Enable remove-dead-values MLIR pass for stablehlo pipeline

[uazizTT]

Enable both remove-dead-values pass by default.

[nsmithtt]

Minor nit otherwise looks good

[mrakitaTT]

Btw just curious, can you please post some example StableHLO graph (or resulting TTIR graph) where you've found the sccp optimization useful?

[uazizTT]

Btw just curious, can you please post some example StableHLO graph (or resulting TTIR graph) where you've found the sccp optimization useful?

I found a couple of opportunities in a Llama model where SCCP will be beneficial, so I created a MLIR test from it, but it seems that SCCP pass is not working even for a simple case. We can merge this patch to have a placeholder in case someone is looking into it in future or else can hold it until me or someone else can figure out why SCCP is not working.

[nsmithtt]

Btw just curious, can you please post some example StableHLO graph (or resulting TTIR graph) where you've found the sccp optimization useful?

I found a couple of opportunities in a Llama model where SCCP will be beneficial, so I created a MLIR test from it, but it seems that SCCP pass is not working even for a simple case. We can merge this patch to have a placeholder in case someone is looking into it in future or else can hold it until me or someone else can figure out why SCCP is not working.

So is it used/working for any case that we have right now? I'd prefer to just do this work before landing this PR unless this PR is blocking something else. We should strive to land it in a working state.