ref: aa8dd897c157952e24ad562babf0c825bd0e29fa
parent: ba05a4c640783dd8a67fb946531940a68d79b906
author: James Yu <james.yu@linaro.org>
date: Sat Feb 8 20:01:36 EST 2014
VP9 common for ARMv8 by using NEON intrinsics 16 Add vp9_reconintra_neon.c - vp9_v_predictor_4x4_neon - vp9_v_predictor_8x8_neon - vp9_v_predictor_16x16_neon - vp9_v_predictor_32x32_neon - vp9_h_predictor_4x4_neon - vp9_h_predictor_8x8_neon - vp9_h_predictor_16x16_neon - vp9_h_predictor_32x32_neon - vp9_tm_predictor_4x4_neon - vp9_tm_predictor_8x8_neon - vp9_tm_predictor_16x16_neon - vp9_tm_predictor_32x32_neon Change-Id: Ib5d54a4766a1b5127169045659974f33aa98376d Signed-off-by: James Yu <james.yu@linaro.org>
--- a/vp9/common/arm/neon/vp9_reconintra_neon.asm
+++ /dev/null
@@ -1,634 +1,0 @@
-;
-; Copyright (c) 2014 The WebM project authors. All Rights Reserved.
-;
-; Use of this source code is governed by a BSD-style license
-; that can be found in the LICENSE file in the root of the source
-; tree. An additional intellectual property rights grant can be found
-; in the file PATENTS. All contributing project authors may
-; be found in the AUTHORS file in the root of the source tree.
-;
-
- EXPORT |vp9_v_predictor_4x4_neon|
- EXPORT |vp9_v_predictor_8x8_neon|
- EXPORT |vp9_v_predictor_16x16_neon|
- EXPORT |vp9_v_predictor_32x32_neon|
- EXPORT |vp9_h_predictor_4x4_neon|
- EXPORT |vp9_h_predictor_8x8_neon|
- EXPORT |vp9_h_predictor_16x16_neon|
- EXPORT |vp9_h_predictor_32x32_neon|
- EXPORT |vp9_tm_predictor_4x4_neon|
- EXPORT |vp9_tm_predictor_8x8_neon|
- EXPORT |vp9_tm_predictor_16x16_neon|
- EXPORT |vp9_tm_predictor_32x32_neon|
- ARM
- REQUIRE8
- PRESERVE8
-
- AREA ||.text||, CODE, READONLY, ALIGN=2
-
-;void vp9_v_predictor_4x4_neon(uint8_t *dst, ptrdiff_t y_stride,
-; const uint8_t *above,
-; const uint8_t *left)
-; r0 uint8_t *dst
-; r1 ptrdiff_t y_stride
-; r2 const uint8_t *above
-; r3 const uint8_t *left
-
-|vp9_v_predictor_4x4_neon| PROC
- vld1.32 {d0[0]}, [r2]- vst1.32 {d0[0]}, [r0], r1- vst1.32 {d0[0]}, [r0], r1- vst1.32 {d0[0]}, [r0], r1- vst1.32 {d0[0]}, [r0], r1- bx lr
- ENDP ; |vp9_v_predictor_4x4_neon|
-
-;void vp9_v_predictor_8x8_neon(uint8_t *dst, ptrdiff_t y_stride,
-; const uint8_t *above,
-; const uint8_t *left)
-; r0 uint8_t *dst
-; r1 ptrdiff_t y_stride
-; r2 const uint8_t *above
-; r3 const uint8_t *left
-
-|vp9_v_predictor_8x8_neon| PROC
- vld1.8 {d0}, [r2]- vst1.8 {d0}, [r0], r1- vst1.8 {d0}, [r0], r1- vst1.8 {d0}, [r0], r1- vst1.8 {d0}, [r0], r1- vst1.8 {d0}, [r0], r1- vst1.8 {d0}, [r0], r1- vst1.8 {d0}, [r0], r1- vst1.8 {d0}, [r0], r1- bx lr
- ENDP ; |vp9_v_predictor_8x8_neon|
-
-;void vp9_v_predictor_16x16_neon(uint8_t *dst, ptrdiff_t y_stride,
-; const uint8_t *above,
-; const uint8_t *left)
-; r0 uint8_t *dst
-; r1 ptrdiff_t y_stride
-; r2 const uint8_t *above
-; r3 const uint8_t *left
-
-|vp9_v_predictor_16x16_neon| PROC
- vld1.8 {q0}, [r2]- vst1.8 {q0}, [r0], r1- vst1.8 {q0}, [r0], r1- vst1.8 {q0}, [r0], r1- vst1.8 {q0}, [r0], r1- vst1.8 {q0}, [r0], r1- vst1.8 {q0}, [r0], r1- vst1.8 {q0}, [r0], r1- vst1.8 {q0}, [r0], r1- vst1.8 {q0}, [r0], r1- vst1.8 {q0}, [r0], r1- vst1.8 {q0}, [r0], r1- vst1.8 {q0}, [r0], r1- vst1.8 {q0}, [r0], r1- vst1.8 {q0}, [r0], r1- vst1.8 {q0}, [r0], r1- vst1.8 {q0}, [r0], r1- bx lr
- ENDP ; |vp9_v_predictor_16x16_neon|
-
-;void vp9_v_predictor_32x32_neon(uint8_t *dst, ptrdiff_t y_stride,
-; const uint8_t *above,
-; const uint8_t *left)
-; r0 uint8_t *dst
-; r1 ptrdiff_t y_stride
-; r2 const uint8_t *above
-; r3 const uint8_t *left
-
-|vp9_v_predictor_32x32_neon| PROC
- vld1.8 {q0, q1}, [r2]- mov r2, #2
-loop_v
- vst1.8 {q0, q1}, [r0], r1- vst1.8 {q0, q1}, [r0], r1- vst1.8 {q0, q1}, [r0], r1- vst1.8 {q0, q1}, [r0], r1- vst1.8 {q0, q1}, [r0], r1- vst1.8 {q0, q1}, [r0], r1- vst1.8 {q0, q1}, [r0], r1- vst1.8 {q0, q1}, [r0], r1- vst1.8 {q0, q1}, [r0], r1- vst1.8 {q0, q1}, [r0], r1- vst1.8 {q0, q1}, [r0], r1- vst1.8 {q0, q1}, [r0], r1- vst1.8 {q0, q1}, [r0], r1- vst1.8 {q0, q1}, [r0], r1- vst1.8 {q0, q1}, [r0], r1- vst1.8 {q0, q1}, [r0], r1- subs r2, r2, #1
- bgt loop_v
- bx lr
- ENDP ; |vp9_v_predictor_32x32_neon|
-
-;void vp9_h_predictor_4x4_neon(uint8_t *dst, ptrdiff_t y_stride,
-; const uint8_t *above,
-; const uint8_t *left)
-; r0 uint8_t *dst
-; r1 ptrdiff_t y_stride
-; r2 const uint8_t *above
-; r3 const uint8_t *left
-
-|vp9_h_predictor_4x4_neon| PROC
- vld1.32 {d1[0]}, [r3]- vdup.8 d0, d1[0]
- vst1.32 {d0[0]}, [r0], r1- vdup.8 d0, d1[1]
- vst1.32 {d0[0]}, [r0], r1- vdup.8 d0, d1[2]
- vst1.32 {d0[0]}, [r0], r1- vdup.8 d0, d1[3]
- vst1.32 {d0[0]}, [r0], r1- bx lr
- ENDP ; |vp9_h_predictor_4x4_neon|
-
-;void vp9_h_predictor_8x8_neon(uint8_t *dst, ptrdiff_t y_stride,
-; const uint8_t *above,
-; const uint8_t *left)
-; r0 uint8_t *dst
-; r1 ptrdiff_t y_stride
-; r2 const uint8_t *above
-; r3 const uint8_t *left
-
-|vp9_h_predictor_8x8_neon| PROC
- vld1.64 {d1}, [r3]- vdup.8 d0, d1[0]
- vst1.64 {d0}, [r0], r1- vdup.8 d0, d1[1]
- vst1.64 {d0}, [r0], r1- vdup.8 d0, d1[2]
- vst1.64 {d0}, [r0], r1- vdup.8 d0, d1[3]
- vst1.64 {d0}, [r0], r1- vdup.8 d0, d1[4]
- vst1.64 {d0}, [r0], r1- vdup.8 d0, d1[5]
- vst1.64 {d0}, [r0], r1- vdup.8 d0, d1[6]
- vst1.64 {d0}, [r0], r1- vdup.8 d0, d1[7]
- vst1.64 {d0}, [r0], r1- bx lr
- ENDP ; |vp9_h_predictor_8x8_neon|
-
-;void vp9_h_predictor_16x16_neon(uint8_t *dst, ptrdiff_t y_stride,
-; const uint8_t *above,
-; const uint8_t *left)
-; r0 uint8_t *dst
-; r1 ptrdiff_t y_stride
-; r2 const uint8_t *above
-; r3 const uint8_t *left
-
-|vp9_h_predictor_16x16_neon| PROC
- vld1.8 {q1}, [r3]- vdup.8 q0, d2[0]
- vst1.8 {q0}, [r0], r1- vdup.8 q0, d2[1]
