abstractdog commented on a change in pull request #1280:
URL: https://github.com/apache/hive/pull/1280#discussion_r469112026
##########
File path:
ql/src/java/org/apache/hadoop/hive/ql/exec/vector/expressions/aggregates/VectorUDAFBloomFilterMerge.java
##########
@@ -77,6 +75,211 @@ public void reset() {
// Do not change the initial bytes which contain
NumHashFunctions/NumBits!
Arrays.fill(bfBytes, BloomKFilter.START_OF_SERIALIZED_LONGS,
bfBytes.length, (byte) 0);
}
+
+ public boolean mergeBloomFilterBytesFromInputColumn(BytesColumnVector
inputColumn,
+ int batchSize, boolean selectedInUse, int[] selected, Configuration
conf) {
+ // already set in previous iterations, no need to call initExecutor again
+ if (numThreads == 0) {
+ return false;
+ }
+ if (executor == null) {
+ initExecutor(conf, batchSize);
+ if (!isParallel) {
+ return false;
+ }
+ }
+
+ // split every bloom filter (represented by a part of a byte[]) across
workers
+ for (int j = 0; j < batchSize; j++) {
+ if (!selectedInUse && inputColumn.noNulls) {
+ splitVectorAcrossWorkers(workers, inputColumn.vector[j],
inputColumn.start[j],
+ inputColumn.length[j]);
+ } else if (!selectedInUse) {
+ if (!inputColumn.isNull[j]) {
+ splitVectorAcrossWorkers(workers, inputColumn.vector[j],
inputColumn.start[j],
+ inputColumn.length[j]);
+ }
+ } else if (inputColumn.noNulls) {
+ int i = selected[j];
+ splitVectorAcrossWorkers(workers, inputColumn.vector[i],
inputColumn.start[i],
+ inputColumn.length[i]);
+ } else {
+ int i = selected[j];
+ if (!inputColumn.isNull[i]) {
+ splitVectorAcrossWorkers(workers, inputColumn.vector[i],
inputColumn.start[i],
+ inputColumn.length[i]);
+ }
+ }
+ }
+
+ return true;
+ }
+
+ private void initExecutor(Configuration conf, int batchSize) {
+ numThreads =
conf.getInt(HiveConf.ConfVars.TEZ_BLOOM_FILTER_MERGE_THREADS.varname,
+ HiveConf.ConfVars.TEZ_BLOOM_FILTER_MERGE_THREADS.defaultIntVal);
+ LOG.info("Number of threads used for bloom filter merge: {}",
numThreads);
+
+ if (numThreads < 0) {
+ throw new RuntimeException(
+ "invalid number of threads for bloom filter merge: " + numThreads);
+ }
+ if (numThreads == 0) { // disable parallel feature
+ return; // this will leave isParallel=false
+ }
+ isParallel = true;
+ executor = Executors.newFixedThreadPool(numThreads);
+
+ workers = new BloomFilterMergeWorker[numThreads];
+ for (int f = 0; f < numThreads; f++) {
Review comment:
cood catch, moving them to a single loop
##########
File path:
ql/src/java/org/apache/hadoop/hive/ql/exec/vector/expressions/aggregates/VectorUDAFBloomFilterMerge.java
##########
@@ -77,6 +75,211 @@ public void reset() {
// Do not change the initial bytes which contain
NumHashFunctions/NumBits!
Arrays.fill(bfBytes, BloomKFilter.START_OF_SERIALIZED_LONGS,
bfBytes.length, (byte) 0);
}
+
+ public boolean mergeBloomFilterBytesFromInputColumn(BytesColumnVector
inputColumn,
+ int batchSize, boolean selectedInUse, int[] selected, Configuration
conf) {
+ // already set in previous iterations, no need to call initExecutor again
+ if (numThreads == 0) {
+ return false;
+ }
+ if (executor == null) {
+ initExecutor(conf, batchSize);
+ if (!isParallel) {
+ return false;
+ }
+ }
+
+ // split every bloom filter (represented by a part of a byte[]) across
workers
+ for (int j = 0; j < batchSize; j++) {
+ if (!selectedInUse && inputColumn.noNulls) {
+ splitVectorAcrossWorkers(workers, inputColumn.vector[j],
inputColumn.start[j],
+ inputColumn.length[j]);
+ } else if (!selectedInUse) {
+ if (!inputColumn.isNull[j]) {
+ splitVectorAcrossWorkers(workers, inputColumn.vector[j],
inputColumn.start[j],
+ inputColumn.length[j]);
+ }
+ } else if (inputColumn.noNulls) {
+ int i = selected[j];
+ splitVectorAcrossWorkers(workers, inputColumn.vector[i],
inputColumn.start[i],
+ inputColumn.length[i]);
+ } else {
+ int i = selected[j];
+ if (!inputColumn.isNull[i]) {
+ splitVectorAcrossWorkers(workers, inputColumn.vector[i],
inputColumn.start[i],
+ inputColumn.length[i]);
+ }
+ }
+ }
+
+ return true;
+ }
+
+ private void initExecutor(Configuration conf, int batchSize) {
+ numThreads =
conf.getInt(HiveConf.ConfVars.TEZ_BLOOM_FILTER_MERGE_THREADS.varname,
+ HiveConf.ConfVars.TEZ_BLOOM_FILTER_MERGE_THREADS.defaultIntVal);
+ LOG.info("Number of threads used for bloom filter merge: {}",
numThreads);
+
+ if (numThreads < 0) {
+ throw new RuntimeException(
+ "invalid number of threads for bloom filter merge: " + numThreads);
+ }
+ if (numThreads == 0) { // disable parallel feature
+ return; // this will leave isParallel=false
+ }
+ isParallel = true;
+ executor = Executors.newFixedThreadPool(numThreads);
+
+ workers = new BloomFilterMergeWorker[numThreads];
+ for (int f = 0; f < numThreads; f++) {
Review comment:
good catch, moving them to a single loop
----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
