Closed enjoy-digital closed 5 years ago
This is an nMigen bug. The problematic lines all are: assign \$verilog_initial_trigger = \$verilog_initial_trigger;
, which is a true positive from Verilator's POV. The reason for \$verilog_initial_trigger
's existence is that a statement like always @* x = 1;
in Verilog will never execute because its implicit event expression list is empty. However, with current Yosys master (specifically, anything that has the proc_prune
pass), such pathological statements will never be emitted. (Statements like always @* if(y) x = 1;
will be, but these are fine, since a conditional does read something.)
I will fix this as it is quite easy.
Let me know if this nMigen patch works for you:
diff --git a/nmigen/back/verilog.py b/nmigen/back/verilog.py
index 8faf58a..eb98a34 100644
--- a/nmigen/back/verilog.py
+++ b/nmigen/back/verilog.py
@@ -39,6 +39,7 @@ def _convert_rtlil_text(rtlil_text, *, strip_internal_attrs=False, write_verilog
read_ilang <<rtlil
{}
rtlil
+{prune}delete w:$verilog_initial_trigger
{prune}proc_prune
proc_init
proc_arst
Thanks for the quick answer/patch, simulation is working correctly with it :+1:
When elaborating Minerva with upstream sources and tools (Minerva, nMigen, Yosys), verilator does not seem able to compile the elaborated verilog:
The CPU has been generated with:
Assuming a RISC-V toolchain and Verilator are installed, the error can be reproduced with:
Here is the elaborated CPU: minerva.zip