[Openmp-commits] [openmp] f7c9299 - [OpenMP] Fix CUDA plugin build after 3817ba13aea3.

[via Openmp-commits]

Author: Abhinav Gaba
Date: 2021-07-23T16:50:00+08:00
New Revision: f7c92995c0e1f95ef5640a0df325e05d12de4658

URL: https://github.com/llvm/llvm-project/commit/f7c92995c0e1f95ef5640a0df325e05d12de4658
DIFF: https://github.com/llvm/llvm-project/commit/f7c92995c0e1f95ef5640a0df325e05d12de4658.diff

LOG: [OpenMP] Fix CUDA plugin build after 3817ba13aea3.

The build was broken on machines that don't have Cuda SDK installed.

See https://reviews.llvm.org/D106627 for the original discussion.

Added: 
    

Modified: 
    openmp/libomptarget/plugins/cuda/dynamic_cuda/cuda.cpp
    openmp/libomptarget/plugins/cuda/dynamic_cuda/cuda.h

Removed: 
    


################################################################################
diff  --git a/openmp/libomptarget/plugins/cuda/dynamic_cuda/cuda.cpp b/openmp/libomptarget/plugins/cuda/dynamic_cuda/cuda.cpp
index c84b3814065e8..235efd2728ded 100644
--- a/openmp/libomptarget/plugins/cuda/dynamic_cuda/cuda.cpp
+++ b/openmp/libomptarget/plugins/cuda/dynamic_cuda/cuda.cpp
@@ -61,6 +61,9 @@ DLWRAP(cuDeviceCanAccessPeer, 3);
 DLWRAP(cuCtxEnablePeerAccess, 2);
 DLWRAP(cuMemcpyPeerAsync, 6);
 
+DLWRAP(cuCtxGetLimit, 2);
+DLWRAP(cuCtxSetLimit, 2);
+
 DLWRAP_FINALIZE();
 
 #ifndef DYNAMIC_CUDA_PATH

diff  --git a/openmp/libomptarget/plugins/cuda/dynamic_cuda/cuda.h b/openmp/libomptarget/plugins/cuda/dynamic_cuda/cuda.h
index 045c39cacc97e..17aa2a12ef6c3 100644
--- a/openmp/libomptarget/plugins/cuda/dynamic_cuda/cuda.h
+++ b/openmp/libomptarget/plugins/cuda/dynamic_cuda/cuda.h
@@ -34,6 +34,17 @@ typedef enum CUstream_flags_enum {
   CU_STREAM_NON_BLOCKING = 0x1,
 } CUstream_flags;
 
+typedef enum CUlimit_enum {
+  CU_LIMIT_STACK_SIZE = 0x0,
+  CU_LIMIT_PRINTF_FIFO_SIZE = 0x1,
+  CU_LIMIT_MALLOC_HEAP_SIZE = 0x2,
+  CU_LIMIT_DEV_RUNTIME_SYNC_DEPTH = 0x3,
+  CU_LIMIT_DEV_RUNTIME_PENDING_LAUNCH_COUNT = 0x4,
+  CU_LIMIT_MAX_L2_FETCH_GRANULARITY = 0x5,
+  CU_LIMIT_PERSISTING_L2_CACHE_SIZE = 0x6,
+  CU_LIMIT_MAX
+} CUlimit;
+
 typedef enum CUdevice_attribute_enum {
   CU_DEVICE_ATTRIBUTE_MAX_BLOCK_DIM_X = 2,
   CU_DEVICE_ATTRIBUTE_MAX_GRID_DIM_X = 5,
@@ -100,4 +111,7 @@ CUresult cuCtxEnablePeerAccess(CUcontext, unsigned);
 CUresult cuMemcpyPeerAsync(CUdeviceptr, CUcontext, CUdeviceptr, CUcontext,
                            size_t, CUstream);
 
+CUresult cuCtxGetLimit(size_t *, CUlimit);
+CUresult cuCtxSetLimit(CUlimit, size_t);
+
 #endif