repo-sync-2024-07-16T11:27:49+0800

anakinxc · Jul 16, 2024 · 59b6657 · 59b6657
1 parent f329677
commit 59b6657
Show file tree

Hide file tree

Showing 187 changed files with 1,197 additions and 994 deletions.
diff --git a/.clang-tidy b/.clang-tidy
@@ -28,7 +28,8 @@ Checks: "abseil-cleanup-ctad,
   -readability-identifier-length,
   -readability-function-cognitive-complexity,
   -readability-magic-numbers,
-  -readability-named-parameter"
+  -readability-named-parameter,
+  -readability-convert-member-functions-to-static"
 
 CheckOptions:
   - key: bugprone-argument-comment.StrictMode

diff --git a/CHANGELOG.md b/CHANGELOG.md
@@ -10,8 +10,9 @@
 >
 > please add your unreleased change here.
 
-## TBD
+## 20240716
 
+- [SPU] 0.9.2b0 release
 - [Feature] Support jax.numpy.bitwise_count
 - [Bugfix] Fix jax.numpy.signbit wrong answer with very large input
 

diff --git a/bazel/repositories.bzl b/bazel/repositories.bzl
@@ -136,8 +136,8 @@ def _bazel_skylib():
     )
 
 def _com_github_openxla_xla():
-    OPENXLA_COMMIT = "9b0dd58c9b625a2e958f4fc7787a1ff5c95dbb40"
-    OPENXLA_SHA256 = "f150c5b49e4d4497aae2c79232f1efe2baccaa72223b21dc8715be73eab74417"
+    OPENXLA_COMMIT = "8533a6869ae02fb3b15a8a12739a982fc3c9f6e7"
+    OPENXLA_SHA256 = "d5b076825c992f59542f6b94e5480c7e7c6c627cd18c80ec60b6d5b295c160d4"
 
     # We need openxla to handle xla/mhlo/stablehlo
     maybe(

diff --git a/docs/development/add_protocols.rst b/docs/development/add_protocols.rst
@@ -248,7 +248,7 @@ When kernels are implemented and registered, a new protocol is finally added.
     auto* prg_state = ctx->getState<PrgState>();
 
     // dispatch the real implementation to different fields
-    return DISPATCH_ALL_FIELDS(field, "aby3.mulAA", [&]() {
+    return DISPATCH_ALL_FIELDS(field, [&]() {
       // the real protocol implementation
       ...
     });

diff --git a/examples/python/ml/jax_lr/README.md b/examples/python/ml/jax_lr/README.md
@@ -5,7 +5,7 @@ This example demonstrates how to use SPU to train a logistic regression model pr
 1. Launch SPU backend runtime
 
     ```sh
-    bazel run -c opt //examples/python/utils:nodectl -- up
+    bazel run -c opt //examples/python/utils:nodectl -- -c examples/python/conf/2pc_semi2k.json up
     ```
 
 2. Run `jax_lr` example

diff --git a/experimental/squirrel/bin_matvec_prot_test.cc b/experimental/squirrel/bin_matvec_prot_test.cc
@@ -110,7 +110,7 @@ TEST_P(BinMatVecProtTest, Basic) {
   });
   NdArrayRef reveal = ring_add(out_shr[0], out_shr[1]);
 
-  DISPATCH_ALL_FIELDS(field, "", [&]() {
+  DISPATCH_ALL_FIELDS(field, [&]() {
     NdArrayView<ring2k_t> _vec(vec);
     auto expected = BinAccumuate<ring2k_t>(_vec, mat);
     NdArrayView<ring2k_t> got(reveal);
@@ -160,7 +160,7 @@ TEST_P(BinMatVecProtTest, WithIndicator) {
   });
   NdArrayRef reveal = ring_add(out_shr[0], out_shr[1]);
 
