Disable Arm Compiler 5 and Use Arm Compiler 6 by default #5406

theotherjimmy · 2017-10-30T22:18:39Z

No description provided.

theotherjimmy · 2017-10-31T15:37:07Z

/morph build

mbed-ci · 2017-10-31T16:29:02Z

Build : FAILURE

Build number : 393
Build artifacts/logs : http://mbed-os.s3-website-eu-west-1.amazonaws.com/?prefix=builds/5406/

studavekar · 2017-10-31T16:34:45Z

@theotherjimmy ARMC6 is no more supported which was used to test ARMC6 ? Should be just use the default ARM now with this PR and moving forward ?

00:01:52.506 mbed test --compile -m ARCH_PRO -t ARMC6 -DMBED_HEAP_STATS_ENABLED=1 -DMBED_STACK_STATS_ENABLED=1 -DMBED_TRAP_ERRORS_ENABLED=1
00:01:52.850 [mbed] WARNING: Could not find mbed program in current path "/builds/ws/mbed-os-build-matrix/393/ARCH_PRO_ARMC6/sources/mbed-os".
00:01:52.850 [mbed] WARNING: You can fix this by calling "mbed new ." in the root of your program.
00:01:52.850 ---
00:01:53.546 usage: test.py [-h] [-m MCU] [-t TOOLCHAIN] [--color] [--cflags CFLAGS]
00:01:53.546                [--asmflags ASMFLAGS] [--ldflags LDFLAGS] [-c]
00:01:53.546                [--profile PROFILE] [--app-config APP_CONFIG] [-D MACROS]
00:01:53.546                [-j JOBS] [--source SOURCE_DIR] [--build BUILD_DIR] [-l]
00:01:53.546                [-p PATHS] [-f FORMAT] [--continue-on-build-fail] [-n NAMES]
00:01:53.546                [--test-config TEST_CONFIG] [--test-spec TEST_SPEC]
00:01:53.546                [--build-report-junit BUILD_REPORT_JUNIT]
00:01:53.546                [--build-data BUILD_DATA] [-v] [--stats-depth STATS_DEPTH]
00:01:53.546 test.py: error: argument -t/--tool: ARMC6 is not a supported toolchain. Supported toolchains are:
00:01:53.546 ARM,     GCC_ARM, IAR

mbed-ci · 2017-10-31T19:09:49Z

Build : FAILURE

Build number : 398
Build artifacts/logs : http://mbed-os.s3-website-eu-west-1.amazonaws.com/?prefix=builds/5406/

theotherjimmy · 2017-10-31T19:11:21Z

Well crud. It still does not fit.

theotherjimmy · 2017-10-31T19:28:48Z

I removed the last commit, as it introduced incompatibility with the STD library. We are just going to have to live with wchar and enum size differences.

theotherjimmy · 2017-11-01T14:31:20Z

/morph build

theotherjimmy · 2017-11-01T14:31:39Z

I removed the only offending test from most targets (it tests the toolchains)

mbed-ci · 2017-11-01T15:04:23Z

Build : FAILURE

Build number : 403
Build artifacts/logs : http://mbed-os.s3-website-eu-west-1.amazonaws.com/?prefix=builds/5406/

pan- · 2017-11-02T13:58:57Z

I look at the C++ standard library size issue. The main offender is the vector class which pulls many things from the library: iostream, some exception support, and so on. On the other hand standard algorithms are fine and don't pull unrelated dependencies.

Looking at libc++ source code and our compilation options vector shouldn't pull all these dependencies. After some digging, I found out that the those symbols are pulled in when the vector base class std::__1::__vector_base_common<1> is imported from an implicit instantiation in libcpp_wnu.l(locale.cpp.o). More specifically, the code is pulled in when the member function __throw_length_error is used.

To avoid import of implicitly instantiated symbols, I've explicitly instantiated the class; templates instantiated implicitly generate weak symbols while template instantiated explicitly generates strong symbols therefore symbols resulting of an explicit instantiation will be picked up at link time.

