May 2019 - Wine-commits - list.winehq.org

Józef Kucia : vkd3d: Use list to store Vulkan semaphores.
by Alexandre Julliard 03 May '19

03 May '19

Module: vkd3d Branch: master Commit: 134f4d0081ce5b65ec7cdc534d29be0877c7fc64 URL: https://source.winehq.org/git/vkd3d.git/?a=commit;h=134f4d0081ce5b65ec7cdc5… Author: Józef Kucia <jkucia(a)codeweavers.com> Date: Thu May 2 16:02:39 2019 +0200 vkd3d: Use list to store Vulkan semaphores. Signed-off-by: Józef Kucia <jkucia(a)codeweavers.com> Signed-off-by: Henri Verbeet <hverbeet(a)codeweavers.com> Signed-off-by: Alexandre Julliard <julliard(a)winehq.org> --- libs/vkd3d/command.c | 81 ++++++++++++++++++++-------------------------- libs/vkd3d/vkd3d_private.h | 18 ++++++----- 2 files changed, 45 insertions(+), 54 deletions(-) diff --git a/libs/vkd3d/command.c b/libs/vkd3d/command.c index c609270..ddd36d4 100644 --- a/libs/vkd3d/command.c +++ b/libs/vkd3d/command.c @@ -446,45 +446,35 @@ static void d3d12_fence_garbage_collect_vk_semaphores_locked(struct d3d12_fence { struct d3d12_device *device = fence->device; const struct vkd3d_vk_device_procs *vk_procs = &device->vk_procs; - VkSemaphore vk_semaphore; - size_t semaphore_count; - unsigned int i, j; + struct vkd3d_signaled_semaphore *current, *p; + unsigned int semaphore_count; semaphore_count = fence->semaphore_count; if (!destroy_all && semaphore_count < VKD3D_MAX_VK_SYNC_OBJECTS_PER_D3D12_FENCE) return; - for (i = 0, j = 0; i < fence->semaphore_count; ++i) + LIST_FOR_EACH_ENTRY_SAFE(current, p, &fence->semaphores, struct vkd3d_signaled_semaphore, entry) { - if (!destroy_all && semaphore_count < VKD3D_MAX_VK_SYNC_OBJECTS_PER_D3D12_FENCE) + if (!destroy_all && fence->semaphore_count < VKD3D_MAX_VK_SYNC_OBJECTS_PER_D3D12_FENCE) break; /* The semaphore doesn't have a pending signal operation if the fence * was signaled. */ - if (!fence->semaphores[i].vk_fence || destroy_all) - { - vk_semaphore = fence->semaphores[i].vk_semaphore; - if (fence->semaphores[i].vk_fence) - WARN("Destroying potentially pending semaphore.\n"); - VK_CALL(vkDestroySemaphore(device->vk_device, vk_semaphore, NULL)); - --semaphore_count; - } - else - { - if (i != j) - fence->semaphores[j] = fence->semaphores[i]; - ++j; - } - } - for (; i < fence->semaphore_count; ++i) - { - fence->semaphores[j++] = fence->semaphores[i]; - } + if (current->vk_fence && !destroy_all) + continue; - if (j != fence->semaphore_count) - TRACE("Destroyed %zu Vulkan semaphores.\n", fence->semaphore_count - j); + if (current->vk_fence) + WARN("Destroying potentially pending semaphore.\n"); + + VK_CALL(vkDestroySemaphore(device->vk_device, current->vk_semaphore, NULL)); + list_remove(&current->entry); + vkd3d_free(current); + + --fence->semaphore_count; + } - fence->semaphore_count = j; + if (semaphore_count != fence->semaphore_count) + TRACE("Destroyed %u Vulkan semaphores.\n", semaphore_count - fence->semaphore_count); } static void d3d12_fence_destroy_vk_objects(struct d3d12_fence *fence) @@ -517,32 +507,33 @@ static void d3d12_fence_destroy_vk_objects(struct d3d12_fence *fence) static HRESULT d3d12_fence_add_vk_semaphore(struct d3d12_fence *fence, VkSemaphore vk_semaphore, VkFence vk_fence, uint64_t value) { + struct vkd3d_signaled_semaphore *semaphore; HRESULT hr = S_OK; int rc; TRACE("fence %p, value %#"PRIx64".\n", fence, value); + if (!(semaphore = vkd3d_malloc(sizeof(*semaphore)))) + { + ERR("Failed to add semaphore.\n"); + return E_OUTOFMEMORY; + } + if ((rc = pthread_mutex_lock(&fence->mutex))) { ERR("Failed to lock mutex, error %d.\n", rc); + vkd3d_free(semaphore); return E_FAIL; } d3d12_fence_garbage_collect_vk_semaphores_locked(fence, false); - if (vkd3d_array_reserve((void **)&fence->semaphores, &fence->semaphores_size, - fence->semaphore_count + 1, sizeof(*fence->semaphores))) - { - fence->semaphores[fence->semaphore_count].value = value; - fence->semaphores[fence->semaphore_count].vk_semaphore = vk_semaphore; - fence->semaphores[fence->semaphore_count].vk_fence = vk_fence; - ++fence->semaphore_count; - } - else - { - ERR("Failed to add semaphore.\n"); - hr = E_OUTOFMEMORY; - } + semaphore->value = value; + semaphore->vk_semaphore = vk_semaphore; + semaphore->vk_fence = vk_fence; + + list_add_tail(&fence->semaphores, &semaphore->entry); + ++fence->semaphore_count; pthread_mutex_unlock(&fence->mutex); @@ -552,6 +543,7 @@ static HRESULT d3d12_fence_add_vk_semaphore(struct d3d12_fence *fence, static HRESULT d3d12_fence_signal(struct d3d12_fence *fence, uint64_t value, VkFence vk_fence) { struct d3d12_device *device = fence->device; + struct vkd3d_signaled_semaphore *current; unsigned int i, j; int rc; @@ -584,10 +576,8 @@ static HRESULT d3d12_fence_signal(struct d3d12_fence *fence, uint64_t value, VkF { const struct vkd3d_vk_device_procs *vk_procs = &device->vk_procs; - for (i = 0; i < fence->semaphore_count; ++i) + LIST_FOR_EACH_ENTRY(current, &fence->semaphores, struct vkd3d_signaled_semaphore, entry) { - struct vkd3d_signaled_semaphore *current = &fence->semaphores[i]; - if (current->vk_fence == vk_fence) current->vk_fence = VK_NULL_HANDLE; } @@ -661,7 +651,6 @@ static ULONG STDMETHODCALLTYPE d3d12_fence_Release(ID3D12Fence *iface) d3d12_fence_destroy_vk_objects(fence); - vkd3d_free(fence->semaphores); vkd3d_free(fence->events); if ((rc = pthread_mutex_destroy(&fence->mutex))) ERR("Failed to destroy mutex, error %d.\n", rc); @@ -851,8 +840,8 @@ static HRESULT d3d12_fence_init(struct d3d12_fence *fence, struct d3d12_device * fence->events = NULL; fence->events_size = 0; fence->event_count = 0; - fence->semaphores = NULL; - fence->semaphores_size = 0; + + list_init(&fence->semaphores); fence->semaphore_count = 0; memset(fence->old_vk_fences, 0, sizeof(fence->old_vk_fences)); diff --git a/libs/vkd3d/vkd3d_private.h b/libs/vkd3d/vkd3d_private.h index 0335822..09ac4e0 100644 --- a/libs/vkd3d/vkd3d_private.h +++ b/libs/vkd3d/vkd3d_private.h @@ -279,6 +279,14 @@ HRESULT vkd3d_set_private_data(struct vkd3d_private_store *store, HRESULT vkd3d_set_private_data_interface(struct vkd3d_private_store *store, const GUID *tag, const IUnknown *object) DECLSPEC_HIDDEN; +struct vkd3d_signaled_semaphore +{ + struct list entry; + uint64_t value; + VkSemaphore vk_semaphore; + VkFence vk_fence; +}; + /* ID3D12Fence */ struct d3d12_fence { @@ -296,14 +304,8 @@ struct d3d12_fence size_t events_size; size_t event_count; - struct vkd3d_signaled_semaphore - { - uint64_t value; - VkSemaphore vk_semaphore; - VkFence vk_fence; - } *semaphores; - size_t semaphores_size; - size_t semaphore_count; + struct list semaphores; + unsigned int semaphore_count; VkFence old_vk_fences[VKD3D_MAX_VK_SYNC_OBJECTS_PER_D3D12_FENCE];

1 0

Józef Kucia : vkd3d: Introduce queue sequence numbers to track unused Vulkan semaphores.
by Alexandre Julliard 03 May '19

03 May '19

Module: vkd3d Branch: master Commit: 205a41aaaae05c03ea44a80d1bcfc2048410dada URL: https://source.winehq.org/git/vkd3d.git/?a=commit;h=205a41aaaae05c03ea44a80… Author: Józef Kucia <jkucia(a)codeweavers.com> Date: Thu May 2 16:02:40 2019 +0200 vkd3d: Introduce queue sequence numbers to track unused Vulkan semaphores. In preparation for the following commmit. Sequence numbers allows us to more easily track progress on queues in comparison to tracking each VkFence individually. Signed-off-by: Józef Kucia <jkucia(a)codeweavers.com> Signed-off-by: Henri Verbeet <hverbeet(a)codeweavers.com> Signed-off-by: Alexandre Julliard <julliard(a)winehq.org> --- libs/vkd3d/command.c | 92 +++++++++++++++++++++++++++++++++++++++++----- libs/vkd3d/device.c | 6 +-- libs/vkd3d/vkd3d_private.h | 19 +++++++++- 3 files changed, 103 insertions(+), 14 deletions(-) diff --git a/libs/vkd3d/command.c b/libs/vkd3d/command.c index ddd36d4..2febca6 100644 --- a/libs/vkd3d/command.c +++ b/libs/vkd3d/command.c @@ -38,10 +38,17 @@ HRESULT vkd3d_queue_create(struct d3d12_device *device, return hresult_from_errno(rc); } + object->completed_sequence_number = 0; + object->submitted_sequence_number = 0; + object->vk_family_index = family_index; object->vk_queue_flags = properties->queueFlags; object->timestamp_bits = properties->timestampValidBits; + object->semaphores = NULL; + object->semaphores_size = 0; + object->semaphore_count = 0; + VK_CALL(vkGetDeviceQueue(device->vk_device, family_index, 0, &object->vk_queue)); TRACE("Created queue %p for queue family index %u.\n", object, family_index); @@ -51,8 +58,23 @@ HRESULT vkd3d_queue_create(struct d3d12_device *device, return S_OK; } -void vkd3d_queue_destroy(struct vkd3d_queue *queue) +void vkd3d_queue_destroy(struct vkd3d_queue *queue, struct d3d12_device *device) { + const struct vkd3d_vk_device_procs *vk_procs = &device->vk_procs; + unsigned int i; + int rc; + + if ((rc = pthread_mutex_lock(&queue->mutex))) + ERR("Failed to lock mutex, error %d.\n", rc); + + for (i = 0; i < queue->semaphore_count; ++i) + VK_CALL(vkDestroySemaphore(device->vk_device, queue->semaphores[i].vk_semaphore, NULL)); + + vkd3d_free(queue->semaphores); + + if (!rc) + pthread_mutex_unlock(&queue->mutex); + pthread_mutex_destroy(&queue->mutex); vkd3d_free(queue); } @@ -103,9 +125,46 @@ static VkResult vkd3d_queue_wait_idle(struct vkd3d_queue *queue, return vr; } +static void vkd3d_queue_update_sequence_number(struct vkd3d_queue *queue, + uint64_t sequence_number, struct d3d12_device *device) +{ + const struct vkd3d_vk_device_procs *vk_procs = &device->vk_procs; + uint64_t completed_sequence_number; + unsigned int i; + int rc; + + if ((rc = pthread_mutex_lock(&queue->mutex))) + { + ERR("Failed to lock mutex, error %d.\n", rc); + return; + } + + completed_sequence_number = queue->completed_sequence_number; + queue->completed_sequence_number = max(sequence_number, queue->completed_sequence_number); + + TRACE("Sequence number %"PRIu64" -> %"PRIu64".\n", completed_sequence_number, queue->completed_sequence_number); + + for (i = 0; i < queue->semaphore_count; ++i) + { + if (queue->semaphores[i].sequence_number > queue->completed_sequence_number) + break; + + VK_CALL(vkDestroySemaphore(device->vk_device, queue->semaphores[i].vk_semaphore, NULL)); + } + if (i > 0) + { + TRACE("Destroyed %u Vulkan semaphores.\n", i); + queue->semaphore_count -= i; + memmove(queue->semaphores, &queue->semaphores[i], queue->semaphore_count * sizeof(*queue->semaphores)); + } + + pthread_mutex_unlock(&queue->mutex); +} + /* Fence worker thread */ static HRESULT vkd3d_enqueue_gpu_fence(struct vkd3d_fence_worker *worker, - VkFence vk_fence, struct d3d12_fence *fence, uint64_t value) + VkFence vk_fence, struct d3d12_fence *fence, uint64_t value, + struct vkd3d_queue *queue, uint64_t queue_sequence_number) { int rc; @@ -135,6 +194,8 @@ static HRESULT vkd3d_enqueue_gpu_fence(struct vkd3d_fence_worker *worker, worker->vk_fences[worker->fence_count] = vk_fence; worker->fences[worker->fence_count].fence = fence; worker->fences[worker->fence_count].value = value; + worker->fences[worker->fence_count].queue = queue; + worker->fences[worker->fence_count].queue_sequence_number = queue_sequence_number; ++worker->fence_count; pthread_cond_signal(&worker->cond); @@ -204,9 +265,12 @@ static void vkd3d_wait_for_gpu_fences(struct vkd3d_fence_worker *worker) if (!(vr = VK_CALL(vkGetFenceStatus(device->vk_device, vk_fence)))) { struct vkd3d_waiting_fence *current = &worker->fences[i]; + TRACE("Signaling fence %p value %#"PRIx64".\n", current->fence, current->value); if (FAILED(hr = d3d12_fence_signal(current->fence, current->value, vk_fence))) ERR("Failed to signal D3D12 fence, hr %#x.