whispercpp 1.3.3 → 1.3.4
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.
- checksums.yaml +4 -4
- data/ext/ruby_whisper_params.c +55 -25
- data/ext/sources/CMakeLists.txt +1 -1
- data/ext/sources/bindings/javascript/package.json +1 -1
- data/ext/sources/build-xcframework.sh +24 -0
- data/ext/sources/examples/CMakeLists.txt +1 -0
- data/ext/sources/examples/addon.node/addon.cpp +19 -19
- data/ext/sources/examples/addon.node/index.js +7 -5
- data/ext/sources/examples/bench/bench.cpp +26 -16
- data/ext/sources/examples/bench.wasm/index-tmpl.html +10 -9
- data/ext/sources/examples/cli/cli.cpp +4 -2
- data/ext/sources/examples/command/command.cpp +26 -24
- data/ext/sources/examples/command.wasm/index-tmpl.html +5 -4
- data/ext/sources/examples/common-ggml.cpp +2 -0
- data/ext/sources/examples/lsp/lsp.cpp +19 -17
- data/ext/sources/examples/server/server.cpp +24 -13
- data/ext/sources/examples/server.py +6 -1
- data/ext/sources/examples/stream/stream.cpp +4 -2
- data/ext/sources/examples/stream.wasm/emscripten.cpp +6 -6
- data/ext/sources/examples/stream.wasm/index-tmpl.html +82 -5
- data/ext/sources/examples/talk-llama/CMakeLists.txt +2 -2
- data/ext/sources/examples/talk-llama/llama-adapter.cpp +101 -4
- data/ext/sources/examples/talk-llama/llama-adapter.h +6 -0
- data/ext/sources/examples/talk-llama/llama-arch.cpp +588 -15
- data/ext/sources/examples/talk-llama/llama-arch.h +58 -1
- data/ext/sources/examples/talk-llama/llama-batch.cpp +103 -71
- data/ext/sources/examples/talk-llama/llama-batch.h +31 -18
- data/ext/sources/examples/talk-llama/llama-chat.cpp +120 -5
- data/ext/sources/examples/talk-llama/llama-chat.h +7 -0
- data/ext/sources/examples/talk-llama/llama-context.cpp +460 -357
- data/ext/sources/examples/talk-llama/llama-context.h +44 -29
- data/ext/sources/examples/talk-llama/llama-cparams.h +4 -4
- data/ext/sources/examples/talk-llama/llama-graph.cpp +543 -271
- data/ext/sources/examples/talk-llama/llama-graph.h +278 -168
- data/ext/sources/examples/talk-llama/llama-hparams.cpp +118 -4
- data/ext/sources/examples/talk-llama/llama-hparams.h +61 -15
- data/ext/sources/examples/talk-llama/llama-impl.h +2 -0
- data/ext/sources/examples/talk-llama/llama-kv-cache-iswa.cpp +326 -0
- data/ext/sources/examples/talk-llama/{llama-kv-cache-unified-iswa.h → llama-kv-cache-iswa.h} +38 -29
- data/ext/sources/examples/talk-llama/llama-kv-cache.cpp +2020 -0
- data/ext/sources/examples/talk-llama/llama-kv-cache.h +358 -27
- data/ext/sources/examples/talk-llama/llama-kv-cells.h +80 -28
- data/ext/sources/examples/talk-llama/llama-memory-hybrid.cpp +56 -36
- data/ext/sources/examples/talk-llama/llama-memory-hybrid.h +30 -29
- data/ext/sources/examples/talk-llama/llama-memory-recurrent.cpp +48 -19
- data/ext/sources/examples/talk-llama/llama-memory-recurrent.h +13 -14
- data/ext/sources/examples/talk-llama/llama-memory.h +16 -10
- data/ext/sources/examples/talk-llama/llama-model-loader.cpp +2 -0
- data/ext/sources/examples/talk-llama/llama-model-loader.h +3 -2
- data/ext/sources/examples/talk-llama/llama-model.cpp +7165 -2336
- data/ext/sources/examples/talk-llama/llama-model.h +60 -9
- data/ext/sources/examples/talk-llama/llama-quant.cpp +48 -10
- data/ext/sources/examples/talk-llama/llama-sampling.cpp +226 -126
- data/ext/sources/examples/talk-llama/llama-vocab.cpp +440 -13
- data/ext/sources/examples/talk-llama/llama-vocab.h +45 -0
- data/ext/sources/examples/talk-llama/llama.cpp +65 -10
- data/ext/sources/examples/talk-llama/llama.h +95 -177
- data/ext/sources/examples/talk-llama/talk-llama.cpp +9 -6
- data/ext/sources/examples/talk-llama/unicode.cpp +207 -0
- data/ext/sources/examples/talk-llama/unicode.h +45 -0
- data/ext/sources/examples/wchess/wchess.cmd/wchess.cmd.cpp +4 -2
- data/ext/sources/examples/whisper.wasm/index-tmpl.html +17 -16
- data/ext/sources/ggml/CMakeLists.txt +59 -31
- data/ext/sources/ggml/cmake/ggml-config.cmake.in +132 -93
- data/ext/sources/ggml/include/ggml-backend.h +17 -1
- data/ext/sources/ggml/include/ggml-cpu.h +1 -1
- data/ext/sources/ggml/include/ggml-metal.h +1 -6
- data/ext/sources/ggml/include/ggml-opt.h +25 -6
- data/ext/sources/ggml/include/ggml-webgpu.h +19 -0
- data/ext/sources/ggml/include/ggml-zdnn.h +17 -0
- data/ext/sources/ggml/include/ggml.h +221 -16
- data/ext/sources/ggml/src/CMakeLists.txt +17 -2
- data/ext/sources/ggml/src/ggml-alloc.c +265 -141
- data/ext/sources/ggml/src/ggml-backend-impl.h +4 -1
- data/ext/sources/ggml/src/ggml-backend-reg.cpp +30 -13
- data/ext/sources/ggml/src/ggml-backend.cpp +221 -38
- data/ext/sources/ggml/src/ggml-blas/CMakeLists.txt +1 -1
- data/ext/sources/ggml/src/ggml-blas/ggml-blas.cpp +5 -4
- data/ext/sources/ggml/src/ggml-cann/CMakeLists.txt +14 -0
- data/ext/sources/ggml/src/ggml-cann/acl_tensor.cpp +3 -1
- data/ext/sources/ggml/src/ggml-cann/aclnn_ops.cpp +903 -717
- data/ext/sources/ggml/src/ggml-cann/aclnn_ops.h +143 -25
- data/ext/sources/ggml/src/ggml-cann/common.h +143 -1
- data/ext/sources/ggml/src/ggml-cann/ggml-cann.cpp +488 -69
- data/ext/sources/ggml/src/ggml-common.h +17 -0
- data/ext/sources/ggml/src/ggml-cpu/CMakeLists.txt +40 -18
- data/ext/sources/ggml/src/ggml-cpu/amx/amx.cpp +4 -2
- data/ext/sources/ggml/src/ggml-cpu/arch/arm/quants.c +132 -596
- data/ext/sources/ggml/src/ggml-cpu/arch/arm/repack.cpp +14 -286
- data/ext/sources/ggml/src/ggml-cpu/arch/loongarch/quants.c +103 -582
- data/ext/sources/ggml/src/ggml-cpu/arch/powerpc/quants.c +162 -589
- data/ext/sources/ggml/src/ggml-cpu/arch/riscv/quants.c +265 -437
- data/ext/sources/ggml/src/ggml-cpu/arch/riscv/repack.cpp +3 -58
- data/ext/sources/ggml/src/ggml-cpu/arch/s390/quants.c +521 -353
- data/ext/sources/ggml/src/ggml-cpu/arch/wasm/quants.c +54 -314
- data/ext/sources/ggml/src/ggml-cpu/arch/x86/quants.c +184 -675
- data/ext/sources/ggml/src/ggml-cpu/arch/x86/repack.cpp +4679 -1657
- data/ext/sources/ggml/src/ggml-cpu/arch-fallback.h +32 -2
- data/ext/sources/ggml/src/ggml-cpu/common.h +14 -0
- data/ext/sources/ggml/src/ggml-cpu/ggml-cpu-impl.h +13 -6
- data/ext/sources/ggml/src/ggml-cpu/ggml-cpu.c +70 -42
- data/ext/sources/ggml/src/ggml-cpu/ggml-cpu.cpp +35 -28
- data/ext/sources/ggml/src/ggml-cpu/kleidiai/kernels.cpp +152 -18
- data/ext/sources/ggml/src/ggml-cpu/kleidiai/kernels.h +7 -1
- data/ext/sources/ggml/src/ggml-cpu/kleidiai/kleidiai.cpp +227 -97
- data/ext/sources/ggml/src/ggml-cpu/llamafile/sgemm.cpp +474 -1116
- data/ext/sources/ggml/src/ggml-cpu/ops.cpp +1587 -1177
- data/ext/sources/ggml/src/ggml-cpu/ops.h +5 -8
- data/ext/sources/ggml/src/ggml-cpu/quants.c +35 -0
- data/ext/sources/ggml/src/ggml-cpu/quants.h +8 -0
- data/ext/sources/ggml/src/ggml-cpu/repack.cpp +458 -47
- data/ext/sources/ggml/src/ggml-cpu/repack.h +22 -0
- data/ext/sources/ggml/src/ggml-cpu/simd-mappings.h +89 -60
- data/ext/sources/ggml/src/ggml-cpu/spacemit/ime.cpp +1024 -0
- data/ext/sources/ggml/src/ggml-cpu/spacemit/ime.h +13 -0
- data/ext/sources/ggml/src/ggml-cpu/spacemit/ime1_kernels.cpp +3196 -0
