Replace crc32 implementation

This is a properly constexpr version for modern c++. Also
added a fast runtime version of it.

include/wrenpp/detail/StringCRC32.hpp

@@ -1,148 +0,0 @@
-#ifndef id42C91D2875AE4E56BA61051619B58C03
-#define id42C91D2875AE4E56BA61051619B58C03
-
-#include <cstdint>
-#include <cstddef>
-
-namespace duckcore {
-	class StringCRC32 {
-		struct FnvHashBase {
-			[[gnu::pure]] [[gnu::always_inline]]
-			static uint32_t Calculate ( const char parChar, uint32_t parCrc );
-		};
-
-		//Update step: crc_32_tab[(crc ^ ch) & 0xff] ^ (crc >> 8)
-		template <uint32_t N, uint32_t I>
-		struct FnvHash : private FnvHashBase {
-			[[gnu::always_inline]]
-			static uint32_t Calculate ( const char (&str)[N], uint32_t crc ) {
-				const uint32_t prevcrc = FnvHash<N, I-1>::Calculate(str, crc);
-				return FnvHashBase::Calculate(str[I-1], prevcrc);
-			}
-		};
-		template <unsigned int N>
-		struct FnvHash<N, 1> : private FnvHashBase {
-			[[gnu::always_inline]]
-			static uint32_t Calculate ( const char (&str)[N], uint32_t crc ) {
-				return FnvHashBase::Calculate(str[0], crc);
-			}
-		};
-
-		class ConstCharWrapper {
-		public:
-			ConstCharWrapper ( const char* parStr ) : str(parStr) { }
-			const char* const str;
-		};
-	public:
-
-		StringCRC32 ( ConstCharWrapper parStr );
-
-		template <unsigned int N>
-		[[gnu::always_inline]]
-		StringCRC32 ( const char (&str)[N] ) :
-			m_hash(~FnvHash<N, N>::Calculate(str, 0xffffffff))
-		{
-		}
-
-		operator uint32_t ( void ) const { return m_hash; }
-
-	private:
-		uint32_t m_hash;
-
-	private:
-		//This is meant to represent a list of bits that are set to 1
-		template <int N, typename T>
-		struct BitArray {
-			enum { value = N };
-			typedef T Next;
-		};
-		//Polynomial from zlib: {0,1,2,4,5,7,8,10,11,12,16,22,23,26} or 0xedb88320UL
-//		for (int poly = 0, n = 0; n < (int)(sizeof(p)/sizeof(unsigned char)); n++)
-//			poly |= (z_crc_t)1 << (31 - p[n]);
-		typedef BitArray<0, BitArray<1, BitArray<2, BitArray<4, BitArray<5,
-				BitArray<7, BitArray<8, BitArray<10, BitArray<11, BitArray<12,
-				BitArray<16, BitArray<22, BitArray<23,
-				BitArray<26, std::nullptr_t> > > > > > > > > > > > > > PolynomialBits;
-		template <typename P, int M>
