aboutsummaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
authorMans Rullgard <mans@mansr.com>2012-08-12 19:45:46 +0100
committerMans Rullgard <mans@mansr.com>2012-08-13 14:51:52 +0100
commit480178a29587df8ed6d5e93bfe79e4a08a61f9e1 (patch)
treee6445d7958e70a7e9f4971179bf80030692a9f15
parent8ec0204ee4ee93218c51a86d2faa24937c8108e7 (diff)
downloadffmpeg-480178a29587df8ed6d5e93bfe79e4a08a61f9e1.tar.gz
x86: yadif: fix asm with suncc
Under some circumstances, suncc will use a single register for the address of all memory operands, inserting lea instructions loading the correct address prior to each memory operand being used in the code. In the yadif code, the branch in the asm block bypasses such an lea instruction, causing an incorrect address to be used in the following load. This patch replaces the tmpX arrays with a single array and uses a register operand to hold its address. Although this prevents using offsets from the stack pointer to access these locations, the code still builds as 32-bit PIC even with old compilers. Signed-off-by: Mans Rullgard <mans@mansr.com>
-rw-r--r--libavfilter/x86/yadif_template.c32
1 files changed, 13 insertions, 19 deletions
diff --git a/libavfilter/x86/yadif_template.c b/libavfilter/x86/yadif_template.c
index 1de0a58e8f..3e45f4fda0 100644
--- a/libavfilter/x86/yadif_template.c
+++ b/libavfilter/x86/yadif_template.c
@@ -107,10 +107,7 @@ static void RENAME(yadif_filter_line)(uint8_t *dst, uint8_t *prev, uint8_t *cur,
uint8_t *next, int w, int prefs,
int mrefs, int parity, int mode)
{
- DECLARE_ALIGNED(16, uint8_t, tmp0)[16];
- DECLARE_ALIGNED(16, uint8_t, tmp1)[16];
- DECLARE_ALIGNED(16, uint8_t, tmp2)[16];
- DECLARE_ALIGNED(16, uint8_t, tmp3)[16];
+ DECLARE_ALIGNED(16, uint8_t, tmp)[16*4];
int x;
#define FILTER\
@@ -124,9 +121,9 @@ static void RENAME(yadif_filter_line)(uint8_t *dst, uint8_t *prev, uint8_t *cur,
MOVQ" "MM"3, "MM"4 \n\t"\
"paddw "MM"2, "MM"3 \n\t"\
"psraw $1, "MM"3 \n\t" /* d = (prev2[x] + next2[x])>>1 */\
- MOVQ" "MM"0, %[tmp0] \n\t" /* c */\
- MOVQ" "MM"3, %[tmp1] \n\t" /* d */\
- MOVQ" "MM"1, %[tmp2] \n\t" /* e */\
+ MOVQ" "MM"0, (%[tmp]) \n\t" /* c */\
+ MOVQ" "MM"3, 16(%[tmp]) \n\t" /* d */\
+ MOVQ" "MM"1, 32(%[tmp]) \n\t" /* e */\
"psubw "MM"4, "MM"2 \n\t"\
PABS( MM"4", MM"2") /* temporal_diff0 */\
LOAD("(%[prev],%[mrefs])", MM"3") /* prev[x-refs] */\
@@ -148,7 +145,7 @@ static void RENAME(yadif_filter_line)(uint8_t *dst, uint8_t *prev, uint8_t *cur,
"paddw "MM"4, "MM"3 \n\t" /* temporal_diff2 */\
"psrlw $1, "MM"3 \n\t"\
"pmaxsw "MM"3, "MM"2 \n\t"\
- MOVQ" "MM"2, %[tmp3] \n\t" /* diff */\
+ MOVQ" "MM"2, 48(%[tmp]) \n\t" /* diff */\
\
"paddw "MM"0, "MM"1 \n\t"\
"paddw "MM"0, "MM"0 \n\t"\
@@ -179,7 +176,7 @@ static void RENAME(yadif_filter_line)(uint8_t *dst, uint8_t *prev, uint8_t *cur,
CHECK2\
\
/* if(p->mode<2) ... */\
- MOVQ" %[tmp3], "MM"6 \n\t" /* diff */\
+ MOVQ" 48(%[tmp]), "MM"6 \n\t" /* diff */\
"cmpl $2, %[mode] \n\t"\
"jge 1f \n\t"\
LOAD("(%["prev2"],%[mrefs],2)", MM"2") /* prev2[x-2*refs] */\
@@ -190,9 +187,9 @@ static void RENAME(yadif_filter_line)(uint8_t *dst, uint8_t *prev, uint8_t *cur,
"paddw "MM"5, "MM"3 \n\t"\
"psrlw $1, "MM"2 \n\t" /* b */\
"psrlw $1, "MM"3 \n\t" /* f */\
- MOVQ" %[tmp0], "MM"4 \n\t" /* c */\
- MOVQ" %[tmp1], "MM"5 \n\t" /* d */\
- MOVQ" %[tmp2], "MM"7 \n\t" /* e */\
+ MOVQ" (%[tmp]), "MM"4 \n\t" /* c */\
+ MOVQ" 16(%[tmp]), "MM"5 \n\t" /* d */\
+ MOVQ" 32(%[tmp]), "MM"7 \n\t" /* e */\
"psubw "MM"4, "MM"2 \n\t" /* b-c */\
"psubw "MM"7, "MM"3 \n\t" /* f-e */\
MOVQ" "MM"5, "MM"0 \n\t"\
@@ -211,7 +208,7 @@ static void RENAME(yadif_filter_line)(uint8_t *dst, uint8_t *prev, uint8_t *cur,
"pmaxsw "MM"4, "MM"6 \n\t" /* diff= MAX3(diff, min, -max); */\
"1: \n\t"\
\
- MOVQ" %[tmp1], "MM"2 \n\t" /* d */\
+ MOVQ" 16(%[tmp]), "MM"2 \n\t" /* d */\
MOVQ" "MM"2, "MM"3 \n\t"\
"psubw "MM"6, "MM"2 \n\t" /* d-diff */\
"paddw "MM"6, "MM"3 \n\t" /* d+diff */\
@@ -219,16 +216,13 @@ static void RENAME(yadif_filter_line)(uint8_t *dst, uint8_t *prev, uint8_t *cur,
"pminsw "MM"3, "MM"1 \n\t" /* d = clip(spatial_pred, d-diff, d+diff); */\
"packuswb "MM"1, "MM"1 \n\t"\
\
- :[tmp0]"=m"(tmp0),\
- [tmp1]"=m"(tmp1),\
- [tmp2]"=m"(tmp2),\
- [tmp3]"=m"(tmp3)\
- :[prev] "r"(prev),\
+ ::[prev] "r"(prev),\
[cur] "r"(cur),\
[next] "r"(next),\
[prefs]"r"((x86_reg)prefs),\
[mrefs]"r"((x86_reg)mrefs),\
- [mode] "g"(mode)\
+ [mode] "g"(mode),\
+ [tmp] "r"(tmp)\
);\
__asm__ volatile(MOV" "MM"1, %0" :"=m"(*dst));\
dst += STEP;\