When there are multiple workgroups contributing to a tile, when using
atomics, there may be round off error in cases where the accumulator
type is not the same as the C type. To compute an error tolerance for
test validation, the Stream-K Tile Partitioner has a function called
estimate_num_wgs_per_tile to estimate the number of workgroups per tile.
That said, this function only provides an estimate. In some cases for
DP+2TSK, the function returns 1 rather than the more accurate value of
2.
Thus, this change updates the estimate_num_wgs_per_tile function to
explicitely return the value of 2 in cases for DP+2TSK to ensure that we
have a better error tolerance to avoid test failures due to round-off
error.
* Remove old CK Tile Stream-K implementation
The original CK Stream-K implementation was based on old CK's Stream-K
block to C tile map. However, this implementation did not align with the
original Stream-K paper. Thus, we implemented a new tile partitioner and
associated Stream-K kernel, which was placed in the reboot namespace.
Now that the new Stream-K implementation is ready, this change removes
all artifacts of the old implementation. Specifically, the following
changes were made:
- Removes old Stream-K tile partitioner from CK Tile
- Removes the reboot namespace such that the new implementation resides
in the ck_tile namespace only.
- Adds tests for bf8 and fp8 using the new implementation
- Removes tests for the old implementation
- Remove the v2 suffix from the new CK Tile Tile Partitioner
derived classes.
- Updates Stream-K Kernel ops file to use /** commenting style.
* Remove v2 from tile partitioner validation function names
1. Enable grouped_gemm_quant and gemm_streamk on gfx12
- test_ck_tile_streamk_smoke is kept on gfx9, since it looks someone is still working on it.
2. Update warp tile size in grouped_gemm_quant and gemm_streamk unit test
3. Reduce gemm tile size to pass the build on gfx12 in test_gemm_streamk_reboot_types.hpp
* Reapply "[CK_TILE] Non-K Major from old CK to CK-Tile (#2442)" (#3017)
This reverts commit e4298e55c7.
* WIP
* take Y2 as the AK1/BK1 value, that is the 'vector size' after shuffle
* use get_n_lds_banks()
* clang-format
---------
Co-authored-by: Adam Osewski <19374865+aosewski@users.noreply.github.com>
* Add missing copyright statements
* Use ck_tile::host_tensor_descriptor instead of a custom lambda
* Refactor use of check_data_type in test classes
* Use TEST_SUITE_NAME with TYPED_TEST_SUITE
* Remove an unused namespace
* Make dim3 const
* Add BF8 x BF8 tests for CompV3 in test_gemm_pipeline_kernel_types.hpp
* Add F8 x BF8 tests for CompV3 in test_gemm_pipeline_kernel_types.hpp
* Add BF16 x I4 tests for CompV3 in test_gemm_pipeline_kernel_types.hpp
* Add BF16 x BF16 tests for CompV3 in test_gemm_pipeline_kernel_types.hpp
* Add BF8 x I4 tests for CompV3 in test_gemm_pipeline_kernel_types.hpp
* Add F8 x I4 tests for CompV3 in test_gemm_pipeline_kernel_types.hpp
* Add F16 x I4 tests for CompV3 in test_gemm_pipeline_kernel_types.hpp
* Skip failing tests of F16 x I4 for CompV3 with K == 2 * K_Tile
* Add missing precision type combinations to CompV4 from CompV3
* Move the INT8 tests around for consistency with KernelTypesCompV3Wmma
* Add missing precision type combinations to CompV3Wmma from CompV3
* Remove the basic and universal tests and their dependencies
* On __gfx950__, avoid using transposed loading of A with datatype pk_int4_t of B
* Use ADataType and BDataType instead of ComputeDataType for WarpGemm
* Explicitly set some return types to void
* Use more general typenames in InterleavedPKTypeLoader
* Add load_interleaved_pk_type.hpp to common.hpp
* Use std::is_same_v in load_int4_tile
* Add handling of LoadTranspose to load_int4_tile
* Factor out common code in several places using load_int4_tile
* Add support for pk_int4_t using load_int4_tile
* Fix formatting
* Introduces the new partitioner to implement the reduction StreamK kernel
* Add more doc text to functions
* Add persistent-dp option to streamk example
* Update example/ck_tile/40_streamk_gemm/README.md
This change replaces pipeline macros like CK_TILE_PIPELINE_COMPUTE_V3,
CK_TILE_PIPELINE_MEMORY, etc in the CK Tile examples with a common enum
called GemmPipeline to reduce code duplication.
