avcodec/mips: MSA (MIPS-SIMD-Arch) optimizations for HEVC bi mc functions

This patch adds MSA (MIPS-SIMD-Arch) optimizations for HEVC bi mc functions (qpel as well as epel) in new file hevc_mc_bi_msa.c
Adds new generic macros (needed for this patch) in libavutil/mips/generic_macros_msa.h
Adds HEVC specific macros (needed for this patch) in libavcodec/mips/hevc_macros_msa.h

Signed-off-by: Shivraj Patil <shivraj.patil@imgtec.com>
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>

1 files changed, 68 insertions, 0 deletions

diff --git a/libavcodec/mips/hevcdsp_init_mips.c b/libavcodec/mips/hevcdsp_init_mips.c
index d2e3c60546..1d8ed9a868 100644
--- a/libavcodec/mips/hevcdsp_init_mips.c
+++ b/libavcodec/mips/hevcdsp_init_mips.c
@@ -96,6 +96,74 @@ static av_cold void hevc_dsp_init_msa(HEVCDSPContext *c,
         c->put_hevc_qpel_uni[7][1][1] = ff_hevc_put_hevc_uni_qpel_hv32_8_msa;
         c->put_hevc_qpel_uni[8][1][1] = ff_hevc_put_hevc_uni_qpel_hv48_8_msa;
         c->put_hevc_qpel_uni[9][1][1] = ff_hevc_put_hevc_uni_qpel_hv64_8_msa;
+
+        c->put_hevc_qpel_bi[1][0][0] = ff_hevc_put_hevc_bi_pel_pixels4_8_msa;
+        c->put_hevc_qpel_bi[3][0][0] = ff_hevc_put_hevc_bi_pel_pixels8_8_msa;
+        c->put_hevc_qpel_bi[4][0][0] = ff_hevc_put_hevc_bi_pel_pixels12_8_msa;
+        c->put_hevc_qpel_bi[5][0][0] = ff_hevc_put_hevc_bi_pel_pixels16_8_msa;
+        c->put_hevc_qpel_bi[6][0][0] = ff_hevc_put_hevc_bi_pel_pixels24_8_msa;
+        c->put_hevc_qpel_bi[7][0][0] = ff_hevc_put_hevc_bi_pel_pixels32_8_msa;
+        c->put_hevc_qpel_bi[8][0][0] = ff_hevc_put_hevc_bi_pel_pixels48_8_msa;
+        c->put_hevc_qpel_bi[9][0][0] = ff_hevc_put_hevc_bi_pel_pixels64_8_msa;
+
+        c->put_hevc_qpel_bi[1][0][1] = ff_hevc_put_hevc_bi_qpel_h4_8_msa;
+        c->put_hevc_qpel_bi[3][0][1] = ff_hevc_put_hevc_bi_qpel_h8_8_msa;
+        c->put_hevc_qpel_bi[4][0][1] = ff_hevc_put_hevc_bi_qpel_h12_8_msa;
+        c->put_hevc_qpel_bi[5][0][1] = ff_hevc_put_hevc_bi_qpel_h16_8_msa;
+        c->put_hevc_qpel_bi[6][0][1] = ff_hevc_put_hevc_bi_qpel_h24_8_msa;
+        c->put_hevc_qpel_bi[7][0][1] = ff_hevc_put_hevc_bi_qpel_h32_8_msa;
+        c->put_hevc_qpel_bi[8][0][1] = ff_hevc_put_hevc_bi_qpel_h48_8_msa;
+        c->put_hevc_qpel_bi[9][0][1] = ff_hevc_put_hevc_bi_qpel_h64_8_msa;
+
+        c->put_hevc_qpel_bi[1][1][0] = ff_hevc_put_hevc_bi_qpel_v4_8_msa;
+        c->put_hevc_qpel_bi[3][1][0] = ff_hevc_put_hevc_bi_qpel_v8_8_msa;
+        c->put_hevc_qpel_bi[4][1][0] = ff_hevc_put_hevc_bi_qpel_v12_8_msa;
+        c->put_hevc_qpel_bi[5][1][0] = ff_hevc_put_hevc_bi_qpel_v16_8_msa;
+        c->put_hevc_qpel_bi[6][1][0] = ff_hevc_put_hevc_bi_qpel_v24_8_msa;
+        c->put_hevc_qpel_bi[7][1][0] = ff_hevc_put_hevc_bi_qpel_v32_8_msa;
+        c->put_hevc_qpel_bi[8][1][0] = ff_hevc_put_hevc_bi_qpel_v48_8_msa;
