How multiprecision optimizes allocation?
I have tested boost::mulitprecision::cpp_int and have tested simplest algorithm for int multiplication My test method is several times slower than boost multiplication because I resized product vector to size fist + size second. This allocation take most time, if I use static array on stack, it will be fast for small numbers. What is the riddle of fast comnputing z = x*y in loop for quite small x and y in Boost cpp_int?
On 05/11/2019 06:12, Andy via Boost-users wrote:
I have tested boost::mulitprecision::cpp_int and have tested simplest algorithm for int multiplication My test method is several times slower than boost multiplication because I resized product vector to size fist + size second. This allocation take most time, if I use static array on stack, it will be fast for small numbers. What is the riddle of fast comnputing z = x*y in loop for quite small x and y in Boost cpp_int?
There's no allocation for small numbers. John. -- This email has been checked for viruses by Avast antivirus software. https://www.avast.com/antivirus
sizeof(boost::multiprecision::cpp_int) is only 24 bytes = 6*32 bit limb This means, that upto 24 bytes is stored at stack on variable itself, and bigger are allocated?
On 05/11/2019 09:12, Andy via Boost-users wrote:
sizeof(boost::multiprecision::cpp_int) is only 24 bytes = 6*32 bit limb This means, that upto 24 bytes is stored at stack on variable itself, and bigger are allocated?
There are other member variables making up that size - the default (as used by type cpp_int) is never less than 2 whole limbs (so 128 bits when __int128 is available), otherwise however many limbs will fit inside sizeof(unsigned)+sizeof(void*), which is 4 32-bit limbs on MSVC. -- This email has been checked for viruses by Avast antivirus software. https://www.avast.com/antivirus
participants (2)
-
Andy -
John Maddock