Vertex Attrib configurations now handled as pointers vs. SkSTArrays

Although this is a less flexible way to set up the vertex attribs I believe we should use it due to the performance improvement.


BEFORE w/ geometry-based AA rect drawing:

drawpointsdash_5_5_aa   GPU: cmsecs =  24.18 gmsecs =  21.15  NULLGPU: cmsecs =  15.29
drawpointsdash_3_1_aa   GPU: cmsecs =  34.48 gmsecs =  31.55  NULLGPU: cmsecs =  23.56
drawpointsdash_1_1_aa   GPU: cmsecs =  75.66 gmsecs =  72.86  NULLGPU: cmsecs =  62.40
              aarects   GPU: cmsecs =  41.18 gmsecs =  37.85  NULLGPU: cmsecs =  34.48


AFTER w/ geometry-based AA rect rendering:

drawpointsdash_5_5_aa   GPU: cmsecs =  22.00 gmsecs =  19.48  NULLGPU: cmsecs =  14.51
drawpointsdash_3_1_aa   GPU: cmsecs =  31.98 gmsecs =  29.20  NULLGPU: cmsecs =  22.46
drawpointsdash_1_1_aa   GPU: cmsecs =  69.42 gmsecs =  67.76  NULLGPU: cmsecs =  58.50
              aarects   GPU: cmsecs =  38.38 gmsecs =  35.25  NULLGPU: cmsecs =  32.45

---------------------------------------------------------

BEFORE w/ shader-based rect rendering:

drawpointsdash_5_5_aa   GPU: cmsecs =  27.30 gmsecs =  24.61  NULLGPU: cmsecs =  18.25
drawpointsdash_3_1_aa   GPU: cmsecs =  40.87 gmsecs =  38.79  NULLGPU: cmsecs =  28.70
drawpointsdash_1_1_aa   GPU: cmsecs =  94.54 gmsecs =  93.06  NULLGPU: cmsecs =  77.06
              aarects   GPU: cmsecs =  55.07 gmsecs =  52.35  NULLGPU: cmsecs =  46.02

AFTER w/ shader-based rect rendering:

drawpointsdash_5_5_aa   GPU: cmsecs =  25.43 gmsecs =  22.67  NULLGPU: cmsecs =  17.00
drawpointsdash_3_1_aa   GPU: cmsecs =  38.22 gmsecs =  35.65  NULLGPU: cmsecs =  26.21
drawpointsdash_1_1_aa   GPU: cmsecs =  85.33 gmsecs =  83.10  NULLGPU: cmsecs =  70.51
              aarects   GPU: cmsecs =  44.93 gmsecs =  43.80  NULLGPU: cmsecs =  39.00

[bsalomon, 2013-04-17 19:04:34]

On 2013/04/17 18:52:26, robertphillips wrote:

This means the vertex attribs always have to a ptr to static const?!

[robertphillips, 2013-04-17 19:13:15]

If you value your sanity - yes. Basically the caller has to guarantee that the
pointer will remain valid as long as required. One could dynamically allocate a
configuration and then hold onto it until the end of time.

These could become ref counted but that seemed overkill for the current usage
pattern.

[robertphillips, 2013-04-19 16:17:26]

PTAL - I have templatized setVertexAttribs on GrDrawState

[bsalomon, 2013-04-19 18:25:13]

Is it possible to declare the arrays const?

https://codereview.chromium.org/14328009/diff/5001/src/gpu/GrDrawState.h
File src/gpu/GrDrawState.h (right):

https://codereview.chromium.org/14328009/diff/5001/src/gpu/GrDrawState.h#newc...
src/gpu/GrDrawState.h:124: *  should be a global GrVertexAttrib that is never
changed.
global GrVertexAttrib ^array that is never changed?

[robertphillips, 2013-04-19 19:04:47]

Nope - that made me sad too. Here is the error message I get (on Windows) when I
try to make one of the arrays const:

1>..\..\src\gpu\GrTextContext.cpp(219): error C2970:
'GrDrawState::setVertexAttribs' : template parameter 'A' :
'`anonymous-namespace'::gTextVertexAttribs' : an expression involving objects
with internal linkage cannot be used as a non-type argument

The MSDN docs on this error say you can't use a static but don't say anything
about consts. My guess is the linkage changes when the const is added.

https://codereview.chromium.org/14328009/diff/5001/src/gpu/GrDrawState.h
File src/gpu/GrDrawState.h (right):

https://codereview.chromium.org/14328009/diff/5001/src/gpu/GrDrawState.h#newc...
src/gpu/GrDrawState.h:124: *  should be a global GrVertexAttrib that is never
changed.
On 2013/04/19 18:25:13, bsalomon wrote:
> global GrVertexAttrib ^array that is never changed?

Done.

[bsalomon, 2013-04-19 19:10:35]

lgtm w/ extern const

[commit-bot: I haz the power, 2013-04-20 11:02:56]

CQ is trying da patch. Follow status at
https://skia-tree-status.appspot.com/cq/robertphillips@google.com/14328009/5003

[robertphillips, 2013-04-20 12:26:40]

committed ar r8787