Add samples to measure vector ldn/stn performance

Add the test cases as follows:
1. 2D vector add and mul for int/short
2. 3D vector add and mul for int/short
3. 4D vector add and mul for int/short
4. RGB To Cmyk for ld3/st4

Change-Id: I68079a8bb2f50bf0d07c3a382a1d7208b95ff9f7

4 files changed, 362 insertions, 0 deletions

diff --git a/src/main/java/org/linaro/benchmarks/TestRGBToCmyk.java b/src/main/java/org/linaro/benchmarks/TestRGBToCmyk.java
new file mode 100644
index 0000000..b9bc897
--- /dev/null
+++ b/src/main/java/org/linaro/benchmarks/TestRGBToCmyk.java
@@ -0,0 +1,53 @@
+package org.linaro.benchmarks;
+
+import org.openjdk.jmh.annotations.*;
+import java.util.concurrent.TimeUnit;
+
+@BenchmarkMode(Mode.AverageTime)
+@OutputTimeUnit(TimeUnit.MICROSECONDS)
+@State(Scope.Benchmark)
+
+public class TestRGBToCmyk
+{
+  static final int LENGTH = 256 * 1024;
+  static byte input[] = new byte[LENGTH*3];
+  static byte output[] = new byte[LENGTH*4];
+
+  public static void TestRGBToCmykInit()
+  {
+    for (int i = 0; i < LENGTH*3; i++) {
+      input[i] = (byte)i;
+    }
+  }
+
+  public static void RGBToCmyk(byte[] rgb, byte[] cmyk, int cnt)
+  {
+    int i;
+    int c, m, y, k;
+    for (i = 0; i < cnt; i++) {
+      /* calculate complementary colors */
+      c = 255 - rgb[i*3];
+      m = 255 - rgb[i*3+1];
+      y = 255 - rgb[i*3+2];
+      /* find the black level k */
+      k = Math.min(Math.min(c, m), y);
+      /* correct complementary color lever based on k */
+      cmyk[i*4] = (byte)(c - k);
+      cmyk[i*4+1] = (byte)(m - k);
+      cmyk[i*4+2] = (byte)(y - k);
+      cmyk[i*4+3] = (byte)(k);
+    }
+  }
+
+  @Setup
+  public void setup()
+  {
+    TestRGBToCmykInit();
+  }
+
+  @Benchmark
+  public void testRGBToCmyk() {
+    RGBToCmyk(input, output, LENGTH);
+  }
+
+}
diff --git a/src/main/java/org/linaro/benchmarks/TestSIMDLd2St2.java b/src/main/java/org/linaro/benchmarks/TestSIMDLd2St2.java
new file mode 100644
index 0000000..68901ab
--- /dev/null
+++ b/src/main/java/org/linaro/benchmarks/TestSIMDLd2St2.java
@@ -0,0 +1,95 @@
+package org.linaro.benchmarks;
+
+import org.openjdk.jmh.annotations.*;
+import java.util.concurrent.TimeUnit;
+
+@BenchmarkMode(Mode.AverageTime)
+@OutputTimeUnit(TimeUnit.MICROSECONDS)
+@State(Scope.Benchmark)
+
+public class TestSIMDLd2St2 {
+  static final int VECT_LENGTH = 256 * 1024;
+  static final int DIM = 2;
+  static final int LENGTH = VECT_LENGTH * DIM;
+  static int [] in = new int[LENGTH];
+  static int [] out = new int[LENGTH];
+  static int [] c = new int[DIM];
+  static short [] sin = new short[LENGTH];
+  static short [] sout = new short[LENGTH];
+  static short [] sc = new short[DIM];
+
+  public static void vect2DInit() {
+    int i;
+    for (i = 0; i < LENGTH; i++) {
+      in[i] = i + 3;
+      sin[i] = (short)(i + 3);
+    }
+    for (i = 0; i < DIM; i++) {
+      c[i] = i + 1;
+      sc[i] = (short)(i + 1);
+    }
+  }
+
+  public static void vect2DAddInt(
+          int[] out,
+          int[] in,
+          int[] c) {
+    int c0 = c[0];
+    int c1 = c[1];
+    for (int i = 0; i < VECT_LENGTH; i++) {
+      out[i*DIM] = in[i*DIM] + c0;
+      out[i*DIM + 1] = in[i*DIM + 1] + c1;
+    }
+  }
+
+  public static void vect2DMulInt(
+          int[] out,
+          int[] in,
+          int[] c) {
+    int c0 = c[0];
+    int c1 = c[1];
+    for (int i = 0; i < VECT_LENGTH; i++) {
+      out[i*DIM] = in[i*DIM] * c0;
