vkd3d: Map timeline semaphore values to fence virtual values and buffer out-of-order waits.

Strictly increasing timeline values must be mapped to fence virtual values
to avoid invalid use of Vulkan timeline semaphores. In particular, non-
increasing values and value jumps of >= 4G are permitted in d3d12.

Different virtual D3D12 command queues may map to the same Vulkan queue.
If a wait of value N is submitted on one command queue, and then a signal
for >= N is submitted on another, but they are sent to the same Vk queue,
the wait will never complete. The solution is to buffer out-of-order waits
and any subsequent queue commands until an unblocking signal value is
submitted to a different D3D12 queue, or signaled on the CPU.

Buffering out-of-order waits also fixes the old fence implementation so it
is fully functional, though a bit less efficient than timeline semaphores.

Based in part on vkd3d-proton patches by Hans-Kristian Arntzen. Unlike the
vkd3d-proton implementation, this patch does not use worker threads for
submissions to the Vulkan queue.

Signed-off-by: Conor McCarthy <cmccarthy@codeweavers.com>
Signed-off-by: Henri Verbeet <hverbeet@codeweavers.com>
Signed-off-by: Alexandre Julliard <julliard@winehq.org>

libs/vkd3d/device.c

@@ -747,7 +747,6 @@ struct vkd3d_physical_device_info
     VkPhysicalDeviceTexelBufferAlignmentPropertiesEXT texel_buffer_alignment_properties;
     VkPhysicalDeviceTransformFeedbackPropertiesEXT xfb_properties;
     VkPhysicalDeviceVertexAttributeDivisorPropertiesEXT vertex_divisor_properties;
-    VkPhysicalDeviceTimelineSemaphorePropertiesKHR timeline_semaphore_properties;
 
     VkPhysicalDeviceProperties2KHR properties2;
 
@@ -772,7 +771,6 @@ static void vkd3d_physical_device_info_init(struct vkd3d_physical_device_info *i
     VkPhysicalDeviceDescriptorIndexingPropertiesEXT *descriptor_indexing_properties;
     VkPhysicalDeviceVertexAttributeDivisorPropertiesEXT *vertex_divisor_properties;
     VkPhysicalDeviceTexelBufferAlignmentPropertiesEXT *buffer_alignment_properties;
-    VkPhysicalDeviceTimelineSemaphorePropertiesKHR *timeline_semaphore_properties;
     VkPhysicalDeviceDescriptorIndexingFeaturesEXT *descriptor_indexing_features;
     VkPhysicalDeviceRobustness2FeaturesEXT *robustness2_features;
     VkPhysicalDeviceVertexAttributeDivisorFeaturesEXT *vertex_divisor_features;
@@ -799,7 +797,6 @@ static void vkd3d_physical_device_info_init(struct vkd3d_physical_device_info *i
     vertex_divisor_features = &info->vertex_divisor_features;
     vertex_divisor_properties = &info->vertex_divisor_properties;
     timeline_semaphore_features = &info->timeline_semaphore_features;
-    timeline_semaphore_properties = &info->timeline_semaphore_properties;
     xfb_features = &info->xfb_features;
     xfb_properties = &info->xfb_properties;
 
@@ -841,8 +838,6 @@ static void vkd3d_physical_device_info_init(struct vkd3d_physical_device_info *i
     vk_prepend_struct(&info->properties2, xfb_properties);
     vertex_divisor_properties->sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_VERTEX_ATTRIBUTE_DIVISOR_PROPERTIES_EXT;
     vk_prepend_struct(&info->properties2, vertex_divisor_properties);
-    timeline_semaphore_properties->sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_TIMELINE_SEMAPHORE_PROPERTIES_KHR;
-    vk_prepend_struct(&info->properties2, timeline_semaphore_properties);
 
