Fix narrowing conversion errors from r11986

Fix narrowing conversion errors from r11986

Although this function only gets used for SkMScalar=float, it gets
compiled in when SkMScalar=double.

R=epoger@google.com
BUG=1764
Committed: http://code.google.com/p/skia/source/detail?r=11991

[enne (OOO), 2013-10-28 21:14:19]

I think this should fix things.

[epoger, 2013-10-28 21:17:33]

LGTM (assuming trybots pass)

Thanks for the fix, and trying it on the affected trybots.

https://codereview.chromium.org/49463005/diff/1/src/utils/SkMatrix44.cpp
File src/utils/SkMatrix44.cpp (right):

https://codereview.chromium.org/49463005/diff/1/src/utils/SkMatrix44.cpp#newc...
src/utils/SkMatrix44.cpp:822: dst4[i] = SkMScalarToFloat(mat[0][i]) * sx +
SkMScalarToFloat(mat[1][i]) * sy + SkMScalarToFloat(mat[3][i]);
please wrap lines at 100 char

[enne (OOO), 2013-10-28 21:25:52]

BuildSkiaLib passed on the Mac10.6 bot, so will CQ this since the tree is red.

[commit-bot: I haz the power, 2013-10-28 21:26:08]

CQ is trying da patch. Follow status at
https://skia-tree-status.appspot.com/cq/enne@chromium.org/49463005/70001

[commit-bot: I haz the power, 2013-10-28 22:05:46]

Change committed as 11991

[reed1, 2013-10-29 14:58:52]

https://codereview.chromium.org/49463005/diff/70001/src/utils/SkMatrix44.cpp
File src/utils/SkMatrix44.cpp (right):

https://codereview.chromium.org/49463005/diff/70001/src/utils/SkMatrix44.cpp#...
src/utils/SkMatrix44.cpp:823: SkMScalarToFloat(mat[0][i]) * sx +
This technique throws away all of the precision benefits of having doubles in
MScalar. Look at map2_af for an alternate technique. It keeps all intermediates
in doubles (MScalar), and just pays the (perf) cost of the conversion when
writing the final answer.

[epoger, 2013-10-29 15:18:29]

https://codereview.chromium.org/49463005/diff/70001/src/utils/SkMatrix44.cpp
File src/utils/SkMatrix44.cpp (right):

https://codereview.chromium.org/49463005/diff/70001/src/utils/SkMatrix44.cpp#...
src/utils/SkMatrix44.cpp:823: SkMScalarToFloat(mat[0][i]) * sx +
On 2013/10/29 14:58:52, reed1 wrote:
> This technique throws away all of the precision benefits of having doubles in
> MScalar. Look at map2_af for an alternate technique. It keeps all
intermediates
> in doubles (MScalar), and just pays the (perf) cost of the conversion when
> writing the final answer.

Thanks for catching this, Mike.

Do you think it's worth having some unit tests to make sure the precision of
these operations is maintained?

[reed1, 2013-10-29 15:49:12]

On 2013/10/29 15:18:29, epoger wrote:
> https://codereview.chromium.org/49463005/diff/70001/src/utils/SkMatrix44.cpp
> File src/utils/SkMatrix44.cpp (right):
> 
>
https://codereview.chromium.org/49463005/diff/70001/src/utils/SkMatrix44.cpp#...
> src/utils/SkMatrix44.cpp:823: SkMScalarToFloat(mat[0][i]) * sx +
> On 2013/10/29 14:58:52, reed1 wrote:
> > This technique throws away all of the precision benefits of having doubles
in
> > MScalar. Look at map2_af for an alternate technique. It keeps all
> intermediates
> > in doubles (MScalar), and just pays the (perf) cost of the conversion when
> > writing the final answer.
> 
> Thanks for catching this, Mike.
> 
> Do you think it's worth having some unit tests to make sure the precision of
> these operations is maintained?

sgtm

[enne (OOO), 2013-10-29 16:56:37]

https://codereview.chromium.org/49463005/diff/70001/src/utils/SkMatrix44.cpp
File src/utils/SkMatrix44.cpp (right):

https://codereview.chromium.org/49463005/diff/70001/src/utils/SkMatrix44.cpp#...
src/utils/SkMatrix44.cpp:823: SkMScalarToFloat(mat[0][i]) * sx +
On 2013/10/29 14:58:52, reed1 wrote:
> This technique throws away all of the precision benefits of having doubles in
> MScalar. Look at map2_af for an alternate technique. It keeps all
intermediates
> in doubles (MScalar), and just pays the (perf) cost of the conversion when
> writing the final answer.

When SkMScalar is a float, map2_af's technique is slow because of all the float
to double conversions.  I think there either need to be #ifdefs or some use of
SkMScalar as intermediates to keep floats as floats and doubles as doubles
without the conversion cost.

[reed1, 2013-10-29 16:58:42]

On 2013/10/29 16:56:37, enne wrote:
> https://codereview.chromium.org/49463005/diff/70001/src/utils/SkMatrix44.cpp
> File src/utils/SkMatrix44.cpp (right):
> 
>
https://codereview.chromium.org/49463005/diff/70001/src/utils/SkMatrix44.cpp#...
> src/utils/SkMatrix44.cpp:823: SkMScalarToFloat(mat[0][i]) * sx +
> On 2013/10/29 14:58:52, reed1 wrote:
> > This technique throws away all of the precision benefits of having doubles
in
> > MScalar. Look at map2_af for an alternate technique. It keeps all
> intermediates
> > in doubles (MScalar), and just pays the (perf) cost of the conversion when
> > writing the final answer.
> 
> When SkMScalar is a float, map2_af's technique is slow because of all the
float
> to double conversions.  I think there either need to be #ifdefs or some use of
> SkMScalar as intermediates to keep floats as floats and doubles as doubles
> without the conversion cost.

Good observation. Perhaps it should have taken the input X,Y as MScalars instead
of explicitly doubles.