packet_data* pd_alloc(uint16_t size, uint16_t block_size)
{
packet_data* pd = NULL;
packet_data* next = NULL;
while(size)
{
uint16_t alloc_size = size > block_size ? block_size : size;
if (next == NULL)
{
pd = bget(align_size(sizeof(packet_data) + alloc_size));
next = pd;
} else
{
next->next = bget(align_size(sizeof(packet_data) + alloc_size));
next = next->next;
}
if (next == NULL)
{
pd_free(pd);
return NULL;
}
pd->data = (uint8_t*)pd + sizeof(packet_data);
pd->next = NULL;
pd->size = alloc_size;
pd->total_size = size;
pd->refs_count = 1;
size -= alloc_size;
}
return pd;
}
static bool grow_buffer_if_needed(struct lwan_strbuf *s, size_t size)
{
if (s->flags & STATIC) {
const size_t aligned_size = align_size(max(size + 1, s->used));
if (UNLIKELY(!aligned_size))
return false;
char *buffer = malloc(aligned_size);
if (UNLIKELY(!buffer))
return false;
memcpy(buffer, s->value.static_buffer, s->used);
buffer[s->used + 1] = '\0';
s->flags &= ~STATIC;
s->value.buffer = buffer;
return true;
}
if (UNLIKELY(!s->used || lwan_nextpow2(s->used) < size)) {
const size_t aligned_size = align_size(size + 1);
if (UNLIKELY(!aligned_size))
return false;
char *buffer = realloc(s->value.buffer, aligned_size);
if (UNLIKELY(!buffer))
return false;
s->value.buffer = buffer;
}
return true;
}
void*
ora_srealloc(void *ptr, size_t size)
{
void *result;
size_t aux_s = 0;
int i;
for (i = 0; i < *list_c; i++)
if (list[i].first_byte_ptr == ptr)
{
if (align_size(size) <= list[i].size)
return ptr;
aux_s = list[i].size;
}
if (aux_s == 0)
ereport(ERROR,
(errcode(ERRCODE_INTERNAL_ERROR),
errmsg("corrupted pointer"),
errdetail("Failed while reallocating memory block in shared memory."),
errhint("Report this bug to autors.")));
if (NULL != (result = ora_salloc(size)))
{
memcpy(result, ptr, aux_s);
ora_sfree(ptr);
}
return result;
}
void* pmem_alloc(size_t size, int *pfd)
{
int fd;
size_t aligned_size;
void* base;
pmem_region region = { 0, 0 };
int err;
if (NULL == pfd)
{
return NULL;
}
*pfd = -1;
fd = open(PMEM_DEVICE_NAME, O_RDWR);
if (INVALID_FD == fd)
{
LOGE("[PMEM] open %s failed!", PMEM_DEVICE_NAME);
goto open_failed;
}
aligned_size = align_size(size);
base = mmap(0, aligned_size, PROT_READ|PROT_WRITE, MAP_SHARED, fd, 0);
if (MAP_FAILED == base)
{
LOGE("[PMEM] mmap size %d failed!", aligned_size);
goto mmap_failed;
}
region.len = aligned_size;
err = ioctl(fd, PMEM_MAP, ®ion);
if (IOCTL_FAILED == err)
{
LOGE("[PMEM] PMEM_MAP size %d failed!", aligned_size);
goto pmem_map_failed;
}
property_get("pm.dumpstack", p_value, "0"); //if not set, disable by default
p_res = atoi(p_value);
if (p_res)
{
LOGE("[PMEM] pmem_alloc: base: 0x%08x, size: %d\n", (int)base, aligned_size);
dump_stack_trace();
}
*pfd = fd;
return base;
insert_failed:
ioctl(fd, PMEM_UNMAP, ®ion);
pmem_map_failed:
munmap(base, aligned_size);
mmap_failed:
close(fd);
open_failed:
return NULL;
}
void Init_Monitor(void)
{
firm_top = (int) &g_pfnVectors;
firm_end = (int) &_etext + (&_edata - &_sdata);
firm_end = align_size(firm_end,1024);
test_adr0= 0;
}
int ofmem_posix_memalign( void **memptr, size_t alignment, size_t size )
{
ofmem_t *ofmem = ofmem_arch_get_private();
alloc_desc_t *d, **pp;
void *ret;
ucell top;
phys_addr_t pa;
if( !size )
return ENOMEM;
if( !ofmem->next_malloc )
ofmem->next_malloc = (char*)ofmem_arch_get_malloc_base();
size = align_size(size + sizeof(alloc_desc_t), alignment);
/* look in the freelist */
for( pp=&ofmem->mfree; *pp && (**pp).size < size; pp = &(**pp).next ) {
}
/* waste at most 4K by taking an entry from the freelist */
if( *pp && (**pp).size > size + 0x1000 ) {
/* Alignment should be on physical not virtual address */
pa = va2pa((uintptr_t)*pp + sizeof(alloc_desc_t));
pa = align_ptr(pa, alignment);
ret = (void *)pa2va(pa);
memset( ret, 0, (**pp).size - sizeof(alloc_desc_t) );
*pp = (**pp).next;
*memptr = ret;
return 0;
}
top = ofmem_arch_get_heap_top();
/* Alignment should be on physical not virtual address */
pa = va2pa((uintptr_t)ofmem->next_malloc + sizeof(alloc_desc_t));
pa = align_ptr(pa, alignment);
ret = (void *)pa2va(pa);
if( pointer2cell(ret) + size > top ) {
printk("out of malloc memory (%x)!\n", size );
return ENOMEM;
}
d = (alloc_desc_t*)((uintptr_t)ret - sizeof(alloc_desc_t));
ofmem->next_malloc += size;
d->next = NULL;
d->size = size;
memset( ret, 0, size - sizeof(alloc_desc_t) );
*memptr = ret;
return 0;
}
void Init_Monitor(void)
{
#if 0
firm_top = (int) &g_pfnVectors;
firm_end = (int) &_etext + (&_edata - &_sdata);
#endif
firm_top = (int) &__cs3_stm32_vector_table;
firm_end = (int) &_etext + (&_edata - &__cs3_region_start_ram);
firm_end = align_size(firm_end,1024);
test_adr0= 0;
}
bool space_try_alloc(space_t *space, size_t size, address_t *memory_out) {
CHECK_FALSE("allocating in empty space", space_is_empty(space));
size_t aligned = align_size(kValueSize, size);
address_t addr = space->next_free;
address_t next = addr + aligned;
if (next <= space->limit) {
// Clear the newly allocated memory to a different value, again to make the
// contents recognizable.
