[FFmpeg-devel] [PATCH] vc1dsp: Change remaining stride parameters to ptrdiff_t

[FFmpeg-devel] [PATCH] vc1dsp: Change remaining stride parameters to ptrdiff_t

[Martin Storsjö]

The existing x86 assembly for loop filters uses the stride as a
full register without clearing/sign extending the upper half
of the registers on x86_64.

This avoids crashes if the caller would have passed nonzero bits
in the previously undefined upper 32 bits of the parameters.

Signed-off-by: Martin Storsjö <martin@martin.st>
---
 libavcodec/vc1dsp.c          | 20 ++++++++++----------
 libavcodec/vc1dsp.h          | 16 ++++++++--------
 libavcodec/x86/vc1dsp_init.c | 16 ++++++++--------
 3 files changed, 26 insertions(+), 26 deletions(-)

diff --git a/libavcodec/vc1dsp.c b/libavcodec/vc1dsp.c
index a29b91bf3d..c567bc43e8 100644
--- a/libavcodec/vc1dsp.c
+++ b/libavcodec/vc1dsp.c
@@ -36,7 +36,7 @@
 #include "startcode.h"
 
 /* Apply overlap transform to horizontal edge */
-static void vc1_v_overlap_c(uint8_t *src, int stride)
+static void vc1_v_overlap_c(uint8_t *src, ptrdiff_t stride)
 {
     int i;
     int a, b, c, d;
@@ -60,7 +60,7 @@ static void vc1_v_overlap_c(uint8_t *src, int stride)
 }
 
 /* Apply overlap transform to vertical edge */
-static void vc1_h_overlap_c(uint8_t *src, int stride)
+static void vc1_h_overlap_c(uint8_t *src, ptrdiff_t stride)
 {
     int i;
     int a, b, c, d;
@@ -146,7 +146,7 @@ static void vc1_h_s_overlap_c(int16_t *left, int16_t *right, int left_stride, in
  * @return whether other 3 pairs should be filtered or not
  * @see 8.6
  */
-static av_always_inline int vc1_filter_line(uint8_t *src, int stride, int pq)
+static av_always_inline int vc1_filter_line(uint8_t *src, ptrdiff_t stride, int pq)
 {
     int a0 = (2 * (src[-2 * stride] - src[1 * stride]) -
               5 * (src[-1 * stride] - src[0 * stride]) + 4) >> 3;
@@ -195,7 +195,7 @@ static av_always_inline int vc1_filter_line(uint8_t *src, int stride, int pq)
  * @param pq block quantizer
  * @see 8.6
  */
-static inline void vc1_loop_filter(uint8_t *src, int step, int stride,
+static inline void vc1_loop_filter(uint8_t *src, int step, ptrdiff_t stride,
                                    int len, int pq)
 {
     int i;
@@ -212,32 +212,32 @@ static inline void vc1_loop_filter(uint8_t *src, int step, int stride,
     }
 }
 
-static void vc1_v_loop_filter4_c(uint8_t *src, int stride, int pq)
+static void vc1_v_loop_filter4_c(uint8_t *src, ptrdiff_t stride, int pq)
 {
     vc1_loop_filter(src, 1, stride, 4, pq);
 }
 
-static void vc1_h_loop_filter4_c(uint8_t *src, int stride, int pq)
+static void vc1_h_loop_filter4_c(uint8_t *src, ptrdiff_t stride, int pq)
 {
     vc1_loop_filter(src, stride, 1, 4, pq);
 }
 
-static void vc1_v_loop_filter8_c(uint8_t *src, int stride, int pq)
+static void vc1_v_loop_filter8_c(uint8_t *src, ptrdiff_t stride, int pq)
 {
     vc1_loop_filter(src, 1, stride, 8, pq);
 }
 
-static void vc1_h_loop_filter8_c(uint8_t *src, int stride, int pq)
+static void vc1_h_loop_filter8_c(uint8_t *src, ptrdiff_t stride, int pq)
 {
     vc1_loop_filter(src, stride, 1, 8, pq);
 }
 
-static void vc1_v_loop_filter16_c(uint8_t *src, int stride, int pq)
+static void vc1_v_loop_filter16_c(uint8_t *src, ptrdiff_t stride, int pq)
 {
     vc1_loop_filter(src, 1, stride, 16, pq);
 }
 
