* Allow selection of mfma_scale instructions
* Read B tensor from LDS to VGPR in chunks of 16 in MFMA order
* Add constexpr and synchronize return type for `get_exponent_value`
* Pass scales by reference and add comments to `mfma_scale_f32_32x32x64`
* Add support for microscaling instructions in `XdlopsGemm`
* Fix `mfma_scale_f32_16x16x128f8f6f4` wrapper
* Remove software implementation of MX GEMM
* Make interface of `intrin_mfma_scale_f32_16x16x128f8f6f4<16, 16>` consistent with the other scale instruction
* Update README
* Updated CHANGELOG
* Remove unused static methods
* Added two kernel for M=32 problem
* Comment the first one
* Enable multiply_multiply for Scale_Block_M = 1 for deepseek
* Modify the a_thread offset since the A data load is different from B.
* edit fp8 ab scale for Scale_Block_M=1
* edit GemmSpec to MNKPadding
* enable blockwise pipelie v1 and v2. v1 is work for small K.
* add instance for gemm_ab_scale
* fix cmakelist of ckProfiler
* optimize blockscale gemm. todo: reduce vgpr usage
* fix a correctness bug
* sanity checked
* revert ckprofiler cmake changes
* clang format
* revert unnecessary changes.
* remove commented codes.
* split weight preshuffle library targets
* bring back enable-post-misched=0
* fix build issues for gemm_multiply_multiply_fp8 instances
* fix clang format
* add verbose build flag when building for all targets
* reduce path names for new instances
* fix paths in cmake
* refactor gemm_multiply_multiply library target
* fix a bug in example
* fix example 65 cmake
* reduce the number of threads when building libs for all targets to 50
* use ninja to build for all targets
* reduce teh number of threads when building for all targets
* reduce the number of threads to 32 when building libs for all targets to 50
---------
Co-authored-by: mtgu0705 <mtgu@amd.com>
Co-authored-by: chenjun <junchen2@amd.com>
Co-authored-by: illsilin <Illia.Silin@amd.com>
Co-authored-by: Illia Silin <98187287+illsilin@users.noreply.github.com>
* Add Gemm fp8xint4 example and kernel, function pass.
* Init Gemm_fp8xint4 Bpreshuffle
* Added gemm_fp8xint4_Bpreshuffle files, function not checked yet
* General fix.
* fp8xint4 bpreshuffle function pass
* fix.
* init b preshuffle dequant in VGPR.
* fix bug, function pass.
* move b thread dequant copy to blockwise.
* fix bug, function now passes.
* modified the tile size to 256, 128x128x128.
* fixed a bug.
* Initial int4 moe, compile pass, function not check.
* fix bug in moe_gemm1.cpp, now function pass.
* test expert = 8 and function pass.
* Added moe_pk_i4_gemm2, function pass.
* Added b preshuffle pipeline v3 support.
* fixed merge issue. fp8xint4 and fp8xint4_bpreshuffle function pass.
* Split the blockwise pipeline for fp8xint4.
* commit missing files
* opt gemm2 to 2x2 wave
* fix swizzle = false
* update int4 moe with latest input changes.
* update tile size.
* enable pipeline v3.
* fix nswizzle = true
* commit a version for compiler debug.
* Updated transfer_v3r1_gather to support pk_i4_t type.
* for int4 moe2 for type_convert support.
* remove some values between mfma instructions.
* fix int4 moe
* Updated transfer_v3r1_gather to support pk_i4_t type.
* i4 support lds multiple shuffle
* fixed int4 moe tflops calculation.
* Modified CshuffleCShuffleMXdlPerWavePerShuffle to 1 to suit C multiple shuffle
* updated gemm2.
* change int4 moe example names
* fix and format code.
* format.
* format codes.
* update fp8xint4 example tile size.
* add <unordered_map> header
* fixed.
* format.
* Added conditional compilation for int4 -> fp8 conversion kernels
---------
Co-authored-by: mtgu0705 <mtgu@amd.com>
Co-authored-by: coderfeli <coderfeli@163.com>
* port all moe changes from ck_moe_gemm branch
* refine codes in the pr
* fix tail odd
* fix clang format
* fix clang format2
* make hot loop scheduler compatible with 16x16 and 32x32
* clang format
* fix per token quant
* rename moe example
* clang format
---------
Co-authored-by: coderfeli <coderfeli@163.com>
* Added two kernel for M=32 problem
* Comment the first one
* Enable multiply_multiply for Scale_Block_M = 1 for deepseek
* Modify the a_thread offset since the A data load is different from B.
