[hip] refactoring cooperative kernel launch APIs (#1737)

This PR is a follow-up on PR# #1698 and it makes two more APIs (hipLaunchCooperativeKernel/hipLaunchCooperativeKernelMultiDevice) inline so that they can work correctly with lazy binding.

hipamd/include/hip/hcc_detail/functional_grid_launch.hpp

@@ -37,6 +37,15 @@ THE SOFTWARE.
 hipError_t ihipExtLaunchMultiKernelMultiDevice(hipLaunchParams* launchParamsList, int numDevices,
                                                unsigned int flags, hip_impl::program_state& ps);
 
+hipError_t ihipLaunchCooperativeKernel(const void* f, dim3 gridDim, dim3 blockDimX, void** kernelParams,
+                unsigned int sharedMemBytes, hipStream_t stream, hip_impl::program_state& ps);
+
+hipError_t ihipLaunchCooperativeKernelMultiDevice(hipLaunchParams* launchParamsList, int  numDevices,
+                unsigned int  flags, hip_impl::program_state& ps);
+
+
+
+
 #pragma GCC visibility push(hidden)
 
 namespace hip_impl {
@@ -190,4 +199,25 @@ hipError_t hipExtLaunchMultiKernelMultiDevice(hipLaunchParams* launchParamsList,
 
 }
 
+template <typename F>
+inline
+__attribute__((visibility("hidden")))
+hipError_t hipLaunchCooperativeKernel(F f, dim3 gridDim, dim3 blockDimX, void** kernelParams,
+                unsigned int sharedMemBytes, hipStream_t stream) {
+
+    hip_impl::hip_init();
+    auto& ps = hip_impl::get_program_state();
+    return ihipLaunchCooperativeKernel(reinterpret_cast<void*>(f), gridDim, blockDimX, kernelParams, sharedMemBytes, stream, ps);
+}
+
+inline
+__attribute__((visibility("hidden")))
+hipError_t hipLaunchCooperativeKernelMultiDevice(hipLaunchParams* launchParamsList, int  numDevices,
+                unsigned int  flags) {
+
+    hip_impl::hip_init();
+    auto& ps = hip_impl::get_program_state();
+    return ihipLaunchCooperativeKernelMultiDevice(launchParamsList, numDevices, flags, ps);
+}
+
 #pragma GCC visibility pop

hipamd/include/hip/hcc_detail/hip_runtime_api.h

@@ -2889,6 +2889,7 @@ hipError_t hipModuleLaunchKernel(hipFunction_t f, unsigned int gridDimX, unsigne
                                  void** kernelParams, void** extra);
 
 
+#if __HIP_VDI__ && !defined(__HCC__)
 /**
  * @brief launches kernel f with launch parameters and shared memory on stream with arguments passed
  * to kernelparams or extra, where thread blocks can cooperate and synchronize as they execute
@@ -2921,6 +2922,8 @@ hipError_t hipLaunchCooperativeKernel(const void* f, dim3 gridDim, dim3 blockDim
 hipError_t hipLaunchCooperativeKernelMultiDevice(hipLaunchParams* launchParamsList,
                                                  int  numDevices, unsigned int  flags);
 
+#endif
+
 /**
  * @brief determine the grid and block sizes to achieves maximum occupancy for a kernel
  *
@@ -3382,7 +3385,7 @@ hipError_t hipBindTextureToMipmappedArray(const texture<T, dim, readMode>& tex,
     return hipSuccess;
 }
 
-
+#if __HIP_VDI__ && !defined(__HCC__)
 template <class T>
 inline hipError_t hipLaunchCooperativeKernel(T f, dim3 gridDim, dim3 blockDim,
                                              void** kernelParams, unsigned int sharedMemBytes, hipStream_t stream) {
@@ -3396,7 +3399,7 @@ inline hipError_t hipLaunchCooperativeKernelMultiDevice(hipLaunchParams* launchP
     return hipLaunchCooperativeKernelMultiDevice(launchParamsList, numDevices, flags);
 }
 
-#if defined(__clang__) && defined(__HIP__)
+
 template <class T>
 inline hipError_t hipExtLaunchMultiKernelMultiDevice(hipLaunchParams* launchParamsList,
                                                      unsigned int  numDevices, unsigned int  flags = 0) {