inline
mask_type operator>=(const short_vec<float, 32>& other) const
{
return
(_mm512_cmp_ps_mask(val1, other.val1, _CMP_GE_OS) << 0) +
(_mm512_cmp_ps_mask(val2, other.val2, _CMP_GE_OS) << 16);
}
// exp()
inline mic_m512_t mic_exp_ps(mic_m512_t x) {
x = _mm512_min_ps(x, _ps_exp_hi);
x = _mm512_max_ps(x, _ps_exp_lo);
mic_m512_t temp_2 = _mm512_fmadd_ps(x, _ps_cephes_LOG2EF, _ps_0point5);
mic_m512_t temp_1 = _mm512_round_ps(temp_2, _MM_FROUND_TO_NEAREST_INT, _MM_EXPADJ_NONE);
mic_m512_t temp_3 = _mm512_sub_ps(temp_1, temp_2);
__mmask16 mask = _mm512_cmp_ps_mask(temp_3, _ps_0, _MM_CMPINT_GT);
temp_2 = _mm512_mask_sub_ps(temp_1, mask, temp_1, _ps_1);
__m512i emm0 = _mm512_cvtfxpnt_round_adjustps_epi32(temp_2, _MM_FROUND_TO_NEAREST_INT, _MM_EXPADJ_NONE);
x = _mm512_fnmadd_ps(temp_2, _ps_cephes_exp_C12, x);
mic_m512_t x2 = _mm512_mul_ps(x, x);
mic_m512_t x3 = _mm512_mul_ps(x2, x);
mic_m512_t x4 = _mm512_mul_ps(x2, x2);
temp_1 = _mm512_add_ps(x, _ps_1);
temp_1 = _mm512_fmadd_ps(x2, _ps_cephes_exp_p5, temp_1);
temp_1 = _mm512_fmadd_ps(x3, _ps_cephes_exp_p4, temp_1);
temp_2 = _mm512_mul_ps(x3, _ps_cephes_exp_p0);
temp_3 = _mm512_mul_ps(x2, _ps_cephes_exp_p1);
mic_m512_t temp_4 = _mm512_mul_ps(x, _ps_cephes_exp_p2);
emm0 = _mm512_add_epi32(emm0, _pi32_0x7f);
temp_2 = _mm512_add_ps(temp_2, temp_3);
temp_3 = _mm512_add_ps(temp_3, temp_4);
temp_2 = _mm512_add_ps(temp_2, temp_3);
emm0 = _mm512_slli_epi32(emm0, 23);
mic_m512_t pow2n = _mm512_castsi512_ps(emm0);
temp_2 = _mm512_mul_ps(temp_2, x4);
mic_m512_t y = _mm512_add_ps(temp_1, temp_2);
y = _mm512_mul_ps(y, pow2n);
return y;
} // newexp_ps()
static batch_bool_type isnan(const batch_type& x)
{
return _mm512_cmp_ps_mask(x, x, _CMP_UNORD_Q);
}
static batch_bool_type lte(const batch_type& lhs, const batch_type& rhs)
{
return _mm512_cmp_ps_mask(lhs, rhs, _CMP_LE_OQ);
}
void
test8bit (void)
{
m512i = _mm512_permutex_epi64 (m512i, 256); /* { dg-error "the last argument must be an 8-bit immediate" } */
m512i = _mm512_mask_permutex_epi64 (m512i, mmask8, m512i, 256); /* { dg-error "the last argument must be an 8-bit immediate" } */
m512i = _mm512_maskz_permutex_epi64 (mmask8, m512i, 256); /* { dg-error "the last argument must be an 8-bit immediate" } */
m512i = _mm512_ternarylogic_epi64 (m512i, m512i, m512i, 256); /* { dg-error "the last argument must be an 8-bit immediate" } */
m512i = _mm512_mask_ternarylogic_epi64 (m512i, mmask8, m512i, m512i, 256); /* { dg-error "the last argument must be an 8-bit immediate" } */
m512i = _mm512_maskz_ternarylogic_epi64 (mmask8, m512i, m512i, m512i, 256); /* { dg-error "the last argument must be an 8-bit immediate" } */
m512i = _mm512_ternarylogic_epi32 (m512i, m512i, m512i, 256); /* { dg-error "the last argument must be an 8-bit immediate" } */
m512i = _mm512_mask_ternarylogic_epi32 (m512i, mmask16, m512i, m512i, 256); /* { dg-error "the last argument must be an 8-bit immediate" } */
m512i = _mm512_maskz_ternarylogic_epi32 (mmask16, m512i, m512i, m512i, 256); /* { dg-error "the last argument must be an 8-bit immediate" } */
m512i = _mm512_shuffle_epi32 (m512i, 256); /* { dg-error "the last argument must be an 8-bit immediate" } */
