GithubStar/QtAV
1
0
Fork 0
mirror of https://github.com/wang-bin/QtAV.git synced 2026-10-29 11:36:51 +08:00
Code Issues Packages Projects Releases Wiki Activity

cedarv: add yuv copy code(neon) from libvdpau-sunxi

Browse Source 
yv12 can not copy correctly. fallback to old one.
This commit is contained in:
wang-bin
2014-12-04 00:49:45 +08:00
parent bac2fd4b81
commit bf47cb3ad2
3 changed files with 241 additions and 46 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/codec/video/VideoDecoderCedarv.cpp
+58 -45
View File
@@ -30,7 +30,19 @@ extern "C" {
}
#include "utils/Logger.h"
// TODO: neon+nv12+opengl crash
// TODO: neon+nv12+opengl crash. tiled_deinterleave_to_planar only copys half-left of image
#ifndef NO_NEON_OPT //Don't HAVE_NEON
extern "C" {
// from libvdpau-sunxi
void tiled_to_planar(void *src, void *dst, unsigned int dst_pitch,
unsigned int width, unsigned int height);
void tiled_deinterleave_to_planar(void *src, void *dst1, void *dst2,
unsigned int dst_pitch,
unsigned int width, unsigned int height);
}
#endif //NO_NEON_OPT
namespace QtAV {
class VideoDecoderCedarvPrivate;
@@ -77,10 +89,10 @@ void RegisterVideoDecoderCedarv_Man()
}
// source data is colum major. every block is 32x32
static void map32x32_to_yuv_Y(unsigned char* srcY, unsigned char* tarY, unsigned int coded_width, unsigned int coded_height)
static void map32x32_to_yuv_Y(void* srcY, void* tarY, unsigned int dst_pitch, unsigned int coded_width, unsigned int coded_height)
{
unsigned long offset;
unsigned char *ptr = srcY;
unsigned char *ptr = (unsigned char *)srcY;
const unsigned int mb_width = (coded_width+15) >> 4;
const unsigned int mb_height = (coded_height+15) >> 4;
const unsigned int twomb_line = (mb_height+1) >> 1;
@@ -90,17 +102,17 @@ static void map32x32_to_yuv_Y(unsigned char* srcY, unsigned char* tarY, unsigned
const unsigned int M = 32*i;
for (unsigned int j = 0; j < recon_width; j+=2) {
const unsigned int n = j*16;
offset = M*coded_width + n;
offset = M*dst_pitch + n;
for (unsigned int l = 0; l < 32; l++) {
if (M+l < coded_height) {
if (n+16 < coded_width) {
//1st & 2nd mb
memcpy(tarY+offset, ptr, 32);
memcpy((unsigned char *)tarY+offset, ptr, 32);
} else if (n<coded_width) {
// 1st mb
memcpy(tarY+offset, ptr, 16);
memcpy((unsigned char *)tarY+offset, ptr, 16);
}
offset += coded_width;
offset += dst_pitch;
}
ptr += 32;
}
@@ -108,11 +120,11 @@ static void map32x32_to_yuv_Y(unsigned char* srcY, unsigned char* tarY, unsigned
}
}
static void map32x32_to_yuv_C(unsigned char* srcC,unsigned char* tarCb,unsigned char* tarCr,unsigned int coded_width,unsigned int coded_height)
static void map32x32_to_yuv_C(void* srcC, void* tarCb, void* tarCr, unsigned int dst_pitch, unsigned int coded_width, unsigned int coded_height)
{
unsigned char line[32];
unsigned long offset;
unsigned char *ptr = srcC;
unsigned char *ptr = (unsigned char *)srcC;
const unsigned int mb_width = (coded_width+7) >> 3;
const unsigned int mb_height = (coded_height+7) >> 3;
const unsigned int fourmb_line = (mb_height+3) >> 2;
@@ -122,7 +134,7 @@ static void map32x32_to_yuv_C(unsigned char* srcC,unsigned char* tarCb,unsigned
const int M = i*32;
for (unsigned int j = 0; j < recon_width; j+=2) {
const unsigned int n = j*8;
offset = M*coded_width + n;
offset = M*dst_pitch + n;
