This is an archive of the discontinued LLVM Phabricator instance.

Differential D92839

[AMDGPU] Add occupancy level tests for GFX10.3. NFC.
ClosedPublic

Authored by foad on Dec 8 2020, 5:18 AM.

Details

Reviewers
arsenm
rampitec
Commits
rG03663e4130d7: [AMDGPU] Add occupancy level tests for GFX10.3. NFC.
Summary

getMaxWavesPerEU and getVGPRAllocGranule both changed in GFX10.3 and
they both affect the occupancy calculation.

Diff Detail

Event Timeline

foad created this revision.Dec 8 2020, 5:18 AM
Herald added subscribers: kerbowa, t-tye, tpr and 5 others. · View Herald TranscriptDec 8 2020, 5:18 AM
foad requested review of this revision.Dec 8 2020, 5:18 AM
Herald added a project: Restricted Project. · View Herald TranscriptDec 8 2020, 5:18 AM
Herald added subscribers: llvm-commits, wdng. · View Herald Transcript
arsenm accepted this revision.Dec 8 2020, 5:46 AM
This revision is now accepted and ready to land.Dec 8 2020, 5:46 AM
Harbormaster completed remote builds in B81427: Diff 310147.Dec 8 2020, 6:03 AM
This revision was landed with ongoing or failed builds.Dec 8 2020, 6:27 AM
Closed by commit rG03663e4130d7: [AMDGPU] Add occupancy level tests for GFX10.3. NFC. (authored by foad). · Explain Why
This revision was automatically updated to reflect the committed changes.
foad added a commit: rG03663e4130d7: [AMDGPU] Add occupancy level tests for GFX10.3. NFC..

Revision Contents

PathSize
llvm/
test/
CodeGen/
AMDGPU/
122 lines

Diff 310172

llvm/test/CodeGen/AMDGPU/occupancy-levels.ll

; RUN: llc -march=amdgcn -mcpu=gfx900 < %s | FileCheck --check-prefixes=GCN,GFX9 %s ; RUN: llc -march=amdgcn -mcpu=gfx900 < %s | FileCheck --check-prefixes=GCN,GFX9 %s
; RUN: llc -march=amdgcn -mcpu=gfx1010 < %s | FileCheck --check-prefixes=GCN,GFX1010,GFX1010W32 %s ; RUN: llc -march=amdgcn -mcpu=gfx1010 < %s | FileCheck --check-prefixes=GCN,GFX10,GFX10W32,GFX1010,GFX1010W32 %s
; RUN: llc -march=amdgcn -mcpu=gfx1010 -mattr=+wavefrontsize64 < %s | FileCheck --check-prefixes=GCN,GFX1010,GFX1010W64 %s ; RUN: llc -march=amdgcn -mcpu=gfx1010 -mattr=+wavefrontsize64 < %s | FileCheck --check-prefixes=GCN,GFX10,GFX10W64,GFX1010,GFX1010W64 %s
; RUN: llc -march=amdgcn -mcpu=gfx1030 < %s | FileCheck --check-prefixes=GCN,GFX10,GFX10W32,GFX1030,GFX1030W32 %s
; RUN: llc -march=amdgcn -mcpu=gfx1030 -mattr=+wavefrontsize64 < %s | FileCheck --check-prefixes=GCN,GFX10,GFX10W64,GFX1030,GFX1030W64 %s
; GCN-LABEL: {{^}}max_occupancy: ; GCN-LABEL: {{^}}max_occupancy:
; GFX9: ; Occupancy: 10 ; GFX9: ; Occupancy: 10
; GFX1010: ; Occupancy: 20 ; GFX1010: ; Occupancy: 20
; GFX1030: ; Occupancy: 16
define amdgpu_kernel void @max_occupancy() { define amdgpu_kernel void @max_occupancy() {
ret void ret void
} }
; GCN-LABEL: {{^}}limited_occupancy_3: ; GCN-LABEL: {{^}}limited_occupancy_3:
