Skip to content

Different mul_mat result between CPU and Vulkan on android device. #1287

Open
@mackong

Description

@mackong

On Android, run mul_mat with matrix A(200x200) and matrix B(200x200), and element values are all 1 of A and B. expect result a 200x200 matrix with all element values 200.

when run with CPU, the result is OK.
when run with Vulkan, the result is wrong. but the result has some special pattern, 0-63 rows is right, 64-127 rows are all 0, 128-191 rows is right, 192-199 rows are all 0.

source code of mul_mat:

#include "ggml.h"
#include "ggml-cpu.h"
#include "ggml-alloc.h"
#include "ggml-backend.h"

#if GGML_USE_VULKAN
#include "ggml-vulkan.h"
#endif

#include <cassert>
#include <cmath>
#include <cstdio>
#include <cstring>
#include <fstream>
#include <map>
#include <string>
#include <vector>

static void ggml_log_callback_default(ggml_log_level level, const char * text, void * user_data) {
    (void) level;
    (void) user_data;
    fputs(text, stderr);
    fflush(stderr);
}

// This is a simple model with two tensors a and b
struct simple_model {
    struct ggml_tensor * a;
    struct ggml_tensor * b;

    // the backend to perform the computation (CPU, CUDA, METAL)
    ggml_backend_t backend = NULL;

    // the backend buffer to storage the tensors data of a and b
    ggml_backend_buffer_t buffer;

    // the context to define the tensor information (dimensions, size, memory address)
    struct ggml_context * ctx;
};

// initialize the tensors of the model in this case two matrices 2x2
void load_model(simple_model & model, float * a, float * b, int rows_A, int cols_A, int rows_B, int cols_B) {
    ggml_log_set(ggml_log_callback_default, nullptr);

    // initialize the backend
#ifdef GGML_USE_VULKAN
    fprintf(stderr, "%s: using Vulkan backend\n", __func__);
    model.backend = ggml_backend_vk_init(0);
    if (!model.backend) {
        fprintf(stderr, "%s: ggml_backend_vk_init(0) failed\n", __func__);
    }
#endif

    // if there aren't GPU Backends fallback to CPU backend
    if (!model.backend) {
        model.backend = ggml_backend_cpu_init();
    }

    int num_tensors = 2;

    struct ggml_init_params params {
            /*.mem_size   =*/ ggml_tensor_overhead() * num_tensors,
            /*.mem_buffer =*/ NULL,
            /*.no_alloc   =*/ true,
    };

    // create context
    model.ctx = ggml_init(params);

    // create tensors
    model.a = ggml_new_tensor_2d(model.ctx, GGML_TYPE_F32, cols_A, rows_A);
    model.b = ggml_new_tensor_2d(model.ctx, GGML_TYPE_F32, cols_B, rows_B);

    // create a backend buffer (backend memory) and alloc the tensors from the context
    model.buffer = ggml_backend_alloc_ctx_tensors(model.ctx, model.backend);

    // load data from cpu memory to backend buffer
    ggml_backend_tensor_set(model.a, a, 0, ggml_nbytes(model.a));
    ggml_backend_tensor_set(model.b, b, 0, ggml_nbytes(model.b));
}

// build the compute graph to perform a matrix multiplication
struct ggml_cgraph * build_graph(const simple_model& model) {
    static size_t buf_size = ggml_tensor_overhead()*GGML_DEFAULT_GRAPH_SIZE + ggml_graph_overhead();
    static std::vector<uint8_t> buf(buf_size);

    struct ggml_init_params params0 = {
        /*.mem_size   =*/ buf_size,
        /*.mem_buffer =*/ buf.data(),
        /*.no_alloc   =*/ true, // the tensors will be allocated later by ggml_allocr_alloc_graph()
    };

    // create a temporally context to build the graph
    struct ggml_context * ctx0 = ggml_init(params0);

    struct ggml_cgraph  * gf = ggml_new_graph(ctx0);

    // result = a*b^T
    struct ggml_tensor * result = ggml_mul_mat(ctx0, model.a, model.b);

    // build operations nodes
    ggml_build_forward_expand(gf, result);

    // delete the temporally context used to build the graph
    ggml_free(ctx0);
    return gf;
}

// compute with backend
struct ggml_tensor * compute(const simple_model & model, ggml_gallocr_t allocr) {
    // reset the allocator to free all the memory allocated during the previous inference

    struct ggml_cgraph * gf = build_graph(model);

    // allocate tensors
    ggml_gallocr_alloc_graph(allocr, gf);

    int n_threads = 1; // number of threads to perform some operations with multi-threading

    if (ggml_backend_is_cpu(model.backend)) {
        ggml_backend_cpu_set_n_threads(model.backend, n_threads);
    }

    ggml_backend_graph_compute(model.backend, gf);

    // in this case, the output tensor is the last one in the graph
    return ggml_graph_node(gf, -1);
}

static void ggml_dump_tensor(uint8_t * data, ggml_type type, const int64_t * ne, const size_t * nb, const std::string & filename) {
    std::ofstream outfile(filename);
    for (int64_t i3 = 0; i3 < ne[3]; i3++) {
        for (int64_t i2 = 0; i2 < ne[2]; i2++) {
            for (int64_t i1 = 0; i1 < ne[1]; i1++) {
                for (int64_t i0 = 0; i0 < ne[0]; i0++) {
                    size_t i = i3 * nb[3] + i2 * nb[2] + i1 * nb[1] + i0 * nb[0];
                    float v;
                    if (type == GGML_TYPE_F16) {
                        v = ggml_fp16_to_fp32(*(ggml_fp16_t *) &data[i]);
                    } else if (type == GGML_TYPE_F32) {
                        v = *(float *) &data[i];
                    } else if (type == GGML_TYPE_I32) {
                        v = (float) *(int32_t *) &data[i];
                    } else if (type == GGML_TYPE_I16) {
                        v = (float) *(int16_t *) &data[i];
                    } else if (type == GGML_TYPE_I8) {
                        v = (float) *(int8_t *) &data[i];
                    } else {
                        GGML_ABORT("fatal error");
                    }
		    if (std::isnan(v)) {
			GGML_ABORT("nan got");
		    }
                    outfile << v << '\n';
                }
            }
        }
    }
}

int main(void) {
    ggml_time_init();

