1; 2; Copyright (c) 2015 The WebM project authors. All Rights Reserved. 3; 4; Use of this source code is governed by a BSD-style license 5; that can be found in the LICENSE file in the root of the source 6; tree. An additional intellectual property rights grant can be found 7; in the file PATENTS. All contributing project authors may 8; be found in the AUTHORS file in the root of the source tree. 9; 10 11%include "third_party/x86inc/x86inc.asm" 12 13SECTION .text 14 15%macro REORDER_INPUTS 0 16 ; a c d b to a b c d 17 SWAP 1, 3, 2 18%endmacro 19 20%macro TRANSFORM_COLS 0 21 ; input: 22 ; m0 a 23 ; m1 b 24 ; m2 c 25 ; m3 d 26 paddw m0, m2 27 psubw m3, m1 28 29 ; wide subtract 30 punpcklwd m4, m0 31 punpcklwd m5, m3 32 psrad m4, 16 33 psrad m5, 16 34 psubd m4, m5 35 psrad m4, 1 36 packssdw m4, m4 ; e 37 38 psubw m5, m4, m1 ; b 39 psubw m4, m2 ; c 40 psubw m0, m5 41 paddw m3, m4 42 ; m0 a 43 SWAP 1, 5 ; m1 b 44 SWAP 2, 4 ; m2 c 45 ; m3 d 46%endmacro 47 48%macro TRANSPOSE_4X4 0 49 punpcklwd m0, m2 50 punpcklwd m1, m3 51 mova m2, m0 52 punpcklwd m0, m1 53 punpckhwd m2, m1 54 pshufd m1, m0, 0x0e 55 pshufd m3, m2, 0x0e 56%endmacro 57 58; transpose a 4x4 int16 matrix in xmm0 and xmm1 to the bottom half of xmm0-xmm3 59%macro TRANSPOSE_4X4_WIDE 0 60 mova m3, m0 61 punpcklwd m0, m1 62 punpckhwd m3, m1 63 mova m2, m0 64 punpcklwd m0, m3 65 punpckhwd m2, m3 66 pshufd m1, m0, 0x0e 67 pshufd m3, m2, 0x0e 68%endmacro 69 70%macro ADD_STORE_4P_2X 5 ; src1, src2, tmp1, tmp2, zero 71 movd m%3, [outputq] 72 movd m%4, [outputq + strideq] 73 punpcklbw m%3, m%5 74 punpcklbw m%4, m%5 75 paddw m%1, m%3 76 paddw m%2, m%4 77 packuswb m%1, m%5 78 packuswb m%2, m%5 79 movd [outputq], m%1 80 movd [outputq + strideq], m%2 81%endmacro 82 83INIT_XMM sse2 84cglobal iwht4x4_16_add, 3, 3, 7, input, output, stride 85 mova m0, [inputq + 0] 86 mova m1, [inputq + 16] 87 88 psraw m0, 2 89 psraw m1, 2 90 91 TRANSPOSE_4X4_WIDE 92 REORDER_INPUTS 93 TRANSFORM_COLS 94 TRANSPOSE_4X4 95 REORDER_INPUTS 96 TRANSFORM_COLS 97 98 pxor m4, m4 99 ADD_STORE_4P_2X 0, 1, 5, 6, 4 100 lea outputq, [outputq + 2 * strideq] 101 ADD_STORE_4P_2X 2, 3, 5, 6, 4 102 103 RET 104