+        c->put_hevc_qpel_bi[9][1][0] = ff_hevc_put_hevc_bi_qpel_v64_8_msa;
+
+        c->put_hevc_qpel_bi[1][1][1] = ff_hevc_put_hevc_bi_qpel_hv4_8_msa;
+        c->put_hevc_qpel_bi[3][1][1] = ff_hevc_put_hevc_bi_qpel_hv8_8_msa;
+        c->put_hevc_qpel_bi[4][1][1] = ff_hevc_put_hevc_bi_qpel_hv12_8_msa;
+        c->put_hevc_qpel_bi[5][1][1] = ff_hevc_put_hevc_bi_qpel_hv16_8_msa;
+        c->put_hevc_qpel_bi[6][1][1] = ff_hevc_put_hevc_bi_qpel_hv24_8_msa;
+        c->put_hevc_qpel_bi[7][1][1] = ff_hevc_put_hevc_bi_qpel_hv32_8_msa;
+        c->put_hevc_qpel_bi[8][1][1] = ff_hevc_put_hevc_bi_qpel_hv48_8_msa;
+        c->put_hevc_qpel_bi[9][1][1] = ff_hevc_put_hevc_bi_qpel_hv64_8_msa;
+
+        c->put_hevc_epel_bi[1][0][0] = ff_hevc_put_hevc_bi_pel_pixels4_8_msa;
+        c->put_hevc_epel_bi[2][0][0] = ff_hevc_put_hevc_bi_pel_pixels6_8_msa;
+        c->put_hevc_epel_bi[3][0][0] = ff_hevc_put_hevc_bi_pel_pixels8_8_msa;
+        c->put_hevc_epel_bi[4][0][0] = ff_hevc_put_hevc_bi_pel_pixels12_8_msa;
+        c->put_hevc_epel_bi[5][0][0] = ff_hevc_put_hevc_bi_pel_pixels16_8_msa;
+        c->put_hevc_epel_bi[6][0][0] = ff_hevc_put_hevc_bi_pel_pixels24_8_msa;
+        c->put_hevc_epel_bi[7][0][0] = ff_hevc_put_hevc_bi_pel_pixels32_8_msa;
+
+        c->put_hevc_epel_bi[1][0][1] = ff_hevc_put_hevc_bi_epel_h4_8_msa;
+        c->put_hevc_epel_bi[2][0][1] = ff_hevc_put_hevc_bi_epel_h6_8_msa;
+        c->put_hevc_epel_bi[3][0][1] = ff_hevc_put_hevc_bi_epel_h8_8_msa;
+        c->put_hevc_epel_bi[4][0][1] = ff_hevc_put_hevc_bi_epel_h12_8_msa;
+        c->put_hevc_epel_bi[5][0][1] = ff_hevc_put_hevc_bi_epel_h16_8_msa;
+        c->put_hevc_epel_bi[6][0][1] = ff_hevc_put_hevc_bi_epel_h24_8_msa;
+        c->put_hevc_epel_bi[7][0][1] = ff_hevc_put_hevc_bi_epel_h32_8_msa;
+
+        c->put_hevc_epel_bi[1][1][0] = ff_hevc_put_hevc_bi_epel_v4_8_msa;
+        c->put_hevc_epel_bi[2][1][0] = ff_hevc_put_hevc_bi_epel_v6_8_msa;
+        c->put_hevc_epel_bi[3][1][0] = ff_hevc_put_hevc_bi_epel_v8_8_msa;
+        c->put_hevc_epel_bi[4][1][0] = ff_hevc_put_hevc_bi_epel_v12_8_msa;
+        c->put_hevc_epel_bi[5][1][0] = ff_hevc_put_hevc_bi_epel_v16_8_msa;
+        c->put_hevc_epel_bi[6][1][0] = ff_hevc_put_hevc_bi_epel_v24_8_msa;
+        c->put_hevc_epel_bi[7][1][0] = ff_hevc_put_hevc_bi_epel_v32_8_msa;
+
+        c->put_hevc_epel_bi[1][1][1] = ff_hevc_put_hevc_bi_epel_hv4_8_msa;
+        c->put_hevc_epel_bi[2][1][1] = ff_hevc_put_hevc_bi_epel_hv6_8_msa;
+        c->put_hevc_epel_bi[3][1][1] = ff_hevc_put_hevc_bi_epel_hv8_8_msa;
+        c->put_hevc_epel_bi[4][1][1] = ff_hevc_put_hevc_bi_epel_hv12_8_msa;
+        c->put_hevc_epel_bi[5][1][1] = ff_hevc_put_hevc_bi_epel_hv16_8_msa;
+        c->put_hevc_epel_bi[6][1][1] = ff_hevc_put_hevc_bi_epel_hv24_8_msa;
+        c->put_hevc_epel_bi[7][1][1] = ff_hevc_put_hevc_bi_epel_hv32_8_msa;
     }
 }
 #endif  // #if HAVE_MSA