Merge branch 'master' of gitlab-gepasp.in.tum.de:gra22s/team199

.vscode/tasks.json

@@ -9,8 +9,11 @@
                 "-g",
                 "${fileDirname}/*.c",
                 "${fileDirname}/md2_impls/*.c",
+                "${fileDirname}/md2_impls/md2_reference/*.c",
                 "-o",
-                "${fileDirname}/../md2"
+                "${fileDirname}/../md2",
+                "-D",
+                "MD2_DETAILED_BENCHMARK"
             ],
             "options": {
                 "cwd": "${fileDirname}"

Implementierung/Makefile

@@ -3,7 +3,11 @@
 SRC = src/main.c src/helper.c src/io.c src/md2.c src/md2_impls/md2_common.c src/md2_impls/md2_0.c src/md2_impls/md2_1.c src/md2_impls/md2_2.c src/md2_impls/md2_3.c src/md2_impls/md2_reference/md2_reference.c
 OBJ = ${subst src,build,${SRC:.c=.o}}
 CC = gcc
-CFLAGS = -Ilib -ggdb -std=c11 -g -Wall -Wextra -no-pie -O3 
+DETAILED_BENCHMARK ?= 0
+CFLAGS = -Ilib -std=c11 -Wall -Wextra -O3
+ifeq ($(DETAILED_BENCHMARK), true)
+	CFLAGS += -DMD2_DETAILED_BENCHMARK
+endif
 LDFLAGS = -pthread
 #TESTFILES = t/1 t/2 t/5 t/7 t/10 t/25 t/50 t/75 t/100 t/250 t/500 t/750 t/1000 t/1500 t/2000 t/2500 t/3000 t/3500 t/4000 t/4500 t/5000 t/5500 
 TESTFILES = t/6000 t/6500 t/7000 t/7500 t/8000 t/8500 t/9000 t/9500 t/10000
@@ -23,9 +27,12 @@ help:
 	@echo
 	@echo Available targets:
 	@echo - all: build everything
+	@echo "       When DETAILED_BENCHMARK=true is passed, the detailed benchmarks are enabled."
+	@echo "       This will decrease the overall performance of the program."
 	@echo - clean: clean distfiles
 	@echo - help: show this help
-	@echo - benchmarks: run benchmarks (only works on linux!)
+	@echo - benchmarks: run benchmarks
+	@echo - valgrind-check: run checks with valgrind
 
 build/%.o: src/%.c
 	@mkdir -p build/md2_impls/md2_reference
@@ -73,7 +80,7 @@ benchmarks.csv: md2 ${TESTFILES}
 
 benchmarks: benchmarks.csv
 
-lint: md2 t/3
+valgrind-check: md2 t/3
 
 	@for i in 0 1 2 3 4; do \
 		echo; \

Implementierung/lib/helper.h

@@ -36,6 +36,8 @@ struct configuration {
 enum argumentParseResult parseArguments(int argc, char** argv,
                                         struct configuration* c);
 
+double current_time();
+
 /**
  * @brief Run an md2_hash_func with benchmark timing
  *

Implementierung/lib/md2_impls/md2_common.h

@@ -9,6 +9,42 @@
 #include <string.h>
 #include <sys/types.h>
 
+#include "../helper.h"
+
+#ifdef MD2_DETAILED_BENCHMARK
+#define md2_process_detailed_benchmark_step_if_defined(step) \
+  md2_process_detailed_benchmark_step(step);
+
+#define md2_print_detailed_benchmark_result_if_defined \
+  md2_print_detailed_benchmark_result();
+#else
+#define md2_process_detailed_benchmark_step_if_defined(step)
+#define md2_print_detailed_benchmark_result_if_defined
+#endif  // MD2_DETAILED_BENCHMARK
+
+#define CHECKSUM_START_MARK \
+  md2_process_detailed_benchmark_step_if_defined(CHECKSUM_START)
+#define CHECKSUM_END_MARK \
+  md2_process_detailed_benchmark_step_if_defined(CHECKSUM_END)
+#define FIRST_LOOP_START_MARK \
+  md2_process_detailed_benchmark_step_if_defined(FIRST_LOOP_START)
+#define FIRST_LOOP_END_MARK \
+  md2_process_detailed_benchmark_step_if_defined(FIRST_LOOP_END)
+#define SECOND_LOOP_START_MARK \
+  md2_process_detailed_benchmark_step_if_defined(SECOND_LOOP_START)
+#define SECOND_LOOP_END_MARK \
+  md2_process_detailed_benchmark_step_if_defined(SECOND_LOOP_END)
+#define END_MARK md2_print_detailed_benchmark_result_if_defined
+
+enum md2_detailed_benchmark_step {
+  CHECKSUM_START = 0,
+  CHECKSUM_END,
+  FIRST_LOOP_START,
+  FIRST_LOOP_END,
+  SECOND_LOOP_START,
+  SECOND_LOOP_END
+};
+
 /**
  * @brief Some digits of pi
  *
@@ -23,4 +59,8 @@ extern unsigned char MD2_PI_SUBST[256];
  */
 void md2_print_buf(size_t len, uint8_t buf[len]);
 
