Analysis of cycle costs for SH4:

-> udiv_le128:            5
-> udiv_ge64k:            6
-> udiv udiv_25:         10
-> pos_divisor:           3
-> pos_result linear:     5
-> pos_result - -:        5
-> div_le128:             7
-> div_ge64k:             9
sdivsi3 -> udiv_25             13
udiv25 -> div_ge64k_end:       15
div_ge64k_end -> rts:          13
div_le128 -> div_le128_2:       2, r1 latency 3
udiv_le128 -> div_le128_2:      2, r1 latency 3
(u)div_le128 -> div_by_1:       9
(u)div_le128 -> rts:           17
div_by_1(_neg) -> rts:          4
div_ge64k -> div_r8:            2
div_ge64k -> div_ge64k_2:       3
udiv_ge64k -> udiv_r8:          3
udiv_ge64k -> div_ge64k_2:      3 + LS
(u)div_ge64k -> div_ge64k_end: 13
div_r8 -> div_r8_2:             2
udiv_r8 -> div_r8_2:            2 + LS
(u)div_r8 -> rts:              21

-> - + neg_result:             5
-> + - neg_result:             5
-> div_le128_neg:              7
-> div_ge64k_neg:              9
-> div_r8_neg:                11
-> <64k div_ge64k_neg_end:    28
-> >=64k div_ge64k_neg_end:   22
div_ge64k_neg_end ft -> rts:  14
div_r8_neg_end -> rts:         4
div_r8_neg -> div_r8_neg_end: 18
div_le128_neg -> div_by_1_neg: 4
div_le128_neg -> rts          18

         sh4-200    absolute divisor range:
            1  [2..128]  [129..64K) [64K..|dividend|/256] >=64K,>|dividend/256|
udiv       18     22         38            32                   30
sdiv pos:  20     24         41            35                   32
sdiv neg:  15     25         42            36                   33

         sh4-300    absolute divisor range:
                 8 bit      16 bit       24 bit              > 24 bit
udiv              15         35            28                   25
sdiv              14         36            34                   31


fp-based:

unsigned: 42 + 3 + 3 (lingering ftrc latency + sts fpul,rx) at caller's site
signed: 33 + 3 + 3 (lingering ftrc latency + sts fpul,rx) at caller's site

call-div1:    divisor range:
              [1..64K)  >= 64K
unsigned:       63        58
signed:         76        76

SFUNC_STATIC call overhead:
mov.l 0f,r1
bsrf r1

SFUNC_GOT call overhead - current:
mov.l 0f,r1
mova 0f,r0
mov.l 1f,r2
add r1,r0
mov.l @(r0,r2),r0
jmp @r0
; 3 cycles worse than SFUNC_STATIC

SFUNC_GOT call overhead - improved assembler:
mov.l 0f,r1
mova 0f,r0
mov.l @(r0,r1),r0
jmp @r0
; 2 cycles worse than SFUNC_STATIC


Copyright (C) 2006-2022 Free Software Foundation, Inc.

Copying and distribution of this file, with or without modification,
are permitted in any medium without royalty provided the copyright
notice and this notice are preserved.