whispercpp 1.3.6 → 1.3.7

data/ext/sources/ggml/src/ggml-cuda/template-instances/generate_cu_files.py

@@ -1,99 +0,0 @@
-#!/usr/bin/env python3
-
-from glob import glob
-import os
-
-HEAD_SIZES_KQ = [40, 64, 72, 80, 96, 112, 128, 256, 576]
-
-TYPES_KV = ["GGML_TYPE_F16", "GGML_TYPE_Q4_0", "GGML_TYPE_Q4_1", "GGML_TYPE_Q5_0", "GGML_TYPE_Q5_1", "GGML_TYPE_Q8_0"]
-
-SOURCE_FATTN_TILE = """// This file has been autogenerated by generate_cu_files.py, do not edit manually.
-
-#include "../fattn-tile.cuh"
-
-DECL_FATTN_TILE_CASE({head_size_kq}, {head_size_v});
-"""
-
-SOURCE_FATTN_VEC = """// This file has been autogenerated by generate_cu_files.py, do not edit manually.
-
-#include "../fattn-vec.cuh"
-
-DECL_FATTN_VEC_CASE( 64, {type_k}, {type_v});
-DECL_FATTN_VEC_CASE(128, {type_k}, {type_v});
-DECL_FATTN_VEC_CASE(256, {type_k}, {type_v});
-"""
-
-SOURCE_FATTN_MMA_START = """// This file has been autogenerated by generate_cu_files.py, do not edit manually.
-
-#include "../fattn-mma-f16.cuh"
-
-"""
-
-SOURCE_FATTN_MMA_CASE = "DECL_FATTN_MMA_F16_CASE({head_size_kq}, {head_size_v}, {ncols1}, {ncols2});\n"
-
-TYPES_MMQ = [
-    "GGML_TYPE_Q4_0", "GGML_TYPE_Q4_1", "GGML_TYPE_Q5_0", "GGML_TYPE_Q5_1", "GGML_TYPE_Q8_0",
-    "GGML_TYPE_Q2_K", "GGML_TYPE_Q3_K", "GGML_TYPE_Q4_K", "GGML_TYPE_Q5_K", "GGML_TYPE_Q6_K",
-    "GGML_TYPE_IQ2_XXS", "GGML_TYPE_IQ2_XS", "GGML_TYPE_IQ2_S", "GGML_TYPE_IQ3_XXS", "GGML_TYPE_IQ3_S",
-    "GGML_TYPE_IQ1_S", "GGML_TYPE_IQ4_NL", "GGML_TYPE_IQ4_XS", "GGML_TYPE_MXFP4"
-]
-
-SOURCE_MMQ = """// This file has been autogenerated by generate_cu_files.py, do not edit manually.
-
-#include "../mmq.cuh"
-
-DECL_MMQ_CASE({type});
-"""
-
-SOURCE_MMF = """// This file has been autogenerated by generate_cu_files.py, do not edit manually.
-
-#include "../mmf.cuh"
-
-DECL_MMF_CASE({type});
-"""
-
-
-def get_short_name(long_quant_name):
-    return long_quant_name.replace("GGML_TYPE_", "").lower()
-
-
-for filename in glob("*.cu"):
-    os.remove(filename)
-
-for head_size_kq in HEAD_SIZES_KQ:
-    head_size_v = head_size_kq if head_size_kq != 576 else 512
-    with open(f"fattn-tile-instance-dkq{head_size_kq}-dv{head_size_v}.cu", "w") as f:
-        f.write(SOURCE_FATTN_TILE.format(head_size_kq=head_size_kq, head_size_v=head_size_v))
-
-for type_k in TYPES_KV:
-    for type_v in TYPES_KV:
-        with open(f"fattn-vec-instance-{get_short_name(type_k)}-{get_short_name(type_v)}.cu", "w") as f:
-            f.write(SOURCE_FATTN_VEC.format(type_k=type_k, type_v=type_v))
-
-for ncols in [8, 16, 32, 64]:
-    for ncols2 in [1, 2, 4, 8, 16, 32]:
-        if ncols2 > ncols:
-            continue
-        ncols1 = ncols // ncols2
-        with open(f"fattn-mma-f16-instance-ncols1_{ncols1}-ncols2_{ncols2}.cu", "w") as f:
-            f.write(SOURCE_FATTN_MMA_START)
-
-            for head_size_kq in HEAD_SIZES_KQ:
-                if head_size_kq == 40:
-                    continue
-                if head_size_kq == 72:
-                    continue
-                if head_size_kq != 576 and ncols2 in (16, 32):
-                    continue
-                if head_size_kq == 576 and ncols2 not in (4, 16, 32):
-                    continue
-                head_size_v = head_size_kq if head_size_kq != 576 else 512
-                f.write(SOURCE_FATTN_MMA_CASE.format(ncols1=ncols1, ncols2=ncols2, head_size_kq=head_size_kq, head_size_v=head_size_v))
-
-for type in TYPES_MMQ:
-    with open(f"mmq-instance-{get_short_name(type)}.cu", "w") as f:
-        f.write(SOURCE_MMQ.format(type=type))
-
-for type in range(1, 17):
-    with open(f"mmf-instance-ncols_{type}.cu", "w") as f:
-        f.write(SOURCE_MMF.format(type=type))

