Derive VertexRoutine from Function<>.
Bug 22652760
Change-Id: I48e9e1f3ff677429eff1aea2f80b1e384a537a14
Reviewed-on: https://swiftshader-review.googlesource.com/4557
Tested-by: Nicolas Capens <capn@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <capn@google.com>
diff --git a/src/Renderer/VertexProcessor.cpp b/src/Renderer/VertexProcessor.cpp
index 30e8cd5..5517820 100644
--- a/src/Renderer/VertexProcessor.cpp
+++ b/src/Renderer/VertexProcessor.cpp
@@ -73,7 +73,7 @@
P = 0;
PB = 0;
PBV = 0;
-
+
for(int i = 0; i < 12; i++)
{
PBVM[i] = 0;
@@ -271,7 +271,7 @@
{
if(light < 8)
{
- ff.attenuationConstant[light] = replicate(constant);
+ ff.attenuationConstant[light] = replicate(constant);
ff.attenuationLinear[light] = replicate(linear);
ff.attenuationQuadratic[light] = replicate(quadratic);
}
@@ -707,7 +707,7 @@
{
PB = P * B;
PBV = PB * V;
-
+
for(int i = 0; i < activeMatrices; i++)
{
PBVM[i] = PBV * M[i];
@@ -723,7 +723,7 @@
{
PB = P * B;
PBV = PB * V;
-
+
for(int i = 0; i < activeMatrices; i++)
{
PBVM[i] = PBV * M[i];
@@ -737,7 +737,7 @@
if(updateViewMatrix)
{
PBV = PB * V;
-
+
for(int i = 0; i < activeMatrices; i++)
{
PBVM[i] = PBV * M[i];
@@ -808,7 +808,7 @@
state.shaderContainsTexldl = context->vertexShader ? context->vertexShader->containsTexldl() : false;
state.positionRegister = context->vertexShader ? context->vertexShader->positionRegister : Pos;
state.pointSizeRegister = context->vertexShader ? context->vertexShader->pointSizeRegister : Pts;
-
+
state.vertexBlendMatrixCount = context->vertexBlendMatrixCountActive();
state.indexedVertexBlendEnable = context->indexedVertexBlendActive();
state.vertexNormalActive = context->vertexNormalActive();
@@ -888,7 +888,7 @@
{
state.output[D0].write = 0xF;
}
-
+
if(context->specularActive())
{
state.output[D1].write = 0xF;
@@ -968,7 +968,7 @@
}
generator->generate();
- routine = generator->getRoutine();
+ routine = (*generator)(L"VertexRoutine_%0.8X", state.shaderID);
delete generator;
routineCache->add(state, routine);
diff --git a/src/Shader/VertexRoutine.cpp b/src/Shader/VertexRoutine.cpp
index b875517..d41b2b7 100644
--- a/src/Shader/VertexRoutine.cpp
+++ b/src/Shader/VertexRoutine.cpp
@@ -23,9 +23,8 @@
extern bool halfIntegerCoordinates; // Pixel centers are not at integer coordinates
extern bool symmetricNormalizedDepth; // [-1, 1] instead of [0, 1]
- VertexRoutine::VertexRoutine(const VertexProcessor::State &state, const VertexShader *shader) : state(state), shader(shader)
+ VertexRoutine::VertexRoutine(const VertexProcessor::State &state, const VertexShader *shader) : r(shader), state(state), shader(shader)
{
- routine = 0;
}
VertexRoutine::~VertexRoutine()
@@ -34,67 +33,56 @@
void VertexRoutine::generate()
{
- Function<Void(Pointer<Byte>, Pointer<Byte>, Pointer<Byte>, Pointer<Byte>)> function;
+ Pointer<Byte> vertex(Arg<0>());
+ Pointer<Byte> batch(Arg<1>());
+ Pointer<Byte> task(Arg<2>());
+ Pointer<Byte> data(Arg<3>());
+
+ const bool texldl = state.shaderContainsTexldl;
+
+ Pointer<Byte> cache = task + OFFSET(VertexTask,vertexCache);
