toy 0.8.0

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 (2107) hide show
  1. checksums.yaml +7 -0
  2. data/CHANGELOG.md +1124 -0
  3. data/LICENSE +21 -0
  4. data/Makefile +2022 -0
  5. data/README.md +154 -0
  6. data/bin/toy +10 -0
  7. data/lib/toy/compute.rb +135 -0
  8. data/lib/toy/compute_cuda.rb +104 -0
  9. data/lib/toy/compute_metal.rb +97 -0
  10. data/lib/toy/core/cli/describe.rb +188 -0
  11. data/lib/toy/core/cli/eval.rb +385 -0
  12. data/lib/toy/core/cli/exit_codes.rb +15 -0
  13. data/lib/toy/core/cli/fetch.rb +238 -0
  14. data/lib/toy/core/cli/infer.rb +268 -0
  15. data/lib/toy/core/cli/install.rb +228 -0
  16. data/lib/toy/core/cli/list.rb +86 -0
  17. data/lib/toy/core/cli/manifest.rb +49 -0
  18. data/lib/toy/core/cli/new.rb +594 -0
  19. data/lib/toy/core/cli/serve.rb +237 -0
  20. data/lib/toy/core/cli/train.rb +471 -0
  21. data/lib/toy/core/cli.rb +165 -0
  22. data/lib/toy/core/config.rb +64 -0
  23. data/lib/toy/core/gguf_meta.rb +161 -0
  24. data/lib/toy/core/model_scan.rb +221 -0
  25. data/lib/toy/core/run_log.rb +94 -0
  26. data/lib/toy/core/toy_root.rb +95 -0
  27. data/lib/toy/dev/toy_card.rb +299 -0
  28. data/lib/toy/dev/toy_describe_flow.rb +412 -0
  29. data/lib/toy/dev/toy_logprobs.rb +86 -0
  30. data/lib/toy/dev/toy_tap.rb +183 -0
  31. data/lib/toy/dev/toy_token_drift.rb +121 -0
  32. data/lib/toy/ffi/tinynn.rb +1491 -0
  33. data/lib/toy/ffi/tinynn_cuda.rb +1124 -0
  34. data/lib/toy/ffi/tinynn_metal.rb +359 -0
  35. data/lib/toy/ffi_manifest.rb +84 -0
  36. data/lib/toy/io/bpe.rb +325 -0
  37. data/lib/toy/io/gguf_kv.rb +35 -0
  38. data/lib/toy/io/gguf_load.rb +331 -0
  39. data/lib/toy/io/loaders/toy_gpt2_loader.rb +70 -0
  40. data/lib/toy/io/loaders/toy_smollm2_loader.rb +754 -0
  41. data/lib/toy/io/model_index.rb +206 -0
  42. data/lib/toy/io/run_bundle.rb +280 -0
  43. data/lib/toy/io/tokenizer.rb +613 -0
  44. data/lib/toy/io/toy_corpus_loader.rb +52 -0
  45. data/lib/toy/io/toy_events.rb +56 -0
  46. data/lib/toy/io/toy_image_loader.rb +48 -0
  47. data/lib/toy/llm/adamw.rb +169 -0
  48. data/lib/toy/llm/archs/llama_arch.rb +233 -0
  49. data/lib/toy/llm/archs/llama_arch_cuda.rb +237 -0
  50. data/lib/toy/llm/archs/llama_arch_metal.rb +237 -0
  51. data/lib/toy/llm/blocks/transformer_block.rb +876 -0
  52. data/lib/toy/llm/blocks/transformer_block_cuda.rb +880 -0
  53. data/lib/toy/llm/blocks/transformer_block_metal.rb +880 -0
  54. data/lib/toy/llm/classify_batch.rb +88 -0
  55. data/lib/toy/llm/engine/gpt2_fwd_engine.rb +360 -0
  56. data/lib/toy/llm/engine/gpt2_fwd_engine_cuda.rb +362 -0
  57. data/lib/toy/llm/engine/gpt2_fwd_engine_metal.rb +362 -0
  58. data/lib/toy/llm/engine/gpt2_kv_engine.rb +346 -0
  59. data/lib/toy/llm/engine/gpt2_kv_engine_cuda.rb +348 -0
  60. data/lib/toy/llm/engine/gpt2_kv_engine_metal.rb +348 -0
  61. data/lib/toy/llm/engine/gpt2_seq_engine.rb +289 -0
  62. data/lib/toy/llm/engine/gpt2_seq_engine_cuda.rb +293 -0
  63. data/lib/toy/llm/engine/gpt2_seq_engine_metal.rb +293 -0
  64. data/lib/toy/llm/engine/llama_kv_engine.rb +1593 -0
  65. data/lib/toy/llm/engine/llama_kv_engine_cuda.rb +1526 -0
  66. data/lib/toy/llm/engine/llama_kv_engine_metal.rb +1526 -0
  67. data/lib/toy/llm/engine/llama_seq_engine.rb +1233 -0
  68. data/lib/toy/llm/engine/llama_seq_engine_cuda.rb +1238 -0
  69. data/lib/toy/llm/engine/llama_seq_engine_metal.rb +1238 -0
  70. data/lib/toy/llm/engine/vit_tiny_engine.rb +467 -0
  71. data/lib/toy/llm/labels.rb +142 -0
  72. data/lib/toy/llm/primitives/gqa.rb +62 -0
  73. data/lib/toy/llm/primitives/gqa_cuda.rb +66 -0
  74. data/lib/toy/llm/primitives/gqa_metal.rb +66 -0
  75. data/lib/toy/llm/primitives/rms_norm.rb +39 -0
  76. data/lib/toy/llm/primitives/rms_norm_cuda.rb +43 -0
  77. data/lib/toy/llm/primitives/rms_norm_metal.rb +43 -0
  78. data/lib/toy/llm/primitives/rope.rb +68 -0
  79. data/lib/toy/llm/primitives/rope_cuda.rb +72 -0
  80. data/lib/toy/llm/primitives/rope_metal.rb +72 -0
  81. data/lib/toy/llm/primitives/swiglu.rb +41 -0
  82. data/lib/toy/llm/primitives/swiglu_cuda.rb +45 -0
  83. data/lib/toy/llm/primitives/swiglu_metal.rb +45 -0
  84. data/lib/toy/llm/recipe_options.rb +71 -0
  85. data/lib/toy/llm/recipes/from_scratch.rb +105 -0
  86. data/lib/toy/llm/recipes/from_scratch_cuda.rb +109 -0
  87. data/lib/toy/llm/recipes/from_scratch_metal.rb +109 -0
  88. data/lib/toy/llm/recipes/lora.rb +110 -0
  89. data/lib/toy/llm/recipes/lora_cuda.rb +114 -0
  90. data/lib/toy/llm/recipes/lora_metal.rb +114 -0
  91. data/lib/toy/llm/recipes/vit_tiny.rb +75 -0
  92. data/lib/toy/llm/recipes/warm_start.rb +235 -0
  93. data/lib/toy/llm/recipes/warm_start_cuda.rb +239 -0
  94. data/lib/toy/llm/recipes/warm_start_metal.rb +239 -0
  95. data/lib/toy/llm/training_batch.rb +133 -0
  96. data/lib/toy/models/arch.rb +253 -0
  97. data/lib/toy/models/gpt2.rb +311 -0
  98. data/lib/toy/models/toy_gpt2.rb +177 -0
  99. data/lib/toy/models/toy_smollm2.rb +393 -0
  100. data/lib/toy/models/toy_vit.rb +83 -0
  101. data/lib/toy/models/transformer.rb +1494 -0
  102. data/lib/toy/models/transformer_lm.rb +298 -0
  103. data/lib/toy/models/transformer_lm_cuda.rb +159 -0
  104. data/lib/toy/models/transformer_lm_metal.rb +142 -0
  105. data/lib/toy/mri.rb +300 -0
  106. data/lib/toy/run/eval.rb +76 -0
  107. data/lib/toy/run/eval_cuda.rb +66 -0
  108. data/lib/toy/run/eval_lmc.rb +334 -0
  109. data/lib/toy/run/eval_metal.rb +67 -0
  110. data/lib/toy/run/infer.rb +130 -0
  111. data/lib/toy/run/infer_cuda.rb +118 -0
  112. data/lib/toy/run/infer_metal.rb +119 -0
  113. data/lib/toy/run/infer_trace.rb +37 -0
  114. data/lib/toy/run/serve.rb +144 -0
  115. data/lib/toy/run/train.rb +404 -0
  116. data/lib/toy/run/train_cuda.rb +397 -0
  117. data/lib/toy/run/train_gpt2.rb +103 -0
  118. data/lib/toy/run/train_gpt2_cuda.rb +85 -0
  119. data/lib/toy/run/train_gpt2_metal.rb +85 -0
  120. data/lib/toy/run/train_lora.rb +207 -0
  121. data/lib/toy/run/train_lora_cuda.rb +219 -0
  122. data/lib/toy/run/train_metal.rb +227 -0
  123. data/lib/toy/run/train_vit.rb +251 -0
  124. data/lib/toy/serve/openai/embeddings_handler.rb +92 -0
  125. data/lib/toy/serve/openai/handlers.rb +143 -0
  126. data/lib/toy/serve/openai/server.rb +159 -0
  127. data/lib/toy/train/sampler.rb +314 -0
  128. data/lib/toy/train/toy_chat_template.rb +179 -0
  129. data/lib/toy/train/toy_drift_grad.rb +176 -0
  130. data/lib/toy/train/toy_gguf_fuse.rb +428 -0
  131. data/lib/toy/train/toy_gguf_writer.rb +100 -0
  132. data/lib/toy/train/toy_lr_schedule.rb +39 -0
  133. data/lib/toy/train/toy_sample.rb +125 -0
  134. data/lib/toy/train/toy_trainer.rb +86 -0
  135. data/lib/toy/train/training.rb +160 -0
  136. data/lib/toy/version.rb +11 -0
  137. data/lib/toy.rb +902 -0
  138. data/prep/progress +118 -0
  139. data/prep/quietly +64 -0
  140. data/sig/toy.rbs +397 -0
  141. data/sig/toy_compute.rbs +450 -0
  142. data/spinel-ext.json +122 -0
  143. data/tinynn/Makefile +71 -0
  144. data/tinynn/tinynn_backend_cuda.c +99 -0
  145. data/tinynn/tinynn_backend_metal.m +75 -0
  146. data/tinynn/tinynn_events.c +122 -0
  147. data/tinynn/tinynn_events.h +83 -0
  148. data/tinynn/tinynn_ggml.c +2460 -0
  149. data/tinynn/tinynn_ggml.h +545 -0
  150. data/tinynn/tinynn_gguf.c +783 -0
  151. data/tinynn/tinynn_gguf.h +167 -0
  152. data/tinynn/tinynn_trace.c +180 -0
  153. data/tinynn/tinynn_trace.h +85 -0
  154. data/vendor/ggml/AUTHORS +335 -0
  155. data/vendor/ggml/CMakeLists.txt +505 -0
  156. data/vendor/ggml/CONTRIBUTING.md +3 -0
  157. data/vendor/ggml/LICENSE +21 -0
  158. data/vendor/ggml/README.md +50 -0
  159. data/vendor/ggml/ci/run.sh +395 -0
  160. data/vendor/ggml/cmake/FindNCCL.cmake +36 -0
  161. data/vendor/ggml/cmake/GitVars.cmake +22 -0
  162. data/vendor/ggml/cmake/common.cmake +50 -0
  163. data/vendor/ggml/cmake/ggml-config.cmake.in +191 -0
  164. data/vendor/ggml/docs/gguf.md +828 -0
  165. data/vendor/ggml/examples/CMakeLists.txt +34 -0
  166. data/vendor/ggml/examples/common-ggml.cpp +244 -0
  167. data/vendor/ggml/examples/common-ggml.h +18 -0
  168. data/vendor/ggml/examples/common.cpp +675 -0
  169. data/vendor/ggml/examples/common.h +322 -0
  170. data/vendor/ggml/examples/gpt-2/CMakeLists.txt +32 -0
  171. data/vendor/ggml/examples/gpt-2/README.md +225 -0
  172. data/vendor/ggml/examples/gpt-2/convert-cerebras-to-ggml.py +183 -0
  173. data/vendor/ggml/examples/gpt-2/convert-ckpt-to-ggml.py +159 -0
  174. data/vendor/ggml/examples/gpt-2/convert-h5-to-ggml.py +195 -0
  175. data/vendor/ggml/examples/gpt-2/download-ggml-model.sh +69 -0
  176. data/vendor/ggml/examples/gpt-2/download-model.sh +48 -0
  177. data/vendor/ggml/examples/gpt-2/main-alloc.cpp +880 -0
  178. data/vendor/ggml/examples/gpt-2/main-backend.cpp +946 -0
  179. data/vendor/ggml/examples/gpt-2/main-batched.cpp +1210 -0
  180. data/vendor/ggml/examples/gpt-2/main-ctx.cpp +840 -0
  181. data/vendor/ggml/examples/gpt-2/main-sched.cpp +1079 -0
  182. data/vendor/ggml/examples/gpt-2/quantize.cpp +184 -0
  183. data/vendor/ggml/examples/gpt-j/CMakeLists.txt +13 -0
  184. data/vendor/ggml/examples/gpt-j/README.md +239 -0
  185. data/vendor/ggml/examples/gpt-j/convert-h5-to-ggml.py +173 -0
  186. data/vendor/ggml/examples/gpt-j/download-ggml-model.sh +69 -0
  187. data/vendor/ggml/examples/gpt-j/download-model.sh +11 -0
  188. data/vendor/ggml/examples/gpt-j/main.cpp +755 -0
  189. data/vendor/ggml/examples/gpt-j/quantize.cpp +182 -0
  190. data/vendor/ggml/examples/magika/CMakeLists.txt +17 -0
  191. data/vendor/ggml/examples/magika/README.md +23 -0
  192. data/vendor/ggml/examples/magika/convert.py +32 -0
  193. data/vendor/ggml/examples/magika/main.cpp +374 -0
  194. data/vendor/ggml/examples/mnist/CMakeLists.txt +58 -0
  195. data/vendor/ggml/examples/mnist/README.md +206 -0
  196. data/vendor/ggml/examples/mnist/mnist-common.cpp +496 -0
  197. data/vendor/ggml/examples/mnist/mnist-common.h +166 -0
  198. data/vendor/ggml/examples/mnist/mnist-eval.cpp +67 -0
  199. data/vendor/ggml/examples/mnist/mnist-train-cnn.py +91 -0
  200. data/vendor/ggml/examples/mnist/mnist-train-fc.py +131 -0
  201. data/vendor/ggml/examples/mnist/mnist-train.cpp +39 -0
  202. data/vendor/ggml/examples/mnist/server.py +36 -0
  203. data/vendor/ggml/examples/mnist/web/index.html +178 -0
  204. data/vendor/ggml/examples/perf-metal/CMakeLists.txt +7 -0
  205. data/vendor/ggml/examples/perf-metal/perf-metal.cpp +152 -0
  206. data/vendor/ggml/examples/prompts/dolly-v2.txt +100 -0
  207. data/vendor/ggml/examples/prompts/gpt-2-chinese.txt +1 -0
  208. data/vendor/ggml/examples/prompts/gpt-2.txt +100 -0
  209. data/vendor/ggml/examples/prompts/gpt-j.txt +100 -0
  210. data/vendor/ggml/examples/prompts/gpt-neox-japanese.txt +1 -0
  211. data/vendor/ggml/examples/prompts/gpt-neox.txt +100 -0
  212. data/vendor/ggml/examples/prompts/polyglot-ko.txt +3 -0
  213. data/vendor/ggml/examples/prompts/replit.txt +100 -0
  214. data/vendor/ggml/examples/prompts/starcoder.txt +100 -0
  215. data/vendor/ggml/examples/prompts/test-cases.txt +110 -0
  216. data/vendor/ggml/examples/prompts/tokenize_huggingface.py +65 -0
  217. data/vendor/ggml/examples/prompts/whisper.txt +100 -0
  218. data/vendor/ggml/examples/python/README.md +115 -0
  219. data/vendor/ggml/examples/python/api.h +14 -0
  220. data/vendor/ggml/examples/python/example_add_quant.py +25 -0
  221. data/vendor/ggml/examples/python/example_test_all_quants.py +68 -0
  222. data/vendor/ggml/examples/python/ggml/__init__.py +58 -0
  223. data/vendor/ggml/examples/python/ggml/__init__.pyi +2406 -0
  224. data/vendor/ggml/examples/python/ggml/cffi.py +11 -0
  225. data/vendor/ggml/examples/python/ggml/ffi/__init__.pyi +7 -0
  226. data/vendor/ggml/examples/python/ggml/utils.py +182 -0
  227. data/vendor/ggml/examples/python/regenerate.py +42 -0
  228. data/vendor/ggml/examples/python/stubs.py +128 -0
  229. data/vendor/ggml/examples/python/test_tensor.py +258 -0
  230. data/vendor/ggml/examples/sam/CMakeLists.txt +13 -0
  231. data/vendor/ggml/examples/sam/README.md +95 -0
  232. data/vendor/ggml/examples/sam/convert-pth-to-ggml.py +147 -0
  233. data/vendor/ggml/examples/sam/example.jpg +0 -0
  234. data/vendor/ggml/examples/sam/sam.cpp +2370 -0
  235. data/vendor/ggml/examples/simple/CMakeLists.txt +21 -0
  236. data/vendor/ggml/examples/simple/README.md +61 -0
  237. data/vendor/ggml/examples/simple/simple-backend.cpp +153 -0
  238. data/vendor/ggml/examples/simple/simple-ctx.cpp +127 -0
  239. data/vendor/ggml/examples/stb_image.h +7987 -0
  240. data/vendor/ggml/examples/stb_image_write.h +1724 -0
  241. data/vendor/ggml/examples/test-cmake/CMakeLists.txt +10 -0
  242. data/vendor/ggml/examples/test-cmake/README.md +3 -0
  243. data/vendor/ggml/examples/test-cmake/test-cmake.cpp +6 -0
  244. data/vendor/ggml/examples/yolo/CMakeLists.txt +6 -0
  245. data/vendor/ggml/examples/yolo/README.md +59 -0
  246. data/vendor/ggml/examples/yolo/convert-yolov3-tiny.py +53 -0
  247. data/vendor/ggml/examples/yolo/data/coco.names +80 -0
  248. data/vendor/ggml/examples/yolo/data/labels/100_0.png +0 -0
  249. data/vendor/ggml/examples/yolo/data/labels/100_1.png +0 -0
  250. data/vendor/ggml/examples/yolo/data/labels/100_2.png +0 -0
  251. data/vendor/ggml/examples/yolo/data/labels/100_3.png +0 -0
  252. data/vendor/ggml/examples/yolo/data/labels/100_4.png +0 -0
  253. data/vendor/ggml/examples/yolo/data/labels/100_5.png +0 -0
  254. data/vendor/ggml/examples/yolo/data/labels/100_6.png +0 -0
  255. data/vendor/ggml/examples/yolo/data/labels/100_7.png +0 -0
  256. data/vendor/ggml/examples/yolo/data/labels/101_0.png +0 -0
  257. data/vendor/ggml/examples/yolo/data/labels/101_1.png +0 -0
  258. data/vendor/ggml/examples/yolo/data/labels/101_2.png +0 -0
  259. data/vendor/ggml/examples/yolo/data/labels/101_3.png +0 -0
  260. data/vendor/ggml/examples/yolo/data/labels/101_4.png +0 -0
  261. data/vendor/ggml/examples/yolo/data/labels/101_5.png +0 -0
  262. data/vendor/ggml/examples/yolo/data/labels/101_6.png +0 -0
  263. data/vendor/ggml/examples/yolo/data/labels/101_7.png +0 -0
  264. data/vendor/ggml/examples/yolo/data/labels/102_0.png +0 -0
  265. data/vendor/ggml/examples/yolo/data/labels/102_1.png +0 -0
  266. data/vendor/ggml/examples/yolo/data/labels/102_2.png +0 -0
  267. data/vendor/ggml/examples/yolo/data/labels/102_3.png +0 -0
  268. data/vendor/ggml/examples/yolo/data/labels/102_4.png +0 -0
  269. data/vendor/ggml/examples/yolo/data/labels/102_5.png +0 -0
  270. data/vendor/ggml/examples/yolo/data/labels/102_6.png +0 -0
  271. data/vendor/ggml/examples/yolo/data/labels/102_7.png +0 -0
  272. data/vendor/ggml/examples/yolo/data/labels/103_0.png +0 -0
  273. data/vendor/ggml/examples/yolo/data/labels/103_1.png +0 -0
  274. data/vendor/ggml/examples/yolo/data/labels/103_2.png +0 -0
  275. data/vendor/ggml/examples/yolo/data/labels/103_3.png +0 -0
  276. data/vendor/ggml/examples/yolo/data/labels/103_4.png +0 -0
  277. data/vendor/ggml/examples/yolo/data/labels/103_5.png +0 -0
  278. data/vendor/ggml/examples/yolo/data/labels/103_6.png +0 -0
  279. data/vendor/ggml/examples/yolo/data/labels/103_7.png +0 -0
  280. data/vendor/ggml/examples/yolo/data/labels/104_0.png +0 -0
  281. data/vendor/ggml/examples/yolo/data/labels/104_1.png +0 -0
  282. data/vendor/ggml/examples/yolo/data/labels/104_2.png +0 -0
  283. data/vendor/ggml/examples/yolo/data/labels/104_3.png +0 -0
  284. data/vendor/ggml/examples/yolo/data/labels/104_4.png +0 -0
  285. data/vendor/ggml/examples/yolo/data/labels/104_5.png +0 -0
  286. data/vendor/ggml/examples/yolo/data/labels/104_6.png +0 -0
  287. data/vendor/ggml/examples/yolo/data/labels/104_7.png +0 -0
  288. data/vendor/ggml/examples/yolo/data/labels/105_0.png +0 -0
  289. data/vendor/ggml/examples/yolo/data/labels/105_1.png +0 -0
  290. data/vendor/ggml/examples/yolo/data/labels/105_2.png +0 -0
  291. data/vendor/ggml/examples/yolo/data/labels/105_3.png +0 -0
  292. data/vendor/ggml/examples/yolo/data/labels/105_4.png +0 -0
  293. data/vendor/ggml/examples/yolo/data/labels/105_5.png +0 -0
  294. data/vendor/ggml/examples/yolo/data/labels/105_6.png +0 -0
  295. data/vendor/ggml/examples/yolo/data/labels/105_7.png +0 -0
  296. data/vendor/ggml/examples/yolo/data/labels/106_0.png +0 -0
  297. data/vendor/ggml/examples/yolo/data/labels/106_1.png +0 -0
  298. data/vendor/ggml/examples/yolo/data/labels/106_2.png +0 -0
  299. data/vendor/ggml/examples/yolo/data/labels/106_3.png +0 -0
  300. data/vendor/ggml/examples/yolo/data/labels/106_4.png +0 -0
  301. data/vendor/ggml/examples/yolo/data/labels/106_5.png +0 -0
  302. data/vendor/ggml/examples/yolo/data/labels/106_6.png +0 -0
  303. data/vendor/ggml/examples/yolo/data/labels/106_7.png +0 -0
  304. data/vendor/ggml/examples/yolo/data/labels/107_0.png +0 -0
  305. data/vendor/ggml/examples/yolo/data/labels/107_1.png +0 -0
  306. data/vendor/ggml/examples/yolo/data/labels/107_2.png +0 -0
  307. data/vendor/ggml/examples/yolo/data/labels/107_3.png +0 -0
  308. data/vendor/ggml/examples/yolo/data/labels/107_4.png +0 -0
  309. data/vendor/ggml/examples/yolo/data/labels/107_5.png +0 -0
  310. data/vendor/ggml/examples/yolo/data/labels/107_6.png +0 -0
  311. data/vendor/ggml/examples/yolo/data/labels/107_7.png +0 -0
  312. data/vendor/ggml/examples/yolo/data/labels/108_0.png +0 -0
  313. data/vendor/ggml/examples/yolo/data/labels/108_1.png +0 -0
  314. data/vendor/ggml/examples/yolo/data/labels/108_2.png +0 -0
  315. data/vendor/ggml/examples/yolo/data/labels/108_3.png +0 -0
  316. data/vendor/ggml/examples/yolo/data/labels/108_4.png +0 -0
  317. data/vendor/ggml/examples/yolo/data/labels/108_5.png +0 -0
  318. data/vendor/ggml/examples/yolo/data/labels/108_6.png +0 -0
  319. data/vendor/ggml/examples/yolo/data/labels/108_7.png +0 -0
  320. data/vendor/ggml/examples/yolo/data/labels/109_0.png +0 -0
  321. data/vendor/ggml/examples/yolo/data/labels/109_1.png +0 -0
  322. data/vendor/ggml/examples/yolo/data/labels/109_2.png +0 -0
  323. data/vendor/ggml/examples/yolo/data/labels/109_3.png +0 -0
  324. data/vendor/ggml/examples/yolo/data/labels/109_4.png +0 -0
  325. data/vendor/ggml/examples/yolo/data/labels/109_5.png +0 -0
  326. data/vendor/ggml/examples/yolo/data/labels/109_6.png +0 -0
  327. data/vendor/ggml/examples/yolo/data/labels/109_7.png +0 -0
  328. data/vendor/ggml/examples/yolo/data/labels/110_0.png +0 -0
  329. data/vendor/ggml/examples/yolo/data/labels/110_1.png +0 -0
  330. data/vendor/ggml/examples/yolo/data/labels/110_2.png +0 -0
  331. data/vendor/ggml/examples/yolo/data/labels/110_3.png +0 -0
  332. data/vendor/ggml/examples/yolo/data/labels/110_4.png +0 -0
  333. data/vendor/ggml/examples/yolo/data/labels/110_5.png +0 -0
  334. data/vendor/ggml/examples/yolo/data/labels/110_6.png +0 -0
  335. data/vendor/ggml/examples/yolo/data/labels/110_7.png +0 -0
  336. data/vendor/ggml/examples/yolo/data/labels/111_0.png +0 -0
  337. data/vendor/ggml/examples/yolo/data/labels/111_1.png +0 -0
  338. data/vendor/ggml/examples/yolo/data/labels/111_2.png +0 -0
  339. data/vendor/ggml/examples/yolo/data/labels/111_3.png +0 -0
  340. data/vendor/ggml/examples/yolo/data/labels/111_4.png +0 -0
  341. data/vendor/ggml/examples/yolo/data/labels/111_5.png +0 -0
  342. data/vendor/ggml/examples/yolo/data/labels/111_6.png +0 -0
  343. data/vendor/ggml/examples/yolo/data/labels/111_7.png +0 -0
  344. data/vendor/ggml/examples/yolo/data/labels/112_0.png +0 -0
  345. data/vendor/ggml/examples/yolo/data/labels/112_1.png +0 -0
  346. data/vendor/ggml/examples/yolo/data/labels/112_2.png +0 -0
  347. data/vendor/ggml/examples/yolo/data/labels/112_3.png +0 -0
  348. data/vendor/ggml/examples/yolo/data/labels/112_4.png +0 -0
  349. data/vendor/ggml/examples/yolo/data/labels/112_5.png +0 -0
  350. data/vendor/ggml/examples/yolo/data/labels/112_6.png +0 -0
  351. data/vendor/ggml/examples/yolo/data/labels/112_7.png +0 -0
  352. data/vendor/ggml/examples/yolo/data/labels/113_0.png +0 -0
  353. data/vendor/ggml/examples/yolo/data/labels/113_1.png +0 -0
  354. data/vendor/ggml/examples/yolo/data/labels/113_2.png +0 -0
  355. data/vendor/ggml/examples/yolo/data/labels/113_3.png +0 -0
  356. data/vendor/ggml/examples/yolo/data/labels/113_4.png +0 -0
  357. data/vendor/ggml/examples/yolo/data/labels/113_5.png +0 -0
  358. data/vendor/ggml/examples/yolo/data/labels/113_6.png +0 -0
  359. data/vendor/ggml/examples/yolo/data/labels/113_7.png +0 -0
  360. data/vendor/ggml/examples/yolo/data/labels/114_0.png +0 -0
  361. data/vendor/ggml/examples/yolo/data/labels/114_1.png +0 -0
  362. data/vendor/ggml/examples/yolo/data/labels/114_2.png +0 -0
  363. data/vendor/ggml/examples/yolo/data/labels/114_3.png +0 -0
  364. data/vendor/ggml/examples/yolo/data/labels/114_4.png +0 -0
  365. data/vendor/ggml/examples/yolo/data/labels/114_5.png +0 -0
  366. data/vendor/ggml/examples/yolo/data/labels/114_6.png +0 -0
  367. data/vendor/ggml/examples/yolo/data/labels/114_7.png +0 -0
  368. data/vendor/ggml/examples/yolo/data/labels/115_0.png +0 -0
  369. data/vendor/ggml/examples/yolo/data/labels/115_1.png +0 -0
  370. data/vendor/ggml/examples/yolo/data/labels/115_2.png +0 -0
  371. data/vendor/ggml/examples/yolo/data/labels/115_3.png +0 -0
  372. data/vendor/ggml/examples/yolo/data/labels/115_4.png +0 -0
  373. data/vendor/ggml/examples/yolo/data/labels/115_5.png +0 -0
  374. data/vendor/ggml/examples/yolo/data/labels/115_6.png +0 -0
  375. data/vendor/ggml/examples/yolo/data/labels/115_7.png +0 -0
  376. data/vendor/ggml/examples/yolo/data/labels/116_0.png +0 -0
  377. data/vendor/ggml/examples/yolo/data/labels/116_1.png +0 -0
  378. data/vendor/ggml/examples/yolo/data/labels/116_2.png +0 -0
  379. data/vendor/ggml/examples/yolo/data/labels/116_3.png +0 -0
  380. data/vendor/ggml/examples/yolo/data/labels/116_4.png +0 -0
  381. data/vendor/ggml/examples/yolo/data/labels/116_5.png +0 -0
  382. data/vendor/ggml/examples/yolo/data/labels/116_6.png +0 -0
  383. data/vendor/ggml/examples/yolo/data/labels/116_7.png +0 -0
  384. data/vendor/ggml/examples/yolo/data/labels/117_0.png +0 -0
  385. data/vendor/ggml/examples/yolo/data/labels/117_1.png +0 -0
  386. data/vendor/ggml/examples/yolo/data/labels/117_2.png +0 -0
  387. data/vendor/ggml/examples/yolo/data/labels/117_3.png +0 -0
  388. data/vendor/ggml/examples/yolo/data/labels/117_4.png +0 -0
  389. data/vendor/ggml/examples/yolo/data/labels/117_5.png +0 -0
  390. data/vendor/ggml/examples/yolo/data/labels/117_6.png +0 -0
  391. data/vendor/ggml/examples/yolo/data/labels/117_7.png +0 -0
  392. data/vendor/ggml/examples/yolo/data/labels/118_0.png +0 -0
  393. data/vendor/ggml/examples/yolo/data/labels/118_1.png +0 -0
  394. data/vendor/ggml/examples/yolo/data/labels/118_2.png +0 -0
  395. data/vendor/ggml/examples/yolo/data/labels/118_3.png +0 -0
  396. data/vendor/ggml/examples/yolo/data/labels/118_4.png +0 -0
  397. data/vendor/ggml/examples/yolo/data/labels/118_5.png +0 -0
  398. data/vendor/ggml/examples/yolo/data/labels/118_6.png +0 -0
  399. data/vendor/ggml/examples/yolo/data/labels/118_7.png +0 -0
  400. data/vendor/ggml/examples/yolo/data/labels/119_0.png +0 -0
  401. data/vendor/ggml/examples/yolo/data/labels/119_1.png +0 -0
  402. data/vendor/ggml/examples/yolo/data/labels/119_2.png +0 -0
  403. data/vendor/ggml/examples/yolo/data/labels/119_3.png +0 -0
  404. data/vendor/ggml/examples/yolo/data/labels/119_4.png +0 -0
  405. data/vendor/ggml/examples/yolo/data/labels/119_5.png +0 -0
  406. data/vendor/ggml/examples/yolo/data/labels/119_6.png +0 -0
  407. data/vendor/ggml/examples/yolo/data/labels/119_7.png +0 -0
  408. data/vendor/ggml/examples/yolo/data/labels/120_0.png +0 -0
  409. data/vendor/ggml/examples/yolo/data/labels/120_1.png +0 -0
  410. data/vendor/ggml/examples/yolo/data/labels/120_2.png +0 -0
  411. data/vendor/ggml/examples/yolo/data/labels/120_3.png +0 -0
  412. data/vendor/ggml/examples/yolo/data/labels/120_4.png +0 -0
  413. data/vendor/ggml/examples/yolo/data/labels/120_5.png +0 -0
  414. data/vendor/ggml/examples/yolo/data/labels/120_6.png +0 -0
  415. data/vendor/ggml/examples/yolo/data/labels/120_7.png +0 -0
  416. data/vendor/ggml/examples/yolo/data/labels/121_0.png +0 -0
  417. data/vendor/ggml/examples/yolo/data/labels/121_1.png +0 -0
  418. data/vendor/ggml/examples/yolo/data/labels/121_2.png +0 -0
  419. data/vendor/ggml/examples/yolo/data/labels/121_3.png +0 -0
  420. data/vendor/ggml/examples/yolo/data/labels/121_4.png +0 -0
  421. data/vendor/ggml/examples/yolo/data/labels/121_5.png +0 -0
  422. data/vendor/ggml/examples/yolo/data/labels/121_6.png +0 -0
  423. data/vendor/ggml/examples/yolo/data/labels/121_7.png +0 -0
  424. data/vendor/ggml/examples/yolo/data/labels/122_0.png +0 -0
  425. data/vendor/ggml/examples/yolo/data/labels/122_1.png +0 -0
  426. data/vendor/ggml/examples/yolo/data/labels/122_2.png +0 -0
  427. data/vendor/ggml/examples/yolo/data/labels/122_3.png +0 -0
  428. data/vendor/ggml/examples/yolo/data/labels/122_4.png +0 -0
  429. data/vendor/ggml/examples/yolo/data/labels/122_5.png +0 -0
  430. data/vendor/ggml/examples/yolo/data/labels/122_6.png +0 -0
  431. data/vendor/ggml/examples/yolo/data/labels/122_7.png +0 -0
  432. data/vendor/ggml/examples/yolo/data/labels/123_0.png +0 -0
  433. data/vendor/ggml/examples/yolo/data/labels/123_1.png +0 -0
  434. data/vendor/ggml/examples/yolo/data/labels/123_2.png +0 -0
  435. data/vendor/ggml/examples/yolo/data/labels/123_3.png +0 -0
  436. data/vendor/ggml/examples/yolo/data/labels/123_4.png +0 -0
  437. data/vendor/ggml/examples/yolo/data/labels/123_5.png +0 -0
  438. data/vendor/ggml/examples/yolo/data/labels/123_6.png +0 -0
  439. data/vendor/ggml/examples/yolo/data/labels/123_7.png +0 -0
  440. data/vendor/ggml/examples/yolo/data/labels/124_0.png +0 -0
  441. data/vendor/ggml/examples/yolo/data/labels/124_1.png +0 -0
  442. data/vendor/ggml/examples/yolo/data/labels/124_2.png +0 -0
  443. data/vendor/ggml/examples/yolo/data/labels/124_3.png +0 -0
  444. data/vendor/ggml/examples/yolo/data/labels/124_4.png +0 -0
  445. data/vendor/ggml/examples/yolo/data/labels/124_5.png +0 -0
  446. data/vendor/ggml/examples/yolo/data/labels/124_6.png +0 -0
  447. data/vendor/ggml/examples/yolo/data/labels/124_7.png +0 -0
  448. data/vendor/ggml/examples/yolo/data/labels/125_0.png +0 -0
  449. data/vendor/ggml/examples/yolo/data/labels/125_1.png +0 -0
  450. data/vendor/ggml/examples/yolo/data/labels/125_2.png +0 -0
  451. data/vendor/ggml/examples/yolo/data/labels/125_3.png +0 -0
  452. data/vendor/ggml/examples/yolo/data/labels/125_4.png +0 -0
  453. data/vendor/ggml/examples/yolo/data/labels/125_5.png +0 -0
  454. data/vendor/ggml/examples/yolo/data/labels/125_6.png +0 -0
  455. data/vendor/ggml/examples/yolo/data/labels/125_7.png +0 -0
  456. data/vendor/ggml/examples/yolo/data/labels/126_0.png +0 -0
  457. data/vendor/ggml/examples/yolo/data/labels/126_1.png +0 -0
  458. data/vendor/ggml/examples/yolo/data/labels/126_2.png +0 -0
  459. data/vendor/ggml/examples/yolo/data/labels/126_3.png +0 -0
  460. data/vendor/ggml/examples/yolo/data/labels/126_4.png +0 -0
  461. data/vendor/ggml/examples/yolo/data/labels/126_5.png +0 -0
  462. data/vendor/ggml/examples/yolo/data/labels/126_6.png +0 -0
  463. data/vendor/ggml/examples/yolo/data/labels/126_7.png +0 -0
  464. data/vendor/ggml/examples/yolo/data/labels/32_0.png +0 -0
  465. data/vendor/ggml/examples/yolo/data/labels/32_1.png +0 -0
  466. data/vendor/ggml/examples/yolo/data/labels/32_2.png +0 -0
  467. data/vendor/ggml/examples/yolo/data/labels/32_3.png +0 -0
  468. data/vendor/ggml/examples/yolo/data/labels/32_4.png +0 -0
  469. data/vendor/ggml/examples/yolo/data/labels/32_5.png +0 -0
  470. data/vendor/ggml/examples/yolo/data/labels/32_6.png +0 -0
  471. data/vendor/ggml/examples/yolo/data/labels/32_7.png +0 -0
  472. data/vendor/ggml/examples/yolo/data/labels/33_0.png +0 -0
  473. data/vendor/ggml/examples/yolo/data/labels/33_1.png +0 -0
  474. data/vendor/ggml/examples/yolo/data/labels/33_2.png +0 -0
  475. data/vendor/ggml/examples/yolo/data/labels/33_3.png +0 -0
  476. data/vendor/ggml/examples/yolo/data/labels/33_4.png +0 -0
  477. data/vendor/ggml/examples/yolo/data/labels/33_5.png +0 -0
  478. data/vendor/ggml/examples/yolo/data/labels/33_6.png +0 -0
  479. data/vendor/ggml/examples/yolo/data/labels/33_7.png +0 -0
  480. data/vendor/ggml/examples/yolo/data/labels/34_0.png +0 -0
  481. data/vendor/ggml/examples/yolo/data/labels/34_1.png +0 -0
  482. data/vendor/ggml/examples/yolo/data/labels/34_2.png +0 -0
  483. data/vendor/ggml/examples/yolo/data/labels/34_3.png +0 -0
  484. data/vendor/ggml/examples/yolo/data/labels/34_4.png +0 -0
  485. data/vendor/ggml/examples/yolo/data/labels/34_5.png +0 -0
  486. data/vendor/ggml/examples/yolo/data/labels/34_6.png +0 -0
  487. data/vendor/ggml/examples/yolo/data/labels/34_7.png +0 -0
  488. data/vendor/ggml/examples/yolo/data/labels/35_0.png +0 -0
  489. data/vendor/ggml/examples/yolo/data/labels/35_1.png +0 -0
  490. data/vendor/ggml/examples/yolo/data/labels/35_2.png +0 -0
  491. data/vendor/ggml/examples/yolo/data/labels/35_3.png +0 -0
  492. data/vendor/ggml/examples/yolo/data/labels/35_4.png +0 -0
  493. data/vendor/ggml/examples/yolo/data/labels/35_5.png +0 -0
  494. data/vendor/ggml/examples/yolo/data/labels/35_6.png +0 -0
  495. data/vendor/ggml/examples/yolo/data/labels/35_7.png +0 -0
  496. data/vendor/ggml/examples/yolo/data/labels/36_0.png +0 -0
  497. data/vendor/ggml/examples/yolo/data/labels/36_1.png +0 -0
  498. data/vendor/ggml/examples/yolo/data/labels/36_2.png +0 -0
  499. data/vendor/ggml/examples/yolo/data/labels/36_3.png +0 -0
  500. data/vendor/ggml/examples/yolo/data/labels/36_4.png +0 -0
  501. data/vendor/ggml/examples/yolo/data/labels/36_5.png +0 -0
  502. data/vendor/ggml/examples/yolo/data/labels/36_6.png +0 -0
  503. data/vendor/ggml/examples/yolo/data/labels/36_7.png +0 -0
  504. data/vendor/ggml/examples/yolo/data/labels/37_0.png +0 -0
  505. data/vendor/ggml/examples/yolo/data/labels/37_1.png +0 -0
  506. data/vendor/ggml/examples/yolo/data/labels/37_2.png +0 -0
  507. data/vendor/ggml/examples/yolo/data/labels/37_3.png +0 -0
  508. data/vendor/ggml/examples/yolo/data/labels/37_4.png +0 -0
  509. data/vendor/ggml/examples/yolo/data/labels/37_5.png +0 -0
  510. data/vendor/ggml/examples/yolo/data/labels/37_6.png +0 -0
  511. data/vendor/ggml/examples/yolo/data/labels/37_7.png +0 -0
  512. data/vendor/ggml/examples/yolo/data/labels/38_0.png +0 -0
  513. data/vendor/ggml/examples/yolo/data/labels/38_1.png +0 -0
  514. data/vendor/ggml/examples/yolo/data/labels/38_2.png +0 -0
  515. data/vendor/ggml/examples/yolo/data/labels/38_3.png +0 -0
  516. data/vendor/ggml/examples/yolo/data/labels/38_4.png +0 -0
  517. data/vendor/ggml/examples/yolo/data/labels/38_5.png +0 -0
  518. data/vendor/ggml/examples/yolo/data/labels/38_6.png +0 -0
  519. data/vendor/ggml/examples/yolo/data/labels/38_7.png +0 -0
  520. data/vendor/ggml/examples/yolo/data/labels/39_0.png +0 -0
  521. data/vendor/ggml/examples/yolo/data/labels/39_1.png +0 -0
  522. data/vendor/ggml/examples/yolo/data/labels/39_2.png +0 -0
  523. data/vendor/ggml/examples/yolo/data/labels/39_3.png +0 -0
  524. data/vendor/ggml/examples/yolo/data/labels/39_4.png +0 -0
  525. data/vendor/ggml/examples/yolo/data/labels/39_5.png +0 -0
  526. data/vendor/ggml/examples/yolo/data/labels/39_6.png +0 -0
  527. data/vendor/ggml/examples/yolo/data/labels/39_7.png +0 -0
  528. data/vendor/ggml/examples/yolo/data/labels/40_0.png +0 -0
  529. data/vendor/ggml/examples/yolo/data/labels/40_1.png +0 -0
  530. data/vendor/ggml/examples/yolo/data/labels/40_2.png +0 -0
  531. data/vendor/ggml/examples/yolo/data/labels/40_3.png +0 -0
  532. data/vendor/ggml/examples/yolo/data/labels/40_4.png +0 -0
  533. data/vendor/ggml/examples/yolo/data/labels/40_5.png +0 -0
  534. data/vendor/ggml/examples/yolo/data/labels/40_6.png +0 -0
  535. data/vendor/ggml/examples/yolo/data/labels/40_7.png +0 -0
  536. data/vendor/ggml/examples/yolo/data/labels/41_0.png +0 -0
  537. data/vendor/ggml/examples/yolo/data/labels/41_1.png +0 -0
  538. data/vendor/ggml/examples/yolo/data/labels/41_2.png +0 -0
  539. data/vendor/ggml/examples/yolo/data/labels/41_3.png +0 -0
  540. data/vendor/ggml/examples/yolo/data/labels/41_4.png +0 -0
  541. data/vendor/ggml/examples/yolo/data/labels/41_5.png +0 -0
  542. data/vendor/ggml/examples/yolo/data/labels/41_6.png +0 -0
  543. data/vendor/ggml/examples/yolo/data/labels/41_7.png +0 -0
  544. data/vendor/ggml/examples/yolo/data/labels/42_0.png +0 -0
  545. data/vendor/ggml/examples/yolo/data/labels/42_1.png +0 -0
  546. data/vendor/ggml/examples/yolo/data/labels/42_2.png +0 -0
  547. data/vendor/ggml/examples/yolo/data/labels/42_3.png +0 -0
  548. data/vendor/ggml/examples/yolo/data/labels/42_4.png +0 -0
  549. data/vendor/ggml/examples/yolo/data/labels/42_5.png +0 -0
  550. data/vendor/ggml/examples/yolo/data/labels/42_6.png +0 -0
  551. data/vendor/ggml/examples/yolo/data/labels/42_7.png +0 -0
  552. data/vendor/ggml/examples/yolo/data/labels/43_0.png +0 -0
  553. data/vendor/ggml/examples/yolo/data/labels/43_1.png +0 -0
  554. data/vendor/ggml/examples/yolo/data/labels/43_2.png +0 -0
  555. data/vendor/ggml/examples/yolo/data/labels/43_3.png +0 -0
  556. data/vendor/ggml/examples/yolo/data/labels/43_4.png +0 -0
  557. data/vendor/ggml/examples/yolo/data/labels/43_5.png +0 -0
  558. data/vendor/ggml/examples/yolo/data/labels/43_6.png +0 -0
  559. data/vendor/ggml/examples/yolo/data/labels/43_7.png +0 -0
  560. data/vendor/ggml/examples/yolo/data/labels/44_0.png +0 -0
  561. data/vendor/ggml/examples/yolo/data/labels/44_1.png +0 -0
  562. data/vendor/ggml/examples/yolo/data/labels/44_2.png +0 -0
  563. data/vendor/ggml/examples/yolo/data/labels/44_3.png +0 -0
  564. data/vendor/ggml/examples/yolo/data/labels/44_4.png +0 -0
  565. data/vendor/ggml/examples/yolo/data/labels/44_5.png +0 -0
  566. data/vendor/ggml/examples/yolo/data/labels/44_6.png +0 -0
  567. data/vendor/ggml/examples/yolo/data/labels/44_7.png +0 -0
  568. data/vendor/ggml/examples/yolo/data/labels/45_0.png +0 -0
  569. data/vendor/ggml/examples/yolo/data/labels/45_1.png +0 -0
  570. data/vendor/ggml/examples/yolo/data/labels/45_2.png +0 -0
  571. data/vendor/ggml/examples/yolo/data/labels/45_3.png +0 -0
  572. data/vendor/ggml/examples/yolo/data/labels/45_4.png +0 -0
  573. data/vendor/ggml/examples/yolo/data/labels/45_5.png +0 -0
  574. data/vendor/ggml/examples/yolo/data/labels/45_6.png +0 -0
  575. data/vendor/ggml/examples/yolo/data/labels/45_7.png +0 -0
  576. data/vendor/ggml/examples/yolo/data/labels/46_0.png +0 -0
  577. data/vendor/ggml/examples/yolo/data/labels/46_1.png +0 -0
  578. data/vendor/ggml/examples/yolo/data/labels/46_2.png +0 -0
  579. data/vendor/ggml/examples/yolo/data/labels/46_3.png +0 -0
  580. data/vendor/ggml/examples/yolo/data/labels/46_4.png +0 -0
  581. data/vendor/ggml/examples/yolo/data/labels/46_5.png +0 -0
  582. data/vendor/ggml/examples/yolo/data/labels/46_6.png +0 -0
  583. data/vendor/ggml/examples/yolo/data/labels/46_7.png +0 -0
  584. data/vendor/ggml/examples/yolo/data/labels/47_0.png +0 -0
  585. data/vendor/ggml/examples/yolo/data/labels/47_1.png +0 -0
  586. data/vendor/ggml/examples/yolo/data/labels/47_2.png +0 -0
  587. data/vendor/ggml/examples/yolo/data/labels/47_3.png +0 -0
  588. data/vendor/ggml/examples/yolo/data/labels/47_4.png +0 -0
  589. data/vendor/ggml/examples/yolo/data/labels/47_5.png +0 -0
  590. data/vendor/ggml/examples/yolo/data/labels/47_6.png +0 -0
  591. data/vendor/ggml/examples/yolo/data/labels/47_7.png +0 -0
  592. data/vendor/ggml/examples/yolo/data/labels/48_0.png +0 -0
  593. data/vendor/ggml/examples/yolo/data/labels/48_1.png +0 -0
  594. data/vendor/ggml/examples/yolo/data/labels/48_2.png +0 -0
  595. data/vendor/ggml/examples/yolo/data/labels/48_3.png +0 -0
  596. data/vendor/ggml/examples/yolo/data/labels/48_4.png +0 -0
  597. data/vendor/ggml/examples/yolo/data/labels/48_5.png +0 -0
  598. data/vendor/ggml/examples/yolo/data/labels/48_6.png +0 -0
  599. data/vendor/ggml/examples/yolo/data/labels/48_7.png +0 -0
  600. data/vendor/ggml/examples/yolo/data/labels/49_0.png +0 -0
  601. data/vendor/ggml/examples/yolo/data/labels/49_1.png +0 -0
  602. data/vendor/ggml/examples/yolo/data/labels/49_2.png +0 -0
  603. data/vendor/ggml/examples/yolo/data/labels/49_3.png +0 -0
  604. data/vendor/ggml/examples/yolo/data/labels/49_4.png +0 -0
  605. data/vendor/ggml/examples/yolo/data/labels/49_5.png +0 -0
  606. data/vendor/ggml/examples/yolo/data/labels/49_6.png +0 -0
  607. data/vendor/ggml/examples/yolo/data/labels/49_7.png +0 -0
  608. data/vendor/ggml/examples/yolo/data/labels/50_0.png +0 -0
  609. data/vendor/ggml/examples/yolo/data/labels/50_1.png +0 -0
  610. data/vendor/ggml/examples/yolo/data/labels/50_2.png +0 -0
  611. data/vendor/ggml/examples/yolo/data/labels/50_3.png +0 -0
  612. data/vendor/ggml/examples/yolo/data/labels/50_4.png +0 -0
  613. data/vendor/ggml/examples/yolo/data/labels/50_5.png +0 -0
  614. data/vendor/ggml/examples/yolo/data/labels/50_6.png +0 -0
  615. data/vendor/ggml/examples/yolo/data/labels/50_7.png +0 -0
  616. data/vendor/ggml/examples/yolo/data/labels/51_0.png +0 -0
  617. data/vendor/ggml/examples/yolo/data/labels/51_1.png +0 -0
  618. data/vendor/ggml/examples/yolo/data/labels/51_2.png +0 -0
  619. data/vendor/ggml/examples/yolo/data/labels/51_3.png +0 -0
  620. data/vendor/ggml/examples/yolo/data/labels/51_4.png +0 -0
  621. data/vendor/ggml/examples/yolo/data/labels/51_5.png +0 -0
  622. data/vendor/ggml/examples/yolo/data/labels/51_6.png +0 -0
  623. data/vendor/ggml/examples/yolo/data/labels/51_7.png +0 -0
  624. data/vendor/ggml/examples/yolo/data/labels/52_0.png +0 -0
  625. data/vendor/ggml/examples/yolo/data/labels/52_1.png +0 -0
  626. data/vendor/ggml/examples/yolo/data/labels/52_2.png +0 -0
  627. data/vendor/ggml/examples/yolo/data/labels/52_3.png +0 -0
  628. data/vendor/ggml/examples/yolo/data/labels/52_4.png +0 -0
  629. data/vendor/ggml/examples/yolo/data/labels/52_5.png +0 -0
  630. data/vendor/ggml/examples/yolo/data/labels/52_6.png +0 -0
  631. data/vendor/ggml/examples/yolo/data/labels/52_7.png +0 -0
  632. data/vendor/ggml/examples/yolo/data/labels/53_0.png +0 -0
