Taichi的GPU并行支持虚函数否

太极图形课专区 太极图形课答疑区
#1

请问，Taichi的GPU并行支持虚函数的调用吗？若支持，是如何实现的？

#2

Taichi可以支持虚函数的调用，可以参考我的代码：

# Python program showing
# Taichi abstract base class work
import taichi as ti
from abc import ABC, abstractmethod

ti.init(arch=ti.cuda)


@ti.data_oriented
class Polygon(ABC):
    @ti.kernel
    @abstractmethod
    def noofsides(self):
        pass


@ti.data_oriented
class Triangle(Polygon):
    def __init__(self):
        super().__init__()
        self.pos = ti.field(ti.f32, shape=3)
        self.init()

    @ti.kernel
    def init(self):
        for i in self.pos:
            self.pos[i] = i


# overriding abstract method

    @ti.kernel
    def noofsides(self):
        for i in self.pos:
            print(self.pos[i])
        print("I have 3 sides")


@ti.data_oriented
class Pentagon(Polygon):
    def __init__(self):
        super().__init__()
        self.vel = ti.field(ti.f32, shape=5)
        self.init()

    @ti.kernel
    def init(self):
        for i in self.vel:
            self.vel[i] = i

    # overriding abstract method
    @ti.kernel
    def noofsides(self):
        for i in self.vel:
            print(self.vel[i])
        print("I have 5 sides")


# Driver code
R = Triangle()
R.noofsides()

R = Pentagon()
R.noofsides()

关于Taichi中kernel函数的并行实现，可以参考：here

3 个赞