readme.1st

Montavista Linux 下的多平台编译软件xdc

------------------------------------------------------------------------------
SDSsq33163
------------------------------------------------------------------------------
Summary  : Dividing zero by zero results in INF rather than NaN

Dividing zero by zero results in INF(Infinite) rather than NaN(Not a 
Number)

const double x = 0.0 / 0.0;
const double y = 1.0 / 0.0;
results in x being assigned the value +Inf (0x7FF0000000000000) rather 
than NaN of some flavour (e.g. 0x7FFC000000000000); y is correctly set 
to +Inf.

------------------------------------------------------------------------------
SDSsq33513
------------------------------------------------------------------------------
Summary  : Optimizer assumes any floating pt value, also NaN ,is equal to
           itself  

The Optimizer, for c6x CCS2.2 CGT v 4.32, appears to assume that any 
floating point value is equal to itself. This is not true - NaN is not 
equal to itself. Thus, as a very simplified example, the function

int isnan(double val)
{
  return (val != val);
}

just returns zero on all occasions: which is wrong.

------------------------------------------------------------------------------
SDSsq35081
------------------------------------------------------------------------------
Summary  : When using long filenames, ar6x may not correctly add file to
           library

If a file with a long file name is added to an archive library using 
the command "ar6x r", it is possible that the file will replace a file 
which already exists in the archive.

Workaround:
-----------

Add new files using the ar6x a command instead of using the r command:

ar6x a library.lib this_is_a_really_long_filename.obj

instead of

ar6x r library.lib this_is_a_really_long_filename.obj

------------------------------------------------------------------------------
SDSsq35348
------------------------------------------------------------------------------
Summary  : Consultant does not properly mark do-while loops

I have a do-while loop in my program that is formatted as follows:

do
{
   /* loop body */
}
while (/* some condition */)

When I view this loop in the consultant, it is not indicating that 
the "while" statement is part of the loop.

------------------------------------------------------------------------------
SDSsq37015
------------------------------------------------------------------------------
Summary  : Data align pragma does not work correctly if constant value is
           0x10000

Data align pragma is not working correcly if constant value is for an
alignment > 32k.

------------------------------------------------------------------------------
SDSsq39228
------------------------------------------------------------------------------
Summary  : assembler expression ~(0x80000000) evaluates as 0x80000000

The assembler may generate unexpected results when evaluating constant 
expressions with large hex constants with the MS bit set.  For example:

        .sslist

        .eval   ~(0xc000000), mask2
x       .macro
        .mmsg   ":mask2:"
        .endm

        x

The mask printed by this message is 0x8000000 when it is expected to 
be 0x4fffffff.

------------------------------------------------------------------------------
SDSsq39254
------------------------------------------------------------------------------
Summary  : Linker -xml_link_info option doesn't work when in a command file

When using --xml_link_info option inside a linker command file,
use the following syntax ...

   --xml_link_info <xml filename>

   Note the use of a space between the option and the filename,
   rather than the '=' shown in documentation.

------------------------------------------------------------------------------
SDSsq39285
------------------------------------------------------------------------------
Summary  : hex6x map file is missing the boot loader section

The generated map file does not indicate that the bootloader has been
stored in a section, instead it indicates that the section was filled with 
zero's.  The hex file does contain the correct bootloader code.  

------------------------------------------------------------------------------
SDSsq39552
------------------------------------------------------------------------------
Summary  : fastrts67x library hangs on rts fmod

Linking the fastrts67x library causes the following code to hang.
When removing the fastrts67x.lib (or divdp.asm from the library) fmod
returns correctly.


------------------------------------------------------------------------------
SDSsq39619
------------------------------------------------------------------------------
Summary  : Float literal conversion to integers incorrectly saturates at 32
           bits

When a floating point constant with a value of larger than pow(2,32) is
converted (explicitly or implicitly) to a target integer type wider than
32 bits, the compiler will incorrectly saturate the value to 32 bits.
Values larger than (pow(2,32)-1) will be saturated incorrectly when
converting to a wide unsigned type, and values outside the range
(-pow(2,31))..(pow(2,31)-1) will be saturated incorrectly when
converting to a wide signed type.

Workaround:
-----------

Use float expressions rather than float constants when they occur as the
argument to an explicit or implicit conversion.  A simple way to do this
is with a function:

float make_float(float x) { return x; }
void fn() { long long y = make_float(5e11); [..] }

------------------------------------------------------------------------------
SDSsq39708
------------------------------------------------------------------------------
Summary  : Compiler should reject bitfields too large for container type

The compiler should reject bitfields larger than the underlying integer
type.  This works fine in most cases, but the compiler fails to reject
bitfields of size 41 through 64 with "long" type.  "long" is a legal
bitfield container type in C++, but it is only 40 bits long.

