Release notes for Arm Compiler 6.14

Table of Contents

1. 1. Introduction
   1. 1.1. Arm Compiler 6 Configuration
   2. 1.2. What's Supported in Arm Compiler 6.14?
2. 2. Installation Instructions
   1. 2.1. Integration into Arm Development Studio 2018.0 or later
   2. 2.2. Integration into Keil MDK 5.22 or later
   3. 2.3. Use as a standalone toolchain installation
   4. 2.4. Installation on Linux
   5. 2.5. Installation on Windows
3. 3. Uninstall
4. 4. Documentation
5. 5. Feedback and Support
6. 6. Release History and Changes

1. Introduction

This is the initial set of release notes provided at the time of the release. For the latest copy of the release notes, see the latest version on https://developer.arm.com.

Arm Compiler 6.14 adds:

• Support for Cortex-A34 and Cortex-M55.
• Armv8.6-A:
  • Support for assembly.
  • Support for intrinsics for the BFloat16 Extension.
  • Support for intrinsics for the Matrix Multiply Extension.
• Armv8-M:
  • Beta support for assembly for the Custom Datapath Extension.
• Armv8.1-M:
  • Support for automatic vectorization for the optional M-profile Vector Extension (MVE).

Arm Compiler 6.14 is intended for use:

• In conjunction with an Arm Development Studio toolkit.
• In conjunction with a Keil MDK toolkit.
• As a standalone toolchain installation, provided you have a license for a suitable toolkit.

Contact your sales representative or visit https://developer.arm.com/tools-and-software/embedded/arm-compiler/buy to enquire about a license. Visit https://developer.arm.com/support/licensing to manage your licenses or access troubleshooting resources.

If you are using a floating license, your license server must be running armlmd and lmgrd version 11.14.1.0 or later. Arm recommends that you always use the latest version of the license server software that is available from https://developer.arm.com/products/software-development-tools/license-management/downloads.

1.1 Arm Compiler 6 Configuration

The Arm Compiler 6.14 toolchain includes the following:

• armclang: Compiler and integrated assembler based on LLVM and Clang technology.
• armasm: Legacy assembler for armasm-syntax assembly code. Use the armclang integrated assembler for all new assembly files.
• armar: Archiver which enables sets of ELF object files to be collected together.
• armlink: Linker that combines objects and libraries to produce an executable.
• fromelf: Image conversion utility and dissembler.
• Arm C++ libraries: Libraries based on the LLVM libc++ project.
• Arm C libraries: Runtime support libraries for embedded systems.

1.2 What's Supported in Arm Compiler 6.14?

The following table lists the target Arm Architectures and Processors supported in Arm Compiler 6.14. The exact set of targets depends on which license and toolkit you are using. The table is correct at the time of the Arm Compiler 6.14 release but may change in the future. Check with your supplier for more information.

Target		License
		Keil MDK (Windows only)				Arm Development Studio (Linux and Windows)
		Lite	Essential	Plus	Professional	Bronze	Silver	Gold	Platinum
Future Arch	-	-	-	-	-	-	-	-	Yes
Armv8.6-A	-	-	-	-	-	-	-	-	Yes
Armv8-A up to 8.5-A	Neoverse N1/E1	-	-	-	-	-	-	-	Yes
	Cortex-A77/A76AE/76/65AE/65	-	-	-	-	-	-	-	Yes
	Cortex-A75/73/72/57/55/53/35/34/32	-	-	-	-	-	-	Yes	Yes
Armv7-A	Cortex-A17/15/12/9/8/7/5	-	-	-	Yes	-	Yes	Yes	Yes
Armv8-R	Cortex-R52	-	-	-	-	-	-	Yes	Yes
Armv7-R	Cortex-R8/7/5	-	-	-	-	-	Yes	Yes	Yes
	Cortex-R4F/4	-	-	Yes	Yes	-	Yes	Yes	Yes
Armv8.1-M	Cortex-M55	-	Non-secure only	Yes	Yes	Yes	Yes	Yes	Yes
Armv8-M	Star	-	-	-	-	-	-	-	Yes
	Cortex-M35P/33/23	-	Non-secure only	Yes	Yes	Yes	Yes	Yes	Yes
Armv7-M	SC300	-	-	Yes	Yes	Yes	Yes	Yes	Yes
	Cortex-M7/4/3	32 Kbyte code limit	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Armv6-M	SC000	-	-	Yes	Yes	Yes	Yes	Yes	Yes
	Cortex-M1/0/0+	32 Kbyte code limit	Yes	Yes	Yes	Yes	Yes	Yes	Yes

For more information about the level of support for these targets and the toolchain features, refer to the product documentation.

Arm Development Studio Platinum Edition is reserved for Arm partners developing the latest IP before it becomes available in devices. It includes all the features of Gold, and additionally has support for the latest announced IP from Arm. Please contact Arm for more information.

2. Installation Instructions

If you received Arm Compiler 6.14 as part of a toolkit, for example Arm Development Studio, the toolkit installer takes care of the installation process. Please refer to the installation instructions for the toolkit in such cases.

For all other cases, you must select an appropriate installation location depending on how you intend to use Arm Compiler 6.14:

• Integrated into Arm Development Studio 2018.0 or later.
• Integrated into Keil MDK 5.22 or later.
• As a standalone toolchain installation.

Unless you are using Keil MDK, Arm Compiler 6.14 can be installed in any location, including the default location, providing this is outside of an Arm Development Studio product installation directory.

Please refer to the important configuration instructions at https://developer.arm.com/tools-and-software/software-development-tools/license-management/resources/product-and-toolkit-configuration and also the relevant sections below.

2.1. Integration into Arm Development Studio 2018.0 or later

You can integrate the toolchain with Arm Development Studio by following the instructions available at https://developer.arm.com/docs/101470/latest/configure-arm-development-studio/register-a-compiler-toolchain.

2.2. Integration into Keil MDK 5.22 or later

Arm Compiler 6.14 must be installed underneath the ARM subdirectory of the Keil MDK installation directory. For example, if your Keil MDK installation directory is in C:\Keil_v5, the recommended installation path is C:\Keil_v5\ARM\ARMCompiler6.14.

You can integrate the toolchain into a Keil MDK project by following the instructions in the tutorial available at http://www.keil.com/support/man/docs/uv4/uv4_armcompilers.htm.

Only the 32-bit variant of Arm Compiler 6.14 for Windows can be used with a Keil Single-User License or Keil Floating-User License.

2.3. Use as a standalone toolchain installation

Ensure that the ARMLMD_LICENSE_FILE environment variable is pointing to your license file or license server.

2.4. Installation on Linux

Arm Compiler 6.14 has been tested on the following supported 64-bit platforms:

• Red Hat Enterprise Linux 6.
• Red Hat Enterprise Linux 7.
• Red Hat Enterprise Linux 8.
• Ubuntu Desktop Edition 16.04 LTS.
• Ubuntu Desktop Edition 18.04 LTS.

Arm Compiler 6.14 is not expected to work on older platforms.

To install Arm Compiler 6.14, run (not source) install_x86_64.sh and follow the on-screen instructions. The installer unpacks Arm Compiler 6.14 into your chosen directory.

The armclang binary is dynamically linked to a copy of libstdc++ installed under your chosen directory as part of Arm Compiler 6.14.

2.5. Installation on Windows

Arm Compiler 6.14 has been tested on the following supported 64-bit platforms:

• Windows Server 2012.
• Windows Server 2016.
• Windows 8.1.
• Windows 10.

Arm Compiler 6.14 is not expected to work on older platforms.

Arm Compiler 6.14 has been tested on a feature update of Windows 10 available as of February 2020. It is expected to work well on later updates.

The 32-bit variant of Arm Compiler 6.14 for Windows is supported if and only if you are using Keil MDK on a supported 64-bit platform. 32-bit Windows platforms are not supported.

To install Arm Compiler 6.14, run win-x86_64\setup.exe or win-x86_32\setup.exe and follow the on-screen instructions. If you have an earlier version of Arm Compiler 6 installed and you wish to perform an upgrade, it is recommended that you uninstall the previous version before installing the new version of Arm Compiler 6.

Arm Compiler 6.14 is built with Microsoft Visual Studio 2017 and requires the Universal C Runtime in Windows to be installed. For further information see https://support.microsoft.com/en-gb/help/2999226/update-for-universal-c-runtime-in-windows.

3. Uninstall

On Linux, delete the Arm Compiler 6.14 installation directory.

On Windows, use Programs and Features in Control Panel, select ARM Compiler 6.14, and click the Uninstall button.

