28   const __m256 vmagic_bias = _mm256_set1_ps(0x1.8000FEp23f);  in xnn_f32_raddstoreexpminusmax_ukernel__avx2_p5_x96_acc6()  local
80     __m256 vn0 = _mm256_fmadd_ps(vx0, vlog2e, vmagic_bias);  in xnn_f32_raddstoreexpminusmax_ukernel__avx2_p5_x96_acc6()
81     __m256 vn1 = _mm256_fmadd_ps(vx1, vlog2e, vmagic_bias);  in xnn_f32_raddstoreexpminusmax_ukernel__avx2_p5_x96_acc6()
82     __m256 vn2 = _mm256_fmadd_ps(vx2, vlog2e, vmagic_bias);  in xnn_f32_raddstoreexpminusmax_ukernel__avx2_p5_x96_acc6()
83     __m256 vn3 = _mm256_fmadd_ps(vx3, vlog2e, vmagic_bias);  in xnn_f32_raddstoreexpminusmax_ukernel__avx2_p5_x96_acc6()
84     __m256 vn4 = _mm256_fmadd_ps(vx4, vlog2e, vmagic_bias);  in xnn_f32_raddstoreexpminusmax_ukernel__avx2_p5_x96_acc6()
85     __m256 vn5 = _mm256_fmadd_ps(vx5, vlog2e, vmagic_bias);  in xnn_f32_raddstoreexpminusmax_ukernel__avx2_p5_x96_acc6()
86     __m256 vn6 = _mm256_fmadd_ps(vx6, vlog2e, vmagic_bias);  in xnn_f32_raddstoreexpminusmax_ukernel__avx2_p5_x96_acc6()
87     __m256 vn7 = _mm256_fmadd_ps(vx7, vlog2e, vmagic_bias);  in xnn_f32_raddstoreexpminusmax_ukernel__avx2_p5_x96_acc6()
88     __m256 vn8 = _mm256_fmadd_ps(vx8, vlog2e, vmagic_bias);  in xnn_f32_raddstoreexpminusmax_ukernel__avx2_p5_x96_acc6()
89     __m256 vn9 = _mm256_fmadd_ps(vx9, vlog2e, vmagic_bias);  in xnn_f32_raddstoreexpminusmax_ukernel__avx2_p5_x96_acc6()
90     __m256 vn10 = _mm256_fmadd_ps(vx10, vlog2e, vmagic_bias);  in xnn_f32_raddstoreexpminusmax_ukernel__avx2_p5_x96_acc6()
91     __m256 vn11 = _mm256_fmadd_ps(vx11, vlog2e, vmagic_bias);  in xnn_f32_raddstoreexpminusmax_ukernel__avx2_p5_x96_acc6()
109     vn0 = _mm256_sub_ps(vn0, vmagic_bias);  in xnn_f32_raddstoreexpminusmax_ukernel__avx2_p5_x96_acc6()
110     vn1 = _mm256_sub_ps(vn1, vmagic_bias);  in xnn_f32_raddstoreexpminusmax_ukernel__avx2_p5_x96_acc6()
111     vn2 = _mm256_sub_ps(vn2, vmagic_bias);  in xnn_f32_raddstoreexpminusmax_ukernel__avx2_p5_x96_acc6()
112     vn3 = _mm256_sub_ps(vn3, vmagic_bias);  in xnn_f32_raddstoreexpminusmax_ukernel__avx2_p5_x96_acc6()
113     vn4 = _mm256_sub_ps(vn4, vmagic_bias);  in xnn_f32_raddstoreexpminusmax_ukernel__avx2_p5_x96_acc6()
114     vn5 = _mm256_sub_ps(vn5, vmagic_bias);  in xnn_f32_raddstoreexpminusmax_ukernel__avx2_p5_x96_acc6()
115     vn6 = _mm256_sub_ps(vn6, vmagic_bias);  in xnn_f32_raddstoreexpminusmax_ukernel__avx2_p5_x96_acc6()
116     vn7 = _mm256_sub_ps(vn7, vmagic_bias);  in xnn_f32_raddstoreexpminusmax_ukernel__avx2_p5_x96_acc6()
117     vn8 = _mm256_sub_ps(vn8, vmagic_bias);  in xnn_f32_raddstoreexpminusmax_ukernel__avx2_p5_x96_acc6()
118     vn9 = _mm256_sub_ps(vn9, vmagic_bias);  in xnn_f32_raddstoreexpminusmax_ukernel__avx2_p5_x96_acc6()
119     vn10 = _mm256_sub_ps(vn10, vmagic_bias);  in xnn_f32_raddstoreexpminusmax_ukernel__avx2_p5_x96_acc6()
120     vn11 = _mm256_sub_ps(vn11, vmagic_bias);  in xnn_f32_raddstoreexpminusmax_ukernel__avx2_p5_x96_acc6()
294     __m256 vn = _mm256_fmadd_ps(vx, vlog2e, vmagic_bias);  in xnn_f32_raddstoreexpminusmax_ukernel__avx2_p5_x96_acc6()
301     vn = _mm256_sub_ps(vn, vmagic_bias);  in xnn_f32_raddstoreexpminusmax_ukernel__avx2_p5_x96_acc6()
344     __m256 vn = _mm256_fmadd_ps(vx, vlog2e, vmagic_bias);  in xnn_f32_raddstoreexpminusmax_ukernel__avx2_p5_x96_acc6()
351     vn = _mm256_sub_ps(vn, vmagic_bias);  in xnn_f32_raddstoreexpminusmax_ukernel__avx2_p5_x96_acc6()