- data/ext/sources/ggml/src/ggml-cpu/spacemit/ime_kernels.h +26 -0
- data/ext/sources/ggml/src/ggml-cpu/traits.cpp +2 -2
- data/ext/sources/ggml/src/ggml-cpu/traits.h +1 -1
- data/ext/sources/ggml/src/ggml-cpu/vec.cpp +170 -26
- data/ext/sources/ggml/src/ggml-cpu/vec.h +506 -63
- data/ext/sources/ggml/src/ggml-cuda/CMakeLists.txt +20 -16
- data/ext/sources/ggml/src/ggml-cuda/add-id.cu +58 -0
- data/ext/sources/ggml/src/ggml-cuda/add-id.cuh +3 -0
- data/ext/sources/ggml/src/ggml-cuda/binbcast.cu +330 -191
- data/ext/sources/ggml/src/ggml-cuda/binbcast.cuh +2 -0
- data/ext/sources/ggml/src/ggml-cuda/common.cuh +250 -63
- data/ext/sources/ggml/src/ggml-cuda/conv-transpose-1d.cu +1 -4
- data/ext/sources/ggml/src/ggml-cuda/conv2d.cu +166 -0
- data/ext/sources/ggml/src/ggml-cuda/conv2d.cuh +5 -0
- data/ext/sources/ggml/src/ggml-cuda/convert.cu +95 -22
- data/ext/sources/ggml/src/ggml-cuda/convert.cuh +15 -0
- data/ext/sources/ggml/src/ggml-cuda/cpy-utils.cuh +217 -0
- data/ext/sources/ggml/src/ggml-cuda/cpy.cu +64 -307
- data/ext/sources/ggml/src/ggml-cuda/cross-entropy-loss.cu +2 -14
- data/ext/sources/ggml/src/ggml-cuda/dequantize.cuh +14 -40
- data/ext/sources/ggml/src/ggml-cuda/fattn-common.cuh +498 -367
- data/ext/sources/ggml/src/ggml-cuda/fattn-mma-f16.cuh +137 -91
- data/ext/sources/ggml/src/ggml-cuda/fattn-tile.cu +755 -0
- data/ext/sources/ggml/src/ggml-cuda/fattn-tile.cuh +3 -0
- data/ext/sources/ggml/src/ggml-cuda/fattn-vec.cuh +593 -0
- data/ext/sources/ggml/src/ggml-cuda/fattn-wmma-f16.cu +86 -50
- data/ext/sources/ggml/src/ggml-cuda/fattn.cu +185 -198
- data/ext/sources/ggml/src/ggml-cuda/fattn.cuh +2 -0
- data/ext/sources/ggml/src/ggml-cuda/getrows.cu +50 -39
- data/ext/sources/ggml/src/ggml-cuda/ggml-cuda.cu +379 -107
- data/ext/sources/ggml/src/ggml-cuda/im2col.cu +196 -35
- data/ext/sources/ggml/src/ggml-cuda/im2col.cuh +1 -0
- data/ext/sources/ggml/src/ggml-cuda/mean.cu +56 -2
- data/ext/sources/ggml/src/ggml-cuda/mma.cuh +198 -45
- data/ext/sources/ggml/src/ggml-cuda/mmf.cu +123 -0
- data/ext/sources/ggml/src/ggml-cuda/mmf.cuh +496 -0
- data/ext/sources/ggml/src/ggml-cuda/mmq.cu +206 -57
- data/ext/sources/ggml/src/ggml-cuda/mmq.cuh +1262 -721
- data/ext/sources/ggml/src/ggml-cuda/{mmv.cu → mmvf.cu} +53 -53
- data/ext/sources/ggml/src/ggml-cuda/{mmv.cuh → mmvf.cuh} +3 -3
- data/ext/sources/ggml/src/ggml-cuda/mmvq.cu +64 -73
- data/ext/sources/ggml/src/ggml-cuda/norm.cu +284 -12
- data/ext/sources/ggml/src/ggml-cuda/norm.cuh +7 -0
- data/ext/sources/ggml/src/ggml-cuda/opt-step-sgd.cu +49 -0
- data/ext/sources/ggml/src/ggml-cuda/opt-step-sgd.cuh +5 -0
- data/ext/sources/ggml/src/ggml-cuda/pad.cu +46 -23
- data/ext/sources/ggml/src/ggml-cuda/pad_reflect_1d.cu +91 -0
- data/ext/sources/ggml/src/ggml-cuda/pad_reflect_1d.cuh +5 -0
- data/ext/sources/ggml/src/ggml-cuda/quantize.cu +12 -10
- data/ext/sources/ggml/src/ggml-cuda/reduce_rows.cuh +53 -0
- data/ext/sources/ggml/src/ggml-cuda/roll.cu +67 -0
- data/ext/sources/ggml/src/ggml-cuda/roll.cuh +5 -0
- data/ext/sources/ggml/src/ggml-cuda/rope.cu +21 -27
- data/ext/sources/ggml/src/ggml-cuda/scale.cu +14 -11
- data/ext/sources/ggml/src/ggml-cuda/set-rows.cu +276 -0
- data/ext/sources/ggml/src/ggml-cuda/set-rows.cuh +7 -0
- data/ext/sources/ggml/src/ggml-cuda/softcap.cu +34 -0
- data/ext/sources/ggml/src/ggml-cuda/softcap.cuh +5 -0
- data/ext/sources/ggml/src/ggml-cuda/softmax.cu +126 -59
- data/ext/sources/ggml/src/ggml-cuda/ssm-conv.cu +10 -2
- data/ext/sources/ggml/src/ggml-cuda/ssm-scan.cu +322 -100
- data/ext/sources/ggml/src/ggml-cuda/sum.cu +6 -10
- data/ext/sources/ggml/src/ggml-cuda/sumrows.cu +21 -4
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-f16-f16.cu +7 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-f16-q4_0.cu +7 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-f16-q4_1.cu +7 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-f16-q5_0.cu +7 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-f16-q5_1.cu +7 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-f16-q8_0.cu +7 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q4_0-f16.cu +7 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q4_0-q4_0.cu +7 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q4_0-q4_1.cu +7 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q4_0-q5_0.cu +7 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q4_0-q5_1.cu +7 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q4_0-q8_0.cu +7 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q4_1-f16.cu +7 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q4_1-q4_0.cu +7 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q4_1-q4_1.cu +7 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q4_1-q5_0.cu +7 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q4_1-q5_1.cu +7 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q4_1-q8_0.cu +7 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q5_0-f16.cu +7 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q5_0-q4_0.cu +7 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q5_0-q4_1.cu +7 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q5_0-q5_0.cu +7 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q5_0-q5_1.cu +7 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q5_0-q8_0.cu +7 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q5_1-f16.cu +7 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q5_1-q4_0.cu +7 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q5_1-q4_1.cu +7 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q5_1-q5_0.cu +7 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q5_1-q5_1.cu +7 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q5_1-q8_0.cu +7 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q8_0-f16.cu +7 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q8_0-q4_0.cu +7 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q8_0-q4_1.cu +7 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q8_0-q5_0.cu +7 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q8_0-q5_1.cu +7 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q8_0-q8_0.cu +7 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/generate_cu_files.py +21 -18
- data/ext/sources/ggml/src/ggml-cuda/template-instances/mmf-instance-ncols_1.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/mmf-instance-ncols_10.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/mmf-instance-ncols_11.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/mmf-instance-ncols_12.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/mmf-instance-ncols_13.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/mmf-instance-ncols_14.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/mmf-instance-ncols_15.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/mmf-instance-ncols_16.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/mmf-instance-ncols_2.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/mmf-instance-ncols_3.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/mmf-instance-ncols_4.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/mmf-instance-ncols_5.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/mmf-instance-ncols_6.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/mmf-instance-ncols_7.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/mmf-instance-ncols_8.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/mmf-instance-ncols_9.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/mmq-instance-mxfp4.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/topk-moe.cu +259 -0
