Commit 64ed3976 authored by Anton Khirnov's avatar Anton Khirnov

vf_yadif: fix out-of line reads

Some changes in the border pixels, visually indistinguishable.
parent ccd70d9c
......@@ -34,15 +34,15 @@
#define PERM_RWP AV_PERM_WRITE | AV_PERM_PRESERVE | AV_PERM_REUSE
#define CHECK(j)\
{ int score = FFABS(cur[mrefs-1+(j)] - cur[prefs-1-(j)])\
{ int score = FFABS(cur[mrefs + off_left + (j)] - cur[prefs + off_left - (j)])\
+ FFABS(cur[mrefs +(j)] - cur[prefs -(j)])\
+ FFABS(cur[mrefs+1+(j)] - cur[prefs+1-(j)]);\
+ FFABS(cur[mrefs + off_right + (j)] - cur[prefs + off_right - (j)]);\
if (score < spatial_score) {\
spatial_score= score;\
spatial_pred= (cur[mrefs +(j)] + cur[prefs -(j)])>>1;\
#define FILTER \
for (x = 0; x < w; x++) { \
#define FILTER(start, end) \
for (x = start; x < end; x++) { \
int c = cur[mrefs]; \
int d = (prev2[0] + next2[0])>>1; \
int e = cur[prefs]; \
......@@ -51,11 +51,15 @@
int temporal_diff2 =(FFABS(next[mrefs] - c) + FFABS(next[prefs] - e) )>>1; \
int diff = FFMAX3(temporal_diff0 >> 1, temporal_diff1, temporal_diff2); \
int spatial_pred = (c+e) >> 1; \
int spatial_score = FFABS(cur[mrefs - 1] - cur[prefs - 1]) + FFABS(c-e) \
+ FFABS(cur[mrefs + 1] - cur[prefs + 1]) - 1; \
int off_right = (x < w - 1) ? 1 : -1;\
int off_left = x ? -1 : 1;\
int spatial_score = FFABS(cur[mrefs + off_left] - cur[prefs + off_left]) + FFABS(c-e) \
+ FFABS(cur[mrefs + off_right] - cur[prefs + off_right]) - 1; \
\
CHECK(-1) CHECK(-2) }} }} \
CHECK( 1) CHECK( 2) }} }} \
if (x > 2 && x < w - 3) {\
CHECK(-1) CHECK(-2) }} }} \
CHECK( 1) CHECK( 2) }} }} \
}\
\
if (mode < 2) { \
int b = (prev2[2 * mrefs] + next2[2 * mrefs])>>1; \
......@@ -93,9 +97,34 @@ static void filter_line_c(void *dst1,
uint8_t *prev2 = parity ? prev : cur ;
uint8_t *next2 = parity ? cur : next;
FILTER
FILTER(0, w)
}
static void filter_edges(void *dst1, void *prev1, void *cur1, void *next1,
int w, int prefs, int mrefs, int parity, int mode,
int l_edge)
{
uint8_t *dst = dst1;
uint8_t *prev = prev1;
uint8_t *cur = cur1;
uint8_t *next = next1;
int x;
uint8_t *prev2 = parity ? prev : cur ;
uint8_t *next2 = parity ? cur : next;
FILTER(0, l_edge)
dst = (uint8_t*)dst1 + w - 3;
prev = (uint8_t*)prev1 + w - 3;
cur = (uint8_t*)cur1 + w - 3;
next = (uint8_t*)next1 + w - 3;
prev2 = (uint8_t*)(parity ? prev : cur);
next2 = (uint8_t*)(parity ? cur : next);
FILTER(w - 3, w)
}
static void filter_line_c_16bit(void *dst1,
void *prev1, void *cur1, void *next1,
int w, int prefs, int mrefs, int parity,
......@@ -111,7 +140,31 @@ static void filter_line_c_16bit(void *dst1,
mrefs /= 2;
prefs /= 2;
FILTER
FILTER(0, w)
}
static void filter_edges_16bit(void *dst1, void *prev1, void *cur1, void *next1,
int w, int prefs, int mrefs, int parity, int mode,
int l_edge)
{
uint16_t *dst = dst1;
uint16_t *prev = prev1;
uint16_t *cur = cur1;
uint16_t *next = next1;
int x;
uint16_t *prev2 = parity ? prev : cur ;
uint16_t *next2 = parity ? cur : next;
FILTER(0, l_edge)
dst = (uint16_t*)dst1 + w - 3;
prev = (uint16_t*)prev1 + w - 3;
cur = (uint16_t*)cur1 + w - 3;
next = (uint16_t*)next1 + w - 3;
prev2 = (uint16_t*)(parity ? prev : cur);
next2 = (uint16_t*)(parity ? cur : next);
FILTER(w - 3, w)
}
