char-fork/llama.cpp
git clone https://git.t4t.associates/char-fork/llama.cpp
ffaafde16
char/aliceai-t5-moe
1#pragma once 2 3// clang-format off 4#include "virtgpu.h" 5#include "ggml-remoting.h" 6#include "backend/shared/apir_backend.h" 7#include "backend/shared/apir_cs_ggml.h" 8#include "ggml-backend-impl.h" 9// clang-format on 10 11#define REMOTE_CALL_PREPARE (gpu_dev_name ,encoder_name ,apir_command_type__ ) \ 12 int32_t REMOTE_CALL_PREPARE_forward_flag = (int32_t) apir_command_type__; \ 13 const char * REMOTE_CALL_PREPARE_command_name = apir_dispatch_command_name(apir_command_type__); \ 14 do { \ 15 encoder_name = remote_call_prepare(gpu_dev_name, APIR_COMMAND_TYPE_FORWARD, REMOTE_CALL_PREPARE_forward_flag); \ 16 if (!encoder_name) { \ 17 GGML_ABORT(GGML_VIRTGPU "%s: failed to prepare the remote call encoder", __func__); \ 18 } \ 19 } while (0) 20 21#define REMOTE_CALL (gpu_dev_name ,encoder_name ,decoder_name ,ret_name ) \ 22 do { \ 23 ret_name = (ApirForwardReturnCode) remote_call(gpu_dev_name, encoder_name, &decoder_name, 0, NULL); \ 24 if (!decoder_name) { \ 25 GGML_ABORT(GGML_VIRTGPU "%s: failed to kick the remote call", __func__); \ 26 } \ 27 if (ret_name < APIR_FORWARD_BASE_INDEX) { \ 28 GGML_ABORT(GGML_VIRTGPU "%s: failed to forward the API call: %s: code %d", __func__, \ 29 apir_forward_error(ret_name), ret_name); \ 30 } \ 31 ret_name = (ApirForwardReturnCode) (ret_name - APIR_FORWARD_BASE_INDEX); \ 32 if (ret_name != 0) { \ 33 GGML_ABORT(GGML_VIRTGPU "backend function '%s' failed (return code: %d)", \ 34 REMOTE_CALL_PREPARE_command_name, ret_name); \ 35 } \ 36 } while (0)