Problem in nvcomp deflate compression

Accelerated Computing GPU-Accelerated Libraries
,
1

I am using nvcomp to perform a gpu accelerated compression in my c++ code. The goal is to replace zlib compress, that runs on cpu, with a gpu alternatives to produce a png image from a bitmap.
I have the zlib version, and it works fine, the image is ok. I perform the compression with deflate in nvcomp and I got a total black image.

Some information:

  1. The data that I pass both to cpu zlib and to gpu nvcomp are exactly the same. I have performed a control gpu to cpu and viceversa and I have compared the two arrays position per position
  2. The compression does not produce any error, also I got a finite and reasonable dimension of the compressed array size.
  3. The input data is the filtered image, where I put a 0 byte before of each scanline. So the dimension of the input array on GPU (input_device_ptr) is (width + 1) * height.

I will provide here the code that I am using

Configuration part

nvcomp::DeflateManager manager{ (size_t)(width + 1) * height, nvcompBatchedDeflateDefaultOpts};
nvcomp::CompressionConfig comp_config = manager.configure_compression((size_t) (width + 1) * height);
int max_size = comp_config.max_compressed_buffer_size;
uint8_t* comp_buffer;
cudaMalloc(&comp_buffer, comp_config.max_compressed_buffer_size);

Compression part and writing the resulting array back on cpu

manager.compress(input_device_ptr, comp_buffer, comp_config);
size_t comp_size = manager.get_compressed_output_size(comp_buffer);
uint8_t* d_ptr_cpu = new uint8_t[comp_size];
cudaMemcpy(comp_buffer, d_ptr_cpu, comp_size, cudaMemcpyDeviceToHost);

First, is my reasoning right? Second, the nvcomp deflate produces the same file format of zlib?