data/ext/sources/ggml/src/ggml-hexagon/htp/htp-msg.h

@@ -1,155 +0,0 @@
-#ifndef HTP_MSG_H
-#define HTP_MSG_H
-
-#include <assert.h>
-
-// ggml-common.h must be included prio to this header
-
-// Mask to enable various stages of the Ops.
-// Used for debugging and profiling.
-enum {
-    HTP_OPMASK_QUEUE    = (1 << 0),  // Enable Queueing (ie calls into the DSP)
-    HTP_OPMASK_QUANTIZE = (1 << 1),  // Enable Quantize
-    HTP_OPMASK_COMPUTE  = (1 << 2),  // Enable Compute
-};
-
-// Op flags
-enum {
-    HTP_OPFLAGS_SKIP_QUANTIZE = (1 << 0),  // Skip dynamic quantization (reuse quantized tensors)
-    HTP_OPFLAGS_SKIP_COMPUTE  = (1 << 1),  // Skip actual computation (used for profiling)
-    HTP_OPFLAGS_EARLY_WAKEUP  = (1 << 2)   // Send early wakeup notification
-};
-
-enum htp_status {
-    HTP_STATUS_OK             = 1,
-    HTP_STATUS_INTERNAL_ERR   = 2,
-    HTP_STATUS_NO_SUPPORT     = 3,
-    HTP_STATUS_INVAL_PARAMS   = 4,
-    HTP_STATUS_VTCM_TOO_SMALL = 5,
-};
-
-// The values must match the ggml_type.
-// Duplicated here because we can't include full ggml.h in the htp build.
-// We have some static_asserts in the cpp code to ensure things are in sync.
-enum htp_data_type {
-    HTP_TYPE_F32   = 0,
-    HTP_TYPE_F16   = 1,
-    HTP_TYPE_Q4_0  = 2,
-    HTP_TYPE_Q8_0  = 8,
-    HTP_TYPE_I32   = 26,
-    HTP_TYPE_I64   = 27,
-    HTP_TYPE_MXFP4 = 39,
-    HTP_TYPE_COUNT
-};
-
-// Do not reorder first 4 (used as an index)
-enum htp_op {
-    HTP_OP_MUL = 0,
-    HTP_OP_ADD = 1,
-    HTP_OP_SUB = 2,
-    HTP_OP_DIV = 3,
-    HTP_OP_MUL_MAT,
-    HTP_OP_MUL_MAT_ID,
-    HTP_OP_RMS_NORM,
-    HTP_OP_UNARY_SILU,
-    HTP_OP_UNARY_GELU,
-    HTP_OP_GLU_SWIGLU,
-    HTP_OP_GLU_SWIGLU_OAI,
-    HTP_OP_GLU_GEGLU,
-    HTP_OP_SOFTMAX,
-    HTP_OP_ADD_ID,
-    HTP_OP_ROPE,
-    HTP_OP_FLASH_ATTN_EXT,
-    HTP_OP_SET_ROWS,
-    HTP_OP_GET_ROWS,
-    HTP_OP_SCALE,
-    HTP_OP_CPY,
-    HTP_OP_ARGSORT,
-    HTP_OP_SQR,
-    HTP_OP_SQRT,
-    HTP_OP_SUM_ROWS,
-    HTP_OP_SSM_CONV,
-    INVALID
-};
-
-static inline size_t htp_t_block_size(uint32_t t) {
-    switch (t) {
-        case HTP_TYPE_F32:
-            return 1;
-        case HTP_TYPE_F16:
-            return 1;
-        case HTP_TYPE_Q4_0:
-            return QK4_0;
-        case HTP_TYPE_Q8_0:
-            return QK8_0;
