npm - @fugood/llama.node - Versions diffs - 0.3.3 → 0.3.5 - Mend

@fugood/llama.node 0.3.3 → 0.3.5

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 (225) hide show

package/CMakeLists.txt +5 -0
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-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 +29 -1
package/package.json +1 -1
package/src/EmbeddingWorker.cpp +15 -5
package/src/EmbeddingWorker.h +2 -1
package/src/LlamaCompletionWorker.cpp +17 -1
package/src/LlamaContext.cpp +86 -18
package/src/LlamaContext.h +2 -0
package/src/llama.cpp/.github/workflows/build.yml +197 -159
package/src/llama.cpp/.github/workflows/docker.yml +5 -8
package/src/llama.cpp/.github/workflows/python-lint.yml +8 -1
package/src/llama.cpp/.github/workflows/server.yml +21 -14
package/src/llama.cpp/CMakeLists.txt +11 -6
package/src/llama.cpp/Sources/llama/llama.h +4 -0
package/src/llama.cpp/cmake/common.cmake +33 -0
package/src/llama.cpp/cmake/x64-windows-llvm.cmake +11 -0
package/src/llama.cpp/common/CMakeLists.txt +6 -2
package/src/llama.cpp/common/arg.cpp +426 -245
package/src/llama.cpp/common/common.cpp +143 -80
package/src/llama.cpp/common/common.h +81 -24
package/src/llama.cpp/common/sampling.cpp +53 -19
package/src/llama.cpp/common/sampling.h +22 -1
package/src/llama.cpp/common/speculative.cpp +274 -0
package/src/llama.cpp/common/speculative.h +28 -0
package/src/llama.cpp/docs/build.md +101 -148
package/src/llama.cpp/examples/CMakeLists.txt +32 -13
package/src/llama.cpp/examples/batched/CMakeLists.txt +1 -1
package/src/llama.cpp/examples/batched/batched.cpp +5 -4
package/src/llama.cpp/examples/batched-bench/CMakeLists.txt +1 -1
package/src/llama.cpp/examples/convert-llama2c-to-ggml/CMakeLists.txt +1 -1
package/src/llama.cpp/examples/cvector-generator/CMakeLists.txt +1 -1
package/src/llama.cpp/examples/deprecation-warning/deprecation-warning.cpp +1 -1
package/src/llama.cpp/examples/embedding/CMakeLists.txt +1 -1
package/src/llama.cpp/examples/eval-callback/CMakeLists.txt +3 -2
package/src/llama.cpp/examples/export-lora/CMakeLists.txt +1 -1
package/src/llama.cpp/examples/gbnf-validator/CMakeLists.txt +1 -1
package/src/llama.cpp/examples/gbnf-validator/gbnf-validator.cpp +4 -7
package/src/llama.cpp/examples/gen-docs/CMakeLists.txt +1 -1
package/src/llama.cpp/examples/gguf/CMakeLists.txt +1 -1
package/src/llama.cpp/examples/gguf-hash/CMakeLists.txt +8 -1
package/src/llama.cpp/examples/gguf-split/CMakeLists.txt +1 -1
package/src/llama.cpp/examples/gguf-split/gguf-split.cpp +2 -2
package/src/llama.cpp/examples/gritlm/CMakeLists.txt +1 -1
package/src/llama.cpp/examples/gritlm/gritlm.cpp +1 -1
package/src/llama.cpp/examples/imatrix/CMakeLists.txt +1 -1
package/src/llama.cpp/examples/imatrix/imatrix.cpp +11 -2
package/src/llama.cpp/examples/infill/CMakeLists.txt +1 -1
package/src/llama.cpp/examples/infill/infill.cpp +1 -1
package/src/llama.cpp/examples/llama-bench/CMakeLists.txt +1 -1
package/src/llama.cpp/examples/llama-bench/llama-bench.cpp +405 -316
package/src/llama.cpp/examples/llama.android/llama/build.gradle.kts +1 -0
package/src/llama.cpp/examples/llava/CMakeLists.txt +10 -3
package/src/llama.cpp/examples/llava/clip.cpp +262 -66
