2 * AltiVec-enhanced yuv2yuvX
4 * Copyright (C) 2004 Romain Dolbeau <romain@dolbeau.org>
5 * based on the equivalent C code in swscale.c
7 * This file is part of FFmpeg.
9 * FFmpeg is free software; you can redistribute it and/or
10 * modify it under the terms of the GNU Lesser General Public
11 * License as published by the Free Software Foundation; either
12 * version 2.1 of the License, or (at your option) any later version.
14 * FFmpeg is distributed in the hope that it will be useful,
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
17 * Lesser General Public License for more details.
19 * You should have received a copy of the GNU Lesser General Public
20 * License along with FFmpeg; if not, write to the Free Software
21 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
27 #include "libswscale/swscale.h"
28 #include "libswscale/swscale_internal.h"
29 #include "libavutil/attributes.h"
30 #include "libavutil/cpu.h"
31 #include "yuv2rgb_altivec.h"
32 #include "libavutil/ppc/util_altivec.h"
35 #define vzero vec_splat_s32(0)
38 #define GET_LS(a,b,c,s) {\
40 a = vec_vsx_ld(((b) << 1) + 16, s);\
43 #define yuv2planeX_8(d1, d2, l1, src, x, perm, filter) do {\
44 vector signed short ls;\
45 GET_LS(l1, x, perm, src);\
46 vector signed int i1 = vec_mule(filter, ls);\
47 vector signed int i2 = vec_mulo(filter, ls);\
48 vector signed int vf1, vf2;\
49 vf1 = vec_mergeh(i1, i2);\
50 vf2 = vec_mergel(i1, i2);\
51 d1 = vec_add(d1, vf1);\
52 d2 = vec_add(d2, vf2);\
55 #define LOAD_FILTER(vf,f) {\
56 vf = vec_vsx_ld(joffset, f);\
58 #define LOAD_L1(ll1,s,p){\
59 ll1 = vec_vsx_ld(xoffset, s);\
62 // The 3 above is 2 (filterSize == 4) + 1 (sizeof(short) == 2).
64 // The neat trick: We only care for half the elements,
65 // high or low depending on (i<<3)%16 (it's 0 or 8 here),
66 // and we're going to use vec_mule, so we choose
67 // carefully how to "unpack" the elements into the even slots.
68 #define GET_VF4(a, vf, f) {\
69 vf = (vector signed short)vec_vsx_ld(a << 3, f);\
70 vf = vec_mergeh(vf, (vector signed short)vzero);\
72 #define FIRST_LOAD(sv, pos, s, per) {}
73 #define UPDATE_PTR(s0, d0, s1, d1) {}
74 #define LOAD_SRCV(pos, a, s, per, v0, v1, vf) {\
75 vf = vec_vsx_ld(pos + a, s);\
77 #define LOAD_SRCV8(pos, a, s, per, v0, v1, vf) LOAD_SRCV(pos, a, s, per, v0, v1, vf)
78 #define GET_VFD(a, b, f, vf0, vf1, per, vf, off) {\
79 vf = vec_vsx_ld((a * 2 * filterSize) + (b * 2) + off, f);\
82 #define FUNC(name) name ## _vsx
83 #include "swscale_ppc_template.c"
86 #endif /* !HAVE_BIGENDIAN */
88 static void yuv2plane1_8_u(const int16_t *src, uint8_t *dest, int dstW,
89 const uint8_t *dither, int offset, int start)
92 for (i = start; i < dstW; i++) {
93 int val = (src[i] + dither[(i + offset) & 7]) >> 7;
94 dest[i] = av_clip_uint8(val);