    // initialize data of matrices to perform matrix multiplication
    const int rows_A = 200, cols_A = 200;
    std::vector<float> A_data(40000, 1);
    float * matrix_A = A_data.data();

    const int rows_B = 200, cols_B = 200;
    std::vector<float> B_data(40000, 1);
    float * matrix_B = B_data.data();

    simple_model model;
    load_model(model, matrix_A, matrix_B, rows_A, cols_A, rows_B, cols_B);

    // calculate the temporaly memory required to compute
    ggml_gallocr_t allocr = NULL;

    {
        allocr = ggml_gallocr_new(ggml_backend_get_default_buffer_type(model.backend));

        // create the worst case graph for memory usage estimation
        struct ggml_cgraph * gf = build_graph(model);
        ggml_gallocr_reserve(allocr, gf);
        size_t mem_size = ggml_gallocr_get_buffer_size(allocr, 0);

        fprintf(stderr, "%s: compute buffer size: %.4f KB\n", __func__, mem_size/1024.0);
    }

    // perform computation
    struct ggml_tensor * result = compute(model, allocr);

    // create a array to print result
    std::vector<float> out_data(ggml_nelements(result));

    // bring the data from the backend memory
    ggml_backend_tensor_get(result, out_data.data(), 0, ggml_nbytes(result));

    ggml_dump_tensor((uint8_t *)out_data.data(), result->type, result->ne, result->nb, "result.txt");

    // release backend memory used for computation
    ggml_gallocr_free(allocr);

    // free memory
    ggml_free(model.ctx);

    // release backend memory and free backend
    ggml_backend_buffer_free(model.buffer);
    ggml_backend_free(model.backend);

