NEW TOOLCHAIN: Add the ARMC6 Compiler #4949

theotherjimmy · 2017-08-21T19:05:39Z

Description

This implementation shares much more code with ARMC5 and comes with an
exporter: make_armc6. This all comes at a cost though: I was required to
remove the CPP invocations from the top of all of the scatter files and
add them back dynamically, I had to replace (or not) the scatter shebang dynamically, based on the toolchain being used.

Design Notes

This variant was designed to minimize the change set. In particular, ~~I removed all of the #!s from the top of Arm Compiler scatter files, and~~ I replace the shebangs at runtime (of the tools). This does impose a requirement that the tools must always be able to do this going forward, but they ~~should (and will, soon)~~ detect if a scatter file does not contain a preprocessor invocation or contains a compatible preprocessor invocation and use it directly without modification. This may lead to linker errors, but I hope that "command armcc not found" and "command armclang not found" will be descriptive enough to users.

Comparison to #4905

Advantages

Exporting is possible, and it even comes with a working uVision5 exporter (armc5 only) and make exporter pair (for both armc5 and armc6).
Coexistence of armc5 and armc6 code is allowed and supported
- Migration can then be done on a single target, single MCU, family or vendor basis

Disadvantages

There is no longer a single Arm toolchain, which might confuse users.
The tools do a bit of "magic" ~~by forcing all scatter files to be preprocessed.~~ that corrects armc5 invocations to armc6 invocations and armc6 invocations to armc5 invocations.

TODO:

Write section detailing this new toolchain in the Handbook
Add ARMC6 to the CI
Run nightly test to verify my implementation of ARMC6
Process dependency information for ARM6
Add small profile
Small error/warning messages
Test skip detection

theotherjimmy · 2017-08-21T19:07:29Z

Note this is an alternate implementation for #4905

c1728p9 · 2017-08-21T21:37:36Z

tools/toolchains/arm.py

-
-        if mem_map:
-            args.extend(["--scatter", mem_map])
+    def make_real_scatter(self, scatter_file):


What about leaving the shebang in as is for ARMC5 so all the linker script stay as they are and have make_real_scatter replace the ARMC5 shebang with the ARMC6 shebang only when building for ARMC6?

Sounds great! I'll do it first thing tomorrow.

0xc0170 · 2017-08-22T08:38:05Z

tools/toolchains/__init__.py

 from tools.toolchains.gcc import GCC_ARM
 from tools.toolchains.iar import IAR

 TOOLCHAIN_CLASSES = {
    'ARM': ARM_STD,
-    'uARM': ARM_MICRO,
+    'ARMC6': ARMC6,


are we removing uARM ,because ARM_MICRO class is still defined ? Is this breaking change or what is the intention here? or I have missed something - most probably

Did not mean to do that. Rebase coming anyway. I'll remove that change.

0xc0170 · 2017-08-22T08:43:01Z

tools/profiles/release.json

@@ -14,6 +14,13 @@
               "-Wl,--wrap,_calloc_r", "-Wl,--wrap,exit", "-Wl,--wrap,atexit",
               "-Wl,-n"]
    },
+    "ARMC6": {
+        "common": ["-c", "--target=arm-arm-none-eabi", "-mthumb", "-Oz", "-DNDEBUG", "-flto"],


isn't this duplicate flto as it is also defined in the ld ?

Curious about choosing - Oz for the release?

No. ld flags are not taken from common.