package/src/llama.cpp/examples/llava/clip.h +8 -2
package/src/llama.cpp/examples/llava/llava-cli.cpp +1 -1
package/src/llama.cpp/examples/llava/llava.cpp +46 -19
package/src/llama.cpp/examples/llava/minicpmv-cli.cpp +1 -1
package/src/llama.cpp/examples/llava/qwen2vl-cli.cpp +581 -0
package/src/llama.cpp/examples/lookahead/CMakeLists.txt +1 -1
package/src/llama.cpp/examples/lookahead/lookahead.cpp +1 -1
package/src/llama.cpp/examples/lookup/CMakeLists.txt +4 -4
package/src/llama.cpp/examples/lookup/lookup-stats.cpp +2 -1
package/src/llama.cpp/examples/lookup/lookup.cpp +2 -2
package/src/llama.cpp/examples/main/CMakeLists.txt +1 -1
package/src/llama.cpp/examples/main/main.cpp +9 -5
package/src/llama.cpp/examples/main-cmake-pkg/CMakeLists.txt +1 -1
package/src/llama.cpp/examples/parallel/CMakeLists.txt +1 -1
package/src/llama.cpp/examples/parallel/parallel.cpp +1 -1
package/src/llama.cpp/examples/passkey/CMakeLists.txt +1 -1
package/src/llama.cpp/examples/perplexity/CMakeLists.txt +1 -1
package/src/llama.cpp/examples/quantize/CMakeLists.txt +1 -1
package/src/llama.cpp/examples/quantize/quantize.cpp +0 -3
package/src/llama.cpp/examples/quantize-stats/CMakeLists.txt +1 -1
package/src/llama.cpp/examples/retrieval/CMakeLists.txt +1 -1
package/src/llama.cpp/examples/retrieval/retrieval.cpp +4 -4
package/src/llama.cpp/examples/run/CMakeLists.txt +5 -0
package/src/llama.cpp/examples/run/run.cpp +911 -0
package/src/llama.cpp/examples/save-load-state/CMakeLists.txt +1 -1
package/src/llama.cpp/examples/save-load-state/save-load-state.cpp +4 -4
package/src/llama.cpp/examples/server/CMakeLists.txt +3 -7
package/src/llama.cpp/examples/server/server.cpp +1758 -886
package/src/llama.cpp/examples/server/tests/requirements.txt +2 -2
package/src/llama.cpp/examples/server/utils.hpp +94 -304
package/src/llama.cpp/examples/simple/CMakeLists.txt +1 -1
package/src/llama.cpp/examples/simple/simple.cpp +4 -0
package/src/llama.cpp/examples/simple-chat/CMakeLists.txt +1 -1
package/src/llama.cpp/examples/simple-chat/simple-chat.cpp +3 -0
package/src/llama.cpp/examples/speculative/CMakeLists.txt +1 -1
package/src/llama.cpp/examples/speculative/speculative.cpp +16 -15
package/src/llama.cpp/examples/speculative-simple/CMakeLists.txt +5 -0
package/src/llama.cpp/examples/speculative-simple/speculative-simple.cpp +265 -0
package/src/llama.cpp/examples/tokenize/CMakeLists.txt +1 -1
package/src/llama.cpp/examples/tokenize/tokenize.cpp +1 -1
package/src/llama.cpp/examples/tts/CMakeLists.txt +5 -0
package/src/llama.cpp/examples/tts/tts.cpp +932 -0
package/src/llama.cpp/ggml/CMakeLists.txt +46 -34
package/src/llama.cpp/ggml/include/ggml-backend.h +16 -0
package/src/llama.cpp/ggml/include/ggml-cpu.h +7 -49
package/src/llama.cpp/ggml/include/ggml-opencl.h +26 -0
package/src/llama.cpp/ggml/include/ggml.h +106 -24
package/src/llama.cpp/ggml/src/CMakeLists.txt +73 -24
package/src/llama.cpp/ggml/src/ggml-alloc.c +0 -1
package/src/llama.cpp/ggml/src/ggml-backend-impl.h +51 -11
package/src/llama.cpp/ggml/src/ggml-backend-reg.cpp +379 -22
package/src/llama.cpp/ggml/src/ggml-backend.cpp +4 -4
package/src/llama.cpp/ggml/src/ggml-blas/CMakeLists.txt +3 -7
package/src/llama.cpp/ggml/src/ggml-blas/ggml-blas.cpp +5 -2
