1.1 --- /dev/null Thu Jan 01 00:00:00 1970 +0000
1.2 +++ b/media/libvpx/vp9/encoder/x86/vp9_subtract_sse2.asm Wed Dec 31 06:09:35 2014 +0100
1.3 @@ -0,0 +1,127 @@
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 +%include "third_party/x86inc/x86inc.asm"
1.15 +
1.16 +SECTION .text
1.17 +
1.18 +; void vp9_subtract_block(int rows, int cols,
1.19 +; int16_t *diff, ptrdiff_t diff_stride,
1.20 +; const uint8_t *src, ptrdiff_t src_stride,
1.21 +; const uint8_t *pred, ptrdiff_t pred_stride)
1.22 +
1.23 +INIT_XMM sse2
1.24 +cglobal subtract_block, 7, 7, 8, \
1.25 + rows, cols, diff, diff_stride, src, src_stride, \
1.26 + pred, pred_stride
1.27 +%define pred_str colsq
1.28 + pxor m7, m7 ; dedicated zero register
1.29 + cmp colsd, 4
1.30 + je .case_4
1.31 + cmp colsd, 8
1.32 + je .case_8
1.33 + cmp colsd, 16
1.34 + je .case_16
1.35 + cmp colsd, 32
1.36 + je .case_32
1.37 +
1.38 +%macro loop16 6
1.39 + mova m0, [srcq+%1]
1.40 + mova m4, [srcq+%2]
1.41 + mova m1, [predq+%3]
1.42 + mova m5, [predq+%4]
1.43 + punpckhbw m2, m0, m7
1.44 + punpckhbw m3, m1, m7
1.45 + punpcklbw m0, m7
1.46 + punpcklbw m1, m7
1.47 + psubw m2, m3
1.48 + psubw m0, m1
1.49 + punpckhbw m1, m4, m7
1.50 + punpckhbw m3, m5, m7
1.51 + punpcklbw m4, m7
1.52 + punpcklbw m5, m7
1.53 + psubw m1, m3
1.54 + psubw m4, m5
1.55 + mova [diffq+mmsize*0+%5], m0
1.56 + mova [diffq+mmsize*1+%5], m2
1.57 + mova [diffq+mmsize*0+%6], m4
1.58 + mova [diffq+mmsize*1+%6], m1
1.59 +%endmacro
1.60 +
1.61 + mov pred_str, pred_stridemp
1.62 +.loop_64:
1.63 + loop16 0*mmsize, 1*mmsize, 0*mmsize, 1*mmsize, 0*mmsize, 2*mmsize
1.64 + loop16 2*mmsize, 3*mmsize, 2*mmsize, 3*mmsize, 4*mmsize, 6*mmsize
1.65 + lea diffq, [diffq+diff_strideq*2]
1.66 + add predq, pred_str
1.67 + add srcq, src_strideq
1.68 + dec rowsd
1.69 + jg .loop_64
1.70 + RET
1.71 +
1.72 +.case_32:
1.73 + mov pred_str, pred_stridemp
1.74 +.loop_32:
1.75 + loop16 0, mmsize, 0, mmsize, 0, 2*mmsize
1.76 + lea diffq, [diffq+diff_strideq*2]
1.77 + add predq, pred_str
1.78 + add srcq, src_strideq
1.79 + dec rowsd
1.80 + jg .loop_32
1.81 + RET
1.82 +
1.83 +.case_16:
1.84 + mov pred_str, pred_stridemp
1.85 +.loop_16:
1.86 + loop16 0, src_strideq, 0, pred_str, 0, diff_strideq*2
1.87 + lea diffq, [diffq+diff_strideq*4]
1.88 + lea predq, [predq+pred_str*2]
1.89 + lea srcq, [srcq+src_strideq*2]
1.90 + sub rowsd, 2
1.91 + jg .loop_16
1.92 + RET
1.93 +
1.94 +%macro loop_h 0
1.95 + movh m0, [srcq]
1.96 + movh m2, [srcq+src_strideq]
1.97 + movh m1, [predq]
1.98 + movh m3, [predq+pred_str]
1.99 + punpcklbw m0, m7
1.100 + punpcklbw m1, m7
1.101 + punpcklbw m2, m7
1.102 + punpcklbw m3, m7
1.103 + psubw m0, m1
1.104 + psubw m2, m3
1.105 + mova [diffq], m0
1.106 + mova [diffq+diff_strideq*2], m2
1.107 +%endmacro
1.108 +
1.109 +.case_8:
1.110 + mov pred_str, pred_stridemp
1.111 +.loop_8:
1.112 + loop_h
1.113 + lea diffq, [diffq+diff_strideq*4]
1.114 + lea srcq, [srcq+src_strideq*2]
1.115 + lea predq, [predq+pred_str*2]
1.116 + sub rowsd, 2
1.117 + jg .loop_8
1.118 + RET
1.119 +
1.120 +INIT_MMX
1.121 +.case_4:
1.122 + mov pred_str, pred_stridemp
1.123 +.loop_4:
1.124 + loop_h
1.125 + lea diffq, [diffq+diff_strideq*4]
1.126 + lea srcq, [srcq+src_strideq*2]
1.127 + lea predq, [predq+pred_str*2]
1.128 + sub rowsd, 2
1.129 + jg .loop_4
1.130 + RET
