Add vectorization sample.

This adds a new generic vectorization sample to replace hello-neon. Most
importantly, it covers non-Neon options for SIMD. One of the useful
things it shows, for example, is that there's actually no reason to
write SIMD code the way that hello-neon does any more.

This also solves a much simpler problem (small matrix multiplication),
which makes it easier to see how to deal with the SIMD features rather
than figuring out what a FIR filter is.

Finally, this sample benchmarks each of the implementations so it's
obvious what is and isn't worth doing. I was sort of surprised that
auto-vectorization didn't do better, and was pleased to learn that
there's no reason at all to write Neon intrinsics.

I'll delete hello-neon after this merges and I've fixed up the doc
links.

android#1011

gradle/libs.versions.toml

@@ -14,9 +14,13 @@ material = "1.12.0"
 jetbrainsKotlinJvm = "1.7.21"
 oboe = "1.8.1"
 
+activityCompose = "1.9.0"
+composeBom = "2023.08.00"
+coreKtx = "1.13.1"
 curl = "7.79.1-beta-1"
 googletest = "1.11.0-beta-1"
 jsoncpp = "1.9.5-beta-1"
+lifecycleRuntimeKtx = "2.7.0"
 openssl = "1.1.1q-beta-1"
 
 [libraries]
@@ -40,6 +44,17 @@ openssl = { group = "com.android.ndk.thirdparty", name = "openssl", version.ref
 
 # build-logic dependencies
 android-gradlePlugin = { group = "com.android.tools.build", name = "gradle", version.ref = "agp" }
+androidx-activity-compose = { group = "androidx.activity", name = "activity-compose", version.ref = "activityCompose" }
+androidx-compose-bom = { group = "androidx.compose", name = "compose-bom", version.ref = "composeBom" }
+androidx-core-ktx = { group = "androidx.core", name = "core-ktx", version.ref = "coreKtx" }
+androidx-lifecycle-runtime-ktx = { group = "androidx.lifecycle", name = "lifecycle-runtime-ktx", version.ref = "lifecycleRuntimeKtx" }
+androidx-material3 = { group = "androidx.compose.material3", name = "material3" }
+androidx-ui = { group = "androidx.compose.ui", name = "ui" }
+androidx-ui-graphics = { group = "androidx.compose.ui", name = "ui-graphics" }
+androidx-ui-test-junit4 = { group = "androidx.compose.ui", name = "ui-test-junit4" }
+androidx-ui-test-manifest = { group = "androidx.compose.ui", name = "ui-test-manifest" }
+androidx-ui-tooling = { group = "androidx.compose.ui", name = "ui-tooling" }
+androidx-ui-tooling-preview = { group = "androidx.compose.ui", name = "ui-tooling-preview" }
 
 [plugins]
 android-application = { id = "com.android.application", version.ref = "agp" }

settings.gradle

@@ -63,3 +63,4 @@ include(":teapots:image-decoder")
 include(":teapots:more-teapots")
 include(":teapots:textured-teapot")
 include(":unit-test:app")
+include(":vectorization")

