* commit 'e3fcb14347466095839c2a3c47ebecff02da891e': dsputil: Split off IDCT bits into their own context Conflicts: configure libavcodec/aic.c libavcodec/arm/Makefile libavcodec/arm/dsputil_init_arm.c libavcodec/arm/dsputil_init_armv6.c libavcodec/asvdec.c libavcodec/dnxhdenc.c libavcodec/dsputil.c libavcodec/dvdec.c libavcodec/dxva2_mpeg2.c libavcodec/intrax8.c libavcodec/mdec.c libavcodec/mjpegdec.c libavcodec/mjpegenc_common.h libavcodec/mpegvideo.c libavcodec/ppc/dsputil_altivec.h libavcodec/ppc/dsputil_ppc.c libavcodec/ppc/idctdsp.c libavcodec/x86/Makefile libavcodec/x86/dsputil_init.c libavcodec/x86/dsputil_mmx.c libavcodec/x86/dsputil_x86.h
Showing the changes to libavcodec/arm (13 of 84 files). Show all
Unified · Split
| @@ -6,10 +6,7 @@ OBJS-$(CONFIG_AC3DSP) += arm/ac3dsp_init_arm.o \ · 5 unmodified lines | |||
| 6 | arm/ac3dsp_arm.o | 6 | arm/ac3dsp_arm.o |
| 7 | OBJS-$(CONFIG_AUDIODSP) += arm/audiodsp_init_arm.o | 7 | OBJS-$(CONFIG_AUDIODSP) += arm/audiodsp_init_arm.o |
| 8 | OBJS-$(CONFIG_BLOCKDSP) += arm/blockdsp_init_arm.o | 8 | OBJS-$(CONFIG_BLOCKDSP) += arm/blockdsp_init_arm.o |
| 9 | OBJS-$(CONFIG_DSPUTIL) += arm/dsputil_init_arm.o \ | 9 | OBJS-$(CONFIG_DSPUTIL) += arm/dsputil_init_arm.o |
| 10 | arm/dsputil_arm.o \ | ||
| 11 | arm/jrevdct_arm.o \ | ||
| 12 | arm/simple_idct_arm.o | ||
| 13 | OBJS-$(CONFIG_FFT) += arm/fft_init_arm.o \ | 10 | OBJS-$(CONFIG_FFT) += arm/fft_init_arm.o \ |
| 14 | arm/fft_fixed_init_arm.o | 11 | arm/fft_fixed_init_arm.o |
| 15 | OBJS-$(CONFIG_H264CHROMA) += arm/h264chroma_init_arm.o | 12 | OBJS-$(CONFIG_H264CHROMA) += arm/h264chroma_init_arm.o |
| @@ -18,6 +15,10 @@ OBJS-$(CONFIG_H264PRED) += arm/h264pred_init_arm.o · 2 unmodified lines | |||
| 18 | OBJS-$(CONFIG_H264QPEL) += arm/h264qpel_init_arm.o | 15 | OBJS-$(CONFIG_H264QPEL) += arm/h264qpel_init_arm.o |
| 19 | OBJS-$(CONFIG_HPELDSP) += arm/hpeldsp_init_arm.o \ | 16 | OBJS-$(CONFIG_HPELDSP) += arm/hpeldsp_init_arm.o \ |
| 20 | arm/hpeldsp_arm.o | 17 | arm/hpeldsp_arm.o |
| 18 | OBJS-$(CONFIG_IDCTDSP) += arm/idctdsp_init_arm.o \ | ||
| 19 | arm/idctdsp_arm.o \ | ||
| 20 | arm/jrevdct_arm.o \ | ||
| 21 | arm/simple_idct_arm.o | ||
| 21 | OBJS-$(CONFIG_LLAUDDSP) += arm/lossless_audiodsp_init_arm.o | 22 | OBJS-$(CONFIG_LLAUDDSP) += arm/lossless_audiodsp_init_arm.o |
| 22 | OBJS-$(CONFIG_MPEGAUDIODSP) += arm/mpegaudiodsp_init_arm.o | 23 | OBJS-$(CONFIG_MPEGAUDIODSP) += arm/mpegaudiodsp_init_arm.o |
| 23 | OBJS-$(CONFIG_MPEGVIDEO) += arm/mpegvideo_arm.o | 24 | OBJS-$(CONFIG_MPEGVIDEO) += arm/mpegvideo_arm.o |
| @@ -41,7 +42,7 @@ OBJS-$(CONFIG_RV30_DECODER) += arm/rv34dsp_init_arm.o · 17 unmodified lines | |||
| 41 | OBJS-$(CONFIG_RV40_DECODER) += arm/rv34dsp_init_arm.o \ | 42 | OBJS-$(CONFIG_RV40_DECODER) += arm/rv34dsp_init_arm.o \ |
| 42 | arm/rv40dsp_init_arm.o | 43 | arm/rv40dsp_init_arm.o |
| 43 | 44 | ||
| 44 | ARMV5TE-OBJS-$(CONFIG_DSPUTIL) += arm/dsputil_init_armv5te.o \ | 45 | ARMV5TE-OBJS-$(CONFIG_IDCTDSP) += arm/idctdsp_init_armv5te.o \ |
| 45 | arm/simple_idct_armv5te.o | 46 | arm/simple_idct_armv5te.o |
| 46 | ARMV5TE-OBJS-$(CONFIG_MPEGVIDEO) += arm/mpegvideo_armv5te.o \ | 47 | ARMV5TE-OBJS-$(CONFIG_MPEGVIDEO) += arm/mpegvideo_armv5te.o \ |
| 47 | arm/mpegvideo_armv5te_s.o | 48 | arm/mpegvideo_armv5te_s.o |
| @@ -52,11 +53,13 @@ ARMV5TE-OBJS-$(CONFIG_MLP_DECODER) += arm/mlpdsp_armv5te.o · 4 unmodified lines | |||