------------------------------------------------------------------------------
SDSsq41626
------------------------------------------------------------------------------
Summary  : Parser generates an INTERNAL ERROR message

Compiling the following code below with any options will result in the
compiler returning an "INTERNAL ERROR" message --

struct foo {char x, y, z[2];};
struct foo f();

void bar(int baz)
{
    f().z[baz] = 1;
}

------------------------------------------------------------------------------
SDSsq42312
------------------------------------------------------------------------------
Summary  : Optimizer removes loop where only output is sat bit

The optimizer may remove a loop whose only output is the saturation 
bit in the CSR.

Workaround:
-----------
The programmer must make something else "live-out" of the loop.

------------------------------------------------------------------------------
SDSsq42346
------------------------------------------------------------------------------
Summary  : Cannot transform saturation intrinsics when the SAT bit is read

The optimizer may sometimes transform saturated intrinsic operations 
in such a way that the SAT bit in the CSR register is no longer set 
when saturation occurs.  If the original code is reading the sat bit 
in the CSR, the sat bit may not have been set correctly.

Workaround:
-----------
Use linear assembly or assembly to write the function that does 
saturation operations and call that function before testing the sat 
bit.

------------------------------------------------------------------------------
SDSsq42776
------------------------------------------------------------------------------
Summary  : .end directive can cause assembler error msgs about DWARF directives

Use of the .end directive in linear assembly source code can lead to 
error messages being emitted by the assembler about illegal debug 
directives --

cl6x <options> foo.sa
"foo.asm", ERROR!   at line 81: [E0000] Illegal specification of a DIE
                                              reference
                 .dwattr $C$DW$1, DW_AT_type(*$C$DW$T$10)

1 Assembly Error, No Assembly Warnings

------------------------------------------------------------------------------
SDSsq43034
------------------------------------------------------------------------------
Summary  : ASM code behavior difference between DA6xx and DA7xx

Hand assembly code written for the C67x/DA6xx which did not have 
functional unit designations for ADDSP, SUBSP, ADDDP, and SUBDP 
instructions may be assigned to the S-unit by the assembler when 
assembling for the C67x+/DA7xx.  This may result in a S-unit resource 
conflict with other sebsequent S-unit instructions that the assembler 
does not detect, resulting in code that runs incorrectly.

Workaround:
-----------

When porting hand coded assembly from the C67x/DA6xx to the 
C67x+/DA7xx architecture make sure that ADDSP, SUBSP, ADDDP, and SUBDP 
instructions are assigned to the L-unit by placinga .L unit specifcy 
on them.

------------------------------------------------------------------------------
SDSsq43742
------------------------------------------------------------------------------
Summary  : STRUCT_ALIGN pragma doesn't work for nested structs

STRUCT_ALIGN pragma doesn't work when applied to nested structures.


------------------------------------------------------------------------------
SDSsq43823
------------------------------------------------------------------------------
Summary  : Linker hangs when trying to create a relocatable output module

Linking with the -r (produce relocatable output module) option can 
cause the linker to hang if the source files were compiled with debug 
(-g) enabled.

cl6x -g file1.c file2.c
cl6x -z -r -o linked.obj file1.obj file2.obj
cl6x -z -r -o linked2.obj linked.obj <-- linker hangs here

Workaround:
-----------

On the second linker invocation, include the -b (disable merging of 
symbolic debug information) linker option

cl6x -z -b -r -o linked2.obj linked.obj

------------------------------------------------------------------------------
SDSsq43826
------------------------------------------------------------------------------
Summary  : Link takes a long time to complete

Linking with the -r (produce relocatable output module) option can cause
the linker to hang if the source files were compiled with debug (-g)
enabled. 

cl6x -g file1.c file2.c 
cl6x -z -r -o linked.obj file1.obj file2.obj 
cl6x -z -r -o linked2.obj linked.obj <-- linker hangs here 

Workaround:
-----------

On the second linker invocation, include the -b (disable merging of
symbolic debug information) linker option 

cl6x -z -b -r -o linked2.obj linked.obj


-------------------------------------------------------------------------------
load6x does not support C64+ and C67+
-------------------------------------------------------------------------------
The standalone simulator load6x does not support running programs build for
C64+ or C67+ architectures.


*******************************************************************************
4. Fixed Issues      
*******************************************************************************

++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
                                Release 6.0.1B2
++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++

------------------------------------------------------------------------------
Fixed SDSsq39117
------------------------------------------------------------------------------
Summary  : Linear assembler does not handle .include/.copy directive

------------------------------------------------------------------------------
Fixed SDSsq42320
------------------------------------------------------------------------------
Summary  : Loop incorrectly transformed at -o3