tree 9c885fa50192b7b6dc90ec4afa19d4bdbabb90af
parent 563b3e73c126a56dcabc8d2b17bf6a27347e37ff
author Nirvedh Meshram <96096277+nirvedhmeshram@users.noreply.github.com> 1729715660 -0500
committer GitHub <noreply@github.com> 1729715660 -0500
gpgsig -----BEGIN PGP SIGNATURE-----
 
 wsFcBAABCAAQBQJnGV3MCRC1aQ7uu5UhlAAA8BAQAFrrxr9KxyUEuHncJaT4It2M
 WFEazUuRO82VXDR71Fox4lwu2KR8Kqx5bsN12Gnldn7+ZzBIwhho4waaq7uebtvN
 9c9ECWCWI/ABSW++1F70LeDHlW8WXvhN7Xv9CZLHbSKNMeSUVUrhl/WiHw8AQbv5
 77YsliUvSxRmKRA75EuCpW4RCdLupW49L6dJYMrbrQNLJxcIuCBqOmjUIrz5kdYK
 BQZbVWP1EUM5Sy2GlJ23gfCJ3igwMnxiQo4NhwHkCgry6Bv8ProeuP6IJvOL9sHg
 DdEBIQif23y3QmFBuk9SFtEO8KegLC07IxWnbwgmOI5f5JZ6lBAnFITOkYMnLAlk
 92CNyTrATrFlfUqctXnYwNILSMILzhbuNZEmakadnlZCQ31smzp4PQUn3t6jq8kf
 j+UxwUJosV1m3Au1nQxY02PQ4BvPyohwQMa2t5fjEt3GNnPkO7gQzpSx7yQLkuje
 cpPLyNTpV/TktrEty2v6GMTEZRJI5IUsaMLnVxP8si71dYZxMwBhFAXl2pinTZqG
 sl6l5QZ8DlCiAbA73egT3rXciiylLeRih0qK3IlyYkqGm5gb5LZwFNFo5Zitj1PU
 uKAY0cDboNiWSF5AzZ1cKguMIGW1kkHgH0TKip4bg6FC9mDSpitJnBMBGUlg3S5k
 do/iMFleCdG3saaHTJq9
 =Nd9V
 -----END PGP SIGNATURE-----
 

[ROCM][NFC] Add option to control SLP vectorization in llvm optimizations (#18865)

We keep SLP vectorization off because it can mask perf issues or create
regressions. However, on ROCM what we have noticed is that we are
hitting issues in several untested paths in the AMDGPU llvm backend
because we dont have SLP vectorization.
Here is an example of an issue that we wouldn't hit if SLP vectorization
was turned on
https://github.com/iree-org/iree/issues/18798

In this PR we are still keeping the exisiting behavior but provide a
flag to toggle it so that we can do the required benchmarking and
analysis.

Signed-off-by: Nirvedh <nirvedh@gmail.com>