Compute shader buffer transfer converter

src/main/java/me/cortex/voxy/client/core/gl/GlFramebuffer.java

@@ -19,6 +19,11 @@ public class GlFramebuffer extends TrackedObject {
         return this;
     }
 
+    public GlFramebuffer bind(int attachment, GlRenderBuffer buffer) {
+        glNamedFramebufferRenderbuffer(this.id, attachment, GL_RENDERBUFFER, buffer.id);
+        return this;
+    }
+
     @Override
     public void free() {
         super.free0();

src/main/java/me/cortex/voxy/client/core/gl/GlRenderBuffer.java

@@ -0,0 +1,21 @@
+package me.cortex.voxy.client.core.gl;
+
+import me.cortex.voxy.common.util.TrackedObject;
+
+import static org.lwjgl.opengl.GL11C.*;
+import static org.lwjgl.opengl.GL45C.*;
+
+public class GlRenderBuffer extends TrackedObject {
+    public final int id;
+
+    public GlRenderBuffer(int format, int width, int height) {
+        this.id = glCreateRenderbuffers();
+        glNamedRenderbufferStorage(this.id, format, width, height);
+    }
+
+    @Override
+    public void free() {
+        super.free0();
+        glDeleteRenderbuffers(this.id);
+    }
+}

src/main/java/me/cortex/voxy/client/core/gl/GlTexture.java

@@ -11,6 +11,7 @@ import static org.lwjgl.opengl.GL45C.*;
 public class GlTexture extends TrackedObject {
     public final int id;
     private final int type;
+    private int format;
     public GlTexture() {
         this(GL_TEXTURE_2D);
     }
@@ -20,7 +21,17 @@ public class GlTexture extends TrackedObject {
         this.type = type;
     }
 
+    private GlTexture(int type, boolean useGenTypes) {
+        if (useGenTypes) {
+            this.id = glGenTextures();
+        } else {
+            this.id = glCreateTextures(type);
+        }
+        this.type = type;
+    }
+
     public GlTexture store(int format, int levels, int width, int height) {
+        this.format = format;
         if (this.type == GL_TEXTURE_2D) {
             glTextureStorage2D(this.id, levels, format, width, height);
         } else {
@@ -29,21 +40,15 @@ public class GlTexture extends TrackedObject {
         return this;
     }
 
+    public GlTexture createView() {
+        var view = new GlTexture(this.type, true);
+        glTextureView(view.id, this.type, this.id, this.format, 0, 1, 0, 1);
+        return view;
+    }
+
     @Override
     public void free() {
         super.free0();
         glDeleteTextures(this.id);
     }
-
-    //TODO: FIXME, glGetTextureParameteri doesnt work
-    public static int getRawTextureType(int texture) {
-        if (!glIsTexture(texture)) {
-            throw new IllegalStateException("Not texture");
-        }
-        int immFormat = glGetTextureParameteri(texture, GL_TEXTURE_IMMUTABLE_FORMAT);
-        if (immFormat == 0) {
-            throw new IllegalStateException("Texture: " + texture + " is not immutable");
-        }
-        return immFormat;
-    }
 }

src/main/java/me/cortex/voxy/client/core/model/ModelBakerySubsystem.java

@@ -18,13 +18,14 @@ public class ModelBakerySubsystem {
     public final ModelFactory factory;
     private final IntLinkedOpenHashSet blockIdQueue = new IntLinkedOpenHashSet();
 
+
     public ModelBakerySubsystem(Mapper mapper) {
         this.factory = new ModelFactory(mapper, this.storage, this.textureDownStream);
     }
 
     public void tick() {
         //There should be a method to access the frame time IIRC, if the user framecap is unlimited lock it to like 60 fps for computation
-        int BUDGET = 50;//TODO: make this computed based on the remaining free time in a frame (and like div by 2 to reduce overhead) (with a min of 1)
+        int BUDGET = 10;//TODO: make this computed based on the remaining free time in a frame (and like div by 2 to reduce overhead) (with a min of 1)
 
         for (int i = 0; i < BUDGET && !this.blockIdQueue.isEmpty(); i++) {
             int blockId = -1;

src/main/java/me/cortex/voxy/client/core/model/ModelFactory.java

@@ -185,16 +185,11 @@ public class ModelFactory {
                 int[] colour = new int[FACE_SIZE];
                 int[] depth = new int[FACE_SIZE];
 
