piwalker/melonDS
1
0
Fork 0
mirror of https://github.com/melonDS-emu/melonDS.git synced 2025-06-28 01:49:42 -06:00
Code Issues Packages Projects Releases Wiki Activity

improve 3D renderer precision. not perfect, but not bad at all.

Browse Source
This commit is contained in:
StapleButter
2017-02-17 04:07:00 +01:00
parent 676e5b32cc
commit abd2cb444b
4 changed files with 133 additions and 103 deletions
Download Patch File Download Diff File Expand all files Collapse all files

40
GPU3D.cpp
View File

@ -34,13 +34,24 @@
// clipping rules:
// * if a shared vertex in a strip is clipped, affected polygons are converted into single polygons
// strip is resumed at the first eligible polygon
//
// clipping exhibits oddities on the real thing. bad precision? fancy algorithm? TODO: investigate.
//
// vertex color precision:
// * vertex colors are kept at 5-bit during clipping. makes for shitty results.
// * vertex colors are converted to 9-bit before drawing, as such:
// if (x > 0) x = (x << 4) + 0xF
// the added bias affects interpolation.
//
// depth buffer:
// Z-buffering mode: val = ((Z * 0x800 * 0x1000) / W) + 0x7FFCFF
// W-buffering mode: val = W - 0x1FF
// TODO: confirm W, because it's weird
namespace GPU3D
{
#define COPYVERTEX(a, b) { *(u64*)&a[0] = *(u64*)&b[0]; *(u64*)&a[2] = *(u64*)&b[2]; }
const u32 CmdNumParams[256] =
{
// 0x00
@ -301,14 +312,6 @@ void MatrixMult4x3(s32* m, s32* s)
s32 tmp[16];
memcpy(tmp, m, 16*4);
/*printf("4x3 matrix\n");
for (int j = 0; j < 12; j += 3)
{
for (int i = 0; i < 3; i++)
printf("%f ", s[i]/4096.0f);
printf("\n");
}*/
// m = s*m
m[0] = ((s64)s[0]*tmp[0] + (s64)s[1]*tmp[4] + (s64)s[2]*tmp[8]) >> 12;
m[1] = ((s64)s[0]*tmp[1] + (s64)s[1]*tmp[5] + (s64)s[2]*tmp[9]) >> 12;
@ -392,11 +395,11 @@ void UpdateClipMatrix()
template<int comp, s32 plane>
void ClipSegment(Vertex* outbuf, Vertex* vout, Vertex* vin)
{
s64 factor = ((s64)(vin->Position[3] - (plane*vin->Position[comp])) << 24) /
((vin->Position[3] - (plane*vin->Position[comp])) - (vout->Position[3] - (plane*vout->Position[comp])));
s64 factor_num = vin->Position[3] - (plane*vin->Position[comp]);
s32 factor_den = factor_num - (vout->Position[3] - (plane*vout->Position[comp]));
Vertex mid;
#define INTERPOLATE(var) mid.var = vin->var + (((vout->var - vin->var) * factor) >> 24);
#define INTERPOLATE(var) mid.var = vin->var + (((vout->var - vin->var) * factor_num) / factor_den);
INTERPOLATE(Position[0]);
INTERPOLATE(Position[1]);
@ -408,6 +411,7 @@ void ClipSegment(Vertex* outbuf, Vertex* vout, Vertex* vin)
INTERPOLATE(Color[2]);
mid.Clipped = true;
mid.ViewportTransformDone = false;
#undef INTERPOLATE
*outbuf = mid;
@ -425,8 +429,6 @@ void SubmitPolygon()
int c;
// culling
//if (!(TempVertexBuffer[0].Color[0]==0 && TempVertexBuffer[0].Color[1]==63 && TempVertexBuffer[0].Color[2]==63))
// return;
// checkme: does it work this way for quads and up?
/*s32 _x1 = TempVertexBuffer[1].Position[0] - TempVertexBuffer[0].Position[0];
@ -770,6 +772,7 @@ void SubmitVertex()
vertextrans->Color[2] = VertexColor[2];
vertextrans->Clipped = false;
vertextrans->ViewportTransformDone = false;
VertexNum++;
VertexNumInPoly++;
@ -1179,9 +1182,9 @@ void ExecuteCommand()
u32 r = c & 0x1F;
u32 g = (c >> 5) & 0x1F;
u32 b = (c >> 10) & 0x1F;
VertexColor[0] = r ? (r<<1)+1 : 0;
VertexColor[1] = g ? (g<<1)+1 : 0;
VertexColor[2] = b ? (b<<1)+1 : 0;
VertexColor[0] = r;
VertexColor[1] = g;
VertexColor[2] = b;
}
break;
@ -1243,6 +1246,7 @@ void ExecuteCommand()
case 0x50:
FlushRequest = 1;//0x80000000 | (ExecParams[0] & 0x3);
CycleCount = 392;
break;
case 0x60: // viewport x1,y1,x2,y2