This is an automated email from the ASF dual-hosted git repository.
baunsgaard pushed a commit to branch main
in repository https://gitbox.apache.org/repos/asf/systemds.git
The following commit(s) were added to refs/heads/main by this push:
new bbdbb1a781 [MINOR] Append perf
bbdbb1a781 is described below
commit bbdbb1a7814d58da15cdd0bd35a08811a635d869
Author: Sebastian Baunsgaard <[email protected]>
AuthorDate: Thu Apr 4 18:08:27 2024 +0200
[MINOR] Append perf
This commit adds a perf script for MCSR appending.
It is manly ment as an example of how to execute a perf script.
````
Me:~/github/systemds$ java -jar target/systemds-3.3.0-SNAPSHOT-perf.jar
1004 1000 100000
Appending rep: 1000 of 100000 distinct append calls (including random and
allocations)
Append all dense: 4.262+- 0.164 ms
Append all zero on empty: 0.198+- 0.004 ms
Append all zero on Scalar: 0.197+- 0.004 ms
Append all zero on Array: 0.203+- 0.013 ms
Append half zero on Array: 4.422+- 0.170 ms
```
Closes #2011
---
.../apache/sysds/runtime/data/SparseBlockMCSR.java | 5 +-
.../java/org/apache/sysds/performance/Main.java | 8 ++
.../java/org/apache/sysds/performance/README.md | 10 +--
.../org/apache/sysds/performance/TimingUtils.java | 14 ++++
.../performance/matrix/MatrixMulPerformance.java | 4 +-
.../sysds/performance/matrix/SparseAppend.java | 89 ++++++++++++++++++++++
6 files changed, 121 insertions(+), 9 deletions(-)
diff --git a/src/main/java/org/apache/sysds/runtime/data/SparseBlockMCSR.java
b/src/main/java/org/apache/sysds/runtime/data/SparseBlockMCSR.java
index 025da10394..52b5d2e338 100644
--- a/src/main/java/org/apache/sysds/runtime/data/SparseBlockMCSR.java
+++ b/src/main/java/org/apache/sysds/runtime/data/SparseBlockMCSR.java
@@ -378,12 +378,13 @@ public class SparseBlockMCSR extends SparseBlock
@Override
public final void append(final int r, final int c, final double v) {
+ // Perf verified in java -jar
target/systemds-3.3.0-SNAPSHOT-perf.jar 1004 1000 100000
if(v == 0)
return;
else if(_rows[r] == null)
_rows[r] = new SparseRowScalar(c, v);
- else
- _rows[r] = _rows[r].append(c, v);
+ else
+ _rows[r] = _rows[r].append(c, v);
}
@Override
diff --git a/src/test/java/org/apache/sysds/performance/Main.java
b/src/test/java/org/apache/sysds/performance/Main.java
index 2fed0d7144..9959192188 100644
--- a/src/test/java/org/apache/sysds/performance/Main.java
+++ b/src/test/java/org/apache/sysds/performance/Main.java
@@ -32,6 +32,7 @@ import org.apache.sysds.performance.generators.IGenerate;
import org.apache.sysds.performance.generators.MatrixFile;
import org.apache.sysds.performance.matrix.MatrixMulPerformance;
import org.apache.sysds.performance.matrix.MatrixStorage;
+import org.apache.sysds.performance.matrix.SparseAppend;
import org.apache.sysds.runtime.data.SparseBlock;
import org.apache.sysds.runtime.frame.data.FrameBlock;
import org.apache.sysds.runtime.matrix.data.MatrixBlock;
@@ -115,6 +116,9 @@ public class Main {
case 1003:
run1003(args);
break;
+ case 1004:
+ run1004(args);
+ break;
default:
break;
}
@@ -319,6 +323,10 @@ public class Main {
ms.testBalancedDims(SparseBlock.Type.DCSR, sparsity,
numEntries, resolution, maxRowColRatio, repetitions);
}
+ private static void run1004(String[] args){
+ new SparseAppend(args);
+ }
+
public static void main(String[] args) {
try {