-static void vc1_h_loop_filter16_c(uint8_t *src, int stride, int pq)
+static void vc1_h_loop_filter16_c(uint8_t *src, ptrdiff_t stride, int pq)
 {
     vc1_loop_filter(src, stride, 1, 16, pq);
 }
diff --git a/libavcodec/vc1dsp.h b/libavcodec/vc1dsp.h
index c6443acb20..34f3c89d85 100644
--- a/libavcodec/vc1dsp.h
+++ b/libavcodec/vc1dsp.h
@@ -42,16 +42,16 @@ typedef struct VC1DSPContext {
     void (*vc1_inv_trans_8x4_dc)(uint8_t *dest, ptrdiff_t stride, int16_t *block);
     void (*vc1_inv_trans_4x8_dc)(uint8_t *dest, ptrdiff_t stride, int16_t *block);
     void (*vc1_inv_trans_4x4_dc)(uint8_t *dest, ptrdiff_t stride, int16_t *block);
-    void (*vc1_v_overlap)(uint8_t *src, int stride);
-    void (*vc1_h_overlap)(uint8_t *src, int stride);
+    void (*vc1_v_overlap)(uint8_t *src, ptrdiff_t stride);
+    void (*vc1_h_overlap)(uint8_t *src, ptrdiff_t stride);
     void (*vc1_v_s_overlap)(int16_t *top,  int16_t *bottom);
     void (*vc1_h_s_overlap)(int16_t *left, int16_t *right, int left_stride, int right_stride, int flags);
-    void (*vc1_v_loop_filter4)(uint8_t *src, int stride, int pq);
-    void (*vc1_h_loop_filter4)(uint8_t *src, int stride, int pq);
-    void (*vc1_v_loop_filter8)(uint8_t *src, int stride, int pq);
-    void (*vc1_h_loop_filter8)(uint8_t *src, int stride, int pq);
-    void (*vc1_v_loop_filter16)(uint8_t *src, int stride, int pq);
-    void (*vc1_h_loop_filter16)(uint8_t *src, int stride, int pq);
+    void (*vc1_v_loop_filter4)(uint8_t *src, ptrdiff_t stride, int pq);
+    void (*vc1_h_loop_filter4)(uint8_t *src, ptrdiff_t stride, int pq);
+    void (*vc1_v_loop_filter8)(uint8_t *src, ptrdiff_t stride, int pq);
+    void (*vc1_h_loop_filter8)(uint8_t *src, ptrdiff_t stride, int pq);
+    void (*vc1_v_loop_filter16)(uint8_t *src, ptrdiff_t stride, int pq);
+    void (*vc1_h_loop_filter16)(uint8_t *src, ptrdiff_t stride, int pq);
 
     /* put 8x8 block with bicubic interpolation and quarterpel precision
      * last argument is actually round value instead of height
diff --git a/libavcodec/x86/vc1dsp_init.c b/libavcodec/x86/vc1dsp_init.c
index 8e0c284cc0..2fbf0b3a74 100644
--- a/libavcodec/x86/vc1dsp_init.c
+++ b/libavcodec/x86/vc1dsp_init.c
@@ -34,18 +34,18 @@
 #include "config.h"
 
 #define LOOP_FILTER(EXT) \
-void ff_vc1_v_loop_filter4_ ## EXT(uint8_t *src, int stride, int pq); \
-void ff_vc1_h_loop_filter4_ ## EXT(uint8_t *src, int stride, int pq); \
-void ff_vc1_v_loop_filter8_ ## EXT(uint8_t *src, int stride, int pq); \
-void ff_vc1_h_loop_filter8_ ## EXT(uint8_t *src, int stride, int pq); \
+void ff_vc1_v_loop_filter4_ ## EXT(uint8_t *src, ptrdiff_t stride, int pq); \
+void ff_vc1_h_loop_filter4_ ## EXT(uint8_t *src, ptrdiff_t stride, int pq); \
+void ff_vc1_v_loop_filter8_ ## EXT(uint8_t *src, ptrdiff_t stride, int pq); \
+void ff_vc1_h_loop_filter8_ ## EXT(uint8_t *src, ptrdiff_t stride, int pq); \
 \
-static void vc1_v_loop_filter16_ ## EXT(uint8_t *src, int stride, int pq) \
+static void vc1_v_loop_filter16_ ## EXT(uint8_t *src, ptrdiff_t stride, int pq) \
 { \
     ff_vc1_v_loop_filter8_ ## EXT(src,   stride, pq); \
     ff_vc1_v_loop_filter8_ ## EXT(src+8, stride, pq); \
 } \
 \
-static void vc1_h_loop_filter16_ ## EXT(uint8_t *src, int stride, int pq) \
+static void vc1_h_loop_filter16_ ## EXT(uint8_t *src, ptrdiff_t stride, int pq) \
 { \
     ff_vc1_h_loop_filter8_ ## EXT(src,          stride, pq); \
     ff_vc1_h_loop_filter8_ ## EXT(src+8*stride, stride, pq); \
@@ -56,9 +56,9 @@ LOOP_FILTER(mmxext)
 LOOP_FILTER(sse2)
 LOOP_FILTER(ssse3)
 
