caio/clang-p2996
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity
Files
1daf2994de49d1ecba4bee4e6842aa8a564cbc96
clang-p2996/llvm/test/CodeGen/AMDGPU/debug-value-scheduler-crash.mir
Nicolai Hähnle 10cef708a7 AMDGPU: Clean up LDS-related occupancy calculations 
Occupancy is expressed as waves per SIMD. This means that we need to
take into account the number of SIMDs per "CU" or, to be more precise,
the number of SIMDs over which a workgroup may be distributed.

getOccupancyWithLocalMemSize was wrong because it didn't take SIMDs
into account at all.

At the same time, we need to take into account that WGP mode offers
access to a larger total amount of LDS, since this can affect how
non-power-of-two LDS allocations are rounded. To make this work
consistently, we distinguish between (available) local memory size and
addressable local memory size (which is always limited by 64kB on
gfx10+, even with WGP mode).

This change results in a massive amount of test churn. A lot of it is
caused by the fact that the default work group size is 1024, which means
that (due to rounding effects) the default occupancy on older hardware
is 8 instead of 10, which affects scheduling via register pressure
estimates. I've adjusted most tests by just running the UTC tools, but
in some cases I manually changed the work group size to 32 or 64 to make
sure that work group size chunkiness has no effect.

Differential Revision: https://reviews.llvm.org/D139468
2023-01-23 21:43:06 +01:00

