Unable to control DSP usage for x_complex multiply/accumulate

[dima2882]

Hi all,

I have some code that basically involves this:

void foo(...) {
    x_complex<float> ch1, ch2, product_0, product_1, combined, C0, C1;

    #pragma HLS RESOURCE variable=product_0 core=FMul_nodsp
    #pragma HLS RESOURCE variable=product_1 core=FMul_nodsp
    #pragma HLS RESOURCE variable=product_0 core=FAddSub_nodsp
    #pragma HLS RESOURCE variable=product_1 core=FAddSub_nodsp
    #pragma HLS RESOURCE variable=combined core=FAddSub_nodsp
    #pragma HLS RESOURCE variable=combined core=FMul_nodsp

    product_0 = C0 * ch1;
    product_1 = C1 * ch2;
    combined = product_0 + product_1;
}

I want no DSP usage here, but the fmul operation ends up using 3 DSP slices each. FAddSub does not. How can DSP slice instantiation be prevented for the multiply case?

[dima2882]

C'mon folks, resource allocation for complex number math is a pretty basic feature - doesn't anyone know?

[dima2882]

Folks - has no one used HLS directives for controlling DSP usage for complex number math? Seriously?!

[dima2882 · Accepted solution]

So, it can be concluded that the Cores utilization directives in HLS simply do not work?