error_sse2.asm (2505B)
1 ; 2 ; Copyright (c) 2016, Alliance for Open Media. All rights reserved. 3 ; 4 ; This source code is subject to the terms of the BSD 2 Clause License and 5 ; the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License 6 ; was not distributed with this source code in the LICENSE file, you can 7 ; obtain it at www.aomedia.org/license/software. If the Alliance for Open 8 ; Media Patent License 1.0 was not distributed with this source code in the 9 ; PATENTS file, you can obtain it at www.aomedia.org/license/patent. 10 ; 11 12 ; 13 14 ; Increment %1 by sizeof() tran_low_t * %2. 15 %macro INCREMENT_ELEMENTS_TRAN_LOW 2 16 lea %1, [%1 + %2 * 4] 17 %endmacro 18 19 ; Load %2 + %3 into m%1. 20 ; %3 is the offset in elements, not bytes. 21 ; If tran_low_t is 16 bits (low bit depth configuration) then load the value 22 ; directly. If tran_low_t is 32 bits (high bit depth configuration) then pack 23 ; the values down to 16 bits. 24 %macro LOAD_TRAN_LOW 3 25 mova m%1, [%2 + (%3) * 4] 26 packssdw m%1, [%2 + (%3) * 4 + 16] 27 %endmacro 28 29 %define private_prefix av1 30 31 %include "third_party/x86inc/x86inc.asm" 32 33 SECTION .text 34 35 ; int64_t av1_block_error(int16_t *coeff, int16_t *dqcoeff, intptr_t block_size, 36 ; int64_t *ssz) 37 38 INIT_XMM sse2 39 cglobal block_error, 3, 3, 8, uqc, dqc, size, ssz 40 pxor m4, m4 ; sse accumulator 41 pxor m6, m6 ; ssz accumulator 42 pxor m5, m5 ; dedicated zero register 43 .loop: 44 LOAD_TRAN_LOW 2, uqcq, 0 45 LOAD_TRAN_LOW 0, dqcq, 0 46 LOAD_TRAN_LOW 3, uqcq, 8 47 LOAD_TRAN_LOW 1, dqcq, 8 48 INCREMENT_ELEMENTS_TRAN_LOW uqcq, 16 49 INCREMENT_ELEMENTS_TRAN_LOW dqcq, 16 50 sub sizeq, 16 51 psubw m0, m2 52 psubw m1, m3 53 ; individual errors are max. 15bit+sign, so squares are 30bit, and 54 ; thus the sum of 2 should fit in a 31bit integer (+ unused sign bit) 55 pmaddwd m0, m0 56 pmaddwd m1, m1 57 pmaddwd m2, m2 58 pmaddwd m3, m3 59 ; the sum of 2 31bit integers will fit in a 32bit unsigned integer 60 paddd m0, m1 61 paddd m2, m3 62 ; accumulate in 64bit 63 punpckldq m7, m0, m5 64 punpckhdq m0, m5 65 paddq m4, m7 66 punpckldq m7, m2, m5 67 paddq m4, m0 68 punpckhdq m2, m5 69 paddq m6, m7 70 paddq m6, m2 71 jg .loop 72 73 ; accumulate horizontally and store in return value 74 movhlps m5, m4 75 movhlps m7, m6 76 paddq m4, m5 77 paddq m6, m7 78 %if AOM_ARCH_X86_64 79 movq rax, m4 80 movq [sszq], m6 81 %else 82 mov eax, sszm 83 pshufd m5, m4, 0x1 84 movq [eax], m6 85 movd eax, m4 86 movd edx, m5 87 %endif 88 RET