VertexRoutine code for Transform Feedback
This cl adds the code that actually performs the copy of the
vertex shader outputs into transform feedback buffers. It
also contains a fix for symmetricNormalizedDepth, which must
be computed after the information was copied into the
transform feedback buffers, when transform feedback is active.
Change-Id: I418f94a15b9425bba0905c840f8cf4828233d0fb
Reviewed-on: https://swiftshader-review.googlesource.com/5172
Tested-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <capn@google.com>
diff --git a/src/Renderer/Renderer.cpp b/src/Renderer/Renderer.cpp
index 2cb19e3..8be8093 100644
--- a/src/Renderer/Renderer.cpp
+++ b/src/Renderer/Renderer.cpp
@@ -240,7 +240,7 @@
if(update || oldMultiSampleMask != context->multiSampleMask)
{
- vertexState = VertexProcessor::update();
+ vertexState = VertexProcessor::update(drawType);
setupState = SetupProcessor::update();
pixelState = PixelProcessor::update();
@@ -1460,12 +1460,8 @@
return;
}
- task->vertexStart = start * 3;
+ task->primitiveStart = start;
task->vertexCount = triangleCount * 3;
- // Note: Quads aren't handled for verticesPerPrimitive, but verticesPerPrimitive is used for transform feedback,
- // which is an OpenGL ES 3.0 feature, and OpenGL ES 3.0 doesn't support quads as a primitive type.
- DrawType type = static_cast<DrawType>(static_cast<unsigned int>(draw->drawType) & 0xF);
- task->verticesPerPrimitive = 1 + (type >= DRAW_LINELIST) + (type >= DRAW_TRIANGLELIST);
vertexRoutine(&triangle->v0, (unsigned int*)&batch, task, data);
}
diff --git a/src/Renderer/VertexProcessor.cpp b/src/Renderer/VertexProcessor.cpp
index 90c954f..73f277b 100644
--- a/src/Renderer/VertexProcessor.cpp
+++ b/src/Renderer/VertexProcessor.cpp
@@ -839,7 +839,7 @@
routineCache = new RoutineCache<State>(clamp(cacheSize, 1, 65536), precacheVertex ? "sw-vertex" : 0);
}
- const VertexProcessor::State VertexProcessor::update()
+ const VertexProcessor::State VertexProcessor::update(DrawType drawType)
{
if(isFixedFunction())
{
@@ -912,6 +912,11 @@
state.transformFeedbackQueryEnabled = context->transformFeedbackQueryEnabled;
state.transformFeedbackEnabled = context->transformFeedbackEnabled;
+ // Note: Quads aren't handled for verticesPerPrimitive, but verticesPerPrimitive is used for transform feedback,
+ // which is an OpenGL ES 3.0 feature, and OpenGL ES 3.0 doesn't support quads as a primitive type.