* add tensorwise quant in grouped gemm
* fix example issue
* update test cases
* format codes
* clang format
* use GTEST_FAIL
* add bquant to grouped_gemm
* add tensorwise quant in grouped gemm
* fix example issue
* update test cases
* format codes
* clang format
* use GTEST_FAIL
* fix a bug in test_grouped_gemm_util
* skip test when use wmma on grouped_quant kernel
* change cmake
* fix a bug in test_grouped_gemm_util
* skip test when use wmma on grouped_quant kernel
* change cmake
* tests(quant_grouped_gemm): add unit tests to cover bquant in grouped_gemm
* Update test/ck_tile/grouped_gemm_quant/test_grouped_gemm_util_quant.hpp
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
* Update example/ck_tile/17_grouped_gemm/quant_grouped_gemm.hpp
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
* feat: add bf8 support
* chore: remove unnecessary decltype usage
* chore: add default quant_mode to function signature as fallback
* fix: pass correct runtime pipeline params in grouped_gemm bquant kernel
Calculate has_hot_loop, num_loop, and tail_number on device side for each
GEMM problem instead of using default values. This fixes incorrect results
when different problems in the group have different K dimensions.
* chore: set default quant mode in function signature
* test: add additional test cases to cover edge case of no hotloop
* change code based on comments
* WIP: bquant preshuffle b compiles but gives numerical error
* feat(grouped_gemm_quant): bquant with preshuffleB support added to grouped_gemm example & kernel
* refactor: refactor code after merge commit
* chore: remove print statements
* test(grouped_gemm): split test cases by quant mode to reduce compilation time and add bquant-preshuffleB mode test cases
---------
Co-authored-by: kyle-256 <Kyle.Zhao@amd.com>
Co-authored-by: ThomasNing <thomas.ning@amd.com>
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
Fixes LDS bank conflicts on gfx950 for universal gemm v3 pipeline
Replaces hardcoded LDS layer calculations with dynamic computation using the new architecture helpers
Adds architecture-specific helper function get_n_lds_banks()
Changes function attributes from CK_TILE_HOST_DEVICE to CK_TILE_DEVICE in universal gemm policy
* HasHotLoop is a constexpr
* Remove an unused function
* Remove some unused include statements
* Add implementation and tests for fp8 x bf8 weight preshuffle GEMM
* Add implementation and tests for fp8 x bf8 in CK Tile basic and universal GEMMs
* Remove two barrier calls that HotLoopScheduler already calls
* No need to suppress a variable that hasn't been declared
* Replace six arg_parser arguments with constexpr literals
* Simplify run_gemm_test_prec_type
* The strides don't need to be passed via arg_parser as we use their default values
* The layouts don't need to be passed as arguments twice
* Pass M N and K as regular arguments, not using the argument parser
* We can now remove the argument parser
* Add a common file for precision types to be used in testing
* Convert basic and universal GEMM tests to use gtest
* Make GemmConfig a test parameter, and form test cases as the cartesian product GemmConfigs x PrecTypes
* Add GemmConfigComputeV4 to the GEMM configs to run the universal tests on
* Added a changelog entry
* Add missing copyright statements
* ifndef-define-endif is not needed with pragma once
* Fix a comment
* Add F8 x BF8 tests for CompV4 in test_gemm_pipeline_kernel_types.hpp
* Disable the unreliable test MoeSortingCase4
---------
Co-authored-by: Adam Osewski <19374865+aosewski@users.noreply.github.com>
* Pass hdim to tile_example_fmha_fwd in fp8 tests
* Add WMMA support to fwd FMHA pipelines
* Tune tile sizes a bit for less spilling
fp16 256 is still quite slow
* Fix Q grad tile distribution for warp size = 32 and hdim >= 256
With AccDataType = float and warp size = 32, K0 becomes 0, K repeat is required to correcty distribute the tile.
