whispercpp 1.3.5 → 1.3.7
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/.document +3 -0
- data/.rdoc_options +2 -0
- data/LICENSE +1 -1
- data/README.md +133 -3
- data/Rakefile +18 -3
- data/ext/dependencies.rb +10 -4
- data/ext/dependencies_for_windows.rb +17 -0
- data/ext/extconf.rb +20 -7
- data/ext/options.rb +54 -14
- data/ext/options_for_windows.rb +51 -0
- data/ext/ruby_whisper.c +56 -46
- data/ext/ruby_whisper.h +165 -2
- data/ext/ruby_whisper_context.c +297 -126
- data/ext/ruby_whisper_context_params.c +163 -0
- data/ext/ruby_whisper_log_queue.c +180 -0
- data/ext/ruby_whisper_log_settable.h +47 -0
- data/ext/ruby_whisper_model.c +0 -1
- data/ext/ruby_whisper_parakeet.c +49 -0
- data/ext/ruby_whisper_parakeet_context.c +304 -0
- data/ext/ruby_whisper_parakeet_context_params.c +117 -0
- data/ext/ruby_whisper_parakeet_model.c +84 -0
- data/ext/ruby_whisper_parakeet_params.c +548 -0
- data/ext/ruby_whisper_parakeet_segment.c +157 -0
- data/ext/ruby_whisper_parakeet_token.c +188 -0
- data/ext/ruby_whisper_parakeet_transcribe.cpp +58 -0
- data/ext/ruby_whisper_params.c +256 -66
- data/ext/ruby_whisper_segment.c +6 -7
- data/ext/ruby_whisper_token.c +29 -9
- data/ext/ruby_whisper_transcribe.cpp +46 -16
- data/ext/ruby_whisper_vad_context.c +48 -1
- data/ext/ruby_whisper_vad_context_detect.cpp +6 -5
- data/ext/ruby_whisper_vad_params.c +0 -1
- data/ext/ruby_whisper_vad_segment.c +0 -1
- data/ext/ruby_whisper_vad_segments.c +0 -1
- data/ext/sources/CMakeLists.txt +41 -3
- data/ext/sources/CMakePresets.json +95 -0
- data/ext/sources/cmake/parakeet-config.cmake.in +30 -0
- data/ext/sources/cmake/parakeet.pc.in +10 -0
- data/ext/sources/cmake/whisper-config.cmake.in +5 -40
- data/ext/sources/cmake/whisper.pc.in +1 -1
- data/ext/sources/examples/CMakeLists.txt +4 -2
- data/ext/sources/examples/bench/bench.cpp +24 -19
- data/ext/sources/examples/cli/cli.cpp +51 -9
- data/ext/sources/examples/common-ggml.cpp +4 -0
- data/ext/sources/examples/common-whisper.cpp +139 -67
- data/ext/sources/examples/common-whisper.h +11 -0
- data/ext/sources/examples/ffmpeg-transcode.cpp +211 -341
- data/ext/sources/examples/miniaudio.h +4507 -2131
- data/ext/sources/examples/parakeet-cli/CMakeLists.txt +8 -0
- data/ext/sources/examples/parakeet-cli/parakeet-cli.cpp +243 -0
- data/ext/sources/examples/parakeet-quantize/CMakeLists.txt +7 -0
- data/ext/sources/examples/parakeet-quantize/parakeet-quantize.cpp +230 -0
- data/ext/sources/examples/server/server.cpp +213 -163
- data/ext/sources/ggml/CMakeLists.txt +29 -15
- data/ext/sources/ggml/cmake/FindNCCL.cmake +36 -0
- data/ext/sources/ggml/cmake/ggml-config.cmake.in +12 -2
- data/ext/sources/ggml/include/ggml-alloc.h +1 -0
- data/ext/sources/ggml/include/ggml-backend.h +73 -11
- data/ext/sources/ggml/include/ggml-cann.h +1 -1
- data/ext/sources/ggml/include/ggml-cpu.h +5 -0
- data/ext/sources/ggml/include/ggml-cuda.h +3 -0
- data/ext/sources/ggml/include/ggml-openvino.h +37 -0
- data/ext/sources/ggml/include/ggml-opt.h +1 -1
- data/ext/sources/ggml/include/ggml-rpc.h +8 -3
- data/ext/sources/ggml/include/ggml-virtgpu.h +14 -0
- data/ext/sources/ggml/include/ggml.h +155 -16
- data/ext/sources/ggml/include/gguf.h +10 -2
- data/ext/sources/ggml/src/CMakeLists.txt +25 -5
- data/ext/sources/ggml/src/ggml-alloc.c +9 -10
- data/ext/sources/ggml/src/ggml-backend-dl.cpp +48 -0
- data/ext/sources/ggml/src/ggml-backend-dl.h +45 -0
- data/ext/sources/ggml/src/ggml-backend-impl.h +22 -2
- data/ext/sources/ggml/src/ggml-backend-meta.cpp +2263 -0
- data/ext/sources/ggml/src/ggml-backend-reg.cpp +40 -86
- data/ext/sources/ggml/src/ggml-backend.cpp +114 -10
- data/ext/sources/ggml/src/ggml-blas/CMakeLists.txt +1 -1
- data/ext/sources/ggml/src/ggml-blas/ggml-blas.cpp +10 -2
- data/ext/sources/ggml/src/ggml-cann/acl_tensor.cpp +1 -1
- data/ext/sources/ggml/src/ggml-cann/acl_tensor.h +1 -1
- data/ext/sources/ggml/src/ggml-cann/aclnn_ops.cpp +1016 -442
- data/ext/sources/ggml/src/ggml-cann/aclnn_ops.h +111 -85
- data/ext/sources/ggml/src/ggml-cann/common.h +23 -14
- data/ext/sources/ggml/src/ggml-cann/ggml-cann.cpp +255 -92
- data/ext/sources/ggml/src/ggml-common.h +22 -0
- data/ext/sources/ggml/src/ggml-cpu/CMakeLists.txt +68 -34
- data/ext/sources/ggml/src/ggml-cpu/amx/amx.cpp +44 -19
- data/ext/sources/ggml/src/ggml-cpu/amx/common.h +34 -10
- data/ext/sources/ggml/src/ggml-cpu/amx/mmq.cpp +101 -101
- data/ext/sources/ggml/src/ggml-cpu/arch/arm/quants.c +194 -1
- data/ext/sources/ggml/src/ggml-cpu/arch/arm/repack.cpp +2874 -613
- data/ext/sources/ggml/src/ggml-cpu/arch/loongarch/quants.c +151 -1
- data/ext/sources/ggml/src/ggml-cpu/arch/powerpc/quants.c +0 -1
- data/ext/sources/ggml/src/ggml-cpu/arch/riscv/quants.c +5480 -840
- data/ext/sources/ggml/src/ggml-cpu/arch/riscv/repack.cpp +1361 -0
- data/ext/sources/ggml/src/ggml-cpu/arch/s390/quants.c +8 -11
- data/ext/sources/ggml/src/ggml-cpu/arch/wasm/quants.c +72 -1
- data/ext/sources/ggml/src/ggml-cpu/arch/x86/quants.c +186 -36
- data/ext/sources/ggml/src/ggml-cpu/arch/x86/repack.cpp +119 -19
- data/ext/sources/ggml/src/ggml-cpu/arch-fallback.h +112 -26
- data/ext/sources/ggml/src/ggml-cpu/binary-ops.cpp +2 -6
- data/ext/sources/ggml/src/ggml-cpu/cmake/FindSMTIME.cmake +32 -0
- data/ext/sources/ggml/src/ggml-cpu/common.h +8 -0
- data/ext/sources/ggml/src/ggml-cpu/ggml-cpu-impl.h +13 -0
- data/ext/sources/ggml/src/ggml-cpu/ggml-cpu.c +153 -16
- data/ext/sources/ggml/src/ggml-cpu/ggml-cpu.cpp +17 -0
- data/ext/sources/ggml/src/ggml-cpu/kleidiai/kernels.cpp +21 -20
- data/ext/sources/ggml/src/ggml-cpu/kleidiai/kleidiai.cpp +976 -251
- data/ext/sources/ggml/src/ggml-cpu/llamafile/sgemm.cpp +671 -266
- data/ext/sources/ggml/src/ggml-cpu/ops.cpp +1277 -263
- data/ext/sources/ggml/src/ggml-cpu/ops.h +4 -0
- data/ext/sources/ggml/src/ggml-cpu/quants.c +95 -0
- data/ext/sources/ggml/src/ggml-cpu/quants.h +6 -0
- data/ext/sources/ggml/src/ggml-cpu/repack.cpp +2893 -679
- data/ext/sources/ggml/src/ggml-cpu/repack.h +119 -8
- data/ext/sources/ggml/src/ggml-cpu/simd-gemm.h +226 -0
- data/ext/sources/ggml/src/ggml-cpu/simd-mappings.h +114 -19
- data/ext/sources/ggml/src/ggml-cpu/spacemit/ime.cpp +1402 -687
- data/ext/sources/ggml/src/ggml-cpu/spacemit/ime.h +8 -0
- data/ext/sources/ggml/src/ggml-cpu/spacemit/ime1_kernels.cpp +597 -2766
- data/ext/sources/ggml/src/ggml-cpu/spacemit/ime2_kernels.cpp +5768 -0
- data/ext/sources/ggml/src/ggml-cpu/spacemit/ime_env.cpp +320 -0
- data/ext/sources/ggml/src/ggml-cpu/spacemit/ime_env.h +55 -0
- data/ext/sources/ggml/src/ggml-cpu/spacemit/ime_kernels.h +182 -19
- data/ext/sources/ggml/src/ggml-cpu/spacemit/repack.cpp +1795 -0
- data/ext/sources/ggml/src/ggml-cpu/spacemit/repack.h +14 -0
- data/ext/sources/ggml/src/ggml-cpu/spacemit/rvv_kernels.cpp +3178 -0
- data/ext/sources/ggml/src/ggml-cpu/spacemit/rvv_kernels.h +95 -0
- data/ext/sources/ggml/src/ggml-cpu/spacemit/spine_barrier.h +34 -0
- data/ext/sources/ggml/src/ggml-cpu/spacemit/spine_mem_pool.cpp +760 -0
- data/ext/sources/ggml/src/ggml-cpu/spacemit/spine_mem_pool.h +32 -0
- data/ext/sources/ggml/src/ggml-cpu/spacemit/spine_tcm.h +409 -0
- data/ext/sources/ggml/src/ggml-cpu/unary-ops.cpp +1 -1
- data/ext/sources/ggml/src/ggml-cpu/vec.cpp +54 -53
- data/ext/sources/ggml/src/ggml-cpu/vec.h +225 -240
- data/ext/sources/ggml/src/ggml-cuda/CMakeLists.txt +18 -8
- data/ext/sources/ggml/src/ggml-cuda/allreduce.cu +971 -0
- data/ext/sources/ggml/src/ggml-cuda/allreduce.cuh +29 -0
- data/ext/sources/ggml/src/ggml-cuda/argsort.cu +73 -28
- data/ext/sources/ggml/src/ggml-cuda/binbcast.cu +69 -41
- data/ext/sources/ggml/src/ggml-cuda/binbcast.cuh +1 -0
- data/ext/sources/ggml/src/ggml-cuda/common.cuh +359 -29
- data/ext/sources/ggml/src/ggml-cuda/concat.cu +120 -114
- data/ext/sources/ggml/src/ggml-cuda/conv2d-transpose.cu +45 -21
- data/ext/sources/ggml/src/ggml-cuda/conv2d-transpose.cuh +1 -0
- data/ext/sources/ggml/src/ggml-cuda/convert.cu +94 -27
- data/ext/sources/ggml/src/ggml-cuda/convert.cuh +10 -0
- data/ext/sources/ggml/src/ggml-cuda/cpy.cu +20 -9
- data/ext/sources/ggml/src/ggml-cuda/dequantize.cuh +22 -0
- data/ext/sources/ggml/src/ggml-cuda/fattn-common.cuh +333 -85
- data/ext/sources/ggml/src/ggml-cuda/fattn-mma-f16.cuh +632 -190
- data/ext/sources/ggml/src/ggml-cuda/fattn-tile.cu +12 -0
- data/ext/sources/ggml/src/ggml-cuda/fattn-tile.cuh +162 -49
- data/ext/sources/ggml/src/ggml-cuda/fattn-vec.cuh +43 -18
- data/ext/sources/ggml/src/ggml-cuda/fattn-wmma-f16.cu +44 -14
- data/ext/sources/ggml/src/ggml-cuda/fattn-wmma-f16.cuh +1 -1
- data/ext/sources/ggml/src/ggml-cuda/fattn.cu +241 -23
- data/ext/sources/ggml/src/ggml-cuda/fattn.cuh +2 -0
- data/ext/sources/ggml/src/ggml-cuda/fwht.cu +101 -0
- data/ext/sources/ggml/src/ggml-cuda/fwht.cuh +4 -0
- data/ext/sources/ggml/src/ggml-cuda/gated_delta_net.cu +312 -0
- data/ext/sources/ggml/src/ggml-cuda/gated_delta_net.cuh +4 -0
- data/ext/sources/ggml/src/ggml-cuda/getrows.cu +34 -12
- data/ext/sources/ggml/src/ggml-cuda/ggml-cuda.cu +1454 -599
- data/ext/sources/ggml/src/ggml-cuda/im2col.cu +32 -29
- data/ext/sources/ggml/src/ggml-cuda/mean.cu +13 -10
- data/ext/sources/ggml/src/ggml-cuda/mma.cuh +397 -183
- data/ext/sources/ggml/src/ggml-cuda/mmf.cu +30 -10
- data/ext/sources/ggml/src/ggml-cuda/mmf.cuh +161 -88
- data/ext/sources/ggml/src/ggml-cuda/mmq.cu +18 -12
- data/ext/sources/ggml/src/ggml-cuda/mmq.cuh +522 -431
- data/ext/sources/ggml/src/ggml-cuda/mmvf.cu +139 -72
- data/ext/sources/ggml/src/ggml-cuda/mmvf.cuh +2 -0
- data/ext/sources/ggml/src/ggml-cuda/mmvq.cu +608 -88
- data/ext/sources/ggml/src/ggml-cuda/mmvq.cuh +6 -0
- data/ext/sources/ggml/src/ggml-cuda/norm.cu +47 -79
- data/ext/sources/ggml/src/ggml-cuda/out-prod.cu +23 -7
- data/ext/sources/ggml/src/ggml-cuda/pad.cu +13 -10
- data/ext/sources/ggml/src/ggml-cuda/quantize.cu +134 -27
- data/ext/sources/ggml/src/ggml-cuda/quantize.cuh +1 -1
- data/ext/sources/ggml/src/ggml-cuda/reduce_rows.cuh +7 -17
- data/ext/sources/ggml/src/ggml-cuda/rope.cu +244 -137
- data/ext/sources/ggml/src/ggml-cuda/scale.cu +4 -1
- data/ext/sources/ggml/src/ggml-cuda/set-rows.cu +14 -6
- data/ext/sources/ggml/src/ggml-cuda/snake.cu +72 -0
- data/ext/sources/ggml/src/ggml-cuda/snake.cuh +8 -0
- data/ext/sources/ggml/src/ggml-cuda/softcap.cu +4 -1
- data/ext/sources/ggml/src/ggml-cuda/softmax.cu +8 -83
- data/ext/sources/ggml/src/ggml-cuda/solve_tri.cu +1 -1
- data/ext/sources/ggml/src/ggml-cuda/ssm-conv.cu +96 -40
- data/ext/sources/ggml/src/ggml-cuda/ssm-conv.cuh +1 -1
- data/ext/sources/ggml/src/ggml-cuda/ssm-scan.cu +40 -18
- data/ext/sources/ggml/src/ggml-cuda/sumrows.cu +8 -4
