[CK_TILE][FMHA] Support gfx11
## Motivation
Add support of gfx11 architectures (RDNA3) to FMHA.
## Technical Details
Distributions (matrix elements to lane registers mapping) of gfx11 WMMA
are completely different from distributions of gfx9 MFMA and gfx12 WMMA.
There are two cases in FMHA where this difference matters:
* usage of results (matrix C) of one GEMM as input (matrix A) of another
GEMM.
* random number generation for dropout (implementation for gfx9 MFMA,
gfx12 WMMA and host validation produce the same results).
Both cases are solved by a special remapping implemented using
`__builtin_amdgcn_permlanex16` and `__builtin_amdgcn_perm`.
Additional changes:
* FMHA tests are now build and run only for those types for which
instances exist (gfx11 supports only fp16 and bf16).
* Two fixes for uninitialized values (`mask.sink` and
`do_fp8_static_quant`): they may contain garbage resulting in incorrect
dispatching logic, sometimes tests report that there are no instance
available for current parameters.
* Small fix to remove expcnt(0) from s_waitcnt instruction on gfx11 when
they are not requested (i.e. every time), likely has no effect on
performance but makes disassembly a bit clearer.
## Test Plan
```
ninja test_ck_tile_fmha
bin/test_ck_tile_fmha_fwd_fp16
bin/test_ck_tile_fmha_fwd_bf16
bin/test_ck_tile_fmha_bwd_fp16
bin/test_ck_tile_fmha_bwd_bf16
```
## Test Result
All tests must pass (some tests may be skipped).
## Submission Checklist
- [x] Look over the contributing guidelines at
https://github.com/ROCm/ROCm/blob/develop/CONTRIBUTING.md#pull-requests.
[CK] Add FP8 KV_BLOCKSCALE support for batch prefill
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit
Implement per-page K/V quantization for paged attention:
- Add KV_BLOCKSCALE enum to BlockAttentionQuantScaleEnum
- Use exp2 shift trick to eliminate explicit P scaling overhead
- Prefetch physical pages offset for KV cache, overlaps with
computations
## Proposed changes
Please describe the motivation behind the pull request, whether it
enables a new feature or fixes a bug. If there are associated pull
requests or issues, please link them to the pull request.
## Checklist
Please put an `x` into the boxes that apply. You can also fill these out
after creating the PR. If you're not sure, please don't hesitate to ask.
- [ ] I have added tests relevant to the introduced functionality, and
the unit tests are passing locally
- [ ] I have added the test to REGRESSION_TESTS list defined at the top
of CMakeLists.txt in tests/CMakeLists.txt, **IF** the test takes more
than 30 seconds to run.
- [ ] I have added inline documentation which enables the maintainers
with understanding the motivation
- [ ] I have removed the stale documentation which is no longer relevant
after this pull request
- [ ] (If this change is user-facing) I have added release notes which
provide the end users with a brief summary of the improvement from this
pull request
- [ ] I have run `clang-format` on all changed files
- [ ] Any dependent changes have been merged
## Discussion
If this is a relatively large or complex change, feel free to start a
discussion by explaining why you chose the solution you did and what
alternatives you considered
* add block scale parameters to kernel
* add block scale to kernel
* add smoke test
* format
* Revert "format"
This reverts commit 356c3c9706.
* only format my code
* format py
* fix auto not allowd in function prototype
* change instance tttt to ttff
* fix structured binding issue
* change s_acc elementwise op
* async pipeline add block scale
* add quantation P using shift exp2
* precompute (m - shift) once per row
* change blk scale seqstrt ptr name
* fix some name
* fix for deduction guide
* fix some comments
* add P scale to qr_ksvs_pipeline
* add comment to idx_identity
* change the method of calculating descale block index
* unify naming style: use block_scale_ as name prefix
* unify naming style
* update the CHANGELOG.md
* Add FP8 block scale quantization support for FMHA forward kernel
---------
Co-authored-by: Illia Silin <98187287+illsilin@users.noreply.github.com>
Co-authored-by: Po Yen Chen <PoYen.Chen@amd.com>
* add page_block_size parameter
* add is_sglang_layout to parameters
* add kv_offset_array_transform to batch async for page size 16
* add kv_last_page_lens to kernel
* change kv layout to [num_total_pages, page_block_size, hdim]
* format
* - enable codegen of batch_prefill kernels
- create new problem struct BlockFmhaBatchPrefillPipelineProblem for
batch prefill kernels
- generate different page sizes of batch prefill kernels (1, 16)
* 1. fix wrong calculation of page id in kv_offset_array_transform in gfx950
2. support page size 1024
* fix python format
* change kv cache layout to [num_blocks, num_kv_heads, head_size/x,
block_size, x] and [num_blocks, num_kv_heads, block_size/X, head_size, X]
* 1. Introduced `kVectorSize` in BlockFmhaBatchPrefillPipelineProblem instead of using hardcode values
