Hi, guys

we aim to self make a FPU which has its own instruction set. For easy 
programming, we prefer c language to assembly code.

here's one of our instruction:
c language format: Bload(vd, value1, value2), we make it like a 
funciton. vd is a special register.
binary format: 31-26 opcode, 25-21 index of vd register, 20-16 opcode, 
15-11 index of the general register which keeps value1, 10-5 opcode, 4-0 
index of the general register which keeps value2.

we use gcc-inline-assembly to convert it to binary code. here's the 
solution:
#define Bload(vd, value1, value2)    \
  __asm__(        \
    "add  r16, r0, %0  \n\t"  \
    "add  r14, r0, %1  \n\t"  \
    ".long  %2  \n\t"    \
    :        \
    : "r"(value2), "r"(value1), 
"i"((0x3b<<26)+(vd<<21)+(14<<11)+(0x2c<<5)+16)  \
    : "r14", "r16");
i fix the general registers(r14, r16) which keep value1 and value2. then 
i make the binary code use "i" in inline-assembly. But disadvantage is 
in evidence: gcc does not understand my instruction, so i can not take 
it's advantage, such as selecting the most suitable general register and 
eliminating unnecessary copy from register to register.

what i want is deleting the first two add instruction, let gcc select 
register for me and tell me which register it selected. so i can make my 
own binary code using .long instruction. But i do not want to modify the 
gcc compiler to let it understand my instruction.

Thank you for your help