+void md2_process_detailed_benchmark_step(enum md2_detailed_benchmark_step step);
+
+void md2_print_detailed_benchmark_result();
+
 #endif  // MD2_COMMON_H

Implementierung/src/helper.c

@@ -30,8 +30,13 @@ void help(char *progname) {
   1: optimized implementation\n\
   2: memory efficient implementation\n\
   3: threaded implementation\n\
-  4: reference implementation\n\
+  4: reference implementation\n"
-  \n",
+#ifdef MD2_DETAILED_BENCHMARK
+          "\n\
+  \n\033[1;33mWARNING: Detailed benchmarking is enabled. This will show detailed benchmaring results\n\
+         after each hash calculation but will decrease the overall performance.\033[0m"
+#endif  // MD2_DETAILED_BENCHMARK
+          "\n",
           progname);
 }
 
@@ -103,7 +108,7 @@ enum argumentParseResult parseArguments(int argc, char **argv,
   return RESULT_OK;
 }
 
-double current_time(void) {
+double current_time() {
   struct timespec t;
   clock_gettime(CLOCK_MONOTONIC, &t);
   return t.tv_sec + t.tv_nsec * 1e-9;

Implementierung/src/md2_impls/md2_0.c

@@ -40,7 +40,9 @@ void md2_hash_0(size_t len, const uint8_t buf[len], uint8_t out[16]) {
   // printBuf(len + 16, newBuf);
 
   // === step 2 ===
+  CHECKSUM_START_MARK
   md2_checksum_0(len, newBuf);
+  CHECKSUM_END_MARK
 
   // printf("buf with cecksum: ");
   // printBuf(len + 16, newBuf);
@@ -51,28 +53,32 @@ void md2_hash_0(size_t len, const uint8_t buf[len], uint8_t out[16]) {
     return;
   }
 
-
   // === step 4 ===
   // <= because we need to hash the last block too
   for (size_t i = 0; i <= (len + 16) / 16 - 1; i++) {
+    FIRST_LOOP_START_MARK
     for (int j = 0; j < 16; j++) {
       messageDigestBuf[16 + j] = newBuf[i * 16 + j];
       messageDigestBuf[32 + j] =
           (messageDigestBuf[16 + j] ^ messageDigestBuf[j]);
     }
+    FIRST_LOOP_END_MARK
 
     u_int8_t t = 0;
 
+    SECOND_LOOP_START_MARK
     for (int j = 0; j < 18; j++) {
       for (int k = 0; k < 48; k++) {
         t = messageDigestBuf[k] = messageDigestBuf[k] ^ MD2_PI_SUBST[t];
       }
       t = (t + j) % 256;
     }
+    SECOND_LOOP_END_MARK
   }
   // printf("messageDigestBuf: \n");
   // printBuf(16, messageDigestBuf);
 
+  END_MARK
   memcpy(out, messageDigestBuf, 16);
 
   free(messageDigestBuf);

Implementierung/src/md2_impls/md2_1.c

@@ -1,17 +1,14 @@
 #include "../../lib/md2_impls/md2_1.h"
 
-#include "../../lib/md2_impls/md2_common.h"
 #include <immintrin.h>
 
+#include "../../lib/md2_impls/md2_common.h"
 
-void md2_checksum_1(size_t len, uint8_t *buf)
+void md2_checksum_1(size_t len, uint8_t *buf) {
-{
   uint8_t l = 0;
 