- data/ext/sources/ggml/src/ggml-cuda/topk-moe.cuh +14 -0
- data/ext/sources/ggml/src/ggml-cuda/tsembd.cu +3 -3
- data/ext/sources/ggml/src/ggml-cuda/unary.cu +90 -0
- data/ext/sources/ggml/src/ggml-cuda/unary.cuh +8 -0
- data/ext/sources/ggml/src/ggml-cuda/upscale.cu +92 -6
- data/ext/sources/ggml/src/ggml-cuda/vecdotq.cuh +110 -22
- data/ext/sources/ggml/src/ggml-cuda/vendors/cuda.h +4 -0
- data/ext/sources/ggml/src/ggml-cuda/vendors/hip.h +58 -36
- data/ext/sources/ggml/src/ggml-cuda/vendors/musa.h +4 -3
- data/ext/sources/ggml/src/ggml-hip/CMakeLists.txt +10 -2
- data/ext/sources/ggml/src/ggml-impl.h +119 -9
- data/ext/sources/ggml/src/ggml-metal/CMakeLists.txt +10 -7
- data/ext/sources/ggml/src/ggml-metal/ggml-metal-common.cpp +446 -0
- data/ext/sources/ggml/src/ggml-metal/ggml-metal-common.h +52 -0
- data/ext/sources/ggml/src/ggml-metal/ggml-metal-context.h +33 -0
- data/ext/sources/ggml/src/ggml-metal/ggml-metal-context.m +600 -0
- data/ext/sources/ggml/src/ggml-metal/ggml-metal-device.cpp +1376 -0
- data/ext/sources/ggml/src/ggml-metal/ggml-metal-device.h +226 -0
- data/ext/sources/ggml/src/ggml-metal/ggml-metal-device.m +1308 -0
- data/ext/sources/ggml/src/ggml-metal/ggml-metal-impl.h +136 -63
- data/ext/sources/ggml/src/ggml-metal/ggml-metal-ops.cpp +3158 -0
- data/ext/sources/ggml/src/ggml-metal/ggml-metal-ops.h +82 -0
- data/ext/sources/ggml/src/ggml-metal/ggml-metal.cpp +718 -0
- data/ext/sources/ggml/src/ggml-metal/ggml-metal.metal +2854 -1503
- data/ext/sources/ggml/src/ggml-musa/CMakeLists.txt +18 -8
- data/ext/sources/ggml/src/ggml-opencl/CMakeLists.txt +18 -0
- data/ext/sources/ggml/src/ggml-opencl/ggml-opencl.cpp +2510 -242
- data/ext/sources/ggml/src/ggml-opencl/kernels/add.cl +107 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/add_id.cl +42 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/conv2d.cl +185 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/conv2d_f16_f32.cl +176 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/cvt.cl +84 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/div.cl +66 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/flash_attn_f16.cl +370 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/flash_attn_f32.cl +370 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/flash_attn_f32_f16.cl +373 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/gelu.cl +27 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/glu.cl +177 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/group_norm.cl +49 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/im2col_f16.cl +1 -1
- data/ext/sources/ggml/src/ggml-opencl/kernels/im2col_f32.cl +1 -1
- data/ext/sources/ggml/src/ggml-opencl/kernels/mul.cl +73 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/mul_mat_f16_f32.cl +130 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/mul_mm_f16_f32_l4_lm.cl +132 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/mul_mm_f32_f32_l4_lm.cl +133 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/mul_mv_id_mxfp4_f32.cl +189 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/mul_mv_id_mxfp4_f32_flat.cl +176 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/mul_mv_id_q8_0_f32.cl +140 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/mul_mv_id_q8_0_f32_flat.cl +222 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/mul_mv_mxfp4_f32.cl +144 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/mul_mv_mxfp4_f32_flat.cl +167 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/mul_mv_q8_0_f32.cl +125 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/mul_mv_q8_0_f32_flat.cl +202 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/norm.cl +80 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/rms_norm.cl +79 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/scale.cl +3 -2
- data/ext/sources/ggml/src/ggml-opencl/kernels/set_rows.cl +189 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/softmax_4_f16.cl +34 -13
- data/ext/sources/ggml/src/ggml-opencl/kernels/softmax_4_f32.cl +34 -13
- data/ext/sources/ggml/src/ggml-opencl/kernels/softmax_f16.cl +34 -13
- data/ext/sources/ggml/src/ggml-opencl/kernels/softmax_f32.cl +34 -13
- data/ext/sources/ggml/src/ggml-opencl/kernels/sub.cl +66 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/transpose.cl +20 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/tsembd.cl +2 -2
- data/ext/sources/ggml/src/ggml-opencl/kernels/upscale.cl +2 -3
- data/ext/sources/ggml/src/ggml-opt.cpp +97 -41
- data/ext/sources/ggml/src/ggml-quants.c +111 -16
- data/ext/sources/ggml/src/ggml-quants.h +6 -0
- data/ext/sources/ggml/src/ggml-rpc/ggml-rpc.cpp +67 -47
- data/ext/sources/ggml/src/ggml-sycl/backend.hpp +2 -0
- data/ext/sources/ggml/src/ggml-sycl/binbcast.cpp +15 -5
- data/ext/sources/ggml/src/ggml-sycl/binbcast.hpp +6 -0
- data/ext/sources/ggml/src/ggml-sycl/concat.cpp +25 -16
- data/ext/sources/ggml/src/ggml-sycl/conv.cpp +10 -4
- data/ext/sources/ggml/src/ggml-sycl/convert.cpp +166 -99
- data/ext/sources/ggml/src/ggml-sycl/cpy.cpp +72 -306
- data/ext/sources/ggml/src/ggml-sycl/cpy.hpp +213 -1
- data/ext/sources/ggml/src/ggml-sycl/dmmv.cpp +67 -49
- data/ext/sources/ggml/src/ggml-sycl/dpct/helper.hpp +1 -31
- data/ext/sources/ggml/src/ggml-sycl/element_wise.cpp +79 -29
- data/ext/sources/ggml/src/ggml-sycl/element_wise.hpp +2 -0
- data/ext/sources/ggml/src/ggml-sycl/gemm.hpp +14 -26
- data/ext/sources/ggml/src/ggml-sycl/getrows.cpp +9 -6
- data/ext/sources/ggml/src/ggml-sycl/ggml-sycl.cpp +328 -323
- data/ext/sources/ggml/src/ggml-sycl/gla.cpp +2 -2
- data/ext/sources/ggml/src/ggml-sycl/im2col.cpp +2 -2
- data/ext/sources/ggml/src/ggml-sycl/mmq.cpp +80 -60
- data/ext/sources/ggml/src/ggml-sycl/mmvq.cpp +201 -132
- data/ext/sources/ggml/src/ggml-sycl/norm.cpp +74 -55
- data/ext/sources/ggml/src/ggml-sycl/quantize.hpp +133 -0
- data/ext/sources/ggml/src/ggml-sycl/quants.hpp +8 -9
- data/ext/sources/ggml/src/ggml-sycl/rope.cpp +35 -42
- data/ext/sources/ggml/src/ggml-sycl/set_rows.cpp +234 -0
- data/ext/sources/ggml/src/ggml-sycl/set_rows.hpp +8 -0
- data/ext/sources/ggml/src/ggml-sycl/softmax.cpp +3 -3
- data/ext/sources/ggml/src/ggml-sycl/tsembd.cpp +12 -6
- data/ext/sources/ggml/src/ggml-sycl/vecdotq.hpp +2 -6
- data/ext/sources/ggml/src/ggml-sycl/wkv.cpp +16 -12
- data/ext/sources/ggml/src/ggml-vulkan/ggml-vulkan.cpp +3492 -883
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/add.comp +41 -1
