whispercpp 1.3.5 → 1.3.6

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (610) hide show
  1. checksums.yaml +4 -4
  2. data/LICENSE +1 -1
  3. data/README.md +99 -2
  4. data/ext/extconf.rb +1 -0
  5. data/ext/ruby_whisper.c +20 -4
  6. data/ext/ruby_whisper.h +30 -2
  7. data/ext/ruby_whisper_context.c +216 -124
  8. data/ext/ruby_whisper_context_params.c +163 -0
  9. data/ext/ruby_whisper_model.c +0 -1
  10. data/ext/ruby_whisper_params.c +0 -1
  11. data/ext/ruby_whisper_segment.c +0 -1
  12. data/ext/ruby_whisper_token.c +29 -9
  13. data/ext/ruby_whisper_transcribe.cpp +4 -1
  14. data/ext/ruby_whisper_vad_context.c +48 -1
  15. data/ext/ruby_whisper_vad_context_detect.cpp +6 -5
  16. data/ext/ruby_whisper_vad_params.c +0 -1
  17. data/ext/ruby_whisper_vad_segment.c +0 -1
  18. data/ext/ruby_whisper_vad_segments.c +0 -1
  19. data/ext/sources/CMakeLists.txt +1 -1
  20. data/ext/sources/bindings/javascript/package.json +1 -1
  21. data/ext/sources/cmake/whisper-config.cmake.in +5 -40
  22. data/ext/sources/examples/bench/bench.cpp +23 -18
  23. data/ext/sources/examples/cli/cli.cpp +8 -0
  24. data/ext/sources/examples/common-ggml.cpp +2 -0
  25. data/ext/sources/examples/miniaudio.h +4507 -2131
  26. data/ext/sources/examples/server/server.cpp +18 -4
  27. data/ext/sources/examples/talk-llama/CMakeLists.txt +3 -2
  28. data/ext/sources/examples/talk-llama/llama-adapter.cpp +7 -13
  29. data/ext/sources/examples/talk-llama/llama-adapter.h +4 -3
  30. data/ext/sources/examples/talk-llama/llama-arch.cpp +335 -17
  31. data/ext/sources/examples/talk-llama/llama-arch.h +42 -0
  32. data/ext/sources/examples/talk-llama/llama-batch.cpp +3 -1
  33. data/ext/sources/examples/talk-llama/llama-chat.cpp +21 -1
  34. data/ext/sources/examples/talk-llama/llama-chat.h +1 -0
  35. data/ext/sources/examples/talk-llama/llama-context.cpp +508 -520
  36. data/ext/sources/examples/talk-llama/llama-context.h +27 -28
  37. data/ext/sources/examples/talk-llama/llama-cparams.h +5 -0
  38. data/ext/sources/examples/talk-llama/llama-ext.h +12 -0
  39. data/ext/sources/examples/talk-llama/llama-grammar.cpp +8 -8
  40. data/ext/sources/examples/talk-llama/llama-graph.cpp +583 -130
  41. data/ext/sources/examples/talk-llama/llama-graph.h +131 -10
  42. data/ext/sources/examples/talk-llama/llama-hparams.cpp +57 -40
  43. data/ext/sources/examples/talk-llama/llama-hparams.h +79 -10
  44. data/ext/sources/examples/talk-llama/llama-impl.cpp +4 -4
  45. data/ext/sources/examples/talk-llama/llama-impl.h +13 -1
  46. data/ext/sources/examples/talk-llama/llama-kv-cache-iswa.cpp +3 -1
  47. data/ext/sources/examples/talk-llama/llama-kv-cache.cpp +274 -89
  48. data/ext/sources/examples/talk-llama/llama-kv-cache.h +2 -3
  49. data/ext/sources/examples/talk-llama/llama-memory-hybrid-iswa.cpp +275 -0
  50. data/ext/sources/examples/talk-llama/llama-memory-hybrid-iswa.h +140 -0
  51. data/ext/sources/examples/talk-llama/llama-memory-recurrent.cpp +11 -13
  52. data/ext/sources/examples/talk-llama/llama-mmap.cpp +28 -11
  53. data/ext/sources/examples/talk-llama/llama-model-loader.cpp +527 -119
  54. data/ext/sources/examples/talk-llama/llama-model-loader.h +35 -5
  55. data/ext/sources/examples/talk-llama/llama-model-saver.cpp +60 -46
  56. data/ext/sources/examples/talk-llama/llama-model-saver.h +5 -2
  57. data/ext/sources/examples/talk-llama/llama-model.cpp +1365 -647
  58. data/ext/sources/examples/talk-llama/llama-model.h +72 -19
  59. data/ext/sources/examples/talk-llama/llama-quant.cpp +578 -346
  60. data/ext/sources/examples/talk-llama/{llama-sampling.cpp → llama-sampler.cpp} +190 -76
  61. data/ext/sources/examples/talk-llama/{llama-sampling.h → llama-sampler.h} +0 -2
  62. data/ext/sources/examples/talk-llama/llama-vocab.cpp +118 -48
  63. data/ext/sources/examples/talk-llama/llama-vocab.h +5 -0
  64. data/ext/sources/examples/talk-llama/llama.cpp +76 -22
  65. data/ext/sources/examples/talk-llama/llama.h +63 -30
  66. data/ext/sources/examples/talk-llama/models/afmoe.cpp +2 -3
  67. data/ext/sources/examples/talk-llama/models/apertus.cpp +3 -3
  68. data/ext/sources/examples/talk-llama/models/arcee.cpp +3 -3
  69. data/ext/sources/examples/talk-llama/models/arctic.cpp +4 -5
  70. data/ext/sources/examples/talk-llama/models/baichuan.cpp +4 -3
  71. data/ext/sources/examples/talk-llama/models/bailingmoe.cpp +1 -2
  72. data/ext/sources/examples/talk-llama/models/bailingmoe2.cpp +3 -5
  73. data/ext/sources/examples/talk-llama/models/bert.cpp +13 -7
  74. data/ext/sources/examples/talk-llama/models/bitnet.cpp +9 -24
  75. data/ext/sources/examples/talk-llama/models/bloom.cpp +2 -2
  76. data/ext/sources/examples/talk-llama/models/chameleon.cpp +3 -3
  77. data/ext/sources/examples/talk-llama/models/chatglm.cpp +2 -2
  78. data/ext/sources/examples/talk-llama/models/codeshell.cpp +3 -3
  79. data/ext/sources/examples/talk-llama/models/cogvlm.cpp +3 -3
  80. data/ext/sources/examples/talk-llama/models/cohere2-iswa.cpp +2 -2
  81. data/ext/sources/examples/talk-llama/models/command-r.cpp +2 -2
  82. data/ext/sources/examples/talk-llama/models/dbrx.cpp +4 -5
  83. data/ext/sources/examples/talk-llama/models/deci.cpp +3 -3
  84. data/ext/sources/examples/talk-llama/models/deepseek.cpp +4 -6
  85. data/ext/sources/examples/talk-llama/models/deepseek2.cpp +24 -21
  86. data/ext/sources/examples/talk-llama/models/delta-net-base.cpp +445 -0
  87. data/ext/sources/examples/talk-llama/models/dots1.cpp +4 -6
  88. data/ext/sources/examples/talk-llama/models/dream.cpp +3 -3
  89. data/ext/sources/examples/talk-llama/models/ernie4-5-moe.cpp +4 -6
  90. data/ext/sources/examples/talk-llama/models/ernie4-5.cpp +3 -3
  91. data/ext/sources/examples/talk-llama/models/eurobert.cpp +97 -0
  92. data/ext/sources/examples/talk-llama/models/exaone-moe.cpp +145 -0
  93. data/ext/sources/examples/talk-llama/models/exaone.cpp +3 -3
  94. data/ext/sources/examples/talk-llama/models/exaone4.cpp +3 -3
  95. data/ext/sources/examples/talk-llama/models/falcon-h1.cpp +2 -4
  96. data/ext/sources/examples/talk-llama/models/falcon.cpp +3 -3
  97. data/ext/sources/examples/talk-llama/models/gemma-embedding.cpp +1 -1
  98. data/ext/sources/examples/talk-llama/models/gemma.cpp +1 -1
  99. data/ext/sources/examples/talk-llama/models/gemma2-iswa.cpp +1 -1
  100. data/ext/sources/examples/talk-llama/models/gemma3.cpp +1 -1
  101. data/ext/sources/examples/talk-llama/models/gemma3n-iswa.cpp +7 -7
  102. data/ext/sources/examples/talk-llama/models/glm4-moe.cpp +3 -3
  103. data/ext/sources/examples/talk-llama/models/glm4.cpp +14 -7
  104. data/ext/sources/examples/talk-llama/models/gpt2.cpp +2 -2
  105. data/ext/sources/examples/talk-llama/models/gptneox.cpp +2 -2
  106. data/ext/sources/examples/talk-llama/models/granite-hybrid.cpp +4 -5
  107. data/ext/sources/examples/talk-llama/models/granite.cpp +4 -5
  108. data/ext/sources/examples/talk-llama/models/grok.cpp +4 -4
  109. data/ext/sources/examples/talk-llama/models/grovemoe.cpp +5 -7
  110. data/ext/sources/examples/talk-llama/models/hunyuan-dense.cpp +3 -3
  111. data/ext/sources/examples/talk-llama/models/hunyuan-moe.cpp +4 -5
  112. data/ext/sources/examples/talk-llama/models/internlm2.cpp +3 -3
  113. data/ext/sources/examples/talk-llama/models/jais.cpp +2 -2
  114. data/ext/sources/examples/talk-llama/models/jais2.cpp +123 -0
  115. data/ext/sources/examples/talk-llama/models/jamba.cpp +3 -3
  116. data/ext/sources/examples/talk-llama/models/kimi-linear.cpp +381 -0
  117. data/ext/sources/examples/talk-llama/models/lfm2.cpp +145 -124
