+$code.=<<___;
+.align 32
+.Linit_loop_avx:
+ vpalignr \$8,$T1,$T2,$T3 # low part is H.lo^H.hi...
+ vmovdqu $T3,-0x10($Htbl) # save Karatsuba "salt"
+___
+ &clmul64x64_avx ($Xhi,$Xi,$Hkey,$HK); # calculate H^3,5,7
+ &reduction_avx ($Xhi,$Xi);
+$code.=<<___;
+.Linit_start_avx:
+ vmovdqa $Xi,$T3
+___
+ &clmul64x64_avx ($Xhi,$Xi,$Hkey,$HK); # calculate H^2,4,6,8
+ &reduction_avx ($Xhi,$Xi);
+$code.=<<___;
+ vpshufd \$0b01001110,$T3,$T1
+ vpshufd \$0b01001110,$Xi,$T2
+ vpxor $T3,$T1,$T1 # Karatsuba pre-processing
+ vmovdqu $T3,0x00($Htbl) # save H^1,3,5,7
+ vpxor $Xi,$T2,$T2 # Karatsuba pre-processing
+ vmovdqu $Xi,0x10($Htbl) # save H^2,4,6,8
+ lea 0x30($Htbl),$Htbl
+ sub \$1,%r10
+ jnz .Linit_loop_avx
+
+ vpalignr \$8,$T2,$T1,$T3 # last "salt" is flipped
+ vmovdqu $T3,-0x10($Htbl)
+
+ vzeroupper
+___
+$code.=<<___ if ($win64);
+ movaps (%rsp),%xmm6
+ lea 0x18(%rsp),%rsp
+.LSEH_end_gcm_init_avx:
+___
+$code.=<<___;
+ ret
+.size gcm_init_avx,.-gcm_init_avx
+___
+} else {
+$code.=<<___;
+ jmp .L_init_clmul
+.size gcm_init_avx,.-gcm_init_avx
+___
+}
+
+$code.=<<___;
+.globl gcm_gmult_avx
+.type gcm_gmult_avx,\@abi-omnipotent
+.align 32
+gcm_gmult_avx:
+ jmp .L_gmult_clmul
+.size gcm_gmult_avx,.-gcm_gmult_avx
+___
+\f
+$code.=<<___;
+.globl gcm_ghash_avx
+.type gcm_ghash_avx,\@abi-omnipotent
+.align 32
+gcm_ghash_avx:
+___
+if ($avx) {
+my ($Xip,$Htbl,$inp,$len)=@_4args;
+my ($Xlo,$Xhi,$Xmi,
+ $Zlo,$Zhi,$Zmi,
+ $Hkey,$HK,$T1,$T2,
+ $Xi,$Xo,$Tred,$bswap,$Ii,$Ij) = map("%xmm$_",(0..15));
+
+$code.=<<___ if ($win64);
+ lea -0x88(%rsp),%rax
+.LSEH_begin_gcm_ghash_avx:
+ # I can't trust assembler to use specific encoding:-(
+ .byte 0x48,0x8d,0x60,0xe0 #lea -0x20(%rax),%rsp
+ .byte 0x0f,0x29,0x70,0xe0 #movaps %xmm6,-0x20(%rax)
+ .byte 0x0f,0x29,0x78,0xf0 #movaps %xmm7,-0x10(%rax)
+ .byte 0x44,0x0f,0x29,0x00 #movaps %xmm8,0(%rax)
+ .byte 0x44,0x0f,0x29,0x48,0x10 #movaps %xmm9,0x10(%rax)
+ .byte 0x44,0x0f,0x29,0x50,0x20 #movaps %xmm10,0x20(%rax)
+ .byte 0x44,0x0f,0x29,0x58,0x30 #movaps %xmm11,0x30(%rax)