for (unsigned int l = 0; l < 32; l++) {
if (M+l < coded_height) {
if (n+8 < coded_width) {
@@ -130,19 +142,19 @@ static void map32x32_to_yuv_C(unsigned char* srcC,unsigned char* tarCb,unsigned
memcpy(line, ptr, 32);
//unsigned char *line = ptr;
for (int k = 0; k < 16; k++) {
*(tarCb + offset + k) = line[2*k];
*(tarCr + offset + k) = line[2*k+1];
*((unsigned char *)tarCb + offset + k) = line[2*k];
*((unsigned char *)tarCr + offset + k) = line[2*k+1];
}
} else if (n < coded_width) {
// 1st mb
memcpy(line, ptr, 16);
//unsigned char *line = ptr;
for (int k = 0; k < 8; k++) {
*(tarCb + offset + k) = line[2*k];
*(tarCr + offset + k) = line[2*k+1];
*((unsigned char *)tarCb + offset + k) = line[2*k];
*((unsigned char *)tarCr + offset + k) = line[2*k+1];
}
}
offset += coded_width;
offset += dst_pitch;
}
ptr += 32;
}
@@ -151,30 +163,29 @@ static void map32x32_to_yuv_C(unsigned char* srcC,unsigned char* tarCb,unsigned
}
#if 0
static void map32x32_to_nv12_UV(unsigned char* srcC,unsigned char* tarUV,unsigned int coded_width,unsigned int coded_height)
static void map32x32_to_nv12_UV(void* srcC, void* tarUV, unsigned int dst_pitch, unsigned int coded_width, unsigned int coded_height)
{
unsigned long offset;
unsigned char *ptr = srcC;
unsigned char *ptr = (unsigned char *)srcC;
const unsigned int mb_width = (coded_width+7) >> 3;
const unsigned int mb_height = (coded_height+7) >> 3;
const unsigned int fourmb_line = (mb_height+3) >> 2;
const unsigned int recon_width = (mb_width+1) & 0xfffffffe;
const unsigned int uv_width = 2*coded_width;
for (unsigned int i = 0; i < fourmb_line; i++) {
const int M = i*32;
for (unsigned int j = 0; j < recon_width; j+=2) {
const unsigned int n = j*16; // 1 logical pixel 2 channel
offset = M*uv_width + n;
offset = M*dst_pitch + n;
for (unsigned int l = 0; l < 32; l++) { // copy 16 logical pixels every time. 16/2*recon_width ~ coded_width
if (M+l < coded_height) {
if (n+8 < coded_width) {
// 1st & 2nd mb
memcpy(tarUV + offset, ptr, 32);
memcpy((unsigned char *)tarUV + offset, ptr, 32);
} else if (n < coded_width) {
// 1st mb
memcpy(tarUV + offset, ptr, 16);
memcpy((unsigned char *)tarUV + offset, ptr, 16);
}
offset += uv_width;
offset += dst_pitch;
}
ptr += 32;
}
@@ -183,12 +194,13 @@ static void map32x32_to_nv12_UV(unsigned char* srcC,unsigned char* tarUV,unsigne
}
#endif
#ifndef NO_NEON_OPT //Don't HAVE_NEON
static void map32x32_to_yuv_Y_neon(unsigned char* srcY,unsigned char* tarY,unsigned int coded_width,unsigned int coded_height)
// use tiled_to_planar instead
static void map32x32_to_yuv_Y_neon(void* srcY, void* tarY, unsigned int dst_pitch, unsigned int coded_width,unsigned int coded_height)
{
unsigned long offset;
unsigned char *dst_asm,*src_asm;
unsigned char *ptr = srcY;
unsigned char *ptr = (unsigned char *)srcY;
const unsigned int mb_width = (coded_width+15) >> 4;
const unsigned int mb_height = (coded_height+15) >> 4;
const unsigned int twomb_line = (mb_height+1) >> 1;
@@ -197,10 +209,10 @@ static void map32x32_to_yuv_Y_neon(unsigned char* srcY,unsigned char* tarY,unsig
const int M = i*32;
for (unsigned int j = 0; j < twomb_width; j++) {
const unsigned int n= j*32;
offset = M*coded_width + n;
offset = M*dst_pitch + n;
for (unsigned int l=0;l<32;l++) {
//first mb
dst_asm = tarY+offset;
