This is an archive of the discontinued LLVM Phabricator instance.

Differential D97315

AMDGPU: Use aligned vgprs/agprs in gfx90a mir tests
ClosedPublic

Authored by arsenm on Feb 23 2021, 11:38 AM.

Details

Reviewers
rampitec
kzhuravl
Summary

These would fail a verifier check in a future change.

Diff Detail

Event Timeline

arsenm created this revision.Feb 23 2021, 11:38 AM
Herald added subscribers: kerbowa, t-tye, tpr and 4 others. · View Herald TranscriptFeb 23 2021, 11:38 AM
arsenm requested review of this revision.Feb 23 2021, 11:38 AM
Herald added a project: Restricted Project. · View Herald TranscriptFeb 23 2021, 11:38 AM
Herald added a subscriber: wdng. · View Herald Transcript
rampitec accepted this revision.Feb 23 2021, 12:25 PM

Thanks Matt!

This revision is now accepted and ready to land.Feb 23 2021, 12:25 PM
Harbormaster completed remote builds in B90447: Diff 325854.Feb 23 2021, 12:54 PM
arsenm closed this revision.Feb 24 2021, 12:08 PM

e844f24a278bf4b3622fb969d06e7b771a6a2aca

Revision Contents

PathSize
llvm/
test/
CodeGen/
AMDGPU/
40 lines
108 lines

Diff 325854

llvm/test/CodeGen/AMDGPU/mai-hazards-gfx90a.mir

Show First 20 Lines • Show All 458 Lines • ▼ Show 20 Lines
# GCN-LABEL: name: dmfma4x4_write_vgpr_flat_read_overlap # GCN-LABEL: name: dmfma4x4_write_vgpr_flat_read_overlap
# GCN: V_MFMA # GCN: V_MFMA
# GCN-NEXT: S_NOP 7 # GCN-NEXT: S_NOP 7
# GCN-NEXT: S_NOP 0 # GCN-NEXT: S_NOP 0
# GCN-NEXT: FLAT_STORE_DWORD # GCN-NEXT: FLAT_STORE_DWORD
name: dmfma4x4_write_vgpr_flat_read_overlap name: dmfma4x4_write_vgpr_flat_read_overlap
body: | body: |
bb.0: bb.0:
$vgpr3_vgpr4 = V_MFMA_F64_4X4X4F64_vgprcd_e64 $vgpr0_vgpr1, $vgpr0_vgpr1, $vgpr0_vgpr1, 0, 0, 0, implicit $mode, implicit $exec $vgpr4_vgpr5 = V_MFMA_F64_4X4X4F64_vgprcd_e64 $vgpr0_vgpr1, $vgpr0_vgpr1, $vgpr0_vgpr1, 0, 0, 0, implicit $mode, implicit $exec
FLAT_STORE_DWORD $vgpr0_vgpr1, $vgpr4, 0, 0, 0, 0, 0, implicit $mode, implicit $exec, implicit $flat_scr FLAT_STORE_DWORD $vgpr0_vgpr1, $vgpr5, 0, 0, 0, 0, 0, implicit $mode, implicit $exec, implicit $flat_scr
... ...
# GCN-LABEL: name: dmfma4x4_write_vgpr_flat_read_full # GCN-LABEL: name: dmfma4x4_write_vgpr_flat_read_full
# GCN: V_MFMA # GCN: V_MFMA
# GCN-NEXT: S_NOP 7 # GCN-NEXT: S_NOP 7
# GCN-NEXT: S_NOP 0 # GCN-NEXT: S_NOP 0
# GCN-NEXT: FLAT_STORE_DWORD # GCN-NEXT: FLAT_STORE_DWORD
name: dmfma4x4_write_vgpr_flat_read_full name: dmfma4x4_write_vgpr_flat_read_full
body: | body: |
bb.0: bb.0:
$vgpr3_vgpr4 = V_MFMA_F64_4X4X4F64_vgprcd_e64 $vgpr0_vgpr1, $vgpr0_vgpr1, $vgpr0_vgpr1, 0, 0, 0, implicit $mode, implicit $exec $vgpr4_vgpr5 = V_MFMA_F64_4X4X4F64_vgprcd_e64 $vgpr0_vgpr1, $vgpr0_vgpr1, $vgpr0_vgpr1, 0, 0, 0, implicit $mode, implicit $exec
FLAT_STORE_DWORDX2 $vgpr0_vgpr1, $vgpr3_vgpr4, 0, 0, 0, 0, 0, implicit $mode, implicit $exec, implicit $flat_scr FLAT_STORE_DWORDX2 $vgpr0_vgpr1, $vgpr4_vgpr5, 0, 0, 0, 0, 0, implicit $mode, implicit $exec, implicit $flat_scr
... ...
# GCN-LABEL: name: dmfma16x16_write_vgpr_flat_read # GCN-LABEL: name: dmfma16x16_write_vgpr_flat_read
# GCN: V_MFMA # GCN: V_MFMA
# GCN-NEXT: S_NOP 7 # GCN-NEXT: S_NOP 7
# GCN-NEXT: S_NOP 7 # GCN-NEXT: S_NOP 7
# GCN-NEXT: S_NOP 1 # GCN-NEXT: S_NOP 1
# GCN-NEXT: FLAT_STORE_DWORD # GCN-NEXT: FLAT_STORE_DWORD
name: dmfma16x16_write_vgpr_flat_read name: dmfma16x16_write_vgpr_flat_read
Show All 37 Lines
... ...
# GCN-LABEL: name: dmfma4x4_write_vgpr_valu_read # GCN-LABEL: name: dmfma4x4_write_vgpr_valu_read
# GCN: V_MFMA # GCN: V_MFMA
# GCN-NEXT: S_NOP 5 # GCN-NEXT: S_NOP 5
# GCN-NEXT: V_MOV_B32 # GCN-NEXT: V_MOV_B32
name: dmfma4x4_write_vgpr_valu_read name: dmfma4x4_write_vgpr_valu_read
body: | body: |
bb.0: bb.0:
