github-mirrors/vortex
1
0
Fork 0
mirror of https://github.com/vortexgpgpu/vortex.git synced 2025-04-24 13:57:17 -04:00
Code Issues Projects Releases Packages Wiki Activity Actions

adding vx_mem_info runtime API

Browse source
This commit is contained in:
Blaise Tine 2023-03-12 20:10:52 -04:00
parent ad110b74d8
commit 5e2b8709bd
9 changed files with 113 additions and 35 deletions
Download patch file Download diff file Expand all files Collapse all files

24
runtime/common/vx_malloc.h
View file

@ -18,6 +18,7 @@ public:
, pageAlign_(pageAlign)
, blockAlign_(blockAlign)
, pages_(nullptr)
, allocated_(0)
{}
~MemoryAllocator() {
@ -98,6 +99,9 @@ public:
// Return the free block address
*addr = freeBlock->addr;
// Update allocated size
allocated_ += size;
return 0;
}
@ -133,6 +137,8 @@ public:
// Insert the block into the free M-list.
currPage->InsertFreeMBlock(usedBlock);
auto size = usedBlock->size;
// Check if we can merge adjacent free blocks from the left.
if (usedBlock->prevFreeM) {
// Calculate the previous address
@ -141,7 +147,7 @@ public:
auto prevBlock = usedBlock->prevFreeM;
// Merge the blocks to the left
prevBlock->size += usedBlock->size;
prevBlock->size += size;
prevBlock->nextFreeM = usedBlock->nextFreeM;
if (prevBlock->nextFreeM) {
prevBlock->nextFreeM->prevFreeM = prevBlock;
@ -159,7 +165,7 @@ public:
// Check if we can merge adjacent free blocks from the right.
if (usedBlock->nextFreeM) {
// Calculate the next allocation start address
auto nextAddr = usedBlock->addr + usedBlock->size;
auto nextAddr = usedBlock->addr + size;
if (usedBlock->nextFreeM->addr == nextAddr) {
auto nextBlock = usedBlock->nextFreeM;
@ -188,7 +194,14 @@ public:
}
return 0;
// update allocated size
allocated_ -= size;
return size;
}
uint64_t allocated() const {
return allocated_;
}
private:
@ -406,8 +419,9 @@ private:
uint64_t nextAddress_;
uint64_t maxAddress_;
uint32_t pageAlign_;
uint32_t blockAlign_;
page_t* pages_;
uint32_t blockAlign_;
page_t* pages_;
uint64_t allocated_;
};
} // namespace vortex

3
runtime/common/vx_utils.h
View file

@ -24,4 +24,5 @@ void perf_remove_device(vx_device_h device);
#define CACHE_BLOCK_SIZE 64
#define ALLOC_BASE_ADDR CACHE_BLOCK_SIZE
#define LOCAL_MEM_SIZE 4294967296 // 4 GB
#define ALLOC_MAX_ADDR 0x080000000 // 2 GB
#define LOCAL_MEM_SIZE 0x100000000 // 4 GB

8
runtime/include/vortex.h
View file

@ -20,9 +20,8 @@ typedef void* vx_buffer_h;
#define VX_CAPS_MAX_THREADS 0x3
#define VX_CAPS_CACHE_LINE_SIZE 0x4
#define VX_CAPS_LOCAL_MEM_SIZE 0x5
#define VX_CAPS_ALLOC_BASE_ADDR 0x6
#define VX_CAPS_KERNEL_BASE_ADDR 0x7
#define VX_CAPS_ISA_FLAGS 0x8
#define VX_CAPS_KERNEL_BASE_ADDR 0x6
#define VX_CAPS_ISA_FLAGS 0x7
// device isa flags
#define VX_ISA_STD_A (1ull << 0)
@ -68,6 +67,9 @@ int vx_mem_alloc(vx_device_h hdevice, uint64_t size, uint64_t* dev_maddr);
// release device memory
int vx_mem_free(vx_device_h hdevice, uint64_t dev_maddr);
// get device memory info
int vx_mem_info(vx_device_h hdevice, uint64_t* mem_free, uint64_t* mem_total);
// Copy bytes from buffer to device local memory
int vx_copy_to_dev(vx_buffer_h hbuffer, uint64_t dev_maddr, uint64_t size, uint64_t src_offset);

2
runtime/opae/common.h
View file

@ -28,7 +28,7 @@ public:
vx_device()
: mem_allocator(
ALLOC_BASE_ADDR,
ALLOC_BASE_ADDR + LOCAL_MEM_SIZE,
ALLOC_MAX_ADDR,
4096,
CACHE_BLOCK_SIZE)
{}

