Apple-2-SW/prog8
1
0
Fork 0
mirror of https://github.com/irmen/prog8.git synced 2025-02-16 22:30:46 +00:00
Code Issues Projects Releases Wiki Activity

optimized pointer access @(pointer - constantoffset)

Browse Source
This commit is contained in:
Irmen de Jong 2025-01-05 06:07:33 +01:00
parent 0ffebc25d0
commit 3b3616afda
3 changed files with 52 additions and 64 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

46
codeGenCpu6502/src/prog8/codegen/cpu6502/AsmGen.kt
View File

@ -1322,34 +1322,19 @@ $repeatLabel""")
}
if(constOffset!=null) {
println("MEMWRITE POINTER - $constOffset ${addressExpr.position}") // TODO
/*
val pointervar = ptrAndIndex.first as? PtIdentifier
if(pointervar!=null && isZpVar(pointervar)) {
val saveA = evalBytevalueWillClobberA(ptrAndIndex.second)
if(saveA) out(" pha")
assignExpressionToRegister(ptrAndIndex.second, RegisterOrPair.Y)
if(saveA) out(" pla")
out(" sta (${asmSymbolName(pointervar)}),y")
val varname = asmSymbolName(pointervar)
out(" ldy #${256-constOffset} ; negative offset $constOffset")
out(" dec $varname+1 | sta ($varname),y | inc $varname+1") // temporarily make MSB 1 less
return true
} else {
// copy the pointer var to zp first
val saveA = evalBytevalueWillClobberA(ptrAndIndex.first) || evalBytevalueWillClobberA(ptrAndIndex.second)
if(saveA) out(" pha")
if(ptrAndIndex.second.isSimple()) {
assignExpressionToVariable(ptrAndIndex.first, "P8ZP_SCRATCH_W2", DataType.forDt(BaseDataType.UWORD))
assignExpressionToRegister(ptrAndIndex.second, RegisterOrPair.Y)
if(saveA) out(" pla")
out(" sta (P8ZP_SCRATCH_W2),y")
} else {
pushCpuStack(BaseDataType.UBYTE, ptrAndIndex.second)
assignExpressionToVariable(ptrAndIndex.first, "P8ZP_SCRATCH_W2", DataType.forDt(BaseDataType.UWORD))
restoreRegisterStack(CpuRegister.Y, true)
if(saveA) out(" pla")
out(" sta (P8ZP_SCRATCH_W2),y")
}
assignExpressionToVariable(ptrAndIndex.first, "P8ZP_SCRATCH_W2", DataType.forDt(BaseDataType.UWORD))
out(" ldy #${256-constOffset} ; negative offset $constOffset")
out(" dec P8ZP_SCRATCH_W2+1 | sta (P8ZP_SCRATCH_W2),y | inc P8ZP_SCRATCH_W2+1") // temporarily make MSB 1 less
return true
}
return true
*/
}
} else {
@ -1367,8 +1352,19 @@ $repeatLabel""")
}
if(constOffset!=null) {
println("MEMREAD POINTER - $constOffset ${addressExpr.position}") // TODO
// TODO optimize more cases
val pointervar = ptrAndIndex.first as? PtIdentifier
if(pointervar!=null && isZpVar(pointervar)) {
val varname = asmSymbolName(pointervar)
out(" ldy #${256-constOffset} ; negative offset $constOffset")
out(" dec $varname+1 | lda ($varname),y | inc $varname+1") // temporarily make MSB 1 less
return true
} else {
// copy the pointer var to zp first
assignExpressionToVariable(ptrAndIndex.first, "P8ZP_SCRATCH_W2", DataType.forDt(BaseDataType.UWORD))
out(" ldy #${256-constOffset} ; negative offset $constOffset")
out(" dec P8ZP_SCRATCH_W2+1 | lda (P8ZP_SCRATCH_W2),y | inc P8ZP_SCRATCH_W2+1") // temporarily make MSB 1 less
return true
}
}
}
}

