avcodec/mips: [loongson] optimize put_hevc_qpel_bi_h_8 with mmi.

Optimize put_hevc_qpel_bi_h_8 with mmi in the case width=4/8/12/16/24/32/48/64.
This optimization improved HEVC decoding performance 2.1%(2.34x to 2.39x, tested on loongson 3A3000).

Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>

3 files changed, 150 insertions, 0 deletions

diff --git a/libavcodec/mips/hevcdsp_init_mips.c b/libavcodec/mips/hevcdsp_init_mips.c
index bf7d823b56..6c0533ca7c 100644
--- a/libavcodec/mips/hevcdsp_init_mips.c
+++ b/libavcodec/mips/hevcdsp_init_mips.c
@@ -34,6 +34,15 @@ static av_cold void hevc_dsp_init_mmi(HEVCDSPContext *c,
         c->put_hevc_qpel[8][1][1] = ff_hevc_put_hevc_qpel_hv48_8_mmi;
         c->put_hevc_qpel[9][1][1] = ff_hevc_put_hevc_qpel_hv64_8_mmi;
 
+        c->put_hevc_qpel_bi[1][0][1] = ff_hevc_put_hevc_qpel_bi_h4_8_mmi;
+        c->put_hevc_qpel_bi[3][0][1] = ff_hevc_put_hevc_qpel_bi_h8_8_mmi;
+        c->put_hevc_qpel_bi[4][0][1] = ff_hevc_put_hevc_qpel_bi_h12_8_mmi;
+        c->put_hevc_qpel_bi[5][0][1] = ff_hevc_put_hevc_qpel_bi_h16_8_mmi;
+        c->put_hevc_qpel_bi[6][0][1] = ff_hevc_put_hevc_qpel_bi_h24_8_mmi;
+        c->put_hevc_qpel_bi[7][0][1] = ff_hevc_put_hevc_qpel_bi_h32_8_mmi;
+        c->put_hevc_qpel_bi[8][0][1] = ff_hevc_put_hevc_qpel_bi_h48_8_mmi;
+        c->put_hevc_qpel_bi[9][0][1] = ff_hevc_put_hevc_qpel_bi_h64_8_mmi;
+
         c->put_hevc_qpel_bi[1][1][1] = ff_hevc_put_hevc_qpel_bi_hv4_8_mmi;
         c->put_hevc_qpel_bi[3][1][1] = ff_hevc_put_hevc_qpel_bi_hv8_8_mmi;
         c->put_hevc_qpel_bi[4][1][1] = ff_hevc_put_hevc_qpel_bi_hv12_8_mmi;
diff --git a/libavcodec/mips/hevcdsp_mips.h b/libavcodec/mips/hevcdsp_mips.h
index fe4faae41a..41f56c8e97 100644
--- a/libavcodec/mips/hevcdsp_mips.h
+++ b/libavcodec/mips/hevcdsp_mips.h
@@ -524,6 +524,15 @@ L_BI_MC(qpel, hv, 32, mmi);
 L_BI_MC(qpel, hv, 48, mmi);
 L_BI_MC(qpel, hv, 64, mmi);
 
+L_BI_MC(qpel, h, 4, mmi);
+L_BI_MC(qpel, h, 8, mmi);
+L_BI_MC(qpel, h, 12, mmi);
+L_BI_MC(qpel, h, 16, mmi);
+L_BI_MC(qpel, h, 24, mmi);
+L_BI_MC(qpel, h, 32, mmi);
+L_BI_MC(qpel, h, 48, mmi);
+L_BI_MC(qpel, h, 64, mmi);
+
 L_BI_MC(epel, hv, 4, mmi);
 L_BI_MC(epel, hv, 8, mmi);
 L_BI_MC(epel, hv, 12, mmi);
diff --git a/libavcodec/mips/hevcdsp_mmi.c b/libavcodec/mips/hevcdsp_mmi.c
index 42795e4d2c..3a7ce54222 100644
--- a/libavcodec/mips/hevcdsp_mmi.c
+++ b/libavcodec/mips/hevcdsp_mmi.c
@@ -217,6 +217,138 @@ PUT_HEVC_QPEL_HV(32, 8, -32, -64);
 PUT_HEVC_QPEL_HV(48, 12, -48, -96);
 PUT_HEVC_QPEL_HV(64, 16, -64, -128);
 
