llvm-project/llvm/test/CodeGen/Hexagon/swp-epilog-phi8.ll
Fangrui Song 2208c97c1b [Hexagon,test] Change llc -march= to -mtriple=
Similar to 806761a7629df268c8aed49657aeccffa6bca449

-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.
2024-12-15 10:20:22 -08:00

46 lines
1.6 KiB
LLVM

; RUN: llc -mtriple=hexagon -mno-pairing -mno-compound -hexagon-initial-cfg-cleanup=0 < %s -pipeliner-experimental-cg=true | FileCheck %s
; XFAIL: *
; Test that we generate the correct phi names in the epilog when the pipeliner
; schedules a phi and it's loop definition in different stages, e.g., a phi is
; scheduled in stage 2, but the loop definition in scheduled in stage 0). The
; code in generateExistingPhis was generating the wrong name for the last
; epilog bock.
; CHECK: endloop0
; CHECK: sub([[REG:r([0-9]+)]],r{{[0-9]+}}):sat
; CHECK-NOT: sub([[REG]],r{{[0-9]+}}):sat
define void @f0() {
b0:
br label %b1
b1: ; preds = %b1, %b0
br i1 undef, label %b2, label %b1
b2: ; preds = %b1
br label %b3
b3: ; preds = %b3, %b2
%v0 = phi i32 [ %v8, %b3 ], [ 7, %b2 ]
%v1 = phi i32 [ %v6, %b3 ], [ undef, %b2 ]
%v2 = phi i32 [ %v1, %b3 ], [ undef, %b2 ]
%v3 = getelementptr inbounds [9 x i32], ptr undef, i32 0, i32 %v0
%v4 = add nsw i32 %v0, -2
%v5 = getelementptr inbounds [9 x i32], ptr undef, i32 0, i32 %v4
%v6 = load i32, ptr %v5, align 4
%v7 = tail call i32 @llvm.hexagon.A2.subsat(i32 %v2, i32 %v6)
store i32 %v7, ptr %v3, align 4
%v8 = add i32 %v0, -1
%v9 = icmp sgt i32 %v8, 1
br i1 %v9, label %b3, label %b4
b4: ; preds = %b3
unreachable
}
; Function Attrs: nounwind readnone
declare i32 @llvm.hexagon.A2.subsat(i32, i32) #0
attributes #0 = { nounwind readnone "target-cpu"="hexagonv60" }