feat: add new functions of CUDA 11.5.1

Signed-off-by: thomassong <[email protected]>

Author: mYmNeo

find_new_lib.sh

@@ -0,0 +1,20 @@
+#!/usr/bin/env bash
+
+set -o errexit
+set -o nounset
+set -o pipefail
+
+CUDA_LIBRARY=$1
+ML_LIBRARY=$2
+
+echo "find new library"
+
+while read item; do
+  grep -q ${item} include/cuda-helper.h || echo "$item,"
+done < <(nm -D ${CUDA_LIBRARY} | grep " T " | awk '{print "CUDA_ENTRY_ENUM("$3")"}')
+
+echo ""
+
+while read item; do
+  grep -q ${item} include/nvml-helper.h || echo "$item,"
+done < <(nm -D ${ML_LIBRARY} | grep " T " | awk '{print "NVML_ENTRY_ENUM("$3")"}')

include/cuda-helper.h

@@ -1078,6 +1078,30 @@ typedef enum {
   CUDA_ENTRY_ENUM(cuMemPoolTrimTo),
   /** cuMipmappedArrayGetSparseProperties */
   CUDA_ENTRY_ENUM(cuMipmappedArrayGetSparseProperties),
+  CUDA_ENTRY_ENUM(cuCtxCreate_v3),
+  CUDA_ENTRY_ENUM(cuCtxGetExecAffinity),
+  CUDA_ENTRY_ENUM(cuDeviceGetExecAffinitySupport),
+  CUDA_ENTRY_ENUM(cuDeviceGetGraphMemAttribute),
+  CUDA_ENTRY_ENUM(cuDeviceGetUuid_v2),
+  CUDA_ENTRY_ENUM(cuDeviceGraphMemTrim),
+  CUDA_ENTRY_ENUM(cuDeviceSetGraphMemAttribute),
+  CUDA_ENTRY_ENUM(cuFlushGPUDirectRDMAWrites),
+  CUDA_ENTRY_ENUM(cuGetProcAddress),
+  CUDA_ENTRY_ENUM(cuGraphAddMemAllocNode),
+  CUDA_ENTRY_ENUM(cuGraphAddMemFreeNode),
+  CUDA_ENTRY_ENUM(cuGraphDebugDotPrint),
+  CUDA_ENTRY_ENUM(cuGraphInstantiateWithFlags),
+  CUDA_ENTRY_ENUM(cuGraphMemAllocNodeGetParams),
+  CUDA_ENTRY_ENUM(cuGraphMemFreeNodeGetParams),
+  CUDA_ENTRY_ENUM(cuGraphReleaseUserObject),
+  CUDA_ENTRY_ENUM(cuGraphRetainUserObject),
+  CUDA_ENTRY_ENUM(cuStreamGetCaptureInfo_v2),
+  CUDA_ENTRY_ENUM(cuStreamGetCaptureInfo_v2_ptsz),
+  CUDA_ENTRY_ENUM(cuStreamUpdateCaptureDependencies),
+  CUDA_ENTRY_ENUM(cuStreamUpdateCaptureDependencies_ptsz),
+  CUDA_ENTRY_ENUM(cuUserObjectCreate),
+  CUDA_ENTRY_ENUM(cuUserObjectRelease),
+  CUDA_ENTRY_ENUM(cuUserObjectRetain),
   CUDA_ENTRY_END
 } cuda_entry_enum_t;
 

include/cuda-subset.h

@@ -2262,6 +2262,105 @@ typedef enum CUmemPool_attribute_enum {
   CU_MEMPOOL_ATTR_RELEASE_THRESHOLD
 } CUmemPool_attribute;
 