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/add_id.comp +42 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/argmax.comp +13 -4
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/argsort.comp +39 -29
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/conv2d_mm.comp +349 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/copy_from_quant.comp +2 -2
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/copy_to_quant.comp +66 -12
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/dequant_funcs.comp +154 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/dequant_funcs_cm2.comp +21 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/dequant_iq2_s.comp +1 -1
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/dequant_iq2_xxs.comp +2 -1
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/dequant_iq3_s.comp +6 -5
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/dequant_iq3_xxs.comp +4 -2
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/dequant_mxfp4.comp +32 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/dequant_q2_k.comp +1 -1
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/dequant_q3_k.comp +1 -1
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/dequant_q4_k.comp +1 -1
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/dequant_q5_k.comp +1 -1
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/dequant_q6_k.comp +1 -1
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/exp.comp +21 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/flash_attn.comp +69 -24
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/flash_attn_base.comp +60 -20
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/flash_attn_cm1.comp +98 -42
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/flash_attn_cm2.comp +64 -27
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/flash_attn_split_k_reduce.comp +74 -13
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/geglu_erf.comp +27 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/geglu_quick.comp +11 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/gelu_erf.comp +39 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/generic_binary_head.comp +4 -17
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/get_rows.comp +19 -10
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/get_rows_quant.comp +25 -15
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/glu_head.comp +4 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/hardsigmoid.comp +22 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/hardswish.comp +22 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/im2col.comp +18 -14
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/im2col_3d.comp +126 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_base.comp +65 -1
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_nc.comp +11 -7
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vecq.comp +140 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mm.comp +144 -531
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mm_cm2.comp +206 -38
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mm_funcs.comp +556 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mmq.comp +12 -5
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mmq_funcs.comp +15 -9
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/multi_add.comp +111 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/opt_step_sgd.comp +22 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/pad.comp +24 -3
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/quantize_q8_1.comp +53 -3
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/rms_norm.comp +55 -11
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/rms_norm_partials.comp +65 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/roll.comp +46 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/rope_head.comp +1 -4
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/rope_multi.comp +7 -9
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/rope_neox.comp +7 -9
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/rope_norm.comp +7 -9
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/rte.comp +5 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/scale.comp +1 -1
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/soft_max.comp +29 -7
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/soft_max_back.comp +4 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/sqrt.comp +17 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/sum_rows.comp +38 -5
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/swiglu_oai.comp +14 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/timestep_embedding.comp +4 -3
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/types.comp +101 -9
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/upscale.comp +69 -5
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/utils.comp +25 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/vulkan-shaders-gen.cpp +335 -77
- data/ext/sources/ggml/src/ggml-webgpu/CMakeLists.txt +54 -0
- data/ext/sources/ggml/src/ggml-webgpu/ggml-webgpu.cpp +1558 -0
- data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/add.tmpl.wgsl +44 -0
- data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/add_in_place.tmpl.wgsl +41 -0
- data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/binary_head.tmpl +45 -0
- data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/common_decls.tmpl +930 -0
- data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/cpy.wgsl +60 -0
- data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/embed_wgsl.py +124 -0
- data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/get_rows.tmpl.wgsl +874 -0
- data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/memset.wgsl +40 -0
- data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/mul.tmpl.wgsl +44 -0
- data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/mul_in_place.tmpl.wgsl +41 -0
- data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/mul_mat.tmpl.wgsl +907 -0
- data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/rms_norm.wgsl +57 -0
- data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/rms_norm_in_place.wgsl +48 -0
- data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/set_rows.wgsl +81 -0
- data/ext/sources/ggml/src/ggml-zdnn/CMakeLists.txt +36 -0
- data/ext/sources/ggml/src/ggml-zdnn/common.hpp +59 -0
- data/ext/sources/ggml/src/ggml-zdnn/ggml-zdnn.cpp +628 -0
- data/ext/sources/ggml/src/ggml-zdnn/mmf.cpp +80 -0
- data/ext/sources/ggml/src/ggml-zdnn/mmf.hpp +12 -0
- data/ext/sources/ggml/src/ggml-zdnn/utils.cpp +79 -0
- data/ext/sources/ggml/src/ggml-zdnn/utils.hpp +19 -0
- data/ext/sources/ggml/src/ggml.c +478 -98
- data/ext/sources/ggml/src/gguf.cpp +8 -1
- data/ext/sources/src/whisper.cpp +23 -46
- data/ext/sources/tests/CMakeLists.txt +8 -1
- data/ext/sources/tests/test-vad-full.cpp +3 -3
- data/ext/sources/tests/test-vad.cpp +2 -2
