Optimize smi multiply by 2 using shl by 1.

Optimize smi multiply by 2 using shl by 1.
Committed: https://code.google.com/p/dart/source/detail?r=20204

[bakster, 2013-03-19 13:22:43]

DBC

https://codereview.chromium.org/12703011/diff/1/runtime/vm/intermediate_langu...
File runtime/vm/intermediate_language_ia32.cc (right):

https://codereview.chromium.org/12703011/diff/1/runtime/vm/intermediate_langu...
runtime/vm/intermediate_language_ia32.cc:2212: if (value == 2) {
What about 4, 8, 16 etc?

[Florian Schneider, 2013-03-19 13:24:33]

https://codereview.chromium.org/12703011/diff/1/runtime/vm/intermediate_langu...
File runtime/vm/intermediate_language_ia32.cc (right):

https://codereview.chromium.org/12703011/diff/1/runtime/vm/intermediate_langu...
runtime/vm/intermediate_language_ia32.cc:2212: if (value == 2) {
On 2013/03/19 13:22:44, bakster wrote:
> What about 4, 8, 16 etc?

shl does not set the overflow flag for other than 1-bit shifts.

[Vyacheslav Egorov (Google), 2013-03-19 13:27:32]

lgtm

[Florian Schneider, 2013-03-19 13:29:56]

Committed patchset #1 manually as r20204 (presubmit successful).

[bakster, 2013-03-19 13:31:25]

https://codereview.chromium.org/12703011/diff/1/runtime/vm/intermediate_langu...
File runtime/vm/intermediate_language_ia32.cc (right):

https://codereview.chromium.org/12703011/diff/1/runtime/vm/intermediate_langu...
runtime/vm/intermediate_language_ia32.cc:2212: if (value == 2) {
On 2013/03/19 13:24:33, Florian Schneider wrote:
> On 2013/03/19 13:22:44, bakster wrote:
> > What about 4, 8, 16 etc?
> 
> shl does not set the overflow flag for other than 1-bit shifts.
I guess you than can unroll into repeated shl(1) :-)

[Florian Schneider, 2013-03-19 14:06:38]

On 2013/03/19 13:31:25, bakster wrote:
>
https://codereview.chromium.org/12703011/diff/1/runtime/vm/intermediate_langu...
> File runtime/vm/intermediate_language_ia32.cc (right):
> 
>
https://codereview.chromium.org/12703011/diff/1/runtime/vm/intermediate_langu...
> runtime/vm/intermediate_language_ia32.cc:2212: if (value == 2) {
> On 2013/03/19 13:24:33, Florian Schneider wrote:
> > On 2013/03/19 13:22:44, bakster wrote:
> > > What about 4, 8, 16 etc?
> > 
> > shl does not set the overflow flag for other than 1-bit shifts.
> I guess you than can unroll into repeated shl(1) :-)

Sure :) But I'm not sure if

shl
jo
shl
jo

is still faster than imul. It's definitely a lot more in code size.

[ahe, 2013-03-19 14:13:11]

https://codereview.chromium.org/12703011/diff/1/runtime/vm/intermediate_langu...
File runtime/vm/intermediate_language_ia32.cc (right):

https://codereview.chromium.org/12703011/diff/1/runtime/vm/intermediate_langu...
runtime/vm/intermediate_language_ia32.cc:2212: if (value == 2) {
On 2013/03/19 13:31:25, bakster wrote:
> On 2013/03/19 13:24:33, Florian Schneider wrote:
> > On 2013/03/19 13:22:44, bakster wrote:
> > > What about 4, 8, 16 etc?
> > 
> > shl does not set the overflow flag for other than 1-bit shifts.
> I guess you than can unroll into repeated shl(1) :-)

What about a compare?

[srdjan, 2013-03-19 15:31:15]

A while ago I implemented the optimization to use shl instead of mul when
possible, but did not see any performance gains, so I abandoned the change then.
I should have posted the result in the circle, though.