package/src/llama.cpp/ggml/src/ggml-cann/CMakeLists.txt +33 -3
package/src/llama.cpp/ggml/src/ggml-cann/aclnn_ops.cpp +456 -111
package/src/llama.cpp/ggml/src/ggml-cann/common.h +6 -3
package/src/llama.cpp/ggml/src/ggml-cann/ggml-cann.cpp +95 -35
package/src/llama.cpp/ggml/src/ggml-cann/kernels/CMakeLists.txt +2 -5
package/src/llama.cpp/ggml/src/ggml-cann/kernels/dup.cpp +22 -9
package/src/llama.cpp/ggml/src/ggml-cann/kernels/get_row_f16.cpp +24 -13
package/src/llama.cpp/ggml/src/ggml-cann/kernels/get_row_f32.cpp +23 -13
package/src/llama.cpp/ggml/src/ggml-cann/kernels/get_row_q4_0.cpp +11 -0
package/src/llama.cpp/ggml/src/ggml-cann/kernels/quantize_f16_q8_0.cpp +10 -0
package/src/llama.cpp/ggml/src/ggml-cann/kernels/quantize_f32_q8_0.cpp +10 -0
package/src/llama.cpp/ggml/src/ggml-cann/kernels/quantize_float_to_q4_0.cpp +17 -0
package/src/llama.cpp/ggml/src/ggml-common.h +42 -42
package/src/llama.cpp/ggml/src/ggml-cpu/CMakeLists.txt +288 -213
package/src/llama.cpp/ggml/src/ggml-cpu/amx/amx.cpp +220 -0
package/src/llama.cpp/ggml/src/ggml-cpu/amx/amx.h +8 -0
package/src/llama.cpp/ggml/src/{ggml-amx → ggml-cpu/amx}/common.h +19 -22
package/src/llama.cpp/ggml/src/{ggml-amx → ggml-cpu/amx}/mmq.cpp +93 -92
package/src/llama.cpp/ggml/src/{ggml-amx → ggml-cpu/amx}/mmq.h +2 -9
package/src/llama.cpp/ggml/src/ggml-cpu/cpu-feats-x86.cpp +323 -0
package/src/llama.cpp/ggml/src/ggml-cpu/{ggml-cpu-aarch64.c → ggml-cpu-aarch64.cpp} +892 -190
package/src/llama.cpp/ggml/src/ggml-cpu/ggml-cpu-aarch64.h +2 -24
package/src/llama.cpp/ggml/src/ggml-cpu/ggml-cpu-hbm.cpp +55 -0
package/src/llama.cpp/ggml/src/ggml-cpu/ggml-cpu-hbm.h +8 -0
package/src/llama.cpp/ggml/src/ggml-cpu/ggml-cpu-impl.h +15 -0
package/src/llama.cpp/ggml/src/ggml-cpu/ggml-cpu-quants.c +38 -25
package/src/llama.cpp/ggml/src/ggml-cpu/ggml-cpu-traits.cpp +36 -0
package/src/llama.cpp/ggml/src/ggml-cpu/ggml-cpu-traits.h +38 -0
package/src/llama.cpp/ggml/src/ggml-cpu/ggml-cpu.c +552 -399
package/src/llama.cpp/ggml/src/ggml-cpu/ggml-cpu.cpp +101 -136
package/src/llama.cpp/ggml/src/ggml-cpu/llamafile/sgemm.cpp +2 -2
package/src/llama.cpp/ggml/src/ggml-cuda/CMakeLists.txt +7 -10
package/src/llama.cpp/ggml/src/ggml-cuda/vendors/hip.h +8 -0
package/src/llama.cpp/ggml/src/ggml-hip/CMakeLists.txt +4 -6
package/src/llama.cpp/ggml/src/ggml-impl.h +32 -11
package/src/llama.cpp/ggml/src/ggml-kompute/CMakeLists.txt +13 -9
package/src/llama.cpp/ggml/src/ggml-kompute/ggml-kompute.cpp +131 -64
package/src/llama.cpp/ggml/src/ggml-metal/CMakeLists.txt +3 -6
package/src/llama.cpp/ggml/src/ggml-metal/ggml-metal-impl.h +39 -0
package/src/llama.cpp/ggml/src/ggml-musa/CMakeLists.txt +14 -7
package/src/llama.cpp/ggml/src/ggml-opencl/CMakeLists.txt +147 -0
package/src/llama.cpp/ggml/src/ggml-opencl/ggml-opencl.cpp +4004 -0
package/src/llama.cpp/ggml/src/ggml-opt.cpp +67 -80
package/src/llama.cpp/ggml/src/ggml-quants.c +0 -9
package/src/llama.cpp/ggml/src/ggml-rpc/CMakeLists.txt +3 -5
package/src/llama.cpp/ggml/src/ggml-rpc/ggml-rpc.cpp +5 -2
package/src/llama.cpp/ggml/src/ggml-sycl/CMakeLists.txt +13 -10
package/src/llama.cpp/ggml/src/ggml-sycl/common.cpp +2 -11
package/src/llama.cpp/ggml/src/ggml-sycl/common.hpp +1 -0
