aes/asm/aesni-x86[_64].pl update.

This addresses

- request for improvement for faster key setup in RT#3576;
- clearing registers and stack in RT#3554 (this is more of a gesture to
see if there will be some traction from compiler side);
- more commentary around input parameters handling and stack layout
(desired when RT#3553 was reviewed);
- minor size and single block performance optimization (was lying around);

Reviewed-by: Matt Caswell <matt@openssl.org>

diff --git a/crypto/aes/asm/aesni-x86.pl b/crypto/aes/asm/aesni-x86.pl
index 3deb86a..847695f 100644 (file)
--- a/crypto/aes/asm/aesni-x86.pl
+++ b/crypto/aes/asm/aesni-x86.pl
@@ -51,7 +51,7 @@
 # Westmere     3.77/1.37       1.37    1.52    1.27
 # * Bridge     5.07/0.98       0.99    1.09    0.91
 # Haswell      4.44/0.80       0.97    1.03    0.72
 # Westmere     3.77/1.37       1.37    1.52    1.27
 # * Bridge     5.07/0.98       0.99    1.09    0.91
 # Haswell      4.44/0.80       0.97    1.03    0.72

-# Atom         5.77/3.56       3.67    4.03    3.46
+# Silvermont   5.77/3.56       3.67    4.03    3.46

 # Bulldozer    5.80/0.98       1.05    1.24    0.93
 
 $PREFIX="aesni";       # if $PREFIX is set to "AES", the script
 # Bulldozer    5.80/0.98       1.05    1.24    0.93
 
 $PREFIX="aesni";       # if $PREFIX is set to "AES", the script


@@ -65,6 +65,9 @@ require "x86asm.pl";
 
 &asm_init($ARGV[0],$0);
 
 
 &asm_init($ARGV[0],$0);
 

+&external_label("OPENSSL_ia32cap_P");
+&static_label("key_const");
+

 if ($PREFIX eq "aesni")        { $movekey=\&movups; }
 else                   { $movekey=\&movups; }
 
 if ($PREFIX eq "aesni")        { $movekey=\&movups; }
 else                   { $movekey=\&movups; }
 


@@ -181,7 +184,10 @@ sub aesni_generate1        # fully unrolled loop
        {   &aesni_inline_generate1("enc");     }
        else
        {   &call       ("_aesni_encrypt1");    }
        {   &aesni_inline_generate1("enc");     }
        else
        {   &call       ("_aesni_encrypt1");    }

+       &pxor   ($rndkey0,$rndkey0);            # clear register bank
+       &pxor   ($rndkey1,$rndkey1);

        &movups (&QWP(0,"eax"),$inout0);
        &movups (&QWP(0,"eax"),$inout0);

+       &pxor   ($inout0,$inout0);

        &ret    ();
 &function_end_B("${PREFIX}_encrypt");
 
        &ret    ();
 &function_end_B("${PREFIX}_encrypt");
 


@@ -197,7 +203,10 @@ sub aesni_generate1        # fully unrolled loop
        {   &aesni_inline_generate1("dec");     }
        else
        {   &call       ("_aesni_decrypt1");    }
        {   &aesni_inline_generate1("dec");     }
        else
        {   &call       ("_aesni_decrypt1");    }

+       &pxor   ($rndkey0,$rndkey0);            # clear register bank
+       &pxor   ($rndkey1,$rndkey1);

        &movups (&QWP(0,"eax"),$inout0);
        &movups (&QWP(0,"eax"),$inout0);

+       &pxor   ($inout0,$inout0);

        &ret    ();
 &function_end_B("${PREFIX}_decrypt");
 
        &ret    ();
 &function_end_B("${PREFIX}_decrypt");
 


@@ -349,17 +358,15 @@ sub aesni_generate6
        &neg            ($rounds);
        eval"&aes${p}   ($inout2,$rndkey1)";
        &pxor           ($inout5,$rndkey0);
        &neg            ($rounds);
        eval"&aes${p}   ($inout2,$rndkey1)";
        &pxor           ($inout5,$rndkey0);

+       &$movekey       ($rndkey0,&QWP(0,$key,$rounds));

        &add            ($rounds,16);
        &add            ($rounds,16);

-       eval"&aes${p}   ($inout3,$rndkey1)";
-       eval"&aes${p}   ($inout4,$rndkey1)";
-       eval"&aes${p}   ($inout5,$rndkey1)";
-       &$movekey       ($rndkey0,&QWP(-16,$key,$rounds));
-       &jmp            (&label("_aesni_${p}rypt6_enter"));
+       &jmp            (&label("_aesni_${p}rypt6_inner"));

 
     &set_label("${p}6_loop",16);
        eval"&aes${p}   ($inout0,$rndkey1)";
        eval"&aes${p}   ($inout1,$rndkey1)";
        eval"&aes${p}   ($inout2,$rndkey1)";
 
     &set_label("${p}6_loop",16);
        eval"&aes${p}   ($inout0,$rndkey1)";
        eval"&aes${p}   ($inout1,$rndkey1)";
        eval"&aes${p}   ($inout2,$rndkey1)";

+    &set_label("_aesni_${p}rypt6_inner");

        eval"&aes${p}   ($inout3,$rndkey1)";
        eval"&aes${p}   ($inout4,$rndkey1)";
        eval"&aes${p}   ($inout5,$rndkey1)";
        eval"&aes${p}   ($inout3,$rndkey1)";
        eval"&aes${p}   ($inout4,$rndkey1)";
        eval"&aes${p}   ($inout5,$rndkey1)";


@@ -615,6 +622,14 @@ if ($PREFIX eq "aesni") {
        &movups (&QWP(0x30,$out),$inout3);
 
 &set_label("ecb_ret");
        &movups (&QWP(0x30,$out),$inout3);
 
 &set_label("ecb_ret");

+       &pxor   ("xmm0","xmm0");                # clear register bank
+       &pxor   ("xmm1","xmm1");
+       &pxor   ("xmm2","xmm2");
+       &pxor   ("xmm3","xmm3");
+       &pxor   ("xmm4","xmm4");
+       &pxor   ("xmm5","xmm5");
+       &pxor   ("xmm6","xmm6");
+       &pxor   ("xmm7","xmm7");

 &function_end("aesni_ecb_encrypt");
 \f
 ######################################################################
 &function_end("aesni_ecb_encrypt");
 \f
 ######################################################################


@@ -704,6 +719,15 @@ if ($PREFIX eq "aesni") {
        &mov    ("esp",&DWP(48,"esp"));
        &mov    ($out,&wparam(5));
        &movups (&QWP(0,$out),$cmac);
        &mov    ("esp",&DWP(48,"esp"));
        &mov    ($out,&wparam(5));
        &movups (&QWP(0,$out),$cmac);

+
+       &pxor   ("xmm0","xmm0");                # clear register bank
+       &pxor   ("xmm1","xmm1");
+       &pxor   ("xmm2","xmm2");
+       &pxor   ("xmm3","xmm3");
+       &pxor   ("xmm4","xmm4");
+       &pxor   ("xmm5","xmm5");
+       &pxor   ("xmm6","xmm6");
+       &pxor   ("xmm7","xmm7");

 &function_end("aesni_ccm64_encrypt_blocks");
 
 &function_begin("aesni_ccm64_decrypt_blocks");
 &function_end("aesni_ccm64_encrypt_blocks");
 
 &function_begin("aesni_ccm64_decrypt_blocks");


@@ -804,6 +828,15 @@ if ($PREFIX eq "aesni") {
        &mov    ("esp",&DWP(48,"esp"));
        &mov    ($out,&wparam(5));
        &movups (&QWP(0,$out),$cmac);
        &mov    ("esp",&DWP(48,"esp"));
        &mov    ($out,&wparam(5));
        &movups (&QWP(0,$out),$cmac);

+
+       &pxor   ("xmm0","xmm0");                # clear register bank
+       &pxor   ("xmm1","xmm1");
+       &pxor   ("xmm2","xmm2");
+       &pxor   ("xmm3","xmm3");
+       &pxor   ("xmm4","xmm4");
+       &pxor   ("xmm5","xmm5");
+       &pxor   ("xmm6","xmm6");
+       &pxor   ("xmm7","xmm7");

 &function_end("aesni_ccm64_decrypt_blocks");
 }
 \f
 &function_end("aesni_ccm64_decrypt_blocks");
 }
 \f


@@ -1053,6 +1086,17 @@ if ($PREFIX eq "aesni") {
        &movups (&QWP(0x30,$out),$inout3);
 
 &set_label("ctr32_ret");
        &movups (&QWP(0x30,$out),$inout3);
 
 &set_label("ctr32_ret");

+       &pxor   ("xmm0","xmm0");                # clear register bank
+       &pxor   ("xmm1","xmm1");
+       &pxor   ("xmm2","xmm2");
+       &pxor   ("xmm3","xmm3");
+       &pxor   ("xmm4","xmm4");
+       &movdqa (&QWP(32,"esp"),"xmm0");        # clear stack
+       &pxor   ("xmm5","xmm5");
+       &movdqa (&QWP(48,"esp"),"xmm0");
+       &pxor   ("xmm6","xmm6");
+       &movdqa (&QWP(64,"esp"),"xmm0");
+       &pxor   ("xmm7","xmm7");

        &mov    ("esp",&DWP(80,"esp"));
 &function_end("aesni_ctr32_encrypt_blocks");
 \f
        &mov    ("esp",&DWP(80,"esp"));
 &function_end("aesni_ctr32_encrypt_blocks");
 \f


@@ -1394,6 +1438,20 @@ if ($PREFIX eq "aesni") {
        &movups (&QWP(-16,$out),$inout0);       # write output
 
 &set_label("xts_enc_ret");
        &movups (&QWP(-16,$out),$inout0);       # write output
 
 &set_label("xts_enc_ret");

+       &pxor   ("xmm0","xmm0");                # clear register bank
+       &pxor   ("xmm1","xmm1");
+       &pxor   ("xmm2","xmm2");
+       &movdqa (&QWP(16*0,"esp"),"xmm0");      # clear stack
+       &pxor   ("xmm3","xmm3");
+       &movdqa (&QWP(16*1,"esp"),"xmm0");
+       &pxor   ("xmm4","xmm4");
+       &movdqa (&QWP(16*2,"esp"),"xmm0");
+       &pxor   ("xmm5","xmm5");
+       &movdqa (&QWP(16*3,"esp"),"xmm0");
+       &pxor   ("xmm6","xmm6");
+       &movdqa (&QWP(16*4,"esp"),"xmm0");
+       &pxor   ("xmm7","xmm7");
+       &movdqa (&QWP(16*5,"esp"),"xmm0");

        &mov    ("esp",&DWP(16*7+4,"esp"));     # restore %esp
 &function_end("aesni_xts_encrypt");
 
        &mov    ("esp",&DWP(16*7+4,"esp"));     # restore %esp
 &function_end("aesni_xts_encrypt");
 


@@ -1756,6 +1814,20 @@ if ($PREFIX eq "aesni") {
        &movups (&QWP(0,$out),$inout0);         # write output
 
 &set_label("xts_dec_ret");
        &movups (&QWP(0,$out),$inout0);         # write output
 
 &set_label("xts_dec_ret");

+       &pxor   ("xmm0","xmm0");                # clear register bank
+       &pxor   ("xmm1","xmm1");
+       &pxor   ("xmm2","xmm2");
+       &movdqa (&QWP(16*0,"esp"),"xmm0");      # clear stack
+       &pxor   ("xmm3","xmm3");
+       &movdqa (&QWP(16*1,"esp"),"xmm0");
+       &pxor   ("xmm4","xmm4");
+       &movdqa (&QWP(16*2,"esp"),"xmm0");
+       &pxor   ("xmm5","xmm5");
+       &movdqa (&QWP(16*3,"esp"),"xmm0");
+       &pxor   ("xmm6","xmm6");
+       &movdqa (&QWP(16*4,"esp"),"xmm0");
+       &pxor   ("xmm7","xmm7");
+       &movdqa (&QWP(16*5,"esp"),"xmm0");

        &mov    ("esp",&DWP(16*7+4,"esp"));     # restore %esp
 &function_end("aesni_xts_decrypt");
 }
        &mov    ("esp",&DWP(16*7+4,"esp"));     # restore %esp
 &function_end("aesni_xts_decrypt");
 }


@@ -1808,6 +1880,7 @@ if ($PREFIX eq "aesni") {
        &add    ($len,16);
        &jnz    (&label("cbc_enc_tail"));
        &movaps ($ivec,$inout0);
        &add    ($len,16);
        &jnz    (&label("cbc_enc_tail"));
        &movaps ($ivec,$inout0);

+       &pxor   ($inout0,$inout0);

        &jmp    (&label("cbc_ret"));
 
 &set_label("cbc_enc_tail");
        &jmp    (&label("cbc_ret"));
 
 &set_label("cbc_enc_tail");


@@ -1871,7 +1944,7 @@ if ($PREFIX eq "aesni") {
        &movaps ($inout0,$inout5);
        &movaps ($ivec,$rndkey0);
        &add    ($len,0x50);
        &movaps ($inout0,$inout5);
        &movaps ($ivec,$rndkey0);
        &add    ($len,0x50);

-       &jle    (&label("cbc_dec_tail_collected"));
+       &jle    (&label("cbc_dec_clear_tail_collected"));

        &movups (&QWP(0,$out),$inout0);
        &lea    ($out,&DWP(0x10,$out));
 &set_label("cbc_dec_tail");
        &movups (&QWP(0,$out),$inout0);
        &lea    ($out,&DWP(0x10,$out));
 &set_label("cbc_dec_tail");


@@ -1910,10 +1983,14 @@ if ($PREFIX eq "aesni") {
        &xorps  ($inout4,$rndkey0);
        &movups (&QWP(0,$out),$inout0);
        &movups (&QWP(0x10,$out),$inout1);
        &xorps  ($inout4,$rndkey0);
        &movups (&QWP(0,$out),$inout0);
        &movups (&QWP(0x10,$out),$inout1);

+       &pxor   ($inout1,$inout1);

        &movups (&QWP(0x20,$out),$inout2);
        &movups (&QWP(0x20,$out),$inout2);

+       &pxor   ($inout2,$inout2);

        &movups (&QWP(0x30,$out),$inout3);
        &movups (&QWP(0x30,$out),$inout3);

+       &pxor   ($inout3,$inout3);

        &lea    ($out,&DWP(0x40,$out));
        &movaps ($inout0,$inout4);
        &lea    ($out,&DWP(0x40,$out));
        &movaps ($inout0,$inout4);

+       &pxor   ($inout4,$inout4);

        &sub    ($len,0x50);
        &jmp    (&label("cbc_dec_tail_collected"));
 
        &sub    ($len,0x50);
        &jmp    (&label("cbc_dec_tail_collected"));
 


@@ -1933,6 +2010,7 @@ if ($PREFIX eq "aesni") {
        &xorps  ($inout1,$in0);
        &movups (&QWP(0,$out),$inout0);
        &movaps ($inout0,$inout1);
        &xorps  ($inout1,$in0);
        &movups (&QWP(0,$out),$inout0);
        &movaps ($inout0,$inout1);

+       &pxor   ($inout1,$inout1);

        &lea    ($out,&DWP(0x10,$out));
        &movaps ($ivec,$in1);
        &sub    ($len,0x20);
        &lea    ($out,&DWP(0x10,$out));
        &movaps ($ivec,$in1);
        &sub    ($len,0x20);


@@ -1945,7 +2023,9 @@ if ($PREFIX eq "aesni") {
        &xorps  ($inout2,$in1);
        &movups (&QWP(0,$out),$inout0);
        &movaps ($inout0,$inout2);
        &xorps  ($inout2,$in1);
        &movups (&QWP(0,$out),$inout0);
        &movaps ($inout0,$inout2);

+       &pxor   ($inout2,$inout2);

        &movups (&QWP(0x10,$out),$inout1);
        &movups (&QWP(0x10,$out),$inout1);

+       &pxor   ($inout1,$inout1);

        &lea    ($out,&DWP(0x20,$out));
        &movups ($ivec,&QWP(0x20,$inp));
        &sub    ($len,0x30);
        &lea    ($out,&DWP(0x20,$out));
        &movups ($ivec,&QWP(0x20,$inp));
        &sub    ($len,0x30);


@@ -1961,29 +2041,44 @@ if ($PREFIX eq "aesni") {
        &movups (&QWP(0,$out),$inout0);
        &xorps  ($inout2,$rndkey1);
        &movups (&QWP(0x10,$out),$inout1);
        &movups (&QWP(0,$out),$inout0);
        &xorps  ($inout2,$rndkey1);
        &movups (&QWP(0x10,$out),$inout1);

+       &pxor   ($inout1,$inout1);

        &xorps  ($inout3,$rndkey0);
        &movups (&QWP(0x20,$out),$inout2);
        &xorps  ($inout3,$rndkey0);
        &movups (&QWP(0x20,$out),$inout2);

+       &pxor   ($inout2,$inout2);

        &lea    ($out,&DWP(0x30,$out));
        &movaps ($inout0,$inout3);
        &lea    ($out,&DWP(0x30,$out));
        &movaps ($inout0,$inout3);

+       &pxor   ($inout3,$inout3);

        &sub    ($len,0x40);
        &sub    ($len,0x40);

+       &jmp    (&label("cbc_dec_tail_collected"));

 
 

+&set_label("cbc_dec_clear_tail_collected",16);
+       &pxor   ($inout1,$inout1);
+       &pxor   ($inout2,$inout2);
+       &pxor   ($inout3,$inout3);
+       &pxor   ($inout4,$inout4);

 &set_label("cbc_dec_tail_collected");
        &and    ($len,15);
        &jnz    (&label("cbc_dec_tail_partial"));
        &movups (&QWP(0,$out),$inout0);
 &set_label("cbc_dec_tail_collected");
        &and    ($len,15);
        &jnz    (&label("cbc_dec_tail_partial"));
        &movups (&QWP(0,$out),$inout0);

+       &pxor   ($rndkey0,$rndkey0);

        &jmp    (&label("cbc_ret"));
 
 &set_label("cbc_dec_tail_partial",16);
        &movaps (&QWP(0,"esp"),$inout0);
        &jmp    (&label("cbc_ret"));
 
 &set_label("cbc_dec_tail_partial",16);
        &movaps (&QWP(0,"esp"),$inout0);

+       &pxor   ($rndkey0,$rndkey0);

        &mov    ("ecx",16);
        &mov    ($inp,"esp");
        &sub    ("ecx",$len);
        &data_word(0xA4F3F689);         # rep movsb
        &mov    ("ecx",16);
        &mov    ($inp,"esp");
        &sub    ("ecx",$len);
        &data_word(0xA4F3F689);         # rep movsb

+       &movdqa (&QWP(0,"esp"),$inout0);

 
 &set_label("cbc_ret");
        &mov    ("esp",&DWP(16,"esp")); # pull original %esp
        &mov    ($key_,&wparam(4));
 
 &set_label("cbc_ret");
        &mov    ("esp",&DWP(16,"esp")); # pull original %esp
        &mov    ($key_,&wparam(4));

+       &pxor   ($inout0,$inout0);
+       &pxor   ($rndkey1,$rndkey1);

        &movups (&QWP(0,$key_),$ivec);  # output IV
        &movups (&QWP(0,$key_),$ivec);  # output IV

+       &pxor   ($ivec,$ivec);

 &set_label("cbc_abort");
 &function_end("${PREFIX}_cbc_encrypt");
 \f
 &set_label("cbc_abort");
 &function_end("${PREFIX}_cbc_encrypt");
 \f


@@ -2000,14 +2095,24 @@ if ($PREFIX eq "aesni") {
 #      $round  rounds
 
 &function_begin_B("_aesni_set_encrypt_key");
 #      $round  rounds
 
 &function_begin_B("_aesni_set_encrypt_key");

+       &push   ("ebp");
+       &push   ("ebx");

        &test   ("eax","eax");
        &jz     (&label("bad_pointer"));
        &test   ($key,$key);
        &jz     (&label("bad_pointer"));
 
        &test   ("eax","eax");
        &jz     (&label("bad_pointer"));
        &test   ($key,$key);
        &jz     (&label("bad_pointer"));
 

