OLD | NEW |
1 #if defined(__x86_64__) | 1 #if defined(__x86_64__) |
2 .text | 2 .text |
3 | 3 |
4 | 4 |
5 | 5 |
6 .globl bn_mul_mont_gather5 | 6 .globl bn_mul_mont_gather5 |
| 7 .hidden bn_mul_mont_gather5 |
7 .type bn_mul_mont_gather5,@function | 8 .type bn_mul_mont_gather5,@function |
8 .align 64 | 9 .align 64 |
9 bn_mul_mont_gather5: | 10 bn_mul_mont_gather5: |
10 testl $7,%r9d | 11 testl $7,%r9d |
11 jnz .Lmul_enter | 12 jnz .Lmul_enter |
12 jmp .Lmul4x_enter | 13 jmp .Lmul4x_enter |
13 | 14 |
14 .align 16 | 15 .align 16 |
15 .Lmul_enter: | 16 .Lmul_enter: |
16 movl %r9d,%r9d | 17 movl %r9d,%r9d |
(...skipping 705 matching lines...) Expand 10 before | Expand all | Expand 10 after Loading... |
722 orq %r15,%rdi | 723 orq %r15,%rdi |
723 xorq $1,%rdi | 724 xorq $1,%rdi |
724 leaq (%r14,%r9,1),%rbx | 725 leaq (%r14,%r9,1),%rbx |
725 leaq (%rcx,%rdi,8),%rbp | 726 leaq (%rcx,%rdi,8),%rbp |
726 movq %r9,%rcx | 727 movq %r9,%rcx |
727 sarq $3+2,%rcx | 728 sarq $3+2,%rcx |
728 movq 56+8(%rsp),%rdi | 729 movq 56+8(%rsp),%rdi |
729 jmp .Lsqr4x_sub | 730 jmp .Lsqr4x_sub |
730 .size mul4x_internal,.-mul4x_internal | 731 .size mul4x_internal,.-mul4x_internal |
731 .globl bn_power5 | 732 .globl bn_power5 |
| 733 .hidden bn_power5 |
732 .type bn_power5,@function | 734 .type bn_power5,@function |
733 .align 32 | 735 .align 32 |
734 bn_power5: | 736 bn_power5: |
735 movq %rsp,%rax | 737 movq %rsp,%rax |
736 pushq %rbx | 738 pushq %rbx |
737 pushq %rbp | 739 pushq %rbp |
738 pushq %r12 | 740 pushq %r12 |
739 pushq %r13 | 741 pushq %r13 |
740 pushq %r14 | 742 pushq %r14 |
741 pushq %r15 | 743 pushq %r15 |
(...skipping 69 matching lines...) Expand 10 before | Expand all | Expand 10 after Loading... |
811 movq -32(%rsi),%r13 | 813 movq -32(%rsi),%r13 |
812 movq -24(%rsi),%r12 | 814 movq -24(%rsi),%r12 |
813 movq -16(%rsi),%rbp | 815 movq -16(%rsi),%rbp |
814 movq -8(%rsi),%rbx | 816 movq -8(%rsi),%rbx |
815 leaq (%rsi),%rsp | 817 leaq (%rsi),%rsp |
816 .Lpower5_epilogue: | 818 .Lpower5_epilogue: |
817 .byte 0xf3,0xc3 | 819 .byte 0xf3,0xc3 |
818 .size bn_power5,.-bn_power5 | 820 .size bn_power5,.-bn_power5 |
819 | 821 |
820 .globl bn_sqr8x_internal | 822 .globl bn_sqr8x_internal |
| 823 .hidden bn_sqr8x_internal |
821 .hidden bn_sqr8x_internal | 824 .hidden bn_sqr8x_internal |
822 .type bn_sqr8x_internal,@function | 825 .type bn_sqr8x_internal,@function |
823 .align 32 | 826 .align 32 |
824 bn_sqr8x_internal: | 827 bn_sqr8x_internal: |
825 __bn_sqr8x_internal: | 828 __bn_sqr8x_internal: |
826 | 829 |
827 | 830 |
828 | 831 |
829 | 832 |
