+added wc-timer
This commit is contained in:
Evghenii
@@ -44,64 +44,79 @@
|
||||
#include "volume_ispc.h"
|
||||
using namespace ispc;
|
||||
|
||||
#include <sys/time.h>
|
||||
|
||||
|
||||
static inline double rtc(void)
|
||||
{
|
||||
struct timeval Tvalue;
|
||||
double etime;
|
||||
struct timezone dummy;
|
||||
|
||||
gettimeofday(&Tvalue,&dummy);
|
||||
etime = (double) Tvalue.tv_sec +
|
||||
1.e-6*((double) Tvalue.tv_usec);
|
||||
return etime;
|
||||
}
|
||||
|
||||
extern void volume_serial(float density[], int nVoxels[3],
|
||||
const float raster2camera[4][4],
|
||||
const float camera2world[4][4],
|
||||
int width, int height, float image[]);
|
||||
const float raster2camera[4][4],
|
||||
const float camera2world[4][4],
|
||||
int width, int height, float image[]);
|
||||
|
||||
/* Write a PPM image file with the image */
|
||||
static void
|
||||
writePPM(float *buf, int width, int height, const char *fn) {
|
||||
FILE *fp = fopen(fn, "wb");
|
||||
fprintf(fp, "P6\n");
|
||||
fprintf(fp, "%d %d\n", width, height);
|
||||
fprintf(fp, "255\n");
|
||||
for (int i = 0; i < width*height; ++i) {
|
||||
float v = buf[i] * 255.f;
|
||||
if (v < 0.f) v = 0.f;
|
||||
else if (v > 255.f) v = 255.f;
|
||||
unsigned char c = (unsigned char)v;
|
||||
for (int j = 0; j < 3; ++j)
|
||||
fputc(c, fp);
|
||||
}
|
||||
fclose(fp);
|
||||
printf("Wrote image file %s\n", fn);
|
||||
FILE *fp = fopen(fn, "wb");
|
||||
fprintf(fp, "P6\n");
|
||||
fprintf(fp, "%d %d\n", width, height);
|
||||
fprintf(fp, "255\n");
|
||||
for (int i = 0; i < width*height; ++i) {
|
||||
float v = buf[i] * 255.f;
|
||||
if (v < 0.f) v = 0.f;
|
||||
else if (v > 255.f) v = 255.f;
|
||||
unsigned char c = (unsigned char)v;
|
||||
for (int j = 0; j < 3; ++j)
|
||||
fputc(c, fp);
|
||||
}
|
||||
fclose(fp);
|
||||
printf("Wrote image file %s\n", fn);
|
||||
}
|
||||
|
||||
|
||||
/* Load image and viewing parameters from a camera data file.
|
||||
FIXME: we should add support to be able to specify viewing parameters
|
||||
in the program here directly. */
|
||||
FIXME: we should add support to be able to specify viewing parameters
|
||||
in the program here directly. */
|
||||
static void
|
||||
loadCamera(const char *fn, int *width, int *height, float raster2camera[4][4],
|
||||
float camera2world[4][4]) {
|
||||
FILE *f = fopen(fn, "r");
|
||||
if (!f) {
|
||||
perror(fn);
|
||||
exit(1);
|
||||
}
|
||||
if (fscanf(f, "%d %d", width, height) != 2) {
|
||||
float camera2world[4][4]) {
|
||||
FILE *f = fopen(fn, "r");
|
||||
if (!f) {
|
||||
perror(fn);
|
||||
exit(1);
|
||||
}
|
||||
if (fscanf(f, "%d %d", width, height) != 2) {
|
||||
fprintf(stderr, "Unexpected end of file in camera file\n");
|
||||
exit(1);
|
||||
}
|
||||
|
||||
for (int i = 0; i < 4; ++i) {
|
||||
for (int j = 0; j < 4; ++j) {
|
||||
if (fscanf(f, "%f", &raster2camera[i][j]) != 1) {
|
||||
fprintf(stderr, "Unexpected end of file in camera file\n");
|
||||
exit(1);
|
||||
}
|
||||
}
|
||||
|
||||
for (int i = 0; i < 4; ++i) {
|
||||
for (int j = 0; j < 4; ++j) {
|
||||
if (fscanf(f, "%f", &raster2camera[i][j]) != 1) {
|
||||
fprintf(stderr, "Unexpected end of file in camera file\n");
|
||||
exit(1);
|
||||
}
|
||||
}
|
||||
}
|
||||
for (int i = 0; i < 4; ++i) {
|
||||
for (int j = 0; j < 4; ++j) {
|
||||
if (fscanf(f, "%f", &camera2world[i][j]) != 1) {