  633. data/vendor/ggml/examples/yolo/data/labels/53_1.png +0 -0
  634. data/vendor/ggml/examples/yolo/data/labels/53_2.png +0 -0
  635. data/vendor/ggml/examples/yolo/data/labels/53_3.png +0 -0
  636. data/vendor/ggml/examples/yolo/data/labels/53_4.png +0 -0
  637. data/vendor/ggml/examples/yolo/data/labels/53_5.png +0 -0
  638. data/vendor/ggml/examples/yolo/data/labels/53_6.png +0 -0
  639. data/vendor/ggml/examples/yolo/data/labels/53_7.png +0 -0
  640. data/vendor/ggml/examples/yolo/data/labels/54_0.png +0 -0
  641. data/vendor/ggml/examples/yolo/data/labels/54_1.png +0 -0
  642. data/vendor/ggml/examples/yolo/data/labels/54_2.png +0 -0
  643. data/vendor/ggml/examples/yolo/data/labels/54_3.png +0 -0
  644. data/vendor/ggml/examples/yolo/data/labels/54_4.png +0 -0
  645. data/vendor/ggml/examples/yolo/data/labels/54_5.png +0 -0
  646. data/vendor/ggml/examples/yolo/data/labels/54_6.png +0 -0
  647. data/vendor/ggml/examples/yolo/data/labels/54_7.png +0 -0
  648. data/vendor/ggml/examples/yolo/data/labels/55_0.png +0 -0
  649. data/vendor/ggml/examples/yolo/data/labels/55_1.png +0 -0
  650. data/vendor/ggml/examples/yolo/data/labels/55_2.png +0 -0
  651. data/vendor/ggml/examples/yolo/data/labels/55_3.png +0 -0
  652. data/vendor/ggml/examples/yolo/data/labels/55_4.png +0 -0
  653. data/vendor/ggml/examples/yolo/data/labels/55_5.png +0 -0
  654. data/vendor/ggml/examples/yolo/data/labels/55_6.png +0 -0
  655. data/vendor/ggml/examples/yolo/data/labels/55_7.png +0 -0
  656. data/vendor/ggml/examples/yolo/data/labels/56_0.png +0 -0
  657. data/vendor/ggml/examples/yolo/data/labels/56_1.png +0 -0
  658. data/vendor/ggml/examples/yolo/data/labels/56_2.png +0 -0
  659. data/vendor/ggml/examples/yolo/data/labels/56_3.png +0 -0
  660. data/vendor/ggml/examples/yolo/data/labels/56_4.png +0 -0
  661. data/vendor/ggml/examples/yolo/data/labels/56_5.png +0 -0
  662. data/vendor/ggml/examples/yolo/data/labels/56_6.png +0 -0
  663. data/vendor/ggml/examples/yolo/data/labels/56_7.png +0 -0
  664. data/vendor/ggml/examples/yolo/data/labels/57_0.png +0 -0
  665. data/vendor/ggml/examples/yolo/data/labels/57_1.png +0 -0
  666. data/vendor/ggml/examples/yolo/data/labels/57_2.png +0 -0
  667. data/vendor/ggml/examples/yolo/data/labels/57_3.png +0 -0
  668. data/vendor/ggml/examples/yolo/data/labels/57_4.png +0 -0
  669. data/vendor/ggml/examples/yolo/data/labels/57_5.png +0 -0
  670. data/vendor/ggml/examples/yolo/data/labels/57_6.png +0 -0
  671. data/vendor/ggml/examples/yolo/data/labels/57_7.png +0 -0
  672. data/vendor/ggml/examples/yolo/data/labels/58_0.png +0 -0
  673. data/vendor/ggml/examples/yolo/data/labels/58_1.png +0 -0
  674. data/vendor/ggml/examples/yolo/data/labels/58_2.png +0 -0
  675. data/vendor/ggml/examples/yolo/data/labels/58_3.png +0 -0
  676. data/vendor/ggml/examples/yolo/data/labels/58_4.png +0 -0
  677. data/vendor/ggml/examples/yolo/data/labels/58_5.png +0 -0
  678. data/vendor/ggml/examples/yolo/data/labels/58_6.png +0 -0
  679. data/vendor/ggml/examples/yolo/data/labels/58_7.png +0 -0
  680. data/vendor/ggml/examples/yolo/data/labels/59_0.png +0 -0
  681. data/vendor/ggml/examples/yolo/data/labels/59_1.png +0 -0
  682. data/vendor/ggml/examples/yolo/data/labels/59_2.png +0 -0
  683. data/vendor/ggml/examples/yolo/data/labels/59_3.png +0 -0
  684. data/vendor/ggml/examples/yolo/data/labels/59_4.png +0 -0
  685. data/vendor/ggml/examples/yolo/data/labels/59_5.png +0 -0
  686. data/vendor/ggml/examples/yolo/data/labels/59_6.png +0 -0
  687. data/vendor/ggml/examples/yolo/data/labels/59_7.png +0 -0
  688. data/vendor/ggml/examples/yolo/data/labels/60_0.png +0 -0
  689. data/vendor/ggml/examples/yolo/data/labels/60_1.png +0 -0
  690. data/vendor/ggml/examples/yolo/data/labels/60_2.png +0 -0
  691. data/vendor/ggml/examples/yolo/data/labels/60_3.png +0 -0
  692. data/vendor/ggml/examples/yolo/data/labels/60_4.png +0 -0
  693. data/vendor/ggml/examples/yolo/data/labels/60_5.png +0 -0
  694. data/vendor/ggml/examples/yolo/data/labels/60_6.png +0 -0
  695. data/vendor/ggml/examples/yolo/data/labels/60_7.png +0 -0
  696. data/vendor/ggml/examples/yolo/data/labels/61_0.png +0 -0
  697. data/vendor/ggml/examples/yolo/data/labels/61_1.png +0 -0
  698. data/vendor/ggml/examples/yolo/data/labels/61_2.png +0 -0
  699. data/vendor/ggml/examples/yolo/data/labels/61_3.png +0 -0
  700. data/vendor/ggml/examples/yolo/data/labels/61_4.png +0 -0
  701. data/vendor/ggml/examples/yolo/data/labels/61_5.png +0 -0
  702. data/vendor/ggml/examples/yolo/data/labels/61_6.png +0 -0
  703. data/vendor/ggml/examples/yolo/data/labels/61_7.png +0 -0
  704. data/vendor/ggml/examples/yolo/data/labels/62_0.png +0 -0
  705. data/vendor/ggml/examples/yolo/data/labels/62_1.png +0 -0
  706. data/vendor/ggml/examples/yolo/data/labels/62_2.png +0 -0
  707. data/vendor/ggml/examples/yolo/data/labels/62_3.png +0 -0
  708. data/vendor/ggml/examples/yolo/data/labels/62_4.png +0 -0
  709. data/vendor/ggml/examples/yolo/data/labels/62_5.png +0 -0
  710. data/vendor/ggml/examples/yolo/data/labels/62_6.png +0 -0
  711. data/vendor/ggml/examples/yolo/data/labels/62_7.png +0 -0
  712. data/vendor/ggml/examples/yolo/data/labels/63_0.png +0 -0
  713. data/vendor/ggml/examples/yolo/data/labels/63_1.png +0 -0
  714. data/vendor/ggml/examples/yolo/data/labels/63_2.png +0 -0
  715. data/vendor/ggml/examples/yolo/data/labels/63_3.png +0 -0
  716. data/vendor/ggml/examples/yolo/data/labels/63_4.png +0 -0
  717. data/vendor/ggml/examples/yolo/data/labels/63_5.png +0 -0
  718. data/vendor/ggml/examples/yolo/data/labels/63_6.png +0 -0
  719. data/vendor/ggml/examples/yolo/data/labels/63_7.png +0 -0
  720. data/vendor/ggml/examples/yolo/data/labels/64_0.png +0 -0
  721. data/vendor/ggml/examples/yolo/data/labels/64_1.png +0 -0
  722. data/vendor/ggml/examples/yolo/data/labels/64_2.png +0 -0
  723. data/vendor/ggml/examples/yolo/data/labels/64_3.png +0 -0
  724. data/vendor/ggml/examples/yolo/data/labels/64_4.png +0 -0
  725. data/vendor/ggml/examples/yolo/data/labels/64_5.png +0 -0
  726. data/vendor/ggml/examples/yolo/data/labels/64_6.png +0 -0
  727. data/vendor/ggml/examples/yolo/data/labels/64_7.png +0 -0
  728. data/vendor/ggml/examples/yolo/data/labels/65_0.png +0 -0
  729. data/vendor/ggml/examples/yolo/data/labels/65_1.png +0 -0
  730. data/vendor/ggml/examples/yolo/data/labels/65_2.png +0 -0
  731. data/vendor/ggml/examples/yolo/data/labels/65_3.png +0 -0
  732. data/vendor/ggml/examples/yolo/data/labels/65_4.png +0 -0
  733. data/vendor/ggml/examples/yolo/data/labels/65_5.png +0 -0
  734. data/vendor/ggml/examples/yolo/data/labels/65_6.png +0 -0
  735. data/vendor/ggml/examples/yolo/data/labels/65_7.png +0 -0
  736. data/vendor/ggml/examples/yolo/data/labels/66_0.png +0 -0
  737. data/vendor/ggml/examples/yolo/data/labels/66_1.png +0 -0
  738. data/vendor/ggml/examples/yolo/data/labels/66_2.png +0 -0
  739. data/vendor/ggml/examples/yolo/data/labels/66_3.png +0 -0
  740. data/vendor/ggml/examples/yolo/data/labels/66_4.png +0 -0
  741. data/vendor/ggml/examples/yolo/data/labels/66_5.png +0 -0
  742. data/vendor/ggml/examples/yolo/data/labels/66_6.png +0 -0
  743. data/vendor/ggml/examples/yolo/data/labels/66_7.png +0 -0
  744. data/vendor/ggml/examples/yolo/data/labels/67_0.png +0 -0
  745. data/vendor/ggml/examples/yolo/data/labels/67_1.png +0 -0
  746. data/vendor/ggml/examples/yolo/data/labels/67_2.png +0 -0
  747. data/vendor/ggml/examples/yolo/data/labels/67_3.png +0 -0
  748. data/vendor/ggml/examples/yolo/data/labels/67_4.png +0 -0
  749. data/vendor/ggml/examples/yolo/data/labels/67_5.png +0 -0
  750. data/vendor/ggml/examples/yolo/data/labels/67_6.png +0 -0
  751. data/vendor/ggml/examples/yolo/data/labels/67_7.png +0 -0
  752. data/vendor/ggml/examples/yolo/data/labels/68_0.png +0 -0
  753. data/vendor/ggml/examples/yolo/data/labels/68_1.png +0 -0
  754. data/vendor/ggml/examples/yolo/data/labels/68_2.png +0 -0
  755. data/vendor/ggml/examples/yolo/data/labels/68_3.png +0 -0
  756. data/vendor/ggml/examples/yolo/data/labels/68_4.png +0 -0
  757. data/vendor/ggml/examples/yolo/data/labels/68_5.png +0 -0
  758. data/vendor/ggml/examples/yolo/data/labels/68_6.png +0 -0
  759. data/vendor/ggml/examples/yolo/data/labels/68_7.png +0 -0
  760. data/vendor/ggml/examples/yolo/data/labels/69_0.png +0 -0
  761. data/vendor/ggml/examples/yolo/data/labels/69_1.png +0 -0
  762. data/vendor/ggml/examples/yolo/data/labels/69_2.png +0 -0
  763. data/vendor/ggml/examples/yolo/data/labels/69_3.png +0 -0
  764. data/vendor/ggml/examples/yolo/data/labels/69_4.png +0 -0
  765. data/vendor/ggml/examples/yolo/data/labels/69_5.png +0 -0
  766. data/vendor/ggml/examples/yolo/data/labels/69_6.png +0 -0
  767. data/vendor/ggml/examples/yolo/data/labels/69_7.png +0 -0
  768. data/vendor/ggml/examples/yolo/data/labels/70_0.png +0 -0
  769. data/vendor/ggml/examples/yolo/data/labels/70_1.png +0 -0
  770. data/vendor/ggml/examples/yolo/data/labels/70_2.png +0 -0
  771. data/vendor/ggml/examples/yolo/data/labels/70_3.png +0 -0
  772. data/vendor/ggml/examples/yolo/data/labels/70_4.png +0 -0
  773. data/vendor/ggml/examples/yolo/data/labels/70_5.png +0 -0
  774. data/vendor/ggml/examples/yolo/data/labels/70_6.png +0 -0
  775. data/vendor/ggml/examples/yolo/data/labels/70_7.png +0 -0
  776. data/vendor/ggml/examples/yolo/data/labels/71_0.png +0 -0
  777. data/vendor/ggml/examples/yolo/data/labels/71_1.png +0 -0
  778. data/vendor/ggml/examples/yolo/data/labels/71_2.png +0 -0
  779. data/vendor/ggml/examples/yolo/data/labels/71_3.png +0 -0
  780. data/vendor/ggml/examples/yolo/data/labels/71_4.png +0 -0
  781. data/vendor/ggml/examples/yolo/data/labels/71_5.png +0 -0
  782. data/vendor/ggml/examples/yolo/data/labels/71_6.png +0 -0
  783. data/vendor/ggml/examples/yolo/data/labels/71_7.png +0 -0
  784. data/vendor/ggml/examples/yolo/data/labels/72_0.png +0 -0
  785. data/vendor/ggml/examples/yolo/data/labels/72_1.png +0 -0
  786. data/vendor/ggml/examples/yolo/data/labels/72_2.png +0 -0
  787. data/vendor/ggml/examples/yolo/data/labels/72_3.png +0 -0
  788. data/vendor/ggml/examples/yolo/data/labels/72_4.png +0 -0
  789. data/vendor/ggml/examples/yolo/data/labels/72_5.png +0 -0
  790. data/vendor/ggml/examples/yolo/data/labels/72_6.png +0 -0
  791. data/vendor/ggml/examples/yolo/data/labels/72_7.png +0 -0
  792. data/vendor/ggml/examples/yolo/data/labels/73_0.png +0 -0
  793. data/vendor/ggml/examples/yolo/data/labels/73_1.png +0 -0
  794. data/vendor/ggml/examples/yolo/data/labels/73_2.png +0 -0
  795. data/vendor/ggml/examples/yolo/data/labels/73_3.png +0 -0
  796. data/vendor/ggml/examples/yolo/data/labels/73_4.png +0 -0
  797. data/vendor/ggml/examples/yolo/data/labels/73_5.png +0 -0
  798. data/vendor/ggml/examples/yolo/data/labels/73_6.png +0 -0
  799. data/vendor/ggml/examples/yolo/data/labels/73_7.png +0 -0
  800. data/vendor/ggml/examples/yolo/data/labels/74_0.png +0 -0
  801. data/vendor/ggml/examples/yolo/data/labels/74_1.png +0 -0
  802. data/vendor/ggml/examples/yolo/data/labels/74_2.png +0 -0
  803. data/vendor/ggml/examples/yolo/data/labels/74_3.png +0 -0
  804. data/vendor/ggml/examples/yolo/data/labels/74_4.png +0 -0
  805. data/vendor/ggml/examples/yolo/data/labels/74_5.png +0 -0
  806. data/vendor/ggml/examples/yolo/data/labels/74_6.png +0 -0
  807. data/vendor/ggml/examples/yolo/data/labels/74_7.png +0 -0
  808. data/vendor/ggml/examples/yolo/data/labels/75_0.png +0 -0
  809. data/vendor/ggml/examples/yolo/data/labels/75_1.png +0 -0
  810. data/vendor/ggml/examples/yolo/data/labels/75_2.png +0 -0
  811. data/vendor/ggml/examples/yolo/data/labels/75_3.png +0 -0
  812. data/vendor/ggml/examples/yolo/data/labels/75_4.png +0 -0
  813. data/vendor/ggml/examples/yolo/data/labels/75_5.png +0 -0
  814. data/vendor/ggml/examples/yolo/data/labels/75_6.png +0 -0
  815. data/vendor/ggml/examples/yolo/data/labels/75_7.png +0 -0
  816. data/vendor/ggml/examples/yolo/data/labels/76_0.png +0 -0
  817. data/vendor/ggml/examples/yolo/data/labels/76_1.png +0 -0
  818. data/vendor/ggml/examples/yolo/data/labels/76_2.png +0 -0
  819. data/vendor/ggml/examples/yolo/data/labels/76_3.png +0 -0
  820. data/vendor/ggml/examples/yolo/data/labels/76_4.png +0 -0
  821. data/vendor/ggml/examples/yolo/data/labels/76_5.png +0 -0
  822. data/vendor/ggml/examples/yolo/data/labels/76_6.png +0 -0
  823. data/vendor/ggml/examples/yolo/data/labels/76_7.png +0 -0
  824. data/vendor/ggml/examples/yolo/data/labels/77_0.png +0 -0
  825. data/vendor/ggml/examples/yolo/data/labels/77_1.png +0 -0
  826. data/vendor/ggml/examples/yolo/data/labels/77_2.png +0 -0
  827. data/vendor/ggml/examples/yolo/data/labels/77_3.png +0 -0
  828. data/vendor/ggml/examples/yolo/data/labels/77_4.png +0 -0
  829. data/vendor/ggml/examples/yolo/data/labels/77_5.png +0 -0
  830. data/vendor/ggml/examples/yolo/data/labels/77_6.png +0 -0
  831. data/vendor/ggml/examples/yolo/data/labels/77_7.png +0 -0
  832. data/vendor/ggml/examples/yolo/data/labels/78_0.png +0 -0
  833. data/vendor/ggml/examples/yolo/data/labels/78_1.png +0 -0
  834. data/vendor/ggml/examples/yolo/data/labels/78_2.png +0 -0
  835. data/vendor/ggml/examples/yolo/data/labels/78_3.png +0 -0
  836. data/vendor/ggml/examples/yolo/data/labels/78_4.png +0 -0
  837. data/vendor/ggml/examples/yolo/data/labels/78_5.png +0 -0
  838. data/vendor/ggml/examples/yolo/data/labels/78_6.png +0 -0
  839. data/vendor/ggml/examples/yolo/data/labels/78_7.png +0 -0
  840. data/vendor/ggml/examples/yolo/data/labels/79_0.png +0 -0
  841. data/vendor/ggml/examples/yolo/data/labels/79_1.png +0 -0
  842. data/vendor/ggml/examples/yolo/data/labels/79_2.png +0 -0
  843. data/vendor/ggml/examples/yolo/data/labels/79_3.png +0 -0
  844. data/vendor/ggml/examples/yolo/data/labels/79_4.png +0 -0
  845. data/vendor/ggml/examples/yolo/data/labels/79_5.png +0 -0
  846. data/vendor/ggml/examples/yolo/data/labels/79_6.png +0 -0
  847. data/vendor/ggml/examples/yolo/data/labels/79_7.png +0 -0
  848. data/vendor/ggml/examples/yolo/data/labels/80_0.png +0 -0
  849. data/vendor/ggml/examples/yolo/data/labels/80_1.png +0 -0
  850. data/vendor/ggml/examples/yolo/data/labels/80_2.png +0 -0
  851. data/vendor/ggml/examples/yolo/data/labels/80_3.png +0 -0
  852. data/vendor/ggml/examples/yolo/data/labels/80_4.png +0 -0
  853. data/vendor/ggml/examples/yolo/data/labels/80_5.png +0 -0
  854. data/vendor/ggml/examples/yolo/data/labels/80_6.png +0 -0
  855. data/vendor/ggml/examples/yolo/data/labels/80_7.png +0 -0
  856. data/vendor/ggml/examples/yolo/data/labels/81_0.png +0 -0
  857. data/vendor/ggml/examples/yolo/data/labels/81_1.png +0 -0
  858. data/vendor/ggml/examples/yolo/data/labels/81_2.png +0 -0
  859. data/vendor/ggml/examples/yolo/data/labels/81_3.png +0 -0
  860. data/vendor/ggml/examples/yolo/data/labels/81_4.png +0 -0
  861. data/vendor/ggml/examples/yolo/data/labels/81_5.png +0 -0
  862. data/vendor/ggml/examples/yolo/data/labels/81_6.png +0 -0
  863. data/vendor/ggml/examples/yolo/data/labels/81_7.png +0 -0
  864. data/vendor/ggml/examples/yolo/data/labels/82_0.png +0 -0
  865. data/vendor/ggml/examples/yolo/data/labels/82_1.png +0 -0
  866. data/vendor/ggml/examples/yolo/data/labels/82_2.png +0 -0
  867. data/vendor/ggml/examples/yolo/data/labels/82_3.png +0 -0
  868. data/vendor/ggml/examples/yolo/data/labels/82_4.png +0 -0
  869. data/vendor/ggml/examples/yolo/data/labels/82_5.png +0 -0
  870. data/vendor/ggml/examples/yolo/data/labels/82_6.png +0 -0
  871. data/vendor/ggml/examples/yolo/data/labels/82_7.png +0 -0
  872. data/vendor/ggml/examples/yolo/data/labels/83_0.png +0 -0
  873. data/vendor/ggml/examples/yolo/data/labels/83_1.png +0 -0
  874. data/vendor/ggml/examples/yolo/data/labels/83_2.png +0 -0
  875. data/vendor/ggml/examples/yolo/data/labels/83_3.png +0 -0
  876. data/vendor/ggml/examples/yolo/data/labels/83_4.png +0 -0
  877. data/vendor/ggml/examples/yolo/data/labels/83_5.png +0 -0
  878. data/vendor/ggml/examples/yolo/data/labels/83_6.png +0 -0
  879. data/vendor/ggml/examples/yolo/data/labels/83_7.png +0 -0
  880. data/vendor/ggml/examples/yolo/data/labels/84_0.png +0 -0
  881. data/vendor/ggml/examples/yolo/data/labels/84_1.png +0 -0
  882. data/vendor/ggml/examples/yolo/data/labels/84_2.png +0 -0
  883. data/vendor/ggml/examples/yolo/data/labels/84_3.png +0 -0
  884. data/vendor/ggml/examples/yolo/data/labels/84_4.png +0 -0
  885. data/vendor/ggml/examples/yolo/data/labels/84_5.png +0 -0
  886. data/vendor/ggml/examples/yolo/data/labels/84_6.png +0 -0
  887. data/vendor/ggml/examples/yolo/data/labels/84_7.png +0 -0
  888. data/vendor/ggml/examples/yolo/data/labels/85_0.png +0 -0
  889. data/vendor/ggml/examples/yolo/data/labels/85_1.png +0 -0
  890. data/vendor/ggml/examples/yolo/data/labels/85_2.png +0 -0
  891. data/vendor/ggml/examples/yolo/data/labels/85_3.png +0 -0
  892. data/vendor/ggml/examples/yolo/data/labels/85_4.png +0 -0
  893. data/vendor/ggml/examples/yolo/data/labels/85_5.png +0 -0
  894. data/vendor/ggml/examples/yolo/data/labels/85_6.png +0 -0
  895. data/vendor/ggml/examples/yolo/data/labels/85_7.png +0 -0
  896. data/vendor/ggml/examples/yolo/data/labels/86_0.png +0 -0
  897. data/vendor/ggml/examples/yolo/data/labels/86_1.png +0 -0
  898. data/vendor/ggml/examples/yolo/data/labels/86_2.png +0 -0
  899. data/vendor/ggml/examples/yolo/data/labels/86_3.png +0 -0
  900. data/vendor/ggml/examples/yolo/data/labels/86_4.png +0 -0
  901. data/vendor/ggml/examples/yolo/data/labels/86_5.png +0 -0
  902. data/vendor/ggml/examples/yolo/data/labels/86_6.png +0 -0
  903. data/vendor/ggml/examples/yolo/data/labels/86_7.png +0 -0
  904. data/vendor/ggml/examples/yolo/data/labels/87_0.png +0 -0
  905. data/vendor/ggml/examples/yolo/data/labels/87_1.png +0 -0
  906. data/vendor/ggml/examples/yolo/data/labels/87_2.png +0 -0
  907. data/vendor/ggml/examples/yolo/data/labels/87_3.png +0 -0
  908. data/vendor/ggml/examples/yolo/data/labels/87_4.png +0 -0
  909. data/vendor/ggml/examples/yolo/data/labels/87_5.png +0 -0
  910. data/vendor/ggml/examples/yolo/data/labels/87_6.png +0 -0
  911. data/vendor/ggml/examples/yolo/data/labels/87_7.png +0 -0
  912. data/vendor/ggml/examples/yolo/data/labels/88_0.png +0 -0
  913. data/vendor/ggml/examples/yolo/data/labels/88_1.png +0 -0
  914. data/vendor/ggml/examples/yolo/data/labels/88_2.png +0 -0
  915. data/vendor/ggml/examples/yolo/data/labels/88_3.png +0 -0
  916. data/vendor/ggml/examples/yolo/data/labels/88_4.png +0 -0
  917. data/vendor/ggml/examples/yolo/data/labels/88_5.png +0 -0
  918. data/vendor/ggml/examples/yolo/data/labels/88_6.png +0 -0
  919. data/vendor/ggml/examples/yolo/data/labels/88_7.png +0 -0
  920. data/vendor/ggml/examples/yolo/data/labels/89_0.png +0 -0
  921. data/vendor/ggml/examples/yolo/data/labels/89_1.png +0 -0
  922. data/vendor/ggml/examples/yolo/data/labels/89_2.png +0 -0
  923. data/vendor/ggml/examples/yolo/data/labels/89_3.png +0 -0
  924. data/vendor/ggml/examples/yolo/data/labels/89_4.png +0 -0
  925. data/vendor/ggml/examples/yolo/data/labels/89_5.png +0 -0
  926. data/vendor/ggml/examples/yolo/data/labels/89_6.png +0 -0
  927. data/vendor/ggml/examples/yolo/data/labels/89_7.png +0 -0
  928. data/vendor/ggml/examples/yolo/data/labels/90_0.png +0 -0
  929. data/vendor/ggml/examples/yolo/data/labels/90_1.png +0 -0
  930. data/vendor/ggml/examples/yolo/data/labels/90_2.png +0 -0
  931. data/vendor/ggml/examples/yolo/data/labels/90_3.png +0 -0
  932. data/vendor/ggml/examples/yolo/data/labels/90_4.png +0 -0
  933. data/vendor/ggml/examples/yolo/data/labels/90_5.png +0 -0
  934. data/vendor/ggml/examples/yolo/data/labels/90_6.png +0 -0
  935. data/vendor/ggml/examples/yolo/data/labels/90_7.png +0 -0
  936. data/vendor/ggml/examples/yolo/data/labels/91_0.png +0 -0
  937. data/vendor/ggml/examples/yolo/data/labels/91_1.png +0 -0
  938. data/vendor/ggml/examples/yolo/data/labels/91_2.png +0 -0
  939. data/vendor/ggml/examples/yolo/data/labels/91_3.png +0 -0
  940. data/vendor/ggml/examples/yolo/data/labels/91_4.png +0 -0
  941. data/vendor/ggml/examples/yolo/data/labels/91_5.png +0 -0
  942. data/vendor/ggml/examples/yolo/data/labels/91_6.png +0 -0
  943. data/vendor/ggml/examples/yolo/data/labels/91_7.png +0 -0
  944. data/vendor/ggml/examples/yolo/data/labels/92_0.png +0 -0
  945. data/vendor/ggml/examples/yolo/data/labels/92_1.png +0 -0
  946. data/vendor/ggml/examples/yolo/data/labels/92_2.png +0 -0
  947. data/vendor/ggml/examples/yolo/data/labels/92_3.png +0 -0
  948. data/vendor/ggml/examples/yolo/data/labels/92_4.png +0 -0
  949. data/vendor/ggml/examples/yolo/data/labels/92_5.png +0 -0
  950. data/vendor/ggml/examples/yolo/data/labels/92_6.png +0 -0
  951. data/vendor/ggml/examples/yolo/data/labels/92_7.png +0 -0
  952. data/vendor/ggml/examples/yolo/data/labels/93_0.png +0 -0
  953. data/vendor/ggml/examples/yolo/data/labels/93_1.png +0 -0
  954. data/vendor/ggml/examples/yolo/data/labels/93_2.png +0 -0
  955. data/vendor/ggml/examples/yolo/data/labels/93_3.png +0 -0
  956. data/vendor/ggml/examples/yolo/data/labels/93_4.png +0 -0
  957. data/vendor/ggml/examples/yolo/data/labels/93_5.png +0 -0
  958. data/vendor/ggml/examples/yolo/data/labels/93_6.png +0 -0
  959. data/vendor/ggml/examples/yolo/data/labels/93_7.png +0 -0
  960. data/vendor/ggml/examples/yolo/data/labels/94_0.png +0 -0
  961. data/vendor/ggml/examples/yolo/data/labels/94_1.png +0 -0
  962. data/vendor/ggml/examples/yolo/data/labels/94_2.png +0 -0
  963. data/vendor/ggml/examples/yolo/data/labels/94_3.png +0 -0
  964. data/vendor/ggml/examples/yolo/data/labels/94_4.png +0 -0
  965. data/vendor/ggml/examples/yolo/data/labels/94_5.png +0 -0
  966. data/vendor/ggml/examples/yolo/data/labels/94_6.png +0 -0
  967. data/vendor/ggml/examples/yolo/data/labels/94_7.png +0 -0
  968. data/vendor/ggml/examples/yolo/data/labels/95_0.png +0 -0
  969. data/vendor/ggml/examples/yolo/data/labels/95_1.png +0 -0
  970. data/vendor/ggml/examples/yolo/data/labels/95_2.png +0 -0
  971. data/vendor/ggml/examples/yolo/data/labels/95_3.png +0 -0
  972. data/vendor/ggml/examples/yolo/data/labels/95_4.png +0 -0
  973. data/vendor/ggml/examples/yolo/data/labels/95_5.png +0 -0
  974. data/vendor/ggml/examples/yolo/data/labels/95_6.png +0 -0
  975. data/vendor/ggml/examples/yolo/data/labels/95_7.png +0 -0
  976. data/vendor/ggml/examples/yolo/data/labels/96_0.png +0 -0
  977. data/vendor/ggml/examples/yolo/data/labels/96_1.png +0 -0
  978. data/vendor/ggml/examples/yolo/data/labels/96_2.png +0 -0
  979. data/vendor/ggml/examples/yolo/data/labels/96_3.png +0 -0
  980. data/vendor/ggml/examples/yolo/data/labels/96_4.png +0 -0
  981. data/vendor/ggml/examples/yolo/data/labels/96_5.png +0 -0
  982. data/vendor/ggml/examples/yolo/data/labels/96_6.png +0 -0
  983. data/vendor/ggml/examples/yolo/data/labels/96_7.png +0 -0
  984. data/vendor/ggml/examples/yolo/data/labels/97_0.png +0 -0
  985. data/vendor/ggml/examples/yolo/data/labels/97_1.png +0 -0
  986. data/vendor/ggml/examples/yolo/data/labels/97_2.png +0 -0
  987. data/vendor/ggml/examples/yolo/data/labels/97_3.png +0 -0
  988. data/vendor/ggml/examples/yolo/data/labels/97_4.png +0 -0
  989. data/vendor/ggml/examples/yolo/data/labels/97_5.png +0 -0
  990. data/vendor/ggml/examples/yolo/data/labels/97_6.png +0 -0
  991. data/vendor/ggml/examples/yolo/data/labels/97_7.png +0 -0
  992. data/vendor/ggml/examples/yolo/data/labels/98_0.png +0 -0
  993. data/vendor/ggml/examples/yolo/data/labels/98_1.png +0 -0
  994. data/vendor/ggml/examples/yolo/data/labels/98_2.png +0 -0
  995. data/vendor/ggml/examples/yolo/data/labels/98_3.png +0 -0
  996. data/vendor/ggml/examples/yolo/data/labels/98_4.png +0 -0
  997. data/vendor/ggml/examples/yolo/data/labels/98_5.png +0 -0
  998. data/vendor/ggml/examples/yolo/data/labels/98_6.png +0 -0
  999. data/vendor/ggml/examples/yolo/data/labels/98_7.png +0 -0
  1000. data/vendor/ggml/examples/yolo/data/labels/99_0.png +0 -0
  1001. data/vendor/ggml/examples/yolo/data/labels/99_1.png +0 -0
  1002. data/vendor/ggml/examples/yolo/data/labels/99_2.png +0 -0
  1003. data/vendor/ggml/examples/yolo/data/labels/99_3.png +0 -0
  1004. data/vendor/ggml/examples/yolo/data/labels/99_4.png +0 -0
  1005. data/vendor/ggml/examples/yolo/data/labels/99_5.png +0 -0
  1006. data/vendor/ggml/examples/yolo/data/labels/99_6.png +0 -0
  1007. data/vendor/ggml/examples/yolo/data/labels/99_7.png +0 -0
  1008. data/vendor/ggml/examples/yolo/yolo-image.cpp +210 -0
  1009. data/vendor/ggml/examples/yolo/yolo-image.h +39 -0
  1010. data/vendor/ggml/examples/yolo/yolov3-tiny.cpp +661 -0
  1011. data/vendor/ggml/ggml.pc.in +10 -0
  1012. data/vendor/ggml/include/ggml-alloc.h +85 -0
  1013. data/vendor/ggml/include/ggml-backend.h +431 -0
  1014. data/vendor/ggml/include/ggml-blas.h +25 -0
  1015. data/vendor/ggml/include/ggml-cann.h +123 -0
  1016. data/vendor/ggml/include/ggml-cpp.h +39 -0
  1017. data/vendor/ggml/include/ggml-cpu.h +151 -0
  1018. data/vendor/ggml/include/ggml-cuda.h +50 -0
  1019. data/vendor/ggml/include/ggml-hexagon.h +19 -0
  1020. data/vendor/ggml/include/ggml-metal.h +61 -0
  1021. data/vendor/ggml/include/ggml-opencl.h +26 -0
  1022. data/vendor/ggml/include/ggml-openvino.h +37 -0
  1023. data/vendor/ggml/include/ggml-opt.h +256 -0
  1024. data/vendor/ggml/include/ggml-rpc.h +35 -0
  1025. data/vendor/ggml/include/ggml-sycl.h +49 -0
  1026. data/vendor/ggml/include/ggml-virtgpu.h +14 -0
  1027. data/vendor/ggml/include/ggml-vulkan.h +29 -0
  1028. data/vendor/ggml/include/ggml-webgpu.h +19 -0
  1029. data/vendor/ggml/include/ggml-zdnn.h +17 -0
  1030. data/vendor/ggml/include/ggml-zendnn.h +22 -0
  1031. data/vendor/ggml/include/ggml.h +2845 -0
  1032. data/vendor/ggml/include/gguf.h +204 -0
  1033. data/vendor/ggml/requirements.txt +12 -0
  1034. data/vendor/ggml/scripts/gen-authors.sh +9 -0
  1035. data/vendor/ggml/scripts/release.sh +296 -0
  1036. data/vendor/ggml/scripts/sync-llama-am.sh +167 -0
  1037. data/vendor/ggml/scripts/sync-llama.last +1 -0
  1038. data/vendor/ggml/scripts/sync-llama.sh +21 -0
  1039. data/vendor/ggml/scripts/sync-whisper-am.sh +138 -0
  1040. data/vendor/ggml/scripts/sync-whisper.last +1 -0
  1041. data/vendor/ggml/scripts/sync-whisper.sh +17 -0
  1042. data/vendor/ggml/src/CMakeLists.txt +493 -0
  1043. data/vendor/ggml/src/ggml-alloc.c +1248 -0
  1044. data/vendor/ggml/src/ggml-backend-dl.cpp +48 -0
  1045. data/vendor/ggml/src/ggml-backend-dl.h +45 -0
  1046. data/vendor/ggml/src/ggml-backend-impl.h +275 -0
  1047. data/vendor/ggml/src/ggml-backend-meta.cpp +2144 -0
  1048. data/vendor/ggml/src/ggml-backend-reg.cpp +586 -0
  1049. data/vendor/ggml/src/ggml-backend.cpp +2371 -0
  1050. data/vendor/ggml/src/ggml-blas/CMakeLists.txt +101 -0
  1051. data/vendor/ggml/src/ggml-blas/ggml-blas.cpp +522 -0
  1052. data/vendor/ggml/src/ggml-cann/CMakeLists.txt +89 -0
  1053. data/vendor/ggml/src/ggml-cann/acl_tensor.cpp +195 -0
  1054. data/vendor/ggml/src/ggml-cann/acl_tensor.h +349 -0
  1055. data/vendor/ggml/src/ggml-cann/aclnn_ops.cpp +4436 -0
  1056. data/vendor/ggml/src/ggml-cann/aclnn_ops.h +1190 -0
  1057. data/vendor/ggml/src/ggml-cann/common.h +651 -0
  1058. data/vendor/ggml/src/ggml-cann/ggml-cann.cpp +3062 -0
  1059. data/vendor/ggml/src/ggml-common.h +1900 -0
  1060. data/vendor/ggml/src/ggml-cpu/CMakeLists.txt +731 -0
  1061. data/vendor/ggml/src/ggml-cpu/amx/amx.cpp +249 -0
  1062. data/vendor/ggml/src/ggml-cpu/amx/amx.h +8 -0
  1063. data/vendor/ggml/src/ggml-cpu/amx/common.h +115 -0
  1064. data/vendor/ggml/src/ggml-cpu/amx/mmq.cpp +2512 -0
  1065. data/vendor/ggml/src/ggml-cpu/amx/mmq.h +10 -0
  1066. data/vendor/ggml/src/ggml-cpu/arch/arm/cpu-feats.cpp +98 -0
  1067. data/vendor/ggml/src/ggml-cpu/arch/arm/quants.c +4245 -0
  1068. data/vendor/ggml/src/ggml-cpu/arch/arm/repack.cpp +5156 -0
  1069. data/vendor/ggml/src/ggml-cpu/arch/loongarch/quants.c +2158 -0
  1070. data/vendor/ggml/src/ggml-cpu/arch/powerpc/cpu-feats.cpp +82 -0
  1071. data/vendor/ggml/src/ggml-cpu/arch/powerpc/quants.c +2304 -0
  1072. data/vendor/ggml/src/ggml-cpu/arch/riscv/cpu-feats.cpp +38 -0
  1073. data/vendor/ggml/src/ggml-cpu/arch/riscv/quants.c +4553 -0
  1074. data/vendor/ggml/src/ggml-cpu/arch/riscv/repack.cpp +1703 -0
  1075. data/vendor/ggml/src/ggml-cpu/arch/s390/cpu-feats.cpp +50 -0
  1076. data/vendor/ggml/src/ggml-cpu/arch/s390/quants.c +1465 -0
  1077. data/vendor/ggml/src/ggml-cpu/arch/wasm/quants.c +1220 -0
  1078. data/vendor/ggml/src/ggml-cpu/arch/x86/cpu-feats.cpp +327 -0
  1079. data/vendor/ggml/src/ggml-cpu/arch/x86/quants.c +3970 -0
  1080. data/vendor/ggml/src/ggml-cpu/arch/x86/repack.cpp +6407 -0
  1081. data/vendor/ggml/src/ggml-cpu/arch-fallback.h +348 -0
  1082. data/vendor/ggml/src/ggml-cpu/binary-ops.cpp +154 -0
  1083. data/vendor/ggml/src/ggml-cpu/binary-ops.h +16 -0
  1084. data/vendor/ggml/src/ggml-cpu/cmake/FindSIMD.cmake +100 -0
  1085. data/vendor/ggml/src/ggml-cpu/cmake/FindSMTIME.cmake +32 -0
  1086. data/vendor/ggml/src/ggml-cpu/common.h +95 -0
  1087. data/vendor/ggml/src/ggml-cpu/ggml-cpu-impl.h +539 -0
  1088. data/vendor/ggml/src/ggml-cpu/ggml-cpu.c +3835 -0
  1089. data/vendor/ggml/src/ggml-cpu/ggml-cpu.cpp +703 -0
  1090. data/vendor/ggml/src/ggml-cpu/hbm.cpp +55 -0
  1091. data/vendor/ggml/src/ggml-cpu/hbm.h +8 -0
  1092. data/vendor/ggml/src/ggml-cpu/kleidiai/kernels.cpp +939 -0
  1093. data/vendor/ggml/src/ggml-cpu/kleidiai/kernels.h +90 -0
  1094. data/vendor/ggml/src/ggml-cpu/kleidiai/kleidiai.cpp +1513 -0
  1095. data/vendor/ggml/src/ggml-cpu/kleidiai/kleidiai.h +17 -0
  1096. data/vendor/ggml/src/ggml-cpu/llamafile/sgemm.cpp +4051 -0
  1097. data/vendor/ggml/src/ggml-cpu/llamafile/sgemm.h +25 -0
  1098. data/vendor/ggml/src/ggml-cpu/ops.cpp +11373 -0
  1099. data/vendor/ggml/src/ggml-cpu/ops.h +119 -0
  1100. data/vendor/ggml/src/ggml-cpu/quants.c +1288 -0
  1101. data/vendor/ggml/src/ggml-cpu/quants.h +103 -0
  1102. data/vendor/ggml/src/ggml-cpu/repack.cpp +4836 -0
  1103. data/vendor/ggml/src/ggml-cpu/repack.h +245 -0
  1104. data/vendor/ggml/src/ggml-cpu/simd-gemm.h +226 -0
  1105. data/vendor/ggml/src/ggml-cpu/simd-mappings.h +1319 -0
  1106. data/vendor/ggml/src/ggml-cpu/spacemit/ime.cpp +1740 -0
  1107. data/vendor/ggml/src/ggml-cpu/spacemit/ime.h +21 -0
  1108. data/vendor/ggml/src/ggml-cpu/spacemit/ime1_kernels.cpp +1027 -0
  1109. data/vendor/ggml/src/ggml-cpu/spacemit/ime2_kernels.cpp +5768 -0
  1110. data/vendor/ggml/src/ggml-cpu/spacemit/ime_env.cpp +320 -0
  1111. data/vendor/ggml/src/ggml-cpu/spacemit/ime_env.h +55 -0
  1112. data/vendor/ggml/src/ggml-cpu/spacemit/ime_kernels.h +189 -0
  1113. data/vendor/ggml/src/ggml-cpu/spacemit/repack.cpp +1795 -0
  1114. data/vendor/ggml/src/ggml-cpu/spacemit/repack.h +14 -0
  1115. data/vendor/ggml/src/ggml-cpu/spacemit/rvv_kernels.cpp +3178 -0
  1116. data/vendor/ggml/src/ggml-cpu/spacemit/rvv_kernels.h +95 -0
  1117. data/vendor/ggml/src/ggml-cpu/spacemit/spine_barrier.h +34 -0
  1118. data/vendor/ggml/src/ggml-cpu/spacemit/spine_mem_pool.cpp +760 -0
  1119. data/vendor/ggml/src/ggml-cpu/spacemit/spine_mem_pool.h +32 -0
  1120. data/vendor/ggml/src/ggml-cpu/spacemit/spine_tcm.h +409 -0
  1121. data/vendor/ggml/src/ggml-cpu/traits.cpp +36 -0
  1122. data/vendor/ggml/src/ggml-cpu/traits.h +38 -0
  1123. data/vendor/ggml/src/ggml-cpu/unary-ops.cpp +337 -0
  1124. data/vendor/ggml/src/ggml-cpu/unary-ops.h +35 -0
  1125. data/vendor/ggml/src/ggml-cpu/vec.cpp +629 -0
  1126. data/vendor/ggml/src/ggml-cpu/vec.h +1588 -0
  1127. data/vendor/ggml/src/ggml-cuda/CMakeLists.txt +268 -0
  1128. data/vendor/ggml/src/ggml-cuda/acc.cu +61 -0
  1129. data/vendor/ggml/src/ggml-cuda/acc.cuh +5 -0
  1130. data/vendor/ggml/src/ggml-cuda/add-id.cu +58 -0
  1131. data/vendor/ggml/src/ggml-cuda/add-id.cuh +3 -0
  1132. data/vendor/ggml/src/ggml-cuda/allreduce.cu +971 -0
  1133. data/vendor/ggml/src/ggml-cuda/allreduce.cuh +29 -0
  1134. data/vendor/ggml/src/ggml-cuda/arange.cu +34 -0
  1135. data/vendor/ggml/src/ggml-cuda/arange.cuh +5 -0
  1136. data/vendor/ggml/src/ggml-cuda/argmax.cu +91 -0
  1137. data/vendor/ggml/src/ggml-cuda/argmax.cuh +3 -0
  1138. data/vendor/ggml/src/ggml-cuda/argsort.cu +266 -0
  1139. data/vendor/ggml/src/ggml-cuda/argsort.cuh +19 -0
  1140. data/vendor/ggml/src/ggml-cuda/binbcast.cu +534 -0
  1141. data/vendor/ggml/src/ggml-cuda/binbcast.cuh +12 -0
  1142. data/vendor/ggml/src/ggml-cuda/clamp.cu +45 -0
  1143. data/vendor/ggml/src/ggml-cuda/clamp.cuh +5 -0
  1144. data/vendor/ggml/src/ggml-cuda/common.cuh +1489 -0
  1145. data/vendor/ggml/src/ggml-cuda/concat.cu +204 -0
  1146. data/vendor/ggml/src/ggml-cuda/concat.cuh +5 -0
  1147. data/vendor/ggml/src/ggml-cuda/conv-transpose-1d.cu +86 -0
  1148. data/vendor/ggml/src/ggml-cuda/conv-transpose-1d.cuh +5 -0
  1149. data/vendor/ggml/src/ggml-cuda/conv2d-dw.cu +161 -0
  1150. data/vendor/ggml/src/ggml-cuda/conv2d-dw.cuh +5 -0
  1151. data/vendor/ggml/src/ggml-cuda/conv2d-transpose.cu +115 -0
  1152. data/vendor/ggml/src/ggml-cuda/conv2d-transpose.cuh +5 -0
  1153. data/vendor/ggml/src/ggml-cuda/conv2d.cu +166 -0
  1154. data/vendor/ggml/src/ggml-cuda/conv2d.cuh +5 -0
  1155. data/vendor/ggml/src/ggml-cuda/convert.cu +892 -0
  1156. data/vendor/ggml/src/ggml-cuda/convert.cuh +66 -0
  1157. data/vendor/ggml/src/ggml-cuda/count-equal.cu +64 -0
  1158. data/vendor/ggml/src/ggml-cuda/count-equal.cuh +5 -0
  1159. data/vendor/ggml/src/ggml-cuda/cp-async.cuh +57 -0
  1160. data/vendor/ggml/src/ggml-cuda/cpy-utils.cuh +217 -0
  1161. data/vendor/ggml/src/ggml-cuda/cpy.cu +558 -0
  1162. data/vendor/ggml/src/ggml-cuda/cpy.cuh +7 -0
  1163. data/vendor/ggml/src/ggml-cuda/cross-entropy-loss.cu +177 -0
  1164. data/vendor/ggml/src/ggml-cuda/cross-entropy-loss.cuh +7 -0
  1165. data/vendor/ggml/src/ggml-cuda/cumsum.cu +307 -0
  1166. data/vendor/ggml/src/ggml-cuda/cumsum.cuh +5 -0
  1167. data/vendor/ggml/src/ggml-cuda/dequantize.cuh +99 -0
  1168. data/vendor/ggml/src/ggml-cuda/diag.cu +77 -0
  1169. data/vendor/ggml/src/ggml-cuda/diag.cuh +5 -0
  1170. data/vendor/ggml/src/ggml-cuda/diagmask.cu +40 -0
  1171. data/vendor/ggml/src/ggml-cuda/diagmask.cuh +5 -0
  1172. data/vendor/ggml/src/ggml-cuda/fattn-common.cuh +1212 -0
  1173. data/vendor/ggml/src/ggml-cuda/fattn-mma-f16.cuh +2020 -0
  1174. data/vendor/ggml/src/ggml-cuda/fattn-tile.cu +61 -0
  1175. data/vendor/ggml/src/ggml-cuda/fattn-tile.cuh +1347 -0
  1176. data/vendor/ggml/src/ggml-cuda/fattn-vec.cuh +600 -0
  1177. data/vendor/ggml/src/ggml-cuda/fattn-wmma-f16.cu +696 -0
  1178. data/vendor/ggml/src/ggml-cuda/fattn-wmma-f16.cuh +51 -0
  1179. data/vendor/ggml/src/ggml-cuda/fattn.cu +562 -0
  1180. data/vendor/ggml/src/ggml-cuda/fattn.cuh +5 -0
  1181. data/vendor/ggml/src/ggml-cuda/fill.cu +37 -0
  1182. data/vendor/ggml/src/ggml-cuda/fill.cuh +3 -0
  1183. data/vendor/ggml/src/ggml-cuda/gated_delta_net.cu +311 -0
  1184. data/vendor/ggml/src/ggml-cuda/gated_delta_net.cuh +4 -0
  1185. data/vendor/ggml/src/ggml-cuda/getrows.cu +300 -0
  1186. data/vendor/ggml/src/ggml-cuda/getrows.cuh +15 -0
  1187. data/vendor/ggml/src/ggml-cuda/ggml-cuda.cu +5684 -0
  1188. data/vendor/ggml/src/ggml-cuda/gla.cu +93 -0
  1189. data/vendor/ggml/src/ggml-cuda/gla.cuh +3 -0
  1190. data/vendor/ggml/src/ggml-cuda/im2col.cu +267 -0
  1191. data/vendor/ggml/src/ggml-cuda/im2col.cuh +6 -0
  1192. data/vendor/ggml/src/ggml-cuda/mean.cu +75 -0
  1193. data/vendor/ggml/src/ggml-cuda/mean.cuh +3 -0
  1194. data/vendor/ggml/src/ggml-cuda/mma.cuh +1456 -0
  1195. data/vendor/ggml/src/ggml-cuda/mmf.cu +191 -0
  1196. data/vendor/ggml/src/ggml-cuda/mmf.cuh +908 -0
  1197. data/vendor/ggml/src/ggml-cuda/mmid.cu +164 -0
  1198. data/vendor/ggml/src/ggml-cuda/mmid.cuh +5 -0
  1199. data/vendor/ggml/src/ggml-cuda/mmq.cu +372 -0
  1200. data/vendor/ggml/src/ggml-cuda/mmq.cuh +4176 -0
  1201. data/vendor/ggml/src/ggml-cuda/mmvf.cu +862 -0
  1202. data/vendor/ggml/src/ggml-cuda/mmvf.cuh +14 -0
  1203. data/vendor/ggml/src/ggml-cuda/mmvq.cu +1161 -0
  1204. data/vendor/ggml/src/ggml-cuda/mmvq.cuh +16 -0
  1205. data/vendor/ggml/src/ggml-cuda/norm.cu +672 -0
  1206. data/vendor/ggml/src/ggml-cuda/norm.cuh +18 -0
  1207. data/vendor/ggml/src/ggml-cuda/opt-step-adamw.cu +78 -0
  1208. data/vendor/ggml/src/ggml-cuda/opt-step-adamw.cuh +5 -0
  1209. data/vendor/ggml/src/ggml-cuda/opt-step-sgd.cu +49 -0
  1210. data/vendor/ggml/src/ggml-cuda/opt-step-sgd.cuh +5 -0
  1211. data/vendor/ggml/src/ggml-cuda/out-prod.cu +84 -0
  1212. data/vendor/ggml/src/ggml-cuda/out-prod.cuh +3 -0
  1213. data/vendor/ggml/src/ggml-cuda/pad.cu +106 -0
  1214. data/vendor/ggml/src/ggml-cuda/pad.cuh +5 -0
  1215. data/vendor/ggml/src/ggml-cuda/pad_reflect_1d.cu +91 -0
  1216. data/vendor/ggml/src/ggml-cuda/pad_reflect_1d.cuh +5 -0
  1217. data/vendor/ggml/src/ggml-cuda/pool2d.cu +94 -0
  1218. data/vendor/ggml/src/ggml-cuda/pool2d.cuh +5 -0
  1219. data/vendor/ggml/src/ggml-cuda/quantize.cu +443 -0
  1220. data/vendor/ggml/src/ggml-cuda/quantize.cuh +41 -0
  1221. data/vendor/ggml/src/ggml-cuda/reduce_rows.cuh +39 -0
  1222. data/vendor/ggml/src/ggml-cuda/roll.cu +67 -0
  1223. data/vendor/ggml/src/ggml-cuda/roll.cuh +5 -0
  1224. data/vendor/ggml/src/ggml-cuda/rope.cu +665 -0
  1225. data/vendor/ggml/src/ggml-cuda/rope.cuh +9 -0
  1226. data/vendor/ggml/src/ggml-cuda/scale.cu +34 -0
  1227. data/vendor/ggml/src/ggml-cuda/scale.cuh +5 -0
  1228. data/vendor/ggml/src/ggml-cuda/set-rows.cu +330 -0
  1229. data/vendor/ggml/src/ggml-cuda/set-rows.cuh +7 -0
  1230. data/vendor/ggml/src/ggml-cuda/set.cu +39 -0
  1231. data/vendor/ggml/src/ggml-cuda/set.cuh +7 -0
  1232. data/vendor/ggml/src/ggml-cuda/snake.cu +72 -0
  1233. data/vendor/ggml/src/ggml-cuda/snake.cuh +8 -0
  1234. data/vendor/ggml/src/ggml-cuda/softcap.cu +34 -0
  1235. data/vendor/ggml/src/ggml-cuda/softcap.cuh +5 -0
  1236. data/vendor/ggml/src/ggml-cuda/softmax.cu +472 -0
  1237. data/vendor/ggml/src/ggml-cuda/softmax.cuh +7 -0
  1238. data/vendor/ggml/src/ggml-cuda/solve_tri.cu +275 -0
  1239. data/vendor/ggml/src/ggml-cuda/solve_tri.cuh +3 -0
  1240. data/vendor/ggml/src/ggml-cuda/ssm-conv.cu +197 -0
  1241. data/vendor/ggml/src/ggml-cuda/ssm-conv.cuh +3 -0
  1242. data/vendor/ggml/src/ggml-cuda/ssm-scan.cu +342 -0
  1243. data/vendor/ggml/src/ggml-cuda/ssm-scan.cuh +3 -0
  1244. data/vendor/ggml/src/ggml-cuda/sum.cu +41 -0
