arm/crypto/poly1305-armv4.pl

495 my ($R0,$R1,$S1,$R2,$S2,$R3,$S3,$R4,$S4) = map("d$_",(0..9));
530 	vdup.32	$R0,r2			@ r^1 in both lanes
554 	vmull.u32	$D0,$R0,${R0}[1]
555 	vmull.u32	$D1,$R1,${R0}[1]
556 	vmull.u32	$D2,$R2,${R0}[1]
557 	vmull.u32	$D3,$R3,${R0}[1]
558 	vmull.u32	$D4,$R4,${R0}[1]
561 	vmlal.u32	$D1,$R0,${R1}[1]
569 	vmlal.u32	$D2,$R0,${R2}[1]
573 	vmlal.u32	$D3,$R0,${R3}[1]
582 	vmlal.u32	$D4,$R0,${R4}[1]
609 	@ H4 = H4*R0 + H3*R1 + H2*R2 + H1*R3 + H0 * R4,
671 	vtrn.32		$R0,$D0#lo		@ r^2:r^1
686 	vst4.32		{${R0}[0],${R1}[0],${S1}[0],${R2}[0]},[$tbl0]!
687 	vst4.32		{${R0}[1],${R1}[1],${S1}[1],${R2}[1]},[$tbl1]!
700 	vmov		$R0,$D0#lo		@ r^4:r^3
714 	vst4.32		{${R0}[0],${R1}[0],${S1}[0],${R2}[0]},[$tbl0]!
715 	vst4.32		{${R0}[1],${R1}[1],${S1}[1],${R2}[1]},[$tbl1]!
878 	vld4.32		{${R0}[1],${R1}[1],${S1}[1],${R2}[1]},[$tbl1]!	@ load r^2
879 	vld4.32		{${R0}[0],${R1}[0],${S1}[0],${R2}[0]},[$tbl0]!	@ load r^4
907 	vmull.u32	$D2,$H2#hi,${R0}[1]
909 	vmull.u32	$D0,$H0#hi,${R0}[1]
911 	vmull.u32	$D3,$H3#hi,${R0}[1]
914 	vmull.u32	$D1,$H1#hi,${R0}[1]
917 	vmull.u32	$D4,$H4#hi,${R0}[1]
951 	vmlal.u32	$D3,$H3#lo,${R0}[0]
952 	vmlal.u32	$D0,$H0#lo,${R0}[0]
953 	vmlal.u32	$D4,$H4#lo,${R0}[0]
954 	vmlal.u32	$D1,$H1#lo,${R0}[0]
955 	vmlal.u32	$D2,$H2#lo,${R0}[0]
1060 	vld4.32		{${R0}[1],${R1}[1],${S1}[1],${R2}[1]},[$tbl1]!	@ load r^1
1061 	vld4.32		{${R0}[0],${R1}[0],${S1}[0],${R2}[0]},[$tbl0]!	@ load r^2
1064 	vmull.u32	$D2,$H2#hi,$R0
1066 	vmull.u32	$D0,$H0#hi,$R0
1068 	vmull.u32	$D3,$H3#hi,$R0
1070 	vmull.u32	$D1,$H1#hi,$R0
1072 	vmull.u32	$D4,$H4#hi,$R0
1113 	vld4.32		{${R0}[1],${R1}[1],${S1}[1],${R2}[1]},[$tbl1]!	@ load r^3
1114 	vld4.32		{${R0}[0],${R1}[0],${S1}[0],${R2}[0]},[$tbl0]!	@ load r^4
1116 	vmlal.u32	$D2,$H2#lo,$R0
1117 	vmlal.u32	$D0,$H0#lo,$R0
1118 	vmlal.u32	$D3,$H3#lo,$R0
1119 	vmlal.u32	$D1,$H1#lo,$R0
1120 	vmlal.u32	$D4,$H4#lo,$R0