Hey guys!
Is there any gain if I've got many bitblt invokes and I move an address to a register?
Say, this is one:
inv BitBlt, hBufferDC, 0, 0, CHAR_LINE_COUNT * CHAR_WIDTH, CHAR_HEIGHT, \
assets.terrain.hdc, trn.frmx[edi], eax, SRCCOPY
Now I use esi to hold assets.terrain.hdc:
push esi
mov esi, assets.terrain.hdc
inv BitBlt, hBufferDC, 0, 0, CHAR_LINE_COUNT * CHAR_WIDTH, CHAR_HEIGHT, \
esi, trn.frmx[edi], eax, SRCCOPY
pop esi
Just consider there is more than one call to bitblt
Cheers
Alex,
With an API call, the total overhead is so high that loading a register does not give you a speed advantage. Use a memory operand as they don't change and get passed to the system DLL as memory in any case. Always try and find a way to use less registers in ordinary code so you have them when you need them.
Quote from: hutch-- on December 11, 2020, 08:28:13 AM
Alex,
With an API call, the total overhead is so high that loading a register does not give you a speed advantage. Use a memory operand as they don't change and get passed to the system DLL as memory in any case. Always try and find a way to use less registers in ordinary code so you have them when you need them.
Cheers Hutch!
Quote from: LordAdef on December 11, 2020, 08:03:09 AMIs there any gain if I've got many bitblt invokes and I move an address to a register?
Speedwise, as Hutch rightly notes, it will not make any difference. Sizewise, if assets.terrain.hdc is a global variable, then you need two invoke xxx, ..., esi to gain one byte less executable size.
include \masm32\MasmBasic\MasmBasic.inc
.data?
rc RECT <>
Init
Cls
repeats=1
rcglobal_s:
REPEAT repeats
invoke GetWindowRect, rv(GetConsoleWindow), addr rc
ENDM
rcglobal_endp:
deb 4, "Rect", rc.left, rc.top, rc.right, rc.bottom
rcreg_s:
push esi
mov esi, offset rc
REPEAT repeats
invoke GetWindowRect, rv(GetConsoleWindow), esi
ENDM
pop esi
rcreg_endp:
deb 4, "Rect", rc.left, rc.top, rc.right, rc.bottom
CodeSize rcglobal
CodeSize rcreg
EndOfCoderepeats=1:
16 bytes for rcglobal
19 bytes for rcreg
repeats=2:
32 bytes for rcglobal
31 bytes for rcreg
repeats=3:
48 bytes for rcglobal
43 bytes for rcreg
... etc ...
now I got curious about related question,which is best?
when calling a PROC that has loop with GDI invokes,calling from WM_PAINT when to structure up lots drawing code,I tried use two ways of make draw PROC's
1:hdc's and other handles are global variables?
2:hdc's are WM_PAINT's local variables and with invoke draw,hdc,handle2,handle3,handle4,its transferred to draw PROC?
would be interesting to time a loop with imageblt'ing on different cpu/gpu's,because its probably gpu depending ,would be good to know
Hi Daydreamer,
Good questions!