4. Documentation

Arm Compiler 6.14 documentation is available on https://developer.arm.com and comprises:

• User Guide.
• Reference Guide.
• Arm C and C++ Libaries and Floating-Point Support User Guide.
• Migration and Compatibility Guide.
• Errors and Warnings Reference Guide.

In January 2018, Arm published information about the Variant 1: bounds check bypass (CVE-2017-5753) vulnerability. Refer to https://developer.arm.com/support/security-update/compiler-support-for-mitigations and use the Migration path recommendations with this Arm Compiler release to help mitigate against the vulnerability. Note that the API of the mitigation is subject to change.

5. Feedback and Support

Your feedback is important to us, and you are welcome to send us defect reports and suggestions for improvement on any aspect of the product. Contact your supplier or visit https://support.developer.arm.com and open a case with feedback or support issues. Where appropriate, please provide the --vsn output from the tool, the complete content of any error message that the tools produce, and include any source code, other files, and command-lines necessary to reproduce the issue.

6. Release History and Changes

The following are the releases to date of the Arm Compiler 6.14 series:

• 6.14 (released February 2020)

Below is a summary of the enhancements and defect fixes when compared to the previous release. The information may include technical inaccuracies or typographical errors, and may change in future editions of the Release notes. Each itemized change is accompanied by a unique SDCOMP-<n> identifier. If you need to contact Arm about a specific issue within these release notes, please quote the appropriate identifier.

Changes in Arm Compiler 6.14

Changes are listed since the previous feature release, Arm Compiler 6.13.

General changes in Arm Compiler 6.14

• [SDCOMP-55145]  Automatic vectorization for the optional M-profile Vector Extension (MVE) in Armv8.1-M is now fully supported.

• [SDCOMP-55144]  Intrinsics for the Matrix Multiply Extension in Armv8.6-A are now fully supported. This extension is optional in Armv8.2-A to Armv8.5-A. To target the Matrix Multiply Extension, select from the following armclang options where <ext> is 2 or later:

  • --target=aarch64-arm-none-eabi -march=armv8.<ext>-a+i8mm to enable matrix multiplication instructions for 8-bit integer operations for AArch64 state.
  • --target=aarch64-arm-none-eabi -march=armv8.<ext>-a+f32mm to enable matrix multiplication instructions for 32-bit single-precision floating-point operations for AArch64 state.
  • --target=aarch64-arm-none-eabi -march=armv8.<ext>-a+f64mm to enable matrix multiplication instructions for 64-bit double-precision floating-point operations for AArch64 state.
  • --target=arm-arm-none-eabi -march=armv8.<ext>-a+i8mm to enable matrix multiplication instructions for 8-bit integer operations for AArch32 state.

• [SDCOMP-55143]  Intrinsics for the BFloat16 Extension in Armv8.6-A are now fully supported. This extension is optional in Armv8.2-A to Armv8.5-A. To target BFloat16, select from the following armclang options where <ext> is 2 or later:

  • --target=aarch64-arm-none-eabi -march=armv8.<ext>-a+bf16 for AArch64 state.
  • --target=arm-arm-none-eabi -march=armv8.<ext>-a+bf16 for AArch32 state.

• [SDCOMP-55142]  Assembly for the Armv8.6-A architecture is now fully supported. To target Armv8.6-A, select from the following armclang options:

  • --target=aarch64-arm-none-eabi -march=armv8.6-a for AArch64 state.
  • --target=arm-arm-none-eabi -march=armv8.6-a for AArch32 state.

• [SDCOMP-54946]  Support has been added for the Cortex-M55 processor. To target a specific feature set configuration of Cortex-M55, select from the following options:

  Integer MVE	Half-precision and Single-precision Floating-point MVE	Scalar Floating-point	armclang	armlink and fromelf
  Included	Included	Included	--target=arm-arm-none-eabi -mcpu=cortex-m55	--cpu=Cortex-M55
  Included	Not included	Included	--target=arm-arm-none-eabi -mcpu=cortex-m55+nomve.fp	--cpu=Cortex-M55.no_mvefp
  Included	Not included	Not included	--target=arm-arm-none-eabi -mcpu=cortex-m55+nofp	--cpu=Cortex-M55.no_fp
  Not included	Not included	Included	--target=arm-arm-none-eabi -mcpu=cortex-m55+nomve	--cpu=Cortex-M55.no_mve
  Not included	Not included	Not included	--target=arm-arm-none-eabi -mcpu=cortex-m55+nofp+nomve	--cpu=Cortex-M55.no_mve.no_fp

• [SDCOMP-54871]  Support has been added for the Cortex-A34 processor. To target Cortex-A34, select from the following options:

  armclang:

  • --target=aarch64-arm-none-eabi -mcpu=cortex-a34

  armasm, armlink, and fromelf:

  • --cpu=8-A.64

• [SDCOMP-54826]  Beta support has been added for assembly for the optional Custom Datapath Extension (CDE) for Armv8-M and later with the Main Extension. To target CDE, select from the following options:

  armclang:

  • --target=arm-arm-none-eabi -march=armv8-m.main+cdecpN, where N is in the range 0-7, for an Armv8-M target with the Main Extension.
  • --target=arm-arm-none-eabi -march=armv8.1-m.main+cdecpN, where N is in the range 0-7, for an Armv8.1-M target with the Main Extension.

  fromelf:

  • --cpu=8-M.main --coprocN=value, where N is in the range 0-7, and value is cde or CDE, for an Armv8-M target with the Main Extension.
  • --cpu=8.1-M.main --coprocN=value, where N is in the range 0-7, and value is cde or CDE, for an Armv8.1-M target with the Main Extension.

Enhancements in Arm Compiler 6.14

Compiler and integrated assembler (armclang)

• [SDCOMP-54955]  Alpha support has been added for automatic loop tail predication when compiling for a target with the M-profile Vector Extension. To enable automatic loop tail predication, compile with -mllvm -disable-mve-tail-predication=false.

• [SDCOMP-49669]  Support has been added for the __FILE_NAME__ predefined macro, which contains the filename part of the value of the __FILE__ predefined macro.

Defect fixes in Arm Compiler 6.14

Compiler and integrated assembler (armclang)

• [SDCOMP-55088]  In certain circumstances, two identical invocations of the compiler could incorrectly result in the compiler generating different but valid sequences of instructions. This has been fixed.

• [SDCOMP-54534]  In certain circumstances, when compiling for AArch64 state, the compiler could incorrectly fail to generate a build attribute which specifies that the output object requires flush-to-zero mode enabled for floating-point operations. Subsequently, the linker could incorrectly select implementations of floating-point library functions that do not have flush-to-zero mode enabled. This has been fixed.

• [SDCOMP-54466]  The compiler incorrectly failed to report an error when the size of the translation unit is greater than the maximum acceptable size of 2GB, including when the size only exceeds this limit during preprocessing. This has been fixed. The compiler now reports fatal error: sorry, this include generates a translation unit too large for Clang to process.

• [SDCOMP-54427]  In rare circumstances, when compiling with -fsanitize=memtag, the compiler could generate incorrect memory tagging code. Subsequently, this could result in spurious memory tagging exceptions against addresses in the stack area. This has been fixed.

• [SDCOMP-54406]  In certain circumstances, when compiling for AArch32 state with -mfloat-abi=soft or for a target that does not support a hardware floating-point unit, and either at -Ofast, with -ffast-math, or with -ffp-mode=fast, the compiler could incorrectly report error: clang frontend command failed due to signal. This has been fixed.

• [SDCOMP-54351]  In certain circumstances, when compiling at any optimization level except -O0, the compiler could incorrectly assume that an offset used in a pointer arithmetic operation O is never negative, and subsequently generate incorrect code for calculations involving O. This has been fixed.

• [SDCOMP-54294]  In certain circumstances, when compiling for AArch64 state, the compiler could incorrectly generate different but valid sequences of branch instructions when compiling with -g or -gdwarf-version versus when compiling without -g or -gdwarf-version. This has been fixed.

• [SDCOMP-54255]  In certain circumstances, when compiling at any optimization level except -O0 for AArch32 state and a target that supports the half-precision floating-point (FP16) extension, and the code contains a variable of _Float16 type, the compiler could incorrectly report fatal error: error in backend: Cannot select: <value>.  This has been fixed.

• [SDCOMP-54190]  When assembling for AArch64 state and an Armv8-A target with the Statistical Profiling Extension (SPE), the inline assembler and integrated assembler incorrectly failed to report an error for an MSR instruction that specifies CurrentEL, ICH_MISR_EL2, PMBIDR_EL1, or PMSIDR_EL1 as the special register to be accessed. This has been fixed. The inline assembler and integrated assembler now report error: expected writable system register or pstate.

