OSSForks/tensorflow
1
0
Fork 0
mirror of https://github.com/zebrajr/tensorflow.git synced 2025-12-06 12:20:11 +01:00
Code Issues Actions 5 Packages Projects Releases Wiki Activity

[xla:codegen] Remove LlvmKernelDefinition alias

Browse Source 
Rely on KernelEmitter<T>::KernelDefinition alias and CTAD to avoid spelling full type.

PiperOrigin-RevId: 825336093
This commit is contained in:
Eugene Zhulenev 2025-10-28 20:39:07 -07:00 committed by TensorFlower Gardener
parent 8e76d82f01
commit 1444679887
23 changed files with 63 additions and 58 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
third_party/xla/xla/backends/cpu/codegen/computation_kernel_emitter.cc vendored
View File

@ -132,7 +132,7 @@ ComputationKernelEmitter::ComputationKernelEmitter(
buffer_assignment_(buffer_assignment),
target_machine_(target_machine) {}
absl::StatusOr<LlvmKernelDefinition>
absl::StatusOr<ComputationKernelEmitter::KernelDefinition>
ComputationKernelEmitter::EmitKernelDefinition() {
VLOG(2) << "Emit Computation host kernel: " << instr_->name();
@ -221,7 +221,7 @@ ComputationKernelEmitter::EmitKernelDefinition() {
std::move(kernel_prototype.result_buffers),
std::move(kernel_prototype.invariant_arguments));
return LlvmKernelDefinition(std::move(spec), std::move(source));
return KernelDefinition(std::move(spec), std::move(source));
}
absl::StatusOr<llvm::Function*> ComputationKernelEmitter::EmitNestedComputation(

2
third_party/xla/xla/backends/cpu/codegen/computation_kernel_emitter.h vendored
View File

@ -51,7 +51,7 @@ class ComputationKernelEmitter final : public KernelEmitter<LlvmKernelSource> {
const TargetMachineFeatures* target_machine);
absl::string_view name() const final { return "computation_kernel_emitter"; }
absl::StatusOr<LlvmKernelDefinition> EmitKernelDefinition() final;
absl::StatusOr<KernelDefinition> EmitKernelDefinition() final;
private:
absl::StatusOr<llvm::Function*> EmitNestedComputation(

5
third_party/xla/xla/backends/cpu/codegen/dot/dot_kernel_emitter.cc vendored
View File

@ -56,7 +56,8 @@ DotKernelEmitter::DotKernelEmitter(const HloInstruction* instr,
buffer_assignment_(buffer_assignment),
target_machine_(target_machine) {}
absl::StatusOr<LlvmKernelDefinition> DotKernelEmitter::EmitKernelDefinition() {
absl::StatusOr<DotKernelEmitter::KernelDefinition>
DotKernelEmitter::EmitKernelDefinition() {
const HloModuleConfig& config = instr_->GetModule()->config();
DotImplementationStrategy strategy = GetDotImplementationStrategy(
@ -111,7 +112,7 @@ absl::StatusOr<LlvmKernelDefinition> DotKernelEmitter::EmitKernelDefinition() {
std::move(kernel_prototype.result_buffers),
std::move(kernel_prototype.invariant_arguments));
return LlvmKernelDefinition(std::move(spec), std::move(source));
return KernelDefinition(std::move(spec), std::move(source));
}
} // namespace xla::cpu

2
third_party/xla/xla/backends/cpu/codegen/dot/dot_kernel_emitter.h vendored
View File

@ -33,7 +33,7 @@ class DotKernelEmitter final : public KernelEmitter<LlvmKernelSource> {
const TargetMachineFeatures* target_machine);
absl::string_view name() const final { return "dot_kernel_emitter"; }
absl::StatusOr<LlvmKernelDefinition> EmitKernelDefinition() override;
absl::StatusOr<KernelDefinition> EmitKernelDefinition() override;
private:
const HloInstruction* instr_;

5
third_party/xla/xla/backends/cpu/codegen/elemental/concatenate_kernel_emitter.cc vendored
View File

