matrix fixed

perflab/matrix/Makefile

@@ -0,0 +1,34 @@
+CC = gcc
+CFLAGS = -Wall -O1 -g
+#LDFLAGS = -lm -lcudart -lcuda
+
+# Source files
+SRCS = rowcol_test.c clock.c cpe.c fcyc.c lsquare.c rowcol_202302723005.c
+#CUDA_SRCS = rowcol.cu
+OBJS = $(SRCS:.c=.o) 
+#rowcol.o
+
+# Target executable
+TARGET = matrix_test
+
+# Default target
+all: $(TARGET)
+
+# Rule to build the executable
+$(TARGET): $(OBJS)
+	$(CC) $(OBJS) -o $(TARGET) $(LDFLAGS)
+
+# Rule to build object files
+%.o: %.c
+	$(CC) $(CFLAGS) -c $< -o $@
+
+# Rule to build CUDA object files
+#rowcol.o: rowcol.cu
+#	$(NVCC) $(CUDA_FLAGS) -c $< -o $@
+
+# Clean rule
+clean:
+	rm -f $(OBJS) $(TARGET)
+
+# Phony targets
+.PHONY: all clean 

perflab/matrix/clock.c

@@ -1,229 +1,196 @@
-/* clock.c
- * Retrofitted to use thread-specific timers
- * and to get clock information from /proc/cpuinfo
- * (C) R. E. Bryant, 2010
- *
- */
-
-/* When this constant is not defined, uses time stamp counter */
-#define USE_POSIX 0
-
-/* Choice to use cpu_gettime call or Intel time stamp counter directly */
-
-#include <stdio.h>
-#include <stdlib.h>
-#include <string.h>
-#include <intrin.h>
-//#include <intrinsics.h>
-#include <windows.h>
-#include <time.h>
-#include "clock.h"
-
-/* Use x86 cycle counter */
-
-/* Initialize the cycle counter */
-static unsigned cyc_hi = 0;
-static unsigned cyc_lo = 0;
-
-/* Set *hi and *lo to the high and low order bits  of the cycle counter.
-   Implementation requires assembly code to use the rdtsc instruction. */
-void access_counter(unsigned *hi, unsigned *lo)
-{
-
-	long long counter;
-
-	counter = __rdtsc();
-	(*hi) = (unsigned int)(counter >> 32);
-	(*lo) = (unsigned int)counter;
-/*
-
-	LARGE_INTEGER lPerformanceCount;
-
-	QueryPerformanceCounter(&lPerformanceCount);
-	(*hi) = (unsigned int)lPerformanceCount.HighPart;
-	(*lo) = (unsigned int)lPerformanceCount.LowPart;
-//	printf("%08X %08X\n",(*hi),(*lo));
-*/
-}
-
-
-/* Record the current value of the cycle counter. */
-void start_counter()
-{
-    access_counter(&cyc_hi, &cyc_lo);
-}
-
-/* Return the number of cycles since the last call to start_counter. */
-double get_counter()
-{
-    unsigned ncyc_hi, ncyc_lo;
-    unsigned hi, lo, borrow;
-    double result;
-
-    /* Get cycle counter */
-    access_counter(&ncyc_hi, &ncyc_lo);
-
-    /* Do double precision subtraction */
-    lo = ncyc_lo - cyc_lo;
-    borrow = cyc_lo > ncyc_lo;
-    hi = ncyc_hi - cyc_hi - borrow;
-    result = (double) hi * (1 << 30) * 4 + lo;
-    return result;
-}
-void make_CPU_busy(void)
-{
-	volatile double old_tick,new_tick;
-	start_counter();
-	old_tick = get_counter();
-	new_tick = get_counter();
-	while (new_tick - old_tick < 1000000000)
-		new_tick = get_counter();
-}
-
-//CPU的频率
-double mhz(int verbose)
-{
-    LARGE_INTEGER lFrequency;
-    LARGE_INTEGER lPerformanceCount_Start;
-    LARGE_INTEGER lPerformanceCount_End;
-	double mhz;
-	double fTime;
-	__int64 _i64StartCpuCounter;
-	__int64 _i64EndCpuCounter;
-    //On a multiprocessor machine, it should not matter which processor is called.
-    //However, you can get different results on different processors due to bugs in
-    //the BIOS or the HAL. To specify processor affinity for a thread, use the SetThreadAffinityMask function.
-    HANDLE hThread=GetCurrentThread();
-    SetThreadAffinityMask(hThread,0x1);
-
-    //主板上高精度定时器的晶振频率
-    //这个定时器应该就是一片8253或者8254
-    //在intel ich7中集成了8254
-    QueryPerformanceFrequency(&lFrequency);
-//    if (verbose>0)
-//    	printf("高精度定时器的晶振频率：%1.0fHz.\n",(double)lFrequency.QuadPart);
-
-    //这个定时器每经过一个时钟周期，其计数器会+1
-    QueryPerformanceCounter(&lPerformanceCount_Start);
-
-    //RDTSC指令:获取CPU经历的时钟周期数
-    _i64StartCpuCounter=__rdtsc();
-
-    //延时长一点,误差会小一点
-    //int nTemp=100000;
-    //while (--nTemp);
-    Sleep(200);
-
-    QueryPerformanceCounter(&lPerformanceCount_End);
-
-    _i64EndCpuCounter=__rdtsc();
-
-    //f=1/T => f=计数次数/(计数次数*T)
-    //这里的“计数次数*T”就是时间差
-    fTime=((double)lPerformanceCount_End.QuadPart-(double)lPerformanceCount_Start.QuadPart)
-        /(double)lFrequency.QuadPart;
-
- 		mhz = (_i64EndCpuCounter-_i64StartCpuCounter)/(fTime*1000000.0);
-    if (verbose>0)
-    	printf("CPU频率为:%1.6fMHz.\n",mhz);
-    return mhz;
-}
-
-double CPU_Factor1(void)
-{
-	double result;
-	int i,j,k,ii,jj,kk;
-	LARGE_INTEGER lStart,lEnd;
-  LARGE_INTEGER lFrequency;
-  HANDLE hThread;
-  double fTime;
-
-  QueryPerformanceFrequency(&lFrequency);
-
-	ii = 43273;
-	kk = 1238;
-	result = 1;
-	jj = 1244;
-
-    hThread=GetCurrentThread();
-    SetThreadAffinityMask(hThread,0x1);
-  QueryPerformanceCounter(&lStart);
-  //_asm("cpuid");
-	start_counter();
-	for (i=0;i<100;i++)
-		for (j=0;j<1000;j++)
-			for (k=0;k<1000;k++)
-				kk += kk*ii+jj;
-
-	result = get_counter();
-	QueryPerformanceCounter(&lEnd);
-  fTime=((double)lEnd.QuadPart-(double)lStart.QuadPart);
-	printf("CPU运行时间为%f",result);
-	printf("\t %f\n",fTime);
-	return result;
-}
-
-double CPU_Factor(void)
-{
- double frequency;
- double multiplier = 1000 * 1000 * 1000;//nano
- LARGE_INTEGER lFrequency;
- LARGE_INTEGER start,stop;
- HANDLE hThread;
- int i;
- const int gigahertz= 1000*1000*1000;
- const int known_instructions_per_loop = 27317; 
-
- int iterations = 100000000;
- int g = 0;
- double normal_ticks_per_second;
-double ticks;
-double time;
-double loops_per_sec;
-double instructions_per_loop;
-double ratio;
-double actual_freq;
-
- QueryPerformanceFrequency(&lFrequency);
- frequency = (double)lFrequency.QuadPart;
-
- hThread=GetCurrentThread();
- SetThreadAffinityMask(hThread,0x1);
- QueryPerformanceCounter(&start);
- for( i = 0; i < iterations; i++)
- {
-   g++;
-   g++;
-   g++;
-   g++;
- }
- QueryPerformanceCounter(&stop);
-
- //normal ticks differs from the WMI data, i.e 3125, when WMI 3201, and CPUZ 3199
- normal_ticks_per_second = frequency * 1000;
- ticks = (double)((double)stop.QuadPart - (double)start.QuadPart);
