Commit graph

33 commits

Author SHA1 Message Date
David Given 79a38ecc08 Instead of using parameterised rsts for stack access, add a huge swathe of
automatically built helper tools. Star Trek goes up from 40243 to 40779 bytes,
but should be a lot faster.
2019-02-13 22:45:22 +01:00
David Given f4b45f1ed7 Fix bad code generation due to not correctly flushing the stack before
comparisons.
2019-02-12 22:19:07 +01:00
David Given 360e1525ae Replace the non-working Z80 style signed comparison sign adjustment with
something which actually works on the 8080. Sadly, Star Trek is up to
40243 bytes again.
2019-02-12 11:57:37 +01:00
David Given f1b128baeb Signed comparisons now work correctly; sadly, Star Trek increases from 39656 to
39726 bytes. Note that the emulator doesn't set P/V correctly and all the tests
fail. Testing in z80ex works fine.
2019-02-11 20:46:06 +01:00
David Given fca54ccc1b Improve subtractions, using dad and inline constants where possible. Shrinks
Star Trek from 39740 to 39656 bytes.
2019-02-10 19:40:48 +01:00
David Given 7d34cd7407 Simplify simple comparisons and use inlie constants where possible. Reduces
Star Trek from 39771 to 39740 bytes.
2019-02-10 19:22:43 +01:00
David Given 35d1f36ca4 Use pop rather than inx;inx to retract the stack; consumes a register but
halves the number of calls. Reduces Star Trek from 39976 to 39771 bytes.
2019-02-10 18:57:20 +01:00
David Given b34ff2bdec Teach the i80 code generator how to do left shifts by using dad hl; prevents it
from calling out to .sli2 for simple shifts. Shrinks Star Trek from 40055 bytes
to 39976.
2019-02-10 18:44:47 +01:00
David Given 10b509d6b6 Add rst 3 to get the address of a frame variable; lots of rule overhaulage.
Reduces Star Trek from 41821 to 41055 bytes.
2019-02-09 13:29:41 +01:00
David Given 0de5001f82 Make sign extension much less horrible. 2019-02-08 23:16:57 +01:00
David Given 0ac145a0d1 Add tests for simple and, or, and xor; drastically improve the i80 code
generation for these.
2019-02-08 22:38:07 +01:00
David Given a91ab5c599 Somehow the implementation of rst 2 got lost, with hilarious results. Also fix
some register corruption issues (I hope).
2019-02-08 21:35:14 +01:00
David Given a5bf2932c8 Add rst 1 to store frame variables. Reduces Star Trek from 42339 to 41821
bytes.
2019-02-07 09:31:38 +01:00
David Given 9d8ea79c7f rst 1 now takes an eight-bit parameter, not a 16 bit one. Reduces Star Trek
again from 43205 to 42339 bytes.
2019-02-07 08:45:20 +01:00
David Given 5ac280026e Experimentally use rst 1 to optimise frame variable loads. Reduces Star Trek
from 43892 to 43205 bytes.
2019-02-07 00:06:32 +01:00
David Given 08b92153a0 Use a helper function for cai --- only saves two bytes per invocation, but is
much less embarassing.
2019-02-06 20:36:23 +01:00
David Given 02779325ab Move the standard ret boilerplate into a helper procedure, to save a little
space.
2019-02-06 08:08:11 +01:00
George Koehler 93e01eb5d1 Teach i80 ncg to use libfp. Enable ack -mcpm -fp.
Old .o files stop working if they use floating point.  One must
recompile those files.  Old files don't call libfp in the correct way,
and may use symbols that I removed from libem.  I don't keep old
symbols in libem/flp.s, because a program that pulls both libfp and
flp.s would get "multiply defined" errors in the linker.

I teach mach/i80/ncg/table to use libfp by copying or adapting the
patterns from mach/i86/ncg/table.  I did not test all the patterns,
but I did use `ack -mcpm -fp -O4` to compile examples/mandelbrot.c,
then I ran it in the emulator YAZE-AG.  It worked, but it was slow.
2018-04-25 16:09:56 -04:00
George Koehler 34cf0c8b63 Kill registers a, de, when i80 ncg does Call libem.
I compiled tests/plat/lib/test.c with ack -mcpm, but i80 ncg did emit
wrong code in writehex(uint32_t) for

    "0123456789abcdef"[code & 0xf]

The code called '.and' to evaluate `code & 0xf`, then tried to call
'.cii' to narrow the result from 4 to 2 bytes, but it passed garbage
instead of 4 to '.cii'.  The rule for '.and' was

    pat and defined($1)
    kills ALL
    uses dereg={const2,$1}
    gen Call {label,".and"}

This failed to kill register de={const2,4}, so ncg pushed de,
expecting to push 4, but actually pushing garbage.

Fix such rules using `mvi a,...` or `lxi de,...` so ncg doesn't track
the token in the register.  This is like the i86 table.  A different
fix would use a dummy instruction `killreg a` or `killreg de` like the
m68020 table.

Also correct 1 to $1 when calling '.exg'.
2017-12-06 22:14:00 -05:00
dtrg df088c184b Added optimised rules for int<<8 and uint>>8 (which can be done with simple register shuffling rather than calling .sli). 2007-04-29 23:00:31 +00:00
dtrg 440d6faadd Fixed a bug where loc lol stf was getting the constant pushed onto the stack *after* stf had popped it off. 2007-04-27 22:15:31 +00:00
ceriel 1db6a3029b Header --> Id 1994-06-24 14:02:31 +00:00
ceriel 6f03cff48f Some minor modifications reflecting some changes in the peephole optimizer 1992-09-01 09:39:15 +00:00
ceriel 564cbc6eba Only call NOP routine when debugging table 1991-09-04 10:55:52 +00:00
ceriel c998623932 Some more adaptions to new semantics of ncgg 1987-11-10 13:49:04 +00:00
ceriel 12d731dfcf adapted table to new semantics of STACK 1987-08-26 13:47:37 +00:00
em 94534b7c15 1- Floating point routines now call routines instead of calling trp. 1985-04-12 14:25:59 +00:00
garde 80e349860b Added RCS id. 1985-03-18 12:55:07 +00:00
garde d52117c8dd Some less important changes 1985-03-18 10:00:30 +00:00
garde 736a2d1022 Many things improved and extended 1985-03-04 13:42:38 +00:00
garde 8e154eb826 Many, many improvements made, ususually small ones.
For example: many em-patterns have got alternative code-rules.
1985-02-01 12:51:56 +00:00
garde a849b751a9 src1 and src2 extended.
Errors in 'pat loc cmi teq' and 'pat loc cmi tne' recovered.
Second stack-pattern for 'pat zlt' and 'pat zge' added.
Patterns for 'lil loc adi sil', 'lil inc sil', 'lil dec sil'
and 'sil lil' added.
1985-01-22 19:55:11 +00:00
garde d8d2068664 Initial revision 1985-01-17 14:31:34 +00:00