 |
|
|
MAEC TOOL NEWS:
MAECT-M3T-CC32R-011201D
Please take note of the following problem in using cross-tool kit M3T-CC32R for the M32R family MCUs.
- On optimization at a function taking 4 arguments or more which calls another function taking 3 arguments or less.
- Versions Concerned
M3T-CC32R V.1.00 Release 1 -- V.3.10 Release 1
- Description
When an optimizing option covering the -04 option is used for a program in which a function taking 4 arguments or more calls another function taking 3 arguments or less, the fourth argument of the former may be treated incorrectly.
- 2.1 Conditions
- This problem may occur if the following six conditions are satisfied:
- (1) An optimizing option covering the -O4 option is used (any of the following options is selected: -O4, -O5, -O6, -O7, -Otime only, or -Ospace only).
- (2) A function have 4 or more arguments and the first four arguments are of respective types that can be passed to registers, (that is, either integer, pointer, or float-type variables). *1)
- (3) Any of the following variables are dynamically initialized or assigned a value in the function described in (2) above:
- (a) a whole structure
- (b) a variable of type double
- (c) an array
- (4) After the processing of (3), a function taking 3 arguments or less is called from the function described in (2).
- (5) After the function call in (4), the fourth argument of the function described in (2) is referenced.
- (6) The dynamic initialization or assignment in (3) is replaced with a _100_builtin_memcopy function at compilation. *2)
- Notes:
- *1) This condition cannot be met in CC32R V.2.10 and earlier because no variables of type float are passed to registers in these versions.
- *2) To check whether Condition (6) is met or not, search for a character string "_100_builtin_memcopy" in the assembly source file generated with the -S option used. For details of how to check this, see the latter half of the following section.
- 2.2 Examples
Example 1: Source file "sample1.c"
--------------------------------------------------------------------
extern int foo1(int, int, int);
struct {int a; int b[30]; int c;} A, B;
int
func1(int a, int b, int c, int d) /* Condition (2) */
{
A = B; /* Condition (3)-(a):
Assigns a structure B to A */
foo1(a, b, c); /* Condition (4) */
return d; /* Condition (5) */
}
--------------------------------------------------------------------
Example 2: Source file "sample2.c"
--------------------------------------------------------------------
extern int foo2(int, int, int)
int
func2(int a,int b,int c,int d) /* Condition (2) */
{
int x[3] = {193,156,119}; /* Condition (3)-(c) */
foo2(x[a], b, c); /* Condition (4) */
return d; /* Condition (5) */
}
--------------------------------------------------------------------
Example of typing commands in M3T-CC32R (a % sign denotes a prompt)
--------------------------------------------------------------------
In "sample1.c"
% cc32R -S -O4 sample1.c ; Condition (1) met.
% find "_100_builtin_memcopy" sample1.ms ; Condition (6) checked.
If Condition (6) met:
---------------------------------------
---------- sample1.ms
.IMPORT $_100_builtin_memcopy
BL $_100_builtin_memcopy
If Condition (6) not met:
---------------------------------------
---------- sample1.ms ; Nothing displayed.
In "sample2.c"
% cc32R -S -Ospace sample2.c ; Condition (1) met.
% find "_100_builtin_memcopy" sample2.ms ; Condition (6) checked.
If Condition (6) met:
---------------------------------------
---------- sample2.ms .IMPORT $_100_builtin_memcopy
BL $_100_builtin_memcopy
If Condition (6) not met:
---------------------------------------
---------- sample2.ms ; Nothing displayed.
--------------------------------------------------------------------
Note: On UNIX-based EWS, use a grep command instead of a find command.
- Workaround
- This problem will be circumvented in any of the following ways:
- (1) Suppress optimization in the -O4 level.
Optimization covering the -O4 option will be performed when optimizing option -04, -O5, -O6, -O7, -Otime only or -Ospace only has been selected. If you want to use -Otime or -Ospace, select any of these options, -O0, -O1, -O2, and -O3, at the same time.
- (2) Change the dynamic initialization of a variable or the assignment of a value to a variable into the call of a memcpy function that produces the equivalent result.
[Example 1: Modification of Source file "sample1.c"]
--------------------------------------------------------------------
#include <string.h> /* Added to call a memcpy function */
extern int foo1(int, int, int);
struct {int a; int b[30]; int c;} A, B;
int
func1(int a, int b, int c, int d)
{
memcpy(&A, &B, sizeof(A)); /* Assigning a structure changed
to calling a memcpy function */
foo1(a, b, c);
return d;
}
--------------------------------------------------------------------
[Example 2: Modification of Source file "sample2.c"]
--------------------------------------------------------------------
#include <string.h> /* Added to call a memcpy function */
extern int foo2(int, int, int)
int
func2(int a,int b,int c,int d)
{ int x[3]; /* Initialization of
array x canceled */
static int x_init[3] = {193,156,119}; /* Added to declare and
initialize array x_init */
memcpy(x, x_init, sizeof(x)); /* A memcpy function called
to initialize array x */
foo2(x[a], b, c);
return d; }
--------------------------------------------------------------------
- (3) Change dynamic initialization to static initialization.
Note that this circumvention cannot be applied to any programs that assume dynamic initialization.
[Example 3: Modification of Source file "sample2.c"]
--------------------------------------------------------------------
extern int foo2(int, int, int)
int
func2(int a,int b,int c,int d)
{
static int x[3] = {193,156,119}; /* Dynamic initialization
changed to static one */
foo2(x[a], b, c);
return d;
}
--------------------------------------------------------------------
- Schedule of Fixing the Problem
We plan to fix this problem in our next release.
|
|
