endif
# Verus hash specific library - optimized
-crypto_libverus_crypto_a_CPPFLAGS = -O3 -Wint-conversion -march=x86-64 -mpclmul -msse4 -msse4.1 -msse4.2 -mssse3 -mavx -maes -g -funroll-loops -fomit-frame-pointer -fPIC $(AM_CPPFLAGS)
-crypto_libverus_crypto_a_CXXFLAGS = -O3 -Wint-conversion -march=x86-64 -mpclmul -msse4 -msse4.1 -msse4.2 -mssse3 -mavx -maes -g -funroll-loops -fomit-frame-pointer -fPIC $(AM_CXXFLAGS)
+if ARCH_ARM
+crypto_libverus_crypto_a_CPPFLAGS = -O3 -Wint-conversion -g -funroll-loops -fomit-frame-pointer -fPIC $(AM_CPPFLAGS)
+crypto_libverus_crypto_a_CXXFLAGS = -O3 -Wint-conversion -g -funroll-loops -fomit-frame-pointer -fPIC $(AM_CXXFLAGS)
+else
+crypto_libverus_crypto_a_CPPFLAGS = -O3 -Wint-conversion -mpclmul -msse4 -msse4.1 -msse4.2 -mssse3 -mavx -maes -g -funroll-loops -fomit-frame-pointer -fPIC $(AM_CPPFLAGS)
+crypto_libverus_crypto_a_CXXFLAGS = -O3 -Wint-conversion -mpclmul -msse4 -msse4.1 -msse4.2 -mssse3 -mavx -maes -g -funroll-loops -fomit-frame-pointer -fPIC $(AM_CXXFLAGS)
+endif
+
crypto_libverus_crypto_a_SOURCES = \
crypto/haraka.h \
crypto/haraka.c \
crypto/verus_clhash.cpp
# Verus hash specific library - portable
+if ARCH_ARM
+crypto_libverus_portable_crypto_a_CPPFLAGS = -O3 -Wint-conversion -g -funroll-loops -fomit-frame-pointer -fPIC $(AM_CPPFLAGS)
+crypto_libverus_portable_crypto_a_CXXFLAGS = -O3 -Wint-conversion -g -funroll-loops -fomit-frame-pointer -fPIC $(AM_CXXFLAGS)
+else
crypto_libverus_portable_crypto_a_CPPFLAGS = -O3 -Wint-conversion -march=x86-64 -g -funroll-loops -fomit-frame-pointer -fPIC $(AM_CPPFLAGS)
crypto_libverus_portable_crypto_a_CXXFLAGS = -O3 -Wint-conversion -march=x86-64 -g -funroll-loops -fomit-frame-pointer -fPIC $(AM_CXXFLAGS)
+endif
+
crypto_libverus_portable_crypto_a_SOURCES = \
crypto/haraka_portable.h \
crypto/haraka_portable.c \
#include <assert.h>
#include <string.h>
-#ifndef _WIN32
-#include <x86intrin.h>
-#else
+#ifdef _WIN32
+#pragma warning (disable : 4146)
#include <intrin.h>
#endif // !WIN32
+#endif
+int __cpuverusoptimized = 0x80;
+
+#if defined(__arm__) || defined(__aarch64__)
+#include "crypto/SSE2NEON.h"
+#else
+#include <x86intrin.h>
+#endif
#ifdef _WIN32
#define posix_memalign(p, a, s) (((*(p)) = _aligned_malloc((s), (a))), *(p) ?0 :errno)
// attempt to workaround horrible mingw/gcc destructor bug on Windows and Mac, which passes garbage in the this pointer
// we use the opportunity of control here to clean up all of our tls variables. we could keep a list, but this is a safe,
// functional hack
+
+#if defined(__arm__) || defined(__aarch64__) //intrinsics not defined in SSE2NEON.h
+ static inline __attribute__((always_inline)) __m128i _mm_set_epi64x(uint64_t hi, uint64_t lo)
+ {
+ __m128i result;
+ ((uint64_t *)&result)[0] = lo;
+ ((uint64_t *)&result)[1] = hi;
+ return result;
+ }
+
+static inline __attribute__((always_inline)) __m128i _mm_mulhrs_epi16(__m128i _a, __m128i _b)
+{
+ int16_t result[8];
+ int16_t *a = (int16_t*)&_a, *b = (int16_t*)&_b;
+ for (int i = 0; i < 8; i++)
+ {
+ result[i] = (int16_t)((((int32_t)(a[i]) * (int32_t)(b[i])) + 0x4000) >> 15);
+ }
+
+ return *(__m128i *)result;
+}
+
+__m128i _mm_cvtsi64_si128(uint64_t lo)
+{
+ __m128i result;
+ ((uint64_t *)&result)[0] = lo;
+ ((uint64_t *)&result)[1] = 0;
+ return result;
+}
+__m128i _mm_aesenc_si128 (__m128i a, __m128i RoundKey)
+{
+ uint8x16_t a1; memcpy(&a1,&a,16);
+ uint8x16_t b1; memcpy(&b1,&RoundKey,16);
+ uint8x16_t c; //FIXME NEEDS -maes compile flags in ARM = vaesmcq_u8(vaeseq_u8(a1, (uint8x16_t){})) ^ b1;
+ __m128i d; memcpy(&d,&c,16);
+ return d;
+}
+__m128i _mm_clmulepi64_si128(const __m128i a, const __m128i b, int imm)
+{
+ __m128i result;
+ uint64x2_t a1 ; memcpy(&a1,&a,16);
+ uint64x2_t b1 ; memcpy(&b1,&b,16);
+ result = a; //FIXME NEEDS -maes compile flags in ARM (__m128i)vmull_p64(vgetq_lane_u64(a1, 1), vgetq_lane_u64(b1,0));
+
+ return result;
+
+}
+
+__m128i _mm_setr_epi8(u_char c0, u_char c1, u_char c2, u_char c3, u_char c4, u_char c5, u_char c6, u_char c7, u_char c8, u_char c9, u_char c10, u_char c11, u_char c12, u_char c13, u_char c14, u_char c15)
+{
+ __m128i result;
+ ((uint8_t *)&result)[0] = c0;
+ ((uint8_t *)&result)[1] = c1;
+ ((uint8_t *)&result)[2] = c2;
+ ((uint8_t *)&result)[3] = c3;
+ ((uint8_t *)&result)[4] = c4;
+ ((uint8_t *)&result)[5] = c5;
+ ((uint8_t *)&result)[6] = c6;
+ ((uint8_t *)&result)[7] = c7;
+ ((uint8_t *)&result)[8] = c8;
+ ((uint8_t *)&result)[9] = c9;
+ ((uint8_t *)&result)[10] = c10;
+ ((uint8_t *)&result)[11] = c11;
+ ((uint8_t *)&result)[12] = c12;
+ ((uint8_t *)&result)[13] = c13;
+ ((uint8_t *)&result)[14] = c14;
+ ((uint8_t *)&result)[15] = c15;
+ return result;
+}
+__m128i _mm_shuffle_epi8(__m128i a, __m128i b)
+{
+ __m128i result;
+ for (int i = 0; i < 16; i++)
+ {
+ if (((uint8_t *)&b)[i] & 0x80)
+ {
+ ((uint8_t *)&result)[i] = 0;
+ }
+ else
+ {
+ ((uint8_t *)&result)[i] = ((uint8_t *)&a)[((uint8_t *)&b)[i] & 0xf];
+ }
+ }
+ return result;
+}
+ int64_t _mm_cvtsi128_si64(__m128i a)
+{
+ return ((int64_t *)&a)[0];
+}
+__m128i _mm_loadl_epi64(__m128i *a)
+{
+ __m128i b = {0}; ((uint64_t*)&b)[0] = ((uint64_t*)a)[0];
+ return b;
+}
+#endif
+
thread_specific_ptr::~thread_specific_ptr() {
if (verusclhasher_key.ptr)
{
void *alloc_aligned_buffer(uint64_t bufSize)
{
void *answer = NULL;
- if (posix_memalign(&answer, sizeof(__m256i), bufSize))
+ if (posix_memalign(&answer, sizeof(__m128i) * 2, bufSize))
{
return NULL;
}
projects / VerusCoin.git / commitdiff