summaryrefslogtreecommitdiffstats
path: root/libtoolame-dab/portableio.h
diff options
context:
space:
mode:
Diffstat (limited to 'libtoolame-dab/portableio.h')
-rw-r--r--libtoolame-dab/portableio.h89
1 files changed, 89 insertions, 0 deletions
diff --git a/libtoolame-dab/portableio.h b/libtoolame-dab/portableio.h
new file mode 100644
index 0000000..70b7d23
--- /dev/null
+++ b/libtoolame-dab/portableio.h
@@ -0,0 +1,89 @@
+#ifndef PORTABLEIO_H__
+#define PORTABLEIO_H__
+/* Copyright (C) 1988-1991 Apple Computer, Inc.
+ * All Rights Reserved.
+ *
+ * Warranty Information
+ * Even though Apple has reviewed this software, Apple makes no warranty
+ * or representation, either express or implied, with respect to this
+ * software, its quality, accuracy, merchantability, or fitness for a
+ * particular purpose. As a result, this software is provided "as is,"
+ * and you, its user, are assuming the entire risk as to its quality
+ * and accuracy.
+ *
+ * This code may be used and freely distributed as long as it includes
+ * this copyright notice and the warranty information.
+ *
+ * Machine-independent I/O routines for 8-, 16-, 24-, and 32-bit integers.
+ *
+ * Motorola processors (Macintosh, Sun, Sparc, MIPS, etc)
+ * pack bytes from high to low (they are big-endian).
+ * Use the HighLow routines to match the native format
+ * of these machines.
+ *
+ * Intel-like machines (PCs, Sequent)
+ * pack bytes from low to high (the are little-endian).
+ * Use the LowHigh routines to match the native format
+ * of these machines.
+ *
+ * These routines have been tested on the following machines:
+ * Apple Macintosh, MPW 3.1 C compiler
+ * Apple Macintosh, THINK C compiler
+ * Silicon Graphics IRIS, MIPS compiler
+ * Cray X/MP and Y/MP
+ * Digital Equipment VAX
+ *
+ *
+ * Implemented by Malcolm Slaney and Ken Turkowski.
+ *
+ * Malcolm Slaney contributions during 1988-1990 include big- and little-
+ * endian file I/O, conversion to and from Motorola's extended 80-bit
+ * FLOATing-point format, and conversions to and from IEEE single-
+ * precision FLOATing-point format.
+ *
+ * In 1991, Ken Turkowski implemented the conversions to and from
+ * IEEE double-precision format, added more precision to the extended
+ * conversions, and accommodated conversions involving +/- infinity,
+ * NaN's, and denormalized numbers.
+ *
+ * $Id: portableio.h,v 2.6 1991/04/30 17:06:02 malcolm Exp $
+ *
+ * $Log: portableio.h,v $
+ * Revision 2.6 91/04/30 17:06:02 malcolm
+ */
+
+#include <stdio.h>
+#include "ieeefloat.h"
+
+int ReadByte (FILE * fp);
+int Read16BitsLowHigh (FILE * fp);
+int Read16BitsHighLow (FILE * fp);
+void Write8Bits (FILE * fp, int i);
+void Write16BitsLowHigh (FILE * fp, int i);
+void Write16BitsHighLow (FILE * fp, int i);
+int Read24BitsHighLow (FILE * fp);
+int Read32Bits (FILE * fp);
+int Read32BitsHighLow (FILE * fp);
+void Write32Bits (FILE * fp, int i);
+void Write32BitsLowHigh (FILE * fp, int i);
+void Write32BitsHighLow (FILE * fp, int i);
+void ReadBytes (FILE * fp, char *p, int n);
+void ReadBytesSwapped (FILE * fp, char *p, int n);
+void WriteBytes (FILE * fp, char *p, int n);
+void WriteBytesSwapped (FILE * fp, char *p, int n);
+double ReadIeeeFloatHighLow (FILE * fp);
+double ReadIeeeFloatLowHigh (FILE * fp);
+double ReadIeeeDoubleHighLow (FILE * fp);
+double ReadIeeeDoubleLowHigh (FILE * fp);
+double ReadIeeeExtendedHighLow (FILE * fp);
+double ReadIeeeExtendedLowHigh (FILE * fp);
+void WriteIeeeFloatLowHigh (FILE * fp, double num);
+void WriteIeeeFloatHighLow (FILE * fp, double num);
+void WriteIeeeDoubleLowHigh (FILE * fp, double num);
+void WriteIeeeDoubleHighLow (FILE * fp, double num);
+void WriteIeeeExtendedLowHigh (FILE * fp, double num);
+void WriteIeeeExtendedHighLow (FILE * fp, double num);
+
+#define Read32BitsLowHigh(f) Read32Bits(f)
+#define WriteString(f,s) fwrite(s,strlen(s),sizeof(char),f)
+#endif