Subzero: Switch to AT&T asm syntax. I give up.

Subzero: Switch to AT&T asm syntax.  I give up.

The main motivation is that -build-on-read introduces Intel-style asm output like:
  mov al, byte ptr [flags]
and llvm-mc misinterprets the global symbol "flags" as the flags register.  Further workarounds will likely cost more effort than switching over to AT&T syntax.

Most of the lit tests don't need changing, since the asm text is generated by assembling and disassembling the llvm2ice asm output.

There some LEAHACK TODOs that can be fixed, but that would change some of the instructions, so that can be a separate CL.

The Operand emit() routines really ought to be moved entirely into the target-specific source files.

BUG= none
R=jvoung@chromium.org

Committed: https://gerrit.chromium.org/gerrit/gitweb?p=native_client/pnacl-subzero.git;a=commit;h=bca2f65551c65fdc18278bf88025c11d8af18579

[Jim Stichnoth, 2014-10-31 20:53:59]

stichnot@chromium.org changed reviewers:
+ jvoung@chromium.org

[Jim Stichnoth, 2014-10-31 20:55:20]

https://codereview.chromium.org/695993004/diff/1/pydir/crosstest.py
File pydir/crosstest.py (right):

https://codereview.chromium.org/695993004/diff/1/pydir/crosstest.py#newcode108
pydir/crosstest.py:108: '-build-on-read=0',
One can add:
  '-ias=0',
to test the AT&T emit syntax with the cross tests.

https://codereview.chromium.org/695993004/diff/1/src/IceInstX8632.cpp
File src/IceInstX8632.cpp (right):

https://codereview.chromium.org/695993004/diff/1/src/IceInstX8632.cpp#newcode446
src/IceInstX8632.cpp:446: void InstX8632Br::emit(const Cfg *Func) const {
No change in branch syntax for AT&T.

https://codereview.chromium.org/695993004/diff/1/src/IceInstX8632.cpp#newcode...
src/IceInstX8632.cpp:1679: void InstX8632Cmpxchg8b::emit(const Cfg *Func) const
{
No syntax changes since only one operand is emitted.

https://codereview.chromium.org/695993004/diff/1/src/IceInstX8632.h
File src/IceInstX8632.h (right):

https://codereview.chromium.org/695993004/diff/1/src/IceInstX8632.h#newcode467
src/IceInstX8632.h:467: void emit(const Cfg *Func) const override {
No changes needed for AT&T syntax.

[jvoung (off chromium), 2014-10-31 22:12:18]

https://codereview.chromium.org/695993004/diff/20001/src/IceInstX8632.cpp
File src/IceInstX8632.cpp (right):

https://codereview.chromium.org/695993004/diff/20001/src/IceInstX8632.cpp#new...
src/IceInstX8632.cpp:2310: << "\n";
\n could fit on the previous line next to %esp

https://codereview.chromium.org/695993004/diff/20001/src/IceInstX8632.cpp#new...
src/IceInstX8632.cpp:2385: << "(%esp)\n";
nit: Could do the full fstp on one line like before.

https://codereview.chromium.org/695993004/diff/20001/src/IceInstX8632.cpp#new...
src/IceInstX8632.cpp:2720: Str << "\txchg\t";
I'm surprised this doesn't have the getWidthString()

https://codereview.chromium.org/695993004/diff/20001/src/IceInstX8632.cpp#new...
src/IceInstX8632.cpp:2767: assert(0);
could do llvm_unreachable("...")

https://codereview.chromium.org/695993004/diff/20001/src/IceTargetLoweringX86...
File src/IceTargetLoweringX8632.cpp (right):

https://codereview.chromium.org/695993004/diff/20001/src/IceTargetLoweringX86...
src/IceTargetLoweringX8632.cpp:4607: // llvm-mc doesn't parse "dword ptr
[.L$foo]".
adjust comment about "dword ptr" ?

[Jim Stichnoth, 2014-11-01 13:44:17]

https://codereview.chromium.org/695993004/diff/20001/src/IceInstX8632.cpp
File src/IceInstX8632.cpp (right):

https://codereview.chromium.org/695993004/diff/20001/src/IceInstX8632.cpp#new...
src/IceInstX8632.cpp:2310: << "\n";
On 2014/10/31 22:12:18, jvoung wrote:
> \n could fit on the previous line next to %esp

Done.

https://codereview.chromium.org/695993004/diff/20001/src/IceInstX8632.cpp#new...
src/IceInstX8632.cpp:2385: << "(%esp)\n";
On 2014/10/31 22:12:18, jvoung wrote:
> nit: Could do the full fstp on one line like before.

Done.

https://codereview.chromium.org/695993004/diff/20001/src/IceInstX8632.cpp#new...
src/IceInstX8632.cpp:2720: Str << "\txchg\t";
On 2014/10/31 22:12:18, jvoung wrote:
> I'm surprised this doesn't have the getWidthString()

Good point.  It turns out that for both llvm-mc and gas (and objdump -d as
well), the b/w/l opcode suffix is optional if it can be inferred from an
explicit register operand.  E.g., "add $1, %ax" is OK, but "add $1, (%esp)" is
not.  Since the xchg instruction always has at least one register operand, it
ends up being OK to never emit the opcode suffix.

All the same, I added it for consistency.

https://codereview.chromium.org/695993004/diff/20001/src/IceInstX8632.cpp#new...
src/IceInstX8632.cpp:2767: assert(0);
On 2014/10/31 22:12:18, jvoung wrote:
> could do llvm_unreachable("...")

Done.

https://codereview.chromium.org/695993004/diff/20001/src/IceTargetLoweringX86...
File src/IceTargetLoweringX8632.cpp (right):

https://codereview.chromium.org/695993004/diff/20001/src/IceTargetLoweringX86...
src/IceTargetLoweringX8632.cpp:4607: // llvm-mc doesn't parse "dword ptr
[.L$foo]".
On 2014/10/31 22:12:18, jvoung wrote:
> adjust comment about "dword ptr" ?

Even better, adjusted the code to use .L$ prefix.

[jvoung (off chromium), 2014-11-01 16:33:13]

LGTM

[Jim Stichnoth, 2014-11-01 17:13:59]

Committed patchset #3 (id:40001) manually as
bca2f65551c65fdc18278bf88025c11d8af18579 (presubmit successful).