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)
94 ; peephole 0g incremented in hl instead of bc.
98 // Catch double and triple incs before later peepholes introduce an ex de,hl in here.
107 ; peephole 0h' incremented in hl instead of de.
119 ; peephole 0h incremented in hl instead of de.
129 ; peephole 0i loaded %2%1 into hl directly instead of going through de.
139 ; peephole 0j used hl for #%2 + %3 instead of bc, not going through a.
142 } if notUsed('a'), notUsed('hl')
149 ; peephole 0j' used hl for #%2 + %3 instead of de, not going through a.
152 } if notUsed('a'), notUsed('hl')
158 ; peephole 0k pushed de directly instead of going through hl.
160 } if notUsed('de'), notUsed('hl')
166 ; peephole 0k' pushed hl directly instead of going through de.
168 } if notUsed('de'), notUsed('hl')
175 ; peephole 0k'' pushed de instead of hl removing a load.
178 } if notUsed('hl'), notUsed('e')
185 ; peephole 0l pushed hl directly instead of going through de.
188 } if notUsed('de'), notUsed('hl')
195 ; peephole 0m pushed bc directly instead of going through hl.
204 ; peephole 0m' pushed bc instead of hl removing a load.
207 } if notUsed('hl'), notUsed('c')
214 ; peephole 0n removed redundant load of a through c.
223 ; peephole 0o used de directly instead of going through bc.
232 ; peephole 0p popped hl directly instead of going through de.
240 ; peephole 0q loaded a into %2 (%3) directly instead of going through %1.
242 } if notVolatile(%1), notUsed(%1)
248 ; peephole 0r loaded %3 from %2 instead of going through %1 (ix).
252 // Don't need to check for volatile, since ix is used for the stack.
258 ; peephole 0s loaded b from %2 (%3) instead of going through %1.
260 } if notVolatile(%1), notUsed(%1)
267 ; peephole 0t loaded %4 from a instead of going through %1.
271 } if notVolatile(%1), operandsNotRelated(%1 %3), operandsNotRelated(%1 %2)
280 ; peephole 0t' loaded %4 from a instead of going through %1.
284 } if notVolatile(%1), operandsNotRelated(%1 %2), operandsNotRelated(%4 %2)
293 ; peephole 0t'' loaded %4 from a instead of going through %1.
303 ; peephole 0u loaded e from (hl) instead of going through %1.
311 ; peephole 0v loaded %2 (%3) from l instead of going through %1.
326 ; peephole 0w shifted (hl) in place.
330 // Don't check for volatile since ix points to the stack.
339 ; peephole 0x pushed de instead of pushing a twice.
342 } if notUsed('d'), notUsed('a')
351 ; peephole 0y pushed bc instead of pushing a twice.
354 } if notUsed('b'), notUsed('a')
360 ; peephole 0z loaded #%1 into d directly instead of going through a.
372 ; peephole 0za pushed %1 (ix), %2(ix) through hl instead of af.
376 } if notUsed('a'), notUsed('hl')
383 ; peephole 0zb pushed hl instead of bc.
387 // Doesn'T work due to bug #1947081
392 // ; peephole 0zc eleminated dead pop/push pair.
399 ; peephole 0zd used hl instead of iy.
402 } if notUsed('iy'), notUsed('hl')
408 ; peephole 0ze used hl instead of iy.
411 } if notUsed('iy'), notUsed('hl')
417 ; peephole 0zf used hl instead of iy.
420 } if notUsed('iy'), notUsed('hl'), operandsNotRelated(%4 'h'), operandsNotRelated(%4 'l')
427 ; peephole 0zg loaded %1 from (hl) directly instead of going through (de).
436 ; peephole 0zh loaded %1 from (hl) directly instead of going through (bc).
445 ; peephole 0zi incremented in hl instead of bc.
455 ; peephole 0zj tested bit of %1 (%2) directly instead of going through a.
464 ; peephole 0zk stored constant #%1 + %2 into hl directly instead of going through bc.
474 ; peephole 0zk' stored %1 (%2) %3 (%4) into hl directly instead of going through bc.
485 ; peephole 3 removed jp by using inverse jump logic
487 } if labelRefCountChange(%1 -1)
495 ; peephole 4 removed jp by using inverse jump logic
497 } if labelRefCountChange(%1 -1)
505 ; peephole 5 removed jp by using inverse jump logic
507 } if labelRefCountChange(%1 -1)
515 ; peephole 6 removed jp by using inverse jump logic
517 } if labelRefCountChange(%1 -1)
523 ; peephole 7 redirected jump-to-jump at %5 by jump to %6
524 } if labelIsUncondJump(), labelRefCountChange(%5 -1), labelRefCountChange(%6 +1)
530 ; peephole 8 redirected jump-to-jump at %5 by jump to %6
531 } if labelIsUncondJump(), labelRefCountChange(%5 -1), labelRefCountChange(%6 +1)
538 ; peephole 10 removed redundant load of 0 into a.