+ DrawType type = static_cast<DrawType>(static_cast<unsigned int>(drawType) & 0xF);
+ state.verticesPerPrimitive = 1 + (type >= DRAW_LINELIST) + (type >= DRAW_TRIANGLELIST);
+
for(int i = 0; i < MAX_VERTEX_INPUTS; i++)
{
state.input[i].type = context->input[i].type;
diff --git a/src/Renderer/VertexProcessor.hpp b/src/Renderer/VertexProcessor.hpp
index 9629c47..3af6690 100644
--- a/src/Renderer/VertexProcessor.hpp
+++ b/src/Renderer/VertexProcessor.hpp
@@ -35,9 +35,8 @@
struct VertexTask
{
- unsigned int vertexStart;
unsigned int vertexCount;
- unsigned int verticesPerPrimitive;
+ unsigned int primitiveStart;
VertexCache vertexCache;
};
@@ -76,6 +75,7 @@
bool pointScaleActive : 1;
bool transformFeedbackQueryEnabled : 1;
uint64_t transformFeedbackEnabled : 64;
+ unsigned char verticesPerPrimitive : 2; // 1 (points), 2 (lines) or 3 (triangles)
bool preTransformed : 1;
bool superSampling : 1;
@@ -275,7 +275,7 @@
const Matrix &getModelTransform(int i);
const Matrix &getViewTransform();
- const State update();
+ const State update(DrawType drawType);
Routine *routine(const State &state);
bool isFixedFunction();
diff --git a/src/Shader/VertexRoutine.cpp b/src/Shader/VertexRoutine.cpp
index dce0dc2..6428cdf 100644
--- a/src/Shader/VertexRoutine.cpp
+++ b/src/Shader/VertexRoutine.cpp
@@ -46,6 +46,8 @@
Pointer<Byte> tagCache = cache + OFFSET(VertexCache,tag);
UInt vertexCount = *Pointer<UInt>(task + OFFSET(VertexTask,vertexCount));
+ UInt primitiveNumber = *Pointer<UInt>(task + OFFSET(VertexTask, primitiveStart));
+ UInt indexInPrimitive = 0;
constants = *Pointer<Pointer<Byte>>(data + OFFSET(DrawData,constants));
@@ -70,11 +72,20 @@
UInt cacheIndex = index & 0x0000003F;
Pointer<Byte> cacheLine = vertexCache + cacheIndex * UInt((int)sizeof(Vertex));
- writeVertex(vertex, cacheLine);
+ writeVertex(vertex, cacheLine, primitiveNumber, indexInPrimitive);
vertex += sizeof(Vertex);
batch += sizeof(unsigned int);
vertexCount--;
+ if(state.transformFeedbackEnabled != 0)
+ {
+ indexInPrimitive++;
+ If(indexInPrimitive == 3)
+ {
+ primitiveNumber++;
+ indexInPrimitive = 0;
+ }
+ }
}
Until(vertexCount == 0)
@@ -96,13 +107,19 @@
{
int pos = state.positionRegister;
+ Float4 outPosZ = o[pos].z;
+ if(state.transformFeedbackEnabled && symmetricNormalizedDepth && !state.fixedFunction)
+ {
+ outPosZ = (outPosZ + o[pos].w) * Float4(0.5f);
+ }
+
Int4 maxX = CmpLT(o[pos].w, o[pos].x);
Int4 maxY = CmpLT(o[pos].w, o[pos].y);
- Int4 maxZ = CmpLT(o[pos].w, o[pos].z);
+ Int4 maxZ = CmpLT(o[pos].w, outPosZ);
Int4 minX = CmpNLE(-o[pos].w, o[pos].x);
Int4 minY = CmpNLE(-o[pos].w, o[pos].y);
- Int4 minZ = CmpNLE(Float4(0.0f), o[pos].z);
+ Int4 minZ = CmpNLE(Float4(0.0f), outPosZ);
Int flags;
@@ -121,7 +138,7 @@
Int4 finiteX = CmpLE(Abs(o[pos].x), *Pointer<Float4>(constants + OFFSET(Constants,maxPos)));
Int4 finiteY = CmpLE(Abs(o[pos].y), *Pointer<Float4>(constants + OFFSET(Constants,maxPos)));
- Int4 finiteZ = CmpLE(Abs(o[pos].z), *Pointer<Float4>(constants + OFFSET(Constants,maxPos)));
+ Int4 finiteZ = CmpLE(Abs(outPosZ), *Pointer<Float4>(constants + OFFSET(Constants,maxPos)));
