1//
 2// MIT license
 3// Copyright (C) 2024 Intel Corporation
 4// SPDX-License-Identifier: MIT
 5//
 6
 7//
 8// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
 9// See https://llvm.org/LICENSE.txt for license information.
10// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
11//
12
13#include "common.hpp"
14
15#include "ggml-backend-impl.h"
16#include "ggml-impl.h"
17
18int get_current_device_id() {
19  return dpct::dev_mgr::instance().current_device_id();
20}
21
22void* ggml_sycl_host_malloc(size_t size) try {
23  if (getenv("GGML_SYCL_NO_PINNED") != nullptr) {
24    return nullptr;
25  }
26
27  void* ptr = nullptr;
28  // allow to use dpct::get_in_order_queue() for host malloc
29  dpct::err0 err = CHECK_TRY_ERROR(
30      ptr = (void*)sycl::malloc_host(size, dpct::get_in_order_queue()));
31
32  if (err != 0) {
33    // clear the error
34    GGML_LOG_ERROR("WARNING: failed to allocate %.2f MB of pinned memory: %s\n", size / 1024.0 / 1024.0,    "syclGetErrorString is not supported");
35    return nullptr;
36  }
37
38  return ptr;
39} catch (sycl::exception const& exc) {
40  std::cerr << exc.what() << "Exception caught at file:" << __FILE__
41            << ", line:" << __LINE__ << std::endl;
42  std::exit(1);
43}
44
45void ggml_sycl_host_free(void* ptr) try {
46  // allow to use dpct::get_in_order_queue() for host malloc
47  SYCL_CHECK(CHECK_TRY_ERROR(sycl::free(ptr, dpct::get_in_order_queue())));
48} catch (sycl::exception const& exc) {
49  std::cerr << exc.what() << "Exception caught at file:" << __FILE__
50            << ", line:" << __LINE__ << std::endl;
51  std::exit(1);
52}
53
54bool gpu_has_xmx(sycl::device &dev) {
55    return dev.has(sycl::aspect::ext_intel_matrix);
56}
57
58int64_t downsample_sycl_global_range(int64_t accumulate_block_num, int64_t block_size) {
59  const int64_t max_range = std::numeric_limits<int>::max();
60  int64_t sycl_down_blk_size = block_size;
61  int64_t global_range = accumulate_block_num * sycl_down_blk_size;
62  while(global_range > max_range) {
63      sycl_down_blk_size /= 2;
64      global_range = accumulate_block_num * sycl_down_blk_size;
65  }
66  return sycl_down_blk_size;
67}
68
69void release_extra_gpu(ggml_tensor_extra_gpu * extra, std::vector<queue_ptr> streams) {
70    for (int i = 0; i < ggml_sycl_info().device_count; ++i) {
71        for (int64_t is = 0; is < GGML_SYCL_MAX_STREAMS; ++is) {
72            if (extra->events[i][is] != nullptr) {
73                SYCL_CHECK(CHECK_TRY_ERROR(dpct::destroy_event(extra->events[i][is])));
74            }
75        }
76        if (extra->data_device[i] != nullptr && streams.size()>0) {
77            ggml_sycl_set_device(i);
78            SYCL_CHECK(
79                CHECK_TRY_ERROR(sycl::free(extra->data_device[i], *(streams[i]))));
80        }
81    }
82    delete extra;
83}