m512i = _mm512_mask_shuffle_epi32 (m512i, mmask16, m512i, 256); /* { dg-error "the last argument must be an 8-bit immediate" } */
m512i = _mm512_maskz_shuffle_epi32 (mmask16, m512i, 256); /* { dg-error "the last argument must be an 8-bit immediate" } */
m512i = _mm512_shuffle_i64x2 (m512i, m512i, 256); /* { dg-error "the last argument must be an 8-bit immediate" } */
m512i = _mm512_mask_shuffle_i64x2 (m512i, mmask8, m512i, m512i, 256); /* { dg-error "the last argument must be an 8-bit immediate" } */
m512i = _mm512_maskz_shuffle_i64x2 (mmask8, m512i, m512i, 256); /* { dg-error "the last argument must be an 8-bit immediate" } */
m512i = _mm512_shuffle_i32x4 (m512i, m512i, 256); /* { dg-error "the last argument must be an 8-bit immediate" } */
m512i = _mm512_mask_shuffle_i32x4 (m512i, mmask16, m512i, m512i, 256); /* { dg-error "the last argument must be an 8-bit immediate" } */
m512i = _mm512_maskz_shuffle_i32x4 (mmask16, m512i, m512i, 256); /* { dg-error "the last argument must be an 8-bit immediate" } */
m512d = _mm512_shuffle_f64x2 (m512d, m512d, 256); /* { dg-error "the last argument must be an 8-bit immediate" } */
m512d = _mm512_mask_shuffle_f64x2 (m512d, mmask8, m512d, m512d, 256); /* { dg-error "the last argument must be an 8-bit immediate" } */
m512d = _mm512_maskz_shuffle_f64x2 (mmask8, m512d, m512d, 256); /* { dg-error "the last argument must be an 8-bit immediate" } */
m512 = _mm512_shuffle_f32x4 (m512, m512, 256); /* { dg-error "the last argument must be an 8-bit immediate" } */
m512 = _mm512_mask_shuffle_f32x4 (m512, mmask16, m512, m512, 256); /* { dg-error "the last argument must be an 8-bit immediate" } */
m512 = _mm512_maskz_shuffle_f32x4 (mmask16, m512, m512, 256); /* { dg-error "the last argument must be an 8-bit immediate" } */
m512d = _mm512_permutex_pd (m512d, 256); /* { dg-error "the last argument must be an 8-bit immediate" } */
m512d = _mm512_mask_permutex_pd (m512d, mmask8, m512d, 256); /* { dg-error "the last argument must be an 8-bit immediate" } */
m512d = _mm512_maskz_permutex_pd (mmask8, m512d, 256); /* { dg-error "the last argument must be an 8-bit immediate" } */
m512d = _mm512_permute_pd (m512d, 256); /* { dg-error "the last argument must be an 8-bit immediate" } */
m512d = _mm512_mask_permute_pd (m512d, mmask8, m512d, 256); /* { dg-error "the last argument must be an 8-bit immediate" } */
m512d = _mm512_maskz_permute_pd (mmask8, m512d, 256); /* { dg-error "the last argument must be an 8-bit immediate" } */
m512 = _mm512_permute_ps (m512, 256); /* { dg-error "the last argument must be an 8-bit immediate" } */
m512 = _mm512_mask_permute_ps (m512, mmask16, m512, 256); /* { dg-error "the last argument must be an 8-bit immediate" } */
m512 = _mm512_maskz_permute_ps (mmask16, m512, 256); /* { dg-error "the last argument must be an 8-bit immediate" } */
m512d = _mm512_shuffle_pd (m512d, m512d, 256); /* { dg-error "the last argument must be an 8-bit immediate" } */
m512d = _mm512_mask_shuffle_pd (m512d, mmask8, m512d, m512d, 256); /* { dg-error "the last argument must be an 8-bit immediate" } */
m512d = _mm512_maskz_shuffle_pd (mmask8, m512d, m512d, 256); /* { dg-error "the last argument must be an 8-bit immediate" } */
