npm - @fugood/llama.node - Versions diffs - 0.3.16 → 0.4.0 - Mend

@fugood/llama.node 0.3.16 → 0.4.0

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.

Files changed (281) hide show

package/CMakeLists.txt +6 -1
package/bin/darwin/arm64/llama-node.node +0 -0
package/bin/darwin/x64/llama-node.node +0 -0
package/bin/linux/arm64/llama-node.node +0 -0
package/bin/linux/x64/llama-node.node +0 -0
package/bin/linux-cuda/arm64/llama-node.node +0 -0
package/bin/linux-cuda/x64/llama-node.node +0 -0
package/bin/linux-vulkan/arm64/llama-node.node +0 -0
package/bin/linux-vulkan/x64/llama-node.node +0 -0
package/bin/win32/arm64/llama-node.node +0 -0
package/bin/win32/arm64/node.lib +0 -0
package/bin/win32/x64/llama-node.node +0 -0
package/bin/win32/x64/node.lib +0 -0
package/bin/win32-vulkan/arm64/llama-node.node +0 -0
package/bin/win32-vulkan/arm64/node.lib +0 -0
package/bin/win32-vulkan/x64/llama-node.node +0 -0
package/bin/win32-vulkan/x64/node.lib +0 -0
package/lib/binding.ts +44 -2
package/lib/index.js +132 -1
package/lib/index.ts +203 -3
package/package.json +2 -1
package/src/EmbeddingWorker.cpp +1 -1
package/src/LlamaCompletionWorker.cpp +374 -19
package/src/LlamaCompletionWorker.h +31 -10
package/src/LlamaContext.cpp +216 -7
package/src/LlamaContext.h +12 -0
package/src/common.hpp +15 -0
package/src/llama.cpp/.github/workflows/build-linux-cross.yml +233 -0
package/src/llama.cpp/.github/workflows/build.yml +89 -767
package/src/llama.cpp/.github/workflows/docker.yml +9 -6
package/src/llama.cpp/.github/workflows/release.yml +716 -0
package/src/llama.cpp/.github/workflows/server.yml +19 -23
package/src/llama.cpp/CMakeLists.txt +11 -1
package/src/llama.cpp/cmake/build-info.cmake +8 -2
package/src/llama.cpp/cmake/x64-windows-llvm.cmake +0 -6
package/src/llama.cpp/common/CMakeLists.txt +35 -4
package/src/llama.cpp/common/arg.cpp +844 -121
package/src/llama.cpp/common/arg.h +9 -0
package/src/llama.cpp/common/chat.cpp +129 -107
package/src/llama.cpp/common/chat.h +2 -0
package/src/llama.cpp/common/common.cpp +64 -518
package/src/llama.cpp/common/common.h +35 -45
package/src/llama.cpp/common/json-schema-to-grammar.cpp +3 -0
package/src/llama.cpp/common/llguidance.cpp +31 -47
package/src/llama.cpp/common/minja/chat-template.hpp +23 -11
package/src/llama.cpp/common/minja/minja.hpp +186 -127
package/src/llama.cpp/common/regex-partial.cpp +204 -0
package/src/llama.cpp/common/regex-partial.h +56 -0
package/src/llama.cpp/common/sampling.cpp +60 -50
package/src/llama.cpp/docs/build.md +122 -7
package/src/llama.cpp/examples/CMakeLists.txt +2 -32
package/src/llama.cpp/examples/batched/batched.cpp +1 -1
package/src/llama.cpp/examples/embedding/embedding.cpp +9 -12
package/src/llama.cpp/examples/gritlm/gritlm.cpp +1 -1
package/src/llama.cpp/examples/llama.android/llama/build.gradle.kts +1 -0
package/src/llama.cpp/examples/parallel/parallel.cpp +89 -15
package/src/llama.cpp/examples/passkey/passkey.cpp +1 -1
package/src/llama.cpp/examples/speculative/speculative.cpp +1 -1
package/src/llama.cpp/examples/speculative-simple/speculative-simple.cpp +1 -1
package/src/llama.cpp/examples/sycl/build.sh +2 -2
package/src/llama.cpp/examples/sycl/win-build-sycl.bat +2 -2
package/src/llama.cpp/examples/training/CMakeLists.txt +5 -0
package/src/llama.cpp/examples/training/finetune.cpp +96 -0
package/src/llama.cpp/ggml/CMakeLists.txt +35 -2
package/src/llama.cpp/ggml/cmake/GitVars.cmake +22 -0
package/src/llama.cpp/ggml/include/ggml-backend.h +4 -4
package/src/llama.cpp/ggml/include/ggml-cpp.h +1 -1
package/src/llama.cpp/ggml/include/ggml-cpu.h +5 -0
package/src/llama.cpp/ggml/include/ggml-opt.h +47 -28
package/src/llama.cpp/ggml/include/ggml-rpc.h +6 -1
package/src/llama.cpp/ggml/include/ggml.h +76 -106
package/src/llama.cpp/ggml/src/CMakeLists.txt +11 -8
package/src/llama.cpp/ggml/src/ggml-alloc.c +4 -1
package/src/llama.cpp/ggml/src/ggml-backend.cpp +9 -5
package/src/llama.cpp/ggml/src/ggml-cann/CMakeLists.txt +0 -2
package/src/llama.cpp/ggml/src/ggml-cann/acl_tensor.cpp +8 -4
package/src/llama.cpp/ggml/src/ggml-cann/acl_tensor.h +5 -5
package/src/llama.cpp/ggml/src/ggml-cann/aclnn_ops.cpp +692 -1534
package/src/llama.cpp/ggml/src/ggml-cann/aclnn_ops.h +613 -122
package/src/llama.cpp/ggml/src/ggml-cann/common.h +135 -1
package/src/llama.cpp/ggml/src/ggml-cann/ggml-cann.cpp +507 -137
package/src/llama.cpp/ggml/src/ggml-common.h +12 -6
package/src/llama.cpp/ggml/src/ggml-cpu/CMakeLists.txt +66 -33
package/src/llama.cpp/ggml/src/ggml-cpu/binary-ops.cpp +158 -0
package/src/llama.cpp/ggml/src/ggml-cpu/binary-ops.h +16 -0
package/src/llama.cpp/ggml/src/ggml-cpu/common.h +72 -0
package/src/llama.cpp/ggml/src/ggml-cpu/cpu-feats-x86.cpp +1 -1
package/src/llama.cpp/ggml/src/ggml-cpu/ggml-cpu-aarch64.cpp +896 -194
package/src/llama.cpp/ggml/src/ggml-cpu/ggml-cpu-impl.h +2 -21
package/src/llama.cpp/ggml/src/ggml-cpu/ggml-cpu-quants.c +1060 -410
package/src/llama.cpp/ggml/src/ggml-cpu/ggml-cpu.c +1008 -13533
