On Mar 22, 2007, at 1:18 PM, Reid Spencer wrote:
> On Thu, 2007-03-22 at 12:35 -0700, Jeff Cohen wrote:
>> How is it any worse than checking for GCC?
>
> I don't like that either :)
FWIW, I don't see anything wrong with Jeff's change. MathExtras may
be suboptimal w.r.t to libsystem, but so is
On Thu, 2007-03-22 at 12:35 -0700, Jeff Cohen wrote:
> How is it any worse than checking for GCC?
I don't like that either :)
>
> It's treated as an intrinsic when optimizations are enable. It directly
> generates a bswap instruction. Same as the conditional code for GCC.
Okay, sounds like i
Reid Spencer wrote:
>
>> -for (uint64_t Shift = 64 >> 1; Shift; Shift >>= 1) {
>> +for (unsigned Shift = 64 >> 1; Shift; Shift >>= 1) {
>>
>
> Why not uint32_t? For conformity with the rest of your changes?
The size of a shift amount is unrelated to the size of the function's
argumen
How is it any worse than checking for GCC?
It's treated as an intrinsic when optimizations are enable. It directly
generates a bswap instruction. Same as the conditional code for GCC.
Reid Spencer wrote:
> Jeff,
>
> On Thu, 2007-03-22 at 14:12 -0500, Jeff Cohen wrote:
>
>> Changes in direct
Jeff,
On Thu, 2007-03-22 at 14:12 -0500, Jeff Cohen wrote:
>
> Changes in directory llvm/include/llvm/Support:
>
> MathExtras.h updated: 1.43 -> 1.44
> ---
> Log message:
>
> Be more explicit concerning argument sizes.
> Use VC++ byteswap intrinsics.
... snip ...
> @@ -93,22 +93,30 @@
>
>