* edit fp8 ab scale for Scale_Block_M=1
* edit GemmSpec to MNKPadding
* enable blockwise pipelie v1 and v2. v1 is work for small K.
* add instance for gemm_ab_scale
* fix cmakelist of ckProfiler
* optimize blockscale gemm. todo: reduce vgpr usage
* fix a correctness bug
* sanity checked
* revert ckprofiler cmake changes
* clang format
* revert unnecessary changes.
* remove commented codes.
---------
Co-authored-by: mtgu0705 <mtgu@amd.com>
Co-authored-by: chenjun <junchen2@amd.com>
* tempsave
* temp save
* tempsave
* tempsave, epilogue optimization for universal gemm done. TODO: mulitpleD epilogue optimization
* temp save
* tempsave
* temp save
* update bf16 instance list
* clang format
* bug fix
* temp save
* tempsave
* revert exp changes.
* add blank line
* add int8 gemm multiply multiply a8w8
* uncomment
* clang-format-12
* Add example_gemm_multiply_multiply_xdl_int8
* Remove shell scripts
* update preprocess number for mi308; bring back printout in ckprofiler
* tempsave
* update ck_a8w8 library, update flush cache timing api
* remove the change in ckprofiler src
* clean the flush_cache api
* reduce prefetch stage in blockwisepipev4
* update tile size for fp8 rowwise
* fix bug in enable f8 gemm inside ckProfiler
* update instance and lds layout strategy
* delete use less files
* fix cmake bug
* update instances
* add configs to fix tunning cases
* port tiles from a8w8
* rm debug used files
* add instances
* remove all non gemm in cmake
* fix build
* sanity bug fix
* add bypass logic and build
* can run
* add double buffer scratch
* remove agpr usage when vgpr usage <256
* add configs to fix tunning cases
* fix build
* fix performance regression on blockgemm v3 pipe
* using develop branch timer
* impl fp16 in ckprofiler
* add cpu shuffle
* fix tail
* use empty hipstream in ckprofiler
* fix missed files and fix clang format
* fix fp16 build
* fix cmake rm compile options
* fix brepeat, kloop and lds two buffer; works ok now
* use new pipeline for b preshuffle, run ok; revert olds to fix ckprofiler
* auto calculate hard code params
* fix warnings and revert cmake and fix clang format
* tempsave
* sanity pass, most tile size enabled. TODO: NWave!=4
* disable N, K Padding, splitk enabled
* add fp16 instances
* use bpreshuffle as independent example
* refine weight preshuffle format.
* tempsave
* optimize software pipeline
* refine blockgemm pipeline version as base struct.
* fp8 add_rmsnorm_dynamic_dequant
* add save_x=true instance
* tempsave
* Add compute-friendly pipeline for bpreshuffle case; remove enable-post-misched=0 flag.
* fix Odd Mrepeat number pipelinev3; Add v3 instances to ckProfiler
* clean the code
* Merge from internal (#1857)
* enable batched_gemm_softmax_gemm_perm_wmma for gfx12
* disable instances with blocksize=256 in attention examples
* debuggging
* debug
* fixed lds_enabled
* debugging
* Fix and add limit to skiplds feature
* Enable skipLds feature and fix compilation bugs
* add ck_tile definitions for gfx12
* fix clang format and test/wmma_op
* updage instances cmake for gfx12
* disable the test_wmma_op on gfx12
* fix the builds for gfx950
* add gfx12 and gfx950 to default target list
* clean-up cmake file
* Initial introduction of OFP8 data types.
* Renamed FP8 and BF8 tests into FP8_FNUZ and BF8_FNUZ.
* Implementation of ConvertFP32Nearest in test_fp8_ocp.
* Remove dependence on possibly undeclared alias.
* Implement FP8OCP test for stochastic rounding mode.
* Implement FP8OCP tests for half_t type conversions.
* enable bf16 atomic add on gfx950
* Implement ConvertFP32Nearest test.
* Implement ConvertFP32Stochastic test.
* Implement ConvertFP16Nearest and ConvertFP16Stochastic tests.
* Refactoring. Move FP8 definitions into a separate header file.
* Enable easy switching between architectures.
* Fix compilation error for gfx942 architecture.