memset(addr, kAllocedHeapMarker, aligned);
*memory_out = addr;
space->next_free = next;
return true;
} else {
return false;
}
}
void *valloc(size_t size)
{
size_t s = alloc_size(size);
void *d = VirtualAlloc(0, s, MEM_RESERVE, PAGE_NOACCESS);
if (!d || !VirtualAlloc(d, s - PAGE_SIZE, MEM_COMMIT, PAGE_READWRITE)) {
Debug("valloc error(%x %d %d)\n", d, s, size);
return NULL;
}
((DWORD *)d)[0] = VALLOC_SIG;
((size_t *)d)[1] = size;
Debug("valloc (%x %d %d)\n", d, s, size);
return (void *)((u_char *)d + s - PAGE_SIZE - align_size(size, ALLOC_ALIGN));
}
void GrVkMemory::FreeImageMemory(const GrVkGpu* gpu, bool linearTiling,
const GrVkAlloc& alloc) {
GrVkHeap* heap;
if (linearTiling) {
heap = gpu->getHeap(GrVkGpu::kLinearImage_Heap);
} else if (alloc.fSize <= kMaxSmallImageSize) {
heap = gpu->getHeap(GrVkGpu::kSmallOptimalImage_Heap);
} else {
heap = gpu->getHeap(GrVkGpu::kOptimalImage_Heap);
}
if (!heap->free(alloc)) {
// must be an adopted allocation
GR_VK_CALL(gpu->vkInterface(), FreeMemory(gpu->device(), alloc.fMemory, nullptr));
} else {
gTotalImageMemory -= alloc.fSize;
VkDeviceSize pageAlignedSize = align_size(alloc.fSize, kMinVulkanPageSize);
gTotalImageMemoryFullPage -= pageAlignedSize;
}
}
packet_data* data_block_alloc(uint16_t size)
{
if (size == 0)
{
return NULL;
}
packet_data* block = bget(align_size(sizeof(packet_data) + size));
if (block == NULL)
{
return NULL;
}
block->data = (uint8_t*)block + sizeof(packet_data);
block->next = NULL;
block->size = size;
block->total_size = size;
return block;
}
laser_packet* lp_alloc(packet_data* data)
{
if (data == NULL)
{
return NULL;
}
uint16_t aligned_size = align_size(sizeof(laser_packet));
laser_packet* lp = bgetz(aligned_size);
if (lp == NULL)
{
return NULL;
}
lp->data = data;
lp->data_size = data->total_size;
return lp;
}
bg_shm_t * bg_shm_alloc_read(int id, int size)
{
void * addr;
bg_shm_t * ret = NULL;
int shm_fd;
char name[SHM_NAME_MAX];
int real_size = get_real_size(size);
gen_name(id, name);
shm_fd = shm_open(name, O_RDWR, 0);
if(shm_fd < 0)
{
bg_log(BG_LOG_ERROR, LOG_DOMAIN,
"shm_open of %s failed: %s", name, strerror(errno));
goto fail;
}
if((addr = mmap(0, real_size, PROT_READ | PROT_WRITE, MAP_SHARED,
shm_fd, 0)) == MAP_FAILED)
{
bg_log(BG_LOG_ERROR, LOG_DOMAIN,
"mmap failed: %s", strerror(errno));
goto fail;
}
ret = calloc(1, sizeof(*ret));
ret->addr = addr;
ret->size = size;
ret->rc = (refcounter_t*)(ret->addr + align_size(size));
ret->id = id;
bg_log(BG_LOG_DEBUG, LOG_DOMAIN,
"created shm segment (read) %s", name);
fail:
if(shm_fd >= 0)
close(shm_fd);
return ret;
}
int pmem_free(void *ptr, size_t size, int fd)
{
int err, ret = 0;
size_t aligned_size = align_size(size);
pmem_region region = { 0, aligned_size };
err = ioctl(fd, PMEM_UNMAP, ®ion);
if (IOCTL_FAILED == err)
{
LOGE("PMEM_UNMAP size %d failed!", size);
ret = err;
}
err = munmap(ptr, aligned_size);
if (UNMAP_FAILED == err)
{
LOGE("mumap size %d failed!", size);
ret = err;
}
err = close(fd);
if (CLOSE_FAILED == err)
{
LOGE("Close file %d failed!", fd);
ret = err;
}
property_get("pm.dumpstack", p_value, "0"); //if not set, disable by default
p_res = atoi(p_value);
if (p_res)
{
LOGE("[PMEM] pmem_free: base: 0x%08x, size: %d\n", (int)ptr, aligned_size);
dump_stack_trace();
}
return ret;
}
t_bool buffer_reserve(t_buffer *buf, size_t avail)
{
char *tmp;
size_t target_len;
size_t target_size;
target_len = buf->len + avail;
if (target_len + 1 <= buf->_size)
return (true);
else
{
target_size = align_size(target_len + 1, BUFFER_BASE_SIZE);
tmp = malloc(target_size);
if (tmp == NULL)
return (false);
ft_strcpy(tmp, buf->str);
free(buf->str);
buf->str = tmp;
buf->_size = target_size;
return (true);
}
}
/**
* Create a memory pool descriptor.