|
||||
fprintf(stderr, "Unexpected end of file in camera file\n");
|
||||
exit(1);
|
||||
}
|
||||
}
|
||||
for (int i = 0; i < 4; ++i) {
|
||||
for (int j = 0; j < 4; ++j) {
|
||||
if (fscanf(f, "%f", &camera2world[i][j]) != 1) {
|
||||
fprintf(stderr, "Unexpected end of file in camera file\n");
|
||||
exit(1);
|
||||
}
|
||||
}
|
||||
}
|
||||
fclose(f);
|
||||
}
|
||||
fclose(f);
|
||||
}
|
||||
|
||||
|
||||
@@ -110,105 +125,108 @@ loadCamera(const char *fn, int *width, int *height, float raster2camera[4][4],
|
||||
floating-point values (also as strings) to give the densities. */
|
||||
static float *
|
||||
loadVolume(const char *fn, int n[3]) {
|
||||
FILE *f = fopen(fn, "r");
|
||||
if (!f) {
|
||||
perror(fn);
|
||||
exit(1);
|
||||
}
|
||||
FILE *f = fopen(fn, "r");
|
||||
if (!f) {
|
||||
perror(fn);
|
||||
exit(1);
|
||||
}
|
||||
|
||||
if (fscanf(f, "%d %d %d", &n[0], &n[1], &n[2]) != 3) {
|
||||
fprintf(stderr, "Couldn't find resolution at start of density file\n");
|
||||
exit(1);
|
||||
}
|
||||
if (fscanf(f, "%d %d %d", &n[0], &n[1], &n[2]) != 3) {
|
||||
fprintf(stderr, "Couldn't find resolution at start of density file\n");
|
||||
exit(1);
|
||||
}
|
||||
|
||||
int count = n[0] * n[1] * n[2];
|
||||
float *v = new float[count];
|
||||
for (int i = 0; i < count; ++i) {
|
||||
if (fscanf(f, "%f", &v[i]) != 1) {
|
||||
fprintf(stderr, "Unexpected end of file at %d'th density value\n", i);
|
||||
exit(1);
|
||||
}
|
||||
int count = n[0] * n[1] * n[2];
|
||||
float *v = new float[count];
|
||||
for (int i = 0; i < count; ++i) {
|
||||
if (fscanf(f, "%f", &v[i]) != 1) {
|
||||
fprintf(stderr, "Unexpected end of file at %d'th density value\n", i);
|
||||
exit(1);
|
||||
}
|
||||
}
|
||||
|
||||
return v;
|
||||
return v;
|
||||
}
|
||||
|
||||
|
||||
int main(int argc, char *argv[]) {
|
||||
if (argc != 3) {
|
||||
fprintf(stderr, "usage: volume <camera.dat> <volume_density.vol>\n");
|
||||
return 1;
|
||||
}
|
||||
if (argc != 3) {
|
||||
fprintf(stderr, "usage: volume <camera.dat> <volume_density.vol>\n");
|
||||
return 1;
|
||||
}
|
||||
|
||||
//
|
||||
// Load viewing data and the volume density data
|
||||
//
|
||||
int width, height;
|
||||
float raster2camera[4][4], camera2world[4][4];
|
||||
loadCamera(argv[1], &width, &height, raster2camera, camera2world);
|
||||
float *image = new float[width*height];
|
||||
//
|
||||
// Load viewing data and the volume density data
|
||||
//
|
||||
int width, height;
|
||||
float raster2camera[4][4], camera2world[4][4];
|
||||
loadCamera(argv[1], &width, &height, raster2camera, camera2world);
|
||||
float *image = new float[width*height];
|
||||
|
||||
int n[3];
|
||||
float *density = loadVolume(argv[2], n);
|
||||
int n[3];
|
||||
float *density = loadVolume(argv[2], n);
|
||||
|
||||
//
|
||||
// Compute the image using the ispc implementation; report the minimum
|
||||
// time of three runs.
|
||||
//
|
||||
double minISPC = 1e30;
|
||||
for (int i = 0; i < 3; ++i) {
|
||||
reset_and_start_timer();
|
||||
volume_ispc(density, n, raster2camera, camera2world,
|
||||
width, height, image);
|
||||
double dt = get_elapsed_mcycles();
|
||||
minISPC = std::min(minISPC, dt);
|
||||
}
|
||||
//
|
||||
// Compute the image using the ispc implementation; report the minimum
|
||||
// time of three runs.