• [SDCOMP-54141]  In certain circumstances, when compiling a program that contains a function or variable that is annotated with __attribute__((section(".ARM.__at_<address>"))), the compiler could incorrectly mark the section containing that function or variable as mergeable. Subsequently, at link-time, the linker could report Error: L6975E: <file>(.ARM.__at_<address>) cannot have a required base and SHF_MERGE. This has been fixed.

• [SDCOMP-54010]  In certain circumstances, when compiling at any optimization level except -O0 and for T32 state, the compiler could incorrectly generate an UNPREDICTABLE ADD or SUB instruction that specifies the stack pointer (SP) as the destination operand but not as the source operand. This has been fixed.

• [SDCOMP-53743]  In certain circumstances, when compiling for AArch64 state with -mbranch-protection=bti or -mbranch-protection=standard, the compiler could incorrectly report fatal error: error in backend: fixup value out of range. This has been fixed.

• [SDCOMP-51231]  When compiling for AArch32 state without -g or -gdwarf-version, the compiler incorrectly failed to generate a .debug_frame section containing stack frame unwinding debug information. This has been fixed.

• [SDCOMP-50543]  In rare circumstances, when compiling with -mcmse and the program contains a function call via a function pointer that is annotated with __attribute__((cmse_nonsecure_call)), the compiler could incorrectly report error: clang frontend command failed due to signal. This has been fixed.

• [SDCOMP-46159]  When compiling with -mcmse, the compiler incorrectly failed to clear the upper 16 bits of a register used for a variable of __fp16 or _Float16 type that is passed as a parameter to a function annotated with __attribute__((cmse_nonsecure_call)) or returned from a function annotated with __attribute__((cmse_nonsecure_entry)). This has been fixed.

• [SDCOMP-30463]  When compiling with -mcmse, the compiler incorrectly failed to clear any unused bits of registers used for a variable of struct type that is passed as a parameter to a function annotated with __attribute__((cmse_nonsecure_call)) or returned from a function annotated with __attribute__((cmse_nonsecure_entry)). This has been fixed.

• [SDCOMP-30440]  In certain circumstances, when compiling with C++ exceptions enabled, with -g or -gdwarf-version, and for AArch64 state, the compiler could incorrectly fail to generate a .debug_frame section containing stack frame unwinding debug information. This has been fixed.

Legacy assembler (armasm)

• [SDCOMP-55207]  When assembling for AArch64 state and an Armv8.2-A or later target, the legacy assembler incorrectly fails to report an error for an MSR instruction that specifies CurrentEL, ICH_MISR_EL2, PMBIDR_EL1, or PMSIDR_EL1 as the special register to be accessed. This has been fixed. The legacy assembler now reports Error: A1616E: Instruction, offset, immediate or register combination is not supported by the current instruction set or Error: A1805E: Register is Read-Only.

Linker (armlink)

• [SDCOMP-46190]  In certain circumstances, when linking with --no_remove, the linker could incorrectly remove ELF sections that contain debug information. This has been fixed.

Libraries and system headers

• [SDCOMP-54710]  The microlib implementations of the snprintf() and vsnprintf() functions incorrectly always returned 0 when called with a buffer size of 0. This has been fixed.

• [SDCOMP-54589]  The arm_mve.h system header incorrectly failed to define the vreinterpretq_s64_*(), vreinterpretq_u64_*(), vreinterpretq_*_s64(), and vreinterpretq_*_u64() M-profile Vector Extension intrinsics. This has been fixed.

Fromelf

• [SDCOMP-54453]  When disassembling an ELF file that contains a BFCSEL instruction and a relocation of type R_ARM_THM_BF12, the fromelf utility incorrectly reported the type of the relocation as R_ARM_THM_BF13. This has been fixed.

Known issues in Arm Compiler 6.14

• [SDCOMP-54724]  In certain circumstances, when compiling with -ffixed-r6, the compiler can generate code that incorrectly fails to fulfil stack alignment requirements.

• [SDCOMP-54513]  In certain circumstances, nested calls to the M-profile Vector Extension intrinsics can result in slow compilation and high memory usage at compile time. To avoid this issue, do not use nested calls to the M-profile Vector Extension intrinsics.

• [SDCOMP-54391]  When compiling with -fstack-protector-strong at -O0, the compiler incorrectly fails to enable stack protection for functions containing certain kinds of address-taking of non-array local variables. To avoid this issue, compile with -fstack-protector-all instead of -fstack-protector-strong.

• [SDCOMP-50470]  The compiler incorrectly fails to report an error for a function call that involves invalid default argument promotion for the _Float16 data type.

Release notes for Arm Compiler 6.13

1. Introduction

Arm Compiler 6.13 adds:

• Support for Cortex-A65, Cortex-A77, and Star processors.
• Early support for Future Architecture Technologies in the A architecture profile:
  • Assembly for the Embedded Trace Extension (ETE).
  • Assembly for the Scalable Vector Extension 2 (SVE2).
  • Assembly for the Trace Buffer Extension (TRBE).
  • Assembly and intrinsics for the Transactional Memory Extension (TME).
• Armv8.6-A:
  • Alpha support for assembly.
  • Alpha support for intrinsics for the BFloat16 Extension.
  • Alpha support for intrinsics for the Matrix Multiply Extension.
• Armv8.5-A:
  • Support for intrinsics for the optional Memory Tagging Extension.
• Armv8.1-M:
  • Support for assembly and intrinsics.
  • Beta support for automatic vectorization for the optional M-profile Vector Extension (MVE).

Arm Compiler 6.13 is intended for use:

• In conjunction with an Arm Development Studio toolkit.
• In conjunction with a Keil MDK toolkit.
• As a standalone toolchain installation, provided you have a license for a suitable toolkit.

Contact your sales representative or visit https://developer.arm.com/buy-arm-products to enquire about a license. Visit https://developer.arm.com/support/licensing to manage your licenses or access troubleshooting resources.

If you are using a floating license, your license server must be running armlmd and lmgrd version 11.14.1.0 or later. Arm recommends that you always use the latest version of the license server software that is available from https://developer.arm.com/products/software-development-tools/license-management/downloads.

1.1 Arm Compiler 6 Configuration

The Arm Compiler 6.13 toolchain includes the following:

• armclang: Compiler and integrated assembler based on LLVM and Clang technology.
• armasm: Legacy assembler for armasm-syntax assembly code. Use the armclang integrated assembler for all new assembly files.
• armar: Archiver which enables sets of ELF object files to be collected together.
• armlink: Linker that combines objects and libraries to produce an executable.
• fromelf: Image conversion utility and dissembler.
• Arm C++ libraries: Libraries based on the LLVM libc++ project.
• Arm C libraries: Runtime support libraries for embedded systems.

1.2 What's Supported in Arm Compiler 6.13?

The following table lists the target Arm Architectures and Processors supported in Arm Compiler 6.13. The exact set of targets depends on which license and toolkit you are using. The table is correct at the time of the Arm Compiler 6.13 release but may change in the future. Check with your supplier for more information.

Target		License
		Keil MDK (Windows Only)				Arm Development Studio (Linux and Windows)
		Lite	Essential	Plus	Professional	Bronze	Silver	Gold	Platinum
Future Arch	-	-	-	-	-	-	-	-	Yes
Armv8.6-A	-	-	-	-	-	-	-	-	Yes
Armv8-A up to 8.5-A	Neoverse N1/E1	-	-	-	-	-	-	-	Yes
	Cortex-A77/76AE/76/65AE/65	-	-	-	-	-	-	-	Yes
	Cortex-A75/73/72/57/55/53/35/32	-	-	-	-	-	-	Yes	Yes
Armv7-A	Cortex-A17/15/12/9/8/7/5	-	-	-	Yes	-	Yes	Yes	Yes
Armv8-R	Cortex-R52	-	-	-	-	-	-	Yes	Yes
Armv7-R	Cortex-R8/7/5	-	-	-	-	-	Yes	Yes	Yes
	Cortex-R4F/4	-	-	Yes	Yes	-	Yes	Yes	Yes
Armv8.1-M	-	-	-	-	-	-	-	-	Yes
Armv8-M	Star	-	-	-	-	-	-	-	Yes
	Cortex-M35P/33/23	-	Non-secure only	Yes	Yes	Yes	Yes	Yes	Yes
Armv7-M	SC300	-	-	Yes	Yes	Yes	Yes	Yes	Yes
	Cortex-M7/4/3	32 Kbyte code limit	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Armv6-M	SC000	-	-	Yes	Yes	Yes	Yes	Yes	Yes
	Cortex-M1/0/0+	32 Kbyte code limit	Yes	Yes	Yes	Yes	Yes	Yes	Yes

For more information about the level of support for these targets and the toolchain features, refer to the product documentation.

