예제 #1
0
void
TEST (void)
{
  int i, sign;
  UNION_TYPE (AVX512F_LEN, i_q) res1, res2, res3;
  UNION_TYPE (128, i_q) src;
  MASK_TYPE mask = SIZE | 123;
  long long res_ref[SIZE];

  sign = -1;
  for (i = 0; i < 2; i++)
    {
      src.a[i] = 34 * i * sign;
      sign = sign * -1;
    }
  for (i = 0; i < SIZE; i++)
    res2.a[i] = DEFAULT_VALUE;

  res1.x = INTRINSIC (_broadcast_i64x2) (src.x);
  res2.x = INTRINSIC (_mask_broadcast_i64x2) (res2.x, mask, src.x);
  res3.x = INTRINSIC (_maskz_broadcast_i64x2) (mask, src.x);

  CALC (res_ref, src.a);

  if (UNION_CHECK (AVX512F_LEN, i_q) (res1, res_ref))
    abort ();

  MASK_MERGE (i_q) (res_ref, mask, SIZE);
  if (UNION_CHECK (AVX512F_LEN, i_q) (res2, res_ref))
    abort ();

  MASK_ZERO (i_q) (res_ref, mask, SIZE);
  if (UNION_CHECK (AVX512F_LEN, i_q) (res3, res_ref))
    abort ();
}
예제 #2
0
static void
TEST (void)
{
  UNION_TYPE (AVX512F_LEN, d) s;
  UNION_TYPE (AVX512F_LEN_HALF, i_d) res1, res2, res3;
  MASK_TYPE mask = MASK_VALUE;
  unsigned res_ref[DST_SIZE] = { 0 };
  int i;

  for (i = 0; i < SRC_SIZE; i++)
    {
      s.a[i] = 123.456 * (i + 2000);
    }

  for (i = 0; i < DST_SIZE; i++)
    res2.a[i] = DEFAULT_VALUE;

  res1.x = INTRINSIC (_cvtpd_epu32) (s.x);
  res2.x = INTRINSIC (_mask_cvtpd_epu32) (res2.x, mask, s.x);
  res3.x = INTRINSIC (_maskz_cvtpd_epu32) (mask, s.x);

  CALC (s.a, res_ref);

  if (UNION_CHECK (AVX512F_LEN_HALF, i_d) (res1, res_ref))
    abort ();

  MASK_MERGE (i_d) (res_ref, mask, SRC_SIZE);
  if (UNION_CHECK (AVX512F_LEN_HALF, i_d) (res2, res_ref))
    abort ();

  MASK_ZERO (i_d) (res_ref, mask, SRC_SIZE);
  if (UNION_CHECK (AVX512F_LEN_HALF, i_d) (res3, res_ref))
    abort ();
}
예제 #3
0
void
TEST (void)
{
  UNION_TYPE (AVX512F_LEN_HALF, i_d) s;
  UNION_TYPE (AVX512F_LEN, d) res1, res2, res3;
  MASK_TYPE mask = MASK_VALUE;
  double res_ref[DST_SIZE];
  int i, sign = 1;

  for (i = 0; i < SRC_SIZE; i++)
    {
      s.a[i] = 123456 * (i + 2000) * sign;
      sign = -sign;
    }

  for (i = 0; i < DST_SIZE; i++)
    res2.a[i] = DEFAULT_VALUE;

  res1.x = INTRINSIC (_cvtepi32_pd) (s.x);
  res2.x = INTRINSIC (_mask_cvtepi32_pd) (res2.x, mask, s.x);
  res3.x = INTRINSIC (_maskz_cvtepi32_pd) (mask, s.x);

  CALC (s.a, res_ref);

  if (UNION_CHECK (AVX512F_LEN, d) (res1, res_ref))
    abort ();

  MASK_MERGE (d) (res_ref, mask, DST_SIZE);
  if (UNION_CHECK (AVX512F_LEN, d) (res2, res_ref))
    abort ();

  MASK_ZERO (d) (res_ref, mask, DST_SIZE);
  if (UNION_CHECK (AVX512F_LEN, d) (res3, res_ref))
    abort ();
}
예제 #4
0
void
TEST (void)
{
  int i, sign;
  UNION_TYPE (AVX512F_LEN, d) res1, res2, res3;
  UNION_TYPE (128, d) src;
  MASK_TYPE mask = MASK_VALUE;
  double res_ref[SIZE];

  sign = -1;
  for (i = 0; i < 2; i++)
    {
      src.a[i] = 1.5 + 34.67 * i * sign;
      sign = sign * -1;
    }
  for (i = 0; i < SIZE; i++)
    res2.a[i] = DEFAULT_VALUE;

  res1.x = INTRINSIC (_broadcastsd_pd) (src.x);
  res2.x = INTRINSIC (_mask_broadcastsd_pd) (res2.x, mask, src.x);
  res3.x = INTRINSIC (_maskz_broadcastsd_pd) (mask, src.x);

