mirror of https://github.com/ROCm/composable_kernel.git synced 2026-04-19 22:39:03 +00:00

Files

Qianfeng c1ef73192e Use ck::half_t for Host Reduction (#195 )

* Add math functions for host

* Change to host reduction to use ck::math:

* Remove the using of half_float::half and half.hpp from reduction example/profiler/ctest

2022-04-20 22:09:26 -05:00

CMakeLists.txt

Reorganize files, Part 1 (#119 )

2022-03-08 21:46:36 -06:00

README.md

Compile for gfx908 and gfx90a (#130 )

2022-03-31 12:33:34 -05:00

reduce_blockwise.cpp

Use ck::half_t for Host Reduction (#195 )

2022-04-20 22:09:26 -05:00

README.md

Instructions for `example_reduce_blockwise`

Run `example_reduce_blockwise`

# -D <xxx> : input 4-d tensor lengths
# -v <x> :   verification (0=no, 1=yes)
#arg1: initialization (0=no init, 1=single integer value, 2=scope integer value, 3=decimal value)
#arg2: run kernel # of times (>1)
./bin/example_reduce_blockwise -D 16,64,32,960 -v 1 1 10

Result

launch_and_time_kernel: grid_dim {240, 1, 1}, block_dim {256, 1, 1} 
Warm up
Start running 3 times...
Perf: 0.23536 ms, 267.32 GB/s, DeviceReduceBlockWise<256,M_C4_S1,K_C64_S1,InSrcVectorDim_0_InSrcVectorSize_1_OutDstVectorSize_1>
error: 0
max_diff: 0, 529, 529
root@dc-smc-18:/data/composable_kernel/Build3# bin/example_reduce_blockwise -D 16,64,32,960 -v 1 1 10
launch_and_time_kernel: grid_dim {240, 1, 1}, block_dim {256, 1, 1} 
Warm up
Start running 10 times...
Perf: 0.23392 ms, 268.966 GB/s, DeviceReduceBlockWise<256,M_C4_S1,K_C64_S1,InSrcVectorDim_0_InSrcVectorSize_1_OutDstVectorSize_1>
error: 0
max_diff: 0, 528, 528

README.md

Instructions for example_reduce_blockwise

Run example_reduce_blockwise

Instructions for `example_reduce_blockwise`

Run `example_reduce_blockwise`