From f2d621e38390a6c69ce2ce86c46837c65a6aa4d6 Mon Sep 17 00:00:00 2001 From: James Zern Date: Fri, 29 May 2015 19:01:39 -0700 Subject: vp9_reconintra_neon: cosmetics: normalize fn params s/y_stride/stride/ Change-Id: Ie98c3fe241dc240b653849eda356a8862bdd52f4 --- vp9/common/arm/neon/vp9_reconintra_neon.c | 110 +++++++++++++++--------------- 1 file changed, 55 insertions(+), 55 deletions(-) (limited to 'vp9') diff --git a/vp9/common/arm/neon/vp9_reconintra_neon.c b/vp9/common/arm/neon/vp9_reconintra_neon.c index 387439fe9..53e640640 100644 --- a/vp9/common/arm/neon/vp9_reconintra_neon.c +++ b/vp9/common/arm/neon/vp9_reconintra_neon.c @@ -163,40 +163,40 @@ void vp9_dc_128_predictor_16x16_neon(uint8_t *dst, ptrdiff_t stride, #if !HAVE_NEON_ASM -void vp9_v_predictor_4x4_neon(uint8_t *dst, ptrdiff_t y_stride, +void vp9_v_predictor_4x4_neon(uint8_t *dst, ptrdiff_t stride, const uint8_t *above, const uint8_t *left) { int i; uint32x2_t d0u32 = vdup_n_u32(0); (void)left; d0u32 = vld1_lane_u32((const uint32_t *)above, d0u32, 0); - for (i = 0; i < 4; i++, dst += y_stride) + for (i = 0; i < 4; i++, dst += stride) vst1_lane_u32((uint32_t *)dst, d0u32, 0); } -void vp9_v_predictor_8x8_neon(uint8_t *dst, ptrdiff_t y_stride, +void vp9_v_predictor_8x8_neon(uint8_t *dst, ptrdiff_t stride, const uint8_t *above, const uint8_t *left) { int i; uint8x8_t d0u8 = vdup_n_u8(0); (void)left; d0u8 = vld1_u8(above); - for (i = 0; i < 8; i++, dst += y_stride) + for (i = 0; i < 8; i++, dst += stride) vst1_u8(dst, d0u8); } -void vp9_v_predictor_16x16_neon(uint8_t *dst, ptrdiff_t y_stride, +void vp9_v_predictor_16x16_neon(uint8_t *dst, ptrdiff_t stride, const uint8_t *above, const uint8_t *left) { int i; uint8x16_t q0u8 = vdupq_n_u8(0); (void)left; q0u8 = vld1q_u8(above); - for (i = 0; i < 16; i++, dst += y_stride) + for (i = 0; i < 16; i++, dst += stride) vst1q_u8(dst, q0u8); } -void vp9_v_predictor_32x32_neon(uint8_t *dst, ptrdiff_t y_stride, +void vp9_v_predictor_32x32_neon(uint8_t *dst, ptrdiff_t stride, const uint8_t *above, const uint8_t *left) { int i; uint8x16_t q0u8 = vdupq_n_u8(0); @@ -205,13 +205,13 @@ void vp9_v_predictor_32x32_neon(uint8_t *dst, ptrdiff_t y_stride, q0u8 = vld1q_u8(above); q1u8 = vld1q_u8(above + 16); - for (i = 0; i < 32; i++, dst += y_stride) { + for (i = 0; i < 32; i++, dst += stride) { vst1q_u8(dst, q0u8); vst1q_u8(dst + 16, q1u8); } } -void vp9_h_predictor_4x4_neon(uint8_t *dst, ptrdiff_t y_stride, +void vp9_h_predictor_4x4_neon(uint8_t *dst, ptrdiff_t stride, const uint8_t *above, const uint8_t *left) { uint8x8_t d0u8 = vdup_n_u8(0); uint32x2_t d1u32 = vdup_n_u32(0); @@ -221,18 +221,18 @@ void vp9_h_predictor_4x4_neon(uint8_t *dst, ptrdiff_t y_stride, d0u8 = vdup_lane_u8(vreinterpret_u8_u32(d1u32), 0); vst1_lane_u32((uint32_t *)dst, vreinterpret_u32_u8(d0u8), 0); - dst += y_stride; + dst += stride; d0u8 = vdup_lane_u8(vreinterpret_u8_u32(d1u32), 1); vst1_lane_u32((uint32_t *)dst, vreinterpret_u32_u8(d0u8), 0); - dst += y_stride; + dst += stride; d0u8 = vdup_lane_u8(vreinterpret_u8_u32(d1u32), 2); vst1_lane_u32((uint32_t *)dst, vreinterpret_u32_u8(d0u8), 0); - dst += y_stride; + dst += stride; d0u8 = vdup_lane_u8(vreinterpret_u8_u32(d1u32), 3); vst1_lane_u32((uint32_t *)dst, vreinterpret_u32_u8(d0u8), 0); } -void vp9_h_predictor_8x8_neon(uint8_t *dst, ptrdiff_t y_stride, +void vp9_h_predictor_8x8_neon(uint8_t *dst, ptrdiff_t stride, const uint8_t *above, const uint8_t *left) { uint8x8_t d0u8 = vdup_n_u8(0); uint64x1_t d1u64 = vdup_n_u64(0); @@ -242,30 +242,30 @@ void vp9_h_predictor_8x8_neon(uint8_t *dst, ptrdiff_t y_stride, d0u8 = vdup_lane_u8(vreinterpret_u8_u64(d1u64), 0); vst1_u8(dst, d0u8); - dst += y_stride; + dst += stride; d0u8 = vdup_lane_u8(vreinterpret_u8_u64(d1u64), 1); vst1_u8(dst, d0u8); - dst += y_stride; + dst += stride; d0u8 = vdup_lane_u8(vreinterpret_u8_u64(d1u64), 2); vst1_u8(dst, d0u8); - dst += y_stride; + dst += stride; d0u8 = vdup_lane_u8(vreinterpret_u8_u64(d1u64), 3); vst1_u8(dst, d0u8); - dst += y_stride; + dst += stride; d0u8 = vdup_lane_u8(vreinterpret_u8_u64(d1u64), 4); vst1_u8(dst, d0u8); - dst += y_stride; + dst += stride; d0u8 = vdup_lane_u8(vreinterpret_u8_u64(d1u64), 5); vst1_u8(dst, d0u8); - dst += y_stride; + dst += stride; d0u8 = vdup_lane_u8(vreinterpret_u8_u64(d1u64), 6); vst1_u8(dst, d0u8); - dst += y_stride; + dst += stride; d0u8 = vdup_lane_u8(vreinterpret_u8_u64(d1u64), 7); vst1_u8(dst, d0u8); } -void vp9_h_predictor_16x16_neon(uint8_t *dst, ptrdiff_t y_stride, +void vp9_h_predictor_16x16_neon(uint8_t *dst, ptrdiff_t stride, const uint8_t *above, const uint8_t *left) { int j; uint8x8_t d2u8 = vdup_n_u8(0); @@ -278,32 +278,32 @@ void vp9_h_predictor_16x16_neon(uint8_t *dst, ptrdiff_t y_stride, for (j = 0; j < 2; j++, d2u8 = vget_high_u8(q1u8)) { q0u8 = vdupq_lane_u8(d2u8, 0); vst1q_u8(dst, q0u8); - dst += y_stride; + dst += stride; q0u8 = vdupq_lane_u8(d2u8, 1); vst1q_u8(dst, q0u8); - dst += y_stride; + dst += stride; q0u8 = vdupq_lane_u8(d2u8, 2); vst1q_u8(dst, q0u8); - dst += y_stride; + dst += stride; q0u8 = vdupq_lane_u8(d2u8, 3); vst1q_u8(dst, q0u8); - dst += y_stride; + dst += stride; q0u8 = vdupq_lane_u8(d2u8, 4); vst1q_u8(dst, q0u8); - dst += y_stride; + dst += stride; q0u8 = vdupq_lane_u8(d2u8, 5); vst1q_u8(dst, q0u8); - dst += y_stride; + dst += stride; q0u8 = vdupq_lane_u8(d2u8, 