Initial commit
commit
1667f3744a
|
@ -0,0 +1,33 @@
|
|||
#
|
||||
# Student makefile for Cache Lab
|
||||
# Note: requires a 64-bit x86-64 system
|
||||
#
|
||||
CC = gcc
|
||||
CFLAGS = -g -Wall -Werror -std=c99 -m64
|
||||
|
||||
all: csim test-trans tracegen
|
||||
# Generate a handin tar file each time you compile
|
||||
-tar -cvf ${USER}-handin.tar csim.c trans.c
|
||||
|
||||
csim: csim.c cachelab.c cachelab.h
|
||||
$(CC) $(CFLAGS) -o csim csim.c cachelab.c -lm
|
||||
|
||||
test-trans: test-trans.c trans.o cachelab.c cachelab.h
|
||||
$(CC) $(CFLAGS) -o test-trans test-trans.c cachelab.c trans.o
|
||||
|
||||
tracegen: tracegen.c trans.o cachelab.c
|
||||
$(CC) $(CFLAGS) -O0 -o tracegen tracegen.c trans.o cachelab.c
|
||||
|
||||
trans.o: trans.c
|
||||
$(CC) $(CFLAGS) -O0 -c trans.c
|
||||
|
||||
#
|
||||
# Clean the src dirctory
|
||||
#
|
||||
clean:
|
||||
rm -rf *.o
|
||||
rm -f *.tar
|
||||
rm -f csim
|
||||
rm -f test-trans tracegen
|
||||
rm -f trace.all trace.f*
|
||||
rm -f .csim_results .marker
|
|
@ -0,0 +1,39 @@
|
|||
This is the handout directory for the CS:APP Cache Lab.
|
||||
|
||||
************************
|
||||
Running the autograders:
|
||||
************************
|
||||
|
||||
Before running the autograders, compile your code:
|
||||
linux> make
|
||||
|
||||
Check the correctness of your simulator:
|
||||
linux> ./test-csim
|
||||
|
||||
Check the correctness and performance of your transpose functions:
|
||||
linux> ./test-trans -M 32 -N 32
|
||||
linux> ./test-trans -M 64 -N 64
|
||||
linux> ./test-trans -M 61 -N 67
|
||||
|
||||
Check everything at once (this is the program that your instructor runs):
|
||||
linux> ./driver.py
|
||||
|
||||
******
|
||||
Files:
|
||||
******
|
||||
|
||||
# You will modifying and handing in these two files
|
||||
csim.c Your cache simulator
|
||||
trans.c Your transpose function
|
||||
|
||||
# Tools for evaluating your simulator and transpose function
|
||||
Makefile Builds the simulator and tools
|
||||
README This file
|
||||
driver.py* The driver program, runs test-csim and test-trans
|
||||
cachelab.c Required helper functions
|
||||
cachelab.h Required header file
|
||||
csim-ref* The executable reference cache simulator
|
||||
test-csim* Tests your cache simulator
|
||||
test-trans.c Tests your transpose function
|
||||
tracegen.c Helper program used by test-trans
|
||||
traces/ Trace files used by test-csim.c
|
|
@ -0,0 +1,83 @@
|
|||
/*
|
||||
* cachelab.c - Cache Lab helper functions
|
||||
*/
|
||||
#include <stdio.h>
|
||||
#include <stdlib.h>
|
||||
#include <assert.h>
|
||||
#include "cachelab.h"
|
||||
#include <time.h>
|
||||
|
||||
trans_func_t func_list[MAX_TRANS_FUNCS];
|
||||
int func_counter = 0;
|
||||
|
||||
/*
|
||||
* printSummary - Summarize the cache simulation statistics. Student cache simulators
|
||||
* must call this function in order to be properly autograded.
|
||||
*/
|
||||
void printSummary(int hits, int misses, int evictions)
|
||||
{
|
||||
printf("hits:%d misses:%d evictions:%d\n", hits, misses, evictions);
|
||||
FILE* output_fp = fopen(".csim_results", "w");
|
||||
assert(output_fp);
|
||||
fprintf(output_fp, "%d %d %d\n", hits, misses, evictions);
|
||||
fclose(output_fp);
|
||||
}
|
||||
|
||||
/*
|
||||
* initMatrix - Initialize the given matrix
|
||||
*/
|
||||
void initMatrix(int M, int N, int A[N][M], int B[M][N])
|
||||
{
|
||||
int i, j;
|
||||
srand(time(NULL));
|
||||
for (i = 0; i < N; i++){
|
||||
for (j = 0; j < M; j++){
|
||||
// A[i][j] = i+j; /* The matrix created this way is symmetric */
|
||||
A[i][j]=rand();
|
||||
B[j][i]=rand();
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
void randMatrix(int M, int N, int A[N][M]) {
|
||||
int i, j;
|
||||
srand(time(NULL));
|
||||
for (i = 0; i < N; i++){
|
||||
for (j = 0; j < M; j++){
|
||||
// A[i][j] = i+j; /* The matrix created this way is symmetric */
|
||||
A[i][j]=rand();
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
/*
|
||||
* correctTrans - baseline transpose function used to evaluate correctness
|
||||
*/
|
||||
void correctTrans(int M, int N, int A[N][M], int B[M][N])
|
||||
{
|
||||
int i, j, tmp;
|
||||
for (i = 0; i < N; i++){
|
||||
for (j = 0; j < M; j++){
|
||||
tmp = A[i][j];
|
||||
B[j][i] = tmp;
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
|
||||
|
||||
/*
|
||||
* registerTransFunction - Add the given trans function into your list
|
||||
* of functions to be tested
|
||||
*/
|
||||
void registerTransFunction(void (*trans)(int M, int N, int[N][M], int[M][N]),
|
||||
char* desc)
|
||||
{
|
||||
func_list[func_counter].func_ptr = trans;
|
||||
func_list[func_counter].description = desc;
|
||||
func_list[func_counter].correct = 0;
|
||||
func_list[func_counter].num_hits = 0;
|
||||
func_list[func_counter].num_misses = 0;
|
||||
func_list[func_counter].num_evictions =0;
|
||||
func_counter++;
|
||||
}
|
|
@ -0,0 +1,37 @@
|
|||
/*
|
||||
* cachelab.h - Prototypes for Cache Lab helper functions
|
||||
*/
|
||||
|
||||
#ifndef CACHELAB_TOOLS_H
|
||||
#define CACHELAB_TOOLS_H
|
||||
|
||||
#define MAX_TRANS_FUNCS 100
|
||||
|
||||
typedef struct trans_func{
|
||||
void (*func_ptr)(int M,int N,int[N][M],int[M][N]);
|
||||
char* description;
|
||||
char correct;
|
||||
unsigned int num_hits;
|
||||
unsigned int num_misses;
|
||||
unsigned int num_evictions;
|
||||
} trans_func_t;
|
||||
|
||||
/*
|
||||
* printSummary - This function provides a standard way for your cache
|
||||
* simulator * to display its final hit and miss statistics
|
||||
*/
|
||||
void printSummary(int hits, /* number of hits */
|
||||
int misses, /* number of misses */
|
||||
int evictions); /* number of evictions */
|
||||
|
||||
/* Fill the matrix with data */
|
||||
void initMatrix(int M, int N, int A[N][M], int B[M][N]);
|
||||
|
||||
/* The baseline trans function that produces correct results. */
|
||||
void correctTrans(int M, int N, int A[N][M], int B[M][N]);
|
||||
|
||||
/* Add the given function to the function list */
|
||||
void registerTransFunction(
|
||||
void (*trans)(int M,int N,int[N][M],int[M][N]), char* desc);
|
||||
|
||||
#endif /* CACHELAB_TOOLS_H */
|
|
@ -0,0 +1,13 @@
|
|||
#include "cachelab.h"
|
||||
|
||||
struct cache_line {
|
||||
unsigned char v;
|
||||
unsigned int tag;
|
||||
unsigned long timestamp;
|
||||
};
|
||||
|
||||
int main()
|
||||
{
|
||||
printSummary(0, 0, 0);
|
||||
return 0;