    return 0;
}

log with GGML_VULKAN_DEBUG=ON

load_model: using Vulkan backend
ggml_backend_vk_init(0)
ggml_vk_init(, 0)
ggml_vk_instance_init()
ggml_vulkan: Found 1 Vulkan devices:
ggml_vk_print_gpu_info(0)
ggml_vulkan: 0 = Turnip Adreno (TM) 750 (turnip Mesa driver) | uma: 1 | fp16: 1 | warp size: 128 | shared memory: 32768 | int dot: 0 | matrix cores: none
ggml_vk_get_device(0)
Initializing new vk_device
ggml_vk_find_queue_family_index()
ggml_vk_find_queue_family_index()
ggml_vk_create_queue()
ggml_vk_load_shaders(Vulkan0)
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=0, src0_type=f32, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=0, src0_type=f32, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=0, src0_type=f32, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=1, src0_type=f32, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=1, src0_type=f32, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=1, src0_type=f32, supported=0
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=0, src0_type=f16, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=0, src0_type=f16, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=0, src0_type=f16, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=1, src0_type=f16, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=1, src0_type=f16, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=1, src0_type=f16, supported=0
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=0, src0_type=q4_0, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=0, src0_type=q4_0, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=0, src0_type=q4_0, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=1, src0_type=q4_0, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=1, src0_type=q4_0, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=1, src0_type=q4_0, supported=0
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=0, src0_type=q4_1, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=0, src0_type=q4_1, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=0, src0_type=q4_1, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=1, src0_type=q4_1, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=1, src0_type=q4_1, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=1, src0_type=q4_1, supported=0
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=0, src0_type=DEPRECATED, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=0, src0_type=DEPRECATED, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=0, src0_type=DEPRECATED, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=1, src0_type=DEPRECATED, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=1, src0_type=DEPRECATED, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=1, src0_type=DEPRECATED, supported=0
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=0, src0_type=DEPRECATED, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=0, src0_type=DEPRECATED, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=0, src0_type=DEPRECATED, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=1, src0_type=DEPRECATED, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=1, src0_type=DEPRECATED, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=1, src0_type=DEPRECATED, supported=0
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=0, src0_type=q5_0, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=0, src0_type=q5_0, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=0, src0_type=q5_0, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=1, src0_type=q5_0, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=1, src0_type=q5_0, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=1, src0_type=q5_0, supported=0
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=0, src0_type=q5_1, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=0, src0_type=q5_1, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=0, src0_type=q5_1, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=1, src0_type=q5_1, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=1, src0_type=q5_1, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=1, src0_type=q5_1, supported=0
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=0, src0_type=q8_0, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=0, src0_type=q8_0, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=0, src0_type=q8_0, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=1, src0_type=q8_0, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=1, src0_type=q8_0, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=1, src0_type=q8_0, supported=0
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=0, src0_type=q8_1, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=0, src0_type=q8_1, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=0, src0_type=q8_1, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=1, src0_type=q8_1, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=1, src0_type=q8_1, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=1, src0_type=q8_1, supported=0
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=0, src0_type=q2_K, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=0, src0_type=q2_K, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=0, src0_type=q2_K, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=1, src0_type=q2_K, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=1, src0_type=q2_K, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=1, src0_type=q2_K, supported=0
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=0, src0_type=q3_K, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=0, src0_type=q3_K, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=0, src0_type=q3_K, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=1, src0_type=q3_K, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=1, src0_type=q3_K, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=1, src0_type=q3_K, supported=0
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=0, src0_type=q4_K, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=0, src0_type=q4_K, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=0, src0_type=q4_K, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=1, src0_type=q4_K, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=1, src0_type=q4_K, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=1, src0_type=q4_K, supported=0
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=0, src0_type=q5_K, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=0, src0_type=q5_K, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=0, src0_type=q5_K, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=1, src0_type=q5_K, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=1, src0_type=q5_K, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=1, src0_type=q5_K, supported=0
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=0, src0_type=q6_K, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=0, src0_type=q6_K, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=0, src0_type=q6_K, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=1, src0_type=q6_K, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=1, src0_type=q6_K, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=1, src0_type=q6_K, supported=0
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=0, src0_type=q8_K, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=0, src0_type=q8_K, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=0, src0_type=q8_K, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=1, src0_type=q8_K, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=1, src0_type=q8_K, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=1, src0_type=q8_K, supported=0