The results are dramatically different:

Implicit instantiation:

+--------------+----------+-----------+------------+-------------+
| name         | target   | toolchain | static_ram | total_flash |
+--------------+----------+-----------+------------+-------------+
| stl_features | NRF52_DK | ARM       |      12576 |      129483 |
+--------------+----------+-----------+------------+-------------+

Explicit instantiation

+--------------+----------+-----------+------------+-------------+
| name         | target   | toolchain | static_ram | total_flash |
+--------------+----------+-----------+------------+-------------+
| stl_features | NRF52_DK | ARM       |       9109 |       49675 |
+--------------+----------+-----------+------------+-------------+

As surprising as it is, the code generated for std::__1::__vector_base_common<1>::__throw_length_error explicitly instantiated in the test source is similar to the one implicitly instantiated in libcpp; it is a simple call to abort.

Maybe there is something wrong with our link parameters which keep contents from locale.cpp.o. The call graph generated by the linker doesn't help, many symbols not referenced are kept in the final executable.

For the record, the problem can be reproduced with the following code:

#include <vector>

#define EXPLICIT_INSTANTIATION

/**
 * @brief provide access to a private class member function or member data
 * @details It is possible to explicitly instantiate a template class with a
 * private pointer to member. When such instantiation occur, a friend function
 * getMember is instantiated. This function will return the pointer to member
 * used at the instantiation.
 *
 * @tparam MemberAccessor Act as a key to resolve the correct overload of getMember
 * @tparam MemberType The type of the pointer to member that will be hijacked
 * @tparam Member The pointer to member hijacked
 */
template<typename MemberAccessor, typename MemberType, MemberType Member>
struct hijacked_member_accessor {
    friend MemberType getMember(MemberAccessor) {
        return Member;
    }
};

/**
 * @brief Provide an accessors to a pointer to member, even if this one is private
 *
 * @param ACCESSOR name of the accessor to create
 * @param MEMBER_TYPE type of the pointer to member
 * @param MEMBER the pointer to member hijacked
 */
#define HIJACK_MEMBER(ACCESSOR, MEMBER_TYPE, MEMBER) \
    struct ACCESSOR##_CLASS__ { \
            ACCESSOR##_CLASS__() { } \
            friend MEMBER_TYPE getMember(ACCESSOR##_CLASS__);\
        };\
    template struct hijacked_member_accessor<ACCESSOR##_CLASS__, MEMBER_TYPE, MEMBER>; \
    static MEMBER_TYPE const volatile ACCESSOR = getMember(ACCESSOR##_CLASS__())

typedef std::__1::__vector_base_common<1> vector_base_common;
typedef __attribute__((noreturn)) void (vector_base_common::*throw_length_error_type)() const;


// set to 1 to explicitly instante std::__1::__vector_base_common<1>
#ifdef EXPLICIT_INSTANTIATION
template class std::__1::__vector_base_common<1>;
#endif

// create an accessor to the member function pointer vector_base_common::__throw_length_error
// named throw_length_error_accessor
HIJACK_MEMBER(
    throw_length_error_accessor,
    throw_length_error_type,
    &vector_base_common::__throw_length_error
);

int main() {
    return 0;
}

The binary sizes are:

EXPLICIT_INSTANTIATION not defined:

