/* Waits for message from PPU to begin work on the current frame */
static void wait_for_begin(uint32_t *mbox_message) {
do {
while (spu_stat_in_mbox() <= 0)
;
*mbox_message = spu_read_in_mbox();
} while (*mbox_message != BEGIN);
}
int main(uint64_t speid, uint64_t argp, uint64_t envp){
unsigned int data[NUM_STREAMS];
unsigned int num_spus = (unsigned int)argp, i, num_images;
struct image my_image __attribute__ ((aligned(16)));
int mode = (int)envp;
speid = speid; //get rid of warning
while(1){
num_images = 0;
for (i = 0; i < NUM_STREAMS / num_spus; i++){
//assume NUM_STREAMS is a multiple of num_spus
while(spu_stat_in_mbox() == 0);
data[i] = spu_read_in_mbox();
if (!data[i])
return 0;
num_images++;
}
for (i = 0; i < num_images; i++){
mfc_get(&my_image, data[i], sizeof(struct image), MY_TAG, 0, 0);
mfc_write_tag_mask(1 << MY_TAG);
mfc_read_tag_status_all();
switch(mode){
default:
case MODE_SIMPLE:
process_image_simple(&my_image);
break;
case MODE_2LINES:
process_image_2lines(&my_image);
break;
case MODE_DOUBLE:
process_image_double(&my_image);
break;
case MODE_DMALIST:
process_image_dmalist(&my_image);
break;
}
}
data[0] = DONE;
spu_write_out_intr_mbox(data[0]);
}
return 0;
}
// -- is there something available in my mailbox? -----------------------------
int as_mbx_avail ()
{
return (spu_stat_in_mbox () > 0);
}
void process_render_tasks(unsigned long eah_render_tasks, unsigned long eal_render_tasks)
{
const vec_uchar16 SHUFFLE_MERGE_BYTES = (vec_uchar16) { // merge lo bytes from unsigned shorts (array)
1,3,5,7,9,11,13,15,17,19,21,23,25,27,29,31
};
const vec_uchar16 SHUFFLE_GET_BUSY_WITH_ONES = (vec_uchar16) { // get busy flag with ones in unused bytes
0xc0, 0xc0, 2, 3, 0xc0,0xc0,0xc0,0xc0, 0xc0,0xc0,0xc0,0xc0
};
const vec_uchar16 ZERO_BYTES = (vec_uchar16) spu_splats(0);
char trianglebuffer[ 256 + TRIANGLE_MAX_SIZE ];
char sync_buffer[128+127];
void* aligned_sync_buffer = (void*) ( ((unsigned long)sync_buffer+127) & ~127 );
RenderableCacheLine* cache = (RenderableCacheLine*) aligned_sync_buffer;
unsigned long long cache_ea;
spu_mfcdma64(&cache_ea, eah_render_tasks, eal_render_tasks, sizeof(cache_ea), 0, MFC_GET_CMD);
mfc_write_tag_mask(1<<0);
mfc_read_tag_status_all();
while (cache_ea) {
// terminate immediately if possible
if (spu_stat_in_mbox())
return;
// read the cache line
spu_mfcdma64(cache, mfc_ea2h(cache_ea), mfc_ea2l(cache_ea), 128, 0, MFC_GETLLAR_CMD);
spu_readch(MFC_RdAtomicStat);
unsigned int endTriangle = cache->endTriangle;
vec_ushort8 testTriangle = spu_splats((unsigned short) endTriangle);
// first look for short chunks
vec_uchar16 next = cache->chunkNext;
vec_uchar16 nextmask = spu_and(next, spu_splats((unsigned char)CHUNKNEXT_MASK));
// change next to word offset, note we don't care what the low bit shifted in is
vec_uchar16 firstshuf = (vec_uchar16) spu_sl( (vec_ushort8)nextmask, 1 );
vec_uchar16 trishufhi = spu_or ( firstshuf, spu_splats((unsigned char) 1));
vec_uchar16 trishuflo = spu_and( firstshuf, spu_splats((unsigned char) 254));
vec_ushort8 start0 = cache->chunkStart[0];
vec_ushort8 start1 = cache->chunkStart[1];
vec_ushort8 nstart0 = spu_shuffle( start0, start1, spu_shuffle( trishuflo, trishufhi, SHUF0 ) );
vec_ushort8 nstart1 = spu_shuffle( start0, start1, spu_shuffle( trishuflo, trishufhi, SHUF1 ) );
vec_ushort8 starteq0 = spu_cmpeq( nstart0, spu_splats((unsigned short)0) );
vec_ushort8 starteq1 = spu_cmpeq( nstart1, spu_splats((unsigned short)0) );
vec_ushort8 end0 = spu_sel( nstart0, spu_splats((unsigned short)4096), starteq0);
vec_ushort8 end1 = spu_sel( nstart1, spu_splats((unsigned short)4096), starteq1);
vec_ushort8 len0 = spu_sub( end0, start0);
vec_ushort8 len1 = spu_sub( end1, start1);
vec_ushort8 small0 = spu_cmpgt( spu_splats((unsigned short)17), len0);
vec_ushort8 small1 = spu_cmpgt( spu_splats((unsigned short)17), len1);
vec_uchar16 small = (vec_uchar16) spu_shuffle( small0, small1, MERGE );
vec_uint4 smallChunkGather = spu_gather(small);
// check to see if chunk is already at the last triangle
