From: Giovanni Mascellani gmascellani@codeweavers.com
d3d12_command_queue_flush_ops() can renter itself while processing signal events. Since we don't use recursive mutexes, we currently have to check some of the queue variables without holding the mutex, which is not safe.
This is solved by allowing the queue to release its mutex while it is processing entries: when flushing, the queue is briefly locked, the is_flushing flag is set, the queue content is copied away and the queue is unlocked again. After having processed the entries, the queue is locked again to check is something else was added in the meantime. This is repeated until the queue is empty (or a wait operation is blocking it).
This should also remove some latency when a thread pushes to the queue while another one is processing it, but I didn't try to measure any impact. While it is expected that with this patch the queue mutex will be locked and unlocked more frequently, it should also remain locked for less time, hopefully creating little contention. --- libs/vkd3d/command.c | 172 +++++++++++++++++++++++++------------ libs/vkd3d/vkd3d_private.h | 18 +++- 2 files changed, 130 insertions(+), 60 deletions(-)
diff --git a/libs/vkd3d/command.c b/libs/vkd3d/command.c index dc2b184e..0e444c42 100644 --- a/libs/vkd3d/command.c +++ b/libs/vkd3d/command.c @@ -5994,6 +5994,11 @@ static ULONG STDMETHODCALLTYPE d3d12_command_queue_AddRef(ID3D12CommandQueue *if return refcount; }
+static void d3d12_command_queue_op_array_destroy(struct d3d12_command_queue_op_array *array) +{ + vkd3d_free(array->ops); +} + static ULONG STDMETHODCALLTYPE d3d12_command_queue_Release(ID3D12CommandQueue *iface) { struct d3d12_command_queue *command_queue = impl_from_ID3D12CommandQueue(iface); @@ -6008,7 +6013,8 @@ static ULONG STDMETHODCALLTYPE d3d12_command_queue_Release(ID3D12CommandQueue *i vkd3d_fence_worker_stop(&command_queue->fence_worker, device);
vkd3d_mutex_destroy(&command_queue->op_mutex); - vkd3d_free(command_queue->ops); + d3d12_command_queue_op_array_destroy(&command_queue->op_queue); + d3d12_command_queue_op_array_destroy(&command_queue->aux_op_queue);
vkd3d_private_store_destroy(&command_queue->private_store);
@@ -6079,14 +6085,6 @@ static HRESULT STDMETHODCALLTYPE d3d12_command_queue_GetDevice(ID3D12CommandQueu return d3d12_device_query_interface(command_queue->device, iid, device); }
-static struct vkd3d_cs_op_data *d3d12_command_queue_require_space_locked(struct d3d12_command_queue *queue) -{ - if (!vkd3d_array_reserve((void **)&queue->ops, &queue->ops_size, queue->ops_count + 1, sizeof(*queue->ops))) - return NULL; - - return &queue->ops[queue->ops_count++]; -} - static void STDMETHODCALLTYPE d3d12_command_queue_UpdateTileMappings(ID3D12CommandQueue *iface, ID3D12Resource *resource, UINT region_count, const D3D12_TILED_RESOURCE_COORDINATE *region_start_coordinates, const D3D12_TILE_REGION_SIZE *region_sizes, @@ -6143,22 +6141,34 @@ static void d3d12_command_queue_execute(struct d3d12_command_queue *command_queu vkd3d_free(buffers); }
+static bool d3d12_command_queue_op_array_append(struct d3d12_command_queue_op_array *array, + size_t count, const struct vkd3d_cs_op_data *new_ops) +{ + if (!vkd3d_array_reserve((void **)&array->ops, &array->size, array->count + count, sizeof(*array->ops))) + { + ERR("Cannot reserve memory for %zu new ops.\n", count); + return false; + } + + memcpy(&array->ops[array->count], new_ops, count * sizeof(*array->ops)); + array->count += count; + + return true; +} + static HRESULT d3d12_command_queue_enqueue_op(struct d3d12_command_queue *queue, const struct vkd3d_cs_op_data *op) { - struct vkd3d_cs_op_data *new_op; bool flushed_any = false;
vkd3d_mutex_lock(&queue->op_mutex);
- if (!(new_op = d3d12_command_queue_require_space_locked(queue))) + if (!d3d12_command_queue_op_array_append(&queue->op_queue, 1, op)) { vkd3d_mutex_unlock(&queue->op_mutex); return E_OUTOFMEMORY; }
- *new_op = *op; - - if (queue->ops_count == 1) + if (queue->op_queue.count == 1 && !queue->is_flushing) return d3d12_command_queue_flush_ops_locked(queue, &flushed_any);
vkd3d_mutex_unlock(&queue->op_mutex); @@ -6723,18 +6733,51 @@ static const struct ID3D12CommandQueueVtbl d3d12_command_queue_vtbl = d3d12_command_queue_GetDesc, };
+static void d3d12_command_queue_swap_queues(struct d3d12_command_queue *queue) +{ + struct d3d12_command_queue_op_array array; + + array = queue->op_queue; + queue->op_queue = queue->aux_op_queue; + queue->aux_op_queue = array; +} + +static HRESULT d3d12_command_queue_fixup_after_flush(struct d3d12_command_queue *queue, unsigned int done_count) +{ + HRESULT hr; + + queue->aux_op_queue.count -= done_count; + memmove(queue->aux_op_queue.ops, &queue->aux_op_queue.ops[done_count], + queue->aux_op_queue.count * sizeof(*queue->aux_op_queue.ops)); + + vkd3d_mutex_lock(&queue->op_mutex); + + d3d12_command_queue_swap_queues(queue); + + d3d12_command_queue_op_array_append(&queue->op_queue, queue->aux_op_queue.count, queue->aux_op_queue.ops); + + queue->aux_op_queue.count = 0; + queue->is_flushing = false; + + hr = d3d12_command_queue_record_as_blocked(queue); + + vkd3d_mutex_unlock(&queue->op_mutex); + + return hr; +} + static HRESULT d3d12_command_queue_flush_ops(struct d3d12_command_queue *queue, bool *flushed_any) { - if (!queue->ops_count) - return S_OK; + vkd3d_mutex_lock(&queue->op_mutex);
/* This function may be re-entered when invoking * d3d12_command_queue_signal(). The first call is responsible * for re-adding the queue to the flush list. */ if (queue->is_flushing) + { + vkd3d_mutex_unlock(&queue->op_mutex); return S_OK; - - vkd3d_mutex_lock(&queue->op_mutex); + }
return d3d12_command_queue_flush_ops_locked(queue, flushed_any); } @@ -6744,55 +6787,68 @@ static HRESULT d3d12_command_queue_flush_ops_locked(struct d3d12_command_queue * { struct vkd3d_cs_op_data *op; struct d3d12_fence *fence; - HRESULT hr = S_OK; unsigned int i;
- /* Currently only required for d3d12_command_queue_signal(), but set it here anyway. */ queue->is_flushing = true;
- for (i = 0; i < queue->ops_count; ++i) + assert(queue->aux_op_queue.count == 0); + + while (queue->op_queue.count != 0) { - op = &queue->ops[i]; - switch (op->opcode) + d3d12_command_queue_swap_queues(queue); + + vkd3d_mutex_unlock(&queue->op_mutex); + + for (i = 0; i < queue->aux_op_queue.count; ++i) { - case VKD3D_CS_OP_WAIT: - fence = op->u.wait.fence; - vkd3d_mutex_lock(&fence->mutex); - if (op->u.wait.value > fence->max_pending_value) - { - vkd3d_mutex_unlock(&fence->mutex); - queue->ops_count -= i; - memmove(queue->ops, op, queue->ops_count * sizeof(*op)); - hr = d3d12_command_queue_record_as_blocked(queue); - goto done; - } - d3d12_command_queue_wait_locked(queue, fence, op->u.wait.value); - d3d12_fence_decref(fence); - break; + op = &queue->aux_op_queue.ops[i]; + switch (op->opcode) + { + case VKD3D_CS_OP_WAIT: + fence = op->u.wait.fence; + vkd3d_mutex_lock(&fence->mutex); + if (op->u.wait.value > fence->max_pending_value) + { + vkd3d_mutex_unlock(&fence->mutex); + return d3d12_command_queue_fixup_after_flush(queue, i); + } + d3d12_command_queue_wait_locked(queue, fence, op->u.wait.value); + d3d12_fence_decref(fence); + break;
