diff --git a/examples/histogram.jl b/examples/histogram.jl
index 3e91c29d..d4a65c50 100644
--- a/examples/histogram.jl
+++ b/examples/histogram.jl
@@ -13,7 +13,7 @@ function create_histogram(input)
 end
 
 # This a 1D histogram kernel where the histogramming happens on shmem
-@kernel function histogram_kernel!(histogram_output, input)
+@kernel implicit_validindex = false function histogram_kernel!(histogram_output, input)
     tid = @index(Global, Linear)
     lid = @index(Local, Linear)
 
diff --git a/src/KernelAbstractions.jl b/src/KernelAbstractions.jl
index ce4a6493..00320904 100644
--- a/src/KernelAbstractions.jl
+++ b/src/KernelAbstractions.jl
@@ -50,7 +50,7 @@ synchronize(backend)
 ```
 """
 macro kernel(expr)
-    return __kernel(expr, #=force_inbounds=# false)
+    return __kernel(expr, #=force_inbounds=# false, #=implicit_validindex=# true)
 end
 
 """
@@ -71,8 +71,9 @@ This allows for two different configurations:
 """
 macro kernel(ex...)
     if length(ex) == 1
-        return __kernel(ex[1], false)
+        return __kernel(ex[1], false, true)
     else
+        implicit_validindex = true
         force_inbounds = false
         for i in 1:(length(ex) - 1)
             if ex[i] isa Expr && ex[i].head == :(=) &&
@@ -81,6 +82,9 @@ macro kernel(ex...)
             elseif ex[i] isa Expr && ex[i].head == :(=) &&
                     ex[i].args[1] == :inbounds && ex[i].args[2] isa Bool
                 force_inbounds = ex[i].args[2]
+            elseif ex[i] isa Expr && ex[i].head == :(=) &&
+                    ex[i].args[1] == :implicit_validindex && ex[i].args[2] isa Bool
+                implicit_validindex = ex[i].args[2]
             else
                 error(
                     "Configuration should be of form:\n" *
@@ -90,7 +94,7 @@ macro kernel(ex...)
                 )
             end
         end
-        return __kernel(ex[end], force_inbounds)
+        return __kernel(ex[end], force_inbounds, implicit_validindex)
     end
 end
 
diff --git a/src/macros.jl b/src/macros.jl
index 205a4888..2431a82c 100644
--- a/src/macros.jl
+++ b/src/macros.jl
@@ -10,7 +10,7 @@ function find_return(stmt)
 end
 
 # XXX: Proper errors
-function __kernel(expr, force_inbounds = false)
+function __kernel(expr, force_inbounds = false, implicit_validindex = true)
     def = splitdef(expr)
     name = def[:name]
     args = def[:args]
@@ -30,7 +30,7 @@ function __kernel(expr, force_inbounds = false)
 
     def_gpu = deepcopy(def)
     def_gpu[:name] = gpu_name = Symbol(:gpu_, name)
-    transform_gpu!(def_gpu, constargs, force_inbounds)
+    transform_gpu!(def_gpu, constargs, force_inbounds, implicit_validindex)
     gpu_function = combinedef(def_gpu)
 
     # create constructor functions
@@ -50,7 +50,7 @@ end
 
 # The easy case, transform the function for GPU execution
 # - mark constant arguments by applying `constify`.
-function transform_gpu!(def, constargs, force_inbounds)
+function transform_gpu!(def, constargs, force_inbounds, implicit_validindex)
     let_constargs = Expr[]
     for (i, arg) in enumerate(def[:args])
         if constargs[i]
@@ -64,11 +64,18 @@ function transform_gpu!(def, constargs, force_inbounds)
             @inbounds $(body)
         end
     end
-    body = quote
-        if $__validindex(__ctx__)
+    if implicit_validindex
+        body = quote
+            if $__validindex(__ctx__)
+                $(body)
+            end
+            return nothing
+        end
+    else
+        body = quote
             $(body)
+            return nothing
         end
-        return nothing
     end
     def[:body] = Expr(
         :let,