From 8b6103da0f0f2ac73cf2698e8f26d7b989a564ac Mon Sep 17 00:00:00 2001 From: Michael Niedermayer Date: Tue, 17 May 2005 11:12:04 +0000 Subject: porting vp3 idct over to lavc idct api Originally committed as revision 4257 to svn://svn.ffmpeg.org/ffmpeg/trunk --- libavcodec/avcodec.h | 1 + libavcodec/dsputil.c | 9 +-- libavcodec/dsputil.h | 14 +---- libavcodec/vp3.c | 34 ++++++++---- libavcodec/vp3dsp.c | 154 ++++++++++++++++++++++++++++++++++----------------- 5 files changed, 136 insertions(+), 76 deletions(-) (limited to 'libavcodec') diff --git a/libavcodec/avcodec.h b/libavcodec/avcodec.h index b9fccc4cff..f849f6b562 100644 --- a/libavcodec/avcodec.h +++ b/libavcodec/avcodec.h @@ -1177,6 +1177,7 @@ typedef struct AVCodecContext { #define FF_IDCT_SH4 9 #define FF_IDCT_SIMPLEARM 10 #define FF_IDCT_H264 11 +#define FF_IDCT_VP3 12 /** * slice count. diff --git a/libavcodec/dsputil.c b/libavcodec/dsputil.c index a19f79d316..5505dff78e 100644 --- a/libavcodec/dsputil.c +++ b/libavcodec/dsputil.c @@ -3643,6 +3643,11 @@ void dsputil_init(DSPContext* c, AVCodecContext *avctx) c->idct_add= ff_jref_idct_add; c->idct = j_rev_dct; c->idct_permutation_type= FF_LIBMPEG2_IDCT_PERM; + }else if(avctx->idct_algo==FF_IDCT_VP3){ + c->idct_put= ff_vp3_idct_put_c; + c->idct_add= ff_vp3_idct_add_c; + c->idct = ff_vp3_idct_c; + c->idct_permutation_type= FF_NO_IDCT_PERM; }else{ //accurate/default c->idct_put= simple_idct_put; c->idct_add= simple_idct_add; @@ -3653,10 +3658,6 @@ void dsputil_init(DSPContext* c, AVCodecContext *avctx) c->h264_idct_add= ff_h264_idct_add_c; - /* VP3 DSP support */ - c->vp3_dsp_init = vp3_dsp_init_c; - c->vp3_idct = vp3_idct_c; - c->get_pixels = get_pixels_c; c->diff_pixels = diff_pixels_c; c->put_pixels_clamped = put_pixels_clamped_c; diff --git a/libavcodec/dsputil.h b/libavcodec/dsputil.h index 57547431ed..002e284639 100644 --- a/libavcodec/dsputil.h +++ b/libavcodec/dsputil.h @@ -68,17 +68,9 @@ extern uint32_t squareTbl[512]; extern uint8_t cropTbl[256 + 2 * MAX_NEG_CROP]; /* VP3 DSP functions */ -void vp3_dsp_init_c(void); -void vp3_idct_c(int16_t *input_data, int16_t *dequant_matrix, - int coeff_count, DCTELEM *output_data); - -void vp3_dsp_init_mmx(void); -void vp3_idct_mmx(int16_t *input_data, int16_t *dequant_matrix, - int coeff_count, DCTELEM *output_data); - -void vp3_dsp_init_sse2(void); -void vp3_idct_sse2(int16_t *input_data, int16_t *dequant_matrix, - int coeff_count, DCTELEM *output_data); +void ff_vp3_idct_c(DCTELEM *block/* align 16*/); +void ff_vp3_idct_put_c(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/); +void ff_vp3_idct_add_c(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/); /* minimum alignment rules ;) if u notice errors in the align stuff, need more alignment for some asm code for some cpu diff --git a/libavcodec/vp3.c b/libavcodec/vp3.c index 7923229dc1..da4f531d29 100644 --- a/libavcodec/vp3.c +++ b/libavcodec/vp3.c @@ -2053,7 +2053,7 @@ static void render_fragments(Vp3DecodeContext *s, int height, int plane /* 0 = Y, 1 = U, 2 = V */) { - int x, y; + int x, y, j; int m, n; int i = first_fragment; int16_t *dequantizer; @@ -2188,21 +2188,32 @@ av_log(s->avctx, AV_LOG_ERROR, " help! got beefy vector! (%X, %X)\n", motion_x, i, s->all_fragments[i].coding_method, s->all_fragments[i].coeffs[0], dequantizer[0]); + if(s->avctx->idct_algo==FF_IDCT_VP3){ + for (j = 0; j < 64; j++) { + s->all_fragments[i].coeffs[j] *= dequantizer[j]; + } + }else{ + for (j = 0; j < 64; j++) { + s->all_fragments[i].coeffs[j]= (dequantizer[j] * s->all_fragments[i].coeffs[j] + 2) >> 2; + } + } + /* invert DCT and place (or add) in final output */ - s->dsp.vp3_idct(s->all_fragments[i].coeffs, - dequantizer, - s->all_fragments[i].coeff_count, - output_samples); - memset(s->all_fragments[i].coeffs, 0, 64*sizeof(DCTELEM)); + if (s->all_fragments[i].coding_method == MODE_INTRA) { - s->dsp.put_signed_pixels_clamped(output_samples, + if(s->avctx->idct_algo!=FF_IDCT_VP3) + s->all_fragments[i].coeffs[0] += 128<<3; + s->dsp.idct_put( output_plane + s->all_fragments[i].first_pixel, - stride); + stride, + s->all_fragments[i].coeffs); } else { - s->dsp.add_pixels_clamped(output_samples, + s->dsp.idct_add( output_plane + s->all_fragments[i].first_pixel, - stride); + stride, + s->all_fragments[i].coeffs); } + memset(s->all_fragments[i].coeffs, 0, 64*sizeof(DCTELEM)); debug_idct("block after idct_%s():\n", (s->all_fragments[i].coding_method == MODE_INTRA)? @@ -2496,8 +2507,9 @@ static int vp3_decode_init(AVCodecContext *avctx) #endif avctx->pix_fmt = PIX_FMT_YUV420P; avctx->has_b_frames = 0; + if(avctx->idct_algo==FF_IDCT_AUTO) + avctx->idct_algo=FF_IDCT_VP3; dsputil_init(&s->dsp, avctx); - s->dsp.vp3_dsp_init(); /* initialize to an impossible value which will force a recalculation * in the first frame decode */ diff --git a/libavcodec/vp3dsp.c b/libavcodec/vp3dsp.c index 8cf77a25af..015f57b570 100644 --- a/libavcodec/vp3dsp.c +++ b/libavcodec/vp3dsp.c @@ -36,29 +36,17 @@ #define xC6S2 25080 #define xC7S1 12785 -void vp3_dsp_init_c(void) +static always_inline void idct(uint8_t *dst, int stride, int16_t *input, int type) { - /* nop */ -} - -void vp3_idct_c(int16_t *input_data, int16_t *dequant_matrix, - int coeff_count, int16_t *output_data) -{ - int32_t dequantized_data[64]; - int32_t *ip = dequantized_data; - int16_t *op = output_data; + int16_t *ip = input; + uint8_t *cm = cropTbl + MAX_NEG_CROP; - int32_t A_, B_, C_, D_, _Ad, _Bd, _Cd, _Dd, E_, F_, G_, H_; - int32_t _Ed, _Gd, _Add, _Bdd, _Fd, _Hd; - int32_t t1, t2; + int A_, B_, C_, D_, _Ad, _Bd, _Cd, _Dd, E_, F_, G_, H_; + int _Ed, _Gd, _Add, _Bdd, _Fd, _Hd; + int t1, t2; int i, j; - /* de-zigzag and dequantize */ - for (i = 0; i < coeff_count; i++) { - dequantized_data[i] = dequant_matrix[i] * input_data[i]; - } - /* Inverse DCT on the rows now */ for (i = 0; i < 8; i++) { /* Check for non-zero values */ @@ -131,28 +119,28 @@ void vp3_idct_c(int16_t *input_data, int16_t *dequant_matrix, _Hd = _Bd + H_; /* Final sequence of operations over-write original inputs. */ - ip[0] = (int16_t)((_Gd + _Cd ) >> 0); - ip[7] = (int16_t)((_Gd - _Cd ) >> 0); + ip[0] = _Gd + _Cd ; + ip[7] = _Gd - _Cd ; - ip[1] = (int16_t)((_Add + _Hd ) >> 0); - ip[2] = (int16_t)((_Add - _Hd ) >> 0); + ip[1] = _Add + _Hd; + ip[2] = _Add - _Hd; - ip[3] = (int16_t)((_Ed + _Dd ) >> 0); - ip[4] = (int16_t)((_Ed - _Dd ) >> 0); + ip[3] = _Ed + _Dd ; + ip[4] = _Ed - _Dd ; - ip[5] = (int16_t)((_Fd + _Bdd ) >> 0); - ip[6] = (int16_t)((_Fd - _Bdd ) >> 0); + ip[5] = _Fd + _Bdd; + ip[6] = _Fd - _Bdd; } ip += 8; /* next row */ } - - ip = dequantized_data; + + ip = input; for ( i = 0; i < 8; i++) { /* Check for non-zero values (bitwise or faster than ||) */ - if ( ip[0 * 