gh0s7/csapp2025
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

matrix fixed

Browse Source
This commit is contained in:
CGH0S7
2025-04-12 11:37:07 +08:00
parent b92e49bd71
commit ba21f80f3b
24 changed files with 1840 additions and 569 deletions
Download Patch File Download Diff File Expand all files Collapse all files

35
perflab/poly/Makefile Normal file
View File

@@ -0,0 +1,35 @@
CC = gcc
NVCC = nvcc
CFLAGS = -Wall -O2 -g
CUDA_FLAGS = -O2 -g
LDFLAGS = -lm -lcudart
# Source files
SRCS = poly_test.c clock.c cpe.c fcyc.c lsquare.c
CUDA_SRCS = poly.cu
OBJS = $(SRCS:.c=.o) poly.o
# Target executable
TARGET = poly_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
poly.o: poly.cu
$(NVCC) $(CUDA_FLAGS) -c $< -o $@
# Clean rule
clean:
rm -f $(OBJS) $(TARGET)
# Phony targets
.PHONY: all clean

314
perflab/poly/clock.c
View File

@@ -13,11 +13,11 @@
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <intrin.h>
//#include <intrinsics.h>
#include <windows.h>
#include <time.h>
#include <x86intrin.h>
// #include <intrinsics.h>
#include "clock.h"
#include <time.h>
#include <windows.h>
/* Use x86 cycle counter */
@@ -27,203 +27,195 @@ 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)
{
void access_counter(unsigned *hi, unsigned *lo) {
long long counter;
long long counter;
counter = __rdtsc();
(*hi) = (unsigned int)(counter >> 32);
(*lo) = (unsigned int)counter;
/*
counter = __rdtsc();
(*hi) = (unsigned int)(counter >> 32);
(*lo) = (unsigned int)counter;
/*
LARGE_INTEGER lPerformanceCount;
LARGE_INTEGER lPerformanceCount;
QueryPerformanceCounter(&lPerformanceCount);
(*hi) = (unsigned int)lPerformanceCount.HighPart;
(*lo) = (unsigned int)lPerformanceCount.LowPart;
// printf("%08X %08X\n",(*hi),(*lo));
*/
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);
}
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;
double get_counter() {
unsigned ncyc_hi, ncyc_lo;
unsigned hi, lo, borrow;
double result;
/* Get cycle counter */
access_counter(&ncyc_hi, &ncyc_lo);
/* 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;
/* 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();
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);
// CPU<EFBFBD><EFBFBD>Ƶ<EFBFBD><EFBFBD>
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);
// <20><><EFBFBD><EFBFBD><EFBFBD>ϸ߾<CFB8><DFBE>ȶ<EFBFBD>ʱ<EFBFBD><CAB1><EFBFBD>ľ<EFBFBD><C4BE><EFBFBD>Ƶ<EFBFBD><C6B5>
// <20><><EFBFBD><EFBFBD><EFBFBD>ʱ<EFBFBD><CAB1>Ӧ<EFBFBD>þ<EFBFBD><C3BE><EFBFBD>һƄ1<C684>78253<35><33><EFBFBD><EFBFBD>8254
// <20><>intel ich7<EFBFBD>м<EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>8254