|
||||
}
|
|
@ -0,0 +1,137 @@
|
|||
#!/usr//bin/python2.7
|
||||
#
|
||||
# driver.py - The driver tests the correctness of the student's cache
|
||||
# simulator and the correctness and performance of their transpose
|
||||
# function. It uses ./test-csim to check the correctness of the
|
||||
# simulator and it runs ./test-trans on three different sized
|
||||
# matrices (32x32, 64x64, and 61x67) to test the correctness and
|
||||
# performance of the transpose function.
|
||||
#
|
||||
import subprocess;
|
||||
import re;
|
||||
import os;
|
||||
import sys;
|
||||
import optparse;
|
||||
|
||||
#
|
||||
# computeMissScore - compute the score depending on the number of
|
||||
# cache misses
|
||||
#
|
||||
def computeMissScore(miss, lower, upper, full_score):
|
||||
if miss <= lower:
|
||||
return full_score
|
||||
if miss >= upper:
|
||||
return 0
|
||||
|
||||
score = (miss - lower) * 1.0
|
||||
range = (upper- lower) * 1.0
|
||||
return round((1 - score / range) * full_score, 1)
|
||||
|
||||
#
|
||||
# main - Main function
|
||||
#
|
||||
def main():
|
||||
|
||||
# Configure maxscores here
|
||||
maxscore= {};
|
||||
maxscore['csim'] = 51
|
||||
maxscore['transc'] = 1
|
||||
maxscore['trans32'] = 15
|
||||
maxscore['trans64'] = 15
|
||||
maxscore['trans61'] = 19
|
||||
|
||||
# Parse the command line arguments
|
||||
p = optparse.OptionParser()
|
||||
p.add_option("-A", action="store_true", dest="autograde",
|
||||
help="emit autoresult string for Autolab");
|
||||
opts, args = p.parse_args()
|
||||
autograde = opts.autograde
|
||||
|
||||
# Check the correctness of the cache simulator
|
||||
print "Part A: Testing cache simulator"
|
||||
print "Running ./test-csim"
|
||||
p = subprocess.Popen("./test-csim",
|
||||
shell=True, stdout=subprocess.PIPE)
|
||||
stdout_data = p.communicate()[0]
|
||||
|
||||
# Emit the output from test-csim
|
||||
stdout_data = re.split('\n', stdout_data)
|
||||
for line in stdout_data:
|
||||
if re.match("TEST_CSIM_RESULTS", line):
|
||||
resultsim = re.findall(r'(\d+)', line)
|
||||
else:
|
||||
print "%s" % (line)
|
||||
|
||||
# Check the correctness and performance of the transpose function
|
||||
# 32x32 transpose
|
||||
print "Part B: Testing transpose function"
|
||||
print "Running ./test-trans -M 32 -N 32"
|
||||
p = subprocess.Popen("./test-trans -M 32 -N 32 | grep TEST_TRANS_RESULTS",
|
||||
shell=True, stdout=subprocess.PIPE)
|
||||
stdout_data = p.communicate()[0]
|
||||
result32 = re.findall(r'(\d+)', stdout_data)
|
||||
|
||||
# 64x64 transpose
|
||||
print "Running ./test-trans -M 64 -N 64"
|
||||
p = subprocess.Popen("./test-trans -M 64 -N 64 | grep TEST_TRANS_RESULTS",
|
||||
shell=True, stdout=subprocess.PIPE)
|
||||
stdout_data = p.communicate()[0]
|
||||
result64 = re.findall(r'(\d+)', stdout_data)
|
||||
|
||||
# 61x67 transpose
|
||||
print "Running ./test-trans -M 61 -N 67"
|
||||
p = subprocess.Popen("./test-trans -M 61 -N 67 | grep TEST_TRANS_RESULTS",
|
||||
shell=True, stdout=subprocess.PIPE)
|
||||
stdout_data = p.communicate()[0]
|
||||
result61 = re.findall(r'(\d+)', stdout_data)
|
||||
|
||||
# Compute the scores for each step
|
||||
csim_cscore = map(int, resultsim[0:1])
|
||||
trans_cscore = int(result32[0]) * int(result64[0]) * int(result61[0]);
|
||||
miss32 = int(result32[1])
|
||||
miss64 = int(result64[1])
|
||||
miss61 = int(result61[1])
|
||||
trans32_score = computeMissScore(miss32, 300, 600, maxscore['trans32']) * int(result32[0])
|
||||
trans64_score = computeMissScore(miss64, 1300, 2000, maxscore['trans64']) * int(result64[0])
|
||||
trans61_score = computeMissScore(miss61, 2000, 3000, maxscore['trans61']) * int(result61[0])
|
||||
total_score = csim_cscore[0] + trans32_score + trans64_score + trans61_score
|
||||
|
||||
# Summarize the results
|
||||
print "\nCache Lab summary:"
|
||||
print "%-22s%8s%10s%12s" % ("", "Points", "Max pts", "Misses")
|
||||
print "%-22s%8.1f%10d" % ("Csim correctness", csim_cscore[0],
|
||||
maxscore['csim'])
|
||||
|
||||
misses = str(miss32)
|
||||
if miss32 == 2**31-1 :
|
||||
misses = "invalid"
|
||||
print "%-22s%8.1f%10d%12s" % ("Trans perf 32x32", trans32_score,
|
||||
maxscore['trans32'], misses)
|
||||
|
||||
misses = str(miss64)
|
||||
if miss64 == 2**31-1 :
|
||||
misses = "invalid"
|
||||
print "%-22s%8.1f%10d%12s" % ("Trans perf 64x64", trans64_score,
|
||||
maxscore['trans64'], misses)
|
||||
|
||||
misses = str(miss61)
|
||||
if miss61 == 2**31-1 :
|
||||
misses = "invalid"
|
||||
print "%-22s%8.1f%10d%12s" % ("Trans perf 61x67", trans61_score,
|
||||
maxscore['trans61'], misses)
|
||||
|
||||
print "%22s%8.1f%10d" % ("Total points", total_score,
|
||||
maxscore['csim'] +
|
||||
maxscore['trans32'] +
|
||||
maxscore['trans64'] +
|
||||
maxscore['trans61'])
|
||||
|
||||
# Emit autoresult string for Autolab if called with -A option
|
||||
if autograde:
|
||||
autoresult="%.1f:%d:%d:%d" % (total_score, miss32, miss64, miss61)
|
||||
print "\nAUTORESULT_STRING=%s" % autoresult
|
||||
|
||||
|
||||
# execute main only if called as a script
|
||||
if __name__ == "__main__":
|
||||
main()
|
|
@ -0,0 +1,261 @@
|
|||
/*
|
||||
* test-trans.c - Checks the correctness and performance of all of the
|
||||
* student's transpose functions and records the results for their
|
||||
* official submitted version as well.