m512 = _mm512_shuffle_ps (m512, m512, 256); /* { dg-error "the last argument must be an 8-bit immediate" } */
m512 = _mm512_mask_shuffle_ps (m512, mmask16, m512, m512, 256); /* { dg-error "the last argument must be an 8-bit immediate" } */
m512 = _mm512_maskz_shuffle_ps (mmask16, m512, m512, 256); /* { dg-error "the last argument must be an 8-bit immediate" } */
m512d = _mm512_fixupimm_pd (m512d, m512d, m512i, 256); /* { dg-error "the immediate argument must be an 8-bit immediate" } */
m512d = _mm512_mask_fixupimm_pd (m512d, mmask8, m512d, m512i, 256); /* { dg-error "the immediate argument must be an 8-bit immediate" } */
m512d = _mm512_maskz_fixupimm_pd (mmask8, m512d, m512d, m512i, 256); /* { dg-error "the immediate argument must be an 8-bit immediate" } */
m512 = _mm512_fixupimm_ps (m512, m512, m512i, 256); /* { dg-error "the immediate argument must be an 8-bit immediate" } */
m512 = _mm512_mask_fixupimm_ps (m512, mmask16, m512, m512i, 256); /* { dg-error "the immediate argument must be an 8-bit immediate" } */
m512 = _mm512_maskz_fixupimm_ps (mmask16, m512, m512, m512i, 256); /* { dg-error "the immediate argument must be an 8-bit immediate" } */
m128d = _mm_fixupimm_sd (m128d, m128d, m128i, 256); /* { dg-error "the immediate argument must be an 8-bit immediate" } */
m128d = _mm_mask_fixupimm_sd (m128d, mmask8, m128d, m128i, 256); /* { dg-error "the immediate argument must be an 8-bit immediate" } */
m128d = _mm_maskz_fixupimm_sd (mmask8, m128d, m128d, m128i, 256); /* { dg-error "the immediate argument must be an 8-bit immediate" } */
m128 = _mm_fixupimm_ss (m128, m128, m128i, 256); /* { dg-error "the immediate argument must be an 8-bit immediate" } */
m128 = _mm_mask_fixupimm_ss (m128, mmask8, m128, m128i, 256); /* { dg-error "the immediate argument must be an 8-bit immediate" } */
m128 = _mm_maskz_fixupimm_ss (mmask8, m128, m128, m128i, 256); /* { dg-error "the immediate argument must be an 8-bit immediate" } */
m512i = _mm512_rol_epi32 (m512i, 256); /* { dg-error "the last argument must be an 8-bit immediate" } */
m512i = _mm512_mask_rol_epi32 (m512i, mmask16, m512i, 256); /* { dg-error "the last argument must be an 8-bit immediate" } */
m512i = _mm512_maskz_rol_epi32 (mmask16, m512i, 256); /* { dg-error "the last argument must be an 8-bit immediate" } */
m512i = _mm512_ror_epi32 (m512i, 256); /* { dg-error "the last argument must be an 8-bit immediate" } */
m512i = _mm512_mask_ror_epi32 (m512i, mmask16, m512i, 256); /* { dg-error "the last argument must be an 8-bit immediate" } */
m512i = _mm512_maskz_ror_epi32 (mmask16, m512i, 256); /* { dg-error "the last argument must be an 8-bit immediate" } */
m512i = _mm512_rol_epi64 (m512i, 256); /* { dg-error "the last argument must be an 8-bit immediate" } */
m512i = _mm512_mask_rol_epi64 (m512i, mmask8, m512i, 256); /* { dg-error "the last argument must be an 8-bit immediate" } */
m512i = _mm512_maskz_rol_epi64 (mmask8, m512i, 256); /* { dg-error "the last argument must be an 8-bit immediate" } */
m512i = _mm512_ror_epi64 (m512i, 256); /* { dg-error "the last argument must be an 8-bit immediate" } */