* Use code based on BlockDropout in BlockDropoutBwd
* Fix split KV combine kernel for gfx12 (warp size 32) and make it more universal
* Fix LSE LDS tensor descriptors: kMaxSplits and kM0 were swapped, it worked on gfx9
because they both equal to 8 while on gfx12 they are 8 and 4;
* Fix Oacc LDS tensor descriptor: it was transposed even though its shape=[4 * kM0, kN1],
it worked on gfx9 because 4 * kM == kN1 == 32;
* Removing these hidden dependecies allows to support:
* any number of warps (power-of-2), not only 4;
* kN1 = 16, not only 32;
* any number of splits;
* Rename ids like o_acc_4 and Oacc4 to eliminate confusion: kNumWarps doesn't have to be 4 now
* Replace hard-coded kN1 in dispatch code with the requested tile size
* Add gfx12-specific tile sizes for split KV
* Pass GPU architecture to kernel generation scripts
This is still a temporary solution.
* Build and run FMHA CI tests for gfx12
* Fix issue after merging
* Fix bwd tile sizes
The current pipelines always read only one tile K and V tile, this
requires bk0 == bhdq and bk2 == bhdv (kK0 == kQKHeaddim and
kK2 == kVHeaddim).
* Use hardware f32->f8 on gfx12, remove v_perm
__builtin_amdgcn_perm is not needed because
__builtin_amdgcn_cvt_pk_fp8_f32 allows to specify which word (16 bit of
32-bit dword) is used to store results (two f8 values).
* Update changelog
* Add WMMA support to pagedkv
* Fix scripts after rebasing
* Support 16x16 (MFMA, WMMA) and 32x32 (MFMA) tiles in fwd and bwd BlockDropout
Add comments with dropout implementation details
Fix performance regression of fwd+dropout
* Remove some usage of type punning (reinterpret_cast with ref or ptr) in Philox;
* "scalarize" seed and offset, they may come either from kernel args or from device memory
(presumably loaded with vector loads).
These changes help the compiler to procude more optimal code and reduce register spilling.
Use WarpGemmDispatcher instead of explicit WarpGemmMfma... to get CWarpDstrEncoding
Use code based on BlockDropout in BlockDropoutBwd
Refactor BlockDropout (fwd)
Implement BlockDropout (fwd) for WMMA
Originally BlockDropout only supported 32x32 tiles (IsWG32 = true),
this version supports 16x16 tiles.
If MPerBlock > MWarp * 16, it can generate numbers for two 16x16 tiles, similarly
to BlockDropoutBwd.
Implement BlockDropoutBwd for WMMA
Remove MakeRandValLds* functions unused in BlockDropoutBwd
Remove unused Run overload from BlockDropoutBwd
* Fix regression with philox seed and offset when they exceed 32-bit int
__builtin_amdgcn_readfirstlane works with 32-bit values, seed and offset
are 64-bit so they get truncated.
* Fix names after cherry-picking
* Fix selection of a fallback tile based on bm0
The assumption that the largest bm0 == 128 is not always true for
current fp32 tiles.
* Do not use filters related to qr_async_trload
They disable tiles/pipelines which are valid for gfx12.
* Use different dstr encoding when C is transposed
* Do not call GetQKBlockGemm (and hence WarpGemmDispatcher) in host code
Some WarpGemmDispatcher instantiations are defined only
for specific archs and undefined on host.
Calculations related to sched barriers are moved from Pipeline's public
fields into pipeline's operator().
* Fix incorrect name WarpGemmMfmaFp8Fp8F32M32N32K16SwizzleBTransposedCDistribution
Correct name is WarpGemmMfmaFp8Fp8F32M32N32K32SwizzleBTransposedCDistribution
because it's 32x32x16 with IterateK = 2 so K = 32, also all tiles used
in codegen scripts are 32, 32, 32.
* Generalize usages of WarpGemmDispatcher for MFMA and WMMA
WarpGemmMfmaFp8Fp8F32M32N32K32SwizzleBTransposedCDistribution is still
used explicitly becaus of swizzle factor = 4.
* Mark has_load_tr as maybe_unused
There are no transpose loading for RDNA.
* Remove CK_TILE_USE_MFMA/WMMA from fmha-related code
* Detect BlockSize on host based on warp size of the current device
If kBlockSize == kNumWarps * get_warp_size(), the kernel is launched with
kBlockSize / 2 because on host get_warp_size() == 64 always.
