#include /* FFT computation */ #ifndef FFTCOMPLEX_T_DEFINED typedef struct FFTComplex { float re, im; } FFTComplex; #define FFTCOMPLEX_T_DEFINED #endif void sbr_qmf_pre_shuffle_sse(float *z); void fft32_interleave_sse(FFTComplex *z); //void vector_fmul_reverse_sse(float *dst, const float *src0, const float *src1, uint32_t len); //float __attribute__ ((aligned (16))) sbr_qmf_window_ds[320]; //float __attribute__ ((aligned (16))) sbr_ldqmf_window_ds[320]; static const float __attribute__((aligned(16))) tcos_analysis[64] = { -0.008677451287989, -0.078057863137826, -0.147250226758303, -0.216087851684923, -0.284404902049773, -0.352036796094658, -0.418820602662969, -0.484595433715113, -0.54920283192189, -0.612487152402081, -0.674295937684597, -0.734480284991882, -0.792895204959746, -0.849399970929445, -0.903858457970525, -0.956139470817702, -1.00611705993175, -1.05367082492298, -1.09868620460631, -1.14105475298919, -1.18067440052753, -1.21744970002014, -1.25129205654939, -1.28211994091418, -1.30985908604087, -1.33444266589917, -1.35581145649195, -1.37391397853099, -1.38870662145509, -1.40015374849177, -1.40822778250935, -1.41290927245269, 1.41418694020244, 1.4120577077451, 1.40652670458816, 1.39760725540267, 1.38532084792299, 1.36969708118088, 1.35077359419891, 1.32859597531472, 1.30321765235473, 1.27469976392184, 1.24311101210715, 1.20852749698062, 1.17103253325936, 1.13071644959515, 1.0876763709648, 1.04201598468759, 0.993845290633451, 0.943280336223609, 0.890442936862206, 0.835460382472311, 0.778465130843354, 0.719594488528702, 0.658990280062152, 0.596798506290201, 0.533168992643221, 0.468255028192876, 0.402212996365318, 0.335201998199818, 0.267383469060427, 0.198920789724042, 0.12997889278181, 0.060723865302061, }; static const FFTComplex __attribute__((aligned(16))) tcos_analysis_ld[16] = { { 1.414107, 0.017355 }, { 1.405597, 0.155878 }, { 1.383550, 0.292900 }, { 1.348178, 0.427101 }, { 1.299823, 0.557189 }, { 1.238950, 0.681911 }, { 1.166145, 0.800066 }, { 1.082110, 0.910515 }, { 0.987653, 1.012196 }, { 0.883685, 1.104129 }, { 0.771206, 1.185429 }, { 0.651300, 1.255312 }, { 0.525122, 1.313106 }, { 0.393886, 1.358254 }, { 0.258857, 1.390321 }, { 0.121336, 1.408999 }, }; static void vector_fmul_reverse_sse(float *dst, const float *src0, const float *src1, int len){ uint64_t i = len*4-32; __asm__ volatile( "1: \n\t" "movaps 16(%1), %%xmm0 \n\t" "movaps (%1), %%xmm1 \n\t" "shufps $0x1b, %%xmm0, %%xmm0 \n\t" "shufps $0x1b, %%xmm1, %%xmm1 \n\t" "mulps (%3,%0), %%xmm0 \n\t" "mulps 16(%3,%0), %%xmm1 \n\t" "movaps %%xmm0, (%2,%0) \n\t" "movaps %%xmm1, 16(%2,%0) \n\t" "add $32, %1 \n\t" "sub $32, %0 \n\t" "jge 1b \n\t" :"+r"(i), "+r"(src1) :"r"(dst), "r"(src0)); } static void sbr_sum64x5_sse(float *z){ uint64_t i = 0; __asm__ volatile( "1: \n\t" "movaps (%1,%0), %%xmm0 \n\t" "addps 256(%1,%0), %%xmm0 \n\t" "addps 512(%1,%0), %%xmm0 \n\t" "addps 768(%1,%0), %%xmm0 \n\t" "addps 1024(%1,%0), %%xmm0 \n\t" "movaps %%xmm0, (%1,%0) \n\t" "add $16, %0 \n\t" "cmp $1024, %0 \n\t" "jne 1b \n\t" :"+r"(i) :"r"(z)); } void sbr_mdct_postrotate_sse(FFTComplex *z, unsigned int n, const float *tcos) { long j, k = n, n2 = k >> 1, n4 = k >> 2; /* post rotation + reinterleave + reorder */ #define CMUL(j,xmm0,xmm1)\ "movups (%2,"#j",2), "#xmm1"\n"\ "movups 16(%2,"#j",2), %%xmm2 \n"\ "movaps "#xmm1", "#xmm0"\n"\ "shufps $0xdd, %%xmm2, "#xmm0"\n"\ "shufps $0x88, %%xmm2, "#xmm1"\n"\ "movaps "#xmm1",%%xmm6 \n"\ "movaps "#xmm0",%%xmm7 \n"\ "mulps (%3,"#j"), %%xmm6 \n"\ "mulps (%3,"#j"), %%xmm7 \n"\ "mulps (%4,"#j"), "#xmm0"\n"\ "mulps (%4,"#j"), "#xmm1"\n"\ "addps %%xmm6, "#xmm0"\n"\ "subps %%xmm7, "#xmm1"\n" j = -n2; k = n2-16; __asm__ volatile( "1: \n" CMUL(%0, %%xmm0, %%xmm1) CMUL(%1, %%xmm4, %%xmm5) "shufps $0x1b, %%xmm1, %%xmm1 \n" "shufps $0x1b, %%xmm5, %%xmm5 \n" "movaps %%xmm4, %%xmm3 \n" "unpckhps %%xmm1, %%xmm3 \n" "unpcklps %%xmm1, %%xmm4 \n" "movaps %%xmm0, %%xmm2 \n" "unpckhps %%xmm5, %%xmm2 \n" "unpcklps %%xmm5, %%xmm0 \n" "movups %%xmm4, (%2,%1,2) \n" "movups %%xmm3, 16(%2,%1,2) \n" "movups %%xmm0, (%2,%0,2) \n" "movups %%xmm2, 16(%2,%0,2) \n" "sub $16, %1 \n" "add $16, %0 \n" "jl 1b \n" :"+&r"(j), "+&r"(k) :"r"(z+n4), "r"(tcos+n4), "r"(tcos+3*n4) :"memory"); } #if 0 void aac_sbr_init(void) { unsigned int n; for (n = 0; n < 320; n++) { sbr_qmf_window_ds[n] = sbr_qmf_window_us[2*n]; } for (n = 0; n < 320; n++) { sbr_ldqmf_window_ds[n] = 0.5f * (sbr_ldqmf_window_us[2*n+1] - sbr_ldqmf_window_us[2*n]) + sbr_ldqmf_window_us[2*n]; } } #endif /** * Analysis QMF Bank (14496-3 sp04 p206) * * @param x pointer to the beginning of the first sample window * @param W array of complex-valued samples split into subbands */ static const uint8_t revtab32_fwd[32] = { 0, 8, 4, 13, 1, 9, 7, 14, 2, 10, 6, 15, 3, 11, 5, 12, 16, 28, 20, 25, 17, 31, 23, 26, 18, 30, 22, 27, 19, 29, 21, 24, }; void sbr_qmf_analysis(void *mdct, float *sbr_qmf_window_ds, float *x, float z[320], FFTComplex W[2][32][32], uint32_t buf_idx) { unsigned int i, j, k; //vector_fmul_copy_sse(x, x+1024, 288); //vector_fmul_copy_sse(x+288, in, 1024); for (i = 0; i < 32; i++) { // numTimeSlots*RATE = 16*2 vector_fmul_reverse_sse(z, sbr_qmf_window_ds, x, 320); sbr_sum64x5_sse(z); sbr_qmf_pre_shuffle_sse(z); /* pre rotation */ for(k = 0; k < 16; k++) { j=revtab32_fwd[k]; /* this is just input * conj(tcos) */ z[2*j+0] = (z[128 - 4*k - 1] * tcos_analysis[32+2*k] + z[64+4*k ] * tcos_analysis[ 2*k]); z[2*j+1] = (z[128 - 4*k - 1] * tcos_analysis[ 2*k] - z[64+4*k ] * tcos_analysis[32+2*k]); j=revtab32_fwd[k+16]; /* this is just input * conj(tcos) */ z[2*j+0] = (z[128 - 4*k - 3] * tcos_analysis[32+2*k+1] + z[64+4*k+2] * tcos_analysis[ 2*k+1]); z[2*j+1] = (z[128 - 4*k - 3] * tcos_analysis[ 2*k+1] - z[64+4*k+2] * tcos_analysis[32+2*k+1]); } fft32_interleave_sse((FFTComplex*)z); sbr_mdct_postrotate_sse((FFTComplex*)z, 64, tcos_analysis); for (k = 0; k < 32; k++) { W[buf_idx][i][k].re = z[k]; W[buf_idx][i][k].im = z[63-k]; } x += 32; } }