zdoom/gzdoom
0
0
Fork 0
mirror of https://github.com/ZDoom/gzdoom.git synced 2024-11-11 23:32:02 +00:00
Code Issues Projects Releases Packages Wiki Activity

Improve performance by 30% from using a better stencil block rejection test

Browse source
This commit is contained in:
Magnus Norddahl 2016-12-01 11:21:14 +01:00
parent e3b3b7a4ce
commit 06362385d6
1 changed files with 16 additions and 16 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

30
tools/drawergen/fixedfunction/drawtrianglecodegen.cpp
View file

@ -238,8 +238,22 @@ void DrawTriangleCodegen::LoopBlockX()
SSAInt c = (c00.zext_int() << 0) | (c10.zext_int() << 1) | (c01.zext_int() << 2) | (c11.zext_int() << 3); SSAInt c = (c00.zext_int() << 0) | (c10.zext_int() << 1) | (c01.zext_int() << 2) | (c11.zext_int() << 3);
// Skip block when outside an edge // Skip block when outside an edge
SSABool process_block = !(a == SSAInt(0) || b == SSAInt(0) || c == SSAInt(0));
SetStencilBlock(x / 8 + y / 8 * stencilPitch);
// Stencil test the whole block, if possible
if (variant == TriDrawVariant::DrawSubsector || variant == TriDrawVariant::FillSubsector || variant == TriDrawVariant::FuzzSubsector || variant == TriDrawVariant::StencilClose)
{
process_block = process_block && (!StencilIsSingleValue() || SSABool::compare_uge(StencilGetSingle(), stencilTestValue));
}
else
{
process_block = process_block && (!StencilIsSingleValue() || StencilGetSingle() == stencilTestValue);
}
SSAIfBlock branch; SSAIfBlock branch;
branch.if_block(!(a == SSAInt(0) || b == SSAInt(0) || c == SSAInt(0))); branch.if_block(process_block);
// Check if block needs clipping // Check if block needs clipping
SSABool clipneeded = (x + q) > clipright || (y + q) > clipbottom; SSABool clipneeded = (x + q) > clipright || (y + q) > clipbottom;
@ -261,8 +275,6 @@ void DrawTriangleCodegen::LoopBlockX()
currentlight = (!is_fixed_light).select(diminishedlight, light); currentlight = (!is_fixed_light).select(diminishedlight, light);
} }
SetStencilBlock(x / 8 + y / 8 * stencilPitch);
SSABool covered = a == SSAInt(0xF) && b == SSAInt(0xF) && c == SSAInt(0xF) && !clipneeded && StencilIsSingleValue(); SSABool covered = a == SSAInt(0xF) && b == SSAInt(0xF) && c == SSAInt(0xF) && !clipneeded && StencilIsSingleValue();
// Accept whole block when totally covered // Accept whole block when totally covered
@ -304,16 +316,6 @@ void DrawTriangleCodegen::SetupAffineBlock()
void DrawTriangleCodegen::LoopFullBlock() void DrawTriangleCodegen::LoopFullBlock()
{ {
SSAIfBlock branch_stenciltest;
if (variant == TriDrawVariant::DrawSubsector || variant == TriDrawVariant::FillSubsector || variant == TriDrawVariant::FuzzSubsector || variant == TriDrawVariant::StencilClose)
{
branch_stenciltest.if_block(SSABool::compare_uge(StencilGetSingle(), stencilTestValue));
}
else
{
branch_stenciltest.if_block(StencilGetSingle() == stencilTestValue);
}
if (variant == TriDrawVariant::Stencil) if (variant == TriDrawVariant::Stencil)
{ {
StencilClear(stencilWriteValue); StencilClear(stencilWriteValue);
@ -418,8 +420,6 @@ void DrawTriangleCodegen::LoopFullBlock()
AffineVaryingPosY[i] = AffineVaryingPosY[i] + gradVaryingY[i]; AffineVaryingPosY[i] = AffineVaryingPosY[i] + gradVaryingY[i];
} }
} }
branch_stenciltest.end_block();
} }
void DrawTriangleCodegen::LoopPartialBlock() void DrawTriangleCodegen::LoopPartialBlock()