From 5bf208f659703895df7926238dcfa8a8175de36b Mon Sep 17 00:00:00 2001 From: Michael Niedermayer Date: Mon, 1 Jul 2013 10:01:22 +0200 Subject: [PATCH] jpeg2000: Use separate fields for int and float codepaths Split stepsize and data into int and float variants. Eliminates a number of casts and simplifies spotting errors. Signed-off-by: Luca Barbato --- libavcodec/jpeg2000.c | 35 ++++++++++++++++++++++------------- libavcodec/jpeg2000.h | 11 ++++------- libavcodec/jpeg2000dec.c | 29 +++++++++++++++-------------- 3 files changed, 41 insertions(+), 34 deletions(-) diff --git a/libavcodec/jpeg2000.c b/libavcodec/jpeg2000.c index 54550bc3b2..73d2a38f28 100644 --- a/libavcodec/jpeg2000.c +++ b/libavcodec/jpeg2000.c @@ -217,9 +217,17 @@ int ff_jpeg2000_init_component(Jpeg2000Component *comp, csize = (comp->coord[0][1] - comp->coord[0][0]) * (comp->coord[1][1] - comp->coord[1][0]); - comp->data = av_malloc_array(csize, sizeof(*comp->data)); - if (!comp->data) - return AVERROR(ENOMEM); + if (codsty->transform == FF_DWT97) { + comp->i_data = NULL; + comp->f_data = av_malloc_array(csize, sizeof(*comp->f_data)); + if (!comp->f_data) + return AVERROR(ENOMEM); + } else { + comp->f_data = NULL; + comp->i_data = av_malloc_array(csize, sizeof(*comp->i_data)); + if (!comp->i_data) + return AVERROR(ENOMEM); + } comp->reslevel = av_malloc_array(codsty->nreslevels, sizeof(*comp->reslevel)); if (!comp->reslevel) return AVERROR(ENOMEM); @@ -290,8 +298,8 @@ int ff_jpeg2000_init_component(Jpeg2000Component *comp, /*TODO: Compute formula to implement. */ numbps = cbps + lut_gain[codsty->transform == FF_DWT53][bandno + (reslevelno > 0)]; - band->stepsize = (float)SHL(2048 + qntsty->mant[gbandno], - 2 + numbps - qntsty->expn[gbandno]); + band->f_stepsize = SHL(2048 + qntsty->mant[gbandno], + 2 + numbps - qntsty->expn[gbandno]); break; case JPEG2000_QSTY_SE: /* Exponent quantization step. @@ -303,20 +311,20 @@ int ff_jpeg2000_init_component(Jpeg2000Component *comp, * but it works (compared to OpenJPEG). Why? * Further investigation needed. */ gain = cbps; - band->stepsize = pow(2.0, gain - qntsty->expn[gbandno]); - band->stepsize *= (float)qntsty->mant[gbandno] / 2048.0 + 1.0; + band->f_stepsize = pow(2.0, gain - qntsty->expn[gbandno]); + band->f_stepsize *= qntsty->mant[gbandno] / 2048.0 + 1.0; break; default: - band->stepsize = 0; + band->f_stepsize = 0; av_log(avctx, AV_LOG_ERROR, "Unknown quantization format\n"); break; } /* FIXME: In openjepg code stespize = stepsize * 0.5. Why? * If not set output of entropic decoder is not correct. */ - band->stepsize *= 0.5; - /* BITEXACT computing case --> convert to int */ - if (avctx->flags & CODEC_FLAG_BITEXACT) - band->stepsize = (int32_t)(band->stepsize * (1 << 16)); + if (!av_codec_is_encoder(avctx->codec)) + band->f_stepsize *= 0.5; + + band->i_stepsize = band->f_stepsize * (1 << 16); /* computation of tbx_0, tbx_1, tby_0, tby_1 * see ISO/IEC 15444-1:2002 B.5 eq. B-15 and tbl B.1 @@ -491,5 +499,6 @@ void ff_jpeg2000_cleanup(Jpeg2000Component *comp, Jpeg2000CodingStyle *codsty) ff_dwt_destroy(&comp->dwt); av_freep(&comp->reslevel); - av_freep(&comp->data); + av_freep(&comp->i_data); + av_freep(&comp->f_data); } diff --git a/libavcodec/jpeg2000.h b/libavcodec/jpeg2000.h index 00bfcdda81..0ff652e56d 100644 --- a/libavcodec/jpeg2000.h +++ b/libavcodec/jpeg2000.h @@ -178,12 +178,11 @@ typedef struct Jpeg2000Prec { uint16_t coord[2][2]; // border coordinates {{x0, x1}, {y0, y1}} } Jpeg2000Prec; // precinct -/* TODO: stepsize can be float or integer depending on - * reversible or irreversible transformation. */ typedef struct Jpeg2000Band { uint16_t coord[2][2]; // border coordinates {{x0, x1}, {y0, y1}} uint16_t log2_cblk_width, log2_cblk_height; - float stepsize; // quantization stepsize + int i_stepsize; // quantization stepsize + float f_stepsize; // quantization stepsize Jpeg2000Prec *prec; } Jpeg2000Band; // subband @@ -195,13 +194,11 @@ typedef struct