static void filter(AVFilterContext *ctx, AVFilterBufferRef *dstpic,
......@@ -125,6 +178,7 @@ static void filter(AVFilterContext *ctx, AVFilterBufferRef *dstpic,
int h = dstpic->video->h;
int refs = yadif->cur->linesize[i];
int df = (yadif->csp->comp[i].depth_minus1 + 8) / 8;
int l_edge, l_edge_pix;
if (i == 1 || i == 2) {
/* Why is this not part of the per-plane description thing? */
......@@ -132,6 +186,12 @@ static void filter(AVFilterContext *ctx, AVFilterBufferRef *dstpic,
h >>= yadif->csp->log2_chroma_h;
}
/* filtering reads 3 pixels to the left/right; to avoid invalid reads,
* we need to call the c variant which avoids this for border pixels
*/
l_edge = yadif->req_align;
l_edge_pix = l_edge / df;
for (y = 0; y < h; y++) {
if ((y ^ parity) & 1) {
uint8_t *prev = &yadif->prev->data[i][y * refs];
......@@ -139,10 +199,22 @@ static void filter(AVFilterContext *ctx, AVFilterBufferRef *dstpic,
uint8_t *next = &yadif->next->data[i][y * refs];
uint8_t *dst = &dstpic->data[i][y * dstpic->linesize[i]];
int mode = y == 1 || y + 2 == h ? 2 : yadif->mode;
yadif->filter_line(dst, prev, cur, next, w,
y + 1 < h ? refs : -refs,
y ? -refs : refs,
parity ^ tff, mode);
if (yadif->req_align) {
yadif->filter_line(dst + l_edge, prev + l_edge, cur + l_edge,
next + l_edge, w - l_edge_pix - 3,
y + 1 < h ? refs : -refs,
y ? -refs : refs,
parity ^ tff, mode);
yadif->filter_edges(dst, prev, cur, next, w,
y + 1 < h ? refs : -refs,
y ? -refs : refs,
parity ^ tff, mode, l_edge_pix);
} else {
yadif->filter_line(dst, prev, cur, next + l_edge, w,
y + 1 < h ? refs : -refs,
y ? -refs : refs,
parity ^ tff, mode);
}
} else {
memcpy(&dstpic->data[i][y * dstpic->linesize[i]],
&yadif->cur->data[i][y * refs], w * df);
......@@ -391,9 +463,11 @@ static int config_props(AVFilterLink *link)
s->csp = av_pix_fmt_desc_get(link->format);
if (s->csp->comp[0].depth_minus1 / 8 == 1) {
s->filter_line = filter_line_c_16bit;
s->filter_line = filter_line_c_16bit;
s->filter_edges = filter_edges_16bit;
} else {
s->filter_line = filter_line_c;
s->filter_line = filter_line_c;
s->filter_edges = filter_edges;
if (ARCH_X86)
ff_yadif_init_x86(s);
......
......@@ -43,12 +43,18 @@ av_cold void ff_yadif_init_x86(YADIFContext *yadif)
#if HAVE_YASM
#if ARCH_X86_32
if (EXTERNAL_MMXEXT(cpu_flags))
if (EXTERNAL_MMXEXT(cpu_flags)) {
yadif->filter_line = ff_yadif_filter_line_mmxext;
yadif->req_align = 8;
}
#endif /* ARCH_X86_32 */
if (EXTERNAL_SSE2(cpu_flags))
if (EXTERNAL_SSE2(cpu_flags)) {
yadif->filter_line = ff_yadif_filter_line_sse2;
if (EXTERNAL_SSSE3(cpu_flags))
yadif->req_align = 16;
}
if (EXTERNAL_SSSE3(cpu_flags)) {
yadif->filter_line = ff_yadif_filter_line_ssse3;
yadif->req_align = 16;
}
#endif /* HAVE_YASM */
}
......@@ -50,9 +50,17 @@ typedef struct YADIFContext {
AVFilterBufferRef *next;
AVFilterBufferRef *prev;
AVFilterBufferRef *out;
/**
* Required alignment for filter_line
*/
int req_align;
void (*filter_line)(void *dst,
void *prev, void *cur, void *next,
int w, int prefs, int mrefs, int parity, int mode);
void (*filter_edges)(void *dst, void *prev, void *cur, void *next,
int w, int prefs, int mrefs, int parity, int mode,
int l_edge);
const AVPixFmtDescriptor *csp;
int eof;
......
#tb 0: 1/180000
0, 64800, 64800, 0, 622080, 0x4440caef
0, 72000, 72000, 0, 622080, 0xce67e69d
0, 79200, 79200, 0, 622080, 0x1dbdc653
0, 86400, 86400, 0, 622080, 0x82c591d1
0, 93600, 93600, 0, 622080, 0x8193740b
0, 100800, 100800, 0, 622080, 0xcb219711
0, 108000, 108000, 0, 622080, 0x1870783b
0, 115200, 115200, 0, 622080, 0x7080590b
0, 122400, 122400, 0, 622080, 0x6df4175d
0, 129600, 129600, 0, 622080, 0x6b530e95
0, 136800, 136800, 0, 622080, 0x7f9d66f7
0, 144000, 144000, 0, 622080, 0x338cda81
0, 151200, 151200, 0, 622080, 0xb13797f8
0, 158400, 158400, 0, 622080, 0xb51e7ca4
0, 165600, 165600, 0, 622080, 0x353eed75
0, 172800, 172800, 0, 622080, 0xf93e92b0
0, 180000, 180000, 0, 622080, 0xd0811094
0, 187200, 187200, 0, 622080, 0xb04a3141
0, 194400, 194400, 0, 622080, 0x4ab84909
0, 201600, 201600, 0, 622080, 0xa0fcb8fb
0, 208800, 208800, 0, 622080, 0x9003aebb
0, 216000, 216000, 0, 622080, 0x153faa3e
0, 223200, 223200, 0, 622080, 0xae724063
0, 230400, 230400, 0, 622080, 0xeb4de77a
0, 237600, 237600, 0, 622080, 0x209ed8c7
0, 244800, 244800, 0, 622080, 0xe2bbac96
0, 252000, 252000, 0, 622080, 0xe945441e
0, 259200, 259200, 0, 622080, 0x8f8cbd5f
0, 266400, 266400, 0, 622080, 0xbc3cf717
0, 273600, 273600, 0, 622080, 0x0109f125
0, 280800, 280800, 0, 622080, 0x230c373f
0, 64800, 64800, 0, 622080, 0x6331caee
0, 72000, 72000, 0, 622080, 0xa459e690
0, 79200, 79200, 0, 622080, 0x6429c648
0, 86400, 86400, 0, 622080, 0xa49891ca
0, 93600, 93600, 0, 622080, 0x2a887404
0, 100800, 100800, 0, 622080, 0xe8d49705
0, 108000, 108000, 0, 622080, 0x1b627835
0, 115200, 115200, 0, 622080, 0x686858fd
0, 122400, 122400, 0, 622080, 0x2675174f
0, 129600, 129600, 0, 622080, 0x78470e7f
0, 136800, 136800, 0, 622080, 0xffb366ec
0, 144000, 144000, 0, 622080, 0xd575da72
0, 151200, 151200, 0, 622080, 0x5fb297f7
0, 158400, 158400, 0, 622080, 0xbac77ca0
0, 165600, 165600, 0, 622080, 0x3276ed72
0, 172800, 172800, 0, 622080, 0x264092b2
0, 180000, 180000, 0, 622080, 0x20ba1094
0, 187200, 187200, 0, 622080, 0x76cc3139
0, 194400, 194400, 0, 622080, 0x469a4902
0, 201600, 201600, 0, 622080, 0x0ed7b8f5
0, 208800, 208800, 0, 622080, 0xdc51aeac
0, 216000, 216000, 0, 622080, 0xee06aa36
0, 223200, 223200, 0, 622080, 0x7372405f
0, 230400, 230400, 0, 622080, 0x9e0ee776
0, 237600, 237600, 0, 622080, 0x39e6d8c9
0, 244800, 244800, 0, 622080, 0x51d9ac9a
0, 252000, 252000, 0, 622080, 0x2b63441d
0, 259200, 259200, 0, 622080, 0x58afbd5e
0, 266400, 266400, 0, 622080, 0xb972f716
0, 273600, 273600, 0, 622080, 0x6a6df129
0, 280800, 280800, 0, 622080, 0x28b1373d
#tb 0: 1/180000
0, 64800, 64800, 0, 622080, 0x4440caef
0, 68400, 68400, 0, 622080, 0xa5cea88b
0, 72000, 72000, 0, 622080, 0xce67e69d
0, 75600, 75600, 0, 622080, 0x9a57891f
0, 79200, 79200, 0, 622080, 0x1dbdc653
0, 82800, 82800, 0, 622080, 0xc171c0c5
0, 86400, 86400, 0, 622080, 0x82c591d1
0, 90000, 90000, 0, 622080, 0x20db9890
0, 93600, 93600, 0, 622080, 0x8193740b
0, 97200, 97200, 0, 622080, 0xdb181d52
0, 100800, 100800, 0, 622080, 0xcb219711
0, 104400, 104400, 0, 622080, 0xc2b913d1
0, 108000, 108000, 0, 622080, 0x1870783b
0, 111600, 111600, 0, 622080, 0xf1d9c5fb
0, 115200, 115200, 0, 622080, 0x7080590b
0, 118800, 118800, 0, 622080, 0x669c5775
0, 122400, 122400, 0, 622080, 0x6df4175d
0, 126000, 126000, 0, 622080, 0x01921a16
0, 129600, 129600, 0, 622080, 0x6b530e95
0, 133200, 133200, 0, 622080, 0xd5047bc9
0, 136800, 136800, 0, 622080, 0x7f9d66f7
0, 140400, 140400, 0, 622080, 0xa8b006eb
0, 144000, 144000, 0, 622080, 0x338cda81
0, 147600, 147600, 0, 622080, 0xf0e125a7
0, 151200, 151200, 0, 622080, 0xb13797f8
0, 154800, 154800, 0, 622080, 0x4afe2976
0, 158400, 158400, 0, 622080, 0xb51e7ca4
0, 162000, 162000, 0, 622080, 0x637fcbfe
0, 165600, 165600, 0, 622080, 0x353eed75
0, 169200, 169200, 0, 622080, 0xd9a8f5ac
0, 172800, 172800, 0, 622080, 0xf93e92b0
0, 176400, 176400, 0, 622080, 0x4540039f
0, 180000, 180000, 0, 622080, 0xd0811094
0, 183600, 183600, 0, 622080, 0x3039906f
0, 187200, 187200, 0, 622080, 0xb04a3141
0, 190800, 190800, 0, 622080, 0x638d2cf5
0, 194400, 194400, 0, 622080, 0x4ab84909
0, 198000, 198000, 0, 622080, 0x82de12ee
0, 201600, 201600, 0, 622080, 0xa0fcb8fb
0, 205200, 205200, 0, 622080, 0x7e849cc9
0, 208800, 208800, 0, 622080, 0x9003aebb
0, 212400, 212400, 0, 622080, 0xffe6f770
0, 216000, 216000, 0, 622080, 0x153faa3e
0, 219600, 219600, 0, 622080, 0xbf023231
0, 223200, 223200, 0, 622080, 0xae724063
0, 226800, 226800, 0, 622080, 0x15fe44b4
0, 230400, 230400, 0, 622080, 0xeb4de77a
0, 234000, 234000, 0, 622080, 0x380f8563
0, 237600, 237600, 0, 622080, 0x209ed8c7
0, 241200, 241200, 0, 622080, 0xb964d70f
0, 244800, 244800, 0, 622080, 0xe2bbac96
0, 248400, 248400, 0, 622080, 0x57e3f7f2
0, 252000, 252000, 0, 622080, 0xe945441e
0, 255600, 255600, 0, 622080, 0xd0afb742
0, 259200, 259200, 0, 622080, 0x8f8cbd5f
0, 262800, 262800, 0, 622080, 0xb9a15294
0, 266400, 266400, 0, 622080, 0xbc3cf717
0, 270000, 270000, 0, 622080, 0xb70b01a9
0, 273600, 273600, 0, 622080, 0x0109f125
0, 277200, 277200, 0, 622080, 0x5806371c
0, 280800, 280800, 0, 622080, 0x230c373f
0, 284400, 284400, 0, 622080, 0x82dfb1f2
0, 64800, 64800, 0, 622080, 0x6331caee
