• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2; RUN: llc -global-isel -march=amdgcn -mcpu=hawaii -mattr=+flat-for-global -verify-machineinstrs < %s | FileCheck -check-prefix=GCN %s
3
4; End to end tests for scalar vs. vector boolean legalization strategies.
5
6define amdgpu_ps float @select_vgpr_sgpr_trunc_cond(i32 inreg %a, i32 %b, i32 %c) {
7; GCN-LABEL: select_vgpr_sgpr_trunc_cond:
8; GCN:       ; %bb.0:
9; GCN-NEXT:    s_and_b32 s0, 1, s0
10; GCN-NEXT:    v_cmp_ne_u32_e64 vcc, 0, s0
11; GCN-NEXT:    v_cndmask_b32_e32 v0, v1, v0, vcc
12; GCN-NEXT:    ; return to shader part epilog
13  %cc = trunc i32 %a to i1
14  %r = select i1 %cc, i32 %b, i32 %c
15  %r.f = bitcast i32 %r to float
16  ret float %r.f
17}
18
19define amdgpu_ps float @select_vgpr_sgpr_trunc_and_cond(i32 inreg %a.0, i32 inreg %a.1, i32 %b, i32 %c) {
20; GCN-LABEL: select_vgpr_sgpr_trunc_and_cond:
21; GCN:       ; %bb.0:
22; GCN-NEXT:    s_and_b32 s0, s0, s1
23; GCN-NEXT:    s_and_b32 s0, 1, s0
24; GCN-NEXT:    v_cmp_ne_u32_e64 vcc, 0, s0
25; GCN-NEXT:    v_cndmask_b32_e32 v0, v1, v0, vcc
26; GCN-NEXT:    ; return to shader part epilog
27  %cc.0 = trunc i32 %a.0 to i1
28  %cc.1 = trunc i32 %a.1 to i1
29  %and = and i1 %cc.0, %cc.1
30  %r = select i1 %and, i32 %b, i32 %c
31  %r.f = bitcast i32 %r to float
32  ret float %r.f
33}
34
35define amdgpu_ps i32 @select_sgpr_trunc_and_cond(i32 inreg %a.0, i32 inreg %a.1, i32 inreg %b, i32 inreg %c) {
36; GCN-LABEL: select_sgpr_trunc_and_cond:
37; GCN:       ; %bb.0:
38; GCN-NEXT:    s_and_b32 s0, s0, s1
39; GCN-NEXT:    s_and_b32 s0, s0, 1
40; GCN-NEXT:    s_cmp_lg_u32 s0, 0
41; GCN-NEXT:    s_cselect_b32 s0, s2, s3
42; GCN-NEXT:    ; return to shader part epilog
43  %cc.0 = trunc i32 %a.0 to i1
44  %cc.1 = trunc i32 %a.1 to i1
45  %and = and i1 %cc.0, %cc.1
46  %r = select i1 %and, i32 %b, i32 %c
47  ret i32 %r
48}
49
50define amdgpu_kernel void @sgpr_trunc_brcond(i32 %cond) {
51; GCN-LABEL: sgpr_trunc_brcond:
52; GCN:       ; %bb.0: ; %entry
53; GCN-NEXT:    s_load_dword s0, s[0:1], 0x9
54; GCN-NEXT:    s_waitcnt lgkmcnt(0)
55; GCN-NEXT:    s_xor_b32 s0, s0, -1
56; GCN-NEXT:    s_and_b32 s0, s0, 1
57; GCN-NEXT:    s_cmp_lg_u32 s0, 0
58; GCN-NEXT:    s_cbranch_scc1 BB3_2
59; GCN-NEXT:  ; %bb.1: ; %bb0
60; GCN-NEXT:    v_mov_b32_e32 v0, 0
61; GCN-NEXT:    flat_store_dword v[0:1], v0
62; GCN-NEXT:  BB3_2: ; %bb1
63; GCN-NEXT:    v_mov_b32_e32 v0, 1
64; GCN-NEXT:    flat_store_dword v[0:1], v0
65entry:
66  %trunc = trunc i32 %cond to i1
67  br i1 %trunc, label %bb0, label %bb1
68
69bb0:
70  store volatile i32 0, i32 addrspace(1)* undef
71  unreachable
72
73bb1:
74  store volatile i32 1, i32 addrspace(1)* undef
75  unreachable
76}
77
78define amdgpu_kernel void @brcond_sgpr_trunc_and(i32 %cond0, i32 %cond1) {
79; GCN-LABEL: brcond_sgpr_trunc_and:
80; GCN:       ; %bb.0: ; %entry
81; GCN-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
82; GCN-NEXT:    s_waitcnt lgkmcnt(0)
83; GCN-NEXT:    s_and_b32 s0, s0, s1
84; GCN-NEXT:    s_xor_b32 s0, s0, -1
85; GCN-NEXT:    s_and_b32 s0, s0, 1
86; GCN-NEXT:    s_cmp_lg_u32 s0, 0
87; GCN-NEXT:    s_cbranch_scc1 BB4_2
88; GCN-NEXT:  ; %bb.1: ; %bb0
89; GCN-NEXT:    v_mov_b32_e32 v0, 0
90; GCN-NEXT:    flat_store_dword v[0:1], v0
91; GCN-NEXT:  BB4_2: ; %bb1
92; GCN-NEXT:    v_mov_b32_e32 v0, 1
93; GCN-NEXT:    flat_store_dword v[0:1], v0
94entry:
95  %trunc0 = trunc i32 %cond0 to i1
96  %trunc1 = trunc i32 %cond1 to i1
97  %and = and i1 %trunc0, %trunc1
98  br i1 %and, label %bb0, label %bb1
99
100bb0:
101  store volatile i32 0, i32 addrspace(1)* undef
102  unreachable
103
104bb1:
105  store volatile i32 1, i32 addrspace(1)* undef
106  unreachable
107}
108