From 6fc5dc37479e5fdecde70fc7654b658e1f4bff31 Mon Sep 17 00:00:00 2001
From: Jin Shang <shangjin1997@gmail.com>
Date: Thu, 22 Jun 2023 13:32:45 +0800
Subject: [PATCH 01/12] add integer round kernels

---
 cpp/src/arrow/compute/kernels/scalar_round.cc | 744 ++++++++++++------
 .../kernels/scalar_round_arithmetic_test.cc   | 312 +++++++-
 docs/source/cpp/compute.rst                   |  28 +-
 3 files changed, 811 insertions(+), 273 deletions(-)

diff --git a/cpp/src/arrow/compute/kernels/scalar_round.cc b/cpp/src/arrow/compute/kernels/scalar_round.cc
index fc2cb5b8a6ee1..522aff38e2cab 100644
--- a/cpp/src/arrow/compute/kernels/scalar_round.cc
+++ b/cpp/src/arrow/compute/kernels/scalar_round.cc
@@ -25,6 +25,7 @@
 #include "arrow/compare.h"
 #include "arrow/compute/api_scalar.h"
 #include "arrow/compute/cast.h"
+#include "arrow/compute/kernel.h"
 #include "arrow/compute/kernels/base_arithmetic_internal.h"
 #include "arrow/compute/kernels/common_internal.h"
 #include "arrow/compute/kernels/util_internal.h"
@@ -34,6 +35,7 @@
 #include "arrow/util/int_util_overflow.h"
 #include "arrow/util/macros.h"
 #include "arrow/visit_scalar_inline.h"
+#include "arrow/visit_type_inline.h"
 
 namespace arrow {
 
@@ -43,8 +45,7 @@ using internal::MultiplyWithOverflow;
 using internal::NegateWithOverflow;
 using internal::SubtractWithOverflow;
 
-namespace compute {
-namespace internal {
+namespace compute::internal {
 
 using applicator::ScalarBinary;
 using applicator::ScalarBinaryEqualTypes;
@@ -56,6 +57,9 @@ using applicator::ScalarUnaryNotNullStateful;
 
 namespace {
 
+// ----------------------------------------------------------------------
+// Begin utility structs for round kernels
+
 // Convenience visitor to detect if a numeric Scalar is positive.
 struct IsPositiveVisitor {
   bool result = false;
@@ -82,9 +86,25 @@ bool IsPositive(const Scalar& scalar) {
 // N.B. take care not to conflict with type_traits.h as that can cause surprises in a
 // unity build
 
+// A constexpr helper struct to compute powers of 10 at compile time
+// Can use a consteval function once we force C++20
+template <int Exp>
+struct Pow10Struct {
+ private:
+  static constexpr uint64_t half_pow = Pow10Struct<Exp / 2>::value;
+
+ public:
+  static constexpr uint64_t value = half_pow * half_pow * (Exp % 2 ? 10 : 1);
+};
+
+template <>
+struct Pow10Struct<0> {
+  static constexpr uint64_t value = 1;
+};
+
 struct RoundUtil {
   // Calculate powers of ten with arbitrary integer exponent
-  template <typename T = double>
+  template <typename T>
   static enable_if_floating_value<T> Pow10(int64_t power) {
     static constexpr T lut[] = {1e0F, 1e1F, 1e2F,  1e3F,  1e4F,  1e5F,  1e6F,  1e7F,
                                 1e8F, 1e9F, 1e10F, 1e11F, 1e12F, 1e13F, 1e14F, 1e15F};
@@ -96,8 +116,30 @@ struct RoundUtil {
     }
     return (power >= 0) ? pow10 : (1 / pow10);
   }
+
+  // Calculate powers of ten with arbitrary integer exponent
+  template <typename T>
+  static enable_if_integer_value<T> Pow10(int64_t power) {
+    DCHECK(power >= 0);
+
+    static constexpr uint64_t lut[] = {
+        Pow10Struct<0>::value,  Pow10Struct<1>::value,  Pow10Struct<2>::value,
+        Pow10Struct<3>::value,  Pow10Struct<4>::value,  Pow10Struct<5>::value,
+        Pow10Struct<6>::value,  Pow10Struct<7>::value,  Pow10Struct<8>::value,
+        Pow10Struct<9>::value,  Pow10Struct<10>::value, Pow10Struct<11>::value,
+        Pow10Struct<12>::value, Pow10Struct<13>::value, Pow10Struct<14>::value,
+        Pow10Struct<15>::value, Pow10Struct<16>::value, Pow10Struct<17>::value,
+        Pow10Struct<18>::value, Pow10Struct<19>::value};
+
+    auto digits10 = std::numeric_limits<T>::digits10;
+    return lut[std::min(power, static_cast<int64_t>(digits10))];
+  }
 };
 
+// End utility structs for round kernels
+// ----------------------------------------------------------------------
+// Begin round implementations for single scalar
+
 // Specializations of rounding implementations for round kernels
 template <typename Type, RoundMode>
 struct RoundImpl;
@@ -117,6 +159,21 @@ struct RoundImpl<Type, RoundMode::DOWN> {
       (*val) -= pow10;
     }
   }
+
+  template <typename T = Type>
+  static enable_if_integer_value<T> Round(const T val, const T floor, const T multiple,
+                                          Status* st) {
+    if constexpr (is_signed_integer_value<T>::value) {
+      if (ARROW_PREDICT_FALSE(val < 0 &&
+                              std::numeric_limits<T>::min() + multiple > floor)) {
+        *st = Status::Invalid("Rounding ", val, " down to multiple of ", multiple,
+                              " would overflow");
+        return val;
+      }
+      return val < 0 ? floor - multiple : floor;
+    }
+    return floor;
+  }
 };
 
 template <typename Type>
@@ -134,6 +191,18 @@ struct RoundImpl<Type, RoundMode::UP> {
       (*val) += pow10;
     }
   }
+
+  template <typename T = Type>
+  static enable_if_integer_value<T> Round(const T val, const T floor, const T multiple,
+                                          Status* st) {
+    if (ARROW_PREDICT_FALSE(val > 0 &&
+                            std::numeric_limits<T>::max() - multiple < floor)) {
+      *st = Status::Invalid("Rounding ", val, " up to multiple of ", multiple,
+                            " would overflow");
+      return val;
+    }
+    return val > 0 ? floor + multiple : floor;
+  }
 };
 
 template <typename Type>
@@ -148,6 +217,12 @@ struct RoundImpl<Type, RoundMode::TOWARDS_ZERO> {
                                                 const T& pow10, const int32_t scale) {
     (*val) -= remainder;
   }
+
+  template <typename T = Type>
+  static enable_if_integer_value<T> Round(const T val, const T floor, const T pow10,
+                                          Status* st) {
+    return floor;
+  }
 };
 
 template <typename Type>
@@ -167,6 +242,32 @@ struct RoundImpl<Type, RoundMode::TOWARDS_INFINITY> {
       (*val) += pow10;
     }
   }
+
+  template <typename T = Type>
+  static enable_if_integer_value<T> Round(const T val, const T floor, const T multiple,
+                                          Status* st) {
+    if constexpr (is_signed_integer_value<T>::value) {
+      if (ARROW_PREDICT_FALSE(val < 0 &&
+                              std::numeric_limits<T>::min() + multiple > floor)) {
+        *st = Status::Invalid("Rounding ", val, " down to multiple of ", multiple,
+                              " would overflow");
+        return val;
+      }
+    }
+
+    if (ARROW_PREDICT_FALSE(val > 0 &&
+                            std::numeric_limits<T>::max() - multiple < floor)) {
+      *st = Status::Invalid("Rounding ", val, " up to multiple of ", multiple,
+                            " would overflow");
+      return val;
+    }
+
+    if constexpr (is_signed_integer_value<T>::value) {
+      return val < 0 ? floor - multiple : floor + multiple;
+    }
+
+    return floor + multiple;
+  }
 };
 
 // NOTE: RoundImpl variants for the HALF_* rounding modes are only
@@ -185,6 +286,12 @@ struct RoundImpl<Type, RoundMode::HALF_DOWN> {
                                                 const T& pow10, const int32_t scale) {
     RoundImpl<T, RoundMode::DOWN>::Round(val, remainder, pow10, scale);
   }
+
+  template <typename T = Type>
+  static constexpr enable_if_integer_value<T> Round(const T val, const T floor,
+                                                    const T multiple, Status* st) {
+    return RoundImpl<T, RoundMode::DOWN>::Round(val, floor, multiple, st);
+  }
 };
 
 template <typename Type>
@@ -199,6 +306,12 @@ struct RoundImpl<Type, RoundMode::HALF_UP> {
                                                 const T& pow10, const int32_t scale) {
     RoundImpl<T, RoundMode::UP>::Round(val, remainder, pow10, scale);
   }
+
+  template <typename T = Type>
+  static constexpr enable_if_integer_value<T> Round(const T val, const T floor,
+                                                    const T multiple, Status* st) {
+    return RoundImpl<T, RoundMode::UP>::Round(val, floor, multiple, st);
+  }
 };
 
 template <typename Type>
@@ -213,6 +326,12 @@ struct RoundImpl<Type, RoundMode::HALF_TOWARDS_ZERO> {
                                                 const T& pow10, const int32_t scale) {
     RoundImpl<T, RoundMode::TOWARDS_ZERO>::Round(val, remainder, pow10, scale);
   }
+
+  template <typename T = Type>
+  static constexpr enable_if_integer_value<T> Round(const T val, const T floor,
+                                                    const T multiple, Status* st) {
+    return RoundImpl<T, RoundMode::TOWARDS_ZERO>::Round(val, floor, multiple, st);
+  }
 };
 
 template <typename Type>
@@ -224,8 +343,14 @@ struct RoundImpl<Type, RoundMode::HALF_TOWARDS_INFINITY> {
 
   template <typename T = Type>
   static enable_if_decimal_value<T, void> Round(T* val, const T& remainder,
-                                                const T& pow10, const int32_t scale) {
-    RoundImpl<T, RoundMode::TOWARDS_INFINITY>::Round(val, remainder, pow10, scale);
+                                                const T& multiple, const int32_t scale) {
+    RoundImpl<T, RoundMode::TOWARDS_INFINITY>::Round(val, remainder, multiple, scale);
+  }
+
+  template <typename T = Type>
+  static constexpr enable_if_integer_value<T> Round(const T val, const T floor,
+                                                    const T multiple, Status* st) {
+    return RoundImpl<T, RoundMode::TOWARDS_INFINITY>::Round(val, floor, multiple, st);
   }
 };
 
@@ -245,6 +370,15 @@ struct RoundImpl<Type, RoundMode::HALF_TO_EVEN> {
     }
     *val = scaled.IncreaseScaleBy(scale);
   }
+
+  template <typename T = Type>
+  static constexpr enable_if_integer_value<T> Round(const T val, const T floor,
+                                                    const T multiple, Status* st) {
+    if ((floor / multiple) % 2 == 0) {
+      return floor;
+    }
+    return RoundImpl<T, RoundMode::TOWARDS_INFINITY>::Round(val, floor, multiple, st);
+  }
 };
 
 template <typename Type>
@@ -263,23 +397,37 @@ struct RoundImpl<Type, RoundMode::HALF_TO_ODD> {
     }
     *val = scaled.IncreaseScaleBy(scale);
   }
+
+  template <typename T = Type>
+  static constexpr enable_if_integer_value<T> Round(const T val, const T floor,
+                                                    const T multiple, Status* st) {
+    if ((floor / multiple) % 2 == 1) {
+      return floor;
+    }
+    return RoundImpl<T, RoundMode::TOWARDS_INFINITY>::Round(val, floor, multiple, st);
+  }
 };
 
+// End round implementations for single scalar
+// ----------------------------------------------------------------------
+// Begin round options wrappers
+
 // Specializations of kernel state for round kernels
-template <typename OptionsType>
+// CType is the physical type used to store pow10
+template <typename OptionsType, typename CType>
 struct RoundOptionsWrapper;
 
-template <>
-struct RoundOptionsWrapper<RoundOptions> : public OptionsWrapper<RoundOptions> {
+template <typename CType>
+struct RoundOptionsWrapper<RoundOptions, CType> : public OptionsWrapper<RoundOptions> {
   using OptionsType = RoundOptions;
-  double pow10;
+  CType pow10;
 
   explicit RoundOptionsWrapper(OptionsType options) : OptionsWrapper(std::move(options)) {
     // Only positive exponents for powers of 10 are used because combining
     // multiply and division operations produced more stable rounding than
     // using multiply-only.  Refer to NumPy's round implementation:
     // https://github.com/numpy/numpy/blob/7b2f20b406d27364c812f7a81a9c901afbd3600c/numpy/core/src/multiarray/calculation.c#L589
-    pow10 = RoundUtil::Pow10(std::abs(options.ndigits));
+    pow10 = RoundUtil::Pow10<CType>(std::abs(options.ndigits));
   }
 
   static Result<std::unique_ptr<KernelState>> Init(KernelContext* ctx,
@@ -292,8 +440,8 @@ struct RoundOptionsWrapper<RoundOptions> : public OptionsWrapper<RoundOptions> {
   }
 };
 
-template <>
-struct RoundOptionsWrapper<RoundBinaryOptions>
+template <typename CType>
+struct RoundOptionsWrapper<RoundBinaryOptions, CType>
     : public OptionsWrapper<RoundBinaryOptions> {
   using OptionsType = RoundBinaryOptions;
 
@@ -310,8 +458,8 @@ struct RoundOptionsWrapper<RoundBinaryOptions>
   }
 };
 
-template <>
-struct RoundOptionsWrapper<RoundToMultipleOptions>
+template <typename CType>
+struct RoundOptionsWrapper<RoundToMultipleOptions, CType>
     : public OptionsWrapper<RoundToMultipleOptions> {
   using OptionsType = RoundToMultipleOptions;
   using OptionsWrapper::OptionsWrapper;
@@ -333,14 +481,8 @@ struct RoundOptionsWrapper<RoundToMultipleOptions>
       return Status::Invalid("Rounding multiple must be positive");
     }
 
-    // Ensure the rounding multiple option matches the kernel's output type.
-    // The output type is not available here so we use the following rule:
-    // If `multiple` is neither a floating-point nor a decimal type, then
-    // cast to float64, else cast to the kernel's input type.
-    std::shared_ptr<DataType> to_type =
-        (!is_floating(multiple->type->id()) && !is_decimal(multiple->type->id()))
-            ? float64()
-            : args.inputs[0].GetSharedPtr();
+    // Ensure the rounding multiple option matches the kernel's input type.
+    std::shared_ptr<DataType> to_type = args.inputs[0].GetSharedPtr();
     if (!multiple->type->Equals(to_type)) {
       ARROW_ASSIGN_OR_RAISE(
           auto casted_multiple,
@@ -355,11 +497,231 @@ struct RoundOptionsWrapper<RoundToMultipleOptions>
   }
 };
 
+template <typename ArrowType, typename Enable = void>
+struct RoundOptionsTrait;
+
+template <typename ArrowType>
+struct RoundOptionsTrait<ArrowType, enable_if_floating_point<ArrowType>> {
+  using CType = double;
+};
+
+template <typename ArrowType>
+struct RoundOptionsTrait<ArrowType, enable_if_decimal<ArrowType>> {
+  using CType = double;
+};
+
+template <typename ArrowType>
+struct RoundOptionsTrait<ArrowType, enable_if_integer<ArrowType>> {
+  using CType = typename ArrowType::c_type;
+};
+
+// End round options wrappers
+// ----------------------------------------------------------------------
+// Begin round op implementations
+
+template <typename ArrowType, RoundMode kRoundMode, typename Enable = void>
+struct RoundToMultiple {
+  using CType = typename TypeTraits<ArrowType>::CType;
+  using State = RoundOptionsWrapper<RoundToMultipleOptions,
+                                    typename RoundOptionsTrait<ArrowType>::CType>;
+
+  CType multiple;
+
+  explicit RoundToMultiple(const State& state, const DataType& out_ty)
+      : multiple(UnboxScalar<ArrowType>::Unbox(*state.options.multiple)) {
+    const auto& options = state.options;
+    DCHECK(options.multiple);
+    DCHECK(options.multiple->is_valid);
+    DCHECK(is_floating(options.multiple->type->id()));
+  }
+
+  template <typename T = ArrowType, typename CType = typename TypeTraits<T>::CType>
+  enable_if_floating_value<CType> Call(KernelContext* ctx, CType arg, Status* st) const {
+    // Do not process Inf or NaN because they will trigger the overflow error at end of
+    // function.
+    if (!std::isfinite(arg)) {
+      return arg;
+    }
+    auto round_val = arg / multiple;
+    auto frac = round_val - std::floor(round_val);
+    if (frac != T(0)) {
+      // Use std::round() if in tie-breaking mode and scaled value is not 0.5.
+      if ((kRoundMode >= RoundMode::HALF_DOWN) && (frac != T(0.5))) {
+        round_val = std::round(round_val);
+      } else {
+        round_val = RoundImpl<CType, kRoundMode>::Round(round_val);
+      }
+      round_val *= multiple;
+      if (!std::isfinite(round_val)) {
+        *st = Status::Invalid("overflow occurred during rounding");
+        return arg;
+      }
+    } else {
+      // If scaled value is an integer, then no rounding is needed.
+      round_val = arg;
+    }
+    return round_val;
+  }
+};
+
+template <typename ArrowType, RoundMode kRoundMode>
+struct RoundToMultiple<ArrowType, kRoundMode, enable_if_decimal<ArrowType>> {
+  using CType = typename TypeTraits<ArrowType>::CType;
+  using State = RoundOptionsWrapper<RoundToMultipleOptions, double>;
+  const ArrowType& ty;
+  CType multiple, half_multiple, neg_half_multiple;
+  bool has_halfway_point;
+
+  explicit RoundToMultiple(const State& state, const DataType& out_ty)
+      : ty(checked_cast<const ArrowType&>(out_ty)),
+        multiple(UnboxScalar<ArrowType>::Unbox(*state.options.multiple)),
+        half_multiple(multiple / 2),
+        neg_half_multiple(-half_multiple),
+        has_halfway_point(multiple.low_bits() % 2 == 0) {
+    const auto& options = state.options;
+    DCHECK(options.multiple);
+    DCHECK(options.multiple->is_valid);
+    DCHECK(options.multiple->type->Equals(out_ty));
+  }
+
+  template <typename T = ArrowType, typename CType = typename TypeTraits<T>::CType>
+  enable_if_decimal_value<CType> Call(KernelContext* ctx, CType arg, Status* st) const {
+    std::pair<CType, CType> pair;
+    *st = arg.Divide(multiple).Value(&pair);
+    if (!st->ok()) return arg;
+    const auto& remainder = pair.second;
+    if (remainder == 0) return arg;
+    if (kRoundMode >= RoundMode::HALF_DOWN) {
+      if (has_halfway_point &&
+          (remainder == half_multiple || remainder == neg_half_multiple)) {
+        // On the halfway point, use tiebreaker
+        // Manually implement rounding since we're not actually rounding a
+        // decimal value, but rather manipulating the multiple
+        switch (kRoundMode) {
+          case RoundMode::HALF_DOWN:
+            if (remainder.Sign() < 0) pair.first -= 1;
+            break;
+          case RoundMode::HALF_UP:
+            if (remainder.Sign() >= 0) pair.first += 1;
+            break;
+          case RoundMode::HALF_TOWARDS_ZERO:
+            // Do nothing
+            break;
+          case RoundMode::HALF_TOWARDS_INFINITY:
+            pair.first += remainder.Sign() >= 0 ? 1 : -1;
+            break;
+          case RoundMode::HALF_TO_EVEN:
+            if (pair.first.low_bits() % 2 != 0) {
+              pair.first += remainder.Sign() >= 0 ? 1 : -1;
+            }
+            break;
+          case RoundMode::HALF_TO_ODD:
+            if (pair.first.low_bits() % 2 == 0) {
+              pair.first += remainder.Sign() >= 0 ? 1 : -1;
+            }
+            break;
+          default:
+            DCHECK(false);
+        }
+      } else if (remainder.Sign() >= 0) {
+        // Positive, round up/down
+        if (remainder > half_multiple) {
+          pair.first += 1;
+        }
+      } else {
+        // Negative, round up/down
+        if (remainder < neg_half_multiple) {
+          pair.first -= 1;
+        }
+      }
+    } else {
+      // Manually implement rounding since we're not actually rounding a
+      // decimal value, but rather manipulating the multiple
+      switch (kRoundMode) {
+        case RoundMode::DOWN:
+          if (remainder.Sign() < 0) pair.first -= 1;
+          break;
+        case RoundMode::UP:
+          if (remainder.Sign() >= 0) pair.first += 1;
+          break;
+        case RoundMode::TOWARDS_ZERO:
+          // Do nothing
+          break;
+        case RoundMode::TOWARDS_INFINITY:
+          pair.first += remainder.Sign() >= 0 ? 1 : -1;
+          break;
+        default:
+          DCHECK(false);
+      }
+    }
+    CType round_val = pair.first * multiple;
+    if (!round_val.FitsInPrecision(ty.precision())) {
+      *st = Status::Invalid("Rounded value ", round_val.ToString(ty.scale()),
+                            " does not fit in precision of ", ty);
+      return 0;
+    }
+    return round_val;
+  }
+};
+
+template <typename ArrowType, RoundMode kRoundMode>
+struct RoundToMultiple<ArrowType, kRoundMode, enable_if_integer<ArrowType>> {
+  using CType = typename TypeTraits<ArrowType>::CType;
+  using State = RoundOptionsWrapper<RoundToMultipleOptions, CType>;
+  CType multiple;
+
+  explicit RoundToMultiple(const State& state, const DataType& out_ty)
+      : multiple(UnboxScalar<ArrowType>::Unbox(*state.options.multiple)) {
+    const auto& options = state.options;
+    DCHECK(options.multiple);
+    DCHECK(options.multiple->is_valid);
+    DCHECK(is_integer(options.multiple->type->id()));
+  }
+
+  explicit RoundToMultiple(const CType multiple, const DataType& out_ty)
+      : multiple(multiple) {}
+
+  template <typename T = ArrowType, typename CType = typename TypeTraits<T>::CType>
+  enable_if_integer_value<CType> Call(KernelContext* ctx, CType arg, Status* st) const {
+    CType floor = arg / multiple * multiple;
+    CType remainder = arg > floor ? arg - floor : floor - arg;
+
+    if (remainder == 0) {
+      return arg;
+    }
+
+    if (kRoundMode >= RoundMode::HALF_DOWN && remainder * 2 != multiple) {
+      // not half way, round to nearest multiple of multiple like std::round
+      if (remainder * 2 > multiple) {
+        if (arg >= 0) {
+          if (ARROW_PREDICT_FALSE(std::numeric_limits<CType>::max() - multiple < floor)) {
+            *st = Status::Invalid("Rounding ", arg, " up to multiples of ", multiple,
+                                  " would overflow");
+            return arg;
+          }
+          return floor + multiple;
+        } else {
+          if (ARROW_PREDICT_FALSE(std::numeric_limits<CType>::min() + multiple > floor)) {
+            *st = Status::Invalid("Rounding ", arg, " down to multiples of ", multiple,
+                                  " would overflow");
+            return arg;
+          }
+          return floor - multiple;
+        }
+      } else {
+        return floor;
+      }
+    } else {
+      return RoundImpl<CType, kRoundMode>::Round(arg, floor, multiple, st);
+    }
+  }
+};
+
 template <typename ArrowType, RoundMode RndMode, typename Enable = void>
 struct Round {
   using CType = typename TypeTraits<ArrowType>::CType;
-  using State = RoundOptionsWrapper<RoundOptions>;
-
+  using State =
+      RoundOptionsWrapper<RoundOptions, typename RoundOptionsTrait<ArrowType>::CType>;
   CType pow10;
   int64_t ndigits;
 
@@ -400,8 +762,7 @@ struct Round {
 template <typename ArrowType, RoundMode kRoundMode>
 struct Round<ArrowType, kRoundMode, enable_if_decimal<ArrowType>> {
   using CType = typename TypeTraits<ArrowType>::CType;
-  using State = RoundOptionsWrapper<RoundOptions>;
-
+  using State = RoundOptionsWrapper<RoundOptions, double>;
   const ArrowType& ty;
   int64_t ndigits;
   int32_t pow;
@@ -470,10 +831,37 @@ struct Round<ArrowType, kRoundMode, enable_if_decimal<ArrowType>> {
   }
 };
 
+template <typename ArrowType, RoundMode kRoundMode>
+struct Round<ArrowType, kRoundMode, enable_if_integer<ArrowType>> {
+  using CType = typename TypeTraits<ArrowType>::CType;
+  using State = RoundOptionsWrapper<RoundOptions, CType>;
+  CType pow10;
+  int64_t ndigits;
+  const DataType& out_ty;
+
+  explicit Round(const State& state, const DataType& out_ty)
+      : pow10(static_cast<CType>(state.pow10)),
+        ndigits(state.options.ndigits),
+        out_ty(out_ty) {}
+
+  template <typename T = ArrowType, typename CType = typename TypeTraits<T>::CType>
+  enable_if_integer_value<CType> Call(KernelContext* ctx, CType arg, Status* st) const {
+    // no-op if ndigits is non-negative
+    if (ndigits >= 0) {
+      return arg;
+    }
+
+    // If ndigits is negative, then round to the nearest multiple of 10^ndigits.
+    RoundToMultiple<ArrowType, kRoundMode> round_to_multiple(pow10, out_ty);
+    return round_to_multiple.Call(ctx, arg, st);
+  }
+};
+
 template <typename ArrowType, RoundMode RndMode, typename Enable = void>
 struct RoundBinary {
   using CType = typename TypeTraits<ArrowType>::CType;
-  using State = RoundOptionsWrapper<RoundBinaryOptions>;
+  using State = RoundOptionsWrapper<RoundBinaryOptions,
+                                    typename RoundOptionsTrait<ArrowType>::CType>;
 
   explicit RoundBinary(const State& state, const DataType& out_ty) {}
 
@@ -491,7 +879,7 @@ struct RoundBinary {
     // multiply and division operations produced more stable rounding than
     // using multiply-only.  Refer to NumPy's round implementation:
     // https://github.com/numpy/numpy/blob/7b2f20b406d27364c812f7a81a9c901afbd3600c/numpy/core/src/multiarray/calculation.c#L589
-    double pow10 = RoundUtil::Pow10(std::abs(arg1));
+    double pow10 = RoundUtil::Pow10<double>(std::abs(arg1));
 
     auto round_val = arg1 >= 0 ? (arg0 * pow10) : (arg0 / pow10);
     auto frac = round_val - std::floor(round_val);
@@ -520,8 +908,7 @@ struct RoundBinary {
 template <typename ArrowType, RoundMode kRoundMode>
 struct RoundBinary<ArrowType, kRoundMode, enable_if_decimal<ArrowType>> {
   using CType = typename TypeTraits<ArrowType>::CType;
-  using State = RoundOptionsWrapper<RoundBinaryOptions>;
-
+  using State = RoundOptionsWrapper<RoundBinaryOptions, double>;
   const ArrowType& ty;
   int32_t pow;
   // pow10 is "1" for the given decimal scale. Similarly half_pow10 is "0.5".
@@ -591,156 +978,27 @@ struct RoundBinary<ArrowType, kRoundMode, enable_if_decimal<ArrowType>> {
   }
 };
 
