Haocong WANG f83e9701e9 [GEMM] Gemm universal device operation (#1154) ...

* Optimize GEMM on MI200/300:
1. Add new blockwise gemm pipeline
2. Add irregular splitk intances

* clang format + typo fix

* Fix a bug

* initial commit

* Add more instances to irregular splitk

* blkgemm pipeline v1~4 prototype

* Sanity Checked. Known issue:
1. Poor performance of splitk
2. Register spill on blkgemmpipeline v3

* Sanity and Performance fix:
1. fix a bug related to sanity in grouped b2c mapping
2. fix a bug related to sanity and performance in splitk offset

* Sanity and API update:
1. Remove prefetch stage
2. Fix valid check bug
3, Add first gemm_universal instance into ckProfiler

* Add NN instances for gemm universal

* 1. Add NT instances for gemm_universal
2. Fix a bug about Kpadding in gemm_universal

* Fix a bug regarding padding Odd K number

* remove kernel print

* Fix KPadding bug...

* Update safety check

* another try to fix kpadding..

* Sanity checked

* new instances..

* clang format+typo fix

* remove clang format script's change

* Add non-hotloop compile option

* 1. Add fp16xfp8 example
2. pull packed convert f8 from pr1150

* Some miscs.. opt and fix

* Add pipeline description docs

* Split universal gemm instance library to cut profiler compiling time

* uncomment cmakefile

* Fix a bug caused by blockwise_gemm_pipe_v2

* reduce default splitk to 1

* Add 224x256x64 tile size

* update, including:
1. Experiment pipeline 5~7
2. Optimization for pipeline 4
3. Organized instance library

* temp save

* temp save

* Permuted lds layout, sanity and function checked

* clang format

* Move OOB check from RunRead to RunWrite, for better software pipeline.
TODO: agpr spill when NN layout

* clangformat

* A/B splitpipe scheduler for v3

* Fix two bugs

* bug fix

* fix a bug in oob check

* Example for mixed fp16_fp8 gemm

* Clean experimental code blocks

* Add mixed precision gemm into profiler

* tempsave

* optimize m/n major lds layout

* Add RRR GEMM  mixed precision instances

* Optimize f8 matrix transpose

* Add test_gemm_universal

* A/B spilt schedule for blkpip v5

* Take ds_read2 into iglp scheduling scheme

* format

* fixed cmake

* Add llvm-option into CI cmake flag

---------

Co-authored-by: Jing Zhang <jizhan@amd.com>

2024-04-13 21:03:18 -05:00

..

device_avgpool3d_bwd_ndhwc_ndhwc.hpp

Average pool backward deviceOP and example (#797)

2023-08-10 12:04:35 +08:00

device_batched_contraction_multiple_d_wmma_cshuffle.hpp

Navi3 rel (#1176)

2024-03-08 17:11:51 -08:00

device_batched_contraction_multiple_d_xdl_cshuffle.hpp

Add support for more Navi2x and Navi3x models. (#1152)

2024-02-02 11:35:26 -08:00

device_batched_gemm_e_permute_xdl.hpp

Add support for more Navi2x and Navi3x models. (#1152)

2024-02-02 11:35:26 -08:00

device_batched_gemm_gemm_xdl_cshuffle.hpp

Add support for more Navi2x and Navi3x models. (#1152)

2024-02-02 11:35:26 -08:00

device_batched_gemm_multi_d_xdl.hpp

Add support for more Navi2x and Navi3x models. (#1152)

2024-02-02 11:35:26 -08:00

device_batched_gemm_multiple_d_dl.hpp

Add support for more Navi2x and Navi3x models. (#1152)

2024-02-02 11:35:26 -08:00

device_batched_gemm_multiple_d_gemm_multiple_d_xdl_cshuffle.hpp

Add support for more Navi2x and Navi3x models. (#1152)

2024-02-02 11:35:26 -08:00

device_batched_gemm_reduce_xdl_cshuffle.hpp

Add support for more Navi2x and Navi3x models. (#1152)

2024-02-02 11:35:26 -08:00

device_batched_gemm_softmax_gemm_permute_wmma_cshuffle.hpp

Navi3 rel (#1176)

2024-03-08 17:11:51 -08:00

device_batched_gemm_softmax_gemm_permute_xdl_cshuffle.hpp

Add support for more Navi2x and Navi3x models. (#1152)

2024-02-02 11:35:26 -08:00

device_batched_gemm_softmax_gemm_xdl_cshuffle.hpp

Add support for more Navi2x and Navi3x models. (#1152)

2024-02-02 11:35:26 -08:00

device_batched_gemm_xdl.hpp

Add support for more Navi2x and Navi3x models. (#1152)

2024-02-02 11:35:26 -08:00

device_batchnorm_backward_impl.hpp

Fixed GroupedGemmFixedNK with hipGraph (#1065)

