hello.
What is the actual precision of long double on Intel 64-bit platforms? is it 80 bits padded to 128 or actual 128 bit?
if former, besides going gmp, is there another option to achieve true 128 precision?
+6
A:
x86-64 precision is the same as regular x86. Extended double is 80 bits, using the x87 ISA, with 6 padding bytes added. There is no 128-bit FP hardware.
A software implementation of quad or extended quad precision might benefit from the x86-64 64x64 => 128 integer multiply instruction, though.
Potatoswatter
2010-05-10 00:16:09
+2
A:
There is a good chance that it's 64 bit for both (depending on the compiler and OS), because the compiler is emitting scalar SSE2 instead of x87 instructions.
x86 doesn't support higher precision than 80 bits, but if you really need more than 64 bits for a FP algorithm most likely you should check your numerics instead of solving the problem with brute force.
Axel Gneiting
2010-05-10 00:36:34
I am doing comparisons between using different precision and accuracy of result.
aaa
2010-05-10 00:39:55
Have you ever observed such behavior? GCC, at least, I don't think can be prohibited from emitting x87.
Potatoswatter
2010-05-10 00:45:37
Potatoswatter: Try `-msse2 -mfpmath=sse`, which should cause it to use SSE instructions for `double` and `float` s - although it likely still uses x87 instructions for `long double` s.
caf
2010-05-10 01:05:10
Microsoft compilers all emit scalar SSE2. x87 is deprecated on Windows for x64 programs.I'm pretty sure that GCC does the same, because it's also faster and x64 CPUs are guaranteed to have support for SSE2.But there might be the case that it still emits x87 for long double.
Axel Gneiting
2010-05-10 01:35:23
@caf: those are the default for x64, but no options affect `long double` that I see in the manpage.
Potatoswatter
2010-05-10 03:32:14
The x86-64 ABI ( http://www.x86-64.org/documentation/abi.pdf ) actually specifies that `long double` is the x87's 80-bit extended precision format.
caf
2010-05-10 04:16:40
Not all systems use the official ABI.
Axel Gneiting
2010-05-10 10:59:27