Jpeg2000ResLevel { Jpeg2000Band *band; } Jpeg2000ResLevel; // resolution level -/* TODO: data can be float of integer depending of reversible/irreversible - * transformation. - */ typedef struct Jpeg2000Component { Jpeg2000ResLevel *reslevel; DWTContext dwt; - float *data; + float *f_data; + int *i_data; uint16_t coord[2][2]; // border coordinates {{x0, x1}, {y0, y1}} -- can be reduced with lowres option uint16_t coord_o[2][2]; // border coordinates {{x0, x1}, {y0, y1}} -- original values from jpeg2000 headers } Jpeg2000Component; diff --git a/libavcodec/jpeg2000dec.c b/libavcodec/jpeg2000dec.c index 3a76f79d66..07ccbe5c8e 100644 --- a/libavcodec/jpeg2000dec.c +++ b/libavcodec/jpeg2000dec.c @@ -990,11 +990,11 @@ static void dequantization_float(int x, int y, Jpeg2000Cblk *cblk, Jpeg2000T1Context *t1, Jpeg2000Band *band) { int i, j, idx; - float *datap = &comp->data[(comp->coord[0][1] - comp->coord[0][0]) * y + x]; + float *datap = &comp->f_data[(comp->coord[0][1] - comp->coord[0][0]) * y + x]; for (j = 0; j < (cblk->coord[1][1] - cblk->coord[1][0]); ++j) for (i = 0; i < (cblk->coord[0][1] - cblk->coord[0][0]); ++i) { idx = (comp->coord[0][1] - comp->coord[0][0]) * j + i; - datap[idx] = (float)(t1->data[j][i]) * ((float)band->stepsize); + datap[idx] = (float)(t1->data[j][i]) * band->f_stepsize; } } @@ -1004,13 +1004,12 @@ static void dequantization_int(int x, int y, Jpeg2000Cblk *cblk, Jpeg2000T1Context *t1, Jpeg2000Band *band) { int i, j, idx; - int32_t *datap = - (int32_t *) &comp->data[(comp->coord[0][1] - comp->coord[0][0]) * y + x]; + int32_t *datap = &comp->i_data[(comp->coord[0][1] - comp->coord[0][0]) * y + x]; for (j = 0; j < (cblk->coord[1][1] - cblk->coord[1][0]); ++j) for (i = 0; i < (cblk->coord[0][1] - cblk->coord[0][0]); ++i) { idx = (comp->coord[0][1] - comp->coord[0][0]) * j + i; datap[idx] = - ((int32_t)(t1->data[j][i]) * ((int32_t)band->stepsize) + (1 << 15)) >> 16; + ((int32_t)(t1->data[j][i]) * band->i_stepsize + (1 << 15)) >> 16; } } @@ -1037,9 +1036,9 @@ static void mct_decode(Jpeg2000DecoderContext *s, Jpeg2000Tile *tile) for (i = 0; i < 3; i++) if (tile->codsty[0].transform == FF_DWT97) - srcf[i] = tile->comp[i].data; + srcf[i] = tile->comp[i].f_data; else - src[i] = (int32_t *)tile->comp[i].data; + src [i] = tile->comp[i].i_data; for (i = 0; i < 2; i++) csize *= tile->comp[0].coord[i][1] - tile->comp[0].coord[i][0]; @@ -1129,21 +1128,23 @@ static int jpeg2000_decode_tile(Jpeg2000DecoderContext *s, Jpeg2000Tile *tile, } /* end reslevel */ /* inverse DWT */ - ff_dwt_decode(&comp->dwt, comp->data); + ff_dwt_decode(&comp->dwt, codsty->transform == FF_DWT97 ? (void*)comp->f_data : (void*)comp->i_data); } /*end comp */ /* inverse MCT transformation */ if (tile->codsty[0].mct) mct_decode(s, tile); - if (s->avctx->pix_fmt == AV_PIX_FMT_BGRA) // RGBA -> BGRA - FFSWAP(float *, tile->comp[0].data, tile->comp[2].data); + if (s->avctx->pix_fmt == AV_PIX_FMT_BGRA) { // RGBA -> BGRA + FFSWAP(float *, tile->comp[0].f_data, tile->comp[2].f_data); + FFSWAP(int *, tile->comp[0].i_data, tile->comp[2].i_data); + } if (s->precision <= 8) { for (compno = 0; compno < s->ncomponents; compno++) { Jpeg2000Component *comp = tile->comp + compno; - float *datap = comp->data; - int32_t *i_datap = (int32_t *) comp->data; + float *datap = comp->f_data; + int32_t *i_datap = comp->i_data; y = tile->comp[compno].coord[1][0] - s->image_offset_y; line = picture->data[0] + y * picture->linesize[0]; for (; y < tile->comp[compno].coord[1][1] - s->image_offset_y; y += s->cdy[compno]) { @@ -1171,8 +1172,8 @@ static int jpeg2000_decode_tile(Jpeg2000DecoderContext *s, Jpeg2000Tile *tile, } else { for (compno = 0; compno < s->ncomponents; compno++) { Jpeg2000Component *comp = tile->comp + compno; - float *datap = comp->data; - int32_t *i_datap = (int32_t *) comp->data; + float *datap = comp->f_data; + int32_t *i_datap = comp->i_data; uint16_t *linel; y = tile->comp[compno].coord[1][0] - s->image_offset_y;