* Fix calculation of grid size for combine kernel with warp size = 32
* Add missing includes and header
* Support multiple archs in one binary for fwd
* Support multiple archs in one binary for fwd_splitkv, fwd_appendkv, pagedkv_prefill
* Support multiple archs in one binary for bwd
* trload kernels are compiled only for gfx950;
* instances with padding are checked after instances without padding so
they can be used as fallbacks (similarly to fwd);
* Extract common code from register_traits
* Revert "Fix regression with philox seed and offset when they exceed 32-bit int"
To simplify merging , the proper fix is in develop already.
* Support new numerical d paddings in trait ordering checks
* Build fp32 tests only on gfx9
* Do not use hardcoded M0 = 64 for dot bwd kernel
* Use textwrap.indent from standard library
* Make fp8 pipelines on gfx12 consistent with gfx9
* Update tests for current pipelines
* Make ninja check more responsive in CI
ninja buffers output so this job looks hanging.
* Support fp8fp32 by limiting O vector size
The fp32 output type requires storing 8 * sizeof(float) = 32 bytes,
which is not implemented (here 8 is the number of C values per lane for
v_wmma_f32_16x16x16...).
* Remove unused cmake options
* Unify including amd_buffer_addressing.hpp/_builtins.hpp
* Temporarily use amd_buffer_addressing.hpp on >=gfx10
amd_buffer_addressing_builtins.hpp uses inline asm for loads/stores
which is not compatible with >=gfx10:
* 1 scalar for exec masks instead of 2,
* gfx12 uses different instruction names etc.
* Update asm in bf16 conversions to work with warp 32
* Do not generate splitkv/appendkv with vlayout=col for consistency with fwd
* Add arch tags to kernels/host funcs, compile for each arch separately
* Add kM0 to fmha_bwd_dot_do_o kernel name to match filename
* Add workaround for miscompilation of bwd with padded hdim
SWDEV-559729: v_wmma instructions can be incorrectly placed in divergent
branches used to store padded tensors (when some lanes are inactive due
to padding). Inline asm with dummy dependencies on VGPRs of the tensors
prevents the compiler doing this.
* Fix add_gtest_executable for absolute paths
Some tests (like gemm_tile_engine) pass absolute paths to source files.
In CI the branch name is a part of the root dir, and if the branch name
contains "wmma", "xdl" etc., files can be incorrectly excluded.
* Run only hdim 128 smoke tests for fp8fp32
There are no instances for hdim 64 and 256.
* Format py with ruff to simplify merging develop
* Fix incorrect var name
* Codegen for gfx9,gfx950 when --targets is not specified
Aiter and Pytorch require changes for passing their targets to the codegen scripts.
With this temporary solution the files are generated but not all of them
have to be really built (depending on the used --offload-arch=).
* Combine arch-related values into ArchTrait
This more centralized approach removes duplication of various formatting templates.
* Try a workaround for Jenkins error "groovyjarjarasm.asm.MethodTooLargeException: Method too large"
Some code is extracted into a function.
* Persistent Stream-K Kernel Implementation
This change implements an operator() function in the
reboot::StreamKKernel class that is enabled when the Persistent flag is
set to true. In this case, the data-parallel portion and the Stream-K
portion of the kernel are fully persistent.
The changes were made in the reboot namespace. A future PR will remove
the old Stream-K kernel class and remove the reboot namespace.
* Unit Tests for Persistent Stream-K Kernel
This change contains the inital test suite for the Persitent Stream-K
Kernel. The files contain "reboot" in the name; a future PR will remove
tests for the old Stream-K Kernel and remove the "reboot" naming.
A future commit will add tests for the non-persistent kernel.
Also added estimate_num_wgs_per_tile to the StreamKTilePartitionerBase
class. This allows us to estimate the number of accumulations done per
macro tile in C to use during validation when computing relative and
absolute tolerance.
* Adding implementation for the Non-Persistent Stream-K kernel
This code is adding the operator() function for the Non-Persistent Stream-K
kernel. Persistency of the kernel is determined through a template argument.
The Non-Persistent kernel will allocate additional workgroups for the data
parallel section, leading to a different structure for processing the data
parallel and Stream-K sections.
There has been an addition to the TilePartitioner to get access to the whether
Persistent has been set to true or false in the StreamKKernel.
* Adding in the tests for the Non-Persistent Stream-K kernel
* Refactor Stream-K Reboot Unit Tests
This commit makes the following changes:
- Update test cases to determine M, N, and K based on the number of CUs.
