[release-branch.go1.20] cmd/compile: fix bswap/load rewrite rules

When combining a byteswap and a load, the resulting combined op
must go in the load's block, not the byteswap's block, as the load
has a memory argument that might only be valid in its original block.

Fixes #59975

Change-Id: Icd84863ef3a9ca1fc22f2bb794a003f2808c746f
Reviewed-on: https://go-review.googlesource.com/c/go/+/492616
Run-TryBot: Keith Randall <khr@golang.org>
TryBot-Result: Gopher Robot <gobot@golang.org>
Reviewed-by: Cherry Mui <cherryyz@google.com>
Reviewed-by: Wayne Zuo <wdvxdr@golangcn.org>
Reviewed-by: Keith Randall <khr@google.com>
(cherry picked from commit 10141676d15665ed8a0255cfc5585ad50444f4ee)
Reviewed-on: https://go-review.googlesource.com/c/go/+/492696

diff --git a/src/cmd/compile/internal/ssa/_gen/AMD64.rules b/src/cmd/compile/internal/ssa/_gen/AMD64.rules
index aa156f61b99250158b74825cc6ff9b52ce63d194..da5ef7ef7f880d1a2b49918fbd4e0d5286851be6 100644 (file)
--- a/src/cmd/compile/internal/ssa/_gen/AMD64.rules
+++ b/src/cmd/compile/internal/ssa/_gen/AMD64.rules
@@ -2180,10 +2180,10 @@
 (BSWAP(Q|L) (BSWAP(Q|L) p)) => p
 
 // CPUID feature: MOVBE.
-(MOV(Q|L)store [i] {s} p x:(BSWAP(Q|L) w) mem) && x.Uses == 1 && buildcfg.GOAMD64 >= 3 => (MOVBE(Q|L)store [i] {s} p w mem)
-(BSWAP(Q|L) x:(MOV(Q|L)load [i] {s} p mem))    && x.Uses == 1 && buildcfg.GOAMD64 >= 3 => (MOVBE(Q|L)load [i] {s} p mem)
-(BSWAP(Q|L) (MOVBE(Q|L)load [i] {s} p m))    => (MOV(Q|L)load [i] {s} p m)
-(MOVBE(Q|L)store [i] {s} p (BSWAP(Q|L) x) m) => (MOV(Q|L)store [i] {s} p x m)
+(MOV(Q|L)store   [i] {s} p x:(BSWAP(Q|L) w) mem) && x.Uses == 1 && buildcfg.GOAMD64 >= 3 => (MOVBE(Q|L)store [i] {s} p w mem)
+(MOVBE(Q|L)store [i] {s} p x:(BSWAP(Q|L) w) mem) && x.Uses == 1                          => (MOV(Q|L)store   [i] {s} p w mem)
+(BSWAP(Q|L) x:(MOV(Q|L)load   [i] {s} p mem))  && x.Uses == 1 && buildcfg.GOAMD64 >= 3 => @x.Block (MOVBE(Q|L)load [i] {s} p mem)
+(BSWAP(Q|L) x:(MOVBE(Q|L)load [i] {s} p mem))  && x.Uses == 1                          => @x.Block (MOV(Q|L)load   [i] {s} p mem)
 (MOVWstore [i] {s} p x:(ROLWconst [8] w) mem)   && x.Uses == 1 && buildcfg.GOAMD64 >= 3 => (MOVBEWstore [i] {s} p w mem)
 (MOVBEWstore [i] {s} p x:(ROLWconst [8] w) mem) && x.Uses == 1 => (MOVWstore [i] {s} p w mem)
 

