Skip to content

[Mthreads] Optimized mm addmm bmm with FlagOSTune #2643

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
huangyiqun merged 14 commits into from
Apr 27, 2026
Merged

[Mthreads] Optimized mm addmm bmm with FlagOSTune #2643

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
14 commits
Select commit Hold shift + click to select a range
b00c233
add flagtune support for mm and addmm operators on mthreads backend
Fattysand Apr 21, 2026
3f54d09
feat: add FlagTune auto-tuning support for mm and addmm operators
Fattysand Apr 21, 2026
d7e9f4e
add flagtune for bmm.py and bmm_mthreads_expand.yaml
Fattysand Apr 22, 2026
a98db0e
modify mm_hopper_expand.yaml
Fattysand Apr 22, 2026
cbf288f
modify SQMMA_ON switch for mthreads mm addmm bmm
Fattysand Apr 22, 2026
79b2611
fix mthreads ops and use configloader logic
Fattysand Apr 23, 2026
37e129b
sync utils.py
Fattysand Apr 23, 2026
3953ba7
fix bug in fp32 for mm/addmm/bmm
Fattysand Apr 24, 2026
c6605d6
modify test_blas_perf_parallel func: test_blas_benchmark
Fattysand Apr 24, 2026
0dcac0b
Modify as required
Fattysand Apr 24, 2026
1939717
fix test_blas_perf_parallel
Fattysand Apr 27, 2026
1dc8970
precommit test_blas_perf_parallel
Fattysand Apr 27, 2026
c410a3f
Merge branch 'flagos-ai:master' into feature/mm-addmm-bmm-flagtune-su…
Fattysand Apr 27, 2026
1537529
Merge branch 'flagos-ai:master' into feature/mm-addmm-bmm-flagtune-su…
Fattysand Apr 27, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
475 changes: 431 additions & 44 deletions benchmark/test_blas_perf_parallel.py
View file
Open in desktop

Large diffs are not rendered by default.

30 changes: 30 additions & 0 deletions src/flag_gems/runtime/backend/_mthreads/addmm_mthreads_expand.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,30 @@
addmm_sqmma:
Comment thread
huangyiqun marked this conversation as resolved.
- config:
param_map:
META:
BLOCK_SIZE_M: block_m
BLOCK_SIZE_N: block_n
BLOCK_SIZE_K: block_k
num_stages: stages
num_warps: warps
block_m:
- 64
- 128
block_n:
- 64
- 128
block_k:
- 64
- 128
stages:
- 3
- 4
- 5
- 6
warps:
- 4
- 8
- strategy:
M: default
N: default
K: default
30 changes: 30 additions & 0 deletions src/flag_gems/runtime/backend/_mthreads/bmm_mthreads_expand.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,30 @@
bmm_sqmma:
- config:
param_map:
META:
BLOCK_SIZE_M: block_m
BLOCK_SIZE_N: block_n
BLOCK_SIZE_K: block_k
num_stages: stages
num_warps: warps
block_m:
- 64
- 128
block_n:
- 64
- 128
block_k:
- 64
- 128
stages:
- 3
- 4
- 5
- 6
warps:
- 4
- 8
- strategy:
M: default
N: default
K: default
95 changes: 95 additions & 0 deletions src/flag_gems/runtime/backend/_mthreads/mm_mthreads_expand.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,95 @@
mm_general_tma:
- config:
param_map:
META:
BLOCK_M: block_m
BLOCK_N: block_n
BLOCK_K: block_k
num_stages: stages
num_warps: warps
block_m:
- 64
- 128
block_n:
- 64
- 128
block_k:
- 64
- 128
stages:
- 1
warps:
- 4
- 8
- strategy:
M: default
N: default
K: default
stride_am: default
stride_bk: default
dtype: default
mm:
- config:
param_map:
META:
BLOCK_M: block_m
BLOCK_N: block_n
BLOCK_K: block_k
num_stages: stages
num_warps: warps
block_m:
- 32
- 64
- 128
block_n:
- 32
- 64
- 128
block_k:
- 64
- 128
stages:
- 3
- 4
- 5
- 6
- 7
- 8
warps:
- 4
- 8
- strategy:
M: default
N: default
K: default
stride_am: default
stride_bk: default
gemv:
- config:
param_map:
META:
BLOCK_M: block_m
BLOCK_K: block_k
num_stages: stages
num_warps: warps
block_m:
- 8
- 16
- 32
- 64
block_k:
- 128
- 256
stages:
- 3
- 4
- 5
- 6
warps:
- 4
- 8
- strategy:
M: default
K: default
stride_am: default
stride_bk: default
Loading
Loading