+ .byte 0x44,0x0f,0x29,0x60,0x40 #movaps %xmm12,0x40(%rax)
+ .byte 0x44,0x0f,0x29,0x68,0x50 #movaps %xmm13,0x50(%rax)
+ .byte 0x44,0x0f,0x29,0x70,0x60 #movaps %xmm14,0x60(%rax)
+ .byte 0x44,0x0f,0x29,0x78,0x70 #movaps %xmm15,0x70(%rax)
+___
+$code.=<<___;
+ vzeroupper
+
+ vmovdqu ($Xip),$Xi # load $Xi
+ lea .L0x1c2_polynomial(%rip),%r10
+ lea 0x40($Htbl),$Htbl # size optimization
+ vmovdqu .Lbswap_mask(%rip),$bswap
+ vpshufb $bswap,$Xi,$Xi
+ cmp \$0x80,$len
+ jb .Lshort_avx
+ sub \$0x80,$len
+
+ vmovdqu 0x70($inp),$Ii # I[7]
+ vmovdqu 0x00-0x40($Htbl),$Hkey # $Hkey^1
+ vpshufb $bswap,$Ii,$Ii
+ vmovdqu 0x20-0x40($Htbl),$HK
+
+ vpunpckhqdq $Ii,$Ii,$T2
+ vmovdqu 0x60($inp),$Ij # I[6]
+ vpclmulqdq \$0x00,$Hkey,$Ii,$Xlo
+ vpxor $Ii,$T2,$T2
+ vpshufb $bswap,$Ij,$Ij
+ vpclmulqdq \$0x11,$Hkey,$Ii,$Xhi
+ vmovdqu 0x10-0x40($Htbl),$Hkey # $Hkey^2
+ vpunpckhqdq $Ij,$Ij,$T1
+ vmovdqu 0x50($inp),$Ii # I[5]
+ vpclmulqdq \$0x00,$HK,$T2,$Xmi
+ vpxor $Ij,$T1,$T1
+
+ vpshufb $bswap,$Ii,$Ii
+ vpclmulqdq \$0x00,$Hkey,$Ij,$Zlo
+ vpunpckhqdq $Ii,$Ii,$T2
+ vpclmulqdq \$0x11,$Hkey,$Ij,$Zhi
+ vmovdqu 0x30-0x40($Htbl),$Hkey # $Hkey^3
+ vpxor $Ii,$T2,$T2
+ vmovdqu 0x40($inp),$Ij # I[4]
+ vpclmulqdq \$0x10,$HK,$T1,$Zmi
+ vmovdqu 0x50-0x40($Htbl),$HK
+
+ vpshufb $bswap,$Ij,$Ij
+ vpxor $Xlo,$Zlo,$Zlo
+ vpclmulqdq \$0x00,$Hkey,$Ii,$Xlo
+ vpxor $Xhi,$Zhi,$Zhi
+ vpunpckhqdq $Ij,$Ij,$T1
+ vpclmulqdq \$0x11,$Hkey,$Ii,$Xhi
+ vmovdqu 0x40-0x40($Htbl),$Hkey # $Hkey^4
+ vpxor $Xmi,$Zmi,$Zmi
+ vpclmulqdq \$0x00,$HK,$T2,$Xmi
+ vpxor $Ij,$T1,$T1
+
+ vmovdqu 0x30($inp),$Ii # I[3]
+ vpxor $Zlo,$Xlo,$Xlo
+ vpclmulqdq \$0x00,$Hkey,$Ij,$Zlo
+ vpxor $Zhi,$Xhi,$Xhi
+ vpshufb $bswap,$Ii,$Ii
+ vpclmulqdq \$0x11,$Hkey,$Ij,$Zhi
+ vmovdqu 0x60-0x40($Htbl),$Hkey # $Hkey^5
+ vpxor $Zmi,$Xmi,$Xmi
+ vpunpckhqdq $Ii,$Ii,$T2
+ vpclmulqdq \$0x10,$HK,$T1,$Zmi
+ vmovdqu 0x80-0x40($Htbl),$HK
+ vpxor $Ii,$T2,$T2
+
+ vmovdqu 0x20($inp),$Ij # I[2]
+ vpxor $Xlo,$Zlo,$Zlo
+ vpclmulqdq \$0x00,$Hkey,$Ii,$Xlo
+ vpxor $Xhi,$Zhi,$Zhi
+ vpshufb $bswap,$Ij,$Ij