dst_asm = (unsigned char *)tarY + offset;
src_asm = ptr;
asm volatile (
"vld1.8 {d0 - d3}, [%[src_asm]] \n\t"
@@ -209,7 +221,7 @@ static void map32x32_to_yuv_Y_neon(unsigned char* srcY,unsigned char* tarY,unsig
: //[srcY] "r" (srcY)
: "cc", "memory", "d0", "d1", "d2", "d3", "d4", "d5", "d6", "d16", "d17", "d18", "d19", "d20", "d21", "d22", "d23", "d24", "d28", "d29", "d30", "d31"
);
offset += coded_width;
offset += dst_pitch;
ptr += 32;
}
}
@@ -218,11 +230,11 @@ static void map32x32_to_yuv_Y_neon(unsigned char* srcY,unsigned char* tarY,unsig
if(mb_width & 1) {
unsigned int j = mb_width-1;
const unsigned int n = j*16;
offset = M*coded_width + n;
offset = M*dst_pitch + n;
for (unsigned int l = 0; l < 32; l++) {
//first mb
if (M+l<coded_height && n<coded_width) {
dst_asm = tarY + offset;
dst_asm = (unsigned char *)tarY + offset;
src_asm = ptr;
asm volatile (
"vld1.8 {d0 - d1}, [%[src_asm]] \n\t"
@@ -232,14 +244,14 @@ static void map32x32_to_yuv_Y_neon(unsigned char* srcY,unsigned char* tarY,unsig
: "cc", "memory", "d0", "d1", "d2", "d3", "d4", "d5", "d6", "d16", "d17", "d18", "d19", "d20", "d21", "d22", "d23", "d24", "d28", "d29", "d30", "d31"
);
}
offset += coded_width;
offset += dst_pitch;
ptr += 32;
}
}
}
}
static void map32x32_to_yuv_C_neon(unsigned char* srcC,unsigned char* tarCb,unsigned char* tarCr,unsigned int coded_width,unsigned int coded_height)
// use tiled_deinterleave_to_planar instead
static void map32x32_to_yuv_C_neon(void* srcC, void* tarCb, void* tarCr, unsigned int dst_pitch, unsigned int coded_width, unsigned int coded_height)
{
unsigned int j,l,k;
unsigned long offset;
@@ -247,7 +259,7 @@ static void map32x32_to_yuv_C_neon(unsigned char* srcC,unsigned char* tarCb,unsi
unsigned char line[16];
int dst_stride = FFALIGN(coded_width, 16);
unsigned char *ptr = srcC;
unsigned char *ptr = (unsigned char *)srcC;
const unsigned int mb_width = (coded_width+7)>>3;
const unsigned int mb_height = (coded_height+7)>>3;
const unsigned int fourmb_line = (mb_height+3)>>2;
@@ -261,8 +273,8 @@ static void map32x32_to_yuv_C_neon(unsigned char* srcC,unsigned char* tarCb,unsi
for (l = 0; l < 32; l++) {
//first mb
if (M+l<coded_height && n<coded_width) {
dst0_asm = tarCb + offset;
dst1_asm = tarCr + offset;
dst0_asm = (unsigned char *)tarCb + offset;
dst1_asm = (unsigned char *)tarCr + offset;
src_asm = ptr;
// for(k=0;k<16;k++)
// {
@@ -282,7 +294,7 @@ static void map32x32_to_yuv_C_neon(unsigned char* srcC,unsigned char* tarCb,unsi
: "cc", "memory", "d0", "d1", "d2", "d3", "d4", "d5", "d6", "d16", "d17", "d18", "d19", "d20", "d21", "d22", "d23", "d24", "d28", "d29", "d30", "d31"
);
}
offset += coded_width;
offset += dst_pitch;
ptr += 32;
}
}
@@ -294,8 +306,8 @@ static void map32x32_to_yuv_C_neon(unsigned char* srcC,unsigned char* tarCb,unsi
if (M+l<coded_height && n<coded_width) {
memcpy(line, ptr, 16);
for(k = 0; k < 8; k++) {
*(tarCb + offset + k) = line[2*k];
*(tarCr + offset + k) = line[2*k+1];
*((unsigned char *)tarCb + offset + k) = line[2*k];
*((unsigned char *)tarCr + offset + k) = line[2*k+1];
}
}
offset += dst_stride;
@@ -403,9 +415,9 @@ public:
CEDARV_DECODER *cedarv;
cedarv_picture_t cedarPicture;
typedef void (*map_y_t)(unsigned char*, unsigned char*, unsigned int, unsigned int);