+       &call   (&label("pic"));
+&set_label("pic");
+       &blindpop("ebx");
+       &lea    ("ebx",&DWP(&label("key_const")."-".&label("pic"),"ebx"));
+
+       &picmeup("ebp","OPENSSL_ia32cap_P","ebx",&label("key_const"));

        &movups ("xmm0",&QWP(0,"eax")); # pull first 128 bits of *userKey
        &xorps  ("xmm4","xmm4");        # low dword of xmm4 is assumed 0
        &movups ("xmm0",&QWP(0,"eax")); # pull first 128 bits of *userKey
        &xorps  ("xmm4","xmm4");        # low dword of xmm4 is assumed 0

+       &mov    ("ebp",&DWP(4,"ebp"));

        &lea    ($key,&DWP(16,$key));
        &lea    ($key,&DWP(16,$key));

+       &and    ("ebp",1<<28|1<<11);    # AVX and XOP bits

        &cmp    ($rounds,256);
        &je     (&label("14rounds"));
        &cmp    ($rounds,192);
        &cmp    ($rounds,256);
        &je     (&label("14rounds"));
        &cmp    ($rounds,192);


@@ -2016,6 +2121,9 @@ if ($PREFIX eq "aesni") {
        &jne    (&label("bad_keybits"));
 
 &set_label("10rounds",16);
        &jne    (&label("bad_keybits"));
 
 &set_label("10rounds",16);

+       &cmp            ("ebp",1<<28);
+       &je             (&label("10rounds_alt"));
+

        &mov            ($rounds,9);
        &$movekey       (&QWP(-16,$key),"xmm0");        # round 0
        &aeskeygenassist("xmm1","xmm0",0x01);           # round 1
        &mov            ($rounds,9);
        &$movekey       (&QWP(-16,$key),"xmm0");        # round 0
        &aeskeygenassist("xmm1","xmm0",0x01);           # round 1


@@ -2040,8 +2148,8 @@ if ($PREFIX eq "aesni") {
        &call           (&label("key_128"));
        &$movekey       (&QWP(0,$key),"xmm0");
        &mov            (&DWP(80,$key),$rounds);
        &call           (&label("key_128"));
        &$movekey       (&QWP(0,$key),"xmm0");
        &mov            (&DWP(80,$key),$rounds);

-       &xor            ("eax","eax");
-       &ret();
+
+       &jmp    (&label("good_key"));

 
 &set_label("key_128",16);
        &$movekey       (&QWP(0,$key),"xmm0");
 
 &set_label("key_128",16);
        &$movekey       (&QWP(0,$key),"xmm0");


@@ -2055,8 +2163,76 @@ if ($PREFIX eq "aesni") {
        &xorps          ("xmm0","xmm1");
        &ret();
 
        &xorps          ("xmm0","xmm1");
        &ret();
 

+&set_label("10rounds_alt",16);
+       &movdqa         ("xmm5",&QWP(0x00,"ebx"));
+       &mov            ($rounds,8);
+       &movdqa         ("xmm4",&QWP(0x20,"ebx"));
+       &movdqa         ("xmm2","xmm0");
+       &movdqu         (&DWP(-16,$key),"xmm0");
+
+&set_label("loop_key128");
+       &pshufb         ("xmm0","xmm5");
+       &aesenclast     ("xmm0","xmm4");
+       &pslld          ("xmm4",1);
+       &lea            ($key,&DWP(16,$key));
+
+       &movdqa         ("xmm3","xmm2");
+       &pslldq         ("xmm2",4);
+       &pxor           ("xmm3","xmm2");
+       &pslldq         ("xmm2",4);
+       &pxor           ("xmm3","xmm2");
+       &pslldq         ("xmm2",4);
+       &pxor           ("xmm2","xmm3");
+
+       &pxor           ("xmm0","xmm2");
+       &movdqu         (&QWP(-16,$key),"xmm0");
+       &movdqa         ("xmm2","xmm0");
+
+       &dec            ($rounds);
+       &jnz            (&label("loop_key128"));
+
+       &movdqa         ("xmm4",&QWP(0x30,"ebx"));
+
+       &pshufb         ("xmm0","xmm5");
+       &aesenclast     ("xmm0","xmm4");
+       &pslld          ("xmm4",1);
+
+       &movdqa         ("xmm3","xmm2");
+       &pslldq         ("xmm2",4);
+       &pxor           ("xmm3","xmm2");
+       &pslldq         ("xmm2",4);
+       &pxor           ("xmm3","xmm2");
+       &pslldq         ("xmm2",4);
+       &pxor           ("xmm2","xmm3");
+
+       &pxor           ("xmm0","xmm2");
+       &movdqu         (&QWP(0,$key),"xmm0");
+
+       &movdqa         ("xmm2","xmm0");
+       &pshufb         ("xmm0","xmm5");
+       &aesenclast     ("xmm0","xmm4");
+
+       &movdqa         ("xmm3","xmm2");
+       &pslldq         ("xmm2",4);
+       &pxor           ("xmm3","xmm2");
+       &pslldq         ("xmm2",4);
+       &pxor           ("xmm3","xmm2");
+       &pslldq         ("xmm2",4);
+       &pxor           ("xmm2","xmm3");
+
+       &pxor           ("xmm0","xmm2");
+       &movdqu         (&QWP(16,$key),"xmm0");
+
+       &mov            ($rounds,9);
+       &mov            (&DWP(96,$key),$rounds);
+
+       &jmp    (&label("good_key"));
+

 &set_label("12rounds",16);
        &movq           ("xmm2",&QWP(16,"eax"));        # remaining 1/3 of *userKey
 &set_label("12rounds",16);
        &movq           ("xmm2",&QWP(16,"eax"));        # remaining 1/3 of *userKey

+       &cmp            ("ebp",1<<28);
+       &je             (&label("12rounds_alt"));
+

        &mov            ($rounds,11);
        &$movekey       (&QWP(-16,$key),"xmm0");        # round 0
        &aeskeygenassist("xmm1","xmm2",0x01);           # round 1,2
        &mov            ($rounds,11);
        &$movekey       (&QWP(-16,$key),"xmm0");        # round 0
        &aeskeygenassist("xmm1","xmm2",0x01);           # round 1,2


@@ -2077,8 +2253,8 @@ if ($PREFIX eq "aesni") {
        &call           (&label("key_192b"));
        &$movekey       (&QWP(0,$key),"xmm0");
        &mov            (&DWP(48,$key),$rounds);
        &call           (&label("key_192b"));
        &$movekey       (&QWP(0,$key),"xmm0");
        &mov            (&DWP(48,$key),$rounds);

-       &xor            ("eax","eax");
-       &ret();
+
+       &jmp    (&label("good_key"));

 
 &set_label("key_192a",16);
        &$movekey       (&QWP(0,$key),"xmm0");
 
 &set_label("key_192a",16);
        &$movekey       (&QWP(0,$key),"xmm0");


@@ -2108,10 +2284,52 @@ if ($PREFIX eq "aesni") {
        &lea            ($key,&DWP(32,$key));
        &jmp            (&label("key_192b_warm"));
 
        &lea            ($key,&DWP(32,$key));
        &jmp            (&label("key_192b_warm"));
 

+&set_label("12rounds_alt",16);
+       &movdqa         ("xmm5",&QWP(0x10,"ebx"));
+       &movdqa         ("xmm4",&QWP(0x20,"ebx"));
+       &mov            ($rounds,8);
+       &movdqu         (&QWP(-16,$key),"xmm0");
+
+&set_label("loop_key192");
+       &movq           (&QWP(0,$key),"xmm2");
+       &movdqa         ("xmm1","xmm2");
+       &pshufb         ("xmm2","xmm5");
+       &aesenclast     ("xmm2","xmm4");
+       &pslld          ("xmm4",1);
+       &lea            ($key,&DWP(24,$key));
+
+       &movdqa         ("xmm3","xmm0");
+       &pslldq         ("xmm0",4);
+       &pxor           ("xmm3","xmm0");
+       &pslldq         ("xmm0",4);
+       &pxor           ("xmm3","xmm0");
+       &pslldq         ("xmm0",4);
+       &pxor           ("xmm0","xmm3");
+
+       &pshufd         ("xmm3","xmm0",0xff);
+       &pxor           ("xmm3","xmm1");
+       &pslldq         ("xmm1",4);
+       &pxor           ("xmm3","xmm1");
+
+       &pxor           ("xmm0","xmm2");
+       &pxor           ("xmm2","xmm3");
+       &movdqu         (&QWP(-16,$key),"xmm0");
+
+       &dec            ($rounds);
+       &jnz            (&label("loop_key192"));
+
+       &mov    ($rounds,11);
+       &mov    (&DWP(32,$key),$rounds);
+
+       &jmp    (&label("good_key"));
+

 &set_label("14rounds",16);
        &movups         ("xmm2",&QWP(16,"eax"));        # remaining half of *userKey
 &set_label("14rounds",16);
        &movups         ("xmm2",&QWP(16,"eax"));        # remaining half of *userKey

-       &mov            ($rounds,13);

        &lea            ($key,&DWP(16,$key));
        &lea            ($key,&DWP(16,$key));

+       &cmp            ("ebp",1<<28);
+       &je             (&label("14rounds_alt"));
+
+       &mov            ($rounds,13);

        &$movekey       (&QWP(-32,$key),"xmm0");        # round 0
        &$movekey       (&QWP(-16,$key),"xmm2");        # round 1
        &aeskeygenassist("xmm1","xmm2",0x01);           # round 2
        &$movekey       (&QWP(-32,$key),"xmm0");        # round 0
        &$movekey       (&QWP(-16,$key),"xmm2");        # round 1
        &aeskeygenassist("xmm1","xmm2",0x01);           # round 2


@@ -2143,7 +2361,8 @@ if ($PREFIX eq "aesni") {
        &$movekey       (&QWP(0,$key),"xmm0");
        &mov            (&DWP(16,$key),$rounds);
        &xor            ("eax","eax");
        &$movekey       (&QWP(0,$key),"xmm0");
        &mov            (&DWP(16,$key),$rounds);
        &xor            ("eax","eax");

-       &ret();
+
+       &jmp    (&label("good_key"));

 
 &set_label("key_256a",16);
        &$movekey       (&QWP(0,$key),"xmm2");
 
 &set_label("key_256a",16);
        &$movekey       (&QWP(0,$key),"xmm2");


@@ -2169,11 +2388,77 @@ if ($PREFIX eq "aesni") {
        &xorps          ("xmm2","xmm1");
        &ret();
 
        &xorps          ("xmm2","xmm1");
        &ret();
 

+&set_label("14rounds_alt",16);
+       &movdqa         ("xmm5",&QWP(0x00,"ebx"));
+       &movdqa         ("xmm4",&QWP(0x20,"ebx"));
+       &mov            ($rounds,7);
+       &movdqu         (&QWP(-32,$key),"xmm0");
+       &movdqa         ("xmm1","xmm2");
+       &movdqu         (&QWP(-16,$key),"xmm2");
+
+&set_label("loop_key256");
+       &pshufb         ("xmm2","xmm5");
+       &aesenclast     ("xmm2","xmm4");
+
+       &movdqa         ("xmm3","xmm0");
+       &pslldq         ("xmm0",4);
+       &pxor           ("xmm3","xmm0");
+       &pslldq         ("xmm0",4);
+       &pxor           ("xmm3","xmm0");
+       &pslldq         ("xmm0",4);
+       &pxor           ("xmm0","xmm3");
+       &pslld          ("xmm4",1);
+
+       &pxor           ("xmm0","xmm2");
+       &movdqu         (&QWP(0,$key),"xmm0");
+
+       &dec            ($rounds);
+       &jz             (&label("done_key256"));
+
+       &pshufd         ("xmm2","xmm0",0xff);
+       &pxor           ("xmm3","xmm3");
+       &aesenclast     ("xmm2","xmm3");
+
+       &movdqa         ("xmm3","xmm1")
+       &pslldq         ("xmm1",4);
+       &pxor           ("xmm3","xmm1");
+       &pslldq         ("xmm1",4);
+       &pxor           ("xmm3","xmm1");
+       &pslldq         ("xmm1",4);
+       &pxor           ("xmm1","xmm3");
+
+       &pxor           ("xmm2","xmm1");
+       &movdqu         (&QWP(16,$key),"xmm2");
+       &lea            ($key,&DWP(32,$key));
+       &movdqa         ("xmm1","xmm2");
+       &jmp            (&label("loop_key256"));
+
+&set_label("done_key256");
+       &mov            ($rounds,13);
+       &mov            (&DWP(16,$key),$rounds);
+
+&set_label("good_key");
+       &pxor   ("xmm0","xmm0");
+       &pxor   ("xmm1","xmm1");
+       &pxor   ("xmm2","xmm2");
+       &pxor   ("xmm3","xmm3");
+       &pxor   ("xmm4","xmm4");
+       &pxor   ("xmm5","xmm5");
+       &xor    ("eax","eax");
+       &pop    ("ebx");
+       &pop    ("ebp");
+       &ret    ();
+

 &set_label("bad_pointer",4);
        &mov    ("eax",-1);
 &set_label("bad_pointer",4);
        &mov    ("eax",-1);

+       &pop    ("ebx");
+       &pop    ("ebp");

        &ret    ();
 &set_label("bad_keybits",4);
        &ret    ();
 &set_label("bad_keybits",4);

+       &pxor   ("xmm0","xmm0");

        &mov    ("eax",-2);
        &mov    ("eax",-2);

+       &pop    ("ebx");
+       &pop    ("ebp");

        &ret    ();
 &function_end_B("_aesni_set_encrypt_key");
 
        &ret    ();
 &function_end_B("_aesni_set_encrypt_key");
 


@@ -2223,10 +2508,18 @@ if ($PREFIX eq "aesni") {
        &aesimc         ("xmm0","xmm0");
        &$movekey       (&QWP(0,$key),"xmm0");
 
        &aesimc         ("xmm0","xmm0");
        &$movekey       (&QWP(0,$key),"xmm0");
 

+       &pxor           ("xmm0","xmm0");
+       &pxor           ("xmm1","xmm1");

        &xor            ("eax","eax");          # return success
 &set_label("dec_key_ret");
        &ret    ();
 &function_end_B("${PREFIX}_set_decrypt_key");
        &xor            ("eax","eax");          # return success
 &set_label("dec_key_ret");
        &ret    ();
 &function_end_B("${PREFIX}_set_decrypt_key");

+
+&set_label("key_const",64);
+&data_word(0x0c0f0e0d,0x0c0f0e0d,0x0c0f0e0d,0x0c0f0e0d);
+&data_word(0x04070605,0x04070605,0x04070605,0x04070605);
+&data_word(1,1,1,1);
+&data_word(0x1b,0x1b,0x1b,0x1b);

 &asciz("AES for Intel AES-NI, CRYPTOGAMS by <appro\@openssl.org>");
 
 &asm_finish();
 &asciz("AES for Intel AES-NI, CRYPTOGAMS by <appro\@openssl.org>");
 
 &asm_finish();

diff --git a/crypto/aes/asm/aesni-x86_64.pl b/crypto/aes/asm/aesni-x86_64.pl
index 5f61746..25ca574 100644 (file)
--- a/crypto/aes/asm/aesni-x86_64.pl
+++ b/crypto/aes/asm/aesni-x86_64.pl
@@ -165,11 +165,11 @@
 # Westmere     3.77/1.25       1.25    1.25    1.26
 # * Bridge     5.07/0.74       0.75    0.90    0.85
 # Haswell      4.44/0.63       0.63    0.73    0.63
 # Westmere     3.77/1.25       1.25    1.25    1.26
 # * Bridge     5.07/0.74       0.75    0.90    0.85
 # Haswell      4.44/0.63       0.63    0.73    0.63

-# Atom         5.75/3.54       3.56    4.12    3.87(*)
+# Silvermont   5.75/3.54       3.56    4.12    3.87(*)

 # Bulldozer    5.77/0.70       0.72    0.90    0.70
 #
 # Bulldozer    5.77/0.70       0.72    0.90    0.70
 #

-# (*)  Atom ECB result is suboptimal because of penalties incurred
-#      by operations on %xmm8-15. As ECB is not considered
+# (*)  Atom Silvermont ECB result is suboptimal because of penalties
+#      incurred by operations on %xmm8-15. As ECB is not considered

 #      critical, nothing was done to mitigate the problem.
 
 $PREFIX="aesni";       # if $PREFIX is set to "AES", the script
 #      critical, nothing was done to mitigate the problem.
 
 $PREFIX="aesni";       # if $PREFIX is set to "AES", the script


@@ -263,7 +263,10 @@ ${PREFIX}_encrypt:
 ___
        &aesni_generate1("enc",$key,$rounds);
 $code.=<<___;
 ___
        &aesni_generate1("enc",$key,$rounds);
 $code.=<<___;

+        pxor   $rndkey0,$rndkey0       # clear register bank
+        pxor   $rndkey1,$rndkey1

        movups  $inout0,($out)          # output
        movups  $inout0,($out)          # output

+        pxor   $inout0,$inout0

        ret
 .size  ${PREFIX}_encrypt,.-${PREFIX}_encrypt
 
        ret
 .size  ${PREFIX}_encrypt,.-${PREFIX}_encrypt
 


@@ -276,7 +279,10 @@ ${PREFIX}_decrypt:
 ___
        &aesni_generate1("dec",$key,$rounds);
 $code.=<<___;
 ___
        &aesni_generate1("dec",$key,$rounds);
 $code.=<<___;

+        pxor   $rndkey0,$rndkey0       # clear register bank
+        pxor   $rndkey1,$rndkey1

        movups  $inout0,($out)          # output
        movups  $inout0,($out)          # output

+        pxor   $inout0,$inout0

        ret
 .size  ${PREFIX}_decrypt, .-${PREFIX}_decrypt
 ___
        ret
 .size  ${PREFIX}_decrypt, .-${PREFIX}_decrypt
 ___


@@ -445,21 +451,18 @@ _aesni_${dir}rypt6:
        pxor            $rndkey0,$inout4
        aes${dir}       $rndkey1,$inout2
        pxor            $rndkey0,$inout5
        pxor            $rndkey0,$inout4
        aes${dir}       $rndkey1,$inout2
        pxor            $rndkey0,$inout5

+       $movkey         ($key,%rax),$rndkey0

        add             \$16,%rax
        add             \$16,%rax

-       aes${dir}       $rndkey1,$inout3
-       aes${dir}       $rndkey1,$inout4
-       aes${dir}       $rndkey1,$inout5
-       $movkey         -16($key,%rax),$rndkey0

        jmp             .L${dir}_loop6_enter
 .align 16
 .L${dir}_loop6:
        aes${dir}       $rndkey1,$inout0
        aes${dir}       $rndkey1,$inout1
        aes${dir}       $rndkey1,$inout2
        jmp             .L${dir}_loop6_enter
 .align 16
 .L${dir}_loop6:
        aes${dir}       $rndkey1,$inout0
        aes${dir}       $rndkey1,$inout1
        aes${dir}       $rndkey1,$inout2

+.L${dir}_loop6_enter:

        aes${dir}       $rndkey1,$inout3
        aes${dir}       $rndkey1,$inout4
        aes${dir}       $rndkey1,$inout5
        aes${dir}       $rndkey1,$inout3
        aes${dir}       $rndkey1,$inout4
        aes${dir}       $rndkey1,$inout5

-.L${dir}_loop6_enter:

        $movkey         ($key,%rax),$rndkey1
        add             \$32,%rax
        aes${dir}       $rndkey0,$inout0
        $movkey         ($key,%rax),$rndkey1
        add             \$32,%rax
        aes${dir}       $rndkey0,$inout0


@@ -506,23 +509,18 @@ _aesni_${dir}rypt8:
        lea             32($key,$rounds),$key
        neg             %rax                    # $rounds
        aes${dir}       $rndkey1,$inout0
        lea             32($key,$rounds),$key
        neg             %rax                    # $rounds
        aes${dir}       $rndkey1,$inout0

