virtualenv/FFmpeg
1
0
Fork 0
mirror of https://github.com/FFmpeg/FFmpeg.git synced 2025-02-04 06:08:26 +02:00
Code Issues Releases Activity

avcodec/vvc/vvc_ps: Use union for luts to avoid unaligned accesses

Browse Source 
These arrays are currently accessed via uint16_t* pointers
although nothing guarantees their alignment. Furthermore,
this is problematic wrt the effective-type rules.
Fix this by using a union of arrays of uint8_t and uint16_t.

Reviewed-by: Nuo Mi <nuomi2021@gmail.com>
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
This commit is contained in:
Andreas Rheinhardt 2024-02-18 19:11:39 +01:00
parent 1eafbc27e2
commit ef04737e18
6 changed files with 14 additions and 13 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
libavcodec/vvc/vvc_filter.c
View File

@ -1328,5 +1328,5 @@ void ff_vvc_lmcs_filter(const VVCLocalContext *lc, const int x, const int y)
const int height = FFMIN(fc->ps.pps->height - y, ctb_size); const int height = FFMIN(fc->ps.pps->height - y, ctb_size);
uint8_t *data = fc->frame->data[LUMA] + y * fc->frame->linesize[LUMA] + (x << fc->ps.sps->pixel_shift); uint8_t *data = fc->frame->data[LUMA] + y * fc->frame->linesize[LUMA] + (x << fc->ps.sps->pixel_shift);
if (sc->sh.r->sh_lmcs_used_flag) if (sc->sh.r->sh_lmcs_used_flag)
fc->vvcdsp.lmcs.filter(data, fc->frame->linesize[LUMA], width, height, fc->ps.lmcs.inv_lut); fc->vvcdsp.lmcs.filter(data, fc->frame->linesize[LUMA], width, height, &fc->ps.lmcs.inv_lut);
} }

4
libavcodec/vvc/vvc_filter_template.c
View File

@ -22,9 +22,9 @@
#include "libavcodec/h26x/h2656_sao_template.c" #include "libavcodec/h26x/h2656_sao_template.c"
static void FUNC(lmcs_filter_luma)(uint8_t *_dst, ptrdiff_t dst_stride, const int width, const int height, const uint8_t *_lut) static void FUNC(lmcs_filter_luma)(uint8_t *_dst, ptrdiff_t dst_stride, const int width, const int height, const void *_lut)
{ {
const pixel *lut = (const pixel *)_lut; const pixel *lut = _lut;
pixel *dst = (pixel*)_dst; pixel *dst = (pixel*)_dst;
dst_stride /= sizeof(pixel); dst_stride /= sizeof(pixel);

4
libavcodec/vvc/vvc_inter.c
View File

@ -571,7 +571,7 @@ static void pred_regular_luma(VVCLocalContext *lc, const int hf_idx, const int v
const int intra_weight = ciip_derive_intra_weight(lc, x0, y0, sbw, sbh); const int intra_weight = ciip_derive_intra_weight(lc, x0, y0, sbw, sbh);
fc->vvcdsp.intra.intra_pred(lc, x0, y0, sbw, sbh, 0); fc->vvcdsp.intra.intra_pred(lc, x0, y0, sbw, sbh, 0);
if (sc->sh.r->sh_lmcs_used_flag) if (sc->sh.r->sh_lmcs_used_flag)
fc->vvcdsp.lmcs.filter(inter, inter_stride, sbw, sbh, fc->ps.lmcs.fwd_lut); fc->vvcdsp.lmcs.filter(inter, inter_stride, sbw, sbh, &fc->ps.lmcs.fwd_lut);
fc->vvcdsp.inter.put_ciip(dst, dst_stride, sbw, sbh, inter, inter_stride, intra_weight); fc->vvcdsp.inter.put_ciip(dst, dst_stride, sbw, sbh, inter, inter_stride, intra_weight);
} }
@ -887,7 +887,7 @@ static void predict_inter(VVCLocalContext *lc)
if (lc->sc->sh.r->sh_lmcs_used_flag && !cu->ciip_flag) { if (lc->sc->sh.r->sh_lmcs_used_flag && !cu->ciip_flag) {
uint8_t* dst0 = POS(0, cu->x0, cu->y0); uint8_t* dst0 = POS(0, cu->x0, cu->y0);
fc->vvcdsp.lmcs.filter(dst0, fc->frame->linesize[LUMA], cu->cb_width, cu->cb_height, fc->ps.lmcs.fwd_lut); fc->vvcdsp.lmcs.filter(dst0, fc->frame->linesize[LUMA], cu->cb_width, cu->cb_height, &fc->ps.lmcs.fwd_lut);
} }
} }

8
libavcodec/vvc/vvc_ps.c
View File

@ -642,9 +642,9 @@ static int lmcs_derive_lut(VVCLMCS *lmcs, const H266RawAPS *rlmcs, const H266Raw
const uint16_t fwd_sample = lmcs_derive_lut_sample(sample, lmcs->pivot, const uint16_t fwd_sample = lmcs_derive_lut_sample(sample, lmcs->pivot,
input_pivot, scale_coeff, idx_y, max); input_pivot, scale_coeff, idx_y, max);
if (bit_depth > 8) if (bit_depth > 8)
((uint16_t *)lmcs->fwd_lut)[sample] = fwd_sample; lmcs->fwd_lut.u16[sample] = fwd_sample;
else else
lmcs->fwd_lut[sample] = fwd_sample; lmcs->fwd_lut.u8 [sample] = fwd_sample;
} }
@ -659,9 +659,9 @@ static int lmcs_derive_lut(VVCLMCS *lmcs, const H266RawAPS *rlmcs, const H266Raw
inv_scale_coeff, i, max); inv_scale_coeff, i, max);
if (bit_depth > 8) if (bit_depth > 8)
((uint16_t *)lmcs->inv_lut)[sample] = inv_sample; lmcs->inv_lut.u16[sample] = inv_sample;
else else
lmcs->inv_lut[sample] = inv_sample; lmcs->inv_lut.u8 [sample] = inv_sample;
} }
return 0; return 0;

7
libavcodec/vvc/vvc_ps.h
View File

@ -191,9 +191,10 @@ typedef struct VVCLMCS {
uint8_t min_bin_idx; uint8_t min_bin_idx;
uint8_t max_bin_idx; uint8_t max_bin_idx;
//*2 for high depth union {
uint8_t fwd_lut[LMCS_MAX_LUT_SIZE * 2]; uint8_t u8[LMCS_MAX_LUT_SIZE];
uint8_t inv_lut[LMCS_MAX_LUT_SIZE * 2]; uint16_t u16[LMCS_MAX_LUT_SIZE]; ///< for high bit-depth
} fwd_lut, inv_lut;
uint16_t pivot[LMCS_MAX_BIN_SIZE + 1]; uint16_t pivot[LMCS_MAX_BIN_SIZE + 1];
uint16_t chroma_scale_coeff[LMCS_MAX_BIN_SIZE]; uint16_t chroma_scale_coeff[LMCS_MAX_BIN_SIZE];

2
libavcodec/vvc/vvcdsp.h
View File

@ -119,7 +119,7 @@ typedef struct VVCItxDSPContext {
} VVCItxDSPContext; } VVCItxDSPContext;
typedef struct VVCLMCSDSPContext { typedef struct VVCLMCSDSPContext {
void (*filter)(uint8_t *dst, ptrdiff_t dst_stride, int width, int height, const uint8_t *lut); void (*filter)(uint8_t *dst, ptrdiff_t dst_stride, int width, int height, const void *lut);
} VVCLMCSDSPContext; } VVCLMCSDSPContext;
typedef struct VVCLFDSPContext { typedef struct VVCLFDSPContext {