plasma-umass/coz
1
1
Fork 0
mirror of https://github.com/plasma-umass/coz.git synced 2024-09-11 04:55:25 +03:00
Code Issues Packages Projects Releases Wiki Activity

Simple loopy test, plus gitignore cleanup

Browse Source
This commit is contained in:
Charlie Curtsinger 2014-07-08 17:13:36 -04:00
parent 3ea479fd4e
commit 095bfae9b9
12 changed files with 170 additions and 58 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

74
.gitignore vendored
View File

@ -13,50 +13,54 @@ lib/support/release
tools/inspect/debug
tools/inspect/release
tests/*/profile.log
tests/*/results.csv
tests/*/Rplots.pdf
tests/loopy/debug
tests/loopy/release
tests/loopy/profile.log
tests/histogram/debug
tests/histogram/release
tests/histogram/histogram_datafiles
benchmarks/*/profile.log
benchmarks/*/results.csv
benchmarks/*/Rplots.pdf
tests/kmeans/debug
tests/kmeans/release
benchmarks/histogram/debug
benchmarks/histogram/release
benchmarks/histogram/histogram_datafiles
tests/linear_regression/debug
tests/libear_regression/release
tests/linear_regression/linear_regression_datafiles
benchmarks/kmeans/debug
benchmarks/kmeans/release
tests/matrix_multiply/debug
tests/matrix_multiply/release
tests/matrix_multiply/matrix_file_A.txt
tests/matrix_multiply/matrix_file_B.txt
tests/matrix_multiply/matrix_file_out_pthreads.txt
benchmarks/linear_regression/debug
benchmarks/libear_regression/release
benchmarks/linear_regression/linear_regression_datafiles
tests/pbzip2/debug
tests/pbzip2/release
tests/pbzip2/data/input.txt
tests/pbzip2/data/output.txt
tests/pbzip2/bzip2-1.0.6
benchmarks/matrix_multiply/debug
benchmarks/matrix_multiply/release
benchmarks/matrix_multiply/matrix_file_A.txt
benchmarks/matrix_multiply/matrix_file_B.txt
benchmarks/matrix_multiply/matrix_file_out_pthreads.txt
tests/pca/debug
tests/pca/release
benchmarks/pbzip2/debug
benchmarks/pbzip2/release
benchmarks/pbzip2/data/input.txt
benchmarks/pbzip2/data/output.txt
benchmarks/pbzip2/bzip2-1.0.6
tests/producer_consumer/debug
tests/producer_consumer/release
benchmarks/pca/debug
benchmarks/pca/release
tests/string_match/debug
tests/string_match/release
tests/string_match/string_match_datafiles
benchmarks/producer_consumer/debug
benchmarks/producer_consumer/release
tests/unbalanced/debug
tests/unbalanced/release
benchmarks/string_match/debug
benchmarks/string_match/release
benchmarks/string_match/string_match_datafiles
tests/word_count/debug
tests/word_count/release
tests/word_count/word_count_datafiles
benchmarks/unbalanced/debug
benchmarks/unbalanced/release
tests/work_queue/debug
tests/work_queue/release
benchmarks/word_count/debug
benchmarks/word_count/release
benchmarks/word_count/word_count_datafiles
benchmarks/work_queue/debug
benchmarks/work_queue/release

2
Makefile
View File

@ -1,5 +1,5 @@
ROOT = .
DIRS = lib tools
DIRS = lib tools benchmarks tests
include $(ROOT)/common.mk

2
benchmarks/Makefile
View File

@ -4,5 +4,3 @@ DIRS = histogram kmeans linear_regression matrix_multiply \
RECURSIVE_TARGETS = debug release bench
include $(ROOT)/common.mk
DIRS = histogram linear_regression matrix_multiply \
pbzip2 pca producer_consumer string_match word_count