-void ff_vc1_h_loop_filter8_sse4(uint8_t *src, int stride, int pq);
+void ff_vc1_h_loop_filter8_sse4(uint8_t *src, ptrdiff_t stride, int pq);
 
-static void vc1_h_loop_filter16_sse4(uint8_t *src, int stride, int pq)
+static void vc1_h_loop_filter16_sse4(uint8_t *src, ptrdiff_t stride, int pq)
 {
     ff_vc1_h_loop_filter8_sse4(src,          stride, pq);
     ff_vc1_h_loop_filter8_sse4(src+8*stride, stride, pq);
-- 
2.25.1

_______________________________________________
ffmpeg-devel mailing list
ffmpeg-devel@ffmpeg.org
https://ffmpeg.org/mailman/listinfo/ffmpeg-devel

To unsubscribe, visit link above, or email
ffmpeg-devel-request@ffmpeg.org with subject "unsubscribe".

Re: [FFmpeg-devel] [PATCH] vc1dsp: Change remaining stride parameters to ptrdiff_t

[Ben Avison]

On 29/03/2022 13:44, Martin Storsjö wrote:
> The existing x86 assembly for loop filters uses the stride as a
> full register without clearing/sign extending the upper half
> of the registers on x86_64.
> 
> This avoids crashes if the caller would have passed nonzero bits
> in the previously undefined upper 32 bits of the parameters.

Looks perfectly reasonable to me. I'll hold off making matching changes 
to my upcoming Arm versions until this gets merged. (My 64-bit Arm 
versions followed the old prototype and sign-extended the strides from 
32 to 64 bits, but that won't be necessary any more.)

Ben
_______________________________________________
ffmpeg-devel mailing list
ffmpeg-devel@ffmpeg.org
https://ffmpeg.org/mailman/listinfo/ffmpeg-devel

To unsubscribe, visit link above, or email
ffmpeg-devel-request@ffmpeg.org with subject "unsubscribe".

[FFmpeg-devel] [PATCH v2] vc1dsp: Change remaining stride parameters to ptrdiff_t

[Martin Storsjö]

The existing x86 assembly for loop filters uses the stride as a
full register without clearing/sign extending the upper half
of the registers on x86_64.

This avoids crashes if the caller would have passed nonzero bits
in the previously undefined upper 32 bits of the parameters.

Signed-off-by: Martin Storsjö <martin@martin.st>
---
Updated function signatures in the mips code too, updated the
left_stride/right_stride parameters in the vc1_h_s_overlap
function too, updated the comments in the x86 assembly.
---
 libavcodec/mips/vc1dsp_mips.h        | 20 ++++++++++----------
 libavcodec/mips/vc1dsp_mmi.c         | 18 +++++++++---------
 libavcodec/vc1dsp.c                  | 22 +++++++++++-----------
 libavcodec/vc1dsp.h                  | 18 +++++++++---------
 libavcodec/x86/vc1dsp_init.c         | 16 ++++++++--------
 libavcodec/x86/vc1dsp_loopfilter.asm | 22 +++++++++++-----------
 6 files changed, 58 insertions(+), 58 deletions(-)

diff --git a/libavcodec/mips/vc1dsp_mips.h b/libavcodec/mips/vc1dsp_mips.h
index 5897daea8c..356153cab6 100644
--- a/libavcodec/mips/vc1dsp_mips.h
+++ b/libavcodec/mips/vc1dsp_mips.h
@@ -166,17 +166,17 @@ void ff_vc1_inv_trans_4x8_dc_mmi(uint8_t *dest, ptrdiff_t linesize, int16_t *blo
 void ff_vc1_inv_trans_8x4_dc_mmi(uint8_t *dest, ptrdiff_t linesize, int16_t *block);
 void ff_vc1_inv_trans_8x8_dc_mmi(uint8_t *dest, ptrdiff_t linesize, int16_t *block);
 
