ESP32 : make option MICROPY_FLOAT_IMPL

[rcolistete]

Implement make option MICROPY_FLOAT_IMPL on ESP32, so it is simpler to build ESP32 firmware with single or double precision float point numbers. For example :
[esp32]$ make -j8 MICROPY_FLOAT_IMPL=double
instead of changing the 'esp32/mpconfigport.h', line :
#define MICROPY_FLOAT_IMPL (MICROPY_FLOAT_IMPL_FLOAT)
to
#define MICROPY_FLOAT_IMPL (MICROPY_FLOAT_IMPL_DOUBLE)
Batch scripts to build many ESP32 firmware variants with combinations of boards (GENERIC, GENERIC_SPIRAM, etc), single/double precision, etc, become possible.

This option MICROPY_FLOAT_IMPL can also be configured by each board via $BOARD/mpconfigboard.mk.

In this way the ESP32 building workflow becomes more compatible with STM32 one.

[stinos]

Does the STM32 also have a flag like this?
Because as far as I know this is the same as doing make CFLAGS_EXTRA="-DMICROPY_FLOAT_IMPL=MICROPY_FLOAT_IMPL_DOUBLE" i.e. it works already like that for all other platforms.

[rcolistete]

Does the STM32 also have a flag like this?
Because as far as I know this is the same as doing make CFLAGS_EXTRA="-DMICROPY_FLOAT_IMPL=MICROPY_FLOAT_IMPL_DOUBLE" i.e. it works already like that for all other platforms.

Yes, micropython/ports/stm32/Makefile has this MICROPY_FLOAT_IMPL flag due to the following code :

# Configure floating point support
ifeq ($(MICROPY_FLOAT_IMPL),double)
CFLAGS += -DMICROPY_FLOAT_IMPL=MICROPY_FLOAT_IMPL_DOUBLE
else
ifeq ($(MICROPY_FLOAT_IMPL),none)
CFLAGS += -DMICROPY_FLOAT_IMPL=MICROPY_FLOAT_IMPL_NONE
else
CFLAGS += -DMICROPY_FLOAT_IMPL=MICROPY_FLOAT_IMPL_FLOAT
CFLAGS += -fsingle-precision-constant
endif
endif

From which I adapted, except the line :
CFLAGS += -fsingle-precision-constant

[rcolistete]

Also, the majority (all?) of people compiling ESP32 with double precision is changing the 'esp32/mpconfigport.h' file :

• "Re: double precision float on ESP8266" (but also refers to ESP32);
• "teuler/robotling - Adding native modules to MicroPython (ESP32)";
• etc;

as they don't see (on ESP32) the make flag/option used by STM32 :

• "Low float precision #4380" issue;
• "Re: Pyboard D double precision" topic;
• etc.

[stinos]

Ah I just figured STM32 has this flag because it also alters the build itself (like adding lib/libm_dbl source when using double) so for STM32 just setting CFLAGS_EXTRA won't cut it.

[rcolistete]

Does the STM32 also have a flag like this?
Because as far as I know this is the same as doing make CFLAGS_EXTRA="-DMICROPY_FLOAT_IMPL=MICROPY_FLOAT_IMPL_DOUBLE" i.e. it works already like that for all other platforms.

After a Internet search, I don't see one example at all using this flag (CFLAGS_EXTRA=...) to compile double precision, besides the Makefile. So this flag is less known by the community, not cited in any docs/tutorial/forum.

Again, it is a lot more practical and known to use :

$ make MICROPY_FLOAT_IMPL=double

[dpgeorge]

This is now obsolete, after the move to CMake.

So I'll close the PR. Feel free to rework it so it works with the latest code and reopen.