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=0, src0_type=iq2_xxs, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=0, src0_type=iq2_xxs, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=0, src0_type=iq2_xxs, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=1, src0_type=iq2_xxs, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=1, src0_type=iq2_xxs, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=1, src0_type=iq2_xxs, supported=0
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=0, src0_type=iq2_xs, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=0, src0_type=iq2_xs, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=0, src0_type=iq2_xs, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=1, src0_type=iq2_xs, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=1, src0_type=iq2_xs, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=1, src0_type=iq2_xs, supported=0
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=0, src0_type=iq3_xxs, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=0, src0_type=iq3_xxs, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=0, src0_type=iq3_xxs, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=1, src0_type=iq3_xxs, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=1, src0_type=iq3_xxs, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=1, src0_type=iq3_xxs, supported=0
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=0, src0_type=iq1_s, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=0, src0_type=iq1_s, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=0, src0_type=iq1_s, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=1, src0_type=iq1_s, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=1, src0_type=iq1_s, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=1, src0_type=iq1_s, supported=0
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=0, src0_type=iq4_nl, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=0, src0_type=iq4_nl, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=0, src0_type=iq4_nl, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=1, src0_type=iq4_nl, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=1, src0_type=iq4_nl, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=1, src0_type=iq4_nl, supported=0
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=0, src0_type=iq3_s, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=0, src0_type=iq3_s, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=0, src0_type=iq3_s, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=1, src0_type=iq3_s, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=1, src0_type=iq3_s, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=1, src0_type=iq3_s, supported=0
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=0, src0_type=iq2_s, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=0, src0_type=iq2_s, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=0, src0_type=iq2_s, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=1, src0_type=iq2_s, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=1, src0_type=iq2_s, supported=0
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=0, src0_type=iq4_xs, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=0, src0_type=iq4_xs, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=0, src0_type=iq4_xs, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=1, src0_type=iq4_xs, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=1, src0_type=iq4_xs, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=1, src0_type=iq4_xs, supported=0
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=0, src0_type=i8, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=0, src0_type=i8, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=0, src0_type=i8, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=1, src0_type=i8, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=1, src0_type=i8, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=1, src0_type=i8, supported=0
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=0, src0_type=i16, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=0, src0_type=i16, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=0, src0_type=i16, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=1, src0_type=i16, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=1, src0_type=i16, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=1, src0_type=i16, supported=0
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=0, src0_type=i32, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=0, src0_type=i32, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=0, src0_type=i32, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=1, src0_type=i32, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=1, src0_type=i32, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=1, src0_type=i32, supported=0
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=0, src0_type=i64, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=0, src0_type=i64, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=0, src0_type=i64, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=1, src0_type=i64, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=1, src0_type=i64, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=1, src0_type=i64, supported=0
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=0, src0_type=f64, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=0, src0_type=f64, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=0, src0_type=f64, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=1, src0_type=f64, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=1, src0_type=f64, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=1, src0_type=f64, supported=0
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=0, src0_type=iq1_m, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=0, src0_type=iq1_m, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=0, src0_type=iq1_m, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=1, src0_type=iq1_m, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=1, src0_type=iq1_m, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=1, src0_type=iq1_m, supported=0
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=0, src0_type=bf16, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=0, src0_type=bf16, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=0, src0_type=bf16, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=1, src0_type=bf16, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=1, src0_type=bf16, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=1, src0_type=bf16, supported=0
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=0, src0_type=TYPE_Q4_0_4_4 REMOVED, use Q4_0 with runtime repacking, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=0, src0_type=TYPE_Q4_0_4_4 REMOVED, use Q4_0 with runtime repacking, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=0, src0_type=TYPE_Q4_0_4_4 REMOVED, use Q4_0 with runtime repacking, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=1, src0_type=TYPE_Q4_0_4_4 REMOVED, use Q4_0 with runtime repacking, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=1, src0_type=TYPE_Q4_0_4_4 REMOVED, use Q4_0 with runtime repacking, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=1, src0_type=TYPE_Q4_0_4_4 REMOVED, use Q4_0 with runtime repacking, supported=0
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=0, src0_type=TYPE_Q4_0_4_8 REMOVED, use Q4_0 with runtime repacking, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=0, src0_type=TYPE_Q4_0_4_8 REMOVED, use Q4_0 with runtime repacking, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=0, src0_type=TYPE_Q4_0_4_8 REMOVED, use Q4_0 with runtime repacking, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=1, src0_type=TYPE_Q4_0_4_8 REMOVED, use Q4_0 with runtime repacking, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=1, src0_type=TYPE_Q4_0_4_8 REMOVED, use Q4_0 with runtime repacking, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=1, src0_type=TYPE_Q4_0_4_8 REMOVED, use Q4_0 with runtime repacking, supported=0
