128 lines · plain
1# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py2# RUN: llc -mtriple=amdgcn -mcpu=gfx900 -verify-machineinstrs -run-pass si-insert-waitcnts -o - %s | FileCheck -check-prefix=GFX9 %s3 4# Two buffer loads with overlapping outputs. No waitcnt required.5---6name: buffer_buffer7tracksRegLiveness: true8body: |9 bb.0:10 liveins: $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr4, $sgpr511 ; GFX9-LABEL: name: buffer_buffer12 ; GFX9: liveins: $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr4, $sgpr513 ; GFX9-NEXT: {{ $}}14 ; GFX9-NEXT: S_WAITCNT 015 ; GFX9-NEXT: $vgpr0_vgpr1_vgpr2_vgpr3 = BUFFER_LOAD_DWORDX4_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr4, 0, 0, 0, implicit $exec16 ; GFX9-NEXT: $vgpr1 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr5, 0, 0, 0, implicit $exec17 $vgpr0_vgpr1_vgpr2_vgpr3 = BUFFER_LOAD_DWORDX4_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr4, 0, 0, 0, implicit $exec18 $vgpr1 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr5, 0, 0, 0, implicit $exec19...20 21# Two tbuffer loads with overlapping outputs. No waitcnt required.22---23name: tbuffer_tbuffer24tracksRegLiveness: true25body: |26 bb.0:27 liveins: $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr4, $sgpr528 ; GFX9-LABEL: name: tbuffer_tbuffer29 ; GFX9: liveins: $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr4, $sgpr530 ; GFX9-NEXT: {{ $}}31 ; GFX9-NEXT: S_WAITCNT 032 ; GFX9-NEXT: $vgpr0_vgpr1_vgpr2 = TBUFFER_LOAD_FORMAT_XYZ_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, 0, 4, 125, 0, 0, implicit $exec33 ; GFX9-NEXT: $vgpr0 = TBUFFER_LOAD_FORMAT_X_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, 0, 16, 116, 0, 0, implicit $exec34 $vgpr0_vgpr1_vgpr2 = TBUFFER_LOAD_FORMAT_XYZ_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, 0, 4, 125, 0, 0, implicit $exec35 $vgpr0 = TBUFFER_LOAD_FORMAT_X_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, 0, 16, 116, 0, 0, implicit $exec36...37 38# Two gathers with overlapping outputs. (Note gathers can't be trimmed because39# dmask means something different.) No waitcnt required.40---41name: gather_gather42tracksRegLiveness: true43body: |44 bb.0:45 liveins: $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7, $sgpr8_sgpr9_sgpr10_sgpr11, $vgpr0_vgpr1_vgpr2, $vgpr3_vgpr4_vgpr546 ; GFX9-LABEL: name: gather_gather47 ; GFX9: liveins: $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7, $sgpr8_sgpr9_sgpr10_sgpr11, $vgpr0_vgpr1_vgpr2, $vgpr3_vgpr4_vgpr548 ; GFX9-NEXT: {{ $}}49 ; GFX9-NEXT: S_WAITCNT 050 ; GFX9-NEXT: $vgpr10_vgpr11_vgpr12_vgpr13 = IMAGE_GATHER4_LZ_O_V4_V3 $vgpr0_vgpr1_vgpr2, $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7, $sgpr8_sgpr9_sgpr10_sgpr11, 1, 0, 0, 0, 0, 0, 0, 0, implicit $exec :: (load (s128))51 ; GFX9-NEXT: $vgpr13_vgpr14_vgpr15_vgpr16 = IMAGE_GATHER4_LZ_O_V4_V3 $vgpr3_vgpr4_vgpr5, $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7, $sgpr8_sgpr9_sgpr10_sgpr11, 1, 0, 0, 0, 0, 0, 0, 0, implicit $exec :: (load (s128))52 $vgpr10_vgpr11_vgpr12_vgpr13 = IMAGE_GATHER4_LZ_O_V4_V3 $vgpr0_vgpr1_vgpr2, $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7, $sgpr8_sgpr9_sgpr10_sgpr11, 1, 0, 0, 0, 0, 0, 0, 0, implicit $exec :: (load (s128))53 $vgpr13_vgpr14_vgpr15_vgpr16 = IMAGE_GATHER4_LZ_O_V4_V3 $vgpr3_vgpr4_vgpr5, $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7, $sgpr8_sgpr9_sgpr10_sgpr11, 1, 0, 0, 0, 0, 0, 0, 0, implicit $exec :: (load (s128))54...55 56# Image load vs image sample. Waitcnt required because they are not guaranteed57# to write their results in order, despite both using the s_waitcnt vmcnt58# counter.59---60name: nosampler_sampler61tracksRegLiveness: true62body: |63 bb.0:64 liveins: $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7, $sgpr8_sgpr9_sgpr10_sgpr11, $vgpr0_vgpr1_vgpr2_vgpr365 ; GFX9-LABEL: name: nosampler_sampler66 ; GFX9: liveins: $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7, $sgpr8_sgpr9_sgpr10_sgpr11, $vgpr0_vgpr1_vgpr2_vgpr367 ; GFX9-NEXT: {{ $}}68 ; GFX9-NEXT: S_WAITCNT 069 ; GFX9-NEXT: $vgpr4 = IMAGE_LOAD_V1_V4 $vgpr0_vgpr1_vgpr2_vgpr3, $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7, 2, -1, 0, 0, 0, 0, 0, 0, implicit $exec :: (load (s128))70 ; GFX9-NEXT: S_WAITCNT 395271 ; GFX9-NEXT: $vgpr4 = IMAGE_SAMPLE_L_V1_V4 $vgpr0_vgpr1_vgpr2_vgpr3, $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7, $sgpr8_sgpr9_sgpr10_sgpr11, 8, 0, 0, 0, 0, 0, -1, 0, implicit $exec :: (load (s128))72 $vgpr4 = IMAGE_LOAD_V1_V4 $vgpr0_vgpr1_vgpr2_vgpr3, $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7, 2, -1, 0, 0, 0, 0, 0, 0, implicit $exec :: (load (s128))73 $vgpr4 = IMAGE_SAMPLE_L_V1_V4 $vgpr0_vgpr1_vgpr2_vgpr3, $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7, $sgpr8_sgpr9_sgpr10_sgpr11, 8, 0, 0, 0, 0, 0, -1, 0, implicit $exec :: (load (s128))74...75# (global_load + scratch_load + buffer_load)76---77name: global_scratch_buffer78tracksRegLiveness: true79body: |80 bb.0:81 liveins: $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr4, $vgpr0_vgpr182 ; GFX9-LABEL: name: global_scratch_buffer83 ; GFX9: liveins: $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr4, $vgpr0_vgpr184 ; GFX9-NEXT: {{ $}}85 ; GFX9-NEXT: S_WAITCNT 086 ; GFX9-NEXT: $vgpr2 = GLOBAL_LOAD_DWORD $vgpr0_vgpr1, 0, 0, implicit $exec, implicit $flat_scr87 ; GFX9-NEXT: $vgpr2 = SCRATCH_LOAD_DWORD $vgpr0, 0, 0, implicit $exec, implicit $flat_scr88 ; GFX9-NEXT: $vgpr2 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr4, 0, 0, 0, implicit $exec89 $vgpr2 = GLOBAL_LOAD_DWORD $vgpr0_vgpr1, 0, 0, implicit $exec, implicit $flat_scr90 $vgpr2 = SCRATCH_LOAD_DWORD $vgpr0, 0, 0, implicit $exec, implicit $flat_scr91 $vgpr2 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr4, 0, 0, 0, implicit $exec92...93# waw between flat and buffer should have a wait inserted between.94# (flat + buffer)95---96name: flat_buffer97tracksRegLiveness: true98body: |99 bb.0:100 liveins: $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr4, $vgpr0_vgpr1101 ; GFX9-LABEL: name: flat_buffer102 ; GFX9: liveins: $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr4, $vgpr0_vgpr1103 ; GFX9-NEXT: {{ $}}104 ; GFX9-NEXT: S_WAITCNT 0105 ; GFX9-NEXT: $vgpr2 = FLAT_LOAD_DWORD $vgpr0_vgpr1, 0, 0, implicit $exec, implicit $flat_scr106 ; GFX9-NEXT: S_WAITCNT 49279107 ; GFX9-NEXT: $vgpr2 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr4, 0, 0, 0, implicit $exec108 $vgpr2 = FLAT_LOAD_DWORD $vgpr0_vgpr1, 0, 0, implicit $exec, implicit $flat_scr109 $vgpr2 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr4, 0, 0, 0, implicit $exec110...111# buffer + flat112---113name: buffer_flat114tracksRegLiveness: true115body: |116 bb.0:117 liveins: $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr4, $vgpr0_vgpr1118 ; GFX9-LABEL: name: buffer_flat119 ; GFX9: liveins: $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr4, $vgpr0_vgpr1120 ; GFX9-NEXT: {{ $}}121 ; GFX9-NEXT: S_WAITCNT 0122 ; GFX9-NEXT: $vgpr2 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr4, 0, 0, 0, implicit $exec123 ; GFX9-NEXT: S_WAITCNT 3952124 ; GFX9-NEXT: $vgpr2 = FLAT_LOAD_DWORD $vgpr0_vgpr1, 0, 0, implicit $exec, implicit $flat_scr125 $vgpr2 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr4, 0, 0, 0, implicit $exec126 $vgpr2 = FLAT_LOAD_DWORD $vgpr0_vgpr1, 0, 0, implicit $exec, implicit $flat_scr127...128