Repo sync (#717)

.clang-tidy

@@ -9,7 +9,7 @@ Checks: "abseil-cleanup-ctad,
   bugprone-*,
   -bugprone-easily-swappable-parameters,
   -bugprone-implicit-widening-of-multiplication-result,
-  -bugprone-narrowing-conversions, # too many false positives around `std::size_t`  vs. `*::difference_type`.
+  -bugprone-narrowing-conversions,
   google-build-using-namespace,
   google-explicit-constructor,
   google-global-names-in-headers,
@@ -20,7 +20,7 @@ Checks: "abseil-cleanup-ctad,
   modernize-*,
   -modernize-use-trailing-return-type,
   -modernize-avoid-c-arrays,
-  -modernize-return-braced-init-list, # can hurt readability
+  -modernize-return-braced-init-list,
   -modernize-use-nodiscard,
   performance-*,
   readability-*,

libspu/BUILD.bazel

@@ -15,6 +15,7 @@
 load("@rules_cc//cc:defs.bzl", "cc_proto_library")
 load("@rules_proto//proto:defs.bzl", "proto_library")
 load("@rules_proto_grpc//python:defs.bzl", "python_proto_compile")
+load("//bazel:spu.bzl", "spu_cc_library")
 
 package(default_visibility = ["//visibility:public"])
 
@@ -34,3 +35,8 @@ python_proto_compile(
     prefix_path = "..",
     protos = ["//libspu:spu_proto"],
 )
+
+spu_cc_library(
+    name = "version",
+    hdrs = ["version.h"],
+)

libspu/core/context.h

@@ -113,7 +113,7 @@ class KernelEvalContext final {
   SPUContext* sctx_;
 
   std::vector<ParamType> params_;
-  ParamType output_;
+  std::vector<ParamType> outputs_;
 
  public:
   explicit KernelEvalContext(SPUContext* sctx) : sctx_(sctx) {}
@@ -139,20 +139,24 @@ class KernelEvalContext final {
   }
 
   size_t numParams() const { return params_.size(); }
+  size_t numOutputs() const { return outputs_.size(); }
 
   // Steal the output from this evaluation context.
   //
   // * usually called by kernel caller.
   template <typename T = Value>
-  T&& stealOutput() {
-    return std::move(std::get<T>(output_));
+  T&& consumeOutput(size_t pos) {
+    SPU_DEBUG_ONLY_ENFORCE(pos < outputs_.size(),
+                           "pos={} exceed num of outputs={}", pos,
+                           outputs_.size());
+    return std::move(std::get<T>(outputs_[pos]));
   }
 
   // Bind an input to this evaluation context.
   //
   // * usually called by kernel caller.
   template <typename T>
-  void bindParam(const T& in) {
+  void pushParam(const T& in) {
     params_.emplace_back(in);
   }
 
@@ -161,27 +165,28 @@ class KernelEvalContext final {
   // * usually called by kernel callee.
   template <typename T>
   const T& getParam(size_t pos) const {
-    SPU_ENFORCE(pos < params_.size(), "pos={} exceed num of inputs={}", pos,
-                params_.size());
+    SPU_DEBUG_ONLY_ENFORCE(pos < params_.size(),
+                           "pos={} exceed num of inputs={}", pos,
+                           params_.size());
     return std::get<T>(params_[pos]);
   }
 
   // Set the output.
   //
   // * usually called by kernel callee.
   template <typename T = Value>
-  void setOutput(T&& out) {
-    output_ = std::forward<T>(out);
+  void pushOutput(T&& out) {
+    outputs_.emplace_back(std::forward<T>(out));
   }
 };
 
 namespace detail {
 
 template <typename First, typename... Args>
 void bindParams(KernelEvalContext* ectx, First&& head, Args&&... tail) {
-  ectx->bindParam(std::forward<First>(head));
+  ectx->pushParam(std::forward<First>(head));
   if constexpr (sizeof...(Args) > 0) {
-    return bindParams(ectx, std::forward<Args>(tail)...);
+    bindParams(ectx, std::forward<Args>(tail)...);
   }
 }
 
@@ -203,7 +208,7 @@ Ret dynDispatch(SPUContext* sctx, const std::string& name, Args&&... args) {
   kernel->evaluate(&ectx);
 
   // 4. steal the result and return it.
-  return ectx.stealOutput<Ret>();
+  return ectx.consumeOutput<Ret>(0);
 }
 
 // helper class

libspu/core/prelude.h

@@ -42,9 +42,11 @@
                __VA_ARGS__)  // NOLINT, readability-simplify-boolean-expr
 