830 | 833 |
(...skipping 789 matching lines...) Expand 10 before | Expand all | Expand 10 after Loading... |
1620 movq %r15,24(%rdi) | 1623 movq %r15,24(%rdi) |
1621 leaq 32(%rdi),%rdi | 1624 leaq 32(%rdi),%rdi |
1622 | 1625 |
1623 incq %rcx | 1626 incq %rcx |
1624 jnz .Lsqr4x_sub | 1627 jnz .Lsqr4x_sub |
1625 movq %r9,%r10 | 1628 movq %r9,%r10 |
1626 negq %r9 | 1629 negq %r9 |
1627 .byte 0xf3,0xc3 | 1630 .byte 0xf3,0xc3 |
1628 .size bn_sqr8x_internal,.-bn_sqr8x_internal | 1631 .size bn_sqr8x_internal,.-bn_sqr8x_internal |
1629 .globl bn_from_montgomery | 1632 .globl bn_from_montgomery |
| 1633 .hidden bn_from_montgomery |
1630 .type bn_from_montgomery,@function | 1634 .type bn_from_montgomery,@function |
1631 .align 32 | 1635 .align 32 |
1632 bn_from_montgomery: | 1636 bn_from_montgomery: |
1633 testl $7,%r9d | 1637 testl $7,%r9d |
1634 jz bn_from_mont8x | 1638 jz bn_from_mont8x |
1635 xorl %eax,%eax | 1639 xorl %eax,%eax |
1636 .byte 0xf3,0xc3 | 1640 .byte 0xf3,0xc3 |
1637 .size bn_from_montgomery,.-bn_from_montgomery | 1641 .size bn_from_montgomery,.-bn_from_montgomery |
1638 | 1642 |
1639 .type bn_from_mont8x,@function | 1643 .type bn_from_mont8x,@function |
(...skipping 105 matching lines...) Expand 10 before | Expand all | Expand 10 after Loading... |
1745 movq -40(%rsi),%r14 | 1749 movq -40(%rsi),%r14 |
1746 movq -32(%rsi),%r13 | 1750 movq -32(%rsi),%r13 |
1747 movq -24(%rsi),%r12 | 1751 movq -24(%rsi),%r12 |
1748 movq -16(%rsi),%rbp | 1752 movq -16(%rsi),%rbp |
1749 movq -8(%rsi),%rbx | 1753 movq -8(%rsi),%rbx |
1750 leaq (%rsi),%rsp | 1754 leaq (%rsi),%rsp |
1751 .Lfrom_epilogue: | 1755 .Lfrom_epilogue: |
1752 .byte 0xf3,0xc3 | 1756 .byte 0xf3,0xc3 |
1753 .size bn_from_mont8x,.-bn_from_mont8x | 1757 .size bn_from_mont8x,.-bn_from_mont8x |
1754 .globl bn_get_bits5 | 1758 .globl bn_get_bits5 |
| 1759 .hidden bn_get_bits5 |
1755 .type bn_get_bits5,@function | 1760 .type bn_get_bits5,@function |
1756 .align 16 | 1761 .align 16 |
1757 bn_get_bits5: | 1762 bn_get_bits5: |
1758 movq %rdi,%r10 | 1763 movq %rdi,%r10 |
1759 movl %esi,%ecx | 1764 movl %esi,%ecx |
1760 shrl $3,%esi | 1765 shrl $3,%esi |
1761 movzwl (%r10,%rsi,1),%eax | 1766 movzwl (%r10,%rsi,1),%eax |
1762 andl $7,%ecx | 1767 andl $7,%ecx |
1763 shrl %cl,%eax | 1768 shrl %cl,%eax |
1764 andl $31,%eax | 1769 andl $31,%eax |
1765 .byte 0xf3,0xc3 | 1770 .byte 0xf3,0xc3 |
1766 .size bn_get_bits5,.-bn_get_bits5 | 1771 .size bn_get_bits5,.-bn_get_bits5 |
1767 | 1772 |
1768 .globl bn_scatter5 | 1773 .globl bn_scatter5 |
| 1774 .hidden bn_scatter5 |