- data/lib/whisper/model/uri.rb +1 -1
- data/sig/whisper.rbs +7 -0
- data/test/test_params.rb +8 -0
- data/test/test_whisper.rb +1 -1
- data/whispercpp.gemspec +1 -1
- metadata +164 -157
- data/ext/sources/examples/talk-llama/llama-kv-cache-unified-iswa.cpp +0 -279
- data/ext/sources/examples/talk-llama/llama-kv-cache-unified.cpp +0 -1841
- data/ext/sources/examples/talk-llama/llama-kv-cache-unified.h +0 -303
- data/ext/sources/ggml/include/ggml-kompute.h +0 -50
- data/ext/sources/ggml/src/ggml-amx/CMakeLists.txt +0 -107
- data/ext/sources/ggml/src/ggml-amx/common.h +0 -94
- data/ext/sources/ggml/src/ggml-amx/ggml-amx.cpp +0 -446
- data/ext/sources/ggml/src/ggml-amx/mmq.cpp +0 -2510
- data/ext/sources/ggml/src/ggml-amx/mmq.h +0 -17
- data/ext/sources/ggml/src/ggml-cann/kernels/CMakeLists.txt +0 -30
- data/ext/sources/ggml/src/ggml-cann/kernels/ascendc_kernels.h +0 -19
- data/ext/sources/ggml/src/ggml-cann/kernels/dup.cpp +0 -234
- data/ext/sources/ggml/src/ggml-cann/kernels/get_row_f16.cpp +0 -197
- data/ext/sources/ggml/src/ggml-cann/kernels/get_row_f32.cpp +0 -190
- data/ext/sources/ggml/src/ggml-cann/kernels/get_row_q4_0.cpp +0 -204
- data/ext/sources/ggml/src/ggml-cann/kernels/get_row_q8_0.cpp +0 -191
- data/ext/sources/ggml/src/ggml-cann/kernels/quantize_f16_q8_0.cpp +0 -218
- data/ext/sources/ggml/src/ggml-cann/kernels/quantize_f32_q8_0.cpp +0 -216
- data/ext/sources/ggml/src/ggml-cann/kernels/quantize_float_to_q4_0.cpp +0 -295
- data/ext/sources/ggml/src/ggml-cuda/fattn-tile-f16.cu +0 -357
- data/ext/sources/ggml/src/ggml-cuda/fattn-tile-f16.cuh +0 -3
- data/ext/sources/ggml/src/ggml-cuda/fattn-tile-f32.cu +0 -365
- data/ext/sources/ggml/src/ggml-cuda/fattn-tile-f32.cuh +0 -3
- data/ext/sources/ggml/src/ggml-cuda/fattn-vec-f16.cuh +0 -482
- data/ext/sources/ggml/src/ggml-cuda/fattn-vec-f32.cuh +0 -472
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-f16-f16.cu +0 -5
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-f16-q4_0.cu +0 -5
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-f16-q4_1.cu +0 -5
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-f16-q5_0.cu +0 -5
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-f16-q5_1.cu +0 -5
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-f16-q8_0.cu +0 -5
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q4_0-f16.cu +0 -5
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q4_0-q4_0.cu +0 -5
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q4_0-q4_1.cu +0 -5
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q4_0-q5_0.cu +0 -5
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q4_0-q5_1.cu +0 -5
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q4_0-q8_0.cu +0 -5
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q4_1-f16.cu +0 -5
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q4_1-q4_0.cu +0 -5
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q4_1-q4_1.cu +0 -5
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q4_1-q5_0.cu +0 -5
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q4_1-q5_1.cu +0 -5
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q4_1-q8_0.cu +0 -5
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q5_0-f16.cu +0 -5
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q5_0-q4_0.cu +0 -5
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q5_0-q4_1.cu +0 -5
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q5_0-q5_0.cu +0 -5
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q5_0-q5_1.cu +0 -5
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q5_0-q8_0.cu +0 -5
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q5_1-f16.cu +0 -5
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q5_1-q4_0.cu +0 -5
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q5_1-q4_1.cu +0 -5
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q5_1-q5_0.cu +0 -5
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q5_1-q5_1.cu +0 -5
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q5_1-q8_0.cu +0 -5
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q8_0-f16.cu +0 -5
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q8_0-q4_0.cu +0 -5
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q8_0-q4_1.cu +0 -5
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q8_0-q5_0.cu +0 -5
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q8_0-q5_1.cu +0 -5
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q8_0-q8_0.cu +0 -5
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs256-f16-f16.cu +0 -5
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs64-f16-f16.cu +0 -5
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs64-f16-q4_0.cu +0 -5
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs64-f16-q4_1.cu +0 -5
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs64-f16-q5_0.cu +0 -5
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs64-f16-q5_1.cu +0 -5
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs64-f16-q8_0.cu +0 -5
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-f16-f16.cu +0 -5
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-f16-q4_0.cu +0 -5
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-f16-q4_1.cu +0 -5
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-f16-q5_0.cu +0 -5
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-f16-q5_1.cu +0 -5
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-f16-q8_0.cu +0 -5
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q4_0-f16.cu +0 -5
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q4_0-q4_0.cu +0 -5
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q4_0-q4_1.cu +0 -5
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q4_0-q5_0.cu +0 -5
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q4_0-q5_1.cu +0 -5
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q4_0-q8_0.cu +0 -5
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q4_1-f16.cu +0 -5
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q4_1-q4_0.cu +0 -5
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q4_1-q4_1.cu +0 -5
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q4_1-q5_0.cu +0 -5
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q4_1-q5_1.cu +0 -5
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q4_1-q8_0.cu +0 -5
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q5_0-f16.cu +0 -5
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q5_0-q4_0.cu +0 -5
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q5_0-q4_1.cu +0 -5
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q5_0-q5_0.cu +0 -5
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q5_0-q5_1.cu +0 -5
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q5_0-q8_0.cu +0 -5
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q5_1-f16.cu +0 -5
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q5_1-q4_0.cu +0 -5
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q5_1-q4_1.cu +0 -5
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q5_1-q5_0.cu +0 -5