2. Makes batch prefill kernel traits structures inherent from fmha fwd
traits
3. Add some static check for Page size, vector size, hdim, ..., etc.
* [Refactor] Replace is_sglang_layout with Enums for KV cache configuration
Refactored `fmha_batch_prefill` to use `BlockAttentionKVCacheMemoryLayoutEnum` (VECTORIZED/LINEAR) and `BlockAttentionKVCacheLookupTableEnum` (SGLANG_1D/VLLM_2D) instead of a single
boolean.
**Changes:**
* Added Enum definitions in `block_attention_kvcache_layout_enum.hpp`.
* Updated Kernel, Pipeline, and Traits to template on these Enums.
* Implemented `kv_offset_array_transform` logic based on `kKVMemoryLayout`.
* Refactored `PageBlockTableKargs` to adapt to `kKVLookupTable`.
* Updated CodeGen scripts to support new parameters.
This decouples memory layout from the paging mechanism, enabling flexible KV cache configurations.
* 1. remove batch prefill pipeline with sk_pad=false
2. correct some comments
3. add static assert to make sure v offsets is in same page within a tile.
* fix vgpr spill count
* remove unnecessary t2s functions
* add fp8 support for receipt 200 and 600 in fmha_bath_prefill.py
* support linear kv cache layout
* Remove block_table_ptr from fwd_batch_prefill_args. Instead, reuse
kv_page_indices as a pointer of the lookup table.
* 1. merge multiple transforms into single transform.
2. add static check to make sure vlayout is row-major.
* move FmhaFwdCommonKargs::seqlen_k_ptr to VllmPageTableKargs.
* update changelog
---------
Co-authored-by: ltqin <letaoqin@amd.com>
Co-authored-by: PoYen, Chen <PoYen.Chen@amd.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.
* [CK_TILE] fmha: Add query padding support to backward pass
Introduces support for query sequence padding (q_padding) in the FMHA backward pass kernels.
- Passing `seqlen_q_ptr` to the backward kernels to distinguish logical from physical sequence lengths.
- Updating `OGradDotO`, `ConvertQGrad`, and `DQDKDV` kernels to respect logical lengths and handle zero-length sequences.
- Aligning LSE indexing in the forward kernel with the padded layout for consistency.
- Adding a new GTest suite (`test_fmha_bwd_kernel_padding.cpp`) with comprehensive tests for various padding scenarios, including zero-length
sequences and deterministic mode.
* fix clang format
* Adapt fmha_bwd_runner.cpp to new q, kv sequence padding
Add backward q/kv sequence padding unit tests.
* [CK_TILE] fmha: Unify sequence length and padding handling
Refactor the handling of sequence lengths and padding in the
FMHA forward and backward kernels to provide a more unified and flexible
interface.
- Replaced `seqstart_padded_*_ptr` with a more robust system that uses
`seqstart_*_ptr` for physical sequence lengths and introduces
`seqlen_*_ptr` and `cu_seqlen_*_ptr` for logical (unpadded) lengths.
- Established a clear order of precedence for determining sequence
length: cumulative lengths (`cu_seqlen_*_ptr`) take priority,
followed by per-sequence lengths (`seqlen_*_ptr`), and finally
physical lengths derived from `seqstart_*_ptr`.
- Clarified the distinction between "group mode" and "batch mode" and
how sequence lengths are handled in each case.
- Renamed `cu_seqlen_kv_ptr` to `cu_seqlen_k_ptr` for consistency.
- Updated comments and documentation to reflect the new argument
structure and usage.
---------
Co-authored-by: illsilin_amdeng <Illia.Silin@amd.com>
* 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
* * [CK_TILE] Add sequence padding and variable length support in fmha (and v3)
- Group Mode Padding: Introduces the `-s_qpad` argument to support
physically padded layouts. Kernels now use padded start pointers
(`seqstart_padded_*_ptr`) for memory addressing.
- Batch Mode Variable Length: Adds `-q_eff_lens` and `-kv_eff_lens`
arguments for efficient processing of variable-length sequences by
passing cumulative effective lengths (`cu_seqlen_*_ptr`) to the kernel.
- FMHA examples: Support padding and variable length both in
group and batch mode. Dispatcher is updated as well (dispatch to
kPadSeqLenK enabled pipeline).