m512i = _mm512_mask_ror_epi64 (m512i, mmask8, m512i, 256); /* { dg-error "the last argument must be an 8-bit immediate" } */
m512i = _mm512_maskz_ror_epi64 (mmask8, m512i, 256); /* { dg-error "the last argument must be an 8-bit immediate" } */
m256i = _mm512_cvtps_ph (m512, 256); /* { dg-error "the last argument must be an 8-bit immediate" } */
m256i = _mm512_mask_cvtps_ph (m256i, mmask16, m512, 256); /* { dg-error "the last argument must be an 8-bit immediate" } */
m256i = _mm512_maskz_cvtps_ph (mmask16, m512, 256); /* { dg-error "the last argument must be an 8-bit immediate" } */
m512d = _mm512_roundscale_pd (m512d, 256); /* { dg-error "the immediate argument must be an 8-bit immediate" } */
m512d = _mm512_mask_roundscale_pd (m512d, mmask8, m512d, 256); /* { dg-error "the immediate argument must be an 8-bit immediate" } */
m512d = _mm512_maskz_roundscale_pd (mmask8, m512d, 256); /* { dg-error "the immediate argument must be an 8-bit immediate" } */
m512 = _mm512_roundscale_ps (m512, 256); /* { dg-error "the immediate argument must be an 8-bit immediate" } */
m512 = _mm512_mask_roundscale_ps (m512, mmask16, m512, 256); /* { dg-error "the immediate argument must be an 8-bit immediate" } */
m512 = _mm512_maskz_roundscale_ps (mmask16, m512, 256); /* { dg-error "the immediate argument must be an 8-bit immediate" } */
m128d = _mm_roundscale_sd (m128d, m128d, 256); /* { dg-error "the immediate argument must be an 8-bit immediate" } */
m128 = _mm_roundscale_ss (m128, m128, 256); /* { dg-error "the immediate argument must be an 8-bit immediate" } */
m512i = _mm512_alignr_epi32 (m512i, m512i, 256); /* { dg-error "the last argument must be an 8-bit immediate" } */
m512i = _mm512_mask_alignr_epi32 (m512i, mmask16, m512i, m512i, 256); /* { dg-error "the last argument must be an 8-bit immediate" } */
m512i = _mm512_maskz_alignr_epi32 (mmask16, m512i, m512i, 256); /* { dg-error "the last argument must be an 8-bit immediate" } */
m512i = _mm512_alignr_epi64 (m512i, m512i, 256); /* { dg-error "the last argument must be an 8-bit immediate" } */
m512i = _mm512_mask_alignr_epi64 (m512i, mmask8, m512i, m512i, 256); /* { dg-error "the last argument must be an 8-bit immediate" } */
m512i = _mm512_maskz_alignr_epi64 (mmask8, m512i, m512i, 256); /* { dg-error "the last argument must be an 8-bit immediate" } */
mmask8 = _mm512_cmp_epi64_mask (m512i, m512i, 256); /* { dg-error "the last argument must be a 3-bit immediate" } */
mmask8 = _mm512_cmp_epi32_mask (m512i, m512i, 256); /* { dg-error "the last argument must be a 3-bit immediate" } */
mmask8 = _mm512_cmp_epu64_mask (m512i, m512i, 256); /* { dg-error "the last argument must be a 3-bit immediate" } */
mmask8 = _mm512_cmp_epu32_mask (m512i, m512i, 256); /* { dg-error "the last argument must be a 3-bit immediate" } */
mmask8 = _mm512_cmp_pd_mask (m512d, m512d, 256); /* { dg-error "the immediate argument must be a 5-bit immediate" } */
mmask8 = _mm512_cmp_ps_mask (m512, m512, 256); /* { dg-error "the immediate argument must be a 5-bit immediate" } */
mmask8 = _mm512_mask_cmp_epi64_mask (2, m512i, m512i, 256); /* { dg-error "the last argument must be a 3-bit immediate" } */
