Make SkMatrix44::invert() check for finite 1/det instead of magic value

Make SkMatrix44::invert() check for finite 1/det instead of magic value

Previously we were checking to see if the magnitude of
determinant of the matrix to be inverted was less than 
1.0e-8, which is a magic number possibly plucked from 
Graphics Gems. After some discussion, it's been determined 
(ha) that we can simply check to see if 1/det is finite and 
if so proceed.

BUG=222926
Committed: http://code.google.com/p/skia/source/detail?r=10758

Committed: http://code.google.com/p/skia/source/detail?r=10818

[jvanverth1, 2013-08-15 19:26:17]

Shawn, just to move things along I've created a new change with the 1/det check.
I can look at the affine change tomorrow.

[reed1, 2013-08-15 19:39:28]

assuming there are no machines that halt on divide-by-zero, lgtm

[reed1, 2013-08-15 19:40:46]

suggested comment for the code

https://codereview.chromium.org/22904003/diff/1/src/utils/SkMatrix44.cpp
File src/utils/SkMatrix44.cpp (right):

https://codereview.chromium.org/22904003/diff/1/src/utils/SkMatrix44.cpp#newc...
src/utils/SkMatrix44.cpp:532: double invdet = 1.0 / det;
// If det is zero, we want to return false. However, we also want to return
false
// if 1/det overflows to infinity (i.e. det is denomralized). Both of these are 
// handled by checking that 1/det is finite.

[shawnsingh, 2013-08-15 19:40:57]

Nice - this is essentially what I have locally too, but you uploaded first =)

[shawnsingh, 2013-08-15 19:42:42]

On 2013/08/15 19:40:57, shawnsingh wrote:
> Nice - this is essentially what I have locally too, but you uploaded first =)

I do have one additional thought though - are we sure isfinite() is supported in
Windows?  I was under the impression that isfinite() is technically not standard
until c++11

[reed1, 2013-08-15 19:45:20]

On 2013/08/15 19:42:42, shawnsingh wrote:
> On 2013/08/15 19:40:57, shawnsingh wrote:
> > Nice - this is essentially what I have locally too, but you uploaded first
=)
> 
> I do have one additional thought though - are we sure isfinite() is supported
in
> Windows?  I was under the impression that isfinite() is technically not
standard
> until c++11

We should write our own anyways -- almost always faster...

bool isfinite(double x) {
    double y = x * 0;
    return y == y;
}

[jvanverth1, 2013-08-15 19:51:51]

On 2013/08/15 19:45:20, reed1 wrote:
> On 2013/08/15 19:42:42, shawnsingh wrote:
> > On 2013/08/15 19:40:57, shawnsingh wrote:
> > > Nice - this is essentially what I have locally too, but you uploaded first
> =)
> > 
> > I do have one additional thought though - are we sure isfinite() is
supported
> in
> > Windows?  I was under the impression that isfinite() is technically not
> standard
> > until c++11
> 
> We should write our own anyways -- almost always faster...
> 
> bool isfinite(double x) {
>     double y = x * 0;
>     return y == y;
> }

I just replaced it with sk_float_isfinite(), which uses _finite() on Windows.
However, I can change it to use that method.

[commit-bot: I haz the power, 2013-08-15 21:24:52]

CQ is trying da patch. Follow status at
https://skia-tree-status.appspot.com/cq/jvanverth@google.com/22904003/15001

[commit-bot: I haz the power, 2013-08-15 21:56:41]

Change committed as 10758

[scroggo, 2013-08-15 23:55:14]

On 2013/08/15 21:56:41, I haz the power (commit-bot) wrote:
> Change committed as 10758

Reverted as 10760, because it broke Matrix44Test on
Test-Ubuntu12-ShuttleA-ATI5770-x86-Release

[jvanverth1, 2013-08-20 13:50:52]

Tweaked one of the matrix tests for i386-Linux. This should
allow this change to go through.

[commit-bot: I haz the power, 2013-08-20 14:57:49]

CQ is trying da patch. Follow status at
https://skia-tree-status.appspot.com/cq/jvanverth@google.com/22904003/22001

[commit-bot: I haz the power, 2013-08-20 15:25:14]

Change committed as 10818