Forum Discussion

Altera_Forum's avatar
Altera_Forum
Icon for Honored Contributor rankHonored Contributor
17 years ago

change logic cell to memory M4K in cylocne 2

How I can say to Quartus 2 (7.2), to replace logic cells to memory blocks M4K in Cyclone 2 device?

I would like to implement following array with memory M4K and not LC

type CoefArrType is array(0 to tapLen + 0 -1) of signed(coefBitWidth-1 downto 0);

Thanx,

Pita

3 Replies

  • Altera_Forum's avatar
    Altera_Forum
    Icon for Honored Contributor rankHonored Contributor
    17 years ago

    I wonder how you achieved to make Quartus use LC for a the signal array? Most likely you have multiple accesses to the signal in your code or are assigning the address expression outside a clock synchronous process. You should consider the limitations of internal RAM hardware.

  • Altera_Forum's avatar
    Altera_Forum
    Icon for Honored Contributor rankHonored Contributor
    17 years ago

    Thanx a lot for replay. This code is generated by Synplicity DSP...

    Below is example of accessing to signal multBufs1.

    CoefArr <= CoefArrTemp;

    multInp <= signed(inp);

    rstBuf <= '0';

    enBuf <= '1';

    outp <= std_logic_vector(sumOut);

    floop: for i in 1 to tapLen-1 generate

    multBufs1(i) <= syn_resize(Shift_Right(mem(i*4-1) * CoefArr(i),

    (coefFrac+inpFrac-dpFrac)), dpBitWidth);

    end generate;

    multBufs1(0) <= syn_resize(Shift_Right(multInp * CoefArr(0),

    (coefFrac+inpFrac-dpFrac)), dpBitWidth);

    gloop: for i in 0 to numfourAdders-1 generate

    sumFLOuts(i) <= signed(syn_resize(multBufs1(i*4) +multBufs1(i*4+1)+multBufs1(i*4+2)+multBufs1(i*4+3), outBitWidth));

    sumFLOutsDelayed(i) <= signed(sumFLOuts(i));

    end generate;

    sumFLOuts(numfourAdders) <= multBufs1(numfourAdders*4) ;

    sumFLOutsDelayed(25) <= signed(sumFLOuts(25));

    sumOut <= sumFLOutsDelayed(0) + sumFLOutsDelayed(1) + sumFLOutsDelayed(2) + sumFLOutsDelayed(3) + sumFLOutsDelayed(4) + sumFLOutsDelayed(5) + sumFLOutsDelayed(6) + sumFLOutsDelayed(7) + sumFLOutsDelayed(8) + sumFLOutsDelayed(9) + sumFLOutsDelayed(10) + sumFLOutsDelayed(11) + sumFLOutsDelayed(12) + sumFLOutsDelayed(13) + sumFLOutsDelayed(14) + sumFLOutsDelayed(15) + sumFLOutsDelayed(16) + sumFLOutsDelayed(17) + sumFLOutsDelayed(18) + sumFLOutsDelayed(19) + sumFLOutsDelayed(20) + sumFLOutsDelayed(21) + sumFLOutsDelayed(22) + sumFLOutsDelayed(23) + sumFLOutsDelayed(24) + sumFLOutsDelayed(25);
  • Altera_Forum's avatar
    Altera_Forum
    Icon for Honored Contributor rankHonored Contributor
    17 years ago

    Hello,

    it's clear to me, that in this case coefarr can't be inferred as internal RAM, cause the array elements are accessed simultanously. RAM storage locations can only be accessed sequentially, one address each clock cycle. E. g. with fast FIR filters, where all coefficients must be accessed each clock cycle, RAM can't be used. You can consult the Altera FIR compiler manual or evaluate by creating a filter design for trial, how Altera FIR filter tools manages coefficient storage.

    Regards,

    Frank