- vst1.8 {q0}, [r0], r1- vdup.8 q0, d2[2]
- vst1.8 {q0}, [r0], r1- vdup.8 q0, d2[3]
- vst1.8 {q0}, [r0], r1- vdup.8 q0, d2[4]
- vst1.8 {q0}, [r0], r1- vdup.8 q0, d2[5]
- vst1.8 {q0}, [r0], r1- vdup.8 q0, d2[6]
- vst1.8 {q0}, [r0], r1- vdup.8 q0, d2[7]
- vst1.8 {q0}, [r0], r1- vdup.8 q0, d3[0]
- vst1.8 {q0}, [r0], r1- vdup.8 q0, d3[1]
- vst1.8 {q0}, [r0], r1- vdup.8 q0, d3[2]
- vst1.8 {q0}, [r0], r1- vdup.8 q0, d3[3]
- vst1.8 {q0}, [r0], r1- vdup.8 q0, d3[4]
- vst1.8 {q0}, [r0], r1- vdup.8 q0, d3[5]
- vst1.8 {q0}, [r0], r1- vdup.8 q0, d3[6]
- vst1.8 {q0}, [r0], r1- vdup.8 q0, d3[7]
- vst1.8 {q0}, [r0], r1- bx lr
- ENDP ; |vp9_h_predictor_16x16_neon|
-
-;void vp9_h_predictor_32x32_neon(uint8_t *dst, ptrdiff_t y_stride,
-; const uint8_t *above,
-; const uint8_t *left)
-; r0 uint8_t *dst
-; r1 ptrdiff_t y_stride
-; r2 const uint8_t *above
-; r3 const uint8_t *left
-
-|vp9_h_predictor_32x32_neon| PROC
- sub r1, r1, #16
- mov r2, #2
-loop_h
- vld1.8 {q1}, [r3]!- vdup.8 q0, d2[0]
- vst1.8 {q0}, [r0]!- vst1.8 {q0}, [r0], r1- vdup.8 q0, d2[1]
- vst1.8 {q0}, [r0]!- vst1.8 {q0}, [r0], r1- vdup.8 q0, d2[2]
- vst1.8 {q0}, [r0]!- vst1.8 {q0}, [r0], r1- vdup.8 q0, d2[3]
- vst1.8 {q0}, [r0]!- vst1.8 {q0}, [r0], r1- vdup.8 q0, d2[4]
- vst1.8 {q0}, [r0]!- vst1.8 {q0}, [r0], r1- vdup.8 q0, d2[5]
- vst1.8 {q0}, [r0]!- vst1.8 {q0}, [r0], r1- vdup.8 q0, d2[6]
- vst1.8 {q0}, [r0]!- vst1.8 {q0}, [r0], r1- vdup.8 q0, d2[7]
- vst1.8 {q0}, [r0]!- vst1.8 {q0}, [r0], r1- vdup.8 q0, d3[0]
- vst1.8 {q0}, [r0]!- vst1.8 {q0}, [r0], r1- vdup.8 q0, d3[1]
- vst1.8 {q0}, [r0]!- vst1.8 {q0}, [r0], r1- vdup.8 q0, d3[2]
- vst1.8 {q0}, [r0]!- vst1.8 {q0}, [r0], r1- vdup.8 q0, d3[3]
- vst1.8 {q0}, [r0]!- vst1.8 {q0}, [r0], r1- vdup.8 q0, d3[4]
- vst1.8 {q0}, [r0]!- vst1.8 {q0}, [r0], r1- vdup.8 q0, d3[5]
- vst1.8 {q0}, [r0]!- vst1.8 {q0}, [r0], r1- vdup.8 q0, d3[6]
- vst1.8 {q0}, [r0]!- vst1.8 {q0}, [r0], r1- vdup.8 q0, d3[7]
- vst1.8 {q0}, [r0]!- vst1.8 {q0}, [r0], r1- subs r2, r2, #1
- bgt loop_h
- bx lr
- ENDP ; |vp9_h_predictor_32x32_neon|
-
-;void vp9_tm_predictor_4x4_neon (uint8_t *dst, ptrdiff_t y_stride,
-; const uint8_t *above,
-; const uint8_t *left)
-; r0 uint8_t *dst
-; r1 ptrdiff_t y_stride
-; r2 const uint8_t *above
-; r3 const uint8_t *left
-
-|vp9_tm_predictor_4x4_neon| PROC
- ; Load ytop_left = above[-1];
- sub r12, r2, #1
- ldrb r12, [r12]
- vdup.u8 d0, r12
-
- ; Load above 4 pixels
- vld1.32 {d2[0]}, [r2]-
- ; Compute above - ytop_left
- vsubl.u8 q3, d2, d0
-
- ; Load left row by row and compute left + (above - ytop_left)
- ; 1st row and 2nd row
- ldrb r12, [r3], #1
- ldrb r2, [r3], #1
- vdup.u16 q1, r12
- vdup.u16 q2, r2
- vadd.s16 q1, q1, q3
- vadd.s16 q2, q2, q3
- vqmovun.s16 d0, q1
- vqmovun.s16 d1, q2
- vst1.32 {d0[0]}, [r0], r1- vst1.32 {d1[0]}, [r0], r1-
- ; 3rd row and 4th row
- ldrb r12, [r3], #1
- ldrb r2, [r3], #1
- vdup.u16 q1, r12
- vdup.u16 q2, r2
- vadd.s16 q1, q1, q3
- vadd.s16 q2, q2, q3
- vqmovun.s16 d0, q1
- vqmovun.s16 d1, q2
- vst1.32 {d0[0]}, [r0], r1- vst1.32 {d1[0]}, [r0], r1- bx lr
- ENDP ; |vp9_tm_predictor_4x4_neon|
-
-;void vp9_tm_predictor_8x8_neon (uint8_t *dst, ptrdiff_t y_stride,
-; const uint8_t *above,
-; const uint8_t *left)
-; r0 uint8_t *dst
-; r1 ptrdiff_t y_stride
-; r2 const uint8_t *above
-; r3 const uint8_t *left
-
-|vp9_tm_predictor_8x8_neon| PROC
- ; Load ytop_left = above[-1];
- sub r12, r2, #1
- ldrb r12, [r12]
- vdup.u8 d0, r12
-
- ; preload 8 left
- vld1.8 {d30}, [r3]-
- ; Load above 8 pixels
- vld1.64 {d2}, [r2]-
- vmovl.u8 q10, d30
-
- ; Compute above - ytop_left
- vsubl.u8 q3, d2, d0
-
- ; Load left row by row and compute left + (above - ytop_left)
- ; 1st row and 2nd row
- vdup.16 q0, d20[0]
- vdup.16 q1, d20[1]
- vadd.s16 q0, q3, q0
- vadd.s16 q1, q3, q1
-
- ; 3rd row and 4th row
- vdup.16 q8, d20[2]
- vdup.16 q9, d20[3]
- vadd.s16 q8, q3, q8
- vadd.s16 q9, q3, q9
-
- vqmovun.s16 d0, q0
- vqmovun.s16 d1, q1
- vqmovun.s16 d2, q8
- vqmovun.s16 d3, q9
-
- vst1.64 {d0}, [r0], r1- vst1.64 {d1}, [r0], r1- vst1.64 {d2}, [r0], r1- vst1.64 {d3}, [r0], r1-
- ; 5th row and 6th row
- vdup.16 q0, d21[0]
- vdup.16 q1, d21[1]
- vadd.s16 q0, q3, q0
- vadd.s16 q1, q3, q1
-
- ; 7th row and 8th row
- vdup.16 q8, d21[2]
- vdup.16 q9, d21[3]
- vadd.s16 q8, q3, q8
- vadd.s16 q9, q3, q9
-
- vqmovun.s16 d0, q0
- vqmovun.s16 d1, q1
- vqmovun.s16 d2, q8
- vqmovun.s16 d3, q9
-
- vst1.64 {d0}, [r0], r1- vst1.64 {d1}, [r0], r1- vst1.64 {d2}, [r0], r1- vst1.64 {d3}, [r0], r1-
- bx lr
- ENDP ; |vp9_tm_predictor_8x8_neon|
-
-;void vp9_tm_predictor_16x16_neon (uint8_t *dst, ptrdiff_t y_stride,
-; const uint8_t *above,
-; const uint8_t *left)
-; r0 uint8_t *dst
-; r1 ptrdiff_t y_stride
-; r2 const uint8_t *above
-; r3 const uint8_t *left
-
-|vp9_tm_predictor_16x16_neon| PROC
- ; Load ytop_left = above[-1];
- sub r12, r2, #1
- ldrb r12, [r12]
- vdup.u8 q0, r12
-
- ; Load above 8 pixels
- vld1.8 {q1}, [r2]-
- ; preload 8 left into r12
- vld1.8 {d18}, [r3]!-
- ; Compute above - ytop_left
- vsubl.u8 q2, d2, d0
- vsubl.u8 q3, d3, d1
-
- vmovl.u8 q10, d18
-
- ; Load left row by row and compute left + (above - ytop_left)
- ; Process 8 rows in each single loop and loop 2 times to process 16 rows.
- mov r2, #2
-
-loop_16x16_neon
- ; Process two rows.
- vdup.16 q0, d20[0]
- vdup.16 q8, d20[1]
- vadd.s16 q1, q0, q2
- vadd.s16 q0, q0, q3
- vadd.s16 q11, q8, q2
- vadd.s16 q8, q8, q3
- vqmovun.s16 d2, q1
- vqmovun.s16 d3, q0
- vqmovun.s16 d22, q11
- vqmovun.s16 d23, q8
- vdup.16 q0, d20[2] ; proload next 2 rows data
- vdup.16 q8, d20[3]
- vst1.64 {d2,d3}, [r0], r1- vst1.64 {d22,d23}, [r0], r1-
- ; Process two rows.
