ARM64: Clean up support for explicit literal load.

ARM64: Clean up support for explicit literal load.

This is the first patch to improve literal pool handling in arm64. Cleans up
assembler/macro-assembler access to literal pools.

BUG=
R=rmcilroy@chromium.org

Committed: https://code.google.com/p/v8/source/detail?r=21723

[ulan, 2014-06-05 16:41:48]

Ross, could you please take over this? I'll be out of office until Tuesday.

[rmcilroy, 2014-06-06 22:48:58]

Lgtm with a couple of suggestions. Thanks.

https://codereview.chromium.org/318773009/diff/1/src/arm64/assembler-arm64.cc
File src/arm64/assembler-arm64.cc (right):

https://codereview.chromium.org/318773009/diff/1/src/arm64/assembler-arm64.cc...
src/arm64/assembler-arm64.cc:1482: Emit(LoadLiteralOpFor(rt) |
ImmLLiteral(imm19) | Rt(rt));
How about having the argument be a byte offset from PC (like the old
LoadLiteral). This seems clearer to me. WDYT?

https://codereview.chromium.org/318773009/diff/1/src/arm64/macro-assembler-ar...
File src/arm64/macro-assembler-arm64.h (right):

https://codereview.chromium.org/318773009/diff/1/src/arm64/macro-assembler-ar...
src/arm64/macro-assembler-arm64.h:397: // Load a literal from an inline pool.
Nit, "Load a literal from the inline constant pool."

https://codereview.chromium.org/318773009/diff/1/src/arm64/macro-assembler-ar...
src/arm64/macro-assembler-arm64.h:399: // helper function for double immediate.
nit - Helper

[Rodolph Perfetta (ARM), 2014-06-09 14:12:21]

https://codereview.chromium.org/318773009/diff/1/src/arm64/assembler-arm64.cc
File src/arm64/assembler-arm64.cc (right):

https://codereview.chromium.org/318773009/diff/1/src/arm64/assembler-arm64.cc...
src/arm64/assembler-arm64.cc:1482: Emit(LoadLiteralOpFor(rt) |
ImmLLiteral(imm19) | Rt(rt));
It is more natural I agree but up to now at the assembler level we tried to
match closely the instruction encoding, for example b and bl have their offset
specified in instructions not in bytes. So I am inclined to leave it as it is. I
am happy to push a follow up CL if we decide to change this.

On 2014/06/06 22:48:58, rmcilroy wrote:
> How about having the argument be a byte offset from PC (like the old
> LoadLiteral). This seems clearer to me. WDYT?

https://codereview.chromium.org/318773009/diff/1/src/arm64/macro-assembler-ar...
File src/arm64/macro-assembler-arm64.h (right):

https://codereview.chromium.org/318773009/diff/1/src/arm64/macro-assembler-ar...
src/arm64/macro-assembler-arm64.h:397: // Load a literal from an inline pool.
On 2014/06/06 22:48:58, rmcilroy wrote:
> Nit, "Load a literal from the inline constant pool."

Done.

https://codereview.chromium.org/318773009/diff/1/src/arm64/macro-assembler-ar...
src/arm64/macro-assembler-arm64.h:399: // helper function for double immediate.
On 2014/06/06 22:48:58, rmcilroy wrote:
> nit - Helper

Done.

[rmcilroy, 2014-06-09 14:22:10]

lgtm

https://codereview.chromium.org/318773009/diff/1/src/arm64/assembler-arm64.cc
File src/arm64/assembler-arm64.cc (right):

https://codereview.chromium.org/318773009/diff/1/src/arm64/assembler-arm64.cc...
src/arm64/assembler-arm64.cc:1482: Emit(LoadLiteralOpFor(rt) |
ImmLLiteral(imm19) | Rt(rt));
On 2014/06/09 14:12:21, Rodolph Perfetta (ARM) wrote:
> It is more natural I agree but up to now at the assembler level we tried to
> match closely the instruction encoding, for example b and bl have their offset
> specified in instructions not in bytes. So I am inclined to leave it as it is.
I
> am happy to push a follow up CL if we decide to change this.
> 
> On 2014/06/06 22:48:58, rmcilroy wrote:
> > How about having the argument be a byte offset from PC (like the old
> > LoadLiteral). This seems clearer to me. WDYT?
> 

I prefer that it stays consistent with what's already here, so keeping it as
you've done it sounds good.

[Rodolph Perfetta (ARM), 2014-06-09 14:24:33]

Committed patchset #1 manually as r21723 (presubmit successful).