+ vpclmulqdq \$0x11,$Hkey,$Ii,$Xhi
+ vmovdqu 0x70-0x40($Htbl),$Hkey # $Hkey^6
+ vpxor $Xmi,$Zmi,$Zmi
+ vpunpckhqdq $Ij,$Ij,$T1
+ vpclmulqdq \$0x00,$HK,$T2,$Xmi
+ vpxor $Ij,$T1,$T1
+
+ vmovdqu 0x10($inp),$Ii # I[1]
+ vpxor $Zlo,$Xlo,$Xlo
+ vpclmulqdq \$0x00,$Hkey,$Ij,$Zlo
+ vpxor $Zhi,$Xhi,$Xhi
+ vpshufb $bswap,$Ii,$Ii
+ vpclmulqdq \$0x11,$Hkey,$Ij,$Zhi
+ vmovdqu 0x90-0x40($Htbl),$Hkey # $Hkey^7
+ vpxor $Zmi,$Xmi,$Xmi
+ vpunpckhqdq $Ii,$Ii,$T2
+ vpclmulqdq \$0x10,$HK,$T1,$Zmi
+ vmovdqu 0xb0-0x40($Htbl),$HK
+ vpxor $Ii,$T2,$T2
+
+ vmovdqu ($inp),$Ij # I[0]
+ vpxor $Xlo,$Zlo,$Zlo
+ vpclmulqdq \$0x00,$Hkey,$Ii,$Xlo
+ vpxor $Xhi,$Zhi,$Zhi
+ vpshufb $bswap,$Ij,$Ij
+ vpclmulqdq \$0x11,$Hkey,$Ii,$Xhi
+ vmovdqu 0xa0-0x40($Htbl),$Hkey # $Hkey^8
+ vpxor $Xmi,$Zmi,$Zmi
+ vpclmulqdq \$0x10,$HK,$T2,$Xmi
+
+ lea 0x80($inp),$inp
+ cmp \$0x80,$len
+ jb .Ltail_avx
+
+ vpxor $Xi,$Ij,$Ij # accumulate $Xi
+ sub \$0x80,$len
+ jmp .Loop8x_avx
+
+.align 32
+.Loop8x_avx:
+ vpunpckhqdq $Ij,$Ij,$T1
+ vmovdqu 0x70($inp),$Ii # I[7]
+ vpxor $Xlo,$Zlo,$Zlo
+ vpxor $Ij,$T1,$T1
+ vpclmulqdq \$0x00,$Hkey,$Ij,$Xi
+ vpshufb $bswap,$Ii,$Ii
+ vpxor $Xhi,$Zhi,$Zhi
+ vpclmulqdq \$0x11,$Hkey,$Ij,$Xo
+ vmovdqu 0x00-0x40($Htbl),$Hkey # $Hkey^1
+ vpunpckhqdq $Ii,$Ii,$T2
+ vpxor $Xmi,$Zmi,$Zmi
+ vpclmulqdq \$0x00,$HK,$T1,$Tred
+ vmovdqu 0x20-0x40($Htbl),$HK
+ vpxor $Ii,$T2,$T2
+
+ vmovdqu 0x60($inp),$Ij # I[6]
+ vpclmulqdq \$0x00,$Hkey,$Ii,$Xlo
+ vpxor $Zlo,$Xi,$Xi # collect result
+ vpshufb $bswap,$Ij,$Ij
+ vpclmulqdq \$0x11,$Hkey,$Ii,$Xhi
+ vxorps $Zhi,$Xo,$Xo
+ vmovdqu 0x10-0x40($Htbl),$Hkey # $Hkey^2
+ vpunpckhqdq $Ij,$Ij,$T1
+ vpclmulqdq \$0x00,$HK, $T2,$Xmi
+ vpxor $Zmi,$Tred,$Tred
+ vxorps $Ij,$T1,$T1
+
+ vmovdqu 0x50($inp),$Ii # I[5]
+ vpxor $Xi,$Tred,$Tred # aggregated Karatsuba post-processing
+ vpclmulqdq \$0x00,$Hkey,$Ij,$Zlo
+ vpxor $Xo,$Tred,$Tred
+ vpslldq \$8,$Tred,$T2
+ vpxor $Xlo,$Zlo,$Zlo
+ vpclmulqdq \$0x11,$Hkey,$Ij,$Zhi
+ vpsrldq \$8,$Tred,$Tred
+ vpxor $T2, $Xi, $Xi
+ vmovdqu 0x30-0x40($Htbl),$Hkey # $Hkey^3
