[PD-cvs] pd/src m_simd_sse_gcc.c,1.1.4.11,1.1.4.12

Update of /cvsroot/pure-data/pd/src In directory sc8-pr-cvs1.sourceforge.net:/tmp/cvs-serv4633

Modified Files: Tag: devel_0_38 m_simd_sse_gcc.c Log Message: changing asm calls (should compile without -msse flag) & removing sumvec stuff

Index: m_simd_sse_gcc.c =================================================================== RCS file: /cvsroot/pure-data/pd/src/Attic/m_simd_sse_gcc.c,v retrieving revision 1.1.4.11 retrieving revision 1.1.4.12 diff -C2 -d -r1.1.4.11 -r1.1.4.12 *** m_simd_sse_gcc.c 18 Jan 2005 15:31:04 -0000 1.1.4.11 --- m_simd_sse_gcc.c 23 Jan 2005 21:13:11 -0000 1.1.4.12 *************** *** 40,44 **** asm( ".set T_FLOAT,4 \n" /* sizeof(t_float) */ ! "shufps $0,%2,%2 \n" /* load value */ "shr $4,%0 \n" --- 40,45 ---- asm( ".set T_FLOAT,4 \n" /* sizeof(t_float) */ ! "movss (%2),%%xmm0 \n" ! "shufps $0,%%xmm0,%%xmm0 \n" /* load value */ "shr $4,%0 \n" *************** *** 46,59 **** /* *dst = v */ "1: \n" ! "movaps %2, (%1) \n" ! "movaps %2, 4*T_FLOAT(%1) \n" ! "movaps %2, 8*T_FLOAT(%1) \n" ! "movaps %2, 12*T_FLOAT(%1) \n" "addl $16*T_FLOAT,%1 \n" "loop 1b \n" : ! :"c"(n),"r"(dst),"x"((t_float)v) ! ); }

--- 47,60 ---- /* *dst = v */ "1: \n" ! "movaps %%xmm0, (%1) \n" ! "movaps %%xmm0, 4*T_FLOAT(%1) \n" ! "movaps %%xmm0, 8*T_FLOAT(%1) \n" ! "movaps %%xmm0, 12*T_FLOAT(%1) \n" "addl $16*T_FLOAT,%1 \n" "loop 1b \n" : ! :"c"(n),"r"(dst),"r"(&v) ! :"%xmm0"); }

*************** *** 366,370 **** ".set T_FLOAT,4 \n" ! "shufps $0, %1, %1 \n" "shrl $4, %3 \n" /* divide by 16 */

--- 367,372 ---- ".set T_FLOAT,4 \n" ! "movss (%1), %%xmm0 \n" ! "shufps $0, %%xmm0, %%xmm0 \n" "shrl $4, %3 \n" /* divide by 16 */

*************** *** 372,388 **** "1: \n" "movaps (%0), %%xmm1 \n" ! "addps %1, %%xmm1 \n" "movaps %%xmm1, (%2) \n"

"movaps 4*T_FLOAT(%0), %%xmm2 \n" ! "addps %1, %%xmm2 \n" "movaps %%xmm2, 4*T_FLOAT(%2) \n" "movaps 8*T_FLOAT(%0), %%xmm3 \n" ! "addps %1, %%xmm3 \n" "movaps %%xmm3, 8*T_FLOAT(%2) \n" "movaps 12*T_FLOAT(%0), %%xmm4 \n" ! "addps %1, %%xmm4 \n" "movaps %%xmm4, 12*T_FLOAT(%2) \n" --- 374,390 ---- "1: \n" "movaps (%0), %%xmm1 \n" ! "addps %%xmm0, %%xmm1 \n" "movaps %%xmm1, (%2) \n"