-       add             \$16,%rax

        pxor            $rndkey0,$inout5
        pxor            $rndkey0,$inout5

-       aes${dir}       $rndkey1,$inout1

        pxor            $rndkey0,$inout6
        pxor            $rndkey0,$inout6

+       aes${dir}       $rndkey1,$inout1

        pxor            $rndkey0,$inout7
        pxor            $rndkey0,$inout7

-       aes${dir}       $rndkey1,$inout2
-       aes${dir}       $rndkey1,$inout3
-       aes${dir}       $rndkey1,$inout4
-       aes${dir}       $rndkey1,$inout5
-       aes${dir}       $rndkey1,$inout6
-       aes${dir}       $rndkey1,$inout7
-       $movkey         -16($key,%rax),$rndkey0
-       jmp             .L${dir}_loop8_enter
+       $movkey         ($key,%rax),$rndkey0
+       add             \$16,%rax
+       jmp             .L${dir}_loop8_inner

 .align 16
 .L${dir}_loop8:
        aes${dir}       $rndkey1,$inout0
        aes${dir}       $rndkey1,$inout1
 .align 16
 .L${dir}_loop8:
        aes${dir}       $rndkey1,$inout0
        aes${dir}       $rndkey1,$inout1

+.L${dir}_loop8_inner:

        aes${dir}       $rndkey1,$inout2
        aes${dir}       $rndkey1,$inout3
        aes${dir}       $rndkey1,$inout4
        aes${dir}       $rndkey1,$inout2
        aes${dir}       $rndkey1,$inout3
        aes${dir}       $rndkey1,$inout4


@@ -587,15 +585,15 @@ aesni_ecb_encrypt:
 ___
 $code.=<<___ if ($win64);
        lea     -0x58(%rsp),%rsp
 ___
 $code.=<<___ if ($win64);
        lea     -0x58(%rsp),%rsp

-       movaps  %xmm6,(%rsp)
+       movaps  %xmm6,(%rsp)            # offload $inout4..7

        movaps  %xmm7,0x10(%rsp)
        movaps  %xmm8,0x20(%rsp)
        movaps  %xmm9,0x30(%rsp)
 .Lecb_enc_body:
 ___
 $code.=<<___;
        movaps  %xmm7,0x10(%rsp)
        movaps  %xmm8,0x20(%rsp)
        movaps  %xmm9,0x30(%rsp)
 .Lecb_enc_body:
 ___
 $code.=<<___;

-       and     \$-16,$len
-       jz      .Lecb_ret
+       and     \$-16,$len              # if ($len<16)
+       jz      .Lecb_ret               # return

 
        mov     240($key),$rounds       # key->rounds
        $movkey ($key),$rndkey0
 
        mov     240($key),$rounds       # key->rounds
        $movkey ($key),$rndkey0


@@ -604,10 +602,10 @@ $code.=<<___;
        test    %r8d,%r8d               # 5th argument
        jz      .Lecb_decrypt
 #--------------------------- ECB ENCRYPT ------------------------------#
        test    %r8d,%r8d               # 5th argument
        jz      .Lecb_decrypt
 #--------------------------- ECB ENCRYPT ------------------------------#

-       cmp     \$0x80,$len
-       jb      .Lecb_enc_tail
+       cmp     \$0x80,$len             # if ($len<8*16)
+       jb      .Lecb_enc_tail          # short input

 
 

-       movdqu  ($inp),$inout0
+       movdqu  ($inp),$inout0          # load 8 input blocks

        movdqu  0x10($inp),$inout1
        movdqu  0x20($inp),$inout2
        movdqu  0x30($inp),$inout3
        movdqu  0x10($inp),$inout1
        movdqu  0x20($inp),$inout2
        movdqu  0x30($inp),$inout3


@@ -615,14 +613,14 @@ $code.=<<___;
        movdqu  0x50($inp),$inout5
        movdqu  0x60($inp),$inout6
        movdqu  0x70($inp),$inout7
        movdqu  0x50($inp),$inout5
        movdqu  0x60($inp),$inout6
        movdqu  0x70($inp),$inout7

-       lea     0x80($inp),$inp
-       sub     \$0x80,$len
+       lea     0x80($inp),$inp         # $inp+=8*16
+       sub     \$0x80,$len             # $len-=8*16 (can be zero)

        jmp     .Lecb_enc_loop8_enter
 .align 16
 .Lecb_enc_loop8:
        jmp     .Lecb_enc_loop8_enter
 .align 16
 .Lecb_enc_loop8:

-       movups  $inout0,($out)
+       movups  $inout0,($out)          # store 8 output blocks

        mov     $key_,$key              # restore $key
        mov     $key_,$key              # restore $key

-       movdqu  ($inp),$inout0
+       movdqu  ($inp),$inout0          # load 8 input blocks

        mov     $rnds_,$rounds          # restore $rounds
        movups  $inout1,0x10($out)
        movdqu  0x10($inp),$inout1
        mov     $rnds_,$rounds          # restore $rounds
        movups  $inout1,0x10($out)
        movdqu  0x10($inp),$inout1


@@ -637,17 +635,17 @@ $code.=<<___;
        movups  $inout6,0x60($out)
        movdqu  0x60($inp),$inout6
        movups  $inout7,0x70($out)
        movups  $inout6,0x60($out)
        movdqu  0x60($inp),$inout6
        movups  $inout7,0x70($out)

-       lea     0x80($out),$out
+       lea     0x80($out),$out         # $out+=8*16

        movdqu  0x70($inp),$inout7
        movdqu  0x70($inp),$inout7

-       lea     0x80($inp),$inp
+       lea     0x80($inp),$inp         # $inp+=8*16

 .Lecb_enc_loop8_enter:
 
        call    _aesni_encrypt8
 
        sub     \$0x80,$len
 .Lecb_enc_loop8_enter:
 
        call    _aesni_encrypt8
 
        sub     \$0x80,$len

-       jnc     .Lecb_enc_loop8
+       jnc     .Lecb_enc_loop8         # loop if $len-=8*16 didn't borrow

 
 

-       movups  $inout0,($out)
+       movups  $inout0,($out)          # store 8 output blocks

        mov     $key_,$key              # restore $key
        movups  $inout1,0x10($out)
        mov     $rnds_,$rounds          # restore $rounds
        mov     $key_,$key              # restore $key
        movups  $inout1,0x10($out)
        mov     $rnds_,$rounds          # restore $rounds


@@ -657,11 +655,11 @@ $code.=<<___;
        movups  $inout5,0x50($out)
        movups  $inout6,0x60($out)
        movups  $inout7,0x70($out)
        movups  $inout5,0x50($out)
        movups  $inout6,0x60($out)
        movups  $inout7,0x70($out)

-       lea     0x80($out),$out
-       add     \$0x80,$len
-       jz      .Lecb_ret
+       lea     0x80($out),$out         # $out+=8*16
+       add     \$0x80,$len             # restore real remaining $len
+       jz      .Lecb_ret               # done if ($len==0)

 
 

-.Lecb_enc_tail:
+.Lecb_enc_tail:                                # $len is less than 8*16

        movups  ($inp),$inout0
        cmp     \$0x20,$len
        jb      .Lecb_enc_one
        movups  ($inp),$inout0
        cmp     \$0x20,$len
        jb      .Lecb_enc_one


@@ -678,8 +676,9 @@ $code.=<<___;
        movups  0x50($inp),$inout5
        je      .Lecb_enc_six
        movdqu  0x60($inp),$inout6
        movups  0x50($inp),$inout5
        je      .Lecb_enc_six
        movdqu  0x60($inp),$inout6

+       xorps   $inout7,$inout7

        call    _aesni_encrypt8
        call    _aesni_encrypt8

-       movups  $inout0,($out)
+       movups  $inout0,($out)          # store 7 output blocks

        movups  $inout1,0x10($out)
        movups  $inout2,0x20($out)
        movups  $inout3,0x30($out)
        movups  $inout1,0x10($out)
        movups  $inout2,0x20($out)
        movups  $inout3,0x30($out)


@@ -692,25 +691,25 @@ $code.=<<___;
 ___
        &aesni_generate1("enc",$key,$rounds);
 $code.=<<___;
 ___
        &aesni_generate1("enc",$key,$rounds);
 $code.=<<___;

-       movups  $inout0,($out)
+       movups  $inout0,($out)          # store one output block

        jmp     .Lecb_ret
 .align 16
 .Lecb_enc_two:
        call    _aesni_encrypt2
        jmp     .Lecb_ret
 .align 16
 .Lecb_enc_two:
        call    _aesni_encrypt2

-       movups  $inout0,($out)
+       movups  $inout0,($out)          # store 2 output blocks

        movups  $inout1,0x10($out)
        jmp     .Lecb_ret
 .align 16
 .Lecb_enc_three:
        call    _aesni_encrypt3
        movups  $inout1,0x10($out)
        jmp     .Lecb_ret
 .align 16
 .Lecb_enc_three:
        call    _aesni_encrypt3

-       movups  $inout0,($out)
+       movups  $inout0,($out)          # store 3 output blocks

        movups  $inout1,0x10($out)
        movups  $inout2,0x20($out)
        jmp     .Lecb_ret
 .align 16
 .Lecb_enc_four:
        call    _aesni_encrypt4
        movups  $inout1,0x10($out)
        movups  $inout2,0x20($out)
        jmp     .Lecb_ret
 .align 16
 .Lecb_enc_four:
        call    _aesni_encrypt4

-       movups  $inout0,($out)
+       movups  $inout0,($out)          # store 4 output blocks

        movups  $inout1,0x10($out)
        movups  $inout2,0x20($out)
        movups  $inout3,0x30($out)
        movups  $inout1,0x10($out)
        movups  $inout2,0x20($out)
        movups  $inout3,0x30($out)


@@ -719,7 +718,7 @@ $code.=<<___;
 .Lecb_enc_five:
        xorps   $inout5,$inout5
        call    _aesni_encrypt6
 .Lecb_enc_five:
        xorps   $inout5,$inout5
        call    _aesni_encrypt6

-       movups  $inout0,($out)
+       movups  $inout0,($out)          # store 5 output blocks

        movups  $inout1,0x10($out)
        movups  $inout2,0x20($out)
        movups  $inout3,0x30($out)
        movups  $inout1,0x10($out)
        movups  $inout2,0x20($out)
        movups  $inout3,0x30($out)


@@ -728,7 +727,7 @@ $code.=<<___;
 .align 16
 .Lecb_enc_six:
        call    _aesni_encrypt6
 .align 16
 .Lecb_enc_six:
        call    _aesni_encrypt6

-       movups  $inout0,($out)
+       movups  $inout0,($out)          # store 6 output blocks

        movups  $inout1,0x10($out)
        movups  $inout2,0x20($out)
        movups  $inout3,0x30($out)
        movups  $inout1,0x10($out)
        movups  $inout2,0x20($out)
        movups  $inout3,0x30($out)


@@ -738,10 +737,10 @@ $code.=<<___;
 \f#--------------------------- ECB DECRYPT ------------------------------#
 .align 16
 .Lecb_decrypt:
 \f#--------------------------- ECB DECRYPT ------------------------------#
 .align 16
 .Lecb_decrypt:

-       cmp     \$0x80,$len
-       jb      .Lecb_dec_tail
+       cmp     \$0x80,$len             # if ($len<8*16)
+       jb      .Lecb_dec_tail          # short input

 
 

-       movdqu  ($inp),$inout0
+       movdqu  ($inp),$inout0          # load 8 input blocks

        movdqu  0x10($inp),$inout1
        movdqu  0x20($inp),$inout2
        movdqu  0x30($inp),$inout3
        movdqu  0x10($inp),$inout1
        movdqu  0x20($inp),$inout2
        movdqu  0x30($inp),$inout3


@@ -749,14 +748,14 @@ $code.=<<___;
        movdqu  0x50($inp),$inout5
        movdqu  0x60($inp),$inout6
        movdqu  0x70($inp),$inout7
        movdqu  0x50($inp),$inout5
        movdqu  0x60($inp),$inout6
        movdqu  0x70($inp),$inout7

-       lea     0x80($inp),$inp
-       sub     \$0x80,$len
+       lea     0x80($inp),$inp         # $inp+=8*16
+       sub     \$0x80,$len             # $len-=8*16 (can be zero)

        jmp     .Lecb_dec_loop8_enter
 .align 16
 .Lecb_dec_loop8:
        jmp     .Lecb_dec_loop8_enter
 .align 16
 .Lecb_dec_loop8:

-       movups  $inout0,($out)
+       movups  $inout0,($out)          # store 8 output blocks

        mov     $key_,$key              # restore $key
        mov     $key_,$key              # restore $key

-       movdqu  ($inp),$inout0
+       movdqu  ($inp),$inout0          # load 8 input blocks

        mov     $rnds_,$rounds          # restore $rounds
        movups  $inout1,0x10($out)
        movdqu  0x10($inp),$inout1
        mov     $rnds_,$rounds          # restore $rounds
        movups  $inout1,0x10($out)
        movdqu  0x10($inp),$inout1


@@ -771,30 +770,38 @@ $code.=<<___;
        movups  $inout6,0x60($out)
        movdqu  0x60($inp),$inout6
        movups  $inout7,0x70($out)
        movups  $inout6,0x60($out)
        movdqu  0x60($inp),$inout6
        movups  $inout7,0x70($out)

-       lea     0x80($out),$out
+       lea     0x80($out),$out         # $out+=8*16

        movdqu  0x70($inp),$inout7
        movdqu  0x70($inp),$inout7

-       lea     0x80($inp),$inp
+       lea     0x80($inp),$inp         # $inp+=8*16

 .Lecb_dec_loop8_enter:
 
        call    _aesni_decrypt8
 
        $movkey ($key_),$rndkey0
        sub     \$0x80,$len
 .Lecb_dec_loop8_enter:
 
        call    _aesni_decrypt8
 
        $movkey ($key_),$rndkey0
        sub     \$0x80,$len

-       jnc     .Lecb_dec_loop8
+       jnc     .Lecb_dec_loop8         # loop if $len-=8*16 didn't borrow

 
 

-       movups  $inout0,($out)
+       movups  $inout0,($out)          # store 8 output blocks
+        pxor   $inout0,$inout0         # clear register bank

        mov     $key_,$key              # restore $key
        movups  $inout1,0x10($out)
        mov     $key_,$key              # restore $key
        movups  $inout1,0x10($out)

+        pxor   $inout1,$inout1

        mov     $rnds_,$rounds          # restore $rounds
        movups  $inout2,0x20($out)
        mov     $rnds_,$rounds          # restore $rounds
        movups  $inout2,0x20($out)

+        pxor   $inout2,$inout2

        movups  $inout3,0x30($out)
        movups  $inout3,0x30($out)

+        pxor   $inout3,$inout3

        movups  $inout4,0x40($out)
        movups  $inout4,0x40($out)

+        pxor   $inout4,$inout4

        movups  $inout5,0x50($out)
        movups  $inout5,0x50($out)

+        pxor   $inout5,$inout5

        movups  $inout6,0x60($out)
        movups  $inout6,0x60($out)

+        pxor   $inout6,$inout6

        movups  $inout7,0x70($out)
        movups  $inout7,0x70($out)

-       lea     0x80($out),$out
-       add     \$0x80,$len
-       jz      .Lecb_ret
+        pxor   $inout7,$inout7
+       lea     0x80($out),$out         # $out+=8*16
+       add     \$0x80,$len             # restore real remaining $len
+       jz      .Lecb_ret               # done if ($len==0)

 
 .Lecb_dec_tail:
        movups  ($inp),$inout0
 
 .Lecb_dec_tail:
        movups  ($inp),$inout0


@@ -814,70 +821,107 @@ $code.=<<___;
        je      .Lecb_dec_six
        movups  0x60($inp),$inout6
        $movkey ($key),$rndkey0
        je      .Lecb_dec_six
        movups  0x60($inp),$inout6
        $movkey ($key),$rndkey0

+       xorps   $inout7,$inout7

        call    _aesni_decrypt8
        call    _aesni_decrypt8

-       movups  $inout0,($out)
+       movups  $inout0,($out)          # store 7 output blocks
+        pxor   $inout0,$inout0         # clear register bank

        movups  $inout1,0x10($out)
        movups  $inout1,0x10($out)

+        pxor   $inout1,$inout1

        movups  $inout2,0x20($out)
        movups  $inout2,0x20($out)

+        pxor   $inout2,$inout2

        movups  $inout3,0x30($out)
        movups  $inout3,0x30($out)

+        pxor   $inout3,$inout3

        movups  $inout4,0x40($out)
        movups  $inout4,0x40($out)

+        pxor   $inout4,$inout4

        movups  $inout5,0x50($out)
        movups  $inout5,0x50($out)

+        pxor   $inout5,$inout5

        movups  $inout6,0x60($out)
        movups  $inout6,0x60($out)

+        pxor   $inout6,$inout6
+        pxor   $inout7,$inout7

        jmp     .Lecb_ret
 .align 16
 .Lecb_dec_one:
 ___
        &aesni_generate1("dec",$key,$rounds);
 $code.=<<___;
        jmp     .Lecb_ret
 .align 16
 .Lecb_dec_one:
 ___
        &aesni_generate1("dec",$key,$rounds);
 $code.=<<___;

-       movups  $inout0,($out)
+       movups  $inout0,($out)          # store one output block
+        pxor   $inout0,$inout0         # clear register bank

        jmp     .Lecb_ret
 .align 16
 .Lecb_dec_two:
        call    _aesni_decrypt2
        jmp     .Lecb_ret
 .align 16
 .Lecb_dec_two:
        call    _aesni_decrypt2

-       movups  $inout0,($out)
+       movups  $inout0,($out)          # store 2 output blocks
+        pxor   $inout0,$inout0         # clear register bank

        movups  $inout1,0x10($out)
        movups  $inout1,0x10($out)

+        pxor   $inout1,$inout1

        jmp     .Lecb_ret
 .align 16
 .Lecb_dec_three:
        call    _aesni_decrypt3
        jmp     .Lecb_ret
 .align 16
 .Lecb_dec_three:
        call    _aesni_decrypt3

-       movups  $inout0,($out)
+       movups  $inout0,($out)          # store 3 output blocks
+        pxor   $inout0,$inout0         # clear register bank

        movups  $inout1,0x10($out)
        movups  $inout1,0x10($out)

+        pxor   $inout1,$inout1

        movups  $inout2,0x20($out)
        movups  $inout2,0x20($out)

+        pxor   $inout2,$inout2

        jmp     .Lecb_ret
 .align 16
 .Lecb_dec_four:
        call    _aesni_decrypt4
        jmp     .Lecb_ret
 .align 16
 .Lecb_dec_four:
        call    _aesni_decrypt4

-       movups  $inout0,($out)
+       movups  $inout0,($out)          # store 4 output blocks
+        pxor   $inout0,$inout0         # clear register bank

        movups  $inout1,0x10($out)
        movups  $inout1,0x10($out)

+        pxor   $inout1,$inout1

        movups  $inout2,0x20($out)
        movups  $inout2,0x20($out)

+        pxor   $inout2,$inout2

        movups  $inout3,0x30($out)
        movups  $inout3,0x30($out)

+        pxor   $inout3,$inout3

        jmp     .Lecb_ret
 .align 16
 .Lecb_dec_five:
        xorps   $inout5,$inout5
        call    _aesni_decrypt6
        jmp     .Lecb_ret
 .align 16
 .Lecb_dec_five:
        xorps   $inout5,$inout5
        call    _aesni_decrypt6

-       movups  $inout0,($out)
+       movups  $inout0,($out)          # store 5 output blocks
+        pxor   $inout0,$inout0         # clear register bank

        movups  $inout1,0x10($out)
        movups  $inout1,0x10($out)

+        pxor   $inout1,$inout1

        movups  $inout2,0x20($out)
        movups  $inout2,0x20($out)

+        pxor   $inout2,$inout2

        movups  $inout3,0x30($out)
        movups  $inout3,0x30($out)

+        pxor   $inout3,$inout3

        movups  $inout4,0x40($out)
        movups  $inout4,0x40($out)

+        pxor   $inout4,$inout4
+        pxor   $inout5,$inout5

        jmp     .Lecb_ret
 .align 16
 .Lecb_dec_six:
        call    _aesni_decrypt6
        jmp     .Lecb_ret
 .align 16
 .Lecb_dec_six:
        call    _aesni_decrypt6

