CC:libav-stable@libav.org
Signed-off-by: Anton Khirnov <anton@khirnov.net>
(cherry picked from commit 5b3c1aecb253828d09fa9825c5a4aed97badf086)
Signed-off-by: Reinhard Tartler <siretart@tauware.de>
... | ... |
@@ -50,10 +50,10 @@ void ff_hqdn3d_row_10_x86(uint8_t *src, uint8_t *dst, uint16_t *line_ant, uint16 |
50 | 50 |
void ff_hqdn3d_row_16_x86(uint8_t *src, uint8_t *dst, uint16_t *line_ant, uint16_t *frame_ant, ptrdiff_t w, int16_t *spatial, int16_t *temporal); |
51 | 51 |
|
52 | 52 |
#define LUT_BITS (depth==16 ? 8 : 4) |
53 |
-#define RIGHTSHIFT(a,b) (((a)+(((1<<(b))-1)>>1))>>(b)) |
|
54 |
-#define LOAD(x) ((depth==8 ? src[x] : AV_RN16A(src+(x)*2)) << (16-depth)) |
|
55 |
-#define STORE(x,val) (depth==8 ? dst[x] = RIGHTSHIFT(val, 16-depth)\ |
|
56 |
- : AV_WN16A(dst+(x)*2, RIGHTSHIFT(val, 16-depth))) |
|
53 |
+#define LOAD(x) (((depth == 8 ? src[x] : AV_RN16A(src + (x) * 2)) << (16 - depth))\ |
|
54 |
+ + (((1 << (16 - depth)) - 1) >> 1)) |
|
55 |
+#define STORE(x,val) (depth == 8 ? dst[x] = (val) >> (16 - depth) : \ |
|
56 |
+ AV_WN16A(dst + (x) * 2, (val) >> (16 - depth))) |
|
57 | 57 |
|
58 | 58 |
av_always_inline |
59 | 59 |
static uint32_t lowpass(int prev, int cur, int16_t *coef, int depth) |
... | ... |
@@ -39,6 +39,7 @@ SECTION .text |
39 | 39 |
%endif |
40 | 40 |
%if %3 != 16 |
41 | 41 |
shl %1, 16-%3 |
42 |
+ add %1, (1<<(15-%3))-1 |
|
42 | 43 |
%endif |
43 | 44 |
%endmacro |
44 | 45 |
|
... | ... |
@@ -86,7 +87,6 @@ ALIGN 16 |
86 | 86 |
mov [frameantq+xq*2], t0w |
87 | 87 |
movifnidn dstq, dstmp |
88 | 88 |
%if %1 != 16 |
89 |
- add t0d, (1<<(15-%1))-1 |
|
90 | 89 |
shr t0d, 16-%1 ; could eliminate this by storing from t0h, but only with some contraints on register allocation |
91 | 90 |
%endif |
92 | 91 |
%if %1 == 8 |