\n", hr); + + vkd3d_queue_update_sequence_number(current->queue, current->queue_sequence_number, device); continue; } @@ -5103,9 +5167,11 @@ static HRESULT STDMETHODCALLTYPE d3d12_command_queue_Signal(ID3D12CommandQueue * VkSemaphore vk_semaphore = VK_NULL_HANDLE; VkSemaphoreCreateInfo semaphore_info; VkFence vk_fence = VK_NULL_HANDLE; + struct vkd3d_queue *vkd3d_queue; struct d3d12_device *device; struct d3d12_fence *fence; VkSubmitInfo submit_info; + uint64_t sequence_number; VkQueue vk_queue; VkResult vr; HRESULT hr; @@ -5114,6 +5180,7 @@ static HRESULT STDMETHODCALLTYPE d3d12_command_queue_Signal(ID3D12CommandQueue * device = command_queue->device; vk_procs = &device->vk_procs; + vkd3d_queue = command_queue->vkd3d_queue; fence = unsafe_impl_from_ID3D12Fence(fence_iface); @@ -5144,17 +5211,23 @@ static HRESULT STDMETHODCALLTYPE d3d12_command_queue_Signal(ID3D12CommandQueue * submit_info.signalSemaphoreCount = 1; submit_info.pSignalSemaphores = &vk_semaphore; - if (!(vk_queue = vkd3d_queue_acquire(command_queue->vkd3d_queue))) + if (!(vk_queue = vkd3d_queue_acquire(vkd3d_queue))) { - ERR("Failed to acquire queue %p.\n", command_queue->vkd3d_queue); + ERR("Failed to acquire queue %p.\n", vkd3d_queue); hr = E_FAIL; goto fail; } - vr = VK_CALL(vkQueueSubmit(vk_queue, 1, &submit_info, vk_fence)); - vkd3d_queue_release(command_queue->vkd3d_queue); + + if ((vr = VK_CALL(vkQueueSubmit(vk_queue, 1, &submit_info, vk_fence))) >= 0) + sequence_number = ++vkd3d_queue->submitted_sequence_number; + + vkd3d_queue_release(vkd3d_queue); + + assert(sequence_number != ~(uint64_t)0); + if (vr < 0) { - WARN("Failed to submit, vr %d.\n", vr); + WARN("Failed to submit signal operation, vr %d.\n", vr); goto fail_vkresult; } @@ -5164,7 +5237,7 @@ static HRESULT STDMETHODCALLTYPE d3d12_command_queue_Signal(ID3D12CommandQueue * vr = VK_CALL(vkGetFenceStatus(device->vk_device, vk_fence)); if (vr == VK_NOT_READY) { - if (SUCCEEDED(hr = vkd3d_enqueue_gpu_fence(&device->fence_worker, vk_fence, fence, value))) + if (SUCCEEDED(hr = vkd3d_enqueue_gpu_fence(&device->fence_worker, vk_fence, fence, value, vkd3d_queue, sequence_number))) vk_fence = VK_NULL_HANDLE; } else if (vr == VK_SUCCESS) @@ -5172,6 +5245,7 @@ static HRESULT STDMETHODCALLTYPE d3d12_command_queue_Signal(ID3D12CommandQueue * TRACE("Already signaled %p, value %#"PRIx64".\n", fence, value); hr = d3d12_fence_signal(fence, value, vk_fence); vk_fence = VK_NULL_HANDLE; + vkd3d_queue_update_sequence_number(vkd3d_queue, sequence_number, device); } else { @@ -5182,7 +5256,7 @@ static HRESULT STDMETHODCALLTYPE d3d12_command_queue_Signal(ID3D12CommandQueue * if (vk_fence || vk_semaphore) { /* In case of an unexpected failure, try to safely destroy Vulkan objects. */ - vkd3d_queue_wait_idle(command_queue->vkd3d_queue, vk_procs); + vkd3d_queue_wait_idle(vkd3d_queue, vk_procs); goto fail; } diff --git a/libs/vkd3d/device.c b/libs/vkd3d/device.c index 41f7ea1..cb2934a 100644 --- a/libs/vkd3d/device.c +++ b/libs/vkd3d/device.c @@ -1425,12 +1425,12 @@ struct vkd3d_device_queue_info static void d3d12_device_destroy_vkd3d_queues(struct d3d12_device *device) { if (device->direct_queue) - vkd3d_queue_destroy(device->direct_queue); + vkd3d_queue_destroy(device->direct_queue, device); if (device->compute_queue && device->compute_queue != device->direct_queue) - vkd3d_queue_destroy(device->compute_queue); + vkd3d_queue_destroy(device->compute_queue, device); if (device->copy_queue && device->copy_queue != device->direct_queue && device->copy_queue != device->compute_queue) - vkd3d_queue_destroy(device->copy_queue); + vkd3d_queue_destroy(device->copy_queue, device); device->direct_queue = NULL; device->compute_queue = NULL; diff --git a/libs/vkd3d/vkd3d_private.h b/libs/vkd3d/vkd3d_private.h index 09ac4e0..f934d0b 100644 --- a/libs/vkd3d/vkd3d_private.h +++ b/libs/vkd3d/vkd3d_private.h @@ -157,7 +157,9 @@ struct vkd3d_fence_worker struct vkd3d_waiting_fence { struct d3d12_fence *fence; - UINT64 value; + uint64_t value; + struct vkd3d_queue *queue; + uint64_t queue_sequence_number; } *fences; size_t fences_size; @@ -882,17 +884,30 @@ struct vkd3d_queue { /* Access to VkQueue must be externally synchronized. */ pthread_mutex_t mutex; + VkQueue vk_queue; + + uint64_t completed_sequence_number; + uint64_t submitted_sequence_number; + uint32_t vk_family_index; VkQueueFlags vk_queue_flags; uint32_t timestamp_bits; + + struct + { + VkSemaphore vk_semaphore; + uint64_t sequence_number; + } *semaphores; + size_t semaphores_size; + size_t semaphore_count; }; VkQueue vkd3d_queue_acquire(struct vkd3d_queue *queue) DECLSPEC_HIDDEN; HRESULT vkd3d_queue_create(struct d3d12_device *device, uint32_t family_index, const VkQueueFamilyProperties *properties, struct vkd3d_queue **queue) DECLSPEC_HIDDEN; -void vkd3d_queue_destroy(struct vkd3d_queue *queue) DECLSPEC_HIDDEN; +void vkd3d_queue_destroy(struct vkd3d_queue *queue, struct d3d12_device *device) DECLSPEC_HIDDEN; void vkd3d_queue_release(struct vkd3d_queue *queue) DECLSPEC_HIDDEN; /* ID3D12CommandQueue */

