diff options
author | Ava Chow <github@achow101.com> | 2024-02-08 13:08:16 -0500 |
---|---|---|
committer | Ava Chow <github@achow101.com> | 2024-02-08 13:30:31 -0500 |
commit | ecbf4bae9cb1815ac0b2e2d4cac15471c9da44d2 (patch) | |
tree | 8f5ce4f0647cfe9089324105f25fa1d7aee9406b | |
parent | 0471aee5077fef81a0e2e032f3c16bdc948a70f2 (diff) | |
parent | fab41697a5448ef2861f65795bd63a4ccdda6a40 (diff) |
Merge bitcoin/bitcoin#29114: util: Faster std::byte (pre)vector (un)serialize
fab41697a5448ef2861f65795bd63a4ccdda6a40 Allow int8_t optimized vector serialization (MarcoFalke)
facaa14785e006c1af5a8b17b10e2722af8d054e Faster std::byte (pre)vector (un)serialize (MarcoFalke)
Pull request description:
Currently, large vectors of `std::byte` are (un)serialized byte-by-byte, which is slow. Fix this, by enabling the already existing optimization for them.
On my system this gives a 10x speedup for `./src/bench/bench_bitcoin --filter=PrevectorDeserializeTrivial`, when `std::byte` are used:
```diff
diff --git a/src/bench/prevector.cpp b/src/bench/prevector.cpp
index 2524e215e4..76b16bc34e 100644
--- a/src/bench/prevector.cpp
+++ b/src/bench/prevector.cpp
@@ -17,7 +17,7 @@ struct nontrivial_t {
static_assert(!std::is_trivially_default_constructible<nontrivial_t>::value,
"expected nontrivial_t to not be trivially constructible");
-typedef unsigned char trivial_t;
+typedef std::byte trivial_t;
static_assert(std::is_trivially_default_constructible<trivial_t>::value,
"expected trivial_t to be trivially constructible");
```
However, the optimization does not cover `signed char`. Fix that as well.
ACKs for top commit:
sipa:
utACK fab41697a5448ef2861f65795bd63a4ccdda6a40
achow101:
ACK fab41697a5448ef2861f65795bd63a4ccdda6a40
TheCharlatan:
ACK fab41697a5448ef2861f65795bd63a4ccdda6a40
Tree-SHA512: a3e20f375fd1d0e0dedb827a8ce528de1173ea69660c8c891ad1343a86b422072f6505096fca0d3f8af4442fbe1378a02e32d5974919d4e88ff06934d0258cba
-rw-r--r-- | src/serialize.h | 16 | ||||
-rw-r--r-- | src/span.h | 2 |
2 files changed, 8 insertions, 10 deletions
diff --git a/src/serialize.h b/src/serialize.h index 19585c630a..a0b012b25c 100644 --- a/src/serialize.h +++ b/src/serialize.h @@ -710,14 +710,12 @@ template<typename Stream, typename C> void Unserialize(Stream& is, std::basic_st /** * prevector - * prevectors of unsigned char are a special case and are intended to be serialized as a single opaque blob. */ template<typename Stream, unsigned int N, typename T> inline void Serialize(Stream& os, const prevector<N, T>& v); template<typename Stream, unsigned int N, typename T> inline void Unserialize(Stream& is, prevector<N, T>& v); /** * vector - * vectors of unsigned char are a special case and are intended to be serialized as a single opaque blob. */ template<typename Stream, typename T, typename A> inline void Serialize(Stream& os, const std::vector<T, A>& v); template<typename Stream, typename T, typename A> inline void Unserialize(Stream& is, std::vector<T, A>& v); @@ -820,10 +818,9 @@ void Unserialize(Stream& is, std::basic_string<C>& str) template <typename Stream, unsigned int N, typename T> void Serialize(Stream& os, const prevector<N, T>& v) { - if constexpr (std::is_same_v<T, unsigned char>) { + if constexpr (BasicByte<T>) { // Use optimized version for unformatted basic bytes WriteCompactSize(os, v.size()); - if (!v.empty()) - os.write(MakeByteSpan(v)); + if (!v.empty()) os.write(MakeByteSpan(v)); } else { Serialize(os, Using<VectorFormatter<DefaultFormatter>>(v)); } @@ -833,7 +830,7 @@ void Serialize(Stream& os, const prevector<N, T>& v) template <typename Stream, unsigned int N, typename T> void Unserialize(Stream& is, prevector<N, T>& v) { - if constexpr (std::is_same_v<T, unsigned char>) { + if constexpr (BasicByte<T>) { // Use optimized version for unformatted basic bytes // Limit size per read so bogus size value won't cause out of memory v.clear(); unsigned int nSize = ReadCompactSize(is); @@ -856,10 +853,9 @@ void Unserialize(Stream& is, prevector<N, T>& v) template <typename Stream, typename T, typename A> void Serialize(Stream& os, const std::vector<T, A>& v) { - if constexpr (std::is_same_v<T, unsigned char>) { + if constexpr (BasicByte<T>) { // Use optimized version for unformatted basic bytes WriteCompactSize(os, v.size()); - if (!v.empty()) - os.write(MakeByteSpan(v)); + if (!v.empty()) os.write(MakeByteSpan(v)); } else if constexpr (std::is_same_v<T, bool>) { // A special case for std::vector<bool>, as dereferencing // std::vector<bool>::const_iterator does not result in a const bool& @@ -877,7 +873,7 @@ void Serialize(Stream& os, const std::vector<T, A>& v) template <typename Stream, typename T, typename A> void Unserialize(Stream& is, std::vector<T, A>& v) { - if constexpr (std::is_same_v<T, unsigned char>) { + if constexpr (BasicByte<T>) { // Use optimized version for unformatted basic bytes // Limit size per read so bogus size value won't cause out of memory v.clear(); unsigned int nSize = ReadCompactSize(is); diff --git a/src/span.h b/src/span.h index 2c27a54fc7..c974c265ce 100644 --- a/src/span.h +++ b/src/span.h @@ -287,9 +287,11 @@ Span<std::byte> MakeWritableByteSpan(V&& v) noexcept // Helper functions to safely cast basic byte pointers to unsigned char pointers. inline unsigned char* UCharCast(char* c) { return reinterpret_cast<unsigned char*>(c); } inline unsigned char* UCharCast(unsigned char* c) { return c; } +inline unsigned char* UCharCast(signed char* c) { return reinterpret_cast<unsigned char*>(c); } inline unsigned char* UCharCast(std::byte* c) { return reinterpret_cast<unsigned char*>(c); } inline const unsigned char* UCharCast(const char* c) { return reinterpret_cast<const unsigned char*>(c); } inline const unsigned char* UCharCast(const unsigned char* c) { return c; } +inline const unsigned char* UCharCast(const signed char* c) { return reinterpret_cast<const unsigned char*>(c); } inline const unsigned char* UCharCast(const std::byte* c) { return reinterpret_cast<const unsigned char*>(c); } // Helper concept for the basic byte types. template <typename B> |