+ vpshufb $bswap,$Ii,$Ii
+ vxorps $Tred,$Xo, $Xo
+ vpxor $Xhi,$Zhi,$Zhi
+ vpunpckhqdq $Ii,$Ii,$T2
+ vpclmulqdq \$0x10,$HK, $T1,$Zmi
+ vmovdqu 0x50-0x40($Htbl),$HK
+ vpxor $Ii,$T2,$T2
+ vpxor $Xmi,$Zmi,$Zmi
+
+ vmovdqu 0x40($inp),$Ij # I[4]
+ vpalignr \$8,$Xi,$Xi,$Tred # 1st phase
+ vpclmulqdq \$0x00,$Hkey,$Ii,$Xlo
+ vpshufb $bswap,$Ij,$Ij
+ vpxor $Zlo,$Xlo,$Xlo
+ vpclmulqdq \$0x11,$Hkey,$Ii,$Xhi
+ vmovdqu 0x40-0x40($Htbl),$Hkey # $Hkey^4
+ vpunpckhqdq $Ij,$Ij,$T1
+ vpxor $Zhi,$Xhi,$Xhi
+ vpclmulqdq \$0x00,$HK, $T2,$Xmi
+ vxorps $Ij,$T1,$T1
+ vpxor $Zmi,$Xmi,$Xmi
+
+ vmovdqu 0x30($inp),$Ii # I[3]
+ vpclmulqdq \$0x10,(%r10),$Xi,$Xi
+ vpclmulqdq \$0x00,$Hkey,$Ij,$Zlo
+ vpshufb $bswap,$Ii,$Ii
+ vpxor $Xlo,$Zlo,$Zlo
+ vpclmulqdq \$0x11,$Hkey,$Ij,$Zhi
+ vmovdqu 0x60-0x40($Htbl),$Hkey # $Hkey^5
+ vpunpckhqdq $Ii,$Ii,$T2
+ vpxor $Xhi,$Zhi,$Zhi
+ vpclmulqdq \$0x10,$HK, $T1,$Zmi
+ vmovdqu 0x80-0x40($Htbl),$HK
+ vpxor $Ii,$T2,$T2
+ vpxor $Xmi,$Zmi,$Zmi
+
+ vmovdqu 0x20($inp),$Ij # I[2]
+ vpclmulqdq \$0x00,$Hkey,$Ii,$Xlo
+ vpshufb $bswap,$Ij,$Ij
+ vpxor $Zlo,$Xlo,$Xlo
+ vpclmulqdq \$0x11,$Hkey,$Ii,$Xhi
+ vmovdqu 0x70-0x40($Htbl),$Hkey # $Hkey^6
+ vpunpckhqdq $Ij,$Ij,$T1
+ vpxor $Zhi,$Xhi,$Xhi
+ vpclmulqdq \$0x00,$HK, $T2,$Xmi
+ vpxor $Ij,$T1,$T1
+ vpxor $Zmi,$Xmi,$Xmi
+ vxorps $Tred,$Xi,$Xi
+
+ vmovdqu 0x10($inp),$Ii # I[1]
+ vpalignr \$8,$Xi,$Xi,$Tred # 2nd phase
+ vpclmulqdq \$0x00,$Hkey,$Ij,$Zlo
+ vpshufb $bswap,$Ii,$Ii
+ vpxor $Xlo,$Zlo,$Zlo
+ vpclmulqdq \$0x11,$Hkey,$Ij,$Zhi
+ vmovdqu 0x90-0x40($Htbl),$Hkey # $Hkey^7
+ vpclmulqdq \$0x10,(%r10),$Xi,$Xi
+ vxorps $Xo,$Tred,$Tred
+ vpunpckhqdq $Ii,$Ii,$T2
+ vpxor $Xhi,$Zhi,$Zhi
+ vpclmulqdq \$0x10,$HK, $T1,$Zmi
+ vmovdqu 0xb0-0x40($Htbl),$HK
+ vpxor $Ii,$T2,$T2
+ vpxor $Xmi,$Zmi,$Zmi
+
+ vmovdqu ($inp),$Ij # I[0]
+ vpclmulqdq \$0x00,$Hkey,$Ii,$Xlo
+ vpshufb $bswap,$Ij,$Ij
+ vpclmulqdq \$0x11,$Hkey,$Ii,$Xhi
+ vmovdqu 0xa0-0x40($Htbl),$Hkey # $Hkey^8
+ vpxor $Tred,$Ij,$Ij
+ vpclmulqdq \$0x10,$HK, $T2,$Xmi
+ vpxor $Xi,$Ij,$Ij # accumulate $Xi
+
+ lea 0x80($inp),$inp
