YosysHQ / yosys

Yosys Open SYnthesis Suite
https://yosyshq.net/yosys/
ISC License
3.51k stars 895 forks source link

Flattening failing #1641

Open AurelienUoU opened 4 years ago

AurelienUoU commented 4 years ago

Dear community,

I encountered issues to flatten a design while I try to synthesize and techmap it with the latest version of Yosys. Design is dual-clock with a wrapper on top to merge read and write clocks. A small case works but the full design doesn't.

The issue appears at the first "synth" step using the flag "-flatten".

All the elements (scripts, netlits etc...) can be found at https://github.com/AurelienUoU/fifo_to_one_clock_dpram.git

Sincerely, Aurelien

daveshah1 commented 4 years ago

synth -run coarse only runs the "coarse" label of synth, which does not include hierarchy and therefore modules aren't elaborated before proc with the proper parameters. Run hierarchy -top ... before synth -run coarse, or use synth -run :fine -top to run both hierarchy and coarse synthesis (:fine means everything up to but not including the fine label)

AurelienUoU commented 4 years ago

Thanks for your quick answer!

I updated the script (and the repo) according to your advice. It appears than my top module "LMAC_CORE_TOP_SYNCH" is pushed down to "LMAC_CORE_TOP". It's not really a big deal and I figured it out relatively quickly. However, it can be painful to often checking this.

eddiehung commented 4 years ago

I updated the script (and the repo) according to your advice. It appears than my top module "LMAC_CORE_TOP_SYNCH" is pushed down to "LMAC_CORE_TOP".

Can you elaborate what you mean here?

AurelienUoU commented 4 years ago

My top module is LMAC_CORE_TOP_SYNTH, it's just a wrapper of LMAC_CORE_TOP who's merging rclk and wclk into clk. However, after all the netlists are read the module LMAC_CORE_TOP_SYNTH is no longer available. For example, "hierarchy -top LMAC_CORE_TOP_SYNTH" fails because this module "doesn't exist". If I use "hierarchy -top LMAC_CORE_TOP" everything works well. The clock merging is propagated into LMAC_CORE_TOP. If I do not read LMAC_CORE_TOP_SYNTH.v and still use "hierarchy -top LMAC_CORE_TOP" my fifo is not correctly map to the mono-clock DPRAM.

AurelienUoU commented 4 years ago

BTW, about the synth command. Documentation shows hierarchy is done as "begin" and as far as I understood synth -run coarse should run "begin" and "coarse" parts. Or, should I do synth -run begin then synth -run coarse ?

daveshah1 commented 4 years ago

synth -run coarse runs just the coarse label part. To run both hierarchy and coarse, use synth -run :fine (everything up to but not including the fine label)

daveshah1 commented 4 years ago

Looking again at the top module bug,

However, after all the netlists are read the module LMAC_CORE_TOP_SYNTH is no longer available. For example, "hierarchy -top LMAC_CORE_TOP_SYNTH" fails because this module "doesn't exist".

Seems like it is not parsing LMAC_CORE_TOP_SYNTH correctly, but something odd is going on as neither Verilator nor iverilog do either

daveshah1 commented 4 years ago

God knows where that file came from, but seems like the problem is some of the spaces are actually UTF-8 U+2008 "punctuation spaces" rather than normal ASCII ones.

iconv -f utf-8 -t ascii//TRANSLIT Verilog_netlists/full_design/LMAC_CORE_TOP_SYNTH.v > Verilog_netlists/full_design/LMAC_CORE_TOP_SYNTH_ascii.v && mv Verilog_netlists/full_design/LMAC_CORE_TOP_SYNTH_ascii.v Verilog_netlists/full_design/LMAC_CORE_TOP_SYNTH.v is a workaround for this, but perhaps Yosys should be failing more gracefully in this case.

mithro commented 4 years ago

Just FYI, the code comes from one of the modules at https://github.com/lewiz-support

daveshah1 commented 4 years ago

I don't think the problematic file (LMAC_CORE_TOP_SYNTH) does - I'm mostly curious what kind of editor is actually inserting these characters...