// *********************************** // interpolate8x8_ia64.s // optimized for IA-64 // Authors : Kai K黨n // Alexander Viehl // last update : 13.7.2002 // *********************************** .file "interpolate8x8_ia64.s" .pred.safe_across_calls p1-p5,p16-p63 .text .align 16 .global interpolate8x8_halfpel_h_ia64# .proc interpolate8x8_halfpel_h_ia64# interpolate8x8_halfpel_h_ia64: LL=3 SL=1 SL2=1 OL=1 OL2=1 AVL=1 AL=1 PSH=1 ML=1 STL=3 alloc r9=ar.pfs,4, 60,0,64 mov r20 = ar.lc mov r21 = pr dep.z r22 = r33,3,3 // rshift of src and r14 = -8,r33 // align src mov r15 = r32 // get dest mov r16 = r34 // stride sub r17 = 1,r35 // 1-rounding ;; add r18 = 8,r14 mux2 r17 = r17, 0x00 // broadcast 1-rounding sub r24 = 64,r22 // lshift of src add r26 = 8,r22 // rshift of src+1 sub r27 = 56,r22 // lshift of src+1 mov ar.lc = 7 // loopcounter mov ar.ec = LL + SL +OL + AVL + STL + 2*PSH + 2*AL + ML // sum of latencies mov pr.rot = 1 ;; .rotr ald1[LL+1],ald2[LL+1],shru1[SL+1],shl1[SL+1],shru2[SL+1],shl2[SL+1],or1[OL+1+PSH],or2[OL+1+PSH],pshb1[PSH+1],pshb2[PSH+1],pshe1[PSH+1],pshe2[PSH+1+AL],pshe3[PSH+1],pshe4[PSH+1+AL],add1[AL+1],add2[AL+1],add3[AL+1],add4[AL+1],avg1[AVL+1],avg2[AVL+1],pmix1[ML+1] .rotp aldp[LL], sh1p[SL], or1p[OL], pshb[PSH], pshe[PSH],addp[AL],add2p[AL],pavg1p[AVL],mixp[ML],stp[STL] .Lloop_interpolate: (aldp[0]) ld8 ald1[0] = [r14],r16 // load aligned src (aldp[0]) ld8 ald2[0] = [r18],r16 // and aligned src+8 (sh1p[0]) shr.u shru1[0] = ald1[LL],r22 // get src (sh1p[0]) shl shl1[0] = ald2[LL],r27 (sh1p[0]) shr.u shru2[0] = ald1[LL],r26 // get src+1 (sh1p[0]) shl shl2[0] = ald2[LL],r24 (or1p[0]) or or1[0] = shru1[SL],shl2[SL] // merge things (or1p[0]) or or2[0] = shru2[SL],shl1[SL] (pshb[0]) pshl2 pshb1[0] = or1[OL],8 (pshb[0]) pshl2 pshb2[0] = or2[OL],8 (pshe[0]) pshr2.u pshe1[0] = pshb1[PSH],8 (pshe[0]) pshr2.u pshe2[0] = pshb2[PSH],8 (pshe[0]) pshr2.u pshe3[0] = or1[PSH+OL],8 (pshe[0]) pshr2.u pshe4[0] = or2[PSH+OL],8 (addp[0]) padd2.sss add1[0] = pshe1[PSH],r17 // add 1-rounding (addp[0]) padd2.sss add2[0] = pshe3[PSH],r17 // add 1-rounding (add2p[0]) padd2.uus add3[0] = pshe2[AL+PSH],add1[AL] (add2p[0]) padd2.uus add4[0] = pshe4[AL+PSH],add2[AL] (pavg1p[0]) pshr2.u avg1[0] = add3[AL],1 // parallel average (pavg1p[0]) pshr2.u avg2[0] = add4[AL],1 // parallel average (mixp[0]) mix1.r pmix1[0] = avg2[AVL],avg1[AVL] (stp[0]) st8 [r15] = pmix1[ML] // store results (stp[0]) add r15 = r15,r16 br.ctop.sptk.few .Lloop_interpolate ;; mov ar.lc = r20 mov pr = r21,-1 br.ret.sptk.many b0 .endp interpolate8x8_halfpel_h_ia64# .align 16 .global interpolate8x8_halfpel_v_ia64# .proc