-template <typename DecimalType, RoundMode kMode, int32_t kDigits>
-Status FixedRoundDecimalExec(KernelContext* ctx, const ExecSpan& batch, ExecResult* out) {
-  using Op = Round<DecimalType, kMode>;
-  return ScalarUnaryNotNullStateful<DecimalType, DecimalType, Op>(
-             Op(kDigits, *out->type()))
-      .Exec(ctx, batch, out);
-}
-
-template <typename ArrowType, RoundMode kRoundMode, typename Enable = void>
-struct RoundToMultiple {
-  using CType = typename TypeTraits<ArrowType>::CType;
-  using State = RoundOptionsWrapper<RoundToMultipleOptions>;
-
-  CType multiple;
-
-  explicit RoundToMultiple(const State& state, const DataType& out_ty)
-      : multiple(UnboxScalar<ArrowType>::Unbox(*state.options.multiple)) {
-    const auto& options = state.options;
-    DCHECK(options.multiple);
-    DCHECK(options.multiple->is_valid);
-    DCHECK(is_floating(options.multiple->type->id()));
-  }
-
-  template <typename T = ArrowType, typename CType = typename TypeTraits<T>::CType>
-  enable_if_floating_value<CType> Call(KernelContext* ctx, CType arg, Status* st) const {
-    // Do not process Inf or NaN because they will trigger the overflow error at end of
-    // function.
-    if (!std::isfinite(arg)) {
-      return arg;
-    }
-    auto round_val = arg / multiple;
-    auto frac = round_val - std::floor(round_val);
-    if (frac != T(0)) {
-      // Use std::round() if in tie-breaking mode and scaled value is not 0.5.
-      if ((kRoundMode >= RoundMode::HALF_DOWN) && (frac != T(0.5))) {
-        round_val = std::round(round_val);
-      } else {
-        round_val = RoundImpl<CType, kRoundMode>::Round(round_val);
-      }
-      round_val *= multiple;
-      if (!std::isfinite(round_val)) {
-        *st = Status::Invalid("overflow occurred during rounding");
-        return arg;
-      }
-    } else {
-      // If scaled value is an integer, then no rounding is needed.
-      round_val = arg;
-    }
-    return round_val;
-  }
-};
-
 template <typename ArrowType, RoundMode kRoundMode>
-struct RoundToMultiple<ArrowType, kRoundMode, enable_if_decimal<ArrowType>> {
+struct RoundBinary<ArrowType, kRoundMode, enable_if_integer<ArrowType>> {
   using CType = typename TypeTraits<ArrowType>::CType;
-  using State = RoundOptionsWrapper<RoundToMultipleOptions>;
+  using State = RoundOptionsWrapper<RoundBinaryOptions, CType>;
 
-  const ArrowType& ty;
-  CType multiple, half_multiple, neg_half_multiple;
-  bool has_halfway_point;
-
-  explicit RoundToMultiple(const State& state, const DataType& out_ty)
-      : ty(checked_cast<const ArrowType&>(out_ty)),
-        multiple(UnboxScalar<ArrowType>::Unbox(*state.options.multiple)),
-        half_multiple(multiple / 2),
-        neg_half_multiple(-half_multiple),
-        has_halfway_point(multiple.low_bits() % 2 == 0) {
-    const auto& options = state.options;
-    DCHECK(options.multiple);
-    DCHECK(options.multiple->is_valid);
-    DCHECK(options.multiple->type->Equals(out_ty));
-  }
+  const DataType& out_ty;
+  explicit RoundBinary(const State& state, const DataType& out_ty) : out_ty(out_ty) {}
 
-  template <typename T = ArrowType, typename CType = typename TypeTraits<T>::CType>
-  enable_if_decimal_value<CType> Call(KernelContext* ctx, CType arg, Status* st) const {
-    std::pair<CType, CType> pair;
-    *st = arg.Divide(multiple).Value(&pair);
-    if (!st->ok()) return arg;
-    const auto& remainder = pair.second;
-    if (remainder == 0) return arg;
-    if (kRoundMode >= RoundMode::HALF_DOWN) {
-      if (has_halfway_point &&
-          (remainder == half_multiple || remainder == neg_half_multiple)) {
-        // On the halfway point, use tiebreaker
-        // Manually implement rounding since we're not actually rounding a
-        // decimal value, but rather manipulating the multiple
-        switch (kRoundMode) {
-          case RoundMode::HALF_DOWN:
-            if (remainder.Sign() < 0) pair.first -= 1;
-            break;
-          case RoundMode::HALF_UP:
-            if (remainder.Sign() >= 0) pair.first += 1;
-            break;
-          case RoundMode::HALF_TOWARDS_ZERO:
-            // Do nothing
-            break;
-          case RoundMode::HALF_TOWARDS_INFINITY:
-            pair.first += remainder.Sign() >= 0 ? 1 : -1;
-            break;
-          case RoundMode::HALF_TO_EVEN:
-            if (pair.first.low_bits() % 2 != 0) {
-              pair.first += remainder.Sign() >= 0 ? 1 : -1;
-            }
-            break;
-          case RoundMode::HALF_TO_ODD:
-            if (pair.first.low_bits() % 2 == 0) {
-              pair.first += remainder.Sign() >= 0 ? 1 : -1;
-            }
-            break;
-          default:
-            DCHECK(false);
-        }
-      } else if (remainder.Sign() >= 0) {
-        // Positive, round up/down
-        if (remainder > half_multiple) {
-          pair.first += 1;
-        }
-      } else {
-        // Negative, round up/down
-        if (remainder < neg_half_multiple) {
-          pair.first -= 1;
-        }
-      }
-    } else {
-      // Manually implement rounding since we're not actually rounding a
-      // decimal value, but rather manipulating the multiple
-      switch (kRoundMode) {
-        case RoundMode::DOWN:
-          if (remainder.Sign() < 0) pair.first -= 1;
-          break;
-        case RoundMode::UP:
-          if (remainder.Sign() >= 0) pair.first += 1;
-          break;
-        case RoundMode::TOWARDS_ZERO:
-          // Do nothing
-          break;
-        case RoundMode::TOWARDS_INFINITY:
-          pair.first += remainder.Sign() >= 0 ? 1 : -1;
-          break;
-        default:
-          DCHECK(false);
-      }
-    }
-    CType round_val = pair.first * multiple;
-    if (!round_val.FitsInPrecision(ty.precision())) {
-      *st = Status::Invalid("Rounded value ", round_val.ToString(ty.scale()),
-                            " does not fit in precision of ", ty);
-      return 0;
+  template <typename T = ArrowType, typename CType0 = typename TypeTraits<T>::CType0,
+            typename CType1 = typename TypeTraits<T>::CType1>
+  enable_if_integer_value<CType> Call(KernelContext* ctx, CType0 arg0, CType1 arg1,
+                                      Status* st) const {
+    // ndigits >= 0 is a no-op
+    if (arg1 >= 0) {
+      return arg0;
     }
-    return round_val;
+
+    // If ndigits is negative, then round to the nearest multiple of 10^ndigits.
+    CType pow10 = RoundUtil::Pow10<CType>(std::abs(arg1));
+    RoundToMultiple<ArrowType, kRoundMode> round_to_multiple(pow10, out_ty);
+    return round_to_multiple.Call(ctx, arg0, st);
   }
 };
 
@@ -748,7 +1006,7 @@ struct Floor {
   template <typename T, typename Arg>
   static constexpr enable_if_floating_value<Arg, T> Call(KernelContext*, Arg arg,
                                                          Status*) {
-    static_assert(std::is_same<T, Arg>::value, "");
+    static_assert(std::is_same<T, Arg>::value);
     return RoundImpl<T, RoundMode::DOWN>::Round(arg);
   }
 };
@@ -757,7 +1015,7 @@ struct Ceil {
   template <typename T, typename Arg>
   static constexpr enable_if_floating_value<Arg, T> Call(KernelContext*, Arg arg,
                                                          Status*) {
-    static_assert(std::is_same<T, Arg>::value, "");
+    static_assert(std::is_same<T, Arg>::value);
     return RoundImpl<T, RoundMode::UP>::Round(arg);
   }
 };
@@ -766,11 +1024,15 @@ struct Trunc {
   template <typename T, typename Arg>
   static constexpr enable_if_floating_value<Arg, T> Call(KernelContext*, Arg arg,
                                                          Status*) {
-    static_assert(std::is_same<T, Arg>::value, "");
+    static_assert(std::is_same<T, Arg>::value);
     return RoundImpl<T, RoundMode::TOWARDS_ZERO>::Round(arg);
   }
 };
 
+// End round op implementations
+// ----------------------------------------------------------------------
+// Begin round functions
+
 struct RoundFunction : ScalarFunction {
   using ScalarFunction::ScalarFunction;
 
@@ -782,6 +1044,10 @@ struct RoundFunction : ScalarFunction {
 
     EnsureDictionaryDecoded(types);
 
+    // for binary round functions, the second scalar must be int32
+    if (types->size() == 2 && (*types)[1].id() != Type::INT32) {
+      (*types)[1] = int32();
+    }
     if (auto kernel = DispatchExactImpl(this, *types)) return kernel;
     return arrow::compute::detail::NoMatchingKernel(this, *types);
   }
@@ -862,6 +1128,10 @@ struct RoundFloatingPointFunction : public RoundFunction {
   }
 };
 
+// End round functions
+// ----------------------------------------------------------------------
+// Begin round kernels
+
 #define ROUND_CASE(MODE)                                                       \
   case RoundMode::MODE: {                                                      \
     using Op = OpImpl<Type, RoundMode::MODE>;                                  \
@@ -874,7 +1144,8 @@ template <typename Type, typename OptionsType,
           template <typename, RoundMode, typename...> class OpImpl>
 struct RoundKernel {
   static Status Exec(KernelContext* ctx, const ExecSpan& batch, ExecResult* out) {
-    using State = RoundOptionsWrapper<OptionsType>;
+    using State =
+        RoundOptionsWrapper<OptionsType, typename RoundOptionsTrait<Type>::CType>;
     const auto& state = static_cast<const State&>(*ctx->state());
     switch (state.options.round_mode) {
       ROUND_CASE(DOWN)
@@ -909,7 +1180,8 @@ template <typename Type, typename OptionsType,
           template <typename, RoundMode, typename...> class OpImpl>
 struct RoundBinaryKernel {
   static Status Exec(KernelContext* ctx, const ExecSpan& batch, ExecResult* out) {
-    using State = RoundOptionsWrapper<OptionsType>;
+    using State =
+        RoundOptionsWrapper<OptionsType, typename RoundOptionsTrait<Type>::CType>;
     const auto& state = static_cast<const State&>(*ctx->state());
     switch (state.options.round_mode) {
       ROUND_BINARY_CASE(DOWN)
@@ -931,38 +1203,59 @@ struct RoundBinaryKernel {
 };
 #undef ROUND_BINARY_CASE
 
+template <typename DecimalType, RoundMode kMode, int32_t kDigits>
+Status FixedRoundDecimalExec(KernelContext* ctx, const ExecSpan& batch, ExecResult* out) {
+  using Op = Round<DecimalType, kMode>;
+  return ScalarUnaryNotNullStateful<DecimalType, DecimalType, Op>(
+             Op(kDigits, *out->type()))
+      .Exec(ctx, batch, out);
+}
+
+// End round kernels
+// ----------------------------------------------------------------------
+// Begin round kernel generation and function registration
+
+template <
+    template <typename, RoundMode, typename...> class Op,
+    template <typename, typename, template <typename, RoundMode, typename...> typename>
+    class Kernel,
+    typename OptionsType>
+struct RoundKernelGenerator {
+  template <typename ArrowType>
+  Status Visit(const ArrowType& type, ArrayKernelExec* exec, KernelInit* init) {
+    if constexpr (is_integer_type<ArrowType>::value ||
+                  (is_floating_type<ArrowType>::value &&
+                   !is_half_float_type<ArrowType>::value) ||
+                  is_decimal_type<ArrowType>::value) {
+      *exec = Kernel<ArrowType, OptionsType, Op>::Exec;
+      *init = RoundOptionsWrapper<OptionsType,
+                                  typename RoundOptionsTrait<ArrowType>::CType>::Init;
+    } else {
+      DCHECK(false);
+      return Status::NotImplemented("Round does not support ", type.ToString());
+    }
+    return Status::OK();
+  }
+};
+
 // For unary rounding functions that control kernel dispatch based on RoundMode, only on
 // non-null output.
 template <template <typename, RoundMode, typename...> class Op, typename OptionsType>
 std::shared_ptr<ScalarFunction> MakeUnaryRoundFunction(std::string name,
                                                        FunctionDoc doc) {
-  using State = RoundOptionsWrapper<OptionsType>;
   static const OptionsType kDefaultOptions = OptionsType::Defaults();
-  auto func = std::make_shared<RoundIntegerToFloatingPointFunction>(
-      name, Arity::Unary(), std::move(doc), &kDefaultOptions);
-  for (const auto& ty : {float32(), float64(), decimal128(1, 0), decimal256(1, 0)}) {
-    auto type_id = ty->id();
-    ArrayKernelExec exec = nullptr;
-    switch (type_id) {
-      case Type::FLOAT:
-        exec = RoundKernel<FloatType, OptionsType, Op>::Exec;
-        break;
-      case Type::DOUBLE:
-        exec = RoundKernel<DoubleType, OptionsType, Op>::Exec;
-        break;
-      case Type::DECIMAL128:
-        exec = RoundKernel<Decimal128Type, OptionsType, Op>::Exec;
-        break;
-      case Type::DECIMAL256:
-        exec = RoundKernel<Decimal256Type, OptionsType, Op>::Exec;
-        break;
-      default:
-        DCHECK(false);
-        break;
+  auto func = std::make_shared<RoundFunction>(name, Arity::Unary(), std::move(doc),
+                                              &kDefaultOptions);
+  RoundKernelGenerator<Op, RoundKernel, OptionsType> kernel_generator;
+  for (const auto& tys : {NumericTypes(), {decimal128(1, 0), decimal256(1, 0)}}) {
+    for (const auto& ty : tys) {
+      ArrayKernelExec exec;
+      KernelInit init;
+      DCHECK_OK(VisitTypeInline(*ty, &kernel_generator, &exec, &init));
+      DCHECK_OK(func->AddKernel(
+          {InputType(ty->id())},
+          is_decimal(ty->id()) ? OutputType(FirstType) : OutputType(ty), exec, init));
     }
-    DCHECK_OK(func->AddKernel(
-        {InputType(type_id)},
-        is_decimal(type_id) ? OutputType(FirstType) : OutputType(ty), exec, State::Init));
   }
   AddNullExec(func.get());
   return func;
@@ -971,33 +1264,19 @@ std::shared_ptr<ScalarFunction> MakeUnaryRoundFunction(std::string name,
 template <template <typename, RoundMode, typename...> class Op, typename OptionsType>
 std::shared_ptr<ScalarFunction> MakeBinaryRoundFunction(const std::string& name,
                                                         FunctionDoc doc) {
-  using State = RoundOptionsWrapper<OptionsType>;
   static const OptionsType kDefaultOptions = OptionsType::Defaults();
-  auto func = std::make_shared<RoundIntegerToFloatingPointFunction>(
-      name, Arity::Binary(), std::move(doc), &kDefaultOptions);
-  for (const auto& ty : {float32(), float64(), decimal128(1, 0), decimal256(1, 0)}) {
-    auto type_id = ty->id();
-    ArrayKernelExec exec = nullptr;
-    switch (type_id) {
-      case Type::FLOAT:
-        exec = RoundBinaryKernel<FloatType, OptionsType, Op>::Exec;
-        break;
-      case Type::DOUBLE:
-        exec = RoundBinaryKernel<DoubleType, OptionsType, Op>::Exec;
-        break;
-      case Type::DECIMAL128:
-        exec = RoundBinaryKernel<Decimal128Type, OptionsType, Op>::Exec;
-        break;
-      case Type::DECIMAL256:
-        exec = RoundBinaryKernel<Decimal256Type, OptionsType, Op>::Exec;
-        break;
-      default:
-        DCHECK(false);
-        break;
+  auto func = std::make_shared<RoundFunction>(name, Arity::Binary(), std::move(doc),
+                                              &kDefaultOptions);
+  RoundKernelGenerator<Op, RoundBinaryKernel, OptionsType> kernel_generator;
+  for (const auto& tys : {NumericTypes(), {decimal128(1, 0), decimal256(1, 0)}}) {
+    for (const auto& ty : tys) {
+      ArrayKernelExec exec;
+      KernelInit init;
+      DCHECK_OK(VisitTypeInline(*ty, &kernel_generator, &exec, &init));
+      DCHECK_OK(func->AddKernel(
+          {InputType(ty->id()), Type::INT32},
+          is_decimal(ty->id()) ? OutputType(FirstType) : OutputType(ty), exec, init));
     }
-    DCHECK_OK(func->AddKernel(
-        {ty, Type::INT32}, is_decimal(type_id) ? OutputType(FirstType) : OutputType(ty),
-        exec, State::Init));
   }
   AddNullExec(func.get());
   return func;
@@ -1101,6 +1380,7 @@ void RegisterScalarRoundArithmetic(FunctionRegistry* registry) {
   DCHECK_OK(registry->AddFunction(std::move(round_to_multiple)));
 }
 
-}  // namespace internal
-}  // namespace compute
+// End round kernel generation and function registration
+// ----------------------------------------------------------------------
+}  // namespace compute::internal
 }  // namespace arrow
diff --git a/cpp/src/arrow/compute/kernels/scalar_round_arithmetic_test.cc b/cpp/src/arrow/compute/kernels/scalar_round_arithmetic_test.cc
index 6f11421610a34..c0a36dd9deb43 100644
--- a/cpp/src/arrow/compute/kernels/scalar_round_arithmetic_test.cc
+++ b/cpp/src/arrow/compute/kernels/scalar_round_arithmetic_test.cc
@@ -34,8 +34,7 @@
 
 #include "arrow/testing/gtest_util.h"
 
-namespace arrow {
-namespace compute {
+namespace arrow::compute {
 
 namespace {
 
@@ -481,13 +480,31 @@ TYPED_TEST_SUITE(TestUnaryRoundArithmeticFloating, FloatingTypes);
 
 TEST(TestUnaryRound, DispatchBestRound) {
   // Integer -> Float64
-  for (std::string name : {"floor", "ceil", "trunc", "round", "round_to_multiple"}) {
+  for (std::string name : {"floor", "ceil", "trunc"}) {
     for (const auto& ty :
          {int8(), int16(), int32(), int64(), uint8(), uint16(), uint32(), uint64()}) {
       CheckDispatchBest(name, {ty}, {float64()});
       CheckDispatchBest(name, {dictionary(int8(), ty)}, {float64()});
     }
   }
+  // Integer -> Integer
+  for (std::string name : {"round", "round_to_multiple"}) {
+    for (const auto& ty :
+         {int8(), int16(), int32(), int64(), uint8(), uint16(), uint32(), uint64()}) {
+      CheckDispatchBest(name, {ty}, {ty});
+      CheckDispatchBest(name, {dictionary(int8(), ty)}, {ty});
+    }
+  }
+
+  // Any -> Int32
+  for (std::string name : {"round_binary"}) {
+    for (const auto& ty1 : NumericTypes()) {
+      for (const auto& ty2 : NumericTypes()) {
+        CheckDispatchBest(name, {ty1, ty2}, {ty1, int32()});
+        CheckDispatchBest(name, {dictionary(int8(), ty1), ty2}, {ty1, int32()});
+      }
+    }
+  }
 }
 
 class TestUnaryRoundArithmeticDecimal : public TestRoundArithmeticDecimal {};
@@ -1014,13 +1031,13 @@ TYPED_TEST(TestUnaryRoundSigned, Round) {
   this->SetRoundNdigits(0);
   for (const auto& round_mode : kRoundModes) {
     this->SetRoundMode(round_mode);
-    this->AssertUnaryOp(Round, values, ArrayFromJSON(float64(), values));
+    this->AssertUnaryOp(Round, values, ArrayFromJSON(this->type_singleton(), values));
   }
 
   // Test different round N-digits for nearest rounding mode
   std::vector<std::pair<int64_t, std::string>> ndigits_and_expected{{
-      {-2, "[0.0, 0.0, -0.0, -100, 100]"},
-      {-1, "[0.0, 0.0, -10, -50, 120]"},
+      {-2, "[0, 0, -0, -100, 100]"},
+      {-1, "[0, 0, -10, -50, 120]"},
       {0, values},
       {1, values},
       {2, values},
@@ -1028,7 +1045,49 @@ TYPED_TEST(TestUnaryRoundSigned, Round) {
   this->SetRoundMode(RoundMode::HALF_TOWARDS_INFINITY);
   for (const auto& pair : ndigits_and_expected) {
     this->SetRoundNdigits(pair.first);
-    this->AssertUnaryOp(Round, values, ArrayFromJSON(float64(), pair.second));
+    this->AssertUnaryOp(Round, values,
+                        ArrayFromJSON(this->type_singleton(), pair.second));
+  }
+
+  // Test different rounding mode
+  // skip int8 because of its small range
+  if constexpr (!std::is_same_v<TypeParam, Int8Type>) {
+    std::string values("[0, 1, -13, -50, 115, -176, 200, 250]");
+    this->SetRoundNdigits(-2);
+    std::vector<std::pair<RoundMode, std::string>> round_modes_and_expected{{
+        {RoundMode::DOWN, "[0, 0, -100, -100, 100, -200, 200, 200]"},
+        {RoundMode::UP, "[0, 100, -0, -0, 200, -100, 200, 300]"},
+        {RoundMode::TOWARDS_ZERO, "[0, 0, -0, -0, 100, -100, 200, 200]"},
+        {RoundMode::TOWARDS_INFINITY, "[0, 100, -100, -100, 200, -200, 200, 300]"},
+        {RoundMode::HALF_DOWN, "[0, 0, -0, -100, 100, -200, 200, 200]"},
+        {RoundMode::HALF_UP, "[0, 0, -0, -0, 100, -200, 200, 300]"},
+        {RoundMode::HALF_TOWARDS_ZERO, "[0, 0, -0, -0, 100, -200, 200, 200]"},
+        {RoundMode::HALF_TOWARDS_INFINITY, "[0, 0, -0, -100, 100, -200, 200, 300]"},
+        {RoundMode::HALF_TO_EVEN, "[0, 0, -0, -0, 100, -200, 200, 200]"},
+        {RoundMode::HALF_TO_ODD, "[0, 0, -0, -100, 100, -200, 200, 300]"},
+    }};
+    for (const auto& pair : round_modes_and_expected) {
+      this->SetRoundMode(pair.first);
+      this->AssertUnaryOp(Round, values,
+                          ArrayFromJSON(this->type_singleton(), pair.second));
+    }
+  }
+
+  // An overly large ndigits would be truncated to type's max digits
+  if constexpr (std::is_same_v<TypeParam, Int8Type>) {
+    this->SetRoundNdigits(-100);
+    this->SetRoundMode(RoundMode::UP);
+    auto values = "[1]";
+    this->AssertUnaryOp(Round, values, ArrayFromJSON(this->type_singleton(), "[100]"));
+  }
+
+  // A larger than double int64 should be correctly handled
+  if constexpr (std::is_same_v<TypeParam, Int64Type>) {
+    this->SetRoundNdigits(-2);
+    this->SetRoundMode(RoundMode::UP);
+    auto values = "[1152921504606846976]";  // 2 ^ 60
+    this->AssertUnaryOp(Round, values,
+                        ArrayFromJSON(this->type_singleton(), "[1152921504606847000]"));
   }
 }
 
@@ -1038,7 +1097,7 @@ TYPED_TEST(TestUnaryRoundUnsigned, Round) {
   this->SetRoundNdigits(0);
   for (const auto& round_mode : kRoundModes) {
     this->SetRoundMode(round_mode);
-    this->AssertUnaryOp(Round, values, ArrayFromJSON(float64(), values));
+    this->AssertUnaryOp(Round, values, ArrayFromJSON(this->type_singleton(), values));
   }
 
   // Test different round N-digits for nearest rounding mode
@@ -1052,7 +1111,49 @@ TYPED_TEST(TestUnaryRoundUnsigned, Round) {
   this->SetRoundMode(RoundMode::HALF_TOWARDS_INFINITY);
   for (const auto& pair : ndigits_and_expected) {
     this->SetRoundNdigits(pair.first);
-    this->AssertUnaryOp(Round, values, ArrayFromJSON(float64(), pair.second));
+    this->AssertUnaryOp(Round, values,
+                        ArrayFromJSON(this->type_singleton(), pair.second));
+  }
+
+  // Test different rounding mode
+  // skip uint8 because of its small range
+  if constexpr (!std::is_same_v<TypeParam, UInt8Type>) {
+    std::string values("[0, 1, 13, 50, 115, 176, 200, 250]");
+    this->SetRoundNdigits(-2);
+    std::vector<std::pair<RoundMode, std::string>> round_modes_and_expected{{
+        {RoundMode::DOWN, "[0, 0, 0, 0, 100, 100, 200, 200]"},
+        {RoundMode::UP, "[0, 100, 100, 100, 200, 200, 200, 300]"},
+        {RoundMode::TOWARDS_ZERO, "[0, 0, 0, 0, 100, 100, 200, 200]"},
+        {RoundMode::TOWARDS_INFINITY, "[0, 100, 100, 100, 200, 200, 200, 300]"},
+        {RoundMode::HALF_DOWN, "[0, 0, 0, 0, 100, 200, 200, 200]"},
+        {RoundMode::HALF_UP, "[0, 0, 0, 100, 100, 200, 200, 300]"},
+        {RoundMode::HALF_TOWARDS_ZERO, "[0, 0, 0, 0, 100, 200, 200, 200]"},
+        {RoundMode::HALF_TOWARDS_INFINITY, "[0, 0, 0, 100, 100, 200, 200, 300]"},
+        {RoundMode::HALF_TO_EVEN, "[0, 0, 0, 0, 100, 200, 200, 200]"},
+        {RoundMode::HALF_TO_ODD, "[0, 0, 0, 100, 100, 200, 200, 300]"},
+    }};
+    for (const auto& pair : round_modes_and_expected) {
+      this->SetRoundMode(pair.first);
+      this->AssertUnaryOp(Round, values,
+                          ArrayFromJSON(this->type_singleton(), pair.second));
+    }
+  }
+
+  // An overly large ndigits would be truncated to type's max digits
+  if constexpr (std::is_same_v<TypeParam, UInt8Type>) {
+    this->SetRoundNdigits(-100);
+    this->SetRoundMode(RoundMode::UP);
+    auto values = "[1]";
+    this->AssertUnaryOp(Round, values, ArrayFromJSON(this->type_singleton(), "[100]"));
+  }
+
+  // A larger than double uint64 should be correctly handled
+  if constexpr (std::is_same_v<TypeParam, Int64Type>) {
+    this->SetRoundNdigits(-2);
+    this->SetRoundMode(RoundMode::UP);
+    auto values = "[1152921504606846976]";  // 2 ^ 60
+    this->AssertUnaryOp(Round, values,
+                        ArrayFromJSON(this->type_singleton(), "[1152921504606847000]"));
   }
 }
 
