- Moved coefficients computation to init function
- Improved documentation

The new function:
- is used to accelerate the 1x1 convolution layer
- does not require temporary buffers
- does not have any limitations on the input dimensions

Added macros for:
__PMU_PRESENT (set to 1 by default)
__PMU_NUM_EVENTCNT (set to 31 by default - max for Armv8.1-M)

The new implementation does not require temporary buffers

Now environment variables can be set and unset per toolchain and core.

Range of input offset is corrected from int8 to -127 to 128

Change-Id: I72e2a436171dd8923aa501d80090bb01c485d2ac

Added ARMCM55 device.

Description of missing kernel_x argument is added as doxygen
comment.

Change-Id: I3fe63abe74aaa097a4315f240dad15fce1b19132

CMSIS-Core documentation: partition_<device>.h includes now partition_gen.h (that is generated using CMSIS-Zone); added FPU settings to partition_<device>.h

Compilation error when ROUNDING=ON in MVE code.

Added a regression script to test several configurations of
the CMSIS-DSP.

Change-Id: If8d6d74e567b849ffb6823db05ec3b5edaed41e8

SVDConv:

fixed:
- Anonymous struct bitfield-members are ansi-c compatible, just the qualifier "__IM" needed to be deleted.

added:
- Cortex-M55
- PMU handling

Change-Id: I8097e29c542e07b61d18d9e77a30eb92ceefafeb

Change-Id: Ieb7c6dd5c1bc46cf73aa6496f199c79f4c8682e4

Change-Id: I3241aa219d633c52d4dcad517900893a76e23918