+ sub \$0x80,$len
+ jnc .Loop8x_avx
+
+ add \$0x80,$len
+ jmp .Ltail_no_xor_avx
+
+.align 32
+.Lshort_avx:
+ vmovdqu -0x10($inp,$len),$Ii # very last word
+ lea ($inp,$len),$inp
+ vmovdqu 0x00-0x40($Htbl),$Hkey # $Hkey^1
+ vmovdqu 0x20-0x40($Htbl),$HK
+ vpshufb $bswap,$Ii,$Ij
+
+ vmovdqa $Xlo,$Zlo # subtle way to zero $Zlo,
+ vmovdqa $Xhi,$Zhi # $Zhi and
+ vmovdqa $Xmi,$Zmi # $Zmi
+ sub \$0x10,$len
+ jz .Ltail_avx
+
+ vpunpckhqdq $Ij,$Ij,$T1
+ vpxor $Xlo,$Zlo,$Zlo
+ vpclmulqdq \$0x00,$Hkey,$Ij,$Xlo
+ vpxor $Ij,$T1,$T1
+ vmovdqu -0x20($inp),$Ii
+ vpxor $Xhi,$Zhi,$Zhi
+ vpclmulqdq \$0x11,$Hkey,$Ij,$Xhi
+ vmovdqu 0x10-0x40($Htbl),$Hkey # $Hkey^2
+ vpshufb $bswap,$Ii,$Ij
+ vpxor $Xmi,$Zmi,$Zmi
+ vpclmulqdq \$0x00,$HK,$T1,$Xmi
+ vpsrldq \$8,$HK,$HK
+ sub \$0x10,$len
+ jz .Ltail_avx
+
+ vpunpckhqdq $Ij,$Ij,$T1
+ vpxor $Xlo,$Zlo,$Zlo
+ vpclmulqdq \$0x00,$Hkey,$Ij,$Xlo
+ vpxor $Ij,$T1,$T1
+ vmovdqu -0x30($inp),$Ii
+ vpxor $Xhi,$Zhi,$Zhi
+ vpclmulqdq \$0x11,$Hkey,$Ij,$Xhi
+ vmovdqu 0x30-0x40($Htbl),$Hkey # $Hkey^3
+ vpshufb $bswap,$Ii,$Ij
+ vpxor $Xmi,$Zmi,$Zmi
+ vpclmulqdq \$0x00,$HK,$T1,$Xmi
+ vmovdqu 0x50-0x40($Htbl),$HK
+ sub \$0x10,$len
+ jz .Ltail_avx
+
+ vpunpckhqdq $Ij,$Ij,$T1
+ vpxor $Xlo,$Zlo,$Zlo
+ vpclmulqdq \$0x00,$Hkey,$Ij,$Xlo
+ vpxor $Ij,$T1,$T1
+ vmovdqu -0x40($inp),$Ii
+ vpxor $Xhi,$Zhi,$Zhi
+ vpclmulqdq \$0x11,$Hkey,$Ij,$Xhi
+ vmovdqu 0x40-0x40($Htbl),$Hkey # $Hkey^4
+ vpshufb $bswap,$Ii,$Ij
+ vpxor $Xmi,$Zmi,$Zmi
+ vpclmulqdq \$0x00,$HK,$T1,$Xmi
+ vpsrldq \$8,$HK,$HK
+ sub \$0x10,$len
+ jz .Ltail_avx
+
+ vpunpckhqdq $Ij,$Ij,$T1
+ vpxor $Xlo,$Zlo,$Zlo
+ vpclmulqdq \$0x00,$Hkey,$Ij,$Xlo
+ vpxor $Ij,$T1,$T1
+ vmovdqu -0x50($inp),$Ii
+ vpxor $Xhi,$Zhi,$Zhi
+ vpclmulqdq \$0x11,$Hkey,$Ij,$Xhi
+ vmovdqu 0x60-0x40($Htbl),$Hkey # $Hkey^5
+ vpshufb $bswap,$Ii,$Ij
+ vpxor $Xmi,$Zmi,$Zmi
+ vpclmulqdq \$0x00,$HK,$T1,$Xmi
+ vmovdqu 0x80-0x40($Htbl),$HK
+ sub \$0x10,$len
+ jz .Ltail_avx
+
+ vpunpckhqdq $Ij,$Ij,$T1
+ vpxor $Xlo,$Zlo,$Zlo
+ vpclmulqdq \$0x00,$Hkey,$Ij,$Xlo
+ vpxor $Ij,$T1,$T1