"movaps 4*T_FLOAT(%0), %%xmm2 \n" ! "addps %%xmm0, %%xmm2 \n" "movaps %%xmm2, 4*T_FLOAT(%2) \n" "movaps 8*T_FLOAT(%0), %%xmm3 \n" ! "addps %%xmm0, %%xmm3 \n" "movaps %%xmm3, 8*T_FLOAT(%2) \n" "movaps 12*T_FLOAT(%0), %%xmm4 \n" ! "addps %%xmm0, %%xmm4 \n" "movaps %%xmm4, 12*T_FLOAT(%2) \n" *************** *** 392,397 **** : /* in, value, out, n */ ! :"r"(w[1]),"x"(*(t_float*)w[2]),"r"(w[3]),"c"(w[4]) ! :"%xmm1","%xmm2","%xmm3","%xmm4" ); return w+5; --- 394,399 ---- : /* in, value, out, n */ ! :"r"(w[1]),"r"(w[2]),"r"(w[3]),"c"(w[4]) ! :"%xmm0", "%xmm1","%xmm2","%xmm3","%xmm4" ); return w+5; *************** *** 442,446 **** ".set T_FLOAT,4 \n" ! "shufps $0, %1, %1 \n" "shrl $4, %3 \n" /* divide by 16 */

--- 444,449 ---- ".set T_FLOAT,4 \n" ! "movss (%1), %%xmm0 \n" ! "shufps $0, %%xmm0, %%xmm0 \n" "shrl $4, %3 \n" /* divide by 16 */

*************** *** 448,464 **** "1: \n" "movaps (%0), %%xmm1 \n" ! "subps %1, %%xmm1 \n" "movaps %%xmm1, (%2) \n"

"movaps 4*T_FLOAT(%0), %%xmm2 \n" ! "subps %1, %%xmm2 \n" "movaps %%xmm2, 4*T_FLOAT(%2) \n" "movaps 8*T_FLOAT(%0), %%xmm3 \n" ! "subps %1, %%xmm3 \n" "movaps %%xmm3, 8*T_FLOAT(%2) \n" "movaps 12*T_FLOAT(%0), %%xmm4 \n" ! "subps %1, %%xmm4 \n" "movaps %%xmm4, 12*T_FLOAT(%2) \n" --- 451,467 ---- "1: \n" "movaps (%0), %%xmm1 \n" ! "subps %%xmm0, %%xmm1 \n" "movaps %%xmm1, (%2) \n"

"movaps 4*T_FLOAT(%0), %%xmm2 \n" ! "subps %%xmm0, %%xmm2 \n" "movaps %%xmm2, 4*T_FLOAT(%2) \n" "movaps 8*T_FLOAT(%0), %%xmm3 \n" ! "subps %%xmm0, %%xmm3 \n" "movaps %%xmm3, 8*T_FLOAT(%2) \n" "movaps 12*T_FLOAT(%0), %%xmm4 \n" ! "subps %%xmm0, %%xmm4 \n" "movaps %%xmm4, 12*T_FLOAT(%2) \n" *************** *** 468,473 **** : /* in, value, out, n */ ! :"r"(w[1]),"x"(*(t_float*)w[2]),"r"(w[3]),"c"(w[4]) ! :"%xmm1","%xmm2","%xmm3","%xmm4" ); return w+5; --- 471,476 ---- : /* in, value, out, n */ ! :"r"(w[1]),"r"(w[2]),"r"(w[3]),"c"(w[4]) ! :"%xmm0","%xmm1","%xmm2","%xmm3","%xmm4" ); return w+5; *************** *** 519,523 **** ".set T_FLOAT,4 \n" ! "shufps $0, %1, %1 \n" "shrl $4, %3 \n" /* divide by 16 */

--- 522,527 ---- ".set T_FLOAT,4 \n" ! "movss (%1), %%xmm0 \n" ! "shufps $0, %%xmm0, %%xmm0 \n" "shrl $4, %3 \n" /* divide by 16 */

*************** *** 525,541 **** "1: \n" "movaps (%0), %%xmm1 \n" ! "mulps %1, %%xmm1 \n" "movaps %%xmm1, (%2) \n"

"movaps 4*T_FLOAT(%0), %%xmm2 \n" ! "mulps %1, %%xmm2 \n" "movaps %%xmm2, 4*T_FLOAT(%2) \n" "movaps 8*T_FLOAT(%0), %%xmm3 \n" ! "mulps %1, %%xmm3 \n" "movaps %%xmm3, 8*T_FLOAT(%2) \n" "movaps 12*T_FLOAT(%0), %%xmm4 \n" ! "mulps %1, %%xmm4 \n" "movaps %%xmm4, 12*T_FLOAT(%2) \n" --- 529,545 ---- "1: \n" "movaps (%0), %%xmm1 \n" ! "mulps %%xmm0, %%xmm1 \n" "movaps %%xmm1, (%2) \n"

