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

This patch adds MSA (MIPS-SIMD-Arch) optimizations for HEVC idct functions in new file hevc_idct_msa.c
Adds new generic macros (needed for this patch) in libavutil/mips/generic_macros_msa.h
Signed-off-by: Shivraj Patil <shivraj.patil@imgtec.com>
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>

libavcodec/mips/Makefile

@@ -25,7 +25,8 @@ MSA-OBJS-$(CONFIG_HEVC_DECODER)           += mips/hevcdsp_msa.o            \
                                              mips/hevc_mc_uni_msa.o        \
                                              mips/hevc_mc_uniw_msa.o       \
                                              mips/hevc_mc_bi_msa.o         \
-                                             mips/hevc_mc_biw_msa.o
+                                             mips/hevc_mc_biw_msa.o        \
+                                             mips/hevc_idct_msa.o
 MSA-OBJS-$(CONFIG_H264DSP)                += mips/h264dsp_msa.o
 LOONGSON3-OBJS-$(CONFIG_H264DSP)          += mips/h264dsp_mmi.o
 LOONGSON3-OBJS-$(CONFIG_H264CHROMA)       += mips/h264chroma_mmi.o

libavcodec/mips/hevcdsp_init_mips.c

@@ -402,6 +402,20 @@ static av_cold void hevc_dsp_init_msa(HEVCDSPContext *c,
         c->put_hevc_epel_bi_w[5][1][1] = ff_hevc_put_hevc_bi_w_epel_hv16_8_msa;
         c->put_hevc_epel_bi_w[6][1][1] = ff_hevc_put_hevc_bi_w_epel_hv24_8_msa;
         c->put_hevc_epel_bi_w[7][1][1] = ff_hevc_put_hevc_bi_w_epel_hv32_8_msa;
+
+        c->idct[0] = ff_hevc_idct_4x4_msa;
+        c->idct[1] = ff_hevc_idct_8x8_msa;
+        c->idct[2] = ff_hevc_idct_16x16_msa;
+        c->idct[3] = ff_hevc_idct_32x32_msa;
+        c->idct_dc[0] = ff_hevc_idct_dc_4x4_msa;
+        c->idct_dc[1] = ff_hevc_idct_dc_8x8_msa;
+        c->idct_dc[2] = ff_hevc_idct_dc_16x16_msa;
+        c->idct_dc[3] = ff_hevc_idct_dc_32x32_msa;
+        c->transform_add[0] = ff_hevc_addblk_4x4_msa;
+        c->transform_add[1] = ff_hevc_addblk_8x8_msa;
+        c->transform_add[2] = ff_hevc_addblk_16x16_msa;
+        c->transform_add[3] = ff_hevc_addblk_32x32_msa;
+        c->idct_4x4_luma = ff_hevc_idct_luma_4x4_msa;
     }
 }
 #endif  // #if HAVE_MSA

libavcodec/mips/hevcdsp_mips.h

@@ -18,6 +18,9 @@
  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  */
 
+#ifndef AVCODEC_MIPS_HEVCDSP_MIPS_H
+#define AVCODEC_MIPS_HEVCDSP_MIPS_H
+
 #include "libavcodec/hevcdsp.h"
 
 #define MC(PEL, DIR, WIDTH)                                                 \
@@ -427,3 +430,23 @@ BI_W_MC(epel, hv, 48);
 BI_W_MC(epel, hv, 64);
 
 #undef BI_W_MC
+
+void ff_hevc_idct_4x4_msa(int16_t *coeffs, int col_limit);
+void ff_hevc_idct_8x8_msa(int16_t *coeffs, int col_limit);
+void ff_hevc_idct_16x16_msa(int16_t *coeffs, int col_limit);
+void ff_hevc_idct_32x32_msa(int16_t *coeffs, int col_limit);
+void ff_hevc_idct_dc_4x4_msa(int16_t *coeffs);
+void ff_hevc_idct_dc_8x8_msa(int16_t *coeffs);
+void ff_hevc_idct_dc_16x16_msa(int16_t *coeffs);
+void ff_hevc_idct_dc_32x32_msa(int16_t *coeffs);
+void ff_hevc_addblk_4x4_msa(uint8_t *dst, int16_t *pi16Coeffs,
+                            ptrdiff_t stride);
+void ff_hevc_addblk_8x8_msa(uint8_t *dst, int16_t *pi16Coeffs,
+                            ptrdiff_t stride);
+void ff_hevc_addblk_16x16_msa(uint8_t *dst, int16_t *pi16Coeffs,
+                              ptrdiff_t stride);
+void ff_hevc_addblk_32x32_msa(uint8_t *dst, int16_t *pi16Coeffs,
+                              ptrdiff_t stride);
+void ff_hevc_idct_luma_4x4_msa(int16_t *pi16Coeffs);
+
+#endif  // #ifndef AVCODEC_MIPS_HEVCDSP_MIPS_H