Code

Pushing FLOPS in Assembly 🏋️‍♂️#21

Code

Merged

ashvardanian merged 21 commits intomainashvardanian/less_slow.cpp:mainfrom asm-matmulashvardanian/less_slow.cpp:asm-matmulCopy head branch name to clipboardJan 20, 2025

+969-16

Commits

Commits on Jan 10, 2025

• Add: Assembly micro-kernels

  ashvardanian

  committed
  2e71e76

  Copy full SHA for 2e71e76

• Add: vfmadd231ps kernels

  ashvardanian

  committed
  7ca3161

  Copy full SHA for 7ca3161

• Add: Arm NEON FMAs

  ashvardanian

  committed
  d0e521e

  Copy full SHA for d0e521e

Commits on Jan 12, 2025

• Add: i8, f16, and bf16 kernels

  Show description for 3f54200
  ashvardanian

  committed
  3f54200

  Copy full SHA for 3f54200

• Merge branch 'main' into asm-matmul

  ashvardanian

  authored
  a268096

  Copy full SHA for a268096

• Add: AVX2 & low-precision AVX-512 TOPS

  ashvardanian

  committed
  0a48108

  Copy full SHA for 0a48108

• Merge branch 'asm-matmul' of https://github.com/ashvardanian/BenchmarkingTutorial into asm-matmul

  ashvardanian

  committed
  6d58708

  Copy full SHA for 6d58708

• Improve: Shorter TOPS names

  ashvardanian

  committed
  be0c94b

  Copy full SHA for be0c94b

• Fix: Missing tops_f64_neon

  ashvardanian

  committed
  4afa7e3

  Copy full SHA for 4afa7e3

• Fix: Missing tops_u8_neon

  ashvardanian

  committed
  d97bbfc

  Copy full SHA for d97bbfc

Commits on Jan 15, 2025

• Add: BLAS & Eigen TOPs benchmarks

  ashvardanian

  committed
  28ca39b

  Copy full SHA for 28ca39b

• Docs: Recommended Eigen macros

  ashvardanian

  committed
  7be2d58

  Copy full SHA for 7be2d58

Commits on Jan 16, 2025

• Fix: Duplicate .global symbols

  ashvardanian

  committed
  c732234

  Copy full SHA for c732234

• Merge branch 'asm-matmul' of https://github.com/ashvardanian/BenchmarkingTutorial into asm-matmul

  ashvardanian

  committed
  1e6a5fb

  Copy full SHA for 1e6a5fb

• Add: Inline Assembly kernels

  ashvardanian

  committed
  89095a6

  Copy full SHA for 89095a6

Commits on Jan 17, 2025

• Add: AMX kernels

  ashvardanian

  committed
  0cb024d

  Copy full SHA for 0cb024d

• Fix: Illegal instruction for AMX

  ashvardanian

  committed
  a7243dd

  Copy full SHA for a7243dd

• Docs: H100 Tensor Cores vs Intel

  ashvardanian

  committed
  fa86663

  Copy full SHA for fa86663

Commits on Jan 18, 2025

• Add: Latency Hiding & Port Interleaving

  ashvardanian

  committed
  086f8d7

  Copy full SHA for 086f8d7

• Merge branch 'main' into asm-matmul

  ashvardanian

  authored
  dc19a6a

  Copy full SHA for dc19a6a

• Docs: Zen4 matmul-benchmarks

  Show description for 2476310
  ashvardanian

  committed
  2476310

  Copy full SHA for 2476310