This ensures that each test case is one of Edge Case, SK Only, DP
Only, or DP + 2 Tile SK regardless of the architecture.
- Since the DP + 2 Tile SK test case takes long to run, this change
moves this case into a separate .inc file and labels it as an extended
test.
- Since the extended test takes > 30 seconds to run, this test is added
to the list of regression tests.
* Fix spelling errors in comments for test cases
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
* Changes based on review
Removed const volatile for typenames
Set up alias for is_tuple_t
Naming changes for clarity: GemmCommon -> BaseGemm
Moved std::enable_if_t out of template parameters and changed to a return type for operator()
Added constructor for StreamKKernelArgs to clarify UniversalGemm inheritance
---------
Co-authored-by: Emily Martins <emily.martins@amd.com>
Co-authored-by: Christopher Millette <63608002+cgmillette@users.noreply.github.com>
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
The following changes were made
- Renamed iter to iter_start
- Renamed tile_iter to tile_iter_start
- Moved documentation from member variables to getters
- Removed double underscore from extra_iters_before_me variable
- Defined parent header in impl file
- Removed unused inlcudes
There are 2 derived structs based on whether Stream-K is persistent or not.
If it's persistent that means that both the data parallel and Stream-K sections
are data parallel. If it's non-persistent that means that only the
Stream-K section is persistent, while the data parallel section will have
separate workgroups allocated for it. Both structs will have a template
argument for Persistent.
The 2 derived classes will inherit common variables and functions from the
Stream-K TilePartitioner base class. There are additional variables for the
differing data parallel sections that will be added to each derived class,
that are in charge of the indexing/bookkeeping for the data parallel sections.
The only additional function that will differ between the 2 structs is GridSize(),
as the non-persistent will allocate extra workgroups for data parallel.
Unit tests for the derived structs are included.
To better align with the original Stream-K paper, this change implements
a new Stream-K tile partitioner base class. This class will handle the
Stream-K setup that is common to both a persistent and non-persistent DP
section. A later change will implement derived classes to handle the
differences between persistent and non-persistent DP.
This change also includes unit tests for the base tile partitioner.
* Enable the adapted LDS B layout for Row-Major
* fix formatting
* Implement specialized col-major A LDS block descriptor
* Fix formatting
* Use VecLoadSize for AK1/BK1
* Fix some thread access pattern values
* Use GetVectorSizeA for A
* Fix formatting
* Add extra condition to avoid division by zero
* disable layout for wave32
* remove extra else
* fix formatting
* Fix formatting
* Rename one remaining TileDistributionEncodingPattern2D
* Use integer ceil division
* revert remod.py changes
* also revert utility.hpp
* use getA/BTileAccessPattern everywhere
* use integer_divide_ceil for AK0 too
---------
Co-authored-by: Adam Osewski <19374865+aosewski@users.noreply.github.com>
Co-authored-by: Adam Osewski <Adam.Osewski@amd.com>
* Add initial fp16_mem_128x128x32_2x2x1_32x32x16_NonPersistent test suite
* Account for stride when computing K offsets for A and B tensor
This change ensures that the correct stride is used when computing the K
offsets into the A and B tensors in the Stream-K Kernel's operator()
function. This ensures that the kernel executes correct regardless of
whether A and B are row or column major.
