deepseek-ai/FlashMLA
File Churn
File change frequency (churn) shows the distribution of file updates (days with at least one commit).
Overview
File Change Frequency Overall
  • There are 97 files with 17,763 lines of code.
    • 0 files changed more than 100 times (0 lines of code)
    • 0 files changed 51-100 times (0 lines of code)
    • 0 files changed 21-50 times (0 lines of code)
    • 1 file changed 6-20 times (321 lines of code)
    • 96 files changed 1-5 times (17,442 lines of code)
0% | 0% | 0% | 1% | 98%
Legend:
101+
51-100
21-50
6-20
1-5

explore: grouped by folders | grouped by update frequency | data
Contributors Count Frequency Overall
  • There are 97 files with 17,763 lines of code.
    • 0 files changed by more than 25 contributors (0 lines of code)
    • 0 files changed by 11-25 contributors (0 lines of code)
    • 1 file changed by 6-10 contributors (321 lines of code)
    • 2 files changed by 2-5 contributors (242 lines of code)
    • 94 files changed by 1 contributor (17,200 lines of code)
0% | 0% | 1% | 1% | 96%
Legend:
26+
11-25
6-10
2-5
1

explore: grouped by folders | grouped by contributors count | data
File Change Frequency per File Extension
h, cu, hpp, py, cuh, cpp, asc, md, txt, gitignore, svg, gitmodules
File Change Frequency per Extension
The number of recorded file updates
101+
51-100
21-50
6-20
1-5
py0% | 0% | 0% | 69% | 30%
hpp0% | 0% | 0% | 0% | 100%
cuh0% | 0% | 0% | 0% | 100%
h0% | 0% | 0% | 0% | 100%
cpp0% | 0% | 0% | 0% | 100%
cu0% | 0% | 0% | 0% | 100%
File Change Frequency per Logical Decomposition
primary
primary (file change frequency)
The number of recorded file updates
101+
51-100
21-50
6-20
1-5
flash_mla0% | 0% | 0% | 76% | 23%
csrc0% | 0% | 0% | 0% | 100%
scripts0% | 0% | 0% | 0% | 100%
Most Frequently Changed Files (Top 50)

See data for all files...

File# lines# unitscreatedlast modified# changes
(days)		# contributorsline
churn		first
contributor		latest
contributor
321 11 2025-02-24 2026-09-30 10 6 +741 / -252 31004720+beginlner@users.no... shengyuliu@deepseek.com
__init__.py
in flash_mla
 19 - 2025-08-01 2026-09-10 5 2 +18 / -1 uchihatmtkinu@gmail.com shengyuliu@deepseek.com
params.h
in csrc
 129 - 2025-09-24 2026-09-30 5 1 +456 / -198 shengyuliu@deepseek.com shengyuliu@deepseek.com
common.h
in csrc/api
 223 14 2026-01-16 2026-09-30 4 2 +319 / -42 shengyuliu@deepseek.com shengyuliu@deepseek.com
api.cpp
in csrc/api
 14 - 2026-01-16 2026-09-30 3 1 +27 / -11 shengyuliu@deepseek.com shengyuliu@deepseek.com
dense_bwd.cpp
in csrc/api
 13 1 2026-09-10 2026-09-30 2 1 +16 / -1 shengyuliu@deepseek.com shengyuliu@deepseek.com
dense_fwd.cpp
in csrc/api
 13 1 2026-09-10 2026-09-30 2 1 +16 / -1 shengyuliu@deepseek.com shengyuliu@deepseek.com
77 4 2026-09-10 2026-09-30 2 1 +237 / -26 shengyuliu@deepseek.com shengyuliu@deepseek.com
195 6 2026-09-10 2026-09-30 2 1 +321 / -89 shengyuliu@deepseek.com shengyuliu@deepseek.com
396 9 2026-09-10 2026-09-30 2 1 +691 / -226 shengyuliu@deepseek.com shengyuliu@deepseek.com
462 6 2026-09-10 2026-09-30 2 1 +589 / -25 shengyuliu@deepseek.com shengyuliu@deepseek.com
phase1_k512.cu
in csrc/cuda_kernels/sm100/prefill/sparse/fwd/head128/instantiations
 2 - 2026-09-30 2026-09-30 1 1 +2 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
v41fp4_h64.cu
in csrc/cuda_kernels/sm100/decode/sparse/head64/instantiations
 5 - 2026-09-30 2026-09-30 1 1 +8 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
v41_h64_no_split.cu
in csrc/cuda_kernels/sm100/decode/sparse/head64/instantiations
 5 - 2026-09-30 2026-09-30 1 1 +8 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
v41fp4_h64_no_split.cu
in csrc/cuda_kernels/sm100/decode/sparse/head64/instantiations
 5 - 2026-09-30 2026-09-30 1 1 +8 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
v41_h64.cu
in csrc/cuda_kernels/sm100/decode/sparse/head64/instantiations
 5 - 2026-09-30 2026-09-30 1 1 +8 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
v41fp4_h64_decode_nonorm.cu
in csrc/cuda_kernels/sm100/prefill/sparse/fused_norm_rope_attn_rope_cast_fwd/core_attn/instantiations
 5 - 2026-09-30 2026-09-30 1 1 +8 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
v41_h64_prefill_nonorm.cu
in csrc/cuda_kernels/sm100/prefill/sparse/fused_norm_rope_attn_rope_cast_fwd/core_attn/instantiations
 5 - 2026-09-30 2026-09-30 1 1 +8 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
v41_h64_prefill_norm.cu
in csrc/cuda_kernels/sm100/prefill/sparse/fused_norm_rope_attn_rope_cast_fwd/core_attn/instantiations
 5 - 2026-09-30 2026-09-30 1 1 +8 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
v41fp4_h64_decode_norm.cu
in csrc/cuda_kernels/sm100/prefill/sparse/fused_norm_rope_attn_rope_cast_fwd/core_attn/instantiations
 5 - 2026-09-30 2026-09-30 1 1 +8 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
v41_h128_prefill_norm.cu
in csrc/cuda_kernels/sm100/prefill/sparse/fused_norm_rope_attn_rope_cast_fwd/core_attn/instantiations
 5 - 2026-09-30 2026-09-30 1 1 +8 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
v41fp4_h128_decode_norm.cu
in csrc/cuda_kernels/sm100/prefill/sparse/fused_norm_rope_attn_rope_cast_fwd/core_attn/instantiations
 5 - 2026-09-30 2026-09-30 1 1 +8 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
v41_h128_decode_nonorm.cu
in csrc/cuda_kernels/sm100/prefill/sparse/fused_norm_rope_attn_rope_cast_fwd/core_attn/instantiations
 5 - 2026-09-30 2026-09-30 1 1 +8 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
v41_h128_prefill_nonorm.cu
in csrc/cuda_kernels/sm100/prefill/sparse/fused_norm_rope_attn_rope_cast_fwd/core_attn/instantiations
 5 - 2026-09-30 2026-09-30 1 1 +8 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
v41_h64_decode_norm.cu
in csrc/cuda_kernels/sm100/prefill/sparse/fused_norm_rope_attn_rope_cast_fwd/core_attn/instantiations
 5 - 2026-09-30 2026-09-30 1 1 +8 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
v41fp4_h128_decode_nonorm.cu
in csrc/cuda_kernels/sm100/prefill/sparse/fused_norm_rope_attn_rope_cast_fwd/core_attn/instantiations
 5 - 2026-09-30 2026-09-30 1 1 +8 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
v41_h128_decode_norm.cu
in csrc/cuda_kernels/sm100/prefill/sparse/fused_norm_rope_attn_rope_cast_fwd/core_attn/instantiations
 5 - 2026-09-30 2026-09-30 1 1 +8 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
v41_h64_decode_nonorm.cu
in csrc/cuda_kernels/sm100/prefill/sparse/fused_norm_rope_attn_rope_cast_fwd/core_attn/instantiations
 5 - 2026-09-30 2026-09-30 1 1 +8 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
phase1_decode_k512_v41fp4.cu
in csrc/cuda_kernels/sm100/prefill/sparse/fwd_for_small_topk/head128/instantiations
 5 - 2026-09-30 2026-09-30 1 1 +8 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
phase1_k512.cu
in csrc/cuda_kernels/sm100/prefill/sparse/fwd_for_small_topk/head128/instantiations
 5 - 2026-09-30 2026-09-30 1 1 +8 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
phase1_decode_k512_v41_splitkv.cu
in csrc/cuda_kernels/sm100/prefill/sparse/fwd_for_small_topk/head128/instantiations
 5 - 2026-09-30 2026-09-30 1 1 +8 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
phase1_decode_k512_v41fp4_splitkv.cu
in csrc/cuda_kernels/sm100/prefill/sparse/fwd_for_small_topk/head128/instantiations
 5 - 2026-09-30 2026-09-30 1 1 +8 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
phase1_decode_k512_v41.cu
in csrc/cuda_kernels/sm100/prefill/sparse/fwd_for_small_topk/head128/instantiations
 5 - 2026-09-30 2026-09-30 1 1 +8 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
phase1.h
in csrc/cuda_kernels/sm100/prefill/sparse/fwd/head128
 6 - 2026-09-30 2026-09-30 1 1 +10 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
phase1.h
in csrc/cuda_kernels/sm100/prefill/sparse/fwd/head64
 6 - 2026-09-30 2026-09-30 1 1 +10 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
mask.cuh
in csrc/cuda_kernels/sm100/prefill/dense/common
 6 - 2026-09-30 2026-09-30 1 1 +8 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
combine.h
in csrc/cuda_kernels/smxx/decode/combine
 6 - 2026-09-30 2026-09-30 1 1 +10 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
get_decoding_sched_meta.h
in csrc/cuda_kernels/smxx/decode/get_decoding_sched_meta
 6 - 2026-09-30 2026-09-30 1 1 +10 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
phase1.h
in csrc/cuda_kernels/sm100/prefill/sparse/fwd_for_small_topk/head128
 7 - 2026-09-30 2026-09-30 1 1 +11 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
phase1_h64_k512.cu
in csrc/cuda_kernels/sm100/prefill/sparse/fwd/head64/instantiations
 7 - 2026-09-30 2026-09-30 1 1 +10 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
defines.h
in csrc/cuda_kernels
 9 - 2026-09-30 2026-09-30 1 1 +12 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
interface.h
in csrc/cuda_kernels/sm100/prefill/dense
 11 - 2026-09-30 2026-09-30 1 1 +14 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
kernel.h
in csrc/cuda_kernels/sm100/decode/sparse/head64
 12 - 2026-09-30 2026-09-30 1 1 +17 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
kernel.h
in csrc/ascend_kernels/prefill/sparse
 16 - 2026-09-30 2026-09-30 1 1 +22 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
kernel.h
in csrc/cuda_kernels/sm100/prefill/sparse/fused_norm_rope_attn_rope_cast_fwd/permute_q_b_proj
 21 - 2026-09-30 2026-09-30 1 1 +31 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
kernel.h
in csrc/cuda_kernels/sm100/prefill/sparse/fused_norm_rope_attn_rope_cast_fwd/permute_wv_proj
 22 - 2026-09-30 2026-09-30 1 1 +32 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
utils.hpp
in csrc/cuda_kernels/sm100/prefill/dense/common
 26 - 2026-09-30 2026-09-30 1 1 +33 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
fast_div_mod.h
in csrc/ascend_kernels/prefill/sparse
 28 2 2026-09-30 2026-09-30 1 1 +37 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
helper.h
in csrc/cuda_kernels/sm100/prefill/dense/common
 28 - 2026-09-30 2026-09-30 1 1 +72 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
kernel.h
in csrc/cuda_kernels/sm100/prefill/sparse/fused_norm_rope_attn_rope_cast_fwd/core_attn
 34 - 2026-09-30 2026-09-30 1 1 +53 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
Files With Most Churn (Top 50)
Based on the number of lines added and deleted across all commits.