* 1. If required, set the allocation size to a power of 2 of the element size.
* 2. Save the given parameters in the pool descriptor, to be used by
* pool_alloc();
* 3. Chain the pool descriptor to the given pool_list, so it can be
* automatically freed.
*/
Pool_desc *pool_new(const char *func, const char *name,
size_t num_elements, size_t element_size,
bool zero_out, bool align, bool exact)
{
Pool_desc *mp = malloc(sizeof(Pool_desc));
mp->func = func;
mp->name = name;
if (align)
{
mp->element_size = align_size(element_size);
mp->alignment = MAX(MIN_ALIGNMENT, mp->element_size);
mp->alignment = MIN(MAX_ALIGNMENT, mp->alignment);
mp->data_size = num_elements * mp->element_size;
mp->block_size = ALIGN(mp->data_size + FLDSIZE_NEXT, mp->alignment);
}
else
{
mp->element_size = element_size;
mp->alignment = MIN_ALIGNMENT;
mp->data_size = num_elements * mp->element_size;
mp->block_size = mp->data_size + FLDSIZE_NEXT;
}
mp->zero_out = zero_out;
mp->exact = exact;
mp->alloc_next = NULL;
mp->chain = NULL;
mp->ring = NULL;
mp->free_list = NULL;
mp->curr_elements = 0;
mp->num_elements = num_elements;
lgdebug(+D_MEMPOOL, "%sElement size %zu, alignment %zu (pool '%s' created in %s())\n",
POOL_ALLOCATOR?"":"(Fake pool allocator) ",
mp->element_size, mp->alignment, mp->name, mp->func);
return mp;
}
bool GrVkHeap::singleAlloc(VkDeviceSize size, VkDeviceSize alignment,
uint32_t memoryTypeIndex, GrVkAlloc* alloc) {
VkDeviceSize alignedSize = align_size(size, alignment);
// first try to find an unallocated subheap that fits our allocation request
int bestFitIndex = -1;
VkDeviceSize bestFitSize = 0x7FFFFFFF;
for (auto i = 0; i < fSubHeaps.count(); ++i) {
if (fSubHeaps[i]->memoryTypeIndex() == memoryTypeIndex && fSubHeaps[i]->unallocated()) {
VkDeviceSize heapSize = fSubHeaps[i]->size();
if (heapSize >= alignedSize && heapSize < bestFitSize) {
bestFitIndex = i;
bestFitSize = heapSize;
}
}
}
if (bestFitIndex >= 0) {
SkASSERT(fSubHeaps[bestFitIndex]->alignment() == alignment);
if (fSubHeaps[bestFitIndex]->alloc(size, alloc)) {
fUsedSize += alloc->fSize;
return true;
}
return false;
}
// need to allocate a new subheap
SkAutoTDelete<GrVkSubHeap>& subHeap = fSubHeaps.push_back();
subHeap.reset(new GrVkSubHeap(fGpu, memoryTypeIndex, alignedSize, alignment));
fAllocSize += alignedSize;
if (subHeap->alloc(size, alloc)) {
fUsedSize += alloc->fSize;
return true;
}
return false;
}
bg_shm_t * bg_shm_alloc_write(int size)
{
int shm_fd = -1;
void * addr;
bg_shm_t * ret = NULL;
char name[SHM_NAME_MAX];
int id = 0;
pthread_mutexattr_t attr;
int real_size = get_real_size(size);
while(1)
{
id++;
gen_name(id, name);
if((shm_fd = shm_open(name, O_RDWR | O_CREAT | O_EXCL,
S_IRUSR | S_IWUSR | S_IRGRP | S_IROTH)) < 0)
{
if(errno != EEXIST)
{
bg_log(BG_LOG_ERROR, LOG_DOMAIN,
"shm_open of %s failed: %s", name, strerror(errno));
return NULL;
}
}
else
break;
}
if(ftruncate(shm_fd, real_size))
{
bg_log(BG_LOG_ERROR, LOG_DOMAIN,
"ftruncate failed: %s", strerror(errno));
goto fail;
}
if((addr = mmap(0, real_size, PROT_READ | PROT_WRITE,
MAP_SHARED, shm_fd, 0)) == MAP_FAILED)
{
bg_log(BG_LOG_ERROR, LOG_DOMAIN,
"mmap failed: %s", strerror(errno));
return NULL;
}
ret = calloc(1, sizeof(*ret));
ret->addr = addr;
ret->size = size;
ret->id = id;
ret->wr = 1;
ret->rc = (refcounter_t*)(ret->addr + align_size(size));
/* Initialize process shared mutex */
pthread_mutexattr_init(&attr);
if(pthread_mutexattr_setpshared(&attr, PTHREAD_PROCESS_SHARED))
{
bg_log(BG_LOG_ERROR, LOG_DOMAIN,
"cannot create process shared mutex: %s", strerror(errno));
goto fail;
}
pthread_mutex_init(&ret->rc->mutex, &attr);
bg_log(BG_LOG_DEBUG, LOG_DOMAIN,
"created shm segment (write) %s", name);
ret->rc->refcount = 0;
fail:
close(shm_fd);
return ret;
}
void Save::save(const Image & in, const Image_Io_Frame_Info & frame)
throw (Error)
{
//DJV_DEBUG("Save::save");
//DJV_DEBUG_PRINT("in = " << in);
// Open the file.
