avcodec: [loongarch] Optimize Hevc_mc_uni/w with LSX.

ffmpeg -i 5_h265_1080p_60fps_3Mbps.mkv -f rawvideo -y /dev/null -an
before: 182fps
after : 191fps
Signed-off-by: Hao Chen <chenhao@loongson.cn>
Reviewed-by: 殷时友 <yinshiyou-hf@loongson.cn>
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>

libavcodec/loongarch/Makefile

@@ -28,4 +28,6 @@ LSX-OBJS-$(CONFIG_VP9_DECODER)        += loongarch/vp9_mc_lsx.o \
 LSX-OBJS-$(CONFIG_HEVC_DECODER)       += loongarch/hevcdsp_lsx.o \
                                          loongarch/hevc_idct_lsx.o \
                                          loongarch/hevc_lpf_sao_lsx.o \
-                                         loongarch/hevc_mc_bi_lsx.o
+                                         loongarch/hevc_mc_bi_lsx.o \
+                                         loongarch/hevc_mc_uni_lsx.o \
+                                         loongarch/hevc_mc_uniw_lsx.o

libavcodec/loongarch/hevcdsp_init_loongarch.c

@@ -137,6 +137,36 @@ void ff_hevc_dsp_init_loongarch(HEVCDSPContext *c, const int bit_depth)
             c->put_hevc_epel_bi[6][1][1] = ff_hevc_put_hevc_bi_epel_hv24_8_lsx;
             c->put_hevc_epel_bi[7][1][1] = ff_hevc_put_hevc_bi_epel_hv32_8_lsx;
 
+            c->put_hevc_qpel_uni[9][0][1] = ff_hevc_put_hevc_uni_qpel_h64_8_lsx;
+
+            c->put_hevc_qpel_uni[6][1][0] = ff_hevc_put_hevc_uni_qpel_v24_8_lsx;
+            c->put_hevc_qpel_uni[7][1][0] = ff_hevc_put_hevc_uni_qpel_v32_8_lsx;
+            c->put_hevc_qpel_uni[8][1][0] = ff_hevc_put_hevc_uni_qpel_v48_8_lsx;
+            c->put_hevc_qpel_uni[9][1][0] = ff_hevc_put_hevc_uni_qpel_v64_8_lsx;
+
+            c->put_hevc_qpel_uni[3][1][1] = ff_hevc_put_hevc_uni_qpel_hv8_8_lsx;
+            c->put_hevc_qpel_uni[5][1][1] = ff_hevc_put_hevc_uni_qpel_hv16_8_lsx;
+            c->put_hevc_qpel_uni[6][1][1] = ff_hevc_put_hevc_uni_qpel_hv24_8_lsx;
+            c->put_hevc_qpel_uni[7][1][1] = ff_hevc_put_hevc_uni_qpel_hv32_8_lsx;
+            c->put_hevc_qpel_uni[8][1][1] = ff_hevc_put_hevc_uni_qpel_hv48_8_lsx;
+            c->put_hevc_qpel_uni[9][1][1] = ff_hevc_put_hevc_uni_qpel_hv64_8_lsx;
+
+            c->put_hevc_epel_uni[6][1][0] = ff_hevc_put_hevc_uni_epel_v24_8_lsx;
+            c->put_hevc_epel_uni[7][1][0] = ff_hevc_put_hevc_uni_epel_v32_8_lsx;
+
+            c->put_hevc_epel_uni[3][1][1] = ff_hevc_put_hevc_uni_epel_hv8_8_lsx;
+            c->put_hevc_epel_uni[4][1][1] = ff_hevc_put_hevc_uni_epel_hv12_8_lsx;
+            c->put_hevc_epel_uni[5][1][1] = ff_hevc_put_hevc_uni_epel_hv16_8_lsx;
+            c->put_hevc_epel_uni[6][1][1] = ff_hevc_put_hevc_uni_epel_hv24_8_lsx;
+            c->put_hevc_epel_uni[7][1][1] = ff_hevc_put_hevc_uni_epel_hv32_8_lsx;
+
+            c->put_hevc_qpel_uni_w[3][1][1] = ff_hevc_put_hevc_uni_w_qpel_hv8_8_lsx;
+            c->put_hevc_qpel_uni_w[5][1][1] = ff_hevc_put_hevc_uni_w_qpel_hv16_8_lsx;
+            c->put_hevc_qpel_uni_w[6][1][1] = ff_hevc_put_hevc_uni_w_qpel_hv24_8_lsx;
+            c->put_hevc_qpel_uni_w[7][1][1] = ff_hevc_put_hevc_uni_w_qpel_hv32_8_lsx;
+            c->put_hevc_qpel_uni_w[8][1][1] = ff_hevc_put_hevc_uni_w_qpel_hv48_8_lsx;
+            c->put_hevc_qpel_uni_w[9][1][1] = ff_hevc_put_hevc_uni_w_qpel_hv64_8_lsx;
+
             c->sao_edge_filter[4] = ff_hevc_sao_edge_filter_8_lsx;
 
             c->hevc_h_loop_filter_luma = ff_hevc_loop_filter_luma_h_8_lsx;

libavcodec/loongarch/hevcdsp_lsx.h

@@ -142,6 +142,65 @@ BI_MC(epel, hv, 32);
 
 #undef BI_MC
 
+#define UNI_MC(PEL, DIR, WIDTH)                                              \
+void ff_hevc_put_hevc_uni_##PEL##_##DIR##WIDTH##_8_lsx(uint8_t *dst,         \
+                                                       ptrdiff_t dst_stride, \
+                                                       uint8_t *src,         \
+                                                       ptrdiff_t src_stride, \
+                                                       int height,           \
+                                                       intptr_t mx,          \
+                                                       intptr_t my,          \
+                                                       int width)
+
+UNI_MC(qpel, h, 64);
+
+UNI_MC(qpel, v, 24);
+UNI_MC(qpel, v, 32);
+UNI_MC(qpel, v, 48);
+UNI_MC(qpel, v, 64);
+
+UNI_MC(qpel, hv, 8);
+UNI_MC(qpel, hv, 16);
+UNI_MC(qpel, hv, 24);
+UNI_MC(qpel, hv, 32);
+UNI_MC(qpel, hv, 48);
+UNI_MC(qpel, hv, 64);
+
+UNI_MC(epel, v, 24);
+UNI_MC(epel, v, 32);
+
+UNI_MC(epel, hv, 8);
+UNI_MC(epel, hv, 12);
+UNI_MC(epel, hv, 16);
+UNI_MC(epel, hv, 24);
+UNI_MC(epel, hv, 32);
+
+#undef UNI_MC
+
+#define UNI_W_MC(PEL, DIR, WIDTH)                                       \
+void ff_hevc_put_hevc_uni_w_##PEL##_##DIR##WIDTH##_8_lsx(uint8_t *dst,  \
+                                                         ptrdiff_t      \
+                                                         dst_stride,    \
+                                                         uint8_t *src,  \
+                                                         ptrdiff_t      \
+                                                         src_stride,    \
+                                                         int height,    \
+                                                         int denom,     \
+                                                         int weight,    \
+                                                         int offset,    \
+                                                         intptr_t mx,   \
+                                                         intptr_t my,   \
+                                                         int width)
+
+UNI_W_MC(qpel, hv, 8);
+UNI_W_MC(qpel, hv, 16);
+UNI_W_MC(qpel, hv, 24);
+UNI_W_MC(qpel, hv, 32);
+UNI_W_MC(qpel, hv, 48);
+UNI_W_MC(qpel, hv, 64);
+
+#undef UNI_W_MC
+
 void ff_hevc_loop_filter_luma_h_8_lsx(uint8_t *src, ptrdiff_t stride,
                                       int32_t beta, int32_t *tc,
                                       uint8_t *p_is_pcm, uint8_t *q_is_pcm);