"movaps 4*T_FLOAT(%0), %%xmm2 \n" ! "mulps %%xmm0, %%xmm2 \n" "movaps %%xmm2, 4*T_FLOAT(%2) \n" "movaps 8*T_FLOAT(%0), %%xmm3 \n" ! "mulps %%xmm0, %%xmm3 \n" "movaps %%xmm3, 8*T_FLOAT(%2) \n" "movaps 12*T_FLOAT(%0), %%xmm4 \n" ! "mulps %%xmm0, %%xmm4 \n" "movaps %%xmm4, 12*T_FLOAT(%2) \n" *************** *** 545,550 **** : /* in, value, out, n */ ! :"r"(w[1]),"x"(*(t_float*)w[2]),"r"(w[3]),"c"(w[4]) ! :"%xmm1","%xmm2","%xmm3","%xmm4" ); return w+5; --- 549,554 ---- : /* in, value, out, n */ ! :"r"(w[1]),"r"(w[2]),"r"(w[3]),"c"(w[4]) ! :"%xmm0","%xmm1","%xmm2","%xmm3","%xmm4" ); return w+5; *************** *** 632,636 **** ".set T_FLOAT,4 \n" ! "shufps $0, %1, %1 \n" "shrl $4, %3 \n" /* divide by 16 */

--- 636,641 ---- ".set T_FLOAT,4 \n" ! "movss (%1), %%xmm0 \n" ! "shufps $0, %%xmm0, %%xmm0 \n" "shrl $4, %3 \n" /* divide by 16 */

*************** *** 638,654 **** "1: \n" "movaps (%0), %%xmm1 \n" ! "divps %1, %%xmm1 \n" "movaps %%xmm1, (%2) \n"

"movaps 4*T_FLOAT(%0), %%xmm2 \n" ! "divps %1, %%xmm2 \n" "movaps %%xmm2, 4*T_FLOAT(%2) \n" "movaps 8*T_FLOAT(%0), %%xmm3 \n" ! "divps %1, %%xmm3 \n" "movaps %%xmm3, 8*T_FLOAT(%2) \n" "movaps 12*T_FLOAT(%0), %%xmm4 \n" ! "divps %1, %%xmm4 \n" "movaps %%xmm4, 12*T_FLOAT(%2) \n" --- 643,659 ---- "1: \n" "movaps (%0), %%xmm1 \n" ! "divps %%xmm0, %%xmm1 \n" "movaps %%xmm1, (%2) \n"

"movaps 4*T_FLOAT(%0), %%xmm2 \n" ! "divps %%xmm0, %%xmm2 \n" "movaps %%xmm2, 4*T_FLOAT(%2) \n" "movaps 8*T_FLOAT(%0), %%xmm3 \n" ! "divps %%xmm0, %%xmm3 \n" "movaps %%xmm3, 8*T_FLOAT(%2) \n" "movaps 12*T_FLOAT(%0), %%xmm4 \n" ! "divps %%xmm0, %%xmm4 \n" "movaps %%xmm4, 12*T_FLOAT(%2) \n" *************** *** 658,663 **** : /* in, value, out, n */ ! :"r"(w[1]),"x"(*(t_float*)w[2]),"r"(w[3]),"c"(w[4]) ! :"%xmm1","%xmm2","%xmm3","%xmm4" ); return w+5; --- 663,668 ---- : /* in, value, out, n */ ! :"r"(w[1]),"r"(w[2]),"r"(w[3]),"c"(w[4]) ! :"%xmm0","%xmm1","%xmm2","%xmm3","%xmm4" ); return w+5; *************** *** 710,714 **** ".set T_FLOAT,4 \n" ! "shufps $0, %1, %1 \n" "shrl $4, %3 \n" /* divide by 16 */

--- 715,720 ---- ".set T_FLOAT,4 \n" ! "movss (%1), %%xmm0 \n" ! "shufps $0, %%xmm0, %%xmm0 \n" "shrl $4, %3 \n" /* divide by 16 */

