Anthony Chang 7db48f9008 Tune & add conflict-free LDS gemm kernels (#159) ...

* retune & add conflict-free bf16/fp16 c-shuffle gemm instances

amend wrong K1 value in some fp16/bf16 kernel instances

* make gemm cshuffle's timing behavior consistent with all other functions

* clang-format

* retune & add conflict-free fp32 c-shuffle gemm instances

* retune & add conflict-free int8 c-shuffle gemm instances

* update the underlying gridwise gemm of all c-shuffle gemm kernels

* typo

2022-03-31 12:58:41 -05:00

..

conv_utils.hpp

Unified implementation of 1d/2d/3d conv bwd-data. fp32/fp16/bfp16/int8 (#134)

2022-03-29 10:52:25 -05:00

convolution_backward_data_specialization.hpp

Compile for gfx908 and gfx90a (#130)

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

convolution_forward_specialization.hpp

Compile for gfx908 and gfx90a (#130)

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

convolution_utility.hpp

Reorganize files, Part 1 (#119)

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

device_base.hpp

Reorganize files, Part 1 (#119)

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

device_batched_gemm_reduce_xdl_cshuffle.hpp

Compile for gfx908 and gfx90a (#130)

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

device_batched_gemm_xdl.hpp

Compile for gfx908 and gfx90a (#130)

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

device_conv2d_backward_weight_xdl_c_shuffle_nhwc_kyxc_nhwk.hpp

Compile for gfx908 and gfx90a (#130)

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

device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk.hpp

Compile for gfx908 and gfx90a (#130)

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

device_conv2d_fwd_xdl_c_shuffle_bias_activation_add_nhwc_kyxc_nhwk.hpp

Compile for gfx908 and gfx90a (#130)

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

device_conv2d_fwd_xdl_c_shuffle_bias_activation_nhwc_kyxc_nhwk.hpp

Compile for gfx908 and gfx90a (#130)

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

device_conv2d_fwd_xdl_c_shuffle_nhwc_kyxc_nhwk.hpp

Compile for gfx908 and gfx90a (#130)

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

device_conv2d_fwd_xdl_nhwc_kyxc_nhwk.hpp

Compile for gfx908 and gfx90a (#130)

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

device_conv3d_fwd_naive_ndhwc_kzyxc_ndhwk.hpp

Reorganize files, Part 1 (#119)

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

device_conv3d_fwd_xdl_ndhwc_kzyxc_ndhwk.hpp

Compile for gfx908 and gfx90a (#130)

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

device_conv_backward_weight.hpp

Reorganize files, Part 1 (#119)

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

device_conv_bwd_data.hpp

Reorganize files, Part 1 (#119)

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

device_conv_fwd_bias_activation_add.hpp

Reorganize files, Part 1 (#119)

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

device_conv_fwd_bias_activation.hpp

Reorganize files, Part 1 (#119)

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

device_conv_fwd.hpp

Reorganize files, Part 1 (#119)

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

device_convnd_bwd_data_xdl_ndhwc_kzyxc_ndhwk.hpp

Compile for gfx908 and gfx90a (#130)

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

device_convnd_fwd_xdl_nhwc_kyxc_nhwk.hpp

Compile for gfx908 and gfx90a (#130)

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

device_gemm_bias_activation_add.hpp

Reorganize files, Part 1 (#119)

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

device_gemm_bias_activation.hpp

Reorganize files, Part 1 (#119)

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

device_gemm_bias.hpp

Gemm+Reduce Fusion (#128)

2022-03-23 22:18:42 -05:00

device_gemm_reduce_xdl_cshuffle.hpp

Compile for gfx908 and gfx90a (#130)

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

device_gemm_reduce.hpp

Batched gemm and reduction (#156)

2022-03-30 11:21:18 -05:00

device_gemm_xdl_c_shuffle_bias_2d.hpp

Compile for gfx908 and gfx90a (#130)

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

device_gemm_xdl_c_shuffle_bias_activation_add.hpp

Compile for gfx908 and gfx90a (#130)

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

device_gemm_xdl_c_shuffle_bias_activation.hpp

Compile for gfx908 and gfx90a (#130)

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

device_gemm_xdl_c_shuffle.hpp

Compile for gfx908 and gfx90a (#130)

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

device_gemm_xdl_cshuffle.hpp

Tune & add conflict-free LDS gemm kernels (#159)

2022-03-31 12:58:41 -05:00

device_gemm_xdl_splitk_c_shuffle.hpp

Compile for gfx908 and gfx90a (#130)

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

device_gemm_xdl_splitk.hpp

Compile for gfx908 and gfx90a (#130)

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

device_gemm_xdl.hpp

Compile for gfx908 and gfx90a (#130)

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

device_gemm.hpp

batched_gemm: use profiler in ctest (#163)

2022-03-30 21:32:49 -05:00

device_grouped_gemm_xdl.hpp

Compile for gfx908 and gfx90a (#130)

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

device_pool2d_fwd_nhwc_nhwc.hpp

Compile for gfx908 and gfx90a (#130)

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

device_pool2d_fwd.hpp

Compile for gfx908 and gfx90a (#130)

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

device_reduce_blockwise_second_call.hpp

Reduction for int8 and bfloat16 (#125)

2022-03-22 14:35:14 -05:00

device_reduce_blockwise.hpp

Reduction for int8 and bfloat16 (#125)

2022-03-22 14:35:14 -05:00

device_reduce_common.hpp

Reduction for int8 and bfloat16 (#125)

2022-03-22 14:35:14 -05:00

device_reduce_multiblock_atomic_add.hpp

Reduction for int8 and bfloat16 (#125)

2022-03-22 14:35:14 -05:00

device_reduce_multiblock_partial_reduce.hpp

Reduction for int8 and bfloat16 (#125)

2022-03-22 14:35:14 -05:00

device_reduce_threadwise.hpp

Reduction for int8 and bfloat16 (#125)

2022-03-22 14:35:14 -05:00

device_reduce.hpp

Reduction for int8 and bfloat16 (#125)

2022-03-22 14:35:14 -05:00

gemm_specialization.hpp

Compile for gfx908 and gfx90a (#130)

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

reduction_operator_mapping.hpp

Compile for gfx908 and gfx90a (#130)

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

tensor_layout.hpp

batched_gemm: use profiler in ctest (#163)

2022-03-30 21:32:49 -05:00