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
55     m = 0x3F800000 - (xinv.i & 0x7F800000);  in vrecip_medprec_f32()
57     xinv.f = 1.41176471f - 0.47058824f * xinv.f;  in vrecip_medprec_f32()
60     b = 2.0f - xinv.f * ax;  in vrecip_medprec_f32()
63     b = 2.0f - xinv.f * ax;  in vrecip_medprec_f32()
66     b = 2.0f - xinv.f * ax;  in vrecip_medprec_f32()
89     m = 0x3F800000 - (xinv.i & 0x7F800000);  in vrecip_hiprec_f32()
91     xinv.f = 1.41176471f - 0.47058824f * xinv.f;  in vrecip_hiprec_f32()
94     b = 2.0f - xinv.f * ax;  in vrecip_hiprec_f32()
97     b = 2.0f - xinv.f * ax;  in vrecip_hiprec_f32()
100     b = 2.0f - xinv.f * ax;  in vrecip_hiprec_f32()
103     b = 2.0f - xinv.f * ax;  in vrecip_hiprec_f32()
122   @brief         Single-precision taylor dev.
145     int32x4_t * e)  in vmant_exp_f32()  argument
152     n = n - 127;  in vmant_exp_f32()
153     r.i = r.i - (n << 23);  in vmant_exp_f32()
155     *e = n;  in vmant_exp_f32()
215     // set log0 down to -inf  in vlogq_f32()
216     vecAcc0 = vdupq_m(vecAcc0, -F32_MAX, vcmpeqq(vecIn, 0.0f));  in vlogq_f32()
223     // Perform range reduction [-log(2),log(2)]  in vexpq_f32()
233     poly = vdupq_m(poly, 0.0f, vcmpltq_n_s32(m, -126));  in vexpq_f32()
240     nb--;  in arm_vec_exponent_f32()
243         nb--;  in arm_vec_exponent_f32()
260     // v.f = v.f * (8 + w * (-28 + w * (56 + w * (-70 + w *(56 + w * (-28 + w * (8 - w)))))));  in vrecip_f32()
262     vecTmp = vfmasq(vecW, vecTmp, -28.0f);  in vrecip_f32()
264     vecTmp = vfmasq(vecW, vecTmp, -70.0f);  in vrecip_f32()
266     vecTmp = vfmasq(vecW, vecTmp, -28.0f);  in vrecip_f32()
282         vminnmq_f32(vmaxnmq_f32(val, vdupq_n_f32(-10.f)), vdupq_n_f32(10.0f));  in vtanhq_f32()
315     nb --;  in arm_vec_exponent_f32()
319         nb--;  in arm_vec_exponent_f32()
328     float32x4_t e = vrsqrteq_f32(x1);  in __arm_vec_sqrt_f32_neon()  local
329     e = vmulq_f32(vrsqrtsq_f32(vmulq_f32(x1, e), e), e);  in __arm_vec_sqrt_f32_neon()
330     e = vmulq_f32(vrsqrtsq_f32(vmulq_f32(x1, e), e), e);  in __arm_vec_sqrt_f32_neon()
331     return vmulq_f32(x, e);  in __arm_vec_sqrt_f32_neon()