Thread enable-modulo-support through TritonToLinalgExperimental (#40)

TritonToStructured accepts enable-modulo-support (since f2eb8c5), but
TritonToLinalgExperimental never declared the option and never forwarded
it to the inner createTritonToStructuredPass() call, so
--triton-to-linalg-experimental=enable-modulo-support=false errored with
"no such option" instead of disabling the modulo split-pointer path.

Author: vpandya-quic

include/triton-shared/Conversion/TritonToLinalgExperimental/Passes.td

@@ -16,6 +16,8 @@ def TritonToLinalgExperimental : Pass<"triton-to-linalg-experimental", "mlir::Mo
   let options = [
       Option<"enableMakeGatherScatterTensorPtr", "enable-make-gather-scatter", "bool", /*default*/"true",
              "Enable make_gather_scatter_tptr support">,
+      Option<"enableModuloSupport", "enable-modulo-support", "bool", /*default*/"true",
+             "Enable modulo operation support in pointer analysis">,
       Option<"enableCollapseShape", "enable-collapse-shape", "bool", /*default*/"false",
              "Enable collapse shape pass">,
   ];

lib/Conversion/TritonToLinalgExperimental/TritonToLinalgExperimentalPass.cpp

@@ -55,7 +55,8 @@ class TritonToLinalgExperimentalPass
     auto moduleOp = getOperation();
     PassManager pm(&getContext(), moduleOp.getOperationName());
 
-    pm.addPass(createTritonToStructuredPass(enableMakeGatherScatterTensorPtr));
+    pm.addPass(createTritonToStructuredPass(enableMakeGatherScatterTensorPtr,
+                                            enableModuloSupport));
 
     // Erase dead code and fold constants created during lowering
     pm.addPass(createCSEPass());