+/**
+ * Execution Affinity Types
+ */
+typedef enum CUexecAffinityType_enum {
+  CU_EXEC_AFFINITY_TYPE_SM_COUNT = 0, /**< Create a context with limited SMs. */
+  CU_EXEC_AFFINITY_TYPE_MAX
+} CUexecAffinityType;
+
+/**
+ * Value for ::CU_EXEC_AFFINITY_TYPE_SM_COUNT
+ */
+typedef struct CUexecAffinitySmCount_st {
+  unsigned int val; /**< The number of SMs the context is limited to use. */
+} CUexecAffinitySmCount_v1;
+typedef CUexecAffinitySmCount_v1 CUexecAffinitySmCount;
+
+/**
+ * Execution Affinity Parameters
+ */
+typedef struct CUexecAffinityParam_st {
+  CUexecAffinityType type;
+  union {
+    CUexecAffinitySmCount
+        smCount; /** Value for ::CU_EXEC_AFFINITY_TYPE_SM_COUNT */
+  } param;
+} CUexecAffinityParam_v1;
+typedef CUexecAffinityParam_v1 CUexecAffinityParam;
+
+typedef enum CUgraphMem_attribute_enum {
+  /**
+   * (value type = cuuint64_t)
+   * Amount of memory, in bytes, currently associated with graphs
+   */
+  CU_GRAPH_MEM_ATTR_USED_MEM_CURRENT,
+
+  /**
+   * (value type = cuuint64_t)
+   * High watermark of memory, in bytes, associated with graphs since the
+   * last time it was reset.  High watermark can only be reset to zero.
+   */
+  CU_GRAPH_MEM_ATTR_USED_MEM_HIGH,
+
+  /**
+   * (value type = cuuint64_t)
+   * Amount of memory, in bytes, currently allocated for use by
+   * the CUDA graphs asynchronous allocator.
+   */
+  CU_GRAPH_MEM_ATTR_RESERVED_MEM_CURRENT,
+
+  /**
+   * (value type = cuuint64_t)
+   * High watermark of memory, in bytes, currently allocated for use by
+   * the CUDA graphs asynchronous allocator.
+   */
+  CU_GRAPH_MEM_ATTR_RESERVED_MEM_HIGH
+} CUgraphMem_attribute;
+
+/**
+ * Memory allocation node parameters
+ */
+typedef struct CUDA_MEM_ALLOC_NODE_PARAMS_st {
+  /**
+   * in: location where the allocation should reside (specified in ::location).
+   * ::handleTypes must be ::CU_MEM_HANDLE_TYPE_NONE. IPC is not supported.
+   */
+  CUmemPoolProps poolProps;
+  const CUmemAccessDesc
+      *accessDescs;       /**< in: array of memory access descriptors. Used to
+                             describe peer GPU access */
+  size_t accessDescCount; /**< in: number of memory access descriptors.  Must
+                             not exceed the number of GPUs. */
+  size_t bytesize;        /**< in: size in bytes of the requested allocation */
+  CUdeviceptr dptr; /**< out: address of the allocation returned by CUDA */
+} CUDA_MEM_ALLOC_NODE_PARAMS;
+
+typedef struct CUuserObject_st
+    *CUuserObject; /**< CUDA user object for graphs */
+
+/**
+ * The targets for ::cuFlushGPUDirectRDMAWrites
+ */
+typedef enum CUflushGPUDirectRDMAWritesTarget_enum {
+  CU_FLUSH_GPU_DIRECT_RDMA_WRITES_TARGET_CURRENT_CTX =
+      0 /**< Sets the target for ::cuFlushGPUDirectRDMAWrites() to the currently
+           active CUDA device context. */
+} CUflushGPUDirectRDMAWritesTarget;
+
+/**
+ * The scopes for ::cuFlushGPUDirectRDMAWrites
+ */
+typedef enum CUflushGPUDirectRDMAWritesScope_enum {
+  CU_FLUSH_GPU_DIRECT_RDMA_WRITES_TO_OWNER =
+      100, /**< Blocks until remote writes are visible to the CUDA device
+              context owning the data. */
+  CU_FLUSH_GPU_DIRECT_RDMA_WRITES_TO_ALL_DEVICES =
+      200 /**< Blocks until remote writes are visible to all CUDA device
+             contexts. */
+} CUflushGPUDirectRDMAWritesScope;
+
 #ifdef __cplusplus
 }
 #endif

include/nvml-helper.h

@@ -540,6 +540,19 @@ typedef enum {
   NVML_ENTRY_ENUM(nvmlVgpuInstanceGetGpuInstanceId),
   /** nvmlVgpuTypeGetGpuInstanceProfileId */
   NVML_ENTRY_ENUM(nvmlVgpuTypeGetGpuInstanceProfileId),
+  NVML_ENTRY_ENUM(nvmlDeviceCreateGpuInstanceWithPlacement),
+  NVML_ENTRY_ENUM(nvmlDeviceGetBusType),
+  NVML_ENTRY_ENUM(nvmlDeviceGetClkMonStatus),
+  NVML_ENTRY_ENUM(nvmlDeviceGetGpuInstancePossiblePlacements_v2),
+  NVML_ENTRY_ENUM(nvmlDeviceGetGridLicensableFeatures_v4),
+  NVML_ENTRY_ENUM(nvmlDeviceGetIrqNum),
+  NVML_ENTRY_ENUM(nvmlDeviceGetMPSComputeRunningProcesses_v2),
+  NVML_ENTRY_ENUM(nvmlDeviceGetNvLinkRemoteDeviceType),
+  NVML_ENTRY_ENUM(nvmlDeviceResetMemoryLockedClocks),
+  NVML_ENTRY_ENUM(nvmlDeviceSetMemoryLockedClocks),
+  NVML_ENTRY_ENUM(nvmlGetExcludedDeviceCount),
+  NVML_ENTRY_ENUM(nvmlGetExcludedDeviceInfoByIndex),
+  NVML_ENTRY_ENUM(nvmlVgpuInstanceGetLicenseInfo),
   NVML_ENTRY_END
 } nvml_entry_enum_t;