See data for all files...

File# lines# unitscreatedlast modified# changes
(days)		# contributorsline
churn		first
contributor		latest
contributor
sm100_fmha_bwd_kernel_tma_warpspecialized.hpp
in csrc/cuda_kernels/sm100/prefill/dense/kernel
 1416 16 2026-09-30 2026-09-30 1 1 +1835 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
sm100_fmha_bwd_mla_kernel_tma_warpspecialized.hpp
in csrc/cuda_kernels/sm100/prefill/dense/kernel
 1414 16 2026-09-30 2026-09-30 1 1 +1829 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
kernel_body.h
in csrc/ascend_kernels/prefill/sparse
 1440 16 2026-09-30 2026-09-30 1 1 +1701 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
kernel.cuh
in csrc/cuda_kernels/sm100/prefill/sparse/fused_norm_rope_attn_rope_cast_fwd/core_attn
 1578 - 2026-09-30 2026-09-30 1 1 +1685 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
sm100_fmha_mla_fwd_mainloop_tma_warpspecialized.hpp
in csrc/cuda_kernels/sm100/prefill/dense/collective
 814 20 2026-09-30 2026-09-30 1 1 +1221 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
sm100_fmha_fwd_mainloop_tma_warpspecialized.hpp
in csrc/cuda_kernels/sm100/prefill/dense/collective
 805 20 2026-09-30 2026-09-30 1 1 +1214 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
phase1.cuh
in csrc/cuda_kernels/sm100/prefill/sparse/fwd_for_small_topk/head128
 974 - 2026-09-30 2026-09-30 1 1 +1075 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
321 11 2025-02-24 2026-09-30 10 6 +741 / -252 31004720+beginlner@users.no... shengyuliu@deepseek.com
396 9 2026-09-10 2026-09-30 2 1 +691 / -226 shengyuliu@deepseek.com shengyuliu@deepseek.com
kernel.cuh
in csrc/cuda_kernels/sm100/decode/sparse/head64
 667 - 2026-09-30 2026-09-30 1 1 +741 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
phase1.cuh
in csrc/cuda_kernels/sm100/prefill/sparse/fwd/head128
 625 - 2026-09-30 2026-09-30 1 1 +711 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
phase1.cuh
in csrc/cuda_kernels/sm100/prefill/sparse/fwd/head64
 598 - 2026-09-30 2026-09-30 1 1 +681 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
params.h
in csrc
 129 - 2025-09-24 2026-09-30 5 1 +456 / -198 shengyuliu@deepseek.com shengyuliu@deepseek.com
sm100_fmha_fwd_kernel_tma_warpspecialized.hpp
in csrc/cuda_kernels/sm100/prefill/dense/kernel
 491 11 2026-09-30 2026-09-30 1 1 +643 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
462 6 2026-09-10 2026-09-30 2 1 +589 / -25 shengyuliu@deepseek.com shengyuliu@deepseek.com
195 6 2026-09-10 2026-09-30 2 1 +321 / -89 shengyuliu@deepseek.com shengyuliu@deepseek.com
fmha_fusion.hpp
in csrc/cuda_kernels/sm100/prefill/dense/collective
 277 21 2026-09-30 2026-09-30 1 1 +396 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
common.h
in csrc/api
 223 14 2026-01-16 2026-09-30 4 2 +319 / -42 shengyuliu@deepseek.com shengyuliu@deepseek.com
config.h
in csrc/cuda_kernels/sm100/prefill/sparse/fused_norm_rope_attn_rope_cast_fwd/core_attn
 198 4 2026-09-30 2026-09-30 1 1 +358 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
fmha_device_bwd.hpp
in csrc/cuda_kernels/sm100/prefill/dense/device
 237 3 2026-09-30 2026-09-30 1 1 +340 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
fmha_cutlass_fwd_sm100.cuh
in csrc/cuda_kernels/sm100/prefill/dense
 278 - 2026-09-30 2026-09-30 1 1 +335 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
sm100_fmha_mla_load_tma_warpspecialized.hpp
in csrc/cuda_kernels/sm100/prefill/dense/collective
 220 5 2026-09-30 2026-09-30 1 1 +326 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
sm100_fmha_load_tma_warpspecialized.hpp
in csrc/cuda_kernels/sm100/prefill/dense/collective
 205 4 2026-09-30 2026-09-30 1 1 +302 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
fmha.hpp
in csrc/cuda_kernels/sm100/prefill/dense/device
 168 3 2026-09-30 2026-09-30 1 1 +281 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
77 4 2026-09-10 2026-09-30 2 1 +237 / -26 shengyuliu@deepseek.com shengyuliu@deepseek.com
pipeline_mla.hpp
in csrc/cuda_kernels/sm100/prefill/dense/common
 174 12 2026-09-30 2026-09-30 1 1 +250 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
combine.cu
in csrc/cuda_kernels/smxx/decode/combine
 208 - 2026-09-30 2026-09-30 1 1 +240 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
sm100_fmha_fwd_epilogue_tma_warpspecialized.hpp
in csrc/cuda_kernels/sm100/prefill/dense/collective
 148 6 2026-09-30 2026-09-30 1 1 +237 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
helpers.h
in csrc/cuda_kernels/sm100
 196 9 2026-09-30 2026-09-30 1 1 +232 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
common_subroutine.h
in csrc/cuda_kernels/sm100
 168 5 2026-09-30 2026-09-30 1 1 +225 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
gather_tensor.hpp
in csrc/cuda_kernels/sm100/prefill/dense/common
 142 3 2026-09-30 2026-09-30 1 1 +215 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
config.h
in csrc/ascend_kernels/prefill/sparse
 139 1 2026-09-30 2026-09-30 1 1 +206 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
fmha_cutlass_bwd_sm100.cuh
in csrc/cuda_kernels/sm100/prefill/dense
 171 - 2026-09-30 2026-09-30 1 1 +203 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
fmha_causal_tile_scheduler.hpp
in csrc/cuda_kernels/sm100/prefill/dense/kernel
 117 8 2026-09-30 2026-09-30 1 1 +197 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
config.h
in csrc/cuda_kernels/sm100/decode/sparse/head64
 151 - 2026-09-30 2026-09-30 1 1 +191 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
fmha_kernel_bwd_sum_OdO.hpp
in csrc/cuda_kernels/sm100/prefill/dense/kernel
 106 7 2026-09-30 2026-09-30 1 1 +168 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
fmha_tile_scheduler.hpp
in csrc/cuda_kernels/sm100/prefill/dense/kernel
 96 8 2026-09-30 2026-09-30 1 1 +162 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
fmha_kernel_bwd_convert.hpp
in csrc/cuda_kernels/sm100/prefill/dense/kernel
 100 8 2026-09-30 2026-09-30 1 1 +160 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
config.h
in csrc/cuda_kernels/sm100/prefill/sparse/fwd_for_small_topk/head128
 113 - 2026-09-30 2026-09-30 1 1 +154 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
get_decoding_sched_meta.cu
in csrc/cuda_kernels/smxx/decode/get_decoding_sched_meta
 132 - 2026-09-30 2026-09-30 1 1 +148 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
config.h
in csrc/cuda_kernels/sm100/prefill/sparse/fwd/head64
 115 - 2026-09-30 2026-09-30 1 1 +147 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
config.h
in csrc/cuda_kernels/sm100/prefill/sparse/fwd/head128
 107 - 2026-09-30 2026-09-30 1 1 +135 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
fmha_common.hpp
in csrc/cuda_kernels/sm100/prefill/dense/collective
 84 5 2026-09-30 2026-09-30 1 1 +127 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
dequant_utils.cuh
in csrc/cuda_kernels/sm100
 107 - 2026-09-30 2026-09-30 1 1 +116 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
kernel.cu
in csrc/cuda_kernels/sm100/prefill/sparse/fused_norm_rope_attn_rope_cast_fwd/permute_wv_proj
 88 - 2026-09-30 2026-09-30 1 1 +106 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
pow_2.hpp
in csrc/cuda_kernels/sm100/prefill/dense/common
 45 3 2026-09-30 2026-09-30 1 1 +92 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
fmha_cutlass_bwd_sm100.cu
in csrc/cuda_kernels/sm100/prefill/dense
 76 - 2026-09-30 2026-09-30 1 1 +87 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
fmha_cutlass_fwd_sm100.cu
in csrc/cuda_kernels/sm100/prefill/dense
 77 - 2026-09-30 2026-09-30 1 1 +87 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
fmha_options.hpp
in csrc/cuda_kernels/sm100/prefill/dense/kernel
 38 - 2026-09-30 2026-09-30 1 1 +85 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
helper.h
in csrc/cuda_kernels/sm100/prefill/dense/common
 28 - 2026-09-30 2026-09-30 1 1 +72 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
Files With Most Contributors (Top 50)
Based on the number of unique email addresses found in commits.

See data for all files...

File# lines# unitscreatedlast modified# changes
(days)		# contributorsline
churn		first
contributor		latest
contributor
321 11 2025-02-24 2026-09-30 10 6 +741 / -252 31004720+beginlner@users.no... shengyuliu@deepseek.com
__init__.py
in flash_mla
 19 - 2025-08-01 2026-09-10 5 2 +18 / -1 uchihatmtkinu@gmail.com shengyuliu@deepseek.com
common.h
in csrc/api
 223 14 2026-01-16 2026-09-30 4 2 +319 / -42 shengyuliu@deepseek.com shengyuliu@deepseek.com
params.h
in csrc
 129 - 2025-09-24 2026-09-30 5 1 +456 / -198 shengyuliu@deepseek.com shengyuliu@deepseek.com
api.cpp
in csrc/api
 14 - 2026-01-16 2026-09-30 3 1 +27 / -11 shengyuliu@deepseek.com shengyuliu@deepseek.com
dense_fwd.cpp
in csrc/api
 13 1 2026-09-10 2026-09-30 2 1 +16 / -1 shengyuliu@deepseek.com shengyuliu@deepseek.com
462 6 2026-09-10 2026-09-30 2 1 +589 / -25 shengyuliu@deepseek.com shengyuliu@deepseek.com
dense_bwd.cpp
in csrc/api
 13 1 2026-09-10 2026-09-30 2 1 +16 / -1 shengyuliu@deepseek.com shengyuliu@deepseek.com
396 9 2026-09-10 2026-09-30 2 1 +691 / -226 shengyuliu@deepseek.com shengyuliu@deepseek.com
195 6 2026-09-10 2026-09-30 2 1 +321 / -89 shengyuliu@deepseek.com shengyuliu@deepseek.com
77 4 2026-09-10 2026-09-30 2 1 +237 / -26 shengyuliu@deepseek.com shengyuliu@deepseek.com
utils.h
in csrc/cuda_kernels
 37 4 2026-09-30 2026-09-30 1 1 +47 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
get_decoding_sched_meta.h
in csrc/cuda_kernels/smxx/decode/get_decoding_sched_meta
 6 - 2026-09-30 2026-09-30 1 1 +10 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
get_decoding_sched_meta.cu
in csrc/cuda_kernels/smxx/decode/get_decoding_sched_meta
 132 - 2026-09-30 2026-09-30 1 1 +148 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
combine.h
in csrc/cuda_kernels/smxx/decode/combine
 6 - 2026-09-30 2026-09-30 1 1 +10 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
combine.cu
in csrc/cuda_kernels/smxx/decode/combine
 208 - 2026-09-30 2026-09-30 1 1 +240 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
defines.h
in csrc/cuda_kernels
 9 - 2026-09-30 2026-09-30 1 1 +12 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
kv_cache_utils.cuh
in csrc/cuda_kernels/sm100
 57 - 2026-09-30 2026-09-30 1 1 +64 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
fmha_cutlass_bwd_sm100.cuh
in csrc/cuda_kernels/sm100/prefill/dense
 171 - 2026-09-30 2026-09-30 1 1 +203 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
sm100_fmha_mla_fwd_mainloop_tma_warpspecialized.hpp
in csrc/cuda_kernels/sm100/prefill/dense/collective
 814 20 2026-09-30 2026-09-30 1 1 +1221 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
sm100_fmha_fwd_mainloop_tma_warpspecialized.hpp
in csrc/cuda_kernels/sm100/prefill/dense/collective
 805 20 2026-09-30 2026-09-30 1 1 +1214 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
fmha_fusion.hpp
in csrc/cuda_kernels/sm100/prefill/dense/collective
 277 21 2026-09-30 2026-09-30 1 1 +396 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
fmha_common.hpp
in csrc/cuda_kernels/sm100/prefill/dense/collective
 84 5 2026-09-30 2026-09-30 1 1 +127 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
sm100_fmha_load_tma_warpspecialized.hpp
in csrc/cuda_kernels/sm100/prefill/dense/collective
 205 4 2026-09-30 2026-09-30 1 1 +302 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
sm100_fmha_fwd_epilogue_tma_warpspecialized.hpp
in csrc/cuda_kernels/sm100/prefill/dense/collective
 148 6 2026-09-30 2026-09-30 1 1 +237 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
sm100_fmha_mla_load_tma_warpspecialized.hpp
in csrc/cuda_kernels/sm100/prefill/dense/collective
 220 5 2026-09-30 2026-09-30 1 1 +326 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
fmha_cutlass_bwd_sm100.cu
in csrc/cuda_kernels/sm100/prefill/dense
 76 - 2026-09-30 2026-09-30 1 1 +87 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
interface.h
in csrc/cuda_kernels/sm100/prefill/dense
 11 - 2026-09-30 2026-09-30 1 1 +14 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
pipeline_mla.hpp
in csrc/cuda_kernels/sm100/prefill/dense/common
 174 12 2026-09-30 2026-09-30 1 1 +250 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
helper.h
in csrc/cuda_kernels/sm100/prefill/dense/common
 28 - 2026-09-30 2026-09-30 1 1 +72 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
mask.cuh
in csrc/cuda_kernels/sm100/prefill/dense/common
 6 - 2026-09-30 2026-09-30 1 1 +8 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
pow_2.hpp
in csrc/cuda_kernels/sm100/prefill/dense/common
 45 3 2026-09-30 2026-09-30 1 1 +92 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
gather_tensor.hpp
in csrc/cuda_kernels/sm100/prefill/dense/common
 142 3 2026-09-30 2026-09-30 1 1 +215 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
utils.hpp
in csrc/cuda_kernels/sm100/prefill/dense/common
 26 - 2026-09-30 2026-09-30 1 1 +33 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
fmha_cutlass_fwd_sm100.cuh
in csrc/cuda_kernels/sm100/prefill/dense
 278 - 2026-09-30 2026-09-30 1 1 +335 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
fmha_device_bwd.hpp
in csrc/cuda_kernels/sm100/prefill/dense/device
 237 3 2026-09-30 2026-09-30 1 1 +340 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
fmha.hpp
in csrc/cuda_kernels/sm100/prefill/dense/device
 168 3 2026-09-30 2026-09-30 1 1 +281 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
fmha_cutlass_fwd_sm100.cu
in csrc/cuda_kernels/sm100/prefill/dense
 77 - 2026-09-30 2026-09-30 1 1 +87 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
fmha_options.hpp
in csrc/cuda_kernels/sm100/prefill/dense/kernel
 38 - 2026-09-30 2026-09-30 1 1 +85 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
fmha_tile_scheduler.hpp
in csrc/cuda_kernels/sm100/prefill/dense/kernel
 96 8 2026-09-30 2026-09-30 1 1 +162 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
sm100_fmha_bwd_mla_kernel_tma_warpspecialized.hpp
in csrc/cuda_kernels/sm100/prefill/dense/kernel
 1414 16 2026-09-30 2026-09-30 1 1 +1829 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
fmha_causal_tile_scheduler.hpp
in csrc/cuda_kernels/sm100/prefill/dense/kernel
 117 8 2026-09-30 2026-09-30 1 1 +197 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
sm100_fmha_bwd_kernel_tma_warpspecialized.hpp
in csrc/cuda_kernels/sm100/prefill/dense/kernel
 1416 16 2026-09-30 2026-09-30 1 1 +1835 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
fmha_kernel_bwd_sum_OdO.hpp
in csrc/cuda_kernels/sm100/prefill/dense/kernel
 106 7 2026-09-30 2026-09-30 1 1 +168 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
sm100_fmha_fwd_kernel_tma_warpspecialized.hpp
in csrc/cuda_kernels/sm100/prefill/dense/kernel
 491 11 2026-09-30 2026-09-30 1 1 +643 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
fmha_kernel_bwd_convert.hpp
in csrc/cuda_kernels/sm100/prefill/dense/kernel
 100 8 2026-09-30 2026-09-30 1 1 +160 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
phase1.cuh
in csrc/cuda_kernels/sm100/prefill/sparse/fwd/head64
 598 - 2026-09-30 2026-09-30 1 1 +681 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
config.h
in csrc/cuda_kernels/sm100/prefill/sparse/fwd/head64
 115 - 2026-09-30 2026-09-30 1 1 +147 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
phase1.h
in csrc/cuda_kernels/sm100/prefill/sparse/fwd/head64
 6 - 2026-09-30 2026-09-30 1 1 +10 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
phase1_h64_k512.cu
in csrc/cuda_kernels/sm100/prefill/sparse/fwd/head64/instantiations
 7 - 2026-09-30 2026-09-30 1 1 +10 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
Files With Least Contributors (Top 50)
Based on the number of unique email addresses found in commits.

See data for all files...

