summaryrefslogtreecommitdiff
path: root/libavcodec/mips/hpeldsp_init_mips.c
diff options
context:
space:
mode:
authorJiaxun Yang <jiaxun.yang@flygoat.com>2020-07-18 23:35:40 +0800
committerMichael Niedermayer <michael@niedermayer.cc>2020-07-23 17:21:58 +0200
commite2fa12e3ae0494a72f5d0d4a67633805746391c1 (patch)
tree4e1a4ae3f6d719a68f63957428845d504913eef9 /libavcodec/mips/hpeldsp_init_mips.c
parente387fcd01cb84d9493f3b96158addd2a85f086c6 (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/hpeldsp_init_mips.c')
-rw-r--r--libavcodec/mips/hpeldsp_init_mips.c180
1 files changed, 85 insertions, 95 deletions
diff --git a/libavcodec/mips/hpeldsp_init_mips.c b/libavcodec/mips/hpeldsp_init_mips.c
index d6f7a9793d..77cbe99fa4 100644
--- a/libavcodec/mips/hpeldsp_init_mips.c
+++ b/libavcodec/mips/hpeldsp_init_mips.c
@@ -19,104 +19,94 @@
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
*/
+#include "libavutil/mips/cpu.h"
#include "../hpeldsp.h"
#include "libavcodec/mips/hpeldsp_mips.h"
-#if HAVE_MSA
-static void ff_hpeldsp_init_msa(HpelDSPContext *c, int flags)
-{
- c->put_pixels_tab[0][0] = ff_put_pixels16_msa;
- c->put_pixels_tab[0][1] = ff_put_pixels16_x2_msa;
- c->put_pixels_tab[0][2] = ff_put_pixels16_y2_msa;
- c->put_pixels_tab[0][3] = ff_put_pixels16_xy2_msa;
-
- c->put_pixels_tab[1][0] = ff_put_pixels8_msa;
- c->put_pixels_tab[1][1] = ff_put_pixels8_x2_msa;
- c->put_pixels_tab[1][2] = ff_put_pixels8_y2_msa;
- c->put_pixels_tab[1][3] = ff_put_pixels8_xy2_msa;
-
- c->put_pixels_tab[2][1] = ff_put_pixels4_x2_msa;
- c->put_pixels_tab[2][2] = ff_put_pixels4_y2_msa;
- c->put_pixels_tab[2][3] = ff_put_pixels4_xy2_msa;
-
- c->put_no_rnd_pixels_tab[0][0] = ff_put_pixels16_msa;
- c->put_no_rnd_pixels_tab[0][1] = ff_put_no_rnd_pixels16_x2_msa;
- c->put_no_rnd_pixels_tab[0][2] = ff_put_no_rnd_pixels16_y2_msa;
- c->put_no_rnd_pixels_tab[0][3] = ff_put_no_rnd_pixels16_xy2_msa;
-
- c->put_no_rnd_pixels_tab[1][0] = ff_put_pixels8_msa;
- c->put_no_rnd_pixels_tab[1][1] = ff_put_no_rnd_pixels8_x2_msa;
- c->put_no_rnd_pixels_tab[1][2] = ff_put_no_rnd_pixels8_y2_msa;
- c->put_no_rnd_pixels_tab[1][3] = ff_put_no_rnd_pixels8_xy2_msa;
-
- c->avg_pixels_tab[0][0] = ff_avg_pixels16_msa;
- c->avg_pixels_tab[0][1] = ff_avg_pixels16_x2_msa;
- c->avg_pixels_tab[0][2] = ff_avg_pixels16_y2_msa;
- c->avg_pixels_tab[0][3] = ff_avg_pixels16_xy2_msa;
-
- c->avg_pixels_tab[1][0] = ff_avg_pixels8_msa;
- c->avg_pixels_tab[1][1] = ff_avg_pixels8_x2_msa;
- c->avg_pixels_tab[1][2] = ff_avg_pixels8_y2_msa;
- c->avg_pixels_tab[1][3] = ff_avg_pixels8_xy2_msa;
-
- c->avg_pixels_tab[2][0] = ff_avg_pixels4_msa;
- c->avg_pixels_tab[2][1] = ff_avg_pixels4_x2_msa;
- c->avg_pixels_tab[2][2] = ff_avg_pixels4_y2_msa;
- c->avg_pixels_tab[2][3] = ff_avg_pixels4_xy2_msa;
-}
-#endif // #if HAVE_MSA
-
-#if HAVE_MMI