Arm Development Studio Platinum Edition is reserved for Arm partners developing the latest IP before it becomes available in devices. It includes all the features of Gold, and additionally has support for the latest announced IP from Arm. Please contact Arm for more information.

2. Installation Instructions

If you received Arm Compiler 6.13 as part of a toolkit, for example Arm Development Studio, the toolkit installer takes care of the installation process. Please refer to the installation instructions for the toolkit in such cases.

For all other cases, you must select an appropriate installation location depending on how you intend to use Arm Compiler 6.13:

• Integrated into Arm Development Studio 2018.0 or later.
• Integrated into Keil MDK 5.22 or later.
• As a standalone toolchain installation.

Unless you are using Keil MDK, Arm Compiler 6.13 can be installed in any location, including the default location, providing this is outside of an Arm Development Studio product installation directory.

Please refer to the configuration instructions at https://developer.arm.com/tools-and-software/software-development-tools/license-management/resources/product-and-toolkit-configuration and also the relevant sections below.

2.1. Integration into Arm Development Studio 2018.0 or later

You can integrate the toolchain with Arm Development Studio 2018.0 or later by following the instructions available at https://developer.arm.com/docs/101470/latest/configure-arm-development-studio/register-a-compiler-toolchain.

2.2. Integration into Keil MDK 5.22 or later

Arm Compiler 6.13 must be installed underneath the ARM subdirectory of the Keil MDK installation directory. For example, if your Keil MDK installation directory is in C:\Keil_v5, the recommended installation path is C:\Keil_v5\ARM\ARMCompiler6.13.

After it is installed, you can integrate the toolchain into a Keil MDK project by following the instructions in the tutorial available at http://www.keil.com/support/man/docs/uv4/uv4_armcompilers.htm.

Only the 32-bit Windows variant of Arm Compiler 6.13 can be used with a Keil Single-User License or Keil Floating-User License.

2.3. Use as a standalone toolchain installation

Ensure that the ARMLMD_LICENSE_FILE environment variable is pointing to your license file or license server.

2.4. Installation on Linux

Arm Compiler 6.13 has been tested on the following supported platforms:

• Red Hat Enterprise Linux 6 Workstation, 64-bit only.
• Red Hat Enterprise Linux 7 Workstation, 64-bit only.
• Ubuntu Desktop Edition 16.04 LTS, 64-bit only.
• Ubuntu Desktop Edition 18.04 LTS, 64-bit only.

Arm Compiler 6.13 is not expected to work on older platforms.

To install Arm Compiler 6.13, run (not source) install_x86_64.sh and follow the on-screen instructions. The installer unpacks Arm Compiler 6.13 into your chosen directory.

The armclang binary is dynamically linked to a copy of libstdc++ installed under your chosen directory as part of Arm Compiler 6.13.

2.5. Installation on Windows

Arm Compiler 6.13 has been tested on the following supported platforms:

• Windows Server 2012, 64-bit only.
• Windows 7 Enterprise SP1.
• Windows 7 Professional SP1.
• Windows 8.1 Enterprise, 64-bit only.
• Windows 8.1 Professional, 64-bit only.
• Windows 10 Enterprise, 64-bit only.
• Windows 10 Professional, 64-bit only.

Arm Compiler 6.13 is not expected to work on older platforms.

Arm Compiler 6.13 has been tested on the latest feature update of Windows 10 available as of October 2019. It is expected to work well on later updates.

To install Arm Compiler 6.13, run win-x86_64\setup.exe on a supported 64-bit Windows platform or win-x86_32\setup.exe on a supported 32-bit Windows platform and follow the on-screen instructions. If you have an earlier version of Arm Compiler 6 installed and you wish to perform an upgrade, it is recommended that you uninstall the previous version before installing the new version of Arm Compiler 6.

Arm Compiler 6.13 is built with Microsoft Visual Studio 2017 and requires the Universal C Runtime in Windows to be installed. For further information see https://support.microsoft.com/en-gb/help/2999226/update-for-universal-c-runtime-in-windows.

3. Uninstall

On Linux, delete the Arm Compiler 6.13 installation directory.

On Windows, use Programs and Features in Control Panel, select ARM Compiler 6.13, and click the Uninstall button.

4. Documentation

Arm Compiler 6.13 documentation is available on https://developer.arm.com and comprises:

• User Guide.
• Reference Guide.
• Arm C and C++ Libraries and Floating-Point Support User Guide.
• Migration and Compatibility Guide.
• Errors and Warnings Reference Guide.

From this release, the Reference Guide has been extended to incorporate reference information for all the Arm Compiler tools. This includes reference information that was provided in separate User Guide documents in previous releases.

In January 2018, Arm published information about the Variant 1: bounds check bypass (CVE-2017-5753) vulnerability. Refer to https://developer.arm.com/support/security-update/compiler-support-for-mitigations and use the Migration path recommendations with this Arm Compiler release to help mitigate against the vulnerability. Note that the API of the mitigation is subject to change.

5. Feedback and Support

Your feedback is important to us, and you are welcome to send us defect reports and suggestions for improvement on any aspect of the product. Contact your supplier or visit https://developer.arm.com/support and open a case with feedback or support issues. Where appropriate, please provide the --vsn output from the tool, the complete content of any error message that the tools produce, and include any source code, other files, and command-lines necessary to reproduce the issue.

6. Release History and Changes

The following are the releases to date of the Arm Compiler 6.13 series:

• 6.13 (released October 2019)

Below is a summary of the enhancements and defect fixes when compared to the previous release. The information may include technical inaccuracies or typographical errors, and may change in future editions of the Release notes. Each itemized change is accompanied by a unique SDCOMP-<n> identifier. If you need to contact Arm about a specific issue within these release notes, please quote the appropriate identifier.

Changes in Arm Compiler 6.13

Changes are listed since the previous feature release, Arm Compiler 6.12.

General changes in Arm Compiler 6.13

• [SDCOMP-54460]  Beta support has been added for automatic vectorization for the optional M-profile Vector Extension (MVE) in Armv8.1-M.

• [SDCOMP-54459]  Alpha support has been added for intrinsics for the Matrix Multiply Extension in Armv8.6-A. To target the Matrix Multiply Extension, select from the following armclang options where <ext> is 2 or later:

  • --target=aarch64-arm-none-eabi -march=armv8.<ext>-a+i8mm to enable matrix multiplication instructions for 8-bit integer operations for AArch64 state.
  • --target=aarch64-arm-none-eabi -march=armv8.<ext>-a+f32mm to enable matrix multiplication instructions for 32-bit single-precision floating-point operations for AArch64 state.
  • --target=aarch64-arm-none-eabi -march=armv8.<ext>-a+f64mm to enable matrix multiplication instructions for 64-bit double-precision floating-point operations for AArch64 state.
  • --target=arm-arm-none-eabi -march=armv8.<ext>-a+i8mm to enable matrix multiplication instructions for 8-bit integer operations for AArch32 state.

• [SDCOMP-54458]  Early support has been added for assembly for the Trace Buffer Extension (TRBE) for Future Architecture Technologies in the A architecture profile. To target TRBE, select the following armclang options:

  • --target=aarch64-arm-none-eabi -march=armv8-a

• [SDCOMP-54457]  Early support has been added for assembly for the Embedded Trace Extension (ETE) for Future Architecture Technologies in the A architecture profile. To target ETE, select the following armclang options:

  • --target=aarch64-arm-none-eabi -march=armv8-a

• [SDCOMP-54456]  Early support has been added for assembly and intrinsics for the Transactional Memory Extension (TME) for Future Architecture Technologies in the A architecture profile. To target TME, select the following armclang options:

  • --target=aarch64-arm-none-eabi -march=armv8-a+tme

• [SDCOMP-54455]  Alpha support has been added for assembly for the Armv8.6-A architecture. To target Armv8.6-A, select from the following armclang options:

  • --target=aarch64-arm-none-eabi -march=armv8.6-a for AArch64 state.
  • --target=arm-arm-none-eabi -march=armv8.6-a for AArch32 state.

• [SDCOMP-54454]  Alpha support has been added for intrinsics for the BFloat16 Extension in Armv8.6-A. To target BFloat16, select from the following armclang options where <ext> is 2 or later:

  • --target=aarch64-arm-none-eabi -march=armv8.<ext>-a+bf16 for AArch64 state.
  • --target=arm-arm-none-eabi -march=armv8.<ext>-a+bf16 for AArch32 state.