2023-11-30 15:09:27 -06:00

device_batchnorm_forward_impl_obsolete.hpp

Fixed GroupedGemmFixedNK with hipGraph (#1065)

2023-11-30 15:09:27 -06:00

device_batchnorm_forward_impl.hpp

Fixed GroupedGemmFixedNK with hipGraph (#1065)

2023-11-30 15:09:27 -06:00

device_cgemm_4gemm_xdl_cshuffle.hpp

Enable f16/f8 mixed precision mode (#820)

2023-08-09 08:44:23 -05:00

device_column_to_image_impl.hpp

Support multi AB for grouped conv fwd xdl (#1027)

2023-11-10 15:54:44 +01:00

device_contraction_multiple_abd_xdl_cshuffle.hpp

Add support for more Navi2x and Navi3x models. (#1152)

2024-02-02 11:35:26 -08:00

device_contraction_multiple_d_xdl_cshuffle.hpp

Add support for more Navi2x and Navi3x models. (#1152)

2024-02-02 11:35:26 -08:00

device_contraction_utils.hpp

SWDEV-439954 - Use hard coded filename rather than using the macro __FILE__ for debug prints. (#1123)

2024-01-09 08:21:47 -08:00

device_conv2d_backward_weight_xdl_c_shuffle_nhwc_kyxc_nhwk.hpp

Use asynchronous version of hipMemset (#850)

2023-08-18 11:14:59 +08:00

device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk.hpp

Disable XDL kernels on unsupported HW Add ck::is_xdl_supported (#768)

2023-07-26 07:19:55 -07:00

device_conv2d_fwd_xdl_c_shuffle_bias_activation_add_nhwc_kyxc_nhwk.hpp

Disable XDL kernels on unsupported HW Add ck::is_xdl_supported (#768)

2023-07-26 07:19:55 -07:00

device_conv2d_fwd_xdl_c_shuffle_bias_activation_nhwc_kyxc_nhwk.hpp

Disable XDL kernels on unsupported HW Add ck::is_xdl_supported (#768)

2023-07-26 07:19:55 -07:00

device_conv2d_fwd_xdl_c_shuffle_nhwc_kyxc_nhwk.hpp

Disable XDL kernels on unsupported HW Add ck::is_xdl_supported (#768)

2023-07-26 07:19:55 -07:00

device_conv2d_fwd_xdl_nhwc_kyxc_nhwk.hpp

Disable XDL kernels on unsupported HW Add ck::is_xdl_supported (#768)

2023-07-26 07:19:55 -07:00

device_conv3d_fwd_naive_ndhwc_kzyxc_ndhwk.hpp

update copyright headers (#726)

2023-05-31 18:46:57 -05:00

device_conv3d_fwd_xdl_ndhwc_kzyxc_ndhwk.hpp

Add support for more Navi2x and Navi3x models. (#1152)

2024-02-02 11:35:26 -08:00

device_convnd_bwd_data_nwc_kxc_nwk_dl.hpp

Add support for more Navi2x and Navi3x models. (#1152)

2024-02-02 11:35:26 -08:00

device_convnd_bwd_data_nwc_kxc_nwk_xdl.hpp

Disable XDL kernels on unsupported HW Add ck::is_xdl_supported (#768)

2023-07-26 07:19:55 -07:00

device_elementwise_2d_impl.hpp

Using number of compute units to set gridSize (#754)

2023-06-15 10:13:59 -05:00

device_elementwise_3d_impl.hpp

Disable transpose device op for MI300 (#1050)

2023-11-29 11:36:40 -06:00

device_elementwise_dynamic_vector_dims_impl.hpp

Add elementwise with dynamic vector dim (#1198)

2024-03-22 10:40:43 +01:00

device_elementwise_impl.hpp

Add Grouped Gemm Multiple D SplitK TwoStage (#1212)

2024-04-04 11:01:33 +02:00

device_elementwise_normalization_impl.hpp

update copyright headers (#726)

2023-05-31 18:46:57 -05:00

device_elementwise_scale_impl.hpp

Extend permute scale support up to 6D (#1168)

2024-02-20 09:56:54 -08:00

device_fpAintB_gemm_wmma.hpp

Navi3 rel (#1176)

2024-03-08 17:11:51 -08:00

device_gemm_bias_add_reduce_xdl_cshuffle.hpp

Disable XDL kernels on unsupported HW Add ck::is_xdl_supported (#768)

2023-07-26 07:19:55 -07:00

device_gemm_dl.hpp

Add support for more Navi2x and Navi3x models. (#1152)

2024-02-02 11:35:26 -08:00

device_gemm_dpp.hpp

Add support for more Navi2x and Navi3x models. (#1152)

