Hi,
On Mon, Oct 12, 2015 at 1:37 PM, Christophe Gisquet <
christophe.gisq...@gmail.com> wrote:
> This should be reused for a generic simple_idct10 function.
> Requires a bit of trickery to declare common constants in C.
> ---
> libavcodec/x86/constants.c | 28 +++
> libavcodec/x86/constants.h | 16 ++
> libavcodec/x86/proresdsp.asm | 263 +-------------------------
> libavcodec/x86/simple_idct10_template.asm | 299
> ++++++++++++++++++++++++++++++
> 4 files changed, 346 insertions(+), 260 deletions(-)
> create mode 100644 libavcodec/x86/simple_idct10_template.asm
>
> diff --git a/libavcodec/x86/constants.c b/libavcodec/x86/constants.c
> index 3f3ee0f..9592fa7 100644
> --- a/libavcodec/x86/constants.c
> +++ b/libavcodec/x86/constants.c
> @@ -87,3 +87,31 @@ DECLARE_ALIGNED(32, const ymm_reg, ff_pd_32) = {
> 0x0000002000000020ULL, 0x000
>
> 0x0000002000000020ULL, 0x0000002000000020ULL };
> DECLARE_ALIGNED(32, const ymm_reg, ff_pd_65535)= {
> 0x0000ffff0000ffffULL, 0x0000ffff0000ffffULL,
>
> 0x0000ffff0000ffffULL, 0x0000ffff0000ffffULL };
> +
> +/* simple idct 10 */
> +#define W1sh2 22725 // W1 = 90901 = 22725<<2 + 1
> +#define W2sh2 21407 // W2 = 85627 = 21407<<2 - 1
> +#define W3sh2 19265 // W3 = 77062 = 19265<<2 + 2
> +#define W4sh2 16384 // W4 = 65535 = 16384<<2 - 1
> +#define W5sh2 12873 // W5 = 51491 = 12873<<2 - 1
> +#define W6sh2 8867 // W6 = 35468 = 8867<<2
> +#define W7sh2 4520 // W7 = 18081 = 4520<<2 + 1
> +
> +#define TIMES4(a, b)
> \
> +
>
> (((b)&0xFFFFLL)<<48)+(((a)&0xFFFFLL)<<32)+(((b)&0xFFFFLL)<<16)+((a)&0xFFFFLL),
> \
> +
> (((b)&0xFFFFLL)<<48)+(((a)&0xFFFFLL)<<32)+(((b)&0xFFFFLL)<<16)+((a)&0xFFFFLL)
> +
> +#if ARCH_X86_64
> +DECLARE_ALIGNED(16, const xmm_reg, ff_w4_plus_w2) = { TIMES4(W4sh2,
> W2sh2) };
> +DECLARE_ALIGNED(16, const xmm_reg, ff_w4_min_w2) = { TIMES4(W4sh2,
> -W2sh2) };
> +DECLARE_ALIGNED(16, const xmm_reg, ff_w4_plus_w6) = { TIMES4(W4sh2,
> +W6sh2) };
> +DECLARE_ALIGNED(16, const xmm_reg, ff_w4_min_w6) = { TIMES4(W4sh2,
> -W6sh2) };
> +DECLARE_ALIGNED(16, const xmm_reg, ff_w1_plus_w3) = { TIMES4(W1sh2,
> +W3sh2) };
> +DECLARE_ALIGNED(16, const xmm_reg, ff_w3_min_w1) = { TIMES4(W3sh2,
> -W1sh2) };
> +DECLARE_ALIGNED(16, const xmm_reg, ff_w7_plus_w3) = { TIMES4(W7sh2,
> +W3sh2) };
> +DECLARE_ALIGNED(16, const xmm_reg, ff_w3_min_w7) = { TIMES4(W3sh2,
> -W7sh2) };
> +DECLARE_ALIGNED(16, const xmm_reg, ff_w1_plus_w5) = { TIMES4(W1sh2,
> +W5sh2) };
> +DECLARE_ALIGNED(16, const xmm_reg, ff_w5_min_w1) = { TIMES4(W5sh2,
> -W1sh2) };
> +DECLARE_ALIGNED(16, const xmm_reg, ff_w5_plus_w7) = { TIMES4(W5sh2,
> +W7sh2) };
> +DECLARE_ALIGNED(16, const xmm_reg, ff_w7_min_w5) = { TIMES4(W7sh2,
> -W5sh2) };
> +#endif
No no no no no no no no no no no no no no no ^1000001.
Check out "const" in x86inc.asm.
%macro mymacro 2
const w4_plus_w2 times 4 dw %1, %2
%endmacro
mymacro(W4sh2, W2sh2)
etc.
And then use cextern in the files where you want to access this constant.
Ronald
_______________________________________________
ffmpeg-devel mailing list
ffmpeg-devel@ffmpeg.org
http://ffmpeg.org/mailman/listinfo/ffmpeg-devel
