/* * Copyright (c) 2012 Apple Computer, Inc. All rights reserved. * * @APPLE_OSREFERENCE_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. The rights granted to you under the License * may not be used to create, or enable the creation or redistribution of, * unlawful or unlicensed copies of an Apple operating system, or to * circumvent, violate, or enable the circumvention or violation of, any * terms of an Apple operating system software license agreement. * * 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_OSREFERENCE_LICENSE_HEADER_END@ * * This file implements the following functions for the arm64 architecture. * * void bcopy(const void * source, * void * destination, * size_t length); * * void *memmove(void * destination, * const void * source, * size_t n); * * void *memcpy(void * restrict destination, * const void * restrict source, * size_t n); * * All copy n successive bytes from source to destination. Memmove and memcpy * return destination, whereas bcopy has no return value. Copying takes place * as if it were through a temporary buffer -- after return destination * contains exactly the bytes from source, even if the buffers overlap (this is * not required of memcpy by the C standard; its behavior is undefined if the * buffers overlap, but we are holding ourselves to the historical behavior of * this function on MacOS). */ #include "asm.h" .globl _bcopy .globl _ovbcopy .globl _memcpy .globl _memmove /***************************************************************************** * Macros * *****************************************************************************/ #define kSmallCopy 64 /***************************************************************************** * Entrypoints * *****************************************************************************/ .text .align 5 _bcopy: _ovbcopy: // Translate bcopy into memcpy by swapping the first and second arguments. mov x3, x0 mov x0, x1 mov x1, x3 .align 4 _memcpy: _memmove: // Our preference is to copy the data in ascending address order, but if the // buffers overlap such that the beginning of the destination buffer aliases // the end of the source buffer, we need to copy in descending address order // instead to preserve the memmove semantics. We detect this case with the // test: // // destination - source < length (unsigned compare) // // If the address of the source buffer is higher than the address of the // destination buffer, this arithmetic can overflow, but the overflowed value // can only be smaller than length if the buffers do not overlap, so we don't // need to worry about false positives due to the overflow (they happen, but // only in cases where copying in either order is correct). PUSH_FRAME sub x3, x0, x1 cmp x3, x2 b.cc L_reverse mov x3, x0 // copy destination pointer cmp x2, #(kSmallCopy) b.cc L_forwardSmallCopy /***************************************************************************** * Forward large copy * *****************************************************************************/ // Load the first 32 bytes from src, and compute the number of bytes to the // first 32-byte aligned location in dst. Even though we are going to copy // 32 bytes, only those preceeding that 32-byte location "count" towards // reducing the length of the buffer or advancing the pointers. We will need // to issue the first load from the advanced src pointer BEFORE the store to // the unmodified dst pointer. add x3, x3, #32 and x3, x3, #-32 // aligned dst ldp x12,x13,[x1] ldp x14,x15,[x1, #16] sub x5, x3, x0 // bytes between original dst and aligned dst add x1, x1, x5 // update src pointer // At this point, data in the following registers is in flight: // // x0 original dst pointer // x1 corresponding location in src buffer. // x2 length from aligned location in dst to end of buffer. This is // guaranteed to be >= (64 - 32). // x3 aligned location in dst buffer. // x12:x15 first 32 bytes of src buffer. // // We now load 32 bytes from x1, and store 32 bytes from x12:x15 to x3. The // store *may* overlap the first 32 bytes of the load, so in order to get // correct memmove semantics, the first 32 byte load must occur before the // store. // // After loading these 32 bytes, we advance x1, and decrement the length by // 64. If the remaining length of the buffer was less than 64, then we jump // directly to the cleanup path. ldp x8, x9, [x1] ldp x10,x11,[x1, #16] add x1, x1, #32 sub x2, x2, x5 // update length stp x12,x13,[x0] // initial unaligned store stp x14,x15,[x0, #16] // initial unaligned store subs x2, x2, #64 b.ls L_forwardCleanup L_forwardCopyLoop: // Main copy loop: // // 