Update SmiShiftLeft, SmiShiftLogicalRight, SmiShiftArithmeticRight and SmiDiv to support x32 port.

Update SmiShiftLeft, SmiShiftLogicalRight, SmiShiftArithmeticRight and SmiDiv to support x32 port.

R=verwaest@chromium.org

Committed: https://code.google.com/p/v8/source/detail?r=21725

[haitao.feng, 2014-05-05 03:34:20]

PTAL.

Once we landed this and https://codereview.chromium.org/261233002/, there is no
difference between the x64 port and x32 port in the macro assembler.

[haitao.feng, 2014-05-14 12:23:25]

Friendly ping.

[Toon Verwaest, 2014-06-02 09:49:08]

Sorry for the long wait!
The CL (well, original code as well) confuses me a bit. Why do we only restore
RCX half of the time?

https://codereview.chromium.org/264973011/diff/1/src/x64/macro-assembler-x64.cc
File src/x64/macro-assembler-x64.cc (right):

https://codereview.chromium.org/264973011/diff/1/src/x64/macro-assembler-x64....
src/x64/macro-assembler-x64.cc:2148: SmiToInteger32(dst, dst);
Why not just SmiToInteger32(dst, src) ?

https://codereview.chromium.org/264973011/diff/1/src/x64/macro-assembler-x64....
src/x64/macro-assembler-x64.cc:2174: SmiToInteger32(dst, dst);
Same here

https://codereview.chromium.org/264973011/diff/1/src/x64/macro-assembler-x64....
src/x64/macro-assembler-x64.cc:2192: if (!dst.is(src1)) {
If src1 is rcx, doing SmiToInteger32 first will destroy that register.

https://codereview.chromium.org/264973011/diff/1/src/x64/macro-assembler-x64....
src/x64/macro-assembler-x64.cc:2216: shll_cl(dst);
Seems like this code could be simplified? Why didn't we need to restore rcx
above?

https://codereview.chromium.org/264973011/diff/1/src/x64/macro-assembler-x64....
src/x64/macro-assembler-x64.cc:2252: movp(dst, src1);
SmiToInteger32(dst, src1);

[haitao.feng, 2014-06-03 03:14:30]

No problem. Danno mentioned that you are on vacation in the last meeting. I
changed the code to use movq to save/restore the rcx register.

https://codereview.chromium.org/264973011/diff/1/src/x64/macro-assembler-x64.cc
File src/x64/macro-assembler-x64.cc (right):

https://codereview.chromium.org/264973011/diff/1/src/x64/macro-assembler-x64....
src/x64/macro-assembler-x64.cc:2148: SmiToInteger32(dst, dst);
On 2014/06/02 09:49:08, Toon Verwaest wrote:
> Why not just SmiToInteger32(dst, src) ?

Done.

https://codereview.chromium.org/264973011/diff/1/src/x64/macro-assembler-x64....
src/x64/macro-assembler-x64.cc:2174: SmiToInteger32(dst, dst);
On 2014/06/02 09:49:08, Toon Verwaest wrote:
> Same here

Done.

https://codereview.chromium.org/264973011/diff/1/src/x64/macro-assembler-x64....
src/x64/macro-assembler-x64.cc:2192: if (!dst.is(src1)) {
On 2014/06/02 09:49:08, Toon Verwaest wrote:
> If src1 is rcx, doing SmiToInteger32 first will destroy that register.

Done.

https://codereview.chromium.org/264973011/diff/1/src/x64/macro-assembler-x64....
src/x64/macro-assembler-x64.cc:2216: shll_cl(dst);
For 31-bit SMI, we might need to jump to the on_not_smi_result label. When we
jump to the on_not_smi_result, we need to keep an invariant that src1 and src2
are intact. We clobber dst and rcx in this function, as we assert dst could not
be src1 or src2, we only need to restore rcx before jumping to the
on_not_smi_result.

We do not need this restoring for 32-bit SMI as the result is always a SMI.

https://codereview.chromium.org/264973011/diff/1/src/x64/macro-assembler-x64....
src/x64/macro-assembler-x64.cc:2252: movp(dst, src1);
On 2014/06/02 09:49:08, Toon Verwaest wrote:
> SmiToInteger32(dst, src1);

Done.

[Toon Verwaest, 2014-06-06 12:22:25]

lgtm

[haitao.feng, 2014-06-10 04:04:08]

Committed patchset #3 manually as r21725.