whispercpp 1.3.5 → 1.3.6

data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/set_rows.wgsl

@@ -1,16 +1,37 @@
 enable f16;
 
+#ifdef DST_F32
+#define DST_INNER_TYPE f32
+#else
+#define DST_INNER_TYPE f16
+#endif
+
+#ifdef VEC4
+#define SRC_TYPE vec4<f32>
+#define DST_TYPE vec4<DST_INNER_TYPE>
+#define VEC_SIZE 4
+#else
+#define SRC_TYPE f32
+#define DST_TYPE DST_INNER_TYPE
+#define VEC_SIZE 1
+#endif
+
 @group(0) @binding(0)
-var<storage, read_write> src: array<f32>;
+var<storage, read_write> src: array<SRC_TYPE>;
 
 @group(0) @binding(1)
 var<storage, read_write> idx: array<u32>;
 
 @group(0) @binding(2)
-var<storage, read_write> dst: array<f16>;
+var<storage, read_write> dst: array<DST_TYPE>;
 
+#ifdef I64_IDX
 @group(0) @binding(3)
 var<storage, read_write> error: atomic<u32>;
+#define PARAMS_BINDING 4
+#else
+#define PARAMS_BINDING 3
+#endif
 
 struct Params {
     offset_src: u32, // in elements
@@ -41,16 +62,19 @@ struct Params {
     idx2: u32,
 };
 
-@group(0) @binding(4)
+@group(0) @binding(PARAMS_BINDING)
 var<uniform> params: Params;
 
-override wg_size: u32;
-@compute @workgroup_size(wg_size)
+@compute @workgroup_size(WG_SIZE)
 fn main(@builtin(global_invocation_id) gid: vec3<u32>) {
-    if (gid.x >= params.n_rows * params.ne2 * params.ne3) {
+    if (gid.x >= (params.ne3 * params.ne2 * params.n_rows * params.ne0) / VEC_SIZE) {
         return;
     }
-    var i = gid.x;
+
+    // getting the row from gid
+    let elems_per_row = params.ne0 / VEC_SIZE;
+    var i = gid.x / elems_per_row;
+
     let i_src3 = i / (params.ne2 * params.n_rows);
 
     i = i % (params.ne2 * params.n_rows);
@@ -61,9 +85,10 @@ fn main(@builtin(global_invocation_id) gid: vec3<u32>) {
     let i_idx1 = i_src2 % params.idx1;
     let i_idx0 = i_src1;
 
+#ifdef I64_IDX
     let idx_high = (params.offset_idx + i_idx0 * params.stride_idx0 + i_idx1 * params.stride_idx1 + i_idx2 * params.stride_idx2) * 2;
 
-    let idx_high_val = idx[idx_high];
+    let idx_val = idx[idx_high];
     let idx_low_val = idx[idx_high + 1];
 
     if (idx_low_val != 0) {
@@ -71,11 +96,14 @@ fn main(@builtin(global_invocation_id) gid: vec3<u32>) {
         atomicStore(&error, 1);
         return;
     }
+#else
+    let idx_i = params.offset_idx + i_idx0 * params.stride_idx0 + i_idx1 * params.stride_idx1 + i_idx2 * params.stride_idx2;
+    let idx_val = idx[idx_i];
+#endif
 
-    let i_dst_row = params.offset_dst + idx_high_val * params.stride_dst1 + i_src2 * params.stride_dst2 + i_src3 * params.stride_dst3;
+    let i_dst_row = params.offset_dst + idx_val * params.stride_dst1 + i_src2 * params.stride_dst2 + i_src3 * params.stride_dst3;
     let i_src_row = params.offset_src + i_src1 * params.stride_src1 + i_src2 * params.stride_src2 + i_src3 * params.stride_src3;
 