* Add fp4 type with constants
* only builf gfx950 branch for gfx950 target by default
* Enable OCP build of example_gemm_xdl_fp8.
* Fix formatting.
* fix the build logic for gfx950
* Improve GEMM example verbosity.
* Add constexpr where applicable.
* fix the logic of enabling XDL and WMMA instances
* Improve GEMM example verbosity.
* Enable build of example_gemm_xdl_fp8_bf8 test.
* Fix tests for gfx1101 architecture.
* Build DPP examples only on gfx103 and gfx11 architectures.
* Optionaly run either CPU or GPU verifications with GEMM examples.
* Extend GeneratorTensor_Sequential to produce values of prescribed data types.
* Add missing constructor.
* Add scale type and mxfp conversions
* Update conversions
* Add conversion tests
* Fix typo
* Improve infrastructure for OFP8 data type support.
* BUGFIX. Should not use FP8 as Compute/Accum data type.
* Add custom target for grouped_convnd_bwd_weight tests.
* Can build `tests` target on gfx950.
* Bugfixes on gfx1101 architecture.
* Fix dependencies.
* Add stochastic rounding tests
* Provide single point of truth for FP8 INF and NAN checks
* Prevent instantiation of operators that are not supported by FP8 data types
* Add FP8 type selection into client_axample CMakeLists.txt
* Prevent sccache server from shutting down during build
* Fix test success reporting logic
* Change default verification method to CPU.
GPU verification takes too much time to complete on the emulator.
* Add scale <-> float conversions
* Add scaled conversions with tests
* Add device conversions
* Make sure all tests and examples are built for gfx950
* Facilitate testing of FP8 data types on the emulator
* Introduce two new tensor generators
* Enable instances built for gfx94 to be built on gfx950
* Verify 35_splitk_gemm on floating point numbers.
splitk gemm appears to be losing precision VS reference implementation when FP numbers are involved.
* Format
* Verify 04_gemm_add_add_fastgelu on floating point numbers
* Verify 20_grouped_conv_bwd_weight on floating point numbers
* Verify 38_grouped_conv_bwd_data_multiple_d on floating point numbers
* Verify more tests on floating point data
* Fix data types and improve testing verbocity.
* Add fp4 vectors
* Add debug tests
* Upgrade to NPI 573 build docker.
* Skip on gemm_universal tests.
The tests take too long to complete on the emulator.
Need to see if it is possible to reduce the scope of the testing to just FP8 data types.
* Add new mfma instructions and examples
* Add preprocessor directives for gfx950 specific code
* Fix gfx1101 build
* Document test availability
* Re-enable fp8 gemms for gfx94/95
* Cherry-pick GEMM Universal tests for FP8 data types
* Cleanup
* Add vector types and tests
* Add check_err function
* Add tensor generators
* CK_USE_GFX94 has already been set on this branch
* Fix
* Address formatting issues and leftovers
* Make fail/pass logic consistent within 01_gemm folder
Removed multiple negations in fail/pass logic to propagate `true` as the success indicator.
* Fix GPU verification reporting logic.
* Update year in copyright notice.
* Cleanup
* Use `enum class` instead of `enum`
* Remove set_property for FP8 tests
* Add vector conversions
* Fix
* Fix linker errror
* Clean up
* Fix gfx950 conversions
* Clean up
* Fix more gfx950 conversions
* Fix even more gfx950 conversions
* Narrowing the scope of PR to OCP FP8 enablement only
* Add tests for OCP FP8 vector_type storage
* Fix client examples build
* Fix typo
* Update e8m0 casting
* Rename E8M0 type
* Update unpack method
* Cleanup merge artifacts
* Enable gemm kernel on all gfx9 architectures (#227)
* clean-up
* Implement `non_native_vector_base` with `ext_vector_type` array. (#232)
* Enable support of 1, 2, 4, and 8-byte custom types in CK.