-        case HTP_TYPE_MXFP4:
-            return QK_MXFP4;
-        default:
-            assert(0 && "unsupported HTP data type");
-    }
-    return 0;
-}
-
-static inline size_t htp_type_nbytes(uint32_t t) {
-    switch (t) {
-        case HTP_TYPE_F32:
-            return 4;
-        case HTP_TYPE_F16:
-            return 2;
-        case HTP_TYPE_Q4_0:
-            return sizeof(block_q4_0);
-        case HTP_TYPE_Q8_0:
-            return sizeof(block_q8_0);
-        case HTP_TYPE_MXFP4:
-            return sizeof(block_mxfp4);
-        default:
-            assert(0 && "unsupported HTP data type");
-    }
-    return 0;
-}
-
-// Internal types
-#define QK_Q4_0x4x2  256  // 4x Q4_0 blocks packed with next 4x Q4_0 blocks (size in bytes 128)
-#define QK_Q8_0x4x2  256  // 4x Q8_0 blocks concat with next 4x Q8_0 blocks
-#define QK_MXFP4x4x2 256  // 4x MXFP4 blocks concat with next 4x MXFP4 blocks
-
-#define HTP_MAX_DIMS 4
-
-struct htp_tensor {
-    uint32_t data;                // Buffer offset in the messages, and data pointer on the NSP
-    uint32_t type;                // Data type
-    uint32_t ne[HTP_MAX_DIMS];    // Number of elements
-    uint32_t nb[HTP_MAX_DIMS];    // Stride in bytes (see ggml.h ggml_tensor)
-};
-
-#define HTP_MAX_OP_PARAMS 64
-
-struct htp_general_req {
-    uint32_t op;  // GGML/HTP Op
-    int32_t  op_params[HTP_MAX_OP_PARAMS / sizeof(int32_t)];
-    // Params for the op, e.g. epsilon of RMS norm
-    uint32_t flags;          // Request flags
-
-    struct htp_tensor src0;  // Input0 tensor
-    struct htp_tensor src1;  // Input1 tensor
-    struct htp_tensor src2;  // Input2 tensor
-    struct htp_tensor src3;  // Input3 tensor
-    struct htp_tensor src4;  // Input4 tensor
-    struct htp_tensor dst;   // Output tensor
-
-    // should be multiple of 64 bytes (cacheline)
-};
-
-struct htp_general_rsp {
-    uint32_t op;           // GGML/HTP Op
-    uint32_t status;       // HTP_STATUS_...
-    uint32_t prof_usecs;   // Number of usec per request
-    uint32_t prof_cycles;  // Number of cycles per request
-    uint32_t prof_pkts;    // Number of instruction packets per request
-    uint8_t  unused[44];   // Pad to 64 bytes
-};
-
-#define HTP_MAX_MESSAGE_SIZE   sizeof(struct htp_general_req)
-#define HTP_MAX_PACKET_BUFFERS 8
-
-#endif /* HTP_MSG_H */

