avcodec/mips/idctdsp_mmi: Version 2 of the optimizations for loongson mmi

    1. no longer use the register names directly and optimized code format
    2. to be compatible with O32, specify type of address variable with mips_reg and handle the address variable with PTR_ operator

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

1 files changed, 154 insertions, 136 deletions

diff --git a/libavcodec/mips/idctdsp_mmi.c b/libavcodec/mips/idctdsp_mmi.c
index 25476f3c40..24beb62df5 100644
--- a/libavcodec/mips/idctdsp_mmi.c
+++ b/libavcodec/mips/idctdsp_mmi.c
@@ -23,63 +23,75 @@
 
 #include "idctdsp_mips.h"
 #include "constants.h"
+#include "libavutil/mips/asmdefs.h"
 
 void ff_put_pixels_clamped_mmi(const int16_t *block,
         uint8_t *av_restrict pixels, ptrdiff_t line_size)
 {
-    const int16_t *p;
-    uint8_t *pix;
-
-    p = block;
-    pix = pixels;
+    double ftmp[8];
+    mips_reg addr[1];
 
     __asm__ volatile (
-        "ldc1 $f0, 0+%3                 \r\n"
-        "ldc1 $f2, 8+%3                 \r\n"
-        "ldc1 $f4, 16+%3                \r\n"
-        "ldc1 $f6, 24+%3                \r\n"
-        "ldc1 $f8, 32+%3                \r\n"
-        "ldc1 $f10, 40+%3               \r\n"
-        "ldc1 $f12, 48+%3               \r\n"
-        "ldc1 $f14, 56+%3               \r\n"
-        "dadd $10, %0, %1               \r\n"
-        "packushb $f0, $f0, $f2         \r\n"
-        "packushb $f4, $f4, $f6         \r\n"
-        "packushb $f8, $f8, $f10        \r\n"
-        "packushb $f12, $f12, $f14      \r\n"
-        "sdc1 $f0, 0(%0)                \r\n"
-        "sdc1 $f4, 0($10)               \r\n"
-        "gssdxc1 $f8, 0($10, %1)        \r\n"
-        "gssdxc1 $f12, 0(%0, %2)        \r\n"
-        ::"r"(pix),"r"((int)line_size),
-          "r"((int)line_size*3),"m"(*p)
-        : "$10","memory"
+        "ldc1       %[ftmp0],   0x00(%[block])                          \n\t"
+        "ldc1       %[ftmp1],   0x08(%[block])                          \n\t"
+        "ldc1       %[ftmp2],   0x10(%[block])                          \n\t"
+        "ldc1       %[ftmp3],   0x18(%[block])                          \n\t"
+        "ldc1       %[ftmp4],   0x20(%[block])                          \n\t"
+        "ldc1       %[ftmp5],   0x28(%[block])                          \n\t"
+        "ldc1       %[ftmp6],   0x30(%[block])                          \n\t"
+        "ldc1       %[ftmp7],   0x38(%[block])                          \n\t"
+        PTR_ADDU   "%[addr0],   %[pixels],      %[line_size]            \n\t"
+        "packushb   %[ftmp0],   %[ftmp0],       %[ftmp1]                \n\t"
+        "packushb   %[ftmp2],   %[ftmp2],       %[ftmp3]                \n\t"
+        "packushb   %[ftmp4],   %[ftmp4],       %[ftmp5]                \n\t"
+        "packushb   %[ftmp6],   %[ftmp6],       %[ftmp7]                \n\t"
+        "sdc1       %[ftmp0],   0x00(%[pixels])                         \n\t"
+        "sdc1       %[ftmp2],   0x00(%[addr0])                          \n\t"
+        "gssdxc1    %[ftmp4],   0x00(%[addr0],  %[line_size])           \n\t"
+        "gssdxc1    %[ftmp6],   0x00(%[pixels], %[line_sizex3])         \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]),
+          [addr0]"=&r"(addr[0]),
+          [pixels]"+&r"(pixels)
+        : [line_size]"r"((mips_reg)line_size),
+          [line_sizex3]"r"((mips_reg)(line_size*3)),
+          [block]"r"(block)
+        : "memory"
     );
 
