Where are the integer math functions implemented?
SRLM
Posts: 5,045
Where are the integer math functions such as division, multiplication, and modulus implemented? I assume addition and subtraction is done by hardware, but we can't do that for */%.
I have a curious timing problem that seems to depend on the particular modulus values, and I think it has to do with the implementation. In my case, I *think* that it's taking less time to do i%550 than i%549 or i%551
I have a curious timing problem that seems to depend on the particular modulus values, and I think it has to do with the implementation. In my case, I *think* that it's taking less time to do i%550 than i%549 or i%551
Comments
In the current default branch the division functions are in gcc/gcc/config/propeller/kernel.ext (they used to be in the various kernels, like crt0_lmm.s). The multiplication routine is still in the kernel. The division and modulus routines try to estimate the number of iterations required by counting the leading 0's of both the dividend and divisor. I don't think i % 550 should take any different amount of time than i %549, since both 550 and 549 have the same number of bits.
Here are the inner loops:
__MULSI mov __TMP0, r0 min __TMP0, r1 max r1, r0 mov r0, #0 __MULSI_loop shr r1, #1 wz, wc IF_C add r0, __TMP0 add __TMP0, __TMP0 IF_NZ jmp #__MULSI_loop __MULSI_ret ret __UDIVSI_impl mov __DIVR, r0 call #__CLZSI_impl neg __DIVCNT, r0 mov r0, r1 wz IF_Z jmp #__UDIV_BY_ZERO call #__CLZSI_impl add __DIVCNT, r0 mov r0, #0 cmps __DIVCNT, #0 wz, wc IF_C jmp #__UDIVSI_done shl r1, __DIVCNT add __DIVCNT, #1 __UDIVSI_loop cmpsub __DIVR, r1 wz, wc addx r0, r0 shr r1, #1 djnz __DIVCNT, #__UDIVSI_loop __UDIVSI_done mov r1, __DIVR __UDIVSI_impl_ret ret