d3d8: Upload sysmem vertex buffers into the streaming buffer in d3d8_device_DrawPrimitive().

Bloodrayne: Terminal Cut (and Bloodrayne 2: Terminal Cut, and probably other games in the series) streams from a SYSTEMMEM index buffer, updating it and drawing from it every frame. This is currently slow on Wine, since each map needs to wait for the previous upload (on the CS) to complete. There are a few ways to avoid waiting, but this patch takes the approach of effectively uploading from the SYSTEMMEM buffer on the client side, while using a dynamic buffer to avoid client/CS synchronization. This brings performance from 20-30 FPS to a (locked) 60, on NVidia GL drivers.

d3d8: Upload sysmem vertex buffers into the streaming buffer in d3d8_device_DrawPrimitive().
fc0f0090 · Zebediah Figura · Alexandre Julliard · 348eebae · fc0f0090 · fc0f0090
Commit fc0f0090 authored Jul 07, 2022 by Zebediah Figura Committed by Alexandre Julliard Jul 13, 2022
Hide whitespace changes
Inline Side-by-side

Showing with 44 additions and 52 deletions

buffer.c dlls/d3d8/buffer.c +5 -24

d3d8_private.h dlls/d3d8/d3d8_private.h +1 -1

device.c dlls/d3d8/device.c +38 -27

No files found.
--- a/dlls/d3d8/buffer.c
+++ b/dlls/d3d8/buffer.c
@@ -377,10 +377,7 @@ static ULONG WINAPI d3d8_indexbuffer_AddRef(IDirect3DIndexBuffer8 *iface)
    if (refcount == 1)
    {
        IDirect3DDevice8_AddRef(buffer->parent_device);
-        if (buffer->draw_buffer)
+        wined3d_buffer_incref(buffer->wined3d_buffer);
-            wined3d_buffer_incref(buffer->draw_buffer);
-        else
-            wined3d_buffer_incref(buffer->wined3d_buffer);
    }
    return refcount;
@@ -395,13 +392,9 @@ static ULONG WINAPI d3d8_indexbuffer_Release(IDirect3DIndexBuffer8 *iface)
    if (!refcount)
    {
-        struct wined3d_buffer *draw_buffer = buffer->draw_buffer;
        IDirect3DDevice8 *device = buffer->parent_device;
-        if (draw_buffer)
+        wined3d_buffer_decref(buffer->wined3d_buffer);
-            wined3d_buffer_decref(draw_buffer);
-        else
-            wined3d_buffer_decref(buffer->wined3d_buffer);
        /* Release the device last, as it may cause the device to be destroyed. */
        IDirect3DDevice8_Release(device);
@@ -584,8 +577,6 @@ static void STDMETHODCALLTYPE d3d8_indexbuffer_wined3d_object_destroyed(void *pa
 {
    struct d3d8_indexbuffer *buffer = parent;
-    if (buffer->draw_buffer)
-        wined3d_buffer_decref(buffer->wined3d_buffer);
    d3d8_resource_cleanup(&buffer->resource);
    heap_free(buffer);
 }
@@ -598,7 +589,6 @@ static const struct wined3d_parent_ops d3d8_indexbuffer_wined3d_parent_ops =
 HRESULT indexbuffer_init(struct d3d8_indexbuffer *buffer, struct d3d8_device *device,
        UINT size, DWORD usage, D3DFORMAT format, D3DPOOL pool)
 {
-    const struct wined3d_parent_ops *parent_ops = &d3d8_null_wined3d_parent_ops;
    struct wined3d_buffer_desc desc;
    HRESULT hr;
@@ -620,26 +610,17 @@ HRESULT indexbuffer_init(struct d3d8_indexbuffer *buffer, struct d3d8_device *de
    desc.structure_byte_stride = 0;
    if (desc.access & WINED3D_RESOURCE_ACCESS_GPU)
-    {
        desc.bind_flags = WINED3D_BIND_INDEX_BUFFER;
-        parent_ops = &d3d8_indexbuffer_wined3d_parent_ops;
-    }
    buffer->IDirect3DIndexBuffer8_iface.lpVtbl = &d3d8_indexbuffer_vtbl;
    d3d8_resource_init(&buffer->resource);
    buffer->format = wined3dformat_from_d3dformat(format);
    buffer->usage = usage;
+    buffer->sysmem = !(desc.access & WINED3D_RESOURCE_ACCESS_GPU);
    wined3d_mutex_lock();
-    hr = wined3d_buffer_create(device->wined3d_device, &desc, NULL, buffer, parent_ops, &buffer->wined3d_buffer);
+    hr = wined3d_buffer_create(device->wined3d_device, &desc, NULL, buffer,
-    if (SUCCEEDED(hr) && !(desc.access & WINED3D_RESOURCE_ACCESS_GPU))
+            &d3d8_indexbuffer_wined3d_parent_ops, &buffer->wined3d_buffer);
-    {
-        desc.bind_flags = WINED3D_BIND_INDEX_BUFFER;
-        desc.access = WINED3D_RESOURCE_ACCESS_GPU;
-        if (FAILED(hr = wined3d_buffer_create(device->wined3d_device, &desc, NULL, buffer,
-                &d3d8_indexbuffer_wined3d_parent_ops, &buffer->draw_buffer)))
-            wined3d_buffer_decref(buffer->wined3d_buffer);
-    }
    wined3d_mutex_unlock();
    if (FAILED(hr))
    {

