draw 3d cubes

.gitignore

@@ -0,0 +1,3 @@
+*.wasm
+*.pyc
+__pycache__

build.sh

@@ -0,0 +1,29 @@
+clang++ \
+    -v \
+    --target=wasm32 \
+    -O3 \
+    -flto \
+    -nostdlib \
+    -Wl,--no-entry \
+    -Wl,--export=rotate \
+    -Wl,--lto-O3 \
+    -Wl,--print-map \
+    -Wl,--import-memory \
+    -D_XM_NO_INTRINSICS_ \
+    -I. \
+    -o rotate.wasm \
+    rotate.c
+
+
+# clang \
+#     -O1 \
+#   --target=wasm32 \
+#   -emit-llvm \
+#   -c \
+#   -S \
+#   test.c
+
+# llc \
+#   -march=wasm32 \
+#   -filetype=obj \
+#   test.ll

compute.wgsl

@@ -1,6 +1,7 @@
 struct Configuration {
   gridSize: vec2f,
-  aspect: vec2f
+  aspect: vec2f,
+  matrix: mat4x4f
 };
 
 @group(0) @binding(0) var<uniform> config: Configuration;

index.js

@@ -15,7 +15,6 @@ canvas.height = canvas.clientHeight
 
 const context = canvas.getContext("webgpu");
 const canvasFormat = navigator.gpu.getPreferredCanvasFormat();
-//console.log(canvasFormat);
 context.configure({
   device: device,
   format: canvasFormat,
@@ -36,30 +35,48 @@ renderPass.end();
 const commandBuffer = encoder.finish();
 device.queue.submit([commandBuffer]);
 
-const vertices = new Float32Array([
-//   X,    Y,
-  -0.8, -0.8, // Triangle 1 (Blue)
-   0.8, -0.8,
-   0.8,  0.8,
 
-  -0.8, -0.8, // Triangle 2 (Red)
+const cubeResponse = await fetch("obj/cube.bin");
-   0.8,  0.8,
+const cubeObj = await cubeResponse.arrayBuffer();
-  -0.8,  0.8,
-]);
 
-const vertexBuffer = device.createBuffer({
+const cubeView = new DataView(cubeObj);
+const indexCount = cubeView.getUint32(0, true);
+const indexBufferOffset = cubeView.getUint32(4, true);
+const indexBufferSize = cubeView.getUint32(8, true);
+const vertexBufferOffset = cubeView.getUint32(12, true);
+const vertexBufferSize = cubeView.getUint32(16, true);
+console.log(`indexBufferOffset ${indexBufferOffset} indexBufferSize ${indexBufferSize}`);
+console.log(`vertexBufferOffset ${vertexBufferOffset} vertexBufferSize ${vertexBufferSize}`);
+const cubeIndexCount = indexBufferSize / 2;
+
+const cubeVerticesStride = 36;
+
+const indexVertexBuffer = device.createBuffer({
   label: "cell vertices",
-  size: vertices.byteLength,
+  size: indexBufferSize + vertexBufferSize,
-  usage: GPUBufferUsage.VERTEX | GPUBufferUsage.COPY_DST,
+  usage: GPUBufferUsage.INDEX | GPUBufferUsage.VERTEX | GPUBufferUsage.COPY_DST,
 });
-device.queue.writeBuffer(vertexBuffer, 0, vertices);
+device.queue.writeBuffer(indexVertexBuffer, 0, cubeObj, indexBufferOffset, indexBufferSize);
+device.queue.writeBuffer(indexVertexBuffer, indexBufferSize, cubeObj, vertexBufferOffset, vertexBufferSize);
 
 const vertexBufferLayout = {
-  arrayStride: 8,
+  arrayStride: cubeVerticesStride,
   attributes: [{
-    format: "float32x2",
+    format: "float32x3",
     offset: 0,
     shaderLocation: 0,
+  }, {
+    format: "float32x2",
+    offset: 12,
+    shaderLocation: 1,
+  }, {
+    format: "float16x4",
+    offset: 20,
+    shaderLocation: 2,
+  }, {
+    format: "float16x4",
+    offset: 28,
+    shaderLocation: 3,
   }],
   stepMode: "vertex",
 };
@@ -115,7 +132,15 @@ const cellPipeline = device.createRenderPipeline({
     targets: [{
       format: canvasFormat,
     }]
-  }
+  },
+  primitive: {
+    topology: 'triangle-list',
+  },
+  depthStencil: {
+    depthWriteEnabled: true,
+    depthCompare: 'less',
+    format: 'depth24plus',
+  },
 });
 
 const computeShaderModule = device.createShaderModule({
@@ -135,13 +160,24 @@ const computePipeline = device.createComputePipeline({
 const gridSize = 32;
 const workGroupSize = 16;
 