package/src/llama.cpp/ggml/src/ggml-sycl/concat.cpp +2 -2
package/src/llama.cpp/ggml/src/ggml-sycl/convert.cpp +1 -1
package/src/llama.cpp/ggml/src/ggml-sycl/dmmv.cpp +5 -5
package/src/llama.cpp/ggml/src/ggml-sycl/dpct/helper.hpp +32 -13
package/src/llama.cpp/ggml/src/ggml-sycl/element_wise.cpp +80 -61
package/src/llama.cpp/ggml/src/ggml-sycl/gemm.hpp +4 -4
package/src/llama.cpp/ggml/src/ggml-sycl/ggml-sycl.cpp +159 -114
package/src/llama.cpp/ggml/src/ggml-sycl/im2col.cpp +3 -2
package/src/llama.cpp/ggml/src/ggml-sycl/mmq.cpp +6 -6
package/src/llama.cpp/ggml/src/ggml-sycl/mmvq.cpp +6 -20
package/src/llama.cpp/ggml/src/ggml-sycl/norm.cpp +4 -3
package/src/llama.cpp/ggml/src/ggml-sycl/outprod.cpp +8 -8
package/src/llama.cpp/ggml/src/ggml-sycl/rope.cpp +4 -3
package/src/llama.cpp/ggml/src/ggml-sycl/softmax.cpp +7 -7
package/src/llama.cpp/ggml/src/ggml-sycl/tsembd.cpp +1 -0
package/src/llama.cpp/ggml/src/ggml-sycl/wkv6.cpp +4 -1
package/src/llama.cpp/ggml/src/ggml-threading.h +4 -2
package/src/llama.cpp/ggml/src/ggml-vulkan/CMakeLists.txt +21 -7
package/src/llama.cpp/ggml/src/ggml-vulkan/ggml-vulkan.cpp +1718 -399
package/src/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/CMakeLists.txt +3 -1
package/src/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/vulkan-shaders-gen.cpp +105 -31
package/src/llama.cpp/ggml/src/ggml.c +367 -207
package/src/llama.cpp/include/llama-cpp.h +25 -0
package/src/llama.cpp/include/llama.h +26 -19
package/src/llama.cpp/models/ggml-vocab-roberta-bpe.gguf.inp +112 -0
package/src/llama.cpp/models/ggml-vocab-roberta-bpe.gguf.out +46 -0
package/src/llama.cpp/pocs/CMakeLists.txt +3 -1
package/src/llama.cpp/pocs/vdot/CMakeLists.txt +2 -2
package/src/llama.cpp/src/CMakeLists.txt +2 -7
package/src/llama.cpp/src/llama-grammar.cpp +15 -15
package/src/llama.cpp/src/llama-grammar.h +2 -5
package/src/llama.cpp/src/llama-sampling.cpp +35 -90
package/src/llama.cpp/src/llama-vocab.cpp +6 -1
package/src/llama.cpp/src/llama.cpp +1748 -640
package/src/llama.cpp/src/unicode.cpp +62 -51
package/src/llama.cpp/src/unicode.h +9 -10
package/src/llama.cpp/tests/CMakeLists.txt +48 -37
package/src/llama.cpp/tests/test-arg-parser.cpp +2 -2
package/src/llama.cpp/tests/test-backend-ops.cpp +140 -21
package/src/llama.cpp/tests/test-chat-template.cpp +50 -4
package/src/llama.cpp/tests/test-gguf.cpp +1303 -0
package/src/llama.cpp/tests/test-grammar-integration.cpp +3 -6
package/src/llama.cpp/tests/test-llama-grammar.cpp +2 -4
package/src/llama.cpp/tests/test-quantize-fns.cpp +3 -3
package/src/llama.cpp/tests/test-rope.cpp +61 -20
package/src/llama.cpp/tests/test-sampling.cpp +2 -2
package/src/llama.cpp/.github/workflows/nix-ci-aarch64.yml +0 -72
package/src/llama.cpp/.github/workflows/nix-ci.yml +0 -79
package/src/llama.cpp/.github/workflows/nix-flake-update.yml +0 -22
package/src/llama.cpp/.github/workflows/nix-publish-flake.yml +0 -36
package/src/llama.cpp/ggml/include/ggml-amx.h +0 -25
package/src/llama.cpp/ggml/src/ggml-aarch64.c +0 -129
package/src/llama.cpp/ggml/src/ggml-aarch64.h +0 -19
package/src/llama.cpp/ggml/src/ggml-amx/CMakeLists.txt +0 -107
package/src/llama.cpp/ggml/src/ggml-amx/ggml-amx.cpp +0 -446

package/src/llama.cpp/ggml/src/ggml-sycl/mmq.cpp CHANGED Viewed

@@ -813,7 +813,7 @@ load_tiles_q4_K(const void *__restrict__ vx, int *__restrict__ x_ql,
         x_ql[i * (WARP_SIZE + 1) + k] = get_int_from_uint8_aligned(bxi->qs, kqsx);
     }
-    const int blocks_per_tile_x_row = WARP_SIZE / QI4_K; // == 1 if QK_K == 256
+    constexpr int blocks_per_tile_x_row = QI4_K > WARP_SIZE ? 1 : WARP_SIZE / QI4_K; // == 1 if QK_K == 256
     const int kbxd = k % blocks_per_tile_x_row;          // == 0 if QK_K == 256
 #pragma unroll
@@ -961,7 +961,7 @@ load_tiles_q5_K(const void *__restrict__ vx, int *__restrict__ x_ql,
         x_ql[i * (2*WARP_SIZE + 1) + kq1] = ql1 | qh1;
     }
-    const int blocks_per_tile_x_row = WARP_SIZE / QI5_K; // == 1 if QK_K == 256
+    constexpr int blocks_per_tile_x_row = QI5_K > WARP_SIZE ? 1 : WARP_SIZE / QI5_K; // == 1 if QK_K == 256
     const int kbxd = k % blocks_per_tile_x_row;          // == 0 if QK_K == 256
 #pragma unroll
@@ -1109,7 +1109,7 @@ load_tiles_q6_K(const void *__restrict__ vx, int *__restrict__ x_ql,
                                                  dpct::sub_sat());
     }
-    const int blocks_per_tile_x_row = WARP_SIZE / QI6_K; // == 1 if QK_K == 256
+    constexpr int blocks_per_tile_x_row = QI6_K > WARP_SIZE ? 1 : WARP_SIZE / QI6_K; // == 1 if QK_K == 256
     const int kbxd = k % blocks_per_tile_x_row;          // == 0 if QK_K == 256
     float * x_dmf = (float *) x_dm;
@@ -3020,9 +3020,9 @@ void ggml_sycl_op_mul_mat_q(
             break;
     }
-    (void) src1;
-    (void) dst;
-    (void) src1_ddf_i;
+    GGML_UNUSED(src1);
+    GGML_UNUSED(dst);
+    GGML_UNUSED(src1_ddf_i);
 }
 catch (sycl::exception const &exc) {
   std::cerr << exc.what() << "Exception caught at file:" << __FILE__

