mirror of
https://github.com/irmen/prog8.git
synced 2024-11-11 18:06:12 +00:00
92 lines
6.9 KiB
ReStructuredText
92 lines
6.9 KiB
ReStructuredText
TODO
|
|
====
|
|
|
|
For next release
|
|
^^^^^^^^^^^^^^^^
|
|
- x16: add new keyboard APIs https://github.com/commanderx16/x16-docs/blob/master/Commander%20X16%20Programmer%27s%20Reference%20Guide.md#keyboard
|
|
- x16: optimize diskio load_raw because headerless files are now supported https://github.com/commanderx16/x16-rom/pull/216
|
|
note: must still work on c64/c128 that don't have this!
|
|
|
|
- vm codegen: When
|
|
- vm codegen: Pipe expression
|
|
- vm codegen: validate that PtFunctionCall translation works okay with resultregister, and multiple paramsters in correct order
|
|
- vm: support no globals re-init option
|
|
- vm: how to remove all unused subroutines? (for asm, 64tass used to do this)
|
|
- vm: rather than being able to jump to any 'address' (IPTR), use 'blocks'
|
|
- vm codegen/assembler: variable memory locations should also be referenced by the variable name instead of just the address
|
|
- when the vm is stable and *if* its language can get promoted to prog8 IL, the variable allocation should be changed.
|
|
It's now done before the vm code generation, but the IL should probably not depend on the allocations already performed.
|
|
So the CodeGen doesn't do VariableAlloc *before* the codegen, but as a last step.
|
|
|
|
...
|
|
|
|
|
|
Need help with
|
|
^^^^^^^^^^^^^^
|
|
- c128 target: various machine specific things (free zp locations, how banking works, getting the floating point routines working, ...)
|
|
- atari target: more details details about the machine, fixing library routines. I have no clue whatsoever.
|
|
- see the :ref:`portingguide` for details on what information is needed.
|
|
|
|
|
|
Future Things and Ideas
|
|
^^^^^^^^^^^^^^^^^^^^^^^
|
|
Compiler:
|
|
|
|
- vm code gen: don't reuse registers and don't pre allocate variables? (except strings + arrays) but instead put them into registers too
|
|
then we ALMOST have Static Single Assignment form in the VM code. But what can we use that for?
|
|
- pipe operator: allow non-unary function calls in the pipe that specify the other argument(s) in the calls.
|
|
- writeAssembly(): make it possible to actually get rid of the VarDecl nodes by fixing the rest of the code mentioned there.
|
|
- make everything an expression? (get rid of Statements. Statements are expressions with void return types?).
|
|
- allow "xxx" * constexpr (where constexpr is not a number literal), now gives expression error not same type
|
|
- make it possible to inline non-asmsub routines that just contain a single statement (return, functioncall, assignment)
|
|
but this requires all identifiers in the inlined expression to be changed to fully scoped names.
|
|
If we can do that why not perhaps also able to inline multi-line subroutines? Why would it be limited to just 1 line? Maybe to protect against code bloat.
|
|
Inlined subroutines cannot contain further nested subroutines!
|
|
- simplifyConditionalExpression() should not split expression if it still results in stack-based evaluation, but how does it know?
|
|
- simplifyConditionalExpression() sometimes introduces needless assignment to r9 tempvar (scenario sought)
|
|
- consider adding McCarthy evaluation to shortcircuit and and or expressions. First do ifs by splitting them up? Then do expressions that compute a value?
|
|
- use more of Result<> to handle errors/ nulls better?
|
|
- make it possible to use cpu opcodes such as 'nop' as variable names by prefixing all asm vars with something such as ``p8v_``? Or not worth it (most 3 letter opcodes as variables are nonsensical anyway)
|
|
then we can get rid of the instruction lists in the machinedefinitions as well?
|
|
- [problematic due to 64tass:] add a compiler option to not remove unused subroutines. this allows for building library programs. But this won't work with 64tass's .proc ...
|
|
Perhaps replace all uses of .proc/.pend by .block/.bend will fix that?
|
|
(but we lose the optimizing aspect of the assembler where it strips out unused code.
|
|
There's not really a dynamic switch possible as all assembly lib code is static and uses one or the other)
|
|
- Zig-like try-based error handling where the V flag could indicate error condition? and/or BRK to jump into monitor on failure? (has to set BRK vector for this)
|
|
- add special (u)word array type (or modifier?) that puts the array into memory as 2 separate byte-arrays 1 for LSB 1 for MSB -> allows for word arrays of length 256
|
|
|
|
Libraries:
|
|
|
|
- fix the problems in c128 target, and flesh out its libraries.
|
|
- fix the problems in atari target, and flesh out its libraries.
|
|
- c64: make the graphics.BITMAP_ADDRESS configurable (VIC banking)
|
|
- optimize several inner loops in gfx2 even further?
|
|
- add modes 2 and 3 to gfx2 (lowres 4 color and 16 color)?
|
|
- add a flood fill routine to gfx2?
|
|
- add a diskio.f_seek() routine for the Cx16 that uses its seek dos api? (only if that's stable)
|
|
|
|
Expressions:
|
|
|
|
- rethink the whole "isAugmentable" business. Because the way this is determined, should always also be exactly mirrorred in the AugmentableAssignmentAsmGen or you'll get a crash at code gen time.
|
|
- can we get rid of pieces of asmgen.AssignmentAsmGen by just reusing the AugmentableAssignment ? generated code should not suffer
|
|
- rewrite expression tree evaluation such that it doesn't use an eval stack but flatten the tree into linear code that uses a fixed number of predetermined value 'variables'?
|
|
"Three address code" was mentioned. https://en.wikipedia.org/wiki/Three-address_code
|
|
these variables have to be unique for each subroutine because they could otherwise be interfered with from irq routines etc.
|
|
- this removes the need for the BinExprSplitter? (which is problematic and very limited now)
|
|
and perhaps as well the assignment splitting in BeforeAsmAstChanger too
|
|
|
|
Optimizations:
|
|
|
|
- various optimizers should/do skip stuff if compTarget.name==VMTarget.NAME. Once (if?) 6502-codegen is no longer done from
|
|
the old CompilerAst, those checks should probably be removed.
|
|
- VariableAllocator: can we think of a smarter strategy for allocating variables into zeropage, rather than first-come-first-served
|
|
- translateUnaryFunctioncall() in BuiltinFunctionsAsmGen: should be able to assign parameters to a builtin function directly from register(s), this will make the use of a builtin function in a pipe expression more efficient without using a temporary variable
|
|
compare ``aa = startvalue(1) |> sin8u() |> cos8u() |> sin8u() |> cos8u()``
|
|
versus: ``aa = cos8u(sin8u(cos8u(sin8u(startvalue(1)))))`` the second one contains no sta cx16.r9L in between.
|
|
- AssignmentAsmGen.assignExpression() -> better code gen for assigning boolean comparison expressions
|
|
- when a for loop's loopvariable isn't referenced in the body, and the iterations are known, replace the loop by a repeatloop
|
|
but we have no efficient way right now to see if the body references a variable.
|
|
- AssignmentAsmGen: "real optimized code for comparison expressions that yield a boolean result value"
|
|
- automatically convert if statements that test for multiple values (if X==1 or X==2..) to if X in [1,2,..] statements, instead of just a warning.
|
|
- introduce byte-index operator to avoid index multiplications in loops over arrays? see github issue #4
|