*************** *** 716,732 **** "1: \n" "movaps (%0), %%xmm1 \n" ! "minps %1, %%xmm1 \n" "movaps %%xmm1, (%2) \n"

"movaps 4*T_FLOAT(%0), %%xmm2 \n" ! "minps %1, %%xmm2 \n" "movaps %%xmm2, 4*T_FLOAT(%2) \n" "movaps 8*T_FLOAT(%0), %%xmm3 \n" ! "minps %1, %%xmm3 \n" "movaps %%xmm3, 8*T_FLOAT(%2) \n" "movaps 12*T_FLOAT(%0), %%xmm4 \n" ! "minps %1, %%xmm4 \n" "movaps %%xmm4, 12*T_FLOAT(%2) \n" --- 722,738 ---- "1: \n" "movaps (%0), %%xmm1 \n" ! "minps %%xmm0, %%xmm1 \n" "movaps %%xmm1, (%2) \n"

"movaps 4*T_FLOAT(%0), %%xmm2 \n" ! "minps %%xmm0, %%xmm2 \n" "movaps %%xmm2, 4*T_FLOAT(%2) \n" "movaps 8*T_FLOAT(%0), %%xmm3 \n" ! "minps %%xmm0, %%xmm3 \n" "movaps %%xmm3, 8*T_FLOAT(%2) \n" "movaps 12*T_FLOAT(%0), %%xmm4 \n" ! "minps %%xmm0, %%xmm4 \n" "movaps %%xmm4, 12*T_FLOAT(%2) \n" *************** *** 736,741 **** : /* in, value, out, n */ ! :"r"(w[1]),"x"(*(t_float*)w[2]),"r"(w[3]),"c"(w[4]) ! :"%xmm1","%xmm2","%xmm3","%xmm4" ); return w+5; --- 742,747 ---- : /* in, value, out, n */ ! :"r"(w[1]),"r"(w[2]),"r"(w[3]),"c"(w[4]) ! :"%xmm0","%xmm1","%xmm2","%xmm3","%xmm4" ); return w+5; *************** *** 788,792 **** ".set T_FLOAT,4 \n" ! "shufps $0, %1, %1 \n" "shrl $4, %3 \n" /* divide by 16 */

--- 794,799 ---- ".set T_FLOAT,4 \n" ! "movss (%1), %%xmm0 \n" ! "shufps $0, %%xmm0, %%xmm0 \n" "shrl $4, %3 \n" /* divide by 16 */

*************** *** 794,810 **** "1: \n" "movaps (%0), %%xmm1 \n" ! "maxps %1, %%xmm1 \n" "movaps %%xmm1, (%2) \n"

"movaps 4*T_FLOAT(%0), %%xmm2 \n" ! "maxps %1, %%xmm2 \n" "movaps %%xmm2, 4*T_FLOAT(%2) \n" "movaps 8*T_FLOAT(%0), %%xmm3 \n" ! "maxps %1, %%xmm3 \n" "movaps %%xmm3, 8*T_FLOAT(%2) \n" "movaps 12*T_FLOAT(%0), %%xmm4 \n" ! "maxps %1, %%xmm4 \n" "movaps %%xmm4, 12*T_FLOAT(%2) \n" --- 801,817 ---- "1: \n" "movaps (%0), %%xmm1 \n" ! "maxps %%xmm0, %%xmm1 \n" "movaps %%xmm1, (%2) \n"

