Skip to content

GitLab

Commit abd2cb44 authored by StapleButter's avatar StapleButter
Browse files

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

parent 676e5b32
Hide whitespace changes
Inline Side-by-side
GPU3D.cpp
View file @ abd2cb44
......@@ -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
......
GPU3D.h
View file @ abd2cb44
......@@ -29,6 +29,14 @@ typedef struct
bool Clipped;
// final vertex attributes.
// allows them to be reused in polygon strips.
s32 FinalPosition[4];
s32 FinalColor[3];
bool ViewportTransformDone;
} Vertex;
typedef struct
......
GPU3D_Soft.cpp
View file @ abd2cb44
......@@ -81,68 +81,74 @@ void RenderPolygon(Polygon* polygon)
int vtop = 0, vbot = 0;
s32 ytop = 191, ybot = 0;
s32 scrcoords[10][4];
// process the vertices, transform to screen coordinates
// find the topmost and bottommost vertices of the polygon
for (int i = 0; i < nverts; i++)
{
Vertex* vtx = polygon->Vertices[i];
s32 posX, posY, posZ, posW;
s32 w = vtx->Position[3];
if (w == 0)
if (!vtx->ViewportTransformDone)
{
posX = 0;
posY = 0;
posZ = 0;
posW = 0x1000;
}
else
{
// TODO: find a way to avoid doing 3 divisions :/
posX = ((s64)vtx->Position[0] << 12) / w;
posY = ((s64)vtx->Position[1] << 12) / w;
posZ = ((s64)vtx->Position[2] << 12) / w;
posW = w;
s32 posX, posY, posZ, posW;
s32 w = vtx->Position[3];
if (w == 0)
{
posX = 0;
posY = 0;
posZ = 0;
posW = 0x1000;
}
else
{
posX = ((s64)vtx->Position[0] << 12) / w;
posY = ((s64)vtx->Position[1] << 12) / w;
// TODO: W-buffering
posZ = (((s64)vtx->Position[2] * 0x800000) / w) + 0x7FFCFF;
posW = w;
}
s32 scrX = (((posX + 0x1000) * Viewport[2]) >> 13) + Viewport[0];
s32 scrY = (((posY + 0x1000) * Viewport[3]) >> 13) + Viewport[1];
if (scrX < 0) scrX = 0;
else if (scrX > 255) scrX = 255;
if (scrY < 0) scrY = 0;
else if (scrY > 191) scrY = 191;
if (posZ < 0) posZ = 0;
else if (posZ > 0xFFFFFF) posZ = 0xFFFFFF;
vtx->FinalPosition[0] = scrX;
vtx->FinalPosition[1] = 191 - scrY;
vtx->FinalPosition[2] = posZ;
vtx->FinalPosition[3] = posW;
vtx->FinalColor[0] = vtx->Color[0] ? ((vtx->Color[0] << 4) + 0xF) : 0;
vtx->FinalColor[1] = vtx->Color[1] ? ((vtx->Color[1] << 4) + 0xF) : 0;
vtx->FinalColor[2] = vtx->Color[2] ? ((vtx->Color[2] << 4) + 0xF) : 0;
vtx->ViewportTransformDone = true;
}
s32 scrX = (((posX + 0x1000) * Viewport[2]) >> 13) + Viewport[0];
s32 scrY = (((posY + 0x1000) * Viewport[3]) >> 13) + Viewport[1];
s32 scrZ = (((s64)(posZ + 0x1000) * 0xFFFFFF) >> 13);
if (scrX > 255) scrX = 255;
if (scrY > 191) scrY = 191;
if (scrZ > 0xFFFFFF) scrZ = 0xFFFFFF;
if (scrX < 0) { printf("!! bad X %d\n", scrX); scrX = 0;}
if (scrY < 0) { printf("!! bad Y %d\n", scrY); scrY = 0;}
if (scrZ < 0) { printf("!! bad Z %d %d\n", scrZ, vtx->Position[2]); scrZ = 0;}
scrcoords[i][0] = scrX;
scrcoords[i][1] = 191 - scrY;
scrcoords[i][2] = scrZ;
