gpu: fix 2 years old regression in blending of backdrop
This commit is contained in:
parent
4244eb69c6
commit
307e0182af
|
@ -2039,19 +2039,13 @@ static void GPU_RenderLine_layer(NDS_Screen * screen, u16 l)
|
|||
//this is currently eating up 2fps or so. it is a reasonable candidate for optimization.
|
||||
gpu->currBgNum = 5;
|
||||
switch(gpu->setFinalColorBck_funcNum) {
|
||||
case 0: case 1: //for backdrops, (even with window enabled) none and blend are both the same: just copy the color
|
||||
//for backdrops, effects arent applied.
|
||||
case 0: case 1:
|
||||
case 2: case 3:
|
||||
memset_u16_le<256>(gpu->currDst,backdrop_color);
|
||||
break;
|
||||
case 2:
|
||||
//for non-windowed fade, we can just fade the color and fill
|
||||
memset_u16_le<256>(gpu->currDst,gpu->currentFadeInColors[backdrop_color]);
|
||||
break;
|
||||
case 3:
|
||||
//likewise for non-windowed fadeout
|
||||
memset_u16_le<256>(gpu->currDst,gpu->currentFadeOutColors[backdrop_color]);
|
||||
break;
|
||||
|
||||
//windowed fades need special treatment
|
||||
//windowed cases apparently need special treatment? why? can we not render the backdrop? how would that even work?
|
||||
case 4: for(int x=0;x<256;x++) gpu->___setFinalColorBck<false,true,4>(backdrop_color,x,1); break;
|
||||
case 5: for(int x=0;x<256;x++) gpu->___setFinalColorBck<false,true,5>(backdrop_color,x,1); break;
|
||||
case 6: for(int x=0;x<256;x++) gpu->___setFinalColorBck<false,true,6>(backdrop_color,x,1); break;
|
||||
|
|
Loading…
Reference in New Issue