RetroWarp

This is a straight forward Vulkan compute shader implementation of a fictional retro GPU. The GPU is implemented with "pure" software rendering, no fixed function units (rasterizer, ROP, texture units) of the target Vulkan GPU is used.

Build

Standard CMake build. Should build on GCC, Clang and MSVC 2015+.

Update submodules

This implementation makes use of my Granite engine. Make shure this is checked out before building.

git submodule update --init --recursive

viewer

A simple test program which renders a glTF 2.0 file in real-time. Don't expect good compatibility as all the meshes have to be translated into a SWRenderableComponent, which implements the bare minimum required to load some models. Some test models I've used are Sponza, Suzanne or Lantern from KhronosGroup/glTF-Sample-Models.

NOTE: Most likely, the application will be CPU bound as all vertex processing is done with unoptimized CPU code unless the "freeze" feature is used.

Controls

• WASD: Move camera around
• Hold right-click and move mouse: Rotate camera
• U: Freeze the frame, no vertex processing on CPU is done, which is useful for testing GPU bound scenario.
• C: Dumps the current frame to retrowarp.dump along with textures. This can be replayed and benchmarked in dump-bench.
• Space: Toggle vsync.

Options

• --width: Control resolution. Maximum is 2048.
• --height: Control resolution. Maximum is 2048.
• --tile-size: Tile size, use 8 or 16.
• --ubershader: Use ubershader rather than split shader architecture.
• --nosubgroup: Disable all subgroup support.
• --async-compute: Enable async compute support.

dump-bench

Takes a .dump file created by the viewer application and replays that frame over and over. At the end, performance metrics are reported in time / iteration (i.e. frame).

Options

• --tile-size: Tile size, use 8 or 16.
• --ubershader: Use ubershader rather than split shader architecture.
• --nosubgroup: Disable all subgroup support.
• --async-compute: Enable async compute support.
• --iterations: Number of iterations.

Resolution is specified in the dump as it contains post-triangle setup data and cannot be rescaled.

Dataset

There is a sample dataset for benchmarking in dataset/.

Implementation

Triangle processing

Triangle processing, clipping and setup is implemented in triangle_converter.hpp and triangle_converter.cpp.

The implementation is not designed to be fast, the focus here was on the rasterization part.

Rasterization

rasterizer_gpu.hpp and rasterizer_gpu.cpp implement the Vulkan side of things. Shaders are contained in assets/shaders.