Another strange icarus expression eval bug (large shifts)

[cliffordwolf]

For b != 0 this should return y[3:2] == 0.

module issue_048(a, b, y);
  input [1:0] a;
  input [2:0] b;
  output [3:0] y;
  assign y = {a >> {22{b}}, a << (0 <<< b)};
endmodule

But iverilog 020e280 returns y[3:2] == y[1:0] instead.

[martinwhitaker]

I can't reproduce this. The following test

module tb();

reg  [1:0] a;
reg  [2:0] b;
wire [3:0] y;

issue_048 dut(a, b, y);

initial begin
  a = 2;
  b = 1;
  #0 $display("%b", y);
end

endmodule

outputs "0010", which seems right. What values of 'a' and 'b' are you using?

[cliffordwolf]

Now this is funny...

The following code (based on your test bench) outputs "11 111 0011":

initial begin
  a = 3; b = 7;
  #0 $display("%b %b %b", a, b, y);
end   

But the following outputs "11 111 1111":

initial begin
  { a, b } = ~0;
  #0 $display("%b %b %b", a, b, y);
end   

[martinwhitaker]

Turns out it is sensitive to the order you assign to a and b. So

a = 3; b = 7;

is OK but

b = 7; a = 3;

causes it to fail.

[martinwhitaker]

Blame this on my fix for large shifts at run-time. Should be fixed now.