diff options
Diffstat (limited to 'libc/string/bfin/memmove.S')
| -rw-r--r-- | libc/string/bfin/memmove.S | 97 |
1 files changed, 54 insertions, 43 deletions
diff --git a/libc/string/bfin/memmove.S b/libc/string/bfin/memmove.S index 3d446f326..da1128387 100644 --- a/libc/string/bfin/memmove.S +++ b/libc/string/bfin/memmove.S @@ -1,5 +1,5 @@ /* memmove.S - * Copyright (C) 2003, 2005, 2006 Analog Devices Inc., All Rights Reserved. + * Copyright (C) 2003-2007 Analog Devices Inc., All Rights Reserved. * * This file is subject to the terms and conditions of the GNU Library General * Public License. See the file "COPYING.LIB" in the main directory of this @@ -9,6 +9,8 @@ * http://www.analog.com/processors/resources/crosscore/visualDspDevSoftware.html */ +#include <sysdep.h> + /* void *memmove(void *dest, const void *src, size_t n); * R0 = To Address (dest) (leave unchanged to form result) * R1 = From Address (src) @@ -21,75 +23,84 @@ .align 2 -.global _memmove -.type _memmove, STT_FUNC -_memmove: + /* We have to bypass the libc-symbols.h machinery to make sure we get + a weak symbol for memcpy (some crummy gcc tests want to redefine + it). */ +.global ___GI_memmove +.type ___GI_memmove, STT_FUNC +___GI_memmove: I1 = P3; - P0 = R0; // P0 = To address - P3 = R1; // P3 = From Address - P2 = R2 ; // P2 = count - CC = P2 == 0; //Check zero count - IF CC JUMP finished; // very unlikely + P0 = R0; /* P0 = To address */ + P3 = R1; /* P3 = From Address */ + P2 = R2; /* P2 = count */ + CC = P2 == 0; /* Check zero count*/ + IF CC JUMP .Lfinished; /* very unlikely */ - CC = R1 < R0 (IU); // From < To - IF !CC JUMP no_overlap; + CC = R1 < R0 (IU); /* From < To */ + IF !CC JUMP .Lno_overlap; R3 = R1 + R2; - CC = R0 <= R3 (IU); // (From+len) >= To - IF CC JUMP overlap; -no_overlap: + CC = R0 <= R3 (IU); /* (From+len) >= To */ + IF CC JUMP .Loverlap; +.Lno_overlap: R3 = 11; CC = R2 <= R3; - IF CC JUMP bytes; - R3 = R1 | R0; // OR addresses together - R3 <<= 30; // check bottom two bits - CC = AZ; // AZ set if zero. - IF !CC JUMP bytes ; // Jump if addrs not aligned. + IF CC JUMP .Lbytes; + R3 = R1 | R0; /* OR addresses together */ + R3 <<= 30; /* check bottom two bits */ + CC = AZ; /* AZ set if zero.*/ + IF !CC JUMP .Lbytes; /* Jump if addrs not aligned.*/ I0 = P3; - P1 = P2 >> 2; // count = n/4 + P1 = P2 >> 2; /* count = n/4 */ P1 += -1; R3 = 3; - R2 = R2 & R3; // remainder - P2 = R2; // set remainder + R2 = R2 & R3; /* remainder */ + P2 = R2; /* set remainder */ R1 = [I0++]; #if !defined(__WORKAROUND_AVOID_DAG1) - LSETUP (quad_loop , quad_loop) LC0=P1; -quad_loop: MNOP || [P0++] = R1 || R1 = [I0++]; + LSETUP (.Lquad_loop, .Lquad_loop) LC0=P1; +.Lquad_loop: MNOP || [P0++] = R1 || R1 = [I0++]; #else - LSETUP (quad_loop_s, quad_loop_e) LC0=P1; -quad_loop_s: [P0++] = R1; -quad_loop_e: R1 = [I0++]; + LSETUP (.Lquad_loop_s, .Lquad_loop_e) LC0=P1; +.Lquad_loop_s: [P0++] = R1; +.Lquad_loop_e: R1 = [I0++]; #endif [P0++] = R1; - CC = P2 == 0; // any remaining bytes? - P3 = I0; // Ammend P3 to updated ptr. - IF !CC JUMP bytes; + CC = P2 == 0; /* any remaining bytes? */ + P3 = I0; /* Ammend P3 to updated ptr. */ + IF !CC JUMP .Lbytes; P3 = I1; RTS; -bytes: LSETUP (byte2_s , byte2_e) LC0=P2; -byte2_s: R1 = B[P3++](Z); -byte2_e: B[P0++] = R1; +.Lbytes: LSETUP (.Lbyte2_s, .Lbyte2_e) LC0=P2; +.Lbyte2_s: R1 = B[P3++](Z); +.Lbyte2_e: B[P0++] = R1; -finished: - P3 = I1; +.Lfinished: P3 = I1; RTS; -overlap: +.Loverlap: P2 += -1; P0 = P0 + P2; P3 = P3 + P2; R1 = B[P3--] (Z); CC = P2 == 0; - IF CC JUMP no_loop; - LSETUP (ol_s, ol_e) LC0 = P2; -ol_s: B[P0--] = R1; -ol_e: R1 = B[P3--] (Z); -no_loop: B[P0] = R1; + IF CC JUMP .Lno_loop; +#if defined(__WORKAROUND_SPECULATIVE_LOADS) + NOP; + NOP; +#endif + LSETUP (.Lol_s, .Lol_e) LC0 = P2; +.Lol_s: B[P0--] = R1; +.Lol_e: R1 = B[P3--] (Z); +.Lno_loop: B[P0] = R1; P3 = I1; RTS; -.size _memmove,.-_memmove -libc_hidden_def (memmove) +.size ___GI_memmove,.-___GI_memmove + +.hidden ___GI_memmove +.weak _memmove +.set _memmove,___GI_memmove |