  118. data/ext/sources/examples/talk-llama/models/llada-moe.cpp +4 -4
  119. data/ext/sources/examples/talk-llama/models/llada.cpp +3 -3
  120. data/ext/sources/examples/talk-llama/models/llama-iswa.cpp +4 -4
  121. data/ext/sources/examples/talk-llama/models/llama.cpp +18 -11
  122. data/ext/sources/examples/talk-llama/models/maincoder.cpp +3 -3
  123. data/ext/sources/examples/talk-llama/models/{graph-context-mamba.cpp → mamba-base.cpp} +9 -3
  124. data/ext/sources/examples/talk-llama/models/mamba.cpp +1 -2
  125. data/ext/sources/examples/talk-llama/models/mimo2-iswa.cpp +11 -5
  126. data/ext/sources/examples/talk-llama/models/minicpm3.cpp +14 -13
  127. data/ext/sources/examples/talk-llama/models/minimax-m2.cpp +4 -5
  128. data/ext/sources/examples/talk-llama/models/mistral3.cpp +4 -4
  129. data/ext/sources/examples/talk-llama/models/models.h +181 -46
  130. data/ext/sources/examples/talk-llama/models/modern-bert.cpp +2 -9
  131. data/ext/sources/examples/talk-llama/models/mpt.cpp +2 -2
  132. data/ext/sources/examples/talk-llama/models/nemotron-h.cpp +26 -14
  133. data/ext/sources/examples/talk-llama/models/nemotron.cpp +3 -3
  134. data/ext/sources/examples/talk-llama/models/neo-bert.cpp +2 -2
  135. data/ext/sources/examples/talk-llama/models/olmo.cpp +3 -3
  136. data/ext/sources/examples/talk-llama/models/olmo2.cpp +3 -3
  137. data/ext/sources/examples/talk-llama/models/olmoe.cpp +4 -4
  138. data/ext/sources/examples/talk-llama/models/openai-moe-iswa.cpp +1 -1
  139. data/ext/sources/examples/talk-llama/models/openelm.cpp +3 -3
  140. data/ext/sources/examples/talk-llama/models/orion.cpp +3 -3
  141. data/ext/sources/examples/talk-llama/models/paddleocr.cpp +122 -0
  142. data/ext/sources/examples/talk-llama/models/pangu-embedded.cpp +3 -3
  143. data/ext/sources/examples/talk-llama/models/phi2.cpp +2 -2
  144. data/ext/sources/examples/talk-llama/models/phi3.cpp +3 -3
  145. data/ext/sources/examples/talk-llama/models/plamo.cpp +3 -3
  146. data/ext/sources/examples/talk-llama/models/plamo2.cpp +9 -5
  147. data/ext/sources/examples/talk-llama/models/plamo3.cpp +2 -2
  148. data/ext/sources/examples/talk-llama/models/plm.cpp +15 -14
  149. data/ext/sources/examples/talk-llama/models/qwen.cpp +2 -2
  150. data/ext/sources/examples/talk-llama/models/qwen2.cpp +3 -3
  151. data/ext/sources/examples/talk-llama/models/qwen2moe.cpp +4 -4
  152. data/ext/sources/examples/talk-llama/models/qwen2vl.cpp +3 -3
  153. data/ext/sources/examples/talk-llama/models/qwen3.cpp +12 -9
  154. data/ext/sources/examples/talk-llama/models/qwen35.cpp +381 -0
  155. data/ext/sources/examples/talk-llama/models/qwen35moe.cpp +422 -0
  156. data/ext/sources/examples/talk-llama/models/qwen3moe.cpp +15 -8
  157. data/ext/sources/examples/talk-llama/models/qwen3next.cpp +84 -432
  158. data/ext/sources/examples/talk-llama/models/qwen3vl-moe.cpp +9 -18
  159. data/ext/sources/examples/talk-llama/models/qwen3vl.cpp +8 -17
  160. data/ext/sources/examples/talk-llama/models/refact.cpp +2 -2
  161. data/ext/sources/examples/talk-llama/models/rnd1.cpp +4 -4
  162. data/ext/sources/examples/talk-llama/models/rwkv6-base.cpp +2 -0
  163. data/ext/sources/examples/talk-llama/models/rwkv7-base.cpp +2 -0
  164. data/ext/sources/examples/talk-llama/models/seed-oss.cpp +3 -3
  165. data/ext/sources/examples/talk-llama/models/smallthinker.cpp +4 -4
  166. data/ext/sources/examples/talk-llama/models/smollm3.cpp +3 -3
  167. data/ext/sources/examples/talk-llama/models/stablelm.cpp +2 -2
  168. data/ext/sources/examples/talk-llama/models/starcoder.cpp +2 -2
  169. data/ext/sources/examples/talk-llama/models/starcoder2.cpp +3 -3
  170. data/ext/sources/examples/talk-llama/models/step35-iswa.cpp +165 -0
  171. data/ext/sources/examples/talk-llama/models/t5-dec.cpp +2 -2
  172. data/ext/sources/examples/talk-llama/models/t5-enc.cpp +2 -2
  173. data/ext/sources/examples/talk-llama/models/xverse.cpp +3 -3
  174. data/ext/sources/examples/talk-llama/unicode.cpp +21 -65
  175. data/ext/sources/ggml/CMakeLists.txt +9 -3
  176. data/ext/sources/ggml/include/ggml-backend.h +1 -1
  177. data/ext/sources/ggml/include/ggml-cann.h +1 -1
  178. data/ext/sources/ggml/include/ggml-cpu.h +5 -0
  179. data/ext/sources/ggml/include/ggml-openvino.h +37 -0
  180. data/ext/sources/ggml/include/ggml-opt.h +1 -1
  181. data/ext/sources/ggml/include/ggml-rpc.h +6 -1
  182. data/ext/sources/ggml/include/ggml-virtgpu.h +14 -0
  183. data/ext/sources/ggml/include/ggml.h +56 -9
  184. data/ext/sources/ggml/src/CMakeLists.txt +3 -0
  185. data/ext/sources/ggml/src/ggml-alloc.c +4 -9
  186. data/ext/sources/ggml/src/ggml-backend-dl.cpp +48 -0
  187. data/ext/sources/ggml/src/ggml-backend-dl.h +45 -0
  188. data/ext/sources/ggml/src/ggml-backend-reg.cpp +28 -86
  189. data/ext/sources/ggml/src/ggml-backend.cpp +5 -2
  190. data/ext/sources/ggml/src/ggml-blas/CMakeLists.txt +1 -1
  191. data/ext/sources/ggml/src/ggml-blas/ggml-blas.cpp +6 -2
  192. data/ext/sources/ggml/src/ggml-cann/acl_tensor.cpp +1 -1
  193. data/ext/sources/ggml/src/ggml-cann/acl_tensor.h +1 -1
  194. data/ext/sources/ggml/src/ggml-cann/aclnn_ops.cpp +348 -189
  195. data/ext/sources/ggml/src/ggml-cann/aclnn_ops.h +40 -85
  196. data/ext/sources/ggml/src/ggml-cann/common.h +3 -4
  197. data/ext/sources/ggml/src/ggml-cann/ggml-cann.cpp +44 -62
  198. data/ext/sources/ggml/src/ggml-common.h +11 -0
  199. data/ext/sources/ggml/src/ggml-cpu/CMakeLists.txt +16 -11
  200. data/ext/sources/ggml/src/ggml-cpu/amx/amx.cpp +42 -19
  201. data/ext/sources/ggml/src/ggml-cpu/amx/common.h +34 -10
  202. data/ext/sources/ggml/src/ggml-cpu/amx/mmq.cpp +85 -85
  203. data/ext/sources/ggml/src/ggml-cpu/arch/arm/quants.c +85 -1
  204. data/ext/sources/ggml/src/ggml-cpu/arch/arm/repack.cpp +2744 -548
  205. data/ext/sources/ggml/src/ggml-cpu/arch/riscv/quants.c +1653 -0
  206. data/ext/sources/ggml/src/ggml-cpu/arch/riscv/repack.cpp +1391 -0
  207. data/ext/sources/ggml/src/ggml-cpu/arch/s390/quants.c +8 -10
  208. data/ext/sources/ggml/src/ggml-cpu/arch/x86/quants.c +9 -9
  209. data/ext/sources/ggml/src/ggml-cpu/arch/x86/repack.cpp +118 -18
  210. data/ext/sources/ggml/src/ggml-cpu/arch-fallback.h +107 -26
  211. data/ext/sources/ggml/src/ggml-cpu/binary-ops.cpp +2 -6
  212. data/ext/sources/ggml/src/ggml-cpu/common.h +8 -0
  213. data/ext/sources/ggml/src/ggml-cpu/ggml-cpu-impl.h +3 -0
  214. data/ext/sources/ggml/src/ggml-cpu/ggml-cpu.c +59 -12
  215. data/ext/sources/ggml/src/ggml-cpu/ggml-cpu.cpp +15 -0
  216. data/ext/sources/ggml/src/ggml-cpu/kleidiai/kernels.cpp +21 -20
  217. data/ext/sources/ggml/src/ggml-cpu/kleidiai/kleidiai.cpp +965 -252
  218. data/ext/sources/ggml/src/ggml-cpu/llamafile/sgemm.cpp +584 -197
  219. data/ext/sources/ggml/src/ggml-cpu/ops.cpp +903 -188
  220. data/ext/sources/ggml/src/ggml-cpu/ops.h +1 -0
  221. data/ext/sources/ggml/src/ggml-cpu/quants.c +40 -0
  222. data/ext/sources/ggml/src/ggml-cpu/quants.h +3 -0
  223. data/ext/sources/ggml/src/ggml-cpu/repack.cpp +2890 -679
  224. data/ext/sources/ggml/src/ggml-cpu/repack.h +119 -8
  225. data/ext/sources/ggml/src/ggml-cpu/simd-gemm.h +136 -0
  226. data/ext/sources/ggml/src/ggml-cpu/simd-mappings.h +111 -3
  227. data/ext/sources/ggml/src/ggml-cpu/unary-ops.cpp +1 -1
  228. data/ext/sources/ggml/src/ggml-cpu/vec.cpp +17 -0
  229. data/ext/sources/ggml/src/ggml-cuda/CMakeLists.txt +1 -1
  230. data/ext/sources/ggml/src/ggml-cuda/argsort.cu +19 -10
  231. data/ext/sources/ggml/src/ggml-cuda/binbcast.cu +32 -30
  232. data/ext/sources/ggml/src/ggml-cuda/common.cuh +134 -18
