File: internal.h

package info (click to toggle)
llvm-toolchain-16 1%3A16.0.6-15~deb12u1
  • links: PTS, VCS
  • area: main
  • in suites: bookworm
  • size: 1,634,792 kB
  • sloc: cpp: 6,179,261; ansic: 1,216,205; asm: 741,319; python: 196,614; objc: 75,325; f90: 49,640; lisp: 32,396; pascal: 12,286; sh: 9,394; perl: 7,442; ml: 5,494; awk: 3,523; makefile: 2,723; javascript: 1,206; xml: 886; fortran: 581; cs: 573
file content (154 lines) | stat: -rw-r--r-- 3,876 bytes parent folder | download | duplicates (2)
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
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
//===--- amdgpu/impl/internal.h ----------------------------------- C++ -*-===//
//
// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
// See https://llvm.org/LICENSE.txt for license information.
// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
//
//===----------------------------------------------------------------------===//
#ifndef SRC_RUNTIME_INCLUDE_INTERNAL_H_
#define SRC_RUNTIME_INCLUDE_INTERNAL_H_
#include <inttypes.h>
#include <pthread.h>
#include <stddef.h>
#include <stdint.h>
#include <stdio.h>
#include <stdlib.h>

#include <cstring>
#include <map>
#include <queue>
#include <string>
#include <utility>
#include <vector>

#include "hsa_api.h"

#include "impl_runtime.h"

#ifndef TARGET_NAME
#error "Missing TARGET_NAME macro"
#endif
#define DEBUG_PREFIX "Target " GETNAME(TARGET_NAME) " RTL"
#include "Debug.h"

#define MAX_NUM_KERNELS (1024 * 16)

// ---------------------- Kernel Start -------------
typedef struct atl_kernel_info_s {
  uint64_t kernel_object;
  uint32_t group_segment_size;
  uint32_t private_segment_size;
  uint32_t sgpr_count;
  uint32_t vgpr_count;
  uint32_t sgpr_spill_count;
  uint32_t vgpr_spill_count;
  uint32_t kernel_segment_size;
  uint32_t explicit_argument_count;
  uint32_t implicit_argument_count;
} atl_kernel_info_t;

typedef struct atl_symbol_info_s {
  uint64_t addr;
  uint32_t size;
} atl_symbol_info_t;

// ---------------------- Kernel End -------------

namespace core {
class TaskgroupImpl;
class TaskImpl;
class Kernel;
class KernelImpl;
} // namespace core

struct SignalPoolT {
  SignalPoolT() {}
  SignalPoolT(const SignalPoolT &) = delete;
  SignalPoolT(SignalPoolT &&) = delete;
  ~SignalPoolT() {
    size_t N = state.size();
    for (size_t i = 0; i < N; i++) {
      hsa_signal_t signal = state.front();
      state.pop();
      hsa_status_t rc = hsa_signal_destroy(signal);
      if (rc != HSA_STATUS_SUCCESS) {
        DP("Signal pool destruction failed\n");
      }
    }
  }
  size_t size() {
    lock l(&mutex);
    return state.size();
  }
  void push(hsa_signal_t s) {
    lock l(&mutex);
    state.push(s);
  }
  hsa_signal_t pop(void) {
    lock l(&mutex);
    if (!state.empty()) {
      hsa_signal_t res = state.front();
      state.pop();
      return res;
    }

    // Pool empty, attempt to create another signal
    hsa_signal_t new_signal;
    hsa_status_t err = hsa_signal_create(0, 0, NULL, &new_signal);
    if (err == HSA_STATUS_SUCCESS) {
      return new_signal;
    }

    // Fail
    return {0};
  }

private:
  static pthread_mutex_t mutex;
  std::queue<hsa_signal_t> state;
  struct lock {
    lock(pthread_mutex_t *m) : m(m) { pthread_mutex_lock(m); }
    ~lock() { pthread_mutex_unlock(m); }
    pthread_mutex_t *m;
  };
};

namespace core {
hsa_status_t atl_init_gpu_context();

hsa_status_t init_hsa();
hsa_status_t finalize_hsa();
/*
 * Generic utils
 */
template <typename T> inline T alignDown(T value, size_t alignment) {
  return (T)(value & ~(alignment - 1));
}

template <typename T> inline T *alignDown(T *value, size_t alignment) {
  return reinterpret_cast<T *>(alignDown((intptr_t)value, alignment));
}

template <typename T> inline T alignUp(T value, size_t alignment) {
  return alignDown((T)(value + alignment - 1), alignment);
}

template <typename T> inline T *alignUp(T *value, size_t alignment) {
  return reinterpret_cast<T *>(
      alignDown((intptr_t)(value + alignment - 1), alignment));
}

extern bool atl_is_impl_initialized();

bool handle_group_signal(hsa_signal_value_t value, void *arg);

hsa_status_t allow_access_to_all_gpu_agents(void *ptr);
} // namespace core

inline const char *get_error_string(hsa_status_t err) {
  const char *res;
  hsa_status_t rc = hsa_status_string(err, &res);
  return (rc == HSA_STATUS_SUCCESS) ? res : "HSA_STATUS UNKNOWN.";
}

#endif // SRC_RUNTIME_INCLUDE_INTERNAL_H_