 #ifdef NDEBUG
-#define SPU_DEBUG_ONLY_THROW static_cast<void>
+#define SPU_DEBUG_ONLY_THROW(...) static_cast<void>(0)
+#define SPU_DEBUG_ONLY_ENFORCE(...) static_cast<void>(0)
 #else
-#define SPU_DEBUG_ONLY_THROW YACL_THROW
+#define SPU_DEBUG_ONLY_THROW(...) YACL_THROW(__VA_ARGS__)
+#define SPU_DEBUG_ONLY_ENFORCE(COND, ...) SPU_ENFORCE(COND, __VA_ARGS__)
 #endif
 
 // Force compiler to inline something regardless of optimization level.

libspu/mpc/aby3/boolean.cc

@@ -36,7 +36,7 @@ void CommonTypeB::evaluate(KernelEvalContext* ctx) const {
   const size_t out_nbits = std::max(lhs_nbits, rhs_nbits);
   const PtType out_btype = calcBShareBacktype(out_nbits);
 
-  ctx->setOutput(makeType<BShrTy>(out_btype, out_nbits));
+  ctx->pushOutput(makeType<BShrTy>(out_btype, out_nbits));
 }
 
 NdArrayRef CastTypeB::proc(KernelEvalContext*, const NdArrayRef& in,

libspu/mpc/aby3/conversion.cc

@@ -922,7 +922,7 @@ void CommonTypeV::evaluate(KernelEvalContext* ctx) const {
   const auto* lhs_v = lhs.as<Priv2kTy>();
   const auto* rhs_v = rhs.as<Priv2kTy>();
 
-  ctx->setOutput(makeType<AShrTy>(std::max(lhs_v->field(), rhs_v->field())));
+  ctx->pushOutput(makeType<AShrTy>(std::max(lhs_v->field(), rhs_v->field())));
 }
 
 }  // namespace spu::mpc::aby3

libspu/mpc/cheetah/boolean_semi2k.cc

@@ -53,7 +53,7 @@ void CommonTypeB::evaluate(KernelEvalContext* ctx) const {
               "cheetah always use same bshare field, lhs={}, rhs={}", lhs_field,
               rhs_field);
 
-  ctx->setOutput(makeType<BShrTy>(lhs_field, std::max(lhs_nbits, rhs_nbits)));
+  ctx->pushOutput(makeType<BShrTy>(lhs_field, std::max(lhs_nbits, rhs_nbits)));
 }
 
 NdArrayRef CastTypeB::proc(KernelEvalContext*, const NdArrayRef& in,

libspu/mpc/cheetah/conversion.cc

@@ -77,7 +77,7 @@ void CommonTypeV::evaluate(KernelEvalContext* ctx) const {
   const auto* lhs_v = lhs.as<Priv2kTy>();
   const auto* rhs_v = rhs.as<Priv2kTy>();
 
-  ctx->setOutput(makeType<AShrTy>(std::max(lhs_v->field(), rhs_v->field())));
+  ctx->pushOutput(makeType<AShrTy>(std::max(lhs_v->field(), rhs_v->field())));
 }
 
 }  // namespace spu::mpc::cheetah

libspu/mpc/common/pv2k.cc

@@ -97,7 +97,7 @@ class MakeP : public Kernel {
   ce::CExpr comm() const override { return ce::Const(0); }
 
   void evaluate(KernelEvalContext* ctx) const override {
-    ctx->setOutput(
+    ctx->pushOutput(
         proc(ctx, ctx->getParam<uint128_t>(0), ctx->getParam<Shape>(1)));
   }
 

libspu/mpc/kernel.cc

@@ -21,15 +21,15 @@ void RandKernel::evaluate(KernelEvalContext* ctx) const {
 
   auto res = proc(ctx, shape);
 
-  ctx->setOutput(WrapValue(res));
+  ctx->pushOutput(WrapValue(res));
 }
 
 void UnaryKernel::evaluate(KernelEvalContext* ctx) const {
   const auto& in = ctx->getParam<Value>(0);
 
   auto res = proc(ctx, UnwrapValue(in));
 
-  ctx->setOutput(WrapValue(res));
+  ctx->pushOutput(WrapValue(res));
 }
 
 void RevealToKernel::evaluate(KernelEvalContext* ctx) const {
@@ -38,7 +38,7 @@ void RevealToKernel::evaluate(KernelEvalContext* ctx) const {
 
   auto res = proc(ctx, UnwrapValue(in), rank);
 
