OpenGL - 从不同角度渲染相同的场景，无需重绘所有内容

Question

I have a scene that I'm trying to render in all directions (later stitched together, outside OpenGL).

Currently for creating each frame - I redraw the entire scene 4 times (front, right, back, left). Can I somehow just render the 4 cubes together without redrawing the entire scene 4 times?

Here is a sample code that I use (Specifically I use PyOpenGL - But that doesn't really matter):

import cv2
import numpy as np
from OpenGL.GL import *
from OpenGL.GLU import *
from OpenGL.GLUT import *
import glfw

# Draws some model (The real code draws a much more complicated model)
def draw_model():
    glBegin(GL_QUADS)
    glColor3f(1.0, 1.0, 1.0)
    glVertex(10, 10, 0); glVertex(10, 10, 3); glVertex(10, -10, 3); glVertex(10, -10, 0)

    glColor3f(1.0, 1.0, 0)
    glVertex(-10, 10, 0); glVertex(-10, 10, 3); glVertex(-10, -10, 3); glVertex(-10, -10, 0)

    glColor3f(1.0, 0, 1.0)
    glVertex(10, 10, 0); glVertex(10, 10, 3); glVertex(-10, 10, 3); glVertex(-10, 10, 0)

    glColor3f(0, 1.0, 0)
    glVertex(10, -10, 0); glVertex(10, -10, 3); glVertex(-10, -10, 3); glVertex(-10, -10, 0)
    glEnd()

# Reads the pixels to NP
def get_display_pixels(rendered_image_width, rendered_image_height):
    data = glReadPixels(0, 0, rendered_image_width, rendered_image_height, OpenGL.GL.GL_RGB, OpenGL.GL.GL_UNSIGNED_BYTE)
    return np.frombuffer(data, dtype=np.uint8).reshape(rendered_image_height, rendered_image_width, 3)[::-1]

DISPLAY_WIDTH = 900
DISPLAY_HEIGHT = 900

glfw.init()
glfw.window_hint(glfw.VISIBLE, False)
window = glfw.create_window(DISPLAY_WIDTH, DISPLAY_HEIGHT, "some window", None, None)
glfw.make_context_current(window)

gluPerspective(90, (DISPLAY_WIDTH / DISPLAY_HEIGHT), 0.01, 30)
glEnable(GL_TEXTURE_2D)
glEnable(GL_DEPTH_TEST)
glDepthFunc(GL_LEQUAL)

position = (0, 3, 1)
# Get cube 1
glPushMatrix()
glClear(GL_COLOR_BUFFER_BIT | GL_DEPTH_BUFFER_BIT)
gluLookAt(*position, 1, 4, 1, 0, 0, 1)
draw_model()
cube1 = get_display_pixels(DISPLAY_WIDTH, DISPLAY_HEIGHT)
glPopMatrix()

# Get cube 2
glPushMatrix()
glClear(GL_COLOR_BUFFER_BIT | GL_DEPTH_BUFFER_BIT)
gluLookAt(*position, 1, 2, 1, 0, 0, 1)
draw_model()
cube2 = get_display_pixels(DISPLAY_WIDTH, DISPLAY_HEIGHT)
glPopMatrix()

# Get cube 3
glPushMatrix()
glClear(GL_COLOR_BUFFER_BIT | GL_DEPTH_BUFFER_BIT)
gluLookAt(*position, -1, 2, 1, 0, 0, 1)
draw_model()
cube3 = get_display_pixels(DISPLAY_WIDTH, DISPLAY_HEIGHT)
glPopMatrix()

# Get cube 4
glPushMatrix()
glClear(GL_COLOR_BUFFER_BIT | GL_DEPTH_BUFFER_BIT)
gluLookAt(*position, -1, 4, 1, 0, 0, 1)
draw_model()
cube4 = get_display_pixels(DISPLAY_WIDTH, DISPLAY_HEIGHT)
glPopMatrix()

cv2.imwrite(r"C:\temp\image1.png", cube1)
cv2.imwrite(r"C:\temp\image2.png", cube2)
cv2.imwrite(r"C:\temp\image3.png", cube3)
cv2.imwrite(r"C:\temp\image4.png", cube4)

glfw.destroy_window(window)
glfw.terminate()

Answer 1

There are ways how one could render four different views at once like Layered Rendering , but that still means that everything after the geometry shader stage is executed once per layer. This could improve the speed when a lot of work is done in the vertex shader stage (or in Tessellation Shader), but won't change much when the bottleneck is the pixel fill rate.

There is no way to render the same scene from a different without any redraw. The whole rasterization (and clipping, etc.) operates in NDC space, which is usually after applying the view and projection matrix, so at least those stages have to be performed once per view.

_{I'm also not sure if layered rendering goes together with the ancient version of OpenGL you are using. The fixed function pipeline is deprecated since a decade and should be avoided as much as possible.}

Answer 2

The solution was achieved using the idea by "Rabbid76" in the comment.

Moving all of the vertices to a VBO and then drawing them all at once gives the needed performance improvement