#define _GNU_SOURCE
#include <stdio.h>
#include <stdlib.h>
#include <sys/mman.h>
#include <stdint.h>
#include <string.h>
#include <omp.h>
#include <x86intrin.h>

#define bind(f, arg) ({typeof(f(arg)) new(void) { return f(arg);} new;})
#define bind2(f, arg, arg2) ({typeof(f(arg, arg2)) new(void) { return f(arg, arg2);} new;})
#define bind3(f, arg, arg2, arg3) ({typeof(f(arg, arg2, arg3)) new(void) { return f(arg, arg2, arg3);} new;})

static inline void * alloc(uint64_t size, uint64_t flags) {
  return mmap(NULL, size, PROT_WRITE | PROT_READ, MAP_PRIVATE | MAP_ANONYMOUS | flags, 0, 0);
}

typedef void*(*vpvf_t)(void);

#define GB (1024ul*1024*1024)
#define KB (1024)

void bench_memcpy(uint64_t total, uint64_t pass, vpvf_t f) {
  uint64_t i = (total/pass);
  double start = omp_get_wtime();
  do {f();} while(--i);
  double time = omp_get_wtime() - start;
  fprintf(stderr, "%fGB/s\n", ((total * 2)/GB)/time);
}

void * memcpy_sphkavx(__m256i * to, __m256i * from, uint64_t n) {
  typeof(from) from_end = ((void *)from + n), ret = to;
  do {
    _mm256_store_si256((to + 0), _mm256_load_si256(from + 0));
    _mm256_store_si256((to + 1), _mm256_load_si256(from + 1));
    _mm256_store_si256((to + 2), _mm256_load_si256(from + 2));
    _mm256_store_si256((to + 3), _mm256_load_si256(from + 3));
  } while((to += 4), (from += 4) != from_end);
  return ret;
}

void * memcpy_sphk(__m128i * to, __m128i * from, uint64_t n) {
  typeof(from) from_end = ((void *)from + n), ret = to;
  do {
    _mm_stream_si128((to + 0), _mm_stream_load_si128(from + 0));
    _mm_stream_si128((to + 1), _mm_stream_load_si128(from + 1));
    _mm_stream_si128((to + 2), _mm_stream_load_si128(from + 2));
    _mm_stream_si128((to + 3), _mm_stream_load_si128(from + 3));
  } while((to += 4), (from += 4) != from_end);
  return ret;
}


int main(void) {
  vpvf_t alloc1gb = bind2(alloc, GB, MAP_POPULATE);
  void * from = alloc1gb(), * to = alloc1gb();
  uint64_t block_size = KB*KB*8, total = GB*2;
  do {
    fprintf(stderr, "\nBLOCK SIZE: %luKB\n", block_size/KB);
    fprintf(stderr, "SPHK:\n");
    if(block_size < (8*KB*KB)) {
      bench_memcpy(total, block_size, bind3(memcpy_sphkavx, to, from, block_size));
      bench_memcpy(total, block_size, bind3(memcpy_sphkavx, to, from, block_size));
    } else {
      bench_memcpy(total, block_size, bind3(memcpy_sphk, to, from, block_size));
      bench_memcpy(total, block_size, bind3(memcpy_sphk, to, from, block_size));
    }
    fprintf(stderr, "memcpy:\n");
    bench_memcpy(total, block_size, bind3(memcpy, to, from, block_size));
    bench_memcpy(total, block_size, bind3(memcpy, to, from, block_size));
  } while((block_size *= 2) != (GB*2));
}