-       movups  $inout0,($out)
+       movups  $inout0,($out)          # store 6 output blocks
+        pxor   $inout0,$inout0         # clear register bank

        movups  $inout1,0x10($out)
        movups  $inout1,0x10($out)

+        pxor   $inout1,$inout1

        movups  $inout2,0x20($out)
        movups  $inout2,0x20($out)

+        pxor   $inout2,$inout2

        movups  $inout3,0x30($out)
        movups  $inout3,0x30($out)

+        pxor   $inout3,$inout3

        movups  $inout4,0x40($out)
        movups  $inout4,0x40($out)

+        pxor   $inout4,$inout4

        movups  $inout5,0x50($out)
        movups  $inout5,0x50($out)

+        pxor   $inout5,$inout5

 
 .Lecb_ret:
 
 .Lecb_ret:

+       xorps   $rndkey0,$rndkey0       # %xmm0
+       pxor    $rndkey1,$rndkey1

 ___
 $code.=<<___ if ($win64);
        movaps  (%rsp),%xmm6
 ___
 $code.=<<___ if ($win64);
        movaps  (%rsp),%xmm6

+       movaps  %xmm0,(%rsp)            # clear stack

        movaps  0x10(%rsp),%xmm7
        movaps  0x10(%rsp),%xmm7

+       movaps  %xmm0,0x10(%rsp)

        movaps  0x20(%rsp),%xmm8
        movaps  0x20(%rsp),%xmm8

+       movaps  %xmm0,0x20(%rsp)

        movaps  0x30(%rsp),%xmm9
        movaps  0x30(%rsp),%xmm9

+       movaps  %xmm0,0x30(%rsp)

        lea     0x58(%rsp),%rsp
 .Lecb_enc_ret:
 ___
        lea     0x58(%rsp),%rsp
 .Lecb_enc_ret:
 ___


@@ -911,10 +955,10 @@ aesni_ccm64_encrypt_blocks:
 ___
 $code.=<<___ if ($win64);
        lea     -0x58(%rsp),%rsp
 ___
 $code.=<<___ if ($win64);
        lea     -0x58(%rsp),%rsp

-       movaps  %xmm6,(%rsp)
-       movaps  %xmm7,0x10(%rsp)
-       movaps  %xmm8,0x20(%rsp)
-       movaps  %xmm9,0x30(%rsp)
+       movaps  %xmm6,(%rsp)            # $iv
+       movaps  %xmm7,0x10(%rsp)        # $bswap_mask
+       movaps  %xmm8,0x20(%rsp)        # $in0
+       movaps  %xmm9,0x30(%rsp)        # $increment

 .Lccm64_enc_body:
 ___
 $code.=<<___;
 .Lccm64_enc_body:
 ___
 $code.=<<___;


@@ -956,7 +1000,7 @@ $code.=<<___;
        aesenc  $rndkey1,$inout0
        aesenc  $rndkey1,$inout1
        paddq   $increment,$iv
        aesenc  $rndkey1,$inout0
        aesenc  $rndkey1,$inout1
        paddq   $increment,$iv

-       dec     $len
+       dec     $len                            # $len-- ($len is in blocks)

        aesenclast      $rndkey0,$inout0
        aesenclast      $rndkey0,$inout1
 
        aesenclast      $rndkey0,$inout0
        aesenclast      $rndkey0,$inout1
 


@@ -965,16 +1009,26 @@ $code.=<<___;
        movdqa  $iv,$inout0
        movups  $in0,($out)                     # save output
        pshufb  $bswap_mask,$inout0
        movdqa  $iv,$inout0
        movups  $in0,($out)                     # save output
        pshufb  $bswap_mask,$inout0

-       lea     16($out),$out
-       jnz     .Lccm64_enc_outer
+       lea     16($out),$out                   # $out+=16
+       jnz     .Lccm64_enc_outer               # loop if ($len!=0)

 
 

-       movups  $inout1,($cmac)
+        pxor   $rndkey0,$rndkey0               # clear register bank
+        pxor   $rndkey1,$rndkey1
+        pxor   $inout0,$inout0
+       movups  $inout1,($cmac)                 # store resulting mac
+        pxor   $inout1,$inout1
+        pxor   $in0,$in0
+        pxor   $iv,$iv

 ___
 $code.=<<___ if ($win64);
        movaps  (%rsp),%xmm6
 ___
 $code.=<<___ if ($win64);
        movaps  (%rsp),%xmm6

+       movaps  %xmm0,(%rsp)                    # clear stack

        movaps  0x10(%rsp),%xmm7
        movaps  0x10(%rsp),%xmm7

+       movaps  %xmm0,0x10(%rsp)

        movaps  0x20(%rsp),%xmm8
        movaps  0x20(%rsp),%xmm8

+       movaps  %xmm0,0x20(%rsp)

        movaps  0x30(%rsp),%xmm9
        movaps  0x30(%rsp),%xmm9

+       movaps  %xmm0,0x30(%rsp)

        lea     0x58(%rsp),%rsp
 .Lccm64_enc_ret:
 ___
        lea     0x58(%rsp),%rsp
 .Lccm64_enc_ret:
 ___


@@ -991,10 +1045,10 @@ aesni_ccm64_decrypt_blocks:
 ___
 $code.=<<___ if ($win64);
        lea     -0x58(%rsp),%rsp
 ___
 $code.=<<___ if ($win64);
        lea     -0x58(%rsp),%rsp

-       movaps  %xmm6,(%rsp)
-       movaps  %xmm7,0x10(%rsp)
-       movaps  %xmm8,0x20(%rsp)
-       movaps  %xmm9,0x30(%rsp)
+       movaps  %xmm6,(%rsp)            # $iv
+       movaps  %xmm7,0x10(%rsp)        # $bswap_mask
+       movaps  %xmm8,0x20(%rsp)        # $in8
+       movaps  %xmm9,0x30(%rsp)        # $increment

 .Lccm64_dec_body:
 ___
 $code.=<<___;
 .Lccm64_dec_body:
 ___
 $code.=<<___;


@@ -1015,7 +1069,7 @@ $code.=<<___;
        mov     \$16,$rounds
        movups  ($inp),$in0                     # load inp
        paddq   $increment,$iv
        mov     \$16,$rounds
        movups  ($inp),$in0                     # load inp
        paddq   $increment,$iv

-       lea     16($inp),$inp
+       lea     16($inp),$inp                   # $inp+=16

        sub     %r10,%rax                       # twisted $rounds
        lea     32($key_,$rnds_),$key           # end of key schedule
        mov     %rax,%r10
        sub     %r10,%rax                       # twisted $rounds
        lea     32($key_,$rnds_),$key           # end of key schedule
        mov     %rax,%r10


@@ -1025,11 +1079,11 @@ $code.=<<___;
        xorps   $inout0,$in0                    # inp ^= E(iv)
        movdqa  $iv,$inout0
        movups  $in0,($out)                     # save output
        xorps   $inout0,$in0                    # inp ^= E(iv)
        movdqa  $iv,$inout0
        movups  $in0,($out)                     # save output

-       lea     16($out),$out
+       lea     16($out),$out                   # $out+=16

        pshufb  $bswap_mask,$inout0
 
        pshufb  $bswap_mask,$inout0
 

-       sub     \$1,$len
-       jz      .Lccm64_dec_break
+       sub     \$1,$len                        # $len-- ($len is in blocks)
+       jz      .Lccm64_dec_break               # if ($len==0) break

 
        $movkey ($key_),$rndkey0
        mov     %r10,%rax
 
        $movkey ($key_),$rndkey0
        mov     %r10,%rax


@@ -1049,13 +1103,13 @@ $code.=<<___;
        aesenc  $rndkey0,$inout1
        $movkey -16($key,%rax),$rndkey0
        jnz     .Lccm64_dec2_loop
        aesenc  $rndkey0,$inout1
        $movkey -16($key,%rax),$rndkey0
        jnz     .Lccm64_dec2_loop

-       movups  ($inp),$in0                     # load inp
+       movups  ($inp),$in0                     # load input

        paddq   $increment,$iv
        aesenc  $rndkey1,$inout0
        aesenc  $rndkey1,$inout1
        aesenclast      $rndkey0,$inout0
        aesenclast      $rndkey0,$inout1
        paddq   $increment,$iv
        aesenc  $rndkey1,$inout0
        aesenc  $rndkey1,$inout1
        aesenclast      $rndkey0,$inout0
        aesenclast      $rndkey0,$inout1

-       lea     16($inp),$inp
+       lea     16($inp),$inp                   # $inp+=16

        jmp     .Lccm64_dec_outer
 
 .align 16
        jmp     .Lccm64_dec_outer
 
 .align 16


@@ -1065,13 +1119,23 @@ $code.=<<___;
 ___
        &aesni_generate1("enc",$key_,$rounds,$inout1,$in0);
 $code.=<<___;
 ___
        &aesni_generate1("enc",$key_,$rounds,$inout1,$in0);
 $code.=<<___;

-       movups  $inout1,($cmac)
+        pxor   $rndkey0,$rndkey0               # clear register bank
+        pxor   $rndkey1,$rndkey1
+        pxor   $inout0,$inout0
+       movups  $inout1,($cmac)                 # store resulting mac
+        pxor   $inout1,$inout1
+        pxor   $in0,$in0
+        pxor   $iv,$iv

 ___
 $code.=<<___ if ($win64);
        movaps  (%rsp),%xmm6
 ___
 $code.=<<___ if ($win64);
        movaps  (%rsp),%xmm6

+       movaps  %xmm0,(%rsp)                    # clear stack

        movaps  0x10(%rsp),%xmm7
        movaps  0x10(%rsp),%xmm7

+       movaps  %xmm0,0x10(%rsp)

        movaps  0x20(%rsp),%xmm8
        movaps  0x20(%rsp),%xmm8

+       movaps  %xmm0,0x20(%rsp)

        movaps  0x30(%rsp),%xmm9
        movaps  0x30(%rsp),%xmm9

+       movaps  %xmm0,0x30(%rsp)

        lea     0x58(%rsp),%rsp
 .Lccm64_dec_ret:
 ___
        lea     0x58(%rsp),%rsp
 .Lccm64_dec_ret:
 ___


@@ -1102,13 +1166,34 @@ $code.=<<___;
 .type  aesni_ctr32_encrypt_blocks,\@function,5
 .align 16
 aesni_ctr32_encrypt_blocks:
 .type  aesni_ctr32_encrypt_blocks,\@function,5
 .align 16
 aesni_ctr32_encrypt_blocks:

+       cmp     \$1,$len
+       jne     .Lctr32_bulk
+
+       # handle single block without allocating stack frame,
+       # useful when handling edges
+       movups  ($ivp),$inout0
+       movups  ($inp),$inout1
+       mov     240($key),%edx                  # key->rounds
+___
+       &aesni_generate1("enc",$key,"%edx");
+$code.=<<___;
+        pxor   $rndkey0,$rndkey0               # clear register bank
+        pxor   $rndkey1,$rndkey1
+       xorps   $inout1,$inout0
+        pxor   $inout1,$inout1
+       movups  $inout0,($out)
+        xorps  $inout0,$inout0
+       jmp     .Lctr32_epilogue
+
+.align 16
+.Lctr32_bulk:

        lea     (%rsp),%rax
        push    %rbp
        sub     \$$frame_size,%rsp
        and     \$-16,%rsp      # Linux kernel stack can be incorrectly seeded
 ___
 $code.=<<___ if ($win64);
        lea     (%rsp),%rax
        push    %rbp
        sub     \$$frame_size,%rsp
        and     \$-16,%rsp      # Linux kernel stack can be incorrectly seeded
 ___
 $code.=<<___ if ($win64);

-       movaps  %xmm6,-0xa8(%rax)
+       movaps  %xmm6,-0xa8(%rax)               # offload everything

        movaps  %xmm7,-0x98(%rax)
        movaps  %xmm8,-0x88(%rax)
        movaps  %xmm9,-0x78(%rax)
        movaps  %xmm7,-0x98(%rax)
        movaps  %xmm8,-0x88(%rax)
        movaps  %xmm9,-0x78(%rax)


@@ -1123,8 +1208,8 @@ ___
 $code.=<<___;
        lea     -8(%rax),%rbp
 
 $code.=<<___;
        lea     -8(%rax),%rbp
 

-       cmp     \$1,$len
-       je      .Lctr32_one_shortcut
+       # 8 16-byte words on top of stack are counter values
+       # xor-ed with zero-round key

 
        movdqu  ($ivp),$inout0
        movdqu  ($key),$rndkey0
 
        movdqu  ($ivp),$inout0
        movdqu  ($key),$rndkey0


@@ -1139,7 +1224,7 @@ $code.=<<___;
        movdqa  $inout0,0x40(%rsp)
        movdqa  $inout0,0x50(%rsp)
        movdqa  $inout0,0x60(%rsp)
        movdqa  $inout0,0x40(%rsp)
        movdqa  $inout0,0x50(%rsp)
        movdqa  $inout0,0x60(%rsp)

-       mov     %rdx,%r10                       # borrow %rdx
+       mov     %rdx,%r10                       # about to borrow %rdx

        movdqa  $inout0,0x70(%rsp)
 
        lea     1($ctr),%rax
        movdqa  $inout0,0x70(%rsp)
 
        lea     1($ctr),%rax


@@ -1183,15 +1268,15 @@ $code.=<<___;
        movdqa  0x40(%rsp),$inout4
        movdqa  0x50(%rsp),$inout5
 
        movdqa  0x40(%rsp),$inout4
        movdqa  0x50(%rsp),$inout5
 

-       cmp     \$8,$len
-       jb      .Lctr32_tail
+       cmp     \$8,$len                # $len is in blocks
+       jb      .Lctr32_tail            # short input if ($len<8)

 
 

-       sub     \$6,$len
+       sub     \$6,$len                # $len is biased by -6

        cmp     \$`1<<22`,%r10d         # check for MOVBE without XSAVE
        cmp     \$`1<<22`,%r10d         # check for MOVBE without XSAVE

-       je      .Lctr32_6x
+       je      .Lctr32_6x              # [which denotes Atom Silvermont]

 
        lea     0x80($key),$key         # size optimization
 
        lea     0x80($key),$key         # size optimization

-       sub     \$2,$len
+       sub     \$2,$len                # $len is biased by -8

        jmp     .Lctr32_loop8
 
 .align 16
        jmp     .Lctr32_loop8
 
 .align 16


@@ -1205,13 +1290,13 @@ $code.=<<___;
 
 .align 16
 .Lctr32_loop6:
 
 .align 16
 .Lctr32_loop6:

-        add    \$6,$ctr
+        add    \$6,$ctr                # next counter value

        $movkey -48($key,$rnds_),$rndkey0
        aesenc  $rndkey1,$inout0
         mov    $ctr,%eax
         xor    $key0,%eax
        aesenc  $rndkey1,$inout1
        $movkey -48($key,$rnds_),$rndkey0
        aesenc  $rndkey1,$inout0
         mov    $ctr,%eax
         xor    $key0,%eax
        aesenc  $rndkey1,$inout1

-        movbe  %eax,`0x00+12`(%rsp)
+        movbe  %eax,`0x00+12`(%rsp)    # store next counter value

         lea    1($ctr),%eax
        aesenc  $rndkey1,$inout2
         xor    $key0,%eax
         lea    1($ctr),%eax
        aesenc  $rndkey1,$inout2
         xor    $key0,%eax


@@ -1244,16 +1329,16 @@ $code.=<<___;
 
        call    .Lenc_loop6
 
 
        call    .Lenc_loop6
 

-       movdqu  ($inp),$inout6
+       movdqu  ($inp),$inout6          # load 6 input blocks

        movdqu  0x10($inp),$inout7
        movdqu  0x20($inp),$in0
        movdqu  0x30($inp),$in1
        movdqu  0x40($inp),$in2
        movdqu  0x50($inp),$in3
        movdqu  0x10($inp),$inout7
        movdqu  0x20($inp),$in0
        movdqu  0x30($inp),$in1
        movdqu  0x40($inp),$in2
        movdqu  0x50($inp),$in3

-       lea     0x60($inp),$inp
+       lea     0x60($inp),$inp         # $inp+=6*16

        $movkey -64($key,$rnds_),$rndkey1
        $movkey -64($key,$rnds_),$rndkey1

-       pxor    $inout0,$inout6
-       movaps  0x00(%rsp),$inout0
+       pxor    $inout0,$inout6         # inp^=E(ctr)
+       movaps  0x00(%rsp),$inout0      # load next counter [xor-ed with 0 round]

        pxor    $inout1,$inout7
        movaps  0x10(%rsp),$inout1
        pxor    $inout2,$in0
        pxor    $inout1,$inout7
        movaps  0x10(%rsp),$inout1
        pxor    $inout2,$in0


@@ -1264,19 +1349,19 @@ $code.=<<___;
        movaps  0x40(%rsp),$inout4
        pxor    $inout5,$in3
        movaps  0x50(%rsp),$inout5
        movaps  0x40(%rsp),$inout4
        pxor    $inout5,$in3
        movaps  0x50(%rsp),$inout5

-       movdqu  $inout6,($out)
+       movdqu  $inout6,($out)          # store 6 output blocks

        movdqu  $inout7,0x10($out)
        movdqu  $in0,0x20($out)
        movdqu  $in1,0x30($out)
        movdqu  $in2,0x40($out)
        movdqu  $in3,0x50($out)
        movdqu  $inout7,0x10($out)
        movdqu  $in0,0x20($out)
        movdqu  $in1,0x30($out)
        movdqu  $in2,0x40($out)
        movdqu  $in3,0x50($out)

-       lea     0x60($out),$out
-       
+       lea     0x60($out),$out         # $out+=6*16
+

        sub     \$6,$len
        sub     \$6,$len

-       jnc     .Lctr32_loop6
+       jnc     .Lctr32_loop6           # loop if $len-=6 didn't borrow

 
 

-       add     \$6,$len
-       jz      .Lctr32_done
+       add     \$6,$len                # restore real remaining $len
+       jz      .Lctr32_done            # done if ($len==0)

 
        lea     -48($rnds_),$rounds
        lea     -80($key,$rnds_),$key   # restore $key
 
        lea     -48($rnds_),$rounds
        lea     -80($key,$rnds_),$key   # restore $key


@@ -1286,7 +1371,7 @@ $code.=<<___;
 
 .align 32
 .Lctr32_loop8:
 
 .align 32
 .Lctr32_loop8:

-        add            \$8,$ctr
+        add            \$8,$ctr                # next counter value

        movdqa          0x60(%rsp),$inout6
        aesenc          $rndkey1,$inout0
         mov            $ctr,%r9d
        movdqa          0x60(%rsp),$inout6
        aesenc          $rndkey1,$inout0
         mov            $ctr,%r9d


@@ -1298,7 +1383,7 @@ $code.=<<___;
         xor            $key0,%r9d
         nop
        aesenc          $rndkey1,$inout3
         xor            $key0,%r9d
         nop
        aesenc          $rndkey1,$inout3

-        mov            %r9d,0x00+12(%rsp)
+        mov            %r9d,0x00+12(%rsp)      # store next counter value

         lea            1($ctr),%r9
        aesenc          $rndkey1,$inout4
        aesenc          $rndkey1,$inout5
         lea            1($ctr),%r9
        aesenc          $rndkey1,$inout4
        aesenc          $rndkey1,$inout5


@@ -1331,7 +1416,7 @@ $code.=<<___;
        aesenc          $rndkey0,$inout1
        aesenc          $rndkey0,$inout2
         xor            $key0,%r9d
        aesenc          $rndkey0,$inout1
        aesenc          $rndkey0,$inout2
         xor            $key0,%r9d

-        movdqu         0x00($inp),$in0
+        movdqu         0x00($inp),$in0         # start loading input

        aesenc          $rndkey0,$inout3
         mov            %r9d,0x70+12(%rsp)
         cmp            \$11,$rounds
        aesenc          $rndkey0,$inout3
         mov            %r9d,0x70+12(%rsp)
         cmp            \$11,$rounds


@@ -1388,7 +1473,7 @@ $code.=<<___;
 .align 16
 .Lctr32_enc_done:
        movdqu          0x10($inp),$in1
 .align 16
 .Lctr32_enc_done:
        movdqu          0x10($inp),$in1

