Specialize mint shift code on ia32 for a constant shift amount.

Specialize mint shift code on ia32 for a constant shift amount.

R=srdjan@google.com

Committed: https://code.google.com/p/dart/source/detail?r=38141

[srdjan, 2014-07-10 23:20:15]

LGTM after adding assembler tests and Dart tests for shifting Mints with
constants (unless they already exist).

https://codereview.chromium.org/387733002/diff/1/runtime/vm/intermediate_lang...
File runtime/vm/intermediate_language_ia32.cc (right):

https://codereview.chromium.org/387733002/diff/1/runtime/vm/intermediate_lang...
runtime/vm/intermediate_language_ia32.cc:5937: if ((shift < 0) || (shift >
kMintShiftCountLimit)) {
For SHR  shouldn't 'shift > kMintShiftCountLimit' produce same shift as with
kMintShiftCountLimit? I.e., 
if (shift > kMintShiftCountLimit) shift = kMintShiftCountLimit;

[regis, 2014-07-10 23:59:27]

Thanks!

I added assembler tests, but not Dart tests, because 78 of them already use the
optimized code shifting mints by a constant amount.

https://codereview.chromium.org/387733002/diff/1/runtime/vm/intermediate_lang...
File runtime/vm/intermediate_language_ia32.cc (right):

https://codereview.chromium.org/387733002/diff/1/runtime/vm/intermediate_lang...
runtime/vm/intermediate_language_ia32.cc:5937: if ((shift < 0) || (shift >
kMintShiftCountLimit)) {
On 2014/07/10 23:20:15, srdjan wrote:
> For SHR  shouldn't 'shift > kMintShiftCountLimit' produce same shift as with
> kMintShiftCountLimit? I.e., 
> if (shift > kMintShiftCountLimit) shift = kMintShiftCountLimit;

I kept the previous semantics. In the case the shift amount is variable and not
known to be in range [0..64[, the function has_shift_count_check() will return
true. This causes range check code to be emitted with a potential deopt jump.
In the case the shift amount is constant and not in range, I can emit the deopt
jump directly. Of course, it would be even better not to optimize, but I did not
want to make changes upstream.

[regis, 2014-07-11 00:00:06]

Committed patchset #2 manually as r38141 (presubmit successful).

[Ivan Posva, 2014-07-16 08:19:05]

What about the other 32-bit architectures?

-Ivan

[regis, 2014-07-16 16:41:57]

On 2014/07/16 08:19:05, Ivan Posva wrote:
> What about the other 32-bit architectures?
> 
> -Ivan

ARM would not benefit as much as IA32 from this specialization, but I will still
do it.
MIPS is behind in regards to mints support. The lack of register pairs is one
issue, but it is on John's todo list.

[regis, 2014-07-17 21:37:25]

On 2014/07/16 16:41:57, regis wrote:
> On 2014/07/16 08:19:05, Ivan Posva wrote:
> > What about the other 32-bit architectures?
> > 
> > -Ivan
> 
> ARM would not benefit as much as IA32 from this specialization, but I will
still
> do it.

See https://codereview.chromium.org/401683003

> MIPS is behind in regards to mints support. The lack of register pairs is one
> issue, but it is on John's todo list.