FFTs should keep the GIL locked, because sharing the interpreter is not funny. Array indices should be sizeof(void*), because following C standard and Python C API is lame. Median should be calculated in O(n log n) instead of O(n) time, because redundant sorting is gymnatics for the CPU. But at least something happened to the tickets, after a year or two :-) Sturla
Fri, 25 Mar 2011 13:23:11 +0100, Sturla Molden wrote:
FFTs should keep the GIL locked, because sharing the interpreter is not funny. Array indices should be sizeof(void*), because following C standard and Python C API is lame. Median should be calculated in O(n log n) instead of O(n) time, because redundant sorting is gymnatics for the CPU.
That the ticket was changed "unscheduled" does not mean it will not be implemented when someone finds the extra time for it. That npy_intp will not be redefined as ssize_t does not mean that the type of array indices could not be changed.
Den 25.03.2011 13:33, skrev Pauli Virtanen:
That npy_intp will not be redefined as ssize_t does not mean that the type of array indices could not be changed.
By the way, what is the resonable array index for AMD64, where 32-bit is the native offset size but size_t is 64 bit? Sturla
Fri, 25 Mar 2011 13:40:54 +0100, Sturla Molden wrote:
Den 25.03.2011 13:33, skrev Pauli Virtanen:
That npy_intp will not be redefined as ssize_t does not mean that the type of array indices could not be changed.
By the way, what is the resonable array index for AMD64, where 32-bit is the native offset size but size_t is 64 bit?
Well, whatever the address space is. Since 64-bit programs on amd64 have in principle 64 bits of address space, size_t should be OK. Pauli
participants (2)
-
Pauli Virtanen
-
Sturla Molden