File# lines# unitscreatedlast modified# changes
(days)		# contributorsline
churn		first
contributor		latest
contributor
kernel.cuh
in csrc/cuda_kernels/sm100/prefill/sparse/fused_norm_rope_attn_rope_cast_fwd/core_attn
 1578 - 2026-09-30 2026-09-30 1 1 +1685 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
kernel_body.h
in csrc/ascend_kernels/prefill/sparse
 1440 16 2026-09-30 2026-09-30 1 1 +1701 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
sm100_fmha_bwd_kernel_tma_warpspecialized.hpp
in csrc/cuda_kernels/sm100/prefill/dense/kernel
 1416 16 2026-09-30 2026-09-30 1 1 +1835 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
sm100_fmha_bwd_mla_kernel_tma_warpspecialized.hpp
in csrc/cuda_kernels/sm100/prefill/dense/kernel
 1414 16 2026-09-30 2026-09-30 1 1 +1829 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
phase1.cuh
in csrc/cuda_kernels/sm100/prefill/sparse/fwd_for_small_topk/head128
 974 - 2026-09-30 2026-09-30 1 1 +1075 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
sm100_fmha_mla_fwd_mainloop_tma_warpspecialized.hpp
in csrc/cuda_kernels/sm100/prefill/dense/collective
 814 20 2026-09-30 2026-09-30 1 1 +1221 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
sm100_fmha_fwd_mainloop_tma_warpspecialized.hpp
in csrc/cuda_kernels/sm100/prefill/dense/collective
 805 20 2026-09-30 2026-09-30 1 1 +1214 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
kernel.cuh
in csrc/cuda_kernels/sm100/decode/sparse/head64
 667 - 2026-09-30 2026-09-30 1 1 +741 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
phase1.cuh
in csrc/cuda_kernels/sm100/prefill/sparse/fwd/head128
 625 - 2026-09-30 2026-09-30 1 1 +711 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
phase1.cuh
in csrc/cuda_kernels/sm100/prefill/sparse/fwd/head64
 598 - 2026-09-30 2026-09-30 1 1 +681 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
sm100_fmha_fwd_kernel_tma_warpspecialized.hpp
in csrc/cuda_kernels/sm100/prefill/dense/kernel
 491 11 2026-09-30 2026-09-30 1 1 +643 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
462 6 2026-09-10 2026-09-30 2 1 +589 / -25 shengyuliu@deepseek.com shengyuliu@deepseek.com
396 9 2026-09-10 2026-09-30 2 1 +691 / -226 shengyuliu@deepseek.com shengyuliu@deepseek.com
fmha_cutlass_fwd_sm100.cuh
in csrc/cuda_kernels/sm100/prefill/dense
 278 - 2026-09-30 2026-09-30 1 1 +335 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
fmha_fusion.hpp
in csrc/cuda_kernels/sm100/prefill/dense/collective
 277 21 2026-09-30 2026-09-30 1 1 +396 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
fmha_device_bwd.hpp
in csrc/cuda_kernels/sm100/prefill/dense/device
 237 3 2026-09-30 2026-09-30 1 1 +340 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
sm100_fmha_mla_load_tma_warpspecialized.hpp
in csrc/cuda_kernels/sm100/prefill/dense/collective
 220 5 2026-09-30 2026-09-30 1 1 +326 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
combine.cu
in csrc/cuda_kernels/smxx/decode/combine
 208 - 2026-09-30 2026-09-30 1 1 +240 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
sm100_fmha_load_tma_warpspecialized.hpp
in csrc/cuda_kernels/sm100/prefill/dense/collective
 205 4 2026-09-30 2026-09-30 1 1 +302 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
config.h
in csrc/cuda_kernels/sm100/prefill/sparse/fused_norm_rope_attn_rope_cast_fwd/core_attn
 198 4 2026-09-30 2026-09-30 1 1 +358 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
helpers.h
in csrc/cuda_kernels/sm100
 196 9 2026-09-30 2026-09-30 1 1 +232 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
195 6 2026-09-10 2026-09-30 2 1 +321 / -89 shengyuliu@deepseek.com shengyuliu@deepseek.com
pipeline_mla.hpp
in csrc/cuda_kernels/sm100/prefill/dense/common
 174 12 2026-09-30 2026-09-30 1 1 +250 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
fmha_cutlass_bwd_sm100.cuh
in csrc/cuda_kernels/sm100/prefill/dense
 171 - 2026-09-30 2026-09-30 1 1 +203 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
fmha.hpp
in csrc/cuda_kernels/sm100/prefill/dense/device
 168 3 2026-09-30 2026-09-30 1 1 +281 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
common_subroutine.h
in csrc/cuda_kernels/sm100
 168 5 2026-09-30 2026-09-30 1 1 +225 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
config.h
in csrc/cuda_kernels/sm100/decode/sparse/head64
 151 - 2026-09-30 2026-09-30 1 1 +191 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
sm100_fmha_fwd_epilogue_tma_warpspecialized.hpp
in csrc/cuda_kernels/sm100/prefill/dense/collective
 148 6 2026-09-30 2026-09-30 1 1 +237 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
gather_tensor.hpp
in csrc/cuda_kernels/sm100/prefill/dense/common
 142 3 2026-09-30 2026-09-30 1 1 +215 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
config.h
in csrc/ascend_kernels/prefill/sparse
 139 1 2026-09-30 2026-09-30 1 1 +206 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
get_decoding_sched_meta.cu
in csrc/cuda_kernels/smxx/decode/get_decoding_sched_meta
 132 - 2026-09-30 2026-09-30 1 1 +148 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
params.h
in csrc
 129 - 2025-09-24 2026-09-30 5 1 +456 / -198 shengyuliu@deepseek.com shengyuliu@deepseek.com
fmha_causal_tile_scheduler.hpp
in csrc/cuda_kernels/sm100/prefill/dense/kernel
 117 8 2026-09-30 2026-09-30 1 1 +197 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
config.h
in csrc/cuda_kernels/sm100/prefill/sparse/fwd/head64
 115 - 2026-09-30 2026-09-30 1 1 +147 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
config.h
in csrc/cuda_kernels/sm100/prefill/sparse/fwd_for_small_topk/head128
 113 - 2026-09-30 2026-09-30 1 1 +154 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
config.h
in csrc/cuda_kernels/sm100/prefill/sparse/fwd/head128
 107 - 2026-09-30 2026-09-30 1 1 +135 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
dequant_utils.cuh
in csrc/cuda_kernels/sm100
 107 - 2026-09-30 2026-09-30 1 1 +116 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
fmha_kernel_bwd_sum_OdO.hpp
in csrc/cuda_kernels/sm100/prefill/dense/kernel
 106 7 2026-09-30 2026-09-30 1 1 +168 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
fmha_kernel_bwd_convert.hpp
in csrc/cuda_kernels/sm100/prefill/dense/kernel
 100 8 2026-09-30 2026-09-30 1 1 +160 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
fmha_tile_scheduler.hpp
in csrc/cuda_kernels/sm100/prefill/dense/kernel
 96 8 2026-09-30 2026-09-30 1 1 +162 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
kernel.cu
in csrc/cuda_kernels/sm100/prefill/sparse/fused_norm_rope_attn_rope_cast_fwd/permute_wv_proj
 88 - 2026-09-30 2026-09-30 1 1 +106 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
fmha_common.hpp
in csrc/cuda_kernels/sm100/prefill/dense/collective
 84 5 2026-09-30 2026-09-30 1 1 +127 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
fmha_cutlass_fwd_sm100.cu
in csrc/cuda_kernels/sm100/prefill/dense
 77 - 2026-09-30 2026-09-30 1 1 +87 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
77 4 2026-09-10 2026-09-30 2 1 +237 / -26 shengyuliu@deepseek.com shengyuliu@deepseek.com
fmha_cutlass_bwd_sm100.cu
in csrc/cuda_kernels/sm100/prefill/dense
 76 - 2026-09-30 2026-09-30 1 1 +87 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
kv_cache_utils.cuh
in csrc/cuda_kernels/sm100
 57 - 2026-09-30 2026-09-30 1 1 +64 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
kernel.cu
in csrc/cuda_kernels/sm100/prefill/sparse/fused_norm_rope_attn_rope_cast_fwd/permute_q_b_proj
 56 - 2026-09-30 2026-09-30 1 1 +71 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
pow_2.hpp
in csrc/cuda_kernels/sm100/prefill/dense/common
 45 3 2026-09-30 2026-09-30 1 1 +92 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
kv_cache_format.h
in csrc/cuda_kernels
 45 2 2026-09-30 2026-09-30 1 1 +59 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
45 2 2026-09-30 2026-09-30 1 1 +64 / -0 shengyuliu@deepseek.com shengyuliu@deepseek.com
Correlations

File Size vs. Number of Changes: 97 points

csrc/api/api.cpp x: 14 lines of code y: 3 # changes csrc/api/common.h x: 223 lines of code y: 4 # changes csrc/api/dense_bwd.cpp x: 13 lines of code y: 2 # changes csrc/api/fused_norm_rope_attn_rope_cast_fwd.cpp x: 462 lines of code y: 2 # changes csrc/api/sparse_decode.cpp x: 396 lines of code y: 2 # changes csrc/api/sparse_prefill.cpp x: 195 lines of code y: 2 # changes csrc/ascend_kernels/prefill/sparse/config.h x: 139 lines of code y: 1 # changes csrc/ascend_kernels/prefill/sparse/fast_div_mod.h x: 28 lines of code y: 1 # changes csrc/ascend_kernels/prefill/sparse/kernel.h x: 16 lines of code y: 1 # changes csrc/ascend_kernels/prefill/sparse/kernel_body.h x: 1440 lines of code y: 1 # changes csrc/cuda_kernels/defines.h x: 9 lines of code y: 1 # changes csrc/cuda_kernels/kv_cache_format.h x: 45 lines of code y: 1 # changes csrc/cuda_kernels/sm100/common_subroutine.h x: 168 lines of code y: 1 # changes csrc/cuda_kernels/sm100/decode/sparse/head64/config.h x: 151 lines of code y: 1 # changes csrc/cuda_kernels/sm100/decode/sparse/head64/instantiations/v41_h64.cu x: 5 lines of code y: 1 # changes csrc/cuda_kernels/sm100/decode/sparse/head64/kernel.cuh x: 667 lines of code y: 1 # changes csrc/cuda_kernels/sm100/decode/sparse/head64/kernel.h x: 12 lines of code y: 1 # changes csrc/cuda_kernels/sm100/dequant_utils.cuh x: 107 lines of code y: 1 # changes csrc/cuda_kernels/sm100/helpers.h x: 196 lines of code y: 1 # changes csrc/cuda_kernels/sm100/kv_cache_utils.cuh x: 57 lines of code y: 1 # changes csrc/cuda_kernels/sm100/prefill/dense/collective/fmha_common.hpp x: 84 lines of code y: 1 # changes csrc/cuda_kernels/sm100/prefill/dense/collective/fmha_fusion.hpp x: 277 lines of code y: 1 # changes csrc/cuda_kernels/sm100/prefill/dense/collective/sm100_fmha_fwd_epilogue_tma_warpspecialized.hpp x: 148 lines of code y: 1 # changes csrc/cuda_kernels/sm100/prefill/dense/collective/sm100_fmha_fwd_mainloop_tma_warpspecialized.hpp x: 805 lines of code y: 1 # changes csrc/cuda_kernels/sm100/prefill/dense/collective/sm100_fmha_load_tma_warpspecialized.hpp x: 205 lines of code y: 1 # changes csrc/cuda_kernels/sm100/prefill/dense/collective/sm100_fmha_mla_fwd_mainloop_tma_warpspecialized.hpp x: 814 lines of code y: 1 # changes csrc/cuda_kernels/sm100/prefill/dense/collective/sm100_fmha_mla_load_tma_warpspecialized.hpp x: 220 lines of code y: 1 # changes csrc/cuda_kernels/sm100/prefill/dense/common/gather_tensor.hpp x: 142 lines of code y: 1 # changes csrc/cuda_kernels/sm100/prefill/dense/common/mask.cuh x: 6 lines of code y: 1 # changes csrc/cuda_kernels/sm100/prefill/dense/common/pipeline_mla.hpp x: 174 lines of code y: 1 # changes csrc/cuda_kernels/sm100/prefill/dense/device/fmha_device_bwd.hpp x: 237 lines of code y: 1 # changes csrc/cuda_kernels/sm100/prefill/dense/fmha_cutlass_bwd_sm100.cu x: 76 lines of code y: 1 # changes csrc/cuda_kernels/sm100/prefill/dense/fmha_cutlass_bwd_sm100.cuh x: 171 lines of code y: 1 # changes csrc/cuda_kernels/sm100/prefill/dense/fmha_cutlass_fwd_sm100.cu x: 77 lines of code y: 1 # changes csrc/cuda_kernels/sm100/prefill/dense/kernel/fmha_causal_tile_scheduler.hpp x: 117 lines of code y: 1 # changes csrc/cuda_kernels/sm100/prefill/dense/kernel/fmha_kernel_bwd_convert.hpp x: 100 lines of code y: 1 # changes csrc/cuda_kernels/sm100/prefill/dense/kernel/fmha_options.hpp x: 38 lines of code y: 1 # changes csrc/cuda_kernels/sm100/prefill/dense/kernel/fmha_tile_scheduler.hpp x: 96 lines of code y: 1 # changes csrc/cuda_kernels/sm100/prefill/dense/kernel/sm100_fmha_bwd_kernel_tma_warpspecialized.hpp x: 1416 lines of code y: 1 # changes csrc/cuda_kernels/sm100/prefill/dense/kernel/sm100_fmha_bwd_mla_kernel_tma_warpspecialized.hpp x: 1414 lines of code y: 1 # changes csrc/cuda_kernels/sm100/prefill/dense/kernel/sm100_fmha_fwd_kernel_tma_warpspecialized.hpp x: 491 lines of code y: 1 # changes csrc/cuda_kernels/sm100/prefill/sparse/fused_norm_rope_attn_rope_cast_fwd/core_attn/config.h x: 198 lines of code y: 1 # changes csrc/cuda_kernels/sm100/prefill/sparse/fused_norm_rope_attn_rope_cast_fwd/core_attn/kernel.cuh x: 1578 lines of code y: 1 # changes csrc/cuda_kernels/sm100/prefill/sparse/fused_norm_rope_attn_rope_cast_fwd/core_attn/kernel.h x: 34 lines of code y: 1 # changes csrc/cuda_kernels/sm100/prefill/sparse/fused_norm_rope_attn_rope_cast_fwd/permute_q_b_proj/kernel.cu x: 56 lines of code y: 1 # changes csrc/cuda_kernels/sm100/prefill/sparse/fused_norm_rope_attn_rope_cast_fwd/permute_q_b_proj/kernel.h x: 21 lines of code y: 1 # changes csrc/cuda_kernels/sm100/prefill/sparse/fused_norm_rope_attn_rope_cast_fwd/permute_wv_proj/kernel.cu x: 88 lines of code y: 1 # changes csrc/cuda_kernels/sm100/prefill/sparse/fwd/head128/instantiations/phase1_k512.cu x: 2 lines of code y: 1 # changes csrc/cuda_kernels/sm100/prefill/sparse/fwd/head128/phase1.cuh x: 625 lines of code y: 1 # changes csrc/cuda_kernels/sm100/prefill/sparse/fwd/head64/config.h x: 115 lines of code y: 1 # changes csrc/cuda_kernels/sm100/prefill/sparse/fwd/head64/phase1.cuh x: 598 lines of code y: 1 # changes csrc/cuda_kernels/sm100/prefill/sparse/fwd_for_small_topk/head128/config.h x: 113 lines of code y: 1 # changes csrc/cuda_kernels/sm100/prefill/sparse/fwd_for_small_topk/head128/phase1.cuh x: 974 lines of code y: 1 # changes csrc/cuda_kernels/smxx/decode/combine/combine.cu x: 208 lines of code y: 1 # changes csrc/cuda_kernels/smxx/decode/get_decoding_sched_meta/get_decoding_sched_meta.cu x: 132 lines of code y: 1 # changes csrc/cuda_kernels/utils.h x: 37 lines of code y: 1 # changes csrc/params.h x: 129 lines of code y: 5 # changes flash_mla/flash_mla_interface.py x: 321 lines of code y: 10 # changes flash_mla/fused_norm_rope_attn_rope_cast.py x: 77 lines of code y: 2 # changes flash_mla/__init__.py x: 19 lines of code y: 5 # changes
10.0
# changes
  min: 1.0
  average: 1.29
  25th percentile: 1.0
  median: 1.0
  75th percentile: 1.0
  max: 10.0
0 1578.0
lines of code
min: 2.0 | average: 183.12 | 25th percentile: 6.0 | median: 56.0 | 75th percentile: 184.5 | max: 1578.0

Number of Contributors vs. Number of Changes: 97 points

csrc/api/api.cpp x: 1 # contributors y: 3 # changes csrc/api/common.h x: 2 # contributors y: 4 # changes csrc/api/dense_bwd.cpp x: 1 # contributors y: 2 # changes csrc/ascend_kernels/prefill/sparse/config.h x: 1 # contributors y: 1 # changes csrc/params.h x: 1 # contributors y: 5 # changes flash_mla/flash_mla_interface.py x: 6 # contributors y: 10 # changes flash_mla/__init__.py x: 2 # contributors y: 5 # changes
10.0
# changes
  min: 1.0
  average: 1.29
  25th percentile: 1.0
  median: 1.0
  75th percentile: 1.0
  max: 10.0
0 6.0
# contributors
min: 1.0 | average: 1.07 | 25th percentile: 1.0 | median: 1.0 | 75th percentile: 1.0 | max: 6.0

Number of Contributors vs. File Size: 97 points

csrc/api/api.cpp x: 1 # contributors y: 14 lines of code csrc/api/common.h x: 2 # contributors y: 223 lines of code csrc/api/fused_norm_rope_attn_rope_cast_fwd.cpp x: 1 # contributors y: 462 lines of code csrc/api/sparse_decode.cpp x: 1 # contributors y: 396 lines of code csrc/api/sparse_prefill.cpp x: 1 # contributors y: 195 lines of code csrc/ascend_kernels/prefill/sparse/config.h x: 1 # contributors y: 139 lines of code csrc/ascend_kernels/prefill/sparse/fast_div_mod.h x: 1 # contributors y: 28 lines of code csrc/ascend_kernels/prefill/sparse/kernel_body.h x: 1 # contributors y: 1440 lines of code csrc/cuda_kernels/defines.h x: 1 # contributors y: 9 lines of code csrc/cuda_kernels/kv_cache_format.h x: 1 # contributors y: 45 lines of code csrc/cuda_kernels/sm100/common_subroutine.h x: 1 # contributors y: 168 lines of code csrc/cuda_kernels/sm100/decode/sparse/head64/config.h x: 1 # contributors y: 151 lines of code csrc/cuda_kernels/sm100/decode/sparse/head64/instantiations/v41_h64.cu x: 1 # contributors y: 5 lines of code csrc/cuda_kernels/sm100/decode/sparse/head64/kernel.cuh x: 1 # contributors y: 667 lines of code csrc/cuda_kernels/sm100/dequant_utils.cuh x: 1 # contributors y: 107 lines of code csrc/cuda_kernels/sm100/kv_cache_utils.cuh x: 1 # contributors y: 57 lines of code csrc/cuda_kernels/sm100/prefill/dense/collective/fmha_common.hpp x: 1 # contributors y: 84 lines of code csrc/cuda_kernels/sm100/prefill/dense/collective/fmha_fusion.hpp x: 1 # contributors y: 277 lines of code csrc/cuda_kernels/sm100/prefill/dense/collective/sm100_fmha_fwd_epilogue_tma_warpspecialized.hpp x: 1 # contributors y: 148 lines of code csrc/cuda_kernels/sm100/prefill/dense/collective/sm100_fmha_fwd_mainloop_tma_warpspecialized.hpp x: 1 # contributors y: 805 lines of code csrc/cuda_kernels/sm100/prefill/dense/collective/sm100_fmha_load_tma_warpspecialized.hpp x: 1 # contributors y: 205 lines of code csrc/cuda_kernels/sm100/prefill/dense/collective/sm100_fmha_mla_fwd_mainloop_tma_warpspecialized.hpp x: 1 # contributors y: 814 lines of code csrc/cuda_kernels/sm100/prefill/dense/collective/sm100_fmha_mla_load_tma_warpspecialized.hpp x: 1 # contributors y: 220 lines of code csrc/cuda_kernels/sm100/prefill/dense/common/pipeline_mla.hpp x: 1 # contributors y: 174 lines of code csrc/cuda_kernels/sm100/prefill/dense/device/fmha_device_bwd.hpp x: 1 # contributors y: 237 lines of code csrc/cuda_kernels/sm100/prefill/dense/fmha_cutlass_bwd_sm100.cu x: 1 # contributors y: 76 lines of code csrc/cuda_kernels/sm100/prefill/dense/kernel/fmha_causal_tile_scheduler.hpp x: 1 # contributors y: 117 lines of code csrc/cuda_kernels/sm100/prefill/dense/kernel/fmha_kernel_bwd_convert.hpp x: 1 # contributors y: 100 lines of code csrc/cuda_kernels/sm100/prefill/dense/kernel/fmha_options.hpp x: 1 # contributors y: 38 lines of code csrc/cuda_kernels/sm100/prefill/dense/kernel/fmha_tile_scheduler.hpp x: 1 # contributors y: 96 lines of code csrc/cuda_kernels/sm100/prefill/dense/kernel/sm100_fmha_bwd_kernel_tma_warpspecialized.hpp x: 1 # contributors y: 1416 lines of code csrc/cuda_kernels/sm100/prefill/dense/kernel/sm100_fmha_fwd_kernel_tma_warpspecialized.hpp x: 1 # contributors y: 491 lines of code csrc/cuda_kernels/sm100/prefill/sparse/fused_norm_rope_attn_rope_cast_fwd/core_attn/kernel.cuh x: 1 # contributors y: 1578 lines of code csrc/cuda_kernels/sm100/prefill/sparse/fused_norm_rope_attn_rope_cast_fwd/core_attn/kernel.h x: 1 # contributors y: 34 lines of code csrc/cuda_kernels/sm100/prefill/sparse/fused_norm_rope_attn_rope_cast_fwd/permute_q_b_proj/kernel.h x: 1 # contributors y: 21 lines of code csrc/cuda_kernels/sm100/prefill/sparse/fused_norm_rope_attn_rope_cast_fwd/permute_wv_proj/kernel.cu x: 1 # contributors y: 88 lines of code csrc/cuda_kernels/sm100/prefill/sparse/fwd/head128/phase1.cuh x: 1 # contributors y: 625 lines of code csrc/cuda_kernels/sm100/prefill/sparse/fwd/head64/phase1.cuh x: 1 # contributors y: 598 lines of code csrc/cuda_kernels/sm100/prefill/sparse/fwd_for_small_topk/head128/phase1.cuh x: 1 # contributors y: 974 lines of code csrc/cuda_kernels/smxx/decode/combine/combine.cu x: 1 # contributors y: 208 lines of code csrc/cuda_kernels/smxx/decode/get_decoding_sched_meta/get_decoding_sched_meta.cu x: 1 # contributors y: 132 lines of code csrc/params.h x: 1 # contributors y: 129 lines of code flash_mla/flash_mla_interface.py x: 6 # contributors y: 321 lines of code flash_mla/__init__.py x: 2 # contributors y: 19 lines of code
1578.0
lines of code
  min: 2.0
  average: 183.12
  25th percentile: 6.0
  median: 56.0
  75th percentile: 184.5
  max: 1578.0
0 6.0
# contributors
min: 1.0 | average: 1.07 | 25th percentile: 1.0 | median: 1.0 | 75th percentile: 1.0 | max: 6.0

Number of Changes vs. Line Churn: 457 points

.gitignore x: 8 # changes y: 11 line churn .gitmodules x: 1 # changes y: 4 line churn README.md x: 14 # changes y: 893 line churn benchmark/bench_flash_mla.py x: 5 # changes y: 1586 line churn benchmark/visualize.py x: 4 # changes y: 95 line churn csrc/3rdparty/kerutils/include/kerutils/common/common.h x: 1 # changes y: 41 line churn csrc/3rdparty/kerutils/include/kerutils/device/cuda/common.h x: 1 # changes y: 75 line churn csrc/3rdparty/kerutils/include/kerutils/device/cuda/sm100/gemm.cuh x: 1 # changes y: 694 line churn csrc/3rdparty/kerutils/include/kerutils/device/cuda/sm100/helpers.cuh x: 1 # changes y: 137 line churn csrc/3rdparty/kerutils/include/kerutils/device/cuda/sm100/intrinsics.cuh x: 1 # changes y: 649 line churn csrc/3rdparty/kerutils/include/kerutils/device/cuda/sm100/tma_cta_group2_nosplit.cuh x: 1 # changes y: 262 line churn csrc/3rdparty/kerutils/include/kerutils/device/cuda/sm80/helpers.cuh x: 1 # changes y: 55 line churn csrc/3rdparty/kerutils/include/kerutils/device/cuda/sm80/intrinsics.cuh x: 1 # changes y: 160 line churn csrc/3rdparty/kerutils/include/kerutils/device/cuda/sm90/helpers.cuh x: 1 # changes y: 110 line churn csrc/3rdparty/kerutils/include/kerutils/device/cuda/sm90/intrinsics.cuh x: 1 # changes y: 107 line churn csrc/3rdparty/kerutils/include/kerutils/host/host.h x: 1 # changes y: 293 line churn csrc/api/api.cpp x: 3 # changes y: 38 line churn csrc/api/common.h x: 4 # changes y: 361 line churn csrc/api/dense_bwd.cpp x: 2 # changes y: 17 line churn csrc/api/dense_decode.cpp x: 2 # changes y: 458 line churn csrc/api/fused_norm_rope_attn_rope_cast_fwd.cpp x: 2 # changes y: 614 line churn csrc/api/sparse_decode.cpp x: 2 # changes y: 917 line churn csrc/api/sparse_prefill.cpp x: 2 # changes y: 410 line churn csrc/ascend_kernels/prefill/sparse/config.h x: 1 # changes y: 206 line churn csrc/ascend_kernels/prefill/sparse/kernel_body.h x: 1 # changes y: 1701 line churn csrc/cuda_kernels/sm100/common_subroutine.h x: 1 # changes y: 225 line churn csrc/cuda_kernels/sm100/decode/sparse/head64/config.h x: 1 # changes y: 191 line churn csrc/cuda_kernels/sm100/decode/sparse/head64/kernel.cuh x: 1 # changes y: 741 line churn csrc/cuda_kernels/sm100/prefill/dense/collective/fmha_fusion.hpp x: 1 # changes y: 396 line churn csrc/cuda_kernels/sm100/prefill/dense/collective/sm100_fmha_fwd_mainloop_tma_warpspecialized.hpp x: 1 # changes y: 1214 line churn csrc/cuda_kernels/sm100/prefill/dense/collective/sm100_fmha_mla_load_tma_warpspecialized.hpp x: 1 # changes y: 326 line churn csrc/cuda_kernels/sm100/prefill/dense/device/fmha.hpp x: 1 # changes y: 281 line churn csrc/cuda_kernels/sm100/prefill/dense/device/fmha_device_bwd.hpp x: 1 # changes y: 340 line churn csrc/cuda_kernels/sm100/prefill/dense/kernel/sm100_fmha_bwd_kernel_tma_warpspecialized.hpp x: 1 # changes y: 1835 line churn csrc/cuda_kernels/sm100/prefill/sparse/fused_norm_rope_attn_rope_cast_fwd/core_attn/config.h x: 1 # changes y: 358 line churn csrc/cuda_kernels/sm100/prefill/sparse/fused_norm_rope_attn_rope_cast_fwd/core_attn/kernel.cuh x: 1 # changes y: 1685 line churn csrc/cuda_kernels/sm100/prefill/sparse/fwd/head128/phase1.cuh x: 1 # changes y: 711 line churn csrc/cuda_kernels/sm100/prefill/sparse/fwd_for_small_topk/head128/phase1.cuh x: 1 # changes y: 1075 line churn csrc/cutlass x: 3 # changes y: 5 line churn csrc/kernels/defines.h x: 2 # changes y: 52 line churn csrc/kernels/kv_cache_format.h x: 2 # changes y: 112 line churn csrc/kernels/params.h x: 4 # changes y: 488 line churn csrc/kernels/sm100/common_subroutine.h x: 2 # changes y: 484 line churn csrc/kernels/sm100/decode/sparse/head64/config.h x: 2 # changes y: 474 line churn csrc/kernels/sm100/decode/sparse/head64/kernel.cuh x: 2 # changes y: 1816 line churn csrc/kernels/sm100/decode/sparse/head64/kernel.h x: 2 # changes y: 32 line churn csrc/kernels/sm100/helpers.h x: 2 # changes y: 334 line churn csrc/kernels/sm100/prefill/dense/collective/fmha_common.hpp x: 2 # changes y: 254 line churn csrc/kernels/sm100/prefill/dense/collective/fmha_fusion.hpp x: 2 # changes y: 792 line churn csrc/kernels/sm100/prefill/dense/collective/sm100_fmha_fwd_mainloop_tma_warpspecialized.hpp x: 2 # changes y: 2428 line churn csrc/kernels/sm100/prefill/dense/collective/sm100_fmha_load_tma_warpspecialized.hpp x: 2 # changes y: 604 line churn csrc/kernels/sm100/prefill/dense/collective/sm100_fmha_mla_fwd_mainloop_tma_warpspecialized.hpp x: 2 # changes y: 2442 line churn csrc/kernels/sm100/prefill/dense/collective/sm100_fmha_mla_load_tma_warpspecialized.hpp x: 2 # changes y: 652 line churn csrc/kernels/sm100/prefill/dense/common/gather_tensor.hpp x: 2 # changes y: 430 line churn csrc/kernels/sm100/prefill/dense/common/helper.h x: 2 # changes y: 144 line churn csrc/kernels/sm100/prefill/dense/common/pow_2.hpp x: 2 # changes y: 184 line churn csrc/kernels/sm100/prefill/dense/common/utils.hpp x: 2 # changes y: 66 line churn csrc/kernels/sm100/prefill/dense/device/fmha.hpp x: 2 # changes y: 562 line churn csrc/kernels/sm100/prefill/dense/device/fmha_device_bwd.hpp x: 2 # changes y: 680 line churn csrc/kernels/sm100/prefill/dense/fmha_cutlass_bwd_sm100.cu x: 2 # changes y: 166 line churn csrc/kernels/sm100/prefill/dense/fmha_cutlass_fwd_sm100.cuh x: 2 # changes y: 670 line churn csrc/kernels/sm100/prefill/dense/kernel/fmha_causal_tile_scheduler.hpp x: 2 # changes y: 394 line churn csrc/kernels/sm100/prefill/dense/kernel/fmha_kernel_bwd_convert.hpp x: 2 # changes y: 320 line churn csrc/kernels/sm100/prefill/dense/kernel/sm100_fmha_bwd_kernel_tma_warpspecialized.hpp x: 2 # changes y: 3670 line churn csrc/kernels/sm100/prefill/dense/kernel/sm100_fmha_fwd_kernel_tma_warpspecialized.hpp x: 2 # changes y: 1286 line churn csrc/kernels/sm100/prefill/sparse/fused_norm_rope_attn_rope_cast_fwd/core_attn/kernel.cuh x: 3 # changes y: 3254 line churn csrc/kernels/sm100/prefill/sparse/fused_norm_rope_attn_rope_cast_fwd/permute_wv_proj/kernel.cu x: 2 # changes y: 212 line churn csrc/kernels/sm100/prefill/sparse/fwd/head128/config.h x: 2 # changes y: 268 line churn csrc/kernels/sm100/prefill/sparse/fwd/head128/phase1.cuh x: 2 # changes y: 1432 line churn csrc/kernels/sm100/prefill/sparse/fwd/head64/phase1.cuh x: 2 # changes y: 1596 line churn csrc/kernels/sm100/prefill/sparse/fwd_for_small_topk/head128/phase1.cuh x: 2 # changes y: 2294 line churn csrc/kernels/sm90/decode/dense/splitkv_mla.cuh x: 2 # changes y: 2710 line churn csrc/kernels/sm90/decode/dense/traits.h x: 2 # changes y: 222 line churn csrc/kernels/sm90/decode/sparse/components/helpers.h x: 3 # changes y: 220 line churn csrc/kernels/sm90/decode/sparse/splitkv_mla.cuh x: 2 # changes y: 1574 line churn csrc/kernels/sm90/prefill/sparse/config.h x: 2 # changes y: 294 line churn csrc/kernels/sm90/prefill/sparse/phase1.cuh x: 2 # changes y: 1292 line churn csrc/kernels/utils.h x: 4 # changes y: 228 line churn csrc/kerutils/include/kerutils/device/common.h x: 3 # changes y: 150 line churn csrc/kerutils/include/kerutils/device/sm100/gemm.cuh x: 3 # changes y: 1414 line churn csrc/kerutils/include/kerutils/device/sm100/intrinsics.cuh x: 3 # changes y: 1298 line churn csrc/kerutils/include/kerutils/device/sm100/tma_cta_group2_nosplit.cuh x: 2 # changes y: 524 line churn csrc/kerutils/include/kerutils/device/sm80/helpers.cuh x: 3 # changes y: 118 line churn csrc/kerutils/include/kerutils/device/sm80/intrinsics.cuh x: 3 # changes y: 324 line churn csrc/kerutils/include/kerutils/device/sm90/intrinsics.cuh x: 3 # changes y: 216 line churn csrc/kerutils/include/kerutils/host/host.h x: 3 # changes y: 412 line churn csrc/params.h x: 5 # changes y: 654 line churn docs/20250422-new-kernel-deep-dive.md x: 3 # changes y: 95 line churn flash_mla/flash_mla_interface.py x: 10 # changes y: 993 line churn setup.py x: 10 # changes y: 915 line churn tests/kernelkit/bench.py x: 2 # changes y: 359 line churn tests/lib.py x: 5 # changes y: 893 line churn tests/quant.py x: 6 # changes y: 663 line churn tests/test-fused-norm-rope-attn-rope-cast.py x: 1 # changes y: 537 line churn tests/test_flash_mla_sparse_decoding.py x: 3 # changes y: 722 line churn tests/test_flash_mla_sparse_prefill.py x: 3 # changes y: 366 line churn tests/test_fused_norm_rope_attn_rope_cast.py x: 2 # changes y: 1084 line churn csrc/api/sparse_decode.h x: 2 # changes y: 990 line churn csrc/sm100/decode/head64/kernel.cuh x: 2 # changes y: 1936 line churn csrc/sm100/prefill/dense/collective/fmha_common.hpp x: 3 # changes y: 381 line churn csrc/sm100/prefill/dense/collective/fmha_fusion.hpp x: 3 # changes y: 1188 line churn csrc/sm100/prefill/dense/collective/sm100_fmha_fwd_mainloop_tma_warpspecialized.hpp x: 3 # changes y: 3650 line churn csrc/sm100/prefill/dense/collective/sm100_fmha_load_tma_warpspecialized.hpp x: 3 # changes y: 906 line churn csrc/sm100/prefill/dense/collective/sm100_fmha_mla_fwd_mainloop_tma_warpspecialized.hpp x: 3 # changes y: 3671 line churn csrc/sm100/prefill/dense/collective/sm100_fmha_mla_load_tma_warpspecialized.hpp x: 3 # changes y: 978 line churn csrc/sm100/prefill/dense/common/gather_tensor.hpp x: 3 # changes y: 645 line churn csrc/sm100/prefill/dense/common/pipeline_mla.hpp x: 3 # changes y: 750 line churn csrc/sm100/prefill/dense/common/pow_2.hpp x: 3 # changes y: 276 line churn csrc/sm100/prefill/dense/device/fmha.hpp x: 3 # changes y: 843 line churn csrc/sm100/prefill/dense/device/fmha_device_bwd.hpp x: 3 # changes y: 1020 line churn csrc/sm100/prefill/dense/fmha_cutlass_bwd_sm100.cu x: 3 # changes y: 249 line churn csrc/sm100/prefill/dense/fmha_cutlass_bwd_sm100.cuh x: 4 # changes y: 605 line churn csrc/sm100/prefill/dense/fmha_cutlass_fwd_sm100.cuh x: 4 # changes y: 1003 line churn csrc/sm100/prefill/dense/kernel/fmha_causal_tile_scheduler.hpp x: 3 # changes y: 593 line churn csrc/sm100/prefill/dense/kernel/fmha_kernel_bwd_convert.hpp x: 5 # changes y: 496 line churn csrc/sm100/prefill/dense/kernel/fmha_kernel_bwd_sum_OdO.hpp x: 4 # changes y: 508 line churn csrc/sm100/prefill/dense/kernel/fmha_tile_scheduler.hpp x: 3 # changes y: 488 line churn csrc/sm100/prefill/dense/kernel/sm100_fmha_bwd_kernel_tma_warpspecialized.hpp x: 5 # changes y: 5560 line churn csrc/sm100/prefill/dense/kernel/sm100_fmha_fwd_kernel_tma_warpspecialized.hpp x: 4 # changes y: 1933 line churn csrc/sm100/prefill/sparse/fwd/head128/phase1.cuh x: 2 # changes y: 1412 line churn csrc/sm100/prefill/sparse/fwd/head64/phase1.cuh x: 2 # changes y: 1346 line churn csrc/sm100/prefill/sparse/fwd_for_small_topk/head128/phase1.cuh x: 2 # changes y: 2214 line churn csrc/sm90/decode/dense/splitkv_mla.h x: 4 # changes y: 32 line churn csrc/sm90/decode/sparse_fp8/components/config.h x: 4 # changes y: 373 line churn csrc/sm90/decode/sparse_fp8/components/helpers.h x: 4 # changes y: 308 line churn csrc/sm90/prefill/sparse/fwd.cu x: 4 # changes y: 2157 line churn csrc/smxx/decode/combine/combine.cu x: 4 # changes y: 468 line churn csrc/utils.h x: 6 # changes y: 508 line churn flash_mla/__init__.py x: 5 # changes y: 19 line churn csrc/pybind.cpp x: 4 # changes y: 1400 line churn csrc/sm100/decode/sparse_fp8/splitkv_mla.cu x: 2 # changes y: 1184 line churn csrc/sm100/intrinsics.h x: 2 # changes y: 922 line churn csrc/sm100/ws_gemm.h x: 2 # changes y: 852 line churn csrc/sm90/decode/dense/splitkv_mla.cu x: 3 # changes y: 4095 line churn csrc/sm90/decode/sparse_fp8/splitkv_mla.cu x: 3 # changes y: 1842 line churn csrc/sm90/prefill/sparse/helpers.h x: 3 # changes y: 531 line churn csrc/smxx/get_mla_metadata.cu x: 3 # changes y: 288 line churn csrc/smxx/mla_combine.cu x: 3 # changes y: 624 line churn tests/test_flash_mla_decoding.py x: 4 # changes y: 1043 line churn tests/test_fmha_sm100.py x: 7 # changes y: 363 line churn csrc/sm100/collective/fmha_fusion.hpp x: 4 # changes y: 1192 line churn csrc/sm100/collective/sm100_fmha_fwd_epilogue_tma_warpspecialized.hpp x: 4 # changes y: 710 line churn csrc/sm100/collective/sm100_fmha_fwd_mainloop_tma_warpspecialized.hpp x: 4 # changes y: 3654 line churn csrc/sm100/collective/sm100_fmha_load_tma_warpspecialized.hpp x: 5 # changes y: 969 line churn csrc/sm100/collective/sm100_fmha_mla_fwd_mainloop_tma_warpspecialized.hpp x: 4 # changes y: 3675 line churn csrc/sm100/collective/sm100_fmha_mla_load_tma_warpspecialized.hpp x: 5 # changes y: 1041 line churn csrc/sm100/common/utils.hpp x: 5 # changes y: 215 line churn csrc/sm100/device/fmha.hpp x: 4 # changes y: 838 line churn csrc/sm100/fmha_cutlass_fwd_sm100.cu x: 4 # changes y: 248 line churn csrc/sm100/fmha_cutlass_fwd_sm100.cuh x: 4 # changes y: 1007 line churn csrc/sm100/kernel/fmha_kernel_bwd_convert.hpp x: 3 # changes y: 459 line churn csrc/sm100/kernel/sm100_fmha_bwd_kernel_tma_warpspecialized.hpp x: 3 # changes y: 5523 line churn csrc/sm100/kernel/sm100_fmha_bwd_mla_kernel_tma_warpspecialized.hpp x: 3 # changes y: 5502 line churn csrc/sm100/kernel/sm100_fmha_fwd_kernel_tma_warpspecialized.hpp x: 4 # changes y: 1912 line churn csrc/sm100/pybind.cu x: 3 # changes y: 51 line churn csrc/sm90/kernels/params.h x: 3 # changes y: 174 line churn csrc/sm90/kernels/splitkv_mla.cu x: 3 # changes y: 4062 line churn csrc/flash_api.cpp x: 5 # changes y: 393 line churn csrc/kernels/splitkv_mla.cu x: 3 # changes y: 2714 line churn tests/test_flash_mla.py x: 4 # changes y: 289 line churn docs/assets/MLA Kernel Sched.drawio.svg x: 1 # changes y: 856 line churn
5560.0
line churn
  min: 1.0
  average: 460.43
  25th percentile: 16.0
  median: 162.0
  75th percentile: 504.0
  max: 5560.0
0 14.0
# changes
min: 1.0 | average: 2.21 | 25th percentile: 1.0 | median: 2.0 | 75th percentile: 3.0 | max: 14.0