typedef void (*map_y_t)(void* src, void* dst, unsigned int dst_pitch, unsigned int w, unsigned int h);
map_y_t map_y;
typedef void (*map_c_t)(unsigned char*,unsigned char*, unsigned char*, unsigned int,unsigned int);
typedef void (*map_c_t)(void* src, void* dst1, void* dst2, unsigned int dst_pitch, unsigned int w, unsigned int h);
map_c_t map_c;
VideoDecoderCedarv::PixFmt pixfmt;
};
@@ -427,8 +439,9 @@ void VideoDecoderCedarv::setNeon(bool value)
DPTR_D(VideoDecoderCedarv);
if (value) {
#ifndef NO_NEON_OPT //Don't HAVE_NEON
d.map_y = map32x32_to_yuv_Y_neon;
d.map_c = map32x32_to_yuv_C_neon;
d.map_y = tiled_to_planar;// map32x32_to_yuv_Y_neon;
// tiled_deinterleave_to_planar now only copys half left of picture
d.map_c = map32x32_to_yuv_C_neon; //tiled_deinterleave_to_planar;// map32x32_to_yuv_C_neon;
#endif
} else {
d.map_y = map32x32_to_yuv_Y;
@@ -620,11 +633,11 @@ VideoFrame VideoDecoderCedarv::frame()
if (nv12)
pitch[1] = dst_y_stride;
d.map_y(d.cedarPicture.y, plane[0], display_w_align, display_h_align);
d.map_y(d.cedarPicture.y, plane[0], display_w_align, pitch[0], display_h_align);
if (nv12)
d.map_y(d.cedarPicture.u, plane[1], display_w_align, display_h_align/2);
d.map_y(d.cedarPicture.u, plane[1], pitch[1], display_w_align, display_h_align/2);
else
d.map_c(d.cedarPicture.u, plane[1], plane[2], display_w_align/2, display_h_align/2);
d.map_c(d.cedarPicture.u, plane[1], plane[2], pitch[1], display_w_align/2, display_h_align/2);
const VideoFormat fmt(nv12 ? VideoFormat::Format_NV12 : VideoFormat::Format_YUV420P);
VideoFrame frame(buf, display_w_align, display_h_align, fmt);
src/codec/video/tiled_yuv.S
+175
View File
@@ -0,0 +1,175 @@
/*
* Copyright (c) 2014 Jens Kuske <jenskuske@gmail.com>
*
* This library is free software; you can redistribute it and/or
* modify it under the terms of the GNU Lesser General Public
* License as published by the Free Software Foundation; either
* version 2.1 of the License, or (at your option) any later version.
*
* This library is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
* Lesser General Public License for more details.
*
* You should have received a copy of the GNU Lesser General Public
* License along with this library; if not, write to the Free Software
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
*
*/
#if defined(__linux__) && defined(__ELF__)
.section .note.GNU-stack,"",%progbits /* mark stack as non-executable */
#endif
.text
.syntax unified
.arch armv7-a
.fpu neon
.thumb
.macro thumb_function fname
.global \fname
#ifdef __ELF__
.hidden \fname
.type \fname, %function
#endif
.thumb_func
\fname:
.endm
.macro end_function fname
#ifdef __ELF__
.size \fname, .-\fname
#endif
.endm
SRC .req r0
DST .req r1
PITCH .req r2
CNT .req r3
TLINE .req r4
HEIGHT .req r5
REST .req r6
NTILES .req r7
TMPSRC .req r8
DST2 .req r9
TSIZE .req r12
NEXTLIN .req lr
thumb_function tiled_to_planar
push {r4, r5, r6, r7, r8, lr}
ldr HEIGHT, [sp, #24]
add NEXTLIN, r3, #31
lsrs NTILES, r3, #5
bic NEXTLIN, NEXTLIN, #31
and REST, r3, #31
lsl NEXTLIN, NEXTLIN, #5
subs PITCH, r2, r3
movs TLINE, #32
rsb NEXTLIN, NEXTLIN, #32
mov TSIZE, #1024
/* y loop */
1: cbz NTILES, 3f
mov CNT, NTILES
/* x loop complete tiles */
2: pld [SRC, TSIZE]
vld1.8 {d0 - d3}, [SRC :256], TSIZE
subs CNT, #1
vst1.8 {d0 - d3}, [DST]!