$vgpr3_vgpr4 = V_MFMA_F64_4X4X4F64_vgprcd_e64 $vgpr0_vgpr1, $vgpr0_vgpr1, $vgpr0_vgpr1, 0, 0, 0, implicit $mode, implicit $exec $vgpr4_vgpr5 = V_MFMA_F64_4X4X4F64_vgprcd_e64 $vgpr0_vgpr1, $vgpr0_vgpr1, $vgpr0_vgpr1, 0, 0, 0, implicit $mode, implicit $exec
$vgpr5 = V_MOV_B32_e32 $vgpr4, implicit $exec $vgpr6 = V_MOV_B32_e32 $vgpr5, implicit $exec
... ...
# GCN-LABEL: name: dmfma16x16_write_vgpr_valu_read # GCN-LABEL: name: dmfma16x16_write_vgpr_valu_read
# GCN: V_MFMA # GCN: V_MFMA
# GCN-NEXT: S_NOP 7 # GCN-NEXT: S_NOP 7
# GCN-NEXT: S_NOP 2 # GCN-NEXT: S_NOP 2
# GCN-NEXT: V_MOV_B32 # GCN-NEXT: V_MOV_B32
name: dmfma16x16_write_vgpr_valu_read name: dmfma16x16_write_vgpr_valu_read
body: | body: |
▲ Show 20 Lines • Show All 46 Lines • ▼ Show 20 Lines
... ...
# GCN-LABEL: name: dmfma4x4_write_vgpr_dot_read # GCN-LABEL: name: dmfma4x4_write_vgpr_dot_read
# GCN: V_MFMA # GCN: V_MFMA
# GCN-NEXT: S_NOP 5 # GCN-NEXT: S_NOP 5
# GCN-NEXT: V_DOT # GCN-NEXT: V_DOT
name: dmfma4x4_write_vgpr_dot_read name: dmfma4x4_write_vgpr_dot_read
body: | body: |
bb.0: bb.0:
$vgpr3_vgpr4 = V_MFMA_F64_4X4X4F64_vgprcd_e64 $vgpr0_vgpr1, $vgpr0_vgpr1, $vgpr0_vgpr1, 0, 0, 0, implicit $mode, implicit $exec $vgpr4_vgpr5 = V_MFMA_F64_4X4X4F64_vgprcd_e64 $vgpr0_vgpr1, $vgpr0_vgpr1, $vgpr0_vgpr1, 0, 0, 0, implicit $mode, implicit $exec
$vgpr1 = V_DOT4C_I32_I8_e32 $vgpr0, $vgpr4, $vgpr1, implicit $exec $vgpr1 = V_DOT4C_I32_I8_e32 $vgpr0, $vgpr5, $vgpr1, implicit $exec
... ...
# GCN-LABEL: name: dmfma16x16_write_vgpr_dot_read # GCN-LABEL: name: dmfma16x16_write_vgpr_dot_read
# GCN: V_MFMA # GCN: V_MFMA
# GCN-NEXT: S_NOP 7 # GCN-NEXT: S_NOP 7
# GCN-NEXT: S_NOP 2 # GCN-NEXT: S_NOP 2
# GCN-NEXT: V_DOT # GCN-NEXT: V_DOT
name: dmfma16x16_write_vgpr_dot_read name: dmfma16x16_write_vgpr_dot_read
body: | body: |
▲ Show 20 Lines • Show All 102 Lines • ▼ Show 20 Lines
... ...
# GCN-LABEL: name: dmfma4x4_write_vgpr_valu_write # GCN-LABEL: name: dmfma4x4_write_vgpr_valu_write
# GCN: V_MFMA # GCN: V_MFMA
# GCN-NEXT: S_NOP 5 # GCN-NEXT: S_NOP 5
# GCN-NEXT: V_MOV_B32 # GCN-NEXT: V_MOV_B32
name: dmfma4x4_write_vgpr_valu_write name: dmfma4x4_write_vgpr_valu_write
body: | body: |
bb.0: bb.0:
$vgpr3_vgpr4 = V_MFMA_F64_4X4X4F64_vgprcd_e64 $vgpr0_vgpr1, $vgpr0_vgpr1, $vgpr0_vgpr1, 0, 0, 0, implicit $mode, implicit $exec $vgpr4_vgpr5 = V_MFMA_F64_4X4X4F64_vgprcd_e64 $vgpr0_vgpr1, $vgpr0_vgpr1, $vgpr0_vgpr1, 0, 0, 0, implicit $mode, implicit $exec
$vgpr3 = V_MOV_B32_e32 0, implicit $mode, implicit $exec $vgpr4 = V_MOV_B32_e32 0, implicit $mode, implicit $exec
... ...
# GCN-LABEL: name: dmfma16x16_write_vgpr_valu_write # GCN-LABEL: name: dmfma16x16_write_vgpr_valu_write
# GCN: V_MFMA # GCN: V_MFMA
# GCN-NEXT: S_NOP 7 # GCN-NEXT: S_NOP 7
# GCN-NEXT: S_NOP 2 # GCN-NEXT: S_NOP 2
# GCN-NEXT: V_MOV_B32 # GCN-NEXT: V_MOV_B32
name: dmfma16x16_write_vgpr_valu_write name: dmfma16x16_write_vgpr_valu_write
body: | body: |
▲ Show 20 Lines • Show All 107 Lines • ▼ Show 20 Lines bb.0:
$vgpr19 = V_MOV_B32_e32 0, implicit $mode, implicit $exec $vgpr19 = V_MOV_B32_e32 0, implicit $mode, implicit $exec
... ...
# GCN-LABEL: name: dmfma4x4_read_srcc_vgpr_valu_write # GCN-LABEL: name: dmfma4x4_read_srcc_vgpr_valu_write
# GCN: V_MFMA # GCN: V_MFMA
# GCN-NEXT: V_MOV_B32 # GCN-NEXT: V_MOV_B32
name: dmfma4x4_read_srcc_vgpr_valu_write name: dmfma4x4_read_srcc_vgpr_valu_write
body: | body: |
bb.0: bb.0:
$vgpr3_vgpr4 = V_MFMA_F64_4X4X4F64_vgprcd_e64 $vgpr0_vgpr1, $vgpr0_vgpr1, $vgpr0_vgpr1, 0, 0, 0, implicit $mode, implicit $exec $vgpr4_vgpr5 = V_MFMA_F64_4X4X4F64_vgprcd_e64 $vgpr0_vgpr1, $vgpr0_vgpr1, $vgpr0_vgpr1, 0, 0, 0, implicit $mode, implicit $exec
$vgpr1 = V_MOV_B32_e32 0, implicit $mode, implicit $exec $vgpr1 = V_MOV_B32_e32 0, implicit $mode, implicit $exec
... ...