+-----------------------+--------+-------+-------+
| Module                |  .text | .data |  .bss |
+-----------------------+--------+-------+-------+
| FileBase.o            |    118 |     0 |    56 |
| FilePath.o            |    158 |     0 |     0 |
| I2C.o                 |    116 |     0 |     0 |
| Mutex.o               |    154 |     0 |     0 |
| Thread.o              |     20 |     0 |     4 |
| TimerEvent.o          |    142 |     0 |     0 |
| [lib]/c_wu.l          |  21389 |    16 |   348 |
| [lib]/fz_wm.l         |   1296 |     0 |     0 |
| [lib]/libcpp_wnu.l    |  69703 |     0 |  3467 |
| [lib]/libcppabi_wnu.l |   2395 |     0 |     0 |
| [lib]/m_wm.l          |    870 |     0 |     0 |
| anon$$obj.o           |     32 |     0 |     0 |
| cmsis_nvic.o          |     16 |     0 |     0 |
| gpio_api.o            |    682 |     0 |   268 |
| irq_cm4f.o            |    184 |     0 |     0 |
| lp_ticker.o           |     94 |     0 |     0 |
| main.o                |     32 |     0 |     8 |
| mbed_assert.o         |     84 |     0 |     0 |
| mbed_board.o          |    370 |     0 |     0 |
| mbed_boot.o           |    440 |     0 |  4260 |
| mbed_critical.o       |     64 |     0 |     0 |
| mbed_error.o          |     56 |     0 |     1 |
| mbed_gpio.o           |    172 |     0 |     0 |
| mbed_lp_ticker_api.o  |     58 |     0 |    64 |
| mbed_retarget.o       |   1353 |     0 |   124 |
| mbed_rtx_handlers.o   |    657 |     0 |     0 |
| mbed_rtx_idle.o       |    658 |     4 |    52 |
| mbed_sleep_manager.o  |    232 |     0 |     2 |
| mbed_ticker_api.o     |    904 |     0 |     0 |
| mbed_us_ticker_api.o  |     58 |     0 |    64 |
| mbed_wait_api_rtos.o  |     90 |     0 |     0 |
| nordic_critical.o     |    330 |     0 |     5 |
| nrf_drv_common.o      |     72 |     0 |     0 |
| nrf_drv_gpiote.o      |   1366 |     0 |   105 |
| reloc_vector_table.o  |     40 |     0 |   216 |
| rt_OsEventObserver.o  |      0 |     0 |     4 |
| rtx_delay.o           |     76 |     0 |     0 |
| rtx_evr.o             |    114 |     0 |     0 |
| rtx_kernel.o          |   1355 |   176 |     0 |
| rtx_lib.o             |    364 |    24 |  2341 |
| rtx_memory.o          |    248 |     0 |     0 |
| rtx_mempool.o         |    272 |     0 |     0 |
| rtx_msgqueue.o        |   1830 |     0 |     0 |
| rtx_mutex.o           |    898 |     0 |     0 |
| rtx_system.o          |    404 |     0 |     0 |
| rtx_thread.o          |   2568 |     0 |     0 |
| rtx_timer.o           |    256 |     0 |     0 |
| serial_api.o          |   1178 |     0 |   136 |
| sleep.o               |    114 |     0 |     0 |
| softdevice_handler.o  |     12 |     0 |    13 |
| startup_nrf52832.o    |    264 |     0 |     0 |
| system_nrf52.o        |    620 |     4 |     0 |
| us_ticker.o           |    552 |     0 |     5 |
| Subtotals             | 115530 |   224 | 11543 |
+-----------------------+--------+-------+-------+
Total Static RAM memory (data + bss): 11767 bytes
Total Flash memory (text + data): 115754 bytes

EXPLICIT_INSTANTION defined:

+-----------------------+-------+-------+------+
| Module                | .text | .data | .bss |
+-----------------------+-------+-------+------+
| FileBase.o            |   118 |     0 |   56 |
| FilePath.o            |   158 |     0 |    0 |
| I2C.o                 |   116 |     0 |    0 |
| Mutex.o               |   154 |     0 |    0 |
| Thread.o              |    20 |     0 |    4 |
| TimerEvent.o          |   142 |     0 |    0 |
| [lib]/c_wu.l          | 11251 |    16 |  348 |
| [lib]/fz_wm.l         |    18 |     0 |    0 |
| [lib]/libcppabi_wnu.l |    44 |     0 |    0 |
| [lib]/m_wm.l          |    48 |     0 |    0 |
| anon$$obj.o           |    32 |     0 |    0 |
| cmsis_nvic.o          |    16 |     0 |    0 |
| gpio_api.o            |   682 |     0 |  268 |
| irq_cm4f.o            |   184 |     0 |    0 |
| lp_ticker.o           |    94 |     0 |    0 |
| main.o                |    36 |     0 |    8 |
| mbed_assert.o         |    84 |     0 |    0 |
| mbed_board.o          |   370 |     0 |    0 |
| mbed_boot.o           |   440 |     0 | 4260 |
| mbed_critical.o       |    64 |     0 |    0 |
| mbed_error.o          |    56 |     0 |    1 |
| mbed_gpio.o           |   172 |     0 |    0 |
| mbed_lp_ticker_api.o  |    58 |     0 |   64 |
| mbed_retarget.o       |   897 |     0 |  124 |
| mbed_rtx_handlers.o   |   657 |     0 |    0 |
| mbed_rtx_idle.o       |   658 |     4 |   52 |
| mbed_sleep_manager.o  |   232 |     0 |    2 |
| mbed_ticker_api.o     |   904 |     0 |    0 |
| mbed_us_ticker_api.o  |    58 |     0 |   64 |
| mbed_wait_api_rtos.o  |    90 |     0 |    0 |
| nordic_critical.o     |   330 |     0 |    5 |
| nrf_drv_common.o      |    72 |     0 |    0 |
| nrf_drv_gpiote.o      |  1366 |     0 |  105 |
| reloc_vector_table.o  |    40 |     0 |  216 |
| rt_OsEventObserver.o  |     0 |     0 |    4 |
| rtx_delay.o           |    76 |     0 |    0 |
| rtx_evr.o             |   114 |     0 |    0 |
| rtx_kernel.o          |  1355 |   176 |    0 |
| rtx_lib.o             |   364 |    24 | 2341 |
| rtx_memory.o          |   248 |     0 |    0 |
| rtx_mempool.o         |   272 |     0 |    0 |
| rtx_msgqueue.o        |  1830 |     0 |    0 |
| rtx_mutex.o           |   898 |     0 |    0 |
| rtx_system.o          |   404 |     0 |    0 |
| rtx_thread.o          |  2568 |     0 |    0 |
| rtx_timer.o           |   256 |     0 |    0 |
| serial_api.o          |  1136 |     0 |  136 |
| sleep.o               |   114 |     0 |    0 |
| softdevice_handler.o  |    12 |     0 |   13 |
| startup_nrf52832.o    |   264 |     0 |    0 |
| system_nrf52.o        |   620 |     4 |    0 |
| us_ticker.o           |   552 |     0 |    5 |
| Subtotals             | 30744 |   224 | 8076 |
+-----------------------+-------+-------+------+
Total Static RAM memory (data + bss): 8300 bytes
Total Flash memory (text + data): 30968 bytes

theotherjimmy · 2017-11-07T19:47:26Z

/morph build

theotherjimmy · 2017-11-07T19:47:51Z

Those two commits should get us testing in Austin CI

theotherjimmy · 2017-11-07T19:48:04Z

now to get Travis working.

mbed-ci · 2017-11-07T19:56:24Z

Build : FAILURE

Build number : 449
Build artifacts/logs : http://mbed-os.s3-website-eu-west-1.amazonaws.com/?prefix=builds/5406/

theotherjimmy · 2017-11-07T19:58:44Z

I don't know what commit that was testing....

theotherjimmy · 2017-11-07T19:59:45Z

Ah, looking at my reflog, it tested the pre-amend version of Add ARMC6 support to REALTEK post-binary hook

theotherjimmy · 2017-11-07T19:59:51Z

/morph build

theotherjimmy · 2017-11-07T20:03:35Z

@pan- I added the Explicit instantiation of the vector thingy to the stl_features test. Do you foresee any issues with that?

mbed-ci · 2017-11-07T20:15:01Z

Build : FAILURE

Build number : 450
Build artifacts/logs : http://mbed-os.s3-website-eu-west-1.amazonaws.com/?prefix=builds/5406/

mbed-ci · 2017-11-07T20:29:55Z

Build : FAILURE

Build number : 451
Build artifacts/logs : http://mbed-os.s3-website-eu-west-1.amazonaws.com/?prefix=builds/5406/

theotherjimmy · 2017-11-07T20:44:59Z

I just made the explicit instantination exclusive to armc6

0xc0170 · 2017-11-09T08:27:43Z

/morph test

mbed-ci · 2017-11-09T10:42:03Z

Test : FAILURE

Build number : 293
Test logs :http://mbed-os-logs.s3-website-us-west-1.amazonaws.com/?prefix=logs/5406/293

