- Use kernel having wall_clock64 for gfx11
- Correct allocated memory size
- Use gpu event for kernel timing in place of std::chrono::system_clock
- Adjust comparison factors to accommodate kernel execution time differences
Change-Id: Ib119e8553a111feaf358693c911e650c1d50633f
[ROCm/hip commit: 7aa4cd6862]
- Use kernel having wall_clock64 for gfx11
- Correct allocated memory size
- Use gpu event for kernel timing in place of std::chrono::system_clock
- Adjust comparison factors to accommodate kernel execution time differences
Change-Id: Ib119e8553a111feaf358693c911e650c1d50633f
- Don't use HIP_PATH for kernel compilation.
- Make the test fail when code object file doesn't get created.
- Disable the test for Nvidia.
Change-Id: I0c78d6c0567aff058cb99fb3696b8d64c9a437ed
[ROCm/hip commit: dcb234c01f]
- Don't use HIP_PATH for kernel compilation.
- Make the test fail when code object file doesn't get created.
- Disable the test for Nvidia.
Change-Id: I0c78d6c0567aff058cb99fb3696b8d64c9a437ed
- Implement basic behavior checks in all copy directions
- Implement synchronization behavior checks for expected behavior based on cuda docs
- Implement positive tests for zero sized width and/or height copies, where no copy is expected to happen
- Implement negative parameter tests
- Implement all of the above for hipDrvMemcpy3D and hipDrvMemcpy3DAsync.
- Disable failing tests on AMD.
- Fix copyright disclaimer.
- Add defect issue numbers.
[ROCm/hip-tests commit: c695f1b146]
- Implement basic behavior checks in all copy directions
- Implement synchronization behavior checks for expected behavior based on cuda docs
- Implement positive tests for zero sized width and/or height copies, where no copy is expected to happen
- Implement negative parameter tests
- Implement all of the above for hipDrvMemcpy3D and hipDrvMemcpy3DAsync.
- Disable failing tests on AMD.
- Fix copyright disclaimer.
- Add defect issue numbers.
Blit manager requires an image view to reduce the amount
of copy kernels. Creation/destruction of a view in ROCr is
an expensive operation. Thus, runtime can cache views for fast access.
Change-Id: Ia67d775b481cc8326d91215ca22d4a73c1dddb59
[ROCm/clr commit: d29755452b]
Blit manager requires an image view to reduce the amount
of copy kernels. Creation/destruction of a view in ROCr is
an expensive operation. Thus, runtime can cache views for fast access.
Change-Id: Ia67d775b481cc8326d91215ca22d4a73c1dddb59
- Implement tests for hipMemcpy3D APIs
- Implement basic behavior checks in all copy directions
- Implement synchronization behavior checks for expected behavior based on cuda docs
- Implement positive tests for zero sized width and/or height copies, where no copy is expected to happen
- Implement negative parameter tests
- Implement all of the above for hipMemcpy3D and hipMemcpy3DAsync.
- Disable failing tests on AMD.
- Fix copyright disclaimer.
[ROCm/hip-tests commit: 113a36c0eb]
- Implement tests for hipMemcpy3D APIs
- Implement basic behavior checks in all copy directions
- Implement synchronization behavior checks for expected behavior based on cuda docs
- Implement positive tests for zero sized width and/or height copies, where no copy is expected to happen
- Implement negative parameter tests
- Implement all of the above for hipMemcpy3D and hipMemcpy3DAsync.
- Disable failing tests on AMD.
- Fix copyright disclaimer.
- Reimplement tests for hipOccupancyMaxActiveBlocksPerMultiprocessor and hipOccupancyMaxPotentialBlockSize APIs
- Add helper file occupancy_common.hh with parameterized templates
- Expand all positive and negative tests to use the templates
- Change section disable macros
- Disable AMD specific test due to defect
- Disable negative test using json file
- Fix disabled files
[ROCm/hip-tests commit: dd9b9b027f]
- Reimplement tests for hipOccupancyMaxActiveBlocksPerMultiprocessor and hipOccupancyMaxPotentialBlockSize APIs
- Add helper file occupancy_common.hh with parameterized templates
- Expand all positive and negative tests to use the templates
- Change section disable macros
- Disable AMD specific test due to defect
- Disable negative test using json file
- Fix disabled files