Currently the warp size is dynamically computed and sent to the kernel through the WARP_SIZE_TEST macro.
Test WARP_SIZE_TEST usage inside the kernel.
Check if WARP_SIZE_TEST and WARP_SIZE can be merged into one single parameter.
Implement any other warp specific feature not yet implemented. For example this line still uses WARP_SIZE:
reduce_energy_pow2(f_buf + (tidx & WARP_SIZE), e_lj, e_el, tidx & ~WARP_SIZE);
WARP_SIZE_TEST not used anymore inside the kernel.
The kernel still uses WARP_SIZE as defined for NVIDIA. It works for AMD as the wavefront size is 64.
Currently the warp size is dynamically computed and sent to the kernel through the WARP_SIZE_TEST macro.
Test WARP_SIZE_TEST usage inside the kernel.
Check if WARP_SIZE_TEST and WARP_SIZE can be merged into one single parameter.
Implement any other warp specific feature not yet implemented. For example this line still uses WARP_SIZE: reduce_energy_pow2(f_buf + (tidx & WARP_SIZE), e_lj, e_el, tidx & ~WARP_SIZE);
Check also #38