[Mlir-commits] [mlir] [MLIR][CUDA] Update export macros in CudaRuntimeWrappers (PR #73932)

[Adam Paszke]

https://github.com/apaszke created https://github.com/llvm/llvm-project/pull/73932

This fixes a few issues present in the current version: 1) The macro doesn't enforce the default visibility on exported
   functions, causing compilation to fail when using
   `-fvisibility=hidden`
2) Not all functions are exported
3) Sometimes the macro ended up weirdly interleaved with `extern "C"`
   declarations

>From 9108ec89e367ca10603a567b08cbf3b198b429b7 Mon Sep 17 00:00:00 2001
From: Adam Paszke <apaszke at google.com>
Date: Thu, 30 Nov 2023 13:01:15 +0000
Subject: [PATCH] [MLIR][CUDA] Update export macros in CudaRuntimeWrappers

This fixes a few issues present in the current version:
1) The macro doesn't enforce the default visibility on exported
   functions, causing compilation to fail when using
   `-fvisibility=hidden`
2) Not all functions are exported
3) Sometimes the macro ended up weirdly interleaved with `extern "C"`
   declarations
---
 .../ExecutionEngine/CudaRuntimeWrappers.cpp   | 25 ++++++++++---------
 1 file changed, 13 insertions(+), 12 deletions(-)

diff --git a/mlir/lib/ExecutionEngine/CudaRuntimeWrappers.cpp b/mlir/lib/ExecutionEngine/CudaRuntimeWrappers.cpp
index d453b8ffe422049..b8ac9ab90a9f3b8 100644
--- a/mlir/lib/ExecutionEngine/CudaRuntimeWrappers.cpp
+++ b/mlir/lib/ExecutionEngine/CudaRuntimeWrappers.cpp
@@ -30,7 +30,7 @@
 #ifdef _WIN32
 #define MLIR_CUDA_WRAPPERS_EXPORT __declspec(dllexport)
 #else
-#define MLIR_CUDA_WRAPPERS_EXPORT
+#define MLIR_CUDA_WRAPPERS_EXPORT __attribute__((visibility("default")))
 #endif // _WIN32
 
 #define CUDA_REPORT_IF_ERROR(expr)                                             \
@@ -226,17 +226,17 @@ extern "C" MLIR_CUDA_WRAPPERS_EXPORT void mgpuEventDestroy(CUevent event) {
   CUDA_REPORT_IF_ERROR(cuEventDestroy(event));
 }
 
-extern MLIR_CUDA_WRAPPERS_EXPORT "C" void mgpuEventSynchronize(CUevent event) {
+extern "C" MLIR_CUDA_WRAPPERS_EXPORT void mgpuEventSynchronize(CUevent event) {
   CUDA_REPORT_IF_ERROR(cuEventSynchronize(event));
 }
 
-extern MLIR_CUDA_WRAPPERS_EXPORT "C" void mgpuEventRecord(CUevent event,
+extern "C" MLIR_CUDA_WRAPPERS_EXPORT void mgpuEventRecord(CUevent event,
                                                           CUstream stream) {
   CUDA_REPORT_IF_ERROR(cuEventRecord(event, stream));
 }
 
-extern "C" void *mgpuMemAlloc(uint64_t sizeBytes, CUstream /*stream*/,
-                              bool /*isHostShared*/) {
+extern "C" MLIR_CUDA_WRAPPERS_EXPORT void *
+mgpuMemAlloc(uint64_t sizeBytes, CUstream /*stream*/, bool /*isHostShared*/) {
   ScopedContext scopedContext;
   CUdeviceptr ptr = 0;
   if (sizeBytes != 0)
@@ -244,25 +244,26 @@ extern "C" void *mgpuMemAlloc(uint64_t sizeBytes, CUstream /*stream*/,
   return reinterpret_cast<void *>(ptr);
 }
 
-extern "C" void mgpuMemFree(void *ptr, CUstream /*stream*/) {
+extern "C" MLIR_CUDA_WRAPPERS_EXPORT void mgpuMemFree(void *ptr,
+                                                      CUstream /*stream*/) {
   CUDA_REPORT_IF_ERROR(cuMemFree(reinterpret_cast<CUdeviceptr>(ptr)));
 }
 
-extern "C" void mgpuMemcpy(void *dst, void *src, size_t sizeBytes,
-                           CUstream stream) {
+extern "C" MLIR_CUDA_WRAPPERS_EXPORT void
+mgpuMemcpy(void *dst, void *src, size_t sizeBytes, CUstream stream) {
   CUDA_REPORT_IF_ERROR(cuMemcpyAsync(reinterpret_cast<CUdeviceptr>(dst),
                                      reinterpret_cast<CUdeviceptr>(src),
                                      sizeBytes, stream));
 }
 
-extern "C" void mgpuMemset32(void *dst, unsigned int value, size_t count,
-                             CUstream stream) {
+extern "C" MLIR_CUDA_WRAPPERS_EXPORT void
+mgpuMemset32(void *dst, unsigned int value, size_t count, CUstream stream) {
   CUDA_REPORT_IF_ERROR(cuMemsetD32Async(reinterpret_cast<CUdeviceptr>(dst),
                                         value, count, stream));
 }
 
-extern "C" void mgpuMemset16(void *dst, unsigned short value, size_t count,
-                             CUstream stream) {
+extern "C" MLIR_CUDA_WRAPPERS_EXPORT void
+mgpuMemset16(void *dst, unsigned short value, size_t count, CUstream stream) {
   CUDA_REPORT_IF_ERROR(cuMemsetD16Async(reinterpret_cast<CUdeviceptr>(dst),
                                         value, count, stream));
 }