|
||||
*/
|
||||
#include <stdio.h>
|
||||
#include <stdlib.h>
|
||||
#include <assert.h>
|
||||
#include <unistd.h>
|
||||
#include <string.h>
|
||||
#include <signal.h>
|
||||
#include <getopt.h>
|
||||
#include <sys/types.h>
|
||||
#include "cachelab.h"
|
||||
#include <sys/wait.h> // fir WEXITSTATUS
|
||||
#include <limits.h> // for INT_MAX
|
||||
|
||||
/* Maximum array dimension */
|
||||
#define MAXN 256
|
||||
|
||||
/* The description string for the transpose_submit() function that the
|
||||
student submits for credit */
|
||||
#define SUBMIT_DESCRIPTION "Transpose submission"
|
||||
|
||||
/* External function defined in trans.c */
|
||||
extern void registerFunctions();
|
||||
|
||||
/* External variables defined in cachelab-tools.c */
|
||||
extern trans_func_t func_list[MAX_TRANS_FUNCS];
|
||||
extern int func_counter;
|
||||
|
||||
/* Globals set on the command line */
|
||||
static int M = 0;
|
||||
static int N = 0;
|
||||
|
||||
/* The correctness and performance for the submitted transpose function */
|
||||
struct results {
|
||||
int funcid;
|
||||
int correct;
|
||||
int misses;
|
||||
};
|
||||
static struct results results = {-1, 0, INT_MAX};
|
||||
|
||||
/*
|
||||
* eval_perf - Evaluate the performance of the registered transpose functions
|
||||
*/
|
||||
void eval_perf(unsigned int s, unsigned int E, unsigned int b)
|
||||
{
|
||||
int i,flag;
|
||||
unsigned int len, hits, misses, evictions;
|
||||
unsigned long long int marker_start, marker_end, addr;
|
||||
char buf[1000], cmd[255];
|
||||
char filename[128];
|
||||
|
||||
registerFunctions();
|
||||
|
||||
/* Open the complete trace file */
|
||||
FILE* full_trace_fp;
|
||||
FILE* part_trace_fp;
|
||||
|
||||
/* Evaluate the performance of each registered transpose function */
|
||||
|
||||
for (i=0; i<func_counter; i++) {
|
||||
if (strcmp(func_list[i].description, SUBMIT_DESCRIPTION) == 0 )
|
||||
results.funcid = i; /* remember which function is the submission */
|
||||
|
||||
|
||||
printf("\nFunction %d (%d total)\nStep 1: Validating and generating memory traces\n",i,func_counter);
|
||||
/* Use valgrind to generate the trace */
|
||||
|
||||
sprintf(cmd, "valgrind --tool=lackey --trace-mem=yes --log-fd=1 -v ./tracegen -M %d -N %d -F %d > trace.tmp", M, N,i);
|
||||
flag=WEXITSTATUS(system(cmd));
|
||||
if (0!=flag) {
|
||||
printf("Validation error at function %d! Run ./tracegen -M %d -N %d -F %d for details.\nSkipping performance evaluation for this function.\n",flag-1,M,N,i);
|
||||
continue;
|
||||
}
|
||||
|
||||
/* Get the start and end marker addresses */
|
||||
FILE* marker_fp = fopen(".marker", "r");
|
||||
assert(marker_fp);
|
||||
fscanf(marker_fp, "%llx %llx", &marker_start, &marker_end);
|
||||
fclose(marker_fp);
|
||||
|
||||
|
||||
func_list[i].correct=1;
|
||||
|
||||
/* Save the correctness of the transpose submission */
|
||||
if (results.funcid == i ) {
|
||||
results.correct = 1;
|
||||
}
|
||||
|
||||
full_trace_fp = fopen("trace.tmp", "r");
|
||||
assert(full_trace_fp);
|
||||
|
||||
|
||||
/* Filtered trace for each transpose function goes in a separate file */
|
||||
sprintf(filename, "trace.f%d", i);
|
||||
part_trace_fp = fopen(filename, "w");
|
||||
assert(part_trace_fp);
|
||||
|
||||
/* Locate trace corresponding to the trans function */
|
||||
flag = 0;
|
||||
while (fgets(buf, 1000, full_trace_fp) != NULL) {
|
||||
|
||||
/* We are only interested in memory access instructions */
|
||||
if (buf[0]==' ' && buf[2]==' ' &&
|
||||
(buf[1]=='S' || buf[1]=='M' || buf[1]=='L' )) {
|
||||
sscanf(buf+3, "%llx,%u", &addr, &len);
|
||||
|
||||
/* If start marker found, set flag */
|
||||
if (addr == marker_start)
|
||||
flag = 1;
|
||||
|
||||
/* Valgrind creates many spurious accesses to the
|
||||
stack that have nothing to do with the students
|
||||
code. At the moment, we are ignoring all stack
|
||||
accesses by using the simple filter of recording
|
||||
accesses to only the low 32-bit portion of the
|
||||
address space. At some point it would be nice to
|
||||
try to do more informed filtering so that would
|
||||
eliminate the valgrind stack references while
|
||||
include the student stack references. */
|
||||
if (flag && addr < 0xffffffff) {
|
||||
fputs(buf, part_trace_fp);
|
||||
}
|
||||
|
