Fast double-to-ascii conversion.

Fast double-to-ascii conversion.

Committed: http://code.google.com/p/v8/source/detail?r=4106

[Florian Loitsch, 2010-03-11 12:10:42]

Next try. Fixed compiler problems on Windows.
Modified the testing routine since atof on Macosx is buggy.
There is now a file "gay_shortest.cc" containing 100.000 precomputed numbers.
Due to the size it has not been uploaded here. (gcl refused the upload).

[William Hesse, 2010-03-11 13:55:25]

LGTM, but with comments.

http://codereview.chromium.org/866002/diff/36001/43002
File src/grisu3.cc (right):

http://codereview.chromium.org/866002/diff/36001/43002#newcode150
src/grisu3.cc:150: bool Grisu3<alpha, gamma>::DigitGen(DiyFp low, DiyFp w, DiyFp
high,
We in the big room insist that there be one parameter per line, all aligned.  It
makes us make mistakes if there are more than one parameter on a line, we have
decided.

http://codereview.chromium.org/866002/diff/36001/39001
File test/cctest/test-diy_fp.cc (right):

http://codereview.chromium.org/866002/diff/36001/39001#newcode14
test/cctest/test-diy_fp.cc:14: static_cast<int64_t>(b))
I think a compiler may complain if you call
CHECK_EQ64(0x9000000000000001, foo), because you are casting a constant outside
the int64_t range to an int64_t.

So this is not a solution either.

[Florian Loitsch, 2010-03-11 14:18:46]

http://codereview.chromium.org/866002/diff/36001/43002
File src/grisu3.cc (right):

http://codereview.chromium.org/866002/diff/36001/43002#newcode150
src/grisu3.cc:150: bool Grisu3<alpha, gamma>::DigitGen(DiyFp low, DiyFp w, DiyFp
high,
On 2010/03/11 13:55:26, William Hesse wrote:
> We in the big room insist that there be one parameter per line, all aligned. 
It
> makes us make mistakes if there are more than one parameter on a line, we have
> decided.

Done.

http://codereview.chromium.org/866002/diff/36001/39001
File test/cctest/test-diy_fp.cc (right):

http://codereview.chromium.org/866002/diff/36001/39001#newcode14
test/cctest/test-diy_fp.cc:14: static_cast<int64_t>(b))
On 2010/03/11 13:55:26, William Hesse wrote:
> I think a compiler may complain if you call
> CHECK_EQ64(0x9000000000000001, foo), because you are casting a constant
outside
> the int64_t range to an int64_t.
> 
> So this is not a solution either.
I have added a CHECK_EQ64 macro to checks.h.

[Florian Loitsch, 2010-03-11 14:49:00]

Finally switched to NOLINT for uint64_t checks.