Message ID | 20240707114448.32270-1-remi@remlab.net |
---|---|
State | New |
Headers | show |
Series | [FFmpeg-devel,1/2] lavc/h264dsp: remove MMI 8-bit chroma DC dequant | expand |
Context | Check | Description |
---|---|---|
yinshiyou/make_loongarch64 | success | Make finished |
yinshiyou/make_fate_loongarch64 | fail | Make fate failed |
andriy/make_x86 | success | Make finished |
andriy/make_fate_x86 | success | Make fate finished |
Le sunnuntaina 7. heinäkuuta 2024, 14.44.47 EEST Rémi Denis-Courmont a écrit : > The function is exactly identical to the C reference, only with the > constant propagated manually. It does not optimise anything. There are presently no (other) implementations for `H264DSPContext.h264_chroma_dc_dequant_idct`, which is used by H.264 and SVQ3 decoders. I don't know if we should keep the pointers anyway or just hard-code the calls to the C implementation. I am not sure if it would be possible to marginally optimise the functions with scalar product or matrix extensions... at least those extensions that do support widening 16x16 multiplications. (This is unfortunately not a given. With the AI/ML hype,16-bit floats, 8-bit and/or 4-bit integers are the more commonly supported types.)
diff --git a/libavcodec/mips/h264dsp_init_mips.c b/libavcodec/mips/h264dsp_init_mips.c index e33df32c71..615b00d142 100644 --- a/libavcodec/mips/h264dsp_init_mips.c +++ b/libavcodec/mips/h264dsp_init_mips.c @@ -46,10 +46,7 @@ av_cold void ff_h264dsp_init_mips(H264DSPContext *c, const int bit_depth, c->h264_luma_dc_dequant_idct = ff_h264_luma_dc_dequant_idct_8_mmi; - if (chroma_format_idc <= 1) - c->h264_chroma_dc_dequant_idct = - ff_h264_chroma_dc_dequant_idct_8_mmi; - else + if (chroma_format_idc > 1) c->h264_chroma_dc_dequant_idct = ff_h264_chroma422_dc_dequant_idct_8_mmi; diff --git a/libavcodec/mips/h264dsp_mips.h b/libavcodec/mips/h264dsp_mips.h index 93a201c66a..7e1dcdd013 100644 --- a/libavcodec/mips/h264dsp_mips.h +++ b/libavcodec/mips/h264dsp_mips.h @@ -340,7 +340,6 @@ void ff_h264_idct_add8_422_8_mmi(uint8_t **dest, const int *block_offset, int16_t *block, int stride, const uint8_t nnzc[15*8]); void ff_h264_luma_dc_dequant_idct_8_mmi(int16_t *output, int16_t *input, int qmul); -void ff_h264_chroma_dc_dequant_idct_8_mmi(int16_t *block, int qmul); void ff_h264_chroma422_dc_dequant_idct_8_mmi(int16_t *block, int qmul); void ff_h264_weight_pixels16_8_mmi(uint8_t *block, ptrdiff_t stride, int height, diff --git a/libavcodec/mips/h264dsp_mmi.c b/libavcodec/mips/h264dsp_mmi.c index dff3d331b1..1ee8ca2f1c 100644 --- a/libavcodec/mips/h264dsp_mmi.c +++ b/libavcodec/mips/h264dsp_mmi.c @@ -1117,20 +1117,6 @@ void ff_h264_chroma422_dc_dequant_idct_8_mmi(int16_t *block, int qmul) block[112]= (t[7]*qmul + 128) >> 8; } -void ff_h264_chroma_dc_dequant_idct_8_mmi(int16_t *block, int qmul) -{ - int a,b,c,d; - - d = block[0] - block[16]; - a = block[0] + block[16]; - b = block[32] - block[48]; - c = block[32] + block[48]; - block[0] = ((a+c)*qmul) >> 7; - block[16]= ((d+b)*qmul) >> 7; - block[32]= ((a-c)*qmul) >> 7; - block[48]= ((d-b)*qmul) >> 7; -} - void ff_h264_weight_pixels16_8_mmi(uint8_t *block, ptrdiff_t stride, int height, int log2_denom, int weight, int offset) {