||||
/* if end marker found, close trace file */
|
||||
if (addr == marker_end) {
|
||||
flag = 0;
|
||||
fclose(part_trace_fp);
|
||||
break;
|
||||
}
|
||||
}
|
||||
}
|
||||
fclose(full_trace_fp);
|
||||
|
||||
/* Run the reference simulator */
|
||||
printf("Step 2: Evaluating performance (s=%d, E=%d, b=%d)\n", s, E, b);
|
||||
char cmd[255];
|
||||
sprintf(cmd, "./csim-ref -s %u -E %u -b %u -t trace.f%d > /dev/null",
|
||||
s, E, b, i);
|
||||
system(cmd);
|
||||
|
||||
/* Collect results from the reference simulator */
|
||||
FILE* in_fp = fopen(".csim_results","r");
|
||||
assert(in_fp);
|
||||
fscanf(in_fp, "%u %u %u", &hits, &misses, &evictions);
|
||||
fclose(in_fp);
|
||||
func_list[i].num_hits = hits;
|
||||
func_list[i].num_misses = misses;
|
||||
func_list[i].num_evictions = evictions;
|
||||
printf("func %u (%s): hits:%u, misses:%u, evictions:%u\n",
|
||||
i, func_list[i].description, hits, misses, evictions);
|
||||
|
||||
/* If it is transpose_submit(), record number of misses */
|
||||
if (results.funcid == i) {
|
||||
results.misses = misses;
|
||||
}
|
||||
}
|
||||
|
||||
}
|
||||
|
||||
/*
|
||||
* usage - Print usage info
|
||||
*/
|
||||
void usage(char *argv[]){
|
||||
printf("Usage: %s [-h] -M <rows> -N <cols>\n", argv[0]);
|
||||
printf("Options:\n");
|
||||
printf(" -h Print this help message.\n");
|
||||
printf(" -M <rows> Number of matrix rows (max %d)\n", MAXN);
|
||||
printf(" -N <cols> Number of matrix columns (max %d)\n", MAXN);
|
||||
printf("Example: %s -M 8 -N 8\n", argv[0]);
|
||||
}
|
||||
|
||||
/*
|
||||
* sigsegv_handler - SIGSEGV handler
|
||||
*/
|
||||
void sigsegv_handler(int signum){
|
||||
printf("Error: Segmentation Fault.\n");
|
||||
printf("TEST_TRANS_RESULTS=0:0\n");
|
||||
fflush(stdout);
|
||||
exit(1);
|
||||
}
|
||||
|
||||
/*
|
||||
* sigalrm_handler - SIGALRM handler
|
||||
*/
|
||||
void sigalrm_handler(int signum){
|
||||
printf("Error: Program timed out.\n");
|
||||
printf("TEST_TRANS_RESULTS=0:0\n");
|
||||
fflush(stdout);
|
||||
exit(1);
|
||||
}
|
||||
|
||||
/*
|
||||
* main - Main routine
|
||||
*/
|
||||
int main(int argc, char* argv[])
|
||||
{
|
||||
char c;
|
||||
|
||||
while ((c = getopt(argc,argv,"M:N:h")) != -1) {
|
||||
switch(c) {
|
||||
case 'M':
|
||||
M = atoi(optarg);
|
||||
break;
|
||||
case 'N':
|
||||
N = atoi(optarg);
|
||||
break;
|
||||
case 'h':
|
||||
usage(argv);
|
||||
exit(0);
|
||||
default:
|
||||
usage(argv);
|
||||
exit(1);
|
||||
}
|
||||
}
|
||||
|
||||
if (M == 0 || N == 0) {
|
||||
printf("Error: Missing required argument\n");
|
||||
usage(argv);
|
||||
exit(1);
|
||||
}
|
||||
|
||||
if (M > MAXN || N > MAXN) {
|
||||
printf("Error: M or N exceeds %d\n", MAXN);
|
||||
usage(argv);
|
||||
exit(1);
|
||||
}
|
||||
|
||||
/* Install SIGSEGV and SIGALRM handlers */
|
||||
if (signal(SIGSEGV, sigsegv_handler) == SIG_ERR) {
|
||||
fprintf(stderr, "Unable to install SIGALRM handler\n");
|
||||
exit(1);
|
||||
}
|
||||
|
||||
if (signal(SIGALRM, sigalrm_handler) == SIG_ERR) {
|
||||
fprintf(stderr, "Unable to install SIGALRM handler\n");
|
||||
exit(1);
|
||||
}
|
||||
|
||||
/* Time out and give up after a while */
|
||||
alarm(120);
|
||||
|
||||
/* Check the performance of the student's transpose function */
|
||||
eval_perf(5, 1, 5);
|
||||
|
||||
/* Emit the results for this particular test */
|
||||
if (results.funcid == -1) {
|
||||
printf("\nError: We could not find your transpose_submit() function\n");
|
||||
printf("Error: Please ensure that description field is exactly \"%s\"\n",
|
||||
SUBMIT_DESCRIPTION);
|
||||
printf("\nTEST_TRANS_RESULTS=0:0\n");
|
||||
}
|
||||
else {
|
||||
printf("\nSummary for official submission (func %d): correctness=%d misses=%d\n",
|
||||
results.funcid, results.correct, results.misses);
|
||||
printf("\nTEST_TRANS_RESULTS=%d:%d\n", results.correct, results.misses);
|
||||
}
|
||||
return 0;
|
||||
}
|
|
@ -0,0 +1,107 @@
|
|||
/*
|
||||
* tracegen.c - Running the binary tracegen with valgrind produces
|
||||
* a memory trace of all of the registered transpose functions.
|
||||
*
|
||||
* The beginning and end of each registered transpose function's trace
|
||||
* is indicated by reading from "marker" addresses. These two marker
|
||||
* addresses are recorded in file for later use.
