⚡️⚡️toy-hgemm library

What's Changed

• [HGEMM] Update RTX 3080 Laptop perf by @DefTruth in #148
• [HGEMM] Update toy-hgemm library 0.1.0 by @DefTruth in #149
• [HGEMM] Update toy-hgemm library 0.1.0 by @DefTruth in #150
• [HGEMM] Update toy-hgemm library 0.1.0 by @DefTruth in #152

Full Changelog: v2.6.4...v2.6.5

toy-hgemm library

What's Changed

• [HGEMM] Release toy-hgemm library 0.1.0 by @DefTruth in #146

Full Changelog: v2.6.3...v2.6.4

toy-hgemm library

What's Changed

• [HGEMM] Release toy-hgemm library 0.1.0 by @DefTruth in #145

Full Changelog: v2.6.2...v2.6.3

CuTe HGEMM Block Swizzle

What's Changed

• [HGEMM] trans mat b from row major -> col major by @DefTruth in #135
• [HGEMM] refactor HGEMM cpp benchmark by @DefTruth in #136
• [HGEMM] Update HGEMM L20/4090 Bench by @DefTruth in #137
• [HGEMM] fix cublas hgemm handle error by @DefTruth in #138
• [HGEMM] Add MMA HGEMM NN C++ benchmark by @DefTruth in #139
• [HGEMM] CuTe HGEMM with Thread Block Swizzle by @DefTruth in #140
• [HGEMM] clear tensor cache avoid OOM by @DefTruth in #141
• [HGEMM] Add gc.collect to HGEMM bench script by @DefTruth in #142
• [HGEMM] Add show_memory option to bench by @DefTruth in #143
• [HGEMM] manually init/destroy cublas handle by @DefTruth in #144

Full Changelog: v2.6.1...v2.6.2

v2.6.1 CuTe HGEMM

What's Changed

• [HGEMM] Add large MNK block swizzle policy by @DefTruth in #132
• [HGEMM] Add CuTe HGEMM with SMEM Swizzle by @DefTruth in #134
• Update embedding.cu by @TheManWhoIsStupid in #133

New Contributors

• @TheManWhoIsStupid made their first contribution in #133

Full Changelog: v2.6...v2.6.1

v2.6 Refactor 7/N

What's Changed

• [HGEMM] Update NVIDIA L20/4090 Perf plots by @DefTruth in #126
• [Blog]图解DeepSpeed-Ulysses&Megatron-LM TP/SP by @DefTruth in #127
• [README] Add contents lists by @DefTruth in #128
• [README] Update README by @DefTruth in #129
• [README] Update README.md by @DefTruth in #130
• Bump up to v2.6 by @DefTruth in #131

Full Changelog: v2.5...v2.6

v2.5

What's Changed

• [HGEMM] Update HGEMM README.md by @DefTruth in #120
• [HGEMM] Add plot tflops function by @DefTruth in #121
• [HGEMM] Add NVIDIA RTX 3090 Laptop perf plot by @DefTruth in #122
• [PERF] Update HGEMM benchmark scripts by @DefTruth in #123
• [HGEMM] Add HGEMM L20/4090 benchmark figures by @DefTruth in #124
• Bump up to v2.5 by @DefTruth in #125

Full Changelog: v2.4.18...v2.5

v2.4.18

What's Changed

• Update README.md by @DefTruth in #115
• [HGEMM] Update HGEMM Supported Matrix by @DefTruth in #116
• [HGEMM] Update HGEMM/SGEMM Supported Matrix by @DefTruth in #117
• [README] Update HGEMM/SGEMM Supported Matrix by @DefTruth in #118
• [HGEMM] Add NVIDIA RTX 4090 benchmark by @DefTruth in #119

Full Changelog: v2.4.17...v2.4.18

v2.4.17

What's Changed

• [NMS] Add nms f32 cuda kernel. by @bear-zd in #102
• [HGEMM] Add some note to collective store by @DefTruth in #103
• [HGEMM] Add HGEMM MMA Col Major Kernel by @DefTruth in #104
• [HGEMM] Update HGEMM benchmark scripts by @DefTruth in #105
• [HGEMM] Add Warp Swizzle as template param by @DefTruth in #106
• [HGEMM] add -Xptxas -v compile flag by @DefTruth in #107
• [HGEMM] Try reduce registers usage by @DefTruth in #108
• [HGEMM] Update HGEMM MMA/WMMA Usage by @DefTruth in #109
• [HGEMM][Docs] Add HGEMM Supported Matrix by @DefTruth in #110
• [HGEMM] Add M=N=K option for benchmark by @DefTruth in #111
• [HGEMM] Update HGEMM/SGEMM Supported Matrix by @DefTruth in #112
• [README] Update HGEMM/SGEMM Supported matrix by @DefTruth in #113
• [Docs] Update HGEMM/SGEMM Supported Matrix by @DefTruth in #114

Full Changelog: v2.4.16...v2.4.17

HGEMM Warp Swizzle/Reg Buffers

What's Changed

• [HGEMM] HGEMM MMA with Reg Double Buffers by @DefTruth in #99
• [HGEMM] ldmatrix.x4.trans with reg double buffers by @DefTruth in #100
• [HGEMM] collective store via warp shfl&reg reuse by @DefTruth in #101

Full Changelog: v2.4.15...v2.4.16