QueryPerformanceFrequency(&lFrequency);
// if (verbose>0)
// printf("<EFBFBD>߾<EFBFBD><EFBFBD>ȶ<EFBFBD>ʱ<EFBFBD><EFBFBD><EFBFBD>ľ<EFBFBD><EFBFBD><EFBFBD>Ƶ<EFBFBD>ʣ<EFBFBD>%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);
// <20><><EFBFBD><EFBFBD><EFBFBD>ʱ<EFBFBD><CAB1>ÿ<EFBFBD><C3BF><EFBFBD><EFBFBD>һ<EFBFBD><D2BB>ʱ<EFBFBD><CAB1><EFBFBD><EFBFBD><EFBFBD>ڣ<EFBFBD><DAA3><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>+1
QueryPerformanceCounter(&lPerformanceCount_Start);
//RDTSC指令:获取CPU经历的时钟周期数
_i64StartCpuCounter=__rdtsc();
// RDTSCָ<EFBFBD><EFBFBD>:<3A><>ȡCPU<50><55><EFBFBD><EFBFBD><EFBFBD><EFBFBD>ʱ<EFBFBD><CAB1><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>
_i64StartCpuCounter = __rdtsc();
//延时长一点,误差会小一点
//int nTemp=100000;
//while (--nTemp);
Sleep(200);
// <20><>ʱ<EFBFBD><CAB1>һ<EFBFBD><D2BB>,<2C><><EFBFBD><EFBFBD>Сһ<D0A1><D2BB>
// int nTemp=100000;
// while (--nTemp);
Sleep(200);
QueryPerformanceCounter(&lPerformanceCount_End);
QueryPerformanceCounter(&lPerformanceCount_End);
_i64EndCpuCounter=__rdtsc();
_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;
// f=1/T => f=<EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>/(<28><><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>*T)
// <20><><EFBFBD><EFBFBD>ġ<EFBFBD><C4A1><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>ￄ1<EFBF84>7*T<><54><EFBFBD><EFBFBD><EFBFBD><EFBFBD>ʱ<EFBFBD><CAB1>ￄ1<EFBF84>7
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;
mhz = (_i64EndCpuCounter - _i64StartCpuCounter) / (fTime * 1000000.0);
if (verbose > 0)
printf("CPUƵ<EFBFBD><EFBFBD>Ϊ:%1.6fMHz.\n", mhz);
return mhz;
}
double CPU_Factor1(void)
{
double result;
int i,j,k,ii,jj,kk;
LARGE_INTEGER lStart,lEnd;
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;
ii = 43273;
kk = 1238;
result = 1;
jj = 1244;
hThread=GetCurrentThread();
SetThreadAffinityMask(hThread,0x1);
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;
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;
result = get_counter();
QueryPerformanceCounter(&lEnd);
fTime = ((double)lEnd.QuadPart - (double)lStart.QuadPart);
printf("CPU<EFBFBD><EFBFBD><EFBFBD><EFBFBD>ʱ<EFBFBD><EFBFBD>Ϊ%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;
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;
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;
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);
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;
// 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;
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;
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;
}