-    pix += line_size*4;
-    p += 32;
+    pixels += line_size*4;
+    block += 32;
 
     __asm__ volatile (
-        "ldc1 $f0, 0+%3                 \r\n"
-        "ldc1 $f2, 8+%3                 \r\n"
-        "ldc1 $f4, 16+%3                \r\n"
-        "ldc1 $f6, 24+%3                \r\n"
-        "ldc1 $f8, 32+%3                \r\n"
-        "ldc1 $f10, 40+%3               \r\n"
-        "ldc1 $f12, 48+%3               \r\n"
-        "ldc1 $f14, 56+%3               \r\n"
-        "dadd $10, %0, %1               \r\n"
-        "packushb $f0, $f0, $f2         \r\n"
-        "packushb $f4, $f4, $f6         \r\n"
-        "packushb $f8, $f8, $f10        \r\n"
-        "packushb $f12, $f12, $f14      \r\n"
-        "sdc1 $f0, 0(%0)                \r\n"
-        "sdc1 $f4, 0($10)               \r\n"
-        "gssdxc1 $f8, 0($10, %1)        \r\n"
-        "gssdxc1 $f12, 0(%0, %2)        \r\n"
-        ::"r"(pix),"r"((int)line_size),
-          "r"((int)line_size*3),"m"(*p)
-        : "$10","memory"
+        "ldc1       %[ftmp0],   0x00(%[block])                          \n\t"
+        "ldc1       %[ftmp1],   0x08(%[block])                          \n\t"
+        "ldc1       %[ftmp2],   0x10(%[block])                          \n\t"
+        "ldc1       %[ftmp3],   0x18(%[block])                          \n\t"
+        "ldc1       %[ftmp4],   0x20(%[block])                          \n\t"
+        "ldc1       %[ftmp5],   0x28(%[block])                          \n\t"
+        "ldc1       %[ftmp6],   0x30(%[block])                          \n\t"
+        "ldc1       %[ftmp7],   0x38(%[block])                          \n\t"
+        PTR_ADDU   "%[addr0],   %[pixels],      %[line_size]            \n\t"
+        "packushb   %[ftmp0],   %[ftmp0],       %[ftmp1]                \n\t"
+        "packushb   %[ftmp2],   %[ftmp2],       %[ftmp3]                \n\t"
+        "packushb   %[ftmp4],   %[ftmp4],       %[ftmp5]                \n\t"
+        "packushb   %[ftmp6],   %[ftmp6],       %[ftmp7]                \n\t"
+        "sdc1       %[ftmp0],   0x00(%[pixels])                         \n\t"
+        "sdc1       %[ftmp2],   0x00(%[addr0])                          \n\t"
+        "gssdxc1    %[ftmp4],   0x00(%[addr0],  %[line_size])           \n\t"
+        "gssdxc1    %[ftmp6],   0x00(%[pixels], %[line_sizex3])         \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]),
+          [addr0]"=&r"(addr[0]),
+          [pixels]"+&r"(pixels)
+        : [line_size]"r"((mips_reg)line_size),
+          [line_sizex3]"r"((mips_reg)(line_size*3)),
+          [block]"r"(block)
+        : "memory"
     );
 }
 
@@ -87,104 +99,110 @@ void ff_put_signed_pixels_clamped_mmi(const int16_t *block,
     uint8_t *av_restrict pixels, ptrdiff_t line_size)
 {
     int64_t line_skip = line_size;
-    int64_t line_skip3;
+    int64_t line_skip3 = 0;
+    double ftmp[5];
+    mips_reg addr[1];
 