"movaps 4*T_FLOAT(%0), %%xmm2 \n" ! "maxps %%xmm0, %%xmm2 \n" "movaps %%xmm2, 4*T_FLOAT(%2) \n" "movaps 8*T_FLOAT(%0), %%xmm3 \n" ! "maxps %%xmm0, %%xmm3 \n" "movaps %%xmm3, 8*T_FLOAT(%2) \n" "movaps 12*T_FLOAT(%0), %%xmm4 \n" ! "maxps %%xmm0, %%xmm4 \n" "movaps %%xmm4, 12*T_FLOAT(%2) \n" *************** *** 814,819 **** : /* in, value, out, n */ ! :"r"(w[1]),"x"(*(t_float*)w[2]),"r"(w[3]),"c"(w[4]) ! :"%xmm1","%xmm2","%xmm3","%xmm4" ); return w+5; --- 821,826 ---- : /* in, value, out, n */ ! :"r"(w[1]),"r"(w[2]),"r"(w[3]),"c"(w[4]) ! :"%xmm0","%xmm1","%xmm2","%xmm3","%xmm4" ); return w+5; *************** *** 825,830 **** ".set T_FLOAT,4 \n" ! "shufps $0, %2, %2 \n" /* lo */ ! "shufps $0, %3, %3 \n" /* hi */ "shrl $4, %4 \n" /* divide by 16 */

--- 832,840 ---- ".set T_FLOAT,4 \n" ! "movss (%2), %%xmm0 \n" /* lo */ ! "shufps $0, %%xmm0, %%xmm0 \n" ! "movss (%3), %%xmm1 \n" /* hi */ ! "shufps $0, %%xmm1, %%xmm1 \n" ! "shrl $4, %4 \n" /* divide by 16 */

*************** *** 832,852 **** "1: \n" "movaps (%0), %%xmm2 \n" ! "maxps %2, %%xmm2 \n" ! "minps %3, %%xmm2 \n" "movaps %%xmm2, (%1) \n"

"movaps 4*T_FLOAT(%0), %%xmm3 \n" ! "maxps %2, %%xmm3 \n" ! "minps %3, %%xmm3 \n" "movaps %%xmm3, 4*T_FLOAT(%1) \n" "movaps 8*T_FLOAT(%0), %%xmm4 \n" ! "maxps %2, %%xmm4 \n" ! "minps %3, %%xmm4 \n" "movaps %%xmm4, 8*T_FLOAT(%1) \n"

"movaps 12*T_FLOAT(%0), %%xmm5 \n" ! "maxps %2, %%xmm5 \n" ! "minps %3, %%xmm5 \n" "movaps %%xmm5, 12*T_FLOAT(%1) \n" --- 842,862 ---- "1: \n" "movaps (%0), %%xmm2 \n" ! "maxps %%xmm0, %%xmm2 \n" ! "minps %%xmm1, %%xmm2 \n" "movaps %%xmm2, (%1) \n"

"movaps 4*T_FLOAT(%0), %%xmm3 \n" ! "maxps %%xmm0, %%xmm3 \n" ! "minps %%xmm1, %%xmm3 \n" "movaps %%xmm3, 4*T_FLOAT(%1) \n" "movaps 8*T_FLOAT(%0), %%xmm4 \n" ! "maxps %%xmm0, %%xmm4 \n" ! "minps %%xmm1, %%xmm4 \n" "movaps %%xmm4, 8*T_FLOAT(%1) \n"

"movaps 12*T_FLOAT(%0), %%xmm5 \n" ! "maxps %%xmm0, %%xmm5 \n" ! "minps %%xmm1, %%xmm5 \n" "movaps %%xmm5, 12*T_FLOAT(%1) \n" *************** *** 856,861 **** : /* in, out, lo, hi, n */ ! :"r"(w[1]),"r"(w[2]),"x"(*(t_float*)w[3]),"x"(*(t_float*)w[4]),"c"(w[5]) ! :"%xmm2","%xmm3","%xmm4","%xmm5" ); return w+6; --- 866,871 ---- : /* in, out, lo, hi, n */ ! :"r"(w[1]),"r"(w[2]),"r"(w[3]),"r"(w[4]),"c"(w[5]) ! :"%xmm0","%xmm1","%xmm2","%xmm3","%xmm4","%xmm5" ); return w+6; *************** *** 945,991 **** }

! float sumvec_simd(t_float* in, t_int n) ! { ! float ret; ! asm( ! ".set T_FLOAT,4 \n" ! "shrl $4, %2 \n" /* divide by 16 */ ! "xorps %%xmm4, %%xmm4 \n" /* zero values */ ! "xorps %%xmm5, %%xmm5 \n" ! "xorps %%xmm6, %%xmm6 \n" ! "xorps %0, %0 \n"