546 ; peephole 11 combined constant loads into register pair.
554 ; peephole 12 combined constant loads into register pair.
562 ; peephole 13 combined constant loads into register pair.
570 ; peephole 14 removed redundant load from %1 into a.
572 // This gives many false negatives and without the test no problems are encountered in the regression tests
573 // Maybe we can try this after 2.7.0 release
580 ; peephole 15 removed redundant load from a into %1.
582 // This gives many false negatives and without the test no problems are encountered in the regression tests
583 // Maybe we can try this after 2.7.0 release
592 ; peephole 17a removed load by reordering or arguments.
602 ; peephole 17b removed load by reordering or arguments.
612 ; peephole 18 used value still in a instead of loading it from %1.
620 ; peephole 19 removed redundant or after or.
628 ; peephole 19a removed redundant or after or.
636 ; peephole 20 removed redundant or after and.
644 ; peephole 20a removed redundant or after and.
652 ; peephole 21 removed redundant or after xor.
660 ; peephole 21a removed redundant or after xor.
669 ; peephole 23 load value in a first and use it next
670 } if notVolatile(%1 %2)
680 ; peephole 24 removed redundant load from %3%1 into %4%2
681 } if notVolatile(%1 %2 %3 %4)
695 ; peephole 31 moved and pushed arguments c and #%2 through bc instead of pushing them individually.
711 ; peephole 32 moved and pushed arguments e and #%2 through de instead of pushing them individually.
724 ; peephole 33 removed load by replacing sub with cp
736 ; peephole 34 removed load by replacing sub with cp
750 ; peephole 35 replaced sub a,#0xFF by inc a.
759 ; peephole 36 replaced sub a,#0xFF by inc a.
776 ; peephole 37 directly used (%1 + %2) in calculation instead of placing it in bc first.
794 ; peephole 38 directly used (%1 + %2) in calculation instead of placing it in de first.
806 ; peephole 39 replaced zero load, rla by and since rlca writes the same value to carry bit and least significant bit.
816 ; peephole 40 moved %2 directly into de instead of going through %1.
826 ; peephole 41 removed lower part of multibyte addition.
837 ; peephole 42 removed loads by exploiting commutativity of addition.
839 } if notVolatile(%1), notUsed(%1), operandsNotRelated(%2 '(bc)' '(de)')
847 ; peephole 42a removed loads by exploiting commutativity of addition.
850 // Don't need to check for volatile, since ix is used to access the stack.
852 // sdcc does not use the H flag. sla resets it, while add sets it.
853 // To ensure that the state of the H flag is not changed by this
854 // peephole uncomment the add %3, %4 at the end (since it overwrite the H flag).
862 ; peephole 42b shifted in accumulator insted of %1
868 // sdcc does not use the H flag. sla resets it, while add sets it.
869 // To ensure that the state of the H flag is not changed by this
870 // peephole uncomment the add %3, %4 at the end (since it overwrite the H flag).
880 ; peephole 42b' shifted in accumulator insted of %1
898 ; peephole 42c incremented in (hl) instead of going through a.
913 ; peephole 42d decremented in (hl) instead of going through a.
922 ; peephole 43 removed load by exploiting commutativity of addition.
924 } if operandsNotRelated(%2 '(bc)' '(de)')
932 ; peephole 43a removed loads by exploiting commutativity of addition.
943 ; peephole 43b removed loads by exploiting commutativity of addition.
954 ; peephole 43c removed loads by exploiting commutativity of addition.
967 ; peephole 44 removed redundant zeroing of a (which has just been tested to be #0x00).
979 ; peephole 45 replaced constant #0x00 by a (which has just been tested to be #0x00).
990 ; peephole 46 replaced constant #0x00 by a (which has just been tested to be #0x00).
1001 ; peephole 47 replaced constant #0x00 by a (which has just been tested to be #0x00).
1012 ; peephole 48 replaced constant #0x00 by a (which has just been tested to be #0x00).
1024 // ; peephole 49 shortened or using a (which has just been tested to be #0x00).
1025 //} if operandsNotSame3(%3 '(bc)' '(de)')
1036 // ; peephole 50 shortened or using a (which has just been tested to be #0x00).
1037 //} if operandsNotSame3(%3 '(bc)' '(de)')
1039 // Commented out sine it triggers for %3 od the form (#...)
1049 // ; peephole 51 shortened or using a (which has just been tested to be #0x00).
1051 //if operandsNotSame3(%3 '(bc)' '(de)')
1062 ; peephole 52 shortened or using a (which has just been tested to be #0x00).
1064 //if operandsNotSame3(%2 '(bc)' '(de)')
1076 ; peephole 53 replaced constant #0x00 by a (which has just been tested to be #0x00).
1089 ; peephole 54 replaced constant #0x00 by a (which has just been tested to be #0x00).
1102 ; peephole 55 replaced constant #0x00 by a (which has just been tested to be #0x00).