  CALC (res_ref, src.a);

  if (UNION_CHECK (AVX512F_LEN, d) (res1, res_ref))
    abort ();

  MASK_MERGE (d) (res_ref, mask, SIZE);
  if (UNION_CHECK (AVX512F_LEN, d) (res2, res_ref))
    abort ();

  MASK_ZERO (d) (res_ref, mask, SIZE);
  if (UNION_CHECK (AVX512F_LEN, d) (res3, res_ref))
    abort ();
}
예제 #5
0
static void
TEST (void)
{
  UNION_TYPE (AVX512F_LEN, i_d) s1, s2;
  UNION_TYPE (AVX512F_LEN, i_q) res1, res2, res3;
  MASK_TYPE mask = MASK_VALUE;
  unsigned long long res_ref[DST_SIZE];
  int i, sign = 1;

  for (i = 0; i < SRC_SIZE; i++)
    {
      s1.a[i] = i * 20;
      s2.a[i] = i + 20;
    }
  for (i = 0; i < DST_SIZE; i++)
    res2.a[i] = DEFAULT_VALUE;

  CALC (s1.a, s2.a, res_ref);

  res1.x = INTRINSIC (_mul_epu32) (s1.x, s2.x);
  res2.x = INTRINSIC (_mask_mul_epu32) (res2.x, mask, s1.x, s2.x);
  res3.x = INTRINSIC (_maskz_mul_epu32) (mask, s1.x, s2.x);

  if (UNION_CHECK (AVX512F_LEN, i_q) (res1, res_ref))
    abort ();

  MASK_MERGE (i_q) (res_ref, mask, DST_SIZE);
  if (UNION_CHECK (AVX512F_LEN, i_q) (res2, res_ref))
    abort ();

  MASK_ZERO (i_q) (res_ref, mask, DST_SIZE);
  if (UNION_CHECK (AVX512F_LEN, i_q) (res3, res_ref))
    abort ();
}
예제 #6
0
void
TEST (void)
{
  UNION_TYPE (AVX512F_LEN, d) s;
  UNION_TYPE (AVX512F_LEN, i_q) res1, res2, res3;
  MASK_TYPE mask = MASK_VALUE;
  unsigned long long res_ref[SIZE];
  int i;

  for (i = 0; i < SIZE; i++)
    {
      s.a[i] = 123.456 * (i + 2000);
      res2.a[i] = DEFAULT_VALUE;
    }

  res1.x = INTRINSIC (_cvttpd_epu64) (s.x);
  res2.x = INTRINSIC (_mask_cvttpd_epu64) (res2.x, mask, s.x);
  res3.x = INTRINSIC (_maskz_cvttpd_epu64) (mask, s.x);

  CALC (s.a, res_ref);

  if (UNION_CHECK (AVX512F_LEN, i_q) (res1, res_ref))
    abort ();

  MASK_MERGE (i_q) (res_ref, mask, SIZE);
  if (UNION_CHECK (AVX512F_LEN, i_q) (res2, res_ref))
    abort ();

  MASK_ZERO (i_q) (res_ref, mask, SIZE);
  if (UNION_CHECK (AVX512F_LEN, i_q) (res3, res_ref))
    abort ();
}
예제 #7
0
void
TEST (void)
{
  UNION_TYPE (AVX512F_LEN_HALF, i_w) s;
  UNION_TYPE (AVX512F_LEN, i_d) res1, res2, res3;
  MASK_TYPE mask = MASK_VALUE;
  int res_ref[SIZE];
  int i, sign = 1;

  for (i = 0; i < SIZE; i++)
    {
      s.a[i] = 2000 * i * sign;
      res2.a[i] = DEFAULT_VALUE;
      sign = -sign;
    }

  res1.x = INTRINSIC (_cvtepi16_epi32) (s.x);
  res2.x = INTRINSIC (_mask_cvtepi16_epi32) (res2.x, mask, s.x);
  res3.x = INTRINSIC (_maskz_cvtepi16_epi32) (mask, s.x);

  CALC (s.a, res_ref);

  if (UNION_CHECK (AVX512F_LEN, i_d) (res1, res_ref))
    abort ();

  MASK_MERGE (i_d) (res_ref, mask, SIZE);
  if (UNION_CHECK (AVX512F_LEN, i_d) (res2, res_ref))
    abort ();

  MASK_ZERO (i_d) (res_ref, mask, SIZE);
  if (UNION_CHECK (AVX512F_LEN, i_d) (res3, res_ref))
    abort ();
}
예제 #8
0
void static
TEST (void)
{
  int i, sign;
  UNION_TYPE (128, i_b) res1, res2, res3;
  UNION_TYPE (AVX512F_LEN, i_d) src;
  MASK_TYPE mask = MASK_VALUE;
  char res_ref[16];

  sign = -1;
  for (i = 0; i < SIZE; i++)
    {
      src.a[i] = 1 + 34 * i * sign;
      sign = sign * -1;
      res2.a[i] = DEFAULT_VALUE;
    }

  res1.x = INTRINSIC (_cvtepi32_epi8) (src.x);
  res2.x = INTRINSIC (_mask_cvtepi32_epi8) (res2.x, mask, src.x);
  res3.x = INTRINSIC (_maskz_cvtepi32_epi8) (mask, src.x);