! "1: \n" ! "movaps (%1), %%xmm0 \n" ! "movaps 4*T_FLOAT(%1), %%xmm1 \n" ! "movaps 8*T_FLOAT(%1), %%xmm2 \n" ! "movaps 12*T_FLOAT(%1), %%xmm3 \n"

! "addps %%xmm0,%%xmm4 \n" ! "addps %%xmm1,%%xmm4 \n" ! "addps %%xmm2,%%xmm4 \n" ! "addps %%xmm3,%%xmm4 \n"

! "addl $16*T_FLOAT,%1 \n" ! "loop 1b \n"

! "movhlps %%xmm4, %%xmm5 \n" ! "movups %%xmm4, %%xmm6 \n" ! "movups %%xmm5, %0 \n" ! "shufps $81, %%xmm6, %%xmm6 \n" ! "shufps $81, %0, %0 \n"

! "addss %%xmm4, %%xmm5 \n" ! "addss %%xmm5, %%xmm6 \n" ! "addss %%xmm6, %0 \n"

! :"=x"(ret) ! :"r"(in),"c"(n) ! :"%xmm0","%xmm1","%xmm2","%xmm3", "%xmm4","%xmm5","%xmm6"); ! return ret; ! }

--- 955,1002 ---- }

! /* tb: is this needed? */ ! /* float sumvec_simd(t_float* in, t_int n) */ ! /* { */ ! /* float ret; */ ! /* asm( */ ! /* ".set T_FLOAT,4 \n" */ ! /* "shrl $4, %2 \n" /* divide by 16 */ */ ! /* "xorps %%xmm4, %%xmm4 \n" /* zero values */ */ ! /* "xorps %%xmm5, %%xmm5 \n" */ ! /* "xorps %%xmm6, %%xmm6 \n" */ ! /* "xorps %0, %0 \n" */

! /* "1: \n" */ ! /* "movaps (%1), %%xmm0 \n" */ ! /* "movaps 4*T_FLOAT(%1), %%xmm1 \n" */ ! /* "movaps 8*T_FLOAT(%1), %%xmm2 \n" */ ! /* "movaps 12*T_FLOAT(%1), %%xmm3 \n" */

! /* "addps %%xmm0,%%xmm4 \n" */ ! /* "addps %%xmm1,%%xmm4 \n" */ ! /* "addps %%xmm2,%%xmm4 \n" */ ! /* "addps %%xmm3,%%xmm4 \n" */

! /* "addl $16*T_FLOAT,%1 \n" */ ! /* "loop 1b \n" */

! /* "movhlps %%xmm4, %%xmm5 \n" */ ! /* "movups %%xmm4, %%xmm6 \n" */ ! /* "movups %%xmm5, %0 \n" */ ! /* "shufps $81, %%xmm6, %%xmm6 \n" */ ! /* "shufps $81, %0, %0 \n" */

! /* "addss %%xmm4, %%xmm5 \n" */ ! /* "addss %%xmm5, %%xmm6 \n" */ ! /* "addss %%xmm6, %0 \n" */

! /* :"=x"(ret) */ ! /* :"r"(in),"c"(n) */ ! /* :"%xmm0","%xmm1","%xmm2","%xmm3", "%xmm4","%xmm5","%xmm6"); */ ! /* return ret; */ ! /* } */

*************** *** 1001,1005 **** "xorps %%xmm3, %%xmm3 \n" "xorps %%xmm4, %%xmm4 \n" ! "xorps %0, %0 \n" "1: \n" --- 1012,1017 ---- "xorps %%xmm3, %%xmm3 \n" "xorps %%xmm4, %%xmm4 \n" ! "xorps %%xmm5, %%xmm5 \n" ! "1: \n" *************** *** 1042,1056 **** "movhlps %%xmm2, %%xmm3 \n" /* unpack xmm0 */ "movups %%xmm2, %%xmm4 \n" ! "movups %%xmm3, %0 \n" "shufps $81, %%xmm4, %%xmm4 \n" ! "shufps $81, %0, %0 \n"

"addss %%xmm2, %%xmm3 \n" "addss %%xmm3, %%xmm4 \n" ! "addss %%xmm4, %0 \n" !

