1 # RUN: llc -march=amdgcn -mcpu=gfx900 -run-pass=machine-scheduler -o - %s | FileCheck %s
4 # Check that the high latency loads are both scheduled first, before the
5 # multiplies, despite the presence of a barrier in the function.
6 # CHECK: BUFFER_LOAD_DWORD_OFFSET
7 # CHECK: BUFFER_LOAD_DWORD_OFFSET
11 tracksRegLiveness: true
14 liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8, $vgpr9
16 undef %43.sub3:vreg_128 = COPY $vgpr9
17 undef %42.sub2:vreg_128 = COPY $vgpr8
18 undef %41.sub1:vreg_128 = COPY $vgpr7
19 undef %26.sub0:vreg_128 = COPY $vgpr6
20 undef %46.sub3:vreg_128 = COPY $vgpr5
21 undef %45.sub2:vreg_128 = COPY $vgpr4
22 undef %44.sub1:vreg_128 = COPY $vgpr3
23 undef %32.sub0:vreg_128 = COPY $vgpr2
24 undef %38.sub1:vreg_64 = COPY $vgpr1
25 %38.sub0:vreg_64 = COPY $vgpr0
29 undef %33.sub0:sgpr_128 = V_READFIRSTLANE_B32 %32.sub0, implicit $exec
30 %33.sub1:sgpr_128 = V_READFIRSTLANE_B32 %44.sub1, implicit $exec
31 %33.sub2:sgpr_128 = V_READFIRSTLANE_B32 %45.sub2, implicit $exec
32 %33.sub3:sgpr_128 = V_READFIRSTLANE_B32 %46.sub3, implicit $exec
33 %15:vgpr_32 = BUFFER_LOAD_DWORD_OFFSET %33, 0, 0, 0, 0, 0, 0, 0, implicit $exec
34 %39:vgpr_32 = V_MUL_LO_U32 %15, %15, implicit $exec
36 undef %27.sub0:sgpr_128 = V_READFIRSTLANE_B32 %26.sub0, implicit $exec
37 %27.sub1:sgpr_128 = V_READFIRSTLANE_B32 %41.sub1, implicit $exec
38 %27.sub2:sgpr_128 = V_READFIRSTLANE_B32 %42.sub2, implicit $exec
39 %27.sub3:sgpr_128 = V_READFIRSTLANE_B32 %43.sub3, implicit $exec
40 %19:vgpr_32 = BUFFER_LOAD_DWORD_OFFSET %27, 0, 0, 0, 0, 0, 0, 0, implicit $exec
41 %40:vgpr_32 = V_MUL_LO_U32 %19, %19, implicit $exec
43 %23:vgpr_32 = V_ADD_U32_e32 %39, %40, implicit $exec
44 GLOBAL_STORE_DWORD %38, %23, 0, 0, 0, 0, implicit $exec