-static void ff_hpeldsp_init_mmi(HpelDSPContext *c, int flags)
-{
- c->put_pixels_tab[0][0] = ff_put_pixels16_8_mmi;
- c->put_pixels_tab[0][1] = ff_put_pixels16_x2_8_mmi;
- c->put_pixels_tab[0][2] = ff_put_pixels16_y2_8_mmi;
- c->put_pixels_tab[0][3] = ff_put_pixels16_xy2_8_mmi;
-
- c->put_pixels_tab[1][0] = ff_put_pixels8_8_mmi;
- c->put_pixels_tab[1][1] = ff_put_pixels8_x2_8_mmi;
- c->put_pixels_tab[1][2] = ff_put_pixels8_y2_8_mmi;
- c->put_pixels_tab[1][3] = ff_put_pixels8_xy2_8_mmi;
-
- c->put_pixels_tab[2][0] = ff_put_pixels4_8_mmi;
- c->put_pixels_tab[2][1] = ff_put_pixels4_x2_8_mmi;
- c->put_pixels_tab[2][2] = ff_put_pixels4_y2_8_mmi;
- c->put_pixels_tab[2][3] = ff_put_pixels4_xy2_8_mmi;
-
- c->put_no_rnd_pixels_tab[0][0] = ff_put_pixels16_8_mmi;
- c->put_no_rnd_pixels_tab[0][1] = ff_put_no_rnd_pixels16_x2_8_mmi;
- c->put_no_rnd_pixels_tab[0][2] = ff_put_no_rnd_pixels16_y2_8_mmi;
- c->put_no_rnd_pixels_tab[0][3] = ff_put_no_rnd_pixels16_xy2_8_mmi;
-
- c->put_no_rnd_pixels_tab[1][0] = ff_put_pixels8_8_mmi;
- c->put_no_rnd_pixels_tab[1][1] = ff_put_no_rnd_pixels8_x2_8_mmi;
- c->put_no_rnd_pixels_tab[1][2] = ff_put_no_rnd_pixels8_y2_8_mmi;
- c->put_no_rnd_pixels_tab[1][3] = ff_put_no_rnd_pixels8_xy2_8_mmi;
-
- c->avg_pixels_tab[0][0] = ff_avg_pixels16_8_mmi;
- c->avg_pixels_tab[0][1] = ff_avg_pixels16_x2_8_mmi;
- c->avg_pixels_tab[0][2] = ff_avg_pixels16_y2_8_mmi;
- c->avg_pixels_tab[0][3] = ff_avg_pixels16_xy2_8_mmi;
-
- c->avg_pixels_tab[1][0] = ff_avg_pixels8_8_mmi;
- c->avg_pixels_tab[1][1] = ff_avg_pixels8_x2_8_mmi;
- c->avg_pixels_tab[1][2] = ff_avg_pixels8_y2_8_mmi;
- c->avg_pixels_tab[1][3] = ff_avg_pixels8_xy2_8_mmi;
-
- c->avg_pixels_tab[2][0] = ff_avg_pixels4_8_mmi;
- c->avg_pixels_tab[2][1] = ff_avg_pixels4_x2_8_mmi;
- c->avg_pixels_tab[2][2] = ff_avg_pixels4_y2_8_mmi;
- c->avg_pixels_tab[2][3] = ff_avg_pixels4_xy2_8_mmi;
-}
-#endif // #if HAVE_MMI
-
void ff_hpeldsp_init_mips(HpelDSPContext *c, int flags)
{
-#if HAVE_MMI
- ff_hpeldsp_init_mmi(c, flags);
-#endif // #if HAVE_MMI
-#if HAVE_MSA
- ff_hpeldsp_init_msa(c, flags);
-#endif // #if HAVE_MSA
+ int cpu_flags = av_get_cpu_flags();
+
+ if (have_mmi(cpu_flags)) {
+ c->put_pixels_tab[0][0] = ff_put_pixels16_8_mmi;
+ c->put_pixels_tab[0][1] = ff_put_pixels16_x2_8_mmi;
+ c->put_pixels_tab[0][2] = ff_put_pixels16_y2_8_mmi;
+ c->put_pixels_tab[0][3] = ff_put_pixels16_xy2_8_mmi;
+
+ c->put_pixels_tab[1][0] = ff_put_pixels8_8_mmi;
+ c->put_pixels_tab[1][1] = ff_put_pixels8_x2_8_mmi;
+ c->put_pixels_tab[1][2] = ff_put_pixels8_y2_8_mmi;
+ c->put_pixels_tab[1][3] = ff_put_pixels8_xy2_8_mmi;
+
+ c->put_pixels_tab[2][0] = ff_put_pixels4_8_mmi;
+ c->put_pixels_tab[2][1] = ff_put_pixels4_x2_8_mmi;
+ c->put_pixels_tab[2][2] = ff_put_pixels4_y2_8_mmi;
+ c->put_pixels_tab[2][3] = ff_put_pixels4_xy2_8_mmi;
+
+ c->put_no_rnd_pixels_tab[0][0] = ff_put_pixels16_8_mmi;
