Add quarters to libcudf datetime (#8779)

Partly addresses #8676.
This PR adds `date_time:: extract_quarter`. The function returns an int `(1,2,3,4)` depending on which quarter of the year the date is in. Also fixed an incorrect date in previous testcase (#8711).

Authors:
  - https://github.com/shaneding

Approvers:
  - Nghia Truong (https://github.com/ttnghia)
  - Conor Hoekstra (https://github.com/codereport)

URL: #8779

cpp/include/cudf/datetime.hpp

@@ -206,6 +206,21 @@ std::unique_ptr<cudf::column> is_leap_year(
   cudf::column_view const& column,
   rmm::mr::device_memory_resource* mr = rmm::mr::get_current_device_resource());
 
+/**
+ * @brief  Returns the quarter of the date
+ *
+ * `output[i]` will be a value from {1, 2, 3, 4} corresponding to the quater of month given by
+ * `column[i]`. It will be null if the input row at `column[i]` is null.
+ *
+ * @throw cudf::logic_error if input column datatype is not a TIMESTAMP
+ *
+ * @param The input column containing datetime values
+ * @return A column of INT16 type indicating which quarter the date is in
+ */
+std::unique_ptr<cudf::column> extract_quarter(
+  cudf::column_view const& column,
+  rmm::mr::device_memory_resource* mr = rmm::mr::get_current_device_resource());
+
 /** @} */  // end of group
 }  // namespace datetime
 }  // namespace cudf

cpp/include/cudf/detail/datetime.hpp

@@ -135,6 +135,11 @@ std::unique_ptr<cudf::column> is_leap_year(
   rmm::cuda_stream_view stream        = rmm::cuda_stream_default,
   rmm::mr::device_memory_resource* mr = rmm::mr::get_current_device_resource());
 
+std::unique_ptr<cudf::column> extract_quarter(
+  cudf::column_view const& column,
+  rmm::cuda_stream_view stream        = rmm::cuda_stream_default,
+  rmm::mr::device_memory_resource* mr = rmm::mr::get_current_device_resource());
+
 }  // namespace detail
 }  // namespace datetime
 }  // namespace cudf

cpp/src/datetime/datetime_ops.cu

@@ -127,6 +127,23 @@ struct extract_day_num_of_year {
   }
 };
 
+// Extract the the quarter to which the timestamp belongs to
+struct extract_quarter_op {
+  template <typename Timestamp>
+  CUDA_DEVICE_CALLABLE int16_t operator()(Timestamp const ts) const
+  {
+    using namespace cuda::std::chrono;
+
+    // Only has the days - time component is chopped off, which is what we want
+    auto const days_since_epoch = floor<days>(ts);
+    auto const date             = year_month_day(days_since_epoch);
+    auto const month            = unsigned{date.month()};
+
+    // (x + y - 1) / y = ceil(x/y), where x and y are unsigned. x = month, y = 3
+    return (month + 2) / 3;
+  }
+};
+
 struct is_leap_year_op {
   template <typename Timestamp>
   CUDA_DEVICE_CALLABLE bool operator()(Timestamp const ts) const
@@ -376,6 +393,13 @@ std::unique_ptr<column> is_leap_year(column_view const& column,
   return apply_datetime_op<is_leap_year_op, type_id::BOOL8>(column, stream, mr);
 }
 
+std::unique_ptr<column> extract_quarter(column_view const& column,
+                                        rmm::cuda_stream_view stream,
+                                        rmm::mr::device_memory_resource* mr)
+{
+  return apply_datetime_op<extract_quarter_op, type_id::INT16>(column, stream, mr);
+}
+
 }  // namespace detail
 
 std::unique_ptr<column> extract_year(column_view const& column, rmm::mr::device_memory_resource* mr)
@@ -452,5 +476,12 @@ std::unique_ptr<column> is_leap_year(column_view const& column, rmm::mr::device_
   return detail::is_leap_year(column, rmm::cuda_stream_default, mr);
 }
 
+std::unique_ptr<column> extract_quarter(column_view const& column,
+                                        rmm::mr::device_memory_resource* mr)
+{
+  CUDF_FUNC_RANGE();
+  return detail::extract_quarter(column, rmm::cuda_stream_default, mr);
+}
+
 }  // namespace datetime
 }  // namespace cudf

cpp/tests/datetime/datetime_ops_test.cpp

@@ -17,6 +17,7 @@
 #include <cudf_test/base_fixture.hpp>
 #include <cudf_test/column_utilities.hpp>
 #include <cudf_test/column_wrapper.hpp>
+#include <cudf_test/iterator_utilities.hpp>
 #include <cudf_test/timestamp_utilities.cuh>
 #include <cudf_test/type_lists.hpp>
 
@@ -552,7 +553,7 @@ TEST_F(BasicDatetimeOpsTest, TestIsLeapYear)
         915148800L,     // 1999-01-01 00:00:00 GMT - non leap year
         -11663029161L,  // 1600-5-31 05:40:39 GMT - leap year
         707904541L,     // 1992-06-07 08:09:01 GMT - leap year
-        2181048447L,    // 1900-11-20 09:12:33 GMT - non leap year
+        -2181005247L,   // 1900-11-20 09:12:33 GMT - non leap year
         0L,             // UNIX EPOCH 1970-01-01 00:00:00 GMT - non leap year
         -12212553600L,  // First full year of Gregorian Calandar 1583-01-01 00:00:00 - non-leap-year
         0L,             // null
@@ -569,4 +570,39 @@ TEST_F(BasicDatetimeOpsTest, TestIsLeapYear)
       {true, false, true, true, true, true, true, true, false, true, true, false}});
 }
 
+TEST_F(BasicDatetimeOpsTest, TestQuarter)
+{
+  using namespace cudf::test;
+  using namespace cudf::datetime;
+  using namespace cuda::std::chrono;
+  using namespace cudf::test::iterators;
+
+  // Time in seconds since epoch
+  // Dates converted using epochconverter.com
+  auto timestamps_s =
+    cudf::test::fixed_width_column_wrapper<cudf::timestamp_s, cudf::timestamp_s::rep>{
+      {
+        1594332839L,    // 2020-07-09 10:13:59 GMT
+        0L,             // null
+        915148800L,     // 1999-01-01 00:00:00 GMT
+        -11663029161L,  // 1600-5-31 05:40:39 GMT
+        707904541L,     // 1992-06-07 08:09:01 GMT
+        -2181005247L,   // 1900-11-20 09:12:33 GMT
+        0L,             // UNIX EPOCH 1970-01-01 00:00:00 GMT
+        -12212553600L,  // First full year of Gregorian Calandar 1583-01-01 00:00:00
+        0L,             // null
+        13591632822L,   // 2400-09-13 13:33:42 GMT
+        4539564243L,    // 2113-11-08 06:04:03 GMT
+        0L,             // null
+        1608581568L,    // 2020-12-21 08:12:48 GMT
+        1584821568L,    // 2020-03-21 08:12:48 GMT
+      },
+      nulls_at({1, 8, 11})};
+
+  auto quarter = cudf::test::fixed_width_column_wrapper<int16_t>{
+    {3, 0 /*null*/, 1, 2, 2, 4, 1, 1, 0 /*null*/, 3, 4, 0 /*null*/, 4, 1}, nulls_at({1, 8, 11})};
+
+  CUDF_TEST_EXPECT_COLUMNS_EQUAL(*extract_quarter(timestamps_s), quarter);
+}
+
 CUDF_TEST_PROGRAM_MAIN()