-       pxor            $rndkey0,$in0
+       pxor            $rndkey0,$in0           # input^=round[last]

        movdqu          0x20($inp),$in2
        pxor            $rndkey0,$in1
        movdqu          0x30($inp),$in3
        movdqu          0x20($inp),$in2
        pxor            $rndkey0,$in1
        movdqu          0x30($inp),$in3


@@ -1406,11 +1491,11 @@ $code.=<<___;
        aesenc          $rndkey1,$inout5
        aesenc          $rndkey1,$inout6
        aesenc          $rndkey1,$inout7
        aesenc          $rndkey1,$inout5
        aesenc          $rndkey1,$inout6
        aesenc          $rndkey1,$inout7

-       movdqu          0x60($inp),$rndkey1
-       lea             0x80($inp),$inp
+       movdqu          0x60($inp),$rndkey1     # borrow $rndkey1 for inp[6]
+       lea             0x80($inp),$inp         # $inp+=8*16

 
 

-       aesenclast      $in0,$inout0
-       pxor            $rndkey0,$rndkey1
+       aesenclast      $in0,$inout0            # $inN is inp[N]^round[last]
+       pxor            $rndkey0,$rndkey1       # borrowed $rndkey

        movdqu          0x70-0x80($inp),$in0
        aesenclast      $in1,$inout1
        pxor            $rndkey0,$in0
        movdqu          0x70-0x80($inp),$in0
        aesenclast      $in1,$inout1
        pxor            $rndkey0,$in0


@@ -1425,10 +1510,10 @@ $code.=<<___;
        movdqa          0x40(%rsp),$in5
        aesenclast      $rndkey1,$inout6
        movdqa          0x50(%rsp),$rndkey0
        movdqa          0x40(%rsp),$in5
        aesenclast      $rndkey1,$inout6
        movdqa          0x50(%rsp),$rndkey0

-       $movkey         0x10-0x80($key),$rndkey1
+       $movkey         0x10-0x80($key),$rndkey1#real 1st-round key

        aesenclast      $in0,$inout7
 
        aesenclast      $in0,$inout7
 

-       movups          $inout0,($out)          # store output
+       movups          $inout0,($out)          # store 8 output blocks

        movdqa          $in1,$inout0
        movups          $inout1,0x10($out)
        movdqa          $in2,$inout1
        movdqa          $in1,$inout0
        movups          $inout1,0x10($out)
        movdqa          $in2,$inout1


@@ -1442,21 +1527,24 @@ $code.=<<___;
        movdqa          $rndkey0,$inout5
        movups          $inout6,0x60($out)
        movups          $inout7,0x70($out)
        movdqa          $rndkey0,$inout5
        movups          $inout6,0x60($out)
        movups          $inout7,0x70($out)

-       lea             0x80($out),$out
-       
+       lea             0x80($out),$out         # $out+=8*16
+

        sub     \$8,$len
        sub     \$8,$len

-       jnc     .Lctr32_loop8
+       jnc     .Lctr32_loop8                   # loop if $len-=8 didn't borrow

 
 

-       add     \$8,$len
-       jz      .Lctr32_done
+       add     \$8,$len                        # restore real remainig $len
+       jz      .Lctr32_done                    # done if ($len==0)

        lea     -0x80($key),$key
 
 .Lctr32_tail:
        lea     -0x80($key),$key
 
 .Lctr32_tail:

+       # note that at this point $inout0..5 are populated with
+       # counter values xor-ed with 0-round key 

        lea     16($key),$key
        cmp     \$4,$len
        jb      .Lctr32_loop3
        je      .Lctr32_loop4
 
        lea     16($key),$key
        cmp     \$4,$len
        jb      .Lctr32_loop3
        je      .Lctr32_loop4
 

+       # if ($len>4) compute 7 E(counter)

        shl             \$4,$rounds
        movdqa          0x60(%rsp),$inout6
        pxor            $inout7,$inout7
        shl             \$4,$rounds
        movdqa          0x60(%rsp),$inout6
        pxor            $inout7,$inout7


@@ -1464,14 +1552,14 @@ $code.=<<___;
        $movkey         16($key),$rndkey0
        aesenc          $rndkey1,$inout0
        aesenc          $rndkey1,$inout1
        $movkey         16($key),$rndkey0
        aesenc          $rndkey1,$inout0
        aesenc          $rndkey1,$inout1

-       lea             32-16($key,$rounds),$key
+       lea             32-16($key,$rounds),$key# prepare for .Lenc_loop8_enter

        neg             %rax
        aesenc          $rndkey1,$inout2
        neg             %rax
        aesenc          $rndkey1,$inout2

-       add             \$16,%rax
+       add             \$16,%rax               # prepare for .Lenc_loop8_enter

         movups         ($inp),$in0
        aesenc          $rndkey1,$inout3
        aesenc          $rndkey1,$inout4
         movups         ($inp),$in0
        aesenc          $rndkey1,$inout3
        aesenc          $rndkey1,$inout4

-        movups         0x10($inp),$in1
+        movups         0x10($inp),$in1         # pre-load input

         movups         0x20($inp),$in2
        aesenc          $rndkey1,$inout5
        aesenc          $rndkey1,$inout6
         movups         0x20($inp),$in2
        aesenc          $rndkey1,$inout5
        aesenc          $rndkey1,$inout6


@@ -1482,7 +1570,7 @@ $code.=<<___;
        pxor    $in0,$inout0
        movdqu  0x40($inp),$in0
        pxor    $in1,$inout1
        pxor    $in0,$inout0
        movdqu  0x40($inp),$in0
        pxor    $in1,$inout1

-       movdqu  $inout0,($out)
+       movdqu  $inout0,($out)                  # store output

        pxor    $in2,$inout2
        movdqu  $inout1,0x10($out)
        pxor    $in3,$inout3
        pxor    $in2,$inout2
        movdqu  $inout1,0x10($out)
        pxor    $in3,$inout3


@@ -1491,17 +1579,17 @@ $code.=<<___;
        movdqu  $inout3,0x30($out)
        movdqu  $inout4,0x40($out)
        cmp     \$6,$len
        movdqu  $inout3,0x30($out)
        movdqu  $inout4,0x40($out)
        cmp     \$6,$len

-       jb      .Lctr32_done
+       jb      .Lctr32_done                    # $len was 5, stop store

 
        movups  0x50($inp),$in1
        xorps   $in1,$inout5
        movups  $inout5,0x50($out)
 
        movups  0x50($inp),$in1
        xorps   $in1,$inout5
        movups  $inout5,0x50($out)

-       je      .Lctr32_done
+       je      .Lctr32_done                    # $len was 6, stop store

 
        movups  0x60($inp),$in2
        xorps   $in2,$inout6
        movups  $inout6,0x60($out)
 
        movups  0x60($inp),$in2
        xorps   $in2,$inout6
        movups  $inout6,0x60($out)

-       jmp     .Lctr32_done
+       jmp     .Lctr32_done                    # $len was 7, stop store

 
 .align 32
 .Lctr32_loop4:
 
 .align 32
 .Lctr32_loop4:


@@ -1515,7 +1603,7 @@ $code.=<<___;
        jnz             .Lctr32_loop4
        aesenclast      $rndkey1,$inout0
        aesenclast      $rndkey1,$inout1
        jnz             .Lctr32_loop4
        aesenclast      $rndkey1,$inout0
        aesenclast      $rndkey1,$inout1

-        movups         ($inp),$in0
+        movups         ($inp),$in0             # load input

         movups         0x10($inp),$in1
        aesenclast      $rndkey1,$inout2
        aesenclast      $rndkey1,$inout3
         movups         0x10($inp),$in1
        aesenclast      $rndkey1,$inout2
        aesenclast      $rndkey1,$inout3


@@ -1523,14 +1611,14 @@ $code.=<<___;
         movups         0x30($inp),$in3
 
        xorps   $in0,$inout0
         movups         0x30($inp),$in3
 
        xorps   $in0,$inout0

-       movups  $inout0,($out)
+       movups  $inout0,($out)                  # store output

        xorps   $in1,$inout1
        movups  $inout1,0x10($out)
        pxor    $in2,$inout2
        movdqu  $inout2,0x20($out)
        pxor    $in3,$inout3
        movdqu  $inout3,0x30($out)
        xorps   $in1,$inout1
        movups  $inout1,0x10($out)
        pxor    $in2,$inout2
        movdqu  $inout2,0x20($out)
        pxor    $in3,$inout3
        movdqu  $inout3,0x30($out)

-       jmp     .Lctr32_done
+       jmp     .Lctr32_done                    # $len was 4, stop store

 
 .align 32
 .Lctr32_loop3:
 
 .align 32
 .Lctr32_loop3:


@@ -1545,48 +1633,79 @@ $code.=<<___;
        aesenclast      $rndkey1,$inout1
        aesenclast      $rndkey1,$inout2
 
        aesenclast      $rndkey1,$inout1
        aesenclast      $rndkey1,$inout2
 

-       movups  ($inp),$in0
+       movups  ($inp),$in0                     # load input

        xorps   $in0,$inout0
        xorps   $in0,$inout0

-       movups  $inout0,($out)
+       movups  $inout0,($out)                  # store output

        cmp     \$2,$len
        cmp     \$2,$len

-       jb      .Lctr32_done
+       jb      .Lctr32_done                    # $len was 1, stop store

 
        movups  0x10($inp),$in1
        xorps   $in1,$inout1
        movups  $inout1,0x10($out)
 
        movups  0x10($inp),$in1
        xorps   $in1,$inout1
        movups  $inout1,0x10($out)

-       je      .Lctr32_done
+       je      .Lctr32_done                    # $len was 2, stop store

 
        movups  0x20($inp),$in2
        xorps   $in2,$inout2
 
        movups  0x20($inp),$in2
        xorps   $in2,$inout2

-       movups  $inout2,0x20($out)
-       jmp     .Lctr32_done
-
-.align 16
-.Lctr32_one_shortcut:
-       movups  ($ivp),$inout0
-       movups  ($inp),$in0
-       mov     240($key),$rounds               # key->rounds
-___
-       &aesni_generate1("enc",$key,$rounds);
-$code.=<<___;
-       xorps   $in0,$inout0
-       movups  $inout0,($out)
-       jmp     .Lctr32_done
+       movups  $inout2,0x20($out)              # $len was 3, stop store

 
 

-.align 16

 .Lctr32_done:
 .Lctr32_done:

+       xorps   %xmm0,%xmm0                     # clear regiser bank
+       xor     $key0,$key0
+       pxor    %xmm1,%xmm1
+       pxor    %xmm2,%xmm2
+       pxor    %xmm3,%xmm3
+       pxor    %xmm4,%xmm4
+       pxor    %xmm5,%xmm5
+___
+$code.=<<___ if (!$win64);
+       pxor    %xmm6,%xmm6
+       pxor    %xmm7,%xmm7
+       movaps  %xmm0,0x00(%rsp)                # clear stack
+       pxor    %xmm8,%xmm8
+       movaps  %xmm0,0x10(%rsp)
+       pxor    %xmm9,%xmm9
+       movaps  %xmm0,0x20(%rsp)
+       pxor    %xmm10,%xmm10
+       movaps  %xmm0,0x30(%rsp)
+       pxor    %xmm11,%xmm11
+       movaps  %xmm0,0x40(%rsp)
+       pxor    %xmm12,%xmm12
+       movaps  %xmm0,0x50(%rsp)
+       pxor    %xmm13,%xmm13
+       movaps  %xmm0,0x60(%rsp)
+       pxor    %xmm14,%xmm14
+       movaps  %xmm0,0x70(%rsp)
+       pxor    %xmm15,%xmm15

 ___
 $code.=<<___ if ($win64);
        movaps  -0xa0(%rbp),%xmm6
 ___
 $code.=<<___ if ($win64);
        movaps  -0xa0(%rbp),%xmm6

+       movaps  %xmm0,-0xa0(%rbp)               # clear stack

        movaps  -0x90(%rbp),%xmm7
        movaps  -0x90(%rbp),%xmm7

+       movaps  %xmm0,-0x90(%rbp)

        movaps  -0x80(%rbp),%xmm8
        movaps  -0x80(%rbp),%xmm8

+       movaps  %xmm0,-0x80(%rbp)

        movaps  -0x70(%rbp),%xmm9
        movaps  -0x70(%rbp),%xmm9

+       movaps  %xmm0,-0x70(%rbp)

        movaps  -0x60(%rbp),%xmm10
        movaps  -0x60(%rbp),%xmm10

+       movaps  %xmm0,-0x60(%rbp)

        movaps  -0x50(%rbp),%xmm11
        movaps  -0x50(%rbp),%xmm11

+       movaps  %xmm0,-0x50(%rbp)

        movaps  -0x40(%rbp),%xmm12
        movaps  -0x40(%rbp),%xmm12

+       movaps  %xmm0,-0x40(%rbp)

        movaps  -0x30(%rbp),%xmm13
        movaps  -0x30(%rbp),%xmm13

+       movaps  %xmm0,-0x30(%rbp)

        movaps  -0x20(%rbp),%xmm14
        movaps  -0x20(%rbp),%xmm14

+       movaps  %xmm0,-0x20(%rbp)

        movaps  -0x10(%rbp),%xmm15
        movaps  -0x10(%rbp),%xmm15

+       movaps  %xmm0,-0x10(%rbp)
+       movaps  %xmm0,0x00(%rsp)
+       movaps  %xmm0,0x10(%rsp)
+       movaps  %xmm0,0x20(%rsp)
+       movaps  %xmm0,0x30(%rsp)
+       movaps  %xmm0,0x40(%rsp)
+       movaps  %xmm0,0x50(%rsp)
+       movaps  %xmm0,0x60(%rsp)
+       movaps  %xmm0,0x70(%rsp)

 ___
 $code.=<<___;
        lea     (%rbp),%rsp
 ___
 $code.=<<___;
        lea     (%rbp),%rsp


@@ -1619,7 +1738,7 @@ aesni_xts_encrypt:
        and     \$-16,%rsp      # Linux kernel stack can be incorrectly seeded
 ___
 $code.=<<___ if ($win64);
        and     \$-16,%rsp      # Linux kernel stack can be incorrectly seeded
 ___
 $code.=<<___ if ($win64);

-       movaps  %xmm6,-0xa8(%rax)
+       movaps  %xmm6,-0xa8(%rax)               # offload everything

        movaps  %xmm7,-0x98(%rax)
        movaps  %xmm8,-0x88(%rax)
        movaps  %xmm9,-0x78(%rax)
        movaps  %xmm7,-0x98(%rax)
        movaps  %xmm8,-0x88(%rax)
        movaps  %xmm9,-0x78(%rax)


@@ -1679,7 +1798,7 @@ $code.=<<___;
        movaps  $rndkey1,0x60(%rsp)             # save round[0]^round[last]
 
        sub     \$16*6,$len
        movaps  $rndkey1,0x60(%rsp)             # save round[0]^round[last]
 
        sub     \$16*6,$len

-       jc      .Lxts_enc_short
+       jc      .Lxts_enc_short                 # if $len-=6*16 borrowed

 
        mov     \$16+96,$rounds
        lea     32($key_,$rnds_),$key           # end of key schedule
 
        mov     \$16+96,$rounds
        lea     32($key_,$rnds_),$key           # end of key schedule


@@ -1694,7 +1813,7 @@ $code.=<<___;
        movdqu  `16*0`($inp),$inout0            # load input
        movdqa  $rndkey0,$twmask
        movdqu  `16*1`($inp),$inout1
        movdqu  `16*0`($inp),$inout0            # load input
        movdqa  $rndkey0,$twmask
        movdqu  `16*1`($inp),$inout1

-       pxor    @tweak[0],$inout0
+       pxor    @tweak[0],$inout0               # input^=tweak^round[0]

        movdqu  `16*2`($inp),$inout2
        pxor    @tweak[1],$inout1
         aesenc         $rndkey1,$inout0
        movdqu  `16*2`($inp),$inout2
        pxor    @tweak[1],$inout1
         aesenc         $rndkey1,$inout0


@@ -1713,10 +1832,10 @@ $code.=<<___;
        lea     `16*6`($inp),$inp
        pxor    $twmask,$inout5
 
        lea     `16*6`($inp),$inp
        pxor    $twmask,$inout5
 

-        pxor   $twres,@tweak[0]
+        pxor   $twres,@tweak[0]                # calclulate tweaks^round[last]

        aesenc          $rndkey1,$inout4
         pxor   $twres,@tweak[1]
        aesenc          $rndkey1,$inout4
         pxor   $twres,@tweak[1]

-        movdqa @tweak[0],`16*0`(%rsp)          # put aside tweaks^last round key
+        movdqa @tweak[0],`16*0`(%rsp)          # put aside tweaks^round[last]

        aesenc          $rndkey1,$inout5
        $movkey         48($key_),$rndkey1
         pxor   $twres,@tweak[2]
        aesenc          $rndkey1,$inout5
        $movkey         48($key_),$rndkey1
         pxor   $twres,@tweak[2]


@@ -1757,7 +1876,7 @@ $code.=<<___;
        $movkey         -80($key,%rax),$rndkey0
        jnz             .Lxts_enc_loop6
 
        $movkey         -80($key,%rax),$rndkey0
        jnz             .Lxts_enc_loop6
 

-       movdqa  (%r8),$twmask
+       movdqa  (%r8),$twmask                   # start calculating next tweak

        movdqa  $twres,$twtmp
        paddd   $twres,$twres
         aesenc         $rndkey1,$inout0
        movdqa  $twres,$twtmp
        paddd   $twres,$twres
         aesenc         $rndkey1,$inout0


@@ -1851,15 +1970,15 @@ $code.=<<___;
         aesenclast     `16*5`(%rsp),$inout5
        pxor    $twres,@tweak[5]
 
         aesenclast     `16*5`(%rsp),$inout5
        pxor    $twres,@tweak[5]
 

-       lea     `16*6`($out),$out
-       movups  $inout0,`-16*6`($out)           # write output
+       lea     `16*6`($out),$out               # $out+=6*16
+       movups  $inout0,`-16*6`($out)           # store 6 output blocks

        movups  $inout1,`-16*5`($out)
        movups  $inout2,`-16*4`($out)
        movups  $inout3,`-16*3`($out)
        movups  $inout4,`-16*2`($out)
        movups  $inout5,`-16*1`($out)
        sub     \$16*6,$len
        movups  $inout1,`-16*5`($out)
        movups  $inout2,`-16*4`($out)
        movups  $inout3,`-16*3`($out)
        movups  $inout4,`-16*2`($out)
        movups  $inout5,`-16*1`($out)
        sub     \$16*6,$len

-       jnc     .Lxts_enc_grandloop
+       jnc     .Lxts_enc_grandloop             # loop if $len-=6*16 didn't borrow

 
        mov     \$16+96,$rounds
        sub     $rnds_,$rounds
 
        mov     \$16+96,$rounds
        sub     $rnds_,$rounds


@@ -1867,34 +1986,36 @@ $code.=<<___;
        shr     \$4,$rounds                     # restore original value
 
 .Lxts_enc_short:
        shr     \$4,$rounds                     # restore original value
 
 .Lxts_enc_short:

+       # at the point @tweak[0..5] are populated with tweak values

        mov     $rounds,$rnds_                  # backup $rounds
        pxor    $rndkey0,@tweak[0]
        mov     $rounds,$rnds_                  # backup $rounds
        pxor    $rndkey0,@tweak[0]

-       add     \$16*6,$len
-       jz      .Lxts_enc_done
+       add     \$16*6,$len                     # restore real remaining $len
+       jz      .Lxts_enc_done                  # done if ($len==0)

 
        pxor    $rndkey0,@tweak[1]
        cmp     \$0x20,$len
 
        pxor    $rndkey0,@tweak[1]
        cmp     \$0x20,$len

-       jb      .Lxts_enc_one
+       jb      .Lxts_enc_one                   # $len is 1*16

        pxor    $rndkey0,@tweak[2]
        pxor    $rndkey0,@tweak[2]