# GCN-LABEL: name: dmfma16x16_read_srcc_vgpr_valu_write # GCN-LABEL: name: dmfma16x16_read_srcc_vgpr_valu_write
# GCN: V_MFMA # GCN: V_MFMA
# GCN-NEXT: V_MOV_B32 # GCN-NEXT: V_MOV_B32
name: dmfma16x16_read_srcc_vgpr_valu_write name: dmfma16x16_read_srcc_vgpr_valu_write
body: | body: |
bb.0: bb.0:
▲ Show 20 Lines • Show All 346 Lines • ▼ Show 20 Lines
# GCN-LABEL: name: dmfma4x4_write_agpr_flat_read_overlap # GCN-LABEL: name: dmfma4x4_write_agpr_flat_read_overlap
# GCN: V_MFMA # GCN: V_MFMA
# GCN-NEXT: S_NOP 7 # GCN-NEXT: S_NOP 7
# GCN-NEXT: S_NOP 0 # GCN-NEXT: S_NOP 0
# GCN-NEXT: FLAT_STORE_DWORD # GCN-NEXT: FLAT_STORE_DWORD
name: dmfma4x4_write_agpr_flat_read_overlap name: dmfma4x4_write_agpr_flat_read_overlap
body: | body: |
bb.0: bb.0:
$agpr3_agpr4 = V_MFMA_F64_4X4X4F64_e64 $vgpr0_vgpr1, $vgpr0_vgpr1, $agpr0_agpr1, 0, 0, 0, implicit $mode, implicit $exec $agpr4_agpr5 = V_MFMA_F64_4X4X4F64_e64 $vgpr0_vgpr1, $vgpr0_vgpr1, $agpr0_agpr1, 0, 0, 0, implicit $mode, implicit $exec
FLAT_STORE_DWORD $vgpr0_vgpr1, $agpr4, 0, 0, 0, 0, 0, implicit $mode, implicit $exec, implicit $flat_scr FLAT_STORE_DWORD $vgpr0_vgpr1, $agpr5, 0, 0, 0, 0, 0, implicit $mode, implicit $exec, implicit $flat_scr
... ...
# GCN-LABEL: name: dmfma4x4_write_agpr_flat_read_full # GCN-LABEL: name: dmfma4x4_write_agpr_flat_read_full
# GCN: V_MFMA # GCN: V_MFMA
# GCN-NEXT: S_NOP 7 # GCN-NEXT: S_NOP 7
# GCN-NEXT: S_NOP 0 # GCN-NEXT: S_NOP 0
# GCN-NEXT: FLAT_STORE_DWORD # GCN-NEXT: FLAT_STORE_DWORD
name: dmfma4x4_write_agpr_flat_read_full name: dmfma4x4_write_agpr_flat_read_full
body: | body: |
bb.0: bb.0:
$agpr3_agpr4 = V_MFMA_F64_4X4X4F64_e64 $vgpr0_vgpr1, $vgpr0_vgpr1, $agpr0_agpr1, 0, 0, 0, implicit $mode, implicit $exec $agpr4_agpr5 = V_MFMA_F64_4X4X4F64_e64 $vgpr0_vgpr1, $vgpr0_vgpr1, $agpr0_agpr1, 0, 0, 0, implicit $mode, implicit $exec
FLAT_STORE_DWORDX2 $vgpr0_vgpr1, $agpr3_agpr4, 0, 0, 0, 0, 0, implicit $mode, implicit $exec, implicit $flat_scr FLAT_STORE_DWORDX2 $vgpr0_vgpr1, $agpr4_agpr5, 0, 0, 0, 0, 0, implicit $mode, implicit $exec, implicit $flat_scr
... ...
# GCN-LABEL: name: dmfma16x16_write_agpr_flat_read # GCN-LABEL: name: dmfma16x16_write_agpr_flat_read
# GCN: V_MFMA # GCN: V_MFMA
# GCN-NEXT: S_NOP 7 # GCN-NEXT: S_NOP 7
# GCN-NEXT: S_NOP 7 # GCN-NEXT: S_NOP 7
# GCN-NEXT: S_NOP 1 # GCN-NEXT: S_NOP 1
# GCN-NEXT: FLAT_STORE_DWORD # GCN-NEXT: FLAT_STORE_DWORD
name: dmfma16x16_write_agpr_flat_read name: dmfma16x16_write_agpr_flat_read
body: | body: |
bb.0: bb.0:
$agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9 = V_MFMA_F64_16X16X4F64_e64 $vgpr10_vgpr11, $vgpr10_vgpr11, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7, 0, 0, 0, implicit $mode, implicit $exec $agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9 = V_MFMA_F64_16X16X4F64_e64 $vgpr10_vgpr11, $vgpr10_vgpr11, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7, 0, 0, 0, implicit $mode, implicit $exec
FLAT_STORE_DWORD $vgpr0_vgpr1, $agpr4, 0, 0, 0, 0, 0, implicit $mode, implicit $exec, implicit $flat_scr FLAT_STORE_DWORD $vgpr0_vgpr1, $agpr4, 0, 0, 0, 0, 0, implicit $mode, implicit $exec, implicit $flat_scr
... ...
# GCN-LABEL: name: dmfma4x4_write_agpr_valu_read # GCN-LABEL: name: dmfma4x4_write_agpr_valu_read
# GCN: V_MFMA # GCN: V_MFMA
# GCN-NEXT: S_NOP 5 # GCN-NEXT: S_NOP 5
# GCN-NEXT: V_ACCVGPR_READ_B32_e64 # GCN-NEXT: V_ACCVGPR_READ_B32_e64
name: dmfma4x4_write_agpr_valu_read name: dmfma4x4_write_agpr_valu_read
body: | body: |
bb.0: bb.0:
$agpr3_agpr4 = V_MFMA_F64_4X4X4F64_e64 $agpr0_agpr1, $vgpr0_vgpr1, $agpr0_agpr1, 0, 0, 0, implicit $mode, implicit $exec $agpr4_agpr5 = V_MFMA_F64_4X4X4F64_e64 $agpr0_agpr1, $vgpr0_vgpr1, $agpr0_agpr1, 0, 0, 0, implicit $mode, implicit $exec
$vgpr5 = V_ACCVGPR_READ_B32_e64 $agpr4, implicit $exec $vgpr5 = V_ACCVGPR_READ_B32_e64 $agpr5, implicit $exec
... ...
# GCN-LABEL: name: dmfma16x16_write_agpr_valu_read # GCN-LABEL: name: dmfma16x16_write_agpr_valu_read
# GCN: V_MFMA # GCN: V_MFMA
# GCN-NEXT: S_NOP 7 # GCN-NEXT: S_NOP 7
# GCN-NEXT: S_NOP 2 # GCN-NEXT: S_NOP 2
# GCN-NEXT: V_ACCVGPR_READ_B32_e64 # GCN-NEXT: V_ACCVGPR_READ_B32_e64
name: dmfma16x16_write_agpr_valu_read name: dmfma16x16_write_agpr_valu_read
body: | body: |
bb.0: bb.0:
$agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9 = V_MFMA_F64_16X16X4F64_e64 $vgpr10_vgpr11, $vgpr10_vgpr11, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7, 0, 0, 0, implicit $mode, implicit $exec $agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9 = V_MFMA_F64_16X16X4F64_e64 $vgpr10_vgpr11, $vgpr10_vgpr11, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7, 0, 0, 0, implicit $mode, implicit $exec
$vgpr12 = V_ACCVGPR_READ_B32_e64 $agpr4, implicit $exec $vgpr12 = V_ACCVGPR_READ_B32_e64 $agpr4, implicit $exec
... ...
# GCN-LABEL: name: dmfma4x4_write_agpr_valu_write # GCN-LABEL: name: dmfma4x4_write_agpr_valu_write
# GCN: V_MFMA # GCN: V_MFMA
# GCN-NEXT: S_NOP 5 # GCN-NEXT: S_NOP 5
# GCN-NEXT: V_ACCVGPR_WRITE_B32_e64 # GCN-NEXT: V_ACCVGPR_WRITE_B32_e64
name: dmfma4x4_write_agpr_valu_write name: dmfma4x4_write_agpr_valu_write
body: | body: |
bb.0: bb.0:
$agpr3_agpr4 = V_MFMA_F64_4X4X4F64_e64 $vgpr0_vgpr1, $vgpr0_vgpr1, $agpr0_agpr1, 0, 0, 0, implicit $mode, implicit $exec $agpr4_agpr5 = V_MFMA_F64_4X4X4F64_e64 $vgpr0_vgpr1, $vgpr0_vgpr1, $agpr0_agpr1, 0, 0, 0, implicit $mode, implicit $exec
$agpr3 = V_ACCVGPR_WRITE_B32_e64 0, implicit $mode, implicit $exec $agpr4 = V_ACCVGPR_WRITE_B32_e64 0, implicit $mode, implicit $exec
... ...
# GCN-LABEL: name: dmfma16x16_write_agpr_valu_write # GCN-LABEL: name: dmfma16x16_write_agpr_valu_write
# GCN: V_MFMA # GCN: V_MFMA
# GCN-NEXT: S_NOP 7 # GCN-NEXT: S_NOP 7
# GCN-NEXT: S_NOP 2 # GCN-NEXT: S_NOP 2
# GCN-NEXT: V_ACCVGPR_WRITE_B32_e64 # GCN-NEXT: V_ACCVGPR_WRITE_B32_e64
name: dmfma16x16_write_agpr_valu_write name: dmfma16x16_write_agpr_valu_write
body: | body: |
bb.0: bb.0:
$agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9 = V_MFMA_F64_16X16X4F64_e64 $vgpr10_vgpr11, $vgpr10_vgpr11, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7, 0, 0, 0, implicit $mode, implicit $exec $agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9 = V_MFMA_F64_16X16X4F64_e64 $vgpr10_vgpr11, $vgpr10_vgpr11, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7, 0, 0, 0, implicit $mode, implicit $exec
$agpr3 = V_ACCVGPR_WRITE_B32_e64 0, implicit $mode, implicit $exec $agpr3 = V_ACCVGPR_WRITE_B32_e64 0, implicit $mode, implicit $exec
... ...
# GCN-LABEL: name: dmfma4x4_read_srcc_agpr_valu_write # GCN-LABEL: name: dmfma4x4_read_srcc_agpr_valu_write
# GCN: V_MFMA # GCN: V_MFMA
# GCN-NEXT: V_ACCVGPR_WRITE_B32_e64 # GCN-NEXT: V_ACCVGPR_WRITE_B32_e64
name: dmfma4x4_read_srcc_agpr_valu_write name: dmfma4x4_read_srcc_agpr_valu_write
body: | body: |
bb.0: bb.0:
$agpr3_agpr4 = V_MFMA_F64_4X4X4F64_e64 $vgpr0_vgpr1, $vgpr0_vgpr1, $agpr0_agpr1, 0, 0, 0, implicit $mode, implicit $exec $agpr4_agpr5 = V_MFMA_F64_4X4X4F64_e64 $vgpr0_vgpr1, $vgpr0_vgpr1, $agpr0_agpr1, 0, 0, 0, implicit $mode, implicit $exec
$agpr1 = V_ACCVGPR_WRITE_B32_e64 0, implicit $mode, implicit $exec $agpr1 = V_ACCVGPR_WRITE_B32_e64 0, implicit $mode, implicit $exec
... ...
# GCN-LABEL: name: dmfma16x16_read_srcc_agpr_valu_write # GCN-LABEL: name: dmfma16x16_read_srcc_agpr_valu_write
# GCN: V_MFMA # GCN: V_MFMA
# GCN-NEXT: V_ACCVGPR_WRITE_B32_e64 # GCN-NEXT: V_ACCVGPR_WRITE_B32_e64
name: dmfma16x16_read_srcc_agpr_valu_write name: dmfma16x16_read_srcc_agpr_valu_write
body: | body: |
bb.0: bb.0:
Show All 23 Lines

