lib/Headers/avxvnniint8intrin.h

26 ///    corresponding signed 8-bit integers in \a __B, producing 4 intermediate
33 /// _mm_dpbssd_epi32(__m128i __W, __m128i __A, __m128i __B);
40 /// \param __B
47 /// 	tmp1.word := SignExtend16(__A.byte[4*j]) * SignExtend16(__B.byte[4*j])
48 /// 	tmp2.word := SignExtend16(__A.byte[4*j+1]) * SignExtend16(__B.byte[4*j+1])
49 /// 	tmp3.word := SignExtend16(__A.byte[4*j+2]) * SignExtend16(__B.byte[4*j+2])
50 /// 	tmp4.word := SignExtend16(__A.byte[4*j+3]) * SignExtend16(__B.byte[4*j+3])
57                                                                  __m128i __B) {
59                                              (__v4si)__B);
63 ///    corresponding signed 8-bit integers in \a __B, producing 4 intermediate
70 /// _mm256_dpbssd_epi32(__m256i __W, __m256i __A, __m256i __B);
77 /// \param __B
84 /// 	tmp1.word := SignExtend16(__A.byte[4*j]) * SignExtend16(__B.byte[4*j])
85 /// 	tmp2.word := SignExtend16(__A.byte[4*j+1]) * SignExtend16(__B.byte[4*j+1])
86 /// 	tmp3.word := SignExtend16(__A.byte[4*j+2]) * SignExtend16(__B.byte[4*j+2])
87 /// 	tmp4.word := SignExtend16(__A.byte[4*j+3]) * SignExtend16(__B.byte[4*j+3])
93 _mm256_dpbssd_epi32(__m256i __W, __m256i __A, __m256i __B) {
95                                              (__v8si)__B);
99 ///    corresponding signed 8-bit integers in \a __B, producing 4 intermediate
107 /// _mm_dpbssds_epi32( __m128i __W, __m128i __A, __m128i __B);
114 /// \param __B
121 /// 	tmp1.word := SignExtend16(__A.byte[4*j]) * SignExtend16(__B.byte[4*j])
122 /// 	tmp2.word := SignExtend16(__A.byte[4*j+1]) * SignExtend16(__B.byte[4*j+1])
123 /// 	tmp3.word := SignExtend16(__A.byte[4*j+2]) * SignExtend16(__B.byte[4*j+2])
124 /// 	tmp4.word := SignExtend16(__A.byte[4*j+3]) * SignExtend16(__B.byte[4*j+3])
131                                                                   __m128i __B) {
133                                               (__v4si)__B);
137 ///    corresponding signed 8-bit integers in \a __B, producing 4 intermediate
145 /// _mm256_dpbssds_epi32(__m256i __W, __m256i __A, __m256i __B);
152 /// \param __B
159 /// 	tmp1.word := SignExtend16(__A.byte[4*j]) * SignExtend16(__B.byte[4*j])
160 /// 	tmp2.word := SignExtend16(__A.byte[4*j+1]) * SignExtend16(__B.byte[4*j+1])
161 /// 	tmp3.word := SignExtend16(__A.byte[4*j+2]) * SignExtend16(__B.byte[4*j+2])
162 /// 	tmp4.word := SignExtend16(__A.byte[4*j+3]) * SignExtend16(__B.byte[4*j+3])
168 _mm256_dpbssds_epi32(__m256i __W, __m256i __A, __m256i __B) {
170                                               (__v8si)__B);
174 ///    corresponding unsigned 8-bit integers in \a __B, producing 4 intermediate
181 /// _mm_dpbsud_epi32(__m128i __W, __m128i __A, __m128i __B);
188 /// \param __B
195 /// 	tmp1.word := Signed(SignExtend16(__A.byte[4*j]) * ZeroExtend16(__B.byte[4*j]))
196 /// 	tmp2.word := Signed(SignExtend16(__A.byte[4*j+1]) * ZeroExtend16(__B.byte[4*j+1]))
197 /// 	tmp3.word := Signed(SignExtend16(__A.byte[4*j+2]) * ZeroExtend16(__B.byte[4*j+2]))
198 /// 	tmp4.word := Signed(SignExtend16(__A.byte[4*j+3]) * ZeroExtend16(__B.byte[4*j+3]))
205                                                                  __m128i __B) {
207                                              (__v4si)__B);
211 ///    corresponding unsigned 8-bit integers in \a __B, producing 4 intermediate
218 /// _mm256_dpbsud_epi32(__m256i __W, __m256i __A, __m256i __B);
225 /// \param __B
232 /// 	tmp1.word := Signed(SignExtend16(__A.byte[4*j]) * ZeroExtend16(__B.byte[4*j]))
233 /// 	tmp2.word := Signed(SignExtend16(__A.byte[4*j+1]) * ZeroExtend16(__B.byte[4*j+1]))
234 /// 	tmp3.word := Signed(SignExtend16(__A.byte[4*j+2]) * ZeroExtend16(__B.byte[4*j+2]))
235 /// 	tmp4.word := Signed(SignExtend16(__A.byte[4*j+3]) * ZeroExtend16(__B.byte[4*j+3]))
241 _mm256_dpbsud_epi32(__m256i __W, __m256i __A, __m256i __B) {
243                                              (__v8si)__B);
