Correct the floating options for Cortex-M7 and Cortex-M4 #9477
Conversation
Cortex-M23 / Cortex-M33 CPU settings for baseline and mainline profile (with optional floating and dsp options) updated.
As per the IAR Development guide, below options for CPU are valid 1. Cortex-M33 2. Cortex-M33.no_dsp (core without integer DSP extension) 3. Cortex-M33.fp (floating-point unit with support for single precision) 4. Cortex-M33.no_se (core without support for TrustZone)
deepikabhavnani
force-pushed
the
iar_fp
branch
from
4064835 to
d5e1860
Compare
As per the IAR Development guide, below options for CPU are valid 1. Cortex-M4 2. Cortex-M4F 3. Cortex-M7 4. Cortex-M7.fp.dp (floating-point unit with support for double precision) 5. Cortex-M7.fp.sp (floating-point unit with support for single precision)
deepikabhavnani
force-pushed
the
iar_fp
branch
from
d5e1860 to
afb559b
Compare
|
@deepikabhavnani, thank you for your changes. |
| #endif | ||
|
|
||
|
|
||
|
|
There was a problem hiding this comment.
they are dummy defines in code, may be old trial code
There was a problem hiding this comment.
I see, but it happens to be in target code which we are modifying/removing, hope its ok with Nuvoton.
| elif target.core == "Cortex-M7F": | ||
| cpuchoice = "Cortex-M7.fp.sp" | ||
| elif target.core.startswith("Cortex-M33FD"): | ||
| cpuchoice = "Cortex-M33.fp" |
There was a problem hiding this comment.
Wonder why this is not fp.dp (double-prec) as in M7? Is this correct, please double-check?
There was a problem hiding this comment.
D in case of Cortex-M33 is DSP and not double floating point.
There was a problem hiding this comment.
Which is a very bad idea, as I believe I've previously commented. Why are we making up non-standard inconsistent naming?
If you really want to have a single-letter name, "M33FE" would make more sense, as "E" is the letter historically used by ARM to represent the DSP Extensions. (Although that makes me think of the ARM7500FE where it meant EDO RAM, so maybe reverse to M33EF - putting the DSP extensions closer to the core than the FP unit is also logical).
But I would really rather go with what all the tools are consistently doing - using the "no_fp" and "no_dsp" suffixes. We're in the odd situation where all 3 toolchains agree on that, and Mbed OS is being different. (Okay, they don't totally agree on polarity of the FP flag)
There was a problem hiding this comment.
While performing some changes recently, I see why Mbed OS used its own letters, as we do not have consistency among all toolchaina and even between assembler and compilers. Flags are +nodsp / .no_dsp / .dsp / +dsp
There was a problem hiding this comment.
single-letter name, "M33FE"
Yes can change this now to some other letter for DSP, as no target is added.
@mbed-os-tools - thoughts on this please
There was a problem hiding this comment.
Please make the core names consistent with the ARM historical single letter abbreviations; use E for DSP.
There was a problem hiding this comment.
@theotherjimmy @kjbracey-arm - One more Query, for DSP and FPU. Should it be F-Eor FE
Floating point - F and FD looks fine, but with addition of E, FDE will be confusing. And will be allow any order here? like FD-E or E-FD?
Description
As per the IAR Development guide, below options for CPU are valid and separate --fpu is not needed.
Commit for this PR is afb559b
Dependent on : #9431
Pull request type
Reviewers
@kjbracey-arm