1115 ; peephole 56 replaced constant #0x00 by a (which has just been tested to be #0x00).
1123 ; peephole 59 moved increment of de to constant.
1131 ; peephole 62 moved increment of bc to constant.
1144 ; peephole 63 moved addition of constant 0x%3%2 to bc to constant.
1152 ; peephole 65 replaced call at end of function by jump.
1163 ; peephole 66 replaced call at end of function by jump moving call beyond pop ix.
1171 ; peephole 67 loaded %2, %3 from a instead of %1.
1175 } if notVolatile(%1)
1184 ; peephole 68 loaded %3 from a instead of %1.
1186 } if notVolatile(%1)
1195 ; peephole 69 used l in or instead of %1.
1197 } if notVolatile(%1)
1206 ; peephole 70 removed load of #%2 into %1 since it's still there.
1207 } if notVolatile(%1), operandsNotRelated(%3 %1)
1213 ; peephole 70a used #%1 from hl for load into de.
1227 ; peephole 71 used ex to get de into hl.
1234 // Lets' hope this is never applied to a jump table. It's fine for calls.
1236 // Should be after 0z.
1245 ; peephole 72 used hl instead of %2 (ix), %1 (ix) to load %4%3.
1248 } if operandsNotRelated('h' %3)
1249 // Don't check for volatile since ix points to the stack.
1260 ; peephole 76 loaded l from a directly instead of going through %1.
1272 ; peephole 77 loaded l from a directly instead of going through %1.
1284 ; peephole 78 removed addition and loads exploiting commutativity of addition.
1292 ; peephole 78a removed ex exploiting commutativity of addition.
1304 ; peephole 79 moved increment to constant.
1309 // These ex-generating rules should be among the last ones since ex counts as a read from both hl and de for notUsed().
1314 ; peephole 80 used ex to load hl into de.
1322 ; peephole 81 used ex to load hl into de.
1330 ; peephole 82 used ex to load de into hl.
1334 // peeph-z80.def - Z80 peephole rules
1337 // (c) Philipp Klaus Krause (pkk@spth.de, philipp@colecovision.eu) 2006 - 2007
1339 // This program is free software; you can redistribute it and/or modify it
1340 // under the terms of the GNU General Public License as published by the
1341 // Free Software Foundation; either version 2, or (at your option) any
1344 // This program is distributed in the hope that it will be useful,
1345 // but WITHOUT ANY WARRANTY; without even the implied warranty of
1346 // MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
1347 // GNU General Public License for more details.
1349 // You should have received a copy of the GNU General Public License
1350 // along with this program; if not, write to the Free Software
1351 // Foundation, 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
1360 ; peephole z1 used #%2 directly instead of going through %1 using indirect addressing.
1372 ; peephole z2 directly used #%2 instead of going through %1 using indirect addressing.
1385 ; peephole z3 used #%2 directly instead of going through %1 using indirect addressing.
1388 } if operandsNotRelated(%3 'a')
1397 ; peephole z4 directly used address #%1 + %2 instead of placing it in bc first.
1406 ld (#%1 + %2 + %6),a
1407 ; peephole z5 directly used address #%1 + %2 + %6 instead of placing it in bc first.
1416 ; peephole z8 moved %1 directly into l instead of going through c.
1428 ; peephole z9 moved hl directly to the stack instead of going through bc.
1440 ; peephole z10 moved hl directly to the stack instead of going through de.
1448 ; peephole z11 replaced jump by return.
1449 } if labelIsReturnOnly(), labelRefCountChange(%5 -1)
1455 ; peephole z11a replaced jump by return.
1456 } if labelIsReturnOnly(), labelRefCountChange(%5 -1)
1458 // Should be one of the last ones. Opens the code to further peephole optimization.
1462 ; peephole z12 removed unused label %1.
1463 } if labelRefCount(%1 0)
1465 // Applying z11 or z11a followed by z12 will often leave a dead ret at the end of the function. Remove it.
1471 ; peephole z13 removed unused ret.
1474 // These four need two be here since the peephole optimizer continues to apply rules further down the file even for replace restart rules.
1481 ; peephole 5' removed jp by using inverse jump logic
1483 } if labelRefCountChange(%1 -1)
1491 ; peephole 5' removed jp by using inverse jump logic
1493 } if labelRefCountChange(%1 -1)
1501 ; peephole 5' removed jp by using inverse jump logic
1503 } if labelRefCountChange(%1 -1)
1511 ; peephole 6' removed jp by using inverse jump logic
1513 } if labelRefCountChange(%1 -1)
1515 // These should be the last rules, so that the peepholes above need to look at jp only.
1520 ; peephole z14 changed absolute to relative unconditional jump.
1527 ; peephole z15 changed absolute to relative conditional jump.
1534 ; peephole z16 changed absolute to relative conditional jump.
1541 ; peephole z17 changed absolute to relative conditional jump.
1548 ; peephole z18 changed absolute to relative conditional jump.