[PATCH 0/2] MR7773: wined3d: Use temporary buffer when stretching a surface to itself with cpu blit.

[Elizabeth Figura (＠zfigura)]

Resend of 7428 with rewritten tests and a few minor formatting tweaks.

[PATCH 1/2] wined3d: Use temporary buffer when stretching a surface to itself with cpu blit.

[Yuri Hérouard]

From: Yuri Hérouard

Wine-Bug: https://bugs.winehq.org/show_bug.cgi?id=57283
---
 dlls/wined3d/surface.c | 29 +++++++++++++++++++++++++++++
 1 file changed, 29 insertions(+)

diff --git a/dlls/wined3d/surface.c b/dlls/wined3d/surface.c
index 14cfebd7074..067b3afc30e 100644
--- a/dlls/wined3d/surface.c
+++ b/dlls/wined3d/surface.c
@@ -609,6 +609,7 @@ static HRESULT surface_cpu_blt(struct wined3d_texture *dst_texture, unsigned int
     struct wined3d_context *context;
     struct wined3d_range dst_range;
     unsigned int texture_level;
+    BYTE *tmp_buffer = NULL;
     HRESULT hr = WINED3D_OK;
     BOOL same_sub_resource;
     BOOL upload = FALSE;
@@ -808,18 +809,43 @@ static HRESULT surface_cpu_blt(struct wined3d_texture *dst_texture, unsigned int
             else
             {
                 /* Stretching in y direction only. */
+
+                if (same_sub_resource)
+                {
+                    /* Use a temporary buffer if blitting a surface to itself. */
+                    if ((tmp_buffer = malloc(src_height * src_map.row_pitch)))
+                    {
+                        memcpy(tmp_buffer, sbase, src_height * src_map.row_pitch);
+                        sbase = tmp_buffer;
+                    }
+                }
+
                 for (y = sy = 0; y < dst_height; ++y, sy += yinc)
                 {
                     sbuf = sbase + (sy >> 16) * src_map.row_pitch;
                     memcpy(dbuf, sbuf, row_byte_count);
                     dbuf += dst_map.row_pitch;
                 }
+
+                if (same_sub_resource)
+                    free(tmp_buffer);
             }
         }
         else
         {
             /* Stretching in X direction. */
             unsigned int last_sy = ~0u;
+
+            if (same_sub_resource)
+            {
+                /* Use a temporary buffer if blitting a surface to itself. */
+                if ((tmp_buffer = malloc(src_height * src_map.row_pitch)))
+                {
+                    memcpy(tmp_buffer, sbase, src_height * src_map.row_pitch);
+                    sbase = tmp_buffer;
+                }
+            }
+
             for (y = sy = 0; y < dst_height; ++y, sy += yinc)
             {
                 sbuf = sbase + (sy >> 16) * src_map.row_pitch;
@@ -878,6 +904,9 @@ do { \
                 dbuf += dst_map.row_pitch;
                 last_sy = sy;
             }
+
+            if (same_sub_resource)
+                free(tmp_buffer);
         }
     }
     else

[PATCH 2/2] ddraw/tests: Test a stretched blit to self with overlap.

[Elizabeth Figura]

From: Elizabeth Figura

---
 dlls/ddraw/tests/ddraw1.c | 99 +++++++++++++++++++++++++++++++++++++++
 dlls/ddraw/tests/ddraw2.c | 99 +++++++++++++++++++++++++++++++++++++++
 dlls/ddraw/tests/ddraw4.c | 99 +++++++++++++++++++++++++++++++++++++++
 dlls/ddraw/tests/ddraw7.c | 99 +++++++++++++++++++++++++++++++++++++++
 4 files changed, 396 insertions(+)

diff --git a/dlls/ddraw/tests/ddraw1.c b/dlls/ddraw/tests/ddraw1.c
index e62d9283aaa..5b482459d3e 100644
--- a/dlls/ddraw/tests/ddraw1.c
+++ b/dlls/ddraw/tests/ddraw1.c
@@ -16119,6 +16119,104 @@ static void test_yuv_blit(void)
     DestroyWindow(window);
 }

