Add benchmark for the sum pooling vectorization pass

benchmarks/DeepLearning/Ops/PoolingNhwcSumOp/BuddyPoolingNhwcSum.mlir

@@ -0,0 +1,10 @@
+func @buddy_pooling_nhwc_sum(%input: memref<1x1024x1024x1xf32>, %filter: memref<3x3xf32>,
+                       %output: memref<1x1022x1022x1xf32>) {
+
+  linalg.pooling_nhwc_sum
+    {strides = dense<1>: tensor<2xi64>, dilations = dense<1>: tensor<2xi64>}
+    ins(%input, %filter: memref<1x1024x1024x1xf32>, memref<3x3xf32>)
+    outs(%output: memref<1x1022x1022x1xf32>)
+
+  return
+}

benchmarks/DeepLearning/Ops/PoolingNhwcSumOp/BuddyPoolingNhwcSumBenchmark.cpp

@@ -0,0 +1,53 @@
+//===- PoolingNhwcSumBenchmark.cpp ----------------------------------------===//
+//
+// 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.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file implements the benchmark for sum pooling (nhwc) operation.
+//
+//===----------------------------------------------------------------------===//
+
+#include "Utils/Container.h"
+#include <benchmark/benchmark.h>
+
+namespace {
+
+// Declare the pooling_nhwc_sum interface.
+extern "C" {
+void _mlir_ciface_buddy_pooling_nhwc_sum(MemRef<float, 4> *input,
+                                   MemRef<float, 2> *kernel,
+                                   MemRef<float, 4> *output);
+}
+
+// Create input, filter, and output.
+MemRef<float, 4> input({1, 1024, 1024, 1}, 0.1);
+MemRef<float, 2> kernel({3, 3}, 1.0);
+intptr_t sizesOutput[4] = {1, 1022, 1022, 1};
+MemRef<float, 4> output(sizesOutput);
+
+// Define benchmark function.
+void Buddy_PoolingNhwcSum(benchmark::State &state) {
+  for (auto _ : state) {
+    for (int i = 0; i < state.range(0); ++i) {
+      _mlir_ciface_buddy_pooling_nhwc_sum(&input, &kernel, &output);
+    }
+  }
+}
+
+} // namespace
+
+// Register benchmarking function with different arguments.
+BENCHMARK(Buddy_PoolingNhwcSum)->Arg(1)->Unit(benchmark::kMillisecond);
+BENCHMARK(Buddy_PoolingNhwcSum)->Arg(4)->Unit(benchmark::kMillisecond);
+

benchmarks/DeepLearning/Ops/PoolingNhwcSumOp/CMakeLists.txt

@@ -1,14 +1,63 @@
-add_buddy_ops_benchmark(pooling-nhwc-sum-benchmark
-  MLIR PoolingNhwcSum.mlir
-  BITCODE pooling-nhwc-sum.o
-  LIBRARY PoolingNhwcSum
-  OPTIONS
-    -convert-linalg-to-loops 
-    -convert-scf-to-cf 
-    -convert-linalg-to-llvm 
-    -lower-affine 
-    --convert-memref-to-llvm 
-    -convert-func-to-llvm='emit-c-wrappers=1' 
-    -reconcile-unrealized-casts
-  SOURCE Main.cpp MLIROptBenchmark.cpp
+#-------------------------------------------------------------------------------
+# MLIR Linalg Dialect PoolingNhwcSum Operation + Upstream Lowering Passes
+#-------------------------------------------------------------------------------
+
+add_custom_command(OUTPUT mlir-pooling-nhwc-sum.o
+  COMMAND ${LLVM_MLIR_BINARY_DIR}/mlir-opt
+          ${CMAKE_CURRENT_SOURCE_DIR}/MLIRPoolingNhwcSum.mlir 
+            -convert-linalg-to-loops
+            -convert-scf-to-cf 
+            -convert-linalg-to-llvm 
+            -lower-affine 
+            --convert-memref-to-llvm 
+            -convert-func-to-llvm='emit-c-wrappers=1' 
+            -reconcile-unrealized-casts | 
+          ${LLVM_MLIR_BINARY_DIR}/mlir-translate --mlir-to-llvmir |
+          ${LLVM_MLIR_BINARY_DIR}/llc -mtriple=${BUDDY_OPT_TRIPLE} 
+            -mattr=${BUDDY_OPT_ATTR} --filetype=obj -o 
+          ${CMAKE_CURRENT_BINARY_DIR}/mlir-pooling-nhwc-sum.o
 )
+add_library(MLIRPoolingNhwcSum ${CMAKE_CURRENT_BINARY_DIR}/mlir-pooling-nhwc-sum.o)
+set_target_properties(MLIRPoolingNhwcSum PROPERTIES LINKER_LANGUAGE CXX)
+
+#-------------------------------------------------------------------------------
+# MLIR Linalg Dialect PoolingNhwcSum Operation + CB Pooling algorithm
+#-------------------------------------------------------------------------------
+
+add_custom_command(OUTPUT buddy-pooling-nhwc-sum.o
+  COMMAND ${BUDDY_OPT_BUILD_DIR}/bin/buddy-opt
+          ${CMAKE_CURRENT_SOURCE_DIR}/BuddyPoolingNhwcSum.mlir 
+            -pooling-vectorization 
+            -lower-affine 
+            -convert-vector-to-scf 
+            -convert-scf-to-cf 
+            -tensor-bufferize 
+            -convert-vector-to-llvm 
+            -convert-memref-to-llvm 
+            -convert-arith-to-llvm 
+            -convert-func-to-llvm='emit-c-wrappers=1' 
+            -reconcile-unrealized-casts | 
+          ${LLVM_MLIR_BINARY_DIR}/mlir-translate --mlir-to-llvmir |
+          ${LLVM_MLIR_BINARY_DIR}/llc -mtriple=${BUDDY_OPT_TRIPLE} 
+            -mattr=${BUDDY_OPT_ATTR} --filetype=obj -o 
+          ${CMAKE_CURRENT_BINARY_DIR}/buddy-pooling-nhwc-sum.o
+)
+add_library(BuddyPoolingNhwcSum ${CMAKE_CURRENT_BINARY_DIR}/buddy-pooling-nhwc-sum.o)
+set_target_properties(BuddyPoolingNhwcSum PROPERTIES LINKER_LANGUAGE CXX)
+
+#-------------------------------------------------------------------------------
+# PoolingNhwcSum Benchmark Target
+#-------------------------------------------------------------------------------
+
+add_executable(pooling-nhwc-sum-benchmark
+  Main.cpp 
+  MLIRPoolingNhwcSumBenchmark.cpp
+  BuddyPoolingNhwcSumBenchmark.cpp
+  )
+
+target_link_libraries(pooling-nhwc-sum-benchmark 
+  GoogleBenchmark
+  MLIRPoolingNhwcSum
+  BuddyPoolingNhwcSum
+  Container
+  )

