2208c97c1b
Similar to 806761a762
-mtriple= specifies the full target triple while -march= merely sets the
architecture part of the default target triple, leaving a target triple which
may not make sense.
Therefore, -march= is error-prone and not recommended for tests without a target
triple. The issue has been benign as we recognize $unknown-apple-darwin as ELF instead
of rejecting it outrightly.
41 lines
1.1 KiB
LLVM
41 lines
1.1 KiB
LLVM
; RUN: llc -mtriple=hexagon -hexagon-hwloop-preheader < %s | FileCheck %s
|
|
|
|
; check for lack of assertion failures.
|
|
|
|
; CHECK: %bb.0
|
|
|
|
; Function Attrs: nounwind readnone
|
|
declare i32 @llvm.hexagon.A2.sath(i32) #0
|
|
|
|
; Function Attrs: nounwind readnone
|
|
declare i32 @llvm.hexagon.A2.neg(i32) #0
|
|
|
|
define void @f0(i16 signext %a0) {
|
|
b0:
|
|
%v0 = icmp slt i16 %a0, 1
|
|
br i1 %v0, label %b1, label %b3
|
|
|
|
b1: ; preds = %b2, %b0
|
|
%v1 = phi i16 [ %v11, %b2 ], [ %a0, %b0 ]
|
|
%v2 = sext i16 %v1 to i32
|
|
%v3 = tail call i32 @llvm.hexagon.A2.neg(i32 %v2)
|
|
%v4 = tail call i32 @llvm.hexagon.A2.sath(i32 %v3)
|
|
%v5 = trunc i32 %v4 to i16
|
|
%v6 = shl i32 %v4, 16
|
|
%v7 = ashr exact i32 %v6, 16
|
|
%v8 = icmp slt i16 %v5, 0
|
|
br i1 %v8, label %b2, label %b3
|
|
|
|
b2: ; preds = %b1
|
|
%v9 = tail call i32 @llvm.hexagon.A2.neg(i32 %v7)
|
|
%v10 = tail call i32 @llvm.hexagon.A2.sath(i32 %v9)
|
|
%v11 = trunc i32 %v10 to i16
|
|
%v12 = icmp slt i16 %v11, 1
|
|
br i1 %v12, label %b1, label %b3
|
|
|
|
b3: ; preds = %b2, %b1, %b0
|
|
ret void
|
|
}
|
|
|
|
attributes #0 = { nounwind readnone }
|