From: Aart Bik <ajcbik@google.com>
Date: Tue, 16 May 2023 17:00:36 +0000 (-0700)
Subject: [mlir][sparse][gpu] set cubin flag when building for cuda
X-Git-Tag: upstream/17.0.6~8290
X-Git-Url: http://review.tizen.org/git/?a=commitdiff_plain;h=195621aacb96a4e2f49f5d888e2fc7af20e6e797;p=platform%2Fupstream%2Fllvm.git

[mlir][sparse][gpu] set cubin flag when building for cuda

Reviewed By: Peiming

Differential Revision: https://reviews.llvm.org/D150692
---

diff --git a/mlir/lib/Dialect/SparseTensor/Pipelines/CMakeLists.txt b/mlir/lib/Dialect/SparseTensor/Pipelines/CMakeLists.txt
index 234a0d8..939d748 100644
--- a/mlir/lib/Dialect/SparseTensor/Pipelines/CMakeLists.txt
+++ b/mlir/lib/Dialect/SparseTensor/Pipelines/CMakeLists.txt
@@ -27,3 +27,11 @@ 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()