Yes, I realize the same question that since the auto_scheduler do the fusion,
the DAG op can't correspond to thoes we wrote in TE and TOPI.But the printed
schedule use the DAG op,so we can't simply match or reuse the printed schedule.
However, when I look into the DAG, I find it doesn't change a lot from TE.So if
I write a new TE according to the DAG,the schedule can work for it, is that
right?
And here's the DAG:
> A = PLACEHOLDER [1, 3, 5, 5]
> PadInput(i0, i1, i2, i3) = tir.if_then_else(((((i2 >= 2) && (i2 < 7)) &&
(i3 >= 2)) && (i3 < 7)), A[i0, i1, (i2 - 2), (i3 - 2)], 10000f)
> T_reshape(ax0, ax1, ax2, ax3) = PadInput[0,
floormod(floordiv(floordiv((((((ax0 + ax1)*9) + ax2)*9) + ax3), 9), 9), 3),
floormod(floordiv((((((ax0 + ax1)*9) + ax2)*9) + ax3), 9), 9),
floormod((((((ax0 + ax1)*9) + ax2)*9) + ax3), 9)]
> pad_temp(i0, i1, i2, i3) = T_reshape[i0, i1, i2, i3]
> temp(i, j) = tir.if_then_else((i == j), 1, 0)
> T_reshape(ax0, ax1, ax2, ax3) = temp[floormod(floordiv((((((ax0 + ax1)*5)
+ ax2)*5) + ax3), 25), 25), floormod((((((ax0 + ax1)*5) + ax2)*5) + ax3), 25)]
> compute(nn, ff, yy, xx) += (pad_temp[nn, rc, (yy + ry), (xx +
rx)]*float32(T_reshape[ff, rc, ry, rx]))
> compute_red(ax0, ax1, ax2) min= compute[ax0, k1, ax1, ax2]
> T_reshape(ax0, ax1, ax2, ax3) =
compute_red[floormod(floordiv(floordiv(((((((ax0*3) + ax1)*5) + ax2)*5) + ax3),
5), 5), 3), floormod(floordiv(((((((ax0*3) + ax1)*5) + ax2)*5) + ax3), 5), 5),
floormod(((((((ax0*3) + ax1)*5) + ax2)*5) + ax3), 5)]
---
[Visit
Topic](https://discuss.tvm.apache.org/t/print-auto-schedule-python-schedule-with-topi-op/11363/6)
to respond.
You are receiving this because you enabled mailing list mode.
To unsubscribe from these emails, [click
here](https://discuss.tvm.apache.org/email/unsubscribe/3da6fffa805ac00efa5747e185daea8e398fd8ca4288e44916abd173ee4f480d).