-  DISPATCH_ALL_FIELDS(field, "", [&]() {
+  DISPATCH_ALL_FIELDS(field, [&]() {
     NdArrayView<ring2k_t> _vec(vec);
     auto expected =
         BinAccumuate<ring2k_t>(_vec, mat, absl::MakeConstSpan(indicator));

diff --git a/experimental/squirrel/objectives.cc b/experimental/squirrel/objectives.cc
@@ -272,9 +272,9 @@ namespace {
     res = NdArrayRef(makeType<mpc::cheetah::AShrTy>(ftype), in.shape());
   }
 
-  return DISPATCH_ALL_FIELDS(field, "cheetah.ring_cast", [&]() {
+  return DISPATCH_ALL_FIELDS(field, [&]() {
     using from_ring2k_t = ring2k_t;
-    return DISPATCH_ALL_FIELDS(ftype, "cheetah.ring_cast", [&]() {
+    return DISPATCH_ALL_FIELDS(ftype, [&]() {
       using to_ring2k_t = ring2k_t;
       NdArrayView<const from_ring2k_t> _in(in);
       NdArrayView<to_ring2k_t> _res(res);
@@ -383,7 +383,7 @@ spu::Value Logistic(spu::SPUContext* ctx, const spu::Value& x) {
 spu::Value Sigmoid(spu::SPUContext* ctx, const spu::Value& x) {
   namespace sk = spu::kernel;
   auto c05 = sk::hlo::Constant(ctx, 0.5F, x.shape());
-  auto half = sk::hal::right_shift_arithmetic(ctx, x, 1);
+  auto half = sk::hal::right_shift_arithmetic(ctx, x, {1});
   auto divisor = sk::hlo::Add(ctx, sk::hlo::Constant(ctx, 1, x.shape()),
                               sk::hal::f_square(ctx, x));
   return sk::hlo::Add(ctx, c05,

diff --git a/experimental/squirrel/tree_build_worker.cc b/experimental/squirrel/tree_build_worker.cc
@@ -230,7 +230,7 @@ void AccumulateHistogram(spu::NdArrayRef buckets_share, size_t nfeatures,
   // The buckets belong to the i-th feature is
   // `buckets[i*bucket_size:(i+1)*bucket_size]`
   auto field = buckets_share.eltype().as<RingTy>()->field();
-  DISPATCH_ALL_FIELDS(field, "AccumulateHistogram", [&]() {
+  DISPATCH_ALL_FIELDS(field, [&]() {
     NdArrayView<ring2k_t> histogram(buckets_share);
     for (size_t j = 0; j < nfeatures; ++j) {
       size_t start = j * bucket_size;

diff --git a/experimental/squirrel/utils.cc b/experimental/squirrel/utils.cc
@@ -167,7 +167,7 @@ spu::Value MulPrivateArithWithPrivateBoolean(spu::SPUContext* ctx,
       &kctx, arith.data(),
       [&](const NdArrayRef& input,
           const std::shared_ptr<spu::mpc::cheetah::BasicOTProtocols>& base_ot) {
-        return DISPATCH_ALL_FIELDS(ft, "ot", [&]() {
+        return DISPATCH_ALL_FIELDS(ft, [&]() {
           NdArrayRef ot_out = spu::mpc::ring_zeros(ft, input.shape());
           auto inp = absl::MakeConstSpan(&input.at<ring2k_t>(0), input.numel());
           auto oup = absl::MakeSpan(&ot_out.at<ring2k_t>(0), ot_out.numel());
@@ -193,7 +193,7 @@ spu::Value MulPrivateArithWithPrivateBoolean(spu::SPUContext* ctx,
       &kctx, boolean,
       [&](absl::Span<const uint8_t> input,
           const std::shared_ptr<spu::mpc::cheetah::BasicOTProtocols>& base_ot) {
-        return DISPATCH_ALL_FIELDS(ft, "ot", [&]() {
+        return DISPATCH_ALL_FIELDS(ft, [&]() {
           NdArrayRef ot_out = spu::mpc::ring_zeros(ft, {(int64_t)input.size()});
           auto oup = absl::MakeSpan(&ot_out.at<ring2k_t>(0), input.size());
           base_ot->GetReceiverCOT()->RecvCAMCC(input, oup);
@@ -222,7 +222,7 @@ spu::Value MulArithShareWithANDBoolShare(spu::SPUContext* ctx,
           std::shared_ptr<spu::mpc::cheetah::BasicOTProtocols> base_ot) {
         NdArrayRef out(x.eltype(), x.shape());
 
