1#------------------------------------------------------------------------------ 2# 3# CopyMem() worker for ARM 4# 5# This file started out as C code that did 64 bit moves if the buffer was 6# 32-bit aligned, else it does a byte copy. It also does a byte copy for 7# any trailing bytes. It was updated to do 32-byte copies using stm/ldm. 8# 9# Copyright (c) 2008 - 2010, Apple Inc. All rights reserved.<BR> 10# This program and the accompanying materials 11# are licensed and made available under the terms and conditions of the BSD License 12# which accompanies this distribution. The full text of the license may be found at 13# http://opensource.org/licenses/bsd-license.php 14# 15# THE PROGRAM IS DISTRIBUTED UNDER THE BSD LICENSE ON AN "AS IS" BASIS, 16# WITHOUT WARRANTIES OR REPRESENTATIONS OF ANY KIND, EITHER EXPRESS OR IMPLIED. 17# 18#------------------------------------------------------------------------------ 19 20/** 21 Copy Length bytes from Source to Destination. Overlap is OK. 22 23 This implementation 24 25 @param Destination Target of copy 26 @param Source Place to copy from 27 @param Length Number of bytes to copy 28 29 @return Destination 30 31 32VOID * 33EFIAPI 34InternalMemCopyMem ( 35 OUT VOID *DestinationBuffer, 36 IN CONST VOID *SourceBuffer, 37 IN UINTN Length 38 ) 39**/ 40.text 41.align 2 42GCC_ASM_EXPORT(InternalMemCopyMem) 43 44ASM_PFX(InternalMemCopyMem): 45 stmfd sp!, {r4-r11, lr} 46 // Save the input parameters in extra registers (r11 = destination, r14 = source, r12 = length) 47 mov r11, r0 48 mov r10, r0 49 mov r12, r2 50 mov r14, r1 51 52memcopy_check_overlapped: 53 cmp r11, r1 54 // If (dest < source) 55 bcc memcopy_check_optim_default 56 // If (dest <= source). But with the previous condition -> If (dest == source) 57 bls memcopy_end 58 59 // If (source + length < dest) 60 rsb r3, r1, r11 61 cmp r12, r3 62 bcc memcopy_check_optim_default 63 64 // If (length == 0) 65 cmp r12, #0 66 beq memcopy_end 67 68 b memcopy_check_optim_overlap 69 70memcopy_check_optim_default: 71 // Check if we can use an optimized path ((length >= 32) && destination word-aligned && source word-aligned) for the memcopy (optimized path if r0 == 1) 72 tst r0, #0xF 73 movne r0, #0 74 bne memcopy_default 75 tst r1, #0xF 76 movne r3, #0 77 moveq r3, #1 78 cmp r2, #31 79 movls r0, #0 80 andhi r0, r3, #1 81 b memcopy_default 82 83memcopy_check_optim_overlap: 84 // r10 = dest_end, r14 = source_end 85 add r10, r11, r12 86 add r14, r12, r1 87 88 // Are we in the optimized case ((length >= 32) && dest_end word-aligned && source_end word-aligned) 89 cmp r2, #31 90 movls r0, #0 91 movhi r0, #1 92 tst r10, #0xF 93 movne r0, #0 94 tst r14, #0xF 95 movne r0, #0 96 b memcopy_overlapped 97 98memcopy_overlapped_non_optim: 99 // We read 1 byte from the end of the source buffer 100 sub r3, r14, #1 101 sub r12, r12, #1 102 ldrb r3, [r3, #0] 103 sub r2, r10, #1 104 cmp r12, #0 105 // We write 1 byte at the end of the dest buffer 106 sub r10, r10, #1 107 sub r14, r14, #1 108 strb r3, [r2, #0] 109 bne memcopy_overlapped_non_optim 110 b memcopy_end 111 112// r10 = dest_end, r14 = source_end 113memcopy_overlapped: 114 // Are we in the optimized case ? 115 cmp r0, #0 116 beq memcopy_overlapped_non_optim 117 118 // Optimized Overlapped - Read 32 bytes 119 sub r14, r14, #32 120 sub r12, r12, #32 121 cmp r12, #31 122 ldmia r14, {r2-r9} 123 124 // If length is less than 32 then disable optim 125 movls r0, #0 126 127 cmp r12, #0 128 129 // Optimized Overlapped - Write 32 bytes 130 sub r10, r10, #32 131 stmia r10, {r2-r9} 132 133 // while (length != 0) 134 bne memcopy_overlapped 135 b memcopy_end 136 137memcopy_default_non_optim: 138 // Byte copy 139 ldrb r3, [r14], #1 140 sub r12, r12, #1 141 strb r3, [r10], #1 142 143memcopy_default: 144 cmp r12, #0 145 beq memcopy_end 146 147// r10 = dest, r14 = source 148memcopy_default_loop: 149 cmp r0, #0 150 beq memcopy_default_non_optim 151 152 // Optimized memcopy - Read 32 Bytes 153 sub r12, r12, #32 154 cmp r12, #31 155 ldmia r14!, {r2-r9} 156 157 // If length is less than 32 then disable optim 158 movls r0, #0 159 160 cmp r12, #0 161 162 // Optimized memcopy - Write 32 Bytes 163 stmia r10!, {r2-r9} 164 165 // while (length != 0) 166 bne memcopy_default_loop 167 168memcopy_end: 169 mov r0, r11 170 ldmfd sp!, {r4-r11, pc} 171 172