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
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
|
//===------------- AMDGPU implementation of timing utils --------*- 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 LLVM_LIBC_UTILS_GPU_TIMING_AMDGPU
#define LLVM_LIBC_UTILS_GPU_TIMING_AMDGPU
#include "hdr/stdint_proxy.h"
#include "src/__support/CPP/algorithm.h"
#include "src/__support/CPP/array.h"
#include "src/__support/CPP/atomic.h"
#include "src/__support/CPP/type_traits.h"
#include "src/__support/GPU/utils.h"
#include "src/__support/macros/attributes.h"
#include "src/__support/macros/config.h"
namespace LIBC_NAMESPACE_DECL {
// Returns the overhead associated with calling the profiling region. This
// allows us to substract the constant-time overhead from the latency to
// obtain a true result. This can vary with system load.
[[gnu::noinline]] static LIBC_INLINE uint64_t overhead() {
cpp::atomic_thread_fence(cpp::MemoryOrder::ACQ_REL);
uint64_t start = gpu::processor_clock();
uint32_t result = 0.0;
asm("v_or_b32 %[v_reg], 0, %[v_reg]\n" ::[v_reg] "v"(result));
asm("" ::"s"(start));
uint64_t stop = gpu::processor_clock();
return stop - start;
}
// Profile a simple function and obtain its latency in clock cycles on the
// system. This function cannot be inlined or else it will disturb the very
// delicate balance of hard-coded dependencies.
template <typename F, typename T>
[[gnu::noinline]] static LIBC_INLINE uint64_t latency(F f, T t) {
// We need to store the input somewhere to guarantee that the compiler
// will not constant propagate it and remove the profiling region.
volatile T storage = t;
T arg = storage;
// The AMDGPU architecture needs to wait on pending results.
cpp::atomic_thread_fence(cpp::MemoryOrder::ACQ_REL);
// Get the current timestamp from the clock.
uint64_t start = gpu::processor_clock();
// This forces the compiler to load the input argument and run the clock
// cycle counter before the profiling region.
asm("" : "+v"(arg) : "s"(start));
// Run the function under test and return its value.
auto result = f(arg);
// This inline assembly performs a no-op which forces the result to both
// be used and prevents us from exiting this region before it's complete.
if constexpr (cpp::is_same_v<decltype(result), char> ||
cpp::is_same_v<decltype(result), bool>)
// AMDGPU does not support input register constraints for i1 and i8, so we
// cast it to a 32-bit integer. This does not add an additional assembly
// instruction (https://godbolt.org/z/zxGqv8G91).
asm("v_or_b32 %[v_reg], 0, %[v_reg]\n" ::[v_reg] "v"(
static_cast<uint32_t>(result)));
else
asm("v_or_b32 %[v_reg], 0, %[v_reg]\n" ::[v_reg] "v"(result));
// Obtain the current timestamp after running the calculation and force
// ordering.
uint64_t stop = gpu::processor_clock();
asm("" ::"s"(stop));
cpp::atomic_thread_fence(cpp::MemoryOrder::ACQ_REL);
// Return the time elapsed.
