#if HAVE_6REGS
static void vector_fmul_window_3dnow2(float *dst, const float *src0, const float *src1,
const float *win, int len){
- x86_reg i = -len*4;
- x86_reg j = len*4-8;
- __asm__ volatile(
- "1: \n"
- "pswapd (%5,%1), %%mm1 \n"
- "movq (%5,%0), %%mm0 \n"
- "pswapd (%4,%1), %%mm5 \n"
- "movq (%3,%0), %%mm4 \n"
- "movq %%mm0, %%mm2 \n"
- "movq %%mm1, %%mm3 \n"
- "pfmul %%mm4, %%mm2 \n" // src0[len+i]*win[len+i]
- "pfmul %%mm5, %%mm3 \n" // src1[ j]*win[len+j]
- "pfmul %%mm4, %%mm1 \n" // src0[len+i]*win[len+j]
- "pfmul %%mm5, %%mm0 \n" // src1[ j]*win[len+i]
- "pfadd %%mm3, %%mm2 \n"
- "pfsub %%mm0, %%mm1 \n"
- "pswapd %%mm2, %%mm2 \n"
- "movq %%mm1, (%2,%0) \n"
- "movq %%mm2, (%2,%1) \n"
- "sub $8, %1 \n"
- "add $8, %0 \n"
- "jl 1b \n"
- "femms \n"
- :"+r"(i), "+r"(j)
- :"r"(dst+len), "r"(src0+len), "r"(src1), "r"(win+len)
- );
+ x86_reg i = -len*4;
+ x86_reg j = len*4-8;
+ __asm__ volatile(
+ "1: \n"
+ "pswapd (%5,%1), %%mm1 \n"
+ "movq (%5,%0), %%mm0 \n"
+ "pswapd (%4,%1), %%mm5 \n"
+ "movq (%3,%0), %%mm4 \n"
+ "movq %%mm0, %%mm2 \n"
+ "movq %%mm1, %%mm3 \n"
+ "pfmul %%mm4, %%mm2 \n" // src0[len+i]*win[len+i]
+ "pfmul %%mm5, %%mm3 \n" // src1[ j]*win[len+j]
+ "pfmul %%mm4, %%mm1 \n" // src0[len+i]*win[len+j]
+ "pfmul %%mm5, %%mm0 \n" // src1[ j]*win[len+i]
+ "pfadd %%mm3, %%mm2 \n"
+ "pfsub %%mm0, %%mm1 \n"
+ "pswapd %%mm2, %%mm2 \n"
+ "movq %%mm1, (%2,%0) \n"
+ "movq %%mm2, (%2,%1) \n"
+ "sub $8, %1 \n"
+ "add $8, %0 \n"
+ "jl 1b \n"
+ "femms \n"
+ :"+r"(i), "+r"(j)
+ :"r"(dst+len), "r"(src0+len), "r"(src1), "r"(win+len)
+ );
}
static void vector_fmul_window_sse(float *dst, const float *src0, const float *src1,
const float *win, int len){
- x86_reg i = -len*4;
- x86_reg j = len*4-16;
- __asm__ volatile(
- "1: \n"
- "movaps (%5,%1), %%xmm1 \n"
- "movaps (%5,%0), %%xmm0 \n"
- "movaps (%4,%1), %%xmm5 \n"
- "movaps (%3,%0), %%xmm4 \n"
- "shufps $0x1b, %%xmm1, %%xmm1 \n"
- "shufps $0x1b, %%xmm5, %%xmm5 \n"
- "movaps %%xmm0, %%xmm2 \n"
- "movaps %%xmm1, %%xmm3 \n"
- "mulps %%xmm4, %%xmm2 \n" // src0[len+i]*win[len+i]
- "mulps %%xmm5, %%xmm3 \n" // src1[ j]*win[len+j]
- "mulps %%xmm4, %%xmm1 \n" // src0[len+i]*win[len+j]
- "mulps %%xmm5, %%xmm0 \n" // src1[ j]*win[len+i]
- "addps %%xmm3, %%xmm2 \n"
- "subps %%xmm0, %%xmm1 \n"
- "shufps $0x1b, %%xmm2, %%xmm2 \n"
- "movaps %%xmm1, (%2,%0) \n"
- "movaps %%xmm2, (%2,%1) \n"
- "sub $16, %1 \n"
- "add $16, %0 \n"
- "jl 1b \n"
- :"+r"(i), "+r"(j)
- :"r"(dst+len), "r"(src0+len), "r"(src1), "r"(win+len)
- );
+ x86_reg i = -len*4;
+ x86_reg j = len*4-16;
+ __asm__ volatile(
+ "1: \n"
+ "movaps (%5,%1), %%xmm1 \n"
+ "movaps (%5,%0), %%xmm0 \n"
+ "movaps (%4,%1), %%xmm5 \n"
+ "movaps (%3,%0), %%xmm4 \n"
+ "shufps $0x1b, %%xmm1, %%xmm1 \n"
+ "shufps $0x1b, %%xmm5, %%xmm5 \n"
+ "movaps %%xmm0, %%xmm2 \n"
+ "movaps %%xmm1, %%xmm3 \n"
+ "mulps %%xmm4, %%xmm2 \n" // src0[len+i]*win[len+i]
+ "mulps %%xmm5, %%xmm3 \n" // src1[ j]*win[len+j]
+ "mulps %%xmm4, %%xmm1 \n" // src0[len+i]*win[len+j]
+ "mulps %%xmm5, %%xmm0 \n" // src1[ j]*win[len+i]
+ "addps %%xmm3, %%xmm2 \n"
+ "subps %%xmm0, %%xmm1 \n"
+ "shufps $0x1b, %%xmm2, %%xmm2 \n"
+ "movaps %%xmm1, (%2,%0) \n"
+ "movaps %%xmm2, (%2,%1) \n"
+ "sub $16, %1 \n"
+ "add $16, %0 \n"
+ "jl 1b \n"
+ :"+r"(i), "+r"(j)
+ :"r"(dst+len), "r"(src0+len), "r"(src1), "r"(win+len)
+ );
}
#endif /* HAVE_6REGS */