package/src/llama.cpp/ggml/src/ggml-sycl/mmvq.cpp CHANGED Viewed

@@ -753,11 +753,7 @@ static void mul_mat_vec_iq2_xs_q8_1_sycl(const void *vx, const void *vy,
     const sycl::range<3> block_nums(1, 1, block_num_y);
     const sycl::range<3> block_dims(1, GGML_SYCL_MMV_Y, QK_WARP_SIZE);
     {
-        stream->submit([&](sycl::handler &cgh) {
-            auto iq2xs_grid_ptr_ct1 = &iq2xs_grid[0];
-            auto ksigns64_ptr_ct1 = &ksigns64[0];
+        stream->submit([&](sycl::handler & cgh) {
             cgh.parallel_for(
                 sycl::nd_range<3>(block_nums * block_dims, block_dims),
                 [=](sycl::nd_item<3> item_ct1)
@@ -780,9 +776,6 @@ static void mul_mat_vec_iq2_s_q8_1_sycl(const void *vx, const void *vy,
     {
         stream->submit([&](sycl::handler &cgh) {
-            auto iq2xs_grid_ptr_ct1 = &iq2xs_grid[0];
-            auto ksigns64_ptr_ct1 = &ksigns64[0];
             cgh.parallel_for(
                 sycl::nd_range<3>(block_nums * block_dims, block_dims),
                 [=](sycl::nd_item<3> item_ct1)
@@ -805,9 +798,6 @@ static void mul_mat_vec_iq3_xxs_q8_1_sycl(const void *vx, const void *vy,
     {
         stream->submit([&](sycl::handler &cgh) {
-            auto iq3xxs_grid_ptr_ct1 = &iq3xxs_grid[0];
-            auto ksigns64_ptr_ct1 = &ksigns64[0];
             cgh.parallel_for(
                 sycl::nd_range<3>(block_nums * block_dims, block_dims),
                 [=](sycl::nd_item<3> item_ct1)
@@ -830,8 +820,6 @@ static void mul_mat_vec_iq3_s_q8_1_sycl(const void *vx, const void *vy,
     {
         stream->submit([&](sycl::handler &cgh) {
-            auto iq3s_grid_ptr_ct1 = &iq3s_grid[0];
             cgh.parallel_for(
                 sycl::nd_range<3>(block_nums * block_dims, block_dims),
                 [=](sycl::nd_item<3> item_ct1)
@@ -854,9 +842,6 @@ static void mul_mat_vec_iq1_s_q8_1_sycl(const void *vx, const void *vy,
     {
         stream->submit([&](sycl::handler &cgh) {
-            auto iq1s_grid_ptr_ct1 = &iq1s_grid_gpu[0];
-            auto ksigns64_ptr_ct1 = &ksigns64[0];
             cgh.parallel_for(
                 sycl::nd_range<3>(block_nums * block_dims, block_dims),
                 [=](sycl::nd_item<3> item_ct1)
@@ -954,7 +939,7 @@ void ggml_sycl_op_mul_mat_vec_q(
     const size_t q8_1_bs = QK8_1;
     // the main device has a larger memory buffer to hold the results from all GPUs
     // nrows_dst == nrows of the matrix that the kernel writes into
-    const int64_t nrows_dst = id == ctx.device ? ne00 : row_diff;
     for (int i = 0; i < src1_ncols; i++)
     {
         const size_t src1_ddq_i_offset = i * src1_padded_col_size * q8_1_ts / q8_1_bs;
@@ -1023,7 +1008,8 @@ void ggml_sycl_op_mul_mat_vec_q(
             break;
         }
     }
-    (void) src1;
-    (void) dst;
-    (void) src1_ddf_i;
+    GGML_UNUSED(src1);
+    GGML_UNUSED(dst);
+    GGML_UNUSED(src1_ddf_i);
+    GGML_UNUSED(ctx);
 }

package/src/llama.cpp/ggml/src/ggml-sycl/norm.cpp CHANGED Viewed

@@ -31,7 +31,7 @@ static void norm_f32(const float* x, float* dst, const int ncols, const float ep
         */
         item_ct1.barrier(sycl::access::fence_space::local_space);
         mean_var = 0.f;
-        int nreduce = nwarps / WARP_SIZE;
+        size_t nreduce = nwarps / WARP_SIZE;
         for (size_t i = 0; i < nreduce; i += 1)
         {
             mean_var += s_sum[lane_id + i * WARP_SIZE];
@@ -55,7 +55,7 @@ static void group_norm_f32(const float* x, float* dst, const int group_size, con
     const int nthreads = item_ct1.get_local_range(2);
     const int nwarps = nthreads / WARP_SIZE;
     start += item_ct1.get_local_id(2);
-    int nreduce = nwarps / WARP_SIZE;
+    size_t nreduce = nwarps / WARP_SIZE;
     if (end >= ne_elements) {
         end = ne_elements;
@@ -163,7 +163,7 @@ static void rms_norm_f32(const float* x, float* dst, const int ncols, const floa
         converged control flow. You may need to adjust the code.
         */
         item_ct1.barrier(sycl::access::fence_space::local_space);
-        int nreduce = nwarps / WARP_SIZE;
+        size_t nreduce = nwarps / WARP_SIZE;
         tmp = 0.f;
         for (size_t i = 0; i < nreduce; i += 1)
         {
@@ -352,6 +352,7 @@ void ggml_sycl_op_group_norm(ggml_backend_sycl_context& ctx, const ggml_tensor*
     (void)src1;
     (void)dst;
     (void)src1_dd;
+    GGML_UNUSED(ctx);
 }
 void ggml_sycl_op_rms_norm(ggml_backend_sycl_context& ctx, const ggml_tensor* src0,

package/src/llama.cpp/ggml/src/ggml-sycl/outprod.cpp CHANGED Viewed

@@ -40,14 +40,14 @@ void ggml_sycl_op_out_prod(ggml_backend_sycl_context& ctx, const ggml_tensor* sr
     try {
         // Perform matrix multiplication using oneMKL GEMM
-        oneapi::mkl::blas::column_major::gemm(*stream,
-            oneapi::mkl::transpose::nontrans, src1_op,
-            ne0, ne1, ne01,
-            alpha,
-            src0_d, ne00,
-            src1_d, ldb,
-            beta,
-            dst_d, ne0);
+#ifdef GGML_SYCL_NVIDIA
+        oneapi::mkl::blas::column_major::gemm(oneapi::mkl::backend_selector<oneapi::mkl::backend::cublas>{ *stream },
+                                              oneapi::mkl::transpose::nontrans, src1_op, ne0, ne1, ne01, alpha, src0_d,
+                                              ne00, src1_d, ldb, beta, dst_d, ne0);
+#else
+        oneapi::mkl::blas::column_major::gemm(*stream, oneapi::mkl::transpose::nontrans, src1_op, ne0, ne1, ne01, alpha,
+                                              src0_d, ne00, src1_d, ldb, beta, dst_d, ne0);
+#endif
     }
     catch (sycl::exception const& exc) {
         std::cerr << exc.what() << std::endl;

package/src/llama.cpp/ggml/src/ggml-sycl/rope.cpp CHANGED Viewed

@@ -269,7 +269,8 @@ void ggml_sycl_op_rope(
         }
     }
-    (void) src1;
-    (void) dst;
-    (void) src1_dd;
+    GGML_UNUSED(src1);
+    GGML_UNUSED(dst);
+    GGML_UNUSED(src1_dd);
+    GGML_UNUSED(ctx);
 }

package/src/llama.cpp/ggml/src/ggml-sycl/softmax.cpp CHANGED Viewed

@@ -16,7 +16,7 @@ static void soft_max_f32(const float * x, const float * mask, float * dst, const
     const int lane_id = item_ct1.get_local_id(2) % WARP_SIZE;
     const int nthreads = block_size;
     const int nwarps = nthreads / WARP_SIZE;
-    int nreduce = nwarps / WARP_SIZE;
+    size_t nreduce = nwarps / WARP_SIZE;
     float slope = 1.0f;
     // ALiBi
@@ -53,8 +53,9 @@ static void soft_max_f32(const float * x, const float * mask, float * dst, const
     if (block_size > WARP_SIZE) {
         if (warp_id == 0) {
             buf[lane_id] = -INFINITY;
-            for (size_t i = 1; i < nreduce; i += 1)
+            for (size_t i = 1; i < nreduce; i += 1) {
                 buf[lane_id + i * WARP_SIZE] = -INFINITY;
+            }
         }
         item_ct1.barrier(sycl::access::fence_space::local_space);
@@ -63,8 +64,7 @@ static void soft_max_f32(const float * x, const float * mask, float * dst, const
         }
         item_ct1.barrier(sycl::access::fence_space::local_space);
         max_val = buf[lane_id];
-        for (size_t i = 1; i < nreduce; i += 1)
-        {
+        for (size_t i = 1; i < nreduce; i += 1) {
             max_val = std::max(max_val, buf[lane_id + i * WARP_SIZE]);
         }
         max_val = warp_reduce_max(max_val, item_ct1);
@@ -89,8 +89,9 @@ static void soft_max_f32(const float * x, const float * mask, float * dst, const
         item_ct1.barrier(sycl::access::fence_space::local_space);
         if (warp_id == 0) {
             buf[lane_id] = 0.f;
-            for (size_t i = 1; i < nreduce; i += 1)
+            for (size_t i = 1; i < nreduce; i += 1) {
                 buf[lane_id + i * WARP_SIZE] = 0.f;
+            }
         }
         item_ct1.barrier(sycl::access::fence_space::local_space);
@@ -100,8 +101,7 @@ static void soft_max_f32(const float * x, const float * mask, float * dst, const
         item_ct1.barrier(sycl::access::fence_space::local_space);
         tmp = buf[lane_id];
-        for (size_t i = 1; i < nreduce; i += 1)
-        {
+        for (size_t i = 1; i < nreduce; i += 1) {
             tmp += buf[lane_id + i * WARP_SIZE];
         }
         tmp = warp_reduce_sum(tmp, item_ct1);