-       je      .Lxts_enc_two
+       je      .Lxts_enc_two                   # $len is 2*16

 
        pxor    $rndkey0,@tweak[3]
        cmp     \$0x40,$len
 
        pxor    $rndkey0,@tweak[3]
        cmp     \$0x40,$len

-       jb      .Lxts_enc_three
+       jb      .Lxts_enc_three                 # $len is 3*16

        pxor    $rndkey0,@tweak[4]
        pxor    $rndkey0,@tweak[4]

-       je      .Lxts_enc_four
+       je      .Lxts_enc_four                  # $len is 4*16

 
 

-       movdqu  ($inp),$inout0
+       movdqu  ($inp),$inout0                  # $len is 5*16

        movdqu  16*1($inp),$inout1
        movdqu  16*2($inp),$inout2
        pxor    @tweak[0],$inout0
        movdqu  16*3($inp),$inout3
        pxor    @tweak[1],$inout1
        movdqu  16*4($inp),$inout4
        movdqu  16*1($inp),$inout1
        movdqu  16*2($inp),$inout2
        pxor    @tweak[0],$inout0
        movdqu  16*3($inp),$inout3
        pxor    @tweak[1],$inout1
        movdqu  16*4($inp),$inout4

-       lea     16*5($inp),$inp
+       lea     16*5($inp),$inp                 # $inp+=5*16

        pxor    @tweak[2],$inout2
        pxor    @tweak[3],$inout3
        pxor    @tweak[4],$inout4
        pxor    @tweak[2],$inout2
        pxor    @tweak[3],$inout3
        pxor    @tweak[4],$inout4

+       pxor    $inout5,$inout5

 
        call    _aesni_encrypt6
 
 
        call    _aesni_encrypt6
 


@@ -1902,35 +2023,35 @@ $code.=<<___;
        movdqa  @tweak[5],@tweak[0]
        xorps   @tweak[1],$inout1
        xorps   @tweak[2],$inout2
        movdqa  @tweak[5],@tweak[0]
        xorps   @tweak[1],$inout1
        xorps   @tweak[2],$inout2

-       movdqu  $inout0,($out)
+       movdqu  $inout0,($out)                  # store 5 output blocks

        xorps   @tweak[3],$inout3
        movdqu  $inout1,16*1($out)
        xorps   @tweak[4],$inout4
        movdqu  $inout2,16*2($out)
        movdqu  $inout3,16*3($out)
        movdqu  $inout4,16*4($out)
        xorps   @tweak[3],$inout3
        movdqu  $inout1,16*1($out)
        xorps   @tweak[4],$inout4
        movdqu  $inout2,16*2($out)
        movdqu  $inout3,16*3($out)
        movdqu  $inout4,16*4($out)

-       lea     16*5($out),$out
+       lea     16*5($out),$out                 # $out+=5*16

        jmp     .Lxts_enc_done
 
 .align 16
 .Lxts_enc_one:
        movups  ($inp),$inout0
        jmp     .Lxts_enc_done
 
 .align 16
 .Lxts_enc_one:
        movups  ($inp),$inout0

-       lea     16*1($inp),$inp
+       lea     16*1($inp),$inp                 # inp+=1*16

        xorps   @tweak[0],$inout0
 ___
        &aesni_generate1("enc",$key,$rounds);
 $code.=<<___;
        xorps   @tweak[0],$inout0
        movdqa  @tweak[1],@tweak[0]
        xorps   @tweak[0],$inout0
 ___
        &aesni_generate1("enc",$key,$rounds);
 $code.=<<___;
        xorps   @tweak[0],$inout0
        movdqa  @tweak[1],@tweak[0]

-       movups  $inout0,($out)
-       lea     16*1($out),$out
+       movups  $inout0,($out)                  # store one output block
+       lea     16*1($out),$out                 # $out+=1*16

        jmp     .Lxts_enc_done
 
 .align 16
 .Lxts_enc_two:
        movups  ($inp),$inout0
        movups  16($inp),$inout1
        jmp     .Lxts_enc_done
 
 .align 16
 .Lxts_enc_two:
        movups  ($inp),$inout0
        movups  16($inp),$inout1

-       lea     32($inp),$inp
+       lea     32($inp),$inp                   # $inp+=2*16

        xorps   @tweak[0],$inout0
        xorps   @tweak[1],$inout1
 
        xorps   @tweak[0],$inout0
        xorps   @tweak[1],$inout1
 


@@ -1939,9 +2060,9 @@ $code.=<<___;
        xorps   @tweak[0],$inout0
        movdqa  @tweak[2],@tweak[0]
        xorps   @tweak[1],$inout1
        xorps   @tweak[0],$inout0
        movdqa  @tweak[2],@tweak[0]
        xorps   @tweak[1],$inout1

-       movups  $inout0,($out)
+       movups  $inout0,($out)                  # store 2 output blocks

        movups  $inout1,16*1($out)
        movups  $inout1,16*1($out)

-       lea     16*2($out),$out
+       lea     16*2($out),$out                 # $out+=2*16

        jmp     .Lxts_enc_done
 
 .align 16
        jmp     .Lxts_enc_done
 
 .align 16


@@ -1949,7 +2070,7 @@ $code.=<<___;
        movups  ($inp),$inout0
        movups  16*1($inp),$inout1
        movups  16*2($inp),$inout2
        movups  ($inp),$inout0
        movups  16*1($inp),$inout1
        movups  16*2($inp),$inout2

-       lea     16*3($inp),$inp
+       lea     16*3($inp),$inp                 # $inp+=3*16

        xorps   @tweak[0],$inout0
        xorps   @tweak[1],$inout1
        xorps   @tweak[2],$inout2
        xorps   @tweak[0],$inout0
        xorps   @tweak[1],$inout1
        xorps   @tweak[2],$inout2


@@ -1960,10 +2081,10 @@ $code.=<<___;
        movdqa  @tweak[3],@tweak[0]
        xorps   @tweak[1],$inout1
        xorps   @tweak[2],$inout2
        movdqa  @tweak[3],@tweak[0]
        xorps   @tweak[1],$inout1
        xorps   @tweak[2],$inout2

-       movups  $inout0,($out)
+       movups  $inout0,($out)                  # store 3 output blocks

        movups  $inout1,16*1($out)
        movups  $inout2,16*2($out)
        movups  $inout1,16*1($out)
        movups  $inout2,16*2($out)

-       lea     16*3($out),$out
+       lea     16*3($out),$out                 # $out+=3*16

        jmp     .Lxts_enc_done
 
 .align 16
        jmp     .Lxts_enc_done
 
 .align 16


@@ -1973,7 +2094,7 @@ $code.=<<___;
        movups  16*2($inp),$inout2
        xorps   @tweak[0],$inout0
        movups  16*3($inp),$inout3
        movups  16*2($inp),$inout2
        xorps   @tweak[0],$inout0
        movups  16*3($inp),$inout3

-       lea     16*4($inp),$inp
+       lea     16*4($inp),$inp                 # $inp+=4*16

        xorps   @tweak[1],$inout1
        xorps   @tweak[2],$inout2
        xorps   @tweak[3],$inout3
        xorps   @tweak[1],$inout1
        xorps   @tweak[2],$inout2
        xorps   @tweak[3],$inout3


@@ -1984,17 +2105,17 @@ $code.=<<___;
        movdqa  @tweak[4],@tweak[0]
        pxor    @tweak[1],$inout1
        pxor    @tweak[2],$inout2
        movdqa  @tweak[4],@tweak[0]
        pxor    @tweak[1],$inout1
        pxor    @tweak[2],$inout2

-       movdqu  $inout0,($out)
+       movdqu  $inout0,($out)                  # store 4 output blocks

        pxor    @tweak[3],$inout3
        movdqu  $inout1,16*1($out)
        movdqu  $inout2,16*2($out)
        movdqu  $inout3,16*3($out)
        pxor    @tweak[3],$inout3
        movdqu  $inout1,16*1($out)
        movdqu  $inout2,16*2($out)
        movdqu  $inout3,16*3($out)

-       lea     16*4($out),$out
+       lea     16*4($out),$out                 # $out+=4*16

        jmp     .Lxts_enc_done
 
 .align 16
 .Lxts_enc_done:
        jmp     .Lxts_enc_done
 
 .align 16
 .Lxts_enc_done:

-       and     \$15,$len_
+       and     \$15,$len_                      # see if $len%16 is 0

        jz      .Lxts_enc_ret
        mov     $len_,$len
 
        jz      .Lxts_enc_ret
        mov     $len_,$len
 


@@ -2021,18 +2142,60 @@ $code.=<<___;
        movups  $inout0,-16($out)
 
 .Lxts_enc_ret:
        movups  $inout0,-16($out)
 
 .Lxts_enc_ret:

+       xorps   %xmm0,%xmm0                     # clear register bank
+       pxor    %xmm1,%xmm1
+       pxor    %xmm2,%xmm2
+       pxor    %xmm3,%xmm3
+       pxor    %xmm4,%xmm4
+       pxor    %xmm5,%xmm5
+___
+$code.=<<___ if (!$win64);
+       pxor    %xmm6,%xmm6
+       pxor    %xmm7,%xmm7
+       movaps  %xmm0,0x00(%rsp)                # clear stack
+       pxor    %xmm8,%xmm8
+       movaps  %xmm0,0x10(%rsp)
+       pxor    %xmm9,%xmm9
+       movaps  %xmm0,0x20(%rsp)
+       pxor    %xmm10,%xmm10
+       movaps  %xmm0,0x30(%rsp)
+       pxor    %xmm11,%xmm11
+       movaps  %xmm0,0x40(%rsp)
+       pxor    %xmm12,%xmm12
+       movaps  %xmm0,0x50(%rsp)
+       pxor    %xmm13,%xmm13
+       movaps  %xmm0,0x60(%rsp)
+       pxor    %xmm14,%xmm14
+       pxor    %xmm15,%xmm15

 ___
 $code.=<<___ if ($win64);
        movaps  -0xa0(%rbp),%xmm6
 ___
 $code.=<<___ if ($win64);
        movaps  -0xa0(%rbp),%xmm6

+       movaps  %xmm0,-0xa0(%rbp)               # clear stack

        movaps  -0x90(%rbp),%xmm7
        movaps  -0x90(%rbp),%xmm7

+       movaps  %xmm0,-0x90(%rbp)

        movaps  -0x80(%rbp),%xmm8
        movaps  -0x80(%rbp),%xmm8

+       movaps  %xmm0,-0x80(%rbp)

        movaps  -0x70(%rbp),%xmm9
        movaps  -0x70(%rbp),%xmm9

+       movaps  %xmm0,-0x70(%rbp)

        movaps  -0x60(%rbp),%xmm10
        movaps  -0x60(%rbp),%xmm10

+       movaps  %xmm0,-0x60(%rbp)

        movaps  -0x50(%rbp),%xmm11
        movaps  -0x50(%rbp),%xmm11

+       movaps  %xmm0,-0x50(%rbp)

        movaps  -0x40(%rbp),%xmm12
        movaps  -0x40(%rbp),%xmm12

+       movaps  %xmm0,-0x40(%rbp)

        movaps  -0x30(%rbp),%xmm13
        movaps  -0x30(%rbp),%xmm13

+       movaps  %xmm0,-0x30(%rbp)

        movaps  -0x20(%rbp),%xmm14
        movaps  -0x20(%rbp),%xmm14

+       movaps  %xmm0,-0x20(%rbp)

        movaps  -0x10(%rbp),%xmm15
        movaps  -0x10(%rbp),%xmm15

+       movaps  %xmm0,-0x10(%rbp)
+       movaps  %xmm0,0x00(%rsp)
+       movaps  %xmm0,0x10(%rsp)
+       movaps  %xmm0,0x20(%rsp)
+       movaps  %xmm0,0x30(%rsp)
+       movaps  %xmm0,0x40(%rsp)
+       movaps  %xmm0,0x50(%rsp)
+       movaps  %xmm0,0x60(%rsp)

 ___
 $code.=<<___;
        lea     (%rbp),%rsp
 ___
 $code.=<<___;
        lea     (%rbp),%rsp


@@ -2053,7 +2216,7 @@ aesni_xts_decrypt:
        and     \$-16,%rsp      # Linux kernel stack can be incorrectly seeded
 ___
 $code.=<<___ if ($win64);
        and     \$-16,%rsp      # Linux kernel stack can be incorrectly seeded
 ___
 $code.=<<___ if ($win64);

-       movaps  %xmm6,-0xa8(%rax)
+       movaps  %xmm6,-0xa8(%rax)               # offload everything

        movaps  %xmm7,-0x98(%rax)
        movaps  %xmm8,-0x88(%rax)
        movaps  %xmm9,-0x78(%rax)
        movaps  %xmm7,-0x98(%rax)
        movaps  %xmm8,-0x88(%rax)
        movaps  %xmm9,-0x78(%rax)


@@ -2116,7 +2279,7 @@ $code.=<<___;
        movaps  $rndkey1,0x60(%rsp)             # save round[0]^round[last]
 
        sub     \$16*6,$len
        movaps  $rndkey1,0x60(%rsp)             # save round[0]^round[last]
 
        sub     \$16*6,$len

-       jc      .Lxts_dec_short
+       jc      .Lxts_dec_short                 # if $len-=6*16 borrowed

 
        mov     \$16+96,$rounds
        lea     32($key_,$rnds_),$key           # end of key schedule
 
        mov     \$16+96,$rounds
        lea     32($key_,$rnds_),$key           # end of key schedule


@@ -2131,7 +2294,7 @@ $code.=<<___;
        movdqu  `16*0`($inp),$inout0            # load input
        movdqa  $rndkey0,$twmask
        movdqu  `16*1`($inp),$inout1
        movdqu  `16*0`($inp),$inout0            # load input
        movdqa  $rndkey0,$twmask
        movdqu  `16*1`($inp),$inout1

-       pxor    @tweak[0],$inout0
+       pxor    @tweak[0],$inout0               # intput^=tweak^round[0]

        movdqu  `16*2`($inp),$inout2
        pxor    @tweak[1],$inout1
         aesdec         $rndkey1,$inout0
        movdqu  `16*2`($inp),$inout2
        pxor    @tweak[1],$inout1
         aesdec         $rndkey1,$inout0


@@ -2150,7 +2313,7 @@ $code.=<<___;
        lea     `16*6`($inp),$inp
        pxor    $twmask,$inout5
 
        lea     `16*6`($inp),$inp
        pxor    $twmask,$inout5
 

-        pxor   $twres,@tweak[0]
+        pxor   $twres,@tweak[0]                # calclulate tweaks^round[last]

        aesdec          $rndkey1,$inout4
         pxor   $twres,@tweak[1]
         movdqa @tweak[0],`16*0`(%rsp)          # put aside tweaks^last round key
        aesdec          $rndkey1,$inout4
         pxor   $twres,@tweak[1]
         movdqa @tweak[0],`16*0`(%rsp)          # put aside tweaks^last round key


@@ -2194,7 +2357,7 @@ $code.=<<___;
        $movkey         -80($key,%rax),$rndkey0
        jnz             .Lxts_dec_loop6
 
        $movkey         -80($key,%rax),$rndkey0
        jnz             .Lxts_dec_loop6
 

-       movdqa  (%r8),$twmask
+       movdqa  (%r8),$twmask                   # start calculating next tweak

        movdqa  $twres,$twtmp
        paddd   $twres,$twres
         aesdec         $rndkey1,$inout0
        movdqa  $twres,$twtmp
        paddd   $twres,$twres
         aesdec         $rndkey1,$inout0


@@ -2288,15 +2451,15 @@ $code.=<<___;
         aesdeclast     `16*5`(%rsp),$inout5
        pxor    $twres,@tweak[5]
 
         aesdeclast     `16*5`(%rsp),$inout5
        pxor    $twres,@tweak[5]
 

-       lea     `16*6`($out),$out
-       movups  $inout0,`-16*6`($out)           # write output
+       lea     `16*6`($out),$out               # $out+=6*16
+       movups  $inout0,`-16*6`($out)           # store 6 output blocks

        movups  $inout1,`-16*5`($out)
        movups  $inout2,`-16*4`($out)
        movups  $inout3,`-16*3`($out)
        movups  $inout4,`-16*2`($out)
        movups  $inout5,`-16*1`($out)
        sub     \$16*6,$len
        movups  $inout1,`-16*5`($out)
        movups  $inout2,`-16*4`($out)
        movups  $inout3,`-16*3`($out)
        movups  $inout4,`-16*2`($out)
        movups  $inout5,`-16*1`($out)
        sub     \$16*6,$len

-       jnc     .Lxts_dec_grandloop
+       jnc     .Lxts_dec_grandloop             # loop if $len-=6*16 didn't borrow

 
        mov     \$16+96,$rounds
        sub     $rnds_,$rounds
 
        mov     \$16+96,$rounds
        sub     $rnds_,$rounds


@@ -2304,31 +2467,32 @@ $code.=<<___;
        shr     \$4,$rounds                     # restore original value
 
 .Lxts_dec_short:
        shr     \$4,$rounds                     # restore original value
 
 .Lxts_dec_short:

+       # at the point @tweak[0..5] are populated with tweak values

        mov     $rounds,$rnds_                  # backup $rounds
        pxor    $rndkey0,@tweak[0]
        pxor    $rndkey0,@tweak[1]
        mov     $rounds,$rnds_                  # backup $rounds
        pxor    $rndkey0,@tweak[0]
        pxor    $rndkey0,@tweak[1]

-       add     \$16*6,$len
-       jz      .Lxts_dec_done
+       add     \$16*6,$len                     # restore real remaining $len
+       jz      .Lxts_dec_done                  # done if ($len==0)

 
        pxor    $rndkey0,@tweak[2]
        cmp     \$0x20,$len
 
        pxor    $rndkey0,@tweak[2]
        cmp     \$0x20,$len

-       jb      .Lxts_dec_one
+       jb      .Lxts_dec_one                   # $len is 1*16

        pxor    $rndkey0,@tweak[3]
        pxor    $rndkey0,@tweak[3]

-       je      .Lxts_dec_two
+       je      .Lxts_dec_two                   # $len is 2*16

 
        pxor    $rndkey0,@tweak[4]
        cmp     \$0x40,$len
 
        pxor    $rndkey0,@tweak[4]
        cmp     \$0x40,$len

-       jb      .Lxts_dec_three
-       je      .Lxts_dec_four
+       jb      .Lxts_dec_three                 # $len is 3*16
+       je      .Lxts_dec_four                  # $len is 4*16

 
 

-       movdqu  ($inp),$inout0
+       movdqu  ($inp),$inout0                  # $len is 5*16

        movdqu  16*1($inp),$inout1
        movdqu  16*2($inp),$inout2
        pxor    @tweak[0],$inout0
        movdqu  16*3($inp),$inout3
        pxor    @tweak[1],$inout1
        movdqu  16*4($inp),$inout4
        movdqu  16*1($inp),$inout1
        movdqu  16*2($inp),$inout2
        pxor    @tweak[0],$inout0
        movdqu  16*3($inp),$inout3
        pxor    @tweak[1],$inout1
        movdqu  16*4($inp),$inout4

-       lea     16*5($inp),$inp
+       lea     16*5($inp),$inp                 # $inp+=5*16

        pxor    @tweak[2],$inout2
        pxor    @tweak[3],$inout3
        pxor    @tweak[4],$inout4
        pxor    @tweak[2],$inout2
        pxor    @tweak[3],$inout3
        pxor    @tweak[4],$inout4


@@ -2338,7 +2502,7 @@ $code.=<<___;
        xorps   @tweak[0],$inout0
        xorps   @tweak[1],$inout1
        xorps   @tweak[2],$inout2
        xorps   @tweak[0],$inout0
        xorps   @tweak[1],$inout1
        xorps   @tweak[2],$inout2

-       movdqu  $inout0,($out)
+       movdqu  $inout0,($out)                  # store 5 output blocks

        xorps   @tweak[3],$inout3
        movdqu  $inout1,16*1($out)
        xorps   @tweak[4],$inout4
        xorps   @tweak[3],$inout3
        movdqu  $inout1,16*1($out)
        xorps   @tweak[4],$inout4