data/ext/sources/ggml/src/ggml-hexagon/op-desc.h

@@ -1,153 +0,0 @@
-#ifndef OP_DESC_H
-#define OP_DESC_H
-
-#define GGML_COMMON_IMPL_CPP
-#include "ggml-backend-impl.h"
-#include "ggml-common.h"
-
-#include <string>
-#include <stdio.h>
-
-struct op_desc {
-    char strides[64 * GGML_MAX_SRC];
-    char dims[64 * GGML_MAX_SRC];
-    char types[16 * GGML_MAX_SRC];
-    char buffs[64 * GGML_MAX_SRC];
-    char names[64 * GGML_MAX_SRC];
-
-    int format_tensor_dims(char * str, const struct ggml_tensor * t) {
-        if (t->ne[2] == 1 && t->ne[3] == 1) {
-            return sprintf(str, "%d:%d", (int) t->ne[0], (int) t->ne[1]);
-        } else {
-            return sprintf(str, "%d:%d:%d:%d", (int) t->ne[0], (int) t->ne[1], (int) t->ne[2], (int) t->ne[3]);
-        }
-    }
-
-    void format_op_dims(char * str, const struct ggml_tensor * t) {
-        char * p = str;
-
-        // append src0 and src1 (if any)
-        if (t->src[0]) {
-            p += format_tensor_dims(p, t->src[0]);
-
-            for (int i = 1; i < GGML_MAX_SRC && t->src[i]; i++) {
-                p += sprintf(p, " x ");
-                p += format_tensor_dims(p, t->src[i]);
-            }
-
-            p += sprintf(p, " -> ");
-        }
-
-        // format self dims separately for better visual alignment
-        char self[64];
-        format_tensor_dims(self, t);
-
-        p += sprintf(p, "%s", self);
-    }
-
-    int format_tensor_strides(char * str, const struct ggml_tensor * t) {
-        const char * c = ggml_is_contiguous(t) ? "" : "!";
-
-        if (t->ne[2] == 1 && t->ne[3] == 1) {
-            return sprintf(str, "%zu:%zu%s", (size_t) t->nb[0], (size_t) t->nb[1], c);
-        } else {
-            return sprintf(str, "%zu:%zu:%zu:%zu%s", (size_t) t->nb[0], (size_t) t->nb[1], (size_t) t->nb[2], (size_t) t->nb[3], c);
-        }
-    }
-
-    void format_op_strides(char * str, const struct ggml_tensor * t) {
-        char * p = str;
-
-        // append src0 and src1 (if any)
-        if (t->src[0]) {
-            p += format_tensor_strides(p, t->src[0]);
-
-            for (int i = 1; i < GGML_MAX_SRC && t->src[i]; i++) {
-                p += sprintf(p, " x ");
-                p += format_tensor_strides(p, t->src[i]);
-            }
-
-            p += sprintf(p, " -> ");
-        }
-
-        // format self dims separately for better visual alignment
-        char self[64];
-        format_tensor_strides(self, t);
-
-        p += sprintf(p, "%s", self);
-    }
-
-    void format_op_types(char * str, const struct ggml_tensor * t) {
-        char * p = str;
-
-        // append src0 and src1 (if any)
-        if (t->src[0]) {
-            p += sprintf(p, "%s", ggml_type_name(t->src[0]->type));
-
-            for (int i = 1; i < GGML_MAX_SRC && t->src[i]; i++) {
-                p += sprintf(p, " x ");
-                p += sprintf(p, "%s", ggml_type_name(t->src[i]->type));
-            }
-
-            p += sprintf(p, " -> ");
-        }
-
-        p += sprintf(p, "%s", ggml_type_name(t->type));
-    }
-
-    const char * tensor_buff_name(const struct ggml_tensor * t) {
-        if (t->buffer) {
-            return ggml_backend_buffer_name(t->buffer);
-        }
-        return "NONE";
-    }
-
-    void format_op_buffs(char * str, const struct ggml_tensor * t) {
-        char * p = str;
-
-        // append src0 and src1 (if any)
-        if (t->src[0]) {
-            p += sprintf(p, "%s", tensor_buff_name(t->src[0]));
-
-            for (int i = 1; i < GGML_MAX_SRC && t->src[i]; i++) {
-                p += sprintf(p, " x ");
-                p += sprintf(p, "%s", tensor_buff_name(t->src[i]));
-            }
-
-            p += sprintf(p, " -> ");
-        }
-
-        p += sprintf(p, "%s", tensor_buff_name(t));
-    }
-
-    void format_op_names(char * str, const struct ggml_tensor * t) {
-        char * p = str;
-
-        // append src0 and src1 (if any)
-        if (t->src[0]) {
-            p += sprintf(p, "%s", t->src[0]->name);
-
-            for (int i = 1; i < GGML_MAX_SRC && t->src[i]; i++) {
-                p += sprintf(p, " x ");
-                p += sprintf(p, "%s", t->src[i]->name);
-            }
-
-            p += sprintf(p, " -> ");
-        }
-
-        p += sprintf(p, "%s", t->name);
-    }
-
-    void format(const ggml_tensor * op) {
-        format_op_dims(dims, op);
-        format_op_strides(strides, op);
-        format_op_types(types, op);
-        format_op_buffs(buffs, op);
-        format_op_names(names, op);
-    }
-
-    op_desc() {}
-    op_desc(const ggml_tensor * op) { format(op); }
-};
-
-#endif // OP_DESC_H

