int test() {
int N = 5000 * SIMDBlockSize, gap;
__m128i * buffer = malloc(SIMDBlockSize * sizeof(uint32_t));
uint32_t * datain = malloc(N * sizeof(uint32_t));
uint32_t * backbuffer = malloc(SIMDBlockSize * sizeof(uint32_t));
for (gap = 1; gap <= 387420489; gap *= 3) {
int k;
printf(" gap = %u \n", gap);
for (k = 0; k < N; ++k)
datain[k] = k * gap;
for (k = 0; k * SIMDBlockSize < N; ++k) {
/*
First part works for general arrays (sorted or unsorted)
*/
int j;
/* we compute the bit width */
const uint32_t b = maxbits(datain + k * SIMDBlockSize);
/* we read 128 integers at "datain + k * SIMDBlockSize" and
write b 128-bit vectors at "buffer" */
simdpackwithoutmask(datain + k * SIMDBlockSize, buffer, b);
/* we read back b1 128-bit vectors at "buffer" and write 128 integers at backbuffer */
simdunpack(buffer, backbuffer, b);/* uncompressed */
for (j = 0; j < SIMDBlockSize; ++j) {
if (backbuffer[j] != datain[k * SIMDBlockSize + j]) {
printf("bug in simdpack\n");
return -2;
}
}
{
/*
next part assumes that the data is sorted (uses differential coding)
*/
uint32_t offset = 0;
/* we compute the bit width */
const uint32_t b1 = simdmaxbitsd1(offset,
datain + k * SIMDBlockSize);
/* we read 128 integers at "datain + k * SIMDBlockSize" and
write b1 128-bit vectors at "buffer" */
simdpackwithoutmaskd1(offset, datain + k * SIMDBlockSize, buffer,
b1);
/* we read back b1 128-bit vectors at "buffer" and write 128 integers at backbuffer */
simdunpackd1(offset, buffer, backbuffer, b1);
for (j = 0; j < SIMDBlockSize; ++j) {
if (backbuffer[j] != datain[k * SIMDBlockSize + j]) {
printf("bug in simdpack d1\n");
return -3;
}
}
offset = datain[k * SIMDBlockSize + SIMDBlockSize - 1];
}
}
}
free(buffer);
free(datain);
free(backbuffer);
printf("Code looks good.\n");
return 0;
}
/* Used below in more_sophisticated_demo ... */
size_t varying_bit_width_compress(uint32_t * datain, size_t length, uint8_t * buffer) {
uint8_t * initout;
size_t k;
if(length/SIMDBlockSize*SIMDBlockSize != length) {
printf("Data length should be a multiple of %i \n",SIMDBlockSize);
}
initout = buffer;
for(k = 0; k < length / SIMDBlockSize; ++k) {
uint32_t b = maxbits(datain);
*buffer++ = b;
simdpackwithoutmask(datain, (__m128i *)buffer, b);
datain += SIMDBlockSize;
buffer += b * sizeof(__m128i);
}
return buffer - initout;
}
//---------------- FastPFor functions ---------------------
unsigned char *simdpackwn(uint32_t *in, uint32_t n, uint32_t b, uint32_t *out) {//checkifdivisibleby(n, 128); const uint32_t * const initout(out); //while(needPaddingTo128Bits(out)) *out++ = 123456;
uint32_t *in_;
for(in_ = in + n; in + 128 <= in_; in += 128, out += 4 * b) simdpackwithoutmask(in, (__m128i *)out, b);
return (unsigned char *)out;
}