|
||||
*/
|
||||
|
||||
#include <stdlib.h>
|
||||
#include <stdio.h>
|
||||
#include <assert.h>
|
||||
#include <unistd.h>
|
||||
#include <getopt.h>
|
||||
#include "cachelab.h"
|
||||
#include <string.h>
|
||||
|
||||
/* External variables declared in cachelab.c */
|
||||
extern trans_func_t func_list[MAX_TRANS_FUNCS];
|
||||
extern int func_counter;
|
||||
|
||||
/* External function from trans.c */
|
||||
extern void registerFunctions();
|
||||
|
||||
/* Markers used to bound trace regions of interest */
|
||||
volatile char MARKER_START, MARKER_END;
|
||||
|
||||
static int A[256][256];
|
||||
static int B[256][256];
|
||||
static int M;
|
||||
static int N;
|
||||
|
||||
|
||||
int validate(int fn,int M, int N, int A[N][M], int B[M][N]) {
|
||||
int C[M][N];
|
||||
memset(C,0,sizeof(C));
|
||||
correctTrans(M,N,A,C);
|
||||
for(int i=0;i<M;i++) {
|
||||
for(int j=0;j<N;j++) {
|
||||
if(B[i][j]!=C[i][j]) {
|
||||
printf("Validation failed on function %d! Expected %d but got %d at B[%d][%d]\n",fn,C[i][j],B[i][j],i,j);
|
||||
return 0;
|
||||
}
|
||||
}
|
||||
}
|
||||
return 1;
|
||||
}
|
||||
|
||||
int main(int argc, char* argv[]){
|
||||
int i;
|
||||
|
||||
char c;
|
||||
int selectedFunc=-1;
|
||||
while( (c=getopt(argc,argv,"M:N:F:")) != -1){
|
||||
switch(c){
|
||||
case 'M':
|
||||
M = atoi(optarg);
|
||||
break;
|
||||
case 'N':
|
||||
N = atoi(optarg);
|
||||
break;
|
||||
case 'F':
|
||||
selectedFunc = atoi(optarg);
|
||||
break;
|
||||
case '?':
|
||||
default:
|
||||
printf("./tracegen failed to parse its options.\n");
|
||||
exit(1);
|
||||
}
|
||||
}
|
||||
|
||||
|
||||
/* Register transpose functions */
|
||||
registerFunctions();
|
||||
|
||||
/* Fill A with data */
|
||||
initMatrix(M,N, A, B);
|
||||
|
||||
/* Record marker addresses */
|
||||
FILE* marker_fp = fopen(".marker","w");
|
||||
assert(marker_fp);
|
||||
fprintf(marker_fp, "%llx %llx",
|
||||
(unsigned long long int) &MARKER_START,
|
||||
(unsigned long long int) &MARKER_END );
|
||||
fclose(marker_fp);
|
||||
|
||||
if (-1==selectedFunc) {
|
||||
/* Invoke registered transpose functions */
|
||||
for (i=0; i < func_counter; i++) {
|
||||
MARKER_START = 33;
|
||||
(*func_list[i].func_ptr)(M, N, A, B);
|
||||
MARKER_END = 34;
|
||||
if (!validate(i,M,N,A,B))
|
||||
return i+1;
|
||||
}
|
||||
} else {
|
||||
MARKER_START = 33;
|
||||
(*func_list[selectedFunc].func_ptr)(M, N, A, B);
|
||||
MARKER_END = 34;
|
||||
if (!validate(selectedFunc,M,N,A,B))
|
||||
return selectedFunc+1;
|
||||
|
||||
}
|
||||
return 0;
|
||||
}
|
||||
|
||||
|
|
@ -0,0 +1,5 @@
|
|||
L 10,4
|
||||
S 18,4
|
||||
L 20,4
|
||||
S 28,4
|
||||
S 50,4
|
File diff suppressed because it is too large
Load Diff
|
@ -0,0 +1,596 @@
|
|||
S 00600aa0,1
|
||||
I 004005b6,5
|
||||
I 004005bb,5
|
||||
I 004005c0,5
|
||||
S 7ff000398,8
|
||||
I 0040051e,1
|
||||
S 7ff000390,8
|
||||
I 0040051f,3
|
||||
I 00400522,4
|
||||
S 7ff000378,8
|
||||
I 00400526,4
|
||||
S 7ff000370,8
|
||||
I 0040052a,7
|
||||
S 7ff000384,4
|
||||
I 00400531,2
|
||||
I 00400581,4
|
||||
L 7ff000384,4
|
||||
I 00400585,2
|
||||
I 00400533,7
|
||||
S 7ff000388,4
|
||||
I 0040053a,2
|
||||
I 00400577,4
|
||||
L 7ff000388,4
|
||||
I 0040057b,2
|
||||
I 0040053c,3
|
||||
L 7ff000384,4
|
||||
I 0040053f,2
|
||||
I 00400541,4
|
||||
I 00400545,3
|
||||
I 00400548,4
|
||||
L 7ff000378,8
|
||||
I 0040054c,3
|
||||
L 7ff000388,4
|
||||
I 0040054f,2
|
||||
I 00400551,3
|
||||
L 00600a20,4
|
||||
I 00400554,3
|
||||
S 7ff00038c,4
|
||||
I 00400557,3
|
||||
L 7ff000388,4
|
||||
I 0040055a,2
|
||||
I 0040055c,4
|
||||
I 00400560,3
|
||||
I 00400563,4
|
||||
L 7ff000370,8
|
||||
I 00400567,3
|
||||
L 7ff000384,4
|
||||
I 0040056a,3
|
||||
I 0040056d,3
|
||||
L 7ff00038c,4
|
||||
I 00400570,3
|
||||
S 00600a60,4
|
||||
I 00400573,4
|
||||
M 7ff000388,4
|
||||
I 00400577,4
|
||||
L 7ff000388,4
|
||||
I 0040057b,2
|
||||
I 0040053c,3
|
||||
L 7ff000384,4
|
||||
I 0040053f,2
|
||||
I 00400541,4
|
||||
I 00400545,3
|
||||
I 00400548,4
|
||||
L 7ff000378,8
|
||||
I 0040054c,3
|
||||
L 7ff000388,4
|
||||
I 0040054f,2
|
||||
I 00400551,3
|
||||
L 00600a24,4
|
||||
I 00400554,3
|
||||
S 7ff00038c,4
|
||||
I 00400557,3
|
||||
L 7ff000388,4
|
||||
I 0040055a,2
|
||||
I 0040055c,4
|
||||
I 00400560,3
|
||||
I 00400563,4
|
||||
L 7ff000370,8
|
||||
I 00400567,3
|
||||
L 7ff000384,4
|
||||
I 0040056a,3
|
||||
I 0040056d,3
|
||||
L 7ff00038c,4
|
||||
I 00400570,3
|
||||
S 00600a70,4
|
||||
I 00400573,4
