Fixed and improved code for integral division. Fixed and extended tests.

Fixed and improved code for integral division. Fixed and extended tests.

Arithmetic right shifting is *not* division in two's complement
representation, only in one's complement. So we convert to one's
complement, shift, and go back to two's complement. By permutating the
last steps, one can get efficient branch-free code. This insight comes
from the paleozoic era of computer science, see the paper from 1976:

   Guy Lewis Steele Jr.: "Arithmetic Shifting Considered Harmful"
   ftp://publications.ai.mit.edu/ai-publications/pdf/AIM-378.pdf

This results in better and more correct code than our previous
"neg/shift/neg" dance.

LOG=y
BUG=v8:3151
R=bmeurer@chromium.org

Committed: https://code.google.com/p/v8/source/detail?r=19434

[Benedikt Meurer, 2014-02-18 09:43:30]

LGTM with nit.

https://codereview.chromium.org/166793002/diff/100001/src/a64/lithium-codegen...
File src/a64/lithium-codegen-a64.cc (right):

https://codereview.chromium.org/166793002/diff/100001/src/a64/lithium-codegen...
src/a64/lithium-codegen-a64.cc:2622: __ Add(result, result, dividend);
Please merge the last Mov and the Add.

https://codereview.chromium.org/166793002/diff/100001/src/arm/lithium-codegen...
File src/arm/lithium-codegen-arm.cc (right):

https://codereview.chromium.org/166793002/diff/100001/src/arm/lithium-codegen...
src/arm/lithium-codegen-arm.cc:1382: __ add(result, result, dividend);
Same as for a64.

[Sven Panne, 2014-02-18 10:45:38]

Committed patchset #5 manually as r19434 (presubmit successful).