[turbofan] Add control and effect inputs to RawMachineAssembler calls.

[turbofan] Add control and effect inputs to RawMachineAssembler calls.

Calls should have control and effect inputs, which were previously missing
for RawMachineAssembler call operations. Add control and effect edges to
graph()->start().

BUG=v8:4280
LOG=N

Committed: https://crrev.com/843d5d5ca408dafb7210db6a97637fe2659ac5f9
Cr-Commit-Position: refs/heads/master@{#30307}

[rmcilroy, 2015-08-18 21:30:30]

The CQ bit was checked by rmcilroy@chromium.org to run a CQ dry run

[commit-bot: I haz the power, 2015-08-18 21:30:42]

Dry run: CQ is trying da patch. Follow status at
 https://chromium-cq-status.appspot.com/patch-status/1283193007/1
View timeline at
 https://chromium-cq-status.appspot.com/patch-timeline/1283193007/1

[rmcilroy, 2015-08-18 21:34:18]

The CQ bit was checked by rmcilroy@chromium.org to run a CQ dry run

[commit-bot: I haz the power, 2015-08-18 21:34:29]

Dry run: CQ is trying da patch. Follow status at
 https://chromium-cq-status.appspot.com/patch-status/1283193007/20001
View timeline at
 https://chromium-cq-status.appspot.com/patch-timeline/1283193007/20001

[commit-bot: I haz the power, 2015-08-18 21:56:03]

The CQ bit was unchecked by commit-bot@chromium.org

[commit-bot: I haz the power, 2015-08-18 21:56:04]

Dry run: This issue passed the CQ dry run.

[rmcilroy, 2015-08-19 06:55:31]

rmcilroy@chromium.org changed reviewers:
+ titzer@chromium.org

[rmcilroy, 2015-08-19 06:55:32]

Ben, please take a look, thanks.

[titzer, 2015-08-19 15:16:17]

On 2015/08/19 06:55:32, rmcilroy wrote:
> Ben, please take a look, thanks.

What's the motivation for this? The graphs built by the raw assembler already
have no hope of being verified because they don't have proper effect chains nor
control chains.

[rmcilroy, 2015-08-19 15:54:17]

On 2015/08/19 15:16:17, titzer wrote:
> On 2015/08/19 06:55:32, rmcilroy wrote:
> > Ben, please take a look, thanks.
> 
> What's the motivation for this? The graphs built by the raw assembler already
> have no hope of being verified because they don't have proper effect chains
nor
> control chains.

There are a couple of motivations
 - The bytecode handlers can get through verification with the changes in
https://codereview.chromium.org/1297203002/, but they wouldn't if I call any of
these functions without this change (note, the callInterpreterDispatch already
has effect/control edges due to the point below)
 - The IsCall node matchers used by interpreter-assembler-unittests expect
control/effect edges and fail if they aren't passed.
 - The Load/Store operations already hook up their control/effect edges to
graph->start().
 - I've ended up with bugs while implementing where I passed x value inputs but
the call descriptor only expected x-1 and it was difficult to work out why the
wrong code was being generated until I figured out one of the value inputs was
being treated as the control edge.

[titzer, 2015-08-21 10:31:52]

*sigh* 

I'll be ok with this patch, but keep in mind that the graphs built by raw
machine assembler are really broken and don't make sense at all without a
schedule.

https://codereview.chromium.org/1283193007/diff/20001/src/compiler/graph.h
File src/compiler/graph.h (right):

https://codereview.chromium.org/1283193007/diff/20001/src/compiler/graph.h#ne...
src/compiler/graph.h:82: Node* NewNode(const Operator* op, Node* n1, Node* n2,
Node* n3, Node* n4,
We've got too many overloads here already, can we use the generic method in the
callers instead?

[rmcilroy, 2015-08-21 13:10:15]

On 2015/08/21 10:31:52, titzer wrote:
> *sigh* 
> 
> I'll be ok with this patch, but keep in mind that the graphs built by raw
> machine assembler are really broken and don't make sense at all without a
> schedule.

Noted, thanks. PTAL.

https://codereview.chromium.org/1283193007/diff/20001/src/compiler/graph.h
File src/compiler/graph.h (right):

https://codereview.chromium.org/1283193007/diff/20001/src/compiler/graph.h#ne...
src/compiler/graph.h:82: Node* NewNode(const Operator* op, Node* n1, Node* n2,
Node* n3, Node* n4,
On 2015/08/21 10:31:52, titzer wrote:
> We've got too many overloads here already, can we use the generic method in
the
> callers instead?

Done.

[titzer, 2015-08-21 13:16:39]

lgtm

[rmcilroy, 2015-08-21 13:26:26]

The CQ bit was checked by rmcilroy@chromium.org

[commit-bot: I haz the power, 2015-08-21 13:26:37]

CQ is trying da patch. Follow status at
 https://chromium-cq-status.appspot.com/patch-status/1283193007/40001
View timeline at
 https://chromium-cq-status.appspot.com/patch-timeline/1283193007/40001

[commit-bot: I haz the power, 2015-08-21 13:52:36]

Committed patchset #3 (id:40001)

[commit-bot: I haz the power, 2015-08-21 13:52:56]

Patchset 3 (id:??) landed as
https://crrev.com/843d5d5ca408dafb7210db6a97637fe2659ac5f9
Cr-Commit-Position: refs/heads/master@{#30307}