  1245. data/vendor/ggml/src/ggml-cuda/sum.cuh +5 -0
  1246. data/vendor/ggml/src/ggml-cuda/sumrows.cu +43 -0
  1247. data/vendor/ggml/src/ggml-cuda/sumrows.cuh +4 -0
  1248. data/vendor/ggml/src/ggml-cuda/template-instances/fattn-mma-f16-instance-ncols1_1-ncols2_16.cu +6 -0
  1249. data/vendor/ggml/src/ggml-cuda/template-instances/fattn-mma-f16-instance-ncols1_1-ncols2_32.cu +6 -0
  1250. data/vendor/ggml/src/ggml-cuda/template-instances/fattn-mma-f16-instance-ncols1_1-ncols2_8.cu +12 -0
  1251. data/vendor/ggml/src/ggml-cuda/template-instances/fattn-mma-f16-instance-ncols1_16-ncols2_1.cu +10 -0
  1252. data/vendor/ggml/src/ggml-cuda/template-instances/fattn-mma-f16-instance-ncols1_16-ncols2_2.cu +10 -0
  1253. data/vendor/ggml/src/ggml-cuda/template-instances/fattn-mma-f16-instance-ncols1_16-ncols2_4.cu +12 -0
  1254. data/vendor/ggml/src/ggml-cuda/template-instances/fattn-mma-f16-instance-ncols1_2-ncols2_16.cu +6 -0
  1255. data/vendor/ggml/src/ggml-cuda/template-instances/fattn-mma-f16-instance-ncols1_2-ncols2_32.cu +6 -0
  1256. data/vendor/ggml/src/ggml-cuda/template-instances/fattn-mma-f16-instance-ncols1_2-ncols2_4.cu +12 -0
  1257. data/vendor/ggml/src/ggml-cuda/template-instances/fattn-mma-f16-instance-ncols1_2-ncols2_8.cu +12 -0
  1258. data/vendor/ggml/src/ggml-cuda/template-instances/fattn-mma-f16-instance-ncols1_32-ncols2_1.cu +10 -0
  1259. data/vendor/ggml/src/ggml-cuda/template-instances/fattn-mma-f16-instance-ncols1_32-ncols2_2.cu +10 -0
  1260. data/vendor/ggml/src/ggml-cuda/template-instances/fattn-mma-f16-instance-ncols1_4-ncols2_16.cu +6 -0
  1261. data/vendor/ggml/src/ggml-cuda/template-instances/fattn-mma-f16-instance-ncols1_4-ncols2_2.cu +10 -0
  1262. data/vendor/ggml/src/ggml-cuda/template-instances/fattn-mma-f16-instance-ncols1_4-ncols2_4.cu +12 -0
  1263. data/vendor/ggml/src/ggml-cuda/template-instances/fattn-mma-f16-instance-ncols1_4-ncols2_8.cu +12 -0
  1264. data/vendor/ggml/src/ggml-cuda/template-instances/fattn-mma-f16-instance-ncols1_64-ncols2_1.cu +10 -0
  1265. data/vendor/ggml/src/ggml-cuda/template-instances/fattn-mma-f16-instance-ncols1_8-ncols2_1.cu +10 -0
  1266. data/vendor/ggml/src/ggml-cuda/template-instances/fattn-mma-f16-instance-ncols1_8-ncols2_2.cu +10 -0
  1267. data/vendor/ggml/src/ggml-cuda/template-instances/fattn-mma-f16-instance-ncols1_8-ncols2_4.cu +12 -0
  1268. data/vendor/ggml/src/ggml-cuda/template-instances/fattn-mma-f16-instance-ncols1_8-ncols2_8.cu +12 -0
  1269. data/vendor/ggml/src/ggml-cuda/template-instances/fattn-tile-instance-dkq112-dv112.cu +5 -0
  1270. data/vendor/ggml/src/ggml-cuda/template-instances/fattn-tile-instance-dkq128-dv128.cu +5 -0
  1271. data/vendor/ggml/src/ggml-cuda/template-instances/fattn-tile-instance-dkq192-dv128.cu +5 -0
  1272. data/vendor/ggml/src/ggml-cuda/template-instances/fattn-tile-instance-dkq256-dv256.cu +5 -0
  1273. data/vendor/ggml/src/ggml-cuda/template-instances/fattn-tile-instance-dkq320-dv256.cu +5 -0
  1274. data/vendor/ggml/src/ggml-cuda/template-instances/fattn-tile-instance-dkq40-dv40.cu +5 -0
  1275. data/vendor/ggml/src/ggml-cuda/template-instances/fattn-tile-instance-dkq512-dv512.cu +5 -0
  1276. data/vendor/ggml/src/ggml-cuda/template-instances/fattn-tile-instance-dkq576-dv512.cu +5 -0
  1277. data/vendor/ggml/src/ggml-cuda/template-instances/fattn-tile-instance-dkq64-dv64.cu +5 -0
  1278. data/vendor/ggml/src/ggml-cuda/template-instances/fattn-tile-instance-dkq72-dv72.cu +5 -0
  1279. data/vendor/ggml/src/ggml-cuda/template-instances/fattn-tile-instance-dkq80-dv80.cu +5 -0
  1280. data/vendor/ggml/src/ggml-cuda/template-instances/fattn-tile-instance-dkq96-dv96.cu +5 -0
  1281. data/vendor/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-bf16-bf16.cu +7 -0
  1282. data/vendor/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-bf16-f16.cu +7 -0
  1283. data/vendor/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-bf16-q4_0.cu +7 -0
  1284. data/vendor/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-bf16-q4_1.cu +7 -0
  1285. data/vendor/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-bf16-q5_0.cu +7 -0
  1286. data/vendor/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-bf16-q5_1.cu +7 -0
  1287. data/vendor/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-bf16-q8_0.cu +7 -0
  1288. data/vendor/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-f16-bf16.cu +7 -0
  1289. data/vendor/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-f16-f16.cu +7 -0
  1290. data/vendor/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-f16-q4_0.cu +7 -0
  1291. data/vendor/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-f16-q4_1.cu +7 -0
  1292. data/vendor/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-f16-q5_0.cu +7 -0
  1293. data/vendor/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-f16-q5_1.cu +7 -0
  1294. data/vendor/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-f16-q8_0.cu +7 -0
  1295. data/vendor/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q4_0-bf16.cu +7 -0
  1296. data/vendor/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q4_0-f16.cu +7 -0
  1297. data/vendor/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q4_0-q4_0.cu +7 -0
  1298. data/vendor/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q4_0-q4_1.cu +7 -0
  1299. data/vendor/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q4_0-q5_0.cu +7 -0
  1300. data/vendor/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q4_0-q5_1.cu +7 -0
  1301. data/vendor/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q4_0-q8_0.cu +7 -0
  1302. data/vendor/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q4_1-bf16.cu +7 -0
  1303. data/vendor/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q4_1-f16.cu +7 -0
  1304. data/vendor/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q4_1-q4_0.cu +7 -0
  1305. data/vendor/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q4_1-q4_1.cu +7 -0
  1306. data/vendor/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q4_1-q5_0.cu +7 -0
  1307. data/vendor/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q4_1-q5_1.cu +7 -0
  1308. data/vendor/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q4_1-q8_0.cu +7 -0
  1309. data/vendor/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q5_0-bf16.cu +7 -0
  1310. data/vendor/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q5_0-f16.cu +7 -0
  1311. data/vendor/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q5_0-q4_0.cu +7 -0
  1312. data/vendor/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q5_0-q4_1.cu +7 -0
  1313. data/vendor/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q5_0-q5_0.cu +7 -0
  1314. data/vendor/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q5_0-q5_1.cu +7 -0
  1315. data/vendor/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q5_0-q8_0.cu +7 -0
  1316. data/vendor/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q5_1-bf16.cu +7 -0
  1317. data/vendor/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q5_1-f16.cu +7 -0
  1318. data/vendor/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q5_1-q4_0.cu +7 -0
  1319. data/vendor/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q5_1-q4_1.cu +7 -0
  1320. data/vendor/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q5_1-q5_0.cu +7 -0
  1321. data/vendor/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q5_1-q5_1.cu +7 -0
  1322. data/vendor/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q5_1-q8_0.cu +7 -0
  1323. data/vendor/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q8_0-bf16.cu +7 -0
  1324. data/vendor/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q8_0-f16.cu +7 -0
  1325. data/vendor/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q8_0-q4_0.cu +7 -0
  1326. data/vendor/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q8_0-q4_1.cu +7 -0
  1327. data/vendor/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q8_0-q5_0.cu +7 -0
  1328. data/vendor/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q8_0-q5_1.cu +7 -0
  1329. data/vendor/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q8_0-q8_0.cu +7 -0
  1330. data/vendor/ggml/src/ggml-cuda/template-instances/generate_cu_files.py +110 -0
  1331. data/vendor/ggml/src/ggml-cuda/template-instances/mmf-instance-ncols_1.cu +5 -0
  1332. data/vendor/ggml/src/ggml-cuda/template-instances/mmf-instance-ncols_10.cu +5 -0
  1333. data/vendor/ggml/src/ggml-cuda/template-instances/mmf-instance-ncols_11.cu +5 -0
  1334. data/vendor/ggml/src/ggml-cuda/template-instances/mmf-instance-ncols_12.cu +5 -0
  1335. data/vendor/ggml/src/ggml-cuda/template-instances/mmf-instance-ncols_13.cu +5 -0
  1336. data/vendor/ggml/src/ggml-cuda/template-instances/mmf-instance-ncols_14.cu +5 -0
  1337. data/vendor/ggml/src/ggml-cuda/template-instances/mmf-instance-ncols_15.cu +5 -0
  1338. data/vendor/ggml/src/ggml-cuda/template-instances/mmf-instance-ncols_16.cu +5 -0
  1339. data/vendor/ggml/src/ggml-cuda/template-instances/mmf-instance-ncols_2.cu +5 -0
  1340. data/vendor/ggml/src/ggml-cuda/template-instances/mmf-instance-ncols_3.cu +5 -0
  1341. data/vendor/ggml/src/ggml-cuda/template-instances/mmf-instance-ncols_4.cu +5 -0
  1342. data/vendor/ggml/src/ggml-cuda/template-instances/mmf-instance-ncols_5.cu +5 -0
  1343. data/vendor/ggml/src/ggml-cuda/template-instances/mmf-instance-ncols_6.cu +5 -0
  1344. data/vendor/ggml/src/ggml-cuda/template-instances/mmf-instance-ncols_7.cu +5 -0
  1345. data/vendor/ggml/src/ggml-cuda/template-instances/mmf-instance-ncols_8.cu +5 -0
  1346. data/vendor/ggml/src/ggml-cuda/template-instances/mmf-instance-ncols_9.cu +5 -0
  1347. data/vendor/ggml/src/ggml-cuda/template-instances/mmq-instance-iq1_s.cu +5 -0
  1348. data/vendor/ggml/src/ggml-cuda/template-instances/mmq-instance-iq2_s.cu +5 -0
  1349. data/vendor/ggml/src/ggml-cuda/template-instances/mmq-instance-iq2_xs.cu +5 -0
  1350. data/vendor/ggml/src/ggml-cuda/template-instances/mmq-instance-iq2_xxs.cu +5 -0
  1351. data/vendor/ggml/src/ggml-cuda/template-instances/mmq-instance-iq3_s.cu +5 -0
  1352. data/vendor/ggml/src/ggml-cuda/template-instances/mmq-instance-iq3_xxs.cu +5 -0
  1353. data/vendor/ggml/src/ggml-cuda/template-instances/mmq-instance-iq4_nl.cu +5 -0
  1354. data/vendor/ggml/src/ggml-cuda/template-instances/mmq-instance-iq4_xs.cu +5 -0
  1355. data/vendor/ggml/src/ggml-cuda/template-instances/mmq-instance-mxfp4.cu +5 -0
  1356. data/vendor/ggml/src/ggml-cuda/template-instances/mmq-instance-nvfp4.cu +5 -0
  1357. data/vendor/ggml/src/ggml-cuda/template-instances/mmq-instance-q1_0.cu +5 -0
  1358. data/vendor/ggml/src/ggml-cuda/template-instances/mmq-instance-q2_k.cu +5 -0
  1359. data/vendor/ggml/src/ggml-cuda/template-instances/mmq-instance-q3_k.cu +5 -0
  1360. data/vendor/ggml/src/ggml-cuda/template-instances/mmq-instance-q4_0.cu +5 -0
  1361. data/vendor/ggml/src/ggml-cuda/template-instances/mmq-instance-q4_1.cu +5 -0
  1362. data/vendor/ggml/src/ggml-cuda/template-instances/mmq-instance-q4_k.cu +5 -0
  1363. data/vendor/ggml/src/ggml-cuda/template-instances/mmq-instance-q5_0.cu +5 -0
  1364. data/vendor/ggml/src/ggml-cuda/template-instances/mmq-instance-q5_1.cu +5 -0
  1365. data/vendor/ggml/src/ggml-cuda/template-instances/mmq-instance-q5_k.cu +5 -0
  1366. data/vendor/ggml/src/ggml-cuda/template-instances/mmq-instance-q6_k.cu +5 -0
  1367. data/vendor/ggml/src/ggml-cuda/template-instances/mmq-instance-q8_0.cu +5 -0
  1368. data/vendor/ggml/src/ggml-cuda/top-k.cu +95 -0
  1369. data/vendor/ggml/src/ggml-cuda/top-k.cuh +3 -0
  1370. data/vendor/ggml/src/ggml-cuda/topk-moe.cu +415 -0
  1371. data/vendor/ggml/src/ggml-cuda/topk-moe.cuh +27 -0
  1372. data/vendor/ggml/src/ggml-cuda/tri.cu +136 -0
  1373. data/vendor/ggml/src/ggml-cuda/tri.cuh +5 -0
  1374. data/vendor/ggml/src/ggml-cuda/tsembd.cu +47 -0
  1375. data/vendor/ggml/src/ggml-cuda/tsembd.cuh +5 -0
  1376. data/vendor/ggml/src/ggml-cuda/unary.cu +640 -0
  1377. data/vendor/ggml/src/ggml-cuda/unary.cuh +114 -0
  1378. data/vendor/ggml/src/ggml-cuda/upscale.cu +293 -0
  1379. data/vendor/ggml/src/ggml-cuda/upscale.cuh +5 -0
  1380. data/vendor/ggml/src/ggml-cuda/vecdotq.cuh +1317 -0
  1381. data/vendor/ggml/src/ggml-cuda/vendors/cuda.h +28 -0
  1382. data/vendor/ggml/src/ggml-cuda/vendors/hip.h +304 -0
  1383. data/vendor/ggml/src/ggml-cuda/vendors/musa.h +150 -0
  1384. data/vendor/ggml/src/ggml-cuda/wkv.cu +199 -0
  1385. data/vendor/ggml/src/ggml-cuda/wkv.cuh +7 -0
  1386. data/vendor/ggml/src/ggml-hexagon/CMakeLists.txt +118 -0
  1387. data/vendor/ggml/src/ggml-hexagon/ggml-hexagon.cpp +3680 -0
  1388. data/vendor/ggml/src/ggml-hexagon/htp/CMakeLists.txt +78 -0
  1389. data/vendor/ggml/src/ggml-hexagon/htp/act-ops.c +782 -0
  1390. data/vendor/ggml/src/ggml-hexagon/htp/argsort-ops.c +293 -0
  1391. data/vendor/ggml/src/ggml-hexagon/htp/binary-ops.c +872 -0
  1392. data/vendor/ggml/src/ggml-hexagon/htp/cmake-toolchain.cmake +157 -0
  1393. data/vendor/ggml/src/ggml-hexagon/htp/cpy-ops.c +275 -0
  1394. data/vendor/ggml/src/ggml-hexagon/htp/cumsum-ops.c +270 -0
  1395. data/vendor/ggml/src/ggml-hexagon/htp/diag-ops.c +216 -0
  1396. data/vendor/ggml/src/ggml-hexagon/htp/fill-ops.c +123 -0
  1397. data/vendor/ggml/src/ggml-hexagon/htp/flash-attn-ops.c +727 -0
  1398. data/vendor/ggml/src/ggml-hexagon/htp/gated-delta-net-ops.c +955 -0
  1399. data/vendor/ggml/src/ggml-hexagon/htp/get-rows-ops.c +124 -0
  1400. data/vendor/ggml/src/ggml-hexagon/htp/hex-dma.c +63 -0
  1401. data/vendor/ggml/src/ggml-hexagon/htp/hex-dma.h +372 -0
  1402. data/vendor/ggml/src/ggml-hexagon/htp/hex-dump.h +86 -0
  1403. data/vendor/ggml/src/ggml-hexagon/htp/hex-fastdiv.h +37 -0
  1404. data/vendor/ggml/src/ggml-hexagon/htp/hex-utils.h +137 -0
  1405. data/vendor/ggml/src/ggml-hexagon/htp/hmx-flash-attn-ops.c +1841 -0
  1406. data/vendor/ggml/src/ggml-hexagon/htp/hmx-matmul-ops.c +1785 -0
  1407. data/vendor/ggml/src/ggml-hexagon/htp/hmx-ops.h +71 -0
  1408. data/vendor/ggml/src/ggml-hexagon/htp/hmx-profile.h +34 -0
  1409. data/vendor/ggml/src/ggml-hexagon/htp/hmx-queue.c +158 -0
  1410. data/vendor/ggml/src/ggml-hexagon/htp/hmx-queue.h +134 -0
  1411. data/vendor/ggml/src/ggml-hexagon/htp/hmx-utils.h +200 -0
  1412. data/vendor/ggml/src/ggml-hexagon/htp/htp-ctx.h +111 -0
  1413. data/vendor/ggml/src/ggml-hexagon/htp/htp-ops.h +181 -0
  1414. data/vendor/ggml/src/ggml-hexagon/htp/htp_iface.idl +22 -0
  1415. data/vendor/ggml/src/ggml-hexagon/htp/hvx-arith.h +443 -0
  1416. data/vendor/ggml/src/ggml-hexagon/htp/hvx-base.h +308 -0
  1417. data/vendor/ggml/src/ggml-hexagon/htp/hvx-copy.h +262 -0
  1418. data/vendor/ggml/src/ggml-hexagon/htp/hvx-div.h +291 -0
  1419. data/vendor/ggml/src/ggml-hexagon/htp/hvx-dump.h +129 -0
  1420. data/vendor/ggml/src/ggml-hexagon/htp/hvx-exp.h +216 -0
  1421. data/vendor/ggml/src/ggml-hexagon/htp/hvx-floor.h +100 -0
  1422. data/vendor/ggml/src/ggml-hexagon/htp/hvx-inverse.h +210 -0
  1423. data/vendor/ggml/src/ggml-hexagon/htp/hvx-reduce.h +296 -0
  1424. data/vendor/ggml/src/ggml-hexagon/htp/hvx-repl.h +74 -0
  1425. data/vendor/ggml/src/ggml-hexagon/htp/hvx-scale.h +133 -0
  1426. data/vendor/ggml/src/ggml-hexagon/htp/hvx-sigmoid.h +142 -0
  1427. data/vendor/ggml/src/ggml-hexagon/htp/hvx-sqrt.h +126 -0
  1428. data/vendor/ggml/src/ggml-hexagon/htp/hvx-types.h +36 -0
  1429. data/vendor/ggml/src/ggml-hexagon/htp/hvx-utils.h +19 -0
  1430. data/vendor/ggml/src/ggml-hexagon/htp/main.c +880 -0
  1431. data/vendor/ggml/src/ggml-hexagon/htp/matmul-ops.c +3173 -0
  1432. data/vendor/ggml/src/ggml-hexagon/htp/repeat-ops.c +148 -0
  1433. data/vendor/ggml/src/ggml-hexagon/htp/rope-ops.c +494 -0
  1434. data/vendor/ggml/src/ggml-hexagon/htp/set-rows-ops.c +184 -0
  1435. data/vendor/ggml/src/ggml-hexagon/htp/softmax-ops.c +407 -0
  1436. data/vendor/ggml/src/ggml-hexagon/htp/solve-tri-ops.c +267 -0
  1437. data/vendor/ggml/src/ggml-hexagon/htp/ssm-conv.c +340 -0
  1438. data/vendor/ggml/src/ggml-hexagon/htp/sum-rows-ops.c +128 -0
  1439. data/vendor/ggml/src/ggml-hexagon/htp/unary-ops.c +657 -0
  1440. data/vendor/ggml/src/ggml-hexagon/htp/vtcm-utils.h +16 -0
  1441. data/vendor/ggml/src/ggml-hexagon/htp/worker-pool.c +293 -0
  1442. data/vendor/ggml/src/ggml-hexagon/htp/worker-pool.h +57 -0
  1443. data/vendor/ggml/src/ggml-hexagon/htp-drv.cpp +418 -0
  1444. data/vendor/ggml/src/ggml-hexagon/htp-drv.h +121 -0
  1445. data/vendor/ggml/src/ggml-hexagon/libdl.h +79 -0
  1446. data/vendor/ggml/src/ggml-hexagon/libggml-htp.inf +40 -0
  1447. data/vendor/ggml/src/ggml-hexagon/op-desc.h +153 -0
  1448. data/vendor/ggml/src/ggml-hip/CMakeLists.txt +157 -0
  1449. data/vendor/ggml/src/ggml-impl.h +783 -0
  1450. data/vendor/ggml/src/ggml-metal/CMakeLists.txt +124 -0
  1451. data/vendor/ggml/src/ggml-metal/ggml-metal-common.cpp +457 -0
  1452. data/vendor/ggml/src/ggml-metal/ggml-metal-common.h +52 -0
  1453. data/vendor/ggml/src/ggml-metal/ggml-metal-context.h +41 -0
  1454. data/vendor/ggml/src/ggml-metal/ggml-metal-context.m +739 -0
  1455. data/vendor/ggml/src/ggml-metal/ggml-metal-device.cpp +2053 -0
  1456. data/vendor/ggml/src/ggml-metal/ggml-metal-device.h +296 -0
  1457. data/vendor/ggml/src/ggml-metal/ggml-metal-device.m +1829 -0
  1458. data/vendor/ggml/src/ggml-metal/ggml-metal-impl.h +1175 -0
  1459. data/vendor/ggml/src/ggml-metal/ggml-metal-ops.cpp +4606 -0
  1460. data/vendor/ggml/src/ggml-metal/ggml-metal-ops.h +97 -0
  1461. data/vendor/ggml/src/ggml-metal/ggml-metal.cpp +950 -0
  1462. data/vendor/ggml/src/ggml-metal/ggml-metal.metal +10679 -0
  1463. data/vendor/ggml/src/ggml-musa/CMakeLists.txt +124 -0
  1464. data/vendor/ggml/src/ggml-musa/mudnn.cu +112 -0
  1465. data/vendor/ggml/src/ggml-musa/mudnn.cuh +12 -0
  1466. data/vendor/ggml/src/ggml-opencl/CMakeLists.txt +189 -0
  1467. data/vendor/ggml/src/ggml-opencl/ggml-opencl.cpp +16374 -0
  1468. data/vendor/ggml/src/ggml-opencl/kernels/add.cl +190 -0
  1469. data/vendor/ggml/src/ggml-opencl/kernels/add_id.cl +42 -0
  1470. data/vendor/ggml/src/ggml-opencl/kernels/argsort.cl +86 -0
  1471. data/vendor/ggml/src/ggml-opencl/kernels/clamp.cl +20 -0
  1472. data/vendor/ggml/src/ggml-opencl/kernels/concat.cl +51 -0
  1473. data/vendor/ggml/src/ggml-opencl/kernels/conv2d.cl +185 -0
  1474. data/vendor/ggml/src/ggml-opencl/kernels/conv2d_f16_f32.cl +176 -0
  1475. data/vendor/ggml/src/ggml-opencl/kernels/cpy.cl +229 -0
  1476. data/vendor/ggml/src/ggml-opencl/kernels/cumsum.cl +139 -0
  1477. data/vendor/ggml/src/ggml-opencl/kernels/cvt.cl +1471 -0
  1478. data/vendor/ggml/src/ggml-opencl/kernels/diag.cl +27 -0
  1479. data/vendor/ggml/src/ggml-opencl/kernels/diag_mask_inf.cl +58 -0
  1480. data/vendor/ggml/src/ggml-opencl/kernels/div.cl +138 -0
  1481. data/vendor/ggml/src/ggml-opencl/kernels/embed_kernel.py +26 -0
  1482. data/vendor/ggml/src/ggml-opencl/kernels/exp.cl +125 -0
  1483. data/vendor/ggml/src/ggml-opencl/kernels/expm1.cl +113 -0
  1484. data/vendor/ggml/src/ggml-opencl/kernels/fill.cl +17 -0
  1485. data/vendor/ggml/src/ggml-opencl/kernels/flash_attn_f16.cl +370 -0
  1486. data/vendor/ggml/src/ggml-opencl/kernels/flash_attn_f32.cl +371 -0
  1487. data/vendor/ggml/src/ggml-opencl/kernels/flash_attn_f32_f16.cl +373 -0
  1488. data/vendor/ggml/src/ggml-opencl/kernels/gelu.cl +89 -0
  1489. data/vendor/ggml/src/ggml-opencl/kernels/gemm_moe_mxfp4_f32.cl +162 -0
  1490. data/vendor/ggml/src/ggml-opencl/kernels/gemm_moe_mxfp4_f32_ns.cl +302 -0
  1491. data/vendor/ggml/src/ggml-opencl/kernels/gemm_moe_q4_0_f32_ns.cl +252 -0
  1492. data/vendor/ggml/src/ggml-opencl/kernels/gemm_moe_q4_1_f32_ns.cl +254 -0
  1493. data/vendor/ggml/src/ggml-opencl/kernels/gemm_moe_q5_0_f32_ns.cl +256 -0
  1494. data/vendor/ggml/src/ggml-opencl/kernels/gemm_moe_q5_1_f32_ns.cl +258 -0
  1495. data/vendor/ggml/src/ggml-opencl/kernels/gemm_noshuffle_iq4_nl_f32.cl +150 -0
  1496. data/vendor/ggml/src/ggml-opencl/kernels/gemm_noshuffle_q4_0_f32.cl +139 -0
  1497. data/vendor/ggml/src/ggml-opencl/kernels/gemm_noshuffle_q4_1_f32.cl +132 -0
  1498. data/vendor/ggml/src/ggml-opencl/kernels/gemm_noshuffle_q4_k_f32.cl +172 -0
  1499. data/vendor/ggml/src/ggml-opencl/kernels/gemm_noshuffle_q5_k_f32.cl +176 -0
  1500. data/vendor/ggml/src/ggml-opencl/kernels/gemm_noshuffle_q6_k_f32.cl +140 -0
  1501. data/vendor/ggml/src/ggml-opencl/kernels/gemm_noshuffle_q8_0_f32.cl +129 -0
  1502. data/vendor/ggml/src/ggml-opencl/kernels/gemm_xmem_f16_f32_os8.cl +233 -0
  1503. data/vendor/ggml/src/ggml-opencl/kernels/gemv_moe_mxfp4_f32.cl +156 -0
  1504. data/vendor/ggml/src/ggml-opencl/kernels/gemv_moe_mxfp4_f32_ns.cl +161 -0
  1505. data/vendor/ggml/src/ggml-opencl/kernels/gemv_moe_q4_0_f32_ns.cl +116 -0
  1506. data/vendor/ggml/src/ggml-opencl/kernels/gemv_moe_q4_1_f32_ns.cl +119 -0
  1507. data/vendor/ggml/src/ggml-opencl/kernels/gemv_moe_q5_0_f32_ns.cl +119 -0
  1508. data/vendor/ggml/src/ggml-opencl/kernels/gemv_moe_q5_1_f32_ns.cl +121 -0
  1509. data/vendor/ggml/src/ggml-opencl/kernels/gemv_noshuffle_iq4_nl_f32.cl +302 -0
  1510. data/vendor/ggml/src/ggml-opencl/kernels/gemv_noshuffle_q4_0_f32.cl +274 -0
  1511. data/vendor/ggml/src/ggml-opencl/kernels/gemv_noshuffle_q4_0_f32_spec.cl +268 -0
  1512. data/vendor/ggml/src/ggml-opencl/kernels/gemv_noshuffle_q4_1_f32.cl +283 -0
  1513. data/vendor/ggml/src/ggml-opencl/kernels/gemv_noshuffle_q4_k_f32.cl +318 -0
  1514. data/vendor/ggml/src/ggml-opencl/kernels/gemv_noshuffle_q5_k_f32.cl +326 -0
  1515. data/vendor/ggml/src/ggml-opencl/kernels/gemv_noshuffle_q6_k_f32.cl +293 -0
  1516. data/vendor/ggml/src/ggml-opencl/kernels/gemv_noshuffle_q8_0_f32.cl +195 -0
  1517. data/vendor/ggml/src/ggml-opencl/kernels/get_rows.cl +187 -0
  1518. data/vendor/ggml/src/ggml-opencl/kernels/glu.cl +378 -0
  1519. data/vendor/ggml/src/ggml-opencl/kernels/group_norm.cl +121 -0
  1520. data/vendor/ggml/src/ggml-opencl/kernels/im2col_f16.cl +57 -0
  1521. data/vendor/ggml/src/ggml-opencl/kernels/im2col_f32.cl +57 -0
  1522. data/vendor/ggml/src/ggml-opencl/kernels/l2_norm.cl +71 -0
  1523. data/vendor/ggml/src/ggml-opencl/kernels/mean.cl +140 -0
  1524. data/vendor/ggml/src/ggml-opencl/kernels/moe_reorder_b.cl +30 -0
  1525. data/vendor/ggml/src/ggml-opencl/kernels/moe_sort_by_expert.cl +82 -0
  1526. data/vendor/ggml/src/ggml-opencl/kernels/mul.cl +152 -0
  1527. data/vendor/ggml/src/ggml-opencl/kernels/mul_mat_f16_f32.cl +130 -0
  1528. data/vendor/ggml/src/ggml-opencl/kernels/mul_mm_f16_f32_kq_kqv.cl +273 -0
  1529. data/vendor/ggml/src/ggml-opencl/kernels/mul_mm_f16_f32_l4_lm.cl +146 -0
  1530. data/vendor/ggml/src/ggml-opencl/kernels/mul_mm_f32_f32_l4_lm.cl +147 -0
  1531. data/vendor/ggml/src/ggml-opencl/kernels/mul_mm_iq4_nl_f32_l4_lm.cl +171 -0
  1532. data/vendor/ggml/src/ggml-opencl/kernels/mul_mm_q4_0_f32_l4_lm.cl +163 -0
  1533. data/vendor/ggml/src/ggml-opencl/kernels/mul_mm_q4_1_f32_l4_lm.cl +165 -0
  1534. data/vendor/ggml/src/ggml-opencl/kernels/mul_mm_q4_k_f32_l4_lm.cl +179 -0
  1535. data/vendor/ggml/src/ggml-opencl/kernels/mul_mm_q5_k_f32_l4_lm.cl +192 -0
  1536. data/vendor/ggml/src/ggml-opencl/kernels/mul_mm_q6_k_f32_l4_lm.cl +158 -0
  1537. data/vendor/ggml/src/ggml-opencl/kernels/mul_mm_q8_0_f32_l4_lm.cl +154 -0
  1538. data/vendor/ggml/src/ggml-opencl/kernels/mul_mv_f16_f16.cl +118 -0
  1539. data/vendor/ggml/src/ggml-opencl/kernels/mul_mv_f16_f32.cl +118 -0
  1540. data/vendor/ggml/src/ggml-opencl/kernels/mul_mv_f16_f32_1row.cl +94 -0
  1541. data/vendor/ggml/src/ggml-opencl/kernels/mul_mv_f16_f32_l4.cl +84 -0
  1542. data/vendor/ggml/src/ggml-opencl/kernels/mul_mv_f32_f32.cl +118 -0
  1543. data/vendor/ggml/src/ggml-opencl/kernels/mul_mv_id_mxfp4_f32.cl +189 -0
  1544. data/vendor/ggml/src/ggml-opencl/kernels/mul_mv_id_mxfp4_f32_flat.cl +176 -0
  1545. data/vendor/ggml/src/ggml-opencl/kernels/mul_mv_id_q4_0_f32_8x_flat.cl +283 -0
  1546. data/vendor/ggml/src/ggml-opencl/kernels/mul_mv_id_q8_0_f32.cl +140 -0
  1547. data/vendor/ggml/src/ggml-opencl/kernels/mul_mv_id_q8_0_f32_flat.cl +222 -0
  1548. data/vendor/ggml/src/ggml-opencl/kernels/mul_mv_iq4_nl_f32.cl +164 -0
  1549. data/vendor/ggml/src/ggml-opencl/kernels/mul_mv_iq4_nl_f32_flat.cl +202 -0
  1550. data/vendor/ggml/src/ggml-opencl/kernels/mul_mv_mxfp4_f32.cl +144 -0
  1551. data/vendor/ggml/src/ggml-opencl/kernels/mul_mv_mxfp4_f32_flat.cl +167 -0
  1552. data/vendor/ggml/src/ggml-opencl/kernels/mul_mv_q4_0_f32.cl +192 -0
  1553. data/vendor/ggml/src/ggml-opencl/kernels/mul_mv_q4_0_f32_1d_16x_flat.cl +307 -0
  1554. data/vendor/ggml/src/ggml-opencl/kernels/mul_mv_q4_0_f32_1d_8x_flat.cl +265 -0
  1555. data/vendor/ggml/src/ggml-opencl/kernels/mul_mv_q4_0_f32_8x_flat.cl +272 -0
  1556. data/vendor/ggml/src/ggml-opencl/kernels/mul_mv_q4_0_f32_v.cl +254 -0
  1557. data/vendor/ggml/src/ggml-opencl/kernels/mul_mv_q4_1_f32.cl +219 -0
  1558. data/vendor/ggml/src/ggml-opencl/kernels/mul_mv_q4_1_f32_flat.cl +229 -0
  1559. data/vendor/ggml/src/ggml-opencl/kernels/mul_mv_q4_k_f32.cl +180 -0
  1560. data/vendor/ggml/src/ggml-opencl/kernels/mul_mv_q4_k_f32_flat.cl +196 -0
  1561. data/vendor/ggml/src/ggml-opencl/kernels/mul_mv_q5_k_f32.cl +187 -0
  1562. data/vendor/ggml/src/ggml-opencl/kernels/mul_mv_q5_k_f32_flat.cl +203 -0
  1563. data/vendor/ggml/src/ggml-opencl/kernels/mul_mv_q6_k_f32.cl +194 -0
  1564. data/vendor/ggml/src/ggml-opencl/kernels/mul_mv_q6_k_f32_flat.cl +194 -0
  1565. data/vendor/ggml/src/ggml-opencl/kernels/mul_mv_q8_0_f32.cl +125 -0
  1566. data/vendor/ggml/src/ggml-opencl/kernels/mul_mv_q8_0_f32_flat.cl +202 -0
  1567. data/vendor/ggml/src/ggml-opencl/kernels/neg.cl +125 -0
  1568. data/vendor/ggml/src/ggml-opencl/kernels/norm.cl +161 -0
  1569. data/vendor/ggml/src/ggml-opencl/kernels/pad.cl +39 -0
  1570. data/vendor/ggml/src/ggml-opencl/kernels/relu.cl +16 -0
  1571. data/vendor/ggml/src/ggml-opencl/kernels/repeat.cl +38 -0
  1572. data/vendor/ggml/src/ggml-opencl/kernels/rms_norm.cl +190 -0
  1573. data/vendor/ggml/src/ggml-opencl/kernels/rope.cl +747 -0
  1574. data/vendor/ggml/src/ggml-opencl/kernels/scale.cl +27 -0
  1575. data/vendor/ggml/src/ggml-opencl/kernels/set_rows.cl +208 -0
  1576. data/vendor/ggml/src/ggml-opencl/kernels/sigmoid.cl +29 -0
  1577. data/vendor/ggml/src/ggml-opencl/kernels/silu.cl +30 -0
  1578. data/vendor/ggml/src/ggml-opencl/kernels/softmax_4_f16.cl +108 -0
  1579. data/vendor/ggml/src/ggml-opencl/kernels/softmax_4_f32.cl +108 -0
  1580. data/vendor/ggml/src/ggml-opencl/kernels/softmax_f16.cl +107 -0
  1581. data/vendor/ggml/src/ggml-opencl/kernels/softmax_f32.cl +107 -0
  1582. data/vendor/ggml/src/ggml-opencl/kernels/softplus.cl +116 -0
  1583. data/vendor/ggml/src/ggml-opencl/kernels/solve_tri.cl +51 -0
  1584. data/vendor/ggml/src/ggml-opencl/kernels/sqr.cl +53 -0
  1585. data/vendor/ggml/src/ggml-opencl/kernels/sqrt.cl +53 -0
  1586. data/vendor/ggml/src/ggml-opencl/kernels/ssm_conv.cl +77 -0
  1587. data/vendor/ggml/src/ggml-opencl/kernels/sub.cl +138 -0
  1588. data/vendor/ggml/src/ggml-opencl/kernels/sum_rows.cl +140 -0
  1589. data/vendor/ggml/src/ggml-opencl/kernels/tanh.cl +109 -0
  1590. data/vendor/ggml/src/ggml-opencl/kernels/transpose.cl +143 -0
  1591. data/vendor/ggml/src/ggml-opencl/kernels/tri.cl +32 -0
  1592. data/vendor/ggml/src/ggml-opencl/kernels/tsembd.cl +48 -0
  1593. data/vendor/ggml/src/ggml-opencl/kernels/upscale.cl +120 -0
  1594. data/vendor/ggml/src/ggml-openvino/CMakeLists.txt +22 -0
  1595. data/vendor/ggml/src/ggml-openvino/ggml-decoder.cpp +985 -0
  1596. data/vendor/ggml/src/ggml-openvino/ggml-decoder.h +294 -0
  1597. data/vendor/ggml/src/ggml-openvino/ggml-openvino-extra.cpp +380 -0
  1598. data/vendor/ggml/src/ggml-openvino/ggml-openvino-extra.h +182 -0
  1599. data/vendor/ggml/src/ggml-openvino/ggml-openvino.cpp +1132 -0
  1600. data/vendor/ggml/src/ggml-openvino/ggml-quants.cpp +956 -0
  1601. data/vendor/ggml/src/ggml-openvino/ggml-quants.h +153 -0
  1602. data/vendor/ggml/src/ggml-openvino/openvino/decoder.h +74 -0
  1603. data/vendor/ggml/src/ggml-openvino/openvino/frontend.cpp +27 -0
  1604. data/vendor/ggml/src/ggml-openvino/openvino/frontend.h +23 -0
  1605. data/vendor/ggml/src/ggml-openvino/openvino/input_model.cpp +17 -0
  1606. data/vendor/ggml/src/ggml-openvino/openvino/input_model.h +29 -0
  1607. data/vendor/ggml/src/ggml-openvino/openvino/node_context.h +112 -0
  1608. data/vendor/ggml/src/ggml-openvino/openvino/op/cont.cpp +48 -0
  1609. data/vendor/ggml/src/ggml-openvino/openvino/op/cpy.cpp +21 -0
  1610. data/vendor/ggml/src/ggml-openvino/openvino/op/flash_attn_ext.cpp +90 -0
  1611. data/vendor/ggml/src/ggml-openvino/openvino/op/get_rows.cpp +69 -0
  1612. data/vendor/ggml/src/ggml-openvino/openvino/op/glu_geglu.cpp +61 -0
  1613. data/vendor/ggml/src/ggml-openvino/openvino/op/glu_swiglu.cpp +62 -0
  1614. data/vendor/ggml/src/ggml-openvino/openvino/op/mulmat.cpp +90 -0
  1615. data/vendor/ggml/src/ggml-openvino/openvino/op/permute.cpp +102 -0
  1616. data/vendor/ggml/src/ggml-openvino/openvino/op/reshape.cpp +83 -0
  1617. data/vendor/ggml/src/ggml-openvino/openvino/op/rms_norm.cpp +46 -0
  1618. data/vendor/ggml/src/ggml-openvino/openvino/op/rope.cpp +149 -0
  1619. data/vendor/ggml/src/ggml-openvino/openvino/op/scale.cpp +41 -0
  1620. data/vendor/ggml/src/ggml-openvino/openvino/op/set_rows.cpp +76 -0
  1621. data/vendor/ggml/src/ggml-openvino/openvino/op/softmax.cpp +89 -0
  1622. data/vendor/ggml/src/ggml-openvino/openvino/op/transpose.cpp +23 -0
  1623. data/vendor/ggml/src/ggml-openvino/openvino/op/unary_gelu.cpp +25 -0
  1624. data/vendor/ggml/src/ggml-openvino/openvino/op/unary_silu.cpp +27 -0
  1625. data/vendor/ggml/src/ggml-openvino/openvino/op/view.cpp +53 -0
  1626. data/vendor/ggml/src/ggml-openvino/openvino/op_table.cpp +47 -0
  1627. data/vendor/ggml/src/ggml-openvino/openvino/op_table.h +40 -0
  1628. data/vendor/ggml/src/ggml-openvino/openvino/pass/fuse_to_sdpa.cpp +60 -0
  1629. data/vendor/ggml/src/ggml-openvino/openvino/pass/fuse_to_sdpa.h +17 -0
  1630. data/vendor/ggml/src/ggml-openvino/openvino/pass/mark_decompression_convert_constant_folding.h +29 -0
  1631. data/vendor/ggml/src/ggml-openvino/openvino/pass/squeeze_matmul.cpp +58 -0
  1632. data/vendor/ggml/src/ggml-openvino/openvino/pass/squeeze_matmul.h +17 -0
  1633. data/vendor/ggml/src/ggml-openvino/openvino/rt_info/weightless_caching_attributes.hpp +41 -0
  1634. data/vendor/ggml/src/ggml-openvino/openvino/translate_session.cpp +317 -0
  1635. data/vendor/ggml/src/ggml-openvino/openvino/translate_session.h +28 -0
  1636. data/vendor/ggml/src/ggml-openvino/openvino/utils.cpp +257 -0
  1637. data/vendor/ggml/src/ggml-openvino/openvino/utils.h +86 -0
  1638. data/vendor/ggml/src/ggml-openvino/utils.cpp +880 -0
  1639. data/vendor/ggml/src/ggml-openvino/utils.h +143 -0
  1640. data/vendor/ggml/src/ggml-opt.cpp +1094 -0
  1641. data/vendor/ggml/src/ggml-quants.c +5491 -0
  1642. data/vendor/ggml/src/ggml-quants.h +112 -0
  1643. data/vendor/ggml/src/ggml-rpc/CMakeLists.txt +33 -0
  1644. data/vendor/ggml/src/ggml-rpc/ggml-rpc.cpp +1974 -0
  1645. data/vendor/ggml/src/ggml-rpc/transport.cpp +683 -0
  1646. data/vendor/ggml/src/ggml-rpc/transport.h +34 -0
  1647. data/vendor/ggml/src/ggml-sycl/CMakeLists.txt +207 -0
  1648. data/vendor/ggml/src/ggml-sycl/add-id.cpp +81 -0
  1649. data/vendor/ggml/src/ggml-sycl/add-id.hpp +8 -0
  1650. data/vendor/ggml/src/ggml-sycl/backend.hpp +48 -0
  1651. data/vendor/ggml/src/ggml-sycl/binbcast.cpp +346 -0
  1652. data/vendor/ggml/src/ggml-sycl/binbcast.hpp +39 -0
  1653. data/vendor/ggml/src/ggml-sycl/common.cpp +155 -0
  1654. data/vendor/ggml/src/ggml-sycl/common.hpp +1002 -0
  1655. data/vendor/ggml/src/ggml-sycl/concat.cpp +202 -0
  1656. data/vendor/ggml/src/ggml-sycl/concat.hpp +20 -0
  1657. data/vendor/ggml/src/ggml-sycl/conv.cpp +101 -0
  1658. data/vendor/ggml/src/ggml-sycl/conv.hpp +20 -0
  1659. data/vendor/ggml/src/ggml-sycl/convert.cpp +825 -0
  1660. data/vendor/ggml/src/ggml-sycl/convert.hpp +64 -0
  1661. data/vendor/ggml/src/ggml-sycl/count-equal.cpp +79 -0
  1662. data/vendor/ggml/src/ggml-sycl/count-equal.hpp +9 -0
  1663. data/vendor/ggml/src/ggml-sycl/cpy.cpp +602 -0
  1664. data/vendor/ggml/src/ggml-sycl/cpy.hpp +223 -0
  1665. data/vendor/ggml/src/ggml-sycl/cumsum.cpp +148 -0
  1666. data/vendor/ggml/src/ggml-sycl/cumsum.hpp +5 -0
  1667. data/vendor/ggml/src/ggml-sycl/dequantize.hpp +975 -0
  1668. data/vendor/ggml/src/ggml-sycl/diag.cpp +67 -0
  1669. data/vendor/ggml/src/ggml-sycl/diag.hpp +5 -0
  1670. data/vendor/ggml/src/ggml-sycl/dmmv.cpp +1579 -0
  1671. data/vendor/ggml/src/ggml-sycl/dmmv.hpp +27 -0
  1672. data/vendor/ggml/src/ggml-sycl/dpct/helper.hpp +3774 -0
  1673. data/vendor/ggml/src/ggml-sycl/element_wise.cpp +1124 -0
  1674. data/vendor/ggml/src/ggml-sycl/element_wise.hpp +94 -0
  1675. data/vendor/ggml/src/ggml-sycl/fattn-buffers.cpp +56 -0
  1676. data/vendor/ggml/src/ggml-sycl/fattn-buffers.hpp +63 -0
  1677. data/vendor/ggml/src/ggml-sycl/fattn-common.hpp +1181 -0
  1678. data/vendor/ggml/src/ggml-sycl/fattn-tile.cpp +59 -0
  1679. data/vendor/ggml/src/ggml-sycl/fattn-tile.hpp +1246 -0
  1680. data/vendor/ggml/src/ggml-sycl/fattn-vec.hpp +674 -0
  1681. data/vendor/ggml/src/ggml-sycl/fattn.cpp +227 -0
  1682. data/vendor/ggml/src/ggml-sycl/fattn.hpp +22 -0
  1683. data/vendor/ggml/src/ggml-sycl/fill.cpp +55 -0
  1684. data/vendor/ggml/src/ggml-sycl/fill.hpp +5 -0
  1685. data/vendor/ggml/src/ggml-sycl/gated_delta_net.cpp +307 -0
  1686. data/vendor/ggml/src/ggml-sycl/gated_delta_net.hpp +9 -0
  1687. data/vendor/ggml/src/ggml-sycl/gemm.hpp +93 -0
  1688. data/vendor/ggml/src/ggml-sycl/getrows.cpp +219 -0
  1689. data/vendor/ggml/src/ggml-sycl/getrows.hpp +20 -0
  1690. data/vendor/ggml/src/ggml-sycl/ggml-sycl.cpp +5520 -0
  1691. data/vendor/ggml/src/ggml-sycl/gla.cpp +106 -0
  1692. data/vendor/ggml/src/ggml-sycl/gla.hpp +8 -0
  1693. data/vendor/ggml/src/ggml-sycl/im2col.cpp +400 -0
  1694. data/vendor/ggml/src/ggml-sycl/im2col.hpp +23 -0
  1695. data/vendor/ggml/src/ggml-sycl/mmq.cpp +3030 -0
  1696. data/vendor/ggml/src/ggml-sycl/mmq.hpp +33 -0
  1697. data/vendor/ggml/src/ggml-sycl/mmvq.cpp +1380 -0
  1698. data/vendor/ggml/src/ggml-sycl/mmvq.hpp +43 -0
  1699. data/vendor/ggml/src/ggml-sycl/norm.cpp +656 -0
  1700. data/vendor/ggml/src/ggml-sycl/norm.hpp +28 -0
  1701. data/vendor/ggml/src/ggml-sycl/outprod.cpp +47 -0
  1702. data/vendor/ggml/src/ggml-sycl/outprod.hpp +10 -0
  1703. data/vendor/ggml/src/ggml-sycl/pad.cpp +97 -0
  1704. data/vendor/ggml/src/ggml-sycl/pad.hpp +24 -0
  1705. data/vendor/ggml/src/ggml-sycl/pad_reflect_1d.cpp +100 -0
  1706. data/vendor/ggml/src/ggml-sycl/pad_reflect_1d.hpp +10 -0
  1707. data/vendor/ggml/src/ggml-sycl/presets.hpp +79 -0
  1708. data/vendor/ggml/src/ggml-sycl/quantize.hpp +133 -0
  1709. data/vendor/ggml/src/ggml-sycl/quants.hpp +156 -0
  1710. data/vendor/ggml/src/ggml-sycl/repeat_back.cpp +76 -0
  1711. data/vendor/ggml/src/ggml-sycl/repeat_back.hpp +8 -0
  1712. data/vendor/ggml/src/ggml-sycl/roll.cpp +122 -0
  1713. data/vendor/ggml/src/ggml-sycl/roll.hpp +20 -0
  1714. data/vendor/ggml/src/ggml-sycl/rope.cpp +641 -0
  1715. data/vendor/ggml/src/ggml-sycl/rope.hpp +26 -0
  1716. data/vendor/ggml/src/ggml-sycl/set.cpp +73 -0
  1717. data/vendor/ggml/src/ggml-sycl/set.hpp +5 -0
  1718. data/vendor/ggml/src/ggml-sycl/set_rows.cpp +240 -0
  1719. data/vendor/ggml/src/ggml-sycl/set_rows.hpp +8 -0
  1720. data/vendor/ggml/src/ggml-sycl/softmax.cpp +426 -0
  1721. data/vendor/ggml/src/ggml-sycl/softmax.hpp +24 -0
  1722. data/vendor/ggml/src/ggml-sycl/solve_tri.cpp +172 -0
  1723. data/vendor/ggml/src/ggml-sycl/solve_tri.hpp +8 -0
  1724. data/vendor/ggml/src/ggml-sycl/ssm_conv.cpp +132 -0
  1725. data/vendor/ggml/src/ggml-sycl/ssm_conv.hpp +5 -0
  1726. data/vendor/ggml/src/ggml-sycl/ssm_scan.cpp +156 -0
  1727. data/vendor/ggml/src/ggml-sycl/ssm_scan.hpp +5 -0
  1728. data/vendor/ggml/src/ggml-sycl/sycl_hw.cpp +67 -0
  1729. data/vendor/ggml/src/ggml-sycl/sycl_hw.hpp +38 -0
  1730. data/vendor/ggml/src/ggml-sycl/template-instances/fattn-tile-instance-dkq112-dv112.cpp +5 -0
  1731. data/vendor/ggml/src/ggml-sycl/template-instances/fattn-tile-instance-dkq128-dv128.cpp +5 -0
  1732. data/vendor/ggml/src/ggml-sycl/template-instances/fattn-tile-instance-dkq256-dv256.cpp +5 -0
  1733. data/vendor/ggml/src/ggml-sycl/template-instances/fattn-tile-instance-dkq40-dv40.cpp +5 -0
  1734. data/vendor/ggml/src/ggml-sycl/template-instances/fattn-tile-instance-dkq512-dv512.cpp +6 -0
  1735. data/vendor/ggml/src/ggml-sycl/template-instances/fattn-tile-instance-dkq576-dv512.cpp +5 -0
  1736. data/vendor/ggml/src/ggml-sycl/template-instances/fattn-tile-instance-dkq64-dv64.cpp +5 -0
  1737. data/vendor/ggml/src/ggml-sycl/template-instances/fattn-tile-instance-dkq72-dv72.cpp +5 -0
  1738. data/vendor/ggml/src/ggml-sycl/template-instances/fattn-tile-instance-dkq80-dv80.cpp +5 -0
  1739. data/vendor/ggml/src/ggml-sycl/template-instances/fattn-tile-instance-dkq96-dv96.cpp +5 -0
  1740. data/vendor/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-f16-f16.cpp +8 -0
  1741. data/vendor/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-f16-q4_0.cpp +8 -0
  1742. data/vendor/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-f16-q4_1.cpp +8 -0
  1743. data/vendor/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-f16-q5_0.cpp +8 -0
  1744. data/vendor/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-f16-q5_1.cpp +8 -0
  1745. data/vendor/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-f16-q8_0.cpp +8 -0
  1746. data/vendor/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q4_0-f16.cpp +8 -0
  1747. data/vendor/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q4_0-q4_0.cpp +8 -0
  1748. data/vendor/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q4_0-q4_1.cpp +8 -0
  1749. data/vendor/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q4_0-q5_0.cpp +8 -0
  1750. data/vendor/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q4_0-q5_1.cpp +8 -0
  1751. data/vendor/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q4_0-q8_0.cpp +8 -0
  1752. data/vendor/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q4_1-f16.cpp +8 -0
  1753. data/vendor/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q4_1-q4_0.cpp +8 -0
