PNaCl: Add NoAlias attributes in ExpandByVal and ExpandVarArgs passes

PNaCl: Add NoAlias attributes in ExpandByVal and ExpandVarArgs passes

This could help prevent these expansion passes from inhibiting
optimisations than run after the expansion.  e.g. It gives the
optimiser more freedom to move around reads from the varargs buffer
because they will not alias writes to other locations.

BUG=https://code.google.com/p/nativeclient/issues/detail?id=3400
TEST=PNaCl toolchain trybots + GCC torture tests + LLVM test suite + Spec2k

Committed: https://gerrit.chromium.org/gerrit/gitweb?p=native_client/pnacl-llvm.git;a=commit;h=0e6d484

[Mark Seaborn, 2013-04-26 15:31:13]

As suggested in https://codereview.chromium.org/14317008/.

[jvoung (off chromium), 2013-04-26 16:43:09]

https://codereview.chromium.org/14060026/diff/1/test/Transforms/NaCl/expand-v...
File test/Transforms/NaCl/expand-varargs-attrs.ll (right):

https://codereview.chromium.org/14060026/diff/1/test/Transforms/NaCl/expand-v...
test/Transforms/NaCl/expand-varargs-attrs.ll:13: ; CHECK: define void
@func_with_arg_attrs(%MyStruct* byval, i8* %varargs) {
Should this new definition have noalias for the %varargs param?

Same with the new version of @varargs_func?

https://codereview.chromium.org/14060026/diff/1/test/Transforms/NaCl/expand-v...
File test/Transforms/NaCl/expand-varargs.ll (right):

https://codereview.chromium.org/14060026/diff/1/test/Transforms/NaCl/expand-v...
test/Transforms/NaCl/expand-varargs.ll:29: ; CHECK: define i32 @varargs_func(i32
%arg, i8* %varargs) {
Same question about noalias for the new %varargs param

[Mark Seaborn, 2013-04-26 16:52:58]

https://codereview.chromium.org/14060026/diff/1/test/Transforms/NaCl/expand-v...
File test/Transforms/NaCl/expand-varargs-attrs.ll (right):

https://codereview.chromium.org/14060026/diff/1/test/Transforms/NaCl/expand-v...
test/Transforms/NaCl/expand-varargs-attrs.ll:13: ; CHECK: define void
@func_with_arg_attrs(%MyStruct* byval, i8* %varargs) {
On 2013/04/26 16:43:09, jvoung (cr) wrote:
> Should this new definition have noalias for the %varargs param?
> 
> Same with the new version of @varargs_func?

Oops, yes, I wasn't thinking.  "noalias" should go on the function definition
instead.  There's no need for it to go on the call.  Fixed.

https://codereview.chromium.org/14060026/diff/1/test/Transforms/NaCl/expand-v...
File test/Transforms/NaCl/expand-varargs.ll (right):

https://codereview.chromium.org/14060026/diff/1/test/Transforms/NaCl/expand-v...
test/Transforms/NaCl/expand-varargs.ll:29: ; CHECK: define i32 @varargs_func(i32
%arg, i8* %varargs) {
On 2013/04/26 16:43:09, jvoung (cr) wrote:
> Same question about noalias for the new %varargs param

Done.

[jvoung (off chromium), 2013-04-26 19:35:09]

lgtm

[jvoung (off chromium), 2013-04-26 19:36:25]

https://codereview.chromium.org/14060026/diff/5001/test/Transforms/NaCl/expan...
File test/Transforms/NaCl/expand-byval.ll (right):

https://codereview.chromium.org/14060026/diff/5001/test/Transforms/NaCl/expan...
test/Transforms/NaCl/expand-byval.ll:33: ; CHECK-NEXT: call void
@ext_byval_func(%MyStruct* noalias %ptr.byval_copy)
Oh, also doesn't *need* to be in the call here?

[Mark Seaborn, 2013-04-26 19:59:09]

https://codereview.chromium.org/14060026/diff/5001/test/Transforms/NaCl/expan...
File test/Transforms/NaCl/expand-byval.ll (right):

https://codereview.chromium.org/14060026/diff/5001/test/Transforms/NaCl/expan...
test/Transforms/NaCl/expand-byval.ll:33: ; CHECK-NEXT: call void
@ext_byval_func(%MyStruct* noalias %ptr.byval_copy)
On 2013/04/26 19:36:25, jvoung (cr) wrote:
> Oh, also doesn't *need* to be in the call here?

Yes, but as we were discussing it doesn't seem to matter either way for the call
sites.  So I've just done what was simplest in the code and not gone out of my
way to either add or remove "noalias" in calls.

[Mark Seaborn, 2013-05-09 15:17:13]

Committed patchset #2 manually as r0e6d484 (presubmit successful).