avcodec/hevc: Add pel_uni_w_pixels4/6/8/12/16/24/32/48/64 asm opt

tests/checkasm/checkasm:           C       LSX     LASX
put_hevc_pel_uni_w_pixels4_8_c:    2.7     1.0
put_hevc_pel_uni_w_pixels6_8_c:    6.2     2.0     1.5
put_hevc_pel_uni_w_pixels8_8_c:    10.7    2.5     1.7
put_hevc_pel_uni_w_pixels12_8_c:   23.0    5.5     5.0
put_hevc_pel_uni_w_pixels16_8_c:   41.0    8.2     5.0
put_hevc_pel_uni_w_pixels24_8_c:   91.0    19.7    13.2
put_hevc_pel_uni_w_pixels32_8_c:   161.7   32.5    16.2
put_hevc_pel_uni_w_pixels48_8_c:   354.5   73.7    43.0
put_hevc_pel_uni_w_pixels64_8_c:   641.5   130.0   64.2

Speedup of decoding H265 4K 30FPS 30Mbps on 3A6000 with
8 threads is 1fps(47fps-->48fps).

Reviewed-by: yinshiyou-hf@loongson.cn
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>

1 files changed, 43 insertions, 0 deletions

diff --git a/libavcodec/loongarch/hevcdsp_init_loongarch.c b/libavcodec/loongarch/hevcdsp_init_loongarch.c
index a8f753dc86..d0ee99d6b5 100644
--- a/libavcodec/loongarch/hevcdsp_init_loongarch.c
+++ b/libavcodec/loongarch/hevcdsp_init_loongarch.c
@@ -22,6 +22,7 @@
 
 #include "libavutil/loongarch/cpu.h"
 #include "hevcdsp_lsx.h"
+#include "hevcdsp_lasx.h"
 
 void ff_hevc_dsp_init_loongarch(HEVCDSPContext *c, const int bit_depth)
 {
@@ -160,6 +161,26 @@ void ff_hevc_dsp_init_loongarch(HEVCDSPContext *c, const int bit_depth)
             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[1][0][0] = ff_hevc_put_hevc_pel_uni_w_pixels4_8_lsx;
+            c->put_hevc_qpel_uni_w[2][0][0] = ff_hevc_put_hevc_pel_uni_w_pixels6_8_lsx;
+            c->put_hevc_qpel_uni_w[3][0][0] = ff_hevc_put_hevc_pel_uni_w_pixels8_8_lsx;
+            c->put_hevc_qpel_uni_w[4][0][0] = ff_hevc_put_hevc_pel_uni_w_pixels12_8_lsx;
+            c->put_hevc_qpel_uni_w[5][0][0] = ff_hevc_put_hevc_pel_uni_w_pixels16_8_lsx;
+            c->put_hevc_qpel_uni_w[6][0][0] = ff_hevc_put_hevc_pel_uni_w_pixels24_8_lsx;
+            c->put_hevc_qpel_uni_w[7][0][0] = ff_hevc_put_hevc_pel_uni_w_pixels32_8_lsx;
+            c->put_hevc_qpel_uni_w[8][0][0] = ff_hevc_put_hevc_pel_uni_w_pixels48_8_lsx;
+            c->put_hevc_qpel_uni_w[9][0][0] = ff_hevc_put_hevc_pel_uni_w_pixels64_8_lsx;
+
+            c->put_hevc_epel_uni_w[1][0][0] = ff_hevc_put_hevc_pel_uni_w_pixels4_8_lsx;
+            c->put_hevc_epel_uni_w[2][0][0] = ff_hevc_put_hevc_pel_uni_w_pixels6_8_lsx;
+            c->put_hevc_epel_uni_w[3][0][0] = ff_hevc_put_hevc_pel_uni_w_pixels8_8_lsx;
+            c->put_hevc_epel_uni_w[4][0][0] = ff_hevc_put_hevc_pel_uni_w_pixels12_8_lsx;
+            c->put_hevc_epel_uni_w[5][0][0] = ff_hevc_put_hevc_pel_uni_w_pixels16_8_lsx;
+            c->put_hevc_epel_uni_w[6][0][0] = ff_hevc_put_hevc_pel_uni_w_pixels24_8_lsx;
+            c->put_hevc_epel_uni_w[7][0][0] = ff_hevc_put_hevc_pel_uni_w_pixels32_8_lsx;
+            c->put_hevc_epel_uni_w[8][0][0] = ff_hevc_put_hevc_pel_uni_w_pixels48_8_lsx;
+            c->put_hevc_epel_uni_w[9][0][0] = ff_hevc_put_hevc_pel_uni_w_pixels64_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;
@@ -196,4 +217,26 @@ void ff_hevc_dsp_init_loongarch(HEVCDSPContext *c, const int bit_depth)
             c->add_residual[3] = ff_hevc_add_residual32x32_8_lsx;
         }
     }
+
+    if (have_lasx(cpu_flags)) {
+        if (bit_depth == 8) {
+            c->put_hevc_qpel_uni_w[2][0][0] = ff_hevc_put_hevc_pel_uni_w_pixels6_8_lasx;
+            c->put_hevc_qpel_uni_w[3][0][0] = ff_hevc_put_hevc_pel_uni_w_pixels8_8_lasx;
+            c->put_hevc_qpel_uni_w[4][0][0] = ff_hevc_put_hevc_pel_uni_w_pixels12_8_lasx;
+            c->put_hevc_qpel_uni_w[5][0][0] = ff_hevc_put_hevc_pel_uni_w_pixels16_8_lasx;
+            c->put_hevc_qpel_uni_w[6][0][0] = ff_hevc_put_hevc_pel_uni_w_pixels24_8_lasx;
+            c->put_hevc_qpel_uni_w[7][0][0] = ff_hevc_put_hevc_pel_uni_w_pixels32_8_lasx;
+            c->put_hevc_qpel_uni_w[8][0][0] = ff_hevc_put_hevc_pel_uni_w_pixels48_8_lasx;
+            c->put_hevc_qpel_uni_w[9][0][0] = ff_hevc_put_hevc_pel_uni_w_pixels64_8_lasx;
+
+            c->put_hevc_epel_uni_w[2][0][0] = ff_hevc_put_hevc_pel_uni_w_pixels6_8_lasx;
+            c->put_hevc_epel_uni_w[3][0][0] = ff_hevc_put_hevc_pel_uni_w_pixels8_8_lasx;
+            c->put_hevc_epel_uni_w[4][0][0] = ff_hevc_put_hevc_pel_uni_w_pixels12_8_lasx;
+            c->put_hevc_epel_uni_w[5][0][0] = ff_hevc_put_hevc_pel_uni_w_pixels16_8_lasx;
+            c->put_hevc_epel_uni_w[6][0][0] = ff_hevc_put_hevc_pel_uni_w_pixels24_8_lasx;
+            c->put_hevc_epel_uni_w[7][0][0] = ff_hevc_put_hevc_pel_uni_w_pixels32_8_lasx;
+            c->put_hevc_epel_uni_w[8][0][0] = ff_hevc_put_hevc_pel_uni_w_pixels48_8_lasx;
+            c->put_hevc_epel_uni_w[9][0][0] = ff_hevc_put_hevc_pel_uni_w_pixels64_8_lasx;
+        }
+    }
 }