10
docs/source/todo.rst
View File

@ -1,7 +1,7 @@
TODO
====
- optimize @(cell_ptr-offset) to use DEC pointer_msb ; LDY #255 ; INC pointer_msb instead. See tryOptimizedPointerAccessWithA()
- min/max(x1,x2) lots of code in a temp word, can sometimes just use the existing variables x1 and x2
- add paypal donation button as well?
@ -21,7 +21,7 @@ Future Things and Ideas
Add a -library $xxxx command line option (and/or some directive) to preselect every setting that is required to make a library at $xxxx rather than a normal loadable and runnable program?
Need to add some way to generate a stable jump table at a given address.
Need library to not call init_system AND init_system_phase2 not either.
Library must not include prog8_program_start stuff either.
Library must not include prog8_program_start stuff either. Must not require 'start' entrypoint either? Although they need some initialization entry point?
- [problematic due to using 64tass:] better support for building library programs, where unused .proc are NOT deleted from the assembly.
Perhaps replace all uses of .proc/.pend/.endproc by .block/.bend will fix that with a compiler flag?
But all library code written in asm uses .proc already..... (textual search/replace when writing the actual asm?)
@ -86,15 +86,15 @@ Optimizations
- if magicwall_enabled and (jiffy_counter & 3 == 1) sounds.magicwall() -> generates shortcut jump to another jump, why not immediately after the if
- explode(x, y+1) pushes x on the stack and pops it, could simply load it in reverse order and not use the stack.normal
- return mkword(attrs[cx16.r2L], object[cx16.r2L]) same as the explode() above
- x = y + z more efficient if rewritten to x=y; x+=z ?
- return peekw(table+64+pos*2) .... or rather .. return <complex expression> -> can this be optimized by using a temporary variable and chop up the expression?
likewise cx16.r0 = (gfx_lores.WIDTH-bmx.width)/2 + (gfx_lores.HEIGHT-bmx.height)/2*gfx_lores.WIDTH a lot of register juggling
- is there a trick to make @(pointer-1) = v more efficient? (like @(pointer+1)=v using Y indexed)
- if sv=="aa" else if sv=="bb" else if sv=="cc" -> needs much code, allow when(stringvar) too to avoid reloading both strings for every case (rockrunner bdcff.p8)
- if cx16.r0L>=97 and cx16.r0L<=122 {...} -> treats the boolean condition as a byte 0/1 result , can't it somehow just act on the carry bit alone?
same with if x1!=x2 or y1!=y2..... but it's because of the way boolean expressions are handled... can this be optimized?
- min(x1,x2) lots of code in a temp word, can just use the existing variables x1 and x2
- this generates empty lines in the resulting asm, why?:
cx16.r0L = cx16.r1L
cx16.r0L += cx16.r2L
- Optimize the IfExpression code generation to be more like regular if-else code. (both 6502 and IR) search for "TODO don't store condition as expression"
- VariableAllocator: can we think of a smarter strategy for allocating variables into zeropage, rather than first-come-first-served?

60
examples/test.p8
View File

@ -4,43 +4,35 @@
main {
sub start() {
ubyte[] array = [11,22,33,44,55,66,77,88,99]
uword @shared ptr = &array[5]
ubyte @shared offset
ubyte @shared bb0, bb1, bb2, bb3, bb4
uword @shared ww0, ww1, ww2, ww3, ww4
bb0 = min(bb1, 100)
bb1 = min(bb1, 100)
bb0 = min(100, bb1)
bb1 = min(100, bb1)
bb0 = min(bb1, bb2)
bb2 = min(bb1+bb2, bb3+bb4)
cx16.r0L = @(&start + 1)
cx16.r1L = @(&start - 1)
@(&start+1) = 99
@(&start-1) = 99
bb0 = max(bb1, 100)
bb1 = max(bb1, 100)
bb0 = max(100, bb1)
bb1 = max(100, bb1)
bb0 = max(bb1, bb2)
bb2 = max(bb1+bb2, bb3+bb4)
; @(ptr+1) = cx16.r0L
; @(ptr+2) = cx16.r0L
; @(ptr+offset) = cx16.r0L
; @(ptr-1) = cx16.r0L
; @(ptr-2) = cx16.r0L
; @(ptr-offset) = cx16.r0L
; cx16.r0L = @(ptr+1)
; cx16.r1L = @(ptr+2)
; cx16.r2L = @(ptr+offset)
; cx16.r3L = @(ptr-1)
; cx16.r4L = @(ptr-2)
; cx16.r5L = @(ptr-offset)
; %asm {{
; dec p8v_ptr+1
; ldy #255
; lda (p8v_ptr),y
; inc p8v_ptr+1
; sta cx16.r0L
; }}
repeat {
}
ww0 = min(ww1, 100)
ww1 = min(ww1, 100)
ww0 = min(100, ww1)
ww1 = min(100, ww1)
ww0 = min(ww1, ww2)
ww2 = min(ww1+ww2, ww3+ww4)
ww0 = max(ww1, 100)
ww1 = max(ww1, 100)
ww0 = max(100, ww1)
ww1 = max(100, ww1)
ww0 = max(ww1, ww2)
ww2 = max(ww1+ww2, ww3+ww4)
}
}