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=0, src0_type=TYPE_Q4_0_8_8 REMOVED, use Q4_0 with runtime repacking, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=0, src0_type=TYPE_Q4_0_8_8 REMOVED, use Q4_0 with runtime repacking, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=0, src0_type=TYPE_Q4_0_8_8 REMOVED, use Q4_0 with runtime repacking, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=1, src0_type=TYPE_Q4_0_8_8 REMOVED, use Q4_0 with runtime repacking, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=1, src0_type=TYPE_Q4_0_8_8 REMOVED, use Q4_0 with runtime repacking, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=1, src0_type=TYPE_Q4_0_8_8 REMOVED, use Q4_0 with runtime repacking, supported=0
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=0, src0_type=tq1_0, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=0, src0_type=tq1_0, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=0, src0_type=tq1_0, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=1, src0_type=tq1_0, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=1, src0_type=tq1_0, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=1, src0_type=tq1_0, supported=0
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=0, src0_type=tq2_0, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=0, src0_type=tq2_0, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=0, src0_type=tq2_0, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=1, src0_type=tq2_0, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=1, src0_type=tq2_0, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=1, src0_type=tq2_0, supported=0
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=0, src0_type=TYPE_IQ4_NL_4_4 REMOVED, use IQ4_NL with runtime repacking, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=0, src0_type=TYPE_IQ4_NL_4_4 REMOVED, use IQ4_NL with runtime repacking, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=0, src0_type=TYPE_IQ4_NL_4_4 REMOVED, use IQ4_NL with runtime repacking, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=1, src0_type=TYPE_IQ4_NL_4_4 REMOVED, use IQ4_NL with runtime repacking, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=1, src0_type=TYPE_IQ4_NL_4_4 REMOVED, use IQ4_NL with runtime repacking, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=1, src0_type=TYPE_IQ4_NL_4_4 REMOVED, use IQ4_NL with runtime repacking, supported=0
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=0, src0_type=TYPE_IQ4_NL_4_8 REMOVED, use IQ4_NL with runtime repacking, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=0, src0_type=TYPE_IQ4_NL_4_8 REMOVED, use IQ4_NL with runtime repacking, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=0, src0_type=TYPE_IQ4_NL_4_8 REMOVED, use IQ4_NL with runtime repacking, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=1, src0_type=TYPE_IQ4_NL_4_8 REMOVED, use IQ4_NL with runtime repacking, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=1, src0_type=TYPE_IQ4_NL_4_8 REMOVED, use IQ4_NL with runtime repacking, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=1, src0_type=TYPE_IQ4_NL_4_8 REMOVED, use IQ4_NL with runtime repacking, supported=0
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=0, src0_type=TYPE_IQ4_NL_8_8 REMOVED, use IQ4_NL with runtime repacking, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=0, src0_type=TYPE_IQ4_NL_8_8 REMOVED, use IQ4_NL with runtime repacking, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=0, src0_type=TYPE_IQ4_NL_8_8 REMOVED, use IQ4_NL with runtime repacking, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=1, src0_type=TYPE_IQ4_NL_8_8 REMOVED, use IQ4_NL with runtime repacking, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=1, src0_type=TYPE_IQ4_NL_8_8 REMOVED, use IQ4_NL with runtime repacking, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=1, src0_type=TYPE_IQ4_NL_8_8 REMOVED, use IQ4_NL with runtime repacking, supported=0
ggml_backend_vk_buffer_type(0)
ggml_vk_get_device(0)
ggml_vulkan memory: ggml_backend_vk_buffer_type_alloc_buffer(320000)
ggml_vk_create_buffer(Vulkan0, 320000, { DeviceLocal }, { HostVisible | HostCoherent })
ggml_vulkan memory: Vulkan0: +312.50 KiB device at 0xb400007cd64716f0. Total device: 312.50 KiB, total host: 0 B
ggml_backend_vk_buffer_init_tensor(0xb400007cd6470c40 (0xb400007e264516a0), 0xb400007e66441460)
ggml_backend_vk_buffer_init_tensor(0xb400007cd6470c40 (0xb400007e264516a0), 0xb400007e664415d0)
ggml_backend_vk_buffer_set_tensor(0xb400007cd6470c40, 0xb400007e66441460, 0xb400007ec8a3b000, 0, 160000)
ggml_vk_buffer_write(160000)
ggml_vk_buffer_write_2d(160000, 1)
ggml_vk_create_temporary_context(0xb400007cd6470478)
ggml_vk_ctx_begin(Vulkan0)
ggml_vk_create_cmd_buffer()
ggml_vk_buffer_write_2d_async(160000, 1)
STAGING
ggml_vulkan memory: ggml_vk_ensure_sync_staging_buffer(160000)
ggml_vk_create_buffer(Vulkan0, 160000, { HostVisible | HostCoherent | HostCached }, { HostVisible | HostCoherent })
ggml_vulkan memory: Vulkan0: +156.25 KiB host at 0xb400007cd6470e80. Total device: 312.50 KiB, total host: 156.25 KiB
ggml_vk_sync_buffers()
ggml_vk_ctx_end(0xb400007cd6470478, 1)
ggml_vk_submit(0xb400007cd6470478, 0xb400007dd643fdd0)
ggml_vk_queue_command_pools_cleanup()
ggml_backend_vk_buffer_set_tensor(0xb400007cd6470c40, 0xb400007e664415d0, 0xb400007ec544d000, 0, 160000)
ggml_vk_buffer_write(160000)
ggml_vk_buffer_write_2d(160000, 1)
ggml_vk_create_temporary_context(0xb400007cd6470478)
ggml_vk_ctx_begin(Vulkan0)
ggml_vk_create_cmd_buffer()
ggml_vk_buffer_write_2d_async(160000, 1)
STAGING
ggml_vk_sync_buffers()
ggml_vk_ctx_end(0xb400007cd6470478, 1)
ggml_vk_submit(0xb400007cd6470478, 0xb400007dd643fdd0)
ggml_vk_queue_command_pools_cleanup()
ggml_backend_vk_buffer_type(0)
ggml_vk_get_device(0)
ggml_vulkan memory: ggml_backend_vk_buffer_type_alloc_buffer(160000)
ggml_vk_create_buffer(Vulkan0, 160000, { DeviceLocal }, { HostVisible | HostCoherent })
ggml_vulkan memory: Vulkan0: +156.25 KiB device at 0xb400007cd6470460. Total device: 468.75 KiB, total host: 156.25 KiB
main: compute buffer size: 156.2500 KB
ggml_backend_vk_buffer_init_tensor(0xb400007cd6471780 (0xb400007e26451e20), 0xb400007c0f3f6020)
ggml_backend_vk_graph_compute(1 nodes)
ggml_vk_build_graph(0xb400007c0f3f6020, MUL_MAT)
ggml_vk_mul_mat(0xb400007e66441460, 0xb400007e664415d0, 0xb400007c0f3f6020)
ggml_vk_mul_mat_q_f16((0xb400007e66441460, name=leaf_0, type=0, ne0=200, ne1=200, ne2=1, ne3=1, nb0=4, nb1=800, nb2=160000, nb3=160000), (0xb400007e664415d0, name=leaf_1, type=0, ne0=200, ne1=200, ne2=1, ne3=1, nb0=4, nb1=800, nb2=160000, nb3=160000), (0xb400007c0f3f6020, name=node_0, type=0, ne0=200, ne1=200, ne2=1, ne3=1, nb0=4, nb1=800, nb2=160000, nb3=160000), dryrun)
ggml_vk_get_mul_mat_mat_pipeline(f32, f32, 0)
ggml_vk_guess_matmul_pipeline_align(200, 200, f32, f32)
ggml_vk_guess_matmul_pipeline(200, 200, 1, f32, f32)
ggml_vk_align_size(200, 128)
ggml_vk_guess_matmul_pipeline(200, 200, 0, f32, f32)
ggml_vk_guess_split_k(200, 200, 200)
ggml_vk_get_to_fp16()
ggml_vk_get_to_fp16()
ggml_pipeline_request_descriptor_sets(matmul_f32_f32_l, 1)
ggml_vk_load_shaders(Vulkan0)
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=0, src0_type=f32, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=0, src0_type=f32, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=0, src0_type=f32, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=1, src0_type=f32, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=1, src0_type=f32, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=1, src0_type=f32, supported=0