-		struct MakePolynomial {
-			enum { value = (1 << (M - P::value)) bitor MakePolynomial<typename P::Next, M>::value };
-		};
-		template <int M>
-		struct MakePolynomial<std::nullptr_t, M> {
-			enum { value = 0 };
-		};
-
-		//CRC32 algorithm from zlib:
-// 		for (int n = 0; n < 256; n++) {
-// 			c = n;
-// 			for (k = 0; k < 8; k++) c = c & 1 ? poly ^ (c >> 1) : c >> 1;
-// 			crc_table[n] = c;
-// 		}
-		template <int C, int S=8, uint32_t P=static_cast<uint32_t>(MakePolynomial<PolynomialBits, 31>::value)>
-		struct TableLookup {
-			static_assert(S > 1);
-			enum {
-				poly = P,
-				value = TableLookup<TableLookup<C, S-1, poly>::value, 1, poly>::value
-			};
-		};
-		template <int C, uint32_t P>
-		struct TableLookup<C, 1, P> {
-			enum {
-				poly = P,
-				value = ((C & 1) == 1 ? poly ^ (static_cast<uint32_t>(C) >> 1) : static_cast<uint32_t>(C) >> 1)
-			};
-		};
-	};
-
-	inline uint32_t StringCRC32::FnvHashBase::Calculate (const char parChar, uint32_t parCrc) {
-		static const uint32_t crc_32_tab[256] = { /* CRC polynomial 0xedb88320 */
-			0x00000000, 0x77073096, 0xee0e612c, 0x990951ba, 0x076dc419, 0x706af48f,
-			0xe963a535, 0x9e6495a3, 0x0edb8832, 0x79dcb8a4, 0xe0d5e91e, 0x97d2d988,
-			0x09b64c2b, 0x7eb17cbd, 0xe7b82d07, 0x90bf1d91, 0x1db71064, 0x6ab020f2,
-			0xf3b97148, 0x84be41de, 0x1adad47d, 0x6ddde4eb, 0xf4d4b551, 0x83d385c7,
-			0x136c9856, 0x646ba8c0, 0xfd62f97a, 0x8a65c9ec, 0x14015c4f, 0x63066cd9,
-			0xfa0f3d63, 0x8d080df5, 0x3b6e20c8, 0x4c69105e, 0xd56041e4, 0xa2677172,
-			0x3c03e4d1, 0x4b04d447, 0xd20d85fd, 0xa50ab56b, 0x35b5a8fa, 0x42b2986c,
-			0xdbbbc9d6, 0xacbcf940, 0x32d86ce3, 0x45df5c75, 0xdcd60dcf, 0xabd13d59,
-			0x26d930ac, 0x51de003a, 0xc8d75180, 0xbfd06116, 0x21b4f4b5, 0x56b3c423,
-			0xcfba9599, 0xb8bda50f, 0x2802b89e, 0x5f058808, 0xc60cd9b2, 0xb10be924,
-			0x2f6f7c87, 0x58684c11, 0xc1611dab, 0xb6662d3d, 0x76dc4190, 0x01db7106,
-			0x98d220bc, 0xefd5102a, 0x71b18589, 0x06b6b51f, 0x9fbfe4a5, 0xe8b8d433,
-			0x7807c9a2, 0x0f00f934, 0x9609a88e, 0xe10e9818, 0x7f6a0dbb, 0x086d3d2d,
-			0x91646c97, 0xe6635c01, 0x6b6b51f4, 0x1c6c6162, 0x856530d8, 0xf262004e,
-			0x6c0695ed, 0x1b01a57b, 0x8208f4c1, 0xf50fc457, 0x65b0d9c6, 0x12b7e950,
-			0x8bbeb8ea, 0xfcb9887c, 0x62dd1ddf, 0x15da2d49, 0x8cd37cf3, 0xfbd44c65,
-			0x4db26158, 0x3ab551ce, 0xa3bc0074, 0xd4bb30e2, 0x4adfa541, 0x3dd895d7,
-			0xa4d1c46d, 0xd3d6f4fb, 0x4369e96a, 0x346ed9fc, 0xad678846, 0xda60b8d0,
-			0x44042d73, 0x33031de5, 0xaa0a4c5f, 0xdd0d7cc9, 0x5005713c, 0x270241aa,
-			0xbe0b1010, 0xc90c2086, 0x5768b525, 0x206f85b3, 0xb966d409, 0xce61e49f,
-			0x5edef90e, 0x29d9c998, 0xb0d09822, 0xc7d7a8b4, 0x59b33d17, 0x2eb40d81,
-			0xb7bd5c3b, 0xc0ba6cad, 0xedb88320, 0x9abfb3b6, 0x03b6e20c, 0x74b1d29a,
-			0xead54739, 0x9dd277af, 0x04db2615, 0x73dc1683, 0xe3630b12, 0x94643b84,
-			0x0d6d6a3e, 0x7a6a5aa8, 0xe40ecf0b, 0x9309ff9d, 0x0a00ae27, 0x7d079eb1,
-			0xf00f9344, 0x8708a3d2, 0x1e01f268, 0x6906c2fe, 0xf762575d, 0x806567cb,
-			0x196c3671, 0x6e6b06e7, 0xfed41b76, 0x89d32be0, 0x10da7a5a, 0x67dd4acc,
-			0xf9b9df6f, 0x8ebeeff9, 0x17b7be43, 0x60b08ed5, 0xd6d6a3e8, 0xa1d1937e,
-			0x38d8c2c4, 0x4fdff252, 0xd1bb67f1, 0xa6bc5767, 0x3fb506dd, 0x48b2364b,
-			0xd80d2bda, 0xaf0a1b4c, 0x36034af6, 0x41047a60, 0xdf60efc3, 0xa867df55,
-			0x316e8eef, 0x4669be79, 0xcb61b38c, 0xbc66831a, 0x256fd2a0, 0x5268e236,
-			0xcc0c7795, 0xbb0b4703, 0x220216b9, 0x5505262f, 0xc5ba3bbe, 0xb2bd0b28,
-			0x2bb45a92, 0x5cb36a04, 0xc2d7ffa7, 0xb5d0cf31, 0x2cd99e8b, 0x5bdeae1d,
-			0x9b64c2b0, 0xec63f226, 0x756aa39c, 0x026d930a, 0x9c0906a9, 0xeb0e363f,
-			0x72076785, 0x05005713, 0x95bf4a82, 0xe2b87a14, 0x7bb12bae, 0x0cb61b38,
-			0x92d28e9b, 0xe5d5be0d, 0x7cdcefb7, 0x0bdbdf21, 0x86d3d2d4, 0xf1d4e242,
-			0x68ddb3f8, 0x1fda836e, 0x81be16cd, 0xf6b9265b, 0x6fb077e1, 0x18b74777,
-			0x88085ae6, 0xff0f6a70, 0x66063bca, 0x11010b5c, 0x8f659eff, 0xf862ae69,
-			0x616bffd3, 0x166ccf45, 0xa00ae278, 0xd70dd2ee, 0x4e048354, 0x3903b3c2,
-			0xa7672661, 0xd06016f7, 0x4969474d, 0x3e6e77db, 0xaed16a4a, 0xd9d65adc,
-			0x40df0b66, 0x37d83bf0, 0xa9bcae53, 0xdebb9ec5, 0x47b2cf7f, 0x30b5ffe9,
-			0xbdbdf21c, 0xcabac28a, 0x53b39330, 0x24b4a3a6, 0xbad03605, 0xcdd70693,
-			0x54de5729, 0x23d967bf, 0xb3667a2e, 0xc4614ab8, 0x5d681b02, 0x2a6f2b94,
-			0xb40bbe37, 0xc30c8ea1, 0x5a05df1b, 0x2d02ef8d
-		};
-		return crc_32_tab[(parCrc ^ parChar) & 0xff] ^ (parCrc >> 8);
-	}
-} //namespace duckcore
-
-#endif

include/wrenpp/detail/crc32.hpp

@@ -0,0 +1,95 @@
+/* Copyright 2020-2022, Michele Santullo
+ * This file is part of wrenpp.
+ *
+ * Wrenpp is free software: you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation, either version 3 of the License, or
+ * (at your option) any later version.
+ *
+ * Wrenpp is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with wrenpp.  If not, see <http://www.gnu.org/licenses/>.
+ */
+
+#pragma once
+
+#include <cstdint>
+#include <cstddef>
+#include <type_traits>
+#include <utility>
+
+namespace wren {
+	namespace detail {
+		[[gnu::const]]
+		std::uint32_t runtime_crc32c (const char* data, std::size_t size, std::uint32_t crc);
+
+		constexpr std::uint32_t g_castagnoli_polynomial = 0x1EDC6F41;
+
+		constexpr std::uint8_t reverse (std::uint8_t b) {
+			//see https://graphics.stanford.edu/~seander/bithacks.html#ReverseByteWith64Bits
+			return ((b * 0x80200802ULL) & 0x0884422110ULL) * 0x0101010101ULL >> 32;
+		}
+
+		constexpr std::uint32_t reverse (std::uint32_t val) {
+			return (reverse(static_cast<std::uint8_t>(val & 0xff)) << 24) |
+				(reverse(static_cast<std::uint8_t>(val >> 8 & 0xff)) << 16) |
+				(reverse(static_cast<std::uint8_t>(val >> 16 & 0xff)) << 8) |
+				reverse(static_cast<std::uint8_t>(val >> 24 & 0xff));
+		}
+
+		consteval std::uint32_t calc_table_entry (std::uint32_t index, std::uint32_t polynomial) {
+			polynomial = reverse(polynomial);
+			auto c = static_cast<std::uint_fast32_t>(index);
+			for (std::uint_fast32_t k = 0; k < 8; ++k) {
+				c = (c >> 1) ^ ((c & 0x1u) ? polynomial : 0);
+			}
+			return static_cast<std::uint32_t>(c);
+		}
+
+		template <std::uint32_t Polynomial, typename T>
+		struct PolynomialTableBase;
+
+		template <std::uint32_t Polynomial, std::uint32_t... Indices>
+		struct PolynomialTableBase<Polynomial, std::integer_sequence<std::uint32_t, Indices...>> {
+			static constexpr std::size_t table_size = sizeof...(Indices);
+			static constexpr std::uint32_t table[table_size] = {calc_table_entry(Indices, Polynomial)...};
+		};
+
+		template <std::uint32_t Polynomial>
+		struct PolynomialTable : public PolynomialTableBase<Polynomial, decltype(std::make_integer_sequence<std::uint32_t, 256>{})> {
+		};
+
+		template <std::uint32_t Polynomial>
+		constexpr inline auto g_polynomial_table = PolynomialTable<Polynomial>::table;
+
+		template <std::uint32_t Polynomial, std::uint32_t XorIn=0xffffffff, std::uint32_t XorOut=XorIn>
+		constexpr std::uint32_t crc32 (const char* data, std::size_t len, std::uint32_t crc) {
+			//static_assert(g_polynomial_table<Polynomial>[0b10000000] == Polynomial);
+			crc ^= XorIn;
+			for (std::size_t z = 0; z < len; ++z) {
+				crc = g_polynomial_table<Polynomial>[(crc ^ static_cast<unsigned char>(data[z])) & 0xff] ^ (crc >> 8);
+			}
+			return crc ^ XorOut;
+		}
+	} //namespace detail
+
+	[[gnu::const]]
+	constexpr std::uint32_t crc32c (const char* data, std::size_t size) {
+		if (not std::is_constant_evaluated()) {
+			return detail::runtime_crc32c(data, size, 0);
+		}
+		else {
+			return detail::crc32<detail::g_castagnoli_polynomial>(data, size, 0);
+		}
+	}
+
+	template <std::size_t N>
+	[[gnu::const]]
+	constexpr std::uint32_t crc32c (const char (&data)[N]) {
+		return crc32c(data, N);
+	}
+} //namespace wren

include/wrenpp/detail/meson.build

@@ -1,12 +1,12 @@
 include_files = [
   'construct_foreign_class.hpp',
+  'crc32.hpp',
   'error_type.hpp',
   'guess_class_name.hpp',
   'has_method.hpp',
   'module_and_name.hpp',
   'setters_getters.hpp',
   'string_bt.hpp',
-  'StringCRC32.hpp',
   'strings_in_vector.hpp',
   'wren_class_name_from_type.hpp',
   'wren_types.hpp',

include/wrenpp/vm.hpp

@@ -19,7 +19,7 @@
 
 #include "detail/has_method.hpp"
 #include "detail/error_type.hpp"
-#include "detail/StringCRC32.hpp"
+#include "detail/crc32.hpp"
 #include "detail/wren_types.hpp"
 #include "handle.hpp"
 #include <memory>
@@ -154,8 +154,8 @@ namespace wren {
 
 		template <typename U>
 		[[gnu::const]]
-		inline constexpr std::uint32_t type_id() {
+		inline consteval std::uint32_t type_id() {
-			return duckcore::StringCRC32(__PRETTY_FUNCTION__);
+			return crc32c(__PRETTY_FUNCTION__);
 		}
 	} //namespace detail
 

meson.build

@@ -44,10 +44,14 @@ compiler_opts = []
 if get_option('wrenpp_with_name_guessing')
   compiler_opts += ['-DWRENPP_WITH_NAME_GUESSING']
 endif
+if get_option('wrenpp_with_sse42')
+  compiler_opts += ['-msse4.2']
+endif
 
 conf.set('POINTER_SIZE', ptr_size)
 conf.set('FUNC_POINTER_SIZE', func_ptr_size)
 conf.set('WRENPP_NAME', meson.project_name())
+conf.set('WRENPP_WITH_SSE42', get_option('wrenpp_with_sse42'))
 
 project_config_file = configure_file(
 	input: 'src/pvt_config.h.in',
@@ -69,6 +73,7 @@ wrenpp = library(meson.project_name(),
   'src/callback_manager.cpp',
   'src/class_manager.cpp',
   'src/wren_class_name_from_type.cpp',
+  'src/crc32.cpp',
   dependencies: [wren_dep],
   include_directories: public_incl,
   install: (not meson.is_subproject() or get_option('default_library')=='shared'),

