fix: arm64 build

Makefile

@@ -20,7 +20,7 @@ OBJET = $(addprefix $(OBJDIR)/, \
 )
 
 CCAP      = 86
-CUDA      = /usr/local/cuda-11.8
+CUDA      = /usr/local/cuda
 CXX       = g++
 CXXCUDA   = /usr/bin/g++
 
@@ -30,19 +30,19 @@ NOSTR_BLOCKS_PER_GRID   = 512
 NOSTR_THREADS_PER_BLOCK = 256
 KEYS_PER_THREAD_BATCH   = 64
 
-CXXFLAGS  = -DWITHGPU -m64 -mssse3 -Wno-write-strings -O2 -I$(SRCDIR) -I$(CUDA)/include \
+CXXFLAGS  = -DWITHGPU -march=native -Wno-write-strings -O2 -I$(SRCDIR) -I$(CUDA)/include \
             -DNOSTR_BLOCKS_PER_GRID=$(NOSTR_BLOCKS_PER_GRID) \
             -DNOSTR_THREADS_PER_BLOCK=$(NOSTR_THREADS_PER_BLOCK) \
             -DKEYS_PER_THREAD_BATCH=$(KEYS_PER_THREAD_BATCH)
-LFLAGS    = /usr/lib/x86_64-linux-gnu/libgmp.so.10 -lpthread -L$(CUDA)/lib64 -lcudart -lcurand
+LFLAGS    = -lgmp -lpthread -L$(CUDA)/lib64 -lcudart -lcurand
 NVCC      = $(CUDA)/bin/nvcc
 
 #--------------------------------------------------------------------
 
 all: rummage
 
 $(OBJDIR)/GPU/GPURummage.o: $(SRCDIR)/GPU/GPURummage.cu
-	$(NVCC) -allow-unsupported-compiler --compile --compiler-options -fPIC -ccbin $(CXXCUDA) -m64 -O2 -I$(SRCDIR) -I$(CUDA)/include \
+	$(NVCC) -allow-unsupported-compiler --compile --compiler-options -fPIC -ccbin $(CXXCUDA) -O2 -I$(SRCDIR) -I$(CUDA)/include \
 	-DNOSTR_BLOCKS_PER_GRID=$(NOSTR_BLOCKS_PER_GRID) \
 	-DNOSTR_THREADS_PER_BLOCK=$(NOSTR_THREADS_PER_BLOCK) \
 	-DKEYS_PER_THREAD_BATCH=$(KEYS_PER_THREAD_BATCH) \

src/CPU/Int.cpp

@@ -18,7 +18,9 @@
 #include "Int.h"
 #include <string.h>
 #include <math.h>
+#if !defined(__aarch64__) && !defined(__arm64__) && !defined(_M_ARM64)
 #include <emmintrin.h>
+#endif
 
 #define MAX(x,y) (((x)>(y))?(x):(y))
 #define MIN(x,y) (((x)<(y))?(x):(y))

