Files
composable_kernel/example/ck_tile/01_fmha/generate.py
Po Yen Chen 0cb2e06ddc [CK_TILE] fmha forward split-kv + combine kernels (#1338)
* FA fwd dropout

* FA bwd

* epilogue reuse

* CMakeLists update

* [CK_TILE] support alibi (#1269)

* 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>

* now fwd/bwd can build

* bwd alibi

* add bwd validation stream_config

* update generated filenames

* update bwd kernel launch

* CK_TILE_HOST_DEVICE in philox

* Transpose -> transpose

* format

* format

* format

* Generate the instance for FA required

* format

* fix error in WarpGemm

* Add num_splits option and dummy split-kv api method

* Generate fmha_fwd_splitkv()

* Add SplitKV kernel codegen logics

* Add SplitKV combine kernel codegen logics

* Fix mismatched return type

* Clean-up code

* Replace sentinel value before storing

* Fix wrong layout of LSE/LSEacc/Oacc

* Format codes

* Fix o_acc memory error

* Fix wrong kBlockSize used in policy

* Reduce # of combine kernels

* Fix split-kv combine kernel name

* Fix wrong LDS indexing logics

* Fix wrong loop counter step logic

* Undo vector size changes

* Remove no-longer used field

* Remove in-consistent comment

* Remove debug statements in example

* Remove more debug statements

* Add constness to local variables

* Clearn up generate.py

* Fix unstable clang-format comment

* Remove unused include directive

* Use shorter template parameter name

* Enable non-split-kv blobs

* Update license date

* Print num_splits conditionally

* Undo disabling data types

* Remove unnessary tile size for fp8

* Fix wrong pipeline args for fp8

* Fix example output format

* Remove more debug code in combine pipeline

* Add stride kernel arguments for LSE/O acc workspace

* Re-order split-kv pipeline call operator arguments

* Pass LSE/O strides in kernel argument

* Re-order pipeline call operator arguments

* Use tensor_descriptor to locate LSEacc elements

* Support providing invalid element for tensor view

* Set invalid element value for LSEacc tensor view

* Remove hand-written store_tile() code

* Remove necessary value-overwrite logic

* Add transposed lds descriptor

* Support load_tile() for tile_window_with_static_lengths<>

* Undo removing necessary value-overwrite logic

* Use read descriptor to locate lds elements

* Simplify pipeline source code

* Add constraint to kMaxSplits

* Default use kMaxSplits=64 in generate.py

* Revert "Add constraint to kMaxSplits"

This reverts commit 0a2132d758.

* Revert "Default use kMaxSplits=64 in generate.py"

This reverts commit c7d9c80b77.

* Decide alignment by the padding parameter

* Remove no-longer used utility functions

* Remove not-working code

* Add comment & remove no-longer used code

* Fix computation errors

* Add heuristic to override num_splits option

* Add constraint to kMaxSplits

* Fix compilation error

* Clean up pipeline code

* Wrap pointer access as lambda function

* Rename confusing methods

* Use kLogMasSplits as template parameter

* Finish splitkv combine kernel codegen

* Update kMaxSplits limit

* Use smaller kM0 for splitkv combine kernel

* Ignore droupout flag in splitkv pipeline

* Unify flag usage

* Add back flag kStoreLSE

* Merge lambda calls in pipeline

* Fix compilation errors

* Avoid all empty splits

* Always check for empty loop in splitkv pipelines

* Re-order parameters

* Remove redundant p_drop option check

* Add traits/problem for fwd splitkv kernel

* Conditionally enable uneven split boundary checks

* Add comment for the splitkv traits field

* Change even split criteria

* Re-order statements

* Refine occupancy value for hdim=128&256

* Refine occupancy value for hdim=32&64

* Remove redundant kernel argument

* Separate fmha bwd codegen logics

* Separate fmha fwd codegen logics

* Remove redundant direction parameter in fwd&bwd codegen logics

* Support generate multiple APIs for an example

* Let 'api' an alias of 'direction' option

* Remove choices for the 'direction' option

* Use dictionary to config all the functions

* Move fmha splitkv codegen logics to other file

* Add fwd_splitkv api for tile_example_fmha_fwd

---------

Co-authored-by: danyao12 <danyao12>
Co-authored-by: carlushuang <carlus.huang@amd.com>
Co-authored-by: rocking <ChunYu.Lai@amd.com>
Co-authored-by: Jing Zhang <jizhan@amd.com>
2024-06-26 17:41:15 +08:00

106 lines
3.2 KiB
Python

# SPDX-License-Identifier: MIT
# Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
# generate kernel instances to speed up compilation
import argparse
from enum import IntEnum
from pathlib import Path
from typing import List, Optional
from codegen.cmake_config import *
from codegen.ops import (
fmha_fwd,
fmha_fwd_splitkv,
fmha_bwd
)
class HandlerId(IntEnum):
LIST_BLOBS = 0
WRITE_BLOBS = 1
handlers = {
'fwd' : (fmha_fwd.list_blobs, fmha_fwd.write_blobs),
'fwd_splitkv' : (fmha_fwd_splitkv.list_blobs, fmha_fwd_splitkv.write_blobs),
'bwd' : (fmha_bwd.list_blobs, fmha_bwd.write_blobs),
}
def write_blobs(output_dir: Optional[str], api_list : List[str], kernel_filter : Optional[str], receipt, mask_impl) -> None:
if output_dir is None:
output_dir = Path(__file__).parent
else:
output_dir = Path(output_dir) / GEN_DIR
output_dir.mkdir(parents=True, exist_ok=True)
for api in api_list:
handler = handlers[api][HandlerId.WRITE_BLOBS]
handler(output_dir, kernel_filter, receipt, mask_impl)
# list all the files that will be generated
def list_blobs(output_file : Optional[str], api_list : List[str], kernel_filter : Optional[str], receipt, mask_impl) -> None:
assert output_file is not None
file_path = Path(output_file)
for api in api_list:
handler = handlers[api][HandlerId.LIST_BLOBS]
handler(file_path, kernel_filter, receipt, mask_impl)
if __name__ == "__main__":
parser = argparse.ArgumentParser(
prog="generate",
description="gen API for CK fmha kernel",
)
parser.add_argument(
"-d",
"--direction", # we keep 'direction' option for backward compatibility
"-a",
"--api",
default='fwd',
required=False,
help="supply API(s) to generate (default: fwd). separated by comma."
)
parser.add_argument(
"-o",
"--output_dir",
required=False,
help="write all the blobs into a directory"
)
parser.add_argument(
"-l",
"--list_blobs",
required=False,
help="list all the kernels to a file"
)
# TODO: if using filter, must apply same value to output_dir and list_blobs
parser.add_argument(
"-f",
"--filter",
required=False,
help="filter out kernels that need to generate, using fnmatch module"
)
parser.add_argument(
"-m",
"--mask",
default="simplified",
required=False,
help="mask implementation, simplified/generic"
)
parser.add_argument(
"-r",
"--receipt",
default=0,
required=False,
help="codegen receipt. 0: generate only 8xhdim coverage\n" + \
" 1: generate more instance to cover all hdim\n" + \
" 2: Only generate instance for Flash attention integration"
)
args = parser.parse_args()
api_list = args.direction.split(',')
if args.list_blobs is not None:
list_blobs(args.list_blobs, api_list, args.filter, int(args.receipt), mask_impl=args.mask)
else:
write_blobs(args.output_dir, api_list, args.filter, int(args.receipt), mask_impl=args.mask)