| 1 | //===-- Benchmark memory specific tools -----------------------------------===// |
| 2 | // |
| 3 | // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. |
| 4 | // See https://llvm.org/LICENSE.txt for license information. |
| 5 | // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception |
| 6 | // |
| 7 | //===----------------------------------------------------------------------===// |
| 8 | |
| 9 | #include "LibcMemoryBenchmark.h" |
| 10 | #include "llvm/ADT/SmallVector.h" |
| 11 | #include "llvm/ADT/Twine.h" |
| 12 | #include "llvm/Support/ErrorHandling.h" |
| 13 | #include "llvm/Support/MathExtras.h" |
| 14 | #include <algorithm> |
| 15 | |
| 16 | namespace llvm { |
| 17 | namespace libc_benchmarks { |
| 18 | |
| 19 | // Returns a distribution that samples the buffer to satisfy the required |
| 20 | // alignment. |
| 21 | // When alignment is set, the distribution is scaled down by `Factor` and scaled |
| 22 | // up again by the same amount during sampling. |
| 23 | static std::uniform_int_distribution<uint32_t> |
| 24 | getOffsetDistribution(size_t BufferSize, size_t MaxSizeValue, |
| 25 | MaybeAlign AccessAlignment) { |
| 26 | if (AccessAlignment && *AccessAlignment > AlignedBuffer::Alignment) |
| 27 | report_fatal_error( |
| 28 | reason: "AccessAlignment must be less or equal to AlignedBuffer::Alignment" ); |
| 29 | if (!AccessAlignment) |
| 30 | return std::uniform_int_distribution<uint32_t>(0, 0); // Always 0. |
| 31 | // If we test up to Size bytes, the returned offset must stay under |
| 32 | // BuffersSize - Size. |
| 33 | int64_t MaxOffset = BufferSize; |
| 34 | MaxOffset -= MaxSizeValue; |
| 35 | MaxOffset -= 1; |
| 36 | if (MaxOffset < 0) |
| 37 | report_fatal_error( |
| 38 | reason: "BufferSize too small to exercise specified Size configuration" ); |
| 39 | MaxOffset /= AccessAlignment->value(); |
| 40 | return std::uniform_int_distribution<uint32_t>(0, MaxOffset); |
| 41 | } |
| 42 | |
| 43 | OffsetDistribution::OffsetDistribution(size_t BufferSize, size_t MaxSizeValue, |
| 44 | MaybeAlign AccessAlignment) |
| 45 | : Distribution( |
| 46 | getOffsetDistribution(BufferSize, MaxSizeValue, AccessAlignment)), |
| 47 | Factor(AccessAlignment.valueOrOne().value()) {} |
| 48 | |
| 49 | // Precomputes offset where to insert mismatches between the two buffers. |
| 50 | MismatchOffsetDistribution::MismatchOffsetDistribution(size_t BufferSize, |
| 51 | size_t MaxSizeValue, |
| 52 | size_t MismatchAt) |
| 53 | : MismatchAt(MismatchAt) { |
| 54 | if (MismatchAt <= 1) |
| 55 | return; |
| 56 | for (size_t I = MaxSizeValue + 1; I < BufferSize; I += MaxSizeValue) |
| 57 | MismatchIndices.push_back(Elt: I); |
| 58 | if (MismatchIndices.empty()) |
| 59 | report_fatal_error(reason: "Unable to generate mismatch" ); |
| 60 | MismatchIndexSelector = |
| 61 | std::uniform_int_distribution<size_t>(0, MismatchIndices.size() - 1); |
| 62 | } |
| 63 | |
| 64 | static size_t getL1DataCacheSize() { |
| 65 | const std::vector<CacheInfo> &CacheInfos = HostState::get().Caches; |
| 66 | const auto IsL1DataCache = [](const CacheInfo &CI) { |
