diff --git a/src/pass/LambdaAnalysis.cpp b/src/pass/LambdaAnalysis.cpp
index 3fb601f0..265ee193 100644
--- a/src/pass/LambdaAnalysis.cpp
+++ b/src/pass/LambdaAnalysis.cpp
@@ -552,8 +552,11 @@ class LambdaPassImpl {
       for (const auto &JitVarInfo : It->second) {
         auto *SlotC = dyn_cast<ConstantInt>(JitVarInfo.Slot);
         auto *OffsetC = dyn_cast<ConstantInt>(JitVarInfo.Offset);
-        auto RCType = getRCTypeForLLVMType(JitVarInfo.Type);
-        if (!SlotC || !OffsetC || !RCType)
+        if (!SlotC || !OffsetC)
+          reportFatalError("Failed to emit lambda schema metadata");
+        auto RCType = getRCTypeForLLVMType(
+            LambdaStorageType->getElementType(SlotC->getZExtValue()));
+        if (!RCType)
           reportFatalError("Failed to emit lambda schema metadata");
 
         LLVMContext &Ctx = M.getContext();
diff --git a/tests/cpu/CMakeLists.txt b/tests/cpu/CMakeLists.txt
index fd8a3009..e33f1c29 100644
--- a/tests/cpu/CMakeLists.txt
+++ b/tests/cpu/CMakeLists.txt
@@ -66,6 +66,7 @@ CREATE_CPU_TEST(types_api types_api.cpp)
 CREATE_CPU_TEST(lambda lambda.cpp)
 CREATE_CPU_TEST(lambda_ptrtoint_capture lambda_ptrtoint_capture.cpp)
 CREATE_CPU_TEST(lambda_def lambda_def.cpp)
+CREATE_CPU_TEST(lambda_cast_capture lambda_cast_capture.cpp)
 CREATE_CPU_TEST(lambda_nested lambda_nested.cpp)
 CREATE_CPU_TEST(annot_nested annot_nested.cpp)
 CREATE_CPU_TEST(lambda_def_register_once lambda_def_register_once.cpp)
diff --git a/tests/cpu/lambda_cast_capture.cpp b/tests/cpu/lambda_cast_capture.cpp
new file mode 100644
index 00000000..f1f3efe7
--- /dev/null
+++ b/tests/cpu/lambda_cast_capture.cpp
@@ -0,0 +1,52 @@
+// clang-format off
+// RUN: rm -rf "%t.$$.proteus"
+// RUN: PROTEUS_CACHE_DIR="%t.$$.proteus" %build/lambda_cast_capture | %FILECHECK %s --check-prefixes=CHECK
+// RUN: rm -rf "%t.$$.proteus"
+// clang-format on
+
+#include <cstdio>
+
+#include <proteus/JitInterface.h>
+
+template <typename F> void run(F &&Func) { proteus::register_lambda(Func)(); }
+
+int main() {
+  int I = -5;
+  unsigned U = 0x80000000u;
+  float F = 1.5f;
+  double D = 2.7;
+
+  auto IntToLong =
+      [=, X = (long)proteus::jit_variable(I)]()
+          __attribute__((annotate("jit"))) { printf("IntToLong %ld\n", X); };
+  run(IntToLong);
+
+  auto UnsignedToULong = [=, X = (unsigned long)proteus::jit_variable(U)]()
+                             __attribute__((annotate("jit"))) {
+                               printf("UnsignedToULong %lu\n", X);
+                             };
+  run(UnsignedToULong);
+
+  auto FloatToDouble =
+      [=, X = (double)proteus::jit_variable(F)]()
+          __attribute__((annotate("jit"))) { printf("FloatToDouble %f\n", X); };
+  run(FloatToDouble);
+
+  auto IntToDouble =
+      [=, X = (double)proteus::jit_variable(I)]()
+          __attribute__((annotate("jit"))) { printf("IntToDouble %f\n", X); };
+  run(IntToDouble);
+
+  auto DoubleToInt =