_open(_file.get(frame.frame));
// Convert the image.
const Pixel_Data * p = ∈
if (in.info() != _info)
{
//DJV_DEBUG_PRINT("convert = " << _image);
_image.zero();
Gl_Image::copy(in, _image);
p = &_image;
}
// Write the file.
const int w = p->w(), h = p->h();
const int channels = Pixel::channels(p->info().pixel);
const int channel_bytes = Pixel::channel_bytes(p->info().pixel);
const int bytes = Pixel::bytes(p->info().pixel);
bool compress = _options.compression ? true : false;
uint32_t length = 0;
//DJV_DEBUG_PRINT("channels = " << channels);
//DJV_DEBUG_PRINT("channel_bytes = " << channel_bytes);
//DJV_DEBUG_PRINT("bytes = " << bytes);
size_t pos = 0;
pos = _io.position();
//DJV_DEBUG_PRINT("position = " << static_cast<int>(pos));
// 'FOR4' type
_io.set_u8('F');
_io.set_u8('O');
_io.set_u8('R');
_io.set_u8('4');
// 'FOR4' length
// NOTE: only reserved for now.
_io.set_u32(length);
// 'TBMP' type
_io.set_u8('T');
_io.set_u8('B');
_io.set_u8('M');
_io.set_u8('P');
// Write tiles.
V2i size = tile_size (w, h);
// Y order.
for (int y = 0; y < size.y; y++)
{
// X order.
for (int x = 0; x < size.x; x++)
{
// Set tile coordinates.
uint16_t xmin, xmax, ymin, ymax;
// Set xmin and xmax.
xmin = x * tile_width();
xmax = Math::min(xmin + tile_width(), w) - 1;
// Set ymin and ymax.
ymin = y * tile_height();
ymax = Math::min(ymin + tile_height(), h) - 1;
// Set width and height.
uint32_t tw = xmax - xmin + 1;
uint32_t th = ymax - ymin + 1;
// Set type.
_io.set_u8('R');
_io.set_u8('G');
_io.set_u8('B');
_io.set_u8('A');
// Length.
uint32_t length = tw * th * bytes;
// Tile length.
uint32_t tile_length = length;
// Align.
length = align_size(length, 4);
// Append xmin, xmax, ymin and ymax.
length += 8;
// Tile compression.
bool tile_compress = compress;
// Set bytes.
Memory_Buffer<uint8_t> tile(tile_length);
uint8_t * out_p = tile();
// Handle 8-bit data.
if (p->info().pixel == Pixel::RGB_U8 ||
p->info().pixel == Pixel::RGBA_U8)
{
if (tile_compress)
{
uint32_t index = 0, size = 0;
// Set bytes.
// NOTE: prevent buffer overrun.
Memory_Buffer<uint8_t> tmp(tile_length * 2);
// Map: RGB(A)8 RGBA to BGRA
for (int c = (channels * channel_bytes) - 1; c >= 0; --c)
{
Memory_Buffer<uint8_t> in(tw * th);
uint8_t * in_p = in();
// Data.
for (uint16_t py = ymin; py <= ymax; py++)
{
const uint8_t * in_dy = p->data(0, py);
for (uint16_t px = xmin; px <= xmax; px++)
{
// Get pixel.
uint8_t pixel;
const uint8_t * in_dx = in_dy + px * bytes + c;
Memory::copy(in_dx, &pixel, 1);
// Set pixel.