     if (vulkan_info->KHR_get_physical_device_properties2)
         VK_CALL(vkGetPhysicalDeviceProperties2KHR(physical_device, &info->properties2));
@@ -1431,7 +1426,6 @@ static HRESULT vkd3d_init_device_caps(struct d3d12_device *device,
     vulkan_info->rasterization_stream = physical_device_info->xfb_properties.transformFeedbackRasterizationStreamSelect;
     vulkan_info->transform_feedback_queries = physical_device_info->xfb_properties.transformFeedbackQueries;
     vulkan_info->max_vertex_attrib_divisor = max(physical_device_info->vertex_divisor_properties.maxVertexAttribDivisor, 1);
-    vulkan_info->timeline_semaphore_properties = physical_device_info->timeline_semaphore_properties;
 
     device->feature_options.DoublePrecisionFloatShaderOps = features->shaderFloat64;
     device->feature_options.OutputMergerLogicOp = features->logicOp;
@@ -1908,75 +1902,6 @@ static bool d3d12_is_64k_msaa_supported(struct d3d12_device *device)
             && info.Alignment <= 0x10000;
 }
 
-/* A lower value can be signalled on a D3D12 fence. Vulkan timeline semaphores
- * do not support this, but test if it works anyway. */
-static bool d3d12_is_timeline_semaphore_supported(const struct d3d12_device *device)
-{
-    const struct vkd3d_vk_device_procs *vk_procs = &device->vk_procs;
-    VkTimelineSemaphoreSubmitInfoKHR timeline_submit_info;
-    VkSemaphore timeline_semaphore;
-    VkSubmitInfo submit_info;
-    bool result = false;
-    uint64_t value = 0;
-    VkQueue vk_queue;
-    VkResult vr;
-
-    if (!device->vk_info.KHR_timeline_semaphore)
-        return false;
-
-    if ((vr = vkd3d_create_timeline_semaphore(device, 1, &timeline_semaphore)) < 0)
-    {
-        WARN("Failed to create timeline semaphore, vr %d.\n", vr);
-        return false;
-    }
-
-    if (!(vk_queue = vkd3d_queue_acquire(device->direct_queue)))
-    {
-        ERR("Failed to acquire queue %p.\n", device->direct_queue);
-        VK_CALL(vkDestroySemaphore(device->vk_device, timeline_semaphore, NULL));
-        return false;
-    }
-
-    submit_info.sType = VK_STRUCTURE_TYPE_SUBMIT_INFO;
-    submit_info.pNext = &timeline_submit_info;
-    submit_info.waitSemaphoreCount = 0;
-    submit_info.pWaitSemaphores = NULL;
-    submit_info.pWaitDstStageMask = NULL;
-    submit_info.commandBufferCount = 0;
-    submit_info.pCommandBuffers = NULL;
-    submit_info.signalSemaphoreCount = 1;
-    submit_info.pSignalSemaphores = &timeline_semaphore;
-
-    timeline_submit_info.sType = VK_STRUCTURE_TYPE_TIMELINE_SEMAPHORE_SUBMIT_INFO_KHR;
-    timeline_submit_info.pNext = NULL;
-    timeline_submit_info.pSignalSemaphoreValues = &value;
-    timeline_submit_info.signalSemaphoreValueCount = 1;
-    timeline_submit_info.waitSemaphoreValueCount = 0;
-    timeline_submit_info.pWaitSemaphoreValues = NULL;
-
-    vr = VK_CALL(vkQueueSubmit(vk_queue, 1, &submit_info, VK_NULL_HANDLE));
-
-    if (vr >= 0)
-    {
-        if ((vr = VK_CALL(vkQueueWaitIdle(vk_queue))) < 0)
-            WARN("Failed to wait for queue, vr %d.\n", vr);
-
-        if ((vr = VK_CALL(vkGetSemaphoreCounterValueKHR(device->vk_device, timeline_semaphore, &value))) < 0)
-            ERR("Failed to get Vulkan semaphore status, vr %d.\n", vr);
-        else if (!(result = !value))
-            WARN("Disabling timeline semaphore use due to incompatible behaviour.\n");
-    }
-    else
-    {
-        WARN("Failed to submit signal operation, vr %d.\n", vr);
-    }
-
-    vkd3d_queue_release(device->direct_queue);
-    VK_CALL(vkDestroySemaphore(device->vk_device, timeline_semaphore, NULL));
-
-    return result;
-}
-
 static HRESULT vkd3d_create_vk_device(struct d3d12_device *device,
         const struct vkd3d_device_create_info *create_info)
 {
@@ -2075,10 +2000,6 @@ static HRESULT vkd3d_create_vk_device(struct d3d12_device *device,
     }
 