142 lines
7.0 KiB
YAML
Raw Blame History

# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
# RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 -run-pass=machine-scheduler -verify-machineinstrs %s -o - | FileCheck %s
--- |
declare void @llvm.dbg.value(metadata, metadata, metadata) #0
define amdgpu_kernel void @could_not_use_debug_inst_to_query_mi2mimap() #1 {
ret void
}
declare hidden float @foo(float, float, float) local_unnamed_addr #1
attributes #0 = { nounwind readnone speculatable }
attributes #1 = {nounwind }
...
---
name: could_not_use_debug_inst_to_query_mi2mimap
tracksRegLiveness: true
frameInfo:
hasCalls: true
body: |
; CHECK-LABEL: name: could_not_use_debug_inst_to_query_mi2mimap
; CHECK: bb.0:
; CHECK-NEXT: successors: %bb.1(0x80000000)
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: [[DEF:%[0-9]+]]:vreg_64 = IMPLICIT_DEF
; CHECK-NEXT: [[DEF1:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
; CHECK-NEXT: [[DEF2:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
; CHECK-NEXT: [[DEF3:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
; CHECK-NEXT: [[DEF4:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
; CHECK-NEXT: [[DEF5:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
; CHECK-NEXT: [[DEF6:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
; CHECK-NEXT: [[DEF7:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
; CHECK-NEXT: [[DEF8:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
; CHECK-NEXT: [[V_MUL_F32_e32_:%[0-9]+]]:vgpr_32 = nofpexcept V_MUL_F32_e32 1082130432, [[DEF1]], implicit $mode, implicit $exec
; CHECK-NEXT: [[DEF9:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
; CHECK-NEXT: [[DEF10:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.1:
; CHECK-NEXT: successors: %bb.2(0x80000000)
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: DBG_VALUE
; CHECK-NEXT: DBG_VALUE
; CHECK-NEXT: DBG_VALUE
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.2:
; CHECK-NEXT: successors: %bb.3(0x80000000)
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: S_BRANCH %bb.3
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.3:
; CHECK-NEXT: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
; CHECK-NEXT: [[DEF11:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
; CHECK-NEXT: [[DEF12:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
; CHECK-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY [[V_MOV_B32_e32_]]
; CHECK-NEXT: [[V_MUL_F32_e32_1:%[0-9]+]]:vgpr_32 = nofpexcept V_MUL_F32_e32 [[DEF7]], [[DEF7]], implicit $mode, implicit $exec
; CHECK-NEXT: [[V_MUL_F32_e32_2:%[0-9]+]]:vgpr_32 = nofpexcept V_MUL_F32_e32 [[DEF7]], [[DEF7]], implicit $mode, implicit $exec
; CHECK-NEXT: [[V_MUL_F32_e32_3:%[0-9]+]]:vgpr_32 = nofpexcept V_MUL_F32_e32 [[V_MOV_B32_e32_]], [[V_MOV_B32_e32_]], implicit $mode, implicit $exec
; CHECK-NEXT: [[V_MOV_B32_e32_1:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 1092616192, implicit $exec
; CHECK-NEXT: [[DEF13:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
; CHECK-NEXT: [[V_ADD_F32_e32_:%[0-9]+]]:vgpr_32 = nofpexcept V_ADD_F32_e32 [[V_MOV_B32_e32_]], [[V_MOV_B32_e32_]], implicit $mode, implicit $exec
; CHECK-NEXT: [[V_MUL_F32_e32_4:%[0-9]+]]:vgpr_32 = nofpexcept V_MUL_F32_e32 [[DEF7]], [[DEF7]], implicit $mode, implicit $exec
; CHECK-NEXT: dead %23:vgpr_32 = nofpexcept V_MUL_F32_e32 [[V_MUL_F32_e32_4]], [[DEF13]], implicit $mode, implicit $exec
; CHECK-NEXT: dead [[V_MOV_B32_e32_1]]:vgpr_32 = nofpexcept V_MAC_F32_e32 [[V_ADD_F32_e32_]], [[COPY]], [[V_MOV_B32_e32_1]], implicit $mode, implicit $exec