diff --git a/src/cmd/compile/internal/ssa/rewriteAMD64.go b/src/cmd/compile/internal/ssa/rewriteAMD64.go
index 7ece8c04ce3f631f772cdb40de75ac04cf5d83db..6025930aea484f499e8abcffcbf554e04478839a 100644 (file)
--- a/src/cmd/compile/internal/ssa/rewriteAMD64.go
+++ b/src/cmd/compile/internal/ssa/rewriteAMD64.go
@@ -3533,6 +3533,8 @@ func rewriteValueAMD64_OpAMD64BSFQ(v *Value) bool {
 }
 func rewriteValueAMD64_OpAMD64BSWAPL(v *Value) bool {
        v_0 := v.Args[0]
+       b := v.Block
+       typ := &b.Func.Config.Types
        // match: (BSWAPL (BSWAPL p))
        // result: p
        for {
@@ -3545,7 +3547,7 @@ func rewriteValueAMD64_OpAMD64BSWAPL(v *Value) bool {
        }
        // match: (BSWAPL x:(MOVLload [i] {s} p mem))
        // cond: x.Uses == 1 && buildcfg.GOAMD64 >= 3
-       // result: (MOVBELload [i] {s} p mem)
+       // result: @x.Block (MOVBELload [i] {s} p mem)
        for {
                x := v_0
                if x.Op != OpAMD64MOVLload {
@@ -3558,32 +3560,43 @@ func rewriteValueAMD64_OpAMD64BSWAPL(v *Value) bool {
                if !(x.Uses == 1 && buildcfg.GOAMD64 >= 3) {
                        break
                }
-               v.reset(OpAMD64MOVBELload)
-               v.AuxInt = int32ToAuxInt(i)
-               v.Aux = symToAux(s)
-               v.AddArg2(p, mem)
+               b = x.Block
+               v0 := b.NewValue0(x.Pos, OpAMD64MOVBELload, typ.UInt32)
+               v.copyOf(v0)
+               v0.AuxInt = int32ToAuxInt(i)
+               v0.Aux = symToAux(s)
+               v0.AddArg2(p, mem)
                return true
        }
-       // match: (BSWAPL (MOVBELload [i] {s} p m))
-       // result: (MOVLload [i] {s} p m)
+       // match: (BSWAPL x:(MOVBELload [i] {s} p mem))
+       // cond: x.Uses == 1
+       // result: @x.Block (MOVLload [i] {s} p mem)
        for {
-               if v_0.Op != OpAMD64MOVBELload {
+               x := v_0
+               if x.Op != OpAMD64MOVBELload {
                        break
                }
-               i := auxIntToInt32(v_0.AuxInt)
-               s := auxToSym(v_0.Aux)
-               m := v_0.Args[1]
-               p := v_0.Args[0]
-               v.reset(OpAMD64MOVLload)
-               v.AuxInt = int32ToAuxInt(i)
-               v.Aux = symToAux(s)
-               v.AddArg2(p, m)
+               i := auxIntToInt32(x.AuxInt)
+               s := auxToSym(x.Aux)
+               mem := x.Args[1]
+               p := x.Args[0]
+               if !(x.Uses == 1) {
+                       break
+               }
+               b = x.Block
+               v0 := b.NewValue0(x.Pos, OpAMD64MOVLload, typ.UInt32)
+               v.copyOf(v0)
+               v0.AuxInt = int32ToAuxInt(i)
+               v0.Aux = symToAux(s)
+               v0.AddArg2(p, mem)
                return true
        }
        return false
 }
 func rewriteValueAMD64_OpAMD64BSWAPQ(v *Value) bool {
        v_0 := v.Args[0]
+       b := v.Block
+       typ := &b.Func.Config.Types
        // match: (BSWAPQ (BSWAPQ p))
        // result: p
        for {
@@ -3596,7 +3609,7 @@ func rewriteValueAMD64_OpAMD64BSWAPQ(v *Value) bool {
        }
        // match: (BSWAPQ x:(MOVQload [i] {s} p mem))
        // cond: x.Uses == 1 && buildcfg.GOAMD64 >= 3
-       // result: (MOVBEQload [i] {s} p mem)
+       // result: @x.Block (MOVBEQload [i] {s} p mem)
        for {
                x := v_0
                if x.Op != OpAMD64MOVQload {
@@ -3609,26 +3622,35 @@ func rewriteValueAMD64_OpAMD64BSWAPQ(v *Value) bool {
                if !(x.Uses == 1 && buildcfg.GOAMD64 >= 3) {
                        break
                }
-               v.reset(OpAMD64MOVBEQload)
-               v.AuxInt = int32ToAuxInt(i)
-               v.Aux = symToAux(s)
-               v.AddArg2(p, mem)
