Review update

- fix gpu_timer in script
- use int64_t
- add const to function

Co-authored-by: Aditya Kashi <aditya.kashi@kit.edu>
Co-authored-by: Pratik Nayak <pratikvn@protonmail.com>
Co-authored-by: Thomas Grützmacher <thomas.gruetzmacher@kit.edu>

benchmark/run_all_benchmarks.sh

@@ -149,7 +149,7 @@ run_conversion_benchmarks() {
     cp "$1" "$1.imd" # make sure we're not loosing the original input
     ./conversions/conversions --backup="$1.bkp" --double_buffer="$1.bkp2" \
                 --executor="${EXECUTOR}" --formats="${FORMATS}" \
-                --device_id="${DEVICE_ID}" --gpu_time=${GPU_TIMER} \
+                --device_id="${DEVICE_ID}" --gpu_timer=${GPU_TIMER} \
                 <"$1.imd" 2>&1 >"$1"
     keep_latest "$1" "$1.bkp" "$1.bkp2" "$1.imd"
 }
@@ -165,7 +165,7 @@ run_spmv_benchmarks() {
     cp "$1" "$1.imd" # make sure we're not loosing the original input
     ./spmv/spmv --backup="$1.bkp" --double_buffer="$1.bkp2" \
                 --executor="${EXECUTOR}" --formats="${FORMATS}" \
-                --device_id="${DEVICE_ID}" --gpu_time=${GPU_TIMER} \
+                --device_id="${DEVICE_ID}" --gpu_timer=${GPU_TIMER} \
                 <"$1.imd" 2>&1 >"$1"
     keep_latest "$1" "$1.bkp" "$1.bkp2" "$1.imd"
 }
@@ -183,7 +183,7 @@ run_solver_benchmarks() {
                     --executor="${EXECUTOR}" --solvers="${SOLVERS}" \
                     --preconditioners="${PRECONDS}" \
                     --max_iters=${SOLVERS_MAX_ITERATIONS} --rel_res_goal=${SOLVERS_PRECISION} \
-                    ${SOLVERS_RHS_FLAG} ${DETAILED_STR} --device_id="${DEVICE_ID}" --gpu_time=${GPU_TIMER} \
+                    ${SOLVERS_RHS_FLAG} ${DETAILED_STR} --device_id="${DEVICE_ID}" --gpu_timer=${GPU_TIMER} \
                     <"$1.imd" 2>&1 >"$1"
     keep_latest "$1" "$1.bkp" "$1.bkp2" "$1.imd"
 }
@@ -209,7 +209,7 @@ run_preconditioner_benchmarks() {
                 --executor="${EXECUTOR}" --preconditioners="jacobi" \
                 --jacobi_max_block_size="${bsize}" \
                 --jacobi_storage="${prec}" \
-                --device_id="${DEVICE_ID}" --gpu_time=${GPU_TIMER} \
+                --device_id="${DEVICE_ID}" --gpu_timer=${GPU_TIMER} \
                 <"$1.imd" 2>&1 >"$1"
             keep_latest "$1" "$1.bkp" "$1.bkp2" "$1.imd"
         done

benchmark/utils/timer.hpp

@@ -62,7 +62,7 @@ OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #include "hip/base/device_guard.hip.hpp"
 
 
-#endif  // HAS_CUDA
+#endif  // HAS_HIP
 
 
 DEFINE_bool(gpu_timer, false,
@@ -79,7 +79,7 @@ class Timer {
         tic_called_ = true;
     }
 
-    std::size_t toc()
+    std::int64_t toc()
     {
         assert(tic_called_ == true);
         auto ns = this->toc_impl();
@@ -88,16 +88,18 @@ class Timer {
         return ns;
     }
 
-    std::size_t get_total_time() { return total_duration_ns_; }
+    std::int64_t get_total_time() const { return total_duration_ns_; }
 
-    std::size_t get_tictoc_num() { return duration_ns_.size(); }
+    std::int64_t get_num_repetitions() const { return duration_ns_.size(); }
 
-    double get_average_time()
+    double get_average_time() const
     {
         return static_cast<double>(this->get_total_time()) /
-               this->get_tictoc_num();
+               this->get_num_repetitions();
     }
 
+    std::vector<std::int64_t> get_time_detail() const{ return duration_ns_; }
+
     void clear()
     {
         duration_ns_.clear();
@@ -108,7 +110,7 @@ class Timer {
     Timer() : tic_called_(false), total_duration_ns_(0) {}
 
 protected:
-    void add_record(std::size_t ns)
+    void add_record(std::int64_t ns)
     {
         // add the result;
         duration_ns_.emplace_back(ns);
@@ -117,12 +119,12 @@ class Timer {
 
     virtual void tic_impl() = 0;
 
-    virtual std::size_t toc_impl() = 0;
+    virtual std::int64_t toc_impl() = 0;
 
 private:
-    std::vector<std::size_t> duration_ns_;
+    std::vector<std::int64_t> duration_ns_;
     bool tic_called_;
-    std::size_t total_duration_ns_;
+    std::int64_t total_duration_ns_;
 };
 
 
@@ -138,14 +140,14 @@ class CpuTimer : public Timer {
         start_ = std::chrono::steady_clock::now();
     }
 
-    std::size_t toc_impl() override
+    std::int64_t toc_impl() override
     {
         exec_->synchronize();
         auto stop = std::chrono::steady_clock::now();
         auto duration_time =
             std::chrono::duration_cast<std::chrono::nanoseconds>(stop - start_)
                 .count();
-        return static_cast<std::size_t>(duration_time);
+        return static_cast<std::int64_t>(duration_time);
     }
 
 private:
@@ -181,7 +183,7 @@ class CudaTimer : public Timer {
         GKO_ASSERT_NO_CUDA_ERRORS(cudaEventRecord(start_));
     }
 
-    std::size_t toc_impl() override
+    std::int64_t toc_impl() override
     {
         gko::cuda::device_guard g{id_};
         // Currently, gko::CudaExecutor always use default stream.
@@ -192,7 +194,7 @@ class CudaTimer : public Timer {
         // resolution of around 0.5 microseconds
         GKO_ASSERT_NO_CUDA_ERRORS(
             cudaEventElapsedTime(&duration_time, start_, stop_));
-        return static_cast<std::size_t>(duration_time * 1e6);
+        return static_cast<std::int64_t>(duration_time * 1e6);
     }
 
 private:
@@ -233,7 +235,7 @@ class HipTimer : public Timer {
         GKO_ASSERT_NO_HIP_ERRORS(hipEventRecord(start_));
     }
 
-    std::size_t toc_impl() override
+    std::int64_t toc_impl() override
     {
         gko::hip::device_guard g{id_};
         // Currently, gko::HipExecutor always use default stream.
@@ -244,7 +246,7 @@ class HipTimer : public Timer {
         // resolution of around 0.5 microseconds
         GKO_ASSERT_NO_HIP_ERRORS(
             hipEventElapsedTime(&duration_time, start_, stop_));
-        return static_cast<std::size_t>(duration_time * 1e6);
+        return static_cast<std::int64_t>(duration_time * 1e6);
     }
 
 private: