Only two changes appear in the compiled RTL, both in SysUtils, but one
is pretty significant. Here's the "IsLeapYear" function under i386-win32:
Trunk:
.section .text.n_sysutils_$$_isleapyear$word$$boolean,"ax"
.balign 16,0x90.globl SYSUTILS_$$_ISLEAPYEAR$WORD$$BOOLEAN
So I've got some pretty good headway so far!
Trunk:
Unsigned 32-bit (n mod 3) = 0 - Pass - average iteration
duration: 0.757 ns
Signed 32-bit (n mod 3) = 0 - Pass - average iteration
duration: 6.403 ns
Unsigned 32-bit (n mod 10) = 0 - Pass - average
I suppose in truth, I can, and that in itself is probably fairly
cross-platform (although I'll stick with x86 for the moment and get that
working). Sometimes the simple solution eludes me! Is there anything I
need to take into account when it comes to range checking (that is, if a
third
Am 10.09.21 um 21:17 schrieb J. Gareth Moreton via fpc-devel:
Hi everyone,
I'm looking at ways to optimise div and mod, starting with x86 and then
probably AArch64. The obvious one is attempting to merge "Q := N div D;
R := N mod D;", where D is a variable (but invariant between the two
Hi everyone,
I'm looking at ways to optimise div and mod, starting with x86 and then
probably AArch64. The obvious one is attempting to merge "Q := N div D;
R := N mod D;", where D is a variable (but invariant between the two
instructions), since DIV returns the quotient in R/EAX and the