  1754. data/vendor/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q4_1-q4_1.cpp +8 -0
  1755. data/vendor/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q4_1-q5_0.cpp +8 -0
  1756. data/vendor/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q4_1-q5_1.cpp +8 -0
  1757. data/vendor/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q4_1-q8_0.cpp +8 -0
  1758. data/vendor/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q5_0-f16.cpp +8 -0
  1759. data/vendor/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q5_0-q4_0.cpp +8 -0
  1760. data/vendor/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q5_0-q4_1.cpp +8 -0
  1761. data/vendor/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q5_0-q5_0.cpp +8 -0
  1762. data/vendor/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q5_0-q5_1.cpp +8 -0
  1763. data/vendor/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q5_0-q8_0.cpp +8 -0
  1764. data/vendor/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q5_1-f16.cpp +8 -0
  1765. data/vendor/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q5_1-q4_0.cpp +8 -0
  1766. data/vendor/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q5_1-q4_1.cpp +8 -0
  1767. data/vendor/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q5_1-q5_0.cpp +8 -0
  1768. data/vendor/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q5_1-q5_1.cpp +8 -0
  1769. data/vendor/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q5_1-q8_0.cpp +8 -0
  1770. data/vendor/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q8_0-f16.cpp +8 -0
  1771. data/vendor/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q8_0-q4_0.cpp +8 -0
  1772. data/vendor/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q8_0-q4_1.cpp +8 -0
  1773. data/vendor/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q8_0-q5_0.cpp +8 -0
  1774. data/vendor/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q8_0-q5_1.cpp +8 -0
  1775. data/vendor/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q8_0-q8_0.cpp +8 -0
  1776. data/vendor/ggml/src/ggml-sycl/tsembd.cpp +73 -0
  1777. data/vendor/ggml/src/ggml-sycl/tsembd.hpp +20 -0
  1778. data/vendor/ggml/src/ggml-sycl/type.hpp +112 -0
  1779. data/vendor/ggml/src/ggml-sycl/upscale.cpp +410 -0
  1780. data/vendor/ggml/src/ggml-sycl/upscale.hpp +9 -0
  1781. data/vendor/ggml/src/ggml-sycl/vecdotq.hpp +1508 -0
  1782. data/vendor/ggml/src/ggml-sycl/wkv.cpp +293 -0
  1783. data/vendor/ggml/src/ggml-sycl/wkv.hpp +10 -0
  1784. data/vendor/ggml/src/ggml-threading.cpp +12 -0
  1785. data/vendor/ggml/src/ggml-threading.h +14 -0
  1786. data/vendor/ggml/src/ggml-virtgpu/CMakeLists.txt +70 -0
  1787. data/vendor/ggml/src/ggml-virtgpu/apir_cs_ggml-rpc-front.cpp +87 -0
  1788. data/vendor/ggml/src/ggml-virtgpu/backend/CMakeLists.txt +21 -0
  1789. data/vendor/ggml/src/ggml-virtgpu/backend/apir_cs_ggml-rpc-back.cpp +115 -0
  1790. data/vendor/ggml/src/ggml-virtgpu/backend/backend-convert.h +13 -0
  1791. data/vendor/ggml/src/ggml-virtgpu/backend/backend-dispatched-backend.cpp +102 -0
  1792. data/vendor/ggml/src/ggml-virtgpu/backend/backend-dispatched-buffer-type.cpp +105 -0
  1793. data/vendor/ggml/src/ggml-virtgpu/backend/backend-dispatched-buffer.cpp +179 -0
  1794. data/vendor/ggml/src/ggml-virtgpu/backend/backend-dispatched-device.cpp +148 -0
  1795. data/vendor/ggml/src/ggml-virtgpu/backend/backend-dispatched.cpp +51 -0
  1796. data/vendor/ggml/src/ggml-virtgpu/backend/backend-dispatched.gen.h +73 -0
  1797. data/vendor/ggml/src/ggml-virtgpu/backend/backend-dispatched.h +27 -0
  1798. data/vendor/ggml/src/ggml-virtgpu/backend/backend-virgl-apir.h +32 -0
  1799. data/vendor/ggml/src/ggml-virtgpu/backend/backend.cpp +144 -0
  1800. data/vendor/ggml/src/ggml-virtgpu/backend/shared/api_remoting.h +95 -0
  1801. data/vendor/ggml/src/ggml-virtgpu/backend/shared/apir_backend.gen.h +94 -0
  1802. data/vendor/ggml/src/ggml-virtgpu/backend/shared/apir_backend.h +50 -0
  1803. data/vendor/ggml/src/ggml-virtgpu/backend/shared/apir_cs.h +378 -0
  1804. data/vendor/ggml/src/ggml-virtgpu/backend/shared/apir_cs_ggml.h +232 -0
  1805. data/vendor/ggml/src/ggml-virtgpu/backend/shared/apir_cs_rpc.h +58 -0
  1806. data/vendor/ggml/src/ggml-virtgpu/ggml-backend-buffer-type.cpp +81 -0
  1807. data/vendor/ggml/src/ggml-virtgpu/ggml-backend-buffer.cpp +123 -0
  1808. data/vendor/ggml/src/ggml-virtgpu/ggml-backend-device.cpp +160 -0
  1809. data/vendor/ggml/src/ggml-virtgpu/ggml-backend-reg.cpp +213 -0
  1810. data/vendor/ggml/src/ggml-virtgpu/ggml-backend.cpp +71 -0
  1811. data/vendor/ggml/src/ggml-virtgpu/ggml-remoting.h +71 -0
  1812. data/vendor/ggml/src/ggml-virtgpu/ggmlremoting_functions.yaml +166 -0
  1813. data/vendor/ggml/src/ggml-virtgpu/include/apir_hw.h +9 -0
  1814. data/vendor/ggml/src/ggml-virtgpu/regenerate_remoting.py +333 -0
  1815. data/vendor/ggml/src/ggml-virtgpu/virtgpu-apir.h +15 -0
  1816. data/vendor/ggml/src/ggml-virtgpu/virtgpu-forward-backend.cpp +58 -0
  1817. data/vendor/ggml/src/ggml-virtgpu/virtgpu-forward-buffer-type.cpp +110 -0
  1818. data/vendor/ggml/src/ggml-virtgpu/virtgpu-forward-buffer.cpp +173 -0
  1819. data/vendor/ggml/src/ggml-virtgpu/virtgpu-forward-device.cpp +192 -0
  1820. data/vendor/ggml/src/ggml-virtgpu/virtgpu-forward-impl.h +36 -0
  1821. data/vendor/ggml/src/ggml-virtgpu/virtgpu-forward.gen.h +53 -0
  1822. data/vendor/ggml/src/ggml-virtgpu/virtgpu-shm.cpp +99 -0
  1823. data/vendor/ggml/src/ggml-virtgpu/virtgpu-shm.h +23 -0
  1824. data/vendor/ggml/src/ggml-virtgpu/virtgpu-utils.cpp +179 -0
  1825. data/vendor/ggml/src/ggml-virtgpu/virtgpu-utils.h +86 -0
  1826. data/vendor/ggml/src/ggml-virtgpu/virtgpu.cpp +545 -0
  1827. data/vendor/ggml/src/ggml-virtgpu/virtgpu.h +115 -0
  1828. data/vendor/ggml/src/ggml-vulkan/CMakeLists.txt +220 -0
  1829. data/vendor/ggml/src/ggml-vulkan/cmake/host-toolchain.cmake.in +15 -0
  1830. data/vendor/ggml/src/ggml-vulkan/ggml-vulkan.cpp +17208 -0
  1831. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/CMakeLists.txt +31 -0
  1832. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/abs.comp +21 -0
  1833. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/acc.comp +37 -0
  1834. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/add.comp +69 -0
  1835. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/add1.comp +28 -0
  1836. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/add_id.comp +42 -0
  1837. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/arange.comp +20 -0
  1838. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/argmax.comp +60 -0
  1839. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/argsort.comp +86 -0
  1840. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/argsort_large.comp +114 -0
  1841. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/ceil.comp +22 -0
  1842. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/clamp.comp +17 -0
  1843. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/concat.comp +41 -0
  1844. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/contig_copy.comp +49 -0
  1845. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/conv2d_dw.comp +105 -0
  1846. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/conv2d_mm.comp +347 -0
  1847. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/conv_transpose_1d.comp +98 -0
  1848. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/copy.comp +23 -0
  1849. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/copy_from_quant.comp +51 -0
  1850. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/copy_to_quant.comp +320 -0
  1851. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/copy_transpose.comp +67 -0
  1852. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/cos.comp +17 -0
  1853. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/count_equal.comp +31 -0
  1854. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/count_experts.comp +51 -0
  1855. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/cumsum.comp +83 -0
  1856. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/cumsum_multipass1.comp +60 -0
  1857. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/cumsum_multipass2.comp +66 -0
  1858. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/dequant_f32.comp +20 -0
  1859. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/dequant_funcs.glsl +653 -0
  1860. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/dequant_funcs_cm2.glsl +768 -0
  1861. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/dequant_head.glsl +13 -0
  1862. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/dequant_iq1_m.comp +42 -0
  1863. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/dequant_iq1_s.comp +35 -0
  1864. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/dequant_iq2_s.comp +44 -0
  1865. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/dequant_iq2_xs.comp +43 -0
  1866. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/dequant_iq2_xxs.comp +49 -0
  1867. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/dequant_iq3_s.comp +40 -0
  1868. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/dequant_iq3_xxs.comp +51 -0
  1869. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/dequant_iq4_nl.comp +32 -0
  1870. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/dequant_iq4_xs.comp +34 -0
  1871. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/dequant_mxfp4.comp +32 -0
  1872. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/dequant_nvfp4.comp +32 -0
  1873. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/dequant_q1_0.comp +29 -0
  1874. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/dequant_q2_k.comp +34 -0
  1875. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/dequant_q3_k.comp +42 -0
  1876. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/dequant_q4_0.comp +30 -0
  1877. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/dequant_q4_1.comp +32 -0
  1878. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/dequant_q4_k.comp +68 -0
  1879. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/dequant_q5_0.comp +34 -0
  1880. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/dequant_q5_1.comp +35 -0
  1881. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/dequant_q5_k.comp +70 -0
  1882. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/dequant_q6_k.comp +33 -0
  1883. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/dequant_q8_0.comp +31 -0
  1884. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/diag.comp +28 -0
  1885. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/diag_mask_inf.comp +34 -0
  1886. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/div.comp +27 -0
  1887. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/elu.comp +27 -0
  1888. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/exp.comp +20 -0
  1889. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/feature-tests/bfloat16.comp +7 -0
  1890. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/feature-tests/coopmat.comp +7 -0
  1891. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/feature-tests/coopmat2.comp +7 -0
  1892. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/feature-tests/integer_dot.comp +7 -0
  1893. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/fill.comp +19 -0
  1894. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/flash_attn.comp +756 -0
  1895. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/flash_attn_base.glsl +255 -0
  1896. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/flash_attn_cm1.comp +626 -0
  1897. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/flash_attn_cm2.comp +427 -0
  1898. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/flash_attn_dequant.glsl +123 -0
  1899. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/flash_attn_mask_opt.comp +162 -0
  1900. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/flash_attn_mmq_funcs.glsl +203 -0
  1901. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/flash_attn_split_k_reduce.comp +121 -0
  1902. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/floor.comp +22 -0
  1903. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/gated_delta_net.comp +190 -0
  1904. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/geglu.comp +13 -0
  1905. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/geglu_erf.comp +27 -0
  1906. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/geglu_quick.comp +11 -0
  1907. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/gelu.comp +25 -0
  1908. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/gelu_erf.comp +39 -0
  1909. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/gelu_quick.comp +23 -0
  1910. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/generic_binary_head.glsl +65 -0
  1911. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/generic_head.glsl +11 -0
  1912. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/generic_unary_head.glsl +83 -0
  1913. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/get_rows.comp +42 -0
  1914. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/get_rows_quant.comp +51 -0
  1915. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/glu_head.glsl +28 -0
  1916. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/glu_main.glsl +39 -0
  1917. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/group_norm.comp +66 -0
  1918. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/hardsigmoid.comp +22 -0
  1919. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/hardswish.comp +22 -0
  1920. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/im2col.comp +93 -0
  1921. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/im2col_3d.comp +124 -0
  1922. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/l2_norm.comp +44 -0
  1923. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/leaky_relu.comp +22 -0
  1924. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/log.comp +17 -0
  1925. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/mul.comp +27 -0
  1926. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_split_k_reduce.comp +48 -0
  1927. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec.comp +169 -0
  1928. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_base.glsl +230 -0
  1929. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_iface.glsl +35 -0
  1930. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_iq1_m.comp +132 -0
  1931. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_iq1_s.comp +95 -0
  1932. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_iq2_s.comp +90 -0
  1933. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_iq2_xs.comp +105 -0
  1934. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_iq2_xxs.comp +87 -0
  1935. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_iq3_s.comp +90 -0
  1936. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_iq3_xxs.comp +88 -0
  1937. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_nc.comp +124 -0
  1938. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_p021.comp +156 -0
  1939. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_q2_k.comp +128 -0
  1940. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_q3_k.comp +132 -0
  1941. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_q4_k.comp +134 -0
  1942. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_q5_k.comp +165 -0
  1943. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_q6_k.comp +130 -0
  1944. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vecq.comp +143 -0
  1945. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vecq_funcs.glsl +503 -0
  1946. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/mul_mm.comp +464 -0
  1947. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/mul_mm_cm2.comp +624 -0
  1948. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/mul_mm_funcs.glsl +600 -0
  1949. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/mul_mm_id_funcs.glsl +74 -0
  1950. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/mul_mmq.comp +311 -0
  1951. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/mul_mmq_funcs.glsl +454 -0
  1952. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/mul_mmq_shmem_types.glsl +93 -0
  1953. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/multi_add.comp +194 -0
  1954. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/neg.comp +20 -0
  1955. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/norm.comp +44 -0
  1956. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/opt_step_adamw.comp +42 -0
  1957. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/opt_step_sgd.comp +22 -0
  1958. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/pad.comp +64 -0
  1959. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/pool2d.comp +74 -0
  1960. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/quantize_q8_1.comp +127 -0
  1961. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/reglu.comp +9 -0
  1962. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/relu.comp +21 -0
  1963. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/repeat.comp +26 -0
  1964. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/repeat_back.comp +37 -0
  1965. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/rms_norm.comp +150 -0
  1966. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/rms_norm_back.comp +55 -0
  1967. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/rms_norm_partials.comp +65 -0
  1968. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/roll.comp +46 -0
  1969. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/rope_funcs.glsl +207 -0
  1970. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/rope_head.glsl +19 -0
  1971. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/rope_multi.comp +17 -0
  1972. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/rope_neox.comp +17 -0
  1973. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/rope_norm.comp +17 -0
  1974. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/rope_params.glsl +31 -0
  1975. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/rope_vision.comp +17 -0
  1976. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/round.comp +29 -0
  1977. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/scale.comp +24 -0
  1978. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/sgn.comp +21 -0
  1979. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/sigmoid.comp +20 -0
  1980. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/silu.comp +22 -0
  1981. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/silu_back.comp +26 -0
  1982. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/sin.comp +17 -0
  1983. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/soft_max.comp +195 -0
  1984. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/soft_max_back.comp +54 -0
  1985. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/soft_max_large1.comp +62 -0
  1986. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/soft_max_large2.comp +79 -0
  1987. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/soft_max_large3.comp +65 -0
  1988. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/soft_max_large_common.glsl +53 -0
  1989. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/softplus.comp +23 -0
  1990. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/solve_tri.comp +81 -0
  1991. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/sqrt.comp +17 -0
  1992. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/square.comp +17 -0
  1993. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/ssm_conv.comp +50 -0
  1994. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/ssm_scan.comp +124 -0
  1995. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/step.comp +22 -0
  1996. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/sub.comp +29 -0
  1997. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/sum_rows.comp +47 -0
  1998. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/sum_rows.glsl +25 -0
  1999. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/swiglu.comp +9 -0
  2000. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/swiglu_oai.comp +14 -0
  2001. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/tanh.comp +20 -0
  2002. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/timestep_embedding.comp +42 -0
  2003. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/topk_argsort.comp +118 -0
  2004. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/topk_moe.comp +213 -0
  2005. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/topk_nary_search.comp +246 -0
  2006. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/tri.comp +42 -0
  2007. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/trunc.comp +22 -0
  2008. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/types.glsl +1846 -0
  2009. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/upscale.comp +178 -0
  2010. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/utils.glsl +25 -0
  2011. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/vulkan-shaders-gen.cpp +1183 -0
  2012. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/wkv6.comp +87 -0
  2013. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/wkv7.comp +91 -0
  2014. data/vendor/ggml/src/ggml-vulkan/vulkan-shaders/xielu.comp +35 -0
  2015. data/vendor/ggml/src/ggml-webgpu/CMakeLists.txt +80 -0
  2016. data/vendor/ggml/src/ggml-webgpu/ggml-webgpu-shader-lib.hpp +3231 -0
  2017. data/vendor/ggml/src/ggml-webgpu/ggml-webgpu.cpp +4461 -0
  2018. data/vendor/ggml/src/ggml-webgpu/pre_wgsl.hpp +778 -0
  2019. data/vendor/ggml/src/ggml-webgpu/wgsl-shaders/add_id.wgsl +64 -0
  2020. data/vendor/ggml/src/ggml-webgpu/wgsl-shaders/argmax.wgsl +72 -0
  2021. data/vendor/ggml/src/ggml-webgpu/wgsl-shaders/argsort.wgsl +106 -0
  2022. data/vendor/ggml/src/ggml-webgpu/wgsl-shaders/argsort_merge.wgsl +134 -0
  2023. data/vendor/ggml/src/ggml-webgpu/wgsl-shaders/binary.wgsl +139 -0
  2024. data/vendor/ggml/src/ggml-webgpu/wgsl-shaders/common_decls.tmpl +905 -0
  2025. data/vendor/ggml/src/ggml-webgpu/wgsl-shaders/concat.wgsl +75 -0
  2026. data/vendor/ggml/src/ggml-webgpu/wgsl-shaders/conv2d.wgsl +165 -0
  2027. data/vendor/ggml/src/ggml-webgpu/wgsl-shaders/cpy.wgsl +81 -0
  2028. data/vendor/ggml/src/ggml-webgpu/wgsl-shaders/cumsum.wgsl +66 -0
  2029. data/vendor/ggml/src/ggml-webgpu/wgsl-shaders/embed_wgsl.py +89 -0
  2030. data/vendor/ggml/src/ggml-webgpu/wgsl-shaders/flash_attn.wgsl +706 -0
  2031. data/vendor/ggml/src/ggml-webgpu/wgsl-shaders/flash_attn_tile.wgsl +351 -0
  2032. data/vendor/ggml/src/ggml-webgpu/wgsl-shaders/flash_attn_vec_blk.wgsl +101 -0
  2033. data/vendor/ggml/src/ggml-webgpu/wgsl-shaders/flash_attn_vec_reduce.wgsl +84 -0
  2034. data/vendor/ggml/src/ggml-webgpu/wgsl-shaders/flash_attn_vec_split.wgsl +720 -0
  2035. data/vendor/ggml/src/ggml-webgpu/wgsl-shaders/gated_delta_net.wgsl +132 -0
  2036. data/vendor/ggml/src/ggml-webgpu/wgsl-shaders/get_rows.wgsl +773 -0
  2037. data/vendor/ggml/src/ggml-webgpu/wgsl-shaders/glu.wgsl +155 -0
  2038. data/vendor/ggml/src/ggml-webgpu/wgsl-shaders/im2col.wgsl +101 -0
  2039. data/vendor/ggml/src/ggml-webgpu/wgsl-shaders/memset.wgsl +40 -0
  2040. data/vendor/ggml/src/ggml-webgpu/wgsl-shaders/mul_mat.wgsl +747 -0
  2041. data/vendor/ggml/src/ggml-webgpu/wgsl-shaders/mul_mat_decls.tmpl +1210 -0
  2042. data/vendor/ggml/src/ggml-webgpu/wgsl-shaders/mul_mat_id.wgsl +195 -0
  2043. data/vendor/ggml/src/ggml-webgpu/wgsl-shaders/mul_mat_id_gather.wgsl +55 -0
  2044. data/vendor/ggml/src/ggml-webgpu/wgsl-shaders/mul_mat_id_vec.wgsl +154 -0
  2045. data/vendor/ggml/src/ggml-webgpu/wgsl-shaders/mul_mat_reg_tile.wgsl +149 -0
  2046. data/vendor/ggml/src/ggml-webgpu/wgsl-shaders/mul_mat_subgroup_matrix.wgsl +200 -0
  2047. data/vendor/ggml/src/ggml-webgpu/wgsl-shaders/mul_mat_vec.wgsl +133 -0
  2048. data/vendor/ggml/src/ggml-webgpu/wgsl-shaders/mul_mat_vec_acc.tmpl +1433 -0
  2049. data/vendor/ggml/src/ggml-webgpu/wgsl-shaders/pad.wgsl +86 -0
  2050. data/vendor/ggml/src/ggml-webgpu/wgsl-shaders/repeat.wgsl +67 -0
  2051. data/vendor/ggml/src/ggml-webgpu/wgsl-shaders/rms_norm_mul.wgsl +152 -0
  2052. data/vendor/ggml/src/ggml-webgpu/wgsl-shaders/rope.wgsl +224 -0
  2053. data/vendor/ggml/src/ggml-webgpu/wgsl-shaders/row_norm.wgsl +153 -0
  2054. data/vendor/ggml/src/ggml-webgpu/wgsl-shaders/scale.wgsl +63 -0
  2055. data/vendor/ggml/src/ggml-webgpu/wgsl-shaders/set.wgsl +109 -0
  2056. data/vendor/ggml/src/ggml-webgpu/wgsl-shaders/set_rows.wgsl +109 -0
  2057. data/vendor/ggml/src/ggml-webgpu/wgsl-shaders/soft_max.wgsl +245 -0
  2058. data/vendor/ggml/src/ggml-webgpu/wgsl-shaders/solve_tri.wgsl +121 -0
  2059. data/vendor/ggml/src/ggml-webgpu/wgsl-shaders/ssm_conv.wgsl +65 -0
  2060. data/vendor/ggml/src/ggml-webgpu/wgsl-shaders/ssm_scan.wgsl +193 -0
  2061. data/vendor/ggml/src/ggml-webgpu/wgsl-shaders/sum_rows.wgsl +55 -0
  2062. data/vendor/ggml/src/ggml-webgpu/wgsl-shaders/unary.wgsl +210 -0
  2063. data/vendor/ggml/src/ggml-webgpu/wgsl-shaders/upscale.wgsl +240 -0
  2064. data/vendor/ggml/src/ggml-zdnn/CMakeLists.txt +36 -0
  2065. data/vendor/ggml/src/ggml-zdnn/common.hpp +59 -0
  2066. data/vendor/ggml/src/ggml-zdnn/ggml-zdnn.cpp +637 -0
  2067. data/vendor/ggml/src/ggml-zdnn/mmf.cpp +80 -0
  2068. data/vendor/ggml/src/ggml-zdnn/mmf.hpp +12 -0
  2069. data/vendor/ggml/src/ggml-zdnn/utils.cpp +79 -0
  2070. data/vendor/ggml/src/ggml-zdnn/utils.hpp +19 -0
  2071. data/vendor/ggml/src/ggml-zendnn/CMakeLists.txt +91 -0
  2072. data/vendor/ggml/src/ggml-zendnn/ggml-zendnn.cpp +669 -0
  2073. data/vendor/ggml/src/ggml.c +7777 -0
  2074. data/vendor/ggml/src/ggml.cpp +26 -0
  2075. data/vendor/ggml/src/gguf.cpp +1556 -0
  2076. data/vendor/ggml/tests/CMakeLists.txt +356 -0
  2077. data/vendor/ggml/tests/test-arange.cpp +100 -0
  2078. data/vendor/ggml/tests/test-backend-ops.cpp +9786 -0
  2079. data/vendor/ggml/tests/test-cont.c +170 -0
  2080. data/vendor/ggml/tests/test-conv-transpose-1d.cpp +691 -0
  2081. data/vendor/ggml/tests/test-conv-transpose.c +248 -0
  2082. data/vendor/ggml/tests/test-conv1d-dw-c1.cpp +243 -0
  2083. data/vendor/ggml/tests/test-conv1d-dw-c2.cpp +243 -0
  2084. data/vendor/ggml/tests/test-conv1d.cpp +289 -0
  2085. data/vendor/ggml/tests/test-conv2d-dw.cpp +153 -0
  2086. data/vendor/ggml/tests/test-conv2d.cpp +391 -0
  2087. data/vendor/ggml/tests/test-customop.c +300 -0
  2088. data/vendor/ggml/tests/test-dup.c +111 -0
  2089. data/vendor/ggml/tests/test-interpolate.cpp +166 -0
  2090. data/vendor/ggml/tests/test-opt.cpp +1003 -0
  2091. data/vendor/ggml/tests/test-pad-reflect-1d.cpp +213 -0
  2092. data/vendor/ggml/tests/test-pool.c +274 -0
  2093. data/vendor/ggml/tests/test-quantize-fns.cpp +196 -0
  2094. data/vendor/ggml/tests/test-quantize-perf.cpp +356 -0
  2095. data/vendor/ggml/tests/test-rel-pos.c +87 -0
  2096. data/vendor/ggml/tests/test-roll.cpp +128 -0
  2097. data/vendor/ggml/tests/test-timestep_embedding.cpp +180 -0
  2098. data/vendor-patches/0001-cuda-buffer_from_ptr.patch +253 -0
  2099. data/vendor-patches/0002-cuda-buffer_from_ptr-reuse-iface.patch +117 -0
  2100. data/vendor-patches/0003-cuda-buffer_from_ptr-copy-mode.patch +128 -0
  2101. data/vendor-patches/0004-cuda-cpy-strided.patch +61 -0
  2102. data/vendor-patches/0005-concat-backward.patch +36 -0
  2103. data/vendor-patches/0006-getrows-back-large-vocab.patch +69 -0
  2104. data/vendor-patches/0007-gpt2-backward-kernels.patch +438 -0
  2105. data/vendor-patches/0008-mul-mat-backward-mixed-precision.patch +50 -0
  2106. data/vendor-patches/0009-sched-unsupported-node-diagnostic.patch +26 -0
  2107. metadata +2161 -0
@@ -0,0 +1,1846 @@
1
+ #if !defined(GGML_TYPES_COMP)
2
+ #define GGML_TYPES_COMP
3
+
4
+ #extension GL_EXT_shader_explicit_arithmetic_types_int64 : require
5
+ #extension GL_EXT_shader_explicit_arithmetic_types_int32 : require
6
+ #extension GL_EXT_shader_explicit_arithmetic_types_int16 : require
7
+ #extension GL_EXT_shader_explicit_arithmetic_types_int8 : require
8
+ #extension GL_EXT_shader_16bit_storage : require
9
+
10
+ #if defined(DATA_A_F32)
11
+ #define QUANT_K 1
12
+ #define QUANT_R 1
13
+
14
+ #if LOAD_VEC_A == 4
15
+ #define A_TYPE vec4
16
+ #elif LOAD_VEC_A == 8
17
+ #define A_TYPE mat2x4
18
+ #else
19
+ #define A_TYPE float
20
+ #endif
21
+ #endif
22
+
23
+ #if defined(DATA_A_F16)
24
+ #define QUANT_K 1
25
+ #define QUANT_R 1
26
+
27
+ #if LOAD_VEC_A == 4
28
+ #define A_TYPE f16vec4
29
+ #elif LOAD_VEC_A == 8
30
+ #define A_TYPE f16mat2x4
31
+ #else
32
+ #define A_TYPE float16_t
33
+ #endif
34
+ #endif
35
+
36
+ #if defined(DATA_A_BF16)
37
+ #define QUANT_K 1
38
+ #define QUANT_R 1
39
+
40
+ #if LOAD_VEC_A == 4
41
+ #define A_TYPE u16vec4
42
+ #elif LOAD_VEC_A == 8
43
+ #error unsupported
44
+ #else
45
+ #define A_TYPE uint16_t
46
+ #endif
47
+ #endif
48
+
49
+ #define QUANT_K_Q4_0 32
50
+ #define QUANT_R_Q4_0 2
51
+
52
+ struct block_q4_0
53
+ {
54
+ float16_t d;
55
+ uint8_t qs[16];
56
+ };
57
+ struct block_q4_0_packed16
58
+ {
59
+ float16_t d;
60
+ uint16_t qs[16/2];
61
+ };
62
+
63
+ #if defined(DATA_A_Q4_0)
64
+ #define QUANT_K QUANT_K_Q4_0
65
+ #define QUANT_R QUANT_R_Q4_0
66
+ #define QUANT_AUXF 1
67
+ #define A_TYPE block_q4_0
68
+ #define A_TYPE_PACKED16 block_q4_0_packed16
69
+ #define DATA_A_QUANT_LEGACY
70
+ #endif
71
+
72
+ #define QUANT_K_Q4_1 32
73
+ #define QUANT_R_Q4_1 2
74
+
75
+ struct block_q4_1
76
+ {
77
+ float16_t d;
78
+ float16_t m;
79
+ uint8_t qs[16];
80
+ };
81
+
82
+ struct block_q4_1_packed16
83
+ {
84
+ float16_t d;
85
+ float16_t m;
86
+ uint16_t qs[16/2];
87
+ };
88
+
89
+ struct block_q4_1_packed32
90
+ {
91
+ f16vec2 dm;
92
+ uint32_t qs[16/4];
93
+ };
94
+
95
+ #if defined(DATA_A_Q4_1)
96
+ #define QUANT_K QUANT_K_Q4_1
97
+ #define QUANT_R QUANT_R_Q4_1
98
+ #define QUANT_AUXF 2
99
+ #define A_TYPE block_q4_1
100
+ #define A_TYPE_PACKED16 block_q4_1_packed16
101
+ #define A_TYPE_PACKED32 block_q4_1_packed32
102
+ #define DATA_A_QUANT_LEGACY
103
+ #endif
104
+
105
+ #define QUANT_K_Q5_0 32
106
+ #define QUANT_R_Q5_0 2
107
+
108
+ struct block_q5_0
109
+ {
110
+ float16_t d;
111
+ uint16_t qh[2];
112
+ uint8_t qs[16];
113
+ };
114
+
115
+ struct block_q5_0_packed16
116
+ {
117
+ float16_t d;
118
+ uint16_t qh[2];
119
+ uint16_t qs[16/2];
120
+ };
121
+
122
+ #if defined(DATA_A_Q5_0)
123
+ #define QUANT_K QUANT_K_Q5_0
124
+ #define QUANT_R QUANT_R_Q5_0
125
+ #define QUANT_AUXF 1
126
+ #define A_TYPE block_q5_0
127
+ #define A_TYPE_PACKED16 block_q5_0_packed16
128
+ #define DATA_A_QUANT_LEGACY
129
+ #endif
130
+
131
+ #define QUANT_K_Q5_1 32
132
+ #define QUANT_R_Q5_1 2
133
+
134
+ struct block_q5_1
135
+ {
136
+ float16_t d;
137
+ float16_t m;
138
+ uint qh;
139
+ uint8_t qs[16];
140
+ };
141
+
142
+ struct block_q5_1_packed16
143
+ {
144
+ float16_t d;
145
+ float16_t m;
146
+ uint qh;
147
+ uint16_t qs[16/2];
148
+ };
149
+
150
+ struct block_q5_1_packed32
151
+ {
152
+ f16vec2 dm;
153
+ uint qh;
154
+ uint32_t qs[16/4];
155
+ };
156
+
157
+ #if defined(DATA_A_Q5_1)
158
+ #define QUANT_K QUANT_K_Q5_1
159
+ #define QUANT_R QUANT_R_Q5_1
160
+ #define QUANT_AUXF 2
161
+ #define A_TYPE block_q5_1
162
+ #define A_TYPE_PACKED16 block_q5_1_packed16
163
+ #define A_TYPE_PACKED32 block_q5_1_packed32
164
+ #define DATA_A_QUANT_LEGACY
165
+ #endif
166
+
167
+ #define QUANT_K_Q8_0 32
168
+ #define QUANT_R_Q8_0 1
169
+
170
+ struct block_q8_0
171
+ {
172
+ float16_t d;
173
+ int8_t qs[32];
174
+ };
175
+
176
+ struct block_q8_0_packed16
177
+ {
178
+ float16_t d;
179
+ int16_t qs[32/2];
180
+ };
181
+
182
+ #if defined(DATA_A_Q8_0)
183
+ #define QUANT_K QUANT_K_Q8_0
184
+ #define QUANT_R QUANT_R_Q8_0
185
+ #define QUANT_AUXF 1
186
+ #define A_TYPE block_q8_0
187
+ #define A_TYPE_PACKED16 block_q8_0_packed16
188
+ #define DATA_A_QUANT_LEGACY
189
+ #endif
190
+
191
+ #define QUANT_K_Q1_0 128
192
+ #define QUANT_R_Q1_0 1
193
+
194
+ struct block_q1_0
195
+ {
196
+ float16_t d;
197
+ uint8_t qs[QUANT_K_Q1_0 / 8];
198
+ };
199
+
200
+ #if defined(DATA_A_Q1_0)
201
+ #define QUANT_K QUANT_K_Q1_0
202
+ #define QUANT_R QUANT_R_Q1_0
203
+ #define QUANT_AUXF 1
204
+ #define A_TYPE block_q1_0
205
+ #endif
206
+
207
+ #define QUANT_K_Q8_1 32
208
+ #define QUANT_R_Q8_1 1
209
+
210
+ struct block_q8_1
211
+ {
212
+ f16vec2 ds;
213
+ int8_t qs[32];
214
+ };
215
+
216
+ struct block_q8_1_packed16
217
+ {
218
+ f16vec2 ds;
219
+ int16_t qs[16];
220
+ };
221
+
222
+ struct block_q8_1_packed32
223
+ {
224
+ f16vec2 ds;
225
+ int32_t qs[8];
226
+ };
227
+
228
+ // 4 blocks in one to allow 16-byte/128-bit alignment and loads
229
+ struct block_q8_1_x4
230
+ {
231
+ f16vec2 ds[4];
232
+ int32_t qs[32];
233
+ };
234
+
235
+ struct block_q8_1_x4_packed128
236
+ {
237
+ f16vec2 ds[4];
238
+ ivec4 qs[8];
239
+ };
240
+
241
+ // K-quants
242
+ #define QUANT_K_Q2_K 256
243
+
244
+ struct block_q2_K
245
+ {
246
+ uint8_t scales[QUANT_K_Q2_K/16];
247
+ uint8_t qs[QUANT_K_Q2_K/4];
248
+ f16vec2 dm;
249
+ };
250
+
251
+ struct block_q2_K_packed16
252
+ {
253
+ uint16_t scales[QUANT_K_Q2_K/16/2];
254
+ uint16_t qs[QUANT_K_Q2_K/4/2];
255
+ f16vec2 dm;
256
+ };
257
+
258
+ struct block_q2_K_packed32
259
+ {
260
+ uint32_t scales[QUANT_K_Q2_K/16/4];
261
+ uint32_t qs[QUANT_K_Q2_K/4/4];
262
+ f16vec2 dm;
263
+ };
264
+
265
+ #if defined(DATA_A_Q2_K)
266
+ #define QUANT_K QUANT_K_Q2_K
267
+ #define QUANT_R 1
268
+ #define A_TYPE block_q2_K
269
+ #define A_TYPE_PACKED16 block_q2_K_packed16
270
+ #define A_TYPE_PACKED32 block_q2_K_packed32
271
+ #define SCALES_PER_32 2
272
+ #define DATA_A_QUANT_K
273
+ #endif
274
+
275
+ #define QUANT_K_Q3_K 256
276
+
277
+ struct block_q3_K
278
+ {
279
+ uint8_t hmask[QUANT_K_Q3_K/8];
280
+ uint8_t qs[QUANT_K_Q3_K/4];
281
+ uint8_t scales[12];
282
+ float16_t d;
283
+ };
284
+
285
+ struct block_q3_K_packed16
286
+ {
287
+ uint16_t hmask[QUANT_K_Q3_K/8/2];
288
+ uint16_t qs[QUANT_K_Q3_K/4/2];
289
+ uint16_t scales[12/2];
290
+ float16_t d;
291
+ };
292
+
293
+ #if defined(DATA_A_Q3_K)
294
+ #define QUANT_K QUANT_K_Q3_K
295
+ #define QUANT_R 1
296
+ #define A_TYPE block_q3_K
297
+ #define A_TYPE_PACKED16 block_q3_K_packed16
298
+ #define DATA_A_QUANT_K
299
+ #endif
300
+
301
+ #define QUANT_K_Q4_K 256
302
+
303
+ struct block_q4_K
304
+ {
305
+ f16vec2 dm;
306
+ uint8_t scales[3*QUANT_K_Q4_K/64];
307
+ uint8_t qs[QUANT_K_Q4_K/2];
308
+ };
309
+
310
+ struct block_q4_K_packed16
311
+ {
312
+ f16vec2 dm;
313
+ uint16_t scales[3*QUANT_K_Q4_K/64/2];
314
+ uint16_t qs[QUANT_K_Q4_K/2/2];
315
+ };
316
+
317
+ struct block_q4_K_packed32
318
+ {
319
+ f16vec2 dm;
320
+ uint32_t scales[3*QUANT_K_Q4_K/64/4];
321
+ uint32_t qs[QUANT_K_Q4_K/2/4];
322
+ };
323
+
324
+ struct block_q4_K_packed128
325
+ {
326
+ uvec4 q4k[9];
327
+ };
328
+
329
+ #if defined(DATA_A_Q4_K)
330
+ #define QUANT_K QUANT_K_Q4_K
331
+ #define QUANT_R 1
332
+ #define A_TYPE block_q4_K
333
+ #define A_TYPE_PACKED16 block_q4_K_packed16
334
+ #define A_TYPE_PACKED32 block_q4_K_packed32
335
+ #define DATA_A_QUANT_K
336
+ #endif
337
+
338
+ #define QUANT_K_Q5_K 256
339
+
340
+ struct block_q5_K
341
+ {
342
+ f16vec2 dm;
343
+ uint8_t scales[12];
344
+ uint8_t qh[QUANT_K_Q5_K/8];
345
+ uint8_t qs[QUANT_K_Q5_K/2];
346
+ };
347
+
348
+ struct block_q5_K_packed16
349
+ {
350
+ f16vec2 dm;
351
+ uint16_t scales[12/2];
352
+ uint16_t qh[QUANT_K_Q5_K/8/2];
353
+ uint16_t qs[QUANT_K_Q5_K/2/2];
354
+ };
355
+
356
+ struct block_q5_K_packed32
357
+ {
358
+ f16vec2 dm;
359
+ uint32_t scales[12/4];
360
+ uint32_t qh[QUANT_K_Q5_K/8/4];
361
+ uint32_t qs[QUANT_K_Q5_K/2/4];
362
+ };
363
+
364
+ struct block_q5_K_packed128
365
+ {
366
+ uvec4 q5k[11];
367
+ };
368
+
369
+ #if defined(DATA_A_Q5_K)
370
+ #define QUANT_K QUANT_K_Q5_K
371
+ #define QUANT_R 1
372
+ #define A_TYPE block_q5_K
373
+ #define A_TYPE_PACKED16 block_q5_K_packed16
374
+ #define A_TYPE_PACKED32 block_q5_K_packed32
375
+ #define DATA_A_QUANT_K
376
+ #endif
377
+
378
+ #define QUANT_K_Q6_K 256
379
+
380
+ struct block_q6_K
381
+ {
382
+ uint8_t ql[QUANT_K_Q6_K/2];
383
+ uint8_t qh[QUANT_K_Q6_K/4];
384
+ int8_t scales[QUANT_K_Q6_K/16];
385
+ float16_t d;
386
+ };
387
+
388
+ struct block_q6_K_packed16
389
+ {
390
+ uint16_t ql[QUANT_K_Q6_K/2/2];
391
+ uint16_t qh[QUANT_K_Q6_K/4/2];
392
+ int16_t scales[QUANT_K_Q6_K/16/2];
393
+ float16_t d;
394
+ };
395
+
396
+ #if defined(DATA_A_Q6_K)
397
+ #define QUANT_K QUANT_K_Q6_K
398
+ #define QUANT_R 1
399
+ #define A_TYPE block_q6_K
400
+ #define A_TYPE_PACKED16 block_q6_K_packed16
401
+ #define DATA_A_QUANT_K
402
+ #endif
403
+
404
+ // IQuants
405
+
406
+ #define QUANT_K_IQ1_S 256
407
+ #define QUANT_R_IQ1_S 1
408
+
409
+ struct block_iq1_s {
410
+ float16_t d;
411
+ uint8_t qs[QUANT_K_IQ1_S/8];
412
+ uint16_t qh[QUANT_K_IQ1_S/32];
413
+ };
414
+
415
+ struct block_iq1_s_packed16 {
416
+ float16_t d;
417
+ uint16_t qs[QUANT_K_IQ1_S/8/2];
418
+ uint16_t qh[QUANT_K_IQ1_S/32];
419
+ };
420
+
421
+ #define QUANT_K_IQ1_M 256
422
+ #define QUANT_R_IQ1_M 1
423
+
424
+ struct block_iq1_m {
425
+ uint8_t qs[QUANT_K_IQ1_M/8];
426
+ uint8_t qh[QUANT_K_IQ1_M/16];
427
+ uint16_t scales[QUANT_K_IQ1_M/64];
428
+ };
429
+
430
+ struct block_iq1_m_packed16 {
431
+ uint16_t qs[QUANT_K_IQ1_M/8/2];
432
+ uint16_t qh[QUANT_K_IQ1_M/16/2];
433
+ uint16_t scales[QUANT_K_IQ1_M/64];
434
+ };
435
+
436
+ struct block_iq1_m_packed32 {
437
+ uint32_t qs[QUANT_K_IQ1_M/8/4];
438
+ uint32_t qh[QUANT_K_IQ1_M/16/4];
439
+ uint32_t scales[QUANT_K_IQ1_M/64/2];
440
+ };
441
+
442
+ struct block_iq1_m_packed64 {
443
+ uint64_t qs[QUANT_K_IQ1_M/8/8];
444
+ uint64_t qh[QUANT_K_IQ1_M/16/8];
445
+ uint64_t scales;
446
+ };
447
+
448
+ #if defined(DATA_A_IQ1_S)
449
+ #define QUANT_K QUANT_K_IQ1_S
450
+ #define QUANT_R QUANT_R_IQ1_S
451
+ #define A_TYPE block_iq1_s
452
+ #define A_TYPE_PACKED16 block_iq1_s_packed16
453
+ #endif
454
+
455
+ #if defined(DATA_A_IQ1_M)
456
+ #define QUANT_K QUANT_K_IQ1_M
457
+ #define QUANT_R QUANT_R_IQ1_M
458
+ #define A_TYPE block_iq1_m
459
+ #define A_TYPE_PACKED16 block_iq1_m_packed16
460
+ #define A_TYPE_PACKED32 block_iq1_m_packed32
461
+ #endif
462
+
463
+ #if defined(DATA_A_IQ1_S) || defined(DATA_A_IQ1_M)
464
+ #define IQ1S_DELTA 0.125f
465
+ #define IQ1M_DELTA 0.125f
466
+
467
+ // Packed IQ1S grid where every 2 vec8 are encoded on 32 bits (2 bits per coordinate).