@ -43,7 +43,6 @@ limitations under the License.
#include "xla/service/cpu/ir_emitter.h"
#include "xla/service/llvm_ir/ir_array.h"
#include "xla/shape.h"
#include "xla/tsl/platform/errors.h"
#include "xla/tsl/platform/statusor.h"
#include "xla/util.h"
@ -67,7 +66,7 @@ ConcatenateKernelEmitter::ConcatenateKernelEmitter(
buffer_assignment_(buffer_assignment),
target_machine_(target_machine) {}
absl::StatusOr<LlvmKernelDefinition>
absl::StatusOr<ConcatenateKernelEmitter::KernelDefinition>
ConcatenateKernelEmitter::EmitKernelDefinition() {
if (absl::Status status = CanDoFastConcatenate(instr_); !status.ok()) {
VLOG(1) << "Could not emit fast concatenate for " << instr_->ToString()
@ -122,7 +121,7 @@ ConcatenateKernelEmitter::EmitKernelDefinition() {
std::move(kernel_prototype.result_buffers),
std::move(kernel_prototype.invariant_arguments));
return LlvmKernelDefinition(std::move(spec), std::move(source));
return KernelDefinition(std::move(spec), std::move(source));
}
} // namespace xla::cpu

2
third_party/xla/xla/backends/cpu/codegen/elemental/concatenate_kernel_emitter.h vendored
View File

@ -33,7 +33,7 @@ class ConcatenateKernelEmitter final : public KernelEmitter<LlvmKernelSource> {
const TargetMachineFeatures* target_machine);
absl::string_view name() const final { return "concatenate_kernel_emitter"; }
absl::StatusOr<LlvmKernelDefinition> EmitKernelDefinition() override;
absl::StatusOr<KernelDefinition> EmitKernelDefinition() override;
private:
const HloInstruction* instr_;

4
third_party/xla/xla/backends/cpu/codegen/elemental/elemental_kernel_emitter.cc vendored
View File

@ -168,7 +168,7 @@ ElementalKernelEmitter::ElementalKernelEmitter(
buffer_assignment_(buffer_assignment),
target_machine_(target_machine) {}
absl::StatusOr<LlvmKernelDefinition>
absl::StatusOr<ElementalKernelEmitter::KernelDefinition>
ElementalKernelEmitter::EmitKernelDefinition() {
VLOG(2) << "Emit elemental host kernel: " << instr_->name();
@ -234,7 +234,7 @@ ElementalKernelEmitter::EmitKernelDefinition() {
std::move(kernel_prototype.result_buffers),
std::move(kernel_prototype.invariant_arguments));
return LlvmKernelDefinition(std::move(spec), std::move(source));
return KernelDefinition(std::move(spec), std::move(source));
}
absl::StatusOr<NumWorkGroups> ElementalKernelEmitter::EmitElementalLoops(

2
third_party/xla/xla/backends/cpu/codegen/elemental/elemental_kernel_emitter.h vendored
View File

@ -39,7 +39,7 @@ class ElementalKernelEmitter final : public KernelEmitter<LlvmKernelSource> {
const TargetMachineFeatures* target_machine);
absl::string_view name() const final { return "elemental_kernel_emitter"; }
absl::StatusOr<LlvmKernelDefinition> EmitKernelDefinition() override;
absl::StatusOr<KernelDefinition> EmitKernelDefinition() override;
private:
// Emits LLVM IR using elemental loop emitter and the given element generator.

2
third_party/xla/xla/backends/cpu/codegen/elemental/elemental_kernel_emitter_test.cc vendored
View File

@ -46,7 +46,7 @@ class ElementalKernelEmitterTest : public HloHardwareIndependentTestBase {
ElementalKernelEmitterTest()
: target_machine_features_([](int64_t size) { return 1; }) {}
absl::StatusOr<LlvmKernelDefinition> EmitKernelDefinition(
absl::StatusOr<KernelDefinition<LlvmKernelSource>> EmitKernelDefinition(
const HloInstruction* instr, const BufferAssignment* buffer_assignment) {
ElementalKernelEmitter emitter(instr, buffer_assignment,
&target_machine_features_);

1
third_party/xla/xla/backends/cpu/testlib/BUILD vendored
View File

@ -48,6 +48,7 @@ cc_library(
"@com_google_absl//absl/log:check",
"@com_google_absl//absl/status",
"@com_google_absl//absl/status:statusor",
"@com_google_absl//absl/strings:string_view",
"@com_google_absl//absl/types:span",
"@llvm-project//llvm:Target",
"@llvm-project//llvm:ir_headers",

17
third_party/xla/xla/backends/cpu/testlib/kernel_runner.cc vendored
View File

@ -16,12 +16,12 @@ limitations under the License.
#include "xla/backends/cpu/testlib/kernel_runner.h"
#include <memory>
#include <string>
#include <utility>
#include <vector>
#include "absl/log/check.h"
#include "absl/status/status.h"
#include "absl/strings/string_view.h"
#include "absl/types/span.h"
#include "llvm/IR/DataLayout.h"
#include "llvm/IR/Module.h"
@ -51,10 +51,9 @@ limitations under the License.
namespace xla::cpu {
absl::StatusOr<KernelRunner> KernelRunner::Create(
LlvmKernelDefinition kernel_definition, JitCompiler compiler) {
auto spec = kernel_definition.spec();
auto thread_safe_module =
std::move(kernel_definition).TakeSource().thread_safe_module();
KernelDefinition<LlvmKernelSource> kernel, JitCompiler compiler) {
auto spec = kernel.spec();
auto thread_safe_module = std::move(kernel).TakeSource().thread_safe_module();
SetModuleMemoryRegionName(*thread_safe_module.getModuleUnlocked(),
"kernel_runner_test");
@ -73,12 +72,12 @@ absl::StatusOr<KernelRunner> KernelRunner::Create(
}
absl::StatusOr<KernelRunner> KernelRunner::Create(
MlirKernelDefinition kernel_definition, JitCompiler compiler) {
auto spec = kernel_definition.spec();
auto source = std::move(kernel_definition).TakeSource();
KernelDefinition<MlirKernelSource> kernel, JitCompiler compiler) {
auto spec = kernel.spec();
auto source = std::move(kernel).TakeSource();
TF_ASSIGN_OR_RETURN(LlvmKernelSource llvm_kernel_source, LowerToLlvm(source));
return Create(LlvmKernelDefinition(spec, std::move(llvm_kernel_source)),
return Create(KernelDefinition(spec, std::move(llvm_kernel_source)),
std::move(compiler));
}

4
third_party/xla/xla/backends/cpu/testlib/kernel_runner.h vendored
View File

@ -37,9 +37,9 @@ namespace xla::cpu {
class KernelRunner final : public xla::KernelRunner {
public:
static absl::StatusOr<KernelRunner> Create(
LlvmKernelDefinition kernel_definition, JitCompiler compiler);
KernelDefinition<LlvmKernelSource> kernel, JitCompiler compiler);
static absl::StatusOr<KernelRunner> Create(
MlirKernelDefinition kernel_definition, JitCompiler compiler);
KernelDefinition<MlirKernelSource> kernel, JitCompiler compiler);
KernelRunner(KernelRunner&&) = default;
KernelRunner& operator=(KernelRunner&&) = default;

5
third_party/xla/xla/backends/cpu/testlib/kernel_runner_extension.cc vendored
View File

@ -258,8 +258,9 @@ NB_MODULE(_extension, kernel_runner_module) {
return *std::move(runner);
})
.def_static(
"create", [](std::unique_ptr<LlvmKernelDefinition,
nb::deleter<LlvmKernelDefinition>>
"create",
[](std::unique_ptr<KernelDefinition<LlvmKernelSource>,
nb::deleter<KernelDefinition<LlvmKernelSource>>>
kernel_definition,
std::unique_ptr<JitCompiler, nb::deleter<JitCompiler>>
jit_compiler) {

4
third_party/xla/xla/backends/cpu/testlib/llvm_ir_kernel_emitter.cc vendored
View File

