vbx_void_t *vbx_sp_malloc_debug( int LINE,const char *FNAME, size_t num_bytes )
{
// print pretty error messages
vbx_mxp_t *this_mxp = VBX_GET_THIS_MXP();
if( !this_mxp || !this_mxp->init ) {
VBX_PRINTF( "ERROR: failed to call _vbx_init().\n" );
VBX_FATAL(LINE,FNAME,-1);
}
// pad to scratchpad width to reduce occurrence of false hazards
size_t padded = VBX_PAD_UP( num_bytes, this_mxp->scratchpad_alignment_bytes );
size_t freesp = (size_t)(this_mxp->scratchpad_end - this_mxp->sp); //VBX_SCRATCHPAD_END - (size_t)vbx_sp; // vbx_sp_getfree();
vbx_void_t *result = NULL;
if( VBX_DEBUG_LEVEL && (num_bytes==0) ) {
print_sp_malloc_null();
} else if( VBX_DEBUG_LEVEL && freesp < padded ) {
print_sp_malloc_full( num_bytes, padded );
} else if( num_bytes > 0 && freesp >= padded ) {
result = this_mxp->sp;
this_mxp->sp += padded;
#if VBX_DEBUG_SP_MALLOC
printf("sp_malloc %d bytes padded to %d, sp=0x%08x\n", num_bytes, padded, this_mxp->sp);
#endif
}
if( !result ) {
VBX_FATAL(LINE,FNAME,-1);
}
return result;
}
vbx_void_t *vbx_sp_malloc_nodebug( size_t num_bytes )
{
if( VBX_DEBUG_LEVEL && 0 ) {
// print pretty error messages
return vbx_sp_malloc_debug( __LINE__, __FILE__, num_bytes );
}
// do it, but do not print pretty error messages
vbx_mxp_t *this_mxp = VBX_GET_THIS_MXP();
// check for valid argument values
if( !this_mxp || num_bytes==0 )
return NULL;
// add padding and allocate
// pad to scratchpad width to reduce occurrence of false hazards
size_t padded = VBX_PAD_UP( num_bytes, this_mxp->scratchpad_alignment_bytes );
vbx_void_t *old_sp = this_mxp->sp;
this_mxp->sp += padded;
// scratchpad full
if( this_mxp->sp > this_mxp->scratchpad_end ) {
this_mxp->sp = old_sp;
return NULL;
}
// success
return old_sp;
}
void *vbx_shared_alloca_nodebug( size_t num_bytes, void *p )
{
void *alloced_ptr;
void *aligned_ptr = NULL;
unsigned int padding = VBX_PADDING();
alloced_ptr = (void *)p;
if( alloced_ptr ) {
aligned_ptr = (void *)VBX_PAD_UP( alloced_ptr, padding );
aligned_ptr = (void *)vbx_remap_uncached_flush( aligned_ptr, num_bytes );
}
return aligned_ptr;
}
void *vbx_shared_malloc( size_t num_bytes )
{
void *alloced_ptr;
void *aligned_ptr = NULL;
unsigned int padding = VBX_PADDING();
#if VBX_DEBUG_MALLOC
printf("shared_malloc %d bytes\n", num_bytes);
#endif
alloced_ptr = (void *)vbx_uncached_malloc(num_bytes+sizeof(void*)+2*padding);
if( alloced_ptr ) {
aligned_ptr = (void *)VBX_PAD_UP( (alloced_ptr+sizeof(void*)), padding );
*((void **)(aligned_ptr-sizeof(void*))) = alloced_ptr;
}
return aligned_ptr;
}
//vector version of rgb converter
void vector_blend(
output_pointer img_out, input_pointer img_in1, input_pointer img_in2,
unsigned int num_row, unsigned int num_column, intermediate_type blending_const )
{
intermediate_type *v_img1[2];
input_type *v_img2[2];
intermediate_type *v_temp;
intermediate_type blending_const_bar = 256-blending_const;
int j;
vbx_mxp_t *this_mxp = VBX_GET_THIS_MXP();
const int VBX_SCRATCHPAD_SIZE = this_mxp->scratchpad_size;
const int VBX_WIDTH_BYTES = this_mxp->vector_lanes * sizeof(int);
const int VBX_DMA_ALIGNMENT = this_mxp->dma_alignment_bytes;
unsigned int chunk_size = VBX_SCRATCHPAD_SIZE/((3*sizeof(intermediate_type))+(2*sizeof(input_type)));
chunk_size = VBX_PAD_UP( chunk_size-(VBX_WIDTH_BYTES-1), VBX_DMA_ALIGNMENT );
unsigned int chunk_size_old = chunk_size;
unsigned int vector_length = chunk_size;
unsigned int vector_length_old = vector_length;
v_img1[0] = (intermediate_type *)vbx_sp_malloc( chunk_size*sizeof(intermediate_type) );
v_img1[1] = (intermediate_type *)vbx_sp_malloc( chunk_size*sizeof(intermediate_type) );
v_img2[0] = (input_type *)vbx_sp_malloc( chunk_size*sizeof(input_type) );
v_img2[1] = (input_type *)vbx_sp_malloc( chunk_size*sizeof(input_type) );
v_temp = (intermediate_type *)vbx_sp_malloc( chunk_size*sizeof(intermediate_type) );
if( v_temp == NULL ) {
VBX_EXIT(0xBADDEAD);
}
int bufselect = 0;
vbx_dma_to_vector( v_img1[bufselect], img_in1, chunk_size*sizeof(input_type) );
vbx_dma_to_vector( v_img2[bufselect], img_in2, chunk_size*sizeof(input_type) );
for( j=0; j<num_row*num_column; j+=vector_length_old ) {
vbx_set_vl(vector_length);
if( j > 0 ) {
vbx_dma_to_host( img_out+j-vector_length_old, v_img1[1-bufselect], chunk_size_old*sizeof(output_type) );
}
if( (j+vector_length_old) < (num_row*num_column-1) ) {
if( (j+vector_length_old*2) >= num_row*num_column ) {
vector_length = num_row*num_column - j - vector_length_old;
chunk_size = vector_length;
}
vbx_dma_to_vector( v_img1[1-bufselect], img_in1+j+vector_length_old, chunk_size*sizeof(input_type) );
vbx_dma_to_vector( v_img2[1-bufselect], img_in2+j+vector_length_old, chunk_size*sizeof(input_type) );
}
vbx( SVBHU, VMULLO, v_temp, blending_const, v_img1[bufselect] );
vbx( SVBHU, VMULLO, v_img1[bufselect], blending_const_bar, v_img2[bufselect] );
vbx( VVHU, VADD, v_img1[bufselect], v_img1[bufselect], v_temp );
vbx( SVHBU, VSHR, v_img1[bufselect], 8, v_img1[bufselect] );
bufselect = 1-bufselect;
}
vbx_dma_to_host( img_out+j-vector_length_old, v_img1[1-bufselect], chunk_size*sizeof(output_type) );
vbx_sp_free();
vbx_sync();
}