- New padding test cases: Add padding test cases to `smoke_test_fwd.sh` and
`test_fmha_fwd.inc`, and add benchmarks to `benchmark_fwd.sh` and
`benchmark_fwd_v3.sh` as well. These test cases and benchmarks that
specifically validate/benchmark the new padding and variable-length
functionalities in both group and batch modes.
* [CK_TILE] Fix build error in fmha unit tests
* [CK_TILE] add mqa, gqa to sequence padding unit tests
* [CI_TILE] Reduce the number of padding seqlen unit tests in FMHA to avoid timeouts in CI
* [CK_TILE] remove unnecessary MageKArgs overload in FmhaFwdV3Kernel and FmhaFwdKernel
* [CK_TILE] Add sequence padding and variable length support in fmha (and v3)
- Group Mode Padding: Introduces the `-s_qpad` argument to support
physically padded layouts. Kernels now use padded start pointers
(`seqstart_padded_*_ptr`) for memory addressing.
- Batch Mode Variable Length: Adds `-q_eff_lens` and `-kv_eff_lens`
arguments for efficient processing of variable-length sequences by
passing cumulative effective lengths (`cu_seqlen_*_ptr`) to the kernel.
- FMHA examples: Support padding and variable length both in
group and batch mode. Dispatcher is updated as well (dispatch to
kPadSeqLenK enabled pipeline).
- New padding test cases: Add padding test cases to `smoke_test_fwd.sh`,
and add benchmarks to `benchmark_fwd.sh` and `benchmark_fwd_v3.sh` as well.
These test cases and benchmarks that specifically validate/benchmark the
new padding and variable-length functionalities in both group and batch modes.
* [CK_TILE] Fix build error in fmha unit tests
---------
Co-authored-by: Po Yen Chen <PoYen.Chen@amd.com>
Co-authored-by: Yi DING <yi.ding@amd.com>
* change host using fp16 to check
* fp8 to fp8 compare
* rewrite input parameters
* add not squant
* remove some output code
* for scale = 1
* format
* saturates only for fp8
* add fp8bf16 data type
* add fp8bf16 data type
* fix test fp8 code
* add run_fp8bf16_tests
* change fmha fwd example parameter(adding fp8bf16)
* Support fp8bf16 for Aiter
* Support aiter fp8bf16 in c++
* fix comment about fp8 in readme.md
* add fp8fp32
* add fp8fp32 test
* remove range_q etc.
* format
* fix test parameters about squant and fmha example input fp8bf16 fp8fp32 data type
* add fp8bf16 to data_type function
* change colmajor to rowmajor in test_ck_tile_fmha_fwd_fp8
* format
* reset atol for fp8
* fix bug for atol
---------
Co-authored-by: rocking <ChunYu.Lai@amd.com>
Co-authored-by: asleepzzz <hanwen.chang@amd.com>
* add prefetching physical block id for pagedkv
* start add pagedkv prefill
* rename pipeline
* add kernel for pagedkv
* add an init version pagedkv prefill
* fix redefine issue
* add struct BlockFmhaFwdPagedKVPipelineProblem and fmha_fwd_pagedkv_args
* generate dispatch code
* add body generating code
* comipling pass
* remove dropout from pagedkv
* set lse to false in generating code
* start changing qr kernel to pagedkv
* init version of kernerl with pagedkv
* change names of file that are generated
* chang host validation for pagedkv prefill
* using iglp to change blockgemm
* add kernel files to op head file
* show parameters
* rewrite print parameter fun
* add fwd
* remove default parameter of GridSize
* format
* fix nhead issue and add seqlen_k_ptr to batch mode
* format code
* remove no-longer used code
* format
* fix some comments
---------
Co-authored-by: ltqin <letaoqin@amd.com>
Co-authored-by: Po Yen Chen <PoYen.Chen@amd.com>
* Update license year
* Add initial code to override decode problem
* Fix splitkv traits/args overriding error
* Reshape and transpose lse for decode
* Remove debug code
* Prettify example code
* Use better function name
* Add kMergeNumHeadGroupsSeqLenQ flag
Kernel user can use this switch to turn on/off optimization for
some problem sizes
* Add missing flag declarations
* Default turn off kMergeNumHeadGroupsSeqLenQ in codegen
* Group similar statements together
* Remove assumption of seqlen_q=1
* Remove kMergeNumHeadGroupsSeqLenQ from splitkv combine kernel
* Support kMergeNumHeadGroupsSeqLenQ=true in fmha splitkv kernel
* Run kMergeNumHeadGroupsSeqLenQ=true kernels when need
* Fix group mode block skip logics
* Undo changes of normal fwd kernel
