diff --git a/Project.toml b/Project.toml
index a34cc83b..134da1ec 100644
--- a/Project.toml
+++ b/Project.toml
@@ -48,7 +48,7 @@ CEnum = "0.5, 1"
 Crayons = "4"
 ExprTools = "0.1"
 GPUArrays = "12.1"
-GPUCompiler = "2.17"
+GPUCompiler = "2.18"
 GPUToolbox = "3.4"
 IntelITT = "0.2"
 KernelAbstractions = "0.10"
diff --git a/src/compiler/execution.jl b/src/compiler/execution.jl
index f39bd4b7..aaab7794 100644
--- a/src/compiler/execution.jl
+++ b/src/compiler/execution.jl
@@ -326,7 +326,11 @@ end
 const zefunction_lock = ReentrantLock()
 
 function zefunction(f::F, tt::TT=Tuple{}; kwargs...) where {F,TT}
-    fun = zefunction_link(methodinstance(F, tt); kwargs...)
+    # the kernel also depends on the task's context and device, see `zefunction_link`
+    kw = values(kwargs)
+    fun = GPUCompiler.cached_lookup(ZeKernel, F, tt, (context(), device(), kw)) do
+        zefunction_link(methodinstance(F, tt); kw...)
+    end
     return HostKernel{F,tt}(f, fun)
 end