RFR: 8291649: multiple tests failing with -Xcomp after JDK-8290034 [v2]

[John R Rose]

On Wed, 24 Aug 2022 05:02:13 GMT, Kim Barrett <kbarrett at openjdk.org> wrote:

>> Jatin Bhateja has updated the pull request incrementally with one additional commit since the last revision:
>> 
>>   8291649: Review comments resolutions.
>
> src/hotspot/share/utilities/moveBits.hpp line 63:
> 
>> 61: /*****************************************************************************
>> 62:  * GCC and compatible (including Clang)
>> 63:  *****************************************************************************/
> 
> Is it really worth the extra code to allow the use of the gcc builtins?

Kim, is there a way to do this in less code?

Seems to me that something like this could work:


template<typename T, size_t S>
T reverse_bytes(T x) {
  if (size == 1)  return x;
  T alt8bs = (T)((uint64_t)-1 / 0xFFFF * 0xFF);  // 0x00FF…00FF
  x = ((x & alt8bs) << 8) | (((uint64_t)x >> 8 & alt8bs);
  if (size == 2)  return x;
  T alt16bs = (T)((uint64_t)-1 / 0xFFFFFFFFll * 0xFFFF);  // 0x0000FFFF…0000FFFF
  …
}


That is, one function bodies handles cases for 1/2/4/8 byte values.

Could it be just a function template, with an explicit instantiation (overriding the general template) where it matches an intrinsic?

-------------

PR: https://git.openjdk.org/jdk/pull/9987