scrcoords[i][3] = posW;
if (scrcoords[i][1] < ytop)
if (vtx->FinalPosition[1] < ytop)
{
ytop = scrcoords[i][1];
ytop = vtx->FinalPosition[1];
vtop = i;
}
if (scrcoords[i][1] > ybot)
if (vtx->FinalPosition[1] > ybot)
{
ybot = scrcoords[i][1];
ybot = vtx->FinalPosition[1];
vbot = i;
}
//if (vtx->Color[0]==63 && vtx->Color[1]==0 && vtx->Color[2]==0)
//printf("v%d: %d,%d Z=%f W=%f %d %d\n", i, scrX, 191-scrY, vtx->Position[2]/4096.0f, vtx->Position[3]/4096.0f,
// polygon->FacingView, vtx->Clipped);
}
// draw, line per line
int lcur = vtop, rcur = vtop;
int lnext, rnext;
s32 lstep, rstep;
//s32 xmin, xmax;
if (polygon->FacingView)
{
......@@ -159,19 +165,11 @@ void RenderPolygon(Polygon* polygon)
if (rnext >= nverts) rnext = 0;
}
/*if ((scrcoords[lnext][1] - scrcoords[lcur][1]) == 0) lstep = 0; else
lstep = ((scrcoords[lnext][0] - scrcoords[lcur][0]) << 12) / (scrcoords[lnext][1] - scrcoords[lcur][1]);
if ((scrcoords[rnext][1] - scrcoords[rcur][1]) == 0) rstep = 0; else
rstep = ((scrcoords[rnext][0] - scrcoords[rcur][0]) << 12) / (scrcoords[rnext][1] - scrcoords[rcur][1]);*/
//xmin = scrcoords[lcur][0] << 12;
//xmax = scrcoords[rcur][0] << 12;
for (s32 y = ytop; y <= ybot; y++)
{
if (y < ybot)
{
while (y == scrcoords[lnext][1])
while (y == polygon->Vertices[lnext]->FinalPosition[1])
{
lcur = lnext;
......@@ -186,12 +184,10 @@ void RenderPolygon(Polygon* polygon)
if (lnext < 0) lnext = nverts - 1;
}
//lstep = ((scrcoords[lnext][0] - scrcoords[lcur][0]) << 12) / (scrcoords[lnext][1] - scrcoords[lcur][1]);
//xmin = scrcoords[lcur][0] << 12;
if (lcur == vbot) break;
}
while (y == scrcoords[rnext][1])
while (y == polygon->Vertices[rnext]->FinalPosition[1])
{
rcur = rnext;
......@@ -206,8 +202,6 @@ void RenderPolygon(Polygon* polygon)
if (rnext >= nverts) rnext = 0;
}
//rstep = ((scrcoords[rnext][0] - scrcoords[rcur][0]) << 12) / (scrcoords[rnext][1] - scrcoords[rcur][1]);
//xmax = scrcoords[rcur][0] << 12;
if (rcur == vbot) break;
}
}
......@@ -219,39 +213,54 @@ void RenderPolygon(Polygon* polygon)
s32 lfactor, rfactor;
if (scrcoords[lnext][1] == scrcoords[lcur][1])
if (vlnext->FinalPosition[1] == vlcur->FinalPosition[1])
lfactor = 0;
else
lfactor = ((y - scrcoords[lcur][1]) << 12) / (scrcoords[lnext][1] - scrcoords[lcur][1]);
lfactor = ((y - vlcur->FinalPosition[1]) << 12) / (vlnext->FinalPosition[1] - vlcur->FinalPosition[1]);
if (scrcoords[rnext][1] == scrcoords[rcur][1])
if (vrnext->FinalPosition[1] == vrcur->FinalPosition[1])
rfactor = 0;
else
rfactor = ((y - scrcoords[rcur][1]) << 12) / (scrcoords[rnext][1] - scrcoords[rcur][1]);
rfactor = ((y - vrcur->FinalPosition[1]) << 12) / (vrnext->FinalPosition[1] - vrcur->FinalPosition[1]);
s32 xl = scrcoords[lcur][0] + (((scrcoords[lnext][0] - scrcoords[lcur][0]) * lfactor) >> 12);