-void ff_vc1_v_overlap_mmi(uint8_t *src, int stride);
-void ff_vc1_h_overlap_mmi(uint8_t *src, int stride);
+void ff_vc1_v_overlap_mmi(uint8_t *src, ptrdiff_t stride);
+void ff_vc1_h_overlap_mmi(uint8_t *src, ptrdiff_t stride);
 void ff_vc1_v_s_overlap_mmi(int16_t *top, int16_t *bottom);
-void ff_vc1_h_s_overlap_mmi(int16_t *left, int16_t *right, int left_stride, int right_stride, int flags);
-
-void ff_vc1_v_loop_filter4_mmi(uint8_t *src, int stride, int pq);
-void ff_vc1_h_loop_filter4_mmi(uint8_t *src, int stride, int pq);
-void ff_vc1_v_loop_filter8_mmi(uint8_t *src, int stride, int pq);
-void ff_vc1_h_loop_filter8_mmi(uint8_t *src, int stride, int pq);
-void ff_vc1_v_loop_filter16_mmi(uint8_t *src, int stride, int pq);
-void ff_vc1_h_loop_filter16_mmi(uint8_t *src, int stride, int pq);
+void ff_vc1_h_s_overlap_mmi(int16_t *left, int16_t *right, ptrdiff_t left_stride, ptrdiff_t right_stride, int flags);
+
+void ff_vc1_v_loop_filter4_mmi(uint8_t *src, ptrdiff_t stride, int pq);
+void ff_vc1_h_loop_filter4_mmi(uint8_t *src, ptrdiff_t stride, int pq);
+void ff_vc1_v_loop_filter8_mmi(uint8_t *src, ptrdiff_t stride, int pq);
+void ff_vc1_h_loop_filter8_mmi(uint8_t *src, ptrdiff_t stride, int pq);
+void ff_vc1_v_loop_filter16_mmi(uint8_t *src, ptrdiff_t stride, int pq);
+void ff_vc1_h_loop_filter16_mmi(uint8_t *src, ptrdiff_t stride, int pq);
 
 void ff_put_no_rnd_vc1_chroma_mc8_mmi(uint8_t *dst /* align 8 */,
                                       uint8_t *src /* align 1 */,
diff --git a/libavcodec/mips/vc1dsp_mmi.c b/libavcodec/mips/vc1dsp_mmi.c
index 27a3c813da..278aa566bc 100644
--- a/libavcodec/mips/vc1dsp_mmi.c
+++ b/libavcodec/mips/vc1dsp_mmi.c
@@ -1375,7 +1375,7 @@ void ff_vc1_inv_trans_4x4_mmi(uint8_t *dest, ptrdiff_t linesize, int16_t *block)
 }
 
 /* Apply overlap transform to horizontal edge */
-void ff_vc1_h_overlap_mmi(uint8_t *src, int stride)
+void ff_vc1_h_overlap_mmi(uint8_t *src, ptrdiff_t stride)
 {
     int i;
     int a, b, c, d;
@@ -1398,7 +1398,7 @@ void ff_vc1_h_overlap_mmi(uint8_t *src, int stride)
     }
 }
 
-void ff_vc1_h_s_overlap_mmi(int16_t *left, int16_t *right, int left_stride, int right_stride, int flags)
+void ff_vc1_h_s_overlap_mmi(int16_t *left, int16_t *right, ptrdiff_t left_stride, ptrdiff_t right_stride, int flags)
 {
     int i;
     int a, b, c, d;
@@ -1428,7 +1428,7 @@ void ff_vc1_h_s_overlap_mmi(int16_t *left, int16_t *right, int left_stride, int
 }
 
 /* Apply overlap transform to vertical edge */
-void ff_vc1_v_overlap_mmi(uint8_t *src, int stride)
+void ff_vc1_v_overlap_mmi(uint8_t *src, ptrdiff_t stride)
 {
     int i;
     int a, b, c, d;
@@ -1551,32 +1551,32 @@ static inline void vc1_loop_filter(uint8_t *src, int step, int stride,
     }
 }
 
-void ff_vc1_v_loop_filter4_mmi(uint8_t *src, int stride, int pq)
+void ff_vc1_v_loop_filter4_mmi(uint8_t *src, ptrdiff_t stride, int pq)
 {
     vc1_loop_filter(src, 1, stride, 4, pq);
 }
 
