bzm3r

# -*- coding: utf-8 -*-
"""
Created on Mon Mar 30 21:49:55 2015

@author: brian
"""

import PyDSTool as dst
from PyDSTool.Toolbox import phaseplane as pp
from matplotlib import pyplot as plt

bzm3r

Despite the ubiquity of text in computing applications, rendering high quality text using the GPU remains a challenging problem. Brief histories of the problem, and work done to date are discussed in this course description, and a recent blog post by Raph Levien and its accompanying HN discussion.

Patrick Walton et al. have produced a promising solution in the form of a Rust library, Pathfinder, whose general design has been beautifully documented by Nicolas Silva. Pathfinder 3 currently uses OpenGL as its GPU API, and we wondered if a port of it to more modern graphics APIs (e.g. Vulkan) might prove interesting in terms of performance results. To this end, I tried to use gfx-rs re-write how Pathfinder accesses/uses the GPU. While [progress was

bzm3r

struct StackElement {
    PietGroupRef group;
    uint index;
    float2 offset; // Maybe pack as short2?
}

kernel1(Buf scene, PietGroupRef root) {
    StackElement stack[MAX_STACK];
    uint stack_ix = 0;
    uint tos_group = root;

bzm3r

import numpy as np
from copy import deepcopy

def subgroup_shuffle_xor(caller_tid, values, mask):
    rd_tid = caller_tid^mask
    return values[rd_tid]


def bnot(n, nbits):
    return (1 << nbits) - 1 - n

bzm3r

#version 450
layout(local_size_x = 32) in;

layout(set = 0, binding = 0) buffer BM {
    uint[32] bm;
};

inline uint shuffle_round(uint a, uint b, uint m, uint s) {
    uint c;
    if ((tix & s) == 0) {

bzm3r

#version 450

#extension GL_KHR_shader_subgroup_shuffle: enable

layout(local_size_x = 32) in;

layout(set = 0, binding = 0) buffer BM {
    uint[32] bm;
};

bzm3r

extern crate gfx_hal;

use std::{fs, ptr, slice, str::FromStr};

use hal::{adapter::MemoryType, buffer, command, memory, pool, prelude::*, pso};

pub struct DispatchTime(f64);

pub fn execute_test<B>(test_kernel: String, input_bm: [u32; 32], num_executions: usize) -> DispatchTime {
    #[cfg(debug_assertions)]

bzm3r

#version 450
#define NUM_BIT_MATRICES ~NUM_BMS~
#define WORKGROUP_SIZE ~WG_SIZE~

layout(local_size_x = WORKGROUP_SIZE, local_size_y = 32) in;

layout(set = 0, binding = 0) buffer BM {
    uint num_bms;
    uint[32] bms[NUM_BIT_MATRICES];
};

bzm3r

#version 450
#extension GL_KHR_shader_subgroup_shuffle: enable

#define WORKGROUP_SIZE ~WG_SIZE~
#define NUM_EXECUTIONS ~NUM_EXECS~

// Unlike the threadgroup case, the Y-dimension of the workgroup size is not used.
// This is because the Y-dimension will be implicit in the number of subgroups in a workgroup.
layout(local_size_x = WORKGROUP_SIZE) in;

bzm3r

#version 450
#extension GL_KHR_shader_subgroup_ballot: enable

#define WORKGROUP_SIZE ~WG_SIZE~
#define NUM_EXECUTIONS ~NUM_EXECS~

layout(local_size_x = WORKGROUP_SIZE) in;

layout(set = 0, binding = 0) buffer BM {
    uint[32] bms[];