llama_cpp 0.16.1 → 0.17.0

data/vendor/tmp/llama.cpp/ggml-backend.h

@@ -1,233 +0,0 @@
-#pragma once
-
-#include "ggml.h"
-#include "ggml-alloc.h"
-
-#ifdef  __cplusplus
-extern "C" {
-#endif
-
-    typedef struct ggml_backend_buffer_type * ggml_backend_buffer_type_t;
-    typedef struct ggml_backend_buffer * ggml_backend_buffer_t;
-    typedef struct ggml_backend_event * ggml_backend_event_t;
-    typedef struct ggml_backend * ggml_backend_t;
-    typedef void * ggml_backend_graph_plan_t;
-
-    //
-    // Backend buffer
-    //
-
-    // buffer type
-    GGML_API           const char *          ggml_backend_buft_name            (ggml_backend_buffer_type_t buft);
-    GGML_API GGML_CALL ggml_backend_buffer_t ggml_backend_buft_alloc_buffer    (ggml_backend_buffer_type_t buft, size_t size);
-    GGML_API           size_t                ggml_backend_buft_get_alignment   (ggml_backend_buffer_type_t buft);
-    GGML_API           size_t                ggml_backend_buft_get_max_size    (ggml_backend_buffer_type_t buft);
-    GGML_API GGML_CALL size_t                ggml_backend_buft_get_alloc_size  (ggml_backend_buffer_type_t buft, struct ggml_tensor * tensor);
-    GGML_API           bool                  ggml_backend_buft_is_host         (ggml_backend_buffer_type_t buft);
-
-    // buffer
-    enum ggml_backend_buffer_usage {
-        GGML_BACKEND_BUFFER_USAGE_ANY = 0,
-        GGML_BACKEND_BUFFER_USAGE_WEIGHTS = 1,
-    };
-
-    GGML_API           const char *               ggml_backend_buffer_name          (ggml_backend_buffer_t buffer);
-    GGML_API           void                       ggml_backend_buffer_free          (ggml_backend_buffer_t buffer);
-    GGML_API           void *                     ggml_backend_buffer_get_base      (ggml_backend_buffer_t buffer);
-    GGML_API           size_t                     ggml_backend_buffer_get_size      (ggml_backend_buffer_t buffer);
-    GGML_API GGML_CALL void                       ggml_backend_buffer_init_tensor   (ggml_backend_buffer_t buffer, struct ggml_tensor * tensor);
-    GGML_API           size_t                     ggml_backend_buffer_get_alignment (ggml_backend_buffer_t buffer);
-    GGML_API           size_t                     ggml_backend_buffer_get_max_size  (ggml_backend_buffer_t buffer);
-    GGML_API           size_t                     ggml_backend_buffer_get_alloc_size(ggml_backend_buffer_t buffer, struct ggml_tensor * tensor);
-    GGML_API           void                       ggml_backend_buffer_clear         (ggml_backend_buffer_t buffer, uint8_t value);
-    GGML_API           bool                       ggml_backend_buffer_is_host       (ggml_backend_buffer_t buffer);
-    GGML_API           void                       ggml_backend_buffer_set_usage     (ggml_backend_buffer_t buffer, enum ggml_backend_buffer_usage usage);
-    GGML_API           ggml_backend_buffer_type_t ggml_backend_buffer_get_type      (ggml_backend_buffer_t buffer);
-    GGML_API           void                       ggml_backend_buffer_reset         (ggml_backend_buffer_t buffer);
-
-    //
-    // Backend
-    //
-
-    GGML_API ggml_guid_t  ggml_backend_guid(ggml_backend_t backend);
-    GGML_API const char * ggml_backend_name(ggml_backend_t backend);
-    GGML_API void         ggml_backend_free(ggml_backend_t backend);
-
-    GGML_API ggml_backend_buffer_type_t ggml_backend_get_default_buffer_type(ggml_backend_t backend);
