diff --git a/src/correlation/shaders/cross_check_filter.comp.glsl b/src/correlation/shaders/cross_check_filter.comp.glsl index 8f8fca5..4eb2803 100644 --- a/src/correlation/shaders/cross_check_filter.comp.glsl +++ b/src/correlation/shaders/cross_check_filter.comp.glsl @@ -38,4 +38,5 @@ void main() { // TODO: remove this debug code img1[0] = ivec2(img1_width, img1_height); + img1[1] = ivec2(img2_width, img2_height); } diff --git a/src/correlation/shaders/init_out_data.comp.glsl b/src/correlation/shaders/init_out_data.comp.glsl index 138da8d..91d53b0 100644 --- a/src/correlation/shaders/init_out_data.comp.glsl +++ b/src/correlation/shaders/init_out_data.comp.glsl @@ -45,7 +45,7 @@ layout(std430, set = 0, binding = 3) buffer Internals_Int { // Layout: // Contains [min, max, neighbor_count] for the corridor range - int internals_int[]; + ivec3 internals_int[]; }; layout(std430, set = 0, binding = 4) buffer Result_Matches { @@ -69,4 +69,5 @@ void main() { // TODO: remove this debug code result_corr[0] = threshold; + result_corr[1] = fundamental_matrix[2][0]; } diff --git a/src/correlation/shaders/init_out_data.spv b/src/correlation/shaders/init_out_data.spv index 0ed0416..d028237 100644 Binary files a/src/correlation/shaders/init_out_data.spv and b/src/correlation/shaders/init_out_data.spv differ diff --git a/src/correlation/vk.rs b/src/correlation/vk.rs index ed81ce7..caea656 100644 --- a/src/correlation/vk.rs +++ b/src/correlation/vk.rs @@ -1,4 +1,4 @@ -use std::{collections::HashMap, error, ffi::CStr, fmt, slice, time::SystemTime}; +use std::{cmp::Ordering, collections::HashMap, error, ffi::CStr, fmt, slice, time::SystemTime}; use ash::{prelude::VkResult, vk}; use nalgebra::Matrix3; @@ -169,6 +169,9 @@ impl GpuContext { } let correlation_values = Grid::new(img1_dimensions.0, img1_dimensions.1, None); + // TODO: remove this debug code + let mut fundamental_matrix = fundamental_matrix.clone(); + fundamental_matrix[(2, 0)] = 9.8765; let params = CorrelationParameters::for_projection(&projection_mode); let result = GpuContext { min_stdev: params.min_stdev, @@ -477,11 +480,13 @@ impl Device { instance.destroy_instance(None); err }; + println!("Created all buffers"); // Init pipelines and shaders. let descriptor_sets = unsafe { Device::create_descriptor_sets(&device, &buffers, CorrelationDirection::Forward) .map_err(cleanup_err)? }; + println!("Created descriptor sets"); let cleanup_err = |err| unsafe { descriptor_sets.destroy(&device); buffers.destroy(&device); @@ -491,6 +496,7 @@ impl Device { }; let pipelines = unsafe { Device::create_pipelines(&device, &descriptor_sets).map_err(cleanup_err)? }; + println!("Created pipelines"); let cleanup_err = |err| unsafe { destroy_pipelines(&device, &pipelines); descriptor_sets.destroy(&device); @@ -502,6 +508,7 @@ impl Device { // Init control struct - queues, fences, command buffer. let control = unsafe { Device::create_control(&device, compute_queue_index).map_err(cleanup_err)? }; + println!("Created control"); let result = Device { _entry: entry, instance, @@ -813,7 +820,11 @@ impl Device { }); } // TODO: remove this debug code - println!("Corr check = {:?}", out_image.val(0, 0)); + println!( + "Corr check = {:?} {:?}", + out_image.val(0, 0), + out_image.val(1, 0) + ); if !buffer.host_coherent { let flush_memory_ranges = vk::MappedMemoryRange::builder() @@ -906,7 +917,7 @@ impl Device { unsafe fn find_device( instance: &ash::Instance, max_buffer_size: usize, - ) -> Result<(vk::PhysicalDevice, &'static str, u32), Box> { + ) -> Result<(vk::PhysicalDevice, String, u32), Box> { let devices = instance.enumerate_physical_devices()?; let device = devices .iter() @@ -923,16 +934,7 @@ impl Device { let queue_index = Device::find_compute_queue(instance, device)?; let device_name = CStr::from_ptr(props.device_name.as_ptr()); - let device_name = device_name.to_str().unwrap(); - println!