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"
36 #define vzero vec_splat_s32(0)
38 #define GET_LS(a,b,c,s) {\
39 vector signed short l2 = vec_ld(((b) << 1) + 16, s);\
40 ls = vec_perm(a, l2, c);\
44 #define yuv2planeX_8(d1, d2, l1, src, x, perm, filter) do {\
45 vector signed short ls;\
46 GET_LS(l1, x, perm, src);\
47 vector signed int i1 = vec_mule(filter, ls);\
48 vector signed int i2 = vec_mulo(filter, ls);\
49 vector signed int vf1, vf2;\
50 vf1 = vec_mergeh(i1, i2);\
51 vf2 = vec_mergel(i1, i2);\
52 d1 = vec_add(d1, vf1);\
53 d2 = vec_add(d2, vf2);\
56 #define LOAD_FILTER(vf,f) {\
57 vector unsigned char perm0 = vec_lvsl(joffset, f);\
58 vf = vec_ld(joffset, f);\
59 vf = vec_perm(vf, vf, perm0);\
61 #define LOAD_L1(ll1,s,p){\
62 p = vec_lvsl(xoffset, s);\
63 ll1 = vec_ld(xoffset, s);\
66 // The 3 above is 2 (filterSize == 4) + 1 (sizeof(short) == 2).
68 // The neat trick: We only care for half the elements,
69 // high or low depending on (i<<3)%16 (it's 0 or 8 here),
70 // and we're going to use vec_mule, so we choose
71 // carefully how to "unpack" the elements into the even slots.
72 #define GET_VF4(a, vf, f) {\
73 vf = vec_ld(a<< 3, f);\
75 vf = vec_mergel(vf, (vector signed short)vzero);\
77 vf = vec_mergeh(vf, (vector signed short)vzero);\
79 #define FIRST_LOAD(sv, pos, s, per) {\
81 per = vec_lvsl(pos, s);\
83 #define UPDATE_PTR(s0, d0, s1, d1) {\
87 #define LOAD_SRCV(pos, a, s, per, v0, v1, vf) {\
88 v1 = vec_ld(pos + a + 16, s);\
89 vf = vec_perm(v0, v1, per);\
91 #define LOAD_SRCV8(pos, a, s, per, v0, v1, vf) {\
92 if ((((uintptr_t)s + pos) % 16) > 8) {\
93 v1 = vec_ld(pos + a + 16, s);\
95 vf = vec_perm(v0, src_v1, per);\
97 #define GET_VFD(a, b, f, vf0, vf1, per, vf, off) {\
98 vf1 = vec_ld((a * 2 * filterSize) + (b * 2) + 16 + off, f);\
99 vf = vec_perm(vf0, vf1, per);\
102 #define FUNC(name) name ## _altivec
103 #include "swscale_ppc_template.c"
108 #endif /* HAVE_BIGENDIAN */
110 #define output_pixel(pos, val, bias, signedness) \
112 AV_WB16(pos, bias + av_clip_ ## signedness ## 16(val >> shift)); \
114 AV_WL16(pos, bias + av_clip_ ## signedness ## 16(val >> shift)); \
118 yuv2plane1_float_u(const int32_t *src, float *dest, int dstW, int start)
120 static const int big_endian = HAVE_BIGENDIAN;
121 static const int shift = 3;
122 static const float float_mult = 1.0f / 65535.0f;
126 for (i = start; i < dstW; ++i){
127 val = src[i] + (1 << (shift - 1));
128 output_pixel(&val_uint, val, 0, uint);
129 dest[i] = float_mult * (float)val_uint;
134 yuv2plane1_float_bswap_u(const int32_t *src, uint32_t *dest, int dstW, int start)
136 static const int big_endian = HAVE_BIGENDIAN;
137 static const int shift = 3;
138 static const float float_mult = 1.0f / 65535.0f;
142 for (i = start; i < dstW; ++i){
143 val = src[i] + (1 << (shift - 1));
144 output_pixel(&val_uint, val, 0, uint);
145 dest[i] = av_bswap32(av_float2int(float_mult * (float)val_uint));
149 static void yuv2plane1_float_altivec(const int32_t *src, float *dest, int dstW)
151 const int dst_u = -(uintptr_t)dest & 3;
153 const int add = (1 << (shift - 1));
154 const int clip = (1 << 16) - 1;
155 const float fmult = 1.0f / 65535.0f;
156 const vector uint32_t vadd = (vector uint32_t) {add, add, add, add};
157 const vector uint32_t vshift = (vector uint32_t) vec_splat_u32(shift);
158 const vector uint32_t vlargest = (vector uint32_t) {clip, clip, clip, clip};
159 const vector float vmul = (vector float) {fmult, fmult, fmult, fmult};
160 const vector float vzero = (vector float) {0, 0, 0, 0};