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=0, src0_type=f16, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=0, src0_type=f16, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=0, src0_type=f16, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=1, src0_type=f16, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=1, src0_type=f16, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=1, src0_type=f16, supported=0
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=0, src0_type=q4_0, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=0, src0_type=q4_0, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=0, src0_type=q4_0, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=1, src0_type=q4_0, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=1, src0_type=q4_0, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=1, src0_type=q4_0, supported=0
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=0, src0_type=q4_1, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=0, src0_type=q4_1, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=0, src0_type=q4_1, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=1, src0_type=q4_1, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=1, src0_type=q4_1, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=1, src0_type=q4_1, supported=0
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=0, src0_type=DEPRECATED, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=0, src0_type=DEPRECATED, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=0, src0_type=DEPRECATED, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=1, src0_type=DEPRECATED, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=1, src0_type=DEPRECATED, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=1, src0_type=DEPRECATED, supported=0
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=0, src0_type=DEPRECATED, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=0, src0_type=DEPRECATED, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=0, src0_type=DEPRECATED, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=1, src0_type=DEPRECATED, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=1, src0_type=DEPRECATED, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=1, src0_type=DEPRECATED, supported=0
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=0, src0_type=q5_0, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=0, src0_type=q5_0, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=0, src0_type=q5_0, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=1, src0_type=q5_0, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=1, src0_type=q5_0, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=1, src0_type=q5_0, supported=0
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=0, src0_type=q5_1, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=0, src0_type=q5_1, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=0, src0_type=q5_1, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=1, src0_type=q5_1, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=1, src0_type=q5_1, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=1, src0_type=q5_1, supported=0
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=0, src0_type=q8_0, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=0, src0_type=q8_0, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=0, src0_type=q8_0, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=1, src0_type=q8_0, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=1, src0_type=q8_0, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=1, src0_type=q8_0, supported=0
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=0, src0_type=q8_1, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=0, src0_type=q8_1, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=0, src0_type=q8_1, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=1, src0_type=q8_1, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=1, src0_type=q8_1, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=1, src0_type=q8_1, supported=0
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=0, src0_type=q2_K, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=0, src0_type=q2_K, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=0, src0_type=q2_K, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=1, src0_type=q2_K, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=1, src0_type=q2_K, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=1, src0_type=q2_K, supported=0
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=0, src0_type=q3_K, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=0, src0_type=q3_K, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=0, src0_type=q3_K, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=1, src0_type=q3_K, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=1, src0_type=q3_K, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=1, src0_type=q3_K, supported=0
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=0, src0_type=q4_K, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=0, src0_type=q4_K, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=0, src0_type=q4_K, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=1, src0_type=q4_K, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=1, src0_type=q4_K, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=1, src0_type=q4_K, supported=0
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=0, src0_type=q5_K, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=0, src0_type=q5_K, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=0, src0_type=q5_K, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=1, src0_type=q5_K, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=1, src0_type=q5_K, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=1, src0_type=q5_K, supported=0
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=0, src0_type=q6_K, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=0, src0_type=q6_K, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=0, src0_type=q6_K, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=1, src0_type=q6_K, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=1, src0_type=q6_K, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=1, src0_type=q6_K, supported=0
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=0, src0_type=q8_K, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=0, src0_type=q8_K, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=0, src0_type=q8_K, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=1, src0_type=q8_K, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=1, src0_type=q8_K, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=1, src0_type=q8_K, supported=0
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=0, src0_type=iq2_xxs, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=0, src0_type=iq2_xxs, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=0, src0_type=iq2_xxs, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=1, src0_type=iq2_xxs, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=1, src0_type=iq2_xxs, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=1, src0_type=iq2_xxs, supported=0
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=0, src0_type=iq2_xs, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=0, src0_type=iq2_xs, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=0, src0_type=iq2_xs, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=1, src0_type=iq2_xs, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=1, src0_type=iq2_xs, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=1, src0_type=iq2_xs, supported=0
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=0, src0_type=iq3_xxs, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=0, src0_type=iq3_xxs, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=0, src0_type=iq3_xxs, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=1, src0_type=iq3_xxs, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=1, src0_type=iq3_xxs, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=1, src0_type=iq3_xxs, supported=0
