Start image rendering

Populates the piet-gpu subdir, with an extremely simple renderer. The
main program saves the image to a PNG.

Contains a few fixes (I was confused about the need for multiple
bindings, as opposed to multiple descriptors within a binding).

Cargo.lock

@@ -1,5 +1,11 @@
 # This file is automatically @generated by Cargo.
 # It is not intended for manual editing.
+[[package]]
+name = "adler32"
+version = "1.0.4"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "5d2e7343e7fc9de883d1b0341e0b13970f764c14101234857d2ddafa1cb1cac2"
+
 [[package]]
 name = "ash"
 version = "0.30.0"
@@ -9,12 +15,58 @@ dependencies = [
  "libloading",
 ]
 
+[[package]]
+name = "bitflags"
+version = "1.2.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "cf1de2fe8c75bc145a2f577add951f8134889b4795d47466a54a5c846d691693"
+
+[[package]]
+name = "byteorder"
+version = "1.3.4"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "08c48aae112d48ed9f069b33538ea9e3e90aa263cfa3d1c24309612b1f7472de"
+
 [[package]]
 name = "cc"
 version = "1.0.50"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "95e28fa049fda1c330bcf9d723be7663a899c4679724b34c81e9f5a326aab8cd"
 
+[[package]]
+name = "cfg-if"
+version = "0.1.10"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "4785bdd1c96b2a846b2bd7cc02e86b6b3dbf14e7e53446c4f54c92a361040822"
+
+[[package]]
+name = "crc32fast"
+version = "1.2.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "ba125de2af0df55319f41944744ad91c71113bf74a4646efff39afe1f6842db1"
+dependencies = [
+ "cfg-if",
+]
+
+[[package]]
+name = "deflate"
+version = "0.8.4"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "e7e5d2a2273fed52a7f947ee55b092c4057025d7a3e04e5ecdbd25d6c3fb1bd7"
+dependencies = [
+ "adler32",
+ "byteorder",
+]
+
+[[package]]
+name = "inflate"
+version = "0.4.5"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "1cdb29978cc5797bd8dcc8e5bf7de604891df2a8dc576973d71a281e916db2ff"
+dependencies = [
+ "adler32",
+]
+
 [[package]]
 name = "libloading"
 version = "0.5.2"
@@ -25,6 +77,14 @@ dependencies = [
  "winapi",
 ]
 
+[[package]]
+name = "piet-gpu"
+version = "0.1.0"
+dependencies = [
+ "piet-gpu-hal",
+ "png",
+]
+
 [[package]]
 name = "piet-gpu-derive"
 version = "0.0.0"
@@ -41,6 +101,18 @@ dependencies = [
  "ash",
 ]
 
+[[package]]
+name = "png"
+version = "0.16.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "910f09135b1ed14bb16be445a8c23ddf0777eca485fbfc7cee00d81fecab158a"
+dependencies = [
+ "bitflags",
+ "crc32fast",
+ "deflate",
+ "inflate",
+]
+
 [[package]]
 name = "proc-macro2"
 version = "1.0.10"

Cargo.toml

@@ -1,6 +1,7 @@
 [workspace]
 
 members = [
+    "piet-gpu",
     "piet-gpu-derive",
     "piet-gpu-hal"
 ]

piet-gpu-hal/src/lib.rs

@@ -74,6 +74,6 @@ pub trait CmdBuf<D: Device> {
     unsafe fn write_timestamp(&mut self, pool: &D::QueryPool, query: u32);
 }
 
-pub trait MemFlags: Sized {
+pub trait MemFlags: Sized + Clone + Copy {
     fn host_coherent() -> Self;
 }

piet-gpu-hal/src/vulkan.rs

@@ -58,6 +58,7 @@ pub struct QueryPool {
     n_queries: u32,
 }
 
+#[derive(Clone, Copy)]
 pub struct MemFlags(vk::MemoryPropertyFlags);
 
