citra-emu
/
citra
Archived
1
0
Fork 0
Code

Rasterizer: Common sub-expression elimination 

Move the computation of some values out of loops so that they're not
constantly recalculated even when they don't change.
This commit is contained in:
Yuri Kunde Schlesner 2014-12-23 10:59:07 -02:00
parent 7e9bc85cc8
commit 2012e1420f
1 changed files with 17 additions and 14 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

31
src/video_core/rasterizer.cpp
View File

@ -106,6 +106,14 @@ void ProcessTriangle(const VertexShader::OutputVertex& v0,
int bias1 = IsRightSideOrFlatBottomEdge(vtxpos[1].xy(), vtxpos[2].xy(), vtxpos[0].xy()) ? -1 : 0; int bias1 = IsRightSideOrFlatBottomEdge(vtxpos[1].xy(), vtxpos[2].xy(), vtxpos[0].xy()) ? -1 : 0;
int bias2 = IsRightSideOrFlatBottomEdge(vtxpos[2].xy(), vtxpos[0].xy(), vtxpos[1].xy()) ? -1 : 0; int bias2 = IsRightSideOrFlatBottomEdge(vtxpos[2].xy(), vtxpos[0].xy(), vtxpos[1].xy()) ? -1 : 0;
const Math::Vec3<float24> w_inverse = Math::MakeVec(
float24::FromFloat32(1.0f) / v0.pos.w,
float24::FromFloat32(1.0f) / v1.pos.w,
float24::FromFloat32(1.0f) / v2.pos.w);
auto textures = registers.GetTextures();
auto tev_stages = registers.GetTevStages();
// TODO: Not sure if looping through x first might be faster // TODO: Not sure if looping through x first might be faster
for (u16 y = min_y; y < max_y; y += 0x10) { for (u16 y = min_y; y < max_y; y += 0x10) {
for (u16 x = min_x; x < max_x; x += 0x10) { for (u16 x = min_x; x < max_x; x += 0x10) {
@ -129,6 +137,11 @@ void ProcessTriangle(const VertexShader::OutputVertex& v0,
if (w0 < 0 || w1 < 0 || w2 < 0) if (w0 < 0 || w1 < 0 || w2 < 0)
continue; continue;
auto baricentric_coordinates = Math::MakeVec(float24::FromFloat32(static_cast<float>(w0)),
float24::FromFloat32(static_cast<float>(w1)),
float24::FromFloat32(static_cast<float>(w2)));
float24 interpolated_w_inverse = float24::FromFloat32(1.0f) / Math::Dot(w_inverse, baricentric_coordinates);
// Perspective correct attribute interpolation: // Perspective correct attribute interpolation:
// Attribute values cannot be calculated by simple linear interpolation since // Attribute values cannot be calculated by simple linear interpolation since
// they are not linear in screen space. For example, when interpolating a // they are not linear in screen space. For example, when interpolating a
@ -145,19 +158,9 @@ void ProcessTriangle(const VertexShader::OutputVertex& v0,
// //
// The generalization to three vertices is straightforward in baricentric coordinates. // The generalization to three vertices is straightforward in baricentric coordinates.
auto GetInterpolatedAttribute = [&](float24 attr0, float24 attr1, float24 attr2) { auto GetInterpolatedAttribute = [&](float24 attr0, float24 attr1, float24 attr2) {
auto attr_over_w = Math::MakeVec(attr0 / v0.pos.w, auto attr_over_w = Math::MakeVec(attr0, attr1, attr2) * w_inverse;
attr1 / v1.pos.w,
attr2 / v2.pos.w);
auto w_inverse = Math::MakeVec(float24::FromFloat32(1.f) / v0.pos.w,
float24::FromFloat32(1.f) / v1.pos.w,
float24::FromFloat32(1.f) / v2.pos.w);
auto baricentric_coordinates = Math::MakeVec(float24::FromFloat32(static_cast<float>(w0)),
float24::FromFloat32(static_cast<float>(w1)),
float24::FromFloat32(static_cast<float>(w2)));
float24 interpolated_attr_over_w = Math::Dot(attr_over_w, baricentric_coordinates); float24 interpolated_attr_over_w = Math::Dot(attr_over_w, baricentric_coordinates);
float24 interpolated_w_inverse = Math::Dot(w_inverse, baricentric_coordinates); return interpolated_attr_over_w * interpolated_w_inverse;
return interpolated_attr_over_w / interpolated_w_inverse;
}; };
Math::Vec4<u8> primary_color{ Math::Vec4<u8> primary_color{
@ -177,7 +180,7 @@ void ProcessTriangle(const VertexShader::OutputVertex& v0,
Math::Vec4<u8> texture_color[3]{}; Math::Vec4<u8> texture_color[3]{};
for (int i = 0; i < 3; ++i) { for (int i = 0; i < 3; ++i) {
auto texture = registers.GetTextures()[i]; const auto& texture = textures[i];
if (!texture.enabled) if (!texture.enabled)
continue; continue;
@ -219,7 +222,7 @@ void ProcessTriangle(const VertexShader::OutputVertex& v0,
// with some basic arithmetic. Alpha combiners can be configured separately but work // with some basic arithmetic. Alpha combiners can be configured separately but work
// analogously. // analogously.
Math::Vec4<u8> combiner_output; Math::Vec4<u8> combiner_output;
for (auto tev_stage : registers.GetTevStages()) { for (const auto& tev_stage : tev_stages) {
using Source = Regs::TevStageConfig::Source; using Source = Regs::TevStageConfig::Source;
using ColorModifier = Regs::TevStageConfig::ColorModifier; using ColorModifier = Regs::TevStageConfig::ColorModifier;
using AlphaModifier = Regs::TevStageConfig::AlphaModifier; using AlphaModifier = Regs::TevStageConfig::AlphaModifier;