1
0
mirror of https://github.com/golang/go synced 2024-11-19 16:24:45 -07:00

cmd/compile/internal/ssa: combine consecutive loads and stores on amd64

Sometimes (often for calls) we generate code like this:

MOVQ  (addr),AX
MOVQ  8(addr),BX
MOVQ  AX,(otheraddr)
MOVQ  BX,8(otheraddr)

Replace it with

MOVUPS (addr),X0
MOVUPS X0,(otheraddr)

For completeness do the same for 8,16,32-bit loads/stores too.
Shaves 1% from code sections of go tool.

/localdisk/itocar/golang/bin/go 10293917
go_old 10334877 [40960 bytes]

read-only data = 682 bytes (0.040769%)
global text (code) = 38961 bytes (1.036503%)
Total difference 39643 bytes (0.674628%)

Updates #6853

Change-Id: I1f0d2f60273a63a079b58927cd1c4e3429d2e7ae
Reviewed-on: https://go-review.googlesource.com/57130
Run-TryBot: Ilya Tocar <ilya.tocar@intel.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
This commit is contained in:
Ilya Tocar 2017-08-18 14:03:33 -05:00
parent b40831b115
commit 9c99512d18
3 changed files with 600 additions and 9 deletions

File diff suppressed because it is too large Load Diff

View File

@ -2327,6 +2327,58 @@
&& clobber(x)
-> (MOVQstoreidx1 [i-4] {s} p (SHLQconst <idx.Type> [2] idx) w0 mem)
(MOVBstore [i] {s} p
x1:(MOVBload [j] {s2} p2 mem)
mem2:(MOVBstore [i-1] {s} p
x2:(MOVBload [j-1] {s2} p2 mem) mem))
&& x1.Uses == 1
&& x2.Uses == 1
&& mem2.Uses == 1
&& clobber(x1)
&& clobber(x2)
&& clobber(mem2)
-> (MOVWstore [i-1] {s} p (MOVWload [j-1] {s2} p2 mem) mem)
(MOVWstore [i] {s} p
x1:(MOVWload [j] {s2} p2 mem)
mem2:(MOVWstore [i-2] {s} p
x2:(MOVWload [j-2] {s2} p2 mem) mem))
&& x1.Uses == 1
&& x2.Uses == 1
&& mem2.Uses == 1
&& clobber(x1)
&& clobber(x2)
&& clobber(mem2)
-> (MOVLstore [i-2] {s} p (MOVLload [j-2] {s2} p2 mem) mem)
(MOVLstore [i] {s} p
x1:(MOVLload [j] {s2} p2 mem)
mem2:(MOVLstore [i-4] {s} p
x2:(MOVLload [j-4] {s2} p2 mem) mem))
&& x1.Uses == 1
&& x2.Uses == 1
&& mem2.Uses == 1
&& clobber(x1)
&& clobber(x2)
&& clobber(mem2)
-> (MOVQstore [i-4] {s} p (MOVQload [j-4] {s2} p2 mem) mem)
// This is somewhat tricky. There may be pointers in SSE registers due to rule below.
// However those register shouldn't live across GC safepoint.
(MOVQstore [i] {s} p
x1:(MOVQload [j] {s2} p2 mem)
mem2:(MOVQstore [i-8] {s} p
x2:(MOVQload [j-8] {s2} p2 mem) mem))
&& x1.Uses == 1
&& x2.Uses == 1
&& mem2.Uses == 1
&& config.useSSE
&& clobber(x1)
&& clobber(x2)
&& clobber(mem2)
-> (MOVOstore [i-8] {s} p (MOVOload [j-8] {s2} p2 mem) mem)
// amd64p32 rules
// same as the rules above, but with 32 instead of 64 bit pointer arithmetic.
// LEAQ,ADDQ -> LEAL,ADDL