package/src/llama.cpp/ggml/src/ggml-cpu/ggml-cpu.cpp +31 -16
package/src/llama.cpp/ggml/src/ggml-cpu/kleidiai/kernels.cpp +90 -12
package/src/llama.cpp/ggml/src/ggml-cpu/kleidiai/kernels.h +47 -13
package/src/llama.cpp/ggml/src/ggml-cpu/kleidiai/kleidiai.cpp +266 -72
package/src/llama.cpp/ggml/src/ggml-cpu/llamafile/sgemm.cpp +1034 -88
package/src/llama.cpp/ggml/src/ggml-cpu/ops.cpp +8796 -0
package/src/llama.cpp/ggml/src/ggml-cpu/ops.h +110 -0
package/src/llama.cpp/ggml/src/ggml-cpu/simd-mappings.h +892 -0
package/src/llama.cpp/ggml/src/ggml-cpu/unary-ops.cpp +186 -0
package/src/llama.cpp/ggml/src/ggml-cpu/unary-ops.h +28 -0
package/src/llama.cpp/ggml/src/ggml-cpu/vec.cpp +252 -0
package/src/llama.cpp/ggml/src/ggml-cpu/vec.h +802 -0
package/src/llama.cpp/ggml/src/ggml-cuda/CMakeLists.txt +23 -4
package/src/llama.cpp/ggml/src/ggml-cuda/vendors/hip.h +7 -0
package/src/llama.cpp/ggml/src/ggml-cuda/vendors/musa.h +1 -0
package/src/llama.cpp/ggml/src/ggml-hip/CMakeLists.txt +0 -4
package/src/llama.cpp/ggml/src/ggml-impl.h +52 -18
package/src/llama.cpp/ggml/src/ggml-metal/ggml-metal-impl.h +106 -14
package/src/llama.cpp/ggml/src/ggml-opencl/CMakeLists.txt +67 -119
package/src/llama.cpp/ggml/src/ggml-opencl/ggml-opencl.cpp +1023 -262
package/src/llama.cpp/ggml/src/ggml-opt.cpp +368 -190
package/src/llama.cpp/ggml/src/ggml-quants.c +0 -6
package/src/llama.cpp/ggml/src/ggml-rpc/ggml-rpc.cpp +307 -40
package/src/llama.cpp/ggml/src/ggml-sycl/CMakeLists.txt +125 -45
package/src/llama.cpp/ggml/src/ggml-sycl/backend.hpp +10 -8
package/src/llama.cpp/ggml/src/ggml-sycl/binbcast.cpp +239 -0
package/src/llama.cpp/ggml/src/ggml-sycl/binbcast.hpp +39 -0
package/src/llama.cpp/ggml/src/ggml-sycl/common.cpp +0 -35
package/src/llama.cpp/ggml/src/ggml-sycl/common.hpp +9 -307
package/src/llama.cpp/ggml/src/ggml-sycl/convert.cpp +72 -25
package/src/llama.cpp/ggml/src/ggml-sycl/convert.hpp +14 -7
package/src/llama.cpp/ggml/src/ggml-sycl/dequantize.hpp +59 -21
package/src/llama.cpp/ggml/src/ggml-sycl/dmmv.cpp +7 -1
package/src/llama.cpp/ggml/src/ggml-sycl/dpct/helper.hpp +79 -90
package/src/llama.cpp/ggml/src/ggml-sycl/element_wise.cpp +944 -438
package/src/llama.cpp/ggml/src/ggml-sycl/element_wise.hpp +22 -23
package/src/llama.cpp/ggml/src/ggml-sycl/gemm.hpp +37 -8
package/src/llama.cpp/ggml/src/ggml-sycl/getrows.cpp +24 -20
package/src/llama.cpp/ggml/src/ggml-sycl/getrows.hpp +1 -4
package/src/llama.cpp/ggml/src/ggml-sycl/ggml-sycl.cpp +507 -411
package/src/llama.cpp/ggml/src/ggml-sycl/im2col.cpp +84 -74
package/src/llama.cpp/ggml/src/ggml-sycl/im2col.hpp +1 -3
package/src/llama.cpp/ggml/src/ggml-sycl/mmvq.cpp +185 -89
package/src/llama.cpp/ggml/src/ggml-sycl/norm.cpp +37 -49
package/src/llama.cpp/ggml/src/ggml-sycl/norm.hpp +7 -22
package/src/llama.cpp/ggml/src/ggml-sycl/outprod.cpp +4 -14
package/src/llama.cpp/ggml/src/ggml-sycl/quants.hpp +83 -0
package/src/llama.cpp/ggml/src/ggml-sycl/rope.cpp +204 -118
package/src/llama.cpp/ggml/src/ggml-sycl/rope.hpp +1 -3
package/src/llama.cpp/ggml/src/ggml-sycl/vecdotq.hpp +128 -53
package/src/llama.cpp/ggml/src/ggml-vulkan/CMakeLists.txt +83 -49
package/src/llama.cpp/ggml/src/ggml-vulkan/ggml-vulkan.cpp +1278 -282
package/src/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/CMakeLists.txt +32 -0
package/src/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/vulkan-shaders-gen.cpp +133 -30
package/src/llama.cpp/ggml/src/ggml.c +170 -265
package/src/llama.cpp/ggml/src/gguf.cpp +34 -33
package/src/llama.cpp/include/llama.h +82 -22
package/src/llama.cpp/models/ggml-vocab-llama4.gguf.inp +112 -0
package/src/llama.cpp/models/ggml-vocab-llama4.gguf.out +46 -0
package/src/llama.cpp/models/ggml-vocab-pixtral.gguf.inp +112 -0
package/src/llama.cpp/models/ggml-vocab-pixtral.gguf.out +46 -0
package/src/llama.cpp/requirements/requirements-all.txt +5 -3
package/src/llama.cpp/requirements/requirements-gguf_editor_gui.txt +3 -0
package/src/llama.cpp/scripts/xxd.cmake +1 -1
package/src/llama.cpp/src/CMakeLists.txt +4 -2
package/src/llama.cpp/src/llama-adapter.cpp +43 -1
package/src/llama.cpp/src/llama-arch.cpp +163 -17
package/src/llama.cpp/src/llama-arch.h +16 -0
package/src/llama.cpp/src/llama-batch.cpp +5 -1
package/src/llama.cpp/src/llama-batch.h +2 -1
package/src/llama.cpp/src/llama-chat.cpp +91 -16
package/src/llama.cpp/src/llama-chat.h +7 -2
package/src/llama.cpp/src/llama-context.cpp +479 -575
package/src/llama.cpp/src/llama-context.h +44 -33
package/src/llama.cpp/src/llama-cparams.h +1 -0
package/src/llama.cpp/src/llama-graph.cpp +209 -157
package/src/llama.cpp/src/llama-graph.h +38 -14
package/src/llama.cpp/src/llama-hparams.h +13 -0
package/src/llama.cpp/src/llama-kv-cache.cpp +1604 -543
package/src/llama.cpp/src/llama-kv-cache.h +283 -171
package/src/llama.cpp/src/llama-memory.h +12 -2
package/src/llama.cpp/src/llama-mmap.cpp +1 -1
package/src/llama.cpp/src/llama-model-loader.cpp +34 -20