@@ -1108,13 +1209,14 @@ TYPED_TEST(TestBinaryRoundSigned, Round) {
   std::string values("[0, 1, -13, -50, 115]");
   for (const auto& round_mode : kRoundModes) {
     this->SetRoundMode(round_mode);
-    this->AssertBinaryOp(RoundBinary, values, 0, ArrayFromJSON(float64(), values));
+    this->AssertBinaryOp(RoundBinary, values, 0,
+                         ArrayFromJSON(this->type_singleton(), values));
   }
 
   // Test different round N-digits for nearest rounding mode
   std::vector<std::pair<int32_t, std::string>> ndigits_and_expected{{
-      {-2, "[0.0, 0.0, -0.0, -100, 100]"},
-      {-1, "[0.0, 0.0, -10, -50, 120]"},
+      {-2, "[0, 0, -0, -100, 100]"},
+      {-1, "[0, 0, -10, -50, 120]"},
       {0, values},
       {1, values},
       {2, values},
@@ -1122,7 +1224,30 @@ TYPED_TEST(TestBinaryRoundSigned, Round) {
   this->SetRoundMode(RoundMode::HALF_TOWARDS_INFINITY);
   for (const auto& pair : ndigits_and_expected) {
     this->AssertBinaryOp(RoundBinary, values, pair.first,
-                         ArrayFromJSON(float64(), pair.second));
+                         ArrayFromJSON(this->type_singleton(), pair.second));
+  }
+
+  // Test different rounding mode
+  // skip int8 because of its small range
+  if constexpr (!std::is_same<TypeParam, Int8Type>::value) {
+    std::string values("[0, 1, -13, -50, 115, -176, 200, 250]");
+    std::vector<std::pair<RoundMode, std::string>> round_modes_and_expected{{
+        {RoundMode::DOWN, "[0, 0, -100, -100, 100, -200, 200, 200]"},
+        {RoundMode::UP, "[0, 100, -0, -0, 200, -100, 200, 300]"},
+        {RoundMode::TOWARDS_ZERO, "[0, 0, -0, -0, 100, -100, 200, 200]"},
+        {RoundMode::TOWARDS_INFINITY, "[0, 100, -100, -100, 200, -200, 200, 300]"},
+        {RoundMode::HALF_DOWN, "[0, 0, -0, -100, 100, -200, 200, 200]"},
+        {RoundMode::HALF_UP, "[0, 0, -0, -0, 100, -200, 200, 300]"},
+        {RoundMode::HALF_TOWARDS_ZERO, "[0, 0, -0, -0, 100, -200, 200, 200]"},
+        {RoundMode::HALF_TOWARDS_INFINITY, "[0, 0, -0, -100, 100, -200, 200, 300]"},
+        {RoundMode::HALF_TO_EVEN, "[0, 0, -0, -0, 100, -200, 200, 200]"},
+        {RoundMode::HALF_TO_ODD, "[0, 0, -0, -100, 100, -200, 200, 300]"},
+    }};
+    for (const auto& pair : round_modes_and_expected) {
+      this->SetRoundMode(pair.first);
+      this->AssertBinaryOp(RoundBinary, values, -2,
+                           ArrayFromJSON(this->type_singleton(), pair.second));
+    }
   }
 }
 
@@ -1131,7 +1256,8 @@ TYPED_TEST(TestBinaryRoundUnsigned, Round) {
   std::string values("[0, 1, 13, 50, 115]");
   for (const auto& round_mode : kRoundModes) {
     this->SetRoundMode(round_mode);
-    this->AssertBinaryOp(RoundBinary, values, 0, ArrayFromJSON(float64(), values));
+    this->AssertBinaryOp(RoundBinary, values, 0,
+                         ArrayFromJSON(this->type_singleton(), values));
   }
 
   // Test different round N-digits for nearest rounding mode
@@ -1145,7 +1271,30 @@ TYPED_TEST(TestBinaryRoundUnsigned, Round) {
   this->SetRoundMode(RoundMode::HALF_TOWARDS_INFINITY);
   for (const auto& pair : ndigits_and_expected) {
     this->AssertBinaryOp(RoundBinary, values, pair.first,
-                         ArrayFromJSON(float64(), pair.second));
+                         ArrayFromJSON(this->type_singleton(), pair.second));
+  }
+
+  // Test different rounding mode
+  // skip uint8 because of its small range
+  if constexpr (!std::is_same<TypeParam, UInt8Type>::value) {
+    std::string values("[0, 1, 13, 50, 115, 176, 200, 250]");
+    std::vector<std::pair<RoundMode, std::string>> round_modes_and_expected{{
+        {RoundMode::DOWN, "[0, 0, 0, 0, 100, 100, 200, 200]"},
+        {RoundMode::UP, "[0, 100, 100, 100, 200, 200, 200, 300]"},
+        {RoundMode::TOWARDS_ZERO, "[0, 0, 0, 0, 100, 100, 200, 200]"},
+        {RoundMode::TOWARDS_INFINITY, "[0, 100, 100, 100, 200, 200, 200, 300]"},
+        {RoundMode::HALF_DOWN, "[0, 0, 0, 0, 100, 200, 200, 200]"},
+        {RoundMode::HALF_UP, "[0, 0, 0, 100, 100, 200, 200, 300]"},
+        {RoundMode::HALF_TOWARDS_ZERO, "[0, 0, 0, 0, 100, 200, 200, 200]"},
+        {RoundMode::HALF_TOWARDS_INFINITY, "[0, 0, 0, 100, 100, 200, 200, 300]"},
+        {RoundMode::HALF_TO_EVEN, "[0, 0, 0, 0, 100, 200, 200, 200]"},
+        {RoundMode::HALF_TO_ODD, "[0, 0, 0, 100, 100, 200, 200, 300]"},
+    }};
+    for (const auto& pair : round_modes_and_expected) {
+      this->SetRoundMode(pair.first);
+      this->AssertBinaryOp(RoundBinary, values, -2,
+                           ArrayFromJSON(this->type_singleton(), pair.second));
+    }
   }
 }
 
@@ -1203,21 +1352,74 @@ TYPED_TEST(TestUnaryRoundToMultipleSigned, RoundToMultiple) {
   this->SetRoundMultiple(1);
   for (const auto& round_mode : kRoundModes) {
     this->SetRoundMode(round_mode);
-    this->AssertUnaryOp(RoundToMultiple, values, ArrayFromJSON(float64(), values));
+    this->AssertUnaryOp(RoundToMultiple, values,
+                        ArrayFromJSON(this->type_singleton(), values));
   }
 
+  // Out of range multiple is not allowed
+  this->SetRoundMultiple(
+      static_cast<double>(
+          std::numeric_limits<typename TypeTraits<TypeParam>::CType>::max()) +
+      1e9);
+  this->AssertUnaryOpRaises(RoundToMultiple, values, "Invalid");
+
   // Test different round multiples for nearest rounding mode
-  std::vector<std::pair<double, std::string>> multiple_and_expected{{
-      {2, "[0.0, 2, -14, -50, 116]"},
-      {0.05, "[0.0, 1, -13, -50, 115]"},
-      {0.1, values},
-      {10, "[0.0, 0.0, -10, -50, 120]"},
-      {100, "[0.0, 0.0, -0.0, -100, 100]"},
+  std::vector<std::pair<int, std::string>> multiple_and_expected{{
+      {2, "[0, 2, -14, -50, 116]"},
+      {10, "[0, 0, -10, -50, 120]"},
+      {100, "[0, 0, -0, -100, 100]"},
   }};
   this->SetRoundMode(RoundMode::HALF_TOWARDS_INFINITY);
   for (const auto& pair : multiple_and_expected) {
     this->SetRoundMultiple(pair.first);
-    this->AssertUnaryOp(RoundToMultiple, values, ArrayFromJSON(float64(), pair.second));
+    this->AssertUnaryOp(RoundToMultiple, values,
+                        ArrayFromJSON(this->type_singleton(), pair.second));
+  }
+
+  // Test different rounding mode
+  values = "[0, -1, 2, -5, 6, -8]";
+  // Even case is tested by round and round_binary so we test an odd case here
+  this->SetRoundMultiple(3);
+  std::vector<std::pair<RoundMode, std::string>> round_modes_and_expected{{
+      {RoundMode::DOWN, "[0, -3, 0, -6, 6, -9]"},
+      {RoundMode::UP, "[0, 0, 3, -3, 6, -6]"},
+      {RoundMode::TOWARDS_ZERO, "[0, 0, 0, -3, 6, -6]"},
+      {RoundMode::TOWARDS_INFINITY, "[0, -3, 3, -6, 6, -9]"},
+      {RoundMode::HALF_DOWN, "[0, 0, 3, -6, 6, -9]"},
+      {RoundMode::HALF_UP, "[0, 0, 3, -6, 6, -9]"},
+      {RoundMode::HALF_TOWARDS_ZERO, "[0, 0, 3, -6, 6, -9]"},
+      {RoundMode::HALF_TOWARDS_INFINITY, "[0, 0, 3, -6, 6, -9]"},
+      {RoundMode::HALF_TO_EVEN, "[0, 0, 3, -6, 6, -9]"},
+      {RoundMode::HALF_TO_ODD, "[0, 0, 3, -6, 6, -9]"},
+  }};
+  for (const auto& pair : round_modes_and_expected) {
+    this->SetRoundMode(pair.first);
+    this->AssertUnaryOp(RoundToMultiple, values,
+                        ArrayFromJSON(this->type_singleton(), pair.second));
+  }
+
+  if constexpr (std::is_same_v<TypeParam, Int32Type>) {
+    // Test overflow handling
+    this->SetRoundMultiple(10);
+    auto input = "[-2147483645]";
+    std::vector<RoundMode> invalid_modes{
+        RoundMode::DOWN, RoundMode::TOWARDS_INFINITY, RoundMode::HALF_DOWN,
+        RoundMode::HALF_TOWARDS_INFINITY, RoundMode::HALF_TO_ODD};
+    std::vector<RoundMode> valid_modes{RoundMode::UP, RoundMode::TOWARDS_ZERO,
+                                       RoundMode::HALF_UP, RoundMode::HALF_TOWARDS_ZERO,
+                                       RoundMode::HALF_TO_EVEN};
+    for (auto mode : invalid_modes) {
+      this->SetRoundMode(mode);
+      this->AssertUnaryOpRaises(
+          RoundToMultiple, input,
+          "Rounding -2147483645 down to multiple of 10 would overflow");
+    }
+
+    for (auto mode : valid_modes) {
+      this->SetRoundMode(mode);
+      this->AssertUnaryOp(RoundToMultiple, input,
+                          ArrayFromJSON(int32(), "[-2147483640]"));
+    }
   }
 }
 
@@ -1227,13 +1429,19 @@ TYPED_TEST(TestUnaryRoundToMultipleUnsigned, RoundToMultiple) {
   this->SetRoundMultiple(1);
   for (const auto& round_mode : kRoundModes) {
     this->SetRoundMode(round_mode);
-    this->AssertUnaryOp(RoundToMultiple, values, ArrayFromJSON(float64(), values));
+    this->AssertUnaryOp(RoundToMultiple, values,
+                        ArrayFromJSON(this->type_singleton(), values));
   }
 
+  // Out of range multiple is not allowed
+  this->SetRoundMultiple(
+      static_cast<double>(
+          std::numeric_limits<typename TypeTraits<TypeParam>::CType>::max()) +
+      1e9);
+  this->AssertUnaryOpRaises(RoundToMultiple, values, "Invalid");
+
   // Test different round multiples for nearest rounding mode
   std::vector<std::pair<double, std::string>> multiple_and_expected{{
-      {0.05, "[0, 1, 13, 50, 115]"},
-      {0.1, values},
       {2, "[0, 2, 14, 50, 116]"},
       {10, "[0, 0, 10, 50, 120]"},
       {100, "[0, 0, 0, 100, 100]"},
@@ -1241,7 +1449,54 @@ TYPED_TEST(TestUnaryRoundToMultipleUnsigned, RoundToMultiple) {
   this->SetRoundMode(RoundMode::HALF_TOWARDS_INFINITY);
   for (const auto& pair : multiple_and_expected) {
     this->SetRoundMultiple(pair.first);
-    this->AssertUnaryOp(RoundToMultiple, values, ArrayFromJSON(float64(), pair.second));
+    this->AssertUnaryOp(RoundToMultiple, values,
+                        ArrayFromJSON(this->type_singleton(), pair.second));
+  }
+
+  // Test different rounding mode
+  values = "[0, 1, 2, 5, 6, 8]";
+  // Even case is tested by round and round_binary so we test an odd case here
+  this->SetRoundMultiple(3);
+  std::vector<std::pair<RoundMode, std::string>> round_modes_and_expected{{
+      {RoundMode::DOWN, "[0, 0, 0, 3, 6, 6]"},
+      {RoundMode::UP, "[0, 3, 3, 6, 6, 9]"},
+      {RoundMode::TOWARDS_ZERO, "[0, 0, 0, 3, 6, 6]"},
+      {RoundMode::TOWARDS_INFINITY, "[0, 3, 3, 6, 6, 9]"},
+      {RoundMode::HALF_DOWN, "[0, 0, 3, 6, 6, 9]"},
+      {RoundMode::HALF_UP, "[0, 0, 3, 6, 6, 9]"},
+      {RoundMode::HALF_TOWARDS_ZERO, "[0, 0, 3, 6, 6, 9]"},
+      {RoundMode::HALF_TOWARDS_INFINITY, "[0, 0, 3, 6, 6, 9]"},
+      {RoundMode::HALF_TO_EVEN, "[0, 0, 3, 6, 6, 9]"},
+      {RoundMode::HALF_TO_ODD, "[0, 0, 3, 6, 6, 9]"},
+  }};
+  for (const auto& pair : round_modes_and_expected) {
+    this->SetRoundMode(pair.first);
+    this->AssertUnaryOp(RoundToMultiple, values,
+                        ArrayFromJSON(this->type_singleton(), pair.second));
+  }
+
+  if constexpr (std::is_same_v<TypeParam, UInt32Type>) {
+    // Test overflow handling
+    this->SetRoundMultiple(10);
+    auto input = "[4294967295]";
+    std::vector<RoundMode> valid_modes{RoundMode::DOWN, RoundMode::TOWARDS_ZERO,
+                                       RoundMode::HALF_DOWN, RoundMode::HALF_TOWARDS_ZERO,
+                                       RoundMode::HALF_TO_ODD};
+    std::vector<RoundMode> invalid_modes{RoundMode::UP, RoundMode::TOWARDS_INFINITY,
+                                         RoundMode::HALF_UP, RoundMode::TOWARDS_INFINITY,
+                                         RoundMode::HALF_TO_EVEN};
+    for (auto mode : invalid_modes) {
+      this->SetRoundMode(mode);
+      this->AssertUnaryOpRaises(
+          RoundToMultiple, input,
+          "Rounding 4294967295 up to multiple of 10 would overflow");
+    }
+
+    for (auto mode : valid_modes) {
+      this->SetRoundMode(mode);
+      this->AssertUnaryOp(RoundToMultiple, input,
+                          ArrayFromJSON(uint32(), "[4294967290]"));
+    }
   }
 }
 
@@ -1444,5 +1699,4 @@ TYPED_TEST(TestUnaryRoundArithmeticFloating, Trunc) {
 }
 
 }  // namespace
-}  // namespace compute
-}  // namespace arrow
+}  // namespace arrow::compute
diff --git a/docs/source/cpp/compute.rst b/docs/source/cpp/compute.rst
index 70c17ae2b96ea..8265188eaadd7 100644
--- a/docs/source/cpp/compute.rst
+++ b/docs/source/cpp/compute.rst
@@ -563,30 +563,34 @@ representation based on the rounding criterion.
 +-------------------+------------+-------------+-------------------------+----------------------------------+--------+
 | floor             | Unary      | Numeric     | Float32/Float64/Decimal |                                  |        |
 +-------------------+------------+-------------+-------------------------+----------------------------------+--------+
-| round             | Unary      | Numeric     | Float32/Float64/Decimal | :struct:`RoundOptions`           | (1)(2) |
+| round             | Unary      | Numeric     | Input Type              | :struct:`RoundOptions`           | (1)(2) |
 +-------------------+------------+-------------+-------------------------+----------------------------------+--------+
-| round_to_multiple | Unary      | Numeric     | Float32/Float64/Decimal | :struct:`RoundToMultipleOptions` | (1)(3) |
+| round_to_multiple | Unary      | Numeric     | Input Type              | :struct:`RoundToMultipleOptions` | (1)(3) |
 +-------------------+------------+-------------+-------------------------+----------------------------------+--------+
 | trunc             | Unary      | Numeric     | Float32/Float64/Decimal |                                  |        |
 +-------------------+------------+-------------+-------------------------+----------------------------------+--------+
 
-* \(1) Output value is a 64-bit floating-point for integral inputs and the
-  retains the same type for floating-point and decimal inputs.  By default
-  rounding functions displace a value to the nearest integer using
-  HALF_TO_EVEN to resolve ties.  Options are available to control the rounding
-  criterion.  Both ``round`` and ``round_to_multiple`` have the ``round_mode``
-  option to set the rounding mode.
+* \(1)  By default rounding functions displace a value to the nearest 
+  integer using HALF_TO_EVEN to resolve ties.  Options are available to control 
+  the rounding criterion.  Both ``round`` and ``round_to_multiple`` have the 
+  ``round_mode`` option to set the rounding mode.
 * \(2) Round to a number of digits where the ``ndigits`` option of
   :struct:`RoundOptions` specifies the rounding precision in terms of number
   of digits.  A negative value corresponds to digits in the non-fractional
   part.  For example, -2 corresponds to rounding to the nearest multiple of
   100 (zeroing the ones and tens digits).  Default value of ``ndigits`` is 0
-  which rounds to the nearest integer.
+  which rounds to the nearest integer. For integer inputs a non-negative 
+  ``ndigits`` value is ignored and the input is returned unchanged. For integer
+  inputs, if ``-ndigits`` is larger than the maximum number of digits the 
+  input type can hold, it is truncated to the maximum digit. For example, 
+  ``round([123], ndigits=-4, round_mode=DOWN)`` returns [100] for ``int8`` type.
+  For integer inputs, an error is returned on overflow.
 * \(3) Round to a multiple where the ``multiple`` option of
   :struct:`RoundToMultipleOptions` specifies the rounding scale.  The rounding
-  multiple has to be a positive value.  For example, 100 corresponds to
-  rounding to the nearest multiple of 100 (zeroing the ones and tens digits).
-  Default value of ``multiple`` is 1 which rounds to the nearest integer.
+  multiple has to be a positive value and can be casted to input type.  
+  For example, 100 corresponds to ounding to the nearest multiple of 100 
+  (zeroing the ones and tens digits). Default value of ``multiple`` is 1 which 
+  rounds to the nearest integer.
 
 For ``round`` and ``round_to_multiple``, the following rounding modes are available.
 Tie-breaking modes are prefixed with HALF and round non-ties to the nearest integer.

From 063e73eab4f789699b8181f99b4df75076c47ff1 Mon Sep 17 00:00:00 2001
From: Jin Shang <shangjin1997@gmail.com>
Date: Thu, 22 Jun 2023 13:43:08 +0800
Subject: [PATCH 02/12] lint

---
 cpp/src/arrow/compute/kernels/scalar_round.cc | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/cpp/src/arrow/compute/kernels/scalar_round.cc b/cpp/src/arrow/compute/kernels/scalar_round.cc
index 522aff38e2cab..3e9515fd02968 100644
--- a/cpp/src/arrow/compute/kernels/scalar_round.cc
+++ b/cpp/src/arrow/compute/kernels/scalar_round.cc
@@ -120,7 +120,7 @@ struct RoundUtil {
   // Calculate powers of ten with arbitrary integer exponent
   template <typename T>
   static enable_if_integer_value<T> Pow10(int64_t power) {
-    DCHECK(power >= 0);
+    DCHECK_GE(power, 0);
 
     static constexpr uint64_t lut[] = {
         Pow10Struct<0>::value,  Pow10Struct<1>::value,  Pow10Struct<2>::value,

From f1a272939a29418dfd8490968536f6afa7433991 Mon Sep 17 00:00:00 2001
From: Jin Shang <shangjin1997@gmail.com>
Date: Thu, 22 Jun 2023 14:04:42 +0800
Subject: [PATCH 03/12] fix warning

---
 cpp/src/arrow/compute/kernels/scalar_round.cc | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/cpp/src/arrow/compute/kernels/scalar_round.cc b/cpp/src/arrow/compute/kernels/scalar_round.cc
index 3e9515fd02968..68f0b554d35e4 100644
--- a/cpp/src/arrow/compute/kernels/scalar_round.cc
+++ b/cpp/src/arrow/compute/kernels/scalar_round.cc
@@ -132,7 +132,7 @@ struct RoundUtil {
         Pow10Struct<18>::value, Pow10Struct<19>::value};
 
     auto digits10 = std::numeric_limits<T>::digits10;
-    return lut[std::min(power, static_cast<int64_t>(digits10))];
+    return static_cast<T>(lut[std::min(power, static_cast<int64_t>(digits10))]);
   }
 };
 

From 627cc1a30528e8f8f6e2bd4bbefced0328446292 Mon Sep 17 00:00:00 2001
From: Jin Shang <shangjin1997@gmail.com>
Date: Thu, 22 Jun 2023 17:55:24 +0800
Subject: [PATCH 04/12] fix existing tests

---
 cpp/src/arrow/engine/substrait/function_test.cc | 2 +-
 python/pyarrow/tests/test_compute.py            | 4 ++--
 2 files changed, 3 insertions(+), 3 deletions(-)

diff --git a/cpp/src/arrow/engine/substrait/function_test.cc b/cpp/src/arrow/engine/substrait/function_test.cc
index eee98ec7199d2..f0d2583e815ae 100644
--- a/cpp/src/arrow/engine/substrait/function_test.cc
+++ b/cpp/src/arrow/engine/substrait/function_test.cc
@@ -490,7 +490,7 @@ TEST(FunctionMapping, ValidCases) {
        {{"rounding", {"TIE_AWAY_FROM_ZERO"}}},
        {int64(), int32()},
        "300",
-       float64()},
+       int64()},
   };
   CheckValidTestCases(valid_test_cases);
 }
diff --git a/python/pyarrow/tests/test_compute.py b/python/pyarrow/tests/test_compute.py
index 38bdeb126348b..f215115f27e07 100644
--- a/python/pyarrow/tests/test_compute.py
+++ b/python/pyarrow/tests/test_compute.py
@@ -1610,9 +1610,9 @@ def test_round_binary():
     scale = pa.scalar(-1, pa.int32())
 
     assert pc.round_binary(
-        5, scale, round_mode="half_towards_zero") == expect_zero
+        5.0, scale, round_mode="half_towards_zero") == expect_zero
     assert pc.round_binary(
-        5, scale, round_mode="half_towards_infinity") == expect_inf
+        5.0, scale, round_mode="half_towards_infinity") == expect_inf
 
 
 def test_is_null():

From eb5a7396227e8267736db4c985c480513744bd91 Mon Sep 17 00:00:00 2001
From: Jin Shang <shangjin1997@gmail.com>
Date: Thu, 6 Jul 2023 17:52:22 +0800
Subject: [PATCH 05/12] pr feedback

---
 .../arrow/compute/kernels/codegen_internal.h  |  5 +++++
 cpp/src/arrow/compute/kernels/scalar_round.cc | 20 ++++++++++++++++---
 .../kernels/scalar_round_arithmetic_test.cc   | 20 +++++++++++++++----
 docs/source/cpp/compute.rst                   |  6 ++----
 4 files changed, 40 insertions(+), 11 deletions(-)

diff --git a/cpp/src/arrow/compute/kernels/codegen_internal.h b/cpp/src/arrow/compute/kernels/codegen_internal.h
index 6224a9fc2af99..b6e960c86c30d 100644
--- a/cpp/src/arrow/compute/kernels/codegen_internal.h
+++ b/cpp/src/arrow/compute/kernels/codegen_internal.h
@@ -198,6 +198,11 @@ template <typename T>
 using is_signed_integer_value =
     std::integral_constant<bool, std::is_integral<T>::value && std::is_signed<T>::value>;
 
+template <typename T>
+using is_integer_value =
+    std::integral_constant<bool, is_signed_integer_value<T>::value ||
+                                     is_unsigned_integer_value<T>::value>;
+
 template <typename T, typename R = T>
 using enable_if_signed_integer_value = enable_if_t<is_signed_integer_value<T>::value, R>;
 
diff --git a/cpp/src/arrow/compute/kernels/scalar_round.cc b/cpp/src/arrow/compute/kernels/scalar_round.cc
index 68f0b554d35e4..d51021749dfd2 100644
--- a/cpp/src/arrow/compute/kernels/scalar_round.cc
+++ b/cpp/src/arrow/compute/kernels/scalar_round.cc
@@ -27,6 +27,7 @@
 #include "arrow/compute/cast.h"
 #include "arrow/compute/kernel.h"
 #include "arrow/compute/kernels/base_arithmetic_internal.h"
+#include "arrow/compute/kernels/codegen_internal.h"
 #include "arrow/compute/kernels/common_internal.h"
 #include "arrow/compute/kernels/util_internal.h"
 #include "arrow/type.h"
@@ -121,7 +122,7 @@ struct RoundUtil {
   template <typename T>
   static enable_if_integer_value<T> Pow10(int64_t power) {
     DCHECK_GE(power, 0);
-
+    DCHECK_LE(power, std::numeric_limits<T>::digits10);
     static constexpr uint64_t lut[] = {
         Pow10Struct<0>::value,  Pow10Struct<1>::value,  Pow10Struct<2>::value,
         Pow10Struct<3>::value,  Pow10Struct<4>::value,  Pow10Struct<5>::value,
@@ -131,8 +132,7 @@ struct RoundUtil {
         Pow10Struct<15>::value, Pow10Struct<16>::value, Pow10Struct<17>::value,
         Pow10Struct<18>::value, Pow10Struct<19>::value};
 
-    auto digits10 = std::numeric_limits<T>::digits10;
-    return static_cast<T>(lut[std::min(power, static_cast<int64_t>(digits10))]);
+    return static_cast<T>(lut[power]);
   }
 };
 
@@ -433,6 +433,13 @@ struct RoundOptionsWrapper<RoundOptions, CType> : public OptionsWrapper<RoundOpt
   static Result<std::unique_ptr<KernelState>> Init(KernelContext* ctx,
                                                    const KernelInitArgs& args) {
     if (auto options = static_cast<const OptionsType*>(args.options)) {
+      if constexpr (is_integer_value<CType>::value) {
+        if (-options->ndigits > std::numeric_limits<CType>::digits10) {
+          return Status::Invalid("Rounding to ", options->ndigits,
+                                 " digits is out of range for type ",
+                                 args.inputs[0].ToString());
+        }
+      }
       return std::make_unique<RoundOptionsWrapper>(*options);
     }
     return Status::Invalid(
@@ -995,6 +1002,13 @@ struct RoundBinary<ArrowType, kRoundMode, enable_if_integer<ArrowType>> {
       return arg0;
     }
 
+    if (-arg1 > std::numeric_limits<CType>::digits10) {
+      // ndigits is larger than the number of digits CType can hold
+      *st = Status::Invalid("Rounding to ", arg1, " digits is out of range for type ",
+                            out_ty.ToString());
+      return arg0;
+    }
+
     // If ndigits is negative, then round to the nearest multiple of 10^ndigits.
     CType pow10 = RoundUtil::Pow10<CType>(std::abs(arg1));
     RoundToMultiple<ArrowType, kRoundMode> round_to_multiple(pow10, out_ty);
diff --git a/cpp/src/arrow/compute/kernels/scalar_round_arithmetic_test.cc b/cpp/src/arrow/compute/kernels/scalar_round_arithmetic_test.cc
index c0a36dd9deb43..3fd737980d6a4 100644
--- a/cpp/src/arrow/compute/kernels/scalar_round_arithmetic_test.cc
+++ b/cpp/src/arrow/compute/kernels/scalar_round_arithmetic_test.cc
@@ -1073,12 +1073,12 @@ TYPED_TEST(TestUnaryRoundSigned, Round) {
     }
   }
 
-  // An overly large ndigits would be truncated to type's max digits
+  // An overly large ndigits would cause an error
   if constexpr (std::is_same_v<TypeParam, Int8Type>) {
     this->SetRoundNdigits(-100);
     this->SetRoundMode(RoundMode::UP);
     auto values = "[1]";
-    this->AssertUnaryOp(Round, values, ArrayFromJSON(this->type_singleton(), "[100]"));
+    this->AssertUnaryOpRaises(Round, values, "out of range");
   }
 
   // A larger than double int64 should be correctly handled
@@ -1139,12 +1139,12 @@ TYPED_TEST(TestUnaryRoundUnsigned, Round) {
     }
   }
 
-  // An overly large ndigits would be truncated to type's max digits
+  // An overly large ndigits would cause an error
   if constexpr (std::is_same_v<TypeParam, UInt8Type>) {
     this->SetRoundNdigits(-100);
     this->SetRoundMode(RoundMode::UP);
     auto values = "[1]";
-    this->AssertUnaryOp(Round, values, ArrayFromJSON(this->type_singleton(), "[100]"));
+    this->AssertUnaryOpRaises(Round, values, "out of range");
   }
 
   // A larger than double uint64 should be correctly handled
@@ -1249,6 +1249,12 @@ TYPED_TEST(TestBinaryRoundSigned, Round) {
                            ArrayFromJSON(this->type_singleton(), pair.second));
     }
   }
+
+  // An overly large ndigits would cause an error
+  if constexpr (std::is_same_v<TypeParam, Int8Type>) {
+    this->SetRoundMode(RoundMode::UP);
+    this->AssertBinaryOpRaises(RoundBinary, "[1]", "[-100]", "out of range");
+  }
 }
 
 TYPED_TEST(TestBinaryRoundUnsigned, Round) {
@@ -1296,6 +1302,12 @@ TYPED_TEST(TestBinaryRoundUnsigned, Round) {
                            ArrayFromJSON(this->type_singleton(), pair.second));
     }
   }
+
+  // An overly large ndigits would cause an error
+  if constexpr (std::is_same_v<TypeParam, UInt8Type>) {
+    this->SetRoundMode(RoundMode::UP);
+    this->AssertBinaryOpRaises(RoundBinary, "[1]", "[-100]", "out of range");
+  }
 }
 
 TYPED_TEST(TestBinaryRoundFloating, Round) {
diff --git a/docs/source/cpp/compute.rst b/docs/source/cpp/compute.rst
index 8265188eaadd7..cfcc69497bd0f 100644
--- a/docs/source/cpp/compute.rst
+++ b/docs/source/cpp/compute.rst
@@ -582,13 +582,11 @@ representation based on the rounding criterion.
   which rounds to the nearest integer. For integer inputs a non-negative 
   ``ndigits`` value is ignored and the input is returned unchanged. For integer
   inputs, if ``-ndigits`` is larger than the maximum number of digits the 
-  input type can hold, it is truncated to the maximum digit. For example, 
-  ``round([123], ndigits=-4, round_mode=DOWN)`` returns [100] for ``int8`` type.
-  For integer inputs, an error is returned on overflow.
+  input type can hold, an error is returned.
 * \(3) Round to a multiple where the ``multiple`` option of
   :struct:`RoundToMultipleOptions` specifies the rounding scale.  The rounding
   multiple has to be a positive value and can be casted to input type.  
-  For example, 100 corresponds to ounding to the nearest multiple of 100 
+  For example, 100 corresponds to rounding to the nearest multiple of 100 
   (zeroing the ones and tens digits). Default value of ``multiple`` is 1 which 
   rounds to the nearest integer.
 

From de3c2892964792724f3813f835122357bc048cea Mon Sep 17 00:00:00 2001
From: Jin Shang <shangjin1997@gmail.com>
Date: Thu, 6 Jul 2023 18:02:58 +0800
Subject: [PATCH 06/12] add overflow test cast

---
 .../kernels/scalar_round_arithmetic_test.cc   | 26 +++++++++++++++++++
 1 file changed, 26 insertions(+)

diff --git a/cpp/src/arrow/compute/kernels/scalar_round_arithmetic_test.cc b/cpp/src/arrow/compute/kernels/scalar_round_arithmetic_test.cc
index 3fd737980d6a4..079c70d040368 100644
--- a/cpp/src/arrow/compute/kernels/scalar_round_arithmetic_test.cc
+++ b/cpp/src/arrow/compute/kernels/scalar_round_arithmetic_test.cc
@@ -1081,6 +1081,13 @@ TYPED_TEST(TestUnaryRoundSigned, Round) {
     this->AssertUnaryOpRaises(Round, values, "out of range");
   }
 
+  // Overflow is also treated as error
+  if constexpr (std::is_same_v<TypeParam, Int8Type>) {
+    this->SetRoundNdigits(-1);
+    this->SetRoundMode(RoundMode::DOWN);
+    this->AssertUnaryOpRaises(Round, "[-127]", "overflow");
+  }
+
   // A larger than double int64 should be correctly handled
   if constexpr (std::is_same_v<TypeParam, Int64Type>) {
     this->SetRoundNdigits(-2);
@@ -1147,6 +1154,13 @@ TYPED_TEST(TestUnaryRoundUnsigned, Round) {
     this->AssertUnaryOpRaises(Round, values, "out of range");
   }
 
+  // Overflow is also treated as error
+  if constexpr (std::is_same_v<TypeParam, UInt8Type>) {
+    this->SetRoundNdigits(-1);
+    this->SetRoundMode(RoundMode::UP);
+    this->AssertUnaryOpRaises(Round, "[255]", "overflow");
+  }
+
   // A larger than double uint64 should be correctly handled
   if constexpr (std::is_same_v<TypeParam, Int64Type>) {
     this->SetRoundNdigits(-2);
@@ -1255,6 +1269,12 @@ TYPED_TEST(TestBinaryRoundSigned, Round) {
     this->SetRoundMode(RoundMode::UP);
     this->AssertBinaryOpRaises(RoundBinary, "[1]", "[-100]", "out of range");
   }
+
+  // Overflow is also treated as error
+  if constexpr (std::is_same_v<TypeParam, Int8Type>) {
+    this->SetRoundMode(RoundMode::DOWN);
+    this->AssertBinaryOpRaises(RoundBinary, "[-127]", "[-1]", "overflow");
+  }
 }
 
 TYPED_TEST(TestBinaryRoundUnsigned, Round) {
@@ -1308,6 +1328,12 @@ TYPED_TEST(TestBinaryRoundUnsigned, Round) {
     this->SetRoundMode(RoundMode::UP);
     this->AssertBinaryOpRaises(RoundBinary, "[1]", "[-100]", "out of range");
   }
+
+  // Overflow is also treated as error
+  if constexpr (std::is_same_v<TypeParam, UInt8Type>) {
+    this->SetRoundMode(RoundMode::UP);
+    this->AssertBinaryOpRaises(RoundBinary, "[255]", "[-1]", "overflow");
+  }
 }
 
 TYPED_TEST(TestBinaryRoundFloating, Round) {

From 89fc535b13512cceec335abc7acb73df292528dc Mon Sep 17 00:00:00 2001
From: Jin Shang <shangjin1997@gmail.com>
Date: Mon, 17 Jul 2023 00:58:04 +0800
Subject: [PATCH 07/12] integer round use naive approach

---
 cpp/src/arrow/compute/kernels/scalar_round.cc | 48 +++++++++----------
 1 file changed, 22 insertions(+), 26 deletions(-)

diff --git a/cpp/src/arrow/compute/kernels/scalar_round.cc b/cpp/src/arrow/compute/kernels/scalar_round.cc
index d51021749dfd2..ed8b0bda0a944 100644
--- a/cpp/src/arrow/compute/kernels/scalar_round.cc
+++ b/cpp/src/arrow/compute/kernels/scalar_round.cc
@@ -84,25 +84,6 @@ bool IsPositive(const Scalar& scalar) {
   return visitor.result;
 }
 
-// N.B. take care not to conflict with type_traits.h as that can cause surprises in a
-// unity build
-
-// A constexpr helper struct to compute powers of 10 at compile time
-// Can use a consteval function once we force C++20
-template <int Exp>
-struct Pow10Struct {
- private:
-  static constexpr uint64_t half_pow = Pow10Struct<Exp / 2>::value;
-
- public:
-  static constexpr uint64_t value = half_pow * half_pow * (Exp % 2 ? 10 : 1);
-};
-
-template <>
-struct Pow10Struct<0> {
-  static constexpr uint64_t value = 1;
-};
-
 struct RoundUtil {
   // Calculate powers of ten with arbitrary integer exponent
   template <typename T>
@@ -124,13 +105,28 @@ struct RoundUtil {
     DCHECK_GE(power, 0);
     DCHECK_LE(power, std::numeric_limits<T>::digits10);
     static constexpr uint64_t lut[] = {
-        Pow10Struct<0>::value,  Pow10Struct<1>::value,  Pow10Struct<2>::value,
-        Pow10Struct<3>::value,  Pow10Struct<4>::value,  Pow10Struct<5>::value,
-        Pow10Struct<6>::value,  Pow10Struct<7>::value,  Pow10Struct<8>::value,
-        Pow10Struct<9>::value,  Pow10Struct<10>::value, Pow10Struct<11>::value,
-        Pow10Struct<12>::value, Pow10Struct<13>::value, Pow10Struct<14>::value,
-        Pow10Struct<15>::value, Pow10Struct<16>::value, Pow10Struct<17>::value,
-        Pow10Struct<18>::value, Pow10Struct<19>::value};
+        // clang-format off
+        1ULL,
+        10ULL,
+        100ULL,
+        1000ULL,
+        10000ULL,
+        100000ULL,
+        1000000ULL,
+        10000000ULL,
+        100000000ULL,
+        1000000000ULL,
+        10000000000ULL,
+        100000000000ULL,
+        1000000000000ULL,
+        10000000000000ULL,
+        100000000000000ULL,
+        1000000000000000ULL,
+        10000000000000000ULL,
+        100000000000000000ULL,
+        1000000000000000000ULL
+        // clang-format on 
+    };
 
     return static_cast<T>(lut[power]);
   }

From 4b4e85a7fc06bc85703f7d7e07e31827ca0a9d9f Mon Sep 17 00:00:00 2001
From: Jin Shang <shangjin1997@gmail.com>
Date: Mon, 17 Jul 2023 14:44:28 +0800
Subject: [PATCH 08/12] Update docs/source/cpp/compute.rst

Co-authored-by: Weston Pace <weston.pace@gmail.com>
---
 docs/source/cpp/compute.rst | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/docs/source/cpp/compute.rst b/docs/source/cpp/compute.rst
index cfcc69497bd0f..2a35b70afc4a3 100644
--- a/docs/source/cpp/compute.rst
+++ b/docs/source/cpp/compute.rst
@@ -570,7 +570,7 @@ representation based on the rounding criterion.
 | trunc             | Unary      | Numeric     | Float32/Float64/Decimal |                                  |        |
 +-------------------+------------+-------------+-------------------------+----------------------------------+--------+
 
-* \(1)  By default rounding functions displace a value to the nearest 
+* \(1)  By default rounding functions change a value to the nearest 
   integer using HALF_TO_EVEN to resolve ties.  Options are available to control 
   the rounding criterion.  Both ``round`` and ``round_to_multiple`` have the 
   ``round_mode`` option to set the rounding mode.

From f1f35f0dbfc1fcde24e07bea405d7189cc61c57a Mon Sep 17 00:00:00 2001
From: Jin Shang <shangjin1997@gmail.com>
Date: Mon, 17 Jul 2023 01:09:57 +0800
Subject: [PATCH 09/12] lint

---
 cpp/src/arrow/compute/kernels/scalar_round.cc | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/cpp/src/arrow/compute/kernels/scalar_round.cc b/cpp/src/arrow/compute/kernels/scalar_round.cc
index ed8b0bda0a944..124daa62ecd40 100644
--- a/cpp/src/arrow/compute/kernels/scalar_round.cc
+++ b/cpp/src/arrow/compute/kernels/scalar_round.cc
@@ -125,7 +125,7 @@ struct RoundUtil {
         10000000000000000ULL,
         100000000000000000ULL,
         1000000000000000000ULL
-        // clang-format on 
+        // clang-format on
     };
 
     return static_cast<T>(lut[power]);

From f0c2eaed4eece383ab50702bd0e910cc3971a6db Mon Sep 17 00:00:00 2001
From: Jin Shang <shangjin1997@gmail.com>
Date: Thu, 27 Jul 2023 09:18:30 +0800
Subject: [PATCH 10/12] review feedback

---
 cpp/src/arrow/compute/kernels/scalar_round.cc |  2 +-
 .../kernels/scalar_round_arithmetic_test.cc   | 62 +++++++++----------
 docs/source/cpp/compute.rst                   | 14 +++--
 3 files changed, 39 insertions(+), 39 deletions(-)

diff --git a/cpp/src/arrow/compute/kernels/scalar_round.cc b/cpp/src/arrow/compute/kernels/scalar_round.cc
index 124daa62ecd40..5dc628f8d95c5 100644
--- a/cpp/src/arrow/compute/kernels/scalar_round.cc
+++ b/cpp/src/arrow/compute/kernels/scalar_round.cc
@@ -397,7 +397,7 @@ struct RoundImpl<Type, RoundMode::HALF_TO_ODD> {
   template <typename T = Type>
   static constexpr enable_if_integer_value<T> Round(const T val, const T floor,
                                                     const T multiple, Status* st) {
-    if ((floor / multiple) % 2 == 1) {
+    if ((floor / multiple) % 2 != 0) {
       return floor;
     }
     return RoundImpl<T, RoundMode::TOWARDS_INFINITY>::Round(val, floor, multiple, st);
diff --git a/cpp/src/arrow/compute/kernels/scalar_round_arithmetic_test.cc b/cpp/src/arrow/compute/kernels/scalar_round_arithmetic_test.cc
index 079c70d040368..3d205b0451dc2 100644
--- a/cpp/src/arrow/compute/kernels/scalar_round_arithmetic_test.cc
+++ b/cpp/src/arrow/compute/kernels/scalar_round_arithmetic_test.cc
@@ -1052,19 +1052,19 @@ TYPED_TEST(TestUnaryRoundSigned, Round) {
   // Test different rounding mode
   // skip int8 because of its small range
   if constexpr (!std::is_same_v<TypeParam, Int8Type>) {
-    std::string values("[0, 1, -13, -50, 115, -176, 200, 250]");
+    std::string values("[0, 1, -13, -50, 115, -150, -176, 200, 250]");
     this->SetRoundNdigits(-2);
     std::vector<std::pair<RoundMode, std::string>> round_modes_and_expected{{
-        {RoundMode::DOWN, "[0, 0, -100, -100, 100, -200, 200, 200]"},
-        {RoundMode::UP, "[0, 100, -0, -0, 200, -100, 200, 300]"},
-        {RoundMode::TOWARDS_ZERO, "[0, 0, -0, -0, 100, -100, 200, 200]"},
-        {RoundMode::TOWARDS_INFINITY, "[0, 100, -100, -100, 200, -200, 200, 300]"},
-        {RoundMode::HALF_DOWN, "[0, 0, -0, -100, 100, -200, 200, 200]"},
-        {RoundMode::HALF_UP, "[0, 0, -0, -0, 100, -200, 200, 300]"},
-        {RoundMode::HALF_TOWARDS_ZERO, "[0, 0, -0, -0, 100, -200, 200, 200]"},
-        {RoundMode::HALF_TOWARDS_INFINITY, "[0, 0, -0, -100, 100, -200, 200, 300]"},
-        {RoundMode::HALF_TO_EVEN, "[0, 0, -0, -0, 100, -200, 200, 200]"},
-        {RoundMode::HALF_TO_ODD, "[0, 0, -0, -100, 100, -200, 200, 300]"},
+        {RoundMode::DOWN, "[0, 0, -100, -100, 100, -200, -200, 200, 200]"},
+        {RoundMode::UP, "[0, 100, -0, -0, 200, -100, -100, 200, 300]"},
+        {RoundMode::TOWARDS_ZERO, "[0, 0, -0, -0, 100, -100, -100, 200, 200]"},
+        {RoundMode::TOWARDS_INFINITY, "[0, 100, -100, -100, 200, -200, -200, 200, 300]"},
+        {RoundMode::HALF_DOWN, "[0, 0, -0, -100, 100, -200, -200, 200, 200]"},
+        {RoundMode::HALF_UP, "[0, 0, -0, -0, 100, -100, -200, 200, 300]"},
+        {RoundMode::HALF_TOWARDS_ZERO, "[0, 0, -0, -0, 100, -100, -200, 200, 200]"},
+        {RoundMode::HALF_TOWARDS_INFINITY, "[0, 0, -0, -100, 100, -200, -200, 200, 300]"},
+        {RoundMode::HALF_TO_EVEN, "[0, 0, -0, -0, 100, -200, -200, 200, 200]"},
+        {RoundMode::HALF_TO_ODD, "[0, 0, -0, -100, 100, -100, -200, 200, 300]"},
     }};
     for (const auto& pair : round_modes_and_expected) {
       this->SetRoundMode(pair.first);
@@ -1074,12 +1074,9 @@ TYPED_TEST(TestUnaryRoundSigned, Round) {
   }
 
   // An overly large ndigits would cause an error
-  if constexpr (std::is_same_v<TypeParam, Int8Type>) {
-    this->SetRoundNdigits(-100);
-    this->SetRoundMode(RoundMode::UP);
-    auto values = "[1]";
-    this->AssertUnaryOpRaises(Round, values, "out of range");
-  }
+  this->SetRoundNdigits(-100);
+  this->SetRoundMode(RoundMode::UP);
+  this->AssertUnaryOpRaises(Round, "[1]", "out of range");
 
   // Overflow is also treated as error
   if constexpr (std::is_same_v<TypeParam, Int8Type>) {
@@ -1125,19 +1122,19 @@ TYPED_TEST(TestUnaryRoundUnsigned, Round) {
   // Test different rounding mode
   // skip uint8 because of its small range
   if constexpr (!std::is_same_v<TypeParam, UInt8Type>) {
-    std::string values("[0, 1, 13, 50, 115, 176, 200, 250]");
+    std::string values("[0, 1, 13, 50, 115, 150, 176, 200, 250]");
     this->SetRoundNdigits(-2);
     std::vector<std::pair<RoundMode, std::string>> round_modes_and_expected{{
-        {RoundMode::DOWN, "[0, 0, 0, 0, 100, 100, 200, 200]"},
-        {RoundMode::UP, "[0, 100, 100, 100, 200, 200, 200, 300]"},
-        {RoundMode::TOWARDS_ZERO, "[0, 0, 0, 0, 100, 100, 200, 200]"},
-        {RoundMode::TOWARDS_INFINITY, "[0, 100, 100, 100, 200, 200, 200, 300]"},
-        {RoundMode::HALF_DOWN, "[0, 0, 0, 0, 100, 200, 200, 200]"},
-        {RoundMode::HALF_UP, "[0, 0, 0, 100, 100, 200, 200, 300]"},
-        {RoundMode::HALF_TOWARDS_ZERO, "[0, 0, 0, 0, 100, 200, 200, 200]"},
-        {RoundMode::HALF_TOWARDS_INFINITY, "[0, 0, 0, 100, 100, 200, 200, 300]"},
-        {RoundMode::HALF_TO_EVEN, "[0, 0, 0, 0, 100, 200, 200, 200]"},
-        {RoundMode::HALF_TO_ODD, "[0, 0, 0, 100, 100, 200, 200, 300]"},
+        {RoundMode::DOWN, "[0, 0, 0, 0, 100, 100, 100, 200, 200]"},
+        {RoundMode::UP, "[0, 100, 100, 100, 200, 200, 200, 200, 300]"},
+        {RoundMode::TOWARDS_ZERO, "[0, 0, 0, 0, 100, 100, 100, 200, 200]"},
+        {RoundMode::TOWARDS_INFINITY, "[0, 100, 100, 100, 200, 200, 200, 200, 300]"},
+        {RoundMode::HALF_DOWN, "[0, 0, 0, 0, 100, 100, 200, 200, 200]"},
+        {RoundMode::HALF_UP, "[0, 0, 0, 100, 100, 200, 200, 200, 300]"},
+        {RoundMode::HALF_TOWARDS_ZERO, "[0, 0, 0, 0, 100, 100, 200, 200, 200]"},
+        {RoundMode::HALF_TOWARDS_INFINITY, "[0, 0, 0, 100, 100, 200, 200, 200, 300]"},
+        {RoundMode::HALF_TO_EVEN, "[0, 0, 0, 0, 100, 200, 200, 200, 200]"},
+        {RoundMode::HALF_TO_ODD, "[0, 0, 0, 100, 100, 100, 200, 200, 300]"},
     }};
     for (const auto& pair : round_modes_and_expected) {
       this->SetRoundMode(pair.first);
@@ -1147,12 +1144,9 @@ TYPED_TEST(TestUnaryRoundUnsigned, Round) {
   }
 
   // An overly large ndigits would cause an error
-  if constexpr (std::is_same_v<TypeParam, UInt8Type>) {
-    this->SetRoundNdigits(-100);
-    this->SetRoundMode(RoundMode::UP);
-    auto values = "[1]";
-    this->AssertUnaryOpRaises(Round, values, "out of range");
-  }
+  this->SetRoundNdigits(-100);
+  this->SetRoundMode(RoundMode::UP);
+  this->AssertUnaryOpRaises(Round, "[1]", "out of range");
 
   // Overflow is also treated as error
   if constexpr (std::is_same_v<TypeParam, UInt8Type>) {
diff --git a/docs/source/cpp/compute.rst b/docs/source/cpp/compute.rst
index 2a35b70afc4a3..d537f349af542 100644
--- a/docs/source/cpp/compute.rst
+++ b/docs/source/cpp/compute.rst
@@ -567,12 +567,14 @@ representation based on the rounding criterion.
 +-------------------+------------+-------------+-------------------------+----------------------------------+--------+
 | round_to_multiple | Unary      | Numeric     | Input Type              | :struct:`RoundToMultipleOptions` | (1)(3) |
 +-------------------+------------+-------------+-------------------------+----------------------------------+--------+
+| round_binary      | Binary     | Numeric     | Input Type              | :struct:`RoundBinaryOptions`     | (1)(4) |
++-------------------+------------+-------------+-------------------------+----------------------------------+--------+
 | trunc             | Unary      | Numeric     | Float32/Float64/Decimal |                                  |        |
 +-------------------+------------+-------------+-------------------------+----------------------------------+--------+
 
 * \(1)  By default rounding functions change a value to the nearest 
   integer using HALF_TO_EVEN to resolve ties.  Options are available to control 
-  the rounding criterion.  Both ``round`` and ``round_to_multiple`` have the 
+  the rounding criterion.  All ``round`` functions have the 
   ``round_mode`` option to set the rounding mode.
 * \(2) Round to a number of digits where the ``ndigits`` option of
   :struct:`RoundOptions` specifies the rounding precision in terms of number
@@ -589,8 +591,12 @@ representation based on the rounding criterion.
   For example, 100 corresponds to rounding to the nearest multiple of 100 
   (zeroing the ones and tens digits). Default value of ``multiple`` is 1 which 
   rounds to the nearest integer.
+* \(4) Round the first input to multiple of the second input. The rounding
+  multiple has to be a positive value and can be casted to the first input type.  
+  For example, 100 corresponds to rounding to the nearest multiple of 100 
+  (zeroing the ones and tens digits).
 
-For ``round`` and ``round_to_multiple``, the following rounding modes are available.
+For ``round`` functions, the following rounding modes are available.
 Tie-breaking modes are prefixed with HALF and round non-ties to the nearest integer.
 The example values are given for default values of ``ndigits`` and ``multiple``.
 
@@ -629,8 +635,8 @@ The example values are given for default values of ``ndigits`` and ``multiple``.
 +-----------------------+--------------------------------------------------------------+---------------------------+
 
 The following table gives examples of how ``ndigits`` (for the ``round``
-function) and ``multiple`` (for ``round_to_multiple``) influence the operance
-performed, respectively.
+function) and ``multiple`` (for ``round_to_multiple`` and ``round_binary``) 
+influence the operance performed, respectively.
 
 +--------------------+-------------------+---------------------------+
 | Round ``multiple`` | Round ``ndigits`` | Operation performed       |

From 88d26195d61ebe77c440ce1a146c2df9a467f203 Mon Sep 17 00:00:00 2001
From: Jin Shang <shangjin1997@gmail.com>
Date: Wed, 9 Aug 2023 22:22:41 +0800
Subject: [PATCH 11/12] Update docs/source/cpp/compute.rst

Co-authored-by: Antoine Pitrou <pitrou@free.fr>
---
 docs/source/cpp/compute.rst | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/docs/source/cpp/compute.rst b/docs/source/cpp/compute.rst
index d537f349af542..e8abf12678895 100644
--- a/docs/source/cpp/compute.rst
+++ b/docs/source/cpp/compute.rst
@@ -636,7 +636,7 @@ The example values are given for default values of ``ndigits`` and ``multiple``.
 
 The following table gives examples of how ``ndigits`` (for the ``round``
 function) and ``multiple`` (for ``round_to_multiple`` and ``round_binary``) 
-influence the operance performed, respectively.
+influence the operation performed, respectively.
 
 +--------------------+-------------------+---------------------------+
 | Round ``multiple`` | Round ``ndigits`` | Operation performed       |

From 0a8d38d530d2974dbed46b6344e4186c5bb1fa13 Mon Sep 17 00:00:00 2001
From: Jin Shang <shangjin1997@gmail.com>
Date: Tue, 8 Aug 2023 00:34:28 +0800
Subject: [PATCH 12/12] fix doc

---
 docs/source/cpp/compute.rst | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/docs/source/cpp/compute.rst b/docs/source/cpp/compute.rst
index e8abf12678895..5f7d499994302 100644
--- a/docs/source/cpp/compute.rst
+++ b/docs/source/cpp/compute.rst
@@ -634,8 +634,8 @@ The example values are given for default values of ``ndigits`` and ``multiple``.
 |                       |                                                              | -3.5 -> -3, -4.5 -> -5    |
 +-----------------------+--------------------------------------------------------------+---------------------------+
 
-The following table gives examples of how ``ndigits`` (for the ``round``
-function) and ``multiple`` (for ``round_to_multiple`` and ``round_binary``) 
+The following table gives examples of how ``ndigits`` (for the ``round`` 
+and ``round_binary`` functions) and ``multiple`` (for ``round_to_multiple``) 
 influence the operation performed, respectively.
 
 +--------------------+-------------------+---------------------------+