char-fork/llama.cpp

git clone https://git.t4t.associates/char-fork/llama.cpp

Kevin Pougetggml-virtgpu: improve the reliability of the code (#19846)ffaafde16

char/aliceai-t5-moe
brancheschar/aliceai-t5-moeall refs →
4.5 KiB73 linesraw
1#pragma once
2
3/* device */
4uint32_t backend_device_get_device_count(apir_encoder * enc, apir_decoder * dec, virgl_apir_context * ctx);
5uint32_t backend_device_get_count(apir_encoder * enc, apir_decoder * dec, virgl_apir_context * ctx);
6uint32_t backend_device_get_name(apir_encoder * enc, apir_decoder * dec, virgl_apir_context * ctx);
7uint32_t backend_device_get_description(apir_encoder * enc, apir_decoder * dec, virgl_apir_context * ctx);
8uint32_t backend_device_get_type(apir_encoder * enc, apir_decoder * dec, virgl_apir_context * ctx);
9uint32_t backend_device_get_memory(apir_encoder * enc, apir_decoder * dec, virgl_apir_context * ctx);
10uint32_t backend_device_supports_op(apir_encoder * enc, apir_decoder * dec, virgl_apir_context * ctx);
11uint32_t backend_device_get_buffer_type(apir_encoder * enc, apir_decoder * dec, virgl_apir_context * ctx);
12uint32_t backend_device_get_props(apir_encoder * enc, apir_decoder * dec, virgl_apir_context * ctx);
13uint32_t backend_device_buffer_from_ptr(apir_encoder * enc, apir_decoder * dec, virgl_apir_context * ctx);
14
15/* buffer-type */
16uint32_t backend_buffer_type_get_name(apir_encoder * enc, apir_decoder * dec, virgl_apir_context * ctx);
17uint32_t backend_buffer_type_get_alignment(apir_encoder * enc, apir_decoder * dec, virgl_apir_context * ctx);
18uint32_t backend_buffer_type_get_max_size(apir_encoder * enc, apir_decoder * dec, virgl_apir_context * ctx);
19/* APIR_COMMAND_TYPE_BUFFER_TYPE_IS_HOST is deprecated. Keeping the handler for backward compatibility. */
20uint32_t backend_buffer_type_is_host(apir_encoder * enc, apir_decoder * dec, virgl_apir_context * ctx);
21uint32_t backend_buffer_type_alloc_buffer(apir_encoder * enc, apir_decoder * dec, virgl_apir_context * ctx);
22uint32_t backend_buffer_type_get_alloc_size(apir_encoder * enc, apir_decoder * dec, virgl_apir_context * ctx);
23
24/* buffer */
25uint32_t backend_buffer_get_base(apir_encoder * enc, apir_decoder * dec, virgl_apir_context * ctx);
26uint32_t backend_buffer_set_tensor(apir_encoder * enc, apir_decoder * dec, virgl_apir_context * ctx);
27uint32_t backend_buffer_get_tensor(apir_encoder * enc, apir_decoder * dec, virgl_apir_context * ctx);
28uint32_t backend_buffer_cpy_tensor(apir_encoder * enc, apir_decoder * dec, virgl_apir_context * ctx);
29uint32_t backend_buffer_clear(apir_encoder * enc, apir_decoder * dec, virgl_apir_context * ctx);
30uint32_t backend_buffer_free_buffer(apir_encoder * enc, apir_decoder * dec, virgl_apir_context * ctx);
31
32/* backend */
33uint32_t backend_backend_graph_compute(apir_encoder * enc, apir_decoder * dec, virgl_apir_context * ctx);
34
35extern "C" {
36static const backend_dispatch_t apir_backend_dispatch_table[APIR_BACKEND_DISPATCH_TABLE_COUNT] = {
37
38    /* device */
39
40    /* APIR_COMMAND_TYPE_DEVICE_GET_DEVICE_COUNT  = */ backend_device_get_device_count,
41    /* APIR_COMMAND_TYPE_DEVICE_GET_COUNT  = */ backend_device_get_count,
42    /* APIR_COMMAND_TYPE_DEVICE_GET_NAME  = */ backend_device_get_name,
43    /* APIR_COMMAND_TYPE_DEVICE_GET_DESCRIPTION  = */ backend_device_get_description,
44    /* APIR_COMMAND_TYPE_DEVICE_GET_TYPE  = */ backend_device_get_type,
45    /* APIR_COMMAND_TYPE_DEVICE_GET_MEMORY  = */ backend_device_get_memory,
46    /* APIR_COMMAND_TYPE_DEVICE_SUPPORTS_OP  = */ backend_device_supports_op,
47    /* APIR_COMMAND_TYPE_DEVICE_GET_BUFFER_TYPE  = */ backend_device_get_buffer_type,
48    /* APIR_COMMAND_TYPE_DEVICE_GET_PROPS  = */ backend_device_get_props,
49    /* APIR_COMMAND_TYPE_DEVICE_BUFFER_FROM_PTR  = */ backend_device_buffer_from_ptr,
50
51    /* buffer-type */
52
53    /* APIR_COMMAND_TYPE_BUFFER_TYPE_GET_NAME  = */ backend_buffer_type_get_name,
54    /* APIR_COMMAND_TYPE_BUFFER_TYPE_GET_ALIGNMENT  = */ backend_buffer_type_get_alignment,
55    /* APIR_COMMAND_TYPE_BUFFER_TYPE_GET_MAX_SIZE  = */ backend_buffer_type_get_max_size,
56    /* APIR_COMMAND_TYPE_BUFFER_TYPE_IS_HOST  = */ backend_buffer_type_is_host /* DEPRECATED */,
57    /* APIR_COMMAND_TYPE_BUFFER_TYPE_ALLOC_BUFFER  = */ backend_buffer_type_alloc_buffer,
58    /* APIR_COMMAND_TYPE_BUFFER_TYPE_GET_ALLOC_SIZE  = */ backend_buffer_type_get_alloc_size,
59
60    /* buffer */
61
62    /* APIR_COMMAND_TYPE_BUFFER_GET_BASE  = */ backend_buffer_get_base,
63    /* APIR_COMMAND_TYPE_BUFFER_SET_TENSOR  = */ backend_buffer_set_tensor,
64    /* APIR_COMMAND_TYPE_BUFFER_GET_TENSOR  = */ backend_buffer_get_tensor,
65    /* APIR_COMMAND_TYPE_BUFFER_CPY_TENSOR  = */ backend_buffer_cpy_tensor,
66    /* APIR_COMMAND_TYPE_BUFFER_CLEAR  = */ backend_buffer_clear,
67    /* APIR_COMMAND_TYPE_BUFFER_FREE_BUFFER  = */ backend_buffer_free_buffer,
68
69    /* backend */
70
71    /* APIR_COMMAND_TYPE_BACKEND_GRAPH_COMPUTE  = */ backend_backend_graph_compute,
72};
73}