-    GGML_API ggml_backend_buffer_t      ggml_backend_alloc_buffer(ggml_backend_t backend, size_t size);
-    GGML_API size_t                     ggml_backend_get_alignment(ggml_backend_t backend);
-    GGML_API size_t                     ggml_backend_get_max_size(ggml_backend_t backend);
-
-    GGML_API void ggml_backend_tensor_set_async(ggml_backend_t backend,       struct ggml_tensor * tensor, const void * data, size_t offset, size_t size);
-    GGML_API void ggml_backend_tensor_get_async(ggml_backend_t backend, const struct ggml_tensor * tensor,       void * data, size_t offset, size_t size);
-
-    GGML_API GGML_CALL void ggml_backend_tensor_set(      struct ggml_tensor * tensor, const void * data, size_t offset, size_t size);
-    GGML_API GGML_CALL void ggml_backend_tensor_get(const struct ggml_tensor * tensor,       void * data, size_t offset, size_t size);
-
-    GGML_API void ggml_backend_synchronize(ggml_backend_t backend);
-
-    GGML_API ggml_backend_graph_plan_t ggml_backend_graph_plan_create(ggml_backend_t backend, struct ggml_cgraph * cgraph);
-    GGML_API void                      ggml_backend_graph_plan_free  (ggml_backend_t backend, ggml_backend_graph_plan_t plan);
-
-    GGML_API enum ggml_status ggml_backend_graph_plan_compute (ggml_backend_t backend, ggml_backend_graph_plan_t plan);
-    GGML_API enum ggml_status ggml_backend_graph_compute      (ggml_backend_t backend, struct ggml_cgraph * cgraph);
-    GGML_API enum ggml_status ggml_backend_graph_compute_async(ggml_backend_t backend, struct ggml_cgraph * cgraph);
-    GGML_API bool ggml_backend_supports_op(ggml_backend_t backend, const struct ggml_tensor * op);
-    GGML_API bool ggml_backend_supports_buft(ggml_backend_t backend, ggml_backend_buffer_type_t buft);
-    GGML_API bool ggml_backend_offload_op(ggml_backend_t backend, const struct ggml_tensor * op);
-
-    // tensor copy between different backends
-    GGML_API void ggml_backend_tensor_copy(struct ggml_tensor * src, struct ggml_tensor * dst);
-
-    // asynchronous copy
-    // the copy is performed after all the currently queued operations in backend_src
-    // backend_dst will wait for the copy to complete before performing other operations
-    // automatic fallback to sync copy if async is not supported
-    GGML_API void ggml_backend_tensor_copy_async(ggml_backend_t backend_src, ggml_backend_t backend_dst, struct ggml_tensor * src, struct ggml_tensor * dst);
-
-    // events
-    GGML_API ggml_backend_event_t   ggml_backend_event_new        (ggml_backend_t backend);
-    GGML_API void                   ggml_backend_event_free       (ggml_backend_event_t event);
-    GGML_API void                   ggml_backend_event_record     (ggml_backend_event_t event);
-    GGML_API void                   ggml_backend_event_synchronize(ggml_backend_event_t event);
-    GGML_API void                   ggml_backend_event_wait       (ggml_backend_t backend, ggml_backend_event_t event);
-
-    //
-    // CPU backend
-    //
-
-    GGML_API ggml_backend_t ggml_backend_cpu_init(void);
-
-    GGML_API GGML_CALL bool ggml_backend_is_cpu                (ggml_backend_t backend);
-    GGML_API           void ggml_backend_cpu_set_n_threads     (ggml_backend_t backend_cpu, int n_threads);
-    GGML_API           void ggml_backend_cpu_set_abort_callback(ggml_backend_t backend_cpu, ggml_abort_callback abort_callback, void * abort_callback_data);
-
-    // Create a backend buffer from an existing pointer
