Add vmov between integers and floats in ARM assembler.

Add vmov between integers and floats in ARM assembler.

Adds vmovrs that implements moving from an integer (GP) register and a
float (S) register to the integrated ARM assembler.

The test also shows that moving from a float (S) register to an
integer (GP) register also works.

BUG= https://bugs.chromium.org/p/nativeclient/issues/detail?id=4334
R=eholk@chromium.org, stichnot@chromium.org

Committed: https://gerrit.chromium.org/gerrit/gitweb?p=native_client/pnacl-subzero.git;a=commit;h=e1b6574f23b67c92fe367128b6597d54c55e88fb

[Karl, 2016-01-27 22:37:50]

Description was changed from

==========
Add vmov between integers and floats in ARM assembler.

Adds vmovrs that implements moving from an integer (GP) register and a
float (S) register to the integrated ARM assembler.

The test also shows that moving from a float (S) register to an
integer (GP) register also works.

BUG= https://bugs.chromium.org/p/nativeclient/issues/detail?id=4334
==========

to

==========
Add vmov between integers and floats in ARM assembler.

Adds vmovrs that implements moving from an integer (GP) register and a
float (S) register to the integrated ARM assembler.

The test also shows that moving from a float (S) register to an
integer (GP) register also works.

BUG= https://bugs.chromium.org/p/nativeclient/issues/detail?id=4334
==========

[Karl, 2016-01-27 22:37:51]

kschimpf@google.com changed reviewers:
+ eholk@chromium.org, jpp@chromium.org, sehr@chromium.org, stichnot@chromium.org

[Jim Stichnoth, 2016-01-27 22:59:22]

lgtm

https://codereview.chromium.org/1647683002/diff/40001/src/DartARM32/assembler...
File src/DartARM32/assembler_arm.h (right):

https://codereview.chromium.org/1647683002/diff/40001/src/DartARM32/assembler...
src/DartARM32/assembler_arm.h:615: static uword GetBreakInstructionFiller() {
Will we ever use this?  If not, maybe document that and combine the #if 0
regions.

https://codereview.chromium.org/1647683002/diff/40001/src/IceInstARM32.cpp
File src/IceInstARM32.cpp (right):

https://codereview.chromium.org/1647683002/diff/40001/src/IceInstARM32.cpp#ne...
src/IceInstARM32.cpp:1213: case IceType_i32:
If you're planning on eventually listing all of the IceTypes, you might want to
just "handle" i64 now (presumably as an error).

https://codereview.chromium.org/1647683002/diff/40001/tests_lit/assembler/arm...
File tests_lit/assembler/arm32/vmov-f2i.ll (right):

https://codereview.chromium.org/1647683002/diff/40001/tests_lit/assembler/arm...
tests_lit/assembler/arm32/vmov-f2i.ll:24: ; DIS-LABEL: {{.*}} <FloatToSignedI1>:
Putting {{.*}} at the beginning or end of the match string is redundant.  Either
omit it entirely, or use {{.+}}.

https://codereview.chromium.org/1647683002/diff/40001/tests_lit/assembler/arm...
tests_lit/assembler/arm32/vmov-f2i.ll:47: 
two newlines

https://codereview.chromium.org/1647683002/diff/40001/tests_lit/assembler/arm...
tests_lit/assembler/arm32/vmov-f2i.ll:79: %v = sitofp i1 1 to float
I would use uitofp for these tests, particularly because a "signed" i1 is either
0 or -1, and it would be just weird to put the "correct" value of -1 here.

[Eric Holk, 2016-01-27 23:29:53]

lgtm

[Karl, 2016-01-27 23:36:21]

Description was changed from

==========
Add vmov between integers and floats in ARM assembler.

Adds vmovrs that implements moving from an integer (GP) register and a
float (S) register to the integrated ARM assembler.

The test also shows that moving from a float (S) register to an
integer (GP) register also works.

BUG= https://bugs.chromium.org/p/nativeclient/issues/detail?id=4334
==========

to

==========
Add vmov between integers and floats in ARM assembler.

Adds vmovrs that implements moving from an integer (GP) register and a
float (S) register to the integrated ARM assembler.

The test also shows that moving from a float (S) register to an
integer (GP) register also works.

BUG= https://bugs.chromium.org/p/nativeclient/issues/detail?id=4334
R=eholk@chromium.org, stichnot@chromium.org

Committed:
https://gerrit.chromium.org/gerrit/gitweb?p=native_client/pnacl-subzero.git;a...
==========

[Karl, 2016-01-27 23:36:22]

Committed patchset #4 (id:60001) manually as
e1b6574f23b67c92fe367128b6597d54c55e88fb (presubmit successful).

[Karl, 2016-01-27 23:36:42]

https://codereview.chromium.org/1647683002/diff/40001/src/DartARM32/assembler...
File src/DartARM32/assembler_arm.h (right):

https://codereview.chromium.org/1647683002/diff/40001/src/DartARM32/assembler...
src/DartARM32/assembler_arm.h:615: static uword GetBreakInstructionFiller() {
On 2016/01/27 22:59:22, stichnot wrote:
> Will we ever use this?  If not, maybe document that and combine the #if 0
> regions.

Done.

https://codereview.chromium.org/1647683002/diff/40001/src/IceInstARM32.cpp
File src/IceInstARM32.cpp (right):

https://codereview.chromium.org/1647683002/diff/40001/src/IceInstARM32.cpp#ne...
src/IceInstARM32.cpp:1213: case IceType_i32:
On 2016/01/27 22:59:22, stichnot wrote:
> If you're planning on eventually listing all of the IceTypes, you might want
to
> just "handle" i64 now (presumably as an error).

Because of "preprocessing" before this method is called, not all cases will be
covered. I64 for example is not. However, I will add that case in, and a TODO to
consider flattening methods into a single switch.

https://codereview.chromium.org/1647683002/diff/40001/tests_lit/assembler/arm...
File tests_lit/assembler/arm32/vmov-f2i.ll (right):

https://codereview.chromium.org/1647683002/diff/40001/tests_lit/assembler/arm...
tests_lit/assembler/arm32/vmov-f2i.ll:24: ; DIS-LABEL: {{.*}} <FloatToSignedI1>:
On 2016/01/27 22:59:22, stichnot wrote:
> Putting {{.*}} at the beginning or end of the match string is redundant. 
Either
> omit it entirely, or use {{.+}}.

Done.

https://codereview.chromium.org/1647683002/diff/40001/tests_lit/assembler/arm...
tests_lit/assembler/arm32/vmov-f2i.ll:47: 
On 2016/01/27 22:59:22, stichnot wrote:
> two newlines

Done.

https://codereview.chromium.org/1647683002/diff/40001/tests_lit/assembler/arm...
tests_lit/assembler/arm32/vmov-f2i.ll:79: %v = sitofp i1 1 to float
On 2016/01/27 22:59:22, stichnot wrote:
> I would use uitofp for these tests, particularly because a "signed" i1 is
either
> 0 or -1, and it would be just weird to put the "correct" value of -1 here.

Done.