Standardize benchmark names (#437)

This PR moves the get_name function in the benchmarks to a common location, ensuring that different backends can obtain uniform benchmark names. --------- Co-authored-by: Alejandro Acosta <[email protected]>
codeplaysoftware · Jul 10, 2023 · 0355a58 · 0355a58
1 parent c0018d9
commit 0355a58
Show file tree

Hide file tree

Showing 110 changed files with 1,176 additions and 1,080 deletions.
diff --git a/benchmark/cublas/blas1/asum.cpp b/benchmark/cublas/blas1/asum.cpp
@@ -25,13 +25,8 @@
 
 #include "../utils.hpp"
 
-template <typename scalar_t>
-std::string get_name(int size) {
-  std::ostringstream str{};
-  str << "BM_Asum<" << blas_benchmark::utils::get_type_name<scalar_t>() << ">/";
-  str << size;
-  return str.str();
-}
+constexpr blas_benchmark::utils::Level1Op benchmark_op =
+    blas_benchmark::utils::Level1Op::asum;
 
 template <typename scalar_t, typename... args_t>
 static inline void cublas_routine(args_t&&... args) {
@@ -140,8 +135,11 @@ void register_benchmark(blas_benchmark::Args& args,
                          index_t size, bool* success) {
       run<scalar_t>(st, cuda_handle_ptr, size, success);
     };
-    benchmark::RegisterBenchmark(get_name<scalar_t>(size).c_str(), BM_lambda,
-                                 cuda_handle_ptr, size, success)
+    benchmark::RegisterBenchmark(
+        blas_benchmark::utils::get_name<benchmark_op, scalar_t>(
+            size, blas_benchmark::utils::MEM_TYPE_USM)
+            .c_str(),
+        BM_lambda, cuda_handle_ptr, size, success)
         ->UseRealTime();
   }
 }

diff --git a/benchmark/cublas/blas1/axpy.cpp b/benchmark/cublas/blas1/axpy.cpp
@@ -25,13 +25,8 @@
 
 #include "../utils.hpp"
 
-template <typename scalar_t>
-std::string get_name(int size) {
-  std::ostringstream str{};
-  str << "BM_Axpy<" << blas_benchmark::utils::get_type_name<scalar_t>() << ">/";
-  str << size;
-  return str.str();
-}
+constexpr blas_benchmark::utils::Level1Op benchmark_op =
+    blas_benchmark::utils::Level1Op::axpy;
 
 template <typename scalar_t, typename... args_t>
 static inline void cublas_routine(args_t&&... args) {
@@ -50,8 +45,8 @@ void run(benchmark::State& state, cublasHandle_t* cuda_handle_ptr, index_t size,
   blas_benchmark::utils::set_benchmark_label<scalar_t>(state);
 
   // init Google-benchmark counters.
-  blas_benchmark::utils::init_level_1_counters<
-      blas_benchmark::utils::Level1Op::axpy, scalar_t>(state, size);
+  blas_benchmark::utils::init_level_1_counters<benchmark_op, scalar_t>(state,
+                                                                       size);
 
   cublasHandle_t& cuda_handle = *cuda_handle_ptr;
 
@@ -139,8 +134,11 @@ void register_benchmark(blas_benchmark::Args& args,
                          index_t size, bool* success) {
       run<scalar_t>(st, cuda_handle_ptr, size, success);
     };
-    benchmark::RegisterBenchmark(get_name<scalar_t>(size).c_str(), BM_lambda,
-                                 cuda_handle_ptr, size, success)
+    benchmark::RegisterBenchmark(
+        blas_benchmark::utils::get_name<benchmark_op, scalar_t>(
+            size, blas_benchmark::utils::MEM_TYPE_USM)
+            .c_str(),
+        BM_lambda, cuda_handle_ptr, size, success)
         ->UseRealTime();
   }
 }

diff --git a/benchmark/cublas/blas1/dot.cpp b/benchmark/cublas/blas1/dot.cpp
@@ -25,13 +25,8 @@
 
 #include "../utils.hpp"
 