-Oz is for minimizing the size. That's what the release profile does. It used to be called small after all.

theotherjimmy · 2017-08-22T14:18:40Z

@c1728p9 @0xc0170 I think I have addressed all of your comments. Please take another look.

theotherjimmy · 2017-08-23T14:50:07Z

targets/targets.json

@@ -600,7 +600,7 @@
    "K64F": {
        "supported_form_factors": ["ARDUINO"],
        "core": "Cortex-M4F",
-        "supported_toolchains": ["ARM", "GCC_ARM", "IAR"],
+        "supported_toolchains": ["ARM", "GCC_ARM", "IAR", "ARMC6"],


Drop this change.

theotherjimmy · 2017-08-23T14:50:52Z

tools/toolchains/__init__.py


 TOOLCHAIN_PATHS = {
    'ARM': ARM_PATH,
-    'uARM': ARM_PATH,


theotherjimmy · 2017-08-24T18:35:53Z

I closed the other implementation, so I'm changing the title and removing the dupe label.

c1728p9 · 2017-08-25T14:46:05Z

...Freescale/TARGET_MCUXpresso_MCUS/TARGET_MCU_K64F/device/TOOLCHAIN_ARM_STD/MK64FN1M0xxx12.sct

@@ -122,4 +122,4 @@ LR_m_text m_interrupts_start m_text_start+m_text_size-m_interrupts_start {   ; l
  }
  RW_IRAM1 ImageLimit(RW_m_data_2) { ; Heap region growing up
  }
-}
+}


nitpick - this file doesn't need to be modified.

Yup. I'll rebase that out.

c1728p9 · 2017-08-25T14:51:40Z

tools/toolchains/arm.py

+        """
+        with open(scatter_file, "rb") as input:
+            lines = input.readlines()
+            if  (lines[0].startswith(self.SHEBANG) or


should the line be normalized before using this check? That way extra spaces won't cause this comparison to fail

The comparison failing would not be that bad: it would just emit another linker script with the correct shebang. I'm not sure I want to make that check complicated, but you're right that it can be improved to reduce the frequency of emitting linker scripts.

Good point. I didn't realize there is no harm done if the comparison fails. I agree, it would be good to keep it simple like it is now.

theotherjimmy · 2017-08-29T18:28:55Z

I need to update this PR to include support for all targets.

studavekar · 2017-09-11T22:45:41Z

@theotherjimmy armc6 looks good except for below 2

for target REALTEK_RTL8195AM,toolchain ARMC6 there are long of warning : http://mbed-ci-master-2.austin.arm.com:8081/job/build_matrix/1138/target=REALTEK_RTL8195AM,toolchain=ARMC6/console
KL46Z,NUCLEO_L073RZ fails http://mbed-ci-master-2.austin.arm.com:8081/job/build_matrix/1138/target=KL46Z,toolchain=ARMC6/consoleFull

00:00:44.012         [DEBUG] Output: ./TESTS/mbed_hal/flash/functional_tests/main.cpp:82:11: error: no flag-preserving variant of this instruction available
00:00:44.012         [DEBUG] Output:     "%=:\n\t"
00:00:44.013         [DEBUG] Output:           ^
00:00:44.013         [DEBUG] Output: <inline asm>:2:2: note: instantiated into assembly here
00:00:44.013         [DEBUG] Output:         SUB  r0, #1

mbed-bot · 2017-09-11T22:54:08Z

Result: FAILURE

Your command has finished executing! Here's what you wrote!

/morph test

Output

mbed Build Number: 1283

Build failed!

adbridge · 2017-09-12T10:01:58Z

/morph test

adbridge · 2017-09-12T14:49:46Z

So I assume this is the outstanding Realtek error

[DEBUG] Output: /home/arm/mbed_jenkins/workspace/bm_wrap/1374/mbed-os/targets/TARGET_Realtek/TARGET_AMEBA/TARGET_RTL8195A/device/rtl8195a_init.c:21:10: fatal error: 'cmsis_gcc.h' file not found
17:52:41         [DEBUG] Output: #include "cmsis_gcc.h"

In this file we have :

#if defined(__CC_ARM)
#include "cmsis_armcc.h"
#elif defined(__GNUC__)
#include "cmsis_gcc.h"
#else
#include <cmsis_iar.h>
#endif

So it looks like GNUC is being defined for ARMCC 6 ? Which is a bit weird in itself. Speaking to Bartek he says it should be including cmsis_armclang.h for ARMCC 6 so I think this needs to look like -

#if defined(__CC_ARM) 
#include "cmsis_armcc.h"
#elif (defined (__ARMCC_VERSION) && __ARMCC_VERSION >= 6010050)
#include "cmsis_armclang.h"
#elif defined(__GNUC__)
#include "cmsis_gcc.h"
#else
#include <cmsis_iar.h>
#endif

mbed-bot · 2017-09-12T15:07:18Z

Result: FAILURE

Your command has finished executing! Here's what you wrote!

