AI/rocm-systems
2
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Implement the hipOccupancyMaxPotentialBlockSize function (#1162)

Browse Source 
* Implement the hipOccupancyMaxPotentialBlockSize function

* Replaced hipGetDeviceProperties() call by ihipGetDeviceProperties() in ihipOccupancyMaxPotentialBlockSize()

* Add test for hipOccupancyMaxPotentialBlockSize in Module API

* Added extern declaration for ihipGetDeviceProperties() to be accessed inside ihipOccupancyMaxPotentialBlockSize()

* fixed hipOccupancyMaxPotentialBlockSize test build issue

* Fix hipOccupancyMaxPotentialBlockSize dtest

* Add BUILD_CMD in hipOccupancyMaxPotentialBlockSize dtest

* Revert "Add BUILD_CMD in hipOccupancyMaxPotentialBlockSize dtest"

This reverts commit 0480ff56f1441fc515d2c26ce33783e303423938.

* Disable hipOccupancyMaxPotentialBlockSize dtest on NVCC

* move extern declaration of ihipGetDeviceProperties to hip_module.cpp

* Update the limiation of 32 wavefronts per CU and 800/512 SGPRs for VI/pre-VI chips to calculate the occupancy
This commit is contained in:
wkwchau
2019-06-19 20:28:29 -04:00
committed by Maneesh Gupta
parent ba323cdef1
commit d492f1fd6b
4 changed files with 246 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files
include/hip/hcc_detail/functional_grid_launch.hpp
+16
View File
@@ -135,6 +135,22 @@ void hipLaunchKernelGGLImpl(
}
} // Namespace hip_impl.
template <typename F>
inline
void hipOccupancyMaxPotentialBlockSize(uint32_t* gridSize, uint32_t* blockSize,
F kernel, size_t dynSharedMemPerBlk, uint32_t blockSizeLimit) {
using namespace hip_impl;
hip_impl::hip_init();
auto f = get_program_state().kernel_descriptor(reinterpret_cast<std::uintptr_t>(kernel),
target_agent(0));
hipOccupancyMaxPotentialBlockSize(gridSize, blockSize, f,
dynSharedMemPerBlk, blockSizeLimit);
}
template <typename... Args, typename F = void (*)(Args...)>
inline
void hipLaunchKernelGGL(F kernel, const dim3& numBlocks, const dim3& dimBlocks,
include/hip/hcc_detail/hip_runtime_api.h
+15 -1
View File
@@ -2744,6 +2744,21 @@ hipError_t hipLaunchCooperativeKernel(const void* f, dim3 gridDim, dim3 blockDim
hipError_t hipLaunchCooperativeKernelMultiDevice(hipLaunchParams* launchParamsList,
int numDevices, unsigned int flags);
/**
* @brief determine the grid and block sizes to achieves maximum occupancy for a kernel
*
* @param [out] gridSize minimum grid size for maximum potential occupancy
* @param [out] blockSize block size for maximum potential occupancy
* @param [in] f kernel to launch
* @param [in] dynSharedMemPerBlk dynamic shared memory usage (in bytes) intended for each block
* @param [in] blockSizeLimit the maximum block size for the kernel, use 0 for no limit
*
* @returns hipSuccess, hipInvalidDevice, hipErrorInvalidValue
*/
hipError_t hipOccupancyMaxPotentialBlockSize(uint32_t* gridSize, uint32_t* blockSize,
hipFunction_t f, size_t dynSharedMemPerBlk,
uint32_t blockSizeLimit);
/**
* @brief Returns occupancy for a device function.
*
@@ -2782,7 +2797,6 @@ hipError_t hipExtLaunchMultiKernelMultiDevice(hipLaunchParams* launchParamsList,
int numDevices, unsigned int flags);
// doxygen end Version Management
/**
* @}