1 // peeph.def - Common Z80 and gbz80 peephole rules
3 // These peepholes could be potentially moved to peeph.def, but a GBZ80 expert
4 // Should have a look at them before.
6 // (c) Philipp Klaus Krause (pkk@spth.de, philipp@colecovision.eu) 2006 - 2008
8 // This program is free software; you can redistribute it and/or modify it
9 // under the terms of the GNU General Public License as published by the
10 // Free Software Foundation; either version 2, or (at your option) any
13 // This program is distributed in the hope that it will be useful,
14 // but WITHOUT ANY WARRANTY; without even the implied warranty of
15 // MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
16 // GNU General Public License for more details.
18 // You should have received a copy of the GNU General Public License
19 // along with this program; if not, write to the Free Software
20 // Foundation, 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
25 ; peephole -1 removed redundant load.
28 // This one doesn't work currently since the peephole optimizer can't match lines generated by multiline macros: Bug #1570701
33 // ; peephole -1a eleminated dead pop.
40 ; peephole 0a removed dead load from %2 into %1.
41 } if notVolatile(%1), notUsed(%1)
42 // Should probably check for notVolatile(%2), too, but gives many false positives and no regression tests fail.
48 ; peephole 0b loaded %2 into a directly instead of going through %1.
50 } if notVolatile(%1), notUsed(%1)
56 ; peephole 0c loaded %2 (%3) into a directly instead of going through %1.
58 } if notVolatile(%1), notUsed(%1)
64 ; peephole 0d loaded %2 into a directly instead of going through %1.
73 ; peephole 0e shifted in a instead of %1.
75 } if notVolatile(%1), notUsed(%1)
83 ; peephole 0f used (hl) in sub directly instead of going through %1.
85 } if notVolatile(%1), notUsed(%1)
95 ; peephole 0g incremented in hl instead of de.
107 ; peephole 0h incremented in hl instead of de.
117 ; peephole 0i loaded %2%1 into hl directly instead of going through de.
127 ; peephole 0j used hl for #%2 + %3 instead of bc, not going through a.
130 } if notUsed('a'), notUsed('hl')
136 ; peephole 0k pushed de directly instead of going through hl.
138 } if notUsed('de'), notUsed('hl')
144 ; peephole 0k' pushed de directly instead of going through hl.
146 } if notUsed('de'), notUsed('hl')
153 ; peephole 0l pushed hl directly instead of going through de.
156 } if notUsed('de'), notUsed('hl')
163 ; peephole 0m pushed bc directly instead of going through hl.
172 ; peephole 0n removed redundant load of a through c.
181 ; peephole 0o used de directly instead of going through bc.
190 ; peephole 0p popped hl directly instead of going through de.
198 ; peephole 0q loaded a into %2 (%3) directly instead of going through %1.
200 } if notVolatile(%1), notUsed(%1)
206 ; peephole 0r loaded %3 from %2 instead of going through %1 (ix).
210 // Don't need to check for volatile, since ix is used for the stack.
216 ; peephole 0s loaded b from %2 (%3) instead of going through %1.
218 } if notVolatile(%1), notUsed(%1)
225 ; peephole 0t loaded %4 from a instead of going through %1.
229 } if notVolatile(%1), operandsNotRelated(%1 %3), operandsNotRelated(%1 %2)
235 ; peephole 0u loaded e from (hl) instead of going through %1.
243 ; peephole 0v loaded %2 (%3) from l instead of going through %1.
258 ; peephole 0w shifted (hl) in place.
262 // Don't check for volatile since ix points to the stack.
271 ; peephole 0x pushed de instead of pushing a twice.
274 } if notUsed('d'), notUsed('a')
283 ; peephole 0y pushed bc instead of pushing a twice.
286 } if notUsed('b'), notUsed('a')
292 ; peephole 0z loaded #%1 into d directly instead of going through a.
304 ; peephole 0za pushed %1 (ix), %2(ix) through hl instead of af.
308 } if notUsed('a'), notUsed('hl')
315 ; peephole 0zb pushed hl instead of bc.
319 // Doesn'T work due to bug #1947081
324 // ; peephole 0zc eleminated dead pop/push pair.
331 ; peephole 0zd used hl instead of iy.
334 } if notUsed('iy'), notUsed('hl')
340 ; peephole 0ze used hl instead of iy.
343 } if notUsed('iy'), notUsed('hl')
349 ; peephole 0zf used hl instead of iy.
352 } if notUsed('iy'), notUsed('hl')
359 ; peephole 0zg loaded %1 from (hl) directly instead of going through (de).
368 ; peephole 0zh loaded %1 from (hl) directly instead of going through (bc).
377 ; peephole 0zi incremented in hl instead of bc.
387 ; peephole 0zj tested bit of %1 (%2) directly instead of going through a.
396 ; peephole 0zk stored constant #%1 + %2 into hl directly instead of going through bc.
