Forum Discussion

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

What can I do for 'Main Optimizer failed' error in Intel HLS?

Hi,

My component has two stream in inputs and one stream out output.

The two inputs pass through the Shift register and perform multiply operations between the elements.

x86 execution is not a problem. An error such as the following occurs when executing build.bat test-fpga.

"Instruction does not dominate all users!"

One stream in is a blocking read, and the other stream in is a non-blocking read.

Changing both data to non-blocking causes the same error.

Shift register has no errors, and errors occur when multiplying the elements.

The data type is ac_fixed, and is an operation of 8bit and 1bit.

I uses Quartus 17.1 version. What can I do?

2 Replies

  • Altera_Forum's avatar
    Altera_Forum
    Icon for Honored Contributor rankHonored Contributor

    Hi,

    Can you share the command prompt log, .cpp & .bat files?

    Best Regards,

    Anand Raj Shankar

    (This message was posted on behalf of Intel Corporation)
  • Altera_Forum's avatar
    Altera_Forum
    Icon for Honored Contributor rankHonored Contributor

    Dear Mr. Anand

    I am sorry. I can't share any files because of our company policy.

    But I add pseudo code. build.bat is changed only cpp filename and -ghdl option for test-fpga.

    # include <>

    ...

    typedef ac_fixed<8,8,false> fixed8;

    typedef ac_fixed<1,1,false> fixed1;

    typedef ac_fixed<10,10,false> fixed10;

    component void func(stream_in<fixed8> a, stream_in<fixed1> b, stream_out<fixed10> c) {

    hls_init_on_reset static fixed8 d[xx];

    hls_init_on_reset static fixed1 e[xx];

    for (){

    // Non blocking read1

    bool success = false;

    fixed8 data = a.tryRead(success);

    if (success) {

    //Shift register

    # pragma unroll

    for ...

    d[] = data;

    }

    // Non blocking read2

    bool success1 = false;

    fixed8 data1 = b.tryRead(success1);

    if (success1) {

    //Shift register

    # pragma unroll

    for ...

    e[] = data1;

    }

    //--------------- The following makes error ----------------

    fixed10 m = d[] *e[];

    ...

    c.write(out);

    }

    }

    int main() {

    //Generate stream in data and buffer

    // Invoke component

    return 0;

    }

    That's all. What can I do?