1 0

Józef Kucia : vkd3d: Implement d3d12_command_queue_Wait().
by Alexandre Julliard 03 May '19

03 May '19

Module: vkd3d Branch: master Commit: 124e6aad4817e9fb4f78c0fd33788fe5d27abe92 URL: https://source.winehq.org/git/vkd3d.git/?a=commit;h=124e6aad4817e9fb4f78c0f… Author: Józef Kucia <jkucia(a)codeweavers.com> Date: Thu May 2 16:02:41 2019 +0200 vkd3d: Implement d3d12_command_queue_Wait(). Waiting before the signal is not supported yet. Signed-off-by: Józef Kucia <jkucia(a)codeweavers.com> Signed-off-by: Henri Verbeet <hverbeet(a)codeweavers.com> Signed-off-by: Alexandre Julliard <julliard(a)winehq.org> --- libs/vkd3d/command.c | 164 ++++++++++++++++++++++++++++++++++++++++++++- libs/vkd3d/vkd3d_private.h | 1 + 2 files changed, 162 insertions(+), 3 deletions(-) diff --git a/libs/vkd3d/command.c b/libs/vkd3d/command.c index 2febca6..c8e23e1 100644 --- a/libs/vkd3d/command.c +++ b/libs/vkd3d/command.c @@ -524,11 +524,12 @@ static void d3d12_fence_garbage_collect_vk_semaphores_locked(struct d3d12_fence /* The semaphore doesn't have a pending signal operation if the fence * was signaled. */ - if (current->vk_fence && !destroy_all) + if ((current->vk_fence || current->is_acquired) && !destroy_all) continue; if (current->vk_fence) WARN("Destroying potentially pending semaphore.\n"); + assert(!current->is_acquired); VK_CALL(vkDestroySemaphore(device->vk_device, current->vk_semaphore, NULL)); list_remove(&current->entry); @@ -568,6 +569,83 @@ static void d3d12_fence_destroy_vk_objects(struct d3d12_fence *fence) pthread_mutex_unlock(&fence->mutex); } +static struct vkd3d_signaled_semaphore *d3d12_fence_acquire_vk_semaphore(struct d3d12_fence *fence, + uint64_t value, uint64_t *completed_value) +{ + struct vkd3d_signaled_semaphore *semaphore; + struct vkd3d_signaled_semaphore *current; + uint64_t semaphore_value; + int rc; + + TRACE("fence %p, value %#"PRIx64".\n", fence, value); + + if ((rc = pthread_mutex_lock(&fence->mutex))) + { + ERR("Failed to lock mutex, error %d.\n", rc); + return VK_NULL_HANDLE; + } + + semaphore = NULL; + semaphore_value = ~(uint64_t)0; + + LIST_FOR_EACH_ENTRY(current, &fence->semaphores, struct vkd3d_signaled_semaphore, entry) + { + /* Prefer a semaphore with the smallest value. */ + if (!current->is_acquired && current->value >= value && semaphore_value >= current->value) + { + semaphore = current; + semaphore_value = current->value; + } + if (semaphore_value == value) + break; + } + + if (semaphore) + semaphore->is_acquired = true; + + *completed_value = fence->value; + + pthread_mutex_unlock(&fence->mutex); + + return semaphore; +} + +static void d3d12_fence_remove_vk_semaphore(struct d3d12_fence *fence, struct vkd3d_signaled_semaphore *semaphore) +{ + int rc; + + if ((rc = pthread_mutex_lock(&fence->mutex))) + { + ERR("Failed to lock mutex, error %d.\n", rc); + return; + } + + assert(semaphore->is_acquired); + + list_remove(&semaphore->entry); + vkd3d_free(semaphore); + + --fence->semaphore_count; + + pthread_mutex_unlock(&fence->mutex); +} + +static void d3d12_fence_release_vk_semaphore(struct d3d12_fence *fence, struct vkd3d_signaled_semaphore *semaphore) +{ + int rc; + + if ((rc = pthread_mutex_lock(&fence->mutex))) + { + ERR("Failed to lock mutex, error %d.\n", rc); + return; + } + + assert(semaphore->is_acquired); + semaphore->is_acquired = false; + + pthread_mutex_unlock(&fence->mutex); +} + static HRESULT d3d12_fence_add_vk_semaphore(struct d3d12_fence *fence, VkSemaphore vk_semaphore, VkFence vk_fence, uint64_t value) { @@ -595,6 +673,7 @@ static HRESULT d3d12_fence_add_vk_semaphore(struct d3d12_fence *fence, semaphore->value = value; semaphore->vk_semaphore = vk_semaphore; semaphore->vk_fence = vk_fence; + semaphore->is_acquired = false; list_add_tail(&fence->semaphores, &semaphore->entry); ++fence->semaphore_count; @@ -5271,11 +5350,90 @@ fail: } static HRESULT STDMETHODCALLTYPE d3d12_command_queue_Wait(ID3D12CommandQueue *iface, - ID3D12Fence *fence, UINT64 value) + ID3D12Fence *fence_iface, UINT64 value) { - FIXME("iface %p, fence %p, value %#"PRIx64" stub!\n", iface, fence, value); + static const VkPipelineStageFlagBits wait_stage_mask = VK_PIPELINE_STAGE_ALL_COMMANDS_BIT; + struct d3d12_command_queue *command_queue = impl_from_ID3D12CommandQueue(iface); + const struct vkd3d_vk_device_procs *vk_procs; + struct vkd3d_signaled_semaphore *semaphore; + VkQueue vk_queue = VK_NULL_HANDLE; + uint64_t completed_value = 0; + struct vkd3d_queue *queue; + struct d3d12_fence *fence; + VkSubmitInfo submit_info; + VkResult vr; + HRESULT hr; + + TRACE("iface %p, fence %p, value %#"PRIx64".\n", iface, fence_iface, value); + + vk_procs = &command_queue->device->vk_procs; + queue = command_queue->vkd3d_queue; + + fence = unsafe_impl_from_ID3D12Fence(fence_iface); + + if (!(semaphore = d3d12_fence_acquire_vk_semaphore(fence, value, &completed_value))) + { + if (completed_value >= value) + { + TRACE("Already signaled %p, value %#"PRIx64".\n", fence, completed_value); + } + else + { + FIXME("Failed to acquire Vulkan semaphore for fence %p, value %#"PRIx64", completed value %#"PRIx64".\n", + fence, value, completed_value); + } + + return S_OK; + } + + submit_info.sType = VK_STRUCTURE_TYPE_SUBMIT_INFO; + submit_info.pNext = NULL; + submit_info.waitSemaphoreCount = 1; + submit_info.pWaitSemaphores = &semaphore->vk_semaphore; + submit_info.pWaitDstStageMask = &wait_stage_mask; + submit_info.commandBufferCount = 0; + submit_info.pCommandBuffers = NULL; + submit_info.signalSemaphoreCount = 0; + submit_info.pSignalSemaphores = NULL; + + if (!(vk_queue = vkd3d_queue_acquire(queue))) + { + ERR("Failed to acquire queue %p.\n", queue); + hr = E_FAIL; + goto fail; + } + + if (!vkd3d_array_reserve((void **)&queue->semaphores, &queue->semaphores_size, + queue->semaphore_count + 1, sizeof(*queue->semaphores))) + { + ERR("Failed to allocate memory for semaphore.\n"); + vkd3d_queue_release(queue); + hr = E_OUTOFMEMORY; + goto fail; + } + if ((vr = VK_CALL(vkQueueSubmit(vk_queue, 1, &submit_info, VK_NULL_HANDLE))) >= 0) + { + queue->semaphores[queue->semaphore_count].vk_semaphore = semaphore->vk_semaphore; + queue->semaphores[queue->semaphore_count].sequence_number = queue->submitted_sequence_number + 1; + ++queue->semaphore_count; + } + + vkd3d_queue_release(queue); + + if (vr < 0) + { + WARN("Failed to submit wait operation, vr %d.\n", vr); + hr = hresult_from_vk_result(vr); + goto fail; + } + + d3d12_fence_remove_vk_semaphore(fence, semaphore); return S_OK; + +fail: + d3d12_fence_release_vk_semaphore(fence, semaphore); + return hr; } static HRESULT STDMETHODCALLTYPE d3d12_command_queue_GetTimestampFrequency(ID3D12CommandQueue *iface, diff --git a/libs/vkd3d/vkd3d_private.h b/libs/vkd3d/vkd3d_private.h index f934d0b..349591d 100644 --- a/libs/vkd3d/vkd3d_private.h +++ b/libs/vkd3d/vkd3d_private.h @@ -287,6 +287,7 @@ struct vkd3d_signaled_semaphore uint64_t value; VkSemaphore vk_semaphore; VkFence vk_fence; + bool is_acquired; }; /* ID3D12Fence */

1 0

Józef Kucia : tests: Add simple test for graphics and compute queue synchronization.
by Alexandre Julliard 03 May '19

03 May '19

Module: vkd3d Branch: master Commit: 1bfe8a2c5415e41ab1a35de01448b4c6972dd41e URL: https://source.winehq.org/git/vkd3d.git/?a=commit;h=1bfe8a2c5415e41ab1a35de… Author: Józef Kucia <jkucia(a)codeweavers.com> Date: Thu May 2 16:02:42 2019 +0200 tests: Add simple test for graphics and compute queue synchronization. Signed-off-by: Józef Kucia <jkucia(a)codeweavers.com> Signed-off-by: Henri Verbeet <hverbeet(a)codeweavers.com> Signed-off-by: Alexandre Julliard <julliard(a)winehq.org> --- tests/d3d12.c | 225 ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 225 insertions(+) Diff: https://source.winehq.org/git/vkd3d.git/?a=commitdiff;h=1bfe8a2c5415e41ab1a…

1 0

Józef Kucia : readme: Update build requirements.
by Alexandre Julliard 03 May '19