*in_p++ = pixel;
}
}
// Compress
size = rle_save(in(), tmp() + index, tw * th);
index += size;
}
// If size exceeds tile length use uncompressed.
if (index < tile_length)
{
Memory::copy(tmp(), tile(), index);
// Set tile length.
tile_length = index;
// Append xmin, xmax, ymin and ymax.
length = index + 8;
// Set length.
uint32_t align = align_size(length, 4);
if (align > length)
{
out_p = tile() + index;
// Pad.
for (int i = 0;
i < static_cast<int>(align - length);
i++)
{
*out_p++ = '\0';
tile_length++;
}
}
}
else
{
tile_compress = false;
}
}
if (!tile_compress)
{
for (uint16_t py = ymin; py <= ymax; py++)
{
const uint8_t * in_dy = p->data(0, py);
for (uint16_t px = xmin; px <= xmax; px++)
{
// Map: RGB(A)8 RGBA to BGRA
for (int c = channels - 1; c >= 0; --c)
{
// Get pixel.
uint8_t pixel;
const uint8_t * in_dx =
in_dy + px * bytes + c * channel_bytes;
Memory::copy(in_dx, &pixel, 1);
// Set pixel.
*out_p++ = pixel;
}
}
}
}
}
// Handle 16-bit data.
else if (
p->info().pixel == Pixel::RGB_U16 ||
p->info().pixel == Pixel::RGBA_U16)
{
if (tile_compress)
{
uint32_t index = 0, size = 0;
// Set bytes.
// NOTE: prevent buffer overrun.
Memory_Buffer<uint8_t> tmp(tile_length * 2);
// Set map.
int* map = NULL;
if (Memory::endian () == Memory::LSB)
{
int rgb16[] = { 0, 2, 4, 1, 3, 5 };
int rgba16[] = { 0, 2, 4, 7, 1, 3, 5, 6 };
if (_info.pixel == Pixel::RGB_U16)
{
map = rgb16;
}
else
{
map = rgba16;
}
}
else
{
int rgb16[] = { 1, 3, 5, 0, 2, 4 };
int rgba16[] = { 1, 3, 5, 7, 0, 2, 4, 6 };
if (_info.pixel == Pixel::RGB_U16)
{
map = rgb16;
}
else
{
map = rgba16;
}
}
// Map: RGB(A)16 RGBA to BGRA
for (int c = (channels * channel_bytes) - 1; c >= 0; --c)
{
int mc = map[c];
Memory_Buffer<uint8_t> in(tw * th);
uint8_t * in_p = in();
// Data.
for (uint16_t py = ymin; py <= ymax; py++)
{
const uint8_t * in_dy = p->data(0, py);
for (uint16_t px = xmin; px <= xmax; px++)
{
// Get pixel.
uint8_t pixel;
const uint8_t * in_dx = in_dy + px * bytes + mc;
Memory::copy(in_dx, &pixel, 1);
// Set pixel.
*in_p++ = pixel;
}
}
// Compress
size = rle_save(in(), tmp() + index, tw * th);
index += size;
}
// If size exceeds tile length use uncompressed.
if (index < tile_length)
{
Memory::copy(tmp(), tile(), index);
// Set tile length.
tile_length = index;
// Append xmin, xmax, ymin and ymax.
length = index + 8;
// Set length.
uint32_t align = align_size(length, 4);
if (align > length)
{
out_p = tile() + index;
// Pad.
for (
int i = 0;
i < static_cast<int>(align - length);
i++)
{
*out_p++ = '\0';
tile_length++;
}
}
}
else
{
tile_compress = false;
}
}
if (!tile_compress)
{
for (uint16_t py = ymin; py <= ymax; py++)
{
const uint8_t * in_dy = p->data(0, py);
for (uint16_t px = xmin; px <= xmax; px++)
{
// Map: RGB(A)16 RGBA to BGRA
for (int c = channels - 1; c >= 0; --c)
{
uint16_t pixel;
const uint8_t * in_dx =
in_dy + px * bytes + c * channel_bytes;
if (Memory::endian () == Memory::LSB)
{
Memory::endian(in_dx, &pixel, 1, 2);
}
else
{
Memory::copy(in_dx, &pixel, 2);
}
// Set pixel.
*out_p++ = pixel;
out_p++;
}
}
}
}
}
// Set length.
_io.set_u32(length);
// Set xmin, xmax, ymin and ymax.
_io.set_u16(xmin);
_io.set_u16(ymin);
_io.set_u16(xmax);
_io.set_u16(ymax);
// Write.
_io.set(tile(), tile_length);
}
}
// Set FOR4 CIMG and FOR4 TBMP size
uint32_t p0 = _io.position() - 8;
uint32_t p1 = p0 - pos;
// NOTE: FOR4 <size> CIMG
_io.position (4);
_io.set_u32 (p0);
// NOTE: FOR4 <size> TBMP
_io.position (pos + 4);
_io.set_u32 (p1);
_io.close();
}
void*
ora_salloc(size_t size)
{
size_t aligned_size;
int repeat_c;
void *ptr = NULL;
aligned_size = align_size(size);
for (repeat_c = 0; repeat_c < 2; repeat_c++)
{
size_t max_min = max_size;
int select = -1;
int i;
/* find first good free block */
for (i = 0; i < *list_c; i++)
{
if (list[i].dispossible)
{
/* If this block is just the right size, return it */
if (list[i].size == aligned_size)
{
list[i].dispossible = false;
ptr = list[i].first_byte_ptr;
/* list[i].context = context; */
return ptr;
}
if (list[i].size > aligned_size && list[i].size < max_min)
{
max_min = list[i].size;
select = i;
}
}
}
/* If no suitable free slot found, defragment and try again. */
if (select == -1 || *list_c == LIST_ITEMS)
{
defragmentation();
continue;
}
/*
* A slot larger than required was found. Divide it to avoid wasting
* space, and return the slot of the right size.
