- Ask a related questionWhat is a related question?A related question is a question created from another question. When the related question is created, it will be automatically linked to the original question.
Part Number: MSP432P4111
Tool/software: TI C/C++ Compiler
I found what appears to be a Compiler bug. When evaluating a simple inequality expression (x < y) using double float numbers, the evaluation is sometimes wrong (e.g. evaluates False when it should be TRUE). The fault case I found involved comparing a constant with a variable (that was set to MIN_DOUBLE). Other cases fail or succeed depending on values used. The issue is only seen when using the ccs TI Compiler, no fault seen using the gcc GNU Compiler. I need help understanding and solving this problem. I included a very simple code example below loaded onto an Evaluation Board (MSP-EXP432P4111) that illustrates the problem.
Components:
Code Composer Studio: CCS ver 7.2.0.00013
Compiler: TI v18.1.4.LTS
Compiler: GNU v 6.3.1 (Linaro)
/*******************************************************************************
*
*
*******************************************************************************/
#include <ti/devices/msp432p4xx/driverlib/driverlib.h>
#include <stdbool.h>
#include <stdio.h>
#define MIN_DOUBLE (double)4.94065645841247E-324
double x;
bool bool1, bool2, bool3, bool4;
int main(void)
{
MAP_WDT_A_holdTimer();
x = MIN_DOUBLE;
bool1 = (0.0 < x); // should evaluate to TRUE, but incorrectly evaluates to FALSE (when using TI Compiler)
bool2 = (0.0 < MIN_DOUBLE); // should evaluate to TRUE, correctly evaluate to TRUE
bool3 = (0.0 > x); // should evaluate to FALSE, correctly evaluate to FALSE
bool4 = (0.0 > MIN_DOUBLE); // should evaluate to FALSE, correctly evaluate to FALSE
printf("bool1 = %d\r", bool1);
printf("bool2 = %d\r", bool2);
printf("bool3 = %d\r", bool3);
printf("bool4 = %d\r", bool4);
fflush (stdout);
while(1);
}
Output from CCS Console
bool1 = 0 // this value is wrong
bool2 = 1
bool3 = 0
bool4 = 0
link: 
Type Size Representation Minimum Maximum
signed char 8 bits ASCII -128 127
char (1) 8 bits ASCII 0 (1) 255 (1)
unsigned char 8 bits ASCII 0 255
bool, _Bool 8 bits ASCII 0 (false) 1(true)
short, signed short 16 bits Binary -32 768 32 767
unsigned short, wchar_t (2) 16 bits Binary 0 65 535
int, signed int 32 bits Binary -2 147 483 648 2 147 483 647
unsigned int 32 bits Binary 0 4 294 967 295
long, signed long 32 bits Binary -2 147 483 648 2 147 483 647
unsigned long 32 bits Binary 0 4 294 967 295
long long, signed long long 64 bits(3) Binary -9 223 372 036 854 775 808 9 223 372 036 854 775 807
unsigned long long 64 bits(3) Binary 0 18 446 744 073 709 551 615
enum (TI_ARM9_ABI and TIABI only) (4) 32 bits Binary -2 147 483 648 2 147 483 647
float 32 bits IEEE 32-bit 1.175 494e-38(5) 3.40 282 346e+38
double 64 bits(3) IEEE 64-bit 2.22 507 385e-308(5) 1.79 769 313e+308
long double 64 bits(3) IEEE 64-bit 2.22 507 385e-308(5) 1.79 769 313e+308
pointers, references, pointer to data members 32 bits Binary 0 0xFFFFFFFF
(1) "Plain" char has the same representation as either signed char or unsigned char. The --plain_char option specifies whether "plain" char is signed or unsigned. The default is unsigned.
(2) This is the default type for wchar_t. You can use the --wchar_t option to change the wchar_t type to a 32-bit unsigned int type.
(3) 64-bit data is aligned on a 64-bit boundary.
(4) For details about the size of an enum type, see Section 5.5.1. Also see Table 5-2 for sizes.
(5) Figures are minimum precision.
Negative values for signed types are represented using two's complement.
What is allowed for subnormalized values? I do not know.
