brintos

brintos / llvm-project-archived public Read only

0
0
Text · 3.8 KiB · 0abf347 Raw
68 lines · plain
1# RUN: llc -mtriple=amdgcn -mcpu=gfx1250 -start-before=amdgpu-insert-delay-alu %s -o - | FileCheck %s2 3---4name: wmma_xdl_twoaddr_trans5tracksRegLiveness: true6body: |7  bb.0:8    ; CHECK-LABEL: {{^}}wmma_xdl_twoaddr_trans:9    ; CHECK: %bb.0:10    ; CHECK-NEXT: v_wmma_f32_16x16x64_fp8_fp8 v[8:15], v[0:7], v[0:7], v[8:15]11    ; CHECK-NEXT: v_exp_f32_e32 v16, v1612    ; CHECK-NEXT: s_delay_alu instid0(TRANS32_DEP_2)13    ; CHECK-NEXT: v_add_nc_u32_e32 v17, v17, v814    liveins: $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7, $vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, $vgpr16, $vgpr1715    $vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15 = V_WMMA_F32_16X16X64_FP8_FP8_w32_twoaddr $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7, $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7, 8, $vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, 0, 0, 0, 0, implicit $exec16    $vgpr16 = V_EXP_F32_e32 $vgpr16, implicit $exec, implicit $mode17    $vgpr17 = V_ADD_U32_e32 $vgpr17, $vgpr8, implicit $exec18...19 20---21name: wmma_xdl_threeaddr_trans22tracksRegLiveness: true23body: |24  bb.0:25    ; CHECK-LABEL: {{^}}wmma_xdl_threeaddr_trans:26    ; CHECK: %bb.0:27    ; CHECK-NEXT: v_wmma_f32_16x16x64_fp8_fp8 v[8:15], v[0:7], v[0:7], v[16:23]28    ; CHECK-NEXT: v_exp_f32_e32 v24, v2429    ; CHECK-NEXT: s_delay_alu instid0(TRANS32_DEP_2)30    ; CHECK-NEXT: v_add_nc_u32_e32 v25, v25, v831    liveins: $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7, $vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23, $vgpr24, $vgpr2532    $vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15 = V_WMMA_F32_16X16X64_FP8_FP8_w32_threeaddr $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7, $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7, 8, $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23, 0, 0, 0, 0, implicit $exec33    $vgpr24 = V_EXP_F32_e32 $vgpr24, implicit $exec, implicit $mode34    $vgpr25 = V_ADD_U32_e32 $vgpr25, $vgpr8, implicit $exec35...36 37name: swmmac_xdl_twoaddr_trans38tracksRegLiveness: true39body: |40  bb.0:41    ; CHECK-LABEL: {{^}}swmmac_xdl_twoaddr_trans:42    ; CHECK: %bb.0:43    ; CHECK-NEXT: v_swmmac_f16_16x16x128_bf8_bf8 v[24:27], v[0:7], v[8:23], v[28:29]44    ; CHECK-NEXT: v_exp_f32_e32 v30, v3045    ; CHECK-NEXT: s_delay_alu instid0(TRANS32_DEP_2)46    ; CHECK-NEXT: v_add_nc_u32_e32 v31, v31, v2447    liveins: $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7, $vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15_vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23, $vgpr24_vgpr25_vgpr26_vgpr27, $vgpr28, $vgpr29, $vgpr30, $vgpr3148    $vgpr24_vgpr25_vgpr26_vgpr27 = V_SWMMAC_F16_16X16X128_BF8_BF8_w32_twoaddr $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7, $vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15_vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23, $vgpr24_vgpr25_vgpr26_vgpr27, $vgpr28_vgpr29, 0, 0, 0, implicit $exec49    $vgpr30 = V_EXP_F32_e32 $vgpr30, implicit $exec, implicit $mode50    $vgpr31 = V_ADD_U32_e32 $vgpr31, $vgpr24, implicit $exec51...52 53name: wmma_non_xdl_large_data_valu54tracksRegLiveness: true55body: |56  bb.0:57    ; CHECK-LABEL: {{^}}wmma_non_xdl_large_data_valu:58    ; CHECK: %bb.0:59    ; CHECK-NEXT: v_wmma_f32_16x16x4_f32 v[4:11], v[0:1], v[2:3], v[4:11] matrix_b_reuse60    ; CHECK-NEXT: v_exp_f32_e32 v12, v1261    ; CHECK-NEXT: s_delay_alu instid0(VALU_DEP_1)62    ; CHECK-NEXT: v_add_nc_u32_e32 v13, v13, v863    liveins: $vgpr0_vgpr1_vgpr2_vgpr3, $vgpr4_vgpr5_vgpr6_vgpr7, $vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11, $vgpr12, $vgpr1364    $vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11 = V_WMMA_F32_16X16X4_F32_w32_twoaddr 8, $vgpr0_vgpr1, 8, $vgpr2_vgpr3, 8, $vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11, 0, -1, 0, 0, implicit $exec65    $vgpr12 = V_EXP_F32_e32 $vgpr12, implicit $exec, implicit $mode66    $vgpr13 = V_ADD_U32_e32 $vgpr13, $vgpr8, implicit $exec67...68