* Update in GridSize() and using GridSize() for splitkv kernel (#1799)
---------
Co-authored-by: Qianfeng <qianfeng.zhang@amd.com>
* Remove using tile partitioner for fmha_fwd_kernel
* Remove using tile partitioner for fmha_fwd_splitkv and splitkv-combine kernels
* Remove using tile partitioner for fmha_fwd_appendkv kernel
* Unify the format of GetTileIndex
* Add check for zero values
* Add static assertions
* Remove invalid option '-e' in smoke_test.sh
* Use correct path of smoke_test.sh
* Avoid zero-sized shared memory array
* Add warning comment
* Replace expr by integer_divide_ceil() call
* Use more readable constant names
* Write down assumption as static assertion
* Add more diagnostic error messages
* Fix wrong BlockWarps when using default pipeline policy
* Add more static assertions for A LDS desc
* Allow using vector size < 8 for data type fp16/bf16
* Align vector size between DRAM dist & LDS desc
* Remove no-longer used func decl
* Fix wrong displayed piepline name
* Undo policy template changes for tile_example_gemm_basic
* Add missing space and make error message stands out
* Unify print precision
* Add missing include directive <iomanip>
* Replace constant 64 by get_warp_size() call
* Replace constant 128 by named variable: BankLength
* Add kAMBlock/kBNBlock attributes
* Allow usig different A/B warp dist for multiple blocks
* Add helper function to get warp dist encodings
* Add 4x64x4 fp16 warp gemm attribute impl
* Complete the A/B warp dist encoding logic
* Fix wrong thread mapping for C matrix
* Use smaller vector size for small tile
* Add static assert to block unsupported warp gemm impl
* Extract common code out as helper method
* Add 4x64x16 fp16 warp gemm type alias
* Add comment to warning developers
* Undo WarpGemmAtrributeMfma<> changes
* Use more clear static assertion error message
* Add trivial wrapper to get warp dstr encodings
* Only transpose warp gemm result if it's square
* Fix compilation error
* Support multi-block warp gemm (on N direction)
* Remove duplicated code
* Fix output encoding of warp gemm
* Fix wrong shape of WarpGemmAtrributeMfmaIterateK<>
* Remove unused code
* Fix wrong shape of WarpGemmAttributeMfmaImplF16F16F32M4N64K4
* Add type config for bf16_t
* Add 4x64x16 bf16 warp gemm
* Update WarpGemmAtrributeMfmaIterateKAndTransposedCDistribution
* Add 64x4x4 fp16/bf16 warp gemm impl
* Add 64x4x16 fp16/bf16 warp gemm
* Add static assertion for better error diagnostic
* Get Q dram dstr directly form block gemm
* Add missing header: fused_moe.hpp
* Allow specifying different warp-gemm for gemm0 & gemm1
* Store P matrix into LDS before gemm1
* Fix inconsistant kernel name
* Remove constraint on gemm0 & gemm1 block warps
* Remove unsupported vector size from checking list
* Allow using 4x64x16 warp gemm for gemm0
* Finish policy customization
* Finish pipeline modification
F#
* Use block warps in codegen
* Fix wrong rank of m_lds_window origin
* Use better distributed tensor
* Make P-store earlier
* Remove duplicated experssions
* Remove unnecessary tile window
* Create new files for new splitkv pipeline
* Separate old/new pipeline codegen logic
* Sync changes form develop
* Undo gemm kernel/pipeline changes
* Undo gemm example changes
* Remove blank lines
* Fix typo
* Use new warp gemm interface
* Fix link error
* Fix wrong pipeline tag
* Fix more link error
* Avoid unnecessary padding
* Always use vector load for K
* Padding on fastest dimension when necessary
* Force padding Q on hdim_q
* Set high dimension padding flag to false
* Re-format headers
* Use warps=<1, 4, 1> for both gemm0 & gemm1
* Fix complilation errors
* Remove m/l shuffle logics
* Ignore duplicate data when write lse_acc
* Use gemm0 block warps as lds tile width
* Remove hard-coded numbers
* Fix wrong distribution width
* Remove unnecessary code
* Add s_barrier before writing to LDS
* Store Q into LDS before gemm0
* Fix wrong Q tile size
* Use simple Q lds descriptor for debuging
* Use more realistic Q lds descriptor
* Add comment & use better variable name
* Make Q lds space not overlapped with others
* Remove unnecessary block_tile_reduce_sync() call
* Move Q load statements
* Move block_sync_lds() right before use
* Re-order instructions
* Remove necessary lambda expression