-  for (size_t i = 0; i < len / 16; i++)
+  for (size_t i = 0; i < len / 16; i++) {
-  {
+    for (int j = 0; j < 16; j++) {
-    for (int j = 0; j < 16; j++)
-    {
       u_int8_t c = buf[i * 16 + j];
       // reference is wrong. It says: Set C[j] to S[c xor L]. But it should be:
       buf[len + j] ^= MD2_PI_SUBST[c ^ l];
@@ -39,18 +36,17 @@ static uint8_t PADDING[17][16] = {
     {15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 0},
     {16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16}};
 
-void md2_hash_1(size_t len, const uint8_t buf[len], uint8_t out[16])
+void md2_hash_1(size_t len, const uint8_t buf[len], uint8_t out[16]) {
-{
   // === step 1 ===
   int paddingNeeded = 16 - (len & 15);
   len += paddingNeeded;
 
-  uint8_t* newBuf = aligned_alloc(16, sizeof(uint8_t)*(len + 16));
+  uint8_t *newBuf = aligned_alloc(16, sizeof(uint8_t) * (len + 16));
   if (newBuf == NULL) {
     return;
   }
 
-  for(size_t i = 0; i < 16; i++) {
+  for (size_t i = 0; i < 16; i++) {
     newBuf[len + i] = 0;
   }
 
@@ -59,7 +55,9 @@ void md2_hash_1(size_t len, const uint8_t buf[len], uint8_t out[16])
   memcpy(newBuf + len - paddingNeeded, PADDING + paddingNeeded, paddingNeeded);
 
   // === step 2 ===
+  CHECKSUM_START_MARK
   md2_checksum_1(len, newBuf);
+  CHECKSUM_END_MARK
 
   // === step 3 ===
   uint8_t *messageDigestBuf = aligned_alloc(16, sizeof(uint8_t) * 48);
@@ -74,27 +72,28 @@ void md2_hash_1(size_t len, const uint8_t buf[len], uint8_t out[16])
   // === step 4 ===
   __m128i vx;
   __m128i vy;
-  for (size_t i = 0; i <= (len + 16) / 16 - 1; i++)
+  for (size_t i = 0; i <= (len + 16) / 16 - 1; i++) {
-  {
+    FIRST_LOOP_START_MARK
-    vx = _mm_load_si128((__m128i*) (newBuf + i * 16));
+    vx = _mm_load_si128((__m128i *)(newBuf + i * 16));
-    _mm_store_si128((__m128i*) (messageDigestBuf + 16), vx);
+    _mm_store_si128((__m128i *)(messageDigestBuf + 16), vx);
-    vy = _mm_load_si128((__m128i*) (messageDigestBuf));
+    vy = _mm_load_si128((__m128i *)(messageDigestBuf));
     vy = _mm_xor_si128(vy, vx);
-    _mm_store_si128((__m128i*) (messageDigestBuf + 32), vy);
+    _mm_store_si128((__m128i *)(messageDigestBuf + 32), vy);
-  
+    FIRST_LOOP_END_MARK
 
     u_int8_t t = 0;
 
-    for (int j = 0; j < 18; j++)
+    SECOND_LOOP_START_MARK
-    {
+    for (int j = 0; j < 18; j++) {
-      for (int k = 0; k < 48; k++)
+      for (int k = 0; k < 48; k++) {
-      {
         t = messageDigestBuf[k] = messageDigestBuf[k] ^ MD2_PI_SUBST[t];
       }
       t = (t + j) & 255;
     }
+    SECOND_LOOP_END_MARK
   }
 
+  END_MARK
   memcpy(out, messageDigestBuf, 16);
 
   free(messageDigestBuf);

Implementierung/src/md2_impls/md2_2.c

@@ -3,19 +3,23 @@
 #include "../../lib/md2_impls/md2_common.h"
 
 void process_block_hash(uint8_t block[16], uint8_t messageDigestBuf[48]) {
+  FIRST_LOOP_START_MARK
   for (int j = 0; j < 16; j++) {
     messageDigestBuf[16 + j] = block[j];
     messageDigestBuf[32 + j] = (messageDigestBuf[16 + j] ^ messageDigestBuf[j]);
   }
+  FIRST_LOOP_END_MARK
 
   u_int8_t t = 0;
 
+  SECOND_LOOP_START_MARK
   for (int j = 0; j < 18; j++) {
     for (int k = 0; k < 48; k++) {
       t = messageDigestBuf[k] = messageDigestBuf[k] ^ MD2_PI_SUBST[t];
     }
     t = (t + j) % 256;
   }
+  SECOND_LOOP_END_MARK
 }
 
 void process_block_checksum(uint8_t block[16], uint8_t checksum[16],
@@ -74,21 +78,24 @@ void md2_hash_2(size_t len, const uint8_t buf[len], uint8_t out[16]) {
       return;
     }
 
