mirror of
https://github.com/RPCS3/llvm-mirror.git
synced 2024-11-23 03:02:36 +01:00
db3c93e416
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
64 lines
2.1 KiB
LLVM
64 lines
2.1 KiB
LLVM
; RUN: llc -march=hexagon -hexagon-expand-condsets=0 -hexagon-initial-cfg-cleanup=0 < %s | FileCheck %s
|
|
;
|
|
; Expand-condsets eliminates the "mux" instruction, which is what this
|
|
; testcase is checking.
|
|
|
|
; Test that we don't generate a new value compare if the operands are
|
|
; the same register.
|
|
|
|
; CHECK-NOT: cmp.eq([[REG0:(r[0-9]+)]].new,[[REG0]])
|
|
; CHECK: cmp.eq([[REG1:(r[0-9]+)]],[[REG1]])
|
|
|
|
%s.0 = type { i16, i8, i32, i8*, i8*, i8*, i8*, i8*, i8*, i32*, [2 x i32], i8*, i8*, i8*, %s.1, i8*, [8 x i8], i8 }
|
|
%s.1 = type { i32, i16, i16 }
|
|
|
|
@g0 = external global %s.0
|
|
@g1 = external unnamed_addr constant [23 x i8], align 8
|
|
|
|
; Function Attrs: nounwind
|
|
declare void @f0(%s.0* nocapture, i8* nocapture readonly, ...) #0
|
|
|
|
define void @f1() #1 {
|
|
b0:
|
|
%v0 = load i32*, i32** undef, align 4
|
|
%v1 = load i32, i32* undef, align 4
|
|
br i1 undef, label %b4, label %b1
|
|
|
|
b1: ; preds = %b0
|
|
%v2 = icmp slt i32 %v1, 0
|
|
%v3 = lshr i32 %v1, 5
|
|
%v4 = add i32 %v3, -134217728
|
|
%v5 = select i1 %v2, i32 %v4, i32 %v3
|
|
%v6 = getelementptr inbounds i32, i32* %v0, i32 %v5
|
|
%v7 = icmp ult i32* %v6, %v0
|
|
%v8 = select i1 %v7, i32 0, i32 1
|
|
br i1 undef, label %b2, label %b4
|
|
|
|
b2: ; preds = %b1
|
|
%v9 = icmp slt i32 %v1, 0
|
|
%v10 = lshr i32 %v1, 5
|
|
%v11 = add i32 %v10, -134217728
|
|
%v12 = select i1 %v9, i32 %v11, i32 %v10
|
|
%v13 = getelementptr inbounds i32, i32* %v0, i32 %v12
|
|
%v14 = icmp ult i32* %v13, %v0
|
|
%v15 = select i1 %v14, i32 0, i32 1
|
|
%v16 = icmp eq i32 %v8, %v15
|
|
br i1 %v16, label %b4, label %b3
|
|
|
|
b3: ; preds = %b2
|
|
call void (%s.0*, i8*, ...) @f0(%s.0* @g0, i8* getelementptr inbounds ([23 x i8], [23 x i8]* @g1, i32 0, i32 0), i32 %v8, i32 %v15) #0
|
|
unreachable
|
|
|
|
b4: ; preds = %b2, %b1, %b0
|
|
br i1 undef, label %b6, label %b5
|
|
|
|
b5: ; preds = %b4
|
|
unreachable
|
|
|
|
b6: ; preds = %b4
|
|
ret void
|
|
}
|
|
|
|
attributes #0 = { nounwind "target-cpu"="hexagonv5" }
|
|
attributes #1 = { "target-cpu"="hexagonv5" }
|