406 ; peephole 3 removed jp by using inverse jump logic
408 } if labelRefCountChange(%1 -1)
416 ; peephole 4 removed jp by using inverse jump logic
418 } if labelRefCountChange(%1 -1)
426 ; peephole 5 removed jp by using inverse jump logic
428 } if labelRefCountChange(%1 -1)
436 ; peephole 6 removed jp by using inverse jump logic
438 } if labelRefCountChange(%1 -1)
444 ; peephole 7 redirected jump-to-jump at %5 by jump to %6
445 } if labelIsUncondJump(), labelRefCountChange(%5 -1), labelRefCountChange(%6 +1)
451 ; peephole 8 redirected jump-to-jump at %5 by jump to %6
452 } if labelIsUncondJump(), labelRefCountChange(%5 -1), labelRefCountChange(%6 +1)
459 ; peephole 10 removed redundant load of 0 into a.
467 ; peephole 11 combined constant loads into register pair.
475 ; peephole 12 combined constant loads into register pair.
483 ; peephole 13 combined constant loads into register pair.
491 ; peephole 14 removed redundant load from %1 into a.
493 // This gives many false negatives and without the test no problems are encountered in the regression tests
494 // Maybe we can try this after 2.7.0 release
501 ; peephole 15 removed redundant load from a into %1.
503 // This gives many false negatives and without the test no problems are encountered in the regression tests
504 // Maybe we can try this after 2.7.0 release
513 ; peephole 17 removed load by reordering or arguments.
523 ; peephole 18 used value still in a instead of loading it from %1.
531 ; peephole 19 removed redundant or after or.
539 ; peephole 19a removed redundant or after or.
547 ; peephole 20 removed redundant or after and.
555 ; peephole 20a removed redundant or after and.
563 ; peephole 21 removed redundant or after xor.
571 ; peephole 21a removed redundant or after xor.
580 ; peephole 23 load value in a first and use it next
581 } if notVolatile(%1 %2)
591 ; peephole 24 removed redundant load from %3%1 into %4%2
592 } if notVolatile(%1 %2 %3 %4)
606 ; peephole 31 moved and pushed arguments c and #%2 through bc instead of pushing them individually.
622 ; peephole 32 moved and pushed arguments e and #%2 through de instead of pushing them individually.
635 ; peephole 33 removed load by replacing sub with cp
647 ; peephole 34 removed load by replacing sub with cp
661 ; peephole 35 replaced sub a,#0xFF by inc a.
670 ; peephole 36 replaced sub a,#0xFF by inc a.
687 ; peephole 37 directly used (%1 + %2) in calculation instead of placing it in bc first.
705 ; peephole 38 directly used (%1 + %2) in calculation instead of placing it in de first.
717 ; peephole 39 replaced zero load, rla by and since rlca writes the same value to carry bit and least significant bit.
727 ; peephole 40 moved %2 directly into de instead of going through %1.
737 ; peephole 41 removed lower part of multibyte addition.
748 ; peephole 42 removed loads by exploiting commutativity of addition.
750 } if notVolatile(%1), notUsed(%1), operandsNotRelated(%2 '(bc)' '(de)')
758 ; peephole 42a removed loads by exploiting commutativity of addition.
761 // Don't need to check for volatile, since ix is used to access the stack.
763 // sdcc does not use the H flag. sla resets it, while add sets it.
764 // To ensure that the state of the H flag is not changed by this
765 // peephole uncomment the add %3, %4 at the end (since it overwrite the H flag).
773 ; peephole 42b shifts in accumulator insted of %1
785 ; peephole 43 removed load by exploiting commutativity of addition.
787 } if operandsNotRelated(%2 '(bc)' '(de)')
795 ; peephole 43a removed loads by exploiting commutativity of addition.
806 ; peephole 43b removed loads by exploiting commutativity of addition.
817 ; peephole 43c removed loads by exploiting commutativity of addition.
830 ; peephole 44 removed redundant zeroing of a (which has just been tested to be #0x00).
842 ; peephole 45 replaced constant #0x00 by a (which has just been tested to be #0x00).
853 ; peephole 46 replaced constant #0x00 by a (which has just been tested to be #0x00).
864 ; peephole 47 replaced constant #0x00 by a (which has just been tested to be #0x00).
875 ; peephole 48 replaced constant #0x00 by a (which has just been tested to be #0x00).
887 // ; peephole 49 shortened or using a (which has just been tested to be #0x00).
888 //} if operandsNotSame3(%3 '(bc)' '(de)')
899 // ; peephole 50 shortened or using a (which has just been tested to be #0x00).
900 //} if operandsNotSame3(%3 '(bc)' '(de)')
911 ; peephole 51 shortened or using a (which has just been tested to be #0x00).
913 //if operandsNotSame3(%3 '(bc)' '(de)')
924 ; peephole 52 shortened or using a (which has just been tested to be #0x00).
