| 1 |
; Copyright (c) 2018, The rav1e contributors. All rights reserved |
| 2 |
; |
| 3 |
; This source code is subject to the terms of the BSD 2 Clause License and |
| 4 |
; the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License |
| 5 |
; was not distributed with this source code in the LICENSE file, you can |
| 6 |
; obtain it at www.aomedia.org/license/software. If the Alliance for Open |
| 7 |
; Media Patent License 1.0 was not distributed with this source code in the |
| 8 |
; PATENTS file, you can obtain it at www.aomedia.org/license/patent. |
| 9 |
|
| 10 |
%include "config.asm" |
| 11 |
%include "ext/x86/x86inc.asm" |
| 12 |
|
| 13 |
SECTION .text |
| 14 |
|
| 15 |
%macro W_ABS_DIFF 8 |
| 16 |
psubw %1, %5 |
| 17 |
psubw %2, %6 |
| 18 |
psubw %3, %7 |
| 19 |
psubw %4, %8 |
| 20 |
pabsw %1, %1 |
| 21 |
pabsw %2, %2 |
| 22 |
pabsw %3, %3 |
| 23 |
pabsw %4, %4 |
| 24 |
%endmacro |
| 25 |
|
| 26 |
INIT_XMM ssse3 |
| 27 |
cglobal sad_4x4_hbd, 4, 6, 8, src, src_stride, dst, dst_stride, \ |
| 28 |
src_stride3, dst_stride3 |
| 29 |
lea src_stride3q, [src_strideq*3] |
| 30 |
lea dst_stride3q, [dst_strideq*3] |
| 31 |
movq m0, [srcq] |
| 32 |
movq m1, [srcq+src_strideq*1] |
| 33 |
movq m2, [srcq+src_strideq*2] |
| 34 |
movq m3, [srcq+src_stride3q] |
| 35 |
movq m4, [dstq] |
| 36 |
movq m5, [dstq+dst_strideq*1] |
| 37 |
movq m6, [dstq+dst_strideq*2] |
| 38 |
movq m7, [dstq+dst_stride3q] |
| 39 |
W_ABS_DIFF m0, m1, m2, m3, m4, m5, m6, m7 |
| 40 |
; Don't convert to 32 bit integers: 4*4 abs diffs of 12-bits fits in 16 bits. |
| 41 |
; Accumulate onto m0 |
| 42 |
%define sum m0 |
| 43 |
paddw sum, m1 |
| 44 |
paddw m2, m3 |
| 45 |
paddw sum, m2 |
| 46 |
; Horizontal reduction |
| 47 |
pshuflw m1, sum, q2323 |
| 48 |
paddw sum, m1 |
| 49 |
pshuflw m1, sum, q1111 |
| 50 |
paddw sum, m1 |
| 51 |
movd eax, sum |
| 52 |
; Convert to 16-bits since the upper half of eax is dirty |
| 53 |
movzx eax, ax |
| 54 |
RET |
| 55 |
|
| 56 |
%if ARCH_X86_64 |
| 57 |
|
| 58 |
INIT_XMM ssse3 |
| 59 |
cglobal sad_16x16_hbd, 4, 5, 10, src, src_stride, dst, dst_stride, \ |
| 60 |
cnt |
| 61 |
mov cntd, 8 |
| 62 |
%define sum m0 |
| 63 |
pxor sum, sum |
| 64 |
pxor m9, m9 |
| 65 |
.loop: |
| 66 |
movu m1, [srcq] |
| 67 |
movu m2, [srcq+16] |
| 68 |
movu m3, [srcq+src_strideq] |
| 69 |
movu m4, [srcq+src_strideq+16] |
| 70 |
lea srcq, [srcq+src_strideq*2] |
| 71 |
movu m5, [dstq] |
| 72 |
movu m6, [dstq+16] |
| 73 |
movu m7, [dstq+dst_strideq] |
| 74 |
movu m8, [dstq+dst_strideq+16] |
| 75 |
lea dstq, [dstq+dst_strideq*2] |
| 76 |
W_ABS_DIFF m1, m2, m3, m4, m5, m6, m7, m8 |
| 77 |
paddw m1, m2 |
| 78 |
paddw m3, m4 |
| 79 |
; Convert to 32-bits |
| 80 |
punpcklwd m2, m1, m9 |
| 81 |
punpcklwd m4, m3, m9 |
| 82 |
punpckhwd m1, m9 |
| 83 |
punpckhwd m3, m9 |
| 84 |
paddd m1, m2 |
| 85 |
paddd m3, m4 |
| 86 |
paddd sum, m1 |
| 87 |
paddd sum, m3 |
| 88 |
dec cntd |
| 89 |
jg .loop |
| 90 |
; Horizontal reduction |
| 91 |
movhlps m1, sum |
| 92 |
paddd sum, m1 |
| 93 |
pshufd m1, sum, q1111 |
| 94 |
paddd sum, m1 |
| 95 |
movd eax, sum |
| 96 |
RET |
| 97 |
|
| 98 |
%endif |