return stop - start;
}
template <typename F, typename T1, typename T2>
[[gnu::noinline]] static LIBC_INLINE uint64_t latency(F f, T1 t1, T2 t2) {
volatile T1 storage1 = t1;
volatile T2 storage2 = t2;
T1 arg1 = storage1;
T2 arg2 = storage2;
cpp::atomic_thread_fence(cpp::MemoryOrder::ACQ_REL);
uint64_t start = gpu::processor_clock();
asm("" ::"s"(start));
auto result = f(arg1, arg2);
if constexpr (cpp::is_same_v<decltype(result), char> ||
cpp::is_same_v<decltype(result), bool>)
asm("v_or_b32 %[v_reg], 0, %[v_reg]\n" ::[v_reg] "v"(
static_cast<uint32_t>(result)));
else
asm("v_or_b32 %[v_reg], 0, %[v_reg]\n" ::[v_reg] "v"(result));
uint64_t stop = gpu::processor_clock();
asm("" ::"s"(stop));
cpp::atomic_thread_fence(cpp::MemoryOrder::ACQ_REL);
return stop - start;
}
// Provides the *baseline* for throughput: measures loop and measurement costs
// without calling the f function
template <typename T, size_t N>
static LIBC_INLINE uint64_t
throughput_baseline(const cpp::array<T, N> &inputs) {
asm("" ::"v"(&inputs));
cpp::atomic_thread_fence(cpp::MemoryOrder::ACQ_REL);
uint64_t start = gpu::processor_clock();
asm("" ::"s"(start));
T result{};
#pragma clang loop unroll(disable)
for (auto input : inputs) {
asm("" ::"v"(input));
result = input;
asm("" ::"v"(result));
}
uint64_t stop = gpu::processor_clock();
asm("" ::"s"(stop));
cpp::atomic_thread_fence(cpp::MemoryOrder::ACQ_REL);
volatile auto output = result;
return stop - start;
}
// Provides throughput benchmarking
template <typename F, typename T, size_t N>
static LIBC_INLINE uint64_t throughput(F f, const cpp::array<T, N> &inputs) {
uint64_t baseline = UINT64_MAX;
for (int i = 0; i < 5; ++i)
baseline = cpp::min(baseline, throughput_baseline<T, N>(inputs));
asm("" ::"v"(&inputs));
cpp::atomic_thread_fence(cpp::MemoryOrder::ACQ_REL);
uint64_t start = gpu::processor_clock();
asm("" ::"s"(start));
T result{};
#pragma clang loop unroll(disable)
for (auto input : inputs) {
asm("" ::"v"(input));
result = f(input);
asm("" ::"v"(result));
}
uint64_t stop = gpu::processor_clock();
asm("" ::"s"(stop));
cpp::atomic_thread_fence(cpp::MemoryOrder::ACQ_REL);
volatile auto output = result;
const uint64_t measured = stop - start;
return measured > baseline ? (measured - baseline) : 0;
}
// Provides the *baseline* for throughput with 2 arguments: measures loop and
// measurement costs without calling the f function
template <typename T, size_t N>
static LIBC_INLINE uint64_t throughput_baseline(
const cpp::array<T, N> &inputs1, const cpp::array<T, N> &inputs2) {
asm("" ::"v"(&inputs1), "v"(&inputs2));
cpp::atomic_thread_fence(cpp::MemoryOrder::ACQ_REL);
uint64_t start = gpu::processor_clock();
asm("" ::"s"(start));
T result{};
#pragma clang loop unroll(disable)
for (size_t i = 0; i < N; i++) {
T x = inputs1[i];
T y = inputs2[i];
asm("" ::"v"(x), "v"(y));
result = x;
asm("" ::"v"(result));
}
uint64_t stop = gpu::processor_clock();
asm("" ::"s"(stop));
cpp::atomic_thread_fence(cpp::MemoryOrder::ACQ_REL);
volatile auto output = result;
return stop - start;
}
// Provides throughput benchmarking for 2 arguments (e.g. atan2())
template <typename F, typename T, size_t N>
static LIBC_INLINE uint64_t throughput(F f, const cpp::array<T, N> &inputs1,
const cpp::array<T, N> &inputs2) {
uint64_t baseline = UINT64_MAX;
for (int i = 0; i < 5; ++i)
baseline = cpp::min(baseline, throughput_baseline<T, N>(inputs1, inputs2));
asm("" ::"v"(&inputs1), "v"(&inputs2));
cpp::atomic_thread_fence(cpp::MemoryOrder::ACQ_REL);
uint64_t start = gpu::processor_clock();
asm("" ::"s"(start));
T result{};
#pragma clang loop unroll(disable)
for (size_t i = 0; i < N; i++) {
T x = inputs1[i];
T y = inputs2[i];
asm("" ::"v"(x), "v"(y));
result = f(x, y);
asm("" ::"v"(result));
}
uint64_t stop = gpu::processor_clock();
asm("" ::"s"(stop));
cpp::atomic_thread_fence(cpp::MemoryOrder::ACQ_REL);
volatile auto output = result;
const uint64_t measured = stop - start;
return measured > baseline ? (measured - baseline) : 0;
}
} // namespace LIBC_NAMESPACE_DECL
#endif // LLVM_LIBC_UTILS_GPU_TIMING_AMDGPU
|