* Use 8 threads on kMaxSplits direction while doing reduction
* Tiny correction for using 8 threads on kMaxSplits direction for combine kernel
* Padding num_split direction of o_acc tile window to 4x
* Update splitkv combine pipeline design
* Add kN1 back to splitkv combine pipeline problem
* Fix compilation errors
* Add missing template parameter
* Fix wrong splitkv combine kernel name
* Fix wrong origin
* Fix wrong LDS descriptor shape
* Fix sync & reduction logics
* Remove unnecessary static assertions
* Extract tile size computation logics
* Make sure we can reuse padding flags in combine kernels
* Rename variables
* Use OaccDataType in BlockFmhaSplitKVCombinePipelineTileSizes<>
* Remove unnecessary static assertion
* Fix function name typo
* Add constraint on kN1 template parameter
* Hide K tile loading latency in earlier iteration
* Fix wrong splitkv kernel name
* Use s_shuffling to replace p_shuffling which removes the needs of cross-warp reduction
* Rename pipeline
* Fix wrong pipeline name attribute
* Add GetAlignmentQ() for NWarpSShuffle pipeline
* Separate Q tile into dram tile & register tile concepts
* Remove non-squre warp gemm transpose c type alias
* Fallback tile size changes for fmha fwd splitkv
* Remove redundant change
* Refine naming for the S tile
* Use better naming of the S tile dstr (read from lds)
* Share Q lds with K lds
* Tiny change
* Fix with using static_for for passing CI checking
---------
Co-authored-by: Qianfeng Zhang <Qianfeng.Zhang@amd.com>
* Generate group mode paged-attn kernel
* Enable paged-kvcache + group mode support
* Add missing header: fused_moe.hpp
* Add comment to explain kernel arg usage
* Make error message more clear
* Add comment for confusing data member names
* Add more comment for confusing variable names
* Fix typo in option description
* Adding seed and offset pointer support to the philox random number generator.
* Separating seed and offset pointer checks with different condition statements.
* Changes include, adding support for device seed and offset pointers, union is used to store seed/offset values and device pointers to minimize device SGPRs.
* Correcting a typo in the readme file
* Re-format files using remod.py
* Use STL type for API parameters
* Use simpler struct design for drop_seed & drop_offset
* Undo unnecessary changes
* Sync kargs style for fmha_fwd.hpp/.cpp
* Use templated union to reduce code
* Use structured binding to make code more readable
---------
Co-authored-by: Sudhir Kylasa <sukylasa@amd.com>
Co-authored-by: Po Yen Chen <PoYen.Chen@amd.com>
* Use same layout for o_acc and o tensor
* Use better param names in partitioner
* Remove redundant kargs 'max_seqlen_q'
* Use better param names in splitkv kernel
* Add comment for additional kernel arguments
* Sync empty loop early return logics between pipelines
* Pass more arguments to cmake in scripts
* Align backslashes
* Fix wrong o_acc tensor view strides
* Change o_acc layout if o_perm=0
* Handle whole row masked via attn_bias
* Use use vector width = 1 for o_acc
* Use more even split sizes
* Use dictionary to config all the functions
* Add init codegen logic for fmha fwd appendkv
* Call HIP_CHECK_ERROR() macro to get real source info
* Setup meaningfull arguments
* Sync kernel name with the codegen
* Add knew/vnew tensors to the kernel argument
* Fix wrong K values after appending
* Fix vnew append errro
* Extract common logics
* Fix Vnew tile dstr for row major case
* Conditionally add fwd_splitkv API in fmha_fwd example
* Conditionally add call to fmha_fwd_splitkv()
* Remove "EXAMPLE_" prefix of cmake variables
* Regsiter API handlers automatically
* Early return if 0 < s_k_new is not supported
* Show message if we are ignoring option
* Unify CMakeLists.txt coding style
* Set num_splits=1 if split-kv is not supported
* Add length/stride getters for HostTensor
* Add RoPE example utilities
* Add reference_rotary_position_embedding() (not implemented)
* Finish reference_rotary_position_embedding() impl
* Fix typo of HostTensor<>::get_length()
* Fix compilation errors
* Fix wrong answer when interleaved=false
* Fix wrong answer when interleaved=true
* Append K/V in the host verification code
* Simplify K appending logics