0, 68400, 68400, 0, 622080, 0x625da883
0, 72000, 72000, 0, 622080, 0xa459e690
0, 75600, 75600, 0, 622080, 0xce5d891e
0, 79200, 79200, 0, 622080, 0x6429c648
0, 82800, 82800, 0, 622080, 0x608cc0ba
0, 86400, 86400, 0, 622080, 0xa49891ca
0, 90000, 90000, 0, 622080, 0x9721987f
0, 93600, 93600, 0, 622080, 0x2a887404
0, 97200, 97200, 0, 622080, 0x60d71d47
0, 100800, 100800, 0, 622080, 0xe8d49705
0, 104400, 104400, 0, 622080, 0x821e13cb
0, 108000, 108000, 0, 622080, 0x1b627835
0, 111600, 111600, 0, 622080, 0x1806c5f4
0, 115200, 115200, 0, 622080, 0x686858fd
0, 118800, 118800, 0, 622080, 0xab865773
0, 122400, 122400, 0, 622080, 0x2675174f
0, 126000, 126000, 0, 622080, 0x43a61a14
0, 129600, 129600, 0, 622080, 0x78470e7f
0, 133200, 133200, 0, 622080, 0xeb877bc6
0, 136800, 136800, 0, 622080, 0xffb366ec
0, 140400, 140400, 0, 622080, 0xda0906e7
0, 144000, 144000, 0, 622080, 0xd575da72
0, 147600, 147600, 0, 622080, 0x23ae25a4
0, 151200, 151200, 0, 622080, 0x5fb297f7
0, 154800, 154800, 0, 622080, 0x99b32978
0, 158400, 158400, 0, 622080, 0xbac77ca0
0, 162000, 162000, 0, 622080, 0xc1cdcbf9
0, 165600, 165600, 0, 622080, 0x3276ed72
0, 169200, 169200, 0, 622080, 0x4061f5ab
0, 172800, 172800, 0, 622080, 0x264092b2
0, 176400, 176400, 0, 622080, 0xa4e2039e
0, 180000, 180000, 0, 622080, 0x20ba1094
0, 183600, 183600, 0, 622080, 0x984e906e
0, 187200, 187200, 0, 622080, 0x76cc3139
0, 190800, 190800, 0, 622080, 0xf70e2cf6
0, 194400, 194400, 0, 622080, 0x469a4902
0, 198000, 198000, 0, 622080, 0x235312e6
0, 201600, 201600, 0, 622080, 0x0ed7b8f5
0, 205200, 205200, 0, 622080, 0xd0269cc3
0, 208800, 208800, 0, 622080, 0xdc51aeac
0, 212400, 212400, 0, 622080, 0x1aa5f76e
0, 216000, 216000, 0, 622080, 0xee06aa36
0, 219600, 219600, 0, 622080, 0xa7103230
0, 223200, 223200, 0, 622080, 0x7372405f
0, 226800, 226800, 0, 622080, 0x8d7a44b5
0, 230400, 230400, 0, 622080, 0x9e0ee776
0, 234000, 234000, 0, 622080, 0xd41e8560
0, 237600, 237600, 0, 622080, 0x39e6d8c9
0, 241200, 241200, 0, 622080, 0x7a23d70c
0, 244800, 244800, 0, 622080, 0x51d9ac9a
0, 248400, 248400, 0, 622080, 0x8eacf7f2
0, 252000, 252000, 0, 622080, 0x2b63441d
0, 255600, 255600, 0, 622080, 0x9f71b742
0, 259200, 259200, 0, 622080, 0x58afbd5e
0, 262800, 262800, 0, 622080, 0x4d645292
0, 266400, 266400, 0, 622080, 0xb972f716
0, 270000, 270000, 0, 622080, 0xbb5d01a2
0, 273600, 273600, 0, 622080, 0x6a6df129
0, 277200, 277200, 0, 622080, 0x9e45371e
0, 280800, 280800, 0, 622080, 0x28b1373d
0, 284400, 284400, 0, 622080, 0xa1cdb1f2
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment