; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx90a -O1 < %s | FileCheck -check-prefix=GCN %s

define amdgpu_kernel void @test_iglp_opt_rev_mfma_gemm(<1 x i64> %L1) {
; GCN-LABEL: test_iglp_opt_rev_mfma_gemm:
; GCN:       ; %bb.0: ; %entry
; GCN-NEXT:    v_mov_b32_e32 v0, 0
; GCN-NEXT:    s_load_dwordx2 s[0:1], s[8:9], 0x0
; GCN-NEXT:    ds_read_b128 v[30:33], v0 offset:112
; GCN-NEXT:    ds_read_b128 v[26:29], v0 offset:96
; GCN-NEXT:    ds_read_b128 v[22:25], v0 offset:80
; GCN-NEXT:    ds_read_b128 v[18:21], v0 offset:64
; GCN-NEXT:    ds_read_b128 v[2:5], v0
; GCN-NEXT:    ds_read_b128 v[6:9], v0 offset:16
; GCN-NEXT:    ds_read_b128 v[10:13], v0 offset:32
; GCN-NEXT:    ds_read_b128 v[14:17], v0 offset:48
; GCN-NEXT:    v_mov_b32_e32 v34, 0
; GCN-NEXT:    v_mov_b32_e32 v35, 0
; GCN-NEXT:    s_waitcnt lgkmcnt(0)
; GCN-NEXT:    s_cmp_lg_u64 s[0:1], 0
; GCN-NEXT:    ; iglp_opt mask(0x00000001)
; GCN-NEXT:    ds_write_b128 v0, v[30:33] offset:112
; GCN-NEXT:    ds_write_b128 v0, v[26:29] offset:96
; GCN-NEXT:    ds_write_b128 v0, v[22:25] offset:80
; GCN-NEXT:    ds_write_b128 v0, v[18:21] offset:64
; GCN-NEXT:    ds_write_b128 v0, v[14:17] offset:48
; GCN-NEXT:    ds_write_b128 v0, v[10:13] offset:32
; GCN-NEXT:    ds_write_b128 v0, v[6:9] offset:16
; GCN-NEXT:    ds_write_b128 v0, v[2:5]
; GCN-NEXT:    ds_write_b64 v0, v[34:35]
; GCN-NEXT:    s_endpgm
entry:
  call void @llvm.amdgcn.iglp.opt(i32 1)
  %load.4 = load <32 x float>, ptr addrspace(3) zeroinitializer, align 128
  %B = urem <1 x i64> zeroinitializer, %L1
  store <32 x float> %load.4, ptr addrspace(3) zeroinitializer, align 128
  store <1 x i64> %B, ptr addrspace(3) zeroinitializer, align 8
  ret void
}

declare void @llvm.amdgcn.iglp.opt(i32 immarg) #0

attributes #0 = { convergent nocallback nofree nounwind willreturn }