[Openexr-devel] Deep image input data memory layout.

[Thomas Loockx]

Hi,

I'm trying to integrate OpenEXR deep pixel rendering into our render engine.

We ran into a crash in copyFromDeepFrameBuffer in this snippet (more
specifically in line 1598 of ImfMisc.cpp).

          case OPENEXR_IMF_INTERNAL_NAMESPACE::FLOAT:

            for (int x = xMin; x <= xMax; x++)
            {
                unsigned int count =
                        sampleCount(sampleCountBase,
                                   sampleCountXStride,
                                   sampleCountYStride,
                                   x - xOffsetForSampleCount,
                                   y - yOffsetForSampleCount);
                const char* ptr = base + (y-yOffsetForData) *
dataYStride + (x-xOffsetForData) * dataXStride;

                const char* readPtr = ((const char**) ptr)[0];
                for (unsigned int i = 0; i < count; i++)
                {
                    Xdr::write <CharPtrIO> (writePtr, *(const float *)
readPtr); // CRASH HERE
                    readPtr += sampleStride;
                }
            }
            break;

My understanding here is that OpenEXR internally interprets a deep
pixel in a channel as a pointer to an array of floating point values.
Unfortunately our memory isn't layed out like that. We allocate a
single memory chunk for the
full deep image. I tried several ways of creating the deep slice but I
can't seem to code my way around it.

Am I forced to copy all my data in arrays (or at least allocate a
separate buffer of float* pointers that point into my data) or is
there a way around this?

thanks,
Thomas Loockx