vec_uint4 doneChunkGather = spu_gather( (vec_uchar16) spu_shuffle(
(vec_uchar16) spu_cmpeq(testTriangle, cache->chunkTriangle[0]),
(vec_uchar16) spu_cmpeq(testTriangle, cache->chunkTriangle[1]),
SHUFFLE_MERGE_BYTES) );
// check if the chunk is free
vec_uint4 freeChunkGather = spu_gather(
spu_cmpeq( spu_splats( (unsigned char) CHUNKNEXT_FREE_BLOCK ), cache->chunkNext ) );
// check to see if the chunk is being processed
vec_uint4 busyChunkGather = spu_gather(
spu_cmpgt( cache->chunkNext, //spu_and(cache->chunkNext, CHUNKNEXT_MASK),
spu_splats( (unsigned char) (CHUNKNEXT_BUSY_BIT-1) ) ) );
// doneChunkGather, freeChunkGather, busyChunkGather - rightmost 16 bits of word 0
// note that if freeChunkGather is true then busyChunkGather must also be true
// done=false, free=false, busy=false -> can process
// free=false, busy=false -> can be merged
// decide which chunk to process
vec_uint4 mayProcessGather = spu_nor( doneChunkGather, busyChunkGather );
vec_uint4 mayProcessShortGather = spu_and( mayProcessGather, smallChunkGather );
vec_uint4 shortSelMask = spu_cmpeq( mayProcessShortGather, spu_splats(0U) );
vec_uint4 mayProcessSelection = spu_sel( mayProcessShortGather, mayProcessGather, shortSelMask );
/*
if (!spu_extract(shortSelMask, 0))
printf("taken short: may=%04x short=%04x mayshort=%04x mask=%04x sel=%04x\n",
spu_extract(mayProcessGather, 0) & 0xffff,
spu_extract(smallChunkGather, 0),
spu_extract(mayProcessShortGather, 0),
spu_extract(shortSelMask, 0) & 0xffff,
spu_extract(mayProcessSelection, 0) & 0xffff );
*/
vec_uint4 mayProcessBits = spu_sl( mayProcessSelection, 16);
unsigned int chunkToProcess = spu_extract( spu_cntlz( mayProcessBits ), 0);
unsigned int freeChunk = spu_extract( spu_cntlz( spu_sl( freeChunkGather, 16 ) ), 0);
// if there's nothing to process, try the next cache line in the rendering tasks list
if (!spu_extract(mayProcessBits, 0)) {
trynextcacheline:
cache_ea = cache->next;
// sleep();
continue;
}
unsigned int chunkStart = cache->chunkStartArray [chunkToProcess];
unsigned int chunkTriangle = cache->chunkTriangleArray[chunkToProcess];
unsigned int chunkNext = cache->chunkNextArray [chunkToProcess] & CHUNKNEXT_MASK;
unsigned int chunkEnd = (cache->chunkStartArray [chunkNext]-1) & (NUMBER_OF_TILES-1);
unsigned int chunkLength = 1 + chunkEnd-chunkStart;
// only need an extra block if the block is especially long
if (chunkLength <= NUMBER_OF_TILES_PER_CHUNK) {
freeChunk = 32;
}
// mark this block as busy
cache->chunkNextArray[chunkToProcess] |= CHUNKNEXT_BUSY_BIT;
// if there's at least one free chunk, claim it
if (freeChunk != 32) {
cache->chunkNextArray[freeChunk] = CHUNKNEXT_RESERVED;
cache->chunkTriangleArray[freeChunk] = chunkTriangle;
}
// write the cache line back
spu_mfcdma64(cache, mfc_ea2h(cache_ea), mfc_ea2l(cache_ea), 128, 0, MFC_PUTLLC_CMD);
if (spu_readch(MFC_RdAtomicStat) & MFC_PUTLLC_STATUS)
continue;
#ifdef INFO
printf("[%d] Claimed chunk %d (%d-%d len %d) at tri %x end %x with free chunk %d\n", _SPUID,
chunkToProcess, chunkStart, chunkEnd, chunkLength, chunkTriangle, endTriangle,
freeChunk!=32 ? freeChunk : -1 );
// debug_render_tasks(cache);
#endif
Triangle* triangle;
int firstTile;
do {
// read the triangle data for the current triangle
unsigned int extra = chunkTriangle & 127;
unsigned long long trianglebuffer_ea = cache_ea + TRIANGLE_OFFSET_FROM_CACHE_LINE + (chunkTriangle & ~127);
triangle = (Triangle*) (trianglebuffer+extra);
unsigned int length = (extra + TRIANGLE_MAX_SIZE + 127) & ~127;
// ensure DMA slot available
do {} while (!spu_readchcnt(MFC_Cmd));
spu_mfcdma64(trianglebuffer, mfc_ea2h(trianglebuffer_ea), mfc_ea2l(trianglebuffer_ea),
length, 0, MFC_GET_CMD);
mfc_write_tag_mask(1<<0);
mfc_read_tag_status_all();
// get the triangle deltas
firstTile = findFirstTriangleTile(triangle, chunkStart, chunkEnd);
if (firstTile>=0)
break;
// no match, try next triangle
chunkTriangle = triangle->next_triangle;
} while (chunkTriangle != endTriangle);
// if we actually have something to process...
