Change Details

IMB_rect_from_float(...) and IMB_float_from_rect(...) are called from threaded code on a shared ibuf. This is a race condition and leads to wrong output when projecting float data onto a uchar texture or vice versa on a multithreaded system. I created a bugreport T80859 earlier.

I created a bugreport T80859 earlier. The problem is in the file source/blender/editors/sculpt_paint/paint_image_proj.c (line numbers against commit d376aea61) at the lines 5309 and 5320 the functions `IMB_rect_from_float(...)` and `IMB_float_from_rect(...)` are called from threaded code on a shared ibuf. However, ps->reproject_ibuf is shared between threads, so this is not thread-safe. This is a race condition and leads to wrong output when projecting float data onto a uchar texture or vice versa on a multithreaded system. The checks on line 5308 and 5319 are already a race condition. I created this patch which fixes the problem , but I'm not sure if I'm overlooking something. This makes sure reproject_ibuf contains the correct formats *before* the threadpool is started. I replaced the original location of the conversions with BLI_asserts(). That may be overkill?

IMB_rect_from_float(...) and IMB_float_from_rect(...) are called from threaded code on a shared ibuf. This is a race condition and leads to wrong output when projecting float data onto a uchar texture or vice versa on a multithreaded system. I created a bugreport T80859 earlier. The problem is in the file source/blender/editors/sculpt_paint/paint_image_proj.c (line numbers against commit d376aea61) at the lines 5309 and 5320 the functions `IMB_rect_from_float(...)` and `IMB_float_from_rect(...)` are called from threaded code on a shared ibuf. However, ps->reproject_ibuf is shared between threads, so this is not thread-safe. This is a race condition and leads to wrong output when projecting float data onto a uchar texture or vice versa on a multithreaded system. The checks on line 5308 and 5319 are already a race condition. I created this patch which fixes the problem , but I'm not sure if I'm overlooking something. This makes sure reproject_ibuf contains the correct formats *before* the threadpool is started. I replaced the original location of the conversions with BLI_asserts(). That may be overkill?