325
perflab/poly/poly.cu Normal file
View File

@@ -0,0 +1,325 @@
/**************************************************************************
多项式计算函数。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. 实现不同版本的多项式计算函数；
3. 编辑peval_fun_rec peval_fun_tab数组，将你的最好的答案
（最小CPE、最小C10）作为数组的前两项
***************************************************************************/
/*
学号：201209054233
姓名：夜半加班狂
*/
#include <stdio.h>
#include <stdlib.h>
#include <cuda_runtime.h>
typedef int (*peval_fun)(int*, int, int);
typedef struct {
peval_fun f;
char *descr;
} peval_fun_rec, *peval_fun_ptr;
/**************************************************************************
Edit this comment to indicate your name and Andrew ID
#ifdef ASSIGN
Submission by Harry Q. Bovik, bovik@andrew.cmu.edu
#else
Instructor's version.
Created by Randal E. Bryant, Randy.Bryant@cs.cmu.edu, 10/07/02
#endif
***************************************************************************/
/*
实现一个指定的常系数多项式计算
第一次，请直接运行程序，以便获知你需要实现的常系数是啥
*/
int const_poly_eval(int *not_use, int not_use2, int x)
{
int result = 0;
/* int i;
int xpwr = 1; // x的幂次
int a[4] = {21,90,42,88};
for (i = 0; i <= 3; i++) {
result += a[i]*xpwr;
xpwr *= x;
}
*/
// 90 = 64 + 32 - 4 - 2
// 42 = 32 + 8 + 2
// 88 = 64 + 16 + 8
int x64,x32,x16,x8,x4,x2;
x64 = x << 6;
x32 = x << 5;
x16 = x << 4;
x8 = x << 3;
x4 = x << 2;
x2 = x << 1;
result = 21 + x64+x32-x4-x2 + ((x32+x8+x2) + (x64+x16+x8)*x)*x;
return result;
}
/* 多项式计算函数。注意：这个只是一个参考实现，你需要实现自己的版本 */
/*
友情提示：lcc支持ATT格式的嵌入式汇编，例如
_asm("movl %eax,%ebx");
_asm("pushl %edx");
可以在lcc中project->configuration->Compiler->Code Generation->Generate .asm，
将其选中后，可以在lcc目录下面生成对应程序的汇编代码实现。通过查看汇编文件，
你可以了解编译器是如何实现你的代码的。有些实现可能非常低效。
你可以在适当的地方加入嵌入式汇编，来大幅度提高计算性能。
*/
int poly_eval(int *a, int degree, int x)
{
int result = 0;
int i;
int xpwr = 1; /* x的幂次 */
// printf("阶=%d\n",degree);
for (i = 0; i <= degree; i++) {
result += a[i]*xpwr;
xpwr *= x;
}
return result;
}
/* CUDA优化的多项式计算函数 - 低CPE版本 */
int cuda_poly_eval_low_cpe(int *a, int degree, int x)
{
// 对于低CPE版本，我们使用CUDA并行计算多项式的各个项
// 然后将结果传回主机进行求和
// 分配设备内存
int *d_a, *d_results;
cudaError_t err;
// 分配内存
err = cudaMalloc(&d_a, (degree + 1) * sizeof(int));
if (err != cudaSuccess) {
printf("CUDA Error: %s\n", cudaGetErrorString(err));
return 0;
}
err = cudaMalloc(&d_results, (degree + 1) * sizeof(int));
if (err != cudaSuccess) {
printf("CUDA Error: %s\n", cudaGetErrorString(err));
cudaFree(d_a);
return 0;
}
// 将系数从主机复制到设备
err = cudaMemcpy(d_a, a, (degree + 1) * sizeof(int), cudaMemcpyHostToDevice);
if (err != cudaSuccess) {
printf("CUDA Error: %s\n", cudaGetErrorString(err));
cudaFree(d_a);
cudaFree(d_results);
return 0;
}
// 定义CUDA核函数
dim3 blockDim(256);
dim3 gridDim((degree + 1 + blockDim.x - 1) / blockDim.x);
// 启动核函数
cudaPolyEvalLowCPE<<<gridDim, blockDim>>>(d_a, degree, x, d_results);
// 检查核函数执行错误
err = cudaGetLastError();
if (err != cudaSuccess) {
printf("CUDA Error: %s\n", cudaGetErrorString(err));
cudaFree(d_a);
cudaFree(d_results);
return 0;
}
// 分配主机内存用于结果
int *h_results = (int *)malloc((degree + 1) * sizeof(int));
if (h_results == NULL) {