+#define PUT_HEVC_QPEL_BI_H(w, x_step, src_step, src2_step, dst_step)    \
+void ff_hevc_put_hevc_qpel_bi_h##w##_8_mmi(uint8_t *_dst,               \
+                                           ptrdiff_t _dststride,        \
+                                           uint8_t *_src,               \
+                                           ptrdiff_t _srcstride,        \
+                                           int16_t *src2, int height,   \
+                                           intptr_t mx, intptr_t my,    \
+                                           int width)                   \
+{                                                                       \
+    int x, y;                                                           \
+    pixel        *src       = (pixel*)_src - 3;                         \
+    ptrdiff_t     srcstride = _srcstride / sizeof(pixel);               \
+    pixel *dst          = (pixel *)_dst;                                \
+    ptrdiff_t dststride = _dststride / sizeof(pixel);                   \
+    const int8_t *filter    = ff_hevc_qpel_filters[mx - 1];             \
+    uint64_t ftmp[20];                                                  \
+    uint64_t rtmp[1];                                                   \
+    int shift = 7;                                                      \
+    int offset = 64;                                                    \
+                                                                        \
+    x = width >> 2;                                                     \
+    y = height;                                                         \
+    __asm__ volatile(                                                   \
+        MMI_LDC1(%[ftmp1], %[filter], 0x00)                             \
+        "li           %[rtmp0],      0x08                       \n\t"   \
+        "dmtc1        %[rtmp0],      %[ftmp0]                   \n\t"   \
+        "punpckhbh    %[ftmp2],      %[ftmp0],      %[ftmp1]    \n\t"   \
+        "punpcklbh    %[ftmp1],      %[ftmp0],      %[ftmp1]    \n\t"   \
+        "psrah        %[ftmp1],      %[ftmp1],      %[ftmp0]    \n\t"   \
+        "psrah        %[ftmp2],      %[ftmp2],      %[ftmp0]    \n\t"   \
+        "xor          %[ftmp0],      %[ftmp0],      %[ftmp0]    \n\t"   \
+        "punpcklhw    %[offset],     %[offset],     %[offset]   \n\t"   \
+        "punpcklwd    %[offset],     %[offset],     %[offset]   \n\t"   \
+                                                                        \
+        "1:                                                     \n\t"   \
+        "li           %[x],        " #x_step "                  \n\t"   \
+        "2:                                                     \n\t"   \
+        "gsldlc1      %[ftmp3],      0x07(%[src])               \n\t"   \
+        "gsldrc1      %[ftmp3],      0x00(%[src])               \n\t"   \
+        "gsldlc1      %[ftmp4],      0x08(%[src])               \n\t"   \
+        "gsldrc1      %[ftmp4],      0x01(%[src])               \n\t"   \
+        "gsldlc1      %[ftmp5],      0x09(%[src])               \n\t"   \
+        "gsldrc1      %[ftmp5],      0x02(%[src])               \n\t"   \