-        DISPATCH_ALL_FIELDS(ft, "camcc", [&]() {
+        DISPATCH_ALL_FIELDS(ft, [&]() {
           spu::NdArrayView<const ring2k_t> _ashr(x);
           auto oup = absl::MakeSpan(&out.at<ring2k_t>(0), y.size());
           std::vector<ring2k_t> corr(y.size());

diff --git a/experimental/squirrel/utils_test.cc b/experimental/squirrel/utils_test.cc
@@ -85,7 +85,7 @@ TEST_F(UtilsTest, ReduceSum) {
         const double fxp = std::pow(2., rt_config.fxp_fraction_bits());
         auto flatten = got.data().reshape({got.numel()});
 
-        DISPATCH_ALL_FIELDS(field, "check", [&]() {
+        DISPATCH_ALL_FIELDS(field, [&]() {
           using s2k = std::make_signed<ring2k_t>::type;
           NdArrayView<s2k> got(flatten);
           for (int64_t i = 0; i < expected.numel(); ++i) {
@@ -136,7 +136,7 @@ TEST_F(UtilsTest, ArgMax) {
       if (lctx->Rank() == 0) {
         auto flatten = got.data().reshape({got.numel()});
 
-        DISPATCH_ALL_FIELDS(field, "check", [&]() {
+        DISPATCH_ALL_FIELDS(field, [&]() {
           NdArrayView<ring2k_t> got(flatten);
           for (size_t i = 0; i < expected.size(); ++i) {
             ASSERT_EQ(expected(i), got[i]);

diff --git a/libspu/compiler/common/compilation_context.h b/libspu/compiler/common/compilation_context.h
@@ -16,7 +16,6 @@
 
 #include <filesystem>
 #include <memory>
-#include <string>
 
 #include "mlir/IR/MLIRContext.h"
 #include "mlir/Pass/PassManager.h"

diff --git a/libspu/compiler/compile.cc b/libspu/compiler/compile.cc
@@ -14,8 +14,6 @@
 
 #include "libspu/compiler/compile.h"
 
-#include "mlir/IR/BuiltinOps.h"
-
 #include "libspu/compiler/codegen/codegen.h"
 #include "libspu/compiler/common/compilation_context.h"
 #include "libspu/compiler/core/core.h"

diff --git a/libspu/compiler/core/core.cc b/libspu/compiler/core/core.cc
@@ -16,7 +16,6 @@
 
 #include "mlir/Dialect/Func/IR/FuncOps.h"
 #include "mlir/IR/BuiltinOps.h"
-#include "mlir/Pass/Pass.h"
 #include "mlir/Pass/PassManager.h"
 #include "mlir/Transforms/Passes.h"
 

diff --git a/libspu/compiler/front_end/fe.cc b/libspu/compiler/front_end/fe.cc
@@ -21,7 +21,6 @@
 #include "mlir/Pass/Pass.h"
 #include "mlir/Pass/PassManager.h"
 #include "mlir/Transforms/Passes.h"
-#include "spdlog/spdlog.h"
 #include "stablehlo/dialect/StablehloOps.h"
 #include "xla/mlir_hlo/mhlo/IR/hlo_ops.h"
 #include "xla/mlir_hlo/mhlo/transforms/passes.h"

diff --git a/libspu/compiler/tests/interpret/abs.mlir b/libspu/compiler/tests/interpret/abs.mlir
@@ -1,6 +1,8 @@
 // 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
+// RUN: spu-translate --protocol_kind=4 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=5 --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
@@ -1,6 +1,8 @@
 // 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
+// RUN: spu-translate --protocol_kind=4 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=5 --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
@@ -1,6 +1,8 @@
 // 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
+// RUN: spu-translate --protocol_kind=4 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=5 --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
@@ -1,6 +1,8 @@
 // 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
+// RUN: spu-translate --protocol_kind=4 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=5 --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
@@ -1,6 +1,8 @@
 // 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
+// RUN: spu-translate --protocol_kind=4 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=5 --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
@@ -1,6 +1,8 @@
 // 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
+// RUN: spu-translate --protocol_kind=4 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=5 --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
@@ -1,6 +1,8 @@
 // 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
+// RUN: spu-translate --protocol_kind=4 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=5 --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
@@ -1,6 +1,8 @@
 // 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
+// RUN: spu-translate --protocol_kind=4 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=5 --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
@@ -1,6 +1,8 @@
 // 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
+// RUN: spu-translate --protocol_kind=4 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=5 --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
@@ -1,6 +1,8 @@
 // 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
+// RUN: spu-translate --protocol_kind=4 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=5 --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
@@ -1,6 +1,8 @@
 // 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
+// RUN: spu-translate --protocol_kind=4 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=5 --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
@@ -1,6 +1,8 @@
 // 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
+// RUN: spu-translate --protocol_kind=4 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=5 --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
@@ -1,6 +1,8 @@
 // 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
+// RUN: spu-translate --protocol_kind=4 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=5 --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
@@ -1,6 +1,8 @@
 // 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
+// RUN: spu-translate --protocol_kind=4 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=5 --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
@@ -1,6 +1,8 @@
 // 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
+// RUN: spu-translate --protocol_kind=4 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=5 --interpret -split-input-file %s
 
 func.func @dynamic_slice() {
   %operand = pphlo.constant dense<[[1, 1, 1],

diff --git a/libspu/compiler/tests/interpret/dynamic_update_slice.mlir b/libspu/compiler/tests/interpret/dynamic_update_slice.mlir
@@ -1,6 +1,8 @@
 // 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
+// RUN: spu-translate --protocol_kind=4 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=5 --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
@@ -1,6 +1,8 @@
 // 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
+// RUN: spu-translate --protocol_kind=4 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=5 --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
@@ -1,6 +1,8 @@
 // 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
+// RUN: spu-translate --protocol_kind=4 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=5 --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
@@ -1,6 +1,8 @@
 // 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
+// RUN: spu-translate --protocol_kind=4 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=5 --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
@@ -1,6 +1,8 @@
 // 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
+// RUN: spu-translate --protocol_kind=4 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=5 --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
@@ -65,6 +65,7 @@ def TestCase(inputs, expected, checker='expect_eq', tol=None):
     "or",
     # "popcnt",
     "power",
+    "reciprocal",
     "reshape",
     "round_afz",
     "rsqrt",
@@ -99,6 +100,14 @@ def TestCase(inputs, expected, checker='expect_eq', tol=None):
         f.write(
             "// RUN: spu-translate --protocol_kind=3 --interpret -split-input-file %s\n"
         )
+        f.write(
+            "// RUN: spu-translate --protocol_kind=4 --interpret -split-input-file %s\n"
+        )
+        # Some test values in max and min are not supported by protocol 5.
+        if test not in ["max", "min"]:
+            f.write(
+                "// RUN: spu-translate --protocol_kind=5 --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
@@ -1,6 +1,8 @@
 // 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
+// RUN: spu-translate --protocol_kind=4 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=5 --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
@@ -1,6 +1,8 @@
 // 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
+// RUN: spu-translate --protocol_kind=4 --interpret -split-input-file %s
+// RUN: spu-translate --protocol_kind=5 --interpret -split-input-file %s
 // AUTO GENERATED, DO NOT EDIT
 
 func.func @greater_equal_op_test_i64_i1_pp() {