flags = SignMask(finiteX & finiteY & finiteZ);
clipFlags |= *Pointer<Int>(constants + OFFSET(Constants,fini) + flags * 4);
@@ -570,7 +587,7 @@
o[pos].y = o[pos].y + *Pointer<Float4>(data + OFFSET(DrawData,YYYY)) * o[pos].w;
}
- if(symmetricNormalizedDepth && !state.fixedFunction)
+ if(!state.transformFeedbackEnabled && symmetricNormalizedDepth && !state.fixedFunction)
{
o[pos].z = (o[pos].z + o[pos].w) * Float4(0.5f);
}
@@ -667,7 +684,7 @@
*Pointer<Float4>(cacheLine + OFFSET(Vertex,X) + sizeof(Vertex) * 3, 16) = v.w;
}
- void VertexRoutine::writeVertex(const Pointer<Byte> &vertex, Pointer<Byte> &cache)
+ void VertexRoutine::writeVertex(const Pointer<Byte> &vertex, Pointer<Byte> &cache, const UInt &primitiveNumber, const UInt &indexInPrimitive)
{
for(int i = 0; i < MAX_VERTEX_OUTPUTS; i++)
{
@@ -679,5 +696,45 @@
*Pointer<Int4>(vertex + OFFSET(Vertex,X)) = *Pointer<Int4>(cache + OFFSET(Vertex,X));
*Pointer<Int>(vertex + OFFSET(Vertex,clipFlags)) = *Pointer<Int>(cache + OFFSET(Vertex,clipFlags));
+
+ if(state.transformFeedbackEnabled != 0)
+ {
+ If(indexInPrimitive < state.verticesPerPrimitive)
+ {
+ UInt tOffset = primitiveNumber * state.verticesPerPrimitive + indexInPrimitive;
+ for(int i = 0; i < MAX_TRANSFORM_FEEDBACK_INTERLEAVED_COMPONENTS; ++i)
+ {
+ if(state.transformFeedbackEnabled & (1ULL << i))
+ {
+ UInt reg = *Pointer<UInt>(data + OFFSET(DrawData, vs.reg[i]));
+ UInt row = *Pointer<UInt>(data + OFFSET(DrawData, vs.row[i]));
+ UInt col = *Pointer<UInt>(data + OFFSET(DrawData, vs.col[i]));
+ UInt str = *Pointer<UInt>(data + OFFSET(DrawData, vs.str[i]));
+
+ Pointer<Byte> t = *Pointer<Pointer<Byte>>(data + OFFSET(DrawData, vs.t[i])) + (tOffset * str * sizeof(float));
+ Pointer<Byte> v = vertex + OFFSET(Vertex, v) + reg * sizeof(float);
+
+ For(UInt r = 0, r < row, r++)
+ {
+ UInt rOffsetX = r * col * sizeof(float);
+ UInt rOffset4 = r * sizeof(float4);
+ For(UInt c = 0, c < col, c++)
+ {
+ UInt cOffset = c * sizeof(float);
+ *Pointer<Float>(t + rOffsetX + cOffset) = *Pointer<Float>(v + rOffset4 + cOffset);
+ }
+ }
+ }
+ }
+ }
+
+ // Make this correction after transform feedback has been outputted
+ if(symmetricNormalizedDepth && !state.fixedFunction && state.output[state.positionRegister].write)
+ {
+ Float z = *Pointer<Float>(vertex + OFFSET(Vertex, v[state.positionRegister]) + 2 * sizeof(float));
+ Float w = *Pointer<Float>(vertex + OFFSET(Vertex, v[state.positionRegister]) + 3 * sizeof(float));
+ *Pointer<Float>(vertex + OFFSET(Vertex, v[state.positionRegister]) + 2 * sizeof(float)) = (z + w) * Float(0.5f);
+ }
+ }
}
}
diff --git a/src/Shader/VertexRoutine.hpp b/src/Shader/VertexRoutine.hpp
index 7f29ff5..eadd795 100644
--- a/src/Shader/VertexRoutine.hpp
+++ b/src/Shader/VertexRoutine.hpp
@@ -63,7 +63,7 @@
void computeClipFlags();
void postTransform();
void writeCache(Pointer<Byte> &cacheLine);
- void writeVertex(const Pointer<Byte> &vertex, Pointer<Byte> &cacheLine);
+ void writeVertex(const Pointer<Byte> &vertex, Pointer<Byte> &cacheLine, const UInt &primitiveNumber, const UInt &indexInPrimitive);
};
}