exec(Integer.parseInt(args[0]), args);
diff --git a/src/test/java/org/apache/sysds/performance/README.md
b/src/test/java/org/apache/sysds/performance/README.md
index 7e7edbb805..7129757f34 100644
--- a/src/test/java/org/apache/sysds/performance/README.md
+++ b/src/test/java/org/apache/sysds/performance/README.md
@@ -28,7 +28,7 @@ mvn package
Example of running it:
```bash
-java -jar target/systemds-3.2.0-SNAPSHOT-perf.jar 1
+java -jar target/systemds-3.3.0-SNAPSHOT-perf.jar 1
```
example result of the above job:
@@ -49,24 +49,24 @@ Running Steam Compression Test
With profiler:
```bash
-java -jar
-agentpath:$HOME/Programs/profiler/lib/libasyncProfiler.so=start,event=cpu,file=temp/log.html
target/systemds-3.2.0-SNAPSHOT-perf.jar 12 10000 100 4 1.0 16 1000 -1
+java -jar
-agentpath:$HOME/Programs/profiler/lib/libasyncProfiler.so=start,event=cpu,file=temp/log.html
target/systemds-3.3.0-SNAPSHOT-perf.jar 12 10000 100 4 1.0 16 1000 -1
```
Take a Matrix and perform serialization
```bash
-java -jar
-agentpath:$HOME/Programs/profiler/lib/libasyncProfiler.so=start,event=cpu,file=temp/log.html
target/systemds-3.2.0-SNAPSHOT-perf.jar 13 16 100 "temp/test.csv" -1
+java -jar
-agentpath:$HOME/Programs/profiler/lib/libasyncProfiler.so=start,event=cpu,file=temp/log.html
target/systemds-3.3.0-SNAPSHOT-perf.jar 13 16 100 "temp/test.csv" -1
```
Take a Frame and transform into a Matrix and perform serialization.
```bash
-java -jar
-agentpath:$HOME/Programs/profiler/lib/libasyncProfiler.so=start,event=cpu,file=temp/log.html
target/systemds-3.2.0-SNAPSHOT-perf.jar 14 16 1000
"src/test/resources/datasets/titanic/titanic.csv"
"src/test/resources/datasets/titanic/tfspec.json" -1
+java -jar
-agentpath:$HOME/Programs/profiler/lib/libasyncProfiler.so=start,event=cpu,file=temp/log.html
target/systemds-3.3.0-SNAPSHOT-perf.jar 14 16 1000
"src/test/resources/datasets/titanic/titanic.csv"
"src/test/resources/datasets/titanic/tfspec.json" -1
```
Frame Operation timings
```bash
-java -jar
-agentpath:$HOME/Programs/profiler/lib/libasyncProfiler.so=start,event=cpu,file=temp/log.html
target/systemds-3.2.0-SNAPSHOT-perf.jar 15 16 10
"src/test/resources/datasets/titanic/titanic.csv"
"src/test/resources/datasets/titanic/tfspec.json"
+java -jar
-agentpath:$HOME/Programs/profiler/lib/libasyncProfiler.so=start,event=cpu,file=temp/log.html
target/systemds-3.3.0-SNAPSHOT-perf.jar 15 16 10
"src/test/resources/datasets/titanic/titanic.csv"
"src/test/resources/datasets/titanic/tfspec.json"
```
diff --git a/src/test/java/org/apache/sysds/performance/TimingUtils.java
b/src/test/java/org/apache/sysds/performance/TimingUtils.java
index 0faf01c9b0..7d055d9afc 100644
--- a/src/test/java/org/apache/sysds/performance/TimingUtils.java
+++ b/src/test/java/org/apache/sysds/performance/TimingUtils.java
@@ -63,6 +63,20 @@ public interface TimingUtils {
// System.out.println(time.stop());
// }
+/**
+ * Time the function given assuming that it should put result into the
given time array at index i.
+ *
+ * @param f The function to time
+ * @param rep the number of repetitions
+ */
+ public static double[] time(F f, int rep) {
+ double[] times = new double[rep];
+ for(int i = 0; i < rep; i ++)
+ time(f, times, i);
+
+ return times;
+ }
+
/**
* Time the function given assuming that it should put result into the
given time array at index i.
