OSStatus host_platform_spi_transfer( bus_transfer_direction_t dir, uint8_t* buffer, uint16_t buffer_length )
{
OSStatus result;
int i;
for(i=0; i<buffer_length; i++) {
buffer_temp_32[i] = SPI0_TXCMD | (uint32_t)buffer[i];;
}
dmaSPITX.dmaChStcd = (edma_software_tcd_t *)mem_align(2 * sizeof(edma_software_tcd_t) * dmaSPITX.period, 32);
dmaSPITX.srcAddr = (uint32_t)buffer_temp_32;
dmaSPITX.length = buffer_length * 4;
dmaSPIRX.dmaChStcd = (edma_software_tcd_t *)mem_align(2 * sizeof(edma_software_tcd_t) * dmaSPIRX.period, 32);
dmaSPIRX.destAddr = (uint32_t)buffer;
dmaSPIRX.length = buffer_length;
MCU_CLOCKS_NEEDED();
SPI0_CS_ENABLE;
setup_edma_loop(&dmaSPITX);
setup_edma_loop(&dmaSPIRX);
EDMA_DRV_StartChannel(dmaSPIRX.dmaCh);
EDMA_DRV_StartChannel(dmaSPITX.dmaCh);
result = mico_rtos_get_semaphore( &spi_transfer_finished_semaphore, 100 );
disable_edma_loop(&dmaSPIRX);
disable_edma_loop(&dmaSPITX);
SPI0_CS_DISABLE;
MCU_CLOCKS_NOT_NEEDED();
free_align(dmaSPITX.dmaChStcd);
free_align(dmaSPIRX.dmaChStcd);
return result;
}
void free_aligned_matrix(int **a, int h)
{
int i;
for (i = 0; i < h; i++)
free_align(a[i]);
free_align(a);
}
void process_image_simple(struct image* img){
unsigned char *input, *output, *temp;
unsigned int addr1, addr2, i, j, k, r, g, b;
int block_nr = img->block_nr;
vector unsigned char *v1, *v2, *v3, *v4, *v5 ;
input = malloc_align(NUM_CHANNELS * SCALE_FACTOR * img->width, 4);
output = malloc_align(NUM_CHANNELS * img->width / SCALE_FACTOR, 4);
temp = malloc_align(NUM_CHANNELS * img->width, 4);
v1 = (vector unsigned char *) &input[0];
v2 = (vector unsigned char *) &input[1 * img->width * NUM_CHANNELS];
v3 = (vector unsigned char *) &input[2 * img->width * NUM_CHANNELS];
v4 = (vector unsigned char *) &input[3 * img->width * NUM_CHANNELS];
v5 = (vector unsigned char *) temp;
addr2 = (unsigned int)img->dst; //start of image
addr2 += (block_nr / NUM_IMAGES_HEIGHT) * img->width * NUM_CHANNELS *
img->height / NUM_IMAGES_HEIGHT; //start line of spu block
addr2 += (block_nr % NUM_IMAGES_WIDTH) * NUM_CHANNELS *
img->width / NUM_IMAGES_WIDTH;
for (i=0; i<img->height / SCALE_FACTOR; i++){
//get 4 lines
addr1 = ((unsigned int)img->src) + i * img->width * NUM_CHANNELS * SCALE_FACTOR;
mfc_get(input, addr1, SCALE_FACTOR * img->width * NUM_CHANNELS, MY_TAG, 0, 0);
mfc_write_tag_mask(1 << MY_TAG);
mfc_read_tag_status_all();
//compute the scaled line
for (j = 0; j < img->width * NUM_CHANNELS / 16; j++){
v5[j] = spu_avg(spu_avg(v1[j], v2[j]), spu_avg(v3[j], v4[j]));
}
for (j=0; j < img->width; j+=SCALE_FACTOR){
r = g = b = 0;
for (k = j; k < j + SCALE_FACTOR; k++) {
r += temp[k * NUM_CHANNELS + 0];
g += temp[k * NUM_CHANNELS + 1];
b += temp[k * NUM_CHANNELS + 2];
}
r /= SCALE_FACTOR;
b /= SCALE_FACTOR;
g /= SCALE_FACTOR;
output[j / SCALE_FACTOR * NUM_CHANNELS + 0] = (unsigned char) r;
output[j / SCALE_FACTOR * NUM_CHANNELS + 1] = (unsigned char) g;
output[j / SCALE_FACTOR * NUM_CHANNELS + 2] = (unsigned char) b;
}
//put the scaled line back
mfc_put(output, addr2, img->width / SCALE_FACTOR * NUM_CHANNELS, MY_TAG, 0, 0);
addr2 += img->width * NUM_CHANNELS; //line inside spu block
mfc_write_tag_mask(1 << MY_TAG);
mfc_read_tag_status_all();
}
free_align(temp);
free_align(input);
free_align(output);
}
void free_patch_id_vector(int **spu_patch_id_vector)
{
int i;
for (i = 0; i < SPU_THREADS; i++)
free_align(spu_patch_id_vector[i]);
free_align(spu_patch_id_vector);
}
static void free_bufs(OLTraceCtx *ctx)
{
if (ctx->tracebuf)
free(ctx->tracebuf);
if (ctx->sb)
free(ctx->sb);
if (ctx->pb)
free(ctx->pb);
if (ctx->k)
free_align(ctx->k);
if (ctx->bibuf)
free_align(ctx->bibuf);
if (ctx->btbuf)
free_align(ctx->btbuf);
if (ctx->sibuf)
free_align(ctx->sibuf);
if (ctx->stbuf)
free_align(ctx->stbuf);
if (ctx->sxbuf)
free_align(ctx->sxbuf);
if (ctx->sybuf)
free_align(ctx->sybuf);
if (ctx->smbuf)
free_align(ctx->smbuf);
}
END_TEST
/*******************************************************************************
* mallac_align/free_align
*/
START_TEST(test_malloc_align)
{
void *ptr[128][256];
int size, align;
