deepseek-ai/DeepGEMM
DeepGEMM: clean and efficient BLAS kernel library on GPU
Cuda★ 8,776+199 that day#6 on trendingD 51.6 health12.61% duplicated
D51.6/100
Duplication36
13.4% in cpp-header, python, bash, perl, cppDead code89
0.8%Complexity59
40%Most complex files
Dead code 0.75% unused · 12 findings
- deep_gemm/locality_domain.py:get_granularityunused-import
- deep_gemm/utils/layout.py:get_tma_aligned_sizeunused-import
- deep_gemm/locality_domain.py:emptyunused-import
- deep_gemm/utils/layout.py:get_mn_major_tma_aligned_tensorunused-import
- deep_gemm/utils/layout.py:get_mn_major_tma_aligned_packed_ue8m0_tensorunused-import
- deep_gemm/utils/layout.py:get_k_grouped_mn_major_tma_aligned_packed_ue8m0_tensorunused-import
- deep_gemm/locality_domain.py:get_sm_locality_domainsunused-import
- deep_gemm/utils/layout.py:set_mk_alignment_for_contiguous_layoutunused-import
- deep_gemm/locality_domain.py:get_balanced_sm_locality_domainsunused-import
- deep_gemm/utils/layout.py:get_theoretical_mk_alignment_for_contiguous_layoutunused-import
107files
20,006lines
166,226tokens
252clones
2,522duplicated lines (12.61%)
20,616duplicated tokens (12.4%)
By format
| Format | Files | Lines | Clones | Duplicated lines | Duplication |
|---|---|---|---|---|---|
| cpp-header | 48 | 10,079 | 230 | 2,311 | 22.93% |
| python | 36 | 7,760 | 21 | 195 | 2.51% |
| yaml | 3 | 375 | 1 | 16 | 4.27% |
| bash | 12 | 821 | 0 | 0 | 0% |
| markdown | 2 | 684 | 0 | 0 | 0% |
| perl | 1 | 168 | 0 | 0 | 0% |
| cpp | 2 | 54 | 0 | 0 | 0% |
| txt | 1 | 34 | 0 | 0 | 0% |
| text | 2 | 31 | 0 | 0 | 0% |
Largest clones
- 31 lines · 327 tokens cpp-headercsrc/jit_kernels/impls/sm100_tf32_hc_prenorm_gemm.hpp:28–58⇆csrc/jit_kernels/impls/sm90_tf32_hc_prenorm_gemm.hpp:30–60
- 14 lines · 204 tokens cpp-headercsrc/jit_kernels/impls/sm100_fp8_fp4_gemm_1d1d.hpp:370–383⇆csrc/jit_kernels/impls/sm100_fp8_fp4_gemm_1d1d.hpp:528–541
- 19 lines · 178 tokens cpp-headercsrc/jit_kernels/impls/sm100_bf16_gemm.hpp:297–315⇆csrc/jit_kernels/impls/sm100_fp8_fp4_gemm_1d1d.hpp:357–375
- 10 lines · 178 tokens pythondeep_gemm/legacy/a_fused_k_grouped_gemm.py:69–78⇆deep_gemm/legacy/b_fused_k_grouped_gemm.py:68–77
- 21 lines · 161 tokens pythondeep_gemm/legacy/a_fused_k_grouped_gemm.py:12–32⇆deep_gemm/legacy/b_fused_k_grouped_gemm.py:12–32
- 17 lines · 154 tokens cpp-headercsrc/jit_kernels/impls/sm100_bf16_gemm.hpp:92–108⇆csrc/jit_kernels/impls/sm100_fp8_fp4_gemm_1d1d.hpp:112–128
- 17 lines · 150 tokens cpp-headercsrc/jit_kernels/impls/sm100_bf16_gemm.hpp:104–120⇆csrc/jit_kernels/impls/sm90_bf16_gemm.hpp:99–115
- 17 lines · 150 tokens cpp-headercsrc/jit_kernels/impls/sm100_bf16_gemm.hpp:234–250⇆csrc/jit_kernels/impls/sm90_bf16_gemm.hpp:231–247
- 12 lines · 146 tokens cpp-headercsrc/apis/gemm.hpp:135–146⇆csrc/apis/gemm.hpp:150–161
- 12 lines · 146 tokens cpp-headercsrc/apis/gemm.hpp:135–146⇆csrc/apis/gemm.hpp:166–177
Trending appearances
| Day | Rank | Stars | Files | Lines | Clones | Duplication | Health | Commit |
|---|---|---|---|---|---|---|---|---|
| Oct 7, 2026 | #6 | 8,776 (+199) | 107 | 20,006 | 252 | 12.61% | D 51.6 | 057ca59 |
Want this for your own project? Install jscpd and run jscpd .