benchmarks/DeepLearning/Ops/PoolingNhwcSumOp/MLIRPoolingNhwcSum.mlir

@@ -0,0 +1,10 @@
+func @pooling_nhwc_sum(%input: memref<1x1024x1024x1xf32>, %filter: memref<3x3xf32>,
+                       %output: memref<1x1022x1022x1xf32>) {
+
+  linalg.pooling_nhwc_sum
+    {strides = dense<1>: tensor<2xi64>, dilations = dense<1>: tensor<2xi64>}
+    ins(%input, %filter: memref<1x1024x1024x1xf32>, memref<3x3xf32>)
+    outs(%output: memref<1x1022x1022x1xf32>)
+
+  return
+}

benchmarks/DeepLearning/Ops/PoolingNhwcSumOp/MLIROptBenchmark.cpp → benchmarks/DeepLearning/Ops/PoolingNhwcSumOp/MLIRPoolingNhwcSumBenchmark.cpp

@@ -1,4 +1,4 @@
-//===- MLIROptBenchmark.cpp -----------------------------------------------===//
+//===- PoolingNhwcSumBenchmark.cpp ----------------------------------------===//
 //
 // Licensed under the Apache License, Version 2.0 (the "License");
 // you may not use this file except in compliance with the License.
@@ -31,13 +31,13 @@ void _mlir_ciface_pooling_nhwc_sum(MemRef<float, 4> *input,
 }
 
 // Create input, filter, and output.
-MemRef<float, 4> input({1, 6, 6, 1}, 1.0);
+MemRef<float, 4> input({1, 1024, 1024, 1}, 0.1);
 MemRef<float, 2> filter({3, 3}, 1.0);
-intptr_t sizesOutput[4] = {1, 3, 3, 1};
+intptr_t sizesOutput[4] = {1, 1022, 1022, 1};
 MemRef<float, 4> output(sizesOutput);
 
 // Define benchmark function.
-void BM_PoolingNhwcSum(benchmark::State &state) {
+void MLIR_PoolingNhwcSum(benchmark::State &state) {
   for (auto _ : state) {
     for (int i = 0; i < state.range(0); ++i) {
       _mlir_ciface_pooling_nhwc_sum(&input, &filter, &output);
@@ -48,18 +48,5 @@ void BM_PoolingNhwcSum(benchmark::State &state) {
 } // namespace
 
 // Register benchmarking function with different arguments.
-BENCHMARK(BM_PoolingNhwcSum)->Arg(1);
-BENCHMARK(BM_PoolingNhwcSum)->Arg(4);
-
-// Print result function.
-void printResult() {
-  // Create the output memref.
-  MemRef<float, 4> output(sizesOutput);
-  // Run the mlir function.
-  _mlir_ciface_pooling_nhwc_sum(&input, &filter, &output);
-  // Print the output.
-  std::cout << "Output: [ ";
-  for (int i = 0; i < 9; ++i)
-    std::cout << output[i] << " ";
-  std::cout << "]" << std::endl;
-}
+BENCHMARK(MLIR_PoolingNhwcSum)->Arg(1)->Unit(benchmark::kMillisecond);
+BENCHMARK(MLIR_PoolingNhwcSum)->Arg(4)->Unit(benchmark::kMillisecond);

benchmarks/DeepLearning/Ops/PoolingNhwcSumOp/Main.cpp

@@ -20,14 +20,10 @@
 
 #include <benchmark/benchmark.h>
 
-void printResult();
-
 int main(int argc, char **argv) {
   // Run benchmarks.
   ::benchmark::Initialize(&argc, argv);
   ::benchmark::RunSpecifiedBenchmarks();
-  // Print result.
-  printResult();
 
   return 0;
 }