1 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2 ; RUN: llc -mtriple=amdgcn -mcpu=verde -verify-machineinstrs < %s | FileCheck -check-prefix=SI %s
3 ; RUN: llc -mtriple=amdgcn -mcpu=tonga -verify-machineinstrs < %s | FileCheck -check-prefix=VI %s
5 define amdgpu_ps float @main(float %arg0, float %arg1) #0 {
8 ; SI-NEXT: image_load v2, v0, s[0:7] dmask:0x1 unorm
9 ; SI-NEXT: v_cvt_i32_f32_e32 v0, v0
10 ; SI-NEXT: v_and_b32_e32 v0, 7, v0
11 ; SI-NEXT: s_waitcnt vmcnt(0)
12 ; SI-NEXT: v_lshrrev_b32_e32 v0, v0, v2
13 ; SI-NEXT: v_and_b32_e32 v0, 1, v0
14 ; SI-NEXT: v_cmp_eq_u32_e32 vcc, 1, v0
15 ; SI-NEXT: v_cndmask_b32_e32 v0, 0, v1, vcc
16 ; SI-NEXT: v_cvt_pkrtz_f16_f32_e32 v0, s0, v0
17 ; SI-NEXT: ; return to shader part epilog
21 ; VI-NEXT: image_load v2, v0, s[0:7] dmask:0x1 unorm
22 ; VI-NEXT: v_cvt_i32_f32_e32 v0, v0
23 ; VI-NEXT: v_and_b32_e32 v0, 7, v0
24 ; VI-NEXT: s_waitcnt vmcnt(0)
25 ; VI-NEXT: v_lshrrev_b32_e32 v0, v0, v2
26 ; VI-NEXT: v_and_b32_e32 v0, 1, v0
27 ; VI-NEXT: v_cmp_eq_u32_e32 vcc, 1, v0
28 ; VI-NEXT: v_cndmask_b32_e32 v0, 0, v1, vcc
29 ; VI-NEXT: v_cvt_pkrtz_f16_f32 v0, s0, v0
30 ; VI-NEXT: ; return to shader part epilog
32 %tmp = fptosi float %arg0 to i32
33 %tmp1 = call <4 x float> @llvm.amdgcn.image.load.1d.v4f32.i32(i32 15, i32 undef, <8 x i32> undef, i32 0, i32 0)
34 %tmp2.f = extractelement <4 x float> %tmp1, i32 0
35 %tmp2 = bitcast float %tmp2.f to i32
36 %tmp3 = and i32 %tmp, 7
37 %tmp4 = shl i32 1, %tmp3
38 %tmp5 = and i32 %tmp2, %tmp4
39 %tmp6 = icmp eq i32 %tmp5, 0
40 %tmp7 = select i1 %tmp6, float 0.000000e+00, float %arg1
41 %tmp8 = call <2 x half> @llvm.amdgcn.cvt.pkrtz(float undef, float %tmp7)
42 %tmp9 = bitcast <2 x half> %tmp8 to float
46 declare <2 x half> @llvm.amdgcn.cvt.pkrtz(float, float) #1
47 declare <4 x float> @llvm.amdgcn.image.load.1d.v4f32.i32(i32, i32, <8 x i32>, i32, i32) #2
49 attributes #0 = { nounwind }
50 attributes #1 = { nounwind readnone }
51 attributes #2 = { nounwind readonly }