| /external/XNNPACK/src/f32-dwconv/gen/ |
| D | up16x9-minmax-fma3-acc2.c | 138 const __m256 vk5x01234567 = _mm256_load_ps(w + 96); in xnn_f32_dwconv_minmax_ukernel_up16x9__fma3_acc2() local
140 vacc01234567p1 = _mm256_fmadd_ps(vi5x01234567, vk5x01234567, vacc01234567p1); in xnn_f32_dwconv_minmax_ukernel_up16x9__fma3_acc2()
221 const __m256 vk5x01234567 = _mm256_load_ps(w + 96); in xnn_f32_dwconv_minmax_ukernel_up16x9__fma3_acc2() local
222 vacc01234567p1 = _mm256_fmadd_ps(vi5x01234567, vk5x01234567, vacc01234567p1); in xnn_f32_dwconv_minmax_ukernel_up16x9__fma3_acc2()
281 const __m256 vk5x01234567 = _mm256_load_ps(w + 96); in xnn_f32_dwconv_minmax_ukernel_up16x9__fma3_acc2() local
282 vacc01234567p1 = _mm256_fmadd_ps(vi5x01234567, vk5x01234567, vacc01234567p1); in xnn_f32_dwconv_minmax_ukernel_up16x9__fma3_acc2()
|
| D | up16x9-minmax-fma3.c | 138 const __m256 vk5x01234567 = _mm256_load_ps(w + 96); in xnn_f32_dwconv_minmax_ukernel_up16x9__fma3() local
140 vacc01234567p0 = _mm256_fmadd_ps(vi5x01234567, vk5x01234567, vacc01234567p0); in xnn_f32_dwconv_minmax_ukernel_up16x9__fma3()
218 const __m256 vk5x01234567 = _mm256_load_ps(w + 96); in xnn_f32_dwconv_minmax_ukernel_up16x9__fma3() local
219 vacc01234567p0 = _mm256_fmadd_ps(vi5x01234567, vk5x01234567, vacc01234567p0); in xnn_f32_dwconv_minmax_ukernel_up16x9__fma3()
276 const __m256 vk5x01234567 = _mm256_load_ps(w + 96); in xnn_f32_dwconv_minmax_ukernel_up16x9__fma3() local
277 vacc01234567p0 = _mm256_fmadd_ps(vi5x01234567, vk5x01234567, vacc01234567p0); in xnn_f32_dwconv_minmax_ukernel_up16x9__fma3()
|
| D | up16x9-minmax-avx.c | 138 const __m256 vk5x01234567 = _mm256_load_ps(w + 96); in xnn_f32_dwconv_minmax_ukernel_up16x9__avx() local
140 vacc01234567p0 = _mm256_add_ps(vacc01234567p0, _mm256_mul_ps(vi5x01234567, vk5x01234567)); in xnn_f32_dwconv_minmax_ukernel_up16x9__avx()
218 const __m256 vk5x01234567 = _mm256_load_ps(w + 96); in xnn_f32_dwconv_minmax_ukernel_up16x9__avx() local
219 vacc01234567p0 = _mm256_add_ps(vacc01234567p0, _mm256_mul_ps(vi5x01234567, vk5x01234567)); in xnn_f32_dwconv_minmax_ukernel_up16x9__avx()
276 const __m256 vk5x01234567 = _mm256_load_ps(w + 96); in xnn_f32_dwconv_minmax_ukernel_up16x9__avx() local
277 vacc01234567p0 = _mm256_add_ps(vacc01234567p0, _mm256_mul_ps(vi5x01234567, vk5x01234567)); in xnn_f32_dwconv_minmax_ukernel_up16x9__avx()
|
| D | up16x9-minmax-avx-acc2.c | 138 const __m256 vk5x01234567 = _mm256_load_ps(w + 96); in xnn_f32_dwconv_minmax_ukernel_up16x9__avx_acc2() local
140 vacc01234567p1 = _mm256_add_ps(vacc01234567p1, _mm256_mul_ps(vi5x01234567, vk5x01234567)); in xnn_f32_dwconv_minmax_ukernel_up16x9__avx_acc2()
221 const __m256 vk5x01234567 = _mm256_load_ps(w + 96); in xnn_f32_dwconv_minmax_ukernel_up16x9__avx_acc2() local