+ Pointer<Byte> vertexCache = cache + OFFSET(VertexCache,vertex);
+ Pointer<Byte> tagCache = cache + OFFSET(VertexCache,tag);
+
+ UInt vertexCount = *Pointer<UInt>(task + OFFSET(VertexTask,vertexCount));
+
+ r.data = data;
+ r.constants = *Pointer<Pointer<Byte> >(data + OFFSET(DrawData,constants));
+ if(shader && shader->instanceIdDeclared)
{
- Pointer<Byte> vertex(function.Arg<0>());
- Pointer<Byte> batch(function.Arg<1>());
- Pointer<Byte> task(function.Arg<2>());
- Pointer<Byte> data(function.Arg<3>());
-
- const bool texldl = state.shaderContainsTexldl;
-
- Pointer<Byte> cache = task + OFFSET(VertexTask,vertexCache);
- Pointer<Byte> vertexCache = cache + OFFSET(VertexCache,vertex);
- Pointer<Byte> tagCache = cache + OFFSET(VertexCache,tag);
-
- UInt vertexCount = *Pointer<UInt>(task + OFFSET(VertexTask,vertexCount));
-
- Registers r(shader);
- r.data = data;
- r.constants = *Pointer<Pointer<Byte> >(data + OFFSET(DrawData,constants));
- if(shader && shader->instanceIdDeclared)
- {
- r.instanceID = *Pointer<Int>(data + OFFSET(DrawData, instanceID));
- }
-
- Do
- {
- UInt index = *Pointer<UInt>(batch);
- UInt tagIndex = index & 0x0000003C;
- UInt indexQ = !texldl ? UInt(index & 0xFFFFFFFC) : index; // FIXME: TEXLDL hack to have independent LODs, hurts performance.
-
- If(*Pointer<UInt>(tagCache + tagIndex) != indexQ)
- {
- *Pointer<UInt>(tagCache + tagIndex) = indexQ;
-
- readInput(r, indexQ);
- pipeline(r);
- postTransform(r);
- computeClipFlags(r);
-
- Pointer<Byte> cacheLine0 = vertexCache + tagIndex * UInt((int)sizeof(Vertex));
- writeCache(cacheLine0, r);
- }
-
- UInt cacheIndex = index & 0x0000003F;
- Pointer<Byte> cacheLine = vertexCache + cacheIndex * UInt((int)sizeof(Vertex));
- writeVertex(vertex, cacheLine);
-
- vertex += sizeof(Vertex);
- batch += sizeof(unsigned int);
- vertexCount--;
- }
- Until(vertexCount == 0)
-
- Return();
+ r.instanceID = *Pointer<Int>(data + OFFSET(DrawData, instanceID));
}
- routine = function(L"VertexRoutine_%0.8X", state.shaderID);
- }
+ Do
+ {
+ UInt index = *Pointer<UInt>(batch);
+ UInt tagIndex = index & 0x0000003C;
+ UInt indexQ = !texldl ? UInt(index & 0xFFFFFFFC) : index; // FIXME: TEXLDL hack to have independent LODs, hurts performance.
- Routine *VertexRoutine::getRoutine()
- {
- return routine;
+ If(*Pointer<UInt>(tagCache + tagIndex) != indexQ)
+ {
+ *Pointer<UInt>(tagCache + tagIndex) = indexQ;
+
+ readInput(r, indexQ);
+ pipeline(r);
+ postTransform(r);
+ computeClipFlags(r);
+
+ Pointer<Byte> cacheLine0 = vertexCache + tagIndex * UInt((int)sizeof(Vertex));
+ writeCache(cacheLine0, r);
+ }
+
+ UInt cacheIndex = index & 0x0000003F;
+ Pointer<Byte> cacheLine = vertexCache + cacheIndex * UInt((int)sizeof(Vertex));
+ writeVertex(vertex, cacheLine);
+
+ vertex += sizeof(Vertex);
+ batch += sizeof(unsigned int);
+ vertexCount--;
+ }
+ Until(vertexCount == 0)
+
+ Return();
}
void VertexRoutine::readInput(Registers &r, UInt &index)
@@ -242,7 +230,7 @@
v.y = Float4(*Pointer<Short4>(source1));