( - "Device {} type {} {}-{}-{}-{}", - device_name, - props.device_type.as_raw(), - props.limits.max_push_constants_size, - props.limits.max_bound_descriptor_sets, - props.limits.max_storage_buffer_range, - max_buffer_size - ); + let device_name = String::from_utf8_lossy(device_name.to_bytes()).to_string(); // TODO: allow to specify a device name filter/regex? let score = match props.device_type { vk::PhysicalDeviceType::DISCRETE_GPU => 3, @@ -941,24 +943,25 @@ impl Device { _ => 0, }; // Prefer real devices instead of dzn emulation. - let dzn_multiplier = if device_name + let is_dzn = device_name .to_lowercase() - .starts_with("microsoft direct3d12") - { - 1 - } else { - 10 - }; - Some((device, device_name, queue_index, score * dzn_multiplier)) + .starts_with("microsoft direct3d12"); + let score = (score, is_dzn); + Some((device, device_name, queue_index, score)) }) - .max_by_key(|(_device, _name, _queue_index, score)| *score); - let (device, name, queue_index) = if let Some((device, name, queue_index, _score)) = device - { + .max_by(|(_, _, _, a), (_, _, _, b)| { + if a.1 && !b.1 { + return Ordering::Less; + } else if !a.1 && b.1 { + return Ordering::Greater; + } + return a.0.cmp(&b.0); + }); + let (device, name, queue_index) = if let Some((device, name, queue_index, score)) = device { (device, name, queue_index) } else { return Err(GpuError::new("Device not found").into()); }; - println!("selected device {}", name); Ok((device, name, queue_index)) } @@ -1009,6 +1012,7 @@ impl Device { let max_pixels = img1_pixels.max(img2_pixels); let mut buffers: Vec = vec![]; let cleanup_err = |buffers: &[Buffer], err| { + println!("buffers count is {}", buffers.len()); buffers.iter().for_each(|buffer| { device.free_memory(buffer.buffer_memory, None); device.destroy_buffer(buffer.buffer, None) @@ -1095,13 +1099,13 @@ impl Device { buffer_type: BufferType, ) -> Result> { let size = size as u64; - let gpu_local = match buffer_type { - BufferType::GpuOnly | BufferType::GpuDestination | BufferType::GpuSource => true, - BufferType::HostSource | BufferType::HostDestination => false, - }; - let host_visible = match buffer_type { - BufferType::HostSource | BufferType::HostDestination => true, - BufferType::GpuOnly | BufferType::GpuDestination | BufferType::GpuSource => false, + let required_memory_properties = match buffer_type { + BufferType::GpuOnly | BufferType::GpuDestination | BufferType::GpuSource => { + vk::MemoryPropertyFlags::DEVICE_LOCAL + } + BufferType::HostSource | BufferType::HostDestination => { + vk::MemoryPropertyFlags::HOST_VISIBLE + } }; let extra_usage_flags = match buffer_type { BufferType::HostSource => vk::BufferUsageFlags::TRANSFER_SRC, @@ -1122,58 +1126,41 @@ impl Device { }; let buffer = device.create_buffer(&buffer_create_info, None)?; let memory_requirements = device.get_buffer_memory_requirements(buffer); - let memory_type_index = memory_properties.memory_types - [..memory_properties.memory_type_count as usize] - .iter() - .enumerate() - .find(|(memory_type_index, memory_type)| { + let buffer_memory = (0..memory_properties.memory_type_count as usize) + .flat_map(|i| { + let memory_type = memory_properties.memory_types[i]; if memory_properties.memory_heaps[memory_type.heap_index as usize].size < memory_requirements.size { - return false; - }; - if (1 << memory_type_index) & memory_requirements.memory_type_bits == 0 { - return false; + return None; } - - if gpu_local - && memory_type - .property_flags - .contains(vk::MemoryPropertyFlags::DEVICE_LOCAL) - { - return true; + if ((1 << i) & memory_requirements.memory_type_bits) == 0 { + return None; } - if host_visible - && memory_type - .property_flags - .contains(vk::MemoryPropertyFlags::HOST_VISIBLE) - { - return true; + let property_flags = memory_type.property_flags; + if !property_flags.contains(required_memory_properties) { + return None; } - false - }); - let memory_type_index = if let Some((index, _)) = memory_type_index { - index as u32 + let host_visible = property_flags.contains(vk::MemoryPropertyFlags::HOST_VISIBLE); + let host_coherent = property_flags.contains(vk::MemoryPropertyFlags::HOST_COHERENT); + let allocate_info = vk::MemoryAllocateInfo { + allocation_size: memory_requirements.size, + memory_type_index: i as u32, + ..Default::default() + }; + // Some buffers may fill up, in this case allocating memory can fail. + let mem = device.allocate_memory(&allocate_info, None).ok()?; + + Some((mem, host_visible, host_coherent)) + }) + .next(); + + let (buffer_memory, host_visible, host_coherent) = if let Some(mem) = buffer_memory { + mem } else { + device.destroy_buffer(buffer, None); return Err(GpuError::new("Cannot find suitable memory").into()); }; - let property_flags = - memory_properties.memory_types[memory_type_index as usize].property_flags; - let host_visible = property_flags.contains(vk::MemoryPropertyFlags::HOST_VISIBLE); - let host_coherent = property_flags.contains(vk::MemoryPropertyFlags::HOST_COHERENT); - let allocate_info = vk::MemoryAllocateInfo { - allocation_size: memory_requirements.size, - memory_type_index, - ..Default::default() - }; - let buffer_memory = device.allocate_memory(&allocate_info, None); - let buffer_memory = match buffer_memory { - Ok(mem) => mem, - Err(err) => { - device.destroy_buffer(buffer, None); - return Err(err.into()); - } - }; let result = Buffer { buffer, buffer_memory, @@ -1205,6 +1192,7 @@ impl Device { vk::DescriptorSetLayoutCreateInfo::builder().bindings(bindings.as_slice()); device.create_descriptor_set_layout(&layout_info, None) }; + println!("creating descriptor pools 1"); let descriptor_pool_size = [vk::DescriptorPoolSize::builder() .ty(vk::DescriptorType::STORAGE_BUFFER) .descriptor_count(2) @@ -1212,19 +1200,23 @@ impl Device { let descriptor_pool_info = vk::DescriptorPoolCreateInfo::builder() .max_sets(1) .pool_sizes(&descriptor_pool_size); + println!("creating descriptor pools 2"); let descriptor_pool = device.create_descriptor_pool(&descriptor_pool_info, None)?; let cleanup_err = |err| { device.destroy_descriptor_pool(descriptor_pool, None); err }; let regular_layout = create_layout_bindings(6).map_err(cleanup_err)?; + println!("creating descriptor pools 3"); let cleanup_err = |err| { device.destroy_descriptor_set_layout(regular_layout, None); device.destroy_descriptor_pool(descriptor_pool, None); err }; let cross_check_layout = create_layout_bindings(2).map_err(cleanup_err)?; + println!("creating descriptor pools 4"); let cleanup_err = |err| { + println!("failed to allocate descriptor set"); device.destroy_descriptor_set_layout(cross_check_layout, None); device.destroy_descriptor_set_layout(regular_layout, None); device.destroy_descriptor_pool(descriptor_pool, None); @@ -1398,6 +1390,7 @@ impl Device { .create_fence(&fence_create_info, None) .map_err(cleanup_err)?; let cleanup_err = |err| { + println!("Failed to alloc command buffer"); device.destroy_command_pool(command_pool, None); device.destroy_fence(fence, None); err