lavfi/vulkan: refactor, fix and fully implement multiple queues

1 files changed, 66 insertions, 37 deletions

diff --git a/libavfilter/vulkan.h b/libavfilter/vulkan.h
index 89b76ba355..9d17d2b14f 100644
--- a/libavfilter/vulkan.h
+++ b/libavfilter/vulkan.h
@@ -20,6 +20,7 @@
 #define AVFILTER_VULKAN_H
 
 #define VK_NO_PROTOTYPES
+#define VK_ENABLE_BETA_EXTENSIONS
 
 #include "avfilter.h"
 #include "libavutil/pixdesc.h"
@@ -52,9 +53,6 @@
             goto fail;                                                         \
     } while (0)
 
-/* Useful for attaching immutable samplers to arrays */
-#define DUP_SAMPLER_ARRAY4(x) (VkSampler []){ x, x, x, x, }
-
 typedef struct FFSPIRVShader {
     const char *name;                       /* Name for id/debugging purposes */
     AVBPrint src;
@@ -62,7 +60,11 @@ typedef struct FFSPIRVShader {
     VkPipelineShaderStageCreateInfo shader;
 } FFSPIRVShader;
 
-typedef struct VulkanDescriptorSetBinding {
+typedef struct FFVkSampler {
+    VkSampler sampler[4];
+} FFVkSampler;
+
+typedef struct FFVulkanDescriptorSetBinding {
     const char         *name;
     VkDescriptorType    type;
     const char         *mem_layout;  /* Storage images (rgba8, etc.) and buffers (std430, etc.) */
@@ -71,9 +73,9 @@ typedef struct VulkanDescriptorSetBinding {
     uint32_t            dimensions;  /* Needed for e.g. sampler%iD */
     uint32_t            elems;       /* 0 - scalar, 1 or more - vector */
     VkShaderStageFlags  stages;
-    const VkSampler    *samplers;    /* Immutable samplers, length - #elems */
+    FFVkSampler        *sampler;     /* Sampler to use for all elems */
     void               *updater;     /* Pointer to VkDescriptor*Info */
-} VulkanDescriptorSetBinding;
+} FFVulkanDescriptorSetBinding;
 
 typedef struct FFVkBuffer {
     VkBuffer buf;
@@ -81,7 +83,15 @@ typedef struct FFVkBuffer {
     VkMemoryPropertyFlagBits flags;
 } FFVkBuffer;
 
-typedef struct VulkanPipeline {
+typedef struct FFVkQueueFamilyCtx {
+    int queue_family;
+    int nb_queues;
+    int cur_queue;
+} FFVkQueueFamilyCtx;
+
+typedef struct FFVulkanPipeline {
+    FFVkQueueFamilyCtx *qf;
+
     VkPipelineBindPoint bind_point;
 
     /* Contexts */
@@ -97,18 +107,21 @@ typedef struct VulkanPipeline {
     int push_consts_num;
 
     /* Descriptors */
-    VkDescriptorSetLayout      *desc_layout;
-    VkDescriptorPool            desc_pool;
-    VkDescriptorSet            *desc_set;
-    VkDescriptorUpdateTemplate *desc_template;
-    int                         desc_layout_num;
-    int                         descriptor_sets_num;
-    int                         pool_size_desc_num;
+    VkDescriptorSetLayout         *desc_layout;
+    VkDescriptorPool               desc_pool;
+    VkDescriptorSet               *desc_set;
+    void                         **desc_staging;
+    VkDescriptorSetLayoutBinding **desc_binding;
+    VkDescriptorUpdateTemplate    *desc_template;
+    int                            desc_layout_num;
+    int                            descriptor_sets_num;
+    int                            total_descriptor_sets;
+    int                            pool_size_desc_num;
 
