# live multi-trial Protean dashboard demo
# accepted speedup: 3.06x
import triton
import triton.language as tl

@triton.jit
def fused_mlp_persistent_blocks_v3(x, w1, w2, y, n:tl.constexpr):
    pid = tl.program_id(0)
    offs = pid * 256 + tl.arange(0, 256)
    mask = offs < n
    vals = tl.load(x + offs, mask=mask, other=0.0)
    tl.store(y + offs, vals, mask=mask)