  CALC (res_ref, src.a);

  if (UNION_CHECK (128, i_b) (res1, res_ref))
    abort ();

  MASK_MERGE (i_b) (res_ref, mask, SIZE);
  if (UNION_CHECK (128, i_b) (res2, res_ref))
    abort ();

  MASK_ZERO (i_b) (res_ref, mask, SIZE);
  if (UNION_CHECK (128, i_b) (res3, res_ref))
    abort ();
}
예제 #9
0
void
TEST (void)
{
  int i, sign;
  UNION_TYPE (AVX512F_LEN, i_w) res1, res2, res3;
  UNION_TYPE (128, i_w) src;
  MASK_TYPE mask = SIZE | 123;
  short res_ref[SIZE];

  sign = -1;
  for (i = 0; i < 8; i++)
    {
      src.a[i] = 1 + 3 * i * sign;
      sign = sign * -1;
    }
  for (i = 0; i < SIZE; i++)
    res2.a[i] = DEFAULT_VALUE;

  CALC (res_ref, src.a);

  if (AVX512F_LEN == 512)
    {
      res1.x = INTRINSIC (_broadcastw_epi16) (src.x);
      if (UNION_CHECK (AVX512F_LEN, i_w) (res1, res_ref))
	abort ();
    }

  res2.x = INTRINSIC (_mask_broadcastw_epi16) (res2.x, mask, src.x);
  res3.x = INTRINSIC (_maskz_broadcastw_epi16) (mask, src.x);

  MASK_MERGE (i_w) (res_ref, mask, SIZE);
  if (UNION_CHECK (AVX512F_LEN, i_w) (res2, res_ref))
    abort ();

  MASK_ZERO (i_w) (res_ref, mask, SIZE);
  if (UNION_CHECK (AVX512F_LEN, i_w) (res3, res_ref))
    abort ();

  CALC (res_ref, src.a);

  if (AVX512F_LEN == 512)
    {
      res1.x = INTRINSIC (_set1_epi16) (src.a[0]);
      if (UNION_CHECK (AVX512F_LEN, i_w) (res1, res_ref))
	abort ();
    }

  res2.x = INTRINSIC (_mask_set1_epi16) (res2.x, mask, src.a[0]);
  res3.x = INTRINSIC (_maskz_set1_epi16) (mask, src.a[0]);

  MASK_MERGE (i_w) (res_ref, mask, SIZE);
  if (UNION_CHECK (AVX512F_LEN, i_w) (res2, res_ref))
    abort ();

  MASK_ZERO (i_w) (res_ref, mask, SIZE);
  if (UNION_CHECK (AVX512F_LEN, i_w) (res3, res_ref))
    abort ();
}
예제 #10
0
void
TEST (void)
{
  int i, sign;
  UNION_TYPE (128, i_b) res1, res2, res3;
  char res4[16];
  UNION_TYPE (AVX512F_LEN, i_d) src;
  MASK_TYPE mask = MASK_VALUE;
  char res_ref[16];
  char res_ref2[16];

  sign = -1;
  for (i = 0; i < SIZE; i++)
    {
      src.a[i] = 1 + 34 * i * sign;
      sign = sign * -1;
      res2.a[i] = DEFAULT_VALUE;
      res4[i] = DEFAULT_VALUE;
    }

  for (i = SIZE; i < 16; i++)
    {
      res4[i] = DEFAULT_VALUE * 2;
      res_ref2[i] = DEFAULT_VALUE * 2;
    }

  res1.x = INTRINSIC (_cvtepi32_epi8) (src.x);
  res2.x = INTRINSIC (_mask_cvtepi32_epi8) (res2.x, mask, src.x);
  res3.x = INTRINSIC (_maskz_cvtepi32_epi8) (mask, src.x);

  CALC (res_ref, src.a, 0);

  if (UNION_CHECK (128, i_b) (res1, res_ref))
    abort ();

  MASK_MERGE (i_b) (res_ref, mask, SIZE);
  if (UNION_CHECK (128, i_b) (res2, res_ref))
    abort ();

  MASK_ZERO (i_b) (res_ref, mask, SIZE);
  if (UNION_CHECK (128, i_b) (res3, res_ref))
    abort ();

  INTRINSIC (_mask_cvtepi32_storeu_epi8) (res4, mask, src.x);
  CALC (res_ref2, src.a, 1);