-    GGML_API GGML_CALL ggml_backend_buffer_t ggml_backend_cpu_buffer_from_ptr(void * ptr, size_t size);
-
-    GGML_API GGML_CALL ggml_backend_buffer_type_t ggml_backend_cpu_buffer_type(void);
-
-#ifdef GGML_USE_CPU_HBM
-    GGML_API ggml_backend_buffer_type_t ggml_backend_cpu_hbm_buffer_type(void);
-#endif
-
-    //
-    // Backend registry
-    //
-
-    // The backend registry is a registry of all the available backends, and allows initializing backends in a generic way
-
-    GGML_API size_t                     ggml_backend_reg_get_count(void);
-    GGML_API size_t                     ggml_backend_reg_find_by_name(const char * name);
-    GGML_API ggml_backend_t             ggml_backend_reg_init_backend_from_str(const char * backend_str); // str is backend_name:params (params is optional)
-    GGML_API const char *               ggml_backend_reg_get_name(size_t i);
-    GGML_API ggml_backend_t             ggml_backend_reg_init_backend(size_t i, const char * params); // params is backend-specific
-    GGML_API ggml_backend_buffer_type_t ggml_backend_reg_get_default_buffer_type(size_t i);
-    GGML_API ggml_backend_buffer_t      ggml_backend_reg_alloc_buffer(size_t i, size_t size);
-
-    //
-    // Backend scheduler
-    //
-
-    // The backend scheduler allows for multiple backends to be used together
-    // Handles compute buffer allocation, assignment of tensors to backends, and copying of tensors between backends
-    // The backends are selected based on:
-    // - the backend that supports the operation
-    // - the location of the pre-allocated tensors (e.g. the weights)
-    /*
-      Example usage:
-
-        // operations that use tensors allocated in a buffer with USAGE_WEIGHTS will be assigned
-        // preferrably to run on the same backend as the buffer
-        ggml_backend_buffer_set_usage(buf_weights, GGML_BACKEND_BUFFER_USAGE_WEIGHTS);
-
-        sched = ggml_backend_sched_new({backend_gpu, backend_gpu2, backend_cpu}, NULL, num_backends, GGML_DEFAULT_GRAPH_SIZE, false);
-
-        // initialize buffers from a max size graph (optional)
-        reserve_graph = build_graph(sched, max_batch_size);
-
-        // manually assign nodes to a backend (optional, should not be needed in most cases)
-        struct ggml_tensor * node = ggml_mul_mat(ctx, ...);
-        ggml_backend_sched_set_tensor_backend(sched, node, backend_gpu);
-
-        ggml_backend_sched_reserve(sched, reserve_graph);
-
-        // compute
-        graph = build_graph(sched);
-        ggml_backend_sched_graph_compute(sched, graph);
-
-        // if there are graph inputs:
-        ggml_backend_sched_reset(sched);
-        ggml_backend_sched_alloc_graph(sched, graph);
-        ggml_backend_tensor_set(input_tensor, ...);
-        ggml_backend_sched_graph_compute(sched, graph);
-    }
-    */
-
-    struct ggml_backend_sched;
-    typedef struct ggml_backend_sched * ggml_backend_sched_t;
-
-    // when ask == true, the scheduler wants to know if the user wants to observe this node
-    // this allows the scheduler to batch nodes together in order to evaluate them in a single call
-    //
-    // when ask == false, the scheduler is passing the node tensor to the user for observation
-    // if the user returns false, the scheduler will cancel the graph compute
-    //
-    typedef bool (*ggml_backend_sched_eval_callback)(struct ggml_tensor * t, bool ask, void * user_data);
-
