/* * Copyright (c) 2005 Apple Computer, Inc. All rights reserved. * * @APPLE_LICENSE_HEADER_START@ * * This file contains Original Code and/or Modifications of Original Code * as defined in and that are subject to the Apple Public Source License * Version 2.0 (the 'License'). You may not use this file except in * compliance with the License. Please obtain a copy of the License at * http://www.opensource.apple.com/apsl/ and read it before using this * file. * * The Original Code and all software distributed under the License are * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES, * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY, * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT. * Please see the License for the specific language governing rights and * limitations under the License. * * @APPLE_LICENSE_HEADER_END@ */ #include // ***************** // * S T R N C P Y * // ***************** // // char *strncpy(const char *dst, const char *src, size_t n); // // We optimize the move by doing it vector parallel. This introduces // a complication: if we blindly did vector load/stores until finding // a 0, we might get a spurious page fault by touching bytes past it. // To avoid this, we never do a load that crosses a page boundary, // and never store a byte we don't have to. // // We align the destination, because unaligned vector stores are slow. // // Recall that strncpy() zero fills the remainder of the dest buffer, // and does not terminate the string if its length is greater than or // equal to n. #define kShort 31 // too short to bother with vector loop .text .globl _strncpy .align 4 _strncpy: // char *strncpy(const char *dst, const char *src, size_t n); movq %rdi,%r8 // preserve destination pointer so we can return it movl %edi,%ecx // copy low 4 bytes of dest ptr negl %ecx andl $15,%ecx // how many bytes to align dest ptr? jnz LCheckShortCopy // align destination first // In order to avoid spurious page faults, we loop until nearing the source page // end. Then we revert to a byte-by-byte loop for 16 bytes until the page is crossed, // then resume the vector loop. // %rsi = source ptr (unaligned) // %rdi = dest ptr (aligned) // %rdx = buffer length remaining LNextChunk: // NB: can drop down to here movl %esi,%eax // copy the low 4 bytes of the source ptr movl $4096,%ecx andl $4095,%eax // get offset into source page subl %eax,%ecx // get #bytes remaining in source page cmpq %rdx,%rcx // will buffer run out before the page end? cmova %rdx,%rcx // get min(length remaining, bytes to page end) shrl $4,%ecx // get #chunks till end of page jnz LLoopOverChunks // enter vector loop // We can't use the chunk loop yet. Check for short and empty buffers, then use byte loop. LCrossPage: // if buffer is large enough, cross source page movl $16,%ecx // move 16 bytes to cross page but keep dest aligned LCheckShortCopy: // we propose to copy %ecx bytes in byte loop cmpq $(kShort),%rdx // much left? ja LLoopOverBytes // yes, loop over bytes then more chunks movl %edx,%ecx // no, use the byte loop for everything testl %edx,%edx // have we filled buffer? jnz LLoopOverBytes // no jmp LDone // Loop over bytes. // %rsi = source ptr // %rdi = dest ptr // %rdx = buffer length remaining // %rcx = count of bytes to loop over (<= buffer length) .align 4,0x90 // align inner loops to optimize I-fetch LLoopOverBytes: movzb (%rsi),%eax // get source byte addq $1,%rsi subq $1,%rdx // decrement length movb %al,(%rdi) // pack into dest addq $1,%rdi testl %eax,%eax // 0? jz LZeroBuffer // yes, we're done copying string subq $1,%rcx // more to go? jnz LLoopOverBytes testq %rdx,%rdx // at end of buffer? jnz LNextChunk // no, xfer chunks jmp LDone // yes // Loop over 16-byte chunks. // %rsi = source ptr (unaligned) // %rdi = dest ptr (aligned) // %rdx = buffer length remaining // %ecx = chunk count .align 4,0x90 // align inner loops to optimize I-fetch LLoopOverChunks: movdqu (%rsi),%xmm1 // get source pxor %xmm0,%xmm0 // get some 0s addq $16,%rsi pcmpeqb %xmm1,%xmm0 // compare source to 0s pmovmskb %xmm0,%eax // get result mask for 0 check testl %eax,%eax // any 0s? jnz LFound0 // yes, exit loop movdqa %xmm1,(%rdi) // no 0s so do aligned store into destination addq $16,%rdi subq $16,%rdx // decrement length remaining subl $1,%ecx // more to go? jnz LLoopOverChunks jmp LCrossPage // cross page but keep dest aligned // Found a zero in the vector. Figure out where it is, and store the bytes // up to it. It is possible that we should check to be sure (%rdx >= 16), and // just do an aligned store of %xmm1 if so. But if we did, we'd be doing byte // stores into the same double quadword in bzero(), which might hit a hazard. // Experimentation needed. // %rdi = dest ptr (aligned) // %eax = result mask // %rdx = buffer length remaining // %xmm1 = source vector LFound0: bsf %eax,%ecx // find first 0 subq %rcx,%rdx // decrement remaining buffer length test $8,%cl // 8-byte store required? jz 4f // no movq %xmm1,(%rdi) // pack in 8 low bytes psrldq $8,%xmm1 // then shift vector down 8 bytes addq $8,%rdi 4: test $4,%cl // 4-byte store required? jz 3f // no movd %xmm1,(%rdi) // pack in 4 low bytes psrldq $4,%xmm1 // then shift vector down 4 bytes addq $4,%rdi 3: andl $3,%ecx // more to go? jz LZeroBuffer // no movd %xmm1,%eax // move remainders out of vector into %eax 1: // loop on up to three bytes movb %al,(%rdi) // pack in next byte shrl $8,%eax // shift next byte into position addq $1,%rdi subl $1,%ecx jnz 1b // We've copied the string. Now zero the rest of the buffer, using commpage bzero(). // %rdi = dest ptr // %rcx = buffer length remaining LZeroBuffer: movq %rdx,%rsi // remaining buffer size (2nd argument) call _bzero LDone: movq %r8,%rax // original dest ptr is return value ret