I'm going to explain why these tests are superfluous because I don't like removing tests. The asserts were all asserting something that does not need to be true: that the toolchain's `name` attribute matched it's key in the `TOOLCHIAN_CLASSES` dictionary. With that out of the way, the `test_instantiation` test did not assert anything, and is therefore not a test.

theotherjimmy · 2017-11-09T15:39:23Z

/morph build

theotherjimmy · 2017-11-09T15:39:44Z

Rebased to get commits from #5405

Arm compiler 6 optimized way too well, and adding 2 more nops did the trick

theotherjimmy · 2017-11-09T15:57:33Z

/morph build

mbed-ci · 2017-11-09T16:07:48Z

Build : SUCCESS

Build number : 480
Build artifacts/logs : http://mbed-os.s3-website-eu-west-1.amazonaws.com/?prefix=builds/5406/

Triggering tests

/morph test
/morph uvisor-test

mbed-ci · 2017-11-09T16:27:10Z

Build : SUCCESS

Build number : 481
Build artifacts/logs : http://mbed-os.s3-website-eu-west-1.amazonaws.com/?prefix=builds/5406/

Triggering tests

/morph test
/morph uvisor-test

mbed-ci · 2017-11-09T19:10:08Z

Test : FAILURE

Build number : 300
Test logs :http://mbed-os-logs.s3-website-us-west-1.amazonaws.com/?prefix=logs/5406/300

mbed-ci · 2017-11-09T20:25:37Z

Test : FAILURE

Build number : 301
Test logs :http://mbed-os-logs.s3-website-us-west-1.amazonaws.com/?prefix=logs/5406/301

studavekar · 2017-11-09T20:27:05Z

@theotherjimmy there is one failure http://mbed-ci-master-2.austin.arm.com:8081/job/test_matrix_raas/301/target=NUCLEO_F746ZG,toolchain=ARM/ , i had to abort this combination.

theotherjimmy · 2017-11-09T20:28:45Z

Thanks. That's a persnickety one, and I'm having trouble reproducing it (My device is not operating well as a serial port).

theotherjimmy · 2017-11-09T21:38:26Z

Alright, I debugged it, and it looks like the FPU is off, leading to a usage fault. I turned the FPU on in a debugger, and was able to get the platform to pass an arbitrary test.

theotherjimmy · 2017-11-10T17:31:28Z

/morph build

mbed-ci · 2017-11-10T18:00:09Z

Build : SUCCESS

Build number : 495
Build artifacts/logs : http://mbed-os.s3-website-eu-west-1.amazonaws.com/?prefix=builds/5406/

Triggering tests

/morph test
/morph uvisor-test
/morph export-build

mbed-ci · 2017-11-10T19:22:54Z

Exporter Build : FAILURE

Build number : 116
Build artifacts/logs : http://mbed-os.s3-website-eu-west-1.amazonaws.com/?prefix=builds/exporter/5406/

mbed-ci · 2017-11-10T19:37:10Z

Test : SUCCESS

Build number : 313
Test logs :http://mbed-os-logs.s3-website-us-west-1.amazonaws.com/?prefix=logs/5406/313