-    // Initialize a backend scheduler
-    GGML_API ggml_backend_sched_t ggml_backend_sched_new(ggml_backend_t * backends, ggml_backend_buffer_type_t * bufts, int n_backends, size_t graph_size, bool parallel);
-    GGML_API void                 ggml_backend_sched_free(ggml_backend_sched_t sched);
-
-    // Initialize backend buffers from a measure graph
-    GGML_API bool                 ggml_backend_sched_reserve(ggml_backend_sched_t sched, struct ggml_cgraph * measure_graph);
-
-    // Get the number of splits of the last graph
-    GGML_API int                  ggml_backend_sched_get_n_splits(ggml_backend_sched_t sched);
-    GGML_API int                  ggml_backend_sched_get_n_copies(ggml_backend_sched_t sched);
-
-    GGML_API size_t               ggml_backend_sched_get_buffer_size(ggml_backend_sched_t sched, ggml_backend_t backend);
-
-    GGML_API void                 ggml_backend_sched_set_tensor_backend(ggml_backend_sched_t sched, struct ggml_tensor * node, ggml_backend_t backend);
-    GGML_API ggml_backend_t       ggml_backend_sched_get_tensor_backend(ggml_backend_sched_t sched, struct ggml_tensor * node);
-
-    // Allocate and compute graph on the backend scheduler
-    GGML_API bool                 ggml_backend_sched_alloc_graph(ggml_backend_sched_t sched, struct ggml_cgraph * graph);
-    GGML_API enum ggml_status     ggml_backend_sched_graph_compute(ggml_backend_sched_t sched, struct ggml_cgraph * graph);
-    GGML_API enum ggml_status     ggml_backend_sched_graph_compute_async(ggml_backend_sched_t sched, struct ggml_cgraph * graph);
-    GGML_API void                 ggml_backend_sched_synchronize(ggml_backend_sched_t sched);
-
-    // Reset all assignments and allocators - must be called before changing the node backends
-    GGML_API void                 ggml_backend_sched_reset(ggml_backend_sched_t sched);
-
-    // Set a callback to be called for each resulting node during graph compute
-    GGML_API void                 ggml_backend_sched_set_eval_callback(ggml_backend_sched_t sched, ggml_backend_sched_eval_callback callback, void * user_data);
-
-    //
-    // Utils
-    //
-
-    struct ggml_backend_graph_copy {
-        ggml_backend_buffer_t buffer;
-        struct ggml_context * ctx_allocated;
-        struct ggml_context * ctx_unallocated;
-        struct ggml_cgraph * graph;
-    };
-
-    // Copy a graph to a different backend
-    GGML_API struct ggml_backend_graph_copy ggml_backend_graph_copy(ggml_backend_t backend, struct ggml_cgraph * graph);
-    GGML_API void                           ggml_backend_graph_copy_free(struct ggml_backend_graph_copy copy);
-
-    typedef bool (*GGML_CALL ggml_backend_eval_callback)(int node_index, struct ggml_tensor * t1, struct ggml_tensor * t2, void * user_data);
-
-    // Compare the output of two backends
-    GGML_API bool ggml_backend_compare_graph_backend(ggml_backend_t backend1, ggml_backend_t backend2, struct ggml_cgraph * graph, ggml_backend_eval_callback callback, void * user_data);
-
-    // Tensor initialization
-    GGML_API void ggml_backend_tensor_alloc(ggml_backend_buffer_t buffer, struct ggml_tensor * tensor, void * addr);
-    GGML_API void ggml_backend_view_init(struct ggml_tensor * tensor);
-
-
-#ifdef  __cplusplus
-}
-#endif

data/vendor/tmp/llama.cpp/ggml-blas.cpp

@@ -1,363 +0,0 @@
-#include "ggml-blas.h"
-#include "ggml-backend-impl.h"
-
-#include <future>
-#include <vector>
-
-#if defined(GGML_USE_ACCELERATE)
-#   include <Accelerate/Accelerate.h>
-#elif defined(GGML_BLAS_USE_MKL)
-#   include <mkl.h>
-#else
-#   include <cblas.h>
-#   ifdef BLIS_ENABLE_CBLAS
-#       include <blis.h>
-#   endif
-#endif
-
-struct ggml_backend_blas_context {
-    int n_threads = GGML_DEFAULT_N_THREADS;
-    std::unique_ptr<char[]> work_data;
-    size_t work_size = 0;
-#ifndef GGML_USE_OPENMP
-    std::vector<std::future<void>> tasks;
-#endif
-};
-
-// helper function to determine if it is better to use BLAS or not
-// for large matrices, BLAS is faster
-static bool ggml_backend_blas_use_blas(const struct ggml_tensor * dst) {
-    const struct ggml_tensor * src0 = dst->src[0];
-    const struct ggml_tensor * src1 = dst->src[1];
-
-    const int64_t ne10 = src1->ne[0];
-
-    const int64_t ne0 = dst->ne[0];
-    const int64_t ne1 = dst->ne[1];
-
-    // TODO: find the optimal values for these
-    if (ggml_is_contiguous(src0) &&
-        ggml_is_contiguous(src1) &&
-        src1->type == GGML_TYPE_F32 &&
-        (ne0 >= 32 && ne1 >= 32 && ne10 >= 32)) {
-
-        /*printf("BLAS: %d %d %d %d %d\n", ne0, ne1, ne10, ne00, ne01);*/
-        return true;
-    }
-
-    return false;
-}
-
-static void ggml_backend_blas_mul_mat(ggml_backend_blas_context * ctx, struct ggml_tensor * dst) {
-    const struct ggml_tensor * src0 = dst->src[0];
-    const struct ggml_tensor * src1 = dst->src[1];
-
-    GGML_TENSOR_BINARY_OP_LOCALS
-
-    const enum ggml_type type = src0->type;
-
-    GGML_ASSERT(ne0 == ne01);
-    GGML_ASSERT(ne1 == ne11);
-    GGML_ASSERT(ne2 == ne12);
-    GGML_ASSERT(ne3 == ne13);
-
-    // we don't support permuted src0 or src1
-    GGML_ASSERT(nb00 == ggml_type_size(type));
-    GGML_ASSERT(nb10 == ggml_type_size(src1->type));
-
-    // dst cannot be transposed or permuted
-    GGML_ASSERT(nb0 == sizeof(float));
-    GGML_ASSERT(nb0 <= nb1);
-    GGML_ASSERT(nb1 <= nb2);
-    GGML_ASSERT(nb2 <= nb3);
-
-    // broadcast factors
-    const int64_t r2 = ne12/ne02;
-    const int64_t r3 = ne13/ne03;
-
-    const int64_t ne_plane      = ne01*ne00;
-    const size_t  desired_wsize = type == GGML_TYPE_F32 ? 0 : ne03*ne02*ne_plane*sizeof(float);
-
-    if (ctx->work_size < desired_wsize) {
-        ctx->work_data.reset(new char[desired_wsize]);
-        ctx->work_size = desired_wsize;
-    }
-    void * wdata = ctx->work_data.get();
-
-    // convert src0 to float
-    if (type != GGML_TYPE_F32) {
-        ggml_type_traits_t type_traits = ggml_internal_get_type_traits(type);
-        ggml_to_float_t const to_float = type_traits.to_float;
-
-        for (int64_t i03 = 0; i03 < ne03; i03++) {
-            for (int64_t i02 = 0; i02 < ne02; i02++) {
-                const void  *       x      = (char *)  src0->data + i02*nb02          + i03*nb03;
-                      float * const wplane = (float *) wdata      + i02*ne_plane      + i03*ne02*ne_plane;
-
-                const int min_cols_per_thread = 4096;
-                const int min_rows_per_thread = std::max((int)(min_cols_per_thread/ne00), 1);
-                const int n_threads = std::max(std::min(ctx->n_threads, (int)(ne01/min_rows_per_thread)), 1);
-
-#ifdef GGML_USE_OPENMP
-                #pragma omp parallel for num_threads(n_threads)
-                for (int64_t i01 = 0; i01 < ne01; i01++) {
-                    to_float((const char *) x + i01*nb01, wplane + i01*ne00, ne00);
-                }
-#else
