Put live register bits in stackmaps.

Put live register bits in stackmaps.

Add a count of live registers to each stackmap and add bits describing the
registers.  This allows untagged values in general purpose registers at
safepoints.

R=vegorov@google.com
BUG=

Committed: https://code.google.com/p/dart/source/detail?r=11595

[Vyacheslav Egorov (Google), 2012-08-27 14:54:51]

lgtm if you remove hack that is used for spilling XMMRegisters on slow paths and
handle them with stack maps.

https://chromiumcodereview.appspot.com/10882055/diff/1/runtime/vm/flow_graph_...
File runtime/vm/flow_graph_compiler.cc (right):

https://chromiumcodereview.appspot.com/10882055/diff/1/runtime/vm/flow_graph_...
runtime/vm/flow_graph_compiler.cc:246: for (intptr_t i = 0; i <
kNumberOfCpuRegisters; ++i) {
This is tightly couple with the way we push registers. Consider adding a comment
for that.

https://chromiumcodereview.appspot.com/10882055/diff/1/runtime/vm/flow_graph_...
runtime/vm/flow_graph_compiler.cc:251: }
Also please handle XMMRegisters here and remove a hack used for this currently.
This will be a good test.

https://chromiumcodereview.appspot.com/10882055/diff/1/runtime/vm/flow_graph_...
File runtime/vm/flow_graph_compiler_x64.cc (right):

https://chromiumcodereview.appspot.com/10882055/diff/1/runtime/vm/flow_graph_...
runtime/vm/flow_graph_compiler_x64.cc:806: 0);
I wonder if it would not be cleaner to pass RegisterSet instead of RegisterCount
into AddEntry?

https://chromiumcodereview.appspot.com/10882055/diff/1/runtime/vm/stack_frame.cc
File runtime/vm/stack_frame.cc (right):

https://chromiumcodereview.appspot.com/10882055/diff/1/runtime/vm/stack_frame...
runtime/vm/stack_frame.cc:70: // Spill slots are at the 'bottom' of the frame.
This only works for calls with no arguments. Consider adding todo and a comment
about that.

[Kevin Millikin (Google), 2012-08-29 12:40:41]

Now handling XMM registers.  Please take a quick look.

https://chromiumcodereview.appspot.com/10882055/diff/1/runtime/vm/flow_graph_...
File runtime/vm/flow_graph_compiler.cc (right):

https://chromiumcodereview.appspot.com/10882055/diff/1/runtime/vm/flow_graph_...
runtime/vm/flow_graph_compiler.cc:246: for (intptr_t i = 0; i <
kNumberOfCpuRegisters; ++i) {
On 2012/08/27 14:54:51, Vyacheslav Egorov (Google) wrote:
> This is tightly couple with the way we push registers. Consider adding a
comment
> for that.

Done.

https://chromiumcodereview.appspot.com/10882055/diff/1/runtime/vm/flow_graph_...
File runtime/vm/flow_graph_compiler_x64.cc (right):

https://chromiumcodereview.appspot.com/10882055/diff/1/runtime/vm/flow_graph_...
runtime/vm/flow_graph_compiler_x64.cc:806: 0);
On 2012/08/27 14:54:51, Vyacheslav Egorov (Google) wrote:
> I wonder if it would not be cleaner to pass RegisterSet instead of
RegisterCount
> into AddEntry?

At some point we have to unpack it into something more low level --- right now
that happens in the code generator and I don't really think it's obviously
better to do it in the stackmap table builder.  It's missing access to the code
generator's state.

https://chromiumcodereview.appspot.com/10882055/diff/1/runtime/vm/stack_frame.cc
File runtime/vm/stack_frame.cc (right):

https://chromiumcodereview.appspot.com/10882055/diff/1/runtime/vm/stack_frame...
runtime/vm/stack_frame.cc:70: // Spill slots are at the 'bottom' of the frame.
On 2012/08/27 14:54:51, Vyacheslav Egorov (Google) wrote:
> This only works for calls with no arguments. Consider adding todo and a
comment
> about that.

Done.

[Vyacheslav Egorov (Google), 2012-08-30 09:24:35]

lgtm

https://chromiumcodereview.appspot.com/10882055/diff/6001/runtime/vm/flow_gra...
File runtime/vm/flow_graph_allocator.h (right):

https://chromiumcodereview.appspot.com/10882055/diff/6001/runtime/vm/flow_gra...
runtime/vm/flow_graph_allocator.h:22: static const intptr_t
kDoubleSpillSlotFactor = kDoubleSize / kWordSize;
you copied the constant but not the comment.

https://chromiumcodereview.appspot.com/10882055/diff/6001/runtime/vm/flow_gra...
File runtime/vm/flow_graph_compiler.cc (right):

https://chromiumcodereview.appspot.com/10882055/diff/6001/runtime/vm/flow_gra...
runtime/vm/flow_graph_compiler.cc:262: // Mark the bits in the stack map in the
same order we push registers
this does not actually match with that order: when we store xmm registers xmm0
comes at the top (if live), but here last one comes at the top.

https://chromiumcodereview.appspot.com/10882055/diff/6001/runtime/vm/flow_gra...
runtime/vm/flow_graph_compiler.cc:267: // assumption that  there are normally
few live  XMM registers, this
s/live  XMM/live XMM/

[Kevin Millikin (Google), 2012-08-30 10:02:21]

https://chromiumcodereview.appspot.com/10882055/diff/6001/runtime/vm/flow_gra...
File runtime/vm/flow_graph_allocator.h (right):

https://chromiumcodereview.appspot.com/10882055/diff/6001/runtime/vm/flow_gra...
runtime/vm/flow_graph_allocator.h:22: static const intptr_t
kDoubleSpillSlotFactor = kDoubleSize / kWordSize;
On 2012/08/30 09:24:35, Vyacheslav Egorov (Google) wrote:
> you copied the constant but not the comment.

I restored the comment.

https://chromiumcodereview.appspot.com/10882055/diff/6001/runtime/vm/flow_gra...
File runtime/vm/flow_graph_compiler.cc (right):

https://chromiumcodereview.appspot.com/10882055/diff/6001/runtime/vm/flow_gra...
runtime/vm/flow_graph_compiler.cc:262: // Mark the bits in the stack map in the
same order we push registers
On 2012/08/30 09:24:35, Vyacheslav Egorov (Google) wrote:
> this does not actually match with that order: when we store xmm registers xmm0
> comes at the top (if live), but here last one comes at the top.

You're right and it's a bit fishy.  I've made it match.

https://chromiumcodereview.appspot.com/10882055/diff/6001/runtime/vm/flow_gra...
runtime/vm/flow_graph_compiler.cc:267: // assumption that  there are normally
few live  XMM registers, this
On 2012/08/30 09:24:35, Vyacheslav Egorov (Google) wrote:
> s/live  XMM/live XMM/

You don't like the fully justified comments?