+      out[i*DIM + 1] = in[i*DIM + 1] * c1;
+    }
+  }
+
+  public static void vect2DAddShort(
+          short[] out,
+          short[] in,
+          short[] c) {
+    short c0 = c[0];
+    short c1 = c[1];
+    for (int i = 0; i < VECT_LENGTH; i++) {
+      out[i*DIM] = (short)(in[i*DIM] + c0);
+      out[i*DIM + 1] = (short)(in[i*DIM + 1] + c1);
+    }
+  }
+
+  public static void vect2DMulShort(
+          short[] out,
+          short[] in,
+          short[] c) {
+    short c0 = c[0];
+    short c1 = c[1];
+    for (int i = 0; i < VECT_LENGTH; i++) {
+      out[i*DIM] = (short)(in[i*DIM] * c0);
+      out[i*DIM + 1] = (short)(in[i*DIM + 1] * c1);
+    }
+  }
+
+  @Setup
+  public void setup()
+  {
+    vect2DInit();
+  }
+
+  @Benchmark
+  public void testVect2D() {
+    vect2DAddInt(out, in, c);
+    vect2DMulInt(out, in, c);
+    vect2DAddShort(sout, sin, sc);
+    vect2DMulShort(sout, sin, sc);
+  }
+
+}
diff --git a/src/main/java/org/linaro/benchmarks/TestSIMDLd3St3.java b/src/main/java/org/linaro/benchmarks/TestSIMDLd3St3.java
new file mode 100644
index 0000000..a691957
--- /dev/null
+++ b/src/main/java/org/linaro/benchmarks/TestSIMDLd3St3.java
@@ -0,0 +1,103 @@
+package org.linaro.benchmarks;
+
+import org.openjdk.jmh.annotations.*;
+import java.util.concurrent.TimeUnit;
+
+@BenchmarkMode(Mode.AverageTime)
+@OutputTimeUnit(TimeUnit.MICROSECONDS)
+@State(Scope.Benchmark)
+
+public class TestSIMDLd3St3 {
+  static final int VECT_LENGTH = 256 * 1024;
+  static final int DIM = 3;
+  static final int LENGTH = VECT_LENGTH * DIM;
+  static int [] in = new int[LENGTH];
+  static int [] out = new int[LENGTH];
+  static int [] c = new int[DIM];
+  static short [] sin = new short[LENGTH];
+  static short [] sout = new short[LENGTH];
+  static short [] sc = new short[DIM];
+
+  public static void vect3DInit() {
+    int i;
+    for (i = 0; i < LENGTH; i++) {
+      in[i] = i + 3;
+      sin[i] = (short)(i + 3);
+    }
+    for (i = 0; i < DIM; i++) {
+      c[i] = i + 1;
+      sc[i] = (short)(i + 1);
+    }
+  }
+
+  public static void vect3DAddInt(
+          int[] out,
+          int[] in,
+          int[] c) {
+    int c0 = c[0];
+    int c1 = c[1];
+    int c2 = c[2];
+    for (int i = 0; i < VECT_LENGTH; i++) {
+      out[i*DIM] = in[i*DIM] + c0;
+      out[i*DIM + 1] = in[i*DIM + 1] + c1;
+      out[i*DIM + 2] = in[i*DIM + 2] + c2;
+    }
+  }
+
+  public static void vect3DMulInt(
+          int[] out,
+          int[] in,
+          int[] c) {
+    int c0 = c[0];
+    int c1 = c[1];
+    int c2 = c[2];
+    for (int i = 0; i < VECT_LENGTH; i++) {
+      out[i*DIM] = in[i*DIM] * c0;
+      out[i*DIM + 1] = in[i*DIM + 1] * c1;
+      out[i*DIM + 2] = in[i*DIM + 2] * c2;
+    }
+  }
+
+  public static void vect3DAddShort(
+          short[] out,
+          short[] in,
+          short[] c) {
+    short c0 = c[0];
+    short c1 = c[1];
+    short c2 = c[2];
+    for (int i = 0; i < VECT_LENGTH; i++) {
+      out[i*DIM] = (short)(in[i*DIM] + c0);
+      out[i*DIM + 1] = (short)(in[i*DIM + 1] + c1);
+      out[i*DIM + 2] = (short)(in[i*DIM + 2] + c2);
+    }
+  }
+
+  public static void vect3DMulShort(
+          short[] out,
+          short[] in,
+          short[] c) {
+    short c0 = c[0];
+    short c1 = c[1];
+    short c2 = c[2];
+    for (int i = 0; i < VECT_LENGTH; i++) {
+      out[i*DIM] = (short)(in[i*DIM] * c0);
+      out[i*DIM + 1] = (short)(in[i*DIM + 1] * c1);
+      out[i*DIM + 2] = (short)(in[i*DIM + 2] * c2);
+    }
+  }
+
+  @Setup
+  public void setup()
+  {
+    vect3DInit();
+  }
+
+  @Benchmark
+  public void testVect3D() {
+    vect3DAddInt(out, in, c);
+    vect3DMulInt(out, in, c);
+    vect3DAddShort(sout, sin, sc);
+    vect3DMulShort(sout, sin, sc);
+  }
+
+}
diff --git a/src/main/java/org/linaro/benchmarks/TestSIMDLd4St4.java b/src/main/java/org/linaro/benchmarks/TestSIMDLd4St4.java
new file mode 100644
index 0000000..fea4808
--- /dev/null
+++ b/src/main/java/org/linaro/benchmarks/TestSIMDLd4St4.java
@@ -0,0 +1,111 @@
+package org.linaro.benchmarks;
+
+import org.openjdk.jmh.annotations.*;
+import java.util.concurrent.TimeUnit;
+
+@BenchmarkMode(Mode.AverageTime)
+@OutputTimeUnit(TimeUnit.MICROSECONDS)
+@State(Scope.Benchmark)
+
+public class TestSIMDLd4St4 {
+  static final int VECT_LENGTH = 256 * 1024;
+  static final int DIM = 4;
+  static final int LENGTH = VECT_LENGTH * DIM;
+  static int [] in = new int[LENGTH];
+  static int [] out = new int[LENGTH];
+  static int [] c = new int[DIM];
+  static short [] sin = new short[LENGTH];
+  static short [] sout = new short[LENGTH];
+  static short [] sc = new short[DIM];
+
+  public static void vect4DInit() {
+    int i;
+    for (i = 0; i < LENGTH; i++) {
+      in[i] = i + 3;
+      sin[i] = (short)(i + 3);
+    }
+    for (i = 0; i < DIM; i++) {
+      c[i] = i + 1;
+      sc[i] = (short)(i + 1);
+    }
+  }
+
+  public static void vect4DAddInt(
+          int[] out,
+          int[] in,
+          int[] c) {
+    int c0 = c[0];
+    int c1 = c[1];
+    int c2 = c[2];
+    int c3 = c[3];
+    for (int i = 0; i < VECT_LENGTH; i++) {
+      out[i*DIM] = in[i*DIM] + c0;
+      out[i*DIM + 1] = in[i*DIM + 1] + c1;
+      out[i*DIM + 2] = in[i*DIM + 2] + c2;
+      out[i*DIM + 3] = in[i*DIM + 3] + c3;
+    }
+  }
+
+  public static void vect4DMulInt(
+          int[] out,
+          int[] in,
+          int[] c) {
+    int c0 = c[0];
+    int c1 = c[1];
+    int c2 = c[2];
+    int c3 = c[3];
+    for (int i = 0; i < VECT_LENGTH; i++) {
+      out[i*DIM] = in[i*DIM] * c0;
+      out[i*DIM + 1] = in[i*DIM + 1] * c1;
+      out[i*DIM + 2] = in[i*DIM + 2] * c2;
+      out[i*DIM + 3] = in[i*DIM + 3] * c3;
+    }
+  }
+
+  public static void vect4DAddShort(
+          short[] out,
+          short[] in,
+          short[] c) {
+    short c0 = c[0];
+    short c1 = c[1];
+    short c2 = c[2];
+    short c3 = c[3];
+    for (int i = 0; i < VECT_LENGTH; i++) {
+      out[i*DIM] = (short)(in[i*DIM] + c0);
+      out[i*DIM + 1] = (short)(in[i*DIM + 1] + c1);
+      out[i*DIM + 2] = (short)(in[i*DIM + 2] + c2);
+      out[i*DIM + 3] = (short)(in[i*DIM + 3] + c3);
+    }
+  }
+
+  public static void vect4DMulShort(
+          short[] out,
+          short[] in,
+          short[] c) {
+    short c0 = c[0];
+    short c1 = c[1];
+    short c2 = c[2];
+    short c3 = c[3];
+    for (int i = 0; i < VECT_LENGTH; i++) {
+      out[i*DIM] = (short)(in[i*DIM] * c0);
+      out[i*DIM + 1] = (short)(in[i*DIM + 1] * c1);
+      out[i*DIM + 2] = (short)(in[i*DIM + 2] * c2);
+      out[i*DIM + 3] = (short)(in[i*DIM + 3] * c3);
+    }
+  }
+
+  @Setup
+  public void setup()
+  {
+    vect4DInit();
+  }
+
+  @Benchmark
+  public void testVect4D() {
+    vect4DAddInt(out, in, c);
+    vect4DMulInt(out, in, c);
+    vect4DAddShort(sout, sin, sc);
+    vect4DMulShort(sout, sin, sc);
+  }
+
+}