[flang][cuda] Fix lowering when step is a variable (#119421)

Add missing conversion.

Author: clementval

flang/lib/Lower/Bridge.cpp

@@ -3037,8 +3037,10 @@ class FirConverter : public Fortran::lower::AbstractConverter {
           fir::getBase(genExprValue(*Fortran::semantics::GetExpr(bounds->upper),
                                     stmtCtx))));
       if (bounds->step)
-        steps.push_back(fir::getBase(
-            genExprValue(*Fortran::semantics::GetExpr(bounds->step), stmtCtx)));
+        steps.push_back(builder->createConvert(
+            crtLoc, idxTy,
+            fir::getBase(genExprValue(
+                *Fortran::semantics::GetExpr(bounds->step), stmtCtx))));
       else // If `step` is not present, assume it is `1`.
         steps.push_back(builder->createIntegerConstant(loc, idxTy, 1));
 

flang/test/Lower/CUDA/cuda-kernel-loop-directive.cuf

@@ -98,3 +98,18 @@ end subroutine
 
 ! CHECK-LABEL: func.func @_QPsub2
 ! CHECK: cuf.kernel
+
+subroutine sub3()
+  integer, device :: a(10), b(10)
+  integer :: lb = 1
+  integer :: n = 10
+  integer :: s = 1
+
+  !$cuf kernel do <<< *, * >>>
+  do i = lb, n, s
+    a(i) = a(i) * b(i)
+  end do
+end
+
+! CHECK-LABEL: func.func @_QPsub3
+! CHECK: cuf.kernel