+static void test_blit_to_self(void)
+{
+    DDSURFACEDESC surface_desc = {sizeof(surface_desc)};
+    const unsigned int width = 64, height = 64;
+    IDirectDrawSurface *surface;
+    unsigned int refcount;
+    IDirectDraw *ddraw;
+    HWND window;
+    HRESULT hr;
+
+    struct
+    {
+        RECT src, dst;
+    }
+    tests[] =
+    {
+        {{16, 16, 24, 24}, {18, 18, 26, 26}},
+        {{16, 16, 24, 24}, {14, 14, 22, 22}},
+        {{16, 16, 24, 24}, {12, 12, 28, 28}},
+        {{16, 16, 24, 24}, {18, 18, 22, 22}},
+    };
+
+    window = create_window();
+    ddraw = create_ddraw();
+    hr = IDirectDraw_SetCooperativeLevel(ddraw, window, DDSCL_NORMAL);
+    ok(hr == S_OK, "Got hr %#lx.\n", hr);
+
+    surface_desc.dwFlags = DDSD_WIDTH | DDSD_HEIGHT | DDSD_CAPS | DDSD_PIXELFORMAT;
+    surface_desc.ddsCaps.dwCaps = DDSCAPS_OFFSCREENPLAIN | DDSCAPS_SYSTEMMEMORY;
+    surface_desc.dwWidth = width;
+    surface_desc.dwHeight = height;
+    init_format_b8g8r8x8(&surface_desc.ddpfPixelFormat);
+
+    hr = IDirectDraw_CreateSurface(ddraw, &surface_desc, &surface, NULL);
+    ok(hr == S_OK, "Got hr %#lx.\n", hr);
+
+    for (unsigned int i = 0; i < ARRAY_SIZE(tests); ++i)
+    {
+        hr = IDirectDrawSurface_Lock(surface, NULL, &surface_desc, DDLOCK_WAIT, NULL);
+        ok(hr == S_OK, "Got hr %#lx.\n", hr);
+        for (unsigned int y = 0; y < height; ++y)
+        {
+            for (unsigned int x = 0; x < width; ++x)
+                ((unsigned int *)surface_desc.lpSurface)[y * width + x] = (y << 8) | x;
+        }
+        hr = IDirectDrawSurface_Unlock(surface, NULL);
+        ok(hr == S_OK, "Got hr %#lx.\n", hr);
+
+        hr = IDirectDrawSurface_Blt(surface, &tests[i].dst, surface, &tests[i].src, DDBLT_WAIT, NULL);
+        ok(hr == S_OK, "Got hr %#lx.\n", hr);
+
+        hr = IDirectDrawSurface_Lock(surface, NULL, &surface_desc, DDLOCK_WAIT | DDLOCK_READONLY, NULL);
+        ok(hr == S_OK, "Got hr %#lx.\n", hr);
+        for (unsigned int y = 0; y < height; ++y)
+        {
+            for (unsigned int x = 0; x < width; ++x)
+            {
+                unsigned int colour = ((unsigned int *)surface_desc.lpSurface)[y * width + x];
+
+                if (x >= tests[i].dst.left && x < tests[i].dst.right
+                        && y >= tests[i].dst.top && y < tests[i].dst.bottom)
+                {
+                    unsigned int src_x = (x - tests[i].dst.left)
+                            * (tests[i].src.right - tests[i].src.left)
+                            / (tests[i].dst.right - tests[i].dst.left)
+                            + tests[i].src.left;
+                    unsigned int src_y = (y - tests[i].dst.top)
+                            * (tests[i].src.bottom - tests[i].src.top)
+                            / (tests[i].dst.bottom - tests[i].dst.top)
+                            + tests[i].src.top;
+                    if (colour != ((src_y << 8) | src_x))
+                    {
+                        ok(0, "Got colour 0x%08x at (%u, %u), expected 0x%08x.\n",
+                                colour, x, y, ((src_y << 8) | src_x));
+                        goto out;
+                    }
+                }
+                else
+                {
+                    if (colour != ((y << 8) | x))
+                    {
+                        ok(0, "Got colour 0x%08x at (%u, %u).\n", colour, x, y);
+                        goto out;
+                    }
+                }
+            }
+        }
+out:
+        hr = IDirectDrawSurface_Unlock(surface, NULL);
+        ok(hr == S_OK, "Got hr %#lx.\n", hr);
+    }
+
+    IDirectDrawSurface_Release(surface);
+    refcount = IDirectDraw_Release(ddraw);
+    ok(!refcount, "Device has %u references left.\n", refcount);
+    DestroyWindow(window);
+}
+
 START_TEST(ddraw1)
 {
     DDDEVICEIDENTIFIER identifier;
@@ -16242,4 +16340,5 @@ START_TEST(ddraw1)
     test_multiple_devices();
     test_sysmem_x_channel();
     test_yuv_blit();
+    test_blit_to_self();
 }
diff --git a/dlls/ddraw/tests/ddraw2.c b/dlls/ddraw/tests/ddraw2.c
index 0f11b27c130..d7daa4e989d 100644
--- a/dlls/ddraw/tests/ddraw2.c
+++ b/dlls/ddraw/tests/ddraw2.c
@@ -17151,6 +17151,104 @@ static void test_yuv_blit(void)
     DestroyWindow(window);
 }