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-mma-f16-instance-ncols1_1-ncols2_16.cu +1 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-mma-f16-instance-ncols1_1-ncols2_32.cu +6 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-mma-f16-instance-ncols1_1-ncols2_8.cu +2 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-mma-f16-instance-ncols1_16-ncols2_4.cu +2 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-mma-f16-instance-ncols1_2-ncols2_16.cu +1 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-mma-f16-instance-ncols1_2-ncols2_32.cu +6 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-mma-f16-instance-ncols1_2-ncols2_4.cu +2 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-mma-f16-instance-ncols1_2-ncols2_8.cu +2 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-mma-f16-instance-ncols1_4-ncols2_16.cu +1 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-mma-f16-instance-ncols1_4-ncols2_4.cu +2 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-mma-f16-instance-ncols1_4-ncols2_8.cu +2 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-mma-f16-instance-ncols1_8-ncols2_4.cu +2 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-mma-f16-instance-ncols1_8-ncols2_8.cu +2 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-tile-instance-dkq192-dv128.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-tile-instance-dkq320-dv256.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-tile-instance-dkq512-dv512.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-bf16-bf16.cu +7 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-bf16-f16.cu +7 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-bf16-q4_0.cu +7 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-bf16-q4_1.cu +7 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-bf16-q5_0.cu +7 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-bf16-q5_1.cu +7 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-bf16-q8_0.cu +7 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-f16-bf16.cu +7 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q4_0-bf16.cu +7 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q4_1-bf16.cu +7 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q5_0-bf16.cu +7 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q5_1-bf16.cu +7 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q8_0-bf16.cu +7 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/mmq-instance-nvfp4.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/mmq-instance-q1_0.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/top-k.cu +5 -5
- data/ext/sources/ggml/src/ggml-cuda/topk-moe.cu +202 -135
- data/ext/sources/ggml/src/ggml-cuda/topk-moe.cuh +20 -14
- data/ext/sources/ggml/src/ggml-cuda/unary.cu +86 -2
- data/ext/sources/ggml/src/ggml-cuda/unary.cuh +4 -0
- data/ext/sources/ggml/src/ggml-cuda/vecdotq.cuh +111 -17
- data/ext/sources/ggml/src/ggml-cuda/vendors/cuda.h +7 -2
- data/ext/sources/ggml/src/ggml-cuda/vendors/hip.h +30 -2
- data/ext/sources/ggml/src/ggml-cuda/vendors/musa.h +3 -0
- data/ext/sources/ggml/src/ggml-hexagon/CMakeLists.txt +84 -46
- data/ext/sources/ggml/src/ggml-hexagon/ggml-hexagon.cpp +1612 -753
- data/ext/sources/ggml/src/ggml-hexagon/htp/CMakeLists.txt +51 -11
- data/ext/sources/ggml/src/ggml-hexagon/htp/act-ops.c +361 -261
- data/ext/sources/ggml/src/ggml-hexagon/htp/argsort-ops.c +294 -0
- data/ext/sources/ggml/src/ggml-hexagon/htp/binary-ops.c +753 -241
- data/ext/sources/ggml/src/ggml-hexagon/htp/cmake-toolchain.cmake +5 -5
- data/ext/sources/ggml/src/ggml-hexagon/htp/concat-ops.c +277 -0
- data/ext/sources/ggml/src/ggml-hexagon/htp/cpy-ops.c +295 -0
- data/ext/sources/ggml/src/ggml-hexagon/htp/cumsum-ops.c +270 -0
- data/ext/sources/ggml/src/ggml-hexagon/htp/diag-ops.c +216 -0
- data/ext/sources/ggml/src/ggml-hexagon/htp/fill-ops.c +123 -0
- data/ext/sources/ggml/src/ggml-hexagon/htp/flash-attn-ops.c +471 -296
- data/ext/sources/ggml/src/ggml-hexagon/htp/gated-delta-net-ops.c +1148 -0
- data/ext/sources/ggml/src/ggml-hexagon/htp/get-rows-ops.c +159 -53
- data/ext/sources/ggml/src/ggml-hexagon/htp/{htp-dma.c → hex-dma.c} +3 -3
- data/ext/sources/ggml/src/ggml-hexagon/htp/hex-dma.h +372 -0
- data/ext/sources/ggml/src/ggml-hexagon/htp/hex-dump.h +86 -0
- data/ext/sources/ggml/src/ggml-hexagon/htp/hex-fastdiv.h +37 -0
- data/ext/sources/ggml/src/ggml-hexagon/htp/hex-utils.h +137 -0
- data/ext/sources/ggml/src/ggml-hexagon/htp/hmx-flash-attn-ops.c +1878 -0
- data/ext/sources/ggml/src/ggml-hexagon/htp/hmx-matmul-ops.c +2066 -0
- data/ext/sources/ggml/src/ggml-hexagon/htp/hmx-ops.c +6 -0
- data/ext/sources/ggml/src/ggml-hexagon/htp/hmx-ops.h +88 -0
- data/ext/sources/ggml/src/ggml-hexagon/htp/hmx-profile.h +34 -0
- data/ext/sources/ggml/src/ggml-hexagon/htp/hmx-queue.c +158 -0
- data/ext/sources/ggml/src/ggml-hexagon/htp/hmx-queue.h +134 -0
- data/ext/sources/ggml/src/ggml-hexagon/htp/hmx-utils.h +200 -0
- data/ext/sources/ggml/src/ggml-hexagon/htp/htp-ctx.h +97 -14
- data/ext/sources/ggml/src/ggml-hexagon/htp/htp-ops.h +163 -67
- data/ext/sources/ggml/src/ggml-hexagon/htp/htp_iface.idl +9 -3
- data/ext/sources/ggml/src/ggml-hexagon/htp/hvx-arith.h +443 -0
- data/ext/sources/ggml/src/ggml-hexagon/htp/hvx-base.h +308 -0
- data/ext/sources/ggml/src/ggml-hexagon/htp/hvx-copy.h +262 -0
- data/ext/sources/ggml/src/ggml-hexagon/htp/hvx-div.h +291 -0
- data/ext/sources/ggml/src/ggml-hexagon/htp/hvx-dump.h +129 -0
- data/ext/sources/ggml/src/ggml-hexagon/htp/hvx-exp.h +216 -0
- data/ext/sources/ggml/src/ggml-hexagon/htp/hvx-flash-attn.h +47 -0
- data/ext/sources/ggml/src/ggml-hexagon/htp/hvx-floor.h +100 -0
- data/ext/sources/ggml/src/ggml-hexagon/htp/hvx-inverse.h +210 -0
- data/ext/sources/ggml/src/ggml-hexagon/htp/hvx-log.h +65 -0
- data/ext/sources/ggml/src/ggml-hexagon/htp/hvx-pow.h +42 -0
- data/ext/sources/ggml/src/ggml-hexagon/htp/hvx-reduce.h +296 -0
- data/ext/sources/ggml/src/ggml-hexagon/htp/hvx-repl.h +74 -0
- data/ext/sources/ggml/src/ggml-hexagon/htp/hvx-scale.h +133 -0
- data/ext/sources/ggml/src/ggml-hexagon/htp/hvx-sigmoid.h +142 -0
- data/ext/sources/ggml/src/ggml-hexagon/htp/hvx-sin-cos.h +90 -0
- data/ext/sources/ggml/src/ggml-hexagon/htp/hvx-sqrt.h +126 -0
- data/ext/sources/ggml/src/ggml-hexagon/htp/hvx-types.h +36 -0
- data/ext/sources/ggml/src/ggml-hexagon/htp/hvx-utils.h +18 -1348
- data/ext/sources/ggml/src/ggml-hexagon/htp/main.c +547 -635
- data/ext/sources/ggml/src/ggml-hexagon/htp/matmul-ops.c +3556 -1101
- data/ext/sources/ggml/src/ggml-hexagon/htp/pad-ops.c +547 -0
- data/ext/sources/ggml/src/ggml-hexagon/htp/repeat-ops.c +148 -0
- data/ext/sources/ggml/src/ggml-hexagon/htp/rope-ops.c +475 -269
- data/ext/sources/ggml/src/ggml-hexagon/htp/set-rows-ops.c +94 -72
- data/ext/sources/ggml/src/ggml-hexagon/htp/softmax-ops.c +222 -217
- data/ext/sources/ggml/src/ggml-hexagon/htp/solve-tri-ops.c +267 -0
- data/ext/sources/ggml/src/ggml-hexagon/htp/ssm-conv.c +432 -0
- data/ext/sources/ggml/src/ggml-hexagon/htp/sum-rows-ops.c +128 -0
- data/ext/sources/ggml/src/ggml-hexagon/htp/unary-ops.c +886 -117
- data/ext/sources/ggml/src/ggml-hexagon/htp/vtcm-utils.h +16 -0
- data/ext/sources/ggml/src/ggml-hexagon/htp/worker-pool.c +1 -5
- data/ext/sources/ggml/src/ggml-hexagon/htp-drv.cpp +418 -0
- data/ext/sources/ggml/src/ggml-hexagon/htp-drv.h +121 -0
- data/ext/sources/ggml/src/ggml-hexagon/htp-opnode.h +272 -0
- data/ext/sources/ggml/src/ggml-hexagon/libdl.h +79 -0
- data/ext/sources/ggml/src/ggml-hexagon/libggml-htp.inf +40 -0
- data/ext/sources/ggml/src/ggml-hip/CMakeLists.txt +28 -9
- data/ext/sources/ggml/src/ggml-impl.h +68 -1
- data/ext/sources/ggml/src/ggml-metal/CMakeLists.txt +10 -10
- data/ext/sources/ggml/src/ggml-metal/ggml-metal-common.cpp +13 -2
- data/ext/sources/ggml/src/ggml-metal/ggml-metal-context.h +8 -0
- data/ext/sources/ggml/src/ggml-metal/ggml-metal-context.m +147 -17
- data/ext/sources/ggml/src/ggml-metal/ggml-metal-device.cpp +409 -83
- data/ext/sources/ggml/src/ggml-metal/ggml-metal-device.h +54 -5
- data/ext/sources/ggml/src/ggml-metal/ggml-metal-device.m +254 -52
- data/ext/sources/ggml/src/ggml-metal/ggml-metal-impl.h +254 -23
- data/ext/sources/ggml/src/ggml-metal/ggml-metal-ops.cpp +756 -285
- data/ext/sources/ggml/src/ggml-metal/ggml-metal-ops.h +7 -4
- data/ext/sources/ggml/src/ggml-metal/ggml-metal.cpp +359 -133
- data/ext/sources/ggml/src/ggml-metal/ggml-metal.metal +1867 -1123
- data/ext/sources/ggml/src/ggml-musa/CMakeLists.txt +5 -6
- data/ext/sources/ggml/src/ggml-opencl/CMakeLists.txt +71 -4
- data/ext/sources/ggml/src/ggml-opencl/ggml-opencl.cpp +14127 -5314
- data/ext/sources/ggml/src/ggml-opencl/kernels/concat.cl +97 -88
- data/ext/sources/ggml/src/ggml-opencl/kernels/cpy.cl +104 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/cumsum.cl +139 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/cvt.cl +1978 -67
- data/ext/sources/ggml/src/ggml-opencl/kernels/diag.cl +27 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/exp.cl +125 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/expm1.cl +87 -56
- data/ext/sources/ggml/src/ggml-opencl/kernels/gated_delta_net.cl +249 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/gemm_moe_mxfp4_f32_ns.cl +306 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/gemm_moe_q4_0_f32_ns.cl +256 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/gemm_moe_q4_1_f32_ns.cl +258 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/gemm_moe_q4_k_f32_ns.cl +283 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/gemm_moe_q5_0_f32_ns.cl +260 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/gemm_moe_q5_1_f32_ns.cl +262 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/gemm_moe_q5_k_f32_ns.cl +288 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/gemm_moe_q6_k_f32_ns.cl +267 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/gemm_noshuffle_iq4_nl_f32.cl +150 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/{mul_mat_Ab_Bi_8x4.cl → gemm_noshuffle_q4_0_f32.cl} +1 -1
- data/ext/sources/ggml/src/ggml-opencl/kernels/gemm_noshuffle_q4_1_f32.cl +132 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/gemm_noshuffle_q4_k_f32.cl +172 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/gemm_noshuffle_q5_0_f32.cl +131 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/gemm_noshuffle_q5_1_f32.cl +134 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/gemm_noshuffle_q5_k_f32.cl +176 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/gemm_noshuffle_q6_k_f32.cl +140 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/gemm_noshuffle_q8_0_f32.cl +129 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/gemm_xmem_f16_f32_os8.cl +233 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/gemv_moe_mxfp4_f32_ns.cl +165 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/gemv_moe_q4_0_f32_ns.cl +120 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/gemv_moe_q4_1_f32_ns.cl +123 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/gemv_moe_q4_k_f32_ns.cl +155 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/gemv_moe_q5_0_f32_ns.cl +123 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/gemv_moe_q5_1_f32_ns.cl +125 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/gemv_moe_q5_k_f32_ns.cl +160 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/gemv_moe_q6_k_f32_ns.cl +141 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/gemv_noshuffle_iq4_nl_f32.cl +302 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/{gemv_noshuffle_general.cl → gemv_noshuffle_q4_0_f32.cl} +5 -5
- data/ext/sources/ggml/src/ggml-opencl/kernels/{gemv_noshuffle.cl → gemv_noshuffle_q4_0_f32_spec.cl} +5 -5
- data/ext/sources/ggml/src/ggml-opencl/kernels/gemv_noshuffle_q4_1_f32.cl +283 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/gemv_noshuffle_q4_k_f32.cl +318 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/gemv_noshuffle_q5_0_f32.cl +291 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/gemv_noshuffle_q5_1_f32.cl +294 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/gemv_noshuffle_q5_k_f32.cl +326 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/gemv_noshuffle_q6_k_f32.cl +293 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/gemv_noshuffle_q8_0_f32.cl +195 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/get_rows.cl +15 -9
