[Gluon] Add SCCP to the compiler pipeline (#7061)

This mainly helps simplify loop arithmetic (phases, indices, etc.)

Author: Mogball

third_party/amd/backend/compiler.py

@@ -280,6 +280,7 @@ def ttgir_opt(src, metadata, options):
         pm.enable_debug()
 
         passes.ttgpuir.add_inliner(pm)
+        passes.common.add_sccp(pm)
         passes.ttir.add_loop_aware_cse(pm)
         passes.ttgpuir.add_canonicalizer(pm)
         passes.ttgpuir.add_combine_tensor_select_and_if(pm)

third_party/nvidia/backend/compiler.py

@@ -304,6 +304,7 @@ def ttgir_opt(self, src, metadata, options, capability):
         pm.enable_debug()
 
         passes.ttgpuir.add_inliner(pm)
+        passes.common.add_sccp(pm)
         passes.ttir.add_loop_aware_cse(pm)
         passes.ttgpuir.add_canonicalizer(pm)
         passes.ttgpuir.add_combine_tensor_select_and_if(pm)