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=0, src0_type=iq1_s, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=0, src0_type=iq1_s, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=0, src0_type=iq1_s, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=1, src0_type=iq1_s, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=1, src0_type=iq1_s, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=1, src0_type=iq1_s, supported=0
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=0, src0_type=iq4_nl, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=0, src0_type=iq4_nl, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=0, src0_type=iq4_nl, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=1, src0_type=iq4_nl, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=1, src0_type=iq4_nl, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=1, src0_type=iq4_nl, supported=0
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=0, src0_type=iq3_s, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=0, src0_type=iq3_s, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=0, src0_type=iq3_s, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=1, src0_type=iq3_s, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=1, src0_type=iq3_s, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=1, src0_type=iq3_s, supported=0
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=0, src0_type=iq2_s, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=0, src0_type=iq2_s, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=0, src0_type=iq2_s, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=1, src0_type=iq2_s, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=1, src0_type=iq2_s, supported=0
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=0, src0_type=iq4_xs, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=0, src0_type=iq4_xs, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=0, src0_type=iq4_xs, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=1, src0_type=iq4_xs, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=1, src0_type=iq4_xs, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=1, src0_type=iq4_xs, supported=0
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=0, src0_type=i8, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=0, src0_type=i8, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=0, src0_type=i8, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=1, src0_type=i8, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=1, src0_type=i8, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=1, src0_type=i8, supported=0
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=0, src0_type=i16, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=0, src0_type=i16, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=0, src0_type=i16, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=1, src0_type=i16, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=1, src0_type=i16, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=1, src0_type=i16, supported=0
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=0, src0_type=i32, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=0, src0_type=i32, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=0, src0_type=i32, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=1, src0_type=i32, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=1, src0_type=i32, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=1, src0_type=i32, supported=0
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=0, src0_type=i64, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=0, src0_type=i64, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=0, src0_type=i64, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=1, src0_type=i64, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=1, src0_type=i64, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=1, src0_type=i64, supported=0
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=0, src0_type=f64, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=0, src0_type=f64, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=0, src0_type=f64, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=1, src0_type=f64, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=1, src0_type=f64, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=1, src0_type=f64, supported=0
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=0, src0_type=iq1_m, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=0, src0_type=iq1_m, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=0, src0_type=iq1_m, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=1, src0_type=iq1_m, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=1, src0_type=iq1_m, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=1, src0_type=iq1_m, supported=0
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=0, src0_type=bf16, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=0, src0_type=bf16, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=0, src0_type=bf16, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=1, src0_type=bf16, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=1, src0_type=bf16, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=1, src0_type=bf16, supported=0
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=0, src0_type=TYPE_Q4_0_4_4 REMOVED, use Q4_0 with runtime repacking, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=0, src0_type=TYPE_Q4_0_4_4 REMOVED, use Q4_0 with runtime repacking, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=0, src0_type=TYPE_Q4_0_4_4 REMOVED, use Q4_0 with runtime repacking, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=1, src0_type=TYPE_Q4_0_4_4 REMOVED, use Q4_0 with runtime repacking, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=1, src0_type=TYPE_Q4_0_4_4 REMOVED, use Q4_0 with runtime repacking, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=1, src0_type=TYPE_Q4_0_4_4 REMOVED, use Q4_0 with runtime repacking, supported=0
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=0, src0_type=TYPE_Q4_0_4_8 REMOVED, use Q4_0 with runtime repacking, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=0, src0_type=TYPE_Q4_0_4_8 REMOVED, use Q4_0 with runtime repacking, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=0, src0_type=TYPE_Q4_0_4_8 REMOVED, use Q4_0 with runtime repacking, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=1, src0_type=TYPE_Q4_0_4_8 REMOVED, use Q4_0 with runtime repacking, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=1, src0_type=TYPE_Q4_0_4_8 REMOVED, use Q4_0 with runtime repacking, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=1, src0_type=TYPE_Q4_0_4_8 REMOVED, use Q4_0 with runtime repacking, supported=0
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=0, src0_type=TYPE_Q4_0_8_8 REMOVED, use Q4_0 with runtime repacking, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=0, src0_type=TYPE_Q4_0_8_8 REMOVED, use Q4_0 with runtime repacking, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=0, src0_type=TYPE_Q4_0_8_8 REMOVED, use Q4_0 with runtime repacking, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=1, src0_type=TYPE_Q4_0_8_8 REMOVED, use Q4_0 with runtime repacking, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=1, src0_type=TYPE_Q4_0_8_8 REMOVED, use Q4_0 with runtime repacking, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=1, src0_type=TYPE_Q4_0_8_8 REMOVED, use Q4_0 with runtime repacking, supported=0
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=0, src0_type=tq1_0, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=0, src0_type=tq1_0, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=0, src0_type=tq1_0, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=1, src0_type=tq1_0, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=1, src0_type=tq1_0, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=1, src0_type=tq1_0, supported=0
