deepseek-ai/DeepGEMM
1
0
Fork 0
mirror of https://github.com/deepseek-ai/DeepGEMM.git synced 2026-11-05 00:47:26 +00:00
Code Issues 162 Projects Packages Wiki Activity

[GH-ISSUE #22] How can I use a custom size? #17

New issue
Closed
opened 2026-10-26 21:10:35 +00:00 by bsakaguchi · 2 comments
bsakaguchi commented 2026-10-26 21:10:35 +00:00
Owner
Copy link

Originally created by @NexusMancer on GitHub (Feb 27, 2025).
Original GitHub issue: https://github.com/deepseek-ai/DeepGEMM/issues/22

I encountered an issue while customizing the size. Could you please advise me on how to resolve it?

while m = 73728, k = 128, n = 36864

def test_gemm() -> None:
    print('Testing GEMM:')
    for m in (73728,):
        for k, n in [(128, 36864)]:
            x_fp8, y_fp8, out, ref_out = construct(m, k, n)
            deep_gemm.gemm_fp8_fp8_bf16_nt(x_fp8, y_fp8, out)
            diff = calc_diff(out, ref_out)
            assert diff < 0.001, f'{m=}, {k=}, {n=}, {diff:.5f}'
            # noinspection PyShadowingNames
            def test_func():
                # Construct new tensors every time to avoid L2 cache acceleration
                x_fp8, y_fp8, out, ref_out = construct(m, k, n)
                deep_gemm.gemm_fp8_fp8_bf16_nt(x_fp8, y_fp8, out)
            t = bench_kineto(test_func, 'fp8_gemm', suppress_kineto_output=True)
            print(f' > Performance (m={m:5}, n={n:5}, k={k:5}): {t * 1e6:4.0f} us | '
                  f'throughput: {2 * m * n * k / t / 1e12:4.0f} TFLOPS, '
                  f'{(m * k + k * n + m * n * 2) / 1e9 / t:4.0f} GB/s')
    print()

issue like

(psekv3) [myname@gpu DeepGEMM]$ python test_core.py 

Library path:
 > ['/home/myname/DeepGEMM/deep_gemm']

Testing GEMM:
lhs.shape: torch.Size([73728, 128])
lhs_scales.shape: torch.Size([73728, 1])
rhs.shape: torch.Size([36864, 128])
rhs_scales.shape: torch.Size([288, 1])
/home/myname/DeepGEMM/deep_gemm/jit/../include/deep_gemm/fp8_gemm.cuh(103): error: static assertion failed with "Misaligned barriers"
      static_assert(not kMustUseUniformedScaleB or SHAPE_K_SCALES % (sizeof(Barrier) / sizeof(float)) == 0, "Misaligned barriers");
      ^
          detected during:
            instantiation of "void deep_gemm::fp8_gemm_kernel<SHAPE_N,SHAPE_K,BLOCK_M,BLOCK_N,BLOCK_K,kNumGroups,kNumStages,kNumTMAThreads,kNumMathThreadsPerGroup,kNumTMAMulticast,kGemmType>(__nv_bfloat16 *, float *, int *, uint32_t, CUtensorMap, CUtensorMap, CUtensorMap, CUtensorMap) [with SHAPE_N=36864U, SHAPE_K=128U, BLOCK_M=128U, BLOCK_N=128U, BLOCK_K=128U, kNumGroups=1U, kNumStages=5U, kNumTMAThreads=128U, kNumMathThreadsPerGroup=128U, kNumTMAMulticast=2U, kGemmType=deep_gemm::GemmType::Normal]" at line 368
            instantiation of "void deep_gemm::Gemm<SHAPE_N, SHAPE_K, BLOCK_M, BLOCK_N, BLOCK_K, kNumGroups, kNumStages, kNumTMAMulticast, kGemmType>::run(__nv_bfloat16 *, float *, int *, uint32_t, const CUtensorMap &, const CUtensorMap &, const CUtensorMap &, const CUtensorMap &, cudaStream_t, int, uint32_t) [with SHAPE_N=36864U, SHAPE_K=128U, BLOCK_M=128U, BLOCK_N=128U, BLOCK_K=128U, kNumGroups=1U, kNumStages=5U, kNumTMAMulticast=2U, kGemmType=deep_gemm::GemmType::Normal]" at line 37 of /home/myname/.deep_gemm/cache/kernel.gemm_fp8_fp8_bf16_nt.21f2d977befb/kernel.cu