-                //TODO: see if there is a memory intrinsic to help here
                 //Copy out colour
                 for (int i = 0; i < FACE_SIZE; i++) {
+                    //De-interpolate results
                     colour[i] = MemoryUtil.memGetInt(faceDataPtr+ (i*4));
-                }
+                    depth[i] = MemoryUtil.memGetInt(faceDataPtr+ (i*4)+4);
-
-                //Shift ptr and copy out depth
-                faceDataPtr += FACE_SIZE*4;
-                for (int i = 0; i < FACE_SIZE; i++) {
-                    depth[i] = MemoryUtil.memGetInt(faceDataPtr+ (i*4));
                 }
 
                 textureData[face] = new ColourDepthTextureData(colour, depth, MODEL_TEXTURE_SIZE, MODEL_TEXTURE_SIZE);

src/main/java/me/cortex/voxy/client/core/model/ModelTextureBakery.java

@@ -1,11 +1,8 @@
 package me.cortex.voxy.client.core.model;
 
-import com.mojang.blaze3d.platform.GlConst;
 import com.mojang.blaze3d.platform.GlStateManager;
-import com.mojang.blaze3d.systems.RenderSystem;
-import com.mojang.blaze3d.systems.VertexSorter;
-import me.cortex.voxy.client.core.gl.GlBuffer;
 import me.cortex.voxy.client.core.gl.GlFramebuffer;
+import me.cortex.voxy.client.core.gl.GlRenderBuffer;
 import me.cortex.voxy.client.core.gl.GlTexture;
 import me.cortex.voxy.client.core.gl.shader.Shader;
 import me.cortex.voxy.client.core.gl.shader.ShaderType;
@@ -30,14 +27,14 @@ import net.minecraft.world.biome.ColorResolver;
 import net.minecraft.world.chunk.light.LightingProvider;
 import org.jetbrains.annotations.Nullable;
 import org.joml.Matrix4f;
+import org.lwjgl.opengl.GL11;
 import org.lwjgl.opengl.GL11C;
 
 import java.util.ArrayList;
 import java.util.List;
 
 import static org.lwjgl.opengl.ARBDirectStateAccess.glGetTextureImage;
-import static org.lwjgl.opengl.ARBImaging.GL_FUNC_ADD;
+import static org.lwjgl.opengl.ARBDirectStateAccess.glTextureParameteri;
-import static org.lwjgl.opengl.ARBImaging.glBlendEquation;
 import static org.lwjgl.opengl.ARBShaderImageLoadStore.GL_FRAMEBUFFER_BARRIER_BIT;
 import static org.lwjgl.opengl.ARBShaderImageLoadStore.glMemoryBarrier;
 import static org.lwjgl.opengl.GL11C.GL_TEXTURE_2D;
@@ -46,7 +43,7 @@ import static org.lwjgl.opengl.GL15C.glBindBuffer;
 import static org.lwjgl.opengl.GL20C.glUniformMatrix4fv;
 import static org.lwjgl.opengl.GL21C.GL_PIXEL_PACK_BUFFER;
 import static org.lwjgl.opengl.GL30.*;
-import static org.lwjgl.opengl.GL43.glCopyImageSubData;
+import static org.lwjgl.opengl.GL43.*;
 
 //Builds a texture for each face of a model
 public class ModelTextureBakery {
@@ -54,6 +51,7 @@ public class ModelTextureBakery {
     private final int height;
     private final GlTexture colourTex;
     private final GlTexture depthTex;
+    private final GlTexture depthTexView;
     private final GlFramebuffer framebuffer;
     private final GlStateCapture glState = GlStateCapture.make()
             .addCapability(GL_DEPTH_TEST)
@@ -62,6 +60,7 @@ public class ModelTextureBakery {
             .addCapability(GL_CULL_FACE)
             .addTexture(GL_TEXTURE0)
             .addTexture(GL_TEXTURE1)
+            .addTexture(GL_TEXTURE2)
             .build()
             ;
     private final Shader rasterShader = Shader.make()
@@ -69,10 +68,9 @@ public class ModelTextureBakery {
             .add(ShaderType.FRAGMENT, "voxy:bakery/position_tex.fsh")
             .compile();
 
-    private static final List<MatrixStack> FACE_VIEWS = new ArrayList<>();
+    private final Shader copyOutShader;
 