- vadd.s16 q1, q0, q2
- vadd.s16 q0, q0, q3
- vadd.s16 q11, q8, q2
- vadd.s16 q8, q8, q3
- vqmovun.s16 d2, q1
- vqmovun.s16 d3, q0
- vqmovun.s16 d22, q11
- vqmovun.s16 d23, q8
- vdup.16 q0, d21[0] ; proload next 2 rows data
- vdup.16 q8, d21[1]
- vst1.64 {d2,d3}, [r0], r1- vst1.64 {d22,d23}, [r0], r1-
- vadd.s16 q1, q0, q2
- vadd.s16 q0, q0, q3
- vadd.s16 q11, q8, q2
- vadd.s16 q8, q8, q3
- vqmovun.s16 d2, q1
- vqmovun.s16 d3, q0
- vqmovun.s16 d22, q11
- vqmovun.s16 d23, q8
- vdup.16 q0, d21[2] ; proload next 2 rows data
- vdup.16 q8, d21[3]
- vst1.64 {d2,d3}, [r0], r1- vst1.64 {d22,d23}, [r0], r1-
-
- vadd.s16 q1, q0, q2
- vadd.s16 q0, q0, q3
- vadd.s16 q11, q8, q2
- vadd.s16 q8, q8, q3
- vqmovun.s16 d2, q1
- vqmovun.s16 d3, q0
- vqmovun.s16 d22, q11
- vqmovun.s16 d23, q8
- vld1.8 {d18}, [r3]! ; preload 8 left into r12- vmovl.u8 q10, d18
- vst1.64 {d2,d3}, [r0], r1- vst1.64 {d22,d23}, [r0], r1-
- subs r2, r2, #1
- bgt loop_16x16_neon
-
- bx lr
- ENDP ; |vp9_tm_predictor_16x16_neon|
-
-;void vp9_tm_predictor_32x32_neon (uint8_t *dst, ptrdiff_t y_stride,
-; const uint8_t *above,
-; const uint8_t *left)
-; r0 uint8_t *dst
-; r1 ptrdiff_t y_stride
-; r2 const uint8_t *above
-; r3 const uint8_t *left
-
-|vp9_tm_predictor_32x32_neon| PROC
- ; Load ytop_left = above[-1];
- sub r12, r2, #1
- ldrb r12, [r12]
- vdup.u8 q0, r12
-
- ; Load above 32 pixels
- vld1.8 {q1}, [r2]!- vld1.8 {q2}, [r2]-
- ; preload 8 left pixels
- vld1.8 {d26}, [r3]!-
- ; Compute above - ytop_left
- vsubl.u8 q8, d2, d0
- vsubl.u8 q9, d3, d1
- vsubl.u8 q10, d4, d0
- vsubl.u8 q11, d5, d1
-
- vmovl.u8 q3, d26
-
- ; Load left row by row and compute left + (above - ytop_left)
- ; Process 8 rows in each single loop and loop 4 times to process 32 rows.
- mov r2, #4
-
-loop_32x32_neon
- ; Process two rows.
- vdup.16 q0, d6[0]
- vdup.16 q2, d6[1]
- vadd.s16 q12, q0, q8
- vadd.s16 q13, q0, q9
- vadd.s16 q14, q0, q10
- vadd.s16 q15, q0, q11
- vqmovun.s16 d0, q12
- vqmovun.s16 d1, q13
- vadd.s16 q12, q2, q8
- vadd.s16 q13, q2, q9
- vqmovun.s16 d2, q14
- vqmovun.s16 d3, q15
- vadd.s16 q14, q2, q10
- vadd.s16 q15, q2, q11
- vst1.64 {d0-d3}, [r0], r1- vqmovun.s16 d24, q12
- vqmovun.s16 d25, q13
- vqmovun.s16 d26, q14
- vqmovun.s16 d27, q15
- vdup.16 q1, d6[2]
- vdup.16 q2, d6[3]
- vst1.64 {d24-d27}, [r0], r1-
- ; Process two rows.
- vadd.s16 q12, q1, q8
- vadd.s16 q13, q1, q9
- vadd.s16 q14, q1, q10
- vadd.s16 q15, q1, q11
- vqmovun.s16 d0, q12
- vqmovun.s16 d1, q13
- vadd.s16 q12, q2, q8
- vadd.s16 q13, q2, q9
- vqmovun.s16 d2, q14
- vqmovun.s16 d3, q15
- vadd.s16 q14, q2, q10
- vadd.s16 q15, q2, q11
- vst1.64 {d0-d3}, [r0], r1- vqmovun.s16 d24, q12
- vqmovun.s16 d25, q13
- vqmovun.s16 d26, q14
- vqmovun.s16 d27, q15
- vdup.16 q0, d7[0]
- vdup.16 q2, d7[1]
- vst1.64 {d24-d27}, [r0], r1-
- ; Process two rows.
- vadd.s16 q12, q0, q8
- vadd.s16 q13, q0, q9
- vadd.s16 q14, q0, q10
- vadd.s16 q15, q0, q11
- vqmovun.s16 d0, q12
- vqmovun.s16 d1, q13
- vadd.s16 q12, q2, q8
- vadd.s16 q13, q2, q9
- vqmovun.s16 d2, q14
- vqmovun.s16 d3, q15
- vadd.s16 q14, q2, q10
- vadd.s16 q15, q2, q11
- vst1.64 {d0-d3}, [r0], r1- vqmovun.s16 d24, q12
- vqmovun.s16 d25, q13
- vqmovun.s16 d26, q14
- vqmovun.s16 d27, q15
- vdup.16 q0, d7[2]
- vdup.16 q2, d7[3]
- vst1.64 {d24-d27}, [r0], r1-
- ; Process two rows.
- vadd.s16 q12, q0, q8
- vadd.s16 q13, q0, q9
- vadd.s16 q14, q0, q10
- vadd.s16 q15, q0, q11
- vqmovun.s16 d0, q12
- vqmovun.s16 d1, q13
- vadd.s16 q12, q2, q8
- vadd.s16 q13, q2, q9
- vqmovun.s16 d2, q14
- vqmovun.s16 d3, q15
- vadd.s16 q14, q2, q10
- vadd.s16 q15, q2, q11
- vst1.64 {d0-d3}, [r0], r1- vqmovun.s16 d24, q12
- vqmovun.s16 d25, q13
- vld1.8 {d0}, [r3]! ; preload 8 left pixels- vqmovun.s16 d26, q14
- vqmovun.s16 d27, q15
- vmovl.u8 q3, d0
- vst1.64 {d24-d27}, [r0], r1-
- subs r2, r2, #1
- bgt loop_32x32_neon
-
- bx lr
- ENDP ; |vp9_tm_predictor_32x32_neon|
-
- END
--- /dev/null
+++ b/vp9/common/arm/neon/vp9_reconintra_neon.c
@@ -1,0 +1,473 @@
+/*
+ * Copyright (c) 2014 The WebM project authors. All Rights Reserved.
+ *
+ * Use of this source code is governed by a BSD-style license
+ * that can be found in the LICENSE file in the root of the source
+ * tree. An additional intellectual property rights grant can be found
+ * in the file PATENTS. All contributing project authors may
+ * be found in the AUTHORS file in the root of the source tree.
+ */
+
+#include <stddef.h>
+#include <arm_neon.h>
+
+void vp9_v_predictor_4x4_neon(
+ uint8_t *dst,
+ ptrdiff_t y_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)
+ vst1_lane_u32((uint32_t *)dst, d0u32, 0);
+ return;
+}
+
+void vp9_v_predictor_8x8_neon(
+ uint8_t *dst,
+ ptrdiff_t y_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)
+ vst1_u8(dst, d0u8);
+ return;
+}
+
+void vp9_v_predictor_16x16_neon(
+ uint8_t *dst,
+ ptrdiff_t y_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)
+ vst1q_u8(dst, q0u8);
+ return;
+}
+
+void vp9_v_predictor_32x32_neon(
+ uint8_t *dst,
+ ptrdiff_t y_stride,
+ const uint8_t *above,
+ const uint8_t *left) {+ int i;
+ uint8x16_t q0u8 = vdupq_n_u8(0);
+ uint8x16_t q1u8 = vdupq_n_u8(0);
+ (void)left;
+
+ q0u8 = vld1q_u8(above);
+ q1u8 = vld1q_u8(above + 16);
+ for (i = 0; i < 32; i++, dst += y_stride) {+ vst1q_u8(dst, q0u8);
+ vst1q_u8(dst + 16, q1u8);
+ }
+ return;
+}
+
+void vp9_h_predictor_4x4_neon(
+ uint8_t *dst,
+ ptrdiff_t y_stride,
+ const uint8_t *above,
+ const uint8_t *left) {+ uint8x8_t d0u8 = vdup_n_u8(0);
+ uint32x2_t d1u32 = vdup_n_u32(0);
+ (void)above;
+
+ d1u32 = vld1_lane_u32((const uint32_t *)left, d1u32, 0);
+
+ d0u8 = vdup_lane_u8(vreinterpret_u8_u32(d1u32), 0);
+ vst1_lane_u32((uint32_t *)dst, vreinterpret_u32_u8(d0u8), 0);
+ dst += y_stride;
+ d0u8 = vdup_lane_u8(vreinterpret_u8_u32(d1u32), 1);
+ vst1_lane_u32((uint32_t *)dst, vreinterpret_u32_u8(d0u8), 0);
+ dst += y_stride;
+ d0u8 = vdup_lane_u8(vreinterpret_u8_u32(d1u32), 2);
+ vst1_lane_u32((uint32_t *)dst, vreinterpret_u32_u8(d0u8), 0);
+ dst += y_stride;
+ d0u8 = vdup_lane_u8(vreinterpret_u8_u32(d1u32), 3);
+ vst1_lane_u32((uint32_t *)dst, vreinterpret_u32_u8(d0u8), 0);
+ return;
+}
+
+void vp9_h_predictor_8x8_neon(
+ uint8_t *dst,
+ ptrdiff_t y_stride,
+ const uint8_t *above,
+ const uint8_t *left) {+ uint8x8_t d0u8 = vdup_n_u8(0);
+ uint64x1_t d1u64 = vdup_n_u64(0);
+ (void)above;
+
+ d1u64 = vld1_u64((const uint64_t *)left);
+
+ d0u8 = vdup_lane_u8(vreinterpret_u8_u64(d1u64), 0);
+ vst1_u8(dst, d0u8);
+ dst += y_stride;
+ d0u8 = vdup_lane_u8(vreinterpret_u8_u64(d1u64), 1);
+ vst1_u8(dst, d0u8);
+ dst += y_stride;
+ d0u8 = vdup_lane_u8(vreinterpret_u8_u64(d1u64), 2);
+ vst1_u8(dst, d0u8);
+ dst += y_stride;
+ d0u8 = vdup_lane_u8(vreinterpret_u8_u64(d1u64), 3);
+ vst1_u8(dst, d0u8);
+ dst += y_stride;
+ d0u8 = vdup_lane_u8(vreinterpret_u8_u64(d1u64), 4);
+ vst1_u8(dst, d0u8);
+ dst += y_stride;
+ d0u8 = vdup_lane_u8(vreinterpret_u8_u64(d1u64), 5);
+ vst1_u8(dst, d0u8);
+ dst += y_stride;
+ d0u8 = vdup_lane_u8(vreinterpret_u8_u64(d1u64), 6);
+ vst1_u8(dst, d0u8);
+ dst += y_stride;
+ d0u8 = vdup_lane_u8(vreinterpret_u8_u64(d1u64), 7);
+ vst1_u8(dst, d0u8);
+ return;
+}
+
+void vp9_h_predictor_16x16_neon(
+ uint8_t *dst,
+ ptrdiff_t y_stride,
+ const uint8_t *above,
+ const uint8_t *left) {+ int j;
+ uint8x8_t d2u8 = vdup_n_u8(0);
+ uint8x16_t q0u8 = vdupq_n_u8(0);
+ uint8x16_t q1u8 = vdupq_n_u8(0);
+ (void)above;
+
+ q1u8 = vld1q_u8(left);
+ d2u8 = vget_low_u8(q1u8);
+ for (j = 0; j < 2; j++, d2u8 = vget_high_u8(q1u8)) {+ q0u8 = vdupq_lane_u8(d2u8, 0);
+ vst1q_u8(dst, q0u8);
+ dst += y_stride;
+ q0u8 = vdupq_lane_u8(d2u8, 1);
+ vst1q_u8(dst, q0u8);
+ dst += y_stride;
+ q0u8 = vdupq_lane_u8(d2u8, 2);
+ vst1q_u8(dst, q0u8);
+ dst += y_stride;
+ q0u8 = vdupq_lane_u8(d2u8, 3);
+ vst1q_u8(dst, q0u8);
+ dst += y_stride;
+ q0u8 = vdupq_lane_u8(d2u8, 4);
+ vst1q_u8(dst, q0u8);
+ dst += y_stride;
+ q0u8 = vdupq_lane_u8(d2u8, 5);
+ vst1q_u8(dst, q0u8);
+ dst += y_stride;
+ q0u8 = vdupq_lane_u8(d2u8, 6);
+ vst1q_u8(dst, q0u8);
+ dst += y_stride;
+ q0u8 = vdupq_lane_u8(d2u8, 7);
+ vst1q_u8(dst, q0u8);
+ dst += y_stride;
+ }
+ return;
+}
+
+void vp9_h_predictor_32x32_neon(
+ uint8_t *dst,
+ ptrdiff_t y_stride,
+ const uint8_t *above,
+ const uint8_t *left) {+ int j, k;
+ uint8x8_t d2u8 = vdup_n_u8(0);
+ uint8x16_t q0u8 = vdupq_n_u8(0);
+ uint8x16_t q1u8 = vdupq_n_u8(0);
+ (void)above;
+
+ for (k = 0; k < 2; k++, left += 16) {+ q1u8 = vld1q_u8(left);
+ d2u8 = vget_low_u8(q1u8);
+ for (j = 0; j < 2; j++, d2u8 = vget_high_u8(q1u8)) {+ q0u8 = vdupq_lane_u8(d2u8, 0);
+ vst1q_u8(dst, q0u8);
+ vst1q_u8(dst + 16, q0u8);
+ dst += y_stride;
+ q0u8 = vdupq_lane_u8(d2u8, 1);
+ vst1q_u8(dst, q0u8);
+ vst1q_u8(dst + 16, q0u8);
+ dst += y_stride;
+ q0u8 = vdupq_lane_u8(d2u8, 2);
+ vst1q_u8(dst, q0u8);
+ vst1q_u8(dst + 16, q0u8);
+ dst += y_stride;
+ q0u8 = vdupq_lane_u8(d2u8, 3);
+ vst1q_u8(dst, q0u8);
+ vst1q_u8(dst + 16, q0u8);
+ dst += y_stride;
+ q0u8 = vdupq_lane_u8(d2u8, 4);
+ vst1q_u8(dst, q0u8);
+ vst1q_u8(dst + 16, q0u8);
+ dst += y_stride;
+ q0u8 = vdupq_lane_u8(d2u8, 5);
+ vst1q_u8(dst, q0u8);
+ vst1q_u8(dst + 16, q0u8);
+ dst += y_stride;
+ q0u8 = vdupq_lane_u8(d2u8, 6);
+ vst1q_u8(dst, q0u8);
+ vst1q_u8(dst + 16, q0u8);
+ dst += y_stride;
+ q0u8 = vdupq_lane_u8(d2u8, 7);
+ vst1q_u8(dst, q0u8);
+ vst1q_u8(dst + 16, q0u8);
+ dst += y_stride;
+ }
+ }
+ return;
+}
+
+void vp9_tm_predictor_4x4_neon(
+ uint8_t *dst,
+ ptrdiff_t y_stride,
+ const uint8_t *above,
+ const uint8_t *left) {+ int i;
+ uint16x8_t q1u16, q3u16;
+ int16x8_t q1s16;
+ uint8x8_t d0u8 = vdup_n_u8(0);
+ uint32x2_t d2u32 = vdup_n_u32(0);
+
+ 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) {+ q1u16 = vdupq_n_u16((uint16_t)left[i]);
+ q1s16 = vaddq_s16(vreinterpretq_s16_u16(q1u16),
+ vreinterpretq_s16_u16(q3u16));
+ d0u8 = vqmovun_s16(q1s16);
+ vst1_lane_u32((uint32_t *)dst, vreinterpret_u32_u8(d0u8), 0);
+ }
+ return;
+}
+
+void vp9_tm_predictor_8x8_neon(
+ uint8_t *dst,
+ ptrdiff_t y_stride,
+ const uint8_t *above,
+ const uint8_t *left) {+ int j;
+ uint16x8_t q0u16, q3u16, q10u16;
+ int16x8_t q0s16;
+ uint16x4_t d20u16;
+ uint8x8_t d0u8, d2u8, d30u8;
+
+ d0u8 = vdup_n_u8(above[-1]);
+ d30u8 = vld1_u8(left);
+ d2u8 = vld1_u8(above);
+ q10u16 = vmovl_u8(d30u8);
+ q3u16 = vsubl_u8(d2u8, d0u8);
+ d20u16 = vget_low_u16(q10u16);
+ for (j = 0; j < 2; j++, d20u16 = vget_high_u16(q10u16)) {+ q0u16 = vdupq_lane_u16(d20u16, 0);
+ 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;
+ 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;
+ 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;
+ 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;
+ }
+ return;
+}
+
+void vp9_tm_predictor_16x16_neon(
+ uint8_t *dst,
+ ptrdiff_t y_stride,
+ const uint8_t *above,
+ const uint8_t *left) {+ int j, k;
+ uint16x8_t q0u16, q2u16, q3u16, q8u16, q10u16;
+ uint8x16_t q0u8, q1u8;
+ int16x8_t q0s16, q1s16, q8s16, q11s16;
+ uint16x4_t d20u16;
+ uint8x8_t d2u8, d3u8, d18u8, d22u8, d23u8;
+
+ q0u8 = vdupq_n_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));
+ for (k = 0; k < 2; k++, left += 8) {+ d18u8 = vld1_u8(left);
+ q10u16 = vmovl_u8(d18u8);
+ d20u16 = vget_low_u16(q10u16);
+ for (j = 0; j < 2; j++, d20u16 = vget_high_u16(q10u16)) {+ q0u16 = vdupq_lane_u16(d20u16, 0);
+ q8u16 = vdupq_lane_u16(d20u16, 1);
+ q1s16 = vaddq_s16(vreinterpretq_s16_u16(q0u16),
+ vreinterpretq_s16_u16(q2u16));
+ q0s16 = vaddq_s16(vreinterpretq_s16_u16(q0u16),
+ vreinterpretq_s16_u16(q3u16));
+ q11s16 = vaddq_s16(vreinterpretq_s16_u16(q8u16),
+ vreinterpretq_s16_u16(q2u16));
+ q8s16 = vaddq_s16(vreinterpretq_s16_u16(q8u16),
+ vreinterpretq_s16_u16(q3u16));
+ d2u8 = vqmovun_s16(q1s16);
+ d3u8 = vqmovun_s16(q0s16);
+ d22u8 = vqmovun_s16(q11s16);
+ 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;
+ vst1_u64((uint64_t *)dst, vreinterpret_u64_u8(d22u8));
+ vst1_u64((uint64_t *)(dst + 8), vreinterpret_u64_u8(d23u8));
+ dst += y_stride;
+
+ q0u16 = vdupq_lane_u16(d20u16, 2);
+ q8u16 = vdupq_lane_u16(d20u16, 3);
+ q1s16 = vaddq_s16(vreinterpretq_s16_u16(q0u16),
+ vreinterpretq_s16_u16(q2u16));
+ q0s16 = vaddq_s16(vreinterpretq_s16_u16(q0u16),
+ vreinterpretq_s16_u16(q3u16));
+ q11s16 = vaddq_s16(vreinterpretq_s16_u16(q8u16),
+ vreinterpretq_s16_u16(q2u16));
+ q8s16 = vaddq_s16(vreinterpretq_s16_u16(q8u16),
+ vreinterpretq_s16_u16(q3u16));
+ d2u8 = vqmovun_s16(q1s16);
+ d3u8 = vqmovun_s16(q0s16);
+ d22u8 = vqmovun_s16(q11s16);
+ 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;
+ vst1_u64((uint64_t *)dst, vreinterpret_u64_u8(d22u8));
+ vst1_u64((uint64_t *)(dst + 8), vreinterpret_u64_u8(d23u8));
+ dst += y_stride;
+ }
+ }
+ return;
+}
+
+void vp9_tm_predictor_32x32_neon(
+ uint8_t *dst,
+ ptrdiff_t y_stride,
+ const uint8_t *above,
+ const uint8_t *left) {+ int j, k;
+ uint16x8_t q0u16, q3u16, q8u16, q9u16, q10u16, q11u16;
+ uint8x16_t q0u8, q1u8, q2u8;
+ int16x8_t q12s16, q13s16, q14s16, q15s16;
+ uint16x4_t d6u16;
+ uint8x8_t d0u8, d1u8, d2u8, d3u8, d26u8;
+
+ q0u8 = vdupq_n_u8(above[-1]);
+ q1u8 = vld1q_u8(above);
+ q2u8 = vld1q_u8(above + 16);
+ q8u16 = vsubl_u8(vget_low_u8(q1u8), vget_low_u8(q0u8));
+ q9u16 = vsubl_u8(vget_high_u8(q1u8), vget_high_u8(q0u8));
+ q10u16 = vsubl_u8(vget_low_u8(q2u8), vget_low_u8(q0u8));
+ q11u16 = vsubl_u8(vget_high_u8(q2u8), vget_high_u8(q0u8));
+ for (k = 0; k < 4; k++, left += 8) {+ d26u8 = vld1_u8(left);
+ q3u16 = vmovl_u8(d26u8);
+ d6u16 = vget_low_u16(q3u16);
+ for (j = 0; j < 2; j++, d6u16 = vget_high_u16(q3u16)) {+ q0u16 = vdupq_lane_u16(d6u16, 0);
+ q12s16 = vaddq_s16(vreinterpretq_s16_u16(q0u16),
+ vreinterpretq_s16_u16(q8u16));
+ q13s16 = vaddq_s16(vreinterpretq_s16_u16(q0u16),
+ vreinterpretq_s16_u16(q9u16));
+ q14s16 = vaddq_s16(vreinterpretq_s16_u16(q0u16),
+ vreinterpretq_s16_u16(q10u16));
+ q15s16 = vaddq_s16(vreinterpretq_s16_u16(q0u16),
+ vreinterpretq_s16_u16(q11u16));
+ d0u8 = vqmovun_s16(q12s16);
+ d1u8 = vqmovun_s16(q13s16);
+ d2u8 = vqmovun_s16(q14s16);
+ d3u8 = vqmovun_s16(q15s16);
+ q0u8 = vcombine_u8(d0u8, d1u8);
+ 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;
+
+ q0u16 = vdupq_lane_u16(d6u16, 1);
+ q12s16 = vaddq_s16(vreinterpretq_s16_u16(q0u16),
+ vreinterpretq_s16_u16(q8u16));
+ q13s16 = vaddq_s16(vreinterpretq_s16_u16(q0u16),
+ vreinterpretq_s16_u16(q9u16));
+ q14s16 = vaddq_s16(vreinterpretq_s16_u16(q0u16),
+ vreinterpretq_s16_u16(q10u16));
+ q15s16 = vaddq_s16(vreinterpretq_s16_u16(q0u16),
+ vreinterpretq_s16_u16(q11u16));
+ d0u8 = vqmovun_s16(q12s16);
+ d1u8 = vqmovun_s16(q13s16);
+ d2u8 = vqmovun_s16(q14s16);
+ d3u8 = vqmovun_s16(q15s16);
+ q0u8 = vcombine_u8(d0u8, d1u8);
+ 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;
+
+ q0u16 = vdupq_lane_u16(d6u16, 2);
+ q12s16 = vaddq_s16(vreinterpretq_s16_u16(q0u16),
+ vreinterpretq_s16_u16(q8u16));
+ q13s16 = vaddq_s16(vreinterpretq_s16_u16(q0u16),
+ vreinterpretq_s16_u16(q9u16));
+ q14s16 = vaddq_s16(vreinterpretq_s16_u16(q0u16),
+ vreinterpretq_s16_u16(q10u16));
+ q15s16 = vaddq_s16(vreinterpretq_s16_u16(q0u16),
+ vreinterpretq_s16_u16(q11u16));
+ d0u8 = vqmovun_s16(q12s16);
+ d1u8 = vqmovun_s16(q13s16);
+ d2u8 = vqmovun_s16(q14s16);
+ d3u8 = vqmovun_s16(q15s16);
+ q0u8 = vcombine_u8(d0u8, d1u8);
+ 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;
+
+ q0u16 = vdupq_lane_u16(d6u16, 3);
+ q12s16 = vaddq_s16(vreinterpretq_s16_u16(q0u16),
+ vreinterpretq_s16_u16(q8u16));
+ q13s16 = vaddq_s16(vreinterpretq_s16_u16(q0u16),
+ vreinterpretq_s16_u16(q9u16));
+ q14s16 = vaddq_s16(vreinterpretq_s16_u16(q0u16),
+ vreinterpretq_s16_u16(q10u16));
+ q15s16 = vaddq_s16(vreinterpretq_s16_u16(q0u16),
+ vreinterpretq_s16_u16(q11u16));
+ d0u8 = vqmovun_s16(q12s16);
+ d1u8 = vqmovun_s16(q13s16);
+ d2u8 = vqmovun_s16(q14s16);
+ d3u8 = vqmovun_s16(q15s16);
+ q0u8 = vcombine_u8(d0u8, d1u8);
+ 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;
+ }
+ }
+ return;
+}
--- /dev/null
+++ b/vp9/common/arm/neon/vp9_reconintra_neon_asm.asm
@@ -1,0 +1,634 @@
+;
+; Copyright (c) 2014 The WebM project authors. All Rights Reserved.
+;
+; Use of this source code is governed by a BSD-style license
+; that can be found in the LICENSE file in the root of the source
+; tree. An additional intellectual property rights grant can be found
+; in the file PATENTS. All contributing project authors may
+; be found in the AUTHORS file in the root of the source tree.
+;
+
+ EXPORT |vp9_v_predictor_4x4_neon|
+ EXPORT |vp9_v_predictor_8x8_neon|
+ EXPORT |vp9_v_predictor_16x16_neon|
+ EXPORT |vp9_v_predictor_32x32_neon|
+ EXPORT |vp9_h_predictor_4x4_neon|
+ EXPORT |vp9_h_predictor_8x8_neon|
+ EXPORT |vp9_h_predictor_16x16_neon|
+ EXPORT |vp9_h_predictor_32x32_neon|
+ EXPORT |vp9_tm_predictor_4x4_neon|
+ EXPORT |vp9_tm_predictor_8x8_neon|
+ EXPORT |vp9_tm_predictor_16x16_neon|
+ EXPORT |vp9_tm_predictor_32x32_neon|
+ ARM
+ REQUIRE8
+ PRESERVE8
+
+ AREA ||.text||, CODE, READONLY, ALIGN=2
+
+;void vp9_v_predictor_4x4_neon(uint8_t *dst, ptrdiff_t y_stride,
+; const uint8_t *above,
+; const uint8_t *left)
+; r0 uint8_t *dst
+; r1 ptrdiff_t y_stride
+; r2 const uint8_t *above
+; r3 const uint8_t *left
+
+|vp9_v_predictor_4x4_neon| PROC
+ vld1.32 {d0[0]}, [r2]+ vst1.32 {d0[0]}, [r0], r1+ vst1.32 {d0[0]}, [r0], r1+ vst1.32 {d0[0]}, [r0], r1+ vst1.32 {d0[0]}, [r0], r1+ bx lr
+ ENDP ; |vp9_v_predictor_4x4_neon|
+
+;void vp9_v_predictor_8x8_neon(uint8_t *dst, ptrdiff_t y_stride,
+; const uint8_t *above,
+; const uint8_t *left)
+; r0 uint8_t *dst
+; r1 ptrdiff_t y_stride
+; r2 const uint8_t *above
+; r3 const uint8_t *left
+
+|vp9_v_predictor_8x8_neon| PROC
+ vld1.8 {d0}, [r2]+ vst1.8 {d0}, [r0], r1+ vst1.8 {d0}, [r0], r1+ vst1.8 {d0}, [r0], r1+ vst1.8 {d0}, [r0], r1+ vst1.8 {d0}, [r0], r1+ vst1.8 {d0}, [r0], r1+ vst1.8 {d0}, [r0], r1+ vst1.8 {d0}, [r0], r1+ bx lr
+ ENDP ; |vp9_v_predictor_8x8_neon|
+
+;void vp9_v_predictor_16x16_neon(uint8_t *dst, ptrdiff_t y_stride,
+; const uint8_t *above,
+; const uint8_t *left)
+; r0 uint8_t *dst
+; r1 ptrdiff_t y_stride
+; r2 const uint8_t *above
+; r3 const uint8_t *left
+
+|vp9_v_predictor_16x16_neon| PROC
+ vld1.8 {q0}, [r2]+ vst1.8 {q0}, [r0], r1+ vst1.8 {q0}, [r0], r1+ vst1.8 {q0}, [r0], r1+ vst1.8 {q0}, [r0], r1+ vst1.8 {q0}, [r0], r1+ vst1.8 {q0}, [r0], r1+ vst1.8 {q0}, [r0], r1+ vst1.8 {q0}, [r0], r1+ vst1.8 {q0}, [r0], r1+ vst1.8 {q0}, [r0], r1+ vst1.8 {q0}, [r0], r1+ vst1.8 {q0}, [r0], r1+ vst1.8 {q0}, [r0], r1+ vst1.8 {q0}, [r0], r1+ vst1.8 {q0}, [r0], r1+ vst1.8 {q0}, [r0], r1+ bx lr
+ ENDP ; |vp9_v_predictor_16x16_neon|
+
+;void vp9_v_predictor_32x32_neon(uint8_t *dst, ptrdiff_t y_stride,
+; const uint8_t *above,
+; const uint8_t *left)
+; r0 uint8_t *dst
+; r1 ptrdiff_t y_stride
+; r2 const uint8_t *above
+; r3 const uint8_t *left
+
+|vp9_v_predictor_32x32_neon| PROC
+ vld1.8 {q0, q1}, [r2]+ mov r2, #2
+loop_v
+ vst1.8 {q0, q1}, [r0], r1+ vst1.8 {q0, q1}, [r0], r1+ vst1.8 {q0, q1}, [r0], r1+ vst1.8 {q0, q1}, [r0], r1+ vst1.8 {q0, q1}, [r0], r1+ vst1.8 {q0, q1}, [r0], r1+ vst1.8 {q0, q1}, [r0], r1+ vst1.8 {q0, q1}, [r0], r1+ vst1.8 {q0, q1}, [r0], r1+ vst1.8 {q0, q1}, [r0], r1+ vst1.8 {q0, q1}, [r0], r1+ vst1.8 {q0, q1}, [r0], r1+ vst1.8 {q0, q1}, [r0], r1+ vst1.8 {q0, q1}, [r0], r1+ vst1.8 {q0, q1}, [r0], r1+ vst1.8 {q0, q1}, [r0], r1+ subs r2, r2, #1
+ bgt loop_v
+ bx lr
+ ENDP ; |vp9_v_predictor_32x32_neon|
+
+;void vp9_h_predictor_4x4_neon(uint8_t *dst, ptrdiff_t y_stride,
+; const uint8_t *above,
+; const uint8_t *left)
+; r0 uint8_t *dst
+; r1 ptrdiff_t y_stride
+; r2 const uint8_t *above
+; r3 const uint8_t *left
+
+|vp9_h_predictor_4x4_neon| PROC
+ vld1.32 {d1[0]}, [r3]+ vdup.8 d0, d1[0]
+ vst1.32 {d0[0]}, [r0], r1+ vdup.8 d0, d1[1]
+ vst1.32 {d0[0]}, [r0], r1+ vdup.8 d0, d1[2]
+ vst1.32 {d0[0]}, [r0], r1+ vdup.8 d0, d1[3]
+ vst1.32 {d0[0]}, [r0], r1+ bx lr
+ ENDP ; |vp9_h_predictor_4x4_neon|
+
+;void vp9_h_predictor_8x8_neon(uint8_t *dst, ptrdiff_t y_stride,
+; const uint8_t *above,
+; const uint8_t *left)
+; r0 uint8_t *dst
+; r1 ptrdiff_t y_stride
+; r2 const uint8_t *above
+; r3 const uint8_t *left
+
+|vp9_h_predictor_8x8_neon| PROC
+ vld1.64 {d1}, [r3]+ vdup.8 d0, d1[0]
+ vst1.64 {d0}, [r0], r1+ vdup.8 d0, d1[1]
+ vst1.64 {d0}, [r0], r1+ vdup.8 d0, d1[2]
+ vst1.64 {d0}, [r0], r1+ vdup.8 d0, d1[3]
+ vst1.64 {d0}, [r0], r1+ vdup.8 d0, d1[4]
+ vst1.64 {d0}, [r0], r1+ vdup.8 d0, d1[5]
+ vst1.64 {d0}, [r0], r1+ vdup.8 d0, d1[6]
+ vst1.64 {d0}, [r0], r1+ vdup.8 d0, d1[7]
+ vst1.64 {d0}, [r0], r1+ bx lr
+ ENDP ; |vp9_h_predictor_8x8_neon|
+
+;void vp9_h_predictor_16x16_neon(uint8_t *dst, ptrdiff_t y_stride,
+; const uint8_t *above,
+; const uint8_t *left)
+; r0 uint8_t *dst
+; r1 ptrdiff_t y_stride
+; r2 const uint8_t *above
+; r3 const uint8_t *left
+
+|vp9_h_predictor_16x16_neon| PROC
+ vld1.8 {q1}, [r3]+ vdup.8 q0, d2[0]
+ vst1.8 {q0}, [r0], r1+ vdup.8 q0, d2[1]
+ vst1.8 {q0}, [r0], r1+ vdup.8 q0, d2[2]
+ vst1.8 {q0}, [r0], r1+ vdup.8 q0, d2[3]
+ vst1.8 {q0}, [r0], r1+ vdup.8 q0, d2[4]
+ vst1.8 {q0}, [r0], r1+ vdup.8 q0, d2[5]
+ vst1.8 {q0}, [r0], r1+ vdup.8 q0, d2[6]
+ vst1.8 {q0}, [r0], r1+ vdup.8 q0, d2[7]
+ vst1.8 {q0}, [r0], r1+ vdup.8 q0, d3[0]
+ vst1.8 {q0}, [r0], r1+ vdup.8 q0, d3[1]
+ vst1.8 {q0}, [r0], r1+ vdup.8 q0, d3[2]
+ vst1.8 {q0}, [r0], r1+ vdup.8 q0, d3[3]
+ vst1.8 {q0}, [r0], r1+ vdup.8 q0, d3[4]
+ vst1.8 {q0}, [r0], r1+ vdup.8 q0, d3[5]
+ vst1.8 {q0}, [r0], r1+ vdup.8 q0, d3[6]
+ vst1.8 {q0}, [r0], r1+ vdup.8 q0, d3[7]
+ vst1.8 {q0}, [r0], r1+ bx lr
+ ENDP ; |vp9_h_predictor_16x16_neon|
+
+;void vp9_h_predictor_32x32_neon(uint8_t *dst, ptrdiff_t y_stride,
+; const uint8_t *above,
+; const uint8_t *left)
+; r0 uint8_t *dst
+; r1 ptrdiff_t y_stride
+; r2 const uint8_t *above
+; r3 const uint8_t *left
+
+|vp9_h_predictor_32x32_neon| PROC
+ sub r1, r1, #16
+ mov r2, #2
+loop_h
+ vld1.8 {q1}, [r3]!+ vdup.8 q0, d2[0]
+ vst1.8 {q0}, [r0]!+ vst1.8 {q0}, [r0], r1+ vdup.8 q0, d2[1]
+ vst1.8 {q0}, [r0]!+ vst1.8 {q0}, [r0], r1+ vdup.8 q0, d2[2]
+ vst1.8 {q0}, [r0]!+ vst1.8 {q0}, [r0], r1+ vdup.8 q0, d2[3]
+ vst1.8 {q0}, [r0]!+ vst1.8 {q0}, [r0], r1+ vdup.8 q0, d2[4]
+ vst1.8 {q0}, [r0]!+ vst1.8 {q0}, [r0], r1+ vdup.8 q0, d2[5]
+ vst1.8 {q0}, [r0]!+ vst1.8 {q0}, [r0], r1+ vdup.8 q0, d2[6]
+ vst1.8 {q0}, [r0]!+ vst1.8 {q0}, [r0], r1+ vdup.8 q0, d2[7]
+ vst1.8 {q0}, [r0]!+ vst1.8 {q0}, [r0], r1+ vdup.8 q0, d3[0]
+ vst1.8 {q0}, [r0]!+ vst1.8 {q0}, [r0], r1+ vdup.8 q0, d3[1]
+ vst1.8 {q0}, [r0]!+ vst1.8 {q0}, [r0], r1+ vdup.8 q0, d3[2]
+ vst1.8 {q0}, [r0]!+ vst1.8 {q0}, [r0], r1+ vdup.8 q0, d3[3]
+ vst1.8 {q0}, [r0]!+ vst1.8 {q0}, [r0], r1+ vdup.8 q0, d3[4]
+ vst1.8 {q0}, [r0]!+ vst1.8 {q0}, [r0], r1+ vdup.8 q0, d3[5]
+ vst1.8 {q0}, [r0]!+ vst1.8 {q0}, [r0], r1+ vdup.8 q0, d3[6]
+ vst1.8 {q0}, [r0]!+ vst1.8 {q0}, [r0], r1+ vdup.8 q0, d3[7]
+ vst1.8 {q0}, [r0]!+ vst1.8 {q0}, [r0], r1+ subs r2, r2, #1
+ bgt loop_h
+ bx lr
+ ENDP ; |vp9_h_predictor_32x32_neon|
+
+;void vp9_tm_predictor_4x4_neon (uint8_t *dst, ptrdiff_t y_stride,
+; const uint8_t *above,
+; const uint8_t *left)
+; r0 uint8_t *dst
+; r1 ptrdiff_t y_stride
+; r2 const uint8_t *above
+; r3 const uint8_t *left
+
+|vp9_tm_predictor_4x4_neon| PROC
+ ; Load ytop_left = above[-1];
+ sub r12, r2, #1
+ ldrb r12, [r12]
+ vdup.u8 d0, r12
+
+ ; Load above 4 pixels
+ vld1.32 {d2[0]}, [r2]+
+ ; Compute above - ytop_left
+ vsubl.u8 q3, d2, d0
+
+ ; Load left row by row and compute left + (above - ytop_left)
+ ; 1st row and 2nd row
+ ldrb r12, [r3], #1
+ ldrb r2, [r3], #1
+ vdup.u16 q1, r12
+ vdup.u16 q2, r2
+ vadd.s16 q1, q1, q3
+ vadd.s16 q2, q2, q3
+ vqmovun.s16 d0, q1
+ vqmovun.s16 d1, q2
+ vst1.32 {d0[0]}, [r0], r1+ vst1.32 {d1[0]}, [r0], r1+
+ ; 3rd row and 4th row
+ ldrb r12, [r3], #1
+ ldrb r2, [r3], #1
+ vdup.u16 q1, r12
+ vdup.u16 q2, r2
+ vadd.s16 q1, q1, q3
+ vadd.s16 q2, q2, q3
+ vqmovun.s16 d0, q1
+ vqmovun.s16 d1, q2