- data/ext/sources/ggml/src/ggml-opencl/kernels/l2_norm.cl +71 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/mean.cl +114 -13
- data/ext/sources/ggml/src/ggml-opencl/kernels/moe_reorder_b.cl +30 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/moe_sort_by_expert.cl +82 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/mul_mm_iq4_nl_f32_l4_lm.cl +171 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/mul_mm_q4_0_f32_l4_lm.cl +163 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/mul_mm_q4_1_f32_l4_lm.cl +165 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/mul_mm_q4_k_f32_l4_lm.cl +179 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/mul_mm_q5_0_f32_l4_lm.cl +173 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/mul_mm_q5_1_f32_l4_lm.cl +175 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/mul_mm_q5_k_f32_l4_lm.cl +192 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/mul_mm_q6_k_f32_l4_lm.cl +158 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/mul_mv_iq4_nl_f32.cl +164 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/mul_mv_iq4_nl_f32_flat.cl +202 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/mul_mv_q4_1_f32.cl +219 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/mul_mv_q4_1_f32_flat.cl +229 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/mul_mv_q4_k_f32.cl +180 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/mul_mv_q4_k_f32_flat.cl +196 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/mul_mv_q5_0_f32.cl +241 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/mul_mv_q5_0_f32_flat.cl +243 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/mul_mv_q5_1_f32.cl +243 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/mul_mv_q5_1_f32_flat.cl +247 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/mul_mv_q5_k_f32.cl +187 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/mul_mv_q5_k_f32_flat.cl +203 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/{mul_mv_q6_k.cl → mul_mv_q6_k_f32.cl} +4 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/mul_mv_q6_k_f32_flat.cl +178 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/neg.cl +125 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/repeat.cl +31 -32
- data/ext/sources/ggml/src/ggml-opencl/kernels/scale.cl +14 -4
- data/ext/sources/ggml/src/ggml-opencl/kernels/softplus.cl +88 -60
- data/ext/sources/ggml/src/ggml-opencl/kernels/solve_tri.cl +51 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/sum_rows.cl +114 -13
- data/ext/sources/ggml/src/ggml-opencl/kernels/tanh.cl +94 -48
- data/ext/sources/ggml/src/ggml-opencl/kernels/transpose.cl +26 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/tri.cl +32 -0
- data/ext/sources/ggml/src/ggml-openvino/.clang-format +154 -0
- data/ext/sources/ggml/src/ggml-openvino/CMakeLists.txt +22 -0
- data/ext/sources/ggml/src/ggml-openvino/ggml-decoder.cpp +985 -0
- data/ext/sources/ggml/src/ggml-openvino/ggml-decoder.h +294 -0
- data/ext/sources/ggml/src/ggml-openvino/ggml-openvino-extra.cpp +380 -0
- data/ext/sources/ggml/src/ggml-openvino/ggml-openvino-extra.h +182 -0
- data/ext/sources/ggml/src/ggml-openvino/ggml-openvino.cpp +1132 -0
- data/ext/sources/ggml/src/ggml-openvino/ggml-quants.cpp +956 -0
- data/ext/sources/ggml/src/ggml-openvino/ggml-quants.h +153 -0
- data/ext/sources/ggml/src/ggml-openvino/openvino/decoder.h +74 -0
- data/ext/sources/ggml/src/ggml-openvino/openvino/frontend.cpp +27 -0
- data/ext/sources/ggml/src/ggml-openvino/openvino/frontend.h +23 -0
- data/ext/sources/ggml/src/ggml-openvino/openvino/input_model.cpp +17 -0
- data/ext/sources/ggml/src/ggml-openvino/openvino/input_model.h +29 -0
- data/ext/sources/ggml/src/ggml-openvino/openvino/node_context.h +112 -0
- data/ext/sources/ggml/src/ggml-openvino/openvino/op/cont.cpp +48 -0
- data/ext/sources/ggml/src/ggml-openvino/openvino/op/cpy.cpp +21 -0
- data/ext/sources/ggml/src/ggml-openvino/openvino/op/flash_attn_ext.cpp +90 -0
- data/ext/sources/ggml/src/ggml-openvino/openvino/op/get_rows.cpp +69 -0
- data/ext/sources/ggml/src/ggml-openvino/openvino/op/glu_geglu.cpp +61 -0
- data/ext/sources/ggml/src/ggml-openvino/openvino/op/glu_swiglu.cpp +62 -0
- data/ext/sources/ggml/src/ggml-openvino/openvino/op/mulmat.cpp +90 -0
- data/ext/sources/ggml/src/ggml-openvino/openvino/op/permute.cpp +102 -0
- data/ext/sources/ggml/src/ggml-openvino/openvino/op/reshape.cpp +83 -0
- data/ext/sources/ggml/src/ggml-openvino/openvino/op/rms_norm.cpp +46 -0
- data/ext/sources/ggml/src/ggml-openvino/openvino/op/rope.cpp +149 -0
- data/ext/sources/ggml/src/ggml-openvino/openvino/op/scale.cpp +41 -0
- data/ext/sources/ggml/src/ggml-openvino/openvino/op/set_rows.cpp +76 -0
- data/ext/sources/ggml/src/ggml-openvino/openvino/op/softmax.cpp +89 -0
- data/ext/sources/ggml/src/ggml-openvino/openvino/op/transpose.cpp +23 -0
- data/ext/sources/ggml/src/ggml-openvino/openvino/op/unary_gelu.cpp +25 -0
- data/ext/sources/ggml/src/ggml-openvino/openvino/op/unary_silu.cpp +27 -0
- data/ext/sources/ggml/src/ggml-openvino/openvino/op/view.cpp +53 -0
- data/ext/sources/ggml/src/ggml-openvino/openvino/op_table.cpp +47 -0
- data/ext/sources/ggml/src/ggml-openvino/openvino/op_table.h +40 -0
- data/ext/sources/ggml/src/ggml-openvino/openvino/pass/fuse_to_sdpa.cpp +60 -0
- data/ext/sources/ggml/src/ggml-openvino/openvino/pass/fuse_to_sdpa.h +17 -0
- data/ext/sources/ggml/src/ggml-openvino/openvino/pass/mark_decompression_convert_constant_folding.h +29 -0
- data/ext/sources/ggml/src/ggml-openvino/openvino/pass/squeeze_matmul.cpp +58 -0
- data/ext/sources/ggml/src/ggml-openvino/openvino/pass/squeeze_matmul.h +17 -0
- data/ext/sources/ggml/src/ggml-openvino/openvino/rt_info/weightless_caching_attributes.hpp +41 -0
- data/ext/sources/ggml/src/ggml-openvino/openvino/translate_session.cpp +317 -0
- data/ext/sources/ggml/src/ggml-openvino/openvino/translate_session.h +28 -0
- data/ext/sources/ggml/src/ggml-openvino/openvino/utils.cpp +257 -0
- data/ext/sources/ggml/src/ggml-openvino/openvino/utils.h +86 -0
- data/ext/sources/ggml/src/ggml-openvino/utils.cpp +880 -0
- data/ext/sources/ggml/src/ggml-openvino/utils.h +143 -0
- data/ext/sources/ggml/src/ggml-opt.cpp +1 -0
- data/ext/sources/ggml/src/ggml-quants.c +385 -119
- data/ext/sources/ggml/src/ggml-quants.h +6 -0
- data/ext/sources/ggml/src/ggml-rpc/CMakeLists.txt +24 -0
- data/ext/sources/ggml/src/ggml-rpc/ggml-rpc.cpp +167 -311
- data/ext/sources/ggml/src/ggml-rpc/transport.cpp +683 -0
- data/ext/sources/ggml/src/ggml-rpc/transport.h +34 -0
- data/ext/sources/ggml/src/ggml-sycl/CMakeLists.txt +64 -91
- data/ext/sources/ggml/src/ggml-sycl/add-id.cpp +5 -1
- data/ext/sources/ggml/src/ggml-sycl/backend.hpp +4 -1
- data/ext/sources/ggml/src/ggml-sycl/binbcast.cpp +21 -20
- data/ext/sources/ggml/src/ggml-sycl/common.cpp +74 -2
- data/ext/sources/ggml/src/ggml-sycl/common.hpp +356 -11
- data/ext/sources/ggml/src/ggml-sycl/convert.cpp +184 -14
- data/ext/sources/ggml/src/ggml-sycl/convert.hpp +31 -1
- data/ext/sources/ggml/src/ggml-sycl/count-equal.cpp +1 -1
- data/ext/sources/ggml/src/ggml-sycl/cumsum.cpp +148 -0
- data/ext/sources/ggml/src/ggml-sycl/cumsum.hpp +5 -0
- data/ext/sources/ggml/src/ggml-sycl/dequantize.hpp +663 -0
- data/ext/sources/ggml/src/ggml-sycl/diag.cpp +67 -0
- data/ext/sources/ggml/src/ggml-sycl/diag.hpp +5 -0
- data/ext/sources/ggml/src/ggml-sycl/dmmv.cpp +586 -6
- data/ext/sources/ggml/src/ggml-sycl/dpct/helper.hpp +791 -47
- data/ext/sources/ggml/src/ggml-sycl/element_wise.cpp +77 -156
- data/ext/sources/ggml/src/ggml-sycl/element_wise.hpp +2 -2
- data/ext/sources/ggml/src/ggml-sycl/fattn-buffers.cpp +56 -0
- data/ext/sources/ggml/src/ggml-sycl/fattn-buffers.hpp +63 -0
- data/ext/sources/ggml/src/ggml-sycl/fattn-common.hpp +1181 -0
- data/ext/sources/ggml/src/ggml-sycl/fattn-tile.cpp +59 -0
- data/ext/sources/ggml/src/ggml-sycl/fattn-tile.hpp +1246 -0
- data/ext/sources/ggml/src/ggml-sycl/fattn-vec.hpp +674 -0
- data/ext/sources/ggml/src/ggml-sycl/fattn.cpp +227 -0
- data/ext/sources/ggml/src/ggml-sycl/fattn.hpp +22 -0
- data/ext/sources/ggml/src/ggml-sycl/fill.cpp +55 -0
- data/ext/sources/ggml/src/ggml-sycl/fill.hpp +5 -0
- data/ext/sources/ggml/src/ggml-sycl/gated_delta_net.cpp +347 -0
- data/ext/sources/ggml/src/ggml-sycl/gated_delta_net.hpp +9 -0
- data/ext/sources/ggml/src/ggml-sycl/gemm.hpp +3 -0
- data/ext/sources/ggml/src/ggml-sycl/getrows.cpp +79 -3
- data/ext/sources/ggml/src/ggml-sycl/ggml-sycl.cpp +1134 -236
- data/ext/sources/ggml/src/ggml-sycl/im2col.cpp +353 -89
- data/ext/sources/ggml/src/ggml-sycl/im2col.hpp +5 -3
- data/ext/sources/ggml/src/ggml-sycl/mmvq.cpp +1344 -26
- data/ext/sources/ggml/src/ggml-sycl/mmvq.hpp +16 -0
- data/ext/sources/ggml/src/ggml-sycl/norm.cpp +65 -66
- data/ext/sources/ggml/src/ggml-sycl/outprod.cpp +3 -3
- data/ext/sources/ggml/src/ggml-sycl/pad.cpp +27 -27
- data/ext/sources/ggml/src/ggml-sycl/presets.hpp +3 -0
- data/ext/sources/ggml/src/ggml-sycl/quants.hpp +72 -1
- data/ext/sources/ggml/src/ggml-sycl/rope.cpp +450 -287
- data/ext/sources/ggml/src/ggml-sycl/rope.hpp +6 -0
- data/ext/sources/ggml/src/ggml-sycl/set_rows.cpp +7 -1
- data/ext/sources/ggml/src/ggml-sycl/softmax.cpp +6 -6
- data/ext/sources/ggml/src/ggml-sycl/solve_tri.cpp +172 -0
- data/ext/sources/ggml/src/ggml-sycl/solve_tri.hpp +8 -0
- data/ext/sources/ggml/src/ggml-sycl/ssm_conv.cpp +6 -1
- data/ext/sources/ggml/src/ggml-sycl/ssm_scan.cpp +156 -0
- data/ext/sources/ggml/src/ggml-sycl/ssm_scan.hpp +5 -0
- data/ext/sources/ggml/src/ggml-sycl/sycl_hw.cpp +62 -10
- data/ext/sources/ggml/src/ggml-sycl/sycl_hw.hpp +18 -6
- data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-tile-instance-dkq112-dv112.cpp +5 -0
- data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-tile-instance-dkq128-dv128.cpp +5 -0
- data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-tile-instance-dkq256-dv256.cpp +5 -0
- data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-tile-instance-dkq40-dv40.cpp +5 -0
- data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-tile-instance-dkq512-dv512.cpp +6 -0
- data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-tile-instance-dkq576-dv512.cpp +5 -0
- data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-tile-instance-dkq64-dv64.cpp +5 -0
- data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-tile-instance-dkq72-dv72.cpp +5 -0
- data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-tile-instance-dkq80-dv80.cpp +5 -0
- data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-tile-instance-dkq96-dv96.cpp +5 -0
- data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-f16-f16.cpp +8 -0
- data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-f16-q4_0.cpp +8 -0
- data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-f16-q4_1.cpp +8 -0
- data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-f16-q5_0.cpp +8 -0
- data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-f16-q5_1.cpp +8 -0
- data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-f16-q8_0.cpp +8 -0
- data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q4_0-f16.cpp +8 -0
- data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q4_0-q4_0.cpp +8 -0