-                for (int i = 1; i < n_threads; i++) {
-                    const int64_t start =       i*ne01/n_threads;
-                    const int64_t end   = (i + 1)*ne01/n_threads;
-                    if (start < end) {
-                        ctx->tasks.push_back(std::async(std::launch::async, [=]() {
-                            for (int64_t i01 = start; i01 < end; i01++) {
-                                to_float((const char *) x + i01*nb01, wplane + i01*ne00, ne00);
-                            }
-                        }));
-                    }
-                }
-                {
-                    // reuse the current thread for the first task
-                    const int64_t start = 0;
-                    const int64_t end   = ne01/n_threads;
-                    for (int64_t i01 = start; i01 < end; i01++) {
-                        to_float((const char *) x + i01*nb01, wplane + i01*ne00, ne00);
-                    }
-                }
-#endif
-            }
-        }
-
-#ifndef GGML_USE_OPENMP
-        // wait for all tasks to finish
-        for (auto & task : ctx->tasks) {
-            task.get();
-        }
-        ctx->tasks.clear();
-#endif
-    }
-
-#if defined(OPENBLAS_VERSION)
-    openblas_set_num_threads(ctx->n_threads);
-#endif
-
-#if defined(BLIS_ENABLE_CBLAS)
-    bli_thread_set_num_threads(ctx->n_threads);
-#endif
-
-    for (int64_t i13 = 0; i13 < ne13; i13++) {
-        for (int64_t i12 = 0; i12 < ne12; i12++) {
-            const int64_t i03 = i13/r3;
-            const int64_t i02 = i12/r2;
-
-            const float * x = (float *) ((char *) src0->data + i02*nb02 + i03*nb03);
-            const float * y = (float *) ((char *) src1->data + i12*nb12 + i13*nb13);
-                  float * d = (float *) ((char *)  dst->data + i12*nb2  + i13*nb3);
-
-            if (type != GGML_TYPE_F32) {
-                x = (float *) wdata + i02*ne_plane + i03*ne02*ne_plane;
-            }
-
-            cblas_sgemm(CblasRowMajor, CblasNoTrans, CblasTrans,
-                        ne1, ne01, ne10,
-                        1.0f,   y, ne10,
-                                x, ne00,
-                        0.0f,   d, ne01);
-        }
-    }
-}
-
-static void ggml_backend_blas_out_prod(ggml_backend_blas_context * ctx, struct ggml_tensor * dst) {
-    const struct ggml_tensor * src0 = dst->src[0];
-    const struct ggml_tensor * src1 = dst->src[1];
-
-    GGML_TENSOR_BINARY_OP_LOCALS
-
-    GGML_ASSERT(ne0  == ne00);
-    GGML_ASSERT(ne1  == ne10);
-    GGML_ASSERT(ne2  == ne02);
-    GGML_ASSERT(ne02 == ne12);
-    GGML_ASSERT(ne3  == ne13);
-    GGML_ASSERT(ne03 == ne13);
-
-    // we don't support permuted src0 or src1
-    GGML_ASSERT(nb00 == sizeof(float));
-
-    // dst cannot be transposed or permuted
-    GGML_ASSERT(nb0 == sizeof(float));
-    // GGML_ASSERT(nb0 <= nb1);
-    // GGML_ASSERT(nb1 <= nb2);
-    // GGML_ASSERT(nb2 <= nb3);
-
-    // Arguments to ggml_compute_forward_out_prod (expressed as major,minor)
-    // src0: (k,n)
-    // src1: (k,m)
-    // dst:  (m,n)
-    //
-    // Arguments to sgemm (see https://github.com/Reference-LAPACK/lapack/blob/master/BLAS/SRC/sgemm.f)
-    // Also expressed as (major,minor)
-    // a: (m,k): so src1 transposed
-    // b: (k,n): so src0
-    // c: (m,n)
-    //
-    // However, if ggml_is_transposed(src1) is true, then
-    // src1->data already contains a transposed version, so sgemm mustn't
-    // transpose it further.
-
-    int n = src0->ne[0];
-    int k = src0->ne[1];
-    int m = src1->ne[0];
-
-    CBLAS_TRANSPOSE transposeA;
-    int lda;
-
-    if (!ggml_is_transposed(src1)) {
-        transposeA = CblasTrans;
-        lda = m;
-    } else {
-        transposeA = CblasNoTrans;
-        lda = k;
-    }
-
-    float * a = (float *) ((char *) src1->data);
-    float * b = (float *) ((char *) src0->data);
-    float * c = (float *) ((char *) dst->data);
-
-    cblas_sgemm(CblasRowMajor, transposeA, CblasNoTrans, m, n, k, 1.0, a, lda, b, n, 0.0, c, n);
-
-    GGML_UNUSED(ctx);
-}
-
-// backend interface
-
-GGML_CALL static const char * ggml_backend_blas_name(ggml_backend_t backend) {
-    return "BLAS";
-
-    GGML_UNUSED(backend);
-}
-
-GGML_CALL static void ggml_backend_blas_free(ggml_backend_t backend) {
-    ggml_backend_blas_context * ctx = (ggml_backend_blas_context *)backend->context;
-    delete ctx;
-    delete backend;
-}
-
-GGML_CALL static ggml_backend_buffer_type_t ggml_backend_blas_get_default_buffer_type(ggml_backend_t backend) {
-    return ggml_backend_cpu_buffer_type();
-
-    GGML_UNUSED(backend);
-}
-
-GGML_CALL static enum ggml_status ggml_backend_blas_graph_compute(ggml_backend_t backend, struct ggml_cgraph * cgraph) {
-    ggml_backend_blas_context * ctx = (ggml_backend_blas_context *)backend->context;
-
-    for (int i = 0; i < cgraph->n_nodes; i++) {
-        struct ggml_tensor * node = cgraph->nodes[i];
-
-        switch (node->op) {
-            case GGML_OP_MUL_MAT:
-                ggml_backend_blas_mul_mat(ctx, node);
-                break;
-
-            case GGML_OP_OUT_PROD:
-                ggml_backend_blas_out_prod(ctx, node);
-                break;
-
-            case GGML_OP_NONE:
-            case GGML_OP_RESHAPE:
-            case GGML_OP_VIEW:
-            case GGML_OP_PERMUTE:
-            case GGML_OP_TRANSPOSE:
-                break;
-
-            default:
-                fprintf(stderr, "%s: unsupported op %s\n", __func__, ggml_op_desc(node));
-                GGML_ASSERT(false);
-        }
-    }
-
-    return GGML_STATUS_SUCCESS;
-
-    GGML_UNUSED(backend);
-}
-
-GGML_CALL static bool ggml_backend_blas_supports_op(ggml_backend_t backend, const struct ggml_tensor * op) {
-    const struct ggml_tensor * src0 = op->src[0];
-    const struct ggml_tensor * src1 = op->src[1];
-
-    return (op->op == GGML_OP_MUL_MAT  && ggml_backend_blas_use_blas(op)) ||
-           (op->op == GGML_OP_OUT_PROD && op->src[0]->type == GGML_TYPE_F32 &&
-                                          op->src[1]->type == GGML_TYPE_F32 &&
-                                          ggml_is_matrix(src0) &&
-                                          ggml_is_matrix(src1) &&
-                                          ggml_is_contiguous(src0) &&
-                                          (ggml_is_contiguous(src1) || ggml_is_transposed(src1)));
-
-    GGML_UNUSED(backend);
-}
-
-GGML_CALL static bool ggml_backend_blas_supports_buft(ggml_backend_t backend, ggml_backend_buffer_type_t buft) {
-    return ggml_backend_buft_is_host(buft);
-
-    GGML_UNUSED(backend);
-}
-
