00001
00002
00003
00004
00005
00006
00007
00008
00009
00010
00011
00012
00013
00014
00015
00016
00017
00018
00019
00020
00021
00022 #include "libavutil/mem.h"
00023 #include "libavutil/x86/asm.h"
00024 #include "libavutil/x86/cpu.h"
00025 #include "dsputil_mmx.h"
00026 #include "libavcodec/ac3.h"
00027 #include "libavcodec/ac3dsp.h"
00028
00029 extern void ff_ac3_exponent_min_mmx (uint8_t *exp, int num_reuse_blocks, int nb_coefs);
00030 extern void ff_ac3_exponent_min_mmxext(uint8_t *exp, int num_reuse_blocks, int nb_coefs);
00031 extern void ff_ac3_exponent_min_sse2 (uint8_t *exp, int num_reuse_blocks, int nb_coefs);
00032
00033 extern int ff_ac3_max_msb_abs_int16_mmx (const int16_t *src, int len);
00034 extern int ff_ac3_max_msb_abs_int16_mmxext(const int16_t *src, int len);
00035 extern int ff_ac3_max_msb_abs_int16_sse2 (const int16_t *src, int len);
00036 extern int ff_ac3_max_msb_abs_int16_ssse3(const int16_t *src, int len);
00037
00038 extern void ff_ac3_lshift_int16_mmx (int16_t *src, unsigned int len, unsigned int shift);
00039 extern void ff_ac3_lshift_int16_sse2(int16_t *src, unsigned int len, unsigned int shift);
00040
00041 extern void ff_ac3_rshift_int32_mmx (int32_t *src, unsigned int len, unsigned int shift);
00042 extern void ff_ac3_rshift_int32_sse2(int32_t *src, unsigned int len, unsigned int shift);
00043
00044 extern void ff_float_to_fixed24_3dnow(int32_t *dst, const float *src, unsigned int len);
00045 extern void ff_float_to_fixed24_sse (int32_t *dst, const float *src, unsigned int len);
00046 extern void ff_float_to_fixed24_sse2 (int32_t *dst, const float *src, unsigned int len);
00047
00048 extern int ff_ac3_compute_mantissa_size_sse2(uint16_t mant_cnt[6][16]);
00049
00050 extern void ff_ac3_extract_exponents_3dnow(uint8_t *exp, int32_t *coef, int nb_coefs);
00051 extern void ff_ac3_extract_exponents_sse2 (uint8_t *exp, int32_t *coef, int nb_coefs);
00052 extern void ff_ac3_extract_exponents_ssse3(uint8_t *exp, int32_t *coef, int nb_coefs);
00053
00054 #if HAVE_SSE_INLINE && HAVE_7REGS
00055
00056 #define IF1(x) x
00057 #define IF0(x)
00058
00059 #define MIX5(mono, stereo) \
00060 __asm__ volatile ( \
00061 "movss 0(%1), %%xmm5 \n" \
00062 "movss 8(%1), %%xmm6 \n" \
00063 "movss 24(%1), %%xmm7 \n" \
00064 "shufps $0, %%xmm5, %%xmm5 \n" \
00065 "shufps $0, %%xmm6, %%xmm6 \n" \
00066 "shufps $0, %%xmm7, %%xmm7 \n" \
00067 "1: \n" \
00068 "movaps (%0, %2), %%xmm0 \n" \
00069 "movaps (%0, %3), %%xmm1 \n" \
00070 "movaps (%0, %4), %%xmm2 \n" \
00071 "movaps (%0, %5), %%xmm3 \n" \
