1.1 --- /dev/null Thu Jan 01 00:00:00 1970 +0000
1.2 +++ b/media/libvpx/vp8/encoder/arm/armv6/vp8_mse16x16_armv6.asm Wed Dec 31 06:09:35 2014 +0100
1.3 @@ -0,0 +1,138 @@
1.4 +;
1.5 +; Copyright (c) 2011 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_armv6|
1.16 +
1.17 + ARM
1.18 +
1.19 + AREA ||.text||, CODE, READONLY, ALIGN=2
1.20 +
1.21 +; r0 unsigned char *src_ptr
1.22 +; r1 int source_stride
1.23 +; r2 unsigned char *ref_ptr
1.24 +; r3 int recon_stride
1.25 +; stack unsigned int *sse
1.26 +;
1.27 +;note: Based on vp8_variance16x16_armv6. In this function, sum is never used.
1.28 +; So, we can remove this part of calculation.
1.29 +
1.30 +|vp8_mse16x16_armv6| PROC
1.31 +
1.32 + push {r4-r9, lr}
1.33 +
1.34 + pld [r0, r1, lsl #0]
1.35 + pld [r2, r3, lsl #0]
1.36 +
1.37 + mov r12, #16 ; set loop counter to 16 (=block height)
1.38 + mov r4, #0 ; initialize sse = 0
1.39 +
1.40 +loop
1.41 + ; 1st 4 pixels
1.42 + ldr r5, [r0, #0x0] ; load 4 src pixels
1.43 + ldr r6, [r2, #0x0] ; load 4 ref pixels
1.44 +
1.45 + mov lr, #0 ; constant zero
1.46 +
1.47 + usub8 r8, r5, r6 ; calculate difference
1.48 + pld [r0, r1, lsl #1]
1.49 + sel r7, r8, lr ; select bytes with positive difference
1.50 + usub8 r9, r6, r5 ; calculate difference with reversed operands
1.51 + pld [r2, r3, lsl #1]
1.52 + sel r8, r9, lr ; select bytes with negative difference
1.53 +
1.54 + ; calculate partial sums
1.55 + usad8 r5, r7, lr ; calculate sum of positive differences
1.56 + usad8 r6, r8, lr ; calculate sum of negative differences
1.57 + orr r8, r8, r7 ; differences of all 4 pixels
1.58 +
1.59 + ldr r5, [r0, #0x4] ; load 4 src pixels
1.60 +
1.61 + ; calculate sse
1.62 + uxtb16 r6, r8 ; byte (two pixels) to halfwords
1.63 + uxtb16 r7, r8, ror #8 ; another two pixels to halfwords
1.64 + smlad r4, r6, r6, r4 ; dual signed multiply, add and accumulate (1)
1.65 +
1.66 + ; 2nd 4 pixels
1.67 + ldr r6, [r2, #0x4] ; load 4 ref pixels
1.68 + smlad r4, r7, r7, r4 ; dual signed multiply, add and accumulate (2)
1.69 +
1.70 + usub8 r8, r5, r6 ; calculate difference
1.71 + sel r7, r8, lr ; select bytes with positive difference
1.72 + usub8 r9, r6, r5 ; calculate difference with reversed operands
1.73 + sel r8, r9, lr ; select bytes with negative difference
1.74 +
1.75 + ; calculate partial sums
1.76 + usad8 r5, r7, lr ; calculate sum of positive differences
1.77 + usad8 r6, r8, lr ; calculate sum of negative differences
1.78 + orr r8, r8, r7 ; differences of all 4 pixels
1.79 + ldr r5, [r0, #0x8] ; load 4 src pixels
1.80 + ; calculate sse
1.81 + uxtb16 r6, r8 ; byte (two pixels) to halfwords
1.82 + uxtb16 r7, r8, ror #8 ; another two pixels to halfwords
1.83 + smlad r4, r6, r6, r4 ; dual signed multiply, add and accumulate (1)
1.84 +
1.85 + ; 3rd 4 pixels
1.86 + ldr r6, [r2, #0x8] ; load 4 ref pixels
1.87 + smlad r4, r7, r7, r4 ; dual signed multiply, add and accumulate (2)
1.88 +
1.89 + usub8 r8, r5, r6 ; calculate difference
1.90 + sel r7, r8, lr ; select bytes with positive difference
1.91 + usub8 r9, r6, r5 ; calculate difference with reversed operands
1.92 + sel r8, r9, lr ; select bytes with negative difference
1.93 +
1.94 + ; calculate partial sums
1.95 + usad8 r5, r7, lr ; calculate sum of positive differences
1.96 + usad8 r6, r8, lr ; calculate sum of negative differences
1.97 + orr r8, r8, r7 ; differences of all 4 pixels
1.98 +
1.99 + ldr r5, [r0, #0xc] ; load 4 src pixels
1.100 +
1.101 + ; calculate sse
1.102 + uxtb16 r6, r8 ; byte (two pixels) to halfwords
1.103 + uxtb16 r7, r8, ror #8 ; another two pixels to halfwords
1.104 + smlad r4, r6, r6, r4 ; dual signed multiply, add and accumulate (1)
1.105 +
1.106 + ; 4th 4 pixels
1.107 + ldr r6, [r2, #0xc] ; load 4 ref pixels
1.108 + smlad r4, r7, r7, r4 ; dual signed multiply, add and accumulate (2)
1.109 +
1.110 + usub8 r8, r5, r6 ; calculate difference
1.111 + add r0, r0, r1 ; set src_ptr to next row
1.112 + sel r7, r8, lr ; select bytes with positive difference
1.113 + usub8 r9, r6, r5 ; calculate difference with reversed operands
1.114 + add r2, r2, r3 ; set dst_ptr to next row
1.115 + sel r8, r9, lr ; select bytes with negative difference
1.116 +
1.117 + ; calculate partial sums
1.118 + usad8 r5, r7, lr ; calculate sum of positive differences
1.119 + usad8 r6, r8, lr ; calculate sum of negative differences
1.120 + orr r8, r8, r7 ; differences of all 4 pixels
1.121 +
1.122 + subs r12, r12, #1 ; next row
1.123 +
1.124 + ; calculate sse
1.125 + uxtb16 r6, r8 ; byte (two pixels) to halfwords
1.126 + uxtb16 r7, r8, ror #8 ; another two pixels to halfwords
1.127 + smlad r4, r6, r6, r4 ; dual signed multiply, add and accumulate (1)
1.128 + smlad r4, r7, r7, r4 ; dual signed multiply, add and accumulate (2)
1.129 +
1.130 + bne loop
1.131 +
1.132 + ; return stuff
1.133 + ldr r1, [sp, #28] ; get address of sse
1.134 + mov r0, r4 ; return sse
1.135 + str r4, [r1] ; store sse
1.136 +
1.137 + pop {r4-r9, pc}
1.138 +
1.139 + ENDP
1.140 +
1.141 + END
