From f3ce7ea800ee9299ad8929474a8accf5d90214d1 Mon Sep 17 00:00:00 2001
From: David Wendt <dwendt@nvidia.com>
Date: Mon, 7 Feb 2022 11:09:13 -0500
Subject: [PATCH] Fix string to decimal128 conversion handling large exponents

---
 .../strings/detail/convert/fixed_point.cuh     | 12 ++++++++----
 cpp/tests/strings/fixed_point_tests.cpp        | 18 +++++++++++++++++-
 2 files changed, 25 insertions(+), 5 deletions(-)

diff --git a/cpp/include/cudf/strings/detail/convert/fixed_point.cuh b/cpp/include/cudf/strings/detail/convert/fixed_point.cuh
index aa3f544202f..7af56f89449 100644
--- a/cpp/include/cudf/strings/detail/convert/fixed_point.cuh
+++ b/cpp/include/cudf/strings/detail/convert/fixed_point.cuh
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2021, NVIDIA CORPORATION.
+ * Copyright (c) 2021-2022, NVIDIA CORPORATION.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
@@ -14,6 +14,8 @@
  * limitations under the License.
  */
 
+#include <cudf/fixed_point/temporary.hpp>
+
 #include <thrust/optional.h>
 #include <thrust/pair.h>
 
@@ -150,9 +152,11 @@ __device__ DecimalType parse_decimal(char const* iter, char const* iter_end, int
   exp_ten += exp_offset;
 
   // shift the output value based on the exp_ten and the scale values
-  value = exp_ten < scale
-            ? value / static_cast<UnsignedDecimalType>(exp10(static_cast<double>(scale - exp_ten)))
-            : value * static_cast<UnsignedDecimalType>(exp10(static_cast<double>(exp_ten - scale)));
+  auto const shift_adjust =
+    abs(scale - exp_ten) > cuda::std::numeric_limits<UnsignedDecimalType>::digits10
+      ? cuda::std::numeric_limits<UnsignedDecimalType>::max()
+      : numeric::detail::exp10<UnsignedDecimalType>(abs(scale - exp_ten));
+  value = exp_ten < scale ? value / shift_adjust : value * shift_adjust;
 
   return static_cast<DecimalType>(value) * (sign == 0 ? 1 : sign);
 }
diff --git a/cpp/tests/strings/fixed_point_tests.cpp b/cpp/tests/strings/fixed_point_tests.cpp
index 5872a9e5bb7..81122b1c5d8 100644
--- a/cpp/tests/strings/fixed_point_tests.cpp
+++ b/cpp/tests/strings/fixed_point_tests.cpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2021, NVIDIA CORPORATION.
+ * Copyright (c) 2021-2022, NVIDIA CORPORATION.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
@@ -112,6 +112,22 @@ TEST_F(StringsConvertTest, ToFixedPointDecimal128)
   CUDF_TEST_EXPECT_COLUMNS_EQUIVALENT(*results, expected);
 }
 
+TEST_F(StringsConvertTest, ToFixedPointLargeScale)
+{
+  using namespace numeric;
+  using RepType    = cudf::device_storage_type_t<decimal128>;
+  using fp_wrapper = cudf::test::fixed_point_column_wrapper<RepType>;
+
+  auto const strings = cudf::test::strings_column_wrapper({"0.05", "0.06", "0.50", "5.01"});
+
+  auto const scale   = scale_type{-25};
+  auto const type    = cudf::data_type{cudf::type_to_id<decimal128>(), scale};
+  auto const results = cudf::strings::to_fixed_point(cudf::strings_column_view(strings), type);
+
+  auto const expected = fp_wrapper{{5, 6, 50, 501}, scale_type{-2}};
+  CUDF_TEST_EXPECT_COLUMNS_EQUIVALENT(*results, expected);
+}
+
 TEST_F(StringsConvertTest, FromFixedPointDecimal128)
 {
   using namespace numeric;