vectorization/.gitignore

@@ -0,0 +1 @@
+/build

vectorization/README.md

@@ -0,0 +1,149 @@
+# Vectorization
+
+This sample shows how to implement matrix multiplication using various
+vectorization approaches.
+
+Note: You should not reuse this matrix library in your application. It was not
+written to be useful beyond the scope of this demo. If you're looking for a
+matrix library, you probably want [GLM] for graphics applications, or a linear
+algebra library for compute applications.
+
+The sample app will benchmark each implementation and display the average run
+time over 1,000,000 runs. The goal of this sample is to illustrate the trade-
+offs of each implementation in terms of flexibility, readability, and
+performance.
+
+Given the relatively small problem size used here (4x4 matrices and vec4s), the
+best performing implementations in this sample are the ones that can best
+improve over the naive implementation without large set up costs. You should not
+take the results of this sample as authoritative: if performance is important to
+you, you **must** benchmark your code for workloads realistic for your app.
+
+If you're not familiar with it [Godbolt] is an invaluable tool for examining
+compiler optimizer behavior. You could also use `llvm-objdump` from the command
+line, but that's just Godbolt with extra steps.
+
+## Implementations
+
+This sample contains the following implementations. Each of their trade-offs are
+discussed briefly, but as mentioned above, you should not rely on the
+performance results measured here to make a decision for your app.
+
+### Auto-vectorization
+
+See [auto_vectorization.h] for the implementation.
+
+This implementation is written in generic C++ and contains no explicit SIMD. The
+only vectorization that will be performed is Clang's auto-vectorization. This
+makes for the most portable code and readable code, but at the cost of
+performance.
+
+See https://llvm.org/docs/Vectorizers.html for Clang's docs about
+auto-vectorization.
+
+### std::simd
+
+This isn't actually available yet. It's an experimental part of the C++ standard
+and is in development in libc++, but NDK r27 happened to catch it right in the
+middle of a rewrite, so it's not currently usable.
+
+See https://en.cppreference.com/w/cpp/experimental/simd/simd.
+
+### Clang vectors
+
+See [clang_vector.h] for the implementation.
+
+This implementation uses Clang's generic vector types. This code is mostly as
+portable as the auto-vectorization implementation, with the only caveat being
+that it is limited by the width of the vector registers for the target hardware.
+To deal with problems that don't fit in the target's vector registers, you would
+need to either alter the algorithm to tile the operations, or use Scalable
+Vector Extensions (AKA [SVE]).
+
+However, the benefit of the portability trade-off is that this does outperform
+the auto-vectorization implementation.
+
+See
+https://clang.llvm.org/docs/LanguageExtensions.html#vectors-and-extended-vectors.
+
+### Neon
+
+See [neon.h] for the implementation.
+
+This implementation uses the Arm-specific Neon vector types. The implementation
+is nearly identical to the one in [clang_vector.h]. The only differences how the
+vector type is specified, and the restriction to only working with matrices of 4
+rows, which was done for the sake of brevity. It doesn't use the Neon intrinsics
+from `arm_neon.h`? Why not? If you look at how the intrinsics in that file are
+defined, all they actually do (for a little endian system, and Android does not
+support big endian, so we can ignore that) is use the `*` operator and leave the
+correct instruction selection up to Clang.
+
+In other words, you should probably never use this. The generated code should be
+identical to code written with Clang's arch-generic vectors. These benchmarks
+show nearly identical results when run on a Pixel 4A (5G), and the differences
+are actually a problem with the benchmark harness itself: Neon sometimes runs
+faster, unless you swap the order of the benchmarks, then the Clang Vector
+implementation runs faster.
+
+See
+https://developer.arm.com/documentation/dht0002/a/Introducing-NEON/Developing-for-NEON/Intrinsics
+for more information.
+
+### Clang matrices
+
+See [matrix.h] for the implementation. This is the default implementation for
+`Matrix::operator*`, so unlike the others that file contains the rest of the
+`Matrix` class as well.
+
+This implementation uses Clang's built-in matrix type. This is an experimental
+feature in Clang, but it has the simplest code (because some kind Clang person
+wrote the hard part) and performs the best by a wide margin. There are
+implementation defined limits on the size of the matrix, but within those limits
+the code is as portable as the auto-vectorization implementation. The docs say
+the feature is still under development and subject to change, so be wary of
+using this in production, and definitely don't use these types as part of your
+ABI.
+
+See https://clang.llvm.org/docs/LanguageExtensions.html#matrix-types for more
+details.
+
+### OpenMP SIMD
+
+See [omp_simd.h] for the implementation.
+
+This implementation uses OpenMP's SIMD directive. For some reason this
+under-performs even the auto-vectorized implementation. There are a lot of
+additional specifiers that can be added to the simd directive that would maybe
+improve this implementation. Patches welcome :)
+
+See https://www.openmp.org/spec-html/5.0/openmpsu42.html for more information.
+
+## Alternatives not shown here
+
+There are other approaches that could be used that aren't shown here.
+
+[SVE] scales SIMD to arbitrarily sized vectors, and the C extensions, while
+making for less concise code than is needed for a constrained vector size like
+we have here, handle windowing of data to fit the hardware vector size for you.
+For problems like the small matrix multiply we do here, it's overkill. For
+portability to a wide variety of (Arm) CPUs, it can reduce the difficulty of
+writing SIMD code.
+
+GPU acceleration is a better fit for large data sets. That approach isn't shown
+here because it's substantially more code to set up the GPU for this
+computation, and our data size is so small that the cost of GPU initialization
+and streaming the data to the GPU is likely to make that a net-loss. If you want
+to learn more about GPU compute, see https://vulkan-tutorial.com/Compute_Shader,
+https://www.khronos.org/opengl/wiki/Compute_Shader, and
+https://www.khronos.org/opencl/ (while OpenCL is not guaranteed to be available
+for all Android devices, it is a very common OEM extension).
+
+[auto_vectorization.h]: src/main/cpp/auto_vectorization.h
+[clang_vector.h]: src/main/cpp/clang_vector.h
+[GLM]: https://github.com/g-truc/glm
+[Gobolt]: https://godbolt.org/
+[matrix.h]: src/main/cpp/matrix.h
+[neon.h]: src/main/cpp/neon.h
+[omp_simd.h]: src/main/cpp/omp_simd.h
+[SVE]: https://developer.arm.com/Architectures/Scalable%20Vector%20Extensions