1 error detected in the compilation of "/home/myname/.deep_gemm/cache/kernel.gemm_fp8_fp8_bf16_nt.21f2d977befb/kernel.cu".
Traceback (most recent call last):
  File "/home/myname/DeepGEMM/test_core.py", line 158, in <module>
    test_gemm()
  File "/home/myname/DeepGEMM/test_core.py", line 72, in test_gemm
    deep_gemm.gemm_fp8_fp8_bf16_nt(x_fp8, y_fp8, out)
  File "/home/myname/DeepGEMM/deep_gemm/jit_kernels/gemm.py", line 160, in gemm_fp8_fp8_bf16_nt
    runtime = jit_tuner.compile_and_tune(
  File "/home/myname/DeepGEMM/deep_gemm/jit_kernels/tuner.py", line 40, in compile_and_tune
    kernels.append((build(name, arg_defs, code), tuned_keys))
  File "/home/myname/DeepGEMM/deep_gemm/jit/compiler.py", line 139, in build
    assert subprocess.check_call(command) == 0, f'Failed to compile {src_path}'
  File "/home/myname/miniconda3/envs/psekv3/lib/python3.10/subprocess.py", line 369, in check_call
    raise CalledProcessError(retcode, cmd)
subprocess.CalledProcessError: Command '['/usr/local/cuda/bin/nvcc', '/home/myname/.deep_gemm/cache/kernel.gemm_fp8_fp8_bf16_nt.21f2d977befb/kernel.cu', '-o', '/home/myname/.deep_gemm/tmp/nvcc.tmp.44fc9af3-79ad-4dbd-9358-04df3cfd478f.1c2307ed5faf.so', '-std=c++17', '-shared', '-O3', '--expt-relaxed-constexpr', '--expt-extended-lambda', '-gencode=arch=compute_90a,code=sm_90a', '--ptxas-options=--register-usage-level=10', '--diag-suppress=177,174,940', '--compiler-options=-fPIC,-O3,-Wno-deprecated-declarations,-Wno-abi', '-I/home/myname/DeepGEMM/deep_gemm/jit/../include']' returned non-zero exit status 1.
Originally created by @NexusMancer on GitHub (Feb 27, 2025). Original GitHub issue: https://github.com/deepseek-ai/DeepGEMM/issues/22 I encountered an issue while customizing the size. Could you please advise me on how to resolve it? **_while m = 73728, k = 128, n = 36864_** ``` def test_gemm() -> None: print('Testing GEMM:') for m in (73728,): for k, n in [(128, 36864)]: x_fp8, y_fp8, out, ref_out = construct(m, k, n) deep_gemm.gemm_fp8_fp8_bf16_nt(x_fp8, y_fp8, out) diff = calc_diff(out, ref_out) assert diff < 0.001, f'{m=}, {k=}, {n=}, {diff:.5f}' # noinspection PyShadowingNames def test_func(): # Construct new tensors every time to avoid L2 cache acceleration x_fp8, y_fp8, out, ref_out = construct(m, k, n) deep_gemm.gemm_fp8_fp8_bf16_nt(x_fp8, y_fp8, out) t = bench_kineto(test_func, 'fp8_gemm', suppress_kineto_output=True) print(f' > Performance (m={m:5}, n={n:5}, k={k:5}): {t * 1e6:4.0f} us | ' f'throughput: {2 * m * n * k / t / 1e12:4.0f} TFLOPS, ' f'{(m * k + k * n + m * n * 2) / 1e9 / t:4.0f} GB/s') print() ``` **issue like** ``` (psekv3) [myname@gpu DeepGEMM]$ python test_core.py Library path: > ['/home/myname/DeepGEMM/deep_gemm'] Testing GEMM: lhs.shape: torch.Size([73728, 128]) lhs_scales.shape: torch.Size([73728, 1]) rhs.shape: torch.Size([36864, 128]) rhs_scales.shape: torch.Size([288, 1]) /home/myname/DeepGEMM/deep_gemm/jit/../include/deep_gemm/fp8_gemm.cuh(103): error: static assertion failed with "Misaligned barriers" static_assert(not kMustUseUniformedScaleB or SHAPE_K_SCALES % (sizeof(Barrier) / sizeof(float)) == 0, "Misaligned barriers"); ^ detected during: instantiation of "void deep_gemm::fp8_gemm_kernel<SHAPE_N,SHAPE_K,BLOCK_M,BLOCK_N,BLOCK_K,kNumGroups,kNumStages,kNumTMAThreads,kNumMathThreadsPerGroup,kNumTMAMulticast,kGemmType>(__nv_bfloat16 *, float *, int *, uint32_t, CUtensorMap, CUtensorMap, CUtensorMap, CUtensorMap) [with SHAPE_N=36864U, SHAPE_K=128U, BLOCK_M=128U, BLOCK_N=128U, BLOCK_K=128U, kNumGroups=1U, kNumStages=5U, kNumTMAThreads=128U, kNumMathThreadsPerGroup=128U, kNumTMAMulticast=2U, kGemmType=deep_gemm::GemmType::Normal]" at line 368 instantiation of "void deep_gemm::Gemm<SHAPE_N, SHAPE_K, BLOCK_M, BLOCK_N, BLOCK_K, kNumGroups, kNumStages, kNumTMAMulticast, kGemmType>::run(__nv_bfloat16 *, float *, int *, uint32_t, const CUtensorMap &, const CUtensorMap &, const CUtensorMap &, const CUtensorMap &, cudaStream_t, int, uint32_t) [with SHAPE_N=36864U, SHAPE_K=128U, BLOCK_M=128U, BLOCK_N=128U, BLOCK_K=128U, kNumGroups=1U, kNumStages=5U, kNumTMAMulticast=2U, kGemmType=deep_gemm::GemmType::Normal]" at line 37 of /home/myname/.deep_gemm/cache/kernel.gemm_fp8_fp8_bf16_nt.21f2d977befb/kernel.cu 1 error detected in the compilation of "/home/myname/.deep_gemm/cache/kernel.gemm_fp8_fp8_bf16_nt.21f2d977befb/kernel.cu". Traceback (most recent call last): File "/home/myname/DeepGEMM/test_core.py", line 158, in <module> test_gemm() File "/home/myname/DeepGEMM/test_core.py", line 72, in test_gemm deep_gemm.gemm_fp8_fp8_bf16_nt(x_fp8, y_fp8, out) File "/home/myname/DeepGEMM/deep_gemm/jit_kernels/gemm.py", line 160, in gemm_fp8_fp8_bf16_nt runtime = jit_tuner.compile_and_tune( File "/home/myname/DeepGEMM/deep_gemm/jit_kernels/tuner.py", line 40, in compile_and_tune kernels.append((build(name, arg_defs, code), tuned_keys)) File "/home/myname/DeepGEMM/deep_gemm/jit/compiler.py", line 139, in build assert subprocess.check_call(command) == 0, f'Failed to compile {src_path}' File "/home/myname/miniconda3/envs/psekv3/lib/python3.10/subprocess.py", line 369, in check_call raise CalledProcessError(retcode, cmd) subprocess.CalledProcessError: Command '['/usr/local/cuda/bin/nvcc', '/home/myname/.deep_gemm/cache/kernel.gemm_fp8_fp8_bf16_nt.21f2d977befb/kernel.cu', '-o', '/home/myname/.deep_gemm/tmp/nvcc.tmp.44fc9af3-79ad-4dbd-9358-04df3cfd478f.1c2307ed5faf.so', '-std=c++17', '-shared', '-O3', '--expt-relaxed-constexpr', '--expt-extended-lambda', '-gencode=arch=compute_90a,code=sm_90a', '--ptxas-options=--register-usage-level=10', '--diag-suppress=177,174,940', '--compiler-options=-fPIC,-O3,-Wno-deprecated-declarations,-Wno-abi', '-I/home/myname/DeepGEMM/deep_gemm/jit/../include']' returned non-zero exit status 1. ```
bsakaguchi commented 2026-10-26 21:10:36 +00:00
Author
Owner
Copy link

@LyricZhao commented on GitHub (Feb 28, 2025):

Fixed in github.com/deepseek-ai/DeepGEMM@6c5da03ba9, thanks for feedback!

BTW, your matrix size is so big which leads to OOM on H800s (calc_diff function uses too much VRAM). So I tested it with smaller M=1024.

<!-- gh-comment-id:2689537301 --> @LyricZhao commented on GitHub (Feb 28, 2025): Fixed in https://github.com/deepseek-ai/DeepGEMM/commit/6c5da03ba9a311b69aaeb3236ceb674e714950c1, thanks for feedback! BTW, your matrix size is so big which leads to OOM on H800s (`calc_diff` function uses too much VRAM). So I tested it with smaller M=1024.
bsakaguchi commented 2026-10-26 21:10:36 +00:00
Author
Owner
Copy link

@NexusMancer commented on GitHub (Feb 28, 2025):

Fixed in 6c5da03, thanks for feedback!

BTW, your matrix size is so big which leads to OOM on H800s (calc_diff function uses too much VRAM). So I tested it with smaller M=1024.

The issue has been successfully resolved. Thank you very much!!!

<!-- gh-comment-id:2690055365 --> @NexusMancer commented on GitHub (Feb 28, 2025): > Fixed in [6c5da03](https://github.com/deepseek-ai/DeepGEMM/commit/6c5da03ba9a311b69aaeb3236ceb674e714950c1), thanks for feedback! > > BTW, your matrix size is so big which leads to OOM on H800s (`calc_diff` function uses too much VRAM). So I tested it with smaller M=1024. The issue has been successfully resolved. Thank you very much!!!
Sign in to join this conversation.
No Branch/Tag specified
Branches Tags
main
public-sync-20261014
0930-release
nv_dev
fix-mega-moe-fence
public-release-2609
public-release-2607
public-release-2606
mega-update
public-release-260416
ko3n1g/ci/build-wheels
nv_dev_f8e8fb5
nv_dev_950d31a
nv_dev_67fc648
nv_dev_c491439
nv_dev_4ff3f54
v2.1.1.post3
v2.1.1.post2
v2.1.1.post1
v2.1.1
v2.1.0
Labels
Clear labels   
pull-request
Mirrored from GitHub Pull Request

No labels
pull-request
Milestone
Clear milestone
No items
No milestone
Projects
Clear projects
No items
No project
Assignees
Clear assignees
No assignees
1 participant
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set.

Reference
deepseek-ai/DeepGEMM#17
No description provided.