- data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q4_0-q4_1.cpp +8 -0
- data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q4_0-q5_0.cpp +8 -0
- data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q4_0-q5_1.cpp +8 -0
- data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q4_0-q8_0.cpp +8 -0
- data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q4_1-f16.cpp +8 -0
- data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q4_1-q4_0.cpp +8 -0
- data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q4_1-q4_1.cpp +8 -0
- data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q4_1-q5_0.cpp +8 -0
- data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q4_1-q5_1.cpp +8 -0
- data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q4_1-q8_0.cpp +8 -0
- data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q5_0-f16.cpp +8 -0
- data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q5_0-q4_0.cpp +8 -0
- data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q5_0-q4_1.cpp +8 -0
- data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q5_0-q5_0.cpp +8 -0
- data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q5_0-q5_1.cpp +8 -0
- data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q5_0-q8_0.cpp +8 -0
- data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q5_1-f16.cpp +8 -0
- data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q5_1-q4_0.cpp +8 -0
- data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q5_1-q4_1.cpp +8 -0
- data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q5_1-q5_0.cpp +8 -0
- data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q5_1-q5_1.cpp +8 -0
- data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q5_1-q8_0.cpp +8 -0
- data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q8_0-f16.cpp +8 -0
- data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q8_0-q4_0.cpp +8 -0
- data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q8_0-q4_1.cpp +8 -0
- data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q8_0-q5_0.cpp +8 -0
- data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q8_0-q5_1.cpp +8 -0
- data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q8_0-q8_0.cpp +8 -0
- data/ext/sources/ggml/src/ggml-sycl/type.hpp +112 -0
- data/ext/sources/ggml/src/ggml-sycl/upscale.cpp +410 -0
- data/ext/sources/ggml/src/ggml-sycl/upscale.hpp +9 -0
- data/ext/sources/ggml/src/ggml-sycl/vecdotq.hpp +228 -53
- data/ext/sources/ggml/src/ggml-sycl/wkv.cpp +1 -1
- data/ext/sources/ggml/src/ggml-virtgpu/CMakeLists.txt +70 -0
- data/ext/sources/ggml/src/ggml-virtgpu/apir_cs_ggml-rpc-front.cpp +87 -0
- data/ext/sources/ggml/src/ggml-virtgpu/backend/CMakeLists.txt +21 -0
- data/ext/sources/ggml/src/ggml-virtgpu/backend/apir_cs_ggml-rpc-back.cpp +115 -0
- data/ext/sources/ggml/src/ggml-virtgpu/backend/backend-convert.h +13 -0
- data/ext/sources/ggml/src/ggml-virtgpu/backend/backend-dispatched-backend.cpp +102 -0
- data/ext/sources/ggml/src/ggml-virtgpu/backend/backend-dispatched-buffer-type.cpp +105 -0
- data/ext/sources/ggml/src/ggml-virtgpu/backend/backend-dispatched-buffer.cpp +179 -0
- data/ext/sources/ggml/src/ggml-virtgpu/backend/backend-dispatched-device.cpp +148 -0
- data/ext/sources/ggml/src/ggml-virtgpu/backend/backend-dispatched.cpp +51 -0
- data/ext/sources/ggml/src/ggml-virtgpu/backend/backend-dispatched.gen.h +73 -0
- data/ext/sources/ggml/src/ggml-virtgpu/backend/backend-dispatched.h +27 -0
- data/ext/sources/ggml/src/ggml-virtgpu/backend/backend-virgl-apir.h +32 -0
- data/ext/sources/ggml/src/ggml-virtgpu/backend/backend.cpp +144 -0
- data/ext/sources/ggml/src/ggml-virtgpu/backend/shared/api_remoting.h +95 -0
- data/ext/sources/ggml/src/ggml-virtgpu/backend/shared/apir_backend.gen.h +94 -0
- data/ext/sources/ggml/src/ggml-virtgpu/backend/shared/apir_backend.h +50 -0
- data/ext/sources/ggml/src/ggml-virtgpu/backend/shared/apir_cs.h +378 -0
- data/ext/sources/ggml/src/ggml-virtgpu/backend/shared/apir_cs_ggml.h +232 -0
- data/ext/sources/ggml/src/ggml-virtgpu/backend/shared/apir_cs_rpc.h +58 -0
- data/ext/sources/ggml/src/ggml-virtgpu/ggml-backend-buffer-type.cpp +81 -0
- data/ext/sources/ggml/src/ggml-virtgpu/ggml-backend-buffer.cpp +123 -0
- data/ext/sources/ggml/src/ggml-virtgpu/ggml-backend-device.cpp +160 -0
- data/ext/sources/ggml/src/ggml-virtgpu/ggml-backend-reg.cpp +213 -0
- data/ext/sources/ggml/src/ggml-virtgpu/ggml-backend.cpp +71 -0
- data/ext/sources/ggml/src/ggml-virtgpu/ggml-remoting.h +71 -0
- data/ext/sources/ggml/src/ggml-virtgpu/ggmlremoting_functions.yaml +166 -0
- data/ext/sources/ggml/src/ggml-virtgpu/include/apir_hw.h +9 -0
- data/ext/sources/ggml/src/ggml-virtgpu/virtgpu-apir.h +15 -0
- data/ext/sources/ggml/src/ggml-virtgpu/virtgpu-forward-backend.cpp +58 -0
- data/ext/sources/ggml/src/ggml-virtgpu/virtgpu-forward-buffer-type.cpp +110 -0
- data/ext/sources/ggml/src/ggml-virtgpu/virtgpu-forward-buffer.cpp +173 -0
- data/ext/sources/ggml/src/ggml-virtgpu/virtgpu-forward-device.cpp +192 -0
- data/ext/sources/ggml/src/ggml-virtgpu/virtgpu-forward-impl.h +36 -0
- data/ext/sources/ggml/src/ggml-virtgpu/virtgpu-forward.gen.h +53 -0
- data/ext/sources/ggml/src/ggml-virtgpu/virtgpu-shm.cpp +99 -0
- data/ext/sources/ggml/src/ggml-virtgpu/virtgpu-shm.h +23 -0
- data/ext/sources/ggml/src/ggml-virtgpu/virtgpu-utils.cpp +179 -0
- data/ext/sources/ggml/src/ggml-virtgpu/virtgpu-utils.h +86 -0
- data/ext/sources/ggml/src/ggml-virtgpu/virtgpu.cpp +545 -0
- data/ext/sources/ggml/src/ggml-virtgpu/virtgpu.h +115 -0
- data/ext/sources/ggml/src/ggml-vulkan/CMakeLists.txt +12 -1
- data/ext/sources/ggml/src/ggml-vulkan/ggml-vulkan.cpp +3250 -940
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/CMakeLists.txt +4 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/acc.comp +16 -8
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/contig_copy.comp +6 -2
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/conv2d_mm.comp +146 -13
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/copy.comp +3 -1
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/copy_from_quant.comp +1 -1
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/copy_to_quant.comp +25 -1
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/dequant_funcs.glsl +88 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/dequant_funcs_cm2.glsl +643 -1
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/dequant_nvfp4.comp +32 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/dequant_q1_0.comp +29 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/diag.comp +0 -1
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/dot_product_funcs.glsl +27 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/elu.comp +27 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/exp.comp +0 -1
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/feature-tests/coopmat2_decode_vector.comp +7 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/flash_attn.comp +533 -180
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/flash_attn_base.glsl +113 -68
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/flash_attn_cm1.comp +412 -222
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/flash_attn_cm2.comp +222 -83
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/flash_attn_dequant.glsl +131 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/flash_attn_mask_opt.comp +162 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/flash_attn_mmq_funcs.glsl +203 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/flash_attn_split_k_reduce.comp +9 -8
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/fwht.comp +115 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/gated_delta_net.comp +189 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/generic_binary_head.glsl +0 -1
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/glu_head.glsl +10 -1
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/glu_main.glsl +16 -6
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/im2col.comp +76 -54
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/im2col_3d.comp +0 -1
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/l2_norm.comp +12 -9
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/log.comp +0 -1
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec.comp +122 -27
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_base.glsl +20 -17
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_iface.glsl +6 -6
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_q2_k.comp +1 -1
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_q4_k.comp +1 -1
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_q5_k.comp +1 -1
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vecq.comp +1 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vecq_funcs.glsl +88 -55
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mm.comp +22 -20
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mm_cm2.comp +51 -14
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mm_funcs.glsl +159 -125
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mm_id_funcs.glsl +3 -1
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mmq.comp +5 -3
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mmq_funcs.glsl +8 -8
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mmq_shmem_types.glsl +24 -9
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/multi_add.comp +0 -1
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/rms_norm.comp +2 -3
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/rope_funcs.glsl +39 -63
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/rope_head.glsl +0 -1
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/rope_multi.comp +7 -4
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/rope_neox.comp +7 -4
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/rope_norm.comp +7 -4
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/rope_params.glsl +13 -7
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/rope_vision.comp +7 -4
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/sgn.comp +21 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/snake.comp +49 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/ssm_conv.comp +27 -11
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/tri.comp +0 -1
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/types.glsl +79 -2
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/vulkan-shaders-gen.cpp +193 -149
- data/ext/sources/ggml/src/ggml-webgpu/CMakeLists.txt +5 -2
- data/ext/sources/ggml/src/ggml-webgpu/ggml-webgpu-shader-lib.hpp +3221 -97
- data/ext/sources/ggml/src/ggml-webgpu/ggml-webgpu.cpp +3493 -1997
- data/ext/sources/ggml/src/ggml-webgpu/pre_wgsl.hpp +37 -7
- data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/add_id.wgsl +64 -0
- data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/argmax.wgsl +72 -0
- data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/argsort.wgsl +106 -0
- data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/argsort_merge.wgsl +134 -0
- data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/binary.wgsl +142 -0
- data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/common_decls.tmpl +115 -141
- data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/concat.wgsl +93 -0
- data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/conv2d.wgsl +165 -0
