Correct SkFixedToDouble.

Correct SkFixedToDouble.

SkFixedToDouble uses the decimal floating point string representation
of 1/(2^16) to provide a constant value. This value is exactly
representable by a double but the decimal constant currently used
does not map to this value.

This change modifies this constant to more decimal digits so that
the actual double value desired is created.

Committed: https://skia.googlesource.com/skia/+/f4fddfc300f983eb221327fa9b1588c3b08262a6

[bungeman-skia, 2015-04-16 20:41:30]

bungeman@google.com changed reviewers:
+ djsollen@google.com, mtklein@google.com, reed@google.com

[bungeman-skia, 2015-04-16 20:41:31]

This is currently being tried on Blink at
https://codereview.chromium.org/1096543003/ . Based on local tests, this may
affect four svg tests, but I'll wait on the bots to be sure.

https://codereview.chromium.org/1097473002/diff/1/include/core/SkFixed.h
File include/core/SkFixed.h (right):

https://codereview.chromium.org/1097473002/diff/1/include/core/SkFixed.h#newc...
include/core/SkFixed.h:31: #define SkFixedToFloat(x)   ((x) * 1.52587890e-5f)
Note that 1.8 is needed to correctly round trip float, and 1.16 is needed to
correctly round trip double.

This value is wrong, as 1.52587891e-5f is more correct (rounded, what %1.8f
would actually produce). For constants like this, C++ doesn't specify which
rounding rule to use, and various compilers are actually different.
1.52587890e-5f can round down depending. 1.52587891e-5f always maps to the
correct value without issue. However, 1.52587890625e-5 is obvious, exact in
decimal, and is correct. So just do that.

https://codereview.chromium.org/1097473002/diff/20001/include/core/SkFixed.h
File include/core/SkFixed.h (right):

https://codereview.chromium.org/1097473002/diff/20001/include/core/SkFixed.h#...
include/core/SkFixed.h:31: #define SkFixedToFloat(x)   ((x) * 1.52587890625e-5f)
Note that this diff might not cause any change, since I think we got lucky here
and all (current) compilers round the resulting value nicely, mapping this to
the same bit pattern. However, if the compiler didn't round so nicely, the old
value may have been off by one ulp.

https://codereview.chromium.org/1097473002/diff/20001/include/core/SkFixed.h#...
include/core/SkFixed.h:53: #define SkFixedToDouble(x)  ((x) * 1.52587890625e-5)
Note that no one is actually using this at the moment, but we should update this
while it's easy to do so (and I have a test that would like to use it).

[bungeman-skia, 2015-04-16 21:18:49]

The CQ bit was checked by bungeman@google.com to run a CQ dry run

[commit-bot: I haz the power, 2015-04-16 21:19:14]

Dry run: CQ is trying da patch. Follow status at
 https://chromium-cq-status.appspot.com/patch-status/1097473002/20001

[commit-bot: I haz the power, 2015-04-16 21:25:47]

The CQ bit was unchecked by commit-bot@chromium.org

[commit-bot: I haz the power, 2015-04-16 21:25:48]

Dry run: This issue passed the CQ dry run.

[bungeman-skia, 2015-04-17 14:55:42]

According to https://codereview.chromium.org/1096543003/ this doesn't affect any
of the current layout tests, so I think this is good to go in.

[mtklein, 2015-04-17 15:18:07]

lgtm

[bsalomon, 2015-04-17 17:04:51]

bsalomon@google.com changed reviewers:
+ bsalomon@google.com

[bsalomon, 2015-04-17 17:04:51]

lgtm

[mtklein, 2015-04-17 17:05:14]

The CQ bit was checked by mtklein@google.com

[commit-bot: I haz the power, 2015-04-17 17:05:36]

CQ is trying da patch. Follow status at
 https://chromium-cq-status.appspot.com/patch-status/1097473002/20001

[commit-bot: I haz the power, 2015-04-17 17:05:48]

Committed patchset #2 (id:20001) as
https://skia.googlesource.com/skia/+/f4fddfc300f983eb221327fa9b1588c3b08262a6