1. store the 32 bytes loaded in the previous loop iteration // 2. advance the destination pointer // 3. load the next 32 bytes // 4. advance the source pointer // 5. subtract 32 from the length // // The loop is terminated when 32 or fewer bytes remain to be loaded. Those // trailing 1-32 bytes will be copied in the loop cleanup. stnp x8, x9, [x3] stnp x10,x11,[x3, #16] add x3, x3, #32 ldnp x8, x9, [x1] ldnp x10,x11,[x1, #16] add x1, x1, #32 subs x2, x2, #32 b.hi L_forwardCopyLoop L_forwardCleanup: // There are 32 bytes in x8-x11 that were loaded in the previous loop // iteration, which need to be stored to [x3,x3+32). In addition, between // 0 and 32 more bytes need to be copied from x1 to x3 + 32. The exact // number of bytes to copy is x2 + 32. Instead of using smaller conditional // copies, we simply copy 32 unaligned bytes from x1+x2 to 64+x3+x2. // This copy may overlap with the first store, so the loads must come before // the store of the data from the previous loop iteration. add x1, x1, x2 ldp x12,x13,[x1] ldp x14,x15,[x1, #16] stp x8, x9, [x3] stp x10,x11,[x3, #16] add x3, x3, x2 stp x12,x13,[x3, #32] stp x14,x15,[x3, #48] POP_FRAME ret /***************************************************************************** * forward small copy * *****************************************************************************/ // Copy one quadword at a time until less than 8 bytes remain to be copied. // At the point of entry to L_forwardSmallCopy, the "calling convention" // is as follows: // // x0 pointer to first byte of destination // x1 pointer to first byte of source // x2 length of buffers // x3 pointer to first byte of destination 0: ldr x6, [x1],#8 str x6, [x3],#8 L_forwardSmallCopy: subs x2, x2, #8 b.cs 0b adds x2, x2, #8 b.eq 2f 1: ldrb w6, [x1],#1 strb w6, [x3],#1 subs x2, x2, #1 b.ne 1b 2: POP_FRAME ret /***************************************************************************** * Reverse copy engines * *****************************************************************************/ // The reverse copy engines are identical in every way to the forward copy // engines, except in that they do everything backwards. For this reason, they // are somewhat more sparsely commented than the forward copy loops. I have // tried to only comment things that might be somewhat surprising in how they // differ from the forward implementation. // // The one important thing to note is that (almost without fail), x1 and x3 // will point to ONE BYTE BEYOND the "right-hand edge" of the active buffer // throughout these copy loops. They are initially advanced to that position // in the L_reverse jump island. Because of this, whereas the forward copy // loops generally follow a "copy data, then advance pointers" scheme, in the // reverse copy loops, we advance the pointers, then copy the data. L_reverse: // As a minor optimization, we early out if dst == src. cbz x3, L_return // advance both pointers to the ends of their respective buffers before // jumping into the appropriate reverse copy loop. add x4, x0, x2 add x1, x1, x2 cmp x2, #(kSmallCopy) b.cc L_reverseSmallCopy /***************************************************************************** * Reverse large copy * *****************************************************************************/ ldp x12,x13,[x1, #-16] ldp x14,x15,[x1, #-32] sub x3, x4, #1 // In the forward copy, we used dst+32 & -32 and x3, x3, #-32 // to find an aligned location in the dest sub x5, x4, x3 // buffer. Here we use dst-1 & -32 instead, sub x1, x1, x5 // because we are going backwards. sub x2, x2, x5 ldp x8, x9, [x1, #-16] ldp x10,x11,[x1, #-32] stp x12,x13,[x4, #-16] stp x14,x15,[x4, #-32] sub x1, x1, #32 subs x2, x2, #64 b.ls L_reverseCleanup L_reverseCopyLoop: stnp x8, x9, [x3, #-16] stnp x10,x11,[x3, #-32] sub x3, x3, #32 ldnp x8, x9, [x1, #-16] ldnp x10,x11,[x1, #-32] sub x1, x1, #32 subs x2, x2, #32 b.hi L_reverseCopyLoop L_reverseCleanup: sub x1, x1, x2 ldp x12,x13,[x1, #-16] ldp x14,x15,[x1, #-32] stp x8, x9, [x3, #-16] stp x10,x11,[x3, #-32] stp x12,x13,[x0, #16] // In the forward copy, we need to compute the stp x14,x15,[x0] // address of these stores, but here we already POP_FRAME // have a pointer to the start of the buffer. ret /***************************************************************************** * reverse small copy * *****************************************************************************/ 0: ldr x6, [x1,#-8]! str x6, [x4,#-8]! L_reverseSmallCopy: subs x2, x2, #8 b.cs 0b adds x2, x2, #8 b.eq 2f 1: ldrb w6, [x1,#-1]! strb w6, [x4,#-1]! subs x2, x2, #1 b.ne 1b 2: POP_FRAME ret L_return: POP_FRAME ret