; GFX9: ; Occupancy: 3 ; GFX9: ; Occupancy: 3
; GFX1010W64: ; Occupancy: 3 ; GFX10W64: ; Occupancy: 3
; GFX1010W32: ; Occupancy: 4 ; GFX10W32: ; Occupancy: 4
define amdgpu_kernel void @limited_occupancy_3() #0 { define amdgpu_kernel void @limited_occupancy_3() #0 {
ret void ret void
} }
; GCN-LABEL: {{^}}limited_occupancy_18: ; GCN-LABEL: {{^}}limited_occupancy_18:
; GFX9: ; Occupancy: 10 ; GFX9: ; Occupancy: 10
; GFX1010: ; Occupancy: 18 ; GFX1010: ; Occupancy: 18
; GFX1030: ; Occupancy: 16
define amdgpu_kernel void @limited_occupancy_18() #1 { define amdgpu_kernel void @limited_occupancy_18() #1 {
ret void ret void
} }
; GCN-LABEL: {{^}}limited_occupancy_19: ; GCN-LABEL: {{^}}limited_occupancy_19:
; GFX9: ; Occupancy: 10 ; GFX9: ; Occupancy: 10
; GFX1010: ; Occupancy: 18 ; GFX1010: ; Occupancy: 18
; GFX1030: ; Occupancy: 16
define amdgpu_kernel void @limited_occupancy_19() #2 { define amdgpu_kernel void @limited_occupancy_19() #2 {
ret void ret void
} }
; GCN-LABEL: {{^}}used_24_vgprs: ; GCN-LABEL: {{^}}used_24_vgprs:
; GFX9: ; Occupancy: 10 ; GFX9: ; Occupancy: 10
; GFX1010: ; Occupancy: 20 ; GFX1010: ; Occupancy: 20
; GFX1030: ; Occupancy: 16
define amdgpu_kernel void @used_24_vgprs() { define amdgpu_kernel void @used_24_vgprs() {
call void asm sideeffect "", "~{v23}" () call void asm sideeffect "", "~{v23}" ()
ret void ret void
} }
; GCN-LABEL: {{^}}used_28_vgprs: ; GCN-LABEL: {{^}}used_28_vgprs:
; GFX9: ; Occupancy: 9 ; GFX9: ; Occupancy: 9
; GFX1010W64: ; Occupancy: 18 ; GFX1010W64: ; Occupancy: 18
; GFX1010W32: ; Occupancy: 20 ; GFX1010W32: ; Occupancy: 20
; GFX1030: ; Occupancy: 16
define amdgpu_kernel void @used_28_vgprs() { define amdgpu_kernel void @used_28_vgprs() {
call void asm sideeffect "", "~{v27}" () call void asm sideeffect "", "~{v27}" ()
ret void ret void
} }
; GCN-LABEL: {{^}}used_32_vgprs: ; GCN-LABEL: {{^}}used_32_vgprs:
; GFX9: ; Occupancy: 8 ; GFX9: ; Occupancy: 8
; GFX1010W64: ; Occupancy: 16 ; GFX10W64: ; Occupancy: 16
; GFX1010W32: ; Occupancy: 20 ; GFX1010W32: ; Occupancy: 20
; GFX1030W32: ; Occupancy: 16
define amdgpu_kernel void @used_32_vgprs() { define amdgpu_kernel void @used_32_vgprs() {
call void asm sideeffect "", "~{v31}" () call void asm sideeffect "", "~{v31}" ()
ret void ret void
} }
; GCN-LABEL: {{^}}used_36_vgprs: ; GCN-LABEL: {{^}}used_36_vgprs:
; GFX9: ; Occupancy: 7 ; GFX9: ; Occupancy: 7
; GFX1010W64: ; Occupancy: 14 ; GFX1010W64: ; Occupancy: 14
; GFX1010W32: ; Occupancy: 20 ; GFX1010W32: ; Occupancy: 20
; GFX1030W64: ; Occupancy: 12
; GFX1030W32: ; Occupancy: 16
define amdgpu_kernel void @used_36_vgprs() { define amdgpu_kernel void @used_36_vgprs() {
call void asm sideeffect "", "~{v35}" () call void asm sideeffect "", "~{v35}" ()
ret void ret void
} }
; GCN-LABEL: {{^}}used_40_vgprs: ; GCN-LABEL: {{^}}used_40_vgprs:
