Lines Matching full:tile
20 __attribute__((__always_inline__, __nodebug__, __target__("amx-tile")))
26 /// Load tile configuration from a 64-byte memory location specified by
27 /// "mem_addr". The tile configuration includes the tile type palette, the
29 /// palette_id is zero, that signifies the init state for both the tile
30 /// config and the tile data, and the tiles are zeroed. Any invalid
44 /// Stores the current tile configuration to a 64-byte memory location
45 /// specified by "mem_addr". The tile configuration includes the tile type
60 /// Release the tile configuration to return to the init state, which
70 /// Load tile rows from memory specifieid by "base" address and "stride" into
71 /// destination tile "dst" using the tile configuration previously configured
79 /// A destination tile. Max size is 1024 Bytes.
88 /// Load tile rows from memory specifieid by "base" address and "stride" into
89 /// destination tile "dst" using the tile configuration previously configured
99 /// A destination tile. Max size is 1024 Bytes.
108 /// Store the tile specified by "src" to memory specifieid by "base" address and
109 /// "stride" using the tile configuration previously configured via
117 /// A destination tile. Max size is 1024 Bytes.
125 /// Zero the tile specified by "tdest".
131 /// \param tile
132 /// The destination tile to be zero. Max size is 1024 Bytes.
133 #define _tile_zero(tile) __builtin_ia32_tilezero((tile))
139 /// and store the 32-bit result back to tile "dst".
146 /// The destination tile. Max size is 1024 Bytes.
148 /// The 1st source tile. Max size is 1024 Bytes.
150 /// The 2nd source tile. Max size is 1024 Bytes.
158 /// in "dst", and store the 32-bit result back to tile "dst".
165 /// The destination tile. Max size is 1024 Bytes.
167 /// The 1st source tile. Max size is 1024 Bytes.
169 /// The 2nd source tile. Max size is 1024 Bytes.
177 /// and store the 32-bit result back to tile "dst".
184 /// The destination tile. Max size is 1024 Bytes.
186 /// The 1st source tile. Max size is 1024 Bytes.
188 /// The 2nd source tile. Max size is 1024 Bytes.
196 /// "dst", and store the 32-bit result back to tile "dst".
203 /// The destination tile. Max size is 1024 Bytes.
205 /// The 1st source tile. Max size is 1024 Bytes.
207 /// The 2nd source tile. Max size is 1024 Bytes.
213 /// elements with elements in "dst", and store the 32-bit result back to tile
221 /// The destination tile. Max size is 1024 Bytes.
223 /// The 1st source tile. Max size is 1024 Bytes.
225 /// The 2nd source tile. Max size is 1024 Bytes.
229 /// AMX tile register size can be configured, the maximum size is 16x64=1024
231 /// represent 2D tile and the fixed size is maximum amx tile register size.
283 __SIZE_TYPE__ stride, _tile1024i tile) {
285 (__SIZE_TYPE__)(stride), tile);
295 /// This struct pack the shape and tile data together for user. We suggest
302 _tile1024i tile;
305 /// Load tile rows from memory specifieid by "base" address and "stride" into
306 /// destination tile "dst".
313 /// A destination tile. Max size is 1024 Bytes.
321 dst->tile = _tile_loadd_internal(dst->row, dst->col, base, stride);
324 /// Load tile rows from memory specifieid by "base" address and "stride" into
325 /// destination tile "dst". This intrinsic provides a hint to the implementation
334 /// A destination tile. Max size is 1024 Bytes.
342 dst->tile = _tile_loaddt1_internal(dst->row, dst->col, base, stride);
349 /// and store the 32-bit result back to tile "dst".
356 /// The destination tile. Max size is 1024 Bytes.
358 /// The 1st source tile. Max size is 1024 Bytes.
360 /// The 2nd source tile. Max size is 1024 Bytes.
364 dst->tile = _tile_dpbssd_internal(src0.row, src1.col, src0.col, dst->tile,
365 src0.tile, src1.tile);
372 /// in "dst", and store the 32-bit result back to tile "dst".
379 /// The destination tile. Max size is 1024 Bytes.
381 /// The 1st source tile. Max size is 1024 Bytes.
383 /// The 2nd source tile. Max size is 1024 Bytes.
387 dst->tile = _tile_dpbsud_internal(src0.row, src1.col, src0.col, dst->tile,
388 src0.tile, src1.tile);
395 /// and store the 32-bit result back to tile "dst".
402 /// The destination tile. Max size is 1024 Bytes.
404 /// The 1st source tile. Max size is 1024 Bytes.
406 /// The 2nd source tile. Max size is 1024 Bytes.
410 dst->tile = _tile_dpbusd_internal(src0.row, src1.col, src0.col, dst->tile,
411 src0.tile, src1.tile);
418 /// "dst", and store the 32-bit result back to tile "dst".
425 /// The destination tile. Max size is 1024 Bytes.
427 /// The 1st source tile. Max size is 1024 Bytes.
429 /// The 2nd source tile. Max size is 1024 Bytes.
433 dst->tile = _tile_dpbuud_internal(src0.row, src1.col, src0.col, dst->tile,
434 src0.tile, src1.tile);
437 /// Store the tile specified by "src" to memory specifieid by "base" address and
451 _tile_stored_internal(src.row, src.col, base, stride, src.tile);
454 /// Zero the tile specified by "dst".
461 /// The destination tile to be zero. Max size is 1024 Bytes.
464 dst->tile = __builtin_ia32_tilezero_internal(dst->row, dst->col);
469 /// elements with elements in "dst", and store the 32-bit result back to tile
477 /// The destination tile. Max size is 1024 Bytes.
479 /// The 1st source tile. Max size is 1024 Bytes.
481 /// The 2nd source tile. Max size is 1024 Bytes.
485 dst->tile = _tile_dpbf16ps_internal(src0.row, src1.col, src0.col, dst->tile,
486 src0.tile, src1.tile);