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q5_1-q5_1.cu +0 -5
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q5_1-q8_0.cu +0 -5
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q8_0-f16.cu +0 -5
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q8_0-q4_0.cu +0 -5
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q8_0-q4_1.cu +0 -5
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q8_0-q5_0.cu +0 -5
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q8_0-q5_1.cu +0 -5
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q8_0-q8_0.cu +0 -5
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs256-f16-f16.cu +0 -5
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs64-f16-f16.cu +0 -5
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs64-f16-q4_0.cu +0 -5
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs64-f16-q4_1.cu +0 -5
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs64-f16-q5_0.cu +0 -5
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs64-f16-q5_1.cu +0 -5
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs64-f16-q8_0.cu +0 -5
- data/ext/sources/ggml/src/ggml-kompute/CMakeLists.txt +0 -166
- data/ext/sources/ggml/src/ggml-kompute/ggml-kompute.cpp +0 -2251
- data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/common.comp +0 -112
- data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_add.comp +0 -58
- data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_addrow.comp +0 -25
- data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_cpy_f16_f16.comp +0 -52
- data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_cpy_f16_f32.comp +0 -52
- data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_cpy_f32_f16.comp +0 -52
- data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_cpy_f32_f32.comp +0 -52
- data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_diagmask.comp +0 -30
- data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_gelu.comp +0 -22
- data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_getrows.comp +0 -17
- data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_getrows_f16.comp +0 -31
- data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_getrows_f32.comp +0 -31
- data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_getrows_q4_0.comp +0 -38
- data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_getrows_q4_1.comp +0 -39
- data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_getrows_q6_k.comp +0 -44
- data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_mul.comp +0 -52
- data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_mul_mat_f16.comp +0 -69
- data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_mul_mat_mat_f32.comp +0 -51
- data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_mul_mat_q4_0.comp +0 -33
- data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_mul_mat_q4_1.comp +0 -35
- data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_mul_mat_q4_k.comp +0 -140
- data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_mul_mat_q6_k.comp +0 -106
- data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_mul_mat_q8_0.comp +0 -73
- data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_mul_mv_q_n.comp +0 -52
- data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_mul_mv_q_n_pre.comp +0 -28
- data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_norm.comp +0 -84
- data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_relu.comp +0 -21
- data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_rmsnorm.comp +0 -53
- data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_rope_neox_f16.comp +0 -52
- data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_rope_neox_f32.comp +0 -52
- data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_rope_norm_f16.comp +0 -52
- data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_rope_norm_f32.comp +0 -52
- data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_scale.comp +0 -19
- data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_scale_8.comp +0 -23
- data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_silu.comp +0 -22
- data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_softmax.comp +0 -72
- data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/rope_common.comp +0 -71
- data/ext/sources/ggml/src/ggml-metal/ggml-metal.m +0 -6280
@@ -544,12 +544,12 @@ static void reorder_mul_mat_vec_q4_0_q8_1_sycl(const void * vx, const void * vy,
|
|
544
544
|
const sycl::range<3> global_size(1, GGML_SYCL_MMV_Y, (block_num_y * WARP_SIZE));
|
545
545
|
const sycl::range<3> workgroup_size(1, GGML_SYCL_MMV_Y, num_subgroups * WARP_SIZE);
|
546
546
|
|
547
|
-
|
548
|
-
|
549
|
-
|
550
|
-
|
551
|
-
|
552
|
-
|
547
|
+
stream->submit([&](sycl::handler & cgh) {
|
548
|
+
cgh.parallel_for(sycl::nd_range<3>(global_size, workgroup_size),
|
549
|
+
[=](sycl::nd_item<3> nd_item) [[sycl::reqd_sub_group_size(WARP_SIZE)]] {
|
550
|
+
mul_mat_vec_q_reorder<reorder_vec_dot_q_sycl<GGML_TYPE_Q4_0>>(vx, vy, dst, ncols, nrows,
|
551
|
+
nd_item);
|
552
|
+
});
|
553
553
|
});
|
554
554
|
}
|
555
555
|
|
@@ -561,12 +561,12 @@ static void mul_mat_vec_q4_0_q8_1_sycl(const void * vx, const void * vy, float *
|
|
561
561
|
const sycl::range<3> block_dims(1, GGML_SYCL_MMV_Y, WARP_SIZE);
|
562
562
|
|
563
563
|
{
|
564
|
-
|
565
|
-
|
566
|
-
|
567
|
-
|
568
|
-
|
569
|
-
|
564
|
+
stream->submit([&](sycl::handler & cgh) {
|
565
|
+
cgh.parallel_for(sycl::nd_range<3>(block_nums * block_dims, block_dims),
|
566
|
+
[=](sycl::nd_item<3> item_ct1) [[sycl::reqd_sub_group_size(WARP_SIZE)]] {
|
567
|
+
mul_mat_vec_q<QK4_0, QI4_0, block_q4_0, VDR_Q4_0_Q8_1_MMVQ, vec_dot_q4_0_q8_1>(
|
568
|
+
vx, vy, dst, ncols, nrows, item_ct1);
|
569
|
+
});
|
570
570
|
});
|
571
571
|
}
|
572
572
|
}
|
@@ -580,12 +580,17 @@ static void mul_mat_vec_q4_1_q8_1_sycl(const void *vx, const void *vy,
|
|
580
580
|
const sycl::range<3> block_nums(1, 1, block_num_y);
|
581
581
|
const sycl::range<3> block_dims(1, GGML_SYCL_MMV_Y, WARP_SIZE);
|
582
582
|
{
|
583
|
-
|
584
|
-
|
585
|
-
|
586
|
-
|
587
|
-
|
588
|
-
|
583
|
+
|
584
|
+
stream->submit([&](sycl::handler &cgh) {
|
585
|
+
|
586
|
+
cgh.parallel_for(
|
587
|
+
sycl::nd_range<3>(block_nums * block_dims, block_dims),
|
588
|
+
[=](sycl::nd_item<3> item_ct1)
|
589
|
+
[[sycl::reqd_sub_group_size(WARP_SIZE)]] {
|
590
|
+
mul_mat_vec_q<QK4_0, QI4_1, block_q4_1,
|
591
|
+
VDR_Q4_1_Q8_1_MMVQ, vec_dot_q4_1_q8_1>(
|
592
|
+
vx, vy, dst, ncols, nrows, item_ct1);
|
593
|
+
});
|
589
594
|
});
|
590
595
|
}
|
591
596
|
}
|
@@ -599,12 +604,17 @@ static void mul_mat_vec_q5_0_q8_1_sycl(const void *vx, const void *vy,
|
|
599
604
|
const sycl::range<3> block_nums(1, 1, block_num_y);
|
600
605
|
const sycl::range<3> block_dims(1, GGML_SYCL_MMV_Y, WARP_SIZE);
|
601
606
|
{
|
602
|
-
|
603
|
-
|
604
|
-
|
605
|
-
|
606
|
-
|
607
|
-
|
607
|
+
|
608
|
+
stream->submit([&](sycl::handler &cgh) {
|
609
|
+
|
610
|
+
cgh.parallel_for(
|
611
|
+
