avcodec/mips: loongson optimize hpeldsp with mmi v1

1.the codes are compatible with O32 ABI
2.use uld and mtc1 to workaround cpu 3A2000 gslwlc1 bug (gslwlc1 instruction extension bug in O32 ABI)

Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>

1 files changed, 49 insertions, 0 deletions

diff --git a/libavcodec/mips/hpeldsp_init_mips.c b/libavcodec/mips/hpeldsp_init_mips.c
index 82f2310985..363a04514f 100644
--- a/libavcodec/mips/hpeldsp_init_mips.c
+++ b/libavcodec/mips/hpeldsp_init_mips.c
@@ -1,5 +1,6 @@
 /*
  * Copyright (c) 2015 Parag Salasakar (Parag.Salasakar@imgtec.com)
+ * Copyright (c) 2016 Zhou Xiaoyong <zhouxiaoyong@loongson.cn>
  *
  * This file is part of FFmpeg.
  *
@@ -65,9 +66,57 @@ static void ff_hpeldsp_init_msa(HpelDSPContext *c, int flags)
 }
 #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_MSA
     ff_hpeldsp_init_msa(c, flags);
 #endif  // #if HAVE_MSA
+#if HAVE_MMI
+    ff_hpeldsp_init_mmi(c, flags);
+#endif  // #if HAVE_MMI
 }