@@ -50,10 +50,10 @@ void ff_hqdn3d_row_10_x86(uint8_t *src, uint8_t *dst, uint16_t *line_ant, uint16

 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);

 #define LUT_BITS (depth==16 ? 8 : 4)

-#define RIGHTSHIFT(a,b) (((a)+(((1<<(b))-1)>>1))>>(b))

-#define LOAD(x) ((depth==8 ? src[x] : AV_RN16A(src+(x)*2)) << (16-depth))

-#define STORE(x,val) (depth==8 ? dst[x] = RIGHTSHIFT(val, 16-depth)\

-                    : AV_WN16A(dst+(x)*2, RIGHTSHIFT(val, 16-depth)))

+#define LOAD(x) (((depth == 8 ? src[x] : AV_RN16A(src + (x) * 2)) << (16 - depth))\

+                 + (((1 << (16 - depth)) - 1) >> 1))

+#define STORE(x,val) (depth == 8 ? dst[x] = (val) >> (16 - depth) : \

+                                   AV_WN16A(dst + (x) * 2, (val) >> (16 - depth)))

 av_always_inline

 static uint32_t lowpass(int prev, int cur, int16_t *coef, int depth)

@@ -39,6 +39,7 @@ SECTION .text

 %endif

 %if %3 != 16

     shl    %1, 16-%3

+    add    %1, (1<<(15-%3))-1

 %endif

 %endmacro

@@ -86,7 +87,6 @@ ALIGN 16

     mov       [frameantq+xq*2], t0w

     movifnidn dstq, dstmp

 %if %1 != 16

-    add    t0d, (1<<(15-%1))-1

     shr    t0d, 16-%1 ; could eliminate this by storing from t0h, but only with some contraints on register allocation

 %endif

 %if %1 == 8