+static void test_blit_to_self(void)
+{
+    DDSURFACEDESC surface_desc = {sizeof(surface_desc)};
+    const unsigned int width = 64, height = 64;
+    IDirectDrawSurface *surface;
+    unsigned int refcount;
+    IDirectDraw2 *ddraw;
+    HWND window;
+    HRESULT hr;
+
+    struct
+    {
+        RECT src, dst;
+    }
+    tests[] =
+    {
+        {{16, 16, 24, 24}, {18, 18, 26, 26}},
+        {{16, 16, 24, 24}, {14, 14, 22, 22}},
+        {{16, 16, 24, 24}, {12, 12, 28, 28}},
+        {{16, 16, 24, 24}, {18, 18, 22, 22}},
+    };
+
+    window = create_window();
+    ddraw = create_ddraw();
+    hr = IDirectDraw2_SetCooperativeLevel(ddraw, window, DDSCL_NORMAL);
+    ok(hr == S_OK, "Got hr %#lx.\n", hr);
+
+    surface_desc.dwFlags = DDSD_WIDTH | DDSD_HEIGHT | DDSD_CAPS | DDSD_PIXELFORMAT;
+    surface_desc.ddsCaps.dwCaps = DDSCAPS_OFFSCREENPLAIN | DDSCAPS_SYSTEMMEMORY;
+    surface_desc.dwWidth = width;
+    surface_desc.dwHeight = height;
+    init_format_b8g8r8x8(&surface_desc.ddpfPixelFormat);
+
+    hr = IDirectDraw2_CreateSurface(ddraw, &surface_desc, &surface, NULL);
+    ok(hr == S_OK, "Got hr %#lx.\n", hr);
+
+    for (unsigned int i = 0; i < ARRAY_SIZE(tests); ++i)
+    {
+        hr = IDirectDrawSurface_Lock(surface, NULL, &surface_desc, DDLOCK_WAIT, NULL);
+        ok(hr == S_OK, "Got hr %#lx.\n", hr);
+        for (unsigned int y = 0; y < height; ++y)
+        {
+            for (unsigned int x = 0; x < width; ++x)
+                ((unsigned int *)surface_desc.lpSurface)[y * width + x] = (y << 8) | x;
+        }
+        hr = IDirectDrawSurface_Unlock(surface, NULL);
+        ok(hr == S_OK, "Got hr %#lx.\n", hr);
+
+        hr = IDirectDrawSurface_Blt(surface, &tests[i].dst, surface, &tests[i].src, DDBLT_WAIT, NULL);
+        ok(hr == S_OK, "Got hr %#lx.\n", hr);
+
+        hr = IDirectDrawSurface_Lock(surface, NULL, &surface_desc, DDLOCK_WAIT | DDLOCK_READONLY, NULL);
+        ok(hr == S_OK, "Got hr %#lx.\n", hr);
+        for (unsigned int y = 0; y < height; ++y)
+        {
+            for (unsigned int x = 0; x < width; ++x)
+            {
+                unsigned int colour = ((unsigned int *)surface_desc.lpSurface)[y * width + x];
+
+                if (x >= tests[i].dst.left && x < tests[i].dst.right
+                        && y >= tests[i].dst.top && y < tests[i].dst.bottom)
+                {
+                    unsigned int src_x = (x - tests[i].dst.left)
+                            * (tests[i].src.right - tests[i].src.left)
+                            / (tests[i].dst.right - tests[i].dst.left)
+                            + tests[i].src.left;
+                    unsigned int src_y = (y - tests[i].dst.top)
+                            * (tests[i].src.bottom - tests[i].src.top)
+                            / (tests[i].dst.bottom - tests[i].dst.top)
+                            + tests[i].src.top;
+                    if (colour != ((src_y << 8) | src_x))
+                    {
+                        ok(0, "Got colour 0x%08x at (%u, %u), expected 0x%08x.\n",
+                                colour, x, y, ((src_y << 8) | src_x));
+                        goto out;
+                    }
+                }
+                else
+                {
+                    if (colour != ((y << 8) | x))
+                    {
+                        ok(0, "Got colour 0x%08x at (%u, %u).\n", colour, x, y);
+                        goto out;
+                    }
+                }
+            }
+        }
+out:
+        hr = IDirectDrawSurface_Unlock(surface, NULL);
+        ok(hr == S_OK, "Got hr %#lx.\n", hr);
+    }
+
+    IDirectDrawSurface_Release(surface);
+    refcount = IDirectDraw2_Release(ddraw);
+    ok(!refcount, "Device has %u references left.\n", refcount);
+    DestroyWindow(window);
+}
+
 START_TEST(ddraw2)
 {
     DDDEVICEIDENTIFIER identifier;
@@ -17280,4 +17378,5 @@ START_TEST(ddraw2)
     test_d3d_state_reset();
     test_sysmem_x_channel();
     test_yuv_blit();
+    test_blit_to_self();
 }
diff --git a/dlls/ddraw/tests/ddraw4.c b/dlls/ddraw/tests/ddraw4.c
index 4571a5426bb..8345b6b3434 100644
--- a/dlls/ddraw/tests/ddraw4.c
+++ b/dlls/ddraw/tests/ddraw4.c
@@ -20224,6 +20224,104 @@ static void test_yuv_blit(void)
     DestroyWindow(window);
 }

