diff options
Diffstat (limited to 'vpx_dsp/arm/mem_neon.h')
| -rw-r--r-- | vpx_dsp/arm/mem_neon.h | 165 |
1 files changed, 161 insertions, 4 deletions
diff --git a/vpx_dsp/arm/mem_neon.h b/vpx_dsp/arm/mem_neon.h index 50aaa94fe..19cfc7c7f 100644 --- a/vpx_dsp/arm/mem_neon.h +++ b/vpx_dsp/arm/mem_neon.h @@ -116,11 +116,11 @@ static INLINE void uint32_to_mem(uint8_t *buf, uint32_t a) { static INLINE uint8x8_t load_unaligned_u8(const uint8_t *buf, ptrdiff_t stride) { uint32_t a; - uint32x2_t a_u32 = vdup_n_u32(0); + uint32x2_t a_u32; if (stride == 4) return vld1_u8(buf); memcpy(&a, buf, 4); buf += stride; - a_u32 = vset_lane_u32(a, a_u32, 0); + a_u32 = vdup_n_u32(a); memcpy(&a, buf, 4); a_u32 = vset_lane_u32(a, a_u32, 1); return vreinterpret_u8_u32(a_u32); @@ -143,11 +143,11 @@ static INLINE void store_unaligned_u8(uint8_t *buf, ptrdiff_t stride, static INLINE uint8x16_t load_unaligned_u8q(const uint8_t *buf, ptrdiff_t stride) { uint32_t a; - uint32x4_t a_u32 = vdupq_n_u32(0); + uint32x4_t a_u32; if (stride == 4) return vld1q_u8(buf); memcpy(&a, buf, 4); buf += stride; - a_u32 = vsetq_lane_u32(a, a_u32, 0); + a_u32 = vdupq_n_u32(a); memcpy(&a, buf, 4); buf += stride; a_u32 = vsetq_lane_u32(a, a_u32, 1); @@ -201,4 +201,161 @@ static INLINE void store_u8(uint8_t *buf, ptrdiff_t stride, const uint8x8_t a) { buf += stride; vst1_lane_u32((uint32_t *)buf, a_u32, 1); } + +static INLINE void load_u8_8x4(const uint8_t *s, const ptrdiff_t p, + uint8x8_t *const s0, uint8x8_t *const s1, + uint8x8_t *const s2, uint8x8_t *const s3) { + *s0 = vld1_u8(s); + s += p; + *s1 = vld1_u8(s); + s += p; + *s2 = vld1_u8(s); + s += p; + *s3 = vld1_u8(s); +} + +static INLINE void store_u8_8x4(uint8_t *s, const ptrdiff_t p, + const uint8x8_t s0, const uint8x8_t s1, + const uint8x8_t s2, const uint8x8_t s3) { + vst1_u8(s, s0); + s += p; + vst1_u8(s, s1); + s += p; + vst1_u8(s, s2); + s += p; + vst1_u8(s, s3); +} + +static INLINE void load_u8_16x4(const uint8_t *s, const ptrdiff_t p, + uint8x16_t *const s0, uint8x16_t *const s1, + uint8x16_t *const s2, uint8x16_t *const s3) { + *s0 = vld1q_u8(s); + s += p; + *s1 = vld1q_u8(s); + s += p; + *s2 = vld1q_u8(s); + s += p; + *s3 = vld1q_u8(s); +} + +static INLINE void store_u8_16x4(uint8_t *s, const ptrdiff_t p, + const uint8x16_t s0, const uint8x16_t s1, + const uint8x16_t s2, const uint8x16_t s3) { + vst1q_u8(s, s0); + s += p; + vst1q_u8(s, s1); + s += p; + vst1q_u8(s, s2); + s += p; + vst1q_u8(s, s3); +} + +static INLINE void load_u8_8x7(const uint8_t *s, const ptrdiff_t p, + uint8x8_t *const s0, uint8x8_t *const s1, + uint8x8_t *const s2, uint8x8_t *const s3, + uint8x8_t *const s4, uint8x8_t *const s5, + uint8x8_t *const s6) { + *s0 = vld1_u8(s); + s += p; + *s1 = vld1_u8(s); + s += p; + *s2 = vld1_u8(s); + s += p; + *s3 = vld1_u8(s); + s += p; + *s4 = vld1_u8(s); + s += p; + *s5 = vld1_u8(s); + s += p; + *s6 = vld1_u8(s); +} + +static INLINE void load_u8_8x8(const uint8_t *s, const ptrdiff_t p, + uint8x8_t *const s0, uint8x8_t *const s1, + uint8x8_t *const s2, uint8x8_t *const s3, + uint8x8_t *const s4, uint8x8_t *const s5, + uint8x8_t *const s6, uint8x8_t *const s7) { + *s0 = vld1_u8(s); + s += p; + *s1 = vld1_u8(s); + s += p; + *s2 = vld1_u8(s); + s += p; + *s3 = vld1_u8(s); + s += p; + *s4 = vld1_u8(s); + s += p; + *s5 = vld1_u8(s); + s += p; + *s6 = vld1_u8(s); + s += p; + *s7 = vld1_u8(s); +} + +static INLINE void store_u8_8x8(uint8_t *s, const ptrdiff_t p, + const uint8x8_t s0, const uint8x8_t s1, + const uint8x8_t s2, const uint8x8_t s3, + const uint8x8_t s4, const uint8x8_t s5, + const uint8x8_t s6, const uint8x8_t s7) { + vst1_u8(s, s0); + s += p; + vst1_u8(s, s1); + s += p; + vst1_u8(s, s2); + s += p; + vst1_u8(s, s3); + s += p; + vst1_u8(s, s4); + s += p; + vst1_u8(s, s5); + s += p; + vst1_u8(s, s6); + s += p; + vst1_u8(s, s7); +} + +static INLINE void load_u8_16x8(const uint8_t *s, const ptrdiff_t p, + uint8x16_t *const s0, uint8x16_t *const s1, + uint8x16_t *const s2, uint8x16_t *const s3, + uint8x16_t *const s4, uint8x16_t *const s5, + uint8x16_t *const s6, uint8x16_t *const s7) { + *s0 = vld1q_u8(s); + s += p; + *s1 = vld1q_u8(s); + s += p; + *s2 = vld1q_u8(s); + s += p; + *s3 = vld1q_u8(s); + s += p; + *s4 = vld1q_u8(s); + s += p; + *s5 = vld1q_u8(s); + s += p; + *s6 = vld1q_u8(s); + s += p; + *s7 = vld1q_u8(s); +} + +static INLINE void store_u8_16x8(uint8_t *s, const ptrdiff_t p, + const uint8x16_t s0, const uint8x16_t s1, + const uint8x16_t s2, const uint8x16_t s3, + const uint8x16_t s4, const uint8x16_t s5, + const uint8x16_t s6, const uint8x16_t s7) { + vst1q_u8(s, s0); + s += p; + vst1q_u8(s, s1); + s += p; + vst1q_u8(s, s2); + s += p; + vst1q_u8(s, s3); + s += p; + vst1q_u8(s, s4); + s += p; + vst1q_u8(s, s5); + s += p; + vst1q_u8(s, s6); + s += p; + vst1q_u8(s, s7); +} + #endif // VPX_VPX_DSP_ARM_MEM_NEON_H_ |