@ -52,7 +52,7 @@ LlvmTestKernelEmitter::LlvmTestKernelEmitter(absl::string_view llvm_ir,
}
}
absl::StatusOr<LlvmKernelDefinition>
absl::StatusOr<LlvmTestKernelEmitter::KernelDefinition>
LlvmTestKernelEmitter::EmitKernelDefinition() {
auto context = std::make_unique<llvm::LLVMContext>();
@ -84,7 +84,7 @@ LlvmTestKernelEmitter::EmitKernelDefinition() {
KernelSpec kernel_spec(kernel_name_, num_workgroups_,
std::move(argument_buffers), std::move(result_buffers),
/*invariant_arguments=*/{});
return LlvmKernelDefinition(std::move(kernel_spec), std::move(source));
return KernelDefinition(std::move(kernel_spec), std::move(source));
}
} // namespace xla::cpu

2
third_party/xla/xla/backends/cpu/testlib/llvm_ir_kernel_emitter.h vendored
View File

@ -51,7 +51,7 @@ class LlvmTestKernelEmitter : public KernelEmitter<LlvmKernelSource> {
absl::Span<const KernelArg> args);
absl::string_view name() const override { return "llvm_test_kernel_emitter"; }
absl::StatusOr<LlvmKernelDefinition> EmitKernelDefinition() final;
absl::StatusOr<KernelDefinition> EmitKernelDefinition() final;
private:
std::string llvm_ir_;

6
third_party/xla/xla/codegen/kernel_emitter.h vendored
View File

@ -58,14 +58,14 @@ class KernelEmitter : public KernelEmitterBase {
static_assert(std::is_base_of_v<KernelSource, Source>,
"Source must be a subclass of KernelSource");
virtual absl::StatusOr<KernelDefinition<Source>> EmitKernelDefinition() = 0;
using KernelDefinition = ::xla::KernelDefinition<Source>;
virtual absl::StatusOr<KernelDefinition> EmitKernelDefinition() = 0;
private:
absl::StatusOr<std::unique_ptr<KernelDefinitionBase>>
EmitKernelDefinitionBase() final {
TF_ASSIGN_OR_RETURN(auto kernel_definition, EmitKernelDefinition());
return std::make_unique<KernelDefinition<Source>>(
std::move(kernel_definition));
return std::make_unique<KernelDefinition>(std::move(kernel_definition));
}
};

3
third_party/xla/xla/codegen/llvm_kernel_source.h vendored
View File

@ -23,7 +23,6 @@ limitations under the License.
#include "llvm/ExecutionEngine/Orc/ThreadSafeModule.h"
#include "llvm/IR/LLVMContext.h"
#include "llvm/IR/Module.h"
#include "xla/codegen/kernel_definition.h"
#include "xla/codegen/kernel_source.h"
namespace xla {
@ -51,8 +50,6 @@ class LlvmKernelSource final : public KernelSource {
llvm::orc::ThreadSafeModule module_;
};
using LlvmKernelDefinition = KernelDefinition<LlvmKernelSource>; // NOLINT
} // namespace xla
#endif // XLA_CODEGEN_LLVM_KERNEL_SOURCE_H_

4
third_party/xla/xla/codegen/testlib/kernel_runner_extension.cc vendored
View File

@ -206,9 +206,9 @@ NB_MODULE(_extension, kernel_runner_module) {
},
nb::rv_policy::reference_internal);
nb::class_<MlirKernelDefinition, KernelDefinitionBase>(
nb::class_<KernelDefinition<MlirKernelSource>, KernelDefinitionBase>(
kernel_runner_module, "MlirKernelDefinition");
nb::class_<LlvmKernelDefinition, KernelDefinitionBase>(
nb::class_<KernelDefinition<LlvmKernelSource>, KernelDefinitionBase>(
kernel_runner_module, "LlvmKernelDefinition");
nb::class_<KernelEmitterBase>(kernel_runner_module, "KernelEmitterBase")

1
third_party/xla/xla/service/cpu/BUILD vendored
View File

@ -2047,6 +2047,7 @@ cc_library(
deps = [
"//xla/backends/cpu/codegen:fusion_compiler",
"//xla/backends/cpu/codegen:fusion_emitter",
"//xla/codegen:kernel_definition",
"//xla/codegen:kernel_spec",
"//xla/codegen:llvm_kernel_source",
"//xla/codegen:mlir_kernel_source",

