Reputation: 3079
This is a follow-up question to Combinatorial synthesis: Better technology mapping results.
I am using Yosys (version 0.5+ (git sha1 f13e387, gcc 5.3.1-8ubuntu2 -O2 -fstack-protector-strong -fPIC -Os)
) with the following synthesis script:
read_liberty -lib my_library.lib
read_verilog test.v
hierarchy -check -top test
proc; opt; memory; opt; fsm -norecode; opt
techmap; opt
dfflibmap -liberty my_library.lib
abc -liberty my_library.lib -script \
+strash;ifraig;scorr;dc2;dretime;strash;&get,-n;&dch,-f;&nf,{D};&put
hilomap -hicell LIB_TIEHI Y -locell LIB_TIELO Y
clean
write_verilog -noattr -noexpr output.v
stat
...to synthesise the following Verilog code (test.v):
module mux4(
input i0, i1, i2, i3,
input s0, s1,
output z);
reg zint;
always @(*) begin
case ({s1, s0})
2'b00: zint = i0;
2'b01: zint = i1;
2'b10: zint = i2;
2'b11: zint = i3;
default: zint = i3;
endcase
end
assign z = zint;
endmodule
module test (
input a,b,c,d,
output result
);
mux4 inst (
.i0(a), .i1(b), .i2(c), .i3(d),
.s0(1'b0), .s1(1'b0), # constants here!
.z(result)
);
endmodule
The synthesis results include a LIB_MUX4
instance with both S0
and S1
tied low by two LIB_TIELO
instances.
Why doesn't Yosys see that S0
and S1
are constant and reduce the output to something like this
module test(a, b, c, d, result);
input a;
input b;
input c;
input d;
output result;
assign result = a;
endmodule
instead?
I tried using the clean -purge
, opt_muxtree
and opt_clean
commands, but without success - the static LIB_MUX
instance is always in the resulting netlist.
Upvotes: 0
Views: 385
Reputation: 8245
You need to run flatten
if you want optimization across hierarchical boundaries.
You probably want to run opt -full
shortly before running techmap
, but after running high-level optimizations like fsm
and share
.
JFYI: If you do not provide all files necessary to run your test case, people will not be able to reproduce what you are talking about. I don't have your my_library.lib
, so I did not even bother trying to run your code.
Upvotes: 1