     /* Temporary, used to store data in between initialization stages */
     VkDescriptorUpdateTemplateCreateInfo *desc_template_info;
     VkDescriptorPoolSize *pool_size_desc;
-} VulkanPipeline;
+} FFVulkanPipeline;
 
 typedef struct FFVkQueueCtx {
     VkFence fence;
@@ -126,6 +139,8 @@ typedef struct FFVkQueueCtx {
 } FFVkQueueCtx;
 
 typedef struct FFVkExecContext {
+    FFVkQueueFamilyCtx *qf;
+
     VkCommandPool pool;
     VkCommandBuffer *bufs;
     FFVkQueueCtx *queues;
@@ -134,7 +149,7 @@ typedef struct FFVkExecContext {
     int *nb_deps;
     int *dep_alloc_size;
 
-    VulkanPipeline *bound_pl;
+    FFVulkanPipeline *bound_pl;
 
     VkSemaphore *sem_wait;
     int sem_wait_alloc; /* Allocated sem_wait */
@@ -152,23 +167,23 @@ typedef struct FFVkExecContext {
 
     uint64_t *sem_sig_val;
     int sem_sig_val_alloc;
+
+    uint64_t **sem_sig_val_dst;
+    int sem_sig_val_dst_alloc;
 } FFVkExecContext;
 
-typedef struct VulkanFilterContext {
+typedef struct FFVulkanContext {
     const AVClass         *class;
     FFVulkanFunctions     vkfn;
     FFVulkanExtensions    extensions;
+    VkPhysicalDeviceProperties props;
+    VkPhysicalDeviceMemoryProperties mprops;
 
     AVBufferRef           *device_ref;
     AVBufferRef           *frames_ref; /* For in-place filtering */
     AVHWDeviceContext     *device;
     AVVulkanDeviceContext *hwctx;
 
-    /* State - mirrored with the exec ctx */
-    int cur_queue_idx;
-    int queue_family_idx;
-    int queue_count;
-
     /* Properties */
     int                 output_width;
     int                output_height;
@@ -176,7 +191,7 @@ typedef struct VulkanFilterContext {
     enum AVPixelFormat  input_format;
 
     /* Samplers */
-    VkSampler **samplers;
+    FFVkSampler **samplers;
     int samplers_num;
 
     /* Exec contexts */
@@ -184,12 +199,12 @@ typedef struct VulkanFilterContext {
     int exec_ctx_num;
 
     /* Pipelines (each can have 1 shader of each type) */
-    VulkanPipeline **pipelines;
+    FFVulkanPipeline **pipelines;
     int pipelines_num;
 
     void *scratch; /* Scratch memory used only in functions */
     unsigned int scratch_size;
-} VulkanFilterContext;
+} FFVulkanContext;
 
 /* Identity mapping - r = r, b = b, g = g, a = a */
 extern const VkComponentMapping ff_comp_identity_map;
@@ -219,10 +234,22 @@ int ff_vk_mt_is_np_rgb(enum AVPixelFormat pix_fmt);
 const char *ff_vk_shader_rep_fmt(enum AVPixelFormat pixfmt);
 
 /**
+ * Initialize a queue family.
+ * A queue limit of 0 means no limit.
+ */
+void ff_vk_qf_init(AVFilterContext *avctx, FFVkQueueFamilyCtx *qf,
+                   VkQueueFlagBits dev_family, int queue_limit);
+
+/**
+ * Rotate through the queues in a queue family.
+ */
+void ff_vk_qf_rotate(FFVkQueueFamilyCtx *qf);
+
+/**
  * Create a Vulkan sampler, will be auto-freed in ff_vk_filter_uninit()
  */
-VkSampler *ff_vk_init_sampler(AVFilterContext *avctx, int unnorm_coords,
-                              VkFilter filt);
+FFVkSampler *ff_vk_init_sampler(AVFilterContext *avctx, int unnorm_coords,
+                                VkFilter filt);
 