-const uniformArray = new Float32Array([gridSize, gridSize, 1, 1]);
+//const uniformArray2 = new Float32Array([gridSize, gridSize, 1, 1]);
+const memory = new WebAssembly.Memory({ initial: 1 });
+const uniformArray = new Float32Array(memory.buffer);
+uniformArray[0] = gridSize;
+uniformArray[1] = gridSize;
+uniformArray[2] = 1;
+uniformArray[3] = 1;
+const uniformArraySize = 4 * 4 + 4 * 4 * 4;
+const rotate = await WebAssembly.instantiateStreaming(fetch("rotate.wasm"), {
+  env: { memory: memory }
+});
+
 const uniformBuffer = device.createBuffer({
   label: "grid uniform",
-  size: uniformArray.byteLength,
+  size: uniformArraySize,
   usage: GPUBufferUsage.UNIFORM | GPUBufferUsage.COPY_DST,
 });
-device.queue.writeBuffer(uniformBuffer, 0, uniformArray);
+device.queue.writeBuffer(uniformBuffer, 0, uniformArray, 0, uniformArraySize / 4);
 
 const cellStateArray = new Uint32Array(gridSize * gridSize);
 
@@ -198,8 +234,28 @@ const bindGroups = [
 let tick = 0;
 let which = 0;
 
+function createDepthTexture() {
+  return device.createTexture({
+    size: [canvas.width, canvas.height],
+    format: 'depth24plus',
+    usage: GPUTextureUsage.RENDER_ATTACHMENT,
+  });
+}
+
+var depthTexture = undefined;
+
 function render()
 {
+  if (canvas.clientWidth !== canvas.width || canvas.clientWidth !== canvas.height) {
+    canvas.width = canvas.clientWidth;
+    canvas.height = canvas.clientHeight;
+
+    if (depthTexture !== undefined) {
+      depthTexture.destroy();
+    }
+    depthTexture = createDepthTexture();
+  }
+
   const encoder = device.createCommandEncoder();
 
   const computePass = encoder.beginComputePass();
@@ -209,15 +265,14 @@ function render()
   computePass.dispatchWorkgroups(workgroupCount);
   computePass.end();
 
+  const aspect = canvas.width / canvas.height;
+  rotate.instance.exports.rotate(tick * 0.01, aspect, 4 * 4);
+
   tick += 1;
   if ((tick % 5) == 0) {
     which ^= 1;
   }
 
-  canvas.width = canvas.clientWidth
-  canvas.height = canvas.clientHeight
-
-  const aspect = canvas.height / canvas.width;
   if (canvas.width > canvas.height) {
     uniformArray[2] = aspect;
     uniformArray[3] = 1;
@@ -225,7 +280,7 @@ function render()
     uniformArray[2] = 1;
     uniformArray[3] = 1 / aspect;
   }
-  device.queue.writeBuffer(uniformBuffer, 0, uniformArray, 0);
+  device.queue.writeBuffer(uniformBuffer, 0, uniformArray, 0, uniformArraySize / 4);
 
   const renderPass = encoder.beginRenderPass({
     colorAttachments: [{
@@ -233,13 +288,23 @@ function render()
       loadOp: "clear",
       clearValue: { r: 0.2, g: 0.2, b: 0.4, a: 1 },
       storeOp: "store",
-    }]
+    }],
+    depthStencilAttachment: {
+      view: depthTexture.createView(),
+      depthClearValue: 1.0,
+      depthLoadOp: 'clear',
+      depthStoreOp: 'store',
+    },
   });
 