; GFX9: ; Occupancy: 6 ; GFX9: ; Occupancy: 6
; GFX1010W64: ; Occupancy: 12 ; GFX10W64: ; Occupancy: 12
; GFX1010W32: ; Occupancy: 20 ; GFX1010W32: ; Occupancy: 20
; GFX1030W32: ; Occupancy: 16
define amdgpu_kernel void @used_40_vgprs() { define amdgpu_kernel void @used_40_vgprs() {
call void asm sideeffect "", "~{v39}" () call void asm sideeffect "", "~{v39}" ()
ret void ret void
} }
; GCN-LABEL: {{^}}used_44_vgprs: ; GCN-LABEL: {{^}}used_44_vgprs:
; GFX9: ; Occupancy: 5 ; GFX9: ; Occupancy: 5
; GFX1010W64: ; Occupancy: 11 ; GFX1010W64: ; Occupancy: 11
; GFX1010W32: ; Occupancy: 20 ; GFX1010W32: ; Occupancy: 20
; GFX1030W64: ; Occupancy: 10
; GFX1030W32: ; Occupancy: 16
define amdgpu_kernel void @used_44_vgprs() { define amdgpu_kernel void @used_44_vgprs() {
call void asm sideeffect "", "~{v43}" () call void asm sideeffect "", "~{v43}" ()
ret void ret void
} }
; GCN-LABEL: {{^}}used_48_vgprs: ; GCN-LABEL: {{^}}used_48_vgprs:
; GFX9: ; Occupancy: 5 ; GFX9: ; Occupancy: 5
; GFX1010W64: ; Occupancy: 10 ; GFX10W64: ; Occupancy: 10
; GFX1010W32: ; Occupancy: 20 ; GFX1010W32: ; Occupancy: 20
; GFX1030W32: ; Occupancy: 16
define amdgpu_kernel void @used_48_vgprs() { define amdgpu_kernel void @used_48_vgprs() {
call void asm sideeffect "", "~{v47}" () call void asm sideeffect "", "~{v47}" ()
ret void ret void
} }
; GCN-LABEL: {{^}}used_56_vgprs: ; GCN-LABEL: {{^}}used_56_vgprs:
; GFX9: ; Occupancy: 4 ; GFX9: ; Occupancy: 4
; GFX1010W64: ; Occupancy: 9 ; GFX10W64: ; Occupancy: 9
; GFX1010W32: ; Occupancy: 18 ; GFX1010W32: ; Occupancy: 18
; GFX1030W32: ; Occupancy: 16
define amdgpu_kernel void @used_56_vgprs() { define amdgpu_kernel void @used_56_vgprs() {
call void asm sideeffect "", "~{v55}" () call void asm sideeffect "", "~{v55}" ()
ret void ret void
} }
; GCN-LABEL: {{^}}used_64_vgprs: ; GCN-LABEL: {{^}}used_64_vgprs:
; GFX9: ; Occupancy: 4 ; GFX9: ; Occupancy: 4
; GFX1010W64: ; Occupancy: 8 ; GFX10W64: ; Occupancy: 8
; GFX1010W32: ; Occupancy: 16 ; GFX10W32: ; Occupancy: 16
define amdgpu_kernel void @used_64_vgprs() { define amdgpu_kernel void @used_64_vgprs() {
call void asm sideeffect "", "~{v63}" () call void asm sideeffect "", "~{v63}" ()
ret void ret void
} }
; GCN-LABEL: {{^}}used_72_vgprs: ; GCN-LABEL: {{^}}used_72_vgprs:
; GFX9: ; Occupancy: 3 ; GFX9: ; Occupancy: 3
; GFX1010W64: ; Occupancy: 7 ; GFX10W64: ; Occupancy: 7
; GFX1010W32: ; Occupancy: 14 ; GFX1010W32: ; Occupancy: 14
; GFX1030W32: ; Occupancy: 12
define amdgpu_kernel void @used_72_vgprs() { define amdgpu_kernel void @used_72_vgprs() {
call void asm sideeffect "", "~{v71}" () call void asm sideeffect "", "~{v71}" ()
ret void ret void
} }
; GCN-LABEL: {{^}}used_80_vgprs: ; GCN-LABEL: {{^}}used_80_vgprs:
; GFX9: ; Occupancy: 3 ; GFX9: ; Occupancy: 3
; GFX1010W64: ; Occupancy: 6 ; GFX10W64: ; Occupancy: 6