2024-02-02 11:35:26 -08:00

device_gemm_multiple_abd_xdl_cshuffle.hpp

Add support for more Navi2x and Navi3x models. (#1152)

2024-02-02 11:35:26 -08:00

device_gemm_multiple_d_dl.hpp

Add support for more Navi2x and Navi3x models. (#1152)

2024-02-02 11:35:26 -08:00

device_gemm_multiple_d_layernorm_xdl_cshuffle.hpp

Add support for more Navi2x and Navi3x models. (#1152)

2024-02-02 11:35:26 -08:00

device_gemm_multiple_d_multiple_r_xdl_cshuffle.hpp

Add support for more Navi2x and Navi3x models. (#1152)

2024-02-02 11:35:26 -08:00

device_gemm_multiple_d_wmma_cshuffle.hpp

Navi3 rel (#1176)

2024-03-08 17:11:51 -08:00

device_gemm_multiple_d_xdl_cshuffle_lds_direct_load.hpp

Add basic support for direct loads from global to LDS (#999)

2023-11-25 13:35:22 +01:00

device_gemm_multiple_d_xdl_cshuffle.hpp

Add host lib (#1134)

2024-03-05 17:08:43 -08:00

device_gemm_reduce_xdl_cshuffle.hpp

Disable XDL kernels on unsupported HW Add ck::is_xdl_supported (#768)

2023-07-26 07:19:55 -07:00

device_gemm_wmma.hpp

Navi3 rel (#1176)

2024-03-08 17:11:51 -08:00

device_gemm_xdl_cshuffle_lds_direct_load.hpp

Add support for double buffering in direct load GEMM kernel (#1052)

2023-12-03 23:08:47 +01:00

device_gemm_xdl_cshuffle_v2.hpp

[GEMM] Optimization for MI200/300. (#1135)

2024-01-19 07:02:22 -06:00

device_gemm_xdl_cshuffle_v3.hpp

[GEMM] Gemm universal device operation (#1154)

2024-04-13 21:03:18 -05:00

device_gemm_xdl_cshuffle.hpp

Add Gemm instances for performance improvement (#1018)

2023-11-07 09:09:58 -06:00

device_gemm_xdl_layernorm_cshuffle.hpp

Disable XDL kernels on unsupported HW Add ck::is_xdl_supported (#768)

2023-07-26 07:19:55 -07:00

device_gemm_xdl_skip_b_lds.hpp

Disable XDL kernels on unsupported HW Add ck::is_xdl_supported (#768)

2023-07-26 07:19:55 -07:00

device_gemm_xdl_splitk_c_shuffle_lds_direct_load.hpp

Implement direct loads split-K GEMM kernel (#1137)

2024-02-07 01:08:34 +01:00

device_gemm_xdl_splitk_c_shuffle.hpp

Optimizing fp8_fp16 mixedprec gemm (#1150)

2024-02-12 09:45:42 -08:00

device_gemm_xdl_streamk.hpp

Add support for more Navi2x and Navi3x models. (#1152)

2024-02-02 11:35:26 -08:00

device_gemm_xdl_waveletmodel_cshuffle.hpp

Add support for more Navi2x and Navi3x models. (#1152)

2024-02-02 11:35:26 -08:00

device_gemm_xdl.hpp

Add support for more Navi2x and Navi3x models. (#1152)

2024-02-02 11:35:26 -08:00

device_grouped_contraction_multiple_d_xdl_cshuffle.hpp

Add support for more Navi2x and Navi3x models. (#1152)

2024-02-02 11:35:26 -08:00

device_grouped_conv_bwd_data_multiple_d_wmma_cshuffle.hpp

Navi3 rel (#1176)

2024-03-08 17:11:51 -08:00

device_grouped_conv_bwd_data_multiple_d_xdl_cshuffle_v1.hpp

Add support for more Navi2x and Navi3x models. (#1152)

2024-02-02 11:35:26 -08:00

device_grouped_conv_bwd_weight_dl.hpp

Add support for more Navi2x and Navi3x models. (#1152)

2024-02-02 11:35:26 -08:00

device_grouped_conv_bwd_weight_wmma_cshuffle.hpp

Navi3 rel (#1176)

2024-03-08 17:11:51 -08:00

device_grouped_conv_bwd_weight_xdl_cshuffle.hpp

Add support for more Navi2x and Navi3x models. (#1152)

2024-02-02 11:35:26 -08:00

device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp

Add support for more Navi2x and Navi3x models. (#1152)

2024-02-02 11:35:26 -08:00

device_grouped_conv_fwd_dl_nhwc_kyxc_nhwk.hpp

Add support for more Navi2x and Navi3x models. (#1152)

