Added missing AVX2 fillers

[itzpr3d4t0r]

This PR is a continuation of #2382 and adds the SUB, MIN, MAX, and MULT blend flags. this should be a 70X to 110X perf improvement over the current fillers with blend flags and is 35% faster than caching a surface with a color and blitting (with AVX2).

Results and test program:
ON MAIN

Flag: BLEND_SUB
fill: 1.8767017999998643
blit: 0.021530719999827853
--------------------
Flag: BLEND_MULT
fill: 1.7614726199999495
blit: 0.03417129999997996
--------------------
Flag: BLEND_MIN
fill: 1.82431628000013
blit: 0.02125239999986661
--------------------
Flag: BLEND_MAX
fill: 1.872223340000346
blit: 0.021590140000080284
--------------------

WITH THIS PR

Flag: BLEND_SUB
fill: 0.01598090000006778
blit: 0.021626579999974638
--------------------
Flag: BLEND_MULT
fill: 0.024974719999954688
blit: 0.03429605999972409
--------------------
Flag: BLEND_MIN
fill: 0.01568092000015895
blit: 0.021365460000197345
--------------------
Flag: BLEND_MAX
fill: 0.015678200000002106
blit: 0.021505839999917953
--------------------

Test Program

from timeit import repeat

import pygame

pygame.init()

surf = pygame.Surface((500, 500))
surf.fill((132, 33, 200))

color = pygame.Surface((500, 500))
color.fill((24, 24, 24))

flags = [
    "BLEND_SUB",
    "BLEND_MULT",
    "BLEND_MIN",
    "BLEND_MAX",
]

G = globals()

for flag in flags:
    print(f"Flag: {flag}")
    teststr = "surf.fill((24, 24, 24), None, pygame." + flag + ")"
    l = [min(repeat(teststr, globals=G, number=1000, repeat=10)) for _ in range(5)]
    print(f"fill: {sum(l) / len(l)}")

    teststr = "surf.blit(color, (0, 0), None, pygame." + flag + ")"
    l = [min(repeat(teststr, globals=G, number=1000, repeat=10)) for _ in range(5)]
    print(f"blit: {sum(l) / len(l)}")
    print("-" * 20)

[MyreMylar]

LGTM 👍

I see the same speed up locally, the tests also all pass locally and the code all makes sense to me - macros very similar to the SSE2 versions from the other PR. Nice work!

[Starbuck5]

Looks good!

Thanks for working this up.