; GFX1010W32: ; Occupancy: 12 ; GFX10W32: ; Occupancy: 12
define amdgpu_kernel void @used_80_vgprs() { define amdgpu_kernel void @used_80_vgprs() {
call void asm sideeffect "", "~{v79}" () call void asm sideeffect "", "~{v79}" ()
ret void ret void
} }
; GCN-LABEL: {{^}}used_84_vgprs: ; GCN-LABEL: {{^}}used_84_vgprs:
; GFX9: ; Occupancy: 3 ; GFX9: ; Occupancy: 3
; GFX1010W64: ; Occupancy: 6 ; GFX1010W64: ; Occupancy: 6
; GFX1010W32: ; Occupancy: 11 ; GFX1010W32: ; Occupancy: 11
; GFX1030W64: ; Occupancy: 5
; GFX1030W32: ; Occupancy: 10
define amdgpu_kernel void @used_84_vgprs() { define amdgpu_kernel void @used_84_vgprs() {
call void asm sideeffect "", "~{v83}" () call void asm sideeffect "", "~{v83}" ()
ret void ret void
} }
; GCN-LABEL: {{^}}used_88_vgprs: ; GCN-LABEL: {{^}}used_88_vgprs:
; GFX9: ; Occupancy: 2 ; GFX9: ; Occupancy: 2
; GFX1010W64: ; Occupancy: 5 ; GFX10W64: ; Occupancy: 5
; GFX1010W32: ; Occupancy: 11 ; GFX1010W32: ; Occupancy: 11
; GFX1030W32: ; Occupancy: 10
define amdgpu_kernel void @used_88_vgprs() { define amdgpu_kernel void @used_88_vgprs() {
call void asm sideeffect "", "~{v87}" () call void asm sideeffect "", "~{v87}" ()
ret void ret void
} }
; GCN-LABEL: {{^}}used_96_vgprs: ; GCN-LABEL: {{^}}used_96_vgprs:
; GFX9: ; Occupancy: 2 ; GFX9: ; Occupancy: 2
; GFX1010W64: ; Occupancy: 5 ; GFX10W64: ; Occupancy: 5
; GFX1010W32: ; Occupancy: 10 ; GFX10W32: ; Occupancy: 10
define amdgpu_kernel void @used_96_vgprs() { define amdgpu_kernel void @used_96_vgprs() {
call void asm sideeffect "", "~{v95}" () call void asm sideeffect "", "~{v95}" ()
ret void ret void
} }
; GCN-LABEL: {{^}}used_100_vgprs: ; GCN-LABEL: {{^}}used_100_vgprs:
; GFX9: ; Occupancy: 2 ; GFX9: ; Occupancy: 2
; GFX1010W64: ; Occupancy: 5 ; GFX1010W64: ; Occupancy: 5
; GFX1010W32: ; Occupancy: 9 ; GFX1030W64: ; Occupancy: 4
; GFX10W32: ; Occupancy: 9
define amdgpu_kernel void @used_100_vgprs() { define amdgpu_kernel void @used_100_vgprs() {
call void asm sideeffect "", "~{v99}" () call void asm sideeffect "", "~{v99}" ()
ret void ret void
} }
; GCN-LABEL: {{^}}used_112_vgprs: ; GCN-LABEL: {{^}}used_112_vgprs:
; GFX9: ; Occupancy: 2 ; GFX9: ; Occupancy: 2
; GFX1010W64: ; Occupancy: 4 ; GFX10W64: ; Occupancy: 4
; GFX1010W32: ; Occupancy: 9 ; GFX10W32: ; Occupancy: 9
define amdgpu_kernel void @used_112_vgprs() { define amdgpu_kernel void @used_112_vgprs() {
call void asm sideeffect "", "~{v111}" () call void asm sideeffect "", "~{v111}" ()
ret void ret void
} }
; GCN-LABEL: {{^}}used_128_vgprs: ; GCN-LABEL: {{^}}used_128_vgprs:
; GFX9: ; Occupancy: 2 ; GFX9: ; Occupancy: 2
; GFX1010W64: ; Occupancy: 4 ; GFX10W64: ; Occupancy: 4
; GFX1010W32: ; Occupancy: 8 ; GFX10W32: ; Occupancy: 8
define amdgpu_kernel void @used_128_vgprs() { define amdgpu_kernel void @used_128_vgprs() {
