[HIP] add cooperative kernel launch APIs on NVCC (#1929)

include/hip/nvcc_detail/hip_runtime_api.h

@@ -186,6 +186,7 @@ typedef struct cudaArray hipArray;
 typedef struct cudaArray* hipArray_t;
 typedef struct cudaArray* hipArray_const_t;
 typedef struct cudaFuncAttributes hipFuncAttributes;
+typedef struct cudaLaunchParams hipLaunchParams;
 #define hipFunction_attribute CUfunction_attribute
 #define hip_Memcpy2D CUDA_MEMCPY2D
 #define hipMemcpy3DParms cudaMemcpy3DParms
@@ -1275,6 +1276,12 @@ inline static hipError_t hipDeviceGetAttribute(int* pi, hipDeviceAttribute_t att
         case hipDeviceAttributeEccEnabled:
             cdattr = cudaDevAttrEccEnabled;
             break;
+        case hipDeviceAttributeCooperativeLaunch:
+            cdattr = cudaDevAttrCooperativeLaunch;
+            break;
+        case hipDeviceAttributeCooperativeMultiDeviceLaunch:
+            cdattr = cudaDevAttrCooperativeMultiDeviceLaunch;
+            break;
         default:
             return hipCUDAErrorTohipError(cudaErrorInvalidValue);
     }
@@ -1683,6 +1690,17 @@ inline static hipError_t hipGetChannelDesc(hipChannelFormatDesc* desc, hipArray_
     return hipCUDAErrorTohipError(cudaGetChannelDesc(desc,array));
 }
 
+inline static hipError_t hipLaunchCooperativeKernel(const void* f, dim3 gridDim, dim3 blockDim,
+                                      void** kernelParams, unsigned int sharedMemBytes,
+                                      hipStream_t stream) {
+    return hipCUDAErrorTohipError(
+            cudaLaunchCooperativeKernel(f, gridDim, blockDim, kernelParams, sharedMemBytes, stream));
+}
+
+inline static hipError_t hipLaunchCooperativeKernelMultiDevice(hipLaunchParams* launchParamsList,
+                                                 int  numDevices, unsigned int  flags) {
+    return hipCUDAErrorTohipError(cudaLaunchCooperativeKernelMultiDevice(launchParamsList, numDevices, flags));
+}
 
 #ifdef __cplusplus
 }
@@ -1746,6 +1764,14 @@ template <class T>
 inline static hipChannelFormatDesc hipCreateChannelDesc() {
     return cudaCreateChannelDesc<T>();
 }
+
+template <class T>
+inline static hipError_t hipLaunchCooperativeKernel(T f, dim3 gridDim, dim3 blockDim,
+                                             void** kernelParams, unsigned int sharedMemBytes, hipStream_t stream) {
+    return hipCUDAErrorTohipError(
+            cudaLaunchCooperativeKernel(f, gridDim, blockDim, kernelParams, sharedMemBytes, stream));
+}
+
 #endif  //__CUDACC__
 
 #endif  // HIP_INCLUDE_HIP_NVCC_DETAIL_HIP_RUNTIME_API_H