Subzero. X86. Refactors initRegisterSet.

Subzero. X86. Refactors initRegisterSet.

initRegisterSet() for x8632 and x8664 were both huge. This CL refactors
those methods to use a pre-initialized table instead of the result of
expanding the x-macros.

BUG=
R=stichnot@chromium.org

Committed: https://gerrit.chromium.org/gerrit/gitweb?p=native_client/pnacl-subzero.git;a=commit;h=4a308cedda6bfe381dc9506f3508c4547a06275f

[John, 2015-12-28 23:37:02]

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

[Jim Stichnoth, 2015-12-30 02:50:24]

lgtm

https://codereview.chromium.org/1546373003/diff/20001/src/IceTargetLoweringX8...
File src/IceTargetLoweringX8632Traits.h (right):

https://codereview.chromium.org/1546373003/diff/20001/src/IceTargetLoweringX8...
src/IceTargetLoweringX8632Traits.h:590: // We need to subtract one to account
for Reg_Invalid.
Any chance this would be unnecessary, and everything else would work out, if you
set Reg_Invalid=-1?

[John, 2015-12-30 17:55:07]

Description was changed from

==========
Subzero. X86. Refactors initRegisterSet.

initRegisterSet() for x8632 and x8664 were both huge. This CL refactors
those methods to use a pre-initialized table instead of the result of
expanding the x-macros.

BUG=
==========

to

==========
Subzero. X86. Refactors initRegisterSet.

initRegisterSet() for x8632 and x8664 were both huge. This CL refactors
those methods to use a pre-initialized table instead of the result of
expanding the x-macros.

BUG=
R=stichnot@chromium.org

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

[John, 2015-12-30 17:55:08]

Committed patchset #4 (id:60001) manually as
4a308cedda6bfe381dc9506f3508c4547a06275f (presubmit successful).

[John, 2015-12-30 17:55:37]

These changes were not trivial, but I have an upcoming CL that depends on this
one. Given that you're on vacation, I'll land as is, but I'll address any
comments you have on a future cl.

https://codereview.chromium.org/1546373003/diff/20001/src/IceTargetLoweringX8...
File src/IceTargetLoweringX8632Traits.h (right):

https://codereview.chromium.org/1546373003/diff/20001/src/IceTargetLoweringX8...
src/IceTargetLoweringX8632Traits.h:590: // We need to subtract one to account
for Reg_Invalid.
On 2015/12/30 02:50:24, stichnot wrote:
> Any chance this would be unnecessary, and everything else would work out, if
you
> set Reg_Invalid=-1?

Yes, and no. I don't know anyway of telling the linker to initializer

SizeT Aliases[4] = {xmm0};

as

SizeT Aliases[4] = {xmm0, -1, -1, -1};

And given that the alias list comes from a macro expansion within a macro, and
that said macro is used by other targets (e.g., ARM32) I decided against
changing the macro.

I could have changed the IceInstX86??.def files so that all register alias list
would have a terminator -- yuck.

I could also upgrade our toolchain to build sz using c++14, which defines
std::initializer_list<>'s ctor as constexpr (Too much work for 12/30).

My approach was to define a helper class

class SizeOf {
  template <...> SizeOf(...) {}
  SizeOf()
};

that computes the size of the initializer list. I abuse the uniform initializer
syntax (i.e., invoking ctors using {args}) to avoid even moar macro trickery for
removing the { and } from the alias list.

in the end, this solution is not as bad as it could've been. and given that this
is a temporary stopgap until c++14, I am not too sad about defining this new
helper class.