Lines Matching +full:3 +full:d1
47 ldrb r7,[r1,#3]
180 ldrb r1,[lr,#-3]
230 adds r2,r2,r1 @ d1+=d0>>32
250 adds r6,lr,r0 @ d2+=d1>>32
259 and r8,r8,#3
364 strb r3,[r1,#3]
403 vdup.32 d1,r3
420 @ d1 = h1*r0 + h0*r1 + h4*5*r2 + h3*5*r3 + h2*5*r4
426 vmull.u32 q6,d1,d0[1]
432 vmlal.u32 q6,d0,d1[1]
433 vmlal.u32 q7,d1,d1[1]
434 vmlal.u32 q8,d3,d1[1]
435 vmlal.u32 q9,d5,d1[1]
439 vmlal.u32 q8,d1,d3[1]
447 vmlal.u32 q9,d1,d5[1]
450 vmlal.u32 q5,d1,d8[1]
491 @ input H0 is limited by (5*4+1)*3 addends, or 58 bits, while
492 @ 5*H4 by 5*5*3, or 59[!] bits. How is this relevant? vmlal.u32
545 vtrn.32 d1,d12
550 vshl.u32 d2,d1,#2
553 vadd.i32 d2,d2,d1
557 vst4.32 {d0[0],d1[0],d2[0],d3[0]},[r6]!
558 vst4.32 {d0[1],d1[1],d2[1],d3[1]},[r7]!
569 add r7,r0,#(48+3*4*9)
571 vmov d0,d10 @ r^4:r^3
573 vmov d1,d12
585 vst4.32 {d0[0],d1[0],d2[0],d3[0]},[r6]!
586 vst4.32 {d0[1],d1[1],d2[1],d3[1]},[r7]!
723 vld4.32 {d21,d23,d25,d27},[r4] @ inp[2:3] (or 0)
727 addhi r6,r0,#(48+3*9*4)
753 vld4.32 {d0[1],d1[1],d2[1],d3[1]},[r7]! @ load r^2
754 vld4.32 {d0[0],d1[0],d2[0],d3[0]},[r6]! @ load r^4
763 @ ((inp[1]*r^4+inp[3]*r^2+inp[5])*r^3+inp[7]*r
766 @ ((inp[1]*r^4+inp[3]*r^2+inp[5])*r^4+inp[7]*r^2+inp[9])*r
769 @ Note that we start with inp[2:3]*r^2. This is because it
775 @ d1 = h1*r0 + h0*r1 + h4*5*r2 + h3*5*r3 + h2*5*r4
779 @ inp[2:3]*r^2
787 vmlal.u32 q7,d23,d1[1]
797 vmlal.u32 q8,d25,d1[1]
799 vmlal.u32 q6,d21,d1[1]
800 vmlal.u32 q9,d27,d1[1]
820 vld4.32 {d21,d23,d25,d27},[r4] @ inp[2:3] (or 0)
833 vmlal.u32 q8,d24,d1[0]
835 vmlal.u32 q9,d26,d1[0]
836 vmlal.u32 q6,d20,d1[0]
837 vmlal.u32 q7,d22,d1[0]
869 @ inp[0:3] previously loaded to q10-q13 and smashed to q10-q14.
919 @ multiply (inp[0:1]+hash) or inp[2:3] by r^2:r^1
935 vld4.32 {d0[1],d1[1],d2[1],d3[1]},[r7]! @ load r^1
936 vld4.32 {d0[0],d1[0],d2[0],d3[0]},[r6]! @ load r^2
951 vmlal.u32 q8,d25,d1
953 vmlal.u32 q6,d21,d1
954 vmlal.u32 q9,d27,d1
955 vmlal.u32 q7,d23,d1
970 addne r6,r0,#(48+3*9*4)
986 @ (hash+inp[0:1])*r^4:r^3 and accumulate
988 vld4.32 {d0[1],d1[1],d2[1],d3[1]},[r7]! @ load r^3
989 vld4.32 {d0[0],d1[0],d2[0],d3[0]},[r6]! @ load r^4
999 vmlal.u32 q8,d24,d1
1001 vmlal.u32 q6,d20,d1
1002 vmlal.u32 q9,d26,d1
1003 vmlal.u32 q7,d22,d1
1106 and r7,r6,#3