+&set_label("SSSE3",32);
+{ my ($cnt,$frame)=("ecx","edx");
+ my @X=map("xmm$_",(0..7));
+ my $j;
+ my $i=0;
+
+ &lea ($frame,&DWP(-64,"esp"));
+ &sub ("esp",256);
+
+ # fixed stack frame layout
+ #
+ # +0 A B C D E F G H # backing store
+ # +64 X[0]+K[i] .. X[15]+K[i] # XMM->MM xfer area
+ # +192 # XMM off-load ring buffer
+ # +256 # saved parameters
+
+ &movdqa (@X[1],&QWP(80*8,$K512)); # byte swap mask
+ &movdqu (@X[0],&QWP(0,"edi"));
+ &pshufb (@X[0],@X[1]);
+ for ($j=0;$j<8;$j++) {
+ &movdqa (&QWP(16*(($j-1)%4),$frame),@X[3]) if ($j>4); # off-load
+ &movdqa (@X[3],&QWP(16*($j%8),$K512));
+ &movdqa (@X[2],@X[1]) if ($j<7); # perpetuate byte swap mask
+ &movdqu (@X[1],&QWP(16*($j+1),"edi")) if ($j<7); # next input
+ &movdqa (@X[1],&QWP(16*(($j+1)%4),$frame)) if ($j==7);# restore @X[0]
+ &paddq (@X[3],@X[0]);
+ &pshufb (@X[1],@X[2]) if ($j<7);
+ &movdqa (&QWP(16*($j%8)-128,$frame),@X[3]); # xfer X[i]+K[i]
+
+ push(@X,shift(@X)); # rotate(@X)
+ }
+ #&jmp (&label("loop_ssse3"));
+ &nop ();
+
+&set_label("loop_ssse3",32);
+ &movdqa (@X[2],&QWP(16*(($j+1)%4),$frame)); # pre-restore @X[1]
+ &movdqa (&QWP(16*(($j-1)%4),$frame),@X[3]); # off-load @X[3]
+ &lea ($K512,&DWP(16*8,$K512));
+
+ #&movq ($Asse2,$A); # off-load A-H
+ &movq ($Bsse2,"mm1");
+ &mov ("ebx","edi");
+ &movq ($Csse2,$BxC);
+ &lea ("edi",&DWP(128,"edi")); # advance input
+ &movq ($Dsse2,"mm3");
+ &cmp ("edi","eax");
+ #&movq ($Esse2,$E);
+ &movq ($Fsse2,"mm5");
+ &cmovb ("ebx","edi");
+ &movq ($Gsse2,"mm6");
+ &mov ("ecx",4); # loop counter
+ &pxor ($BxC,"mm1"); # magic
+ &movq ($Hsse2,"mm7");
+ &pxor ("mm3","mm3"); # magic
+
+ &jmp (&label("00_47_ssse3"));
+
+sub BODY_00_15_ssse3 { # "phase-less" copy of BODY_00_15_sse2
+ (
+ '&movq ("mm1",$E)', # %mm1 is sliding right
+ '&movq ("mm7",&QWP(((-8*$i)%128)-128,$frame))',# X[i]+K[i]
+ '&pxor ("mm5","mm6")', # f^=g
+ '&psrlq ("mm1",14)',
+ '&movq (&QWP(8*($i+4)%64,"esp"),$E)', # modulo-scheduled save e
+ '&pand ("mm5",$E)', # f&=e
+ '&psllq ($E,23)', # $E is sliding left
+ '&paddq ($A,"mm3")', # [h+=Maj(a,b,c)]
+ '&movq ("mm3","mm1")', # %mm3 is T1
+ '&psrlq("mm1",4)',
+ '&pxor ("mm5","mm6")', # Ch(e,f,g)
+ '&pxor ("mm3",$E)',