8] | ip[1 * 8] | ip[2 * 8] | ip[3 * 8] | + if ( ip[1 * 8] | ip[2 * 8] | ip[3 * 8] | ip[4 * 8] | ip[5 * 8] | ip[6 * 8] | ip[7 * 8] ) { t1 = (int32_t)(xC1S7 * ip[1*8]); @@ -222,37 +210,103 @@ void vp3_idct_c(int16_t *input_data, int16_t *dequant_matrix, _Fd = F_ - _Ad; _Hd = _Bd + H_; + if(type==1){ //HACK + _Gd += 16*128; + _Add+= 16*128; + _Ed += 16*128; + _Fd += 16*128; + } _Gd += IdctAdjustBeforeShift; _Add += IdctAdjustBeforeShift; _Ed += IdctAdjustBeforeShift; _Fd += IdctAdjustBeforeShift; /* Final sequence of operations over-write original inputs. */ - op[0*8] = (int16_t)((_Gd + _Cd ) >> 4); - op[7*8] = (int16_t)((_Gd - _Cd ) >> 4); - - op[1*8] = (int16_t)((_Add + _Hd ) >> 4); - op[2*8] = (int16_t)((_Add - _Hd ) >> 4); - - op[3*8] = (int16_t)((_Ed + _Dd ) >> 4); - op[4*8] = (int16_t)((_Ed - _Dd ) >> 4); - - op[5*8] = (int16_t)((_Fd + _Bdd ) >> 4); - op[6*8] = (int16_t)((_Fd - _Bdd ) >> 4); + if(type==0){ + ip[0*8] = (_Gd + _Cd ) >> 4; + ip[7*8] = (_Gd - _Cd ) >> 4; + + ip[1*8] = (_Add + _Hd ) >> 4; + ip[2*8] = (_Add - _Hd ) >> 4; + + ip[3*8] = (_Ed + _Dd ) >> 4; + ip[4*8] = (_Ed - _Dd ) >> 4; + + ip[5*8] = (_Fd + _Bdd ) >> 4; + ip[6*8] = (_Fd - _Bdd ) >> 4; + }else if(type==1){ + dst[0*stride] = cm[(_Gd + _Cd ) >> 4]; + dst[7*stride] = cm[(_Gd - _Cd ) >> 4]; + + dst[1*stride] = cm[(_Add + _Hd ) >> 4]; + dst[2*stride] = cm[(_Add - _Hd ) >> 4]; + + dst[3*stride] = cm[(_Ed + _Dd ) >> 4]; + dst[4*stride] = cm[(_Ed - _Dd ) >> 4]; + + dst[5*stride] = cm[(_Fd + _Bdd ) >> 4]; + dst[6*stride] = cm[(_Fd - _Bdd ) >> 4]; + }else{ + dst[0*stride] = cm[dst[0*stride] + ((_Gd + _Cd ) >> 4)]; + dst[7*stride] = cm[dst[7*stride] + ((_Gd - _Cd ) >> 4)]; + + dst[1*stride] = cm[dst[1*stride] + ((_Add + _Hd ) >> 4)]; + dst[2*stride] = cm[dst[2*stride] + ((_Add - _Hd ) >> 4)]; + + dst[3*stride] = cm[dst[3*stride] + ((_Ed + _Dd ) >> 4)]; + dst[4*stride] = cm[dst[4*stride] + ((_Ed - _Dd ) >> 4)]; + + dst[5*stride] = cm[dst[5*stride] + ((_Fd + _Bdd ) >> 4)]; + dst[6*stride] = cm[dst[6*stride] + ((_Fd - _Bdd ) >> 4)]; + } } else { - - op[0*8] = 0; - op[7*8] = 0; - op[1*8] = 0; - op[2*8] = 0; - op[3*8] = 0; - op[4*8] = 0; - op[5*8] = 0; - op[6*8] = 0; + if(type==0){ + ip[0*8] = + ip[1*8] = + ip[2*8] = + ip[3*8] = + ip[4*8] = + ip[5*8] = + ip[6*8] = + ip[7*8] = ((xC4S4 * ip[0*8] + (IdctAdjustBeforeShift<<16))>>20); + }else if(type==1){ + dst[0*stride]= + dst[1*stride]= + dst[2*stride]= + dst[3*stride]= + dst[4*stride]= + dst[5*stride]= + dst[6*stride]= + dst[7*stride]= 128 + ((xC4S4 * ip[0*8] + (IdctAdjustBeforeShift<<16))>>20); + }else{ + if(ip[0*8]){ + int v= ((xC4S4 * ip[0*8] + (IdctAdjustBeforeShift<<16))>>20); + dst[0*stride] = cm[dst[0*stride] + v]; + dst[1*stride] = cm[dst[1*stride] + v]; + dst[2*stride] = cm[dst[2*stride] + v]; + dst[3*stride] = cm[dst[3*stride] + v]; + dst[4*stride] = cm[dst[4*stride] + v]; + dst[5*stride] = cm[dst[5*stride] + v]; + dst[6*stride] = cm[dst[6*stride] + v]; + dst[7*stride] = cm[dst[7*stride] + v]; + } + } } ip++; /* next column */ - op++; + dst++; } } + +void ff_vp3_idct_c(DCTELEM *block/* align 16*/){ + idct(NULL, 0, block, 0); +} + +void ff_vp3_idct_put_c(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/){ + idct(dest, line_size, block, 1); +} + +void ff_vp3_idct_add_c(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/){ + idct(dest, line_size, block, 2); +} -- cgit v1.2.3