00072 "movaps (%0, %6), %%xmm4 \n" \
00073 "mulps %%xmm5, %%xmm0 \n" \
00074 "mulps %%xmm6, %%xmm1 \n" \
00075 "mulps %%xmm5, %%xmm2 \n" \
00076 "mulps %%xmm7, %%xmm3 \n" \
00077 "mulps %%xmm7, %%xmm4 \n" \
00078 stereo("addps %%xmm1, %%xmm0 \n") \
00079 "addps %%xmm1, %%xmm2 \n" \
00080 "addps %%xmm3, %%xmm0 \n" \
00081 "addps %%xmm4, %%xmm2 \n" \
00082 mono("addps %%xmm2, %%xmm0 \n") \
00083 "movaps %%xmm0, (%0, %2) \n" \
00084 stereo("movaps %%xmm2, (%0, %3) \n") \
00085 "add $16, %0 \n" \
00086 "jl 1b \n" \
00087 : "+&r"(i) \
00088 : "r"(matrix), \
00089 "r"(samples[0] + len), \
00090 "r"(samples[1] + len), \
00091 "r"(samples[2] + len), \
00092 "r"(samples[3] + len), \
00093 "r"(samples[4] + len) \
00094 : XMM_CLOBBERS("%xmm0", "%xmm1", "%xmm2", "%xmm3", \
00095 "%xmm4", "%xmm5", "%xmm6", "%xmm7",) \
00096 "memory" \
00097 );
00098
00099 #define MIX_MISC(stereo) \
00100 __asm__ volatile ( \
00101 "mov %5, %2 \n" \
00102 "1: \n" \
00103 "mov -%c7(%6, %2, %c8), %3 \n" \
00104 "movaps (%3, %0), %%xmm0 \n" \
00105 stereo("movaps %%xmm0, %%xmm1 \n") \
00106 "mulps %%xmm4, %%xmm0 \n" \
00107 stereo("mulps %%xmm5, %%xmm1 \n") \
00108 "2: \n" \
00109 "mov (%6, %2, %c8), %1 \n" \
00110 "movaps (%1, %0), %%xmm2 \n" \
00111 stereo("movaps %%xmm2, %%xmm3 \n") \
00112 "mulps (%4, %2, 8), %%xmm2 \n" \
00113 stereo("mulps 16(%4, %2, 8), %%xmm3 \n") \
00114 "addps %%xmm2, %%xmm0 \n" \
00115 stereo("addps %%xmm3, %%xmm1 \n") \
00116 "add $4, %2 \n" \
00117 "jl 2b \n" \
00118 "mov %5, %2 \n" \
00119 stereo("mov (%6, %2, %c8), %1 \n") \
00120 "movaps %%xmm0, (%3, %0) \n" \
00121 stereo("movaps %%xmm1, (%1, %0) \n") \
00122 "add $16, %0 \n" \
00123 "jl 1b \n" \
00124 : "+&r"(i), "=&r"(j), "=&r"(k), "=&r"(m) \
00125 : "r"(matrix_simd + in_ch), \
00126 "g"((intptr_t) - 4 * (in_ch - 1)), \
00127 "r"(samp + in_ch), \
00128 "i"(sizeof(float *)), "i"(sizeof(float *)/4) \
00129 : "memory" \
00130 );
00131
00132 static void ac3_downmix_sse(float **samples, float (*matrix)[2],
00133 int out_ch, int in_ch, int len)
00134 {
00135 int (*matrix_cmp)[2] = (int(*)[2])matrix;
00136 intptr_t i, j, k, m;
00137
00138 i = -len * sizeof(float);
00139 if (in_ch == 5 && out_ch == 2 &&
00140 !(matrix_cmp[0][1] | matrix_cmp[2][0] |
00141 matrix_cmp[3][1] | matrix_cmp[4][0] |
00142 (matrix_cmp[1][0] ^ matrix_cmp[1][1]) |
00143 (matrix_cmp[0][0] ^ matrix_cmp[2][1]))) {
00144 MIX5(IF0, IF1);
00145 } else if (in_ch == 5 && out_ch == 1 &&
00146 matrix_cmp[0][0] == matrix_cmp[2][0] &&
00147 matrix_cmp[3][0] == matrix_cmp[4][0]) {