call void asm sideeffect "", "~{v127}" () call void asm sideeffect "", "~{v127}" ()
ret void ret void
} }
; GCN-LABEL: {{^}}used_144_vgprs: ; GCN-LABEL: {{^}}used_144_vgprs:
; GFX9: ; Occupancy: 1 ; GFX9: ; Occupancy: 1
; GFX1010W64: ; Occupancy: 3 ; GFX10W64: ; Occupancy: 3
; GFX1010W32: ; Occupancy: 7 ; GFX10W32: ; Occupancy: 7
define amdgpu_kernel void @used_144_vgprs() { define amdgpu_kernel void @used_144_vgprs() {
call void asm sideeffect "", "~{v143}" () call void asm sideeffect "", "~{v143}" ()
ret void ret void
} }
; GCN-LABEL: {{^}}used_168_vgprs: ; GCN-LABEL: {{^}}used_168_vgprs:
; GFX9: ; Occupancy: 1 ; GFX9: ; Occupancy: 1
; GFX1010W64: ; Occupancy: 3 ; GFX10W64: ; Occupancy: 3
; GFX1010W32: ; Occupancy: 6 ; GFX1010W32: ; Occupancy: 6
; GFX1030W32: ; Occupancy: 5
define amdgpu_kernel void @used_168_vgprs() { define amdgpu_kernel void @used_168_vgprs() {
call void asm sideeffect "", "~{v167}" () call void asm sideeffect "", "~{v167}" ()
ret void ret void
} }
; GCN-LABEL: {{^}}used_200_vgprs: ; GCN-LABEL: {{^}}used_200_vgprs:
; GFX9: ; Occupancy: 1 ; GFX9: ; Occupancy: 1
; GFX1010W64: ; Occupancy: 2 ; GFX10W64: ; Occupancy: 2
; GFX1010W32: ; Occupancy: 5 ; GFX1010W32: ; Occupancy: 5
; GFX1030W32: ; Occupancy: 4
define amdgpu_kernel void @used_200_vgprs() { define amdgpu_kernel void @used_200_vgprs() {
call void asm sideeffect "", "~{v199}" () call void asm sideeffect "", "~{v199}" ()
ret void ret void
} }
; GCN-LABEL: {{^}}used_256_vgprs: ; GCN-LABEL: {{^}}used_256_vgprs:
; GFX9: ; Occupancy: 1 ; GFX9: ; Occupancy: 1
; GFX1010W64: ; Occupancy: 2 ; GFX10W64: ; Occupancy: 2
; GFX1010W32: ; Occupancy: 4 ; GFX10W32: ; Occupancy: 4
define amdgpu_kernel void @used_256_vgprs() { define amdgpu_kernel void @used_256_vgprs() {
call void asm sideeffect "", "~{v255}" () call void asm sideeffect "", "~{v255}" ()
ret void ret void
} }
; GCN-LABEL: {{^}}used_80_sgprs: ; GCN-LABEL: {{^}}used_80_sgprs:
; GFX9: ; Occupancy: 10 ; GFX9: ; Occupancy: 10
; GFX1010: ; Occupancy: 20 ; GFX1010: ; Occupancy: 20
; GFX1030: ; Occupancy: 16
define amdgpu_kernel void @used_80_sgprs() { define amdgpu_kernel void @used_80_sgprs() {
call void asm sideeffect "", "~{s79}" () call void asm sideeffect "", "~{s79}" ()
ret void ret void
} }
; GCN-LABEL: {{^}}used_88_sgprs: ; GCN-LABEL: {{^}}used_88_sgprs:
; GFX9: ; Occupancy: 9 ; GFX9: ; Occupancy: 9
; GFX1010: ; Occupancy: 20 ; GFX1010: ; Occupancy: 20
; GFX1030: ; Occupancy: 16
define amdgpu_kernel void @used_88_sgprs() { define amdgpu_kernel void @used_88_sgprs() {
call void asm sideeffect "", "~{s87}" () call void asm sideeffect "", "~{s87}" ()
ret void ret void
} }
; GCN-LABEL: {{^}}used_100_sgprs: ; GCN-LABEL: {{^}}used_100_sgprs:
; GFX9: ; Occupancy: 8 ; GFX9: ; Occupancy: 8
; GFX1010: ; Occupancy: 20 ; GFX1010: ; Occupancy: 20
