cmd/compile: optimize BSRQ with nonzero input

Same as BSFQ, if the input is known to be nonzero we don't need a CMOV to correct for the weird all-zero behavior of the BSRQ instruction. Fixes #51964 Change-Id: Ic1d14fdf3e7e698ff2f47efecfd761d8a036012a Reviewed-on: https://go-review.googlesource.com/c/go/+/396034 Trust: Keith Randall <[email protected]> Run-TryBot: Keith Randall <[email protected]> TryBot-Result: Gopher Robot <[email protected]> Reviewed-by: David Chase <[email protected]>
cloudflare · Mar 27, 2022 · 7d3b64a · 7d3b64a
1 parent 56400fc
commit 7d3b64a
Show file tree

Hide file tree

Showing 2 changed files with 24 additions and 1 deletion.
diff --git a/src/cmd/compile/internal/ssa/gen/AMD64.rules b/src/cmd/compile/internal/ssa/gen/AMD64.rules
@@ -2073,7 +2073,7 @@
  (CMPXCHGLlock [off1+off2] {sym} ptr old new_ mem)
 
 // We don't need the conditional move if we know the arg of BSF is not zero.
-(CMOVQEQ x _ (Select1 (BSFQ (ORQconst [c] _)))) && c != 0 => x
+(CMOVQEQ x _ (Select1 (BS(F|R)Q (ORQconst [c] _)))) && c != 0 => x
 // Extension is unnecessary for trailing zeros.
 (BSFQ (ORQconst <t> [1<<8] (MOVBQZX x))) => (BSFQ (ORQconst <t> [1<<8] x))
 (BSFQ (ORQconst <t> [1<<16] (MOVWQZX x))) => (BSFQ (ORQconst <t> [1<<16] x))

diff --git a/src/cmd/compile/internal/ssa/rewriteAMD64.go b/src/cmd/compile/internal/ssa/rewriteAMD64.go