-template <typename scalar_t>
-std::string get_name(int size) {
-  std::ostringstream str{};
-  str << "BM_Dot<" << blas_benchmark::utils::get_type_name<scalar_t>() << ">/";
-  str << size;
-  return str.str();
-}
+constexpr blas_benchmark::utils::Level1Op benchmark_op =
+    blas_benchmark::utils::Level1Op::dot;
 
 template <typename scalar_t, typename... args_t>
 static inline void cublas_routine(args_t&&... args) {
@@ -141,8 +136,11 @@ void register_benchmark(blas_benchmark::Args& args,
                          index_t size, bool* success) {
       run<scalar_t>(st, cuda_handle_ptr, size, success);
     };
-    benchmark::RegisterBenchmark(get_name<scalar_t>(size).c_str(), BM_lambda,
-                                 cuda_handle_ptr, size, success)
+    benchmark::RegisterBenchmark(
+        blas_benchmark::utils::get_name<benchmark_op, scalar_t>(
+            size, blas_benchmark::utils::MEM_TYPE_USM)
+            .c_str(),
+        BM_lambda, cuda_handle_ptr, size, success)
         ->UseRealTime();
   }
 }

diff --git a/benchmark/cublas/blas1/iamax.cpp b/benchmark/cublas/blas1/iamax.cpp
@@ -26,13 +26,8 @@
 #include "../utils.hpp"
 #include "common/common_utils.hpp"
 
-template <typename scalar_t>
-std::string get_name(int size) {
-  std::ostringstream str{};
-  str << "BM_Iamax<" << blas_benchmark::utils::get_type_name<scalar_t>();
-  str << ">/" << size;
-  return str.str();
-}
+constexpr blas_benchmark::utils::Level1Op benchmark_op =
+    blas_benchmark::utils::Level1Op::iamax;
 
 template <typename scalar_t, typename... args_t>
 static inline void cublas_routine(args_t&&... args) {
@@ -147,8 +142,11 @@ void register_benchmark(blas_benchmark::Args& args,
                          index_t size, bool* success) {
       run<scalar_t>(st, cuda_handle_ptr, size, success);
     };
-    benchmark::RegisterBenchmark(get_name<scalar_t>(size).c_str(), BM_lambda,
-                                 cuda_handle_ptr, size, success)
+    benchmark::RegisterBenchmark(
+        blas_benchmark::utils::get_name<benchmark_op, scalar_t>(
+            size, blas_benchmark::utils::MEM_TYPE_USM)
+            .c_str(),
+        BM_lambda, cuda_handle_ptr, size, success)
         ->UseRealTime();
   }
 }

diff --git a/benchmark/cublas/blas1/iamin.cpp b/benchmark/cublas/blas1/iamin.cpp
@@ -26,13 +26,8 @@
 #include "../utils.hpp"
 #include "common/common_utils.hpp"
 
-template <typename scalar_t>
-std::string get_name(int size) {
-  std::ostringstream str{};
-  str << "BM_Iamin<" << blas_benchmark::utils::get_type_name<scalar_t>();
-  str << ">/" << size;
-  return str.str();
-}
+constexpr blas_benchmark::utils::Level1Op benchmark_op =
+    blas_benchmark::utils::Level1Op::iamin;
 
 template <typename scalar_t, typename... args_t>
 static inline void cublas_routine(args_t&&... args) {
@@ -146,8 +141,11 @@ void register_benchmark(blas_benchmark::Args& args,
                          index_t size, bool* success) {
       run<scalar_t>(st, cuda_handle_ptr, size, success);
     };
-    benchmark::RegisterBenchmark(get_name<scalar_t>(size).c_str(), BM_lambda,
-                                 cuda_handle_ptr, size, success)
+    benchmark::RegisterBenchmark(
+        blas_benchmark::utils::get_name<benchmark_op, scalar_t>(
+            size, blas_benchmark::utils::MEM_TYPE_USM)
+            .c_str(),
+        BM_lambda, cuda_handle_ptr, size, success)
         ->UseRealTime();
   }
 }

diff --git a/benchmark/cublas/blas1/nrm2.cpp b/benchmark/cublas/blas1/nrm2.cpp
@@ -26,13 +26,8 @@
 #include "../utils.hpp"
 #include <type_traits>
 
-template <typename scalar_t>
-std::string get_name(int size) {
-  std::ostringstream str{};
-  str << "BM_Nrm2<" << blas_benchmark::utils::get_type_name<scalar_t>() << ">/";
-  str << size;
-  return str.str();
-}
+constexpr blas_benchmark::utils::Level1Op benchmark_op =
+    blas_benchmark::utils::Level1Op::nrm2;
 
 template <typename scalar_t, typename... args_t>
 static inline void cublas_routine(args_t&&... args) {
@@ -141,8 +136,11 @@ void register_benchmark(blas_benchmark::Args& args,
                          index_t size, bool* success) {
       run<scalar_t>(st, cuda_handle_ptr, size, success);
     };
-    benchmark::RegisterBenchmark(get_name<scalar_t>(size).c_str(), BM_lambda,
-                                 cuda_handle_ptr, size, success)
+    benchmark::RegisterBenchmark(
+        blas_benchmark::utils::get_name<benchmark_op, scalar_t>(
+            size, blas_benchmark::utils::MEM_TYPE_USM)
+            .c_str(),
+        BM_lambda, cuda_handle_ptr, size, success)
         ->UseRealTime();
   }
 }

diff --git a/benchmark/cublas/blas1/rotg.cpp b/benchmark/cublas/blas1/rotg.cpp
@@ -25,12 +25,8 @@
 
