llvm/test/CodeGen/AMDGPU/power-sched-no-cycle.mir

# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
# RUN: llc -mtriple=amdgcn -mcpu=gfx90a -run-pass=postmisched %s -o - -amdgpu-enable-power-sched=true  2>&1 | FileCheck %s
# This test represents a pattern which caused power-sched to introduce cycles into the Scheduling graph. By virtue of this test not crashing indicates it has completed succesfully.

---
name:            power_sched_cycle_condition
tracksRegLiveness: true
body:             |
  bb.0:
    liveins: $vgpr1, $sgpr2_sgpr3, $sgpr1, $vgpr72_vgpr73, $vgpr2_vgpr3, $sgpr22, $sgpr6_sgpr7, $sgpr10_sgpr11
    ; CHECK-LABEL: name: power_sched_cycle_condition
    ; CHECK: liveins: $vgpr1, $sgpr2_sgpr3, $sgpr1, $vgpr72_vgpr73, $vgpr2_vgpr3, $sgpr22, $sgpr6_sgpr7, $sgpr10_sgpr11
    ; CHECK-NEXT: {{  $}}
    ; CHECK-NEXT: $sgpr1 = S_LOAD_DWORD_IMM $sgpr2_sgpr3, 56, 0 :: (dereferenceable invariant load (s64), addrspace 4)
    ; CHECK-NEXT: $sgpr4 = S_LSHL_B32 killed $sgpr22, 1, implicit-def dead $scc
    ; CHECK-NEXT: $sgpr22_sgpr23 = S_LOAD_DWORDX2_IMM killed $sgpr2_sgpr3, 36, 0 :: (dereferenceable invariant load (s64), addrspace 4)
    ; CHECK-NEXT: $vgpr2 = nsw V_MUL_LO_U32_e64 killed $vgpr1, killed $sgpr1, implicit $exec
    ; CHECK-NEXT: early-clobber $vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15_vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21 = V_MFMA_F32_32X32X8F16_vgprcd_e64 killed $vgpr2_vgpr3, $vgpr72_vgpr73, 0, 0, 0, 0, implicit $mode, implicit $exec
    $sgpr1 = S_LOAD_DWORD_IMM $sgpr2_sgpr3, 56, 0 :: (dereferenceable invariant load (s64), addrspace 4)
    $vgpr2 = nsw V_MUL_LO_U32_e64 $vgpr1, $sgpr1, implicit $exec
    $vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15_vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21 =  V_MFMA_F32_32X32X8F16_vgprcd_e64 $vgpr2_vgpr3, $vgpr72_vgpr73, 0, 0, 0, 0, implicit $mode, implicit $exec
    $sgpr4 = S_LSHL_B32 $sgpr22, 1, implicit-def dead $scc
    $sgpr22_sgpr23 = S_LOAD_DWORDX2_IMM $sgpr2_sgpr3, 36, 0 :: (dereferenceable invariant load (s64), addrspace 4)
...