! :"=x"(ret) ! :"r"(in),"c"(n), "r"(hp) :"%xmm0","%xmm1","%xmm2","%xmm3", "%xmm4", "%xmm5"); return ret; --- 1054,1069 ---- "movhlps %%xmm2, %%xmm3 \n" /* unpack xmm0 */ "movups %%xmm2, %%xmm4 \n" ! "movups %%xmm3, %%xmm5 \n" "shufps $81, %%xmm4, %%xmm4 \n" ! "shufps $81, %%xmm5, %%xmm5 \n"

"addss %%xmm2, %%xmm3 \n" "addss %%xmm3, %%xmm4 \n" ! "addss %%xmm4, %%xmm5 \n" ! ! "movss %%xmm5, (%0) \n"

! : ! :"r"(&ret),"r"(in),"c"(n), "r"(hp) :"%xmm0","%xmm1","%xmm2","%xmm3", "%xmm4", "%xmm5"); return ret; *************** *** 1074,1101 **** "shrl $4, %2 \n" /* divide by 16 */ "movaps (2b), %%xmm0 \n" ! "shufps $0, %0, %0 \n"

"1: \n" "movaps (%1), %%xmm1 \n" "andps %%xmm0, %%xmm1 \n" ! "maxps %%xmm1, %0 \n"

"movaps 4*T_FLOAT(%1), %%xmm1 \n" "andps %%xmm0, %%xmm1 \n" ! "maxps %%xmm1, %0 \n"

"movaps 8*T_FLOAT(%1), %%xmm1 \n" "andps %%xmm0, %%xmm1 \n" ! "maxps %%xmm1, %0 \n"

"movaps 12*T_FLOAT(%1), %%xmm1 \n" "andps %%xmm0, %%xmm1 \n" ! "maxps %%xmm1, %0 \n"

"addl $16*T_FLOAT, %1 \n" "loop 1b \n" ! "movhlps %0, %%xmm2 \n" ! "movaps %0, %%xmm3 \n" "movaps %%xmm2, %%xmm4 \n" "shufps $81, %%xmm3, %%xmm3 \n" --- 1087,1116 ---- "shrl $4, %2 \n" /* divide by 16 */ "movaps (2b), %%xmm0 \n" ! ! "movss (%0), %%xmm5 \n" /* cur_max */ ! "shufps $0, %%xmm5, %%xmm5 \n"

"1: \n" "movaps (%1), %%xmm1 \n" "andps %%xmm0, %%xmm1 \n" ! "maxps %%xmm1, %%xmm5 \n"

"movaps 4*T_FLOAT(%1), %%xmm1 \n" "andps %%xmm0, %%xmm1 \n" ! "maxps %%xmm1, %%xmm5 \n"

"movaps 8*T_FLOAT(%1), %%xmm1 \n" "andps %%xmm0, %%xmm1 \n" ! "maxps %%xmm1, %%xmm5 \n"

"movaps 12*T_FLOAT(%1), %%xmm1 \n" "andps %%xmm0, %%xmm1 \n" ! "maxps %%xmm1, %%xmm5 \n"

"addl $16*T_FLOAT, %1 \n" "loop 1b \n" ! "movhlps %%xmm5, %%xmm2 \n" ! "movaps %%xmm5, %%xmm3 \n" "movaps %%xmm2, %%xmm4 \n" "shufps $81, %%xmm3, %%xmm3 \n" *************** *** 1104,1111 **** "maxss %%xmm2, %%xmm3 \n" "maxss %%xmm3, %%xmm4 \n" ! "maxss %%xmm4, %0 \n"

! :"=x"(cur_max) ! :"r"(vec),"c"(n), "0"(cur_max) :"%xmm0","%xmm1","%xmm2","%xmm3", "%xmm4", "%xmm5");

--- 1119,1128 ---- "maxss %%xmm2, %%xmm3 \n" "maxss %%xmm3, %%xmm4 \n" ! "maxss %%xmm4, %%xmm5 \n"

! "movss %%xmm5, (%0) \n" ! ! : ! :"r"(&cur_max), "r"(vec),"c"(n) :"%xmm0","%xmm1","%xmm2","%xmm3", "%xmm4", "%xmm5");