- time = (ticks * multiplier) /frequency;
- loops_per_sec = iterations / (time/multiplier);
- instructions_per_loop = normal_ticks_per_second  / loops_per_sec;
-
- ratio = (instructions_per_loop / known_instructions_per_loop);
- actual_freq = normal_ticks_per_second / ratio;
-/* 
- actual_freq = normal_ticks_per_second / ratio;
- actual_freq = known_instructions_per_loop*iterations*multiplier/time;
-
-	2293 = x/time;
-	
-	2292.599713*1191533038.809362=known_instructions_per_loop*100000000*1000
- loops_per_sec = iterations*frequency / ticks
- 
- instructions_per_loop =   / loops_per_sec;
-*/ 
- printf("Perf counter freq: %f\n", normal_ticks_per_second);
- printf("Loops per sec:      %f\n", loops_per_sec);
- printf("Perf counter freq div loops per sec: %f\n", instructions_per_loop);
- printf("Presumed freq: %f\n", actual_freq);
- printf("ratio: %f\n", ratio);
- printf("time=%f\n",time);
- return ratio;
-}
+/* clock.c
+ * Retrofitted to use thread-specific timers
+ * and to get clock information from /proc/cpuinfo
+ * (C) R. E. Bryant, 2010
+ * Modified for cross-platform compatibility
+ */
+
+#define _GNU_SOURCE // For sched_setaffinity on Linux
+#include <stdint.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+
+#ifdef _WIN32
+#include <intrin.h>
+#include <windows.h>
+#else
+#include <sched.h>
+#include <time.h>
+#include <unistd.h>
+#include <x86intrin.h>
+typedef struct {
+  uint64_t QuadPart;
+} LARGE_INTEGER;
+typedef void *HANDLE;
+#define __int64 long long
+#define Sleep(ms) usleep((ms) * 1000)
+#endif
+
+#include "clock.h"
+
+/* Use x86 cycle counter */
+static unsigned cyc_hi = 0;
+static unsigned cyc_lo = 0;
+
+void access_counter(unsigned *hi, unsigned *lo) {
+  uint64_t counter = __rdtsc();
+  *hi = (unsigned)(counter >> 32);
+  *lo = (unsigned)counter;
+}
+
+void start_counter() { access_counter(&cyc_hi, &cyc_lo); }
+
+double get_counter() {
+  unsigned ncyc_hi, ncyc_lo;
+  access_counter(&ncyc_hi, &ncyc_lo);
+  uint64_t start = ((uint64_t)cyc_hi << 32) | cyc_lo;
+  uint64_t end = ((uint64_t)ncyc_hi << 32) | ncyc_lo;
+  return (double)(end - start);
+}
+
+void make_CPU_busy(void) {
+  volatile double old_tick = get_counter();
+  volatile double new_tick;
+  while ((new_tick - old_tick) < 1000000000) {
+    new_tick = get_counter();
+  }
+}
+
+#ifdef _WIN32
+#define GET_TIME(dest) QueryPerformanceCounter(dest)
+#else
+static inline void GET_TIME(LARGE_INTEGER *dest) {
+  struct timespec ts;
+  clock_gettime(CLOCK_MONOTONIC, &ts);
+  dest->QuadPart = (uint64_t)ts.tv_sec * 1000000000 + ts.tv_nsec;
+}
+#define QueryPerformanceFrequency(freq) ((freq)->QuadPart = 1000000000)
+#endif
+
+double mhz(int verbose) {
+  LARGE_INTEGER lFrequency;
+  LARGE_INTEGER lPerformanceCount_Start;
+  LARGE_INTEGER lPerformanceCount_End;
+  double mhz;
+  double fTime;
+  __int64 _i64StartCpuCounter;
+  __int64 _i64EndCpuCounter;
+
+#ifdef _WIN32
+  HANDLE hThread = GetCurrentThread();
+  SetThreadAffinityMask(hThread, 0x1);
+#else
+  cpu_set_t cpuset;
+  CPU_ZERO(&cpuset);
+  CPU_SET(0, &cpuset);
+  sched_setaffinity(0, sizeof(cpuset), &cpuset);
+#endif
+
+  QueryPerformanceFrequency(&lFrequency);
+  GET_TIME(&lPerformanceCount_Start);
+  _i64StartCpuCounter = __rdtsc();
+  Sleep(200);
+  GET_TIME(&lPerformanceCount_End);
+  _i64EndCpuCounter = __rdtsc();
+
+  fTime = (lPerformanceCount_End.QuadPart - lPerformanceCount_Start.QuadPart) /
+          (double)lFrequency.QuadPart;
+  mhz = (_i64EndCpuCounter - _i64StartCpuCounter) / (fTime * 1000000.0);
+
+  if (verbose > 0) {
+    printf("CPU频率为: %.6fMHz.\n", mhz);
+  }
+  return mhz;
+}
+
+double CPU_Factor1(void) {
+  double result;
+  int i, j, k;
+  LARGE_INTEGER lStart, lEnd;
+  LARGE_INTEGER lFrequency;
+  double fTime;
+
+#ifdef _WIN32
+  HANDLE hThread = GetCurrentThread();
+  SetThreadAffinityMask(hThread, 0x1);
+#else
+  cpu_set_t cpuset;
+  CPU_ZERO(&cpuset);
+  CPU_SET(0, &cpuset);
+  sched_setaffinity(0, sizeof(cpuset), &cpuset);
+#endif
+
+  QueryPerformanceFrequency(&lFrequency);
+  GET_TIME(&lStart);
+  start_counter();
+
+  for (i = 0; i < 100; i++)
+    for (j = 0; j < 1000; j++)
+      for (k = 0; k < 1000; k++)
+        ;
+
+  result = get_counter();
+  GET_TIME(&lEnd);
+
+  fTime = (lEnd.QuadPart - lStart.QuadPart) / (double)lFrequency.QuadPart;
+  printf("CPU计算时长为: %f", result);
+  printf("\t %f\n", fTime);
+  return result;
+}
+
+double CPU_Factor(void) {
+  double frequency;
+  double multiplier = 1000 * 1000 * 1000; // nano
+  LARGE_INTEGER lFrequency;
+  LARGE_INTEGER start, stop;
+  int i;
+  const int known_instructions_per_loop = 27317;
+  int iterations = 100000000;
+  int g = 0;
+  double normal_ticks_per_second;
+  double ticks;
+  double time;
+  double loops_per_sec;
+  double instructions_per_loop;
+  double ratio;
+  double actual_freq;
+
+#ifdef _WIN32
+  HANDLE hThread = GetCurrentThread();
+  SetThreadAffinityMask(hThread, 0x1);
+#else
+  cpu_set_t cpuset;
+  CPU_ZERO(&cpuset);
+  CPU_SET(0, &cpuset);
+  sched_setaffinity(0, sizeof(cpuset), &cpuset);
+#endif
+
+  QueryPerformanceFrequency(&lFrequency);
+  frequency = (double)lFrequency.QuadPart;
+  GET_TIME(&start);
+
+  for (i = 0; i < iterations; i++) {
+    g++;
+    g++;
+    g++;
+    g++;
+  }
+
+  GET_TIME(&stop);
+  normal_ticks_per_second = frequency * 1000;
+  ticks = (double)(stop.QuadPart - start.QuadPart);
+  time = (ticks * multiplier) / frequency;
+  loops_per_sec = iterations / (time / multiplier);
+  instructions_per_loop = normal_ticks_per_second / loops_per_sec;
+  ratio = instructions_per_loop / known_instructions_per_loop;
+  actual_freq = normal_ticks_per_second / ratio;
+
+  printf("Perf counter freq: %f\n", normal_ticks_per_second);
+  printf("Loops per sec:      %f\n", loops_per_sec);
+  printf("Perf counter freq div loops per sec: %f\n", instructions_per_loop);
+  printf("Presumed freq: %f\n", actual_freq);
+  printf("ratio: %f\n", ratio);
+  printf("time=%f\n", time);
+  return ratio;
+}

