diff --git a/libspu/compiler/codegen/BUILD.bazel b/libspu/compiler/codegen/BUILD.bazel
index 2585dae3..18c81585 100644
--- a/libspu/compiler/codegen/BUILD.bazel
+++ b/libspu/compiler/codegen/BUILD.bazel
@@ -23,6 +23,7 @@ spu_cc_library(
     srcs = ["codegen.cc"],
     hdrs = ["codegen.h"],
     deps = [
+        "//libspu:version",
         "//libspu/dialect/pphlo/IR:dialect",
     ],
 )
diff --git a/libspu/compiler/codegen/codegen.cc b/libspu/compiler/codegen/codegen.cc
index 1c628b9c..b0c25d21 100644
--- a/libspu/compiler/codegen/codegen.cc
+++ b/libspu/compiler/codegen/codegen.cc
@@ -16,9 +16,15 @@
 
 #include "llvm/Support/raw_ostream.h"
 
+#include "libspu/version.h"
+
 namespace spu::compiler {
 
 std::string CodeGen::doit(mlir::ModuleOp module) {
+  // Add ir version attr
+  module->setAttr("pphlo.version",
+                  mlir::StringAttr::get(module->getContext(), getVersionStr()));
+  // Emit module
   std::string ir_dump;
   llvm::raw_string_ostream stream(ir_dump);
   module.print(stream);
diff --git a/libspu/compiler/tests/interpret/abs.mlir b/libspu/compiler/tests/interpret/abs.mlir
index 584fc87a..dd655bf1 100644
--- a/libspu/compiler/tests/interpret/abs.mlir
+++ b/libspu/compiler/tests/interpret/abs.mlir
@@ -1,4 +1,6 @@
-// RUN: spu-translate --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=1 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=2 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=3 --interpret -split-input-file %s
 // AUTO GENERATED, DO NOT EDIT
 
 func.func @abs_op_test_i64_i64_p() {
diff --git a/libspu/compiler/tests/interpret/add.mlir b/libspu/compiler/tests/interpret/add.mlir
index 9244b346..a763c0ae 100644
--- a/libspu/compiler/tests/interpret/add.mlir
+++ b/libspu/compiler/tests/interpret/add.mlir
@@ -1,4 +1,6 @@
-// RUN: spu-translate --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=1 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=2 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=3 --interpret -split-input-file %s
 // AUTO GENERATED, DO NOT EDIT
 
 func.func @add_op_test_i8_i8_pp() {
diff --git a/libspu/compiler/tests/interpret/and.mlir b/libspu/compiler/tests/interpret/and.mlir
index c06920e1..6e3cdfa6 100644
--- a/libspu/compiler/tests/interpret/and.mlir
+++ b/libspu/compiler/tests/interpret/and.mlir
@@ -1,4 +1,6 @@
-// RUN: spu-translate --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=1 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=2 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=3 --interpret -split-input-file %s
 // AUTO GENERATED, DO NOT EDIT
 
 func.func @and_op_test_i8_i8_pp() {
diff --git a/libspu/compiler/tests/interpret/atan2.mlir b/libspu/compiler/tests/interpret/atan2.mlir
index 73acce5f..e1938852 100644
--- a/libspu/compiler/tests/interpret/atan2.mlir
+++ b/libspu/compiler/tests/interpret/atan2.mlir
@@ -1,4 +1,6 @@
-// RUN: spu-translate --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=1 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=2 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=3 --interpret -split-input-file %s
 // AUTO GENERATED, DO NOT EDIT
 
 func.func @atan2_op_test_f64_f64_pp() {
diff --git a/libspu/compiler/tests/interpret/broadcast.mlir b/libspu/compiler/tests/interpret/broadcast.mlir
index 34a1d934..1d0e9a63 100644
--- a/libspu/compiler/tests/interpret/broadcast.mlir
+++ b/libspu/compiler/tests/interpret/broadcast.mlir
@@ -1,4 +1,6 @@
-// RUN: spu-translate --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=1 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=2 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=3 --interpret -split-input-file %s
 
 func.func @broadcast_in_dim() {
   %operand = pphlo.constant dense<[[1], [2], [3]]> : tensor<3x1xi64>
diff --git a/libspu/compiler/tests/interpret/case.mlir b/libspu/compiler/tests/interpret/case.mlir
index 9beae82b..543fbf86 100644
--- a/libspu/compiler/tests/interpret/case.mlir
+++ b/libspu/compiler/tests/interpret/case.mlir
@@ -1,4 +1,6 @@
-// RUN: spu-translate --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=1 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=2 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=3 --interpret -split-input-file %s
 
 func.func @case_negative_index_default() {
   %index = pphlo.constant dense<-1> : tensor<i32>
diff --git a/libspu/compiler/tests/interpret/ceil.mlir b/libspu/compiler/tests/interpret/ceil.mlir
index 5a7d2e35..d4f74b8b 100644
--- a/libspu/compiler/tests/interpret/ceil.mlir
+++ b/libspu/compiler/tests/interpret/ceil.mlir
@@ -1,4 +1,6 @@
-// RUN: spu-translate --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=1 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=2 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=3 --interpret -split-input-file %s
 // AUTO GENERATED, DO NOT EDIT
 
 func.func @ceil_op_test_f16_f16_p() {
diff --git a/libspu/compiler/tests/interpret/clamp.mlir b/libspu/compiler/tests/interpret/clamp.mlir
index cf193911..69c05696 100644
--- a/libspu/compiler/tests/interpret/clamp.mlir
+++ b/libspu/compiler/tests/interpret/clamp.mlir
@@ -1,4 +1,6 @@
-// RUN: spu-translate --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=1 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=2 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=3 --interpret -split-input-file %s
 
 func.func @clamp_op_test_si64() {
   %min = pphlo.constant dense<[1, 5, -5]> : tensor<3xi64>
diff --git a/libspu/compiler/tests/interpret/concatenate.mlir b/libspu/compiler/tests/interpret/concatenate.mlir
index e16c5e1f..1d0d7655 100644
--- a/libspu/compiler/tests/interpret/concatenate.mlir
+++ b/libspu/compiler/tests/interpret/concatenate.mlir
@@ -1,4 +1,6 @@
-// RUN: spu-translate --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=1 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=2 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=3 --interpret -split-input-file %s
 