* Simplify v_host_ref definition
* Reduce input/output dimensions
* Rename function: add "batched" prefix
* Apply RoPE on host side
* Rename RoPE utility function
* Fix wrong tensor size
* Avoid invoking deprecated method 'find_module'
* Pass RoPE kernel args
* Create Rotary Cos/Sin tile windows in kernel
* Add compute data type alias for RoPE
* Randomly generate seqlen_knew if needed
* Fix seqlen_knew enabling check logic
* Add minimum seqlen_k to generate compliance kvcache
* Fix compilation error in debug mode
* Fix wrong boundaries
* Fix wrong seqlen_k for kvcache
* Rename variables used in distributio encoding
* Fix rotary cos/sin tensor/tile size
* Add constraint to the rotary_dim option
* Remove unused inner namespace
* Add dram distribution for rotary_cos/rotary_sin (interleaved)
* Only apply interleaved RoPE on Knew for now
* Fix wrong thread starting offset
* Instantiate multiple kernels for RoPE approaches
* Clean-up pipeline
* Fix error in RoPE host reference
* Handle RoPE half-rotated logics
* Support 8x rotary_dim under half-rotated RoPE
* Add comment
* Apply elementwise function to the loaded tiles
* Unify parameter/variable naming style
* Remove constness from q_ptr
* Add code blocks for q_tile
* Apply RoPE to q_tile
* Remove debug print code in kernel
* Fix wrong knew/vnew appending positions
* Use better naming for tile indices
* Add make_tile_window() for adding distribution only
* Skip code if # of block is more than needed
* Move thread locating logics into policy
* Remove always true static_assert()
* Rename header
* Rename RotaryEmbeddingEnum
* Extract rotary embedding logic out
* Re-order parameters
* Align naming of some tile size constants
* Rename more tile size constants
* Fix wrong grid size
* Fix wrong shape of knew_host/vnew_host
* Fix wrong index into knew_host/vnew_host
* Fix wrong rotary_cos/rotary_sin memory size for Q
* Extract Q/Knew vector size to helper methods
* Use different rotary_cos/rotary_sin distr for Q/Knew
* Update host/device specifiers
* Fix wrong data type for Q rotary_cos/rotary_sin
* Remove RoPEComputeDataType type alias
* Shift rotary_cos/rotary_sin by cache_seqlen_k
* Add comment for why I just 't' for all padding flags
* Align commit message to the real comment
* Fix wrong pipeline
* Rename utility function
* Disable host verification if API not exist
* Fix wrong rope key for fp8 pipeline
* Allow only apply RoPE on Q (without append KV)
* Add append-kv smoke tests
* Remove debug statements
* Remove more debug statements
* Re-arrange the 'set +x' command
* Remove no-longer used method in pipeline
* Add missing init code
* Refine pipeline padding settings
* Enlarge rotary_dim limit (8 -> 16)
* Enlarge KPerThread for rotary_interleaved=false
* Update rotary_dim range in smoke_test_fwd.sh
* Add template argument 'kIsPagedKV' for splitkv kernels
* Launch splitkv kernel if given page_block_size
* Fix wrong kernel name
* Fix seqlen_k_min for pre-fill case (1 -> 0)
* Add copy_const<> type trait
* Add another make_tile_window()
* Introduce 'TileWindowNavigator' types
* Simplify TileWindowNavigator interfaces
* Fix tile window navigation bugs
* Disable calling fmha_fwd()
* Remove ununnecessary data members
* Simplify more make_tile_window() overloads
* Move V tile through TileWindowNavigator
* Fix uneven split checking logic
* Move code after decide seqlen_q/seqlen_k
* Make sure we always start reading complete tile
* Use 128 as minimus page_block_size
* Fix wrong origin for bias
* Add batch_stride_k/batch_stride_v in group mode
* Unify origin
* Add missing kernel arguments for group mode
* Add paged-kv codegen logic for appendkv kernels
* Add block_table kernel args for appendkv kernel
* Add tile navigators to the appendkv kernel
* Fix wrong tensor descriptor lengths
* Pass re-created tile window to pipeline
* Fix wrong strides for appendkv kernel
* Allow transit tile_window to another page-block
* Handle cross-page-block write
* Donot perform write again if already in last page-block
* Always add fmha_fwd() api
* Add missing group mode argument
* Remove debug macro usages
* Rename option s_k_new to s_knew
* Separate splitkv/non-splitkv args/traits
* Remove fmha_fwd_dispatch()
* Fix compilation errors