Number of Contributors vs. Line Churn: 97 points

csrc/api/api.cpp x: 1 # contributors y: 38 line churn csrc/api/common.h x: 2 # contributors y: 361 line churn csrc/api/dense_bwd.cpp x: 1 # contributors y: 17 line churn csrc/api/fused_norm_rope_attn_rope_cast_fwd.cpp x: 1 # contributors y: 614 line churn csrc/api/sparse_decode.cpp x: 1 # contributors y: 917 line churn csrc/api/sparse_prefill.cpp x: 1 # contributors y: 410 line churn csrc/ascend_kernels/prefill/sparse/config.h x: 1 # contributors y: 206 line churn csrc/ascend_kernels/prefill/sparse/kernel.h x: 1 # contributors y: 22 line churn csrc/ascend_kernels/prefill/sparse/kernel_body.h x: 1 # contributors y: 1701 line churn csrc/cuda_kernels/defines.h x: 1 # contributors y: 12 line churn csrc/cuda_kernels/kv_cache_format.h x: 1 # contributors y: 59 line churn csrc/cuda_kernels/sm100/common_subroutine.h x: 1 # contributors y: 225 line churn csrc/cuda_kernels/sm100/decode/sparse/head64/config.h x: 1 # contributors y: 191 line churn csrc/cuda_kernels/sm100/decode/sparse/head64/kernel.cuh x: 1 # contributors y: 741 line churn csrc/cuda_kernels/sm100/dequant_utils.cuh x: 1 # contributors y: 116 line churn csrc/cuda_kernels/sm100/helpers.h x: 1 # contributors y: 232 line churn csrc/cuda_kernels/sm100/prefill/dense/collective/fmha_common.hpp x: 1 # contributors y: 127 line churn csrc/cuda_kernels/sm100/prefill/dense/collective/fmha_fusion.hpp x: 1 # contributors y: 396 line churn csrc/cuda_kernels/sm100/prefill/dense/collective/sm100_fmha_fwd_mainloop_tma_warpspecialized.hpp x: 1 # contributors y: 1214 line churn csrc/cuda_kernels/sm100/prefill/dense/collective/sm100_fmha_load_tma_warpspecialized.hpp x: 1 # contributors y: 302 line churn csrc/cuda_kernels/sm100/prefill/dense/collective/sm100_fmha_mla_fwd_mainloop_tma_warpspecialized.hpp x: 1 # contributors y: 1221 line churn csrc/cuda_kernels/sm100/prefill/dense/collective/sm100_fmha_mla_load_tma_warpspecialized.hpp x: 1 # contributors y: 326 line churn csrc/cuda_kernels/sm100/prefill/dense/common/gather_tensor.hpp x: 1 # contributors y: 215 line churn csrc/cuda_kernels/sm100/prefill/dense/common/helper.h x: 1 # contributors y: 72 line churn csrc/cuda_kernels/sm100/prefill/dense/common/pipeline_mla.hpp x: 1 # contributors y: 250 line churn csrc/cuda_kernels/sm100/prefill/dense/common/pow_2.hpp x: 1 # contributors y: 92 line churn csrc/cuda_kernels/sm100/prefill/dense/common/utils.hpp x: 1 # contributors y: 33 line churn csrc/cuda_kernels/sm100/prefill/dense/device/fmha.hpp x: 1 # contributors y: 281 line churn csrc/cuda_kernels/sm100/prefill/dense/device/fmha_device_bwd.hpp x: 1 # contributors y: 340 line churn csrc/cuda_kernels/sm100/prefill/dense/fmha_cutlass_fwd_sm100.cuh x: 1 # contributors y: 335 line churn csrc/cuda_kernels/sm100/prefill/dense/kernel/fmha_causal_tile_scheduler.hpp x: 1 # contributors y: 197 line churn csrc/cuda_kernels/sm100/prefill/dense/kernel/fmha_kernel_bwd_convert.hpp x: 1 # contributors y: 160 line churn csrc/cuda_kernels/sm100/prefill/dense/kernel/fmha_kernel_bwd_sum_OdO.hpp x: 1 # contributors y: 168 line churn csrc/cuda_kernels/sm100/prefill/dense/kernel/fmha_options.hpp x: 1 # contributors y: 85 line churn csrc/cuda_kernels/sm100/prefill/dense/kernel/sm100_fmha_bwd_kernel_tma_warpspecialized.hpp x: 1 # contributors y: 1835 line churn csrc/cuda_kernels/sm100/prefill/dense/kernel/sm100_fmha_fwd_kernel_tma_warpspecialized.hpp x: 1 # contributors y: 643 line churn csrc/cuda_kernels/sm100/prefill/sparse/fused_norm_rope_attn_rope_cast_fwd/core_attn/config.h x: 1 # contributors y: 358 line churn csrc/cuda_kernels/sm100/prefill/sparse/fused_norm_rope_attn_rope_cast_fwd/core_attn/kernel.cuh x: 1 # contributors y: 1685 line churn csrc/cuda_kernels/sm100/prefill/sparse/fused_norm_rope_attn_rope_cast_fwd/core_attn/kernel.h x: 1 # contributors y: 53 line churn csrc/cuda_kernels/sm100/prefill/sparse/fused_norm_rope_attn_rope_cast_fwd/permute_wv_proj/kernel.cu x: 1 # contributors y: 106 line churn csrc/cuda_kernels/sm100/prefill/sparse/fwd/head128/config.h x: 1 # contributors y: 135 line churn csrc/cuda_kernels/sm100/prefill/sparse/fwd/head128/instantiations/phase1_k512.cu x: 1 # contributors y: 2 line churn csrc/cuda_kernels/sm100/prefill/sparse/fwd/head128/phase1.cuh x: 1 # contributors y: 711 line churn csrc/cuda_kernels/sm100/prefill/sparse/fwd/head64/config.h x: 1 # contributors y: 147 line churn csrc/cuda_kernels/sm100/prefill/sparse/fwd/head64/phase1.cuh x: 1 # contributors y: 681 line churn csrc/cuda_kernels/sm100/prefill/sparse/fwd_for_small_topk/head128/config.h x: 1 # contributors y: 154 line churn csrc/cuda_kernels/sm100/prefill/sparse/fwd_for_small_topk/head128/phase1.cuh x: 1 # contributors y: 1075 line churn csrc/cuda_kernels/smxx/decode/combine/combine.cu x: 1 # contributors y: 240 line churn csrc/cuda_kernels/utils.h x: 1 # contributors y: 47 line churn csrc/params.h x: 1 # contributors y: 654 line churn flash_mla/flash_mla_interface.py x: 6 # contributors y: 993 line churn flash_mla/fused_norm_rope_attn_rope_cast.py x: 1 # contributors y: 263 line churn flash_mla/__init__.py x: 2 # contributors y: 19 line churn
1835.0
line churn
  min: 2.0
  average: 255.81
  25th percentile: 10.0
  median: 85.0
  75th percentile: 272.0
  max: 1835.0
0 6.0
# contributors
min: 1.0 | average: 1.07 | 25th percentile: 1.0 | median: 1.0 | 75th percentile: 1.0 | max: 6.0