-  ctx->setOutput(WrapValue(res));
+  ctx->pushOutput(WrapValue(res));
 }
 
 void ShiftKernel::evaluate(KernelEvalContext* ctx) const {
@@ -47,7 +47,7 @@ void ShiftKernel::evaluate(KernelEvalContext* ctx) const {
 
   auto res = proc(ctx, UnwrapValue(in), bits);
 
-  ctx->setOutput(WrapValue(res));
+  ctx->pushOutput(WrapValue(res));
 }
 
 void BinaryKernel::evaluate(KernelEvalContext* ctx) const {
@@ -59,7 +59,7 @@ void BinaryKernel::evaluate(KernelEvalContext* ctx) const {
 
   auto z = proc(ctx, UnwrapValue(lhs), UnwrapValue(rhs));
 
-  ctx->setOutput(WrapValue(z));
+  ctx->pushOutput(WrapValue(z));
 }
 
 void MatmulKernel::evaluate(KernelEvalContext* ctx) const {
@@ -69,7 +69,7 @@ void MatmulKernel::evaluate(KernelEvalContext* ctx) const {
   SPU_ENFORCE(lhs.shape()[1] == rhs.shape()[0], "invalid shape {} {}", lhs,
               rhs);
 
-  ctx->setOutput(WrapValue(proc(ctx, lhs.data(), rhs.data())));
+  ctx->pushOutput(WrapValue(proc(ctx, lhs.data(), rhs.data())));
 }
 
 void Conv2DKernel::evaluate(KernelEvalContext* ctx) const {
@@ -80,7 +80,7 @@ void Conv2DKernel::evaluate(KernelEvalContext* ctx) const {
 
   auto z = proc(ctx, UnwrapValue(lhs), UnwrapValue(rhs), stride_h, stride_w);
 
-  ctx->setOutput(WrapValue(z));
+  ctx->pushOutput(WrapValue(z));
 }
 
 void BitrevKernel::evaluate(KernelEvalContext* ctx) const {
@@ -90,7 +90,7 @@ void BitrevKernel::evaluate(KernelEvalContext* ctx) const {
 
   auto z = proc(ctx, UnwrapValue(in), start, end);
 
-  ctx->setOutput(WrapValue(z));
+  ctx->pushOutput(WrapValue(z));
 }
 
 void TruncAKernel::evaluate(KernelEvalContext* ctx) const {
@@ -100,7 +100,7 @@ void TruncAKernel::evaluate(KernelEvalContext* ctx) const {
 
   auto z = proc(ctx, UnwrapValue(in), bits, sign);
 
-  ctx->setOutput(WrapValue(z));
+  ctx->pushOutput(WrapValue(z));
 }
 
 void BitSplitKernel::evaluate(KernelEvalContext* ctx) const {
@@ -109,7 +109,7 @@ void BitSplitKernel::evaluate(KernelEvalContext* ctx) const {
 
   auto res = proc(ctx, UnwrapValue(in), stride);
 
-  ctx->setOutput(WrapValue(res));
+  ctx->pushOutput(WrapValue(res));
 }
 
 void CastTypeKernel::evaluate(KernelEvalContext* ctx) const {
@@ -118,7 +118,7 @@ void CastTypeKernel::evaluate(KernelEvalContext* ctx) const {
 
   auto res = proc(ctx, UnwrapValue(val), to_type);
 
-  ctx->setOutput(WrapValue(res));
+  ctx->pushOutput(WrapValue(res));
 }
 
 void PermKernel::evaluate(KernelEvalContext* ctx) const {
@@ -131,7 +131,7 @@ void PermKernel::evaluate(KernelEvalContext* ctx) const {
 
   auto z = proc(ctx, UnwrapValue(x), UnwrapValue(y));
 
-  ctx->setOutput(WrapValue(z));
+  ctx->pushOutput(WrapValue(z));
 }
 
 void GenInvPermKernel::evaluate(KernelEvalContext* ctx) const {
@@ -141,7 +141,7 @@ void GenInvPermKernel::evaluate(KernelEvalContext* ctx) const {
 
   auto y = proc(ctx, UnwrapValue(in), is_ascending);
 
-  ctx->setOutput(WrapValue(y));
+  ctx->pushOutput(WrapValue(y));
 }
 
 void MergeKeysKernel::evaluate(KernelEvalContext* ctx) const {
@@ -153,7 +153,7 @@ void MergeKeysKernel::evaluate(KernelEvalContext* ctx) const {
   }
   auto y = proc(ctx, inputs, is_ascending);
 
-  ctx->setOutput(WrapValue(y));
+  ctx->pushOutput(WrapValue(y));
 }
 
 void BroadcastKernel::evaluate(KernelEvalContext* ctx) const {
@@ -163,7 +163,7 @@ void BroadcastKernel::evaluate(KernelEvalContext* ctx) const {
 
   auto z = proc(ctx, UnwrapValue(in), to_shape, in_dims);
 
-  ctx->setOutput(WrapValue(z));
+  ctx->pushOutput(WrapValue(z));
 }
 
 void DimsBasedKernel::evaluate(KernelEvalContext* ctx) const {
@@ -172,7 +172,7 @@ void DimsBasedKernel::evaluate(KernelEvalContext* ctx) const {
 
   auto z = proc(ctx, UnwrapValue(in), axes);
 
-  ctx->setOutput(WrapValue(z));
+  ctx->pushOutput(WrapValue(z));
 }
 
 void ShapeBasedKernel::evaluate(KernelEvalContext* ctx) const {
@@ -181,7 +181,7 @@ void ShapeBasedKernel::evaluate(KernelEvalContext* ctx) const {
 
   auto z = proc(ctx, UnwrapValue(in), to_shape);
 
-  ctx->setOutput(WrapValue(z));
+  ctx->pushOutput(WrapValue(z));
 }
 
 void ExtractSliceKernel::evaluate(KernelEvalContext* ctx) const {
@@ -192,7 +192,7 @@ void ExtractSliceKernel::evaluate(KernelEvalContext* ctx) const {
 
   auto z = proc(ctx, UnwrapValue(in), start, end, strides);
 
-  ctx->setOutput(WrapValue(z));
+  ctx->pushOutput(WrapValue(z));
 }
 
 void UpdateSliceKernel::evaluate(KernelEvalContext* ctx) const {
@@ -202,7 +202,7 @@ void UpdateSliceKernel::evaluate(KernelEvalContext* ctx) const {
 
   auto z = proc(ctx, UnwrapValue(in), UnwrapValue(update), start);
 
-  ctx->setOutput(WrapValue(z));
+  ctx->pushOutput(WrapValue(z));
 }
 
 void PadKernel::evaluate(KernelEvalContext* ctx) const {
@@ -215,7 +215,7 @@ void PadKernel::evaluate(KernelEvalContext* ctx) const {
   auto z = proc(ctx, UnwrapValue(in), UnwrapValue(padding_value), edge_low,
                 edge_high, interior_padding);
 
-  ctx->setOutput(WrapValue(z));
+  ctx->pushOutput(WrapValue(z));
 }
 
 void ConcateKernel::evaluate(KernelEvalContext* ctx) const {
@@ -230,7 +230,7 @@ void ConcateKernel::evaluate(KernelEvalContext* ctx) const {
 
   auto z = proc(ctx, unwrapped, axis);
 
-  ctx->setOutput(WrapValue(z));
+  ctx->pushOutput(WrapValue(z));
 }
 
 void OramOneHotKernel::evaluate(KernelEvalContext* ctx) const {
@@ -242,7 +242,7 @@ void OramOneHotKernel::evaluate(KernelEvalContext* ctx) const {
 
   auto res = proc(ctx, UnwrapValue(target), s);
 
-  ctx->setOutput(WrapValue(res));
+  ctx->pushOutput(WrapValue(res));
 }
 
 void OramReadKernel::evaluate(KernelEvalContext* ctx) const {
@@ -256,7 +256,7 @@ void OramReadKernel::evaluate(KernelEvalContext* ctx) const {
   SPU_ENFORCE(onehot.shape()[1] == db.shape()[0],
               "onehot and database shape mismatch");
 
-  ctx->setOutput(
+  ctx->pushOutput(
       WrapValue(proc(ctx, UnwrapValue(onehot), UnwrapValue(db), offset)));
 }
 

libspu/mpc/ref2k/ref2k.cc

@@ -58,7 +58,7 @@ class Ref2kCommonTypeS : public Kernel {
     SPU_TRACE_MPC_DISP(ctx, lhs, rhs);
     SPU_ENFORCE(lhs.isa<Ref2kSecrTy>(), "invalid type, got={}", lhs);
     SPU_ENFORCE(rhs.isa<Ref2kSecrTy>(), "invalid type, got={}", rhs);
-    ctx->setOutput(lhs);
+    ctx->pushOutput(lhs);
   }
 };
 
@@ -79,7 +79,7 @@ class Ref2kCommonTypeV : public Kernel {
     const auto* lhs_v = lhs.as<Priv2kTy>();
     const auto* rhs_v = rhs.as<Priv2kTy>();
 
-    ctx->setOutput(
+    ctx->pushOutput(
         makeType<Ref2kSecrTy>(std::max(lhs_v->field(), rhs_v->field())));
   }
 };