if (firstTile>=0) {
// the "normal" splitting will now become:
// chunkStart .. (firstTile-1) -> triangle->next_triangle
// firstTile .. (firstTile+NUM-1) -> chunkTriangle (BUSY)
// (firstTile+NUM) .. chunkEnd -> chunkTriangle (FREE)
int tailChunk;
int thisChunk;
int nextBlockStart;
int thisBlockStart;
int realBlockStart;
do {
retry:
// read the cache line
spu_mfcdma64(cache, mfc_ea2h(cache_ea), mfc_ea2l(cache_ea), 128, 0, MFC_GETLLAR_CMD);
spu_readch(MFC_RdAtomicStat);
// calculate start of next block
nextBlockStart = firstTile + NUMBER_OF_TILES_PER_CHUNK;
if (nextBlockStart > chunkEnd)
nextBlockStart = chunkEnd+1;
// calculate start of block to mark as busy
thisBlockStart = nextBlockStart - NUMBER_OF_TILES_PER_CHUNK;
if (thisBlockStart < chunkStart)
thisBlockStart = chunkStart;
realBlockStart = thisBlockStart;
#ifdef INFO
printf("[%d] nextBlockStart=%d, realBlockStart=%d, thisBlockStart=%d, chunkStart=%d\n", _SPUID,
nextBlockStart, realBlockStart, thisBlockStart, chunkStart);
#endif
// allocate some more free chunks
vec_uint4 freeChunkGather2 = spu_sl(spu_gather(spu_cmpeq(
spu_splats((unsigned char)CHUNKNEXT_FREE_BLOCK), cache->chunkNext)), 16);
unsigned int freeChunk2 = spu_extract(spu_cntlz(freeChunkGather2), 0);
if (freeChunk == 32) {
// if we didn't have one before, try again
freeChunk = freeChunk2;
// and try to get the second one
freeChunkGather2 = spu_andc( freeChunkGather2, spu_promote(0x80000000>>freeChunk2, 0) );
freeChunk2 = spu_extract(spu_cntlz(freeChunkGather2), 0);
} else {
// speculatively clear the free chunk just in case we don't need it
cache->chunkNextArray[freeChunk] = CHUNKNEXT_FREE_BLOCK;
}
#ifdef INFO
printf("[%d] Free chunks %d and %d, cN=%d, nBS=%d, cE=%d, tBS=%d, cS=%d\n",
_SPUID, freeChunk, freeChunk2, chunkNext, nextBlockStart, chunkEnd, thisBlockStart, chunkStart );
#endif
// mark region after as available for processing if required
if (nextBlockStart < chunkEnd) {
if (freeChunk==32) {
// if no free chunk, relinquish entire block and write back
cache->chunkNextArray[chunkToProcess] = chunkNext;
spu_mfcdma64(cache, mfc_ea2h(cache_ea), mfc_ea2l(cache_ea), 128, 0, MFC_PUTLLC_CMD);
// if writeback failed, we *might* have a free block, retry
if (spu_readch(MFC_RdAtomicStat) & MFC_PUTLLC_STATUS)
goto retry;
// otherwise give up and try the next cache line
goto trynextcacheline;
}
cache->chunkStartArray[freeChunk] = nextBlockStart;
cache->chunkNextArray[freeChunk] = chunkNext;
cache->chunkTriangleArray[freeChunk] = chunkTriangle;
cache->chunkNextArray[chunkToProcess] = freeChunk | CHUNKNEXT_BUSY_BIT;
tailChunk = freeChunk;
#ifdef INFO
printf("[%d] Insert tail, tailChunk=%d, chunkNext=%d, chunkToProcess=%d\n", _SPUID, tailChunk, chunkNext, chunkToProcess);
debug_render_tasks(cache);
#endif
} else {
// we're gonna use freeChunk2 for the "in front" block, as we've not
// used freeChunk, let's use it as it's more likely to have a free chunk
freeChunk2 = freeChunk;
tailChunk = chunkNext;
}
// mark region before as available if required and possible
thisChunk = chunkToProcess;
if (thisBlockStart > chunkStart) {
if (freeChunk2 != 32) {
// mark this region as busy
cache->chunkStartArray[freeChunk2]=thisBlockStart;
cache->chunkNextArray[freeChunk2]=tailChunk | CHUNKNEXT_BUSY_BIT;
cache->chunkTriangleArray[freeChunk2]=chunkTriangle;
// mark region before as available for processing
cache->chunkNextArray[chunkToProcess]=freeChunk2;
cache->chunkTriangleArray[chunkToProcess]=triangle->next_triangle;
thisChunk = freeChunk2;
#ifdef INFO
printf("[%d] Insert new head, tailChunk=%d, chunkNext=%d, thisChunk=%d\n", _SPUID, tailChunk, chunkNext, thisChunk);
