6  * Target Processor: Cortex-M and Cortex-A cores
9  * Copyright (c) 2010-2021 Arm Limited or its affiliates. All rights reserved.
11  * SPDX-License-Identifier: Apache-2.0
17  * www.apache.org/licenses/LICENSE-2.0
57     m = 0x03c00 - (xinv.i & 0x07c00);  in vrecip_medprec_f16()
59     xinv.f = 1.41176471f16 - 0.47058824f16 * xinv.f;  in vrecip_medprec_f16()
62     b = 2.0f16 - xinv.f * ax;  in vrecip_medprec_f16()
65     b = 2.0f16 - xinv.f * ax;  in vrecip_medprec_f16()
68     b = 2.0f16 - xinv.f * ax;  in vrecip_medprec_f16()
91     m = 0x03c00 - (xinv.i & 0x07c00);  in vrecip_hiprec_f16()
93     xinv.f = 1.41176471f16 - 0.47058824f16 * xinv.f;  in vrecip_hiprec_f16()
96     b = 2.0f16 - xinv.f * ax;  in vrecip_hiprec_f16()
99     b = 2.0f16 - xinv.f * ax;  in vrecip_hiprec_f16()
102     b = 2.0f16 - xinv.f * ax;  in vrecip_hiprec_f16()
105     b = 2.0f16 - xinv.f * ax;  in vrecip_hiprec_f16()
125   @brief         Single-precision taylor dev.
147     any16x8_t       r;    \
150     r.f = x;              \
151     n = r.i >> 10;        \
152     n = n - 15;           \
153     r.i = r.i - (n << 10);\
156     vecTmpFlt1 = r.f;
172      * a = (__logf_lut_f16[4] * r.f) + (__logf_lut_f16[0]);  in vlogq_f16()
177      * b = (__logf_lut_f16[6] * r.f) + (__logf_lut_f16[2]);  in vlogq_f16()
182      * c = (__logf_lut_f16[5] * r.f) + (__logf_lut_f16[1]);  in vlogq_f16()
187      * d = (__logf_lut_f16[7] * r.f) + (__logf_lut_f16[3]);  in vlogq_f16()
205      * r.f = a + c * xx;  in vlogq_f16()
210      * r.f = r.f + ((float32_t) m) * __logf_rng_f16;  in vlogq_f16()
213     // set log0 down to -inf  in vlogq_f16()
214     vecAcc0 = vdupq_m_n_f16(vecAcc0, -(_Float16)F16INFINITY, vcmpeqq_n_f16(vecIn, 0.0f));  in vlogq_f16()
221     // Perform range reduction [-log(2),log(2)]  in vexpq_f16()
231     poly = vdupq_m_n_f16(poly, 0.0f16, vcmpltq_n_s16(m, -14));  in vexpq_f16()
237     float16x8_t         r = x;  in arm_vec_exponent_f16()  local
238     nb--;  in arm_vec_exponent_f16()
240         r = vmulq(r, x);  in arm_vec_exponent_f16()
241         nb--;  in arm_vec_exponent_f16()
243     return (r);  in arm_vec_exponent_f16()
267     // v.f = v.f * (8 + w * (-28 + w * (56 + w * (-70 + w *(56 + w * (-28 + w * (8 - w)))))));  in vrecip_f16()
269     vecTmp = vfmasq_n_f16(vecW, vecTmp, -28.0f16);  in vrecip_f16()
271     vecTmp = vfmasq_n_f16(vecW, vecTmp, -70.0f16);  in vrecip_f16()
273     vecTmp = vfmasq_n_f16(vecW, vecTmp, -28.0f16);  in vrecip_f16()
289         vminnmq_f16(vmaxnmq_f16(val, vdupq_n_f16(-10.f16)), vdupq_n_f16(10.0f16));  in vtanhq_f16()