• [SDCOMP-53664]  Support has been added for the Star processor. To target Star, select from the following armclang options:

  • --target=arm-arm-none-eabi -mcpu=star for a variant with DSP and FP.
  • --target=arm-arm-none-eabi -mcpu=star+nodsp for a variant without DSP but with FP.
  • --target=arm-arm-none-eabi -mcpu=star -mfloat-abi=soft for a variant with DSP but without FP.
  • --target=arm-arm-none-eabi -mcpu=star+nodsp -mfloat-abi=soft for a variant without DSP and FP.

• [SDCOMP-53663]  Support has been added for the Cortex-A65 processor. To target Cortex-A65, select the following armclang options:

  • --target=aarch64-arm-none-eabi -mcpu=cortex-a65

• [SDCOMP-53662]  Support has been added for the Cortex-A77 processor. To target Cortex-A77, select from the following armclang options:

  • --target=aarch64-arm-none-eabi -mcpu=cortex-a77 for AArch64 state.
  • --target=arm-arm-none-eabi -mcpu=cortex-a77 for AArch32 state.

• [SDCOMP-53600]  Early support has been added for assembly for the Scalar Vector Extension 2 (SVE2) for Future Architecture Technologies in the A architecture profile. To target SVE2, select from the following armclang options:

  • --target=aarch64-arm-none-eabi -march=armv8-a+sve2
  • --target=aarch64-arm-none-eabi -march=armv8-a+sve2-aes
  • --target=aarch64-arm-none-eabi -march=armv8-a+sve2-bitperm
  • --target=aarch64-arm-none-eabi -march=armv8-a+sve2-sha3
  • --target=aarch64-arm-none-eabi -march=armv8-a+sve2-sm4

• [SDCOMP-53277]  Support has been added for the SysV dynamic linking model. Use the following options to enable this model:

  • armclang:
    • -fsysv to enable the generation of code suitable for the SysV linking model.
  • armlink:
    • --sysv to create a System V (SysV) formatted ELF executable file.

  For more information about these options, refer to the -fsysv, -fno-sysv and --sysv sections of the Reference Guide.

• [SDCOMP-53102]  Support has been added for Arm C Language Extensions (ACLE) intrinsics for the optional Memory Tagging Extension in Armv8.5-A.

• [SDCOMP-52927]  Support has been added for the memory tagging stack protection (stack tagging) feature for the optional Memory Tagging Extension in Armv8.5-A. The memory tagging stack protection feature can be used with the following options:

  • armclang:
    • -fsanitize=memtag to enable the generation of memory tagging code for protecting the memory allocations on the stack.
  • armlink:
    • --library_security=protection to override the automatic selection of protected libraries for branch protection and memory tagging stack protection.

  For more information about these options, refer to the -fsanitize and --library_security=protection sections of the Reference Guide.

• [SDCOMP-51237]  In the C++ Thread Porting API, the name of the timespec structure has been changed to __ARM_TPL_timespec_t to avoid a namespace violation. Additionally, the type of the tv_nsec member of this structure has been changed from unsigned long to long.

  For more information about the C++ Thread Porting API, refer to the Multithreaded support in Arm C++ libraries [ALPHA] section of the Arm C and C++ Libraries and Floating-Point Support User Guide.

• [SDCOMP-49562]  Support has been added for assembly and intrinsics for the Armv8.1-M architecture, including the optional M-profile Vector Extension (MVE). To target Armv8.1-M, select from the following armclang options:

  • --target=arm-arm-none-eabi -march=armv8.1-m.main for a variant without MVE.
  • --target=arm-arm-none-eabi -march=armv8.1-m.main+mve to enable MVE instructions for integer operations.
  • --target=arm-arm-none-eabi -march=armv8.1-m.main+mve.fp to enable MVE instructions for integer and single-precision floating-point operations.
  • --target=arm-arm-none-eabi -march=armv8.1-m.main+mve.fp+fp.dp to enable MVE instructions for integer, single-precision, and double-precision floating-point operations.

Enhancements in Arm Compiler 6.13

Compiler and integrated assembler (armclang)

• [SDCOMP-51844]  Support has been added for the small and tiny code model types when compiling for AArch64 state. Use one of the following options to compile for these memory models:

  • -mcmodel=small to generate code for the small code model. The program and its statically defined symbols must be within 4GB of each other. This is the default code model.
  • -mcmodel=tiny to generate code for the tiny code model. The program and its statically defined symbols must be within 1MB of each other.

  Alpha support has been added for the large code model when compiling for AArch64 state. Use the following option to compile for the large memory model:

  • -mcmodel=large to generate code for the large code model. The compiler makes no assumptions about addresses and sizes of sections.

  For more information, refer to the -mcmodel section of the Reference Guide.

Defect fixes in Arm Compiler 6.13

Compiler and integrated assembler (armclang)

• [SDCOMP-54132]  In certain circumstances, when compiling with -fstack-protector, -fstack-protector-all, or -fstack-protector-strong, the compiler could generate code that incorrectly failed to properly protect against buffer overflow. This has been fixed.

  In July 2019, cert.org published information about this vulnerability as LLVMs Arm stack protection feature can be rendered ineffective (VU#129209). Arm Compiler 6.12 is affected by this vulnerability. Versions before Arm Compiler 6.12 are not affected because the stack protection feature was not supported until Arm Compiler 6.12.

• [SDCOMP-53862]  In certain circumstances, when compiling with C++ exceptions enabled for an Armv6-M target or an Armv8-M target without the Main Extension, the compiler could generate incorrect exception unwinding information for an inline assembly statement that contains one of the following registers in its clobber list:

  • R8.
  • R9.
  • R10.
  • R11.
  • R12.
  • R13 (SP).
  • R14 (LR).

  Subsequently, when an exception is thrown, the variables of the function could be corrupted. This has been fixed.

• [SDCOMP-53783]  In certain circumstances, when compiling at any optimization level except -O0 for AArch32 state, the compiler could generate incorrect code for a variadic function arguments list that contains a struct that has a member of double type and is annotated with __attribute__((packed)). This has been fixed.

• [SDCOMP-53771]  In certain circumstances, when compiling for AArch64 state, the compiler could generate incorrect code for a call to a vsqadd*() or vuqadd*() Neon intrinsic that passes a value of floating-point type as the second argument. This has been fixed.

• [SDCOMP-53728]  In certain circumstances, when compiling with -MD or -MMD, and not with -MF, and the output filename does not contain a dot, the dependency information could be written to a file with an incorrect name. This has been fixed.

• [SDCOMP-53569]  In certain circumstances, when compiling at any optimization level except -O0 for AArch32 state, the compiler could generate incorrect code for an expression that casts an integer type with a size smaller than 32 bits to a floating-point type. This has been fixed.

• [SDCOMP-53506]  In certain circumstances, when compiling a program that uses a variadic function arguments list that contains an overaligned or underaligned parameter P, the compiler could generate code that incorrectly misaligned P. Such code did not confirm to the Parameter Passing section of the Procedure Call for the Arm Architecture ABI release 2.09 for AArch32 state, or the Parameter Passing section of the Procedure Call Standard for the Arm 64-bit Architecture ABI release 1.0 for AArch64 state. This has been fixed.

• [SDCOMP-53434]  In certain circumstances, when compiling at any optimization level except -O0, the compiler could generate incorrect code for a function F that contains conditional accesses to the stack. Such incorrect code prematurely released stack memory that must remain reserved throughout the execution of F. This has been fixed.

• [SDCOMP-53407]  In certain circumstances, when assembling for AArch32 state, the inline assembler and integrated assembler could incorrectly report error: invalid instruction or error: invalid operand for instruction for a CDP, CDP2, LDC, LDC2, MCR, MCR2, MCRR, MCRR2, MRC, MRC2, MRRC, MRRC2, STP, or STP2 instruction that specifies a coprocessor name (P<n>) or coprocessor register (C<n>) using an uppercase letter. This has been fixed.

• [SDCOMP-53279]  In rare circumstances, when compiling at any optimization level except -O0 for T32 state, the compiler could incorrectly generate an invalid IT block. This has been fixed.

• [SDCOMP-53271]  The inline assembler and integrated assembler incorrectly failed to report an error for a VFMAL or VFMSL instruction with a second source register operand that is not in the range D0-D7 or S0-S15. This has been fixed. The inline assembler and integrated assembler now report error: invalid instruction and either note: operand must be a register in range [d0, d7] or note: operand must be a register in range [s0, s15].