468
+ const uint[1024] iq1s_grid_const = {
469
+ 0xfffdffff, 0xfff7fff0, 0xffccfff5, 0xffdfffc0, 0xffd7ffdd, 0xff30ffd5, 0xff03ff0c, 0xff10ff01,
470
+ 0xff7dff7f, 0xff75ff77, 0xff5fff40, 0xff57ff5d, 0xfcf3ff55, 0xfcccfcf0, 0xfcc1fcc3, 0xfcc5fcc4,
471
+ 0xfc3cfcd0, 0xfc34fc31, 0xfc00fc0d, 0xfc1cfc05, 0xfc11fc13, 0xfc70fc17, 0xfc43fc4c, 0xfc50fc41,
472
+ 0xfdfdfdff, 0xfdf5fdf7, 0xfddffdc0, 0xfdd7fddd, 0xfd30fdd5, 0xfd04fd0c, 0xfd14fd13, 0xfd7dfd7f,
473
+ 0xfd75fd77, 0xfd40fd4c, 0xfd5ffd44, 0xfd57fd5d, 0xf3ccfd55, 0xf3c1f3c3, 0xf33cf3d0, 0xf300f334,
474
+ 0xf313f305, 0xf34cf310, 0xf350f344, 0xf0f3f0fc, 0xf0f1f0f0, 0xf0c7f0c0, 0xf0d4f0c5, 0xf030f03f,
475
+ 0xf00ff035, 0xf003f00c, 0xf001f000, 0xf01ff004, 0xf010f01d, 0xf015f017, 0xf04cf07c, 0xf047f040,
476
+ 0xf05cf045, 0xf050f053, 0xf054f051, 0xf1c4f1c3, 0xf133f13c, 0xf10df10f, 0xf107f100, 0xf11cf11f,
477
+ 0xf114f111, 0xf14cf170, 0xf144f143, 0xf7fdf7ff, 0xf7f5f7f7, 0xf7dff7c0, 0xf7d7f7dd, 0xf730f7d5,
478
+ 0xf701f70c, 0xf77ff710, 0xf777f77d, 0xf740f775, 0xf75df75f, 0xf755f757, 0xf4ccf4f0, 0xf4c4f4c3,
479
+ 0xf4d0f4d3, 0xf40ff43c, 0xf400f40c, 0xf413f41c, 0xf44cf414, 0xf441f443, 0xf450f444, 0xf5fdf5ff,
480
+ 0xf5f5f5f7, 0xf5dff5c0, 0xf5d7f5dd, 0xf530f5d5, 0xf504f50c, 0xf510f51c, 0xf57df57f, 0xf577f570,
481
+ 0xf540f575, 0xf55df55f, 0xf555f557, 0xcfcccfcf, 0xcfc4cfc3, 0xcfd0cfd3, 0xcf33cf3c, 0xcf00cf0f,
482
+ 0xcf1ccf07, 0xcf10cf13, 0xcf4ccf14, 0xcf41cf43, 0xcf50cf5c, 0xccf3ccfc, 0xccf4ccf1, 0xcccdcccf,
483
+ 0xccc7ccc0, 0xccd3ccdc, 0xcc30ccd4, 0xcc0fcc35, 0xcc0dcc0c, 0xcc00cc03, 0xcc04cc01, 0xcc10cc1f,
484
+ 0xcc4dcc73, 0xcc5ccc40, 0xcdcccc53, 0xcdc1cdc3, 0xcd3fcdd0, 0xcd34cd31, 0xcd00cd0d, 0xcd05cd07,
485
+ 0xcd11cd13, 0xcd4ccd70, 0xcd41cd43, 0xc3fccd50, 0xc3f4c3f1, 0xc3c0c3c3, 0xc3c4c3c7, 0xc3d1c3dc,
486
+ 0xc330c33c, 0xc337c331, 0xc30cc335, 0xc300c303, 0xc304c301, 0xc310c31d, 0xc373c317, 0xc34fc374,
487
+ 0xc340c343, 0xc344c347, 0xc35cc345, 0xc350c353, 0xc0fdc354, 0xc0f5c0f0, 0xc0c3c0cc, 0xc0c1c0c0,
488
+ 0xc0dfc0c4, 0xc0d0c0dd, 0xc0d5c0d7, 0xc033c03c, 0xc031c030, 0xc00dc00c, 0xc000c003, 0xc004c001,
489
+ 0xc01cc005, 0xc010c013, 0xc014c011, 0xc07dc07f, 0xc070c073, 0xc075c077, 0xc04cc04f, 0xc040c043,
490
+ 0xc044c041, 0xc05fc045, 0xc050c05d, 0xc1f3c1fc, 0xc1f1c1f0, 0xc1c1c1c0, 0xc1c5c1c7, 0xc1d1c1dc,
491
+ 0xc13dc13f, 0xc130c133, 0xc135c137, 0xc100c10c, 0xc107c101, 0xc11cc104, 0xc110c113, 0xc114c117,
492
+ 0xc171c115, 0xc14dc175, 0xc153c140, 0xc7ccc154, 0xc7d0c7c1, 0xc733c73c, 0xc734c731, 0xc700c70f,
493
+ 0xc705c707, 0xc71cc71f, 0xc711c713, 0xc770c714, 0xc743c74c, 0xc4cfc750, 0xc4c0c4cd, 0xc4dcc4c5,
494
+ 0xc43dc4d0, 0xc430c433, 0xc40cc437, 0xc400c403, 0xc404c401, 0xc41fc405, 0xc415c410, 0xc44cc474,
495
+ 0xc440c44d, 0xc45cc447, 0xc454c451, 0xc5c1c5f4, 0xc5d1c5d3, 0xc531c533, 0xc50fc534, 0xc500c50d,
496
+ 0xc51cc507, 0xc514c511, 0xc54cc570, 0xc545c541, 0xdffddfff, 0xdff5dff7, 0xdfdfdfc0, 0xdfd0dfdd,
497
+ 0xdfd5dfd7, 0xdf0cdf30, 0xdf1cdf04, 0xdf7fdf10, 0xdf77df7d, 0xdf40df75, 0xdf5ddf5f, 0xdf57df50,
498
+ 0xdcf0df55, 0xdcc3dccc, 0xdcd0dcc4, 0xdc33dc3d, 0xdc00dc34, 0xdc05dc07, 0xdc13dc1c, 0xdc11dc10,
499
+ 0xdc4fdc70, 0xdc44dc41, 0xddfcdc50, 0xddf5ddf7, 0xddc0ddcc, 0xdddddddf, 0xddd5ddd7, 0xdd0cdd30,
500
+ 0xdd04dd01, 0xdd7cdd10, 0xdd75dd77, 0xdd40dd4c, 0xdd5ddd5f, 0xdd55dd57, 0xd3c3d3f0, 0xd3c4d3c1,
501
+ 0xd333d3d0, 0xd331d330, 0xd30dd334, 0xd307d300, 0xd311d305, 0xd34cd370, 0xd344d343, 0xd350d35c,
502
+ 0xd0c0d0f4, 0xd0d4d0dc, 0xd030d03f, 0xd00cd037, 0xd000d003, 0xd01dd004, 0xd017d010, 0xd04fd074,
503
+ 0xd040d043, 0xd045d047, 0xd053d05c, 0xd054d051, 0xd1cfd1f0, 0xd1c4d1cd, 0xd13cd1d0, 0xd100d134,
504
+ 0xd11cd11f, 0xd173d114, 0xd14fd171, 0xd7ffd145, 0xd7f7d7fd, 0xd7c0d7f5, 0xd7ddd7df, 0xd7d5d7d7,
505
+ 0xd70cd730, 0xd710d703, 0xd77dd77f, 0xd775d777, 0xd75dd75f, 0xd755d757, 0xd4ccd4f4, 0xd4c4d4c3,
506
+ 0xd431d4d0, 0xd40dd434, 0xd41cd400, 0xd411d413, 0xd470d414, 0xd441d44f, 0xd453d444, 0xd5ffd450,
507
+ 0xd5f7d5fd, 0xd5dfd5f5, 0xd5d7d5dd, 0xd530d5d5, 0xd501d50c, 0xd510d504, 0xd57dd57f, 0xd575d577,
508
+ 0xd55fd540, 0xd557d55d, 0x3ff0d555, 0x3fc13fcc, 0x3f343fd0, 0x3f003f0d, 0x3f053f07, 0x3f133f1c,
509
+ 0x3f433f11, 0x3f5c3f44, 0x3cff3f51, 0x3cf33cfc, 0x3cf43cf1, 0x3cc03ccd, 0x3cc73cc1, 0x3cdc3cc5,
510
+ 0x3cd43cd1, 0x3c373c30, 0x3c0c3c35, 0x3c003c03, 0x3c043c01, 0x3c103c05, 0x3c153c17, 0x3c733c7c,
511
+ 0x3c4f3c71, 0x3c403c4d, 0x3c5c3c5f, 0x3df03c5d, 0x3dc33dcc, 0x3dd03dc1, 0x3d0d3d3c, 0x3d053d00,
512
+ 0x3d143d13, 0x3d433d74, 0x33fc3d50, 0x33c433c0, 0x333033d4, 0x33353337, 0x3303330c, 0x33013300,
513
+ 0x331d331c, 0x33173310, 0x337c3315, 0x33743371, 0x334d334f, 0x335f3340, 0x3354335c, 0x30fd30fc,
514
+ 0x30f530f0, 0x30c330cc, 0x30c130c0, 0x30df30c4, 0x30d530d0, 0x3033303c, 0x30313030, 0x300f3034,
515
+ 0x3003300c, 0x30013000, 0x30043007, 0x3013301c, 0x30113010, 0x307d3014, 0x30703073, 0x304c3077,
516
+ 0x30403043, 0x30443041, 0x30503045, 0x30553057, 0x31f031fc, 0x31c331f4, 0x31c731c0, 0x31dc31c5,
517
+ 0x31d431d3, 0x313d313f, 0x31373130, 0x310c310f, 0x3100310d, 0x31043101, 0x3110311d, 0x317c3117,
518
+ 0x31753170, 0x31403143, 0x3153315c, 0x37f03151, 0x37c037cc, 0x37d037c5, 0x3734373d, 0x3700370f,
519
+ 0x371c3707, 0x37113713, 0x37703714, 0x3743374c, 0x37443741, 0x34fc3750, 0x34f134f0, 0x34cf34f5,
520
+ 0x34c034c3, 0x34dc34c7, 0x34d134d3, 0x3430343f, 0x340c3435, 0x3403340d, 0x34013400, 0x341f3404,
521
+ 0x3410341d, 0x34153411, 0x34743471, 0x3440344d, 0x34473441, 0x3453345c, 0x34543451, 0x353335c1,
522
+ 0x35343531, 0x35073500, 0x35133505, 0x35433514, 0x0ffc3550, 0x0ff00ff3, 0x0ff40ff1, 0x0fc00fcd,
523
+ 0x0fdc0fc5, 0x0fd40fd3, 0x0f300f3f, 0x0f0c0f37, 0x0f000f03, 0x0f040f01, 0x0f170f10, 0x0f740f71,
524
+ 0x0f470f40, 0x0f5c0f5f, 0x0f540f51, 0x0cf70cf0, 0x0cf50cf4, 0x0cc30ccc, 0x0cc10cc0, 0x0cc40cc7,
525
+ 0x0cd00cdf, 0x0cd70cd1, 0x0c3c0cd5, 0x0c300c33, 0x0c340c31, 0x0c0c0c0f, 0x0c030c0d, 0x0c010c00,
526
+ 0x0c040c07, 0x0c1c0c05, 0x0c100c13, 0x0c140c11, 0x0c700c7d, 0x0c430c4c, 0x0c410c40, 0x0c5f0c44,
527
+ 0x0c550c50, 0x0df10dfc, 0x0dc00dcd, 0x0ddc0dc5, 0x0d3d0dd3, 0x0d350d30, 0x0d030d0c, 0x0d010d00,
528
+ 0x0d1d0d04, 0x0d700d10, 0x0d4d0d4f, 0x0d440d40, 0x0d530d45, 0x03f003f3, 0x03c303cc, 0x03c103c0,
529
+ 0x03c403c7, 0x03d003dc, 0x03d503d7, 0x0333033c, 0x03310330, 0x03350334, 0x030c030f, 0x03000303,
530
+ 0x03070301, 0x03050304, 0x031d031c, 0x03100313, 0x03140311, 0x0377037f, 0x034c0375, 0x03400343,
531
+ 0x03440341, 0x0353035c, 0x03550350, 0x00fd00fc, 0x00f000f3, 0x00f400f1, 0x00cc00cf, 0x00c300cd,
532
+ 0x00c100c0, 0x00c500c4, 0x00d300dc, 0x00d100d0, 0x003f00d4, 0x003d003c, 0x00300033, 0x00370031,
533
+ 0x000f0034, 0x000d000c, 0x00000003, 0x00070001, 0x00050004, 0x001c001f, 0x00100013, 0x00170011,
534
+ 0x00150014, 0x0073007c, 0x00740070, 0x004f0075, 0x0043004c, 0x00410040, 0x00440047, 0x0053005c,
535
+ 0x00510050, 0x01ff0054, 0x01fd01fc, 0x01f101f3, 0x01f401f7, 0x01c301cc, 0x01c701c0, 0x01df01c4,
536
+ 0x01dd01dc, 0x01d001d3, 0x01d701d1, 0x013c01d4, 0x01310130, 0x01340137, 0x010f0135, 0x010d010c,
537
+ 0x01000103, 0x01070101, 0x01050104, 0x0113011c, 0x01140110, 0x0170017d, 0x01770171, 0x01750174,
538
+ 0x0140014c, 0x015d0145, 0x01510150, 0x01540157, 0x07f007f3, 0x07f407f1, 0x07c007cf, 0x07dc07c7,
539
+ 0x073007d5, 0x07350737, 0x0703070c, 0x07010700, 0x07040707, 0x071d071f, 0x07100713, 0x0774077d,
540
+ 0x074d074f, 0x07470740, 0x0754075c, 0x04fd04fc, 0x04f504f0, 0x04c304cc, 0x04c104c0, 0x04d004c4,
541
+ 0x0433043c, 0x04310430, 0x040f0434, 0x040d040c, 0x04000403, 0x04070401, 0x04050404, 0x0413041c,
542
+ 0x04110410, 0x047c0414, 0x04740470, 0x0443044c, 0x04410440, 0x04440447, 0x05f30450, 0x05c005f7,
543
+ 0x05df05c5, 0x05d105d0, 0x053005d4, 0x05340537, 0x0500050c, 0x05070501, 0x051d0504, 0x05170510,
544
+ 0x057c0515, 0x054d0575, 0x05410540, 0x05450547, 0x1ff0055c, 0x1fc11fc3, 0x1fd01fc4, 0x1f0f1f33,
545
+ 0x1f011f00, 0x1f051f07, 0x1f131f1c, 0x1f141f11, 0x1f411f7c, 0x1cfc1f50, 0x1cf11cf3, 0x1ccd1cf4,
546
+ 0x1cdc1cc0, 0x1cd11cdd, 0x1c301cd4, 0x1c0c1c34, 0x1c011c00, 0x1c101c04, 0x1c151c11, 0x1c751c73,
547
+ 0x1c401c4d, 0x1c511c5c, 0x1dcc1c54, 0x1dc41dc1, 0x1d3c1d3f, 0x1d001d31, 0x1d071d01, 0x1d701d1f,
548
+ 0x1d411d4c, 0x13cc1d50, 0x13c013cd, 0x13c513c1, 0x13d113dc, 0x133f13d4, 0x1330133d, 0x13351337,
549
+ 0x1303130c, 0x13011300, 0x13051304, 0x131d131f, 0x13731310, 0x13741370, 0x134d134f, 0x13401343,
550
+ 0x13471341, 0x135c1345, 0x13541353, 0x10f710f0, 0x10cc10f5, 0x10c110c0, 0x103310c4, 0x10311030,
551
+ 0x100f1034, 0x1003100c, 0x10011000, 0x101c1004, 0x10101013, 0x10141011, 0x10741071, 0x104c1075,
552
+ 0x10411040, 0x10451044, 0x1050105d, 0x10571051, 0x11f411fd, 0x11df11c0, 0x11d711d1, 0x113f11d4,
553
+ 0x11371130, 0x110c1135, 0x11001103, 0x11071101, 0x111f1105, 0x11171110, 0x117d117f, 0x11751170,
554
+ 0x11411143, 0x11441147, 0x1153115f, 0x11551151, 0x17c417c1, 0x173c17d0, 0x1700170d, 0x171c1705,
555
+ 0x17701714, 0x1747174c, 0x14fc1751, 0x14cf14f3, 0x14dc14c0, 0x14d114d3, 0x143f14d4, 0x1430143c,
556
+ 0x14371431, 0x1403140c, 0x14011400, 0x141f1404, 0x14151410, 0x1473147d, 0x14401475, 0x1453145c,
557
+ 0x14541450, 0x15c115cc, 0x153c15c7, 0x15341533, 0x1500150f, 0x15051507, 0x15101513, 0x15711514,
558
+ 0x15471543, 0x15511545, 0x7ffd7fff, 0x7ff57ff7, 0x7fdd7fdf, 0x7fd57fd7, 0x7f0f7f30, 0x7f037f0c,
559
+ 0x7f047f01, 0x7f7f7f10, 0x7f777f7d, 0x7f407f75, 0x7f5d7f5f, 0x7f557f57, 0x7ccc7cf0, 0x7cc17cc3,
560
+ 0x7cd07cc4, 0x7c337c3c, 0x7c0f7c34, 0x7c007c0d, 0x7c077c01, 0x7c137c04, 0x7c147c11, 0x7c747c70,
561
+ 0x7c417c43, 0x7c507c44, 0x7dfd7dff, 0x7df57df7, 0x7ddf7dc0, 0x7dd77ddd, 0x7d0c7dd5, 0x7d047d03,
562
+ 0x7d7f7d10, 0x7d777d7d, 0x7d407d75, 0x7d5d7d5f, 0x7d557d57, 0x73c473c3, 0x7333733c, 0x7300730c,
563
+ 0x731c7305, 0x73147313, 0x73447343, 0x70f470fc, 0x70c070cd, 0x70d170c5, 0x703f70d4, 0x7030703c,
564
+ 0x700c7037, 0x70007003, 0x70047001, 0x70107005, 0x70177011, 0x707c7015, 0x70717073, 0x704f7074,
565
+ 0x7040704d, 0x70517047, 0x71c171cc, 0x71d071c4, 0x7133713c, 0x71357134, 0x7100710f, 0x71057104,
566
+ 0x7111711c, 0x71707115, 0x7145714c, 0x77ff7153, 0x77f777fd, 0x77c077f5, 0x77dd77df, 0x77d577d7,
567
+ 0x7730773c, 0x7703770c, 0x77107704, 0x777f7714, 0x7777777d, 0x77407775, 0x775d775f, 0x77557757,
568
+ 0x74f174f0, 0x74c374cc, 0x74d074c1, 0x7433743c, 0x74347431, 0x740d740f, 0x74057400, 0x7413741c,
569
+ 0x74417470, 0x74507444, 0x75fd75ff, 0x75f575f7, 0x75df75c0, 0x75d775dd, 0x753075d5, 0x7503750c,
570
+ 0x757f7501, 0x7577757d, 0x75407575, 0x755d755f, 0x75557557, 0x4fcc4ff0, 0x4fc74fc1, 0x4fd04fc4,
571
+ 0x4f314f3c, 0x4f004f34, 0x4f054f07, 0x4f154f14, 0x4f4c4f70, 0x4f414f43, 0x4f504f44, 0x4cf34cfc,
572
+ 0x4cf44cf1, 0x4cc04ccf, 0x4cc54cc7, 0x4cd34cdc, 0x4cd44cd1, 0x4c304c3f, 0x4c0c4c0f, 0x4c004c03,
573
+ 0x4c044c01, 0x4c104c1d, 0x4c714c73, 0x4c404c4d, 0x4c5c4c47, 0x4c514c53, 0x4df04c54, 0x4dc34dcc,
574
+ 0x4dd04dc4, 0x4d314d33, 0x4d0f4d34, 0x4d004d0d, 0x4d114d07, 0x4d704d14, 0x4d414d43, 0x43fc4d54,
575
+ 0x43f143f3, 0x43c043cf, 0x43d143c7, 0x4335433f, 0x4303430c, 0x43014300, 0x43044307, 0x431c431f,
576
+ 0x4310431d, 0x43714373, 0x4343434d, 0x43474340, 0x4354435c, 0x40f040ff, 0x40f540f7, 0x40cc40cf,
577
+ 0x40c040c3, 0x40c440c1, 0x40d040dc, 0x40d540d4, 0x4033403c, 0x40314030, 0x400f4034, 0x400d400c,
578
+ 0x40004003, 0x40074001, 0x40054004, 0x4013401c, 0x40114010, 0x407c4014, 0x40774070, 0x404d404c,
579
+ 0x40404043, 0x40444041, 0x405f4045, 0x4050405d, 0x40554057, 0x41f341fc, 0x41c041cf, 0x41df41c4,
580
+ 0x41d441d1, 0x41374130, 0x410c4134, 0x4100410d, 0x41044101, 0x41174110, 0x4173417d, 0x41754174,
581
+ 0x4143414d, 0x41534140, 0x41544151, 0x47c147f0, 0x47d047c4, 0x4731473c, 0x470d470f, 0x47014700,
582
+ 0x47134705, 0x47704710, 0x4741474c, 0x47504744, 0x44f144f3, 0x44cf44f4, 0x44c044cd, 0x44c544c7,
583
+ 0x44dc44df, 0x44d144d3, 0x443d443f, 0x44374430, 0x440c4435, 0x44004403, 0x44044401, 0x4410441d,
584
+ 0x44154411, 0x4473447c, 0x444d444f, 0x44454440, 0x4451445c, 0x45c045f0, 0x453345d0, 0x45344531,
585
+ 0x4500450f, 0x451c4507, 0x454c4570, 0x45404543, 0x5fff4541, 0x5ff75ffd, 0x5fc05ff5, 0x5fdd5fdf,
586
+ 0x5fd55fd7, 0x5f0c5f30, 0x5f015f03, 0x5f7f5f04, 0x5f775f7d, 0x5f405f75, 0x5f5d5f5f, 0x5f555f57,
587
+ 0x5cf45cf0, 0x5cc35ccc, 0x5cc45cc1, 0x5c315cc5, 0x5c0c5c34, 0x5c075c00, 0x5c1c5c05, 0x5c705c13,
588
+ 0x5c4d5c4f, 0x5c445c41, 0x5df75dfd, 0x5dcf5df5, 0x5ddd5dc4, 0x5dd55dd7, 0x5d0c5d30, 0x5d045d01,
589
+ 0x5d7f5d10, 0x5d775d7d, 0x5d405d75, 0x5d5d5d5f, 0x5d555d57, 0x53d053c4, 0x5333533c, 0x5303530f,
590
+ 0x53075300, 0x531c5305, 0x53115310, 0x53145317, 0x50f15370, 0x50cf50f4, 0x50c050cd, 0x50d150c7,
591
+ 0x503d50d4, 0x500c5030, 0x50005003, 0x50045001, 0x50155010, 0x5073507c, 0x50715070, 0x504d5074,
592
+ 0x50475040, 0x51cc51f0, 0x51c551c1, 0x51d051dc, 0x51315133, 0x510d5135, 0x51015100, 0x511f5107,
593
+ 0x5171511d, 0x5140514f, 0x51445141, 0x5153515c, 0x57ff5151, 0x57f757fd, 0x57df57f5, 0x57d757dd,
594
+ 0x570c57d5, 0x57015703, 0x577f5704, 0x5777577d, 0x57405775, 0x575d575f, 0x57555757, 0x54c354f0,
595
+ 0x54dc54c4, 0x543c54d0, 0x5400540f, 0x541c5405, 0x54145411, 0x5441544f, 0x55fd55ff, 0x55f555f7,
596
+ 0x55dd55df, 0x55d555d7, 0x5503550c, 0x557f5501, 0x5577557d, 0x55405575, 0x555d555f, 0x55555557
597
+ };
598
+
599
+ // Same content as iq1s_grid_const except each 2-bit value is expanded to 4-bit
600
+ // and has 1 added to it (allows packed values to be extracted with & 0x0F0F0F0F
601
+ // and 0xF0F0F0F0).
602
+ const uint32_t[2048] iq1s_grid_gpu_const = {
603
+ 0x00000000, 0x00000002, 0x00000101, 0x00000200, 0x00000202, 0x00010001, 0x00010101, 0x00020000,
604
+ 0x00020002, 0x00020200, 0x00020202, 0x01000101, 0x01010001, 0x01010100, 0x01010102, 0x01020101,
605
+ 0x02000000, 0x02000002, 0x02000200, 0x02000202, 0x02010101, 0x02020000, 0x02020002, 0x02020200,
606
+ 0x02020202, 0x00000110, 0x00000111, 0x00010011, 0x00010110, 0x00010112, 0x00010211, 0x00010212,
607
+ 0x00020111, 0x01000011, 0x01000112, 0x01000211, 0x01010012, 0x01010111, 0x01010212, 0x01020011,
608
+ 0x01020110, 0x01020112, 0x01020210, 0x02000111, 0x02010011, 0x02010110, 0x02010112, 0x02020111,
609
+ 0x00000020, 0x00000022, 0x00000220, 0x00000222, 0x00010121, 0x00020020, 0x00020022, 0x00020220,
610
+ 0x00020222, 0x01000121, 0x01010021, 0x01010221, 0x01020120, 0x01020221, 0x02000020, 0x02000022,
611
+ 0x02000220, 0x02000222, 0x02010021, 0x02010121, 0x02010221, 0x02020020, 0x02020022, 0x02020220,
612
+ 0x02020222, 0x00011001, 0x00011100, 0x00011102, 0x00021101, 0x01001001, 0x01001201, 0x01011101,
613
+ 0x01011202, 0x01021100, 0x01021101, 0x02011001, 0x02011201, 0x02021101, 0x00001011, 0x00001110,
614
+ 0x00001111, 0x00001112, 0x00011111, 0x00011210, 0x00011212, 0x00021211, 0x01001010, 0x01001111,
615
+ 0x01001212, 0x01011010, 0x01011011, 0x01011110, 0x01011111, 0x01011112, 0x01011211, 0x01021010,
616
+ 0x01021012, 0x01021111, 0x01021210, 0x01021212, 0x02001011, 0x02011011, 0x02011111, 0x02011210,
617
+ 0x02011212, 0x02021011, 0x02021110, 0x02021111, 0x02021112, 0x02021211, 0x00011120, 0x00011221,
618
+ 0x01001021, 0x01001120, 0x01011020, 0x01011022, 0x01011121, 0x01011220, 0x01021020, 0x01021021,
619
+ 0x01021122, 0x01021221, 0x02001121, 0x02011021, 0x02011120, 0x02011221, 0x00002000, 0x00002002,
620
+ 0x00002200, 0x00002202, 0x00012101, 0x00022000, 0x00022002, 0x00022200, 0x00022202, 0x01002101,
621
+ 0x01012001, 0x01012102, 0x01022101, 0x02002000, 0x02002002, 0x02002200, 0x02002202, 0x02012101,
622
+ 0x02022000, 0x02022002, 0x02022200, 0x02022202, 0x00002111, 0x00012011, 0x00012110, 0x00012211,
623
+ 0x00022110, 0x00022111, 0x01002011, 0x01012010, 0x01012011, 0x01012111, 0x01022011, 0x01022110,
624
+ 0x01022211, 0x02012011, 0x02012110, 0x02012112, 0x02012211, 0x02022111, 0x00002020, 0x00002022,
625
+ 0x00002220, 0x00002222, 0x00012121, 0x00022020, 0x00022022, 0x00022220, 0x00022222, 0x01002121,
626
+ 0x01012021, 0x01012221, 0x01022021, 0x01022121, 0x02002020, 0x02002022, 0x02002121, 0x02002220,
627
+ 0x02002222, 0x02012121, 0x02022020, 0x02022022, 0x02022220, 0x02022222, 0x00110000, 0x00110001,
628
+ 0x00110100, 0x00110201, 0x00120100, 0x00120101, 0x01100001, 0x01100100, 0x01110000, 0x01110101,
629
+ 0x01110200, 0x01120001, 0x01120100, 0x01120101, 0x01120201, 0x02110001, 0x02110100, 0x02110102,
630
+ 0x02120001, 0x02120101, 0x00100011, 0x00100110, 0x00100112, 0x00100211, 0x00110010, 0x00110012,
631
+ 0x00110111, 0x00110210, 0x00120011, 0x00120110, 0x00120211, 0x01100111, 0x01100212, 0x01110010,
632
+ 0x01110011, 0x01110012, 0x01110110, 0x01110111, 0x01110112, 0x01110211, 0x01120010, 0x01120111,
633
+ 0x02100110, 0x02110012, 0x02110111, 0x02120011, 0x02120110, 0x00110021, 0x00110120, 0x00110122,
634
+ 0x00120121, 0x01100020, 0x01100122, 0x01100221, 0x01110022, 0x01110121, 0x01110220, 0x01110222,
635
+ 0x01120120, 0x01120122, 0x02100121, 0x02110021, 0x02110120, 0x02110122, 0x02120121, 0x00101001,
636
+ 0x00101102, 0x00101201, 0x00111100, 0x00111101, 0x00111200, 0x00111201, 0x00121001, 0x00121102,
637
+ 0x01101001, 0x01101101, 0x01101102, 0x01101200, 0x01101202, 0x01111001, 0x01111100, 0x01111101,
638
+ 0x01111102, 0x01111201, 0x01121002, 0x01121101, 0x01121200, 0x02101100, 0x02101201, 0x02111000,
639
+ 0x02111100, 0x02111101, 0x02111200, 0x02111201, 0x02111202, 0x02121001, 0x02121100, 0x02121101,
640
+ 0x02121201, 0x00101012, 0x00101111, 0x00101212, 0x00111011, 0x00111110, 0x00111111, 0x00111112,
641
+ 0x00111211, 0x00121010, 0x00121012, 0x00121111, 0x00121210, 0x00121212, 0x01101011, 0x01101110,
642
+ 0x01101111, 0x01101112, 0x01111011, 0x01111012, 0x01111110, 0x01111111, 0x01111112, 0x01111211,
643
+ 0x01111212, 0x01121011, 0x01121110, 0x01121111, 0x01121112, 0x01121211, 0x02101010, 0x02101012,
644
+ 0x02101110, 0x02101111, 0x02101210, 0x02101212, 0x02111010, 0x02111011, 0x02111110, 0x02111111,
645
+ 0x02111112, 0x02111211, 0x02111212, 0x02121010, 0x02121012, 0x02121111, 0x00101021, 0x00101120,
646
+ 0x00101121, 0x00101122, 0x00111121, 0x00111122, 0x00111220, 0x00111222, 0x00121021, 0x00121122,
647
+ 0x01101020, 0x01101022, 0x01101120, 0x01101121, 0x01101220, 0x01101222, 0x01111021, 0x01111121,
648
+ 0x01111122, 0x01111220, 0x01111221, 0x01121021, 0x01121120, 0x01121121, 0x01121220, 0x01121221,
649
+ 0x01121222, 0x02101122, 0x02101222, 0x02111022, 0x02111121, 0x02121120, 0x02121221, 0x00112001,
650
+ 0x00112102, 0x00122101, 0x01102001, 0x01102100, 0x01102102, 0x01102201, 0x01112000, 0x01112101,
651
+ 0x01112200, 0x01112202, 0x01122000, 0x01122001, 0x01122100, 0x01122102, 0x01122201, 0x02102101,
652
+ 0x02112001, 0x02112100, 0x02122101, 0x00112010, 0x00112012, 0x00112111, 0x00112212, 0x00122011,
653
+ 0x00122111, 0x01102012, 0x01102110, 0x01102111, 0x01102210, 0x01112011, 0x01112110, 0x01112111,
654
+ 0x01112112, 0x01112211, 0x01112212, 0x01122010, 0x01122111, 0x01122212, 0x02102211, 0x02112011,
655
+ 0x02112012, 0x02112111, 0x02112210, 0x02122011, 0x02122112, 0x02122211, 0x00102221, 0x00112122,
656
+ 0x00122120, 0x00122122, 0x01102120, 0x01102122, 0x01102221, 0x01112020, 0x01112022, 0x01112121,
657
+ 0x01112220, 0x01122021, 0x01122122, 0x01122221, 0x02102121, 0x02112021, 0x02112122, 0x02112222,
658
+ 0x00200000, 0x00200002, 0x00200200, 0x00200202, 0x00210101, 0x00220000, 0x00220002, 0x00220101,
659
+ 0x00220200, 0x00220202, 0x01200101, 0x01210001, 0x01210201, 0x01220001, 0x01220101, 0x02200000,
660
+ 0x02200002, 0x02200200, 0x02200202, 0x02210101, 0x02220000, 0x02220002, 0x02220101, 0x02220200,
661
+ 0x02220202, 0x00200111, 0x00210011, 0x00210110, 0x00210211, 0x00220111, 0x01200012, 0x01200110,
662
+ 0x01200211, 0x01210111, 0x01210210, 0x01210212, 0x01220011, 0x01220110, 0x01220111, 0x01220112,
663
+ 0x02200111, 0x02210010, 0x02210112, 0x02210211, 0x02220111, 0x00200021, 0x00200220, 0x00200222,
664
+ 0x00210021, 0x00210121, 0x00220020, 0x00220022, 0x00220220, 0x00220222, 0x01200121, 0x01210021,
665
+ 0x01210122, 0x01210221, 0x01220121, 0x02200021, 0x02200220, 0x02200222, 0x02210021, 0x02210121,
666
+ 0x02220020, 0x02220022, 0x02220220, 0x02220222, 0x00201101, 0x00211100, 0x00211102, 0x00211201,
667
+ 0x00221101, 0x01201100, 0x01201101, 0x01201102, 0x01201201, 0x01211002, 0x01211101, 0x01211200,
668
+ 0x01211202, 0x01221102, 0x02201101, 0x02211001, 0x02211100, 0x02211201, 0x02221001, 0x02221101,
669
+ 0x00201211, 0x00211111, 0x00221011, 0x00221211, 0x01201010, 0x01201111, 0x01201210, 0x01211011,
670
+ 0x01211110, 0x01211111, 0x01211211, 0x01221012, 0x01221111, 0x01221210, 0x02201211, 0x02211010,
671
+ 0x02211110, 0x02211111, 0x02211210, 0x02211212, 0x02221011, 0x02221110, 0x02221112, 0x02221211,
672
+ 0x00201121, 0x00211020, 0x00211022, 0x00211221, 0x00221121, 0x01201021, 0x01201221, 0x01211121,
673
+ 0x01221020, 0x01221021, 0x01221221, 0x02201120, 0x02201122, 0x02211020, 0x02211222, 0x00202000,
674
+ 0x00202002, 0x00202200, 0x00202202, 0x00212101, 0x00222000, 0x00222002, 0x00222200, 0x00222202,
675
+ 0x01202101, 0x01212001, 0x01212100, 0x01222101, 0x02202000, 0x02202002, 0x02202200, 0x02202202,
676
+ 0x02222000, 0x02222002, 0x02222200, 0x02222202, 0x00202211, 0x00212011, 0x00212110, 0x00212211,
677
+ 0x00222111, 0x01202112, 0x01202211, 0x01212012, 0x01212111, 0x01222011, 0x01222110, 0x01222112,
678
+ 0x01222211, 0x02202111, 0x02212010, 0x02212112, 0x02212211, 0x02222110, 0x02222111, 0x00202020,
679
+ 0x00202022, 0x00202220, 0x00202222, 0x00222020, 0x00222022, 0x00222220, 0x00222222, 0x01202121,
680
+ 0x01212021, 0x01212122, 0x01212221, 0x01222121, 0x02202020, 0x02202022, 0x02202220, 0x02202222,
681
+ 0x02212121, 0x02222020, 0x02222022, 0x02222220, 0x02222222, 0x10000101, 0x10010001, 0x10010102,
682
+ 0x10020101, 0x11000201, 0x11010002, 0x11010101, 0x11010200, 0x11010202, 0x11020001, 0x11020100,
683
+ 0x11020102, 0x12010100, 0x12010201, 0x12020001, 0x12020102, 0x10000010, 0x10000011, 0x10000110,
684
+ 0x10000112, 0x10000211, 0x10010012, 0x10010111, 0x10010112, 0x10010210, 0x10010212, 0x10020011,
685
+ 0x10020112, 0x10020211, 0x11000111, 0x11000210, 0x11000212, 0x11010011, 0x11010110, 0x11010111,
686
+ 0x11010112, 0x11010211, 0x11010212, 0x11020111, 0x11020210, 0x11020212, 0x12000011, 0x12000110,
687
+ 0x12000112, 0x12010010, 0x12010012, 0x12010111, 0x12020010, 0x12020011, 0x12020012, 0x10000121,
688
+ 0x10010021, 0x10010120, 0x10010122, 0x10020121, 0x11000021, 0x11010022, 0x11010121, 0x11010222,
689
+ 0x11020120, 0x11020221, 0x12000221, 0x12010120, 0x12020121, 0x10001001, 0x10011101, 0x10011201,
690
+ 0x10021201, 0x11001101, 0x11001200, 0x11001202, 0x11011001, 0x11011100, 0x11011101, 0x11011102,
691
+ 0x11021001, 0x11021002, 0x11021101, 0x11021200, 0x11021202, 0x12001001, 0x12001102, 0x12001201,
692
+ 0x12011000, 0x12011002, 0x12011101, 0x12021000, 0x12021001, 0x12021201, 0x10001011, 0x10001012,
693
+ 0x10001111, 0x10001212, 0x10011011, 0x10011110, 0x10011111, 0x10011112, 0x10011211, 0x10021010,
694
+ 0x10021111, 0x10021212, 0x11001011, 0x11001110, 0x11001111, 0x11001112, 0x11001211, 0x11011010,
695
+ 0x11011011, 0x11011110, 0x11011111, 0x11011112, 0x11011210, 0x11011211, 0x11021011, 0x11021110,
696
+ 0x11021111, 0x11021112, 0x11021211, 0x12001012, 0x12001110, 0x12001111, 0x12001210, 0x12011011,
697
+ 0x12011110, 0x12011111, 0x12011112, 0x12011211, 0x12011212, 0x12021111, 0x12021210, 0x12021212,
698
+ 0x10001021, 0x10001121, 0x10001221, 0x10011120, 0x10011121, 0x10011220, 0x10011222, 0x10021021,
699
+ 0x10021120, 0x10021221, 0x11001020, 0x11001022, 0x11001121, 0x11001220, 0x11011020, 0x11011021,
700
+ 0x11011022, 0x11011121, 0x11011122, 0x11011221, 0x11021022, 0x11021121, 0x11021220, 0x12001021,
701
+ 0x12001121, 0x12001222, 0x12011120, 0x12011121, 0x12021021, 0x12021120, 0x12021122, 0x10002101,
702
+ 0x10012001, 0x10012101, 0x10012202, 0x10022101, 0x11002002, 0x11002201, 0x11012000, 0x11012101,
703
+ 0x11012200, 0x11022001, 0x11022100, 0x11022102, 0x11022201, 0x12002101, 0x12012001, 0x12012100,
704
+ 0x12012102, 0x12012201, 0x12022101, 0x10002011, 0x10002111, 0x10002112, 0x10002212, 0x10012010,
705
+ 0x10012110, 0x10012111, 0x10012210, 0x10022011, 0x10022110, 0x10022112, 0x11002010, 0x11002111,
706
+ 0x11002212, 0x11012011, 0x11012012, 0x11012110, 0x11012111, 0x11012112, 0x11012211, 0x11022010,
707
+ 0x11022012, 0x11022111, 0x11022112, 0x11022212, 0x12002112, 0x12002211, 0x12012012, 0x12012111,
708
+ 0x12012112, 0x12012210, 0x12022011, 0x12022110, 0x12022112, 0x12022211, 0x10012122, 0x11002120,
709
+ 0x11002122, 0x11002221, 0x11012121, 0x11012220, 0x11012222, 0x11022120, 0x11022221, 0x12012120,
710
+ 0x12022121, 0x10100001, 0x10100100, 0x10100101, 0x10100102, 0x10100201, 0x10110002, 0x10110101,
711
+ 0x10110202, 0x10120001, 0x10120100, 0x10120201, 0x11100000, 0x11100101, 0x11100200, 0x11110001,
712
+ 0x11110100, 0x11110101, 0x11110102, 0x11110201, 0x11120101, 0x11120200, 0x12100102, 0x12100201,
713
+ 0x12110101, 0x12110200, 0x12120000, 0x12120001, 0x12120102, 0x12120201, 0x10100111, 0x10100210,
714
+ 0x10100211, 0x10100212, 0x10110011, 0x10110110, 0x10110111, 0x10110112, 0x10110210, 0x10110211,
715
+ 0x10120010, 0x10120111, 0x10120112, 0x10120210, 0x10120212, 0x11100011, 0x11100110, 0x11100111,
716
+ 0x11100112, 0x11100211, 0x11110010, 0x11110011, 0x11110012, 0x11110110, 0x11110111, 0x11110112,
717
+ 0x11110210, 0x11110211, 0x11110212, 0x11120011, 0x11120110, 0x11120111, 0x11120112, 0x11120211,
718
+ 0x12100012, 0x12100111, 0x12110011, 0x12110110, 0x12110111, 0x12110112, 0x12110211, 0x12120010,
719
+ 0x12120111, 0x12120212, 0x10100021, 0x10100122, 0x10110022, 0x10110121, 0x10110222, 0x10120021,
720
+ 0x10120120, 0x11100022, 0x11100121, 0x11100222, 0x11110021, 0x11110120, 0x11110121, 0x11110122,
721
+ 0x11110221, 0x11120022, 0x11120121, 0x12100121, 0x12110020, 0x12110022, 0x12110121, 0x12110221,
722
+ 0x12110222, 0x12120120, 0x10101100, 0x10101101, 0x10111001, 0x10111100, 0x10111101, 0x10111102,
723
+ 0x10111200, 0x10111201, 0x10121001, 0x10121101, 0x10121200, 0x10121202, 0x11101001, 0x11101100,
724
+ 0x11101101, 0x11101102, 0x11101201, 0x11101202, 0x11111000, 0x11111001, 0x11111100, 0x11111101,
725
+ 0x11111102, 0x11111200, 0x11111201, 0x11111202, 0x11121001, 0x11121002, 0x11121100, 0x11121101,
726
+ 0x11121102, 0x11121201, 0x12101000, 0x12101200, 0x12101202, 0x12111001, 0x12111100, 0x12111101,
727
+ 0x12111102, 0x12111201, 0x12121001, 0x12121100, 0x12121101, 0x12121202, 0x10101011, 0x10101012,
728
+ 0x10101110, 0x10101111, 0x10101112, 0x10101211, 0x10111010, 0x10111011, 0x10111012, 0x10111110,
729
+ 0x10111111, 0x10111112, 0x10111211, 0x10111212, 0x10121011, 0x10121110, 0x10121111, 0x10121112,
730
+ 0x10121211, 0x11101010, 0x11101011, 0x11101012, 0x11101110, 0x11101111, 0x11101112, 0x11101210,
731
+ 0x11101211, 0x11111010, 0x11111011, 0x11111012, 0x11111110, 0x11111111, 0x11111112, 0x11111210,
732
+ 0x11111211, 0x11111212, 0x11121010, 0x11121011, 0x11121110, 0x11121111, 0x11121112, 0x11121210,
733
+ 0x11121211, 0x11121212, 0x12101011, 0x12101110, 0x12101111, 0x12101211, 0x12101212, 0x12111010,
734
+ 0x12111011, 0x12111110, 0x12111111, 0x12111112, 0x12111210, 0x12111211, 0x12121011, 0x12121110,
735
+ 0x12121111, 0x12121112, 0x12121211, 0x10101020, 0x10101021, 0x10101022, 0x10101120, 0x10101122,
736
+ 0x10101220, 0x10101221, 0x10111021, 0x10111120, 0x10111121, 0x10111220, 0x10111221, 0x10121020,
737
+ 0x10121021, 0x10121022, 0x10121120, 0x10121121, 0x10121122, 0x10121220, 0x10121221, 0x11101021,
738
+ 0x11101121, 0x11101122, 0x11101220, 0x11101221, 0x11101222, 0x11111020, 0x11111021, 0x11111022,
739
+ 0x11111120, 0x11111121, 0x11111122, 0x11111220, 0x11111221, 0x11111222, 0x11121021, 0x11121120,
740
+ 0x11121121, 0x11121221, 0x12101022, 0x12101121, 0x12101122, 0x12101220, 0x12101221, 0x12101222,
741
+ 0x12111021, 0x12111121, 0x12111222, 0x12121022, 0x12121121, 0x12121122, 0x12121220, 0x12121221,
742
+ 0x10102100, 0x10102101, 0x10102102, 0x10102201, 0x10112000, 0x10112101, 0x10112200, 0x10122001,
743
+ 0x10122202, 0x11102101, 0x11102200, 0x11102202, 0x11112001, 0x11112100, 0x11112101, 0x11112102,
744
+ 0x11112200, 0x11112201, 0x11122000, 0x11122002, 0x11122100, 0x11122101, 0x12102002, 0x12102201,
745
+ 0x12112000, 0x12112002, 0x12112101, 0x12112200, 0x12122001, 0x12122201, 0x10102011, 0x10102012,
746
+ 0x10102111, 0x10102212, 0x10112011, 0x10112110, 0x10112111, 0x10112112, 0x10112211, 0x10122111,
747
+ 0x11102011, 0x11102110, 0x11102111, 0x11102112, 0x11102211, 0x11112010, 0x11112011, 0x11112012,
748
+ 0x11112110, 0x11112111, 0x11112112, 0x11112210, 0x11112211, 0x11112212, 0x11122011, 0x11122110,
749
+ 0x11122111, 0x11122112, 0x11122211, 0x12102011, 0x12102111, 0x12102211, 0x12112011, 0x12112110,
750
+ 0x12112111, 0x12112112, 0x12112210, 0x12112211, 0x12122111, 0x10102120, 0x10102220, 0x10112121,
751
+ 0x10112222, 0x10122020, 0x10122121, 0x10122122, 0x10122221, 0x11102121, 0x11102220, 0x11102221,
752
+ 0x11112021, 0x11112121, 0x11112122, 0x11112220, 0x11112221, 0x11122022, 0x11122121, 0x11122220,
753
+ 0x11122222, 0x12102021, 0x12102222, 0x12112022, 0x12112121, 0x12112122, 0x12112220, 0x12112222,
754
+ 0x12122021, 0x10200101, 0x10210100, 0x10210102, 0x10210201, 0x10220101, 0x11200100, 0x11210000,
755
+ 0x11210101, 0x11210102, 0x11210200, 0x11210202, 0x11220001, 0x11220100, 0x11220102, 0x11220201,
756
+ 0x12200001, 0x12210102, 0x12220101, 0x10200011, 0x10200110, 0x10200112, 0x10200211, 0x10210012,
757
+ 0x10210111, 0x10220011, 0x10220012, 0x10220112, 0x10220211, 0x11200111, 0x11200211, 0x11210011,
758
+ 0x11210111, 0x11210112, 0x11210211, 0x11220111, 0x11220112, 0x11220212, 0x12200110, 0x12200212,
759
+ 0x12210012, 0x12210111, 0x12220011, 0x12220112, 0x12220211, 0x10210021, 0x10210122, 0x10210221,
760
+ 0x11200020, 0x11200021, 0x11200122, 0x11210121, 0x11210122, 0x11210220, 0x11220020, 0x12200121,
761
+ 0x12210021, 0x12210122, 0x12220121, 0x10211001, 0x10211002, 0x10211101, 0x10211102, 0x10211202,
762
+ 0x10221001, 0x10221102, 0x10221201, 0x11201000, 0x11201002, 0x11201101, 0x11201200, 0x11201202,
763
+ 0x11211001, 0x11211100, 0x11211101, 0x11211102, 0x11211201, 0x11211202, 0x11221000, 0x11221002,
764
+ 0x11221101, 0x12201100, 0x12201101, 0x12201201, 0x12211000, 0x12211002, 0x12211100, 0x12211101,
765
+ 0x12211102, 0x12211200, 0x12211202, 0x12221001, 0x12221100, 0x12221201, 0x10201111, 0x10201210,
766
+ 0x10201212, 0x10211011, 0x10211111, 0x10211112, 0x10211211, 0x11201110, 0x11201111, 0x11201112,
767
+ 0x11201211, 0x11211010, 0x11211011, 0x11211110, 0x11211111, 0x11211112, 0x11211211, 0x11221011,