* Move helper code to test_gemm_streamk_util.hpp
* Separate tests into smoke/regression/extended. Add bf16 datatype
* Run clang-format
* Refactor combinatorial macro expansion and naming
* Adjust the initialization values to account for better tolerance on bf16
* Correct BF16 datatypes in comments
* Move the extended tests under the REGRESSION_TESTS label
* Apply suggestions from code review
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
---------
Co-authored-by: Emily Martins <emily.martins@amd.com>
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
* reuse local prefetch logic from compute v4 pipeline
add single-tile test
explicit lambda capture
reuse lds block descriptors from base policy for the transposed case
match the test case kernel configuration with compute v4
* add comments
* WIP: add memory pipeline boiler plate code that compiles and works for one block
* WIP: tail handling works for memory pipeline
* WIP: numerical errors appears to have gone by adding block_sync_lds()
* fix: numerical error with memory pipeline by adding block_sync_lds() and new tail handler
* refactror: remove debug print statements and lints
* fix: remove redundant sync barriars
* chore: remove lint
* fix: remove unused code from tile handler and remove redundant block_sync_lds()
* fix: correct parent struct name for memory pipeline
* fix: remove static assert check from parent struct and add it to child struct because not all child structs needs to static assert
* fix: defer block sync lds to just before prefill
* check in pipeline and policy
for async load in mi350, need to make sure TileAccessPattern is warp_raked or block_raked
solve merge conflicts
* fix cmakelists
* make it build
* fix? buffer async fence
* relax fences; it appears it only is needed between pairs of ping-pongs
* remove fences
* remove fences
* cleanup and reformat
* add steps annotations
* comment all pipeline steps / remove unexplainable syncs
* clang-format
* add comment
* cleanup kernel types for test
* fix comment
* fix hardcoded warp size
* faithfully copy block gemm from compute v4 policy to async policy
* make async test gfx950 only
* fix cmake logic
* set separate compile options for async
* refine comment in policy
* try update hotloop scheduler
* cleanup comments
* test more K block sizes
* unhardcode Ks, sort of
* add large odd test case
* fix build for quant
* add comment to hot loop scheduler and rename enum
* reformat
* reword the pipeline description
* reformat
* address review / add static asserts / typo fix
* update changelog
* feat(grouped_gemm_multi_d): add new example that integrates grouped_gemm and multi_d_gemm feature
* refactor: grouped_gemm_multi_d relies on grouped_gemm_kernel
* tests(grouped_gemm): grouped_gemm test suite passes with minor adjustments
* fix: segfault fix by passing correct parameters for d tensors
* style: clang format
* WIP: host code for grouped_gemm_multi_d persistent kernel compiles but segfaults
* feat(grouped_gemm_multi_d): add functionality to run persistant kernel
* feat(grouped_gemm_multi_d): add new example that integrates grouped_gemm and multi_d_gemm feature
* refactor: grouped_gemm_multi_d relies on grouped_gemm_kernel
* tests(grouped_gemm): grouped_gemm test suite passes with minor adjustments
* fix: segfault fix by passing correct parameters for d tensors
* style: clang format
* fix: incorrect validation method and Dtensor layout in test suite
* docs: improved README text based on review comments
* fix: parameterize NumDTensor in GroupedGemmHostArgs and remove lint
* initial commit
* remove extra files
* fixing errors
* updated ReadMe file for mapping of diff quants with diff configs
* addressing review comments
* addressing review comments
* Resolved merge conflicts
* [CK TILE GEMM] Replace get_preshuffle_or with is_quantpreshuffle_enabled
The get_preshuffle_or was not working as expected, which led to incorrect behavior
in the quantization preshuffle process. This change replaces it with the more reliable
is_quantpreshuffle_enabled function to properly determine when preshuffle should be applied.
* initial commit
* debugging
* working fp8 for init constant
* fp8 working with all inits
* updated block level code with comments
* changing the loop iter
* debugging
* debugging
* debugging
* code fix
* code clean up
* clang formatted
* Add comment
* code cleanup
* clang formatted
* merge conflicts fixes
* applying the latest int4 changes to the piepline
* fixing test code for updated traits
* Adding gtest
* review comments addressed
* addressing review comments
* remove c++20 code
* added flush cache changes
---------
Co-authored-by: Cong Ma <congma13@amd.com>
Co-authored-by: root <root@banff-cyxtera-s73-2.ctr.dcgpu>
* Support 16x16 (MFMA, WMMA) and 32x32 (MFMA) tiles in fwd and bwd BlockDropout
Add comments with dropout implementation details
Fix performance regression of fwd+dropout
* Remove some usage of type punning (reinterpret_cast with ref or ptr) in Philox;
* "scalarize" seed and offset, they may come either from kernel args or from device memory
(presumably loaded with vector loads).
These changes help the compiler to procude more optimal code and reduce register spilling.
Use WarpGemmDispatcher instead of explicit WarpGemmMfma... to get CWarpDstrEncoding
Use code based on BlockDropout in BlockDropoutBwd
Refactor BlockDropout (fwd)
Implement BlockDropout (fwd) for WMMA
Originally BlockDropout only supported 32x32 tiles (IsWG32 = true),
this version supports 16x16 tiles.