s32 xr = scrcoords[rcur][0] + (((scrcoords[rnext][0] - scrcoords[rcur][0]) * rfactor) >> 12);
s32 xl = vlcur->FinalPosition[0] + (((vlnext->FinalPosition[0] - vlcur->FinalPosition[0]) * lfactor) >> 12);
s32 xr = vrcur->FinalPosition[0] + (((vrnext->FinalPosition[0] - vrcur->FinalPosition[0]) * rfactor) >> 12);
if (xl<0 || xr>255) continue; // hax
if (xl<0 || xr>255)
{
printf("!! BAD X %d %d\n", xl, xr);
continue; // hax
}
s32 zl = scrcoords[lcur][2] + (((s64)(scrcoords[lnext][2] - scrcoords[lcur][2]) * lfactor) >> 12);
s32 zr = scrcoords[rcur][2] + (((s64)(scrcoords[rnext][2] - scrcoords[rcur][2]) * rfactor) >> 12);
s32 zl = vlcur->FinalPosition[2] + (((s64)(vlnext->FinalPosition[2] -vlcur->FinalPosition[2]) * lfactor) >> 12);
s32 zr = vrcur->FinalPosition[2] + (((s64)(vrnext->FinalPosition[2] - vrcur->FinalPosition[2]) * rfactor) >> 12);
s32 wl = scrcoords[lcur][3] + (((s64)(scrcoords[lnext][3] - scrcoords[lcur][3]) * lfactor) >> 12);
s32 wr = scrcoords[rcur][3] + (((s64)(scrcoords[rnext][3] - scrcoords[rcur][3]) * rfactor) >> 12);
s32 wl = vlcur->FinalPosition[3] + (((s64)(vlnext->FinalPosition[3] - vlcur->FinalPosition[3]) * lfactor) >> 12);
s32 wr = vrcur->FinalPosition[3] + (((s64)(vrnext->FinalPosition[3] - vrcur->FinalPosition[3]) * rfactor) >> 12);
s64 perspfactorl1 = ((s64)(0x1000 - lfactor) << 12) / scrcoords[lcur][3];
s64 perspfactorl2 = ((s64)lfactor << 12) / scrcoords[lnext][3];
s64 perspfactorr1 = ((s64)(0x1000 - rfactor) << 12) / scrcoords[rcur][3];
s64 perspfactorr2 = ((s64)rfactor << 12) / scrcoords[rnext][3];
s64 perspfactorl1 = ((s64)(0x1000 - lfactor) << 12) / vlcur->FinalPosition[3];
s64 perspfactorl2 = ((s64)lfactor << 12) / vlnext->FinalPosition[3];
s64 perspfactorr1 = ((s64)(0x1000 - rfactor) << 12) / vrcur->FinalPosition[3];
s64 perspfactorr2 = ((s64)rfactor << 12) / vrnext->FinalPosition[3];
u32 rl = (((perspfactorl1 * vlcur->Color[0]) + (perspfactorl2 * vlnext->Color[0])) << 12) / (perspfactorl1 + perspfactorl2);
u32 gl = (((perspfactorl1 * vlcur->Color[1]) + (perspfactorl2 * vlnext->Color[1])) << 12) / (perspfactorl1 + perspfactorl2);
u32 bl = (((perspfactorl1 * vlcur->Color[2]) + (perspfactorl2 * vlnext->Color[2])) << 12) / (perspfactorl1 + perspfactorl2);
if (perspfactorl1 + perspfactorl2 == 0)
{
perspfactorl1 = 0x1000;
perspfactorl2 = 0;
}
if (perspfactorr1 + perspfactorr2 == 0)
{
perspfactorr1 = 0x1000;
perspfactorr2 = 0;
}
s32 rl = ((perspfactorl1 * vlcur->FinalColor[0]) + (perspfactorl2 * vlnext->FinalColor[0])) / (perspfactorl1 + perspfactorl2);
s32 gl = ((perspfactorl1 * vlcur->FinalColor[1]) + (perspfactorl2 * vlnext->FinalColor[1])) / (perspfactorl1 + perspfactorl2);
s32 bl = ((perspfactorl1 * vlcur->FinalColor[2]) + (perspfactorl2 * vlnext->FinalColor[2])) / (perspfactorl1 + perspfactorl2);
u32 rr = (((perspfactorr1 * vrcur->Color[0]) + (perspfactorr2 * vrnext->Color[0])) << 12) / (perspfactorr1 + perspfactorr2);
