Open tissue3 opened 4 years ago
mattfel1
commented
4 years ago This looks like there are old packages hanging around. Can you try deleting these directories to get a totally fresh start and trying again?
~/.ivy2/cache/edu.stanford.cs.dawn ~/.ivy2/local/edu.stanford.cs.dawn ~/spatial-quickstart/target ~/spatial-quickstart/project/target ~/spatial-quickstart/gen
?
tissue3
commented
4 years ago It works now.
mattfel1
commented
4 years ago Just caught a mistake in the synth makefile (if you are using our scripts to synthesize instead of your own). I will push a fix, but in the short term you should edit zynq.hw-resources/build/Makefile. Change line 24 from sed -i 's/design_1_SpatialIP_0_0/Top/' design_1.v to sed -i 's/design_1_SpatialIP_0_0/SpatialIP/' design_1.v`
tissue3
commented
4 years ago Thanks for reminding. I manually changed it and reran the flow. Everything works fine till Phase 3 of vivado synthesis. "The packing of instances into the device could not be obeyed." Any insight that why this would occur? I am using Vivado v2017.2_sdx.
Phase 4 Add Constraints new method
Phase 4.1 Build CellView Core
Phase 4.1 Build CellView Core | Checksum: 418df461
Time (s): cpu = 00:06:24 ; elapsed = 00:03:31 . Memory (MB): peak = 3162.660 ; gain = 33.109 ; free physical = 4739 ; free virtual = 49878
Phase 4.2 Add User PBlocks
Phase 4.2 Add User PBlocks | Checksum: 418df461
Time (s): cpu = 00:06:24 ; elapsed = 00:03:31 . Memory (MB): peak = 3162.660 ; gain = 33.109 ; free physical = 4739 ; free virtual = 49878
Phase 4.3 Apply User PBlocks
Phase 4.3 Apply User PBlocks | Checksum: 418df461
Time (s): cpu = 00:06:24 ; elapsed = 00:03:31 . Memory (MB): peak = 3162.660 ; gain = 33.109 ; free physical = 4739 ; free virtual = 49878
Phase 4.4 Add terminal Constraints
Phase 4.4 Add terminal Constraints | Checksum: 418df461
Time (s): cpu = 00:06:24 ; elapsed = 00:03:31 . Memory (MB): peak = 3162.660 ; gain = 33.109 ; free physical = 4739 ; free virtual = 49878
Phase 4 Add Constraints new method | Checksum: 418df461
Time (s): cpu = 00:06:24 ; elapsed = 00:03:31 . Memory (MB): peak = 3162.660 ; gain = 33.109 ; free physical = 4739 ; free virtual = 49878
Ending Placer Task | Checksum: b5354623
Time (s): cpu = 00:06:25 ; elapsed = 00:03:31 . Memory (MB): peak = 3162.660 ; gain = 33.109 ; free physical = 4838 ; free virtual = 49977
42 Infos, 1 Warnings, 0 Critical Warnings and 3 Errors encountered.
place_design failed
ERROR: [Common 17-69] Command failed: Placer could not place all instances
INFO: [Common 17-206] Exiting Vivado at Tue Oct 29 18:10:35 2019...It looks like the design didn't fit on the board. There should be a synth_utilization_hierarchical.rpt in the verilog/ directory that may give hints as to what parts of the app are taking a lot of resources. We can look at that and try to figure out if there are opportunities for improvement in the app. The best way is to either open the rpt in Vivado, or parse it as a csv and sort by whichever resource you want to tackle first (probably LUTs)
tissue3
commented
4 years ago I parse it as a csv and sort by LUT usage and got this:
(top) | 55186(103.73%)
-- | --
design_1 | 55186(103.73%)
SpatialIP | 38967(73.25%)
FringeZynq | 33842(63.61%)
Fringe | 33428(62.83%)
DRAMArbiter_1_817 | 15176(28.53%)
StreamControllerGather | 13878(26.09%)
GatherBuffer | 13256(24.92%)
FIFO_61 | 13256(24.92%)
FFRAM_1 | 13058(24.55%)
DRAMArbiter | 8598(16.16%)
AccelUnit | 5125(9.63%)
... (there are more lines)It seems no particular component used too much resources but when they addup, it becomes unacceptably large. Again, I am using HelloSpatial and didn't modify any line in the code. The design seems not particularly large, why would it be too large?
mattfel1
commented
4 years ago Thanks for pointing this out. It seems like we've never actually tried to synthesize this app before :grimacing: . We have usually just used it to show how to do miscellaneous things in spatial, but the load+store+scatter+gather streams seem to add up to a DMA that is way too big for the Zynq (Fringe is usually around 25% of this board if I recall correctly. 62.83% is huge. Most of the area for real apps should go into the AccelUnit). If your goal is to just get the flow working from start to finish, you can just comment out the scatter and gather lines (49 and 51) and I think it should fit (testing now).
Following the advice in #277 , I tried to use Zynq as target device. However, this result in errors at compilation stage. I used spatial-quickstart run the following commands:
This results in a lot of errors.