*/
list[*list_c].size = list[select].size - aligned_size;
list[*list_c].first_byte_ptr = (char*)list[select].first_byte_ptr + aligned_size;
list[*list_c].dispossible = true;
list[select].size = aligned_size;
list[select].dispossible = false;
/* list[select].context = context; */
ptr = list[select].first_byte_ptr;
*list_c += 1;
break;
}
return ptr;
}
void display(void) {
struct timeval time1, time2;
double total_time, rate, pixel_rate;
const unsigned pixel_size = formats[cur_format].num_comp * types[cur_type].byte_size / types[cur_type].num_comp;
const unsigned buf_size = align_size(screen_width * pixel_size, ALIGNMENT) * screen_height;
unsigned char* buf = (unsigned char*) malloc(buf_size);
assert(buf);
if (stabilizing_rounds) {
printf("Stabilizing round %i\n", stabilizing_rounds);
}
glPixelStorei(GL_PACK_ALIGNMENT, ALIGNMENT);
gettimeofday(&time1, NULL);
for (int i = 0; i < NUM_READBACKS; ++i) {
glReadPixels(0, 0, screen_width, screen_height, formats[cur_format].format, types[cur_type].type, buf);
}
gettimeofday(&time2, NULL);
total_time = (time2.tv_sec + 1e-6 * time2.tv_usec - time1.tv_sec - 1e-6 * time1.tv_usec);
rate = 1e-6 * NUM_READBACKS * buf_size / total_time;
pixel_rate = 1e-6 * NUM_READBACKS * screen_width * screen_height / total_time;
if (stabilizing_rounds) {
--stabilizing_rounds;
} else {
results[cur_type * NUM_FORMATS + cur_format].type = cur_type;
results[cur_type * NUM_FORMATS + cur_format].format = cur_format;
results[cur_type * NUM_FORMATS + cur_format].rate = rate;
results[cur_type * NUM_FORMATS + cur_format].pixel_rate = pixel_rate;
printf("%s %s %g MB/s %g Mp/s\n",
formats[cur_format].desc, types[cur_type].desc, rate, pixel_rate);
/* Find the next type that is compatible with the current format */
do {
++cur_type;
} while (!compatible(formats[cur_format], types[cur_type]) && cur_type < NUM_TYPES);
/* Go to the next type/format */
if (cur_type == NUM_TYPES) {
cur_type = 0;
if (++cur_format == NUM_FORMATS) {
printf("\nSorted by data rate:\n");
qsort(results, NUM_FORMATS * NUM_TYPES, sizeof(result_t), compare_data_rate);
for (int i = 0; i < NUM_FORMATS * NUM_TYPES; ++i) {
if (results[i].rate >= 0)
printf("%s %s %g MB/s\n",
formats[results[i].format].desc,
types[results[i].type].desc,
results[i].rate);
}
printf("\nSorted by pixel rate:\n");
qsort(results, NUM_FORMATS * NUM_TYPES, sizeof(result_t), compare_pixel_rate);
for (int i = 0; i < NUM_FORMATS * NUM_TYPES; ++i) {
if (results[i].rate >= 0)
printf("%s %s %g Mp/s\n",
formats[results[i].format].desc,
types[results[i].type].desc,
results[i].pixel_rate);
}
exit(0);
}
}
}
free(buf);
glutReportErrors();
glutPostRedisplay();
}
struct rte_hash *
rte_hash_create(const struct rte_hash_parameters *params)
{
struct rte_hash *h = NULL;
uint32_t num_buckets, sig_bucket_size, key_size,
hash_tbl_size, sig_tbl_size, key_tbl_size, mem_size;
char hash_name[RTE_HASH_NAMESIZE];
struct rte_hash_list *hash_list;
/* check that we have an initialised tail queue */
if ((hash_list =
RTE_TAILQ_LOOKUP_BY_IDX(RTE_TAILQ_HASH, rte_hash_list)) == NULL) {
rte_errno = E_RTE_NO_TAILQ;
return NULL;
}
/* Check for valid parameters */
if ((params == NULL) ||
(params->entries > RTE_HASH_ENTRIES_MAX) ||
(params->bucket_entries > RTE_HASH_BUCKET_ENTRIES_MAX) ||
(params->entries < params->bucket_entries) ||
!rte_is_power_of_2(params->entries) ||
!rte_is_power_of_2(params->bucket_entries) ||
(params->key_len == 0) ||
(params->key_len > RTE_HASH_KEY_LENGTH_MAX)) {
rte_errno = EINVAL;
RTE_LOG(ERR, HASH, "rte_hash_create has invalid parameters\n");
return NULL;
}
rte_snprintf(hash_name, sizeof(hash_name), "HT_%s", params->name);
/* Calculate hash dimensions */
num_buckets = params->entries / params->bucket_entries;
sig_bucket_size = align_size(params->bucket_entries *
sizeof(hash_sig_t), SIG_BUCKET_ALIGNMENT);
key_size = align_size(params->key_len, KEY_ALIGNMENT);
hash_tbl_size = align_size(sizeof(struct rte_hash), CACHE_LINE_SIZE);