package/src/llama.cpp/ggml/src/ggml-sycl/tsembd.cpp CHANGED Viewed

@@ -68,4 +68,5 @@ void ggml_sycl_op_timestep_embedding(ggml_backend_sycl_context & ctx, const ggml
     const int max_period = dst->op_params[1];
     timestep_embedding_f32_sycl(src0_d, dst_d, src0->ne[0], dst->nb[1], dim, max_period, stream);
+    GGML_UNUSED(src1);
 }

package/src/llama.cpp/ggml/src/ggml-sycl/wkv6.cpp CHANGED Viewed

@@ -59,7 +59,7 @@ static void rwkv_wkv_f32_kernel(
         float y = 0;
         // Process in chunks of 4 for better vectorization
-        sycl::float4 k4, r4, tf4, td4, s4, kv4;
+        sycl::float4 k4, r4, tf4, td4, s4;
         #pragma unroll
         for (int j = 0; j < head_size; j += 4) {
             // Load data in vec4 chunks
@@ -135,4 +135,7 @@ void ggml_sycl_op_rwkv_wkv6(ggml_backend_sycl_context& ctx, const ggml_tensor* s
                 );
             });
     });
+    GGML_UNUSED(src0);
+    GGML_UNUSED(src1);
 }

package/src/llama.cpp/ggml/src/ggml-threading.h CHANGED Viewed

@@ -1,11 +1,13 @@
 #pragma once
+#include "ggml.h"
 #ifdef __cplusplus
 extern "C" {
 #endif
-void ggml_critical_section_start(void);
-void ggml_critical_section_end(void);
+GGML_API void ggml_critical_section_start(void);
+GGML_API void ggml_critical_section_end(void);
 #ifdef __cplusplus
 }

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

@@ -3,13 +3,27 @@ find_package(Vulkan COMPONENTS glslc REQUIRED)
 if (Vulkan_FOUND)
     message(STATUS "Vulkan found")
-    add_library(ggml-vulkan
-                ggml-vulkan.cpp
-                ../../include/ggml-vulkan.h
-                )
+    ggml_add_backend_library(ggml-vulkan
+                             ggml-vulkan.cpp
+                             ../../include/ggml-vulkan.h
+                            )
+    # Compile a test shader to determine whether GL_NV_cooperative_matrix2 is supported.
+    # If it's not, there will be an error to stderr.
+    # If it's supported, set a define to indicate that we should compile those shaders
+    execute_process(COMMAND ${Vulkan_GLSLC_EXECUTABLE} -o - -fshader-stage=compute --target-env=vulkan1.3 "${CMAKE_CURRENT_SOURCE_DIR}/vulkan-shaders/test_coopmat2_support.comp"
+                    OUTPUT_VARIABLE glslc_output
+                    ERROR_VARIABLE glslc_error)
+    if (${glslc_error} MATCHES ".*extension not supported: GL_NV_cooperative_matrix2.*")
+        message(STATUS "GL_NV_cooperative_matrix2 not supported by glslc")
+    else()
+        message(STATUS "GL_NV_cooperative_matrix2 supported by glslc")
+        add_compile_definitions(GGML_VULKAN_COOPMAT2_GLSLC_SUPPORT)
+    endif()
-    target_link_libraries(ggml-vulkan PRIVATE ggml-base Vulkan::Vulkan)
-    target_include_directories(ggml-vulkan PRIVATE . .. ${CMAKE_CURRENT_BINARY_DIR})
+    target_link_libraries(ggml-vulkan PRIVATE Vulkan::Vulkan)
+    target_include_directories(ggml-vulkan PRIVATE ${CMAKE_CURRENT_BINARY_DIR})
     # Workaround to the "can't dereference invalidated vector iterator" bug in clang-cl debug build
     # Posssibly relevant: https://stackoverflow.com/questions/74748276/visual-studio-no-displays-the-correct-length-of-stdvector
@@ -67,7 +81,7 @@ if (Vulkan_FOUND)
             --target-cpp ${_ggml_vk_source}
             --no-clean
-        DEPENDS ${_ggml_vk_shader_deps}
+        DEPENDS ${_ggml_vk_shader_deps} ${_ggml_vk_genshaders_cmd}
         COMMENT "Generate vulkan shaders"
     )