| 52 | 53 | ||
| 53 | ARMV6-OBJS-$(CONFIG_AC3DSP) += arm/ac3dsp_armv6.o | 54 | ARMV6-OBJS-$(CONFIG_AC3DSP) += arm/ac3dsp_armv6.o |
| 54 | ARMV6-OBJS-$(CONFIG_DSPUTIL) += arm/dsputil_init_armv6.o \ | 55 | ARMV6-OBJS-$(CONFIG_DSPUTIL) += arm/dsputil_init_armv6.o \ |
| 55 | arm/dsputil_armv6.o \ | 56 | arm/dsputil_armv6.o |
| 56 | arm/simple_idct_armv6.o | ||
| 57 | ARMV6-OBJS-$(CONFIG_H264DSP) += arm/startcode_armv6.o | 57 | ARMV6-OBJS-$(CONFIG_H264DSP) += arm/startcode_armv6.o |
| 58 | ARMV6-OBJS-$(CONFIG_HPELDSP) += arm/hpeldsp_init_armv6.o \ | 58 | ARMV6-OBJS-$(CONFIG_HPELDSP) += arm/hpeldsp_init_armv6.o \ |
| 59 | arm/hpeldsp_armv6.o | 59 | arm/hpeldsp_armv6.o |
| 60 | ARMV6-OBJS-$(CONFIG_IDCTDSP) += arm/idctdsp_init_armv6.o \ | ||
| 61 | arm/idctdsp_armv6.o \ | ||
| 62 | arm/simple_idct_armv6.o | ||
| 60 | ARMV6-OBJS-$(CONFIG_MPEGAUDIODSP) += arm/mpegaudiodsp_fixed_armv6.o | 63 | ARMV6-OBJS-$(CONFIG_MPEGAUDIODSP) += arm/mpegaudiodsp_fixed_armv6.o |
| 61 | 64 | ||
| 62 | ARMV6-OBJS-$(CONFIG_MLP_DECODER) += arm/mlpdsp_armv6.o | 65 | ARMV6-OBJS-$(CONFIG_MLP_DECODER) += arm/mlpdsp_armv6.o |
| @@ -86,9 +89,6 @@ NEON-OBJS-$(CONFIG_AUDIODSP) += arm/audiodsp_init_neon.o \ · 23 unmodified lines | |||
| 86 | arm/int_neon.o | 89 | arm/int_neon.o |
| 87 | NEON-OBJS-$(CONFIG_BLOCKDSP) += arm/blockdsp_init_neon.o \ | 90 | NEON-OBJS-$(CONFIG_BLOCKDSP) += arm/blockdsp_init_neon.o \ |
| 88 | arm/blockdsp_neon.o | 91 | arm/blockdsp_neon.o |
| 89 | NEON-OBJS-$(CONFIG_DSPUTIL) += arm/dsputil_init_neon.o \ | ||
| 90 | arm/dsputil_neon.o \ | ||
| 91 | arm/simple_idct_neon.o | ||
| 92 | NEON-OBJS-$(CONFIG_FFT) += arm/fft_neon.o \ | 92 | NEON-OBJS-$(CONFIG_FFT) += arm/fft_neon.o \ |
| 93 | arm/fft_fixed_neon.o | 93 | arm/fft_fixed_neon.o |
| 94 | NEON-OBJS-$(CONFIG_H264CHROMA) += arm/h264cmc_neon.o | 94 | NEON-OBJS-$(CONFIG_H264CHROMA) += arm/h264cmc_neon.o |
| @@ -99,6 +99,9 @@ NEON-OBJS-$(CONFIG_H264QPEL) += arm/h264qpel_neon.o \ · 4 unmodified lines | |||
| 99 | arm/hpeldsp_neon.o | 99 | arm/hpeldsp_neon.o |
| 100 | NEON-OBJS-$(CONFIG_HPELDSP) += arm/hpeldsp_init_neon.o \ | 100 | NEON-OBJS-$(CONFIG_HPELDSP) += arm/hpeldsp_init_neon.o \ |
| 101 | arm/hpeldsp_neon.o | 101 | arm/hpeldsp_neon.o |
| 102 | NEON-OBJS-$(CONFIG_IDCTDSP) += arm/idctdsp_init_neon.o \ | ||
| 103 | arm/idctdsp_neon.o \ | ||
| 104 | arm/simple_idct_neon.o | ||
| 102 | NEON-OBJS-$(CONFIG_MDCT) += arm/mdct_neon.o \ | 105 | NEON-OBJS-$(CONFIG_MDCT) += arm/mdct_neon.o \ |
| 103 | arm/mdct_fixed_neon.o | 106 | arm/mdct_fixed_neon.o |
| 104 | NEON-OBJS-$(CONFIG_MPEGVIDEO) += arm/mpegvideo_neon.o | 107 | NEON-OBJS-$(CONFIG_MPEGVIDEO) += arm/mpegvideo_neon.o |
| @@ -24,11 +24,7 @@ · 23 unmodified lines | |||
| 24 | #include "libavcodec/avcodec.h" | 24 | #include "libavcodec/avcodec.h" |
| 25 | #include "libavcodec/dsputil.h" | 25 | #include "libavcodec/dsputil.h" |
| 26 | 26 | ||
| 27 | void ff_dsputil_init_armv5te(DSPContext *c, AVCodecContext *avctx, | ||
| 28 | unsigned high_bit_depth); | ||
| 29 | void ff_dsputil_init_armv6(DSPContext *c, AVCodecContext *avctx, | 27 | void ff_dsputil_init_armv6(DSPContext *c, AVCodecContext *avctx, |
| 30 | unsigned high_bit_depth); | 28 | unsigned high_bit_depth); |
| 31 | void ff_dsputil_init_neon(DSPContext *c, AVCodecContext *avctx, | ||
| 32 | unsigned high_bit_depth); | ||
| 33 | 29 | ||
| 34 | #endif /* AVCODEC_ARM_DSPUTIL_ARM_H */ | 30 | #endif /* AVCODEC_ARM_DSPUTIL_ARM_H */ |
| @@ -20,33 +20,6 @@ · 19 unmodified lines | |||
| 20 | 20 | ||
| 21 | #include "libavutil/arm/asm.S" | 21 | #include "libavutil/arm/asm.S" |
| 22 | 22 | ||
| 23 | function ff_add_pixels_clamped_armv6, export=1 | ||
| 24 | push {r4-r8,lr} | ||
| 25 | mov r3, #8 | ||
| 26 | 1: | ||
| 27 | ldm r0!, {r4,r5,r12,lr} | ||
| 28 | ldrd r6, r7, [r1] | ||
| 29 | pkhbt r8, r4, r5, lsl #16 | ||
| 30 | pkhtb r5, r5, r4, asr #16 | ||
| 31 | pkhbt r4, r12, lr, lsl #16 | ||
| 32 | pkhtb lr, lr, r12, asr #16 | ||
| 33 | pld [r1, r2] | ||
| 34 | uxtab16 r8, r8, r6 | ||
| 35 | uxtab16 r5, r5, r6, ror #8 | ||
| 36 | uxtab16 r4, r4, r7 | ||
| 37 | uxtab16 lr, lr, r7, ror #8 | ||
| 38 | usat16 r8, #8, r8 | ||
| 39 | usat16 r5, #8, r5 | ||
| 40 | usat16 r4, #8, r4 | ||
| 41 | usat16 lr, #8, lr | ||
| 42 | orr r6, r8, r5, lsl #8 | ||
| 43 | orr r7, r4, lr, lsl #8 | ||
| 44 | subs r3, r3, #1 | ||
| 45 | strd_post r6, r7, r1, r2 | ||
| 46 | bgt 1b | ||
| 47 | pop {r4-r8,pc} | ||
| 48 | endfunc | ||
| 49 | |||
| 50 | function ff_get_pixels_armv6, export=1 | 23 | function ff_get_pixels_armv6, export=1 |
| 51 | pld [r1, r2] | 24 | pld [r1, r2] |
| 52 | push {r4-r8, lr} | 25 | push {r4-r8, lr} |
| @@ -28,71 +28,11 @@ · 27 unmodified lines | |||
| 28 | #include "libavcodec/dsputil.h" | 28 | #include "libavcodec/dsputil.h" |
| 29 | #include "dsputil_arm.h" | 29 | #include "dsputil_arm.h" |
| 30 | 30 | ||
| 31 | void ff_j_rev_dct_arm(int16_t *data); | ||
| 32 | void ff_simple_idct_arm(int16_t *data); | ||
| 33 | |||
| 34 | /* XXX: local hack */ | ||
| 35 | static void (*ff_put_pixels_clamped)(const int16_t *block, uint8_t *pixels, int line_size); | ||
| 36 | static void (*ff_add_pixels_clamped)(const int16_t *block, uint8_t *pixels, int line_size); | ||
| 37 | |||
| 38 | void ff_add_pixels_clamped_arm(const int16_t *block, uint8_t *dest, | ||
| 39 | int line_size); | ||
| 40 | |||
| 41 | /* XXX: those functions should be suppressed ASAP when all IDCTs are | ||
| 42 | * converted */ | ||
| 43 | static void j_rev_dct_arm_put(uint8_t *dest, int line_size, int16_t *block) | ||
| 44 | { | ||
| 45 | ff_j_rev_dct_arm(block); | ||
| 46 | ff_put_pixels_clamped(block, dest, line_size); | ||
| 47 | } | ||
| 48 | |||
| 49 | static void j_rev_dct_arm_add(uint8_t *dest, int line_size, int16_t *block) | ||
| 50 | { | ||
| 51 | ff_j_rev_dct_arm(block); | ||
| 52 | ff_add_pixels_clamped(block, dest, line_size); | ||
| 53 | } | ||
| 54 | |||
| 55 | static void simple_idct_arm_put(uint8_t *dest, int line_size, int16_t *block) | ||
| 56 | { | ||
| 57 | ff_simple_idct_arm(block); | ||
| 58 | ff_put_pixels_clamped(block, dest, line_size); | ||
| 59 | } | ||
| 60 | |||
| 61 | static void simple_idct_arm_add(uint8_t *dest, int line_size, int16_t *block) | ||
| 62 | { | ||
| 63 | ff_simple_idct_arm(block); | ||
| 64 | ff_add_pixels_clamped(block, dest, line_size); | ||
| 65 | } | ||
| 66 | |||
| 67 | av_cold void ff_dsputil_init_arm(DSPContext *c, AVCodecContext *avctx, | 31 | av_cold void ff_dsputil_init_arm(DSPContext *c, AVCodecContext *avctx, |
| 68 | unsigned high_bit_depth) | 32 | unsigned high_bit_depth) |
| 69 | { | 33 | { |
| 70 | int cpu_flags = av_get_cpu_flags(); | 34 | int cpu_flags = av_get_cpu_flags(); |
| 71 | 35 | ||
| 72 | ff_put_pixels_clamped = c->put_pixels_clamped; | ||
| 73 | ff_add_pixels_clamped = c->add_pixels_clamped; | ||
| 74 | |||
| 75 | if (!avctx->lowres && !high_bit_depth) { | ||
| 76 | if (avctx->idct_algo == FF_IDCT_AUTO || | ||
| 77 | avctx->idct_algo == FF_IDCT_ARM) { | ||
| 78 | c->idct_put = j_rev_dct_arm_put; | ||
| 79 | c->idct_add = j_rev_dct_arm_add; | ||
| 80 | c->idct = ff_j_rev_dct_arm; | ||
| 81 | c->idct_permutation_type = FF_LIBMPEG2_IDCT_PERM; | ||
| 82 | } else if (avctx->idct_algo == FF_IDCT_SIMPLEARM) { | ||
| 83 | c->idct_put = simple_idct_arm_put; | ||
| 84 | c->idct_add = simple_idct_arm_add; | ||
| 85 | c->idct = ff_simple_idct_arm; | ||
| 86 | c->idct_permutation_type = FF_NO_IDCT_PERM; | ||
| 87 | } | ||
| 88 | } | ||
| 89 | |||
| 90 | c->add_pixels_clamped = ff_add_pixels_clamped_arm; | ||
| 91 | |||
| 92 | if (have_armv5te(cpu_flags)) | ||
| 93 | ff_dsputil_init_armv5te(c, avctx, high_bit_depth); | ||
| 94 | if (have_armv6(cpu_flags)) | 36 | if (have_armv6(cpu_flags)) |
| 95 | ff_dsputil_init_armv6(c, avctx, high_bit_depth); | 37 | ff_dsputil_init_armv6(c, avctx, high_bit_depth); |
| 96 | if (have_neon(cpu_flags)) | ||
| 97 | ff_dsputil_init_neon(c, avctx, high_bit_depth); | ||
| 98 | } | 38 | } |
| @@ -26,13 +26,6 @@ · 25 unmodified lines | |||
| 26 | #include "libavcodec/mpegvideo.h" | 26 | #include "libavcodec/mpegvideo.h" |
| 27 | #include "dsputil_arm.h" | 27 | #include "dsputil_arm.h" |
| 28 | 28 | ||
| 29 | void ff_simple_idct_armv6(int16_t *data); | ||
| 30 | void ff_simple_idct_put_armv6(uint8_t *dest, int line_size, int16_t *data); | ||
| 31 | void ff_simple_idct_add_armv6(uint8_t *dest, int line_size, int16_t *data); | ||
| 32 | |||
| 33 | void ff_add_pixels_clamped_armv6(const int16_t *block, uint8_t *pixels, | ||
| 34 | int line_size); | ||
| 35 | |||
| 36 | void ff_get_pixels_armv6(int16_t *block, const uint8_t *pixels, int stride); | 29 | void ff_get_pixels_armv6(int16_t *block, const uint8_t *pixels, int stride); |
| 37 | void ff_diff_pixels_armv6(int16_t *block, const uint8_t *s1, | 30 | void ff_diff_pixels_armv6(int16_t *block, const uint8_t *s1, |
| 38 | const uint8_t *s2, int stride); | 31 | const uint8_t *s2, int stride); |
| @@ -56,17 +49,6 @@ int ff_pix_sum_armv6(uint8_t *pix, int line_size); · 17 unmodified lines | |||
| 56 | av_cold void ff_dsputil_init_armv6(DSPContext *c, AVCodecContext *avctx, | 49 | av_cold void ff_dsputil_init_armv6(DSPContext *c, AVCodecContext *avctx, |
| 57 | unsigned high_bit_depth) | 50 | unsigned high_bit_depth) |
| 58 | { | 51 | { |
| 59 | if (!avctx->lowres && !high_bit_depth) { | ||
| 60 | if (avctx->idct_algo == FF_IDCT_AUTO || | ||
| 61 | avctx->idct_algo == FF_IDCT_SIMPLEARMV6) { | ||
| 62 | c->idct_put = ff_simple_idct_put_armv6; | ||
| 63 | c->idct_add = ff_simple_idct_add_armv6; | ||
| 64 | c->idct = ff_simple_idct_armv6; | ||
| 65 | c->idct_permutation_type = FF_LIBMPEG2_IDCT_PERM; | ||
| 66 | } | ||
| 67 | } | ||
| 68 | c->add_pixels_clamped = ff_add_pixels_clamped_armv6; | ||
| 69 | |||
| 70 | if (!high_bit_depth) | 52 | if (!high_bit_depth) |
| 71 | c->get_pixels = ff_get_pixels_armv6; | 53 | c->get_pixels = ff_get_pixels_armv6; |
| 72 | c->diff_pixels = ff_diff_pixels_armv6; | 54 | c->diff_pixels = ff_diff_pixels_armv6; |
| @@ -1,5 +1,5 @@ | |||
| 1 | @ | 1 | @ |
| 2 | @ ARMv4 optimized DSP utils | 2 | @ ARMv4-optimized IDCT functions |
| 3 | @ Copyright (c) 2004 AGAWA Koji <i (AT) atty (DOT) jp> | 3 | @ Copyright (c) 2004 AGAWA Koji <i (AT) atty (DOT) jp> |
| 4 | @ | 4 | @ |
| 5 | @ This file is part of FFmpeg. | 5 | @ This file is part of FFmpeg. |
| @@ -0,0 +1,34 @@ | |||
| 1 | /* | ||
| 2 | * Copyright (c) 2009 Mans Rullgard <mans@mansr.com> | ||
| 3 | * | ||
| 4 | * This file is part of FFmpeg. | ||
| 5 | * | ||
| 6 | * FFmpeg is free software; you can redistribute it and/or | ||
| 7 | * modify it under the terms of the GNU Lesser General Public | ||
| 8 | * License as published by the Free Software Foundation; either | ||
| 9 | * version 2.1 of the License, or (at your option) any later version. | ||
| 10 | * | ||
| 11 | * FFmpeg is distributed in the hope that it will be useful, | ||
| 12 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | ||
| 13 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | ||
| 14 | * Lesser General Public License for more details. | ||
| 15 | * | ||
| 16 | * You should have received a copy of the GNU Lesser General Public | ||
| 17 | * License along with FFmpeg; if not, write to the Free Software | ||
| 18 | * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA | ||
| 19 | */ | ||
| 20 | |||
| 21 | #ifndef AVCODEC_ARM_IDCTDSP_ARM_H | ||
| 22 | #define AVCODEC_ARM_IDCTDSP_ARM_H | ||
| 23 | |||
| 24 | #include "libavcodec/avcodec.h" | ||
| 25 | #include "libavcodec/idctdsp.h" | ||
| 26 | |||
| 27 | void ff_idctdsp_init_armv5te(IDCTDSPContext *c, AVCodecContext *avctx, | ||
| 28 | unsigned high_bit_depth); | ||
| 29 | void ff_idctdsp_init_armv6(IDCTDSPContext *c, AVCodecContext *avctx, | ||
| 30 | unsigned high_bit_depth); | ||
| 31 | void ff_idctdsp_init_neon(IDCTDSPContext *c, AVCodecContext *avctx, | ||
| 32 | unsigned high_bit_depth); | ||
| 33 | |||
| 34 | #endif /* AVCODEC_ARM_IDCTDSP_ARM_H */ | ||
| @@ -0,0 +1,48 @@ | |||
| 1 | /* | ||
| 2 | * Copyright (c) 2009 Mans Rullgard <mans@mansr.com> | ||
| 3 | * | ||
| 4 | * This file is part of FFmpeg. | ||
| 5 | * | ||
| 6 | * FFmpeg is free software; you can redistribute it and/or | ||
| 7 | * modify it under the terms of the GNU Lesser General Public | ||
| 8 | * License as published by the Free Software Foundation; either | ||
| 9 | * version 2.1 of the License, or (at your option) any later version. | ||
| 10 | * | ||
| 11 | * FFmpeg is distributed in the hope that it will be useful, | ||
| 12 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | ||
| 13 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | ||
| 14 | * Lesser General Public License for more details. | ||
| 15 | * | ||
| 16 | * You should have received a copy of the GNU Lesser General Public | ||
| 17 | * License along with FFmpeg; if not, write to the Free Software | ||
| 18 | * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA | ||
| 19 | */ | ||
| 20 | |||
| 21 | #include "libavutil/arm/asm.S" | ||
| 22 | |||
| 23 | function ff_add_pixels_clamped_armv6, export=1 | ||
| 24 | push {r4-r8,lr} | ||
| 25 | mov r3, #8 | ||
| 26 | 1: | ||
| 27 | ldm r0!, {r4,r5,r12,lr} | ||
| 28 | ldrd r6, r7, [r1] | ||
| 29 | pkhbt r8, r4, r5, lsl #16 | ||
| 30 | pkhtb r5, r5, r4, asr #16 | ||
| 31 | pkhbt r4, r12, lr, lsl #16 | ||
| 32 | pkhtb lr, lr, r12, asr #16 | ||
| 33 | pld [r1, r2] | ||
| 34 | uxtab16 r8, r8, r6 | ||
| 35 | uxtab16 r5, r5, r6, ror #8 | ||
| 36 | uxtab16 r4, r4, r7 | ||
| 37 | uxtab16 lr, lr, r7, ror #8 | ||
| 38 | usat16 r8, #8, r8 | ||
| 39 | usat16 r5, #8, r5 | ||
| 40 | usat16 r4, #8, r4 | ||
| 41 | usat16 lr, #8, lr | ||
| 42 | orr r6, r8, r5, lsl #8 | ||
| 43 | orr r7, r4, lr, lsl #8 | ||
| 44 | subs r3, r3, #1 | ||
| 45 | strd_post r6, r7, r1, r2 | ||
| 46 | bgt 1b | ||
| 47 | pop {r4-r8,pc} | ||
| 48 | endfunc | ||
| @@ -0,0 +1,98 @@ | |||
| 1 | /* | ||
| 2 | * ARM-optimized IDCT functions | ||
| 3 | * Copyright (c) 2001 Lionel Ulmer | ||
| 4 | * | ||
| 5 | * This file is part of FFmpeg. | ||
| 6 | * | ||
| 7 | * FFmpeg is free software; you can redistribute it and/or | ||
| 8 | * modify it under the terms of the GNU Lesser General Public | ||
| 9 | * License as published by the Free Software Foundation; either | ||
| 10 | * version 2.1 of the License, or (at your option) any later version. | ||
| 11 | * | ||
| 12 | * FFmpeg is distributed in the hope that it will be useful, | ||
| 13 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | ||
| 14 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | ||
| 15 | * Lesser General Public License for more details. | ||
| 16 | * | ||
| 17 | * You should have received a copy of the GNU Lesser General Public | ||
| 18 | * License along with FFmpeg; if not, write to the Free Software | ||
| 19 | * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA | ||
| 20 | */ | ||
| 21 | |||
| 22 | #include <stdint.h> | ||
| 23 | |||
| 24 | #include "libavutil/attributes.h" | ||
| 25 | #include "libavutil/cpu.h" | ||
| 26 | #include "libavutil/arm/cpu.h" | ||
| 27 | #include "libavcodec/avcodec.h" | ||
| 28 | #include "libavcodec/idctdsp.h" | ||
| 29 | #include "idctdsp_arm.h" | ||
| 30 | |||
| 31 | void ff_j_rev_dct_arm(int16_t *data); | ||
| 32 | void ff_simple_idct_arm(int16_t *data); | ||
| 33 | |||
| 34 | /* XXX: local hack */ | ||
| 35 | static void (*ff_put_pixels_clamped)(const int16_t *block, uint8_t *pixels, int line_size); | ||
| 36 | static void (*ff_add_pixels_clamped)(const int16_t *block, uint8_t *pixels, int line_size); | ||
| 37 | |||
| 38 | void ff_add_pixels_clamped_arm(const int16_t *block, uint8_t *dest, | ||
| 39 | int line_size); | ||
| 40 | |||
| 41 | /* XXX: those functions should be suppressed ASAP when all IDCTs are | ||
| 42 | * converted */ | ||
| 43 | static void j_rev_dct_arm_put(uint8_t *dest, int line_size, int16_t *block) | ||
| 44 | { | ||
| 45 | ff_j_rev_dct_arm(block); | ||
| 46 | ff_put_pixels_clamped(block, dest, line_size); | ||
| 47 | } | ||
| 48 | |||
| 49 | static void j_rev_dct_arm_add(uint8_t *dest, int line_size, int16_t *block) | ||
| 50 | { | ||
| 51 | ff_j_rev_dct_arm(block); | ||
| 52 | ff_add_pixels_clamped(block, dest, line_size); | ||
| 53 | } | ||
| 54 | |||
| 55 | static void simple_idct_arm_put(uint8_t *dest, int line_size, int16_t *block) | ||
| 56 | { | ||
| 57 | ff_simple_idct_arm(block); | ||
| 58 | ff_put_pixels_clamped(block, dest, line_size); | ||
| 59 | } | ||
| 60 | |||
| 61 | static void simple_idct_arm_add(uint8_t *dest, int line_size, int16_t *block) | ||
| 62 | { | ||
| 63 | ff_simple_idct_arm(block); | ||
| 64 | ff_add_pixels_clamped(block, dest, line_size); | ||
| 65 | } | ||
| 66 | |||
| 67 | av_cold void ff_idctdsp_init_arm(IDCTDSPContext *c, AVCodecContext *avctx, | ||
| 68 | unsigned high_bit_depth) | ||
| 69 | { | ||
| 70 | int cpu_flags = av_get_cpu_flags(); | ||