sycl::nd_range<3>(block_nums * block_dims, block_dims),
|
612
|
+
[=](sycl::nd_item<3> item_ct1)
|
613
|
+
[[sycl::reqd_sub_group_size(WARP_SIZE)]] {
|
614
|
+
mul_mat_vec_q<QK5_0, QI5_0, block_q5_0,
|
615
|
+
VDR_Q5_0_Q8_1_MMVQ, vec_dot_q5_0_q8_1>(
|
616
|
+
vx, vy, dst, ncols, nrows, item_ct1);
|
617
|
+
});
|
608
618
|
});
|
609
619
|
}
|
610
620
|
}
|
@@ -618,12 +628,17 @@ static void mul_mat_vec_q5_1_q8_1_sycl(const void *vx, const void *vy,
|
|
618
628
|
const sycl::range<3> block_nums(1, 1, block_num_y);
|
619
629
|
const sycl::range<3> block_dims(1, GGML_SYCL_MMV_Y, WARP_SIZE);
|
620
630
|
{
|
621
|
-
|
622
|
-
|
623
|
-
|
624
|
-
|
625
|
-
|
626
|
-
|
631
|
+
|
632
|
+
stream->submit([&](sycl::handler &cgh) {
|
633
|
+
|
634
|
+
cgh.parallel_for(
|
635
|
+
sycl::nd_range<3>(block_nums * block_dims, block_dims),
|
636
|
+
[=](sycl::nd_item<3> item_ct1)
|
637
|
+
[[sycl::reqd_sub_group_size(WARP_SIZE)]] {
|
638
|
+
mul_mat_vec_q<QK5_1, QI5_1, block_q5_1,
|
639
|
+
VDR_Q5_1_Q8_1_MMVQ, vec_dot_q5_1_q8_1>(
|
640
|
+
vx, vy, dst, ncols, nrows, item_ct1);
|
641
|
+
});
|
627
642
|
});
|
628
643
|
}
|
629
644
|
}
|
@@ -637,12 +652,17 @@ static void mul_mat_vec_q8_0_q8_1_sycl(const void *vx, const void *vy,
|
|
637
652
|
const sycl::range<3> block_nums(1, 1, block_num_y);
|
638
653
|
const sycl::range<3> block_dims(1, GGML_SYCL_MMV_Y, WARP_SIZE);
|
639
654
|
{
|
640
|
-
|
641
|
-
|
642
|
-
|
643
|
-
|
644
|
-
|
645
|
-
|
655
|
+
|
656
|
+
stream->submit([&](sycl::handler &cgh) {
|
657
|
+
|
658
|
+
cgh.parallel_for(
|
659
|
+
sycl::nd_range<3>(block_nums * block_dims, block_dims),
|
660
|
+
[=](sycl::nd_item<3> item_ct1)
|
661
|
+
[[sycl::reqd_sub_group_size(WARP_SIZE)]] {
|
662
|
+
mul_mat_vec_q<QK8_0, QI8_0, block_q8_0,
|
663
|
+
VDR_Q8_0_Q8_1_MMVQ, vec_dot_q8_0_q8_1>(
|
664
|
+
vx, vy, dst, ncols, nrows, item_ct1);
|
665
|
+
});
|
646
666
|
});
|
647
667
|
}
|
648
668
|
}
|
@@ -656,12 +676,17 @@ static void mul_mat_vec_q2_K_q8_1_sycl(const void *vx, const void *vy,
|
|
656
676
|
const sycl::range<3> block_nums(1, 1, block_num_y);
|
657
677
|
const sycl::range<3> block_dims(1, GGML_SYCL_MMV_Y, WARP_SIZE);
|
658
678
|
{
|
659
|
-
|
660
|
-
|
661
|
-
|
662
|
-
|
663
|
-
|
664
|
-
|
679
|
+
|
680
|
+
stream->submit([&](sycl::handler &cgh) {
|
681
|
+
|
682
|
+
cgh.parallel_for(
|
683
|
+
sycl::nd_range<3>(block_nums * block_dims, block_dims),
|
684
|
+
[=](sycl::nd_item<3> item_ct1)
|
685
|
+
[[sycl::reqd_sub_group_size(WARP_SIZE)]] {
|
686
|
+
mul_mat_vec_q<QK_K, QI2_K, block_q2_K,
|
687
|
+
VDR_Q2_K_Q8_1_MMVQ, vec_dot_q2_K_q8_1>(
|
688
|
+
vx, vy, dst, ncols, nrows, item_ct1);
|
689
|
+
});
|
665
690
|
});
|
666
691
|
}
|
667
692
|
}
|
@@ -675,12 +700,17 @@ static void mul_mat_vec_q3_K_q8_1_sycl(const void *vx, const void *vy,
|
|
675
700
|
const sycl::range<3> block_nums(1, 1, block_num_y);
|
676
701
|
const sycl::range<3> block_dims(1, GGML_SYCL_MMV_Y, WARP_SIZE);
|
677
702
|
{
|
678
|
-
|
679
|
-
|
680
|
-
|
681
|
-
|
682
|
-
|
683
|
-
|
703
|
+
|
704
|
+
stream->submit([&](sycl::handler &cgh) {
|
705
|
+
|
706
|
+
cgh.parallel_for(
|
707
|
+
sycl::nd_range<3>(block_nums * block_dims, block_dims),
|
708
|
+
[=](sycl::nd_item<3> item_ct1)
|
709
|
+
[[sycl::reqd_sub_group_size(WARP_SIZE)]] {
|
710
|
+
mul_mat_vec_q<QK_K, QI3_K, block_q3_K,
|
711
|
+
VDR_Q3_K_Q8_1_MMVQ, vec_dot_q3_K_q8_1>(
|
712
|
+
vx, vy, dst, ncols, nrows, item_ct1);
|
713
|
+
});
|
684
714
|
});
|
685
715
|
}
|
686
716
|
}
|
@@ -694,12 +724,17 @@ static void mul_mat_vec_q4_K_q8_1_sycl(const void *vx, const void *vy,
|
|
694
724
|
const sycl::range<3> block_nums(1, 1, block_num_y);
|
695
725
|
const sycl::range<3> block_dims(1, GGML_SYCL_MMV_Y, WARP_SIZE);
|
696
726
|
{
|
697
|
-
|
698
|
-
|
699
|
-
|
700
|
-
|
701
|
-
|
702
|
-
|
727
|
+
|
728
|
+
stream->submit([&](sycl::handler &cgh) {
|
729
|
+
|
730
|
+
cgh.parallel_for(
|
731
|
+
sycl::nd_range<3>(block_nums * block_dims, block_dims),
|
732
|
+
[=](sycl::nd_item<3> item_ct1)
|
733
|
+
[[sycl::reqd_sub_group_size(WARP_SIZE)]] {
|
734
|
+
mul_mat_vec_q<QK_K, QI4_K, block_q4_K,
|
735
|
+
VDR_Q4_K_Q8_1_MMVQ, vec_dot_q4_K_q8_1>(
|
736
|
+
vx, vy, dst, ncols, nrows, item_ct1);
|
737
|
+
});
|
703
738
|
});
|
704
739
|
}
|
705
740
|
}
|
@@ -715,12 +750,12 @@ static void reorder_mul_mat_vec_q4_k_q8_1_sycl(const void * vx, const void * vy,
|
|
715
750
|
const sycl::range<3> global_size(1, GGML_SYCL_MMV_Y, block_num_y * WARP_SIZE);
|
716
751
|
const sycl::range<3> workgroup_size(1, GGML_SYCL_MMV_Y, num_subgroups * WARP_SIZE);
|
717
752
|
|
718
|
-
|
719
|
-
|
720
|
-
|
721
|
-
|
722
|
-
nd_item);
|
723
|
-
|
753
|
+
stream->submit([&](sycl::handler & cgh) {
|
754
|
+
cgh.parallel_for(sycl::nd_range<3>(global_size, workgroup_size),
|
755
|
+
[=](sycl::nd_item<3> nd_item) [[sycl::reqd_sub_group_size(WARP_SIZE)]] {
|
756
|
+
mul_mat_vec_q_reorder<reorder_vec_dot_q_sycl<GGML_TYPE_Q4_K>>(vx, vy, dst, ncols,
|
757
|
+
nrows, nd_item);
|
758
|
+
});
|
724
759
|
});
|
725
760
|
}
|
726
761
|
|
@@ -734,12 +769,17 @@ static void mul_mat_vec_q5_K_q8_1_sycl(const void *vx, const void *vy,
|
|
734
769
|
const sycl::range<3> block_nums(1, 1, block_num_y);
|
735
770
|
const sycl::range<3> block_dims(1, GGML_SYCL_MMV_Y, WARP_SIZE);
|
736
771
|
{
|
737
|
-
|
738
|
-
|
739
|
-
|
740
|
-
|
741
|
-
|
742
|
-
|
772
|
+
|
773
|
+
stream->submit([&](sycl::handler &cgh) {
|
774
|
+
|
775
|
+
cgh.parallel_for(
|
776
|
+
sycl::nd_range<3>(block_nums * block_dims, block_dims),
|
777
|
+
[=](sycl::nd_item<3> item_ct1)
|
778
|
+
[[sycl::reqd_sub_group_size(WARP_SIZE)]] {
|
779
|
+
mul_mat_vec_q<QK_K, QI5_K, block_q5_K,
|
780
|
+
VDR_Q5_K_Q8_1_MMVQ, vec_dot_q5_K_q8_1>(
|
781
|
+
vx, vy, dst, ncols, nrows, item_ct1);
|
782
|
+
});
|
743
783
|
});
|
744
784
|
}
|
745
785
|
}
|
@@ -754,12 +794,12 @@ static void reorder_mul_mat_vec_q6_k_q8_1_sycl(const void * vx, const void * vy,
|
|
754
794
|
const sycl::range<3> global_size(1, GGML_SYCL_MMV_Y, block_num_y * WARP_SIZE);
|
755
795
|
const sycl::range<3> workgroup_size(1, GGML_SYCL_MMV_Y, num_subgroups * WARP_SIZE);
|
756
796
|
|
757
|
-
|
758
|
-
|
759
|
-
|
760
|
-
|
761
|
-
|
762
|
-
|
797
|
+
stream->submit([&](sycl::handler & cgh) {
|
798
|
+
cgh.parallel_for(sycl::nd_range<3>(global_size, workgroup_size),
|
799
|
+
[=](sycl::nd_item<3> nd_item) [[sycl::reqd_sub_group_size(WARP_SIZE)]] {
|
800
|
+
mul_mat_vec_q_reorder<reorder_vec_dot_q_sycl<GGML_TYPE_Q6_K>>(vx, vy, dst, ncols, nrows,
|
801
|
+
nd_item);
|
802
|
+
});
|
763
803
|
});
|
764
804
|
}
|
765
805
|
static void mul_mat_vec_q6_K_q8_1_sycl(const void *vx, const void *vy,
|
@@ -771,12 +811,17 @@ static void mul_mat_vec_q6_K_q8_1_sycl(const void *vx, const void *vy,
|
|
771
811
|
const sycl::range<3> block_nums(1, 1, block_num_y);
|
772
812
|
const sycl::range<3> block_dims(1, GGML_SYCL_MMV_Y, WARP_SIZE);
|
773
813
|
{
|
774
|
-
|
775
|
-
|
776
|
-
|
777
|
-
|
778
|
-
|
779
|
-
|
814
|
+
|
815
|
+
stream->submit([&](sycl::handler &cgh) {
|
816
|
+
|
817
|
+
cgh.parallel_for(
|