-    //A truly terrible hackfix for nvidia drivers murderizing performance with PBO readout with a depth texture
+    private static final List<MatrixStack> FACE_VIEWS = new ArrayList<>();
-    private static final GlTexture TEMPORARY = new GlTexture().store(GL_R32UI, 1, ModelFactory.MODEL_TEXTURE_SIZE, ModelFactory.MODEL_TEXTURE_SIZE);
 
 
     public ModelTextureBakery(int width, int height) {
@@ -82,8 +80,19 @@ public class ModelTextureBakery {
         this.height = height;
         this.colourTex = new GlTexture().store(GL_RGBA8, 1, width, height);
         this.depthTex = new GlTexture().store(GL_DEPTH24_STENCIL8, 1, width, height);
+        this.depthTexView = this.depthTex.createView();
+
         this.framebuffer = new GlFramebuffer().bind(GL_COLOR_ATTACHMENT0, this.colourTex).bind(GL_DEPTH_STENCIL_ATTACHMENT, this.depthTex).verify();
 
+        glTextureParameteri(this.depthTex.id, GL_DEPTH_STENCIL_TEXTURE_MODE, GL_DEPTH_COMPONENT);
+        glTextureParameteri(this.depthTexView.id, GL_DEPTH_STENCIL_TEXTURE_MODE, GL_STENCIL_INDEX);
+
+        this.copyOutShader = Shader.make()
+                .define("WIDTH", width)
+                .define("HEIGHT", height)
+                .add(ShaderType.COMPUTE, "voxy:bakery/buffercopy.comp")
+                .compile();
+
         //This is done to help make debugging easier
         FACE_VIEWS.clear();
         AddViews();
@@ -175,15 +184,13 @@ public class ModelTextureBakery {
         glStencilFunc(GL_ALWAYS, 1, 0xFF);
         glStencilMask(0xFF);
 
-        this.rasterShader.bind();
+
-        glActiveTexture(GL_TEXTURE0);
         int texId = MinecraftClient.getInstance().getTextureManager().getTexture(Identifier.of("minecraft", "textures/atlas/blocks.png")).getGlId();
-        GlUniform.uniform1(0, 0);
 
         final int TEXTURE_SIZE = this.width*this.height *4;//NOTE! assume here that both depth and colour are 4 bytes in size
         for (int i = 0; i < FACE_VIEWS.size(); i++) {
             int faceOffset = streamBaseOffset + TEXTURE_SIZE*i*2;
-            captureViewToStream(state, model, entityModel, FACE_VIEWS.get(i), randomValue, i, renderFluid, texId, projection, streamBuffer, faceOffset, faceOffset + TEXTURE_SIZE);
+            captureViewToStream(state, model, entityModel, FACE_VIEWS.get(i), randomValue, i, renderFluid, texId, projection, streamBuffer, faceOffset);
         }
 
         renderLayer.endDrawing();
@@ -199,7 +206,11 @@ public class ModelTextureBakery {
     }
 
     private final BufferAllocator allocator = new BufferAllocator(786432);
-    private void captureViewToStream(BlockState state, BakedModel model, BakedBlockEntityModel blockEntityModel, MatrixStack stack, long randomValue, int face, boolean renderFluid, int textureId, Matrix4f projection, int streamBuffer, int streamColourOffset, int streamDepthOffset) {
+    private void captureViewToStream(BlockState state, BakedModel model, BakedBlockEntityModel blockEntityModel, MatrixStack stack, long randomValue, int face, boolean renderFluid, int textureId, Matrix4f projection, int streamBuffer, int streamOffset) {
+        this.rasterShader.bind();
+        glActiveTexture(GL_TEXTURE0);
+        GlUniform.uniform1(0, 0);
+
         glClear(GL_COLOR_BUFFER_BIT | GL_DEPTH_BUFFER_BIT | GL_STENCIL_BUFFER_BIT);
         float[] mat = new float[4*4];
         new Matrix4f(projection).mul(stack.peek().getPositionMatrix()).get(mat);
@@ -292,27 +303,24 @@ public class ModelTextureBakery {
 
         glMemoryBarrier(GL_FRAMEBUFFER_BARRIER_BIT);
 
-        GlStateManager._pixelStore(GL_PACK_ROW_LENGTH, 0);
+        this.emitToStream(streamBuffer, streamOffset);
-        GlStateManager._pixelStore(GL_PACK_SKIP_PIXELS, 0);
-        GlStateManager._pixelStore(GL_PACK_SKIP_ROWS, 0);
-        GlStateManager._pixelStore(GL_PACK_ALIGNMENT, 4);
-
-        glBindBuffer(GL_PIXEL_PACK_BUFFER, streamBuffer);
-        {//Copy colour
-            glGetTextureImage(this.colourTex.id, 0, GL_RGBA, GL_UNSIGNED_BYTE, this.width*this.height*4, streamColourOffset);
     }
 
