start drafting nv gpu mlir pipeline

Author: tkarna

src/jit/mlir.cpp

@@ -757,31 +757,47 @@ static const std::string gpu_pipeline =
     "func.func(convert-parallel-loops-to-gpu),"
     // insert-gpu-allocs pass can have client-api = opencl or vulkan args
     "func.func(insert-gpu-allocs{in-regions=1}),"
+    // ** imex GPU passes
+    // "drop-regions,"
+    // "canonicalize,"
+    // // "normalize-memrefs,"
+    // // "gpu-decompose-memrefs,"
+    // "func.func(lower-affine),"
+    // "gpu-kernel-outlining,"
+    // "canonicalize,"
+    // "cse,"
+    // // The following set-spirv-* passes can have client-api = opencl or
+    // vulkan
+    // // args
+    // "set-spirv-capabilities{client-api=opencl},"
+    // "gpu.module(set-spirv-abi-attrs{client-api=opencl}),"
+    // "canonicalize,"
+    // "fold-memref-alias-ops,"
+    // "imex-convert-gpu-to-spirv{enable-vc-intrinsic=1},"
+    // "spirv.module(spirv-lower-abi-attrs),"
+    // "spirv.module(spirv-update-vce),"
+    // // "func.func(llvm-request-c-wrappers),"
+    // "serialize-spirv,"
+    // "expand-strided-metadata,"
+    // "lower-affine,"
+    // "convert-gpu-to-gpux,"
+    // "convert-func-to-llvm,"
+    // "convert-math-to-llvm,"
+    // "convert-gpux-to-llvm,"
+    // "finalize-memref-to-llvm,"
+    // "reconcile-unrealized-casts";
+    // ** nv GPU passes
+    "func.func(insert-gpu-copy),"
     "drop-regions,"
     "canonicalize,"
-    // "normalize-memrefs,"
-    // "gpu-decompose-memrefs,"
-    "func.func(lower-affine),"
     "gpu-kernel-outlining,"
-    "canonicalize,"
-    "cse,"
-    // The following set-spirv-* passes can have client-api = opencl or vulkan
-    // args
-    "set-spirv-capabilities{client-api=opencl},"
-    "gpu.module(set-spirv-abi-attrs{client-api=opencl}),"
-    "canonicalize,"
-    "fold-memref-alias-ops,"
-    "imex-convert-gpu-to-spirv{enable-vc-intrinsic=1},"
-    "spirv.module(spirv-lower-abi-attrs),"
-    "spirv.module(spirv-update-vce),"
-    // "func.func(llvm-request-c-wrappers),"
-    "serialize-spirv,"
-    "expand-strided-metadata,"
-    "lower-affine,"
-    "convert-gpu-to-gpux,"
+    "gpu.module(strip-debuginfo,convert-gpu-to-nvvm),"
+    "nvvm-attach-target,"
+    "func.func(gpu-async-region),"
+    "gpu-to-llvm,"
+    "gpu-module-to-binary{format=fatbin},"
     "convert-func-to-llvm,"
     "convert-math-to-llvm,"
-    "convert-gpux-to-llvm,"
     "finalize-memref-to-llvm,"
     "reconcile-unrealized-casts";