@@ -238,6 +238,24 @@ _tile_dpbssd_internal(unsigned short m, unsigned short n, unsigned short k,
238238 return __builtin_ia32_tdpbssd_internal (m , n , k , dst , src1 , src2 );
239239}
240240
241+ static __inline__ _tile1024i __DEFAULT_FN_ATTRS_INT8
242+ _tile_dpbsud_internal (unsigned short m , unsigned short n , unsigned short k ,
243+ _tile1024i dst , _tile1024i src1 , _tile1024i src2 ) {
244+ return __builtin_ia32_tdpbsud_internal (m , n , k , dst , src1 , src2 );
245+ }
246+
247+ static __inline__ _tile1024i __DEFAULT_FN_ATTRS_INT8
248+ _tile_dpbusd_internal (unsigned short m , unsigned short n , unsigned short k ,
249+ _tile1024i dst , _tile1024i src1 , _tile1024i src2 ) {
250+ return __builtin_ia32_tdpbusd_internal (m , n , k , dst , src1 , src2 );
251+ }
252+
253+ static __inline__ _tile1024i __DEFAULT_FN_ATTRS_INT8
254+ _tile_dpbuud_internal (unsigned short m , unsigned short n , unsigned short k ,
255+ _tile1024i dst , _tile1024i src1 , _tile1024i src2 ) {
256+ return __builtin_ia32_tdpbuud_internal (m , n , k , dst , src1 , src2 );
257+ }
258+
241259static __inline__ void __DEFAULT_FN_ATTRS_INT8
242260_tile_stored_internal (unsigned short m , unsigned short n , void * base ,
243261 __SIZE_TYPE__ stride , _tile1024i tile ) {
@@ -264,6 +282,27 @@ static void __tile_dpbssd(__tile1024i *dst, __tile1024i src1,
264282 src1 .tile , src2 .tile );
265283}
266284
285+ __DEFAULT_FN_ATTRS_INT8
286+ static void __tile_dpbsud (__tile1024i * dst , __tile1024i src1 ,
287+ __tile1024i src2 ) {
288+ dst -> tile = _tile_dpbsud_internal (src1 .row , src2 .col , src1 .col , dst -> tile ,
289+ src1 .tile , src2 .tile );
290+ }
291+
292+ __DEFAULT_FN_ATTRS_INT8
293+ static void __tile_dpbusd (__tile1024i * dst , __tile1024i src1 ,
294+ __tile1024i src2 ) {
295+ dst -> tile = _tile_dpbusd_internal (src1 .row , src2 .col , src1 .col , dst -> tile ,
296+ src1 .tile , src2 .tile );
297+ }
298+
299+ __DEFAULT_FN_ATTRS_INT8
300+ static void __tile_dpbuud (__tile1024i * dst , __tile1024i src1 ,
301+ __tile1024i src2 ) {
302+ dst -> tile = _tile_dpbuud_internal (src1 .row , src2 .col , src1 .col , dst -> tile ,
303+ src1 .tile , src2 .tile );
304+ }
305+
267306__DEFAULT_FN_ATTRS_TILE
268307static void __tile_stored (void * base , __SIZE_TYPE__ stride , __tile1024i src ) {
269308 _tile_stored_internal (src .row , src .col , base , stride , src .tile );
0 commit comments