03 May '19

Module: vkd3d Branch: master Commit: 9e2ecaf7f4625e56ae388fb27587341636047172 URL: https://source.winehq.org/git/vkd3d.git/?a=commit;h=9e2ecaf7f4625e56ae388fb… Author: Józef Kucia <jkucia(a)codeweavers.com> Date: Thu May 2 16:02:35 2019 +0200 readme: Update build requirements. Signed-off-by: Józef Kucia <jkucia(a)codeweavers.com> Signed-off-by: Henri Verbeet <hverbeet(a)codeweavers.com> Signed-off-by: Alexandre Julliard <julliard(a)winehq.org> --- README | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-) diff --git a/README b/README index bec91a9..be329e1 100644 --- a/README +++ b/README @@ -9,14 +9,14 @@ similar, but not identical, to Direct3D 12. Building vkd3d ============== -Vkd3d depends on SPIRV-Headers and Vulkan-Headers (>= 1.1.88). +Vkd3d depends on SPIRV-Headers and Vulkan-Headers (>= 1.1.101). Vkd3d generates some of its headers from IDL files. If you are using the release tarballs, then these headers are pre-generated and are included. If you are building from git, then they will be generated at build-time using widl. By default, vkd3d will use the widl found in `PATH'. If widl is not -available or is not recent (>= 3.0 or so), then you can build Wine with -`make tools/widl' to avoid building all of Wine. You can then point vkd3d's +available or is not recent (>= 3.20), then you can build Wine with `make +tools/widl' to avoid building all of Wine. You can then point vkd3d's configure at that widl binary with `WIDL="/path/to/widl"'. ===========

1 0

Józef Kucia : vkd3d: Allow simultaneous buffer access from multiple queues.
by Alexandre Julliard 03 May '19

03 May '19

Module: vkd3d Branch: master Commit: 2126614a03b7862a71b0999ef8f1c64e23262f15 URL: https://source.winehq.org/git/vkd3d.git/?a=commit;h=2126614a03b7862a71b0999… Author: Józef Kucia <jkucia(a)codeweavers.com> Date: Thu May 2 16:02:36 2019 +0200 vkd3d: Allow simultaneous buffer access from multiple queues. Signed-off-by: Józef Kucia <jkucia(a)codeweavers.com> Signed-off-by: Henri Verbeet <hverbeet(a)codeweavers.com> Signed-off-by: Alexandre Julliard <julliard(a)winehq.org> --- libs/vkd3d/device.c | 15 ++++++++++++--- libs/vkd3d/resource.c | 8 ++++---- libs/vkd3d/vkd3d_private.h | 7 +++++-- 3 files changed, 21 insertions(+), 9 deletions(-) diff --git a/libs/vkd3d/device.c b/libs/vkd3d/device.c index 052d4f6..41f7ea1 100644 --- a/libs/vkd3d/device.c +++ b/libs/vkd3d/device.c @@ -1449,22 +1449,31 @@ static HRESULT d3d12_device_create_vkd3d_queues(struct d3d12_device *device, device->compute_queue = NULL; device->copy_queue = NULL; - if (FAILED((hr = vkd3d_queue_create(device, direct_family_index, + device->queue_family_count = 0; + memset(device->queue_family_indices, 0, sizeof(device->queue_family_indices)); + + if (SUCCEEDED((hr = vkd3d_queue_create(device, direct_family_index, &queue_info->vk_properties[VKD3D_QUEUE_FAMILY_DIRECT], &device->direct_queue)))) + device->queue_family_indices[device->queue_family_count++] = direct_family_index; + else goto out_destroy_queues; if (compute_family_index == direct_family_index) device->compute_queue = device->direct_queue; - else if (FAILED(hr = vkd3d_queue_create(device, compute_family_index, + else if (SUCCEEDED(hr = vkd3d_queue_create(device, compute_family_index, &queue_info->vk_properties[VKD3D_QUEUE_FAMILY_COMPUTE], &device->compute_queue))) + device->queue_family_indices[device->queue_family_count++] = compute_family_index; + else goto out_destroy_queues; if (transfer_family_index == direct_family_index) device->copy_queue = device->direct_queue; else if (transfer_family_index == compute_family_index) device->copy_queue = device->compute_queue; - else if (FAILED(hr = vkd3d_queue_create(device, transfer_family_index, + else if (SUCCEEDED(hr = vkd3d_queue_create(device, transfer_family_index, &queue_info->vk_properties[VKD3D_QUEUE_FAMILY_TRANSFER], &device->copy_queue))) + device->queue_family_indices[device->queue_family_count++] = transfer_family_index; + else goto out_destroy_queues; return S_OK; diff --git a/libs/vkd3d/resource.c b/libs/vkd3d/resource.c index 8485585..9e6216a 100644 --- a/libs/vkd3d/resource.c +++ b/libs/vkd3d/resource.c @@ -529,10 +529,10 @@ HRESULT vkd3d_create_buffer(struct d3d12_device *device, if (!(desc->Flags & D3D12_RESOURCE_FLAG_DENY_SHADER_RESOURCE)) buffer_info.usage |= VK_BUFFER_USAGE_UNIFORM_TEXEL_BUFFER_BIT; - /* FIXME: Buffers always can be accessed from multiple queues. */ - buffer_info.sharingMode = VK_SHARING_MODE_EXCLUSIVE; - buffer_info.queueFamilyIndexCount = 0; - buffer_info.pQueueFamilyIndices = 0; + /* Buffers always have properties of D3D12_RESOURCE_FLAG_ALLOW_SIMULTANEOUS_ACCESS. */ + buffer_info.sharingMode = VK_SHARING_MODE_CONCURRENT; + buffer_info.queueFamilyIndexCount = device->queue_family_count; + buffer_info.pQueueFamilyIndices = device->queue_family_indices; if ((vr = VK_CALL(vkCreateBuffer(device->vk_device, &buffer_info, NULL, vk_buffer))) < 0) { diff --git a/libs/vkd3d/vkd3d_private.h b/libs/vkd3d/vkd3d_private.h index b333182..0335822 100644 --- a/libs/vkd3d/vkd3d_private.h +++ b/libs/vkd3d/vkd3d_private.h @@ -48,8 +48,9 @@ #define VKD3D_DESCRIPTOR_MAGIC_DSV 0x00565344u #define VKD3D_DESCRIPTOR_MAGIC_RTV 0x00565452u -#define VKD3D_MAX_SHADER_STAGES 5u -#define VKD3D_MAX_VK_SYNC_OBJECTS_PER_D3D12_FENCE 4 +#define VKD3D_MAX_QUEUE_FAMILY_COUNT 3u +#define VKD3D_MAX_SHADER_STAGES 5u +#define VKD3D_MAX_VK_SYNC_OBJECTS_PER_D3D12_FENCE 4u struct d3d12_command_list; struct d3d12_device; @@ -970,6 +971,8 @@ struct d3d12_device struct vkd3d_queue *direct_queue; struct vkd3d_queue *compute_queue; struct vkd3d_queue *copy_queue; + uint32_t queue_family_indices[VKD3D_MAX_QUEUE_FAMILY_COUNT]; + unsigned int queue_family_count; struct vkd3d_instance *vkd3d_instance;

1 0

Józef Kucia : vkd3d: Translate D3D12_RESOURCE_FLAG_ALLOW_SIMULTANEOUS_ACCESS to VK_SHARING_MODE_CONCURRENT.
by Alexandre Julliard 03 May '19

03 May '19

Module: vkd3d Branch: master Commit: 756931f8494ab195f7832b70a23e9cf7e0817310 URL: https://source.winehq.org/git/vkd3d.git/?a=commit;h=756931f8494ab195f7832b7… Author: Józef Kucia <jkucia(a)codeweavers.com> Date: Thu May 2 16:02:37 2019 +0200 vkd3d: Translate D3D12_RESOURCE_FLAG_ALLOW_SIMULTANEOUS_ACCESS to VK_SHARING_MODE_CONCURRENT. Signed-off-by: Józef Kucia <jkucia(a)codeweavers.com> Signed-off-by: Henri Verbeet <hverbeet(a)codeweavers.com> Signed-off-by: Alexandre Julliard <julliard(a)winehq.org> --- libs/vkd3d/resource.c | 17 ++++++++++++----- 1 file changed, 12 insertions(+), 5 deletions(-) diff --git a/libs/vkd3d/resource.c b/libs/vkd3d/resource.c index 9e6216a..ac2ff38 100644 --- a/libs/vkd3d/resource.c +++ b/libs/vkd3d/resource.c @@ -621,11 +621,18 @@ static HRESULT vkd3d_create_image(struct d3d12_device *device, image_info.usage |= VK_IMAGE_USAGE_SAMPLED_BIT; if (desc->Flags & D3D12_RESOURCE_FLAG_ALLOW_SIMULTANEOUS_ACCESS) - FIXME("Ignoring D3D12_RESOURCE_FLAG_ALLOW_SIMULTANEOUS_ACCESS.\n"); - - image_info.sharingMode = VK_SHARING_MODE_EXCLUSIVE; - image_info.queueFamilyIndexCount = 0; - image_info.pQueueFamilyIndices = NULL; + { + TRACE("Creating image with VK_SHARING_MODE_CONCURRENT.\n"); + image_info.sharingMode = VK_SHARING_MODE_CONCURRENT; + image_info.queueFamilyIndexCount = device->queue_family_count; + image_info.pQueueFamilyIndices = device->queue_family_indices; + } + else + { + image_info.sharingMode = VK_SHARING_MODE_EXCLUSIVE; + image_info.queueFamilyIndexCount = 0; + image_info.pQueueFamilyIndices = NULL; + } image_info.initialLayout = is_cpu_accessible_heap(heap_properties) ? VK_IMAGE_LAYOUT_PREINITIALIZED : VK_IMAGE_LAYOUT_UNDEFINED;

1 0

Henri Verbeet : wined3d: Introduce a separate structure for OpenGL context information.
by Alexandre Julliard 02 May '19

02 May '19

Module: wine Branch: master Commit: 5cd787b341673b945af19edfd53685ca657089d1 URL: https://source.winehq.org/git/wine.git/?a=commit;h=5cd787b341673b945af19edf… Author: Henri Verbeet <hverbeet(a)codeweavers.com> Date: Thu May 2 17:28:51 2019 +0430 wined3d: Introduce a separate structure for OpenGL context information. Signed-off-by: Henri Verbeet <hverbeet(a)codeweavers.com> Signed-off-by: Alexandre Julliard <julliard(a)winehq.org> --- dlls/wined3d/context.c | 6 ++++-- dlls/wined3d/wined3d_private.h | 5 +++++ 2 files changed, 9 insertions(+), 2 deletions(-) diff --git a/dlls/wined3d/context.c b/dlls/wined3d/context.c index 5015586..4b6aaa9 100644 --- a/dlls/wined3d/context.c +++ b/dlls/wined3d/context.c @@ -1886,6 +1886,7 @@ struct wined3d_context *context_create(struct wined3d_swapchain *swapchain, struct wined3d_texture *target, const struct wined3d_format *ds_format) { struct wined3d_device *device = swapchain->device; + struct wined3d_context_gl *context_gl; struct wined3d_context *context; DWORD state; @@ -1893,8 +1894,9 @@ struct wined3d_context *context_create(struct wined3d_swapchain *swapchain, wined3d_from_cs(device->cs); - if (!(context = heap_alloc_zero(sizeof(*context)))) + if (!(context_gl = heap_alloc_zero(sizeof(*context_gl)))) return NULL; + context = &context_gl->c; list_init(&context->timestamp_queries); list_init(&context->occlusion_queries); @@ -1977,7 +1979,7 @@ out: wined3d_release_dc(swapchain->win_handle, context->hdc); device->shader_backend->shader_free_context_data(context); device->adapter->fragment_pipe->free_context_data(context); - heap_free(context); + heap_free(context_gl); return NULL; } diff --git a/dlls/wined3d/wined3d_private.h b/dlls/wined3d/wined3d_private.h index daba669..138cefb 100644 --- a/dlls/wined3d/wined3d_private.h +++ b/dlls/wined3d/wined3d_private.h @@ -2029,6 +2029,11 @@ struct wined3d_context unsigned int scissor_rect_count; }; +struct wined3d_context_gl +{ + struct wined3d_context c; +}; + struct wined3d_fb_state { struct wined3d_rendertarget_view *render_targets[MAX_RENDER_TARGET_VIEWS];

1 0

Henri Verbeet : wined3d: Introduce wined3d_context_init().
by Alexandre Julliard 02 May '19