debug_render_tasks(cache);
#endif
} else {
// need to keep whole block, update info and mark bust
cache->chunkTriangleArray[chunkToProcess]=chunkTriangle;
cache->chunkNextArray[chunkToProcess]=tailChunk | CHUNKNEXT_BUSY_BIT;
realBlockStart = chunkStart;
printf("[%d] Keep whole block, tailChunk=%d, chunkNext=%d, thisChunk=%d\n", _SPUID, tailChunk, chunkNext, thisChunk);
debug_render_tasks(cache);
#ifdef INFO
#endif
sleep();
}
}
// merge chunks
merge_cache_blocks(cache);
// write the cache line back
spu_mfcdma64(cache, mfc_ea2h(cache_ea), mfc_ea2l(cache_ea), 128, 0, MFC_PUTLLC_CMD);
} while (spu_readch(MFC_RdAtomicStat) & MFC_PUTLLC_STATUS);
// finally after the write succeeded, update the variables
chunkNext = tailChunk;
chunkToProcess = thisChunk;
chunkStart = firstTile; //thisBlockStart;
chunkLength = nextBlockStart - firstTile;
chunkEnd = chunkStart + chunkLength - 1;
freeChunk = 32;
// now we can process the block up to endTriangle
initTileBuffers(thisBlockStart, chunkEnd);
int ok=0;
while (chunkTriangle != endTriangle) {
#ifdef INFO
printf("[%d] Processing chunk %d at %4d len %4d, triangle %04x first=%d tbs=%d\n",
_SPUID, chunkToProcess, chunkStart, chunkLength,
chunkTriangle, firstTile, thisBlockStart);
#endif
// and actually process that triangle on these chunks
processTriangleChunks(triangle, cache, thisBlockStart, chunkEnd, chunkTriangle, ok);
ok=1;
#ifdef PAUSE
sleep();
#endif
// and advance to the next-triangle
chunkTriangle = triangle->next_triangle;
// this should only ever happen if we're running really low on cache line slots
// basically, if we pick up a block with more than NUMBER_OF_TILES_PER_CHUNK and
// there's no slot to store the pre-NUMBER_OF_TILES_PER_CHUNK tiles.
// in this case, we process from thisBlockStart only (because we know that from
// chunkStart to there has no result) and then we only process one triangle
if (chunkStart != realBlockStart) {
/*
printf("[%d] chunkStart=%d != realBlockStart %d, chunkEnd=%d, "
"firstTile=%d chunk=%d\n",
_SPUID, chunkStart, realBlockStart, chunkEnd,
firstTile, chunkToProcess);
debug_render_tasks(cache);
*/
// abort the while loop
break;
}
// read the next triangle
unsigned int extra = chunkTriangle & 127;
unsigned long long trianglebuffer_ea = cache_ea + TRIANGLE_OFFSET_FROM_CACHE_LINE + (chunkTriangle & ~127);
triangle = (Triangle*) (trianglebuffer+extra);
unsigned int length = (extra + TRIANGLE_MAX_SIZE + 127) & ~127;
// ensure DMA slot available
do {} while (!spu_readchcnt(MFC_Cmd));
spu_mfcdma64(trianglebuffer, mfc_ea2h(trianglebuffer_ea),
mfc_ea2l(trianglebuffer_ea), length, 0, MFC_GET_CMD);
mfc_write_tag_mask(1<<0);
mfc_read_tag_status_all();
} // until chunkTriangle == endTriangle
// flush any output buffers
flushTileBuffers(thisBlockStart, chunkEnd);
} // firstTile>=0
int main(unsigned long long speid, unsigned long long argp, unsigned long long envp)
{
int tgiy0[2];
int tgiy1[2];
int tgiu0[2];
int tgiu1[2];
int tgiv0[2];
int tgiv1[2];
int tgo0[2];
int tgo1[2];
tgiu1[0]=1;
tgiu1[1]=2;
tgo0[0]=3;
tgo0[1]=4;
tgiy0[0]=5;
tgiy0[1]=6;
tgiy1[0]=7;
tgiy1[1]=8;
tgiu0[0]=9;
tgiu0[1]=10;
tgiv0[0]=11;
tgiv0[1]=12;
tgiv1[1]=13;
tgiv1[1]=14;
tgo1[0]=15;
tgo1[1]=16;
int selOut = 0;
int selIn = 0;
int tag = 31;
int LineSelIn=0;
int LineSelOut=0;
int selY0In = 0;
int selY1In = 0;
int selCrIn = 0;
struct img_args *iargs;
iargs =(struct img_args*)memalign(128,sizeof(*iargs));
unsigned long long Cp;
int first=1;
int waiting=0;
unsigned long long Op;
unsigned int msg;
unsigned long long YIp,UIp,VIp,YOp;
int crblock0;
int crblock1;
int srcsmallcroma=0;
;
int noscale=1;
static int crblockdst1;
static int crblockdst0;