sig_tbl_size = align_size(num_buckets * sig_bucket_size,
CACHE_LINE_SIZE);
key_tbl_size = align_size(num_buckets * key_size *
params->bucket_entries, CACHE_LINE_SIZE);
/* Total memory required for hash context */
mem_size = hash_tbl_size + sig_tbl_size + key_tbl_size;
rte_rwlock_write_lock(RTE_EAL_TAILQ_RWLOCK);
/* guarantee there's no existing */
TAILQ_FOREACH(h, hash_list, next) {
if (strncmp(params->name, h->name, RTE_HASH_NAMESIZE) == 0)
break;
}
if (h != NULL)
goto exit;
h = (struct rte_hash *)rte_zmalloc_socket(hash_name, mem_size,
CACHE_LINE_SIZE, params->socket_id);
if (h == NULL) {
RTE_LOG(ERR, HASH, "memory allocation failed\n");
goto exit;
}
/* Setup hash context */
rte_snprintf(h->name, sizeof(h->name), "%s", params->name);
h->entries = params->entries;
h->bucket_entries = params->bucket_entries;
h->key_len = params->key_len;
h->hash_func_init_val = params->hash_func_init_val;
h->num_buckets = num_buckets;
h->bucket_bitmask = h->num_buckets - 1;
h->sig_msb = 1 << (sizeof(hash_sig_t) * 8 - 1);
h->sig_tbl = (uint8_t *)h + hash_tbl_size;
h->sig_tbl_bucket_size = sig_bucket_size;
h->key_tbl = h->sig_tbl + sig_tbl_size;
h->key_tbl_key_size = key_size;
h->hash_func = (params->hash_func == NULL) ?
DEFAULT_HASH_FUNC : params->hash_func;
TAILQ_INSERT_TAIL(hash_list, h, next);
exit:
rte_rwlock_write_unlock(RTE_EAL_TAILQ_RWLOCK);
return h;
}
inline size_t alloc_size(size_t size) {
return align_size((align_size(size, ALLOC_ALIGN) + 16 + PAGE_SIZE), PAGE_SIZE);
}
bool GrVkFreeListAlloc::alloc(VkDeviceSize requestedSize,
VkDeviceSize* allocOffset, VkDeviceSize* allocSize) {
VkDeviceSize alignedSize = align_size(requestedSize, fAlignment);
// find the smallest block big enough for our allocation
FreeList::Iter iter = fFreeList.headIter();
FreeList::Iter bestFitIter;
VkDeviceSize bestFitSize = fSize + 1;
VkDeviceSize secondLargestSize = 0;
VkDeviceSize secondLargestOffset = 0;
while (iter.get()) {
Block* block = iter.get();
// need to adjust size to match desired alignment
SkASSERT(align_size(block->fOffset, fAlignment) - block->fOffset == 0);
if (block->fSize >= alignedSize && block->fSize < bestFitSize) {
bestFitIter = iter;
bestFitSize = block->fSize;
}
if (secondLargestSize < block->fSize && block->fOffset != fLargestBlockOffset) {
secondLargestSize = block->fSize;
secondLargestOffset = block->fOffset;
}
iter.next();
}
SkASSERT(secondLargestSize <= fLargestBlockSize);
Block* bestFit = bestFitIter.get();
if (bestFit) {
SkASSERT(align_size(bestFit->fOffset, fAlignment) == bestFit->fOffset);
*allocOffset = bestFit->fOffset;
*allocSize = alignedSize;
// adjust or remove current block
VkDeviceSize originalBestFitOffset = bestFit->fOffset;
if (bestFit->fSize > alignedSize) {
bestFit->fOffset += alignedSize;
bestFit->fSize -= alignedSize;
if (fLargestBlockOffset == originalBestFitOffset) {
if (bestFit->fSize >= secondLargestSize) {
fLargestBlockSize = bestFit->fSize;
fLargestBlockOffset = bestFit->fOffset;
} else {
fLargestBlockSize = secondLargestSize;
fLargestBlockOffset = secondLargestOffset;
}
}
#ifdef SK_DEBUG
VkDeviceSize largestSize = 0;
iter = fFreeList.headIter();
while (iter.get()) {
Block* block = iter.get();
if (largestSize < block->fSize) {
largestSize = block->fSize;
}
iter.next();
}
SkASSERT(largestSize == fLargestBlockSize);
#endif
} else {
SkASSERT(bestFit->fSize == alignedSize);
if (fLargestBlockOffset == originalBestFitOffset) {
fLargestBlockSize = secondLargestSize;
fLargestBlockOffset = secondLargestOffset;
}
fFreeList.remove(bestFit);
#ifdef SK_DEBUG
VkDeviceSize largestSize = 0;
iter = fFreeList.headIter();
while (iter.get()) {
Block* block = iter.get();
if (largestSize < block->fSize) {
largestSize = block->fSize;
}
iter.next();
}
SkASSERT(largestSize == fLargestBlockSize);
#endif
}
fFreeSize -= alignedSize;
SkASSERT(*allocSize > 0);
return true;
}