+ '&psllq($E,23)',
+ '&pxor ("mm3","mm1")',
+ '&movq (&QWP(8*$i%64,"esp"),$A)', # modulo-scheduled save a
+ '&paddq("mm7","mm5")', # X[i]+=Ch(e,f,g)
+ '&pxor ("mm3",$E)',
+ '&psrlq("mm1",23)',
+ '&paddq("mm7",&QWP(8*($i+7)%64,"esp"))', # X[i]+=h
+ '&pxor ("mm3","mm1")',
+ '&psllq($E,4)',
+ '&pxor ("mm3",$E)', # T1=Sigma1_512(e)
+
+ '&movq ($E,&QWP(8*($i+3)%64,"esp"))', # e = load d, e in next round
+ '&paddq ("mm3","mm7")', # T1+=X[i]
+ '&movq ("mm5",$A)', # %mm5 is sliding right
+ '&psrlq("mm5",28)',
+ '&paddq ($E,"mm3")', # d += T1
+ '&movq ("mm6",$A)', # %mm6 is sliding left
+ '&movq ("mm7","mm5")',
+ '&psllq("mm6",25)',
+ '&movq ("mm1",&QWP(8*($i+1)%64,"esp"))', # load b
+ '&psrlq("mm5",6)',
+ '&pxor ("mm7","mm6")',
+ '&psllq("mm6",5)',
+ '&pxor ("mm7","mm5")',
+ '&pxor ($A,"mm1")', # a^b, b^c in next round
+ '&psrlq("mm5",5)',
+ '&pxor ("mm7","mm6")',
+ '&pand ($BxC,$A)', # (b^c)&(a^b)
+ '&psllq("mm6",6)',
+ '&pxor ("mm7","mm5")',
+ '&pxor ($BxC,"mm1")', # [h=]Maj(a,b,c)
+ '&pxor ("mm6","mm7")', # Sigma0_512(a)
+ '&movq ("mm5",&QWP(8*($i+5-1)%64,"esp"))', # pre-load f
+ '&paddq ($BxC,"mm6")', # h+=Sigma0(a)
+ '&movq ("mm6",&QWP(8*($i+6-1)%64,"esp"))', # pre-load g
+
+ '($A,$BxC) = ($BxC,$A); $i--;'
+ );
+}
+
+&set_label("00_47_ssse3",32);
+
+ for(;$j<16;$j++) {
+ my ($t0,$t2,$t1)=@X[2..4];
+ my @insns = (&BODY_00_15_ssse3(),&BODY_00_15_ssse3());
+
+ &movdqa ($t2,@X[5]);
+ &movdqa (@X[1],$t0); # restore @X[1]
+ &palignr ($t0,@X[0],8); # X[1..2]
+ &movdqa (&QWP(16*($j%4),$frame),@X[4]); # off-load @X[4]
+ &palignr ($t2,@X[4],8); # X[9..10]
+
+ &movdqa ($t1,$t0);
+ &psrlq ($t0,7);
+ &paddq (@X[0],$t2); # X[0..1] += X[9..10]
+ &movdqa ($t2,$t1);
+ &psrlq ($t1,1);
+ &psllq ($t2,64-8);
+ &pxor ($t0,$t1);
+ &psrlq ($t1,8-1);
+ &pxor ($t0,$t2);
+ &psllq ($t2,8-1);
+ &pxor ($t0,$t1);
+ &movdqa ($t1,@X[7]);
+ &pxor ($t0,$t2); # sigma0(X[1..2])
+ &movdqa ($t2,@X[7]);
+ &psrlq ($t1,6);
+ &paddq (@X[0],$t0); # X[0..1] += sigma0(X[1..2])
+
+ &movdqa ($t0,@X[7]);
+ &psrlq ($t2,19);
+ &psllq ($t0,64-61);
+ &pxor ($t1,$t2);
+ &psrlq ($t2,61-19);
+ &pxor ($t1,$t0);
+ &psllq ($t0,61-19);
+ &pxor ($t1,$t2);
+ &movdqa ($t2,&QWP(16*(($j+2)%4),$frame));# pre-restore @X[1]
