Skip to content

A lightweight WebAssembly runtime that is fast, secure, and standards-compliant

x64: Shrink size of `x64_not` in `icmp` peephole optimization (#12876)

Alex Crichton8da1acec7e5e
diff --git a/cranelift/codegen/src/isa/x64/lower.isle b/cranelift/codegen/src/isa/x64/lower.isle
index 1ce3df5753..2cea2d5511 100644
--- a/cranelift/codegen/src/isa/x64/lower.isle
+++ b/cranelift/codegen/src/isa/x64/lower.isle
@@ -1978,11 +1978,11 @@
;; Peephole optimization for `0 <= x`, when x is a signed 32 bit value
(rule 2 (lower (has_type $I8 (icmp _ (IntCC.SignedLessThanOrEqual) (u64_from_iconst 0) x @ (value_type $I32))))
- (x64_shrl_mi (x64_not $I64 x) 31))
+ (x64_shrl_mi (x64_not $I32 x) 31))
;; Peephole optimization for `x >= 0`, when x is a signed 32 bit value
(rule 2 (lower (has_type $I8 (icmp _ (IntCC.SignedGreaterThanOrEqual) x @ (value_type $I32) (u64_from_iconst 0))))
- (x64_shrl_mi (x64_not $I64 x) 31))
+ (x64_shrl_mi (x64_not $I32 x) 31))
;; For XMM-held values, we lower to `PCMP*` instructions, sometimes more than
;; one. To note: what is different here about the output values is that each
diff --git a/cranelift/filetests/filetests/isa/x64/conditional-values.clif b/cranelift/filetests/filetests/isa/x64/conditional-values.clif
index 231cbd8d90..24f048c1e2 100644
--- a/cranelift/filetests/filetests/isa/x64/conditional-values.clif
+++ b/cranelift/filetests/filetests/isa/x64/conditional-values.clif
@@ -377,7 +377,7 @@ block0(v0: i32):
; pushq %rbp
; movq %rsp, %rbp
; block0:
-; notq %rdi
+; notl %edi
; movq %rdi, %rax
; shrl $0x1f, %eax
; movq %rbp, %rsp
@@ -389,7 +389,7 @@ block0(v0: i32):
; pushq %rbp
; movq %rsp, %rbp
; block1: ; offset 0x4
-; notq %rdi
+; notl %edi
; movq %rdi, %rax
; shrl $0x1f, %eax
; movq %rbp, %rsp
@@ -437,7 +437,7 @@ block0(v0: i32):
; pushq %rbp
; movq %rsp, %rbp
; block0:
-; notq %rdi
+; notl %edi
; movq %rdi, %rax
; shrl $0x1f, %eax
; movq %rbp, %rsp
@@ -449,7 +449,7 @@ block0(v0: i32):
; pushq %rbp
; movq %rsp, %rbp
; block1: ; offset 0x4
-; notq %rdi
+; notl %edi
; movq %rdi, %rax
; shrl $0x1f, %eax
; movq %rbp, %rsp