+               b = x.Block
+               v0 := b.NewValue0(x.Pos, OpAMD64MOVBEQload, typ.UInt64)
+               v.copyOf(v0)
+               v0.AuxInt = int32ToAuxInt(i)
+               v0.Aux = symToAux(s)
+               v0.AddArg2(p, mem)
                return true
        }
-       // match: (BSWAPQ (MOVBEQload [i] {s} p m))
-       // result: (MOVQload [i] {s} p m)
+       // match: (BSWAPQ x:(MOVBEQload [i] {s} p mem))
+       // cond: x.Uses == 1
+       // result: @x.Block (MOVQload [i] {s} p mem)
        for {
-               if v_0.Op != OpAMD64MOVBEQload {
+               x := v_0
+               if x.Op != OpAMD64MOVBEQload {
                        break
                }
-               i := auxIntToInt32(v_0.AuxInt)
-               s := auxToSym(v_0.Aux)
-               m := v_0.Args[1]
-               p := v_0.Args[0]
-               v.reset(OpAMD64MOVQload)
-               v.AuxInt = int32ToAuxInt(i)
-               v.Aux = symToAux(s)
-               v.AddArg2(p, m)
+               i := auxIntToInt32(x.AuxInt)
+               s := auxToSym(x.Aux)
+               mem := x.Args[1]
+               p := x.Args[0]
+               if !(x.Uses == 1) {
+                       break
+               }
+               b = x.Block
+               v0 := b.NewValue0(x.Pos, OpAMD64MOVQload, typ.UInt64)
+               v.copyOf(v0)
+               v0.AuxInt = int32ToAuxInt(i)
+               v0.Aux = symToAux(s)
+               v0.AddArg2(p, mem)
                return true
        }
        return false
@@ -9398,21 +9420,26 @@ func rewriteValueAMD64_OpAMD64MOVBELstore(v *Value) bool {
        v_2 := v.Args[2]
        v_1 := v.Args[1]
        v_0 := v.Args[0]
-       // match: (MOVBELstore [i] {s} p (BSWAPL x) m)
-       // result: (MOVLstore [i] {s} p x m)
+       // match: (MOVBELstore [i] {s} p x:(BSWAPL w) mem)
+       // cond: x.Uses == 1
+       // result: (MOVLstore [i] {s} p w mem)
        for {
                i := auxIntToInt32(v.AuxInt)
                s := auxToSym(v.Aux)
                p := v_0
-               if v_1.Op != OpAMD64BSWAPL {
+               x := v_1
+               if x.Op != OpAMD64BSWAPL {
+                       break
+               }
+               w := x.Args[0]
+               mem := v_2
+               if !(x.Uses == 1) {
                        break
                }
-               x := v_1.Args[0]
-               m := v_2
                v.reset(OpAMD64MOVLstore)
                v.AuxInt = int32ToAuxInt(i)
                v.Aux = symToAux(s)
-               v.AddArg3(p, x, m)
+               v.AddArg3(p, w, mem)
                return true
        }
        return false
@@ -9421,21 +9448,26 @@ func rewriteValueAMD64_OpAMD64MOVBEQstore(v *Value) bool {
        v_2 := v.Args[2]
        v_1 := v.Args[1]
        v_0 := v.Args[0]
-       // match: (MOVBEQstore [i] {s} p (BSWAPQ x) m)
-       // result: (MOVQstore [i] {s} p x m)
+       // match: (MOVBEQstore [i] {s} p x:(BSWAPQ w) mem)
+       // cond: x.Uses == 1
+       // result: (MOVQstore [i] {s} p w mem)
        for {
                i := auxIntToInt32(v.AuxInt)
                s := auxToSym(v.Aux)
                p := v_0
-               if v_1.Op != OpAMD64BSWAPQ {
+               x := v_1
+               if x.Op != OpAMD64BSWAPQ {
+                       break
+               }
+               w := x.Args[0]
+               mem := v_2
+               if !(x.Uses == 1) {
                        break
                }
-               x := v_1.Args[0]
-               m := v_2
                v.reset(OpAMD64MOVQstore)
                v.AuxInt = int32ToAuxInt(i)
                v.Aux = symToAux(s)
-               v.AddArg3(p, x, m)
+               v.AddArg3(p, w, mem)
                return true
        }
        return false