13
benchmarks/matrix_multiply/matrix_multiply-pthread.c
View File

@ -80,8 +80,7 @@ void matrixmult_splitter(void *data_in)
assert(data->matrix_B);
assert(data->output);
//CHECK_ERROR((num_procs = sysconf(_SC_NPROCESSORS_ONLN)) <= 0);
num_procs = 4;
CHECK_ERROR((num_procs = sysconf(_SC_NPROCESSORS_ONLN)) <= 0);
dprintf("THe number of processors is %d\n", num_procs);
tid = (pthread_t *)MALLOC(num_procs * sizeof(pthread_t));
@ -148,9 +147,7 @@ void *matrixmult_map(void *args_in)
value = 0;
for(j=0;j<data->matrix_len ; j++)
{
for(int k = 0; k <= 8 - SKIP_COUNT; k++) {
value += ( a_ptr[j] * (*b_ptr));
}
value += ( a_ptr[j] * (*b_ptr));
b_ptr+= data->matrix_len;
}
x_loc = (data->row_num + row_count);
@ -286,14 +283,14 @@ int main(int argc, char *argv[]) {
fprintf(stderr, "runtime = %f\n", (usec / 1000000));
/*for(i=0;i<matrix_len*matrix_len;i++)
for(i=0;i<matrix_len*matrix_len;i++)
{
if(i%matrix_len == 0)
dprintf("\n");
dprintf("%d ",mm_data.output[i]);
//write(fd_out,&(mm_data.output[i]),sizeof(int));
write(fd_out,&(mm_data.output[i]),sizeof(int));
}
dprintf("\n");
@ -307,7 +304,7 @@ int main(int argc, char *argv[]) {
CHECK_ERROR(munmap(fdata_B, file_size + 1) < 0);
CHECK_ERROR(close(fd_B) < 0);
CHECK_ERROR(close(fd_out) < 0);*/
CHECK_ERROR(close(fd_out) < 0);
return 0;
}

14
include/causal.h
View File

@ -18,9 +18,11 @@ static void __init_counter(int kind, size_t* ctr, const char* name) {
#define CAUSAL_INCREMENT_COUNTER(kind, name) \
if(1) { \
static unsigned char __causal_counter_initialized = 0; \
static size_t __causal_counter = 0; \
if(__atomic_exchange_n(&__causal_counter_initialized, 1, __ATOMIC_SEQ_CST) == 0) { \
static unsigned int __causal_counter_initialized; \
static size_t __causal_counter; \
if(__causal_counter_initialized != 0xDEADBEEF && \
__atomic_exchange_n(&__causal_counter_initialized, 0xDEADBEEF, __ATOMIC_SEQ_CST) != 0xDEADBEEF) { \
__causal_counter = 0; \
__init_counter(kind, &__causal_counter, name); \
} \
__atomic_fetch_add(&__causal_counter, 1, __ATOMIC_SEQ_CST); \
@ -33,9 +35,9 @@ static void __init_counter(int kind, size_t* ctr, const char* name) {
#define STR2(x) #x
#define STR(x) STR2(x)
#define CAUSAL_PROGRESS CAUSAL_INCREMENT_COUNTER(PROGRESS_COUNTER, __FILE__ " line " STR(__LINE__))
#define CAUSAL_BEGIN CAUSAL_INCREMENT_COUNTER(BEGIN_COUNTER, __FILE__ " line " STR(__LINE__))
#define CAUSAL_END CAUSAL_INCREMENT_COUNTER(END_COUNTER, __FILE__ " line " STR(__LINE__))
#define CAUSAL_PROGRESS CAUSAL_INCREMENT_COUNTER(PROGRESS_COUNTER, __FILE__ ":" STR(__LINE__))
#define CAUSAL_BEGIN CAUSAL_INCREMENT_COUNTER(BEGIN_COUNTER, __FILE__ ":" STR(__LINE__))
#define CAUSAL_END CAUSAL_INCREMENT_COUNTER(END_COUNTER, __FILE__ ":" STR(__LINE__))
#if defined(__cplusplus)
}

4
lib/runtime/Makefile
View File

@ -9,4 +9,8 @@ CFLAGS =
include $(ROOT)/common.mk
debug:: $(ROOT)/debug/lib/libcausal_support.a
release:: $(ROOT)/release/lib/libcausal_support.a
CXXFLAGS += --std=c++11 -fPIC

2
lib/runtime/include/profiler.h
View File

@ -10,7 +10,7 @@
enum {
PauseSignal = 42,
SamplePeriod = 1000000, // 1ms
SamplePeriod = 2100000, // 1ms
SampleWakeupCount = 1,
MinRoundSamples = 1000
};

