Skip to content

Commit 9613a63

Browse files
dr2chasedr2chase
committed
cmd/compile: fix subword store/load elision for amd64, x86, arm
Replacing byteload-of-bytestore-of-x with x is incorrect when x contains a larger-than-byte value (and so on for 16 and 32-bit load/store pairs). Replace "x" with the appropriate zero/sign extension of x, which if unnecessary will be repaired by other rules. Made logic for arm match x86 and amd64; yields minor extra optimization, plus I am (much) more confident it's correct, despite inability to reproduce bug on arm. Ppc64 lacks this optimization, hence lacks this problem. See related https://golang.org/cl/37154/ Fixes #20530. Change-Id: I6af9cac2ad43bee99cafdcb04725ce7e55a43323 Reviewed-on: https://go-review.googlesource.com/44355 Run-TryBot: David Chase <drchase@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Keith Randall <khr@golang.org>
1 parent d10549f commit 9613a63

File tree

7 files changed

+182
-47
lines changed

7 files changed

+182
-47
lines changed

‎src/cmd/compile/internal/ssa/gen/386.rules

Lines changed: 5 additions & 3 deletions
Original file line numberDiff line numberDiff line change
@@ -615,10 +615,12 @@
615615
(MOVWLZX x:(MOVWloadidx1 [off] {sym} ptr idx mem)) && x.Uses == 1 && clobber(x) -> @x.Block (MOVWloadidx1 <v.Type> [off] {sym} ptr idx mem)
616616
(MOVWLZX x:(MOVWloadidx2 [off] {sym} ptr idx mem)) && x.Uses == 1 && clobber(x) -> @x.Block (MOVWloadidx2 <v.Type> [off] {sym} ptr idx mem)
617617

618-
// replace load from same location as preceding store with copy
619-
(MOVBload [off] {sym} ptr (MOVBstore [off2] {sym2} ptr2 x _)) && sym == sym2 && off == off2 && isSamePtr(ptr, ptr2) -> x
620-
(MOVWload [off] {sym} ptr (MOVWstore [off2] {sym2} ptr2 x _)) && sym == sym2 && off == off2 && isSamePtr(ptr, ptr2) -> x
618+
// replace load from same location as preceding store with zero/sign extension (or copy in case of full width)
619+
(MOVBload [off] {sym} ptr (MOVBstore [off2] {sym2} ptr2 x _)) && sym == sym2 && off == off2 && isSamePtr(ptr, ptr2) -> (MOVBLZX x)
620+
(MOVWload [off] {sym} ptr (MOVWstore [off2] {sym2} ptr2 x _)) && sym == sym2 && off == off2 && isSamePtr(ptr, ptr2) -> (MOVWLZX x)
621621
(MOVLload [off] {sym} ptr (MOVLstore [off2] {sym2} ptr2 x _)) && sym == sym2 && off == off2 && isSamePtr(ptr, ptr2) -> x
622+
(MOVBLSXload [off] {sym} ptr (MOVBstore [off2] {sym2} ptr2 x _)) && sym == sym2 && off == off2 && isSamePtr(ptr, ptr2) -> (MOVBLSX x)
623+
(MOVWLSXload [off] {sym} ptr (MOVWstore [off2] {sym2} ptr2 x _)) && sym == sym2 && off == off2 && isSamePtr(ptr, ptr2) -> (MOVWLSX x)
622624

623625
// Fold extensions and ANDs together.
624626
(MOVBLZX (ANDLconst [c] x)) -> (ANDLconst [c & 0xff] x)

‎src/cmd/compile/internal/ssa/gen/AMD64.rules

Lines changed: 7 additions & 4 deletions
Original file line numberDiff line numberDiff line change
@@ -965,11 +965,14 @@
965965
(MOVLQZX x:(MOVLloadidx1 [off] {sym} ptr idx mem)) && x.Uses == 1 && clobber(x) -> @x.Block (MOVLloadidx1 <v.Type> [off] {sym} ptr idx mem)
966966
(MOVLQZX x:(MOVLloadidx4 [off] {sym} ptr idx mem)) && x.Uses == 1 && clobber(x) -> @x.Block (MOVLloadidx4 <v.Type> [off] {sym} ptr idx mem)
967967