u32 gr = (((perspfactorr1 * vrcur->Color[1]) + (perspfactorr2 * vrnext->Color[1])) << 12) / (perspfactorr1 + perspfactorr2);
u32 br = (((perspfactorr1 * vrcur->Color[2]) + (perspfactorr2 * vrnext->Color[2])) << 12) / (perspfactorr1 + perspfactorr2);
s32 rr = ((perspfactorr1 * vrcur->FinalColor[0]) + (perspfactorr2 * vrnext->FinalColor[0])) / (perspfactorr1 + perspfactorr2);
s32 gr = ((perspfactorr1 * vrcur->FinalColor[1]) + (perspfactorr2 * vrnext->FinalColor[1])) / (perspfactorr1 + perspfactorr2);
s32 br = ((perspfactorr1 * vrcur->FinalColor[2]) + (perspfactorr2 * vrnext->FinalColor[2])) / (perspfactorr1 + perspfactorr2);
s32 xdiv;
if (xr == xl)
......@@ -264,16 +273,25 @@ void RenderPolygon(Polygon* polygon)
s32 xfactor = (x - xl) * xdiv;
s32 z = zl + (((s64)(zr - zl) * xfactor) >> 12);
//z = wl + (((s64)(wr - wl) * xfactor) >> 12);
//z -= 0x1FF;
//if (z < 0) z = 0;
s32 perspfactor1 = ((0x1000 - xfactor) << 12) / wl;
s32 perspfactor2 = (xfactor << 12) / wr;
if (perspfactor1 + perspfactor2 == 0)
{
perspfactor1 = 0x1000;
perspfactor2 = 0;
}
// possible optimization: only do color interpolation if the depth test passes
u32 vr = (s64)((perspfactor1 * rl) + (perspfactor2 * rr)) / (perspfactor1 + perspfactor2);
u32 vg = (s64)((perspfactor1 * gl) + (perspfactor2 * gr)) / (perspfactor1 + perspfactor2);
u32 vb = (s64)((perspfactor1 * bl) + (perspfactor2 * br)) / (perspfactor1 + perspfactor2);
u32 vr = ((perspfactor1 * rl) + (perspfactor2 * rr)) / (perspfactor1 + perspfactor2);
u32 vg = ((perspfactor1 * gl) + (perspfactor2 * gr)) / (perspfactor1 + perspfactor2);
u32 vb = ((perspfactor1 * bl) + (perspfactor2 * br)) / (perspfactor1 + perspfactor2);
RenderPixel(polygon->Attr, x, y, z, vr>>12, vg>>12, vb>>12);
RenderPixel(polygon->Attr, x, y, z, vr>>3, vg>>3, vb>>3);
// Z debug
/*u8 zerp = (w * 63) / 0xFFFFFF;
......
melonDS.depend
View file @ abd2cb44
# depslib dependency file v1.0
1487173011 source:c:\documents\sources\melonds\main.cpp
1487292827 source:c:\documents\sources\melonds\main.cpp
<stdio.h>
<windows.h>
"NDS.h"
......@@ -10,7 +10,7 @@
1481161027 c:\documents\sources\melonds\types.h
1487177244 source:c:\documents\sources\melonds\nds.cpp
1487299879 source:c:\documents\sources\melonds\nds.cpp
<stdio.h>
<string.h>
"NDS.h"
......@@ -146,16 +146,16 @@
"NDS.h"
"NDSCart.h"
1487016725 c:\documents\sources\melonds\gpu3d.h
1487287868 c:\documents\sources\melonds\gpu3d.h
1487036430 source:c:\documents\sources\melonds\gpu3d.cpp
1487299939 source:c:\documents\sources\melonds\gpu3d.cpp
<stdio.h>
<string.h>
"NDS.h"
"GPU.h"
"FIFO.h"
1487177182 source:c:\documents\sources\melonds\gpu3d_soft.cpp
1487300098 source:c:\documents\sources\melonds\gpu3d_soft.cpp
<stdio.h>
<string.h>
"NDS.h"
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment