Pica: Cleanup clipping code and change screenspace z to range from -1..0.

The change in depth range seems to reflect better to what applications are expecting, and makes for cleaner code overall (hence is more likely to reflect hardware behavior).
This commit is contained in:
Tony Wasserka 2015-01-02 20:37:25 +01:00
parent 70a764d992
commit 365236fa4c
2 changed files with 42 additions and 53 deletions

View file

@ -15,30 +15,18 @@ namespace Clipper {
struct ClippingEdge {
public:
enum Type {
POS_X = 0,
NEG_X = 1,
POS_Y = 2,
NEG_Y = 3,
POS_Z = 4,
NEG_Z = 5,
};
ClippingEdge(Type type, float24 position) : type(type), pos(position) {}
ClippingEdge(Math::Vec4<float24> coeffs,
Math::Vec4<float24> bias = Math::Vec4<float24>(float24::FromFloat32(0),
float24::FromFloat32(0),
float24::FromFloat32(0),
float24::FromFloat32(0)))
: coeffs(coeffs),
bias(bias)
{
}
bool IsInside(const OutputVertex& vertex) const {
switch (type) {
case POS_X: return vertex.pos.x <= pos * vertex.pos.w;
case NEG_X: return vertex.pos.x >= pos * vertex.pos.w;
case POS_Y: return vertex.pos.y <= pos * vertex.pos.w;
case NEG_Y: return vertex.pos.y >= pos * vertex.pos.w;
// TODO: Check z compares ... should be 0..1 instead?
case POS_Z: return vertex.pos.z <= pos * vertex.pos.w;
default:
case NEG_Z: return vertex.pos.z >= pos * vertex.pos.w;
}
return Math::Dot(vertex.pos + bias, coeffs) <= float24::FromFloat32(0);
}
bool IsOutSide(const OutputVertex& vertex) const {
@ -46,31 +34,17 @@ public:
}
OutputVertex GetIntersection(const OutputVertex& v0, const OutputVertex& v1) const {
auto dotpr = [this](const OutputVertex& vtx) {
switch (type) {
case POS_X: return vtx.pos.x - vtx.pos.w;
case NEG_X: return -vtx.pos.x - vtx.pos.w;
case POS_Y: return vtx.pos.y - vtx.pos.w;
case NEG_Y: return -vtx.pos.y - vtx.pos.w;
// TODO: Verify z clipping
case POS_Z: return vtx.pos.z - vtx.pos.w;
default:
case NEG_Z: return -vtx.pos.w;
}
};
float24 dp = dotpr(v0);
float24 dp_prev = dotpr(v1);
float24 dp = Math::Dot(v0.pos + bias, coeffs);
float24 dp_prev = Math::Dot(v1.pos + bias, coeffs);
float24 factor = dp_prev / (dp_prev - dp);
return OutputVertex::Lerp(factor, v0, v1);
}
private:
Type type;
float24 pos;
Math::Vec4<float24> coeffs;
Math::Vec4<float24> bias;
};
static void InitScreenCoordinates(OutputVertex& vtx)
@ -98,10 +72,9 @@ static void InitScreenCoordinates(OutputVertex& vtx)
vtx.tc2 *= inv_w;
vtx.pos.w = inv_w;
// TODO: Not sure why the viewport width needs to be divided by 2 but the viewport height does not
vtx.screenpos[0] = (vtx.pos.x * inv_w + float24::FromFloat32(1.0)) * viewport.halfsize_x + viewport.offset_x;
vtx.screenpos[1] = (vtx.pos.y * inv_w + float24::FromFloat32(1.0)) * viewport.halfsize_y + viewport.offset_y;
vtx.screenpos[2] = viewport.offset_z - vtx.pos.z * inv_w * viewport.zscale;
vtx.screenpos[2] = viewport.offset_z + vtx.pos.z * inv_w * viewport.zscale;
}
void ProcessTriangle(OutputVertex &v0, OutputVertex &v1, OutputVertex &v2) {
@ -117,14 +90,29 @@ void ProcessTriangle(OutputVertex &v0, OutputVertex &v1, OutputVertex &v2) {
auto* output_list = &buffer_a;
auto* input_list = &buffer_b;
// NOTE: We clip against a w=epsilon plane to guarantee that the output has a positive w value.
// TODO: Not sure if this is a valid approach. Also should probably instead use the smallest
// epsilon possible within float24 accuracy.
static const float24 EPSILON = float24::FromFloat32(0.00001);
static const float24 f0 = float24::FromFloat32(0.0);
static const float24 f1 = float24::FromFloat32(1.0);
static const std::array<ClippingEdge, 7> clipping_edges = {{
{ Math::MakeVec( f1, f0, f0, -f1) }, // x = +w
{ Math::MakeVec(-f1, f0, f0, -f1) }, // x = -w
{ Math::MakeVec( f0, f1, f0, -f1) }, // y = +w
{ Math::MakeVec( f0, -f1, f0, -f1) }, // y = -w
{ Math::MakeVec( f0, f0, f1, f0) }, // z = 0
{ Math::MakeVec( f0, f0, -f1, -f1) }, // z = -w
{ Math::MakeVec( f0, f0, f0, -f1), Math::Vec4<float24>(f0, f0, f0, EPSILON) }, // w = EPSILON
}};
// TODO: If one vertex lies outside one of the depth clipping planes, some platforms (e.g. Wii)
// drop the whole primitive instead of clipping the primitive properly. We should test if
// this happens on the 3DS, too.
// Simple implementation of the Sutherland-Hodgman clipping algorithm.
// TODO: Make this less inefficient (currently lots of useless buffering overhead happens here)
for (auto edge : { ClippingEdge(ClippingEdge::POS_X, float24::FromFloat32(+1.0)),
ClippingEdge(ClippingEdge::NEG_X, float24::FromFloat32(-1.0)),
ClippingEdge(ClippingEdge::POS_Y, float24::FromFloat32(+1.0)),
ClippingEdge(ClippingEdge::NEG_Y, float24::FromFloat32(-1.0)),
ClippingEdge(ClippingEdge::POS_Z, float24::FromFloat32(+1.0)),
ClippingEdge(ClippingEdge::NEG_Z, float24::FromFloat32(-1.0)) }) {
for (auto edge : clipping_edges) {
std::swap(input_list, output_list);
output_list->clear();

View file

@ -106,16 +106,17 @@ void ProcessTriangle(const VertexShader::OutputVertex& v0,
ScreenToRasterizerCoordinates(v1.screenpos),
ScreenToRasterizerCoordinates(v2.screenpos) };
if (registers.cull_mode == Regs::CullMode::KeepClockWise) {
// Reverse vertex order and use the CCW code path.
if (registers.cull_mode == Regs::CullMode::KeepCounterClockWise) {
// Reverse vertex order and use the CW code path.
std::swap(vtxpos[1], vtxpos[2]);
}
if (registers.cull_mode != Regs::CullMode::KeepAll) {
// Cull away triangles which are wound clockwise.
// TODO: A check for degenerate triangles ("== 0") should be considered for CullMode::KeepAll
// Cull away triangles which are wound counter-clockwise.
if (SignedArea(vtxpos[0].xy(), vtxpos[1].xy(), vtxpos[2].xy()) <= 0)
return;
} else {
// TODO: Consider A check for degenerate triangles ("SignedArea == 0")
}
// TODO: Proper scissor rect test!
@ -475,7 +476,7 @@ void ProcessTriangle(const VertexShader::OutputVertex& v0,
// TODO: Does depth indeed only get written even if depth testing is enabled?
if (registers.output_merger.depth_test_enable) {
u16 z = (u16)(-(v0.screenpos[2].ToFloat32() * w0 +
u16 z = (u16)((v0.screenpos[2].ToFloat32() * w0 +
v1.screenpos[2].ToFloat32() * w1 +
v2.screenpos[2].ToFloat32() * w2) * 65535.f / wsum);
u16 ref_z = GetDepth(x >> 4, y >> 4);