* Fix pool tests for OCP FP8 data type
* Fix build
* Add ckProfiler gemm instances for new mfma instructions and fix ckProfiler build on MI350
* fix clang format
* Add new mfma instructions and examples
* Add preprocessor directives for gfx950 specific code
* Add ckProfiler gemm instances for new mfma instructions and fix ckProfiler build on MI350
* fix clang format
* Fix clang format for the newly merged files
* Use the existing example instances for fp16 bf16 and int8
* Remove comment on new mfma instructions in MfmaInstr
* Update include/ck/tensor_operation/gpu/grid/gridwise_batched_gemm_gemm_xdl_cshuffle_v1.hpp
Co-authored-by: Andriy Roshchenko <107577548+andriy-ca@users.noreply.github.com>
* merge from public repo
* Fix ck build
* Fix ck build
* Use double for max_abs_in_val
* Move scaled_type_convert functions to a separate header (#251)
* re-enable building mha lib and gemm_universal_f8 instances for gfx950
* Update library/src/tensor_operation_instance/gpu/CMakeLists.txt
Co-authored-by: Andriy Roshchenko <107577548+andriy-ca@users.noreply.github.com>
* fix typo for CK_USE_OCP_FP8
* fix typo for CK_USE_OCP_FP8
* Add FP6 and BF6 types (#261)
* Add a rounding flag
* Add FP6 and BF6
* Add tests
Co-authored-by: Andriy Roshchenko <107577548+andriy-ca@users.noreply.github.com>
* Clean up
---------
Co-authored-by: Andriy Roshchenko <107577548+andriy-ca@users.noreply.github.com>
* fix one more typo
* Refactor E8M0 scale implementation (#262)
* Refactor E8M0 scale implementation
* Add MXFP6 and MXBF6 conversion methods (#270)
* Add conversions
* Add tests
* Add docstrings
* Add scaled conversions
* Add fp6/bf6 tests
* Remove misleading fp4 test case
* Add docstrings
* Clean up
* Address comments
* Set stricter tolerances for RNE tests
* Add missing tests
* Add native conversions to float
* Revert "Add native conversions to float"
This reverts commit 09467111f73b753c8cc3d597533b187940353dab.
* Update copyright years
* replace the fp6 with bf6 convert calls in test_bf6
* fix test_bf6
* enable smfmac test
* [MX FP8] Add Scaled Type Convert Functions for OCP FP8/BF8 data types (#271)
* Move scaled_type_convert functions to a separate header
* Introduce MX data tests
* Build MX tests only on relevant architectures
* Refactor E8M0 scale implementation
* Fix `config.h` typo
* Cleanup deprecated symbols
* Refactor `amd_ck_fp8.hpp`
* `scaled_type_convert` for `f8_ocp_t`
* Implement test for MX FP8 scaled type convert
* Implement test for MX BF8 scaled type convert
* Scaled type convert for vectors of 2 FP8 elements
* Scaled type convert for vectors of 16 FP8 elements
* Implementation of scaled conversion from F32 to F8
* Add tests for scaled conversions from FP32 to FP8
* Add documentation to the test functions
* Implementation of scaled conversion from F32x2 to F8x2
* Implementation of scaled conversion from F32x16 to F8x16
* Implementation of scaled conversion from F32x32 to F8x32
* Implementation of scaled conversion from F8x32 to F32x32
* Verified on the emulator
* MX FP GEMM - Example Template (#277)
Temporarily uses `DeviceGemmMultiD_ABScale_Xdl_CShuffle_V3` kernel and 128x128 scaling matrices.
Must be modified to use MX-native GEMM kernell with 16 or 32 component vectors per scale.
Verified on the emulator.
* Add vector support
* Add tests
* Add missing type aliases
* Fix test naming
* only build mx example for gfx950
* disable CK_USE_AMD_MFMA_GFX950 by default
* fic build for multiple archs
* fix typo
* fix typo
* Update unpack signature
* Fix merge
* Add size checks in pack function
* Add a flag
* Add conversions
* Fix build logic
* Update pack/unpack methods
* Remove unneeded AsType accessors
* Add docstrings
* Add a flag to config file
* Test the functionality of V_MFMA_F32_16X16X128_F8F6F4 and V_MFMA_F32_32X32X64_F8F6F4 instructions. (#293)
* Introduced MFMA tests
* Verified f8f6f4 MFMA Instructions
* Move flag logic to scaled_type_convert header
* Use pointers instead of array indices
* Fix a typo
* Update tests and pack functions
* Fix gemm gemm on gfx950
* Fix clang format
* restore the default gput target lists
* fix the jenkinsfile
* add missing ifdef
---------
Co-authored-by: Jing Zhang <jizhan@amd.com>
Co-authored-by: aska-0096 <haocwang@amd.com>
Co-authored-by: Jun Liu <Liu.Jun@amd.com>
Co-authored-by: Andriy Roshchenko <andriy.roshchenko@amd.com>
Co-authored-by: Rostyslav Geyyer <rosty.geyyer@amd.com>
Co-authored-by: Rostyslav Geyyer <46627076+geyyer@users.noreply.github.com>
Co-authored-by: root <root@banff-cyxtera-s83-2.ctr.dcgpu>
Co-authored-by: Andriy Roshchenko <107577548+andriy-ca@users.noreply.github.com>
Co-authored-by: jefyang1 <146495389+jefyang1@users.noreply.github.com>
Co-authored-by: jefyang1 <Jeffreyj.Yang@amd.com>
* clang format
* fix errors
* fix errors
* remove compile flags in example
* fix error
* restore cron trigger (#1863)
* recover enable-post-misched=0 for sanity issue
* add vectorloads on non-k dim for memory pipelines (#1856)
* Support for dtypes (fp8, bf8, bf16 and fp16) for the ck_tile/03_gemm example. (#1845)
* Support bf16/fb8/bf8 datatypes for ck_tile/gemm
* remove commented out code.
* Addressing code review comments and enabling universal_gemm for all the supported data types.
* Merge conflict resolution.
* Solve the memory pipeline compilation error. Merge with the new change of CShuffle
* finish the feature, pass the tests
* Fix the pipeline and add the benchmark script for other data types
---------
Co-authored-by: ThomasNing <thomas.ning@amd.com>
* revert blockwisegemm modification
* revert blkgemm pipe v2 changes.
* CK Tile - small fix to hotloop scheduler & KPack value. (#1867)
* Use SmemPack in HotLoop scheduler
* Additional debug print information
* Change KPack value.
Hardcode for now, as without AK1/BK1 there's no good way to determine
its value.
* Fix HotLoopScheduler MFMA instr parameters.
* Add a host mx gemm reference kernel (#1864)
* Add mx gemm reference kernel
* Update copyright year
* Update mx gemm example
* Use element-wise ops in the reference gemm
* External CI: enable amd-develop branch trigger (#1859)
* Apply suggestions from code review
Co-authored-by: John Afaganis <john.afaganis@amd.com>
* hotfix for ckprofiler operator
* add the 16x16 mfma instances
---------
Co-authored-by: chenjun <junchen2@amd.com>
Co-authored-by: coderfeli <coderfeli@163.com>
Co-authored-by: Illia Silin <98187287+illsilin@users.noreply.github.com>
Co-authored-by: Jing Zhang <jizhan@amd.com>
Co-authored-by: Jun Liu <Liu.Jun@amd.com>
Co-authored-by: Andriy Roshchenko <andriy.roshchenko@amd.com>
Co-authored-by: Rostyslav Geyyer <rosty.geyyer@amd.com>
Co-authored-by: Rostyslav Geyyer <46627076+geyyer@users.noreply.github.com>
Co-authored-by: root <root@banff-cyxtera-s83-2.ctr.dcgpu>
Co-authored-by: Andriy Roshchenko <107577548+andriy-ca@users.noreply.github.com>
Co-authored-by: jefyang1 <146495389+jefyang1@users.noreply.github.com>
Co-authored-by: jefyang1 <Jeffreyj.Yang@amd.com>
Co-authored-by: jakpiase <jakub.piasecki@amd.com>
Co-authored-by: kylasa <sudhir.kylasa@amd.com>
Co-authored-by: ThomasNing <thomas.ning@amd.com>
Co-authored-by: Adam Osewski <19374865+aosewski@users.noreply.github.com>
Co-authored-by: Daniel Su <danielsu@amd.com>
Co-authored-by: John Afaganis <john.afaganis@amd.com>
* updating codegen build for MIOpen access: adding .cmake for codegen component
* updating CMake
* adding in header guards for some headers due to issues with hiprtc compilation in MIOpen
* some more header guards
* putting env file in header guard
* cleaning up some includes
* updated types file for hiprtc purposes
* fixed types file: bit-wise/memcpy issue
* updating multiple utility files to deal with standard header inclusion for hiprtc
* added some more header guards in the utility files, replacing some standard header functionality
* added some more header guards
* fixing some conflicts in utility files, another round of header guards
* fixing errors in data type file
* resolved conflict errors in a few utility files
* added header guards/replicated functionality in device files
* resolved issues with standard headers in device files: device_base and device_grouped_conv_fwd_multiple_abd
* resolved issues with standard headers in device files: device_base.hpp, device_grouped_conv_fwd_multiple_abd.hpp, device_grouped_conv_fwd_multiple_abd_xdl_cshuffle.hpp
* added header guards for gridwise gemm files: gridwise_gemm_multiple_abd_xdl_cshuffle.hpp and gridwise_gemm_multiple_d_xdl_cshuffle.hpp
* fixed issue with numerics header, removed from transform_conv_fwd_to_gemm and added to device_column_to_image_impl, device_grouped_conv_fwd_multiple_abd_xdl_cshuffle, device_grouped_conv_fwd_multiple_abd_xdl_cshuffle_v3, device_image_to_column_impl
* replaced standard header usage and added header guards in block to ctile map and gridwise_gemm_pipeline_selector
* resolved errors in device_gemm_xdl_splitk_c_shuffle files in regards to replacement of standard headers in previous commit
* added replicated functionality for standard header methods in utility files
* replaced standard header functionality in threadwise tensor slice transfer files and added header guards in element_wise_operation.hpp
* temp fix for namespace error in MIOpen
* remove standard header usage in codegen device op
* removed standard header usage in elementwise files, resolved namespace errors
* formatting fix
* changed codegen argument to ON for testing
* temporarily removing codegen compiler flag for testing purposes
* added codegen flag again, set default to ON
* set codegen flag default back to OFF
* replaced enable_if_t standard header usage in data_type.hpp
