f0dd12ec5c
The first attempt missed changing test files for tools (update_llc_test_checks.py). Original commit message: This implements the main suggested change from issue #56498. Using the shorter (non-extending) instruction with only -Oz ("minsize") rather than -Os ("optsize") is left as a possible follow-up. As noted in the bug report, the zero-extending load may have shorter latency/better throughput across a wide range of x86 micro-arches, and it avoids a potential false dependency. The cost is an extra instruction byte. This could cause perf ups and downs from secondary effects, but I don't think it is possible to account for those in advance, and that will likely also depend on exact micro-arch. This does bring LLVM x86 codegen more in line with existing gcc codegen, so if problems are exposed they are more likely to occur for both compilers. Differential Revision: https://reviews.llvm.org/D129775
43 lines
952 B
LLVM
43 lines
952 B
LLVM
; RUN: llc < %s -mtriple=i686-apple-darwin10 -fast-isel -fast-isel-abort=1 | FileCheck %s
|
|
; RUN: llc < %s -mtriple=x86_64-apple-darwin10 -fast-isel -fast-isel-abort=1 | FileCheck %s
|
|
|
|
declare i32 @test1a(i32)
|
|
|
|
define i32 @test1(i32 %x) nounwind {
|
|
; CHECK-LABEL: test1:
|
|
; CHECK: andb $1, %
|
|
%y = add i32 %x, -3
|
|
%t = call i32 @test1a(i32 %y)
|
|
%s = mul i32 %t, 77
|
|
%z = trunc i32 %s to i1
|
|
br label %next
|
|
|
|
next: ; preds = %0
|
|
%u = zext i1 %z to i32
|
|
%v = add i32 %u, 1999
|
|
br label %exit
|
|
|
|
exit: ; preds = %next
|
|
ret i32 %v
|
|
}
|
|
|
|
define void @test2(ptr %a) nounwind {
|
|
entry:
|
|
; clang uses i8 constants for booleans, so we test with an i8 1.
|
|
; CHECK-LABEL: test2:
|
|
; CHECK: movzbl {{.*}} %eax
|
|
; CHECK-NEXT: xorb $1, %al
|
|
; CHECK-NEXT: testb $1
|
|
%tmp = load i8, ptr %a, align 1
|
|
%xor = xor i8 %tmp, 1
|
|
%tobool = trunc i8 %xor to i1
|
|
br i1 %tobool, label %if.then, label %if.end
|
|
|
|
if.then:
|
|
call void @test2(ptr null)
|
|
br label %if.end
|
|
|
|
if.end:
|
|
ret void
|
|
}
|