 func.func @concatenate() {
   %input0 = pphlo.constant dense<[[1, 2], [3, 4], [5, 6]]> : tensor<3x2xi64>
diff --git a/libspu/compiler/tests/interpret/convert.mlir b/libspu/compiler/tests/interpret/convert.mlir
index 64c6fa15..747047ee 100644
--- a/libspu/compiler/tests/interpret/convert.mlir
+++ b/libspu/compiler/tests/interpret/convert.mlir
@@ -1,4 +1,6 @@
-// RUN: spu-translate --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=1 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=2 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=3 --interpret -split-input-file %s
 
 func.func @convert_op_test_1() {
   %0 = pphlo.constant dense<[0, 1, 8, -9, 0]> : tensor<5xi32>
diff --git a/libspu/compiler/tests/interpret/convolution.mlir b/libspu/compiler/tests/interpret/convolution.mlir
index 81a12c05..294cdd36 100644
--- a/libspu/compiler/tests/interpret/convolution.mlir
+++ b/libspu/compiler/tests/interpret/convolution.mlir
@@ -1,4 +1,6 @@
-// RUN: spu-translate --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=1 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=2 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=3 --interpret -split-input-file %s
 
 func.func @main() {
     %0 = pphlo.constant dense<[[[[ 1.0, 2.0, 3.0, 4.0],
diff --git a/libspu/compiler/tests/interpret/cosine.mlir b/libspu/compiler/tests/interpret/cosine.mlir
index 2b89d421..72d6f249 100644
--- a/libspu/compiler/tests/interpret/cosine.mlir
+++ b/libspu/compiler/tests/interpret/cosine.mlir
@@ -1,4 +1,6 @@
-// RUN: spu-translate --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=1 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=2 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=3 --interpret -split-input-file %s
 // AUTO GENERATED, DO NOT EDIT
 
 func.func @cosine_op_test_f16_f16_p() {
diff --git a/libspu/compiler/tests/interpret/divide.mlir b/libspu/compiler/tests/interpret/divide.mlir
index 6b5a95e1..56ce53c2 100644
--- a/libspu/compiler/tests/interpret/divide.mlir
+++ b/libspu/compiler/tests/interpret/divide.mlir
@@ -1,4 +1,6 @@
-// RUN: spu-translate --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=1 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=2 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=3 --interpret -split-input-file %s
 // AUTO GENERATED, DO NOT EDIT
 
 func.func @divide_op_test_i64_i64_pp() {
diff --git a/libspu/compiler/tests/interpret/dot_general.mlir b/libspu/compiler/tests/interpret/dot_general.mlir
index b97f06f7..29023c81 100644
--- a/libspu/compiler/tests/interpret/dot_general.mlir
+++ b/libspu/compiler/tests/interpret/dot_general.mlir
@@ -1,4 +1,6 @@
-// RUN: spu-translate --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=1 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=2 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=3 --interpret -split-input-file %s
 
 func.func @dot_general_op_test_si64() {
   %lhs = pphlo.constant dense<[[[1, 2], [3, 4]],
diff --git a/libspu/compiler/tests/interpret/dynamic_slice.mlir b/libspu/compiler/tests/interpret/dynamic_slice.mlir
index a72577c6..672ace7c 100644
--- a/libspu/compiler/tests/interpret/dynamic_slice.mlir
+++ b/libspu/compiler/tests/interpret/dynamic_slice.mlir
@@ -1,4 +1,6 @@
-// RUN: spu-translate --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=1 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=2 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=3 --interpret -split-input-file %s
 
 func.func @dynamic_slice() {
   %operand = pphlo.constant dense<[[1, 1, 1],
@@ -12,3 +14,22 @@ func.func @dynamic_slice() {
   pphlo.custom_call @expect_eq (%result, %operand) : (tensor<3x3xi64>,tensor<3x3xi64>)->()
   func.return
 }
+
+// -----
+
+func.func @dynamic_slice() {
+  %operand = pphlo.constant dense<[[1, 2, 3],
+                                   [4, 5, 6],
+                                   [7, 8, 9]]> : tensor<3x3xi64>
+  %i0 = pphlo.constant dense<1> : tensor<i64>
+  %start_indices0 = pphlo.convert %i0 : (tensor<i64>) -> tensor<!pphlo.secret<i64>>
+  %start_indices1 = pphlo.constant dense<1> : tensor<i64>
+  %r = "pphlo.dynamic_slice"(%operand, %start_indices0, %start_indices1) {
+    slice_sizes = array<i64: 2, 2>
+  } : (tensor<3x3xi64>, tensor<!pphlo.secret<i64>>, tensor<i64>) -> tensor<2x2x!pphlo.secret<i64>>
+  %result = pphlo.convert %r : (tensor<2x2x!pphlo.secret<i64>>) -> tensor<2x2xi64>
+  %expected = pphlo.constant dense<[[5, 6],
+                                    [8, 9]]> : tensor<2x2xi64>
+  pphlo.custom_call @expect_eq (%result, %expected) : (tensor<2x2xi64>,tensor<2x2xi64>)->()
+  func.return
+}
diff --git a/libspu/compiler/tests/interpret/dynamic_update_slice.mlir b/libspu/compiler/tests/interpret/dynamic_update_slice.mlir
index 9177f938..a7053428 100644
--- a/libspu/compiler/tests/interpret/dynamic_update_slice.mlir
+++ b/libspu/compiler/tests/interpret/dynamic_update_slice.mlir
@@ -1,4 +1,6 @@
-// RUN: spu-translate --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=1 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=2 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=3 --interpret -split-input-file %s
 
 func.func @dynamic_update_slice() {
   %operand = pphlo.constant dense<[[1, 1, 1, 1],
diff --git a/libspu/compiler/tests/interpret/equal.mlir b/libspu/compiler/tests/interpret/equal.mlir
index c29d51e4..f5364638 100644
--- a/libspu/compiler/tests/interpret/equal.mlir
+++ b/libspu/compiler/tests/interpret/equal.mlir
@@ -1,4 +1,6 @@
-// RUN: spu-translate --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=1 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=2 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=3 --interpret -split-input-file %s
 // AUTO GENERATED, DO NOT EDIT
 