; GFX1030: ; Occupancy: 16
define amdgpu_kernel void @used_100_sgprs() { define amdgpu_kernel void @used_100_sgprs() {
call void asm sideeffect "", "~{s99}" () call void asm sideeffect "", "~{s99}" ()
ret void ret void
} }
; GCN-LABEL: {{^}}used_101_sgprs: ; GCN-LABEL: {{^}}used_101_sgprs:
; GFX9: ; Occupancy: 7 ; GFX9: ; Occupancy: 7
; GFX1010: ; Occupancy: 20 ; GFX1010: ; Occupancy: 20
; GFX1030: ; Occupancy: 16
define amdgpu_kernel void @used_101_sgprs() { define amdgpu_kernel void @used_101_sgprs() {
call void asm sideeffect "", "~{s100}" () call void asm sideeffect "", "~{s100}" ()
ret void ret void
} }
; GCN-LABEL: {{^}}used_lds_6552: ; GCN-LABEL: {{^}}used_lds_6552:
; GFX9: ; Occupancy: 10 ; GFX9: ; Occupancy: 10
; GFX1010: ; Occupancy: 20 ; GFX1010: ; Occupancy: 20
; GFX1030: ; Occupancy: 16
@lds6552 = internal addrspace(3) global [6552 x i8] undef, align 4 @lds6552 = internal addrspace(3) global [6552 x i8] undef, align 4
define amdgpu_kernel void @used_lds_6552() { define amdgpu_kernel void @used_lds_6552() {
%p = bitcast [6552 x i8] addrspace(3)* @lds6552 to i8 addrspace(3)* %p = bitcast [6552 x i8] addrspace(3)* @lds6552 to i8 addrspace(3)*
store volatile i8 1, i8 addrspace(3)* %p store volatile i8 1, i8 addrspace(3)* %p
ret void ret void
} }
; GCN-LABEL: {{^}}used_lds_6556: ; GCN-LABEL: {{^}}used_lds_6556:
; GFX9: ; Occupancy: 10 ; GFX9: ; Occupancy: 10
; GFX1010W64: ; Occupancy: 20 ; GFX1010: ; Occupancy: 20
; GFX1010W32: ; Occupancy: 20 ; GFX1030: ; Occupancy: 16
@lds6556 = internal addrspace(3) global [6556 x i8] undef, align 4 @lds6556 = internal addrspace(3) global [6556 x i8] undef, align 4
define amdgpu_kernel void @used_lds_6556() { define amdgpu_kernel void @used_lds_6556() {
%p = bitcast [6556 x i8] addrspace(3)* @lds6556 to i8 addrspace(3)* %p = bitcast [6556 x i8] addrspace(3)* @lds6556 to i8 addrspace(3)*
store volatile i8 1, i8 addrspace(3)* %p store volatile i8 1, i8 addrspace(3)* %p
ret void ret void
} }
; GCN-LABEL: {{^}}used_lds_13112: ; GCN-LABEL: {{^}}used_lds_13112:
; GFX9: ; Occupancy: 10 ; GFX9: ; Occupancy: 10
; GFX1010W64: ; Occupancy: 20 ; GFX1010: ; Occupancy: 20
; GFX1010W32: ; Occupancy: 20 ; GFX1030: ; Occupancy: 16
@lds13112 = internal addrspace(3) global [13112 x i8] undef, align 4 @lds13112 = internal addrspace(3) global [13112 x i8] undef, align 4
define amdgpu_kernel void @used_lds_13112() { define amdgpu_kernel void @used_lds_13112() {
%p = bitcast [13112 x i8] addrspace(3)* @lds13112 to i8 addrspace(3)* %p = bitcast [13112 x i8] addrspace(3)* @lds13112 to i8 addrspace(3)*
store volatile i8 1, i8 addrspace(3)* %p store volatile i8 1, i8 addrspace(3)* %p
ret void ret void
} }
; GCN-LABEL: {{^}}used_lds_8252_max_group_size_64: ; GCN-LABEL: {{^}}used_lds_8252_max_group_size_64:
; GFX9: ; Occupancy: 7{{$}} ; GFX9: ; Occupancy: 7{{$}}
; GFX101064: ; Occupancy: 7{{$}} ; GFX10W64: ; Occupancy: 7{{$}}
; GFX1010W32: ; Occupancy: 14{{$}} ; GFX10W32: ; Occupancy: 14{{$}}
@lds8252 = internal addrspace(3) global [8252 x i8] undef, align 4 @lds8252 = internal addrspace(3) global [8252 x i8] undef, align 4
define amdgpu_kernel void @used_lds_8252_max_group_size_64() #3 { define amdgpu_kernel void @used_lds_8252_max_group_size_64() #3 {
%p = bitcast [8252 x i8] addrspace(3)* @lds8252 to i8 addrspace(3)* %p = bitcast [8252 x i8] addrspace(3)* @lds8252 to i8 addrspace(3)*
store volatile i8 1, i8 addrspace(3)* %p store volatile i8 1, i8 addrspace(3)* %p
ret void ret void
} }
; GCN-LABEL: {{^}}used_lds_8252_max_group_size_96: ; GCN-LABEL: {{^}}used_lds_8252_max_group_size_96:
; GFX9: ; Occupancy: 10{{$}} ; GFX9: ; Occupancy: 10{{$}}
; GFX1010W64: ; Occupancy: 14{{$}} ; GFX10W64: ; Occupancy: 14{{$}}
; GFX1010W32: ; Occupancy: 20{{$}} ; GFX1010W32: ; Occupancy: 20{{$}}
; GFX1030W32: ; Occupancy: 16{{$}}
define amdgpu_kernel void @used_lds_8252_max_group_size_96() #4 { define amdgpu_kernel void @used_lds_8252_max_group_size_96() #4 {
%p = bitcast [8252 x i8] addrspace(3)* @lds8252 to i8 addrspace(3)* %p = bitcast [8252 x i8] addrspace(3)* @lds8252 to i8 addrspace(3)*
store volatile i8 1, i8 addrspace(3)* %p store volatile i8 1, i8 addrspace(3)* %p
ret void ret void
} }
; GCN-LABEL: {{^}}used_lds_8252_max_group_size_128: ; GCN-LABEL: {{^}}used_lds_8252_max_group_size_128:
; GFX9: ; Occupancy: 10{{$}} ; GFX9: ; Occupancy: 10{{$}}
; GFX1010W64: ; Occupancy: 14{{$}} ; GFX10W64: ; Occupancy: 14{{$}}
; GFX1010W32: ; Occupancy: 20{{$}} ; GFX1010W32: ; Occupancy: 20{{$}}
; GFX1030W32: ; Occupancy: 16{{$}}
define amdgpu_kernel void @used_lds_8252_max_group_size_128() #5 { define amdgpu_kernel void @used_lds_8252_max_group_size_128() #5 {
%p = bitcast [8252 x i8] addrspace(3)* @lds8252 to i8 addrspace(3)* %p = bitcast [8252 x i8] addrspace(3)* @lds8252 to i8 addrspace(3)*
store volatile i8 1, i8 addrspace(3)* %p store volatile i8 1, i8 addrspace(3)* %p
ret void ret void
} }
; GCN-LABEL: {{^}}used_lds_8252_max_group_size_192: ; GCN-LABEL: {{^}}used_lds_8252_max_group_size_192:
; GFX9: ; Occupancy: 10{{$}} ; GFX9: ; Occupancy: 10{{$}}
; GFX1010W64: ; Occupancy: 20{{$}} ; GFX1010: ; Occupancy: 20{{$}}
; GFX1010W32: ; Occupancy: 20{{$}} ; GFX1030: ; Occupancy: 16{{$}}
define amdgpu_kernel void @used_lds_8252_max_group_size_192() #6 { define amdgpu_kernel void @used_lds_8252_max_group_size_192() #6 {
%p = bitcast [8252 x i8] addrspace(3)* @lds8252 to i8 addrspace(3)* %p = bitcast [8252 x i8] addrspace(3)* @lds8252 to i8 addrspace(3)*
store volatile i8 1, i8 addrspace(3)* %p store volatile i8 1, i8 addrspace(3)* %p
ret void ret void
} }
