Age | Commit message (Collapse) | Author | |
---|---|---|---|
2020-10-13 | Replace QS8 4x8 with 2x8 neon microkernel. | Frank Barchard | |
Improves performance for aarch32. PiperOrigin-RevId: 336945809 | |||
2020-10-12 | Cortex A55r1 QS8 GEMM microkernel | Frank Barchard | |
PiperOrigin-RevId: 336803747 | |||
2020-10-12 | Add RUY benchmark to qs8_gemm_bench | Frank Barchard | |
PiperOrigin-RevId: 336711804 | |||
2020-10-10 | Rename QS8 assembly GEMM kernels to ld64 | Frank Barchard | |
PiperOrigin-RevId: 336494103 | |||
2020-10-06 | 1x16 QS8 GEMM AARCH64 assembly microkernel using dot product. | Frank Barchard | |
PiperOrigin-RevId: 335801231 | |||
2020-10-06 | 4x16 QS8 GEMM AARCH64 assembly microkernel using dot product. | Frank Barchard | |
PiperOrigin-RevId: 335584437 | |||
2020-09-24 | 4x8, 6x8 and 8x16 Neon dot product GEMM microkernels | Frank Barchard | |
PiperOrigin-RevId: 333462985 | |||
2020-09-22 | 6x16 QS8 GEMM for Neon dot product | Frank Barchard | |
PiperOrigin-RevId: 333099606 | |||
2020-08-13 | Add xnn_qs8_gemm_minmax_ukernel_${MR}x${NR}c4__neondot (ARMv8.2+dotprod). | Benoit Jacob | |
PiperOrigin-RevId: 326503942 | |||
2020-08-10 | AVX512 variants of QS8 GEMM and IGEMM microkernels | Marat Dukhan | |
PiperOrigin-RevId: 325850791 | |||
2020-08-05 | Benchmark ARM NEON versions of QS8 GEMM microkernels | Marat Dukhan | |
PiperOrigin-RevId: 325127096 | |||
2020-08-05 | WAsm SIMD versions of QS8 GEMM and IGEMM microkernels | Marat Dukhan | |
PiperOrigin-RevId: 325112592 | |||
2020-08-03 | Rename s8rng/s32rng -> i8rng/i32rng | Marat Dukhan | |
PiperOrigin-RevId: 324746732 | |||
2020-08-03 | XW (eXtended Weights) optimization for QS8 GEMM microkernel | Marat Dukhan | |
PiperOrigin-RevId: 324734460 | |||
2020-08-03 | Add 3x4c8 variants of SSE2/SSSE3/SSE4.1/XOP GEMM/IGEMM microkernels | Marat Dukhan | |
PiperOrigin-RevId: 324710790 | |||
2020-08-02 | AVX2 version of QS8 GEMM and IGEMM microkernels | Marat Dukhan | |
PiperOrigin-RevId: 324543693 | |||
2020-08-02 | XOP versions of QS8 GEMM/IGEMM microkernels | Marat Dukhan | |
PiperOrigin-RevId: 324541139 | |||
2020-08-02 | Bind RNG by reference in microbenchmarks | Marat Dukhan | |
PiperOrigin-RevId: 324540550 | |||
2020-07-31 | LD128 versions of QS8 GEMM SSE2/SSSE3/SSE4.1 microkernels | Marat Dukhan | |
PiperOrigin-RevId: 324316646 | |||
2020-07-31 | Add LD64 suffix in QS8 GEMM/IGEMM microkernels | Marat Dukhan | |
LD64 denotes that weights are loaded 64 bits at a time and sign-extended to 128 bits PiperOrigin-RevId: 324305250 | |||
2020-07-31 | QS8 GEMM MRx4c8 SSE2/SSSE3/SSE4.1 microkernels | Marat Dukhan | |
PiperOrigin-RevId: 324300862 | |||
2020-07-31 | QS8 GEMM microkernels and infrastructure | Marat Dukhan | |
- QS8 GEMM microkernels for SSE2/SSSE3/SSE4.1 - Updated unit test generator to support SSSE3 ISA - Updated GEMM tester to support QS8 GEMM - Updated weights packing functions to support QS8 GEMM - Microbenchmark for QS8 GEMM microkernels PiperOrigin-RevId: 324231357 |