  233. data/ext/sources/ggml/src/ggml-cuda/convert.cu +41 -27
  234. data/ext/sources/ggml/src/ggml-cuda/cpy.cu +6 -3
  235. data/ext/sources/ggml/src/ggml-cuda/fattn-common.cuh +78 -64
  236. data/ext/sources/ggml/src/ggml-cuda/fattn-mma-f16.cuh +384 -143
  237. data/ext/sources/ggml/src/ggml-cuda/fattn-tile.cuh +36 -22
  238. data/ext/sources/ggml/src/ggml-cuda/fattn-vec.cuh +3 -3
  239. data/ext/sources/ggml/src/ggml-cuda/fattn-wmma-f16.cu +26 -5
  240. data/ext/sources/ggml/src/ggml-cuda/fattn-wmma-f16.cuh +1 -1
  241. data/ext/sources/ggml/src/ggml-cuda/fattn.cu +127 -12
  242. data/ext/sources/ggml/src/ggml-cuda/gated_delta_net.cu +263 -0
  243. data/ext/sources/ggml/src/ggml-cuda/gated_delta_net.cuh +4 -0
  244. data/ext/sources/ggml/src/ggml-cuda/ggml-cuda.cu +595 -200
  245. data/ext/sources/ggml/src/ggml-cuda/mean.cu +9 -8
  246. data/ext/sources/ggml/src/ggml-cuda/mma.cuh +173 -6
  247. data/ext/sources/ggml/src/ggml-cuda/mmf.cu +30 -10
  248. data/ext/sources/ggml/src/ggml-cuda/mmf.cuh +158 -85
  249. data/ext/sources/ggml/src/ggml-cuda/mmq.cuh +34 -22
  250. data/ext/sources/ggml/src/ggml-cuda/mmvf.cu +127 -67
  251. data/ext/sources/ggml/src/ggml-cuda/mmvf.cuh +2 -0
  252. data/ext/sources/ggml/src/ggml-cuda/mmvq.cu +157 -65
  253. data/ext/sources/ggml/src/ggml-cuda/mmvq.cuh +1 -0
  254. data/ext/sources/ggml/src/ggml-cuda/norm.cu +18 -76
  255. data/ext/sources/ggml/src/ggml-cuda/pad.cu +13 -10
  256. data/ext/sources/ggml/src/ggml-cuda/quantize.cu +1 -1
  257. data/ext/sources/ggml/src/ggml-cuda/reduce_rows.cuh +2 -16
  258. data/ext/sources/ggml/src/ggml-cuda/rope.cu +233 -133
  259. data/ext/sources/ggml/src/ggml-cuda/softmax.cu +8 -83
  260. data/ext/sources/ggml/src/ggml-cuda/solve_tri.cu +1 -1
  261. data/ext/sources/ggml/src/ggml-cuda/ssm-conv.cu +56 -32
  262. data/ext/sources/ggml/src/ggml-cuda/ssm-conv.cuh +1 -1
  263. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-mma-f16-instance-ncols1_1-ncols2_32.cu +5 -0
  264. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-mma-f16-instance-ncols1_16-ncols2_4.cu +1 -0
  265. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-mma-f16-instance-ncols1_2-ncols2_32.cu +5 -0
  266. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-mma-f16-instance-ncols1_2-ncols2_4.cu +1 -0
  267. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-mma-f16-instance-ncols1_4-ncols2_4.cu +1 -0
  268. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-mma-f16-instance-ncols1_8-ncols2_4.cu +1 -0
  269. data/ext/sources/ggml/src/ggml-cuda/template-instances/generate_cu_files.py +3 -3
  270. data/ext/sources/ggml/src/ggml-cuda/top-k.cu +0 -1
  271. data/ext/sources/ggml/src/ggml-cuda/topk-moe.cu +199 -135
  272. data/ext/sources/ggml/src/ggml-cuda/topk-moe.cuh +20 -14
  273. data/ext/sources/ggml/src/ggml-cuda/unary.cu +55 -0
  274. data/ext/sources/ggml/src/ggml-cuda/unary.cuh +2 -0
  275. data/ext/sources/ggml/src/ggml-cuda/vecdotq.cuh +31 -17
  276. data/ext/sources/ggml/src/ggml-cuda/vendors/hip.h +10 -0
  277. data/ext/sources/ggml/src/ggml-hexagon/CMakeLists.txt +82 -45
  278. data/ext/sources/ggml/src/ggml-hexagon/ggml-hexagon.cpp +334 -160
  279. data/ext/sources/ggml/src/ggml-hexagon/htp/CMakeLists.txt +7 -5
  280. data/ext/sources/ggml/src/ggml-hexagon/htp/act-ops.c +328 -197
  281. data/ext/sources/ggml/src/ggml-hexagon/htp/argsort-ops.c +281 -0
  282. data/ext/sources/ggml/src/ggml-hexagon/htp/binary-ops.c +765 -234
  283. data/ext/sources/ggml/src/ggml-hexagon/htp/cpy-ops.c +252 -0
  284. data/ext/sources/ggml/src/ggml-hexagon/htp/flash-attn-ops.c +412 -265
  285. data/ext/sources/ggml/src/ggml-hexagon/htp/get-rows-ops.c +23 -23
  286. data/ext/sources/ggml/src/ggml-hexagon/htp/{htp-dma.c → hex-dma.c} +1 -1
  287. data/ext/sources/ggml/src/ggml-hexagon/htp/{htp-dma.h → hex-dma.h} +28 -3
  288. data/ext/sources/ggml/src/ggml-hexagon/htp/hex-dump.h +77 -0
  289. data/ext/sources/ggml/src/ggml-hexagon/htp/hex-fastdiv.h +37 -0
  290. data/ext/sources/ggml/src/ggml-hexagon/htp/hex-utils.h +51 -0
  291. data/ext/sources/ggml/src/ggml-hexagon/htp/htp-ctx.h +1 -1
  292. data/ext/sources/ggml/src/ggml-hexagon/htp/htp-msg.h +27 -37
  293. data/ext/sources/ggml/src/ggml-hexagon/htp/htp-ops.h +6 -35
  294. data/ext/sources/ggml/src/ggml-hexagon/htp/hvx-arith.h +443 -0
  295. data/ext/sources/ggml/src/ggml-hexagon/htp/hvx-base.h +240 -0
  296. data/ext/sources/ggml/src/ggml-hexagon/htp/hvx-copy.h +245 -0
  297. data/ext/sources/ggml/src/ggml-hexagon/htp/hvx-div.h +251 -0
  298. data/ext/sources/ggml/src/ggml-hexagon/htp/hvx-dump.h +129 -0
  299. data/ext/sources/ggml/src/ggml-hexagon/htp/hvx-exp.h +215 -0
  300. data/ext/sources/ggml/src/ggml-hexagon/htp/hvx-floor.h +100 -0
  301. data/ext/sources/ggml/src/ggml-hexagon/htp/hvx-inverse.h +210 -0
  302. data/ext/sources/ggml/src/ggml-hexagon/htp/hvx-reduce.h +296 -0
  303. data/ext/sources/ggml/src/ggml-hexagon/htp/hvx-scale.h +133 -0
  304. data/ext/sources/ggml/src/ggml-hexagon/htp/hvx-sigmoid.h +141 -0
  305. data/ext/sources/ggml/src/ggml-hexagon/htp/hvx-sqrt.h +126 -0
  306. data/ext/sources/ggml/src/ggml-hexagon/htp/hvx-types.h +36 -0
  307. data/ext/sources/ggml/src/ggml-hexagon/htp/hvx-utils.h +20 -1347
  308. data/ext/sources/ggml/src/ggml-hexagon/htp/main.c +211 -13
  309. data/ext/sources/ggml/src/ggml-hexagon/htp/matmul-ops.c +1119 -952
  310. data/ext/sources/ggml/src/ggml-hexagon/htp/rope-ops.c +254 -244
  311. data/ext/sources/ggml/src/ggml-hexagon/htp/set-rows-ops.c +36 -36
  312. data/ext/sources/ggml/src/ggml-hexagon/htp/softmax-ops.c +155 -138
  313. data/ext/sources/ggml/src/ggml-hexagon/htp/ssm-conv.c +339 -0
  314. data/ext/sources/ggml/src/ggml-hexagon/htp/sum-rows-ops.c +128 -0
  315. data/ext/sources/ggml/src/ggml-hexagon/htp/unary-ops.c +209 -114
  316. data/ext/sources/ggml/src/ggml-hexagon/htp/worker-pool.c +1 -5
  317. data/ext/sources/ggml/src/ggml-hexagon/htp-drv.cpp +418 -0
  318. data/ext/sources/ggml/src/ggml-hexagon/htp-drv.h +121 -0
  319. data/ext/sources/ggml/src/ggml-hexagon/libdl.h +79 -0
  320. data/ext/sources/ggml/src/ggml-hexagon/libggml-htp.inf +38 -0
  321. data/ext/sources/ggml/src/ggml-hip/CMakeLists.txt +6 -0
  322. data/ext/sources/ggml/src/ggml-impl.h +62 -0
  323. data/ext/sources/ggml/src/ggml-metal/CMakeLists.txt +10 -10
  324. data/ext/sources/ggml/src/ggml-metal/ggml-metal-common.cpp +13 -2
  325. data/ext/sources/ggml/src/ggml-metal/ggml-metal-context.h +8 -0
  326. data/ext/sources/ggml/src/ggml-metal/ggml-metal-context.m +147 -17
  327. data/ext/sources/ggml/src/ggml-metal/ggml-metal-device.cpp +274 -73
  328. data/ext/sources/ggml/src/ggml-metal/ggml-metal-device.h +22 -4
  329. data/ext/sources/ggml/src/ggml-metal/ggml-metal-device.m +102 -36
  330. data/ext/sources/ggml/src/ggml-metal/ggml-metal-impl.h +174 -23
  331. data/ext/sources/ggml/src/ggml-metal/ggml-metal-ops.cpp +580 -280
  332. data/ext/sources/ggml/src/ggml-metal/ggml-metal-ops.h +5 -4
  333. data/ext/sources/ggml/src/ggml-metal/ggml-metal.cpp +320 -107
  334. data/ext/sources/ggml/src/ggml-metal/ggml-metal.metal +1068 -825
  335. data/ext/sources/ggml/src/ggml-opencl/CMakeLists.txt +19 -1
  336. data/ext/sources/ggml/src/ggml-opencl/ggml-opencl.cpp +3108 -636
  337. data/ext/sources/ggml/src/ggml-opencl/kernels/concat.cl +41 -99
  338. data/ext/sources/ggml/src/ggml-opencl/kernels/cpy.cl +45 -0
  339. data/ext/sources/ggml/src/ggml-opencl/kernels/cumsum.cl +139 -0
  340. data/ext/sources/ggml/src/ggml-opencl/kernels/cvt.cl +204 -0
  341. data/ext/sources/ggml/src/ggml-opencl/kernels/diag.cl +27 -0