 impl VkInstance {
@@ -169,23 +170,25 @@ impl crate::Device for VkDevice {
 
     /// This creates a pipeline that runs over the buffer.
     ///
-    /// The code is included from "../comp.spv", and the descriptor set layout is just some
-    /// number of buffers.
+    /// The descriptor set layout is just some number of buffers (this will change).
     unsafe fn create_simple_compute_pipeline(
         &self,
         code: &[u8],
         n_buffers: u32,
     ) -> Result<Pipeline, Error> {
         let device = &self.device.device;
-        let descriptor_set_layout = device.create_descriptor_set_layout(
-            &vk::DescriptorSetLayoutCreateInfo::builder().bindings(&[
+        let bindings = (0..n_buffers)
+            .map(|i| {
                 vk::DescriptorSetLayoutBinding::builder()
-                    .binding(0)
+                    .binding(i)
                     .descriptor_type(vk::DescriptorType::STORAGE_BUFFER)
-                    .descriptor_count(n_buffers)
+                    .descriptor_count(1)
                     .stage_flags(vk::ShaderStageFlags::COMPUTE)
-                    .build(),
-            ]),
+                    .build()
+            })
+            .collect::<Vec<_>>();
+        let descriptor_set_layout = device.create_descriptor_set_layout(
+            &vk::DescriptorSetLayoutCreateInfo::builder().bindings(&bindings),
             None,
         )?;
 
@@ -248,25 +251,22 @@ impl crate::Device for VkDevice {
                     .set_layouts(&descriptor_set_layouts),
             )
             .unwrap();
-        let buf_infos = bufs
-            .iter()
-            .map(|buf| {
-                vk::DescriptorBufferInfo::builder()
-                    .buffer(buf.buffer)
-                    .offset(0)
-                    .range(vk::WHOLE_SIZE)
-                    .build()
-            })
-            .collect::<Vec<_>>();
-        device.update_descriptor_sets(
-            &[vk::WriteDescriptorSet::builder()
-                .dst_set(descriptor_sets[0])
-                .dst_binding(0)
-                .descriptor_type(vk::DescriptorType::STORAGE_BUFFER)
-                .buffer_info(&buf_infos)
-                .build()],
-            &[],
-        );
+        for (i, buf) in bufs.iter().enumerate() {
+            let buf_info = vk::DescriptorBufferInfo::builder()
+                .buffer(buf.buffer)
+                .offset(0)
+                .range(vk::WHOLE_SIZE)
+                .build();
+            device.update_descriptor_sets(
+                &[vk::WriteDescriptorSet::builder()
+                    .dst_set(descriptor_sets[0])
+                    .dst_binding(i as u32)
+                    .descriptor_type(vk::DescriptorType::STORAGE_BUFFER)
+                    .buffer_info(&[buf_info])
+                    .build()],
+                &[],
+            );
+        }
         Ok(DescriptorSet {
             descriptor_set: descriptor_sets[0],
         })
@@ -321,7 +321,10 @@ impl crate::Device for VkDevice {
         device.destroy_query_pool(pool.pool, None);
         let ts0 = buf[0];
         let tsp = self.timestamp_period as f64 * 1e-9;
-        let result = buf[1..].iter().map(|ts| ts.wrapping_sub(ts0) as f64 * tsp).collect();
+        let result = buf[1..]
+            .iter()
+            .map(|ts| ts.wrapping_sub(ts0) as f64 * tsp)
+            .collect();
         Ok(result)
     }
 
@@ -354,7 +357,7 @@ impl crate::Device for VkDevice {
         result: &mut Vec<T>,
     ) -> Result<(), Error> {
         let device = &self.device.device;
-        let size = buffer.size as usize;
+        let size = buffer.size as usize / std::mem::size_of::<T>();
         let buf = device.map_memory(
             buffer.buffer_memory,
             0,

piet-gpu/Cargo.toml

@@ -0,0 +1,13 @@
+[package]
+name = "piet-gpu"
+version = "0.1.0"
+authors = ["Raph Levien <raph.levien@gmail.com>"]
+description = "A compute-centric GPU 2D renderer."
+license = "MIT/Apache-2.0"
+edition = "2018"
+
+[dependencies.piet-gpu-hal]
+path = "../piet-gpu-hal"
+
+[dependencies]
+png = "0.16.2"

piet-gpu/shader/build.ninja

@@ -0,0 +1,10 @@
+# Build file for shaders.
+
+# You must have glslangValidator in your path, or patch here.
+
+glslang_validator = glslangValidator
+
+rule glsl
+  command = $glslang_validator -V -o $out $in
+
+build image.spv: glsl image.comp

piet-gpu/shader/image.comp

@@ -0,0 +1,31 @@
+// A simple kernel to create an image.
+
+// Right now, this kernel stores the image in a buffer, but a better
+// plan is to use a texture. This is because of limited support.
+
+#version 450
+layout(local_size_x = 16, local_size_y = 16) in;
+
+layout(set = 0, binding = 0) readonly buffer SceneBuf {
+    uint[] scene;
+};
+
+layout(set = 0, binding = 1) buffer ImageBuf {
+    uint[] image;
+};
+
+// TODO: make the image size dynamic.
+#define IMAGE_WIDTH 2048
+#define IMAGE_HEIGHT 1535
+
+void main() {
+    uvec2 xy = gl_GlobalInvocationID.xy;
+    vec2 uv = vec2(xy) * vec2(1.0 / IMAGE_WIDTH, 1.0 / IMAGE_HEIGHT);
+    vec4 rgba = vec4(uv.xyy, 1.0);
+    uvec4 s = uvec4(round(rgba * 255.0));
+    uint rgba_packed = s.x | (s.y << 8) | (s.z << 16) | (s.w << 24);
+    image[xy.y * IMAGE_WIDTH + xy.x] = rgba_packed;
+    if (xy.y == 0 && xy.x < 8) {
+        image[xy.x] = scene[xy.x];
+    }
+}

piet-gpu/src/main.rs

@@ -0,0 +1,63 @@
+use std::path::Path;
+use std::fs::File;
+use std::io::BufWriter;
+
+use piet_gpu_hal::vulkan::VkInstance;
+use piet_gpu_hal::{CmdBuf, Device, MemFlags};
+
+const WIDTH: usize = 2048;
+const HEIGHT: usize = 1536;
+
+const TILE_W: usize = 16;
+const TILE_H: usize = 16;
+
+fn main() {
+    let instance = VkInstance::new().unwrap();
+    unsafe {
+        let device = instance.device().unwrap();
+        let mem_flags = MemFlags::host_coherent();
+        let src = (0..256).map(|x| x + 1).collect::<Vec<u32>>();
+        let scene_buf = device
+            .create_buffer(std::mem::size_of_val(&src[..]) as u64, mem_flags)
+            .unwrap();
+        device.write_buffer(&scene_buf, &src).unwrap();
+        let image_buf = device
+            .create_buffer((WIDTH * HEIGHT * 4) as u64, mem_flags)
+            .unwrap();
+        let code = include_bytes!("../shader/image.spv");
+        let pipeline = device.create_simple_compute_pipeline(code, 2).unwrap();
+        let descriptor_set = device
+            .create_descriptor_set(&pipeline, &[&scene_buf, &image_buf])
+            .unwrap();
+        let query_pool = device.create_query_pool(2).unwrap();
+        let mut cmd_buf = device.create_cmd_buf().unwrap();
+        cmd_buf.begin();
+        cmd_buf.write_timestamp(&query_pool, 0);
+        cmd_buf.dispatch(
+            &pipeline,
+            &descriptor_set,
+            ((WIDTH / TILE_W) as u32, (HEIGHT / TILE_H) as u32, 1),
+        );
+        cmd_buf.write_timestamp(&query_pool, 1);
+        cmd_buf.finish();
+        device.run_cmd_buf(&cmd_buf).unwrap();
+        let timestamps = device.reap_query_pool(query_pool).unwrap();
+        println!("Render time: {:.3}ms", timestamps[0] * 1e3);
+        let mut img_data: Vec<u8> = Default::default();
+        // Note: because png can use a `&[u8]` slice, we could avoid an extra copy
+        // (probably passing a slice into a closure). But for now: keep it simple.
+        device.read_buffer(&image_buf, &mut img_data).unwrap();
+
+        // Write image as PNG file.
+        let path = Path::new("image.png");
+        let file = File::create(path).unwrap();
+        let ref mut w = BufWriter::new(file);
+        
+        let mut encoder = png::Encoder::new(w, WIDTH as u32, HEIGHT as u32);
+        encoder.set_color(png::ColorType::RGBA);
+        encoder.set_depth(png::BitDepth::Eight);
+        let mut writer = encoder.write_header().unwrap();
+        
+        writer.write_image_data(&img_data).unwrap();
+    }
+}