-    for (var i: u32 = 0; i < params.ne0; i++) {
-      dst[i_dst_row + i] = f16(src[i_src_row + i]);
-    }
+    let col_idx = (gid.x % elems_per_row);
+    dst[i_dst_row/VEC_SIZE + col_idx] = DST_TYPE(src[i_src_row/VEC_SIZE + col_idx]);
 }

data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/sum_rows.wgsl

@@ -0,0 +1,55 @@
+@group(0) @binding(0)
+var<storage, read_write> src: array<f32>;
+
+@group(0) @binding(1)
+var<storage, read_write> dst: array<f32>;
+
+struct Params {
+    offset_src: u32, // in elements
+    offset_dst: u32, // in elements
+
+    // Strides (in elements)
+    stride_src1: u32,
+    stride_src2: u32,
+    stride_src3: u32,
+
+    ne0: u32,
+    ne1: u32,
+    ne2: u32
+};
+
+@group(0) @binding(2)
+var<uniform> params: Params;
+
+var<workgroup> shared_sum: array<f32, WG_SIZE>;
+
+@compute @workgroup_size(WG_SIZE)
+fn main(@builtin(workgroup_id) wid: vec3<u32>,
+        @builtin(local_invocation_id) lid: vec3<u32>) {
+
+    var i = wid.x;
+    let i3 = i / (params.ne2 * params.ne1);
+    i = i % (params.ne2 * params.ne1);
+    let i2 = i / params.ne1;
+    let i1 = i % params.ne1;
+    let i_src_row = params.offset_src + i3 * params.stride_src3 + i2 * params.stride_src2 + i1 * params.stride_src1;
+    var local_sum: f32 = 0.0;
+    for (var col = lid.x; col < params.ne0; col += WG_SIZE) {
+        local_sum += src[i_src_row + col];
+    }
+    shared_sum[lid.x] = local_sum;
+    workgroupBarrier();
+    // reduce within workgroup
+    var offset: u32 = WG_SIZE >> 1;
+    while (offset > 0) {
+        if (lid.x < offset) {
+            shared_sum[lid.x] = shared_sum[lid.x] + shared_sum[lid.x + offset];
+        }
+        workgroupBarrier();
+        offset >>= 1;
+    }
+
+    if (lid.x == 0) {
+        dst[params.offset_dst + wid.x] = shared_sum[0];
+    }
+}

