From aa3ab062fa9eb0d557afcb84e615ffbcd6867c26 Mon Sep 17 00:00:00 2001 From: liyuan <84758614+nvliyuan@users.noreply.github.com> Date: Tue, 26 Sep 2023 09:06:43 +0800 Subject: [PATCH] Update examples/UDF-Examples/RAPIDS-accelerated-UDFs/README.md Co-authored-by: Jason Lowe --- examples/UDF-Examples/RAPIDS-accelerated-UDFs/README.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/examples/UDF-Examples/RAPIDS-accelerated-UDFs/README.md b/examples/UDF-Examples/RAPIDS-accelerated-UDFs/README.md index 7afa7edee..7e7bf7266 100644 --- a/examples/UDF-Examples/RAPIDS-accelerated-UDFs/README.md +++ b/examples/UDF-Examples/RAPIDS-accelerated-UDFs/README.md @@ -30,7 +30,7 @@ If there is no existing simple Java API we could leverage, we can write native c The Java class for the UDF is similar as the previous URLDecode/URLEncode demo, we need to implement a cosineSimilarity function in C++ code and goes into the native code as quickly as possible, because it is easier to write the code safely. In the native code, it `reinterpret_cast` the input to columnview, do some sanity checking and convert to list -columnviews, then compute the cosine similarity, finally return the unique pointer to a column, release the underlying resources. +column views, then compute the cosine similarity, finally return the unique pointer to a column, release the underlying resources. On Java side we are going to wrap it in a columnvector and own that resource. In `cosine_similarity.cu` we implement the computation as the actual CUDA kernel. In CUDA kernel part, we can leverage thrust template library to write the standard algorithms for GPU parallelizing code.