580 lines · plain
1# RUN: llc -mtriple=amdgcn -mcpu=carrizo -verify-machineinstrs -run-pass post-RA-hazard-rec %s -o - | FileCheck -check-prefixes=GCN,XNACK %s2 3# RUN: llc -mtriple=amdgcn -verify-machineinstrs -mattr=+volcanic-islands -run-pass post-RA-hazard-rec %s -o - | FileCheck -check-prefixes=GCN %s4# RUN: llc -mtriple=amdgcn -mcpu=fiji -verify-machineinstrs -run-pass post-RA-hazard-rec %s -o - | FileCheck --check-prefix=GCN %s5---6# Trivial clause at beginning of program7name: trivial_clause_load_flat4_x18 9body: |10 bb.0:11 ; GCN-LABEL: name: trivial_clause_load_flat4_x112 ; GCN: $vgpr0 = FLAT_LOAD_DWORD $vgpr2_vgpr3, 0, 0, implicit $exec, implicit $flat_scr13 ; GCN-NEXT: S_ENDPGM 014 15 $vgpr0 = FLAT_LOAD_DWORD $vgpr2_vgpr3, 0, 0, implicit $exec, implicit $flat_scr16 S_ENDPGM 017...18---19# Trivial clause at beginning of program20name: trivial_clause_load_flat4_x221 22body: |23 bb.0:24 ; GCN-LABEL: name: trivial_clause_load_flat4_x225 ; GCN: $vgpr0 = FLAT_LOAD_DWORD $vgpr2_vgpr3, 0, 0, implicit $exec, implicit $flat_scr26 ; GCN-NEXT: $vgpr1 = FLAT_LOAD_DWORD $vgpr4_vgpr5, 0, 0, implicit $exec, implicit $flat_scr27 ; GCN-NEXT: S_ENDPGM 028 29 $vgpr0 = FLAT_LOAD_DWORD $vgpr2_vgpr3, 0, 0, implicit $exec, implicit $flat_scr30 $vgpr1 = FLAT_LOAD_DWORD $vgpr4_vgpr5, 0, 0, implicit $exec, implicit $flat_scr31 S_ENDPGM 032...33---34# Trivial clause at beginning of program35name: trivial_clause_load_flat4_x336 37body: |38 bb.0:39 ; GCN-LABEL: name: trivial_clause_load_flat4_x340 ; GCN: $vgpr0 = FLAT_LOAD_DWORD $vgpr3_vgpr4, 0, 0, implicit $exec, implicit $flat_scr41 ; GCN-NEXT: $vgpr1 = FLAT_LOAD_DWORD $vgpr5_vgpr6, 0, 0, implicit $exec, implicit $flat_scr42 ; GCN-NEXT: $vgpr2 = FLAT_LOAD_DWORD $vgpr7_vgpr8, 0, 0, implicit $exec, implicit $flat_scr43 ; GCN-NEXT: S_ENDPGM 044 45 $vgpr0 = FLAT_LOAD_DWORD $vgpr3_vgpr4, 0, 0, implicit $exec, implicit $flat_scr46 $vgpr1 = FLAT_LOAD_DWORD $vgpr5_vgpr6, 0, 0, implicit $exec, implicit $flat_scr47 $vgpr2 = FLAT_LOAD_DWORD $vgpr7_vgpr8, 0, 0, implicit $exec, implicit $flat_scr48 S_ENDPGM 049...50---51# Trivial clause at beginning of program52name: trivial_clause_load_flat4_x453 54body: |55 bb.0:56 ; GCN-LABEL: name: trivial_clause_load_flat4_x457 ; GCN: $vgpr0 = FLAT_LOAD_DWORD $vgpr4_vgpr5, 0, 0, implicit $exec, implicit $flat_scr58 ; GCN-NEXT: $vgpr1 = FLAT_LOAD_DWORD $vgpr6_vgpr7, 0, 0, implicit $exec, implicit $flat_scr59 ; GCN-NEXT: $vgpr2 = FLAT_LOAD_DWORD $vgpr8_vgpr9, 0, 0, implicit $exec, implicit $flat_scr60 ; GCN-NEXT: $vgpr3 = FLAT_LOAD_DWORD $vgpr10_vgpr11, 0, 0, implicit $exec, implicit $flat_scr61 ; GCN-NEXT: S_ENDPGM 062 63 $vgpr0 = FLAT_LOAD_DWORD $vgpr4_vgpr5, 0, 0, implicit $exec, implicit $flat_scr64 $vgpr1 = FLAT_LOAD_DWORD $vgpr6_vgpr7, 0, 0, implicit $exec, implicit $flat_scr65 $vgpr2 = FLAT_LOAD_DWORD $vgpr8_vgpr9, 0, 0, implicit $exec, implicit $flat_scr66 $vgpr3 = FLAT_LOAD_DWORD $vgpr10_vgpr11, 0, 0, implicit $exec, implicit $flat_scr67 S_ENDPGM 068...69---70# Reuse of same input pointer is OK71 72name: trivial_clause_load_flat4_x2_sameptr73body: |74 bb.0:75 ; GCN-LABEL: name: trivial_clause_load_flat4_x2_sameptr76 ; GCN: $vgpr0 = FLAT_LOAD_DWORD $vgpr2_vgpr3, 0, 0, implicit $exec, implicit $flat_scr77 ; GCN-NEXT: $vgpr1 = FLAT_LOAD_DWORD $vgpr2_vgpr3, 0, 0, implicit $exec, implicit $flat_scr78 ; GCN-NEXT: S_ENDPGM 079 80 $vgpr0 = FLAT_LOAD_DWORD $vgpr2_vgpr3, 0, 0, implicit $exec, implicit $flat_scr81 $vgpr1 = FLAT_LOAD_DWORD $vgpr2_vgpr3, 0, 0, implicit $exec, implicit $flat_scr82 S_ENDPGM 083...84---85# 32-bit load partially clobbers its own ptr reg86name: flat_load4_overwrite_ptr_lo87 88body: |89 bb.0:90 ; GCN-LABEL: name: flat_load4_overwrite_ptr_lo91 ; GCN: $vgpr0 = FLAT_LOAD_DWORD $vgpr0_vgpr1, 0, 0, implicit $exec, implicit $flat_scr92 ; GCN-NEXT: S_ENDPGM 093 94 $vgpr0 = FLAT_LOAD_DWORD $vgpr0_vgpr1, 0, 0, implicit $exec, implicit $flat_scr95 S_ENDPGM 096...97---98# 32-bit load partially clobbers its own ptr reg99name: flat_load4_overwrite_ptr_hi100 101body: |102 bb.0:103 ; GCN-LABEL: name: flat_load4_overwrite_ptr_hi104 ; GCN: $vgpr1 = FLAT_LOAD_DWORD $vgpr0_vgpr1, 0, 0, implicit $exec, implicit $flat_scr105 ; GCN-NEXT: S_ENDPGM 0106 107 $vgpr1 = FLAT_LOAD_DWORD $vgpr0_vgpr1, 0, 0, implicit $exec, implicit $flat_scr108 S_ENDPGM 0109...110---111# 64-bit load clobbers its own ptr reg112name: flat_load8_overwrite_ptr113 114body: |115 bb.0:116 ; GCN-LABEL: name: flat_load8_overwrite_ptr117 ; GCN: $vgpr2_vgpr3 = FLAT_LOAD_DWORDX2 $vgpr2_vgpr3, 0, 0, implicit $exec, implicit $flat_scr118 ; GCN-NEXT: S_ENDPGM 0119 120 $vgpr2_vgpr3 = FLAT_LOAD_DWORDX2 $vgpr2_vgpr3, 0, 0, implicit $exec, implicit $flat_scr121 S_ENDPGM 0122...123---124# vmcnt has 4 bits, so maximum 16 outstanding loads. The waitcnt125# breaks the clause.126 127 128name: break_clause_at_max_clause_size_flat_load4129 130body: |131 bb.0:132 ; GCN-LABEL: name: break_clause_at_max_clause_size_flat_load4133 ; GCN: $vgpr2 = FLAT_LOAD_DWORD $vgpr0_vgpr1, 0, 0, implicit $exec, implicit $flat_scr134 ; GCN-NEXT: $vgpr3 = FLAT_LOAD_DWORD $vgpr0_vgpr1, 0, 0, implicit $exec, implicit $flat_scr135 ; GCN-NEXT: $vgpr4 = FLAT_LOAD_DWORD $vgpr0_vgpr1, 0, 0, implicit $exec, implicit $flat_scr136 ; GCN-NEXT: $vgpr5 = FLAT_LOAD_DWORD $vgpr0_vgpr1, 0, 0, implicit $exec, implicit $flat_scr137 ; GCN-NEXT: $vgpr6 = FLAT_LOAD_DWORD $vgpr0_vgpr1, 0, 0, implicit $exec, implicit $flat_scr138 ; GCN-NEXT: $vgpr7 = FLAT_LOAD_DWORD $vgpr0_vgpr1, 0, 0, implicit $exec, implicit $flat_scr139 ; GCN-NEXT: $vgpr8 = FLAT_LOAD_DWORD $vgpr0_vgpr1, 0, 0, implicit $exec, implicit $flat_scr140 ; GCN-NEXT: $vgpr9 = FLAT_LOAD_DWORD $vgpr0_vgpr1, 0, 0, implicit $exec, implicit $flat_scr141 ; GCN-NEXT: $vgpr10 = FLAT_LOAD_DWORD $vgpr0_vgpr1, 0, 0, implicit $exec, implicit $flat_scr142 ; GCN-NEXT: $vgpr11 = FLAT_LOAD_DWORD $vgpr0_vgpr1, 0, 0, implicit $exec, implicit $flat_scr143 ; GCN-NEXT: $vgpr12 = FLAT_LOAD_DWORD $vgpr0_vgpr1, 0, 0, implicit $exec, implicit $flat_scr144 ; GCN-NEXT: $vgpr13 = FLAT_LOAD_DWORD $vgpr0_vgpr1, 0, 0, implicit $exec, implicit $flat_scr145 ; GCN-NEXT: $vgpr14 = FLAT_LOAD_DWORD $vgpr0_vgpr1, 0, 0, implicit $exec, implicit $flat_scr146 ; GCN-NEXT: $vgpr15 = FLAT_LOAD_DWORD $vgpr0_vgpr1, 0, 0, implicit $exec, implicit $flat_scr147 ; GCN-NEXT: $vgpr16 = FLAT_LOAD_DWORD $vgpr0_vgpr1, 0, 0, implicit $exec, implicit $flat_scr148 ; GCN-NEXT: $vgpr17 = FLAT_LOAD_DWORD $vgpr0_vgpr1, 0, 0, implicit $exec, implicit $flat_scr149 ; XNACK-NEXT: S_NOP 0150 ; GCN-NEXT: $vgpr0 = FLAT_LOAD_DWORD $vgpr2_vgpr3, 0, 0, implicit $exec, implicit $flat_scr151 ; GCN-NEXT: $sgpr0 = S_MOV_B32 $sgpr0, implicit $vgpr2, implicit $vgpr3, implicit $vgpr4, implicit $vgpr5, implicit $vgpr6, implicit $vgpr7, implicit $vgpr8, implicit $vgpr9, implicit $vgpr10, implicit $vgpr11, implicit $vgpr12, implicit $vgpr13, implicit $vgpr14, implicit $vgpr15, implicit $vgpr16, implicit $vgpr17, implicit $vgpr18152 ; GCN-NEXT: S_ENDPGM 0153 154 $vgpr2 = FLAT_LOAD_DWORD $vgpr0_vgpr1, 0, 0, implicit $exec, implicit $flat_scr155 $vgpr3 = FLAT_LOAD_DWORD $vgpr0_vgpr1, 0, 0, implicit $exec, implicit $flat_scr156 $vgpr4 = FLAT_LOAD_DWORD $vgpr0_vgpr1, 0, 0, implicit $exec, implicit $flat_scr157 $vgpr5 = FLAT_LOAD_DWORD $vgpr0_vgpr1, 0, 0, implicit $exec, implicit $flat_scr158 159 $vgpr6 = FLAT_LOAD_DWORD $vgpr0_vgpr1, 0, 0, implicit $exec, implicit $flat_scr160 $vgpr7 = FLAT_LOAD_DWORD $vgpr0_vgpr1, 0, 0, implicit $exec, implicit $flat_scr161 $vgpr8 = FLAT_LOAD_DWORD $vgpr0_vgpr1, 0, 0, implicit $exec, implicit $flat_scr162 $vgpr9 = FLAT_LOAD_DWORD $vgpr0_vgpr1, 0, 0, implicit $exec, implicit $flat_scr163 164 $vgpr10 = FLAT_LOAD_DWORD $vgpr0_vgpr1, 0, 0, implicit $exec, implicit $flat_scr165 $vgpr11 = FLAT_LOAD_DWORD $vgpr0_vgpr1, 0, 0, implicit $exec, implicit $flat_scr166 $vgpr12 = FLAT_LOAD_DWORD $vgpr0_vgpr1, 0, 0, implicit $exec, implicit $flat_scr167 $vgpr13 = FLAT_LOAD_DWORD $vgpr0_vgpr1, 0, 0, implicit $exec, implicit $flat_scr168 169 $vgpr14 = FLAT_LOAD_DWORD $vgpr0_vgpr1, 0, 0, implicit $exec, implicit $flat_scr170 $vgpr15 = FLAT_LOAD_DWORD $vgpr0_vgpr1, 0, 0, implicit $exec, implicit $flat_scr171 $vgpr16 = FLAT_LOAD_DWORD $vgpr0_vgpr1, 0, 0, implicit $exec, implicit $flat_scr172 $vgpr17 = FLAT_LOAD_DWORD $vgpr0_vgpr1, 0, 0, implicit $exec, implicit $flat_scr173 174 $vgpr0 = FLAT_LOAD_DWORD $vgpr2_vgpr3, 0, 0, implicit $exec, implicit $flat_scr175 $sgpr0 = S_MOV_B32 $sgpr0, implicit $vgpr2, implicit $vgpr3, implicit $vgpr4, implicit $vgpr5, implicit $vgpr6, implicit $vgpr7, implicit $vgpr8, implicit $vgpr9, implicit $vgpr10, implicit $vgpr11, implicit $vgpr12, implicit $vgpr13, implicit $vgpr14, implicit $vgpr15, implicit $vgpr16, implicit $vgpr17, implicit $vgpr18176 S_ENDPGM 0177...178---179 180name: break_clause_simple_load_flat4_lo_ptr181 182body: |183 bb.0:184 ; GCN-LABEL: name: break_clause_simple_load_flat4_lo_ptr185 ; GCN: $vgpr0 = FLAT_LOAD_DWORD $vgpr2_vgpr3, 0, 0, implicit $exec, implicit $flat_scr186 ; XNACK-NEXT: S_NOP 0187 ; GCN-NEXT: $vgpr2 = FLAT_LOAD_DWORD $vgpr4_vgpr5, 0, 0, implicit $exec, implicit $flat_scr188 ; GCN-NEXT: S_ENDPGM 0189 190 $vgpr0 = FLAT_LOAD_DWORD $vgpr2_vgpr3, 0, 0, implicit $exec, implicit $flat_scr191 $vgpr2 = FLAT_LOAD_DWORD $vgpr4_vgpr5, 0, 0, implicit $exec, implicit $flat_scr192 S_ENDPGM 0193...194---195 196name: break_clause_simple_load_flat4_hi_ptr197 198body: |199 bb.0:200 ; GCN-LABEL: name: break_clause_simple_load_flat4_hi_ptr201 ; GCN: $vgpr0 = FLAT_LOAD_DWORD $vgpr2_vgpr3, 0, 0, implicit $exec, implicit $flat_scr202 ; XNACK-NEXT: S_NOP 0203 ; GCN-NEXT: $vgpr3 = FLAT_LOAD_DWORD $vgpr4_vgpr5, 0, 0, implicit $exec, implicit $flat_scr204 ; GCN-NEXT: S_ENDPGM 0205 206 $vgpr0 = FLAT_LOAD_DWORD $vgpr2_vgpr3, 0, 0, implicit $exec, implicit $flat_scr207 $vgpr3 = FLAT_LOAD_DWORD $vgpr4_vgpr5, 0, 0, implicit $exec, implicit $flat_scr208 S_ENDPGM 0209...210---211 212name: break_clause_simple_load_flat8_ptr213 214body: |215 bb.0:216 ; GCN-LABEL: name: break_clause_simple_load_flat8_ptr217 ; GCN: $vgpr0_vgpr1 = FLAT_LOAD_DWORDX2 $vgpr2_vgpr3, 0, 0, implicit $exec, implicit $flat_scr218 ; XNACK-NEXT: S_NOP 0219 ; GCN-NEXT: $vgpr2_vgpr3 = FLAT_LOAD_DWORDX2 $vgpr4_vgpr5, 0, 0, implicit $exec, implicit $flat_scr220 ; GCN-NEXT: S_ENDPGM 0221 222 $vgpr0_vgpr1 = FLAT_LOAD_DWORDX2 $vgpr2_vgpr3, 0, 0, implicit $exec, implicit $flat_scr223 $vgpr2_vgpr3 = FLAT_LOAD_DWORDX2 $vgpr4_vgpr5, 0, 0, implicit $exec, implicit $flat_scr224 S_ENDPGM 0225...226---227 228 229name: break_clause_simple_load_flat16_ptr230 231body: |232 bb.0:233 ; GCN-LABEL: name: break_clause_simple_load_flat16_ptr234 ; GCN: $vgpr0_vgpr1 = FLAT_LOAD_DWORDX2 $vgpr2_vgpr3, 0, 0, implicit $exec, implicit $flat_scr235 ; XNACK-NEXT: S_NOP 0236 ; GCN-NEXT: $vgpr2_vgpr3_vgpr4_vgpr5 = FLAT_LOAD_DWORDX4 $vgpr6_vgpr7, 0, 0, implicit $exec, implicit $flat_scr237 ; GCN-NEXT: S_ENDPGM 0238 $vgpr0_vgpr1 = FLAT_LOAD_DWORDX2 $vgpr2_vgpr3, 0, 0, implicit $exec, implicit $flat_scr239 $vgpr2_vgpr3_vgpr4_vgpr5 = FLAT_LOAD_DWORDX4 $vgpr6_vgpr7, 0, 0, implicit $exec, implicit $flat_scr240 S_ENDPGM 0241...242---243 244# The clause is broken by the waitcnt inserted at the end of the245# block, so no nop is needed.246 247 248name: break_clause_block_boundary_load_flat8_ptr249 250body: |251 ; GCN-LABEL: name: break_clause_block_boundary_load_flat8_ptr252 ; GCN: bb.0:253 ; GCN-NEXT: successors: %bb.1(0x80000000)254 ; GCN: $vgpr0_vgpr1 = FLAT_LOAD_DWORDX2 $vgpr2_vgpr3, 0, 0, implicit $exec, implicit $flat_scr255 ; GCN: bb.1:256 ; XNACK-NEXT: S_NOP 0257 ; GCN-NEXT: $vgpr2_vgpr3 = FLAT_LOAD_DWORDX2 $vgpr4_vgpr5, 0, 0, implicit $exec, implicit $flat_scr258 ; GCN-NEXT: S_ENDPGM 0259 260 bb.0:261 $vgpr0_vgpr1 = FLAT_LOAD_DWORDX2 $vgpr2_vgpr3, 0, 0, implicit $exec, implicit $flat_scr262 263 bb.1:264 $vgpr2_vgpr3 = FLAT_LOAD_DWORDX2 $vgpr4_vgpr5, 0, 0, implicit $exec, implicit $flat_scr265 S_ENDPGM 0266...267---268# The load clobbers the pointer of the store, so it needs to break.269 270name: break_clause_store_load_into_ptr_flat4271 272body: |273 bb.0:274 ; GCN-LABEL: name: break_clause_store_load_into_ptr_flat4275 ; GCN: FLAT_STORE_DWORD $vgpr2_vgpr3, $vgpr0, 0, 0, implicit $exec, implicit $flat_scr276 ; GCN-NEXT: $vgpr2 = FLAT_LOAD_DWORD $vgpr4_vgpr5, 0, 0, implicit $exec, implicit $flat_scr277 ; GCN-NEXT: S_ENDPGM 0278 279 FLAT_STORE_DWORD $vgpr2_vgpr3, $vgpr0, 0, 0, implicit $exec, implicit $flat_scr280 $vgpr2 = FLAT_LOAD_DWORD $vgpr4_vgpr5, 0, 0, implicit $exec, implicit $flat_scr281 S_ENDPGM 0282...283---284# The load clobbers