Skip to content

[tutorial] Update the matmul kernel to run different WS modes. #19

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
htyu merged 1 commit into from
Jan 6, 2025
Merged

[tutorial] Update the matmul kernel to run different WS modes. #19

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
22 changes: 19 additions & 3 deletions python/tutorials/10-warp-specialized-matmul.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -105,14 +105,29 @@ def get_tma_descriptor_kernel_param(self, name):
"BLOCK_SIZE_N": 256,
"BLOCK_SIZE_K": 64,
"GROUP_SIZE_M": 8,
"NUM_CONSUMER_GROUPS": 2,
},
num_stages=2,
num_warps=4,
num_consumer_groups=2,
num_buffers_warp_spec=3,
),
triton.Config(
{
"BLOCK_SIZE_M": 64,
"BLOCK_SIZE_N": 64,
"BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 8,
"NUM_CONSUMER_GROUPS": 1,
},
num_stages=3,
num_warps=4,
num_consumer_groups=0, # disable warp specialization
num_buffers_warp_spec=3,
),
],
key=["M", "N", "K"],
use_cuda_graph=True,
)
@triton.jit
def matmul_persistent_tma_ws_cooperative_kernel(
Expand All @@ -126,6 +141,7 @@ def matmul_persistent_tma_ws_cooperative_kernel(
BLOCK_SIZE_N: tl.constexpr,
BLOCK_SIZE_K: tl.constexpr, #
GROUP_SIZE_M: tl.constexpr, #
NUM_CONSUMER_GROUPS: tl.constexpr,
):
"""Kernel for computing the matmul C = A x B.
A has shape (M, K), B has shape (K, N) and C has shape (M, N)
Expand Down Expand Up @@ -171,7 +187,7 @@ def matmul_persistent_tma_ws_cooperative_kernel(

c = accumulator.to(tl.float16)

with tl.async_task([1, 2]):
with tl.async_task([1, NUM_CONSUMER_GROUPS]):
tl._experimental_descriptor_store(c_ptr, c, [offs_am, offs_bn])


Expand Down Expand Up @@ -203,7 +219,7 @@ def grid(META):
a.data_ptr(),
M,
K,
META["BLOCK_SIZE_M"] // 2,
META["BLOCK_SIZE_M"] // META["NUM_CONSUMER_GROUPS"],
META["BLOCK_SIZE_K"],
a.element_size(),
)
Expand All @@ -222,7 +238,7 @@ def grid(META):
c.data_ptr(),
M,
N,
META["BLOCK_SIZE_M"] // 2,
META["BLOCK_SIZE_M"] // META["NUM_CONSUMER_GROUPS"],
META["BLOCK_SIZE_N"],
c.element_size(),
)
Expand Down