Documentation, comments and code formatting improvements
Small documentation corrections.
Improve comments in codes, spelling corrections, comment formatting
Wrap long source code lines so there are fewer lines over 80 columns and very few over 120 columns
NO changes to actual source or interface
diff --git a/src/ieee754.c b/src/ieee754.c
index 6fdfda8..ea44e1d 100644
--- a/src/ieee754.c
+++ b/src/ieee754.c
@@ -1,35 +1,42 @@
/*==============================================================================
- ieee754.c -- floating point conversion between half, double and single precision
+ ieee754.c -- floating-point conversion between half, double & single-precision
- Copyright (c) 2018-2019, Laurence Lundblade. All rights reserved.
+ Copyright (c) 2018-2020, Laurence Lundblade. All rights reserved.
SPDX-License-Identifier: BSD-3-Clause
See BSD-3-Clause license in README.md
Created on 7/23/18
- ==============================================================================*/
+ =============================================================================*/
#include "ieee754.h"
#include <string.h> // For memcpy()
/*
- This code is written for clarity and verifiability, not for size, on the assumption
- that the optimizer will do a good job. The LLVM optimizer, -Os, does seem to do the
- job and the resulting object code is smaller from combining code for the many different
- cases (normal, subnormal, infinity, zero...) for the conversions.
+ This code is written for clarity and verifiability, not for size, on
+ the assumption that the optimizer will do a good job. The LLVM
+ optimizer, -Os, does seem to do the job and the resulting object code
+ is smaller from combining code for the many different cases (normal,
+ subnormal, infinity, zero...) for the conversions.
- Dead stripping is also really helpful to get code size down when floating point
- encoding is not needed.
+ This code has really long lines and is much easier to read because of
+ them. Some coding guidelines prefer 80 column lines (can they not afford
+ big displays?). It would make this code much worse even to wrap at 120
+ columns.
- This code works solely using shifts and masks and thus has no dependency on
- any math libraries. It can even work if the CPU doesn't have any floating
- point support, though that isn't the most useful thing to do.
+ Dead stripping is also really helpful to get code size down when
+ floating-point encoding is not needed.
- The memcpy() dependency is only for CopyFloatToUint32() and friends which only
- is needed to avoid type punning when converting the actual float bits to
- an unsigned value so the bit shifts and masks can work.
+ This code works solely using shifts and masks and thus has no
+ dependency on any math libraries. It can even work if the CPU doesn't
+ have any floating-point support, though that isn't the most useful
+ thing to do.
+
+ The memcpy() dependency is only for CopyFloatToUint32() and friends
+ which only is needed to avoid type punning when converting the actual
+ float bits to an unsigned value so the bit shifts and masks can work.
*/
/*
@@ -69,7 +76,7 @@
#define HALF_EXPONENT_INF_OR_NAN (HALF_EXPONENT_BIAS+1) // 16 Unbiased
-// ------ Single Precision --------
+// ------ Single-Precision --------
#define SINGLE_NUM_SIGNIFICAND_BITS (23)
#define SINGLE_NUM_EXPONENT_BITS (8)
#define SINGLE_NUM_SIGN_BITS (1)
@@ -96,7 +103,7 @@
#define SINGLE_EXPONENT_INF_OR_NAN (SINGLE_EXPONENT_BIAS+1) // 128 unbiased
-// --------- Double Precision ----------
+// --------- Double-Precision ----------
#define DOUBLE_NUM_SIGNIFICAND_BITS (52)
#define DOUBLE_NUM_EXPONENT_BITS (11)
#define DOUBLE_NUM_SIGN_BITS (1)
@@ -349,7 +356,7 @@
uSingleSign = uHalfSign;
- // Shift the three parts of the single precision into place
+ // Shift the three parts of the single-precision into place
const uint32_t uSinglePrecision = uSingleSignificand |
(uSingleBiasedExponent << SINGLE_EXPONENT_SHIFT) |
(uSingleSign << SINGLE_SIGN_SHIFT);