Skip to content

refine bf16 group gemm dispatch policy #177

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

Open
xinyu-intel wants to merge 1 commit into
base: main
Choose a base branch
from
+9 −3
Open

refine bf16 group gemm dispatch policy #177

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
4 changes: 2 additions & 2 deletions csrc/xpu/grouped_gemm/xe_2/gemm_xe2_policy.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -8,8 +8,8 @@ using namespace cute;

class xe_gemm_policy_base {
public:
using WGTile = Shape<_256, _256, _32>;
using SGLayout = Layout<Shape<_8, _4, _1>, Stride<_4, _1, _0>>;
using WGTile = Shape<_256, _128, _32>;
using SGLayout = Layout<Shape<_8, _2, _1>, Stride<_2, _1, _0>>;

// Copy can be turned for better performance
using GmemTiledCopyA = void; // same as make_block_2d_copy_A
Expand Down
8 changes: 7 additions & 1 deletion csrc/xpu/grouped_gemm/xe_2/grouped_gemm_xe2_interface.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -346,9 +346,15 @@ at::Tensor cutlass_grouped_gemm_xe2_impl(
MoEGEMMLauncherCallER('R', 'R', policy, scalar_t, scalar_t, scalar_t); \
}

if (A_avg_M <= 4) {
if (A_avg_M <= 8) {
using policy = w16a16_policy_m_8;
W16A16LauncherCallER(policy);
} else if (A_avg_M <= 16) {
using policy = w16a16_policy_m_16;
W16A16LauncherCallER(policy);
} else if (A_avg_M <= 32) {
using policy = w16a16_policy_m_32;
W16A16LauncherCallER(policy);
} else {
using policy = w16a16_policy;
W16A16LauncherCallER(policy);
Expand Down