perflab/matrix/clock.c.bak

@@ -0,0 +1,229 @@
+/* clock.c
+ * Retrofitted to use thread-specific timers
+ * and to get clock information from /proc/cpuinfo
+ * (C) R. E. Bryant, 2010
+ *
+ */
+
+/* When this constant is not defined, uses time stamp counter */
+#define USE_POSIX 0
+
+/* Choice to use cpu_gettime call or Intel time stamp counter directly */
+
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <x86intrin.h>
+//#include <intrinsics.h>
+//#include <windows.h>
+#include <time.h>
+#include "clock.h"
+
+/* Use x86 cycle counter */
+
+/* Initialize the cycle counter */
+static unsigned cyc_hi = 0;
+static unsigned cyc_lo = 0;
+
+/* Set *hi and *lo to the high and low order bits  of the cycle counter.
+   Implementation requires assembly code to use the rdtsc instruction. */
+void access_counter(unsigned *hi, unsigned *lo)
+{
+
+	long long counter;
+
+	counter = __rdtsc();
+	(*hi) = (unsigned int)(counter >> 32);
+	(*lo) = (unsigned int)counter;
+/*
+
+	LARGE_INTEGER lPerformanceCount;
+
+	QueryPerformanceCounter(&lPerformanceCount);
+	(*hi) = (unsigned int)lPerformanceCount.HighPart;
+	(*lo) = (unsigned int)lPerformanceCount.LowPart;
+//	printf("%08X %08X\n",(*hi),(*lo));
+*/
+}
+
+
+/* Record the current value of the cycle counter. */
+void start_counter()
+{
+    access_counter(&cyc_hi, &cyc_lo);
+}
+
+/* Return the number of cycles since the last call to start_counter. */
+double get_counter()
+{
+    unsigned ncyc_hi, ncyc_lo;
+    unsigned hi, lo, borrow;
+    double result;
+
+    /* Get cycle counter */
+    access_counter(&ncyc_hi, &ncyc_lo);
+
+    /* Do double precision subtraction */
+    lo = ncyc_lo - cyc_lo;
+    borrow = cyc_lo > ncyc_lo;
+    hi = ncyc_hi - cyc_hi - borrow;
+    result = (double) hi * (1 << 30) * 4 + lo;
+    return result;
+}
+void make_CPU_busy(void)
+{
+	volatile double old_tick,new_tick;
+	start_counter();
+	old_tick = get_counter();
+	new_tick = get_counter();
+	while (new_tick - old_tick < 1000000000)
+		new_tick = get_counter();
+}
+
+//CPU的频率
+double mhz(int verbose)
+{
+    LARGE_INTEGER lFrequency;
+    LARGE_INTEGER lPerformanceCount_Start;
+    LARGE_INTEGER lPerformanceCount_End;
+	double mhz;
+	double fTime;
+	__int64 _i64StartCpuCounter;
+	__int64 _i64EndCpuCounter;
+    //On a multiprocessor machine, it should not matter which processor is called.
+    //However, you can get different results on different processors due to bugs in
+    //the BIOS or the HAL. To specify processor affinity for a thread, use the SetThreadAffinityMask function.
+    HANDLE hThread=GetCurrentThread();
+    SetThreadAffinityMask(hThread,0x1);
+
+    //主板上高精度定时器的晶振频率
+    //这个定时器应该就是一片8253或者8254
+    //在intel ich7中集成了8254
+    QueryPerformanceFrequency(&lFrequency);
+//    if (verbose>0)
+//    	printf("高精度定时器的晶振频率：%1.0fHz.\n",(double)lFrequency.QuadPart);
+
+    //这个定时器每经过一个时钟周期，其计数器会+1
+    QueryPerformanceCounter(&lPerformanceCount_Start);
+
+    //RDTSC指令:获取CPU经历的时钟周期数
+    _i64StartCpuCounter=__rdtsc();
+
+    //延时长一点,误差会小一点
+    //int nTemp=100000;
+    //while (--nTemp);
+    Sleep(200);
+
+    QueryPerformanceCounter(&lPerformanceCount_End);
+
+    _i64EndCpuCounter=__rdtsc();
+
+    //f=1/T => f=计数次数/(计数次数*T)
+    //这里的“计数次数*T”就是时间差
+    fTime=((double)lPerformanceCount_End.QuadPart-(double)lPerformanceCount_Start.QuadPart)
+        /(double)lFrequency.QuadPart;
+
+ 		mhz = (_i64EndCpuCounter-_i64StartCpuCounter)/(fTime*1000000.0);
+    if (verbose>0)
+    	printf("CPU频率为:%1.6fMHz.\n",mhz);
+    return mhz;
+}
+
+double CPU_Factor1(void)
+{
+	double result;
+	int i,j,k,ii,jj,kk;
+	LARGE_INTEGER lStart,lEnd;
+  LARGE_INTEGER lFrequency;
+  HANDLE hThread;
+  double fTime;
+
+  QueryPerformanceFrequency(&lFrequency);
+
+	ii = 43273;
+	kk = 1238;
+	result = 1;
+	jj = 1244;
+
+    hThread=GetCurrentThread();
+    SetThreadAffinityMask(hThread,0x1);
+  QueryPerformanceCounter(&lStart);
+  //_asm("cpuid");
+	start_counter();
+	for (i=0;i<100;i++)
+		for (j=0;j<1000;j++)
+			for (k=0;k<1000;k++)
+				kk += kk*ii+jj;
+
+	result = get_counter();
+	QueryPerformanceCounter(&lEnd);
+  fTime=((double)lEnd.QuadPart-(double)lStart.QuadPart);
+	printf("CPU运行时间为%f",result);
+	printf("\t %f\n",fTime);
+	return result;
+}
+
+double CPU_Factor(void)
+{
+ double frequency;
+ double multiplier = 1000 * 1000 * 1000;//nano
+ LARGE_INTEGER lFrequency;
+ LARGE_INTEGER start,stop;
+ HANDLE hThread;
+ int i;
+ const int gigahertz= 1000*1000*1000;
+ const int known_instructions_per_loop = 27317; 
+
+ int iterations = 100000000;
+ int g = 0;
+ double normal_ticks_per_second;
+double ticks;
+double time;
+double loops_per_sec;
+double instructions_per_loop;
+double ratio;
+double actual_freq;
+
+ QueryPerformanceFrequency(&lFrequency);
+ frequency = (double)lFrequency.QuadPart;
+
+ hThread=GetCurrentThread();
+ SetThreadAffinityMask(hThread,0x1);
+ QueryPerformanceCounter(&start);
+ for( i = 0; i < iterations; i++)
+ {
+   g++;
+   g++;
+   g++;
+   g++;
+ }
+ QueryPerformanceCounter(&stop);
+
+ //normal ticks differs from the WMI data, i.e 3125, when WMI 3201, and CPUZ 3199
+ normal_ticks_per_second = frequency * 1000;
+ ticks = (double)((double)stop.QuadPart - (double)start.QuadPart);
+ time = (ticks * multiplier) /frequency;
+ loops_per_sec = iterations / (time/multiplier);
+ instructions_per_loop = normal_ticks_per_second  / loops_per_sec;
+
+ ratio = (instructions_per_loop / known_instructions_per_loop);
+ actual_freq = normal_ticks_per_second / ratio;
+/* 
+ actual_freq = normal_ticks_per_second / ratio;
+ actual_freq = known_instructions_per_loop*iterations*multiplier/time;
+
+	2293 = x/time;
+	
+	2292.599713*1191533038.809362=known_instructions_per_loop*100000000*1000
+ loops_per_sec = iterations*frequency / ticks
+ 
+ instructions_per_loop =   / loops_per_sec;
+*/ 
+ printf("Perf counter freq: %f\n", normal_ticks_per_second);
+ printf("Loops per sec:      %f\n", loops_per_sec);
+ printf("Perf counter freq div loops per sec: %f\n", instructions_per_loop);
+ printf("Presumed freq: %f\n", actual_freq);
+ printf("ratio: %f\n", ratio);
+ printf("time=%f\n",time);
+ return ratio;
+}

perflab/matrix/fcyc.c

@@ -119,7 +119,7 @@ double fcyc(test_funct f, int *params)
 	    if (clear_cache)
 		clear();
 	    start_counter();
-	    f(params);
+	    f((long*)params);
 	    cyc = get_counter();
 	    if (cyc > 0.0)
 		add_sample(cyc);
@@ -131,7 +131,7 @@ double fcyc(test_funct f, int *params)
 		clear();
 	    start_counter();
 	    for (i=0;i<MAX_ITER_TIMES;i++)
-    		f(params);
+    		f((long*)params);
 	    cyc = get_counter()/MAX_ITER_TIMES;
 	    if (cyc > 0.0)
 		add_sample(cyc);

perflab/matrix/rowcol.c

@@ -1,77 +1,69 @@
 /**************************************************************************
-	行/列求和函数。按下面的要求编辑此文件：
-	1. 将你的学号、姓名，以注释的方式写到下面；
-	2. 实现不同版本的行列求和函数；
-	3. 编辑rc_fun_rec rc_fun_tab数组，将你的最好的答案
-		（最好的行和列求和、最好的列求和）作为数组的前两项
+        ??/???????????????????????????????
+        1. ???????????????????????????????
+        2. ??????????????????????
+        3. ??rc_fun_rec rc_fun_tab??????????????????
+                ???????????????????????????????????????????
 ***************************************************************************/
-   
+
 /*
-	学号：201209054233
-	姓名：夜半加班狂
+        ????201209054233
+        ??????????????
 */
 