llvm/test/CodeGen/AMDGPU/waitcnt-agpr.mir

Show First 20 Lines • Show All 53 Lines • ▼ Show 20 Lines
name: flat_zero_waitcnt name: flat_zero_waitcnt
body: | body: |
; GCN-LABEL: name: flat_zero_waitcnt ; GCN-LABEL: name: flat_zero_waitcnt
; GCN: bb.0: ; GCN: bb.0:
; GCN: successors: %bb.1(0x80000000) ; GCN: successors: %bb.1(0x80000000)
; GCN: S_WAITCNT 0 ; GCN: S_WAITCNT 0
; GCN: $agpr0 = FLAT_LOAD_DWORD $vgpr1_vgpr2, 0, 0, 0, 0, 0, implicit $exec, implicit $flat_scr :: (load 4 from %ir.global4, addrspace 1) ; GCN: $agpr0 = FLAT_LOAD_DWORD $vgpr2_vgpr3, 0, 0, 0, 0, 0, implicit $exec, implicit $flat_scr :: (load 4 from %ir.global4, addrspace 1)
; GCN: $agpr3_agpr4_agpr5_agpr6 = FLAT_LOAD_DWORDX4 $vgpr7_vgpr8, 0, 0, 0, 0, 0, implicit $exec, implicit $flat_scr :: (load 16 from %ir.global16, addrspace 1) ; GCN: $agpr4_agpr5_agpr6_agpr7 = FLAT_LOAD_DWORDX4 $vgpr8_vgpr9, 0, 0, 0, 0, 0, implicit $exec, implicit $flat_scr :: (load 16 from %ir.global16, addrspace 1)
; GCN: S_WAITCNT 3953 ; GCN: S_WAITCNT 3953
; GCN: $agpr0 = V_ACCVGPR_MOV_B32 $agpr1, implicit $exec ; GCN: $agpr0 = V_ACCVGPR_MOV_B32 $agpr1, implicit $exec
; GCN: S_BRANCH %bb.1 ; GCN: S_BRANCH %bb.1
; GCN: bb.1: ; GCN: bb.1:
; GCN: successors: %bb.2(0x80000000) ; GCN: successors: %bb.2(0x80000000)
; GCN: $agpr0 = FLAT_LOAD_DWORD $vgpr1_vgpr2, 0, 0, 0, 0, 0, implicit $exec, implicit $flat_scr ; GCN: $agpr0 = FLAT_LOAD_DWORD $vgpr2_vgpr3, 0, 0, 0, 0, 0, implicit $exec, implicit $flat_scr
; GCN: S_WAITCNT 3952 ; GCN: S_WAITCNT 3952
; GCN: $agpr3_agpr4_agpr5_agpr6 = FLAT_LOAD_DWORDX4 $vgpr7_vgpr8, 0, 0, 0, 0, 0, implicit $exec, implicit $flat_scr :: (load 16 from %ir.global16, addrspace 1) ; GCN: $agpr4_agpr5_agpr6_agpr7 = FLAT_LOAD_DWORDX4 $vgpr8_vgpr9, 0, 0, 0, 0, 0, implicit $exec, implicit $flat_scr :: (load 16 from %ir.global16, addrspace 1)
; GCN: $vgpr0 = V_ACCVGPR_READ_B32_e64 $agpr1, implicit $exec ; GCN: $vgpr0 = V_ACCVGPR_READ_B32_e64 $agpr1, implicit $exec
; GCN: S_BRANCH %bb.2 ; GCN: S_BRANCH %bb.2
; GCN: bb.2: ; GCN: bb.2:
; GCN: S_WAITCNT 49279 ; GCN: S_WAITCNT 49279
; GCN: $agpr0 = FLAT_LOAD_DWORD $vgpr1_vgpr2, 0, 0, 0, 0, 0, implicit $exec, implicit $flat_scr :: (load 4 from %ir.flat4) ; GCN: $agpr0 = FLAT_LOAD_DWORD $vgpr2_vgpr3, 0, 0, 0, 0, 0, implicit $exec, implicit $flat_scr :: (load 4 from %ir.flat4)
; GCN: S_WAITCNT 3952 ; GCN: S_WAITCNT 3952
; GCN: $agpr3_agpr4_agpr5_agpr6 = FLAT_LOAD_DWORDX4 $vgpr7_vgpr8, 0, 0, 0, 0, 0, implicit $exec, implicit $flat_scr :: (load 16 from %ir.flat16) ; GCN: $agpr4_agpr5_agpr6_agpr7 = FLAT_LOAD_DWORDX4 $vgpr8_vgpr9, 0, 0, 0, 0, 0, implicit $exec, implicit $flat_scr :: (load 16 from %ir.flat16)
; GCN: $vgpr0 = V_ACCVGPR_READ_B32_e64 $agpr1, implicit $exec ; GCN: $vgpr0 = V_ACCVGPR_READ_B32_e64 $agpr1, implicit $exec
; GCN: S_ENDPGM 0 ; GCN: S_ENDPGM 0
bb.0: bb.0:
successors: %bb.1 successors: %bb.1
$agpr0 = FLAT_LOAD_DWORD $vgpr1_vgpr2, 0, 0, 0, 0, 0, implicit $exec, implicit $flat_scr :: (load 4 from %ir.global4) $agpr0 = FLAT_LOAD_DWORD $vgpr2_vgpr3, 0, 0, 0, 0, 0, implicit $exec, implicit $flat_scr :: (load 4 from %ir.global4)
$agpr3_agpr4_agpr5_agpr6 = FLAT_LOAD_DWORDX4 $vgpr7_vgpr8, 0, 0, 0, 0, 0, implicit $exec, implicit $flat_scr :: (load 16 from %ir.global16) $agpr4_agpr5_agpr6_agpr7 = FLAT_LOAD_DWORDX4 $vgpr8_vgpr9, 0, 0, 0, 0, 0, implicit $exec, implicit $flat_scr :: (load 16 from %ir.global16)
$agpr0 = V_ACCVGPR_MOV_B32 $agpr1, implicit $exec $agpr0 = V_ACCVGPR_MOV_B32 $agpr1, implicit $exec
S_BRANCH %bb.1 S_BRANCH %bb.1
bb.1: bb.1:
successors: %bb.2 successors: %bb.2
$agpr0 = FLAT_LOAD_DWORD $vgpr1_vgpr2, 0, 0, 0, 0, 0, implicit $exec, implicit $flat_scr $agpr0 = FLAT_LOAD_DWORD $vgpr2_vgpr3, 0, 0, 0, 0, 0, implicit $exec, implicit $flat_scr
$agpr3_agpr4_agpr5_agpr6 = FLAT_LOAD_DWORDX4 $vgpr7_vgpr8, 0, 0, 0, 0, 0, implicit $exec, implicit $flat_scr :: (load 16 from %ir.global16) $agpr4_agpr5_agpr6_agpr7 = FLAT_LOAD_DWORDX4 $vgpr8_vgpr9, 0, 0, 0, 0, 0, implicit $exec, implicit $flat_scr :: (load 16 from %ir.global16)
$vgpr0 = V_ACCVGPR_READ_B32_e64 $agpr1, implicit $exec $vgpr0 = V_ACCVGPR_READ_B32_e64 $agpr1, implicit $exec
S_BRANCH %bb.2 S_BRANCH %bb.2
bb.2: bb.2:
$agpr0 = FLAT_LOAD_DWORD $vgpr1_vgpr2, 0, 0, 0, 0, 0, implicit $exec, implicit $flat_scr :: (load 4 from %ir.flat4) $agpr0 = FLAT_LOAD_DWORD $vgpr2_vgpr3, 0, 0, 0, 0, 0, implicit $exec, implicit $flat_scr :: (load 4 from %ir.flat4)