data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/unary.wgsl

@@ -0,0 +1,193 @@
+#ifdef TYPE_F16
+enable f16;
+#define TYPE f16
+#else
+#define TYPE f32
+#endif
+
+
+@group(0) @binding(0)
+var<storage, read_write> src: array<TYPE>;
+
+#ifndef INPLACE
+@group(0) @binding(1)
+var<storage, read_write> dst: array<TYPE>;
+#define PARAMS_BINDING 2
+#else
+#define PARAMS_BINDING 1
+#endif
+
+struct Params {
+    ne: u32,            // total number of elements
+    offset_src: u32,    // in elements
+    offset_dst: u32,    // in elements
+
+    // Strides (in elements)
+    stride_src0: u32,
+    stride_src1: u32,
+    stride_src2: u32,
+    stride_src3: u32,
+
+    // Logical shapes
+    ne0: u32,
+    ne1: u32,
+    ne2: u32,
+#ifdef CLAMP
+    clamp_min: f32,
+    clamp_max: f32,
+#endif
+#ifdef FILL
+    fill_val: f32,
+#endif
+#ifdef XIELU
+    alpha_n: f32,
+    alpha_p: f32,
+    beta: f32,
+    eps: f32,
+#endif
+
+};
+
+@group(0) @binding(PARAMS_BINDING)
+var<uniform> params: Params;
+
+@compute @workgroup_size(WG_SIZE)
+fn main(@builtin(global_invocation_id) gid: vec3<u32>) {
+    if (gid.x >= params.ne) {
+      return;
+    }
+    var i = gid.x;
+    let i3 = i / (params.ne2 * params.ne1 * params.ne0);
+    i = i % (params.ne2 * params.ne1 * params.ne0);
+    let i2 = i / (params.ne1 * params.ne0);
+    i = i % (params.ne1 * params.ne0);
+    let i1 = i / params.ne0;
+    let i0 = i % params.ne0;
+
+    let src_idx = i0 * params.stride_src0 + i1 * params.stride_src1 +
+                  i2 * params.stride_src2 + i3 * params.stride_src3;
+
+#ifdef ABS
+    let res = abs(src[params.offset_src + src_idx]);
+#endif
+#ifdef SGN
+    let res = select(TYPE(select(0.0, -1.0, src[params.offset_src + src_idx] < 0.0)), TYPE(1.0),
+                     src[params.offset_src + src_idx] > 0.0);
+#endif
+#ifdef NEG
+    let res = -src[params.offset_src + src_idx];
+#endif
+#ifdef STEP
+    let res = TYPE(select(0.0, 1.0, src[params.offset_src + src_idx] > 0.0));
+#endif
+#ifdef TANH
+    let res = tanh(clamp(src[params.offset_src + src_idx], -9.010913, 9.010913));
+#endif
+#ifdef RELU
+    let res = select(0.0, src[params.offset_src + src_idx], src[params.offset_src + src_idx] > 0.0);
+#endif
+#ifdef ELU
+    let res = select(exp(src[params.offset_src + src_idx]) - 1.0, src[params.offset_src + src_idx],
+                     src[params.offset_src + src_idx] > 0.0);
+#endif
+#ifdef HARDSIGMOID
+    let res = min(1.0, max(0.0, (src[params.offset_src + src_idx] + 3.0) / 6.0));
+#endif
+#ifdef SIGMOID
+    let res = 1.0 / (1.0 + exp(-src[params.offset_src + src_idx]));
+#endif
+#ifdef SILU
+    let res = src[params.offset_src + src_idx] / (1.0 + exp(-src[params.offset_src + src_idx]));
+#endif
+#ifdef EXP
+    let res = exp(src[params.offset_src + src_idx]);
+#endif
+#ifdef LOG
+    let res = TYPE(log(f32(src[params.offset_src + src_idx])));
+#endif
+#ifdef CLAMP
+    let res = clamp(src[params.offset_src + src_idx], TYPE(params.clamp_min), TYPE(params.clamp_max));
+#endif
+#ifdef FILL
+    let res = TYPE(params.fill_val);
+#endif
+#ifdef HARDSWISH
+    let res = src[params.offset_src + src_idx] *
+              min(1.0, max(0.0, (src[params.offset_src + src_idx] + 3.0) / 6.0));
+#endif
+#ifdef GELU
+    let res = 0.5 * src[params.offset_src + src_idx] *
+              (1.0 + tanh(clamp(sqrt(2.0 / 3.14159265) *
+                               (src[params.offset_src + src_idx] +
+                                0.044715 * pow(src[params.offset_src + src_idx], 3.0)),
+                               -9.010913, 9.010913)));
+#endif
+#ifdef GELU_QUICK
+    let res = src[params.offset_src + src_idx] * 0.5 *
+              (1.0 + tanh(clamp(0.79788456 *
+                               (src[params.offset_src + src_idx] +
+                                0.044715 * src[params.offset_src + src_idx] *
+                                    src[params.offset_src + src_idx] * src[params.offset_src + src_idx]),
+                               -9.010913, 9.010913)));
+#endif
+#ifdef GELU_ERF
+    let res = 0.5 * src[params.offset_src + src_idx] *
+              (1.0 + tanh(clamp(0.79788456 *
+                               (src[params.offset_src + src_idx] +
+                                0.044715 * src[params.offset_src + src_idx] *
+                                    src[params.offset_src + src_idx] * src[params.offset_src + src_idx]),
+                               -9.010913, 9.010913)));
+#endif
+#ifdef XIELU
+    let res =
+        select(((exp(min(src[params.offset_src + src_idx], TYPE(params.eps))) - 1.0) -
+                src[params.offset_src + src_idx]) *
+                   TYPE(params.alpha_n) +
+               TYPE(params.beta) * src[params.offset_src + src_idx],
+               TYPE(params.alpha_p) * src[params.offset_src + src_idx] *
+                   src[params.offset_src + src_idx] +
+                   TYPE(params.beta) * src[params.offset_src + src_idx],
+               src[params.offset_src + src_idx] > 0.0);
+#endif
+#ifdef SOFTPLUS
+    let src_f32 = f32(src[params.offset_src + src_idx]);
+    let res = TYPE(select(log(1.0 + exp(src_f32)), src_f32, src_f32 > 20.0));
+#endif
+#ifdef EXPM1
+    let res = exp(src[params.offset_src + src_idx]) - 1.0;
+#endif
+#ifdef FLOOR
+    let res = floor(src[params.offset_src + src_idx]);
+#endif
+#ifdef CEIL
+    let res = ceil(src[params.offset_src + src_idx]);
+#endif
+#ifdef ROUND
+    let src_f32 = f32(src[params.offset_src + src_idx]);
+    let result = select(ceil(src_f32 - 0.5), floor(src_f32 + 0.5), src_f32 >= 0.0);
+    let res = TYPE(result);
+#endif
+#ifdef TRUNC
+    let res = trunc(src[params.offset_src + src_idx]);
+#endif
+#ifdef SQR
+    let res = src[params.offset_src + src_idx] * src[params.offset_src + src_idx];
+#endif
+#ifdef SQRT
+    let res = sqrt(src[params.offset_src + src_idx]);
+#endif
+#ifdef SIN
+    let res_f32 = sin(f32(src[params.offset_src + src_idx]));
+    let res = TYPE(res_f32);
+#endif
+#ifdef COS
+    let res_f32 = cos(f32(src[params.offset_src + src_idx]));
+    let res = TYPE(res_f32);
+#endif
+
+#ifdef INPLACE
+    src[params.offset_src + src_idx] = res;
+#else
+    dst[params.offset_dst + gid.x] = res;
+#endif
+}

data/ext/sources/ggml/src/ggml-zdnn/ggml-zdnn.cpp

@@ -58,6 +58,10 @@ static enum ggml_status ggml_zdnn_graph_compute(ggml_backend_t backend, ggml_cgr
             continue;
         }
 
+        if ((node->flags & GGML_TENSOR_FLAG_COMPUTE) == 0) {
+            continue;
+        }
+
         bool ok = ggml_zdnn_compute_forward(ctx, node);
         if (!ok) {
             GGML_LOG_ERROR("%s: unsupported op %s (%s)\n",
@@ -368,7 +372,8 @@ static size_t ggml_backend_zdnn_buffer_type_get_alignment(ggml_backend_buffer_ty
 }
 
 static bool ggml_backend_zdnn_buffer_type_is_host(ggml_backend_buffer_type_t buft) {
-    return true;
+    /* while it resides in host memory, additional transformation is needed */
+    return false;
 
     GGML_UNUSED(buft);
 }

data/ext/sources/ggml/src/ggml-zendnn/CMakeLists.txt

@@ -1,12 +1,19 @@
 ggml_add_backend_library(ggml-zendnn
                          ggml-zendnn.cpp)
 
-# Get ZenDNN path
 if (NOT DEFINED ZENDNN_ROOT OR ZENDNN_ROOT STREQUAL "")
     set(ZENDNN_ROOT "$ENV{ZENDNN_ROOT}")
 endif()
 