If MPerBlock > MWarp * 16, it can generate numbers for two 16x16 tiles, similarly
to BlockDropoutBwd.
Implement BlockDropoutBwd for WMMA
Remove MakeRandValLds* functions unused in BlockDropoutBwd
Remove unused Run overload from BlockDropoutBwd
* Fix regression with philox seed and offset when they exceed 32-bit int
__builtin_amdgcn_readfirstlane works with 32-bit values, seed and offset
are 64-bit so they get truncated.
* Add F32 MFMA warp gemms
* Support f32 in fwd FMHA
* Implement transpose_vectors for 4-byte types (float)
* Fix unexpected implicit f32->uint32 cast in buffer_store<4>
__builtin_amdgcn_raw_buffer_store_b32 expects unsigned int but float was passed (implicitly casted to uint).
mbuf_t types in other buffer_store<> are changed for consistency.
* Support F32 in bwd FMHA
hdim = 256 is disabled for now because it uses too much memory on gfx90a
* Support Headdim = 48 (divisible by 16) in fwd
* Add fp32-specific receipts (800 and 801)
* Tune fwd tiles
* Tune bwd tiles
* Use small tiles only for small seqlen_q
* Fix after rebasing
* Fix selection of a fallback tile based on bm0
The assumption that the largest bm0 == 128 is not always true for
current fp32 tiles.
* Remove constraints and adjust filtering for fp32
Custom constraints are no longer needed because now the smallest tile
is selected automtically based on seqlen_q.
Filters related to qr_async_trload disabled valid fp32 tiles.
* Add fp32 tests
* Make splitkv and appendkv compile for fp32 only
There are no instances yet, but API still must compile when only fp32 is
requested.
* Remove unimportant f32 instances
* Add test_ck_tile_fmha_*_fp32 to REGRESSION_TESTS
* Replace magic numbers with a constant, improve comments for dropout
* Update changelog
* Fix condition that dq_acc must be set to zero when mask is used
The change was introduced in #2799
* Replace warp_uniform with recently added amd_wave_read_first_lane
* Add hdim = 96 and 192 to fwd
* feat(grouped_gemm_multi_d): add new example that integrates grouped_gemm and multi_d_gemm feature
* feat: generalized grouped_gemm_kernel.hpp
* feat: generalized grouped_gemm_kernel.hpp even further by removing hardcoded 0
* refactor: grouped_gemm_multi_d relies on grouped_gemm_kernel
* tests(grouped_gemm): grouped_gemm test suite passes with minor adjustments
* fix: segfault fix by passing correct parameters for d tensors
* docs: add multi d info and trim down outdated content
* tests: add unit tests for grouped_gemm_multi_d and minor changes in grouped_gemm related test for compatibility
* style: clang format
* fix: incorrect validation method and Dtensor layout in test suite
* Update grouped_gemm example and pipeline
* find the root cause error in did not enable the transpose in gfx950 correctly
* Fix v3 pipeline, row and col major
* Disable f8 datatype tests, it fails on gfx950
* fix the abd test by clear the runtime argument unsupported
---------
Co-authored-by: AviralGoelAMD <aviral.goel@amd.com>
Co-authored-by: Mateusz Ozga <mateusz.ozga@amd.com>
* Have a workable version for SGPR
* have a workable version for atomic add
* Revert "have a workable version for atomic add"
This reverts commit 792377a590c26cfff9c8f545d9a9e8484a7422eb.
* substitute with the new sgpr read api
* update the CHANGELOG
* have a workable version for atomic add
* Revert "have a workable version for atomic add"
This reverts commit 792377a590c26cfff9c8f545d9a9e8484a7422eb.
* change to static for logic
* have a workable version for atomic add
* Revert "have a workable version for atomic add"
This reverts commit 792377a590c26cfff9c8f545d9a9e8484a7422eb.
* Factor out the three separate copies of load_interleaved_pk_type into a common utility class
* Add preprocessing with optional cache flushing and clearing of output for k_batch > 1 to the weight preshuffle GEMM example
* Remove a duplicate function
* Add support for B tensor type pk_int4_t for the weight preshuffle GEMM, with tests included
* I4 support introduced more failing test cases that mirror the existing ones for F8
* Simplify the check for which tests to skip (they all have F8 as A tensor type)
* Add a changelog entry
* add the test for v2 wp pipeline, polish the code, add the support of int4 for v2 wp pipeline
* have a workable version for atomic add
* Revert "have a workable version for atomic add"
This reverts commit 792377a590c26cfff9c8f545d9a9e8484a7422eb.