1769 .type bn_scatter5,@function | 1775 .type bn_scatter5,@function |
1770 .align 16 | 1776 .align 16 |
1771 bn_scatter5: | 1777 bn_scatter5: |
1772 cmpl $0,%esi | 1778 cmpl $0,%esi |
1773 jz .Lscatter_epilogue | 1779 jz .Lscatter_epilogue |
1774 leaq (%rdx,%rcx,8),%rdx | 1780 leaq (%rdx,%rcx,8),%rdx |
1775 .Lscatter: | 1781 .Lscatter: |
1776 movq (%rdi),%rax | 1782 movq (%rdi),%rax |
1777 leaq 8(%rdi),%rdi | 1783 leaq 8(%rdi),%rdi |
1778 movq %rax,(%rdx) | 1784 movq %rax,(%rdx) |
1779 leaq 256(%rdx),%rdx | 1785 leaq 256(%rdx),%rdx |
1780 subl $1,%esi | 1786 subl $1,%esi |
1781 jnz .Lscatter | 1787 jnz .Lscatter |
1782 .Lscatter_epilogue: | 1788 .Lscatter_epilogue: |
1783 .byte 0xf3,0xc3 | 1789 .byte 0xf3,0xc3 |
1784 .size bn_scatter5,.-bn_scatter5 | 1790 .size bn_scatter5,.-bn_scatter5 |
1785 | 1791 |
1786 .globl bn_gather5 | 1792 .globl bn_gather5 |
| 1793 .hidden bn_gather5 |
1787 .type bn_gather5,@function | 1794 .type bn_gather5,@function |
1788 .align 16 | 1795 .align 16 |
1789 bn_gather5: | 1796 bn_gather5: |
1790 movl %ecx,%r11d | 1797 movl %ecx,%r11d |
1791 shrl $3,%ecx | 1798 shrl $3,%ecx |
1792 andq $7,%r11 | 1799 andq $7,%r11 |
1793 notl %ecx | 1800 notl %ecx |
1794 leaq .Lmagic_masks(%rip),%rax | 1801 leaq .Lmagic_masks(%rip),%rax |
1795 andl $3,%ecx | 1802 andl $3,%ecx |
1796 leaq 128(%rdx,%r11,8),%rdx | 1803 leaq 128(%rdx,%r11,8),%rdx |
(...skipping 24 matching lines...) Expand all Loading... |
1821 jnz .Lgather | 1828 jnz .Lgather |
1822 .byte 0xf3,0xc3 | 1829 .byte 0xf3,0xc3 |
1823 .LSEH_end_bn_gather5: | 1830 .LSEH_end_bn_gather5: |
1824 .size bn_gather5,.-bn_gather5 | 1831 .size bn_gather5,.-bn_gather5 |
1825 .align 64 | 1832 .align 64 |
1826 .Lmagic_masks: | 1833 .Lmagic_masks: |
1827 .long 0,0, 0,0, 0,0, -1,-1 | 1834 .long 0,0, 0,0, 0,0, -1,-1 |
1828 .long 0,0, 0,0, 0,0, 0,0 | 1835 .long 0,0, 0,0, 0,0, 0,0 |
1829 .byte 77,111,110,116,103,111,109,101,114,121,32,77,117,108,116,105,112,108,105
,99,97,116,105,111,110,32,119,105,116,104,32,115,99,97,116,116,101,114,47,103,97
,116,104,101,114,32,102,111,114,32,120,56,54,95,54,52,44,32,67,82,89,80,84,79,71
,65,77,83,32,98,121,32,60,97,112,112,114,111,64,111,112,101,110,115,115,108,46,1
11,114,103,62,0 | 1836 .byte 77,111,110,116,103,111,109,101,114,121,32,77,117,108,116,105,112,108,105
,99,97,116,105,111,110,32,119,105,116,104,32,115,99,97,116,116,101,114,47,103,97
,116,104,101,114,32,102,111,114,32,120,56,54,95,54,52,44,32,67,82,89,80,84,79,71
,65,77,83,32,98,121,32,60,97,112,112,114,111,64,111,112,101,110,115,115,108,46,1
11,114,103,62,0 |
1830 #endif | 1837 #endif |
OLD | NEW |