768
+ 0x11221110, 0x11221111, 0x11221112, 0x11221211, 0x12201112, 0x12201211, 0x12201212, 0x12211011,
769
+ 0x12211111, 0x12211112, 0x12211211, 0x12211212, 0x12221012, 0x12221111, 0x12221112, 0x12221210,
770
+ 0x10201022, 0x10201221, 0x10211121, 0x10221020, 0x10221122, 0x10221220, 0x10221221, 0x11201020,
771
+ 0x11201121, 0x11201220, 0x11201222, 0x11211021, 0x11211120, 0x11211121, 0x11211122, 0x11211220,
772
+ 0x11211222, 0x11221020, 0x11221121, 0x11221220, 0x12201020, 0x12201022, 0x12201121, 0x12201222,
773
+ 0x12211120, 0x12211122, 0x12211220, 0x12211221, 0x12221020, 0x12221120, 0x12221122, 0x12221222,
774
+ 0x10212102, 0x10212201, 0x10222101, 0x11202001, 0x11212002, 0x11212101, 0x11212202, 0x11222001,
775
+ 0x11222201, 0x12202101, 0x12212001, 0x12212200, 0x12222102, 0x10202011, 0x10202110, 0x10212010,
776
+ 0x10212111, 0x10222011, 0x10222110, 0x10222112, 0x10222211, 0x11202010, 0x11202011, 0x11202111,
777
+ 0x11202112, 0x11202210, 0x11212011, 0x11212110, 0x11212111, 0x11212112, 0x11212211, 0x11222010,
778
+ 0x11222111, 0x11222212, 0x12202012, 0x12202110, 0x12202212, 0x12212111, 0x12222011, 0x12222110,
779
+ 0x12222111, 0x12222211, 0x10212021, 0x10212122, 0x10212220, 0x11202021, 0x11202120, 0x11202221,
780
+ 0x11212020, 0x11212121, 0x11212220, 0x11212222, 0x11222120, 0x11222121, 0x11222221, 0x12202122,
781
+ 0x12212120, 0x12212220, 0x12212222, 0x12222122, 0x20000000, 0x20000002, 0x20000200, 0x20000202,
782
+ 0x20020000, 0x20020002, 0x20020200, 0x20020202, 0x21000101, 0x21010000, 0x21010001, 0x21010100,
783
+ 0x21010102, 0x21010201, 0x21020101, 0x22000000, 0x22000002, 0x22000200, 0x22000202, 0x22010101,
784
+ 0x22020000, 0x22020002, 0x22020200, 0x22020202, 0x20000111, 0x20010011, 0x20010110, 0x20010112,
785
+ 0x20010211, 0x20020111, 0x21000011, 0x21000110, 0x21000211, 0x21010010, 0x21010012, 0x21010111,
786
+ 0x21010112, 0x21010210, 0x21010211, 0x21020110, 0x21020112, 0x21020211, 0x22000111, 0x22000211,
787
+ 0x22010110, 0x22010112, 0x22010211, 0x22020111, 0x20000020, 0x20000022, 0x20000220, 0x20000222,
788
+ 0x20010121, 0x20020020, 0x20020022, 0x20020220, 0x20020222, 0x21010021, 0x21010120, 0x21010221,
789
+ 0x21020121, 0x22000020, 0x22000022, 0x22000220, 0x22000222, 0x22010121, 0x22020020, 0x22020022,
790
+ 0x22020220, 0x22020222, 0x20011100, 0x20011201, 0x21001001, 0x21001100, 0x21011001, 0x21011101,
791
+ 0x21011202, 0x21021001, 0x21021100, 0x21021201, 0x22011100, 0x22011201, 0x20001011, 0x20001211,
792
+ 0x20011012, 0x20011111, 0x20011212, 0x20021112, 0x20021211, 0x21001010, 0x21001011, 0x21001111,
793
+ 0x21001210, 0x21011011, 0x21011110, 0x21011111, 0x21011112, 0x21011211, 0x21011212, 0x21021111,
794
+ 0x21021112, 0x21021210, 0x21021212, 0x22001011, 0x22001110, 0x22001112, 0x22001211, 0x22011010,
795
+ 0x22011012, 0x22011111, 0x22011210, 0x22021112, 0x20011021, 0x20011122, 0x20011221, 0x20021121,
796
+ 0x21001021, 0x21001120, 0x21001221, 0x21001222, 0x21011020, 0x21011121, 0x21011221, 0x21011222,
797
+ 0x21021021, 0x21021122, 0x21021222, 0x22001121, 0x22011021, 0x22011222, 0x22021120, 0x20002000,
798
+ 0x20002002, 0x20002200, 0x20002202, 0x20012101, 0x20022000, 0x20022002, 0x20022200, 0x20022202,
799
+ 0x21002001, 0x21002101, 0x21012001, 0x21012100, 0x21012201, 0x21022101, 0x21022201, 0x22002000,
800
+ 0x22002002, 0x22002200, 0x22002202, 0x22012101, 0x22022000, 0x22022002, 0x22022200, 0x22022202,
801
+ 0x20002111, 0x20002112, 0x20012011, 0x20012110, 0x20012112, 0x20022111, 0x21002011, 0x21002110,
802
+ 0x21002112, 0x21002211, 0x21012010, 0x21012012, 0x21012111, 0x21012212, 0x21022011, 0x21022110,
803
+ 0x22002111, 0x22012112, 0x22012211, 0x22022111, 0x20002020, 0x20002022, 0x20002220, 0x20002222,
804
+ 0x20012121, 0x20022020, 0x20022022, 0x20022220, 0x20022222, 0x21002121, 0x21012021, 0x21012120,
805
+ 0x21012122, 0x22002020, 0x22002022, 0x22002220, 0x22002222, 0x22012121, 0x22022020, 0x22022022,
806
+ 0x22022220, 0x22022222, 0x20100101, 0x20110001, 0x20110102, 0x20110200, 0x20110201, 0x20120101,
807
+ 0x21100001, 0x21100102, 0x21100201, 0x21110101, 0x21110200, 0x21110202, 0x21120201, 0x21120202,
808
+ 0x22100101, 0x22110001, 0x22110100, 0x22110102, 0x22110201, 0x22120101, 0x20100011, 0x20100110,
809
+ 0x20100112, 0x20100211, 0x20110010, 0x20110111, 0x20110210, 0x20110212, 0x20120011, 0x20120110,
810
+ 0x20120112, 0x20120211, 0x21100010, 0x21100111, 0x21110010, 0x21110011, 0x21110110, 0x21110111,
811
+ 0x21110112, 0x21110211, 0x21120012, 0x21120111, 0x22100110, 0x22100112, 0x22110012, 0x22110111,
812
+ 0x22110210, 0x22120011, 0x22120110, 0x22120112, 0x22120211, 0x20100121, 0x20110021, 0x20110120,
813
+ 0x20110221, 0x20120121, 0x21100120, 0x21100122, 0x21100221, 0x21110020, 0x21110022, 0x21110121,
814
+ 0x21110220, 0x21120122, 0x21120221, 0x22100121, 0x22110120, 0x22110122, 0x22120221, 0x20101001,
815
+ 0x20101100, 0x20101102, 0x20111000, 0x20111101, 0x20111200, 0x20121102, 0x21101000, 0x21101202,
816
+ 0x21111001, 0x21111100, 0x21111101, 0x21111102, 0x21111200, 0x21111201, 0x21121000, 0x21121001,
817
+ 0x21121002, 0x21121101, 0x22101100, 0x22101102, 0x22111002, 0x22111100, 0x22111101, 0x22111200,
818
+ 0x22121001, 0x22121201, 0x20101010, 0x20101111, 0x20101210, 0x20101212, 0x20111010, 0x20111011,
819
+ 0x20111110, 0x20111111, 0x20111112, 0x20111211, 0x20121011, 0x20121111, 0x20121211, 0x20121212,
820
+ 0x21101011, 0x21101110, 0x21101111, 0x21101112, 0x21101211, 0x21111010, 0x21111011, 0x21111012,
821
+ 0x21111110, 0x21111111, 0x21111112, 0x21111210, 0x21111211, 0x21111212, 0x21121011, 0x21121110,
822
+ 0x21121111, 0x21121112, 0x21121211, 0x22101011, 0x22101111, 0x22101210, 0x22111011, 0x22111012,
823
+ 0x22111110, 0x22111111, 0x22111112, 0x22111211, 0x22111212, 0x22121010, 0x22121012, 0x22121111,
824
+ 0x22121210, 0x22121212, 0x20101021, 0x20101120, 0x20111020, 0x20111121, 0x20111221, 0x20121020,
825
+ 0x20121122, 0x20121221, 0x21101121, 0x21101220, 0x21101221, 0x21111021, 0x21111022, 0x21111121,
826
+ 0x21111122, 0x21111221, 0x21121121, 0x21121220, 0x22101022, 0x22101120, 0x22101221, 0x22101222,
827
+ 0x22111022, 0x22111120, 0x22111121, 0x22121120, 0x22121122, 0x22121221, 0x20102101, 0x20112102,
828
+ 0x20112201, 0x20122101, 0x21102001, 0x21102102, 0x21112000, 0x21112002, 0x21112101, 0x21112102,
829
+ 0x21112202, 0x21122100, 0x21122101, 0x22102101, 0x22112001, 0x22112102, 0x22112201, 0x22122101,
830
+ 0x20102110, 0x20102112, 0x20102211, 0x20112010, 0x20112012, 0x20112111, 0x20112210, 0x20112212,
831
+ 0x20122010, 0x20122011, 0x20122110, 0x20122112, 0x21102010, 0x21102012, 0x21102111, 0x21102210,
832
+ 0x21102212, 0x21112011, 0x21112110, 0x21112111, 0x21112112, 0x21112211, 0x21122012, 0x21122111,
833
+ 0x21122112, 0x21122212, 0x22102011, 0x22102110, 0x22112010, 0x22112012, 0x22112111, 0x22112212,
834
+ 0x22122011, 0x22122112, 0x20102121, 0x20112121, 0x20122121, 0x21102120, 0x21102122, 0x21102221,
835
+ 0x21112020, 0x21112121, 0x21112220, 0x21122021, 0x22102121, 0x22112021, 0x22112120, 0x22112121,
836
+ 0x22112122, 0x20200000, 0x20200002, 0x20200200, 0x20200202, 0x20210101, 0x20220000, 0x20220002,
837
+ 0x20220200, 0x20220202, 0x21200101, 0x21210001, 0x21210100, 0x21210102, 0x21210201, 0x22200000,
838
+ 0x22200002, 0x22200200, 0x22200202, 0x22210101, 0x22220000, 0x22220002, 0x22220200, 0x22220202,
839
+ 0x20200111, 0x20200211, 0x20210011, 0x20210110, 0x20210112, 0x20210211, 0x20210212, 0x21200112,
840
+ 0x21200211, 0x21210011, 0x21210111, 0x21210210, 0x21210212, 0x21220011, 0x21220110, 0x22200111,
841
+ 0x22210010, 0x22210012, 0x22210112, 0x22210211, 0x20200022, 0x20200220, 0x20200222, 0x20210020,
842
+ 0x20210221, 0x20220022, 0x20220220, 0x20220222, 0x21200121, 0x21210021, 0x21210122, 0x21210221,
843
+ 0x21220121, 0x22200020, 0x22200022, 0x22200220, 0x22200222, 0x22210121, 0x22220020, 0x22220022,
844
+ 0x22220220, 0x22220222, 0x20211201, 0x20221101, 0x21201001, 0x21201100, 0x21211000, 0x21211100,
845
+ 0x21211101, 0x21211200, 0x21211202, 0x21221001, 0x21221101, 0x21221102, 0x21221200, 0x21221201,
846
+ 0x22201101, 0x20201112, 0x20201211, 0x20211010, 0x20211012, 0x20211111, 0x20211210, 0x20221112,
847
+ 0x20221211, 0x21201012, 0x21201111, 0x21211011, 0x21211110, 0x21211111, 0x21211112, 0x21211211,
848
+ 0x21221111, 0x21221212, 0x22201011, 0x22201110, 0x22201111, 0x22201112, 0x22201211, 0x22211012,
849
+ 0x22211111, 0x22211210, 0x20201121, 0x20211021, 0x20211122, 0x20211222, 0x20221021, 0x20221121,
850
+ 0x21201120, 0x21201122, 0x21201222, 0x21211022, 0x21211121, 0x21211122, 0x21211220, 0x21221020,
851
+ 0x21221022, 0x22201122, 0x22211020, 0x22211121, 0x22211122, 0x22211221, 0x22221021, 0x22221120,
852
+ 0x22221122, 0x20202000, 0x20202002, 0x20202200, 0x20202202, 0x20222000, 0x20222002, 0x20222200,
853
+ 0x20222202, 0x21212001, 0x21212100, 0x21212102, 0x21212201, 0x22202000, 0x22202002, 0x22202200,
854
+ 0x22202202, 0x22212101, 0x22222000, 0x22222002, 0x22222200, 0x22222202, 0x20202111, 0x20212110,
855
+ 0x20212211, 0x20222011, 0x20222111, 0x21202011, 0x21212010, 0x21212111, 0x21212212, 0x21222011,
856
+ 0x21222112, 0x21222211, 0x22212010, 0x22212112, 0x20202020, 0x20202022, 0x20202220, 0x20202222,
857
+ 0x20222020, 0x20222022, 0x20222220, 0x20222222, 0x21212021, 0x21212120, 0x21212122, 0x22202020,
858
+ 0x22202022, 0x22202220, 0x22202222, 0x22212121, 0x22222020, 0x22222022, 0x22222220, 0x22222222,
859
+ };
860
+
861
+ shared uint16_t iq1s_grid[2048];
862
+ shared uint32_t iq1s_grid_gpu[2048];
863
+
864
+ #define NEEDS_INIT_IQ_SHMEM
865
+ void init_iq_shmem(uvec3 wgsize)
866
+ {
867
+ // copy the table into shared memory and sync
868
+ [[unroll]] for (uint i = 0; i < iq1s_grid_const.length(); i += wgsize.x) {
869
+ uint idx = i + gl_LocalInvocationIndex.x;
870
+ if (iq1s_grid_const.length() % wgsize.x == 0 || idx < iq1s_grid_const.length()) {
871
+ u16vec2 g = unpack16(iq1s_grid_const[idx]);
872
+ iq1s_grid[2*idx+0] = g.x;
873
+ iq1s_grid[2*idx+1] = g.y;
874
+ }
875
+ }
876
+ [[unroll]] for (uint i = 0; i < iq1s_grid_gpu_const.length(); i += wgsize.x) {
877
+ uint idx = i + gl_LocalInvocationIndex.x;
878
+ if (iq1s_grid_gpu_const.length() % wgsize.x == 0 || idx < iq1s_grid_gpu_const.length()) {
879
+ iq1s_grid_gpu[idx] = iq1s_grid_gpu_const[idx];
880
+ }
881
+ }
882
+ barrier();
883
+ }
884
+ #endif
885
+
886
+ #define QUANT_K_IQ2_XXS 256
887
+ #define QUANT_R_IQ2_XXS 1
888
+
889
+ struct block_iq2_xxs
890
+ {
891
+ float16_t d;
892
+ uint8_t qs[QUANT_K_IQ2_XXS/4];
893
+ };
894
+
895
+ struct block_iq2_xxs_packed16
896
+ {
897
+ float16_t d;
898
+ uint16_t qs[QUANT_K_IQ2_XXS/8];
899
+ };
900
+
901
+ #if defined(DATA_A_IQ2_XXS)
902
+
903
+ const uvec2[256] iq2xxs_grid_const = {
904
+ uvec2(0x08080808, 0x08080808), uvec2(0x0808082b, 0x08080808), uvec2(0x08081919, 0x08080808), uvec2(0x08082b08, 0x08080808),
905
+ uvec2(0x08082b2b, 0x08080808), uvec2(0x08190819, 0x08080808), uvec2(0x08191908, 0x08080808), uvec2(0x082b0808, 0x08080808),
906
+ uvec2(0x082b082b, 0x08080808), uvec2(0x082b2b08, 0x08080808), uvec2(0x082b2b2b, 0x08080808), uvec2(0x19080819, 0x08080808),
907
+ uvec2(0x19081908, 0x08080808), uvec2(0x19190808, 0x08080808), uvec2(0x19192b08, 0x08080808), uvec2(0x192b0819, 0x08080808),
908
+ uvec2(0x192b1908, 0x08080808), uvec2(0x2b080808, 0x08080808), uvec2(0x2b08082b, 0x08080808), uvec2(0x2b082b2b, 0x08080808),
909
+ uvec2(0x2b2b082b, 0x08080808), uvec2(0x08080819, 0x08080819), uvec2(0x08081908, 0x08080819), uvec2(0x08190808, 0x08080819),
910
+ uvec2(0x08191919, 0x08080819), uvec2(0x19080808, 0x08080819), uvec2(0x2b081908, 0x08080819), uvec2(0x2b192b08, 0x08080819),
911
+ uvec2(0x08080808, 0x0808082b), uvec2(0x0808082b, 0x0808082b), uvec2(0x082b082b, 0x0808082b), uvec2(0x2b08082b, 0x0808082b),
912
+ uvec2(0x08080819, 0x08081908), uvec2(0x08081908, 0x08081908), uvec2(0x08190808, 0x08081908), uvec2(0x082b0819, 0x08081908),
913
+ uvec2(0x082b1908, 0x08081908), uvec2(0x19080808, 0x08081908), uvec2(0x1908082b, 0x08081908), uvec2(0x19082b08, 0x08081908),
914
+ uvec2(0x192b0808, 0x08081908), uvec2(0x2b080819, 0x08081908), uvec2(0x2b081908, 0x08081908), uvec2(0x2b190808, 0x08081908),
915
+ uvec2(0x2b2b1908, 0x08081908), uvec2(0x08080808, 0x08081919), uvec2(0x0808082b, 0x08081919), uvec2(0x08082b08, 0x08081919),
916
+ uvec2(0x082b0808, 0x08081919), uvec2(0x1908192b, 0x08081919), uvec2(0x192b2b19, 0x08081919), uvec2(0x2b080808, 0x08081919),
917
+ uvec2(0x2b190819, 0x08081919), uvec2(0x08082b19, 0x0808192b), uvec2(0x08190808, 0x0808192b), uvec2(0x19080808, 0x0808192b),
918
+ uvec2(0x2b081908, 0x0808192b), uvec2(0x2b2b1908, 0x0808192b), uvec2(0x08080808, 0x08082b08), uvec2(0x08081919, 0x08082b08),
919
+ uvec2(0x08082b08, 0x08082b08), uvec2(0x08191908, 0x08082b08), uvec2(0x082b2b08, 0x08082b08), uvec2(0x19080819, 0x08082b08),
920
+ uvec2(0x19081908, 0x08082b08), uvec2(0x19190808, 0x08082b08), uvec2(0x1919082b, 0x08082b08), uvec2(0x2b082b08, 0x08082b08),
921
+ uvec2(0x08081908, 0x08082b19), uvec2(0x19080808, 0x08082b19), uvec2(0x0808082b, 0x08082b2b), uvec2(0x08191908, 0x08082b2b),
922
+ uvec2(0x08080819, 0x08190808), uvec2(0x08081908, 0x08190808), uvec2(0x08190808, 0x08190808), uvec2(0x082b0819, 0x08190808),
923
+ uvec2(0x19080808, 0x08190808), uvec2(0x192b0808, 0x08190808), uvec2(0x2b081908, 0x08190808), uvec2(0x2b190808, 0x08190808),
924
+ uvec2(0x2b191919, 0x08190808), uvec2(0x08080808, 0x08190819), uvec2(0x08082b08, 0x08190819), uvec2(0x082b0808, 0x08190819),
925
+ uvec2(0x19190808, 0x08190819), uvec2(0x19192b2b, 0x08190819), uvec2(0x2b080808, 0x08190819), uvec2(0x082b1908, 0x0819082b),
926
+ uvec2(0x19081919, 0x0819082b), uvec2(0x08080808, 0x08191908), uvec2(0x08082b08, 0x08191908), uvec2(0x082b0808, 0x08191908),
927
+ uvec2(0x082b1919, 0x08191908), uvec2(0x19082b19, 0x08191908), uvec2(0x2b080808, 0x08191908), uvec2(0x08192b08, 0x08191919),
928
+ uvec2(0x192b082b, 0x08191919), uvec2(0x08080808, 0x0819192b), uvec2(0x0819192b, 0x0819192b), uvec2(0x08080819, 0x08192b08),
929
+ uvec2(0x08081908, 0x08192b08), uvec2(0x08190808, 0x08192b08), uvec2(0x19080808, 0x08192b08), uvec2(0x2b080819, 0x08192b08),
930
+ uvec2(0x08080808, 0x08192b19), uvec2(0x08081919, 0x08192b19), uvec2(0x2b2b0808, 0x08192b19), uvec2(0x19190819, 0x08192b2b),
931
+ uvec2(0x08080808, 0x082b0808), uvec2(0x0808082b, 0x082b0808), uvec2(0x08082b2b, 0x082b0808), uvec2(0x19081908, 0x082b0808),
932
+ uvec2(0x192b0819, 0x082b0808), uvec2(0x2b080808, 0x082b0808), uvec2(0x2b08082b, 0x082b0808), uvec2(0x082b2b19, 0x082b0819),
933
+ uvec2(0x19082b08, 0x082b0819), uvec2(0x08080808, 0x082b082b), uvec2(0x0808082b, 0x082b082b), uvec2(0x08080819, 0x082b1908),
934
+ uvec2(0x08081908, 0x082b1908), uvec2(0x08190808, 0x082b1908), uvec2(0x19080808, 0x082b1908), uvec2(0x1919192b, 0x082b1908),
935
+ uvec2(0x08080808, 0x082b1919), uvec2(0x19080819, 0x082b1919), uvec2(0x192b1908, 0x082b1919), uvec2(0x2b190808, 0x082b192b),
936
+ uvec2(0x08082b08, 0x082b2b08), uvec2(0x082b0808, 0x082b2b08), uvec2(0x2b191908, 0x082b2b08), uvec2(0x19081908, 0x082b2b2b),
937
+ uvec2(0x08080819, 0x19080808), uvec2(0x08081908, 0x19080808), uvec2(0x08190808, 0x19080808), uvec2(0x08192b08, 0x19080808),
938
+ uvec2(0x082b0819, 0x19080808), uvec2(0x082b1908, 0x19080808), uvec2(0x19080808, 0x19080808), uvec2(0x19082b08, 0x19080808),
939
+ uvec2(0x1919192b, 0x19080808), uvec2(0x192b0808, 0x19080808), uvec2(0x2b080819, 0x19080808), uvec2(0x2b081908, 0x19080808),
940
+ uvec2(0x2b190808, 0x19080808), uvec2(0x08080808, 0x19080819), uvec2(0x082b0808, 0x19080819), uvec2(0x192b0819, 0x19080819),
941
+ uvec2(0x2b080808, 0x19080819), uvec2(0x2b081919, 0x19080819), uvec2(0x08080819, 0x1908082b), uvec2(0x08190808, 0x1908082b),
942
+ uvec2(0x19082b08, 0x1908082b), uvec2(0x1919192b, 0x1908082b), uvec2(0x192b2b08, 0x1908082b), uvec2(0x08080808, 0x19081908),
943
+ uvec2(0x08082b08, 0x19081908), uvec2(0x082b0808, 0x19081908), uvec2(0x2b080808, 0x19081908), uvec2(0x2b192b19, 0x19081908),
944
+ uvec2(0x0819082b, 0x19081919), uvec2(0x082b1908, 0x19081919), uvec2(0x08080808, 0x1908192b), uvec2(0x08080819, 0x19082b08),
945
+ uvec2(0x08081908, 0x19082b08), uvec2(0x08190808, 0x19082b08), uvec2(0x19080808, 0x19082b08), uvec2(0x19081919, 0x19082b08),
946
+ uvec2(0x08080808, 0x19082b19), uvec2(0x19192b08, 0x19082b19), uvec2(0x192b0819, 0x19082b19), uvec2(0x2b08082b, 0x19082b19),
947
+ uvec2(0x19081919, 0x19082b2b), uvec2(0x2b190808, 0x19082b2b), uvec2(0x08080808, 0x19190808), uvec2(0x08082b08, 0x19190808),
948
+ uvec2(0x08190819, 0x19190808), uvec2(0x08192b19, 0x19190808), uvec2(0x082b0808, 0x19190808), uvec2(0x2b080808, 0x19190808),
949
+ uvec2(0x2b082b08, 0x19190808), uvec2(0x08081908, 0x19190819), uvec2(0x1908082b, 0x19190819), uvec2(0x2b2b1908, 0x19190819),
950
+ uvec2(0x2b190819, 0x1919082b), uvec2(0x2b190808, 0x19191908), uvec2(0x2b19082b, 0x19191908), uvec2(0x08082b2b, 0x19191919),
951
+ uvec2(0x08080819, 0x1919192b), uvec2(0x19191908, 0x1919192b), uvec2(0x08080808, 0x19192b08), uvec2(0x08190819, 0x19192b08),
952
+ uvec2(0x08192b19, 0x19192b08), uvec2(0x192b1908, 0x19192b08), uvec2(0x19080808, 0x19192b19), uvec2(0x08082b08, 0x19192b2b),
953
+ uvec2(0x08081908, 0x192b0808), uvec2(0x08190808, 0x192b0808), uvec2(0x19080808, 0x192b0808), uvec2(0x192b2b08, 0x192b0808),
954
+ uvec2(0x08080808, 0x192b0819), uvec2(0x19191919, 0x192b0819), uvec2(0x08192b08, 0x192b082b), uvec2(0x192b0808, 0x192b082b),
955
+ uvec2(0x08080808, 0x192b1908), uvec2(0x08081919, 0x192b1908), uvec2(0x08190808, 0x192b1919), uvec2(0x0819082b, 0x192b1919),
956
+ uvec2(0x2b081908, 0x192b1919), uvec2(0x1908082b, 0x192b2b08), uvec2(0x08080808, 0x2b080808), uvec2(0x0808082b, 0x2b080808),
957
+ uvec2(0x08082b2b, 0x2b080808), uvec2(0x19080819, 0x2b080808), uvec2(0x2b08082b, 0x2b080808), uvec2(0x08081908, 0x2b080819),
958
+ uvec2(0x08192b08, 0x2b080819), uvec2(0x19080808, 0x2b080819), uvec2(0x08190819, 0x2b08082b), uvec2(0x08080819, 0x2b081908),
959
+ uvec2(0x08081908, 0x2b081908), uvec2(0x08190808, 0x2b081908), uvec2(0x08191919, 0x2b081908), uvec2(0x19080808, 0x2b081908),
960
+ uvec2(0x192b0808, 0x2b081908), uvec2(0x08080808, 0x2b081919), uvec2(0x1908192b, 0x2b081919), uvec2(0x2b191908, 0x2b081919),
961
+ uvec2(0x08082b19, 0x2b08192b), uvec2(0x19080808, 0x2b08192b), uvec2(0x192b0808, 0x2b08192b), uvec2(0x0808082b, 0x2b082b08),
962
+ uvec2(0x08081908, 0x2b082b19), uvec2(0x08190819, 0x2b082b2b), uvec2(0x08081908, 0x2b190808), uvec2(0x08190808, 0x2b190808),
963
+ uvec2(0x082b1908, 0x2b190808), uvec2(0x19080808, 0x2b190808), uvec2(0x2b2b0819, 0x2b190808), uvec2(0x0819192b, 0x2b190819),
964
+ uvec2(0x2b080808, 0x2b190819), uvec2(0x19081919, 0x2b19082b), uvec2(0x08080808, 0x2b191908), uvec2(0x082b082b, 0x2b191908),
965
+ uvec2(0x19081908, 0x2b191908), uvec2(0x19190819, 0x2b191919), uvec2(0x2b080819, 0x2b192b08), uvec2(0x082b0808, 0x2b192b19),
966
+ uvec2(0x0808082b, 0x2b2b0808), uvec2(0x19190808, 0x2b2b0808), uvec2(0x2b081919, 0x2b2b0808), uvec2(0x08082b19, 0x2b2b0819),
967
+ uvec2(0x08080808, 0x2b2b082b), uvec2(0x08192b08, 0x2b2b1908), uvec2(0x19190808, 0x2b2b2b08), uvec2(0x08081908, 0x2b2b2b19)
968
+ };
969
+
970
+ shared uvec2 iq2xxs_grid[256];
971
+
972
+ #define NEEDS_INIT_IQ_SHMEM
973
+ void init_iq_shmem(uvec3 wgsize)
974
+ {
975
+ // copy the table into shared memory and sync
976
+ [[unroll]] for (uint i = 0; i < iq2xxs_grid.length(); i += wgsize.x) {
977
+ if (iq2xxs_grid_const.length() % wgsize.x == 0 || i + gl_LocalInvocationIndex.x < iq2xxs_grid_const.length()) {
978
+ iq2xxs_grid[i + gl_LocalInvocationIndex.x] = iq2xxs_grid_const[i + gl_LocalInvocationIndex.x];
979
+ }
980
+ }
981
+ barrier();
982
+ }
983
+
984
+ #define QUANT_K QUANT_K_IQ2_XXS
985
+ #define QUANT_R QUANT_R_IQ2_XXS
986
+ #define A_TYPE block_iq2_xxs
987
+ #define A_TYPE_PACKED16 block_iq2_xxs_packed16
988
+ #endif
989
+
990
+ #define QUANT_K_IQ2_XS 256
991
+ #define QUANT_R_IQ2_XS 1
992
+
993
+ struct block_iq2_xs
994
+ {
995
+ float16_t d;
996
+ uint16_t qs[QUANT_K_IQ2_XS/8];
997
+ uint8_t scales[QUANT_K_IQ2_XS/32];
998
+ };
999
+
1000
+ struct block_iq2_xs_packed16
1001
+ {
1002
+ float16_t d;
1003
+ uint16_t qs[QUANT_K_IQ2_XS/8];
1004
+ uint16_t scales[QUANT_K_IQ2_XS/64];
1005
+ };
1006
+
1007
+ #if defined(DATA_A_IQ2_XS)
1008
+
1009
+ const uvec2 iq2xs_grid_const[512] = {
1010
+ uvec2(0x08080808, 0x08080808), uvec2(0x0808082b, 0x08080808), uvec2(0x08081919, 0x08080808), uvec2(0x08082b08, 0x08080808),
1011
+ uvec2(0x08082b2b, 0x08080808), uvec2(0x08190819, 0x08080808), uvec2(0x08191908, 0x08080808), uvec2(0x0819192b, 0x08080808),
1012
+ uvec2(0x08192b19, 0x08080808), uvec2(0x082b0808, 0x08080808), uvec2(0x082b082b, 0x08080808), uvec2(0x082b1919, 0x08080808),
1013
+ uvec2(0x082b2b08, 0x08080808), uvec2(0x19080819, 0x08080808), uvec2(0x19081908, 0x08080808), uvec2(0x1908192b, 0x08080808),
1014
+ uvec2(0x19082b19, 0x08080808), uvec2(0x19190808, 0x08080808), uvec2(0x1919082b, 0x08080808), uvec2(0x19191919, 0x08080808),
1015
+ uvec2(0x19192b08, 0x08080808), uvec2(0x192b0819, 0x08080808), uvec2(0x192b1908, 0x08080808), uvec2(0x2b080808, 0x08080808),
1016
+ uvec2(0x2b08082b, 0x08080808), uvec2(0x2b081919, 0x08080808), uvec2(0x2b082b08, 0x08080808), uvec2(0x2b190819, 0x08080808),
1017
+ uvec2(0x2b191908, 0x08080808), uvec2(0x2b192b19, 0x08080808), uvec2(0x2b2b0808, 0x08080808), uvec2(0x08080819, 0x08080819),
1018
+ uvec2(0x08081908, 0x08080819), uvec2(0x0808192b, 0x08080819), uvec2(0x08082b19, 0x08080819), uvec2(0x08190808, 0x08080819),
1019
+ uvec2(0x0819082b, 0x08080819), uvec2(0x08191919, 0x08080819), uvec2(0x08192b08, 0x08080819), uvec2(0x08192b2b, 0x08080819),
1020
+ uvec2(0x082b0819, 0x08080819), uvec2(0x082b1908, 0x08080819), uvec2(0x19080808, 0x08080819), uvec2(0x1908082b, 0x08080819),
1021
+ uvec2(0x19081919, 0x08080819), uvec2(0x19082b08, 0x08080819), uvec2(0x19190819, 0x08080819), uvec2(0x19191908, 0x08080819),
1022
+ uvec2(0x192b0808, 0x08080819), uvec2(0x192b2b08, 0x08080819), uvec2(0x2b080819, 0x08080819), uvec2(0x2b081908, 0x08080819),
1023
+ uvec2(0x2b190808, 0x08080819), uvec2(0x08080808, 0x0808082b), uvec2(0x0808082b, 0x0808082b), uvec2(0x08081919, 0x0808082b),
1024
+ uvec2(0x08082b08, 0x0808082b), uvec2(0x08190819, 0x0808082b), uvec2(0x08191908, 0x0808082b), uvec2(0x082b0808, 0x0808082b),
1025
+ uvec2(0x19080819, 0x0808082b), uvec2(0x19081908, 0x0808082b), uvec2(0x19190808, 0x0808082b), uvec2(0x19191919, 0x0808082b),
1026
+ uvec2(0x2b080808, 0x0808082b), uvec2(0x2b082b2b, 0x0808082b), uvec2(0x08080819, 0x08081908), uvec2(0x08081908, 0x08081908),
1027
+ uvec2(0x0808192b, 0x08081908), uvec2(0x08082b19, 0x08081908), uvec2(0x08190808, 0x08081908), uvec2(0x0819082b, 0x08081908),
1028
+ uvec2(0x08191919, 0x08081908), uvec2(0x08192b08, 0x08081908), uvec2(0x082b0819, 0x08081908), uvec2(0x082b1908, 0x08081908),
1029
+ uvec2(0x19080808, 0x08081908), uvec2(0x1908082b, 0x08081908), uvec2(0x19081919, 0x08081908), uvec2(0x19082b08, 0x08081908),
1030
+ uvec2(0x19190819, 0x08081908), uvec2(0x19191908, 0x08081908), uvec2(0x1919192b, 0x08081908), uvec2(0x192b0808, 0x08081908),
1031
+ uvec2(0x2b080819, 0x08081908), uvec2(0x2b081908, 0x08081908), uvec2(0x2b190808, 0x08081908), uvec2(0x08080808, 0x08081919),
1032
+ uvec2(0x0808082b, 0x08081919), uvec2(0x08081919, 0x08081919), uvec2(0x08082b08, 0x08081919), uvec2(0x08190819, 0x08081919),
1033
+ uvec2(0x08191908, 0x08081919), uvec2(0x082b0808, 0x08081919), uvec2(0x19080819, 0x08081919), uvec2(0x19081908, 0x08081919),
1034
+ uvec2(0x19190808, 0x08081919), uvec2(0x192b0819, 0x08081919), uvec2(0x2b080808, 0x08081919), uvec2(0x08080819, 0x0808192b),
1035
+ uvec2(0x08081908, 0x0808192b), uvec2(0x08190808, 0x0808192b), uvec2(0x082b192b, 0x0808192b), uvec2(0x19080808, 0x0808192b),
1036
+ uvec2(0x1908082b, 0x0808192b), uvec2(0x2b081908, 0x0808192b), uvec2(0x08080808, 0x08082b08), uvec2(0x0808082b, 0x08082b08),
1037
+ uvec2(0x08081919, 0x08082b08), uvec2(0x08082b08, 0x08082b08), uvec2(0x08082b2b, 0x08082b08), uvec2(0x08190819, 0x08082b08),
1038
+ uvec2(0x08191908, 0x08082b08), uvec2(0x082b0808, 0x08082b08), uvec2(0x082b1919, 0x08082b08), uvec2(0x19080819, 0x08082b08),
1039
+ uvec2(0x19081908, 0x08082b08), uvec2(0x19190808, 0x08082b08), uvec2(0x19192b08, 0x08082b08), uvec2(0x2b080808, 0x08082b08),
1040
+ uvec2(0x2b2b0808, 0x08082b08), uvec2(0x2b2b2b2b, 0x08082b08), uvec2(0x08080819, 0x08082b19), uvec2(0x08081908, 0x08082b19),
1041
+ uvec2(0x08190808, 0x08082b19), uvec2(0x19080808, 0x08082b19), uvec2(0x2b080819, 0x08082b19), uvec2(0x2b082b19, 0x08082b19),
1042
+ uvec2(0x08080808, 0x08082b2b), uvec2(0x082b0808, 0x08082b2b), uvec2(0x082b2b08, 0x08082b2b), uvec2(0x2b19192b, 0x08082b2b),
1043
+ uvec2(0x2b2b0808, 0x08082b2b), uvec2(0x08080819, 0x08190808), uvec2(0x08081908, 0x08190808), uvec2(0x0808192b, 0x08190808),
1044
+ uvec2(0x08082b19, 0x08190808), uvec2(0x08190808, 0x08190808), uvec2(0x0819082b, 0x08190808), uvec2(0x08191919, 0x08190808),
1045
+ uvec2(0x08192b08, 0x08190808), uvec2(0x082b0819, 0x08190808), uvec2(0x082b1908, 0x08190808), uvec2(0x19080808, 0x08190808),
1046
+ uvec2(0x1908082b, 0x08190808), uvec2(0x19081919, 0x08190808), uvec2(0x19082b08, 0x08190808), uvec2(0x19190819, 0x08190808),
1047
+ uvec2(0x19191908, 0x08190808), uvec2(0x192b0808, 0x08190808), uvec2(0x192b2b2b, 0x08190808), uvec2(0x2b080819, 0x08190808),
1048
+ uvec2(0x2b081908, 0x08190808), uvec2(0x2b190808, 0x08190808), uvec2(0x08080808, 0x08190819), uvec2(0x0808082b, 0x08190819),
1049
+ uvec2(0x08081919, 0x08190819), uvec2(0x08082b08, 0x08190819), uvec2(0x08190819, 0x08190819), uvec2(0x08191908, 0x08190819),
1050
+ uvec2(0x082b0808, 0x08190819), uvec2(0x19080819, 0x08190819), uvec2(0x19081908, 0x08190819), uvec2(0x19190808, 0x08190819),
1051
+ uvec2(0x2b080808, 0x08190819), uvec2(0x2b191908, 0x08190819), uvec2(0x2b19192b, 0x08190819), uvec2(0x08080819, 0x0819082b),
1052
+ uvec2(0x08081908, 0x0819082b), uvec2(0x0808192b, 0x0819082b), uvec2(0x08190808, 0x0819082b), uvec2(0x19080808, 0x0819082b),
1053
+ uvec2(0x192b0808, 0x0819082b), uvec2(0x08080808, 0x08191908), uvec2(0x0808082b, 0x08191908), uvec2(0x08081919, 0x08191908),
1054
+ uvec2(0x08082b08, 0x08191908), uvec2(0x08190819, 0x08191908), uvec2(0x08191908, 0x08191908), uvec2(0x082b0808, 0x08191908),
1055
+ uvec2(0x19080819, 0x08191908), uvec2(0x19081908, 0x08191908), uvec2(0x19082b19, 0x08191908), uvec2(0x19190808, 0x08191908),
1056
+ uvec2(0x192b1908, 0x08191908), uvec2(0x2b080808, 0x08191908), uvec2(0x08080819, 0x08191919), uvec2(0x08081908, 0x08191919),
1057
+ uvec2(0x08190808, 0x08191919), uvec2(0x19080808, 0x08191919), uvec2(0x08080808, 0x0819192b), uvec2(0x08191908, 0x0819192b),
1058
+ uvec2(0x19082b19, 0x0819192b), uvec2(0x08080819, 0x08192b08), uvec2(0x08081908, 0x08192b08), uvec2(0x08190808, 0x08192b08),
1059
+ uvec2(0x0819082b, 0x08192b08), uvec2(0x19080808, 0x08192b08), uvec2(0x19191908, 0x08192b08), uvec2(0x2b08192b, 0x08192b08),
1060
+ uvec2(0x08080808, 0x08192b19), uvec2(0x08081919, 0x08192b19), uvec2(0x192b192b, 0x08192b19), uvec2(0x19190819, 0x08192b2b),
1061
+ uvec2(0x2b2b2b19, 0x08192b2b), uvec2(0x08080808, 0x082b0808), uvec2(0x0808082b, 0x082b0808), uvec2(0x08081919, 0x082b0808),
1062
+ uvec2(0x08082b08, 0x082b0808), uvec2(0x08082b2b, 0x082b0808), uvec2(0x08190819, 0x082b0808), uvec2(0x08191908, 0x082b0808),
1063
+ uvec2(0x082b0808, 0x082b0808), uvec2(0x19080819, 0x082b0808), uvec2(0x19081908, 0x082b0808), uvec2(0x19190808, 0x082b0808),
1064
+ uvec2(0x2b080808, 0x082b0808), uvec2(0x2b2b0808, 0x082b0808), uvec2(0x08080819, 0x082b0819), uvec2(0x08081908, 0x082b0819),
1065
+ uvec2(0x08190808, 0x082b0819), uvec2(0x19080808, 0x082b0819), uvec2(0x19082b08, 0x082b0819), uvec2(0x192b1919, 0x082b0819),
1066
+ uvec2(0x08080808, 0x082b082b), uvec2(0x082b082b, 0x082b082b), uvec2(0x2b080808, 0x082b082b), uvec2(0x2b2b2b08, 0x082b082b),
1067
+ uvec2(0x08080819, 0x082b1908), uvec2(0x08081908, 0x082b1908), uvec2(0x08190808, 0x082b1908), uvec2(0x082b2b19, 0x082b1908),
1068
+ uvec2(0x19080808, 0x082b1908), uvec2(0x08080808, 0x082b1919), uvec2(0x19080819, 0x082b1919), uvec2(0x1919082b, 0x082b1919),
1069
+ uvec2(0x2b192b19, 0x082b1919), uvec2(0x08080819, 0x082b192b), uvec2(0x08192b2b, 0x082b192b), uvec2(0x2b2b192b, 0x082b192b),
1070
+ uvec2(0x08080808, 0x082b2b08), uvec2(0x08082b08, 0x082b2b08), uvec2(0x08082b2b, 0x082b2b08), uvec2(0x082b0808, 0x082b2b08),
1071
+ uvec2(0x19191919, 0x082b2b08), uvec2(0x2b082b08, 0x082b2b08), uvec2(0x2b2b082b, 0x082b2b08), uvec2(0x192b2b08, 0x082b2b19),
1072
+ uvec2(0x2b190808, 0x082b2b19), uvec2(0x08082b08, 0x082b2b2b), uvec2(0x082b0808, 0x082b2b2b), uvec2(0x2b08082b, 0x082b2b2b),
1073
+ uvec2(0x2b082b08, 0x082b2b2b), uvec2(0x2b082b2b, 0x082b2b2b), uvec2(0x08080819, 0x19080808), uvec2(0x08081908, 0x19080808),
1074
+ uvec2(0x0808192b, 0x19080808), uvec2(0x08082b19, 0x19080808), uvec2(0x08190808, 0x19080808), uvec2(0x0819082b, 0x19080808),
1075
+ uvec2(0x08191919, 0x19080808), uvec2(0x08192b08, 0x19080808), uvec2(0x082b0819, 0x19080808), uvec2(0x082b1908, 0x19080808),
1076
+ uvec2(0x19080808, 0x19080808), uvec2(0x1908082b, 0x19080808), uvec2(0x19081919, 0x19080808), uvec2(0x19082b08, 0x19080808),
1077
+ uvec2(0x19082b2b, 0x19080808), uvec2(0x19190819, 0x19080808), uvec2(0x19191908, 0x19080808), uvec2(0x192b0808, 0x19080808),
1078
+ uvec2(0x192b1919, 0x19080808), uvec2(0x2b080819, 0x19080808), uvec2(0x2b081908, 0x19080808), uvec2(0x2b190808, 0x19080808),
1079
+ uvec2(0x08080808, 0x19080819), uvec2(0x0808082b, 0x19080819), uvec2(0x08081919, 0x19080819), uvec2(0x08082b08, 0x19080819),
1080
+ uvec2(0x08190819, 0x19080819), uvec2(0x08191908, 0x19080819), uvec2(0x082b0808, 0x19080819), uvec2(0x19080819, 0x19080819),
1081
+ uvec2(0x19081908, 0x19080819), uvec2(0x19190808, 0x19080819), uvec2(0x2b080808, 0x19080819), uvec2(0x2b081919, 0x19080819),
1082
+ uvec2(0x2b2b082b, 0x19080819), uvec2(0x08080819, 0x1908082b), uvec2(0x08081908, 0x1908082b), uvec2(0x08190808, 0x1908082b),
1083
+ uvec2(0x0819082b, 0x1908082b), uvec2(0x082b2b19, 0x1908082b), uvec2(0x19080808, 0x1908082b), uvec2(0x08080808, 0x19081908),
1084
+ uvec2(0x0808082b, 0x19081908), uvec2(0x08081919, 0x19081908), uvec2(0x08082b08, 0x19081908), uvec2(0x08190819, 0x19081908),
1085
+ uvec2(0x08191908, 0x19081908), uvec2(0x08192b19, 0x19081908), uvec2(0x082b0808, 0x19081908), uvec2(0x19080819, 0x19081908),
1086
+ uvec2(0x19081908, 0x19081908), uvec2(0x19190808, 0x19081908), uvec2(0x2b080808, 0x19081908), uvec2(0x2b191908, 0x19081908),
