[PR #173] [MERGED] [Fix] fix B200 cu128 NVCC compilation failed #277

New issue

Closed

opened 2026-10-26 21:15:39 +00:00 by bsakaguchi · 0 comments

bsakaguchi commented

2026-10-26 21:15:39 +00:00

Owner

📋 Pull Request Information

Original PR: https://github.com/deepseek-ai/DeepGEMM/pull/173
Author: @FlamingoPg
Created: 8/26/2025
Status: ✅ Merged
Merged: 8/27/2025
Merged by: @LyricZhao

Base: main ← Head: main

📝 Commits (1)

2074fa7 fix B200 cu128 NVCC compilation failed

📊 Changes

1 file changed (+10 additions, -2 deletions)

View changed files

📝 csrc/jit/compiler.hpp (+10 -2)

📄 Description

cu128 B200 failed with

➜  DeepGEMM git:(main) python3 /sgl-workspace/DeepGEMM/tests/test_bf16.py                                           
Library path:
 > ['/usr/local/lib/python3.12/dist-packages/deep_gemm']

Testing GEMM:
Warning: please use at least NVCC 12.9 for the best DeepGEMM performanceNVCC compilation failed: nvcc fatal   : Unsupported gpu architecture 'sm_100f'

Traceback (most recent call last):
  File "/sgl-workspace/DeepGEMM/tests/test_bf16.py", line 123, in <module>
    test_gemm()
  File "/sgl-workspace/DeepGEMM/tests/test_bf16.py", line 30, in test_gemm
    getattr(deep_gemm, func_name)(a, b, d, c=c)
RuntimeError: Assertion error (csrc/apis/../jit_kernels/impls/../../jit/compiler.hpp:176): false and "NVCC compilation failed"

After fix

➜  DeepGEMM git:(main) ✗ python3 /sgl-workspace/DeepGEMM/tests/test_bf16.py                                         
Library path:
 > ['/usr/local/lib/python3.12/dist-packages/deep_gemm']

Testing GEMM:
 > Perf (m=  128, n= 2112, k= 7168, layout=NT, BF16, acc=0):   18 us |  221 TFLOPS | 1860 GB/s | 0.79x cuBLAS
 > Perf (m=  128, n= 2112, k= 7168, layout=NT, FP32, acc=0):   18 us |  212 TFLOPS | 1815 GB/s | 0.00x cuBLAS
 > Perf (m=  128, n= 2112, k= 7168, layout=NT, FP32, acc=1):   18 us |  212 TFLOPS | 1871 GB/s | 0.00x cuBLAS
 > Perf (m=  128, n= 2112, k= 7168, layout=NN, BF16, acc=0):   18 us |  218 TFLOPS | 1833 GB/s | 0.78x cuBLAS
 > Perf (m=  128, n= 2112, k= 7168, layout=NN, FP32, acc=0):   19 us |  207 TFLOPS | 1774 GB/s | 0.00x cuBLAS
 > Perf (m=  128, n= 2112, k= 7168, layout=NN, FP32, acc=1):   19 us |  206 TFLOPS | 1821 GB/s | 0.00x cuBLAS
 > Perf (m=  128, n= 2112, k= 7168, layout=TT, BF16, acc=0):   18 us |  220 TFLOPS | 1851 GB/s | 0.79x cuBLAS
 > Perf (m=  128, n= 2112, k= 7168, layout=TT, FP32, acc=0):   18 us |  212 TFLOPS | 1820 GB/s | 0.00x cuBLAS
 > Perf (m=  128, n= 2112, k= 7168, layout=TT, FP32, acc=1):   18 us |  210 TFLOPS | 1855 GB/s | 0.00x cuBLAS
 > Perf (m=  128, n= 2112, k= 7168, layout=TN, BF16, acc=0):   18 us |  218 TFLOPS | 1837 GB/s | 0.77x cuBLAS

_{🔄 This issue represents a GitHub Pull Request. It cannot be merged through Gitea due to API limitations.}

## 📋 Pull Request Information **Original PR:** https://github.com/deepseek-ai/DeepGEMM/pull/173 **Author:** [@FlamingoPg](https://github.com/FlamingoPg) **Created:** 8/26/2025 **Status:** ✅ Merged **Merged:** 8/27/2025 **Merged by:** [@LyricZhao](https://github.com/LyricZhao) **Base:** `main` ← **Head:** `main` --- ### 📝 Commits (1) - [`2074fa7`](https://github.com/deepseek-ai/DeepGEMM/commit/2074fa75b3942883404debe23ef5d95347425caa) fix B200 cu128 NVCC compilation failed ### 📊 Changes **1 file changed** (+10 additions, -2 deletions) <details> <summary>View changed files</summary> 📝 `csrc/jit/compiler.hpp` (+10 -2) </details> ### 📄 Description cu128 B200 failed with ```bash ➜ DeepGEMM git:(main) python3 /sgl-workspace/DeepGEMM/tests/test_bf16.py Library path: > ['/usr/local/lib/python3.12/dist-packages/deep_gemm'] Testing GEMM: Warning: please use at least NVCC 12.9 for the best DeepGEMM performanceNVCC compilation failed: nvcc fatal : Unsupported gpu architecture 'sm_100f' Traceback (most recent call last): File "/sgl-workspace/DeepGEMM/tests/test_bf16.py", line 123, in <module> test_gemm() File "/sgl-workspace/DeepGEMM/tests/test_bf16.py", line 30, in test_gemm getattr(deep_gemm, func_name)(a, b, d, c=c) RuntimeError: Assertion error (csrc/apis/../jit_kernels/impls/../../jit/compiler.hpp:176): false and "NVCC compilation failed" ``` After fix ```bash ➜ DeepGEMM git:(main) ✗ python3 /sgl-workspace/DeepGEMM/tests/test_bf16.py Library path: > ['/usr/local/lib/python3.12/dist-packages/deep_gemm'] Testing GEMM: > Perf (m= 128, n= 2112, k= 7168, layout=NT, BF16, acc=0): 18 us | 221 TFLOPS | 1860 GB/s | 0.79x cuBLAS > Perf (m= 128, n= 2112, k= 7168, layout=NT, FP32, acc=0): 18 us | 212 TFLOPS | 1815 GB/s | 0.00x cuBLAS > Perf (m= 128, n= 2112, k= 7168, layout=NT, FP32, acc=1): 18 us | 212 TFLOPS | 1871 GB/s | 0.00x cuBLAS > Perf (m= 128, n= 2112, k= 7168, layout=NN, BF16, acc=0): 18 us | 218 TFLOPS | 1833 GB/s | 0.78x cuBLAS > Perf (m= 128, n= 2112, k= 7168, layout=NN, FP32, acc=0): 19 us | 207 TFLOPS | 1774 GB/s | 0.00x cuBLAS > Perf (m= 128, n= 2112, k= 7168, layout=NN, FP32, acc=1): 19 us | 206 TFLOPS | 1821 GB/s | 0.00x cuBLAS > Perf (m= 128, n= 2112, k= 7168, layout=TT, BF16, acc=0): 18 us | 220 TFLOPS | 1851 GB/s | 0.79x cuBLAS > Perf (m= 128, n= 2112, k= 7168, layout=TT, FP32, acc=0): 18 us | 212 TFLOPS | 1820 GB/s | 0.00x cuBLAS > Perf (m= 128, n= 2112, k= 7168, layout=TT, FP32, acc=1): 18 us | 210 TFLOPS | 1855 GB/s | 0.00x cuBLAS > Perf (m= 128, n= 2112, k= 7168, layout=TN, BF16, acc=0): 18 us | 218 TFLOPS | 1837 GB/s | 0.77x cuBLAS ``` --- <sub>🔄 This issue represents a GitHub Pull Request. It cannot be merged through Gitea due to API limitations.</sub>

bsakaguchi

2026-10-26 21:15:39 +00:00

closed this issue
added the
pull-request
label

No labels

pull-request

No milestone

No project

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#277

No description provided.

Rows
Columns