968-
// replace load from same location as preceding store with copy
969-
(MOVBload [off] {sym} ptr (MOVBstore [off2] {sym2} ptr2 x _)) && sym == sym2 && off == off2 && isSamePtr(ptr, ptr2) -> x
970-
(MOVWload [off] {sym} ptr (MOVWstore [off2] {sym2} ptr2 x _)) && sym == sym2 && off == off2 && isSamePtr(ptr, ptr2) -> x
971-
(MOVLload [off] {sym} ptr (MOVLstore [off2] {sym2} ptr2 x _)) && sym == sym2 && off == off2 && isSamePtr(ptr, ptr2) -> x
968+
// replace load from same location as preceding store with zero/sign extension (or copy in case of full width)
969+
(MOVBload [off] {sym} ptr (MOVBstore [off2] {sym2} ptr2 x _)) && sym == sym2 && off == off2 && isSamePtr(ptr, ptr2) -> (MOVBQZX x)
970+
(MOVWload [off] {sym} ptr (MOVWstore [off2] {sym2} ptr2 x _)) && sym == sym2 && off == off2 && isSamePtr(ptr, ptr2) -> (MOVWQZX x)
971+
(MOVLload [off] {sym} ptr (MOVLstore [off2] {sym2} ptr2 x _)) && sym == sym2 && off == off2 && isSamePtr(ptr, ptr2) -> (MOVLQZX x)
972972
(MOVQload [off] {sym} ptr (MOVQstore [off2] {sym2} ptr2 x _)) && sym == sym2 && off == off2 && isSamePtr(ptr, ptr2) -> x
973+
(MOVBQSXload [off] {sym} ptr (MOVBstore [off2] {sym2} ptr2 x _)) && sym == sym2 && off == off2 && isSamePtr(ptr, ptr2) -> (MOVBQSX x)
974+
(MOVWQSXload [off] {sym} ptr (MOVWstore [off2] {sym2} ptr2 x _)) && sym == sym2 && off == off2 && isSamePtr(ptr, ptr2) -> (MOVWQSX x)
975+
(MOVLQSXload [off] {sym} ptr (MOVLstore [off2] {sym2} ptr2 x _)) && sym == sym2 && off == off2 && isSamePtr(ptr, ptr2) -> (MOVLQSX x)
973976

974977
// Fold extensions and ANDs together.
975978
(MOVBQZX (ANDLconst [c] x)) -> (ANDLconst [c & 0xff] x)

‎src/cmd/compile/internal/ssa/gen/ARM.rules

Lines changed: 6 additions & 5 deletions
Original file line numberDiff line numberDiff line change
@@ -485,12 +485,13 @@
485485
(MOVDstore [off1] {sym1} (MOVWaddr [off2] {sym2} ptr) val mem) && canMergeSym(sym1,sym2) ->
486486
(MOVDstore [off1+off2] {mergeSym(sym1,sym2)} ptr val mem)
487487

488-
// replace load from same location as preceding store with copy
489-
(MOVBload [off] {sym} ptr (MOVBstore [off2] {sym2} ptr2 x _)) && sym == sym2 && off == off2 && isSamePtr(ptr, ptr2) && isSigned(x.Type) -> x
490-
(MOVBUload [off] {sym} ptr (MOVBstore [off2] {sym2} ptr2 x _)) && sym == sym2 && off == off2 && isSamePtr(ptr, ptr2) && !isSigned(x.Type) -> x
491-
(MOVHload [off] {sym} ptr (MOVHstore [off2] {sym2} ptr2 x _)) && sym == sym2 && off == off2 && isSamePtr(ptr, ptr2) && isSigned(x.Type) -> x
492-
(MOVHUload [off] {sym} ptr (MOVHstore [off2] {sym2} ptr2 x _)) && sym == sym2 && off == off2 && isSamePtr(ptr, ptr2) && !isSigned(x.Type) -> x
488+
// replace load from same location as preceding store with zero/sign extension (or copy in case of full width)
489+
(MOVBload [off] {sym} ptr (MOVBstore [off2] {sym2} ptr2 x _)) && sym == sym2 && off == off2 && isSamePtr(ptr, ptr2) -> (MOVBreg x)
490+
(MOVBUload [off] {sym} ptr (MOVBstore [off2] {sym2} ptr2 x _)) && sym == sym2 && off == off2 && isSamePtr(ptr, ptr2) -> (MOVBUreg x)
491+
(MOVHload [off] {sym} ptr (MOVHstore [off2] {sym2} ptr2 x _)) && sym == sym2 && off == off2 && isSamePtr(ptr, ptr2) -> (MOVHreg x)
492+
(MOVHUload [off] {sym} ptr (MOVHstore [off2] {sym2} ptr2 x _)) && sym == sym2 && off == off2 && isSamePtr(ptr, ptr2) -> (MOVHUreg x)
493493
(MOVWload [off] {sym} ptr (MOVWstore [off2] {sym2} ptr2 x _)) && sym == sym2 && off == off2 && isSamePtr(ptr, ptr2) -> x
494+
494495
(MOVFload [off] {sym} ptr (MOVFstore [off2] {sym2} ptr2 x _)) && sym == sym2 && off == off2 && isSamePtr(ptr, ptr2) -> x
495496
(MOVDload [off] {sym} ptr (MOVDstore [off2] {sym2} ptr2 x _)) && sym == sym2 && off == off2 && isSamePtr(ptr, ptr2) -> x
496497

‎src/cmd/compile/internal/ssa/rewrite386.go

Lines changed: 52 additions & 6 deletions
Some generated files are not rendered by default. Learn more about customizing how changed files appear on GitHub.

‎src/cmd/compile/internal/ssa/rewriteAMD64.go

Lines changed: 78 additions & 9 deletions
Some generated files are not rendered by default. Learn more about customizing how changed files appear on GitHub.

0 commit comments

Comments
 (0)