--- a/dlls/d3d8/d3d8_private.h
+++ b/dlls/d3d8/d3d8_private.h
@@ -229,9 +229,9 @@ struct d3d8_indexbuffer
    struct d3d8_resource resource;
    struct wined3d_buffer *wined3d_buffer;
    IDirect3DDevice8 *parent_device;
-    struct wined3d_buffer *draw_buffer;
    enum wined3d_format_id format;
    DWORD usage;
+    bool sysmem;
 };
 HRESULT indexbuffer_init(struct d3d8_indexbuffer *buffer, struct d3d8_device *device,

--- a/dlls/d3d8/device.c
+++ b/dlls/d3d8/device.c
@@ -913,7 +913,7 @@ static HRESULT CDECL reset_enum_callback(struct wined3d_resource *resource)
        if (desc.bind_flags & WINED3D_BIND_INDEX_BUFFER)
        {
            index_buffer = wined3d_resource_get_parent(resource);
-            if (index_buffer && index_buffer->draw_buffer)
+            if (index_buffer && index_buffer->sysmem)
                return D3D_OK;
        }
@@ -2452,34 +2452,41 @@ static void d3d8_device_upload_sysmem_vertex_buffers(struct d3d8_device *device,
    }
 }
-static void d3d8_device_upload_sysmem_index_buffer(struct d3d8_device *device,
+static HRESULT d3d8_device_upload_sysmem_index_buffer(struct d3d8_device *device,
-        unsigned int start_idx, unsigned int idx_count)
+        unsigned int *start_idx, unsigned int idx_count)
 {
    const struct wined3d_stateblock_state *state = device->stateblock_state;
-    struct wined3d_box box = {0, 0, 0, 1, 0, 1};
+    unsigned int src_offset, idx_size, buffer_size, pos;
-    struct wined3d_resource *dst_resource;
+    struct wined3d_resource_desc resource_desc;
-    struct d3d8_indexbuffer *d3d8_buffer;
+    struct wined3d_resource *index_buffer;
-    struct wined3d_buffer *dst_buffer;
+    struct wined3d_map_desc map_desc;
-    struct wined3d_resource_desc desc;
+    struct wined3d_box box;
-    enum wined3d_format_id format;
-    unsigned int idx_size;
    HRESULT hr;
    if (!device->sysmem_ib)
-        return;
+        return S_OK;
-    dst_buffer = state->index_buffer;
+    index_buffer = wined3d_buffer_get_resource(state->index_buffer);
-    format = state->index_format;
+    wined3d_resource_get_desc(index_buffer, &resource_desc);
-    d3d8_buffer = wined3d_buffer_get_parent(dst_buffer);
+    idx_size = (state->index_format == WINED3DFMT_R16_UINT) ? 2 : 4;
-    dst_resource = wined3d_buffer_get_resource(dst_buffer);
-    wined3d_resource_get_desc(dst_resource, &desc);
+    src_offset = (*start_idx) * idx_size;
-    idx_size = format == WINED3DFMT_R16_UINT ? 2 : 4;
+    buffer_size = min(idx_count * idx_size, resource_desc.size - src_offset);
-    box.left = start_idx * idx_size;
-    box.right = min(box.left + idx_count * idx_size, desc.size);
+    wined3d_box_set(&box, src_offset, 0, buffer_size, 1, 0, 1);
-    if (FAILED(hr = wined3d_device_context_copy_sub_resource_region(device->immediate_context,
+    if (FAILED(hr = wined3d_resource_map(index_buffer, 0, &map_desc, &box, WINED3D_MAP_READ)))
-            dst_resource, 0, box.left, 0, 0,
+    {
-            wined3d_buffer_get_resource(d3d8_buffer->wined3d_buffer), 0, &box, 0)))
+        wined3d_mutex_unlock();
-        ERR("Failed to update buffer.\n");
+        return hr;
+    }
+    wined3d_streaming_buffer_upload(device->wined3d_device, &device->index_buffer,
+            map_desc.data, buffer_size, idx_size, &pos);
+    wined3d_resource_unmap(index_buffer, 0);
+    wined3d_device_context_set_index_buffer(device->immediate_context,
+            device->index_buffer.buffer, state->index_format, pos);
+    *start_idx = 0;
+    return S_OK;
 }
 static HRESULT WINAPI d3d8_device_DrawPrimitive(IDirect3DDevice8 *iface,
@@ -2510,6 +2517,7 @@ static HRESULT WINAPI d3d8_device_DrawIndexedPrimitive(IDirect3DDevice8 *iface,
    struct d3d8_device *device = impl_from_IDirect3DDevice8(iface);
    unsigned int index_count;
    int base_vertex_index;
+    HRESULT hr;
    TRACE("iface %p, primitive_type %#x, min_vertex_idx %u, vertex_count %u, start_idx %u, primitive_count %u.\n",
            iface, primitive_type, min_vertex_idx, vertex_count, start_idx, primitive_count);
@@ -2524,11 +2532,16 @@ static HRESULT WINAPI d3d8_device_DrawIndexedPrimitive(IDirect3DDevice8 *iface,
    }
    base_vertex_index = device->stateblock_state->base_vertex_index;
    d3d8_device_upload_sysmem_vertex_buffers(device, base_vertex_index + min_vertex_idx, vertex_count);
-    d3d8_device_upload_sysmem_index_buffer(device, start_idx, index_count);
    wined3d_device_context_set_primitive_type(device->immediate_context,
            wined3d_primitive_type_from_d3d(primitive_type), 0);
    wined3d_device_apply_stateblock(device->wined3d_device, device->state);
+    if (FAILED(hr = d3d8_device_upload_sysmem_index_buffer(device, &start_idx, index_count)))
+    {
+        wined3d_mutex_unlock();
+        return hr;
+    }
    wined3d_device_context_draw_indexed(device->immediate_context, base_vertex_index, start_idx, index_count, 0, 0);
    wined3d_mutex_unlock();
    return D3D_OK;
@@ -3022,8 +3035,6 @@ static HRESULT WINAPI d3d8_device_SetIndices(IDirect3DDevice8 *iface,
    if (!ib)
        wined3d_buffer = NULL;
-    else if (ib->draw_buffer)
-        wined3d_buffer = ib->draw_buffer;
    else
        wined3d_buffer = ib->wined3d_buffer;
@@ -3037,7 +3048,7 @@ static HRESULT WINAPI d3d8_device_SetIndices(IDirect3DDevice8 *iface,
    wined3d_stateblock_set_base_vertex_index(device->update_state, base_vertex_idx);
    wined3d_stateblock_set_index_buffer(device->update_state, wined3d_buffer, ib ? ib->format : WINED3DFMT_UNKNOWN);
    if (!device->recording)
-        device->sysmem_ib = ib && ib->draw_buffer;
+        device->sysmem_ib = ib && ib->sysmem;
    wined3d_mutex_unlock();
    return D3D_OK;