+ c->put_no_rnd_pixels_tab[0][1] = ff_put_no_rnd_pixels16_x2_8_mmi;
+ c->put_no_rnd_pixels_tab[0][2] = ff_put_no_rnd_pixels16_y2_8_mmi;
+ c->put_no_rnd_pixels_tab[0][3] = ff_put_no_rnd_pixels16_xy2_8_mmi;
+
+ c->put_no_rnd_pixels_tab[1][0] = ff_put_pixels8_8_mmi;
+ c->put_no_rnd_pixels_tab[1][1] = ff_put_no_rnd_pixels8_x2_8_mmi;
+ c->put_no_rnd_pixels_tab[1][2] = ff_put_no_rnd_pixels8_y2_8_mmi;
+ c->put_no_rnd_pixels_tab[1][3] = ff_put_no_rnd_pixels8_xy2_8_mmi;
+
+ c->avg_pixels_tab[0][0] = ff_avg_pixels16_8_mmi;
+ c->avg_pixels_tab[0][1] = ff_avg_pixels16_x2_8_mmi;
+ c->avg_pixels_tab[0][2] = ff_avg_pixels16_y2_8_mmi;
+ c->avg_pixels_tab[0][3] = ff_avg_pixels16_xy2_8_mmi;
+
+ c->avg_pixels_tab[1][0] = ff_avg_pixels8_8_mmi;
+ c->avg_pixels_tab[1][1] = ff_avg_pixels8_x2_8_mmi;
+ c->avg_pixels_tab[1][2] = ff_avg_pixels8_y2_8_mmi;
+ c->avg_pixels_tab[1][3] = ff_avg_pixels8_xy2_8_mmi;
+
+ c->avg_pixels_tab[2][0] = ff_avg_pixels4_8_mmi;
+ c->avg_pixels_tab[2][1] = ff_avg_pixels4_x2_8_mmi;
+ c->avg_pixels_tab[2][2] = ff_avg_pixels4_y2_8_mmi;
+ c->avg_pixels_tab[2][3] = ff_avg_pixels4_xy2_8_mmi;
+ }
+
+ if (have_msa(cpu_flags)) {
+ c->put_pixels_tab[0][0] = ff_put_pixels16_msa;
+ c->put_pixels_tab[0][1] = ff_put_pixels16_x2_msa;
+ c->put_pixels_tab[0][2] = ff_put_pixels16_y2_msa;
+ c->put_pixels_tab[0][3] = ff_put_pixels16_xy2_msa;
+
+ c->put_pixels_tab[1][0] = ff_put_pixels8_msa;
+ c->put_pixels_tab[1][1] = ff_put_pixels8_x2_msa;
+ c->put_pixels_tab[1][2] = ff_put_pixels8_y2_msa;
+ c->put_pixels_tab[1][3] = ff_put_pixels8_xy2_msa;
+
+ c->put_pixels_tab[2][1] = ff_put_pixels4_x2_msa;
+ c->put_pixels_tab[2][2] = ff_put_pixels4_y2_msa;
+ c->put_pixels_tab[2][3] = ff_put_pixels4_xy2_msa;
+
+ c->put_no_rnd_pixels_tab[0][0] = ff_put_pixels16_msa;
+ c->put_no_rnd_pixels_tab[0][1] = ff_put_no_rnd_pixels16_x2_msa;
+ c->put_no_rnd_pixels_tab[0][2] = ff_put_no_rnd_pixels16_y2_msa;
+ c->put_no_rnd_pixels_tab[0][3] = ff_put_no_rnd_pixels16_xy2_msa;
+
+ c->put_no_rnd_pixels_tab[1][0] = ff_put_pixels8_msa;
+ c->put_no_rnd_pixels_tab[1][1] = ff_put_no_rnd_pixels8_x2_msa;
+ c->put_no_rnd_pixels_tab[1][2] = ff_put_no_rnd_pixels8_y2_msa;
+ c->put_no_rnd_pixels_tab[1][3] = ff_put_no_rnd_pixels8_xy2_msa;
+
+ c->avg_pixels_tab[0][0] = ff_avg_pixels16_msa;
+ c->avg_pixels_tab[0][1] = ff_avg_pixels16_x2_msa;
+ c->avg_pixels_tab[0][2] = ff_avg_pixels16_y2_msa;
+ c->avg_pixels_tab[0][3] = ff_avg_pixels16_xy2_msa;
+
+ c->avg_pixels_tab[1][0] = ff_avg_pixels8_msa;
+ c->avg_pixels_tab[1][1] = ff_avg_pixels8_x2_msa;
+ c->avg_pixels_tab[1][2] = ff_avg_pixels8_y2_msa;
+ c->avg_pixels_tab[1][3] = ff_avg_pixels8_xy2_msa;
+
+ c->avg_pixels_tab[2][0] = ff_avg_pixels4_msa;
+ c->avg_pixels_tab[2][1] = ff_avg_pixels4_x2_msa;
+ c->avg_pixels_tab[2][2] = ff_avg_pixels4_y2_msa;
+ c->avg_pixels_tab[2][3] = ff_avg_pixels4_xy2_msa;
+ }
}