-# Check if path is still empty or OFF
+if (BUILD_SHARED_LIBS)
+    set(ZENDNN_SHARED_LIB ON)
+    set(ZENDNN_ARCHIVE_LIB OFF)
+else()
+    set(ZENDNN_SHARED_LIB OFF)
+    set(ZENDNN_ARCHIVE_LIB ON)
+endif()
+
+# Download and build ZenDNN if not provided
 if (NOT ZENDNN_ROOT OR ZENDNN_ROOT STREQUAL "" OR ZENDNN_ROOT STREQUAL "OFF")
     message(STATUS "ZENDNN_ROOT not set. Automatically downloading and building ZenDNN...")
     message(STATUS "This will take several minutes on first build...")
@@ -21,7 +28,7 @@ if (NOT ZENDNN_ROOT OR ZENDNN_ROOT STREQUAL "" OR ZENDNN_ROOT STREQUAL "OFF")
     ExternalProject_Add(
         zendnn
         GIT_REPOSITORY https://github.com/amd/ZenDNN.git
-        GIT_TAG zendnnl
+        GIT_TAG a18adf8c605fb5f5e52cefd7eda08a7b18febbaf    # ZenDNN-2026-WW08
         PREFIX      ${ZENDNN_PREFIX}
         SOURCE_DIR  ${ZENDNN_SOURCE_DIR}
         BINARY_DIR  ${ZENDNN_BUILD_DIR}
@@ -32,7 +39,9 @@ if (NOT ZENDNN_ROOT OR ZENDNN_ROOT STREQUAL "" OR ZENDNN_ROOT STREQUAL "OFF")
             -DZENDNNL_BUILD_DOXYGEN=OFF
             -DZENDNNL_BUILD_GTEST=OFF
             -DZENDNNL_BUILD_BENCHDNN=OFF
-            # Enable ALL matmul algorithm backends
+            -DZENDNNL_DEPENDS_FBGEMM=OFF
+            -DZENDNNL_LIB_BUILD_ARCHIVE=${ZENDNN_ARCHIVE_LIB}
+            -DZENDNNL_LIB_BUILD_SHARED=${ZENDNN_SHARED_LIB}
             -DZENDNNL_DEPENDS_AOCLDLP=ON
             -DZENDNNL_DEPENDS_ONEDNN=ON
             -DZENDNNL_DEPENDS_LIBXSMM=ON
@@ -45,47 +54,37 @@ if (NOT ZENDNN_ROOT OR ZENDNN_ROOT STREQUAL "" OR ZENDNN_ROOT STREQUAL "OFF")
         LOG_INSTALL ON
     )
 
-    # Add dependency so ZenDNN builds before our library
     add_dependencies(ggml-zendnn zendnn)
-
-    # Set ZENDNN_ROOT to the installation directory
     set(ZENDNN_ROOT ${ZENDNN_INSTALL_DIR})
-
     message(STATUS "ZenDNN will be built to: ${ZENDNN_ROOT}")
 else()
     message(STATUS "Using custom ZenDNN installation at: ${ZENDNN_ROOT}")
 endif()
 
-# ZenDNN headers + libs
 target_include_directories(ggml-zendnn PRIVATE
     ${ZENDNN_ROOT}/zendnnl/include
-    ${ZENDNN_ROOT}/deps/aocldlp/include
-    ${ZENDNN_ROOT}/deps/aoclutils/include
     ${ZENDNN_ROOT}/deps/json/include
-    ${ZENDNN_ROOT}/deps/libxsmm/include
+    ${ZENDNN_ROOT}/deps/aoclutils/include
+    ${ZENDNN_ROOT}/deps/aocldlp/include
     ${ZENDNN_ROOT}/deps/onednn/include
-)
+    ${ZENDNN_ROOT}/deps/libxsmm/include)
 
-target_link_directories(ggml-zendnn PRIVATE
-    ${ZENDNN_ROOT}/zendnnl/lib
-    ${ZENDNN_ROOT}/deps/aocldlp/lib
-    ${ZENDNN_ROOT}/deps/aoclutils/lib
-    ${ZENDNN_ROOT}/deps/libxsmm/lib
-    ${ZENDNN_ROOT}/deps/onednn/lib
-)
+if (ZENDNN_SHARED_LIB)
+    target_link_directories(ggml-zendnn PRIVATE ${ZENDNN_ROOT}/zendnnl/lib)
+    target_link_libraries(ggml-zendnn PRIVATE zendnnl)
+elseif (ZENDNN_ARCHIVE_LIB)
+    target_link_libraries(ggml-zendnn PRIVATE
+        ${ZENDNN_ROOT}/zendnnl/lib/libzendnnl_archive.a
+        ${ZENDNN_ROOT}/deps/aoclutils/${CMAKE_INSTALL_LIBDIR}/libaoclutils.a
+        ${ZENDNN_ROOT}/deps/aoclutils/${CMAKE_INSTALL_LIBDIR}/libau_cpuid.a
+        ${ZENDNN_ROOT}/deps/aocldlp/lib/libaocl-dlp.a
+        ${ZENDNN_ROOT}/deps/onednn/${CMAKE_INSTALL_LIBDIR}/libdnnl.a
+        ${ZENDNN_ROOT}/deps/libxsmm/lib/libxsmm.a
+        ${ZENDNN_ROOT}/deps/libxsmm/lib/libxsmmext.a
+        ${ZENDNN_ROOT}/deps/libxsmm/lib/libxsmmnoblas.a)
+endif()
 