222 vacc01234567p1 = _mm256_add_ps(vacc01234567p1, _mm256_mul_ps(vi5x01234567, vk5x01234567)); in xnn_f32_dwconv_minmax_ukernel_up16x9__avx_acc2()
281 const __m256 vk5x01234567 = _mm256_load_ps(w + 96); in xnn_f32_dwconv_minmax_ukernel_up16x9__avx_acc2() local
282 vacc01234567p1 = _mm256_add_ps(vacc01234567p1, _mm256_mul_ps(vi5x01234567, vk5x01234567)); in xnn_f32_dwconv_minmax_ukernel_up16x9__avx_acc2()
|
| D | up8x9-minmax-fma3.c | 121 const __m256 vk5x01234567 = _mm256_load_ps(w + 48); in xnn_f32_dwconv_minmax_ukernel_up8x9__fma3() local
122 vacc01234567p0 = _mm256_fmadd_ps(vi5x01234567, vk5x01234567, vacc01234567p0); in xnn_f32_dwconv_minmax_ukernel_up8x9__fma3()
179 const __m256 vk5x01234567 = _mm256_load_ps(w + 48); in xnn_f32_dwconv_minmax_ukernel_up8x9__fma3() local
180 vacc01234567p0 = _mm256_fmadd_ps(vi5x01234567, vk5x01234567, vacc01234567p0); in xnn_f32_dwconv_minmax_ukernel_up8x9__fma3()
|
| D | up8x9-minmax-avx.c | 121 const __m256 vk5x01234567 = _mm256_load_ps(w + 48); in xnn_f32_dwconv_minmax_ukernel_up8x9__avx() local
122 vacc01234567p0 = _mm256_add_ps(vacc01234567p0, _mm256_mul_ps(vi5x01234567, vk5x01234567)); in xnn_f32_dwconv_minmax_ukernel_up8x9__avx()
179 const __m256 vk5x01234567 = _mm256_load_ps(w + 48); in xnn_f32_dwconv_minmax_ukernel_up8x9__avx() local
180 vacc01234567p0 = _mm256_add_ps(vacc01234567p0, _mm256_mul_ps(vi5x01234567, vk5x01234567)); in xnn_f32_dwconv_minmax_ukernel_up8x9__avx()
|
| D | up8x9-minmax-fma3-acc2.c | 121 const __m256 vk5x01234567 = _mm256_load_ps(w + 48); in xnn_f32_dwconv_minmax_ukernel_up8x9__fma3_acc2() local
122 vacc01234567p1 = _mm256_fmadd_ps(vi5x01234567, vk5x01234567, vacc01234567p1); in xnn_f32_dwconv_minmax_ukernel_up8x9__fma3_acc2()
181 const __m256 vk5x01234567 = _mm256_load_ps(w + 48); in xnn_f32_dwconv_minmax_ukernel_up8x9__fma3_acc2() local
182 vacc01234567p1 = _mm256_fmadd_ps(vi5x01234567, vk5x01234567, vacc01234567p1); in xnn_f32_dwconv_minmax_ukernel_up8x9__fma3_acc2()
|
| D | up8x9-minmax-avx-acc2.c | 121 const __m256 vk5x01234567 = _mm256_load_ps(w + 48); in xnn_f32_dwconv_minmax_ukernel_up8x9__avx_acc2() local
122 vacc01234567p1 = _mm256_add_ps(vacc01234567p1, _mm256_mul_ps(vi5x01234567, vk5x01234567)); in xnn_f32_dwconv_minmax_ukernel_up8x9__avx_acc2()
181 const __m256 vk5x01234567 = _mm256_load_ps(w + 48); in xnn_f32_dwconv_minmax_ukernel_up8x9__avx_acc2() local
182 vacc01234567p1 = _mm256_add_ps(vacc01234567p1, _mm256_mul_ps(vi5x01234567, vk5x01234567)); in xnn_f32_dwconv_minmax_ukernel_up8x9__avx_acc2()
|
| /external/XNNPACK/src/f16-dwconv/gen/ |
| D | up16x9-minmax-neonfp16arith-acc2.c | 128 const float16x8_t vk5x01234567 = vld1q_f16(w); w += 8; in xnn_f16_dwconv_minmax_ukernel_up16x9__neonfp16arith_acc2() local
130 vacc01234567p1 = vfmaq_f16(vacc01234567p1, vi5x01234567, vk5x01234567); in xnn_f16_dwconv_minmax_ukernel_up16x9__neonfp16arith_acc2()