---------
Co-authored-by: ThomasNing <thomas.ning@amd.com>
* fix(grouped_gemm): numerical errors on gfx950 by correctly calculating the tail num
* WIP: add temp config to stress test numerical error correction
* refactor: remove comments
* Change splitk_batch_offset parameter to k_size in UniversalGemmKernel::MakeGemmTensorViews function
Prior to this change, the splitk_batch_offset parameter of
MakeGemmTensorViews had type SplitKBatchOffset. But, the only member
variable of the SplitKBatchOffset class used in the MakeGemmTensorViews
function was splitted_k (an int32_t). The splitted_k value was used as
part of defining the dimensions of the tensor view. That said, for
Stream K, we do not need to use the SplitKBatchOffset class since we are
not using Split K. Thus, this commit changes the splitk_batch_offset
parameter to a int32_t called k_size. This will avoid the constraint of
requiring a caller of MakeGemmTensorViews to use the SplitKBatchOffset
class while still providing the same functionality. Calls to
UniversalGemmKernel::MakeGemmTensorViews have been updated accordingly.
* StreamK Kernel RunGemm Implementation
Stream K cannot simply use UniversalGemmKernel's RunGemm for the
following reasons:
1. The UniversalGemmKernel::RunGemm function computes num_loop based on
a static function of the TilePartitioner. That said, for Stream K,
num_loop must be computed using a member function (namely
GetCurrentIterLength from PR #2708).
2. The UniversalGemmKernel::RunGemm function requires the use of a
SplitKBatchOffset object which is not used for Stream K since we are
not using Split K.
Thus, this change adds a RunGemm function in the StreamKKernel class.
* initial implementation for operator() for StreamKKernel: adding stream-k algorithm and calls to RunGemm
* Fix indexing and offset issues for StreamK
These changes do the following:
- Ensure offsets along the M and N dimensions are multiplied by
MPerblock or NPerBlock, respectively. This ensures tile window origins
are at the correct locations.
- Fix bug in the tile partitioner's GetTileIdxWithOffset. Now, we apply
divmod to the given references to ensure correct values are available
to the caller.
- Added documentation in the Stream-K operator()
* Initial gtests for Stream-K
These changes add an initial gtest suite for the CK Tile Stream-K
kernel. Currently, due to bugs in the StreamKTilePartitioner (which will
be handled in a future PR), there are validation issues for certain
cases which may differ on different architectures. Thus, we opted to run
cases that are only fully data-parallel (skipping others). A guard was
added to Stream-K's IsSupportedArgument method to ensure that callers
are aware of this constraint. Additionally, to ensure testing
reproducibility, options for setting the number of CUs and occupancy
were added to MakeKernelArgs.
* Use GemmPipeline operator() variant that takes hot loop and tail num
In Stream-K, the num_loop value varies per WG and per iteration of a
Stream-K loop. So instead, we use the version of the GemmPipeline's
operator() function that takes in has_hot_loop and tail_num. This is
similar to what is done in Grouped GEMM.
* changes from review: comments, move readfirstlane, remove ifndef
* Switch direction of C tensor traversal & add padding guard
Prior to this change, WGs travelled backwards through their assigned
macro tiles in the C tensor. For instance, if WG0 is responsible for C
tiles 0 and 1, it would first visit tile 1 then tile 0. This means that
the iter_end decrements in each iteration of the stream-K while loop.
Since we are working with unsigned integers, the subtraction operation
may not be safe. Thus, this change makes is such that WGs travel forward
so that their iter_start is incremented and their iter_end remains
fixed.
Additionally, we added a guard against WGs that are neither sk_blocks
nor dp_blocks to ensure such WGs do not participate in the GEMM.
Together, these changes make is such that the algorithm is correct when
sk_blocks is greater than zero.
* Disable StreamK_M256_N256_K256_SKBlocks12 test case
This instance involves >=3 WGs contributing to each macro tile in C. Due
to the use of atomics, this is resulting in precision errors. These
errors will not persist once the reduction strategy is implemented. We
will re-enable this test then.
---------
Co-authored-by: Astha Rai <astha.rai713@gmail.com>