| 67 | return CI.Type == "Data" && CI.Level == 1; |
| 68 | }; |
| 69 | const auto CacheIt = find_if(Range: CacheInfos, P: IsL1DataCache); |
| 70 | if (CacheIt != CacheInfos.end()) |
| 71 | return CacheIt->Size; |
| 72 | report_fatal_error(reason: "Unable to read L1 Cache Data Size" ); |
| 73 | } |
| 74 | |
| 75 | static constexpr int64_t KiB = 1024; |
| 76 | static constexpr int64_t ParameterStorageBytes = 4 * KiB; |
| 77 | static constexpr int64_t L1LeftAsideBytes = 1 * KiB; |
| 78 | |
| 79 | static size_t getAvailableBufferSize() { |
| 80 | return getL1DataCacheSize() - L1LeftAsideBytes - ParameterStorageBytes; |
| 81 | } |
| 82 | |
| 83 | ParameterBatch::ParameterBatch(size_t BufferCount) |
| 84 | : BufferSize(getAvailableBufferSize() / BufferCount), |
| 85 | BatchSize(ParameterStorageBytes / sizeof(ParameterType)), |
| 86 | Parameters(BatchSize) { |
| 87 | if (BufferSize <= 0 || BatchSize < 100) |
| 88 | report_fatal_error(reason: "Not enough L1 cache" ); |
| 89 | const size_t ParameterBytes = Parameters.size() * sizeof(ParameterType); |
| 90 | const size_t BufferBytes = BufferSize * BufferCount; |
| 91 | if (ParameterBytes + BufferBytes + L1LeftAsideBytes > getL1DataCacheSize()) |
| 92 | report_fatal_error( |
| 93 | reason: "We're splitting a buffer of the size of the L1 cache between a data " |
| 94 | "buffer and a benchmark parameters buffer, so by construction the " |
| 95 | "total should not exceed the size of the L1 cache" ); |
| 96 | } |
| 97 | |
| 98 | size_t ParameterBatch::getBatchBytes() const { |
| 99 | size_t BatchBytes = 0; |
| 100 | for (auto &P : Parameters) |
| 101 | BatchBytes += P.SizeBytes; |
| 102 | return BatchBytes; |
| 103 | } |
| 104 | |
| 105 | void ParameterBatch::checkValid(const ParameterType &P) const { |
| 106 | if (P.OffsetBytes + P.SizeBytes >= BufferSize) |
| 107 | report_fatal_error( |
| 108 | reason: llvm::Twine("Call would result in buffer overflow: Offset=" ) |
| 109 | .concat(Suffix: llvm::Twine(P.OffsetBytes)) |
| 110 | .concat(Suffix: ", Size=" ) |
| 111 | .concat(Suffix: llvm::Twine(P.SizeBytes)) |
| 112 | .concat(Suffix: ", BufferSize=" ) |
| 113 | .concat(Suffix: llvm::Twine(BufferSize))); |
| 114 | } |
| 115 | |
| 116 | CopySetup::CopySetup() |
| 117 | : ParameterBatch(2), SrcBuffer(ParameterBatch::BufferSize), |
| 118 | DstBuffer(ParameterBatch::BufferSize) {} |
| 119 | |
| 120 | MoveSetup::MoveSetup() |
| 121 | : ParameterBatch(3), Buffer(ParameterBatch::BufferSize * 3) {} |
| 122 | |
| 123 | ComparisonSetup::ComparisonSetup() |
| 124 | : ParameterBatch(2), LhsBuffer(ParameterBatch::BufferSize), |
| 125 | RhsBuffer(ParameterBatch::BufferSize) { |
| 126 | // The memcmp buffers always compare equal. |
| 127 | memset(s: LhsBuffer.begin(), c: 0xF, n: BufferSize); |
| 128 | memset(s: RhsBuffer.begin(), c: 0xF, n: BufferSize); |
| 129 | } |
| 130 | |
| 131 | SetSetup::SetSetup() |
| 132 | : ParameterBatch(1), DstBuffer(ParameterBatch::BufferSize) {} |
| 133 | |
| 134 | } // namespace libc_benchmarks |
| 135 | } // namespace llvm |
| 136 | |