add exrmetrics tool #1769
add exrmetrics tool #1769
Conversation
Signed-off-by: Peter Hillman <[email protected]>
Signed-off-by: Peter Hillman <[email protected]>
Signed-off-by: Peter Hillman <[email protected]>
Signed-off-by: Peter Hillman <[email protected]>
Signed-off-by: Peter Hillman <[email protected]>
Signed-off-by: Peter Hillman <[email protected]>
Signed-off-by: Peter Hillman <[email protected]>
src/bin/exrmetrics/main.cpp
Outdated
| std::string compressionNames; | ||
| getCompressionNamesString ("/", compressionNames); | ||
| stream | ||
| << "Read a part OpenEXR image from infile, write an identical copy to outfile" |
There was a problem hiding this comment.
do you mean "Read an OpenEXR image part from infile..."?
| "\n" | ||
| "Options:\n" | ||
| "\n" | ||
| " -p n part number to copy (only one part will be written to output file)\n" |
There was a problem hiding this comment.
Any reason to not just do all parts? Seems like in the field you'd like to run this on all parts, and observe the effect on the files as a whole, even if the compression stats are specific to individual parts.
There was a problem hiding this comment.
It's certainly doable, but there are a bunch of issues: since the parts can have different types, and different compression codecs, generating meaningful output gets messy. The C++ API doesn't give access to the size of individual compressed tiles/scanlines, so the total file size is quoted instead. It would be tricky to quote the compressed size of each part, but some data only makes sense if it's part-specific. A custom stream, or the Core API, might do a better job of that.
Currently, for all types, exrmetrics reads all the data in one go, then writes them all out, timing the total read and write time, since that should give the most reliable statistics. That gets messy to implement if it has to read data all parts first, then write them all out again, using part-specific code for each type.
Signed-off-by: Peter Hillman <[email protected]>
peterhillman
changed the title
Add tool to measure read/write times of OpenEXR images and compression performance, by reading entire file into memory, then writing it back out again. Zip and DWA compression level can be adjusted. Currently only regular and deep scanline images are supported; tiled images are not implemented yet.
multithreading is not used because that would give inconsistent timings (and is dependent on threading strategy)
The expected usage would be to run exrmetrics twice for each file: first, reading the source image and writing a temporary file with
-z <test_codec>, to get write speed of the and file size of the codec on the compressed file, then use that output as input for a run with-z noneto get the read speed of the test codec and baseline file size and write speed.