- case VKD3D_CS_OP_SIGNAL: - d3d12_command_queue_signal(queue, op->u.signal.fence, op->u.signal.value); - d3d12_fence_decref(op->u.signal.fence); - break; + case VKD3D_CS_OP_SIGNAL: + d3d12_command_queue_signal(queue, op->u.signal.fence, op->u.signal.value); + d3d12_fence_decref(op->u.signal.fence); + break;
- case VKD3D_CS_OP_EXECUTE: - d3d12_command_queue_execute(queue, op->u.execute.buffers, op->u.execute.buffer_count); - break; + case VKD3D_CS_OP_EXECUTE: + d3d12_command_queue_execute(queue, op->u.execute.buffers, op->u.execute.buffer_count); + break;
- default: - FIXME("Unhandled op type %u.\n", op->opcode); - break; + default: + vkd3d_unreachable(); + } + + *flushed_any |= true; } - *flushed_any |= true; - }
- queue->ops_count = 0; + queue->aux_op_queue.count = 0; + + vkd3d_mutex_lock(&queue->op_mutex); + }
-done: queue->is_flushing = false;
vkd3d_mutex_unlock(&queue->op_mutex); - return hr; + + return S_OK; +} + +static void d3d12_command_queue_op_array_init(struct d3d12_command_queue_op_array *array) +{ + array->ops = NULL; + array->count = 0; + array->size = 0; }
static HRESULT d3d12_command_queue_init(struct d3d12_command_queue *queue, @@ -6813,11 +6869,11 @@ static HRESULT d3d12_command_queue_init(struct d3d12_command_queue *queue, queue->last_waited_fence = NULL; queue->last_waited_fence_value = 0;
- queue->ops = NULL; - queue->ops_count = 0; - queue->ops_size = 0; + d3d12_command_queue_op_array_init(&queue->op_queue); queue->is_flushing = false;
+ d3d12_command_queue_op_array_init(&queue->aux_op_queue); + if (desc->Priority == D3D12_COMMAND_QUEUE_PRIORITY_GLOBAL_REALTIME) { FIXME("Global realtime priority is not implemented.\n"); @@ -6881,8 +6937,10 @@ VkQueue vkd3d_acquire_vk_queue(ID3D12CommandQueue *queue) struct d3d12_command_queue *d3d12_queue = impl_from_ID3D12CommandQueue(queue); VkQueue vk_queue = vkd3d_queue_acquire(d3d12_queue->vkd3d_queue);
- if (d3d12_queue->ops_count) - WARN("Acquired command queue %p with %zu remaining ops.\n", d3d12_queue, d3d12_queue->ops_count); + if (d3d12_queue->op_queue.count) + WARN("Acquired command queue %p with %zu remaining ops.\n", d3d12_queue, d3d12_queue->op_queue.count); + else if (d3d12_queue->is_flushing) + WARN("Acquired command queue %p which is flushing.\n", d3d12_queue);
return vk_queue; } diff --git a/libs/vkd3d/vkd3d_private.h b/libs/vkd3d/vkd3d_private.h index 56677648..075b67c2 100644 --- a/libs/vkd3d/vkd3d_private.h +++ b/libs/vkd3d/vkd3d_private.h @@ -1370,6 +1370,13 @@ struct vkd3d_cs_op_data } u; };
+struct d3d12_command_queue_op_array +{ + struct vkd3d_cs_op_data *ops; + size_t count; + size_t size; +}; + /* ID3D12CommandQueue */ struct d3d12_command_queue { @@ -1387,11 +1394,16 @@ 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; + + /* These fields are protected by op_mutex. */ + struct d3d12_command_queue_op_array op_queue; bool is_flushing;
+ /* This field is not protected by op_mutex, but can only be used + * by the thread that set is_flushing; when is_flushing is not + * set, aux_op_queue.count must be zero. */ + struct d3d12_command_queue_op_array aux_op_queue; + struct vkd3d_private_store private_store; };