/morph test

Output

mbed Build Number: 1288

Build failed!

studavekar · 2017-09-12T15:54:14Z

REALTEK_RTL8195AM,toolchain ARMC6 there are long of warning : http://mbed-ci-master-2.austin.arm.com:8081/job/build_matrix/1138/target=REALTEK_RTL8195AM,toolchain=ARMC6/console

Build time for REALTEK_RTL8195AM :

ARM 5 - 4 min
ARM 6 - 1 hr 27 min

The temp fix to this is reduce the verbosity of logs, caveat to reduced info in case of failure.

theotherjimmy · 2017-09-12T15:54:37Z

So it looks like __GNUC__ is being defined for ARMCC 6 ?

Yep

theotherjimmy · 2017-09-12T15:55:48Z

@adbridge Yes, that's the correct fix. Thanks for pushing it!

adbridge · 2017-09-12T16:00:03Z

/morph test

studavekar · 2017-09-12T16:26:26Z

/morph test

mbed-bot · 2017-09-12T16:27:10Z

Result: FAILURE

Your command has finished executing! Here's what you wrote!

/morph test

Output

mbed Build Number: 1289

Build failed!

studavekar · 2017-09-12T16:30:58Z

/morph test

mbed-bot · 2017-09-12T19:53:53Z

Result: FAILURE

Your command has finished executing! Here's what you wrote!

/morph test

Output

mbed Build Number: 1291

Example Build failed!

studavekar · 2017-09-12T20:33:37Z

/morph test

mbed-bot · 2017-09-13T06:12:42Z

Result: FAILURE

Your command has finished executing! Here's what you wrote!

/morph test

Output

mbed Build Number: 1294

Example Build failed!

studavekar · 2017-09-13T06:18:08Z

Run with ARM 6 build disabled since we have following issue to be resolved

Build time for REALTEK device : NEW TOOLCHAIN: Add the ARMC6 Compiler #4949 (comment)
NORDIC application are not build properly missing boot loader resulting in device fails to boot the firmware.

/morph test

studavekar · 2017-09-13T06:43:38Z

/morph test

mbed-bot · 2017-09-13T08:55:48Z

Result: SUCCESS

Your command has finished executing! Here's what you wrote!

/morph test

Output

mbed Build Number: 1298

All builds and test passed!

adbridge · 2017-09-13T09:37:27Z

@studavekar Thanks for getting this through CI.

theotherjimmy added type: duplicate needs: review labels Aug 21, 2017

theotherjimmy changed the title ~~NEW TOOLCHAIN: ARMC6, alternate implementation~~ NEW TOOLCHAIN: Add the ARMC6 Compiler, alternate implementation Aug 21, 2017

c1728p9 reviewed Aug 21, 2017

View reviewed changes

0xc0170 reviewed Aug 22, 2017

View reviewed changes

theotherjimmy force-pushed the feature-armc5+armc6 branch 2 times, most recently from d3591e7 to 6a50eb2 Compare August 22, 2017 14:17

theotherjimmy commented Aug 23, 2017

View reviewed changes

tools/toolchains/__init__.py Outdated

TOOLCHAIN_PATHS = {

'ARM': ARM_PATH,

'uARM': ARM_PATH,

Copy link

Contributor Author

theotherjimmy Aug 23, 2017

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Fix.

