Commit Graph

  • 7f7bc3a37f r4_nrcy_16: iq3_k_r4, iq4_k_r4, iq4_ks_r4, iq5_k_r4 ik/r4_nrcy_16 Iwan Kawrakow 2024-12-22 10:25:01 +02:00
  • 31ff6ec368 nrc_y = 16: AVX2 iq2_xxs, iq2_xs, iq3_xxs Iwan Kawrakow 2024-12-22 08:53:35 +02:00
  • 0406b39b9e nrc_y = 16: NEON iq2_xxs, iq2_xs, iq3_xxs Iwan Kawrakow 2024-12-21 19:04:35 +01:00
  • 4aabff46a2 nrc_y = 16: NEON iq2_s Iwan Kawrakow 2024-12-21 18:34:22 +01:00
  • e6a0358686 nrc_y = 16: AVX2 iq2_s Iwan Kawrakow 2024-12-21 19:26:03 +02:00
  • 5607bae2a6 Add nrc_y = 16 implementation. Iwan Kawrakow 2024-12-21 19:07:46 +02:00
  • 7598ec79a2 IQ2_S_R4 (#156) Kawrakow 2024-12-21 11:26:35 +01:00
  • 93419de68f IQ2_S_R4 (#156) Kawrakow 2024-12-21 11:26:35 +01:00
  • fe8eda7b47 iq2_s_r4: NEON ik/iq2_s_r4 Iwan Kawrakow 2024-12-21 11:13:58 +01:00
  • 74ee045b06 Minor Iwan Kawrakow 2024-12-21 11:24:02 +02:00
  • 964c5f9e82 iq2_s_r4: Zen4 Iwan Kawrakow 2024-12-21 11:10:13 +02:00
  • 6892554e43 IQ2_XS_R4 (#155) Kawrakow 2024-12-21 08:32:39 +01:00
  • a867b919ca IQ2_XS_R4 (#155) Kawrakow 2024-12-21 08:32:39 +01:00
  • 6dddd0832a iq2_xs_r4: slightly better NEON ik/iq2_xs_r4 Iwan Kawrakow 2024-12-21 08:15:35 +01:00
  • 2f2c94de78 iq2_xs_r4: NEON - not much better than iq2_xs Iwan Kawrakow 2024-12-20 19:00:39 +01:00
  • ce2d053286 iq2_xs_r4: slightly better matrix x vector on AVX2 Iwan Kawrakow 2024-12-20 18:33:02 +02:00
  • d8a274a18f iq2_xs_r4: AVX2 Iwan Kawrakow 2024-12-20 18:06:02 +02:00
  • 51e5a87382 iq2_xs_r4: Zen4 Iwan Kawrakow 2024-12-20 17:44:54 +02:00
  • 9dfd69bd93 IQ2_XXS_R4 (#154) Kawrakow 2024-12-20 12:02:42 +01:00
  • 4b53bc876e IQ2_XXS_R4 (#154) Kawrakow 2024-12-20 12:02:42 +01:00
  • 9904f8f691 fix typo (#151) Nexes the Elder 2024-12-20 12:02:15 +01:00
  • 7254f6d340 fix typo (#151) Nexes the Elder 2024-12-20 12:02:15 +01:00
  • c3e9a0e26f iq2_xxs_r4: NEON ik/iq2_xxs_r4 Iwan Kawrakow 2024-12-20 11:46:37 +01:00
  • 6c0e88ef51 Minor Iwan Kawrakow 2024-12-20 12:11:40 +02:00
  • 6862db04b0 iq2_xxs_r4: Zen4 Iwan Kawrakow 2024-12-20 11:55:49 +02:00
  • 310f8b1d22 IQ3_XXS_R4 (#153) Kawrakow 2024-12-20 09:12:48 +01:00
  • f4170f78bd IQ3_XXS_R4 (#153) Kawrakow 2024-12-20 09:12:48 +01:00
  • 3681093127 iq3_xxs_r4: NEON ik/iq3_xxs_r4_v2 Iwan Kawrakow 2024-12-19 17:47:39 +01:00
  • 5243b4d847 iq4_xxs_r4: 1st shot at AVX2 Iwan Kawrakow 2024-12-19 14:52:46 +02:00
  • 327b3abb38 iq4_xxs_r4: WIP Iwan Kawrakow 2024-12-19 14:15:12 +02:00
  • ef0eeacadb iq3_xxs_r4: 1st shot on Zen4 Iwan Kawrakow 2024-12-19 12:19:34 +02:00
  • 1875b1d8e6 iq3_xxs_r4: NEON ik/iq3_xxs_r4 Iwan Kawrakow 2024-12-19 17:47:39 +01:00
  • b33e128d33 iq4_xxs_r4: 1st shot at AVX2 Iwan Kawrakow 2024-12-19 14:52:46 +02:00
  • e68c9f4e69 iq4_xxs_r4: WIP Iwan Kawrakow 2024-12-19 14:15:12 +02:00
  • 19acd7ce1e iq3_xxs_r4: 1st shot on Zen4 Iwan Kawrakow 2024-12-19 12:19:34 +02:00
