29   const __m256 vmagic_bias = _mm256_set1_ps(0x1.8000FEp23f);  in xnn_f32_vscaleexpminusmax_ukernel__avx2_p5_x96()  local
76     __m256 vn0 = _mm256_fmadd_ps(vx0, vlog2e, vmagic_bias);  in xnn_f32_vscaleexpminusmax_ukernel__avx2_p5_x96()
77     __m256 vn1 = _mm256_fmadd_ps(vx1, vlog2e, vmagic_bias);  in xnn_f32_vscaleexpminusmax_ukernel__avx2_p5_x96()
78     __m256 vn2 = _mm256_fmadd_ps(vx2, vlog2e, vmagic_bias);  in xnn_f32_vscaleexpminusmax_ukernel__avx2_p5_x96()
79     __m256 vn3 = _mm256_fmadd_ps(vx3, vlog2e, vmagic_bias);  in xnn_f32_vscaleexpminusmax_ukernel__avx2_p5_x96()
80     __m256 vn4 = _mm256_fmadd_ps(vx4, vlog2e, vmagic_bias);  in xnn_f32_vscaleexpminusmax_ukernel__avx2_p5_x96()
81     __m256 vn5 = _mm256_fmadd_ps(vx5, vlog2e, vmagic_bias);  in xnn_f32_vscaleexpminusmax_ukernel__avx2_p5_x96()
82     __m256 vn6 = _mm256_fmadd_ps(vx6, vlog2e, vmagic_bias);  in xnn_f32_vscaleexpminusmax_ukernel__avx2_p5_x96()
83     __m256 vn7 = _mm256_fmadd_ps(vx7, vlog2e, vmagic_bias);  in xnn_f32_vscaleexpminusmax_ukernel__avx2_p5_x96()
84     __m256 vn8 = _mm256_fmadd_ps(vx8, vlog2e, vmagic_bias);  in xnn_f32_vscaleexpminusmax_ukernel__avx2_p5_x96()
85     __m256 vn9 = _mm256_fmadd_ps(vx9, vlog2e, vmagic_bias);  in xnn_f32_vscaleexpminusmax_ukernel__avx2_p5_x96()
86     __m256 vn10 = _mm256_fmadd_ps(vx10, vlog2e, vmagic_bias);  in xnn_f32_vscaleexpminusmax_ukernel__avx2_p5_x96()
87     __m256 vn11 = _mm256_fmadd_ps(vx11, vlog2e, vmagic_bias);  in xnn_f32_vscaleexpminusmax_ukernel__avx2_p5_x96()
105     vn0 = _mm256_sub_ps(vn0, vmagic_bias);  in xnn_f32_vscaleexpminusmax_ukernel__avx2_p5_x96()
106     vn1 = _mm256_sub_ps(vn1, vmagic_bias);  in xnn_f32_vscaleexpminusmax_ukernel__avx2_p5_x96()
107     vn2 = _mm256_sub_ps(vn2, vmagic_bias);  in xnn_f32_vscaleexpminusmax_ukernel__avx2_p5_x96()
108     vn3 = _mm256_sub_ps(vn3, vmagic_bias);  in xnn_f32_vscaleexpminusmax_ukernel__avx2_p5_x96()
109     vn4 = _mm256_sub_ps(vn4, vmagic_bias);  in xnn_f32_vscaleexpminusmax_ukernel__avx2_p5_x96()
110     vn5 = _mm256_sub_ps(vn5, vmagic_bias);  in xnn_f32_vscaleexpminusmax_ukernel__avx2_p5_x96()
111     vn6 = _mm256_sub_ps(vn6, vmagic_bias);  in xnn_f32_vscaleexpminusmax_ukernel__avx2_p5_x96()
112     vn7 = _mm256_sub_ps(vn7, vmagic_bias);  in xnn_f32_vscaleexpminusmax_ukernel__avx2_p5_x96()
113     vn8 = _mm256_sub_ps(vn8, vmagic_bias);  in xnn_f32_vscaleexpminusmax_ukernel__avx2_p5_x96()
114     vn9 = _mm256_sub_ps(vn9, vmagic_bias);  in xnn_f32_vscaleexpminusmax_ukernel__avx2_p5_x96()
115     vn10 = _mm256_sub_ps(vn10, vmagic_bias);  in xnn_f32_vscaleexpminusmax_ukernel__avx2_p5_x96()
116     vn11 = _mm256_sub_ps(vn11, vmagic_bias);  in xnn_f32_vscaleexpminusmax_ukernel__avx2_p5_x96()
282     __m256 vn = _mm256_fmadd_ps(vx, vlog2e, vmagic_bias);  in xnn_f32_vscaleexpminusmax_ukernel__avx2_p5_x96()
289     vn = _mm256_sub_ps(vn, vmagic_bias);  in xnn_f32_vscaleexpminusmax_ukernel__avx2_p5_x96()
332     __m256 vn = _mm256_fmadd_ps(vx, vlog2e, vmagic_bias);  in xnn_f32_vscaleexpminusmax_ukernel__avx2_p5_x96()
339     vn = _mm256_sub_ps(vn, vmagic_bias);  in xnn_f32_vscaleexpminusmax_ukernel__avx2_p5_x96()