GitLab

* get rid of cudaMalloc and cudaFree

* minor fix

---------
Co-authored-by: Jiashi Li <js.li@high-flyer.cn>

* Add more GPU architctures support

* Merge fmha and mla runner

* add varlen & non varlen support, and add incontiguous tensor support

* update readme

* add varlen api

---------
Co-authored-by: dianzhangc <dianzhangc@nvidia.com>

Thank you for open source FlashMLA! Just read the write up and very amazing
work! Found some very minor mistakes regarding to typos, and the link
to the FlashAttention-3 paper is wrong as that is the original FlashAttention
paper, so I just send the PR here. Thanks again!
Signed-off-by: Hollow Man <hollowman@opensuse.org>

* Fix benchmark script

* Performance optimization for compute-bound cases

* Add new testcase (s_k = 16384)

* Update README.md

* Update comment

* Update README.md

* Add the deep-dive blog

* Add background color for MLA Kernel Sched.drawio.svg

* Use relative path for the schedule image

* Move flash_mla.h to kernels/params.h

fix(benchmark): store 'compare' and 'one' perf results in csv files and visualize them

Support FP16 dtype in FlashMLA kenrel

feat: add benchmark for flash_infer vs flash_mla

Update docstring

support Windows build

minor fix test

tests: Triton 3.2.0 had remove the fast_flush parameter from do_bench 

i