package/src/llama.cpp/src/llama-model-loader.h +5 -3
package/src/llama.cpp/src/llama-model-saver.cpp +281 -0
package/src/llama.cpp/src/llama-model-saver.h +37 -0
package/src/llama.cpp/src/llama-model.cpp +1803 -330
package/src/llama.cpp/src/llama-model.h +21 -2
package/src/llama.cpp/src/llama-quant.cpp +33 -10
package/src/llama.cpp/src/llama-sampling.cpp +25 -7
package/src/llama.cpp/src/llama-vocab.cpp +86 -10
package/src/llama.cpp/src/llama-vocab.h +6 -0
package/src/llama.cpp/src/llama.cpp +15 -1
package/src/llama.cpp/tests/CMakeLists.txt +52 -31
package/src/llama.cpp/tests/test-arg-parser.cpp +51 -4
package/src/llama.cpp/tests/test-backend-ops.cpp +189 -90
package/src/llama.cpp/tests/test-chat-template.cpp +26 -6
package/src/llama.cpp/tests/test-chat.cpp +15 -3
package/src/llama.cpp/{examples/gbnf-validator/gbnf-validator.cpp → tests/test-gbnf-validator.cpp} +2 -2
package/src/llama.cpp/tests/test-grammar-integration.cpp +3 -2
package/src/llama.cpp/tests/test-grammar-llguidance.cpp +63 -2
package/src/llama.cpp/tests/test-grammar-parser.cpp +3 -1
package/src/llama.cpp/tests/test-json-schema-to-grammar.cpp +17 -1
package/src/llama.cpp/tests/test-llama-grammar.cpp +2 -1
package/src/llama.cpp/tests/test-mtmd-c-api.c +63 -0
package/src/llama.cpp/tests/test-opt.cpp +33 -21
package/src/llama.cpp/{examples/quantize-stats/quantize-stats.cpp → tests/test-quantize-stats.cpp} +3 -1
package/src/llama.cpp/tests/test-regex-partial.cpp +288 -0
package/src/llama.cpp/tests/test-sampling.cpp +1 -1
package/src/llama.cpp/tests/test-tokenizer-1-bpe.cpp +2 -1
package/src/llama.cpp/tests/test-tokenizer-1-spm.cpp +2 -1
package/src/llama.cpp/tools/CMakeLists.txt +39 -0
package/src/llama.cpp/{examples → tools}/batched-bench/batched-bench.cpp +3 -3
package/src/llama.cpp/{examples → tools}/export-lora/export-lora.cpp +1 -1
package/src/llama.cpp/{examples → tools}/gguf-split/gguf-split.cpp +15 -16
package/src/llama.cpp/{examples → tools}/imatrix/imatrix.cpp +11 -9
package/src/llama.cpp/{examples → tools}/llama-bench/llama-bench.cpp +623 -274
package/src/llama.cpp/{examples → tools}/main/main.cpp +22 -14
package/src/llama.cpp/tools/mtmd/CMakeLists.txt +47 -0
package/src/llama.cpp/tools/mtmd/clip-impl.h +365 -0
package/src/llama.cpp/tools/mtmd/clip.cpp +3646 -0
package/src/llama.cpp/tools/mtmd/clip.h +99 -0
package/src/llama.cpp/tools/mtmd/deprecation-warning.cpp +22 -0
package/src/llama.cpp/tools/mtmd/mtmd-cli.cpp +370 -0
package/src/llama.cpp/tools/mtmd/mtmd-helper.cpp +310 -0
package/src/llama.cpp/tools/mtmd/mtmd.cpp +678 -0
package/src/llama.cpp/tools/mtmd/mtmd.h +331 -0
package/src/llama.cpp/{examples → tools}/perplexity/perplexity.cpp +21 -5
package/src/llama.cpp/{examples → tools}/quantize/quantize.cpp +53 -3
package/src/llama.cpp/tools/rpc/CMakeLists.txt +4 -0
package/src/llama.cpp/tools/rpc/rpc-server.cpp +322 -0
package/src/llama.cpp/tools/run/CMakeLists.txt +16 -0
package/src/llama.cpp/{examples → tools}/run/run.cpp +30 -30
package/src/llama.cpp/{examples → tools}/server/CMakeLists.txt +2 -1
package/src/llama.cpp/{examples → tools}/server/httplib.h +313 -247
package/src/llama.cpp/{examples → tools}/server/server.cpp +529 -215
package/src/llama.cpp/{examples → tools}/server/utils.hpp +427 -6
package/src/llama.cpp/{examples → tools}/tts/tts.cpp +6 -9
package/src/llama.cpp/cmake/arm64-windows-msvc.cmake +0 -6
package/src/llama.cpp/examples/gbnf-validator/CMakeLists.txt +0 -5
package/src/llama.cpp/examples/infill/CMakeLists.txt +0 -5
package/src/llama.cpp/examples/infill/infill.cpp +0 -590
package/src/llama.cpp/examples/llava/CMakeLists.txt +0 -66
package/src/llama.cpp/examples/llava/android/build_64.sh +0 -8
package/src/llama.cpp/examples/llava/clip-quantize-cli.cpp +0 -59
package/src/llama.cpp/examples/llava/clip.cpp +0 -3206
package/src/llama.cpp/examples/llava/clip.h +0 -118
package/src/llama.cpp/examples/llava/gemma3-cli.cpp +0 -341
package/src/llama.cpp/examples/llava/llava-cli.cpp +0 -332
package/src/llama.cpp/examples/llava/llava.cpp +0 -574
package/src/llama.cpp/examples/llava/llava.h +0 -49
package/src/llama.cpp/examples/llava/minicpmv-cli.cpp +0 -354
package/src/llama.cpp/examples/llava/qwen2vl-cli.cpp +0 -584
package/src/llama.cpp/examples/quantize-stats/CMakeLists.txt +0 -6
package/src/llama.cpp/examples/rpc/CMakeLists.txt +0 -2
package/src/llama.cpp/examples/rpc/rpc-server.cpp +0 -171
package/src/llama.cpp/examples/run/CMakeLists.txt +0 -5
package/src/llama.cpp/ggml/src/ggml-cann/kernels/CMakeLists.txt +0 -30
package/src/llama.cpp/ggml/src/ggml-cann/kernels/ascendc_kernels.h +0 -19
package/src/llama.cpp/ggml/src/ggml-cann/kernels/dup.cpp +0 -234
package/src/llama.cpp/ggml/src/ggml-cann/kernels/get_row_f16.cpp +0 -197
package/src/llama.cpp/ggml/src/ggml-cann/kernels/get_row_f32.cpp +0 -190
package/src/llama.cpp/ggml/src/ggml-cann/kernels/get_row_q4_0.cpp +0 -204
package/src/llama.cpp/ggml/src/ggml-cann/kernels/get_row_q8_0.cpp +0 -191
package/src/llama.cpp/ggml/src/ggml-cann/kernels/quantize_f16_q8_0.cpp +0 -218