- data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/{cpy.tmpl.wgsl → cpy.wgsl} +25 -44
- data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/cumsum.wgsl +66 -0
- data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/flash_attn.wgsl +198 -230
- data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/flash_attn_quant_staging.tmpl +124 -0
- data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/flash_attn_tile.wgsl +397 -0
- data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/flash_attn_vec_blk.wgsl +101 -0
- data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/flash_attn_vec_reduce.wgsl +84 -0
- data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/flash_attn_vec_split.wgsl +619 -0
- data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/gated_delta_net.wgsl +149 -0
- data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/{get_rows.tmpl.wgsl → get_rows.wgsl} +234 -335
- data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/glu.wgsl +155 -0
- data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/im2col.wgsl +101 -0
- data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/mul_mat_decls.tmpl +871 -42
- data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/mul_mat_id.wgsl +195 -0
- data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/mul_mat_id_gather.wgsl +52 -0
- data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/mul_mat_id_vec.wgsl +154 -0
- data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/mul_mat_reg_tile.wgsl +149 -0
- data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/{mul_mat_subgroup_matrix.tmpl.wgsl → mul_mat_subgroup_matrix.wgsl} +36 -138
- data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/mul_mat_vec.wgsl +151 -0
- data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/mul_mat_vec_acc.tmpl +1432 -0
- data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/mul_mat_vec_q_acc.tmpl +303 -0
- data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/pad.wgsl +86 -0
- data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/quant_inner_loops.tmpl +21 -0
- data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/quantize_q8.wgsl +173 -0
- data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/repeat.wgsl +67 -0
- data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/rms_norm_mul.wgsl +152 -0
- data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/{rope.tmpl.wgsl → rope.wgsl} +71 -142
- data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/row_norm.wgsl +153 -0
- data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/{scale.tmpl.wgsl → scale.wgsl} +15 -40
- data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/set.wgsl +109 -0
- data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/set_rows.wgsl +39 -12
- data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/set_rows_quant.wgsl +224 -0
- data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/{soft_max.tmpl.wgsl → soft_max.wgsl} +106 -206
- data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/solve_tri.wgsl +121 -0
- data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/ssm_conv.wgsl +65 -0
- data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/ssm_scan.wgsl +193 -0
- data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/sum_rows.wgsl +55 -0
- data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/unary.wgsl +213 -0
- data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/upscale.wgsl +240 -0
- data/ext/sources/ggml/src/ggml-zdnn/ggml-zdnn.cpp +24 -15
- data/ext/sources/ggml/src/ggml-zendnn/CMakeLists.txt +31 -32
- data/ext/sources/ggml/src/ggml-zendnn/ggml-zendnn.cpp +253 -16
- data/ext/sources/ggml/src/ggml.c +268 -52
- data/ext/sources/ggml/src/gguf.cpp +377 -47
- data/ext/sources/include/parakeet.h +342 -0
- data/ext/sources/include/whisper.h +10 -0
- data/ext/sources/media/matmul.png +0 -0
- data/ext/sources/src/CMakeLists.txt +23 -0
- data/ext/sources/src/parakeet-arch.h +188 -0
- data/ext/sources/src/parakeet.cpp +3838 -0
- data/ext/sources/src/whisper.cpp +62 -40
- data/extsources.rb +26 -10
- data/lib/whisper/log_settable.rb +36 -0
- data/lib/whisper/model/uri.rb +13 -1
- data/lib/whisper/output.rb +74 -0
- data/sig/whisper.rbs +445 -55
- data/test/helper.rb +2 -0
- data/test/jfk_reader/jfk_reader.c +50 -7
- data/test/test_callback.rb +1 -0
- data/test/test_context_params.rb +82 -0
- data/test/test_package.rb +6 -5
- data/test/test_parakeet.rb +28 -0
- data/test/test_parakeet_callback.rb +107 -0
- data/test/test_parakeet_context.rb +116 -0
- data/test/test_parakeet_context_params.rb +24 -0
- data/test/test_parakeet_model.rb +21 -0
- data/test/test_parakeet_params.rb +78 -0
- data/test/test_parakeet_segment.rb +42 -0
- data/test/test_parakeet_token.rb +73 -0
- data/test/test_params.rb +2 -0
- data/test/test_token.rb +11 -0
- data/test/test_vad_context.rb +58 -8
- data/test/test_vad_segment.rb +1 -1
- data/test/test_whisper.rb +44 -6
- data/whispercpp.gemspec +2 -2
- metadata +426 -280
- data/ext/sources/bindings/javascript/CMakeLists.txt +0 -41
- data/ext/sources/bindings/javascript/emscripten.cpp +0 -93
- data/ext/sources/bindings/javascript/libwhisper.worker.js +0 -1
- data/ext/sources/bindings/javascript/package.json +0 -26
- data/ext/sources/bindings/javascript/whisper.js +0 -19
- data/ext/sources/examples/addon.node/CMakeLists.txt +0 -31
- data/ext/sources/examples/addon.node/__test__/whisper.spec.js +0 -133
- data/ext/sources/examples/addon.node/addon.cpp +0 -557
- data/ext/sources/examples/addon.node/index.js +0 -59
- data/ext/sources/examples/addon.node/package.json +0 -16
- data/ext/sources/examples/addon.node/vad-example.js +0 -132
- data/ext/sources/examples/bench.wasm/CMakeLists.txt +0 -49
- data/ext/sources/examples/bench.wasm/emscripten.cpp +0 -87
- data/ext/sources/examples/bench.wasm/index-tmpl.html +0 -285
- data/ext/sources/examples/coi-serviceworker.js +0 -146
- data/ext/sources/examples/command/CMakeLists.txt +0 -10
- data/ext/sources/examples/command/command.cpp +0 -802
- data/ext/sources/examples/command/commands.txt +0 -9
- data/ext/sources/examples/command.wasm/CMakeLists.txt +0 -50
- data/ext/sources/examples/command.wasm/emscripten.cpp +0 -327
- data/ext/sources/examples/command.wasm/index-tmpl.html +0 -415
- data/ext/sources/examples/generate-karaoke.sh +0 -57
- data/ext/sources/examples/helpers.js +0 -191
- data/ext/sources/examples/livestream.sh +0 -112
- data/ext/sources/examples/lsp/CMakeLists.txt +0 -10
- data/ext/sources/examples/lsp/lsp.cpp +0 -471
- data/ext/sources/examples/lsp/whisper.vim +0 -362
- data/ext/sources/examples/python/test_whisper_processor.py +0 -7
- data/ext/sources/examples/python/whisper_processor.py +0 -54
- data/ext/sources/examples/server/bench.js +0 -29
- data/ext/sources/examples/server.py +0 -120
- data/ext/sources/examples/stream/CMakeLists.txt +0 -10
- data/ext/sources/examples/stream/stream.cpp +0 -437
- data/ext/sources/examples/stream.wasm/CMakeLists.txt +0 -49
- data/ext/sources/examples/stream.wasm/emscripten.cpp +0 -216
- data/ext/sources/examples/stream.wasm/index-tmpl.html +0 -491
- data/ext/sources/examples/sycl/CMakeLists.txt +0 -9
- data/ext/sources/examples/sycl/build.sh +0 -22
- data/ext/sources/examples/sycl/ls-sycl-device.cpp +0 -11
- data/ext/sources/examples/sycl/run-whisper.sh +0 -17
- data/ext/sources/examples/talk-llama/CMakeLists.txt +0 -47
- data/ext/sources/examples/talk-llama/eleven-labs.py +0 -80
- data/ext/sources/examples/talk-llama/llama-adapter.cpp +0 -494
- data/ext/sources/examples/talk-llama/llama-adapter.h +0 -88
- data/ext/sources/examples/talk-llama/llama-arch.cpp +0 -2559
- data/ext/sources/examples/talk-llama/llama-arch.h +0 -586
- data/ext/sources/examples/talk-llama/llama-batch.cpp +0 -917
- data/ext/sources/examples/talk-llama/llama-batch.h +0 -173
- data/ext/sources/examples/talk-llama/llama-chat.cpp +0 -876
- data/ext/sources/examples/talk-llama/llama-chat.h +0 -70
- data/ext/sources/examples/talk-llama/llama-context.cpp +0 -3645
- data/ext/sources/examples/talk-llama/llama-context.h +0 -360
- data/ext/sources/examples/talk-llama/llama-cparams.cpp +0 -5
- data/ext/sources/examples/talk-llama/llama-cparams.h +0 -42
- data/ext/sources/examples/talk-llama/llama-grammar.cpp +0 -1464
- data/ext/sources/examples/talk-llama/llama-grammar.h +0 -194
- data/ext/sources/examples/talk-llama/llama-graph.cpp +0 -2282
- data/ext/sources/examples/talk-llama/llama-graph.h +0 -910
- data/ext/sources/examples/talk-llama/llama-hparams.cpp +0 -241
- data/ext/sources/examples/talk-llama/llama-hparams.h +0 -284
- data/ext/sources/examples/talk-llama/llama-impl.cpp +0 -171
- data/ext/sources/examples/talk-llama/llama-impl.h +0 -63
- data/ext/sources/examples/talk-llama/llama-io.cpp +0 -15
- data/ext/sources/examples/talk-llama/llama-io.h +0 -35
- data/ext/sources/examples/talk-llama/llama-kv-cache-iswa.cpp +0 -328
- data/ext/sources/examples/talk-llama/llama-kv-cache-iswa.h +0 -137
- data/ext/sources/examples/talk-llama/llama-kv-cache.cpp +0 -2100
- data/ext/sources/examples/talk-llama/llama-kv-cache.h +0 -390
- data/ext/sources/examples/talk-llama/llama-kv-cells.h +0 -533
- data/ext/sources/examples/talk-llama/llama-memory-hybrid.cpp +0 -268
- data/ext/sources/examples/talk-llama/llama-memory-hybrid.h +0 -139
- data/ext/sources/examples/talk-llama/llama-memory-recurrent.cpp +0 -1167
- data/ext/sources/examples/talk-llama/llama-memory-recurrent.h +0 -182
- data/ext/sources/examples/talk-llama/llama-memory.cpp +0 -59
- data/ext/sources/examples/talk-llama/llama-memory.h +0 -122
- data/ext/sources/examples/talk-llama/llama-mmap.cpp +0 -735
- data/ext/sources/examples/talk-llama/llama-mmap.h +0 -73
- data/ext/sources/examples/talk-llama/llama-model-loader.cpp +0 -1247
- data/ext/sources/examples/talk-llama/llama-model-loader.h +0 -176
- data/ext/sources/examples/talk-llama/llama-model-saver.cpp +0 -285
- data/ext/sources/examples/talk-llama/llama-model-saver.h +0 -37
- data/ext/sources/examples/talk-llama/llama-model.cpp +0 -8338
- data/ext/sources/examples/talk-llama/llama-model.h +0 -544
- data/ext/sources/examples/talk-llama/llama-quant.cpp +0 -1072
- data/ext/sources/examples/talk-llama/llama-quant.h +0 -1
- data/ext/sources/examples/talk-llama/llama-sampling.cpp +0 -3771
- data/ext/sources/examples/talk-llama/llama-sampling.h +0 -44
- data/ext/sources/examples/talk-llama/llama-vocab.cpp +0 -3900
- data/ext/sources/examples/talk-llama/llama-vocab.h +0 -182
- data/ext/sources/examples/talk-llama/llama.cpp +0 -1140
- data/ext/sources/examples/talk-llama/llama.h +0 -1540
- data/ext/sources/examples/talk-llama/models/afmoe.cpp +0 -191
- data/ext/sources/examples/talk-llama/models/apertus.cpp +0 -125
- data/ext/sources/examples/talk-llama/models/arcee.cpp +0 -135
- data/ext/sources/examples/talk-llama/models/arctic.cpp +0 -138
- data/ext/sources/examples/talk-llama/models/arwkv7.cpp +0 -86
- data/ext/sources/examples/talk-llama/models/baichuan.cpp +0 -122
- data/ext/sources/examples/talk-llama/models/bailingmoe.cpp +0 -144
- data/ext/sources/examples/talk-llama/models/bailingmoe2.cpp +0 -135
- data/ext/sources/examples/talk-llama/models/bert.cpp +0 -178
- data/ext/sources/examples/talk-llama/models/bitnet.cpp +0 -160
- data/ext/sources/examples/talk-llama/models/bloom.cpp +0 -101
- data/ext/sources/examples/talk-llama/models/chameleon.cpp +0 -178
- data/ext/sources/examples/talk-llama/models/chatglm.cpp +0 -132
