Skip to content

Commit 1531619

Browse files
mbouronubitux
mbouron
authored and
ubitux
committed
swscale: add a neon version for the nv12 to rgba conversion v2
1 parent f6e5b17 commit 1531619

File tree

4 files changed

+240
-3
lines changed

4 files changed

+240
-3
lines changed

‎libswscale/arm/Makefile‎

Lines changed: 2 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -1,4 +1,5 @@
1-
# OBJS += arm/swscale_unscaled.o
1+
OBJS += arm/swscale_unscaled.o
22

33
# NEON-OBJS += arm/rgb2yuv_neon_32.o
44
# NEON-OBJS += arm/rgb2yuv_neon_16.o
5+
NEON-OBJS += arm/yuv2rgb_neon.o

‎libswscale/arm/swscale_unscaled.c‎

Lines changed: 70 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -23,6 +23,7 @@
2323
#include "libswscale/swscale_internal.h"
2424
#include "libavutil/arm/cpu.h"
2525

26+
#if 0
2627
extern void rgbx_to_nv12_neon_32(const uint8_t *src, uint8_t *y, uint8_t *chroma,
2728
int width, int height,
2829
int y_stride, int c_stride, int src_stride,
@@ -60,15 +61,84 @@ static int rgbx_to_nv12_neon_16_wrapper(SwsContext *context, const uint8_t *src[
6061

6162
return 0;
6263
}
64+
#endif
65+
66+
#define DECLARE_FF_NVX_TO_RGBX_FUNCS(ifmt, ofmt) \
67+
int ff_##ifmt##_to_##ofmt##_neon(int w, int h, \
68+
uint8_t *dst, int linesize, \
69+
const uint8_t *srcY, int linesizeY, \
70+
const uint8_t *srcC, int linesizeC, \
71+
const int16_t *table, \
72+
int y_offset, \
73+
int y_coeff); \
74+
\
75+
static int ifmt##_to_##ofmt##_neon_wrapper(SwsContext *c, const uint8_t *src[], \
76+
int srcStride[], int srcSliceY, int srcSliceH, \
77+
uint8_t *dst[], int dstStride[]) { \
78+
const int16_t yuv2rgb_table[] = { \
79+
c->yuv2rgb_v2r_coeff, \
80+
c->yuv2rgb_u2g_coeff, \
81+
c->yuv2rgb_v2g_coeff, \
82+
c->yuv2rgb_u2b_coeff, \
83+
}; \
84+
\
85+
ff_##ifmt##_to_##ofmt##_neon(c->srcW, srcSliceH, \
86+
dst[0] + srcSliceY * dstStride[0], dstStride[0], \
87+
src[0] + srcSliceY * srcStride[0], srcStride[0], \
88+
src[1] + (srcSliceY / 2) * srcStride[1], \
89+
srcStride[1], \
90+
yuv2rgb_table, \
91+
c->yuv2rgb_y_offset, \
92+
c->yuv2rgb_y_coeff >> 9); \
93+
\
94+
return 0; \
95+
}
96+
97+
#define DECLARE_FF_NVX_TO_ALL_RGBX_FUNCS(nvx) \
98+
DECLARE_FF_NVX_TO_RGBX_FUNCS(nvx, argb) \
99+
DECLARE_FF_NVX_TO_RGBX_FUNCS(nvx, rgba) \
100+
DECLARE_FF_NVX_TO_RGBX_FUNCS(nvx, abgr) \
101+
DECLARE_FF_NVX_TO_RGBX_FUNCS(nvx, bgra) \
102+
DECLARE_FF_NVX_TO_RGBX_FUNCS(nvx, 0rgb) \
103+
DECLARE_FF_NVX_TO_RGBX_FUNCS(nvx, rgb0) \
104+
DECLARE_FF_NVX_TO_RGBX_FUNCS(nvx, 0bgr) \
105+
DECLARE_FF_NVX_TO_RGBX_FUNCS(nvx, bgr0) \
106+
107+
DECLARE_FF_NVX_TO_ALL_RGBX_FUNCS(nv12)
108+
DECLARE_FF_NVX_TO_ALL_RGBX_FUNCS(nv21)
109+
110+
#define SET_FF_NVX_TO_RGBX_FUNC(ifmt, IFMT, ofmt, OFMT) do { \
111+
if (c->srcFormat == AV_PIX_FMT_##IFMT \
112+
&& c->dstFormat == AV_PIX_FMT_##OFMT \
113+
&& !(c->srcH & 1)) { \
114+
c->swscale = ifmt##_to_##ofmt##_neon_wrapper; \
115+
} \
116+
} while (0)
117+
118+
#define SET_FF_NVX_TO_ALL_RGBX_FUNC(nvx, NVX) do { \
119+
SET_FF_NVX_TO_RGBX_FUNC(nvx, NVX, argb, ARGB); \
120+
SET_FF_NVX_TO_RGBX_FUNC(nvx, NVX, rgba, RGBA); \
121+
SET_FF_NVX_TO_RGBX_FUNC(nvx, NVX, abgr, ABGR); \
122+
SET_FF_NVX_TO_RGBX_FUNC(nvx, NVX, bgra, BGRA); \
123+
SET_FF_NVX_TO_RGBX_FUNC(nvx, NVX, 0rgb, 0RGB); \
124+
SET_FF_NVX_TO_RGBX_FUNC(nvx, NVX, rgb0, RGB0); \
125+
SET_FF_NVX_TO_RGBX_FUNC(nvx, NVX, 0bgr, 0BGR); \
126+
SET_FF_NVX_TO_RGBX_FUNC(nvx, NVX, bgr0, BGR0); \
127+
} while (0)
63128