+ &pxor ($t1,$t0); # sigma0(X[1..2])
+ &movdqa ($t0,&QWP(16*($j%8),$K512));
+ eval(shift(@insns));
+ &paddq (@X[0],$t1); # X[0..1] += sigma0(X[14..15])
+ eval(shift(@insns));
+ eval(shift(@insns));
+ eval(shift(@insns));
+ eval(shift(@insns));
+ &paddq ($t0,@X[0]);
+ foreach(@insns) { eval; }
+ &movdqa (&QWP(16*($j%8)-128,$frame),$t0);# xfer X[i]+K[i]
+
+ push(@X,shift(@X)); # rotate(@X)
+ }
+ &lea ($K512,&DWP(16*8,$K512));
+ &dec ("ecx");
+ &jnz (&label("00_47_ssse3"));
+
+ &movdqa (@X[1],&QWP(0,$K512)); # byte swap mask
+ &lea ($K512,&DWP(-80*8,$K512)); # rewind
+ &movdqu (@X[0],&QWP(0,"ebx"));
+ &pshufb (@X[0],@X[1]);
+
+ for ($j=0;$j<8;$j++) { # load next or same block
+ my @insns = (&BODY_00_15_ssse3(),&BODY_00_15_ssse3());
+
+ &movdqa (&QWP(16*(($j-1)%4),$frame),@X[3]) if ($j>4); # off-load
+ &movdqa (@X[3],&QWP(16*($j%8),$K512));
+ &movdqa (@X[2],@X[1]) if ($j<7); # perpetuate byte swap mask
+ &movdqu (@X[1],&QWP(16*($j+1),"ebx")) if ($j<7); # next input
+ &movdqa (@X[1],&QWP(16*(($j+1)%4),$frame)) if ($j==7);# restore @X[0]
+ &paddq (@X[3],@X[0]);
+ &pshufb (@X[1],@X[2]) if ($j<7);
+ foreach(@insns) { eval; }
+ &movdqa (&QWP(16*($j%8)-128,$frame),@X[3]);# xfer X[i]+K[i]
+
+ push(@X,shift(@X)); # rotate(@X)
+ }
+
+ #&movq ($A,$Asse2); # load A-H
+ &movq ("mm1",$Bsse2);
+ &paddq ($A,"mm3"); # from BODY_00_15
+ #&movq ($BxC,$Csse2);
+ &movq ("mm3",$Dsse2);
+ #&movq ($E,$Esse2);
+ #&movq ("mm5",$Fsse2);
+ #&movq ("mm6",$Gsse2);
+ &movq ("mm7",$Hsse2);
+
+ &pxor ($BxC,"mm1"); # de-magic
+ &paddq ($A,&QWP(0,"esi"));
+ &paddq ("mm1",&QWP(8,"esi"));
+ &paddq ($BxC,&QWP(16,"esi"));
+ &paddq ("mm3",&QWP(24,"esi"));
+ &paddq ($E,&QWP(32,"esi"));
+ &paddq ("mm5",&QWP(40,"esi"));
+ &paddq ("mm6",&QWP(48,"esi"));
+ &paddq ("mm7",&QWP(56,"esi"));
+
+ &movq (&QWP(0,"esi"),$A);
+ &movq (&QWP(8,"esi"),"mm1");
+ &movq (&QWP(16,"esi"),$BxC);
+ &movq (&QWP(24,"esi"),"mm3");
+ &movq (&QWP(32,"esi"),$E);
+ &movq (&QWP(40,"esi"),"mm5");
+ &movq (&QWP(48,"esi"),"mm6");
+ &movq (&QWP(56,"esi"),"mm7");
+
+ &cmp ("edi","eax") # are we done yet?
+ &jb (&label("loop_ssse3"));
+
+ &mov ("esp",&DWP(64+12,$frame)); # restore sp
+ &emms ();
+}
+&function_end_A();
+}