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=0, src0_type=tq2_0, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=0, src0_type=tq2_0, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=0, src0_type=tq2_0, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=1, src0_type=tq2_0, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=1, src0_type=tq2_0, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=1, src0_type=tq2_0, supported=0
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=0, src0_type=TYPE_IQ4_NL_4_4 REMOVED, use IQ4_NL with runtime repacking, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=0, src0_type=TYPE_IQ4_NL_4_4 REMOVED, use IQ4_NL with runtime repacking, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=0, src0_type=TYPE_IQ4_NL_4_4 REMOVED, use IQ4_NL with runtime repacking, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=1, src0_type=TYPE_IQ4_NL_4_4 REMOVED, use IQ4_NL with runtime repacking, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=1, src0_type=TYPE_IQ4_NL_4_4 REMOVED, use IQ4_NL with runtime repacking, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=1, src0_type=TYPE_IQ4_NL_4_4 REMOVED, use IQ4_NL with runtime repacking, supported=0
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=0, src0_type=TYPE_IQ4_NL_4_8 REMOVED, use IQ4_NL with runtime repacking, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=0, src0_type=TYPE_IQ4_NL_4_8 REMOVED, use IQ4_NL with runtime repacking, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=0, src0_type=TYPE_IQ4_NL_4_8 REMOVED, use IQ4_NL with runtime repacking, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=1, src0_type=TYPE_IQ4_NL_4_8 REMOVED, use IQ4_NL with runtime repacking, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=1, src0_type=TYPE_IQ4_NL_4_8 REMOVED, use IQ4_NL with runtime repacking, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=1, src0_type=TYPE_IQ4_NL_4_8 REMOVED, use IQ4_NL with runtime repacking, supported=0
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=0, src0_type=TYPE_IQ4_NL_8_8 REMOVED, use IQ4_NL with runtime repacking, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=0, src0_type=TYPE_IQ4_NL_8_8 REMOVED, use IQ4_NL with runtime repacking, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=0, src0_type=TYPE_IQ4_NL_8_8 REMOVED, use IQ4_NL with runtime repacking, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,32,32), mul_mat_id=1, src0_type=TYPE_IQ4_NL_8_8 REMOVED, use IQ4_NL with runtime repacking, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,64,64), mul_mat_id=1, src0_type=TYPE_IQ4_NL_8_8 REMOVED, use IQ4_NL with runtime repacking, supported=1
ggml_vk_matmul_shmem_support(warptile=(128,128,128), mul_mat_id=1, src0_type=TYPE_IQ4_NL_8_8 REMOVED, use IQ4_NL with runtime repacking, supported=0
ggml_vk_create_pipeline(Vulkan0, matmul_f32_f32_l, main, 3, (128,128,1), specialization_constants, 0, 0, 0)
ggml_vulkan memory: ggml_vk_preallocate_buffers(x_size: 0)
ggml_vk_create_buffer(Vulkan0, 0, { DeviceLocal }, { HostVisible | HostCoherent })
ggml_vulkan memory: ggml_vk_preallocate_buffers(y_size: 0)
ggml_vk_create_buffer(Vulkan0, 0, { DeviceLocal }, { HostVisible | HostCoherent })
ggml_vulkan memory: ggml_vk_preallocate_buffers(split_k_size: 0)
ggml_vk_create_buffer(Vulkan0, 0, { DeviceLocal }, { HostVisible | HostCoherent })
ggml_vk_build_graph(0xb400007c0f3f6020, MUL_MAT)
ggml_vk_create_context(0xb400007cd646fae8)
ggml_vk_ctx_begin(Vulkan0)
ggml_vk_create_cmd_buffer()
ggml_vk_mul_mat(0xb400007e66441460, 0xb400007e664415d0, 0xb400007c0f3f6020)
ggml_vk_mul_mat_q_f16((0xb400007e66441460, name=leaf_0, type=0, ne0=200, ne1=200, ne2=1, ne3=1, nb0=4, nb1=800, nb2=160000, nb3=160000), (0xb400007e664415d0, name=leaf_1, type=0, ne0=200, ne1=200, ne2=1, ne3=1, nb0=4, nb1=800, nb2=160000, nb3=160000), (0xb400007c0f3f6020, name=node_0, type=0, ne0=200, ne1=200, ne2=1, ne3=1, nb0=4, nb1=800, nb2=160000, nb3=160000), )
ggml_vk_get_mul_mat_mat_pipeline(f32, f32, 0)
ggml_vk_guess_matmul_pipeline_align(200, 200, f32, f32)
ggml_vk_guess_matmul_pipeline(200, 200, 1, f32, f32)
ggml_vk_align_size(200, 128)
ggml_vk_guess_matmul_pipeline(200, 200, 0, f32, f32)
ggml_vk_guess_split_k(200, 200, 200)
ggml_vk_get_to_fp16()
ggml_vk_get_to_fp16()
ggml_vk_matmul(a: (0xb400007cd64716f0, 0, 160000), b: (0xb400007cd64716f0, 160000, 160000), d: (0xb400007cd6470460, 0, 160000), split_k: (0x0, 0, 160000), m: 200, n: 200, k: 200, stride_a: 200, stride_b: 200, stride_d: 200, batch_stride_a: 40000, batch_stride_b: 40000, batch_stride_d: 40000, split_k: 1, batch: 1, ne02: 1, ne12: 1, broadcast2: 1, broadcast3: 1, padded_n: 200)
ggml_vk_sync_buffers()
ggml_vk_dispatch_pipeline(matmul_f32_f32_l, {(0xb400007cd64716f0, 0, 160000), (0xb400007cd64716f0, 160000, 160000), (0xb400007cd6470460, 0, 160000), }, (2,2,1))
ggml_vk_ctx_end(0xb400007cd646fae8, 1)
ggml_vk_compute_forward(0xb400007c0f3f6020, name=node_0, op=MUL_MAT, type=0, ne0=200, ne1=200, ne2=1, ne3=1, nb0=4, nb1=800, nb2=160000, nb3=160000, view_src=0x0, view_offs=0)
ggml_vk_check_results_0(node_0)
ggml_vk_buffer_read(0xb400007cd64716f0, 0, 160000)
ggml_vk_create_temporary_context(0xb400007cd6471678)
ggml_vk_ctx_begin(Vulkan0)
ggml_vk_create_cmd_buffer()
ggml_vk_buffer_read_2d_async(offset=0, width=160000, height=1)
STAGING
ggml_vk_sync_buffers()
ggml_vk_ctx_end(0xb400007cd6471678, 1)
ggml_vk_submit(0xb400007cd6471678, 0xb400007dd643fdd0)
ggml_vk_queue_command_pools_cleanup()
ggml_vk_buffer_read(0xb400007cd64716f0, 160000, 160000)
ggml_vk_create_temporary_context(0xb400007cd6471678)
ggml_vk_ctx_begin(Vulkan0)
ggml_vk_create_cmd_buffer()
ggml_vk_buffer_read_2d_async(offset=160000, width=160000, height=1)
STAGING
ggml_vk_sync_buffers()
ggml_vk_ctx_end(0xb400007cd6471678, 1)
ggml_vk_submit(0xb400007cd6471678, 0xb400007dd643fdd0)
ggml_vk_queue_command_pools_cleanup()
END ggml_vk_check_results_0(node_0)
ggml_vk_submit(0xb400007cd646fae8, 0xb400007dd6441390)
ggml_vk_check_results_1(node_0)
ggml_vk_buffer_read(0xb400007cd6470460, 0, 160000)
ggml_vk_create_temporary_context(0xb400007cd6471678)
ggml_vk_ctx_begin(Vulkan0)
ggml_vk_create_cmd_buffer()
ggml_vk_buffer_read_2d_async(offset=0, width=160000, height=1)
STAGING
ggml_vk_sync_buffers()
ggml_vk_ctx_end(0xb400007cd6471678, 1)
ggml_vk_submit(0xb400007cd6471678, 0xb400007dd643fdd0)
ggml_vk_queue_command_pools_cleanup()
1 node_0 op=MUL_MAT avg_err=0.36
END ggml_vk_check_results_1(node_0)
ggml_vk_graph_cleanup()
ggml_vk_command_pool_cleanup()
ggml_vk_command_pool_cleanup()
ggml_backend_vk_buffer_get_tensor(0xb400007cd6471780, 0xb400007c0f3f6020, 0xb400007ec3213f00, 0, 160000)
ggml_vk_buffer_read(0xb400007cd6470460, 0, 160000)
ggml_vk_create_temporary_context(0xb400007cd646fae8)
ggml_vk_ctx_begin(Vulkan0)
ggml_vk_create_cmd_buffer()
ggml_vk_buffer_read_2d_async(offset=0, width=160000, height=1)
STAGING
ggml_vk_sync_buffers()
ggml_vk_ctx_end(0xb400007cd646fae8, 1)
ggml_vk_submit(0xb400007cd646fae8, 0xb400007dd643fdd0)
ggml_vk_queue_command_pools_cleanup()
ggml_vulkan memory: ggml_backend_vk_buffer_free_buffer()
ggml_vulkan memory: Vulkan0: -156.25 KiB device at 0xb400007cd6470460. Total device: 312.50 KiB, total host: 156.25 KiB
~vk_buffer_struct(0xb400007cd6470460, 160000)
ggml_vulkan memory: ggml_backend_vk_buffer_free_buffer()
ggml_vulkan memory: Vulkan0: -312.50 KiB device at 0xb400007cd64716f0. Total device: 0 B, total host: 156.25 KiB
~vk_buffer_struct(0xb400007cd64716f0, 320000)
ggml_backend_vk_free(Vulkan0)
ggml_vk_cleanup(Vulkan0)
ggml_vk_graph_cleanup()
ggml_vk_command_pool_cleanup()
ggml_vk_command_pool_cleanup()

result matrix data:

result.txt

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions