#ifndef BLAKE3_IMPL_H #define BLAKE3_IMPL_H #include <assert.h> #include <stdbool.h> #include <stddef.h> #include <stdint.h> #include <string.h> #include "llvm-c/blake3.h" // For \p LLVM_LIBRARY_VISIBILITY #include "llvm/Support/Compiler.h" #include "llvm_blake3_prefix.h" // internal flags enum blake3_flags { … }; // This C implementation tries to support recent versions of GCC, Clang, and // MSVC. #if defined(_MSC_VER) #define INLINE … #else #define INLINE … #endif #if defined(__x86_64__) || defined(_M_X64) #define IS_X86 #define IS_X86_64 #endif #if defined(__i386__) || defined(_M_IX86) #define IS_X86 #define IS_X86_32 #endif #if defined(__aarch64__) || defined(_M_ARM64) #define IS_AARCH64 #endif #if defined(IS_X86) #if defined(_MSC_VER) #include <intrin.h> #endif #include <immintrin.h> #endif #if !defined(BLAKE3_USE_NEON) // If BLAKE3_USE_NEON not manually set, autodetect based on // AArch64ness and endianness. #if defined(IS_AARCH64) && !defined(__ARM_BIG_ENDIAN) #define BLAKE3_USE_NEON … #else #define BLAKE3_USE_NEON … #endif #endif #if defined(IS_X86) #define MAX_SIMD_DEGREE … #elif BLAKE3_USE_NEON == 1 #define MAX_SIMD_DEGREE … #else #define MAX_SIMD_DEGREE … #endif // There are some places where we want a static size that's equal to the // MAX_SIMD_DEGREE, but also at least 2. #define MAX_SIMD_DEGREE_OR_2 … static const uint32_t IV[8] = …; static const uint8_t MSG_SCHEDULE[7][16] = …; /* Find index of the highest set bit */ /* x is assumed to be nonzero. */ static unsigned int highest_one(uint64_t x) { … } // Count the number of 1 bits. INLINE unsigned int popcnt(uint64_t x) { … } // Largest power of two less than or equal to x. As a special case, returns 1 // when x is 0. INLINE uint64_t round_down_to_power_of_2(uint64_t x) { … } INLINE uint32_t counter_low(uint64_t counter) { … } INLINE uint32_t counter_high(uint64_t counter) { … } INLINE uint32_t load32(const void *src) { … } INLINE void load_key_words(const uint8_t key[BLAKE3_KEY_LEN], uint32_t key_words[8]) { … } INLINE void store32(void *dst, uint32_t w) { … } INLINE void store_cv_words(uint8_t bytes_out[32], uint32_t cv_words[8]) { … } LLVM_LIBRARY_VISIBILITY void blake3_compress_in_place(uint32_t cv[8], const uint8_t block[BLAKE3_BLOCK_LEN], uint8_t block_len, uint64_t counter, uint8_t flags); LLVM_LIBRARY_VISIBILITY void blake3_compress_xof(const uint32_t cv[8], const uint8_t block[BLAKE3_BLOCK_LEN], uint8_t block_len, uint64_t counter, uint8_t flags, uint8_t out[64]); LLVM_LIBRARY_VISIBILITY void blake3_hash_many(const uint8_t *const *inputs, size_t num_inputs, size_t blocks, const uint32_t key[8], uint64_t counter, bool increment_counter, uint8_t flags, uint8_t flags_start, uint8_t flags_end, uint8_t *out); LLVM_LIBRARY_VISIBILITY size_t blake3_simd_degree(void); // Declarations for implementation-specific functions. LLVM_LIBRARY_VISIBILITY void blake3_compress_in_place_portable(uint32_t cv[8], const uint8_t block[BLAKE3_BLOCK_LEN], uint8_t block_len, uint64_t counter, uint8_t flags); LLVM_LIBRARY_VISIBILITY void blake3_compress_xof_portable(const uint32_t cv[8], const uint8_t block[BLAKE3_BLOCK_LEN], uint8_t block_len, uint64_t counter, uint8_t flags, uint8_t out[64]); LLVM_LIBRARY_VISIBILITY void blake3_hash_many_portable(const uint8_t *const *inputs, size_t num_inputs, size_t blocks, const uint32_t key[8], uint64_t counter, bool increment_counter, uint8_t flags, uint8_t flags_start, uint8_t flags_end, uint8_t *out); #if defined(IS_X86) #if !defined(BLAKE3_NO_SSE2) LLVM_LIBRARY_VISIBILITY void blake3_compress_in_place_sse2(uint32_t cv[8], const uint8_t block[BLAKE3_BLOCK_LEN], uint8_t block_len, uint64_t counter, uint8_t flags); LLVM_LIBRARY_VISIBILITY void blake3_compress_xof_sse2(const uint32_t cv[8], const uint8_t block[BLAKE3_BLOCK_LEN], uint8_t block_len, uint64_t counter, uint8_t flags, uint8_t out[64]); LLVM_LIBRARY_VISIBILITY void blake3_hash_many_sse2(const uint8_t *const *inputs, size_t num_inputs, size_t blocks, const uint32_t key[8], uint64_t counter, bool increment_counter, uint8_t flags, uint8_t flags_start, uint8_t flags_end, uint8_t *out); #endif #if !defined(BLAKE3_NO_SSE41) LLVM_LIBRARY_VISIBILITY void blake3_compress_in_place_sse41(uint32_t cv[8], const uint8_t block[BLAKE3_BLOCK_LEN], uint8_t block_len, uint64_t counter, uint8_t flags); LLVM_LIBRARY_VISIBILITY void blake3_compress_xof_sse41(const uint32_t cv[8], const uint8_t block[BLAKE3_BLOCK_LEN], uint8_t block_len, uint64_t counter, uint8_t flags, uint8_t out[64]); LLVM_LIBRARY_VISIBILITY void blake3_hash_many_sse41(const uint8_t *const *inputs, size_t num_inputs, size_t blocks, const uint32_t key[8], uint64_t counter, bool increment_counter, uint8_t flags, uint8_t flags_start, uint8_t flags_end, uint8_t *out); #endif #if !defined(BLAKE3_NO_AVX2) LLVM_LIBRARY_VISIBILITY void blake3_hash_many_avx2(const uint8_t *const *inputs, size_t num_inputs, size_t blocks, const uint32_t key[8], uint64_t counter, bool increment_counter, uint8_t flags, uint8_t flags_start, uint8_t flags_end, uint8_t *out); #endif #if !defined(BLAKE3_NO_AVX512) LLVM_LIBRARY_VISIBILITY void blake3_compress_in_place_avx512(uint32_t cv[8], const uint8_t block[BLAKE3_BLOCK_LEN], uint8_t block_len, uint64_t counter, uint8_t flags); LLVM_LIBRARY_VISIBILITY void blake3_compress_xof_avx512(const uint32_t cv[8], const uint8_t block[BLAKE3_BLOCK_LEN], uint8_t block_len, uint64_t counter, uint8_t flags, uint8_t out[64]); LLVM_LIBRARY_VISIBILITY void blake3_hash_many_avx512(const uint8_t *const *inputs, size_t num_inputs, size_t blocks, const uint32_t key[8], uint64_t counter, bool increment_counter, uint8_t flags, uint8_t flags_start, uint8_t flags_end, uint8_t *out); #endif #endif #if BLAKE3_USE_NEON == 1 LLVM_LIBRARY_VISIBILITY void blake3_hash_many_neon(const uint8_t *const *inputs, size_t num_inputs, size_t blocks, const uint32_t key[8], uint64_t counter, bool increment_counter, uint8_t flags, uint8_t flags_start, uint8_t flags_end, uint8_t *out); #endif #endif /* BLAKE3_IMPL_H */
Codebase Browser
llvm