- data/ext/sources/examples/talk-llama/models/codeshell.cpp +0 -111
- data/ext/sources/examples/talk-llama/models/cogvlm.cpp +0 -102
- data/ext/sources/examples/talk-llama/models/cohere2-iswa.cpp +0 -134
- data/ext/sources/examples/talk-llama/models/command-r.cpp +0 -122
- data/ext/sources/examples/talk-llama/models/dbrx.cpp +0 -123
- data/ext/sources/examples/talk-llama/models/deci.cpp +0 -135
- data/ext/sources/examples/talk-llama/models/deepseek.cpp +0 -144
- data/ext/sources/examples/talk-llama/models/deepseek2.cpp +0 -259
- data/ext/sources/examples/talk-llama/models/dots1.cpp +0 -134
- data/ext/sources/examples/talk-llama/models/dream.cpp +0 -105
- data/ext/sources/examples/talk-llama/models/ernie4-5-moe.cpp +0 -150
- data/ext/sources/examples/talk-llama/models/ernie4-5.cpp +0 -110
- data/ext/sources/examples/talk-llama/models/exaone.cpp +0 -114
- data/ext/sources/examples/talk-llama/models/exaone4.cpp +0 -123
- data/ext/sources/examples/talk-llama/models/falcon-h1.cpp +0 -113
- data/ext/sources/examples/talk-llama/models/falcon.cpp +0 -120
- data/ext/sources/examples/talk-llama/models/gemma-embedding.cpp +0 -116
- data/ext/sources/examples/talk-llama/models/gemma.cpp +0 -112
- data/ext/sources/examples/talk-llama/models/gemma2-iswa.cpp +0 -128
- data/ext/sources/examples/talk-llama/models/gemma3.cpp +0 -155
- data/ext/sources/examples/talk-llama/models/gemma3n-iswa.cpp +0 -384
- data/ext/sources/examples/talk-llama/models/glm4-moe.cpp +0 -170
- data/ext/sources/examples/talk-llama/models/glm4.cpp +0 -150
- data/ext/sources/examples/talk-llama/models/gpt2.cpp +0 -105
- data/ext/sources/examples/talk-llama/models/gptneox.cpp +0 -144
- data/ext/sources/examples/talk-llama/models/granite-hybrid.cpp +0 -196
- data/ext/sources/examples/talk-llama/models/granite.cpp +0 -211
- data/ext/sources/examples/talk-llama/models/graph-context-mamba.cpp +0 -283
- data/ext/sources/examples/talk-llama/models/grok.cpp +0 -159
- data/ext/sources/examples/talk-llama/models/grovemoe.cpp +0 -141
- data/ext/sources/examples/talk-llama/models/hunyuan-dense.cpp +0 -132
- data/ext/sources/examples/talk-llama/models/hunyuan-moe.cpp +0 -154
- data/ext/sources/examples/talk-llama/models/internlm2.cpp +0 -120
- data/ext/sources/examples/talk-llama/models/jais.cpp +0 -86
- data/ext/sources/examples/talk-llama/models/jamba.cpp +0 -106
- data/ext/sources/examples/talk-llama/models/lfm2.cpp +0 -175
- data/ext/sources/examples/talk-llama/models/llada-moe.cpp +0 -122
- data/ext/sources/examples/talk-llama/models/llada.cpp +0 -99
- data/ext/sources/examples/talk-llama/models/llama-iswa.cpp +0 -178
- data/ext/sources/examples/talk-llama/models/llama.cpp +0 -168
- data/ext/sources/examples/talk-llama/models/maincoder.cpp +0 -117
- data/ext/sources/examples/talk-llama/models/mamba.cpp +0 -55
- data/ext/sources/examples/talk-llama/models/mimo2-iswa.cpp +0 -123
- data/ext/sources/examples/talk-llama/models/minicpm3.cpp +0 -199
- data/ext/sources/examples/talk-llama/models/minimax-m2.cpp +0 -124
- data/ext/sources/examples/talk-llama/models/mistral3.cpp +0 -160
- data/ext/sources/examples/talk-llama/models/models.h +0 -569
- data/ext/sources/examples/talk-llama/models/modern-bert.cpp +0 -116
- data/ext/sources/examples/talk-llama/models/mpt.cpp +0 -126
- data/ext/sources/examples/talk-llama/models/nemotron-h.cpp +0 -150
- data/ext/sources/examples/talk-llama/models/nemotron.cpp +0 -122
- data/ext/sources/examples/talk-llama/models/neo-bert.cpp +0 -104
- data/ext/sources/examples/talk-llama/models/olmo.cpp +0 -121
- data/ext/sources/examples/talk-llama/models/olmo2.cpp +0 -150
- data/ext/sources/examples/talk-llama/models/olmoe.cpp +0 -124
- data/ext/sources/examples/talk-llama/models/openai-moe-iswa.cpp +0 -127
- data/ext/sources/examples/talk-llama/models/openelm.cpp +0 -124
- data/ext/sources/examples/talk-llama/models/orion.cpp +0 -123
- data/ext/sources/examples/talk-llama/models/pangu-embedded.cpp +0 -121
- data/ext/sources/examples/talk-llama/models/phi2.cpp +0 -121
- data/ext/sources/examples/talk-llama/models/phi3.cpp +0 -152
- data/ext/sources/examples/talk-llama/models/plamo.cpp +0 -110
- data/ext/sources/examples/talk-llama/models/plamo2.cpp +0 -316
- data/ext/sources/examples/talk-llama/models/plamo3.cpp +0 -128
- data/ext/sources/examples/talk-llama/models/plm.cpp +0 -168
- data/ext/sources/examples/talk-llama/models/qwen.cpp +0 -108
- data/ext/sources/examples/talk-llama/models/qwen2.cpp +0 -126
- data/ext/sources/examples/talk-llama/models/qwen2moe.cpp +0 -151
- data/ext/sources/examples/talk-llama/models/qwen2vl.cpp +0 -117
- data/ext/sources/examples/talk-llama/models/qwen3.cpp +0 -117
- data/ext/sources/examples/talk-llama/models/qwen3moe.cpp +0 -124
- data/ext/sources/examples/talk-llama/models/qwen3next.cpp +0 -873
- data/ext/sources/examples/talk-llama/models/qwen3vl-moe.cpp +0 -149
- data/ext/sources/examples/talk-llama/models/qwen3vl.cpp +0 -141
- data/ext/sources/examples/talk-llama/models/refact.cpp +0 -94
- data/ext/sources/examples/talk-llama/models/rnd1.cpp +0 -126
- data/ext/sources/examples/talk-llama/models/rwkv6-base.cpp +0 -162
- data/ext/sources/examples/talk-llama/models/rwkv6.cpp +0 -94
- data/ext/sources/examples/talk-llama/models/rwkv6qwen2.cpp +0 -86
- data/ext/sources/examples/talk-llama/models/rwkv7-base.cpp +0 -135
- data/ext/sources/examples/talk-llama/models/rwkv7.cpp +0 -90
- data/ext/sources/examples/talk-llama/models/seed-oss.cpp +0 -124
- data/ext/sources/examples/talk-llama/models/smallthinker.cpp +0 -126
- data/ext/sources/examples/talk-llama/models/smollm3.cpp +0 -128
- data/ext/sources/examples/talk-llama/models/stablelm.cpp +0 -146
- data/ext/sources/examples/talk-llama/models/starcoder.cpp +0 -100
- data/ext/sources/examples/talk-llama/models/starcoder2.cpp +0 -121
- data/ext/sources/examples/talk-llama/models/t5-dec.cpp +0 -166
- data/ext/sources/examples/talk-llama/models/t5-enc.cpp +0 -96
- data/ext/sources/examples/talk-llama/models/wavtokenizer-dec.cpp +0 -149
- data/ext/sources/examples/talk-llama/models/xverse.cpp +0 -108
- data/ext/sources/examples/talk-llama/prompts/talk-alpaca.txt +0 -23
- data/ext/sources/examples/talk-llama/speak +0 -40
- data/ext/sources/examples/talk-llama/speak.bat +0 -1
- data/ext/sources/examples/talk-llama/speak.ps1 +0 -14
- data/ext/sources/examples/talk-llama/talk-llama.cpp +0 -813
- data/ext/sources/examples/talk-llama/unicode-data.cpp +0 -7034
- data/ext/sources/examples/talk-llama/unicode-data.h +0 -20
- data/ext/sources/examples/talk-llama/unicode.cpp +0 -1147
- data/ext/sources/examples/talk-llama/unicode.h +0 -111
- data/ext/sources/examples/wchess/CMakeLists.txt +0 -10
- data/ext/sources/examples/wchess/libwchess/CMakeLists.txt +0 -19
- data/ext/sources/examples/wchess/libwchess/Chessboard.cpp +0 -803
- data/ext/sources/examples/wchess/libwchess/Chessboard.h +0 -33
- data/ext/sources/examples/wchess/libwchess/WChess.cpp +0 -193
- data/ext/sources/examples/wchess/libwchess/WChess.h +0 -63
- data/ext/sources/examples/wchess/libwchess/test-chessboard.cpp +0 -117
- data/ext/sources/examples/wchess/wchess.cmd/CMakeLists.txt +0 -8
- data/ext/sources/examples/wchess/wchess.cmd/wchess.cmd.cpp +0 -253
- data/ext/sources/examples/whisper.wasm/CMakeLists.txt +0 -50
- data/ext/sources/examples/whisper.wasm/emscripten.cpp +0 -118
- data/ext/sources/examples/whisper.wasm/index-tmpl.html +0 -659
- data/ext/sources/ggml/cmake/BuildTypes.cmake +0 -54
- data/ext/sources/ggml/src/ggml-cpu/llamafile/sgemm-ppc.h +0 -333
- data/ext/sources/ggml/src/ggml-cuda/template-instances/generate_cu_files.py +0 -99
- data/ext/sources/ggml/src/ggml-hexagon/htp/htp-dma.h +0 -157
- data/ext/sources/ggml/src/ggml-hexagon/htp/htp-msg.h +0 -165
- data/ext/sources/ggml/src/ggml-hexagon/htp/hvx-exp.c +0 -94
- data/ext/sources/ggml/src/ggml-hexagon/htp/hvx-inverse.c +0 -72
- data/ext/sources/ggml/src/ggml-hexagon/htp/hvx-sigmoid.c +0 -49
- data/ext/sources/ggml/src/ggml-hexagon/htp/hvx-utils.c +0 -1020
- data/ext/sources/ggml/src/ggml-hexagon/htp/ops-utils.h +0 -149
- data/ext/sources/ggml/src/ggml-hexagon/htp-utils.c +0 -454
- data/ext/sources/ggml/src/ggml-hexagon/htp-utils.h +0 -221
- data/ext/sources/ggml/src/ggml-hexagon/op-desc.h +0 -153
- data/ext/sources/ggml/src/ggml-opencl/kernels/embed_kernel.py +0 -26
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/rte.glsl +0 -5
- data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/bin_op.tmpl.wgsl +0 -188
- data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/binary_head.tmpl +0 -45
- data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/embed_wgsl.py +0 -147
- data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/glu.tmpl.wgsl +0 -323
- data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/mul_mat.tmpl.wgsl +0 -907
- data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/mul_mat_reg_tile.tmpl.wgsl +0 -247
- data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/mul_mat_vec.tmpl.wgsl +0 -267
- data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/rms_norm.wgsl +0 -123
- data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/set_rows.tmpl.wgsl +0 -112
- data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/unary_op.wgsl +0 -483
- data/ext/sources/tests/CMakeLists.txt +0 -112
- data/ext/sources/tests/earnings21/eval.mk +0 -58
- data/ext/sources/tests/earnings21/eval.py +0 -68
- data/ext/sources/tests/earnings21/normalizers/__init__.py +0 -2
- data/ext/sources/tests/earnings21/normalizers/basic.py +0 -80
- data/ext/sources/tests/earnings21/normalizers/english.json +0 -1741
- data/ext/sources/tests/earnings21/normalizers/english.py +0 -550
- data/ext/sources/tests/earnings21/requirements.txt +0 -6
- data/ext/sources/tests/en-0-ref.txt +0 -1
- data/ext/sources/tests/en-1-ref.txt +0 -1
- data/ext/sources/tests/en-2-ref.txt +0 -1
- data/ext/sources/tests/es-0-ref.txt +0 -1
- data/ext/sources/tests/librispeech/eval.mk +0 -39
- data/ext/sources/tests/librispeech/eval.py +0 -47
- data/ext/sources/tests/librispeech/normalizers/__init__.py +0 -2
- data/ext/sources/tests/librispeech/normalizers/basic.py +0 -80
- data/ext/sources/tests/librispeech/normalizers/english.json +0 -1741
- data/ext/sources/tests/librispeech/normalizers/english.py +0 -550
- data/ext/sources/tests/librispeech/requirements.txt +0 -6
- data/ext/sources/tests/run-tests.sh +0 -130
- data/ext/sources/tests/test-c.c +0 -3
- data/ext/sources/tests/test-vad-full.cpp +0 -56
- data/ext/sources/tests/test-vad.cpp +0 -83
- data/ext/sources/tests/test-whisper.js +0 -58
- data/lib/whisper/context.rb +0 -15
- data/lib/whisper/segment.rb +0 -58
|
@@ -0,0 +1,16 @@
|
|
|
1
|
+
#ifndef VTCM_UTILS_H
|
|
2
|
+
#define VTCM_UTILS_H
|
|
3
|
+
|
|
4
|
+
#include "hex-utils.h"
|
|
5
|
+
|
|
6
|
+
#include <assert.h>
|
|
7
|
+
#include <stdint.h>
|
|
8
|
+
#include <hexagon_types.h>
|
|
9
|
+
|
|
10
|
+
static inline uint8_t *vtcm_seq_alloc(uint8_t **vtcm_ptr, size_t size) {
|
|
11
|
+
uint8_t *p = *vtcm_ptr;
|
|
12
|
+
*vtcm_ptr += size;
|
|
13
|
+
return p;
|
|
14
|
+
}
|
|
15
|
+
|
|
16
|
+
#endif // VTCM_UTILS_H
|
|
@@ -7,10 +7,6 @@
|
|
|
7
7
|
#include <stdlib.h>
|
|
8
8
|
#include <string.h>
|
|
9
9
|
|
|
10
|
-
#ifdef HTP_DEBUG
|
|
11
|
-
# define FARF_HIGH 1
|
|
12
|
-
#endif
|
|
13
|
-
|
|
14
10
|
#include "HAP_farf.h"
|
|
15
11
|
|
|
16
12
|
#define WORKER_THREAD_STACK_SZ (2 * 16384)
|
|
