1
0
mirror of https://github.com/RPCS3/llvm-mirror.git synced 2024-11-22 18:54:02 +01:00
llvm-mirror/test/CodeGen/X86/2012-08-16-setcc.ll
Simon Pilgrim 6aa28351a5 [X86] Fold cmpeq/ne(trunc(logic(x)),0) --> cmpeq/ne(logic(x),0)
Fixes the issues noted in PR48768, where the and/or/xor instruction had been promoted to avoid i8/i16 partial-dependencies, but the test against zero had not.

We can almost certainly relax this fold to work for any truncation, although it breaks a number of existing folds (notable movmsk folds which tend to rely on the truncate to determine the demanded bits/elts in the source vector).

There is a reverse combine in TargetLowering.SimplifySetCC so we must wait until after legalization before attempting this.
2021-04-12 16:05:34 +01:00

54 lines
1.3 KiB
LLVM

; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc < %s -mtriple=x86_64-unknown-unknown | FileCheck %s
; rdar://12081007
define i32 @and_1(i8 zeroext %a, i8 zeroext %b, i32 %x) {
; CHECK-LABEL: and_1:
; CHECK: # %bb.0:
; CHECK-NEXT: movl %edi, %eax
; CHECK-NEXT: andl %esi, %eax
; CHECK-NEXT: cmovnel %edx, %eax
; CHECK-NEXT: retq
%1 = and i8 %b, %a
%2 = icmp ne i8 %1, 0
%3 = select i1 %2, i32 %x, i32 0
ret i32 %3
}
define zeroext i1 @and_2(i8 zeroext %a, i8 zeroext %b) {
; CHECK-LABEL: and_2:
; CHECK: # %bb.0:
; CHECK-NEXT: testl %edi, %esi
; CHECK-NEXT: setne %al
; CHECK-NEXT: retq
%1 = and i8 %b, %a
%2 = icmp ne i8 %1, 0
ret i1 %2
}
define i32 @xor_1(i8 zeroext %a, i8 zeroext %b, i32 %x) {
; CHECK-LABEL: xor_1:
; CHECK: # %bb.0:
; CHECK-NEXT: xorl %eax, %eax
; CHECK-NEXT: xorl %esi, %edi
; CHECK-NEXT: cmovnel %edx, %eax
; CHECK-NEXT: retq
%1 = xor i8 %b, %a
%2 = icmp ne i8 %1, 0
%3 = select i1 %2, i32 %x, i32 0
ret i32 %3
}
define zeroext i1 @xor_2(i8 zeroext %a, i8 zeroext %b) {
; CHECK-LABEL: xor_2:
; CHECK: # %bb.0:
; CHECK-NEXT: xorl %esi, %edi
; CHECK-NEXT: setne %al
; CHECK-NEXT: retq
%1 = xor i8 %b, %a
%2 = icmp ne i8 %1, 0
ret i1 %2
}