10
lib/runtime/profiler.cpp
View File

@ -187,8 +187,8 @@ namespace profiler {
/// The perf event configuration
struct perf_event_attr pe = {
.type = PERF_TYPE_SOFTWARE,
.config = PERF_COUNT_SW_TASK_CLOCK,
.type = PERF_TYPE_HARDWARE,
.config = PERF_COUNT_HW_CPU_CYCLES,
.sample_type = PERF_SAMPLE_IP | PERF_SAMPLE_TID | PERF_SAMPLE_TIME | PERF_SAMPLE_CALLCHAIN,
.sample_period = SamplePeriod,
.wakeup_events = SampleWakeupCount
@ -207,7 +207,11 @@ namespace profiler {
// If the baseline round has run long enough and we're in a known line,
// use this line for the next speedup round
if(roundSamples >= MinRoundSamples) {
selectedLine = l;
if(fixed_line) {
selectedLine = fixed_line;
} else {
selectedLine = l;
}
}
} else if(mode == Speedup) {
// Check if the sample is in the selected line

15
lib/support/libsuport.cpp
View File

@ -254,8 +254,19 @@ namespace causal_support {
}
shared_ptr<line> memory_map::find_line(const string& name) {
WARNING << "Line name searches are not yet implemented!";
return shared_ptr<line>();
string::size_type colon_pos = name.find_first_of(':');
if(colon_pos == string::npos) {
WARNING << "Could not identify file name in input " << name;
return shared_ptr<line>();
}
string filename = name.substr(0, colon_pos);
string line_no_str = name.substr(colon_pos + 1);
size_t line_no;
stringstream(line_no_str) >> line_no;
return memory_map::get_file(filename)->get_line(line_no);
}
shared_ptr<line> memory_map::find_line(uintptr_t addr) {

5
tests/Makefile Normal file
View File

@ -0,0 +1,5 @@
ROOT = ..
DIRS = loopy
RECURSIVE_TARGETS = debug release test
include $(ROOT)/common.mk

8
tests/loopy/Makefile Normal file
View File

@ -0,0 +1,8 @@
ROOT = ../..
TARGETS = loopy
LIBS = pthread dl
include $(ROOT)/common.mk
test:: debug/bin/loopy
debug/bin/loopy

79
tests/loopy/loopy.cpp Normal file
View File

@ -0,0 +1,79 @@
#include <pthread.h>
#include <stdint.h>
#include <stdio.h>
#include <stdlib.h>
#include <time.h>
#include <causal.h>
#define WIDTH 40000
#define HEIGHT 40000
#define MAX_SKIP_COUNT 8
struct thread_arg {
size_t index;
size_t thread_count;
size_t skip_count;
int* data;
thread_arg(size_t i, size_t t, size_t s, int* d) :
index(i), thread_count(t), skip_count(s), data(d) {}
};
void* thread_fn(void* arg) {
thread_arg* a = reinterpret_cast<thread_arg*>(arg);
for(size_t i = a->index; i < HEIGHT; i += a->thread_count) {
for(size_t k = 0; k < MAX_SKIP_COUNT; k++) {
if(k >= a->skip_count) {
for(size_t j = 0; j < WIDTH; j++) {
a->data[i * HEIGHT + j] *= 2;
}
}
}
CAUSAL_PROGRESS;
}
delete a;
return NULL;
}
size_t get_nanos() {
struct timespec ts;
if(clock_gettime(CLOCK_REALTIME, &ts)) {
fprintf(stderr, "Failed to get start time\n");
exit(2);
}
return ts.tv_sec * 1000000000 + ts.tv_nsec;
}
int main(int argc, char** argv) {
if(argc != 3) {
fprintf(stderr, "Must specify thread count and skip count on command line\n");
exit(1);
}
size_t num_threads = atoi(argv[1]);
size_t skip_count = atoi(argv[2]);
int* data = new int[WIDTH * HEIGHT];
size_t start_nanos = get_nanos();
pthread_t threads[num_threads];
for(size_t i = 0; i < num_threads; i++) {
thread_arg* arg = new thread_arg(i, num_threads, skip_count, data);
pthread_create(&threads[i], NULL, thread_fn, arg);
}
for(size_t i = 0; i < num_threads; i++) {
while(pthread_tryjoin_np(threads[i], NULL)) {}
}
size_t nanos = get_nanos() - start_nanos;
delete[] data;
fprintf(stderr, "runtime: %lu\n", nanos);
return 0;
}