165 yuv2plane1_float_u(src, dest, dst_u, 0);
167 for (i = dst_u; i < dstW - 3; i += 4) {
168 v = vec_ld(0, (const uint32_t *) &src[i]);
169 v = vec_add(v, vadd);
170 v = vec_sr(v, vshift);
171 v = vec_min(v, vlargest);
174 vd = vec_madd(vd, vmul, vzero);
176 vec_st(vd, 0, &dest[i]);
179 yuv2plane1_float_u(src, dest, dstW, i);
182 static void yuv2plane1_float_bswap_altivec(const int32_t *src, uint32_t *dest, int dstW)
184 const int dst_u = -(uintptr_t)dest & 3;
186 const int add = (1 << (shift - 1));
187 const int clip = (1 << 16) - 1;
188 const float fmult = 1.0f / 65535.0f;
189 const vector uint32_t vadd = (vector uint32_t) {add, add, add, add};
190 const vector uint32_t vshift = (vector uint32_t) vec_splat_u32(shift);
191 const vector uint32_t vlargest = (vector uint32_t) {clip, clip, clip, clip};
192 const vector float vmul = (vector float) {fmult, fmult, fmult, fmult};
193 const vector float vzero = (vector float) {0, 0, 0, 0};
194 const vector uint32_t vswapbig = (vector uint32_t) {16, 16, 16, 16};
195 const vector uint16_t vswapsmall = vec_splat_u16(8);
200 yuv2plane1_float_bswap_u(src, dest, dst_u, 0);
202 for (i = dst_u; i < dstW - 3; i += 4) {
203 v = vec_ld(0, (const uint32_t *) &src[i]);
204 v = vec_add(v, vadd);
205 v = vec_sr(v, vshift);
206 v = vec_min(v, vlargest);
209 vd = vec_madd(vd, vmul, vzero);
211 vd = (vector float) vec_rl((vector uint32_t) vd, vswapbig);
212 vd = (vector float) vec_rl((vector uint16_t) vd, vswapsmall);
214 vec_st(vd, 0, (float *) &dest[i]);
217 yuv2plane1_float_bswap_u(src, dest, dstW, i);
220 #define yuv2plane1_float(template, dest_type, BE_LE) \
221 static void yuv2plane1_float ## BE_LE ## _altivec(const int16_t *src, uint8_t *dest, \
223 const uint8_t *dither, int offset) \
225 template((const int32_t *)src, (dest_type *)dest, dstW); \
229 yuv2plane1_float(yuv2plane1_float_altivec, float, BE)
230 yuv2plane1_float(yuv2plane1_float_bswap_altivec, uint32_t, LE)
232 yuv2plane1_float(yuv2plane1_float_altivec, float, LE)
233 yuv2plane1_float(yuv2plane1_float_bswap_altivec, uint32_t, BE)
236 #endif /* HAVE_ALTIVEC */
238 av_cold void ff_sws_init_swscale_ppc(SwsContext *c)
241 enum AVPixelFormat dstFormat = c->dstFormat;
243 if (!(av_get_cpu_flags() & AV_CPU_FLAG_ALTIVEC))
247 if (c->srcBpc == 8 && c->dstBpc <= 14) {
248 c->hyScale = c->hcScale = hScale_real_altivec;
250 if (!is16BPS(dstFormat) && !isNBPS(dstFormat) &&
251 dstFormat != AV_PIX_FMT_NV12 && dstFormat != AV_PIX_FMT_NV21 &&
252 dstFormat != AV_PIX_FMT_GRAYF32BE && dstFormat != AV_PIX_FMT_GRAYF32LE &&
254 c->yuv2planeX = yuv2planeX_altivec;
258 if (dstFormat == AV_PIX_FMT_GRAYF32BE) {
259 c->yuv2plane1 = yuv2plane1_floatBE_altivec;
260 } else if (dstFormat == AV_PIX_FMT_GRAYF32LE) {
261 c->yuv2plane1 = yuv2plane1_floatLE_altivec;
264 /* The following list of supported dstFormat values should
265 * match what's found in the body of ff_yuv2packedX_altivec() */
266 if (!(c->flags & (SWS_BITEXACT | SWS_FULL_CHR_H_INT)) && !c->needAlpha) {
267 switch (c->dstFormat) {
268 case AV_PIX_FMT_ABGR:
269 c->yuv2packedX = ff_yuv2abgr_X_altivec;
271 case AV_PIX_FMT_BGRA:
272 c->yuv2packedX = ff_yuv2bgra_X_altivec;
274 case AV_PIX_FMT_ARGB:
275 c->yuv2packedX = ff_yuv2argb_X_altivec;
277 case AV_PIX_FMT_RGBA:
278 c->yuv2packedX = ff_yuv2rgba_X_altivec;
280 case AV_PIX_FMT_BGR24:
281 c->yuv2packedX = ff_yuv2bgr24_X_altivec;
283 case AV_PIX_FMT_RGB24:
284 c->yuv2packedX = ff_yuv2rgb24_X_altivec;
288 #endif /* HAVE_ALTIVEC */
290 ff_sws_init_swscale_vsx(c);