191 const float16x8_t vk5x01234567 = vld1q_f16(w + 88); in xnn_f16_dwconv_minmax_ukernel_up16x9__neonfp16arith_acc2() local
192 vacc01234567p1 = vfmaq_f16(vacc01234567p1, vi5x01234567, vk5x01234567); in xnn_f16_dwconv_minmax_ukernel_up16x9__neonfp16arith_acc2()
239 const float16x8_t vk5x01234567 = vld1q_f16(w + 96); in xnn_f16_dwconv_minmax_ukernel_up16x9__neonfp16arith_acc2() local
240 vacc01234567p1 = vfmaq_f16(vacc01234567p1, vi5x01234567, vk5x01234567); in xnn_f16_dwconv_minmax_ukernel_up16x9__neonfp16arith_acc2()
|
| D | up16x9-minmax-neonfp16arith.c | 128 const float16x8_t vk5x01234567 = vld1q_f16(w); w += 8; in xnn_f16_dwconv_minmax_ukernel_up16x9__neonfp16arith() local
130 vacc01234567p0 = vfmaq_f16(vacc01234567p0, vi5x01234567, vk5x01234567); in xnn_f16_dwconv_minmax_ukernel_up16x9__neonfp16arith()
188 const float16x8_t vk5x01234567 = vld1q_f16(w + 88); in xnn_f16_dwconv_minmax_ukernel_up16x9__neonfp16arith() local
189 vacc01234567p0 = vfmaq_f16(vacc01234567p0, vi5x01234567, vk5x01234567); in xnn_f16_dwconv_minmax_ukernel_up16x9__neonfp16arith()
234 const float16x8_t vk5x01234567 = vld1q_f16(w + 96); in xnn_f16_dwconv_minmax_ukernel_up16x9__neonfp16arith() local
235 vacc01234567p0 = vfmaq_f16(vacc01234567p0, vi5x01234567, vk5x01234567); in xnn_f16_dwconv_minmax_ukernel_up16x9__neonfp16arith()
|
| D | up8x9-minmax-neonfp16arith.c | 111 const float16x8_t vk5x01234567 = vld1q_f16(w); w += 8; in xnn_f16_dwconv_minmax_ukernel_up8x9__neonfp16arith() local
112 vacc01234567p0 = vfmaq_f16(vacc01234567p0, vi5x01234567, vk5x01234567); in xnn_f16_dwconv_minmax_ukernel_up8x9__neonfp16arith()
157 const float16x8_t vk5x01234567 = vld1q_f16(w); w += 8; in xnn_f16_dwconv_minmax_ukernel_up8x9__neonfp16arith() local
158 vacc01234567p0 = vfmaq_f16(vacc01234567p0, vi5x01234567, vk5x01234567); in xnn_f16_dwconv_minmax_ukernel_up8x9__neonfp16arith()
|
| D | up8x9-minmax-neonfp16arith-acc2.c | 111 const float16x8_t vk5x01234567 = vld1q_f16(w); w += 8; in xnn_f16_dwconv_minmax_ukernel_up8x9__neonfp16arith_acc2() local
112 vacc01234567p1 = vfmaq_f16(vacc01234567p1, vi5x01234567, vk5x01234567); in xnn_f16_dwconv_minmax_ukernel_up8x9__neonfp16arith_acc2()
159 const float16x8_t vk5x01234567 = vld1q_f16(w); w += 8; in xnn_f16_dwconv_minmax_ukernel_up8x9__neonfp16arith_acc2() local
160 vacc01234567p1 = vfmaq_f16(vacc01234567p1, vi5x01234567, vk5x01234567); in xnn_f16_dwconv_minmax_ukernel_up8x9__neonfp16arith_acc2()
|
| D | up16x9-minmax-fma3.c | 141 const __m256 vk5x01234567 = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) (w + 96))); in xnn_f16_dwconv_minmax_ukernel_up16x9__fma3() local
143 …vacc01234567p0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi5x01234567, vk5x01234567, vacc0… in xnn_f16_dwconv_minmax_ukernel_up16x9__fma3()
221 const __m256 vk5x01234567 = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 96))); in xnn_f16_dwconv_minmax_ukernel_up16x9__fma3() local
222 …vacc01234567p0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi5x01234567, vk5x01234567, vacc0… in xnn_f16_dwconv_minmax_ukernel_up16x9__fma3()
284 const __m256 vk5x01234567 = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 96))); in xnn_f16_dwconv_minmax_ukernel_up16x9__fma3() local
285 …vacc01234567p0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi5x01234567, vk5x01234567, vacc0… in xnn_f16_dwconv_minmax_ukernel_up16x9__fma3()
|
| D | up8x9-minmax-fma3-acc2.c | 124 const __m256 vk5x01234567 = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) (w + 48))); in xnn_f16_dwconv_minmax_ukernel_up8x9__fma3_acc2() local
125 …vacc01234567p1 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi5x01234567, vk5x01234567, vacc0… in xnn_f16_dwconv_minmax_ukernel_up8x9__fma3_acc2()
189 const __m256 vk5x01234567 = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 48))); in xnn_f16_dwconv_minmax_ukernel_up8x9__fma3_acc2() local
190 …vacc01234567p1 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi5x01234567, vk5x01234567, vacc0… in xnn_f16_dwconv_minmax_ukernel_up8x9__fma3_acc2()
|
| D | up8x9-minmax-fma3.c | 124 const __m256 vk5x01234567 = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) (w + 48))); in xnn_f16_dwconv_minmax_ukernel_up8x9__fma3() local
125 …vacc01234567p0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi5x01234567, vk5x01234567, vacc0… in xnn_f16_dwconv_minmax_ukernel_up8x9__fma3()
187 const __m256 vk5x01234567 = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 48))); in xnn_f16_dwconv_minmax_ukernel_up8x9__fma3() local
188 …vacc01234567p0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi5x01234567, vk5x01234567, vacc0… in xnn_f16_dwconv_minmax_ukernel_up8x9__fma3()
|
| D | up32x9-minmax-neonfp16arith-acc2.c | 162 const float16x8_t vk5x01234567 = vld1q_f16(w); w += 8; in xnn_f16_dwconv_minmax_ukernel_up32x9__neonfp16arith_acc2() local
166 vacc01234567p1 = vfmaq_f16(vacc01234567p1, vi5x01234567, vk5x01234567); in xnn_f16_dwconv_minmax_ukernel_up32x9__neonfp16arith_acc2()
255 const float16x8_t vk5x01234567 = vld1q_f16(w + 184); in xnn_f16_dwconv_minmax_ukernel_up32x9__neonfp16arith_acc2() local
256 vacc01234567p1 = vfmaq_f16(vacc01234567p1, vi5x01234567, vk5x01234567); in xnn_f16_dwconv_minmax_ukernel_up32x9__neonfp16arith_acc2()
303 const float16x8_t vk5x01234567 = vld1q_f16(w + 192); in xnn_f16_dwconv_minmax_ukernel_up32x9__neonfp16arith_acc2() local
304 vacc01234567p1 = vfmaq_f16(vacc01234567p1, vi5x01234567, vk5x01234567); in xnn_f16_dwconv_minmax_ukernel_up32x9__neonfp16arith_acc2()
|
| /external/XNNPACK/src/qs8-dwconv/gen/ |
| D | up8x9-minmax-fp32-neonv8-mul16.c | 123 … const int16x8_t vk5x01234567 = vmovl_s8(vld1_s8(w)); w = (const void*) ((const int8_t*) w + 8); in xnn_qs8_dwconv_minmax_fp32_ukernel_up8x9__neonv8_mul16() local