   renderPass.setPipeline(cellPipeline);
-  renderPass.setVertexBuffer(0, vertexBuffer);
+  const vertexOffset = indexBufferSize;
+  renderPass.setVertexBuffer(0, indexVertexBuffer, vertexOffset, vertexBufferSize);
+  renderPass.setIndexBuffer(indexVertexBuffer, "uint16", 0, indexBufferSize);
   renderPass.setBindGroup(0, bindGroups[which]);
-  renderPass.draw(vertices.length / 2, gridSize * gridSize);
+  //renderPass.draw(cubeVerticesCount, gridSize * gridSize);
+  renderPass.drawIndexed(cubeIndexCount, gridSize * gridSize);
+  //renderPass.drawIndexed(cubeIndexCount);
 
   renderPass.end();
 

index.wgsl

@@ -1,23 +1,29 @@
 struct Configuration {
   gridSize: vec2f,
-  aspect: vec2f
+  aspect: vec2f,
+  matrix: mat4x4f
 };
 
 @group(0) @binding(0) var<uniform> config: Configuration;
 @group(0) @binding(1) var<storage> state: array<u32>;
 
 struct VertexInput {
-  @location(0) position: vec2f,
+  @location(0) position: vec3f,
+  @location(1) texture: vec2f,
+  @location(2) normal: vec3f,
+  @location(3) tangent: vec4f,
   @builtin(instance_index) instance: u32,
 };
 
 struct VertexOutput {
   @builtin(position) position: vec4f,
   @location(0) cell: vec2f,
+  @location(1) normal: vec3f,
 };
 
 struct FragmentInput {
   @location(0) cell: vec2f,
+  @location(1) normal: vec3f,
 };
 
 @vertex
@@ -28,16 +34,21 @@ fn vertexMain(input: VertexInput) -> VertexOutput
 
   let state = f32(state[input.instance]);
   let cell = vec2f(instance % config.gridSize.x, floor(instance / config.gridSize.y));
-  let grid = (position * state + cell) / config.gridSize;
+  let grid = (position * state + vec3f(cell, 0)) / config.gridSize.xxx;
 
   var output: VertexOutput;
-  output.position = vec4f((grid * 2 - 1) * config.aspect, 0, 1);
+  //output.position = (config.matrix * vec4f((grid * 2 - 1), 1)) * vec4f(config.aspect, 1, 1);
+  output.position = config.matrix * vec4f((grid * 2 - 1), 1);
+
+  //output.position = config.matrix * vec4f(input.position, 1);
+
   output.cell = cell / config.gridSize;
+  output.normal = input.normal;
   return output;
 }
 
 @fragment
 fn fragmentMain(input: FragmentInput) -> @location(0) vec4f
 {
-  return vec4f(input.cell, 0, 1);
+  return vec4f(input.normal * 0.5 + 0.5, 1);
 }

obj/cube.obj

@@ -0,0 +1,36 @@
+# Blender 5.2.0 LTS
+# www.blender.org
+mtllib Untitled.mtl
+o Cube
+v 1.000000 1.000000 -1.000000
+v 1.000000 -1.000000 -1.000000
+v 1.000000 1.000000 1.000000
+v 1.000000 -1.000000 1.000000
+v -1.000000 1.000000 -1.000000
+v -1.000000 -1.000000 -1.000000
+v -1.000000 1.000000 1.000000
+v -1.000000 -1.000000 1.000000
+vn -0.0000 1.0000 -0.0000
+vn -0.0000 -0.0000 1.0000
+vn -1.0000 -0.0000 -0.0000
+vn -0.0000 -1.0000 -0.0000
+vn 1.0000 -0.0000 -0.0000
+vn -0.0000 -0.0000 -1.0000
+vt 0.000000 1.000000
+vt 1.000000 0.000000
+vt 1.000000 1.000000
+vt 0.000000 0.000000
+s 0
+usemtl Material
+f 5/1/1 3/2/1 1/3/1
+f 3/3/2 8/4/2 4/2/2
+f 7/1/3 6/2/3 8/4/3
+f 2/3/4 8/4/4 6/1/4
+f 1/3/5 4/4/5 2/2/5
+f 5/1/6 2/2/6 6/4/6
+f 5/1/1 7/4/1 3/2/1
+f 3/3/2 7/1/2 8/4/2
+f 7/1/3 5/3/3 6/2/3
+f 2/3/4 4/2/4 8/4/4
+f 1/3/5 3/1/5 4/4/5
+f 5/1/6 1/3/6 2/2/6

rotate.c

@@ -0,0 +1,25 @@
+#include "directxmath/directxmath.h"
+
+extern "C" void rotate(float angle, float aspect, float * buffer);
+
+void rotate(float angle, float aspect, float * buffer)
+{
+  //XMMATRIX mat = XMMatrixRotationZ(angle);
+  //XMMATRIX mat2 = XMMatrixRotationX(angle);
+  //XMMATRIX mat2 = XMMatrixTranslation(0, 0, 1);
+  //XMStoreFloat4x4((XMFLOAT4X4*)buffer, mat * mat2);
+
+
+  XMVECTOR eye = XMVectorSet(3, 3, 3, 0);
+  XMVECTOR at = XMVectorSet(0, 0, 0, 0);
+  XMVECTOR up = XMVectorSet(0, 1, 0, 0);
+  XMMATRIX view = XMMatrixLookAtRH(eye, at, up);
+
+  float fov_angle_y = XMConvertToRadians(45);
+  float near_z = 0.01;
+  float far_z = 1000.0;
+  XMMATRIX projection = XMMatrixPerspectiveFovRH(fov_angle_y, aspect, near_z, far_z);
+
+  XMMATRIX rot = XMMatrixRotationY(angle);
+  XMStoreFloat4x4((XMFLOAT4X4*)buffer, rot * view * projection);
+}

tools/obj/obj.py

@@ -0,0 +1,128 @@
+import sys
+import struct
+from itertools import chain
+from dataclasses import dataclass
+
+from obj import tangent
+
+with open(sys.argv[1], 'r') as f:
+    lines = f.read().split('\n')
+
+position = []
+texture = []
+normal = []
+faces = []
+
+def parse_vertex(prefix, s):
+    return s.removeprefix(prefix).strip().split()
+
+def float_vector(l, n):
+    v = list(map(float, l))
+    assert len(v) == n
+    return v
+
+vertex_types = [
+    ("v ", position, 3),
+    ("vn ", normal, 3),
+    ("vt ", texture, 2),
+]
+
+ignore_prefixes = ["#", "mtllib ", "usemtl ", "o ", "s "]
+
+def parse_line(line):
+    if not line.strip():
+        return
+
+    for prefix in ignore_prefixes:
+        if line.startswith(prefix):
+            return
+
+    for prefix, out, n in vertex_types:
+        if line.startswith(prefix):
+            out.append(float_vector(parse_vertex(prefix, line), n))
+            return
+
+    if line.startswith("f "):
+        vertices = [
+            tuple(int(i)-1 for i in vertex.split("/"))
+            for vertex in parse_vertex("f ", line)
+        ]
+        assert len(vertices) == 3, vertices # is not triangle
+        faces.append(vertices)
+        return
+
+    assert False, repr(line) # unhandled line prefix
+
+for line in lines:
+    parse_line(line)
+
+vertex_buffer_keys = dict()
+vertex_buffer = []
+index_buffer = []
+
+for face in faces:
+    for vertex in face:
+        if vertex not in vertex_buffer_keys:
+            vertex_buffer_keys[vertex] = len(vertex_buffer)
+            vertex_buffer.append(tuple(chain(
+                position[vertex[0]],
+                texture[vertex[1]],
+                normal[vertex[2]],
+            )))
+
+        index_buffer.append(vertex_buffer_keys[vertex])
+
+def render_vertex(f, vertex, tangent):
+    # extra padding zero for normal R16G16B16A16
+    # position texture normal tangent
+    stride = struct.pack("<" + "fff" + "ff" + "eeee" + "eeee", *vertex, 0, *tangent)
+    print(vertex, tangent)
+    #print(len(stride))
+    f.write(stride)
+
+minv = [0, 0, 0]
+maxv = [0, 0, 0]
+for v in vertex_buffer:
+    for i in range(3):
+        if v[i] > maxv[i]:
+            maxv[i] = v[i]
+        if v[i] < minv[i]:
+            minv[i] = v[i]
+
+with open(sys.argv[2], 'wb') as f:
+
+    header_size = 4 * 5 + 4 * 6
+
+    index_size = 2 * 1
+    index_buffer_size = index_size * len(index_buffer)
+
+    vertex_size = (4 * 3) + (4 * 2) + (2 * 4) + (2 * 4)
+    vertex_buffer_size = vertex_size * len(vertex_buffer)
+
+    index_count = len(index_buffer)
+    index_buffer_offset = header_size
+    vertex_buffer_offset = header_size + index_buffer_size
+
+    tangents = tangent.generate_tangents(vertex_buffer, index_buffer)
+
+    header = struct.pack("<IIIIIffffff",
+                         index_count,
+                         index_buffer_offset,
+                         index_buffer_size,
+                         vertex_buffer_offset,
+                         vertex_buffer_size,
+                         *minv,
+                         *maxv)
+    f.write(header)
+
+    assert f.tell() == index_buffer_offset
+    for index in index_buffer:
+        print(index, vertex_buffer[index])
+        f.write(struct.pack("<H", index))
+
+    assert f.tell() == vertex_buffer_offset
+
+    for vertex, tangent in zip(vertex_buffer, tangents):
+        render_vertex(f, vertex, tangent)
+
+    assert f.tell() == vertex_buffer_offset + vertex_buffer_size

tools/obj/tangent.py

@@ -0,0 +1,81 @@
+import numpy as np
+
+PNT = list[tuple[float, float, float, # position
+                 float, float, # texture
+                 float, float, float, float]] # normal
+
+def get_position(vertex_buffer: list[PNT], index):
+    return vertex_buffer[index][0:3]
+
+def get_texture(vertex_buffer: list[PNT], index):
+    return vertex_buffer[index][3:5]
+
+def get_normal(vertex_buffer: list[PNT], index):
+    return vertex_buffer[index][5:8]
+
+def normalize(v):
+    return v / np.linalg.norm(v)
+
+def generate_tangents(vertex_buffer: list[PNT],
+                      index_buffer: list[int]):
+    vertex_count = len(vertex_buffer)
+
+    triangle_count = len(index_buffer) // 3
+
+    tan_s = np.zeros((vertex_count, 3,))
+    tan_t = np.zeros((vertex_count, 3,))
+
+    for t in range(triangle_count):
+        i0 = index_buffer[t * 3 + 0]
+        i1 = index_buffer[t * 3 + 1]
+        i2 = index_buffer[t * 3 + 2]
+
+        p0 = get_position(vertex_buffer, i0)
+        p1 = get_position(vertex_buffer, i1)
+        p2 = get_position(vertex_buffer, i2)
+
+        uv0 = get_texture(vertex_buffer, i0)
+        uv1 = get_texture(vertex_buffer, i1)
+        uv2 = get_texture(vertex_buffer, i2)
+
+        x1 = p1[0] - p0[0]
+        x2 = p2[0] - p0[0]
+        y1 = p1[1] - p0[1]
+        y2 = p2[1] - p0[1]
+        z1 = p1[2] - p0[2]
+        z2 = p2[2] - p0[2]
+
+        s1 = uv1[0] - uv0[0]
+        s2 = uv2[0] - uv0[0]
+
+        t1 = uv1[1] - uv0[1]
+        t2 = uv2[1] - uv0[1]
+
+        d = (s1 * t2 - s2 * t1)
+        r = 1.0 / d if d != 0 else 0
+
+        sdir = np.array([(t2 * x1 - t1 * x2) * r,
+                         (t2 * y1 - t1 * y2) * r,
+                         (t2 * z1 - t1 * z2) * r])
+
+        tdir = np.array([(s1 * x2 - s2 * x1) * r,
+                         (s1 * y2 - s2 * y1) * r,
+                         (s1 * z2 - s2 * z1) * r])
+
+        tan_s += sdir
+        tan_t += tdir
+
+    tangents = []
+    for v in range(vertex_count):
+        n = np.array(get_normal(vertex_buffer, v))
+        s = tan_s[v]
+
+        # orthagonalize
+        tangent = normalize(s - n * np.dot(n, s))
+        h = np.dot(np.cross(n, s), tan_t[v])
+        w = -1.0 if h < 0.0 else 1.0
+        tangent_w = (tangent[0], tangent[1], tangent[2], w)
+        #print(tangent_w)
+        tangents.append(tangent_w)
+
+    return tangents