+#include "rowcol.h"
+#include <math.h>
+#include <stdio.h>
+#include <stdlib.h>
 
-#include  <stdio.h>
-#include  <stdlib.h>
-#include  "rowcol.h"
-#include  <math.h>
-
-/* 参考的列求和函数实现 */
-/* 计算矩阵中的每一列的和。请注意对于行和列求和来说，调用参数是
-	一样的，只是第2个参数不会用到而已
+/* ????????????????? */
+/* ???????????????????????????????????????????????
+        ??????????2?????????????????
 */
 
-void c_sum(matrix_t M, vector_t rowsum, vector_t colsum)
-{
-    int i,j;
+void c_sum(matrix_t M, vector_t rowsum, vector_t colsum) {
+  int i, j;
+  for (j = 0; j < N; j++) {
+    colsum[j] = 0;
+    for (i = 0; i < N; i++)
+      colsum[j] += M[i][j];
+  }
+}
+
+/* ???????????????????? */
+/* ??????????????????????? */
+
+void rc_sum(matrix_t M, vector_t rowsum, vector_t colsum) {
+  int i, j;
+  for (i = 0; i < N; i++) {
+    rowsum[i] = colsum[i] = 0;
     for (j = 0; j < N; j++) {
-	colsum[j] = 0;
-	for (i = 0; i < N; i++)
-	    colsum[j] += M[i][j];
+      rowsum[i] += M[i][j];
+      colsum[i] += M[j][i];
     }
+  }
 }
 
-
-/* 参考的列和行求和函数实现 */
-/* 计算矩阵中的每一行、每一列的和。 */
-
-void rc_sum(matrix_t M, vector_t rowsum, vector_t colsum)
-{
-    int i,j;
-    for (i = 0; i < N; i++) {
-	rowsum[i] = colsum[i] = 0;
-	for (j = 0; j < N; j++) {
-	    rowsum[i] += M[i][j];
-	    colsum[i] += M[j][i];
-	}
-    }
-}
-
-
-
-/* 
-	这个表格包含多个数组元素，每一组元素（函数名字, COL/ROWCOL, "描述字符串"）
-	COL表示该函数仅仅计算每一列的和
-	ROWCOL表示该函数计算每一行、每一列的和
-	将你认为最好的两个实现，放在最前面。
-	比如：
-	{my_c_sum1, "超级垃圾列求和实现"},
-	{my_rc_sum2, "好一点的行列求和实现"},
+/*
+        ????????????????????????????????????????, COL/ROWCOL, "?????????"??
+        COL??????????????????????
+        ROWCOL???????????????????????
+        ?????????????????????????????
+        ????
+        {my_c_sum1, "?????????????????"},
+        {my_rc_sum2, "??????????????????"},
 */
 
-rc_fun_rec rc_fun_tab[] = 
-{
+rc_fun_rec rc_fun_tab[] = {
 
-  /* 第一项，应当是你写的最好列求和的函数实现 */
+    /* ???????????????????????????????? */
     {c_sum, COL, "Best column sum"},
-  /* 第二项，应当是你写的最好行列求和的函数实现 */
+    /* ?????????????????????????????????? */
     {rc_sum, ROWCOL, "Best row and column sum"},
 
     {c_sum, COL, "Column sum, reference implementation"},
 
     {rc_sum, ROWCOL, "Row and column sum, reference implementation"},
 
- /* 下面的代码不能修改或者删除！！表明数组列表结束 */
-    {NULL,ROWCOL,NULL}
-};
+    /* ??????????????????????????????????????? */
+    {NULL, ROWCOL, NULL}};

perflab/matrix/rowcol.c~

@@ -0,0 +1,162 @@
+/**************************************************************************
+	行/列求和函数。按下面的要求编辑此文件：
+	1. 将你的学号、姓名，以注释的方式写到下面；
+	2. 实现不同版本的行列求和函数；
+	3. 编辑rc_fun_rec rc_fun_tab数组，将你的最好的答案
+		（最好的行和列求和、最好的列求和）作为数组的前两项
+***************************************************************************/
+   
+/*
+	学号：202302723005
+	姓名：程景愉
+*/
+
+
+#include  <stdio.h>
+#include  <stdlib.h>
+#include  "rowcol.h"
+#include  <math.h>
+#include  <cuda_runtime.h>
+
+/* 参考的列求和函数实现 */
+/* 计算矩阵中的每一列的和。请注意对于行和列求和来说，调用参数是
+	一样的，只是第2个参数不会用到而已
+*/
+
+void c_sum(matrix_t M, vector_t rowsum, vector_t colsum)
+{
+    int i,j;
+    for (j = 0; j < N; j++) {
+	colsum[j] = 0;
+	for (i = 0; i < N; i++)
+	    colsum[j] += M[i][j];
+    }
+}
+
+
+/* 参考的列和行求和函数实现 */
+/* 计算矩阵中的每一行、每一列的和。 */
+
+void rc_sum(matrix_t M, vector_t rowsum, vector_t colsum)
+{
+    int i,j;
+    for (i = 0; i < N; i++) {
+	rowsum[i] = colsum[i] = 0;
+	for (j = 0; j < N; j++) {
+	    rowsum[i] += M[i][j];
+	    colsum[i] += M[j][i];
+	}
+    }
+}
+
+/* CUDA优化的列求和函数 */
+void cuda_c_sum(matrix_t M, vector_t rowsum, vector_t colsum)
+{
+    // 分配设备内存
+    int *d_M, *d_colsum;
+    cudaMalloc(&d_M, N * N * sizeof(int));
+    cudaMalloc(&d_colsum, N * sizeof(int));
+    
+    // 将数据从主机复制到设备
+    cudaMemcpy(d_M, M, N * N * sizeof(int), cudaMemcpyHostToDevice);
+    
+    // 定义CUDA核函数
+    dim3 blockDim(256);
+    dim3 gridDim((N + blockDim.x - 1) / blockDim.x);
+    
+    // 启动核函数
+    cudaColumnSum<<<gridDim, blockDim>>>(d_M, d_colsum);
+    
+    // 将结果从设备复制回主机
+    cudaMemcpy(colsum, d_colsum, N * sizeof(int), cudaMemcpyDeviceToHost);
+    
+    // 释放设备内存
+    cudaFree(d_M);
+    cudaFree(d_colsum);
+}
+
+/* CUDA优化的行列求和函数 */
+void cuda_rc_sum(matrix_t M, vector_t rowsum, vector_t colsum)
+{
+    // 分配设备内存
+    int *d_M, *d_rowsum, *d_colsum;
+    cudaMalloc(&d_M, N * N * sizeof(int));
+    cudaMalloc(&d_rowsum, N * sizeof(int));
+    cudaMalloc(&d_colsum, N * sizeof(int));
+    
+    // 将数据从主机复制到设备
+    cudaMemcpy(d_M, M, N * N * sizeof(int), cudaMemcpyHostToDevice);
+    
+    // 定义CUDA核函数
+    dim3 blockDim(256);
+    dim3 gridDim((N + blockDim.x - 1) / blockDim.x);
+    
+    // 启动核函数
+    cudaRowColSum<<<gridDim, blockDim>>>(d_M, d_rowsum, d_colsum);
+    
+    // 将结果从设备复制回主机
+    cudaMemcpy(rowsum, d_rowsum, N * sizeof(int), cudaMemcpyDeviceToHost);
+    cudaMemcpy(colsum, d_colsum, N * sizeof(int), cudaMemcpyDeviceToHost);
+    
+    // 释放设备内存
+    cudaFree(d_M);
+    cudaFree(d_rowsum);
+    cudaFree(d_colsum);
+}
+
+/* CUDA核函数 - 列求和 */
+__global__ void cudaColumnSum(int *M, int *colsum)
+{
+    int col = blockIdx.x * blockDim.x + threadIdx.x;
+    if (col < N) {
+        colsum[col] = 0;
+        for (int row = 0; row < N; row++) {
+            colsum[col] += M[row * N + col];
+        }
+    }
+}
+
+/* CUDA核函数 - 行列求和 */
+__global__ void cudaRowColSum(int *M, int *rowsum, int *colsum)
+{
+    int idx = blockIdx.x * blockDim.x + threadIdx.x;
+    if (idx < N) {
+        // 计算行和
+        rowsum[idx] = 0;
+        for (int j = 0; j < N; j++) {
+            rowsum[idx] += M[idx * N + j];
+        }
+        
+        // 计算列和
+        colsum[idx] = 0;
+        for (int i = 0; i < N; i++) {
+            colsum[idx] += M[i * N + idx];
+        }
+    }
+}
+
+/* 
+	这个表格包含多个数组元素，每一组元素（函数名字, COL/ROWCOL, "描述字符串"）
+	COL表示该函数仅仅计算每一列的和
+	ROWCOL表示该函数计算每一行、每一列的和
+	将你认为最好的两个实现，放在最前面。
+	比如：
+	{my_c_sum1, "超级垃圾列求和实现"},
+	{my_rc_sum2, "好一点的行列求和实现"},
+*/
+
+rc_fun_rec rc_fun_tab[] = 
+{
+
+  /* 第一项，应当是你写的最好列求和的函数实现 */
+    {cuda_c_sum, COL, "CUDA optimized column sum"},
+  /* 第二项，应当是你写的最好行列求和的函数实现 */
+    {cuda_rc_sum, ROWCOL, "CUDA optimized row and column sum"},
+
+    {c_sum, COL, "Column sum, reference implementation"},
+
+    {rc_sum, ROWCOL, "Row and column sum, reference implementation"},
+
+ /* 下面的代码不能修改或者删除！！表明数组列表结束 */
+    {NULL,ROWCOL,NULL}
+};

