[mlir][gpu] Deprecate gpu::Serialization* passes. (#65857)

Deprecate the `gpu-to-cubin` & `gpu-to-hsaco` passes in favor of the
`TargetAttr` workflow. This patch removes remaining upstream uses of the
aforementioned passes, including the option to use them in `mlir-opt`. A
future patch will remove these passes entirely.

The passes can be re-enabled in `mlir-opt` by adding the CMake flag: `-DMLIR_ENABLE_DEPRECATED_GPU_SERIALIZATION=1`.

Author: fabianmcg

mlir/CMakeLists.txt

@@ -114,6 +114,7 @@ else()
 endif()
 add_definitions(-DMLIR_ROCM_CONVERSIONS_ENABLED=${MLIR_ENABLE_ROCM_CONVERSIONS})
 
+set(MLIR_ENABLE_DEPRECATED_GPU_SERIALIZATION 0 CACHE BOOL "Enable deprecated GPU serialization passes")
 set(MLIR_ENABLE_CUDA_RUNNER 0 CACHE BOOL "Enable building the mlir CUDA runner")
 set(MLIR_ENABLE_ROCM_RUNNER 0 CACHE BOOL "Enable building the mlir ROCm runner")
 set(MLIR_ENABLE_SPIRV_CPU_RUNNER 0 CACHE BOOL "Enable building the mlir SPIR-V cpu runner")

mlir/include/mlir/Dialect/GPU/Transforms/Passes.h

@@ -134,14 +134,17 @@ class SerializeToBlobPass : public OperationPass<gpu::GPUModuleOp> {
 
 /// Register pass to serialize GPU kernel functions to a CUBIN binary
 /// annotation.
+LLVM_DEPRECATED("use Target attributes instead", "")
 void registerGpuSerializeToCubinPass();
 
 /// Register pass to serialize GPU kernel functions to a HSAco binary
 /// annotation.
+LLVM_DEPRECATED("use Target attributes instead", "")
 void registerGpuSerializeToHsacoPass();
 
 /// Create an instance of the GPU kernel function to CUBIN binary serialization
 /// pass with optLevel (default level 2).
+LLVM_DEPRECATED("use Target attributes instead", "")
 std::unique_ptr<Pass> createGpuSerializeToCubinPass(StringRef triple,
                                                     StringRef chip,
                                                     StringRef features,
@@ -150,6 +153,7 @@ std::unique_ptr<Pass> createGpuSerializeToCubinPass(StringRef triple,
 
 /// Create an instance of the GPU kernel function to HSAco binary serialization
 /// pass.
+LLVM_DEPRECATED("use Target attributes instead", "")
 std::unique_ptr<Pass> createGpuSerializeToHsacoPass(StringRef triple,
                                                     StringRef arch,
                                                     StringRef features,

mlir/include/mlir/InitAllPasses.h

@@ -65,8 +65,6 @@ inline void registerAllPasses() {
   bufferization::registerBufferizationPasses();
   func::registerFuncPasses();
   registerGPUPasses();
-  registerGpuSerializeToCubinPass();
-  registerGpuSerializeToHsacoPass();
   registerLinalgPasses();
   registerNVGPUPasses();
   registerSparseTensorPasses();

mlir/lib/Dialect/SparseTensor/Pipelines/CMakeLists.txt

@@ -27,11 +27,3 @@ add_mlir_dialect_library(MLIRSparseTensorPipelines
   MLIRVectorToLLVM
   MLIRVectorTransforms
 )
-
-if(MLIR_ENABLE_CUDA_RUNNER)
-  # Enable gpu-to-cubin pass.
-  target_compile_definitions(obj.MLIRSparseTensorPipelines
-    PRIVATE
-    MLIR_GPU_TO_CUBIN_PASS_ENABLE=1
-  )
-endif()

mlir/lib/Dialect/SparseTensor/Pipelines/SparseTensorPipelines.cpp

@@ -78,11 +78,13 @@ void mlir::sparse_tensor::buildSparseCompiler(
 
   // Finalize GPU code generation.
   if (gpuCodegen) {
-#if MLIR_GPU_TO_CUBIN_PASS_ENABLE
-    pm.addNestedPass<gpu::GPUModuleOp>(createGpuSerializeToCubinPass(
-        options.gpuTriple, options.gpuChip, options.gpuFeatures));
-#endif
+    GpuNVVMAttachTargetOptions nvvmTargetOptions;
+    nvvmTargetOptions.triple = options.gpuTriple;
+    nvvmTargetOptions.chip = options.gpuChip;
+    nvvmTargetOptions.features = options.gpuFeatures;
+    pm.addPass(createGpuNVVMAttachTarget(nvvmTargetOptions));
     pm.addPass(createGpuToLLVMConversionPass());
+    pm.addPass(createGpuModuleToBinaryPass());
   }
 
   pm.addPass(createReconcileUnrealizedCastsPass());

mlir/test/Conversion/GPUToCUDA/lit.local.cfg

@@ -31,8 +31,6 @@ set(LIBS
   )
 
 add_mlir_library(MLIRGPUTestPasses
-  TestConvertGPUKernelToCubin.cpp
-  TestConvertGPUKernelToHsaco.cpp
   TestGpuMemoryPromotion.cpp
   TestGpuRewrite.cpp
   TestLowerToNVVM.cpp
@@ -43,12 +41,3 @@ add_mlir_library(MLIRGPUTestPasses
   ${LIBS}
   )
 
-# This is how it is defined in mlir/lib/Dialect/GPU/CMakeLists.txt
-# We probably want something better project-wide
-if(MLIR_ENABLE_CUDA_RUNNER)
-  # Enable gpu-to-cubin pass.
-  target_compile_definitions(MLIRGPUTestPasses
-    PRIVATE
-    MLIR_GPU_TO_CUBIN_PASS_ENABLE=1
-  )
-endif()