theotherjimmy force-pushed the feature-armc5+armc6 branch from 61818a5 to 9c779d9 Compare August 23, 2017 15:06

This was referenced Aug 24, 2017

NEW TOOLCHAIN: Add the ARMC6 compiler #4905

Closed

Add ARMC6 where appropriate ARMmbed/mbed-os-5-docs#205

Merged

theotherjimmy removed the type: duplicate label Aug 24, 2017

theotherjimmy changed the title ~~NEW TOOLCHAIN: Add the ARMC6 Compiler, alternate implementation~~ NEW TOOLCHAIN: Add the ARMC6 Compiler Aug 24, 2017

c1728p9 reviewed Aug 25, 2017

View reviewed changes

theotherjimmy force-pushed the feature-armc5+armc6 branch from 9c779d9 to 8bd662a Compare August 25, 2017 15:13

c1728p9 approved these changes Aug 25, 2017

View reviewed changes

theotherjimmy force-pushed the feature-armc5+armc6 branch from 8bd662a to cb7c509 Compare August 28, 2017 20:46

theotherjimmy added needs: work and removed needs: review labels Aug 29, 2017

deepikabhavnani mentioned this pull request Aug 29, 2017

Add Nuvoton m2351device #4892

Closed

theotherjimmy force-pushed the feature-armc5+armc6 branch 5 times, most recently from 45c3dc8 to b7c1072 Compare August 29, 2017 22:16

studavekar mentioned this pull request Sep 12, 2017

use gcc assembly for arm 6 #5073

Merged

Fix include file for ARMCC 6, should use cmsis_armclang.h

32bd25e

ARMmbed deleted a comment from mbed-bot Sep 12, 2017

add ARMC6 in supported compiler

fec2346

adbridge added ready for merge and removed needs: CI labels Sep 13, 2017

adbridge merged commit 7b42891 into ARMmbed:master Sep 13, 2017

adbridge removed the ready for merge label Sep 13, 2017

theotherjimmy deleted the feature-armc5+armc6 branch September 13, 2017 14:02

jeromecoutant mentioned this pull request Oct 5, 2017

NUCLEO_L496ZG: Add new Platform #4650

Merged

0xc0170 pushed a commit that referenced this pull request Nov 3, 2017

remove sys.cpp file of the ARM_STD toolchain as done in the MR #4949

3198920

NEW TOOLCHAIN: Add the ARMC6 Compiler #4949

NEW TOOLCHAIN: Add the ARMC6 Compiler #4949

Uh oh!

Conversation

theotherjimmy commented Aug 21, 2017 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Description

Design Notes

Comparison to #4905

Advantages

Disadvantages

TODO:

Uh oh!

theotherjimmy commented Aug 21, 2017

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

theotherjimmy commented Aug 22, 2017

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

theotherjimmy commented Aug 24, 2017

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

theotherjimmy commented Aug 29, 2017

Uh oh!

studavekar commented Sep 11, 2017 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

mbed-bot commented Sep 11, 2017

Output

Uh oh!

adbridge commented Sep 12, 2017

Uh oh!

adbridge commented Sep 12, 2017 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

mbed-bot commented Sep 12, 2017

Output

Uh oh!

studavekar commented Sep 12, 2017

Uh oh!

theotherjimmy commented Sep 12, 2017

Uh oh!

theotherjimmy commented Sep 12, 2017

Uh oh!

adbridge commented Sep 12, 2017

Uh oh!

studavekar commented Sep 12, 2017

Uh oh!

mbed-bot commented Sep 12, 2017

Output

Uh oh!

studavekar commented Sep 12, 2017

Uh oh!

mbed-bot commented Sep 12, 2017

Output

Uh oh!

studavekar commented Sep 12, 2017

theotherjimmy commented Aug 21, 2017 •

edited

Loading

studavekar commented Sep 11, 2017 •

edited

Loading

adbridge commented Sep 12, 2017 •

edited

Loading