caio/clang-p2996
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity
Files
8fcb1263f42657ecbc355beff12500dfbcddee17
clang-p2996/llvm/test/CodeGen/AMDGPU/memory_clause.mir
Lucas Ramirez 6206f5444f [AMDGPU] Occupancy w.r.t. workgroup size range is also a range (#123748) 
Occupancy (i.e., the number of waves per EU) depends, in addition to
register usage, on per-workgroup LDS usage as well as on the range of
possible workgroup sizes. Mirroring the latter, occupancy should
therefore be expressed as a range since different group sizes generally
yield different achievable occupancies.

`getOccupancyWithLocalMemSize` currently returns a scalar occupancy
based on the maximum workgroup size and LDS usage. With respect to the
workgroup size range, this scalar can be the minimum, the maximum, or
neither of the two of the range of achievable occupancies. This commit
fixes the function by making it compute and return the range of
achievable occupancies w.r.t. workgroup size and LDS usage; it also
renames it to `getOccupancyWithWorkGroupSizes` since it is the range of
workgroup sizes that produces the range of achievable occupancies.

Computing the achievable occupancy range is surprisingly involved.
Minimum/maximum workgroup sizes do not necessarily yield maximum/minimum
occupancies i.e., sometimes workgroup sizes inside the range yield the
occupancy bounds. The implementation finds these sizes in constant time;
heavy documentation explains the rationale behind the sometimes
relatively obscure calculations.

As a justifying example, consider a target with 10 waves / EU, 4 EUs/CU,
64-wide waves. Also consider a function with no LDS usage and a flat
workgroup size range of [513,1024].

- A group of 513 items requires 9 waves per group. Only 4 groups made up
of 9 waves each can fit fully on a CU at any given time, for a total of
36 waves on the CU, or 9 per EU. However, filling as much as possible
the remaining 40-36=4 wave slots without decreasing the number of groups
reveals that a larger group of 640 items yields 40 waves on the CU, or
10 per EU.
- Similarly, a group of 1024 items requires 16 waves per group. Only 2
groups made up of 16 waves each can fit fully on a CU ay any given time,
for a total of 32 waves on the CU, or 8 per EU. However, removing as
many waves as possible from the groups without being able to fit another
equal-sized group on the CU reveals that a smaller group of 896 items
yields 28 waves on the CU, or 7 per EU.

Therefore the achievable occupancy range for this function is not [8,9]
as the group size bounds directly yield, but [7,10].

Naturally this change causes a lot of test churn as instruction
scheduling is driven by achievable occupancy estimates. In most unit
tests the flat workgroup size range is the default [1,1024] which,
ignoring potential LDS limitations, would previously produce a scalar
occupancy of 8 (derived from 1024) on a lot of targets, whereas we now
consider the maximum occupancy to be 10 in such cases. Most tests are
updated automatically and checked manually for sanity. I also manually
changed some non-automatically generated assertions when necessary.

Fixes #118220.
2025-01-23 16:07:57 +01:00

553 lines
23 KiB
YAML
Raw Blame History

# RUN: llc -mtriple=amdgcn -mcpu=gfx902 -verify-machineinstrs -run-pass=si-form-memory-clauses %s -o - | FileCheck -check-prefix=GCN %s
# GCN-LABEL: {{^}}name: vector_clause{{$}}
# GCN: %1:vreg_128 = GLOBAL_LOAD_DWORDX4 %0, 0, 0, implicit $exec
# GCN-NEXT: %2:vreg_128 = GLOBAL_LOAD_DWORDX4 %0, 16, 0, implicit $exec
# GCN-NEXT: %3:vreg_128 = GLOBAL_LOAD_DWORDX4 %0, 32, 0, implicit $exec
# GCN-NEXT: %4:vreg_128 = GLOBAL_LOAD_DWORDX4 %0, 48, 0, implicit $exec
# GCN-NEXT: KILL %0{{$}}
# GCN-NEXT: %5:vreg_64 = IMPLICIT_DEF
# GCN-NEXT: GLOBAL_STORE_DWORDX4 %5, %1, 0, 0, implicit $exec
---
name: vector_clause
tracksRegLiveness: true
body: |
bb.0:
%0:vreg_64 = IMPLICIT_DEF
%1:vreg_128 = GLOBAL_LOAD_DWORDX4 %0, 0, 0, implicit $exec
%2:vreg_128 = GLOBAL_LOAD_DWORDX4 %0, 16, 0, implicit $exec
%3:vreg_128 = GLOBAL_LOAD_DWORDX4 %0, 32, 0, implicit $exec
%4:vreg_128 = GLOBAL_LOAD_DWORDX4 %0, 48, 0, implicit $exec
%5:vreg_64 = IMPLICIT_DEF
GLOBAL_STORE_DWORDX4 %5, %1, 0, 0, implicit $exec
GLOBAL_STORE_DWORDX4 %5, %2, 16, 0, implicit $exec
GLOBAL_STORE_DWORDX4 %5, %3, 32, 0, implicit $exec
GLOBAL_STORE_DWORDX4 %5, %4, 48, 0, implicit $exec
...
