Wand threads option not working (?)

[EucliTs0]

Greetings,

Using Wand to convert PDF to PNG we want to speed up the process. We enabled the threads in the policy.xml file, but with any number of threads, there is not any significant difference in the processing.

We also modified the threads using the wand.resource module in python (limits['thread']) (the policy.xml) in this case is untouched). We print the time processing using different threads:

Threads	Time
1	40.6 s
2	44.5 s
3	48.1 s
4	42.4 s
5	40.4 s
6	42.4 s
7	43.8 s
8	43.8 s

Wand 0.6.3
ImageMagick 6.9.10-23 Q16 x86_64 20190101 https://imagemagick.org

Ubuntu 20.04
Python 3.8

We expected that with increasing threads we would get less time processing.

import io 
from wand import color, image
from wand.resource import limits
from time import time


def pdf_to_image(pdf_blob):

    fmt = ''

    with image.Image(blob=pdf_blob, resolution=300, depth=8, format=fmt) \
            as all_pages:
        images_seq = []
        for page in all_pages.sequence:
            with image.Image(page) as img:
                img.format = 'png'
                img.background_color = color.Color('white')
                img.alpha_channel = 'remove'
                images_seq.append(io.BytesIO(img.make_blob("PNG")))

        return images_seq


pdf_file = 'TAX_ASSESSMENT_0059.pdf'

with open(pdf_file, 'rb') as f:
    pdf_bytes = f.read()
    

limits['thread'] = 8 # Change it from 1 to 8

print(list(limits.items()))

start = time()
result = pdf_to_image(pdf_bytes)
print(time() - start)

The PDF has 13 pages, but due to confidentiality I cannot attach it.

Does someone has any insights on that?

Thank you in advance.

[emcconville]

Yep. The number of threads has no impact with decoding.

Wand interacts with ImageMagick, and for all PDF encoding/decoding tasks are offloaded to Ghostscript. As you're not doing any raster manipulation, you may be able to speed up the process by working directly with the gs utility.

gs -sstdout=%stderr -dQUIET -dSAFER -dBATCH -dNOPAUSE -dNOPROMPT \
   -dMaxBitmap=500000000 -dAlignToPixels=0 -dGridFitTT=2 -sDEVICE=png16m' \
   -r300x300  -sOutputFile=/tmp/page_%04d.png TAX_ASSESSMENT_0059.pdf

For Python, I would suggest the following...

• Create a temporary directory to hold the PNG files.
• Invoke Ghostscript (like the example above) via Subprocess.
• Glob the temporary directory for any generated PNG files.

[EucliTs0]

Thank you for your answer. Since the number of threads has no impact, is there any way possible to use several CPUs when converting PDF to images?

[emcconville]

... Updated answer with more detail.

[...] is there any way possible to use several CPUs when converting PDF to images?

I think there's a ghostscript option -dNumRenderingThreads=8 but If I remember correctly, the overhang of working with threads doesn't really start to speed things up until your working with dozens of pages. But try it out!

[EucliTs0]

Thank you for the detailed answers! I will try them and I will come back if I have other question.