+    CHECKSUM_START_MARK
     process_block_checksum(data, checksum, &l);
+    CHECKSUM_END_MARK
     process_block_hash(data, messageDigestBuf);
 
     bytes_left_to_read -= bytes_left_to_process;
   };
 
-  fclose(file);
-
   apply_padding(bytes_left_to_process % 16, data);
   process_block_checksum(data, checksum, &l);
   process_block_hash(data, messageDigestBuf);
 
   process_block_hash(checksum, messageDigestBuf);
+
+  END_MARK
   memcpy(out, messageDigestBuf, 16);
 
+  fclose(file);
   free(data);
   free(messageDigestBuf);
   free(checksum);

Implementierung/src/md2_impls/md2_3.c

@@ -10,20 +10,24 @@ struct thread_args {
 void process_nothread_hash(size_t len, const uint8_t buf[len],
                            uint8_t messageDigestBuf[48]) {
   for (size_t i = 0; i < (len + 16) / 16 - 1; i++) {
+    FIRST_LOOP_START_MARK
     for (int j = 0; j < 16; j++) {
       messageDigestBuf[16 + j] = buf[i * 16 + j];
       messageDigestBuf[32 + j] =
           (messageDigestBuf[16 + j] ^ messageDigestBuf[j]);
     }
+    FIRST_LOOP_END_MARK
 
     u_int8_t t = 0;
 
+    SECOND_LOOP_START_MARK
     for (int j = 0; j < 18; j++) {
       for (int k = 0; k < 48; k++) {
         t = messageDigestBuf[k] = messageDigestBuf[k] ^ MD2_PI_SUBST[t];
       }
       t = (t + j) % 256;
     }
+    SECOND_LOOP_END_MARK
   }
 }
 
@@ -48,7 +52,7 @@ void* process_checksum(void* threasdArgs) {
   }
 
   uint8_t l = 0;
-
+  CHECKSUM_START_MARK
   for (size_t i = 0; i < args->len / 16; i++) {
     for (int j = 0; j < 16; j++) {
       u_int8_t c = args->buf[i * 16 + j];
@@ -56,6 +60,7 @@ void* process_checksum(void* threasdArgs) {
       l = checksum[j] ^= MD2_PI_SUBST[c ^ l];
     }
   }
+  CHECKSUM_END_MARK
   pthread_exit(checksum);
 }
 
@@ -107,6 +112,8 @@ void md2_hash_3(size_t len, const uint8_t buf[len], uint8_t out[16]) {
   }
 
   process_nothread_hash(16, checksum, messageDigestBuf);
+
+  END_MARK
   memcpy(out, messageDigestBuf, 16);
 
   free(messageDigestBuf);

Implementierung/src/md2_impls/md2_common.c

@@ -27,3 +27,34 @@ void md2_print_buf(size_t len, uint8_t buf[len]) {
   }
   printf("\n");
 }
+
+static double detailed_benchmark_timestamps[6] = {0, 0, 0, 0, 0, 0};
+static double detailed_benchmark_times[3] = {0, 0, 0};
+
+void md2_process_detailed_benchmark_step(
+    enum md2_detailed_benchmark_step step) {
+  switch (step) {
+    case CHECKSUM_START:
+    case FIRST_LOOP_START:
+    case SECOND_LOOP_START:
+      detailed_benchmark_timestamps[step] = current_time();
+      return;
+
+    case CHECKSUM_END:
+    case FIRST_LOOP_END:
+    case SECOND_LOOP_END:
+      detailed_benchmark_timestamps[step] = current_time();
+      detailed_benchmark_times[step / 2] +=
+          detailed_benchmark_timestamps[step] -
+          detailed_benchmark_timestamps[step - 1];
+      return;
+  }
+}
+
+void md2_print_detailed_benchmark_result() {
+  printf("Detailed benchmarking results:\n");
+  printf("  Checksum: %f\n", detailed_benchmark_times[0]);
+  printf("  First loop: %f\n", detailed_benchmark_times[1]);
+  printf("  Second loop: %f\n", detailed_benchmark_times[2]);
+  printf("\n");
+}