# This would be a valid soft clause, but there's no need for a KILL
# since the pointer uses are live beyond the end the clause.
# GCN-LABEL: {{^}}name: vector_clause_no_kill{{$}}
# GCN: %1:vreg_128 = GLOBAL_LOAD_DWORDX4 %0, 0, 0, implicit $exec
# GCN-NEXT: %2:vreg_128 = GLOBAL_LOAD_DWORDX4 %0, 16, 0, implicit $exec
# GCN-NEXT: %3:vreg_128 = GLOBAL_LOAD_DWORDX4 %0, 32, 0, implicit $exec
# GCN-NEXT: %4:vreg_128 = GLOBAL_LOAD_DWORDX4 %0, 48, 0, implicit $exec
# GCN-NEXT: GLOBAL_STORE_DWORDX4 %0, %1, 0, 0, implicit $exec
---
name: vector_clause_no_kill
tracksRegLiveness: true
body: |
bb.0:
%0:vreg_64 = IMPLICIT_DEF
%1:vreg_128 = GLOBAL_LOAD_DWORDX4 %0, 0, 0, implicit $exec
%2:vreg_128 = GLOBAL_LOAD_DWORDX4 %0, 16, 0, implicit $exec
%3:vreg_128 = GLOBAL_LOAD_DWORDX4 %0, 32, 0, implicit $exec
%4:vreg_128 = GLOBAL_LOAD_DWORDX4 %0, 48, 0, implicit $exec
GLOBAL_STORE_DWORDX4 %0, %1, 0, 0, implicit $exec
GLOBAL_STORE_DWORDX4 %0, %2, 16, 0, implicit $exec
GLOBAL_STORE_DWORDX4 %0, %3, 32, 0, implicit $exec
GLOBAL_STORE_DWORDX4 %0, %4, 48, 0, implicit $exec
...
# GCN-LABEL: {{^}}name: subreg_full{{$}}
# GCN: undef %1.sub0:vreg_128 = GLOBAL_LOAD_DWORD %0.sub0_sub1, 0, 0, implicit $exec
# GCN-NEXT: %1.sub1:vreg_128 = GLOBAL_LOAD_DWORD %0.sub1_sub2, 16, 0, implicit $exec
# GCN-NEXT: %1.sub2:vreg_128 = GLOBAL_LOAD_DWORD %0.sub2_sub3, 32, 0, implicit $exec
# GCN-NEXT: %1.sub3:vreg_128 = GLOBAL_LOAD_DWORD %0.sub2_sub3, 32, 0, implicit $exec
# GCN-NEXT: KILL %0.sub2_sub3{{$}}
# GCN-NEXT: GLOBAL_STORE_DWORDX4 %0.sub0_sub1, %1, 0, 0, implicit $exec
---
name: subreg_full
tracksRegLiveness: true
registers:
- { id: 0, class: vreg_128 }
- { id: 1, class: vreg_128 }
body: |
bb.0:
%0 = IMPLICIT_DEF
undef %1.sub0:vreg_128 = GLOBAL_LOAD_DWORD %0.sub0_sub1, 0, 0, implicit $exec
%1.sub1:vreg_128 = GLOBAL_LOAD_DWORD %0.sub1_sub2, 16, 0, implicit $exec
%1.sub2:vreg_128 = GLOBAL_LOAD_DWORD %0.sub2_sub3, 32, 0, implicit $exec
%1.sub3:vreg_128 = GLOBAL_LOAD_DWORD %0.sub2_sub3, 32, 0, implicit $exec
GLOBAL_STORE_DWORDX4 %0.sub0_sub1, %1, 0, 0, implicit $exec
...
# GCN-LABEL: {{^}}name: subreg_part{{$}}
# GCN: undef %1.sub0:vreg_128 = GLOBAL_LOAD_DWORD %0.sub0_sub1, 0, 0, implicit $exec
# GCN-NEXT: %1.sub1:vreg_128 = GLOBAL_LOAD_DWORD %0.sub1_sub2, 16, 0, implicit $exec
# GCN-NEXT: %1.sub3:vreg_128 = GLOBAL_LOAD_DWORD %0.sub2_sub3, 32, 0, implicit $exec
# GCN-NEXT: KILL %0.sub2_sub3{{$}}
# GCN-NEXT: GLOBAL_STORE_DWORDX4 %0.sub0_sub1, %1, 0, 0, implicit $exec
---
name: subreg_part
tracksRegLiveness: true
registers:
- { id: 0, class: vreg_128 }
- { id: 1, class: vreg_128 }
body: |
bb.0:
%0 = IMPLICIT_DEF
undef %1.sub0:vreg_128 = GLOBAL_LOAD_DWORD %0.sub0_sub1, 0, 0, implicit $exec
%1.sub1:vreg_128 = GLOBAL_LOAD_DWORD %0.sub1_sub2, 16, 0, implicit $exec
%1.sub3:vreg_128 = GLOBAL_LOAD_DWORD %0.sub2_sub3, 32, 0, implicit $exec
GLOBAL_STORE_DWORDX4 %0.sub0_sub1, %1, 0, 0, implicit $exec
...
# GCN-LABEL: {{^}}name: dead{{$}}
# GCN: dead %1:vreg_128 = GLOBAL_LOAD_DWORDX4 %0, 0, 0, implicit $exec
# GCN-NEXT: %2:vreg_128 = GLOBAL_LOAD_DWORDX4 %0, 16, 0, implicit $exec
# GCN-NEXT: dead %3:vreg_128 = GLOBAL_LOAD_DWORDX4 %0, 32, 0, implicit $exec
# GCN-NEXT: dead %4:vreg_128 = GLOBAL_LOAD_DWORDX4 %0, 48, 0, implicit $exec
# GCN-NEXT: KILL %0{{$}}
---
name: dead
tracksRegLiveness: true
registers:
- { id: 0, class: vreg_64 }
- { id: 1, class: vreg_128 }
- { id: 2, class: vreg_128 }
- { id: 3, class: vreg_128 }
- { id: 4, class: vreg_128 }
body: |
bb.0:
%0 = IMPLICIT_DEF
dead %1:vreg_128 = GLOBAL_LOAD_DWORDX4 %0, 0, 0, implicit $exec
dead %2:vreg_128 = GLOBAL_LOAD_DWORDX4 %0, 16, 0, implicit $exec
dead %3:vreg_128 = GLOBAL_LOAD_DWORDX4 %0, 32, 0, implicit $exec
dead %4:vreg_128 = GLOBAL_LOAD_DWORDX4 %0, 48, 0, implicit $exec
...
# GCN-LABEL: {{^}}name: subreg_dead{{$}}
# GCN: undef %2.sub0:vreg_64 = GLOBAL_LOAD_DWORD %0, 16, 0, implicit $exec
# GCN-NEXT: dead %2.sub1:vreg_64 = GLOBAL_LOAD_DWORD %0, 32, 0, implicit $exec
# GCN-NEXT: KILL %0{{$}}
# GCN-NEXT: GLOBAL_STORE_DWORD %1, %2.sub0, 0, 0, implicit $exec
---
name: subreg_dead
tracksRegLiveness: true
body: |
bb.0:
%0:vreg_64 = IMPLICIT_DEF
%1:vreg_64 = IMPLICIT_DEF
undef %2.sub0:vreg_64 = GLOBAL_LOAD_DWORD %0, 16, 0, implicit $exec
dead %2.sub1:vreg_64 = GLOBAL_LOAD_DWORD %0, 32, 0, implicit $exec
GLOBAL_STORE_DWORD %1, %2.sub0, 0, 0, implicit $exec
...
# GCN-LABEL: {{^}}name: kill{{$}}
# GCN: %3:vreg_128 = GLOBAL_LOAD_DWORDX4 %0, 0, 0, implicit $exec
# GCN-NEXT: %4:vreg_128 = GLOBAL_LOAD_DWORDX4 %1, 16, 0, implicit $exec
# GCN-NEXT: KILL %1{{$}}
# GCN-NEXT: KILL %0{{$}}
---
name: kill
tracksRegLiveness: true
body: |
bb.0:
%0:vreg_64 = IMPLICIT_DEF
%1:vreg_64 = IMPLICIT_DEF
%2:vreg_64 = IMPLICIT_DEF
%3:vreg_128 = GLOBAL_LOAD_DWORDX4 %0, 0, 0, implicit $exec
%4:vreg_128 = GLOBAL_LOAD_DWORDX4 killed %1, 16, 0, implicit $exec
GLOBAL_STORE_DWORDX4 %2, %3, 0, 0, implicit $exec
GLOBAL_STORE_DWORDX4 %2, %4, 16, 0, implicit $exec
...
# GCN-LABEL: {{^}}name: indirect{{$}}
# GCN: %1:vreg_64 = GLOBAL_LOAD_DWORDX2 %0, 0, 0, implicit $exec
# GCN-NEXT: %2:vreg_128 = GLOBAL_LOAD_DWORDX4 %1, 0, 0, implicit $exec
# GCN-NEXT: %3:vreg_128 = GLOBAL_LOAD_DWORDX4 %1, 16, 0, implicit $exec
# GCN-NEXT: KILL %1{{$}}
---
name: indirect
tracksRegLiveness: true
registers:
- { id: 0, class: vreg_64 }
- { id: 1, class: vreg_64 }
- { id: 2, class: vreg_128 }
- { id: 3, class: vreg_128 }
body: |
bb.0:
%0 = IMPLICIT_DEF
%1:vreg_64 = GLOBAL_LOAD_DWORDX2 %0, 0, 0, implicit $exec
%2:vreg_128 = GLOBAL_LOAD_DWORDX4 %1, 0, 0, implicit $exec
%3:vreg_128 = GLOBAL_LOAD_DWORDX4 %1, 16, 0, implicit $exec
GLOBAL_STORE_DWORDX4 %0, %2, 0, 0, implicit $exec
GLOBAL_STORE_DWORDX4 %0, %3, 16, 0, implicit $exec
...