  342. data/ext/sources/ggml/src/ggml-opencl/kernels/exp.cl +125 -0
  343. data/ext/sources/ggml/src/ggml-opencl/kernels/expm1.cl +87 -56
  344. data/ext/sources/ggml/src/ggml-opencl/kernels/gemm_noshuffle_q4_1_f32.cl +132 -0
  345. data/ext/sources/ggml/src/ggml-opencl/kernels/gemv_noshuffle_general_q8_0_f32.cl +195 -0
  346. data/ext/sources/ggml/src/ggml-opencl/kernels/gemv_noshuffle_q4_1_f32.cl +283 -0
  347. data/ext/sources/ggml/src/ggml-opencl/kernels/l2_norm.cl +71 -0
  348. data/ext/sources/ggml/src/ggml-opencl/kernels/mean.cl +114 -13
  349. data/ext/sources/ggml/src/ggml-opencl/kernels/mul_mm_q4_0_f32_l4_lm.cl +163 -0
  350. data/ext/sources/ggml/src/ggml-opencl/kernels/mul_mm_q4_1_f32_l4_lm.cl +165 -0
  351. data/ext/sources/ggml/src/ggml-opencl/kernels/mul_mm_q6_k_f32_l4_lm.cl +158 -0
  352. data/ext/sources/ggml/src/ggml-opencl/kernels/mul_mm_q8_0_f32_8x4.cl +129 -0
  353. data/ext/sources/ggml/src/ggml-opencl/kernels/mul_mv_q4_1_f32.cl +219 -0
  354. data/ext/sources/ggml/src/ggml-opencl/kernels/mul_mv_q4_1_f32_flat.cl +229 -0
  355. data/ext/sources/ggml/src/ggml-opencl/kernels/mul_mv_q4_k_f32.cl +180 -0
  356. data/ext/sources/ggml/src/ggml-opencl/kernels/{mul_mv_q6_k.cl → mul_mv_q6_k_f32.cl} +4 -0
  357. data/ext/sources/ggml/src/ggml-opencl/kernels/mul_mv_q6_k_f32_flat.cl +194 -0
  358. data/ext/sources/ggml/src/ggml-opencl/kernels/neg.cl +125 -0
  359. data/ext/sources/ggml/src/ggml-opencl/kernels/repeat.cl +31 -32
  360. data/ext/sources/ggml/src/ggml-opencl/kernels/scale.cl +14 -4
  361. data/ext/sources/ggml/src/ggml-opencl/kernels/softplus.cl +88 -60
  362. data/ext/sources/ggml/src/ggml-opencl/kernels/solve_tri.cl +51 -0
  363. data/ext/sources/ggml/src/ggml-opencl/kernels/sum_rows.cl +114 -13
  364. data/ext/sources/ggml/src/ggml-opencl/kernels/tanh.cl +94 -48
  365. data/ext/sources/ggml/src/ggml-opencl/kernels/transpose.cl +26 -0
  366. data/ext/sources/ggml/src/ggml-opencl/kernels/tri.cl +32 -0
  367. data/ext/sources/ggml/src/ggml-openvino/.clang-format +154 -0
  368. data/ext/sources/ggml/src/ggml-openvino/CMakeLists.txt +22 -0
  369. data/ext/sources/ggml/src/ggml-openvino/ggml-decoder.cpp +975 -0
  370. data/ext/sources/ggml/src/ggml-openvino/ggml-decoder.h +294 -0
  371. data/ext/sources/ggml/src/ggml-openvino/ggml-openvino-extra.cpp +373 -0
  372. data/ext/sources/ggml/src/ggml-openvino/ggml-openvino-extra.h +182 -0
  373. data/ext/sources/ggml/src/ggml-openvino/ggml-openvino.cpp +1110 -0
  374. data/ext/sources/ggml/src/ggml-openvino/ggml-quants.cpp +884 -0
  375. data/ext/sources/ggml/src/ggml-openvino/ggml-quants.h +153 -0
  376. data/ext/sources/ggml/src/ggml-openvino/openvino/decoder.h +74 -0
  377. data/ext/sources/ggml/src/ggml-openvino/openvino/frontend.cpp +27 -0
  378. data/ext/sources/ggml/src/ggml-openvino/openvino/frontend.h +23 -0
  379. data/ext/sources/ggml/src/ggml-openvino/openvino/input_model.cpp +17 -0
  380. data/ext/sources/ggml/src/ggml-openvino/openvino/input_model.h +29 -0
  381. data/ext/sources/ggml/src/ggml-openvino/openvino/node_context.h +112 -0
  382. data/ext/sources/ggml/src/ggml-openvino/openvino/op/cont.cpp +48 -0
  383. data/ext/sources/ggml/src/ggml-openvino/openvino/op/cpy.cpp +21 -0
  384. data/ext/sources/ggml/src/ggml-openvino/openvino/op/flash_attn_ext.cpp +90 -0
  385. data/ext/sources/ggml/src/ggml-openvino/openvino/op/get_rows.cpp +69 -0
  386. data/ext/sources/ggml/src/ggml-openvino/openvino/op/glu_geglu.cpp +61 -0
  387. data/ext/sources/ggml/src/ggml-openvino/openvino/op/glu_swiglu.cpp +62 -0
  388. data/ext/sources/ggml/src/ggml-openvino/openvino/op/mulmat.cpp +90 -0
  389. data/ext/sources/ggml/src/ggml-openvino/openvino/op/permute.cpp +102 -0
  390. data/ext/sources/ggml/src/ggml-openvino/openvino/op/reshape.cpp +83 -0
  391. data/ext/sources/ggml/src/ggml-openvino/openvino/op/rms_norm.cpp +46 -0
  392. data/ext/sources/ggml/src/ggml-openvino/openvino/op/rope.cpp +123 -0
  393. data/ext/sources/ggml/src/ggml-openvino/openvino/op/scale.cpp +41 -0
  394. data/ext/sources/ggml/src/ggml-openvino/openvino/op/set_rows.cpp +76 -0
  395. data/ext/sources/ggml/src/ggml-openvino/openvino/op/softmax.cpp +89 -0
  396. data/ext/sources/ggml/src/ggml-openvino/openvino/op/transpose.cpp +23 -0
  397. data/ext/sources/ggml/src/ggml-openvino/openvino/op/unary_silu.cpp +27 -0
  398. data/ext/sources/ggml/src/ggml-openvino/openvino/op/view.cpp +53 -0
  399. data/ext/sources/ggml/src/ggml-openvino/openvino/op_table.cpp +46 -0
  400. data/ext/sources/ggml/src/ggml-openvino/openvino/op_table.h +39 -0
  401. data/ext/sources/ggml/src/ggml-openvino/openvino/pass/eliminate_zp.cpp +123 -0
  402. data/ext/sources/ggml/src/ggml-openvino/openvino/pass/eliminate_zp.h +17 -0
  403. data/ext/sources/ggml/src/ggml-openvino/openvino/pass/fuse_to_sdpa.cpp +60 -0
  404. data/ext/sources/ggml/src/ggml-openvino/openvino/pass/fuse_to_sdpa.h +17 -0
  405. data/ext/sources/ggml/src/ggml-openvino/openvino/pass/mark_decompression_convert_constant_folding.h +29 -0
  406. data/ext/sources/ggml/src/ggml-openvino/openvino/pass/squeeze_matmul.cpp +58 -0
  407. data/ext/sources/ggml/src/ggml-openvino/openvino/pass/squeeze_matmul.h +17 -0
  408. data/ext/sources/ggml/src/ggml-openvino/openvino/translate_session.cpp +293 -0
  409. data/ext/sources/ggml/src/ggml-openvino/openvino/translate_session.h +28 -0
  410. data/ext/sources/ggml/src/ggml-openvino/openvino/utils.cpp +226 -0
  411. data/ext/sources/ggml/src/ggml-openvino/openvino/utils.h +85 -0
  412. data/ext/sources/ggml/src/ggml-openvino/utils.cpp +823 -0
  413. data/ext/sources/ggml/src/ggml-openvino/utils.h +123 -0
  414. data/ext/sources/ggml/src/ggml-quants.c +96 -5
  415. data/ext/sources/ggml/src/ggml-quants.h +3 -0
  416. data/ext/sources/ggml/src/ggml-sycl/CMakeLists.txt +15 -88
  417. data/ext/sources/ggml/src/ggml-sycl/add-id.cpp +5 -1
  418. data/ext/sources/ggml/src/ggml-sycl/backend.hpp +1 -0
  419. data/ext/sources/ggml/src/ggml-sycl/binbcast.cpp +21 -20
  420. data/ext/sources/ggml/src/ggml-sycl/common.hpp +315 -10
  421. data/ext/sources/ggml/src/ggml-sycl/convert.cpp +69 -1
  422. data/ext/sources/ggml/src/ggml-sycl/convert.hpp +22 -1
  423. data/ext/sources/ggml/src/ggml-sycl/count-equal.cpp +1 -1
  424. data/ext/sources/ggml/src/ggml-sycl/dpct/helper.hpp +791 -47
  425. data/ext/sources/ggml/src/ggml-sycl/element_wise.cpp +78 -68
  426. data/ext/sources/ggml/src/ggml-sycl/element_wise.hpp +2 -0
  427. data/ext/sources/ggml/src/ggml-sycl/fattn-common.hpp +1179 -0
  428. data/ext/sources/ggml/src/ggml-sycl/fattn-tile.cpp +55 -0
  429. data/ext/sources/ggml/src/ggml-sycl/fattn-tile.hpp +1338 -0
  430. data/ext/sources/ggml/src/ggml-sycl/fattn-vec.hpp +667 -0
  431. data/ext/sources/ggml/src/ggml-sycl/fattn.cpp +225 -0
  432. data/ext/sources/ggml/src/ggml-sycl/fattn.hpp +22 -0
  433. data/ext/sources/ggml/src/ggml-sycl/gated_delta_net.cpp +309 -0
  434. data/ext/sources/ggml/src/ggml-sycl/gated_delta_net.hpp +8 -0
  435. data/ext/sources/ggml/src/ggml-sycl/ggml-sycl.cpp +316 -51
  436. data/ext/sources/ggml/src/ggml-sycl/norm.cpp +65 -66
  437. data/ext/sources/ggml/src/ggml-sycl/outprod.cpp +3 -3
  438. data/ext/sources/ggml/src/ggml-sycl/presets.hpp +3 -0
  439. data/ext/sources/ggml/src/ggml-sycl/quants.hpp +1 -1
  440. data/ext/sources/ggml/src/ggml-sycl/rope.cpp +450 -287
  441. data/ext/sources/ggml/src/ggml-sycl/rope.hpp +6 -0
  442. data/ext/sources/ggml/src/ggml-sycl/softmax.cpp +6 -6
  443. data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-tile-instance-dkq112-dv112.cpp +5 -0