00148 MIX5(IF1, IF0);
00149 } else {
00150 DECLARE_ALIGNED(16, float, matrix_simd)[AC3_MAX_CHANNELS][2][4];
00151 float *samp[AC3_MAX_CHANNELS];
00152
00153 for (j = 0; j < in_ch; j++)
00154 samp[j] = samples[j] + len;
00155
00156 j = 2 * in_ch * sizeof(float);
00157 __asm__ volatile (
00158 "1: \n"
00159 "sub $8, %0 \n"
00160 "movss (%2, %0), %%xmm4 \n"
00161 "movss 4(%2, %0), %%xmm5 \n"
00162 "shufps $0, %%xmm4, %%xmm4 \n"
00163 "shufps $0, %%xmm5, %%xmm5 \n"
00164 "movaps %%xmm4, (%1, %0, 4) \n"
00165 "movaps %%xmm5, 16(%1, %0, 4) \n"
00166 "jg 1b \n"
00167 : "+&r"(j)
00168 : "r"(matrix_simd), "r"(matrix)
00169 : "memory"
00170 );
00171 if (out_ch == 2) {
00172 MIX_MISC(IF1);
00173 } else {
00174 MIX_MISC(IF0);
00175 }
00176 }
00177 }
00178
00179 #endif
00180
00181 av_cold void ff_ac3dsp_init_x86(AC3DSPContext *c, int bit_exact)
00182 {
00183 int mm_flags = av_get_cpu_flags();
00184
00185 if (EXTERNAL_MMX(mm_flags)) {
00186 c->ac3_exponent_min = ff_ac3_exponent_min_mmx;
00187 c->ac3_max_msb_abs_int16 = ff_ac3_max_msb_abs_int16_mmx;
00188 c->ac3_lshift_int16 = ff_ac3_lshift_int16_mmx;
00189 c->ac3_rshift_int32 = ff_ac3_rshift_int32_mmx;
00190 }
00191 if (EXTERNAL_AMD3DNOW(mm_flags)) {
00192 c->extract_exponents = ff_ac3_extract_exponents_3dnow;
00193 if (!bit_exact) {
00194 c->float_to_fixed24 = ff_float_to_fixed24_3dnow;
00195 }
00196 }
00197 if (EXTERNAL_MMXEXT(mm_flags)) {
00198 c->ac3_exponent_min = ff_ac3_exponent_min_mmxext;
00199 c->ac3_max_msb_abs_int16 = ff_ac3_max_msb_abs_int16_mmxext;
00200 }
00201 if (EXTERNAL_SSE(mm_flags)) {
00202 c->float_to_fixed24 = ff_float_to_fixed24_sse;
00203 }
00204 if (EXTERNAL_SSE2(mm_flags)) {
00205 c->ac3_exponent_min = ff_ac3_exponent_min_sse2;
00206 c->ac3_max_msb_abs_int16 = ff_ac3_max_msb_abs_int16_sse2;
00207 c->float_to_fixed24 = ff_float_to_fixed24_sse2;
00208 c->compute_mantissa_size = ff_ac3_compute_mantissa_size_sse2;
00209 c->extract_exponents = ff_ac3_extract_exponents_sse2;
00210 if (!(mm_flags & AV_CPU_FLAG_SSE2SLOW)) {
00211 c->ac3_lshift_int16 = ff_ac3_lshift_int16_sse2;
00212 c->ac3_rshift_int32 = ff_ac3_rshift_int32_sse2;
00213 }
00214 }
00215 if (EXTERNAL_SSSE3(mm_flags)) {
00216 c->ac3_max_msb_abs_int16 = ff_ac3_max_msb_abs_int16_ssse3;
00217 if (!(mm_flags & AV_CPU_FLAG_ATOM)) {
00218 c->extract_exponents = ff_ac3_extract_exponents_ssse3;
00219 }
00220 }
00221
00222 #if HAVE_SSE_INLINE && HAVE_7REGS
00223 if (INLINE_SSE(mm_flags)) {
00224 c->downmix = ac3_downmix_sse;
00225 }
00226 #endif
00227 }
