yading@10: /* yading@10: * Copyright (c) 2011 Mans Rullgard yading@10: * yading@10: * This file is part of Libav. yading@10: * yading@10: * Libav is free software; you can redistribute it and/or yading@10: * modify it under the terms of the GNU Lesser General Public yading@10: * License as published by the Free Software Foundation; either yading@10: * version 2.1 of the License, or (at your option) any later version. yading@10: * yading@10: * Libav is distributed in the hope that it will be useful, yading@10: * but WITHOUT ANY WARRANTY; without even the implied warranty of yading@10: * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU yading@10: * Lesser General Public License for more details. yading@10: * yading@10: * You should have received a copy of the GNU Lesser General Public yading@10: * License along with Libav; if not, write to the Free Software yading@10: * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA yading@10: */ yading@10: yading@10: #include "libavutil/arm/asm.S" yading@10: yading@10: .macro bflies d0, d1, r0, r1 yading@10: vrev64.32 \r0, \d1 @ t5, t6, t1, t2 yading@10: vhsub.s16 \r1, \d1, \r0 @ t1-t5, t2-t6, t5-t1, t6-t2 yading@10: vhadd.s16 \r0, \d1, \r0 @ t1+t5, t2+t6, t5+t1, t6+t2 yading@10: vext.16 \r1, \r1, \r1, #1 @ t2-t6, t5-t1, t6-t2, t1-t5 yading@10: vtrn.32 \r0, \r1 @ t1+t5, t2+t6, t2-t6, t5-t1 yading@10: @ t5, t6, t4, t3 yading@10: vhsub.s16 \d1, \d0, \r0 yading@10: vhadd.s16 \d0, \d0, \r0 yading@10: .endm yading@10: yading@10: .macro transform01 q0, q1, d3, c0, c1, r0, w0, w1 yading@10: vrev32.16 \r0, \d3 yading@10: vmull.s16 \w0, \d3, \c0 yading@10: vmlal.s16 \w0, \r0, \c1 yading@10: vshrn.s32 \d3, \w0, #15 yading@10: bflies \q0, \q1, \w0, \w1 yading@10: .endm yading@10: yading@10: .macro transform2 d0, d1, d2, d3, q0, q1, c0, c1, c2, c3, \ yading@10: r0, r1, w0, w1 yading@10: vrev32.16 \r0, \d1 yading@10: vrev32.16 \r1, \d3 yading@10: vmull.s16 \w0, \d1, \c0 yading@10: vmlal.s16 \w0, \r0, \c1 yading@10: vmull.s16 \w1, \d3, \c2 yading@10: vmlal.s16 \w1, \r1, \c3 yading@10: vshrn.s32 \d1, \w0, #15 yading@10: vshrn.s32 \d3, \w1, #15 yading@10: bflies \q0, \q1, \w0, \w1 yading@10: .endm yading@10: yading@10: .macro fft4 d0, d1, r0, r1 yading@10: vhsub.s16 \r0, \d0, \d1 @ t3, t4, t8, t7 yading@10: vhsub.s16 \r1, \d1, \d0 yading@10: vhadd.s16 \d0, \d0, \d1 @ t1, t2, t6, t5 yading@10: vmov.i64 \d1, #0xffff00000000 yading@10: vbit \r0, \r1, \d1 yading@10: vrev64.16 \r1, \r0 @ t7, t8, t4, t3 yading@10: vtrn.32 \r0, \r1 @ t3, t4, t7, t8 yading@10: vtrn.32 \d0, \r0 @ t1, t2, t3, t4, t6, t5, t8, t7 yading@10: vhsub.s16 \d1, \d0, \r0 @ r2, i2, r3, i1 yading@10: vhadd.s16 \d0, \d0, \r0 @ r0, i0, r1, i3 yading@10: .endm yading@10: yading@10: .macro fft8 d0, d1, d2, d3, q0, q1, c0, c1, r0, r1, w0, w1 yading@10: fft4 \d0, \d1, \r0, \r1 yading@10: vtrn.32 \d0, \d1 @ z0, z2, z1, z3 yading@10: vhadd.s16 \r0, \d2, \d3 @ t1, t2, t3, t4 yading@10: vhsub.s16 \d3, \d2, \d3 @ z5, z7 yading@10: vmov \d2, \r0 yading@10: transform01 \q0, \q1, \d3, \c0, \c1, \r0, \w0, \w1 yading@10: .endm yading@10: yading@10: function fft4_neon yading@10: vld1.16 {d0-d1}, [r0] yading@10: fft4 d0, d1, d2, d3 yading@10: vst1.16 {d0-d1}, [r0] yading@10: bx lr yading@10: endfunc yading@10: yading@10: function fft8_neon yading@10: vld1.16 {d0-d3}, [r0,:128] yading@10: movrel r1, coefs yading@10: vld1.16 {d30}, [r1,:64] yading@10: vdup.16 d31, d30[0] yading@10: fft8 d0, d1, d2, d3, q0, q1, d31, d30, d20, d21, q8, q9 yading@10: vtrn.32 d0, d1 yading@10: vtrn.32 d2, d3 yading@10: vst1.16 {d0-d3}, [r0,:128] yading@10: bx lr yading@10: endfunc yading@10: yading@10: function fft16_neon yading@10: vld1.16 {d0-d3}, [r0,:128]! yading@10: vld1.16 {d4-d7}, [r0,:128] yading@10: movrel r1, coefs yading@10: sub r0, r0, #32 yading@10: vld1.16 {d28-d31},[r1,:128] yading@10: vdup.16 d31, d28[0] yading@10: fft8 d0, d1, d2, d3, q0, q1, d31, d28, d20, d21, q8, q9 yading@10: vswp d5, d6 yading@10: fft4 q2, q3, q8, q9 yading@10: vswp d5, d6 yading@10: vtrn.32 q0, q1 @ z0, z4, z2, z6, z1, z5, z3, z7 yading@10: vtrn.32 q2, q3 @ z8, z12,z10,z14,z9, z13,z11,z15 yading@10: vswp d1, d2 yading@10: vdup.16 d31, d28[0] yading@10: transform01 q0, q2, d5, d31, d28, d20, q8, q9 yading@10: vdup.16 d26, d29[0] yading@10: vdup.16 d27, d30[0] yading@10: transform2 d2, d6, d3, d7, q1, q3, d26, d30, d27, d29, \ yading@10: d20, d21, q8, q9 yading@10: vtrn.32 q0, q1 yading@10: vtrn.32 q2, q3 yading@10: vst1.16 {d0-d3}, [r0,:128]! yading@10: vst1.16 {d4-d7}, [r0,:128] yading@10: bx lr yading@10: endfunc yading@10: yading@10: function fft_pass_neon yading@10: push {r4,lr} yading@10: movrel lr, coefs+24 yading@10: vld1.16 {d30}, [lr,:64] yading@10: lsl r12, r2, #3 yading@10: vmov d31, d30 yading@10: add r3, r1, r2, lsl #2 yading@10: mov lr, #-8 yading@10: sub r3, r3, #2 yading@10: mov r4, r0 yading@10: vld1.16 {d27[]}, [r3,:16] yading@10: sub r3, r3, #6 yading@10: vld1.16 {q0}, [r4,:128], r12 yading@10: vld1.16 {q1}, [r4,:128], r12 yading@10: vld1.16 {q2}, [r4,:128], r12 yading@10: vld1.16 {q3}, [r4,:128], r12 yading@10: vld1.16 {d28}, [r1,:64]! yading@10: vld1.16 {d29}, [r3,:64], lr yading@10: vswp d1, d2 yading@10: vswp d5, d6 yading@10: vtrn.32 d0, d1 yading@10: vtrn.32 d4, d5 yading@10: vdup.16 d25, d28[1] yading@10: vmul.s16 d27, d27, d31 yading@10: transform01 q0, q2, d5, d25, d27, d20, q8, q9 yading@10: b 2f yading@10: 1: yading@10: mov r4, r0 yading@10: vdup.16 d26, d29[0] yading@10: vld1.16 {q0}, [r4,:128], r12 yading@10: vld1.16 {q1}, [r4,:128], r12 yading@10: vld1.16 {q2}, [r4,:128], r12 yading@10: vld1.16 {q3}, [r4,:128], r12 yading@10: vld1.16 {d28}, [r1,:64]! yading@10: vld1.16 {d29}, [r3,:64], lr yading@10: vswp d1, d2 yading@10: vswp d5, d6 yading@10: vtrn.32 d0, d1 yading@10: vtrn.32 d4, d5 yading@10: vdup.16 d24, d28[0] yading@10: vdup.16 d25, d28[1] yading@10: vdup.16 d27, d29[3] yading@10: vmul.s16 q13, q13, q15 yading@10: transform2 d0, d4, d1, d5, q0, q2, d24, d26, d25, d27, \ yading@10: d16, d17, q9, q10 yading@10: 2: yading@10: vtrn.32 d2, d3 yading@10: vtrn.32 d6, d7 yading@10: vdup.16 d24, d28[2] yading@10: vdup.16 d26, d29[2] yading@10: vdup.16 d25, d28[3] yading@10: vdup.16 d27, d29[1] yading@10: vmul.s16 q13, q13, q15 yading@10: transform2 d2, d6, d3, d7, q1, q3, d24, d26, d25, d27, \ yading@10: d16, d17, q9, q10 yading@10: vtrn.32 d0, d1 yading@10: vtrn.32 d2, d3 yading@10: vtrn.32 d4, d5 yading@10: vtrn.32 d6, d7 yading@10: vswp d1, d2 yading@10: vswp d5, d6 yading@10: mov r4, r0 yading@10: vst1.16 {q0}, [r4,:128], r12 yading@10: vst1.16 {q1}, [r4,:128], r12 yading@10: vst1.16 {q2}, [r4,:128], r12 yading@10: vst1.16 {q3}, [r4,:128], r12 yading@10: add r0, r0, #16 yading@10: subs r2, r2, #2 yading@10: bgt 1b yading@10: pop {r4,pc} yading@10: endfunc yading@10: yading@10: #define F_SQRT1_2 23170 yading@10: #define F_COS_16_1 30274 yading@10: #define F_COS_16_3 12540 yading@10: yading@10: const coefs, align=4 yading@10: .short F_SQRT1_2, -F_SQRT1_2, -F_SQRT1_2, F_SQRT1_2 yading@10: .short F_COS_16_1,-F_COS_16_1,-F_COS_16_1, F_COS_16_1 yading@10: .short F_COS_16_3,-F_COS_16_3,-F_COS_16_3, F_COS_16_3 yading@10: .short 1, -1, -1, 1 yading@10: endconst yading@10: yading@10: .macro def_fft n, n2, n4 yading@10: function fft\n\()_neon yading@10: push {r4, lr} yading@10: mov r4, r0 yading@10: bl fft\n2\()_neon yading@10: add r0, r4, #\n4*2*4 yading@10: bl fft\n4\()_neon yading@10: add r0, r4, #\n4*3*4 yading@10: bl fft\n4\()_neon yading@10: mov r0, r4 yading@10: pop {r4, lr} yading@10: movrelx r1, X(ff_cos_\n\()_fixed) yading@10: mov r2, #\n4/2 yading@10: b fft_pass_neon yading@10: endfunc yading@10: .endm yading@10: yading@10: def_fft 32, 16, 8 yading@10: def_fft 64, 32, 16 yading@10: def_fft 128, 64, 32 yading@10: def_fft 256, 128, 64 yading@10: def_fft 512, 256, 128 yading@10: def_fft 1024, 512, 256 yading@10: def_fft 2048, 1024, 512 yading@10: def_fft 4096, 2048, 1024 yading@10: def_fft 8192, 4096, 2048 yading@10: def_fft 16384, 8192, 4096 yading@10: def_fft 32768, 16384, 8192 yading@10: def_fft 65536, 32768, 16384 yading@10: yading@10: function ff_fft_fixed_calc_neon, export=1 yading@10: ldr r2, [r0] yading@10: sub r2, r2, #2 yading@10: movrel r3, fft_fixed_tab_neon yading@10: ldr r3, [r3, r2, lsl #2] yading@10: mov r0, r1 yading@10: bx r3 yading@10: endfunc yading@10: yading@10: const fft_fixed_tab_neon yading@10: .word fft4_neon yading@10: .word fft8_neon yading@10: .word fft16_neon yading@10: .word fft32_neon yading@10: .word fft64_neon yading@10: .word fft128_neon yading@10: .word fft256_neon yading@10: .word fft512_neon yading@10: .word fft1024_neon yading@10: .word fft2048_neon yading@10: .word fft4096_neon yading@10: .word fft8192_neon yading@10: .word fft16384_neon yading@10: .word fft32768_neon yading@10: .word fft65536_neon yading@10: endconst