13
third_party/xla/xla/service/cpu/parallel_fusion_emitter.cc vendored
View File

@ -31,6 +31,7 @@ limitations under the License.
#include "mlir/IR/MLIRContext.h"
#include "xla/backends/cpu/codegen/fusion_compiler.h"
#include "xla/backends/cpu/codegen/fusion_emitter.h"
#include "xla/codegen/kernel_definition.h"
#include "xla/codegen/kernel_spec.h"
#include "xla/codegen/llvm_kernel_source.h"
#include "xla/codegen/mlir_kernel_source.h"
@ -190,7 +191,7 @@ absl::StatusOr<KernelSpec> ParallelFusionEmitter::AddFusion(
return spec;
}
absl::StatusOr<std::vector<LlvmKernelDefinition>>
absl::StatusOr<std::vector<KernelDefinition<LlvmKernelSource>>>
ParallelFusionEmitter::ConsumeKernels() {
absl::MutexLock lock(kernels_mutex_);
@ -203,8 +204,8 @@ ParallelFusionEmitter::ConsumeKernels() {
}
// Sort the kernels by name to ensure a deterministic order.
absl::c_sort(kernels_, [](const LlvmKernelDefinition& lhs,
const LlvmKernelDefinition& rhs) {
absl::c_sort(kernels_, [](const KernelDefinition<LlvmKernelSource>& lhs,
const KernelDefinition<LlvmKernelSource>& rhs) {
return lhs.spec().name() < rhs.spec().name();
});
@ -212,12 +213,12 @@ ParallelFusionEmitter::ConsumeKernels() {
}
void ParallelFusionEmitter::CompileFusion(
std::shared_ptr<MlirKernelDefinition> mlir_kernel_definition,
std::shared_ptr<KernelDefinition<MlirKernelSource>> mlir_kernel,
std::shared_ptr<CompilerInstance> compiler_instance) {
KernelSpec spec = mlir_kernel_definition->spec();
KernelSpec spec = mlir_kernel->spec();
absl::StatusOr<LlvmKernelSource> llvm_kernel_source =
compiler_instance->compiler->Compile(
std::move(*mlir_kernel_definition).TakeSource());
std::move(*mlir_kernel).TakeSource());
absl::MutexLock lock(kernels_mutex_);
outstanding_kernels_--;

10
third_party/xla/xla/service/cpu/parallel_fusion_emitter.h vendored
View File

@ -25,8 +25,10 @@ limitations under the License.
#include "absl/status/statusor.h"
#include "absl/synchronization/mutex.h"
#include "xla/backends/cpu/codegen/fusion_compiler.h"
#include "xla/codegen/kernel_definition.h"
#include "xla/codegen/kernel_spec.h"
#include "xla/codegen/llvm_kernel_source.h"
#include "xla/codegen/mlir_kernel_source.h"
#include "xla/hlo/ir/hlo_instructions.h"
#include "xla/service/buffer_assignment.h"
#include "xla/tsl/platform/threadpool.h"
@ -52,14 +54,15 @@ class ParallelFusionEmitter {
// Returns the kernels for all the added fusions, blocks until all kernels
// have been compiled.
absl::StatusOr<std::vector<LlvmKernelDefinition>> ConsumeKernels();
absl::StatusOr<std::vector<KernelDefinition<LlvmKernelSource>>>
ConsumeKernels();
private:
struct CompilerInstance;
class FusionCompilerPool;
void CompileFusion(
std::shared_ptr<MlirKernelDefinition> mlir_kernel_definition,
std::shared_ptr<KernelDefinition<MlirKernelSource>> mlir_kernel,
std::shared_ptr<CompilerInstance> compiler_instance);
tsl::thread::ThreadPool& thread_pool_;
@ -70,7 +73,8 @@ class ParallelFusionEmitter {
absl::Mutex kernels_mutex_;
int64_t outstanding_kernels_ ABSL_GUARDED_BY(kernels_mutex_) = 0;
absl::Status kernels_status_ ABSL_GUARDED_BY(kernels_mutex_);
std::vector<LlvmKernelDefinition> kernels_ ABSL_GUARDED_BY(kernels_mutex_);
std::vector<KernelDefinition<LlvmKernelSource>> kernels_
ABSL_GUARDED_BY(kernels_mutex_);
};
} // namespace xla::cpu

