Skip to content

[AMDGPU][Attributor] Fix the merge of two range states of flat workgroup size #125889

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Closed
shiltian wants to merge 1 commit into from
Closed

[AMDGPU][Attributor] Fix the merge of two range states of flat workgroup size #125889

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
35 changes: 29 additions & 6 deletions llvm/lib/Target/AMDGPU/AMDGPUAttributor.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -799,7 +799,31 @@ AAAMDAttributes &AAAMDAttributes::createForPosition(const IRPosition &IRP,
llvm_unreachable("AAAMDAttributes is only valid for function position");
}

/// Base class to derive different size ranges.
/// Merge two range states and indicate changes. The range can apply to
/// flat-workgroup-size and waves-per-eu attributes. The growth/merge of these
/// two attributes is neither a union nor an intersection. Instead, it follows
/// the rule below:
///
/// Lower = max(S.Lower, R.Lower)
/// Upper = max(S.Upper, R.Upper)
ChangeStatus mergeRangeStateAndIndicateChange(IntegerRangeState &S,
const IntegerRangeState &R) {
auto Assumed = S.getAssumed();
unsigned Min = R.getAssumed().getLower().getZExtValue();
unsigned Max = R.getAssumed().getUpper().getZExtValue();
Min = std::max(Min, static_cast<unsigned>(Assumed.getLower().getZExtValue()));
Max = std::max(Max, static_cast<unsigned>(Assumed.getUpper().getZExtValue()));
ConstantRange Range(APInt(32, Min), APInt(32, Max));
IntegerRangeState RangeState(Range);
S = RangeState;
return S == Assumed ? ChangeStatus::UNCHANGED : ChangeStatus::CHANGED;
}

/// Base class for deriving different size ranges. This size range is
/// specifically intended for flat-workgroup-size and waves-per-eu attributes.
/// Therefore, updating the attribute should use
/// \p mergeRangeStateAndIndicateChange. Given this, the attribute may not be
/// applicable to other potential size ranges in the future.
struct AAAMDSizeRangeAttribute
: public StateWrapper<IntegerRangeState, AbstractAttribute, uint32_t> {
using Base = StateWrapper<IntegerRangeState, AbstractAttribute, uint32_t>;
Expand All @@ -826,8 +850,8 @@ struct AAAMDSizeRangeAttribute
if (!CallerInfo || !CallerInfo->isValidState())
return false;

Change |=
clampStateAndIndicateChange(this->getState(), CallerInfo->getState());
Change |= mergeRangeStateAndIndicateChange(this->getState(),
CallerInfo->getState());

return true;
};
Expand Down Expand Up @@ -903,15 +927,14 @@ struct AAAMDFlatWorkGroupSize : public AAAMDSizeRangeAttribute {
}
}

// We don't want to directly clamp the state if it's the max range because
// We don't want to directly merge the state if it's the max range because
// that is basically the worst state.
if (Range == MaxRange)
return;

auto [Min, Max] = Range;
ConstantRange CR(APInt(32, Min), APInt(32, Max + 1));
IntegerRangeState IRS(CR);
clampStateAndIndicateChange(this->getState(), IRS);
this->getState() = IntegerRangeState(CR);

if (HasAttr || AMDGPU::isEntryFunctionCC(F->getCallingConv()))
indicateOptimisticFixpoint();
Expand Down
12 changes: 6 additions & 6 deletions llvm/test/CodeGen/AMDGPU/propagate-flat-work-group-size.ll
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -86,7 +86,7 @@ define internal void @flat_group_512_1024() #4 {

define amdgpu_kernel void @kernel_128_512() #5 {
; CHECK-LABEL: define {{[^@]+}}@kernel_128_512
; CHECK-SAME: () #[[ATTR2]] {
; CHECK-SAME: () #[[ATTR6:[0-9]+]] {
; CHECK-NEXT: call void @default_to_128_512()
; CHECK-NEXT: call void @flat_group_64_64()
; CHECK-NEXT: ret void
Expand All @@ -98,7 +98,7 @@ define amdgpu_kernel void @kernel_128_512() #5 {