+static void test_blit_to_self(void)
+{
+    DDSURFACEDESC2 surface_desc = {sizeof(surface_desc)};
+    const unsigned int width = 64, height = 64;
+    IDirectDrawSurface4 *surface;
+    unsigned int refcount;
+    IDirectDraw4 *ddraw;
+    HWND window;
+    HRESULT hr;
+
+    struct
+    {
+        RECT src, dst;
+    }
+    tests[] =
+    {
+        {{16, 16, 24, 24}, {18, 18, 26, 26}},
+        {{16, 16, 24, 24}, {14, 14, 22, 22}},
+        {{16, 16, 24, 24}, {12, 12, 28, 28}},
+        {{16, 16, 24, 24}, {18, 18, 22, 22}},
+    };
+
+    window = create_window();
+    ddraw = create_ddraw();
+    hr = IDirectDraw4_SetCooperativeLevel(ddraw, window, DDSCL_NORMAL);
+    ok(hr == S_OK, "Got hr %#lx.\n", hr);
+
+    surface_desc.dwFlags = DDSD_WIDTH | DDSD_HEIGHT | DDSD_CAPS | DDSD_PIXELFORMAT;
+    surface_desc.ddsCaps.dwCaps = DDSCAPS_OFFSCREENPLAIN | DDSCAPS_SYSTEMMEMORY;
+    surface_desc.dwWidth = width;
+    surface_desc.dwHeight = height;
+    init_format_b8g8r8x8(&surface_desc.ddpfPixelFormat);
+
+    hr = IDirectDraw4_CreateSurface(ddraw, &surface_desc, &surface, NULL);
+    ok(hr == S_OK, "Got hr %#lx.\n", hr);
+
+    for (unsigned int i = 0; i < ARRAY_SIZE(tests); ++i)
+    {
+        hr = IDirectDrawSurface4_Lock(surface, NULL, &surface_desc, DDLOCK_WAIT, NULL);
+        ok(hr == S_OK, "Got hr %#lx.\n", hr);
+        for (unsigned int y = 0; y < height; ++y)
+        {
+            for (unsigned int x = 0; x < width; ++x)
+                ((unsigned int *)surface_desc.lpSurface)[y * width + x] = (y << 8) | x;
+        }
+        hr = IDirectDrawSurface4_Unlock(surface, NULL);
+        ok(hr == S_OK, "Got hr %#lx.\n", hr);
+
+        hr = IDirectDrawSurface4_Blt(surface, &tests[i].dst, surface, &tests[i].src, DDBLT_WAIT, NULL);
+        ok(hr == S_OK, "Got hr %#lx.\n", hr);
+
+        hr = IDirectDrawSurface4_Lock(surface, NULL, &surface_desc, DDLOCK_WAIT | DDLOCK_READONLY, NULL);
+        ok(hr == S_OK, "Got hr %#lx.\n", hr);
+        for (unsigned int y = 0; y < height; ++y)
+        {
+            for (unsigned int x = 0; x < width; ++x)
+            {
+                unsigned int colour = ((unsigned int *)surface_desc.lpSurface)[y * width + x];
+
+                if (x >= tests[i].dst.left && x < tests[i].dst.right
+                        && y >= tests[i].dst.top && y < tests[i].dst.bottom)
+                {
+                    unsigned int src_x = (x - tests[i].dst.left)
+                            * (tests[i].src.right - tests[i].src.left)
+                            / (tests[i].dst.right - tests[i].dst.left)
+                            + tests[i].src.left;
+                    unsigned int src_y = (y - tests[i].dst.top)
+                            * (tests[i].src.bottom - tests[i].src.top)
+                            / (tests[i].dst.bottom - tests[i].dst.top)
+                            + tests[i].src.top;
+                    if (colour != ((src_y << 8) | src_x))
+                    {
+                        ok(0, "Got colour 0x%08x at (%u, %u), expected 0x%08x.\n",
+                                colour, x, y, ((src_y << 8) | src_x));
+                        goto out;
+                    }
+                }
+                else
+                {
+                    if (colour != ((y << 8) | x))
+                    {
+                        ok(0, "Got colour 0x%08x at (%u, %u).\n", colour, x, y);
+                        goto out;
+                    }
+                }
+            }
+        }
+out:
+        hr = IDirectDrawSurface4_Unlock(surface, NULL);
+        ok(hr == S_OK, "Got hr %#lx.\n", hr);
+    }
+
+    IDirectDrawSurface4_Release(surface);
+    refcount = IDirectDraw4_Release(ddraw);
+    ok(!refcount, "Device has %u references left.\n", refcount);
+    DestroyWindow(window);
+}
+
 START_TEST(ddraw4)
 {
     DDDEVICEIDENTIFIER identifier;
@@ -20370,4 +20468,5 @@ START_TEST(ddraw4)
     test_d3d_state_reset();
     test_sysmem_x_channel();
     test_yuv_blit();
+    test_blit_to_self();
 }
diff --git a/dlls/ddraw/tests/ddraw7.c b/dlls/ddraw/tests/ddraw7.c
index d4980250a8c..aebca7727b9 100644
--- a/dlls/ddraw/tests/ddraw7.c
+++ b/dlls/ddraw/tests/ddraw7.c
@@ -20659,6 +20659,104 @@ static void test_yuv_blit(void)
     DestroyWindow(window);
 }

+static void test_blit_to_self(void)
+{
+    DDSURFACEDESC2 surface_desc = {sizeof(surface_desc)};
+    const unsigned int width = 64, height = 64;
+    IDirectDrawSurface7 *surface;
+    unsigned int refcount;
+    IDirectDraw7 *ddraw;
+    HWND window;
+    HRESULT hr;
+
+    struct
+    {
+        RECT src, dst;
+    }
+    tests[] =
+    {
+        {{16, 16, 24, 24}, {18, 18, 26, 26}},
+        {{16, 16, 24, 24}, {14, 14, 22, 22}},
+        {{16, 16, 24, 24}, {12, 12, 28, 28}},
+        {{16, 16, 24, 24}, {18, 18, 22, 22}},
+    };
+
+    window = create_window();
+    ddraw = create_ddraw();
+    hr = IDirectDraw7_SetCooperativeLevel(ddraw, window, DDSCL_NORMAL);
+    ok(hr == S_OK, "Got hr %#lx.\n", hr);
+
+    surface_desc.dwFlags = DDSD_WIDTH | DDSD_HEIGHT | DDSD_CAPS | DDSD_PIXELFORMAT;
+    surface_desc.ddsCaps.dwCaps = DDSCAPS_OFFSCREENPLAIN | DDSCAPS_SYSTEMMEMORY;
+    surface_desc.dwWidth = width;
+    surface_desc.dwHeight = height;
+    init_format_b8g8r8x8(&surface_desc.ddpfPixelFormat);
+
+    hr = IDirectDraw7_CreateSurface(ddraw, &surface_desc, &surface, NULL);
+    ok(hr == S_OK, "Got hr %#lx.\n", hr);
+
+    for (unsigned int i = 0; i < ARRAY_SIZE(tests); ++i)
+    {
+        hr = IDirectDrawSurface7_Lock(surface, NULL, &surface_desc, DDLOCK_WAIT, NULL);
+        ok(hr == S_OK, "Got hr %#lx.\n", hr);
+        for (unsigned int y = 0; y < height; ++y)
+        {
+            for (unsigned int x = 0; x < width; ++x)
+                ((unsigned int *)surface_desc.lpSurface)[y * width + x] = (y << 8) | x;
+        }
+        hr = IDirectDrawSurface7_Unlock(surface, NULL);
+        ok(hr == S_OK, "Got hr %#lx.\n", hr);
+
+        hr = IDirectDrawSurface7_Blt(surface, &tests[i].dst, surface, &tests[i].src, DDBLT_WAIT, NULL);
+        ok(hr == S_OK, "Got hr %#lx.\n", hr);
+
+        hr = IDirectDrawSurface7_Lock(surface, NULL, &surface_desc, DDLOCK_WAIT | DDLOCK_READONLY, NULL);
+        ok(hr == S_OK, "Got hr %#lx.\n", hr);
+        for (unsigned int y = 0; y < height; ++y)
+        {
+            for (unsigned int x = 0; x < width; ++x)
+            {
+                unsigned int colour = ((unsigned int *)surface_desc.lpSurface)[y * width + x];
+