for (size = 0; size < countof(ptr); size++)
{
for (align = 0; align < countof(ptr[0]); align++)
{
ptr[size][align] = malloc_align(size, align);
if (align)
{
ck_assert((uintptr_t)ptr[size][align] % align == 0);
}
if (size)
{
ck_assert(ptr[size][align]);
memset(ptr[size][align], 0xEF, size);
}
}
}
for (size = 0; size < countof(ptr); size++)
{
for (align = 0; align < countof(ptr[0]); align++)
{
free_align(ptr[size][align]);
}
}
}
/* free image data */
void free_image(struct image* img) {
if (img != NULL) {
//free(img->data);
free_align(img->data);
img->data = NULL;
}
}
int main(int argc, char* argv[]) {
printf("coucou\n");
int i = 42;
int* i_p = &i;
long* l_p = (long*) i_p;
char* c_p = (char*) i_p;
/*
bitprint((long)i_p);
bitprint((long)(i_p+1)); // + 4
bitprint((long)l_p);
bitprint((long)(l_p+1)); // + 8
bitprint((long)c_p);
bitprint((long)(c_p+1)); // + 8
*/
int align = 24;
if (argc > 1)
align = atoi(argv[1]);
free_align(malloc_align(1024, align));
exit(0);
}
int main(){
int i;
int N=1024;
float pi=0.0;
pthread_t pthreads[SPU_THREADS];
context ctxs[SPU_THREADS] __attribute__ ((aligned(16)));
for(i=0;i<SPU_THREADS;i++){
ctxs[i].N=N;
ctxs[i].Nstart=(N/SPU_THREADS)*i;
ctxs[i].Nend=(N/SPU_THREADS)*(i+1);
ctxs[i].pi=(float*) malloc_align(sizeof(float),7);
pthread_create(&pthreads[i], NULL, &pthread_run_spe, &ctxs[i]);
}
for (i=0; i<SPU_THREADS; i++)
pthread_join (pthreads[i], NULL);
for(i=0;i<SPU_THREADS;i++)
pi+=*(ctxs[i].pi);
for(i=0;i<SPU_THREADS;i++)
free_align(ctxs[i].pi);
printf("PI = %f\n",pi);
return (0);
}
void disable_edma_loop(edma_loop_setup_t *loopSetup)
{
EDMA_DRV_StopChannel(loopSetup->dmaCh);
#if (defined(__ICCARM__) || defined(__CC_ARM))
free_align(loopSetup->dmaChStcd);
#elif defined(__GNUC__)
//OSA_MemFree(loopSetup->dmaChStcd);
#endif
print_edma_ch_erq(DMA0, loopSetup->dmaChanNum);
//OSA_MemFree(loopSetup);
}
void write_btc(char* path, struct c_img* out_img){
int i, nr_blocks, j, fd, k;
struct bits tmp;
char *buf;
fd = _open_for_write(path);
write(fd, &out_img->width, sizeof(int));
write(fd, &out_img->height, sizeof(int));
nr_blocks = out_img->width * out_img->height / (BLOCK_SIZE * BLOCK_SIZE);
buf = _alloc(nr_blocks * (2 + BLOCK_SIZE * BLOCK_SIZE / BITS_IN_BYTE));
k = 0;
for (i=0; i<nr_blocks; i++){
//write a and b
buf[k++] = out_img->blocks[i].a;
buf[k++] = out_img->blocks[i].b;
//from bytes to bits
j = 0;
while (j < BLOCK_SIZE * BLOCK_SIZE){
tmp.bit0 = out_img->blocks[i].bitplane[j++];
tmp.bit1 = out_img->blocks[i].bitplane[j++];
tmp.bit2 = out_img->blocks[i].bitplane[j++];
tmp.bit3 = out_img->blocks[i].bitplane[j++];
tmp.bit4 = out_img->blocks[i].bitplane[j++];
tmp.bit5 = out_img->blocks[i].bitplane[j++];
tmp.bit6 = out_img->blocks[i].bitplane[j++];
tmp.bit7 = out_img->blocks[i].bitplane[j++];
buf[k++] = *((char*)&tmp);
}
//write bitplane
}
_write_buffer(fd, buf,
nr_blocks * (2 + BLOCK_SIZE * BLOCK_SIZE / BITS_IN_BYTE));
free_align(buf);
close(fd);
}
void read_btc(char* path, struct c_img* out_img){
int fd, nr_blocks, i, j = 0, k, ii;
char *big_buf;
struct bits tmp;
fd = _open_for_read(path);
read(fd, &out_img->width, sizeof(int));
read(fd, &out_img->height, sizeof(int));
nr_blocks = out_img->width * out_img->height / (BLOCK_SIZE * BLOCK_SIZE);
out_img->blocks = (struct block*) _alloc(nr_blocks * sizeof(struct block));
big_buf = (char*) _alloc(nr_blocks * (2 + BLOCK_SIZE * BLOCK_SIZE / BITS_IN_BYTE));
_read_buffer(fd, big_buf, nr_blocks * (2 + BLOCK_SIZE * BLOCK_SIZE / BITS_IN_BYTE));
for (i=0; i<nr_blocks; i++){
//read a and b
out_img->blocks[i].a = big_buf[j++];
out_img->blocks[i].b = big_buf[j++];
//read bitplane
k = 0;
for (ii=0; ii<BLOCK_SIZE * BLOCK_SIZE / BITS_IN_BYTE; ii++){
tmp = *((struct bits*)&big_buf[j++]);
out_img->blocks[i].bitplane[k++] = tmp.bit0;
out_img->blocks[i].bitplane[k++] = tmp.bit1;
out_img->blocks[i].bitplane[k++] = tmp.bit2;
out_img->blocks[i].bitplane[k++] = tmp.bit3;
out_img->blocks[i].bitplane[k++] = tmp.bit4;
out_img->blocks[i].bitplane[k++] = tmp.bit5;
out_img->blocks[i].bitplane[k++] = tmp.bit6;
out_img->blocks[i].bitplane[k++] = tmp.bit7;
