avcodec/mips: [loongson] optimize put_hevc_qpel_uni_hv_8 with mmi.

Optimize put_hevc_qpel_uni_hv_8 with mmi in the case width=4/8/12/16/24/32/48/64.
This optimization improved HEVC decoding performance 2.7%(2.24x to 2.30x, tested on loongson 3A3000).
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>

libavcodec/mips/hevcdsp_init_mips.c

@@ -54,6 +54,15 @@ static av_cold void hevc_dsp_init_mmi(HEVCDSPContext *c,
         c->put_hevc_epel_bi[5][0][0] = ff_hevc_put_hevc_pel_bi_pixels16_8_mmi;
         c->put_hevc_epel_bi[6][0][0] = ff_hevc_put_hevc_pel_bi_pixels24_8_mmi;
         c->put_hevc_epel_bi[7][0][0] = ff_hevc_put_hevc_pel_bi_pixels32_8_mmi;
+
+        c->put_hevc_qpel_uni[1][1][1] = ff_hevc_put_hevc_qpel_uni_hv4_8_mmi;
+        c->put_hevc_qpel_uni[3][1][1] = ff_hevc_put_hevc_qpel_uni_hv8_8_mmi;
+        c->put_hevc_qpel_uni[4][1][1] = ff_hevc_put_hevc_qpel_uni_hv12_8_mmi;
+        c->put_hevc_qpel_uni[5][1][1] = ff_hevc_put_hevc_qpel_uni_hv16_8_mmi;
+        c->put_hevc_qpel_uni[6][1][1] = ff_hevc_put_hevc_qpel_uni_hv24_8_mmi;
+        c->put_hevc_qpel_uni[7][1][1] = ff_hevc_put_hevc_qpel_uni_hv32_8_mmi;
+        c->put_hevc_qpel_uni[8][1][1] = ff_hevc_put_hevc_qpel_uni_hv48_8_mmi;
+        c->put_hevc_qpel_uni[9][1][1] = ff_hevc_put_hevc_qpel_uni_hv64_8_mmi;
     }
 }
 #endif // #if HAVE_MMI

libavcodec/mips/hevcdsp_mips.h

@@ -525,4 +525,25 @@ L_BI_MC(qpel, hv, 48, mmi);
 L_BI_MC(qpel, hv, 64, mmi);
 
 #undef L_BI_MC
+
+#define L_UNI_MC(PEL, DIR, WIDTH, TYPE)                                         \
+void ff_hevc_put_hevc_##PEL##_uni_##DIR##WIDTH##_8_##TYPE(uint8_t *dst,         \
+                                                          ptrdiff_t dst_stride, \
+                                                          uint8_t *src,         \
+                                                          ptrdiff_t src_stride, \
+                                                          int height,           \
+                                                          intptr_t mx,          \
+                                                          intptr_t my,          \
+                                                          int width)
+
+L_UNI_MC(qpel, hv, 4, mmi);
+L_UNI_MC(qpel, hv, 8, mmi);
+L_UNI_MC(qpel, hv, 12, mmi);
+L_UNI_MC(qpel, hv, 16, mmi);
+L_UNI_MC(qpel, hv, 24, mmi);
+L_UNI_MC(qpel, hv, 32, mmi);
+L_UNI_MC(qpel, hv, 48, mmi);
+L_UNI_MC(qpel, hv, 64, mmi);
+#undef L_UNI_MC
+
 #endif  // #ifndef AVCODEC_MIPS_HEVCDSP_MIPS_H