125 vacc0123 = vmlal_s16(vacc0123, vget_low_s16(vi5x01234567), vget_low_s16(vk5x01234567)); in xnn_qs8_dwconv_minmax_fp32_ukernel_up8x9__neonv8_mul16()
126 vacc4567 = vmlal_s16(vacc4567, vget_high_s16(vi5x01234567), vget_high_s16(vk5x01234567)); in xnn_qs8_dwconv_minmax_fp32_ukernel_up8x9__neonv8_mul16()
206 const int16x8_t vk5x01234567 = vmovl_s8(vld1_s8((const void*) ((const int8_t*) w + 40))); in xnn_qs8_dwconv_minmax_fp32_ukernel_up8x9__neonv8_mul16() local
208 vacc0123 = vmlal_s16(vacc0123, vget_low_s16(vi5x01234567), vget_low_s16(vk5x01234567)); in xnn_qs8_dwconv_minmax_fp32_ukernel_up8x9__neonv8_mul16()
209 vacc4567 = vmlal_s16(vacc4567, vget_high_s16(vi5x01234567), vget_high_s16(vk5x01234567)); in xnn_qs8_dwconv_minmax_fp32_ukernel_up8x9__neonv8_mul16()
|
| D | up8x9-minmax-rndnu-neon-mul16.c | 124 … const int16x8_t vk5x01234567 = vmovl_s8(vld1_s8(w)); w = (const void*) ((const int8_t*) w + 8); in xnn_qs8_dwconv_minmax_rndnu_ukernel_up8x9__neon_mul16() local
126 vacc0123 = vmlal_s16(vacc0123, vget_low_s16(vi5x01234567), vget_low_s16(vk5x01234567)); in xnn_qs8_dwconv_minmax_rndnu_ukernel_up8x9__neon_mul16()
127 vacc4567 = vmlal_s16(vacc4567, vget_high_s16(vi5x01234567), vget_high_s16(vk5x01234567)); in xnn_qs8_dwconv_minmax_rndnu_ukernel_up8x9__neon_mul16()
207 const int16x8_t vk5x01234567 = vmovl_s8(vld1_s8((const void*) ((const int8_t*) w + 40))); in xnn_qs8_dwconv_minmax_rndnu_ukernel_up8x9__neon_mul16() local
209 vacc0123 = vmlal_s16(vacc0123, vget_low_s16(vi5x01234567), vget_low_s16(vk5x01234567)); in xnn_qs8_dwconv_minmax_rndnu_ukernel_up8x9__neon_mul16()
210 vacc4567 = vmlal_s16(vacc4567, vget_high_s16(vi5x01234567), vget_high_s16(vk5x01234567)); in xnn_qs8_dwconv_minmax_rndnu_ukernel_up8x9__neon_mul16()
|
| D | up8x9-minmax-fp32-neon-mul16.c | 123 … const int16x8_t vk5x01234567 = vmovl_s8(vld1_s8(w)); w = (const void*) ((const int8_t*) w + 8); in xnn_qs8_dwconv_minmax_fp32_ukernel_up8x9__neon_mul16() local
125 vacc0123 = vmlal_s16(vacc0123, vget_low_s16(vi5x01234567), vget_low_s16(vk5x01234567)); in xnn_qs8_dwconv_minmax_fp32_ukernel_up8x9__neon_mul16()
126 vacc4567 = vmlal_s16(vacc4567, vget_high_s16(vi5x01234567), vget_high_s16(vk5x01234567)); in xnn_qs8_dwconv_minmax_fp32_ukernel_up8x9__neon_mul16()
207 const int16x8_t vk5x01234567 = vmovl_s8(vld1_s8((const void*) ((const int8_t*) w + 40))); in xnn_qs8_dwconv_minmax_fp32_ukernel_up8x9__neon_mul16() local
209 vacc0123 = vmlal_s16(vacc0123, vget_low_s16(vi5x01234567), vget_low_s16(vk5x01234567)); in xnn_qs8_dwconv_minmax_fp32_ukernel_up8x9__neon_mul16()
210 vacc4567 = vmlal_s16(vacc4567, vget_high_s16(vi5x01234567), vget_high_s16(vk5x01234567)); in xnn_qs8_dwconv_minmax_fp32_ukernel_up8x9__neon_mul16()
|
| /external/XNNPACK/src/qc8-dwconv/gen/ |
| D | up8x9-minmax-fp32-neonv8-mul16.c | 122 … const int16x8_t vk5x01234567 = vmovl_s8(vld1_s8(w)); w = (const void*) ((const int8_t*) w + 8); in xnn_qc8_dwconv_minmax_fp32_ukernel_up8x9__neonv8_mul16() local
124 vacc0123 = vmlal_s16(vacc0123, vget_low_s16(vi5x01234567), vget_low_s16(vk5x01234567)); in xnn_qc8_dwconv_minmax_fp32_ukernel_up8x9__neonv8_mul16()
125 vacc4567 = vmlal_s16(vacc4567, vget_high_s16(vi5x01234567), vget_high_s16(vk5x01234567)); in xnn_qc8_dwconv_minmax_fp32_ukernel_up8x9__neonv8_mul16()
208 const int16x8_t vk5x01234567 = vmovl_s8(vld1_s8((const void*) ((const int8_t*) w + 40))); in xnn_qc8_dwconv_minmax_fp32_ukernel_up8x9__neonv8_mul16() local
210 vacc0123 = vmlal_s16(vacc0123, vget_low_s16(vi5x01234567), vget_low_s16(vk5x01234567)); in xnn_qc8_dwconv_minmax_fp32_ukernel_up8x9__neonv8_mul16()
211 vacc4567 = vmlal_s16(vacc4567, vget_high_s16(vi5x01234567), vget_high_s16(vk5x01234567)); in xnn_qc8_dwconv_minmax_fp32_ukernel_up8x9__neonv8_mul16()
|
| D | up8x9-minmax-fp32-neon-mul16.c | 122 … const int16x8_t vk5x01234567 = vmovl_s8(vld1_s8(w)); w = (const void*) ((const int8_t*) w + 8); in xnn_qc8_dwconv_minmax_fp32_ukernel_up8x9__neon_mul16() local
124 vacc0123 = vmlal_s16(vacc0123, vget_low_s16(vi5x01234567), vget_low_s16(vk5x01234567)); in xnn_qc8_dwconv_minmax_fp32_ukernel_up8x9__neon_mul16()
125 vacc4567 = vmlal_s16(vacc4567, vget_high_s16(vi5x01234567), vget_high_s16(vk5x01234567)); in xnn_qc8_dwconv_minmax_fp32_ukernel_up8x9__neon_mul16()
209 const int16x8_t vk5x01234567 = vmovl_s8(vld1_s8((const void*) ((const int8_t*) w + 40))); in xnn_qc8_dwconv_minmax_fp32_ukernel_up8x9__neon_mul16() local
211 vacc0123 = vmlal_s16(vacc0123, vget_low_s16(vi5x01234567), vget_low_s16(vk5x01234567)); in xnn_qc8_dwconv_minmax_fp32_ukernel_up8x9__neon_mul16()
212 vacc4567 = vmlal_s16(vacc4567, vget_high_s16(vi5x01234567), vget_high_s16(vk5x01234567)); in xnn_qc8_dwconv_minmax_fp32_ukernel_up8x9__neon_mul16()
|
| D | up8x9-minmax-fp32-sse2-mul16-add16.c | 145 …const __m128i vk5x01234567 = _mm_loadl_epi64((const __m128i*) ((uintptr_t) w + 8 * sizeof(int32_t)… in xnn_qc8_dwconv_minmax_fp32_ukernel_up8x9__sse2_mul16_add16() local
149 … const __m128i vxk5x01234567 = _mm_srai_epi16(_mm_unpacklo_epi8(vk5x01234567, vk5x01234567), 8); in xnn_qc8_dwconv_minmax_fp32_ukernel_up8x9__sse2_mul16_add16()
281 …const __m128i vk5x01234567 = _mm_loadl_epi64((const __m128i*) ((uintptr_t) w + 8 * sizeof(int32_t)… in xnn_qc8_dwconv_minmax_fp32_ukernel_up8x9__sse2_mul16_add16() local
284 … const __m128i vxk5x01234567 = _mm_srai_epi16(_mm_unpacklo_epi8(vk5x01234567, vk5x01234567), 8); in xnn_qc8_dwconv_minmax_fp32_ukernel_up8x9__sse2_mul16_add16()
|
| /external/XNNPACK/src/qu8-dwconv/gen/ |
| D | up8x9-minmax-fp32-neonv8-mul16.c | 124 …const int16x8_t vk5x01234567 = vreinterpretq_s16_u16(vsubl_u8(vld1_u8(w), vkernel_zero_point)); w … in xnn_qu8_dwconv_minmax_fp32_ukernel_up8x9__neonv8_mul16() local
