d91a9e27a9
This will remove suboptimal branching from the generated ll/sc loops. The extra simplification pass affects a lot of testcases, which have been modified to accommodate this change: either by modifying the test to become immune to the CFG simplification, or (less preferablt) by adding option -hexagon-initial-cfg-clenaup=0. llvm-svn: 338774
18 lines
284 B
LLVM
18 lines
284 B
LLVM
; RUN: llc -march=hexagon -O2 < %s | FileCheck %s
|
|
|
|
; CHECK-NOT: p1 =
|
|
|
|
define i32 @f0(i32 %a0, i32 %a1) #0 {
|
|
b0:
|
|
%v0 = icmp slt i32 %a0, %a1
|
|
br i1 %v0, label %b1, label %b2
|
|
|
|
b1:
|
|
ret i32 0
|
|
|
|
b2:
|
|
%v1 = icmp slt i32 %a1, 100
|
|
%v2 = select i1 %v1, i32 123, i32 321
|
|
ret i32 %v2
|
|
}
|