def test_vk_staging_allocator():
if not hasattr(ncnn, "get_gpu_count"):
return
vkdev = ncnn.get_gpu_device(0)
assert vkdev is not None
allocator = ncnn.VkStagingAllocator(vkdev)
assert allocator.buffer_memory_type_index >= 0
assert allocator.image_memory_type_index >= 0
mappable = allocator.mappable
allocator.mappable = not mappable
assert allocator.mappable == (not mappable)
coherent = allocator.coherent
allocator.coherent = not coherent
assert allocator.coherent == (not coherent)
bufmem = allocator.fastMalloc(10 * 1024)
assert bufmem is not None
allocator.fastFree(bufmem)
imgmem = allocator.fastMalloc(4, 4, 3, 4, 1)
assert imgmem is not None
allocator.fastFree(imgmem)
def test_vulkan_device_vkdev():
if not hasattr(ncnn, "get_gpu_count"):
return
net = ncnn.Net()
assert net.vulkan_device() is None
vkdev = ncnn.get_gpu_device(0)
net.set_vulkan_device(vkdev)
assert net.vulkan_device() is not None
def test_gpu_api():
if not hasattr(ncnn, "get_gpu_count"):
return
assert ncnn.create_gpu_instance() == 0
assert ncnn.get_gpu_count() > 0
assert ncnn.get_default_gpu_index() >= 0
gpuinfo = ncnn.get_gpu_info(0)
check_gpuinfo(gpuinfo)
vkdev = ncnn.get_gpu_device(0)
assert vkdev is not None
gpuinfo = vkdev.info()
check_gpuinfo(gpuinfo)
ncnn.destroy_gpu_instance()
if argc >= 6:
cooling_down = int(sys.argv[5])
use_vulkan_compute = gpu_device != -1
g_enable_cooling_down = cooling_down != 0
g_loop_count = loop_count
g_blob_pool_allocator.set_size_compare_ratio(0.0)
g_workspace_pool_allocator.set_size_compare_ratio(0.5)
if use_vulkan_compute:
g_warmup_loop_count = 10
g_vkdev = ncnn.get_gpu_device(gpu_device)
g_blob_vkallocator = ncnn.VkBlobAllocator(g_vkdev)
g_staging_vkallocator = ncnn.VkStagingAllocator(g_vkdev)
opt = ncnn.Option()
opt.lightmode = True
opt.num_threads = num_threads
opt.blob_allocator = g_blob_pool_allocator
opt.workspace_allocator = g_workspace_pool_allocator
if use_vulkan_compute:
opt.blob_vkallocator = g_blob_vkallocator
opt.workspace_vkallocator = g_blob_vkallocator
opt.staging_vkallocator = g_staging_vkallocator
opt.use_winograd_convolution = True
opt.use_sgemm_convolution = True