1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
|
//
// Copyright (C) 2023-2024 The ggml authors
// Copyright (C) 2024 Iwan Kawrakow
// MIT license
// SPDX-License-Identifier: MIT
//
#pragma once
#include "common.cuh"
#include "mmq.cuh"
#include <cstdint>
#define CUDA_QUANTIZE_BLOCK_SIZE 256
#define CUDA_QUANTIZE_BLOCK_SIZE_MMQ 128
static_assert(MATRIX_ROW_PADDING % CUDA_QUANTIZE_BLOCK_SIZE == 0, "Risk of out-of-bounds access.");
static_assert(MATRIX_ROW_PADDING % (4*CUDA_QUANTIZE_BLOCK_SIZE_MMQ) == 0, "Risk of out-of-bounds access.");
typedef void (*quantize_cuda_t)(
const float * x, void * vy, const int64_t kx0, const int64_t kx1, const int64_t channels, const int64_t kx0_padded,
const ggml_type type_x, cudaStream_t stream);
void quantize_row_q8_1_cuda(
const float * x, void * vy, const int64_t kx0, const int64_t kx1, const int64_t channels, const int64_t kx0_padded,
const ggml_type type_x, cudaStream_t stream);
void quantize_mmq_q8_1_cuda(
const float * x, void * vy, const int64_t kx0, const int64_t kx1, const int64_t channels, const int64_t kx0_padded,
const ggml_type type_x, cudaStream_t stream);
// For now only applicable for tensors with ne[1] = 1, ne[3] = 1, and useful if ne[2] > 1
void quantize_tensor_q8_1_cuda(const struct ggml_tensor * src, void * vy, const enum ggml_type type, cudaStream_t stream);
|