+      [=, X = (int)proteus::jit_variable(D)]()
+          __attribute__((annotate("jit"))) { printf("DoubleToInt %d\n", X); };
+  run(DoubleToInt);
+
+  return 0;
+}
+
+// CHECK: IntToLong -5
+// CHECK: UnsignedToULong 2147483648
+// CHECK: FloatToDouble 1.500000
+// CHECK: IntToDouble -5.000000
+// CHECK: DoubleToInt 2
diff --git a/tests/gpu/CMakeLists.txt b/tests/gpu/CMakeLists.txt
index 5e9b321d..0e122d55 100644
--- a/tests/gpu/CMakeLists.txt
+++ b/tests/gpu/CMakeLists.txt
@@ -262,6 +262,7 @@ CREATE_GPU_TEST(shared_array shared_array.cpp)
 CREATE_GPU_TEST(enable_disable enable_disable.cpp)
 CREATE_GPU_TEST(lambda_multiple lambda_multiple.cpp)
 CREATE_GPU_TEST(lambda_def lambda_def.cpp)
+CREATE_GPU_TEST(lambda_cast_capture lambda_cast_capture.cpp)
 CREATE_GPU_TEST(lambda_factory lambda_factory.cpp)
 CREATE_GPU_TEST(lambda_factory_forwarded lambda_factory_forwarded.cpp)
 CREATE_GPU_TEST(lambda_thread_privatizer lambda_thread_privatizer.cpp)
diff --git a/tests/gpu/lambda_cast_capture.cpp b/tests/gpu/lambda_cast_capture.cpp
new file mode 100644
index 00000000..61224def
--- /dev/null
+++ b/tests/gpu/lambda_cast_capture.cpp
@@ -0,0 +1,63 @@
+// clang-format off
+// RUN: rm -rf "%t.$$.proteus"
+// RUN: PROTEUS_CACHE_DIR="%t.$$.proteus" %build/lambda_cast_capture.%ext | %FILECHECK %s --check-prefixes=CHECK
+// RUN: rm -rf "%t.$$.proteus"
+// clang-format on
+
+#include <cstdio>
+
+#include "gpu_common.h"
+#include <proteus/JitInterface.h>
+
+template <typename T>
+__global__ __attribute__((annotate("jit"))) void kernel(T LB) {
+  LB();
+}
+
+template <typename T> void run(T &&LB) {
+  kernel<<<1, 1>>>(proteus::register_lambda(LB));
+  gpuErrCheck(gpuDeviceSynchronize());
+  fflush(stdout);
+}
+
+int main() {
+  int I = -5;
+  unsigned U = 0x80000000u;
+  float F = 1.5f;
+  double D = 2.7;
+
+  auto IntToLong =
+      [=, X = (long)proteus::jit_variable(I)] __device__()
+          __attribute__((annotate("jit"))) { printf("IntToLong %ld\n", X); };
+  run(IntToLong);
+
+  auto UnsignedToULong =
+      [=, X = (unsigned long)proteus::jit_variable(U)] __device__()
+          __attribute__((annotate("jit"))) {
+            printf("UnsignedToULong %lu\n", X);
+          };
+  run(UnsignedToULong);
+
+  auto FloatToDouble =
+      [=, X = (double)proteus::jit_variable(F)] __device__()
+          __attribute__((annotate("jit"))) { printf("FloatToDouble %f\n", X); };
+  run(FloatToDouble);
+
+  auto IntToDouble =
+      [=, X = (double)proteus::jit_variable(I)] __device__()
+          __attribute__((annotate("jit"))) { printf("IntToDouble %f\n", X); };
+  run(IntToDouble);
+
+  auto DoubleToInt =
+      [=, X = (int)proteus::jit_variable(D)] __device__()
+          __attribute__((annotate("jit"))) { printf("DoubleToInt %d\n", X); };
+  run(DoubleToInt);
+
+  return 0;
+}
+
+// CHECK: IntToLong -5
+// CHECK: UnsignedToULong 2147483648
+// CHECK: FloatToDouble 1.500000
+// CHECK: IntToDouble -5.000000
+// CHECK: DoubleToInt 2