Why does my pipeline take double the number clock cycles than expected?

[ElVale]

Hi, I have this generate block for an unrolled pipeline, notice the 0 to 63 range.

rounds: for i in 0 to 63 generate
	 begin
	 	process(clk)
	 	begin
		if (clk = '1' and clk'event) then 
			 T1(i) <= h(i)+ e1(e(i)) + ch(e(i), f(i), g(i)) + K(i) + W(i);
			 T2(i) <= e0(a(i)) + maj(a(i),b(i),c(i));
			 h(i+1) <= g(i);
			 g(i+1) <= f(i);
			 f(i+1) <= e(i);
			 e(i+1) <= d(i) + T1(i);
			 d(i+1) <= c(i);
			 c(i+1) <= b(i);
			 b(i+1) <= a(i);
			 a(i+1) <= T1(i) + T2(i);
		end if;
		end process;
	 end generate rounds;

                process(clk)
	        begin
		if (clk = '1' and clk'event) then 
			H1_out <= H1_in + a(64);
			H2_out <= H2_in + b(64);
			H3_out <= H3_in + c(64);
			H4_out <= H4_in + d(64);	
			H5_out <= H5_in + e(64);
			H6_out <= H6_in + f(64);
			H7_out <= H7_in + g(64);
			H8_out <= H8_in + h(64);	
		end if;
	        end process;

When I simulate I get 128 clock cycles, when I was expecting 64, I'm using a 100MHz clock and I get 1280 ns in total. Am I missing something here? thanks a lot for your help.

 

[ads-ee]

The loop looks okay to me. Are you absolutely sure the clock is correctly defined and really has a 10 ns period? It's pretty easy to forget to divide the clock period by two in the testbench for generating a clock.

Regards,

 

[ElVale]

I'm totally positive, I'm using:

CLK_GEN: process
	begin
		clk <= '0';
		wait for 5 ns;
		clk <= '1';
		wait for 5 ns;
	end process;

I'll try with another simulator and see, thanks for confirming.

 

[permute]

T1, T2 should be variables or placed outside of the process. It currently takes one cycle to update T1/T2, and 1 cycle for the logic after these. I'm assuming the logic is also incorrect for some cases that you haven't tested yet.

 

[ElVale]

T1, T2 should be variables or placed outside of the process. It currently takes one cycle to update T1/T2, and 1 cycle for the logic after these. I'm assuming the logic is also incorrect for some cases that you haven't tested yet.

I made T1 and T2 variables and BAM! 64 cycles only, yet I don't quite understand what's the theory behind this, anyway thanks a lot!

 

[permute]

look up blocking vs non-blocking assignments. There are many threads on this subject. (or use google on the same subject)

short version ":=" evaluates and assigns immediately. "<=" evaluates the expression, but assigns at the end of the process. (or wait)

 

[ftsolutions]

permute nailed it!