• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1; XFAIL: *
2; RUN: llc -verify-machineinstrs -march=amdgcn -mcpu=SI -mattr=-promote-alloca < %s | FileCheck -check-prefix=SI %s
3; RUN: llc -verify-machineinstrs -march=amdgcn -mcpu=tonga -mattr=-promote-alloca < %s | FileCheck -check-prefix=SI %s
4
5; 64-bit select was originally lowered with a build_pair, and this
6; could be simplified to 1 cndmask instead of 2, but that broken when
7; it started being implemented with a v2i32 build_vector and
8; bitcasting.
9define void @trunc_select_i64(i32 addrspace(1)* %out, i64 %a, i64 %b, i32 %c) {
10  %cmp = icmp eq i32 %c, 0
11  %select = select i1 %cmp, i64 %a, i64 %b
12  %trunc = trunc i64 %select to i32
13  store i32 %trunc, i32 addrspace(1)* %out, align 4
14  ret void
15}
16
17; FIXME: Fix truncating store for local memory
18; SI-LABEL: {{^}}trunc_load_alloca_i64:
19; SI: v_movrels_b32
20; SI-NOT: v_movrels_b32
21; SI: s_endpgm
22define void @trunc_load_alloca_i64(i64 addrspace(1)* %out, i32 %a, i32 %b) {
23  %idx = add i32 %a, %b
24  %alloca = alloca i64, i32 4
25  %gep0 = getelementptr i64, i64* %alloca, i64 0
26  %gep1 = getelementptr i64, i64* %alloca, i64 1
27  %gep2 = getelementptr i64, i64* %alloca, i64 2
28  %gep3 = getelementptr i64, i64* %alloca, i64 3
29  store i64 24, i64* %gep0, align 8
30  store i64 9334, i64* %gep1, align 8
31  store i64 3935, i64* %gep2, align 8
32  store i64 9342, i64* %gep3, align 8
33  %gep = getelementptr i64, i64* %alloca, i32 %idx
34  %load = load i64, i64* %gep, align 8
35  %mask = and i64 %load, 4294967296
36  %add = add i64 %mask, -1
37  store i64 %add, i64 addrspace(1)* %out, align 4
38  ret void
39}
40