| 71 | |||
| 72 | ff_put_pixels_clamped = c->put_pixels_clamped; | ||
| 73 | ff_add_pixels_clamped = c->add_pixels_clamped; | ||
| 74 | |||
| 75 | if (!avctx->lowres && !high_bit_depth) { | ||
| 76 | if (avctx->idct_algo == FF_IDCT_AUTO || | ||
| 77 | avctx->idct_algo == FF_IDCT_ARM) { | ||
| 78 | c->idct_put = j_rev_dct_arm_put; | ||
| 79 | c->idct_add = j_rev_dct_arm_add; | ||
| 80 | c->idct = ff_j_rev_dct_arm; | ||
| 81 | c->idct_permutation_type = FF_LIBMPEG2_IDCT_PERM; | ||
| 82 | } else if (avctx->idct_algo == FF_IDCT_SIMPLEARM) { | ||
| 83 | c->idct_put = simple_idct_arm_put; | ||
| 84 | c->idct_add = simple_idct_arm_add; | ||
| 85 | c->idct = ff_simple_idct_arm; | ||
| 86 | c->idct_permutation_type = FF_NO_IDCT_PERM; | ||
| 87 | } | ||
| 88 | } | ||
| 89 | |||
| 90 | c->add_pixels_clamped = ff_add_pixels_clamped_arm; | ||
| 91 | |||
| 92 | if (have_armv5te(cpu_flags)) | ||
| 93 | ff_idctdsp_init_armv5te(c, avctx, high_bit_depth); | ||
| 94 | if (have_armv6(cpu_flags)) | ||
| 95 | ff_idctdsp_init_armv6(c, avctx, high_bit_depth); | ||
| 96 | if (have_neon(cpu_flags)) | ||
| 97 | ff_idctdsp_init_neon(c, avctx, high_bit_depth); | ||
| 98 | } | ||
| @@ -22,14 +22,14 @@ · 21 unmodified lines | |||
| 22 | 22 | ||
| 23 | #include "libavutil/attributes.h" | 23 | #include "libavutil/attributes.h" |
| 24 | #include "libavcodec/avcodec.h" | 24 | #include "libavcodec/avcodec.h" |
| 25 | #include "libavcodec/dsputil.h" | 25 | #include "libavcodec/idctdsp.h" |
| 26 | #include "dsputil_arm.h" | 26 | #include "idctdsp_arm.h" |
| 27 | 27 | ||
| 28 | void ff_simple_idct_armv5te(int16_t *data); | 28 | void ff_simple_idct_armv5te(int16_t *data); |
| 29 | void ff_simple_idct_put_armv5te(uint8_t *dest, int line_size, int16_t *data); | 29 | void ff_simple_idct_put_armv5te(uint8_t *dest, int line_size, int16_t *data); |
| 30 | void ff_simple_idct_add_armv5te(uint8_t *dest, int line_size, int16_t *data); | 30 | void ff_simple_idct_add_armv5te(uint8_t *dest, int line_size, int16_t *data); |
| 31 | 31 | ||
| 32 | av_cold void ff_dsputil_init_armv5te(DSPContext *c, AVCodecContext *avctx, | 32 | av_cold void ff_idctdsp_init_armv5te(IDCTDSPContext *c, AVCodecContext *avctx, |
| 33 | unsigned high_bit_depth) | 33 | unsigned high_bit_depth) |
| 34 | { | 34 | { |
| 35 | if (!avctx->lowres && !high_bit_depth && | 35 | if (!avctx->lowres && !high_bit_depth && |
| @@ -0,0 +1,48 @@ | |||
| 1 | /* | ||
| 2 | * Copyright (c) 2009 Mans Rullgard <mans@mansr.com> | ||
| 3 | * | ||
| 4 | * This file is part of FFmpeg. | ||
| 5 | * | ||
| 6 | * FFmpeg is free software; you can redistribute it and/or | ||
| 7 | * modify it under the terms of the GNU Lesser General Public | ||
| 8 | * License as published by the Free Software Foundation; either | ||
| 9 | * version 2.1 of the License, or (at your option) any later version. | ||
| 10 | * | ||
| 11 | * FFmpeg is distributed in the hope that it will be useful, | ||
| 12 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | ||
| 13 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | ||
| 14 | * Lesser General Public License for more details. | ||
| 15 | * | ||
| 16 | * You should have received a copy of the GNU Lesser General Public | ||
| 17 | * License along with FFmpeg; if not, write to the Free Software | ||
| 18 | * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA | ||
