Process two 32-bit digits as one 64-bit digit in all bigint intrinsics on x64.

Process two 32-bit digits as one 64-bit digit in all bigint intrinsics on x64.

R=zra@google.com

Committed: https://code.google.com/p/dart/source/detail?r=42607

[regis, 2014-12-31 01:10:11]

regis@google.com changed reviewers:
+ rmacnak@google.com, zra@google.com

[regis, 2014-12-31 01:10:12]

This speeds up the RSA benchmark by a factor two on X64.
The same change for ARM64 is coming next.

However, a new function is crashing the fuzzer, so I black listed it for now. I
already spent quite some time debugging this, but I cannot figure out how the
fuzzer is seing a bigint of value 1 being parsed as literal. I'll figure it out
next year.

Happy new year!

[zra, 2015-01-05 16:39:22]

lgtm

Are there any instructions that were added for the one-digit-at-a-time version
no longer used in the two-digits-at-a-time version? If so, should they be left
in or removed?

https://codereview.chromium.org/786933009/diff/1/runtime/vm/intrinsifier_x64.cc
File runtime/vm/intrinsifier_x64.cc (right):

https://codereview.chromium.org/786933009/diff/1/runtime/vm/intrinsifier_x64....
runtime/vm/intrinsifier_x64.cc:946: //     uint364_t aj = *ajp;
uint64_t?

https://codereview.chromium.org/786933009/diff/1/tests/lib/mirrors/invocation...
File tests/lib/mirrors/invocation_fuzz_test.dart (right):

https://codereview.chromium.org/786933009/diff/1/tests/lib/mirrors/invocation...
tests/lib/mirrors/invocation_fuzz_test.dart:6: // invoke them will various basic
values (nulls, ints, etc). This may result in
will -> with

[regis, 2015-01-05 19:43:24]

On 2015/01/05 16:39:22, zra wrote:
> lgtm
> 
> Are there any instructions that were added for the one-digit-at-a-time version
> no longer used in the two-digits-at-a-time version? If so, should they be left
> in or removed?
> 
>
https://codereview.chromium.org/786933009/diff/1/runtime/vm/intrinsifier_x64.cc
> File runtime/vm/intrinsifier_x64.cc (right):
> 
>
https://codereview.chromium.org/786933009/diff/1/runtime/vm/intrinsifier_x64....
> runtime/vm/intrinsifier_x64.cc:946: //     uint364_t aj = *ajp;
> uint64_t?
> 
>
https://codereview.chromium.org/786933009/diff/1/tests/lib/mirrors/invocation...
> File tests/lib/mirrors/invocation_fuzz_test.dart (right):
> 
>
https://codereview.chromium.org/786933009/diff/1/tests/lib/mirrors/invocation...
> tests/lib/mirrors/invocation_fuzz_test.dart:6: // invoke them will various
basic
> values (nulls, ints, etc). This may result in
> will -> with

Yes, there were some instructions added for the 32-bit version. I feel that
there is not much cost in keeping them in. Adding new instructions is painful,
because of the required assembler tests. I'd prefer to leave them in.

[regis, 2015-01-05 19:44:08]

Committed patchset #2 (id:20001) manually as r42607 (presubmit successful).

[regis, 2017-01-13 02:34:46]

https://codereview.chromium.org/786933009/diff/1/runtime/vm/intrinsifier_x64.cc
File runtime/vm/intrinsifier_x64.cc (right):

https://codereview.chromium.org/786933009/diff/1/runtime/vm/intrinsifier_x64....
runtime/vm/intrinsifier_x64.cc:946: //     uint364_t aj = *ajp;
On 2015/01/05 16:39:22, zra wrote:
> uint64_t?

Done.

https://codereview.chromium.org/786933009/diff/1/tests/lib/mirrors/invocation...
File tests/lib/mirrors/invocation_fuzz_test.dart (right):

https://codereview.chromium.org/786933009/diff/1/tests/lib/mirrors/invocation...
tests/lib/mirrors/invocation_fuzz_test.dart:6: // invoke them will various basic
values (nulls, ints, etc). This may result in
On 2015/01/05 16:39:22, zra wrote:
> will -> with

Done.