17
runtime/opae/vortex.cpp
View file

@ -65,9 +65,6 @@ extern int vx_dev_caps(vx_device_h hdevice, uint32_t caps_id, uint64_t *value) {
case VX_CAPS_LOCAL_MEM_SIZE:
*value = LOCAL_MEM_SIZE;
break;
case VX_CAPS_ALLOC_BASE_ADDR:
*value = ALLOC_BASE_ADDR;
break;
case VX_CAPS_KERNEL_BASE_ADDR:
*value = device->dcrs.read(DCR_BASE_STARTUP_ADDR);
break;
@ -252,6 +249,20 @@ extern int vx_mem_free(vx_device_h hdevice, uint64_t dev_maddr) {
return device->mem_allocator.release(dev_maddr);
}
extern int vx_mem_info(vx_device_h hdevice, uint64_t* mem_free, uint64_t* mem_total) {
if (nullptr == hdevice)
return -1;
auto device = ((vx_device*)hdevice);
if (mem_free) {
*mem_free = (ALLOC_MAX_ADDR - ALLOC_BASE_ADDR) - device->mem_allocator.allocated();
}
if (mem_total) {
*mem_total = (ALLOC_MAX_ADDR - ALLOC_BASE_ADDR);
}
return 0;
}
extern int vx_buf_alloc(vx_device_h hdevice, uint64_t size, vx_buffer_h* hbuffer) {
void* host_ptr;
uint64_t wsid;

31
runtime/rtlsim/vortex.cpp
View file

@ -69,7 +69,7 @@ public:
: ram_(RAM_PAGE_SIZE)
, mem_allocator_(
ALLOC_BASE_ADDR,
ALLOC_BASE_ADDR + LOCAL_MEM_SIZE,
ALLOC_MAX_ADDR,
RAM_PAGE_SIZE,
CACHE_BLOCK_SIZE)
{
@ -82,14 +82,18 @@ public:
}
}
int alloc_local_mem(uint64_t size, uint64_t* dev_maddr) {
int mem_alloc(uint64_t size, uint64_t* dev_maddr) {
return mem_allocator_.allocate(size, dev_maddr);
}
int free_local_mem(uint64_t dev_maddr) {
int mem_free(uint64_t dev_maddr) {
return mem_allocator_.release(dev_maddr);
}
uint64_t mem_used() const {
return mem_allocator_.allocated();
}
int upload(const void* src, uint64_t dest_addr, uint64_t size, uint64_t src_offset) {
uint64_t asize = aligned_size(size, CACHE_BLOCK_SIZE);
if (dest_addr + asize > LOCAL_MEM_SIZE)
@ -203,9 +207,6 @@ extern int vx_dev_caps(vx_device_h hdevice, uint32_t caps_id, uint64_t *value) {
case VX_CAPS_LOCAL_MEM_SIZE:
*value = LOCAL_MEM_SIZE;
break;
case VX_CAPS_ALLOC_BASE_ADDR:
*value = ALLOC_BASE_ADDR;
break;
case VX_CAPS_KERNEL_BASE_ADDR:
*value = device->read_dcr(DCR_BASE_STARTUP_ADDR);
break;
@ -266,7 +267,7 @@ extern int vx_mem_alloc(vx_device_h hdevice, uint64_t size, uint64_t* dev_maddr)
return -1;
vx_device *device = ((vx_device*)hdevice);
return device->alloc_local_mem(size, dev_maddr);
return device->mem_alloc(size, dev_maddr);
}
extern int vx_mem_free(vx_device_h hdevice, uint64_t dev_maddr) {
@ -274,7 +275,21 @@ extern int vx_mem_free(vx_device_h hdevice, uint64_t dev_maddr) {
return -1;
vx_device *device = ((vx_device*)hdevice);
return device->free_local_mem(dev_maddr);
return device->mem_free(dev_maddr);
}
extern int vx_mem_info(vx_device_h hdevice, uint64_t* mem_free, uint64_t* mem_total) {
if (nullptr == hdevice)
return -1;
auto device = ((vx_device*)hdevice);
if (mem_free) {
*mem_free = (ALLOC_MAX_ADDR - ALLOC_BASE_ADDR) - device->mem_used();
}
if (mem_total) {
*mem_total = (ALLOC_MAX_ADDR - ALLOC_BASE_ADDR);
}
return 0;
}
extern int vx_buf_alloc(vx_device_h hdevice, uint64_t size, vx_buffer_h* hbuffer) {

