[Piglit] [PATCH 4/7] glsl-1.10 / glsl-1.20: Don't use ortho in variable indexing tests.

[Matt Turner]

On Tue, Nov 5, 2013 at 10:57 AM, Kenneth Graunke <kenneth at whitecape.org> wrote:
> On 11/04/2013 10:27 PM, Matt Turner wrote:
>> Using ortho makes shrinking the window size much more difficult, which
>> makes running tests in simulation take much longer.
>> ---
>>  tests/spec/glsl-1.10/variable-index-read.sh  | 18 +++---------------
>>  tests/spec/glsl-1.10/variable-index-write.sh | 17 +++--------------
>>  2 files changed, 6 insertions(+), 29 deletions(-)
>
> I'm actually not a huge fan of patches 4-5.  These tests draw multiple
> squares, each of which represents something different.  Currently, you
> can run the test and see each of them, side-by-side.
>
> With these patches, all the squares get drawn on top of each other,
> which makes it impossible to visually inspect failures without editing
> the test.  That's a pretty big drop in usability, IMHO.

Except that you have to parse a bunch of widths/heights and x/y
coordinates to figure out which square is doing what. Every time I've
worked with a test like this I wind up commenting out the passing
cases any way to help debug.

> A large window size actually isn't necessarily awful for simulation,
> either.  You pay for the clear, but that's not very complicated.  It's
> drawing the 250x250 square that's awful, since it requires 62,500
> non-trivial pixel shader invocations, each of which probably does
> control flow, variable indexing, or such.  Drawing a 10x10 square only
> requires 100 PS invocations, which is obviously much cheaper.

My claim wasn't that this speeds up simulation. It was that by
removing the ortho use we can more easily shrink the window sizes
without having to reason about whether it's safe.

But I did this for you, so if you don't think it's useful I'll throw them out.