mmask8 = _mm512_mask_cmp_epi32_mask (2, m512i, m512i, 256); /* { dg-error "the last argument must be a 3-bit immediate" } */
mmask8 = _mm512_mask_cmp_epu64_mask (2, m512i, m512i, 256); /* { dg-error "the last argument must be a 3-bit immediate" } */
mmask8 = _mm512_mask_cmp_epu32_mask (2, m512i, m512i, 256); /* { dg-error "the last argument must be a 3-bit immediate" } */
mmask8 = _mm512_mask_cmp_pd_mask (2, m512d, m512d, 256); /* { dg-error "the immediate argument must be a 5-bit immediate" } */
mmask8 = _mm512_mask_cmp_ps_mask (2, m512, m512, 256); /* { dg-error "the immediate argument must be a 5-bit immediate" } */
mmask8 = _mm_cmp_sd_mask (m128d, m128d, 256); /* { dg-error "the immediate argument must be a 5-bit immediate" } */
mmask8 = _mm_cmp_ss_mask (m128, m128, 256); /* { dg-error "the immediate argument must be a 5-bit immediate" } */
mmask8 = _mm_mask_cmp_sd_mask (1, m128d, m128d, 256); /* { dg-error "the immediate argument must be a 5-bit immediate" } */
mmask8 = _mm_mask_cmp_ss_mask (1, m128, m128, 256); /* { dg-error "the immediate argument must be a 5-bit immediate" } */
}
void AVX512BW_mandelbrot(
float Re_min, float Re_max,
float Im_min, float Im_max,
float threshold,
int maxiters,
int width, int height,
uint8_t *data)
{
float dRe, dIm;
int x, y;
__m128i* ptr = (__m128i*)data;
// step on Re and Im axis
dRe = (Re_max - Re_min)/width;
dIm = (Im_max - Im_min)/height;
// prepare vectors
// 1. threshold
const __m512 vec_threshold = _mm512_set1_ps(threshold);
// 2. Cim
__m512 Cim = _mm512_set1_ps(Im_min);
// 3. Re advance every x iteration
const __m512 vec_dRe = _mm512_set1_ps(16*dRe);
// 4. Im advance every y iteration
const __m512 vec_dIm = _mm512_set1_ps(dIm);
// calculations
for (y=0; y < height; y++) {
__m512 Cre = _mm512_setr_ps(
Re_min + 0*dRe, Re_min + 1*dRe, Re_min + 2*dRe, Re_min + 3*dRe,
Re_min + 4*dRe, Re_min + 5*dRe, Re_min + 6*dRe, Re_min + 7*dRe,
Re_min + 8*dRe, Re_min + 9*dRe, Re_min + 10*dRe, Re_min + 11*dRe,
Re_min + 12*dRe, Re_min + 13*dRe, Re_min + 14*dRe, Re_min + 15*dRe
);
for (x=0; x < width; x+=16) {
__m512 Xre = _mm512_setzero_ps();
__m512 Xim = _mm512_setzero_ps();
__m128i itercount = _mm_setzero_si128();
int i;
for (i=0; i < maxiters; i++) {
// Tre = Xre^2 - Xim^2 + Cim
const __m512 Xre2 = _mm512_mul_ps(Xre, Xre);
const __m512 Xim2 = _mm512_mul_ps(Xim, Xim);
const __m512 Tre = _mm512_add_ps(Cre, _mm512_sub_ps(Xre2, Xim2));
// Tim = 2*Xre*Xim + Cre
const __m512 t1 = _mm512_mul_ps(Xre, Xim);
const __m512 Tim = _mm512_add_ps(Cim, _mm512_add_ps(t1, t1));
// sqr_dist = Tre^2 + Tim^2
__m512 Tre2 = _mm512_mul_ps(Tre, Tre);
__m512 Tim2 = _mm512_mul_ps(Tim, Tim);
__m512 sqr_dist = _mm512_add_ps(Tre2, Tim2);
// sqr_dist < threshold => 16-bit mask
__mmask16 mask = _mm512_cmp_ps_mask(sqr_dist, vec_threshold, _CMP_LE_OS);
if (mask == 0) {
break;
}
// Note: unlike SSE/AVX2 versions itercount is a packed byte vector,
// thus conversion packed dword -> byte is not needed.
itercount = _mm_sub_epi8(itercount, _mm_movm_epi8(mask));
Xre = Tre;
Xim = Tim;
} // for
*ptr++ = itercount;
// advance Cre vector
Cre = _mm512_add_ps(Cre, vec_dRe);
}
// advance Cim vector
Cim = _mm512_add_ps(Cim, vec_dIm);
}
}