}
}
free_align(big_buf);
close(fd);
}
static void
print_res(result_p_t res, int rev, seq_p_t seq1, seq_p_t seq2)
{
unsigned int i;
if (res->st.nmatches >= options.minScore_cutoff) {
printf("\n%s%s\n", seq1->header, seq2->header);
if (rev)
printf("(complement)\n\n");
switch (options.ali_flag) {
case 0:
print_exons(&res->eCol, res->direction);
break;
case 1:
print_align_lat(seq1->seq, seq2->seq, res);
break;
case 3:
print_exons(&res->eCol, res->direction);
print_align_lat(seq1->seq, seq2->seq, res);
break;
case 4:
print_exons(&res->eCol, res->direction);
print_polyA_info(seq1, seq2, &res->eCol, &res->st);
print_align_lat(seq1->seq, seq2->seq, res);
break;
default:
fatal("Unrecognized option for alignment output.\n");
}
printf("\n");
}
for (i = 0; i < res->eCol.nb; i++)
free(res->eCol.e.elt[i]);
free(res->eCol.e.elt);
if (res->sList)
free_align(res->sList);
free(res);
}
void process_image_2lines(struct image* img){
unsigned char *input, *output, *output2, *temp;
unsigned int addr1, addr2, i, j, k, r1, g1, b1, r2, g2, b2;
int block_nr = img->block_nr;
vector unsigned char *v1_1, *v1_2, *v1_3, *v1_4, *v1_5;
vector unsigned char *v2_1, *v2_2, *v2_3, *v2_4, *v2_5;
// optimization
unsigned int num_channels_X_img_width = NUM_CHANNELS * img->width;
unsigned int num_channels_X_img_width_X_SCALE_FACTOR = num_channels_X_img_width * SCALE_FACTOR;
input = malloc_align(2 * num_channels_X_img_width_X_SCALE_FACTOR, 4);
output = malloc_align(num_channels_X_img_width / SCALE_FACTOR, 4);
output2 = malloc_align(num_channels_X_img_width / SCALE_FACTOR, 4);
temp = malloc_align(2 * NUM_CHANNELS * img->width, 4);
// first line
v1_1 = (vector unsigned char *) &input[0];
v1_2 = (vector unsigned char *) &input[1 * num_channels_X_img_width];
v1_3 = (vector unsigned char *) &input[2 * num_channels_X_img_width];
v1_4 = (vector unsigned char *) &input[3 * num_channels_X_img_width];
v1_5 = (vector unsigned char *) temp;
// second line
v2_1 = (vector unsigned char *) &input[4 * num_channels_X_img_width];
v2_2 = (vector unsigned char *) &input[5 * num_channels_X_img_width];
v2_3 = (vector unsigned char *) &input[6 * num_channels_X_img_width];
v2_4 = (vector unsigned char *) &input[7 * num_channels_X_img_width];
v2_5 = (vector unsigned char *) &temp[num_channels_X_img_width];
addr2 = (unsigned int)img->dst; //start of image
addr2 += (block_nr / NUM_IMAGES_HEIGHT) * img->width * NUM_CHANNELS *
img->height / NUM_IMAGES_HEIGHT; //start line of spu block
addr2 += (block_nr % NUM_IMAGES_WIDTH) * NUM_CHANNELS *
img->width / NUM_IMAGES_WIDTH;
for (i = 0; i<img->height / SCALE_FACTOR / 2; i++){
// get 8 lines
addr1 = ((unsigned int)img->src) + 2 * i * num_channels_X_img_width_X_SCALE_FACTOR;
mfc_get(input, addr1, 2 * num_channels_X_img_width * SCALE_FACTOR, MY_TAG, 0, 0);
mfc_write_tag_mask(1 << MY_TAG);
mfc_read_tag_status_all();
// compute the 2 scaled line
for (j = 0; j < num_channels_X_img_width / 16; j++){
v1_5[j] = spu_avg(spu_avg(v1_1[j], v1_2[j]), spu_avg(v1_3[j], v1_4[j]));
v2_5[j] = spu_avg(spu_avg(v2_1[j], v2_2[j]), spu_avg(v2_3[j], v2_4[j]));
}
for (j = 0; j < img->width; j += SCALE_FACTOR){
r1 = g1 = b1 = 0;
r2 = b2 = g2 = 0;
for (k = j; k < j + SCALE_FACTOR; k++) {
unsigned int k_X_NUM_CHANNELS = k * NUM_CHANNELS;
r1 += temp[k_X_NUM_CHANNELS + 0];
g1 += temp[k_X_NUM_CHANNELS + 1];
b1 += temp[k_X_NUM_CHANNELS + 2];
r2 += temp[num_channels_X_img_width + k_X_NUM_CHANNELS + 0];
g2 += temp[num_channels_X_img_width + k_X_NUM_CHANNELS + 1];
b2 += temp[num_channels_X_img_width + k_X_NUM_CHANNELS + 2];
}
r1 /= SCALE_FACTOR;
b1 /= SCALE_FACTOR;
g1 /= SCALE_FACTOR;
r2 /= SCALE_FACTOR;
b2 /= SCALE_FACTOR;
g2 /= SCALE_FACTOR;
output[j / SCALE_FACTOR * NUM_CHANNELS + 0] = (unsigned char) r1;
output[j / SCALE_FACTOR * NUM_CHANNELS + 1] = (unsigned char) g1;
output[j / SCALE_FACTOR * NUM_CHANNELS + 2] = (unsigned char) b1;
output2[j / SCALE_FACTOR * NUM_CHANNELS + 0] = (unsigned char) r2;
output2[j / SCALE_FACTOR * NUM_CHANNELS + 1] = (unsigned char) g2;
output2[j / SCALE_FACTOR * NUM_CHANNELS + 2] = (unsigned char) b2;