*
diff --git
a/src/test/java/org/apache/sysds/performance/matrix/MatrixMulPerformance.java
b/src/test/java/org/apache/sysds/performance/matrix/MatrixMulPerformance.java
index 05b91d0ebe..fea49cfb72 100644
---
a/src/test/java/org/apache/sysds/performance/matrix/MatrixMulPerformance.java
+++
b/src/test/java/org/apache/sysds/performance/matrix/MatrixMulPerformance.java
@@ -19,6 +19,8 @@
package org.apache.sysds.performance.matrix;
+import java.util.Arrays;
+
import org.apache.sysds.runtime.data.SparseBlock;
import org.apache.sysds.runtime.data.SparseBlockCOO;
import org.apache.sysds.runtime.data.SparseBlockCSR;
@@ -29,8 +31,6 @@ import org.apache.sysds.runtime.matrix.data.MatrixBlock;
import org.apache.sysds.runtime.util.DataConverter;
import org.apache.sysds.test.TestUtils;
-import java.util.Arrays;
-
public class MatrixMulPerformance {
private final int _rl;
diff --git
a/src/test/java/org/apache/sysds/performance/matrix/SparseAppend.java
b/src/test/java/org/apache/sysds/performance/matrix/SparseAppend.java
new file mode 100644
index 0000000000..7930fd3275
--- /dev/null
+++ b/src/test/java/org/apache/sysds/performance/matrix/SparseAppend.java
@@ -0,0 +1,89 @@
+package org.apache.sysds.performance.matrix;
+
+import java.util.Random;
+
+import org.apache.sysds.performance.TimingUtils;
+import org.apache.sysds.runtime.data.SparseBlockMCSR;
+
+public class SparseAppend {
+
+ public SparseAppend(String[] args) {
+ // Ignore args 1 it is containing the Id to run this test.
+
+ final int rep = Integer.parseInt(args[1]);
+ final int size = Integer.parseInt(args[2]);
+ final Random r = new Random(42);
+
+ System.out.println("Appending rep: " + rep + " of " + size + "
distinct append calls (including random and allocations)");
+ double[] times;
+ // Warmup
+ times = TimingUtils.time(() -> {
+ SparseBlockMCSR sb = new SparseBlockMCSR(100);
+ for(int i = 0; i < size; i++) {
+ sb.append(i % 100, i, r.nextDouble());
+ }
+ }, rep);
+
+ times = TimingUtils.time(() -> {
+ SparseBlockMCSR sb = new SparseBlockMCSR(100);
+ for(int i = 0; i < size; i++) {
+ sb.append(i % 100, i, r.nextDouble());
+ }
+ }, rep);
+
+ System.out.println("Append all dense: " +
TimingUtils.stats(times));
+
+ times = TimingUtils.time(() -> {
+ SparseBlockMCSR sb = new SparseBlockMCSR(100);
+ for(int i = 0; i < size; i++) {
+ sb.append(i % 100, i, 0);
+ }
+ }, rep);
+
+ System.out.println("Append all zero on empty: " +
TimingUtils.stats(times));
+
+ times = TimingUtils.time(() -> {
+ SparseBlockMCSR sb = new SparseBlockMCSR(100);
+ for(int i = 0; i < 100; i ++){
+ sb.append(i, i, 1);
+ }
+ for(int i = 0; i < size; i++) {
+ sb.append(i % 100, i, 0);
+ }
+ }, rep);
+
+ System.out.println("Append all zero on Scalar: " +
TimingUtils.stats(times));
+
+ times = TimingUtils.time(() -> {
+ SparseBlockMCSR sb = new SparseBlockMCSR(100);
+ for(int i = 0; i < 100; i ++){
+ sb.append(i, i, 1);
+ sb.append(i, i, 1);
+ }
+ for(int i = 0; i < size; i++) {
+ sb.append(i % 100, i, 0);
+ }
+ }, rep);
+
+ System.out.println("Append all zero on Array: " +
TimingUtils.stats(times));
+
+ times = TimingUtils.time(() -> {
+ SparseBlockMCSR sb = new SparseBlockMCSR(100);
+ for(int i = 0; i < 100; i ++){
+ sb.append(i, i, 1);
+ sb.append(i, i, 1);
+ }
+ for(int i = 0; i < size; i++) {
+ double d = r.nextDouble();
+ d = d > 0.5 ? d : 0;
+ sb.append(i % 100, i, d);
+ }
+ }, rep);
+
+ System.out.println("Append half zero on Array: " +
TimingUtils.stats(times));
+ }
+
+ public static void main(String[] args) {
+ new SparseAppend(new String[] {"1004", "10000", "10000"});
+ }
+}