  444. data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-tile-instance-dkq128-dv128.cpp +5 -0
  445. data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-tile-instance-dkq256-dv256.cpp +5 -0
  446. data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-tile-instance-dkq40-dv40.cpp +5 -0
  447. data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-tile-instance-dkq576-dv512.cpp +5 -0
  448. data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-tile-instance-dkq64-dv64.cpp +5 -0
  449. data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-tile-instance-dkq72-dv72.cpp +5 -0
  450. data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-tile-instance-dkq80-dv80.cpp +5 -0
  451. data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-tile-instance-dkq96-dv96.cpp +5 -0
  452. data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-f16-f16.cpp +7 -0
  453. data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-f16-q4_0.cpp +7 -0
  454. data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-f16-q4_1.cpp +7 -0
  455. data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-f16-q5_0.cpp +7 -0
  456. data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-f16-q5_1.cpp +7 -0
  457. data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-f16-q8_0.cpp +7 -0
  458. data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q4_0-f16.cpp +7 -0
  459. data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q4_0-q4_0.cpp +7 -0
  460. data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q4_0-q4_1.cpp +7 -0
  461. data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q4_0-q5_0.cpp +7 -0
  462. data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q4_0-q5_1.cpp +7 -0
  463. data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q4_0-q8_0.cpp +7 -0
  464. data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q4_1-f16.cpp +7 -0
  465. data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q4_1-q4_0.cpp +7 -0
  466. data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q4_1-q4_1.cpp +7 -0
  467. data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q4_1-q5_0.cpp +7 -0
  468. data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q4_1-q5_1.cpp +7 -0
  469. data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q4_1-q8_0.cpp +7 -0
  470. data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q5_0-f16.cpp +7 -0
  471. data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q5_0-q4_0.cpp +7 -0
  472. data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q5_0-q4_1.cpp +7 -0
  473. data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q5_0-q5_0.cpp +7 -0
  474. data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q5_0-q5_1.cpp +7 -0
  475. data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q5_0-q8_0.cpp +7 -0
  476. data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q5_1-f16.cpp +7 -0
  477. data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q5_1-q4_0.cpp +7 -0
  478. data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q5_1-q4_1.cpp +7 -0
  479. data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q5_1-q5_0.cpp +7 -0
  480. data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q5_1-q5_1.cpp +7 -0
  481. data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q5_1-q8_0.cpp +7 -0
  482. data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q8_0-f16.cpp +7 -0
  483. data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q8_0-q4_0.cpp +7 -0
  484. data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q8_0-q4_1.cpp +7 -0
  485. data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q8_0-q5_0.cpp +7 -0
  486. data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q8_0-q5_1.cpp +7 -0
  487. data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q8_0-q8_0.cpp +7 -0
  488. data/ext/sources/ggml/src/ggml-sycl/vecdotq.hpp +13 -0
  489. data/ext/sources/ggml/src/ggml-sycl/wkv.cpp +1 -1
  490. data/ext/sources/ggml/src/ggml-virtgpu/CMakeLists.txt +70 -0
  491. data/ext/sources/ggml/src/ggml-virtgpu/apir_cs_ggml-rpc-front.cpp +87 -0
  492. data/ext/sources/ggml/src/ggml-virtgpu/backend/CMakeLists.txt +21 -0
  493. data/ext/sources/ggml/src/ggml-virtgpu/backend/apir_cs_ggml-rpc-back.cpp +115 -0
  494. data/ext/sources/ggml/src/ggml-virtgpu/backend/backend-convert.h +13 -0
  495. data/ext/sources/ggml/src/ggml-virtgpu/backend/backend-dispatched-backend.cpp +102 -0
  496. data/ext/sources/ggml/src/ggml-virtgpu/backend/backend-dispatched-buffer-type.cpp +105 -0
  497. data/ext/sources/ggml/src/ggml-virtgpu/backend/backend-dispatched-buffer.cpp +179 -0
  498. data/ext/sources/ggml/src/ggml-virtgpu/backend/backend-dispatched-device.cpp +148 -0
  499. data/ext/sources/ggml/src/ggml-virtgpu/backend/backend-dispatched.cpp +51 -0
  500. data/ext/sources/ggml/src/ggml-virtgpu/backend/backend-dispatched.gen.h +73 -0
  501. data/ext/sources/ggml/src/ggml-virtgpu/backend/backend-dispatched.h +27 -0
  502. data/ext/sources/ggml/src/ggml-virtgpu/backend/backend-virgl-apir.h +32 -0
  503. data/ext/sources/ggml/src/ggml-virtgpu/backend/backend.cpp +144 -0
  504. data/ext/sources/ggml/src/ggml-virtgpu/backend/shared/api_remoting.h +95 -0
  505. data/ext/sources/ggml/src/ggml-virtgpu/backend/shared/apir_backend.gen.h +94 -0
  506. data/ext/sources/ggml/src/ggml-virtgpu/backend/shared/apir_backend.h +50 -0
  507. data/ext/sources/ggml/src/ggml-virtgpu/backend/shared/apir_cs.h +378 -0
  508. data/ext/sources/ggml/src/ggml-virtgpu/backend/shared/apir_cs_ggml.h +232 -0
  509. data/ext/sources/ggml/src/ggml-virtgpu/backend/shared/apir_cs_rpc.h +58 -0
  510. data/ext/sources/ggml/src/ggml-virtgpu/ggml-backend-buffer-type.cpp +81 -0
  511. data/ext/sources/ggml/src/ggml-virtgpu/ggml-backend-buffer.cpp +119 -0
  512. data/ext/sources/ggml/src/ggml-virtgpu/ggml-backend-device.cpp +158 -0
  513. data/ext/sources/ggml/src/ggml-virtgpu/ggml-backend-reg.cpp +213 -0
  514. data/ext/sources/ggml/src/ggml-virtgpu/ggml-backend.cpp +69 -0
  515. data/ext/sources/ggml/src/ggml-virtgpu/ggml-remoting.h +71 -0
  516. data/ext/sources/ggml/src/ggml-virtgpu/ggmlremoting_functions.yaml +166 -0
  517. data/ext/sources/ggml/src/ggml-virtgpu/include/apir_hw.h +9 -0
  518. data/ext/sources/ggml/src/ggml-virtgpu/regenerate_remoting.py +333 -0
  519. data/ext/sources/ggml/src/ggml-virtgpu/virtgpu-apir.h +15 -0
  520. data/ext/sources/ggml/src/ggml-virtgpu/virtgpu-forward-backend.cpp +58 -0
  521. data/ext/sources/ggml/src/ggml-virtgpu/virtgpu-forward-buffer-type.cpp +110 -0
  522. data/ext/sources/ggml/src/ggml-virtgpu/virtgpu-forward-buffer.cpp +173 -0
  523. data/ext/sources/ggml/src/ggml-virtgpu/virtgpu-forward-device.cpp +192 -0
  524. data/ext/sources/ggml/src/ggml-virtgpu/virtgpu-forward-impl.h +36 -0
  525. data/ext/sources/ggml/src/ggml-virtgpu/virtgpu-forward.gen.h +53 -0
  526. data/ext/sources/ggml/src/ggml-virtgpu/virtgpu-shm.cpp +98 -0
  527. data/ext/sources/ggml/src/ggml-virtgpu/virtgpu-shm.h +23 -0
  528. data/ext/sources/ggml/src/ggml-virtgpu/virtgpu-utils.cpp +179 -0
  529. data/ext/sources/ggml/src/ggml-virtgpu/virtgpu-utils.h +86 -0
  530. data/ext/sources/ggml/src/ggml-virtgpu/virtgpu.cpp +544 -0
  531. data/ext/sources/ggml/src/ggml-virtgpu/virtgpu.h +117 -0
  532. data/ext/sources/ggml/src/ggml-vulkan/CMakeLists.txt +1 -1
  533. data/ext/sources/ggml/src/ggml-vulkan/ggml-vulkan.cpp +1250 -465
  534. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/acc.comp +16 -8
  535. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/elu.comp +27 -0
  536. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/flash_attn.comp +374 -170
