| // This file is generated from a similarly-named Perl script in the BoringSSL |
| // source tree. Do not edit by hand. |
| |
| #include <openssl/asm_base.h> |
| |
| #if !defined(OPENSSL_NO_ASM) && defined(OPENSSL_X86) && defined(__APPLE__) |
| .text |
| .globl _bn_mul_mont_words |
| .private_extern _bn_mul_mont_words |
| .align 4 |
| _bn_mul_mont_words: |
| L_bn_mul_mont_words_begin: |
| pushl %ebp |
| pushl %ebx |
| pushl %esi |
| pushl %edi |
| movl 40(%esp),%edi |
| leal 20(%esp),%esi |
| leal 24(%esp),%edx |
| addl $2,%edi |
| negl %edi |
| leal -32(%esp,%edi,4),%ebp |
| negl %edi |
| movl %ebp,%eax |
| subl %edx,%eax |
| andl $2047,%eax |
| subl %eax,%ebp |
| xorl %ebp,%edx |
| andl $2048,%edx |
| xorl $2048,%edx |
| subl %edx,%ebp |
| andl $-64,%ebp |
| movl %esp,%eax |
| subl %ebp,%eax |
| andl $-4096,%eax |
| movl %esp,%edx |
| leal (%ebp,%eax,1),%esp |
| movl (%esp),%eax |
| cmpl %ebp,%esp |
| ja L000page_walk |
| jmp L001page_walk_done |
| .align 4,0x90 |
| L000page_walk: |
| leal -4096(%esp),%esp |
| movl (%esp),%eax |
| cmpl %ebp,%esp |
| ja L000page_walk |
| L001page_walk_done: |
| movl (%esi),%eax |
| movl 4(%esi),%ebx |
| movl 8(%esi),%ecx |
| movl 12(%esi),%ebp |
| movl 16(%esi),%esi |
| movl (%esi),%esi |
| movl %eax,4(%esp) |
| movl %ebx,8(%esp) |
| movl %ecx,12(%esp) |
| movl %ebp,16(%esp) |
| movl %esi,20(%esp) |
| leal -3(%edi),%ebx |
| movl %edx,24(%esp) |
| movl $-1,%eax |
| movd %eax,%mm7 |
| movl 8(%esp),%esi |
| movl 12(%esp),%edi |
| movl 16(%esp),%ebp |
| xorl %edx,%edx |
| xorl %ecx,%ecx |
| movd (%edi),%mm4 |
| movd (%esi),%mm5 |
| movd (%ebp),%mm3 |
| pmuludq %mm4,%mm5 |
| movq %mm5,%mm2 |
| movq %mm5,%mm0 |
| pand %mm7,%mm0 |
| pmuludq 20(%esp),%mm5 |
| pmuludq %mm5,%mm3 |
| paddq %mm0,%mm3 |
| movd 4(%ebp),%mm1 |
| movd 4(%esi),%mm0 |
| psrlq $32,%mm2 |
| psrlq $32,%mm3 |
| incl %ecx |
| .align 4,0x90 |
| L0021st: |
| pmuludq %mm4,%mm0 |
| pmuludq %mm5,%mm1 |
| paddq %mm0,%mm2 |
| paddq %mm1,%mm3 |
| movq %mm2,%mm0 |
| pand %mm7,%mm0 |
| movd 4(%ebp,%ecx,4),%mm1 |
| paddq %mm0,%mm3 |
| movd 4(%esi,%ecx,4),%mm0 |
| psrlq $32,%mm2 |
| movd %mm3,28(%esp,%ecx,4) |
| psrlq $32,%mm3 |
| leal 1(%ecx),%ecx |
| cmpl %ebx,%ecx |
| jl L0021st |
| pmuludq %mm4,%mm0 |
| pmuludq %mm5,%mm1 |
| paddq %mm0,%mm2 |
| paddq %mm1,%mm3 |
| movq %mm2,%mm0 |
| pand %mm7,%mm0 |
| paddq %mm0,%mm3 |
| movd %mm3,28(%esp,%ecx,4) |
| psrlq $32,%mm2 |
| psrlq $32,%mm3 |
