[flang][cuda] Fix signatures after argument change (#138081)

Author: clementval

flang-rt/lib/cuda/allocatable.cpp

@@ -23,7 +23,7 @@ namespace Fortran::runtime::cuda {
 extern "C" {
 RT_EXT_API_GROUP_BEGIN
 
-int RTDEF(CUFAllocatableAllocateSync)(Descriptor &desc, int64_t stream,
+int RTDEF(CUFAllocatableAllocateSync)(Descriptor &desc, int64_t *stream,
     bool *pinned, bool hasStat, const Descriptor *errMsg,
     const char *sourceFile, int sourceLine) {
   int stat{RTNAME(CUFAllocatableAllocate)(
@@ -41,7 +41,7 @@ int RTDEF(CUFAllocatableAllocateSync)(Descriptor &desc, int64_t stream,
   return stat;
 }
 
-int RTDEF(CUFAllocatableAllocate)(Descriptor &desc, int64_t stream,
+int RTDEF(CUFAllocatableAllocate)(Descriptor &desc, int64_t *stream,
     bool *pinned, bool hasStat, const Descriptor *errMsg,
     const char *sourceFile, int sourceLine) {
   if (desc.HasAddendum()) {
@@ -63,7 +63,7 @@ int RTDEF(CUFAllocatableAllocate)(Descriptor &desc, int64_t stream,
 }
 
 int RTDEF(CUFAllocatableAllocateSource)(Descriptor &alloc,
-    const Descriptor &source, int64_t stream, bool *pinned, bool hasStat,
+    const Descriptor &source, int64_t *stream, bool *pinned, bool hasStat,
     const Descriptor *errMsg, const char *sourceFile, int sourceLine) {
   int stat{RTNAME(CUFAllocatableAllocate)(
       alloc, stream, pinned, hasStat, errMsg, sourceFile, sourceLine)};
@@ -76,7 +76,7 @@ int RTDEF(CUFAllocatableAllocateSource)(Descriptor &alloc,
 }
 
 int RTDEF(CUFAllocatableAllocateSourceSync)(Descriptor &alloc,
-    const Descriptor &source, int64_t stream, bool *pinned, bool hasStat,
+    const Descriptor &source, int64_t *stream, bool *pinned, bool hasStat,
     const Descriptor *errMsg, const char *sourceFile, int sourceLine) {
   int stat{RTNAME(CUFAllocatableAllocateSync)(
       alloc, stream, pinned, hasStat, errMsg, sourceFile, sourceLine)};

flang-rt/lib/cuda/allocator.cpp

@@ -98,15 +98,15 @@ static unsigned findAllocation(void *ptr) {
   return allocNotFound;
 }
 
-static void insertAllocation(void *ptr, std::size_t size, std::int64_t stream) {
+static void insertAllocation(void *ptr, std::size_t size, cudaStream_t stream) {
   CriticalSection critical{lock};
   initAllocations();
   if (numDeviceAllocations >= maxDeviceAllocations) {
     doubleAllocationArray();
   }
   deviceAllocations[numDeviceAllocations].ptr = ptr;
   deviceAllocations[numDeviceAllocations].size = size;
-  deviceAllocations[numDeviceAllocations].stream = (cudaStream_t)stream;
+  deviceAllocations[numDeviceAllocations].stream = stream;
   ++numDeviceAllocations;
   qsort(deviceAllocations, numDeviceAllocations, sizeof(DeviceAllocation),
       compareDeviceAlloc);

flang/include/flang/Runtime/CUDA/allocator.h

@@ -20,16 +20,16 @@ extern "C" {
 void RTDECL(CUFRegisterAllocator)();
 }
 
-void *CUFAllocPinned(std::size_t, std::int64_t);
+void *CUFAllocPinned(std::size_t, std::int64_t*);
 void CUFFreePinned(void *);
 
-void *CUFAllocDevice(std::size_t, std::int64_t);
+void *CUFAllocDevice(std::size_t, std::int64_t*);
 void CUFFreeDevice(void *);
 
-void *CUFAllocManaged(std::size_t, std::int64_t);
+void *CUFAllocManaged(std::size_t, std::int64_t*);
 void CUFFreeManaged(void *);
 
-void *CUFAllocUnified(std::size_t, std::int64_t);
+void *CUFAllocUnified(std::size_t, std::int64_t*);
 void CUFFreeUnified(void *);
 
 } // namespace Fortran::runtime::cuda