• [SDCOMP-53213]  In certain circumstances, when assembling for AArch64 state, the inline assembler and integrated assembler could incorrectly report:

  • error: expected readable system register for an MRS instruction that specifies TTBR0_EL2 as the special register to be accessed.
  • error: expected writable system register or pstate for an MSR instruction that specifies TTBR0_EL2 as the special register to be accessed.

  This has been fixed.

• [SDCOMP-53007]  In certain circumstances, when compiling at -Os or -Oz for AArch32 state, the compiler could incorrectly report fatal error: error in backend: ran out of registers during register allocation. This has been fixed.

• [SDCOMP-52959]  In certain circumstances, when compiling at -Oz and with a -mbranch-protection option that enables branch protection using Branch Target Identification, the compiler could generate code containing a BTI instruction in an incorrect location. This has been fixed.

• [SDCOMP-52933]  In certain circumstances, when compiling at any optimization level except -O0 for AArch32 state, the compiler could incorrectly fail to truncate a 16-bit integer to an 8-bit integer when required to do so. This has been fixed.

• [SDCOMP-52918]  In certain circumstances, when compiling for AArch32 state and a target that supports a hardware floating-point unit but not the half-precision floating-point (FP16) extension, the compiler could generate incorrect code for a call to function which has a parameter of struct type that has a member of half-precision floating-point vector type. This has been fixed.

• [SDCOMP-52910]  In certain circumstances, when compiling at any optimization level except -O0 for AArch32 state and a target that supports the half-precision floating-point (FP16) extension, the compiler could incorrectly report fatal error: error in backend: Cannot select: <value>. This has been fixed.

• [SDCOMP-52828]  In certain circumstances, when compiling for T32 state and an Armv8-A target that supports a hardware floating-point unit, the compiler could generate an IT block that incorrectly contains a floating-point instruction. This has been fixed.

• [SDCOMP-52632]  In certain circumstances, when compiling with -mfloat-abi=value, the code contains a function F with a homogenous aggregate parameter, and F is annotated with __attribute__((pcs("calling_convention"))), the compiler could generate code that incorrectly does not conform to the Procedure Call Standard for the Arm Architecture ABI release 2.09. This has been fixed.

• [SDCOMP-51612]  In certain circumstances, when compiling at any optimization level except -O0 for AArch32 state and a target that supports a hardware floating-point unit, the compiler could generate incorrect code for throw statements that specify an exception object of a vector type that has an alignment requirement of 8 bytes. This has been fixed.

• [SDCOMP-51181]  In certain circumstances, the compiler could generate incorrect code for a function which throws a C++ exception E when it can be proven that code following E is unreachable. This has been fixed.

• [SDCOMP-51026]  In certain circumstances, when compiling at any optimization level except -O0 for AArch32 state, the compiler could generate incorrect code for a function that has a parameter of struct type that is more than 80 bytes in size, and performs multiple operations involving a variable that is 8 bytes in size. This has been fixed.

• [SDCOMP-50736]  In certain circumstances, when compiling at -O0 for AArch32 state and a target that supports the half-precision floating-point (FP16) extension, the compiler could generate incorrect code for a struct that has a member of half-precision floating-point type. This has been fixed.

• [SDCOMP-50723]  In certain circumstances, when compiling for AArch32 state and a target that supports the half-precision floating-point (FP16) extension, the compiler could generate a VLDR{<c>}{<q>}.16 or VSTR{<c>}{<q>}.16 instruction with an incorrect immediate offset. This has been fixed.

• [SDCOMP-50722]  In certain circumstances, when compiling code at any optimization level except -O0 for AArch32 state and a big-endian target that supports the half-precision floating-point (FP16) extension, the compiler could incorrectly report error: clang frontend command failed due to signal. This has been fixed.

• [SDCOMP-50507]  In certain circumstances, when compiling for a target that supports the half-precision floating-point (FP16) extension, the compiler could incorrectly report fatal error: error in backend or error: cannot compile this builtin function yet for a program that contains a call to a floating-point absolute compare intrinsic or a floating-point integer conversion intrinsic. This has been fixed.

  Additionally, support has been added for the floating-point multiply extended (by element) intrinsics, vmulxh_lane_f16() and vmulxh_laneq_f16().

• [SDCOMP-50396]  In certain circumstances, when compiling for an Armv6-M target or an Armv8-M target without the Main Extension, the compiler could generate code that incorrectly corrupts a register by overwriting it with the value from R12. This has been fixed.

Linker (armlink)

• [SDCOMP-53507]  In certain circumstances, when linking without --no_merge_litpools for T32 state, the linker could generate an output file containing symbol information that incorrectly failed to account for a function that changes in size as a result of the merging of identical constants. This has been fixed.

• [SDCOMP-53500]  In certain circumstances, when linking with --paged, the linker could calculate the number of load segments incorrectly and subsequently generate a corrupted image. This has been fixed.

• [SDCOMP-52081]  In certain circumstances, when linking a program with RW data compression enabled for AArch32 state, and the program contains a compressible execution region which references a symbol in a compressible execution region, the linker could generate an image that incorrectly does not contain a required function. This has been fixed.

• [SDCOMP-51704]  In certain circumstances, when linking with a scatter file that contains a .ANY selector, the linker could incorrectly report Internal fault: [0xb3b91b:<ver>] or a Windows APPCRASH error. This has been fixed.

• [SDCOMP-51507]  The linker incorrectly failed to report an error when the input objects contain conflicting stack or heap setup methods. Subsequently, this could result in unexpected run time behavior. This has been fixed. The linker now reports Error: 6915E: Library reports error: Conflicting stack/heap set up methods in use. __user_setup_stackheap, __user_initial_stackheap cannot be used with __initial_sp, __heap_base and __heap_limit.

• [SDCOMP-51114]  In certain circumstances, the linker could incorrectly report Warning: L6412W: Disabling merging for <objname>(<secname>), unsupported relocation <type> from <srcobj>(<srcsec>). This has been fixed.

• [SDCOMP-30573]  In certain circumstances, when linking with --partial, and an input object that has been compiled in a C++ source language mode contains a COMDAT group and either debug information or a .eh_frame section, the linker could incorrectly report Internal fault: [0x13429b:<ver>], Internal fault: [0x67fec6:<ver>], or Internal fault: [0xe81a5a:<ver>]. This has been fixed.

Libraries and system headers

• [SDCOMP-54483]  The arm_neon.h system header incorrectly failed to define the poly64_t, poly64x1_t, and poly64x2_t types when compiling for AArch32 state. This has been fixed.

• [SDCOMP-53777]  In certain circumstances, when the program is built for a target that supports a hardware floating-point unit, the Arm C++ library implementation of the std::binomial_distribution::operator() function could return an incorrect result. This has been fixed.

• [SDCOMP-53677]  In certain circumstances, when using microlib and the program contains a call to the scanf() function with a non-constant format string, an issue in microlib could incorrectly cause the linker to report Error: L6218E: Undefined symbol. This has been fixed.

• [SDCOMP-52285]  The Arm C library implementations of certain functions specified parameters that incorrectly did not conform to the Application Binary Interface (ABI) for the Arm Architecture for calls to the __aeabi_idiv0() or __aeabi_ldiv0() functions. This has been fixed.

• [SDCOMP-52050]  In certain circumstances, the Arm C library implementation of the atan2f() function could incorrectly return positive zero instead of negative zero. This has been fixed.

• [SDCOMP-51695]  The Arm C library implementation of the freopen() function incorrectly failed to return a null pointer when called with a null pointer as the filename parameter. This has been fixed.

• [SDCOMP-50456]  The Arm C++ library variant built with C++ exception handling, for AArch32 state and a big-endian target, incorrectly failed to catch C++ exceptions. This has been fixed.

• [SDCOMP-30358]  In certain circumstances, the Arm C++ library implementation of regex_traits::isctype() could return an incorrect result when the character class is the word character class. This has been fixed.

• [SDCOMP-30357]  The Arm C++ library implementation of std::regex_traits<T>::lookup_classname() returned an incorrect value when its argument is equal to punct. This has been fixed.

Known issues in Arm Compiler 6.13

• [SDCOMP-54453]  When disassembling an ELF file that contains a BFCSEL instruction and a relocation of type R_ARM_THM_BF12, the fromelf utility incorrectly reports the type of the relocation as R_ARM_THM_BF13.

• [SDCOMP-54427]  In rare circumstances, when compiling with -fsanitize=memtag, the compiler can generate incorrect memory tagging code. Subsequently, this can result in spurious memory tagging exceptions against addresses in the stack area.