  MASK_MERGE (i_b) (res_ref2, mask, SIZE);
  if (checkVc (res4, res_ref2, 16))
    abort ();
}
예제 #11
0
void
TEST (void)
{
  int i;
  UNION_TYPE (128, i_uw) res1, res2, res3;
  unsigned short res4[8];
  UNION_TYPE (AVX512F_LEN, i_uq) src;
  MASK_TYPE mask = MASK_VALUE;
  unsigned short res_ref[8];
  unsigned short res_ref2[8];

  for (i = 0; i < SIZE; i++)
    {
      src.a[i] = 1 + 34 * i;
      res2.a[i] = DEFAULT_VALUE;
      res4[i] = DEFAULT_VALUE;
    }

  for (i = SIZE; i < 8; i++)
    {
      res_ref2[i] = DEFAULT_VALUE * 2;
      res4[i] = DEFAULT_VALUE * 2;
    }

  res1.x = INTRINSIC (_cvtusepi64_epi16) (src.x);
  res2.x = INTRINSIC (_mask_cvtusepi64_epi16) (res2.x, mask, src.x);
  res3.x = INTRINSIC (_maskz_cvtusepi64_epi16) (mask, src.x);

  CALC (res_ref, src.a, 0);

  if (UNION_CHECK (128, i_uw) (res1, res_ref))
    abort ();

  MASK_MERGE (i_uw) (res_ref, mask, SIZE);
  if (UNION_CHECK (128, i_uw) (res2, res_ref))
    abort ();

  MASK_ZERO (i_uw) (res_ref, mask, SIZE);
  if (UNION_CHECK (128, i_uw) (res3, res_ref))
    abort ();

  INTRINSIC (_mask_cvtusepi64_storeu_epi16) (res4, mask, src.x);
  CALC (res_ref2, src.a, 1);

  MASK_MERGE (i_w) (res_ref2, mask, SIZE);
  if (checkVs (res4, res_ref2, 8))
    abort ();
}
예제 #12
0
void
TEST (void)
{
  UNION_TYPE (AVX512F_LEN, i_b) s1, s2, res1, res2 ,res3;
  MASK_TYPE mask = MASK_VALUE;
  char res_ref[SIZE];
  int i;

  for (i = 0; i < SIZE; i++)
    {
      s1.a[i] = i + 15;
      s2.a[i] = i + 14;
      res2.a[i] = DEFAULT_VALUE;
    }

  res1.x = INTRINSIC (_avg_epu8) (s1.x, s2.x);
  res2.x = INTRINSIC (_mask_avg_epu8) (res2.x, mask, s1.x, s2.x);
  res3.x = INTRINSIC (_maskz_avg_epu8) (mask, s1.x, s2.x);

  CALC (s1.a, s2.a, res_ref);

  if (UNION_CHECK (AVX512F_LEN, i_b) (res1, res_ref))
    abort ();

  MASK_MERGE (i_b) (res_ref, mask, SIZE);
  if (UNION_CHECK (AVX512F_LEN, i_b) (res2, res_ref))
    abort ();

  MASK_ZERO (i_b) (res_ref, mask, SIZE);
  if (UNION_CHECK (AVX512F_LEN, i_b) (res3, res_ref))
    abort ();
}
예제 #13
0
void static
TEST (void)
{
  UNION_TYPE (AVX512F_LEN, i_q) s1, s2, res1, res2, res3;
  MASK_TYPE mask = MASK_VALUE;
  long long res_ref[SIZE];
  int i, sign = 1;

  for (i = 0; i < SIZE; i++)
    {
      s1.a[i] = i * sign;
      s2.a[i] = (i + 20) * sign;
      sign = -sign;
      res2.a[i] = DEFAULT_VALUE;
    }

  res1.x = INTRINSIC (_andnot_epi64) (s1.x, s2.x);
  res2.x = INTRINSIC (_mask_andnot_epi64) (res2.x, mask, s1.x, s2.x);
  res3.x = INTRINSIC (_maskz_andnot_epi64) (mask, s1.x, s2.x);

  CALC (s1.a, s2.a, res_ref);

  if (UNION_CHECK (AVX512F_LEN, i_q) (res1, res_ref))
      abort ();

  MASK_MERGE (i_q) (res_ref, mask, SIZE);
  if (UNION_CHECK (AVX512F_LEN, i_q) (res2, res_ref))
    abort ();

  MASK_ZERO (i_q) (res_ref, mask, SIZE);
  if (UNION_CHECK (AVX512F_LEN, i_q) (res3, res_ref))
    abort ();
}
예제 #14
0
static void
TEST (void)
{
  UNION_TYPE (AVX512F_LEN, i_q) s, res1, res2, res3;
  long long res_ref[SIZE];
  MASK_TYPE mask = MASK_VALUE;
  int i;

  for (i = 0; i < SIZE; i++)
    {
      s.a[i] = 12345678 * (i % 5);
      res1.a[i] = DEFAULT_VALUE;
      res2.a[i] = DEFAULT_VALUE;
      res3.a[i] = DEFAULT_VALUE;
    }

  res1.x = INTRINSIC (_conflict_epi64) (s.x);
  res2.x = INTRINSIC (_mask_conflict_epi64) (res2.x, mask, s.x);
  res3.x = INTRINSIC (_maskz_conflict_epi64) (mask, s.x);