     __asm__ volatile (
-        "dmtc1 %4, $f0                  \n\t"
-        "daddu %1, %3, %3               \n\t"
-        "ldc1 $f2, 0(%2)                \n\t"
-        "ldc1 $f10, 8(%2)               \n\t"
-        "packsshb $f2, $f2, $f10        \n\t"
-        "ldc1 $f4, 16(%2)               \n\t"
-        "ldc1 $f10, 24(%2)              \n\t"
-        "packsshb $f4, $f4, $f10        \n\t"
-        "ldc1 $f6, 32(%2)               \n\t"
-        "ldc1 $f10, 40(%2)              \n\t"
-        "packsshb $f6, $f6, $f10        \n\t"
-        "ldc1 $f8, 48(%2)               \n\t"
-        "ldc1 $f10, 56(%2)              \n\t"
-        "packsshb $f8, $f8, $f10        \n\t"
-        "paddb $f2, $f2, $f0            \n\t"
-        "paddb $f4, $f4, $f0            \n\t"
-        "paddb $f6, $f6, $f0            \n\t"
-        "paddb $f8, $f8, $f0            \n\t"
-        "sdc1 $f2, 0(%0)                \n\t"
-        "gssdxc1 $f4, 0(%0, %3)         \n\t"
-        "gssdxc1 $f6, 0(%0, %1)         \n\t"
-        "daddu %1, %1, %3               \n\t"
-        "gssdxc1 $f8, 0(%0, %1)         \n\t"
-        "daddu $10, %1, %3              \n\t"
-        "daddu %0, %0, $10              \n\t"
-        "ldc1 $f2, 64(%2)               \n\t"
-        "ldc1 $f10, 8+64(%2)            \n\t"
-        "packsshb  $f2, $f2, $f10       \n\t"
-        "ldc1 $f4, 16+64(%2)            \n\t"
-        "ldc1 $f10, 24+64(%2)           \n\t"
-        "packsshb $f4, $f4, $f10        \n\t"
-        "ldc1 $f6, 32+64(%2)            \n\t"
-        "ldc1 $f10, 40+64(%2)           \n\t"
-        "packsshb $f6, $f6, $f10        \n\t"
-        "ldc1 $f8, 48+64(%2)            \n\t"
-        "ldc1 $f10, 56+64(%2)           \n\t"
-        "packsshb $f8, $f8, $f10        \n\t"
-        "paddb $f2, $f2, $f0            \n\t"
-        "paddb $f4, $f4, $f0            \n\t"
-        "paddb $f6, $f6, $f0            \n\t"
-        "paddb $f8, $f8, $f0            \n\t"
-        "sdc1 $f2, 0(%0)                \n\t"
-        "gssdxc1 $f4, 0(%0, %3)         \n\t"
-        "daddu $10, %3, %3              \n\t"
-        "gssdxc1 $f6, 0(%0, $10)        \n\t"
-        "gssdxc1 $f8, 0(%0, %1)         \n\t"
-        : "+&r"(pixels),"=&r"(line_skip3)
-        : "r"(block),"r"(line_skip),"r"(ff_pb_80)
-        : "$10","memory"
+        PTR_ADDU   "%[line_skip3],  %[line_skip],   %[line_skip]        \n\t"
+        "ldc1       %[ftmp1],       0x00(%[block])                      \n\t"
+        "ldc1       %[ftmp0],       0x08(%[block])                      \n\t"
+        "packsshb   %[ftmp1],       %[ftmp1],       %[ftmp0]            \n\t"
+        "ldc1       %[ftmp2],       0x10(%[block])                      \n\t"
+        "ldc1       %[ftmp0],       0x18(%[block])                      \n\t"
+        "packsshb   %[ftmp2],       %[ftmp2],       %[ftmp0]            \n\t"
+        "ldc1       %[ftmp3],       0x20(%[block])                      \n\t"
+        "ldc1       %[ftmp0],       0x28(%[block])                      \n\t"
+        "packsshb   %[ftmp3],       %[ftmp3],       %[ftmp0]            \n\t"
+        "ldc1       %[ftmp4],       48(%[block])                        \n\t"
+        "ldc1       %[ftmp0],       56(%[block])                        \n\t"
+        "packsshb   %[ftmp4],       %[ftmp4],       %[ftmp0]            \n\t"
+        "paddb      %[ftmp1],       %[ftmp1],       %[ff_pb_80]         \n\t"
+        "paddb      %[ftmp2],       %[ftmp2],       %[ff_pb_80]         \n\t"
+        "paddb      %[ftmp3],       %[ftmp3],       %[ff_pb_80]         \n\t"
+        "paddb      %[ftmp4],       %[ftmp4],       %[ff_pb_80]         \n\t"
+        "sdc1       %[ftmp1],       0x00(%[pixels])                     \n\t"
+        "gssdxc1    %[ftmp2],       0x00(%[pixels], %[line_skip])       \n\t"
+        "gssdxc1    %[ftmp3],       0x00(%[pixels], %[line_skip3])      \n\t"
+        PTR_ADDU   "%[line_skip3],  %[line_skip3],  %[line_skip]        \n\t"
+        "gssdxc1    %[ftmp4],       0x00(%[pixels], %[line_skip3])      \n\t"
+        PTR_ADDU   "%[addr0],       %[line_skip3],  %[line_skip]        \n\t"
+        PTR_ADDU   "%[pixels],      %[pixels],      %[addr0]            \n\t"
+        "ldc1       %[ftmp1],       0x40(%[block])                      \n\t"
+        "ldc1       %[ftmp0],       0x48(%[block])                      \n\t"
+        "packsshb   %[ftmp1],       %[ftmp1],       %[ftmp0]            \n\t"
+        "ldc1       %[ftmp2],       0x50(%[block])                      \n\t"
+        "ldc1       %[ftmp0],       0x58(%[block])                      \n\t"
+        "packsshb   %[ftmp2],       %[ftmp2],       %[ftmp0]            \n\t"
+        "ldc1       %[ftmp3],       0x60(%[block])                      \n\t"
+        "ldc1       %[ftmp0],       0x68(%[block])                      \n\t"
+        "packsshb   %[ftmp3],       %[ftmp3],       %[ftmp0]            \n\t"
+        "ldc1       %[ftmp4],       0x70(%[block])                      \n\t"
+        "ldc1       %[ftmp0],       0x78(%[block])                      \n\t"
+        "packsshb   %[ftmp4],       %[ftmp4],       %[ftmp0]            \n\t"
+        "paddb      %[ftmp1],       %[ftmp1],       %[ff_pb_80]         \n\t"
+        "paddb      %[ftmp2],       %[ftmp2],       %[ff_pb_80]         \n\t"
+        "paddb      %[ftmp3],       %[ftmp3],       %[ff_pb_80]         \n\t"
+        "paddb      %[ftmp4],       %[ftmp4],       %[ff_pb_80]         \n\t"
+        "sdc1       %[ftmp1],       0x00(%[pixels])                     \n\t"
+        "gssdxc1    %[ftmp2],       0x00(%[pixels], %[line_skip])       \n\t"
+        PTR_ADDU   "%[addr0],       %[line_skip],   %[line_skip]        \n\t"
+        "gssdxc1    %[ftmp3],       0x00(%[pixels], %[addr0])           \n\t"
+        "gssdxc1    %[ftmp4],       0x00(%[pixels], %[line_skip3])      \n\t"
+        : [ftmp0]"=&f"(ftmp[0]),            [ftmp1]"=&f"(ftmp[1]),
+          [ftmp2]"=&f"(ftmp[2]),            [ftmp3]"=&f"(ftmp[3]),
+          [ftmp4]"=&f"(ftmp[4]),
+          [addr0]"=&r"(addr[0]),
+          [pixels]"+&r"(pixels),            [line_skip3]"+&r"(line_skip3)
+        : [block]"r"(block),
+          [line_skip]"r"((mips_reg)line_skip),
+          [ff_pb_80]"f"(ff_pb_80)
+        : "memory"
     );
 }
 
