diff options
Diffstat (limited to 'vp9/encoder/arm')
| -rw-r--r-- | vp9/encoder/arm/neon/vp9_denoiser_neon.c | 2 | ||||
| -rw-r--r-- | vp9/encoder/arm/neon/vp9_diamond_search_sad_neon.c | 14 | ||||
| -rw-r--r-- | vp9/encoder/arm/neon/vp9_quantize_neon.c | 6 |
3 files changed, 11 insertions, 11 deletions
diff --git a/vp9/encoder/arm/neon/vp9_denoiser_neon.c b/vp9/encoder/arm/neon/vp9_denoiser_neon.c index 53e8c7e49..d631cd437 100644 --- a/vp9/encoder/arm/neon/vp9_denoiser_neon.c +++ b/vp9/encoder/arm/neon/vp9_denoiser_neon.c @@ -21,7 +21,7 @@ // Compute the sum of all pixel differences of this MB. static INLINE int horizontal_add_s8x16(const int8x16_t v_sum_diff_total) { -#if defined(__aarch64__) +#if VPX_ARCH_AARCH64 return vaddlvq_s8(v_sum_diff_total); #else const int16x8_t fe_dc_ba_98_76_54_32_10 = vpaddlq_s8(v_sum_diff_total); diff --git a/vp9/encoder/arm/neon/vp9_diamond_search_sad_neon.c b/vp9/encoder/arm/neon/vp9_diamond_search_sad_neon.c index 255e6fbc4..b82b3f9db 100644 --- a/vp9/encoder/arm/neon/vp9_diamond_search_sad_neon.c +++ b/vp9/encoder/arm/neon/vp9_diamond_search_sad_neon.c @@ -94,7 +94,7 @@ int vp9_diamond_search_sad_neon(const MACROBLOCK *x, // Work out the start point for the search const uint8_t *best_address = in_what; const uint8_t *new_best_address = best_address; -#if defined(__aarch64__) +#if VPX_ARCH_AARCH64 int64x2_t v_ba_q = vdupq_n_s64((intptr_t)best_address); #else int32x4_t v_ba_d = vdupq_n_s32((intptr_t)best_address); @@ -117,7 +117,7 @@ int vp9_diamond_search_sad_neon(const MACROBLOCK *x, int8x16_t v_inside_d; uint32x4_t v_outside_d; int32x4_t v_cost_d, v_sad_d; -#if defined(__aarch64__) +#if VPX_ARCH_AARCH64 int64x2_t v_blocka[2]; #else int32x4_t v_blocka[1]; @@ -138,7 +138,7 @@ int vp9_diamond_search_sad_neon(const MACROBLOCK *x, vreinterpretq_s32_s16(v_these_mv_w))); // If none of them are inside, then move on -#if defined(__aarch64__) +#if VPX_ARCH_AARCH64 horiz_max = vmaxvq_u32(vreinterpretq_u32_s8(v_inside_d)); #else horiz_max_0 = vmax_u32(vget_low_u32(vreinterpretq_u32_s8(v_inside_d)), @@ -167,7 +167,7 @@ int vp9_diamond_search_sad_neon(const MACROBLOCK *x, // Compute the SIMD pointer offsets. { -#if defined(__aarch64__) // sizeof(intptr_t) == 8 +#if VPX_ARCH_AARCH64 // sizeof(intptr_t) == 8 // Load the offsets int64x2_t v_bo10_q = vld1q_s64((const int64_t *)&ss_os[i + 0]); int64x2_t v_bo32_q = vld1q_s64((const int64_t *)&ss_os[i + 2]); @@ -234,7 +234,7 @@ int vp9_diamond_search_sad_neon(const MACROBLOCK *x, // Find the minimum value and index horizontally in v_sad_d { uint32_t local_best_sad; -#if defined(__aarch64__) +#if VPX_ARCH_AARCH64 local_best_sad = vminvq_u32(vreinterpretq_u32_s32(v_sad_d)); #else uint32x2_t horiz_min_0 = @@ -256,7 +256,7 @@ int vp9_diamond_search_sad_neon(const MACROBLOCK *x, uint32x4_t v_mask_d = vandq_u32(v_sel_d, v_idx_d); v_mask_d = vbslq_u32(v_sel_d, v_mask_d, vdupq_n_u32(0xffffffff)); -#if defined(__aarch64__) +#if VPX_ARCH_AARCH64 local_best_idx = vminvq_u32(v_mask_d); #else horiz_min_0 = @@ -280,7 +280,7 @@ int vp9_diamond_search_sad_neon(const MACROBLOCK *x, best_address = new_best_address; v_bmv_w = vreinterpretq_s16_s32(vdupq_n_s32(bmv.as_int)); -#if defined(__aarch64__) +#if VPX_ARCH_AARCH64 v_ba_q = vdupq_n_s64((intptr_t)best_address); #else v_ba_d = vdupq_n_s32((intptr_t)best_address); diff --git a/vp9/encoder/arm/neon/vp9_quantize_neon.c b/vp9/encoder/arm/neon/vp9_quantize_neon.c index c2b55fcba..97ab13628 100644 --- a/vp9/encoder/arm/neon/vp9_quantize_neon.c +++ b/vp9/encoder/arm/neon/vp9_quantize_neon.c @@ -50,7 +50,7 @@ static VPX_FORCE_INLINE int16x8_t get_max_lane_eob(const int16_t *iscan_ptr, } static VPX_FORCE_INLINE uint16_t get_max_eob(int16x8_t v_eobmax) { -#ifdef __aarch64__ +#if VPX_ARCH_AARCH64 return (uint16_t)vmaxvq_s16(v_eobmax); #else const int16x4_t v_eobmax_3210 = @@ -65,7 +65,7 @@ static VPX_FORCE_INLINE uint16_t get_max_eob(int16x8_t v_eobmax) { vmax_s16(v_eobmax_tmp, vreinterpret_s16_s64(v_eobmax_xxx3)); return (uint16_t)vget_lane_s16(v_eobmax_final, 0); -#endif // __aarch64__ +#endif // VPX_ARCH_AARCH64 } static VPX_FORCE_INLINE void load_fp_values(const int16_t *round_ptr, @@ -81,7 +81,7 @@ static VPX_FORCE_INLINE void load_fp_values(const int16_t *round_ptr, static VPX_FORCE_INLINE void update_fp_values(int16x8_t *v_round, int16x8_t *v_quant, int16x8_t *v_dequant) { -#ifdef __aarch64__ +#if VPX_ARCH_AARCH64 *v_round = vdupq_laneq_s16(*v_round, 1); *v_quant = vdupq_laneq_s16(*v_quant, 1); *v_dequant = vdupq_laneq_s16(*v_dequant, 1); |