|
||||
//
|
||||
double minISPC = 1e30;
|
||||
#if 0
|
||||
for (int i = 0; i < 3; ++i) {
|
||||
reset_and_start_timer();
|
||||
volume_ispc(density, n, raster2camera, camera2world,
|
||||
width, height, image);
|
||||
double dt = get_elapsed_mcycles();
|
||||
minISPC = std::min(minISPC, dt);
|
||||
}
|
||||
#endif
|
||||
|
||||
printf("[volume ispc 1 core]:\t\t[%.3f] million cycles\n", minISPC);
|
||||
writePPM(image, width, height, "volume-ispc-1core.ppm");
|
||||
printf("[volume ispc 1 core]:\t\t[%.3f] million cycles\n", minISPC);
|
||||
writePPM(image, width, height, "volume-ispc-1core.ppm");
|
||||
|
||||
// Clear out the buffer
|
||||
for (int i = 0; i < width * height; ++i)
|
||||
image[i] = 0.;
|
||||
// Clear out the buffer
|
||||
for (int i = 0; i < width * height; ++i)
|
||||
image[i] = 0.;
|
||||
|
||||
//
|
||||
// Compute the image using the ispc implementation that also uses
|
||||
// tasks; report the minimum time of three runs.
|
||||
//
|
||||
double minISPCtasks = 1e30;
|
||||
for (int i = 0; i < 3; ++i) {
|
||||
reset_and_start_timer();
|
||||
volume_ispc_tasks(density, n, raster2camera, camera2world,
|
||||
width, height, image);
|
||||
double dt = get_elapsed_mcycles();
|
||||
minISPCtasks = std::min(minISPCtasks, dt);
|
||||
}
|
||||
//
|
||||
// Compute the image using the ispc implementation that also uses
|
||||
// tasks; report the minimum time of three runs.
|
||||
//
|
||||
double minISPCtasks = 1e30;
|
||||
for (int i = 0; i < 3; ++i) {
|
||||
reset_and_start_timer();
|
||||
const double t0 = rtc();
|
||||
volume_ispc_tasks(density, n, raster2camera, camera2world,
|
||||
width, height, image);
|
||||
double dt = rtc() - t0; //get_elapsed_mcycles();
|
||||
minISPCtasks = std::min(minISPCtasks, dt);
|
||||
}
|
||||
|
||||
printf("[volume ispc + tasks]:\t\t[%.3f] million cycles\n", minISPCtasks);
|
||||
writePPM(image, width, height, "volume-ispc-tasks.ppm");
|
||||
printf("[volume ispc + tasks]:\t\t[%.3f] million cycles\n", minISPCtasks);
|
||||
writePPM(image, width, height, "volume-ispc-tasks.ppm");
|
||||
|
||||
// Clear out the buffer
|
||||
for (int i = 0; i < width * height; ++i)
|
||||
image[i] = 0.;
|
||||
// Clear out the buffer
|
||||
for (int i = 0; i < width * height; ++i)
|
||||
image[i] = 0.;
|
||||
|
||||
//
|
||||
// And run the serial implementation 3 times, again reporting the
|
||||
// minimum time.
|
||||
//
|
||||
double minSerial = 1e30;
|
||||
for (int i = 0; i < 3; ++i) {
|
||||
reset_and_start_timer();
|
||||
volume_serial(density, n, raster2camera, camera2world,
|
||||
width, height, image);
|
||||
double dt = get_elapsed_mcycles();
|
||||
minSerial = std::min(minSerial, dt);
|
||||
}
|
||||
//
|
||||
// And run the serial implementation 3 times, again reporting the
|
||||
// minimum time.
|
||||
//
|
||||
double minSerial = 1e30;
|
||||
for (int i = 0; i < 3; ++i) {
|
||||
reset_and_start_timer();
|
||||
volume_serial(density, n, raster2camera, camera2world,
|
||||
width, height, image);
|
||||
double dt = get_elapsed_mcycles();
|
||||
minSerial = std::min(minSerial, dt);
|
||||
}
|
||||
|
||||
printf("[volume serial]:\t\t[%.3f] million cycles\n", minSerial);
|
||||
writePPM(image, width, height, "volume-serial.ppm");
|
||||
printf("[volume serial]:\t\t[%.3f] million cycles\n", minSerial);
|
||||
writePPM(image, width, height, "volume-serial.ppm");
|
||||
|
||||
printf("\t\t\t\t(%.2fx speedup from ISPC, %.2fx speedup from ISPC + tasks)\n",
|
||||
minSerial/minISPC, minSerial / minISPCtasks);
|
||||
printf("\t\t\t\t(%.2fx speedup from ISPC, %.2fx speedup from ISPC + tasks)\n",
|
||||
minSerial/minISPC, minSerial / minISPCtasks);
|
||||
|
||||
return 0;
|
||||
return 0;
|
||||
}
|
||||
|
||||
Reference in New Issue
Block a user