 /**
  * Create an imageview.
@@ -237,19 +264,20 @@ int ff_vk_create_imageview(AVFilterContext *avctx, FFVkExecContext *e,
  * Define a push constant for a given stage into a pipeline.
  * Must be called before the pipeline layout has been initialized.
  */
-int ff_vk_add_push_constant(AVFilterContext *avctx, VulkanPipeline *pl,
+int ff_vk_add_push_constant(AVFilterContext *avctx, FFVulkanPipeline *pl,
                             int offset, int size, VkShaderStageFlagBits stage);
 
 /**
  * Inits a pipeline. Everything in it will be auto-freed when calling
  * ff_vk_filter_uninit().
  */
-VulkanPipeline *ff_vk_create_pipeline(AVFilterContext *avctx);
+FFVulkanPipeline *ff_vk_create_pipeline(AVFilterContext *avctx,
+                                      FFVkQueueFamilyCtx *qf);
 
 /**
  * Inits a shader for a specific pipeline. Will be auto-freed on uninit.
  */
-FFSPIRVShader *ff_vk_init_shader(AVFilterContext *avctx, VulkanPipeline *pl,
+FFSPIRVShader *ff_vk_init_shader(AVFilterContext *avctx, FFVulkanPipeline *pl,
                                  const char *name, VkShaderStageFlags stage);
 
 /**
@@ -261,8 +289,8 @@ void ff_vk_set_compute_shader_sizes(AVFilterContext *avctx, FFSPIRVShader *shd,
 /**
  * Adds a descriptor set to the shader and registers them in the pipeline.
  */
-int ff_vk_add_descriptor_set(AVFilterContext *avctx, VulkanPipeline *pl,
-                             FFSPIRVShader *shd, VulkanDescriptorSetBinding *desc,
+int ff_vk_add_descriptor_set(AVFilterContext *avctx, FFVulkanPipeline *pl,
+                             FFSPIRVShader *shd, FFVulkanDescriptorSetBinding *desc,
                              int num, int only_print_to_shader);
 
 /**
@@ -280,27 +308,28 @@ void ff_vk_print_shader(AVFilterContext *avctx, FFSPIRVShader *shd, int prio);
  * Initializes the pipeline layout after all shaders and descriptor sets have
  * been finished.
  */
-int ff_vk_init_pipeline_layout(AVFilterContext *avctx, VulkanPipeline *pl);
+int ff_vk_init_pipeline_layout(AVFilterContext *avctx, FFVulkanPipeline *pl);
 
 /**
  * Initializes a compute pipeline. Will pick the first shader with the
  * COMPUTE flag set.
  */
-int ff_vk_init_compute_pipeline(AVFilterContext *avctx, VulkanPipeline *pl);
+int ff_vk_init_compute_pipeline(AVFilterContext *avctx, FFVulkanPipeline *pl);
 
 /**
  * Updates a descriptor set via the updaters defined.
  * Can be called immediately after pipeline creation, but must be called
  * at least once before queue submission.
  */
-void ff_vk_update_descriptor_set(AVFilterContext *avctx, VulkanPipeline *pl,
+void ff_vk_update_descriptor_set(AVFilterContext *avctx, FFVulkanPipeline *pl,
                                  int set_id);
 
 /**
  * Init an execution context for command recording and queue submission.
  * WIll be auto-freed on uninit.
  */
-int ff_vk_create_exec_ctx(AVFilterContext *avctx, FFVkExecContext **ctx);
+int ff_vk_create_exec_ctx(AVFilterContext *avctx, FFVkExecContext **ctx,
+                          FFVkQueueFamilyCtx *qf);
 
 /**
  * Begin recording to the command buffer. Previous execution must have been
@@ -313,7 +342,7 @@ int ff_vk_start_exec_recording(AVFilterContext *avctx, FFVkExecContext *e);
  * Must be called after ff_vk_start_exec_recording() and before submission.
  */
 void ff_vk_bind_pipeline_exec(AVFilterContext *avctx, FFVkExecContext *e,
-                              VulkanPipeline *pl);
+                              FFVulkanPipeline *pl);
 
 /**
  * Updates push constants.