xeptore · xeptore · May 11, 2021 · May 11, 2021 · May 11, 2021 · May 11, 2021
diff --git a/Benchmarks.md b/Benchmarks.md
@@ -0,0 +1,102 @@
+# Benchmark Results
+
+## System Specification
+
+### Hardware
+
+```txt
+Memory: RAM: total: 30.62 GiB
+CPU: Topology: 16-Core (4-Die) model: AMD EPYC (with IBPB) bits: 64 type: MCP MCM arch: Zen rev: 2
+ L2 cache: 8192 KiB
+ flags: avx avx2 lm nx pae sse sse2 sse3 sse4_1 sse4_2 sse4a ssse3 bogomips: 79849
+ Speed: 2495 MHz min/max: N/A Core speeds (MHz): 1: 2495 2: 2495 3: 2495 4: 2495 5: 2495 6: 2495 7: 2495
+ 8: 2495 9: 2495 10: 2495 11: 2495 12: 2495 13: 2495 14: 2495 15: 2495 16: 2495
+```
+
+### Software
+
+```txt
+System: Host: ubuntu-32gb-nbg1-1 Kernel: 5.4.0-72-generic x86_64 bits: 64 compiler: gcc v: 9.3.0 Console: N/A
+ dm: N/A Distro: Ubuntu 20.04.2 LTS (Focal Fossa)
+Compiler: Ubuntu clang version 11.0.0-2~ubuntu20.04.1
+ Target: x86_64-pc-linux-gnu
+ Thread model: posix
+```
+
+## Results
+
+Execution times are the average of processing an image with dimensions **43680px x 4160px** with **100 times** iterations.
+
+### pthread
+
+![pthread benchmark diagram](/benchmark_diagrams/pthread.png)
+
+### OpenMP
+
+![OpenMP benchmark diagram](/benchmark_diagrams/openmp.png)
+
+### pthread vs. OpenMP
+
+![pthread vs. OpenMP 1 workers benchmark diagram](/benchmark_diagrams/01w.png)
+
+![pthread vs. OpenMP 2 workers benchmark diagram](/benchmark_diagrams/02w.png)
+
+![pthread vs. OpenMP 3 workers benchmark diagram](/benchmark_diagrams/03w.png)
+
+![pthread vs. OpenMP 4 workers benchmark diagram](/benchmark_diagrams/04w.png)
+
+![pthread vs. OpenMP 5 workers benchmark diagram](/benchmark_diagrams/05w.png)
+
+![pthread vs. OpenMP 6 workers benchmark diagram](/benchmark_diagrams/06w.png)
+
+![pthread vs. OpenMP 7 workers benchmark diagram](/benchmark_diagrams/07w.png)
+
+![pthread vs. OpenMP 8 workers benchmark diagram](/benchmark_diagrams/08w.png)
+
+![pthread vs. OpenMP 9 workers benchmark diagram](/benchmark_diagrams/09w.png)
+
+![pthread vs. OpenMP 10 workers benchmark diagram](/benchmark_diagrams/10w.png)
+
+![pthread vs. OpenMP 11 workers benchmark diagram](/benchmark_diagrams/11w.png)
+
+![pthread vs. OpenMP 12 workers benchmark diagram](/benchmark_diagrams/12w.png)
+
+![pthread vs. OpenMP 13 workers benchmark diagram](/benchmark_diagrams/13w.png)
+
+![pthread vs. OpenMP 14 workers benchmark diagram](/benchmark_diagrams/14w.png)
+
+![pthread vs. OpenMP 15 workers benchmark diagram](/benchmark_diagrams/15w.png)
+
+![pthread vs. OpenMP 16 workers benchmark diagram](/benchmark_diagrams/16w.png)
+
+![pthread vs. OpenMP 17 workers benchmark diagram](/benchmark_diagrams/17w.png)
+
+![pthread vs. OpenMP 18 workers benchmark diagram](/benchmark_diagrams/18w.png)
+
+![pthread vs. OpenMP 19 workers benchmark diagram](/benchmark_diagrams/19w.png)
+
+![pthread vs. OpenMP 20 workers benchmark diagram](/benchmark_diagrams/20w.png)
+
+![pthread vs. OpenMP 21 workers benchmark diagram](/benchmark_diagrams/21w.png)
+
+![pthread vs. OpenMP 22 workers benchmark diagram](/benchmark_diagrams/22w.png)
+
+![pthread vs. OpenMP 23 workers benchmark diagram](/benchmark_diagrams/23w.png)
+
+![pthread vs. OpenMP 24 workers benchmark diagram](/benchmark_diagrams/24w.png)
+
+![pthread vs. OpenMP 25 workers benchmark diagram](/benchmark_diagrams/25w.png)
+
+![pthread vs. OpenMP 26 workers benchmark diagram](/benchmark_diagrams/26w.png)
+
+![pthread vs. OpenMP 27 workers benchmark diagram](/benchmark_diagrams/27w.png)
+
+![pthread vs. OpenMP 28 workers benchmark diagram](/benchmark_diagrams/28w.png)
+
+![pthread vs. OpenMP 29 workers benchmark diagram](/benchmark_diagrams/29w.png)
+
+![pthread vs. OpenMP 30 workers benchmark diagram](/benchmark_diagrams/30w.png)
+
+![pthread vs. OpenMP 31 workers benchmark diagram](/benchmark_diagrams/31w.png)
+
+![pthread vs. OpenMP 32 workers benchmark diagram](/benchmark_diagrams/32w.png)
diff --git a/CMakeLists.txt b/CMakeLists.txt
@@ -4,6 +4,10 @@ set(CMAKE_C_COMPILER /usr/bin/clang)
 set(CMAKE_CXX_COMPILER /usr/bin/clang++)
 set(CMAKE_EXPORT_COMPILE_COMMANDS true)
 
