From: Mans Rullgard Date: Thu, 26 Jun 2008 18:37:40 +0000 (+0100) Subject: ARM: NEON optimised simple_idct X-Git-Url: http://git.mansr.com/?p=ffmpeg.mru;a=commitdiff_plain;h=215b9eaa8cf0195908c92f373c018320736ec106 ARM: NEON optimised simple_idct --- diff --git a/libavcodec/Makefile b/libavcodec/Makefile index 27746df..7fa02fa 100644 --- a/libavcodec/Makefile +++ b/libavcodec/Makefile @@ -436,6 +436,7 @@ ASM_OBJS-$(HAVE_ARMV6) += armv4l/simple_idct_armv6.o \ OBJS-$(HAVE_NEON) += armv4l/dsputil_neon.o \ ASM_OBJS-$(HAVE_NEON) += armv4l/dsputil_neon_s.o \ + armv4l/simple_idct_neon.o \ OBJS-$(HAVE_VIS) += sparc/dsputil_vis.o \ sparc/simple_idct_vis.o \ diff --git a/libavcodec/armv4l/dsputil_arm.c b/libavcodec/armv4l/dsputil_arm.c index 89b51e7..942c0de 100644 --- a/libavcodec/armv4l/dsputil_arm.c +++ b/libavcodec/armv4l/dsputil_arm.c @@ -43,6 +43,12 @@ extern void ff_simple_idct_put_armv6(uint8_t *dest, int line_size, extern void ff_simple_idct_add_armv6(uint8_t *dest, int line_size, DCTELEM *data); +extern void ff_simple_idct_neon(DCTELEM *data); +extern void ff_simple_idct_put_neon(uint8_t *dest, int line_size, + DCTELEM *data); +extern void ff_simple_idct_add_neon(uint8_t *dest, int line_size, + DCTELEM *data); + /* XXX: local hack */ static void (*ff_put_pixels_clamped)(const DCTELEM *block, uint8_t *pixels, int line_size); static void (*ff_add_pixels_clamped)(const DCTELEM *block, uint8_t *pixels, int line_size); @@ -233,6 +239,8 @@ void dsputil_init_armv4l(DSPContext* c, AVCodecContext *avctx) if(idct_algo == FF_IDCT_AUTO){ #if defined(HAVE_IPP) idct_algo = FF_IDCT_IPP; +#elif defined(HAVE_NEON) + idct_algo = FF_IDCT_SIMPLENEON; #elif defined(HAVE_ARMV6) idct_algo = FF_IDCT_SIMPLEARMV6; #elif defined(HAVE_ARMV5TE) @@ -273,6 +281,13 @@ void dsputil_init_armv4l(DSPContext* c, AVCodecContext *avctx) c->idct = simple_idct_ipp; c->idct_permutation_type= FF_NO_IDCT_PERM; #endif +#ifdef HAVE_NEON + } else if (idct_algo==FF_IDCT_SIMPLENEON){ + c->idct_put= ff_simple_idct_put_neon; + c->idct_add= ff_simple_idct_add_neon; + c->idct = ff_simple_idct_neon; + c->idct_permutation_type = FF_PARTTRANS_IDCT_PERM; +#endif } } diff --git a/libavcodec/armv4l/simple_idct_neon.S b/libavcodec/armv4l/simple_idct_neon.S new file mode 100644 index 0000000..44701f8 --- /dev/null +++ b/libavcodec/armv4l/simple_idct_neon.S @@ -0,0 +1,411 @@ +/* + * ARM NEON IDCT + * + * Copyright (c) 2008 Mans Rullgard + * + * Based on Simple IDCT + * Copyright (c) 2001 Michael Niedermayer + * + * This file is part of FFmpeg. + * + * FFmpeg is free software; you can redistribute it and/or + * modify it under the terms of the GNU Lesser General Public + * License as published by the Free Software Foundation; either + * version 2.1 of the License, or (at your option) any later version. + * + * FFmpeg is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + * Lesser General Public License for more details. + * + * You should have received a copy of the GNU Lesser General Public + * License along with FFmpeg; if not, write to the Free Software + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA + */ + +#define W1 22725 //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5 +#define W2 21407 //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5 +#define W3 19266 //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5 +#define W4 16383 //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5 +#define W5 12873 //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5 +#define W6 8867 //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5 +#define W7 4520 //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5 +#define W4c ((1<<(COL_SHIFT-1))/W4) +#define ROW_SHIFT 11 +#define COL_SHIFT 20 + +#define w1 d0[0] +#define w2 d0[1] +#define w3 d0[2] +#define w4 d0[3] +#define w5 d1[0] +#define w6 d1[1] +#define w7 d1[2] +#define w4c d1[3] + + .fpu neon + + .macro idct_col4_top + vmull.s16 q7, d6, w2 /* q9 = W2 * col[2] */ + vmull.s16 q8, d6, w6 /* q10 = W6 * col[2] */ + vmull.s16 q9, d4, w1 /* q9 = W1 * col[1] */ + vadd.i32 q11, q15, q7 + vmull.s16 q10, d4, w3 /* q10 = W3 * col[1] */ + vadd.i32 q12, q15, q8 + vmull.s16 q5, d4, w5 /* q5 = W5 * col[1] */ + vsub.i32 q13, q15, q8 + vmull.s16 q6, d4, w7 /* q6 = W7 * col[1] */ + vsub.i32 q14, q15, q7 + + vmlal.s16 q9, d8, w3 /* q9 += W3 * col[3] */ + vmlsl.s16 q10, d8, w7 /* q10 -= W7 * col[3] */ + vmlsl.s16 q5, d8, w1 /* q5 -= W1 * col[3] */ + vmlsl.s16 q6, d8, w5 /* q6 -= W5 * col[3] */ + .endm + + .text + .align + .type idct_row4_neon, %function + .func idct_row4_neon +idct_row4_neon: + vmov.i32 q15, #(1<<(ROW_SHIFT-1)) + vld1.64 {d2-d5}, [a3,:128]! + vmlal.s16 q15, d2, w4 /* q15 += W4 * col[0] */ + vld1.64 {d6,d7}, [a3,:128]! + vorr d10, d3, d5 + vld1.64 {d8,d9}, [a3,:128]! + add a3, a3, #-64 + + vorr d11, d7, d9 + vorr d10, d10, d11 + vmov a4, v1, d10 + + idct_col4_top + + orrs a4, a4, v1 + beq 1f + + vmull.s16 q7, d3, w4 /* q7 = W4 * col[4] */ + vmlal.s16 q9, d5, w5 /* q9 += W5 * col[5] */ + vmlsl.s16 q10, d5, w1 /* q10 -= W1 * col[5] */ + vmull.s16 q8, d7, w2 /* q8 = W2 * col[6] */ + vmlal.s16 q5, d5, w7 /* q5 += W7 * col[5] */ + vadd.i32 q11, q11, q7 + vsub.i32 q12, q12, q7 + vsub.i32 q13, q13, q7 + vadd.i32 q14, q14, q7 + vmlal.s16 q6, d5, w3 /* q6 += W3 * col[5] */ + vmull.s16 q7, d7, w6 /* q7 = W6 * col[6] */ + vmlal.s16 q9, d9, w7 + vmlsl.s16 q10, d9, w5 + vmlal.s16 q5, d9, w3 + vmlsl.s16 q6, d9, w1 + vadd.i32 q11, q11, q7 + vsub.i32 q12, q12, q8 + vadd.i32 q13, q13, q8 + vsub.i32 q14, q14, q7 + +1: vadd.i32 q3, q11, q9 + vadd.i32 q4, q12, q10 + vshrn.i32 d2, q3, #ROW_SHIFT + vshrn.i32 d4, q4, #ROW_SHIFT + vadd.i32 q7, q13, q5 + vadd.i32 q8, q14, q6 + vtrn.16 d2, d4 + vshrn.i32 d6, q7, #ROW_SHIFT + vshrn.i32 d8, q8, #ROW_SHIFT + vsub.i32 q14, q14, q6 + vsub.i32 q11, q11, q9 + vtrn.16 d6, d8 + vsub.i32 q13, q13, q5 + vshrn.i32 d3, q14, #ROW_SHIFT + vtrn.32 d2, d6 + vsub.i32 q12, q12, q10 + vtrn.32 d4, d8 + vshrn.i32 d5, q13, #ROW_SHIFT + vshrn.i32 d7, q12, #ROW_SHIFT + vshrn.i32 d9, q11, #ROW_SHIFT + + vtrn.16 d3, d5 + vtrn.16 d7, d9 + vtrn.32 d3, d7 + vtrn.32 d5, d9 + + vst1.64 {d2-d5}, [a3,:128]! + vst1.64 {d6-d9}, [a3,:128]! + + bx lr + .endfunc + + .align + .type idct_col4_neon, %function + .func idct_col4_neon +idct_col4_neon: + mov ip, #16 + vld1.64 {d2}, [a3,:64], ip /* d2 = col[0] */ + vdup.16 d30, w4c + vld1.64 {d4}, [a3,:64], ip /* d3 = col[1] */ + vadd.i16 d30, d30, d2 + vld1.64 {d6}, [a3,:64], ip /* d4 = col[2] */ + vmull.s16 q15, d30, w4 /* q15 = W4*(col[0]+(1<