meson_options.txt

@@ -3,3 +3,4 @@ option('build_examples', type: 'boolean', value: false)
 option('wren_with_rand', type: 'boolean', value: false)
 option('wren_with_meta', type: 'boolean', value: false)
 option('wrenpp_with_name_guessing', type: 'boolean', value: true)
+option('wrenpp_with_sse42', type: 'boolean', value: true)

src/crc32.cpp

@@ -0,0 +1,92 @@
+/* Copyright 2020-2022, Michele Santullo
+ * This file is part of wrenpp.
+ *
+ * Wrenpp is free software: you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation, either version 3 of the License, or
+ * (at your option) any later version.
+ *
+ * Wrenpp is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with wrenpp.  If not, see <http://www.gnu.org/licenses/>.
+ */
+
+#include "wrenpp/detail/crc32.hpp"
+#include "pvt_config.h"
+#if defined(WRENPP_WITH_SSE42)
+#	if defined(__aarch64__)
+#		include "sse2neon.h"
+#	else
+#		include <smmintrin.h>
+#		include <immintrin.h>
+#	endif
+#endif
+
+#if defined(WRENPP_WITH_SSE42)
+	// Byte-boundary alignment issues
+#	define CALC_CRC(op, crc, type, buf, len)							   \
+  do {																  \
+	for (; (len) >= sizeof (type); (len) -= sizeof(type), buf += sizeof (type)) { \
+	  (crc) = op((crc), *(type *) (buf));							   \
+	}																   \
+  } while(0)
+#endif
+
+namespace wren {
+	namespace {
+		//Some useful polynomials:
+		// zlib: 0x04C11DB7
+		// castagnoli (intel):  0x1EDC6F41
+
+#if defined(WRENPP_WITH_SSE42)
+		/* Compute CRC-32C using the Intel hardware instruction. */
+		/* for better parallelization with bigger buffers see
+		   http://www.drdobbs.com/parallel/fast-parallelized-crc-computation-using/229401411 */
+		std::uint32_t crc32c_hw(const void *input, std::size_t len, std::uint32_t crc)
+		{
+			//see https://github.com/rurban/smhasher/blob/master/crc32_hw.c
+			constexpr std::size_t align_size = alignof(std::uint64_t);
+			constexpr std::size_t align_mask = align_size - 1;
+
+			const char* buf = static_cast<const char*>(input);
+
+			// XOR the initial CRC with INT_MAX
+			crc ^= 0xFFFFFFFF;
+
+			// Align the input to the word boundary
+			for (; (len > 0) && (reinterpret_cast<std::uintptr_t>(buf) & align_mask); len--, buf++) {
+				crc = _mm_crc32_u8(crc, *buf);
+			}
+
+			// Blast off the CRC32 calculation
+#if defined(__x86_64__) || defined(__aarch64__)
+			CALC_CRC(_mm_crc32_u64, crc, std::uint64_t, buf, len);
+#endif
+			CALC_CRC(_mm_crc32_u32, crc, std::uint32_t, buf, len);
+			CALC_CRC(_mm_crc32_u16, crc, std::uint16_t, buf, len);
+			CALC_CRC(_mm_crc32_u8, crc, std::uint8_t, buf, len);
+
+			// Post-process the crc
+			return ~crc;
+		}
+#endif
+	} //unnamed namespace
+
+	[[gnu::const]]
+	std::uint32_t runtime_crc32c (const char* data, std::size_t size, std::uint32_t crc) {
+#if defined(WRENPP_WITH_SSE42)
+		//if (_may_i_use_cpu_feature(_FEATURE_SSE4_2)) {
+		if (__builtin_cpu_supports("sse4.2")) {
+			return crc32c_hw(data, size, crc);
+		}
+		else
+#endif
+		{
+			return detail::crc32<detail::g_castagnoli_polynomial>(data, size, crc);
+		}
+	}
+} //namespace wren

src/pvt_config.h.in

@@ -27,3 +27,5 @@ static_assert(sizeof(void*) == ASM_PTR_SIZE, "Build system reports an unexpected
 
 static_assert(sizeof(void(*)(int)) == ASM_FUNC_PTR_SIZE, "Build system reports an unexpected function pointer size, please ensure assembly code is correct");
 #endif
+
+#mesondefine WRENPP_WITH_SSE42