• [SDCOMP-54391]  When compiling with -fstack-protector-strong at -O0, the compiler incorrectly fails to enable stack protection for functions containing certain kinds of address-taking of non-array local variables. To avoid this issue, compile with -fstack-protector-all instead of -fstack-protector-strong.

• [SDCOMP-50470]  The compiler incorrectly fails to report an error for a function call that involves invalid default argument promotion for the _Float16 data type.

Release notes for Arm Compiler 6.12

1. Introduction

Arm Compiler 6.12 adds:

• Support for Cortex-A65AE, Neoverse E1, and Neoverse N1.
• Alpha support for intrinsics for the optional Memory Tagging Extension in Armv8.5-A.
• Support for the heap protection with memory tagging feature for the optional Memory Tagging Extension in Armv8.5-A.
• Alpha support for the memory tagging stack protection feature for the optional Memory Tagging Extension in Armv8.5-A.
• Support for the stack protection feature.
• Support for the global named register variables feature.

Arm Compiler 6.12 is intended for use:

• In conjunction with Arm Development Studio.
• In conjunction with Keil MDK.
• As a standalone toolchain installation.

A suitable license from one of these products must be available. Contact your sales representative or visit https://developer.arm.com/buy-arm-products to enquire about a license.

If you are using a floating license, your license server must be running armlmd and lmgrd version 11.14.1.0 or later. Arm recommends that you always use the latest version of the license server software that is available from https://developer.arm.com/products/software-development-tools/license-management/downloads.

EULA
Arm Compiler has moved away from its own compiler-specific EULA. The Arm Compiler 6.12 release adopts a unified EULA, which is used by Arm Development Studio and other Arm products.

LLVM Relicensing
Arm Compiler 6.12 includes components from the LLVM project, including the C++ libraries (libc++, libc++abi, and libunwind) that can be linked into user code. The license terms for the code included for the libraries has changed, as part of LLVM relicensing, to the Apache 2.0 license with LLVM exceptions. You should check your obligations under the terms of the new licenses (https://llvm.org/LICENSE.txt). More information on the LLVM relicensing can be found at http://www.llvm.org/foundation/relicensing/.

1.1 Arm Compiler 6 Configuration

Arm Compiler 6 is the successor to Arm Compiler 5 and includes the components listed below. See the Migration and Compatibility Guide in the product documentation for more information on migrating projects from previous versions.

• armclang
  • armclang is the successor to armcc and is based on LLVM and Clang technology.
• armlink, armasm, fromelf, armar
  • armlink, armasm, fromelf, and armar have been extended to support Armv8 and behave similarly to Arm Compiler 5.
• Arm C and C++ libraries for embedded systems
  • The standard Arm Compiler embedded libraries have been extended to support Armv8 and behave similarly to those found in Arm Compiler 5.
  • Arm Compiler 6 includes the libc++ library as the C++ Standard Template Library.

Note regarding assemblers:

• Arm Compiler 6 adopts the LLVM integrated assembler as default because it aligns more closely with GNU assembler syntax, improving portability between the GNU and Arm Compiler toolchains. The LLVM integrated assembler is called by default by armclang. A side effect is that Arm Compiler 6 will not compile C/C++ source files which contain legacy armcc inline or embedded assembler.
• The legacy assembler (armasm) is not called by default, but is included in Arm Compiler 6 for assembling source code written in armasm syntax. The legacy assembler is included to help with migration of existing projects from Arm Compiler 5 or earlier. Although the legacy assembler has been extended to support Armv8, this assembler does not include all the features of the latest Armv8 updates, and Arm has no plans to add further architectural support to armasm. For new projects, Arm recommends the armclang inline assembler and integrated assembler.

1.2 What's Supported in Arm Compiler 6.12?

The following table describes the level of support in Arm Compiler 6.12 for each of the listed Arm Architectures and Processors. The exact set of Architectures and Processors you are able to target using Arm Compiler 6.12 depends on which license and toolkit you are using. Check with your supplier for more information.

Architectures and Processors		Support Level	License
			Keil MDK				Arm Development Studio
			Lite	Essential	Plus	Professional	Bronze	Silver	Gold
Armv8-A up to 8.5-A	Neoverse N1/E1	Product feature	-	-	-	-	-	-	(*)
	Cortex-A76AE/76/65AE	Product feature	-	-	-	-	-	-	(*)
	Cortex-A75/73/72/57/55/53/35/32	Product feature	-	-	-	-	-	-	Yes
Armv7-A	Cortex-A17/15/12/9/8/7/5	Product feature	-	-	-	Yes	-	Yes	Yes
Armv8-R	Cortex-R52	Product feature	-	-	-	-	-	-	Yes
Armv7-R	Cortex-R8/7/5	Product feature	-	-	-	-	-	Yes	Yes
	Cortex-R4F/4	Product feature	-	-	Yes	Yes	-	Yes	Yes
Armv8-M	Cortex-M35P/33/23	Product feature	-	Non-secure only	Yes	Yes	Yes	Yes	Yes
Armv7-M	SC300	Product feature	-	-	Yes	Yes	Yes	Yes	Yes
	Cortex-M7/4/3	Product feature	32 Kbyte code limit	Yes	Yes	Yes	Yes	Yes	Yes
Armv6-M	SC000	Product feature	-	-	Yes	Yes	Yes	Yes	Yes
	Cortex-M1/0/0+	Product feature	32 Kbyte code limit	Yes	Yes	Yes	Yes	Yes	Yes
Arm architectures earlier than Armv6-M		Unsupported	-	-	-	-	-	-	-
Non Arm architectures		Unsupported	-	-	-	-	-	-	-

* When processors are available in physical devices. Contact your supplier for more information.

Support Level	Description
Product features	Production quality. Highest support priority.
Beta product features	Implementation complete, but not thoroughly tested. User experimentation and feedback is welcomed.
Alpha product features	Implementation is not complete and not thoroughly tested. User experimentation and feedback is welcomed.
Community features	Additional features that are available in the open-source technology Arm Compiler 6 is built on. Arm makes no claims about the quality level or the degree of functionality of these features. User experimentation and feedback is welcomed.
Unsupported features	Features that are either not present in the toolchain or have been deprecated. These features are completely untested. Use entirely at your own risk.

For more information on the supported features and level of support for features, see the product documentation.

2. Installation Instructions

If you received Arm Compiler 6.12 as part of a toolkit, for example Arm Development Studio, the toolkit installer takes care of the installation process. Please refer to the installation instructions for the toolkit in such cases.

For all other cases, you must select an appropriate installation location depending on how you intend to use Arm Compiler 6.12:

• Integrated into Arm Development Studio 2018.0 or later.
• Integrated into Keil MDK 5.22 or later.
• As a standalone product.

2.1. Integration into Arm Development Studio 2018.0 or later

Arm Compiler 6.12 can be installed in any location, including the default location, that is outside of the Arm Development Studio installation directory.

After it is installed, you can integrate the toolchain with Arm Development Studio 2018.0 or later by following the instructions available at https://developer.arm.com/docs/101470/latest/configure-arm-development-studio/register-a-compiler-toolchain.

2.2. Integration into Keil MDK 5.22 or later

Arm Compiler 6.12 must be installed within the ARM subdirectory of the Keil MDK installation directory. For example, if your Keil MDK installation directory is in C:\Keil_v5, the recommended installation path is C:\Keil_v5\ARM\ARMCompiler6.12.

Only the 32-bit Windows variant of Arm Compiler 6.12 can be used with a Keil Single-User License or Keil Floating-User License.

After it is installed, you can integrate the toolchain into a Keil MDK project by following the instructions in the tutorial available at http://www.keil.com/support/man/docs/uv4/uv4_armcompilers.htm.

2.3. Use as a standalone product

Arm Compiler 6.12 can be installed standalone, that is without requiring Arm Development Studio nor Keil MDK to be installed on that machine.

Licensing must be set up manually:

• Ensure that the ARMLMD_LICENSE_FILE environment variable is pointing to your license file or license server.
• Depending on the license that you have, you will need to configure the environment further. Please visit https://developer.arm.com/products/software-development-tools/license-management/resources/product-and-toolkit-configuration for instructions on how to configure your environment.

2.4. Installation on Linux

Arm Compiler 6.12 has been tested on the following supported platforms:

• Red Hat Enterprise Linux 6 Workstation, 64-bit only.
• Red Hat Enterprise Linux 7 Workstation, 64-bit only.
• Ubuntu Desktop Edition 14.04 LTS, 64-bit only.
• Ubuntu Desktop Edition 16.04 LTS, 64-bit only.

