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.

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

Convert a subset of GPU dialect ops to the OpenCL GPU runtime calls #333

Merged
merged 8 commits into from
Sep 24, 2024
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
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
29 changes: 29 additions & 0 deletions include/gc/ExecutionEngine/GPURuntime/GpuOclRuntime.h
Original file line number Diff line number Diff line change
@@ -0,0 +1,29 @@
//===-- GpuOclRuntime.h - GPU OpenCL runtime --------------------*- C++ -*-===//
//
// This file is licensed under the Apache License v2.0 with LLVM Exceptions.
// See https://llvm.org/LICENSE.txt for license information.
// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
//
//===----------------------------------------------------------------------===//

#ifndef GC_GPUOCLRUNTIME_H
#define GC_GPUOCLRUNTIME_H

namespace mlir::gc::gpu {
AndreyPavlenko marked this conversation as resolved.
Show resolved Hide resolved
constexpr char GPU_OCL_MALLOC[] = "gcGpuOclMalloc";
constexpr char GPU_OCL_DEALLOC[] = "gcGpuOclDealloc";
constexpr char GPU_OCL_MEMCPY[] = "gcGpuOclMemcpy";
constexpr char GPU_OCL_KERNEL_CREATE[] = "gcGpuOclKernelCreate";
constexpr char GPU_OCL_KERNEL_DESTROY[] = "gcGpuOclKernelDestroy";
constexpr char GPU_OCL_KERNEL_LAUNCH[] = "gcGpuOclKernelLaunch";
constexpr char GPU_OCL_MOD_DESTRUCTOR[] = "gcGpuOclModuleDestructor";
} // namespace mlir::gc::gpu

#ifndef GC_GPU_OCL_CONST_ONLY

// TBD

#else
#undef GC_GPU_OCL_CONST_ONLY
Comment on lines +22 to +27
Copy link
Contributor

Choose a reason for hiding this comment

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

dead code

Copy link
Contributor Author

Choose a reason for hiding this comment

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

This is a placeholder for #343. It also undefs this one

#define GC_GPU_OCL_CONST_ONLY

#endif
#endif
14 changes: 14 additions & 0 deletions include/gc/Transforms/Passes.td
Original file line number Diff line number Diff line change
Expand Up @@ -93,6 +93,20 @@ def LinalgToXeGPU : Pass<"linalg-to-xegpu", "func::FuncOp"> {
"DPAS register block sizes MxNxK">,
];
}

def AddContextArg : Pass<"add-ctx-arg", "func::FuncOp"> {
Copy link
Contributor

Choose a reason for hiding this comment

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

Should it be applied to kernels only?

Copy link
Contributor Author

Choose a reason for hiding this comment

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

No, it should be applied to the functions, that launch the kernels. The argument is passed to the runtime.

let summary = "Add a context argument.";
let description = [{
Add a new memref argument to the function, that could be used to pass some context.
}];
}

def GpuToGpuOcl : Pass<"gpu-to-gpuocl", "ModuleOp"> {
let summary = "Convert the GPU operations to GpuOclRuntime calls.";
let description = [{
Convert the gpu alloc, dealloc, memcpy and launch operations to GpuOclRuntime calls.
}];
}
#endif // GC_USE_IMEX

def IterativeTilingAndFusion : Pass<"iterative-tiling-and-fusion",
Expand Down
54 changes: 54 additions & 0 deletions lib/gc/Transforms/GPU/AddContextArg.cpp
Original file line number Diff line number Diff line change
@@ -0,0 +1,54 @@
//===-- AddContextArg.cpp - Add context argument ----------------*- C++ -*-===//
//
// This file is licensed under the Apache License v2.0 with LLVM Exceptions.
// See https://llvm.org/LICENSE.txt for license information.
// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
//
//===----------------------------------------------------------------------===//
#include "mlir/Conversion/Passes.h"
#include "mlir/Dialect/Func/IR/FuncOps.h"

namespace mlir::gc {
#define GEN_PASS_DECL_ADDCONTEXTARG
#define GEN_PASS_DEF_ADDCONTEXTARG
#include "gc/Transforms/Passes.h.inc"
} // namespace mlir::gc

using namespace mlir;

namespace {
struct AddContextArg final : gc::impl::AddContextArgBase<AddContextArg> {
void runOnOperation() override {
auto func = getOperation();
if (func.isExternal()) {
return;
}

auto funcType = func.getFunctionType();
auto argTypes = llvm::to_vector<8>(funcType.getInputs());
auto resultTypes = llvm::to_vector<1>(funcType.getResults());
auto ctx = func->getContext();
auto newArgType = MemRefType::get({}, IntegerType::get(ctx, 8));
argTypes.emplace_back(newArgType);
auto newFuncType = FunctionType::get(ctx, argTypes, resultTypes);
func.setType(newFuncType);
func.getBody().front().addArgument(newArgType, func.getLoc());

// Find all function calls and append the last argument of the current
// function to the call.
auto module = func->getParentOfType<ModuleOp>();
func.walk([&](func::CallOp call) {
// If the function to be called is defined in the current module, then the
// context arg will be added to this function signature either and, thus,
// wee need add the context arg to the function call.
if (auto callee = module.lookupSymbol<func::FuncOp>(call.getCallee());
!callee || callee.isExternal()) {
return;
}
auto args = llvm::to_vector<8>(call.getOperands());
args.emplace_back(func.getArgument(func.getNumArguments() - 1));
call->setOperands(args);
});
}
};
} // namespace
2 changes: 2 additions & 0 deletions lib/gc/Transforms/GPU/CMakeLists.txt
Original file line number Diff line number Diff line change
@@ -1,4 +1,6 @@
gc_add_mlir_library(GcGpuPasses
AddContextArg.cpp
GpuToGpuOcl.cpp
LinalgToXeGPU.cpp
Pipeline.cpp

Expand Down
Loading