package/src/llama.cpp/ggml/src/ggml-cann/kernels/quantize_f32_q8_0.cpp +0 -216
package/src/llama.cpp/ggml/src/ggml-cann/kernels/quantize_float_to_q4_0.cpp +0 -295
/package/src/llama.cpp/{examples → tools}/batched-bench/CMakeLists.txt +0 -0
/package/src/llama.cpp/{examples → tools}/cvector-generator/CMakeLists.txt +0 -0
/package/src/llama.cpp/{examples → tools}/cvector-generator/completions.txt +0 -0
/package/src/llama.cpp/{examples → tools}/cvector-generator/cvector-generator.cpp +0 -0
/package/src/llama.cpp/{examples → tools}/cvector-generator/mean.hpp +0 -0
/package/src/llama.cpp/{examples → tools}/cvector-generator/negative.txt +0 -0
/package/src/llama.cpp/{examples → tools}/cvector-generator/pca.hpp +0 -0
/package/src/llama.cpp/{examples → tools}/cvector-generator/positive.txt +0 -0
/package/src/llama.cpp/{examples → tools}/export-lora/CMakeLists.txt +0 -0
/package/src/llama.cpp/{examples → tools}/gguf-split/CMakeLists.txt +0 -0
/package/src/llama.cpp/{examples → tools}/imatrix/CMakeLists.txt +0 -0
/package/src/llama.cpp/{examples → tools}/llama-bench/CMakeLists.txt +0 -0
/package/src/llama.cpp/{examples → tools}/main/CMakeLists.txt +0 -0
/package/src/llama.cpp/{examples/llava → tools/mtmd}/requirements.txt +0 -0
/package/src/llama.cpp/{examples → tools}/perplexity/CMakeLists.txt +0 -0
/package/src/llama.cpp/{examples → tools}/quantize/CMakeLists.txt +0 -0
/package/src/llama.cpp/{examples → tools}/run/linenoise.cpp/linenoise.cpp +0 -0
/package/src/llama.cpp/{examples → tools}/run/linenoise.cpp/linenoise.h +0 -0
/package/src/llama.cpp/{examples → tools}/server/bench/requirements.txt +0 -0
/package/src/llama.cpp/{examples → tools}/server/tests/requirements.txt +0 -0
/package/src/llama.cpp/{examples → tools}/tokenize/CMakeLists.txt +0 -0
/package/src/llama.cpp/{examples → tools}/tokenize/tokenize.cpp +0 -0
/package/src/llama.cpp/{examples → tools}/tts/CMakeLists.txt +0 -0

package/src/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/CMakeLists.txt CHANGED Viewed

@@ -1,7 +1,39 @@
+cmake_minimum_required(VERSION 3.19)
+project("vulkan-shaders-gen" C CXX)
 find_package (Threads REQUIRED)
+if (GGML_VULKAN_COOPMAT_GLSLC_SUPPORT)
+    add_compile_definitions(GGML_VULKAN_COOPMAT_GLSLC_SUPPORT)
+    message(STATUS "Enabling coopmat glslc support")
+endif()
+if (GGML_VULKAN_COOPMAT2_GLSLC_SUPPORT)
+    add_compile_definitions(GGML_VULKAN_COOPMAT2_GLSLC_SUPPORT)
+    message(STATUS "Enabling coopmat2 glslc support")
+endif()
+if (GGML_VULKAN_INTEGER_DOT_GLSLC_SUPPORT)
+    add_compile_definitions(GGML_VULKAN_INTEGER_DOT_GLSLC_SUPPORT)
+    message(STATUS "Enabling dot glslc support")
+endif()
+if (GGML_VULKAN_BFLOAT16_GLSLC_SUPPORT)
+    add_compile_definitions(GGML_VULKAN_BFLOAT16_GLSLC_SUPPORT)
+    message(STATUS "Enabling bfloat16 glslc support")
+endif()
 set(TARGET vulkan-shaders-gen)
 add_executable(${TARGET} vulkan-shaders-gen.cpp)
 install(TARGETS ${TARGET} RUNTIME)
 target_compile_features(${TARGET} PRIVATE cxx_std_17)
 target_link_libraries(vulkan-shaders-gen PUBLIC Threads::Threads)
+# Configure output directories for MSVC builds
+if(MSVC)
+    # Get the main project's runtime output directory if possible
+    if(DEFINED CMAKE_RUNTIME_OUTPUT_DIRECTORY)
+        foreach(CONFIG ${CMAKE_CONFIGURATION_TYPES})
+            string(TOUPPER ${CONFIG} CONFIG)
+            set_target_properties(${TARGET} PROPERTIES
+                RUNTIME_OUTPUT_DIRECTORY_${CONFIG} ${CMAKE_RUNTIME_OUTPUT_DIRECTORY})
+        endforeach()
+    endif()
+endif()

package/src/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/vulkan-shaders-gen.cpp CHANGED Viewed

@@ -63,7 +63,8 @@ const std::vector<std::string> type_names = {
     "iq3_xxs",
     "iq3_s",
     "iq4_xs",
-    "iq4_nl"
+    "iq4_nl",
+    "bf16",
 };
 namespace {
@@ -214,7 +215,7 @@ static std::mutex compile_count_mutex;
 static std::condition_variable compile_count_cond;
 void string_to_spv_func(const std::string& _name, const std::string& in_fname, const std::map<std::string, std::string>& defines, bool fp16 = true, bool coopmat = false, bool coopmat2 = false, bool f16acc = false) {
-    std::string name = _name + (f16acc ? "_f16acc" : "") + (coopmat ? "_coopmat" : "") + (coopmat2 ? "_cm2" : (fp16 ? "" : "_fp32"));
+    std::string name = _name + (f16acc ? "_f16acc" : "") + (coopmat ? "_cm1" : "") + (coopmat2 ? "_cm2" : (fp16 ? "" : "_fp32"));
     std::string out_fname = join_paths(output_dir, name + ".spv");
     std::string in_path = join_paths(input_dir, in_fname);
@@ -295,7 +296,9 @@ void matmul_shaders(bool fp16, bool matmul_id, bool coopmat, bool coopmat2, bool
     std::string aligned_b_type_f32 = coopmat2 ? "float" : fp16 ? "mat2x4" : "vec4";
     std::string aligned_b_type_f16 = coopmat2 ? "float16_t" : fp16 ? "f16mat2x4" : "f16vec4";
-    std::map<std::string, std::string> base_dict = {{"FLOAT_TYPE", (coopmat2 || fp16) ? "float16_t" : "float"}};