 func.func @equal_op_test_i64_i1_pp() {
diff --git a/libspu/compiler/tests/interpret/exponential.mlir b/libspu/compiler/tests/interpret/exponential.mlir
index 4b505e63..a85c5bb6 100644
--- a/libspu/compiler/tests/interpret/exponential.mlir
+++ b/libspu/compiler/tests/interpret/exponential.mlir
@@ -1,4 +1,6 @@
-// RUN: spu-translate --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=1 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=2 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=3 --interpret -split-input-file %s
 // AUTO GENERATED, DO NOT EDIT
 
 func.func @exponential_op_test_f64_f64_p() {
diff --git a/libspu/compiler/tests/interpret/exponential_minus_one.mlir b/libspu/compiler/tests/interpret/exponential_minus_one.mlir
index 46cf91f1..1a337a21 100644
--- a/libspu/compiler/tests/interpret/exponential_minus_one.mlir
+++ b/libspu/compiler/tests/interpret/exponential_minus_one.mlir
@@ -1,4 +1,6 @@
-// RUN: spu-translate --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=1 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=2 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=3 --interpret -split-input-file %s
 // AUTO GENERATED, DO NOT EDIT
 
 func.func @exponential_minus_one_op_test_f64_f64_p() {
diff --git a/libspu/compiler/tests/interpret/floor.mlir b/libspu/compiler/tests/interpret/floor.mlir
index 734266b9..97ccdb1e 100644
--- a/libspu/compiler/tests/interpret/floor.mlir
+++ b/libspu/compiler/tests/interpret/floor.mlir
@@ -1,4 +1,6 @@
-// RUN: spu-translate --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=1 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=2 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=3 --interpret -split-input-file %s
 // AUTO GENERATED, DO NOT EDIT
 
 func.func @floor_op_test_f16_f16_p() {
diff --git a/libspu/compiler/tests/interpret/generate_mlir_tests.py b/libspu/compiler/tests/interpret/generate_mlir_tests.py
index 114e66a0..f9bdef86 100755
--- a/libspu/compiler/tests/interpret/generate_mlir_tests.py
+++ b/libspu/compiler/tests/interpret/generate_mlir_tests.py
@@ -90,7 +90,15 @@ def TestCase(inputs, expected, checker='expect_eq', tol=None):
 
     with open(f"{test_name}.mlir", "w+") as f:
         # emit run command
-        f.write("// RUN: spu-translate --interpret -split-input-file %s\n")
+        f.write(
+            "// RUN: spu-translate --protocol_kind=1 --interpret -split-input-file %s\n"
+        )
+        f.write(
+            "// RUN: spu-translate --protocol_kind=2 --interpret -split-input-file %s\n"
+        )
+        f.write(
+            "// RUN: spu-translate --protocol_kind=3 --interpret -split-input-file %s\n"
+        )
         f.write("// AUTO GENERATED, DO NOT EDIT\n\n")
 
         # Emit cases
diff --git a/libspu/compiler/tests/interpret/greater.mlir b/libspu/compiler/tests/interpret/greater.mlir
index e6b6b0d1..7f8e76be 100644
--- a/libspu/compiler/tests/interpret/greater.mlir
+++ b/libspu/compiler/tests/interpret/greater.mlir
@@ -1,4 +1,6 @@
-// RUN: spu-translate --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=1 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=2 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=3 --interpret -split-input-file %s
 // AUTO GENERATED, DO NOT EDIT
 
 func.func @greater_op_test_i64_i1_pp() {
diff --git a/libspu/compiler/tests/interpret/greater_equal.mlir b/libspu/compiler/tests/interpret/greater_equal.mlir
index b1b32c14..305aaff5 100644
--- a/libspu/compiler/tests/interpret/greater_equal.mlir
+++ b/libspu/compiler/tests/interpret/greater_equal.mlir
@@ -1,4 +1,6 @@
-// RUN: spu-translate --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=1 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=2 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=3 --interpret -split-input-file %s
 // AUTO GENERATED, DO NOT EDIT
 
 func.func @greater_equal_op_test_i64_i1_pp() {
diff --git a/libspu/compiler/tests/interpret/if.mlir b/libspu/compiler/tests/interpret/if.mlir
index 7f98d8b0..cc98b65d 100644
--- a/libspu/compiler/tests/interpret/if.mlir
+++ b/libspu/compiler/tests/interpret/if.mlir
@@ -1,4 +1,6 @@
-// RUN: spu-translate --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=1 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=2 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=3 --interpret -split-input-file %s
 
 func.func @if_ops_true_branch() {
   %pred = pphlo.constant dense<true> : tensor<i1>
diff --git a/libspu/compiler/tests/interpret/iota.mlir b/libspu/compiler/tests/interpret/iota.mlir
index 40f7800d..a7ee86ed 100644
--- a/libspu/compiler/tests/interpret/iota.mlir
+++ b/libspu/compiler/tests/interpret/iota.mlir
@@ -1,4 +1,6 @@
-// RUN: spu-translate --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=1 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=2 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=3 --interpret -split-input-file %s
 
 func.func @iota_op_test_si8_dim_0() {
   %0 = pphlo.iota dim = 0 : tensor<3x4xi8>
diff --git a/libspu/compiler/tests/interpret/less.mlir b/libspu/compiler/tests/interpret/less.mlir
index 21f12d45..58444a29 100644
--- a/libspu/compiler/tests/interpret/less.mlir
+++ b/libspu/compiler/tests/interpret/less.mlir
@@ -1,4 +1,6 @@
-// RUN: spu-translate --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=1 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=2 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=3 --interpret -split-input-file %s
 // AUTO GENERATED, DO NOT EDIT
 
 func.func @less_op_test_i64_i1_pp() {
diff --git a/libspu/compiler/tests/interpret/less_equal.mlir b/libspu/compiler/tests/interpret/less_equal.mlir
index 2715e3c6..9951a569 100644
--- a/libspu/compiler/tests/interpret/less_equal.mlir
+++ b/libspu/compiler/tests/interpret/less_equal.mlir
@@ -1,4 +1,6 @@
-// RUN: spu-translate --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=1 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=2 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=3 --interpret -split-input-file %s
 // AUTO GENERATED, DO NOT EDIT
 
 func.func @less_equal_op_test_i64_i1_pp() {
diff --git a/libspu/compiler/tests/interpret/log.mlir b/libspu/compiler/tests/interpret/log.mlir
index 56c51db7..af61d86a 100644
--- a/libspu/compiler/tests/interpret/log.mlir
+++ b/libspu/compiler/tests/interpret/log.mlir
@@ -1,4 +1,6 @@
-// RUN: spu-translate --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=1 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=2 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=3 --interpret -split-input-file %s
 // AUTO GENERATED, DO NOT EDIT
 
 func.func @log_op_test_f64_f64_p() {
diff --git a/libspu/compiler/tests/interpret/log_plus_one.mlir b/libspu/compiler/tests/interpret/log_plus_one.mlir
index 721f72b9..3aec9184 100644
--- a/libspu/compiler/tests/interpret/log_plus_one.mlir
+++ b/libspu/compiler/tests/interpret/log_plus_one.mlir
@@ -1,4 +1,6 @@
-// RUN: spu-translate --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=1 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=2 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=3 --interpret -split-input-file %s
 // AUTO GENERATED, DO NOT EDIT
 
 func.func @log_plus_one_op_test_f64_f64_p() {
diff --git a/libspu/compiler/tests/interpret/logistic.mlir b/libspu/compiler/tests/interpret/logistic.mlir
index 886ef64a..eeac6fab 100644
--- a/libspu/compiler/tests/interpret/logistic.mlir
+++ b/libspu/compiler/tests/interpret/logistic.mlir
@@ -1,4 +1,6 @@
-// RUN: spu-translate --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=1 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=2 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=3 --interpret -split-input-file %s
 // AUTO GENERATED, DO NOT EDIT
 
 func.func @logistic_op_test_f64_f64_p() {
diff --git a/libspu/compiler/tests/interpret/maximum.mlir b/libspu/compiler/tests/interpret/maximum.mlir
index f05a0048..b90553e7 100644
--- a/libspu/compiler/tests/interpret/maximum.mlir
+++ b/libspu/compiler/tests/interpret/maximum.mlir
@@ -1,4 +1,6 @@
-// RUN: spu-translate --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=1 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=2 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=3 --interpret -split-input-file %s
 // AUTO GENERATED, DO NOT EDIT
 
 func.func @maximum_op_test_i8_i8_pp() {
diff --git a/libspu/compiler/tests/interpret/minimum.mlir b/libspu/compiler/tests/interpret/minimum.mlir
index 763aa66b..74bb2b83 100644
--- a/libspu/compiler/tests/interpret/minimum.mlir
+++ b/libspu/compiler/tests/interpret/minimum.mlir
@@ -1,4 +1,6 @@
-// RUN: spu-translate --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=1 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=2 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=3 --interpret -split-input-file %s
 // AUTO GENERATED, DO NOT EDIT
 
 func.func @minimum_op_test_i8_i8_pp() {
diff --git a/libspu/compiler/tests/interpret/multiply.mlir b/libspu/compiler/tests/interpret/multiply.mlir
index 930b6458..82b780f3 100644
--- a/libspu/compiler/tests/interpret/multiply.mlir
+++ b/libspu/compiler/tests/interpret/multiply.mlir
@@ -1,4 +1,6 @@
-// RUN: spu-translate --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=1 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=2 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=3 --interpret -split-input-file %s
 // AUTO GENERATED, DO NOT EDIT
 
 func.func @multiply_op_test_i8_i8_pp() {
diff --git a/libspu/compiler/tests/interpret/negate.mlir b/libspu/compiler/tests/interpret/negate.mlir
index fa5fe4c3..8c5e74c3 100644
--- a/libspu/compiler/tests/interpret/negate.mlir
+++ b/libspu/compiler/tests/interpret/negate.mlir
@@ -1,4 +1,6 @@
-// RUN: spu-translate --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=1 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=2 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=3 --interpret -split-input-file %s
 // AUTO GENERATED, DO NOT EDIT
 
 func.func @negate_op_test_i8_i8_p() {
diff --git a/libspu/compiler/tests/interpret/not.mlir b/libspu/compiler/tests/interpret/not.mlir
index 77c91d25..0fdf44e4 100644
--- a/libspu/compiler/tests/interpret/not.mlir
+++ b/libspu/compiler/tests/interpret/not.mlir
@@ -1,4 +1,6 @@
-// RUN: spu-translate --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=1 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=2 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=3 --interpret -split-input-file %s
 // AUTO GENERATED, DO NOT EDIT
 
 func.func @not_op_test_i8_i8_p() {
diff --git a/libspu/compiler/tests/interpret/not_equal.mlir b/libspu/compiler/tests/interpret/not_equal.mlir
index aabacc0b..1bbbd5b3 100644
--- a/libspu/compiler/tests/interpret/not_equal.mlir
+++ b/libspu/compiler/tests/interpret/not_equal.mlir
@@ -1,4 +1,6 @@
-// RUN: spu-translate --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=1 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=2 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=3 --interpret -split-input-file %s
 // AUTO GENERATED, DO NOT EDIT
 
 func.func @not_equal_op_test_i64_i1_pp() {
diff --git a/libspu/compiler/tests/interpret/or.mlir b/libspu/compiler/tests/interpret/or.mlir
index 2856bba6..79836813 100644
--- a/libspu/compiler/tests/interpret/or.mlir
+++ b/libspu/compiler/tests/interpret/or.mlir
@@ -1,4 +1,6 @@
-// RUN: spu-translate --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=1 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=2 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=3 --interpret -split-input-file %s
 // AUTO GENERATED, DO NOT EDIT
 
 func.func @or_op_test_i8_i8_pp() {
diff --git a/libspu/compiler/tests/interpret/pad.mlir b/libspu/compiler/tests/interpret/pad.mlir
index 85436480..abedc73a 100644
--- a/libspu/compiler/tests/interpret/pad.mlir
+++ b/libspu/compiler/tests/interpret/pad.mlir
@@ -1,4 +1,6 @@
-// RUN: spu-translate --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=1 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=2 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=3 --interpret -split-input-file %s
 