0xc0170 added the needs: work label Oct 31, 2017

theotherjimmy force-pushed the disable-armc5 branch from 4aba329 to 3bf7863 Compare October 31, 2017 19:28

theotherjimmy force-pushed the disable-armc5 branch from 1d8715f to 3bf7863 Compare November 1, 2017 16:56

theotherjimmy force-pushed the disable-armc5 branch from 1997b70 to 0154f7d Compare November 7, 2017 19:42

theotherjimmy closed this Nov 7, 2017

theotherjimmy reopened this Nov 7, 2017

theotherjimmy added needs: work and removed needs: work labels Nov 7, 2017

theotherjimmy force-pushed the disable-armc5 branch from 656f596 to 6448d67 Compare November 7, 2017 20:44

theotherjimmy added 4 commits November 9, 2017 09:38

Disable Arm Compiler 5 and Use Arm Compiler 6 by default

1cfaed1

Add ARMC6 support to REALTEK post-binary hook

3f9330c

Explicitly instantiate vector base class

9c68057

theotherjimmy force-pushed the disable-armc5 branch from 4896e26 to 2c653ad Compare November 9, 2017 15:38

Workaround for NCS36510 deepsleep bug

727aada

Arm compiler 6 optimized way too well, and adding 2 more nops did the trick

theotherjimmy mentioned this pull request Nov 9, 2017

Tracking issue for Arm Compiler 6 library updates #5383

Closed

10 tasks

theotherjimmy added 2 commits November 10, 2017 11:30

Correct fpu passing for Armlink 6

146d0d9

Correct Arm Compiler 6 FPU present check

5839c1b

0xc0170 added needs: work and removed needs: CI labels Nov 16, 2017

theotherjimmy merged commit 5839c1b into ARMmbed:feature-armc6 Nov 16, 2017

theotherjimmy removed the needs: work label Nov 16, 2017

theotherjimmy deleted the disable-armc5 branch November 16, 2017 20:19

Disable Arm Compiler 5 and Use Arm Compiler 6 by default #5406

Disable Arm Compiler 5 and Use Arm Compiler 6 by default #5406

Uh oh!

Conversation

theotherjimmy commented Oct 30, 2017

Uh oh!

theotherjimmy commented Oct 31, 2017

Uh oh!

mbed-ci commented Oct 31, 2017

Uh oh!

studavekar commented Oct 31, 2017

Uh oh!

mbed-ci commented Oct 31, 2017

Uh oh!

theotherjimmy commented Oct 31, 2017

Uh oh!

theotherjimmy commented Oct 31, 2017

Uh oh!

theotherjimmy commented Nov 1, 2017

Uh oh!

theotherjimmy commented Nov 1, 2017

Uh oh!

mbed-ci commented Nov 1, 2017

Uh oh!

pan- commented Nov 2, 2017

Uh oh!

theotherjimmy commented Nov 7, 2017

Uh oh!

theotherjimmy commented Nov 7, 2017

Uh oh!

theotherjimmy commented Nov 7, 2017

Uh oh!

mbed-ci commented Nov 7, 2017

Uh oh!

theotherjimmy commented Nov 7, 2017

Uh oh!

theotherjimmy commented Nov 7, 2017

Uh oh!

theotherjimmy commented Nov 7, 2017

Uh oh!

theotherjimmy commented Nov 7, 2017

Uh oh!

mbed-ci commented Nov 7, 2017

Uh oh!

mbed-ci commented Nov 7, 2017

Uh oh!

theotherjimmy commented Nov 7, 2017

Uh oh!

0xc0170 commented Nov 9, 2017

Uh oh!

mbed-ci commented Nov 9, 2017

Uh oh!

theotherjimmy commented Nov 9, 2017

Uh oh!

theotherjimmy commented Nov 9, 2017

Uh oh!

theotherjimmy commented Nov 9, 2017

Uh oh!

mbed-ci commented Nov 9, 2017

Triggering tests

Uh oh!

mbed-ci commented Nov 9, 2017

Triggering tests

Uh oh!

mbed-ci commented Nov 9, 2017

Uh oh!

mbed-ci commented Nov 9, 2017

Uh oh!

studavekar commented Nov 9, 2017

Uh oh!

theotherjimmy commented Nov 9, 2017

Uh oh!

theotherjimmy commented Nov 9, 2017

Uh oh!

theotherjimmy commented Nov 10, 2017

Uh oh!

mbed-ci commented Nov 10, 2017

Triggering tests

Uh oh!

mbed-ci commented Nov 10, 2017