* Remove dropout code in splitkv kernel
* Allow problem types without define kHasDropout attr
* Use generic lambda to init traits objects
* Separate more non-splitkv & splitkv traits/args
* Display more info for specific kernels
* Show more detailed warning message
* Rename 'max_num_blocks' to 'max_num_page_blocks'
* Remove no-longer used pipeline files
* Wrap code by #if directives
* Move functors to the begining of validation code
* Use generic lambda to init all the api traits/args
* Fix wrong seqlen for kvcache
* Add missing comment
* Rename TileWindowNavigator to PageBlockNavigator
* Only expose necessary methods (not attributes)
* Re-order pipeline paremeters
* Refine smoke_test_fwd.sh
* Fix wrong arugment count
* Make tile window directly via PageBlockNavigator
* Remove unused template paremeter
* Remove group mode from appendkv kernel
* Fix skcheck logic
* Fix wrong syntax in skcheck expr
* Use meaningful options in smoke test
* Remove options
* Fix formatting
* Fix more format
* Re-organize bash functions
* Pass cache_batch_idx to kernels
* Support cache_batch_idx in example
* Fix compilation error
* Add more appendkv test
* Add more case for appendkv
* Fix unexisted attribute
* Remove 0 < seqlen_knew constraint
* Clarify the case in warning message
* Remove macro checking
* Force batch mode when invoking appendkv & splitkv apis
* Fix mode overriding logics
* Fix wrong parameter name
* Randomize seqlen_k if use kvcache
* Use randomized seqlen_k for kvcache
* Avoid using too small rotary_cos & rotary_sin
* Rename parameter
* Add seqlen_q & seqlen_k rules
* Add comment
* Add more comments
* Fix compilation errors
* Fix typo in comment
* Remove type argument
* Avoid seqlen_k=0 for kvcache
* Revert "Avoid seqlen_k=0 for kvcache"
This reverts commit 21c4df89e4.
* Fix wrong uneven split checking logics
* Only randomize kvcache seqlen_k if 1 < batch
* Return earlier if split is empty
* Revert "Only randomize kvcache seqlen_k if 1 < batch"
This reverts commit b9a4ab0d7e.
* Re-order seqlen_k_start adjustment logics
* Fix compilation errors
* Re-format script
* Find executable from folder automatically
* Fix kvcache seqlen_k generating logic
* Make comment more clear
* Fix wrong knew/vew appending logic on host
* Add s_barrier to sync threads
* Revert "Add s_barrier to sync threads"
This reverts commit d3f550f30c.
* Support only using 1 row of rotary_cos/rotary_sin
* Rotate Q in different way
* Unify tensor view creation logics
* Fix wrong argument
* Add mask to switch how we use the rotary_cos/sin
* Move attr from traits to problem
* Move has_mask to fmha_fwd_appendkv_args
* Support use uint32_t as SAD operand in Alibi<>
* Use sad_u32() in splitkv kernels
* Store tensor views in PageBlockNavigator
* Use stored tensor view to update tile windows
* Enlarge tensor view size
* Remove debug code
* Fix wrong tensor view size
* Wrap tensor view into PageBlockNavigator
* Add DataType member to PageBlockNavigator
* Remove unnecessary member functions
* Refind macro use
* Fix typo
* Add blank line between directives and actual code
* Re-format files
* Remove type in comment
---------
Co-authored-by: carlushuang <carlus.huang@amd.com>
Co-authored-by: rocking <ChunYu.Lai@amd.com>
* add alibi support
* fix code
* update code based on comment
* Support more hdim
* fix fp8 bias
* support seqlen_k=0 case
* remove unused printf
* fix format
---------
Co-authored-by: rocking <ChunYu.Lai@amd.com>
* enable gfx940
* switch between intrinsic mfma routines on mi100/200 and mi300
* fix mfma_int8 on MI300
* disable 2 int8 examples on MI300
* Update cmake-ck-dev.sh
* restore gitignore file
* modify Jenkinsfile to the internal repo
* Bump rocm-docs-core from 0.24.0 to 0.29.0 in /docs/sphinx
Bumps [rocm-docs-core](https://github.com/RadeonOpenCompute/rocm-docs-core) from 0.24.0 to 0.29.0.
- [Release notes](https://github.com/RadeonOpenCompute/rocm-docs-core/releases)
- [Changelog](https://github.com/RadeonOpenCompute/rocm-docs-core/blob/develop/CHANGELOG.md)
- [Commits](https://github.com/RadeonOpenCompute/rocm-docs-core/compare/v0.24.0...v0.29.0)
---
updated-dependencies:
- dependency-name: rocm-docs-core
dependency-type: direct:production
update-type: version-update:semver-minor
...