  537. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/flash_attn_base.glsl +66 -22
  538. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/flash_attn_cm1.comp +389 -201
  539. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/flash_attn_cm2.comp +106 -58
  540. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/flash_attn_mask_opt.comp +162 -0
  541. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/flash_attn_split_k_reduce.comp +9 -8
  542. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/gated_delta_net.comp +128 -0
  543. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/l2_norm.comp +12 -9
  544. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_base.glsl +20 -17
  545. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mm.comp +11 -3
  546. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mm_cm2.comp +8 -4
  547. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mm_id_funcs.glsl +3 -1
  548. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mmq.comp +5 -3
  549. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mmq_funcs.glsl +3 -3
  550. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/rms_norm.comp +2 -3
  551. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/rope_funcs.glsl +36 -63
  552. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/rope_multi.comp +7 -4
  553. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/rope_neox.comp +7 -4
  554. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/rope_norm.comp +7 -4
  555. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/rope_params.glsl +10 -5
  556. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/rope_vision.comp +7 -4
  557. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/sgn.comp +21 -0
  558. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/ssm_conv.comp +16 -10
  559. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/vulkan-shaders-gen.cpp +55 -35
  560. data/ext/sources/ggml/src/ggml-webgpu/ggml-webgpu-shader-lib.hpp +1314 -109
  561. data/ext/sources/ggml/src/ggml-webgpu/ggml-webgpu.cpp +1660 -1371
  562. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/argmax.wgsl +72 -0
  563. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/argsort.wgsl +106 -0
  564. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/argsort_merge.wgsl +134 -0
  565. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/binary.wgsl +141 -0
  566. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/common_decls.tmpl +65 -72
  567. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/concat.wgsl +75 -0
  568. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/cpy.tmpl.wgsl +6 -0
  569. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/cumsum.wgsl +66 -0
  570. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/embed_wgsl.py +40 -5
  571. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/flash_attn.wgsl +105 -60
  572. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/{get_rows.tmpl.wgsl → get_rows.wgsl} +53 -259
  573. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/{mul_mat.tmpl.wgsl → mul_mat.wgsl} +68 -257
  574. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/mul_mat_decls.tmpl +692 -23
  575. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/{mul_mat_reg_tile.tmpl.wgsl → mul_mat_reg_tile.wgsl} +28 -128
  576. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/{mul_mat_subgroup_matrix.tmpl.wgsl → mul_mat_subgroup_matrix.wgsl} +31 -137
  577. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/mul_mat_vec.wgsl +480 -0
  578. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/pad.wgsl +86 -0
  579. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/repeat.wgsl +67 -0
  580. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/{scale.tmpl.wgsl → scale.wgsl} +9 -36
  581. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/set_rows.wgsl +40 -12
  582. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/sum_rows.wgsl +55 -0
  583. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/unary.wgsl +193 -0
  584. data/ext/sources/ggml/src/ggml-zdnn/ggml-zdnn.cpp +6 -1
  585. data/ext/sources/ggml/src/ggml-zendnn/CMakeLists.txt +31 -32
  586. data/ext/sources/ggml/src/ggml-zendnn/ggml-zendnn.cpp +9 -6
  587. data/ext/sources/ggml/src/ggml.c +167 -33
  588. data/ext/sources/ggml/src/gguf.cpp +229 -44
  589. data/ext/sources/src/whisper.cpp +6 -28
  590. data/sig/whisper.rbs +43 -2
  591. data/test/test_context_params.rb +82 -0
  592. data/test/test_token.rb +11 -0
  593. data/test/test_vad_context.rb +58 -8
  594. data/test/test_whisper.rb +20 -0
  595. data/whispercpp.gemspec +1 -1
  596. metadata +240 -28
  597. data/ext/sources/ggml/cmake/BuildTypes.cmake +0 -54
  598. data/ext/sources/ggml/src/ggml-cpu/llamafile/sgemm-ppc.h +0 -333
  599. data/ext/sources/ggml/src/ggml-hexagon/htp/hvx-exp.c +0 -94
  600. data/ext/sources/ggml/src/ggml-hexagon/htp/hvx-inverse.c +0 -72
  601. data/ext/sources/ggml/src/ggml-hexagon/htp/hvx-sigmoid.c +0 -49
  602. data/ext/sources/ggml/src/ggml-hexagon/htp/hvx-utils.c +0 -1020
  603. data/ext/sources/ggml/src/ggml-hexagon/htp/ops-utils.h +0 -149
  604. data/ext/sources/ggml/src/ggml-hexagon/htp-utils.c +0 -454
  605. data/ext/sources/ggml/src/ggml-hexagon/htp-utils.h +0 -221
  606. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/bin_op.tmpl.wgsl +0 -188
  607. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/binary_head.tmpl +0 -45
  608. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/mul_mat_vec.tmpl.wgsl +0 -267
  609. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/set_rows.tmpl.wgsl +0 -112
  610. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/unary_op.wgsl +0 -483
@@ -1,483 +0,0 @@
1
- #define(REPL_TEMPLATES)
2
-
3
- {
4
- "XIELU_FUNC": "{{MUTATE}}[dst_i] = select(((exp(min(src[src_i], {{TYPE}}(params.eps))) - 1.0) - src[src_i]) * {{TYPE}}(params.alpha_n) + {{TYPE}}(params.beta) * src[src_i], {{TYPE}}(params.alpha_p) * src[src_i] * src[src_i] + {{TYPE}}(params.beta) * src[src_i], src[src_i] > 0.0);",
5
- "ABS_FUNC": "{{MUTATE}}[dst_i] = abs(src[src_i]);",
6
- "SGN_FUNC": "{{MUTATE}}[dst_i] = select({{TYPE}}(select(0.0, -1.0, src[src_i] < 0.0)), {{TYPE}}(1.0), src[src_i] > 0.0);",
7
- "NEG_FUNC": "{{MUTATE}}[dst_i] = -src[src_i];",
8
- "STEP_FUNC": "{{MUTATE}}[dst_i] = {{TYPE}}(select(0.0, 1.0, src[src_i] > 0.0));",
9
- "TANH_FUNC": "{{MUTATE}}[dst_i] = tanh(clamp(src[src_i], -9.010913, 9.010913)); // Regarding tanh() domain restrictions in wgsl https://github.com/gpuweb/gpuweb/issues/4458",
10
- "RELU_FUNC": "{{MUTATE}}[dst_i] = select(0.0, src[src_i], src[src_i] > 0.0);",
11
- "ELU_FUNC": "{{MUTATE}}[dst_i] = select(exp(src[src_i]) - 1.0, src[src_i], src[src_i] > 0.0);",
12
- "HARDSIGMOID_FUNC": "{{MUTATE}}[dst_i] = min(1.0, max(0.0, (src[src_i] + 3.0) / 6.0));",
13
- "SIGMOID_FUNC": "{{MUTATE}}[dst_i] = 1.0 / (1.0 + exp(-src[src_i]));",
14
- "SILU_FUNC": "{{MUTATE}}[dst_i] = src[src_i] / (1.0 + exp(-src[src_i]));",
15
- "EXP_FUNC": "{{MUTATE}}[dst_i] = exp(src[src_i]);",
16
- "HARDSWISH_FUNC": "{{MUTATE}}[dst_i] = src[src_i] * min(1.0, max(0.0, (src[src_i] + 3.0) / 6.0));",
17
- "GELU_FUNC": "{{MUTATE}}[dst_i] = 0.5 * src[src_i] * (1.0 + tanh(clamp(sqrt(2.0 / 3.14159265) * (src[src_i] + 0.044715 * pow(src[src_i], 3.0)), -9.010913, 9.010913))); // Regarding tanh() domain restrictions in wgsl https://github.com/gpuweb/gpuweb/issues/4458",
18
- "GELU_QUICK_FUNC": "{{MUTATE}}[dst_i] = src[src_i] * 0.5 * (1.0 + tanh(clamp(0.79788456 * (src[src_i] + 0.044715 * src[src_i] * src[src_i] * src[src_i]), -9.010913, 9.010913))); // Regarding tanh() domain restrictions in wgsl https://github.com/gpuweb/gpuweb/issues/4458",
