get tests closer to passing for SKIA_SCALAR=fixed

See http://code.google.com/p/skia/issues/detail?id=263 ('fix libskia build for SKIA_SCALAR=fixed')

Still doesn't quite pass with SKIA_SCALAR=fixed, but getting closer.

To reproduce the build/run:

export SKIA_DEBUG=true; export SKIA_SCALAR=fixed ; make clean ; make out/libskia.a ; make tests ; out/tests/tests

[epoger, 2011-05-17 14:41:30]

Ready for review, for reals this time.

[reed1, 2011-05-17 14:53:08]

http://codereview.appspot.com/4532064/diff/6001/gpu/src/GrPathUtils.cpp
File gpu/src/GrPathUtils.cpp (left):

http://codereview.appspot.com/4532064/diff/6001/gpu/src/GrPathUtils.cpp#oldco...
gpu/src/GrPathUtils.cpp:72: d = ceilf(sqrtf(d/tol));
SkScalarCeil returns an int, not a float. That is probably OK, since the
compiler will auto-promote it back to float for d...

In fixed, making d an int is a error, since we just effectively reduced it by
1/65536. That basically speaks as much to this line as to the next...
GrNextPow2((uint32_t)d)

http://codereview.appspot.com/4532064/diff/6001/tests/MatrixTest.cpp
File tests/MatrixTest.cpp (left):

http://codereview.appspot.com/4532064/diff/6001/tests/MatrixTest.cpp#oldcode8
tests/MatrixTest.cpp:8: const int32_t tolerance = 8;
Can also express this (and the float version btw) as a fraction...

SK_Scalar1 / 1000 (or whatever)

[epoger, 2011-05-17 15:44:29]

http://codereview.appspot.com/4532064/diff/6001/gpu/src/GrPathUtils.cpp
File gpu/src/GrPathUtils.cpp (left):

http://codereview.appspot.com/4532064/diff/6001/gpu/src/GrPathUtils.cpp#oldco...
gpu/src/GrPathUtils.cpp:72: d = ceilf(sqrtf(d/tol));
On 2011/05/17 14:53:08, reed1 wrote:
> SkScalarCeil returns an int, not a float. That is probably OK, since the
> compiler will auto-promote it back to float for d...
> 
> In fixed, making d an int is a error, since we just effectively reduced it by
> 1/65536. That basically speaks as much to this line as to the next...
> GrNextPow2((uint32_t)d)

Done.

http://codereview.appspot.com/4532064/diff/6001/tests/MatrixTest.cpp
File tests/MatrixTest.cpp (left):

http://codereview.appspot.com/4532064/diff/6001/tests/MatrixTest.cpp#oldcode8
tests/MatrixTest.cpp:8: const int32_t tolerance = 8;
On 2011/05/17 14:53:08, reed1 wrote:
> Can also express this (and the float version btw) as a fraction...
> 
> SK_Scalar1 / 1000 (or whatever)

Done.

[epoger, 2011-05-17 16:01:47]

Rookie error: I forgot to rerun upload.py last time.

[reed1, 2011-05-17 16:22:29]

LGTM

[epoger, 2011-05-17 17:37:44]

committed as r1351