; GCN-LABEL: {{^}}used_lds_8252_max_group_size_256: ; GCN-LABEL: {{^}}used_lds_8252_max_group_size_256:
; GFX9: ; Occupancy: 10{{$}} ; GFX9: ; Occupancy: 10{{$}}
; GFX1010W64: ; Occupancy: 20{{$}} ; GFX1010: ; Occupancy: 20{{$}}
; GFX1010W32: ; Occupancy: 20{{$}} ; GFX1030: ; Occupancy: 16{{$}}
define amdgpu_kernel void @used_lds_8252_max_group_size_256() #7 { define amdgpu_kernel void @used_lds_8252_max_group_size_256() #7 {
%p = bitcast [8252 x i8] addrspace(3)* @lds8252 to i8 addrspace(3)* %p = bitcast [8252 x i8] addrspace(3)* @lds8252 to i8 addrspace(3)*
store volatile i8 1, i8 addrspace(3)* %p store volatile i8 1, i8 addrspace(3)* %p
ret void ret void
} }
; GCN-LABEL: {{^}}used_lds_8252_max_group_size_512: ; GCN-LABEL: {{^}}used_lds_8252_max_group_size_512:
; GFX9: ; Occupancy: 10{{$}} ; GFX9: ; Occupancy: 10{{$}}
; GFX1010W64: ; Occupancy: 20{{$}} ; GFX1010: ; Occupancy: 20{{$}}
; GFX1010W32: ; Occupancy: 20{{$}} ; GFX1030: ; Occupancy: 16{{$}}
define amdgpu_kernel void @used_lds_8252_max_group_size_512() #8 { define amdgpu_kernel void @used_lds_8252_max_group_size_512() #8 {
%p = bitcast [8252 x i8] addrspace(3)* @lds8252 to i8 addrspace(3)* %p = bitcast [8252 x i8] addrspace(3)* @lds8252 to i8 addrspace(3)*
store volatile i8 1, i8 addrspace(3)* %p store volatile i8 1, i8 addrspace(3)* %p
ret void ret void
} }
; GCN-LABEL: {{^}}used_lds_8252_max_group_size_1024: ; GCN-LABEL: {{^}}used_lds_8252_max_group_size_1024:
; GFX9: ; Occupancy: 10{{$}} ; GFX9: ; Occupancy: 10{{$}}
; GFX1010W64: ; Occupancy: 20{{$}} ; GFX1010: ; Occupancy: 20{{$}}
; GFX1010W32: ; Occupancy: 20{{$}} ; GFX1030: ; Occupancy: 16{{$}}
define amdgpu_kernel void @used_lds_8252_max_group_size_1024() #9 { define amdgpu_kernel void @used_lds_8252_max_group_size_1024() #9 {
%p = bitcast [8252 x i8] addrspace(3)* @lds8252 to i8 addrspace(3)* %p = bitcast [8252 x i8] addrspace(3)* @lds8252 to i8 addrspace(3)*
store volatile i8 1, i8 addrspace(3)* %p store volatile i8 1, i8 addrspace(3)* %p
ret void ret void
} }
; GCN-LABEL: {{^}}used_lds_8252_max_group_size_32: ; GCN-LABEL: {{^}}used_lds_8252_max_group_size_32:
; GFX9: ; Occupancy: 7{{$}} ; GFX9: ; Occupancy: 7{{$}}
; GFX1010W64: ; Occupancy: 7{{$}} ; GFX10: ; Occupancy: 7{{$}}
; GFX1010W32: ; Occupancy: 7{{$}}
define amdgpu_kernel void @used_lds_8252_max_group_size_32() #10 { define amdgpu_kernel void @used_lds_8252_max_group_size_32() #10 {
%p = bitcast [8252 x i8] addrspace(3)* @lds8252 to i8 addrspace(3)* %p = bitcast [8252 x i8] addrspace(3)* @lds8252 to i8 addrspace(3)*
store volatile i8 1, i8 addrspace(3)* %p store volatile i8 1, i8 addrspace(3)* %p
ret void ret void
} }
attributes #0 = { "amdgpu-waves-per-eu"="2,3" } attributes #0 = { "amdgpu-waves-per-eu"="2,3" }
attributes #1 = { "amdgpu-waves-per-eu"="18,18" } attributes #1 = { "amdgpu-waves-per-eu"="18,18" }
Show All 9 Lines