scaler_settings_t sc;
while (spu_stat_in_mbox() == 0);
msg=spu_read_in_mbox();
if (msg==RUN){
fprintf(stderr,"spu_yuv2argb_scaler: Starting Up\n");
}
dmaGetnWait(iargs,(unsigned int)argp,(int)envp,tag); //getting neccesary data to process image
printf("spu_yuv2argb_scaler: SRC width %d,DST width %d\n",iargs->srcW,iargs->dstW);
printf("spu_yuv2argb_scaler: SRC height %d,DST height %d\n",iargs->srcH,iargs->dstH);
printf("spu_yuv2argb_scaler: DST offset %d\n",iargs->offset);
// bad fix for centering image on 1080p)
//iargs->offset=(iargs->maxwidth-iargs->dstW)/2 + iargs->maxwidth*(1080-iargs->dstH)/2;
vector unsigned char *widthfilter0=(vector unsigned char*)memalign(128,MAXWIDTH*4+16);
vector unsigned char *widthfilter1=(vector unsigned char*)memalign(128,MAXWIDTH*4+16);
vector unsigned char *crwidthfilter0=(vector unsigned char*)memalign(128,MAXWIDTH*2+16);
vector unsigned char *crwidthfilter1=(vector unsigned char*)memalign(128,MAXWIDTH*2+16);
vector float * weightWfilter0=(vector float*)memalign(128,MAXWIDTH*4+16);
vector float * weightWfilter1=(vector float*)memalign(128,MAXWIDTH*4+16);
float weightHfilter[MAXHEIGHT+1];
unsigned long long dmapos[MAXHEIGHT+2];
unsigned long long dmacromapos[MAXHEIGHT+2];
vector float * Ytemp0=(vector float *)memalign(128,MAXWIDTH*4+16);
vector float * Ytemp1=(vector float *)memalign(128,MAXWIDTH*4+16);
vector float * Utemp=(vector float *)memalign(128,MAXWIDTH*2+16);
vector float * Vtemp=(vector float *)memalign(128,MAXWIDTH*2+16);
int wfilterpos[MAXWIDTH+2];
int hfilterpos0[MAXHEIGHT+2];
int hfilterpos1[MAXHEIGHT+2];
int crwfilterpos[MAXWIDTH/2+2];
vector unsigned char *InputY0[2];
InputY0[0]=(vector unsigned char*)memalign(128,MAXWIDTH);
InputY0[1]=(vector unsigned char*)memalign(128,MAXWIDTH);
vector unsigned char *InputU0[2];
InputU0[0]=(vector unsigned char*)memalign(128,MAXWIDTH/2+16);
InputU0[1]=(vector unsigned char*)memalign(128,MAXWIDTH/2+16);
vector unsigned char *InputV0[2];
InputV0[0]=(vector unsigned char*)memalign(128,MAXWIDTH/2+16);
InputV0[1]=(vector unsigned char*)memalign(128,MAXWIDTH/2+16);
vector unsigned char *InputY1[2];
InputY1[0]=(vector unsigned char*)memalign(128,MAXWIDTH);
InputY1[1]=(vector unsigned char*)memalign(128,MAXWIDTH);
vector unsigned char *InputU1[2];
InputU1[0]=(vector unsigned char*)memalign(128,MAXWIDTH/2+16);
InputU1[1]=(vector unsigned char*)memalign(128,MAXWIDTH/2+16);
vector unsigned char *InputV1[2];
InputV1[0]=(vector unsigned char*)memalign(128,MAXWIDTH/2+16);
InputV1[1]=(vector unsigned char*)memalign(128,MAXWIDTH/2+16);
vector unsigned char* Output0[2];
Output0[0]=(vector unsigned char*)memalign(128,MAXWIDTH*4); // 1line output
Output0[1]=(vector unsigned char*)memalign(128,MAXWIDTH*4); // 1line output
vector unsigned char* Output1[2];
Output1[0]=(vector unsigned char*)memalign(128,MAXWIDTH*4); // 1line output
Output1[1]=(vector unsigned char*)memalign(128,MAXWIDTH*4); // 1line output
while (msg!=STOP)
{
int h=0;
int i;
if (first)
{
crblock0=(iargs->srcW>>1)&~15; // rounded down
crblock1=((iargs->srcW>>1) + 15)&~15; //rounded up
crblockdst1=((iargs->dstW>>1) + 15)&~15;//destination size rounded up.
crblockdst0=((iargs->dstW>>1) + 7)&~7;//destination size rounded up.
initHFilter(iargs->srcW,iargs->srcH,iargs->dstH,hfilterpos0,hfilterpos1,weightHfilter,dmapos,dmacromapos);
// printf("line :%d, dmapos :%f, dmacromapos :%f \n",i,dmapos[hfilterpos1[1]]/16.0,dmacromapos[hfilterpos1[0]]/16.0);
// printf("line :%d, dmapos :%f, dmacromapos :%f \n",i,dmapos[hfilterpos1[1]]/16.0,dmacromapos[hfilterpos1[1]]/16.0);
//
// for (i=0;i < iargs->dstH>>1;i++)
// {
