1.1 --- /dev/null Thu Jan 01 00:00:00 1970 +0000
1.2 +++ b/media/libyuv/source/row_x86.asm Wed Dec 31 06:09:35 2014 +0100
1.3 @@ -0,0 +1,146 @@
1.4 +;
1.5 +; Copyright 2012 The LibYuv 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 +%ifdef __YASM_VERSION_ID__
1.15 +%if __YASM_VERSION_ID__ < 01020000h
1.16 +%error AVX2 is supported only by yasm 1.2.0 or later.
1.17 +%endif
1.18 +%endif
1.19 +%include "x86inc.asm"
1.20 +
1.21 +SECTION .text
1.22 +
1.23 +; cglobal numeric constants are parameters, gpr regs, mm regs
1.24 +
1.25 +; void YUY2ToYRow_SSE2(const uint8* src_yuy2, uint8* dst_y, int pix)
1.26 +
1.27 +%macro YUY2TOYROW 2-3
1.28 +cglobal %1ToYRow%3, 3, 3, 3, src_yuy2, dst_y, pix
1.29 +%ifidn %1,YUY2
1.30 + pcmpeqb m2, m2, m2 ; generate mask 0x00ff00ff
1.31 + psrlw m2, m2, 8
1.32 +%endif
1.33 +
1.34 + ALIGN 4
1.35 +.convertloop:
1.36 + mov%2 m0, [src_yuy2q]
1.37 + mov%2 m1, [src_yuy2q + mmsize]
1.38 + lea src_yuy2q, [src_yuy2q + mmsize * 2]
1.39 +%ifidn %1,YUY2
1.40 + pand m0, m0, m2 ; YUY2 even bytes are Y
1.41 + pand m1, m1, m2
1.42 +%else
1.43 + psrlw m0, m0, 8 ; UYVY odd bytes are Y
1.44 + psrlw m1, m1, 8
1.45 +%endif
1.46 + packuswb m0, m0, m1
1.47 +%if cpuflag(AVX2)
1.48 + vpermq m0, m0, 0xd8
1.49 +%endif
1.50 + sub pixd, mmsize
1.51 + mov%2 [dst_yq], m0
1.52 + lea dst_yq, [dst_yq + mmsize]
1.53 + jg .convertloop
1.54 + REP_RET
1.55 +%endmacro
1.56 +
1.57 +; TODO(fbarchard): Remove MMX. Add SSSE3 pshufb version.
1.58 +INIT_MMX MMX
1.59 +YUY2TOYROW YUY2,a,
1.60 +YUY2TOYROW YUY2,u,_Unaligned
1.61 +YUY2TOYROW UYVY,a,
1.62 +YUY2TOYROW UYVY,u,_Unaligned
1.63 +INIT_XMM SSE2
1.64 +YUY2TOYROW YUY2,a,
1.65 +YUY2TOYROW YUY2,u,_Unaligned
1.66 +YUY2TOYROW UYVY,a,
1.67 +YUY2TOYROW UYVY,u,_Unaligned
1.68 +INIT_YMM AVX2
1.69 +YUY2TOYROW YUY2,a,
1.70 +YUY2TOYROW UYVY,a,
1.71 +
1.72 +; void SplitUVRow_SSE2(const uint8* src_uv, uint8* dst_u, uint8* dst_v, int pix)
1.73 +
1.74 +%macro SplitUVRow 1-2
1.75 +cglobal SplitUVRow%2, 4, 4, 5, src_uv, dst_u, dst_v, pix
1.76 + pcmpeqb m4, m4, m4 ; generate mask 0x00ff00ff
1.77 + psrlw m4, m4, 8
1.78 + sub dst_vq, dst_uq
1.79 +
1.80 + ALIGN 4
1.81 +.convertloop:
1.82 + mov%1 m0, [src_uvq]
1.83 + mov%1 m1, [src_uvq + mmsize]
1.84 + lea src_uvq, [src_uvq + mmsize * 2]
1.85 + psrlw m2, m0, 8 ; odd bytes
1.86 + psrlw m3, m1, 8
1.87 + pand m0, m0, m4 ; even bytes
1.88 + pand m1, m1, m4
1.89 + packuswb m0, m0, m1
1.90 + packuswb m2, m2, m3
1.91 +%if cpuflag(AVX2)
1.92 + vpermq m0, m0, 0xd8
1.93 + vpermq m2, m2, 0xd8
1.94 +%endif
1.95 + mov%1 [dst_uq], m0
1.96 + mov%1 [dst_uq + dst_vq], m2
1.97 + lea dst_uq, [dst_uq + mmsize]
1.98 + sub pixd, mmsize
1.99 + jg .convertloop
1.100 + REP_RET
1.101 +%endmacro
1.102 +
1.103 +INIT_MMX MMX
1.104 +SplitUVRow a,
1.105 +SplitUVRow u,_Unaligned
1.106 +INIT_XMM SSE2
1.107 +SplitUVRow a,
1.108 +SplitUVRow u,_Unaligned
1.109 +INIT_YMM AVX2
1.110 +SplitUVRow a,
1.111 +
1.112 +; void MergeUVRow_SSE2(const uint8* src_u, const uint8* src_v, uint8* dst_uv,
1.113 +; int width);
1.114 +
1.115 +%macro MergeUVRow_ 1-2
1.116 +cglobal MergeUVRow_%2, 4, 4, 3, src_u, src_v, dst_uv, pix
1.117 + sub src_vq, src_uq
1.118 +
1.119 + ALIGN 4
1.120 +.convertloop:
1.121 + mov%1 m0, [src_uq]
1.122 + mov%1 m1, [src_vq]
1.123 + lea src_uq, [src_uq + mmsize]
1.124 + punpcklbw m2, m0, m1 // first 8 UV pairs
1.125 + punpckhbw m0, m0, m1 // next 8 UV pairs
1.126 +%if cpuflag(AVX2)
1.127 + vperm2i128 m1, m2, m0, 0x20 // low 128 of ymm2 and low 128 of ymm0
1.128 + vperm2i128 m2, m2, m0, 0x31 // high 128 of ymm2 and high 128 of ymm0
1.129 + mov%1 [dst_uvq], m1
1.130 + mov%1 [dst_uvq + mmsize], m2
1.131 +%else
1.132 + mov%1 [dst_uvq], m2
1.133 + mov%1 [dst_uvq + mmsize], m0
1.134 +%endif
1.135 + lea dst_uvq, [dst_uvq + mmsize * 2]
1.136 + sub pixd, mmsize
1.137 + jg .convertloop
1.138 + REP_RET
1.139 +%endmacro
1.140 +
1.141 +INIT_MMX MMX
1.142 +MergeUVRow_ a,
1.143 +MergeUVRow_ u,_Unaligned
1.144 +INIT_XMM SSE2
1.145 +MergeUVRow_ a,
1.146 +MergeUVRow_ u,_Unaligned
1.147 +INIT_YMM AVX2
1.148 +MergeUVRow_ a,
1.149 +