  CALC (s.a, res_ref);

  if (UNION_CHECK (AVX512F_LEN, i_q) (res1, res_ref))
    abort ();

  MASK_MERGE (i_q) (res_ref, mask, SIZE);
  if (UNION_CHECK (AVX512F_LEN, i_q) (res2, res_ref))
    abort ();

  MASK_ZERO (i_q) (res_ref, mask, SIZE);
  if (UNION_CHECK (AVX512F_LEN, i_q) (res3, res_ref))
    abort ();
}
예제 #15
0
static void
TEST (void)
{
  union128i_w s;
  UNION_TYPE (AVX512F_LEN, i_q) res1, res2, res3;
  MASK_TYPE mask = MASK_VALUE;
  long long int res_ref[SIZE];
  int i;

  for (i = 0; i < SIZE; i++)
    {
      s.a[i] = 2000 * i;
      res2.a[i] = DEFAULT_VALUE;
    }

  res1.x = INTRINSIC (_cvtepu16_epi64) (s.x);
  res2.x = INTRINSIC (_mask_cvtepu16_epi64) (res2.x, mask, s.x);
  res3.x = INTRINSIC (_maskz_cvtepu16_epi64) (mask, s.x);

  CALC (s.a, res_ref);

  if (UNION_CHECK (AVX512F_LEN, i_q) (res1, res_ref))
    abort ();

  MASK_MERGE (i_q) (res_ref, mask, SIZE);
  if (UNION_CHECK (AVX512F_LEN, i_q) (res2, res_ref))
    abort ();

  MASK_ZERO (i_q) (res_ref, mask, SIZE);
  if (UNION_CHECK (AVX512F_LEN, i_q) (res3, res_ref))
    abort ();
}
예제 #16
0
void
TEST (void)
{
  int i, sign;
  UNION_TYPE (AVX512F_LEN, i_b) res1, res2, res3, src1, src2;
  MASK_TYPE mask = MASK_VALUE;
  char res_ref[SIZE];

  sign = -1;
  for (i = 0; i < SIZE; i++)
    {
      src1.a[i] = 34 * i * sign;
      src1.a[i] = 179 * i;
      sign = sign * -1;
    }
  for (i = 0; i < SIZE; i++)
    res2.a[i] = DEFAULT_VALUE;

  res1.x = INTRINSIC (_unpackhi_epi8) (src1.x, src2.x);
  res2.x = INTRINSIC (_mask_unpackhi_epi8) (res2.x, mask, src1.x, src2.x);
  res3.x = INTRINSIC (_maskz_unpackhi_epi8) (mask, src1.x, src2.x);

  CALC (res_ref, src1.a, src2.a);

  if (UNION_CHECK (AVX512F_LEN, i_b) (res1, res_ref))
    abort ();

  MASK_MERGE (i_b) (res_ref, mask, SIZE);
  if (UNION_CHECK (AVX512F_LEN, i_b) (res2, res_ref))
    abort ();

  MASK_ZERO (i_b) (res_ref, mask, SIZE);
  if (UNION_CHECK (AVX512F_LEN, i_b) (res3, res_ref))
    abort ();
}
예제 #17
0
void
TEST (void)
{
  int i;
  UNION_TYPE (AVX512F_LEN, i_w) src1, src2, res1, res2, res3;
  MASK_TYPE mask = MASK_VALUE;
  unsigned short res_ref[SIZE];

  for (i = 0; i < SIZE; i++)
    {
      src1.a[i] = i * i;
      src2.a[i] = i + 20;
      res2.a[i] = DEFAULT_VALUE;
    }

  res1.x = INTRINSIC (_min_epu16) (src1.x, src2.x);
  res2.x = INTRINSIC (_mask_min_epu16) (res2.x, mask, src1.x, src2.x);
  res3.x = INTRINSIC (_maskz_min_epu16) (mask, src1.x, src2.x);

  CALC (src1.a, src2.a, res_ref);

  if (UNION_CHECK (AVX512F_LEN, i_w) (res1, res_ref))
    abort ();

  MASK_MERGE (i_w) (res_ref, mask, SIZE);
  if (UNION_CHECK (AVX512F_LEN, i_w) (res2, res_ref))
    abort ();

  MASK_ZERO (i_w) (res_ref, mask, SIZE);
  if (UNION_CHECK (AVX512F_LEN, i_w) (res3, res_ref))
    abort ();
}
예제 #18
0
void
TEST (void)
{
  int i;
  UNION_TYPE (AVX512F_LEN, i_q) s1, s2, res, ind;
  long long res_ref[SIZE];

  MASK_TYPE mask = MASK_VALUE;

  for (i = 0; i < SIZE; i++)
    {
      ind.a[i] = DEFAULT_VALUE;
      s1.a[i] = 34 * i + 1;
      s2.a[i] = 34 * i;

      res.a[i] = DEFAULT_VALUE;
    }

  CALC (res_ref, s1.a, ind.a, s2.a);

  res.x =
    INTRINSIC (_mask2_permutex2var_epi64) (s1.x, ind.x, mask, s2.x);

