Reviewers: , Description: By definitions, closures and non-floats can't have derivs. Oops. Please review this at http://codereview.appspot.com/904047/show Affected files: src/liboslexec/exec.cpp src/l

Guys, I just noticed a difference in normalization for the closures in liboslexec and the code snippets provided in the specification. For instance, WardClosure::eval_reflect looks like it implements

Question for the wonderful OSL developers: I'm wondering why in a surface shader N is read-write, but u and v are read-only. Is there a specific benefit gained by doing this? In general, I'm wondering

Reviewers: , Description: An OSL construct such as shader foo ( float bar[4] = {0} ) where fewer array elements than the total were given defaults, would result in only the number of defaults specifie

Reviewers: , Description: First attempt at inter-node runtime optimizations. First, there is some refactoring in runtimeoptimize.cpp, I hope it's relatively straightforward to read. Also, I added opti

Yes, I want to extend the functionality to do the right thing for constant spaces, too. It may be ugly in some eyes, but we do exactly this in many other places (including the interpretive shadeop for

Reviewers: , Description: Several miscellaneous fixes to runtime optimization: * Constant fold float*triple and triple*float (previously we only folded f*f and t*t). * Make constant folding of divisio

All the other constant-folding routines want to know the local constant aliases of their *readable* arguments, and doesn't care about the writable arguments (in fact, we want writable args to have had

Reviewers: , Description: Another round of runtime optimization improvements: * A=B-B => A=0, even if B is not a constant * More type cases for div to constant fold (i/i, f/i, i/f, v/f, f/v) * Make ne

Reviewers: , Description: Refactor the increasingly cluttered runtime optimization into a RuntimeOptimizer class that does most of the dirty work, and by maintaining some state, now no longer needs to

Reviewers: , Description: New runtime optimization: Track which messages are set by setmessage(), including if a non-constant named message is ever set (could be anything). Then for getmessage() of a

Remember that we're making multiple passes over each layer, whittling it down. On the first pass, we may see a setattribute, and if we mark it in "global" list of potentially set attributes, it's ther

Reviewers: , Description: Make a safe_pow that gracefully handles under/overflow. Also try to do pow with derivs as sanely as possible, and fix a bug in the Dual pow wherein pow(x<0,y integer) would e

Reviewers: , Description: Otherwise it crashes when testing shaders that access these globals. Also renamed the strage gP_dx, gP_dy to gdP_dx, gdP_dy. Please review this at http://codereview.appspot.c

Reviewers: , Description: Constant folding for pow. Note that this depends on a separate review for "safe_pow". Please review this at http://codereview.appspot.com/1124041/show Affected files: src/lib

Reviewers: dev-osl_imageworks.com, osl-dev_googlegroups.com, Description: Here at Imageworks we're building OSL for different versions of Linux. This patch simplifies the process of managing those bui

This clause is in the u<0 case. That's only well defined for integral values of v. The function is discontinuous when u<0, even for values of v where it is defined. So there are no derivs to be had. I

Hello OSLers, Mainly just for my fun(To understand OSO format), I've made simple python script which visualizes the dependences of OSO instructions using graphviz. http://github.com/syoyo/OSOgraph AFA

thanks. You mean for non-constant X? I dunno. What transformation would you recommend to speed up constant integral Y and unknown x? pow(x,0) and pow(x,1) are already handled. The only other common on

Reviewers: osl-dev_googlegroups.com, Description: This is our regular expression engine for light expressions. The automaton is created using AccumAutomata (which is immutable once you call compile) a