Skip to content

Commit b32a422

Browse files
mbouronmbouron
committed
swscale/arm/yuv2rgb: add ff_yuv420p_to_{argb,rgba,abgr,bgra}_neon_{16,32}
1 parent e0dc22b commit b32a422

File tree

2 files changed

+118
-11
lines changed

2 files changed

+118
-11
lines changed

‎libswscale/arm/swscale_unscaled.c‎

Lines changed: 46 additions & 6 deletions
Original file line numberDiff line numberDiff line change
@@ -63,6 +63,50 @@ static int rgbx_to_nv12_neon_16_wrapper(SwsContext *context, const uint8_t *src[
6363
}
6464
#endif
6565

66+
#define YUV_TO_RGB_TABLE(precision) \
67+
c->yuv2rgb_v2r_coeff / ((precision) == 16 ? 1 << 7 : 1), \
68+
c->yuv2rgb_u2g_coeff / ((precision) == 16 ? 1 << 7 : 1), \
69+
c->yuv2rgb_v2g_coeff / ((precision) == 16 ? 1 << 7 : 1), \
70+
c->yuv2rgb_u2b_coeff / ((precision) == 16 ? 1 << 7 : 1), \
71+
72+
#define DECLARE_FF_YUV420P_TO_RGBX_FUNCS(ofmt, precision) \
73+
int ff_yuv420p_to_##ofmt##_neon_##precision(int w, int h, \
74+
uint8_t *dst, int linesize, \
75+
const uint8_t *srcY, int linesizeY, \
76+
const uint8_t *srcU, int linesizeU, \
77+
const uint8_t *srcV, int linesizeV, \
78+
const int16_t *table, \
79+
int y_offset, \
80+
int y_coeff); \
81+
\
82+
static int yuv420p_to_##ofmt##_neon_wrapper_##precision(SwsContext *c, const uint8_t *src[],\
83+
int srcStride[], int srcSliceY, int srcSliceH, \
84+
uint8_t *dst[], int dstStride[]) { \
85+
const int16_t yuv2rgb_table[] = { YUV_TO_RGB_TABLE(precision) }; \
86+
\
87+
ff_yuv420p_to_##ofmt##_neon_##precision(c->srcW, srcSliceH, \
88+
dst[0] + srcSliceY * dstStride[0], dstStride[0], \
89+
src[0], srcStride[0], \
90+
src[1], srcStride[1], \
91+
src[2], srcStride[2], \
92+
yuv2rgb_table, \
93+
c->yuv2rgb_y_offset >> 9, \
94+
c->yuv2rgb_y_coeff / ((precision) == 16 ? 1 << 7 : 1)); \
95+
\
96+
return 0; \
97+
} \
98+
99+
#define DECLARE_FF_YUV420P_TO_ALL_RGBX_FUNCS(precision) \
100+
DECLARE_FF_YUV420P_TO_RGBX_FUNCS(argb, precision) \
101+
DECLARE_FF_YUV420P_TO_RGBX_FUNCS(rgba, precision) \
102+
DECLARE_FF_YUV420P_TO_RGBX_FUNCS(abgr, precision) \
103+
DECLARE_FF_YUV420P_TO_RGBX_FUNCS(bgra, precision) \
104+
105+
#define DECLARE_FF_YUV420P_TO_ALL_RGBX_ALL_PRECISION_FUNCS \
106+
DECLARE_FF_YUV420P_TO_ALL_RGBX_FUNCS(16) \
107+
108+
DECLARE_FF_YUV420P_TO_ALL_RGBX_ALL_PRECISION_FUNCS
109+
66110
#define DECLARE_FF_NVX_TO_RGBX_FUNCS(ifmt, ofmt, precision) \
67111
int ff_##ifmt##_to_##ofmt##_neon_##precision(int w, int h, \
68112
uint8_t *dst, int linesize, \
@@ -75,12 +119,7 @@ int ff_##ifmt##_to_##ofmt##_neon_##precision(int w, int h,
75119
static int ifmt##_to_##ofmt##_neon_wrapper_##precision(SwsContext *c, const uint8_t *src[], \
76120
int srcStride[], int srcSliceY, int srcSliceH, \
77121
uint8_t *dst[], int dstStride[]) { \
78-
const int16_t yuv2rgb_table[] = { \
79-
c->yuv2rgb_v2r_coeff / ((precision) == 16 ? 1 << 7 : 1), \
80-
c->yuv2rgb_u2g_coeff / ((precision) == 16 ? 1 << 7 : 1), \
81-
c->yuv2rgb_v2g_coeff / ((precision) == 16 ? 1 << 7 : 1), \
82-
c->yuv2rgb_u2b_coeff / ((precision) == 16 ? 1 << 7 : 1), \
83-
}; \
122+
const int16_t yuv2rgb_table[] = { YUV_TO_RGB_TABLE(precision) }; \
84123
\
85124
ff_##ifmt##_to_##ofmt##_neon_##precision(c->srcW, srcSliceH, \
86125
dst[0] + srcSliceY * dstStride[0], dstStride[0], \
@@ -138,6 +177,7 @@ static void get_unscaled_swscale_neon(SwsContext *c) {
138177

139178
SET_FF_NVX_TO_ALL_RGBX_FUNC(nv12, NV12, accurate_rnd);
140179
SET_FF_NVX_TO_ALL_RGBX_FUNC(nv21, NV21, accurate_rnd);
180+
SET_FF_NVX_TO_ALL_RGBX_FUNC(yuv420p, YUV420P, accurate_rnd);
141181
}
142182

143183
void ff_get_unscaled_swscale_arm(SwsContext *c)