+    std::map<std::string, std::string> base_dict = {
+        {"FLOAT_TYPE_VEC2", (coopmat2 || fp16) ? "f16vec2" : "vec2"},
+    };
     std::string shader_name = "matmul";
     if (matmul_id) {
@@ -313,16 +316,47 @@ void matmul_shaders(bool fp16, bool matmul_id, bool coopmat, bool coopmat2, bool
         base_dict["COOPMAT"] = "1";
     }
-    base_dict["ACC_TYPE"] = f16acc ? "float16_t" : "float";
+    const std::string source_name = coopmat2 ? "mul_mm_cm2.comp" : "mul_mm.comp";
-    std::string source_name = coopmat2 ? "mul_mm_cm2.comp" : "mul_mm.comp";
+    auto const &FLOAT_TYPE = [&](const std::string &t) -> std::string {
+        if (t == "bf16") {
+            // scalar path promotes to float
+            if (!coopmat && !coopmat2) {
+                return "float";
+            }
+            return "bfloat16_t";
+        }
+        if (coopmat2 || fp16) {
+            return "float16_t";
+        }
+        return "float";
+    };
     // Shaders with f16 B_TYPE
-    string_to_spv(shader_name + "_f32_f16", source_name, merge_maps(base_dict, {{"DATA_A_F32", "1"}, {"B_TYPE", "float16_t"}, {"D_TYPE", "float"}, }), fp16, coopmat, coopmat2, f16acc);
-    string_to_spv(shader_name + "_f32_f16_aligned", source_name, merge_maps(base_dict, {{"DATA_A_F32", "1"}, {"LOAD_VEC_A", load_vec}, {"LOAD_VEC_B", load_vec}, {"B_TYPE", aligned_b_type_f16}, {"D_TYPE", "float"}, {"ALIGNED", "1"}}), fp16, coopmat, coopmat2, f16acc);
+    string_to_spv(shader_name + "_f32_f16", source_name, merge_maps(base_dict, {{"FLOAT_TYPE", FLOAT_TYPE("f16")}, {"DATA_A_F32", "1"}, {"B_TYPE", "float16_t"}, {"D_TYPE", "float"}, }), fp16, coopmat, coopmat2, f16acc);
+    string_to_spv(shader_name + "_f32_f16_aligned", source_name, merge_maps(base_dict, {{"FLOAT_TYPE", FLOAT_TYPE("f16")}, {"DATA_A_F32", "1"}, {"LOAD_VEC_A", load_vec}, {"LOAD_VEC_B", load_vec}, {"B_TYPE", aligned_b_type_f16}, {"D_TYPE", "float"}, {"ALIGNED", "1"}}), fp16, coopmat, coopmat2, f16acc);
+    string_to_spv(shader_name + "_f16_aligned", source_name, merge_maps(base_dict, {{"FLOAT_TYPE", FLOAT_TYPE("f16")}, {"DATA_A_F16", "1"}, {"LOAD_VEC_A", load_vec}, {"LOAD_VEC_B", load_vec}, {"B_TYPE", aligned_b_type_f16}, {"D_TYPE", "float"}, {"ALIGNED", "1"}}), fp16, coopmat, coopmat2, f16acc);
+    string_to_spv(shader_name + "_f16", source_name, merge_maps(base_dict, {{"FLOAT_TYPE", FLOAT_TYPE("f16")}, {"DATA_A_F16", "1"}, {"B_TYPE", "float16_t"}, {"D_TYPE", "float"}}), fp16, coopmat, coopmat2, f16acc);
-    string_to_spv(shader_name + "_f16_aligned", source_name, merge_maps(base_dict, {{"DATA_A_F16", "1"}, {"LOAD_VEC_A", load_vec}, {"LOAD_VEC_B", load_vec}, {"B_TYPE", aligned_b_type_f16}, {"D_TYPE", "float"}, {"ALIGNED", "1"}}), fp16, coopmat, coopmat2, f16acc);
-    string_to_spv(shader_name + "_f16", source_name, merge_maps(base_dict, {{"DATA_A_F16", "1"}, {"B_TYPE", "float16_t"}, {"D_TYPE", "float"}}), fp16, coopmat, coopmat2, f16acc);
+    // bf16
+    {
+        std::string load_vec_a_unaligned = "1";
+        // For aligned matmul loads
+        std::string load_vec_a = coopmat2 ? "1" : "4";
+        // scalar path promotes to float
+        std::string to_float_type = (coopmat || coopmat2) ? "uintBitsToBFloat16EXT" : "bf16_to_fp32";
+        // If bfloat16 is not supported, then only compile the scalar (promote to fp32) shader
+#if !defined(GGML_VULKAN_BFLOAT16_GLSLC_SUPPORT)
+        if (!(coopmat || coopmat2))
+#endif
+        {
+            string_to_spv(shader_name + "_bf16_aligned", source_name, merge_maps(base_dict, {{"FLOAT_TYPE", FLOAT_TYPE("bf16")}, {"TO_FLOAT_TYPE", to_float_type}, {"DATA_A_BF16", "1"}, {"LOAD_VEC_A", load_vec_a},           {"LOAD_VEC_B", "4"}, {"B_TYPE", coopmat2 ? "bfloat16_t" : "u16vec4"},   {"D_TYPE", "float"}, {"B_IS_FLOAT", "1"}, {"ALIGNED", "1"}}), fp16, coopmat, coopmat2, f16acc);
+            string_to_spv(shader_name + "_bf16",         source_name, merge_maps(base_dict, {{"FLOAT_TYPE", FLOAT_TYPE("bf16")}, {"TO_FLOAT_TYPE", to_float_type}, {"DATA_A_BF16", "1"}, {"LOAD_VEC_A", load_vec_a_unaligned},                      {"B_TYPE", coopmat2 ? "bfloat16_t" : "uint16_t"},                          {"D_TYPE", "float"}, {"B_IS_FLOAT", "1"}}),                   fp16, coopmat, coopmat2, f16acc);
+        }
+    }
     for (const auto& tname : type_names) {
         std::string load_vec_quant = "2";
@@ -331,22 +365,32 @@ void matmul_shaders(bool fp16, bool matmul_id, bool coopmat, bool coopmat2, bool
         else if ((tname == "q5_0") || (tname == "q5_1") || (tname == "q8_0") || (tname == "iq4_nl"))
             load_vec_quant = "4";
+        if (tname == "bf16") {
+            continue;
+        }
         std::string data_a_key = "DATA_A_" + to_uppercase(tname);
         // For unaligned, load one at a time for f32/f16, or two at a time for quants
-        std::string load_vec_a_unaligned = (coopmat2 || tname == "f32" || tname == "f16") ? "1" : load_vec_quant;