+        "gsldlc1      %[ftmp6],      0x0a(%[src])               \n\t"   \
+        "gsldrc1      %[ftmp6],      0x03(%[src])               \n\t"   \
+        "punpcklbh    %[ftmp7],      %[ftmp3],      %[ftmp0]    \n\t"   \
+        "punpckhbh    %[ftmp8],      %[ftmp3],      %[ftmp0]    \n\t"   \
+        "pmullh       %[ftmp7],      %[ftmp7],      %[ftmp1]    \n\t"   \
+        "pmullh       %[ftmp8],      %[ftmp8],      %[ftmp2]    \n\t"   \
+        "paddh        %[ftmp3],      %[ftmp7],      %[ftmp8]    \n\t"   \
+        "punpcklbh    %[ftmp7],      %[ftmp4],      %[ftmp0]    \n\t"   \
+        "punpckhbh    %[ftmp8],      %[ftmp4],      %[ftmp0]    \n\t"   \
+        "pmullh       %[ftmp7],      %[ftmp7],      %[ftmp1]    \n\t"   \
+        "pmullh       %[ftmp8],      %[ftmp8],      %[ftmp2]    \n\t"   \
+        "paddh        %[ftmp4],      %[ftmp7],      %[ftmp8]    \n\t"   \
+        "punpcklbh    %[ftmp7],      %[ftmp5],      %[ftmp0]    \n\t"   \
+        "punpckhbh    %[ftmp8],      %[ftmp5],      %[ftmp0]    \n\t"   \
+        "pmullh       %[ftmp7],      %[ftmp7],      %[ftmp1]    \n\t"   \
+        "pmullh       %[ftmp8],      %[ftmp8],      %[ftmp2]    \n\t"   \
+        "paddh        %[ftmp5],      %[ftmp7],      %[ftmp8]    \n\t"   \
+        "punpcklbh    %[ftmp7],      %[ftmp6],      %[ftmp0]    \n\t"   \
+        "punpckhbh    %[ftmp8],      %[ftmp6],      %[ftmp0]    \n\t"   \
+        "pmullh       %[ftmp7],      %[ftmp7],      %[ftmp1]    \n\t"   \
+        "pmullh       %[ftmp8],      %[ftmp8],      %[ftmp2]    \n\t"   \
+        "paddh        %[ftmp6],      %[ftmp7],      %[ftmp8]    \n\t"   \
+        TRANSPOSE_4H(%[ftmp3], %[ftmp4], %[ftmp5], %[ftmp6],            \
+                     %[ftmp7], %[ftmp8], %[ftmp9], %[ftmp10])           \
+        "paddh        %[ftmp3],      %[ftmp3],      %[ftmp4]    \n\t"   \
+        "paddh        %[ftmp5],      %[ftmp5],      %[ftmp6]    \n\t"   \
+        "paddh        %[ftmp3],      %[ftmp3],      %[ftmp5]    \n\t"   \
+        "paddh        %[ftmp3],      %[ftmp3],      %[offset]   \n\t"   \
+        "gsldlc1      %[ftmp4],      0x07(%[src2])              \n\t"   \
+        "gsldrc1      %[ftmp4],      0x00(%[src2])              \n\t"   \
+        "li           %[rtmp0],      0x10                       \n\t"   \
+        "dmtc1        %[rtmp0],      %[ftmp8]                   \n\t"   \
+        "punpcklhw    %[ftmp5],      %[ftmp0],      %[ftmp3]    \n\t"   \
+        "punpckhhw    %[ftmp6],      %[ftmp0],      %[ftmp3]    \n\t"   \
+        "punpckhhw    %[ftmp3],      %[ftmp0],      %[ftmp4]    \n\t"   \
+        "punpcklhw    %[ftmp4],      %[ftmp0],      %[ftmp4]    \n\t"   \
+        "psraw        %[ftmp5],      %[ftmp5],      %[ftmp8]    \n\t"   \
+        "psraw        %[ftmp6],      %[ftmp6],      %[ftmp8]    \n\t"   \
+        "psraw        %[ftmp3],      %[ftmp3],      %[ftmp8]    \n\t"   \
+        "psraw        %[ftmp4],      %[ftmp4],      %[ftmp8]    \n\t"   \
+        "paddw        %[ftmp5],      %[ftmp5],      %[ftmp4]    \n\t"   \
+        "paddw        %[ftmp6],      %[ftmp6],      %[ftmp3]    \n\t"   \
+        "psraw        %[ftmp5],      %[ftmp5],      %[shift]    \n\t"   \
+        "psraw        %[ftmp6],      %[ftmp6],      %[shift]    \n\t"   \
+        "packsswh     %[ftmp5],      %[ftmp5],      %[ftmp6]    \n\t"   \
+        "pcmpgth      %[ftmp7],      %[ftmp5],      %[ftmp0]    \n\t"   \
+        "and          %[ftmp3],      %[ftmp5],      %[ftmp7]    \n\t"   \
+        "packushb     %[ftmp3],      %[ftmp3],      %[ftmp3]    \n\t"   \
+        "gsswlc1      %[ftmp3],      0x03(%[dst])               \n\t"   \
+        "gsswrc1      %[ftmp3],      0x00(%[dst])               \n\t"   \
+                                                                        \
+        "daddi        %[x],          %[x],         -0x01        \n\t"   \
+        PTR_ADDIU    "%[src],        %[src],        0x04        \n\t"   \
+        PTR_ADDIU    "%[dst],        %[dst],        0x04        \n\t"   \
+        PTR_ADDIU    "%[src2],       %[src2],       0x08        \n\t"   \
+        "bnez         %[x],          2b                         \n\t"   \
+                                                                        \
+        "daddi        %[y],          %[y],         -0x01        \n\t"   \
+        PTR_ADDIU    "%[src],        %[src],     " #src_step "  \n\t"   \
+        PTR_ADDIU    "%[dst],        %[dst],     " #dst_step "  \n\t"   \
+        PTR_ADDIU    "%[src2],       %[src2],    " #src2_step " \n\t"   \
+        PTR_ADDU     "%[src],        %[src],    %[src_stride]   \n\t"   \
+        PTR_ADDU     "%[dst],        %[dst],    %[dst_stride]   \n\t"   \
+        PTR_ADDIU    "%[src2],       %[src2],       0x80        \n\t"   \
+        "bnez         %[y],          1b                         \n\t"   \
+        : [ftmp0]"=&f"(ftmp[0]), [ftmp1]"=&f"(ftmp[1]),                 \
+          [ftmp2]"=&f"(ftmp[2]), [ftmp3]"=&f"(ftmp[3]),                 \
+          [ftmp4]"=&f"(ftmp[4]), [ftmp5]"=&f"(ftmp[5]),                 \
+          [ftmp6]"=&f"(ftmp[6]), [ftmp7]"=&f"(ftmp[7]),                 \
+          [ftmp8]"=&f"(ftmp[8]), [ftmp9]"=&f"(ftmp[9]),                 \
+          [ftmp10]"=&f"(ftmp[10]), [ftmp11]"=&f"(ftmp[11]),             \
+          [ftmp12]"=&f"(ftmp[12]), [src2]"+&r"(src2),                   \
+          [dst]"+&r"(dst), [src]"+&r"(src), [y]"+&r"(y), [x]"=&r"(x),   \
+          [offset]"+&f"(offset), [rtmp0]"=&r"(rtmp[0])                  \
+        : [src_stride]"r"(srcstride), [dst_stride]"r"(dststride),       \
+          [filter]"r"(filter), [shift]"f"(shift)                        \
+        : "memory"                                                      \
+    );                                                                  \
+}
+
+PUT_HEVC_QPEL_BI_H(4, 1, -4, -8, -4);
+PUT_HEVC_QPEL_BI_H(8, 2, -8, -16, -8);
+PUT_HEVC_QPEL_BI_H(12, 3, -12, -24, -12);
+PUT_HEVC_QPEL_BI_H(16, 4, -16, -32, -16);
+PUT_HEVC_QPEL_BI_H(24, 6, -24, -48, -24);
+PUT_HEVC_QPEL_BI_H(32, 8, -32, -64, -32);
+PUT_HEVC_QPEL_BI_H(48, 12, -48, -96, -48);
+PUT_HEVC_QPEL_BI_H(64, 16, -64, -128, -64);
+
 #define PUT_HEVC_QPEL_BI_HV(w, x_step, src_step, src2_step, dst_step)   \
 void ff_hevc_put_hevc_qpel_bi_hv##w##_8_mmi(uint8_t *_dst,              \
                                             ptrdiff_t _dststride,       \