interpolate8x8_halfpel_v_ia64# interpolate8x8_halfpel_v_ia64: LL=3 SL=1 SL2=1 OL=1 OL2=1 AVL=1 AL=1 PSH=1 ML=1 STL=3 alloc r9=ar.pfs,4, 60,0,64 mov r20 = ar.lc mov r21 = pr dep.z r22 = r33,3,3 and r14 = -8,r33 mov r15 = r32 mov r16 = r34 sub r17 = 1,r35 ;; add r18 = 8,r14 add r19 = r14,r16 // src + stride mux2 r17 = r17, 0x00 sub r24 = 64,r22 ;; add r26 = 8,r19 // src + stride + 8 mov ar.lc = 7 mov ar.ec = LL + SL +OL + AVL + STL + 2*PSH + 2*AL + ML // sum of latencies mov pr.rot = 1 ;; .rotr ald1[LL+1],ald2[LL+1],ald3[LL+1],ald4[LL+1],shru1[SL+1],shl1[SL+1],shru2[SL+1],shl2[SL+1],or1[OL+1+PSH],or2[OL+1+PSH],pshb1[PSH+1],pshb2[PSH+1],pshe1[PSH+1],pshe2[PSH+1+AL],pshe3[PSH+1],pshe4[PSH+1+AL],add1[AL+1],add2[AL+1],add3[AL+1],add4[AL+1],avg1[AVL+1],avg2[AVL+1],pmix1[ML+1] .rotp aldp[LL], sh1p[SL], or1p[OL], pshb[PSH], pshe[PSH],addp[AL],add2p[AL],pavg1p[AVL],mixp[ML],stp[STL] .Lloop_interpolate2: (aldp[0]) ld8 ald1[0] = [r14],r16 (aldp[0]) ld8 ald2[0] = [r18],r16 (aldp[0]) ld8 ald3[0] = [r19],r16 (aldp[0]) ld8 ald4[0] = [r26],r16 (sh1p[0]) shr.u shru1[0] = ald1[LL],r22 (sh1p[0]) shl shl1[0] = ald2[LL],r24 (sh1p[0]) shr.u shru2[0] = ald3[LL],r22 (sh1p[0]) shl shl2[0] = ald4[LL],r24 (or1p[0]) or or1[0] = shru1[SL],shl1[SL] (or1p[0]) or or2[0] = shru2[SL],shl2[SL] (pshb[0]) pshl2 pshb1[0] = or1[OL],8 (pshb[0]) pshl2 pshb2[0] = or2[OL],8 (pshe[0]) pshr2.u pshe1[0] = pshb1[PSH],8 (pshe[0]) pshr2.u pshe2[0] = pshb2[PSH],8 (pshe[0]) pshr2.u pshe3[0] = or1[PSH+OL],8 (pshe[0]) pshr2.u pshe4[0] = or2[PSH+OL],8 (addp[0]) padd2.sss add1[0] = pshe1[PSH],r17 // add 1-rounding (addp[0]) padd2.sss add2[0] = pshe3[PSH],r17 // add 1-rounding (add2p[0]) padd2.uus add3[0] = pshe2[AL+PSH],add1[AL] (add2p[0]) padd2.uus add4[0] = pshe4[AL+PSH],add2[AL] (pavg1p[0]) pshr2.u avg1[0] = add3[AL],1 // parallel average (pavg1p[0]) pshr2.u avg2[0] = add4[AL],1 // parallel average (mixp[0]) mix1.r pmix1[0] = avg2[AVL],avg1[AVL] (stp[0]) st8 [r15] = pmix1[ML] (stp[0]) add r15 = r15,r16 br.ctop.sptk.few .Lloop_interpolate2 ;; mov ar.lc = r20 mov pr = r21,-1 br.ret.sptk.many b0 .endp interpolate8x8_halfpel_v_ia64# .align 16 .global interpolate8x8_halfpel_hv_ia64# .proc interpolate8x8_halfpel_hv_ia64# interpolate8x8_halfpel_hv_ia64: LL=3 SL=1 SL2=1 OL=1 OL2=1 AVL=1 AL=1 PSH=1 ML=1 STL=3 alloc r9=ar.pfs,4, 92,0,96 mov r20 = ar.lc mov r21 = pr dep.z r22 = r33,3,3 and r14 = -8,r33 mov r15 = r32 mov r16 = r34 sub r17 = 2,r35 ;; add r18 = 8,r14 add r19 = r14,r16 mux2 r17 = r17, 0x00 add r27 = 8,r22 sub r28 = 56,r22 sub r24 = 64,r22 ;; add r26 = 8,r19 mov ar.lc = 7 mov ar.ec = LL + SL +OL + AVL + STL + 2*PSH + 2*AL + ML // sum of latencies mov pr.rot = 1 ;; .rotr ald1[LL+1],ald2[LL+1],ald3[LL+1],ald4[LL+1],shru1[SL+1],shl1[SL+1],shru2[SL+1],shl2[SL+1],shl3[SL+1],shru3[SL+1],shl4[SL+1],shru4[SL+1],or1[OL+1+PSH],or2[OL+1+PSH],or3[OL+1+PSH],or4[OL+1+PSH],pshb1[PSH+1],pshb2[PSH+1],pshb3[PSH+1],pshb4[PSH+1],pshe1[PSH+1],pshe2[PSH+1+AL],pshe3[PSH+1],pshe4[PSH+1+AL],pshe5[PSH+1],pshe6[PSH+1+AL],pshe7[PSH+1],pshe8[PSH+1+AL],add1[AL+1],add2[AL+1],add3[AL+1],add4[AL+1],add5[AL+1],add6[AL+1],add7[AL+1],add8[AL+1],avg1[AVL+1],avg2[AVL+1],pmix1[ML+1] .rotp aldp[LL], sh1p[SL],or1p[OL],pshb[PSH],pshe[PSH],addp[AL],add2p[AL],add3p[AL],pavg1p[AVL],mixp[ML],stp[STL] .Lloop_interpolate3: (aldp[0]) ld8 ald1[0] = [r14],r16 (aldp[0]) ld8 ald2[0] = [r18],r16 (aldp[0]) ld8 ald3[0] = [r19],r16 (aldp[0]) ld8 ald4[0] = [r26],r16 (sh1p[0]) shr.u shru1[0] = ald1[LL],r22 (sh1p[0]) shl shl1[0] = ald2[LL],r24 (sh1p[0]) shr.u shru2[0] = ald3[LL],r22 (sh1p[0]) shl shl2[0] = ald4[LL],r24 (sh1p[0]) shr.u shru3[0] = ald1[LL],r27 (sh1p[0]) shl shl3[0] = ald2[LL],r28 (sh1p[0]) shr.u shru4[0] = ald3[LL],r27 (sh1p[0]) shl shl4[0] = ald4[LL],r28 (or1p[0]) or or1[0] = shru1[SL],shl1[SL] (or1p[0]) or or2[0] = shru2[SL],shl2[SL] (or1p[0]) or or3[0] = shru3[SL],shl3[SL] (or1p[0]) or or4[0] = shru4[SL],shl4[SL] (pshb[0]) pshl2 pshb1[0] = or1[OL],8 (pshb[0]) pshl2 pshb2[0] = or2[OL],8 (pshb[0]) pshl2 pshb3[0] = or3[OL],8 (pshb[0]) pshl2 pshb4[0] = or4[OL],8 (pshe[0]) pshr2.u pshe1[0] = pshb1[PSH],8 (pshe[0]) pshr2.u pshe2[0] = pshb2[PSH],8 (pshe[0]) pshr2.u pshe3[0] = or1[PSH+OL],8 (pshe[0]) pshr2.u pshe4[0] = or2[PSH+OL],8 (pshe[0]) pshr2.u pshe5[0] = pshb3[PSH],8 (pshe[0]) pshr2.u pshe6[0] = pshb4[PSH],8 (pshe[0]) pshr2.u pshe7[0] = or3[PSH+OL],8 (pshe[0]) pshr2.u pshe8[0] = or4[PSH+OL],8 (addp[0]) padd2.sss add1[0] = pshe1[PSH],pshe2[PSH] // add 1-rounding (addp[0]) padd2.sss add2[0] = pshe3[PSH],pshe4[PSH] // add 1-rounding (addp[0]) padd2.sss add5[0] = pshe5[PSH],pshe6[PSH] // add 1-rounding (addp[0]) padd2.sss add6[0] = pshe7[PSH],pshe8[PSH] // add 1-rounding (add2p[0]) padd2.uus add3[0] = add1[AL],add5[AL] (add2p[0]) padd2.uus add4[0] = add2[AL],add6[AL] (add3p[0]) padd2.uus add7[0] = add3[AL],r17 (add3p[0]) padd2.uus add8[0] = add4[AL],r17 (pavg1p[0]) pshr2.u avg1[0] = add7[AL],2 // parallel average (pavg1p[0]) pshr2.u avg2[0] = add8[AL],2 // parallel average (mixp[0]) mix1.r pmix1[0] = avg2[AVL],avg1[AVL] (stp[0]) st8 [r15] = pmix1[ML] (stp[0]) add r15 = r15,r16 br.ctop.sptk.few .Lloop_interpolate3 ;; mov ar.lc = r20 mov pr = r21,-1 br.ret.sptk.many b0 .endp interpolate8x8_halfpel_hv_ia64#