perflab/matrix/rowcol.y~

@@ -0,0 +1,240 @@
+/**************************************************************************
+        靠/靠靠靠靠靠靠靠靠靠靠靠靠靠靠靠<E99DA0>
+        1. 靠靠靠靠靠靠靠靠靠靠靠靠靠靠靠<E99DA0>
+        2. 靠靠靠靠靠靠靠靠靠靠靠
+        3. 靠rc_fun_rec rc_fun_tab靠靠靠靠靠靠旷靠靠
+                靠靠旷靠靠靠靠靠靠靠蹩靠靠靠靠靠靠靠靠靠靠<E99DA0>
+***************************************************************************/
+   
+/*
+        靠靠201209054233
+        靠靠靠靠靠靠靠
+*/
+
+
+#include  <stdio.h>
+#include  <stdlib.h>
+#include  "rowcol.h"
+#include  <math.h>
+
+/* 靠靠靠靠靠靠靠靠<E99DA0> */
+/* 靠靠靠靠靠<E99DA0><E99DA0>靠靠靠靠靠靠靠靠靠靠靠靠靠靠靠靠蚩靠靠
+        靠靠靠靠靠2靠靠靠靠靠旷靠靠<E99DA0>
+*/
+
+void c_sum(matrix_t M, vector_t rowsum, vector_t colsum)
+{
+    int i,j;
+    for (j = 0; j < N; j++) {
+        colsum[j] = 0;
+        for (i = 0; i < N; i++)
+            colsum[j] += M[i][j];
+    }
+}
+
+
+/* 靠靠靠靠靠靠靠靠靠靠 */
+/* 靠靠靠靠靠<E99DA0><E99DA0>靠靠<E99DA0>靠靠靠 */
+
+void rc_sum(matrix_t M, vector_t rowsum, vector_t colsum)
+{
+    int i,j;
+    for (i = 0; i < N; i++) {
+        rowsum[i] = colsum[i] = 0;
+        for (j = 0; j < N; j++) {
+            rowsum[i] += M[i][j];
+            colsum[i] += M[j][i];
+        }
+    }
+}
+
+
+
+/* 
+        靠靠靠靠靠靠靠靠靠靠靠靠<E99DA0>靠靠靠靠靠靠靠<E99DA0>, COL/ROWCOL, "靠靠靠靠<E99DA0>"靠
+        COL靠靠<E99DA0>靠靠靠靠靠<E99DA0>靠靠<E99DA0>
+        ROWCOL靠靠<E99DA0>靠靠靠<E99DA0>靠靠<E99DA0>靠靠<E99DA0>
+        靠靠靠靠靠蹩靠靠靠靠靠靠靠靠<E99DA0>
+        靠靠
+        {my_c_sum1, "靠靠靠靠靠靠靠靠<E99DA0>"},
+        {my_rc_sum2, "靠靠靠靠靠靠靠靠靠"},
+*/
+
+rc_fun_rec rc_fun_tab[] = 
+{
+
+  /* 靠靠靠靠靠靠靠靠靠靠靠靠靠靠靠靠 */
+    {c_sum, COL, "Best column sum"},
+  /* 靠靠靠靠靠靠靠靠靠靠靠靠靠靠靠靠靠 */
+    {rc_sum, ROWCOL, "Best row and column sum"},
+
+    {c_sum, COL, "Column sum, reference implementation"},
+
+    {rc_sum, ROWCOL, "Row and column sum, reference implementation"},
+
+ /* 靠靠靠靠靠靠靠靠靠靠靠靠靠靠靠靠靠靠靠<E99DA0> */
+    {NULL,ROWCOL,NULL}
+};
+
+// /**************************************************************************
+// 	行/列求和函数。按下面的要求编辑此文件：
+// 	1. 将你的学号、姓名，以注释的方式写到下面；
+// 	2. 实现不同版本的行列求和函数；
+// 	3. 编辑rc_fun_rec rc_fun_tab数组，将你的最好的答案
+// 		（最好的行和列求和、最好的列求和）作为数组的前两项
+// ***************************************************************************/
+//
+// /*
+// 	学号：202302723005
+// 	姓名：程景愉
+// */
+//
+//
+// #include  <stdio.h>
+// #include  <stdlib.h>
+// #include  "rowcol.h"
+// #include  <math.h>
+// #include  <cuda_runtime.h>
+//
+// /* 参考的列求和函数实现 */
+// /* 计算矩阵中的每一列的和。请注意对于行和列求和来说，调用参数是
+// 	一样的，只是第2个参数不会用到而已
+// */
+//
+// void c_sum(matrix_t M, vector_t rowsum, vector_t colsum)
+// {
+//     int i,j;
+//     for (j = 0; j < N; j++) {
+// 	colsum[j] = 0;
+// 	for (i = 0; i < N; i++)
+// 	    colsum[j] += M[i][j];
+//     }
+// }
+//
+//
+// /* 参考的列和行求和函数实现 */
+// /* 计算矩阵中的每一行、每一列的和。 */
+//
+// void rc_sum(matrix_t M, vector_t rowsum, vector_t colsum)
+// {
+//     int i,j;
+//     for (i = 0; i < N; i++) {
+// 	rowsum[i] = colsum[i] = 0;
+// 	for (j = 0; j < N; j++) {
+// 	    rowsum[i] += M[i][j];
+// 	    colsum[i] += M[j][i];
+// 	}
+//     }
+// }
+//
+// /* CUDA优化的列求和函数 */
+// void cuda_c_sum(matrix_t M, vector_t rowsum, vector_t colsum)
+// {
+//     // 分配设备内存
+//     int *d_M, *d_colsum;
+//     cudaMalloc(&d_M, N * N * sizeof(int));
+//     cudaMalloc(&d_colsum, N * sizeof(int));
+//
+//     // 将数据从主机复制到设备
+//     cudaMemcpy(d_M, M, N * N * sizeof(int), cudaMemcpyHostToDevice);
+//
+//     // 定义CUDA核函数
+//     dim3 blockDim(256);
+//     dim3 gridDim((N + blockDim.x - 1) / blockDim.x);
+//
+//     // 启动核函数
+//     cudaColumnSum<<<gridDim, blockDim>>>(d_M, d_colsum);
+//
+//     // 将结果从设备复制回主机
+//     cudaMemcpy(colsum, d_colsum, N * sizeof(int), cudaMemcpyDeviceToHost);
+//
+//     // 释放设备内存
+//     cudaFree(d_M);
+//     cudaFree(d_colsum);
+// }
+//
+// /* CUDA优化的行列求和函数 */
+// void cuda_rc_sum(matrix_t M, vector_t rowsum, vector_t colsum)
+// {
+//     // 分配设备内存
+//     int *d_M, *d_rowsum, *d_colsum;
+//     cudaMalloc(&d_M, N * N * sizeof(int));
+//     cudaMalloc(&d_rowsum, N * sizeof(int));
+//     cudaMalloc(&d_colsum, N * sizeof(int));
+//
+//     // 将数据从主机复制到设备
+//     cudaMemcpy(d_M, M, N * N * sizeof(int), cudaMemcpyHostToDevice);
+//
+//     // 定义CUDA核函数
+//     dim3 blockDim(256);
+//     dim3 gridDim((N + blockDim.x - 1) / blockDim.x);
+//
+//     // 启动核函数
+//     cudaRowColSum<<<gridDim, blockDim>>>(d_M, d_rowsum, d_colsum);
+//
+//     // 将结果从设备复制回主机
+//     cudaMemcpy(rowsum, d_rowsum, N * sizeof(int), cudaMemcpyDeviceToHost);
+//     cudaMemcpy(colsum, d_colsum, N * sizeof(int), cudaMemcpyDeviceToHost);
+//
+//     // 释放设备内存
+//     cudaFree(d_M);
+//     cudaFree(d_rowsum);
+//     cudaFree(d_colsum);
+// }
+//
+// /* CUDA核函数 - 列求和 */
+// __global__ void cudaColumnSum(int *M, int *colsum)
+// {
+//     int col = blockIdx.x * blockDim.x + threadIdx.x;
+//     if (col < N) {
+//         colsum[col] = 0;
+//         for (int row = 0; row < N; row++) {
+//             colsum[col] += M[row * N + col];
+//         }
+//     }
+// }
+//
+// /* CUDA核函数 - 行列求和 */
+// __global__ void cudaRowColSum(int *M, int *rowsum, int *colsum)
+// {
+//     int idx = blockIdx.x * blockDim.x + threadIdx.x;
+//     if (idx < N) {
+//         // 计算行和
+//         rowsum[idx] = 0;
+//         for (int j = 0; j < N; j++) {
+//             rowsum[idx] += M[idx * N + j];
+//         }
+//
+//         // 计算列和
+//         colsum[idx] = 0;
+//         for (int i = 0; i < N; i++) {
+//             colsum[idx] += M[i * N + idx];
+//         }
+//     }
+// }
+//
+// /* 
+// 	这个表格包含多个数组元素，每一组元素（函数名字, COL/ROWCOL, "描述字符串"）
+// 	COL表示该函数仅仅计算每一列的和
+// 	ROWCOL表示该函数计算每一行、每一列的和
+// 	将你认为最好的两个实现，放在最前面。
+// 	比如：
+// 	{my_c_sum1, "超级垃圾列求和实现"},
+// 	{my_rc_sum2, "好一点的行列求和实现"},
+// */
+//
+// rc_fun_rec rc_fun_tab[] = 
+// {
+//
+//   /* 第一项，应当是你写的最好列求和的函数实现 */
+//     {cuda_c_sum, COL, "CUDA optimized column sum"},
+//   /* 第二项，应当是你写的最好行列求和的函数实现 */
+//     {cuda_rc_sum, ROWCOL, "CUDA optimized row and column sum"},
+//
+//     {c_sum, COL, "Column sum, reference implementation"},
+//
+//     {rc_sum, ROWCOL, "Row and column sum, reference implementation"},
+//
+//  /* 下面的代码不能修改或者删除！！表明数组列表结束 */
+//     {NULL,ROWCOL,NULL}
+// };

perflab/matrix/rowcol.z~

@@ -0,0 +1,240 @@
+/**************************************************************************
+        靠/靠靠靠靠靠靠靠靠靠靠靠靠靠靠靠<E99DA0>
+        1. 靠靠靠靠靠靠靠靠靠靠靠靠靠靠靠<E99DA0>
+        2. 靠靠靠靠靠靠靠靠靠靠靠
+        3. 靠rc_fun_rec rc_fun_tab靠靠靠靠靠靠旷靠靠
+                靠靠旷靠靠靠靠靠靠靠蹩靠靠靠靠靠靠靠靠靠靠<E99DA0>
+***************************************************************************/
+   
+/*
+        靠靠201209054233
+        靠靠靠靠靠靠靠
+*/
+
+
+#include  <stdio.h>
+#include  <stdlib.h>
+#include  "rowcol.h"
+#include  <math.h>
+
+/* 靠靠靠靠靠靠靠靠<E99DA0> */
+/* 靠靠靠靠靠<E99DA0><E99DA0>靠靠靠靠靠靠靠靠靠靠靠靠靠靠靠靠蚩靠靠
+        靠靠靠靠靠2靠靠靠靠靠旷靠靠<E99DA0>
+*/
+
+void c_sum(matrix_t M, vector_t rowsum, vector_t colsum)
+{
+    int i,j;
+    for (j = 0; j < N; j++) {
+        colsum[j] = 0;
+        for (i = 0; i < N; i++)
+            colsum[j] += M[i][j];
+    }
+}
+
+
+/* 靠靠靠靠靠靠靠靠靠靠 */
+/* 靠靠靠靠靠<E99DA0><E99DA0>靠靠<E99DA0>靠靠靠 */
+
+void rc_sum(matrix_t M, vector_t rowsum, vector_t colsum)
+{
+    int i,j;
+    for (i = 0; i < N; i++) {
+        rowsum[i] = colsum[i] = 0;
+        for (j = 0; j < N; j++) {
+            rowsum[i] += M[i][j];
+            colsum[i] += M[j][i];
+        }
+    }
+}
+
+
+
+/* 
+        靠靠靠靠靠靠靠靠靠靠靠靠<E99DA0>靠靠靠靠靠靠靠<E99DA0>, COL/ROWCOL, "靠靠靠靠<E99DA0>"靠
+        COL靠靠<E99DA0>靠靠靠靠靠<E99DA0>靠靠<E99DA0>
+        ROWCOL靠靠<E99DA0>靠靠靠<E99DA0>靠靠<E99DA0>靠靠<E99DA0>
+        靠靠靠靠靠蹩靠靠靠靠靠靠靠靠<E99DA0>
+        靠靠
+        {my_c_sum1, "靠靠靠靠靠靠靠靠<E99DA0>"},
+        {my_rc_sum2, "靠靠靠靠靠靠靠靠靠"},
+*/
+
+rc_fun_rec rc_fun_tab[] = 
+{
+
+  /* 靠靠靠靠靠靠靠靠靠靠靠靠靠靠靠靠 */
+    {c_sum, COL, "Best column sum"},
+  /* 靠靠靠靠靠靠靠靠靠靠靠靠靠靠靠靠靠 */
+    {rc_sum, ROWCOL, "Best row and column sum"},
+
+    {c_sum, COL, "Column sum, reference implementation"},
+
+    {rc_sum, ROWCOL, "Row and column sum, reference implementation"},
+
+ /* 靠靠靠靠靠靠靠靠靠靠靠靠靠靠靠靠靠靠靠<E99DA0> */
+    {NULL,ROWCOL,NULL}
+};
+
+// /**************************************************************************
+// 	行/列求和函数。按下面的要求编辑此文件：
+// 	1. 将你的学号、姓名，以注释的方式写到下面；
+// 	2. 实现不同版本的行列求和函数；
+// 	3. 编辑rc_fun_rec rc_fun_tab数组，将你的最好的答案
+// 		（最好的行和列求和、最好的列求和）作为数组的前两项
+// ***************************************************************************/
+//
+// /*
+// 	学号：202302723005
+// 	姓名：程景愉
+// */
+//
+//
+// #include  <stdio.h>
+// #include  <stdlib.h>
+// #include  "rowcol.h"
+// #include  <math.h>
+// #include  <cuda_runtime.h>
+//
+// /* 参考的列求和函数实现 */
+// /* 计算矩阵中的每一列的和。请注意对于行和列求和来说，调用参数是
+// 	一样的，只是第2个参数不会用到而已
+// */
+//
+// void c_sum(matrix_t M, vector_t rowsum, vector_t colsum)
+// {
+//     int i,j;
+//     for (j = 0; j < N; j++) {
+// 	colsum[j] = 0;
+// 	for (i = 0; i < N; i++)
+// 	    colsum[j] += M[i][j];
+//     }
+// }
+//
+//
+// /* 参考的列和行求和函数实现 */
+// /* 计算矩阵中的每一行、每一列的和。 */
+//
+// void rc_sum(matrix_t M, vector_t rowsum, vector_t colsum)
+// {
+//     int i,j;
+//     for (i = 0; i < N; i++) {
+// 	rowsum[i] = colsum[i] = 0;
+// 	for (j = 0; j < N; j++) {
+// 	    rowsum[i] += M[i][j];
+// 	    colsum[i] += M[j][i];
+// 	}
+//     }
+// }
+//
+// /* CUDA优化的列求和函数 */
+// void cuda_c_sum(matrix_t M, vector_t rowsum, vector_t colsum)
+// {
+//     // 分配设备内存
+//     int *d_M, *d_colsum;
+//     cudaMalloc(&d_M, N * N * sizeof(int));
+//     cudaMalloc(&d_colsum, N * sizeof(int));
+//
+//     // 将数据从主机复制到设备
+//     cudaMemcpy(d_M, M, N * N * sizeof(int), cudaMemcpyHostToDevice);
+//
+//     // 定义CUDA核函数
+//     dim3 blockDim(256);
+//     dim3 gridDim((N + blockDim.x - 1) / blockDim.x);
+//
+//     // 启动核函数
+//     cudaColumnSum<<<gridDim, blockDim>>>(d_M, d_colsum);
+//
+//     // 将结果从设备复制回主机
+//     cudaMemcpy(colsum, d_colsum, N * sizeof(int), cudaMemcpyDeviceToHost);
+//
+//     // 释放设备内存
+//     cudaFree(d_M);
+//     cudaFree(d_colsum);
+// }
+//
+// /* CUDA优化的行列求和函数 */
+// void cuda_rc_sum(matrix_t M, vector_t rowsum, vector_t colsum)
+// {
+//     // 分配设备内存
+//     int *d_M, *d_rowsum, *d_colsum;
+//     cudaMalloc(&d_M, N * N * sizeof(int));
+//     cudaMalloc(&d_rowsum, N * sizeof(int));
+//     cudaMalloc(&d_colsum, N * sizeof(int));
+//
+//     // 将数据从主机复制到设备
+//     cudaMemcpy(d_M, M, N * N * sizeof(int), cudaMemcpyHostToDevice);
+//
+//     // 定义CUDA核函数
+//     dim3 blockDim(256);
+//     dim3 gridDim((N + blockDim.x - 1) / blockDim.x);
+//
+//     // 启动核函数
+//     cudaRowColSum<<<gridDim, blockDim>>>(d_M, d_rowsum, d_colsum);
+//
+//     // 将结果从设备复制回主机
+//     cudaMemcpy(rowsum, d_rowsum, N * sizeof(int), cudaMemcpyDeviceToHost);
+//     cudaMemcpy(colsum, d_colsum, N * sizeof(int), cudaMemcpyDeviceToHost);
+//
+//     // 释放设备内存
+//     cudaFree(d_M);
+//     cudaFree(d_rowsum);
+//     cudaFree(d_colsum);
+// }
+//
+// /* CUDA核函数 - 列求和 */
+// __global__ void cudaColumnSum(int *M, int *colsum)
+// {
+//     int col = blockIdx.x * blockDim.x + threadIdx.x;
+//     if (col < N) {
+//         colsum[col] = 0;
+//         for (int row = 0; row < N; row++) {
+//             colsum[col] += M[row * N + col];
+//         }
+//     }
+// }
+//
+// /* CUDA核函数 - 行列求和 */
+// __global__ void cudaRowColSum(int *M, int *rowsum, int *colsum)
+// {
+//     int idx = blockIdx.x * blockDim.x + threadIdx.x;
+//     if (idx < N) {
+//         // 计算行和
+//         rowsum[idx] = 0;
+//         for (int j = 0; j < N; j++) {
+//             rowsum[idx] += M[idx * N + j];
+//         }
+//
+//         // 计算列和
+//         colsum[idx] = 0;
+//         for (int i = 0; i < N; i++) {
+//             colsum[idx] += M[i * N + idx];
+//         }
+//     }
+// }
+//
+// /* 
+// 	这个表格包含多个数组元素，每一组元素（函数名字, COL/ROWCOL, "描述字符串"）
+// 	COL表示该函数仅仅计算每一列的和
+// 	ROWCOL表示该函数计算每一行、每一列的和
+// 	将你认为最好的两个实现，放在最前面。
+// 	比如：
+// 	{my_c_sum1, "超级垃圾列求和实现"},
+// 	{my_rc_sum2, "好一点的行列求和实现"},
+// */
+//
+// rc_fun_rec rc_fun_tab[] = 
+// {
+//
+//   /* 第一项，应当是你写的最好列求和的函数实现 */
+//     {cuda_c_sum, COL, "CUDA optimized column sum"},
+//   /* 第二项，应当是你写的最好行列求和的函数实现 */
+//     {cuda_rc_sum, ROWCOL, "CUDA optimized row and column sum"},
+//
+//     {c_sum, COL, "Column sum, reference implementation"},
+//
+//     {rc_sum, ROWCOL, "Row and column sum, reference implementation"},
+//
+//  /* 下面的代码不能修改或者删除！！表明数组列表结束 */
+//     {NULL,ROWCOL,NULL}
+// };