1087
+ uvec2(0x08080819, 0x19081919), uvec2(0x08081908, 0x19081919), uvec2(0x08190808, 0x19081919), uvec2(0x082b1908, 0x19081919),
1088
+ uvec2(0x19080808, 0x19081919), uvec2(0x2b192b2b, 0x19081919), uvec2(0x08080808, 0x1908192b), uvec2(0x08082b2b, 0x1908192b),
1089
+ uvec2(0x19081908, 0x1908192b), uvec2(0x19190808, 0x1908192b), uvec2(0x08080819, 0x19082b08), uvec2(0x08081908, 0x19082b08),
1090
+ uvec2(0x08190808, 0x19082b08), uvec2(0x19080808, 0x19082b08), uvec2(0x19081919, 0x19082b08), uvec2(0x19191908, 0x19082b08),
1091
+ uvec2(0x192b082b, 0x19082b08), uvec2(0x08080808, 0x19082b19), uvec2(0x08190819, 0x19082b19), uvec2(0x19081908, 0x19082b19),
1092
+ uvec2(0x19190808, 0x19082b19), uvec2(0x192b2b19, 0x19082b19), uvec2(0x08081908, 0x19082b2b), uvec2(0x08080808, 0x19190808),
1093
+ uvec2(0x0808082b, 0x19190808), uvec2(0x08081919, 0x19190808), uvec2(0x08082b08, 0x19190808), uvec2(0x08190819, 0x19190808),
1094
+ uvec2(0x08191908, 0x19190808), uvec2(0x082b0808, 0x19190808), uvec2(0x082b2b08, 0x19190808), uvec2(0x19080819, 0x19190808),
1095
+ uvec2(0x19081908, 0x19190808), uvec2(0x19190808, 0x19190808), uvec2(0x2b080808, 0x19190808), uvec2(0x08080819, 0x19190819),
1096
+ uvec2(0x08081908, 0x19190819), uvec2(0x08190808, 0x19190819), uvec2(0x08191919, 0x19190819), uvec2(0x19080808, 0x19190819),
1097
+ uvec2(0x1908082b, 0x19190819), uvec2(0x08080808, 0x1919082b), uvec2(0x19081908, 0x1919082b), uvec2(0x2b2b2b2b, 0x1919082b),
1098
+ uvec2(0x08080819, 0x19191908), uvec2(0x08081908, 0x19191908), uvec2(0x08190808, 0x19191908), uvec2(0x082b0819, 0x19191908),
1099
+ uvec2(0x19080808, 0x19191908), uvec2(0x192b0808, 0x19191908), uvec2(0x2b080819, 0x19191908), uvec2(0x2b2b0819, 0x19191908),
1100
+ uvec2(0x08080808, 0x19191919), uvec2(0x08082b08, 0x19191919), uvec2(0x2b080808, 0x19191919), uvec2(0x2b082b08, 0x19191919),
1101
+ uvec2(0x082b0819, 0x1919192b), uvec2(0x192b2b08, 0x1919192b), uvec2(0x2b2b0819, 0x1919192b), uvec2(0x08080808, 0x19192b08),
1102
+ uvec2(0x08191908, 0x19192b08), uvec2(0x19080819, 0x19192b08), uvec2(0x19190808, 0x19192b08), uvec2(0x2b192b19, 0x19192b08),
1103
+ uvec2(0x08192b2b, 0x19192b19), uvec2(0x19080808, 0x19192b19), uvec2(0x1908082b, 0x19192b19), uvec2(0x2b081919, 0x19192b2b),
1104
+ uvec2(0x08080819, 0x192b0808), uvec2(0x08081908, 0x192b0808), uvec2(0x08190808, 0x192b0808), uvec2(0x19080808, 0x192b0808),
1105
+ uvec2(0x19191908, 0x192b0808), uvec2(0x192b082b, 0x192b0808), uvec2(0x2b08192b, 0x192b0808), uvec2(0x2b2b2b19, 0x192b0808),
1106
+ uvec2(0x08080808, 0x192b0819), uvec2(0x082b1908, 0x192b082b), uvec2(0x19082b2b, 0x192b082b), uvec2(0x2b19082b, 0x192b082b),
1107
+ uvec2(0x08080808, 0x192b1908), uvec2(0x0819192b, 0x192b1908), uvec2(0x08190808, 0x192b1919), uvec2(0x19080808, 0x192b1919),
1108
+ uvec2(0x19081919, 0x192b1919), uvec2(0x2b2b1908, 0x192b1919), uvec2(0x08080819, 0x192b2b08), uvec2(0x192b2b2b, 0x192b2b08),
1109
+ uvec2(0x082b1919, 0x192b2b19), uvec2(0x0808192b, 0x192b2b2b), uvec2(0x19191908, 0x192b2b2b), uvec2(0x192b082b, 0x192b2b2b),
1110
+ uvec2(0x08080808, 0x2b080808), uvec2(0x0808082b, 0x2b080808), uvec2(0x08081919, 0x2b080808), uvec2(0x08082b08, 0x2b080808),
1111
+ uvec2(0x08190819, 0x2b080808), uvec2(0x08191908, 0x2b080808), uvec2(0x082b0808, 0x2b080808), uvec2(0x082b2b2b, 0x2b080808),
1112
+ uvec2(0x19080819, 0x2b080808), uvec2(0x19081908, 0x2b080808), uvec2(0x19190808, 0x2b080808), uvec2(0x2b080808, 0x2b080808),
1113
+ uvec2(0x2b08082b, 0x2b080808), uvec2(0x2b2b2b08, 0x2b080808), uvec2(0x2b2b2b2b, 0x2b080808), uvec2(0x08080819, 0x2b080819),
1114
+ uvec2(0x08081908, 0x2b080819), uvec2(0x0808192b, 0x2b080819), uvec2(0x08190808, 0x2b080819), uvec2(0x19080808, 0x2b080819),
1115
+ uvec2(0x19190819, 0x2b080819), uvec2(0x19192b19, 0x2b080819), uvec2(0x08080808, 0x2b08082b), uvec2(0x082b0808, 0x2b08082b),
1116
+ uvec2(0x2b080808, 0x2b08082b), uvec2(0x2b08082b, 0x2b08082b), uvec2(0x2b2b0808, 0x2b08082b), uvec2(0x2b2b2b08, 0x2b08082b),
1117
+ uvec2(0x08080819, 0x2b081908), uvec2(0x08081908, 0x2b081908), uvec2(0x08190808, 0x2b081908), uvec2(0x0819082b, 0x2b081908),
1118
+ uvec2(0x08191919, 0x2b081908), uvec2(0x19080808, 0x2b081908), uvec2(0x192b0808, 0x2b081908), uvec2(0x2b082b19, 0x2b081908),
1119
+ uvec2(0x08080808, 0x2b081919), uvec2(0x19081908, 0x2b081919), uvec2(0x2b2b1919, 0x2b081919), uvec2(0x08192b08, 0x2b08192b),
1120
+ uvec2(0x192b2b2b, 0x2b08192b), uvec2(0x08080808, 0x2b082b08), uvec2(0x08082b08, 0x2b082b08), uvec2(0x082b1919, 0x2b082b08),
1121
+ uvec2(0x19192b2b, 0x2b082b08), uvec2(0x2b080808, 0x2b082b08), uvec2(0x2b08082b, 0x2b082b08), uvec2(0x2b2b2b08, 0x2b082b08),
1122
+ uvec2(0x0808192b, 0x2b082b19), uvec2(0x082b082b, 0x2b082b2b), uvec2(0x2b080808, 0x2b082b2b), uvec2(0x2b082b08, 0x2b082b2b),
1123
+ uvec2(0x2b19192b, 0x2b082b2b), uvec2(0x2b2b2b08, 0x2b082b2b), uvec2(0x08080819, 0x2b190808), uvec2(0x08081908, 0x2b190808),
1124
+ uvec2(0x08190808, 0x2b190808), uvec2(0x19080808, 0x2b190808), uvec2(0x1919192b, 0x2b190808), uvec2(0x2b081908, 0x2b190808),
1125
+ uvec2(0x08080808, 0x2b190819), uvec2(0x082b082b, 0x2b190819), uvec2(0x192b1908, 0x2b190819), uvec2(0x1919192b, 0x2b19082b),
1126
+ uvec2(0x2b082b19, 0x2b19082b), uvec2(0x08080808, 0x2b191908), uvec2(0x08081919, 0x2b191908), uvec2(0x19081908, 0x2b191908),
1127
+ uvec2(0x19190808, 0x2b191908), uvec2(0x19192b08, 0x2b191908), uvec2(0x082b2b19, 0x2b191919), uvec2(0x2b190808, 0x2b191919),
1128
+ uvec2(0x2b19082b, 0x2b191919), uvec2(0x19080819, 0x2b19192b), uvec2(0x19190819, 0x2b192b08), uvec2(0x2b2b192b, 0x2b192b08),
1129
+ uvec2(0x19082b19, 0x2b192b19), uvec2(0x08191919, 0x2b192b2b), uvec2(0x192b0808, 0x2b192b2b), uvec2(0x08080808, 0x2b2b0808),
1130
+ uvec2(0x0808082b, 0x2b2b0808), uvec2(0x08082b08, 0x2b2b0808), uvec2(0x08082b2b, 0x2b2b0808), uvec2(0x082b0808, 0x2b2b0808),
1131
+ uvec2(0x082b2b2b, 0x2b2b0808), uvec2(0x2b2b0808, 0x2b2b0808), uvec2(0x19190819, 0x2b2b0819), uvec2(0x19192b19, 0x2b2b0819),
1132
+ uvec2(0x2b2b192b, 0x2b2b0819), uvec2(0x08080808, 0x2b2b082b), uvec2(0x0808082b, 0x2b2b082b), uvec2(0x08082b08, 0x2b2b082b),
1133
+ uvec2(0x082b2b2b, 0x2b2b082b), uvec2(0x2b080808, 0x2b2b082b), uvec2(0x2b2b0808, 0x2b2b082b), uvec2(0x19080808, 0x2b2b1908),
1134
+ uvec2(0x2b191919, 0x2b2b1908), uvec2(0x192b1919, 0x2b2b192b), uvec2(0x2b192b08, 0x2b2b192b), uvec2(0x08082b2b, 0x2b2b2b08),
1135
+ uvec2(0x082b0808, 0x2b2b2b08), uvec2(0x082b082b, 0x2b2b2b08), uvec2(0x082b2b08, 0x2b2b2b08), uvec2(0x2b2b0808, 0x2b2b2b08),
1136
+ uvec2(0x2b2b2b08, 0x2b2b2b08), uvec2(0x08081908, 0x2b2b2b19), uvec2(0x2b081908, 0x2b2b2b19), uvec2(0x2b08192b, 0x2b2b2b19),
1137
+ uvec2(0x082b2b08, 0x2b2b2b2b), uvec2(0x082b2b2b, 0x2b2b2b2b), uvec2(0x2b190819, 0x2b2b2b2b), uvec2(0x2b2b2b2b, 0x2b2b2b2b),
1138
+ };
1139
+
1140
+ shared uvec2 iq2xs_grid[512];
1141
+
1142
+ #define NEEDS_INIT_IQ_SHMEM
1143
+ void init_iq_shmem(uvec3 wgsize)
1144
+ {
1145
+ // copy the table into shared memory and sync
1146
+ [[unroll]] for (uint i = 0; i < iq2xs_grid.length(); i += wgsize.x) {
1147
+ if (iq2xs_grid.length() % wgsize.x == 0 || i + gl_LocalInvocationIndex.x < iq2xs_grid_const.length()) {
1148
+ iq2xs_grid[i + gl_LocalInvocationIndex.x] = iq2xs_grid_const[i + gl_LocalInvocationIndex.x];
1149
+ }
1150
+ }
1151
+ barrier();
1152
+ }
1153
+
1154
+ #define QUANT_K QUANT_K_IQ2_XS
1155
+ #define QUANT_R QUANT_R_IQ2_XS
1156
+ #define A_TYPE block_iq2_xs
1157
+ #define A_TYPE_PACKED16 block_iq2_xs_packed16
1158
+ #endif
1159
+
1160
+ #define QUANT_K_IQ2_S 256
1161
+ #define QUANT_R_IQ2_S 1
1162
+
1163
+ struct block_iq2_s
1164
+ {
1165
+ float16_t d;
1166
+ uint8_t qs[QUANT_K_IQ2_S/4];
1167
+ uint8_t qh[QUANT_K_IQ2_S/32];
1168
+ uint8_t scales[QUANT_K_IQ2_S/32];
1169
+ };
1170
+
1171
+ struct block_iq2_s_packed16
1172
+ {
1173
+ float16_t d;
1174
+ uint16_t qs[QUANT_K_IQ2_S/8];
1175
+ uint16_t qh[QUANT_K_IQ2_S/64];
1176
+ uint16_t scales[QUANT_K_IQ2_S/64];
1177
+ };
1178
+
1179
+ #if defined(DATA_A_IQ2_S)
1180
+
1181
+ const uvec2 iq2s_grid_const[1024] = {
1182
+ uvec2(0x08080808, 0x08080808), uvec2(0x0808082b, 0x08080808), uvec2(0x08081919, 0x08080808), uvec2(0x08082b08, 0x08080808),
1183
+ uvec2(0x08082b2b, 0x08080808), uvec2(0x08190819, 0x08080808), uvec2(0x08191908, 0x08080808), uvec2(0x0819192b, 0x08080808),
1184
+ uvec2(0x08192b19, 0x08080808), uvec2(0x082b0808, 0x08080808), uvec2(0x082b082b, 0x08080808), uvec2(0x082b1919, 0x08080808),
1185
+ uvec2(0x082b2b08, 0x08080808), uvec2(0x19080819, 0x08080808), uvec2(0x19081908, 0x08080808), uvec2(0x1908192b, 0x08080808),
1186
+ uvec2(0x19082b19, 0x08080808), uvec2(0x19190808, 0x08080808), uvec2(0x1919082b, 0x08080808), uvec2(0x19191919, 0x08080808),
1187
+ uvec2(0x19192b08, 0x08080808), uvec2(0x192b0819, 0x08080808), uvec2(0x192b1908, 0x08080808), uvec2(0x192b192b, 0x08080808),
1188
+ uvec2(0x192b2b19, 0x08080808), uvec2(0x2b080808, 0x08080808), uvec2(0x2b08082b, 0x08080808), uvec2(0x2b081919, 0x08080808),
1189
+ uvec2(0x2b082b08, 0x08080808), uvec2(0x2b190819, 0x08080808), uvec2(0x2b191908, 0x08080808), uvec2(0x2b2b0808, 0x08080808),
1190
+ uvec2(0x2b2b1919, 0x08080808), uvec2(0x2b2b2b2b, 0x08080808), uvec2(0x08080819, 0x08080819), uvec2(0x08081908, 0x08080819),
1191
+ uvec2(0x0808192b, 0x08080819), uvec2(0x08082b19, 0x08080819), uvec2(0x08190808, 0x08080819), uvec2(0x0819082b, 0x08080819),
1192
+ uvec2(0x08191919, 0x08080819), uvec2(0x08192b08, 0x08080819), uvec2(0x082b0819, 0x08080819), uvec2(0x082b1908, 0x08080819),
1193
+ uvec2(0x19080808, 0x08080819), uvec2(0x1908082b, 0x08080819), uvec2(0x19081919, 0x08080819), uvec2(0x19082b08, 0x08080819),
1194
+ uvec2(0x19190819, 0x08080819), uvec2(0x19191908, 0x08080819), uvec2(0x1919192b, 0x08080819), uvec2(0x19192b19, 0x08080819),
1195
+ uvec2(0x192b0808, 0x08080819), uvec2(0x192b1919, 0x08080819), uvec2(0x192b2b08, 0x08080819), uvec2(0x2b080819, 0x08080819),
1196
+ uvec2(0x2b081908, 0x08080819), uvec2(0x2b190808, 0x08080819), uvec2(0x2b19082b, 0x08080819), uvec2(0x2b191919, 0x08080819),
1197
+ uvec2(0x2b2b0819, 0x08080819), uvec2(0x2b2b1908, 0x08080819), uvec2(0x08080808, 0x0808082b), uvec2(0x0808082b, 0x0808082b),
1198
+ uvec2(0x08081919, 0x0808082b), uvec2(0x08082b08, 0x0808082b), uvec2(0x08190819, 0x0808082b), uvec2(0x08191908, 0x0808082b),
1199
+ uvec2(0x082b0808, 0x0808082b), uvec2(0x082b2b2b, 0x0808082b), uvec2(0x19080819, 0x0808082b), uvec2(0x19081908, 0x0808082b),
1200
+ uvec2(0x1908192b, 0x0808082b), uvec2(0x19082b19, 0x0808082b), uvec2(0x19190808, 0x0808082b), uvec2(0x19191919, 0x0808082b),
1201
+ uvec2(0x2b080808, 0x0808082b), uvec2(0x2b081919, 0x0808082b), uvec2(0x2b082b2b, 0x0808082b), uvec2(0x2b191908, 0x0808082b),
1202
+ uvec2(0x2b2b082b, 0x0808082b), uvec2(0x08080819, 0x08081908), uvec2(0x08081908, 0x08081908), uvec2(0x0808192b, 0x08081908),
1203
+ uvec2(0x08082b19, 0x08081908), uvec2(0x08190808, 0x08081908), uvec2(0x0819082b, 0x08081908), uvec2(0x08191919, 0x08081908),
1204
+ uvec2(0x08192b08, 0x08081908), uvec2(0x082b0819, 0x08081908), uvec2(0x082b1908, 0x08081908), uvec2(0x082b192b, 0x08081908),
1205
+ uvec2(0x082b2b19, 0x08081908), uvec2(0x19080808, 0x08081908), uvec2(0x1908082b, 0x08081908), uvec2(0x19081919, 0x08081908),
1206
+ uvec2(0x19082b08, 0x08081908), uvec2(0x19082b2b, 0x08081908), uvec2(0x19190819, 0x08081908), uvec2(0x19191908, 0x08081908),
1207
+ uvec2(0x1919192b, 0x08081908), uvec2(0x19192b19, 0x08081908), uvec2(0x192b0808, 0x08081908), uvec2(0x192b082b, 0x08081908),
1208
+ uvec2(0x192b1919, 0x08081908), uvec2(0x2b080819, 0x08081908), uvec2(0x2b081908, 0x08081908), uvec2(0x2b08192b, 0x08081908),
1209
+ uvec2(0x2b082b19, 0x08081908), uvec2(0x2b190808, 0x08081908), uvec2(0x2b191919, 0x08081908), uvec2(0x2b192b08, 0x08081908),
1210
+ uvec2(0x2b2b0819, 0x08081908), uvec2(0x2b2b1908, 0x08081908), uvec2(0x08080808, 0x08081919), uvec2(0x0808082b, 0x08081919),
1211
+ uvec2(0x08081919, 0x08081919), uvec2(0x08082b08, 0x08081919), uvec2(0x08082b2b, 0x08081919), uvec2(0x08190819, 0x08081919),
1212
+ uvec2(0x08191908, 0x08081919), uvec2(0x0819192b, 0x08081919), uvec2(0x08192b19, 0x08081919), uvec2(0x082b0808, 0x08081919),
1213
+ uvec2(0x082b1919, 0x08081919), uvec2(0x082b2b08, 0x08081919), uvec2(0x19080819, 0x08081919), uvec2(0x19081908, 0x08081919),
1214
+ uvec2(0x1908192b, 0x08081919), uvec2(0x19082b19, 0x08081919), uvec2(0x19190808, 0x08081919), uvec2(0x1919082b, 0x08081919),
1215
+ uvec2(0x19191919, 0x08081919), uvec2(0x19192b08, 0x08081919), uvec2(0x192b0819, 0x08081919), uvec2(0x192b1908, 0x08081919),
1216
+ uvec2(0x2b080808, 0x08081919), uvec2(0x2b08082b, 0x08081919), uvec2(0x2b081919, 0x08081919), uvec2(0x2b082b08, 0x08081919),
1217
+ uvec2(0x2b190819, 0x08081919), uvec2(0x2b191908, 0x08081919), uvec2(0x2b2b0808, 0x08081919), uvec2(0x08080819, 0x0808192b),
1218
+ uvec2(0x08081908, 0x0808192b), uvec2(0x0808192b, 0x0808192b), uvec2(0x08082b19, 0x0808192b), uvec2(0x08190808, 0x0808192b),
1219
+ uvec2(0x08191919, 0x0808192b), uvec2(0x19080808, 0x0808192b), uvec2(0x19081919, 0x0808192b), uvec2(0x19082b08, 0x0808192b),
1220
+ uvec2(0x19190819, 0x0808192b), uvec2(0x19191908, 0x0808192b), uvec2(0x192b0808, 0x0808192b), uvec2(0x2b080819, 0x0808192b),
1221
+ uvec2(0x2b081908, 0x0808192b), uvec2(0x2b190808, 0x0808192b), uvec2(0x08080808, 0x08082b08), uvec2(0x0808082b, 0x08082b08),
1222
+ uvec2(0x08081919, 0x08082b08), uvec2(0x08082b08, 0x08082b08), uvec2(0x08190819, 0x08082b08), uvec2(0x08191908, 0x08082b08),
1223
+ uvec2(0x0819192b, 0x08082b08), uvec2(0x08192b19, 0x08082b08), uvec2(0x082b0808, 0x08082b08), uvec2(0x082b1919, 0x08082b08),
1224
+ uvec2(0x082b2b2b, 0x08082b08), uvec2(0x19080819, 0x08082b08), uvec2(0x19081908, 0x08082b08), uvec2(0x1908192b, 0x08082b08),
1225
+ uvec2(0x19082b19, 0x08082b08), uvec2(0x19190808, 0x08082b08), uvec2(0x1919082b, 0x08082b08), uvec2(0x19191919, 0x08082b08),
1226
+ uvec2(0x19192b08, 0x08082b08), uvec2(0x192b0819, 0x08082b08), uvec2(0x192b1908, 0x08082b08), uvec2(0x2b080808, 0x08082b08),
1227
+ uvec2(0x2b081919, 0x08082b08), uvec2(0x2b191908, 0x08082b08), uvec2(0x2b2b2b2b, 0x08082b08), uvec2(0x08080819, 0x08082b19),
1228
+ uvec2(0x08081908, 0x08082b19), uvec2(0x08190808, 0x08082b19), uvec2(0x0819082b, 0x08082b19), uvec2(0x08191919, 0x08082b19),
1229
+ uvec2(0x08192b08, 0x08082b19), uvec2(0x082b0819, 0x08082b19), uvec2(0x19080808, 0x08082b19), uvec2(0x19081919, 0x08082b19),
1230
+ uvec2(0x19082b08, 0x08082b19), uvec2(0x19190819, 0x08082b19), uvec2(0x19191908, 0x08082b19), uvec2(0x192b0808, 0x08082b19),
1231
+ uvec2(0x2b080819, 0x08082b19), uvec2(0x2b190808, 0x08082b19), uvec2(0x08080808, 0x08082b2b), uvec2(0x08190819, 0x08082b2b),
1232
+ uvec2(0x08191908, 0x08082b2b), uvec2(0x082b082b, 0x08082b2b), uvec2(0x082b2b08, 0x08082b2b), uvec2(0x082b2b2b, 0x08082b2b),
1233
+ uvec2(0x19190808, 0x08082b2b), uvec2(0x2b192b19, 0x08082b2b), uvec2(0x08080819, 0x08190808), uvec2(0x08081908, 0x08190808),
1234
+ uvec2(0x0808192b, 0x08190808), uvec2(0x08082b19, 0x08190808), uvec2(0x08190808, 0x08190808), uvec2(0x0819082b, 0x08190808),
1235
+ uvec2(0x08191919, 0x08190808), uvec2(0x08192b08, 0x08190808), uvec2(0x082b0819, 0x08190808), uvec2(0x082b1908, 0x08190808),
1236
+ uvec2(0x082b192b, 0x08190808), uvec2(0x19080808, 0x08190808), uvec2(0x1908082b, 0x08190808), uvec2(0x19081919, 0x08190808),
1237
+ uvec2(0x19082b08, 0x08190808), uvec2(0x19190819, 0x08190808), uvec2(0x19191908, 0x08190808), uvec2(0x1919192b, 0x08190808),
1238
+ uvec2(0x19192b19, 0x08190808), uvec2(0x192b0808, 0x08190808), uvec2(0x192b082b, 0x08190808), uvec2(0x192b1919, 0x08190808),
1239
+ uvec2(0x192b2b08, 0x08190808), uvec2(0x2b080819, 0x08190808), uvec2(0x2b081908, 0x08190808), uvec2(0x2b08192b, 0x08190808),
1240
+ uvec2(0x2b190808, 0x08190808), uvec2(0x2b191919, 0x08190808), uvec2(0x2b192b08, 0x08190808), uvec2(0x2b2b0819, 0x08190808),
1241
+ uvec2(0x2b2b1908, 0x08190808), uvec2(0x08080808, 0x08190819), uvec2(0x0808082b, 0x08190819), uvec2(0x08081919, 0x08190819),
1242
+ uvec2(0x08082b08, 0x08190819), uvec2(0x08082b2b, 0x08190819), uvec2(0x08190819, 0x08190819), uvec2(0x08191908, 0x08190819),
1243
+ uvec2(0x0819192b, 0x08190819), uvec2(0x08192b19, 0x08190819), uvec2(0x082b0808, 0x08190819), uvec2(0x082b082b, 0x08190819),
1244
+ uvec2(0x082b1919, 0x08190819), uvec2(0x082b2b08, 0x08190819), uvec2(0x19080819, 0x08190819), uvec2(0x19081908, 0x08190819),
1245
+ uvec2(0x1908192b, 0x08190819), uvec2(0x19082b19, 0x08190819), uvec2(0x19190808, 0x08190819), uvec2(0x1919082b, 0x08190819),
1246
+ uvec2(0x19191919, 0x08190819), uvec2(0x19192b08, 0x08190819), uvec2(0x192b0819, 0x08190819), uvec2(0x192b1908, 0x08190819),
1247
+ uvec2(0x2b080808, 0x08190819), uvec2(0x2b08082b, 0x08190819), uvec2(0x2b081919, 0x08190819), uvec2(0x2b082b08, 0x08190819),
1248
+ uvec2(0x2b190819, 0x08190819), uvec2(0x2b191908, 0x08190819), uvec2(0x08080819, 0x0819082b), uvec2(0x08081908, 0x0819082b),
1249
+ uvec2(0x08082b19, 0x0819082b), uvec2(0x08190808, 0x0819082b), uvec2(0x08191919, 0x0819082b), uvec2(0x082b0819, 0x0819082b),
1250
+ uvec2(0x082b1908, 0x0819082b), uvec2(0x19080808, 0x0819082b), uvec2(0x19081919, 0x0819082b), uvec2(0x19190819, 0x0819082b),
1251
+ uvec2(0x19191908, 0x0819082b), uvec2(0x2b080819, 0x0819082b), uvec2(0x2b081908, 0x0819082b), uvec2(0x2b190808, 0x0819082b),
1252
+ uvec2(0x08080808, 0x08191908), uvec2(0x0808082b, 0x08191908), uvec2(0x08081919, 0x08191908), uvec2(0x08082b08, 0x08191908),
1253
+ uvec2(0x08190819, 0x08191908), uvec2(0x08191908, 0x08191908), uvec2(0x0819192b, 0x08191908), uvec2(0x08192b19, 0x08191908),
1254
+ uvec2(0x082b0808, 0x08191908), uvec2(0x082b1919, 0x08191908), uvec2(0x082b2b08, 0x08191908), uvec2(0x19080819, 0x08191908),
1255
+ uvec2(0x19081908, 0x08191908), uvec2(0x1908192b, 0x08191908), uvec2(0x19082b19, 0x08191908), uvec2(0x19190808, 0x08191908),
1256
+ uvec2(0x1919082b, 0x08191908), uvec2(0x19191919, 0x08191908), uvec2(0x19192b08, 0x08191908), uvec2(0x192b0819, 0x08191908),
1257
+ uvec2(0x192b1908, 0x08191908), uvec2(0x2b080808, 0x08191908), uvec2(0x2b08082b, 0x08191908), uvec2(0x2b081919, 0x08191908),
1258
+ uvec2(0x2b082b08, 0x08191908), uvec2(0x2b190819, 0x08191908), uvec2(0x2b191908, 0x08191908), uvec2(0x2b2b0808, 0x08191908),
1259
+ uvec2(0x08080819, 0x08191919), uvec2(0x08081908, 0x08191919), uvec2(0x0808192b, 0x08191919), uvec2(0x08082b19, 0x08191919),
1260
+ uvec2(0x08190808, 0x08191919), uvec2(0x0819082b, 0x08191919), uvec2(0x08191919, 0x08191919), uvec2(0x08192b08, 0x08191919),
1261
+ uvec2(0x082b0819, 0x08191919), uvec2(0x082b1908, 0x08191919), uvec2(0x19080808, 0x08191919), uvec2(0x1908082b, 0x08191919),
1262
+ uvec2(0x19081919, 0x08191919), uvec2(0x19082b08, 0x08191919), uvec2(0x19190819, 0x08191919), uvec2(0x19191908, 0x08191919),
1263
+ uvec2(0x192b0808, 0x08191919), uvec2(0x2b080819, 0x08191919), uvec2(0x2b081908, 0x08191919), uvec2(0x2b190808, 0x08191919),
1264
+ uvec2(0x08080808, 0x0819192b), uvec2(0x08081919, 0x0819192b), uvec2(0x08082b08, 0x0819192b), uvec2(0x08190819, 0x0819192b),
1265
+ uvec2(0x08191908, 0x0819192b), uvec2(0x082b0808, 0x0819192b), uvec2(0x19080819, 0x0819192b), uvec2(0x19081908, 0x0819192b),
1266
+ uvec2(0x19190808, 0x0819192b), uvec2(0x2b080808, 0x0819192b), uvec2(0x2b2b2b2b, 0x0819192b), uvec2(0x08080819, 0x08192b08),
1267
+ uvec2(0x08081908, 0x08192b08), uvec2(0x0808192b, 0x08192b08), uvec2(0x08082b19, 0x08192b08), uvec2(0x08190808, 0x08192b08),
1268
+ uvec2(0x08191919, 0x08192b08), uvec2(0x08192b08, 0x08192b08), uvec2(0x082b0819, 0x08192b08), uvec2(0x19080808, 0x08192b08),
1269
+ uvec2(0x1908082b, 0x08192b08), uvec2(0x19081919, 0x08192b08), uvec2(0x19082b08, 0x08192b08), uvec2(0x19190819, 0x08192b08),
1270
+ uvec2(0x19191908, 0x08192b08), uvec2(0x192b0808, 0x08192b08), uvec2(0x2b080819, 0x08192b08), uvec2(0x2b081908, 0x08192b08),
1271
+ uvec2(0x08080808, 0x08192b19), uvec2(0x0808082b, 0x08192b19), uvec2(0x08081919, 0x08192b19), uvec2(0x08082b08, 0x08192b19),
1272
+ uvec2(0x08190819, 0x08192b19), uvec2(0x08191908, 0x08192b19), uvec2(0x082b0808, 0x08192b19), uvec2(0x19080819, 0x08192b19),
1273
+ uvec2(0x19081908, 0x08192b19), uvec2(0x19190808, 0x08192b19), uvec2(0x192b2b19, 0x08192b19), uvec2(0x2b2b082b, 0x08192b19),
1274
+ uvec2(0x08081908, 0x08192b2b), uvec2(0x08190808, 0x08192b2b), uvec2(0x19080808, 0x08192b2b), uvec2(0x1919192b, 0x08192b2b),
1275
+ uvec2(0x08080808, 0x082b0808), uvec2(0x0808082b, 0x082b0808), uvec2(0x08081919, 0x082b0808), uvec2(0x08082b08, 0x082b0808),
1276
+ uvec2(0x08190819, 0x082b0808), uvec2(0x08191908, 0x082b0808), uvec2(0x0819192b, 0x082b0808), uvec2(0x08192b19, 0x082b0808),
1277
+ uvec2(0x082b0808, 0x082b0808), uvec2(0x082b1919, 0x082b0808), uvec2(0x082b2b2b, 0x082b0808), uvec2(0x19080819, 0x082b0808),
1278
+ uvec2(0x19081908, 0x082b0808), uvec2(0x19190808, 0x082b0808), uvec2(0x1919082b, 0x082b0808), uvec2(0x19191919, 0x082b0808),
1279
+ uvec2(0x192b1908, 0x082b0808), uvec2(0x2b080808, 0x082b0808), uvec2(0x2b082b2b, 0x082b0808), uvec2(0x2b191908, 0x082b0808),
1280
+ uvec2(0x2b2b2b2b, 0x082b0808), uvec2(0x08080819, 0x082b0819), uvec2(0x08081908, 0x082b0819), uvec2(0x08190808, 0x082b0819),
1281
+ uvec2(0x0819082b, 0x082b0819), uvec2(0x08191919, 0x082b0819), uvec2(0x082b0819, 0x082b0819), uvec2(0x19080808, 0x082b0819),
1282
+ uvec2(0x1908082b, 0x082b0819), uvec2(0x19081919, 0x082b0819), uvec2(0x19190819, 0x082b0819), uvec2(0x19191908, 0x082b0819),
1283
+ uvec2(0x192b0808, 0x082b0819), uvec2(0x2b080819, 0x082b0819), uvec2(0x2b081908, 0x082b0819), uvec2(0x2b190808, 0x082b0819),
1284
+ uvec2(0x08080808, 0x082b082b), uvec2(0x08082b2b, 0x082b082b), uvec2(0x082b082b, 0x082b082b), uvec2(0x082b2b08, 0x082b082b),
1285
+ uvec2(0x082b2b2b, 0x082b082b), uvec2(0x19081908, 0x082b082b), uvec2(0x19190808, 0x082b082b), uvec2(0x2b082b08, 0x082b082b),
1286
+ uvec2(0x2b082b2b, 0x082b082b), uvec2(0x2b2b2b08, 0x082b082b), uvec2(0x08080819, 0x082b1908), uvec2(0x08081908, 0x082b1908),
1287
+ uvec2(0x0808192b, 0x082b1908), uvec2(0x08082b19, 0x082b1908), uvec2(0x08190808, 0x082b1908), uvec2(0x08191919, 0x082b1908),
1288
+ uvec2(0x08192b08, 0x082b1908), uvec2(0x082b0819, 0x082b1908), uvec2(0x082b1908, 0x082b1908), uvec2(0x19080808, 0x082b1908),
1289
+ uvec2(0x1908082b, 0x082b1908), uvec2(0x19081919, 0x082b1908), uvec2(0x19082b08, 0x082b1908), uvec2(0x19190819, 0x082b1908),
1290
+ uvec2(0x19191908, 0x082b1908), uvec2(0x192b0808, 0x082b1908), uvec2(0x2b080819, 0x082b1908), uvec2(0x2b081908, 0x082b1908),
1291
+ uvec2(0x2b190808, 0x082b1908), uvec2(0x08080808, 0x082b1919), uvec2(0x08081919, 0x082b1919), uvec2(0x08082b08, 0x082b1919),
1292
+ uvec2(0x08190819, 0x082b1919), uvec2(0x08191908, 0x082b1919), uvec2(0x082b0808, 0x082b1919), uvec2(0x19080819, 0x082b1919),
1293
+ uvec2(0x19081908, 0x082b1919), uvec2(0x19190808, 0x082b1919), uvec2(0x192b192b, 0x082b1919), uvec2(0x2b080808, 0x082b1919),
1294
+ uvec2(0x08080819, 0x082b192b), uvec2(0x08081908, 0x082b192b), uvec2(0x08190808, 0x082b192b), uvec2(0x19080808, 0x082b192b),
1295
+ uvec2(0x19192b19, 0x082b192b), uvec2(0x08080808, 0x082b2b08), uvec2(0x08081919, 0x082b2b08), uvec2(0x08190819, 0x082b2b08),
1296
+ uvec2(0x08191908, 0x082b2b08), uvec2(0x19080819, 0x082b2b08), uvec2(0x19081908, 0x082b2b08), uvec2(0x19190808, 0x082b2b08),
1297
+ uvec2(0x2b082b2b, 0x082b2b08), uvec2(0x2b2b2b2b, 0x082b2b08), uvec2(0x08080819, 0x082b2b19), uvec2(0x08081908, 0x082b2b19),
1298
+ uvec2(0x08190808, 0x082b2b19), uvec2(0x2b191919, 0x082b2b19), uvec2(0x08082b2b, 0x082b2b2b), uvec2(0x082b082b, 0x082b2b2b),
1299
+ uvec2(0x192b1908, 0x082b2b2b), uvec2(0x2b082b08, 0x082b2b2b), uvec2(0x2b082b2b, 0x082b2b2b), uvec2(0x08080819, 0x19080808),
1300
+ uvec2(0x08081908, 0x19080808), uvec2(0x0808192b, 0x19080808), uvec2(0x08082b19, 0x19080808), uvec2(0x08190808, 0x19080808),
1301
+ uvec2(0x0819082b, 0x19080808), uvec2(0x08191919, 0x19080808), uvec2(0x08192b08, 0x19080808), uvec2(0x08192b2b, 0x19080808),
1302
+ uvec2(0x082b0819, 0x19080808), uvec2(0x082b1908, 0x19080808), uvec2(0x082b192b, 0x19080808), uvec2(0x19080808, 0x19080808),
1303
+ uvec2(0x1908082b, 0x19080808), uvec2(0x19081919, 0x19080808), uvec2(0x19082b08, 0x19080808), uvec2(0x19082b2b, 0x19080808),
1304
+ uvec2(0x19190819, 0x19080808), uvec2(0x19191908, 0x19080808), uvec2(0x1919192b, 0x19080808), uvec2(0x19192b19, 0x19080808),
1305
+ uvec2(0x192b0808, 0x19080808), uvec2(0x192b082b, 0x19080808), uvec2(0x192b1919, 0x19080808), uvec2(0x2b080819, 0x19080808),
1306
+ uvec2(0x2b081908, 0x19080808), uvec2(0x2b190808, 0x19080808), uvec2(0x2b191919, 0x19080808), uvec2(0x2b192b08, 0x19080808),
1307
+ uvec2(0x2b2b0819, 0x19080808), uvec2(0x2b2b1908, 0x19080808), uvec2(0x08080808, 0x19080819), uvec2(0x0808082b, 0x19080819),
1308
+ uvec2(0x08081919, 0x19080819), uvec2(0x08082b08, 0x19080819), uvec2(0x08190819, 0x19080819), uvec2(0x08191908, 0x19080819),
1309
+ uvec2(0x0819192b, 0x19080819), uvec2(0x08192b19, 0x19080819), uvec2(0x082b0808, 0x19080819), uvec2(0x082b082b, 0x19080819),
1310
+ uvec2(0x082b1919, 0x19080819), uvec2(0x19080819, 0x19080819), uvec2(0x19081908, 0x19080819), uvec2(0x1908192b, 0x19080819),
1311
+ uvec2(0x19082b19, 0x19080819), uvec2(0x19190808, 0x19080819), uvec2(0x1919082b, 0x19080819), uvec2(0x19191919, 0x19080819),
1312
+ uvec2(0x19192b08, 0x19080819), uvec2(0x192b0819, 0x19080819), uvec2(0x192b1908, 0x19080819), uvec2(0x2b080808, 0x19080819),
1313
+ uvec2(0x2b08082b, 0x19080819), uvec2(0x2b081919, 0x19080819), uvec2(0x2b082b08, 0x19080819), uvec2(0x2b190819, 0x19080819),
1314
+ uvec2(0x2b191908, 0x19080819), uvec2(0x2b2b0808, 0x19080819), uvec2(0x08080819, 0x1908082b), uvec2(0x08081908, 0x1908082b),
1315
+ uvec2(0x08190808, 0x1908082b), uvec2(0x0819082b, 0x1908082b), uvec2(0x08191919, 0x1908082b), uvec2(0x08192b08, 0x1908082b),
1316
+ uvec2(0x082b1908, 0x1908082b), uvec2(0x19080808, 0x1908082b), uvec2(0x19081919, 0x1908082b), uvec2(0x19082b08, 0x1908082b),
1317
+ uvec2(0x19190819, 0x1908082b), uvec2(0x19191908, 0x1908082b), uvec2(0x192b0808, 0x1908082b), uvec2(0x2b080819, 0x1908082b),
1318
+ uvec2(0x2b081908, 0x1908082b), uvec2(0x08080808, 0x19081908), uvec2(0x0808082b, 0x19081908), uvec2(0x08081919, 0x19081908),
1319
+ uvec2(0x08082b08, 0x19081908), uvec2(0x08082b2b, 0x19081908), uvec2(0x08190819, 0x19081908), uvec2(0x08191908, 0x19081908),
1320
+ uvec2(0x0819192b, 0x19081908), uvec2(0x08192b19, 0x19081908), uvec2(0x082b0808, 0x19081908), uvec2(0x082b082b, 0x19081908),
1321
+ uvec2(0x082b1919, 0x19081908), uvec2(0x082b2b08, 0x19081908), uvec2(0x19080819, 0x19081908), uvec2(0x19081908, 0x19081908),
1322
+ uvec2(0x1908192b, 0x19081908), uvec2(0x19082b19, 0x19081908), uvec2(0x19190808, 0x19081908), uvec2(0x1919082b, 0x19081908),
1323
+ uvec2(0x19191919, 0x19081908), uvec2(0x19192b08, 0x19081908), uvec2(0x192b0819, 0x19081908), uvec2(0x192b1908, 0x19081908),
1324
+ uvec2(0x2b080808, 0x19081908), uvec2(0x2b08082b, 0x19081908), uvec2(0x2b081919, 0x19081908), uvec2(0x2b082b08, 0x19081908),
1325
+ uvec2(0x2b190819, 0x19081908), uvec2(0x2b191908, 0x19081908), uvec2(0x2b2b0808, 0x19081908), uvec2(0x08080819, 0x19081919),
1326
+ uvec2(0x08081908, 0x19081919), uvec2(0x0808192b, 0x19081919), uvec2(0x08082b19, 0x19081919), uvec2(0x08190808, 0x19081919),
1327
+ uvec2(0x0819082b, 0x19081919), uvec2(0x08191919, 0x19081919), uvec2(0x08192b08, 0x19081919), uvec2(0x082b0819, 0x19081919),
1328
+ uvec2(0x082b1908, 0x19081919), uvec2(0x19080808, 0x19081919), uvec2(0x1908082b, 0x19081919), uvec2(0x19081919, 0x19081919),
1329
+ uvec2(0x19082b08, 0x19081919), uvec2(0x19190819, 0x19081919), uvec2(0x19191908, 0x19081919), uvec2(0x192b0808, 0x19081919),
1330
+ uvec2(0x192b2b2b, 0x19081919), uvec2(0x2b080819, 0x19081919), uvec2(0x2b081908, 0x19081919), uvec2(0x2b190808, 0x19081919),
1331
+ uvec2(0x08080808, 0x1908192b), uvec2(0x0808082b, 0x1908192b), uvec2(0x08081919, 0x1908192b), uvec2(0x08082b08, 0x1908192b),
1332
+ uvec2(0x08190819, 0x1908192b), uvec2(0x08191908, 0x1908192b), uvec2(0x082b0808, 0x1908192b), uvec2(0x19080819, 0x1908192b),
1333
+ uvec2(0x19081908, 0x1908192b), uvec2(0x19190808, 0x1908192b), uvec2(0x2b080808, 0x1908192b), uvec2(0x2b2b1919, 0x1908192b),
1334
+ uvec2(0x08080819, 0x19082b08), uvec2(0x08081908, 0x19082b08), uvec2(0x08082b19, 0x19082b08), uvec2(0x08190808, 0x19082b08),
1335
+ uvec2(0x0819082b, 0x19082b08), uvec2(0x08191919, 0x19082b08), uvec2(0x08192b08, 0x19082b08), uvec2(0x082b0819, 0x19082b08),
1336
+ uvec2(0x082b1908, 0x19082b08), uvec2(0x19080808, 0x19082b08), uvec2(0x1908082b, 0x19082b08), uvec2(0x19081919, 0x19082b08),
1337
+ uvec2(0x19082b08, 0x19082b08), uvec2(0x19190819, 0x19082b08), uvec2(0x19191908, 0x19082b08), uvec2(0x192b0808, 0x19082b08),
1338
+ uvec2(0x2b081908, 0x19082b08), uvec2(0x2b190808, 0x19082b08), uvec2(0x08080808, 0x19082b19), uvec2(0x0808082b, 0x19082b19),
1339
+ uvec2(0x08081919, 0x19082b19), uvec2(0x08082b08, 0x19082b19), uvec2(0x08190819, 0x19082b19), uvec2(0x08191908, 0x19082b19),
1340
+ uvec2(0x082b0808, 0x19082b19), uvec2(0x19080819, 0x19082b19), uvec2(0x19081908, 0x19082b19), uvec2(0x19190808, 0x19082b19),
1341
+ uvec2(0x2b080808, 0x19082b19), uvec2(0x2b19192b, 0x19082b19), uvec2(0x08080819, 0x19082b2b), uvec2(0x08081908, 0x19082b2b),
1342
+ uvec2(0x08190808, 0x19082b2b), uvec2(0x19080808, 0x19082b2b), uvec2(0x08080808, 0x19190808), uvec2(0x0808082b, 0x19190808),
1343
+ uvec2(0x08081919, 0x19190808), uvec2(0x08082b08, 0x19190808), uvec2(0x08190819, 0x19190808), uvec2(0x08191908, 0x19190808),
1344
+ uvec2(0x0819192b, 0x19190808), uvec2(0x08192b19, 0x19190808), uvec2(0x082b0808, 0x19190808), uvec2(0x082b082b, 0x19190808),
1345
+ uvec2(0x082b1919, 0x19190808), uvec2(0x082b2b08, 0x19190808), uvec2(0x19080819, 0x19190808), uvec2(0x19081908, 0x19190808),
1346
+ uvec2(0x1908192b, 0x19190808), uvec2(0x19082b19, 0x19190808), uvec2(0x19190808, 0x19190808), uvec2(0x1919082b, 0x19190808),
1347
+ uvec2(0x19191919, 0x19190808), uvec2(0x19192b08, 0x19190808), uvec2(0x192b0819, 0x19190808), uvec2(0x192b1908, 0x19190808),
1348
+ uvec2(0x2b080808, 0x19190808), uvec2(0x2b08082b, 0x19190808), uvec2(0x2b081919, 0x19190808), uvec2(0x2b082b08, 0x19190808),
1349
+ uvec2(0x2b190819, 0x19190808), uvec2(0x2b191908, 0x19190808), uvec2(0x08080819, 0x19190819), uvec2(0x08081908, 0x19190819),
1350
+ uvec2(0x0808192b, 0x19190819), uvec2(0x08082b19, 0x19190819), uvec2(0x08190808, 0x19190819), uvec2(0x0819082b, 0x19190819),
1351