‎libswscale/arm/yuv2rgb_neon.S‎

Lines changed: 72 additions & 5 deletions
Original file line numberDiff line numberDiff line change
@@ -103,7 +103,8 @@
103103
vmovl.u8 q15, \y1 @ 8px of y
104104

105105
vdup.16 q5, r9 @ q5 = y_offset
106-
vdup.16 q7, r10 @ q7 = y_coeff
106+
vmov d14, d0 @ q7 = y_coeff
107+
vmov d15, d0 @ q7 = y_coeff
107108

108109
vsub.s16 q14, q5
109110
vsub.s16 q15, q5
@@ -184,7 +185,7 @@
184185
compute_8px_32 r11, d30, \ofmt
185186
.endm
186187

187-
.macro load_args
188+
.macro load_args_nvx
188189
push {r4-r12, lr}
189190
vpush {q4-q7}
190191
ldr r4, [sp, #104] @ r4 = srcY
@@ -206,26 +207,73 @@
206207
sub r7, r7, r0 @ r7 = linesizeC - width (paddingC)
207208
.endm
208209

210+
.macro load_args_yuv420p
211+
push {r4-r12, lr}
212+
vpush {q4-q7}
213+
ldr r4, [sp, #104] @ r4 = srcY
214+
ldr r5, [sp, #108] @ r5 = linesizeY
215+
ldr r6, [sp, #112] @ r6 = srcU
216+
ldr r8, [sp, #128] @ r8 = table
217+
ldr r9, [sp, #132] @ r9 = y_offset
218+
ldr r10,[sp, #136] @ r10 = y_coeff
219+
vdup.16 d0, r10 @ d0 = y_coeff
220+
vld1.16 {d1}, [r8] @ d1 = *table
221+
add r11, r2, r3 @ r11 = dst + linesize (dst2)
222+
add r12, r4, r5 @ r12 = srcY + linesizeY (srcY2)
223+
lsl r3, r3, #1
224+
lsl r5, r5, #1
225+
lsl r8, r0, #2
226+
sub r3, r3, r8 @ r3 = linesize * 2 - width * 4 (padding)
227+
sub r5, r5, r0 @ r5 = linesizeY * 2 - width (paddingY)
228+
ldr r10,[sp, #120] @ r10 = srcV
229+
.endm
230+
209231
.macro declare_func ifmt ofmt precision
210232
function ff_\ifmt\()_to_\ofmt\()_neon_\precision\(), export=1
211-
load_args
233+
234+
.ifc \ifmt,nv12
235+
load_args_nvx
236+
.endif
237+
238+
.ifc \ifmt,nv21
239+
load_args_nvx
240+
.endif
241+
242+
.ifc \ifmt,yuv420p
243+
load_args_yuv420p
244+
.endif
245+
212246
1:
213247
mov r8, r0 @ r8 = width
214248
2:
215249
pld [r6, #64*3]
216250
pld [r4, #64*3]
217251
pld [r12, #64*3]
218252

219-
vld2.8 {d2, d3}, [r6]! @ q1: interleaved chroma line
220253
vmov.i8 d10, #128
254+
221255
.ifc \ifmt,nv12
256+
vld2.8 {d2, d3}, [r6]! @ q1: interleaved chroma line
222257
vsubl.u8 q14, d2, d10 @ q14 = U - 128
223258
vsubl.u8 q15, d3, d10 @ q15 = V - 128
224-
.else
259+
.endif
260+
261+
.ifc \ifmt,nv21
262+
vld2.8 {d2, d3}, [r6]! @ q1: interleaved chroma line
225263
vsubl.u8 q14, d3, d10 @ q14 = U - 128
226264
vsubl.u8 q15, d2, d10 @ q15 = V - 128
227265
.endif
228266

267+
.ifc \ifmt,yuv420p
268+
pld [r10, #64*3]
269+
270+
vld1.8 d2, [r6]! @ d2: chroma red line
271+
vld1.8 d3, [r10]! @ d3: chroma blue line
272+
vsubl.u8 q14, d2, d10 @ q14 = U - 128
273+
vsubl.u8 q15, d3, d10 @ q15 = V - 128
274+
.endif
275+
276+
229277
process_16px_\precision \ofmt
230278

231279
subs r8, r8, #16 @ width -= 16
@@ -235,7 +283,24 @@ function ff_\ifmt\()_to_\ofmt\()_neon_\precision\(), export=1
235283
add r4, r4, r5 @ srcY += paddingY
236284
add r11, r11, r3 @ dst2 += padding
237285
add r12, r12, r5 @ srcY2 += paddingY
286+
287+
.ifc \ifmt,nv12
238288
add r6, r6, r7 @ srcC += paddingC
289+
.endif
290+
291+
.ifc \ifmt,nv21
292+
add r6, r6, r7 @ srcC += paddingC
293+
.endif
294+
295+
.ifc \ifmt,yuv420p
296+
ldr r7, [sp, #116] @ r7 = linesizeU
297+
sub r7, r7, r0, lsr #1 @ r7 = linesizeU - width / 2 (paddingU)
298+
add r6, r6, r7 @ srcU += paddingU
299+
300+
ldr r7, [sp, #124] @ r7 = linesizeV
301+
sub r7, r7, r0, lsr #1 @ r7 = linesizeV - width / 2 (paddingV)
302+
add r10, r10, r7 @ srcU += paddingV
303+
.endif
239304

240305
subs r1, r1, #2 @ height -= 2
241306
bgt 1b
@@ -257,3 +322,5 @@ declare_rgb_funcs nv12, 16
257322
declare_rgb_funcs nv21, 16
258323
declare_rgb_funcs nv12, 32
259324
declare_rgb_funcs nv21, 32
325+
declare_rgb_funcs yuv420p, 16
326+
declare_rgb_funcs yuv420p, 32

0 commit comments

Comments
 (0)