From a30b7c0158fe744678ca9e4c53b3f343d9f387ff Mon Sep 17 00:00:00 2001 From: Lynne Date: Tue, 16 Jul 2024 17:16:34 +0200 Subject: [PATCH] hwcontext_vulkan: initialize optical flow queues if available Lets us implement FPS conversion. --- libavutil/hwcontext_vulkan.c | 18 +++++++++++++++--- libavutil/vulkan.c | 6 +++++- libavutil/vulkan.h | 1 + libavutil/vulkan_functions.h | 8 ++++++++ libavutil/vulkan_loader.h | 1 + 5 files changed, 30 insertions(+), 4 deletions(-) diff --git a/libavutil/hwcontext_vulkan.c b/libavutil/hwcontext_vulkan.c index 5baf68660a..c81fc95af2 100644 --- a/libavutil/hwcontext_vulkan.c +++ b/libavutil/hwcontext_vulkan.c @@ -105,6 +105,7 @@ typedef struct VulkanDevicePriv { VkPhysicalDeviceDescriptorBufferFeaturesEXT desc_buf_features; VkPhysicalDeviceShaderAtomicFloatFeaturesEXT atomic_float_features; VkPhysicalDeviceCooperativeMatrixFeaturesKHR coop_matrix_features; + VkPhysicalDeviceOpticalFlowFeaturesNV optical_flow_features; /* Queues */ pthread_mutex_t **qf_mutex; @@ -429,6 +430,7 @@ static const VulkanOptExtension optional_device_exts[] = { { VK_EXT_PHYSICAL_DEVICE_DRM_EXTENSION_NAME, FF_VK_EXT_DEVICE_DRM }, { VK_EXT_SHADER_ATOMIC_FLOAT_EXTENSION_NAME, FF_VK_EXT_ATOMIC_FLOAT }, { VK_KHR_COOPERATIVE_MATRIX_EXTENSION_NAME, FF_VK_EXT_COOP_MATRIX }, + { VK_NV_OPTICAL_FLOW_EXTENSION_NAME, FF_VK_EXT_OPTICAL_FLOW }, /* Imports/exports */ { VK_KHR_EXTERNAL_MEMORY_FD_EXTENSION_NAME, FF_VK_EXT_EXTERNAL_FD_MEMORY }, @@ -1127,13 +1129,14 @@ static int setup_queue_families(AVHWDeviceContext *ctx, VkDeviceCreateInfo *cd) av_log(ctx, AV_LOG_VERBOSE, "Queue families:\n"); for (int i = 0; i < num; i++) { - av_log(ctx, AV_LOG_VERBOSE, " %i:%s%s%s%s%s%s%s (queues: %i)\n", i, + av_log(ctx, AV_LOG_VERBOSE, " %i:%s%s%s%s%s%s%s%s (queues: %i)\n", i, ((qf[i].queueFamilyProperties.queueFlags) & VK_QUEUE_GRAPHICS_BIT) ? " graphics" : "", ((qf[i].queueFamilyProperties.queueFlags) & VK_QUEUE_COMPUTE_BIT) ? " compute" : "", ((qf[i].queueFamilyProperties.queueFlags) & VK_QUEUE_TRANSFER_BIT) ? " transfer" : "", ((qf[i].queueFamilyProperties.queueFlags) & VK_QUEUE_VIDEO_ENCODE_BIT_KHR) ? " encode" : "", ((qf[i].queueFamilyProperties.queueFlags) & VK_QUEUE_VIDEO_DECODE_BIT_KHR) ? " decode" : "", ((qf[i].queueFamilyProperties.queueFlags) & VK_QUEUE_SPARSE_BINDING_BIT) ? " sparse" : "", + ((qf[i].queueFamilyProperties.queueFlags) & VK_QUEUE_OPTICAL_FLOW_BIT_NV) ? " optical_flow" : "", ((qf[i].queueFamilyProperties.queueFlags) & VK_QUEUE_PROTECTED_BIT) ? " protected" : "", qf[i].queueFamilyProperties.queueCount); @@ -1177,6 +1180,7 @@ static int setup_queue_families(AVHWDeviceContext *ctx, VkDeviceCreateInfo *cd) PICK_QF(VK_QUEUE_GRAPHICS_BIT, VK_VIDEO_CODEC_OPERATION_NONE_KHR); PICK_QF(VK_QUEUE_COMPUTE_BIT, VK_VIDEO_CODEC_OPERATION_NONE_KHR); PICK_QF(VK_QUEUE_TRANSFER_BIT, VK_VIDEO_CODEC_OPERATION_NONE_KHR); + PICK_QF(VK_QUEUE_OPTICAL_FLOW_BIT_NV, VK_VIDEO_CODEC_OPERATION_NONE_KHR); PICK_QF(VK_QUEUE_VIDEO_ENCODE_BIT_KHR, VK_VIDEO_CODEC_OPERATION_ENCODE_H264_BIT_KHR); PICK_QF(VK_QUEUE_VIDEO_DECODE_BIT_KHR, VK_VIDEO_CODEC_OPERATION_DECODE_H264_BIT_KHR); @@ -1318,9 +1322,13 @@ static int vulkan_device_create_internal(AVHWDeviceContext *ctx, VkPhysicalDeviceTimelineSemaphoreFeatures timeline_features = { .sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_TIMELINE_SEMAPHORE_FEATURES, }; + VkPhysicalDeviceOpticalFlowFeaturesNV optical_flow_features = { + .sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_OPTICAL_FLOW_FEATURES_NV, + .pNext = &timeline_features, + }; VkPhysicalDeviceCooperativeMatrixFeaturesKHR coop_matrix_features = { .sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_COOPERATIVE_MATRIX_FEATURES_KHR, - .pNext = &timeline_features, + .pNext = &optical_flow_features, }; VkPhysicalDeviceShaderAtomicFloatFeaturesEXT atomic_float_features = { .sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SHADER_ATOMIC_FLOAT_FEATURES_EXT, @@ -1364,7 +1372,9 @@ static int vulkan_device_create_internal(AVHWDeviceContext *ctx, p->atomic_float_features.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SHADER_ATOMIC_FLOAT_FEATURES_EXT; p->atomic_float_features.pNext = &p->coop_matrix_features; p->coop_matrix_features.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_COOPERATIVE_MATRIX_FEATURES_KHR; - p->coop_matrix_features.pNext = NULL; + p->coop_matrix_features.pNext = &p->optical_flow_features; + p->optical_flow_features.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_OPTICAL_FLOW_FEATURES_NV; + p->optical_flow_features.pNext = NULL; ctx->free = vulkan_device_free; @@ -1428,6 +1438,8 @@ static int vulkan_device_create_internal(AVHWDeviceContext *ctx, p->coop_matrix_features.cooperativeMatrix = coop_matrix_features.cooperativeMatrix; + p->optical_flow_features.opticalFlow = optical_flow_features.opticalFlow; + dev_info.pNext = &hwctx->device_features; /* Setup queue family */ diff --git a/libavutil/vulkan.c b/libavutil/vulkan.c index cec8354ba6..ade8d482b9 100644 --- a/libavutil/vulkan.c +++ b/libavutil/vulkan.c @@ -90,9 +90,13 @@ int ff_vk_load_props(FFVulkanContext *s) s->hprops = (VkPhysicalDeviceExternalMemoryHostPropertiesEXT) { .sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_EXTERNAL_MEMORY_HOST_PROPERTIES_EXT, }; + s->optical_flow_props = (VkPhysicalDeviceOpticalFlowPropertiesNV) { + .sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_OPTICAL_FLOW_PROPERTIES_NV, + .pNext = &s->hprops, + }; s->coop_matrix_props = (VkPhysicalDeviceCooperativeMatrixPropertiesKHR) { .sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_COOPERATIVE_MATRIX_PROPERTIES_KHR, - .pNext = &s->hprops, + .pNext = &s->optical_flow_props, }; s->subgroup_props = (VkPhysicalDeviceSubgroupSizeControlProperties) { .sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SUBGROUP_SIZE_CONTROL_PROPERTIES, diff --git a/libavutil/vulkan.h b/libavutil/vulkan.h index 4b4705a25e..ee0ea18fac 100644 --- a/libavutil/vulkan.h +++ b/libavutil/vulkan.h @@ -238,6 +238,7 @@ typedef struct FFVulkanContext { VkPhysicalDeviceDescriptorBufferPropertiesEXT desc_buf_props; VkPhysicalDeviceSubgroupSizeControlProperties subgroup_props; VkPhysicalDeviceCooperativeMatrixPropertiesKHR coop_matrix_props; + VkPhysicalDeviceOpticalFlowPropertiesNV optical_flow_props; VkQueueFamilyQueryResultStatusPropertiesKHR *query_props; VkQueueFamilyVideoPropertiesKHR *video_props; VkQueueFamilyProperties2 *qf_props; diff --git a/libavutil/vulkan_functions.h b/libavutil/vulkan_functions.h index 6b379acf93..3f0b96f77e 100644 --- a/libavutil/vulkan_functions.h +++ b/libavutil/vulkan_functions.h @@ -46,6 +46,7 @@ typedef enum FFVulkanExtensions { FF_VK_EXT_VIDEO_DECODE_AV1 = 1ULL << 14, /* VK_KHR_video_decode_av1 */ FF_VK_EXT_ATOMIC_FLOAT = 1ULL << 15, /* VK_EXT_shader_atomic_float */ FF_VK_EXT_COOP_MATRIX = 1ULL << 16, /* VK_KHR_cooperative_matrix */ + FF_VK_EXT_OPTICAL_FLOW = 1ULL << 17, /* VK_NV_optical_flow */ FF_VK_EXT_NO_FLAG = 1ULL << 31, } FFVulkanExtensions; @@ -208,6 +209,13 @@ typedef enum FFVulkanExtensions { MACRO(1, 1, FF_VK_EXT_NO_FLAG, DestroySamplerYcbcrConversion) \ MACRO(1, 1, FF_VK_EXT_NO_FLAG, CreateSampler) \ MACRO(1, 1, FF_VK_EXT_NO_FLAG, DestroySampler) \ + \ + /* Optical flow */ \ + MACRO(1, 1, FF_VK_EXT_OPTICAL_FLOW, BindOpticalFlowSessionImageNV) \ + MACRO(1, 1, FF_VK_EXT_OPTICAL_FLOW, CmdOpticalFlowExecuteNV) \ + MACRO(1, 1, FF_VK_EXT_OPTICAL_FLOW, CreateOpticalFlowSessionNV) \ + MACRO(1, 1, FF_VK_EXT_OPTICAL_FLOW, DestroyOpticalFlowSessionNV) \ + MACRO(1, 0, FF_VK_EXT_OPTICAL_FLOW, GetPhysicalDeviceOpticalFlowImageFormatsNV)\ \ /* Shaders */ \ MACRO(1, 1, FF_VK_EXT_NO_FLAG, CreateShaderModule) \ diff --git a/libavutil/vulkan_loader.h b/libavutil/vulkan_loader.h index 73cf03935d..bb22591957 100644 --- a/libavutil/vulkan_loader.h +++ b/libavutil/vulkan_loader.h @@ -49,6 +49,7 @@ static inline uint64_t ff_vk_extensions_to_mask(const char * const *extensions, { VK_EXT_PHYSICAL_DEVICE_DRM_EXTENSION_NAME, FF_VK_EXT_DEVICE_DRM }, { VK_EXT_SHADER_ATOMIC_FLOAT_EXTENSION_NAME, FF_VK_EXT_ATOMIC_FLOAT }, { VK_KHR_COOPERATIVE_MATRIX_EXTENSION_NAME, FF_VK_EXT_COOP_MATRIX }, + { VK_NV_OPTICAL_FLOW_EXTENSION_NAME, FF_VK_EXT_OPTICAL_FLOW }, #ifdef _WIN32 { VK_KHR_EXTERNAL_MEMORY_WIN32_EXTENSION_NAME, FF_VK_EXT_EXTERNAL_WIN32_MEMORY }, { VK_KHR_EXTERNAL_SEMAPHORE_WIN32_EXTENSION_NAME, FF_VK_EXT_EXTERNAL_WIN32_SEM },