|
||||
M 7ff000388,4
|
||||
I 00400577,4
|
||||
L 7ff000388,4
|
||||
I 0040057b,2
|
||||
I 0040053c,3
|
||||
L 7ff000384,4
|
||||
I 0040053f,2
|
||||
I 00400541,4
|
||||
I 00400545,3
|
||||
I 00400548,4
|
||||
L 7ff000378,8
|
||||
I 0040054c,3
|
||||
L 7ff000388,4
|
||||
I 0040054f,2
|
||||
I 00400551,3
|
||||
L 00600a28,4
|
||||
I 00400554,3
|
||||
S 7ff00038c,4
|
||||
I 00400557,3
|
||||
L 7ff000388,4
|
||||
I 0040055a,2
|
||||
I 0040055c,4
|
||||
I 00400560,3
|
||||
I 00400563,4
|
||||
L 7ff000370,8
|
||||
I 00400567,3
|
||||
L 7ff000384,4
|
||||
I 0040056a,3
|
||||
I 0040056d,3
|
||||
L 7ff00038c,4
|
||||
I 00400570,3
|
||||
S 00600a80,4
|
||||
I 00400573,4
|
||||
M 7ff000388,4
|
||||
I 00400577,4
|
||||
L 7ff000388,4
|
||||
I 0040057b,2
|
||||
I 0040053c,3
|
||||
L 7ff000384,4
|
||||
I 0040053f,2
|
||||
I 00400541,4
|
||||
I 00400545,3
|
||||
I 00400548,4
|
||||
L 7ff000378,8
|
||||
I 0040054c,3
|
||||
L 7ff000388,4
|
||||
I 0040054f,2
|
||||
I 00400551,3
|
||||
L 00600a2c,4
|
||||
I 00400554,3
|
||||
S 7ff00038c,4
|
||||
I 00400557,3
|
||||
L 7ff000388,4
|
||||
I 0040055a,2
|
||||
I 0040055c,4
|
||||
I 00400560,3
|
||||
I 00400563,4
|
||||
L 7ff000370,8
|
||||
I 00400567,3
|
||||
L 7ff000384,4
|
||||
I 0040056a,3
|
||||
I 0040056d,3
|
||||
L 7ff00038c,4
|
||||
I 00400570,3
|
||||
S 00600a90,4
|
||||
I 00400573,4
|
||||
M 7ff000388,4
|
||||
I 00400577,4
|
||||
L 7ff000388,4
|
||||
I 0040057b,2
|
||||
I 0040057d,4
|
||||
M 7ff000384,4
|
||||
I 00400581,4
|
||||
L 7ff000384,4
|
||||
I 00400585,2
|
||||
I 00400533,7
|
||||
S 7ff000388,4
|
||||
I 0040053a,2
|
||||
I 00400577,4
|
||||
L 7ff000388,4
|
||||
I 0040057b,2
|
||||
I 0040053c,3
|
||||
L 7ff000384,4
|
||||
I 0040053f,2
|
||||
I 00400541,4
|
||||
I 00400545,3
|
||||
I 00400548,4
|
||||
L 7ff000378,8
|
||||
I 0040054c,3
|
||||
L 7ff000388,4
|
||||
I 0040054f,2
|
||||
I 00400551,3
|
||||
L 00600a30,4
|
||||
I 00400554,3
|
||||
S 7ff00038c,4
|
||||
I 00400557,3
|
||||
L 7ff000388,4
|
||||
I 0040055a,2
|
||||
I 0040055c,4
|
||||
I 00400560,3
|
||||
I 00400563,4
|
||||
L 7ff000370,8
|
||||
I 00400567,3
|
||||
L 7ff000384,4
|
||||
I 0040056a,3
|
||||
I 0040056d,3
|
||||
L 7ff00038c,4
|
||||
I 00400570,3
|
||||
S 00600a64,4
|
||||
I 00400573,4
|
||||
M 7ff000388,4
|
||||
I 00400577,4
|
||||
L 7ff000388,4
|
||||
I 0040057b,2
|
||||
I 0040053c,3
|
||||
L 7ff000384,4
|
||||
I 0040053f,2
|
||||
I 00400541,4
|
||||
I 00400545,3
|
||||
I 00400548,4
|
||||
L 7ff000378,8
|
||||
I 0040054c,3
|
||||
L 7ff000388,4
|
||||
I 0040054f,2
|
||||
I 00400551,3
|
||||
L 00600a34,4
|
||||
I 00400554,3
|
||||
S 7ff00038c,4
|
||||
I 00400557,3
|
||||
L 7ff000388,4
|
||||
I 0040055a,2
|
||||
I 0040055c,4
|
||||
I 00400560,3
|
||||
I 00400563,4
|
||||
L 7ff000370,8
|
||||
I 00400567,3
|
||||
L 7ff000384,4
|
||||
I 0040056a,3
|
||||
I 0040056d,3
|
||||
L 7ff00038c,4
|
||||
I 00400570,3
|
||||
S 00600a74,4
|
||||
I 00400573,4
|
||||
M 7ff000388,4
|
||||
I 00400577,4
|
||||
L 7ff000388,4
|
||||
I 0040057b,2
|
||||
I 0040053c,3
|
||||
L 7ff000384,4
|
||||
I 0040053f,2
|
||||
I 00400541,4
|
||||
I 00400545,3
|
||||
I 00400548,4
|
||||
L 7ff000378,8
|
||||
I 0040054c,3
|
||||
L 7ff000388,4
|
||||
I 0040054f,2
|
||||
I 00400551,3
|
||||
L 00600a38,4
|
||||
I 00400554,3
|
||||
S 7ff00038c,4
|
||||
I 00400557,3
|
||||
L 7ff000388,4
|
||||
I 0040055a,2
|
||||
I 0040055c,4
|
||||
I 00400560,3
|
||||
I 00400563,4
|
||||
L 7ff000370,8
|
||||
I 00400567,3
|
||||
L 7ff000384,4
|
||||
I 0040056a,3
|
||||
I 0040056d,3
|
||||
L 7ff00038c,4
|
||||
I 00400570,3
|
||||
S 00600a84,4
|
||||
I 00400573,4
|
||||
M 7ff000388,4
|
||||
I 00400577,4
|
||||
L 7ff000388,4
|
||||
I 0040057b,2
|
||||
I 0040053c,3
|
||||
L 7ff000384,4
|
||||
I 0040053f,2
|
||||
I 00400541,4
|
||||
I 00400545,3
|
||||
I 00400548,4
|
||||
L 7ff000378,8
|
||||
I 0040054c,3
|
||||
L 7ff000388,4
|
||||
I 0040054f,2
|
||||
I 00400551,3
|
||||
L 00600a3c,4
|
||||
I 00400554,3
|
||||
S 7ff00038c,4
|
||||
I 00400557,3
|
||||
L 7ff000388,4
|
||||
I 0040055a,2
|
||||
I 0040055c,4
|
||||
I 00400560,3
|
||||
I 00400563,4
|
||||
L 7ff000370,8
|
||||
I 00400567,3
|
||||
L 7ff000384,4
|
||||
I 0040056a,3
|
||||
I 0040056d,3
|
||||
L 7ff00038c,4
|
||||
I 00400570,3
|
||||
S 00600a94,4
|
||||
I 00400573,4
|
||||
M 7ff000388,4
|
||||
I 00400577,4
|
||||
L 7ff000388,4
|
||||
I 0040057b,2
|
||||
I 0040057d,4
|
||||
M 7ff000384,4
|
||||
I 00400581,4
|
||||
L 7ff000384,4
|
||||
I 00400585,2
|
||||
I 00400533,7
|
||||
S 7ff000388,4
|
||||
I 0040053a,2
|
||||
I 00400577,4
|
||||
L 7ff000388,4
|
||||
I 0040057b,2
|
||||
I 0040053c,3
|
||||
L 7ff000384,4
|
||||
I 0040053f,2
|
||||
I 00400541,4
|
||||
I 00400545,3
|
||||
I 00400548,4
|
||||
L 7ff000378,8
|
||||
I 0040054c,3
|
||||
L 7ff000388,4
|
||||
I 0040054f,2
|
||||
I 00400551,3
|
||||
L 00600a40,4
|
||||
I 00400554,3
|
||||
S 7ff00038c,4
|
||||
I 00400557,3
|
||||
L 7ff000388,4
|
||||
I 0040055a,2
|
||||
I 0040055c,4
|
||||
I 00400560,3
|
||||
I 00400563,4
|
||||
L 7ff000370,8
|
||||
I 00400567,3
|
||||
L 7ff000384,4
|
||||
I 0040056a,3
|
||||
I 0040056d,3
|
||||
L 7ff00038c,4
|
||||
I 00400570,3
|
||||
S 00600a68,4
|
||||
I 00400573,4
|
||||
M 7ff000388,4
|
||||
I 00400577,4
|
||||
L 7ff000388,4
|
||||
I 0040057b,2
|
||||
I 0040053c,3
|
||||
L 7ff000384,4
|
||||
I 0040053f,2
|
||||
I 00400541,4
|
||||
I 00400545,3
|
||||
I 00400548,4
|
||||
L 7ff000378,8
|
||||
I 0040054c,3
|
||||
L 7ff000388,4
|
||||
I 0040054f,2
|
||||
I 00400551,3
|
||||
L 00600a44,4
|
||||
I 00400554,3
|
||||
S 7ff00038c,4
|
||||
I 00400557,3
|
||||
L 7ff000388,4
|
||||
I 0040055a,2
|
||||
I 0040055c,4
|
||||
I 00400560,3
|
||||
I 00400563,4
|
||||
L 7ff000370,8
|
||||
I 00400567,3
|
||||
L 7ff000384,4
|
||||
I 0040056a,3
|
||||
I 0040056d,3
|
||||
L 7ff00038c,4
|
||||
I 00400570,3
|
||||
S 00600a78,4
|
||||
I 00400573,4
|
||||
M 7ff000388,4
|
||||
I 00400577,4
|
||||
L 7ff000388,4
|
||||
I 0040057b,2
|
||||
I 0040053c,3
|
||||
L 7ff000384,4
|
||||
I 0040053f,2
|
||||
I 00400541,4
|
||||
I 00400545,3
|
||||
I 00400548,4
|
||||
L 7ff000378,8
|
||||
I 0040054c,3
|
||||
L 7ff000388,4
|
||||
I 0040054f,2
|
||||
I 00400551,3
|
||||
L 00600a48,4
|
||||
I 00400554,3
|
||||
S 7ff00038c,4
|
||||
I 00400557,3
|
||||
L 7ff000388,4
|
||||
I 0040055a,2
|
||||
I 0040055c,4
|
||||
I 00400560,3
|
||||
I 00400563,4
|
||||
L 7ff000370,8
|
||||
I 00400567,3
|
||||
L 7ff000384,4
|
||||
I 0040056a,3
|
||||
I 0040056d,3
|
||||
L 7ff00038c,4
|
||||
I 00400570,3
|
||||
S 00600a88,4
|
||||
I 00400573,4
|
||||
M 7ff000388,4
|
||||
I 00400577,4
|
||||
L 7ff000388,4
|
||||
I 0040057b,2
|
||||
I 0040053c,3
|
||||
L 7ff000384,4
|
||||
I 0040053f,2
|
||||
I 00400541,4
|
||||
I 00400545,3
|
||||
I 00400548,4
|
||||
L 7ff000378,8
|
||||
I 0040054c,3
|
||||
L 7ff000388,4
|
||||
I 0040054f,2
|
||||
I 00400551,3
|
||||
L 00600a4c,4
|
||||
I 00400554,3
|
||||
S 7ff00038c,4
|
||||
I 00400557,3
|
||||
L 7ff000388,4
|
||||
I 0040055a,2
|
||||
I 0040055c,4
|
||||
I 00400560,3
|
||||
I 00400563,4
|
||||
L 7ff000370,8
|
||||
I 00400567,3
|
||||
L 7ff000384,4
|
||||
I 0040056a,3
|
||||
I 0040056d,3
|
||||
L 7ff00038c,4
|
||||
I 00400570,3
|
||||
S 00600a98,4
|
||||
I 00400573,4
|
||||
M 7ff000388,4
|
||||
I 00400577,4
|
||||
L 7ff000388,4
|
||||
I 0040057b,2
|
||||
I 0040057d,4
|
||||
M 7ff000384,4
|
||||
I 00400581,4
|
||||
L 7ff000384,4
|
||||
I 00400585,2
|
||||
I 00400533,7
|
||||
S 7ff000388,4
|
||||
I 0040053a,2
|
||||
I 00400577,4
|
||||
L 7ff000388,4
|
||||
I 0040057b,2
|
||||
I 0040053c,3
|
||||
L 7ff000384,4
|
||||
I 0040053f,2
|
||||
I 00400541,4
|
||||
I 00400545,3
|
||||
I 00400548,4
|
||||
L 7ff000378,8
|
||||
I 0040054c,3
|
||||
L 7ff000388,4
|
||||
I 0040054f,2
|
||||
I 00400551,3
|
||||
L 00600a50,4
|
||||
I 00400554,3
|
||||
S 7ff00038c,4
|
||||
I 00400557,3
|
||||
L 7ff000388,4
|
||||
I 0040055a,2
|
||||
I 0040055c,4
|
||||
I 00400560,3
|
||||
I 00400563,4
|
||||
L 7ff000370,8
|
||||
I 00400567,3
|
||||
L 7ff000384,4
|
||||
I 0040056a,3
|
||||
I 0040056d,3
|
||||
L 7ff00038c,4
|
||||
I 00400570,3
|
||||
S 00600a6c,4
|
||||
I 00400573,4
|
||||
M 7ff000388,4
|
||||
I 00400577,4
|
||||
L 7ff000388,4
|
||||
I 0040057b,2
|
||||
I 0040053c,3
|
||||
L 7ff000384,4
|
||||
I 0040053f,2
|
||||
I 00400541,4
|
||||
I 00400545,3
|
||||
I 00400548,4
|
||||
L 7ff000378,8
|
||||
I 0040054c,3
|
||||
L 7ff000388,4
|
||||
I 0040054f,2
|
||||
I 00400551,3
|
||||
L 00600a54,4
|
||||
I 00400554,3
|
||||
S 7ff00038c,4
|
||||
I 00400557,3
|
||||
L 7ff000388,4
|
||||
I 0040055a,2
|
||||
I 0040055c,4
|
||||
I 00400560,3
|
||||
I 00400563,4
|
||||
L 7ff000370,8
|
||||
I 00400567,3
|
||||
L 7ff000384,4
|
||||
I 0040056a,3
|
||||
I 0040056d,3
|
||||
L 7ff00038c,4
|
||||
I 00400570,3
|
||||
S 00600a7c,4
|
||||
I 00400573,4
|
||||
M 7ff000388,4
|
||||
I 00400577,4
|
||||
L 7ff000388,4
|
||||
I 0040057b,2
|
||||
I 0040053c,3
|
||||
L 7ff000384,4
|
||||
I 0040053f,2
|
||||
I 00400541,4
|
||||
I 00400545,3
|
||||
I 00400548,4
|
||||
L 7ff000378,8
|
||||