the data of the store, so it needs to break.285# FIXME: Would it be better to s_nop and wait later?286 287name: break_clause_store_load_into_data_flat4288 289body: |290 bb.0:291 ; GCN-LABEL: name: break_clause_store_load_into_data_flat4292 ; GCN: FLAT_STORE_DWORD $vgpr2_vgpr3, $vgpr0, 0, 0, implicit $exec, implicit $flat_scr293 ; GCN-NEXT: $vgpr0 = FLAT_LOAD_DWORD $vgpr4_vgpr5, 0, 0, implicit $exec, implicit $flat_scr294 ; GCN-NEXT: S_ENDPGM 0295 296 FLAT_STORE_DWORD $vgpr2_vgpr3, $vgpr0, 0, 0, implicit $exec, implicit $flat_scr297 $vgpr0 = FLAT_LOAD_DWORD $vgpr4_vgpr5, 0, 0, implicit $exec, implicit $flat_scr298 S_ENDPGM 0299...300---301# Regular VALU instruction breaks clause, no nop needed302 303name: valu_inst_breaks_clause304 305body: |306 bb.0:307 ; GCN-LABEL: name: valu_inst_breaks_clause308 ; GCN: $vgpr0 = FLAT_LOAD_DWORD $vgpr2_vgpr3, 0, 0, implicit $exec, implicit $flat_scr309 ; GCN-NEXT: $vgpr8 = V_MOV_B32_e32 0, implicit $exec310 ; GCN-NEXT: $vgpr2 = FLAT_LOAD_DWORD $vgpr4_vgpr5, 0, 0, implicit $exec, implicit $flat_scr311 ; GCN-NEXT: S_ENDPGM 0312 313 $vgpr0 = FLAT_LOAD_DWORD $vgpr2_vgpr3, 0, 0, implicit $exec, implicit $flat_scr314 $vgpr8 = V_MOV_B32_e32 0, implicit $exec315 $vgpr2 = FLAT_LOAD_DWORD $vgpr4_vgpr5, 0, 0, implicit $exec, implicit $flat_scr316 S_ENDPGM 0317...318---319# Regular SALU instruction breaks clause, no nop needed320 321name: salu_inst_breaks_clause322 323body: |324 bb.0:325 ; GCN-LABEL: name: salu_inst_breaks_clause326 ; GCN: $vgpr0 = FLAT_LOAD_DWORD $vgpr2_vgpr3, 0, 0, implicit $exec, implicit $flat_scr327 ; GCN-NEXT: $sgpr8 = S_MOV_B32 0328 ; GCN-NEXT: $vgpr2 = FLAT_LOAD_DWORD $vgpr4_vgpr5, 0, 0, implicit $exec, implicit $flat_scr329 ; GCN-NEXT: S_ENDPGM 0330 331 $vgpr0 = FLAT_LOAD_DWORD $vgpr2_vgpr3, 0, 0, implicit $exec, implicit $flat_scr332 $sgpr8 = S_MOV_B32 0333 $vgpr2 = FLAT_LOAD_DWORD $vgpr4_vgpr5, 0, 0, implicit $exec, implicit $flat_scr334 S_ENDPGM 0335...336---337 338name: ds_inst_breaks_clause339 340body: |341 bb.0:342 ; GCN-LABEL: name: ds_inst_breaks_clause343 ; GCN: $vgpr0 = FLAT_LOAD_DWORD $vgpr2_vgpr3, 0, 0, implicit $exec, implicit $flat_scr344 ; GCN-NEXT: $vgpr8 = DS_READ_B32 $vgpr9, 0, 0, implicit $m0, implicit $exec345 ; GCN-NEXT: $vgpr2 = FLAT_LOAD_DWORD $vgpr4_vgpr5, 0, 0, implicit $exec, implicit $flat_scr346 ; GCN-NEXT: S_ENDPGM 0347 348 $vgpr0 = FLAT_LOAD_DWORD $vgpr2_vgpr3, 0, 0, implicit $exec, implicit $flat_scr349 $vgpr8 = DS_READ_B32 $vgpr9, 0, 0, implicit $m0, implicit $exec350 $vgpr2 = FLAT_LOAD_DWORD $vgpr4_vgpr5, 0, 0, implicit $exec, implicit $flat_scr351 S_ENDPGM 0352...353---354 355name: smrd_inst_breaks_clause356 357body: |358 bb.0:359 ; GCN-LABEL: name: smrd_inst_breaks_clause360 ; GCN: $vgpr0 = FLAT_LOAD_DWORD $vgpr2_vgpr3, 0, 0, implicit $exec, implicit $flat_scr361 ; GCN-NEXT: $sgpr8 = S_LOAD_DWORD_IMM $sgpr0_sgpr1, 0, 0362 ; GCN-NEXT: $vgpr2 = FLAT_LOAD_DWORD $vgpr4_vgpr5, 0, 0, implicit $exec, implicit $flat_scr363 ; GCN-NEXT: S_ENDPGM 0364 365 $vgpr0 = FLAT_LOAD_DWORD $vgpr2_vgpr3, 0, 0, implicit $exec, implicit $flat_scr366 $sgpr8 = S_LOAD_DWORD_IMM $sgpr0_sgpr1, 0, 0367 $vgpr2 = FLAT_LOAD_DWORD $vgpr4_vgpr5, 0, 0, implicit $exec, implicit $flat_scr368 S_ENDPGM 0369...370---371# FIXME: Should this be handled?372name: implicit_use_breaks_clause373 374body: |375 bb.0:376 ; GCN-LABEL: name: implicit_use_breaks_clause377 ; GCN: $vgpr0_vgpr1 = FLAT_LOAD_DWORDX2 $vgpr2_vgpr3, 0, 0, implicit $exec, implicit $flat_scr, implicit $vgpr4_vgpr5378 ; XNACK-NEXT: S_NOP 0379 ; GCN-NEXT: $vgpr4_vgpr5 = FLAT_LOAD_DWORDX2 $vgpr6_vgpr7, 0, 0, implicit $exec, implicit $flat_scr380 ; GCN-NEXT: S_ENDPGM 0381 382 $vgpr0_vgpr1 = FLAT_LOAD_DWORDX2 $vgpr2_vgpr3, 0, 0, implicit $exec, implicit $flat_scr, implicit $vgpr4_vgpr5383 $vgpr4_vgpr5 = FLAT_LOAD_DWORDX2 $vgpr6_vgpr7, 0, 0, implicit $exec, implicit $flat_scr384 S_ENDPGM 0385...386---387name: trivial_clause_load_mubuf4_x2388 389body: |390 bb.0:391 ; GCN-LABEL: name: trivial_clause_load_mubuf4_x2392 ; GCN: $vgpr1 = BUFFER_LOAD_DWORD_OFFEN $vgpr2, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr4, 0, 0, 0, implicit $exec393 ; GCN-NEXT: $vgpr3 = BUFFER_LOAD_DWORD_OFFEN $vgpr4, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr4, 0, 0, 0, implicit $exec394 ; GCN-NEXT: S_ENDPGM 0395 396 $vgpr1 = BUFFER_LOAD_DWORD_OFFEN $vgpr2, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr4, 0, 0, 0, implicit $exec397 $vgpr3 = BUFFER_LOAD_DWORD_OFFEN $vgpr4, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr4, 0, 0, 0, implicit $exec398 S_ENDPGM 0399...400---401name: break_clause_simple_load_mubuf_offen_ptr402 403body: |404 bb.0:405 ; GCN-LABEL: name: break_clause_simple_load_mubuf_offen_ptr406 ; GCN: $vgpr1 = BUFFER_LOAD_DWORD_OFFEN $vgpr2, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr4, 0, 0, 0, implicit $exec407 ; XNACK-NEXT: S_NOP 0408 ; GCN-NEXT: $vgpr2 = BUFFER_LOAD_DWORD_OFFEN $vgpr3, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr4, 0, 0, 0, implicit $exec409 ; GCN-NEXT: S_ENDPGM 0410 411 $vgpr1 = BUFFER_LOAD_DWORD_OFFEN $vgpr2, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr4, 0, 0, 0, implicit $exec412 $vgpr2 = BUFFER_LOAD_DWORD_OFFEN $vgpr3, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr4, 0, 0, 0, implicit $exec413 S_ENDPGM 0414...415---416# BUFFER instructions overwriting their own inputs is supposedly OK.417 418name: mubuf_load4_overwrite_ptr419 420body: |421 bb.0:422 ; GCN-LABEL: name: mubuf_load4_overwrite_ptr423 ; GCN: $vgpr0 = BUFFER_LOAD_DWORD_OFFEN $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr4, 0, 0, 0, implicit $exec424 ; GCN-NEXT: $vgpr1 = V_MOV_B32_e32 0, implicit $exec425 ; GCN-NEXT: $vgpr2 = V_MOV_B32_e32 $vgpr0, implicit $exec426 ; GCN-NEXT: S_ENDPGM 0427 $vgpr0 = BUFFER_LOAD_DWORD_OFFEN $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr4, 0, 0, 0, implicit $exec428 $vgpr1 = V_MOV_B32_e32 0, implicit $exec429 $vgpr2 = V_MOV_B32_e32 $vgpr0, implicit $exec430 S_ENDPGM 0431...432---433# Break a clause from interference between mubuf and flat instructions434 435name: break_clause_flat_load_mubuf_load436 437body: |438 bb.0:439 ; GCN-LABEL: name: break_clause_flat_load_mubuf_load440 ; GCN: $vgpr0 = FLAT_LOAD_DWORD $vgpr2_vgpr3, 0, 0, implicit $exec, implicit $flat_scr441 ; XNACK-NEXT: S_NOP 0442 ; GCN-NEXT: $vgpr2 = BUFFER_LOAD_DWORD_OFFEN $vgpr1, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr4, 0, 0, 0, implicit $exec443 ; GCN-NEXT: S_ENDPGM 0444 445 $vgpr0 = FLAT_LOAD_DWORD $vgpr2_vgpr3, 0, 0, implicit $exec, implicit $flat_scr446 $vgpr2 = BUFFER_LOAD_DWORD_OFFEN $vgpr1, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr4, 0, 0, 0, implicit $exec447 S_ENDPGM 0448...449# Break a clause from interference between mubuf and flat instructions450 451# GCN-LABEL: name: break_clause_mubuf_load_flat_load452# GCN: bb.0:453# GCN-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFEN $vgpr1, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr4454# XNACK-NEXT: S_NOP 0455# GCN-NEXT: $vgpr1 = FLAT_LOAD_DWORD $vgpr2_vgpr3456# GCN-NEXT: S_ENDPGM 0457name: break_clause_mubuf_load_flat_load458 459body: |460 bb.0:461 $vgpr0 = BUFFER_LOAD_DWORD_OFFEN $vgpr1, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr4, 0, 0, 0, implicit $exec462 $vgpr1 = FLAT_LOAD_DWORD $vgpr2_vgpr3, 0, 0, implicit $exec, implicit $flat_scr463 464 S_ENDPGM 0465...466---467 468name: break_clause_atomic_rtn_into_ptr_flat4469 470body: |471 bb.0:472 ; GCN-LABEL: name: break_clause_atomic_rtn_into_ptr_flat4473 ; GCN: $vgpr2 = FLAT_LOAD_DWORD $vgpr4_vgpr5, 0, 0, implicit $exec, implicit $flat_scr474 ; XNACK-NEXT: S_NOP 0475 ; GCN-NEXT: $vgpr4 = FLAT_ATOMIC_ADD_RTN $vgpr5_vgpr6, $vgpr7, 0, 1, implicit $exec, implicit $flat_scr476 ; GCN-NEXT: S_ENDPGM 0477 478 $vgpr2 = FLAT_LOAD_DWORD $vgpr4_vgpr5, 0, 0, implicit $exec, implicit $flat_scr479 $vgpr4 = FLAT_ATOMIC_ADD_RTN $vgpr5_vgpr6, $vgpr7, 0, 1, implicit $exec, implicit $flat_scr480 S_ENDPGM 0481...482---483name: break_clause_atomic_nortn_ptr_load_flat4484 485body: |486 bb.0:487 ; GCN-LABEL: name: break_clause_atomic_nortn_ptr_load_flat4488 ; GCN: FLAT_ATOMIC_ADD $vgpr0_vgpr1, $vgpr2, 0, 0, implicit $exec, implicit $flat_scr489 ; GCN-NEXT: $vgpr2 = FLAT_LOAD_DWORD $vgpr3_vgpr4, 0, 0, implicit $exec, implicit $flat_scr490 ; GCN-NEXT: S_ENDPGM 0491 492 FLAT_ATOMIC_ADD $vgpr0_vgpr1, $vgpr2, 0, 0, implicit $exec, implicit $flat_scr493 $vgpr2 = FLAT_LOAD_DWORD $vgpr3_vgpr4, 0, 0, implicit $exec, implicit $flat_scr494 S_ENDPGM 0495...496---497 498name: break_clause_atomic_rtn_into_ptr_mubuf4499 500body: |501 bb.0:502 ; GCN-LABEL: name: break_clause_atomic_rtn_into_ptr_mubuf4503 ; GCN: $vgpr1 = BUFFER_LOAD_DWORD_OFFEN $vgpr2, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr4, 0, 0, 0, implicit $exec504 ; XNACK-NEXT: S_NOP 0505 ; GCN-NEXT: $vgpr2 = BUFFER_ATOMIC_ADD_OFFEN_RTN $vgpr2, $vgpr5, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr4, 0, 1, implicit $exec506 ; GCN-NEXT: S_ENDPGM 0507 508 $vgpr1 = BUFFER_LOAD_DWORD_OFFEN $vgpr2, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr4, 0, 0, 0, implicit $exec509 $vgpr2 = BUFFER_ATOMIC_ADD_OFFEN_RTN $vgpr2, $vgpr5, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr4, 0, 1, implicit $exec510 S_ENDPGM 0511...512---513 514name: break_clause_atomic_nortn_ptr_load_mubuf4515 516body: |517 bb.0:518 ; GCN-LABEL: name: break_clause_atomic_nortn_ptr_load_mubuf4519 ; GCN: BUFFER_ATOMIC_ADD_OFFEN $vgpr0, $vgpr1, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr4, 0, 0, implicit $exec520 ; GCN-NEXT: $vgpr1 = BUFFER_LOAD_DWORD_OFFEN $vgpr2, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr4, 0, 0, 0, implicit $exec521 ; GCN-NEXT: S_ENDPGM 0522 523 BUFFER_ATOMIC_ADD_OFFEN $vgpr0, $vgpr1, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr4, 0, 0, implicit $exec524 $vgpr1 = BUFFER_LOAD_DWORD_OFFEN $vgpr2, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr4, 0, 0, 0, implicit $exec525 S_ENDPGM 0526...527---528# Make sure there is no assert on mubuf instructions which do not have529# vaddr, and don't add register to track.530name: no_break_clause_mubuf_load_novaddr531 532body: |533 bb.0:534 ; GCN-LABEL: name: no_break_clause_mubuf_load_novaddr535 ; GCN: $vgpr1 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr4, 0, 0, 0, implicit $exec536 ; GCN-NEXT: $vgpr3 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr4, 0, 0, 0, implicit $exec537 ; GCN-NEXT: S_ENDPGM 0538 $vgpr1 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr4, 0, 0, 0, implicit $exec539 $vgpr3 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr4, 0, 0, 0, implicit $exec540 S_ENDPGM 0541...542---543# Loads and stores using different addresses theoretically does not544# need a nop545name: mix_load_store_clause546body: |547 bb.0:548 ; GCN-LABEL: name: mix_load_store_clause549 ; GCN: FLAT_STORE_DWORD $vgpr0_vgpr1, $vgpr5, 0, 0, implicit $exec, implicit $flat_scr550 ; GCN-NEXT: $vgpr10 = FLAT_LOAD_DWORD $vgpr2_vgpr3, 0, 0, implicit $exec, implicit $flat_scr551 ; XNACK-NEXT: S_NOP 0552 ; GCN-NEXT: FLAT_STORE_DWORD $vgpr2_vgpr3, $vgpr6, 0, 0, implicit $exec, implicit $flat_scr553 ; GCN-NEXT: $vgpr11 = FLAT_LOAD_DWORD $vgpr4_vgpr5, 0, 0, implicit $exec, implicit $flat_scr554 555 FLAT_STORE_DWORD $vgpr0_vgpr1, $vgpr5, 0, 0, implicit $exec, implicit $flat_scr556 $vgpr10 = FLAT_LOAD_DWORD $vgpr2_vgpr3, 0, 0, implicit $exec, implicit $flat_scr557 FLAT_STORE_DWORD $vgpr2_vgpr3, $vgpr6, 0, 0, implicit $exec, implicit $flat_scr558 $vgpr11 = FLAT_LOAD_DWORD $vgpr4_vgpr5, 0, 0, implicit $exec, implicit $flat_scr559 S_ENDPGM 0560...561---562# Loads and stores using the same address needs a nop.563 564name: mix_load_store_clause_same_address565body: |566 bb.0:567 ; GCN-LABEL: name: mix_load_store_clause_same_address568 ; GCN: FLAT_STORE_DWORD $vgpr0_vgpr1, $vgpr5, 0, 0, implicit $exec, implicit $flat_scr569 ; GCN-NEXT: $vgpr10 = FLAT_LOAD_DWORD $vgpr2_vgpr3, 0, 0, implicit $exec, implicit $flat_scr570 ; XNACK-NEXT: S_NOP 0571 ; GCN-NEXT: FLAT_STORE_DWORD $vgpr0_vgpr1, $vgpr6, 0, 0, implicit $exec, implicit $flat_scr572 ; GCN-NEXT: $vgpr11 = FLAT_LOAD_DWORD $vgpr4_vgpr5, 0, 0, implicit $exec, implicit $flat_scr573 574 FLAT_STORE_DWORD $vgpr0_vgpr1, $vgpr5, 0, 0, implicit $exec, implicit $flat_scr575 $vgpr10 = FLAT_LOAD_DWORD $vgpr2_vgpr3, 0, 0, implicit $exec, implicit $flat_scr576 FLAT_STORE_DWORD $vgpr0_vgpr1, $vgpr6, 0, 0, implicit $exec, implicit $flat_scr577 $vgpr11 = FLAT_LOAD_DWORD $vgpr4_vgpr5, 0, 0, implicit $exec, implicit $flat_scr578 S_ENDPGM 0579...580