Sparse MM benchmarkyouki/benchmarksparse

3 files changed, 48 insertions, 39 deletions

diff --git a/bench/PackedFloatInOutBenchmark.cc b/bench/PackedFloatInOutBenchmark.cc
index 0c953f8..b53d7c5 100644
--- a/bench/PackedFloatInOutBenchmark.cc
+++ b/bench/PackedFloatInOutBenchmark.cc
@@ -32,40 +32,44 @@ void performance_test() {
   vector<vector<int>> shapes = {
     // NOTE: clang-format wants to use a different formatting but the current
     // formatting should be easier to read.
-    {1, 128, 512},
-    {1, 1024, 256},
+    {1, 512, 512},
+    {1, 512, 512},
     {1, 2048, 512},
-    {1, 4096, 1024},
-
-    {6, 256, 1024},
-    {6, 256, 2048},
-    {6, 512, 512},
-    {6, 1024, 256},
-    {6, 2048, 256},
-    {6, 2048, 512},
-    {6, 4096, 256},
-    {6, 4096, 1024},
-    {6, 4096, 2048},
-
-    {10, 2048, 256},
-    {10, 4096, 1024},
-
-    {20, 2048, 256},
-    {20, 4096, 1024},
-
-    {102, 1024, 512},
-    {102, 2323, 256},
-    {102, 512, 256},
-
-    {1, 800, 3200},
-    {1, 800, 8000},
-
-    {16, 256, 1500},
-    {16, 256, 1567},
-    {1, 128, 2876},
-    {16, 128, 1567},
-    {1, 128, 2722},
-    {16, 256, 512},
+    {1, 32000, 512},
+    // {1, 128, 512},
+    // {1, 1024, 256},
+    // {1, 2048, 512},
+    // {1, 4096, 1024},
+
+    // {6, 256, 1024},
+    // {6, 256, 2048},
+    // {6, 512, 512},
+    // {6, 1024, 256},
+    // {6, 2048, 256},
+    // {6, 2048, 512},
+    // {6, 4096, 256},
+    // {6, 4096, 1024},
+    // {6, 4096, 2048},
+
+    // {10, 2048, 256},
+    // {10, 4096, 1024},
+
+    // {20, 2048, 256},
+    // {20, 4096, 1024},
+
+    // {102, 1024, 512},
+    // {102, 2323, 256},
+    // {102, 512, 256},
+
+    // {1, 800, 3200},
+    // {1, 800, 8000},
+
+    // {16, 256, 1500},
+    // {16, 256, 1567},
+    // {1, 128, 2876},
+    // {16, 128, 1567},
+    // {1, 128, 2722},
+    // {16, 256, 512},
   };
   // clang-format on
   bool flush = true;
diff --git a/bench/SpMMFP32Benchmark.cc b/bench/SpMMFP32Benchmark.cc
index 6f6892a..21167a5 100644
--- a/bench/SpMMFP32Benchmark.cc
+++ b/bench/SpMMFP32Benchmark.cc
@@ -7,9 +7,11 @@ using namespace std;
 using namespace fbgemm;
 
 int main(int, char**) {
-  vector<char> llc(128 * 1024 * 1024);
+//  vector<char> llc(128 * 1024 * 1024);
+  vector<char> llc(1 * 512 * 2048);
 
-  vector<vector<int>> shapes = {{1024, 128, 1024}};
+//  vector<vector<int>> shapes = {{1024, 128, 1024}};
+  vector<vector<int>> shapes = {{2048, 1, 512}, {512, 1, 512}, {32000, 1, 512}};
 
   // C is MxN -> CT is NxM
   // A is MxK -> BT is KxM
@@ -38,10 +40,11 @@ int main(int, char**) {
       constexpr int NWARMUP = 5;
       constexpr int NITER = 32;
       auto secs = measureWithWarmup(
-          [&]() { fn(bData.data(), cData.data(), 0); }, NWARMUP, NITER, &llc);
+          [&]() { //fn = generateSpMM<float>(m, n, k, aData.data(), lda, ldb, ldc);
+            fn(bData.data(), cData.data(), 0); }, NWARMUP, NITER, &llc);
 
       auto secs_varying_n = measureWithWarmup(
-          [&]() {
+          [&]() { //fn_varying_n = generateSpMM<float>(m, k, aData.data(), lda);
             fn_varying_n(
                 bData.data(),
                 cData.data(),
diff --git a/bench/SpMMI8Benchmark.cc b/bench/SpMMI8Benchmark.cc
index 212fd09..484f0cc 100644
--- a/bench/SpMMI8Benchmark.cc
+++ b/bench/SpMMI8Benchmark.cc
@@ -8,12 +8,14 @@ using namespace std;
 using namespace fbgemm;
 
 int main(int, char**) {
-  vector<char> llc(128 * 1024 * 1024);
+  //vector<char> llc(128 * 1024 * 1024);
+  vector<char> llc(1 * 512 * 2048);
 
   // vector<vector<unsigned>> shapes = {{64, 64, 64}};
   // vector<vector<unsigned>> shapes = {{1, 16, 4}};
 
-  vector<vector<unsigned>> shapes = {{1024, 128, 1024}};
+  // vector<vector<unsigned>> shapes = {{1024, 128, 1024}};
+  vector<vector<unsigned>> shapes = {{2048, 1, 512}, {512, 1, 512}, {32000, 1, 512}};
 
   // C is MxN -> CT is NxM
   // A is MxK -> BT is KxM