src/CPU/Int.h

@@ -215,6 +215,57 @@ class Int {
 
 #ifndef WIN64
 
+#if defined(__aarch64__) || defined(__arm64__) || defined(_M_ARM64)
+
+// ===== ARM64 IMPLEMENTATIONS =====
+
+static uint64_t inline _umul128(uint64_t a, uint64_t b, uint64_t *h) {
+  unsigned __int128 res = (unsigned __int128)a * (unsigned __int128)b;
+  *h = (uint64_t)(res >> 64);
+  return (uint64_t)res;
+}
+
+static int64_t inline _mul128(int64_t a, int64_t b, int64_t *h) {
+  __int128 res = (__int128)a * (__int128)b;
+  *h = (int64_t)(res >> 64);
+  return (int64_t)res;
+}
+
+static uint64_t inline _udiv128(uint64_t hi, uint64_t lo, uint64_t d, uint64_t *r) {
+  unsigned __int128 dividend = ((unsigned __int128)hi << 64) | lo;
+  *r = (uint64_t)(dividend % d);
+  return (uint64_t)(dividend / d);
+}
+
+static uint64_t inline __rdtsc() {
+  uint64_t val;
+  __asm__ volatile("mrs %0, cntvct_el0" : "=r"(val));
+  return val;
+}
+
+#define __shiftright128(a,b,n) ((a)>>(n))|((b)<<(64-(n)))
+#define __shiftleft128(a,b,n) ((b)<<(n))|((a)>>(64-(n)))
+
+static inline unsigned char _addcarry_u64(unsigned char c_in, uint64_t a, uint64_t b, uint64_t *out) {
+  unsigned __int128 sum = (unsigned __int128)a + (unsigned __int128)b + (unsigned __int128)c_in;
+  *out = (uint64_t)sum;
+  return (unsigned char)(sum >> 64);
+}
+
+static inline unsigned char _subborrow_u64(unsigned char b_in, uint64_t a, uint64_t b, uint64_t *out) {
+  unsigned __int128 diff = (unsigned __int128)a - (unsigned __int128)b - (unsigned __int128)b_in;
+  *out = (uint64_t)diff;
+  return (unsigned char)((diff >> 64) & 1);
+}
+
+#define _byteswap_uint64 __builtin_bswap64
+#define LZC(x) __builtin_clzll(x)
+#define TZC(x) __builtin_ctzll(x)
+
+#else
+
+// ===== X86-64 IMPLEMENTATIONS =====
+
 // Missing intrinsics
 static uint64_t inline _umul128(uint64_t a, uint64_t b, uint64_t *h) {
   uint64_t rhi;
@@ -257,6 +308,8 @@ static uint64_t inline __rdtsc() {
 #define LZC(x) __builtin_clzll(x)
 #define TZC(x) __builtin_ctzll(x)
 
+#endif  // ARM64 vs x86-64
+
 #else
 
 #include <intrin.h>

src/CPU/IntMod.cpp

@@ -16,7 +16,15 @@
 */
 
 #include "Int.h"
-#include <emmintrin.h>
+#if defined(__aarch64__) || defined(__arm64__) || defined(_M_ARM64)
+  #include <arm_neon.h>
+  typedef int64x2_t __m128i;
+  #define _mm_slli_epi64(a, count) vshlq_n_s64(a, count)
+  #define _mm_add_epi64(a, b) vaddq_s64(a, b)
+  #define _mm_sub_epi64(a, b) vsubq_s64(a, b)
+#else
+  #include <emmintrin.h>
+#endif
 #include <string.h>
 
 #define MAX(x,y) (((x)>(y))?(x):(y))
@@ -234,6 +242,9 @@ void Int::DivStep62(Int* u,Int* v,int64_t* eta,int* pos,int64_t* uu,int64_t* uv,
   _u.m128i_u64[1] = 0;
   _v.m128i_u64[0] = 0;
   _v.m128i_u64[1] = 1;
+#elif defined(__aarch64__) || defined(__arm64__) || defined(_M_ARM64)
+  _u = vsetq_lane_s64(1, vdupq_n_s64(0), 0);
+  _v = vsetq_lane_s64(1, vdupq_n_s64(0), 1);
 #else
   ((int64_t *)&_u)[0] = 1;
   ((int64_t *)&_u)[1] = 0;
@@ -271,6 +282,11 @@ void Int::DivStep62(Int* u,Int* v,int64_t* eta,int* pos,int64_t* uu,int64_t* uv,
   *uv = _u.m128i_u64[1];
   *vu = _v.m128i_u64[0];
   *vv = _v.m128i_u64[1];
+#elif defined(__aarch64__) || defined(__arm64__) || defined(_M_ARM64)
+  *uu = vgetq_lane_s64(_u, 0);
+  *uv = vgetq_lane_s64(_u, 1);
+  *vu = vgetq_lane_s64(_v, 0);
+  *vv = vgetq_lane_s64(_v, 1);
 #else
   *uu = ((int64_t *)&_u)[0];
   *uv = ((int64_t *)&_u)[1];