1. Product and Versions Concerned
The C/C++ compiler package--M3T-CC32R--for the M32R MCU family
V.4.00 Release 1 through V.5.01 Release 00
2. Description
If the -m32re5 and -fminst options, which enables using of the FMSUB
instruction, and an optimizing option are selected in compilation, an
incorrect operand of the FMSUB instruction may be generated for the
operations consisting of multiplication and subtraction both of
type float.
2.1 Conditions
This problem may occur if the following conditions are all satisfied:
(1) The -m32re5 and -fminst options are both selected in compilation.
(2) The optimizing option used in compilation is either of the
following:
(a) Any of the options -O, -O1, -O3, -O5, and -O7
(b) Either -Ospace or -Otime with none of the options -O0, -O2,
-O4, and -O6 being selected.
(3) The following float-type operations are both performed:
(a) A multiplication whose both terms are of type float
(b) A subtraction where a value of type float is subtracted from
the result of (a)
(4) The subtraction in (3)-(b) refers to the auto variable to which
the result in (3)-(a) is assigned.
2.2 Examples
Source code (sample.c):
-------------------------------------------------------------------
float A, B, C;
float func(void)
{
float var;
float answer;
. . . . . . .
var = A * B; /* Condition (3)-(a) */
answer = var - C; /* Conditions (3)-(b) and (4) */
. . . . . . .
}
-------------------------------------------------------------------
In the above example, the problem may not occur depending on the
content of the code written in the omitted lines.
Command line:
-------------------------------------------------------------------
cc32R -c -O7 -m32re5 -fminst sample.c (Conditions (1) and (2))
-------------------------------------------------------------------
In the above example of source code, the correct expression for
calculating "answer" is A * B - C. However, in the case when the
problem arises, an incorrect instruction "FMSUB A,B,C" is generated,
which means A - B * C.
3. Workaround
To avoid generating incorrect FMSUB instructions, compile the source
program by using the internal control option -Qa-Xs16.
Example of command line:
-------------------------------------------------------------------
cc32R -c -O7 -m32re5 -fminst -Qa-Xs16 sample.c
-------------------------------------------------------------------
