diff options
author | Jiaxun Yang <jiaxun.yang@flygoat.com> | 2020-07-18 23:35:40 +0800 |
---|---|---|
committer | Michael Niedermayer <michael@niedermayer.cc> | 2020-07-23 17:21:58 +0200 |
commit | e2fa12e3ae0494a72f5d0d4a67633805746391c1 (patch) | |
tree | 4e1a4ae3f6d719a68f63957428845d504913eef9 /libavcodec/mips/idctdsp_init_mips.c | |
parent | e387fcd01cb84d9493f3b96158addd2a85f086c6 (diff) |
libavcodec: Enable runtime detection for MIPS MMI & MSA
Apply optimized functions according to cpuflags.
MSA is usually put after MMI as it's generally faster than MMI.
Signed-off-by: Jiaxun Yang <jiaxun.yang@flygoat.com>
Reviewed-by: Shiyou Yin <yinshiyou-hf@loongson.cn>
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
Diffstat (limited to 'libavcodec/mips/idctdsp_init_mips.c')
-rw-r--r-- | libavcodec/mips/idctdsp_init_mips.c | 74 |
1 files changed, 31 insertions, 43 deletions
diff --git a/libavcodec/mips/idctdsp_init_mips.c b/libavcodec/mips/idctdsp_init_mips.c index 85b76ca478..23efd9ed58 100644 --- a/libavcodec/mips/idctdsp_init_mips.c +++ b/libavcodec/mips/idctdsp_init_mips.c @@ -19,56 +19,44 @@ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA */ +#include "libavutil/mips/cpu.h" #include "idctdsp_mips.h" #include "xvididct_mips.h" -#if HAVE_MSA -static av_cold void idctdsp_init_msa(IDCTDSPContext *c, AVCodecContext *avctx, - unsigned high_bit_depth) +av_cold void ff_idctdsp_init_mips(IDCTDSPContext *c, AVCodecContext *avctx, + unsigned high_bit_depth) { - if ((avctx->lowres != 1) && (avctx->lowres != 2) && (avctx->lowres != 3) && - (avctx->bits_per_raw_sample != 10) && - (avctx->bits_per_raw_sample != 12) && - (avctx->idct_algo == FF_IDCT_AUTO)) { - c->idct_put = ff_simple_idct_put_msa; - c->idct_add = ff_simple_idct_add_msa; - c->idct = ff_simple_idct_msa; - c->perm_type = FF_IDCT_PERM_NONE; - } + int cpu_flags = av_get_cpu_flags(); - c->put_pixels_clamped = ff_put_pixels_clamped_msa; - c->put_signed_pixels_clamped = ff_put_signed_pixels_clamped_msa; - c->add_pixels_clamped = ff_add_pixels_clamped_msa; -} -#endif // #if HAVE_MSA + if (have_mmi(cpu_flags)) { + if ((avctx->lowres != 1) && (avctx->lowres != 2) && (avctx->lowres != 3) && + (avctx->bits_per_raw_sample != 10) && + (avctx->bits_per_raw_sample != 12) && + ((avctx->idct_algo == FF_IDCT_AUTO) || (avctx->idct_algo == FF_IDCT_SIMPLE))) { + c->idct_put = ff_simple_idct_put_8_mmi; + c->idct_add = ff_simple_idct_add_8_mmi; + c->idct = ff_simple_idct_8_mmi; + c->perm_type = FF_IDCT_PERM_NONE; + } -#if HAVE_MMI -static av_cold void idctdsp_init_mmi(IDCTDSPContext *c, AVCodecContext *avctx, - unsigned high_bit_depth) -{ - if ((avctx->lowres != 1) && (avctx->lowres != 2) && (avctx->lowres != 3) && - (avctx->bits_per_raw_sample != 10) && - (avctx->bits_per_raw_sample != 12) && - ((avctx->idct_algo == FF_IDCT_AUTO) || (avctx->idct_algo == FF_IDCT_SIMPLE))) { - c->idct_put = ff_simple_idct_put_8_mmi; - c->idct_add = ff_simple_idct_add_8_mmi; - c->idct = ff_simple_idct_8_mmi; - c->perm_type = FF_IDCT_PERM_NONE; + c->put_pixels_clamped = ff_put_pixels_clamped_mmi; + c->add_pixels_clamped = ff_add_pixels_clamped_mmi; + c->put_signed_pixels_clamped = ff_put_signed_pixels_clamped_mmi; } - c->put_pixels_clamped = ff_put_pixels_clamped_mmi; - c->add_pixels_clamped = ff_add_pixels_clamped_mmi; - c->put_signed_pixels_clamped = ff_put_signed_pixels_clamped_mmi; -} -#endif /* HAVE_MMI */ + if (have_msa(cpu_flags)) { + if ((avctx->lowres != 1) && (avctx->lowres != 2) && (avctx->lowres != 3) && + (avctx->bits_per_raw_sample != 10) && + (avctx->bits_per_raw_sample != 12) && + (avctx->idct_algo == FF_IDCT_AUTO)) { + c->idct_put = ff_simple_idct_put_msa; + c->idct_add = ff_simple_idct_add_msa; + c->idct = ff_simple_idct_msa; + c->perm_type = FF_IDCT_PERM_NONE; + } -av_cold void ff_idctdsp_init_mips(IDCTDSPContext *c, AVCodecContext *avctx, - unsigned high_bit_depth) -{ -#if HAVE_MMI - idctdsp_init_mmi(c, avctx, high_bit_depth); -#endif /* HAVE_MMI */ -#if HAVE_MSA - idctdsp_init_msa(c, avctx, high_bit_depth); -#endif // #if HAVE_MSA + c->put_pixels_clamped = ff_put_pixels_clamped_msa; + c->put_signed_pixels_clamped = ff_put_signed_pixels_clamped_msa; + c->add_pixels_clamped = ff_add_pixels_clamped_msa; + } } |