# GCN-LABEL: {{^}}name: stack{{$}}
# GCN: %0:vreg_64 = IMPLICIT_DEF
# GCN-NEXT: %1:vreg_128 = GLOBAL_LOAD_DWORDX4 %stack.0, 0, 0, implicit $exec
# GCN-NEXT: %2:vreg_128 = GLOBAL_LOAD_DWORDX4 %stack.0, 16, 0, implicit $exec
# GCN-NEXT: GLOBAL_STORE_DWORDX4 %0, %1, 0, 0, implicit $exec
---
name: stack
tracksRegLiveness: true
registers:
- { id: 0, class: vreg_64 }
- { id: 1, class: vreg_128 }
- { id: 2, class: vreg_128 }
stack:
- { id: 0, type: default, offset: 0, size: 64, alignment: 8 }
body: |
bb.0:
%0 = IMPLICIT_DEF
%1:vreg_128 = GLOBAL_LOAD_DWORDX4 %stack.0, 0, 0, implicit $exec
%2:vreg_128 = GLOBAL_LOAD_DWORDX4 %stack.0, 16, 0, implicit $exec
GLOBAL_STORE_DWORDX4 %0, %1, 0, 0, implicit $exec
GLOBAL_STORE_DWORDX4 %0, %2, 16, 0, implicit $exec
...
# GCN-LABEL: {{^}}name: overflow_counter{{$}}
# GCN: dead %2:vgpr_32 = GLOBAL_LOAD_DWORD %0, 0, 0, implicit $exec
# GCN-NEXT: dead %3:vgpr_32 = GLOBAL_LOAD_DWORD %0, 4, 0, implicit $exec
# GCN-NEXT: dead %4:vgpr_32 = GLOBAL_LOAD_DWORD %0, 8, 0, implicit $exec
# GCN-NEXT: dead %5:vgpr_32 = GLOBAL_LOAD_DWORD %0, 12, 0, implicit $exec
# GCN-NEXT: dead %6:vgpr_32 = GLOBAL_LOAD_DWORD %0, 16, 0, implicit $exec
# GCN-NEXT: dead %7:vgpr_32 = GLOBAL_LOAD_DWORD %0, 20, 0, implicit $exec
# GCN-NEXT: dead %8:vgpr_32 = GLOBAL_LOAD_DWORD %0, 24, 0, implicit $exec
# GCN-NEXT: dead %9:vgpr_32 = GLOBAL_LOAD_DWORD %0, 28, 0, implicit $exec
# GCN-NEXT: dead %10:vgpr_32 = GLOBAL_LOAD_DWORD %0, 32, 0, implicit $exec
# GCN-NEXT: dead %11:vgpr_32 = GLOBAL_LOAD_DWORD %0, 36, 0, implicit $exec
# GCN-NEXT: dead %12:vgpr_32 = GLOBAL_LOAD_DWORD %0, 40, 0, implicit $exec
# GCN-NEXT: dead %13:vgpr_32 = GLOBAL_LOAD_DWORD %0, 44, 0, implicit $exec
# GCN-NEXT: dead %14:vgpr_32 = GLOBAL_LOAD_DWORD %0, 48, 0, implicit $exec
# GCN-NEXT: dead %15:vgpr_32 = GLOBAL_LOAD_DWORD %0, 52, 0, implicit $exec
# GCN-NEXT: dead %16:vgpr_32 = GLOBAL_LOAD_DWORD %0, 56, 0, implicit $exec
# GCN-NEXT: KILL %0{{$}}
# GCN-NEXT: dead %17:vgpr_32 = GLOBAL_LOAD_DWORD %1, 60, 0, implicit $exec
# GCN-NEXT: dead %18:vgpr_32 = GLOBAL_LOAD_DWORD %1, 64, 0, implicit $exec
# GCN-NEXT: KILL %1{{$}}
---
name: overflow_counter
tracksRegLiveness: true
body: |
bb.0:
%0:vreg_64 = IMPLICIT_DEF
%1:vreg_64 = IMPLICIT_DEF
%2:vgpr_32 = GLOBAL_LOAD_DWORD %0, 0, 0, implicit $exec
%3:vgpr_32 = GLOBAL_LOAD_DWORD %0, 4, 0, implicit $exec
%4:vgpr_32 = GLOBAL_LOAD_DWORD %0, 8, 0, implicit $exec
%5:vgpr_32 = GLOBAL_LOAD_DWORD %0, 12, 0, implicit $exec
%6:vgpr_32 = GLOBAL_LOAD_DWORD %0, 16, 0, implicit $exec
%7:vgpr_32 = GLOBAL_LOAD_DWORD %0, 20, 0, implicit $exec
%8:vgpr_32 = GLOBAL_LOAD_DWORD %0, 24, 0, implicit $exec
%9:vgpr_32 = GLOBAL_LOAD_DWORD %0, 28, 0, implicit $exec
%10:vgpr_32 = GLOBAL_LOAD_DWORD %0, 32, 0, implicit $exec
%11:vgpr_32 = GLOBAL_LOAD_DWORD %0, 36, 0, implicit $exec
%12:vgpr_32 = GLOBAL_LOAD_DWORD %0, 40, 0, implicit $exec
%13:vgpr_32 = GLOBAL_LOAD_DWORD %0, 44, 0, implicit $exec
%14:vgpr_32 = GLOBAL_LOAD_DWORD %0, 48, 0, implicit $exec
%15:vgpr_32 = GLOBAL_LOAD_DWORD %0, 52, 0, implicit $exec
%16:vgpr_32 = GLOBAL_LOAD_DWORD %0, 56, 0, implicit $exec
%17:vgpr_32 = GLOBAL_LOAD_DWORD %1, 60, 0, implicit $exec
%18:vgpr_32 = GLOBAL_LOAD_DWORD %1, 64, 0, implicit $exec
...
# GCN-LABEL: {{^}}name: reg_pressure{{$}}
# GCN: dead %2:vreg_128 = GLOBAL_LOAD_DWORDX4 %0, 0, 0, implicit $exec
# GCN-NEXT: dead %3:vreg_128 = GLOBAL_LOAD_DWORDX4 %0, 16, 0, implicit $exec
# GCN-NEXT: dead %4:vreg_128 = GLOBAL_LOAD_DWORDX4 %0, 32, 0, implicit $exec
# GCN-NEXT: dead %5:vreg_128 = GLOBAL_LOAD_DWORDX4 %0, 48, 0, implicit $exec
# GCN-NEXT: dead %6:vreg_128 = GLOBAL_LOAD_DWORDX4 %0, 64, 0, implicit $exec
# GCN-NEXT: dead %7:vreg_128 = GLOBAL_LOAD_DWORDX4 %0, 80, 0, implicit $exec
# GCN-NEXT: dead %8:vreg_128 = GLOBAL_LOAD_DWORDX4 %0, 96, 0, implicit $exec
# GCN-NEXT: dead %9:vreg_128 = GLOBAL_LOAD_DWORDX4 %1, 128, 0, implicit $exec
# GCN-NEXT: dead %10:vreg_128 = GLOBAL_LOAD_DWORDX4 %1, 144, 0, implicit $exec
# GCN-NEXT: KILL %1{{$}}
# GCN-NEXT: KILL %0{{$}}
---
name: reg_pressure
tracksRegLiveness: true
body: |
bb.0:
%0:vreg_64 = IMPLICIT_DEF
%1:vreg_64 = IMPLICIT_DEF
%2:vreg_128 = GLOBAL_LOAD_DWORDX4 %0, 0, 0, implicit $exec
%3:vreg_128 = GLOBAL_LOAD_DWORDX4 %0, 16, 0, implicit $exec
%4:vreg_128 = GLOBAL_LOAD_DWORDX4 %0, 32, 0, implicit $exec
%5:vreg_128 = GLOBAL_LOAD_DWORDX4 %0, 48, 0, implicit $exec
%6:vreg_128 = GLOBAL_LOAD_DWORDX4 %0, 64, 0, implicit $exec
%7:vreg_128 = GLOBAL_LOAD_DWORDX4 %0, 80, 0, implicit $exec
%8:vreg_128 = GLOBAL_LOAD_DWORDX4 %0, 96, 0, implicit $exec
%9:vreg_128 = GLOBAL_LOAD_DWORDX4 %1, 128, 0, implicit $exec
%10:vreg_128 = GLOBAL_LOAD_DWORDX4 %1, 144, 0, implicit $exec
...
# GCN-LABEL: {{^}}name: image_clause{{$}}
# GCN: %3:vreg_128 = IMAGE_SAMPLE_LZ_V4_V2 %0, %1, undef %2:sgpr_128, 15, 0, 0, 0, 0, 0, 0, 0, implicit $exec
# GCN-NEXT: %4:vreg_128 = IMAGE_SAMPLE_LZ_V4_V2 %0, %1, undef %2:sgpr_128, 15, 0, 0, 0, 0, 0, 0, 0, implicit $exec
# GCN-NEXT: %5:vreg_128 = IMAGE_SAMPLE_LZ_V4_V2 %0, %1, undef %2:sgpr_128, 15, 0, 0, 0, 0, 0, 0, 0, implicit $exec
# GCN-NEXT: KILL undef %2:sgpr_128{{$}}
# GCN-NEXT: IMAGE_STORE_V4_V2 %3, %0, %1, 15, -1, 0, 0, 0, 0, 0, 0, implicit $exec
---
name: image_clause
tracksRegLiveness: true
registers:
- { id: 0, class: vreg_64 }
- { id: 1, class: sgpr_256 }
- { id: 2, class: sgpr_128 }
- { id: 3, class: vreg_128 }
- { id: 4, class: vreg_128 }
- { id: 5, class: vreg_128 }
body: |
bb.0:
%0 = IMPLICIT_DEF
%1 = IMPLICIT_DEF
%3:vreg_128 = IMAGE_SAMPLE_LZ_V4_V2 %0, %1, undef %2:sgpr_128, 15, 0, 0, 0, 0, 0, 0, 0, implicit $exec :: (load (s128))
%4:vreg_128 = IMAGE_SAMPLE_LZ_V4_V2 %0, %1, undef %2:sgpr_128, 15, 0, 0, 0, 0, 0, 0, 0, implicit $exec :: (load (s128))
%5:vreg_128 = IMAGE_SAMPLE_LZ_V4_V2 %0, %1, undef %2:sgpr_128, 15, 0, 0, 0, 0, 0, 0, 0, implicit $exec :: (load (s128))
IMAGE_STORE_V4_V2 %3, %0, %1, 15, -1, 0, 0, 0, 0, 0, 0, implicit $exec :: (store (s128))
IMAGE_STORE_V4_V2 %4, %0, %1, 15, -1, 0, 0, 0, 0, 0, 0, implicit $exec :: (store (s128))
IMAGE_STORE_V4_V2 %5, %0, %1, 15, -1, 0, 0, 0, 0, 0, 0, implicit $exec :: (store (s128))
...
# GCN-LABEL: {{^}}name: mixed_clause{{$}}
# GCN: dead %3:vreg_128 = IMAGE_SAMPLE_LZ_V4_V2 %0, %1, %2, 15, 0, 0, 0, 0, 0, 0, 0, implicit $exec
# GCN-NEXT: dead %4:vreg_128 = GLOBAL_LOAD_DWORDX4 %0, 0, 0, implicit $exec
# GCN-NEXT: dead %5:vgpr_32 = BUFFER_LOAD_DWORD_OFFSET %2, 0, 0, 0, 0, implicit $exec
# GCN-NEXT: KILL %1{{$}}
# GCN-NEXT: KILL %2{{$}}
# GCN-NEXT: KILL %0{{$}}
---
name: mixed_clause
tracksRegLiveness: true
registers:
- { id: 0, class: vreg_64 }
- { id: 1, class: sgpr_256 }
- { id: 2, class: sgpr_128 }
- { id: 3, class: vreg_128 }
- { id: 4, class: vreg_128 }
- { id: 5, class: vgpr_32 }
body: |
bb.0:
%0 = IMPLICIT_DEF
%1 = IMPLICIT_DEF
%2 = IMPLICIT_DEF
%3:vreg_128 = IMAGE_SAMPLE_LZ_V4_V2 %0, %1, %2, 15, 0, 0, 0, 0, 0, 0, 0, implicit $exec :: (load (s128))
%4:vreg_128 = GLOBAL_LOAD_DWORDX4 %0, 0, 0, implicit $exec
%5:vgpr_32 = BUFFER_LOAD_DWORD_OFFSET %2, 0, 0, 0, 0, implicit $exec
...
# GCN-LABEL: {{^}}name: atomic{{$}}
# GCN: %1:vgpr_32 = IMPLICIT_DEF
# GCN-NEXT: dead %2:vgpr_32 = FLAT_ATOMIC_ADD_RTN %0, %1, 0, 2, implicit $exec, implicit $flat_scr
# GCN-NEXT: dead %3:vgpr_32 = FLAT_ATOMIC_ADD_RTN %0, %1, 0, 2, implicit $exec, implicit $flat_scr
# GCN-NEXT: FLAT_ATOMIC_ADD %0, %1, 0, 0, implicit $exec, implicit $flat_scr
# GCN-NEXT: FLAT_ATOMIC_ADD %0, %1, 0, 0, implicit $exec, implicit $flat_scr
# GCN-NEXT: S_ENDPGM 0
---
name: atomic
tracksRegLiveness: true
registers:
- { id: 0, class: vreg_64 }
- { id: 1, class: vgpr_32 }
- { id: 2, class: vgpr_32 }
- { id: 3, class: vgpr_32 }
body: |
bb.0:
%0 = IMPLICIT_DEF
%1 = IMPLICIT_DEF
%2:vgpr_32 = FLAT_ATOMIC_ADD_RTN %0, %1, 0, 2, implicit $exec, implicit $flat_scr
%3:vgpr_32 = FLAT_ATOMIC_ADD_RTN %0, %1, 0, 2, implicit $exec, implicit $flat_scr
FLAT_ATOMIC_ADD %0, %1, 0, 0, implicit $exec, implicit $flat_scr
FLAT_ATOMIC_ADD %0, %1, 0, 0, implicit $exec, implicit $flat_scr
S_ENDPGM 0
...
# One of the pointers has an additional use after the clause, but one
# doesn't. Only the final use should be killed.
# GCN-LABEL: {{^}}name: ptr_use_after_clause{{$}}
# GCN: dead %2:vreg_128 = GLOBAL_LOAD_DWORDX4 %0, 0, 0, implicit $exec
# GCN-NEXT: dead %3:vreg_128 = GLOBAL_LOAD_DWORDX4 %1, 16, 0, implicit $exec
# GCN-NEXT: dead %4:vreg_128 = GLOBAL_LOAD_DWORDX4 %0, 32, 0, implicit $exec
# GCN-NEXT: dead %5:vreg_128 = GLOBAL_LOAD_DWORDX4 %1, 48, 0, implicit $exec
# GCN-NEXT: KILL %1{{$}}
# GCN-NEXT: S_NOP 0, implicit %0
---
name: ptr_use_after_clause
tracksRegLiveness: true
body: |
bb.0:
liveins: $vgpr0_vgpr1, $vgpr2_vgpr3
%0:vreg_64 = COPY $vgpr0_vgpr1
%1:vreg_64 = COPY $vgpr2_vgpr3
%2:vreg_128 = GLOBAL_LOAD_DWORDX4 %0, 0, 0, implicit $exec
%3:vreg_128 = GLOBAL_LOAD_DWORDX4 %1, 16, 0, implicit $exec
%4:vreg_128 = GLOBAL_LOAD_DWORDX4 %0, 32, 0, implicit $exec
%5:vreg_128 = GLOBAL_LOAD_DWORDX4 %1, 48, 0, implicit $exec
S_NOP 0, implicit %0
...
# Only part of the register is really live past the clause.
# GCN-LABEL: {{^}}name: ptr_use_after_clause_subreg{{$}}
# GCN: dead %1:vreg_128 = GLOBAL_LOAD_DWORDX4 %0.sub0_sub1, 0, 0, implicit $exec
# GCN-NEXT: dead %2:vreg_128 = GLOBAL_LOAD_DWORDX4 %0.sub2_sub3, 16, 0, implicit $exec
# GCN-NEXT: dead %3:vreg_128 = GLOBAL_LOAD_DWORDX4 %0.sub0_sub1, 32, 0, implicit $exec
# GCN-NEXT: dead %4:vreg_128 = GLOBAL_LOAD_DWORDX4 %0.sub2_sub3, 48, 0, implicit $exec
# GCN-NEXT: KILL %0.sub2_sub3{{$}}
# GCN-NEXT: S_NOP 0, implicit %0.sub0_sub1{{$}}
---
name: ptr_use_after_clause_subreg
tracksRegLiveness: true
body: |
bb.0:
liveins: $vgpr0_vgpr1_vgpr2_vgpr3
%0:vreg_128 = COPY $vgpr0_vgpr1_vgpr2_vgpr3
%1:vreg_128 = GLOBAL_LOAD_DWORDX4 %0.sub0_sub1, 0, 0, implicit $exec
%2:vreg_128 = GLOBAL_LOAD_DWORDX4 %0.sub2_sub3, 16, 0, implicit $exec
%3:vreg_128 = GLOBAL_LOAD_DWORDX4 %0.sub0_sub1, 32, 0, implicit $exec
%4:vreg_128 = GLOBAL_LOAD_DWORDX4 %0.sub2_sub3, 48, 0, implicit $exec
S_NOP 0, implicit %0.sub0_sub1
...
# More complex situation where only some of the use subregisters live
# beyond the clause.
# GCN-LABEL: {{^}}name: ptr_use_after_clause_subreg_multi{{$}}
# GCN: dead %1:vreg_128 = GLOBAL_LOAD_DWORDX4 %0.sub0_sub1, 0, 0, implicit $exec
# GCN-NEXT: dead %2:vreg_128 = GLOBAL_LOAD_DWORDX4 %0.sub2_sub3, 16, 0, implicit $exec
# GCN-NEXT: dead %3:vreg_128 = GLOBAL_LOAD_DWORDX4 %0.sub3_sub4, 32, 0, implicit $exec
# GCN-NEXT: dead %4:vreg_128 = GLOBAL_LOAD_DWORDX4 %0.sub6_sub7, 48, 0, implicit $exec
# GCN-NEXT: KILL %0.sub2_sub3_sub4, %0.sub7{{$}}
# GCN-NEXT: S_NOP 0, implicit %0.sub0_sub1, implicit %0.sub5_sub6
---
name: ptr_use_after_clause_subreg_multi
tracksRegLiveness: true
body: |
bb.0:
liveins: $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7
%0:vreg_256 = COPY $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7
%1:vreg_128 = GLOBAL_LOAD_DWORDX4 %0.sub0_sub1, 0, 0, implicit $exec
%2:vreg_128 = GLOBAL_LOAD_DWORDX4 %0.sub2_sub3, 16, 0, implicit $exec
%3:vreg_128 = GLOBAL_LOAD_DWORDX4 %0.sub3_sub4, 32, 0, implicit $exec
%4:vreg_128 = GLOBAL_LOAD_DWORDX4 %0.sub6_sub7, 48, 0, implicit $exec
S_NOP 0, implicit %0.sub0_sub1, implicit %0.sub5_sub6
...
# Have subranges, but none of them are killed
# GCN-LABEL: {{^}}name: no_killed_subranges{{$}}
# GCN: dead %1:vreg_128 = GLOBAL_LOAD_DWORDX4 %0.sub0_sub1, 0, 0, implicit $exec
# GCN-NEXT: dead %2:vreg_128 = GLOBAL_LOAD_DWORDX4 %0.sub2_sub3, 16, 0, implicit $exec
# GCN-NEXT: dead %3:vreg_128 = GLOBAL_LOAD_DWORDX4 %0.sub0_sub1, 32, 0, implicit $exec
# GCN-NEXT: dead %4:vreg_128 = GLOBAL_LOAD_DWORDX4 %0.sub2_sub3, 48, 0, implicit $exec
# GCN-NEXT: S_NOP 0, implicit %0.sub0_sub1{{$}}
# GCN-NEXT: S_NOP 0, implicit %0.sub2_sub3{{$}}
---
name: no_killed_subranges
tracksRegLiveness: true
body: |
bb.0:
liveins: $vgpr0_vgpr1_vgpr2_vgpr3
%0:vreg_128 = COPY $vgpr0_vgpr1_vgpr2_vgpr3
%1:vreg_128 = GLOBAL_LOAD_DWORDX4 %0.sub0_sub1, 0, 0, implicit $exec
%2:vreg_128 = GLOBAL_LOAD_DWORDX4 %0.sub2_sub3, 16, 0, implicit $exec
%3:vreg_128 = GLOBAL_LOAD_DWORDX4 %0.sub0_sub1, 32, 0, implicit $exec
%4:vreg_128 = GLOBAL_LOAD_DWORDX4 %0.sub2_sub3, 48, 0, implicit $exec
S_NOP 0, implicit %0.sub0_sub1
S_NOP 0, implicit %0.sub2_sub3
...
# sub2 is undef at entry to the soft clause. It should not be have its
# live range extended.
# GCN-LABEL: name: no_killed_undef_subrange_use
# GCN: dead %1:vreg_128 = GLOBAL_LOAD_DWORDX4 %0.sub0_sub1, 0, 0, implicit $exec
# GCN-NEXT: dead %2:vreg_128 = GLOBAL_LOAD_DWORDX4 %0.sub0_sub1, 32, 0, implicit $exec
# GCN-NEXT: KILL %0.sub0_sub1{{$}}
# GCN-NEXT: %0.sub2:vreg_128 = V_MOV_B32_e32 0, implicit $exec
# GCN-NEXT: S_NOP 0, implicit %0.sub2
---
name: no_killed_undef_subrange_use
tracksRegLiveness: true
body: |
bb.0:
liveins: $vgpr0_vgpr1
undef %0.sub0_sub1:vreg_128 = COPY $vgpr0_vgpr1
%1:vreg_128 = GLOBAL_LOAD_DWORDX4 %0.sub0_sub1, 0, 0, implicit $exec
%3:vreg_128 = GLOBAL_LOAD_DWORDX4 %0.sub0_sub1, 32, 0, implicit $exec
%0.sub2:vreg_128 = V_MOV_B32_e32 0, implicit $exec
S_NOP 0, implicit %0.sub2
...
# Make sure intervening implicit_defs are not treated as breaking the
# clause
#
# GCN-LABEL: {{^}}name: implicit_def_no_break{{$}}
# GCN: %1:vreg_128 = GLOBAL_LOAD_DWORDX4 %0, 0, 0, implicit $exec
# GCN-NEXT: %2:vreg_128 = GLOBAL_LOAD_DWORDX4 %0, 16, 0, implicit $exec
# GCN-NEXT: %3:vreg_64 = IMPLICIT_DEF
# GCN-NEXT: %4:vreg_128 = GLOBAL_LOAD_DWORDX4 %3, 32, 0, implicit $exec
# GCN-NEXT: %5:vreg_128 = GLOBAL_LOAD_DWORDX4 %3, 48, 0, implicit $exec
# GCN-NEXT: KILL %3{{$}}
# GCN-NEXT: KILL %0{{$}}
---
name: implicit_def_no_break
tracksRegLiveness: true
body: |
bb.0:
liveins: $vgpr0_vgpr1
%0:vreg_64 = COPY $vgpr0_vgpr1
%1:vreg_128 = GLOBAL_LOAD_DWORDX4 %0, 0, 0, implicit $exec
%2:vreg_128 = GLOBAL_LOAD_DWORDX4 %0, 16, 0, implicit $exec
%3:vreg_64 = IMPLICIT_DEF
%4:vreg_128 = GLOBAL_LOAD_DWORDX4 %3, 32, 0, implicit $exec
%5:vreg_128 = GLOBAL_LOAD_DWORDX4 %3, 48, 0, implicit $exec
S_NOP 0, implicit %1, implicit %2, implicit %4, implicit %5
...
# GCN-LABEL: {{^}}name: kill_part_subreg{{$}}
# GCN: undef %1.sub0:vreg_128 = GLOBAL_LOAD_DWORD %0.sub0_sub1, 0, 0, implicit $exec
# GCN-NEXT: %1.sub1:vreg_128 = GLOBAL_LOAD_DWORD %0.sub1_sub2, 16, 0, implicit $exec
# GCN-NEXT: %1.sub3:vreg_128 = GLOBAL_LOAD_DWORD %0.sub2_sub3, 32, 0, implicit $exec
# GCN-NEXT: KILL %0.sub0_sub1_sub2, %0.sub3
---
name: kill_part_subreg
tracksRegLiveness: true
body: |
bb.0:
liveins: $vgpr0_vgpr1_vgpr2_vgpr3
%0:vreg_128 = COPY $vgpr0_vgpr1_vgpr2_vgpr3
undef %1.sub0:vreg_128 = GLOBAL_LOAD_DWORD %0.sub0_sub1, 0, 0, implicit $exec
%1.sub1:vreg_128 = GLOBAL_LOAD_DWORD %0.sub1_sub2, 16, 0, implicit $exec
%1.sub3:vreg_128 = GLOBAL_LOAD_DWORD %0.sub2_sub3, 32, 0, implicit $exec
S_NOP 0, implicit %1
...
# GCN-LABEL: {{^}}name: mem_clause_sreg256_used_stack{{$}}
# GCN: undef %0.sub7:sgpr_256 = S_LOAD_DWORD_IMM undef %1:sgpr_64(p4), 8, 0
# GCN-NEXT: dead %0.sub3:sgpr_256 = S_LOAD_DWORD_IMM undef %1:sgpr_64(p4), 24, 0
# GCN-NEXT: KILL undef %1
---
name: mem_clause_sreg256_used_stack
stack:
- { id: 0, type: default, offset: 0, size: 40, alignment: 8 }
machineFunctionInfo:
isEntryFunction: true
scratchRSrcReg: '$sgpr0_sgpr1_sgpr2_sgpr3'
stackPtrOffsetReg: '$sgpr32'
body: |
bb.0:
undef %0.sub7:sgpr_256 = S_LOAD_DWORD_IMM undef %1:sgpr_64(p4), 8, 0
%0.sub3:sgpr_256 = S_LOAD_DWORD_IMM undef %1:sgpr_64(p4), 24, 0
S_ENDPGM 0
...
Reference in New Issue View Git Blame Copy Permalink