}
//put the scaled line back
mfc_put(output, addr2, img->width / SCALE_FACTOR * NUM_CHANNELS, MY_TAG, 0, 0);
addr2 += img->width * NUM_CHANNELS; //line inside spu block
// trimite si al 2-lea set
mfc_put(output2, addr2, img->width / SCALE_FACTOR * NUM_CHANNELS, MY_TAG, 0, 0);
addr2 += img->width * NUM_CHANNELS; //line inside spu block
mfc_write_tag_mask(1 << MY_TAG);
mfc_read_tag_status_all();
}
free_align(temp);
free_align(input);
free_align(output);
free_align(output2);
}
void process_image_double(struct image* img){
unsigned char *input[2], *output, *temp;
unsigned int addr1, addr2, i, j, k, r, g, b;
int block_nr = img->block_nr;
vector unsigned char *v1[2], *v2[2], *v3[2], *v4[2], *v5;
int buf, nxt_buf; //index of the buffer (0/1)
input[0] = malloc_align(NUM_CHANNELS * SCALE_FACTOR * img->width, 4);
input[1] = malloc_align(NUM_CHANNELS * SCALE_FACTOR * img->width, 4);
output = malloc_align(NUM_CHANNELS * img->width / SCALE_FACTOR, 4);
temp = malloc_align(NUM_CHANNELS * img->width, 4);
//optimization
unsigned int num_channels_X_img_width = NUM_CHANNELS * img->width;
v1[0] = (vector unsigned char *) &input[0][0];
v2[0] = (vector unsigned char *) &input[0][1 * num_channels_X_img_width];
v3[0] = (vector unsigned char *) &input[0][2 * num_channels_X_img_width];
v4[0] = (vector unsigned char *) &input[0][3 * num_channels_X_img_width];
v5 = (vector unsigned char *) temp;
v1[1] = (vector unsigned char *) &input[1][0];
v2[1] = (vector unsigned char *) &input[1][1 * num_channels_X_img_width];
v3[1] = (vector unsigned char *) &input[1][2 * num_channels_X_img_width];
v4[1] = (vector unsigned char *) &input[1][3 * num_channels_X_img_width];
addr2 = (unsigned int)img->dst; //start of image
addr2 += (block_nr / NUM_IMAGES_HEIGHT) * num_channels_X_img_width *
img->height / NUM_IMAGES_HEIGHT; //start line of spu block
addr2 += (block_nr % NUM_IMAGES_WIDTH) * num_channels_X_img_width / NUM_IMAGES_WIDTH;
addr1 = ((unsigned int)img->src);
buf = 0; // first data transfer
mfc_getb(input[buf], addr1, SCALE_FACTOR * num_channels_X_img_width, 0, 0, 0);
for (i = 1; i<img->height / SCALE_FACTOR; i++){
// get 4 lines
nxt_buf = buf ^ 1; //ask for next data buffer from PPU
//mfg_get with barrier
addr1 = ((unsigned int)img->src) + i * num_channels_X_img_width * SCALE_FACTOR;
mfc_getb(input[nxt_buf], addr1, SCALE_FACTOR * num_channels_X_img_width, nxt_buf, 0, 0);
mfc_write_tag_mask(1 << buf);
mfc_read_tag_status_all();
// process current buffer
for (j = 0; j < img->width * NUM_CHANNELS / 16; j++){
v5[j] = spu_avg(spu_avg(v1[buf][j], v2[buf][j]), spu_avg(v3[buf][j], v4[buf][j]));
}
for (j = 0; j < img->width; j+=SCALE_FACTOR){
r = g = b = 0;
for (k = j; k < j + SCALE_FACTOR; k++) {
r += temp[k * NUM_CHANNELS + 0];
g += temp[k * NUM_CHANNELS + 1];
b += temp[k * NUM_CHANNELS + 2];
}
r /= SCALE_FACTOR;
b /= SCALE_FACTOR;
g /= SCALE_FACTOR;
output[j / SCALE_FACTOR * NUM_CHANNELS + 0] = (unsigned char) r;
output[j / SCALE_FACTOR * NUM_CHANNELS + 1] = (unsigned char) g;
output[j / SCALE_FACTOR * NUM_CHANNELS + 2] = (unsigned char) b;
}
// sent precedent buffer to PPU
mfc_put(output, addr2, img->width / SCALE_FACTOR * NUM_CHANNELS, MY_TAG, 0, 0);
addr2 += img->width * NUM_CHANNELS; //line inside spu block
mfc_write_tag_mask(1 << MY_TAG);
mfc_read_tag_status_all();
buf = nxt_buf; //prepare next iteration
}
mfc_write_tag_mask(1 << buf);
mfc_read_tag_status_all();
// process last buffer
for (j = 0; j < img->width * NUM_CHANNELS / 16; j++){
v5[j] = spu_avg(spu_avg(v1[buf][j], v2[buf][j]), spu_avg(v3[buf][j], v4[buf][j]));
}
for (j=0; j < img->width; j+=SCALE_FACTOR){
r = g = b = 0;
for (k = j; k < j + SCALE_FACTOR; k++) {
r += temp[k * NUM_CHANNELS + 0];
g += temp[k * NUM_CHANNELS + 1];
b += temp[k * NUM_CHANNELS + 2];
}
r /= SCALE_FACTOR;
b /= SCALE_FACTOR;
g /= SCALE_FACTOR;
output[j / SCALE_FACTOR * NUM_CHANNELS + 0] = (unsigned char) r;
output[j / SCALE_FACTOR * NUM_CHANNELS + 1] = (unsigned char) g;
output[j / SCALE_FACTOR * NUM_CHANNELS + 2] = (unsigned char) b;
}
// send last buffer to PPU
mfc_put(output, addr2, img->width / SCALE_FACTOR * NUM_CHANNELS, MY_TAG, 0, 0);
addr2 += img->width * NUM_CHANNELS;
mfc_write_tag_mask(1 << MY_TAG);
mfc_read_tag_status_all();
free_align(temp);
free_align(input[0]);
free_align(input[1]);
free_align(output);
}
static void deallocate(type data)
{
free_align(data);
}
void free_btc(struct c_img* image){
free_align(image->blocks);
}
void free_seed_vector(int* rand_seed)
{
free_align(rand_seed);
}
void free_img(image img)
{
free_align(img->buf);
free_align(img);
}
main(int argc, char **argv)
{
int i, j, k, l, m, n;
int dist[20];
int reads;
int num_vertex, num_class, num_edge;
int *len_seq, num_seq, num_remain;
int **num_pa;
char **src_seq, **src_name;
char temp[100];
ALIGN **eq_class, *align;
EDGE **edge, *edge1, *edge2, *bal_edge1, *bal_edge2;
PATH *path;
int num_path;
NODES **vertex, *begin, *node, *node_next, **start_node;
LIST **list;
READINTERVAL *readinterval;
POSITION *position;
FILE *fp, *fp1;
readpar();
random1(&idum);
initenv(argc, argv);
printf("%d %d %d\n", sizeof(POSITION), sizeof(NODES), sizeof(LIST));
/* Input the length of the genome (required) */
len_seq = (int *) ckalloc(2 * MAX_NUM * sizeof(int));
src_name = alloc_name(MAX_NUM, 100);
fp = ckopen(lenfile, "r");
num_seq = readlen(fp, len_seq, src_name);
fclose(fp);
src_seq = (char **) ckalloc(2 * num_seq * sizeof(char *));
l = 0;
printf("Genome length: ");
for(i = 0; i < num_seq; i ++) {
l += len_seq[i];
printf("%d ", len_seq[i]);
}
printf("\n");
printf("Total length: %d\n", l);
/* Make reverse complements of input sequences rev(i) --> i + num_seq */
for(i = 0; i < num_seq; i ++) {
len_seq[i + num_seq] = len_seq[i];
src_seq[i] = (char *) ckalloc(len_seq[i] * sizeof(char));
src_seq[i + num_seq] = (char *) ckalloc(len_seq[i] * sizeof(char));
for(j = 0; j < len_seq[i]; j ++) {
src_seq[num_seq + i][j] = rev(src_seq[i][len_seq[i] - j - 1]);
}
}
/* Input equivalent readintervales between reads --
see the format of the equivalent readinterval files */
printf("Read equivalent readintervales...\n");
eq_class = (ALIGN **) ckalloc(2 * num_seq * sizeof(ALIGN *));
fp = ckopen(inpfile, "r");
num_class = readclass(eq_class, num_seq, fp);
fclose(fp);
printf("# equivalent readintervales input: %d\n", num_class);
/*
for(i = 0; i < 2 * num_seq; i ++) {
align = eq_class[i];
while(align) {
printf("See: \n");
output_align(align, src_name, src_seq, len_seq, num_seq);
getchar();
align = align -> next;
}
}
*/
/* Initialize the nodes: each position in each read is assigned
as a new node. An array of "list" is set up for each read */
list = (LIST **) ckalloc(2 * num_seq * sizeof(LIST *));
for(i = 0; i < 2 * num_seq; i ++) {
list[i] = (LIST *) ckalloc(len_seq[i] * sizeof(LIST));
}
printf("intitialize nodes...\n");
initialize(list, len_seq, num_seq);
printf("done.\n");
n = countnode(list, len_seq, 2 * num_seq);
printf("# of nodes before merge: %d\n", n);
/* Glue together two nodes if their corresponding positions are defined
as equivalent in a pairwise alignment */
printf("Merge...\n");
merge(num_seq, len_seq, eq_class, num_class, list);
printf("done.\n");
for(i = 0; i < num_seq; i ++) {
while(eq_class[i]) {
eq_class[i] = free_align(eq_class[i]);
}
}
free((void **) eq_class);
/* Compute the width of each node */
for(i = 0; i < 2 * num_seq; i ++) {
for(j = 0; j < len_seq[i]; j ++) {
if(!list[i][j].node -> visit) {
list[i][j].node -> num_path = countthickness(list[i][j].node);
list[i][j].node -> visit = 1;
}
}
}
cleannode(list, len_seq, 2 * num_seq);
n = countnode(list, len_seq, 2 * num_seq);
printf("# of nodes after merge: %d\n", n);
/* Add edges to the graph */
edge = (EDGE **) ckalloc(n * sizeof(EDGE *));
num_edge = graph(num_seq, len_seq, list, edge);
printf("# edges: %d\n", num_edge);
start_node = (NODES **) ckalloc(num_seq * sizeof(NODES *));
for(i = 0; i < num_seq; i ++) {
if(len_seq[i] > 0) {
start_node[i] = list[i][0].node;
} else {
start_node[i] = (NODES *) NULL;
}
}
for(i = 0; i < 2 * num_seq; i ++) {
free((void *) list[i]);
}
free((void **) list);
vertex = (NODES **) ckalloc(2 * num_edge * sizeof(NODES *));