     device->feature_options4.MSAA64KBAlignedTextureSupported = d3d12_is_64k_msaa_supported(device);
-    device->use_timeline_semaphores = d3d12_is_timeline_semaphore_supported(device)
-            && vkd3d_queue_init_timeline_semaphore(device->direct_queue, device)
-            && vkd3d_queue_init_timeline_semaphore(device->compute_queue, device)
-            && vkd3d_queue_init_timeline_semaphore(device->copy_queue, device);
 
     TRACE("Created Vulkan device %p.\n", vk_device);
 
@@ -4362,6 +4283,8 @@ static HRESULT d3d12_device_init(struct d3d12_device *device,
     vkd3d_gpu_va_allocator_init(&device->gpu_va_allocator);
     vkd3d_time_domains_init(device);
 
+    device->blocked_queue_count = 0;
+
     for (i = 0; i < ARRAY_SIZE(device->desc_mutex); ++i)
         vkd3d_mutex_init(&device->desc_mutex[i]);
 

libs/vkd3d/vkd3d_private.h

@@ -59,7 +59,7 @@
 #define VKD3D_MAX_SHADER_EXTENSIONS       3u
 #define VKD3D_MAX_SHADER_STAGES           5u
 #define VKD3D_MAX_VK_SYNC_OBJECTS         4u
-#define VKD3D_MAX_FENCE_WAITING_QUEUES    4u
+#define VKD3D_MAX_DEVICE_BLOCKED_QUEUES  16u
 #define VKD3D_MAX_DESCRIPTOR_SETS        64u
 /* D3D12 binding tier 3 has a limit of 2048 samplers. */
 #define VKD3D_MAX_DESCRIPTOR_SET_SAMPLERS 2048u
@@ -152,8 +152,6 @@ struct vkd3d_vulkan_info
 
     VkPhysicalDeviceTexelBufferAlignmentPropertiesEXT texel_buffer_alignment_properties;
 
-    VkPhysicalDeviceTimelineSemaphorePropertiesKHR timeline_semaphore_properties;
-
     unsigned int shader_extension_count;
     enum vkd3d_shader_spirv_extension shader_extensions[VKD3D_MAX_SHADER_EXTENSIONS];
 
@@ -502,15 +500,17 @@ HRESULT vkd3d_set_private_data_interface(struct vkd3d_private_store *store, cons
 struct vkd3d_signaled_semaphore
 {
     uint64_t value;
+    union
+    {
+        struct
+        {
             VkSemaphore vk_semaphore;
             VkFence vk_fence;
             bool is_acquired;
-};
+        } binary;
-
+        uint64_t timeline_value;
-struct vkd3d_pending_fence_wait
+    } u;
-{
+    const struct vkd3d_queue *signalling_queue;
-    const struct vkd3d_queue *queue;
-    uint64_t pending_value;
 };
 
 /* ID3D12Fence */
@@ -521,6 +521,7 @@ struct d3d12_fence
     LONG refcount;
 
     uint64_t value;
+    uint64_t max_pending_value;
     struct vkd3d_mutex mutex;
     struct vkd3d_cond null_event_cond;
 
@@ -534,9 +535,8 @@ struct d3d12_fence
     size_t event_count;
 
     VkSemaphore timeline_semaphore;
+    uint64_t timeline_value;
     uint64_t pending_timeline_value;
-    struct vkd3d_pending_fence_wait gpu_waits[VKD3D_MAX_FENCE_WAITING_QUEUES];
-    unsigned int gpu_wait_count;
 
     struct vkd3d_signaled_semaphore *semaphores;
     size_t semaphores_size;
@@ -1294,9 +1294,6 @@ struct vkd3d_queue
     VkQueueFlags vk_queue_flags;
     uint32_t timestamp_bits;
 
-    VkSemaphore wait_completion_semaphore;
-    uint64_t pending_wait_completion_value;
-
     struct
     {
         VkSemaphore vk_semaphore;
@@ -1311,10 +1308,45 @@ struct vkd3d_queue
 VkQueue vkd3d_queue_acquire(struct vkd3d_queue *queue);
 HRESULT vkd3d_queue_create(struct d3d12_device *device, uint32_t family_index,
         const VkQueueFamilyProperties *properties, struct vkd3d_queue **queue);
-bool vkd3d_queue_init_timeline_semaphore(struct vkd3d_queue *queue, struct d3d12_device *device);
 void vkd3d_queue_destroy(struct vkd3d_queue *queue, struct d3d12_device *device);
 void vkd3d_queue_release(struct vkd3d_queue *queue);
 
+enum vkd3d_cs_op
+{
+    VKD3D_CS_OP_WAIT,
+    VKD3D_CS_OP_SIGNAL,
+    VKD3D_CS_OP_EXECUTE,
+};
+
+struct vkd3d_cs_wait
+{
+    struct d3d12_fence *fence;
+    uint64_t value;
+};
+
+struct vkd3d_cs_signal
+{
+    struct d3d12_fence *fence;
+    uint64_t value;
+};
+
+struct vkd3d_cs_execute
+{
+    VkCommandBuffer *buffers;
+    unsigned int buffer_count;
+};
+
+struct vkd3d_cs_op_data
+{
+    enum vkd3d_cs_op opcode;
+    union
+    {
+        struct vkd3d_cs_wait wait;
+        struct vkd3d_cs_signal signal;
+        struct vkd3d_cs_execute execute;
+    } u;
+};
+
 /* ID3D12CommandQueue */
 struct d3d12_command_queue
 {
@@ -1331,6 +1363,12 @@ struct d3d12_command_queue
 
     struct d3d12_device *device;
 
+    struct vkd3d_mutex op_mutex;
+    struct vkd3d_cs_op_data *ops;
+    size_t ops_count;
+    size_t ops_size;
+    bool is_flushing;
+
     struct vkd3d_private_store private_store;
 };
 
@@ -1452,6 +1490,9 @@ struct d3d12_device
     unsigned int queue_family_count;
     VkTimeDomainEXT vk_host_time_domain;
 
+    struct d3d12_command_queue *blocked_queues[VKD3D_MAX_DEVICE_BLOCKED_QUEUES];
+    unsigned int blocked_queue_count;
+
     struct vkd3d_instance *vkd3d_instance;
 
     IUnknown *parent;
@@ -1470,7 +1511,6 @@ struct d3d12_device
     VkDescriptorPoolSize vk_pool_sizes[VKD3D_DESCRIPTOR_POOL_COUNT];
     struct vkd3d_vk_descriptor_heap_layout vk_descriptor_heap_layouts[VKD3D_SET_INDEX_COUNT];
     bool use_vk_heaps;
-    bool use_timeline_semaphores;
 };
 
 HRESULT d3d12_device_create(struct vkd3d_instance *instance,

tests/d3d12.c

@@ -33224,9 +33224,7 @@ static void test_queue_wait(void)
     command_list = context.list;
     queue = context.queue;
 
-    /* 'queue2' must not map to the same command queue as 'queue', or Wait() before GPU signal will fail.
+    queue2 = create_command_queue(device, D3D12_COMMAND_LIST_TYPE_DIRECT, D3D12_COMMAND_QUEUE_PRIORITY_NORMAL);
-     * Using a compute queue fixes this on most hardware, but it may still fail on low spec hardware. */
-    queue2 = create_command_queue(device, D3D12_COMMAND_LIST_TYPE_COMPUTE, D3D12_COMMAND_QUEUE_PRIORITY_NORMAL);
 
     event = create_event();
     ok(event, "Failed to create event.\n");