1.1 --- /dev/null Thu Jan 01 00:00:00 1970 +0000
1.2 +++ b/media/libvpx/vp8/encoder/arm/neon/vp8_mse16x16_neon.asm Wed Dec 31 06:09:35 2014 +0100
1.3 @@ -0,0 +1,116 @@
1.4 +;
1.5 +; Copyright (c) 2010 The WebM project authors. All Rights Reserved.
1.6 +;
1.7 +; Use of this source code is governed by a BSD-style license
1.8 +; that can be found in the LICENSE file in the root of the source
1.9 +; tree. An additional intellectual property rights grant can be found
1.10 +; in the file PATENTS. All contributing project authors may
1.11 +; be found in the AUTHORS file in the root of the source tree.
1.12 +;
1.13 +
1.14 +
1.15 + EXPORT |vp8_mse16x16_neon|
1.16 + EXPORT |vp8_get4x4sse_cs_neon|
1.17 +
1.18 + ARM
1.19 + REQUIRE8
1.20 + PRESERVE8
1.21 +
1.22 + AREA ||.text||, CODE, READONLY, ALIGN=2
1.23 +;============================
1.24 +; r0 unsigned char *src_ptr
1.25 +; r1 int source_stride
1.26 +; r2 unsigned char *ref_ptr
1.27 +; r3 int recon_stride
1.28 +; stack unsigned int *sse
1.29 +;note: in this function, sum is never used. So, we can remove this part of calculation
1.30 +;from vp8_variance().
1.31 +
1.32 +|vp8_mse16x16_neon| PROC
1.33 + vmov.i8 q7, #0 ;q7, q8, q9, q10 - sse
1.34 + vmov.i8 q8, #0
1.35 + vmov.i8 q9, #0
1.36 + vmov.i8 q10, #0
1.37 +
1.38 + mov r12, #8
1.39 +
1.40 +mse16x16_neon_loop
1.41 + vld1.8 {q0}, [r0], r1 ;Load up source and reference
1.42 + vld1.8 {q2}, [r2], r3
1.43 + vld1.8 {q1}, [r0], r1
1.44 + vld1.8 {q3}, [r2], r3
1.45 +
1.46 + vsubl.u8 q11, d0, d4
1.47 + vsubl.u8 q12, d1, d5
1.48 + vsubl.u8 q13, d2, d6
1.49 + vsubl.u8 q14, d3, d7
1.50 +
1.51 + vmlal.s16 q7, d22, d22
1.52 + vmlal.s16 q8, d23, d23
1.53 +
1.54 + subs r12, r12, #1
1.55 +
1.56 + vmlal.s16 q9, d24, d24
1.57 + vmlal.s16 q10, d25, d25
1.58 + vmlal.s16 q7, d26, d26
1.59 + vmlal.s16 q8, d27, d27
1.60 + vmlal.s16 q9, d28, d28
1.61 + vmlal.s16 q10, d29, d29
1.62 +
1.63 + bne mse16x16_neon_loop
1.64 +
1.65 + vadd.u32 q7, q7, q8
1.66 + vadd.u32 q9, q9, q10
1.67 +
1.68 + ldr r12, [sp] ;load *sse from stack
1.69 +
1.70 + vadd.u32 q10, q7, q9
1.71 + vpaddl.u32 q1, q10
1.72 + vadd.u64 d0, d2, d3
1.73 +
1.74 + vst1.32 {d0[0]}, [r12]
1.75 + vmov.32 r0, d0[0]
1.76 +
1.77 + bx lr
1.78 +
1.79 + ENDP
1.80 +
1.81 +
1.82 +;=============================
1.83 +; r0 unsigned char *src_ptr,
1.84 +; r1 int source_stride,
1.85 +; r2 unsigned char *ref_ptr,
1.86 +; r3 int recon_stride
1.87 +|vp8_get4x4sse_cs_neon| PROC
1.88 + vld1.8 {d0}, [r0], r1 ;Load up source and reference
1.89 + vld1.8 {d4}, [r2], r3
1.90 + vld1.8 {d1}, [r0], r1
1.91 + vld1.8 {d5}, [r2], r3
1.92 + vld1.8 {d2}, [r0], r1
1.93 + vld1.8 {d6}, [r2], r3
1.94 + vld1.8 {d3}, [r0], r1
1.95 + vld1.8 {d7}, [r2], r3
1.96 +
1.97 + vsubl.u8 q11, d0, d4
1.98 + vsubl.u8 q12, d1, d5
1.99 + vsubl.u8 q13, d2, d6
1.100 + vsubl.u8 q14, d3, d7
1.101 +
1.102 + vmull.s16 q7, d22, d22
1.103 + vmull.s16 q8, d24, d24
1.104 + vmull.s16 q9, d26, d26
1.105 + vmull.s16 q10, d28, d28
1.106 +
1.107 + vadd.u32 q7, q7, q8
1.108 + vadd.u32 q9, q9, q10
1.109 + vadd.u32 q9, q7, q9
1.110 +
1.111 + vpaddl.u32 q1, q9
1.112 + vadd.u64 d0, d2, d3
1.113 +
1.114 + vmov.32 r0, d0[0]
1.115 + bx lr
1.116 +
1.117 + ENDP
1.118 +
1.119 + END