| 19 | */ | ||
| 20 | |||
| 21 | #include <stdint.h> | ||
| 22 | |||
| 23 | #include "libavutil/attributes.h" | ||
| 24 | #include "libavcodec/avcodec.h" | ||
| 25 | #include "libavcodec/idctdsp.h" | ||
| 26 | #include "idctdsp_arm.h" | ||
| 27 | |||
| 28 | void ff_simple_idct_armv6(int16_t *data); | ||
| 29 | void ff_simple_idct_put_armv6(uint8_t *dest, int line_size, int16_t *data); | ||
| 30 | void ff_simple_idct_add_armv6(uint8_t *dest, int line_size, int16_t *data); | ||
| 31 | |||
| 32 | void ff_add_pixels_clamped_armv6(const int16_t *block, uint8_t *pixels, | ||
| 33 | int line_size); | ||
| 34 | |||
| 35 | av_cold void ff_idctdsp_init_armv6(IDCTDSPContext *c, AVCodecContext *avctx, | ||
| 36 | unsigned high_bit_depth) | ||
| 37 | { | ||
| 38 | if (!avctx->lowres && !high_bit_depth) { | ||
| 39 | if (avctx->idct_algo == FF_IDCT_AUTO || | ||
| 40 | avctx->idct_algo == FF_IDCT_SIMPLEARMV6) { | ||
| 41 | c->idct_put = ff_simple_idct_put_armv6; | ||
| 42 | c->idct_add = ff_simple_idct_add_armv6; | ||
| 43 | c->idct = ff_simple_idct_armv6; | ||
| 44 | c->idct_permutation_type = FF_LIBMPEG2_IDCT_PERM; | ||
| 45 | } | ||
| 46 | } | ||
| 47 | c->add_pixels_clamped = ff_add_pixels_clamped_armv6; | ||
| 48 | } | ||
| @@ -1,5 +1,5 @@ | |||
| 1 | /* | 1 | /* |
| 2 | * ARM NEON optimised DSP functions | 2 | * ARM-NEON-optimized IDCT functions |
| 3 | * Copyright (c) 2008 Mans Rullgard <mans@mansr.com> | 3 | * Copyright (c) 2008 Mans Rullgard <mans@mansr.com> |
| 4 | * | 4 | * |
| 5 | * This file is part of FFmpeg. | 5 | * This file is part of FFmpeg. |
| @@ -23,8 +23,8 @@ · 17 unmodified lines | |||
| 23 | 23 | ||
| 24 | #include "libavutil/attributes.h" | 24 | #include "libavutil/attributes.h" |
| 25 | #include "libavcodec/avcodec.h" | 25 | #include "libavcodec/avcodec.h" |
| 26 | #include "libavcodec/dsputil.h" | 26 | #include "libavcodec/idctdsp.h" |
| 27 | #include "dsputil_arm.h" | 27 | #include "idctdsp_arm.h" |
| 28 | 28 | ||
| 29 | void ff_simple_idct_neon(int16_t *data); | 29 | void ff_simple_idct_neon(int16_t *data); |
| 30 | void ff_simple_idct_put_neon(uint8_t *dest, int line_size, int16_t *data); | 30 | void ff_simple_idct_put_neon(uint8_t *dest, int line_size, int16_t *data); |
| @@ -34,7 +34,7 @@ void ff_add_pixels_clamped_neon(const int16_t *, uint8_t *, int); · 3 unmodified lines | |||
| 34 | void ff_put_pixels_clamped_neon(const int16_t *, uint8_t *, int); | 34 | void ff_put_pixels_clamped_neon(const int16_t *, uint8_t *, int); |
| 35 | void ff_put_signed_pixels_clamped_neon(const int16_t *, uint8_t *, int); | 35 | void ff_put_signed_pixels_clamped_neon(const int16_t *, uint8_t *, int); |
| 36 | 36 | ||
| 37 | av_cold void ff_dsputil_init_neon(DSPContext *c, AVCodecContext *avctx, | 37 | av_cold void ff_idctdsp_init_neon(IDCTDSPContext *c, AVCodecContext *avctx, |
| 38 | unsigned high_bit_depth) | 38 | unsigned high_bit_depth) |
| 39 | { | 39 | { |
| 40 | if (!avctx->lowres && !high_bit_depth) { | 40 | if (!avctx->lowres && !high_bit_depth) { |
| @@ -1,5 +1,5 @@ | |||
| 1 | /* | 1 | /* |
| 2 | * ARM NEON optimised DSP functions | 2 | * ARM-NEON-optimized IDCT functions |
| 3 | * Copyright (c) 2008 Mans Rullgard <mans@mansr.com> | 3 | * Copyright (c) 2008 Mans Rullgard <mans@mansr.com> |
| 4 | * | 4 | * |
| 5 | * This file is part of FFmpeg. | 5 | * This file is part of FFmpeg. |