126 vacc0123 = vmlal_s16(vacc0123, vget_low_s16(vi5x01234567), vget_low_s16(vk5x01234567)); in xnn_qu8_dwconv_minmax_fp32_ukernel_up8x9__neonv8_mul16()
127 vacc4567 = vmlal_s16(vacc4567, vget_high_s16(vi5x01234567), vget_high_s16(vk5x01234567)); in xnn_qu8_dwconv_minmax_fp32_ukernel_up8x9__neonv8_mul16()
207 …const int16x8_t vk5x01234567 = vreinterpretq_s16_u16(vsubl_u8(vld1_u8((const void*) ((const uint8_… in xnn_qu8_dwconv_minmax_fp32_ukernel_up8x9__neonv8_mul16() local
209 vacc0123 = vmlal_s16(vacc0123, vget_low_s16(vi5x01234567), vget_low_s16(vk5x01234567)); in xnn_qu8_dwconv_minmax_fp32_ukernel_up8x9__neonv8_mul16()
210 vacc4567 = vmlal_s16(vacc4567, vget_high_s16(vi5x01234567), vget_high_s16(vk5x01234567)); in xnn_qu8_dwconv_minmax_fp32_ukernel_up8x9__neonv8_mul16()
|
| D | up8x9-minmax-rndnu-neon-mul16.c | 125 …const int16x8_t vk5x01234567 = vreinterpretq_s16_u16(vsubl_u8(vld1_u8(w), vkernel_zero_point)); w … in xnn_qu8_dwconv_minmax_rndnu_ukernel_up8x9__neon_mul16() local
127 vacc0123 = vmlal_s16(vacc0123, vget_low_s16(vi5x01234567), vget_low_s16(vk5x01234567)); in xnn_qu8_dwconv_minmax_rndnu_ukernel_up8x9__neon_mul16()
128 vacc4567 = vmlal_s16(vacc4567, vget_high_s16(vi5x01234567), vget_high_s16(vk5x01234567)); in xnn_qu8_dwconv_minmax_rndnu_ukernel_up8x9__neon_mul16()
208 …const int16x8_t vk5x01234567 = vreinterpretq_s16_u16(vsubl_u8(vld1_u8((const void*) ((const uint8_… in xnn_qu8_dwconv_minmax_rndnu_ukernel_up8x9__neon_mul16() local
210 vacc0123 = vmlal_s16(vacc0123, vget_low_s16(vi5x01234567), vget_low_s16(vk5x01234567)); in xnn_qu8_dwconv_minmax_rndnu_ukernel_up8x9__neon_mul16()
211 vacc4567 = vmlal_s16(vacc4567, vget_high_s16(vi5x01234567), vget_high_s16(vk5x01234567)); in xnn_qu8_dwconv_minmax_rndnu_ukernel_up8x9__neon_mul16()
|
| D | up8x9-minmax-fp32-neon-mul16.c | 124 …const int16x8_t vk5x01234567 = vreinterpretq_s16_u16(vsubl_u8(vld1_u8(w), vkernel_zero_point)); w … in xnn_qu8_dwconv_minmax_fp32_ukernel_up8x9__neon_mul16() local
126 vacc0123 = vmlal_s16(vacc0123, vget_low_s16(vi5x01234567), vget_low_s16(vk5x01234567)); in xnn_qu8_dwconv_minmax_fp32_ukernel_up8x9__neon_mul16()
127 vacc4567 = vmlal_s16(vacc4567, vget_high_s16(vi5x01234567), vget_high_s16(vk5x01234567)); in xnn_qu8_dwconv_minmax_fp32_ukernel_up8x9__neon_mul16()
208 …const int16x8_t vk5x01234567 = vreinterpretq_s16_u16(vsubl_u8(vld1_u8((const void*) ((const uint8_… in xnn_qu8_dwconv_minmax_fp32_ukernel_up8x9__neon_mul16() local
210 vacc0123 = vmlal_s16(vacc0123, vget_low_s16(vi5x01234567), vget_low_s16(vk5x01234567)); in xnn_qu8_dwconv_minmax_fp32_ukernel_up8x9__neon_mul16()
211 vacc4567 = vmlal_s16(vacc4567, vget_high_s16(vi5x01234567), vget_high_s16(vk5x01234567)); in xnn_qu8_dwconv_minmax_fp32_ukernel_up8x9__neon_mul16()
|