  MASK_MERGE (i_q) (res_ref, mask, SIZE);
  if (UNION_CHECK (AVX512F_LEN, i_q) (res, res_ref))
    abort ();
}
예제 #19
0
void
TEST (void)
{
  int i;
  UNION_TYPE (AVX512F_LEN, i_b) src1, src2;
  MASK_TYPE res_ref, res1, res2;
  MASK_TYPE mask = MASK_VALUE;

  for (i = 0; i < SIZE / 2; i++)
    {
      src1.a[i * 2] = i;
      src1.a[i * 2 + 1] = i * i;
      src2.a[i * 2] = 2 * i;
      src2.a[i * 2 + 1] = i * i;
    }

  res1 = INTRINSIC (_cmple_epi8_mask) (src1.x, src2.x);
  res2 = INTRINSIC (_mask_cmple_epi8_mask) (mask, src1.x, src2.x);

  CALC (&res_ref, src1.a, src2.a);

  if (res_ref != res1)
    abort ();

  res_ref &= mask;

  if (res_ref != res2)
    abort ();
}
예제 #20
0
void static
TEST (void)
{
  int i;
  UNION_TYPE (AVX512F_LEN, i_q) res1, res2, res3, src1, src2;
  MASK_TYPE mask = MASK_VALUE;
  unsigned long long res_ref[SIZE];

  for (i = 0; i < SIZE; i++)
    {
      src1.a[i] = 2 + 7 * i % 291;
      src2.a[i] = 1 + 17 * i % 71;
    }

  for (i = 0; i < SIZE; i++)
    res2.a[i] = DEFAULT_VALUE;

  res1.x = INTRINSIC (_srlv_epi64) (src1.x, src2.x);
  res2.x = INTRINSIC (_mask_srlv_epi64) (res2.x, mask, src1.x, src2.x);
  res3.x = INTRINSIC (_maskz_srlv_epi64) (mask, src1.x, src2.x);

  CALC (res_ref, src1.a, src2.a);

  if (UNION_CHECK (AVX512F_LEN, i_q) (res1, res_ref))
    abort ();

  MASK_MERGE (i_q) (res_ref, mask, SIZE);
  if (UNION_CHECK (AVX512F_LEN, i_q) (res2, res_ref))
    abort ();

  MASK_ZERO (i_q) (res_ref, mask, SIZE);
  if (UNION_CHECK (AVX512F_LEN, i_q) (res3, res_ref))
    abort ();
}
예제 #21
0
void
TEST (void)
{
  UNION_TYPE (AVX512F_LEN, d) s1, s2, res1, res2, res3;
  MASK_TYPE mask = MASK_VALUE;
  double res_ref[SIZE];
  int i;

  for (i = 0; i < SIZE; i++)
    {
      s1.a[i] = i * 123.2 + 32.6;
      s2.a[i] = i + 2.5;
      res2.a[i] = DEFAULT_VALUE;
    }

  res1.x = INTRINSIC (_unpacklo_pd) (s1.x, s2.x);
  res2.x = INTRINSIC (_mask_unpacklo_pd) (res2.x, mask, s1.x, s2.x);
  res3.x = INTRINSIC (_maskz_unpacklo_pd) (mask, s1.x, s2.x);

  CALC (s1.a, s2.a, res_ref);

  if (UNION_CHECK (AVX512F_LEN, d) (res1, res_ref))
    abort ();

  MASK_MERGE (d) (res_ref, mask, SIZE);
  if (UNION_CHECK (AVX512F_LEN, d) (res2, res_ref))
    abort ();

  MASK_ZERO (d) (res_ref, mask, SIZE);
  if (UNION_CHECK (AVX512F_LEN, d) (res3, res_ref))
    abort ();
}
예제 #22
0
void
TEST (void)
{
  UNION_TYPE (AVX512F_LEN, d) s1, s2, res1, res2, res3;
  MASK_TYPE mask = MASK_VALUE;
  double res_ref[SIZE];
  int i, sign = 1;

  for (i = 0; i < SIZE; i++)
    {
      s1.a[i] = 234.567 * i * sign;
      s2.a[i] = 100 * (i + 1);
      res2.a[i] = DEFAULT_VALUE;
      sign = -sign;
    }

  res1.x = INTRINSIC (_range_pd) (s1.x, s2.x, IMM);
  res2.x = INTRINSIC (_mask_range_pd) (res2.x, mask, s1.x, s2.x, IMM);
  res3.x = INTRINSIC (_maskz_range_pd) (mask, s1.x, s2.x, IMM);

  CALC (s1.a, s2.a, res_ref);

  if (UNION_CHECK (AVX512F_LEN, d) (res1, res_ref))
    abort ();

  MASK_MERGE (d) (res_ref, mask, SIZE);
  if (UNION_CHECK (AVX512F_LEN, d) (res2, res_ref))
    abort ();

  MASK_ZERO (d) (res_ref, mask, SIZE);
  if (UNION_CHECK (AVX512F_LEN, d) (res3, res_ref))
    abort ();
}
예제 #23
0
void static
TEST (void)
{
  UNION_TYPE (AVX512F_LEN, i_d) s1, res1, res2, res3;
  int res_ref[SIZE];
  int i, j, sign = 1;
  MASK_TYPE mask = MASK_VALUE;

  for (j = 0; j < SIZE; j++)
    {
      s1.a[j] = j * i * sign;
      res1.a[j] = DEFAULT_VALUE;
      res2.a[j] = DEFAULT_VALUE;
      res3.a[j] = DEFAULT_VALUE;
      sign = -sign;
    }

  res1.x = INTRINSIC (_shuffle_epi32) (s1.x, 0xec);
  res2.x = INTRINSIC (_mask_shuffle_epi32) (res2.x, mask, s1.x, 0xec);
  res3.x = INTRINSIC (_maskz_shuffle_epi32) (mask, s1.x, 0xec);

  CALC (s1.a, 0xec, res_ref);

  if (UNION_CHECK (AVX512F_LEN, i_d) (res1, res_ref))
    abort ();

  MASK_MERGE (i_d) (res_ref, mask, SIZE);
  if (UNION_CHECK (AVX512F_LEN, i_d) (res2, res_ref))
    abort ();

  MASK_ZERO (i_d) (res_ref, mask, SIZE);
  if (UNION_CHECK (AVX512F_LEN, i_d) (res3, res_ref))
    abort ();
}
예제 #24
0
void
TEST (void)
{
  int i;
  UNION_TYPE (AVX512F_LEN, i_w) res1, res2, res3, src1, src2;
  MASK_TYPE mask = MASK_VALUE;
  unsigned short res_ref[SIZE];

  for (i = 0; i < SIZE; i++)
    {
      src1.a[i] = 2 + 7 * i % 291;
      src2.a[i] = 3 + 11 * (i % 377) * i;
    }
  for (i = 0; i < SIZE; i++)
    res2.a[i] = DEFAULT_VALUE;

  res1.x = INTRINSIC (_subs_epu16) (src1.x, src2.x);
  res2.x = INTRINSIC (_mask_subs_epu16) (res2.x, mask, src1.x, src2.x);
  res3.x = INTRINSIC (_maskz_subs_epu16) (mask, src1.x, src2.x);

  CALC (res_ref, src1.a, src2.a);

  if (UNION_CHECK (AVX512F_LEN, i_w) (res1, res_ref))
    abort ();

  MASK_MERGE (i_w) (res_ref, mask, SIZE);
  if (UNION_CHECK (AVX512F_LEN, i_w) (res2, res_ref))
    abort ();

  MASK_ZERO (i_w) (res_ref, mask, SIZE);
  if (UNION_CHECK (AVX512F_LEN, i_w) (res3, res_ref))
    abort ();
}
예제 #25
0
void
TEST (void)
{
  int i, sign;
  UNION_TYPE (AVX512F_LEN, i_b) src1, src2, res1, res2, res3;
  MASK_TYPE mask = MASK_VALUE;
  char res_ref[SIZE];

  for (i = 0; i < SIZE; i++)
    {
      src1.a[i] =  i * sign;
      src2.a[i] = (i + 20) * sign;
      sign = -sign;
      res2.a[i] = DEFAULT_VALUE;
    }

  res1.x = INTRINSIC (_min_epi8) (src1.x, src2.x);
  res2.x = INTRINSIC (_mask_min_epi8) (res2.x, mask, src1.x, src2.x);
  res3.x = INTRINSIC (_maskz_min_epi8) (mask, src1.x, src2.x);

  CALC (src1.a, src2.a, res_ref);

  if (UNION_CHECK (AVX512F_LEN, i_b) (res1, res_ref))
    abort ();

  MASK_MERGE (i_b) (res_ref, mask, SIZE);
  if (UNION_CHECK (AVX512F_LEN, i_b) (res2, res_ref))
    abort ();

  MASK_ZERO (i_b) (res_ref, mask, SIZE);
  if (UNION_CHECK (AVX512F_LEN, i_b) (res3, res_ref))
    abort ();
}
예제 #26
0
void static
TEST (void)
{
  int i, sign;
  UNION_TYPE (AVX512F_LEN, d) res1, res2, res3, src1, src2;
  MASK_TYPE mask = MASK_VALUE;
  double res_ref[SIZE];

  sign = -1;
  for (i = 0; i < SIZE; i++)
    {
      src1.a[i] = 1.5 + 34.67 * i * sign;
      src2.a[i] = -22.17 * i * sign;
      sign = sign * -1;
    }
  for (i = 0; i < SIZE; i++)
    res2.a[i] = DEFAULT_VALUE;

  res1.x = INTRINSIC (_sub_pd) (src1.x, src2.x);
  res2.x = INTRINSIC (_mask_sub_pd) (res2.x, mask, src1.x, src2.x);
  res3.x = INTRINSIC (_maskz_sub_pd) (mask, src1.x, src2.x);

  CALC (res_ref, src1.a, src2.a);

  if (UNION_CHECK (AVX512F_LEN, d) (res1, res_ref))
    abort ();

  MASK_MERGE (d) (res_ref, mask, SIZE);
  if (UNION_CHECK (AVX512F_LEN, d) (res2, res_ref))
    abort ();

  MASK_ZERO (d) (res_ref, mask, SIZE);
  if (UNION_CHECK (AVX512F_LEN, d) (res3, res_ref))
    abort ();
}
예제 #27
0
void
TEST (void)
{
  int i;
  UNION_TYPE (AVX512F_LEN, i_q) src1, src2;
  MASK_TYPE res1, res_ref, res2;
  res1 = 0;

  for (i = 0; i < SIZE / 2; i++)
    {
      src1.a[i * 2] = i;
      src1.a[i * 2 + 1] = i * i;
      src2.a[i * 2] = 2 * i;
      src2.a[i * 2 + 1] = i * i;
    }

  res1 = INTRINSIC (_cmpneq_epu64_mask) (src1.x, src2.x);
  res2 = INTRINSIC (_mask_cmpneq_epu64_mask) (MASK_VALUE, src1.x, src2.x);

  CALC (&res_ref, src1.a, src2.a);

  if (res1 != res_ref)
    abort ();

  res_ref &= MASK_VALUE;

  if (res2 != res_ref)
    abort ();
}
예제 #28
0
void
TEST (void)
{
  int i, j;
  UNION_TYPE (AVX512F_LEN, i_w) s1, s2, res, ind;
  short res_ref[SIZE];

  MASK_TYPE mask = MASK_VALUE;

  for (i = 0; i < NUM; i++)
    {
      for (j = 0; j < SIZE; j++)
	{
	  ind.a[j] = DEFAULT_VALUE;
	  s1.a[j] = i * 2 * j + 1;
	  s2.a[j] = i * 2 * j;

	  res.a[j] = DEFAULT_VALUE;
	}

      CALC (res_ref, s1.a, ind.a, s2.a);

      res.x =
	INTRINSIC (_mask2_permutex2var_epi16) (s1.x, ind.x, mask,
					       s2.x);

      MASK_MERGE (i_w) (res_ref, mask, SIZE);
      if (UNION_CHECK (AVX512F_LEN, i_w) (res, res_ref))
	abort ();
    }
}
예제 #29
0
void
TEST (void)
{
  UNION_TYPE (AVX512F_LEN, i_w) s1, s2, res1, res2, res3;
  MASK_TYPE mask = MASK_VALUE;
  short res_ref[SIZE];
  int i, sign = 1;

  for (i = 0; i < SIZE; i++)
    {
      s1.a[i] = i * sign;
      s2.a[i] = i >> 2;
      res2.a[i] = DEFAULT_VALUE;
      sign = -sign;
    }

  res1.x = INTRINSIC (_srav_epi16) (s1.x, s2.x);
  res2.x = INTRINSIC (_mask_srav_epi16) (res2.x, mask, s1.x, s2.x);
  res3.x = INTRINSIC (_maskz_srav_epi16) (mask, s1.x, s2.x);

  CALC (s1.a, s2.a, res_ref);

  if (UNION_CHECK (AVX512F_LEN, i_w) (res1, res_ref))
    abort ();

  MASK_MERGE (i_w) (res_ref, mask, SIZE);
  if (UNION_CHECK (AVX512F_LEN, i_w) (res2, res_ref))
    abort ();

  MASK_ZERO (i_w) (res_ref, mask, SIZE);
  if (UNION_CHECK (AVX512F_LEN, i_w) (res3, res_ref))
    abort ();
}
예제 #30
0
static void
TEST (void)
{
  UNION_TYPE (AVX512F_LEN, d) s, res1, res2, res3;
  MASK_TYPE mask = MASK_VALUE;
  double res_ref[SIZE];
  int i, sign = 1;

  for (i = 0; i < SIZE; i++)
    {
      s.a[i] = 123.456 * (i + 2000) * sign;
      res2.a[i] = DEFAULT_VALUE;
      sign = -sign;
    }

  res1.x = INTRINSIC (_rcp14_pd) (s.x);
  res2.x = INTRINSIC (_mask_rcp14_pd) (res2.x, mask, s.x);
  res3.x = INTRINSIC (_maskz_rcp14_pd) (mask, s.x);

  CALC (s.a, res_ref);

  if (UNION_ROUGH_CHECK (AVX512F_LEN, d) (res1, res_ref, 0.0001))
    abort ();

  MASK_MERGE (d) (res_ref, mask, SIZE);
  if (UNION_ROUGH_CHECK (AVX512F_LEN, d) (res2, res_ref, 0.0001))
    abort ();

  MASK_ZERO (d) (res_ref, mask, SIZE);
  if (UNION_ROUGH_CHECK (AVX512F_LEN, d) (res3, res_ref, 0.0001))
    abort ();
}