+ uvec2(0x08191919, 0x19190819), uvec2(0x08192b08, 0x19190819), uvec2(0x082b0819, 0x19190819), uvec2(0x082b1908, 0x19190819),
1352
+ uvec2(0x19080808, 0x19190819), uvec2(0x1908082b, 0x19190819), uvec2(0x19081919, 0x19190819), uvec2(0x19082b08, 0x19190819),
1353
+ uvec2(0x19190819, 0x19190819), uvec2(0x19191908, 0x19190819), uvec2(0x192b0808, 0x19190819), uvec2(0x2b080819, 0x19190819),
1354
+ uvec2(0x2b081908, 0x19190819), uvec2(0x2b190808, 0x19190819), uvec2(0x08080808, 0x1919082b), uvec2(0x08081919, 0x1919082b),
1355
+ uvec2(0x08082b08, 0x1919082b), uvec2(0x08190819, 0x1919082b), uvec2(0x08191908, 0x1919082b), uvec2(0x082b0808, 0x1919082b),
1356
+ uvec2(0x19080819, 0x1919082b), uvec2(0x19081908, 0x1919082b), uvec2(0x19190808, 0x1919082b), uvec2(0x192b2b19, 0x1919082b),
1357
+ uvec2(0x2b080808, 0x1919082b), uvec2(0x08080819, 0x19191908), uvec2(0x08081908, 0x19191908), uvec2(0x0808192b, 0x19191908),
1358
+ uvec2(0x08082b19, 0x19191908), uvec2(0x08190808, 0x19191908), uvec2(0x0819082b, 0x19191908), uvec2(0x08191919, 0x19191908),
1359
+ uvec2(0x08192b08, 0x19191908), uvec2(0x082b0819, 0x19191908), uvec2(0x082b1908, 0x19191908), uvec2(0x19080808, 0x19191908),
1360
+ uvec2(0x1908082b, 0x19191908), uvec2(0x19081919, 0x19191908), uvec2(0x19082b08, 0x19191908), uvec2(0x19190819, 0x19191908),
1361
+ uvec2(0x19191908, 0x19191908), uvec2(0x192b0808, 0x19191908), uvec2(0x2b080819, 0x19191908), uvec2(0x2b081908, 0x19191908),
1362
+ uvec2(0x2b190808, 0x19191908), uvec2(0x08080808, 0x19191919), uvec2(0x0808082b, 0x19191919), uvec2(0x08081919, 0x19191919),
1363
+ uvec2(0x08082b08, 0x19191919), uvec2(0x08190819, 0x19191919), uvec2(0x08191908, 0x19191919), uvec2(0x082b0808, 0x19191919),
1364
+ uvec2(0x19080819, 0x19191919), uvec2(0x19081908, 0x19191919), uvec2(0x19190808, 0x19191919), uvec2(0x2b080808, 0x19191919),
1365
+ uvec2(0x08080819, 0x1919192b), uvec2(0x08081908, 0x1919192b), uvec2(0x08190808, 0x1919192b), uvec2(0x082b192b, 0x1919192b),
1366
+ uvec2(0x19080808, 0x1919192b), uvec2(0x08080808, 0x19192b08), uvec2(0x0808082b, 0x19192b08), uvec2(0x08081919, 0x19192b08),
1367
+ uvec2(0x08082b08, 0x19192b08), uvec2(0x08190819, 0x19192b08), uvec2(0x08191908, 0x19192b08), uvec2(0x082b0808, 0x19192b08),
1368
+ uvec2(0x19080819, 0x19192b08), uvec2(0x19081908, 0x19192b08), uvec2(0x19190808, 0x19192b08), uvec2(0x19192b2b, 0x19192b08),
1369
+ uvec2(0x2b080808, 0x19192b08), uvec2(0x08080819, 0x19192b19), uvec2(0x08081908, 0x19192b19), uvec2(0x08190808, 0x19192b19),
1370
+ uvec2(0x19080808, 0x19192b19), uvec2(0x08080808, 0x19192b2b), uvec2(0x08192b19, 0x19192b2b), uvec2(0x2b081919, 0x19192b2b),
1371
+ uvec2(0x2b2b2b08, 0x19192b2b), uvec2(0x08080819, 0x192b0808), uvec2(0x08081908, 0x192b0808), uvec2(0x0808192b, 0x192b0808),
1372
+ uvec2(0x08190808, 0x192b0808), uvec2(0x0819082b, 0x192b0808), uvec2(0x08191919, 0x192b0808), uvec2(0x08192b08, 0x192b0808),
1373
+ uvec2(0x082b0819, 0x192b0808), uvec2(0x082b1908, 0x192b0808), uvec2(0x19080808, 0x192b0808), uvec2(0x19081919, 0x192b0808),
1374
+ uvec2(0x19082b08, 0x192b0808), uvec2(0x19190819, 0x192b0808), uvec2(0x19191908, 0x192b0808), uvec2(0x192b0808, 0x192b0808),
1375
+ uvec2(0x2b081908, 0x192b0808), uvec2(0x2b190808, 0x192b0808), uvec2(0x08080808, 0x192b0819), uvec2(0x0808082b, 0x192b0819),
1376
+ uvec2(0x08081919, 0x192b0819), uvec2(0x08082b08, 0x192b0819), uvec2(0x08190819, 0x192b0819), uvec2(0x08191908, 0x192b0819),
1377
+ uvec2(0x082b0808, 0x192b0819), uvec2(0x19080819, 0x192b0819), uvec2(0x19081908, 0x192b0819), uvec2(0x19190808, 0x192b0819),
1378
+ uvec2(0x2b080808, 0x192b0819), uvec2(0x2b192b19, 0x192b0819), uvec2(0x08081908, 0x192b082b), uvec2(0x08190808, 0x192b082b),
1379
+ uvec2(0x19080808, 0x192b082b), uvec2(0x1919192b, 0x192b082b), uvec2(0x2b2b0819, 0x192b082b), uvec2(0x08080808, 0x192b1908),
1380
+ uvec2(0x08081919, 0x192b1908), uvec2(0x08082b08, 0x192b1908), uvec2(0x08190819, 0x192b1908), uvec2(0x08191908, 0x192b1908),
1381
+ uvec2(0x082b0808, 0x192b1908), uvec2(0x19080819, 0x192b1908), uvec2(0x19081908, 0x192b1908), uvec2(0x19190808, 0x192b1908),
1382
+ uvec2(0x2b080808, 0x192b1908), uvec2(0x08080819, 0x192b1919), uvec2(0x08081908, 0x192b1919), uvec2(0x08190808, 0x192b1919),
1383
+ uvec2(0x19080808, 0x192b1919), uvec2(0x19082b2b, 0x192b1919), uvec2(0x192b2b08, 0x192b1919), uvec2(0x2b19082b, 0x192b1919),
1384
+ uvec2(0x08080808, 0x192b192b), uvec2(0x2b191908, 0x192b192b), uvec2(0x08080819, 0x192b2b08), uvec2(0x08081908, 0x192b2b08),
1385
+ uvec2(0x08190808, 0x192b2b08), uvec2(0x192b1919, 0x192b2b08), uvec2(0x2b192b08, 0x192b2b08), uvec2(0x08080808, 0x192b2b19),
1386
+ uvec2(0x082b2b2b, 0x192b2b19), uvec2(0x1908082b, 0x192b2b2b), uvec2(0x2b2b0819, 0x192b2b2b), uvec2(0x08080808, 0x2b080808),
1387
+ uvec2(0x0808082b, 0x2b080808), uvec2(0x08081919, 0x2b080808), uvec2(0x08082b08, 0x2b080808), uvec2(0x08190819, 0x2b080808),
1388
+ uvec2(0x08191908, 0x2b080808), uvec2(0x08192b19, 0x2b080808), uvec2(0x082b0808, 0x2b080808), uvec2(0x082b1919, 0x2b080808),
1389
+ uvec2(0x19080819, 0x2b080808), uvec2(0x19081908, 0x2b080808), uvec2(0x19190808, 0x2b080808), uvec2(0x1919082b, 0x2b080808),
1390
+ uvec2(0x19191919, 0x2b080808), uvec2(0x19192b08, 0x2b080808), uvec2(0x192b0819, 0x2b080808), uvec2(0x2b080808, 0x2b080808),
1391
+ uvec2(0x2b081919, 0x2b080808), uvec2(0x2b190819, 0x2b080808), uvec2(0x2b191908, 0x2b080808), uvec2(0x08080819, 0x2b080819),
1392
+ uvec2(0x08081908, 0x2b080819), uvec2(0x08082b19, 0x2b080819), uvec2(0x08190808, 0x2b080819), uvec2(0x0819082b, 0x2b080819),
1393
+ uvec2(0x08191919, 0x2b080819), uvec2(0x08192b08, 0x2b080819), uvec2(0x082b0819, 0x2b080819), uvec2(0x082b1908, 0x2b080819),
1394
+ uvec2(0x19080808, 0x2b080819), uvec2(0x1908082b, 0x2b080819), uvec2(0x19081919, 0x2b080819), uvec2(0x19082b08, 0x2b080819),
1395
+ uvec2(0x19190819, 0x2b080819), uvec2(0x19191908, 0x2b080819), uvec2(0x2b080819, 0x2b080819), uvec2(0x2b081908, 0x2b080819),
1396
+ uvec2(0x2b190808, 0x2b080819), uvec2(0x2b2b2b19, 0x2b080819), uvec2(0x08080808, 0x2b08082b), uvec2(0x08081919, 0x2b08082b),
1397
+ uvec2(0x08082b2b, 0x2b08082b), uvec2(0x08190819, 0x2b08082b), uvec2(0x08191908, 0x2b08082b), uvec2(0x19080819, 0x2b08082b),
1398
+ uvec2(0x19081908, 0x2b08082b), uvec2(0x19190808, 0x2b08082b), uvec2(0x08080819, 0x2b081908), uvec2(0x08081908, 0x2b081908),
1399
+ uvec2(0x0808192b, 0x2b081908), uvec2(0x08082b19, 0x2b081908), uvec2(0x08190808, 0x2b081908), uvec2(0x0819082b, 0x2b081908),
1400
+ uvec2(0x08191919, 0x2b081908), uvec2(0x08192b08, 0x2b081908), uvec2(0x082b0819, 0x2b081908), uvec2(0x19080808, 0x2b081908),
1401
+ uvec2(0x1908082b, 0x2b081908), uvec2(0x19081919, 0x2b081908), uvec2(0x19082b08, 0x2b081908), uvec2(0x19190819, 0x2b081908),
1402
+ uvec2(0x19191908, 0x2b081908), uvec2(0x192b0808, 0x2b081908), uvec2(0x2b080819, 0x2b081908), uvec2(0x2b081908, 0x2b081908),
1403
+ uvec2(0x2b190808, 0x2b081908), uvec2(0x08080808, 0x2b081919), uvec2(0x0808082b, 0x2b081919), uvec2(0x08081919, 0x2b081919),
1404
+ uvec2(0x08082b08, 0x2b081919), uvec2(0x08190819, 0x2b081919), uvec2(0x08191908, 0x2b081919), uvec2(0x082b0808, 0x2b081919),
1405
+ uvec2(0x19080819, 0x2b081919), uvec2(0x19081908, 0x2b081919), uvec2(0x19190808, 0x2b081919), uvec2(0x2b080808, 0x2b081919),
1406
+ uvec2(0x2b082b2b, 0x2b081919), uvec2(0x08080819, 0x2b08192b), uvec2(0x08081908, 0x2b08192b), uvec2(0x08190808, 0x2b08192b),
1407
+ uvec2(0x082b2b19, 0x2b08192b), uvec2(0x19080808, 0x2b08192b), uvec2(0x08080808, 0x2b082b08), uvec2(0x08081919, 0x2b082b08),
1408
+ uvec2(0x08190819, 0x2b082b08), uvec2(0x08191908, 0x2b082b08), uvec2(0x19080819, 0x2b082b08), uvec2(0x19081908, 0x2b082b08),
1409
+ uvec2(0x19190808, 0x2b082b08), uvec2(0x2b2b082b, 0x2b082b08), uvec2(0x08080819, 0x2b082b19), uvec2(0x08081908, 0x2b082b19),
1410
+ uvec2(0x19080808, 0x2b082b19), uvec2(0x192b1919, 0x2b082b19), uvec2(0x082b082b, 0x2b082b2b), uvec2(0x19192b08, 0x2b082b2b),
1411
+ uvec2(0x19192b2b, 0x2b082b2b), uvec2(0x2b08082b, 0x2b082b2b), uvec2(0x2b2b082b, 0x2b082b2b), uvec2(0x08080819, 0x2b190808),
1412
+ uvec2(0x08081908, 0x2b190808), uvec2(0x08082b19, 0x2b190808), uvec2(0x08190808, 0x2b190808), uvec2(0x0819082b, 0x2b190808),
1413
+ uvec2(0x08191919, 0x2b190808), uvec2(0x08192b08, 0x2b190808), uvec2(0x082b1908, 0x2b190808), uvec2(0x19080808, 0x2b190808),
1414
+ uvec2(0x1908082b, 0x2b190808), uvec2(0x19081919, 0x2b190808), uvec2(0x19082b08, 0x2b190808), uvec2(0x19190819, 0x2b190808),
1415
+ uvec2(0x19191908, 0x2b190808), uvec2(0x192b0808, 0x2b190808), uvec2(0x2b080819, 0x2b190808), uvec2(0x2b081908, 0x2b190808),
1416
+ uvec2(0x2b190808, 0x2b190808), uvec2(0x08080808, 0x2b190819), uvec2(0x08081919, 0x2b190819), uvec2(0x08190819, 0x2b190819),
1417
+ uvec2(0x08191908, 0x2b190819), uvec2(0x19080819, 0x2b190819), uvec2(0x19081908, 0x2b190819), uvec2(0x19190808, 0x2b190819),
1418
+ uvec2(0x19192b2b, 0x2b190819), uvec2(0x08080819, 0x2b19082b), uvec2(0x08081908, 0x2b19082b), uvec2(0x08190808, 0x2b19082b),
1419
+ uvec2(0x19080808, 0x2b19082b), uvec2(0x2b2b192b, 0x2b19082b), uvec2(0x08080808, 0x2b191908), uvec2(0x0808082b, 0x2b191908),
1420
+ uvec2(0x08081919, 0x2b191908), uvec2(0x08082b08, 0x2b191908), uvec2(0x08190819, 0x2b191908), uvec2(0x08191908, 0x2b191908),
1421
+ uvec2(0x082b0808, 0x2b191908), uvec2(0x19080819, 0x2b191908), uvec2(0x19081908, 0x2b191908), uvec2(0x19190808, 0x2b191908),
1422
+ uvec2(0x2b080808, 0x2b191908), uvec2(0x2b19192b, 0x2b191908), uvec2(0x08080819, 0x2b191919), uvec2(0x08081908, 0x2b191919),
1423
+ uvec2(0x08190808, 0x2b191919), uvec2(0x19080808, 0x2b191919), uvec2(0x2b192b08, 0x2b191919), uvec2(0x2b2b0819, 0x2b191919),
1424
+ uvec2(0x08080808, 0x2b19192b), uvec2(0x1908192b, 0x2b19192b), uvec2(0x192b1908, 0x2b19192b), uvec2(0x08080819, 0x2b192b08),
1425
+ uvec2(0x08081908, 0x2b192b08), uvec2(0x08190808, 0x2b192b08), uvec2(0x082b192b, 0x2b192b08), uvec2(0x19080808, 0x2b192b08),
1426
+ uvec2(0x2b2b2b19, 0x2b192b08), uvec2(0x08080808, 0x2b192b19), uvec2(0x19082b19, 0x2b192b19), uvec2(0x1919082b, 0x2b192b19),
1427
+ uvec2(0x2b190808, 0x2b192b2b), uvec2(0x08080808, 0x2b2b0808), uvec2(0x08081919, 0x2b2b0808), uvec2(0x08082b2b, 0x2b2b0808),
1428
+ uvec2(0x08191908, 0x2b2b0808), uvec2(0x082b082b, 0x2b2b0808), uvec2(0x082b2b2b, 0x2b2b0808), uvec2(0x19080819, 0x2b2b0808),
1429
+ uvec2(0x19081908, 0x2b2b0808), uvec2(0x19190808, 0x2b2b0808), uvec2(0x2b2b082b, 0x2b2b0808), uvec2(0x2b2b2b2b, 0x2b2b0808),
1430
+ uvec2(0x19080808, 0x2b2b0819), uvec2(0x192b1919, 0x2b2b0819), uvec2(0x0808082b, 0x2b2b082b), uvec2(0x08082b2b, 0x2b2b082b),
1431
+ uvec2(0x082b082b, 0x2b2b082b), uvec2(0x082b2b08, 0x2b2b082b), uvec2(0x082b2b2b, 0x2b2b082b), uvec2(0x2b08082b, 0x2b2b082b),
1432
+ uvec2(0x2b082b08, 0x2b2b082b), uvec2(0x2b082b2b, 0x2b2b082b), uvec2(0x2b2b2b08, 0x2b2b082b), uvec2(0x08080819, 0x2b2b1908),
1433
+ uvec2(0x08081908, 0x2b2b1908), uvec2(0x08190808, 0x2b2b1908), uvec2(0x19080808, 0x2b2b1908), uvec2(0x2b082b19, 0x2b2b1908),
1434
+ uvec2(0x2b2b1908, 0x2b2b1908), uvec2(0x08080808, 0x2b2b1919), uvec2(0x08192b19, 0x2b2b1919), uvec2(0x19190819, 0x2b2b192b),
1435
+ uvec2(0x08082b2b, 0x2b2b2b08), uvec2(0x082b2b08, 0x2b2b2b08), uvec2(0x2b2b082b, 0x2b2b2b08), uvec2(0x19191908, 0x2b2b2b19),
1436
+ uvec2(0x2b08192b, 0x2b2b2b19), uvec2(0x08082b08, 0x2b2b2b2b), uvec2(0x08082b2b, 0x2b2b2b2b), uvec2(0x082b0808, 0x2b2b2b2b),
1437
+ uvec2(0x082b082b, 0x2b2b2b2b), uvec2(0x082b2b08, 0x2b2b2b2b), uvec2(0x2b082b08, 0x2b2b2b2b), uvec2(0x2b2b2b2b, 0x2b2b2b2b)
1438
+ };
1439
+
1440
+ shared uvec2 iq2s_grid[1024];
1441
+
1442
+ #define NEEDS_INIT_IQ_SHMEM
1443
+ void init_iq_shmem(uvec3 wgsize)
1444
+ {
1445
+ // copy the table into shared memory and sync
1446
+ [[unroll]] for (uint i = 0; i < iq2s_grid.length(); i += wgsize.x) {
1447
+ if (iq2s_grid.length() % wgsize.x == 0 || i + gl_LocalInvocationIndex.x < iq2s_grid_const.length()) {
1448
+ iq2s_grid[i + gl_LocalInvocationIndex.x] = iq2s_grid_const[i + gl_LocalInvocationIndex.x];
1449
+ }
1450
+ }
1451
+ barrier();
1452
+ }
1453
+
1454
+ #define QUANT_K QUANT_K_IQ2_S
1455
+ #define QUANT_R QUANT_R_IQ2_S
1456
+ #define A_TYPE block_iq2_s
1457
+ #define A_TYPE_PACKED16 block_iq2_s_packed16
1458
+ #endif
1459
+
1460
+ #define QUANT_K_IQ3_XXS 256
1461
+ #define QUANT_R_IQ3_XXS 1
1462
+
1463
+ struct block_iq3_xxs
1464
+ {
1465
+ float16_t d;
1466
+ uint8_t qs[QUANT_K_IQ3_XXS/4 + QUANT_K_IQ3_XXS/8];
1467
+ };
1468
+
1469
+ struct block_iq3_xxs_packed16
1470
+ {
1471
+ float16_t d;
1472
+ uint16_t qs[QUANT_K_IQ3_XXS/8 + QUANT_K_IQ3_XXS/16];
1473
+ };
1474
+
1475
+ #if defined(DATA_A_IQ3_XXS)
1476
+
1477
+ const uint32_t iq3xxs_grid_const[256] = {
1478
+ 0x04040404, 0x04040414, 0x04040424, 0x04040c0c, 0x04040c1c, 0x04040c3e, 0x04041404, 0x04041414,
1479
+ 0x04041c0c, 0x04042414, 0x04043e1c, 0x04043e2c, 0x040c040c, 0x040c041c, 0x040c0c04, 0x040c0c14,
1480
+ 0x040c140c, 0x040c142c, 0x040c1c04, 0x040c1c14, 0x040c240c, 0x040c2c24, 0x040c3e04, 0x04140404,
1481
+ 0x04140414, 0x04140424, 0x04140c0c, 0x04141404, 0x04141414, 0x04141c0c, 0x04141c1c, 0x04141c3e,
1482
+ 0x04142c0c, 0x04142c3e, 0x04143e2c, 0x041c040c, 0x041c043e, 0x041c0c04, 0x041c0c14, 0x041c142c,
1483
+ 0x041c3e04, 0x04240c1c, 0x04241c3e, 0x04242424, 0x04242c3e, 0x04243e1c, 0x04243e2c, 0x042c040c,
1484
+ 0x042c043e, 0x042c1c14, 0x042c2c14, 0x04341c2c, 0x04343424, 0x043e0c04, 0x043e0c24, 0x043e0c34,
1485
+ 0x043e241c, 0x043e340c, 0x0c04040c, 0x0c04041c, 0x0c040c04, 0x0c040c14, 0x0c04140c, 0x0c04141c,
1486
+ 0x0c041c04, 0x0c041c14, 0x0c041c24, 0x0c04243e, 0x0c042c04, 0x0c0c0404, 0x0c0c0414, 0x0c0c0c0c,
1487
+ 0x0c0c1404, 0x0c0c1414, 0x0c14040c, 0x0c14041c, 0x0c140c04, 0x0c140c14, 0x0c14140c, 0x0c141c04,
1488
+ 0x0c143e14, 0x0c1c0404, 0x0c1c0414, 0x0c1c1404, 0x0c1c1c0c, 0x0c1c2434, 0x0c1c3434, 0x0c24040c,
1489
+ 0x0c24042c, 0x0c242c04, 0x0c2c1404, 0x0c2c1424, 0x0c2c2434, 0x0c2c3e0c, 0x0c34042c, 0x0c3e1414,
1490
+ 0x0c3e2404, 0x14040404, 0x14040414, 0x14040c0c, 0x14040c1c, 0x14041404, 0x14041414, 0x14041434,
1491
+ 0x14041c0c, 0x14042414, 0x140c040c, 0x140c041c, 0x140c042c, 0x140c0c04, 0x140c0c14, 0x140c140c,
1492
+ 0x140c1c04, 0x140c341c, 0x140c343e, 0x140c3e04, 0x14140404, 0x14140414, 0x14140c0c, 0x14140c3e,
1493
+ 0x14141404, 0x14141414, 0x14141c3e, 0x14142404, 0x14142c2c, 0x141c040c, 0x141c0c04, 0x141c0c24,
1494
+ 0x141c3e04, 0x141c3e24, 0x14241c2c, 0x14242c1c, 0x142c041c, 0x142c143e, 0x142c240c, 0x142c3e24,
1495
+ 0x143e040c, 0x143e041c, 0x143e0c34, 0x143e242c, 0x1c04040c, 0x1c040c04, 0x1c040c14, 0x1c04140c,
1496
+ 0x1c04141c, 0x1c042c04, 0x1c04342c, 0x1c043e14, 0x1c0c0404, 0x1c0c0414, 0x1c0c1404, 0x1c0c1c0c,
1497
+ 0x1c0c2424, 0x1c0c2434, 0x1c14040c, 0x1c14041c, 0x1c140c04, 0x1c14142c, 0x1c142c14, 0x1c143e14,
1498
+ 0x1c1c0c0c, 0x1c1c1c1c, 0x1c241c04, 0x1c24243e, 0x1c243e14, 0x1c2c0404, 0x1c2c0434, 0x1c2c1414,
1499
+ 0x1c2c2c2c, 0x1c340c24, 0x1c341c34, 0x1c34341c, 0x1c3e1c1c, 0x1c3e3404, 0x24040424, 0x24040c3e,
1500
+ 0x24041c2c, 0x24041c3e, 0x24042c1c, 0x24042c3e, 0x240c3e24, 0x24141404, 0x24141c3e, 0x24142404,
1501
+ 0x24143404, 0x24143434, 0x241c043e, 0x241c242c, 0x24240424, 0x24242c0c, 0x24243424, 0x242c142c,
1502
+ 0x242c241c, 0x242c3e04, 0x243e042c, 0x243e0c04, 0x243e0c14, 0x243e1c04, 0x2c040c14, 0x2c04240c,
1503
+ 0x2c043e04, 0x2c0c0404, 0x2c0c0434, 0x2c0c1434, 0x2c0c2c2c, 0x2c140c24, 0x2c141c14, 0x2c143e14,
1504
+ 0x2c1c0414, 0x2c1c2c1c, 0x2c240c04, 0x2c24141c, 0x2c24143e, 0x2c243e14, 0x2c2c0414, 0x2c2c1c0c,
1505
+ 0x2c342c04, 0x2c3e1424, 0x2c3e2414, 0x34041424, 0x34042424, 0x34042434, 0x34043424, 0x340c140c,
1506
+ 0x340c340c, 0x34140c3e, 0x34143424, 0x341c1c04, 0x341c1c34, 0x34242424, 0x342c042c, 0x342c2c14,
1507
+ 0x34341c1c, 0x343e041c, 0x343e140c, 0x3e04041c, 0x3e04042c, 0x3e04043e, 0x3e040c04, 0x3e041c14,
1508
+ 0x3e042c14, 0x3e0c1434, 0x3e0c2404, 0x3e140c14, 0x3e14242c, 0x3e142c14, 0x3e1c0404, 0x3e1c0c2c,
1509
+ 0x3e1c1c1c, 0x3e1c3404, 0x3e24140c, 0x3e24240c, 0x3e2c0404, 0x3e2c0414, 0x3e2c1424, 0x3e341c04,
1510
+ };
1511
+
1512
+ shared uint32_t iq3xxs_grid[256];
1513
+
1514
+ #define NEEDS_INIT_IQ_SHMEM
1515
+ void init_iq_shmem(uvec3 wgsize)
1516
+ {
1517
+ // copy the table into shared memory and sync
1518
+ [[unroll]] for (uint i = 0; i < iq3xxs_grid.length(); i += wgsize.x) {
1519
+ if (iq3xxs_grid.length() % wgsize.x == 0 || i + gl_LocalInvocationIndex.x < iq3xxs_grid.length()) {
1520
+ iq3xxs_grid[i + gl_LocalInvocationIndex.x] = iq3xxs_grid_const[i + gl_LocalInvocationIndex.x];
1521
+ }
1522
+ }
1523
+ barrier();
1524
+ }
1525
+
1526
+ #define QUANT_K QUANT_K_IQ3_XXS
1527
+ #define QUANT_R QUANT_R_IQ3_XXS
1528
+ #define A_TYPE block_iq3_xxs
1529
+ #define A_TYPE_PACKED16 block_iq3_xxs_packed16
1530
+ #endif
1531
+
1532
+ #define QUANT_K_IQ3_S 256
1533
+ #define QUANT_R_IQ3_S 1
1534
+
1535
+ struct block_iq3_s
1536
+ {
1537
+ float16_t d;
1538
+ uint8_t qs[QUANT_K_IQ3_S/4];
1539
+ uint8_t qh[QUANT_K_IQ3_S/32];
1540
+ uint8_t signs[QUANT_K_IQ3_S/8];
1541
+ uint8_t scales[QUANT_K_IQ3_S/64];
1542
+ };
1543
+
1544
+ struct block_iq3_s_packed16
1545
+ {
1546
+ float16_t d;
1547
+ uint16_t qs[QUANT_K_IQ3_S/4/2];
1548
+ uint16_t qh[QUANT_K_IQ3_S/32/2];
1549
+ uint16_t signs[QUANT_K_IQ3_S/8/2];
1550
+ uint16_t scales[QUANT_K_IQ3_S/64/2];
1551
+ };
1552
+
1553
+ #if defined(DATA_A_IQ3_S)
1554
+
1555
+ const uint32_t iq3s_grid_const[512] = {
1556
+ 0x01010101, 0x01010103, 0x01010105, 0x0101010b, 0x0101010f, 0x01010301, 0x01010303, 0x01010305,
1557
+ 0x01010309, 0x0101030d, 0x01010501, 0x01010503, 0x0101050b, 0x01010707, 0x01010901, 0x01010905,
1558
+ 0x0101090b, 0x0101090f, 0x01010b03, 0x01010b07, 0x01010d01, 0x01010d05, 0x01010f03, 0x01010f09,
1559
+ 0x01010f0f, 0x01030101, 0x01030103, 0x01030105, 0x01030109, 0x01030301, 0x01030303, 0x0103030b,
1560
+ 0x01030501, 0x01030507, 0x0103050f, 0x01030703, 0x0103070b, 0x01030909, 0x01030d03, 0x01030d0b,
1561
+ 0x01030f05, 0x01050101, 0x01050103, 0x0105010b, 0x0105010f, 0x01050301, 0x01050307, 0x0105030d,
1562
+ 0x01050503, 0x0105050b, 0x01050701, 0x01050709, 0x01050905, 0x0105090b, 0x0105090f, 0x01050b03,
1563
+ 0x01050b07, 0x01050f01, 0x01050f07, 0x01070107, 0x01070303, 0x0107030b, 0x01070501, 0x01070505,
1564
+ 0x01070703, 0x01070707, 0x0107070d, 0x01070909, 0x01070b01, 0x01070b05, 0x01070d0f, 0x01070f03,
1565
+ 0x01070f0b, 0x01090101, 0x01090307, 0x0109030f, 0x01090503, 0x01090509, 0x01090705, 0x01090901,
1566
+ 0x01090907, 0x01090b03, 0x01090f01, 0x010b0105, 0x010b0109, 0x010b0501, 0x010b0505, 0x010b050d,
1567
+ 0x010b0707, 0x010b0903, 0x010b090b, 0x010b090f, 0x010b0d0d, 0x010b0f07, 0x010d010d, 0x010d0303,
1568
+ 0x010d0307, 0x010d0703, 0x010d0b05, 0x010d0f03, 0x010f0101, 0x010f0105, 0x010f0109, 0x010f0501,
1569
+ 0x010f0505, 0x010f050d, 0x010f0707, 0x010f0b01, 0x010f0b09, 0x03010101, 0x03010103, 0x03010105,
1570
+ 0x03010109, 0x03010301, 0x03010303, 0x03010307, 0x0301030b, 0x0301030f, 0x03010501, 0x03010505,
1571
+ 0x03010703, 0x03010709, 0x0301070d, 0x03010b09, 0x03010b0d, 0x03010d03, 0x03010f05, 0x03030101,
1572
+ 0x03030103, 0x03030107, 0x0303010d, 0x03030301, 0x03030309, 0x03030503, 0x03030701, 0x03030707,
1573
+ 0x03030903, 0x03030b01, 0x03030b05, 0x03030f01, 0x03030f0d, 0x03050101, 0x03050305, 0x0305030b,
1574
+ 0x0305030f, 0x03050501, 0x03050509, 0x03050705, 0x03050901, 0x03050907, 0x03050b0b, 0x03050d01,
1575
+ 0x03050f05, 0x03070103, 0x03070109, 0x0307010f, 0x03070301, 0x03070307, 0x03070503, 0x0307050f,
1576
+ 0x03070701, 0x03070709, 0x03070903, 0x03070d05, 0x03070f01, 0x03090107, 0x0309010b, 0x03090305,
1577
+ 0x03090309, 0x03090703, 0x03090707, 0x03090905, 0x0309090d, 0x03090b01, 0x03090b09, 0x030b0103,
1578
+ 0x030b0301, 0x030b0307, 0x030b0503, 0x030b0701, 0x030b0705, 0x030b0b03, 0x030d0501, 0x030d0509,
1579
+ 0x030d050f, 0x030d0909, 0x030d090d, 0x030f0103, 0x030f0107, 0x030f0301, 0x030f0305, 0x030f0503,
1580
+ 0x030f070b, 0x030f0903, 0x030f0d05, 0x030f0f01, 0x05010101, 0x05010103, 0x05010107, 0x0501010b,
1581
+ 0x0501010f, 0x05010301, 0x05010305, 0x05010309, 0x0501030d, 0x05010503, 0x05010507, 0x0501050f,
1582
+ 0x05010701, 0x05010705, 0x05010903, 0x05010907, 0x0501090b, 0x05010b01, 0x05010b05, 0x05010d0f,
1583
+ 0x05010f01, 0x05010f07, 0x05010f0b, 0x05030101, 0x05030105, 0x05030301, 0x05030307, 0x0503030f,
1584
+ 0x05030505, 0x0503050b, 0x05030703, 0x05030709, 0x05030905, 0x05030b03, 0x05050103, 0x05050109,
1585
+ 0x0505010f, 0x05050503, 0x05050507, 0x05050701, 0x0505070f, 0x05050903, 0x05050b07, 0x05050b0f,
1586
+ 0x05050f03, 0x05050f09, 0x05070101, 0x05070105, 0x0507010b, 0x05070303, 0x05070505, 0x05070509,
1587
+ 0x05070703, 0x05070707, 0x05070905, 0x05070b01, 0x05070d0d, 0x05090103, 0x0509010f, 0x05090501,
1588
+ 0x05090507, 0x05090705, 0x0509070b, 0x05090903, 0x05090f05, 0x05090f0b, 0x050b0109, 0x050b0303,
1589
+ 0x050b0505, 0x050b070f, 0x050b0901, 0x050b0b07, 0x050b0f01, 0x050d0101, 0x050d0105, 0x050d010f,
1590
+ 0x050d0503, 0x050d0b0b, 0x050d0d03, 0x050f010b, 0x050f0303, 0x050f050d, 0x050f0701, 0x050f0907,
1591
+ 0x050f0b01, 0x07010105, 0x07010303, 0x07010307, 0x0701030b, 0x0701030f, 0x07010505, 0x07010703,
1592
+ 0x07010707, 0x0701070b, 0x07010905, 0x07010909, 0x0701090f, 0x07010b03, 0x07010d07, 0x07010f03,
1593
+ 0x07030103, 0x07030107, 0x0703010b, 0x07030309, 0x07030503, 0x07030507, 0x07030901, 0x07030d01,
1594
+ 0x07030f05, 0x07030f0d, 0x07050101, 0x07050305, 0x07050501, 0x07050705, 0x07050709, 0x07050b01,
1595
+ 0x07070103, 0x07070301, 0x07070309, 0x07070503, 0x07070507, 0x0707050f, 0x07070701, 0x07070903,
1596
+ 0x07070907, 0x0707090f, 0x07070b0b, 0x07070f07, 0x07090107, 0x07090303, 0x0709030d, 0x07090505,
1597
+ 0x07090703, 0x07090b05, 0x07090d01, 0x07090d09, 0x070b0103, 0x070b0301, 0x070b0305, 0x070b050b,
1598
+ 0x070b0705, 0x070b0909, 0x070b0b0d, 0x070b0f07, 0x070d030d, 0x070d0903, 0x070f0103, 0x070f0107,
1599
+ 0x070f0501, 0x070f0505, 0x070f070b, 0x09010101, 0x09010109, 0x09010305, 0x09010501, 0x09010509,
1600
+ 0x0901050f, 0x09010705, 0x09010903, 0x09010b01, 0x09010f01, 0x09030105, 0x0903010f, 0x09030303,
1601
+ 0x09030307, 0x09030505, 0x09030701, 0x0903070b, 0x09030907, 0x09030b03, 0x09030b0b, 0x09050103,
1602
+ 0x09050107, 0x09050301, 0x0905030b, 0x09050503, 0x09050707, 0x09050901, 0x09050b0f, 0x09050d05,
1603
+ 0x09050f01, 0x09070109, 0x09070303, 0x09070307, 0x09070501, 0x09070505, 0x09070703, 0x0907070b,
1604
+ 0x09090101, 0x09090105, 0x09090509, 0x0909070f, 0x09090901, 0x09090f03, 0x090b010b, 0x090b010f,
1605
+ 0x090b0503, 0x090b0d05, 0x090d0307, 0x090d0709, 0x090d0d01, 0x090f0301, 0x090f030b, 0x090f0701,
1606
+ 0x090f0907, 0x090f0b03, 0x0b010105, 0x0b010301, 0x0b010309, 0x0b010505, 0x0b010901, 0x0b010909,
1607
+ 0x0b01090f, 0x0b010b05, 0x0b010d0d, 0x0b010f09, 0x0b030103, 0x0b030107, 0x0b03010b, 0x0b030305,
1608
+ 0x0b030503, 0x0b030705, 0x0b030f05, 0x0b050101, 0x0b050303, 0x0b050507, 0x0b050701, 0x0b05070d,
1609
+ 0x0b050b07, 0x0b070105, 0x0b07010f, 0x0b070301, 0x0b07050f, 0x0b070909, 0x0b070b03, 0x0b070d0b,
1610
+ 0x0b070f07, 0x0b090103, 0x0b090109, 0x0b090501, 0x0b090705, 0x0b09090d, 0x0b0b0305, 0x0b0b050d,
1611
+ 0x0b0b0b03, 0x0b0b0b07, 0x0b0d0905, 0x0b0f0105, 0x0b0f0109, 0x0b0f0505, 0x0d010303, 0x0d010307,
1612
+ 0x0d01030b, 0x0d010703, 0x0d010707, 0x0d010d01, 0x0d030101, 0x0d030501, 0x0d03050f, 0x0d030d09,
1613
+ 0x0d050305, 0x0d050709, 0x0d050905, 0x0d050b0b, 0x0d050d05, 0x0d050f01, 0x0d070101, 0x0d070309,
1614
+ 0x0d070503, 0x0d070901, 0x0d09050b, 0x0d090907, 0x0d090d05, 0x0d0b0101, 0x0d0b0107, 0x0d0b0709,
1615
+ 0x0d0b0d01, 0x0d0d010b, 0x0d0d0901, 0x0d0f0303, 0x0d0f0307, 0x0f010101, 0x0f010109, 0x0f01010f,
1616
+ 0x0f010501, 0x0f010505, 0x0f01070d, 0x0f010901, 0x0f010b09, 0x0f010d05, 0x0f030105, 0x0f030303,
1617
+ 0x0f030509, 0x0f030907, 0x0f03090b, 0x0f050103, 0x0f050109, 0x0f050301, 0x0f05030d, 0x0f050503,
1618
+ 0x0f050701, 0x0f050b03, 0x0f070105, 0x0f070705, 0x0f07070b, 0x0f070b07, 0x0f090103, 0x0f09010b,
1619
+ 0x0f090307, 0x0f090501, 0x0f090b01, 0x0f0b0505, 0x0f0b0905, 0x0f0d0105, 0x0f0d0703, 0x0f0f0101,
1620
+ };
1621
+
1622
+ shared uint32_t iq3s_grid[512];
1623
+
1624
+ #define NEEDS_INIT_IQ_SHMEM
1625
+ void init_iq_shmem(uvec3 wgsize)
1626
+ {
1627
+ // copy the table into shared memory and sync
1628
+ [[unroll]] for (uint i = 0; i < iq3s_grid.length(); i += wgsize.x) {
1629
+ if (iq3s_grid.length() % wgsize.x == 0 || i + gl_LocalInvocationIndex.x < iq3s_grid.length()) {
1630
+ iq3s_grid[i + gl_LocalInvocationIndex.x] = iq3s_grid_const[i + gl_LocalInvocationIndex.x];
1631
+ }
1632
+ }
1633
+ barrier();
1634
+ }
1635
+
1636
+ #define QUANT_K QUANT_K_IQ3_S
1637
+ #define QUANT_R QUANT_R_IQ3_S
1638
+ #define A_TYPE block_iq3_s
1639
+ #define A_TYPE_PACKED16 block_iq3_s_packed16
1640
+ #endif
1641
+
1642
+ #define QUANT_K_IQ4_XS 256
1643
+ #define QUANT_R_IQ4_XS 1
1644
+
1645
+ struct block_iq4_xs
1646
+ {
1647
+ float16_t d;
1648
+ uint16_t scales_h;
1649
+ uint8_t scales_l[QUANT_K_IQ4_XS/64];
1650
+ uint8_t qs[QUANT_K_IQ4_XS/2];
1651
+ };
1652
+
1653
+ struct block_iq4_xs_packed16
1654
+ {
1655
+ float16_t d;
1656
+ uint16_t scales_h;
1657
+ uint16_t scales_l[QUANT_K_IQ4_XS/128];
1658
+ uint16_t qs[QUANT_K_IQ4_XS/4];
1659
+ };
1660
+
1661
+ struct block_iq4_xs_packed32
1662
+ {
1663
+ float16_t d;
1664
+ uint16_t scales_h;
1665
+ uint32_t scales_l;
1666
+ uint32_t qs[QUANT_K_IQ4_XS/8];
1667
+ };
1668
+
1669
+ #if defined(DATA_A_IQ4_XS)
1670
+ #define QUANT_K QUANT_K_IQ4_XS
1671
+ #define QUANT_R QUANT_R_IQ4_XS
1672
+ #define A_TYPE block_iq4_xs
1673
+ #define A_TYPE_PACKED16 block_iq4_xs_packed16
1674
+ #define A_TYPE_PACKED32 block_iq4_xs_packed32
1675
+ #endif
1676
+
1677
+ #define QUANT_K_IQ4_NL 32
1678
+ #define QUANT_R_IQ4_NL 2
1679
+
1680
+ struct block_iq4_nl
1681
+ {
1682
+ float16_t d;
1683
+ uint8_t qs[QUANT_K_IQ4_NL/2];
1684
+ };
1685
+
1686
+ struct block_iq4_nl_packed16
1687
+ {
1688
+ float16_t d;
1689
+ uint16_t qs[QUANT_K_IQ4_NL/2/2];
1690
+ };
1691
+
1692
+ #if defined(DATA_A_IQ4_NL)
1693
+ #define QUANT_K QUANT_K_IQ4_NL
1694
+ #define QUANT_R QUANT_R_IQ4_NL
1695
+ #define QUANT_AUXF 1
1696
+ #define A_TYPE block_iq4_nl
1697
+ #define A_TYPE_PACKED16 block_iq4_nl_packed16
1698
+ #endif
1699
+
1700
+ #define QUANT_K_MXFP4 32
1701
+ #define QUANT_R_MXFP4 2
1702
+
1703
+ struct block_mxfp4
1704
+ {
1705
+ uint8_t e;
1706
+ uint8_t qs[QUANT_K_MXFP4/2];
1707
+ };
1708
+
1709
+ #if defined(DATA_A_MXFP4)
1710
+ #define QUANT_K QUANT_K_MXFP4
1711
+ #define QUANT_R QUANT_R_MXFP4
1712
+ #define QUANT_AUXF 1
1713
+ #define A_TYPE block_mxfp4
1714
+ #endif
1715
+
1716
+ #define QUANT_K_NVFP4 64
1717
+ #define QUANT_R_NVFP4 1
1718
+
1719
+ struct block_nvfp4
1720
+ {
1721
+ uint8_t d[QUANT_K_NVFP4 / 16];
1722
+ uint8_t qs[QUANT_K_NVFP4 / 2];
1723
+ };
1724
+
1725
+ #if defined(DATA_A_NVFP4)
1726
+ #define QUANT_K QUANT_K_NVFP4
1727
+ #define QUANT_R QUANT_R_NVFP4
1728
+ #define QUANT_AUXF 1
1729
+ #define A_TYPE block_nvfp4
1730
+ #endif
1731
+
1732
+ #if defined(DATA_A_IQ4_NL) || defined(DATA_A_IQ4_XS)
1733
+ const int8_t kvalues_iq4nl_const[16] = {
1734
+ int8_t(-127), int8_t(-104), int8_t(-83), int8_t(-65), int8_t(-49), int8_t(-35), int8_t(-22), int8_t(-10),
1735
+ int8_t(1), int8_t(13), int8_t(25), int8_t(38), int8_t(53), int8_t(69), int8_t(89), int8_t(113)
1736
+ };
1737
+
1738
+ shared FLOAT_TYPE kvalues_iq4nl[16];
1739
+
1740
+ #define NEEDS_INIT_IQ_SHMEM
1741
+ void init_iq_shmem(uvec3 wgsize)
1742
+ {
1743
+ // copy the table into shared memory and sync
1744
+ for (uint i = gl_LocalInvocationIndex.x; i < kvalues_iq4nl.length(); i += wgsize.x) {
1745
+ kvalues_iq4nl[i] = FLOAT_TYPE(kvalues_iq4nl_const[i]);
1746
+ }
1747
+ barrier();
1748
+ }
1749
+ #endif
1750
+
1751
+ #if defined(DATA_A_MXFP4) || defined(DATA_A_NVFP4)
1752
+ const int8_t kvalues_mxfp4_const[16] = {
1753
+ int8_t(0), int8_t(1), int8_t(2), int8_t(3), int8_t(4), int8_t(6), int8_t(8), int8_t(12),
1754
+ int8_t(0), int8_t(-1), int8_t(-2), int8_t(-3), int8_t(-4), int8_t(-6), int8_t(-8), int8_t(-12),
1755
+ };
1756
+
1757
+ shared int8_t kvalues_mxfp4[16];
1758
+
1759
+ #if defined(DATA_A_NVFP4)
1760
+ // UE4M3 scale in NVFP4 blocks use only 7 bits; sign (bit 7) is always zero.
1761
+ shared float ue4m3_fp32_lut[128];
1762
+
1763
+ float ue4m3_to_fp32_build(uint u) {
1764
+ if (u == 0u || u == 127u) {
1765
+ return 0.0;
1766
+ }
1767
+ const uint exp = (u >> 3) & 15u;
1768
+ const uint man = u & 7u;
1769
+ if (exp == 0u) {
1770
+ return float(man) * (1.0 / 512.0);
1771
+ }
1772
+ const uint bits = (exp + 120u) << 23 | (man << 20);
1773
+ return uintBitsToFloat(bits);
1774
+ }
1775
+ #endif
1776
+
1777
+ #define NEEDS_INIT_IQ_SHMEM
1778
+ void init_iq_shmem(uvec3 wgsize)
1779
+ {
1780
+ // copy the table into shared memory and sync
1781
+ for (uint i = gl_LocalInvocationIndex.x; i < kvalues_mxfp4.length(); i += wgsize.x) {
1782
+ kvalues_mxfp4[i] = kvalues_mxfp4_const[i];
1783
+ }
1784
+ #if defined(DATA_A_NVFP4)
1785
+ for (uint i = gl_LocalInvocationIndex.x; i < 128u; i += wgsize.x) {
1786
+ ue4m3_fp32_lut[i] = ue4m3_to_fp32_build(i);
1787
+ }
1788
+ #endif
1789
+ barrier();
1790
+ }
1791
+ #endif
1792
+
1793
+ // returns the bfloat value in the low 16b.
1794
+ // See ggml_compute_fp32_to_bf16
1795
+ uint32_t fp32_to_bf16(float f)
1796
+ {
1797
+ uint32_t u = floatBitsToUint(f);
1798
+ u = (u + (0x7fff + ((u >> 16) & 1))) >> 16;
1799
+ return u;
1800
+ }
1801
+
1802
+ float bf16_to_fp32(uint32_t u)
1803
+ {
1804
+ return uintBitsToFloat(u << 16);
1805
+ }
1806
+
1807
+ vec4 bf16_to_fp32(uvec4 u)
1808
+ {
1809
+ return vec4(bf16_to_fp32(u.x), bf16_to_fp32(u.y), bf16_to_fp32(u.z), bf16_to_fp32(u.w));
1810
+ }
1811
+
1812
+ float e8m0_to_fp32(uint8_t x) {
1813
+ uint32_t bits;
1814
+
1815
+ if (x == 0) {
1816
+ bits = 0x00400000;
1817
+ } else {
1818
+ bits = x;
1819
+ bits = bits << 23;
1820
+ }
1821
+
1822
+ return uintBitsToFloat(bits);
1823
+ }
1824
+
1825
+ #if defined(DATA_A_NVFP4)
1826
+ float ue4m3_to_fp32(uint8_t x) {
1827
+ return ue4m3_fp32_lut[uint(x)];
1828
+ }
1829
+ #endif
1830
+
1831
+ #if BDA
1832
+
1833
+ #extension GL_EXT_buffer_reference : enable
1834
+ #extension GL_EXT_shader_explicit_arithmetic_types_int64 : enable
1835
+
1836
+ #define BDA_STORAGE_T uint64_t
1837
+ #define BDA_OFFSET_T uint64_t
1838
+
1839
+ #else
1840
+
1841
+ #define BDA_STORAGE_T uvec2
1842
+ #define BDA_OFFSET_T uint
1843
+
1844
+ #endif
1845
+
1846
+ #endif // !defined(GGML_TYPES_COMP)