Signed-off-by: dependabot[bot] <support@github.com>
* initial enablement of gfx950
* fix clang format
* disable examples 31 and 41 int8 on gfx950
* add code
* fix build wip
* fix xx
* now can build
* naming
* minor fix
* wip fix
* fix macro for exp2; fix warpgemm a/b in transposedC
* unify as tuple_array
* Update the required Python version to 3.9
* Update executable name in test scripts
* re-structure tuple/array to avoid spill
* Merge function templates
* Fix format
* Add constraint to array<> ctor
* Re-use function
* Some minor changes
* remove wrong code in store_raw()
* fix compile issue in transpose
* Rename enum
Rename 'cood_transform_enum' to 'coord_transform_enum'
* let more integral_constant->constant, and formating
* make sure thread_buffer can be tuple/array
* temp fix buffer_store spill
* not using custom data type by default, now we can have ISA-level same code as opt_padding
* fix compile error, fp8 not ready now
* fix fp8 duplicated move/shift/and/or problem
* Default use CK_TILE_FLOAT_TO_FP8_STOCHASTIC rounding mode
* fix scratch in fp8 kernel
* update some readme
* fix merge from upstream
* sync with upstream
* sync upstream again
* sync 22
* remove unused
* fix clang-format
* update README of ck_tile example
* fix several issue
* let python version to be 3.8 as minimal
* remove ck_tile example from default cmake target like all/install/check
* remove mistake
* 1).support receipe in generate.py 2).use simplified mask type 3).change left/right to pass into karg
* fix some bug in group-mode masking and codegen. update README
* F8 quantization for FMHA forward (#1224)
* Add SAccElementFunction, PComputeElementFunction, OAccElementFunction in pipeline
* Add element function to fmha api
* Adjust P elementwise function
* Fix bug of elementwise op, our elementwise op is not inout
* Add some elementwise op, prepare to quantization
* Let generate.py can generate different elementwise function
* To prevent compiler issue, remove the elementwise function we have not used.
* Remove f8 pipeline, we should share the same pipeline even in f8
* Remove remove_cvref_t
* Avoid warning
* Fix wrong fp8 QK/KV block gemm setting
* Check fp8 rounding error in check_err()
* Set fp8 rounding error for check_err()
* Use CK_TILE_FLOAT_TO_FP8_STANDARD as default fp8 rounding mode
* 1. codgen the f8 api and kernel
2. f8 host code
* prevent warning in filter mode
* Remove not-in-use elementwise function kargs
* Remove more not-in-use elementwise function kargs
* Small refinements in C++ source files
* Use conditional_t<> to simplify code
* Support heterogeneous argument for binary function types
* Re-use already-existing scales<> functor template
* Fix wrong value produced by saturating
* Generalize the composes<> template
* Unify saturates<> implementation
* Fix type errors in composes<>
* Extend less_equal<>
* Reuse the existing template less_equal<> in check_err()
* Add equal<float> & equal<double>
* Rename check_err() parameter
* Rename check_err() parameter
* Add FIXME comment for adding new macro in future
* Remove unnecessary cast to void
* Eliminate duplicated code
* Avoid dividing api pool into more than 2 groups
* Use more clear variable names
* Use affirmative condition in if stmt
* Remove blank lines
* Donot perfect forwarding in composes<>
* To fix compile error, revert generate.py back to 4439cc107d
* Fix bug of p element function
* Add compute element op to host softmax
* Remove element function in api interface
* Extract user parameter
* Rename pscale and oscale variable
* rename f8 to fp8
* rename more f8 to fp8
* Add pipeline::operator() without element_functor
* 1. Remove deprecated pipeline enum
2. Refine host code parameter
* Use quantization range as input
* 1. Rename max_dtype to dtype_max.
2. Rename scale to scale_s
3.Add init description
* Refine description
* prevent early return
* unify _squant kernel name in cpp, update README
* Adjust the default range.
* Refine error message and bias range
* Add fp8 benchmark and smoke test
* fix fp8 swizzle_factor=4 case
---------
Co-authored-by: Po Yen Chen <PoYen.Chen@amd.com>
Co-authored-by: carlushuang <carlus.huang@amd.com>
---------
Signed-off-by: dependabot[bot] <support@github.com>
Co-authored-by: illsilin <Illia.Silin@amd.com>
Co-authored-by: Illia Silin <98187287+illsilin@users.noreply.github.com>
Co-authored-by: Jing Zhang <jizha@amd.com>
Co-authored-by: zjing14 <zhangjing14@gmail.com>
Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com>
Co-authored-by: Po-Yen, Chen <PoYen.Chen@amd.com>
Co-authored-by: rocking <ChunYu.Lai@amd.com>