-target_link_libraries(ggml-zendnn PRIVATE
-    zendnnl_archive    # ZenDNN main
-    aocl-dlp           # AOCL libraries
-    aoclutils
-    au_cpuid
-    dnnl               # OneDNN
-    xsmm               # libxsmm small matrix math
-    xsmmext
-    xsmmnoblas
-    m
-    pthread
-)
+target_link_libraries(ggml-zendnn PRIVATE m pthread)
 
 if (GGML_OPENMP)
     target_link_libraries(ggml-zendnn PRIVATE OpenMP::OpenMP_CXX)

data/ext/sources/ggml/src/ggml-zendnn/ggml-zendnn.cpp

@@ -2,7 +2,6 @@
 
 #include "ggml-backend-impl.h"
 #include "ggml-impl.h"
-#include "ggml-cpu.h"
 #include "zendnnl.hpp"
 
 #include <cstring>
@@ -42,13 +41,13 @@ static bool ggml_zendnn_matmul(ggml_backend_zendnn_context * ctx, int64_t m, int
                                const TA * A, int64_t lda, const TB * B, int64_t ldb, TC * C,
                                int64_t ldc) {
 
-    zendnnl::lowoha::lowoha_params params;
+    zendnnl::lowoha::matmul::matmul_params params;
     params.dtypes.src = ggml_to_zendnn_type<TB>();
     params.dtypes.wei = ggml_to_zendnn_type<TA>();
     params.dtypes.dst = ggml_to_zendnn_type<TC>();
     params.num_threads = ctx->n_threads;
 
-    zendnnl::lowoha::status_t status = zendnnl::lowoha::matmul_direct(
+    zendnnl::error_handling::status_t status = zendnnl::lowoha::matmul::matmul_direct(
         'r', false, true,   // row-major, don't transpose B, transpose A (because it's column-major)
         n,                  // M: rows of B and C
         m,                  // N: cols of A^T and C
@@ -64,7 +63,7 @@ static bool ggml_zendnn_matmul(ggml_backend_zendnn_context * ctx, int64_t m, int
         params              // params
     );
 
-    if (status != zendnnl::lowoha::status_t::success) {
+    if (status != zendnnl::error_handling::status_t::success) {
         GGML_LOG_ERROR("%s, ZenDNN matmul failed: status=%d\n", __func__, static_cast<int>(status));
         return false;
     }
@@ -122,8 +121,8 @@ static void ggml_zendnn_compute_forward_mul_mat(
 
     GGML_TENSOR_BINARY_OP_LOCALS
 
-    ggml_type         const vec_dot_type = ggml_get_type_traits_cpu(src0->type)->vec_dot_type;
-    ggml_from_float_t const from_float = ggml_get_type_traits_cpu(vec_dot_type)->from_float;
+    ggml_type         const vec_dot_type = src0->type;
+    ggml_from_float_t const from_float = ggml_get_type_traits(vec_dot_type)->from_float_ref;
 
     GGML_ASSERT(ne0 == ne01);
     GGML_ASSERT(ne1 == ne11);
@@ -211,6 +210,10 @@ static ggml_status ggml_backend_zendnn_graph_compute(ggml_backend_t backend, ggm
     for (int i = 0; i < cgraph->n_nodes; i++) {
         struct ggml_tensor * node = cgraph->nodes[i];
 
+        if ((node->flags & GGML_TENSOR_FLAG_COMPUTE) == 0) {
+            continue;
+        }
+
         switch (node->op) {
             case GGML_OP_MUL_MAT:
                 ggml_zendnn_compute_forward_mul_mat(ctx, node);