-void ff_vc1_h_loop_filter4_mmi(uint8_t *src, int stride, int pq)
+void ff_vc1_h_loop_filter4_mmi(uint8_t *src, ptrdiff_t stride, int pq)
 {
     vc1_loop_filter(src, stride, 1, 4, pq);
 }
 
-void ff_vc1_v_loop_filter8_mmi(uint8_t *src, int stride, int pq)
+void ff_vc1_v_loop_filter8_mmi(uint8_t *src, ptrdiff_t stride, int pq)
 {
     vc1_loop_filter(src, 1, stride, 8, pq);
 }
 
-void ff_vc1_h_loop_filter8_mmi(uint8_t *src, int stride, int pq)
+void ff_vc1_h_loop_filter8_mmi(uint8_t *src, ptrdiff_t stride, int pq)
 {
     vc1_loop_filter(src, stride, 1, 8, pq);
 }
 
-void ff_vc1_v_loop_filter16_mmi(uint8_t *src, int stride, int pq)
+void ff_vc1_v_loop_filter16_mmi(uint8_t *src, ptrdiff_t stride, int pq)
 {
     vc1_loop_filter(src, 1, stride, 16, pq);
 }
 
-void ff_vc1_h_loop_filter16_mmi(uint8_t *src, int stride, int pq)
+void ff_vc1_h_loop_filter16_mmi(uint8_t *src, ptrdiff_t stride, int pq)
 {
     vc1_loop_filter(src, stride, 1, 16, pq);
 }
diff --git a/libavcodec/vc1dsp.c b/libavcodec/vc1dsp.c
index a29b91bf3d..f651d7d461 100644
--- a/libavcodec/vc1dsp.c
+++ b/libavcodec/vc1dsp.c
@@ -36,7 +36,7 @@
 #include "startcode.h"
 
 /* Apply overlap transform to horizontal edge */
-static void vc1_v_overlap_c(uint8_t *src, int stride)
+static void vc1_v_overlap_c(uint8_t *src, ptrdiff_t stride)
 {
     int i;
     int a, b, c, d;
@@ -60,7 +60,7 @@ static void vc1_v_overlap_c(uint8_t *src, int stride)
 }
 
 /* Apply overlap transform to vertical edge */
-static void vc1_h_overlap_c(uint8_t *src, int stride)
+static void vc1_h_overlap_c(uint8_t *src, ptrdiff_t stride)
 {
     int i;
     int a, b, c, d;
@@ -109,7 +109,7 @@ static void vc1_v_s_overlap_c(int16_t *top, int16_t *bottom)
     }
 }
 
-static void vc1_h_s_overlap_c(int16_t *left, int16_t *right, int left_stride, int right_stride, int flags)
+static void vc1_h_s_overlap_c(int16_t *left, int16_t *right, ptrdiff_t left_stride, ptrdiff_t right_stride, int flags)
 {
     int i;
     int a, b, c, d;
@@ -146,7 +146,7 @@ static void vc1_h_s_overlap_c(int16_t *left, int16_t *right, int left_stride, in
  * @return whether other 3 pairs should be filtered or not
  * @see 8.6
  */
-static av_always_inline int vc1_filter_line(uint8_t *src, int stride, int pq)
+static av_always_inline int vc1_filter_line(uint8_t *src, ptrdiff_t stride, int pq)
 {
     int a0 = (2 * (src[-2 * stride] - src[1 * stride]) -
               5 * (src[-1 * stride] - src[0 * stride]) + 4) >> 3;
@@ -195,7 +195,7 @@ static av_always_inline int vc1_filter_line(uint8_t *src, int stride, int pq)
  * @param pq block quantizer
  * @see 8.6
  */
-static inline void vc1_loop_filter(uint8_t *src, int step, int stride,
+static inline void vc1_loop_filter(uint8_t *src, int step, ptrdiff_t stride,
                                    int len, int pq)
 {
     int i;
@@ -212,32 +212,32 @@ static inline void vc1_loop_filter(uint8_t *src, int step, int stride,
     }
 }
 
-static void vc1_v_loop_filter4_c(uint8_t *src, int stride, int pq)
+static void vc1_v_loop_filter4_c(uint8_t *src, ptrdiff_t stride, int pq)
 {
     vc1_loop_filter(src, 1, stride, 4, pq);
 }
 
-static void vc1_h_loop_filter4_c(uint8_t *src, int stride, int pq)
+static void vc1_h_loop_filter4_c(uint8_t *src, ptrdiff_t stride, int pq)
 {
     vc1_loop_filter(src, stride, 1, 4, pq);
 }
 