$agpr3_agpr4_agpr5_agpr6 = FLAT_LOAD_DWORDX4 $vgpr7_vgpr8, 0, 0, 0, 0, 0, implicit $exec, implicit $flat_scr :: (load 16 from %ir.flat16) $agpr4_agpr5_agpr6_agpr7 = FLAT_LOAD_DWORDX4 $vgpr8_vgpr9, 0, 0, 0, 0, 0, implicit $exec, implicit $flat_scr :: (load 16 from %ir.flat16)
$vgpr0 = V_ACCVGPR_READ_B32_e64 $agpr1, implicit $exec $vgpr0 = V_ACCVGPR_READ_B32_e64 $agpr1, implicit $exec
S_ENDPGM 0 S_ENDPGM 0
... ...
--- ---
# There is only a single fallthrough successor block, so there's no # There is only a single fallthrough successor block, so there's no
# need to wait immediately. # need to wait immediately.
name: single_fallthrough_successor_no_end_block_wait name: single_fallthrough_successor_no_end_block_wait
body: | body: |
; GCN-LABEL: name: single_fallthrough_successor_no_end_block_wait ; GCN-LABEL: name: single_fallthrough_successor_no_end_block_wait
; GCN: bb.0: ; GCN: bb.0:
; GCN: successors: %bb.1(0x80000000) ; GCN: successors: %bb.1(0x80000000)
; GCN: S_WAITCNT 0 ; GCN: S_WAITCNT 0
; GCN: $agpr0 = FLAT_LOAD_DWORD $vgpr1_vgpr2, 0, 0, 0, 0, 0, implicit $exec, implicit $flat_scr ; GCN: $agpr0 = FLAT_LOAD_DWORD $vgpr2_vgpr3, 0, 0, 0, 0, 0, implicit $exec, implicit $flat_scr
; GCN: bb.1: ; GCN: bb.1:
; GCN: $vgpr3_vgpr4 = V_LSHLREV_B64_e64 4, $vgpr7_vgpr8, implicit $exec ; GCN: $vgpr4_vgpr5 = V_LSHLREV_B64_e64 4, $vgpr8_vgpr9, implicit $exec
; GCN: S_WAITCNT 112 ; GCN: S_WAITCNT 112
; GCN: FLAT_STORE_DWORD $vgpr3_vgpr4, $agpr0, 0, 0, 0, 0, 0, implicit $exec, implicit $flat_scr ; GCN: FLAT_STORE_DWORD $vgpr4_vgpr5, $agpr0, 0, 0, 0, 0, 0, implicit $exec, implicit $flat_scr
; GCN: S_ENDPGM 0 ; GCN: S_ENDPGM 0
bb.0: bb.0:
successors: %bb.1 successors: %bb.1
$agpr0 = FLAT_LOAD_DWORD $vgpr1_vgpr2, 0, 0, 0, 0, 0, implicit $exec, implicit $flat_scr $agpr0 = FLAT_LOAD_DWORD $vgpr2_vgpr3, 0, 0, 0, 0, 0, implicit $exec, implicit $flat_scr
bb.1: bb.1:
$vgpr3_vgpr4 = V_LSHLREV_B64_e64 4, $vgpr7_vgpr8, implicit $exec $vgpr4_vgpr5 = V_LSHLREV_B64_e64 4, $vgpr8_vgpr9, implicit $exec
FLAT_STORE_DWORD $vgpr3_vgpr4, $agpr0, 0, 0, 0, 0, 0, implicit $exec, implicit $flat_scr FLAT_STORE_DWORD $vgpr4_vgpr5, $agpr0, 0, 0, 0, 0, 0, implicit $exec, implicit $flat_scr
S_ENDPGM 0 S_ENDPGM 0
... ...
--- ---
# The block has a single predecessor with a single successor, but it # The block has a single predecessor with a single successor, but it
# is not the next block so it's non-obvious that the wait is not needed. # is not the next block so it's non-obvious that the wait is not needed.
name: single_branch_successor_not_next_block name: single_branch_successor_not_next_block
body: | body: |
; GCN-LABEL: name: single_branch_successor_not_next_block ; GCN-LABEL: name: single_branch_successor_not_next_block
; GCN: bb.0: ; GCN: bb.0:
; GCN: successors: %bb.2(0x80000000) ; GCN: successors: %bb.2(0x80000000)
; GCN: S_WAITCNT 0 ; GCN: S_WAITCNT 0
; GCN: $agpr0 = FLAT_LOAD_DWORD $vgpr1_vgpr2, 0, 0, 0, 0, 0, implicit $exec, implicit $flat_scr ; GCN: $agpr0 = FLAT_LOAD_DWORD $vgpr2_vgpr3, 0, 0, 0, 0, 0, implicit $exec, implicit $flat_scr
; GCN: S_BRANCH %bb.2 ; GCN: S_BRANCH %bb.2
; GCN: bb.1: ; GCN: bb.1:
; GCN: FLAT_STORE_DWORD $vgpr8_vgpr9, $agpr10, 0, 0, 0, 0, 0, implicit $exec, implicit $flat_scr ; GCN: FLAT_STORE_DWORD $vgpr8_vgpr9, $agpr10, 0, 0, 0, 0, 0, implicit $exec, implicit $flat_scr
; GCN: S_ENDPGM 0 ; GCN: S_ENDPGM 0
; GCN: bb.2: ; GCN: bb.2:
; GCN: $vgpr3_vgpr4 = V_LSHLREV_B64_e64 4, $vgpr7_vgpr8, implicit $exec ; GCN: $vgpr4_vgpr5 = V_LSHLREV_B64_e64 4, $vgpr8_vgpr9, implicit $exec
; GCN: S_WAITCNT 112 ; GCN: S_WAITCNT 112
; GCN: FLAT_STORE_DWORD $vgpr3_vgpr4, $agpr0, 0, 0, 0, 0, 0, implicit $exec, implicit $flat_scr ; GCN: FLAT_STORE_DWORD $vgpr4_vgpr5, $agpr0, 0, 0, 0, 0, 0, implicit $exec, implicit $flat_scr
; GCN: S_ENDPGM 0 ; GCN: S_ENDPGM 0
bb.0: bb.0:
successors: %bb.2 successors: %bb.2
$agpr0 = FLAT_LOAD_DWORD $vgpr1_vgpr2, 0, 0, 0, 0, 0, implicit $exec, implicit $flat_scr $agpr0 = FLAT_LOAD_DWORD $vgpr2_vgpr3, 0, 0, 0, 0, 0, implicit $exec, implicit $flat_scr
S_BRANCH %bb.2 S_BRANCH %bb.2
bb.1: bb.1:
FLAT_STORE_DWORD $vgpr8_vgpr9, $agpr10, 0, 0, 0, 0, 0, implicit $exec, implicit $flat_scr FLAT_STORE_DWORD $vgpr8_vgpr9, $agpr10, 0, 0, 0, 0, 0, implicit $exec, implicit $flat_scr
S_ENDPGM 0 S_ENDPGM 0
bb.2: bb.2:
$vgpr3_vgpr4 = V_LSHLREV_B64_e64 4, $vgpr7_vgpr8, implicit $exec $vgpr4_vgpr5 = V_LSHLREV_B64_e64 4, $vgpr8_vgpr9, implicit $exec