// // printf("Hfilterpos0 dst: %d, src:%d, weight:%f\n",i,hfilterpos0[i],weightHfilter[i]);
// // printf("Hfilterpos1 dst: %d, src:%d, weight:%f\n",i,hfilterpos1[i],1.0-weightHfilter[i]);
// printf("line :%d, dmapos :%f, dmacromapos :%f \n",i,dmapos[hfilterpos1[2*i+2]]/16.0,dmacromapos[hfilterpos1[2*i+2]]/16.0);
// printf("line :%d, dmapos :%f, dmacromapos :%f \n",i,dmapos[hfilterpos1[2*i+3]]/16.0,dmacromapos[hfilterpos1[2*i+3]]/16.0);
// }
if ((iargs->srcW==iargs->dstW)&&(iargs->srcH==iargs->dstH))
{
printf("spu_yuv2argb_scaler: No scaling proceeding with direct csc\n");
noscale=1;
if ((iargs->srcW%32) != 0)
{
srcsmallcroma=1;
sc.smallcroma=1;
}
} else {
noscale=0;
printf("spu_yuv2argb_scaler: Scaling, computing shuffle filters\n");
initWFilter(iargs->srcW,iargs->dstW,1,wfilterpos,widthfilter0,widthfilter1,weightWfilter0,weightWfilter1);
/* for (i=0;i < iargs->dstW/4;i++)
{
printf("filterpos dst: %d, src:%d\n",i,wfilterpos[i]);
printcharvec("widthfilter0",widthfilter0[i]);
printcharvec("widthfilter1",widthfilter1[i]);
printfvec("weightWfilter0",weightWfilter0[i]);
printfvec("weightWfilter1",weightWfilter1[i]);
}*/
srcsmallcroma=0;
sc.smallcroma=0;
if ((iargs->srcW%32) != 0)
{
sc.smallcroma=1;
srcsmallcroma=1;
initWcrFilter(iargs->srcW/2,iargs->dstW/2,1,crwfilterpos,crwidthfilter0,crwidthfilter1);
printf("spu_yuv2argb_scaler: Computing Crshuffle filter\n");
// for (i=0;i < (iargs->dstW>>1)/4;i++)
// {
// printf("crwfilterpos dst: %d, src:%d, weight:%f\n",i,crwfilterpos[i]);
// printcharvec("crwidthfilter0",crwidthfilter0[i]);
// printcharvec("crwidthfilter1",crwidthfilter1[i]);
// printfvec("weightWfilter0",weightWfilter0[i]);
// printfvec("weightWfilter1",weightWfilter1[i]);
//
// }
}
sc.wWfilter0=weightWfilter0;
sc.wWfilter1=weightWfilter1;
sc.wfilterpos=wfilterpos;
sc.sWfilter0=widthfilter0;
sc.sWfilter1=widthfilter1;
sc.crsWfilter0=crwidthfilter0;
sc.crsWfilter1=crwidthfilter1;
sc.crfilterpos=crwfilterpos;
sc.smallcromaline0=0;
sc.smallcromaline1=0;
}
first=0;
printf("spu_yuv2argb_scaler: Initiation completed\n");
}
YIp = iargs->Ystart[selIn];
UIp = iargs->Ustart[selIn];
VIp = iargs->Vstart[selIn];
Op = iargs->Output[selOut] + iargs->offset*4;
LineSelOut=0;
selY0In=0;
selY1In=0;
selCrIn=0;
dmaGet(InputY0[0],YIp+dmapos[hfilterpos0[0]],iargs->srcW,tgiy0[0]);
dmaGet(InputY1[0],YIp+dmapos[hfilterpos1[0]],iargs->srcW,tgiy1[0]);
dmaGet(InputY0[1],YIp+dmapos[hfilterpos0[1]],iargs->srcW,tgiy0[1]);
dmaGet(InputY1[1],YIp+dmapos[hfilterpos1[1]],iargs->srcW,tgiy1[1]);
dmaGet(InputU0[0],UIp+dmacromapos[hfilterpos0[0]],crblock1,tgiu0[0]);
dmaGet(InputU0[1],UIp+dmacromapos[hfilterpos0[1]],crblock1,tgiu0[1]);
dmaGet(InputU1[0],UIp+dmacromapos[hfilterpos1[0]],crblock1,tgiu1[0]);
dmaGet(InputU1[1],UIp+dmacromapos[hfilterpos1[1]],crblock1,tgiu1[1]);
//
dmaGet(InputV0[0],VIp+dmacromapos[hfilterpos0[0]],crblock1,tgiv0[0]);
dmaGet(InputV0[1],VIp+dmacromapos[hfilterpos0[1]],crblock1,tgiv0[1]);
dmaGet(InputV1[0],VIp+dmacromapos[hfilterpos1[0]],crblock1,tgiv1[0]);
dmaGet(InputV1[1],VIp+dmacromapos[hfilterpos1[1]],crblock1,tgiv1[1]);
LineSelOut=0;
selY0In=0;
selY1In=0;
selCrIn=0;
// printf("New image\n");
for (h=0; h < iargs->dstH>>1; h++) //we asume that output is allways h/2
{
sc.width=iargs->dstW;
sc.smallcroma=0;
sc.smallcromaline0=0;
sc.smallcromaline1=0;
sc.wHfilter=weightHfilter[2*h];
dmaWaitTag(tgiy0[selY0In]);
// printf("dma: %d\n",2*h+2);
dmaWaitTag(tgiy1[selY1In]);
// printf("dma: %d\n",2*h+2);
sc.source00=InputY0[selY0In];
sc.source01=InputY1[selY1In];
sc.Output=Ytemp0;
if (noscale) {
unpack(&sc);
} else {
scale(&sc);
}
//first Y line scaled
dmaGet(InputY0[selY0In],YIp+dmapos[hfilterpos0[2*h+2]],iargs->srcW,tgiy0[selY0In]);
// printf("dma: %d\n",2*h+2);
if (!noscale) { //if we are scaling we also need the second line
dmaGet(InputY1[selY1In],YIp+dmapos[hfilterpos1[2*h+2]],iargs->srcW,tgiy1[selY1In]);
}
// printf("dma: %d\n",2*h+2);
selY0In=selY0In^1;
selY1In=selY1In^1;
sc.wHfilter=weightHfilter[2*h+1];
dmaWaitTag(tgiy0[selY0In]);
dmaWaitTag(tgiy1[selY0In]);
sc.source00=InputY0[selY0In];
sc.source01=InputY1[selY0In];
sc.Output=Ytemp1;
if (noscale) {
unpack(&sc);
} else {
scale(&sc);
} //second Y line scaled
dmaGet(InputY0[selY0In],YIp+dmapos[hfilterpos0[2*h+3]],iargs->srcW,tgiy0[selY0In]);
if(!noscale) { //if we are scaling we also need the second line
dmaGet(InputY1[selY1In],YIp+dmapos[hfilterpos1[2*h+3]],iargs->srcW,tgiy1[selY1In]);
}
selY0In=selY0In^1;
selY1In=selY1In^1;
// printf("dma: %d\n",2*h+3);
if (srcsmallcroma) //these settings applly for both U and V
{
sc.smallcroma=1;
if ((hfilterpos0[h]&1)==1) {
sc.smallcromaline0=1;
} else {
sc.smallcromaline0=0;
}
if ((hfilterpos1[h]&1)==1){
sc.smallcromaline1=1;
} else {
sc.smallcromaline1=0;
}
if (((hfilterpos0[h]&1)==0)&&((hfilterpos1[h]&1)==0))
{
sc.smallcroma=0; //both lines are 128 bit alligned only when doing extreme downscaling can this happen
}
}
// if (noscale) {
// sc.width=crblockdst0;//crblockdst1;
// } else {
// sc.width=crblockdst0;
// }
sc.width=iargs->dstW>>1;
sc.wHfilter=weightHfilter[h];
dmaWaitTag(tgiu0[selCrIn]);
dmaWaitTag(tgiu1[selCrIn]);
sc.Output=Utemp;
sc.source00=InputU0[selCrIn];
sc.source01=InputU1[selCrIn];
if (noscale) {
unpack(&sc);
} else {
scale(&sc);
}
dmaWaitTag(tgiv0[selCrIn]);
dmaWaitTag(tgiv1[selCrIn]);
sc.Output=Vtemp;
sc.source00=InputV0[selCrIn];
sc.source01=InputV1[selCrIn];
if (noscale) {
unpack(&sc);
} else {
scale(&sc);
}
dmaGet(InputV0[selCrIn],VIp+dmacromapos[hfilterpos0[h+2]],crblock1,tgiu0[selCrIn]); //this is allways pos 0
dmaGet(InputU0[selCrIn],UIp+dmacromapos[hfilterpos0[h+2]],crblock1,tgiv0[selCrIn]);
if(!noscale) { //if we are scaling we also need the second line
dmaGet(InputV1[selCrIn],VIp+dmacromapos[hfilterpos1[h+2]],crblock1,tgiu1[selCrIn]);
dmaGet(InputU1[selCrIn],UIp+dmacromapos[hfilterpos1[h+2]],crblock1,tgiv1[selCrIn]);
}
selCrIn=selCrIn^1;
dmaWaitTag(tgo0[LineSelOut]);
dmaWaitTag(tgo1[LineSelOut]);
yuv420toARGBfloat(Ytemp0,Ytemp1,Utemp,Vtemp,Output0[LineSelOut],Output1[LineSelOut],iargs->dstW,iargs->maxwidth); //colorspace convert results
dmaPut(Output0[LineSelOut],Op,iargs->dstW*4,tgo0[LineSelOut]);
Op=Op+iargs->maxwidth*4;
dmaPut(Output1[LineSelOut],Op,iargs->dstW*4,tgo1[LineSelOut]);
Op=Op+iargs->maxwidth*4;
LineSelOut=LineSelOut^1;
}
dmaWaitTag(tgo0[LineSelOut^1]); //wait for last write.
dmaWaitTag(tgo1[LineSelOut^1]); //wait for last write.
// printf("Image done\n");
if (iargs->MessageForm==INTR)
{
while (spu_stat_out_intr_mbox() == 0);
msg=RDY;
spu_writech(SPU_WrOutIntrMbox, msg);
waiting=1;
}
if (iargs->MessageForm==HARD)
{
while (spu_stat_out_mbox() == 0);
msg=RDY;
spu_write_out_mbox(msg);
waiting=1;
}
// fprintf(stderr,"spu_yuvscaler: Waiting\n");
while (waiting){
while (spu_stat_in_mbox() == 0);
msg=spu_read_in_mbox();
if (msg == RUN){
selOut = selOut ^ 1; // flips the output buffer pointers
selIn = selIn ^ 1; // flips the input buffer pointers
waiting=0;
}
else if (msg == STOP)
{
// fprintf(stderr,"spu_yuvscaler: Stopping\n");
waiting=0;
}
else if (msg == UPDATE)
{
// fprintf(stderr,"spu_yuvscaler: Update\n");
dmaGetnWait(iargs,(unsigned int)argp,(int)envp,tag); //getting neccesary data to process the new image
first=1; // update filters to reflect the new image!
// selOut=0; // no need to change these. that can be done by the run.
// selIn=0;
}
}
}
return 0;
}
int main(unsigned long long speid, unsigned long long argp, unsigned long long envp)
{
int tgi0[2];
int tgo0[2];
int tgio0[2];
tgi0[0]=1;
tgi0[1]=2;
tgio0[0]=11;
tgio0[1]=12;
tgo0[0]=13;
tgo0[1]=14;
/* tgo1[0]=15;
tgo1[1]=16;*/
int selOut = 0;
int selIn = 0;
int msg=RUN;
int waiting=0;
int tag = 31;
struct img_args *iargs;
iargs =(struct img_args*)memalign(128,sizeof(*iargs));
dmaGetnWait(iargs,(unsigned int)argp,(int)envp,tag);
printf("spu_blit_yuv422_to_argb: SRC width %d,DST width %d\n",iargs->src_w,iargs->drw_w);
printf("spu_blit_yuv422_to_argb: SRC height %d,DST height %d\n",iargs->src_h,iargs->drw_h);
while (spu_stat_in_mbox() == 0);
msg=spu_read_in_mbox();
// first=0;
vector unsigned char *InOutbuffer[2];
vector unsigned char *Inbuffer[2];
vector unsigned char *Outbuffer[2];
int Outwidth=(4*iargs->drw_w+3)&~3;
int Inwidth=(2*iargs->src_w+7)&~7;
Inbuffer[0]=(vector unsigned char*)memalign(128,Inwidth);
Inbuffer[1]=(vector unsigned char*)memalign(128,Inwidth);
if (iargs->BLEND)
{
InOutbuffer[0]=(vector unsigned char*)memalign(128,Outwidth);
InOutbuffer[1]=(vector unsigned char*)memalign(128,Outwidth);
}
Outbuffer[0]=(vector unsigned char*)memalign(128,Outwidth);
Outbuffer[1]=(vector unsigned char*)memalign(128,Outwidth);
unsigned long long Inp,Outp,InOutp;
int i=0;
// int update=1;
while (msg!=STOP)
{
selOut = 0;
selIn = 0;
Inp=iargs->Inp0[0];
InOutp=iargs->Outp0[0];
Outp=iargs->Outp0[0];
dmaGet(Inbuffer[0],Inp,Inwidth,tgi0[0]);
Inp=Inp+iargs->Istride[0]*2;
dmaGet(Inbuffer[1],Inp,Inwidth,tgi0[1]);
Inp=Inp+iargs->Istride[0]*2;
// if (iargs->BLEND)
// {
// dmaGet(InOutbuffer[0],InOutp,Outwidth,tgio0[0]);
// InOutp=InOutp+iargs->Ostride[0]*4;
// dmaGet(InOutbuffer[1],InOutp,Outwidth,tgio0[1]);
// InOutp=InOutp+iargs->Ostride[0]*4;
// }
selIn=0;
selOut=0;
for (i=0;i < iargs->drw_h ;i++) {
dmaWaitTag(tgi0[selIn]);
// if (iargs->BLEND)
// dmaWaitTag(tgio0[selIn]);
dmaWaitTag(tgo0[selOut]);
if (iargs->SourceFormat==YUY2||iargs->SourceFormat==YUYV422)
{
yuv422_to_argb(Inbuffer[selIn],Outbuffer[selOut],iargs->drw_w);
// printf("spe_blitter: YUV422->ARGB\n");
}
//yuv420_to_yuv2(Yinbuffer[selIn],Uinbuffer[selIn],Vinbuffer[selIn],Outbuffer[selOut],iargs->Istride[0]);
// if (iargs->BLEND)
// blend(InOutbuffer[selIn],OutBuffer[selOut],iargs->ALPHA,iargs->SourceFormat);
dmaPut(Outbuffer[selOut],Outp,Outwidth,tgo0[selOut]);
// if (iargs->BLEND){
// dmaGet(InOutbuffer[selIn],InOutp,Outwidth,tgio0[selIn]);
// InOutp=InOutp+iargs->Ostride[0];
//
// }
dmaGet(Inbuffer[selIn],Inp,Inwidth,tgi0[selIn]);
Inp=Inp+iargs->Istride[0]*2;
Outp=Outp+iargs->Ostride[0]*4;
selIn=selIn^1;
selOut=selOut^1;
}
while (spu_stat_out_intr_mbox() == 0);
msg=RDY;
spu_writech(SPU_WrOutIntrMbox, msg);
waiting=1;
while (waiting){
while (spu_stat_in_mbox() == 0);
msg=spu_read_in_mbox();
if (msg == RUN){
waiting=0;
}
else if (msg == STOP)
{
waiting=0;
}
else if (msg == UPDATE)
{
tag=30;
dmaGetnWait(iargs,(unsigned int)argp,(int)envp,tag); //getting neccesary data to process the new image
// // update=1; // update filters to reflect the new image!
// Outwidth=(iargs->drw_w+3)&~3;
// Inwidth=(iargs->src_w+7)&~7;
// free(Inbuffer[0]);
// free(Inbuffer[1]);
//
// free(Outbuffer[0]);
// free(Outbuffer[1]);
//
// Inbuffer[0]=(vector unsigned char*)memalign(128,Inwidth);
// Inbuffer[1]=(vector unsigned char*)memalign(128,Inwidth);
//
// if (iargs->BLEND)
// {
// free(InOutbuffer[0]);
// free(InOutbuffer[1]);
// InOutbuffer[0]=(vector unsigned char*)memalign(128,Outwidth);
// InOutbuffer[1]=(vector unsigned char*)memalign(128,Outwidth);
// }
//
// Outbuffer[0]=(vector unsigned char*)memalign(128,Outwidth);
// Outbuffer[1]=(vector unsigned char*)memalign(128,Outwidth);
}
}
}
return 0;
}