-static void vc1_v_loop_filter8_c(uint8_t *src, int stride, int pq)
+static void vc1_v_loop_filter8_c(uint8_t *src, ptrdiff_t stride, int pq)
 {
     vc1_loop_filter(src, 1, stride, 8, pq);
 }
 
-static void vc1_h_loop_filter8_c(uint8_t *src, int stride, int pq)
+static void vc1_h_loop_filter8_c(uint8_t *src, ptrdiff_t stride, int pq)
 {
     vc1_loop_filter(src, stride, 1, 8, pq);
 }
 
-static void vc1_v_loop_filter16_c(uint8_t *src, int stride, int pq)
+static void vc1_v_loop_filter16_c(uint8_t *src, ptrdiff_t stride, int pq)
 {
     vc1_loop_filter(src, 1, stride, 16, pq);
 }
 
-static void vc1_h_loop_filter16_c(uint8_t *src, int stride, int pq)
+static void vc1_h_loop_filter16_c(uint8_t *src, ptrdiff_t stride, int pq)
 {
     vc1_loop_filter(src, stride, 1, 16, pq);
 }
diff --git a/libavcodec/vc1dsp.h b/libavcodec/vc1dsp.h
index c6443acb20..fe60025a2a 100644
--- a/libavcodec/vc1dsp.h
+++ b/libavcodec/vc1dsp.h
@@ -42,16 +42,16 @@ typedef struct VC1DSPContext {
     void (*vc1_inv_trans_8x4_dc)(uint8_t *dest, ptrdiff_t stride, int16_t *block);
     void (*vc1_inv_trans_4x8_dc)(uint8_t *dest, ptrdiff_t stride, int16_t *block);
     void (*vc1_inv_trans_4x4_dc)(uint8_t *dest, ptrdiff_t stride, int16_t *block);
-    void (*vc1_v_overlap)(uint8_t *src, int stride);
-    void (*vc1_h_overlap)(uint8_t *src, int stride);
+    void (*vc1_v_overlap)(uint8_t *src, ptrdiff_t stride);
+    void (*vc1_h_overlap)(uint8_t *src, ptrdiff_t stride);
     void (*vc1_v_s_overlap)(int16_t *top,  int16_t *bottom);
-    void (*vc1_h_s_overlap)(int16_t *left, int16_t *right, int left_stride, int right_stride, int flags);
-    void (*vc1_v_loop_filter4)(uint8_t *src, int stride, int pq);
-    void (*vc1_h_loop_filter4)(uint8_t *src, int stride, int pq);
-    void (*vc1_v_loop_filter8)(uint8_t *src, int stride, int pq);
-    void (*vc1_h_loop_filter8)(uint8_t *src, int stride, int pq);
-    void (*vc1_v_loop_filter16)(uint8_t *src, int stride, int pq);
-    void (*vc1_h_loop_filter16)(uint8_t *src, int stride, int pq);
+    void (*vc1_h_s_overlap)(int16_t *left, int16_t *right, ptrdiff_t left_stride, ptrdiff_t right_stride, int flags);
+    void (*vc1_v_loop_filter4)(uint8_t *src, ptrdiff_t stride, int pq);
+    void (*vc1_h_loop_filter4)(uint8_t *src, ptrdiff_t stride, int pq);
+    void (*vc1_v_loop_filter8)(uint8_t *src, ptrdiff_t stride, int pq);
+    void (*vc1_h_loop_filter8)(uint8_t *src, ptrdiff_t stride, int pq);
+    void (*vc1_v_loop_filter16)(uint8_t *src, ptrdiff_t stride, int pq);
+    void (*vc1_h_loop_filter16)(uint8_t *src, ptrdiff_t stride, int pq);
 
     /* put 8x8 block with bicubic interpolation and quarterpel precision
      * last argument is actually round value instead of height
diff --git a/libavcodec/x86/vc1dsp_init.c b/libavcodec/x86/vc1dsp_init.c
index 8e0c284cc0..2fbf0b3a74 100644
--- a/libavcodec/x86/vc1dsp_init.c
+++ b/libavcodec/x86/vc1dsp_init.c
@@ -34,18 +34,18 @@
 #include "config.h"
 
