nengel@2: /*
nengel@2:  * This file is part of FFmpeg.
nengel@2:  *
nengel@2:  * FFmpeg is free software; you can redistribute it and/or
nengel@2:  * modify it under the terms of the GNU Lesser General Public
nengel@2:  * License as published by the Free Software Foundation; either
nengel@2:  * version 2.1 of the License, or (at your option) any later version.
nengel@2:  *
nengel@2:  * FFmpeg is distributed in the hope that it will be useful,
nengel@2:  * but WITHOUT ANY WARRANTY; without even the implied warranty of
nengel@2:  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
nengel@2:  * Lesser General Public License for more details.
nengel@2:  *
nengel@2:  * You should have received a copy of the GNU Lesser General Public
nengel@2:  * License along with FFmpeg; if not, write to the Free Software
nengel@2:  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
nengel@2:  */
nengel@2: 
nengel@2: /**
nengel@2:  * @file
nengel@2:  * Contains misc utility macros and inline functions
nengel@2:  */
nengel@2: 
nengel@2: #ifndef AVCODEC_PPC_UTIL_ALTIVEC_H
nengel@2: #define AVCODEC_PPC_UTIL_ALTIVEC_H
nengel@2: 
nengel@2: #include <stdint.h>
nengel@2: 
nengel@2: #include "config.h"
nengel@2: 
nengel@2: #if HAVE_ALTIVEC_H
nengel@2: #include <altivec.h>
nengel@2: #endif
nengel@2: 
nengel@2: // used to build registers permutation vectors (vcprm)
nengel@2: // the 's' are for words in the _s_econd vector
nengel@2: #define WORD_0 0x00,0x01,0x02,0x03
nengel@2: #define WORD_1 0x04,0x05,0x06,0x07
nengel@2: #define WORD_2 0x08,0x09,0x0a,0x0b
nengel@2: #define WORD_3 0x0c,0x0d,0x0e,0x0f
nengel@2: #define WORD_s0 0x10,0x11,0x12,0x13
nengel@2: #define WORD_s1 0x14,0x15,0x16,0x17
nengel@2: #define WORD_s2 0x18,0x19,0x1a,0x1b
nengel@2: #define WORD_s3 0x1c,0x1d,0x1e,0x1f
nengel@2: 
nengel@2: #define vcprm(a,b,c,d) (const vector unsigned char){WORD_ ## a, WORD_ ## b, WORD_ ## c, WORD_ ## d}
nengel@2: #define vcii(a,b,c,d) (const vector float){FLOAT_ ## a, FLOAT_ ## b, FLOAT_ ## c, FLOAT_ ## d}
nengel@2: 
nengel@2: // vcprmle is used to keep the same index as in the SSE version.
nengel@2: // it's the same as vcprm, with the index inversed
nengel@2: // ('le' is Little Endian)
nengel@2: #define vcprmle(a,b,c,d) vcprm(d,c,b,a)
nengel@2: 
nengel@2: // used to build inverse/identity vectors (vcii)
nengel@2: // n is _n_egative, p is _p_ositive
nengel@2: #define FLOAT_n -1.
nengel@2: #define FLOAT_p 1.
nengel@2: 
nengel@2: 
nengel@2: // Transpose 8x8 matrix of 16-bit elements (in-place)
nengel@2: #define TRANSPOSE8(a,b,c,d,e,f,g,h) \
nengel@2: do { \
nengel@2:     vector signed short A1, B1, C1, D1, E1, F1, G1, H1; \
nengel@2:     vector signed short A2, B2, C2, D2, E2, F2, G2, H2; \
nengel@2:  \
nengel@2:     A1 = vec_mergeh (a, e); \
nengel@2:     B1 = vec_mergel (a, e); \
nengel@2:     C1 = vec_mergeh (b, f); \
nengel@2:     D1 = vec_mergel (b, f); \
nengel@2:     E1 = vec_mergeh (c, g); \
nengel@2:     F1 = vec_mergel (c, g); \
nengel@2:     G1 = vec_mergeh (d, h); \
nengel@2:     H1 = vec_mergel (d, h); \
nengel@2:  \
nengel@2:     A2 = vec_mergeh (A1, E1); \
nengel@2:     B2 = vec_mergel (A1, E1); \
nengel@2:     C2 = vec_mergeh (B1, F1); \
nengel@2:     D2 = vec_mergel (B1, F1); \
nengel@2:     E2 = vec_mergeh (C1, G1); \
nengel@2:     F2 = vec_mergel (C1, G1); \
nengel@2:     G2 = vec_mergeh (D1, H1); \
nengel@2:     H2 = vec_mergel (D1, H1); \
nengel@2:  \
nengel@2:     a = vec_mergeh (A2, E2); \
nengel@2:     b = vec_mergel (A2, E2); \
nengel@2:     c = vec_mergeh (B2, F2); \
nengel@2:     d = vec_mergel (B2, F2); \
nengel@2:     e = vec_mergeh (C2, G2); \
nengel@2:     f = vec_mergel (C2, G2); \
nengel@2:     g = vec_mergeh (D2, H2); \
nengel@2:     h = vec_mergel (D2, H2); \
nengel@2: } while (0)
nengel@2: 
nengel@2: 
nengel@2: /** \brief loads unaligned vector \a *src with offset \a offset
nengel@2:     and returns it */
nengel@2: static inline vector unsigned char unaligned_load(int offset, uint8_t *src)
nengel@2: {
nengel@2:     register vector unsigned char first = vec_ld(offset, src);
nengel@2:     register vector unsigned char second = vec_ld(offset+15, src);
nengel@2:     register vector unsigned char mask = vec_lvsl(offset, src);
nengel@2:     return vec_perm(first, second, mask);
nengel@2: }
nengel@2: 
nengel@2: #endif /* AVCODEC_PPC_UTIL_ALTIVEC_H */