Re: [gem5-users] question about reading out a whole DRAM row data

Tue, 29 Jan 2013 07:00:49 -0800 

Hi Andreas,

Thanks a lot for the information. I will take a look.

-Tao

On 01/29/2013 03:47 AM, Andreas Hansson wrote:

Hi Tao, Shuchang,
Please note that there are uncacheable requests that do not necessarily have the size 64 bytes. Thus, rather use bytesPerCacheLine than pkt->getSize() for this. 

Andreas
From: ???????????? <[email protected] <mailto:[email protected]>> Reply-To: gem5 users mailing list <[email protected] <mailto:[email protected]>> 
Date: Tuesday, 29 January 2013 01:58
To: gem5 users mailing list <[email protected] <mailto:[email protected]>> 
Subject: Re: [gem5-users] question about reading out a whole DRAM row data

Hi Tao:

Thanks for your reply. :)


Shuchang
2013/1/28 Tao Zhang <[email protected] <mailto:[email protected]>> 

    Hi Shuchang,

    If we assume the memory space in both host machine and simulator
    is continuous, then we can calculate the row address as follows.

    unsigned myRowSize = pkt->getSize() * linesPerRow;

    uint8_t* myRowAddr = pmemAddr + (pkt->getAddr() - range.start) /
    myRowSize * myRowSize;

    or

    uint8_t* myRowAddr = pmemAddr + (pkt->getAddr() - pkt->getAddr() %
    myRowSize) - range.start;

    -Tao


    On 01/28/2013 03:18 AM, ???????????? wrote:

    Hi Tao, thanks for your quickly reply.

    I have read the codes of function access(PacketPtr pkt) in
    abstract_mem.cc. The main difference is that, in the given
    function access(), the operation is the granularity of a cache
    line (64-Byte, actually it is a specified column of a whole row;
    while for my codes, I want to copy the whole row in to data
    structure row_data (That's why I need to multiply another 64.
    Here the 64 equals the parameter linesPerRowBuffer).

    In other words, in the function access(), memory access is to
    only one column of a row, while I want to access the entire row.

    My concern is that, whether myHostAddr can point to the start
    address of the row (given pkt->getAddr)?


    /=======================================================================/
    /uint8_t *myHostAddr = pmemAddr + (pkt->getAddr() -
    pkt->getAddr() % linesPerRow) - range.start;/
    /=======================================================================/
    /
    /
    Regards,
    Shuchang

    2013/1/28 Tao Zhang <[email protected]
    <mailto:[email protected]>>

        Hi Shuchang,

        In my opinion, your code are almost correct expect that the
        pkt->getSize() does return the cacheline size (64 in byte).
        You don???t need to multiply another 64 again.

        In fact, AbstractMemory has the function ???access(PacketPtr
        pkt)???in which the data is also copied. You can read the
        code between line 352 and 372 in abstract_mem.cc to see how
        the data is copied. It does almost the same thing as you do.
        As a result, you can simply extend the AbstractMemory and use
        the access() to make sure the data is in
        pkt->getPtr<uint8_t>(); What you only need to do is to copy
        this data to your raw data pointer.

        -Tao

        *From:*[email protected]
        <mailto:[email protected]>
        [mailto:[email protected]
        <mailto:[email protected]>] *On Behalf Of *????????????
        *Sent:* Sunday, January 27, 2013 6:31 AM
        *To:* [email protected] <mailto:[email protected]>
        *Subject:* [gem5-users] question about reading out a whole
        DRAM row data

        Hi all,

        I am trying to modify the SimpleDRAM module to profile the
        data contents in phymem.

        In simple_dram.cc, the /docodeAddr()/ function shows how to
        get the rank and the bank addresses for given request /pkt/.

        I wonder, given/pkt->getAddr()/, how to calculate the row
        address?

        It seems that, a dram row contains 64 (linesPerRow = 64)
        64-Byte data blocks, and I want to read out the whole row
        data into a self-defined data structure ( namely row_data).

        The detailed codes are like this:

        /uint8_t *myHostAddr = pmemAddr + (pkt->getAddr() -
        pkt->getAddr() % linesPerRow) - range.start;/

        /int row_size = pkt->getSize()*64;/

        /uint8_t row_data[row_size];/

        /memcpy(row_data, myHostAddr , row_size ); // read the whole
        MEM row data out into the array structure: row_data/

        I am not sure whether the calculation of the row address is
        right. Any advice will be appreciated.

        Regards,

        shuchang


        _______________________________________________
        gem5-users mailing list
        [email protected] <mailto:[email protected]>
        http://m5sim.org/cgi-bin/mailman/listinfo/gem5-users



    _______________________________________________
    gem5-users mailing list
    [email protected]  
<mailto:[email protected]>http://m5sim.org/cgi-bin/mailman/listinfo/gem5-users


    _______________________________________________
    gem5-users mailing list
    [email protected] <mailto:[email protected]>
    http://m5sim.org/cgi-bin/mailman/listinfo/gem5-users
-- IMPORTANT NOTICE: The contents of this email and any attachments are confidential and may also be privileged. If you are not the intended recipient, please notify the sender immediately and do not disclose the contents to any other person, use it for any purpose, or store or copy the information in any medium. Thank you. 


_______________________________________________
gem5-users mailing list
[email protected]
http://m5sim.org/cgi-bin/mailman/listinfo/gem5-users

_______________________________________________
gem5-users mailing list
[email protected]
http://m5sim.org/cgi-bin/mailman/listinfo/gem5-users

Reply via email to