Merge pull request #350 from YonatanNachum/cuda_13_fix

Fix use of cuCtxCreate with Cuda 13.0 and newer

Author: sshaulnv

src/cuda_loader.c

@@ -10,7 +10,11 @@ CUresult (*p_cuDeviceGetCount)(int *) = NULL;
 CUresult (*p_cuDeviceGet)(CUdevice *, int) = NULL;
 CUresult (*p_cuDeviceGetAttribute)(int *, CUdevice_attribute, CUdevice) = NULL;
 CUresult (*p_cuDeviceGetName)(char *, int, CUdevice) = NULL;
+#if CUDA_VER >= 13000
+CUresult (*p_cuCtxCreate)(CUcontext *, CUctxCreateParams *, unsigned int, CUdevice) = NULL;
+#else
 CUresult (*p_cuCtxCreate)(CUcontext *, unsigned int, CUdevice) = NULL;
+#endif
 CUresult (*p_cuDevicePrimaryCtxRetain)(CUcontext *, CUdevice) = NULL;
 CUresult (*p_cuCtxSetCurrent)(CUcontext) = NULL;
 CUresult (*p_cuCtxDestroy)(CUcontext) = NULL;

src/cuda_loader.h

@@ -30,7 +30,11 @@ extern CUresult (*p_cuDeviceGetCount)(int *);
 extern CUresult (*p_cuDeviceGet)(CUdevice *, int);
 extern CUresult (*p_cuDeviceGetAttribute)(int *, CUdevice_attribute, CUdevice);
 extern CUresult (*p_cuDeviceGetName)(char *, int, CUdevice);
+#if CUDA_VER >= 13000
+extern CUresult (*p_cuCtxCreate)(CUcontext *, CUctxCreateParams *, unsigned int, CUdevice);
+#else
 extern CUresult (*p_cuCtxCreate)(CUcontext *, unsigned int, CUdevice);
+#endif
 extern CUresult (*p_cuDevicePrimaryCtxRetain)(CUcontext *, CUdevice);
 extern CUresult (*p_cuCtxSetCurrent)(CUcontext);
 extern CUresult (*p_cuCtxDestroy)(CUcontext);
@@ -61,4 +65,4 @@ void unload_cuda_library(void);
 }
 #endif
 
-#endif // CUDA_LOADER_H
+#endif // CUDA_LOADER_H

src/cuda_memory.c

@@ -1,6 +1,6 @@
 /* SPDX-License-Identifier: GPL-2.0 OR BSD-2-Clause */
 /*
- * Copyright 2023 Amazon.com, Inc. or its affiliates. All rights reserved.
+ * Copyright 2023-2025 Amazon.com, Inc. or its affiliates. All rights reserved.
  */
 
 #include <stdio.h>
@@ -96,7 +96,11 @@ static int init_gpu(struct cuda_memory_ctx *ctx)
 printf("creating CUDA Ctx\n");
 
 /* Create context */
+#if CUDA_VER >= 13000
+error = p_cuCtxCreate(&ctx->cuContext, NULL, CU_CTX_MAP_HOST, ctx->cuDevice);
+#else
 error = p_cuCtxCreate(&ctx->cuContext, CU_CTX_MAP_HOST, ctx->cuDevice);
+#endif
 if (error != CUDA_SUCCESS) {
 printf("cuCtxCreate() error=%d\n", error);
 return FAILURE;