2024-02-02 11:35:26 -08:00

device_grouped_conv_fwd_multiple_abd_xdl_cshuffle.hpp

Add instances for conv_scale with fp8@bf8->fp8 (#1220)

2024-04-03 09:08:08 -05:00

device_grouped_conv_fwd_multiple_d_multiple_r_xdl_cshuffle.hpp

Add support for more Navi2x and Navi3x models. (#1152)

2024-02-02 11:35:26 -08:00

device_grouped_conv_fwd_multiple_d_multiple_r.hpp

update copyright headers (#726)

2023-05-31 18:46:57 -05:00

device_grouped_conv_fwd_multiple_d_wmma_cshuffle.hpp

Navi3 rel (#1176)

2024-03-08 17:11:51 -08:00

device_grouped_conv_fwd_multiple_d_xdl_cshuffle.hpp

Add instances for conv_scale with fp8@bf8->fp8 (#1220)

2024-04-03 09:08:08 -05:00

device_grouped_conv_utils.hpp

Support multi AB for grouped conv fwd xdl (#1027)

2023-11-10 15:54:44 +01:00

device_grouped_gemm_multiple_d_dl.hpp

Add support for more Navi2x and Navi3x models. (#1152)

2024-02-02 11:35:26 -08:00

device_grouped_gemm_multiple_d_splitk_xdl_cshuffle_two_stage.hpp

Add Grouped Gemm Multiple D SplitK TwoStage (#1212)

2024-04-04 11:01:33 +02:00

device_grouped_gemm_softmax_gemm_permute_xdl_cshuffle.hpp

Add support for more Navi2x and Navi3x models. (#1152)

2024-02-02 11:35:26 -08:00

device_grouped_gemm_xdl_fixed_nk.hpp

improved zeroing (#1221)

2024-04-02 11:02:52 -05:00

device_grouped_gemm_xdl_splitk_cshuffle.hpp

Add Grouped Gemm Multiple D SplitK TwoStage (#1212)

2024-04-04 11:01:33 +02:00

device_grouped_gemm_xdl.hpp

Add support for more Navi2x and Navi3x models. (#1152)

2024-02-02 11:35:26 -08:00

device_grouped_query_attention_forward_wmma.hpp

Navi3 rel (#1176)

2024-03-08 17:11:51 -08:00

device_image_to_column_impl.hpp

Support multi AB for grouped conv fwd xdl (#1027)

2023-11-10 15:54:44 +01:00

device_max_pool_bwd_impl.hpp

MaxPool & AvgPool bwd instances, test, ckProfiler, client example (#861)

2023-08-31 21:01:50 +08:00

device_multi_query_attention_forward_wmma.hpp

Navi3 rel (#1176)

2024-03-08 17:11:51 -08:00

device_multiple_reduce_multiblock.hpp

update copyright headers (#726)

2023-05-31 18:46:57 -05:00

device_multiple_reduce_threadwise.hpp

update copyright headers (#726)

2023-05-31 18:46:57 -05:00

device_normalization_bwd_data_impl.hpp

layernorm and groupnorm backward data (#1083)

2023-12-19 04:23:11 +08:00

device_normalization_bwd_gamma_beta_impl.hpp

layernorm and groupnorm backward data (#1083)

2023-12-19 04:23:11 +08:00

device_normalization_fwd_impl.hpp

layernorm and groupnorm backward data (#1083)

2023-12-19 04:23:11 +08:00

device_normalization_fwd_splitk_impl.hpp

layernorm and groupnorm backward data (#1083)

2023-12-19 04:23:11 +08:00

device_permute_impl.hpp

update copyright headers (#726)

2023-05-31 18:46:57 -05:00

device_pool2d_fwd_nhwc_nhwc.hpp

Refactor pool fwd (#815)

2023-08-15 02:25:28 +08:00

device_pool3d_fwd_ndhwc_ndhwc.hpp

Refactor pool fwd (#815)

2023-08-15 02:25:28 +08:00

device_put_element_impl.hpp

Maxpool bwd (#750)

2023-06-19 09:44:22 -05:00

device_reduce_common.hpp

update copyright headers (#726)

2023-05-31 18:46:57 -05:00

device_reduce_multiblock.hpp

update copyright headers (#726)

2023-05-31 18:46:57 -05:00

device_reduce_threadwise.hpp

update copyright headers (#726)

2023-05-31 18:46:57 -05:00

device_softmax_impl.hpp

Revert "Grouped Gemm with looping over the tiles. (#788)" (#982)

2023-10-11 14:27:29 -05:00

device_sparse_embeddings_forward_layernorm.hpp

update copyright headers (#726)

2023-05-31 18:46:57 -05:00

device_splitk_contraction_multiple_d_xdl_cshuffle.hpp

Add support for more Navi2x and Navi3x models. (#1152)

2024-02-02 11:35:26 -08:00