num_vertex = count_vertex(edge, num_edge, vertex);
free((void **) edge);
num_pa = (int **) ckalloc(MAX_BRA * sizeof(int *));
for(i = 0; i < MAX_BRA; i ++) {
num_pa[i] = (int *) ckalloc(MAX_BRA * sizeof(int));
}
num_edge = count_edge_simp(vertex, num_vertex, num_pa);
printf("%d vertices %d edges (%d source %d sinks) remained.\n", num_vertex, num_edge,
num_pa[0][1], num_pa[1][0]);
/* Assign the complementary edges of each edge */
for(i = 0; i < num_vertex; i ++) {
for(j = 0; j < vertex[i] -> num_nextedge; j ++) {
edge1 = vertex[i] -> nextedge[j];
edge1 -> bal_edge = find_bal_edge(edge1, len_seq, num_seq, i);
}
}
/* Remove bulges in the graph */
printf("Shave...\n");
num_vertex = shave_graph(vertex, num_vertex);
printf("done.\n");
/* Remove cycles shorter than some threshold in the graph */
/*
printf("Shaving graph...\n");
num_vertex = rem_cycle(vertex, num_vertex);
printf("done.\n%d vertices remained.\n", num_vertex);
*/
/* remove short edges */
/*
printf("Remove shortedges...\n");
num_vertex = rem_short_edge(vertex, num_vertex, len_seq);
printf("done.\n%d vertices remained.\n", num_vertex);
fflush(stdout);
*/
num_edge = count_edge_simp(vertex, num_vertex, num_pa);
printf("%d vertices %d edges (%d source %d sinks) remained.\n", num_vertex, num_edge,
num_pa[0][1], num_pa[1][0]);
fflush(stdout);
/* Allocate the spaces for paths */
printf("Allocating paths...\n");
for(i = 0; i < num_vertex; i ++) {
vertex[i] -> num_path = 0;
}
/* Build sequence paths */
printf("Define paths...\n");
m = 0;
for(i = 0; i < num_vertex; i ++) {
for(j = 0; j < vertex[i] -> num_nextedge; j ++) {
m += vertex[i] -> nextedge[j] -> multip;
}
}
path = (PATH *) ckalloc(2 * num_seq * sizeof(PATH));
for(i = 0; i < 2 * num_seq; i ++) {
path[i].edge = (EDGE **) ckalloc(m * sizeof(EDGE *));
}
num_path = readpath(start_node, path, num_seq);
free((void **) start_node);
num_edge = count_edge_simp(vertex, num_vertex, num_pa);
m = l = 0;
for(i = 0; i < num_vertex; i ++) {
for(j = 0; j < vertex[i] -> num_nextedge; j ++) {
l += vertex[i] -> nextedge[j] -> length;
if(vertex[i] -> nextedge[j] -> length > m) {
m = vertex[i] -> nextedge[j] -> length;
}
}
}
printf("%d vertics %d edges (%d source %d sinks) remained: total length %d (maximal %d).\n", num_vertex, num_edge,
num_pa[0][1], num_pa[1][0], l, m);
fflush(stdout);
/* Make consensus of edges */
initial_edge(vertex, num_vertex, src_seq, num_seq);
printf("edge initialed\n");
/* Output sequence path */
n = 0;
for(i = 0; i < num_vertex; i ++) {
vertex[i] -> visit = i;
for(j = 0; j < vertex[i] -> num_nextedge; j ++) {
vertex[i] -> nextedge[j] -> start_cover = n;
n ++;
}
}
for(m = 0; m < num_seq; m ++) {
printf("len_path %d\n", path[m].len_path);
printf("Sequence%d: ", m + 1);
for(i = 0; i < path[m].len_path; i ++) {
printf("%d -- %d(%d,%d) --> ", path[m].edge[i] -> begin -> visit,
path[m].edge[i] -> start_cover, path[m].edge[i] -> multip,
path[m].edge[i] -> length);
if(i % 5 == 4) {
printf("\n");
}
}
if(path[m].len_path > 0) {
printf("%d\n", path[m].edge[i - 1] -> end -> visit);
} else {
printf("\n");
}
fflush(stdout);
}
/* Output graph & contigs */
sprintf(temp, "%s.edge", seqfile);
fp = ckopen(temp, "w");
sprintf(temp, "%s.graph", seqfile);
fp1 = ckopen(temp, "w");
write_graph(vertex, num_vertex, fp, fp1);
fclose(fp);
fclose(fp1);
/* Output read intervals in each edge */
sprintf(temp, "%s.intv", seqfile);
fp = ckopen(temp, "w");
write_interval(vertex, num_vertex, fp);
fclose(fp);
/* Output graphviz format graph */
sprintf(temp, "%s", outfile);
fp = ckopen(temp, "w");
output_graph(vertex, num_vertex, fp);
fclose(fp);
for(i = 0; i < MAX_BRA; i ++) {
free((void *) num_pa[i]);
}
free((void **) num_pa);
for(i = 0; i < 2 * num_seq; i ++) {
if(path[i].len_path > 0) {
free((void **) path[i].edge);
}
}
free((void *) path);
free_graph(vertex, num_vertex);
for(i = 0; i < 2 * num_seq; i ++) {
free((void *) src_seq[i]);
}
free((void **) src_seq);
free_name(src_name, MAX_NUM);
free((void *) len_seq);
}
int main(int argc, char **argv)
{
if (argc != 8) {
printf("Usage: ./tema3 mod_vect mod_dma num_spus in.pgm out.cmp out.pgm results.txt");
return -1;
}
int mod_vect = atoi(argv[1]);
int mod_dma = atoi(argv[2]);
int num_spus = atoi(argv[3]);
char *inpgm = argv[4];
char *outcmp = argv[5];
char *outpgm = argv[6];
char *results = argv[7];
int i;
struct img initial_image, decompressed_image;
struct c_img compressed_image;
struct timeval start_total, end_total, start_op, end_op;
double total_time = 0, op_time = 0;
gettimeofday(&start_total, NULL);
// citeste imaginea initiala
read_pgm(inpgm, &initial_image);
gettimeofday(&start_op, NULL);
compressed_image.width = initial_image.width;
compressed_image.height = initial_image.height;
int nr_cmp_blocks = (1LL * initial_image.width * initial_image.height) / (BLOCK_SIZE * BLOCK_SIZE);
compressed_image.blocks = (struct block *)malloc_align(nr_cmp_blocks * sizeof(struct block), 7);
pthread_t *compress_threads = (pthread_t*)malloc_align(num_spus * sizeof(pthread_t), 7);
struct package_t *cthread_arg = (struct package_t *)malloc_align(num_spus * sizeof(struct package_t), 7);
int nr_of_blocks = (initial_image.width * initial_image.height) / (BLOCK_SIZE * BLOCK_SIZE);
int average_blocks = nr_of_blocks / num_spus;
int rest_blocks = nr_of_blocks % num_spus;
int offset = 0;
for(i = 0; i < num_spus; i++) {
/* completeaza structura package_t de trimis la spu pentru fiecare spu*/
cthread_arg[i].action_type = 0;
cthread_arg[i].mod_vect = mod_vect;
cthread_arg[i].mod_dma = mod_dma;
cthread_arg[i].num_spus = num_spus;
cthread_arg[i].nr_blocks = average_blocks;
cthread_arg[i].index_block = offset;
cthread_arg[i].img_pgm.width = initial_image.width;
cthread_arg[i].img_pgm.height = initial_image.height;
cthread_arg[i].img_pgm.pixels = initial_image.pixels + ((offset / (initial_image.width / BLOCK_SIZE)) * BLOCK_SIZE * initial_image.width + (offset % (initial_image.width / BLOCK_SIZE)) * BLOCK_SIZE);
cthread_arg[i].img_cmp.width = compressed_image.width;
cthread_arg[i].img_cmp.height = compressed_image.height;
cthread_arg[i].img_cmp.blocks = compressed_image.blocks + ((offset / (initial_image.width / BLOCK_SIZE)) * (initial_image.width / BLOCK_SIZE) + (offset % (initial_image.width / BLOCK_SIZE)));
offset += average_blocks;
nr_of_blocks -= average_blocks;
if (rest_blocks != 0 && i != num_spus - 1) {
average_blocks = nr_of_blocks / (num_spus - 1 - i);
rest_blocks = nr_of_blocks % (num_spus - 1 - i);
}
/* Create thread for each SPE context */
if (pthread_create (&compress_threads[i], NULL, &ppu_pthread_function, &cthread_arg[i])) {
perror ("Failed creating thread");
exit (1);
}
}
/* Wait for SPU-thread to complete execution. */
for (i = 0; i < num_spus; i++) {
if (pthread_join (compress_threads[i], NULL)) {
perror("Failed pthread_join");
exit (1);
}
}
free_align(compress_threads);
free_align(cthread_arg);
decompressed_image.width = initial_image.width;
decompressed_image.height = initial_image.height;
int nr_dec_blocks = (1LL * initial_image.width * initial_image.height) / (BLOCK_SIZE * BLOCK_SIZE);
decompressed_image.pixels = (unsigned char *)malloc_align(initial_image.height * initial_image.width * sizeof(unsigned char), 7);
pthread_t *decompress_threads = (pthread_t*)malloc_align(num_spus * sizeof(pthread_t), 7);
struct package_t *dthread_arg = (struct package_t *)malloc_align(num_spus * sizeof(struct package_t), 7);
int dec_average_blocks = nr_dec_blocks / num_spus;
int dec_rest_blocks = nr_dec_blocks % num_spus;
int dec_offset = 0;
for(i = 0; i < num_spus; i++) {
/* completeaza structura package_t de trimis la spu pentru fiecare spu*/
dthread_arg[i].action_type = 1;
dthread_arg[i].mod_vect = mod_vect;
dthread_arg[i].mod_dma = mod_dma;
dthread_arg[i].num_spus = num_spus;
dthread_arg[i].nr_blocks = dec_average_blocks;
dthread_arg[i].index_block = dec_offset;
dthread_arg[i].img_pgm.width = initial_image.width;
dthread_arg[i].img_pgm.height = initial_image.height;
dthread_arg[i].img_pgm.pixels = decompressed_image.pixels + ((dec_offset / (initial_image.width / BLOCK_SIZE)) * BLOCK_SIZE * initial_image.width + (dec_offset % (initial_image.width / BLOCK_SIZE)) * BLOCK_SIZE);