+set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -fopenmp")
+set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -fopenmp")
+set(CMAKE_EXE_LINKER_FLAGS "${CMAKE_EXE_LINKER_FLAGS} -fopenmp")
+
 project(blurrifier VERSION 0.1.0)
 
 add_executable(blurrifier main.c)

diff --git a/benchmark_diagrams/01w.png b/benchmark_diagrams/01w.png
diff --git a/benchmark_diagrams/02w.png b/benchmark_diagrams/02w.png
diff --git a/benchmark_diagrams/03w.png b/benchmark_diagrams/03w.png
diff --git a/benchmark_diagrams/04w.png b/benchmark_diagrams/04w.png
diff --git a/benchmark_diagrams/05w.png b/benchmark_diagrams/05w.png
diff --git a/benchmark_diagrams/06w.png b/benchmark_diagrams/06w.png
diff --git a/benchmark_diagrams/07w.png b/benchmark_diagrams/07w.png
diff --git a/benchmark_diagrams/08w.png b/benchmark_diagrams/08w.png
diff --git a/benchmark_diagrams/09w.png b/benchmark_diagrams/09w.png
diff --git a/benchmark_diagrams/10w.png b/benchmark_diagrams/10w.png
diff --git a/benchmark_diagrams/11w.png b/benchmark_diagrams/11w.png
diff --git a/benchmark_diagrams/12w.png b/benchmark_diagrams/12w.png
diff --git a/benchmark_diagrams/13w.png b/benchmark_diagrams/13w.png
diff --git a/benchmark_diagrams/14w.png b/benchmark_diagrams/14w.png
diff --git a/benchmark_diagrams/15w.png b/benchmark_diagrams/15w.png
diff --git a/benchmark_diagrams/16w.png b/benchmark_diagrams/16w.png
diff --git a/benchmark_diagrams/17w.png b/benchmark_diagrams/17w.png
diff --git a/benchmark_diagrams/18w.png b/benchmark_diagrams/18w.png
diff --git a/benchmark_diagrams/19w.png b/benchmark_diagrams/19w.png
diff --git a/benchmark_diagrams/20w.png b/benchmark_diagrams/20w.png
diff --git a/benchmark_diagrams/21w.png b/benchmark_diagrams/21w.png
diff --git a/benchmark_diagrams/22w.png b/benchmark_diagrams/22w.png
diff --git a/benchmark_diagrams/23w.png b/benchmark_diagrams/23w.png
diff --git a/benchmark_diagrams/24w.png b/benchmark_diagrams/24w.png
diff --git a/benchmark_diagrams/25w.png b/benchmark_diagrams/25w.png
diff --git a/benchmark_diagrams/26w.png b/benchmark_diagrams/26w.png
diff --git a/benchmark_diagrams/27w.png b/benchmark_diagrams/27w.png
diff --git a/benchmark_diagrams/28w.png b/benchmark_diagrams/28w.png
diff --git a/benchmark_diagrams/29w.png b/benchmark_diagrams/29w.png
diff --git a/benchmark_diagrams/30w.png b/benchmark_diagrams/30w.png
diff --git a/benchmark_diagrams/31w.png b/benchmark_diagrams/31w.png
diff --git a/benchmark_diagrams/32w.png b/benchmark_diagrams/32w.png
diff --git a/benchmark_diagrams/openmp.png b/benchmark_diagrams/openmp.png
diff --git a/benchmark_diagrams/pthread.png b/benchmark_diagrams/pthread.png
diff --git a/diagram.py b/diagram.py
@@ -1,16 +1,58 @@
 import plotly.express as px
+import plotly.io as pio
 import pandas as pd
 
-df = pd.read_csv("./results.csv")
+pio.kaleido.scope.default_width = 700 * 1
+pio.kaleido.scope.default_height = 500 * 1
+pio.kaleido.scope.default_width = 900 * 1
+pio.kaleido.scope.default_height = 1200 * 1
+pio.kaleido.scope.default_scale = 5
 
-df['time'] = df['time'] * 1e-9
+pthread_dataframe = pd.read_csv("./results.csv")
 
-fig = px.line(
- df,
- title="Kernel Radius Per Number of Workers Processing Times",
+pthread_dataframe['time'] = pthread_dataframe['time'] * 1e-9
+
+pthread_fig = px.line(
+ pthread_dataframe,
+ title="Processing Time For Different Kernel Radii (pthread)",
  x="radius",
  y="time",
- color="workers",
- labels={ "time": "Duration (seconds)", "radius": "Kernel Radius" }
+ color="Workers",
+ labels={"time": "Duration (seconds)", "radius": "Kernel Radius"},
+ range_x=[2, 20],
+ range_y=[0, 700]
 )
-fig.show()
+pthread_fig.write_image(f"./benchmark_diagrams/pthread.png")
+
+pthread_dataframe["Threading"] = "pthread"
+
+openmp_dataframe = pd.read_csv("./openmp.csv")
+openmp_dataframe['time'] = openmp_dataframe['time'] * 1e-9
+
+openmp_fig = px.line(
+ openmp_dataframe,
+ title=f"Processing Time For Different Kernel Radii (OpenMP)",
+ x="radius",
+ y="time",
+ color="Workers",
+ labels={"time": "Duration (seconds)", "radius": "Kernel Radius"},
+ range_x=[2, 20],
+ range_y=[0, 700]
+)
+openmp_fig.write_image(f"./benchmark_diagrams/openmp.png")
+
+openmp_dataframe["Threading"] = "OpenMP"
+
+for i in range(32):
+ fig = px.line(
+ pd.concat([pthread_dataframe.where(pthread_dataframe["Workers"] == i + 1).loc[i*10:(i+1) * 10 - 1],
+ openmp_dataframe.where(openmp_dataframe["Workers"] == i + 1).loc[i*10:(i+1) * 10 - 1]]),
+ title=f"Processing Time For Different Kernel Radii Using {i + 1} Worker{'s' if i > 0 else ''}",
+ x="radius",
+ y="time",
+ color="Threading",
+ labels={"time": "Duration (seconds)", "radius": "Kernel Radius"},
+ range_x=[2, 20],
+ range_y=[0, 700]
+ )
+ fig.write_image(f"./benchmark_diagrams/{i + 1:02}w.png")
diff --git a/main.c b/main.c
@@ -5,6 +5,7 @@
 #include <memory.h>
 #include <jpeglib.h>
 #include <math.h>
+#include <omp.h>
 #include <pthread.h>
 #include <sys/time.h>
 #include "config.h"
@@ -114,9 +115,7 @@ void copy_kernel(double destination[KERNEL_HEIGHT][KERNEL_WIDTH], const double s
  };
 }
 
-void *transform_rows(void *serialized_params) {
- struct transform_row_params *params = (struct transform_row_params *)serialized_params;
-
+void transform_rows(struct transform_row_params *params) {
  for (size_t i = params->start_row; i < params->start_row + params->num_rows; i++) {
  for (size_t j = 0; j < params->IMAGE_WIDTH; j++) {
  struct pixel_components components_multiplication_sum = {
@@ -145,8 +144,6 @@ void *transform_rows(void *serialized_params) {
  params->output_image[i][INPUT_IMAGE_COMPONENTS_NUMBER * j + 2] = round(components_multiplication_sum.blue / kernel_cells_sum);
  }
  }
-
- return NULL;
 }
 
 int transform(
@@ -215,6 +212,12 @@ int transform(
  const unsigned int remainder = decompressor.image_height % NUM_THREADS;
 
  unsigned long total_assigned_rows = 0U;
+
+ struct timespec start_time, end;
+
+ timespec_get(&start_time, TIME_UTC);
+
+ #pragma omp parallel for
  for (size_t i = 0; i < NUM_THREADS; i++) {
  const unsigned long int worker_quotient = (i < remainder) ? (quotient + 1) : (quotient);
  struct transform_row_params *params = (struct transform_row_params *)&buffer[2 * IMAGE_SIZE_IN_BYTES + i * sizeof(struct transform_row_params)];
@@ -225,25 +228,13 @@ int transform(
  params->IMAGE_WIDTH = IMAGE_WIDTH;
  params->num_rows = worker_quotient;
  params->start_row = total_assigned_rows;
- thread_params_refs[i] = params;
  total_assigned_rows += worker_quotient;
- }
-
- struct timespec start, end;
-
- timespec_get(&start, TIME_UTC);
-
- for (size_t i = 0; i < NUM_THREADS; i++) {
- (void)pthread_create(&thread_ids[i], NULL, transform_rows, thread_params_refs[i]);
- }
-
- for (size_t i = 0; i < NUM_THREADS; i++) {
- (void)pthread_join(thread_ids[i], NULL);
+ transform_rows(params);
  }
 
  timespec_get(&end, TIME_UTC);
 
- unsigned long int time_in_nano_seconds = (end.tv_sec - start.tv_sec) * 1e9 + (end.tv_nsec - start.tv_nsec);
+ unsigned long int time_in_nano_seconds = (end.tv_sec - start_time.tv_sec) * 1e9 + (end.tv_nsec - start_time.tv_nsec);
  printf("total:%lu", time_in_nano_seconds);
 
  while (compressor.next_scanline < compressor.image_height) {
@@ -266,6 +257,8 @@ int transform(
 }
 
 int main() {
+ omp_set_num_threads(NUM_THREADS);
+
  return transform(
  INPUT_IMAGE_FILENAME,
  OUTPUT_IMAGE_FILENAME,