avcodec/mips: [loongson] optimize put_hevc_qpel_h_8 with mmi.

Optimize put_hevc_qpel_h_8 with mmi in the case width=4/8/12/16/24/32/48/64.
This optimization improved HEVC decoding performance 2%(2.39x to 2.44x, tested on loongson 3A3000).

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

3 files changed, 115 insertions, 0 deletions

diff --git a/libavcodec/mips/hevcdsp_init_mips.c b/libavcodec/mips/hevcdsp_init_mips.c
index 6c0533ca7c..88337f462e 100644
--- a/libavcodec/mips/hevcdsp_init_mips.c
+++ b/libavcodec/mips/hevcdsp_init_mips.c
@@ -25,6 +25,15 @@ static av_cold void hevc_dsp_init_mmi(HEVCDSPContext *c,
                                       const int bit_depth)
 {
     if (8 == bit_depth) {
+        c->put_hevc_qpel[1][0][1] = ff_hevc_put_hevc_qpel_h4_8_mmi;
+        c->put_hevc_qpel[3][0][1] = ff_hevc_put_hevc_qpel_h8_8_mmi;
+        c->put_hevc_qpel[4][0][1] = ff_hevc_put_hevc_qpel_h12_8_mmi;
+        c->put_hevc_qpel[5][0][1] = ff_hevc_put_hevc_qpel_h16_8_mmi;
+        c->put_hevc_qpel[6][0][1] = ff_hevc_put_hevc_qpel_h24_8_mmi;
+        c->put_hevc_qpel[7][0][1] = ff_hevc_put_hevc_qpel_h32_8_mmi;
+        c->put_hevc_qpel[8][0][1] = ff_hevc_put_hevc_qpel_h48_8_mmi;
+        c->put_hevc_qpel[9][0][1] = ff_hevc_put_hevc_qpel_h64_8_mmi;
+
         c->put_hevc_qpel[1][1][1] = ff_hevc_put_hevc_qpel_hv4_8_mmi;
         c->put_hevc_qpel[3][1][1] = ff_hevc_put_hevc_qpel_hv8_8_mmi;
         c->put_hevc_qpel[4][1][1] = ff_hevc_put_hevc_qpel_hv12_8_mmi;
diff --git a/libavcodec/mips/hevcdsp_mips.h b/libavcodec/mips/hevcdsp_mips.h
index 41f56c8e97..c84e08d160 100644
--- a/libavcodec/mips/hevcdsp_mips.h
+++ b/libavcodec/mips/hevcdsp_mips.h
@@ -488,6 +488,15 @@ void ff_hevc_put_hevc_##PEL##_##DIR##WIDTH##_8_##TYPE(int16_t *dst,          \
                                                       intptr_t mx,           \
                                                       intptr_t my,           \
                                                       int width)
+L_MC(qpel, h, 4, mmi);
+L_MC(qpel, h, 8, mmi);
+L_MC(qpel, h, 12, mmi);
+L_MC(qpel, h, 16, mmi);
+L_MC(qpel, h, 24, mmi);
+L_MC(qpel, h, 32, mmi);
+L_MC(qpel, h, 48, mmi);
+L_MC(qpel, h, 64, mmi);
+
 L_MC(qpel, hv, 4, mmi);
 L_MC(qpel, hv, 8, mmi);
 L_MC(qpel, hv, 12, mmi);
diff --git a/libavcodec/mips/hevcdsp_mmi.c b/libavcodec/mips/hevcdsp_mmi.c
index 3a7ce54222..aa83e1f9ad 100644
--- a/libavcodec/mips/hevcdsp_mmi.c
+++ b/libavcodec/mips/hevcdsp_mmi.c
@@ -23,6 +23,103 @@
 #include "libavcodec/mips/hevcdsp_mips.h"
 #include "libavutil/mips/mmiutils.h"
 
+#define PUT_HEVC_QPEL_H(w, x_step, src_step, dst_step)                   \
+void ff_hevc_put_hevc_qpel_h##w##_8_mmi(int16_t *dst, uint8_t *_src,     \
+                                        ptrdiff_t _srcstride,            \
+                                        int height, intptr_t mx,         \
+                                        intptr_t my, int width)          \
+{                                                                        \
+    int x, y;                                                            \
+    pixel *src = (pixel*)_src - 3;                                       \
+    ptrdiff_t srcstride = _srcstride / sizeof(pixel);                    \
+    uint64_t ftmp[15];                                                   \
+    uint64_t rtmp[1];                                                    \
+    const int8_t *filter = ff_hevc_qpel_filters[mx - 1];                 \
+                                                                         \
+    x = x_step;                                                          \
+    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"    \
+                                                                         \
+        "1:                                                     \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"    \
+        "gssdlc1      %[ftmp3],      0x07(%[dst])               \n\t"    \
+        "gssdrc1      %[ftmp3],      0x00(%[dst])               \n\t"    \
+                                                                         \
+        "daddi        %[x],          %[x],         -0x01        \n\t"    \
+        PTR_ADDIU    "%[src],        %[src],        0x04        \n\t"    \
+        PTR_ADDIU    "%[dst],        %[dst],        0x08        \n\t"    \
+        "bnez         %[x],          2b                         \n\t"    \
+                                                                         \
+        "daddi        %[y],          %[y],         -0x01        \n\t"    \
+        "li           %[x],        " #x_step "                  \n\t"    \
+        PTR_ADDIU    "%[src],        %[src],     " #src_step "  \n\t"    \
+        PTR_ADDIU    "%[dst],        %[dst],     " #dst_step "  \n\t"    \
+        PTR_ADDU     "%[src],        %[src],        %[stride]   \n\t"    \
+        PTR_ADDIU    "%[dst],        %[dst],        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]), [rtmp0]"=&r"(rtmp[0]),                \
+          [src]"+&r"(src), [dst]"+&r"(dst), [y]"+&r"(y),                 \
+          [x]"+&r"(x)                                                    \
+        : [filter]"r"(filter), [stride]"r"(srcstride)                    \
+        : "memory"                                                       \
+    );                                                                   \
+}
+
+PUT_HEVC_QPEL_H(4, 1, -4, -8);
+PUT_HEVC_QPEL_H(8, 2, -8, -16);
+PUT_HEVC_QPEL_H(12, 3, -12, -24);
+PUT_HEVC_QPEL_H(16, 4, -16, -32);
+PUT_HEVC_QPEL_H(24, 6, -24, -48);
+PUT_HEVC_QPEL_H(32, 8, -32, -64);
+PUT_HEVC_QPEL_H(48, 12, -48, -96);
+PUT_HEVC_QPEL_H(64, 16, -64, -128);
+
 #define PUT_HEVC_QPEL_HV(w, x_step, src_step, dst_step)                  \
 void ff_hevc_put_hevc_qpel_hv##w##_8_mmi(int16_t *dst, uint8_t *_src,    \
                                      ptrdiff_t _srcstride,               \