* added some debug prints to pinpoint issues in MIOpen
* added print outs to debug in MIOpen
* removed debug print outs from device op
* resolved stdexcept include error
* formatting fix
* adding includes to new fp8 file to resolve ck::enable_if_t errors
* made changes to amd_wave_read_first_lane
* updated functionality in type utility file
* fixed end of file issue
* resovled errors in type utility file, added functionality to array utility file
* fixed standard header usage replication in data_type file, resolves error with failing examples on navi3x
* formatting fix
* replaced standard header usage in amd_ck_fp8 file
* added include to random_gen file
* removed and replicated standard header usage from data_type and type_convert files for fp8 changes
* replicated standard unsigned integer types in random_gen
* resolved comments from review: put calls to reinterpret_cast for size_t in header guards
* updated/added copyright headers
* removed duplicate header
* fixed typo in header guard
* updated copyright headers
---------
Co-authored-by: Illia Silin <98187287+illsilin@users.noreply.github.com>
* enable int4 scale (weight only) kernel
* format some files
* Add unit test for int4 weight only
* fixed and formatted code
* fixed
* formated
* formated
* fixed
* fixed a bug in the ckProfiler, and formatted the code
---------
Co-authored-by: mtgu0705 <mtgu@amd.com>
Without this change, the following diagnostic is generated:
a template argument list is expected after a name prefixed by the template
keyword [-Wmissing-template-arg-list-after-template-kw]
See C++17 spec [temp.names] p5.
* revert ckprofiler change
* temp save
* Add test and test pass
* test pass
* Fix bug inside rotating buffer when tensor is not packed
* bug fix
* clang format
---------
Co-authored-by: Illia Silin <98187287+illsilin@users.noreply.github.com>
* add ab_scale init support
* enabled interwave
* add scale type; update isSupport
* adjust example
* clean
* enable f8 pure gemm rcr ckprofiler
* Add gemm_multiply_multiply instances
* clang format
* Optimize for ScaleBlockMNK=128
* enable abscale f8 gemm ck profiler
* Add pure f8 gemm test suite
* Reverting to the state of project at f60fd77
* update copyright
* clang format
* update copyright
---------
Co-authored-by: root <jizhan@amd.com>
* Add bf16 instances
* Add bf16 gemm universal example
* tempsave
* Add guard to navi compilation
* workground on a specific mixed gemm instance ( bring back it when compiler fix upload)
* fix formatting condition statement issue
* solve conflict
---------
Co-authored-by: Jun Liu <Liu.Jun@amd.com>
* Add basic support for direct loads from global to LDS
* Clean the code and comments
* Add support for fp16
* Add comments
* Add check for thread cluster lengths
* Align non-direct-load fp16 example
* Small fixes
* Extend IsSupported to check for supported GPU gens
* Build examples only on the supported HW
* Do not throw when instance not supported in 04 example
* Review: Apply review suggestions
* Review: small fix
* Review: small fix
* Redesign the DPP8 GEMM kernel to use warp-wise component
* Review: Improve error messages
* Review: Remove unnecessary empty lines
* Review: Fix M, N per thread names
* Review: Rename mfma_input_type to dpp_input_type
* Review: Fix tensor adaptor; remove unnecessary element
* Review: Remove calls to dpp_gemm's MakeCDescriptor
* Review: Add blockwise doc, change function names to include dimension names
* Review: Remove duplicated code; Move Block2CtileMap alias to the top of the file
* Review: Add __restrict__ keywords
* Review: Use MatrixPadder for padding A, B, C matrices
* Review: Remove hardcoded datatypes
* Review: Change names from FloatX to XDataType
* Review: Introduce AK0 and BK0 instead of a single K0
* Review: Remove construction of dpp_datatypes object
* Review: Rename DppInstrRunner to DppLanegroupGemm
* experiment with config file
* experiment with version.h config
* add more info to version.h
* minor updates
* minor updates
* fix case where DTYPE is not used
* large amount of files but minor changes
* remove white space
* minor changes to add more MACROs
* fix cmakedefine01
* fix issue with CK internal conflict
* fix define and define value
* fix clang-format
* fix formatting issue
* experiment with cmake
* clang format v12 to be consistent with miopen
* avoid clang-format for config file
* initial stream-k implementation with example
* fix unexpected change in err
* improve a little bit performance by reorganize pipeline.
