diff options
Diffstat (limited to 'thirdparty/libwebp/src/dsp/dec_neon.c')
| -rw-r--r-- | thirdparty/libwebp/src/dsp/dec_neon.c | 9 |
1 files changed, 8 insertions, 1 deletions
diff --git a/thirdparty/libwebp/src/dsp/dec_neon.c b/thirdparty/libwebp/src/dsp/dec_neon.c index ffa697fcf9..239ec4167e 100644 --- a/thirdparty/libwebp/src/dsp/dec_neon.c +++ b/thirdparty/libwebp/src/dsp/dec_neon.c @@ -1361,7 +1361,8 @@ static void RD4_NEON(uint8_t* dst) { // Down-right const uint32_t J = dst[-1 + 1 * BPS]; const uint32_t K = dst[-1 + 2 * BPS]; const uint32_t L = dst[-1 + 3 * BPS]; - const uint64x1_t LKJI____ = vcreate_u64(L | (K << 8) | (J << 16) | (I << 24)); + const uint64x1_t LKJI____ = + vcreate_u64((uint64_t)L | (K << 8) | (J << 16) | (I << 24)); const uint64x1_t LKJIXABC = vorr_u64(LKJI____, ____XABC); const uint8x8_t KJIXABC_ = vreinterpret_u8_u64(vshr_n_u64(LKJIXABC, 8)); const uint8x8_t JIXABC__ = vreinterpret_u8_u64(vshr_n_u64(LKJIXABC, 16)); @@ -1427,10 +1428,16 @@ static WEBP_INLINE void DC8_NEON(uint8_t* dst, int do_top, int do_left) { if (do_top) { const uint8x8_t A = vld1_u8(dst - BPS); // top row +#if defined(__aarch64__) + const uint16x8_t B = vmovl_u8(A); + const uint16_t p2 = vaddvq_u16(B); + sum_top = vdupq_n_u16(p2); +#else const uint16x4_t p0 = vpaddl_u8(A); // cascading summation of the top const uint16x4_t p1 = vpadd_u16(p0, p0); const uint16x4_t p2 = vpadd_u16(p1, p1); sum_top = vcombine_u16(p2, p2); +#endif } if (do_left) { |