 void ff_add_pixels_clamped_mmi(const int16_t *block,
         uint8_t *av_restrict pixels, ptrdiff_t line_size)
 {
-    const int16_t *p;
-    uint8_t *pix;
-    int i = 4;
-
-    p = block;
-    pix = pixels;
+    double ftmp[8];
+    uint64_t tmp[1];
 
     __asm__ volatile (
-        "xor $f14, $f14, $f14           \r\n"
-        ::
+        "li         %[tmp0],    0x04                                    \n\t"
+        "xor        %[ftmp0],   %[ftmp0],       %[ftmp0]                \n\t"
+        "1:                                                             \n\t"
+        "ldc1       %[ftmp1],   0x00(%[block])                          \n\t"
+        "ldc1       %[ftmp2],   0x08(%[block])                          \n\t"
+        "ldc1       %[ftmp3],   0x10(%[block])                          \n\t"
+        "ldc1       %[ftmp4],   0x18(%[block])                          \n\t"
+        "ldc1       %[ftmp5],   0x00(%[pixels])                         \n\t"
+        "gsldxc1    %[ftmp6],   0x00(%[pixels], %[line_size])           \n\t"
+        "mov.d      %[ftmp7],   %[ftmp5]                                \n\t"
+        "punpcklbh  %[ftmp5],   %[ftmp5],       %[ftmp0]                \n\t"
+        "punpckhbh  %[ftmp7],   %[ftmp7],       %[ftmp0]                \n\t"
+        "paddh      %[ftmp1],   %[ftmp1],       %[ftmp5]                \n\t"
+        "paddh      %[ftmp2],   %[ftmp2],       %[ftmp7]                \n\t"
+        "mov.d      %[ftmp7],   %[ftmp6]                                \n\t"
+        "punpcklbh  %[ftmp6],   %[ftmp6],       %[ftmp0]                \n\t"
+        "punpckhbh  %[ftmp7],   %[ftmp7],       %[ftmp0]                \n\t"
+        "paddh      %[ftmp3],   %[ftmp3],       %[ftmp6]                \n\t"
+        "paddh      %[ftmp4],   %[ftmp4],       %[ftmp7]                \n\t"
+        "packushb   %[ftmp1],   %[ftmp1],       %[ftmp2]                \n\t"
+        "packushb   %[ftmp3],   %[ftmp3],       %[ftmp4]                \n\t"
+        "sdc1       %[ftmp1],   0x00(%[pixels])                         \n\t"
+        "gssdxc1    %[ftmp3],   0x00(%[pixels], %[line_size])           \n\t"
+        "addi       %[tmp0],    %[tmp0],        -0x01                   \n\t"
+        PTR_ADDIU  "%[block],   %[block],       0x20                    \n\t"
+        PTR_ADDU   "%[pixels],  %[pixels],      %[line_size]            \n\t"
+        PTR_ADDU   "%[pixels],  %[pixels],      %[line_size]            \n\t"
+        "bnez       %[tmp0],    1b"
+        : [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]),
+          [tmp0]"=&r"(tmp[0]),
+          [pixels]"+&r"(pixels),            [block]"+&r"(block)
+        : [line_size]"r"((mips_reg)line_size)
+        : "memory"
     );
-
-    do {
-        __asm__ volatile (
-            "ldc1 $f0, 0+%2             \r\n"
-            "ldc1 $f2, 8+%2             \r\n"
-            "ldc1 $f4, 16+%2            \r\n"
-            "ldc1 $f6, 24+%2            \r\n"
-            "ldc1 $f8, %0               \r\n"
-            "ldc1 $f12, %1              \r\n"
-            "mov.d $f10, $f8            \r\n"
-            "punpcklbh $f8, $f8, $f14   \r\n"
-            "punpckhbh $f10, $f10, $f14 \r\n"
-            "paddsh $f0, $f0, $f8       \r\n"
-            "paddsh $f2, $f2, $f10      \r\n"
-            "mov.d $f10, $f12           \r\n"
-            "punpcklbh $f12, $f12, $f14 \r\n"
-            "punpckhbh $f10, $f10, $f14 \r\n"
-            "paddsh $f4, $f4, $f12      \r\n"
-            "paddsh $f6, $f6, $f10      \r\n"
-            "packushb $f0, $f0, $f2     \r\n"
-            "packushb $f4, $f4, $f6     \r\n"
-            "sdc1 $f0, %0               \r\n"
-            "sdc1 $f4, %1               \r\n"
-            : "+m"(*pix),"+m"(*(pix+line_size))
-            : "m"(*p)
-            : "memory"
-        );
-
-        pix += line_size*2;
-        p += 16;
-    } while (--i);
 }