6); vst1q_u8(dst, q0u8); - dst += y_stride; + dst += stride; q0u8 = vdupq_lane_u8(d2u8, 7); vst1q_u8(dst, q0u8); - dst += y_stride; + dst += stride; } } -void vp9_h_predictor_32x32_neon(uint8_t *dst, ptrdiff_t y_stride, +void vp9_h_predictor_32x32_neon(uint8_t *dst, ptrdiff_t stride, const uint8_t *above, const uint8_t *left) { int j, k; uint8x8_t d2u8 = vdup_n_u8(0); @@ -318,40 +318,40 @@ void vp9_h_predictor_32x32_neon(uint8_t *dst, ptrdiff_t y_stride, q0u8 = vdupq_lane_u8(d2u8, 0); vst1q_u8(dst, q0u8); vst1q_u8(dst + 16, q0u8); - dst += y_stride; + dst += stride; q0u8 = vdupq_lane_u8(d2u8, 1); vst1q_u8(dst, q0u8); vst1q_u8(dst + 16, q0u8); - dst += y_stride; + dst += stride; q0u8 = vdupq_lane_u8(d2u8, 2); vst1q_u8(dst, q0u8); vst1q_u8(dst + 16, q0u8); - dst += y_stride; + dst += stride; q0u8 = vdupq_lane_u8(d2u8, 3); vst1q_u8(dst, q0u8); vst1q_u8(dst + 16, q0u8); - dst += y_stride; + dst += stride; q0u8 = vdupq_lane_u8(d2u8, 4); vst1q_u8(dst, q0u8); vst1q_u8(dst + 16, q0u8); - dst += y_stride; + dst += stride; q0u8 = vdupq_lane_u8(d2u8, 5); vst1q_u8(dst, q0u8); vst1q_u8(dst + 16, q0u8); - dst += y_stride; + dst += stride; q0u8 = vdupq_lane_u8(d2u8, 6); vst1q_u8(dst, q0u8); vst1q_u8(dst + 16, q0u8); - dst += y_stride; + dst += stride; q0u8 = vdupq_lane_u8(d2u8, 7); vst1q_u8(dst, q0u8); vst1q_u8(dst + 16, q0u8); - dst += y_stride; + dst += stride; } } } -void vp9_tm_predictor_4x4_neon(uint8_t *dst, ptrdiff_t y_stride, +void vp9_tm_predictor_4x4_neon(uint8_t *dst, ptrdiff_t stride, const uint8_t *above, const uint8_t *left) { int i; uint16x8_t q1u16, q3u16; @@ -362,7 +362,7 @@ void vp9_tm_predictor_4x4_neon(uint8_t *dst, ptrdiff_t y_stride, d0u8 = vdup_n_u8(above[-1]); d2u32 = vld1_lane_u32((const uint32_t *)above, d2u32, 0); q3u16 = vsubl_u8(vreinterpret_u8_u32(d2u32), d0u8); - for (i = 0; i < 4; i++, dst += y_stride) { + for (i = 0; i < 4; i++, dst += stride) { q1u16 = vdupq_n_u16((uint16_t)left[i]); q1s16 = vaddq_s16(vreinterpretq_s16_u16(q1u16), vreinterpretq_s16_u16(q3u16)); @@ -371,7 +371,7 @@ void vp9_tm_predictor_4x4_neon(uint8_t *dst, ptrdiff_t y_stride, } } -void vp9_tm_predictor_8x8_neon(uint8_t *dst, ptrdiff_t y_stride, +void vp9_tm_predictor_8x8_neon(uint8_t *dst, ptrdiff_t stride, const uint8_t *above, const uint8_t *left) { int j; uint16x8_t q0u16, q3u16, q10u16; @@ -391,29 +391,29 @@ void vp9_tm_predictor_8x8_neon(uint8_t *dst, ptrdiff_t y_stride, vreinterpretq_s16_u16(q0u16)); d0u8 = vqmovun_s16(q0s16); vst1_u64((uint64_t *)dst, vreinterpret_u64_u8(d0u8)); - dst += y_stride; + dst += stride; q0u16 = vdupq_lane_u16(d20u16, 1); q0s16 = vaddq_s16(vreinterpretq_s16_u16(q3u16), vreinterpretq_s16_u16(q0u16)); d0u8 = vqmovun_s16(q0s16); vst1_u64((uint64_t *)dst, vreinterpret_u64_u8(d0u8)); - dst += y_stride; + dst += stride; q0u16 = vdupq_lane_u16(d20u16, 2); q0s16 = vaddq_s16(vreinterpretq_s16_u16(q3u16), vreinterpretq_s16_u16(q0u16)); d0u8 = vqmovun_s16(q0s16); vst1_u64((uint64_t *)dst, vreinterpret_u64_u8(d0u8)); - dst += y_stride; + dst += stride; q0u16 = vdupq_lane_u16(d20u16, 3); q0s16 = vaddq_s16(vreinterpretq_s16_u16(q3u16), vreinterpretq_s16_u16(q0u16)); d0u8 = vqmovun_s16(q0s16); vst1_u64((uint64_t *)dst, vreinterpret_u64_u8(d0u8)); - dst += y_stride; + dst += stride; } } -void vp9_tm_predictor_16x16_neon(uint8_t *dst, ptrdiff_t y_stride, +void vp9_tm_predictor_16x16_neon(uint8_t *dst, ptrdiff_t stride, const uint8_t *above, const uint8_t *left) { int j, k; uint16x8_t q0u16, q2u16, q3u16, q8u16, q10u16; @@ -447,10 +447,10 @@ void vp9_tm_predictor_16x16_neon(uint8_t *dst, ptrdiff_t y_stride, d23u8 = vqmovun_s16(q8s16); vst1_u64((uint64_t *)dst, vreinterpret_u64_u8(d2u8)); vst1_u64((uint64_t *)(dst + 8), vreinterpret_u64_u8(d3u8)); - dst += y_stride; + dst += stride; vst1_u64((uint64_t *)dst, vreinterpret_u64_u8(d22u8)); vst1_u64((uint64_t *)(dst + 8), vreinterpret_u64_u8(d23u8)); - dst += y_stride; + dst += stride; q0u16 = vdupq_lane_u16(d20u16, 2); q8u16 = vdupq_lane_u16(d20u16, 3); @@ -468,15 +468,15 @@ void vp9_tm_predictor_16x16_neon(uint8_t *dst, ptrdiff_t y_stride, d23u8 = vqmovun_s16(q8s16); vst1_u64((uint64_t *)dst, vreinterpret_u64_u8(d2u8)); vst1_u64((uint64_t *)(dst + 8), vreinterpret_u64_u8(d3u8)); - dst += y_stride; + dst += stride; vst1_u64((uint64_t *)dst, vreinterpret_u64_u8(d22u8)); vst1_u64((uint64_t *)(dst + 8), vreinterpret_u64_u8(d23u8)); - dst += y_stride; + dst += stride; } } } -void vp9_tm_predictor_32x32_neon(uint8_t *dst, ptrdiff_t y_stride, +void vp9_tm_predictor_32x32_neon(uint8_t *dst, ptrdiff_t stride, const uint8_t *above, const uint8_t *left) { int j, k; uint16x8_t q0u16, q3u16, q8u16, q9u16, q10u16, q11u16; @@ -514,7 +514,7 @@ void vp9_tm_predictor_32x32_neon(uint8_t *dst, ptrdiff_t y_stride, q1u8 = vcombine_u8(d2u8, d3u8); vst1q_u64((uint64_t *)dst, vreinterpretq_u64_u8(q0u8)); vst1q_u64((uint64_t *)(dst + 16), vreinterpretq_u64_u8(q1u8)); - dst += y_stride; + dst += stride; q0u16 = vdupq_lane_u16(d6u16, 1); q12s16 = vaddq_s16(vreinterpretq_s16_u16(q0u16), @@ -533,7 +533,7 @@ void vp9_tm_predictor_32x32_neon(uint8_t *dst, ptrdiff_t y_stride, q1u8 = vcombine_u8(d2u8, d3u8); vst1q_u64((uint64_t *)dst, vreinterpretq_u64_u8(q0u8)); vst1q_u64((uint64_t *)(dst + 16), vreinterpretq_u64_u8(q1u8)); - dst += y_stride; + dst += stride; q0u16 = vdupq_lane_u16(d6u16, 2); q12s16 = vaddq_s16(vreinterpretq_s16_u16(q0u16), @@ -552,7 +552,7 @@ void vp9_tm_predictor_32x32_neon(uint8_t *dst, ptrdiff_t y_stride, q1u8 = vcombine_u8(d2u8, d3u8); vst1q_u64((uint64_t *)dst, vreinterpretq_u64_u8(q0u8)); vst1q_u64((uint64_t *)(dst + 16), vreinterpretq_u64_u8(q1u8)); - dst += y_stride; + dst += stride; q0u16 = vdupq_lane_u16(d6u16, 3); q12s16 = vaddq_s16(vreinterpretq_s16_u16(q0u16), @@ -571,7 +571,7 @@ void vp9_tm_predictor_32x32_neon(uint8_t *dst, ptrdiff_t y_stride, q1u8 = vcombine_u8(d2u8, d3u8); vst1q_u64((uint64_t *)dst, vreinterpretq_u64_u8(q0u8)); vst1q_u64((uint64_t *)(dst + 16), vreinterpretq_u64_u8(q1u8)); - dst += y_stride; + dst += stride; } } } -- cgit v1.2.3 From 7621b48a1c89287529b86266d9a4e7ee3ac28667 Mon Sep 17 00:00:00 2001 From: James Zern Date: Fri, 29 May 2015 19:10:27 -0700 Subject: vp9_reconintra_neon/tm: improve above_left load use vld1?_dup_u8 over vdup?_n_u8, reduces general register use; mildly faster Change-Id: Ie0e4e550849a207b34b378541196b553c9f12011 --- vp9/common/arm/neon/vp9_reconintra_neon.c | 8 ++++---- 1 file changed, 4 insertions(+), 4 deletions(-) (limited to 'vp9') diff --git a/vp9/common/arm/neon/vp9_reconintra_neon.c b/vp9/common/arm/neon/vp9_reconintra_neon.c index 53e640640..499c42ac3 100644 --- a/vp9/common/arm/neon/vp9_reconintra_neon.c +++ b/vp9/common/arm/neon/vp9_reconintra_neon.c @@ -359,7 +359,7 @@ void vp9_tm_predictor_4x4_neon(uint8_t *dst, ptrdiff_t stride, uint8x8_t d0u8 = vdup_n_u8(0); uint32x2_t d2u32 = vdup_n_u32(0); - d0u8 = vdup_n_u8(above[-1]); + d0u8 = vld1_dup_u8(above - 1); d2u32 = vld1_lane_u32((const uint32_t *)above, d2u32, 0); q3u16 = vsubl_u8(vreinterpret_u8_u32(d2u32), d0u8); for (i = 0; i < 4; i++, dst += stride) { @@ -379,7 +379,7 @@ void vp9_tm_predictor_8x8_neon(uint8_t *dst, ptrdiff_t stride, uint16x4_t d20u16; uint8x8_t d0u8, d2u8, d30u8; - d0u8 = vdup_n_u8(above[-1]); + d0u8 = vld1_dup_u8(above - 1); d30u8 = vld1_u8(left); d2u8 = vld1_u8(above); q10u16 = vmovl_u8(d30u8); @@ -422,7 +422,7 @@ void vp9_tm_predictor_16x16_neon(uint8_t *dst, ptrdiff_t stride, uint16x4_t d20u16; uint8x8_t d2u8, d3u8, d18u8, d22u8, d23u8; - q0u8 = vdupq_n_u8(above[-1]); + q0u8 = vld1q_dup_u8(above - 1); q1u8 = vld1q_u8(above); q2u16 = vsubl_u8(vget_low_u8(q1u8), vget_low_u8(q0u8)); q3u16 = vsubl_u8(vget_high_u8(q1u8), vget_high_u8(q0u8)); @@ -485,7 +485,7 @@ void vp9_tm_predictor_32x32_neon(uint8_t *dst, ptrdiff_t stride, uint16x4_t d6u16; uint8x8_t d0u8, d1u8, d2u8, d3u8, d26u8; - q0u8 = vdupq_n_u8(above[-1]); + q0u8 = vld1q_dup_u8(above - 1); q1u8 = vld1q_u8(above); q2u8 = vld1q_u8(above + 16); q8u16 = vsubl_u8(vget_low_u8(q1u8), vget_low_u8(q0u8)); -- cgit v1.2.3