+        std::string load_vec_a_unaligned = (coopmat2 || tname == "f32" || tname == "f16" || tname == "bf16") ? "1" : load_vec_quant;
         // For aligned matmul loads
-        std::string load_vec_a = (coopmat2 || tname == "f32" || tname == "f16") ? load_vec : load_vec_quant;
+        std::string load_vec_a = (coopmat2 || tname == "f32" || tname == "f16" || tname == "bf16") ? load_vec : load_vec_quant;
         // don't generate f32 variants for coopmat2
         if (!coopmat2) {
-            string_to_spv(shader_name + "_" + tname + "_f32", source_name, merge_maps(base_dict, {{data_a_key, "1"}, {"LOAD_VEC_A", load_vec_a_unaligned}, {"B_TYPE", "float"}, {"D_TYPE", "float"}, {"B_IS_FLOAT", "1"}}), fp16, coopmat, coopmat2, f16acc);
-            string_to_spv(shader_name + "_" + tname + "_f32_aligned", source_name, merge_maps(base_dict, {{data_a_key, "1"}, {"LOAD_VEC_A", load_vec_a}, {"LOAD_VEC_B", load_vec}, {"B_TYPE", aligned_b_type_f32}, {"D_TYPE", "float"}, {"B_IS_FLOAT", "1"}, {"ALIGNED", "1"}}), fp16, coopmat, coopmat2, f16acc);
+            string_to_spv(shader_name + "_" + tname + "_f32",         source_name, merge_maps(base_dict, {{"FLOAT_TYPE", FLOAT_TYPE(tname)}, {data_a_key, "1"}, {"LOAD_VEC_A", load_vec_a_unaligned},                           {"B_TYPE", "float"},            {"D_TYPE", "float"}}), fp16, coopmat, coopmat2, f16acc);
+            string_to_spv(shader_name + "_" + tname + "_f32_aligned", source_name, merge_maps(base_dict, {{"FLOAT_TYPE", FLOAT_TYPE(tname)}, {data_a_key, "1"}, {"LOAD_VEC_A", load_vec_a},           {"LOAD_VEC_B", load_vec}, {"B_TYPE", aligned_b_type_f32}, {"D_TYPE", "float"}, {"ALIGNED", "1"}}), fp16, coopmat, coopmat2, f16acc);
         }
         if (tname != "f16" && tname != "f32") {
-            string_to_spv(shader_name + "_" + tname + "_f16", source_name,          merge_maps(base_dict, {{data_a_key, "1"}, {"LOAD_VEC_A", load_vec_a_unaligned},                           {"B_TYPE", "float16_t"},        {"D_TYPE", "float"}, {"B_IS_FLOAT", "1"}}), fp16, coopmat, coopmat2, f16acc);
-            string_to_spv(shader_name + "_" + tname + "_f16_aligned", source_name,  merge_maps(base_dict, {{data_a_key, "1"}, {"LOAD_VEC_A", load_vec_a},           {"LOAD_VEC_B", load_vec}, {"B_TYPE", aligned_b_type_f16}, {"D_TYPE", "float"}, {"B_IS_FLOAT", "1"}, {"ALIGNED", "1"}}), fp16, coopmat, coopmat2, f16acc);
+            string_to_spv(shader_name + "_" + tname + "_f16",         source_name,  merge_maps(base_dict, {{"FLOAT_TYPE", FLOAT_TYPE(tname)}, {data_a_key, "1"}, {"LOAD_VEC_A", load_vec_a_unaligned},                           {"B_TYPE", "float16_t"},        {"D_TYPE", "float"}}), fp16, coopmat, coopmat2, f16acc);
+            string_to_spv(shader_name + "_" + tname + "_f16_aligned", source_name,  merge_maps(base_dict, {{"FLOAT_TYPE", FLOAT_TYPE(tname)}, {data_a_key, "1"}, {"LOAD_VEC_A", load_vec_a},           {"LOAD_VEC_B", load_vec}, {"B_TYPE", aligned_b_type_f16}, {"D_TYPE", "float"}, {"ALIGNED", "1"}}), fp16, coopmat, coopmat2, f16acc);
         }
+#if defined(GGML_VULKAN_INTEGER_DOT_GLSLC_SUPPORT)
+        if (!coopmat && !coopmat2 && !matmul_id && (tname == "q4_0" || tname == "q4_1" || tname == "q5_0" || tname == "q5_1" || tname == "q8_0")) {
+            string_to_spv(shader_name + "_" + tname + "_q8_1", "mul_mmq.comp", merge_maps(base_dict, {{"FLOAT_TYPE", FLOAT_TYPE(tname)}, {data_a_key, "1"}, {"D_TYPE", "float"},}), fp16, coopmat, coopmat2, f16acc);
+        }
+#endif
     }
 }
@@ -377,16 +421,18 @@ void process_shaders() {
 #endif
     }
-#if defined(GGML_VULKAN_COOPMAT2_GLSLC_SUPPORT)
     // flash attention
     for (const auto& f16acc : {false, true}) {
         std::string acctype = f16acc ? "float16_t" : "float";
+        std::string acctypev4 = f16acc ? "f16vec4" : "vec4";
         for (const auto& tname : type_names) {
             if (tname == "f32") {
                 continue;
             }
+            if (tname == "bf16") continue;
+#if defined(GGML_VULKAN_COOPMAT2_GLSLC_SUPPORT)
             if (tname == "f16") {
                 string_to_spv("flash_attn_f32_f16_" + tname, "flash_attn_cm2.comp",
                     merge_maps(base_dict, {{"Q_TYPE", "float"}, {"D_TYPE", "float"}, {"ACC_TYPE", acctype}}), true, false, true, f16acc);
@@ -395,9 +441,27 @@ void process_shaders() {
                 string_to_spv("flash_attn_f32_f16_" + tname, "flash_attn_cm2.comp",
                     merge_maps(base_dict, {{data_a_key, "1"}, {"Q_TYPE", "float"}, {"D_TYPE", "float"}, {"ACC_TYPE", acctype}, {"DEQUANTFUNC", "dequantFunc"+to_uppercase(tname) }, {"BLOCK_SIZE", "QUANT_K_"+to_uppercase(tname) }}), true, false, true, f16acc);
             }
+#endif
+#if defined(GGML_VULKAN_COOPMAT_GLSLC_SUPPORT)
+            if (tname == "f16") {
+                string_to_spv("flash_attn_f32_f16_" + tname, "flash_attn_cm1.comp",
+                    merge_maps(base_dict, {{"Q_TYPE", "float"}, {"D_TYPE", "float"}, {"ACC_TYPE", acctype}, {"ACC_TYPEV4", acctypev4}, {"COOPMAT", "1"}}), true, true, false, f16acc);
+            } else if (tname == "q4_0" || tname == "q8_0") {
+                std::string data_a_key = "DATA_A_" + to_uppercase(tname);
+                string_to_spv("flash_attn_f32_f16_" + tname, "flash_attn_cm1.comp",
+                    merge_maps(base_dict, {{data_a_key, "1"}, {"Q_TYPE", "float"}, {"D_TYPE", "float"}, {"ACC_TYPE", acctype}, {"ACC_TYPEV4", acctypev4}, {"BLOCK_SIZE", "QUANT_K_"+to_uppercase(tname)}, {"COOPMAT", "1"}}), true, true, false, f16acc);
+            }
+#endif
+            if (tname == "f16") {
+                string_to_spv("flash_attn_f32_f16_" + tname, "flash_attn.comp",
+                    merge_maps(base_dict, {{"Q_TYPE", "float"}, {"D_TYPE", "float"}, {"ACC_TYPE", acctype}}), true, false, false, f16acc);
+            } else if (tname == "q4_0" || tname == "q8_0") {
+                std::string data_a_key = "DATA_A_" + to_uppercase(tname);
+                string_to_spv("flash_attn_f32_f16_" + tname, "flash_attn.comp",
+                    merge_maps(base_dict, {{data_a_key, "1"}, {"Q_TYPE", "float"}, {"D_TYPE", "float"}, {"ACC_TYPE", acctype}, {"BLOCK_SIZE", "QUANT_K_"+to_uppercase(tname) }}), true, false, false, f16acc);
+            }
         }
     }
-#endif
     for (const auto& tname : type_names) {
         // mul mat vec
@@ -410,12 +474,12 @@ void process_shaders() {
         string_to_spv("mul_mat_vec_id_" + tname + "_f32", shader, merge_maps(base_dict, {{"MUL_MAT_ID", "1"}, {data_a_key, "1"}, {"B_TYPE", "float"}, {"B_TYPE_VEC2", "vec2"}, {"B_TYPE_VEC4", "vec4"}, {"D_TYPE", "float"}}));
         // Dequant shaders
-        if (tname != "f16") {
+        if (tname != "f16" && tname != "bf16") {
             string_to_spv("dequant_" + tname, "dequant_" + tname + ".comp", merge_maps(base_dict, {{data_a_key, "1"}, {"D_TYPE", "float16_t"}}));
         }
         if (!string_ends_with(tname, "_k")) {
-            shader = (tname == "f32" || tname == "f16") ? "get_rows.comp" : "get_rows_quant.comp";
+            shader = (tname == "f32" || tname == "f16" || tname == "bf16") ? "get_rows.comp" : "get_rows_quant.comp";
             if (tname == "f16") {
                 string_to_spv("get_rows_" + tname, shader, merge_maps(base_dict, {{data_a_key, "1"}, {"B_TYPE", "int"}, {"D_TYPE", "float16_t"}, {"OPTIMIZATION_ERROR_WORKAROUND", "1"}}));
@@ -440,9 +504,13 @@ void process_shaders() {
     string_to_spv("cpy_f32_f32", "copy.comp", {{"A_TYPE", "float"}, {"D_TYPE", "float"}});
     string_to_spv("cpy_f32_f16", "copy.comp", {{"A_TYPE", "float"}, {"D_TYPE", "float16_t"}});
     string_to_spv("cpy_f16_f16", "copy.comp", {{"A_TYPE", "float16_t"}, {"D_TYPE", "float16_t"}, {"OPTIMIZATION_ERROR_WORKAROUND", "1"}});
+    string_to_spv("cpy_f16_f32", "copy.comp", {{"A_TYPE", "float16_t"}, {"D_TYPE", "float"}, {"OPTIMIZATION_ERROR_WORKAROUND", "1"}});
+    string_to_spv("cpy_f32_bf16","copy.comp", {{"A_TYPE", "float"}, {"D_TYPE", "uint16_t"}, {"DATA_D_BF16", "1"}});
     string_to_spv("contig_cpy_f32_f32", "contig_copy.comp", {{"A_TYPE", "float"}, {"D_TYPE", "float"}});
     string_to_spv("contig_cpy_f32_f16", "contig_copy.comp", {{"A_TYPE", "float"}, {"D_TYPE", "float16_t"}});
     string_to_spv("contig_cpy_f16_f16", "contig_copy.comp", {{"A_TYPE", "float16_t"}, {"D_TYPE", "float16_t"}, {"OPTIMIZATION_ERROR_WORKAROUND", "1"}});
+    string_to_spv("contig_cpy_f16_f32", "contig_copy.comp", {{"A_TYPE", "float16_t"}, {"D_TYPE", "float"}, {"OPTIMIZATION_ERROR_WORKAROUND", "1"}});
+    string_to_spv("contig_cpy_f32_bf16","contig_copy.comp",{{"A_TYPE", "float"}, {"D_TYPE", "uint16_t"}, {"DATA_D_BF16", "1"}});
     for (std::string t : {"q4_0", "q4_1", "q5_0", "q5_1", "q8_0", "iq4_nl"}) {
         string_to_spv("cpy_f32_" + t, "copy_to_quant.comp", {{"DATA_A_" + to_uppercase(t), "1"}, {"D_TYPE", "float"}, {"FLOAT_TYPE", "float"}});
@@ -450,14 +518,34 @@ void process_shaders() {
         string_to_spv("cpy_" + t + "_f32", "copy_from_quant.comp", {{"DATA_A_" + to_uppercase(t), "1"}, {"D_TYPE", "float"}, {"FLOAT_TYPE", "float"}});
     }
-    string_to_spv("add_f32", "add.comp", {{"A_TYPE", "float"}, {"B_TYPE", "float"}, {"D_TYPE", "float"}, {"FLOAT_TYPE", "float"}});
-    string_to_spv("add_f16_f32_f16", "add.comp", {{"A_TYPE", "float16_t"}, {"B_TYPE", "float"}, {"D_TYPE", "float16_t"}, {"FLOAT_TYPE", "float"}});
+    auto get_type_str = [](bool f16) {
+        return f16 ? "float16_t" : "float";
+    };
+    auto get_suffix = [](bool src0_f16, bool src1_f16, bool dst_f16) {
+        std::string s;
+        s += std::string(src0_f16 ? "_f16" : "_f32");
+        s += std::string(src1_f16 ? "_f16" : "_f32");
+        s += std::string(dst_f16 ? "_f16" : "_f32");
+        return s;
+    };
+    for (std::string op : {"add", "sub", "mul", "div"}) {
+    for (auto src0_f16 : {false, true}) {
+    for (auto src1_f16 : {false, true}) {
+    for (auto dst_f16  : {false, true}) {
+        auto name = op + get_suffix(src0_f16, src1_f16, dst_f16);
+        string_to_spv(name.c_str(), op + ".comp", {{"A_TYPE", get_type_str(src0_f16)}, {"B_TYPE", get_type_str(src1_f16)}, {"D_TYPE", get_type_str(dst_f16)}, {"FLOAT_TYPE", "float"}});
+    }
+    }
+    }
+    }
     string_to_spv("sub_f32", "sub.comp", {{"A_TYPE", "float"}, {"B_TYPE", "float"}, {"D_TYPE", "float"}, {"FLOAT_TYPE", "float"}});
     string_to_spv("acc_f32", "acc.comp", {{"A_TYPE", "float"}, {"B_TYPE", "float"}, {"D_TYPE", "float"}, {"FLOAT_TYPE", "float"}});
     string_to_spv("split_k_reduce", "mul_mat_split_k_reduce.comp", {});
+    string_to_spv("fa_split_k_reduce", "flash_attn_split_k_reduce.comp", {});
+    string_to_spv("quantize_q8_1", "quantize_q8_1.comp", {});
     string_to_spv("mul_f32", "mul.comp", {{"A_TYPE", "float"}, {"B_TYPE", "float"}, {"D_TYPE", "float"}, {"FLOAT_TYPE", "float"}});
@@ -484,14 +572,21 @@ void process_shaders() {
     string_to_spv("upscale_f32", "upscale.comp", {{"A_TYPE", "float"}, {"B_TYPE", "float"}, {"D_TYPE", "float"}});
-    string_to_spv("gelu_f32", "gelu.comp", {{"A_TYPE", "float"}, {"D_TYPE", "float"}});
-    string_to_spv("gelu_quick_f32", "gelu_quick.comp", {{"A_TYPE", "float"}, {"D_TYPE", "float"}});
-    string_to_spv("silu_f32", "silu.comp", {{"A_TYPE", "float"}, {"D_TYPE", "float"}});
-    string_to_spv("silu_back_f32", "silu_back.comp", {{"A_TYPE", "float"}, {"B_TYPE", "float"}, {"D_TYPE", "float"}});
-    string_to_spv("relu_f32", "relu.comp", {{"A_TYPE", "float"}, {"D_TYPE", "float"}});
-    string_to_spv("leaky_relu_f32", "leaky_relu.comp", {{"A_TYPE", "float"}, {"D_TYPE", "float"}});
-    string_to_spv("tanh_f32", "tanh.comp", {{"A_TYPE", "float"}, {"D_TYPE", "float"}});
-    string_to_spv("sigmoid_f32", "sigmoid.comp", {{"A_TYPE", "float"}, {"D_TYPE", "float"}});
+    string_to_spv("gelu_f16",       "gelu.comp",        {{"A_TYPE", "float16_t"},   {"D_TYPE", "float16_t"}});
+    string_to_spv("gelu_f32",       "gelu.comp",        {{"A_TYPE", "float"},       {"D_TYPE", "float"}});
+    string_to_spv("gelu_quick_f16", "gelu_quick.comp",  {{"A_TYPE", "float16_t"},   {"D_TYPE", "float16_t"}});
+    string_to_spv("gelu_quick_f32", "gelu_quick.comp",  {{"A_TYPE", "float"},       {"D_TYPE", "float"}});
+    string_to_spv("silu_f16",       "silu.comp",        {{"A_TYPE", "float16_t"},   {"D_TYPE", "float16_t"}});
+    string_to_spv("silu_f32",       "silu.comp",        {{"A_TYPE", "float"},       {"D_TYPE", "float"}});
+    string_to_spv("relu_f16",       "relu.comp",        {{"A_TYPE", "float16_t"},   {"D_TYPE", "float16_t"}});
+    string_to_spv("relu_f32",       "relu.comp",        {{"A_TYPE", "float"},       {"D_TYPE", "float"}});
+    string_to_spv("tanh_f16",       "tanh.comp",        {{"A_TYPE", "float16_t"},   {"D_TYPE", "float16_t"}});
+    string_to_spv("tanh_f32",       "tanh.comp",        {{"A_TYPE", "float"},       {"D_TYPE", "float"}});
+    string_to_spv("sigmoid_f16",    "sigmoid.comp",     {{"A_TYPE", "float16_t"},   {"D_TYPE", "float16_t"}});
+    string_to_spv("sigmoid_f32",    "sigmoid.comp",     {{"A_TYPE", "float"},       {"D_TYPE", "float"}});
+    string_to_spv("leaky_relu_f32", "leaky_relu.comp",  {{"A_TYPE", "float"}, {"D_TYPE", "float"}});
+    string_to_spv("silu_back_f32",  "silu_back.comp",   {{"A_TYPE", "float"}, {"B_TYPE", "float"}, {"D_TYPE", "float"}});
     string_to_spv("diag_mask_inf_f32", "diag_mask_inf.comp", {{"A_TYPE", "float"}, {"D_TYPE", "float"}});
@@ -535,6 +630,9 @@ void process_shaders() {
     string_to_spv("opt_step_adamw_f32", "opt_step_adamw.comp", merge_maps(base_dict, {{"A_TYPE", "float"}}));
+    string_to_spv("conv2d_dw_whcn_f32", "conv2d_dw.comp", merge_maps(base_dict, {{"A_TYPE", "float"}, {"B_TYPE", "float"}, {"D_TYPE", "float"}, {"WHCN", "1"}}));
+    string_to_spv("conv2d_dw_cwhn_f32", "conv2d_dw.comp", merge_maps(base_dict, {{"A_TYPE", "float"}, {"B_TYPE", "float"}, {"D_TYPE", "float"}, {"CWHN", "1"}}));
     for (auto &c : compiles) {
         c.wait();
     }
@@ -589,7 +687,12 @@ void write_output_files() {
             std::remove(path.c_str());
         }
     }
+    for (const char *op : {"add", "sub", "mul", "div"}) {
+        fprintf(hdr, "extern unsigned char *%s_data[2][2][2];\n", op);
+        fprintf(hdr, "extern uint64_t %s_len[2][2][2];\n", op);
+        fprintf(src, "unsigned char *%s_data[2][2][2] = {{{%s_f32_f32_f32_data, %s_f32_f32_f16_data}, {%s_f32_f16_f32_data, %s_f32_f16_f16_data}}, {{%s_f16_f32_f32_data, %s_f16_f32_f16_data}, {%s_f16_f16_f32_data, %s_f16_f16_f16_data}}};\n", op, op, op, op, op, op, op, op, op);
+        fprintf(src, "uint64_t %s_len[2][2][2] = {{{%s_f32_f32_f32_len, %s_f32_f32_f16_len}, {%s_f32_f16_f32_len, %s_f32_f16_f16_len}}, {{%s_f16_f32_f32_len, %s_f16_f32_f16_len}, {%s_f16_f16_f32_len, %s_f16_f16_f16_len}}};\n", op, op, op, op, op, op, op, op, op);
+    }
     fclose(hdr);
     fclose(src);
 }