define amdgpu_kernel void @kernel_512_512() #6 {
; CHECK-LABEL: define {{[^@]+}}@kernel_512_512
; CHECK-SAME: () #[[ATTR6:[0-9]+]] {
; CHECK-SAME: () #[[ATTR2]] {
; CHECK-NEXT: call void @default_to_128_512()
; CHECK-NEXT: call void @flat_group_512_1024()
; CHECK-NEXT: ret void
Expand All @@ -111,7 +111,7 @@ define amdgpu_kernel void @kernel_512_512() #6 {
; Called from kernels with 128,256 and 64,128 => 64,256
Copy link
Contributor Author

@shiltian shiltian Feb 5, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I should also update the comments

define internal void @default_to_64_256() {
; CHECK-LABEL: define {{[^@]+}}@default_to_64_256
; CHECK-SAME: () #[[ATTR7:[0-9]+]] {
; CHECK-SAME: () #[[ATTR4]] {
; CHECK-NEXT: ret void
;
ret void
Expand Down Expand Up @@ -153,7 +153,7 @@ define internal void @merge_cycle_1() #3 {

define amdgpu_kernel void @kernel_64_256() #7 {
; CHECK-LABEL: define {{[^@]+}}@kernel_64_256
; CHECK-SAME: () #[[ATTR7]] {
; CHECK-SAME: () #[[ATTR7:[0-9]+]] {
; CHECK-NEXT: call void @merge_cycle_0()
; CHECK-NEXT: call void @default_captured_address()
; CHECK-NEXT: call void @externally_visible_default()
Expand Down Expand Up @@ -205,11 +205,11 @@ attributes #7 = { "amdgpu-flat-work-group-size"="64,256" }
;.
; CHECK: attributes #[[ATTR0]] = { "amdgpu-flat-work-group-size"="1,256" "amdgpu-no-agpr" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" }
; CHECK: attributes #[[ATTR1]] = { "amdgpu-flat-work-group-size"="64,128" "amdgpu-no-agpr" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" }
; CHECK: attributes #[[ATTR2]] = { "amdgpu-flat-work-group-size"="128,512" "amdgpu-no-agpr" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "amdgpu-waves-per-eu"="2,10" "uniform-work-group-size"="false" }
; CHECK: attributes #[[ATTR2]] = { "amdgpu-flat-work-group-size"="512,512" "amdgpu-no-agpr" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "amdgpu-waves-per-eu"="2,10" "uniform-work-group-size"="false" }
; CHECK: attributes #[[ATTR3]] = { "amdgpu-flat-work-group-size"="64,64" "amdgpu-no-agpr" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" }
; CHECK: attributes #[[ATTR4]] = { "amdgpu-flat-work-group-size"="128,256" "amdgpu-no-agpr" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" }
; CHECK: attributes #[[ATTR5]] = { "amdgpu-flat-work-group-size"="512,1024" "amdgpu-no-agpr" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "amdgpu-waves-per-eu"="4,10" "uniform-work-group-size"="false" }
; CHECK: attributes #[[ATTR6]] = { "amdgpu-flat-work-group-size"="512,512" "amdgpu-no-agpr" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "amdgpu-waves-per-eu"="2,10" "uniform-work-group-size"="false" }
; CHECK: attributes #[[ATTR6]] = { "amdgpu-flat-work-group-size"="128,512" "amdgpu-no-agpr" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "amdgpu-waves-per-eu"="2,10" "uniform-work-group-size"="false" }
; CHECK: attributes #[[ATTR7]] = { "amdgpu-flat-work-group-size"="64,256" "amdgpu-no-agpr" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" }
; CHECK: attributes #[[ATTR8]] = { "amdgpu-no-agpr" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" }
;.