Emit atomic memory order other than seq_cst on demand only

Emit atomic memory order other than seq_cst on demand only

The following CL:
  https://codereview.chromium.org/791053006/
Taught PNaCl to support atomic memory orderings other than seq_cst (which was
the only supported memory ordering at launch). This is desirable for
performance, but it causes issues for users when rolling out the new SDK: old
versions of the translator don't accept anything but seq_cst and the canary SDK
emits them, which makes testing harder. This CL makes the SDK default to
emitting only seq_cst memory ordering, but leaves the translator as is (i.e. it
still accepts the new memory ordering). This makes it easier to roll out this
new feature, and users can still opt-in to emitting the new memory order by
specifying -pnacl-memory-order-seq-cst-only=false. The flag currently defaults
to seq_cst only, and for Chrome 43 it will be made to default to false
(i.e. acq/rel/acq_rel/seq_cst can be emitted).

R=jvoung@chromium.org
BUG= https://code.google.com/p/nativeclient/issues/detail?id=4029
TEST= make check

Committed: https://gerrit.chromium.org/gerrit/gitweb?p=native_client/pnacl-llvm.git;a=commit;h=0cebd5960dcd213d379ab504b6ae21b6e0837c93

[jvoung (off chromium), 2015-01-20 22:28:32]

LGTM -- from what I remember, there might be plumbing needed if a user wants to
set the opt flag from the pnacl-ld/pnacl-clang++ drivers, so something to watch
out for.

https://codereview.chromium.org/857403002/diff/1/test/Transforms/NaCl/atomic/...
File test/Transforms/NaCl/atomic/atomic_others.ll (right):

https://codereview.chromium.org/857403002/diff/1/test/Transforms/NaCl/atomic/...
test/Transforms/NaCl/atomic/atomic_others.ll:43: ; CHECK-NEXT: %res = call i32
@llvm.nacl.atomic.load.i32(i32* %ptr, i32 3)
Could you add a test run w/ the flag being default also, and check that 3/4/5 go
to 6?

[JF, 2015-01-21 03:50:03]

On 2015/01/20 22:28:32, jvoung wrote:
> LGTM -- from what I remember, there might be plumbing needed if a user wants
to
> set the opt flag from the pnacl-ld/pnacl-clang++ drivers, so something to
watch
> out for.

Oh right, thanks for reminding me. I'll do that with the COMPONENT_REVISION rev.

https://codereview.chromium.org/857403002/diff/1/test/Transforms/NaCl/atomic/...
File test/Transforms/NaCl/atomic/atomic_others.ll (right):

https://codereview.chromium.org/857403002/diff/1/test/Transforms/NaCl/atomic/...
test/Transforms/NaCl/atomic/atomic_others.ll:43: ; CHECK-NEXT: %res = call i32
@llvm.nacl.atomic.load.i32(i32* %ptr, i32 3)
On 2015/01/20 22:28:31, jvoung wrote:
> Could you add a test run w/ the flag being default also, and check that 3/4/5
go
> to 6?

Done.

[JF, 2015-01-21 03:50:55]

Committed patchset #2 (id:20001) manually as
0cebd5960dcd213d379ab504b6ae21b6e0837c93 (presubmit successful).