  • 8352f12275 IQ4_KS_R4 (#150) Kawrakow 2024-12-18 19:58:21 +01:00
  • dfa12b7f91 IQ4_KS_R4 (#150) Kawrakow 2024-12-18 19:58:21 +01:00
  • baa9ed4a5e Minor ik/iq4_ks_r4 Iwan Kawrakow 2024-12-18 19:14:40 +01:00
  • 918ac19eed iq4_ks_r4: NEON Iwan Kawrakow 2024-12-18 19:02:49 +01:00
  • 6b1a1b9880 iq4_ks_r4: slightly better Zen4 Iwan Kawrakow 2024-12-18 19:03:16 +02:00
  • 00452d4683 iq4_ks_r4: slightly better Zen4 Iwan Kawrakow 2024-12-18 18:45:08 +02:00
  • e232c1b785 iq4_ks_r4: WIP Iwan Kawrakow 2024-12-18 18:09:56 +02:00
  • e7bafd469e iq4_ks_r4: AVX2 Iwan Kawrakow 2024-12-18 17:44:27 +02:00
  • 16af231c12 iq4_ks_r4: Zen4 Iwan Kawrakow 2024-12-18 16:40:19 +02:00
  • c208fec2f2 IQ5_K_R4 (#149) Kawrakow 2024-12-18 13:29:25 +01:00
  • 59d742b00f IQ5_K_R4 (#149) Kawrakow 2024-12-18 13:29:25 +01:00
  • 6c935be1c0 iq5_k_r4: NEON ik/iq5_k_r4 Iwan Kawrakow 2024-12-18 13:21:30 +01:00
  • f9d182ff44 iq5_k_r4: slightly better AVX2 Iwan Kawrakow 2024-12-18 11:54:01 +02:00
  • ece6d0a52b iq5_k_r4: better Zen4 Iwan Kawrakow 2024-12-18 11:35:40 +02:00
  • a8b0f8148c iq5_k_r4: fix AVX2 nrc_y = 1 case Iwan Kawrakow 2024-12-18 11:14:45 +02:00
  • f1b37f6bc0 Minor Iwan Kawrakow 2024-12-18 10:46:32 +02:00
  • 27e987d959 iq5_k_r5: WIP Iwan Kawrakow 2024-12-18 10:26:47 +02:00
  • 5eac4edc90 iq5_k_r4: Zen4 Iwan Kawrakow 2024-12-18 09:17:30 +02:00
  • 2247afa967 Slightly better matrix x vector on Zen4/AVX2 for iq2_k_r4, iq3_k_r4, iq4_k_r4 (#148) Kawrakow 2024-12-17 18:55:38 +01:00
  • 9b6d14a299 Slightly better matrix x vector on Zen4/AVX2 for iq2_k_r4, iq3_k_r4, iq4_k_r4 (#148) Kawrakow 2024-12-17 18:55:38 +01:00
  • c367b15c5f Minor ik/avx2_r4_tweaks Iwan Kawrakow 2024-12-17 19:53:44 +02:00
  • 2190af4dde Slightly better matrix x vector on Zen4/AVX2 for iq2_k_r4, iq3_k_r4, iq4_k_r4 Iwan Kawrakow 2024-12-17 19:07:01 +02:00
  • a648191c2c Be able to repack tensors at run time (#147) Kawrakow 2024-12-17 14:16:34 +01:00
  • 514ae08620 Be able to repack tensors at run time (#147) Kawrakow 2024-12-17 14:16:34 +01:00
  • e716ae7778 Repack: make sure number of rows is a multiple of the packing ik/run_time_repack Iwan Kawrakow 2024-12-17 15:14:36 +02:00
  • b0e826d5b4 Repack: also add bf16 as repackable type Iwan Kawrakow 2024-12-17 15:12:27 +02:00
  • 5008d055c0 Be able to repack tensors at run time Iwan Kawrakow 2024-12-17 15:00:13 +02:00
  • c16d352915 IQ2_K_R4 (#146) Kawrakow 2024-12-17 10:18:33 +01:00
  • 4ade4c568c IQ2_K_R4 (#146) Kawrakow 2024-12-17 10:18:33 +01:00
  • a75a790e8f iq2_k_r4: better matrix x vector multiplication on NEON ik/iq2_k_r4 Iwan Kawrakow 2024-12-17 10:06:02 +01:00
  • bb4cc44472 iq2_k_r4: NEON Iwan Kawrakow 2024-12-17 09:37:41 +01:00
  • d5d94bbe66 iq2_k_r4: Zen4 Iwan Kawrakow 2024-12-17 09:54:22 +02:00
  • b52e2e2934 IQ3_K_R4 (#145) Kawrakow 2024-12-17 07:51:11 +01:00
