Merge streaming into collide shaderpattern/ab

i.e. implement the A-B pattern.

Dispatching only one compute shader per interaction-less simulation step
already yields very noticeable performance gains. All cell types are now
fully handled by the collide shader which further simplifies the code.

3 files changed, 31 insertions, 102 deletions

diff --git a/src/main.cc b/src/main.cc
index 71fbc88..fb3bfd0 100644
--- a/src/main.cc
+++ b/src/main.cc
@@ -20,7 +20,6 @@
 
 #include "shader/code/interact.glsl"
 #include "shader/code/collide.glsl"
-#include "shader/code/stream.glsl"
 
 #include "timer.h"
 
@@ -100,7 +99,6 @@ int render(bool open_boundaries) {
 
 	std::unique_ptr<ComputeShader> interact_shader;
 	std::unique_ptr<ComputeShader> collide_shader;
-	std::unique_ptr<ComputeShader> stream_shader;
 
 	window.init([&]() {
 		scene_shader = std::make_unique<GraphicShader>(
@@ -112,10 +110,9 @@ int render(bool open_boundaries) {
 
 		interact_shader = std::make_unique<ComputeShader>(INTERACT_SHADER_CODE);
 		collide_shader  = std::make_unique<ComputeShader>(COLLIDE_SHADER_CODE);
-		stream_shader   = std::make_unique<ComputeShader>(STREAM_SHADER_CODE);
 	});
 
-	if ( !interact_shader->isGood() || !collide_shader->isGood() || !stream_shader->isGood() ) {
+	if ( !interact_shader->isGood() || !collide_shader->isGood() ) {
 		std::cerr << "Compute shader error." << std::endl;
 		return -1;
 	}
@@ -151,14 +148,12 @@ int render(bool open_boundaries) {
 		if ( update_lattice ) {
 			if ( timer::millisecondsSince(last_frame) >= 1000/lups ) {
 				if ( tick ) {
-					collide_shader->workOn(tick_buffers);
-					stream_shader->workOn(tick_buffers);
 					interact_shader->workOn(tick_buffers);
+					collide_shader->workOn(tick_buffers);
 					tick = false;
 				} else {
-					collide_shader->workOn(tock_buffers);
-					stream_shader->workOn(tock_buffers);
 					interact_shader->workOn(tock_buffers);
+					collide_shader->workOn(tock_buffers);
 					tick = true;
 				}
 
@@ -193,10 +188,6 @@ int render(bool open_boundaries) {
 					auto guard = collide_shader->use();
 					collide_shader->dispatch(nX, nY);
 				}
-				{
-					auto guard = stream_shader->use();
-					stream_shader->dispatch(nX, nY);
-				}
 
 				last_frame = timer::now();
 			}
diff --git a/src/shader/code/collide.glsl b/src/shader/code/collide.glsl
index 673a285..4efd7fb 100644
--- a/src/shader/code/collide.glsl
+++ b/src/shader/code/collide.glsl
@@ -61,7 +61,7 @@ float get(uint x, uint y, int i, int j) {
 }
 
 void set(uint x, uint y, int i, int j, float v) {
-	collideCells[indexOfLatticeCell(x,y) + indexOfDirection(i,j)] = v;
+	streamCells[indexOfLatticeCell(x,y) + indexOfDirection(i,j)] = v;
 }
 
 void setFluid(uint x, uint y, vec2 v) {
@@ -105,6 +105,18 @@ bool isBulkFluidCell(int material) {
 	return material == 1 || material == 4 || material == 5 || material == 6;
 }
 
+bool isBounceBackCell(int material) {
+	return material == 2 || material == 3;
+}
+
+bool isInflowCell(int material) {
+	return material == 5;
+}
+
+bool isOutflowCell(int material) {
+	return material == 6;
+}
+
 float getExternalMassInflux(int material) {
 	if ( material == 4 ) {
 		return 1.5;
@@ -113,7 +125,7 @@ float getExternalMassInflux(int material) {
 	}
 };
 
-/// Actual collide kernel
+/// Actual collide&stream kernel
 
 void main() {
 	const uint x = gl_GlobalInvocationID.x;
@@ -125,14 +137,14 @@ void main() {
 
 	const int material = getMaterial(x,y);
 