-static struct ggml_backend_i blas_backend_i = {
-    /* .get_name                = */ ggml_backend_blas_name,
-    /* .free                    = */ ggml_backend_blas_free,
-    /* .get_default_buffer_type = */ ggml_backend_blas_get_default_buffer_type,
-    /* .set_tensor_async        = */ NULL,
-    /* .get_tensor_async        = */ NULL,
-    /* .cpy_tensor_async        = */ NULL,
-    /* .synchronize             = */ NULL,
-    /* .graph_plan_create       = */ NULL,
-    /* .graph_plan_free         = */ NULL,
-    /* .graph_plan_update       = */ NULL,
-    /* .graph_plan_compute      = */ NULL,
-    /* .graph_compute           = */ ggml_backend_blas_graph_compute,
-    /* .supports_op             = */ ggml_backend_blas_supports_op,
-    /* .supports_buft           = */ ggml_backend_blas_supports_buft,
-    /* .offload_op              = */ NULL,
-    /* .event_new               = */ NULL,
-    /* .event_free              = */ NULL,
-    /* .event_record            = */ NULL,
-    /* .event_wait              = */ NULL,
-    /* .event_synchronize       = */ NULL,
-};
-
-static ggml_guid_t ggml_backend_blas_guid(void) {
-    static ggml_guid guid = { 0x12, 0xa8, 0xae, 0xf4, 0xc0, 0x1e, 0x61, 0x97, 0x8f, 0xeb, 0x33, 0x04, 0xa1, 0x33, 0x51, 0x2d };
-    return &guid;
-}
-
-ggml_backend_t ggml_backend_blas_init(void) {
-    ggml_backend_blas_context * ctx = new ggml_backend_blas_context;
-
-    ggml_backend_t backend = new ggml_backend {
-        /* .guid      = */ ggml_backend_blas_guid(),
-        /* .interface = */ blas_backend_i,
-        /* .context   = */ ctx,
-    };
-
-#if !defined(NDEBUG) && defined(OPENBLAS_VERSION) && defined(GGML_USE_OPENMP)
-    if (openblas_get_parallel() != OPENBLAS_OPENMP) {
-        fprintf(stderr, "%s: warning: ggml is using OpenMP, but OpenBLAS was compiled without OpenMP support\n", __func__);
-    }
-#endif
-
-#if !defined(NDEBUG) && defined(BLIS_ENABLE_CBLAS) && defined(GGML_USE_OPENMP) && !defined(BLIS_ENABLE_OPENMP)
-    fprintf(stderr, "%s: warning: ggml is using OpenMP, but BLIS was compiled without OpenMP support\n", __func__);
-#endif
-
-    return backend;
-}
-
-GGML_CALL bool ggml_backend_is_blas(ggml_backend_t backend) {
-    return backend != NULL && ggml_guid_matches(backend->guid, ggml_backend_blas_guid());
-}
-
-void ggml_backend_blas_set_n_threads(ggml_backend_t backend_blas, int n_threads) {
-    GGML_ASSERT(ggml_backend_is_blas(backend_blas));
-
-    ggml_backend_blas_context * ctx = (ggml_backend_blas_context *)backend_blas->context;
-    ctx->n_threads = n_threads;
-}

data/vendor/tmp/llama.cpp/ggml-blas.h

@@ -1,23 +0,0 @@
-#pragma once
-
-#include "ggml.h"
-#include "ggml-backend.h"
-
-
-#ifdef  __cplusplus
-extern "C" {
-#endif
-
-// backend API
-GGML_API GGML_CALL ggml_backend_t ggml_backend_blas_init(void);
-
-GGML_API GGML_CALL bool ggml_backend_is_blas(ggml_backend_t backend);
-
-// number of threads used for conversion to float
-// for openblas and blis, this will also set the number of threads used for blas operations
-GGML_API GGML_CALL void ggml_backend_blas_set_n_threads(ggml_backend_t backend_blas, int n_threads);
-
-
-#ifdef  __cplusplus
-}
-#endif