@@ -2347,7 +2511,7 @@ $code.=<<___;
        movdqu  $inout3,16*3($out)
         pcmpgtd        @tweak[5],$twtmp
        movdqu  $inout4,16*4($out)
        movdqu  $inout3,16*3($out)
         pcmpgtd        @tweak[5],$twtmp
        movdqu  $inout4,16*4($out)

-       lea     16*5($out),$out
+       lea     16*5($out),$out                 # $out+=5*16

         pshufd         \$0x13,$twtmp,@tweak[1] # $twres
        and     \$15,$len_
        jz      .Lxts_dec_ret
         pshufd         \$0x13,$twtmp,@tweak[1] # $twres
        and     \$15,$len_
        jz      .Lxts_dec_ret


@@ -2361,23 +2525,23 @@ $code.=<<___;
 .align 16
 .Lxts_dec_one:
        movups  ($inp),$inout0
 .align 16
 .Lxts_dec_one:
        movups  ($inp),$inout0

-       lea     16*1($inp),$inp
+       lea     16*1($inp),$inp                 # $inp+=1*16

        xorps   @tweak[0],$inout0
 ___
        &aesni_generate1("dec",$key,$rounds);
 $code.=<<___;
        xorps   @tweak[0],$inout0
        movdqa  @tweak[1],@tweak[0]
        xorps   @tweak[0],$inout0
 ___
        &aesni_generate1("dec",$key,$rounds);
 $code.=<<___;
        xorps   @tweak[0],$inout0
        movdqa  @tweak[1],@tweak[0]

-       movups  $inout0,($out)
+       movups  $inout0,($out)                  # store one output block

        movdqa  @tweak[2],@tweak[1]
        movdqa  @tweak[2],@tweak[1]

-       lea     16*1($out),$out
+       lea     16*1($out),$out                 # $out+=1*16

        jmp     .Lxts_dec_done
 
 .align 16
 .Lxts_dec_two:
        movups  ($inp),$inout0
        movups  16($inp),$inout1
        jmp     .Lxts_dec_done
 
 .align 16
 .Lxts_dec_two:
        movups  ($inp),$inout0
        movups  16($inp),$inout1

-       lea     32($inp),$inp
+       lea     32($inp),$inp                   # $inp+=2*16

        xorps   @tweak[0],$inout0
        xorps   @tweak[1],$inout1
 
        xorps   @tweak[0],$inout0
        xorps   @tweak[1],$inout1
 


@@ -2387,9 +2551,9 @@ $code.=<<___;
        movdqa  @tweak[2],@tweak[0]
        xorps   @tweak[1],$inout1
        movdqa  @tweak[3],@tweak[1]
        movdqa  @tweak[2],@tweak[0]
        xorps   @tweak[1],$inout1
        movdqa  @tweak[3],@tweak[1]

-       movups  $inout0,($out)
+       movups  $inout0,($out)                  # store 2 output blocks

        movups  $inout1,16*1($out)
        movups  $inout1,16*1($out)

-       lea     16*2($out),$out
+       lea     16*2($out),$out                 # $out+=2*16

        jmp     .Lxts_dec_done
 
 .align 16
        jmp     .Lxts_dec_done
 
 .align 16


@@ -2397,7 +2561,7 @@ $code.=<<___;
        movups  ($inp),$inout0
        movups  16*1($inp),$inout1
        movups  16*2($inp),$inout2
        movups  ($inp),$inout0
        movups  16*1($inp),$inout1
        movups  16*2($inp),$inout2

-       lea     16*3($inp),$inp
+       lea     16*3($inp),$inp                 # $inp+=3*16

        xorps   @tweak[0],$inout0
        xorps   @tweak[1],$inout1
        xorps   @tweak[2],$inout2
        xorps   @tweak[0],$inout0
        xorps   @tweak[1],$inout1
        xorps   @tweak[2],$inout2


@@ -2409,10 +2573,10 @@ $code.=<<___;
        xorps   @tweak[1],$inout1
        movdqa  @tweak[4],@tweak[1]
        xorps   @tweak[2],$inout2
        xorps   @tweak[1],$inout1
        movdqa  @tweak[4],@tweak[1]
        xorps   @tweak[2],$inout2

-       movups  $inout0,($out)
+       movups  $inout0,($out)                  # store 3 output blocks

        movups  $inout1,16*1($out)
        movups  $inout2,16*2($out)
        movups  $inout1,16*1($out)
        movups  $inout2,16*2($out)

-       lea     16*3($out),$out
+       lea     16*3($out),$out                 # $out+=3*16

        jmp     .Lxts_dec_done
 
 .align 16
        jmp     .Lxts_dec_done
 
 .align 16


@@ -2422,7 +2586,7 @@ $code.=<<___;
        movups  16*2($inp),$inout2
        xorps   @tweak[0],$inout0
        movups  16*3($inp),$inout3
        movups  16*2($inp),$inout2
        xorps   @tweak[0],$inout0
        movups  16*3($inp),$inout3

-       lea     16*4($inp),$inp
+       lea     16*4($inp),$inp                 # $inp+=4*16

        xorps   @tweak[1],$inout1
        xorps   @tweak[2],$inout2
        xorps   @tweak[3],$inout3
        xorps   @tweak[1],$inout1
        xorps   @tweak[2],$inout2
        xorps   @tweak[3],$inout3


@@ -2434,17 +2598,17 @@ $code.=<<___;
        pxor    @tweak[1],$inout1
        movdqa  @tweak[5],@tweak[1]
        pxor    @tweak[2],$inout2
        pxor    @tweak[1],$inout1
        movdqa  @tweak[5],@tweak[1]
        pxor    @tweak[2],$inout2

-       movdqu  $inout0,($out)
+       movdqu  $inout0,($out)                  # store 4 output blocks

        pxor    @tweak[3],$inout3
        movdqu  $inout1,16*1($out)
        movdqu  $inout2,16*2($out)
        movdqu  $inout3,16*3($out)
        pxor    @tweak[3],$inout3
        movdqu  $inout1,16*1($out)
        movdqu  $inout2,16*2($out)
        movdqu  $inout3,16*3($out)

-       lea     16*4($out),$out
+       lea     16*4($out),$out                 # $out+=4*16

        jmp     .Lxts_dec_done
 
 .align 16
 .Lxts_dec_done:
        jmp     .Lxts_dec_done
 
 .align 16
 .Lxts_dec_done:

-       and     \$15,$len_
+       and     \$15,$len_                      # see if $len%16 is 0

        jz      .Lxts_dec_ret
 .Lxts_dec_done2:
        mov     $len_,$len
        jz      .Lxts_dec_ret
 .Lxts_dec_done2:
        mov     $len_,$len


@@ -2482,18 +2646,60 @@ $code.=<<___;
        movups  $inout0,($out)
 
 .Lxts_dec_ret:
        movups  $inout0,($out)
 
 .Lxts_dec_ret:

+       xorps   %xmm0,%xmm0                     # clear register bank
+       pxor    %xmm1,%xmm1
+       pxor    %xmm2,%xmm2
+       pxor    %xmm3,%xmm3
+       pxor    %xmm4,%xmm4
+       pxor    %xmm5,%xmm5
+___
+$code.=<<___ if (!$win64);
+       pxor    %xmm6,%xmm6
+       pxor    %xmm7,%xmm7
+       movaps  %xmm0,0x00(%rsp)                # clear stack
+       pxor    %xmm8,%xmm8
+       movaps  %xmm0,0x10(%rsp)
+       pxor    %xmm9,%xmm9
+       movaps  %xmm0,0x20(%rsp)
+       pxor    %xmm10,%xmm10
+       movaps  %xmm0,0x30(%rsp)
+       pxor    %xmm11,%xmm11
+       movaps  %xmm0,0x40(%rsp)
+       pxor    %xmm12,%xmm12
+       movaps  %xmm0,0x50(%rsp)
+       pxor    %xmm13,%xmm13
+       movaps  %xmm0,0x60(%rsp)
+       pxor    %xmm14,%xmm14
+       pxor    %xmm15,%xmm15

 ___
 $code.=<<___ if ($win64);
        movaps  -0xa0(%rbp),%xmm6
 ___
 $code.=<<___ if ($win64);
        movaps  -0xa0(%rbp),%xmm6

+       movaps  %xmm0,-0xa0(%rbp)               # clear stack

        movaps  -0x90(%rbp),%xmm7
        movaps  -0x90(%rbp),%xmm7

+       movaps  %xmm0,-0x90(%rbp)

        movaps  -0x80(%rbp),%xmm8
        movaps  -0x80(%rbp),%xmm8

+       movaps  %xmm0,-0x80(%rbp)

        movaps  -0x70(%rbp),%xmm9
        movaps  -0x70(%rbp),%xmm9

+       movaps  %xmm0,-0x70(%rbp)

        movaps  -0x60(%rbp),%xmm10
        movaps  -0x60(%rbp),%xmm10

+       movaps  %xmm0,-0x60(%rbp)

        movaps  -0x50(%rbp),%xmm11
        movaps  -0x50(%rbp),%xmm11

+       movaps  %xmm0,-0x50(%rbp)

        movaps  -0x40(%rbp),%xmm12
        movaps  -0x40(%rbp),%xmm12

+       movaps  %xmm0,-0x40(%rbp)

        movaps  -0x30(%rbp),%xmm13
        movaps  -0x30(%rbp),%xmm13

+       movaps  %xmm0,-0x30(%rbp)

        movaps  -0x20(%rbp),%xmm14
        movaps  -0x20(%rbp),%xmm14

+       movaps  %xmm0,-0x20(%rbp)

        movaps  -0x10(%rbp),%xmm15
        movaps  -0x10(%rbp),%xmm15

+       movaps  %xmm0,-0x10(%rbp)
+       movaps  %xmm0,0x00(%rsp)
+       movaps  %xmm0,0x10(%rsp)
+       movaps  %xmm0,0x20(%rsp)
+       movaps  %xmm0,0x30(%rsp)
+       movaps  %xmm0,0x40(%rsp)
+       movaps  %xmm0,0x50(%rsp)
+       movaps  %xmm0,0x60(%rsp)

 ___
 $code.=<<___;
        lea     (%rbp),%rsp
 ___
 $code.=<<___;
        lea     (%rbp),%rsp


@@ -2548,7 +2754,11 @@ $code.=<<___;
        jnc     .Lcbc_enc_loop
        add     \$16,$len
        jnz     .Lcbc_enc_tail
        jnc     .Lcbc_enc_loop
        add     \$16,$len
        jnz     .Lcbc_enc_tail

+        pxor   $rndkey0,$rndkey0       # clear register bank
+        pxor   $rndkey1,$rndkey1

        movups  $inout0,($ivp)
        movups  $inout0,($ivp)

+        pxor   $inout0,$inout0
+        pxor   $inout1,$inout1

        jmp     .Lcbc_ret
 
 .Lcbc_enc_tail:
        jmp     .Lcbc_ret
 
 .Lcbc_enc_tail:


@@ -2568,6 +2778,27 @@ $code.=<<___;
 \f#--------------------------- CBC DECRYPT ------------------------------#
 .align 16
 .Lcbc_decrypt:
 \f#--------------------------- CBC DECRYPT ------------------------------#
 .align 16
 .Lcbc_decrypt:

+       cmp     \$16,$len
+       jne     .Lcbc_decrypt_bulk
+
+       # handle single block without allocating stack frame,
+       # useful in ciphertext stealing mode
+       movdqu  ($inp),$inout0          # load input
+       movdqu  ($ivp),$inout1          # load iv
+       movdqa  $inout0,$inout2         # future iv
+___
+       &aesni_generate1("dec",$key,$rnds_);
+$code.=<<___;
+        pxor   $rndkey0,$rndkey0       # clear register bank
+        pxor   $rndkey1,$rndkey1
+       movdqu  $inout2,($ivp)          # store iv
+       xorps   $inout1,$inout0         # ^=iv
+        pxor   $inout1,$inout1
+       movups  $inout0,($out)          # store output
+        pxor   $inout0,$inout0
+       jmp     .Lcbc_ret
+.align 16
+.Lcbc_decrypt_bulk:

        lea     (%rsp),%rax
        push    %rbp
        sub     \$$frame_size,%rsp
        lea     (%rsp),%rax
        push    %rbp
        sub     \$$frame_size,%rsp


@@ -2609,11 +2840,11 @@ $code.=<<___;
        cmp     \$0x70,$len
        jbe     .Lcbc_dec_six_or_seven
 
        cmp     \$0x70,$len
        jbe     .Lcbc_dec_six_or_seven
 

-       and     \$`1<<26|1<<22`,%r9d    # isolate XSAVE+MOVBE   
-       sub     \$0x50,$len
+       and     \$`1<<26|1<<22`,%r9d    # isolate XSAVE+MOVBE
+       sub     \$0x50,$len             # $len is biased by -5*16

        cmp     \$`1<<22`,%r9d          # check for MOVBE without XSAVE
        cmp     \$`1<<22`,%r9d          # check for MOVBE without XSAVE

-       je      .Lcbc_dec_loop6_enter
-       sub     \$0x20,$len
+       je      .Lcbc_dec_loop6_enter   # [which denotes Atom Silvermont]
+       sub     \$0x20,$len             # $len is biased by -7*16

        lea     0x70($key),$key         # size optimization
        jmp     .Lcbc_dec_loop8_enter
 .align 16
        lea     0x70($key),$key         # size optimization
        jmp     .Lcbc_dec_loop8_enter
 .align 16


@@ -2740,7 +2971,7 @@ $code.=<<___;
        movaps  $inout7,$inout0
        lea     -0x70($key),$key
        add     \$0x70,$len
        movaps  $inout7,$inout0
        lea     -0x70($key),$key
        add     \$0x70,$len

-       jle     .Lcbc_dec_tail_collected
+       jle     .Lcbc_dec_clear_tail_collected

        movups  $inout7,($out)
        lea     0x10($out),$out
        cmp     \$0x50,$len
        movups  $inout7,($out)
        lea     0x10($out),$out
        cmp     \$0x50,$len


@@ -2759,14 +2990,19 @@ $code.=<<___;
        movdqu  $inout0,($out)
        pxor    $in1,$inout2
        movdqu  $inout1,0x10($out)
        movdqu  $inout0,($out)
        pxor    $in1,$inout2
        movdqu  $inout1,0x10($out)

+        pxor   $inout1,$inout1         # clear register bank

        pxor    $in2,$inout3
        movdqu  $inout2,0x20($out)
        pxor    $in2,$inout3
        movdqu  $inout2,0x20($out)

+        pxor   $inout2,$inout2

        pxor    $in3,$inout4
        movdqu  $inout3,0x30($out)
        pxor    $in3,$inout4
        movdqu  $inout3,0x30($out)

+        pxor   $inout3,$inout3

        pxor    $in4,$inout5
        movdqu  $inout4,0x40($out)
        pxor    $in4,$inout5
        movdqu  $inout4,0x40($out)

+        pxor   $inout4,$inout4

        lea     0x50($out),$out
        movdqa  $inout5,$inout0
        lea     0x50($out),$out
        movdqa  $inout5,$inout0

+        pxor   $inout5,$inout5

        jmp     .Lcbc_dec_tail_collected
 
 .align 16
        jmp     .Lcbc_dec_tail_collected
 
 .align 16


@@ -2781,16 +3017,23 @@ $code.=<<___;
        movdqu  $inout0,($out)
        pxor    $in1,$inout2
        movdqu  $inout1,0x10($out)
        movdqu  $inout0,($out)
        pxor    $in1,$inout2
        movdqu  $inout1,0x10($out)

+        pxor   $inout1,$inout1         # clear register bank

        pxor    $in2,$inout3
        movdqu  $inout2,0x20($out)
        pxor    $in2,$inout3
        movdqu  $inout2,0x20($out)

+        pxor   $inout2,$inout2

        pxor    $in3,$inout4
        movdqu  $inout3,0x30($out)
        pxor    $in3,$inout4
        movdqu  $inout3,0x30($out)

+        pxor   $inout3,$inout3

        pxor    $in4,$inout5
        movdqu  $inout4,0x40($out)
        pxor    $in4,$inout5
        movdqu  $inout4,0x40($out)

+        pxor   $inout4,$inout4

        pxor    $inout7,$inout6
        movdqu  $inout5,0x50($out)
        pxor    $inout7,$inout6
        movdqu  $inout5,0x50($out)

+        pxor   $inout5,$inout5

        lea     0x60($out),$out
        movdqa  $inout6,$inout0
        lea     0x60($out),$out
        movdqa  $inout6,$inout0

+        pxor   $inout6,$inout6
+        pxor   $inout7,$inout7

        jmp     .Lcbc_dec_tail_collected
 
 .align 16
        jmp     .Lcbc_dec_tail_collected
 
 .align 16


@@ -2834,31 +3077,31 @@ $code.=<<___;
 
        movdqa  $inout5,$inout0
        add     \$0x50,$len
 
        movdqa  $inout5,$inout0
        add     \$0x50,$len

-       jle     .Lcbc_dec_tail_collected
+       jle     .Lcbc_dec_clear_tail_collected

        movups  $inout5,($out)
        lea     0x10($out),$out
 
 .Lcbc_dec_tail:
        movups  ($inp),$inout0
        sub     \$0x10,$len
        movups  $inout5,($out)
        lea     0x10($out),$out
 
 .Lcbc_dec_tail:
        movups  ($inp),$inout0
        sub     \$0x10,$len

-       jbe     .Lcbc_dec_one
+       jbe     .Lcbc_dec_one           # $len is 1*16 or less

 
        movups  0x10($inp),$inout1
        movaps  $inout0,$in0
        sub     \$0x10,$len
 
        movups  0x10($inp),$inout1
        movaps  $inout0,$in0
        sub     \$0x10,$len

-       jbe     .Lcbc_dec_two
+       jbe     .Lcbc_dec_two           # $len is 2*16 or less

 
        movups  0x20($inp),$inout2
        movaps  $inout1,$in1
        sub     \$0x10,$len
 
        movups  0x20($inp),$inout2
        movaps  $inout1,$in1
        sub     \$0x10,$len

-       jbe     .Lcbc_dec_three
+       jbe     .Lcbc_dec_three         # $len is 3*16 or less

 
        movups  0x30($inp),$inout3
        movaps  $inout2,$in2
        sub     \$0x10,$len
 
        movups  0x30($inp),$inout3
        movaps  $inout2,$in2
        sub     \$0x10,$len

-       jbe     .Lcbc_dec_four
+       jbe     .Lcbc_dec_four          # $len is 4*16 or less

 
 

-       movups  0x40($inp),$inout4
+       movups  0x40($inp),$inout4      # $len is 5*16 or less

        movaps  $inout3,$in3
        movaps  $inout4,$in4
        xorps   $inout5,$inout5
        movaps  $inout3,$in3
        movaps  $inout4,$in4
        xorps   $inout5,$inout5


@@ -2869,12 +3112,17 @@ $code.=<<___;
        movdqu  $inout0,($out)
        pxor    $in1,$inout2
        movdqu  $inout1,0x10($out)
        movdqu  $inout0,($out)
        pxor    $in1,$inout2
        movdqu  $inout1,0x10($out)

+        pxor   $inout1,$inout1         # clear register bank

        pxor    $in2,$inout3
        movdqu  $inout2,0x20($out)
        pxor    $in2,$inout3
        movdqu  $inout2,0x20($out)

+        pxor   $inout2,$inout2

        pxor    $in3,$inout4
        movdqu  $inout3,0x30($out)
        pxor    $in3,$inout4
        movdqu  $inout3,0x30($out)

+        pxor   $inout3,$inout3

        lea     0x40($out),$out
        movdqa  $inout4,$inout0
        lea     0x40($out),$out
        movdqa  $inout4,$inout0

+        pxor   $inout4,$inout4
+        pxor   $inout5,$inout5

        sub     \$0x10,$len
        jmp     .Lcbc_dec_tail_collected
 
        sub     \$0x10,$len
        jmp     .Lcbc_dec_tail_collected
 


@@ -2896,6 +3144,7 @@ $code.=<<___;
        pxor    $in0,$inout1
        movdqu  $inout0,($out)
        movdqa  $inout1,$inout0
        pxor    $in0,$inout1
        movdqu  $inout0,($out)
        movdqa  $inout1,$inout0