data/ext/sources/ggml/src/ggml-opencl/kernels/embed_kernel.py

@@ -1,26 +0,0 @@
-#
-
-import sys
-import logging
-logger = logging.getLogger("opencl-embed-kernel")
-
-
-def main():
-    logging.basicConfig(level=logging.INFO)
-
-    if len(sys.argv) != 3:
-        logger.info("Usage: python embed_kernel.py <input_file> <output_file>")
-        sys.exit(1)
-
-    ifile = open(sys.argv[1], "r")
-    ofile = open(sys.argv[2], "w")
-
-    for i in ifile:
-        ofile.write('R"({})"\n'.format(i))
-
-    ifile.close()
-    ofile.close()
-
-
-if __name__ == "__main__":
-    main()

data/ext/sources/ggml/src/ggml-openvino/openvino/pass/eliminate_zp.cpp

@@ -1,123 +0,0 @@
-#include "eliminate_zp.h"
-
-#include <openvino/core/graph_util.hpp>
-#include <openvino/core/parallel.hpp>
-#include <openvino/core/rt_info.hpp>
-#include <openvino/op/constant.hpp>
-#include <openvino/op/convert.hpp>
-#include <openvino/op/multiply.hpp>
-#include <openvino/op/subtract.hpp>
-#include <openvino/pass/pattern/op/label.hpp>
-#include <openvino/pass/pattern/op/pattern.hpp>
-#include <openvino/pass/pattern/op/wrap_type.hpp>
-
-namespace ov {
-namespace frontend {
-namespace ggml {
-namespace pass {
-
-EliminateZeroPoints::EliminateZeroPoints() {
-    // Find pattern:
-    // (Multiply Any(scale)
-    //           (Subtract (Convert Constant(data)))
-    //                     (Convert Constant(zero_point)))
-    // where zero_point is a scalar
-    // If data is u4 and zp value is 8 (q4_0), Replace the Subtract with an i4 Constant whose value is data - zp_val
-    // If data is u8 and zp value is 128 (q8_0) or 32 (q6_k), Replace the Subtract with an i8 Constant
-
-    auto m_data_constant = ov::pass::pattern::wrap_type<ov::op::v0::Constant>();
-    auto m_data_convert = ov::pass::pattern::wrap_type<ov::op::v0::Convert>({m_data_constant});
-
-    auto m_zp_constant = ov::pass::pattern::wrap_type<ov::op::v0::Constant>();
-    auto m_zp_convert = ov::pass::pattern::wrap_type<ov::op::v0::Convert>({m_zp_constant});
-
-    auto m_subtract = ov::pass::pattern::wrap_type<ov::op::v1::Subtract>({m_data_convert, m_zp_convert});
-    auto m_scale = ov::pass::pattern::any_input();
-    auto m_multiply = ov::pass::pattern::wrap_type<ov::op::v1::Multiply>({m_scale, m_subtract});
-
-    const auto callback = [=](ov::pass::pattern::Matcher & m) {
-        const auto & pattern_map = m.get_pattern_value_map();
-
-        auto multiply_node =
-            std::dynamic_pointer_cast<ov::op::v1::Multiply>(pattern_map.at(m_multiply).get_node_shared_ptr());
-        auto subtract_node =
-            std::dynamic_pointer_cast<ov::op::v1::Subtract>(pattern_map.at(m_subtract).get_node_shared_ptr());
-        auto data_constant =
-            std::dynamic_pointer_cast<ov::op::v0::Constant>(pattern_map.at(m_data_constant).get_node_shared_ptr());
-        auto zp_constant =
-            std::dynamic_pointer_cast<ov::op::v0::Constant>(pattern_map.at(m_zp_constant).get_node_shared_ptr());
-
-        if (!multiply_node || !subtract_node || !data_constant || !zp_constant) {
