Skip to content
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

Jump to bottom

[LLVMGPU] Extract subgroup size from export op to use for vector distribution #14826

Merged
merged 3 commits into from
Aug 26, 2023
Merged

[LLVMGPU] Extract subgroup size from export op to use for vector distribution #14826

Show file tree
Hide file tree
Changes from 1 commit
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
697e162
[LLVMGPU] Extract subgroup size from export op to use for vector dist…
qedawkins Aug 25, 2023
4e09303
Use getEntryPoint
qedawkins Aug 25, 2023
3a2493b
clang format
qedawkins Aug 25, 2023
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
41 changes: 30 additions & 11 deletions compiler/src/iree/compiler/Codegen/LLVMGPU/TransformExtensions/LLVMGPUExtensions.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -574,11 +574,13 @@ static Value simpleWarpShuffleFunction(Location loc, OpBuilder &builder,

static void populatePropagateVectorDistribution(Operation *target,
RewritePatternSet &patterns,
PatternBenefit benefit) {
auto groupReductionFn = [](Location loc, OpBuilder &builder, Value input,
vector::CombiningKind kind, uint32_t size) {
PatternBenefit benefit,
unsigned subgroupSize) {
auto groupReductionFn = [subgroupSize](
Location loc, OpBuilder &builder, Value input,
vector::CombiningKind kind, uint32_t size) {
return mlir::iree_compiler::emitGPUGroupReduction(loc, builder, input, kind,
size, 32);
size, subgroupSize);
};
assert(target->hasTrait<OpTrait::IsIsolatedFromAbove>());
vector::populatePropagateWarpVectorDistributionPatterns(
Expand All @@ -604,14 +606,30 @@ static void populateWarpExecuteOnLane0ToScf(

DiagnosedSilenceableFailure
transform_dialect::VectorWarpDistributionOp::applyToOne(
transform::TransformRewriter &rewriter, Operation *target,
transform::TransformRewriter &rewriter, func::FuncOp target,
transform::ApplyToEachResultList &results,
transform::TransformState &state) {
if (!target->hasTrait<OpTrait::IsIsolatedFromAbove>()) {
target->emitOpError(
"applies only to isolated-from-above targets because it "
"needs to apply "
"patterns greedily");
if (!isa<HAL::ExecutableOp, HAL::ExecutableVariantOp>(state.getTopLevel())) {
qedawkins marked this conversation as resolved.
Show resolved Hide resolved
state.getTopLevel()->emitOpError(
"requires HAL::ExecutableOp or HAL::ExecutableVariantOp "
"toplevel to extract subgroup size information");
return emitDefaultDefiniteFailure(target);
}

IREE::HAL::ExecutableExportOp exportOp;
state.getTopLevel()->walk([&](IREE::HAL::ExecutableExportOp op) {
if (op.getSymName() == target.getName())
exportOp = op;
});
if (!exportOp) {
state.getTopLevel()->emitOpError("no IREE::HAL::ExecutableExportOp found");
return emitDefaultDefiniteFailure(target);
}

std::optional<llvm::APInt> subgroupSize = exportOp.getSubgroupSize();
if (!subgroupSize) {
state.getTopLevel()->emitOpError(
"could not extract subgroup size from IREE::HAL::ExecutableExportOp");
return emitDefaultDefiniteFailure(target);
}

Expand Down Expand Up @@ -645,7 +663,8 @@ transform_dialect::VectorWarpDistributionOp::applyToOne(
populateVectorTransferWriteDistribution(target, patterns,
/*benefit=*/2);
populatePropagateVectorDistribution(target, patterns,
/*benefit=*/1);
/*benefit=*/1,
subgroupSize->getSExtValue());
if (failed(
applyPatternsAndFoldGreedily(target, std::move(patterns), config))) {
return mlir::emitDefiniteFailure(
Expand Down
2 changes: 1 addition & 1 deletion compiler/src/iree/compiler/Codegen/LLVMGPU/TransformExtensions/LLVMGPUExtensionsOps.td
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -325,7 +325,7 @@ def VectorWarpDistributionOp : Op<Transform_Dialect, "iree.vector.warp_distribut
let extraClassDeclaration = [{
::mlir::DiagnosedSilenceableFailure applyToOne(
::mlir::transform::TransformRewriter &rewriter,
::mlir::Operation *target,
::mlir::func::FuncOp target,
::mlir::transform::ApplyToEachResultList &results,
::mlir::transform::TransformState &state);
}];
Expand Down
2 changes: 1 addition & 1 deletion compiler/src/iree/compiler/Codegen/LLVMGPU/test/transform_dialect_vector_distribution.mlir
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -11,7 +11,7 @@

hal.executable private @reduce_dispatch_0 {
hal.executable.variant public @cuda_nvptx_fb, target = #executable_target_cuda_nvptx_fb {
hal.executable.export public @reduce_dispatch_0 ordinal(0) layout(#pipeline_layout) attributes { workgroup_size = [64: index, 1: index, 1: index] }
hal.executable.export public @reduce_dispatch_0 ordinal(0) layout(#pipeline_layout) attributes { workgroup_size = [64: index, 1: index, 1: index], subgroup_size = 32 : index }
builtin.module {
func.func @reduce_dispatch_0() {
%c0 = arith.constant 0 : index
Expand Down
Loading