problems with gcc inline assembly using xmm registers

[David Palao <david.palao@xxxxx>]

Hi!

I'm newbie using assembly, but I'm working in computational physics and we 
need to design a very fast function doing an operation which is repeated 
billions of times in our calculations.
In order to gain performance we decided to use some of the SSE features 
(properly used, xmm registers should provide a very nice increase in 
performance).
However, I'm having serious troubles using xmm[0...7] registers with gcc 
inlines.
One example of the code in which I got these troubles is at the end of the 
message.

The error I get is:

../Libraries/fermiqcd_fermi_actions_sse2.h:818: can't find a register in class
   `GENERAL_REGS' while reloading `asm'


what I understand is that the function I'm trying to compile uses too many 
`GENERAL_REGS' registers but...
...do xmm registers belong to this group? aren't they a special set of 
registers in the SSE area?
...Could it be that the inlines use too many intermediate registers from the 
general registers group? In that case, how could I avoid such behaviour?

The problem occurs whenever I try to use the xmm intensively; if I don't use 
SSE extensions eveything goes ok, but I NEED xmm registers!

Thanks in advance!!!
Best regards


David


PS gcc -v:
gcc version 3.2.3 20030502 (Red Hat Linux 3.2.3-34)




CODE:

__asm__ __volatile__ ("movsd %0, %%xmm3 \n\t" \
      "movsd %1, %%xmm6 \n\t" \
      "movsd %2, %%xmm4 \n\t" \
      "movsd %3, %%xmm7 \n\t" \
      "movsd %4, %%xmm5 \n\t" \
      "unpcklpd %%xmm3, %%xmm3 \n\t" \
      "unpcklpd %%xmm6, %%xmm6 \n\t" \
      "unpcklpd %%xmm4, %%xmm4 \n\t" \
      "mulpd %%xmm0, %%xmm3 \n\t" \
      "unpcklpd %%xmm7, %%xmm7 \n\t" \
      "mulpd %%xmm1, %%xmm6 \n\t" \
      "unpcklpd %%xmm5, %%xmm5 \n\t" \
      "mulpd %%xmm0, %%xmm4 \n\t" \
      "addpd %%xmm6, %%xmm3 \n\t" \
      "mulpd %%xmm2, %%xmm7 \n\t" \
      "mulpd %%xmm0, %%xmm5 \n\t" \
      "addpd %%xmm7, %%xmm4 \n\t" \
      "movsd %5, %%xmm6 \n\t" \
      "movsd %6, %%xmm7 \n\t" \
      "unpcklpd %%xmm6, %%xmm6 \n\t" \
      "unpcklpd %%xmm7, %%xmm7 \n\t" \
      "mulpd %%xmm1, %%xmm6 \n\t" \
      "mulpd %%xmm2, %%xmm7 \n\t" \
      "addpd %%xmm6, %%xmm5 \n\t" \
      "addpd %%xmm7, %%xmm3 \n\t" \
      "movsd %7, %%xmm6 \n\t" \
      "movsd %8, %%xmm7 \n\t" \
      "unpcklpd %%xmm6, %%xmm6 \n\t" \
      "unpcklpd %%xmm7, %%xmm7 \n\t" \
      "mulpd %%xmm1, %%xmm6 \n\t" \
      "mulpd %%xmm2, %%xmm7 \n\t" \
      "addpd %%xmm6, %%xmm4 \n\t" \
      "addpd %%xmm7, %%xmm5" \
      : \
      : \
      "m" ((u).c11.real()), \
      "m" ((u).c12.real()), \
      "m" ((u).c21.real()), \
      "m" ((u).c23.real()), \
      "m" ((u).c31.real()), \
      "m" ((u).c32.real()), \
      "m" ((u).c13.real()), \
      "m" ((u).c22.real()), \
      "m" ((u).c33.real())); \

__asm__ __volatile__ ("movsd %0, %%xmm6 \n\t" \
      "movsd %1, %%xmm7 \n\t" \
      "shufpd $0x1, %%xmm0, %%xmm0 \n\t" \
      "shufpd $0x1, %%xmm1, %%xmm1 \n\t" \
      "shufpd $0x1, %%xmm2, %%xmm2 \n\t" \
      "unpcklpd %%xmm6, %%xmm6 \n\t" \
      "unpcklpd %%xmm7, %%xmm7 \n\t" \
      "xorpd %9, %%xmm0 \n\t" \
      "xorpd %9, %%xmm1 \n\t" \
      "xorpd %9, %%xmm2 \n\t" \
      "mulpd %%xmm0, %%xmm6 \n\t" \
      "mulpd %%xmm1, %%xmm7 \n\t" \
      "addpd %%xmm6, %%xmm3 \n\t" \
      "addpd %%xmm7, %%xmm4 \n\t" \
      "movsd %2, %%xmm6 \n\t" \
      "movsd %3, %%xmm7 \n\t" \
      "unpcklpd %%xmm6, %%xmm6 \n\t" \
      "unpcklpd %%xmm7, %%xmm7 \n\t" \
      "mulpd %%xmm2, %%xmm6 \n\t" \
      "mulpd %%xmm0, %%xmm7 \n\t" \
      "addpd %%xmm6, %%xmm5 \n\t" \
      "addpd %%xmm7, %%xmm4 \n\t" \
      "movsd %4, %%xmm6 \n\t" \
      "movsd %5, %%xmm7 \n\t" \
      "unpcklpd %%xmm6, %%xmm6 \n\t" \
      "unpcklpd %%xmm7, %%xmm7 \n\t" \
      "mulpd %%xmm1, %%xmm6 \n\t" \
      "mulpd %%xmm0, %%xmm7 \n\t" \
      "addpd %%xmm6, %%xmm3 \n\t" \
      "addpd %%xmm7, %%xmm5 \n\t" \
      "movsd %6, %%xmm0 \n\t" \
      "movsd %7, %%xmm6 \n\t" \
      "movsd %8, %%xmm7 \n\t" \
      "unpcklpd %%xmm0, %%xmm0 \n\t" \
      "unpcklpd %%xmm6, %%xmm6 \n\t" \
      "unpcklpd %%xmm7, %%xmm7 \n\t" \
      "mulpd %%xmm2, %%xmm0 \n\t" \
      "mulpd %%xmm1, %%xmm6 \n\t" \
      "mulpd %%xmm2, %%xmm7 \n\t" \
      "addpd %%xmm0, %%xmm3 \n\t" \
      "addpd %%xmm6, %%xmm5 \n\t" \
      "addpd %%xmm7, %%xmm4" \
      : \
      : \
      "m" ((u).c11.imag()), \
      "m" ((u).c22.imag()), \
      "m" ((u).c33.imag()), \
      "m" ((u).c21.imag()), \
      "m" ((u).c12.imag()), \
      "m" ((u).c31.imag()), \
      "m" ((u).c13.imag()), \
      "m" ((u).c32.imag()), \
      "m" ((u).c23.imag()), \
      "m" (_sse_double_sgn));