From d8426d6f121e0b2b1e634e61104ccfed05cd99f8 Mon Sep 17 00:00:00 2001 From: Alex Converse Date: Mon, 13 Jul 2015 11:12:45 -0700 Subject: Add an SSE2 version of vp9_iwht4x4_16_add Roughly half as many cycles as plain C. Change-Id: I8c16c29940b76d54ee7e4fb874c328ce90bff5d4 --- vp9/common/vp9_rtcd_defs.pl | 2 +- vp9/common/x86/vp9_idct_sse2.asm | 102 +++++++++++++++++++++++++++++++++++++++ vp9/vp9_common.mk | 1 + 3 files changed, 104 insertions(+), 1 deletion(-) create mode 100644 vp9/common/x86/vp9_idct_sse2.asm (limited to 'vp9') diff --git a/vp9/common/vp9_rtcd_defs.pl b/vp9/common/vp9_rtcd_defs.pl index 524e79f0b..538f1ed7e 100644 --- a/vp9/common/vp9_rtcd_defs.pl +++ b/vp9/common/vp9_rtcd_defs.pl @@ -475,7 +475,7 @@ if (vpx_config("CONFIG_VP9_HIGHBITDEPTH") eq "yes") { specialize qw/vp9_iwht4x4_1_add msa/; add_proto qw/void vp9_iwht4x4_16_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride"; - specialize qw/vp9_iwht4x4_16_add msa/; + specialize qw/vp9_iwht4x4_16_add msa/, "$sse2_x86inc"; } } diff --git a/vp9/common/x86/vp9_idct_sse2.asm b/vp9/common/x86/vp9_idct_sse2.asm new file mode 100644 index 000000000..69b68e6d8 --- /dev/null +++ b/vp9/common/x86/vp9_idct_sse2.asm @@ -0,0 +1,102 @@ +; +; Copyright (c) 2015 The WebM project authors. All Rights Reserved. +; +; Use of this source code is governed by a BSD-style license +; that can be found in the LICENSE file in the root of the source +; tree. An additional intellectual property rights grant can be found +; in the file PATENTS. All contributing project authors may +; be found in the AUTHORS file in the root of the source tree. +; +%include "third_party/x86inc/x86inc.asm" + +SECTION .text + +%macro REORDER_INPUTS 0 + ; a c d b to a b c d + SWAP 1, 3, 2 +%endmacro + +%macro TRANSFORM_COLS 0 + ; input: + ; m0 a + ; m1 b + ; m2 c + ; m3 d + paddw m0, m2 + psubw m3, m1 + + ; wide subtract + punpcklwd m4, m0 + punpcklwd m5, m3 + psrad m4, 16 + psrad m5, 16 + psubd m4, m5 + psrad m4, 1 + packssdw m4, m4 ; e + + psubw m5, m4, m1 ; b + psubw m4, m2 ; c + psubw m0, m5 + paddw m3, m4 + ; m0 a + SWAP 1, 5 ; m1 b + SWAP 2, 4 ; m2 c + ; m3 d +%endmacro + +%macro TRANSPOSE_4X4 0 + punpcklwd m0, m2 + punpcklwd m1, m3 + mova m2, m0 + punpcklwd m0, m1 + punpckhwd m2, m1 + pshufd m1, m0, 0x0e + pshufd m3, m2, 0x0e +%endmacro + +; transpose a 4x4 int16 matrix in xmm0 and xmm1 to the bottom half of xmm0-xmm3 +%macro TRANSPOSE_4X4_WIDE 0 + mova m3, m0 + punpcklwd m0, m1 + punpckhwd m3, m1 + mova m2, m0 + punpcklwd m0, m3 + punpckhwd m2, m3 + pshufd m1, m0, 0x0e + pshufd m3, m2, 0x0e +%endmacro + +%macro ADD_STORE_4P_2X 5 ; src1, src2, tmp1, tmp2, zero + movq m%3, [outputq] + movq m%4, [outputq + strideq] + punpcklbw m%3, m%5 + punpcklbw m%4, m%5 + paddw m%1, m%3 + paddw m%2, m%4 + packuswb m%1, m%5 + packuswb m%2, m%5 + movd [outputq], m%1 + movd [outputq + strideq], m%2 +%endmacro + +INIT_XMM sse2 +cglobal iwht4x4_16_add, 3, 3, 7, input, output, stride + mova m0, [inputq + 0] + mova m1, [inputq + 16] + + psraw m0, 2 + psraw m1, 2 + + TRANSPOSE_4X4_WIDE + REORDER_INPUTS + TRANSFORM_COLS + TRANSPOSE_4X4 + REORDER_INPUTS + TRANSFORM_COLS + + pxor m4, m4 + ADD_STORE_4P_2X 0, 1, 5, 6, 4 + lea outputq, [outputq + 2 * strideq] + ADD_STORE_4P_2X 2, 3, 5, 6, 4 + + RET diff --git a/vp9/vp9_common.mk b/vp9/vp9_common.mk index 2e43c272a..bcab5582c 100644 --- a/vp9/vp9_common.mk +++ b/vp9/vp9_common.mk @@ -75,6 +75,7 @@ VP9_COMMON_SRCS-$(CONFIG_VP9_POSTPROC) += common/vp9_postproc.h VP9_COMMON_SRCS-$(CONFIG_VP9_POSTPROC) += common/vp9_postproc.c VP9_COMMON_SRCS-$(CONFIG_VP9_POSTPROC) += common/vp9_mfqe.h VP9_COMMON_SRCS-$(CONFIG_VP9_POSTPROC) += common/vp9_mfqe.c +VP9_COMMON_SRCS-$(HAVE_SSE2) += common/x86/vp9_idct_sse2.asm VP9_COMMON_SRCS-$(HAVE_MMX) += common/x86/vp9_loopfilter_mmx.asm VP9_COMMON_SRCS-$(HAVE_SSE2) += common/x86/vp9_subpixel_8t_sse2.asm VP9_COMMON_SRCS-$(HAVE_SSE2) += common/x86/vp9_subpixel_bilinear_sse2.asm -- cgit v1.2.3