dthread_arg[i].img_cmp.width = compressed_image.width;
dthread_arg[i].img_cmp.height = compressed_image.height;
dthread_arg[i].img_cmp.blocks = compressed_image.blocks + ((dec_offset / (initial_image.width / BLOCK_SIZE)) * (initial_image.width / BLOCK_SIZE) + (dec_offset % (initial_image.width / BLOCK_SIZE)));
dec_offset += dec_average_blocks;
nr_dec_blocks -= dec_average_blocks;
if (dec_rest_blocks != 0 && i != num_spus - 1) {
dec_average_blocks = nr_dec_blocks / (num_spus - 1 - i);
dec_rest_blocks = nr_dec_blocks % (num_spus - 1 - i);
}
/* Create thread for each SPE context */
if (pthread_create (&decompress_threads[i], NULL, &ppu_pthread_function, &dthread_arg[i])) {
perror ("Failed creating thread");
exit (1);
}
}
/* Wait for SPU-thread to complete execution. */
for (i = 0; i < num_spus; i++) {
if (pthread_join (decompress_threads[i], NULL)) {
perror("Failed pthread_join");
exit (1);
}
}
gettimeofday(&end_op, NULL);
write_cmp(outcmp, &compressed_image);
write_pgm(outpgm, &decompressed_image);
free_align(compressed_image.blocks);
free_align(decompressed_image.pixels);
free_align(decompress_threads);
free_align(dthread_arg);
gettimeofday(&end_total, NULL);
total_time += GET_TIME_DELTA(start_total, end_total);
op_time += GET_TIME_DELTA(start_op, end_op);
freopen(results, "a+", stdout);
printf("%i %lf %lf\n", num_spus, op_time, total_time);
fclose(stdout);
return 0;
}
/* Does the actual processing of the frame */
static void do_work(ppu_data_t ppu_data) {
struct image input;
struct image big_image;
dprintf("SPU[%d] ppu_data.input:%p ppu_big_img:%p sizeof(struct image):%lu\n",
ppu_data.spe_id, (void *)ppu_data.input,
(void *)ppu_data.big_image, sizeof(struct image));
/* Get input image and big_image details */
mfc_get((void *)(&input), (uint32_t)(ppu_data.input),
(uint32_t)(sizeof(struct image)), tag_id, 0, 0);
mfc_get((void *)(&big_image), (uint32_t)(ppu_data.big_image),
(uint32_t)(sizeof(struct image)), tag_id, 0, 0);
waittag(tag_id);
dprintf("SPU[%d] got structs\n"\
"input.width=%u\tinput.height=%u\n"\
"big_image.width=%u\tbig_image.height=%u\n"\
"input.data=%p\tbig_image.data=%p\n",
ppu_data.spe_id, input.width, input.height, big_image.width,
big_image.height, (void *)input.data, (void *)big_image.data);
struct image img_chunk;
unsigned int buf_line_sz = input.width * NUM_CHANNELS;
int transfer_sz = 4 * buf_line_sz;
img_chunk.width = input.width;
img_chunk.height = 4;
alloc_image(&img_chunk);
struct image img_scaled_line;
img_scaled_line.width = input.width / SCALE_FACTOR;
img_scaled_line.height = 1;
/* Hack for memory align of local image data to have the same 4 bits in its
* address as the remote corresponding address in PPU
*/
int left_padding = (ppu_data.spe_id % 4) * 4;
unsigned char* addr_to_free = malloc_align(NUM_CHANNELS * 3 * sizeof(char) +
left_padding, 4);
img_scaled_line.data = addr_to_free + left_padding;
unsigned int i;
/* Process 4 lines from the initial image at a time */
for (i = 0; i < input.height / img_chunk.height; ++i) {
/* Get the image chunk from PPU through DMA transfer */
dprintf("SPU[%d] getting image_chunk %d of size %d\n",
ppu_data.spe_id, i, transfer_sz);
dprintf("SPU[%d] input.data=%p img_chunk.data=%p "\
"start_addr=%p\n", ppu_data.spe_id, (void *)input.data,
(void *)img_chunk.data, (void *)((uint32_t)(input.data) + i * transfer_sz));
mfc_get((void *)(img_chunk.data), (uint32_t)(input.data) + i * transfer_sz,
(uint32_t)(transfer_sz), tag_id, 0, 0);
waittag(tag_id);
dprintf("SPU[%d] got image_chunk %d\n", ppu_data.spe_id, i);
compute_lines_average(&img_chunk, buf_line_sz);
/* Make average for column. avg = (c0.r + c1.r) / 2 etc*/
compute_columns_average(&img_chunk, &img_scaled_line);
store_line(&img_scaled_line, ppu_data, &big_image, i);
}
free_image(&img_chunk);
free_align(addr_to_free);
}
static void deallocate(type data)
{
free_align(data.first);
free_align(data.second);
}
void closeSortedMap(SortedMap *map) {
free_align(map->keys);
free_align(map->values);
free_align(map);
}