 func.func @pad() {
   %operand = pphlo.constant dense<[[0, 0, 0, 0],
diff --git a/libspu/compiler/tests/interpret/popcnt.mlir b/libspu/compiler/tests/interpret/popcnt.mlir
index 8d741134..e5f83152 100644
--- a/libspu/compiler/tests/interpret/popcnt.mlir
+++ b/libspu/compiler/tests/interpret/popcnt.mlir
@@ -1,4 +1,6 @@
-// RUN: spu-translate --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=1 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=2 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=3 --interpret -split-input-file %s
 // AUTO GENERATED, DO NOT EDIT
 
 func.func @popcnt_op_test_i64_i64_p() {
diff --git a/libspu/compiler/tests/interpret/power.mlir b/libspu/compiler/tests/interpret/power.mlir
index 8041418a..dd950ef1 100644
--- a/libspu/compiler/tests/interpret/power.mlir
+++ b/libspu/compiler/tests/interpret/power.mlir
@@ -1,4 +1,6 @@
-// RUN: spu-translate --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=1 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=2 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=3 --interpret -split-input-file %s
 // AUTO GENERATED, DO NOT EDIT
 
 func.func @power_op_test_i64_i64_pp() {
diff --git a/libspu/compiler/tests/interpret/reduce.mlir b/libspu/compiler/tests/interpret/reduce.mlir
index 8fc4ced7..3b34bf3e 100644
--- a/libspu/compiler/tests/interpret/reduce.mlir
+++ b/libspu/compiler/tests/interpret/reduce.mlir
@@ -1,4 +1,6 @@
-// RUN: spu-translate --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=1 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=2 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=3 --interpret -split-input-file %s
 
 func.func @reduce() {
   %input = pphlo.constant dense<[[0, 1, 2, 3, 4, 5]]> : tensor<1x6xi64>
diff --git a/libspu/compiler/tests/interpret/reduce_window.mlir b/libspu/compiler/tests/interpret/reduce_window.mlir
index 1b28aef1..5385ab04 100644
--- a/libspu/compiler/tests/interpret/reduce_window.mlir
+++ b/libspu/compiler/tests/interpret/reduce_window.mlir
@@ -1,4 +1,6 @@
-// RUN: spu-translate --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=1 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=2 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=3 --interpret -split-input-file %s
 
 func.func @reduce_window() {
   %input = pphlo.constant dense<[[1, 2], [3, 4], [5, 6]]> : tensor<3x2xi64>
diff --git a/libspu/compiler/tests/interpret/reshape.mlir b/libspu/compiler/tests/interpret/reshape.mlir
index fa3551c6..9e483b77 100644
--- a/libspu/compiler/tests/interpret/reshape.mlir
+++ b/libspu/compiler/tests/interpret/reshape.mlir
@@ -1,4 +1,6 @@
-// RUN: spu-translate --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=1 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=2 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=3 --interpret -split-input-file %s
 // AUTO GENERATED, DO NOT EDIT
 
 func.func @reshape_op_test_i32_i32_p() {
diff --git a/libspu/compiler/tests/interpret/reverse.mlir b/libspu/compiler/tests/interpret/reverse.mlir
index 0489311b..63ab9590 100644
--- a/libspu/compiler/tests/interpret/reverse.mlir
+++ b/libspu/compiler/tests/interpret/reverse.mlir
@@ -1,4 +1,6 @@
-// RUN: spu-translate --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=1 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=2 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=3 --interpret -split-input-file %s
 
 func.func @reverse() {
   %operand = pphlo.constant dense<[[1, 2], [3, 4], [5, 6]]> : tensor<3x2xi64>
diff --git a/libspu/compiler/tests/interpret/ring_cast.mlir b/libspu/compiler/tests/interpret/ring_cast.mlir
new file mode 100644
index 00000000..94b53241
--- /dev/null
+++ b/libspu/compiler/tests/interpret/ring_cast.mlir
@@ -0,0 +1,13 @@
+// RUN: spu-translate --protocol_kind=1 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=2 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=3 --interpret -split-input-file %s
+
+func.func @cast_1() {
+  %c0 = pphlo.constant dense<[[1, 2], [3, 4], [5, 6]]> : tensor<3x2xi32>
+  %operand = pphlo.convert %c0 : (tensor<3x2xi32>)->tensor<3x2x!pphlo.secret<i32>>
+  %r = pphlo.convert %operand : (tensor<3x2x!pphlo.secret<i32>>) -> tensor<3x2x!pphlo.secret<i64>>
+  %expected = pphlo.constant dense<[[1, 2], [3, 4], [5, 6]]> : tensor<3x2xi64>
+  %result = pphlo.convert %r : (tensor<3x2x!pphlo.secret<i64>>)->tensor<3x2xi64>
+  pphlo.custom_call @expect_eq (%result, %expected) : (tensor<3x2xi64>,tensor<3x2xi64>)->()
+  func.return
+}
diff --git a/libspu/compiler/tests/interpret/round_nearest_afz.mlir b/libspu/compiler/tests/interpret/round_nearest_afz.mlir
index 79eba66d..6601fa1d 100644
--- a/libspu/compiler/tests/interpret/round_nearest_afz.mlir
+++ b/libspu/compiler/tests/interpret/round_nearest_afz.mlir
@@ -1,4 +1,6 @@
-// RUN: spu-translate --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=1 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=2 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=3 --interpret -split-input-file %s
 // AUTO GENERATED, DO NOT EDIT
 
 func.func @round_nearest_afz_op_test_f64_f64_p() {
diff --git a/libspu/compiler/tests/interpret/rsqrt.mlir b/libspu/compiler/tests/interpret/rsqrt.mlir
index 0ef05654..ef6470c9 100644
--- a/libspu/compiler/tests/interpret/rsqrt.mlir
+++ b/libspu/compiler/tests/interpret/rsqrt.mlir
@@ -1,4 +1,6 @@
-// RUN: spu-translate --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=1 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=2 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=3 --interpret -split-input-file %s
 // AUTO GENERATED, DO NOT EDIT
 
 func.func @rsqrt_op_test_f64_f64_p() {
diff --git a/libspu/compiler/tests/interpret/select.mlir b/libspu/compiler/tests/interpret/select.mlir
index eb15e544..33e9a9c7 100644
--- a/libspu/compiler/tests/interpret/select.mlir
+++ b/libspu/compiler/tests/interpret/select.mlir
@@ -1,4 +1,6 @@
-// RUN: spu-translate --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=1 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=2 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=3 --interpret -split-input-file %s
 
 func.func @select_op_test_si64() {
   %pred = pphlo.constant dense<[true, false, true]> : tensor<3xi1>
diff --git a/libspu/compiler/tests/interpret/select_and_scatter.mlir b/libspu/compiler/tests/interpret/select_and_scatter.mlir
index 6ae7fc19..5e55dc59 100644
--- a/libspu/compiler/tests/interpret/select_and_scatter.mlir
+++ b/libspu/compiler/tests/interpret/select_and_scatter.mlir
@@ -1,4 +1,6 @@
-// RUN: spu-translate --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=1 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=2 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=3 --interpret -split-input-file %s
 
 // FIXME
 func.func @select_and_scatter_op_test() {
diff --git a/libspu/compiler/tests/interpret/shift_right_arithmetic.mlir b/libspu/compiler/tests/interpret/shift_right_arithmetic.mlir
index 7cc71a93..494666ae 100644
--- a/libspu/compiler/tests/interpret/shift_right_arithmetic.mlir
+++ b/libspu/compiler/tests/interpret/shift_right_arithmetic.mlir
@@ -1,4 +1,6 @@
-// RUN: spu-translate --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=1 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=2 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=3 --interpret -split-input-file %s
 // AUTO GENERATED, DO NOT EDIT
 
 func.func @shift_right_arithmetic_op_test_i64_i64_pp() {
diff --git a/libspu/compiler/tests/interpret/shift_right_logical.mlir b/libspu/compiler/tests/interpret/shift_right_logical.mlir
index f7331033..253d6cc3 100644
--- a/libspu/compiler/tests/interpret/shift_right_logical.mlir
+++ b/libspu/compiler/tests/interpret/shift_right_logical.mlir
@@ -1,4 +1,6 @@
-// RUN: spu-translate --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=1 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=2 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=3 --interpret -split-input-file %s
 // AUTO GENERATED, DO NOT EDIT
 
 func.func @shift_right_logical_op_test_i64_i64_pp() {
diff --git a/libspu/compiler/tests/interpret/sign.mlir b/libspu/compiler/tests/interpret/sign.mlir
index 3c1b2bed..b153a373 100644
--- a/libspu/compiler/tests/interpret/sign.mlir
+++ b/libspu/compiler/tests/interpret/sign.mlir
@@ -1,4 +1,6 @@
-// RUN: spu-translate --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=1 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=2 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=3 --interpret -split-input-file %s
 // AUTO GENERATED, DO NOT EDIT
 
 func.func @sign_op_test_i64_i64_p() {
diff --git a/libspu/compiler/tests/interpret/sine.mlir b/libspu/compiler/tests/interpret/sine.mlir
index 52825da5..f0b59e5c 100644
--- a/libspu/compiler/tests/interpret/sine.mlir
+++ b/libspu/compiler/tests/interpret/sine.mlir
@@ -1,4 +1,6 @@
-// RUN: spu-translate --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=1 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=2 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=3 --interpret -split-input-file %s
 // AUTO GENERATED, DO NOT EDIT
 
 func.func @sine_op_test_f16_f16_p() {
diff --git a/libspu/compiler/tests/interpret/slice.mlir b/libspu/compiler/tests/interpret/slice.mlir
index e6ce68b2..b4b8bdbe 100644
--- a/libspu/compiler/tests/interpret/slice.mlir
+++ b/libspu/compiler/tests/interpret/slice.mlir
@@ -1,4 +1,6 @@
-// RUN: spu-translate --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=1 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=2 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=3 --interpret -split-input-file %s
 
 func.func @slice_op() {
   %operand = pphlo.constant dense<[[0, 0, 1, 0, 0, 1],
diff --git a/libspu/compiler/tests/interpret/sort.mlir b/libspu/compiler/tests/interpret/sort.mlir
index c0def6c4..837ded37 100644
--- a/libspu/compiler/tests/interpret/sort.mlir
+++ b/libspu/compiler/tests/interpret/sort.mlir
@@ -1,4 +1,6 @@
-// RUN: spu-translate --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=1 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=2 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=3 --interpret -split-input-file %s
 
 func.func @sort_stable() {
   %input0 = pphlo.constant dense<[[1, 2, 3], [3, 2, 1]]> : tensor<2x3xi64>
diff --git a/libspu/compiler/tests/interpret/sqrt.mlir b/libspu/compiler/tests/interpret/sqrt.mlir
index 249355fa..9248077a 100644
--- a/libspu/compiler/tests/interpret/sqrt.mlir
+++ b/libspu/compiler/tests/interpret/sqrt.mlir
@@ -1,4 +1,6 @@
-// RUN: spu-translate --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=1 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=2 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=3 --interpret -split-input-file %s
 // AUTO GENERATED, DO NOT EDIT
 
 func.func @sqrt_op_test_f64_f64_p() {
diff --git a/libspu/compiler/tests/interpret/subtract.mlir b/libspu/compiler/tests/interpret/subtract.mlir
index 3cab82eb..ce8b9633 100644
--- a/libspu/compiler/tests/interpret/subtract.mlir
+++ b/libspu/compiler/tests/interpret/subtract.mlir
@@ -1,4 +1,6 @@
-// RUN: spu-translate --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=1 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=2 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=3 --interpret -split-input-file %s
 // AUTO GENERATED, DO NOT EDIT
 
 func.func @subtract_op_test_i8_i8_pp() {
diff --git a/libspu/compiler/tests/interpret/tanh.mlir b/libspu/compiler/tests/interpret/tanh.mlir
index 4f42bbf7..413dc6d2 100644
--- a/libspu/compiler/tests/interpret/tanh.mlir
+++ b/libspu/compiler/tests/interpret/tanh.mlir
@@ -1,4 +1,6 @@
-// RUN: spu-translate --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=1 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=2 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=3 --interpret -split-input-file %s
 // AUTO GENERATED, DO NOT EDIT
 