printf("Memory allocation error\n");
cudaFree(d_a);
cudaFree(d_results);
return 0;
}
// 将结果从设备复制回主机
err = cudaMemcpy(h_results, d_results, (degree + 1) * sizeof(int), cudaMemcpyDeviceToHost);
if (err != cudaSuccess) {
printf("CUDA Error: %s\n", cudaGetErrorString(err));
free(h_results);
cudaFree(d_a);
cudaFree(d_results);
return 0;
}
// 在主机上求和
int result = 0;
for (int i = 0; i <= degree; i++) {
result += h_results[i];
}
// 释放内存
free(h_results);
cudaFree(d_a);
cudaFree(d_results);
return result;
}
/* CUDA优化的多项式计算函数 - 10阶优化版本 */
int cuda_poly_eval_degree10(int *a, int degree, int x)
{
// 对于10阶多项式，我们可以使用更优化的方法
// 使用CUDA并行计算，但针对10阶多项式进行特殊优化
// 分配设备内存
int *d_a, *d_result;
cudaError_t err;
// 分配内存
err = cudaMalloc(&d_a, (degree + 1) * sizeof(int));
if (err != cudaSuccess) {
printf("CUDA Error: %s\n", cudaGetErrorString(err));
return 0;
}
err = cudaMalloc(&d_result, sizeof(int));
if (err != cudaSuccess) {
printf("CUDA Error: %s\n", cudaGetErrorString(err));
cudaFree(d_a);
return 0;
}
// 将系数从主机复制到设备
err = cudaMemcpy(d_a, a, (degree + 1) * sizeof(int), cudaMemcpyHostToDevice);
if (err != cudaSuccess) {
printf("CUDA Error: %s\n", cudaGetErrorString(err));
cudaFree(d_a);
cudaFree(d_result);
return 0;
}
// 定义CUDA核函数
dim3 blockDim(256);
dim3 gridDim(1); // 只需要一个块，因为我们只需要一个结果
// 启动核函数
cudaPolyEvalDegree10<<<gridDim, blockDim>>>(d_a, degree, x, d_result);
// 检查核函数执行错误
err = cudaGetLastError();
if (err != cudaSuccess) {
printf("CUDA Error: %s\n", cudaGetErrorString(err));
cudaFree(d_a);
cudaFree(d_result);
return 0;
}
// 获取结果
int result;
err = cudaMemcpy(&result, d_result, sizeof(int), cudaMemcpyDeviceToHost);
if (err != cudaSuccess) {
printf("CUDA Error: %s\n", cudaGetErrorString(err));
cudaFree(d_a);
cudaFree(d_result);
return 0;
}
// 释放内存
cudaFree(d_a);
cudaFree(d_result);
return result;
}
/* CUDA核函数 - 低CPE版本 */
__global__ void cudaPolyEvalLowCPE(int *a, int degree, int x, int *results)
{
int idx = blockIdx.x * blockDim.x + threadIdx.x;
if (idx <= degree) {
// 计算x的幂
int xpwr = 1;
for (int i = 0; i < idx; i++) {
xpwr *= x;
}
// 计算这一项的结果
results[idx] = a[idx] * xpwr;
}
}
/* CUDA核函数 - 10阶优化版本 */
__global__ void cudaPolyEvalDegree10(int *a, int degree, int x, int *result)
{
// 使用共享内存来存储中间结果
__shared__ int shared_result;
// 只有第一个线程初始化共享结果
if (threadIdx.x == 0) {
shared_result = 0;
}
__syncthreads();
// 每个线程计算一部分项
int local_result = 0;
int xpwr = 1;
// 计算x的幂
for (int i = 0; i < threadIdx.x; i++) {
xpwr *= x;
}
// 计算这一项的结果
if (threadIdx.x <= degree) {
local_result = a[threadIdx.x] * xpwr;
}
// 使用原子操作累加结果
atomicAdd(&shared_result, local_result);
// 同步所有线程
__syncthreads();
// 只有第一个线程将结果写回全局内存
if (threadIdx.x == 0) {
*result = shared_result;
}
}
/*
这个表格包含多个数组元素，每一组元素（函数名字, "描述字符串"）
将你认为最好的两个实现，放在最前面。
比如：
{my_poly_eval1, "超级垃圾实现"},
{my_poly_eval2, "好一点的实现"},
*/
peval_fun_rec peval_fun_tab[] =
{
/* 第一项，应当是你写的最好CPE的函数实现 */
{cuda_poly_eval_low_cpe, "CUDA optimized low CPE implementation"},
/* 第二项，应当是你写的在10阶时具有最好性能的实现 */
{cuda_poly_eval_degree10, "CUDA optimized degree 10 implementation"},
{poly_eval, "poly_eval: 参考实现"},
/* 下面的代码不能修改或者删除！！表明数组列表结束 */
{NULL, ""}
};