FLAT_STORE_DWORD $vgpr3_vgpr4, $agpr0, 0, 0, 0, 0, 0, implicit $exec, implicit $flat_scr FLAT_STORE_DWORD $vgpr4_vgpr5, $agpr0, 0, 0, 0, 0, 0, implicit $exec, implicit $flat_scr
S_ENDPGM 0 S_ENDPGM 0
... ...
# GCN-LABEL: name: preexisting_waitcnt{{$}} # GCN-LABEL: name: preexisting_waitcnt{{$}}
# GCN: FLAT_LOAD_DWORD # GCN: FLAT_LOAD_DWORD
# GCN-NEXT: S_WAITCNT 0 # GCN-NEXT: S_WAITCNT 0
# GCN-NOT: S_WAITCNT # GCN-NOT: S_WAITCNT
name: preexisting_waitcnt name: preexisting_waitcnt
tracksRegLiveness: true tracksRegLiveness: true
machineFunctionInfo: machineFunctionInfo:
isEntryFunction: true isEntryFunction: true
body: | body: |
bb.0: bb.0:
liveins: $vgpr1_vgpr2 liveins: $vgpr1_vgpr2
$agpr0 = FLAT_LOAD_DWORD $vgpr1_vgpr2, 0, 0, 0, 0, 0, implicit $exec, implicit $flat_scr $agpr0 = FLAT_LOAD_DWORD $vgpr2_vgpr3, 0, 0, 0, 0, 0, implicit $exec, implicit $flat_scr
S_WAITCNT 0 S_WAITCNT 0
FLAT_STORE_DWORD $vgpr1_vgpr2, $agpr0, 0, 0, 0, 0, 0, implicit $exec, implicit $flat_scr FLAT_STORE_DWORD $vgpr2_vgpr3, $agpr0, 0, 0, 0, 0, 0, implicit $exec, implicit $flat_scr
... ...
--- ---
name: bundle_no_waitcnt name: bundle_no_waitcnt
tracksRegLiveness: true tracksRegLiveness: true
machineFunctionInfo: machineFunctionInfo:
isEntryFunction: true isEntryFunction: true
body: | body: |
bb.0: bb.0:
liveins: $vgpr1_vgpr2 liveins: $vgpr1_vgpr2
; GCN-LABEL: name: bundle_no_waitcnt ; GCN-LABEL: name: bundle_no_waitcnt
; GCN: liveins: $vgpr1_vgpr2 ; GCN: liveins: $vgpr1_vgpr2
; GCN: $agpr0 = FLAT_LOAD_DWORD $vgpr1_vgpr2, 0, 0, 0, 0, 0, implicit $exec, implicit $flat_scr ; GCN: $agpr0 = FLAT_LOAD_DWORD $vgpr2_vgpr3, 0, 0, 0, 0, 0, implicit $exec, implicit $flat_scr
; GCN: BUNDLE { ; GCN: BUNDLE {
; GCN: S_NOP 0 ; GCN: S_NOP 0
; GCN: S_NOP 0 ; GCN: S_NOP 0
; GCN: } ; GCN: }
; GCN: S_WAITCNT 112 ; GCN: S_WAITCNT 112
; GCN: FLAT_STORE_DWORD $vgpr1_vgpr2, $agpr0, 0, 0, 0, 0, 0, implicit $exec, implicit $flat_scr ; GCN: FLAT_STORE_DWORD $vgpr2_vgpr3, $agpr0, 0, 0, 0, 0, 0, implicit $exec, implicit $flat_scr
$agpr0 = FLAT_LOAD_DWORD $vgpr1_vgpr2, 0, 0, 0, 0, 0, implicit $exec, implicit $flat_scr $agpr0 = FLAT_LOAD_DWORD $vgpr2_vgpr3, 0, 0, 0, 0, 0, implicit $exec, implicit $flat_scr
BUNDLE { BUNDLE {
S_NOP 0 S_NOP 0
S_NOP 0 S_NOP 0
} }
FLAT_STORE_DWORD $vgpr1_vgpr2, $agpr0, 0, 0, 0, 0, 0, implicit $exec, implicit $flat_scr FLAT_STORE_DWORD $vgpr2_vgpr3, $agpr0, 0, 0, 0, 0, 0, implicit $exec, implicit $flat_scr
... ...
--- ---
# See the waitcnt inside the bundle and don't insert an extra # See the waitcnt inside the bundle and don't insert an extra
name: preexisting_waitcnt_in_bundle name: preexisting_waitcnt_in_bundle
tracksRegLiveness: true tracksRegLiveness: true
machineFunctionInfo: machineFunctionInfo:
isEntryFunction: true isEntryFunction: true
body: | body: |
bb.0: bb.0:
liveins: $vgpr1_vgpr2 liveins: $vgpr1_vgpr2
; GCN-LABEL: name: preexisting_waitcnt_in_bundle ; GCN-LABEL: name: preexisting_waitcnt_in_bundle
; GCN: liveins: $vgpr1_vgpr2 ; GCN: liveins: $vgpr1_vgpr2
; GCN: $agpr0 = FLAT_LOAD_DWORD $vgpr1_vgpr2, 0, 0, 0, 0, 0, implicit $exec, implicit $flat_scr ; GCN: $agpr0 = FLAT_LOAD_DWORD $vgpr2_vgpr3, 0, 0, 0, 0, 0, implicit $exec, implicit $flat_scr
; GCN: BUNDLE { ; GCN: BUNDLE {
; GCN: S_NOP 0 ; GCN: S_NOP 0
; GCN: S_WAITCNT 0 ; GCN: S_WAITCNT 0
; GCN: } ; GCN: }
; GCN: FLAT_STORE_DWORD $vgpr1_vgpr2, $agpr0, 0, 0, 0, 0, 0, implicit $exec, implicit $flat_scr ; GCN: FLAT_STORE_DWORD $vgpr2_vgpr3, $agpr0, 0, 0, 0, 0, 0, implicit $exec, implicit $flat_scr
$agpr0 = FLAT_LOAD_DWORD $vgpr1_vgpr2, 0, 0, 0, 0, 0, implicit $exec, implicit $flat_scr $agpr0 = FLAT_LOAD_DWORD $vgpr2_vgpr3, 0, 0, 0, 0, 0, implicit $exec, implicit $flat_scr
BUNDLE { BUNDLE {
S_NOP 0 S_NOP 0
S_WAITCNT 0 S_WAITCNT 0
} }
FLAT_STORE_DWORD $vgpr1_vgpr2, $agpr0, 0, 0, 0, 0, 0, implicit $exec, implicit $flat_scr FLAT_STORE_DWORD $vgpr2_vgpr3, $agpr0, 0, 0, 0, 0, 0, implicit $exec, implicit $flat_scr
... ...
--- ---
# Def and use inside bundle # Def and use inside bundle
name: insert_in_bundle name: insert_in_bundle
tracksRegLiveness: true tracksRegLiveness: true
machineFunctionInfo: machineFunctionInfo:
isEntryFunction: true isEntryFunction: true
body: | body: |
bb.0: bb.0:
liveins: $vgpr1_vgpr2 liveins: $vgpr1_vgpr2
; GCN-LABEL: name: insert_in_bundle ; GCN-LABEL: name: insert_in_bundle