247 ///    corresponding unsigned 8-bit integers in \a __B, producing 4 intermediate
255 /// _mm_dpbsuds_epi32( __m128i __W, __m128i __A, __m128i __B);
262 /// \param __B
269 /// 	tmp1.word := Signed(SignExtend16(__A.byte[4*j]) * ZeroExtend16(__B.byte[4*j]))
270 /// 	tmp2.word := Signed(SignExtend16(__A.byte[4*j+1]) * ZeroExtend16(__B.byte[4*j+1]))
271 /// 	tmp3.word := Signed(SignExtend16(__A.byte[4*j+2]) * ZeroExtend16(__B.byte[4*j+2]))
272 /// 	tmp4.word := Signed(SignExtend16(__A.byte[4*j+3]) * ZeroExtend16(__B.byte[4*j+3]))
279                                                                   __m128i __B) {
281                                               (__v4si)__B);
285 ///    corresponding unsigned 8-bit integers in \a __B, producing 4 intermediate
293 /// _mm256_dpbsuds_epi32(__m256i __W, __m256i __A, __m256i __B);
300 /// \param __B
307 /// 	tmp1.word := Signed(SignExtend16(__A.byte[4*j]) * ZeroExtend16(__B.byte[4*j]))
308 /// 	tmp2.word := Signed(SignExtend16(__A.byte[4*j+1]) * ZeroExtend16(__B.byte[4*j+1]))
309 /// 	tmp3.word := Signed(SignExtend16(__A.byte[4*j+2]) * ZeroExtend16(__B.byte[4*j+2]))
310 /// 	tmp4.word := Signed(SignExtend16(__A.byte[4*j+3]) * ZeroExtend16(__B.byte[4*j+3]))
316 _mm256_dpbsuds_epi32(__m256i __W, __m256i __A, __m256i __B) {
318                                               (__v8si)__B);
322 ///    corresponding unsigned 8-bit integers in \a __B, producing 4 intermediate
329 /// _mm_dpbuud_epi32(__m128i __W, __m128i __A, __m128i __B);
336 /// \param __B
343 /// 	tmp1.word := ZeroExtend16(__A.byte[4*j]) * ZeroExtend16(__B.byte[4*j])
344 /// 	tmp2.word := ZeroExtend16(__A.byte[4*j+1]) * ZeroExtend16(__B.byte[4*j+1])
345 /// 	tmp3.word := ZeroExtend16(__A.byte[4*j+2]) * ZeroExtend16(__B.byte[4*j+2])
346 /// 	tmp4.word := ZeroExtend16(__A.byte[4*j+3]) * ZeroExtend16(__B.byte[4*j+3])
353                                                                  __m128i __B) {
355                                              (__v4si)__B);
359 ///    corresponding unsigned 8-bit integers in \a __B, producing 4 intermediate
366 /// _mm256_dpbuud_epi32(__m256i __W, __m256i __A, __m256i __B);
373 /// \param __B
380 /// 	tmp1.word := ZeroExtend16(__A.byte[4*j]) * ZeroExtend16(__B.byte[4*j])
381 /// 	tmp2.word := ZeroExtend16(__A.byte[4*j+1]) * ZeroExtend16(__B.byte[4*j+1])
382 /// 	tmp3.word := ZeroExtend16(__A.byte[4*j+2]) * ZeroExtend16(__B.byte[4*j+2])
383 /// 	tmp4.word := ZeroExtend16(__A.byte[4*j+3]) * ZeroExtend16(__B.byte[4*j+3])
389 _mm256_dpbuud_epi32(__m256i __W, __m256i __A, __m256i __B) {
391                                              (__v8si)__B);
395 ///    corresponding unsigned 8-bit integers in \a __B, producing 4 intermediate
403 /// _mm_dpbuuds_epi32( __m128i __W, __m128i __A, __m128i __B);
410 /// \param __B
417 /// 	tmp1.word := ZeroExtend16(__A.byte[4*j]) * ZeroExtend16(__B.byte[4*j])
418 /// 	tmp2.word := ZeroExtend16(__A.byte[4*j+1]) * ZeroExtend16(__B.byte[4*j+1])
419 /// 	tmp3.word := ZeroExtend16(__A.byte[4*j+2]) * ZeroExtend16(__B.byte[4*j+2])
420 /// 	tmp4.word := ZeroExtend16(__A.byte[4*j+3]) * ZeroExtend16(__B.byte[4*j+3])
427                                                                   __m128i __B) {
429                                               (__v4si)__B);
433 ///    corresponding unsigned 8-bit integers in \a __B, producing 4 intermediate
441 /// _mm256_dpbuuds_epi32(__m256i __W, __m256i __A, __m256i __B);
448 /// \param __B
455 /// 	tmp1.word := ZeroExtend16(__A.byte[4*j]) * ZeroExtend16(__B.byte[4*j])
456 /// 	tmp2.word := ZeroExtend16(__A.byte[4*j+1]) * ZeroExtend16(__B.byte[4*j+1])
457 /// 	tmp3.word := ZeroExtend16(__A.byte[4*j+2]) * ZeroExtend16(__B.byte[4*j+2])
458 /// 	tmp4.word := ZeroExtend16(__A.byte[4*j+3]) * ZeroExtend16(__B.byte[4*j+3])
464 _mm256_dpbuuds_epi32(__m256i __W, __m256i __A, __m256i __B) {
466                                               (__v8si)__B);