-	if ( isBulkFluidCell(material) ) {
-		float d = max(density(x,y), getExternalMassInflux(material));
-		vec2  v = velocity(x,y,d);
+	float d = max(density(x,y), getExternalMassInflux(material));
+	vec2  v = velocity(x,y,d);
 
-		if ( material == 5 ) {
+	if ( isBulkFluidCell(material) ) {
+		if ( isInflowCell(material) ) {
 			v = vec2(0.1,0.0);
 		}
-		if ( material == 6 ) {
+		if ( isOutflowCell(material) ) {
 			d = 1.0;
 		}
 
@@ -140,7 +152,15 @@ void main() {
 
 		for ( int i = -1; i <= 1; ++i ) {
 			for ( int j = -1; j <= 1; ++j ) {
-				set(x,y,i,j, get(x,y,i,j) + omega * (equilibrium(d,v,i,j) - get(x,y,i,j)));
+				set(x+i,y+j,i,j, get(x,y,i,j) + omega * (equilibrium(d,v,i,j) - get(x,y,i,j)));
+			}
+		}
+	}
+
+	if ( isBounceBackCell(material) ) {
+		for ( int i = -1; i <= 1; ++i ) {
+			for ( int j = -1; j <= 1; ++j ) {
+				set(x+(-1)*i,y+(-1)*j,(-1)*i,(-1)*j, get(x,y,i,j) + omega * (equilibrium(d,v,i,j) - get(x,y,i,j)));
 			}
 		}
 	}
diff --git a/src/shader/code/stream.glsl b/src/shader/code/stream.glsl
deleted file mode 100644
index 59aed98..0000000
--- a/src/shader/code/stream.glsl
+++ /dev/null
@@ -1,82 +0,0 @@
-static const std::string STREAM_SHADER_CODE = R"(
-#version 430
-
-layout (local_size_x = 1, local_size_y = 1) in;
-
-layout (std430, binding=1) buffer bufferCollide  { float collideCells[];  };
-layout (std430, binding=2) buffer bufferStream   { float streamCells[];   };
-layout (std430, binding=3) buffer bufferFluid    { float fluidCells[];    };
-
-/// LBM constants
-
-uniform uint nX;
-uniform uint nY;
-
-const uint q = 9;
-
-// Array indexing
-
-uint indexOfDirection(int i, int j) {
-	return 3*(i+1) + (j+1);
-}
-
-uint indexOfLatticeCell(uint x, uint y) {
-	return q*nX*y + q*x;
-}
-
-uint indexOfFluidVertex(uint x, uint y) {
-	return 3*nX*y + 3*x;
-}
-
-/// Data access
-
-float get(uint x, uint y, int i, int j) {
-	return collideCells[indexOfLatticeCell(x,y) + indexOfDirection(i,j)];
-}
-
-void set(uint x, uint y, int i, int j, float v) {
-	streamCells[indexOfLatticeCell(x,y) + indexOfDirection(i,j)] = v;
-}
-
-int getMaterial(uint x, uint y) {
-	const uint idx = indexOfFluidVertex(x, y);
-	return int(fluidCells[idx + 2]);
-}
-
-/// Boundary conditions
-
-void bounceBack(uint x, uint y) {
-	for ( int i = -1; i <= 1; ++i ) {
-		for ( int j = -1; j <= 1; ++j ) {
-			set(x,y,i,j, get(x,y,(-1)*i,(-1)*j));
-		}
-	}
-}
-
-/// Actual stream kernel
-
-void main() {
-	const uint x = gl_GlobalInvocationID.x;
-	const uint y = gl_GlobalInvocationID.y;
-
-	if ( !(x < nX && y < nY) ) {
-		return;
-	}
-
-	const int material = getMaterial(x,y);
-
-	if ( material == 0 ) {
-		return;
-	}
-
-	if ( material == 2 || material == 3 ) {
-		bounceBack(x,y);
-	}
-
-	for ( int i = -1; i <= 1; ++i ) {
-		for ( int j = -1; j <= 1; ++j ) {
-			set(x+i,y+j,i,j, get(x,y,i,j));
-		}
-	}
-}
-)";