View File

@ -154,7 +154,7 @@ func rewriteValueAMD64(v *Value) bool {
case OpAMD64MOVQloadidx8:
return rewriteValueAMD64_OpAMD64MOVQloadidx8_0(v)
case OpAMD64MOVQstore:
return rewriteValueAMD64_OpAMD64MOVQstore_0(v)
return rewriteValueAMD64_OpAMD64MOVQstore_0(v) || rewriteValueAMD64_OpAMD64MOVQstore_10(v)
case OpAMD64MOVQstoreconst:
return rewriteValueAMD64_OpAMD64MOVQstoreconst_0(v)
case OpAMD64MOVQstoreconstidx1:
@ -5690,6 +5690,10 @@ func rewriteValueAMD64_OpAMD64MOVBstore_0(v *Value) bool {
return false
}
func rewriteValueAMD64_OpAMD64MOVBstore_10(v *Value) bool {
b := v.Block
_ = b
typ := &b.Func.Config.Types
_ = typ
// match: (MOVBstore [i] {s} p (SHRQconst [8] w) x:(MOVBstore [i-1] {s} p w mem))
// cond: x.Uses == 1 && clobber(x)
// result: (MOVWstore [i-1] {s} p w mem)
@ -5785,6 +5789,73 @@ func rewriteValueAMD64_OpAMD64MOVBstore_10(v *Value) bool {
v.AddArg(mem)
return true
}
// match: (MOVBstore [i] {s} p x1:(MOVBload [j] {s2} p2 mem) mem2:(MOVBstore [i-1] {s} p x2:(MOVBload [j-1] {s2} p2 mem) mem))
// cond: x1.Uses == 1 && x2.Uses == 1 && mem2.Uses == 1 && clobber(x1) && clobber(x2) && clobber(mem2)
// result: (MOVWstore [i-1] {s} p (MOVWload [j-1] {s2} p2 mem) mem)
for {
i := v.AuxInt
s := v.Aux
_ = v.Args[2]
p := v.Args[0]
x1 := v.Args[1]
if x1.Op != OpAMD64MOVBload {
break
}
j := x1.AuxInt
s2 := x1.Aux
_ = x1.Args[1]
p2 := x1.Args[0]
mem := x1.Args[1]
mem2 := v.Args[2]
if mem2.Op != OpAMD64MOVBstore {
break
}
if mem2.AuxInt != i-1 {
break
}
if mem2.Aux != s {
break
}
_ = mem2.Args[2]
if p != mem2.Args[0] {
break
}
x2 := mem2.Args[1]
if x2.Op != OpAMD64MOVBload {
break
}
if x2.AuxInt != j-1 {
break
}
if x2.Aux != s2 {
break
}
_ = x2.Args[1]
if p2 != x2.Args[0] {
break
}
if mem != x2.Args[1] {
break
}
if mem != mem2.Args[2] {
break
}
if !(x1.Uses == 1 && x2.Uses == 1 && mem2.Uses == 1 && clobber(x1) && clobber(x2) && clobber(mem2)) {
break
}
v.reset(OpAMD64MOVWstore)
v.AuxInt = i - 1
v.Aux = s
v.AddArg(p)
v0 := b.NewValue0(v.Pos, OpAMD64MOVWload, typ.UInt16)
v0.AuxInt = j - 1
v0.Aux = s2
v0.AddArg(p2)
v0.AddArg(mem)
v.AddArg(v0)
v.AddArg(mem)
return true
}
// match: (MOVBstore [off1] {sym1} (LEAL [off2] {sym2} base) val mem)
// cond: canMergeSym(sym1, sym2)
// result: (MOVBstore [off1+off2] {mergeSym(sym1,sym2)} base val mem)
@ -7810,6 +7881,77 @@ func rewriteValueAMD64_OpAMD64MOVLstore_0(v *Value) bool {
return false
}
func rewriteValueAMD64_OpAMD64MOVLstore_10(v *Value) bool {
b := v.Block
_ = b
typ := &b.Func.Config.Types
_ = typ
// match: (MOVLstore [i] {s} p x1:(MOVLload [j] {s2} p2 mem) mem2:(MOVLstore [i-4] {s} p x2:(MOVLload [j-4] {s2} p2 mem) mem))
// cond: x1.Uses == 1 && x2.Uses == 1 && mem2.Uses == 1 && clobber(x1) && clobber(x2) && clobber(mem2)
// result: (MOVQstore [i-4] {s} p (MOVQload [j-4] {s2} p2 mem) mem)
for {
i := v.AuxInt
s := v.Aux
_ = v.Args[2]
p := v.Args[0]
x1 := v.Args[1]
if x1.Op != OpAMD64MOVLload {
break
}
j := x1.AuxInt
s2 := x1.Aux
_ = x1.Args[1]
p2 := x1.Args[0]
mem := x1.Args[1]
mem2 := v.Args[2]
if mem2.Op != OpAMD64MOVLstore {
break
}
if mem2.AuxInt != i-4 {
break
}
if mem2.Aux != s {
break
}
_ = mem2.Args[2]
if p != mem2.Args[0] {
break
}
x2 := mem2.Args[1]
if x2.Op != OpAMD64MOVLload {
break
}
if x2.AuxInt != j-4 {
break
}
if x2.Aux != s2 {
break
}
_ = x2.Args[1]
if p2 != x2.Args[0] {
break
}
if mem != x2.Args[1] {
break
}
if mem != mem2.Args[2] {
break
}
if !(x1.Uses == 1 && x2.Uses == 1 && mem2.Uses == 1 && clobber(x1) && clobber(x2) && clobber(mem2)) {
break
}
v.reset(OpAMD64MOVQstore)
v.AuxInt = i - 4
v.Aux = s
v.AddArg(p)
v0 := b.NewValue0(v.Pos, OpAMD64MOVQload, typ.UInt64)
v0.AuxInt = j - 4
v0.Aux = s2
v0.AddArg(p2)
v0.AddArg(mem)
v.AddArg(v0)
v.AddArg(mem)
return true
}
// match: (MOVLstore [off1] {sym1} (LEAL [off2] {sym2} base) val mem)
// cond: canMergeSym(sym1, sym2)
// result: (MOVLstore [off1+off2] {mergeSym(sym1,sym2)} base val mem)
@ -9345,6 +9487,10 @@ func rewriteValueAMD64_OpAMD64MOVQloadidx8_0(v *Value) bool {
return false
}
func rewriteValueAMD64_OpAMD64MOVQstore_0(v *Value) bool {
b := v.Block
_ = b
config := b.Func.Config
_ = config
// match: (MOVQstore [off1] {sym} (ADDQconst [off2] ptr) val mem)
// cond: is32Bit(off1+off2)
// result: (MOVQstore [off1+off2] {sym} ptr val mem)
@ -9510,6 +9656,73 @@ func rewriteValueAMD64_OpAMD64MOVQstore_0(v *Value) bool {
v.AddArg(mem)
return true
}
// match: (MOVQstore [i] {s} p x1:(MOVQload [j] {s2} p2 mem) mem2:(MOVQstore [i-8] {s} p x2:(MOVQload [j-8] {s2} p2 mem) mem))
// cond: x1.Uses == 1 && x2.Uses == 1 && mem2.Uses == 1 && config.useSSE && clobber(x1) && clobber(x2) && clobber(mem2)
// result: (MOVOstore [i-8] {s} p (MOVOload [j-8] {s2} p2 mem) mem)
for {
i := v.AuxInt
s := v.Aux
_ = v.Args[2]
p := v.Args[0]
x1 := v.Args[1]
if x1.Op != OpAMD64MOVQload {
break
}
j := x1.AuxInt
s2 := x1.Aux
_ = x1.Args[1]
p2 := x1.Args[0]
mem := x1.Args[1]
mem2 := v.Args[2]
if mem2.Op != OpAMD64MOVQstore {
break
}
if mem2.AuxInt != i-8 {
break
}
if mem2.Aux != s {
break
}
_ = mem2.Args[2]
if p != mem2.Args[0] {
break
}
x2 := mem2.Args[1]
if x2.Op != OpAMD64MOVQload {
break
}
if x2.AuxInt != j-8 {
break
}
if x2.Aux != s2 {
break
}
_ = x2.Args[1]
if p2 != x2.Args[0] {
break
}
if mem != x2.Args[1] {
break
}
if mem != mem2.Args[2] {
break
}
if !(x1.Uses == 1 && x2.Uses == 1 && mem2.Uses == 1 && config.useSSE && clobber(x1) && clobber(x2) && clobber(mem2)) {
break
}
v.reset(OpAMD64MOVOstore)
v.AuxInt = i - 8
v.Aux = s
v.AddArg(p)
v0 := b.NewValue0(v.Pos, OpAMD64MOVOload, types.TypeInt128)
v0.AuxInt = j - 8
v0.Aux = s2
v0.AddArg(p2)
v0.AddArg(mem)
v.AddArg(v0)
v.AddArg(mem)
return true
}
// match: (MOVQstore [off1] {sym1} (LEAL [off2] {sym2} base) val mem)
// cond: canMergeSym(sym1, sym2)
// result: (MOVQstore [off1+off2] {mergeSym(sym1,sym2)} base val mem)
@ -9602,6 +9815,9 @@ func rewriteValueAMD64_OpAMD64MOVQstore_0(v *Value) bool {
v.AddArg(mem)
return true
}
return false
}
func rewriteValueAMD64_OpAMD64MOVQstore_10(v *Value) bool {
// match: (MOVQstore [off] {sym} ptr (MOVQf2i val) mem)
// cond:
// result: (MOVSDstore [off] {sym} ptr val mem)
@ -12334,6 +12550,77 @@ func rewriteValueAMD64_OpAMD64MOVWstore_0(v *Value) bool {
return false
}
func rewriteValueAMD64_OpAMD64MOVWstore_10(v *Value) bool {
b := v.Block
_ = b
typ := &b.Func.Config.Types
_ = typ
// match: (MOVWstore [i] {s} p x1:(MOVWload [j] {s2} p2 mem) mem2:(MOVWstore [i-2] {s} p x2:(MOVWload [j-2] {s2} p2 mem) mem))
// cond: x1.Uses == 1 && x2.Uses == 1 && mem2.Uses == 1 && clobber(x1) && clobber(x2) && clobber(mem2)
// result: (MOVLstore [i-2] {s} p (MOVLload [j-2] {s2} p2 mem) mem)
for {
i := v.AuxInt
s := v.Aux
_ = v.Args[2]
p := v.Args[0]
x1 := v.Args[1]
if x1.Op != OpAMD64MOVWload {
break
}
j := x1.AuxInt
s2 := x1.Aux
_ = x1.Args[1]
p2 := x1.Args[0]
mem := x1.Args[1]
mem2 := v.Args[2]
if mem2.Op != OpAMD64MOVWstore {
break
}
if mem2.AuxInt != i-2 {
break
}
if mem2.Aux != s {
break
}
_ = mem2.Args[2]
if p != mem2.Args[0] {
break
}
x2 := mem2.Args[1]
if x2.Op != OpAMD64MOVWload {
break
}
if x2.AuxInt != j-2 {
break
}
if x2.Aux != s2 {
break
}
_ = x2.Args[1]
if p2 != x2.Args[0] {
break
}
if mem != x2.Args[1] {
break
}
if mem != mem2.Args[2] {
break
}
if !(x1.Uses == 1 && x2.Uses == 1 && mem2.Uses == 1 && clobber(x1) && clobber(x2) && clobber(mem2)) {
break
}
v.reset(OpAMD64MOVLstore)
v.AuxInt = i - 2
v.Aux = s
v.AddArg(p)
v0 := b.NewValue0(v.Pos, OpAMD64MOVLload, typ.UInt32)
v0.AuxInt = j - 2
v0.Aux = s2
v0.AddArg(p2)
v0.AddArg(mem)
v.AddArg(v0)
v.AddArg(mem)
return true
}
// match: (MOVWstore [off1] {sym1} (LEAL [off2] {sym2} base) val mem)
// cond: canMergeSym(sym1, sym2)
// result: (MOVWstore [off1+off2] {mergeSym(sym1,sym2)} base val mem)