Add JNI backend for Spark SQL function conv for (hexa)decimals

src/main/cpp/src/CastStringJni.cpp

@@ -15,6 +15,13 @@
  */
 
 #include "cast_string.hpp"
+#include <cudf/column/column_factories.hpp>
+#include <cudf/replace.hpp>
+#include <cudf/scalar/scalar_factories.hpp>
+#include <cudf/strings/extract.hpp>
+#include <cudf/strings/convert/convert_integers.hpp>
+#include <cudf/strings/strings_column_view.hpp>
+#include <cudf/strings/regex/regex_program.hpp>
 
 #include "cudf_jni_apis.hpp"
 #include "dtype_utils.hpp"
@@ -111,4 +118,72 @@ JNIEXPORT jlong JNICALL Java_com_nvidia_spark_rapids_jni_CastStrings_fromDecimal
   }
   CATCH_CAST_EXCEPTION(env, 0);
 }
+
+JNIEXPORT jlong JNICALL Java_com_nvidia_spark_rapids_jni_CastStrings_toIntegersWithBase(
+  JNIEnv* env, jclass, jlong input_column, jint base)
+{
+  JNI_NULL_CHECK(env, input_column, "input column is null", 0);
+  using namespace cudf;
+  try {
+    jni::auto_set_device(env);
+    auto const res_data_type = data_type(type_id::UINT64);
+
+    auto const input_view{*reinterpret_cast<column_view const*>(input_column)};
+    const strings_column_view input_strings{input_view};
+    auto integer_view = [&] {
+      switch (base) {
+        case 10: {
+          // TODO implement it in the kernel
+          auto const regex = strings::regex_program::create(R"(^\s*([0-9]+).*)");
+          auto const dec_str_table = strings::extract(input_strings, *regex);
+          const strings_column_view dec_str_view{dec_str_table->get_column(0)};
+          return strings::to_integers(dec_str_view, res_data_type);
+        } break;
+        case 16: {
+          auto const regex = strings::regex_program::create(R"(^\s*([0-9a-fA-F]+).*)");
+          auto const hex_str_table = strings::extract(input_strings, *regex);
+          const strings_column_view hex_str_view{hex_str_table->get_column(0)};
+          return strings::hex_to_integers(hex_str_view, res_data_type);
+        }
+        default: {
+          auto const error_msg = "Bases supported 10, 16; Actual: " + std::to_string(base);
+          throw spark_rapids_jni::cast_error(0, error_msg);
+        }
+      }
+    }();
+
+    return jni::release_as_jlong(integer_view);
+  }
+  CATCH_CAST_EXCEPTION(env, 0);
+}
+
+JNIEXPORT jlong JNICALL Java_com_nvidia_spark_rapids_jni_CastStrings_fromIntegersWithBase(
+  JNIEnv* env, jclass, jlong input_column, jint base)
+{
+  JNI_NULL_CHECK(env, input_column, "input column is null", 0);
+  using namespace cudf;
+  try {
+    jni::auto_set_device(env);
+    auto input_view{*reinterpret_cast<column_view const*>(input_column)};
+    auto result = [&] {
+      switch (base) {
+        case 10: {
+          return strings::from_integers(input_view);
+        } break;
+        case 16: {
+          auto pre_res = strings::integers_to_hex(input_view);
+          auto const regex = strings::regex_program::create("^0?([0-9a-fA-F]+)$");
+          auto const wo_leading_zeros = strings::extract(strings_column_view(*pre_res), *regex);
+          return std::move(wo_leading_zeros->release()[0]);
+        }
+        default: {
+          auto const error_msg = "Bases supported 10, 16; Actual: " + std::to_string(base);
+          throw spark_rapids_jni::cast_error(0, error_msg);
+        }
+      }
+    }();
+    return jni::release_as_jlong(result);
+  }
+  CATCH_CAST_EXCEPTION(env, 0);
+}
 }

src/main/cpp/src/row_conversion.cu

@@ -181,8 +181,8 @@ struct tile_info {
  *
  */
 struct row_batch {
-  size_type num_bytes;                      // number of bytes in this batch
-  size_type row_count;                      // number of rows in the batch
+  size_type num_bytes;                    // number of bytes in this batch
+  size_type row_count;                    // number of rows in the batch
   device_uvector<size_type> row_offsets;  // offsets column of output cudf column
 };
 

src/main/java/com/nvidia/spark/rapids/jni/CastStrings.java

@@ -82,7 +82,7 @@ public static ColumnVector toDecimal(ColumnView cv, boolean ansiMode, boolean st
 
   /**
    * Convert a decimal column to a string column.
-   * 
+   *
    * @param cv the column data to process
    * @return the converted column
    */
@@ -102,10 +102,21 @@ public static ColumnVector toFloat(ColumnView cv, boolean ansiMode, DType type)
     return new ColumnVector(toFloat(cv.getNativeView(), ansiMode, type.getTypeId().getNativeId()));
   }
 
+
+  public static ColumnVector toIntegersWithBase(ColumnView cv, int base) {
+    return new ColumnVector(toIntegersWithBase(cv.getNativeView(), base));
+  }
+
+  public static ColumnVector fromIntegersWithBase(ColumnView cv, int base) {
+    return new ColumnVector(fromIntegersWithBase(cv.getNativeView(), base));
+  }
+
   private static native long toInteger(long nativeColumnView, boolean ansi_enabled, boolean strip,
       int dtype);
   private static native long toDecimal(long nativeColumnView, boolean ansi_enabled, boolean strip,
       int precision, int scale);
   private static native long toFloat(long nativeColumnView, boolean ansi_enabled, int dtype);
   private static native long fromDecimal(long nativeColumnView);
+  private static native long toIntegersWithBase(long nativeColumnView, int base);
+  private static native long fromIntegersWithBase(long nativeColumnView, int base);
 }