v.z = Float4(*Pointer<Short4>(source2));
v.w = Float4(*Pointer<Short4>(source3));
-
+
transpose4xN(v.x, v.y, v.z, v.w, stream.count);
if(stream.normalized)
@@ -251,7 +239,7 @@
if(stream.count >= 2) v.y *= *Pointer<Float4>(r.constants + OFFSET(Constants,unscaleShort));
if(stream.count >= 3) v.z *= *Pointer<Float4>(r.constants + OFFSET(Constants,unscaleShort));
if(stream.count >= 4) v.w *= *Pointer<Float4>(r.constants + OFFSET(Constants,unscaleShort));
- }
+ }
}
break;
case STREAMTYPE_USHORT:
@@ -260,7 +248,7 @@
v.y = Float4(*Pointer<UShort4>(source1));
v.z = Float4(*Pointer<UShort4>(source2));
v.w = Float4(*Pointer<UShort4>(source3));
-
+
transpose4xN(v.x, v.y, v.z, v.w, stream.count);
if(stream.normalized)
@@ -277,7 +265,7 @@
// FIXME: Vectorize
{
Int x, y, z;
-
+
x = y = z = *Pointer<Int>(source0);
v.x.x = Float(x & 0x000003FF);
@@ -287,7 +275,7 @@
{
Int x, y, z;
-
+
x = y = z = *Pointer<Int>(source1);
v.y.x = Float(x & 0x000003FF);
@@ -297,7 +285,7 @@
{
Int x, y, z;
-
+
x = y = z = *Pointer<Int>(source2);
v.z.x = Float(x & 0x000003FF);
@@ -307,7 +295,7 @@
{
Int x, y, z;
-
+
x = y = z = *Pointer<Int>(source3);
v.w.x = Float(x & 0x000003FF);
@@ -326,7 +314,7 @@
// FIXME: Vectorize
{
Int x, y, z;
-
+
x = y = z = *Pointer<Int>(source0);
v.x.x = Float((x << 22) & 0xFFC00000);
@@ -336,7 +324,7 @@
{
Int x, y, z;
-
+
x = y = z = *Pointer<Int>(source1);
v.y.x = Float((x << 22) & 0xFFC00000);
@@ -346,7 +334,7 @@
{
Int x, y, z;
-
+
x = y = z = *Pointer<Int>(source2);
v.z.x = Float((x << 22) & 0xFFC00000);
@@ -356,7 +344,7 @@
{
Int x, y, z;
-
+
x = y = z = *Pointer<Int>(source3);
v.w.x = Float((x << 22) & 0xFFC00000);
diff --git a/src/Shader/VertexRoutine.hpp b/src/Shader/VertexRoutine.hpp
index 26663d7..55b1018 100644
--- a/src/Shader/VertexRoutine.hpp
+++ b/src/Shader/VertexRoutine.hpp
@@ -19,7 +19,7 @@
namespace sw
{
- class VertexRoutine
+ class VertexRoutine : public Function<Void(Pointer<Byte>, Pointer<Byte>, Pointer<Byte>, Pointer<Byte>)>
{
protected:
struct Registers
@@ -31,7 +31,7 @@
{
loopDepth = -1;
enableStack[0] = Int4(0xFFFFFFFF, 0xFFFFFFFF, 0xFFFFFFFF, 0xFFFFFFFF);
-
+
if(shader && shader->containsBreakInstruction())
{
enableBreak = Int4(0xFFFFFFFF, 0xFFFFFFFF, 0xFFFFFFFF, 0xFFFFFFFF);
@@ -71,31 +71,30 @@
Int instanceID;
};
+ Registers r;
+
public:
VertexRoutine(const VertexProcessor::State &state, const VertexShader *shader);
virtual ~VertexRoutine();
void generate();
- Routine *getRoutine();
protected:
const VertexProcessor::State &state;
const VertexShader *const shader;
- private:
+ private:
virtual void pipeline(Registers &r) = 0;
typedef VertexProcessor::State::Input Stream;
-
+
Vector4f readStream(Registers &r, Pointer<Byte> &buffer, UInt &stride, const Stream &stream, const UInt &index);
void readInput(Registers &r, UInt &index);
void computeClipFlags(Registers &r);
void postTransform(Registers &r);
void writeCache(Pointer<Byte> &cacheLine, Registers &r);
void writeVertex(Pointer<Byte> &vertex, Pointer<Byte> &cacheLine);
-
- Routine *routine;
};
}
