload [8] src mem)
                (MOVQstore dst (MOVQload src mem) mem))
+(Move [11] dst src mem) =>
+       (MOVLstore [7] dst (MOVLload [7] src mem)
+               (MOVQstore dst (MOVQload src mem) mem))
 (Move [12] dst src mem) =>
        (MOVLstore [8] dst (MOVLload [8] src mem)
                (MOVQstore dst (MOVQload src mem) mem))
-(Move [s] dst src mem) && s == 11 || s >= 13 && s <= 15 =>
+(Move [s] dst src mem) && s >= 13 && s <= 15 =>
        (MOVQstore [int32(s-8)] dst (MOVQload [int32(s-8)] src mem)
                (MOVQstore dst (MOVQload src mem) mem))
 

diff --git a/src/cmd/compile/internal/ssa/rewriteAMD64.go b/src/cmd/compile/internal/ssa/rewriteAMD64.go
index d3e009d644043b08019a2d3784c8defd1cb8af97..5cf5425fdcf14603f6de5a091ea7c8783e59b881 100644 (file)
--- a/src/cmd/compile/internal/ssa/rewriteAMD64.go
+++ b/src/cmd/compile/internal/ssa/rewriteAMD64.go
@@ -27456,6 +27456,27 @@ func rewriteValueAMD64_OpMove(v *Value) bool {
                v.AddArg3(dst, v0, v1)
                return true
        }
+       // match: (Move [11] dst src mem)
+       // result: (MOVLstore [7] dst (MOVLload [7] src mem) (MOVQstore dst (MOVQload src mem) mem))
+       for {
+               if auxIntToInt64(v.AuxInt) != 11 {
+                       break
+               }
+               dst := v_0
+               src := v_1
+               mem := v_2
+               v.reset(OpAMD64MOVLstore)
+               v.AuxInt = int32ToAuxInt(7)
+               v0 := b.NewValue0(v.Pos, OpAMD64MOVLload, typ.UInt32)
+               v0.AuxInt = int32ToAuxInt(7)
+               v0.AddArg2(src, mem)
+               v1 := b.NewValue0(v.Pos, OpAMD64MOVQstore, types.TypeMem)
+               v2 := b.NewValue0(v.Pos, OpAMD64MOVQload, typ.UInt64)
+               v2.AddArg2(src, mem)
+               v1.AddArg3(dst, v2, mem)
+               v.AddArg3(dst, v0, v1)
+               return true
+       }
        // match: (Move [12] dst src mem)
        // result: (MOVLstore [8] dst (MOVLload [8] src mem) (MOVQstore dst (MOVQload src mem) mem))
        for {
@@ -27478,14 +27499,14 @@ func rewriteValueAMD64_OpMove(v *Value) bool {
                return true
        }
        // match: (Move [s] dst src mem)
-       // cond: s == 11 || s >= 13 && s <= 15
+       // cond: s >= 13 && s <= 15
        // result: (MOVQstore [int32(s-8)] dst (MOVQload [int32(s-8)] src mem) (MOVQstore dst (MOVQload src mem) mem))
        for {
                s := auxIntToInt64(v.AuxInt)
                dst := v_0
                src := v_1
                mem := v_2
-               if !(s == 11 || s >= 13 && s <= 15) {
+               if !(s >= 13 && s <= 15) {
                        break
                }
                v.reset(OpAMD64MOVQstore)