1 // Copyright 2019 Google LLC
2 //
3 // This source code is licensed under the BSD-style license found in the
4 // LICENSE file in the root directory of this source tree.
5
6 #pragma once
7
8 #include <cstddef>
9 #include <cstdint>
10
11 #include <benchmark/benchmark.h>
12
13 namespace benchmark {
14 namespace utils {
15
16 uint32_t WipeCache();
17 uint32_t PrefetchToL1(const void* ptr, size_t size);
18
19 // Disable support for denormalized numbers in floating-point units.
20 void DisableDenormals();
21
22 // Return clock rate, in Hz, for the currently used logical processor.
23 uint64_t GetCurrentCpuFrequency();
24
25 // Return maximum (across all cores/clusters/sockets) last level cache size.
26 // Can overestimate, but not underestimate LLC size.
27 size_t GetMaxCacheSize();
28
29 // Set multi-threading parameters appropriate for the processor.
30 void MultiThreadingParameters(benchmark::internal::Benchmark* benchmark);
31
32 typedef bool (*IsaCheckFunction)(benchmark::State& state);
33
34 // Check if either ARM VFPv2 or VFPv3 extension is supported.
35 // If VFP is unsupported, report error in benchmark state, and return false.
36 bool CheckVFP(benchmark::State& state);
37
38 // Check if ARM NEON-FP16-ARITH extension is supported.
39 // If NEON-FP16-ARITH is unsupported, report error in benchmark state, and return false.
40 bool CheckNEONFP16ARITH(benchmark::State& state);
41
42 // Check if ARM NEON extension is supported.
43 // If NEON is unsupported, report error in benchmark state, and return false.
44 bool CheckNEON(benchmark::State& state);
45
46 // Check if ARM NEON-FMA extension is supported.
47 // If NEON-FMA is unsupported, report error in benchmark state, and return false.
48 bool CheckNEONFMA(benchmark::State& state);
49
50 // Check if ARM DOT extension is supported.
51 // If DOT is unsupported, report error in benchmark state, and return false.
52 bool CheckNEONDOT(benchmark::State& state);
53
54 // Check if x86 SSSE3 extension is supported.
55 // If SSSE3 is unsupported, report error in benchmark state, and return false.
56 bool CheckSSSE3(benchmark::State& state);
57
58 // Check if x86 SSE4.1 extension is supported.
59 // If SSE4.1 is unsupported, report error in benchmark state, and return false.
60 bool CheckSSE41(benchmark::State& state);
61
62 // Check if x86 AVX extension is supported.
63 // If AVX is unsupported, report error in benchmark state, and return false.
64 bool CheckAVX(benchmark::State& state);
65
66 // Check if x86 XOP extension is supported.
67 // If XOP is unsupported, report error in benchmark state, and return false.
68 bool CheckXOP(benchmark::State& state);
69
70 // Check if x86 FMA3 extension is supported.
71 // If FMA3 is unsupported, report error in benchmark state, and return false.
72 bool CheckFMA3(benchmark::State& state);
73
74 // Check if x86 AVX2 extension is supported.
75 // If AVX2 is unsupported, report error in benchmark state, and return false.
76 bool CheckAVX2(benchmark::State& state);
77
78 // Check if x86 AVX512F extension is supported.
79 // If AVX512F is unsupported, report error in benchmark state, and return false.
80 bool CheckAVX512F(benchmark::State& state);
81
82 // Check if x86 SKX-level AVX512 extensions (AVX512F, AVX512CD, AVX512BW, AVX512DQ, and AVX512VL) are supported.
83 // If SKX-level AVX512 extensions are unsupported, report error in benchmark state, and return false.
84 bool CheckAVX512SKX(benchmark::State& state);
85
86 template <class T>
DivideRoundUp(T x,T q)87 inline T DivideRoundUp(T x, T q) {
88 return x / q + T(x % q != 0);
89 }
90
91 template <class T>
RoundUp(T x,T q)92 inline T RoundUp(T x, T q) {
93 return q * DivideRoundUp(x, q);
94 }
95
96 template <class T>
Doz(T a,T b)97 inline T Doz(T a, T b) {
98 return a >= b ? a - b : T(0);
99 }
100
101 } // namespace utils
102 } // namespace benchmark
103