98 static void yuv2plane1_8_vsx(const int16_t *src, uint8_t *dest, int dstW,
99 const uint8_t *dither, int offset)
101 const int dst_u = -(uintptr_t)dest & 15;
103 LOCAL_ALIGNED(16, int16_t, val, [16]);
104 const vector uint16_t shifts = (vector uint16_t) {7, 7, 7, 7, 7, 7, 7, 7};
105 vector int16_t vi, vileft, ditherleft, ditherright;
108 for (j = 0; j < 16; j++) {
109 val[j] = dither[(dst_u + offset + j) & 7];
112 ditherleft = vec_ld(0, val);
113 ditherright = vec_ld(0, &val[8]);
115 yuv2plane1_8_u(src, dest, dst_u, dither, offset, 0);
117 for (i = dst_u; i < dstW - 15; i += 16) {
119 vi = vec_vsx_ld(0, &src[i]);
120 vi = vec_adds(ditherleft, vi);
121 vileft = vec_sra(vi, shifts);
123 vi = vec_vsx_ld(0, &src[i + 8]);
124 vi = vec_adds(ditherright, vi);
125 vi = vec_sra(vi, shifts);
127 vd = vec_packsu(vileft, vi);
128 vec_st(vd, 0, &dest[i]);
131 yuv2plane1_8_u(src, dest, dstW, dither, offset, i);
136 #define output_pixel(pos, val) \
138 AV_WB16(pos, av_clip_uintp2(val >> shift, output_bits)); \
140 AV_WL16(pos, av_clip_uintp2(val >> shift, output_bits)); \
143 static void yuv2plane1_nbps_u(const int16_t *src, uint16_t *dest, int dstW,
144 int big_endian, int output_bits, int start)
147 int shift = 15 - output_bits;
149 for (i = start; i < dstW; i++) {
150 int val = src[i] + (1 << (shift - 1));
151 output_pixel(&dest[i], val);
155 static void yuv2plane1_nbps_vsx(const int16_t *src, uint16_t *dest, int dstW,
156 int big_endian, int output_bits)
158 const int dst_u = -(uintptr_t)dest & 7;
159 const int shift = 15 - output_bits;
160 const int add = (1 << (shift - 1));
161 const int clip = (1 << output_bits) - 1;
162 const vector uint16_t vadd = (vector uint16_t) {add, add, add, add, add, add, add, add};
163 const vector uint16_t vswap = (vector uint16_t) vec_splat_u16(big_endian ? 8 : 0);
164 const vector uint16_t vshift = (vector uint16_t) vec_splat_u16(shift);
165 const vector uint16_t vlargest = (vector uint16_t) {clip, clip, clip, clip, clip, clip, clip, clip};
169 yuv2plane1_nbps_u(src, dest, dst_u, big_endian, output_bits, 0);
171 for (i = dst_u; i < dstW - 7; i += 8) {
172 v = vec_vsx_ld(0, (const uint16_t *) &src[i]);
173 v = vec_add(v, vadd);
174 v = vec_sr(v, vshift);
175 v = vec_min(v, vlargest);
176 v = vec_rl(v, vswap);
177 vec_st(v, 0, &dest[i]);
180 yuv2plane1_nbps_u(src, dest, dstW, big_endian, output_bits, i);
185 #define output_pixel(pos, val, bias, signedness) \
187 AV_WB16(pos, bias + av_clip_ ## signedness ## 16(val >> shift)); \
189 AV_WL16(pos, bias + av_clip_ ## signedness ## 16(val >> shift)); \
192 static void yuv2plane1_16_u(const int32_t *src, uint16_t *dest, int dstW,
193 int big_endian, int output_bits, int start)
198 for (i = start; i < dstW; i++) {
199 int val = src[i] + (1 << (shift - 1));
