decouple IMEX and GPU

Author: akroviakov

CMakeLists.txt

@@ -1,5 +1,5 @@
 ################################################################################
-# Copyright (C) 2024 Intel Corporation
+# Copyright (C) 2025 Intel Corporation
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
@@ -45,7 +45,8 @@ option(GC_ENABLE_TEST_DNNL_API "Build the dnnl tests" ${GC_ENABLE_DNNL_API})
 option(GC_ENABLE_TEST_MLIR "Build the mlir tests" ON)
 option(GC_ENABLE_TOOLS "Build the tools" ON)
 option(GC_ENABLE_OPT "Build gc-opt" ${GC_ENABLE_TOOLS})
-option(GC_ENABLE_IMEX "Enable Intel® Extension for MLIR" OFF)
+option(GC_ENABLE_IMEX "Enable Intel® Extension for MLIR (implicitly enables GPU compilation)" OFF)
+option(GC_ENABLE_GPU "Enable GPU runtime and tools components" OFF)
 option(GC_ENABLE_BINDINGS_PYTHON "Enable Graph Complier Python Binding" ON)
 option(GC_DEV_LINK_LLVM_DYLIB "Link dynamic libraries of LLVM and MLIR. For developers only. Do not use it in packing the library." OFF)
 option(GC_ENABLE_RUNTIME_NAIVE_BRGEMM "Use naive BRGEMM as runtime backend for debug purpose." OFF)
@@ -55,6 +56,10 @@ if(GC_ENABLE_LEGACY)
   add_subdirectory(legacy/core)
 endif()
 
+if (GC_ENABLE_GPU)
+  set(GC_ENABLE_GPU ON)
+endif()
+
 if (GC_ENABLE_IMEX)
   # normalize the value for lit config
   set(GC_ENABLE_IMEX ON)
@@ -70,6 +75,9 @@ endif()
 ############################## Targets #########################################
 # All common options, includes etc. are added to this interface target.
 add_library(GcInterface INTERFACE)
+if (GC_ENABLE_GPU)
+  target_compile_options(GcInterface INTERFACE -DGC_USE_GPU)
+endif()
 target_compile_features(GcInterface INTERFACE cxx_std_17)
 target_include_directories(GcInterface INTERFACE
   $<BUILD_INTERFACE:${PROJECT_BINARY_DIR}/include>

README.md

@@ -76,5 +76,6 @@ Graph Compiler supports the following build-time options.
 | GC_ENABLE_TEST            | **ON**, OFF                            | Controls building the tests                                     |
 | GC_DEV_LINK_LLVM_DYLIB    | ON, **OFF**                            | Controls dynamic link LLVM/MLIR libraries, mainly for developer |
 | GC_ENABLE_BINDINGS_PYTHON | **ON**, OFF                            | Controls building the Python API                                |
-| GC_ENABLE_IMEX            | ON, **OFF**                            | Whether to enable the GPU components                            |
+| GC_ENABLE_IMEX            | ON, **OFF**                            | Whether to enable the IMEX components                           |
+| GC_ENABLE_GPU             | ON, **OFF**                            | Whether to enable the GPU tools and components                            |
 

include/gc/Conversion/Passes.h

@@ -10,6 +10,7 @@
 #define GC_CONVERSION_PASSES_H
 
 #include "gc/Conversion/XeVMToLLVM/XeVMToLLVM.h"
+#include "mlir/Pass/Pass.h"
 
 namespace mlir {
 

include/gc/Transforms/CMakeLists.txt

@@ -1,8 +1,11 @@
 if(GC_ENABLE_DNNL_API)
     list(APPEND TABLEGEN_MACROS -DGC_HAS_ONEDNN_DIALECT)
 endif()
+if(GC_ENABLE_GPU)
+    list(APPEND TABLEGEN_MACROS -DGC_USE_GPU)
+endif()
 if(GC_ENABLE_IMEX)
-    list(APPEND TABLEGEN_MACROS -DGC_USE_IMEX)
+    list(APPEND TABLEGEN_MACROS -DGC_USE_IMEX -DGC_USE_GPU)
 endif()
 
 set(LLVM_TARGET_DEFINITIONS Passes.td)

include/gc/Transforms/Passes.h

@@ -115,7 +115,6 @@ std::unique_ptr<Pass> createMergeAllocPass();
 void populateFrontendPasses(mlir::OpPassManager &);
 void populateCPUPipeline(mlir::OpPassManager &);
 