File Size vs. Line Churn: 97 points

csrc/api/api.cpp x: 14 lines of code y: 38 line churn csrc/api/common.h x: 223 lines of code y: 361 line churn csrc/api/dense_bwd.cpp x: 13 lines of code y: 17 line churn csrc/api/fused_norm_rope_attn_rope_cast_fwd.cpp x: 462 lines of code y: 614 line churn csrc/api/sparse_decode.cpp x: 396 lines of code y: 917 line churn csrc/api/sparse_prefill.cpp x: 195 lines of code y: 410 line churn csrc/ascend_kernels/prefill/sparse/config.h x: 139 lines of code y: 206 line churn csrc/ascend_kernels/prefill/sparse/fast_div_mod.h x: 28 lines of code y: 37 line churn csrc/ascend_kernels/prefill/sparse/kernel.h x: 16 lines of code y: 22 line churn csrc/ascend_kernels/prefill/sparse/kernel_body.h x: 1440 lines of code y: 1701 line churn csrc/cuda_kernels/defines.h x: 9 lines of code y: 12 line churn csrc/cuda_kernels/kv_cache_format.h x: 45 lines of code y: 59 line churn csrc/cuda_kernels/sm100/common_subroutine.h x: 168 lines of code y: 225 line churn csrc/cuda_kernels/sm100/decode/sparse/head64/config.h x: 151 lines of code y: 191 line churn csrc/cuda_kernels/sm100/decode/sparse/head64/instantiations/v41_h64.cu x: 5 lines of code y: 8 line churn csrc/cuda_kernels/sm100/decode/sparse/head64/kernel.cuh x: 667 lines of code y: 741 line churn csrc/cuda_kernels/sm100/dequant_utils.cuh x: 107 lines of code y: 116 line churn csrc/cuda_kernels/sm100/helpers.h x: 196 lines of code y: 232 line churn csrc/cuda_kernels/sm100/kv_cache_utils.cuh x: 57 lines of code y: 64 line churn csrc/cuda_kernels/sm100/prefill/dense/collective/fmha_common.hpp x: 84 lines of code y: 127 line churn csrc/cuda_kernels/sm100/prefill/dense/collective/fmha_fusion.hpp x: 277 lines of code y: 396 line churn csrc/cuda_kernels/sm100/prefill/dense/collective/sm100_fmha_fwd_epilogue_tma_warpspecialized.hpp x: 148 lines of code y: 237 line churn csrc/cuda_kernels/sm100/prefill/dense/collective/sm100_fmha_fwd_mainloop_tma_warpspecialized.hpp x: 805 lines of code y: 1214 line churn csrc/cuda_kernels/sm100/prefill/dense/collective/sm100_fmha_load_tma_warpspecialized.hpp x: 205 lines of code y: 302 line churn csrc/cuda_kernels/sm100/prefill/dense/collective/sm100_fmha_mla_fwd_mainloop_tma_warpspecialized.hpp x: 814 lines of code y: 1221 line churn csrc/cuda_kernels/sm100/prefill/dense/collective/sm100_fmha_mla_load_tma_warpspecialized.hpp x: 220 lines of code y: 326 line churn csrc/cuda_kernels/sm100/prefill/dense/common/gather_tensor.hpp x: 142 lines of code y: 215 line churn csrc/cuda_kernels/sm100/prefill/dense/common/helper.h x: 28 lines of code y: 72 line churn csrc/cuda_kernels/sm100/prefill/dense/common/mask.cuh x: 6 lines of code y: 8 line churn csrc/cuda_kernels/sm100/prefill/dense/common/pipeline_mla.hpp x: 174 lines of code y: 250 line churn csrc/cuda_kernels/sm100/prefill/dense/common/pow_2.hpp x: 45 lines of code y: 92 line churn csrc/cuda_kernels/sm100/prefill/dense/common/utils.hpp x: 26 lines of code y: 33 line churn csrc/cuda_kernels/sm100/prefill/dense/device/fmha.hpp x: 168 lines of code y: 281 line churn csrc/cuda_kernels/sm100/prefill/dense/device/fmha_device_bwd.hpp x: 237 lines of code y: 340 line churn csrc/cuda_kernels/sm100/prefill/dense/fmha_cutlass_bwd_sm100.cu x: 76 lines of code y: 87 line churn csrc/cuda_kernels/sm100/prefill/dense/fmha_cutlass_bwd_sm100.cuh x: 171 lines of code y: 203 line churn csrc/cuda_kernels/sm100/prefill/dense/fmha_cutlass_fwd_sm100.cu x: 77 lines of code y: 87 line churn csrc/cuda_kernels/sm100/prefill/dense/fmha_cutlass_fwd_sm100.cuh x: 278 lines of code y: 335 line churn csrc/cuda_kernels/sm100/prefill/dense/kernel/fmha_causal_tile_scheduler.hpp x: 117 lines of code y: 197 line churn csrc/cuda_kernels/sm100/prefill/dense/kernel/fmha_kernel_bwd_convert.hpp x: 100 lines of code y: 160 line churn csrc/cuda_kernels/sm100/prefill/dense/kernel/fmha_kernel_bwd_sum_OdO.hpp x: 106 lines of code y: 168 line churn csrc/cuda_kernels/sm100/prefill/dense/kernel/fmha_options.hpp x: 38 lines of code y: 85 line churn csrc/cuda_kernels/sm100/prefill/dense/kernel/fmha_tile_scheduler.hpp x: 96 lines of code y: 162 line churn csrc/cuda_kernels/sm100/prefill/dense/kernel/sm100_fmha_bwd_kernel_tma_warpspecialized.hpp x: 1416 lines of code y: 1835 line churn csrc/cuda_kernels/sm100/prefill/dense/kernel/sm100_fmha_bwd_mla_kernel_tma_warpspecialized.hpp x: 1414 lines of code y: 1829 line churn csrc/cuda_kernels/sm100/prefill/dense/kernel/sm100_fmha_fwd_kernel_tma_warpspecialized.hpp x: 491 lines of code y: 643 line churn csrc/cuda_kernels/sm100/prefill/sparse/fused_norm_rope_attn_rope_cast_fwd/core_attn/config.h x: 198 lines of code y: 358 line churn csrc/cuda_kernels/sm100/prefill/sparse/fused_norm_rope_attn_rope_cast_fwd/core_attn/kernel.cuh x: 1578 lines of code y: 1685 line churn csrc/cuda_kernels/sm100/prefill/sparse/fused_norm_rope_attn_rope_cast_fwd/core_attn/kernel.h x: 34 lines of code y: 53 line churn csrc/cuda_kernels/sm100/prefill/sparse/fused_norm_rope_attn_rope_cast_fwd/permute_q_b_proj/kernel.cu x: 56 lines of code y: 71 line churn csrc/cuda_kernels/sm100/prefill/sparse/fused_norm_rope_attn_rope_cast_fwd/permute_q_b_proj/kernel.h x: 21 lines of code y: 31 line churn csrc/cuda_kernels/sm100/prefill/sparse/fused_norm_rope_attn_rope_cast_fwd/permute_wv_proj/kernel.cu x: 88 lines of code y: 106 line churn csrc/cuda_kernels/sm100/prefill/sparse/fwd/head128/config.h x: 107 lines of code y: 135 line churn csrc/cuda_kernels/sm100/prefill/sparse/fwd/head128/instantiations/phase1_k512.cu x: 2 lines of code y: 2 line churn csrc/cuda_kernels/sm100/prefill/sparse/fwd/head128/phase1.cuh x: 625 lines of code y: 711 line churn csrc/cuda_kernels/sm100/prefill/sparse/fwd/head64/config.h x: 115 lines of code y: 147 line churn csrc/cuda_kernels/sm100/prefill/sparse/fwd/head64/phase1.cuh x: 598 lines of code y: 681 line churn csrc/cuda_kernels/sm100/prefill/sparse/fwd_for_small_topk/head128/config.h x: 113 lines of code y: 154 line churn csrc/cuda_kernels/sm100/prefill/sparse/fwd_for_small_topk/head128/phase1.cuh x: 974 lines of code y: 1075 line churn csrc/cuda_kernels/smxx/decode/combine/combine.cu x: 208 lines of code y: 240 line churn csrc/cuda_kernels/smxx/decode/get_decoding_sched_meta/get_decoding_sched_meta.cu x: 132 lines of code y: 148 line churn csrc/cuda_kernels/utils.h x: 37 lines of code y: 47 line churn csrc/params.h x: 129 lines of code y: 654 line churn flash_mla/flash_mla_interface.py x: 321 lines of code y: 993 line churn flash_mla/fused_norm_rope_attn_rope_cast.py x: 77 lines of code y: 263 line churn flash_mla/__init__.py x: 19 lines of code y: 19 line churn
1835.0
line churn
  min: 2.0
  average: 255.81
  25th percentile: 10.0
  median: 85.0
  75th percentile: 272.0
  max: 1835.0
0 1578.0
lines of code
min: 2.0 | average: 183.12 | 25th percentile: 6.0 | median: 56.0 | 75th percentile: 184.5 | max: 1578.0