-            return false;
-        }
-
-        if (ov::shape_size(zp_constant->get_shape()) != 1) {
-            return false;
-        }
-
-        auto data_type = data_constant->get_element_type();
-        auto zp_data = zp_constant->cast_vector<int>();
-
-        if (zp_data.empty()) {
-            return false;
-        }
-
-        int zp_value = zp_data[0];
-
-        bool should_eliminate = false;
-        ov::element::Type target_type;
-
-        if (data_type == ov::element::u4 && zp_value == 8) {
-            should_eliminate = true;
-            target_type = ov::element::i4;
-        } else if (data_type == ov::element::u8 && (zp_value == 128 || zp_value == 32)) {
-            should_eliminate = true;
-            target_type = ov::element::i8;
-        }
-
-        if (!should_eliminate) {
-            return false;
-        }
-
-        auto data_shape = data_constant->get_shape();
-        size_t total_elements = ov::shape_size(data_shape);
-
-        std::shared_ptr<ov::op::v0::Constant> new_constant;
-
-        // TODO improve performance
-        if (data_type == ov::element::u4) {
-            auto data_values = data_constant->cast_vector<uint8_t>();
-            std::vector<int8_t> adjusted_values(total_elements);
-
-            ov::parallel_for(total_elements, [&](size_t i) {
-                adjusted_values[i] = static_cast<int8_t>(static_cast<int>(data_values[i]) - 8);
-            });
-
-            new_constant = std::make_shared<ov::op::v0::Constant>(target_type, data_shape, adjusted_values);
-        } else if (data_type == ov::element::u8) {
-            auto data_values = data_constant->cast_vector<uint8_t>();
-            std::vector<int8_t> adjusted_values(total_elements);
-
-            ov::parallel_for(total_elements, [&, zp_value](size_t i) {
-                adjusted_values[i] = static_cast<int8_t>(static_cast<int>(data_values[i]) - zp_value);
-            });
-
-            new_constant = std::make_shared<ov::op::v0::Constant>(target_type, data_shape, adjusted_values);
-        }
-
-        auto new_convert =
-            std::make_shared<ov::op::v0::Convert>(new_constant, subtract_node->get_output_element_type(0));
-        ov::replace_node(subtract_node, new_convert);
-
-        return true;
-    };
-
-    register_matcher(
-        std::make_shared<ov::pass::pattern::Matcher>(m_multiply, "ov::frontend::ggml::pass::EliminateZeroPoints"),
-        callback);
-}
-
-}  // namespace pass
-}  // namespace ggml
-}  // namespace frontend
-}  // namespace ov

data/ext/sources/ggml/src/ggml-openvino/openvino/pass/eliminate_zp.h

@@ -1,17 +0,0 @@
-#include "openvino/pass/matcher_pass.hpp"
-
-namespace ov {
-namespace frontend {
-namespace ggml {
-namespace pass {
-
-class EliminateZeroPoints : public ov::pass::MatcherPass {
-public:
-    OPENVINO_MATCHER_PASS_RTTI("ov::frontend::ggml::pass::EliminateZeroPoints")
-    EliminateZeroPoints();
-};
-
-}  // namespace pass
-}  // namespace ggml
-}  // namespace frontend
-}  // namespace ov