 func.func @tanh_op_test_f16_f16_p() {
diff --git a/libspu/compiler/tests/interpret/transpose.mlir b/libspu/compiler/tests/interpret/transpose.mlir
index 126660db..d5ce9b6e 100644
--- a/libspu/compiler/tests/interpret/transpose.mlir
+++ b/libspu/compiler/tests/interpret/transpose.mlir
@@ -1,4 +1,6 @@
-// RUN: spu-translate --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=1 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=2 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=3 --interpret -split-input-file %s
 
 func.func @transpose_op_test_si32() {
   %0 = pphlo.constant dense<[[[1,2],[3,4],[5,6]], [[7,8],[9,10],[11,12]]]> : tensor<2x3x2xi32>
diff --git a/libspu/compiler/tests/interpret/while.mlir b/libspu/compiler/tests/interpret/while.mlir
index 2bcbd3ff..32789fae 100644
--- a/libspu/compiler/tests/interpret/while.mlir
+++ b/libspu/compiler/tests/interpret/while.mlir
@@ -1,4 +1,6 @@
-// RUN: spu-translate --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=1 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=2 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=3 --interpret -split-input-file %s
 
 func.func @while() {
   // int i = 0;
diff --git a/libspu/compiler/tests/interpret/xor.mlir b/libspu/compiler/tests/interpret/xor.mlir
index 21d3484d..c2e1f1ee 100644
--- a/libspu/compiler/tests/interpret/xor.mlir
+++ b/libspu/compiler/tests/interpret/xor.mlir
@@ -1,4 +1,6 @@
-// RUN: spu-translate --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=1 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=2 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=3 --interpret -split-input-file %s
 // AUTO GENERATED, DO NOT EDIT
 
 func.func @xor_op_test_i8_i8_pp() {
diff --git a/libspu/compiler/tools/spu-translate.cc b/libspu/compiler/tools/spu-translate.cc
index a37bd465..423321a5 100644
--- a/libspu/compiler/tools/spu-translate.cc
+++ b/libspu/compiler/tools/spu-translate.cc
@@ -46,6 +46,10 @@
 template <typename T>
 struct fmt::formatter<xt::xarray<T>> : ostream_formatter {};
 
+llvm::cl::opt<uint32_t> ProtocolKind(
+    "protocol_kind", llvm::cl::init(1),
+    llvm::cl::desc("1 for REF2k, 2 for SEMI2k, 3 for ABY3, 4 for Cheetah"));
+
 namespace mlir {
 
 namespace {
@@ -215,9 +219,29 @@ void evalModule(ModuleOp module) {
   runPasses(module);
 
   ::spu::RuntimeConfig conf;
-  conf.set_protocol(::spu::REF2K);
   conf.set_field(::spu::FM64);
   conf.set_enable_type_checker(true);
+  int numParties = 1;
+
+  switch (ProtocolKind.getValue()) {
+  case 1: {
+    conf.set_protocol(::spu::REF2K);
+    numParties = 1;
+    break;
+  }
+  case 2: {
+    conf.set_protocol(::spu::SEMI2K);
+    numParties = 2;
+    break;
+  }
+  case 3: {
+    conf.set_protocol(::spu::ABY3);
+    numParties = 3;
+    break;
+  }
+  }
+
+  SPDLOG_INFO(conf.DebugString());
 
   auto entry_function = spu::get_entrypoint(module);
   SPU_ENFORCE(entry_function, "main module not found");
@@ -227,7 +251,7 @@ void evalModule(ModuleOp module) {
   ::spu::device::ExecutionOptions opts;
 
   ::spu::mpc::utils::simulate(
-      1, [&](const std::shared_ptr<yacl::link::Context> &lctx) {
+      numParties, [&](const std::shared_ptr<yacl::link::Context> &lctx) {
         auto sctx = ::spu::kernel::test::makeSPUContext(conf, lctx);
 
         runRegion(&executor, &sctx, nullptr, entry_function.getBody(), {},
diff --git a/libspu/device/BUILD.bazel b/libspu/device/BUILD.bazel
index 1a071162..a821980c 100644
--- a/libspu/device/BUILD.bazel
+++ b/libspu/device/BUILD.bazel
@@ -53,6 +53,7 @@ spu_cc_library(
     srcs = ["executor.cc"],
     hdrs = ["executor.h"],
     deps = [
+        ":intrinsic_table",
         ":symbol_table",
         "//libspu:spu_cc_proto",
         "//libspu/core:context",
@@ -68,6 +69,7 @@ spu_cc_library(
     hdrs = ["api.h"],
     deps = [
         ":executor",
+        "//libspu:version",
         "//libspu/device/pphlo:pphlo_executor",
         "//libspu/device/utils:debug_dump_constant",
         "//libspu/dialect/utils",
diff --git a/libspu/device/api.cc b/libspu/device/api.cc
index 97077a93..5535ad85 100644
--- a/libspu/device/api.cc
+++ b/libspu/device/api.cc
@@ -29,6 +29,7 @@
 #include "libspu/device/utils/debug_dump_constant.h"
 #include "libspu/dialect/pphlo/IR/dialect.h"
 #include "libspu/dialect/utils/utils.h"
+#include "libspu/version.h"
 
 namespace spu::device {
 namespace {
@@ -290,6 +291,22 @@ void executeImpl(OpExecutor *executor, spu::SPUContext *sctx,
 
     SPU_ENFORCE(moduleOpRef, "MLIR parser failure");
 
+    if (!moduleOpRef.get()->hasAttr("pphlo.version")) {
+      // There are tests that has no version attributes.
+      // So treats this as a warning
+      SPDLOG_WARN("Missing ir version");
+    } else {
+      auto ir_version = mlir::dyn_cast<mlir::StringAttr>(
+                            moduleOpRef.get()->getAttr("pphlo.version"))
+                            .str();
+      if (ir_version != getVersionStr()) {
+        SPU_THROW(
+            "IR was generted by compiler {} and does not match current runtime "
+            "{}",
+            ir_version, getVersionStr());
+      }
+    }
+
     auto entry_function = mlir::spu::get_entrypoint(moduleOpRef.get());
     SPU_ENFORCE(entry_function, "main module not found");
 
diff --git a/libspu/device/executor.cc b/libspu/device/executor.cc
index 9093b628..5d03db2d 100644
--- a/libspu/device/executor.cc
+++ b/libspu/device/executor.cc
@@ -24,6 +24,8 @@
 #include "libspu/core/context.h"
 #include "libspu/core/prelude.h"
 #include "libspu/core/value.h"
+#include "libspu/device/intrinsic_table.h"
+#include "libspu/dialect/pphlo/IR/ops.h"
 
