Fix dictionary element load to pass correct elements kind.

Fix dictionary element load to pass correct elements kind.

Using FAST_SMI_ELEMENTS triggers optimization on 64-bit architectures that load
only the higher 32 bits of the element. If the element is a pointer to undefined
that has 0 in the higher half than it is erroneously treated as SMI 0.

BUG=v8:3158
LOG=N
TEST=mjsunit/sparse-array-reverse,mjsunit/regress/regress-3158.js
R=danno@chromium.org, ishell@chromium.org

Committed: https://code.google.com/p/v8/source/detail?r=19387

[ulan, 2014-02-14 14:14:20]

PTAL

[danno, 2014-02-14 14:17:00]

lgtm

[Igor Sheludko, 2014-02-14 14:25:18]

I think it is worth adding a debug-mode check that whenever we use optimized smi
keyed load we actually loading a smi. At least for x64 where we have a scratch
register for free.

[ulan, 2014-02-14 15:03:31]

Igor, I added check in x64, ptal.

[Igor Sheludko, 2014-02-14 15:16:31]

lgtm with a nit:

https://codereview.chromium.org/166653005/diff/110001/src/x64/lithium-codegen...
File src/x64/lithium-codegen-x64.cc (right):

https://codereview.chromium.org/166653005/diff/110001/src/x64/lithium-codegen...
src/x64/lithium-codegen-x64.cc:2784: if (FLAG_debug_code) {
I think it is better to use #ifdef DEBUG as we did for check in StoreNamedField.

https://codereview.chromium.org/166653005/diff/110001/src/x64/lithium-codegen...
src/x64/lithium-codegen-x64.cc:3034: if (FLAG_debug_code) {
Same as above.

[ulan, 2014-02-14 15:22:38]

Thanks, landing.

https://codereview.chromium.org/166653005/diff/110001/src/x64/lithium-codegen...
File src/x64/lithium-codegen-x64.cc (right):

https://codereview.chromium.org/166653005/diff/110001/src/x64/lithium-codegen...
src/x64/lithium-codegen-x64.cc:2784: if (FLAG_debug_code) {
On 2014/02/14 15:16:31, Igor Sheludko wrote:
> I think it is better to use #ifdef DEBUG as we did for check in
StoreNamedField.

Done.

https://codereview.chromium.org/166653005/diff/110001/src/x64/lithium-codegen...
src/x64/lithium-codegen-x64.cc:3034: if (FLAG_debug_code) {
On 2014/02/14 15:16:31, Igor Sheludko wrote:
> Same as above.

Done.

[ulan, 2014-02-14 15:52:32]

Committed patchset #5 manually as r19387 (presubmit successful).