* improve perf a little bit by swizzle block idx
* add profiler
* update example
* fix spelling
* shrink karg for streamk
* support dynamic buffer using memory coherence glc_slc bit from template
* control memory coherence while construct dynamic buffer
* update reduction for streamk(not ready yet)
* Add template parameter to make_dynamic_buffer to support amd_buffer coherence setting
* fix build issue
* fix several bug
* now result is correct, everything works (but has scratch)
* remove scratch by manually reset coordinate
* update device code
* fix a bug in final reduce
* fix something in example
* update async memset
* fix enum as camel case
* modify coherence enum name
* clean code and use atomic streamk by default
* remove unused var
* throw exception if have empty pointer
* fix format
* fix CI warning
* fix type in init
* modify CI error
* filter out on gfx10+
* restore changed example code
---------
Co-authored-by: Qianfeng Zhang <Qianfeng.Zhang@amd.com>
* Use dim 0 as faster dim for writing mean/var/count workspace in batchnorm multiblock method [performance]
* Add CountDataType as template parameter in blockwise_welford
* Add utility/get_shift.hpp
* Add BatchNorm multiblock single-kernel implementation
* Add smem inline assembly based implementation of gms_init/gms_barrier/gms_reset for gfx90a
* Renaming in device_batchnorm_forward_impl.hpp
* Tiny fix in the batchnorm_fwd profiler
* Revert "Add smem inline assembly based implementation of gms_init/gms_barrier/gms_reset for gfx90a"
This reverts commit d16d00919c.
* Use the old two-kernel batchnorm multiblock method for gfx1030
* Use the old two-kernel batchnorm multiblock method for gfx908
* use the single-kernel batchnorm multiblock method only for gfx90a
* Remove get_wave_id() from utility/get_id.hpp since it is not used
* Set true for testing running mean/variance and saving mean/invvariance in the examples
* Fix to copy-right words
* Remove un-needed including in utility/get_id.hpp
* Add comments to workgroup_synchronization.hpp
* Remove un-used codes in gridwise_multiblock_batchnorm_forward.hpp
* Renaming in the kernels
* Remove un-used kernel file
* Modify Doxygen config to pick up include directories recursively
* Add DeviceMem struct to API Reference guide
* Add classes that are used in Flash Attention kernel
* Add a reference and config for generating bibliography
Co-authored-by: Philip Maybank <Philip.Maybank@amd.com>
* wmma_op + unit test
* add arch limitation to wmma test
* change arch limitation
* Refactor + Add all type unit test(int4 compile failed)
* Add f32_16x16x16_bf16 unit test
* tempsave
* tempsave
* tempsave
* runtime bug, cannot find symbol
* workaround for incorrect HIP warpSize return value
* debugging
* tempsave
* Correctness OK, waiting for optimization
* Tidy up + format
* temp save
* temp save, reproduce the v_bfi_b32 issue
* add inline asm for wmmaop test
* tidy up
* clean some debug purpose code
* discard some codes
* clang format
* clang format
* compiler issue fixed + increase tile size
* Add gridwise gemm pipeline v1/v2 selector
* Pipeline selector working, test-wise add pipeline options to one instance
* Add gemm instances
* Add debug info to DeviceGemmXdl
* Add debug info to DeviceGemmXdl_CShuffle
* Add debug info to DeviceGemmXdl_CShuffle and instances to gemm_add_add_fastgelu
* Minor fix
* Add debug info to DeviceBatchedGemmXdl and instances to batched_gemm
* set up inter-wave configuration
* use defualt loop scheduling for supported gemm ops
for blanket-applying interwave scheduling for all supported gemm ops, define macro CK_EXPERIMENTAL_DEFAULT_TO_INTER_WAVE_SCHEDULING=1. this should be discouraged though as it is not covered by CI
* Add enum PipelineVersion
* Update instances
* Format
