]> Cypherpunks repositories - gostls13.git/commit
cmd/compile: fold LEAQ with constant scale into LEA
authorGiovanni Bajo <rasky@develer.com>
Mon, 23 Mar 2020 08:21:44 +0000 (09:21 +0100)
committerGiovanni Bajo <rasky@develer.com>
Mon, 23 Mar 2020 17:47:29 +0000 (17:47 +0000)
commit24925c7ed9da15fe780d0d2fce21bbaa707e8751
tree3e985653337a8feb00921ccc7bea81c7562e3b96
parent5d47f870a64982173f3da8a82a6a793b83075d6a
cmd/compile: fold LEAQ with constant scale into LEA

Discovered this after rebasing CL196679 (use poset bounds in prove).
Some tests fail with that CL applied:

codegen/smallintiface.go:11: linux/amd64/: opcode not found: "^LEAQ\\truntime.staticuint64s\\+8\\(SB\\)"
codegen/smallintiface.go:16: linux/amd64/: opcode not found: "^LEAQ\\truntime.staticuint64s\\+2024\\(SB\\)"
codegen/smallintiface.go:21: linux/amd64/: opcode not found: "^LEAQ\\truntime.staticuint64s\\+24\\(SB\\)"

The only difference in prove SSA dumps is that a single Lsh64x64
op with constant shift (<< 3) is marked as bounded. This triggers
a different rule matching sequence in lower, which manages to generate
worse code for the above testcases.

This CL fixes the above test after CL196679 is applied. Right now,
these rules never trigger (this CL passes toolstash -cmp), so I can't
write a test.

Change-Id: I353f1c79c1875cac1da82cd8afa1e05e42684f1c
Reviewed-on: https://go-review.googlesource.com/c/go/+/224877
Run-TryBot: Giovanni Bajo <rasky@develer.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
src/cmd/compile/internal/ssa/gen/AMD64.rules
src/cmd/compile/internal/ssa/rewriteAMD64.go