926 //if operandsNotSame3(%2 '(bc)' '(de)')
938 ; peephole 53 replaced constant #0x00 by a (which has just been tested to be #0x00).
951 ; peephole 54 replaced constant #0x00 by a (which has just been tested to be #0x00).
964 ; peephole 55 replaced constant #0x00 by a (which has just been tested to be #0x00).
977 ; peephole 56 replaced constant #0x00 by a (which has just been tested to be #0x00).
985 ; peephole 59 moved increment of de to constant.
993 ; peephole 62 moved increment of bc to constant.
1006 ; peephole 63 moved addition of constant 0x%3%2 to bc to constant.
1014 ; peephole 65 replaced call at end of function by jump.
1025 ; peephole 66 replaced call at end of function by jump moving call beyond pop ix.
1033 ; peephole 67 loaded %2, %3 from a instead of %1.
1037 } if notVolatile(%1)
1046 ; peephole 68 loaded %3 from a instead of %1.
1048 } if notVolatile(%1)
1057 ; peephole 69 used l in or instead of %1.
1059 } if notVolatile(%1)
1068 ; peephole 70 removed load of #%2 into %1 since it's still there.
1069 } if notVolatile(%1), operandsNotRelated(%3 %1)
1075 ; peephole 70a used #%1 from hl for load into de.
1089 ; peephole 71 used ex to get de into hl.
1096 // Lets' hope this is never applied to a jump table. It's fine for calls.
1098 // Should be after 0z.
1107 ; peephole 72 used hl instead of %2 (ix), %1 (ix) to load %4%3.
1110 } if operandsNotRelated('h' %3)
1111 // Don't check for volatile since ix points to the stack.
1113 // These ex-generating rules should be among the last ones since ex counts as a read from both hl and de for notUsed().
1118 ; peephole 73 used ex to load hl into de.
1126 ; peephole 74 used ex to load hl into de.
1134 ; peephole 75 used ex to load de into hl.
1138 // peeph-z80.def - Z80 peephole rules
1141 // (c) Philipp Klaus Krause (pkk@spth.de, philipp@colecovision.eu) 2006 - 2007
1143 // This program is free software; you can redistribute it and/or modify it
1144 // under the terms of the GNU General Public License as published by the
1145 // Free Software Foundation; either version 2, or (at your option) any
1148 // This program is distributed in the hope that it will be useful,
1149 // but WITHOUT ANY WARRANTY; without even the implied warranty of
1150 // MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
1151 // GNU General Public License for more details.
1153 // You should have received a copy of the GNU General Public License
1154 // along with this program; if not, write to the Free Software
1155 // Foundation, 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
1164 ; peephole z1 used #%2 directly instead of going through %1 using indirect addressing.
1176 ; peephole z2 directly used #%2 instead of going through %1 using indirect addressing.
1189 ; peephole z3 used #%2 directly instead of going through %1 using indirect addressing.
1192 } if operandsNotRelated(%3 'a')
1201 ; peephole z4 directly used address #%1 + %2 instead of placing it in bc first.
1210 ld (#%1 + %2 + %6),a
1211 ; peephole z5 directly used address #%1 + %2 + %6 instead of placing it in bc first.
1220 ; peephole z8 moved %1 directly into l instead of going through c.
1232 ; peephole z9 moved hl directly to the stack instead of going through bc.
1244 ; peephole z10 moved hl directly to the stack instead of going through de.
1252 ; peephole z11 replaced jump by return.
1253 } if labelIsReturnOnly(), labelRefCountChange(%5 -1)
1259 ; peephole z11a replaced jump by return.
1260 } if labelIsReturnOnly(), labelRefCountChange(%5 -1)
1262 // Should be one of the last ones. Opens the code to further peephole optimization.
1266 ; peephole z12 removed unused label %1.
1267 } if labelRefCount(%1 0)
1269 // Applying z11 or z11a followed by z12 will often leave a dead ret at the end of the function. Remove it.
1275 ; peephole z13 removed unused ret.
1278 // These four need two be here since the peephole optimizer continues to apply rules further down the file even for replace restart rules.
1285 ; peephole 5' removed jp by using inverse jump logic
1287 } if labelRefCountChange(%1 -1)
1295 ; peephole 5' removed jp by using inverse jump logic
1297 } if labelRefCountChange(%1 -1)
1305 ; peephole 5' removed jp by using inverse jump logic
1307 } if labelRefCountChange(%1 -1)
1315 ; peephole 6' removed jp by using inverse jump logic
1317 } if labelRefCountChange(%1 -1)
1319 // These should be the last rules, so that the peepholes above need to look at jp only.
1324 ; peephole z14 changed absolute to relative unconditional jump.
1331 ; peephole z15 changed absolute to relative conditional jump.
1338 ; peephole z16 changed absolute to relative conditional jump.
1345 ; peephole z17 changed absolute to relative conditional jump.
1352 ; peephole z18 changed absolute to relative conditional jump.