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
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
|
//===- CRunnerUtils.cpp - Utils for MLIR execution ------------------------===//
//
// 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
//
//===----------------------------------------------------------------------===//
//
// This file implements basic functions to manipulate structured MLIR types at
// runtime. Entities in this file are meant to be retargetable, including on
// targets without a C++ runtime, and must be kept C compatible.
//
//===----------------------------------------------------------------------===//
#include "mlir/ExecutionEngine/CRunnerUtils.h"
#include "mlir/ExecutionEngine/Msan.h"
#ifndef _WIN32
#if defined(__FreeBSD__) || defined(__NetBSD__) || defined(__OpenBSD__) || \
defined(__DragonFly__)
#include <cstdlib>
#else
#include <alloca.h>
#endif
#include <sys/time.h>
#else
#include "malloc.h"
#endif // _WIN32
#include <algorithm>
#include <cinttypes>
#include <cstdio>
#include <cstdlib>
#include <numeric>
#include <random>
#include <string.h>
#ifdef MLIR_CRUNNERUTILS_DEFINE_FUNCTIONS
namespace {
template <typename V>
void stdSort(uint64_t n, V *p) {
std::sort(p, p + n);
}
} // namespace
// Small runtime support "lib" for vector.print lowering.
// By providing elementary printing methods only, this
// library can remain fully unaware of low-level implementation
// details of our vectors. Also useful for direct LLVM IR output.
extern "C" void printI64(int64_t i) { fprintf(stdout, "%" PRId64, i); }
extern "C" void printU64(uint64_t u) { fprintf(stdout, "%" PRIu64, u); }
extern "C" void printF32(float f) {
if (std::isnan(f) && std::signbit(f)) {
fprintf(stdout, "-nan");
} else {
fprintf(stdout, "%g", f);
}
}
extern "C" void printF64(double d) {
if (std::isnan(d) && std::signbit(d)) {
fprintf(stdout, "-nan");
} else {
fprintf(stdout, "%lg", d);
}
}
extern "C" void printString(char const *s) { fputs(s, stdout); }
extern "C" void printOpen() { fputs("( ", stdout); }
extern "C" void printClose() { fputs(" )", stdout); }
extern "C" void printComma() { fputs(", ", stdout); }
extern "C" void printNewline() { fputc('\n', stdout); }
extern "C" void memrefCopy(int64_t elemSize, UnrankedMemRefType<char> *srcArg,
UnrankedMemRefType<char> *dstArg) {
DynamicMemRefType<char> src(*srcArg);
DynamicMemRefType<char> dst(*dstArg);
int64_t rank = src.rank;
MLIR_MSAN_MEMORY_IS_INITIALIZED(src.sizes, rank * sizeof(int64_t));
// Handle empty shapes -> nothing to copy.
for (int rankp = 0; rankp < rank; ++rankp)
if (src.sizes[rankp] == 0)
return;
char *srcPtr = src.data + src.offset * elemSize;
char *dstPtr = dst.data + dst.offset * elemSize;
if (rank == 0) {
memcpy(dstPtr, srcPtr, elemSize);
return;
}
int64_t *indices = static_cast<int64_t *>(alloca(sizeof(int64_t) * rank));
int64_t *srcStrides = static_cast<int64_t *>(alloca(sizeof(int64_t) * rank));
int64_t *dstStrides = static_cast<int64_t *>(alloca(sizeof(int64_t) * rank));
// Initialize index and scale strides.
for (int rankp = 0; rankp < rank; ++rankp) {
indices[rankp] = 0;
srcStrides[rankp] = src.strides[rankp] * elemSize;
dstStrides[rankp] = dst.strides[rankp] * elemSize;
}
int64_t readIndex = 0, writeIndex = 0;
for (;;) {
// Copy over the element, byte by byte.
memcpy(dstPtr + writeIndex, srcPtr + readIndex, elemSize);
// Advance index and read position.
for (int64_t axis = rank - 1; axis >= 0; --axis) {
// Advance at current axis.
auto newIndex = ++indices[axis];
readIndex += srcStrides[axis];
writeIndex += dstStrides[axis];
// If this is a valid index, we have our next index, so continue copying.
if (src.sizes[axis] != newIndex)
break;
// We reached the end of this axis. If this is axis 0, we are done.
if (axis == 0)
return;
// Else, reset to 0 and undo the advancement of the linear index that
// this axis had. Then continue with the axis one outer.
indices[axis] = 0;
readIndex -= src.sizes[axis] * srcStrides[axis];
writeIndex -= dst.sizes[axis] * dstStrides[axis];
}
}
}
/// Prints GFLOPS rating.
extern "C" void printFlops(double flops) {
fprintf(stderr, "%lf GFLOPS\n", flops / 1.0E9);
}
/// Returns the number of seconds since Epoch 1970-01-01 00:00:00 +0000 (UTC).
extern "C" double rtclock() {
#ifndef _WIN32
struct timeval tp;
int stat = gettimeofday(&tp, nullptr);
if (stat != 0)
fprintf(stderr, "Error returning time from gettimeofday: %d\n", stat);
return (tp.tv_sec + tp.tv_usec * 1.0e-6);
#else
fprintf(stderr, "Timing utility not implemented on Windows\n");
return 0.0;
#endif // _WIN32
}
extern "C" void *mlirAlloc(uint64_t size) { return malloc(size); }
extern "C" void *mlirAlignedAlloc(uint64_t alignment, uint64_t size) {
#ifdef _WIN32
return _aligned_malloc(size, alignment);
#elif defined(__APPLE__)
// aligned_alloc was added in MacOS 10.15. Fall back to posix_memalign to also
// support older versions.
void *result = nullptr;
(void)::posix_memalign(&result, alignment, size);
return result;
#else
return aligned_alloc(alignment, size);
#endif
}
extern "C" void mlirFree(void *ptr) { free(ptr); }
extern "C" void mlirAlignedFree(void *ptr) {
#ifdef _WIN32
_aligned_free(ptr);
#else
free(ptr);
#endif
}
extern "C" void *rtsrand(uint64_t s) {
// Standard mersenne_twister_engine seeded with s.
return new std::mt19937(s);
}
extern "C" uint64_t rtrand(void *g, uint64_t m) {
std::mt19937 *generator = static_cast<std::mt19937 *>(g);
std::uniform_int_distribution<uint64_t> distrib(0, m);
return distrib(*generator);
}
extern "C" void rtdrand(void *g) {
std::mt19937 *generator = static_cast<std::mt19937 *>(g);
delete generator;
}
extern "C" void _mlir_ciface_shuffle(StridedMemRefType<uint64_t, 1> *mref,
void *g) {
assert(mref);
assert(mref->strides[0] == 1); // consecutive
std::mt19937 *generator = static_cast<std::mt19937 *>(g);
uint64_t s = mref->sizes[0];
uint64_t *data = mref->data + mref->offset;
std::iota(data, data + s, 0);
std::shuffle(data, data + s, *generator);
}
#define IMPL_STDSORT(VNAME, V) \
extern "C" void _mlir_ciface_stdSort##VNAME(uint64_t n, \
StridedMemRefType<V, 1> *vref) { \
assert(vref); \
assert(vref->strides[0] == 1); \
V *values = vref->data + vref->offset; \
stdSort(n, values); \
}
IMPL_STDSORT(I64, int64_t)
IMPL_STDSORT(F64, double)
IMPL_STDSORT(F32, float)
#undef IMPL_STDSORT
#endif // MLIR_CRUNNERUTILS_DEFINE_FUNCTIONS
|