 #include "../utils.hpp"
 
-template <typename scalar_t>
-std::string get_name() {
-  std::ostringstream str{};
-  str << "BM_Rotg<" << blas_benchmark::utils::get_type_name<scalar_t>() << ">/";
-  return str.str();
-}
+constexpr blas_benchmark::utils::Level1Op benchmark_op =
+    blas_benchmark::utils::Level1Op::rotg;
 
 template <typename scalar_t, typename... args_t>
 static inline void cublas_routine(args_t&&... args) {
@@ -158,8 +154,11 @@ void register_benchmark(blas_benchmark::Args& args,
                        bool* success) {
     run<scalar_t>(st, cuda_handle_ptr, success);
   };
-  benchmark::RegisterBenchmark(get_name<scalar_t>().c_str(), BM_lambda,
-                               cuda_handle_ptr, success)
+  benchmark::RegisterBenchmark(
+      blas_benchmark::utils::get_name<benchmark_op, scalar_t>(
+          blas_benchmark::utils::MEM_TYPE_USM)
+          .c_str(),
+      BM_lambda, cuda_handle_ptr, success)
       ->UseRealTime();
 }
 

diff --git a/benchmark/cublas/blas1/rotm.cpp b/benchmark/cublas/blas1/rotm.cpp
@@ -25,13 +25,8 @@
 
 #include "../utils.hpp"
 
-template <typename scalar_t>
-std::string get_name(int size) {
-  std::ostringstream str{};
-  str << "BM_Rotm<" << blas_benchmark::utils::get_type_name<scalar_t>() << ">/";
-  str << size;
-  return str.str();
-}
+constexpr blas_benchmark::utils::Level1Op benchmark_op =
+    blas_benchmark::utils::Level1Op::rotm;
 
 template <typename scalar_t, typename... args_t>
 static inline void cublas_routine(args_t&&... args) {
@@ -160,8 +155,11 @@ void register_benchmark(blas_benchmark::Args& args,
                          index_t size, bool* success) {
       run<scalar_t>(st, cuda_handle_ptr, size, success);
     };
-    benchmark::RegisterBenchmark(get_name<scalar_t>(size).c_str(), BM_lambda,
-                                 cuda_handle_ptr, size, success)
+    benchmark::RegisterBenchmark(
+        blas_benchmark::utils::get_name<benchmark_op, scalar_t>(
+            size, blas_benchmark::utils::MEM_TYPE_USM)
+            .c_str(),
+        BM_lambda, cuda_handle_ptr, size, success)
         ->UseRealTime();
   }
 }

diff --git a/benchmark/cublas/blas1/rotmg.cpp b/benchmark/cublas/blas1/rotmg.cpp
@@ -26,13 +26,8 @@
 #include "../utils.hpp"
 #include "common/float_comparison.hpp"
 
-template <typename scalar_t>
-std::string get_name() {
-  std::ostringstream str{};
-  str << "BM_Rotmg<" << blas_benchmark::utils::get_type_name<scalar_t>()
-      << ">/";
-  return str.str();
-}
+constexpr blas_benchmark::utils::Level1Op benchmark_op =
+    blas_benchmark::utils::Level1Op::rotmg;
 
 template <typename scalar_t, typename... args_t>
 static inline void cublas_routine(args_t&&... args) {
@@ -164,8 +159,11 @@ void register_benchmark(blas_benchmark::Args& args,
                        bool* success) {
     run<scalar_t>(st, cuda_handle_ptr, success);
   };
-  benchmark::RegisterBenchmark(get_name<scalar_t>().c_str(), BM_lambda,
-                               cuda_handle_ptr, success)
+  benchmark::RegisterBenchmark(
+      blas_benchmark::utils::get_name<benchmark_op, scalar_t>(
+          blas_benchmark::utils::MEM_TYPE_USM)
+          .c_str(),
+      BM_lambda, cuda_handle_ptr, success)
       ->UseRealTime();
 }
 

diff --git a/benchmark/cublas/blas1/scal.cpp b/benchmark/cublas/blas1/scal.cpp
@@ -25,13 +25,8 @@
 
 #include "../utils.hpp"
 
-template <typename scalar_t>
-std::string get_name(int size) {
-  std::ostringstream str{};
-  str << "BM_Scal<" << blas_benchmark::utils::get_type_name<scalar_t>() << ">/";
-  str << size;
-  return str.str();
-}
+constexpr blas_benchmark::utils::Level1Op benchmark_op =
+    blas_benchmark::utils::Level1Op::scal;
 
 template <typename scalar_t, typename... args_t>
 static inline void cublas_routine(args_t&&... args) {
@@ -138,8 +133,11 @@ void register_benchmark(blas_benchmark::Args& args,
                          index_t size, bool* success) {
       run<scalar_t>(st, cuda_handle_ptr, size, success);
     };
-    benchmark::RegisterBenchmark(get_name<scalar_t>(size).c_str(), BM_lambda,
-                                 cuda_handle_ptr, size, success)
+    benchmark::RegisterBenchmark(
+        blas_benchmark::utils::get_name<benchmark_op, scalar_t>(
+            size, blas_benchmark::utils::MEM_TYPE_USM)
+            .c_str(),
+        BM_lambda, cuda_handle_ptr, size, success)
         ->UseRealTime();
   }
 }

diff --git a/benchmark/cublas/blas2/gbmv.cpp b/benchmark/cublas/blas2/gbmv.cpp
@@ -25,13 +25,8 @@
 
 #include "../utils.hpp"
 
-template <typename scalar_t>
-std::string get_name(std::string t, int m, int n, int kl, int ku) {
-  std::ostringstream str{};
-  str << "BM_Gbmv<" << blas_benchmark::utils::get_type_name<scalar_t>() << ">/"
-      << t << "/" << m << "/" << n << "/" << kl << "/" << ku;
-  return str.str();
-}
+constexpr blas_benchmark::utils::Level2Op benchmark_op =
+    blas_benchmark::utils::Level2Op::gbmv;
 
 template <typename scalar_t, typename... args_t>
 static inline void cublas_routine(args_t&&... args) {
@@ -166,9 +161,11 @@ void register_benchmark(blas_benchmark::Args& args,
                          scalar_t alpha, scalar_t beta, bool* success) {
       run<scalar_t>(st, cuda_handle_ptr, t, m, n, kl, ku, alpha, beta, success);
     };
-    benchmark::RegisterBenchmark(get_name<scalar_t>(ts, m, n, kl, ku).c_str(),
-                                 BM_lambda, cuda_handle_ptr, t, m, n, kl, ku,
-                                 alpha, beta, success)
+    benchmark::RegisterBenchmark(
+        blas_benchmark::utils::get_name<benchmark_op, scalar_t>(
+            ts, m, n, kl, ku, blas_benchmark::utils::MEM_TYPE_USM)
+            .c_str(),
+        BM_lambda, cuda_handle_ptr, t, m, n, kl, ku, alpha, beta, success)
         ->UseRealTime();
   }
 }

diff --git a/benchmark/cublas/blas2/gemv.cpp b/benchmark/cublas/blas2/gemv.cpp
@@ -25,13 +25,8 @@
 
 #include "../utils.hpp"
 
-template <typename scalar_t>
-std::string get_name(std::string t, int m, int n) {
-  std::ostringstream str{};
-  str << "BM_Gemv<" << blas_benchmark::utils::get_type_name<scalar_t>() << ">/"
-      << t << "/" << m << "/" << n;
-  return str.str();
-}
+constexpr blas_benchmark::utils::Level2Op benchmark_op =
+    blas_benchmark::utils::Level2Op::gemv;
 
 template <typename scalar_t, typename... args_t>
 static inline void cublas_routine(args_t&&... args) {
@@ -164,9 +159,11 @@ void register_benchmark(blas_benchmark::Args& args,
                          scalar_t beta, bool* success) {
       run<scalar_t>(st, cuda_handle_ptr, t, m, n, alpha, beta, success);
     };
-    benchmark::RegisterBenchmark(get_name<scalar_t>(ts, m, n).c_str(),
-                                 BM_lambda, cuda_handle_ptr, t, m, n, alpha,
-                                 beta, success)
+    benchmark::RegisterBenchmark(
+        blas_benchmark::utils::get_name<benchmark_op, scalar_t>(
+            ts, m, n, blas_benchmark::utils::MEM_TYPE_USM)
+            .c_str(),
+        BM_lambda, cuda_handle_ptr, t, m, n, alpha, beta, success)
         ->UseRealTime();
   }
 }