VM: Fix bugs with missing deoptimization environment and int32x4 constructor

VM: Fix bugs with missing deoptimization environment and int32x4 constructor

It can occur that we don't have exact type information at UnboxUint32 instructions,
but range information concludes that the instruction can't deoptimize. In this case,
we can assume the input must be a smi or mint if the range fits into mint range.

The generated coe for this case was missing on x64 and arm64. It was already added
for ia32, ARM and MIPS (see https://codereview.chromium.org//770303002)

The optimized instruction of the Int32x4 constructor was incorrectly using uint32 as
input representation, instead of int32.

R=iposva@google.com, vegorov@google.com

Committed: https://github.com/dart-lang/sdk/commit/96e4c4d3c6dac4a217356f785c95afbdf7aa51b1

[Florian Schneider, 2015-05-27 15:07:22]

fschneider@google.com changed reviewers:
+ vegorov@google.com

[Ivan Posva, 2015-05-27 15:57:52]

iposva@google.com changed reviewers:
+ iposva@google.com

[Ivan Posva, 2015-05-27 15:57:53]

LGTM -ip

https://codereview.chromium.org/1148943004/diff/40001/runtime/lib/typed_data....
File runtime/lib/typed_data.dart (right):

https://codereview.chromium.org/1148943004/diff/40001/runtime/lib/typed_data....
runtime/lib/typed_data.dart:1269: _setUint32(index *
Uint32List.BYTES_PER_ELEMENT, value);
Thanks!

[Vyacheslav Egorov (Google), 2015-05-27 16:17:35]

lgtm

https://codereview.chromium.org/1148943004/diff/40001/runtime/vm/intermediate...
File runtime/vm/intermediate_language.cc (right):

https://codereview.chromium.org/1148943004/diff/40001/runtime/vm/intermediate...
runtime/vm/intermediate_language.cc:1268: if (value_range != NULL &&
no need for != NULL check. Fits does it for you (it calls IsUnknown which checks
for NULL)

you can just

return !RangeUtils::Fits(..., RangeBoundary::kRangeBoundaryInt64)

On the other hand RangeUtils::Fits(range, RangeBoundary::kRangeBoundaryInt64) is
equivalent to !Range::IsUnknown(range) because we never assign infinity ranges
to instructions, we only use them during range calculations.

[Florian Schneider, 2015-05-28 08:09:56]

Committed patchset #4 (id:60001) manually as
96e4c4d3c6dac4a217356f785c95afbdf7aa51b1 (presubmit successful).

[Florian Schneider, 2015-05-28 08:11:09]

https://codereview.chromium.org/1148943004/diff/40001/runtime/vm/intermediate...
File runtime/vm/intermediate_language.cc (right):

https://codereview.chromium.org/1148943004/diff/40001/runtime/vm/intermediate...
runtime/vm/intermediate_language.cc:1268: if (value_range != NULL &&
On 2015/05/27 16:17:35, Vyacheslav Egorov (Google) wrote:
> no need for != NULL check. Fits does it for you (it calls IsUnknown which
checks
> for NULL)
> 
> you can just
> 
> return !RangeUtils::Fits(..., RangeBoundary::kRangeBoundaryInt64)
> 
> On the other hand RangeUtils::Fits(range, RangeBoundary::kRangeBoundaryInt64)
is
> equivalent to !Range::IsUnknown(range) because we never assign infinity ranges
> to instructions, we only use them during range calculations. 

Done.