Lines Matching refs:MUBUF
1 …pu=hawaii -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefixes=GCN,CI,MUBUF %s
2 …=gfx900 -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefixes=GCN,GFX9,MUBUF %s
16 ; MUBUF-NEXT: s_mov_b32 [[FP_COPY:s4]], s33
37 ; MUBUF-NEXT: buffer_store_dword v0, off, s[0:3], s32{{$}}
52 ; MUBUF-NEXT: s_mov_b32 [[FP_COPY:s4]], s33
55 ; MUBUF-NEXT: s_add_u32 s32, s32, 0x200
58 ; MUBUF-NEXT: buffer_store_dword v0, off, s[0:3], s33 offset:4{{$}}
60 ; MUBUF-NEXT: s_sub_u32 s32, s32, 0x200
75 ; MUBUF-NEXT: buffer_store_dword v0, off, s[0:3], s32{{$}}
89 ; MUBUF-NEXT: buffer_store_dword [[CSR_VGPR:v[0-9]+]], off, s[0:3], s32 offset:4 ; 4-byte Folded …
94 ; MUBUF-DAG: s_add_u32 s32, s32, 0x400{{$}}
100 ; MUBUF-DAG: buffer_store_dword [[ZERO]], off, s[0:3], s33{{$}}
105 ; MUBUF-DAG: v_readlane_b32 s5, [[CSR_VGPR]]
106 ; MUBUF-DAG: v_readlane_b32 s4, [[CSR_VGPR]]
110 ; MUBUF: s_sub_u32 s32, s32, 0x400{{$}}
114 ; MUBUF-NEXT: buffer_load_dword [[CSR_VGPR]], off, s[0:3], s32 offset:4 ; 4-byte Folded Reload
136 ; MUBUF-NEXT: buffer_store_dword [[CSR_VGPR:v[0-9]+]], off, s[0:3], s32 ; 4-byte Folded Spill
139 ; MUBUF-DAG: s_add_u32 s32, s32, 0x400
147 ; MUBUF-DAG: v_readlane_b32 s4, v40, 0
148 ; MUBUF-DAG: v_readlane_b32 s5, v40, 1
152 ; MUBUF: s_sub_u32 s32, s32, 0x400
156 ; MUBUF-NEXT: buffer_load_dword [[CSR_VGPR]], off, s[0:3], s32 ; 4-byte Folded Reload
173 ; MUBUF-NEXT: buffer_store_dword [[CSR_VGPR:v[0-9]+]], off, s[0:3], s32 ; 4-byte Folded Spill
184 ; MUBUF-NEXT: buffer_load_dword [[CSR_VGPR]], off, s[0:3], s32 ; 4-byte Folded Reload
234 ; MUBUF-DAG: buffer_store_dword v41, off, s[0:3], s33 ; 4-byte Folded Spill
236 ; MUBUF-DAG: buffer_store_dword [[ZERO]], off, s[0:3], s33 offset:8
243 ; MUBUF: buffer_load_dword v41, off, s[0:3], s33 ; 4-byte Folded Reload
245 ; MUBUF: s_add_u32 s32, s32, 0x300
246 ; MUBUF-NEXT: s_sub_u32 s32, s32, 0x300
247 ; MUBUF-NEXT: s_mov_b32 s33, s4
267 ; MUBUF: buffer_store_dword v41, off, s[0:3], s33 ; 4-byte Folded Spill
269 ; MUBUF: buffer_store_dword v{{[0-9]+}}, off, s[0:3], s33 offset:8
274 ; MUBUF: s_add_u32 s32, s32, 0x300
275 ; MUBUF: s_sub_u32 s32, s32, 0x300
304 ; MUBUF: buffer_store_dword v41, off, s[0:3], s33 ; 4-byte Folded Spill
306 ; MUBUF: buffer_store_dword
310 ; MUBUF: buffer_load_dword v41, off, s[0:3], s33 ; 4-byte Folded Reload
312 ; MUBUF: s_add_u32 s32, s32, 0x300
315 ; MUBUF-NEXT: s_sub_u32 s32, s32, 0x300
338 ; MUBUF-NEXT: s_add_u32 [[SCRATCH:s[0-9]+]], s32, 0x7ffc0
340 ; MUBUF-NEXT: s_mov_b32 [[FP_COPY:s4]], s33
342 ; MUBUF-NEXT: s_and_b32 s33, [[SCRATCH]], 0xfff80000
344 ; MUBUF-NEXT: s_add_u32 s32, s32, 0x100000
347 ; MUBUF-NEXT: buffer_store_dword [[ZERO]], off, s[0:3], s33
349 ; MUBUF-NEXT: s_sub_u32 s32, s32, 0x100000
367 ; MUBUF: buffer_store_dword [[ZERO]], off, s[0:3], s33 offset:4