31
runtime/simx/vortex.cpp
View file

@ -79,7 +79,7 @@ public:
, processor_(arch_)
, mem_allocator_(
ALLOC_BASE_ADDR,
ALLOC_BASE_ADDR + LOCAL_MEM_SIZE,
ALLOC_MAX_ADDR,
RAM_PAGE_SIZE,
CACHE_BLOCK_SIZE)
{
@ -93,14 +93,18 @@ public:
}
}
int alloc_local_mem(uint64_t size, uint64_t* dev_maddr) {
int mem_alloc(uint64_t size, uint64_t* dev_maddr) {
return mem_allocator_.allocate(size, dev_maddr);
}
int free_local_mem(uint64_t dev_maddr) {
int mem_free(uint64_t dev_maddr) {
return mem_allocator_.release(dev_maddr);
}
uint64_t mem_used() const {
return mem_allocator_.allocated();
}
int upload(const void* src, uint64_t dest_addr, uint64_t size, uint64_t src_offset) {
uint64_t asize = aligned_size(size, CACHE_BLOCK_SIZE);
if (dest_addr + asize > LOCAL_MEM_SIZE)
@ -251,9 +255,6 @@ extern int vx_dev_caps(vx_device_h hdevice, uint32_t caps_id, uint64_t *value) {
case VX_CAPS_LOCAL_MEM_SIZE:
*value = LOCAL_MEM_SIZE;
break;
case VX_CAPS_ALLOC_BASE_ADDR:
*value = ALLOC_BASE_ADDR;
break;
case VX_CAPS_KERNEL_BASE_ADDR:
*value = device->read_dcr(DCR_BASE_STARTUP_ADDR);
break;
@ -276,7 +277,7 @@ extern int vx_mem_alloc(vx_device_h hdevice, uint64_t size, uint64_t* dev_maddr)
return -1;
vx_device *device = ((vx_device*)hdevice);
return device->alloc_local_mem(size, dev_maddr);
return device->mem_alloc(size, dev_maddr);
}
extern int vx_mem_free(vx_device_h hdevice, uint64_t dev_maddr) {
@ -284,7 +285,21 @@ extern int vx_mem_free(vx_device_h hdevice, uint64_t dev_maddr) {
return -1;
vx_device *device = ((vx_device*)hdevice);
return device->free_local_mem(dev_maddr);
return device->mem_free(dev_maddr);
}
extern int vx_mem_info(vx_device_h hdevice, uint64_t* mem_free, uint64_t* mem_total) {
if (nullptr == hdevice)
return -1;
auto device = ((vx_device*)hdevice);
if (mem_free) {
*mem_free = (ALLOC_MAX_ADDR - ALLOC_BASE_ADDR) - device->mem_used();
}
if (mem_total) {
*mem_total = (ALLOC_MAX_ADDR - ALLOC_BASE_ADDR);
}
return 0;
}
extern int vx_buf_alloc(vx_device_h hdevice, uint64_t size, vx_buffer_h* hbuffer) {

7
runtime/stub/vortex.cpp
View file

@ -16,10 +16,15 @@ extern int vx_mem_alloc(vx_device_h /*hdevice*/, uint64_t /*size*/, uint64_t* /*
return -1;
}
int vx_mem_free(vx_device_h /*hdevice*/, uint64_t /*dev_maddr*/) {
extern int vx_mem_free(vx_device_h /*hdevice*/, uint64_t /*dev_maddr*/) {
return -1;
}
// get device memory info
extern int vx_mem_info(vx_device_h /*hdevice*/, uint64_t* /*mem_free*/, uint64_t* /*mem_total*/) {
return 0;
}
extern int vx_buf_alloc(vx_device_h /*hdevice*/, uint64_t /*size*/, vx_buffer_h* /*hbuffer*/) {
return -1;
}

25
runtime/xrt/vortex.cpp
View file

@ -133,7 +133,7 @@ public:
, bank_size_(bank_size)
, mem_allocator_(
ALLOC_BASE_ADDR,
ALLOC_BASE_ADDR + LOCAL_MEM_SIZE,
ALLOC_MAX_ADDR,
4096,
CACHE_BLOCK_SIZE)
{}
@ -207,6 +207,10 @@ public:
return 0;
}
uint64_t mem_used() const {
return mem_allocator_.allocated();
}
int get_buffer(uint32_t bank_id, xrt_buffer_t* pBuf) {
auto it = xrtBuffers_.find(bank_id);
if (it != xrtBuffers_.end()) {
@ -324,9 +328,6 @@ extern int vx_dev_caps(vx_device_h hdevice, uint32_t caps_id, uint64_t *value) {
case VX_CAPS_LOCAL_MEM_SIZE:
*value = LOCAL_MEM_SIZE;
break;
case VX_CAPS_ALLOC_BASE_ADDR:
*value = ALLOC_BASE_ADDR;
break;
case VX_CAPS_KERNEL_BASE_ADDR:
*value = device->dcrs.read(DCR_BASE_STARTUP_ADDR);
break;
@ -559,7 +560,7 @@ extern int vx_mem_alloc(vx_device_h hdevice, uint64_t size, uint64_t* dev_maddr)
return device->mem_alloc(size, dev_maddr);
}
int vx_mem_free(vx_device_h hdevice, uint64_t dev_maddr) {
extern int vx_mem_free(vx_device_h hdevice, uint64_t dev_maddr) {
if (nullptr == hdevice)
return -1;
@ -567,6 +568,20 @@ int vx_mem_free(vx_device_h hdevice, uint64_t dev_maddr) {
return device->mem_free(dev_maddr);
}
extern int vx_mem_info(vx_device_h hdevice, uint64_t* mem_free, uint64_t* mem_total) {
if (nullptr == hdevice)
return -1;
auto device = (vx_device*)hdevice;
if (mem_free) {
*mem_free = (ALLOC_MAX_ADDR - ALLOC_BASE_ADDR) - device->mem_used();
}
if (mem_total) {
*mem_total = (ALLOC_MAX_ADDR - ALLOC_BASE_ADDR);
}
return 0;
}
extern int vx_buf_alloc(vx_device_h hdevice, uint64_t size, vx_buffer_h* hbuffer) {
if (nullptr == hdevice)
return -1;