* Fix the merge conflict
* Add flags to disable added instances
* Test disable flag check
* Disable flag check
* Enable the instances
Co-authored-by: Anthony Chang <ac.chang@outlook.com>
* reopen masking att instance due to CI is upgraded
* re-enable instances previously failed on 9110
* enable ksize-kpadding pair validity test
* add non-masked attention+permute test; expose masking boolean to attention kernel handles
* disable bench
* fix test
* move files
* bulk rename batched_gemm_masking_scale_softmax_gemm_permute to batched_gemm_softmax_gemm_permute
* format
* amend rename
* disable bench in test
* add mask/no-mask test for non-permute attention kernels
* disable broken kernel instance
* example working
add non-permuted problem statement
evaluating whether overhead comes from permutation or the extra kernel arg
* interface for bias addition without implementing it
* test and profiler running
* tidy
* mask type determined by enum class
* unify example code
* move masking specialization to its own header
* align formats
* extract helper functions
* experiment merging dims for attn w/ permute; shows perf parity with attn wo/ permute
* add tensor specialization to template args
since tensor spec packed shows perf parity when permutation isn't needed
remove redundant template args
comment on 'packed' tensor specialization
* grouped attention with input/output permute example
* format
* clean up
* refactor acc0 tile visitor
Co-authored-by: shaojiewang <wsjmessi@163.com>
Co-authored-by: Chao Liu <chao.liu2@amd.com>
* modify comment
* trim unnecessary check
* add gemm spec in kernel name
* add TNTT gemm_gemm + atten kernel instances
* refactor attention padding to better fit in unit tests
This streamlines usage where "ResetNaNToMinusInf" is now hidden from user facing device op.
Also added compile-time conditionals that load OOB value as NaN only after padding is enabled
* add adhoc padding test for atten
* shrink input value range for attention kernel validation to avoid occasional error by 1e-3
Still unsure whether this kind of deterministic floating point accurary issue is expected
or not. May want to try exact same approach as the GPU kernel in the host reference
GEMM+Softmax+GEMM function to see if the accuracy discrepancy goes away. Until then,
shrink the input value range as it is less likely to produce errors of around ~1e-3.
* attention kernel proper granular padding for all 4 dims
* IsSupportedArgument checks
* test more padded cases
* block PadK specialization in attention kernels
* workaround clang crash for gfx908
(gfx908 only) workaround for compiler crash in fused kernels on mainline #9110; #10738 seems ok
error message was "fatal error: error in backend: Error while trying to spill VGPR0 from class
VGPR_32: Cannot scavenge register without an emergency spill slot!"
this fall back to less ideal way of handle NPadding in fused attention kernel
* comment out kernels giving wrong results on MI100; MI200 doesn't seem affected
* Add threadwise and blockwise welford
* Rename gridwise op, prepare to add welford version
* implement welford and integrate welford into layernorm
* Take care of tail loop
* Fix buf when ThreadSliceK > 1
* Fix bug of merging of two empty set
* Rename clip to clamp
* 1. Fix type of count
2. Remove useless static_assert
* Do not inherit Reduction::Argument
* [What] replace __syncthreads() with block_sync_lds()
[Why] __syncthreads might wait both lgkmcnt(0) and vmcnt(0)
* Add y stride
* Rename.
DeviceLayernorm -> DeviceLayernormImpl
DeviceNormalization2 -> DeviceLayernorm
* Move literal ""_uz & ""_zu into namespace 'literals'
* Move namespace 'literals' as 'ck::literals'
Co-authored-by: Po-Yen, Chen <PoYen.Chen@amd.com>
Co-authored-by: Chao Liu <chao.liu2@amd.com>
