Skip to content
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.

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

Compilation failure with modified tuning parameters on CK Tile GEMM #1641

Open
zjing14 opened this issue Nov 5, 2024 · 6 comments
Open

Compilation failure with modified tuning parameters on CK Tile GEMM #1641

zjing14 opened this issue Nov 5, 2024 · 6 comments

Comments

@zjing14
Copy link
Contributor

zjing14 commented Nov 5, 2024

constexpr ck_tile::index_t M_Tile = 128;
constexpr ck_tile::index_t N_Tile = 128;
constexpr ck_tile::index_t K_Tile = 32;
constexpr ck_tile::index_t M_Warp = 2;
constexpr ck_tile::index_t N_Warp = 2;
constexpr ck_tile::index_t K_Warp = 1;
constexpr ck_tile::index_t M_Warp_Tile = 32;
constexpr ck_tile::index_t N_Warp_Tile = 32;
constexpr ck_tile::index_t K_Warp_Tile = 8;

Got the compilation failure with the following changes

    constexpr ck_tile::index_t M_Tile = 16;
    constexpr ck_tile::index_t N_Tile = 16;
    constexpr ck_tile::index_t K_Tile = 64;

    constexpr ck_tile::index_t M_Warp = 1;
    constexpr ck_tile::index_t N_Warp = 1;
    constexpr ck_tile::index_t K_Warp = 1;

    constexpr ck_tile::index_t M_Warp_Tile = 16;
    constexpr ck_tile::index_t N_Warp_Tile = 16;
    constexpr ck_tile::index_t K_Warp_Tile = 8;
@zjing14
Copy link
Contributor Author

zjing14 commented Nov 5, 2024

@aosewski @carlushuang

@zjing14 zjing14 changed the title Compilation failure with modified tuning parameters Compilation failure with modified tuning parameters on CK Tile GEMM Nov 5, 2024
@illsilin
Copy link
Collaborator

@zjing14 Which compiler version did you use and what was the error message?

@poyenc
Copy link
Contributor

poyenc commented Nov 12, 2024

I think we only have 16x16x16 warp gemm for now (v_mfma_f32_16x16x16f16/v_mfma_f32_16x16x16bf16_1k). And 16x16x8 warp gemm is only for bf16 (v_mfma_f32_16x16x8bf16), but we haven't added it in ck_tile.

@ThomasNing
Copy link
Contributor

ThomasNing commented Nov 13, 2024

Poyen is right. Just add a little bit, alternatively, we could have the v_mfma_f32_32x32x8_f16: 32x32x8 warp tile and v_mfma_f32_16x16x4_4b_f16: 16x16x4 for 4 matrix batches. We could add the first instruction set support to the ck_tile gemm first.

@aosewski
Copy link
Collaborator

@zjing14 If you would change k warp tile to 16: K_Warp_Tile = 16; and change in example to use universal gemm pipeline as in gemm_basic.cpp example then it should work. This example just doesn't have yet enabled universal gemm pipeline as of the date of writing. Will soon change all examples to use universal gemm pipeline when we will be sure that it works best.

@zjing14
Copy link
Contributor Author

zjing14 commented Nov 13, 2024

@poyenc @ThomasNing @illsilin @aosewski Thank you all very much. I will give a try with new changes.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Projects
None yet
Development

No branches or pull requests

6 participants