BIN
perflab/poly/poly.o Normal file
View File

Binary file not shown.

35
perflab/poly/poly_test.c
View File

@@ -6,6 +6,7 @@
#include "poly.h"
#include "cpe.h"
#include "clock.h"
#include <time.h>
double CPU_Mhz;
@@ -17,7 +18,7 @@ static int coeff[MAXDEGREE+1];
#define MAX_ITER_COUNT 100
#define REF_CPU_MHZ 2292.6 // 这是我的处理器主频
#define REF_CPU_MHZ 2292.6 // <EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>ҵĴ<EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>Ƶ
/* Define performance standards */
static struct {
@@ -26,7 +27,7 @@ static struct {
} cstandard[3] =
{{4.00, 1.75}, /* CPE */
{50, 43}, /* C(10) */
{57,31} /* 常系数多项式计算 */
{57,31} /* <EFBFBD><EFBFBD>ϵ<EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>ʽ<EFBFBD><EFBFBD><EFBFBD><EFBFBD> */
};
int coeff_const[4];
@@ -82,7 +83,7 @@ static void init_const_poly(void)
coeff_const[i] = rand_div+10;
}
printf("你需要修改poly.c的const_poly_eval函数，实现下面的常数多项式计算！\n");
printf("<EFBFBD><EFBFBD><EFBFBD><EFBFBD>Ҫ<EFBFBD>޸<EFBFBD>poly.c<EFBFBD><EFBFBD>const_poly_eval<EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>ʵ<EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>ij<EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>ʽ<EFBFBD><EFBFBD><EFBFBD>㣡\n");
printf("\tresult=%d+%d*x+%d*x^2+%d*x^3\n",coeff_const[0],coeff_const[1],coeff_const[2],coeff_const[3]);
fixval_const = ref_poly_eval(coeff_const, 3, xval);
@@ -97,15 +98,15 @@ void test_const_poly(void)
int my_cal = const_poly_eval(coeff_const, 3, xval);
if (fixval_const != my_cal)
{
printf("常系数多项式计算const_poly_eval实现错误（x=%d），预期结果是%d，但是计算得到的是%d\n",xval,fixval_const,my_cal);
printf("<EFBFBD><EFBFBD>ϵ<EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>ʽ<EFBFBD><EFBFBD><EFBFBD><EFBFBD>const_poly_evalʵ<EFBFBD>ִ<EFBFBD><EFBFBD><EFBFBD>x=%d<><64><EFBFBD><EFBFBD>Ԥ<EFBFBD>ڽ<EFBFBD><DABD><EFBFBD><EFBFBD>%d<><64><EFBFBD><EFBFBD><EFBFBD>Ǽ<EFBFBD><C7BC><EFBFBD>õ<EFBFBD><C3B5><EFBFBD><EFBFBD><EFBFBD>%d\n",xval,fixval_const,my_cal);
exit(0);
}
fix_time = 0;
for (i=0;i<MAX_ITER_COUNT;i++)
fix_time += measure_function(run_fun_const, 3);
fix_time = fix_time / MAX_ITER_COUNT;
printf(" 常系数多项式计算时间 = %.1f\n", fix_time);
printf(" 最高的常系数多项式计算得分 ============== %.0f\n",
printf(" <EFBFBD><EFBFBD>ϵ<EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>ʽ<EFBFBD><EFBFBD><EFBFBD><EFBFBD>ʱ<EFBFBD><EFBFBD> = %.1f\n", fix_time);
printf(" <EFBFBD><EFBFBD>ߵij<EFBFBD>ϵ<EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>ʽ<EFBFBD><EFBFBD><EFBFBD><EFBFBD>÷<EFBFBD> ============== %.0f\n",
compute_score(fix_time, cstandard[2].cref, cstandard[2].cbest));
}
@@ -132,7 +133,7 @@ int test_poly(peval_fun f, FILE *rpt) {
ok = 0;
if (rpt) {
fprintf(rpt,
"错误！多项式计算不对！阶=%d时，计算的值是%d，而正确值是%d\n",
"<EFBFBD><EFBFBD><EFBFBD>󣡶<EFBFBD><EFBFBD><EFBFBD>ʽ<EFBFBD><EFBFBD><EFBFBD>㲻<EFBFBD>ԣ<EFBFBD><EFBFBD><EFBFBD>=%dʱ<64><CAB1><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>ֵ<EFBFBD><D6B5>%d<><64><EFBFBD><EFBFBD><EFBFBD><EFBFBD>ȷֵ<C8B7><D6B5>%d\n",
MAXDEGREE-i, v, pval[i]);
}
}
@@ -142,7 +143,7 @@ int test_poly(peval_fun f, FILE *rpt) {
ok = 0;
if (rpt) {
fprintf(rpt,
"错误！多项式计算不对！阶=%d时，计算的值是%d，而正确值是%d\n",