  • d69344f8ea IQ3_K_R4 (#145) Kawrakow 2024-12-17 07:51:11 +01:00
  • 2243a998d3 iq3_k_r4: faster matrix x vector multiplication on NEON ik/iq3_k_r4_v2 Iwan Kawrakow 2024-12-17 06:42:30 +01:00
  • 370628b7b4 iq3_k_r4: NEON Iwan Kawrakow 2024-12-17 05:53:53 +01:00
  • 751fbaaa8c iq3_k_r4: AVX2 Iwan Kawrakow 2024-12-16 19:51:51 +02:00
  • d73efc787f iq3_k_r4: Zen4 Iwan Kawrakow 2024-12-16 19:17:05 +02:00
  • 72330eba1d iq3_k_r4 WIP Iwan Kawrakow 2024-12-16 18:20:11 +02:00
  • 1095cd27a9 Slightly faster IQ4_K_R4 on AVX2/Zen4 (#144) Kawrakow 2024-12-16 16:09:15 +01:00
  • 1714e46f13 Slightly faster IQ4_K_R4 on AVX2/Zen4 (#144) Kawrakow 2024-12-16 16:09:15 +01:00
  • 5415228878 iq4_k_r4: slightly better Zen4 ik/iq4_k_r4_avx2 Iwan Kawrakow 2024-12-16 17:03:07 +02:00
  • 8bfd5cff31 iq4_k_r4: slightly better AVX2 Iwan Kawrakow 2024-12-16 16:38:14 +02:00
  • 93cfa3e662 Slightly faster IQ4_XS_R4 on AVX2 (#143) Kawrakow 2024-12-16 14:40:10 +01:00
  • 84ab873385 Slightly faster IQ4_XS_R4 on AVX2 (#143) Kawrakow 2024-12-16 14:40:10 +01:00
  • f8a6c31617 Delete unused stuff ik/iq4_xs_r4_avx2 Iwan Kawrakow 2024-12-16 15:39:12 +02:00
  • 9ffada4231 Minor Iwan Kawrakow 2024-12-16 15:32:34 +02:00
  • b3d52ee884 iq4_xs_r4: slightly faster and correct AVX2 implementation Iwan Kawrakow 2024-12-16 14:32:47 +02:00
  • 3283fbeb82 q8_k_r8: this change for NEON got lost? Kawrakow 2024-12-16 13:31:21 +01:00
  • 987a43c531 q8_k_r8: this change for NEON got lost? Iwan Kawrakow 2024-12-16 13:31:21 +01:00
  • dd84b1e462 iq4_xs_r8: better AVX2 ik/iq4_xs_r8 Iwan Kawrakow 2024-12-16 09:54:23 +02:00
  • 7481ec634f iq4_xs_r4 -> iq4_xs_r8 (AVX2) Iwan Kawrakow 2024-12-16 09:34:46 +02:00
  • a8dcde8156 iq4_xs_r4 -> iq4_xs_r8 (Zen4) Iwan Kawrakow 2024-12-16 09:14:23 +02:00
  • e811de75e9 BF16_R16 - 16 interleaved bf16 rows (#142) Kawrakow 2024-12-15 09:54:21 +01:00
  • 85c5a1a995 BF16_R16 - 16 interleaved bf16 rows (#142) Kawrakow 2024-12-15 09:54:21 +01:00
  • 6058dd735a Cleanup unused stuff ik/bf16_r4 Iwan Kawrakow 2024-12-14 19:30:28 +02:00
  • 042ea88616 Rename bf16_r4 to bf16_r16 Iwan Kawrakow 2024-12-14 19:06:19 +02:00
  • 99b580d73c bf16_rx: Very slightly faster by interleaving 16 rows Iwan Kawrakow 2024-12-14 18:58:10 +02:00
  • 4b9350b353 Adding bf16_r8 Iwan Kawrakow 2024-12-14 17:28:29 +02:00
  • 69689d60e5 Not working bf16_r4 Iwan Kawrakow 2024-12-05 06:53:41 +02:00
  • e885c1e59b Q8_K_R8: Fastest quantized matrix multiplications (#141) Kawrakow 2024-12-14 09:24:30 +01:00
  • 20758edcae Q8_K_R8: Fastest quantized matrix multiplications (#141) Kawrakow 2024-12-14 09:24:30 +01:00
  • e65f7e8c8b Shutup useless compiler warnings ik/q8_k_r8 Iwan Kawrakow 2024-12-14 08:56:48 +02:00
  • 48c73c7b0f q8_k_r4: go to signed ints Iwan Kawrakow 2024-12-14 07:31:52 +02:00
  • 89510ccce7 q8_k_r4: NEON Iwan Kawrakow 2024-12-13 18:57:07 +01:00