+ vmovdqu -0x60($inp),$Ii
+ vpxor $Xhi,$Zhi,$Zhi
+ vpclmulqdq \$0x11,$Hkey,$Ij,$Xhi
+ vmovdqu 0x70-0x40($Htbl),$Hkey # $Hkey^6
+ vpshufb $bswap,$Ii,$Ij
+ vpxor $Xmi,$Zmi,$Zmi
+ vpclmulqdq \$0x00,$HK,$T1,$Xmi
+ vpsrldq \$8,$HK,$HK
+ sub \$0x10,$len
+ jz .Ltail_avx
+
+ vpunpckhqdq $Ij,$Ij,$T1
+ vpxor $Xlo,$Zlo,$Zlo
+ vpclmulqdq \$0x00,$Hkey,$Ij,$Xlo
+ vpxor $Ij,$T1,$T1
+ vmovdqu -0x70($inp),$Ii
+ vpxor $Xhi,$Zhi,$Zhi
+ vpclmulqdq \$0x11,$Hkey,$Ij,$Xhi
+ vmovdqu 0x90-0x40($Htbl),$Hkey # $Hkey^7
+ vpshufb $bswap,$Ii,$Ij
+ vpxor $Xmi,$Zmi,$Zmi
+ vpclmulqdq \$0x00,$HK,$T1,$Xmi
+ vmovq 0xb8-0x40($Htbl),$HK
+ sub \$0x10,$len
+ jmp .Ltail_avx
+
+.align 32
+.Ltail_avx:
+ vpxor $Xi,$Ij,$Ij # accumulate $Xi
+.Ltail_no_xor_avx:
+ vpunpckhqdq $Ij,$Ij,$T1
+ vpxor $Xlo,$Zlo,$Zlo
+ vpclmulqdq \$0x00,$Hkey,$Ij,$Xlo
+ vpxor $Ij,$T1,$T1
+ vpxor $Xhi,$Zhi,$Zhi
+ vpclmulqdq \$0x11,$Hkey,$Ij,$Xhi
+ vpxor $Xmi,$Zmi,$Zmi
+ vpclmulqdq \$0x00,$HK,$T1,$Xmi
+
+ vmovdqu (%r10),$Tred
+
+ vpxor $Xlo,$Zlo,$Xi
+ vpxor $Xhi,$Zhi,$Xo
+ vpxor $Xmi,$Zmi,$Zmi
+
+ vpxor $Xi, $Zmi,$Zmi # aggregated Karatsuba post-processing
+ vpxor $Xo, $Zmi,$Zmi
+ vpslldq \$8, $Zmi,$T2
+ vpsrldq \$8, $Zmi,$Zmi
+ vpxor $T2, $Xi, $Xi
+ vpxor $Zmi,$Xo, $Xo
+
+ vpclmulqdq \$0x10,$Tred,$Xi,$T2 # 1st phase
+ vpalignr \$8,$Xi,$Xi,$Xi
+ vpxor $T2,$Xi,$Xi
+
+ vpclmulqdq \$0x10,$Tred,$Xi,$T2 # 2nd phase
+ vpalignr \$8,$Xi,$Xi,$Xi
+ vpxor $Xo,$Xi,$Xi
+ vpxor $T2,$Xi,$Xi
+
+ cmp \$0,$len
+ jne .Lshort_avx
+
+ vpshufb $bswap,$Xi,$Xi
+ vmovdqu $Xi,($Xip)
+ vzeroupper
+___
+$code.=<<___ if ($win64);
+ movaps (%rsp),%xmm6
+ movaps 0x10(%rsp),%xmm7
+ movaps 0x20(%rsp),%xmm8
+ movaps 0x30(%rsp),%xmm9
+ movaps 0x40(%rsp),%xmm10
+ movaps 0x50(%rsp),%xmm11
+ movaps 0x60(%rsp),%xmm12
+ movaps 0x70(%rsp),%xmm13
+ movaps 0x80(%rsp),%xmm14
+ movaps 0x90(%rsp),%xmm15
+ lea 0xa8(%rsp),%rsp
+.LSEH_end_gcm_ghash_avx:
+___
+$code.=<<___;
+ ret
+.size gcm_ghash_avx,.-gcm_ghash_avx
+___
+} else {
+$code.=<<___;
+ jmp .L_ghash_clmul
+.size gcm_ghash_avx,.-gcm_ghash_avx
+___
+}
+\f