[Mlir-commits] [mlir] [mlir][gpu] Deprecate gpu::Serialization* passes. (PR #65857)

[Aart Bik]

================
@@ -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());
----------------
aartbik wrote:

I think these are currently failing on main branch in OSS:

  MLIR :: Integration/Dialect/SparseTensor/GPU/CUDA/sparse-matvec-const.mlir
  MLIR :: Integration/Dialect/SparseTensor/GPU/CUDA/sparse-matvec.mlir

with

'cuModuleLoadData(&module, data)' failed with 'CUDA_ERROR_NO_BINARY_FOR_GPU'





https://github.com/llvm/llvm-project/pull/65857