I 0040054c,3
|
||||
L 7ff000388,4
|
||||
I 0040054f,2
|
||||
I 00400551,3
|
||||
L 00600a58,4
|
||||
I 00400554,3
|
||||
S 7ff00038c,4
|
||||
I 00400557,3
|
||||
L 7ff000388,4
|
||||
I 0040055a,2
|
||||
I 0040055c,4
|
||||
I 00400560,3
|
||||
I 00400563,4
|
||||
L 7ff000370,8
|
||||
I 00400567,3
|
||||
L 7ff000384,4
|
||||
I 0040056a,3
|
||||
I 0040056d,3
|
||||
L 7ff00038c,4
|
||||
I 00400570,3
|
||||
S 00600a8c,4
|
||||
I 00400573,4
|
||||
M 7ff000388,4
|
||||
I 00400577,4
|
||||
L 7ff000388,4
|
||||
I 0040057b,2
|
||||
I 0040053c,3
|
||||
L 7ff000384,4
|
||||
I 0040053f,2
|
||||
I 00400541,4
|
||||
I 00400545,3
|
||||
I 00400548,4
|
||||
L 7ff000378,8
|
||||
I 0040054c,3
|
||||
L 7ff000388,4
|
||||
I 0040054f,2
|
||||
I 00400551,3
|
||||
L 00600a5c,4
|
||||
I 00400554,3
|
||||
S 7ff00038c,4
|
||||
I 00400557,3
|
||||
L 7ff000388,4
|
||||
I 0040055a,2
|
||||
I 0040055c,4
|
||||
I 00400560,3
|
||||
I 00400563,4
|
||||
L 7ff000370,8
|
||||
I 00400567,3
|
||||
L 7ff000384,4
|
||||
I 0040056a,3
|
||||
I 0040056d,3
|
||||
L 7ff00038c,4
|
||||
I 00400570,3
|
||||
S 00600a9c,4
|
||||
I 00400573,4
|
||||
M 7ff000388,4
|
||||
I 00400577,4
|
||||
L 7ff000388,4
|
||||
I 0040057b,2
|
||||
I 0040057d,4
|
||||
M 7ff000384,4
|
||||
I 00400581,4
|
||||
L 7ff000384,4
|
||||
I 00400585,2
|
||||
I 00400587,1
|
||||
L 7ff000390,8
|
||||
I 00400588,1
|
||||
L 7ff000398,8
|
||||
I 004005c5,7
|
||||
L 00600aa0,1
|
|
@ -0,0 +1,7 @@
|
|||
L 10,1
|
||||
M 20,1
|
||||
L 22,1
|
||||
S 18,1
|
||||
L 110,1
|
||||
L 210,1
|
||||
M 12,1
|
|
@ -0,0 +1,16 @@
|
|||
L 0,1
|
||||
L 1,1
|
||||
L 2,1
|
||||
L 3,1
|
||||
S 4,1
|
||||
L 5,1
|
||||
S 6,1
|
||||
L 7,1
|
||||
S 8,1
|
||||
L 9,1
|
||||
S a,1
|
||||
L b,1
|
||||
S c,1
|
||||
L d,1
|
||||
S e,1
|
||||
M f,1
|
|
@ -0,0 +1,84 @@
|
|||
/*
|
||||
* trans.c - Matrix transpose B = A^T
|
||||
*
|
||||
* Each transpose function must have a prototype of the form:
|
||||
* void trans(int M, int N, int A[N][M], int B[M][N]);
|
||||
*
|
||||
* A transpose function is evaluated by counting the number of misses
|
||||
* on a 1KB direct mapped cache with a block size of 32 bytes.
|
||||
*/
|
||||
#include <stdio.h>
|
||||
#include "cachelab.h"
|
||||
|
||||
int is_transpose(int M, int N, int A[N][M], int B[M][N]);
|
||||
|
||||
/*
|
||||
* transpose_submit - This is the solution transpose function that you
|
||||
* will be graded on for Part B of the assignment. Do not change
|
||||
* the description string "Transpose submission", as the driver
|
||||
* searches for that string to identify the transpose function to
|
||||
* be graded.
|
||||
*/
|
||||
char transpose_submit_desc[] = "Transpose submission";
|
||||
void transpose_submit(int M, int N, int A[N][M], int B[M][N])
|
||||
{
|
||||
}
|
||||
|
||||
/*
|
||||
* You can define additional transpose functions below. We've defined
|
||||
* a simple one below to help you get started.
|
||||
*/
|
||||
|
||||
/*
|
||||
* trans - A simple baseline transpose function, not optimized for the cache.
|
||||
*/
|
||||
char trans_desc[] = "Simple row-wise scan transpose";
|
||||
void trans(int M, int N, int A[N][M], int B[M][N])
|
||||
{
|
||||
int i, j, tmp;
|
||||
|
||||
for (i = 0; i < N; i++) {
|
||||
for (j = 0; j < M; j++) {
|
||||
tmp = A[i][j];
|
||||
B[j][i] = tmp;
|
||||
}
|
||||
}
|
||||
|
||||
}
|
||||
|
||||
/*
|
||||
* registerFunctions - This function registers your transpose
|
||||
* functions with the driver. At runtime, the driver will
|
||||
* evaluate each of the registered functions and summarize their
|
||||
* performance. This is a handy way to experiment with different
|
||||
* transpose strategies.
|
||||
*/
|
||||
void registerFunctions()
|
||||
{
|
||||
/* Register your solution function */
|
||||
registerTransFunction(transpose_submit, transpose_submit_desc);
|
||||
|
||||
/* Register any additional transpose functions */
|
||||
registerTransFunction(trans, trans_desc);
|
||||
|
||||
}
|
||||
|
||||
/*
|
||||
* is_transpose - This helper function checks if B is the transpose of
|
||||
* A. You can check the correctness of your transpose by calling
|
||||
* it before returning from the transpose function.
|
||||
*/
|
||||
int is_transpose(int M, int N, int A[N][M], int B[M][N])
|
||||
{
|
||||
int i, j;
|
||||
|
||||
for (i = 0; i < N; i++) {
|
||||
for (j = 0; j < M; ++j) {
|
||||
if (A[i][j] != B[j][i]) {
|
||||
return 0;
|
||||
}
|
||||
}
|
||||
}
|
||||
return 1;
|
||||
}
|
||||
|
Loading…
Reference in New Issue