+ vst1.32 {d0[0]}, [r0], r1+ vst1.32 {d1[0]}, [r0], r1+ bx lr
+ ENDP ; |vp9_tm_predictor_4x4_neon|
+
+;void vp9_tm_predictor_8x8_neon (uint8_t *dst, ptrdiff_t y_stride,
+; const uint8_t *above,
+; const uint8_t *left)
+; r0 uint8_t *dst
+; r1 ptrdiff_t y_stride
+; r2 const uint8_t *above
+; r3 const uint8_t *left
+
+|vp9_tm_predictor_8x8_neon| PROC
+ ; Load ytop_left = above[-1];
+ sub r12, r2, #1
+ ldrb r12, [r12]
+ vdup.u8 d0, r12
+
+ ; preload 8 left
+ vld1.8 {d30}, [r3]+
+ ; Load above 8 pixels
+ vld1.64 {d2}, [r2]+
+ vmovl.u8 q10, d30
+
+ ; Compute above - ytop_left
+ vsubl.u8 q3, d2, d0
+
+ ; Load left row by row and compute left + (above - ytop_left)
+ ; 1st row and 2nd row
+ vdup.16 q0, d20[0]
+ vdup.16 q1, d20[1]
+ vadd.s16 q0, q3, q0
+ vadd.s16 q1, q3, q1
+
+ ; 3rd row and 4th row
+ vdup.16 q8, d20[2]
+ vdup.16 q9, d20[3]
+ vadd.s16 q8, q3, q8
+ vadd.s16 q9, q3, q9
+
+ vqmovun.s16 d0, q0
+ vqmovun.s16 d1, q1
+ vqmovun.s16 d2, q8
+ vqmovun.s16 d3, q9
+
+ vst1.64 {d0}, [r0], r1+ vst1.64 {d1}, [r0], r1+ vst1.64 {d2}, [r0], r1+ vst1.64 {d3}, [r0], r1+
+ ; 5th row and 6th row
+ vdup.16 q0, d21[0]
+ vdup.16 q1, d21[1]
+ vadd.s16 q0, q3, q0
+ vadd.s16 q1, q3, q1
+
+ ; 7th row and 8th row
+ vdup.16 q8, d21[2]
+ vdup.16 q9, d21[3]
+ vadd.s16 q8, q3, q8
+ vadd.s16 q9, q3, q9
+
+ vqmovun.s16 d0, q0
+ vqmovun.s16 d1, q1
+ vqmovun.s16 d2, q8
+ vqmovun.s16 d3, q9
+
+ vst1.64 {d0}, [r0], r1+ vst1.64 {d1}, [r0], r1+ vst1.64 {d2}, [r0], r1+ vst1.64 {d3}, [r0], r1+
+ bx lr
+ ENDP ; |vp9_tm_predictor_8x8_neon|
+
+;void vp9_tm_predictor_16x16_neon (uint8_t *dst, ptrdiff_t y_stride,
+; const uint8_t *above,
+; const uint8_t *left)
+; r0 uint8_t *dst
+; r1 ptrdiff_t y_stride
+; r2 const uint8_t *above
+; r3 const uint8_t *left
+
+|vp9_tm_predictor_16x16_neon| PROC
+ ; Load ytop_left = above[-1];
+ sub r12, r2, #1
+ ldrb r12, [r12]
+ vdup.u8 q0, r12
+
+ ; Load above 8 pixels
+ vld1.8 {q1}, [r2]+
+ ; preload 8 left into r12
+ vld1.8 {d18}, [r3]!+
+ ; Compute above - ytop_left
+ vsubl.u8 q2, d2, d0
+ vsubl.u8 q3, d3, d1
+
+ vmovl.u8 q10, d18
+
+ ; Load left row by row and compute left + (above - ytop_left)
+ ; Process 8 rows in each single loop and loop 2 times to process 16 rows.
+ mov r2, #2
+
+loop_16x16_neon
+ ; Process two rows.
+ vdup.16 q0, d20[0]
+ vdup.16 q8, d20[1]
+ vadd.s16 q1, q0, q2
+ vadd.s16 q0, q0, q3
+ vadd.s16 q11, q8, q2
+ vadd.s16 q8, q8, q3
+ vqmovun.s16 d2, q1
+ vqmovun.s16 d3, q0
+ vqmovun.s16 d22, q11
+ vqmovun.s16 d23, q8
+ vdup.16 q0, d20[2] ; proload next 2 rows data
+ vdup.16 q8, d20[3]
+ vst1.64 {d2,d3}, [r0], r1+ vst1.64 {d22,d23}, [r0], r1+
+ ; Process two rows.
+ vadd.s16 q1, q0, q2
+ vadd.s16 q0, q0, q3
+ vadd.s16 q11, q8, q2
+ vadd.s16 q8, q8, q3
+ vqmovun.s16 d2, q1
+ vqmovun.s16 d3, q0
+ vqmovun.s16 d22, q11
+ vqmovun.s16 d23, q8
+ vdup.16 q0, d21[0] ; proload next 2 rows data
+ vdup.16 q8, d21[1]
+ vst1.64 {d2,d3}, [r0], r1+ vst1.64 {d22,d23}, [r0], r1+
+ vadd.s16 q1, q0, q2
+ vadd.s16 q0, q0, q3
+ vadd.s16 q11, q8, q2
+ vadd.s16 q8, q8, q3
+ vqmovun.s16 d2, q1
+ vqmovun.s16 d3, q0
+ vqmovun.s16 d22, q11
+ vqmovun.s16 d23, q8
+ vdup.16 q0, d21[2] ; proload next 2 rows data
+ vdup.16 q8, d21[3]
+ vst1.64 {d2,d3}, [r0], r1+ vst1.64 {d22,d23}, [r0], r1+
+
+ vadd.s16 q1, q0, q2
+ vadd.s16 q0, q0, q3
+ vadd.s16 q11, q8, q2
+ vadd.s16 q8, q8, q3
+ vqmovun.s16 d2, q1
+ vqmovun.s16 d3, q0
+ vqmovun.s16 d22, q11
+ vqmovun.s16 d23, q8
+ vld1.8 {d18}, [r3]! ; preload 8 left into r12+ vmovl.u8 q10, d18
+ vst1.64 {d2,d3}, [r0], r1+ vst1.64 {d22,d23}, [r0], r1+
+ subs r2, r2, #1
+ bgt loop_16x16_neon
+
+ bx lr
+ ENDP ; |vp9_tm_predictor_16x16_neon|
+
+;void vp9_tm_predictor_32x32_neon (uint8_t *dst, ptrdiff_t y_stride,
+; const uint8_t *above,
+; const uint8_t *left)
+; r0 uint8_t *dst
+; r1 ptrdiff_t y_stride
+; r2 const uint8_t *above
+; r3 const uint8_t *left
+
+|vp9_tm_predictor_32x32_neon| PROC
+ ; Load ytop_left = above[-1];
+ sub r12, r2, #1
+ ldrb r12, [r12]
+ vdup.u8 q0, r12
+
+ ; Load above 32 pixels
+ vld1.8 {q1}, [r2]!+ vld1.8 {q2}, [r2]+
+ ; preload 8 left pixels
+ vld1.8 {d26}, [r3]!+
+ ; Compute above - ytop_left
+ vsubl.u8 q8, d2, d0
+ vsubl.u8 q9, d3, d1
+ vsubl.u8 q10, d4, d0
+ vsubl.u8 q11, d5, d1
+
+ vmovl.u8 q3, d26
+
+ ; Load left row by row and compute left + (above - ytop_left)
+ ; Process 8 rows in each single loop and loop 4 times to process 32 rows.
+ mov r2, #4
+
+loop_32x32_neon
+ ; Process two rows.
+ vdup.16 q0, d6[0]
+ vdup.16 q2, d6[1]
+ vadd.s16 q12, q0, q8
+ vadd.s16 q13, q0, q9
+ vadd.s16 q14, q0, q10
+ vadd.s16 q15, q0, q11
+ vqmovun.s16 d0, q12
+ vqmovun.s16 d1, q13
+ vadd.s16 q12, q2, q8
+ vadd.s16 q13, q2, q9
+ vqmovun.s16 d2, q14
+ vqmovun.s16 d3, q15
+ vadd.s16 q14, q2, q10
+ vadd.s16 q15, q2, q11
+ vst1.64 {d0-d3}, [r0], r1+ vqmovun.s16 d24, q12
+ vqmovun.s16 d25, q13
+ vqmovun.s16 d26, q14
+ vqmovun.s16 d27, q15
+ vdup.16 q1, d6[2]
+ vdup.16 q2, d6[3]
+ vst1.64 {d24-d27}, [r0], r1+
+ ; Process two rows.
+ vadd.s16 q12, q1, q8
+ vadd.s16 q13, q1, q9
+ vadd.s16 q14, q1, q10
+ vadd.s16 q15, q1, q11
+ vqmovun.s16 d0, q12
+ vqmovun.s16 d1, q13
+ vadd.s16 q12, q2, q8
+ vadd.s16 q13, q2, q9
+ vqmovun.s16 d2, q14
+ vqmovun.s16 d3, q15
+ vadd.s16 q14, q2, q10
+ vadd.s16 q15, q2, q11
+ vst1.64 {d0-d3}, [r0], r1+ vqmovun.s16 d24, q12
+ vqmovun.s16 d25, q13
+ vqmovun.s16 d26, q14
+ vqmovun.s16 d27, q15
+ vdup.16 q0, d7[0]
+ vdup.16 q2, d7[1]
+ vst1.64 {d24-d27}, [r0], r1+
+ ; Process two rows.
+ vadd.s16 q12, q0, q8
+ vadd.s16 q13, q0, q9
+ vadd.s16 q14, q0, q10
+ vadd.s16 q15, q0, q11
+ vqmovun.s16 d0, q12
+ vqmovun.s16 d1, q13
+ vadd.s16 q12, q2, q8
+ vadd.s16 q13, q2, q9
+ vqmovun.s16 d2, q14
+ vqmovun.s16 d3, q15
+ vadd.s16 q14, q2, q10
+ vadd.s16 q15, q2, q11
+ vst1.64 {d0-d3}, [r0], r1+ vqmovun.s16 d24, q12
+ vqmovun.s16 d25, q13
+ vqmovun.s16 d26, q14
+ vqmovun.s16 d27, q15
+ vdup.16 q0, d7[2]
+ vdup.16 q2, d7[3]
+ vst1.64 {d24-d27}, [r0], r1+
+ ; Process two rows.