perflab/matrix/rowcol_202302723005.c

@@ -0,0 +1,69 @@
+/**************************************************************************
+        ??/???????????????????????????????
+        1. ???????????????????????????????
+        2. ??????????????????????
+        3. ??rc_fun_rec rc_fun_tab??????????????????
+                ???????????????????????????????????????????
+***************************************************************************/
+
+/*
+        ????201209054233
+        ??????????????
+*/
+
+#include "rowcol.h"
+#include <math.h>
+#include <stdio.h>
+#include <stdlib.h>
+
+/* ????????????????? */
+/* ???????????????????????????????????????????????
+        ??????????2?????????????????
+*/
+
+void c_sum(matrix_t M, vector_t rowsum, vector_t colsum) {
+  int i, j;
+  for (j = 0; j < N; j++) {
+    colsum[j] = 0;
+    for (i = 0; i < N; i++)
+      colsum[j] += M[i][j];
+  }
+}
+
+/* ???????????????????? */
+/* ??????????????????????? */
+
+void rc_sum(matrix_t M, vector_t rowsum, vector_t colsum) {
+  int i, j;
+  for (i = 0; i < N; i++) {
+    rowsum[i] = colsum[i] = 0;
+    for (j = 0; j < N; j++) {
+      rowsum[i] += M[i][j];
+      colsum[i] += M[j][i];
+    }
+  }
+}
+
+/*
+        ????????????????????????????????????????, COL/ROWCOL, "?????????"??
+        COL??????????????????????
+        ROWCOL???????????????????????
+        ?????????????????????????????
+        ????
+        {my_c_sum1, "?????????????????"},
+        {my_rc_sum2, "??????????????????"},
+*/
+
+rc_fun_rec rc_fun_tab[] = {
+
+    /* ???????????????????????????????? */
+    {c_sum, COL, "Best column sum"},
+    /* ?????????????????????????????????? */
+    {rc_sum, ROWCOL, "Best row and column sum"},
+
+    {c_sum, COL, "Column sum, reference implementation"},
+
+    {rc_sum, ROWCOL, "Row and column sum, reference implementation"},
+
+    /* ??????????????????????????????????????? */
+    {NULL, ROWCOL, NULL}};

perflab/matrix/rowcol_test.c

@@ -1,9 +1,9 @@
 #include <stdio.h>
 #include <stdlib.h>
-//#include <random.h>
-#include "rowcol.h"
-#include "fcyc.h"
+// #include <random.h>
 #include "clock.h"
+#include "fcyc.h"
+#include "rowcol.h"
 
 #define MAX_ITER_COUNT 100
 
@@ -11,9 +11,9 @@
 static struct {
   double cref;  /* Cycles taken by reference solution */
   double cbest; /* Cycles taken by our best implementation */
-} cstandard[2] = 
-{{7.7, 6.40}, /* Column Sum */
- {9.75, 6.60} /* Row & Column Sum */
+} cstandard[2] = {
+    {7.7, 6.40}, /* Column Sum */
+    {9.75, 6.60} /* Row & Column Sum */
 };
 
 /* Put in code to align matrix so that it starts on a cache block boundary.
@@ -26,7 +26,7 @@ static struct {
 #define WPB 16
 
 int verbose = 1;
-int data[N*N+WPB];
+int data[N * N + WPB];
 int *mstart;
 
 typedef vector_t *row_t;
@@ -37,137 +37,122 @@ vector_t rsref, csref, rcomp, ccomp;
 static void init_tests(void);
 extern void make_CPU_busy(void);
 
-static void init_tests(void)
-{
-    int i, j;
-    size_t bytes_per_block = sizeof(int) * WPB;
-    /* round mstart up to nearest block boundary */
-    mstart = (int *)
-      (((size_t) data + bytes_per_block-1) / bytes_per_block * bytes_per_block);
-    for (i = 0; i < N; i++) {
-	rsref[i] = csref[i] = 0;
-    }
-    for (i = 0; i < N; i++) {
-	for (j = 0; j < N; j++) {
-	    int val = rand();
-	    mstart[i*N+j] = val;
-	    rsref[i] += val;
-	    csref[j] += val;
-	}
+static void init_tests(void) {
+  int i, j;
+  size_t bytes_per_block = sizeof(int) * WPB;
+  /* round mstart up to nearest block boundary */
+  mstart = (int *)(((size_t)data + bytes_per_block - 1) / bytes_per_block *
+                   bytes_per_block);
+  for (i = 0; i < N; i++) {
+    rsref[i] = csref[i] = 0;
+  }
+  for (i = 0; i < N; i++) {
+    for (j = 0; j < N; j++) {
+      int val = rand();
+      mstart[i * N + j] = val;
+      rsref[i] += val;
+      csref[j] += val;
     }
+  }
 }
 
-
 /* Test function on all values */
 int test_rc(rc_fun f, FILE *rpt, rc_comp_t rc_type) {
-    int i;
-    int ok = 1;
+  int i;
+  int ok = 1;
 
-    for (i = 0; i < N; i++)
-	rcomp[i] = ccomp[i] = 0xDEADBEEF;
-    f((row_t)mstart, rcomp, ccomp);
-
-    for (i = 0; ok && i < N; i++) {
-	if (rc_type == ROWCOL
-	    && rsref[i] != rcomp[i]) {
-	    ok = 0;
-	    if (rpt)
-		fprintf(rpt,
-			"对第%d行的计算出错！正确结果是%d，但是计算得到%d\n",
-			i, rsref[i], rcomp[i]);
-	}
-	if ((rc_type == ROWCOL || rc_type == COL)
-		 && csref[i] != ccomp[i]) {
-	    ok = 0;
-	    if (rpt)
-		fprintf(rpt,
-			"对第%d列的计算出错！正确结果是%d，但是计算得到%d\n",
-			i, csref[i], ccomp[i]);
-	}
+  for (i = 0; i < N; i++)
+    rcomp[i] = ccomp[i] = 0xDEADBEEF;
+  f((row_t)mstart, rcomp, ccomp);
 
+  for (i = 0; ok && i < N; i++) {
+    if (rc_type == ROWCOL && rsref[i] != rcomp[i]) {
+      ok = 0;
+      if (rpt)
+        fprintf(rpt, "对第%d行的计算出错！正确结果是%d，但是计算得到%d\n", i,
+                rsref[i], rcomp[i]);
     }
-    return ok;
+    if ((rc_type == ROWCOL || rc_type == COL) && csref[i] != ccomp[i]) {
+      ok = 0;
+      if (rpt)
+        fprintf(rpt, "对第%d列的计算出错！正确结果是%d，但是计算得到%d\n", i,
+                csref[i], ccomp[i]);
+    }
+  }
+  return ok;
 }
 
 /* Kludgy way to interface to cycle measuring code */
-void do_test(int *intf)
-{
-  rc_fun f = (rc_fun) intf;
+void do_test(int *intf) {
+  rc_fun f = (rc_fun)intf;
   f((row_t)mstart, rcomp, ccomp);
 }
 
-void time_rc(rc_fun f, rc_comp_t rc_type, char *descr, double *cycp)
-{
-	int i;
-  int *intf = (int *) f;
+void time_rc(rc_fun f, rc_comp_t rc_type, char *descr, double *cycp) {
+  int i;
+  int *intf = (int *)f;
   double t, cme;
   t = 0;
-  if (verbose) printf("函数：%s\n", descr);
+  if (verbose)
+    printf("函数：%s\n", descr);
   if (test_rc(f, stdout, rc_type)) {
-  	make_CPU_busy();
-  	for (i=0;i<MAX_ITER_COUNT;i++)
-    	t += fcyc(do_test, intf);
-    t = t/MAX_ITER_COUNT;
-    cme = t/(N*N);
-    if (verbose) printf("  总周期数 = %.2f, 平均周期/元素 = %.2f\n",
-	   t, cme);
+    make_CPU_busy();
+    for (i = 0; i < MAX_ITER_COUNT; i++)
+      t += fcyc((void (*)(long *))do_test, intf);
+    t = t / MAX_ITER_COUNT;
+    cme = t / (N * N);
+    if (verbose)
+      printf("  总周期数 = %.2f, 平均周期/元素 = %.2f\n", t, cme);
     if (cycp)
       *cycp = cme;
   }
 }
 
 /* Compute the grade achieved by function */
-static double compute_score(double cmeas, double cref, double cbest)
-{
-  double sbest = cref/cbest;
-  double smeas = cref/cmeas;
-  if (smeas < 0.1*(sbest-1)+1)
+static double compute_score(double cmeas, double cref, double cbest) {
+  double sbest = cref / cbest;
+  double smeas = cref / cmeas;
+  if (smeas < 0.1 * (sbest - 1) + 1)
     return 0;
-  if (smeas > 1.1*(sbest-1)+1)
+  if (smeas > 1.1 * (sbest - 1) + 1)
     return 120;
-  return 100*((smeas-1.0)/(sbest-1.0) + 0.1);
+  return 100 * ((smeas - 1.0) / (sbest - 1.0) + 0.1);
 }
 
-int main(int argc, char *argv[])
-{
+int main(int argc, char *argv[]) {
   int i;
   double cme;
-  double cme_c,cme_rc;
-  int EnableScore=0;
-  
-  if (argc == 3)
-  {
-  	EnableScore = 1;
-  	verbose = 0;
+  double cme_c, cme_rc;
+  int EnableScore = 0;
+
+  if (argc == 3) {
+    EnableScore = 1;
+    verbose = 0;
   }
   init_tests();
-  set_fcyc_clear_cache(1);  /* Set so that clears cache between runs */
+  set_fcyc_clear_cache(1); /* Set so that clears cache between runs */
   for (i = 0; rc_fun_tab[i].f != NULL; i++) {
-      cme = 100.0;
-      time_rc(rc_fun_tab[i].f,
-	    rc_fun_tab[i].rc_type, rc_fun_tab[i].descr, &cme);
-    if (i == 0)
-    {
-    	cme_c = cme;
-    	if (EnableScore==0)
-    	{
-      printf("  最高\"列求和\"得分   ======================== %.0f\n",
-	     compute_score(cme, cstandard[0].cref, cstandard[0].cbest));
-	    }
-	  }
-    if (i == 1)
-    {
-    	cme_rc = cme;
-    	if (EnableScore==0)
-    	{
-      printf("  最高\"行和列求和\"得分 ====================== %.0f\n",
-	     compute_score(cme, cstandard[1].cref, cstandard[1].cbest));
-	    }
-	  }
+    cme = 100.0;
+    time_rc(rc_fun_tab[i].f, rc_fun_tab[i].rc_type, rc_fun_tab[i].descr, &cme);
+    if (i == 0) {
+      cme_c = cme;
+      if (EnableScore == 0) {
+        printf("  最高\"列求和\"得分   ======================== %.0f\n",
+               compute_score(cme, cstandard[0].cref, cstandard[0].cbest));
+      }
+    }
+    if (i == 1) {
+      cme_rc = cme;
+      if (EnableScore == 0) {
+        printf("  最高\"行和列求和\"得分 ====================== %.0f\n",
+               compute_score(cme, cstandard[1].cref, cstandard[1].cbest));
+      }
+    }
   }
-  
+
   if (EnableScore)
-  	printf("%.2f\t %.0f\t %.2f\t %.0f\t 0\t 0\n",cme_c,compute_score(cme_c, cstandard[0].cref, cstandard[0].cbest),
-  	cme_rc,compute_score(cme_rc, cstandard[1].cref, cstandard[1].cbest));
+    printf("%.2f\t %.0f\t %.2f\t %.0f\t 0\t 0\n", cme_c,
+           compute_score(cme_c, cstandard[0].cref, cstandard[0].cbest), cme_rc,
+           compute_score(cme_rc, cstandard[1].cref, cstandard[1].cbest));
   return 0;
 }