liangrj2014
commented
11 months ago -
We expect that the input is limited to .net/.cap formats. And evaluator.cpp includes parsers for the .net/.cap formats.
-
Concerning the "mempool_group" design, the xSize is indicated as 1782 in the "GCELLGRID" field of the .def file. However, it possesses 1781= (1782 - 1) GCell edges, as shown in the .cap file. This behavior aligns with expectations, as illustrated in the following figure. Additionally, it is anticipated that the sum of hEdge lengths (1995 + 3800 * 1779 + 3325) equals 6765520, which is less than the width of the die (6767040). This conforms to the expectation that the sum of hEdge lengths is less than the sum of the widths of GCells = the width of the die.
Could you please confirm whether LEF/DEF files will be accepted, or if the input is limited to .net/.cap formats?
I found the .cap files have some mismatch with the .def file. For example, for design "mempool_group", the hEdge length in the .cap file is "1995 3800 3800 ... 3800 3325" and xSize is 1782. According to evaluator.cpp, I think there should be 1781 grids on X direction. The summation hEdge length is 1995 + 3800 1779 + 3325 = 6765520, which does NOT match "DIE_AREA ( 0 0 ) ( 6767040 6762560 ) ;" in mempool_group.def (6767040 != 6765520). And the "GCELLGRID" field in .def file is not matched with .cap file. Besides, for design "mempool_group", in .cap file, xSize = 1782, ySize = 2417, I think there should be a 1781 2416 grid map. And the access points should be in range [0, 1780] for x and [0, 2415] for y. However, I found for net "tcdm_master_req_o[2975]", one of the access point is [(2, 1781, 665)], where 1781 > 1780.