| paddq %mm2,%mm3 |
| movq %mm3,32(%esp,%ebx,4) |
| incl %edx |
| L003outer: |
| xorl %ecx,%ecx |
| movd (%edi,%edx,4),%mm4 |
| movd (%esi),%mm5 |
| movd 32(%esp),%mm6 |
| movd (%ebp),%mm3 |
| pmuludq %mm4,%mm5 |
| paddq %mm6,%mm5 |
| movq %mm5,%mm0 |
| movq %mm5,%mm2 |
| pand %mm7,%mm0 |
| pmuludq 20(%esp),%mm5 |
| pmuludq %mm5,%mm3 |
| paddq %mm0,%mm3 |
| movd 36(%esp),%mm6 |
| movd 4(%ebp),%mm1 |
| movd 4(%esi),%mm0 |
| psrlq $32,%mm2 |
| psrlq $32,%mm3 |
| paddq %mm6,%mm2 |
| incl %ecx |
| decl %ebx |
| L004inner: |
| pmuludq %mm4,%mm0 |
| pmuludq %mm5,%mm1 |
| paddq %mm0,%mm2 |
| paddq %mm1,%mm3 |
| movq %mm2,%mm0 |
| movd 36(%esp,%ecx,4),%mm6 |
| pand %mm7,%mm0 |
| movd 4(%ebp,%ecx,4),%mm1 |
| paddq %mm0,%mm3 |
| movd 4(%esi,%ecx,4),%mm0 |
| psrlq $32,%mm2 |
| movd %mm3,28(%esp,%ecx,4) |
| psrlq $32,%mm3 |
| paddq %mm6,%mm2 |
| decl %ebx |
| leal 1(%ecx),%ecx |
| jnz L004inner |
| movl %ecx,%ebx |
| pmuludq %mm4,%mm0 |
| pmuludq %mm5,%mm1 |
| paddq %mm0,%mm2 |
| paddq %mm1,%mm3 |
| movq %mm2,%mm0 |
| pand %mm7,%mm0 |
| paddq %mm0,%mm3 |
| movd %mm3,28(%esp,%ecx,4) |
| psrlq $32,%mm2 |
| psrlq $32,%mm3 |
| movd 36(%esp,%ebx,4),%mm6 |
| paddq %mm2,%mm3 |
| paddq %mm6,%mm3 |
| movq %mm3,32(%esp,%ebx,4) |
| leal 1(%edx),%edx |
| cmpl %ebx,%edx |
| jle L003outer |
| emms |
| jmp L005common_tail |
| .align 4,0x90 |
| L005common_tail: |
| movl 16(%esp),%ebp |
| movl 4(%esp),%edi |
| leal 32(%esp),%esi |
| movl (%esi),%eax |
| movl %ebx,%ecx |
| xorl %edx,%edx |
| .align 4,0x90 |
| L006sub: |
| sbbl (%ebp,%edx,4),%eax |
| movl %eax,(%edi,%edx,4) |
| decl %ecx |
| movl 4(%esi,%edx,4),%eax |
| leal 1(%edx),%edx |
| jge L006sub |
| sbbl $0,%eax |
| movl $-1,%edx |
| xorl %eax,%edx |
| jmp L007copy |
| .align 4,0x90 |
| L007copy: |
| movl 32(%esp,%ebx,4),%esi |
| movl (%edi,%ebx,4),%ebp |
| movl %ecx,32(%esp,%ebx,4) |
| andl %eax,%esi |
| andl %edx,%ebp |
| orl %esi,%ebp |
| movl %ebp,(%edi,%ebx,4) |
| decl %ebx |
| jge L007copy |
| movl 24(%esp),%esp |
| popl %edi |
| popl %esi |
| popl %ebx |
| popl %ebp |
| ret |
| .byte 77,111,110,116,103,111,109,101,114,121,32,77,117,108,116,105 |
| .byte 112,108,105,99,97,116,105,111,110,32,102,111,114,32,120,56 |
| .byte 54,44,32,67,82,89,80,84,79,71,65,77,83,32,98,121 |
| .byte 32,60,97,112,112,114,111,64,111,112,101,110,115,115,108,46 |
| .byte 111,114,103,62,0 |
| #endif // !defined(OPENSSL_NO_ASM) && defined(OPENSSL_X86) && defined(__APPLE__) |
