From: Conor McCarthy cmccarthy@codeweavers.com
Provides a simple way to disable Vulkan writes for non-shader-visible heaps. Also there is a chance of avoiding access to the d3d12_device object which helps memory cache performance. --- libs/vkd3d/device.c | 4 +++- libs/vkd3d/resource.c | 25 ++++++++++++++++--------- libs/vkd3d/vkd3d_private.h | 6 +++++- 3 files changed, 24 insertions(+), 11 deletions(-)
diff --git a/libs/vkd3d/device.c b/libs/vkd3d/device.c index b9a8943c..025a60fc 100644 --- a/libs/vkd3d/device.c +++ b/libs/vkd3d/device.c @@ -3414,6 +3414,7 @@ static void STDMETHODCALLTYPE d3d12_device_CopyDescriptors(ID3D12Device *iface, struct d3d12_device *device = impl_from_ID3D12Device(iface); unsigned int dst_range_idx, dst_idx, src_range_idx, src_idx; unsigned int dst_range_size, src_range_size; + struct d3d12_descriptor_heap *dst_heap; const struct d3d12_desc *src; struct d3d12_desc *dst;
@@ -3443,13 +3444,14 @@ static void STDMETHODCALLTYPE d3d12_device_CopyDescriptors(ID3D12Device *iface, src_range_size = src_descriptor_range_sizes ? src_descriptor_range_sizes[src_range_idx] : 1;
dst = d3d12_desc_from_cpu_handle(dst_descriptor_range_offsets[dst_range_idx]); + dst_heap = d3d12_desc_get_descriptor_heap(dst); src = d3d12_desc_from_cpu_handle(src_descriptor_range_offsets[src_range_idx]);
for (; dst_idx < dst_range_size && src_idx < src_range_size; ++dst_idx, ++src_idx) { if (dst[dst_idx].s.u.object == src[src_idx].s.u.object) continue; - d3d12_desc_copy(&dst[dst_idx], &src[src_idx], device); + d3d12_desc_copy(&dst[dst_idx], &src[src_idx], dst_heap, device); }
if (dst_idx >= dst_range_size) diff --git a/libs/vkd3d/resource.c b/libs/vkd3d/resource.c index 4c07d326..72663afe 100644 --- a/libs/vkd3d/resource.c +++ b/libs/vkd3d/resource.c @@ -2411,13 +2411,11 @@ void d3d12_desc_flush_vk_heap_updates_locked(struct d3d12_descriptor_heap *descr descriptor_writes_free_object_refs(&writes, device); }
-static void d3d12_desc_mark_as_modified(struct d3d12_desc *dst) +static void d3d12_desc_mark_as_modified(struct d3d12_desc *dst, struct d3d12_descriptor_heap *descriptor_heap) { - struct d3d12_descriptor_heap *descriptor_heap; unsigned int i, head;
i = dst->index; - descriptor_heap = d3d12_desc_get_descriptor_heap(dst); head = descriptor_heap->dirty_list_head;
/* Only one thread can swap the value away from zero. */ @@ -2431,14 +2429,20 @@ static void d3d12_desc_mark_as_modified(struct d3d12_desc *dst) } }
-void d3d12_desc_write_atomic(struct d3d12_desc *dst, const struct d3d12_desc *src, - struct d3d12_device *device) +static inline void descriptor_heap_write_atomic(struct d3d12_descriptor_heap *descriptor_heap, struct d3d12_desc *dst, + const struct d3d12_desc *src, struct d3d12_device *device) { void *object = src->s.u.object;
d3d12_desc_replace(dst, object, device); - if (device->use_vk_heaps && object && !dst->next) - d3d12_desc_mark_as_modified(dst); + if (descriptor_heap->use_vk_heaps && object && !dst->next) + d3d12_desc_mark_as_modified(dst, descriptor_heap); +} + +void d3d12_desc_write_atomic(struct d3d12_desc *dst, const struct d3d12_desc *src, + struct d3d12_device *device) +{ + descriptor_heap_write_atomic(d3d12_desc_get_descriptor_heap(dst), dst, src, device); }
static void d3d12_desc_destroy(struct d3d12_desc *descriptor, struct d3d12_device *device) @@ -2446,7 +2450,9 @@ static void d3d12_desc_destroy(struct d3d12_desc *descriptor, struct d3d12_devic d3d12_desc_replace(descriptor, NULL, device); }
-void d3d12_desc_copy(struct d3d12_desc *dst, const struct d3d12_desc *src, +/* This is a major performance bottleneck for some games, so do not load the device + * pointer from dst_heap. In some cases device will not be used. */ +void d3d12_desc_copy(struct d3d12_desc *dst, const struct d3d12_desc *src, struct d3d12_descriptor_heap *dst_heap, struct d3d12_device *device) { struct d3d12_desc tmp; @@ -2454,7 +2460,7 @@ void d3d12_desc_copy(struct d3d12_desc *dst, const struct d3d12_desc *src, assert(dst != src);
tmp.s.u.object = d3d12_desc_get_object_ref(src, device); - d3d12_desc_write_atomic(dst, &tmp, device); + descriptor_heap_write_atomic(dst_heap, dst, &tmp, device); }
static VkDeviceSize vkd3d_get_required_texel_buffer_alignment(const struct d3d12_device *device, @@ -3987,6 +3993,7 @@ static HRESULT d3d12_descriptor_heap_init(struct d3d12_descriptor_heap *descript if (FAILED(hr = vkd3d_private_store_init(&descriptor_heap->private_store))) return hr;
+ descriptor_heap->use_vk_heaps = device->use_vk_heaps; d3d12_descriptor_heap_vk_descriptor_sets_init(descriptor_heap, device, desc); vkd3d_mutex_init(&descriptor_heap->vk_sets_mutex);
diff --git a/libs/vkd3d/vkd3d_private.h b/libs/vkd3d/vkd3d_private.h index c5259420..53a9620b 100644 --- a/libs/vkd3d/vkd3d_private.h +++ b/libs/vkd3d/vkd3d_private.h @@ -895,7 +895,10 @@ static inline void d3d12_desc_copy_raw(struct d3d12_desc *dst, const struct d3d1 dst->s = src->s; }
-void d3d12_desc_copy(struct d3d12_desc *dst, const struct d3d12_desc *src, struct d3d12_device *device); +struct d3d12_descriptor_heap; + +void d3d12_desc_copy(struct d3d12_desc *dst, const struct d3d12_desc *src, struct d3d12_descriptor_heap *dst_heap, + struct d3d12_device *device); void d3d12_desc_create_cbv(struct d3d12_desc *descriptor, struct d3d12_device *device, const D3D12_CONSTANT_BUFFER_VIEW_DESC *desc); void d3d12_desc_create_srv(struct d3d12_desc *descriptor, @@ -998,6 +1001,7 @@ struct d3d12_descriptor_heap D3D12_DESCRIPTOR_HEAP_DESC desc;
struct d3d12_device *device; + bool use_vk_heaps;
struct vkd3d_private_store private_store;