@@ -60,7 +56,7 @@ static void worker_pool_main(void * context) {
|
|
|
60
56
|
unsigned int n = atomic_load(&pool->n_jobs);
|
|
61
57
|
unsigned int i = atomic_fetch_add(&pool->next_job, 1);
|
|
62
58
|
if (i >= n) {
|
|
63
|
-
//
|
|
59
|
+
// Spurious wakeup
|
|
64
60
|
continue;
|
|
65
61
|
}
|
|
66
62
|
|
|
@@ -0,0 +1,418 @@
|
|
|
1
|
+
// sample drv interface
|
|
2
|
+
|
|
3
|
+
#pragma clang diagnostic ignored "-Wgnu-anonymous-struct"
|
|
4
|
+
#pragma clang diagnostic ignored "-Wmissing-prototypes"
|
|
5
|
+
#pragma clang diagnostic ignored "-Wsign-compare"
|
|
6
|
+
|
|
7
|
+
#include <filesystem>
|
|
8
|
+
#include <set>
|
|
9
|
+
#include <sstream>
|
|
10
|
+
#include <string>
|
|
11
|
+
#ifdef _WIN32
|
|
12
|
+
# define WIN32_LEAN_AND_MEAN
|
|
13
|
+
# ifndef NOMINMAX
|
|
14
|
+
# define NOMINMAX
|
|
15
|
+
# endif
|
|
16
|
+
# include <windows.h>
|
|
17
|
+
# include <winevt.h>
|
|
18
|
+
#else
|
|
19
|
+
# include <dlfcn.h>
|
|
20
|
+
# include <unistd.h>
|
|
21
|
+
#endif
|
|
22
|
+
#include "ggml-impl.h"
|
|
23
|
+
#include "htp-drv.h"
|
|
24
|
+
#include "libdl.h"
|
|
25
|
+
|
|
26
|
+
#include <domain.h>
|
|
27
|
+
|
|
28
|
+
//
|
|
29
|
+
// Driver API types
|
|
30
|
+
//
|
|
31
|
+
|
|
32
|
+
typedef void * (*rpcmem_alloc_pfn_t)(int heapid, uint32_t flags, int size);
|
|
33
|
+
typedef void * (*rpcmem_alloc2_pfn_t)(int heapid, uint32_t flags, size_t size);
|
|
34
|
+
typedef void (*rpcmem_free_pfn_t)(void * po);
|
|
35
|
+
typedef int (*rpcmem_to_fd_pfn_t)(void * po);
|
|
36
|
+
|
|
37
|
+
typedef AEEResult (*dspqueue_create_pfn_t)(int domain,
|
|
38
|
+
uint32_t flags,
|
|
39
|
+
uint32_t req_queue_size,
|
|
40
|
+
uint32_t resp_queue_size,
|
|
41
|
+
dspqueue_callback_t packet_callback,
|
|
42
|
+
dspqueue_callback_t error_callback,
|
|
43
|
+
void * callback_context,
|
|
44
|
+
dspqueue_t * queue);
|
|
45
|
+
typedef AEEResult (*dspqueue_close_pfn_t)(dspqueue_t queue);
|
|
46
|
+
typedef AEEResult (*dspqueue_export_pfn_t)(dspqueue_t queue, uint64_t *queue_id);
|
|
47
|
+
typedef AEEResult (*dspqueue_write_pfn_t)(dspqueue_t queue, uint32_t flags,
|
|
48
|
+
uint32_t num_buffers,
|
|
49
|
+
struct dspqueue_buffer *buffers,
|
|
50
|
+
uint32_t message_length,
|
|
51
|
+
const uint8_t *message,
|
|
52
|
+
uint32_t timeout_us);
|
|
53
|
+
typedef AEEResult (*dspqueue_read_pfn_t)(dspqueue_t queue, uint32_t *flags,
|
|
54
|
+
uint32_t max_buffers, uint32_t *num_buffers,
|
|
55
|
+
struct dspqueue_buffer *buffers,
|
|
56
|
+
uint32_t max_message_length,
|
|
57
|
+
uint32_t *message_length, uint8_t *message,
|
|
58
|
+
uint32_t timeout_us);
|
|
59
|
+
|
|
60
|
+
typedef int (*fastrpc_mmap_pfn_t)(int domain, int fd, void *addr, int offset, size_t length, enum fastrpc_map_flags flags);
|
|
61
|
+
typedef int (*fastrpc_munmap_pfn_t)(int domain, int fd, void *addr, size_t length);
|
|
62
|
+
|
|
63
|
+
typedef int (*remote_handle64_open_pfn_t)(const char* name, remote_handle64 *ph);
|
|
64
|
+
typedef int (*remote_handle64_invoke_pfn_t)(remote_handle64 h, uint32_t dwScalars, remote_arg *pra);
|
|
65
|
+
typedef int (*remote_handle64_close_pfn_t)(remote_handle h);
|
|
66
|
+
typedef int (*remote_handle_control_pfn_t)(uint32_t req, void* data, uint32_t datalen);
|
|
67
|
+
typedef int (*remote_handle64_control_pfn_t)(remote_handle64 h, uint32_t req, void* data, uint32_t datalen);
|
|
68
|
+
typedef int (*remote_session_control_pfn_t)(uint32_t req, void *data, uint32_t datalen);
|
|
69
|
+
|
|
70
|
+
//
|
|
71
|
+
// Driver API pfns
|
|
72
|
+
//
|
|
73
|
+
|
|
74
|
+
rpcmem_alloc_pfn_t rpcmem_alloc_pfn = nullptr;
|
|
75
|
+
rpcmem_alloc2_pfn_t rpcmem_alloc2_pfn = nullptr;
|
|
76
|
+
rpcmem_free_pfn_t rpcmem_free_pfn = nullptr;
|
|
77
|
+
rpcmem_to_fd_pfn_t rpcmem_to_fd_pfn = nullptr;
|
|
78
|
+
|
|
79
|
+
fastrpc_mmap_pfn_t fastrpc_mmap_pfn = nullptr;
|
|
80
|
+
fastrpc_munmap_pfn_t fastrpc_munmap_pfn = nullptr;
|
|
81
|
+
|
|
82
|
+
dspqueue_create_pfn_t dspqueue_create_pfn = nullptr;
|
|
83
|
+
dspqueue_close_pfn_t dspqueue_close_pfn = nullptr;
|
|
84
|
+
dspqueue_export_pfn_t dspqueue_export_pfn = nullptr;
|
|
85
|
+
dspqueue_write_pfn_t dspqueue_write_pfn = nullptr;
|
|
86
|
+
dspqueue_read_pfn_t dspqueue_read_pfn = nullptr;
|
|
87
|
+
|
|
88
|
+
remote_handle64_open_pfn_t remote_handle64_open_pfn = nullptr;
|
|
89
|
+
remote_handle64_invoke_pfn_t remote_handle64_invoke_pfn = nullptr;
|
|
90
|
+
remote_handle64_close_pfn_t remote_handle64_close_pfn = nullptr;
|
|
91
|
+
remote_handle_control_pfn_t remote_handle_control_pfn = nullptr;
|
|
92
|
+
remote_handle64_control_pfn_t remote_handle64_control_pfn = nullptr;
|
|
93
|
+
remote_session_control_pfn_t remote_session_control_pfn = nullptr;
|
|
94
|
+
|
|
95
|
+
//
|
|
96
|
+
// Driver API
|
|
97
|
+
//
|
|
98
|
+
|
|
99
|
+
void * rpcmem_alloc(int heapid, uint32_t flags, int size) {
|
|
100
|
+
return rpcmem_alloc_pfn(heapid, flags, size);
|
|
101
|
+
}
|
|
102
|
+
|
|
103
|
+
void * rpcmem_alloc2(int heapid, uint32_t flags, size_t size) {
|
|
104
|
+
if (rpcmem_alloc2_pfn) {
|
|
105
|
+
return rpcmem_alloc2_pfn(heapid, flags, size);
|
|
106
|
+
} else {
|
|
107
|
+
GGML_LOG_INFO("ggml-hex: rpcmem_alloc2 not found, falling back to rpcmem_alloc\n");
|
|
108
|
+
return rpcmem_alloc_pfn(heapid, flags, size);
|
|
109
|
+
}
|
|
110
|
+
}
|
|
111
|
+
|
|
112
|
+
void rpcmem_free(void * po) {
|
|
113
|
+
return rpcmem_free_pfn(po);
|
|
114
|
+
}
|
|
115
|
+
|
|
116
|
+
int rpcmem_to_fd(void * po) {
|
|
117
|
+
return rpcmem_to_fd_pfn(po);
|
|
118
|
+
}
|
|
119
|
+
|
|
120
|
+
HTPDRV_API int fastrpc_mmap(int domain, int fd, void * addr, int offset, size_t length, enum fastrpc_map_flags flags) {
|
|
121
|
+
return fastrpc_mmap_pfn(domain, fd, addr, offset, length, flags);
|
|
122
|
+
}
|
|
123
|
+
|
|
124
|
+
HTPDRV_API int fastrpc_munmap(int domain, int fd, void * addr, size_t length) {
|
|
125
|
+
return fastrpc_munmap_pfn(domain, fd, addr, length);
|
|
126
|
+
}
|
|
127
|
+
|
|
128
|
+
AEEResult dspqueue_create(int domain,
|
|
129
|
+
uint32_t flags,
|
|
130
|
+
uint32_t req_queue_size,
|
|
131
|
+
uint32_t resp_queue_size,
|
|
132
|
+
dspqueue_callback_t packet_callback,
|
|
133
|
+
dspqueue_callback_t error_callback,
|
|
134
|
+
void * callback_context,
|
|
135
|
+
dspqueue_t * queue) {
|
|
136
|
+
return dspqueue_create_pfn(domain, flags, req_queue_size, resp_queue_size, packet_callback, error_callback,
|
|
137
|
+
callback_context, queue);
|
|
138
|
+
}
|
|
139
|
+
|
|
140
|
+
AEEResult dspqueue_close(dspqueue_t queue) {
|
|
141
|
+
return dspqueue_close_pfn(queue);
|
|
142
|
+
}
|
|
143
|
+
|
|
144
|
+
AEEResult dspqueue_export(dspqueue_t queue, uint64_t * queue_id) {
|
|
145
|
+
return dspqueue_export_pfn(queue, queue_id);
|
|
146
|
+
}
|
|
147
|
+
|
|
148
|
+
AEEResult dspqueue_write(dspqueue_t queue,
|
|
149
|
+
uint32_t flags,
|
|
150
|
+
uint32_t num_buffers,
|
|
151
|
+
struct dspqueue_buffer * buffers,
|
|
152
|
+
uint32_t message_length,
|
|
153
|
+
const uint8_t * message,
|
|
154
|
+
uint32_t timeout_us) {
|
|
155
|
+
return dspqueue_write_pfn(queue, flags, num_buffers, buffers, message_length, message, timeout_us);
|
|
156
|
+
}
|
|
157
|
+
|
|
158
|
+
AEEResult dspqueue_read(dspqueue_t queue,
|
|
159
|
+
uint32_t * flags,
|
|
160
|
+
uint32_t max_buffers,
|
|
161
|
+
uint32_t * num_buffers,
|
|
162
|
+
struct dspqueue_buffer * buffers,
|
|
163
|
+
uint32_t max_message_length,
|
|
164
|
+
uint32_t * message_length,
|
|
165
|
+
uint8_t * message,
|
|
166
|
+
uint32_t timeout_us) {
|
|
167
|
+
return dspqueue_read_pfn(queue, flags, max_buffers, num_buffers, buffers, max_message_length, message_length,
|
|
168
|
+
message, timeout_us);
|
|
169
|
+
}
|
|
170
|
+
|
|
171
|
+
HTPDRV_API int remote_handle64_open(const char * name, remote_handle64 * ph) {
|
|
172
|
+
return remote_handle64_open_pfn(name, ph);
|
|
173
|
+
}
|
|
174
|
+
|
|
175
|
+
HTPDRV_API int remote_handle64_invoke(remote_handle64 h, uint32_t dwScalars, remote_arg * pra) {
|
|
176
|
+
return remote_handle64_invoke_pfn(h, dwScalars, pra);
|
|
177
|
+
}
|
|
178
|
+
|
|
179
|
+
HTPDRV_API int remote_handle64_close(remote_handle64 h) {
|
|
180
|
+
return remote_handle64_close_pfn(h);
|
|
181
|
+
}
|
|
182
|
+
|
|
183
|
+
HTPDRV_API int remote_handle_control(uint32_t req, void * data, uint32_t datalen) {
|
|
184
|
+
return remote_handle_control_pfn(req, data, datalen);
|
|
185
|
+
}
|
|
186
|
+
|
|
187
|
+
HTPDRV_API int remote_handle64_control(remote_handle64 h, uint32_t req, void * data, uint32_t datalen) {
|
|
188
|
+
return remote_handle64_control_pfn(h, req, data, datalen);
|
|
189
|
+
}
|
|
190
|
+
|
|
191
|
+
HTPDRV_API int remote_session_control(uint32_t req, void * data, uint32_t datalen) {
|
|
192
|
+
return remote_session_control_pfn(req, data, datalen);
|
|
193
|
+
}
|
|
194
|
+
|
|
195
|
+
#ifdef _WIN32
|
|
196
|
+
|
|
197
|
+
static std::string wstr_to_str(std::wstring_view wstr) {
|
|
198
|
+
std::string result;
|
|
199
|
+
if (wstr.empty()) {
|
|
200
|
+
return result;
|
|
201
|
+
}
|
|
202
|
+
auto bytes_needed = WideCharToMultiByte(CP_UTF8, WC_ERR_INVALID_CHARS,
|
|
203
|
+
wstr.data(), (int) wstr.size(),
|
|
204
|
+
nullptr, 0, nullptr, nullptr);
|
|
205
|
+
if (bytes_needed == 0) {
|
|
206
|
+
GGML_LOG_ERROR("ggml-hex: WideCharToMultiByte failed. Error %lu\n", GetLastError());
|
|
207
|
+
throw std::runtime_error("Invalid wstring input");
|
|
208
|
+
}
|
|
209
|
+
|
|
210
|
+
result.resize(bytes_needed, '\0');
|
|
211
|
+
int bytes_written = WideCharToMultiByte(CP_UTF8, WC_ERR_INVALID_CHARS,
|
|
212
|
+
wstr.data(), (int) wstr.size(),
|
|
213
|
+
result.data(), bytes_needed,
|
|
214
|
+
nullptr, nullptr);
|
|
215
|
+
if (bytes_written == 0) {
|
|
216
|
+
GGML_LOG_ERROR("ggml-hex: WideCharToMultiByte failed. Error %lu\n", GetLastError());
|
|
217
|
+
throw std::runtime_error("Wstring conversion failed");
|
|
218
|
+
}
|
|
219
|
+
return result;
|
|
220
|
+
}
|
|
221
|
+
|
|
222
|
+
static std::string get_driver_path() {
|
|
223
|
+
std::wstring serviceName = L"qcnspmcdm";
|
|
224
|
+
std::string result;
|
|
225
|
+
|
|
226
|
+
// Get a handle to the SCM database.
|
|
227
|
+
SC_HANDLE schSCManager = OpenSCManagerW(NULL, NULL, STANDARD_RIGHTS_READ);
|
|
228
|
+
if (nullptr == schSCManager) {
|
|
229
|
+
GGML_LOG_ERROR("ggml-hex: Failed to open SCManager. Error: %lu\n", GetLastError());
|
|
230
|
+
return result;
|
|
231
|
+
}
|
|
232
|
+
|
|
233
|
+
// Get a handle to the service.
|
|
234
|
+
SC_HANDLE schService = OpenServiceW(schSCManager, // SCM database
|
|
235
|
+
serviceName.c_str(), // name of service
|
|
236
|
+
SERVICE_QUERY_CONFIG); // need query config access
|
|
237
|
+
|
|
238
|
+
if (nullptr == schService) {
|
|
239
|
+
GGML_LOG_ERROR("ggml-hex: Failed to open qcnspmcdm service. Error: %lu\n", GetLastError());
|
|
240
|
+
CloseServiceHandle(schSCManager);
|
|
241
|
+
return result;
|
|
242
|
+
}
|
|
243
|
+
|
|
244
|
+
// Store the size of buffer used as an output.
|
|
245
|
+
DWORD bufferSize;
|
|
246
|
+
if (!QueryServiceConfigW(schService, NULL, 0, &bufferSize) &&
|
|
247
|
+
(GetLastError() != ERROR_INSUFFICIENT_BUFFER)) {
|
|
248
|
+
GGML_LOG_ERROR("ggml-hex: Failed to query service config. Error: %lu\n", GetLastError());
|
|
249
|
+
CloseServiceHandle(schService);
|
|
250
|
+
CloseServiceHandle(schSCManager);
|
|
251
|
+
return result;
|
|
252
|
+
}
|
|
253
|
+
// Get the configuration of the service.