+        pxor   $inout1,$inout1         # clear register bank

        lea     0x10($out),$out
        jmp     .Lcbc_dec_tail_collected
 .align 16
        lea     0x10($out),$out
        jmp     .Lcbc_dec_tail_collected
 .align 16


@@ -2908,7 +3157,9 @@ $code.=<<___;
        movdqu  $inout0,($out)
        pxor    $in1,$inout2
        movdqu  $inout1,0x10($out)
        movdqu  $inout0,($out)
        pxor    $in1,$inout2
        movdqu  $inout1,0x10($out)

+        pxor   $inout1,$inout1         # clear register bank

        movdqa  $inout2,$inout0
        movdqa  $inout2,$inout0

+        pxor   $inout2,$inout2

        lea     0x20($out),$out
        jmp     .Lcbc_dec_tail_collected
 .align 16
        lea     0x20($out),$out
        jmp     .Lcbc_dec_tail_collected
 .align 16


@@ -2921,41 +3172,71 @@ $code.=<<___;
        movdqu  $inout0,($out)
        pxor    $in1,$inout2
        movdqu  $inout1,0x10($out)
        movdqu  $inout0,($out)
        pxor    $in1,$inout2
        movdqu  $inout1,0x10($out)

+        pxor   $inout1,$inout1         # clear register bank

        pxor    $in2,$inout3
        movdqu  $inout2,0x20($out)
        pxor    $in2,$inout3
        movdqu  $inout2,0x20($out)

+        pxor   $inout2,$inout2

        movdqa  $inout3,$inout0
        movdqa  $inout3,$inout0

+        pxor   $inout3,$inout3

        lea     0x30($out),$out
        jmp     .Lcbc_dec_tail_collected
 
 .align 16
        lea     0x30($out),$out
        jmp     .Lcbc_dec_tail_collected
 
 .align 16

+.Lcbc_dec_clear_tail_collected:
+       pxor    $inout1,$inout1         # clear register bank
+       pxor    $inout2,$inout2
+       pxor    $inout3,$inout3
+___
+$code.=<<___ if (!$win64);
+       pxor    $inout4,$inout4         # %xmm6..9
+       pxor    $inout5,$inout5
+       pxor    $inout6,$inout6
+       pxor    $inout7,$inout7
+___
+$code.=<<___;

 .Lcbc_dec_tail_collected:
        movups  $iv,($ivp)
        and     \$15,$len
        jnz     .Lcbc_dec_tail_partial
        movups  $inout0,($out)
 .Lcbc_dec_tail_collected:
        movups  $iv,($ivp)
        and     \$15,$len
        jnz     .Lcbc_dec_tail_partial
        movups  $inout0,($out)

+       pxor    $inout0,$inout0

        jmp     .Lcbc_dec_ret
 .align 16
 .Lcbc_dec_tail_partial:
        movaps  $inout0,(%rsp)
        jmp     .Lcbc_dec_ret
 .align 16
 .Lcbc_dec_tail_partial:
        movaps  $inout0,(%rsp)

+       pxor    $inout0,$inout0

        mov     \$16,%rcx
        mov     $out,%rdi
        sub     $len,%rcx
        lea     (%rsp),%rsi
        mov     \$16,%rcx
        mov     $out,%rdi
        sub     $len,%rcx
        lea     (%rsp),%rsi

-       .long   0x9066A4F3      # rep movsb
+       .long   0x9066A4F3              # rep movsb
+       movdqa  $inout0,(%rsp)

 
 .Lcbc_dec_ret:
 
 .Lcbc_dec_ret:

+       xorps   $rndkey0,$rndkey0       # %xmm0
+       pxor    $rndkey1,$rndkey1

 ___
 $code.=<<___ if ($win64);
        movaps  0x10(%rsp),%xmm6
 ___
 $code.=<<___ if ($win64);
        movaps  0x10(%rsp),%xmm6

+       movaps  %xmm0,0x10(%rsp)        # clear stack

        movaps  0x20(%rsp),%xmm7
        movaps  0x20(%rsp),%xmm7

+       movaps  %xmm0,0x20(%rsp)

        movaps  0x30(%rsp),%xmm8
        movaps  0x30(%rsp),%xmm8

+       movaps  %xmm0,0x30(%rsp)

        movaps  0x40(%rsp),%xmm9
        movaps  0x40(%rsp),%xmm9

+       movaps  %xmm0,0x40(%rsp)

        movaps  0x50(%rsp),%xmm10
        movaps  0x50(%rsp),%xmm10

+       movaps  %xmm0,0x50(%rsp)

        movaps  0x60(%rsp),%xmm11
        movaps  0x60(%rsp),%xmm11

+       movaps  %xmm0,0x60(%rsp)

        movaps  0x70(%rsp),%xmm12
        movaps  0x70(%rsp),%xmm12

+       movaps  %xmm0,0x70(%rsp)

        movaps  0x80(%rsp),%xmm13
        movaps  0x80(%rsp),%xmm13

+       movaps  %xmm0,0x80(%rsp)

        movaps  0x90(%rsp),%xmm14
        movaps  0x90(%rsp),%xmm14

+       movaps  %xmm0,0x90(%rsp)

        movaps  0xa0(%rsp),%xmm15
        movaps  0xa0(%rsp),%xmm15

+       movaps  %xmm0,0xa0(%rsp)

 ___
 $code.=<<___;
        lea     (%rbp),%rsp
 ___
 $code.=<<___;
        lea     (%rbp),%rsp


@@ -2965,8 +3246,15 @@ $code.=<<___;
 .size  ${PREFIX}_cbc_encrypt,.-${PREFIX}_cbc_encrypt
 ___
 } \f
 .size  ${PREFIX}_cbc_encrypt,.-${PREFIX}_cbc_encrypt
 ___
 } \f

-# int $PREFIX_set_[en|de]crypt_key (const unsigned char *userKey,
+# int ${PREFIX}_set_decrypt_key(const unsigned char *inp,

 #                              int bits, AES_KEY *key)
 #                              int bits, AES_KEY *key)

+#
+# input:       $inp    user-supplied key
+#              $bits   $inp length in bits
+#              $key    pointer to key schedule
+# output:      %eax    0 denoting success, -1 or -2 - failure (see C)
+#              *$key   key schedule
+#

 { my ($inp,$bits,$key) = @_4args;
   $bits =~ s/%r/%e/;
 
 { my ($inp,$bits,$key) = @_4args;
   $bits =~ s/%r/%e/;
 


@@ -3003,7 +3291,9 @@ ${PREFIX}_set_decrypt_key:
 
        $movkey ($key),%xmm0            # inverse middle
        aesimc  %xmm0,%xmm0
 
        $movkey ($key),%xmm0            # inverse middle
        aesimc  %xmm0,%xmm0

+       pxor    %xmm1,%xmm1

        $movkey %xmm0,($inp)
        $movkey %xmm0,($inp)

+       pxor    %xmm0,%xmm0

 .Ldec_key_ret:
        add     \$8,%rsp
        ret
 .Ldec_key_ret:
        add     \$8,%rsp
        ret


@@ -3020,6 +3310,22 @@ ___
 # Agressively optimized in respect to aeskeygenassist's critical path
 # and is contained in %xmm0-5 to meet Win64 ABI requirement.
 #
 # Agressively optimized in respect to aeskeygenassist's critical path
 # and is contained in %xmm0-5 to meet Win64 ABI requirement.
 #

+# int ${PREFIX}_set_encrypt_key(const unsigned char *inp,
+#                              int bits, AES_KEY * const key);
+#
+# input:       $inp    user-supplied key
+#              $bits   $inp length in bits
+#              $key    pointer to key schedule
+# output:      %eax    0 denoting success, -1 or -2 - failure (see C)
+#              $bits   rounds-1 (used in aesni_set_decrypt_key)
+#              *$key   key schedule
+#              $key    pointer to key schedule (used in
+#                      aesni_set_decrypt_key)
+#
+# Subroutine is frame-less, which means that only volatile registers
+# are used. Note that it's declared "abi-omnipotent", which means that
+# amount of volatile registers is smaller on Windows.
+#

 $code.=<<___;
 .globl ${PREFIX}_set_encrypt_key
 .type  ${PREFIX}_set_encrypt_key,\@abi-omnipotent
 $code.=<<___;
 .globl ${PREFIX}_set_encrypt_key
 .type  ${PREFIX}_set_encrypt_key,\@abi-omnipotent


@@ -3033,9 +3339,11 @@ __aesni_set_encrypt_key:
        test    $key,$key
        jz      .Lenc_key_ret
 
        test    $key,$key
        jz      .Lenc_key_ret
 

+       mov     \$`1<<28|1<<11`,%r10d   # AVX and XOP bits

        movups  ($inp),%xmm0            # pull first 128 bits of *userKey
        xorps   %xmm4,%xmm4             # low dword of xmm4 is assumed 0
        movups  ($inp),%xmm0            # pull first 128 bits of *userKey
        xorps   %xmm4,%xmm4             # low dword of xmm4 is assumed 0

-       lea     16($key),%rax
+       and     OPENSSL_ia32cap_P+4(%rip),%r10d
+       lea     16($key),%rax           # %rax is used as modifiable copy of $key

        cmp     \$256,$bits
        je      .L14rounds
        cmp     \$192,$bits
        cmp     \$256,$bits
        je      .L14rounds
        cmp     \$192,$bits


@@ -3045,6 +3353,9 @@ __aesni_set_encrypt_key:
 
 .L10rounds:
        mov     \$9,$bits                       # 10 rounds for 128-bit key
 
 .L10rounds:
        mov     \$9,$bits                       # 10 rounds for 128-bit key

+       cmp     \$`1<<28`,%r10d                 # AVX, bit no XOP
+       je      .L10rounds_alt
+

        $movkey %xmm0,($key)                    # round 0
        aeskeygenassist \$0x1,%xmm0,%xmm1       # round 1
        call            .Lkey_expansion_128_cold
        $movkey %xmm0,($key)                    # round 0
        aeskeygenassist \$0x1,%xmm0,%xmm1       # round 1
        call            .Lkey_expansion_128_cold


@@ -3071,10 +3382,80 @@ __aesni_set_encrypt_key:
        xor     %eax,%eax
        jmp     .Lenc_key_ret
 
        xor     %eax,%eax
        jmp     .Lenc_key_ret
 

+.align 16
+.L10rounds_alt:
+       movdqa  .Lkey_rotate(%rip),%xmm5
+       mov     \$8,%r10d
+       movdqa  .Lkey_rcon1(%rip),%xmm4
+       movdqa  %xmm0,%xmm2
+       movdqu  %xmm0,($key)
+       jmp     .Loop_key128
+
+.align 16
+.Loop_key128:
+       pshufb          %xmm5,%xmm0
+       aesenclast      %xmm4,%xmm0
+       pslld           \$1,%xmm4
+       lea             16(%rax),%rax
+
+       movdqa          %xmm2,%xmm3
+       pslldq          \$4,%xmm2
+       pxor            %xmm2,%xmm3
+       pslldq          \$4,%xmm2
+       pxor            %xmm2,%xmm3
+       pslldq          \$4,%xmm2
+       pxor            %xmm3,%xmm2
+
+       pxor            %xmm2,%xmm0
+       movdqu          %xmm0,-16(%rax)
+       movdqa          %xmm0,%xmm2
+
+       dec     %r10d
+       jnz     .Loop_key128
+
+       movdqa          .Lkey_rcon1b(%rip),%xmm4
+
+       pshufb          %xmm5,%xmm0
+       aesenclast      %xmm4,%xmm0
+       pslld           \$1,%xmm4
+
+       movdqa          %xmm2,%xmm3
+       pslldq          \$4,%xmm2
+       pxor            %xmm2,%xmm3
+       pslldq          \$4,%xmm2
+       pxor            %xmm2,%xmm3
+       pslldq          \$4,%xmm2
+       pxor            %xmm3,%xmm2
+
+       pxor            %xmm2,%xmm0
+       movdqu          %xmm0,(%rax)
+
+       movdqa          %xmm0,%xmm2
+       pshufb          %xmm5,%xmm0
+       aesenclast      %xmm4,%xmm0
+
+       movdqa          %xmm2,%xmm3
+       pslldq          \$4,%xmm2
+       pxor            %xmm2,%xmm3
+       pslldq          \$4,%xmm2
+       pxor            %xmm2,%xmm3
+       pslldq          \$4,%xmm2
+       pxor            %xmm3,%xmm2
+
+       pxor            %xmm2,%xmm0
+       movdqu          %xmm0,16(%rax)
+
+       mov     $bits,96(%rax)  # 240($key)
+       xor     %eax,%eax
+       jmp     .Lenc_key_ret
+

 .align 16
 .L12rounds:
        movq    16($inp),%xmm2                  # remaining 1/3 of *userKey
        mov     \$11,$bits                      # 12 rounds for 192
 .align 16
 .L12rounds:
        movq    16($inp),%xmm2                  # remaining 1/3 of *userKey
        mov     \$11,$bits                      # 12 rounds for 192

+       cmp     \$`1<<28`,%r10d                 # AVX, but no XOP
+       je      .L12rounds_alt
+

        $movkey %xmm0,($key)                    # round 0
        aeskeygenassist \$0x1,%xmm2,%xmm1       # round 1,2
        call            .Lkey_expansion_192a_cold
        $movkey %xmm0,($key)                    # round 0
        aeskeygenassist \$0x1,%xmm2,%xmm1       # round 1,2
        call            .Lkey_expansion_192a_cold


@@ -3097,11 +3478,55 @@ __aesni_set_encrypt_key:
        xor     %rax, %rax
        jmp     .Lenc_key_ret
 
        xor     %rax, %rax
        jmp     .Lenc_key_ret
 

+.align 16
+.L12rounds_alt:
+       movdqa  .Lkey_rotate192(%rip),%xmm5
+       movdqa  .Lkey_rcon1(%rip),%xmm4
+       mov     \$8,%r10d
+       movdqu  %xmm0,($key)
+       jmp     .Loop_key192
+
+.align 16
+.Loop_key192:
+       movq            %xmm2,0(%rax)
+       movdqa          %xmm2,%xmm1
+       pshufb          %xmm5,%xmm2
+       aesenclast      %xmm4,%xmm2
+       pslld           \$1, %xmm4
+       lea             24(%rax),%rax
+
+       movdqa          %xmm0,%xmm3
+       pslldq          \$4,%xmm0
+       pxor            %xmm0,%xmm3
+       pslldq          \$4,%xmm0
+       pxor            %xmm0,%xmm3
+       pslldq          \$4,%xmm0
+       pxor            %xmm3,%xmm0
+
+       pshufd          \$0xff,%xmm0,%xmm3
+       pxor            %xmm1,%xmm3
+       pslldq          \$4,%xmm1
+       pxor            %xmm1,%xmm3
+
+       pxor            %xmm2,%xmm0
+       pxor            %xmm3,%xmm2
+       movdqu          %xmm0,-16(%rax)
+
+       dec     %r10d
+       jnz     .Loop_key192
+
+       mov     $bits,32(%rax)  # 240($key)
+       xor     %eax,%eax
+       jmp     .Lenc_key_ret
+

 .align 16
 .L14rounds:
        movups  16($inp),%xmm2                  # remaning half of *userKey
        mov     \$13,$bits                      # 14 rounds for 256
        lea     16(%rax),%rax
 .align 16
 .L14rounds:
        movups  16($inp),%xmm2                  # remaning half of *userKey
        mov     \$13,$bits                      # 14 rounds for 256
        lea     16(%rax),%rax

+       cmp     \$`1<<28`,%r10d                 # AVX, but no XOP
+       je      .L14rounds_alt
+

        $movkey %xmm0,($key)                    # round 0
        $movkey %xmm2,16($key)                  # round 1
        aeskeygenassist \$0x1,%xmm2,%xmm1       # round 2
        $movkey %xmm0,($key)                    # round 0
        $movkey %xmm2,16($key)                  # round 1
        aeskeygenassist \$0x1,%xmm2,%xmm1       # round 2


@@ -3135,10 +3560,70 @@ __aesni_set_encrypt_key:
        xor     %rax,%rax
        jmp     .Lenc_key_ret
 
        xor     %rax,%rax
        jmp     .Lenc_key_ret
 

+.align 16
+.L14rounds_alt:
+       movdqa  .Lkey_rotate(%rip),%xmm5
+       movdqa  .Lkey_rcon1(%rip),%xmm4
+       mov     \$7,%r10d
+       movdqu  %xmm0,0($key)
+       movdqa  %xmm2,%xmm1
+       movdqu  %xmm2,16($key)
+       jmp     .Loop_key256
+
+.align 16
+.Loop_key256:
+       pshufb          %xmm5,%xmm2
+       aesenclast      %xmm4,%xmm2
+
+       movdqa          %xmm0,%xmm3
+       pslldq          \$4,%xmm0
+       pxor            %xmm0,%xmm3
+       pslldq          \$4,%xmm0
+       pxor            %xmm0,%xmm3
+       pslldq          \$4,%xmm0
+       pxor            %xmm3,%xmm0
+       pslld           \$1,%xmm4
+
+       pxor            %xmm2,%xmm0
+       movdqu          %xmm0,(%rax)
+
+       dec     %r10d
+       jz      .Ldone_key256
+
+       pshufd          \$0xff,%xmm0,%xmm2
+       pxor            %xmm3,%xmm3
+       aesenclast      %xmm3,%xmm2
+
+       movdqa          %xmm1,%xmm3
+       pslldq          \$4,%xmm1
+       pxor            %xmm1,%xmm3
+       pslldq          \$4,%xmm1
+       pxor            %xmm1,%xmm3
+       pslldq          \$4,%xmm1
+       pxor            %xmm3,%xmm1
+
+       pxor            %xmm1,%xmm2
+       movdqu          %xmm2,16(%rax)
+       lea             32(%rax),%rax
+       movdqa          %xmm2,%xmm1
+
+       jmp     .Loop_key256
+
+.Ldone_key256:
+       mov     $bits,16(%rax)  # 240($key)
+       xor     %eax,%eax
+       jmp     .Lenc_key_ret
+

 .align 16
 .Lbad_keybits:
        mov     \$-2,%rax
 .Lenc_key_ret:
 .align 16
 .Lbad_keybits:
        mov     \$-2,%rax
 .Lenc_key_ret:

+       pxor    %xmm0,%xmm0
+       pxor    %xmm1,%xmm1
+       pxor    %xmm2,%xmm2
+       pxor    %xmm3,%xmm3
+       pxor    %xmm4,%xmm4
+       pxor    %xmm5,%xmm5

        add     \$8,%rsp
        ret
 .LSEH_end_set_encrypt_key:
        add     \$8,%rsp
        ret
 .LSEH_end_set_encrypt_key:


@@ -3228,6 +3713,14 @@ $code.=<<___;
        .long   0x87,0,1,0
 .Lincrement1:
        .byte   0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1
        .long   0x87,0,1,0
 .Lincrement1:
        .byte   0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1

+.Lkey_rotate:
+       .long   0x0c0f0e0d,0x0c0f0e0d,0x0c0f0e0d,0x0c0f0e0d
+.Lkey_rotate192:
+       .long   0x04070605,0x04070605,0x04070605,0x04070605
+.Lkey_rcon1:
+       .long   1,1,1,1
+.Lkey_rcon1b:
+       .long   0x1b,0x1b,0x1b,0x1b

 
 .asciz  "AES for Intel AES-NI, CRYPTOGAMS by <appro\@openssl.org>"
 .align 64
 
 .asciz  "AES for Intel AES-NI, CRYPTOGAMS by <appro\@openssl.org>"
 .align 64


@@ -3345,7 +3838,7 @@ cbc_se_handler:
        mov     152($context),%rax      # pull context->Rsp
        mov     248($context),%rbx      # pull context->Rip
 
        mov     152($context),%rax      # pull context->Rsp
        mov     248($context),%rbx      # pull context->Rip
 

-       lea     .Lcbc_decrypt(%rip),%r10
+       lea     .Lcbc_decrypt_bulk(%rip),%r10

        cmp     %r10,%rbx               # context->Rip<"prologue" label
        jb      .Lcommon_seh_tail
 
        cmp     %r10,%rbx               # context->Rip<"prologue" label
        jb      .Lcommon_seh_tail