 #define LOOP_FILTER(EXT) \
-void ff_vc1_v_loop_filter4_ ## EXT(uint8_t *src, int stride, int pq); \
-void ff_vc1_h_loop_filter4_ ## EXT(uint8_t *src, int stride, int pq); \
-void ff_vc1_v_loop_filter8_ ## EXT(uint8_t *src, int stride, int pq); \
-void ff_vc1_h_loop_filter8_ ## EXT(uint8_t *src, int stride, int pq); \
+void ff_vc1_v_loop_filter4_ ## EXT(uint8_t *src, ptrdiff_t stride, int pq); \
+void ff_vc1_h_loop_filter4_ ## EXT(uint8_t *src, ptrdiff_t stride, int pq); \
+void ff_vc1_v_loop_filter8_ ## EXT(uint8_t *src, ptrdiff_t stride, int pq); \
+void ff_vc1_h_loop_filter8_ ## EXT(uint8_t *src, ptrdiff_t stride, int pq); \
 \
-static void vc1_v_loop_filter16_ ## EXT(uint8_t *src, int stride, int pq) \
+static void vc1_v_loop_filter16_ ## EXT(uint8_t *src, ptrdiff_t stride, int pq) \
 { \
     ff_vc1_v_loop_filter8_ ## EXT(src,   stride, pq); \
     ff_vc1_v_loop_filter8_ ## EXT(src+8, stride, pq); \
 } \
 \
-static void vc1_h_loop_filter16_ ## EXT(uint8_t *src, int stride, int pq) \
+static void vc1_h_loop_filter16_ ## EXT(uint8_t *src, ptrdiff_t stride, int pq) \
 { \
     ff_vc1_h_loop_filter8_ ## EXT(src,          stride, pq); \
     ff_vc1_h_loop_filter8_ ## EXT(src+8*stride, stride, pq); \
@@ -56,9 +56,9 @@ LOOP_FILTER(mmxext)
 LOOP_FILTER(sse2)
 LOOP_FILTER(ssse3)
 
-void ff_vc1_h_loop_filter8_sse4(uint8_t *src, int stride, int pq);
+void ff_vc1_h_loop_filter8_sse4(uint8_t *src, ptrdiff_t stride, int pq);
 
-static void vc1_h_loop_filter16_sse4(uint8_t *src, int stride, int pq)
+static void vc1_h_loop_filter16_sse4(uint8_t *src, ptrdiff_t stride, int pq)
 {
     ff_vc1_h_loop_filter8_sse4(src,          stride, pq);
     ff_vc1_h_loop_filter8_sse4(src+8*stride, stride, pq);
diff --git a/libavcodec/x86/vc1dsp_loopfilter.asm b/libavcodec/x86/vc1dsp_loopfilter.asm
index fd33bd13dc..74360949dc 100644
--- a/libavcodec/x86/vc1dsp_loopfilter.asm
+++ b/libavcodec/x86/vc1dsp_loopfilter.asm
@@ -237,19 +237,19 @@ cglobal vc1_h_loop_filter_internal
     VC1_H_LOOP_FILTER 4, r4
     ret
 
-; void ff_vc1_v_loop_filter4_mmxext(uint8_t *src, int stride, int pq)
+; void ff_vc1_v_loop_filter4_mmxext(uint8_t *src, ptrdiff_t stride, int pq)
 cglobal vc1_v_loop_filter4, 3,5,0
     START_V_FILTER
     call vc1_v_loop_filter_internal
     RET
 
-; void ff_vc1_h_loop_filter4_mmxext(uint8_t *src, int stride, int pq)
+; void ff_vc1_h_loop_filter4_mmxext(uint8_t *src, ptrdiff_t stride, int pq)
 cglobal vc1_h_loop_filter4, 3,5,0
     START_H_FILTER 4
     call vc1_h_loop_filter_internal
     RET
 
-; void ff_vc1_v_loop_filter8_mmxext(uint8_t *src, int stride, int pq)
+; void ff_vc1_v_loop_filter8_mmxext(uint8_t *src, ptrdiff_t stride, int pq)
 cglobal vc1_v_loop_filter8, 3,5,0
     START_V_FILTER
     call vc1_v_loop_filter_internal
@@ -258,7 +258,7 @@ cglobal vc1_v_loop_filter8, 3,5,0
     call vc1_v_loop_filter_internal
     RET
 
-; void ff_vc1_h_loop_filter8_mmxext(uint8_t *src, int stride, int pq)
+; void ff_vc1_h_loop_filter8_mmxext(uint8_t *src, ptrdiff_t stride, int pq)
 cglobal vc1_h_loop_filter8, 3,5,0
     START_H_FILTER 4
     call vc1_h_loop_filter_internal
@@ -271,46 +271,46 @@ INIT_MMX mmxext
 VC1_LF
 