+ vadd.s16 q12, q0, q8
+ vadd.s16 q13, q0, q9
+ vadd.s16 q14, q0, q10
+ vadd.s16 q15, q0, q11
+ vqmovun.s16 d0, q12
+ vqmovun.s16 d1, q13
+ vadd.s16 q12, q2, q8
+ vadd.s16 q13, q2, q9
+ vqmovun.s16 d2, q14
+ vqmovun.s16 d3, q15
+ vadd.s16 q14, q2, q10
+ vadd.s16 q15, q2, q11
+ vst1.64 {d0-d3}, [r0], r1+ vqmovun.s16 d24, q12
+ vqmovun.s16 d25, q13
+ vld1.8 {d0}, [r3]! ; preload 8 left pixels+ vqmovun.s16 d26, q14
+ vqmovun.s16 d27, q15
+ vmovl.u8 q3, d0
+ vst1.64 {d24-d27}, [r0], r1+
+ subs r2, r2, #1
+ bgt loop_32x32_neon
+
+ bx lr
+ ENDP ; |vp9_tm_predictor_32x32_neon|
+
+ END
--- a/vp9/common/vp9_rtcd_defs.pl
+++ b/vp9/common/vp9_rtcd_defs.pl
@@ -66,8 +66,7 @@
specialize qw/vp9_d63_predictor_4x4/, "$ssse3_x86inc";
add_proto qw/void vp9_h_predictor_4x4/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
-specialize qw/vp9_h_predictor_4x4 neon_asm dspr2/, "$ssse3_x86inc";
-$vp9_h_predictor_4x4_neon_asm=vp9_h_predictor_4x4_neon;
+specialize qw/vp9_h_predictor_4x4 neon dspr2/, "$ssse3_x86inc";
add_proto qw/void vp9_d117_predictor_4x4/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
specialize qw/vp9_d117_predictor_4x4/;
@@ -79,12 +78,10 @@
specialize qw/vp9_d153_predictor_4x4/, "$ssse3_x86inc";
add_proto qw/void vp9_v_predictor_4x4/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
-specialize qw/vp9_v_predictor_4x4 neon_asm/, "$sse_x86inc";
-$vp9_v_predictor_4x4_neon_asm=vp9_v_predictor_4x4_neon;
+specialize qw/vp9_v_predictor_4x4 neon/, "$sse_x86inc";
add_proto qw/void vp9_tm_predictor_4x4/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
-specialize qw/vp9_tm_predictor_4x4 neon_asm dspr2/, "$sse_x86inc";
-$vp9_tm_predictor_4x4_neon_asm=vp9_tm_predictor_4x4_neon;
+specialize qw/vp9_tm_predictor_4x4 neon dspr2/, "$sse_x86inc";
add_proto qw/void vp9_dc_predictor_4x4/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
specialize qw/vp9_dc_predictor_4x4 dspr2/, "$sse_x86inc";
@@ -108,8 +105,7 @@
specialize qw/vp9_d63_predictor_8x8/, "$ssse3_x86inc";
add_proto qw/void vp9_h_predictor_8x8/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
-specialize qw/vp9_h_predictor_8x8 neon_asm dspr2/, "$ssse3_x86inc";
-$vp9_h_predictor_8x8_neon_asm=vp9_h_predictor_8x8_neon;
+specialize qw/vp9_h_predictor_8x8 neon dspr2/, "$ssse3_x86inc";
add_proto qw/void vp9_d117_predictor_8x8/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
specialize qw/vp9_d117_predictor_8x8/;
@@ -121,12 +117,10 @@
specialize qw/vp9_d153_predictor_8x8/, "$ssse3_x86inc";
add_proto qw/void vp9_v_predictor_8x8/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
-specialize qw/vp9_v_predictor_8x8 neon_asm/, "$sse_x86inc";
-$vp9_v_predictor_8x8_neon_asm=vp9_v_predictor_8x8_neon;
+specialize qw/vp9_v_predictor_8x8 neon/, "$sse_x86inc";
add_proto qw/void vp9_tm_predictor_8x8/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
-specialize qw/vp9_tm_predictor_8x8 neon_asm dspr2/, "$sse2_x86inc";
-$vp9_tm_predictor_8x8_neon_asm=vp9_tm_predictor_8x8_neon;
+specialize qw/vp9_tm_predictor_8x8 neon dspr2/, "$sse2_x86inc";
add_proto qw/void vp9_dc_predictor_8x8/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
specialize qw/vp9_dc_predictor_8x8 dspr2/, "$sse_x86inc";
@@ -150,8 +144,7 @@
specialize qw/vp9_d63_predictor_16x16/, "$ssse3_x86inc";
add_proto qw/void vp9_h_predictor_16x16/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
-specialize qw/vp9_h_predictor_16x16 neon_asm dspr2/, "$ssse3_x86inc";
-$vp9_h_predictor_16x16_neon_asm=vp9_h_predictor_16x16_neon;
+specialize qw/vp9_h_predictor_16x16 neon dspr2/, "$ssse3_x86inc";
add_proto qw/void vp9_d117_predictor_16x16/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
specialize qw/vp9_d117_predictor_16x16/;
@@ -163,12 +156,10 @@
specialize qw/vp9_d153_predictor_16x16/, "$ssse3_x86inc";
add_proto qw/void vp9_v_predictor_16x16/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
-specialize qw/vp9_v_predictor_16x16 neon_asm/, "$sse2_x86inc";
-$vp9_v_predictor_16x16_neon_asm=vp9_v_predictor_16x16_neon;
+specialize qw/vp9_v_predictor_16x16 neon/, "$sse2_x86inc";
add_proto qw/void vp9_tm_predictor_16x16/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
-specialize qw/vp9_tm_predictor_16x16 neon_asm/, "$sse2_x86inc";
-$vp9_tm_predictor_16x16_neon_asm=vp9_tm_predictor_16x16_neon;
+specialize qw/vp9_tm_predictor_16x16 neon/, "$sse2_x86inc";
add_proto qw/void vp9_dc_predictor_16x16/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
specialize qw/vp9_dc_predictor_16x16 dspr2/, "$sse2_x86inc";
@@ -192,8 +183,7 @@
specialize qw/vp9_d63_predictor_32x32/, "$ssse3_x86inc";
add_proto qw/void vp9_h_predictor_32x32/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
-specialize qw/vp9_h_predictor_32x32 neon_asm/, "$ssse3_x86inc";
-$vp9_h_predictor_32x32_neon_asm=vp9_h_predictor_32x32_neon;
+specialize qw/vp9_h_predictor_32x32 neon/, "$ssse3_x86inc";
add_proto qw/void vp9_d117_predictor_32x32/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
specialize qw/vp9_d117_predictor_32x32/;
@@ -205,12 +195,10 @@
specialize qw/vp9_d153_predictor_32x32/;
add_proto qw/void vp9_v_predictor_32x32/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
-specialize qw/vp9_v_predictor_32x32 neon_asm/, "$sse2_x86inc";
-$vp9_v_predictor_32x32_neon_asm=vp9_v_predictor_32x32_neon;
+specialize qw/vp9_v_predictor_32x32 neon/, "$sse2_x86inc";
add_proto qw/void vp9_tm_predictor_32x32/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
-specialize qw/vp9_tm_predictor_32x32 neon_asm/, "$sse2_x86_64";
-$vp9_tm_predictor_32x32_neon_asm=vp9_tm_predictor_32x32_neon;
+specialize qw/vp9_tm_predictor_32x32 neon/, "$sse2_x86_64";
add_proto qw/void vp9_dc_predictor_32x32/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
specialize qw/vp9_dc_predictor_32x32/, "$sse2_x86inc";
--- a/vp9/vp9_common.mk
+++ b/vp9/vp9_common.mk
@@ -136,7 +136,6 @@
VP9_COMMON_SRCS-$(HAVE_NEON_ASM) += common/arm/neon/vp9_loopfilter_16_neon_asm$(ASM)
VP9_COMMON_SRCS-$(HAVE_NEON_ASM) += common/arm/neon/vp9_mb_lpf_neon$(ASM)
VP9_COMMON_SRCS-$(HAVE_NEON_ASM) += common/arm/neon/vp9_save_reg_neon$(ASM)
-VP9_COMMON_SRCS-$(HAVE_NEON_ASM) += common/arm/neon/vp9_reconintra_neon$(ASM)
VP9_COMMON_SRCS-$(HAVE_NEON) += common/arm/neon/vp9_iht4x4_add_neon.c
VP9_COMMON_SRCS-$(HAVE_NEON) += common/arm/neon/vp9_iht8x8_add_neon.c
@@ -160,6 +159,7 @@
VP9_COMMON_SRCS-yes += common/arm/neon/vp9_idct8x8_add_neon_asm$(ASM)
VP9_COMMON_SRCS-yes += common/arm/neon/vp9_loopfilter_neon_asm$(ASM)
VP9_COMMON_SRCS-yes += common/arm/neon/vp9_loopfilter_16_neon.c
+VP9_COMMON_SRCS-yes += common/arm/neon/vp9_reconintra_neon_asm$(ASM)
else
ifeq ($(HAVE_NEON), yes)
VP9_COMMON_SRCS-yes += common/arm/neon/vp9_avg_neon.c
@@ -178,6 +178,7 @@
VP9_COMMON_SRCS-yes += common/arm/neon/vp9_idct8x8_add_neon.c
VP9_COMMON_SRCS-yes += common/arm/neon/vp9_loopfilter_neon.c
VP9_COMMON_SRCS-yes += common/arm/neon/vp9_loopfilter_16_neon.c
+VP9_COMMON_SRCS-yes += common/arm/neon/vp9_reconintra_neon.c
endif # HAVE_NEON
endif # HAVE_NEON_ASM