SkDebugf("Can't allocate %d bytes, %d bytes available, largest free block %d\n", alignedSize, fFreeSize, fLargestBlockSize);
return false;
}
static int get_real_size(int size)
{
return align_size(size) + sizeof(refcounter_t);
}
bool GrVkMemory::AllocAndBindImageMemory(const GrVkGpu* gpu,
VkImage image,
bool linearTiling,
GrVkAlloc* alloc) {
const GrVkInterface* iface = gpu->vkInterface();
VkDevice device = gpu->device();
VkMemoryRequirements memReqs;
GR_VK_CALL(iface, GetImageMemoryRequirements(device, image, &memReqs));
uint32_t typeIndex = 0;
GrVkHeap* heap;
if (linearTiling) {
VkMemoryPropertyFlags desiredMemProps = VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT |
VK_MEMORY_PROPERTY_HOST_COHERENT_BIT |
VK_MEMORY_PROPERTY_HOST_CACHED_BIT;
if (!get_valid_memory_type_index(gpu->physicalDeviceMemoryProperties(),
memReqs.memoryTypeBits,
desiredMemProps,
&typeIndex)) {
// this memory type should always be available
SkASSERT_RELEASE(get_valid_memory_type_index(gpu->physicalDeviceMemoryProperties(),
memReqs.memoryTypeBits,
VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT |
VK_MEMORY_PROPERTY_HOST_COHERENT_BIT,
&typeIndex));
}
heap = gpu->getHeap(GrVkGpu::kLinearImage_Heap);
} else {
// this memory type should always be available
SkASSERT_RELEASE(get_valid_memory_type_index(gpu->physicalDeviceMemoryProperties(),
memReqs.memoryTypeBits,
VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT,
&typeIndex));
if (memReqs.size <= kMaxSmallImageSize) {
heap = gpu->getHeap(GrVkGpu::kSmallOptimalImage_Heap);
} else {
heap = gpu->getHeap(GrVkGpu::kOptimalImage_Heap);
}
}
if (!heap->alloc(memReqs.size, memReqs.alignment, typeIndex, alloc)) {
SkDebugf("Failed to alloc image\n");
return false;
}
// Bind Memory to device
VkResult err = GR_VK_CALL(iface, BindImageMemory(device, image,
alloc->fMemory, alloc->fOffset));
if (err) {
SkASSERT_RELEASE(heap->free(*alloc));
return false;
}
gTotalImageMemory += alloc->fSize;
VkDeviceSize pageAlignedSize = align_size(alloc->fSize, kMinVulkanPageSize);
gTotalImageMemoryFullPage += pageAlignedSize;
return true;
}
bool GrVkHeap::subAlloc(VkDeviceSize size, VkDeviceSize alignment,
uint32_t memoryTypeIndex, GrVkAlloc* alloc) {
VkDeviceSize alignedSize = align_size(size, alignment);
// if requested is larger than our subheap allocation, just alloc directly
if (alignedSize > fSubHeapSize) {
VkMemoryAllocateInfo allocInfo = {
VK_STRUCTURE_TYPE_MEMORY_ALLOCATE_INFO, // sType
NULL, // pNext
size, // allocationSize
memoryTypeIndex, // memoryTypeIndex
};
VkResult err = GR_VK_CALL(fGpu->vkInterface(), AllocateMemory(fGpu->device(),
&allocInfo,
nullptr,
&alloc->fMemory));
if (VK_SUCCESS != err) {
return false;
}
alloc->fOffset = 0;
alloc->fSize = 0; // hint that this is not a subheap allocation
return true;
}
// first try to find a subheap that fits our allocation request
int bestFitIndex = -1;
VkDeviceSize bestFitSize = 0x7FFFFFFF;
for (auto i = 0; i < fSubHeaps.count(); ++i) {
if (fSubHeaps[i]->memoryTypeIndex() == memoryTypeIndex) {
VkDeviceSize heapSize = fSubHeaps[i]->largestBlockSize();
if (heapSize >= alignedSize && heapSize < bestFitSize) {
bestFitIndex = i;
bestFitSize = heapSize;
}
}
}
if (bestFitIndex >= 0) {
SkASSERT(fSubHeaps[bestFitIndex]->alignment() == alignment);
if (fSubHeaps[bestFitIndex]->alloc(size, alloc)) {
fUsedSize += alloc->fSize;
return true;
}
return false;
}
// need to allocate a new subheap
SkAutoTDelete<GrVkSubHeap>& subHeap = fSubHeaps.push_back();
subHeap.reset(new GrVkSubHeap(fGpu, memoryTypeIndex, fSubHeapSize, alignment));
// try to recover from failed allocation by only allocating what we need
if (subHeap->size() == 0) {
VkDeviceSize alignedSize = align_size(size, alignment);
subHeap.reset(new GrVkSubHeap(fGpu, memoryTypeIndex, alignedSize, alignment));
if (subHeap->size() == 0) {
return false;
}
}
fAllocSize += fSubHeapSize;
if (subHeap->alloc(size, alloc)) {
fUsedSize += alloc->fSize;
return true;
}
return false;
}