 namespace spu::device {
 
@@ -166,12 +168,15 @@ class OpExecTask final {
   OpExecTask() = default;
   explicit OpExecTask(std::unique_ptr<SPUContext> sctx, OpExecutor *executor,
                       SymbolScope *sscope, mlir::Operation *op,
-                      SymbolTableEvent *event)
+                      SymbolTableEvent *event,
+                      const llvm::SmallVector<mlir::Value> &extra_dependencies)
       : sctx_(std::move(sctx)),
         executor_(executor),
         sscope_(sscope),
         op_(op),
-        event_(event) {
+        event_(event),
+        extra_dependencies_(extra_dependencies.begin(),
+                            extra_dependencies.end()) {
     // If a op has nested regions, it may depend on more values than operands
     // FIXME: (azheng) Implement a better notify mechanism
     if (op->getNumRegions() > 0) {
@@ -228,8 +233,21 @@ class BlockParallelRunner final {
 
   std::vector<spu::Value> run(mlir::Block &block) {
     SymbolTableEvent st_event;
+    llvm::SmallVector<mlir::Value> extra_dependencies;
     for (auto &op : block.without_terminator()) {
-      task_queue_.emplace(sctx_->fork(), executor_, sscope_, &op, &st_event);
+      task_queue_.emplace(sctx_->fork(), executor_, sscope_, &op, &st_event,
+                          extra_dependencies);
+      // FIXME(jimi): DBG_PRINT has side effect but has no outputs. We should
+      // use more formal scheduling policy
+      if (auto custom_call = llvm::dyn_cast<mlir::spu::pphlo::CustomCallOp>(op);
+          custom_call && custom_call.getCallTargetName() == DBG_PRINT) {
+        continue;
+      }
+      auto hasSideEffect = op.getAttrOfType<mlir::BoolAttr>("has_side_effect");
+      if (hasSideEffect && hasSideEffect.getValue()) {
+        extra_dependencies.append(op.getResults().begin(),
+                                  op.getResults().end());
+      }
     }
 
     threads_.reserve(opts_.concurrency);
diff --git a/libspu/device/intrinsic_table.h b/libspu/device/intrinsic_table.h
index d56c5be9..c83d74d7 100644
--- a/libspu/device/intrinsic_table.h
+++ b/libspu/device/intrinsic_table.h
@@ -22,7 +22,8 @@
 #define    PREFER_A         "spu.prefer_a"
 #define    DBG_PRINT        "spu.dbg_print"
 #define    GATHER           "spu.gather"
-#define    MAKE_CACHE_VAR   "spu.make_cache_var"
+// should be consistent with python level
+#define    MAKE_CACHED_VAR  "spu.make_cached_var"
 #define    DROP_CACHED_VAR  "spu.drop_cached_var"
 
 // clang-format on
diff --git a/libspu/device/pphlo/pphlo_intrinsic_executor.cc b/libspu/device/pphlo/pphlo_intrinsic_executor.cc
index 01cb2565..7bfafc3b 100644
--- a/libspu/device/pphlo/pphlo_intrinsic_executor.cc
+++ b/libspu/device/pphlo/pphlo_intrinsic_executor.cc
@@ -50,7 +50,7 @@ std::vector<Value> intrinsic_dispatcher(SPUContext* ctx,
     return {zeros};
   }
 
-  if (name == MAKE_CACHE_VAR) {
+  if (name == MAKE_CACHED_VAR) {
     if (ctx->hasKernel("beaver_cache")) {
       SPU_ENFORCE(inputs.size() == 1);
       dynDispatch(ctx, "beaver_cache", inputs[0], true);
diff --git a/libspu/mpc/api.cc b/libspu/mpc/api.cc
index a27e81d0..2008e8aa 100644
--- a/libspu/mpc/api.cc
+++ b/libspu/mpc/api.cc
@@ -225,7 +225,7 @@ Type common_type_s(SPUContext* ctx, const Type& a, const Type& b) {
   } else if (a.isa<BShare>() && b.isa<AShare>()) {
     return b;
   } else if (a.isa<BShare>() && b.isa<BShare>()) {
-    return common_type_b(ctx, b, b);
+    return common_type_b(ctx, a, b);
   } else {
     SPU_THROW("should not be here, a={}, b={}", a, b);
   }
diff --git a/requirements.txt b/requirements.txt
index 695a2f1f..bea5a7d1 100644
--- a/requirements.txt
+++ b/requirements.txt
@@ -1,5 +1,5 @@
 grpcio>=1.42.0,!=1.48.0
-numpy>=1.22.0
+numpy>=1.22.0, < 2
 protobuf>=4, <5
 cloudpickle>=2.0.0
 multiprocess>=0.70.12.2
diff --git a/spu/tests/spu_runtime_test.py b/spu/tests/spu_runtime_test.py
index 938b4672..84d1956b 100644
--- a/spu/tests/spu_runtime_test.py
+++ b/spu/tests/spu_runtime_test.py
@@ -76,6 +76,31 @@ def test_raise(self):
         with self.assertRaisesRegex(RuntimeError, "stacktrace:.*"):
             sim(executable, x, y)
 
+    def test_wrong_version(self):
+        wsize = 1
+        config = spu_pb2.RuntimeConfig(
+            protocol=spu_pb2.ProtocolKind.REF2K,
+            field=spu_pb2.FieldType.FM64,
+        )
+
+        sim = Simulator(wsize, config)
+
+        # Give some insane ir
+        code = """
+module @no_version attributes {pphlo.version = "0.0.1"} {
+  func.func @main() -> tensor<1xf32> {
+    %0 = pphlo.constant dense<[1.0]> : tensor<1xf32>
+    return %0 : tensor<1xf32>
+  }
+}"""
+        executable = spu_pb2.ExecutableProto(
+            name="test",
+            code=code.encode(),
+        )
+
+        with self.assertRaisesRegex(RuntimeError, "IR was generted by compiler.*"):
+            sim(executable)
+
 
 if __name__ == '__main__':
     unittest.main()