92a286ae5a
Summary:
When conditional branches with complex conditions are split into
multiple branches in SelectionDAGBuilder::FindMergedConditions, also
handle inverted conditions. These may sometimes appear without having
been optimized by InstCombine when CodeGenPrepare decides to sink and
duplicate cmp instructions, causing them to have only one use. This
problem can be increased by e.g. GVNHoist hiding more cmps from
InstCombine by combining equivalent cmps from different blocks.
For example codegen X & !(Y | Z) as:
jmp_if_X TmpBB
jmp FBB
TmpBB:
jmp_if_notY Tmp2BB
jmp FBB
Tmp2BB:
jmp_if_notZ TBB
jmp FBB
Reviewers: bogner, MatzeB, qcolombet
Subscribers: llvm-commits, hiraditya, mcrosier, sebpop
Differential Revision: https://reviews.llvm.org/D28380
llvm-svn: 292944
49 lines
1.2 KiB
LLVM
49 lines
1.2 KiB
LLVM
; RUN: llc < %s | FileCheck %s
|
|
|
|
target datalayout = "e-m:o-i64:64-f80:128-n8:16:32:64-S128"
|
|
target triple = "x86_64-apple-macosx10.10.0"
|
|
|
|
; Function Attrs: nounwind
|
|
declare i32 @printf(i8* nocapture readonly, ...)
|
|
|
|
; On X86 1 is true and 0 is false, so we can't perform the combine:
|
|
; (and (setgt X, true), (setgt Y, true)) -> (setgt (or X, Y), true)
|
|
; This combine only works if the true value is -1.
|
|
|
|
|
|
;CHECK: cmpl
|
|
;CHECK: setl
|
|
;CHECK: cmpl
|
|
;CHECK: setl
|
|
;CHECK: orb
|
|
;CHECK: je
|
|
|
|
@.str = private unnamed_addr constant [4 x i8] c"%d\0A\00", align 1
|
|
; Function Attrs: optsize ssp uwtable
|
|
define i32 @foo(i32 %a, i32 %b, i32 * %c) {
|
|
if.else429:
|
|
%cmp.i1144 = icmp eq i32* %c, null
|
|
%cmp430 = icmp slt i32 %a, 2
|
|
%cmp432 = icmp slt i32 %b, 2
|
|
%or.cond710 = or i1 %cmp430, %cmp432
|
|
%or.cond710.not = xor i1 %or.cond710, true
|
|
%brmerge1448 = or i1 %cmp.i1144, %or.cond710.not
|
|
br i1 %brmerge1448, label %ret1, label %ret2
|
|
|
|
ret1:
|
|
ret i32 0
|
|
|
|
ret2:
|
|
ret i32 1
|
|
}
|
|
|
|
define i32 @main(i32 %argc, i8** nocapture readnone %argv) {
|
|
%res = alloca i32, align 4
|
|
%t = call i32 @foo(i32 1, i32 2, i32* %res) #3
|
|
%v = call i32 (i8*, ...) @printf(i8* getelementptr inbounds ([4 x i8], [4 x i8]* @.str, i64 0, i64 0), i32 %t)
|
|
ret i32 0
|
|
}
|
|
|
|
|
|
|