projects / thirdparty / gcc.git / blame
| Commit | Line | Data |
|---|---|---|
| b368d6b8 R |
1 | Analysis of cycle costs for SH4: |
| 2 | ||
| 3 | -> udiv_le128: 5 | |
| 4 | -> udiv_ge64k: 6 | |
| 5 | -> udiv udiv_25: 10 | |
| 6 | -> pos_divisor: 3 | |
| 7 | -> pos_result linear: 5 | |
| 8 | -> pos_result - -: 5 | |
| 9 | -> div_le128: 7 | |
| 10 | -> div_ge64k: 9 | |
| 11 | sdivsi3 -> udiv_25 13 | |
| 12 | udiv25 -> div_ge64k_end: 15 | |
| 13 | div_ge64k_end -> rts: 13 | |
| 14 | div_le128 -> div_le128_2: 2, r1 latency 3 | |
| 15 | udiv_le128 -> div_le128_2: 2, r1 latency 3 | |
| 16 | (u)div_le128 -> div_by_1: 9 | |
| 17 | (u)div_le128 -> rts: 17 | |
| 18 | div_by_1(_neg) -> rts: 4 | |
| 19 | div_ge64k -> div_r8: 2 | |
| 20 | div_ge64k -> div_ge64k_2: 3 | |
| 21 | udiv_ge64k -> udiv_r8: 3 | |
| 22 | udiv_ge64k -> div_ge64k_2: 3 + LS | |
| 23 | (u)div_ge64k -> div_ge64k_end: 13 | |
| 24 | div_r8 -> div_r8_2: 2 | |
| 25 | udiv_r8 -> div_r8_2: 2 + LS | |
| 26 | (u)div_r8 -> rts: 21 | |
| 27 | ||
| 28 | -> - + neg_result: 5 | |
| 29 | -> + - neg_result: 5 | |
| 30 | -> div_le128_neg: 7 | |
| 31 | -> div_ge64k_neg: 9 | |
| 32 | -> div_r8_neg: 11 | |
| 33 | -> <64k div_ge64k_neg_end: 28 | |
| 34 | -> >=64k div_ge64k_neg_end: 22 | |
| 35 | div_ge64k_neg_end ft -> rts: 14 | |
| 36 | div_r8_neg_end -> rts: 4 | |
| 37 | div_r8_neg -> div_r8_neg_end: 18 | |
| 38 | div_le128_neg -> div_by_1_neg: 4 | |
| 39 | div_le128_neg -> rts 18 | |
| 40 | ||
| 78d310c2 | 41 | sh4-200 absolute divisor range: |
| 0d52bcc1 | 42 | 1 [2..128] [129..64K) [64K..|dividend|/256] >=64K,>|dividend/256| |
| b368d6b8 R |
43 | udiv 18 22 38 32 30 |
| 44 | sdiv pos: 20 24 41 35 32 | |
| 45 | sdiv neg: 15 25 42 36 33 | |
| 46 | ||
| 78d310c2 R |
47 | sh4-300 absolute divisor range: |
| 48 | 8 bit 16 bit 24 bit > 24 bit | |
| 49 | udiv 15 35 28 25 | |
| 50 | sdiv 14 36 34 31 | |
| 51 | ||
| b368d6b8 R |
52 | |
| 53 | fp-based: | |
| 54 | ||
| 55 | unsigned: 42 + 3 + 3 (lingering ftrc latency + sts fpul,rx) at caller's site | |
| 56 | signed: 33 + 3 + 3 (lingering ftrc latency + sts fpul,rx) at caller's site | |
| 57 | ||
| 58 | call-div1: divisor range: | |
| 59 | [1..64K) >= 64K | |
| 60 | unsigned: 63 58 | |
| 61 | signed: 76 76 | |
| 62 | ||
| 63 | SFUNC_STATIC call overhead: | |
| 64 | mov.l 0f,r1 | |
| 65 | bsrf r1 | |
| 66 | ||
| 67 | SFUNC_GOT call overhead - current: | |
| 68 | mov.l 0f,r1 | |
| 69 | mova 0f,r0 | |
| 70 | mov.l 1f,r2 | |
| 71 | add r1,r0 | |
| 72 | mov.l @(r0,r2),r0 | |
| 73 | jmp @r0 | |
| 74 | ; 3 cycles worse than SFUNC_STATIC | |
| 75 | ||
| 76 | SFUNC_GOT call overhead - improved assembler: | |
| 77 | mov.l 0f,r1 | |
| 78 | mova 0f,r0 | |
| 79 | mov.l @(r0,r1),r0 | |
| 80 | jmp @r0 | |
| 81 | ; 2 cycles worse than SFUNC_STATIC | |
| ad41bd84 JM |
82 | |
| 83 | \f | |
| 8d9254fc | 84 | Copyright (C) 2006-2020 Free Software Foundation, Inc. |
| ad41bd84 JM |
85 | |
| 86 | Copying and distribution of this file, with or without modification, | |
| 87 | are permitted in any medium without royalty provided the copyright | |
| 88 | notice and this notice are preserved. |