-#ifdef GC_USE_IMEX
 struct GPUPipelineOptions : PassPipelineOptions<GPUPipelineOptions> {
   Option<bool> isUsmArgs{
       *this, "is-usm-args",
@@ -136,6 +135,8 @@ struct GPUPipelineOptions : PassPipelineOptions<GPUPipelineOptions> {
       llvm::cl::init(false)};
 };
 void populateGPUPipeline(mlir::OpPassManager &, const GPUPipelineOptions &);
+#ifdef GC_USE_IMEX
+void populateIMEXPipeline(mlir::OpPassManager &, const GPUPipelineOptions &);
 #endif
 
 #define GEN_PASS_DECL

include/gc/Transforms/Passes.td

@@ -74,6 +74,7 @@ def ConvertMemRefToCPURuntime : Pass<"convert-memref-to-cpuruntime", "func::Func
       "cpuruntime::CPURuntimeDialect"
   ];
 }
+#ifdef GC_USE_GPU
 
 #ifdef GC_USE_IMEX
 def LinalgToXeGPU : Pass<"linalg-to-xegpu", "func::FuncOp"> {
@@ -93,6 +94,20 @@ def LinalgToXeGPU : Pass<"linalg-to-xegpu", "func::FuncOp"> {
                "DPAS register block sizes MxNxK">,
   ];
 }
+#endif
+
+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.
+  }];
+  let options = [
+    Option<"callFinish", "call-finish", "bool",
+           /*default=*/"false",
+           "Call finish() after each kernel launch.">
+    ];
+}
+#endif // GC_USE_GPU
 
 def AddContextArg : Pass<"add-ctx-arg", "func::FuncOp"> {
   let summary = "Add a context argument.";
@@ -109,17 +124,6 @@ def AllocsToSLM : Pass<"allocs-to-slm", "func::FuncOp"> {
   ];
 }
 
-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.
-  }];
-  let options = [
-    Option<"callFinish", "call-finish", "bool",
-           /*default=*/"false",
-           "Call finish() after each kernel launch.">
-    ];
-}
 
 def GpuTilingAndFusion : Pass<"gpu-tiling", "func::FuncOp"> {
   let summary = "GPU tiling and fusion path.";
@@ -185,7 +189,6 @@ def GpuXeVMAttachTarget: Pass<"xevm-attach-target", ""> {
   ];
 }
 
-#endif // GC_USE_IMEX
 
 def IterativeTilingAndFusion : Pass<"iterative-tiling-and-fusion",
                                         "func::FuncOp"> {

lib/gc/CAPI/CMakeLists.txt

@@ -4,9 +4,7 @@ set(GC_ALL_LIBS
   GcAnalysis
   MLIRCPURuntimeTransforms)
 
-if(GC_ENABLE_IMEX)
-  list(APPEND GC_ALL_LIBS GcGpuPasses)
-endif()
+list(APPEND GC_ALL_LIBS GcGpuPasses)
 
 add_mlir_public_c_api_library(GcCAPI
   Dialects.cpp

lib/gc/ExecutionEngine/CMakeLists.txt

@@ -1,6 +1,6 @@
 add_subdirectory(CPURuntime)
 add_subdirectory(Driver)
-if(GC_ENABLE_IMEX)
+if(GC_ENABLE_GPU)
     add_subdirectory(GPURuntime)
     add_subdirectory(OpenCLRuntime)
 endif()

lib/gc/ExecutionEngine/Driver/CMakeLists.txt

@@ -27,10 +27,7 @@ else()
 endif()
 get_property(extension_libs GLOBAL PROPERTY MLIR_EXTENSION_LIBS)
 
-set(GC_PASSES GcInterface GcPasses)
-if(GC_ENABLE_IMEX)
-  list(APPEND GC_PASSES GcGpuPasses)
-endif()
+set(GC_PASSES GcInterface GcPasses GcGpuPasses)
 
 gc_add_mlir_library(GcJitWrapper
   Driver.cpp

lib/gc/Transforms/CMakeLists.txt

@@ -40,8 +40,6 @@ gc_add_mlir_library(GcPasses
     MLIRMicrokernelTransforms
   )
 
-if(GC_ENABLE_IMEX)
-  add_subdirectory(GPU)
-endif()
+add_subdirectory(GPU)
 
 add_subdirectory(Microkernel)