[UT] enable common loop fusion optimization

[quintinwang5]

FuseNestedLoopsPass is a common pass independent with backend, so we just enable it in our backend.

[whitneywhtsang]

What is the performance impact? Is there a cost model inside the pass, or is it only enabled by user?

[quintinwang5]

What is the performance impact? Is there a cost model inside the pass, or is it only enabled by user?

Now, it's controled by a fronted flag flatten=True. From the original PR, the performance is slightly better.

[whitneywhtsang]

What is the performance impact? Is there a cost model inside the pass, or is it only enabled by user?

Now, it's controled by a fronted flag flatten=True. From the original PR, the performance is slightly better.

icic, can you please measure performance impact of this PR, as this PR also adds two other passes in the pipeline?

[quintinwang5]

What is the performance impact? Is there a cost model inside the pass, or is it only enabled by user?

Now, it's controled by a fronted flag flatten=True. From the original PR, the performance is slightly better.

icic, can you please measure performance impact of this PR, as this PR also adds two other passes in the pipeline?

What's the scope of the performance test, just benchmarks?

[whitneywhtsang]

What is the performance impact? Is there a cost model inside the pass, or is it only enabled by user?

Now, it's controled by a fronted flag flatten=True. From the original PR, the performance is slightly better.

icic, can you please measure performance impact of this PR, as this PR also adds two other passes in the pipeline?

What's the scope of the performance test, just benchmarks?

I am only expecting performance of benchmarks, you can use CI to launch it.

[quintinwang5]

What is the performance impact? Is there a cost model inside the pass, or is it only enabled by user?

Now, it's controled by a fronted flag flatten=True. From the original PR, the performance is slightly better.

icic, can you please measure performance impact of this PR, as this PR also adds two other passes in the pipeline?

What's the scope of the performance test, just benchmarks?

I am only expecting performance of benchmarks, you can use CI to launch it.

It's hard to draw a simple conclusion. Because we have some cases better but some worse. So I'll try another CI without add_licm