[PATCH 0/1] MR2699: evr: Don't flip image vertically.

[eric pouech (＠epo)]

This is actually fixing two issues:
- if we get a negative stride, then the video transform should have
  already flipped vertically the image, so we don't need to flip it
  again.
  (This is another side effect for handling negative strides in
  wg_transform).
- it also fixes a crash in AoEII, as the image source line offsets
  were incorrectly computed with unsigned arithmetic (while signed
  arithmetic was expected), resulting in incorrect sign propagation
  from 32bit to 64bit integers.

Signed-off-by: Eric Pouech

[PATCH 1/1] evr: Don't flip image vertically.

[Eric Pouech]

From: Eric Pouech

This is actually fixing two issues:
- if we get a negative stride, then the video transform should have
  already flipped vertically the image, so we don't need to flip it
  again.
  (This is another side effect for handling negative strides in
  wg_transform).
- it also fixes a crash in AoEII, as the image source line offsets
  were incorrectly computed with unsigned arithmetic (while signed
  arithmetic was expected), resulting in incorrect sign propagation
  from 32bit to 64bit integers.

Signed-off-by: Eric Pouech 
---
 dlls/evr/evr.c | 3 +--
 1 file changed, 1 insertion(+), 2 deletions(-)

diff --git a/dlls/evr/evr.c b/dlls/evr/evr.c
index 26a2a3a1eb5..25785ea97ab 100644
--- a/dlls/evr/evr.c
+++ b/dlls/evr/evr.c
@@ -382,8 +382,7 @@ static HRESULT evr_copy_sample_buffer(struct evr *filter, IMediaSample *input_sa
         {
             if (SUCCEEDED(hr = IDirect3DSurface9_LockRect(surface, &locked_rect, NULL, D3DLOCK_DISCARD)))
             {
-                if (src_stride < 0) src -= src_stride * (lines - 1);
-                MFCopyImage(locked_rect.pBits, locked_rect.Pitch, src, src_stride, width * 4, lines);
+                MFCopyImage(locked_rect.pBits, locked_rect.Pitch, src, abs(src_stride), width * 4, lines);
                 IDirect3DSurface9_UnlockRect(surface);
             }

[Nikolay Sivov (＠nsivov)]

> This is actually fixing two issues:
> - if we get a negative stride, then the video transform should have already flipped vertically the image, so we don't need to flip it again. (This is another side effect for handling negative strides in wg_transform).

This does not explain much, because EVR does not require a transform.
> - it also fixes a crash in AoEII, as the image source line offsets were incorrectly computed with unsigned arithmetic (while signed arithmetic was expected), resulting in incorrect sign propagation from 32bit to 64bit integers.

Is it possible to test EVR in isolation to validate this change?

[eric pouech (＠epo)]

closing (bug I was chasing was due to an incorrect integration of other changes)

[PATCH 0/1] MR2699: evr: Don't flip image vertically. - closed

[eric pouech (＠epo)]

This merge request was closed by eric pouech.