These variables and methods should make it easier to loop through buffers elements/pixels. They take into account single element buffers.
Single element buffers can be used for set operations to reduce memory usage.
Usage example: P2078
Details
Details
- Reviewers
Jeroen Bakker (jbakker) - Group Reviewers
Compositing - Commits
- rBf966f6ed55f7: Compositor: Add vars and methods for easier image looping
Diff Detail
Diff Detail
- Repository
- rB Blender
- Branch
- cmp-loop-image-vars (branched from master)
- Build Status
Buildable 14270 Build 14270: arc lint + arc unit
Event Timeline
Comment Actions
What do you think of adding some BLI_asserts?
stride and offset are more common words than jumps.
get_offset might then become something else.
naming single elem is a
We need to see how we could eliminate some logic so the compiler could optimize them. (but that is not related to this patch).
Some methods should be implemented in the header so the compiler could optimize these better.
Another idea is to use a template based on the num of channels.
| source/blender/compositor/intern/COM_MemoryBuffer.cc | ||
|---|---|---|
| 39 | can't we decode is_single_elem from the rect. or do we mean constant data. | |
| source/blender/compositor/intern/COM_MemoryBuffer.cc | ||
|---|---|---|
| 39 | Cannot because when is_a_single_elem=true, buffer may virtually have any resolution. For example a SetColorOperation buffer may be: m_rect={0,500,0,500}, getWidth()=500, get_memory_width()=1, elem_offset=0, row_offset=0, is_a_single_elem=trueThis allows writing an output buffer without worrying if inputs are single elem buffers, just need to correctly use their offset variables. | |
Comment Actions
I agree. I'd like to make it generic too, not only for floats. And maybe moving pixel operations methods like readEWA and read bilinear to a header specific for pixel operations. We can see this later on.
Comment Actions
Naming is important so good spend just another round to have this correct and consistent. Rest of the patch is fine.
| source/blender/compositor/intern/COM_MemoryBuffer.h | ||
|---|---|---|
| 61 | Isn't an offset, but a stride. elem_stride | |
| 70 | same here. eg buffer_index = y * buffer.row_stride + x * buffer.elem_stride; In this case an offset can be used as a channel selector. (offset of second channel) buffer_index = y * buffer.row_stride + x * buffer.elem_stride + offset_channel; Just to keep things aligned to other areas. | |