 INIT_XMM sse2
-; void ff_vc1_v_loop_filter8_sse2(uint8_t *src, int stride, int pq)
+; void ff_vc1_v_loop_filter8_sse2(uint8_t *src, ptrdiff_t stride, int pq)
 cglobal vc1_v_loop_filter8, 3,5,8
     START_V_FILTER
     VC1_V_LOOP_FILTER 8, q
     RET
 
-; void ff_vc1_h_loop_filter8_sse2(uint8_t *src, int stride, int pq)
+; void ff_vc1_h_loop_filter8_sse2(uint8_t *src, ptrdiff_t stride, int pq)
 cglobal vc1_h_loop_filter8, 3,6,8
     START_H_FILTER 8
     VC1_H_LOOP_FILTER 8, r5
     RET
 
 INIT_MMX ssse3
-; void ff_vc1_v_loop_filter4_ssse3(uint8_t *src, int stride, int pq)
+; void ff_vc1_v_loop_filter4_ssse3(uint8_t *src, ptrdiff_t stride, int pq)
 cglobal vc1_v_loop_filter4, 3,5,0
     START_V_FILTER
     VC1_V_LOOP_FILTER 4, d
     RET
 
-; void ff_vc1_h_loop_filter4_ssse3(uint8_t *src, int stride, int pq)
+; void ff_vc1_h_loop_filter4_ssse3(uint8_t *src, ptrdiff_t stride, int pq)
 cglobal vc1_h_loop_filter4, 3,5,0
     START_H_FILTER 4
     VC1_H_LOOP_FILTER 4, r4
     RET
 
 INIT_XMM ssse3
-; void ff_vc1_v_loop_filter8_ssse3(uint8_t *src, int stride, int pq)
+; void ff_vc1_v_loop_filter8_ssse3(uint8_t *src, ptrdiff_t stride, int pq)
 cglobal vc1_v_loop_filter8, 3,5,8
     START_V_FILTER
     VC1_V_LOOP_FILTER 8, q
     RET
 
-; void ff_vc1_h_loop_filter8_ssse3(uint8_t *src, int stride, int pq)
+; void ff_vc1_h_loop_filter8_ssse3(uint8_t *src, ptrdiff_t stride, int pq)
 cglobal vc1_h_loop_filter8, 3,6,8
     START_H_FILTER 8
     VC1_H_LOOP_FILTER 8, r5
     RET
 
 INIT_XMM sse4
-; void ff_vc1_h_loop_filter8_sse4(uint8_t *src, int stride, int pq)
+; void ff_vc1_h_loop_filter8_sse4(uint8_t *src, ptrdiff_t stride, int pq)
 cglobal vc1_h_loop_filter8, 3,5,8
     START_H_FILTER 8
     VC1_H_LOOP_FILTER 8
-- 
2.25.1

_______________________________________________
ffmpeg-devel mailing list
ffmpeg-devel@ffmpeg.org
https://ffmpeg.org/mailman/listinfo/ffmpeg-devel

To unsubscribe, visit link above, or email
ffmpeg-devel-request@ffmpeg.org with subject "unsubscribe".

Re: [FFmpeg-devel] [PATCH] vc1dsp: Change remaining stride parameters to ptrdiff_t

[Martin Storsjö]

On Tue, 29 Mar 2022, Ben Avison wrote:

> On 29/03/2022 13:44, Martin Storsjö wrote:
>> The existing x86 assembly for loop filters uses the stride as a
>> full register without clearing/sign extending the upper half
>> of the registers on x86_64.
>> 
>> This avoids crashes if the caller would have passed nonzero bits
>> in the previously undefined upper 32 bits of the parameters.
>
> Looks perfectly reasonable to me. I'll hold off making matching changes 
> to my upcoming Arm versions until this gets merged. (My 64-bit Arm 
> versions followed the old prototype and sign-extended the strides from 
> 32 to 64 bits, but that won't be necessary any more.)

Acked by Lynne on irc too - so I pushed it now.

// Martin

_______________________________________________
ffmpeg-devel mailing list
ffmpeg-devel@ffmpeg.org
https://ffmpeg.org/mailman/listinfo/ffmpeg-devel

To unsubscribe, visit link above, or email
ffmpeg-devel-request@ffmpeg.org with subject "unsubscribe".