; GCN: liveins: $vgpr1_vgpr2 ; GCN: liveins: $vgpr1_vgpr2
; GCN: BUNDLE implicit-def $agpr0, implicit $vgpr1_vgpr2 { ; GCN: BUNDLE implicit-def $agpr0, implicit $vgpr2_vgpr3 {
; GCN: $agpr0 = FLAT_LOAD_DWORD $vgpr1_vgpr2, 0, 0, 0, 0, 0, implicit $exec, implicit $flat_scr ; GCN: $agpr0 = FLAT_LOAD_DWORD $vgpr2_vgpr3, 0, 0, 0, 0, 0, implicit $exec, implicit $flat_scr
; GCN: S_WAITCNT 112 ; GCN: S_WAITCNT 112
; GCN: FLAT_STORE_DWORD $vgpr1_vgpr2, internal $agpr0, 0, 0, 0, 0, 0, implicit $exec, implicit $flat_scr ; GCN: FLAT_STORE_DWORD $vgpr2_vgpr3, internal $agpr0, 0, 0, 0, 0, 0, implicit $exec, implicit $flat_scr
; GCN: } ; GCN: }
BUNDLE implicit-def $agpr0, implicit $vgpr1_vgpr2 { BUNDLE implicit-def $agpr0, implicit $vgpr2_vgpr3 {
$agpr0 = FLAT_LOAD_DWORD $vgpr1_vgpr2, 0, 0, 0, 0, 0, implicit $exec, implicit $flat_scr $agpr0 = FLAT_LOAD_DWORD $vgpr2_vgpr3, 0, 0, 0, 0, 0, implicit $exec, implicit $flat_scr
FLAT_STORE_DWORD $vgpr1_vgpr2, internal $agpr0, 0, 0, 0, 0, 0, implicit $exec, implicit $flat_scr FLAT_STORE_DWORD $vgpr2_vgpr3, internal $agpr0, 0, 0, 0, 0, 0, implicit $exec, implicit $flat_scr
} }
... ...
--- ---
# Def is last instruction in bundle, use is outside bundle # Def is last instruction in bundle, use is outside bundle
name: exit_bundle name: exit_bundle
tracksRegLiveness: true tracksRegLiveness: true
machineFunctionInfo: machineFunctionInfo:
isEntryFunction: true isEntryFunction: true
body: | body: |
bb.0: bb.0:
liveins: $vgpr1_vgpr2 liveins: $vgpr1_vgpr2
; GCN-LABEL: name: exit_bundle ; GCN-LABEL: name: exit_bundle
; GCN: liveins: $vgpr1_vgpr2 ; GCN: liveins: $vgpr1_vgpr2
; GCN: BUNDLE implicit-def $agpr0, implicit $vgpr1_vgpr2 { ; GCN: BUNDLE implicit-def $agpr0, implicit $vgpr2_vgpr3 {
; GCN: $agpr0 = FLAT_LOAD_DWORD $vgpr1_vgpr2, 0, 0, 0, 0, 0, implicit $exec, implicit $flat_scr ; GCN: $agpr0 = FLAT_LOAD_DWORD $vgpr2_vgpr3, 0, 0, 0, 0, 0, implicit $exec, implicit $flat_scr
; GCN: } ; GCN: }
; GCN: S_WAITCNT 112 ; GCN: S_WAITCNT 112
; GCN: FLAT_STORE_DWORD $vgpr1_vgpr2, $agpr0, 0, 0, 0, 0, 0, implicit $exec, implicit $flat_scr ; GCN: FLAT_STORE_DWORD $vgpr2_vgpr3, $agpr0, 0, 0, 0, 0, 0, implicit $exec, implicit $flat_scr
BUNDLE implicit-def $agpr0, implicit $vgpr1_vgpr2 { BUNDLE implicit-def $agpr0, implicit $vgpr2_vgpr3 {
$agpr0 = FLAT_LOAD_DWORD $vgpr1_vgpr2, 0, 0, 0, 0, 0, implicit $exec, implicit $flat_scr $agpr0 = FLAT_LOAD_DWORD $vgpr2_vgpr3, 0, 0, 0, 0, 0, implicit $exec, implicit $flat_scr
} }
FLAT_STORE_DWORD $vgpr1_vgpr2, $agpr0, 0, 0, 0, 0, 0, implicit $exec, implicit $flat_scr FLAT_STORE_DWORD $vgpr2_vgpr3, $agpr0, 0, 0, 0, 0, 0, implicit $exec, implicit $flat_scr
... ...
--- ---
# Def is in bundle, use is in another bundle # Def is in bundle, use is in another bundle
name: cross_bundle name: cross_bundle
tracksRegLiveness: true tracksRegLiveness: true
machineFunctionInfo: machineFunctionInfo:
isEntryFunction: true isEntryFunction: true
body: | body: |
bb.0: bb.0:
liveins: $vgpr1_vgpr2 liveins: $vgpr1_vgpr2
; GCN-LABEL: name: cross_bundle ; GCN-LABEL: name: cross_bundle
; GCN: liveins: $vgpr1_vgpr2 ; GCN: liveins: $vgpr1_vgpr2
; GCN: BUNDLE implicit-def $agpr0, implicit $vgpr1_vgpr2 { ; GCN: BUNDLE implicit-def $agpr0, implicit $vgpr2_vgpr3 {
; GCN: $agpr0 = FLAT_LOAD_DWORD $vgpr1_vgpr2, 0, 0, 0, 0, 0, implicit $exec, implicit $flat_scr ; GCN: $agpr0 = FLAT_LOAD_DWORD $vgpr2_vgpr3, 0, 0, 0, 0, 0, implicit $exec, implicit $flat_scr
; GCN: } ; GCN: }
; GCN: S_WAITCNT 112 ; GCN: S_WAITCNT 112
; GCN: BUNDLE implicit $agpr0, implicit $vgpr1_vgpr2 { ; GCN: BUNDLE implicit $agpr0, implicit $vgpr2_vgpr3 {
; GCN: FLAT_STORE_DWORD $vgpr1_vgpr2, $agpr0, 0, 0, 0, 0, 0, implicit $exec, implicit $flat_scr ; GCN: FLAT_STORE_DWORD $vgpr2_vgpr3, $agpr0, 0, 0, 0, 0, 0, implicit $exec, implicit $flat_scr
; GCN: } ; GCN: }
BUNDLE implicit-def $agpr0, implicit $vgpr1_vgpr2 { BUNDLE implicit-def $agpr0, implicit $vgpr2_vgpr3 {
$agpr0 = FLAT_LOAD_DWORD $vgpr1_vgpr2, 0, 0, 0, 0, 0, implicit $exec, implicit $flat_scr $agpr0 = FLAT_LOAD_DWORD $vgpr2_vgpr3, 0, 0, 0, 0, 0, implicit $exec, implicit $flat_scr
} }
BUNDLE implicit $agpr0, implicit $vgpr1_vgpr2 { BUNDLE implicit $agpr0, implicit $vgpr2_vgpr3 {
FLAT_STORE_DWORD $vgpr1_vgpr2, $agpr0, 0, 0, 0, 0, 0, implicit $exec, implicit $flat_scr FLAT_STORE_DWORD $vgpr2_vgpr3, $agpr0, 0, 0, 0, 0, 0, implicit $exec, implicit $flat_scr
} }
... ...