370 ; MUBUF: v_readlane_b32 s4, v1, 0
371 ; MUBUF-NEXT: s_add_u32 s32, s32, 0x200
372 ; MUBUF-NEXT: v_readlane_b32 s5, v1, 1
376 ; MUBUF-NEXT: s_sub_u32 s32, s32, 0x200
380 ; MUBUF-NEXT: s_setpc_b64 s[4:5]
400 ; MUBUF-NEXT: buffer_store_dword [[CSR_VGPR:v[0-9]+]], off, s[0:3], s32 offset:8 ; 4-byte Folded …
408 ; MUBUF-DAG: buffer_store_dword
410 ; MUBUF: s_add_u32 s32, s32, 0x300{{$}}
413 ; MUBUF: v_readlane_b32 s4, [[CSR_VGPR]], 0
416 ; MUBUF: v_readlane_b32 s5, [[CSR_VGPR]], 1
418 ; MUBUF-NEXT: s_sub_u32 s32, s32, 0x300{{$}}
422 ; MUBUF-NEXT: buffer_load_dword [[CSR_VGPR]], off, s[0:3], s32 offset:8 ; 4-byte Folded Reload
447 ; The byval argument exceeds the MUBUF constant offset, so a scratch
452 ; MUBUF-NEXT: v_mov_b32_e32 [[SCRATCH_VGPR:v[0-9]+]], 0x1008
453 ; MUBUF-NEXT: buffer_store_dword [[CSR_VGPR:v[0-9]+]], [[SCRATCH_VGPR]], s[0:3], s32 offen ; 4-byte…
461 ; MUBUF-DAG: s_add_u32 s32, s32, 0x40300{{$}}
463 ; MUBUF-DAG: buffer_store_dword
466 ; MUBUF: v_readlane_b32 s4, [[CSR_VGPR]], 0
469 ; MUBUF: v_readlane_b32 s5, [[CSR_VGPR]], 1
471 ; MUBUF-NEXT: s_sub_u32 s32, s32, 0x40300{{$}}
475 ; MUBUF-NEXT: v_mov_b32_e32 [[SCRATCH_VGPR:v[0-9]+]], 0x1008
476 ; MUBUF-NEXT: buffer_load_dword [[CSR_VGPR]], [[SCRATCH_VGPR]], s[0:3], s32 offen ; 4-byte Folded R…
515 ; MUBUF: s_add_u32 s32, s32, 0x400
517 ; MUBUF: buffer_store_dword v{{[0-9]+}}, off, s[0:3], s33{{$}}
520 ; MUBUF: s_sub_u32 s32, s32, 0x400
532 ; MUBUF: s_or_saveexec_b64 [[COPY_EXEC1:s\[[0-9]+:[0-9]+\]]], -1{{$}}
533 ; MUBUF: v_mov_b32_e32 [[TMP_VGPR1:v[0-9]+]], s33
534 ; MUBUF: buffer_store_dword [[TMP_VGPR1]], off, s[0:3], s32 offset:4
535 ; MUBUF: s_mov_b64 exec, [[COPY_EXEC1]]
538 ; MUBUF: s_or_saveexec_b64 [[COPY_EXEC2:s\[[0-9]+:[0-9]+\]]], -1{{$}}
539 ; MUBUF: buffer_load_dword [[TMP_VGPR2:v[0-9]+]], off, s[0:3], s32 offset:4
541 ; MUBUF: s_waitcnt vmcnt(0)
542 ; MUBUF: v_readfirstlane_b32 s33, [[TMP_VGPR2]]
543 ; MUBUF: s_mov_b64 exec, [[COPY_EXEC2]]
545 ; MUBUF: ScratchSize: 8
566 ; MUBUF: s_or_saveexec_b64 [[COPY_EXEC1:s\[[0-9]+:[0-9]+\]]], -1{{$}}
567 ; MUBUF: v_mov_b32_e32 [[TMP_VGPR1:v[0-9]+]], s33
568 ; MUBUF: buffer_store_dword [[TMP_VGPR1]], off, s[0:3], s32 offset:[[OFF:[0-9]+]]
569 ; MUBUF: s_mov_b64 exec, [[COPY_EXEC1]]
571 ; MUBUF: s_mov_b32 s33, s32
574 ; MUBUF: s_or_saveexec_b64 [[COPY_EXEC2:s\[[0-9]+:[0-9]+\]]], -1{{$}}
575 ; MUBUF: buffer_load_dword [[TMP_VGPR2:v[0-9]+]], off, s[0:3], s32 offset:[[OFF]]
576 ; MUBUF: v_readfirstlane_b32 s33, [[TMP_VGPR2]]
577 ; MUBUF: s_mov_b64 exec, [[COPY_EXEC2]]
600 ; If the size of the offset exceeds the MUBUF offset field we need another
603 ; MUBUF: s_or_saveexec_b64 s[4:5], -1
604 ; MUBUF: v_mov_b32_e32 v0, s33
606 ; MUBUF-NEXT: v_mov_b32_e32 v1, 0x1008
607 ; MUBUF-NEXT: buffer_store_dword v0, v1, s[0:3], s32 offen ; 4-byte Folded Spill