Roderick Colenbrander : wined3d: Fix a bug in BtlFast which was uncovered by the 32-bit dibsection support.

[Alexandre Julliard]

Module: wine
Branch: master
Commit: 6d8c4572efae3b7e8e64100504c10270b2168079
URL:    http://source.winehq.org/git/wine.git/?a=commit;h=6d8c4572efae3b7e8e64100504c10270b2168079

Author: Roderick Colenbrander <thunderbird2k at gmail.com>
Date:   Sat Nov 14 16:30:25 2009 +0100

wined3d: Fix a bug in BtlFast which was uncovered by the 32-bit dibsection support.

---

 dlls/ddraw/tests/dsurface.c |   34 ++++++++++++++++++++++++++++++++++
 dlls/wined3d/surface_base.c |   10 +++++++++-
 2 files changed, 43 insertions(+), 1 deletions(-)

diff --git a/dlls/ddraw/tests/dsurface.c b/dlls/ddraw/tests/dsurface.c
index ab0408b..d53d6a0 100644
--- a/dlls/ddraw/tests/dsurface.c
+++ b/dlls/ddraw/tests/dsurface.c
@@ -304,6 +304,40 @@ static void SrcColorKey32BlitTest(void)
     rc = IDirectDrawSurface_Unlock(lpDst, NULL);
     ok(rc==DD_OK,"Unlock returned: %x\n",rc);
 
+    /* Below we repeat the same test as above but now using BltFast instead of Blt. Before
+     * we can carry out the test we need to restore the color of the destination surface.
+     */
+    rc = IDirectDrawSurface_Lock(lpDst, NULL, &ddsd2, DDLOCK_WAIT, NULL);
+    ok(rc==DD_OK,"Lock returned: %x\n",rc);
+    lpData = ddsd2.lpSurface;
+    lpData[0] = 0xCCCCCCCC;
+    lpData[1] = 0xCCCCCCCC;
+    lpData[2] = 0xCCCCCCCC;
+    lpData[3] = 0xCCCCCCCC;
+    rc = IDirectDrawSurface_Unlock(lpDst, NULL);
+    ok(rc==DD_OK,"Unlock returned: %x\n",rc);
+
+    IDirectDrawSurface_BltFast(lpDst, 0, 0, lpSrc, NULL, DDBLTFAST_SRCCOLORKEY);
+
+    rc = IDirectDrawSurface_Lock(lpDst, NULL, &ddsd2, DDLOCK_WAIT, NULL);
+    ok(rc==DD_OK,"Lock returned: %x\n",rc);
+    ok((ddsd2.dwFlags & DDSD_LPSURFACE) == 0, "Surface desc has LPSURFACE Flags set\n");
+    lpData = ddsd2.lpSurface;
+    /* Different behavior on some drivers / windows versions. Some versions ignore the X channel when
+     * color keying, but copy it to the destination surface. Others apply it for color keying, but
+     * do not copy it into the destination surface.
+     */
+    if(lpData[0]==0x00010203) {
+        trace("X channel was not copied into the destination surface\n");
+        ok((lpData[0]==0x00010203)&&(lpData[1]==0x00010203)&&(lpData[2]==0x00FF00FF)&&(lpData[3]==0xCCCCCCCC),
+           "Destination data after blitting is not correct\n");
+    } else {
+        ok((lpData[0]==0x77010203)&&(lpData[1]==0x00010203)&&(lpData[2]==0xCCCCCCCC)&&(lpData[3]==0xCCCCCCCC),
+           "Destination data after blitting is not correct\n");
+    }
+    rc = IDirectDrawSurface_Unlock(lpDst, NULL);
+    ok(rc==DD_OK,"Unlock returned: %x\n",rc);
+
     /* Also test SetColorKey */
     IDirectDrawSurface_GetColorKey(lpSrc, DDCKEY_SRCBLT, &DDColorKey);
     ok(DDColorKey.dwColorSpaceLowValue == 0xFF00FF && DDColorKey.dwColorSpaceHighValue == 0xFF00FF,
diff --git a/dlls/wined3d/surface_base.c b/dlls/wined3d/surface_base.c
index c737f99..dc29ad2 100644
--- a/dlls/wined3d/surface_base.c
+++ b/dlls/wined3d/surface_base.c
@@ -1695,6 +1695,14 @@ HRESULT WINAPI IWineD3DBaseSurfaceImpl_BltFast(IWineD3DSurface *iface, DWORD dst
     if (trans & (WINEDDBLTFAST_SRCCOLORKEY | WINEDDBLTFAST_DESTCOLORKEY))
     {
         DWORD keylow, keyhigh;
+        DWORD mask = Src->resource.format_desc->red_mask |
+                     Src->resource.format_desc->green_mask |
+                     Src->resource.format_desc->blue_mask;
+
+        /* For some 8-bit formats like L8 and P8 color masks don't make sense */
+        if(!mask && bpp==1)
+            mask = 0xff;
+
         TRACE("Color keyed copy\n");
         if (trans & WINEDDBLTFAST_SRCCOLORKEY)
         {
@@ -1716,7 +1724,7 @@ HRESULT WINAPI IWineD3DBaseSurfaceImpl_BltFast(IWineD3DSurface *iface, DWORD dst
         for (y = 0; y < h; y++) { \
         for (x = 0; x < w; x++) { \
         tmp = s[x]; \
-        if (tmp < keylow || tmp > keyhigh) d[x] = tmp; \
+        if ((tmp & mask) < keylow || (tmp & mask) > keyhigh) d[x] = tmp; \
     } \
         s = (const type *)((const BYTE *)s + slock.Pitch); \
         d = (type *)((BYTE *)d + dlock.Pitch); \