vectorization/build.gradle.kts

@@ -0,0 +1,45 @@
+plugins {
+    id("ndksamples.android.application")
+    id("ndksamples.android.kotlin")
+}
+
+android {
+    namespace = "com.android.ndk.samples.vectorization"
+
+    defaultConfig {
+        applicationId = "com.android.ndk.samples.vectorization"
+
+        vectorDrawables {
+            useSupportLibrary = true
+        }
+    }
+
+    externalNativeBuild {
+        cmake {
+            path = file("src/main/cpp/CMakeLists.txt")
+        }
+    }
+
+    buildFeatures {
+        compose = true
+        prefab = true
+    }
+
+    composeOptions {
+        kotlinCompilerExtensionVersion = "1.5.1"
+    }
+}
+
+dependencies {
+    implementation(project(":base"))
+    implementation(libs.androidx.core.ktx)
+    implementation(libs.androidx.lifecycle.runtime.ktx)
+    implementation(libs.androidx.activity.compose)
+    implementation(platform(libs.androidx.compose.bom))
+    implementation(libs.androidx.ui)
+    implementation(libs.androidx.ui.graphics)
+    implementation(libs.androidx.ui.tooling.preview)
+    implementation(libs.androidx.material3)
+    debugImplementation(libs.androidx.ui.tooling)
+    debugImplementation(libs.androidx.ui.test.manifest)
+}

vectorization/src/main/AndroidManifest.xml

@@ -0,0 +1,24 @@
+<?xml version="1.0" encoding="utf-8"?>
+<manifest xmlns:android="http://schemas.android.com/apk/res/android">
+
+    <application
+        android:allowBackup="true"
+        android:icon="@mipmap/ic_launcher"
+        android:label="@string/app_name"
+        android:roundIcon="@mipmap/ic_launcher_round"
+        android:supportsRtl="true"
+        android:theme="@style/Theme.NDKSamples">
+        <activity
+            android:name=".VectorizationActivity"
+            android:exported="true"
+            android:label="@string/app_name"
+            android:theme="@style/Theme.NDKSamples">
+            <intent-filter>
+                <action android:name="android.intent.action.MAIN" />
+
+                <category android:name="android.intent.category.LAUNCHER" />
+            </intent-filter>
+        </activity>
+    </application>
+
+</manifest>

vectorization/src/main/cpp/CMakeLists.txt

@@ -0,0 +1,32 @@
+cmake_minimum_required(VERSION 3.22.1)
+project(Vectorization LANGUAGES CXX)
+
+add_compile_options(-Wall -Wextra -Werror)
+
+find_package(base REQUIRED CONFIG)
+
+add_library(app
+    SHARED
+    benchmark.cpp
+    jni.cpp
+)
+
+target_compile_features(app PUBLIC cxx_std_23)
+target_compile_options(app PUBLIC -fenable-matrix -fopenmp)
+
+target_link_libraries(app
+    PRIVATE
+    base::base
+    log
+)
+
+target_link_options(app
+    PRIVATE
+    -flto
+    -Wl,--version-script,${CMAKE_SOURCE_DIR}/libapp.map.txt
+)
+
+set_target_properties(app
+    PROPERTIES
+    LINK_DEPENDS ${CMAKE_SOURCE_DIR}/libapp.map.txt
+)

vectorization/src/main/cpp/auto_vectorization.h

@@ -0,0 +1,56 @@
+/*
+ * Copyright (C) 2024 The Android Open Source Project
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#pragma once
+
+#include <stdint.h>
+
+#include "matrix.h"
+
+namespace samples::vectorization {
+
+/**
+ * Multiplies two compatible matrices and returns the result.
+ *
+ * @tparam T The type of each matrix cell.
+ * @tparam M The number of rows in the left operand and the result.
+ * @tparam N The number of columns in the left operand, and the rows in the
+ * right operand.
+ * @tparam P The number of columns in the right operand and the result.
+ * @param lhs The left operand.
+ * @param rhs The right operand.
+ * @return The result of lhs * rhs.
+ */
+template <typename T, size_t M, size_t N, size_t P>
+Matrix<M, P, T> MultiplyWithAutoVectorization(const Matrix<M, N, T>& lhs,
+                                              const Matrix<N, P, T>& rhs) {
+  // This may look like an unfair benchmark because this implementation uses the
+  // less vector friendly one than the others, however, using the vector
+  // friendly algorithm here actually made performance worse.
+  Matrix<M, P, T> result;
+  for (auto i = 0U; i < M; i++) {
+    for (auto j = 0U; j < P; j++) {
+      T sum = {};
+      for (auto k = 0U; k < N; k++) {
+        sum += lhs.get(i, k) * rhs[k, j];
+      }
+      result[i, j] = sum;
+    }
+  }
+  return result;
+}
+
+}  // namespace samples::vectorization