*/
asm (
/* clear */
- "pxor %%xmm0, %%xmm0 \n"
- "pxor %%xmm1, %%xmm1 \n"
- "pxor %%xmm2, %%xmm2 \n"
- "pxor %%xmm3, %%xmm3 \n"
+ "pxor %%xmm0, %%xmm0 \n"
+ "pxor %%xmm1, %%xmm1 \n"
+ "pxor %%xmm2, %%xmm2 \n"
+ "pxor %%xmm3, %%xmm3 \n"
/* main loop */
- ".lbl: \n"
+ ".lbl: \n"
/* a zero is useful during unpacking */
- "pxor %%xmm4, %%xmm4 \n"
+ "pxor %%xmm4, %%xmm4 \n"
/* fetch all 16 source bytes */
- "movups (%0), %%xmm5 \n"
- "prefetcht0 (%0,%3,4) \n"
+ "movups (%0), %%xmm5 \n"
+ "prefetcht0 (%0,%3,4) \n"
/* unpack into words (xmm5, xmm7) */
- "movaps %%xmm5, %%xmm7 \n"
- "punpcklbw %%xmm4, %%xmm5 \n"
- "punpckhbw %%xmm4, %%xmm7 \n"
+ "movaps %%xmm5, %%xmm7 \n"
+ "punpcklbw %%xmm4, %%xmm5 \n"
+ "punpckhbw %%xmm4, %%xmm7 \n"
/* unpack xmm5 into dwords (xmm5, xmm6) */
- "movaps %%xmm5, %%xmm6 \n"
- "punpcklwd %%xmm4, %%xmm5 \n"
- "punpckhwd %%xmm4, %%xmm6 \n"
+ "movaps %%xmm5, %%xmm6 \n"
+ "punpcklwd %%xmm4, %%xmm5 \n"
+ "punpckhwd %%xmm4, %%xmm6 \n"
/* convert xmm5, xmm6 to floats */
- "cvtdq2ps %%xmm5, %%xmm5 \n"
- "cvtdq2ps %%xmm6, %%xmm6 \n"
+ "cvtdq2ps %%xmm5, %%xmm5 \n"
+ "cvtdq2ps %%xmm6, %%xmm6 \n"
/* fetch the coefficient */
- "movss (%2), %%xmm4 \n"
- "shufps $0x0, %%xmm4, %%xmm4 \n"
+ "movss (%2), %%xmm4 \n"
+ "shufps $0x0, %%xmm4, %%xmm4 \n"
/* do the muls for xmm5 and xmm6 */
- "mulps %%xmm4, %%xmm5 \n"
- "mulps %%xmm4, %%xmm6 \n"
- "addps %%xmm5, %%xmm0 \n"
- "addps %%xmm6, %%xmm1 \n"
+ "mulps %%xmm4, %%xmm5 \n"
+ "mulps %%xmm4, %%xmm6 \n"
+ "addps %%xmm5, %%xmm0 \n"
+ "addps %%xmm6, %%xmm1 \n"
/* get the zero back again */
- "pxor %%xmm4, %%xmm4 \n"
+ "pxor %%xmm4, %%xmm4 \n"
/* unpack xmm7 into dwords (xmm7, xmm6) */
- "movaps %%xmm7, %%xmm6 \n"
- "punpcklwd %%xmm4, %%xmm7 \n"
- "punpckhwd %%xmm4, %%xmm6 \n"
+ "movaps %%xmm7, %%xmm6 \n"
+ "punpcklwd %%xmm4, %%xmm7 \n"
+ "punpckhwd %%xmm4, %%xmm6 \n"
/* convert xmm7, xmm6 to floats */
- "cvtdq2ps %%xmm7, %%xmm7 \n"
- "cvtdq2ps %%xmm6, %%xmm6 \n"
+ "cvtdq2ps %%xmm7, %%xmm7 \n"
+ "cvtdq2ps %%xmm6, %%xmm6 \n"
/* fetch the coefficient */
- "movss (%2), %%xmm4 \n"
- "shufps $0x0, %%xmm4, %%xmm4 \n"
+ "movss (%2), %%xmm4 \n"
+ "shufps $0x0, %%xmm4, %%xmm4 \n"
/* do the second set of muls */
- "mulps %%xmm4, %%xmm7 \n"
- "mulps %%xmm4, %%xmm6 \n"
- "addps %%xmm7, %%xmm2 \n"
- "addps %%xmm6, %%xmm3 \n"
+ "mulps %%xmm4, %%xmm7 \n"
+ "mulps %%xmm4, %%xmm6 \n"
+ "addps %%xmm7, %%xmm2 \n"
+ "addps %%xmm6, %%xmm3 \n"
/* move along, and loop */
- "add $4, %2 \n"
- "add %3, %0 \n"
- "dec %1 \n"
- "jnz .lbl \n"
+ "add $4, %2 \n"
+ "add %3, %0 \n"
+ "dec %1 \n"
+ "jnz .lbl \n"
/* store the values */
- "movaps %%xmm0, (%4) \n"
- "movaps %%xmm1, 16(%4) \n"
- "movaps %%xmm2, 32(%4) \n"
- "movaps %%xmm3, 48(%4) \n"
+ "movaps %%xmm0, (%4) \n"
+ "movaps %%xmm1, 16(%4) \n"
+ "movaps %%xmm2, 32(%4) \n"
+ "movaps %%xmm3, 48(%4) \n"
: :
"r" (&sptr[pd[y].start * w]), /* 0: srcptr base */
"r" (pd[y].end - pd[y].start + 1), /* 1: filter len */