2
third_party/xla/xla/service/cpu/parallel_fusion_emitter_test.cc vendored
View File

@ -123,7 +123,7 @@ TEST_F(ParallelFusionEmitterTest, HappyPathSingleFusion) {
TF_ASSERT_OK_AND_ASSIGN(auto kernels, fussion_emitter.ConsumeKernels());
ASSERT_EQ(kernels.size(), 1);
LlvmKernelDefinition& lowered_kernel = kernels[0];
KernelDefinition<LlvmKernelSource>& lowered_kernel = kernels[0];
EXPECT_EQ(lowered_kernel.spec().name(), expected_name);
auto source = std::move(lowered_kernel).TakeSource();

13
third_party/xla/xla/service/cpu/thunk_emitter.cc vendored
View File

@ -234,11 +234,12 @@ absl::StatusOr<ThunkSequence> ThunkEmitter::EmitEntryComputation(
absl::StatusOr<std::vector<ThunkEmitter::EmittedKernel>>
ThunkEmitter::ConsumeKernels() {
tsl::profiler::TraceMe trace("ThunkEmitter::ConsumeKernels");
TF_ASSIGN_OR_RETURN(std::vector<LlvmKernelDefinition> fusion_kernels,
TF_ASSIGN_OR_RETURN(
std::vector<KernelDefinition<LlvmKernelSource>> fusion_kernels,
parallel_fusion_emitter_.ConsumeKernels());
kernels_.reserve(kernels_.size() + fusion_kernels.size());
for (LlvmKernelDefinition& kernel : fusion_kernels) {
for (KernelDefinition<LlvmKernelSource>& kernel : fusion_kernels) {
std::string name(kernel.spec().name());
auto source = std::move(kernel).TakeSource();
kernels_.push_back({name, std::move(source).thread_safe_module()});
@ -688,7 +689,7 @@ absl::StatusOr<ThunkSequence> ThunkEmitter::EmitCallThunk(
maybe_small_call.has_value() && *maybe_small_call == "true") {
ComputationKernelEmitter emitter(instruction, &buffer_assignment_,
&target_machine_features_);
TF_ASSIGN_OR_RETURN(LlvmKernelDefinition kernel_definition,
TF_ASSIGN_OR_RETURN(KernelDefinition kernel_definition,
emitter.EmitKernelDefinition());
auto kernel_spec = kernel_definition.spec();
@ -712,7 +713,7 @@ absl::StatusOr<ThunkSequence> ThunkEmitter::EmitConcatenateKernelThunk(
const HloInstruction* instruction) {
ConcatenateKernelEmitter emitter(instruction, &buffer_assignment_,
&target_machine_features_);
TF_ASSIGN_OR_RETURN(LlvmKernelDefinition kernel_definition,
TF_ASSIGN_OR_RETURN(KernelDefinition kernel_definition,
emitter.EmitKernelDefinition());
auto kernel_spec = kernel_definition.spec();
@ -818,7 +819,7 @@ absl::StatusOr<ThunkSequence> ThunkEmitter::EmitElementalKernelThunk(
const HloInstruction* instruction) {
ElementalKernelEmitter emitter(instruction, &buffer_assignment_,
&target_machine_features_);
TF_ASSIGN_OR_RETURN(LlvmKernelDefinition kernel_definition,
TF_ASSIGN_OR_RETURN(KernelDefinition kernel_definition,
emitter.EmitKernelDefinition());
auto kernel_spec = kernel_definition.spec();
@ -1062,7 +1063,7 @@ absl::StatusOr<ThunkSequence> ThunkEmitter::EmitDotThunk(
case DotImplementationStrategy::kTiledLlvmIrGemv: {
DotKernelEmitter emitter(instruction, &buffer_assignment_,
&target_machine_features_);
TF_ASSIGN_OR_RETURN(LlvmKernelDefinition kernel_definition,
TF_ASSIGN_OR_RETURN(KernelDefinition kernel_definition,
emitter.EmitKernelDefinition());
auto kernel_spec = kernel_definition.spec();