- "dli $2, 8 \n\t"
- "xor $f14, $f14, $f14 \n\t"
- "1: \n\t"
- "gsldlc1 $f0, 7(%1) \n\t"
- "gsldrc1 $f0, 0(%1) \n\t"
- "or $f2, $f0, $f0 \n\t"
- "gsldlc1 $f4, 7(%2) \n\t"
- "gsldrc1 $f4, 0(%2) \n\t"
- "or $f6, $f4, $f4 \n\t"
- "punpcklbh $f0, $f0, $f14 \n\t"
- "punpckhbh $f2, $f2, $f14 \n\t"
- "punpcklbh $f4, $f4, $f14 \n\t"
- "punpckhbh $f6, $f6, $f14 \n\t"
- "psubh $f0, $f0, $f4 \n\t"
- "psubh $f2, $f2, $f6 \n\t"
- "gssdlc1 $f0, 7(%0) \n\t"
- "gssdrc1 $f0, 0(%0) \n\t"
- "gssdlc1 $f2, 15(%0) \n\t"
- "gssdrc1 $f2, 8(%0) \n\t"
- "daddi %0, %0, 16 \n\t"
- "daddu %1, %1, %3 \n\t"
- "daddu %2, %2, %3 \n\t"
- "daddi $2, $2, -1 \n\t"
- "bgtz $2, 1b \n\t"
- ::"r"(block),"r"(src1),"r"(src2),"r"(stride)
- : "$2","memory"
+ "li %[tmp0], 0x08 \n\t"
+ "xor %[ftmp4], %[ftmp4], %[ftmp4] \n\t"
+ "1: \n\t"
+ "gsldlc1 %[ftmp0], 0x07(%[src1]) \n\t"
+ "gsldrc1 %[ftmp0], 0x00(%[src1]) \n\t"
+ "or %[ftmp1], %[ftmp0], %[ftmp0] \n\t"
+ "gsldlc1 %[ftmp2], 0x07(%[src2]) \n\t"
+ "gsldrc1 %[ftmp2], 0x00(%[src2]) \n\t"
+ "or %[ftmp3], %[ftmp2], %[ftmp2] \n\t"
+ "punpcklbh %[ftmp0], %[ftmp0], %[ftmp4] \n\t"
+ "punpckhbh %[ftmp1], %[ftmp1], %[ftmp4] \n\t"
+ "punpcklbh %[ftmp2], %[ftmp2], %[ftmp4] \n\t"
+ "punpckhbh %[ftmp3], %[ftmp3], %[ftmp4] \n\t"
+ "psubh %[ftmp0], %[ftmp0], %[ftmp2] \n\t"
+ "psubh %[ftmp1], %[ftmp1], %[ftmp3] \n\t"
+ "gssdlc1 %[ftmp0], 0x07(%[block]) \n\t"
+ "gssdrc1 %[ftmp0], 0x00(%[block]) \n\t"
+ "gssdlc1 %[ftmp1], 0x0f(%[block]) \n\t"
+ "gssdrc1 %[ftmp1], 0x08(%[block]) \n\t"
+ PTR_ADDI "%[tmp0], %[tmp0], -0x01 \n\t"
+ PTR_ADDIU "%[block], %[block], 0x10 \n\t"
+ PTR_ADDU "%[src1], %[src1], %[stride] \n\t"
+ PTR_ADDU "%[src2], %[src2], %[stride] \n\t"
+ "bgtz %[tmp0], 1b \n\t"
+ : [ftmp0]"=&f"(ftmp[0]), [ftmp1]"=&f"(ftmp[1]),
+ [ftmp2]"=&f"(ftmp[2]), [ftmp3]"=&f"(ftmp[3]),
+ [ftmp4]"=&f"(ftmp[4]),
+ [tmp0]"=&r"(tmp[0]),
+ [block]"+&r"(block), [src1]"+&r"(src1),
+ [src2]"+&r"(src2)
+ : [stride]"r"((mips_reg)stride)
+ : "memory"