diff options
| -rw-r--r-- | vp8/common/x86/loopfilter_x86.c | 17 |
1 files changed, 8 insertions, 9 deletions
diff --git a/vp8/common/x86/loopfilter_x86.c b/vp8/common/x86/loopfilter_x86.c index 1fa8ed431..4d490a613 100644 --- a/vp8/common/x86/loopfilter_x86.c +++ b/vp8/common/x86/loopfilter_x86.c @@ -335,7 +335,6 @@ static __inline void transpose(unsigned char *src[], int in_p, int num_8x8_to_transpose) { int idx8x8 = 0; __m128i x0, x1, x2, x3, x4, x5, x6, x7; - do { unsigned char *in = src[idx8x8]; unsigned char *out = dst[idx8x8]; @@ -366,13 +365,13 @@ static __inline void transpose(unsigned char *src[], int in_p, x7 = _mm_unpackhi_epi32(x4, x5); _mm_storel_pd((double *)(out + 0*out_p), - (__m128d)x6); // 00 10 20 30 40 50 60 70 + _mm_castsi128_pd(x6)); // 00 10 20 30 40 50 60 70 _mm_storeh_pd((double *)(out + 1*out_p), - (__m128d)x6); // 01 11 21 31 41 51 61 71 + _mm_castsi128_pd(x6)); // 01 11 21 31 41 51 61 71 _mm_storel_pd((double *)(out + 2*out_p), - (__m128d)x7); // 02 12 22 32 42 52 62 72 + _mm_castsi128_pd(x7)); // 02 12 22 32 42 52 62 72 _mm_storeh_pd((double *)(out + 3*out_p), - (__m128d)x7); // 03 13 23 33 43 53 63 73 + _mm_castsi128_pd(x7)); // 03 13 23 33 43 53 63 73 // 04 14 24 34 05 15 25 35 06 16 26 36 07 17 27 37 x4 = _mm_unpackhi_epi16(x0, x1); @@ -384,13 +383,13 @@ static __inline void transpose(unsigned char *src[], int in_p, x7 = _mm_unpackhi_epi32(x4, x5); _mm_storel_pd((double *)(out + 4*out_p), - (__m128d)x6); // 04 14 24 34 44 54 64 74 + _mm_castsi128_pd(x6)); // 04 14 24 34 44 54 64 74 _mm_storeh_pd((double *)(out + 5*out_p), - (__m128d)x6); // 05 15 25 35 45 55 65 75 + _mm_castsi128_pd(x6)); // 05 15 25 35 45 55 65 75 _mm_storel_pd((double *)(out + 6*out_p), - (__m128d)x7); // 06 16 26 36 46 56 66 76 + _mm_castsi128_pd(x7)); // 06 16 26 36 46 56 66 76 _mm_storeh_pd((double *)(out + 7*out_p), - (__m128d)x7); // 07 17 27 37 47 57 67 77 + _mm_castsi128_pd(x7)); // 07 17 27 37 47 57 67 77 } while (++idx8x8 < num_8x8_to_transpose); } void vp8_mbloop_filter_vertical_edge_c_sse2 |