"<EFBFBD><EFBFBD><EFBFBD>󣡶<EFBFBD><EFBFBD><EFBFBD>ʽ<EFBFBD><EFBFBD><EFBFBD>㲻<EFBFBD>ԣ<EFBFBD><EFBFBD><EFBFBD>=%dʱ<64><CAB1><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>ֵ<EFBFBD><D6B5>%d<><64><EFBFBD><EFBFBD><EFBFBD><EFBFBD>ȷֵ<C8B7><D6B5>%d\n",
FIXDEGREE, v, fixval);
}
}
@@ -175,7 +176,7 @@ void run_poly(peval_fun f, char *descr, double *cpep, double *cfixp)
double cpe=0;
double fix_time=0;
pfun = f;
printf("函数：%s\n", descr);
printf("<EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>%s\n", descr);
if (test_poly(f, stdout)) {
cpe = 0;
for (i=0;i<MAX_ITER_COUNT;i++)
@@ -206,7 +207,7 @@ static double compute_score(double cmeas, double cref, double cbest)
return 100*((smeas-1.0)/(sbest-1.0) + 0.1);
}
/* 产生一个0~divv-1之间的随机数，同时更新随机数种子 */
/* <EFBFBD><EFBFBD><EFBFBD><EFBFBD>һ<EFBFBD><EFBFBD>0~divv-1֮<EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>ͬʱ<EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD> */
void GenerateRandomNumber(unsigned long divv)
{
unsigned long long x = rand1_h;
@@ -230,18 +231,18 @@ int main(int argc, char *argv[])
// CPU_Factor();
// GetCpuClock();
printf("\t2015多项式优化实验，欢迎你！\n");
printf("\t2015<EFBFBD><EFBFBD><EFBFBD><EFBFBD>ʽ<EFBFBD>Ż<EFBFBD>ʵ<EFBFBD>飬<EFBFBD><EFBFBD>ӭ<EFBFBD>㣡\n");
printf("============================\n");
if (argc == 1)
{
printf("使用方法：%s 学号后6位 [学号后6位] [学号后6位] ...\n",argv[0]);
printf("你需要依据提示改写poly.c程序，实现一个常系数多项式的计算，尽可能快哦....\n");
printf("另外，你需要改写poly.c程序，实现任意阶的多项式计算和10阶的多项式计算，要快！\n");
printf("ʹ<EFBFBD>÷<EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>%s ѧ<>ź<EFBFBD>6λ [ѧ<>ź<EFBFBD>6λ] [ѧ<>ź<EFBFBD>6λ] ...\n",argv[0]);
printf("<EFBFBD><EFBFBD><EFBFBD><EFBFBD>Ҫ<EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>ʾ<EFBFBD><EFBFBD>дpoly.c<><63><EFBFBD><EFBFBD>ʵ<EFBFBD><CAB5>һ<EFBFBD><D2BB><EFBFBD><EFBFBD>ϵ<EFBFBD><CFB5><EFBFBD><EFBFBD><EFBFBD><EFBFBD>ʽ<EFBFBD>ļ<EFBFBD><C4BC>㣬<EFBFBD><E3A3AC><EFBFBD><EFBFBD><EFBFBD>ܿ<EFBFBD>Ŷ....\n");
printf("<EFBFBD><EFBFBD><EFBFBD>⣬<EFBFBD><EFBFBD><EFBFBD><EFBFBD>Ҫ<EFBFBD><EFBFBD>дpoly.c<><63><EFBFBD><EFBFBD>ʵ<EFBFBD><CAB5><EFBFBD><EFBFBD><EFBFBD><EFBFBD>׵Ķ<D7B5><C4B6><EFBFBD>ʽ<EFBFBD><CABD><EFBFBD><EFBFBD><EFBFBD>10<31>׵Ķ<D7B5><C4B6><EFBFBD>ʽ<EFBFBD><CABD><EFBFBD>㣬Ҫ<E3A3AC>죡\n");
return 0;
}
/*依据学号，初始化一个随机数发生器*/
/*<EFBFBD><EFBFBD><EFBFBD><EFBFBD>ѧ<EFBFBD>ţ<EFBFBD><EFBFBD><EFBFBD>ʼ<EFBFBD><EFBFBD>һ<EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>*/
rand1_h = (unsigned long)atoi(argv[1]);
rand1_l=0x29A;
GenerateRandomNumber(0);
@@ -266,10 +267,10 @@ int main(int argc, char *argv[])
//make_CPU_busy();
run_poly(peval_fun_tab[i].f, peval_fun_tab[i].descr, &cpe, &cfix);
if (i == 0)
printf(" 最高的CPE得分 =========================== %.0f\n",
printf(" <EFBFBD><EFBFBD>ߵ<EFBFBD>CPE<EFBFBD>÷<EFBFBD> =========================== %.0f\n",
compute_score(cpe, cstandard[0].cref, cstandard[0].cbest));
if (i == 1)
printf(" 最高的C(10)得分 ========================= %.0f\n",
printf(" <EFBFBD><EFBFBD>ߵ<EFBFBD>C(10)<EFBFBD>÷<EFBFBD> ========================= %.0f\n",
compute_score(cfix, cstandard[1].cref, cstandard[1].cbest));
}
return 0;

BIN
perflab/poly/poly_test.o Normal file
View File

Binary file not shown.