[llvm] [AMDGPU] Remove AMDGPUISD::FFBH_I32 and add ISD::CTLS lowering (PR #187694)

[Matt Arsenault via llvm-commits]

================
@@ -1385,10 +1385,10 @@ AMDGPULegalizerInfo::AMDGPULegalizerInfo(const GCNSubtarget &ST_,
       .widenScalarToNextPow2(1, 32);
 
   getActionDefinitionsBuilder(G_CTLS)
-      .customFor({{S32, S32}})
+      .scalarize(0)
----------------
arsenm wrote:

None of these actions mutate the instructions. It informs the legalizer what action to perform for that operation. The legalizer will incrementally progress the function. Later transformations are starting fresh on a mutated instruction 

https://github.com/llvm/llvm-project/pull/187694