[BACKEND] Fix reductions when number of unique element is smaller than layout (#1913)

Fix calculation of unique number of threads within a warp. We need to
consider the number of elements per thread in the calculation. Also
change the layout test to integer sum in order to catch bugs with unique
data as max reduction may hide those kind of problems.

lib/Analysis/AxisInfo.cpp

@@ -918,7 +918,8 @@ unsigned ModuleAxisInfoAnalysis::getPtrContiguity(Value ptr) {
   auto order = triton::gpu::getOrder(layout);
   unsigned align = getPtrAlignment(ptr);
 
-  auto uniqueContigPerThread = triton::gpu::getUniqueContigPerThread(tensorTy);
+  auto uniqueContigPerThread =
+      triton::gpu::getUniqueContigPerThread(layout, tensorTy.getShape());
   assert(order[0] < uniqueContigPerThread.size() &&
          "Unxpected uniqueContigPerThread size");
   unsigned contiguity = uniqueContigPerThread[order[0]];