83       __m128i vacc0123 = _mm_loadu_si128((const __m128i*) w);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
84       __m128i vacc4567 = _mm_loadu_si128((const __m128i*) ((uintptr_t) w + 4 * sizeof(int32_t)));  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
85       __m128i vacc89AB = _mm_loadu_si128((const __m128i*) ((uintptr_t) w + 8 * sizeof(int32_t)));  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
86       __m128i vaccCDEF = _mm_loadu_si128((const __m128i*) ((uintptr_t) w + 12 * sizeof(int32_t)));  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
89       const __m128i vi0x01234567 = _mm_loadl_epi64((const __m128i*) i0);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
90       const __m128i vxi0x01234567 = _mm_cvtepi8_epi16(vi0x01234567);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
91 …const __m128i vk0x01234567 = _mm_loadl_epi64((const __m128i*) ((uintptr_t) w + 16 * sizeof(int32_t…  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
92       const __m128i vxk0x01234567 = _mm_cvtepi8_epi16(vk0x01234567);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
93       const __m128i vi0x89ABCDEF = _mm_loadl_epi64((const __m128i*) (i0 + 8));  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
94       const __m128i vxi0x89ABCDEF = _mm_cvtepi8_epi16(vi0x89ABCDEF);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
95 …const __m128i vk0x89ABCDEF = _mm_loadl_epi64((const __m128i*) ((uintptr_t) w + 16 * sizeof(int32_t…  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
96       const __m128i vxk0x89ABCDEF = _mm_cvtepi8_epi16(vk0x89ABCDEF);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
100       const __m128i vp0x01234567lo = _mm_mullo_epi16(vxi0x01234567, vxk0x01234567);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
101       const __m128i vp0x01234567hi = _mm_mulhi_epi16(vxi0x01234567, vxk0x01234567);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
102       const __m128i vp0x89ABCDEFlo = _mm_mullo_epi16(vxi0x89ABCDEF, vxk0x89ABCDEF);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
103       const __m128i vp0x89ABCDEFhi = _mm_mulhi_epi16(vxi0x89ABCDEF, vxk0x89ABCDEF);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
110       const __m128i vi1x01234567 = _mm_loadl_epi64((const __m128i*) i1);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
111       const __m128i vxi1x01234567 = _mm_cvtepi8_epi16(vi1x01234567);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
112 …const __m128i vk1x01234567 = _mm_loadl_epi64((const __m128i*) ((uintptr_t) w + 16 * sizeof(int32_t…  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
113       const __m128i vxk1x01234567 = _mm_cvtepi8_epi16(vk1x01234567);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
114       const __m128i vi1x89ABCDEF = _mm_loadl_epi64((const __m128i*) (i1 + 8));  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
115       const __m128i vxi1x89ABCDEF = _mm_cvtepi8_epi16(vi1x89ABCDEF);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
116 …const __m128i vk1x89ABCDEF = _mm_loadl_epi64((const __m128i*) ((uintptr_t) w + 16 * sizeof(int32_t…  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
117       const __m128i vxk1x89ABCDEF = _mm_cvtepi8_epi16(vk1x89ABCDEF);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
121       const __m128i vp1x01234567lo = _mm_mullo_epi16(vxi1x01234567, vxk1x01234567);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
122       const __m128i vp1x01234567hi = _mm_mulhi_epi16(vxi1x01234567, vxk1x01234567);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
123       const __m128i vp1x89ABCDEFlo = _mm_mullo_epi16(vxi1x89ABCDEF, vxk1x89ABCDEF);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
124       const __m128i vp1x89ABCDEFhi = _mm_mulhi_epi16(vxi1x89ABCDEF, vxk1x89ABCDEF);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
131       const __m128i vi2x01234567 = _mm_loadl_epi64((const __m128i*) i2);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
132       const __m128i vxi2x01234567 = _mm_cvtepi8_epi16(vi2x01234567);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
133 …const __m128i vk2x01234567 = _mm_loadl_epi64((const __m128i*) ((uintptr_t) w + 16 * sizeof(int32_t…  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
134       const __m128i vxk2x01234567 = _mm_cvtepi8_epi16(vk2x01234567);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
135       const __m128i vi2x89ABCDEF = _mm_loadl_epi64((const __m128i*) (i2 + 8));  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
136       const __m128i vxi2x89ABCDEF = _mm_cvtepi8_epi16(vi2x89ABCDEF);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
137 …const __m128i vk2x89ABCDEF = _mm_loadl_epi64((const __m128i*) ((uintptr_t) w + 16 * sizeof(int32_t…  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
138       const __m128i vxk2x89ABCDEF = _mm_cvtepi8_epi16(vk2x89ABCDEF);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
142       const __m128i vp2x01234567lo = _mm_mullo_epi16(vxi2x01234567, vxk2x01234567);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
143       const __m128i vp2x01234567hi = _mm_mulhi_epi16(vxi2x01234567, vxk2x01234567);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
144       const __m128i vp2x89ABCDEFlo = _mm_mullo_epi16(vxi2x89ABCDEF, vxk2x89ABCDEF);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
145       const __m128i vp2x89ABCDEFhi = _mm_mulhi_epi16(vxi2x89ABCDEF, vxk2x89ABCDEF);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
152       const __m128i vi3x01234567 = _mm_loadl_epi64((const __m128i*) i3);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
153       const __m128i vxi3x01234567 = _mm_cvtepi8_epi16(vi3x01234567);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
154 …const __m128i vk3x01234567 = _mm_loadl_epi64((const __m128i*) ((uintptr_t) w + 16 * sizeof(int32_t…  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
155       const __m128i vxk3x01234567 = _mm_cvtepi8_epi16(vk3x01234567);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
156       const __m128i vi3x89ABCDEF = _mm_loadl_epi64((const __m128i*) (i3 + 8));  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
157       const __m128i vxi3x89ABCDEF = _mm_cvtepi8_epi16(vi3x89ABCDEF);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
158 …const __m128i vk3x89ABCDEF = _mm_loadl_epi64((const __m128i*) ((uintptr_t) w + 16 * sizeof(int32_t…  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
159       const __m128i vxk3x89ABCDEF = _mm_cvtepi8_epi16(vk3x89ABCDEF);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
163       const __m128i vp3x01234567lo = _mm_mullo_epi16(vxi3x01234567, vxk3x01234567);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
164       const __m128i vp3x01234567hi = _mm_mulhi_epi16(vxi3x01234567, vxk3x01234567);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
165       const __m128i vp3x89ABCDEFlo = _mm_mullo_epi16(vxi3x89ABCDEF, vxk3x89ABCDEF);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
166       const __m128i vp3x89ABCDEFhi = _mm_mulhi_epi16(vxi3x89ABCDEF, vxk3x89ABCDEF);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
173       const __m128i vi4x01234567 = _mm_loadl_epi64((const __m128i*) i4);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
174       const __m128i vxi4x01234567 = _mm_cvtepi8_epi16(vi4x01234567);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
175 …const __m128i vk4x01234567 = _mm_loadl_epi64((const __m128i*) ((uintptr_t) w + 16 * sizeof(int32_t…  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
176       const __m128i vxk4x01234567 = _mm_cvtepi8_epi16(vk4x01234567);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
177       const __m128i vi4x89ABCDEF = _mm_loadl_epi64((const __m128i*) (i4 + 8));  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
178       const __m128i vxi4x89ABCDEF = _mm_cvtepi8_epi16(vi4x89ABCDEF);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
179 …const __m128i vk4x89ABCDEF = _mm_loadl_epi64((const __m128i*) ((uintptr_t) w + 16 * sizeof(int32_t…  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
180       const __m128i vxk4x89ABCDEF = _mm_cvtepi8_epi16(vk4x89ABCDEF);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
184       const __m128i vp4x01234567lo = _mm_mullo_epi16(vxi4x01234567, vxk4x01234567);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
185       const __m128i vp4x01234567hi = _mm_mulhi_epi16(vxi4x01234567, vxk4x01234567);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
186       const __m128i vp4x89ABCDEFlo = _mm_mullo_epi16(vxi4x89ABCDEF, vxk4x89ABCDEF);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
187       const __m128i vp4x89ABCDEFhi = _mm_mulhi_epi16(vxi4x89ABCDEF, vxk4x89ABCDEF);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
194       const __m128i vi5x01234567 = _mm_loadl_epi64((const __m128i*) i5);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
195       const __m128i vxi5x01234567 = _mm_cvtepi8_epi16(vi5x01234567);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
196 …const __m128i vk5x01234567 = _mm_loadl_epi64((const __m128i*) ((uintptr_t) w + 16 * sizeof(int32_t…  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
197       const __m128i vxk5x01234567 = _mm_cvtepi8_epi16(vk5x01234567);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
198       const __m128i vi5x89ABCDEF = _mm_loadl_epi64((const __m128i*) (i5 + 8));  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
199       const __m128i vxi5x89ABCDEF = _mm_cvtepi8_epi16(vi5x89ABCDEF);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
200 …const __m128i vk5x89ABCDEF = _mm_loadl_epi64((const __m128i*) ((uintptr_t) w + 16 * sizeof(int32_t…  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
201       const __m128i vxk5x89ABCDEF = _mm_cvtepi8_epi16(vk5x89ABCDEF);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
205       const __m128i vp5x01234567lo = _mm_mullo_epi16(vxi5x01234567, vxk5x01234567);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
206       const __m128i vp5x01234567hi = _mm_mulhi_epi16(vxi5x01234567, vxk5x01234567);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
207       const __m128i vp5x89ABCDEFlo = _mm_mullo_epi16(vxi5x89ABCDEF, vxk5x89ABCDEF);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
208       const __m128i vp5x89ABCDEFhi = _mm_mulhi_epi16(vxi5x89ABCDEF, vxk5x89ABCDEF);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
215       const __m128i vi6x01234567 = _mm_loadl_epi64((const __m128i*) i6);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
216       const __m128i vxi6x01234567 = _mm_cvtepi8_epi16(vi6x01234567);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
217 …const __m128i vk6x01234567 = _mm_loadl_epi64((const __m128i*) ((uintptr_t) w + 16 * sizeof(int32_t…  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
218       const __m128i vxk6x01234567 = _mm_cvtepi8_epi16(vk6x01234567);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
219       const __m128i vi6x89ABCDEF = _mm_loadl_epi64((const __m128i*) (i6 + 8));  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
220       const __m128i vxi6x89ABCDEF = _mm_cvtepi8_epi16(vi6x89ABCDEF);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
221 …const __m128i vk6x89ABCDEF = _mm_loadl_epi64((const __m128i*) ((uintptr_t) w + 16 * sizeof(int32_t…  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
222       const __m128i vxk6x89ABCDEF = _mm_cvtepi8_epi16(vk6x89ABCDEF);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
226       const __m128i vp6x01234567lo = _mm_mullo_epi16(vxi6x01234567, vxk6x01234567);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
227       const __m128i vp6x01234567hi = _mm_mulhi_epi16(vxi6x01234567, vxk6x01234567);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
228       const __m128i vp6x89ABCDEFlo = _mm_mullo_epi16(vxi6x89ABCDEF, vxk6x89ABCDEF);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
229       const __m128i vp6x89ABCDEFhi = _mm_mulhi_epi16(vxi6x89ABCDEF, vxk6x89ABCDEF);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
236       const __m128i vi7x01234567 = _mm_loadl_epi64((const __m128i*) i7);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
237       const __m128i vxi7x01234567 = _mm_cvtepi8_epi16(vi7x01234567);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
238 …const __m128i vk7x01234567 = _mm_loadl_epi64((const __m128i*) ((uintptr_t) w + 16 * sizeof(int32_t…  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
239       const __m128i vxk7x01234567 = _mm_cvtepi8_epi16(vk7x01234567);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
240       const __m128i vi7x89ABCDEF = _mm_loadl_epi64((const __m128i*) (i7 + 8));  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
241       const __m128i vxi7x89ABCDEF = _mm_cvtepi8_epi16(vi7x89ABCDEF);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
242 …const __m128i vk7x89ABCDEF = _mm_loadl_epi64((const __m128i*) ((uintptr_t) w + 16 * sizeof(int32_t…  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
243       const __m128i vxk7x89ABCDEF = _mm_cvtepi8_epi16(vk7x89ABCDEF);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
247       const __m128i vp7x01234567lo = _mm_mullo_epi16(vxi7x01234567, vxk7x01234567);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
248       const __m128i vp7x01234567hi = _mm_mulhi_epi16(vxi7x01234567, vxk7x01234567);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
249       const __m128i vp7x89ABCDEFlo = _mm_mullo_epi16(vxi7x89ABCDEF, vxk7x89ABCDEF);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
250       const __m128i vp7x89ABCDEFhi = _mm_mulhi_epi16(vxi7x89ABCDEF, vxk7x89ABCDEF);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
257       const __m128i vi8x01234567 = _mm_loadl_epi64((const __m128i*) i8);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
258       const __m128i vxi8x01234567 = _mm_cvtepi8_epi16(vi8x01234567);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
259 …const __m128i vk8x01234567 = _mm_loadl_epi64((const __m128i*) ((uintptr_t) w + 16 * sizeof(int32_t…  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
260       const __m128i vxk8x01234567 = _mm_cvtepi8_epi16(vk8x01234567);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
261       const __m128i vi8x89ABCDEF = _mm_loadl_epi64((const __m128i*) (i8 + 8));  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
262       const __m128i vxi8x89ABCDEF = _mm_cvtepi8_epi16(vi8x89ABCDEF);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
263 …const __m128i vk8x89ABCDEF = _mm_loadl_epi64((const __m128i*) ((uintptr_t) w + 16 * sizeof(int32_t…  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
264       const __m128i vxk8x89ABCDEF = _mm_cvtepi8_epi16(vk8x89ABCDEF);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
268       const __m128i vp8x01234567lo = _mm_mullo_epi16(vxi8x01234567, vxk8x01234567);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
269       const __m128i vp8x01234567hi = _mm_mulhi_epi16(vxi8x01234567, vxk8x01234567);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
270       const __m128i vp8x89ABCDEFlo = _mm_mullo_epi16(vxi8x89ABCDEF, vxk8x89ABCDEF);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
271       const __m128i vp8x89ABCDEFhi = _mm_mulhi_epi16(vxi8x89ABCDEF, vxk8x89ABCDEF);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
280       const __m128i vmultiplier = _mm_load_si128((const __m128i*) params->sse2.multiplier);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
281       const __m128i vrounding = _mm_load_si128((const __m128i*) params->sse2.rounding);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
283       const __m128i vacc13 = _mm_shuffle_epi32(vacc0123, _MM_SHUFFLE(3, 3, 1, 1));  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
284       const __m128i vprod02 = _mm_add_epi64(_mm_mul_epi32(vacc0123, vmultiplier), vrounding);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
285       const __m128i vprod13 = _mm_add_epi64(_mm_mul_epi32(vacc13, vmultiplier), vrounding);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
286       const __m128i vacc57 = _mm_shuffle_epi32(vacc4567, _MM_SHUFFLE(3, 3, 1, 1));  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
287       const __m128i vprod46 = _mm_add_epi64(_mm_mul_epi32(vacc4567, vmultiplier), vrounding);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
288       const __m128i vprod57 = _mm_add_epi64(_mm_mul_epi32(vacc57, vmultiplier), vrounding);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
289       const __m128i vacc9B = _mm_shuffle_epi32(vacc89AB, _MM_SHUFFLE(3, 3, 1, 1));  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
290       const __m128i vprod8A = _mm_add_epi64(_mm_mul_epi32(vacc89AB, vmultiplier), vrounding);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
291       const __m128i vprod9B = _mm_add_epi64(_mm_mul_epi32(vacc9B, vmultiplier), vrounding);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
292       const __m128i vaccDF = _mm_shuffle_epi32(vaccCDEF, _MM_SHUFFLE(3, 3, 1, 1));  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
293       const __m128i vprodCE = _mm_add_epi64(_mm_mul_epi32(vaccCDEF, vmultiplier), vrounding);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
294       const __m128i vprodDF = _mm_add_epi64(_mm_mul_epi32(vaccDF, vmultiplier), vrounding);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
296       const __m128i vq31prod02 = _mm_srli_epi64(vprod02, 31);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
297       const __m128i vq31prod13 = _mm_add_epi64(vprod13, vprod13);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
298       const __m128i vq31prod46 = _mm_srli_epi64(vprod46, 31);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
299       const __m128i vq31prod57 = _mm_add_epi64(vprod57, vprod57);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
300       const __m128i vq31prod8A = _mm_srli_epi64(vprod8A, 31);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
301       const __m128i vq31prod9B = _mm_add_epi64(vprod9B, vprod9B);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
302       const __m128i vq31prodCE = _mm_srli_epi64(vprodCE, 31);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
303       const __m128i vq31prodDF = _mm_add_epi64(vprodDF, vprodDF);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
305       const __m128i vq31prod0123 = _mm_blend_epi16(vq31prod02, vq31prod13, 0xCC);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
306       const __m128i vq31prod4567 = _mm_blend_epi16(vq31prod46, vq31prod57, 0xCC);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
307       const __m128i vq31prod89AB = _mm_blend_epi16(vq31prod8A, vq31prod9B, 0xCC);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
308       const __m128i vq31prodCDEF = _mm_blend_epi16(vq31prodCE, vq31prodDF, 0xCC);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
310       const __m128i vremainder_mask = _mm_load_si128((const __m128i*) params->sse2.remainder_mask);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
311       const __m128i vrem0123 =  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
313       const __m128i vrem4567 =  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
315       const __m128i vrem89AB =  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
317       const __m128i vremCDEF =  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
320 …const __m128i vremainder_threshold = _mm_load_si128((const __m128i*) params->sse2.remainder_thresh…  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
321       const __m128i vshift = _mm_load_si128((const __m128i*) params->sse2.shift);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
331 …const __m128i voutput_zero_point = _mm_load_si128((const __m128i*) params->sse2.output_zero_point);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
332 …    __m128i vout01234567 = _mm_adds_epi16(_mm_packs_epi32(vacc0123, vacc4567), voutput_zero_point);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
333 …    __m128i vout89ABCDEF = _mm_adds_epi16(_mm_packs_epi32(vacc89AB, vaccCDEF), voutput_zero_point);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
335       const __m128i voutput_min = _mm_load_si128((const __m128i*) params->sse2.output_min);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
336       const __m128i voutput_max = _mm_load_si128((const __m128i*) params->sse2.output_max);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
340       __m128i vout0123456789ABCDEF = _mm_packs_epi16(vout01234567, vout89ABCDEF);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
342       _mm_storeu_si128((__m128i*) output, vout0123456789ABCDEF);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
348         __m128i vacc0123 = _mm_loadu_si128((const __m128i*) w);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
349         __m128i vacc4567 = _mm_loadu_si128((const __m128i*) ((uintptr_t) w + 4 * sizeof(int32_t)));  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
352         const __m128i vi0x01234567 = _mm_loadl_epi64((const __m128i*) i0);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
353         const __m128i vxi0x01234567 = _mm_cvtepi8_epi16(vi0x01234567);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
354         const __m128i vk0x01234567 = _mm_loadl_epi64((const __m128i*) k);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
355         const __m128i vxk0x01234567 = _mm_cvtepi8_epi16(vk0x01234567);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
359         const __m128i vp0x01234567lo = _mm_mullo_epi16(vxi0x01234567, vxk0x01234567);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
360         const __m128i vp0x01234567hi = _mm_mulhi_epi16(vxi0x01234567, vxk0x01234567);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
365         const __m128i vi1x01234567 = _mm_loadl_epi64((const __m128i*) i1);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
366         const __m128i vxi1x01234567 = _mm_cvtepi8_epi16(vi1x01234567);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
367         const __m128i vk1x01234567 = _mm_loadl_epi64((const __m128i*) (k + 16));  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
368         const __m128i vxk1x01234567 = _mm_cvtepi8_epi16(vk1x01234567);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
372         const __m128i vp1x01234567lo = _mm_mullo_epi16(vxi1x01234567, vxk1x01234567);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
373         const __m128i vp1x01234567hi = _mm_mulhi_epi16(vxi1x01234567, vxk1x01234567);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
378         const __m128i vi2x01234567 = _mm_loadl_epi64((const __m128i*) i2);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
379         const __m128i vxi2x01234567 = _mm_cvtepi8_epi16(vi2x01234567);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
380         const __m128i vk2x01234567 = _mm_loadl_epi64((const __m128i*) (k + 32));  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
381         const __m128i vxk2x01234567 = _mm_cvtepi8_epi16(vk2x01234567);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
385         const __m128i vp2x01234567lo = _mm_mullo_epi16(vxi2x01234567, vxk2x01234567);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
386         const __m128i vp2x01234567hi = _mm_mulhi_epi16(vxi2x01234567, vxk2x01234567);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
391         const __m128i vi3x01234567 = _mm_loadl_epi64((const __m128i*) i3);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
392         const __m128i vxi3x01234567 = _mm_cvtepi8_epi16(vi3x01234567);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
393         const __m128i vk3x01234567 = _mm_loadl_epi64((const __m128i*) (k + 48));  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
394         const __m128i vxk3x01234567 = _mm_cvtepi8_epi16(vk3x01234567);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
398         const __m128i vp3x01234567lo = _mm_mullo_epi16(vxi3x01234567, vxk3x01234567);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
399         const __m128i vp3x01234567hi = _mm_mulhi_epi16(vxi3x01234567, vxk3x01234567);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
404         const __m128i vi4x01234567 = _mm_loadl_epi64((const __m128i*) i4);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
405         const __m128i vxi4x01234567 = _mm_cvtepi8_epi16(vi4x01234567);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
406         const __m128i vk4x01234567 = _mm_loadl_epi64((const __m128i*) (k + 64));  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
407         const __m128i vxk4x01234567 = _mm_cvtepi8_epi16(vk4x01234567);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
411         const __m128i vp4x01234567lo = _mm_mullo_epi16(vxi4x01234567, vxk4x01234567);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
412         const __m128i vp4x01234567hi = _mm_mulhi_epi16(vxi4x01234567, vxk4x01234567);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
417         const __m128i vi5x01234567 = _mm_loadl_epi64((const __m128i*) i5);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
418         const __m128i vxi5x01234567 = _mm_cvtepi8_epi16(vi5x01234567);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
419         const __m128i vk5x01234567 = _mm_loadl_epi64((const __m128i*) (k + 80));  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
420         const __m128i vxk5x01234567 = _mm_cvtepi8_epi16(vk5x01234567);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
424         const __m128i vp5x01234567lo = _mm_mullo_epi16(vxi5x01234567, vxk5x01234567);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
425         const __m128i vp5x01234567hi = _mm_mulhi_epi16(vxi5x01234567, vxk5x01234567);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
430         const __m128i vi6x01234567 = _mm_loadl_epi64((const __m128i*) i6);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
431         const __m128i vxi6x01234567 = _mm_cvtepi8_epi16(vi6x01234567);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
432         const __m128i vk6x01234567 = _mm_loadl_epi64((const __m128i*) (k + 96));  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
433         const __m128i vxk6x01234567 = _mm_cvtepi8_epi16(vk6x01234567);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
437         const __m128i vp6x01234567lo = _mm_mullo_epi16(vxi6x01234567, vxk6x01234567);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
438         const __m128i vp6x01234567hi = _mm_mulhi_epi16(vxi6x01234567, vxk6x01234567);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
443         const __m128i vi7x01234567 = _mm_loadl_epi64((const __m128i*) i7);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
444         const __m128i vxi7x01234567 = _mm_cvtepi8_epi16(vi7x01234567);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
445         const __m128i vk7x01234567 = _mm_loadl_epi64((const __m128i*) (k + 112));  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
446         const __m128i vxk7x01234567 = _mm_cvtepi8_epi16(vk7x01234567);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
450         const __m128i vp7x01234567lo = _mm_mullo_epi16(vxi7x01234567, vxk7x01234567);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
451         const __m128i vp7x01234567hi = _mm_mulhi_epi16(vxi7x01234567, vxk7x01234567);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
456         const __m128i vi8x01234567 = _mm_loadl_epi64((const __m128i*) i8);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
457         const __m128i vxi8x01234567 = _mm_cvtepi8_epi16(vi8x01234567);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
458         const __m128i vk8x01234567 = _mm_loadl_epi64((const __m128i*) (k + 128));  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
459         const __m128i vxk8x01234567 = _mm_cvtepi8_epi16(vk8x01234567);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
463         const __m128i vp8x01234567lo = _mm_mullo_epi16(vxi8x01234567, vxk8x01234567);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
464         const __m128i vp8x01234567hi = _mm_mulhi_epi16(vxi8x01234567, vxk8x01234567);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
472         const __m128i vmultiplier = _mm_load_si128((const __m128i*) params->sse2.multiplier);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
473         const __m128i vrounding = _mm_load_si128((const __m128i*) params->sse2.rounding);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
475         const __m128i vacc13 = _mm_shuffle_epi32(vacc0123, _MM_SHUFFLE(3, 3, 1, 1));  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
476         const __m128i vacc57 = _mm_shuffle_epi32(vacc4567, _MM_SHUFFLE(3, 3, 1, 1));  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
478         const __m128i vprod02 = _mm_add_epi64(_mm_mul_epi32(vacc0123, vmultiplier), vrounding);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
479         const __m128i vprod46 = _mm_add_epi64(_mm_mul_epi32(vacc4567, vmultiplier), vrounding);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
481         const __m128i vprod13 = _mm_add_epi64(_mm_mul_epi32(vacc13, vmultiplier), vrounding);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
482         const __m128i vprod57 = _mm_add_epi64(_mm_mul_epi32(vacc57, vmultiplier), vrounding);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
484         const __m128i vq31prod02 = _mm_srli_epi64(vprod02, 31);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
485         const __m128i vq31prod13 = _mm_add_epi64(vprod13, vprod13);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
486         const __m128i vq31prod46 = _mm_srli_epi64(vprod46, 31);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
487         const __m128i vq31prod57 = _mm_add_epi64(vprod57, vprod57);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
489         const __m128i vq31prod0123 = _mm_blend_epi16(vq31prod02, vq31prod13, 0xCC);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
490         const __m128i vq31prod4567 = _mm_blend_epi16(vq31prod46, vq31prod57, 0xCC);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
492 …      const __m128i vremainder_mask = _mm_load_si128((const __m128i*) params->sse2.remainder_mask);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
493         const __m128i vrem0123 =  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
495         const __m128i vrem4567 =  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
498 …const __m128i vremainder_threshold = _mm_load_si128((const __m128i*) params->sse2.remainder_thresh…  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
499         const __m128i vshift = _mm_load_si128((const __m128i*) params->sse2.shift);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
505 …const __m128i voutput_zero_point = _mm_load_si128((const __m128i*) params->sse2.output_zero_point);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
506 …    __m128i vout01234567 = _mm_adds_epi16(_mm_packs_epi32(vacc0123, vacc4567), voutput_zero_point);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
508         const __m128i voutput_min = _mm_load_si128((const __m128i*) params->sse2.output_min);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
509         const __m128i voutput_max = _mm_load_si128((const __m128i*) params->sse2.output_max);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
512         __m128i vout0123456701234567 = _mm_packs_epi16(vout01234567, vout01234567);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()
515           _mm_storel_epi64((__m128i*) output, vout0123456701234567);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__sse41_mul16()