64129
static void get_unscaled_swscale_neon(SwsContext *c) {
130+
#if 0
65131
int accurate_rnd = c->flags & SWS_ACCURATE_RND;
66132
if (c->srcFormat == AV_PIX_FMT_RGBA
67133
&& c->dstFormat == AV_PIX_FMT_NV12
68134
&& (c->srcW >= 16)) {
69135
c->swscale = accurate_rnd ? rgbx_to_nv12_neon_32_wrapper
70136
: rgbx_to_nv12_neon_16_wrapper;
71137
}
138+
#endif
139+
140+
SET_FF_NVX_TO_ALL_RGBX_FUNC(nv12, NV12);
141+
SET_FF_NVX_TO_ALL_RGBX_FUNC(nv21, NV21);
72142
}
73143

74144
void ff_get_unscaled_swscale_arm(SwsContext *c)

‎libswscale/arm/yuv2rgb_neon.S‎

Lines changed: 166 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,166 @@
1+
/*
2+
* Copyright (c) 2015 Stupéflix
3+
*
4+
* This file is part of FFmpeg.
5+
*
6+
* FFmpeg is free software; you can redistribute it and/or
7+
* modify it under the terms of the GNU Lesser General Public
8+
* License as published by the Free Software Foundation; either
9+
* version 2.1 of the License, or (at your option) any later version.
10+
*
11+
* FFmpeg is distributed in the hope that it will be useful,
12+
* but WITHOUT ANY WARRANTY; without even the implied warranty of
13+
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14+
* Lesser General Public License for more details.
15+
*
16+
* You should have received a copy of the GNU Lesser General Public
17+
* License along with FFmpeg; if not, write to the Free Software
18+
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
19+
*/
20+
21+
#include "libavutil/arm/asm.S"
22+
23+
.macro compute_premult half_u half_v
24+
vmov d2, \half_u @ copy left q14 to left q1
25+
vmov d3, \half_u @ copy left q14 to right q1
26+
vmov d4, \half_v @ copy left q15 to left q2
27+
vmov d5, \half_v @ copy left q15 to right q2
28+
29+
vzip.16 d2, d3 @ U1U1U2U2U3U3U4U4
30+
vzip.16 d4, d5 @ V1V1V2V2V3V3V4V4
31+
32+
vmull.s16 q8, d4, d1[0] @ V * v2r (left, red)
33+
vmull.s16 q9, d5, d1[0] @ V * v2r (right, red)
34+
vmull.s16 q10, d2, d1[1] @ U * u2g
35+
vmull.s16 q11, d3, d1[1] @ U * u2g
36+
vmlal.s16 q10, d4, d1[2] @ U * u2g + V * v2g (left, green)
37+
vmlal.s16 q11, d5, d1[2] @ U * u2g + V * v2g (right, green)
38+
vmull.s16 q12, d2, d1[3] @ U * u2b (left, blue)
39+
vmull.s16 q13, d3, d1[3] @ U * u2b (right, blue)
40+
.endm
41+
42+
.macro compute_color dst_comp pre1 pre2
43+
vadd.s32 q3, q1, \pre1
44+
vadd.s32 q4, q2, \pre2
45+
vqrshrun.s32 d10, q3, #13
46+
vqrshrun.s32 d11, q4, #13 @ q5 = ({q3,q4} + (1<<12)) >> 13
47+
vqmovn.u16 \dst_comp, q5 @ saturate 16bit -> 8bit
48+
.endm
49+
50+
.macro compute_rgbx a b c d alpha
51+
compute_color \a, q8, q9 @ r
52+
compute_color \b, q10, q11 @ g
53+
compute_color \c, q12, q13 @ b
54+
vmov.u8 \d, #\alpha @ x
55+
.endm
56+
57+
.macro compute_half_line dst half_y rgb alpha
58+
vmovl.u8 q7, \half_y @ 8px of Y
59+
vdup.16 q5, r9
60+
@vsub.s16 q7, q5
61+
vmull.s16 q1, d14, d0 @ q1 = (srcY - y_offset) * y_coeff (left)
62+
vmull.s16 q2, d15, d0 @ q2 = (srcY - y_offset) * y_coeff (right)
63+
64+
.ifc \rgb,xrgb
65+
compute_rgbx d15, d12, d13, d14, \alpha
66+
.endif
67+
68+
.ifc \rgb,rgbx
69+
compute_rgbx d12, d13, d14, d15, \alpha
70+
.endif
71+
72+
.ifc \rgb,xbgr
73+
compute_rgbx d15, d14, d13, d12, \alpha
74+
.endif
75+
76+
.ifc \rgb,bgrx
77+
compute_rgbx d14, d13, d12, d15, \alpha
78+
.endif
79+
80+
vst4.u8 {d12, d13, d14, d15}, [\dst]!
81+
.endm
82+
83+
.macro declare_func ifmt ofmt rgb alpha
84+
function ff_\ifmt\()_to_\ofmt\()_neon, export=1
85+
push {r4-r12, lr}
86+
vpush {q4-q7}
87+
ldr r4, [sp, #104] @ r4 = srcY
88+
ldr r5, [sp, #108] @ r5 = linesizeY
89+
ldr r6, [sp, #112] @ r6 = srcC
90+
ldr r7, [sp, #116] @ r7 = linesizeC
91+
ldr r8, [sp, #120] @ r8 = table
92+
ldr r9, [sp, #124] @ r9 = y_offset
93+
ldr r10,[sp, #128] @ r10 = y_coeff
94+
vdup.16 d0, r10 @ d0 = y_coeff
95+
vld1.16 {d1}, [r8] @ d1 = *table
96+
add r11, r2, r3 @ r11 = dst + linesize (dst2)
97+
add r12, r4, r5 @ r12 = srcY + linesizeY (srcY2)
98+
lsl r3, r3, #1
99+
lsl r5, r5, #1
100+
lsl r8, r0, #2
101+
sub r3, r3, r8 @ r3 = linesize * 2 - width * 4 (padding)
102+
sub r5, r5, r0 @ r5 = linesizeY * 2 - width (paddingY)
103+
sub r7, r7, r0 @ r7 = linesizeC * 2 - width (paddingC)
104+
1:
105+
mov r8, r0 @ r8 = width
106+
2:
107+
pld [r6, #64*3]
108+
pld [r4, #64*3]
109+
pld [r12, #64*3]
110+
111+
vld2.8 {d2, d3}, [r6]! @ q1: interleaved chroma line
112+
vmov.i8 d10, #128
113+
.ifc \ifmt,nv12
114+
vsubl.u8 q14, d2, d10 @ q14 = U - 128
115+
vsubl.u8 q15, d3, d10 @ q15 = V - 128
116+
.else
117+
vsubl.u8 q14, d3, d10 @ q14 = U - 128
118+
vsubl.u8 q15, d2, d10 @ q15 = V - 128
119+
.endif
120+
121+
compute_premult d28 d30
122+
123+
vld1.8 {q7}, [r4]! @ first line of luma
124+
vmov d28, d15 @ save right of the first line of luma for later use
125+
compute_half_line r2, d14, \rgb, \alpha
126+
127+
vld1.8 {q7}, [r12]! @ second line of luma
128+
vmov d30, d15 @ save right of the second line of luma for later use
129+
compute_half_line r11, d14, \rgb, \alpha
130+
131+
compute_premult d29, d31
132+
compute_half_line r2, d28, \rgb, \alpha
133+
compute_half_line r11, d30, \rgb, \alpha
134+
135+
subs r8, r8, #16 @ width -= 16
136+
bgt 2b
137+
138+
add r2, r2, r3 @ dst += padding
139+
add r11, r11, r3 @ dst2 += padding
140+
add r4, r4, r5 @ srcY += paddingY
141+
add r6, r6, r7 @ srcC += paddingC
142+
add r12, r12, r5 @ srcY2 += paddingY
143+
144+
subs r1, r1, #2 @ height -= 2
145+
bgt 1b
146+
147+
vpop {q4-q7}
148+
pop {r4-r12, lr}
149+
mov pc, lr
150+
endfunc
151+
.endm
152+
153+
.macro declare_rgb_funcs ifmt
154+
declare_func \ifmt, argb, xrgb, 255
155+
declare_func \ifmt, rgba, rgbx, 255
156+
declare_func \ifmt, abgr, xbgr, 255
157+
declare_func \ifmt, bgra, bgrx, 255
158+
159+
declare_func \ifmt, 0rgb, xrgb, 0
160+
declare_func \ifmt, rgb0, rgbx, 0
161+
declare_func \ifmt, 0bgr, xbgr, 0
162+
declare_func \ifmt, bgr0, bgrx, 0
163+
.endm
164+
165+
declare_rgb_funcs nv12
166+
declare_rgb_funcs nv21

‎libswscale/swscale_unscaled.c‎

Lines changed: 2 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -1762,8 +1762,8 @@ void ff_get_unscaled_swscale(SwsContext *c)
17621762

17631763
if (ARCH_PPC)
17641764
ff_get_unscaled_swscale_ppc(c);
1765-
// if (ARCH_ARM)
1766-
// ff_get_unscaled_swscale_arm(c);
1765+
if (ARCH_ARM)
1766+
ff_get_unscaled_swscale_arm(c);
17671767
}
17681768

17691769
/* Convert the palette to the same packed 32-bit format as the palette */

0 commit comments

Comments
 (0)