200 output_pixel(&dest[i], val, 0, uint);
204 static void yuv2plane1_16_vsx(const int32_t *src, uint16_t *dest, int dstW,
205 int big_endian, int output_bits)
207 const int dst_u = -(uintptr_t)dest & 7;
209 const int add = (1 << (shift - 1));
210 const vector uint32_t vadd = (vector uint32_t) {add, add, add, add};
211 const vector uint16_t vswap = (vector uint16_t) vec_splat_u16(big_endian ? 8 : 0);
212 const vector uint32_t vshift = (vector uint32_t) vec_splat_u32(shift);
213 vector uint32_t v, v2;
217 yuv2plane1_16_u(src, dest, dst_u, big_endian, output_bits, 0);
219 for (i = dst_u; i < dstW - 7; i += 8) {
220 v = vec_vsx_ld(0, (const uint32_t *) &src[i]);
221 v = vec_add(v, vadd);
222 v = vec_sr(v, vshift);
224 v2 = vec_vsx_ld(0, (const uint32_t *) &src[i + 4]);
225 v2 = vec_add(v2, vadd);
226 v2 = vec_sr(v2, vshift);
228 vd = vec_packsu(v, v2);
229 vd = vec_rl(vd, vswap);
231 vec_st(vd, 0, &dest[i]);
234 yuv2plane1_16_u(src, dest, dstW, big_endian, output_bits, i);
237 #define yuv2NBPS(bits, BE_LE, is_be, template_size, typeX_t) \
238 static void yuv2plane1_ ## bits ## BE_LE ## _vsx(const int16_t *src, \
239 uint8_t *dest, int dstW, \
240 const uint8_t *dither, int offset) \
242 yuv2plane1_ ## template_size ## _vsx((const typeX_t *) src, \
243 (uint16_t *) dest, dstW, is_be, bits); \
246 yuv2NBPS( 9, BE, 1, nbps, int16_t)
247 yuv2NBPS( 9, LE, 0, nbps, int16_t)
248 yuv2NBPS(10, BE, 1, nbps, int16_t)
249 yuv2NBPS(10, LE, 0, nbps, int16_t)
250 yuv2NBPS(12, BE, 1, nbps, int16_t)
251 yuv2NBPS(12, LE, 0, nbps, int16_t)
252 yuv2NBPS(14, BE, 1, nbps, int16_t)
253 yuv2NBPS(14, LE, 0, nbps, int16_t)
254 yuv2NBPS(16, BE, 1, 16, int32_t)
255 yuv2NBPS(16, LE, 0, 16, int32_t)
257 #endif /* !HAVE_BIGENDIAN */
259 #endif /* HAVE_VSX */
261 av_cold void ff_sws_init_swscale_vsx(SwsContext *c)
264 enum AVPixelFormat dstFormat = c->dstFormat;
266 if (!(av_get_cpu_flags() & AV_CPU_FLAG_VSX))
270 if (c->srcBpc == 8 && c->dstBpc <= 14) {
271 c->hyScale = c->hcScale = hScale_real_vsx;
273 if (!is16BPS(dstFormat) && !isNBPS(dstFormat) &&
274 dstFormat != AV_PIX_FMT_NV12 && dstFormat != AV_PIX_FMT_NV21 &&
275 dstFormat != AV_PIX_FMT_GRAYF32BE && dstFormat != AV_PIX_FMT_GRAYF32LE &&
277 c->yuv2planeX = yuv2planeX_vsx;
281 if (!(c->flags & (SWS_BITEXACT | SWS_FULL_CHR_H_INT)) && !c->needAlpha) {
284 c->yuv2plane1 = yuv2plane1_8_vsx;
288 c->yuv2plane1 = isBE(dstFormat) ? yuv2plane1_9BE_vsx : yuv2plane1_9LE_vsx;
291 c->yuv2plane1 = isBE(dstFormat) ? yuv2plane1_10BE_vsx : yuv2plane1_10LE_vsx;
294 c->yuv2plane1 = isBE(dstFormat) ? yuv2plane1_12BE_vsx : yuv2plane1_12LE_vsx;
297 c->yuv2plane1 = isBE(dstFormat) ? yuv2plane1_14BE_vsx : yuv2plane1_14LE_vsx;
300 c->yuv2plane1 = isBE(dstFormat) ? yuv2plane1_16BE_vsx : yuv2plane1_16LE_vsx;
305 #endif /* HAVE_VSX */