1 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2 ; RUN: llc -verify-machineinstrs -mtriple=powerpc64le-unknown-unknown \
3 ; RUN: -mcpu=pwr6 -ppc-asm-full-reg-names -mattr=-vsx \
4 ; RUN: -ppc-vsr-nums-as-vr < %s | FileCheck %s
6 ; There is code in the SDAG to expand FMAX/FMIN with fast flags to SELECT_CC.
7 ; On PPC, we had SELECT_CC legalized using Promote for all vector types
8 ; (including the type that they are all promoted to - which caused an infinite
9 ; loop in legalization). This test just ensures that we terminate on such input.
10 define dso_local void @_ZN1a1bEv(<4 x float> %in) local_unnamed_addr #0 align 2 {
11 ; CHECK-LABEL: _ZN1a1bEv:
13 ; CHECK-NEXT: bc 12, 4*cr5+lt, .LBB0_6
14 ; CHECK-NEXT: b .LBB0_1
15 ; CHECK-NEXT: .LBB0_1: # %.preheader
16 ; CHECK-NEXT: b .LBB0_2
17 ; CHECK-NEXT: .LBB0_2:
18 ; CHECK-NEXT: b .LBB0_3
19 ; CHECK-NEXT: .LBB0_3:
20 ; CHECK-NEXT: addis r3, r2, .LCPI0_0@toc@ha
21 ; CHECK-NEXT: addi r3, r3, .LCPI0_0@toc@l
22 ; CHECK-NEXT: lvx v3, 0, r3
23 ; CHECK-NEXT: vperm v2, v2, v2, v3
24 ; CHECK-NEXT: vxor v3, v3, v3
25 ; CHECK-NEXT: addi r3, r1, -48
26 ; CHECK-NEXT: stvx v3, 0, r3
27 ; CHECK-NEXT: addi r3, r1, -32
28 ; CHECK-NEXT: stvx v2, 0, r3
29 ; CHECK-NEXT: lwz r3, -48(r1)
30 ; CHECK-NEXT: lwz r4, -32(r1)
31 ; CHECK-NEXT: cmpw r4, r3
32 ; CHECK-NEXT: bc 12, gt, .LBB0_4
33 ; CHECK-NEXT: b .LBB0_5
34 ; CHECK-NEXT: .LBB0_4:
35 ; CHECK-NEXT: addi r3, r4, 0
36 ; CHECK-NEXT: .LBB0_5:
37 ; CHECK-NEXT: cmpw r3, r3
38 ; CHECK-NEXT: stw r3, -64(r1)
39 ; CHECK-NEXT: addi r3, r1, -64
40 ; CHECK-NEXT: lvx v2, 0, r3
41 ; CHECK-NEXT: addi r3, r1, -16
42 ; CHECK-NEXT: stvx v2, 0, r3
43 ; CHECK-NEXT: lfs f0, -16(r1)
44 ; CHECK-NEXT: .LBB0_6:
46 br i1 undef, label %7, label %1
49 br i1 undef, label %2, label %1
52 %3 = shufflevector <4 x float> %in, <4 x float> undef, <4 x i32> <i32 2, i32 3, i32 1, i32 0>
53 %4 = call fast <4 x float> @llvm.maxnum.v4f32(<4 x float> %3, <4 x float> zeroinitializer)
54 %5 = call fast <4 x float> @llvm.maxnum.v4f32(<4 x float> %4, <4 x float> undef)
55 %6 = extractelement <4 x float> %5, i32 0
59 %8 = phi float [ %6, %2 ], [ undef, %0 ]
60 %9 = fcmp fast une float %8, 0.000000e+00
64 declare <4 x float> @llvm.maxnum.v4f32(<4 x float>, <4 x float>) #0
66 attributes #0 = { nounwind optnone noinline }