|
|
254
|
+
LPQUERY_SERVICE_CONFIGW serviceConfig =
|
|
255
|
+
static_cast<LPQUERY_SERVICE_CONFIGW>(LocalAlloc(LMEM_FIXED, bufferSize));
|
|
256
|
+
if (!QueryServiceConfigW(schService, serviceConfig, bufferSize, &bufferSize)) {
|
|
257
|
+
fprintf(stderr, "ggml-hex: Failed to query service config. Error: %lu\n", GetLastError());
|
|
258
|
+
LocalFree(serviceConfig);
|
|
259
|
+
CloseServiceHandle(schService);
|
|
260
|
+
CloseServiceHandle(schSCManager);
|
|
261
|
+
return result;
|
|
262
|
+
}
|
|
263
|
+
|
|
264
|
+
// Read the driver file path get its parent directory
|
|
265
|
+
std::wstring driverPath = std::wstring(serviceConfig->lpBinaryPathName);
|
|
266
|
+
driverPath = driverPath.substr(0, driverPath.find_last_of(L"\\"));
|
|
267
|
+
|
|
268
|
+
// Clean up resources
|
|
269
|
+
LocalFree(serviceConfig);
|
|
270
|
+
CloseServiceHandle(schService);
|
|
271
|
+
CloseServiceHandle(schSCManager);
|
|
272
|
+
|
|
273
|
+
// Driver path would contain invalid path string, like:
|
|
274
|
+
// \SystemRoot\System32\DriverStore\FileRepository\qcadsprpc8280.inf_arm64_c2b9460c9a072f37
|
|
275
|
+
// "\SystemRoot" should be replace with a correct one (e.g. C:\Windows)
|
|
276
|
+
const std::wstring systemRootPlaceholder = L"\\SystemRoot";
|
|
277
|
+
if (0 != driverPath.compare(0, systemRootPlaceholder.length(), systemRootPlaceholder)) {
|
|
278
|
+
GGML_LOG_ERROR("ggml-hex: String pattern not found in driver path.\n");
|
|
279
|
+
return result;
|
|
280
|
+
}
|
|
281
|
+
|
|
282
|
+
// Replace \SystemRoot with an absolute path from system ENV windir
|
|
283
|
+
const std::wstring systemRootEnv = L"windir";
|
|
284
|
+
|
|
285
|
+
// Query the number of wide characters this variable requires
|
|
286
|
+
DWORD numWords = GetEnvironmentVariableW(systemRootEnv.c_str(), NULL, 0);
|
|
287
|
+
if (numWords == 0) {
|
|
288
|
+
GGML_LOG_ERROR("ggml-hex: Failed get systemRoot environment variable\n");
|
|
289
|
+
return result;
|
|
290
|
+
}
|
|
291
|
+
|
|
292
|
+
// Query the actual system root name from environment variable
|
|
293
|
+
std::vector<wchar_t> systemRoot(numWords + 1);
|
|
294
|
+
numWords = GetEnvironmentVariableW(systemRootEnv.c_str(), systemRoot.data(), numWords + 1);
|
|
295
|
+
if (numWords == 0) {
|
|
296
|
+
GGML_LOG_ERROR("ggml-hex: Failed to read windir environment variable\n");
|
|
297
|
+
return result;
|
|
298
|
+
}
|
|
299
|
+
driverPath.replace(0, systemRootPlaceholder.length(), std::wstring(systemRoot.data()));
|
|
300
|
+
|
|
301
|
+
return wstr_to_str(driverPath);
|
|
302
|
+
}
|
|
303
|
+
|
|
304
|
+
#endif
|
|
305
|
+
|
|
306
|
+
using dl_handle_ptr = std::unique_ptr<dl_handle, dl_handle_deleter>;
|
|
307
|
+
|
|
308
|
+
int htpdrv_init() {
|
|
309
|
+
static dl_handle_ptr lib_cdsp_rpc_handle = nullptr;
|
|
310
|
+
static bool initialized = false;
|
|
311
|
+
#ifdef _WIN32
|
|
312
|
+
std::string drv_path = get_driver_path() + "\\" + "libcdsprpc.dll";
|
|
313
|
+
#else
|
|
314
|
+
std::string drv_path = "libcdsprpc.so";
|
|
315
|
+
#endif
|
|
316
|
+
if (initialized) {
|
|
317
|
+
GGML_LOG_INFO("ggml-hex: Driver already loaded\n");
|
|
318
|
+
return AEE_SUCCESS;
|
|
319
|
+
}
|
|
320
|
+
GGML_LOG_INFO("ggml-hex: Loading driver %s\n", drv_path.c_str());
|
|
321
|
+
|
|
322
|
+
fs::path path{ drv_path.c_str() };
|
|
323
|
+
dl_handle_ptr handle { dl_load_library(path) };
|
|
324
|
+
if (!handle) {
|
|
325
|
+
GGML_LOG_ERROR("ggml-hex: failed to load %s: %s\n", path.u8string().c_str(), dl_error());
|
|
326
|
+
return AEE_EUNABLETOLOAD;
|
|
327
|
+
}
|
|
328
|
+
|
|
329
|
+
#define dlsym(drv, type, pfn, symbol, ignore) \
|
|
330
|
+
do { \
|
|
331
|
+
pfn = (type) dl_get_sym(drv, #symbol); \
|
|
332
|
+
if (!ignore && nullptr == pfn) { \
|
|
333
|
+
GGML_LOG_ERROR("ggml-hex: failed to dlsym %s\n", #symbol); \
|
|
334
|
+
return AEE_EUNABLETOLOAD; \
|
|
335
|
+
} \
|
|
336
|
+
} while (0)
|
|
337
|
+
|
|
338
|
+
dlsym(handle.get(), rpcmem_alloc_pfn_t, rpcmem_alloc_pfn, rpcmem_alloc, false);
|
|
339
|
+
dlsym(handle.get(), rpcmem_alloc2_pfn_t, rpcmem_alloc2_pfn, rpcmem_alloc2, true);
|
|
340
|
+
dlsym(handle.get(), rpcmem_free_pfn_t, rpcmem_free_pfn, rpcmem_free, false);
|
|
341
|
+
dlsym(handle.get(), rpcmem_to_fd_pfn_t, rpcmem_to_fd_pfn, rpcmem_to_fd, false);
|
|
342
|
+
dlsym(handle.get(), fastrpc_mmap_pfn_t, fastrpc_mmap_pfn, fastrpc_mmap, false);
|
|
343
|
+
dlsym(handle.get(), fastrpc_munmap_pfn_t, fastrpc_munmap_pfn, fastrpc_munmap, false);
|
|
344
|
+
dlsym(handle.get(), dspqueue_create_pfn_t, dspqueue_create_pfn, dspqueue_create, false);
|
|
345
|
+
dlsym(handle.get(), dspqueue_close_pfn_t, dspqueue_close_pfn, dspqueue_close, false);
|
|
346
|
+
dlsym(handle.get(), dspqueue_export_pfn_t, dspqueue_export_pfn, dspqueue_export, false);
|
|
347
|
+
dlsym(handle.get(), dspqueue_write_pfn_t, dspqueue_write_pfn, dspqueue_write, false);
|
|
348
|
+
dlsym(handle.get(), dspqueue_read_pfn_t, dspqueue_read_pfn, dspqueue_read, false);
|
|
349
|
+
dlsym(handle.get(), remote_handle64_open_pfn_t, remote_handle64_open_pfn, remote_handle64_open, false);
|
|
350
|
+
dlsym(handle.get(), remote_handle64_invoke_pfn_t, remote_handle64_invoke_pfn, remote_handle64_invoke, false);
|
|
351
|
+
dlsym(handle.get(), remote_handle_control_pfn_t, remote_handle_control_pfn, remote_handle_control, false);
|
|
352
|
+
dlsym(handle.get(), remote_handle64_control_pfn_t, remote_handle64_control_pfn, remote_handle64_control, false);
|
|
353
|
+
dlsym(handle.get(), remote_session_control_pfn_t, remote_session_control_pfn, remote_session_control, false);
|
|
354
|
+
dlsym(handle.get(), remote_handle64_close_pfn_t, remote_handle64_close_pfn, remote_handle64_close, false);
|
|
355
|
+
|
|
356
|
+
lib_cdsp_rpc_handle = std::move(handle);
|
|
357
|
+
initialized = true;
|
|
358
|
+
|
|
359
|
+
return AEE_SUCCESS;
|
|
360
|
+
}
|
|
361
|
+
|
|
362
|
+
domain * get_domain(int domain_id) {
|
|
363
|
+
int i = 0;
|
|
364
|
+
int size = sizeof(supported_domains) / sizeof(domain);
|
|
365
|
+
|
|
366
|
+
for (i = 0; i < size; i++) {
|
|
367
|
+
if (supported_domains[i].id == domain_id) {
|
|
368
|
+
return &supported_domains[i];
|
|
369
|
+
}
|
|
370
|
+
}
|
|
371
|
+
|
|
372
|
+
return NULL;
|
|
373
|
+
}
|
|
374
|
+
|
|
375
|
+
int get_hex_arch_ver(int domain, int * arch) {
|
|
376
|
+
if (!remote_handle_control_pfn) {
|
|
377
|
+
GGML_LOG_ERROR("ggml-hex: remote_handle_control is not supported on this device\n");
|
|
378
|
+
return AEE_EUNSUPPORTEDAPI;
|
|
379
|
+
}
|
|
380
|
+
|
|
381
|
+
struct remote_dsp_capability arch_ver;
|
|
382
|
+
arch_ver.domain = (uint32_t) domain;
|
|
383
|
+
arch_ver.attribute_ID = ARCH_VER;
|
|
384
|
+
arch_ver.capability = (uint32_t) 0;
|
|
385
|
+
|
|
386
|
+
int err = remote_handle_control(DSPRPC_GET_DSP_INFO, &arch_ver, sizeof(arch_ver));
|
|
387
|
+
if ((err & 0xff) == (AEE_EUNSUPPORTEDAPI & 0xff)) {
|
|
388
|
+
GGML_LOG_ERROR("ggml-hex: FastRPC capability API is not supported on this device\n");
|
|
389
|
+
return AEE_EUNSUPPORTEDAPI;
|
|
390
|
+
}
|
|
391
|
+
|
|
392
|
+
if (err != AEE_SUCCESS) {
|
|
393
|
+
GGML_LOG_ERROR("ggml-hex: FastRPC capability query failed (err %d)\n", err);
|
|
394
|
+
return err;
|
|
395
|
+
}
|
|
396
|
+
|
|
397
|
+
switch (arch_ver.capability & 0xff) {
|
|
398
|
+
case 0x68:
|
|
399
|
+
*arch = 68;
|
|
400
|
+
return 0;
|
|
401
|
+
case 0x69:
|
|
402
|
+
*arch = 69;
|
|
403
|
+
return 0;
|
|
404
|
+
case 0x73:
|
|
405
|
+
*arch = 73;
|
|
406
|
+
return 0;
|
|
407
|
+
case 0x75:
|
|
408
|
+
*arch = 75;
|
|
409
|
+
return 0;
|
|
410
|
+
case 0x79:
|
|
411
|
+
*arch = 79;
|
|
412
|
+
return 0;
|
|
413
|
+
case 0x81:
|
|
414
|
+
*arch = 81;
|
|
415
|
+
return 0;
|
|
416
|
+
}
|
|
417
|
+
return -1;
|
|
418
|
+
}
|
|
@@ -0,0 +1,121 @@
|
|
|
1
|
+
#pragma once
|
|
2
|
+
|
|
3
|
+
#ifdef __cplusplus
|
|
4
|
+
extern "C" {
|
|
5
|
+
#endif
|
|
6
|
+
|
|
7
|
+
#ifdef _WIN32
|
|
8
|
+
# pragma clang diagnostic ignored "-Wignored-attributes"
|
|
9
|
+
#endif
|
|
10
|
+
|
|
11
|
+
#include <AEEStdErr.h>
|
|
12
|
+
#include <rpcmem.h>
|
|
13
|
+
#include <remote.h>
|
|
14
|
+
#include <dspqueue.h>
|
|
15
|
+
|
|
16
|
+
#if defined(_WIN32) && !defined(__MINGW32__)
|
|
17
|
+
# ifdef GGML_BACKEND_BUILD
|
|
18
|
+
# define HTPDRV_API __declspec(dllexport) extern
|
|
19
|
+
# else
|
|
20
|
+
# define HTPDRV_API __declspec(dllimport) extern
|
|
21
|
+
# endif
|
|
22
|
+
#else
|
|
23
|
+
# define HTPDRV_API __attribute__ ((visibility ("default"))) extern
|
|
24
|
+
#endif
|
|
25
|
+
|
|
26
|
+
/* Offset to differentiate HLOS and Hexagon error codes.
|
|
27
|
+
Stores the value of AEE_EOFFSET for Hexagon. */
|
|
28
|
+
#ifndef DSP_OFFSET
|
|
29
|
+
# define DSP_OFFSET 0x80000400
|
|
30
|
+
#endif
|
|
31
|
+
|
|
32
|
+
/* Errno for connection reset by peer. */
|
|
33
|
+
#ifndef ECONNRESET
|
|
34
|
+
# ifdef __hexagon__
|
|
35
|
+
# define ECONNRESET 104
|
|
36
|
+
# endif
|
|
37
|
+
#endif
|
|
38
|
+
|
|
39
|
+
/* Abstraction of different OS specific sleep APIs.
|
|
40
|
+
SLEEP accepts input in seconds. */
|
|
41
|
+
#ifndef SLEEP
|
|
42
|
+
# ifdef __hexagon__
|
|
43
|
+
# define SLEEP(x) \
|
|
44
|
+
{ /* Do nothing for simulator. */ \
|
|
45
|
+
}
|
|
46
|
+
# else
|
|
47
|
+
# ifdef _WIN32
|
|
48
|
+
# define SLEEP(x) Sleep(1000 * x) /* Sleep accepts input in milliseconds. */
|
|
49
|
+
# else
|
|
50
|
+
# define SLEEP(x) sleep(x) /* sleep accepts input in seconds. */
|
|
51
|
+
# endif
|
|
52
|
+
# endif
|
|
53
|
+
#endif
|
|
54
|
+
|
|
55
|
+
/* Include windows specific header files. */
|
|
56
|
+
#ifdef _WIN32
|
|
57
|
+
# include <windows.h>
|
|
58
|
+
# include <sysinfoapi.h>
|
|
59
|
+
# define _CRT_SECURE_NO_WARNINGS 1
|
|
60
|
+
# define _WINSOCK_DEPRECATED_NO_WARNINGS 1
|
|
61
|
+
#endif
|
|
62
|
+
|
|
63
|
+
/* Includes and defines for all HLOS except windows */
|
|
64
|
+
#if !defined(__hexagon__) && !defined(_WIN32)
|
|
65
|
+
# include "unistd.h"
|
|
66
|
+
|
|
67
|
+
# include <sys/time.h>
|
|
68
|
+
#endif
|
|
69
|
+
|
|
70
|
+
/* Includes and defines for Hexagon and all HLOS except Windows. */
|
|
71
|
+
#if !defined(_WIN32)
|
|
72
|
+
/* Weak reference to remote symbol for compilation. */
|
|
73
|
+
# pragma weak remote_session_control
|
|
74
|
+
# pragma weak remote_handle_control
|
|
75
|
+
# pragma weak remote_handle64_control
|
|
76
|
+
# pragma weak fastrpc_mmap
|
|
77
|
+
# pragma weak fastrpc_munmap
|
|
78
|
+
# pragma weak rpcmem_alloc2
|
|
79
|
+
#endif
|
|
80
|
+
|
|
81
|
+
#if !defined(_WIN32)
|
|
82
|
+
# pragma weak remote_system_request
|
|
83
|
+
#endif
|
|
84
|
+
|
|
85
|
+
#ifdef _WIN32
|
|
86
|
+
# define DSPQUEUE_TIMEOUT DSPQUEUE_TIMEOUT_NONE
|
|
87
|
+
#else
|
|
88
|
+
# define DSPQUEUE_TIMEOUT 1000000
|
|
89
|
+
#endif
|
|
90
|
+
|
|
91
|
+
/**
|
|
92
|
+
* htpdrv_init API: driver interface entry point
|
|
93
|
+
*
|
|
94
|
+
* @return Return AEE error codes as defined in Hexagon SDK.
|
|
95
|
+
*/
|
|
96
|
+
HTPDRV_API int htpdrv_init(void);
|
|
97
|
+
|
|
98
|
+
/**
|
|
99
|
+
* get_domain API: get domain struct from domain value.
|
|
100
|
+
*
|
|
101
|
+
* @param[in] domain value of a domain
|
|
102
|
+
* @return Returns domain struct of the domain if it is supported or else
|
|
103
|
+
* returns NULL.
|
|
104
|
+
*
|
|
105
|
+
*/
|
|
106
|
+
HTPDRV_API domain * get_domain(int domain_id);
|
|
107
|
+
|
|
108
|
+
/**
|
|
109
|
+
* get_hex_arch_ver API: query the Hexagon processor architecture version information
|
|
110
|
+
*
|
|
111
|
+
* @param[in] domain_id value of a domain
|
|
112
|
+
* @param[out] Arch version (73, 75, ...)
|
|
113
|
+
* @return 0 if query is successful.
|
|
114
|
+
* non-zero if error, return value points to the error.
|
|
115
|
+
*
|
|
116
|
+
*/
|
|
117
|
+
HTPDRV_API int get_hex_arch_ver(int domain, int * arch);
|
|
118
|
+
|
|
119
|
+
#ifdef __cplusplus
|
|
120
|
+
}
|
|
121
|
+
#endif
|