02 May '19

Module: wine Branch: master Commit: cfe9314b646abbde60d1b1d05f92c43f82a27c30 URL: https://source.winehq.org/git/wine.git/?a=commit;h=cfe9314b646abbde60d1b1d0… Author: Henri Verbeet <hverbeet(a)codeweavers.com> Date: Thu May 2 17:28:52 2019 +0430 wined3d: Introduce wined3d_context_init(). Signed-off-by: Henri Verbeet <hverbeet(a)codeweavers.com> Signed-off-by: Alexandre Julliard <julliard(a)winehq.org> --- dlls/wined3d/context.c | 83 ++++++++++++++++++++++++++++++-------------------- 1 file changed, 50 insertions(+), 33 deletions(-) diff --git a/dlls/wined3d/context.c b/dlls/wined3d/context.c index 4b6aaa9..223e6b2 100644 --- a/dlls/wined3d/context.c +++ b/dlls/wined3d/context.c @@ -1882,22 +1882,12 @@ HGLRC context_create_wgl_attribs(const struct wined3d_gl_info *gl_info, HDC hdc, return ctx; } -struct wined3d_context *context_create(struct wined3d_swapchain *swapchain, +static BOOL wined3d_context_init(struct wined3d_context *context, struct wined3d_swapchain *swapchain, struct wined3d_texture *target, const struct wined3d_format *ds_format) { struct wined3d_device *device = swapchain->device; - struct wined3d_context_gl *context_gl; - struct wined3d_context *context; DWORD state; - TRACE("swapchain %p, target %p, window %p.\n", swapchain, target, swapchain->win_handle); - - wined3d_from_cs(device->cs); - - if (!(context_gl = heap_alloc_zero(sizeof(*context_gl)))) - return NULL; - context = &context_gl->c; - list_init(&context->timestamp_queries); list_init(&context->occlusion_queries); list_init(&context->fences); @@ -1910,31 +1900,28 @@ struct wined3d_context *context_create(struct wined3d_swapchain *swapchain, if (!device->shader_backend->shader_allocate_context_data(context)) { ERR("Failed to allocate shader backend context data.\n"); - goto out; + return FALSE; } + if (!device->adapter->fragment_pipe->allocate_context_data(context)) { ERR("Failed to allocate fragment pipeline context data.\n"); - goto out; + device->shader_backend->shader_free_context_data(context); + return FALSE; } if (!(context->hdc = GetDCEx(swapchain->win_handle, 0, DCX_USESTYLE | DCX_CACHE))) { WARN("Failed to retrieve device context, trying swapchain backup.\n"); - if ((context->hdc = swapchain_get_backup_dc(swapchain))) - context->hdc_is_private = TRUE; - else + if (!(context->hdc = swapchain_get_backup_dc(swapchain))) { ERR("Failed to retrieve a device context.\n"); - goto out; + device->shader_backend->shader_free_context_data(context); + device->adapter->fragment_pipe->free_context_data(context); + return FALSE; } - } - - if (!device_context_add(device, context)) - { - ERR("Failed to add the newly created context to the context list\n"); - goto out; + context->hdc_is_private = TRUE; } context->win_handle = swapchain->win_handle; @@ -1956,13 +1943,6 @@ struct wined3d_context *context_create(struct wined3d_swapchain *swapchain, context->current_rt.sub_resource_idx = 0; context->tid = GetCurrentThreadId(); - if (!(device->adapter->adapter_ops->adapter_create_context(context, target, ds_format))) - { - device_context_remove(device, context); - goto out; - } - - device->shader_backend->shader_init_context_state(context); context->shader_update_mask = (1u << WINED3D_SHADER_TYPE_PIXEL) | (1u << WINED3D_SHADER_TYPE_VERTEX) | (1u << WINED3D_SHADER_TYPE_GEOMETRY) @@ -1970,13 +1950,50 @@ struct wined3d_context *context_create(struct wined3d_swapchain *swapchain, | (1u << WINED3D_SHADER_TYPE_DOMAIN) | (1u << WINED3D_SHADER_TYPE_COMPUTE); + return TRUE; +} + +struct wined3d_context *context_create(struct wined3d_swapchain *swapchain, + struct wined3d_texture *target, const struct wined3d_format *ds_format) +{ + struct wined3d_device *device = swapchain->device; + struct wined3d_context_gl *context_gl; + struct wined3d_context *context; + + TRACE("swapchain %p, target %p, window %p.\n", swapchain, target, swapchain->win_handle); + + wined3d_from_cs(device->cs); + + if (!(context_gl = heap_alloc_zero(sizeof(*context_gl)))) + return NULL; + context = &context_gl->c; + + if (!(wined3d_context_init(context, swapchain, target, ds_format))) + { + heap_free(context_gl); + return NULL; + } + + if (!device_context_add(device, context)) + { + ERR("Failed to add the newly created context to the context list\n"); + goto fail; + } + + if (!(device->adapter->adapter_ops->adapter_create_context(context, target, ds_format))) + { + device_context_remove(device, context); + goto fail; + } + + device->shader_backend->shader_init_context_state(context); + TRACE("Created context %p.\n", context); return context; -out: - if (context->hdc) - wined3d_release_dc(swapchain->win_handle, context->hdc); +fail: + wined3d_release_dc(context->win_handle, context->hdc); device->shader_backend->shader_free_context_data(context); device->adapter->fragment_pipe->free_context_data(context); heap_free(context_gl);

1 0

Nikolay Sivov : mfplat: Implement async file stream creation API.
by Alexandre Julliard 02 May '19

02 May '19

Module: wine Branch: master Commit: 1cf021f45eeafe126bd1225dc450b7bab0d2961e URL: https://source.winehq.org/git/wine.git/?a=commit;h=1cf021f45eeafe126bd1225d… Author: Nikolay Sivov <nsivov(a)codeweavers.com> Date: Thu May 2 17:09:07 2019 +0300 mfplat: Implement async file stream creation API. Signed-off-by: Nikolay Sivov <nsivov(a)codeweavers.com> Signed-off-by: Alexandre Julliard <julliard(a)winehq.org> --- dlls/mfplat/main.c | 257 +++++++++++++++++++++++++++++++++++++++++++++ dlls/mfplat/mfplat.spec | 6 +- dlls/mfplat/tests/mfplat.c | 67 ++++++++++++ include/mfapi.h | 4 + 4 files changed, 331 insertions(+), 3 deletions(-) Diff: https://source.winehq.org/git/wine.git/?a=commitdiff;h=1cf021f45eeafe126bd1…

1 0