bne 2b
3: cbnz REST, 4f
/* fix up dest pointer if pitch != width */
7: add DST, PITCH
/* fix up src pointer at end of line */
subs TLINE, #1
itee ne
addne SRC, NEXTLIN
subeq SRC, #992
moveq TLINE, #32
subs HEIGHT, #1
bne 1b
pop {r4, r5, r6, r7, r8, pc}
/* partly copy last tile of line */
4: mov TMPSRC, SRC
tst REST, #16
beq 5f
vld1.8 {d0 - d1}, [TMPSRC :128]!
vst1.8 {d0 - d1}, [DST]!
5: add SRC, TSIZE
ands CNT, REST, #15
beq 7b
6: vld1.8 {d0[0]}, [TMPSRC]!
subs CNT, #1
vst1.8 {d0[0]}, [DST]!
bne 6b
b 7b
end_function tiled_to_planar
thumb_function tiled_deinterleave_to_planar
push {r4, r5, r6, r7, r8, r9, lr}
mov DST2, r2
ldr HEIGHT, [sp, #32]
ldr r4, [sp, #28]
add NEXTLIN, r4, #31
lsrs NTILES, r4, #5
bic NEXTLIN, NEXTLIN, #31
ubfx REST, r4, #1, #4
lsl NEXTLIN, NEXTLIN, #5
sub PITCH, r3, r4, lsr #1
movs TLINE, #32
rsb NEXTLIN, NEXTLIN, #32
mov TSIZE, #1024
/* y loop */
1: cbz NTILES, 3f
mov CNT, NTILES
/* x loop complete tiles */
2: pld [SRC, TSIZE]
vld2.8 {d0 - d3}, [SRC :256], TSIZE
subs CNT, #1
vst1.8 {d0 - d1}, [DST]!
vst1.8 {d2 - d3}, [DST2]!
bne 2b
3: cbnz REST, 4f
/* fix up dest pointer if pitch != width */
7: add DST, PITCH
add DST2, PITCH
/* fix up src pointer at end of line */
subs TLINE, #1
itee ne
addne SRC, NEXTLIN
subeq SRC, #992
moveq TLINE, #32
subs HEIGHT, #1
bne 1b
pop {r4, r5, r6, r7, r8, r9, pc}
/* partly copy last tile of line */
4: mov TMPSRC, SRC
tst REST, #8
beq 5f
vld2.8 {d0 - d1}, [TMPSRC :128]!
vst1.8 {d0}, [DST]!
vst1.8 {d1}, [DST2]!
5: add SRC, TSIZE
ands CNT, REST, #7
beq 7b
6: vld2.8 {d0[0], d1[0]}, [TMPSRC]!
subs CNT, #1
vst1.8 {d0[0]}, [DST]!
vst1.8 {d1[0]}, [DST2]!
bne 6b
b 7b
end_function tiled_deinterleave_to_planar
src/libQtAV.pro
+8 -1
View File
@@ -2,6 +2,8 @@ TEMPLATE = lib
MODULE_INCNAME = QtAV # for mac framework. also used in install_sdk.pro
TARGET = QtAV
QT += core gui
config_libcedarv: CONFIG += neon #need by qt4 addSimdCompiler()
greaterThan(QT_MAJOR_VERSION, 4) {
qtHaveModule(widgets):!no_widgets {
QT += widgets
@@ -219,9 +221,14 @@ config_vaapi* {
}
config_libcedarv {
DEFINES *= QTAV_HAVE_CEDARV=1
QMAKE_CXXFLAGS *= -march=armv7-a -mfpu=neon
QMAKE_CXXFLAGS *= -march=armv7-a
neon: QMAKE_CXXFLAGS *= $$QMAKE_CFLAGS_NEON
SOURCES += codec/video/VideoDecoderCedarv.cpp
CONFIG += simd #addSimdCompiler xxx_ASM
CONFIG += no_clang_integrated_as #see qtbase/src/gui/painting/painting.pri. add -fno-integrated-as from simd.prf
NEON_ASM += codec/video/tiled_yuv.S #from libvdpau-sunxi
LIBS += -lvecore -lcedarv
OTHER_FILES += $$NEON_ASM
}
macx:!ios: CONFIG += config_vda
config_vda {