; CHECK-NEXT: [[DEF14:%[0-9]+]]:sreg_64 = IMPLICIT_DEF
; CHECK-NEXT: $sgpr4 = IMPLICIT_DEF
; CHECK-NEXT: $vgpr0 = COPY [[DEF11]]
; CHECK-NEXT: $vgpr0 = COPY [[V_MOV_B32_e32_]]
; CHECK-NEXT: $vgpr1 = COPY [[DEF7]]
; CHECK-NEXT: $vgpr0 = COPY [[V_MUL_F32_e32_1]]
; CHECK-NEXT: $vgpr1 = COPY [[V_MUL_F32_e32_2]]
; CHECK-NEXT: $vgpr2 = COPY [[V_MUL_F32_e32_3]]
; CHECK-NEXT: dead $sgpr30_sgpr31 = SI_CALL [[DEF14]], @foo, csr_amdgpu, implicit undef $sgpr0_sgpr1_sgpr2_sgpr3, implicit killed $sgpr4, implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit-def $vgpr0
; CHECK-NEXT: [[V_ADD_F32_e32_1:%[0-9]+]]:vgpr_32 = nofpexcept V_ADD_F32_e32 [[V_MUL_F32_e32_]], [[DEF8]], implicit $mode, implicit $exec
; CHECK-NEXT: [[V_MAC_F32_e32_:%[0-9]+]]:vgpr_32 = nofpexcept V_MAC_F32_e32 [[DEF12]], [[DEF9]], [[V_MAC_F32_e32_]], implicit $mode, implicit $exec
; CHECK-NEXT: dead %26:vgpr_32 = nofpexcept V_MAD_F32_e64 0, [[V_MAC_F32_e32_]], 0, [[DEF4]], 0, [[DEF1]], 0, 0, implicit $mode, implicit $exec
; CHECK-NEXT: dead %27:vgpr_32 = nofpexcept V_MAD_F32_e64 0, [[V_MAC_F32_e32_]], 0, [[DEF5]], 0, [[DEF2]], 0, 0, implicit $mode, implicit $exec
; CHECK-NEXT: dead %28:vgpr_32 = nofpexcept V_MAD_F32_e64 0, [[V_MAC_F32_e32_]], 0, [[DEF6]], 0, [[DEF3]], 0, 0, implicit $mode, implicit $exec
; CHECK-NEXT: GLOBAL_STORE_DWORD [[DEF]], [[DEF10]], 0, 0, implicit $exec
; CHECK-NEXT: S_ENDPGM 0
bb.0:
successors: %bb.1
%0:vreg_64 = IMPLICIT_DEF
%1:vgpr_32 = IMPLICIT_DEF
%2:vgpr_32 = IMPLICIT_DEF
%3:vgpr_32 = IMPLICIT_DEF
%4:vgpr_32 = IMPLICIT_DEF
%5:vgpr_32 = IMPLICIT_DEF
%6:vgpr_32 = IMPLICIT_DEF
%7:vgpr_32 = IMPLICIT_DEF
%8:vgpr_32 = IMPLICIT_DEF
%9:vgpr_32 = nofpexcept V_MUL_F32_e32 1082130432, %1, implicit $mode, implicit $exec
%10:vgpr_32 = IMPLICIT_DEF
%11:vgpr_32 = IMPLICIT_DEF
bb.1:
successors: %bb.2
DBG_VALUE
DBG_VALUE
DBG_VALUE
bb.2:
successors: %bb.3
S_BRANCH %bb.3
bb.3:
%12:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
%13:vgpr_32 = COPY %12
%14:vgpr_32 = IMPLICIT_DEF
%15:vgpr_32 = IMPLICIT_DEF
%16:vgpr_32 = nofpexcept V_MUL_F32_e32 %7, %7, implicit $mode, implicit $exec
%17:vgpr_32 = nofpexcept V_MUL_F32_e32 %7, %7, implicit $mode, implicit $exec
%18:vgpr_32 = nofpexcept V_MUL_F32_e32 %12, %12, implicit $mode, implicit $exec
%19:vgpr_32 = V_MOV_B32_e32 1092616192, implicit $exec
%20:vgpr_32 = IMPLICIT_DEF
%21:vgpr_32 = nofpexcept V_ADD_F32_e32 %12, %12, implicit $mode, implicit $exec
%22:vgpr_32 = nofpexcept V_MUL_F32_e32 %7, %7, implicit $mode, implicit $exec
%23:vgpr_32 = nofpexcept V_MUL_F32_e32 %22, %20, implicit $mode, implicit $exec
%19:vgpr_32 = nofpexcept V_MAC_F32_e32 %21, %13, %19, implicit $mode, implicit $exec
%24:sreg_64 = IMPLICIT_DEF
$vgpr0 = COPY %14
$vgpr0 = COPY %12
$vgpr1 = COPY %7
$vgpr0 = COPY %16
$vgpr1 = COPY %17
$vgpr2 = COPY %18
$sgpr4 = IMPLICIT_DEF
dead $sgpr30_sgpr31 = SI_CALL %24, @foo, csr_amdgpu, implicit undef $sgpr0_sgpr1_sgpr2_sgpr3, implicit killed $sgpr4, implicit killed $vgpr0, implicit killed $vgpr1, implicit killed $vgpr2, implicit-def $vgpr0
%25:vgpr_32 = nofpexcept V_ADD_F32_e32 %9, %8, implicit $mode, implicit $exec
%25:vgpr_32 = nofpexcept V_MAC_F32_e32 %15, %10, %25, implicit $mode, implicit $exec
%26:vgpr_32 = nofpexcept V_MAD_F32_e64 0, %25, 0, %4, 0, %1, 0, 0, implicit $mode, implicit $exec
%27:vgpr_32 = nofpexcept V_MAD_F32_e64 0, %25, 0, %5, 0, %2, 0, 0, implicit $mode, implicit $exec
%28:vgpr_32 = nofpexcept V_MAD_F32_e64 0, %25, 0, %6, 0, %3, 0, 0, implicit $mode, implicit $exec
GLOBAL_STORE_DWORD %0, %11, 0, 0, implicit $exec
S_ENDPGM 0
...
Reference in New Issue View Git Blame Copy Permalink