TMU load experiment

test.py

import numpy as np
import time

from videocore.assembler import qpu
from videocore.driver import Driver

@qpu
def hello_world(asm):
    mov(r0, uniform)
    ldi(r2, 3000000)

    L.loop
    mov(tmu0_s, r0)
    # 1. 
    # if we insert nops here, 0.6456s -> 0.6461s
    # for i in range(0, 4):
    #     nop()
    nop(sig='load tmu0')
    # 2.
    # if we insert nops here, 0.6456s -> 0.8060s
    # for i in range(0, 4):
    #     nop()
    isub(r2, r2, 1, set_flags=True)
    jzc(L.loop)
    nop()
    nop()
    nop()

    # Finish the thread
    exit()

with Driver() as drv:
    # Input vectors
    a = np.random.random(16).astype('float32')

    # Copy vectors to shared memory for DMA transfer
    inp = drv.copy(np.r_[a])

    # Run the program
    start = time.perf_counter()
    drv.execute(
        n_threads=1,
        program=drv.program(hello_world),
        uniforms=[inp.address]
    )
    end = time.perf_counter()

    print('{:.4f}'.format(end - start))