Home | History | Annotate | Download | only in Headers

Lines Matching refs:src0

136 /// Multiply groups of 4 adjacent pairs of signed 8-bit integers in src0 with
147 /// \param src0
151 #define _tile_dpbssd(dst, src0, src1)                                          \
152   __builtin_ia32_tdpbssd((dst), (src0), (src1))
155 /// Multiply groups of 4 adjacent pairs of signed 8-bit integers in src0 with
166 /// \param src0
170 #define _tile_dpbsud(dst, src0, src1)                                          \
171   __builtin_ia32_tdpbsud((dst), (src0), (src1))
174 /// Multiply groups of 4 adjacent pairs of unsigned 8-bit integers in src0 with
185 /// \param src0
189 #define _tile_dpbusd(dst, src0, src1)                                          \
190   __builtin_ia32_tdpbusd((dst), (src0), (src1))
193 /// Multiply groups of 4 adjacent pairs of unsigned 8-bit integers in src0 with
204 /// \param src0
208 #define _tile_dpbuud(dst, src0, src1)                                          \
209   __builtin_ia32_tdpbuud((dst), (src0), (src1))
211 /// Compute dot-product of BF16 (16-bit) floating-point pairs in tiles src0 and
222 /// \param src0
226 #define _tile_dpbf16ps(dst, src0, src1)                                        \
227   __builtin_ia32_tdpbf16ps((dst), (src0), (src1))
315 /// Multiply groups of 4 adjacent pairs of signed 8-bit integers in src0 with
326 /// \param src0
331 static void __tile_dpbssd(__tile1024i *dst, __tile1024i src0,
333   dst->tile = _tile_dpbssd_internal(src0.row, src1.col, src0.col, dst->tile,
334                                     src0.tile, src1.tile);
338 /// Multiply groups of 4 adjacent pairs of signed 8-bit integers in src0 with
349 /// \param src0
354 static void __tile_dpbsud(__tile1024i *dst, __tile1024i src0,
356   dst->tile = _tile_dpbsud_internal(src0.row, src1.col, src0.col, dst->tile,
357                                     src0.tile, src1.tile);
361 /// Multiply groups of 4 adjacent pairs of unsigned 8-bit integers in src0 with
372 /// \param src0
377 static void __tile_dpbusd(__tile1024i *dst, __tile1024i src0,
379   dst->tile = _tile_dpbusd_internal(src0.row, src1.col, src0.col, dst->tile,
380                                     src0.tile, src1.tile);
384 /// Multiply groups of 4 adjacent pairs of unsigned 8-bit integers in src0 with
395 /// \param src0
400 static void __tile_dpbuud(__tile1024i *dst, __tile1024i src0,
402   dst->tile = _tile_dpbuud_internal(src0.row, src1.col, src0.col, dst->tile,
403                                     src0.tile, src1.tile);
437 /// Compute dot-product of BF16 (16-bit) floating-point pairs in tiles src0 and
448 /// \param src0
453 static void __tile_dpbf16ps(__tile1024i *dst, __tile1024i src0,
455   dst->tile = _tile_dpbf16ps_internal(src0.row, src1.col, src0.col, dst->tile,
456                                       src0.tile, src1.tile);