This commit removes the previous tdesc_cortex_a, tdesc_cortex_m, and
tdesc_cortex_mf XML string literals used for target description to GDB,
now instead programmatically generating them at runtime to significantly
deduplicate the characters that get embedded into the binary.
Output of ld's --print-memory-usage during final link before:
Memory region Used Size Region Size %age Used
rom: 116388 B 128 KB 88.80%
ram: 3348 B 20 KB 16.35%
Output of ld's --print-memory-usage during final link now:
Memory region Used Size Region Size %age Used
rom: 113032 B 128 KB 86.24%
ram: 3376 B 20 KB 16.48%
So all in all this saves 3356 bytes of flash.
Note: the exact size saved when compiled on your machine may differ, as
the size of the build seems at least partially non-deterministic.
We've gotten slightly different sizes (within 15 bytes of each other)
at different times, with the only differences being things like which
files were rebuilt in an incremental rebuild, or the order object files
were given to the linker command line. The numbers given above were the
numbers we got when testing the final builds from scratch, but all the
sizes we got were extremely similar to the sizes listed above.
Ubuntu's default /bin/sh is dash, which does not support the `&>`
redirection syntax. This commit moves version.h generation back into the
Makefile, as 8afaedd had it, but restores compatibility with
GNU Make < 4.0, which 8afaedd, broke. This also fixes building on macOS,
as macOS bundles GNU Make 3.81.
The target was building but not including the BMDA binary, as the
build system does not expect a binary that does not end with `.bin`.
Additionally this corrects the BMDA Makefile.inc that was missing the
`all` and `host_clean` targets.
More recent versions of GNU make remove the `-` prefix from
`$(MAKEFLAGS)`, which breaks the build:
```bash
# output truncated to interesting parts
> make -C src --trace all_platforms
set -e ;\
mkdir -p artifacts/v1.7.1-91-g98b4ec5 ;\
echo "<html><body><ul>" > artifacts/index.html ;\
for i in platforms/*/Makefile.inc ; do \
...\
make w --trace --no-print-directory clean ;\
make w --trace --no-print-directory;\
...\
done ;\
echo "</ul></body></html>" >> artifacts/index.html ;\
cp artifacts/*.bin artifacts/v1.7.1-91-g98b4ec5
Building for hardware platform: f4discovery
make[1]: *** No rule to make target 'w'. Stop.
```
Per the
[documentation](https://www.gnu.org/software/make/manual/html_node/Options_002fRecursion.html),
there's no need to manually apply insert `$(MAKEFLAGS)`, it should be
automatically included when using `$(MAKE)` when recursing.
I tested on my machine (make 4.2.1) and a docker image based on
ubuntu:trusty (make 3.81), both behaved the same (directories were not
printed when recursing by default);
Adds a target driver for Microchip SAM D51 / E5x family.
Tested on SAMD51G19A and SAMD51J19A. According to the datasheet, the
D51 / E5x family share the same core feature set, differing only in the
addition of CAN (E51) or ethernet controllers (E53/54). All members of
the family should be equivalent from a debug and programming perspective.
-Wall on gcc8 otherwise warns without -Wno-cast-function-type but older
GCCs/CLang choke on that argument:
error: unknown warning option '-Wno-cast-function-type'; did you mean
'-Wno-bad-function-cast'? [-Werror,-Wunknown-warning-option]
This adds 24 byte to the binary, as some functions are now called with
additional dummy arguments:
"Pushing and popping garbage to keep the system happy"
After update of GNU make the tests for Windows OS are no longer needed. Note that GNU make 4.2.1 is used to build BMP under Windows OS.
Make native working
Again, the newer version of GCC make does not need the OS tests.
This allows the platform to specify another optimization level than
the default (like -Os if space is limited)
Signed-off-by: Sylvain Munaut <tnt@246tNt.com>
Support for Kinetis KE04 8KiB, 64KiB and 128KiB variants in nxpke04.c
Target monitor commands for sector and mass erase.
Changes to kinetis.c MDM-AP target to support KE04.
Only KE04Z8 tested in HW.