44 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 52; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx90a -O1 < %s | FileCheck -check-prefix=GCN %s3 4define amdgpu_kernel void @test_iglp_opt_rev_mfma_gemm(<1 x i64> %L1) {5; GCN-LABEL: test_iglp_opt_rev_mfma_gemm:6; GCN: ; %bb.0: ; %entry7; GCN-NEXT: v_mov_b32_e32 v0, 08; GCN-NEXT: ds_read_b128 v[2:5], v09; GCN-NEXT: s_load_dwordx2 s[0:1], s[8:9], 0x010; GCN-NEXT: ds_read_b128 v[30:33], v0 offset:11211; GCN-NEXT: ds_read_b128 v[26:29], v0 offset:9612; GCN-NEXT: ds_read_b128 v[22:25], v0 offset:8013; GCN-NEXT: ds_read_b128 v[18:21], v0 offset:6414; GCN-NEXT: ds_read_b128 v[6:9], v0 offset:1615; GCN-NEXT: ds_read_b128 v[10:13], v0 offset:3216; GCN-NEXT: ds_read_b128 v[14:17], v0 offset:4817; GCN-NEXT: s_waitcnt lgkmcnt(0)18; GCN-NEXT: ds_write_b128 v0, v[2:5]19; GCN-NEXT: v_mov_b32_e32 v2, 020; GCN-NEXT: v_mov_b32_e32 v3, 021; GCN-NEXT: s_cmp_lg_u64 s[0:1], 022; GCN-NEXT: ; iglp_opt mask(0x00000001)23; GCN-NEXT: ds_write_b128 v0, v[30:33] offset:11224; GCN-NEXT: ds_write_b128 v0, v[26:29] offset:9625; GCN-NEXT: ds_write_b128 v0, v[22:25] offset:8026; GCN-NEXT: ds_write_b128 v0, v[18:21] offset:6427; GCN-NEXT: ds_write_b128 v0, v[14:17] offset:4828; GCN-NEXT: ds_write_b128 v0, v[10:13] offset:3229; GCN-NEXT: ds_write_b128 v0, v[6:9] offset:1630; GCN-NEXT: ds_write_b64 v0, v[2:3]31; GCN-NEXT: s_endpgm32entry:33 call void @llvm.amdgcn.iglp.opt(i32 1)34 %load.4 = load <32 x float>, ptr addrspace(3) null, align 12835 %B = urem <1 x i64> zeroinitializer, %L136 store <32 x float> %load.4, ptr addrspace(3) null, align 12837 store <1 x i64> %B, ptr addrspace(3) null, align 838 ret void39}40 41declare void @llvm.amdgcn.iglp.opt(i32 immarg) #042 43attributes #0 = { convergent nocallback nofree nounwind willreturn }44