clock cycles on + - * /

[Altera_Forum]

Hei, I try to figure out how many clock cycles that is needed to do + - * / on both int and float.  

 I have used the : nios2_51\examples\vhdl\niosII_cycloneII_2c35\full_featured nios2_51\examples\vhdl\niosII_cycloneII_2c35\standard  The bothe show the same result  I made an counter in hardware: library ieee, std; use ieee.std_logic_1164.all; use ieee.numeric_std.all; use std.textio.all; entity counter is  port    (      clk     : in std_logic;      reset_n : in std_logic;      chipselect : in std_logic;      address    : in std_logic_vector(0 downto 0);      write     : in std_logic;      writedata : in std_logic_vector(31 downto 0);      read     : in  std_logic;      readdata : out std_logic_vector(31 downto 0);     irq : out std_logic      ); end counter; ------------------------------------------------------------------------------- -- Architecture: behavior ------------------------------------------------------------------------------- architecture behavior of counter is     signal counter : unsigned(31 downto 0); begin  readdata <= std_logic_vector(counter);  pro_clock : process (clk, reset_n)  begin    if reset_n = &#39;0&#39; then      counter   <= (others => &#39;0&#39;);   irq <= &#39;0&#39;;    elsif rising_edge(clk) then  if chipselect = &#39;1&#39; and write = &#39;1&#39; and address = "0" then      counter <= unsigned(writedata);      else      counter <= counter + 1;  end if;    if chipselect = &#39;1&#39; and write = &#39;1&#39; and address = "1" then      irq <= writedata(0);      end if;      end if;  end process pro_clock; end architecture behavior;  And I code In C:  #include "sys/alt_irq.h"# include "system.h"# include "stdio.h" # include <io.h> # define IOADDR_PRE_COUNTER(base)              __IO_CALC_ADDRESS_NATIVE(base, 0)# define IORD_PRE_AVALON_COUNTER(base)                IORD(base, 0) # define IOWR_PRE_AVALON_COUNTER(base, data)          IOWR(base, 0, data) # define IOADDR_PRE_IRQ(base)              __IO_CALC_ADDRESS_NATIVE(base, 0)# define IORD_PRE_AVALON_IRQ(base)                IORD(base, 0) # define IOWR_PRE_AVALON_IRQ(base, data)          IOWR(base, 0, data) int main(void){ register volatile float a,b = 1,c = 2; volatile int data0, data1, data2,i; alt_irq_context context; for(i = 0, a = 0, b = 0; i < 100; i ++, b = b + a, a = a + 1){   context = alt_irq_disable_all();   IOWR_PRE_AVALON_COUNTER(0x01211160, 0);   data0 = IORD_PRE_AVALON_COUNTER(0x01211160);   c = b * a;   data1 = IORD_PRE_AVALON_COUNTER(0x01211160);   alt_irq_enable_all(context);   data2 = data1 - data0;   printf("%d \n", data2); } return 0; }  And I have tested it with the run on harware.   I get: 30 clock cycles on * int 30 clock cycles on + int 200 clock cycles on / int  600 clock cycles on * float 300 clock cycles on + float 600 clock cycles on / float  Can this be right?  In http://www.altera.com/literature/hb/nios2/n2cpu_nii51015.pdf (http://www.altera.com/literature/hb/nios2/n2cpu_nii51015.pdf) it says that the standard core use 5 clock cycles on embedded multiply.  Accessing the counter tok around 3 clock cycles.   I have tried to disable the interrupt, I think I Have done it right. I do 100 operation on eatch and all show almost the same. I don't belive some interupt have managed to jump in between 30 clock cycles on all the hundred.   I also did the compilation with release optimation and not debug optimation.   I also remeber some other in this forum tell that the use 87 clock cycles on the float.  Another ting wile I am writing, one time I tried to enable the hardware divde on the full design eksample. But the divide didn't work. I didn't chec the timing in quartus that time. But the core was running as it was set up in the desing on 85 MHz. It is so that the hardware divide don't manage 85 MHz?  Have someone some information on the floating point instruction you can download under the IP place in this forum. How fast is it, how much space does it take? I did look into the file, but there was only some of the operation that information was there.

[Altera_Forum]

I change in code with the same result: 

  for(i = 0; i < 100; i ++){   for(j = 0; j < 100; j ++){     a = a + 1;     b = b + a;   }     context = alt_irq_disable_all();   IOWR_PRE_AVALON_COUNTER(0x01211160, 0);   data0 = IORD_PRE_AVALON_COUNTER(0x01211160);     for(j = 0; j < 100; j ++){       c = b + a;     }   data1 = IORD_PRE_AVALON_COUNTER(0x01211160);   alt_irq_enable_all(context);   data2 = (data1 - data0)/100;   printf("%d \n", data2); }

[Altera_Forum]

TO_BE_DONE

[Altera_Forum]

hi, 

 all these mesurement is good http://forum.niosforum.com/work2/style_emoticons/<#EMO_DIR#>/smile.gif   but did u see how do access to the avalon switch interface ? http://forum.niosforum.com/work2/style_emoticons/<#EMO_DIR#>/rolleyes.gif   if not, double click on your counter in SOPC builder, and see cycle definition that u specify, and reply me http://forum.niosforum.com/work2/style_emoticons/<#EMO_DIR#>/wink.gif

[Altera_Forum]

--- Quote Start ---  originally posted by sdhnain@May 1 2006, 03:53 PM hi,  all these mesurement is good  http://forum.niosforum.com/work2/style_emoticons/<#emo_dir#>/smile.gif   but did u see how do access to the avalon switch interface ? http://forum.niosforum.com/work2/style_emoticons/<#emo_dir#>/rolleyes.gif   if not, double click on your counter in sopc builder, and see cycle definition that u specify, and reply me  http://forum.niosforum.com/work2/style_emoticons/<#emo_dir#>/wink.gif  <div align='right'><{post_snapback}> (index.php?act=findpost&pid=14897)  --- quote end ---    --- Quote End ---    I use only 1 clock cycle for reading and writing the counter. And register interface.

[Altera_Forum]

to doofus64: thanks for this topic. it&#39;s realy help me. 

 doofus64 can you show code of progamm with  **** Floating Point Arithmetic Test ****.  Is&#39;s offtop. sorry.

[Altera_Forum]

http://forum.niosforum.com/work2/style_emoticons/<#EMO_DIR#>/ph34r.gif