+                if (x >= tests[i].dst.left && x < tests[i].dst.right
+                        && y >= tests[i].dst.top && y < tests[i].dst.bottom)
+                {
+                    unsigned int src_x = (x - tests[i].dst.left)
+                            * (tests[i].src.right - tests[i].src.left)
+                            / (tests[i].dst.right - tests[i].dst.left)
+                            + tests[i].src.left;
+                    unsigned int src_y = (y - tests[i].dst.top)
+                            * (tests[i].src.bottom - tests[i].src.top)
+                            / (tests[i].dst.bottom - tests[i].dst.top)
+                            + tests[i].src.top;
+                    if (colour != ((src_y << 8) | src_x))
+                    {
+                        ok(0, "Got colour 0x%08x at (%u, %u), expected 0x%08x.\n",
+                                colour, x, y, ((src_y << 8) | src_x));
+                        goto out;
+                    }
+                }
+                else
+                {
+                    if (colour != ((y << 8) | x))
+                    {
+                        ok(0, "Got colour 0x%08x at (%u, %u).\n", colour, x, y);
+                        goto out;
+                    }
+                }
+            }
+        }
+out:
+        hr = IDirectDrawSurface7_Unlock(surface, NULL);
+        ok(hr == S_OK, "Got hr %#lx.\n", hr);
+    }
+
+    IDirectDrawSurface7_Release(surface);
+    refcount = IDirectDraw7_Release(ddraw);
+    ok(!refcount, "Device has %u references left.\n", refcount);
+    DestroyWindow(window);
+}
+
 START_TEST(ddraw7)
 {
     DDDEVICEIDENTIFIER2 identifier;
@@ -20839,4 +20937,5 @@ START_TEST(ddraw7)
     test_d3d_state_reset();
     test_sysmem_x_channel();
     test_yuv_blit();
+    test_blit_to_self();
 }

[PATCH 0/2] MR7773: wined3d: Use temporary buffer when stretching a surface to itself with cpu blit. - approved

[Elizabeth Figura (＠zfigura)]

This merge request was approved by Elizabeth Figura.

[PATCH 0/2] MR7773: wined3d: Use temporary buffer when stretching a surface to itself with cpu blit. - approved

[Jan Sikorski (＠jsikorski)]

This merge request was approved by Jan Sikorski.