-        //TODO: fixme!! only do this dodgy double copy if the driver is nvidia
+    private void emitToStream(int streamBuffer, int streamOffset) {
-        {//Copy depth
+        if (streamOffset%4 != 0) {
-            //First copy to the temporary buffer
+            throw new IllegalArgumentException();
-            glCopyImageSubData(textureId, GL_TEXTURE_2D, 0,0,0,0,
-                    TEMPORARY.id, GL_TEXTURE_2D, 0,0,0,0,
-                    ModelFactory.MODEL_TEXTURE_SIZE, ModelFactory.MODEL_TEXTURE_SIZE, 1);
-
-            //Then download
-            glGetTextureImage(TEMPORARY.id, 0, GL_RED_INTEGER, GL_UNSIGNED_INT, this.width*this.height*4, streamDepthOffset);
         }
-        glBindBuffer(GL_PIXEL_PACK_BUFFER, 0);
+        this.copyOutShader.bind();
+        glActiveTexture(GL_TEXTURE0);
+        GL11C.glBindTexture(GL11.GL_TEXTURE_2D, this.colourTex.id);
+        glActiveTexture(GL_TEXTURE1);
+        glBindTexture(GL11.GL_TEXTURE_2D, this.depthTex.id);
+        glActiveTexture(GL_TEXTURE2);
+        glBindTexture(GL11.GL_TEXTURE_2D, this.depthTexView.id);
+        glBindBufferBase(GL_SHADER_STORAGE_BUFFER, 3, streamBuffer);
+        glUniform1ui(4, streamOffset/4);
+
+        glDispatchCompute(1,1,1);
     }
 
     private static void renderQuads(BufferBuilder builder, BlockState state, BakedModel model, MatrixStack stack, long randomValue) {
@@ -329,8 +337,10 @@ public class ModelTextureBakery {
     public void free() {
         this.framebuffer.free();
         this.colourTex.free();
+        this.depthTexView.free();
         this.depthTex.free();
         this.rasterShader.free();
+        this.copyOutShader.free();
         this.allocator.close();
     }
 }

src/main/java/me/cortex/voxy/client/core/model/TextureUtils.java

@@ -94,7 +94,8 @@ public class TextureUtils {
         //https://registry.khronos.org/OpenGL-Refpages/gl4/html/glDepthRange.xhtml
         // due to this and the unsigned bullshit, i believe the depth value needs to get multiplied by 2
 
-        depthF *= 2;
+        //Shouldent be needed due to the compute bake copy
+        //depthF *= 2;
         if (depthF > 1.00001f) {
             System.err.println("Warning: Depth greater than 1");
             depthF = 1.0f;
@@ -180,6 +181,7 @@ public class TextureUtils {
         return weightedAverageColor(weightedAverageColor(one, two), weightedAverageColor(three, four));
     }
 
+    //TODO: FIXME!!! ITS READING IT AS ABGR??? isnt the format RGBA??
     private static int weightedAverageColor(int one, int two) {
         int alphaOne = ColorHelper.Abgr.getAlpha(one);
         int alphaTwo = ColorHelper.Abgr.getAlpha(two);

src/main/resources/assets/voxy/shaders/bakery/buffercopy.comp

@@ -0,0 +1,28 @@
+#version 450
+
+layout(local_size_x = WIDTH, local_size_y = HEIGHT) in;
+
+layout(binding = 0) uniform sampler2D colourTexIn;
+layout(binding = 1) uniform sampler2D depthTexIn;
+layout(binding = 2) uniform usampler2D stencilTexIn;
+layout(binding = 3, std430) writeonly restrict buffer OutBuffer {
+    uint[] outBuffer;
+};
+
+
+layout(location=4) uniform uint bufferOffset;
+
+
+void main() {
+    ivec2 point = ivec2(gl_GlobalInvocationID.xy);
+    uint writeIndex = ((gl_GlobalInvocationID.x+(gl_GlobalInvocationID.y*HEIGHT))*2)+bufferOffset;
+    uvec4 colour = clamp(uvec4(texelFetch(colourTexIn, point, 0)*255), uvec4(0), uvec4(255));//TODO: check that this actually gets to the range of 255
+    colour <<= uvec4(0,8,16,24);//ABGR format!!!
+    outBuffer[writeIndex] = colour.r|colour.g|colour.b|colour.a;
+
+    float depth = clamp(texelFetch(depthTexIn, point, 0).r, 0, 1);//Opengl grumble grumble
+    uint stencil = texelFetch(stencilTexIn, point, 0).r;
+    uint value = uint(depth*((1<<24)-1))<<8;
+    value |= stencil;
+    outBuffer[writeIndex+1] = value;
+}