19
- "GELU_ERF_FUNC": "{{MUTATE}}[dst_i] = 0.5 * src[src_i] * (1.0 + tanh(clamp(0.79788456 * (src[src_i] + 0.044715 * src[src_i] * src[src_i] * src[src_i]), -9.010913, 9.010913))); // Regarding tanh() domain restrictions in wgsl https://github.com/gpuweb/gpuweb/issues/4458",
20
- "CEIL_FUNC": "{{MUTATE}}[dst_i] = ceil(src[src_i]);"
21
- }
22
-
23
- #end(REPL_TEMPLATES)
24
-
25
- #define(VARIANTS)
26
-
27
- [
28
- {
29
- "SHADER_NAME": "abs_f32",
30
- "REPLS": { "TYPE": "f32", "FUNC": "ABS_FUNC", "EXT_PARAMS": "", "MUTATE": "dst" },
31
- "DECLS": ["NOT_INPLACE"]
32
- },
33
- {
34
- "SHADER_NAME": "abs_f16",
35
- "REPLS": { "TYPE": "f16", "FUNC": "ABS_FUNC", "EXT_PARAMS": "", "MUTATE": "dst" },
36
- "DECLS": ["NOT_INPLACE"]
37
- },
38
- {
39
- "SHADER_NAME": "abs_inplace_f32",
40
- "REPLS": { "TYPE": "f32", "FUNC": "ABS_FUNC", "EXT_PARAMS": "", "MUTATE": "src" },
41
- "DECLS": ["INPLACE"]
42
- },
43
- {
44
- "SHADER_NAME": "abs_inplace_f16",
45
- "REPLS": { "TYPE": "f16", "FUNC": "ABS_FUNC", "EXT_PARAMS": "", "MUTATE": "src" },
46
- "DECLS": ["INPLACE"]
47
- },
48
-
49
- {
50
- "SHADER_NAME": "sgn_f32",
51
- "REPLS": { "TYPE": "f32", "FUNC": "SGN_FUNC", "EXT_PARAMS": "", "MUTATE": "dst" },
52
- "DECLS": ["NOT_INPLACE"]
53
- },
54
- {
55
- "SHADER_NAME": "sgn_f16",
56
- "REPLS": { "TYPE": "f16", "FUNC": "SGN_FUNC", "EXT_PARAMS": "", "MUTATE": "dst" },
57
- "DECLS": ["NOT_INPLACE"]
58
- },
59
- {
60
- "SHADER_NAME": "sgn_inplace_f32",
61
- "REPLS": { "TYPE": "f32", "FUNC": "SGN_FUNC", "EXT_PARAMS": "", "MUTATE": "src" },
62
- "DECLS": ["INPLACE"]
63
- },
64
- {
65
- "SHADER_NAME": "sgn_inplace_f16",
66
- "REPLS": { "TYPE": "f16", "FUNC": "SGN_FUNC", "EXT_PARAMS": "", "MUTATE": "src" },
67
- "DECLS": ["INPLACE"]
68
- },
69
-
70
- {
71
- "SHADER_NAME": "neg_f32",
72
- "REPLS": { "TYPE": "f32", "FUNC": "NEG_FUNC", "EXT_PARAMS": "", "MUTATE": "dst" },
73
- "DECLS": ["NOT_INPLACE"]
74
- },
75
- {
76
- "SHADER_NAME": "neg_f16",
77
- "REPLS": { "TYPE": "f16", "FUNC": "NEG_FUNC", "EXT_PARAMS": "", "MUTATE": "dst" },
78
- "DECLS": ["NOT_INPLACE"]
79
- },
80
- {
81
- "SHADER_NAME": "neg_inplace_f32",
82
- "REPLS": { "TYPE": "f32", "FUNC": "NEG_FUNC", "EXT_PARAMS": "", "MUTATE": "src" },
83
- "DECLS": ["INPLACE"]
84
- },
85
- {
86
- "SHADER_NAME": "neg_inplace_f16",
87
- "REPLS": { "TYPE": "f16", "FUNC": "NEG_FUNC", "EXT_PARAMS": "", "MUTATE": "src" },
88
- "DECLS": ["INPLACE"]
89
- },
90
-
91
- {
92
- "SHADER_NAME": "step_f32",
93
- "REPLS": { "TYPE": "f32", "FUNC": "STEP_FUNC", "EXT_PARAMS": "", "MUTATE": "dst" },
94
- "DECLS": ["NOT_INPLACE"]
95
- },
96
- {
97
- "SHADER_NAME": "step_f16",
98
- "REPLS": { "TYPE": "f16", "FUNC": "STEP_FUNC", "EXT_PARAMS": "", "MUTATE": "dst" },
99
- "DECLS": ["NOT_INPLACE"]
100
- },
101
- {
102
- "SHADER_NAME": "step_inplace_f32",
103
- "REPLS": { "TYPE": "f32", "FUNC": "STEP_FUNC", "EXT_PARAMS": "", "MUTATE": "src" },
104
- "DECLS": ["INPLACE"]
105
- },
106
- {
107
- "SHADER_NAME": "step_inplace_f16",
108
- "REPLS": { "TYPE": "f16", "FUNC": "STEP_FUNC", "EXT_PARAMS": "", "MUTATE": "src" },
109
- "DECLS": ["INPLACE"]
110
- },
111
-
112
- {
113
- "SHADER_NAME": "tanh_f32",
114
- "REPLS": { "TYPE": "f32", "FUNC": "TANH_FUNC", "EXT_PARAMS": "", "MUTATE": "dst" },
115
- "DECLS": ["NOT_INPLACE"]
116
- },
117
- {
118
- "SHADER_NAME": "tanh_f16",
119
- "REPLS": { "TYPE": "f16", "FUNC": "TANH_FUNC", "EXT_PARAMS": "", "MUTATE": "dst" },
120
- "DECLS": ["NOT_INPLACE"]
121
- },
122
- {
123
- "SHADER_NAME": "tanh_inplace_f32",
124
- "REPLS": { "TYPE": "f32", "FUNC": "TANH_FUNC", "EXT_PARAMS": "", "MUTATE": "src" },
125
- "DECLS": ["INPLACE"]
126
- },
127
- {
128
- "SHADER_NAME": "tanh_inplace_f16",
129
- "REPLS": { "TYPE": "f16", "FUNC": "TANH_FUNC", "EXT_PARAMS": "", "MUTATE": "src" },
130
- "DECLS": ["INPLACE"]
131
- },
132
-
133
- {
134
- "SHADER_NAME": "elu_f32",
135
- "REPLS": { "TYPE": "f32", "FUNC": "ELU_FUNC", "EXT_PARAMS": "", "MUTATE": "dst" },
136
- "DECLS": ["NOT_INPLACE"]
137
- },
138
- {
139
- "SHADER_NAME": "elu_f16",
140
- "REPLS": { "TYPE": "f16", "FUNC": "ELU_FUNC", "EXT_PARAMS": "", "MUTATE": "dst" },
141
- "DECLS": ["NOT_INPLACE"]
142
- },
143
- {
144
- "SHADER_NAME": "elu_inplace_f32",
145
- "REPLS": { "TYPE": "f32", "FUNC": "ELU_FUNC", "EXT_PARAMS": "", "MUTATE": "src" },
146
- "DECLS": ["INPLACE"]
147
- },
148
- {
149
- "SHADER_NAME": "elu_inplace_f16",
150
- "REPLS": { "TYPE": "f16", "FUNC": "ELU_FUNC", "EXT_PARAMS": "", "MUTATE": "src" },
151
- "DECLS": ["INPLACE"]
152
- },
153
-
154
- {
155
- "SHADER_NAME": "relu_f32",
156
- "REPLS": { "TYPE": "f32", "FUNC": "RELU_FUNC", "EXT_PARAMS": "", "MUTATE": "dst" },
157
- "DECLS": ["NOT_INPLACE"]
158
- },
159
- {
160
- "SHADER_NAME": "relu_f16",
161
- "REPLS": { "TYPE": "f16", "FUNC": "RELU_FUNC", "EXT_PARAMS": "", "MUTATE": "dst" },
162
- "DECLS": ["NOT_INPLACE"]
163
- },
164
- {
165
- "SHADER_NAME": "relu_inplace_f32",
166
- "REPLS": { "TYPE": "f32", "FUNC": "RELU_FUNC", "EXT_PARAMS": "", "MUTATE": "src" },
167
- "DECLS": ["INPLACE"]
168
- },
169
- {
170
- "SHADER_NAME": "relu_inplace_f16",
171
- "REPLS": { "TYPE": "f16", "FUNC": "RELU_FUNC", "EXT_PARAMS": "", "MUTATE": "src" },
172
- "DECLS": ["INPLACE"]
173
- },
174
-
175
- {
176
- "SHADER_NAME": "sigmoid_f32",
177
- "REPLS": { "TYPE": "f32", "FUNC": "SIGMOID_FUNC", "EXT_PARAMS": "", "MUTATE": "dst" },
178
- "DECLS": ["NOT_INPLACE"]
179
- },
180
- {
181
- "SHADER_NAME": "sigmoid_f16",
182
- "REPLS": { "TYPE": "f16", "FUNC": "SIGMOID_FUNC", "EXT_PARAMS": "", "MUTATE": "dst" },
183
- "DECLS": ["NOT_INPLACE"]
184
- },
185
- {
186
- "SHADER_NAME": "sigmoid_inplace_f32",
187
- "REPLS": { "TYPE": "f32", "FUNC": "SIGMOID_FUNC", "EXT_PARAMS": "", "MUTATE": "src" },
188
- "DECLS": ["INPLACE"]
189
- },
190
- {
191
- "SHADER_NAME": "sigmoid_inplace_f16",
192
- "REPLS": { "TYPE": "f16", "FUNC": "SIGMOID_FUNC", "EXT_PARAMS": "", "MUTATE": "src" },
193
- "DECLS": ["INPLACE"]
194
- },
195
-
196
- {
197
- "SHADER_NAME": "silu_f32",
198
- "REPLS": { "TYPE": "f32", "FUNC": "SILU_FUNC", "EXT_PARAMS": "", "MUTATE": "dst" },
199
- "DECLS": ["NOT_INPLACE"]
200
- },
201
- {
202
- "SHADER_NAME": "silu_f16",
203
- "REPLS": { "TYPE": "f16", "FUNC": "SILU_FUNC", "EXT_PARAMS": "", "MUTATE": "dst" },
204
- "DECLS": ["NOT_INPLACE"]
205
- },
206
- {
207
- "SHADER_NAME": "silu_inplace_f32",
208
- "REPLS": { "TYPE": "f32", "FUNC": "SILU_FUNC", "EXT_PARAMS": "", "MUTATE": "src" },
209
- "DECLS": ["INPLACE"]
210
- },
211
- {
212
- "SHADER_NAME": "silu_inplace_f16",
213
- "REPLS": { "TYPE": "f16", "FUNC": "SILU_FUNC", "EXT_PARAMS": "", "MUTATE": "src" },
214
- "DECLS": ["INPLACE"]
215
- },
216
-
217
- {
218
- "SHADER_NAME": "exp_f32",
219
- "REPLS": { "TYPE": "f32", "FUNC": "EXP_FUNC", "EXT_PARAMS": "", "MUTATE": "dst" },
220
- "DECLS": ["NOT_INPLACE"]
221
- },
222
- {
223
- "SHADER_NAME": "exp_f16",
224
- "REPLS": { "TYPE": "f16", "FUNC": "EXP_FUNC", "EXT_PARAMS": "", "MUTATE": "dst" },
225
- "DECLS": ["NOT_INPLACE"]
226
- },
227
- {
228
- "SHADER_NAME": "exp_inplace_f32",
229
- "REPLS": { "TYPE": "f32", "FUNC": "EXP_FUNC", "EXT_PARAMS": "", "MUTATE": "src" },
230
- "DECLS": ["INPLACE"]
231
- },
232
- {
233
- "SHADER_NAME": "exp_inplace_f16",
234
- "REPLS": { "TYPE": "f16", "FUNC": "EXP_FUNC", "EXT_PARAMS": "", "MUTATE": "src" },
235
- "DECLS": ["INPLACE"]
236
- },
237
-
238
- {
239
- "SHADER_NAME": "hardsigmoid_f32",
240
- "REPLS": { "TYPE": "f32", "FUNC": "HARDSIGMOID_FUNC", "EXT_PARAMS": "", "MUTATE": "dst" },
241
- "DECLS": ["NOT_INPLACE"]
242
- },
243
- {
244
- "SHADER_NAME": "hardsigmoid_f16",
245
- "REPLS": { "TYPE": "f16", "FUNC": "HARDSIGMOID_FUNC", "EXT_PARAMS": "", "MUTATE": "dst" },
246
- "DECLS": ["NOT_INPLACE"]
247
- },
248
- {
249
- "SHADER_NAME": "hardsigmoid_inplace_f32",
250
- "REPLS": { "TYPE": "f32", "FUNC": "HARDSIGMOID_FUNC", "EXT_PARAMS": "", "MUTATE": "src" },
251
- "DECLS": ["INPLACE"]
252
- },
253
- {
254
- "SHADER_NAME": "hardsigmoid_inplace_f16",
255
- "REPLS": { "TYPE": "f16", "FUNC": "HARDSIGMOID_FUNC", "EXT_PARAMS": "", "MUTATE": "src" },
256
- "DECLS": ["INPLACE"]
257
- },
258
-
259
- {
260
- "SHADER_NAME": "hardswish_f32",
261
- "REPLS": { "TYPE": "f32", "FUNC": "HARDSWISH_FUNC", "EXT_PARAMS": "", "MUTATE": "dst" },
262
- "DECLS": ["NOT_INPLACE"]
263
- },
264
- {
265
- "SHADER_NAME": "hardswish_f16",
266
- "REPLS": { "TYPE": "f16", "FUNC": "HARDSWISH_FUNC", "EXT_PARAMS": "", "MUTATE": "dst" },
267
- "DECLS": ["NOT_INPLACE"]
268
- },
269
- {
270
- "SHADER_NAME": "hardswish_inplace_f32",
271
- "REPLS": { "TYPE": "f32", "FUNC": "HARDSWISH_FUNC", "EXT_PARAMS": "", "MUTATE": "src" },
272
- "DECLS": ["INPLACE"]
273
- },
274
- {
275
- "SHADER_NAME": "hardswish_inplace_f16",
276
- "REPLS": { "TYPE": "f16", "FUNC": "HARDSWISH_FUNC", "EXT_PARAMS": "", "MUTATE": "src" },
277
- "DECLS": ["INPLACE"]
278
- },
279
-
280
- {
281
- "SHADER_NAME": "gelu_f32",
282
- "REPLS": { "TYPE": "f32", "FUNC": "GELU_FUNC", "EXT_PARAMS": "", "MUTATE": "dst" },
283
- "DECLS": ["NOT_INPLACE"]
284
- },
285
- {
286
- "SHADER_NAME": "gelu_f16",
287
- "REPLS": { "TYPE": "f16", "FUNC": "GELU_FUNC", "EXT_PARAMS": "", "MUTATE": "dst" },
288
- "DECLS": ["NOT_INPLACE"]
289
- },
290
- {
291
- "SHADER_NAME": "gelu_inplace_f32",
292
- "REPLS": { "TYPE": "f32", "FUNC": "GELU_FUNC", "EXT_PARAMS": "", "MUTATE": "src" },
293
- "DECLS": ["INPLACE"]
294
- },
295
- {
296
- "SHADER_NAME": "gelu_inplace_f16",
297
- "REPLS": { "TYPE": "f16", "FUNC": "GELU_FUNC", "EXT_PARAMS": "", "MUTATE": "src" },
298
- "DECLS": ["INPLACE"]
299
- },
300
-
301
- {
302
- "SHADER_NAME": "gelu_quick_f32",
303
- "REPLS": { "TYPE": "f32", "FUNC": "GELU_QUICK_FUNC", "EXT_PARAMS": "", "MUTATE": "dst" },
304
- "DECLS": ["NOT_INPLACE"]
305
- },
306
- {
307
- "SHADER_NAME": "gelu_quick_f16",
308
- "REPLS": { "TYPE": "f16", "FUNC": "GELU_QUICK_FUNC", "EXT_PARAMS": "", "MUTATE": "dst" },
309
- "DECLS": ["NOT_INPLACE"]
310
- },
311
- {
312
- "SHADER_NAME": "gelu_quick_inplace_f32",
313
- "REPLS": { "TYPE": "f32", "FUNC": "GELU_QUICK_FUNC", "EXT_PARAMS": "", "MUTATE": "src" },
314
- "DECLS": ["INPLACE"]
315
- },
316
- {
317
- "SHADER_NAME": "gelu_quick_inplace_f16",
318
- "REPLS": { "TYPE": "f16", "FUNC": "GELU_QUICK_FUNC", "EXT_PARAMS": "", "MUTATE": "src" },
319
- "DECLS": ["INPLACE"]
320
- },
321
-
322
- {
323
- "SHADER_NAME": "xielu_f32",
324
- "REPLS": { "TYPE": "f32", "FUNC": "XIELU_FUNC", "EXT_PARAMS": "alpha_n: f32, alpha_p: f32, beta: f32, eps: f32", "MUTATE": "dst" },
325
- "DECLS": ["NOT_INPLACE"]
326
- },
327
- {
328
- "SHADER_NAME": "xielu_f16",
329
- "REPLS": { "TYPE": "f16", "FUNC": "XIELU_FUNC", "EXT_PARAMS": "alpha_n: f32, alpha_p: f32, beta: f32, eps: f32", "MUTATE": "dst" },
330
- "DECLS": ["NOT_INPLACE"]
331
- },
332
- {
333
- "SHADER_NAME": "xielu_inplace_f32",
334
- "REPLS": { "TYPE": "f32", "FUNC": "XIELU_FUNC", "EXT_PARAMS": "alpha_n: f32, alpha_p: f32, beta: f32, eps: f32", "MUTATE": "src" },
335
- "DECLS": ["INPLACE"]
336
- },
337
- {
338
- "SHADER_NAME": "xielu_inplace_f16",
339
- "REPLS": { "TYPE": "f16", "FUNC": "XIELU_FUNC", "EXT_PARAMS": "alpha_n: f32, alpha_p: f32, beta: f32, eps: f32", "MUTATE": "src" },
340
- "DECLS": ["INPLACE"]
341
- },
342
- {
343
- "SHADER_NAME": "gelu_erf_f32",
344
- "REPLS": { "TYPE": "f32", "FUNC": "GELU_ERF_FUNC", "EXT_PARAMS": "", "MUTATE": "dst" },
345
- "DECLS": ["NOT_INPLACE"]
346
- },
347
- {
348
- "SHADER_NAME": "gelu_erf_f16",
349
- "REPLS": { "TYPE": "f16", "FUNC": "GELU_ERF_FUNC", "EXT_PARAMS": "", "MUTATE": "dst" },
350
- "DECLS": ["NOT_INPLACE"]
351
- },
352
- {
353
- "SHADER_NAME": "gelu_erf_inplace_f32",
354
- "REPLS": { "TYPE": "f32", "FUNC": "GELU_ERF_FUNC", "EXT_PARAMS": "", "MUTATE": "src" },
355
- "DECLS": ["INPLACE"]
356
- },
357
- {
358
- "SHADER_NAME": "gelu_erf_inplace_f16",
359
- "REPLS": { "TYPE": "f16", "FUNC": "GELU_ERF_FUNC", "EXT_PARAMS": "", "MUTATE": "src" },
360
- "DECLS": ["INPLACE"]
361
- },
362
-
363
- {
364
- "SHADER_NAME": "ceil_f32",
365
- "REPLS": { "TYPE": "f32", "FUNC": "CEIL_FUNC", "EXT_PARAMS": "", "MUTATE": "dst" },
366
- "DECLS": ["NOT_INPLACE"]
367
- },
368
- {
369
- "SHADER_NAME": "ceil_f16",
370
- "REPLS": { "TYPE": "f16", "FUNC": "CEIL_FUNC", "EXT_PARAMS": "", "MUTATE": "dst" },
371
- "DECLS": ["NOT_INPLACE"]
372
- },
373
- {
374
- "SHADER_NAME": "ceil_inplace_f32",
375
- "REPLS": { "TYPE": "f32", "FUNC": "CEIL_FUNC", "EXT_PARAMS": "", "MUTATE": "src" },
376
- "DECLS": ["INPLACE"]
377
- },
378
- {
379
- "SHADER_NAME": "ceil_inplace_f16",
380
- "REPLS": { "TYPE": "f16", "FUNC": "CEIL_FUNC", "EXT_PARAMS": "", "MUTATE": "src" },
381
- "DECLS": ["INPLACE"]
382
- }
383
- ]
384
-
385
- #end(VARIANTS)
386
-
387
- #define(DECLS)
388
-
389
- #decl(INPLACE)
390
-
391
- @group(0) @binding(1)
392
- var<uniform> params: Params;
393
-
394
- #enddecl(INPLACE)
395
-
396
- #decl(NOT_INPLACE)
397
-
398
- @group(0) @binding(1)
399
- var<storage, read_write> dst: array<{{TYPE}}>;
400
-
401
- @group(0) @binding(2)
402
- var<uniform> params: Params;
403
-
404
- #enddecl(NOT_INPLACE)
405
-
406
- #end(DECLS)
407
-
408
- #define(SHADER)
409
-
410
- enable f16;
411
-
412
- fn update(dst_i: u32, src_i: u32) {
413
- {{FUNC}}
414
- }
415
-
416
- @group(0) @binding(0)
417
- var<storage, read_write> src: array<{{TYPE}}>;
418
-
419
- DECLS
420
-
421
- struct Params {
422
- ne: u32, // total number of elements
423
- offset_src: u32, // in elements
424
- offset_dst: u32, // in elements
425
-
426
- // Strides (in elements) — may be permuted
427
- stride_src0: u32,
428
- stride_src1: u32,
429
- stride_src2: u32,
430
- stride_src3: u32,
431
-
432
- stride_dst0: u32,
433
- stride_dst1: u32,
434
- stride_dst2: u32,
435
- stride_dst3: u32,
436
-
437
- // Logical shapes
438
- src_ne0: u32,
439
- src_ne1: u32,
440
- src_ne2: u32,
441
-
442
- dst_ne0: u32,
443
- dst_ne1: u32,
444
- dst_ne2: u32,
445
-
446
- {{EXT_PARAMS}}
447
- };
448
-
449
- override wg_size: u32;
450
- @compute @workgroup_size(wg_size)
451
- fn main(@builtin(global_invocation_id) gid: vec3<u32>) {
452
- if (gid.x >= params.ne) {
453
- return;
454
- }
455
-
456
- var i = gid.x;
457
- let i3 = i / (params.src_ne2 * params.src_ne1 * params.src_ne0);
458
- i = i % (params.src_ne2 * params.src_ne1 * params.src_ne0);
459
- let i2 = i / (params.src_ne1 * params.src_ne0);
460
- i = i % (params.src_ne1 * params.src_ne0);
461
- let i1 = i / params.src_ne0;
462
- let i0 = i % params.src_ne0;
463
-
464
- var j = gid.x;
465
- let j3 = j / (params.dst_ne2 * params.dst_ne1 * params.dst_ne0);
466
- j = j % (params.dst_ne2 * params.dst_ne1 * params.dst_ne0);
467
- let j2 = j / (params.dst_ne1 * params.dst_ne0);
468
- j = j % (params.dst_ne1 * params.dst_ne0);
469
- let j1 = j / params.dst_ne0;
470
- let j0 = j % params.dst_ne0;
471
-
472
- let src_idx = i0 * params.stride_src0 + i1 * params.stride_src1 +
473
- i2 * params.stride_src2 + i3 * params.stride_src3;
474
-
475
- let dst_idx = j0 * params.stride_dst0 + j1 * params.stride_dst1 +
476
- j2 * params.stride_dst2 + j3 * params.stride_dst3;
477
-
478
-
479
- update(params.offset_dst + dst_idx, params.offset_src + src_idx);
480
- }
481
-
482
- #end(SHADER)
483
-