[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:

Yeah, I changed exactly that line to

--sparse-compiler="enable-runtime-library=false parallelization-strategy=dense-outer-loop gpu-triple=nvptx64-nvidia-cuda gpu-chip=sm_50 gpu-features=+ptx60"

to make the test pass. The "problem" is that I lazily put *all* the tests under the mlir_run_cuda_sm80_tests flag, since A100s are sort of the "minimum" GPUs we are interested in. But that gave the pass-before/fail-after behavior for me (which was not really a fail since I knew my desktop was not up-to-standard). So all is good!

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