Blender Git Loki

Fix OpenGL backtrace build errors, without disabling warnings.

Tentative compile fix after recent gpu debug changes

OpenGL: debug context on Windows

Enable based on --debug-gpu at the command line. Linux already works
this way.

This commit in master (2.78 timeframe) is the smallest change that gets
the desired result. I did a similar commit in blender2.8. Most ongoing
GL debug work will go into 2.8, not 2.7x.

In support of T49089

OpenGL: backtrace on errors (--debug-gpu)

Backtrace so we can pinpoint where the GL error came from. Then fflush
on severe errors in case it's severe enough to crash Blender.

OpenGL: enhance debug output

When running blender --debug-gpu

Display which debug facilities are available. One of these, in order of preference:
- OpenGL 4.3
- KHR_debug
- ARB_debug_output
- AMD_debug_output

All messages are logged now, not just errors. Will probably turn some of these off later.

GL_DEBUG_OUTPUT_SYNCHRONOUS lets us break on errors and backtrace to the exact trouble spot.

Callers of GPU_string_marker no longer pass in a message length, just the message itself (null terminated).

Apple provides no GL debug logging features.

OpenGL: backtrace on errors (--debug-gpu)

Backtrace so we can pinpoint where the GL error came from. Then fflush
on severe errors in case it's severe enough to crash Blender.

Revert "Cycles: Use half storage for 16bit files (e.g. 16bit pngs) as well."

After discussion in IRC, 16bit int should not be converted to half, too much precision loss.

Fix CUDA compilation after OpenCL changes.

Cycles: Use half storage for 16bit files (e.g. 16bit pngs) as well.

Note: This only works for textures loaded from disk via OIIO, not packed textures. That's still a ToDo.

Cycles microdisplacement: Move call to tessellate() from addon to Cycles

By calling `tessellate()` from the mesh manager in Cycles we can do pre/post
processing or even threaded tessellation without concerning client side code
with the details.

Cycles: Add single channel texture support for OpenCL.

This way OpenCL devices can also benefit from a smaller memory footprint, when using e.g. bumpmaps (greyscale, 1 channel).

Additional target for my GSoC 2016.

get latest OpenGL version on Windows + AMD

When we ask for GL 3.2 compatibility profile:
AMD (Radeon HD 6970) gives us exactly this version
NVIDIA (Quadro K600) gives at least this version
Still need to check Intel behavior

We want *at least* the version requested, plus more recent features if
available.

Both GPUs tested & mentioned above are capable of GL 4.5. With this
commit they both give 4.5 to Blender.

early out for ortho grid drawing

Helps most when real-world units are used.

Previous code started at the smallest visible unit (e.g. Inches) then
followed to Feet, Yards, Chains, Furlongs, Miles. Always to the largest
unit of the set, even though most would be way off screen.

New code knows whether it skipped any grid lines for the next unit to
fill in, can stop once all lines are on screen.

Mac fixes for new ortho grid drawing

Previous commit works on Windows, found some issues after trying on Mac.

- benign warnings about && within ||
- replaced nearbyint() with round() to avoid floating point environment
surprises
- remquo function appears to be broken on Mac (!) results were way way
off. Replaced with simple division.
- minor tweaks to debug output

draw ortho grid with new immediate mode

Work toward T49043, with a side of client vertex arrays.

Not a straightforward port from glVertex to immVertex since Gawain needs
to know how many vertices we'll be drawing *before* we start drawing.

Fixed these not-so-great aspects of grid drawing:
- coarse grids would draw atop some lines from the finer grids
- visible axes would draw atop lines from coarse grid
- axes were drawn even if they weren't in view
- terrible misuse of vertex arrays
- each line issued its own draw call

New code draws each line exactly once. The entire grid is one draw call.

Bonus: I had to / got to learn how the units system works!

Gawain: increase size of immediate mode buffer

New value of 4MB should handle our needs without taking up too many GPU
resources.

Old value of 1KB was for observing what happens when the buffer fills up
and we need to flush and start a new one.

Prevent max stiffness values from going under normal stiffness values in cloth stiffness scaling.

When updating the max values under stiffness scaling, they clip at the normal stiffness values
as expected, however when updating stiffness values, you could set them higher than the max
values, and the max values weren't updated accordingly. As the stiffness scaling computes using
the absolute difference between the max values and the stiffness values, you got higher
stiffnesses in scaled areas even though your max is actually lower than the normal stiffness.

This diff fixes that behaviour, by updating the max values to be equal to the stiffness whenever
you set a higher stiffness than the max value.

Also, I have initialized the max values to the same as the stiffnesses, as they were previously
just set to zero, and caused the same problem described above.

Reviewers: lukastoenne

Reviewed By: lukastoenne

Tags: #physics

Differential Revision: https://developer.blender.org/D2147

GPencil: Cleanup code

Fix T49082: Intermediate update call when making links can change group socket layout.

Node tree update calls in the middle of a socket loop are dangerous, they can change sockets
on group nodes and link instances in particular. Updates should only happen after the operator
has finished.

Simply removed the extra convenience check for validity now. Worst case an invalid (red) link
is created which can be removed by the user as well and should simply be ignored by node systems.

The update system in nodes needs a complete rewrite to handle complex cases like this, where an
operator may need to react to changes during its execution.

Add DNA_struct_find (useful for version patching)