818
|
+
sycl::nd_range<3>(block_nums * block_dims, block_dims),
|
819
|
+
[=](sycl::nd_item<3> item_ct1)
|
820
|
+
[[sycl::reqd_sub_group_size(WARP_SIZE)]] {
|
821
|
+
mul_mat_vec_q<QK_K, QI6_K, block_q6_K,
|
822
|
+
VDR_Q6_K_Q8_1_MMVQ, vec_dot_q6_K_q8_1>(
|
823
|
+
vx, vy, dst, ncols, nrows, item_ct1);
|
824
|
+
});
|
780
825
|
});
|
781
826
|
}
|
782
827
|
}
|
@@ -791,12 +836,14 @@ static void mul_mat_vec_iq2_xxs_q8_1_sycl(const void *vx, const void *vy,
|
|
791
836
|
const sycl::range<3> block_nums(1, 1, block_num_y);
|
792
837
|
const sycl::range<3> block_dims(1, GGML_SYCL_MMV_Y, WARP_SIZE);
|
793
838
|
{
|
794
|
-
|
795
|
-
|
796
|
-
|
797
|
-
|
798
|
-
|
799
|
-
|
839
|
+
stream->submit([&](sycl::handler &cgh) {
|
840
|
+
cgh.parallel_for(
|
841
|
+
sycl::nd_range<3>(block_nums * block_dims, block_dims),
|
842
|
+
[=](sycl::nd_item<3> item_ct1)
|
843
|
+
[[sycl::reqd_sub_group_size(WARP_SIZE)]] {
|
844
|
+
mul_mat_vec_q_iq2_xxs_q8_1<QK_K, QI2_XXS/2, block_iq2_xxs, 1>(
|
845
|
+
vx, vy, dst, ncols, nrows, item_ct1);
|
846
|
+
});
|
800
847
|
});
|
801
848
|
}
|
802
849
|
}
|
@@ -810,12 +857,14 @@ static void mul_mat_vec_iq2_xs_q8_1_sycl(const void *vx, const void *vy,
|
|
810
857
|
const sycl::range<3> block_nums(1, 1, block_num_y);
|
811
858
|
const sycl::range<3> block_dims(1, GGML_SYCL_MMV_Y, WARP_SIZE);
|
812
859
|
{
|
813
|
-
|
814
|
-
|
815
|
-
|
816
|
-
|
817
|
-
|
818
|
-
|
860
|
+
stream->submit([&](sycl::handler & cgh) {
|
861
|
+
cgh.parallel_for(
|
862
|
+
sycl::nd_range<3>(block_nums * block_dims, block_dims),
|
863
|
+
[=](sycl::nd_item<3> item_ct1)
|
864
|
+
[[sycl::reqd_sub_group_size(WARP_SIZE)]] {
|
865
|
+
mul_mat_vec_q_iq2_xs_q8_1<QK_K, QI2_XS/2, block_iq2_xs, 1>(
|
866
|
+
vx, vy, dst, ncols, nrows, item_ct1);
|
867
|
+
});
|
819
868
|
});
|
820
869
|
}
|
821
870
|
}
|
@@ -829,12 +878,15 @@ static void mul_mat_vec_iq2_s_q8_1_sycl(const void *vx, const void *vy,
|
|
829
878
|
const sycl::range<3> block_nums(1, 1, block_num_y);
|
830
879
|
const sycl::range<3> block_dims(1, GGML_SYCL_MMV_Y, WARP_SIZE);
|
831
880
|
{
|
832
|
-
|
833
|
-
|
834
|
-
|
835
|
-
|
836
|
-
|
837
|
-
|
881
|
+
|
882
|
+
stream->submit([&](sycl::handler &cgh) {
|
883
|
+
cgh.parallel_for(
|
884
|
+
sycl::nd_range<3>(block_nums * block_dims, block_dims),
|
885
|
+
[=](sycl::nd_item<3> item_ct1)
|
886
|
+
[[sycl::reqd_sub_group_size(WARP_SIZE)]] {
|
887
|
+
mul_mat_vec_q_iq2_s_q8_1<QK_K, QI2_S/2, block_iq2_s, 1>(
|
888
|
+
vx, vy, dst, ncols, nrows, item_ct1);
|
889
|
+
});
|
838
890
|
});
|
839
891
|
}
|
840
892
|
}
|
@@ -848,12 +900,15 @@ static void mul_mat_vec_iq3_xxs_q8_1_sycl(const void *vx, const void *vy,
|
|
848
900
|
const sycl::range<3> block_nums(1, 1, block_num_y);
|
849
901
|
const sycl::range<3> block_dims(1, GGML_SYCL_MMV_Y, WARP_SIZE);
|
850
902
|
{
|
851
|
-
|
852
|
-
|
853
|
-
|
854
|
-
|
855
|
-
|
856
|
-
|
903
|
+
|
904
|
+
stream->submit([&](sycl::handler &cgh) {
|
905
|
+
cgh.parallel_for(
|
906
|
+
sycl::nd_range<3>(block_nums * block_dims, block_dims),
|
907
|
+
[=](sycl::nd_item<3> item_ct1)
|
908
|
+
[[sycl::reqd_sub_group_size(WARP_SIZE)]] {
|
909
|
+
mul_mat_vec_q_iq3_xxs_q8_1<QK_K, QI3_XXS/2, block_iq3_xxs, 1>(
|
910
|
+
vx, vy, dst, ncols, nrows, item_ct1);
|
911
|
+
});
|
857
912
|
});
|
858
913
|
}
|
859
914
|
}
|
@@ -867,12 +922,15 @@ static void mul_mat_vec_iq3_s_q8_1_sycl(const void *vx, const void *vy,
|
|
867
922
|
const sycl::range<3> block_nums(1, 1, block_num_y);
|
868
923
|
const sycl::range<3> block_dims(1, GGML_SYCL_MMV_Y, WARP_SIZE);
|
869
924
|
{
|
870
|
-
|
871
|
-
|
872
|
-
|
873
|
-
|
874
|
-
|
875
|
-
|
925
|
+
|
926
|
+
stream->submit([&](sycl::handler &cgh) {
|
927
|
+
cgh.parallel_for(
|
928
|
+
sycl::nd_range<3>(block_nums * block_dims, block_dims),
|
929
|
+
[=](sycl::nd_item<3> item_ct1)
|
930
|
+
[[sycl::reqd_sub_group_size(WARP_SIZE)]] {
|
931
|
+
mul_mat_vec_q_iq3_s_q8_1<QK_K, QI3_S/2, block_iq3_s, 1>(
|
932
|
+
vx, vy, dst, ncols, nrows, item_ct1);
|
933
|
+
});
|
876
934
|
});
|
877
935
|
}
|
878
936
|
}
|
@@ -886,12 +944,15 @@ static void mul_mat_vec_iq1_s_q8_1_sycl(const void *vx, const void *vy,
|
|
886
944
|
const sycl::range<3> block_nums(1, 1, block_num_y);
|
887
945
|
const sycl::range<3> block_dims(1, GGML_SYCL_MMV_Y, WARP_SIZE);
|
888
946
|
{
|
889
|
-
|
890
|
-
|
891
|
-
|
892
|
-
|
893
|
-
|
894
|
-
|
947
|
+
|
948
|
+
stream->submit([&](sycl::handler &cgh) {
|
949
|
+
cgh.parallel_for(
|
950
|
+
sycl::nd_range<3>(block_nums * block_dims, block_dims),
|
951
|
+
[=](sycl::nd_item<3> item_ct1)
|
952
|
+
[[sycl::reqd_sub_group_size(WARP_SIZE)]] {
|
953
|
+
mul_mat_vec_q_iq1_s_q8_1<QK_K, QI1_S, block_iq1_s, 1>(
|
954
|
+
vx, vy, dst, ncols, nrows, item_ct1);
|
955
|
+
});
|
895
956
|
});
|
896
957
|
}
|
897
958
|
}
|
@@ -905,12 +966,14 @@ static void mul_mat_vec_iq1_m_q8_1_sycl(const void *vx, const void *vy,
|
|
905
966
|
const sycl::range<3> block_nums(1, 1, block_num_y);
|
906
967
|
const sycl::range<3> block_dims(1, GGML_SYCL_MMV_Y, WARP_SIZE);
|
907
968
|
{
|
908
|
-
|
909
|
-
|
910
|
-
|
911
|
-
|
912
|
-
|
913
|
-
|
969
|
+
stream->submit([&](sycl::handler &cgh) {
|
970
|
+
cgh.parallel_for(
|
971
|
+
sycl::nd_range<3>(block_nums * block_dims, block_dims),
|
972
|
+
[=](sycl::nd_item<3> item_ct1)
|
973
|
+
[[sycl::reqd_sub_group_size(WARP_SIZE)]] {
|
974
|
+
mul_mat_vec_q_iq1_m_q8_1<QK_K, QI1_S, block_iq1_m, 1>(
|
975
|
+
vx, vy, dst, ncols, nrows, item_ct1);
|
976
|
+
});
|
914
977
|
});
|
915
978
|
}
|
916
979
|
}
|
@@ -924,12 +987,15 @@ static void mul_mat_vec_iq4_nl_q8_1_sycl(const void *vx, const void *vy,
|
|
924
987
|
const sycl::range<3> block_nums(1, 1, block_num_y);
|
925
988
|
const sycl::range<3> block_dims(1, GGML_SYCL_MMV_Y, WARP_SIZE);
|
926
989
|
{
|
927
|
-
|
928
|
-
|
929
|
-
|
930
|
-
|
931
|
-
|
932
|
-
|
990
|
+
|
991
|
+
stream->submit([&](sycl::handler &cgh) {
|
992
|
+
cgh.parallel_for(
|
993
|
+
sycl::nd_range<3>(block_nums * block_dims, block_dims),
|
994
|
+
[=](sycl::nd_item<3> item_ct1)
|
995
|
+
[[sycl::reqd_sub_group_size(WARP_SIZE)]] {
|
996
|
+
mul_mat_vec_q_iq4_nl_q8_1<QK4_NL, QI4_NL, block_iq4_nl, 2>(
|
997
|
+
vx, vy, dst, ncols, nrows, item_ct1);
|
998
|
+
});
|
933
999
|
});
|
934
1000
|
}
|
935
1001
|
}
|
@@ -943,12 +1009,15 @@ static void mul_mat_vec_iq4_xs_q8_1_sycl(const void *vx, const void *vy,
|
|
943
1009
|
const sycl::range<3> block_nums(1, 1, block_num_y);
|
944
1010
|
const sycl::range<3> block_dims(1, GGML_SYCL_MMV_Y, WARP_SIZE);
|
945
1011
|
{
|
946
|
-
|
947
|
-
|
948
|
-
|
949
|
-
|
950
|
-
|
951
|
-
|
1012
|
+
|
1013
|
+
stream->submit([&](sycl::handler &cgh) {
|
1014
|
+
cgh.parallel_for(
|
1015
|
+
sycl::nd_range<3>(block_nums * block_dims, block_dims),
|
1016
|
+
[=](sycl::nd_item<3> item_ct1)
|
1017
|
+
[[sycl::reqd_sub_group_size(WARP_SIZE)]] {
|
1018
|
+
mul_mat_vec_q_iq4_xs_q8_1<QK_K, QI4_XS/4, block_iq4_xs, 1>(
|
1019
|
+
vx, vy, dst, ncols, nrows, item_ct1);
|
1020
|
+
});
|
952
1021
|
});
|
953
1022
|
}
|
954
1023
|
}
|