1.1 --- /dev/null Thu Jan 01 00:00:00 1970 +0000
1.2 +++ b/media/libvpx/vp9/encoder/x86/vp9_error_sse2.asm Wed Dec 31 06:09:35 2014 +0100
1.3 @@ -0,0 +1,74 @@
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 +; int64_t vp9_block_error(int16_t *coeff, int16_t *dqcoeff, intptr_t block_size,
1.19 +; int64_t *ssz)
1.20 +
1.21 +INIT_XMM sse2
1.22 +cglobal block_error, 3, 3, 8, uqc, dqc, size, ssz
1.23 + pxor m4, m4 ; sse accumulator
1.24 + pxor m6, m6 ; ssz accumulator
1.25 + pxor m5, m5 ; dedicated zero register
1.26 + lea uqcq, [uqcq+sizeq*2]
1.27 + lea dqcq, [dqcq+sizeq*2]
1.28 + neg sizeq
1.29 +.loop:
1.30 + mova m2, [uqcq+sizeq*2]
1.31 + mova m0, [dqcq+sizeq*2]
1.32 + mova m3, [uqcq+sizeq*2+mmsize]
1.33 + mova m1, [dqcq+sizeq*2+mmsize]
1.34 + psubw m0, m2
1.35 + psubw m1, m3
1.36 + ; individual errors are max. 15bit+sign, so squares are 30bit, and
1.37 + ; thus the sum of 2 should fit in a 31bit integer (+ unused sign bit)
1.38 + pmaddwd m0, m0
1.39 + pmaddwd m1, m1
1.40 + pmaddwd m2, m2
1.41 + pmaddwd m3, m3
1.42 + ; accumulate in 64bit
1.43 + punpckldq m7, m0, m5
1.44 + punpckhdq m0, m5
1.45 + paddq m4, m7
1.46 + punpckldq m7, m1, m5
1.47 + paddq m4, m0
1.48 + punpckhdq m1, m5
1.49 + paddq m4, m7
1.50 + punpckldq m7, m2, m5
1.51 + paddq m4, m1
1.52 + punpckhdq m2, m5
1.53 + paddq m6, m7
1.54 + punpckldq m7, m3, m5
1.55 + paddq m6, m2
1.56 + punpckhdq m3, m5
1.57 + paddq m6, m7
1.58 + paddq m6, m3
1.59 + add sizeq, mmsize
1.60 + jl .loop
1.61 +
1.62 + ; accumulate horizontally and store in return value
1.63 + movhlps m5, m4
1.64 + movhlps m7, m6
1.65 + paddq m4, m5
1.66 + paddq m6, m7
1.67 +%if ARCH_X86_64
1.68 + movq rax, m4
1.69 + movq [sszq], m6
1.70 +%else
1.71 + mov eax, sszm
1.72 + pshufd m5, m4, 0x1
1.73 + movq [eax], m6
1.74 + movd eax, m4
1.75 + movd edx, m5
1.76 +%endif
1.77 + RET
