• Docs: OpenBLAS installation on MacOS (be4a0be)
• Fix: Missing const qualifiers in strided_ptr (9120723)
• Fix: Can't std::format(time) on macOS (4d00aba)

Thanks to @ab-10 for spotting 🤗

Release: v0.10.7 [skip ci]

Patch

• Improve: Include Asm tests into macOS Arm builds (#45) (ecff6e3)

• Docs: Notes on #pragma regions (ab7bf3f)
• Fix: Aligned allocation (#42) (a66cfe2)

Thanks to @bmanga 🤗

Release: v0.10.5 [skip ci]

Patch

• Docs: Intro (60e18d8)

Release: v0.10.4 [skip ci]

Patch

• Improve: Detecting CUDA availability (21dfdf3)

Release: v0.10.3 [skip ci]

Patch

• Docs: Cleaner stance on std::sin approximation (d4cbe85)

• Improve: Horner method (cab8824)
• Make: Default to -O2 (56016d5)
• Fix: Compiling w/out Intel TBB (2346e03)
• Docs: Typo (#39) (99a91ba)
• Improve: Stricter range limits & fast-math (7ae2c01)
• Make: Formatting CMake (0e3c916)
• Improve: Detecting CUDA availability (91c5f4e)

Thanks to @corneliusroemer, @dzaima, @DrChr 🤗

Release: v0.10.1 [skip ci]

Patch

• Fix: Destroy CUDA events (c50e2e7)

DeepSeek has just released their mixed-precision FP8 GEMM implementation, and it felt like a good time to introduce some cuBLASLt snippets as a baseline for such work. On Nvidia H200, the results for different input sizes look like this:

--------------------------------------------------------------------------------------------------
Benchmark                                        Time             CPU   Iterations UserCounters...
--------------------------------------------------------------------------------------------------
cublaslt_tops<fp8_e4m3_t, float>/256         12496 ns        12496 ns        56284 TOP=2.67999T/s
cublaslt_tops<fp8_e4m3_t, float>/512         13089 ns        13089 ns        53100 TOP=20.4883T/s
cublaslt_tops<fp8_e4m3_t, float>/1024        14882 ns        14882 ns        46918 TOP=144.23T/s
cublaslt_tops<fp8_e4m3_t, float>/2048        25802 ns        25802 ns        26869 TOP=665.679T/s
cublaslt_tops<fp8_e4m3_t, float>/4096       109316 ns       109313 ns         6021 TOP=1.25715P/s
cublaslt_tops<fp8_e4m3_t, float>/8192       821080 ns       821050 ns          629 TOP=1.33907P/s
cublaslt_tops<fp8_e4m3_t, float>/16384     7135472 ns      7135461 ns           93 TOP=1.23269P/s
cublaslt_tops<fp8_e4m3_t, float>_BigO         0.00 N^3        0.00 N^3  
cublaslt_tops<fp8_e4m3_t, float>_RMS             2 %             2 % 

The advertised throughput for H100 and H200 in the SXM form factor is 2 Peta-Ops, and cuBLASLt achieves around 67% of that in the shared benchmarks. So, one should definitely be able to squeeze more.

I haven't tried implementing synthetic ALU benchmarks for different FP8-oriented PTX instructions, so if you have time and want to try something new - feel free to submit a PR 🤗

Release: v0.9.2 [skip ci]

Patch

• Docs: Counting PTX as Assembly lines (cb470dd)