Arm Compiler 6.12 is not expected to work on older platforms.

To install Arm Compiler 6.12, run (not source) install_x86_64.sh and follow the on-screen instructions. The installer unpacks Arm Compiler 6.12 into your chosen directory.

The armclang binary is dynamically linked to a copy of libstdc++ installed under your chosen directory as part of Arm Compiler 6.12.

2.5. Installation on Windows

Arm Compiler 6.12 has been tested on the following supported platforms:

• Windows Server 2012, 64-bit only.
• Windows 7 Enterprise SP1.
• Windows 7 Professional SP1.
• Windows 8.1 Enterprise, 64-bit only.
• Windows 8.1 Professional, 64-bit only.
• Windows 10 Enterprise, 64-bit only.
• Windows 10 Professional, 64-bit only.

Arm Compiler 6.12 is not expected to work on older platforms.

Arm Compiler 6.12 has been tested on the latest feature update of Windows 10 available as of February 2019. It is expected to work well on later updates.

To install Arm Compiler 6.12, run win-x86_64\setup.exe on a supported 64-bit Windows platform or win-x86_32\setup.exe on a supported 32-bit Windows platform and follow the on-screen instructions. If you have an earlier version of Arm Compiler 6 installed and you wish to perform an upgrade, it is recommended that you uninstall the previous version before installing the new version of Arm Compiler 6.

Arm Compiler 6 is built with Microsoft Visual Studio 2015 and requires the Universal C Runtime in Windows to be installed. For further information see https://support.microsoft.com/en-gb/help/2999226/update-for-universal-c-runtime-in-windows.

3. Uninstall

On Linux, delete the Arm Compiler 6.12 installation directory.

On Windows, use Programs and Features in Control Panel, select ARM Compiler 6.12, and click the Uninstall button.

4. Documentation

The following documentation is available for Arm Compiler 6.12:

• User Guide.
• armar User Guide.
• armasm User Guide.
• armclang Reference Guide.
• armlink User Guide.
• fromelf User Guide.
• Arm C and C++ Libraries and Floating-Point Support User Guide.
• Migration and Compatibility Guide.
• Errors and Warnings Reference Guide.
• Scalable Vector Extension User Guide.

For more information, please see Arm Compiler 6 documentation in developer.arm.com.

In January 2018, Arm published information about the Variant 1: bounds check bypass (CVE-2017-5753) vulnerability. Refer to https://developer.arm.com/support/security-update/compiler-support-for-mitigations and use the Migration path recommendations with this Arm Compiler release to help mitigate against the vulnerability. Note that the API of the mitigation is subject to change.

5. Feedback and Support

Your feedback is important to us, and you are welcome to send us defect reports and suggestions for improvement on any aspect of the product. Contact your supplier or visit https://support.developer.arm.com and open a case with feedback or support issues. Where appropriate, please provide the --vsn output from the tool, the complete content of any error message that the tools produce, and include any source code, other files, and command-lines necessary to reproduce the issue.

6. Release History and Changes

The following are the releases to date of the Arm Compiler 6.12 series:

• 6.12 (released February 2019)

Below is a summary of the changes in each release, including new features and defect fixes. Changes are listed since the previous release in each case unless otherwise specified. Each itemized change is accompanied by a unique SDCOMP-<NNNNN> identifier. If you need to contact Arm about a specific issue within these release notes, please quote the appropriate identifier.

Changes in Arm Compiler 6.12

Changes are listed since the previous feature release, Arm Compiler 6.11.

General changes in Arm Compiler 6.12

• [SDCOMP-52966]  Support has been added for the Prediction Restriction by Context, Speculation Barrier, and Speculative Store Bypass Safe instructions in Armv8-A and later targets. These instructions are mandatory in Armv8.5-A and later targets, and optional in Armv8.4-A and earlier targets. The Prediction Restriction by Context and Speculative Store Bypass Safe instructions are only available in AArch64 state.

  To target these instructions, select from the following options:

  Architecture	armclang	armlink and fromelf
  Armv8.5-A or later with the Prediction Restriction by Context, Speculation Barrier, and Speculative Store Bypass Safe instructions	-march=armv8.5-a	None required
  Armv8.<ver>-A with the Prediction Restriction by Context instructions, where <ver> is 1, 2, 3, or 4	-march=armv8.<ver>-a+predres	None required
  Armv8.<ver>-A with the Speculation Barrier instruction, where <ver> is 1, 2, 3, or 4	-march=armv8.<ver>-a+sb	None required
  Armv8.<ver>-A with the Speculative Store Bypass Safe instructions, where <ver> is 1, 2, 3, or 4	-march=armv8.<ver>-a+ssbs	None required
  Armv8-A with the Prediction Restriction by Context instructions	-march=armv8-a+predres	None required
  Armv8-A with the Speculation Barrier instruction	-march=armv8-a+sb	None required
  Armv8-A with the Speculative Store Bypass Safe instructions	-march=armv8-a+ssbs	None required

  The legacy assembler (armasm) does not support these instructions.

  For more information about these options, refer to the -march and -mcpu sections of the armclang Reference Guide.

• [SDCOMP-53105]  Support has been added for the Cortex-A65AE processor. To target Cortex-A65AE, select from the following options:

  armclang:

  • --target=aarch64-arm-none-eabi -mcpu=cortex-a65ae.

  armasm, armlink, and fromelf:

  • --cpu=8.2-A.64.

• [SDCOMP-53106]  Support has been added for the Neoverse E1 processor. To target Neoverse E1, select from the following options:

  armclang:

  • --target=aarch64-arm-none-eabi -mcpu=neoverse-e1.

  armasm, armlink, and fromelf:

  • --cpu=8.2-A.64.

• [SDCOMP-53107]  Support has been added for the Neoverse N1 processor. To target Neoverse N1, select from the following options:

  armclang:

  • --target=aarch64-arm-none-eabi -mcpu=neoverse-n1 for AArch64 state.
  • --target=arm-arm-none-eabi -mcpu=neoverse-n1 for AArch32 state.

  armasm, armlink, and fromelf:

  • --cpu=8.2-A.64 for AArch64 state.
  • --cpu=8.2-A.32 for AArch32 state.

Enhancements in Arm Compiler 6.12

Compiler and integrated assembler (armclang)

• [SDCOMP-53080]  Support has been added for the global named register variables feature for AArch32 state.

  For more information about this feature, refer to the Global named register variables and -ffixed-rN sections of the armclang Reference Guide.

  This enables migration of source code that previously used the legacy armcc global named register variables feature.

  For information on replacing uses of the legacy armcc option --global_reg=reg_name and the __global_reg(N) keyword, refer to the Migration and Compatibility Guide.

• [SDCOMP-52575]  Support has been added for the stack protection feature. Use one of the following options to enable the stack protection feature:

  • -fstack-protector.
  • -fstack-protector-all.
  • -fstack-protector-strong.

  For more information about these options, refer to the -fstack-protector, -fstack-protector-strong, -fstack-protector-all, -fno-stack-protector section of the armclang Reference Guide.

  This enables migration of source code that previously used the legacy armcc options --protect_stack and --protect_stack_all.

  For information on replacing uses of the legacy armcc options --protect_stack and --protect_stack_all, refer to the Migration and Compatibility Guide.

• [SDCOMP-52165]  Support has been added for Arm C Language Extensions (ACLE) special register intrinsics that can access a system register containing a floating-point value. To access a system register containing a floating-point value, select from the following intrinsics:

  • __arm_rsrf() to read a 32-bit system register containing a single-precision floating-point value.
  • __arm_wsrf() to write a 32-bit system register containing a single-precision floating-point value.
  • __arm_rsrf64() to read a 64-bit system register containing a double-precision floating-point value.
  • __arm_wsrf64() to write a 64-bit system register containing a double-precision floating-point value.

• [SDCOMP-51418]  Alpha support has been added for the memory tagging stack protection (stack tagging) feature for the optional Memory Tagging Extension in Armv8.5-A. The memory tagging stack protection feature can be used with the following options:

  armclang:

  • -mmemtag-stack to enable the generation of stack protection code that uses the Memory Tagging Extension.

  armlink:

  • --library_security=protection to control the selection of protected libraries for memory tagging stack protection.

  For more information about these options, refer to the -mmemtag-stack, -mno-memtag-stack section of the armclang Reference Guide, and the --library_security=protection section of the armlink User Guide.

• [SDCOMP-51359]  Alpha support has been added for Arm C Language Extensions (ACLE) intrinsics for the optional Memory Tagging Extension in Armv8.5-A.

Libraries and system headers

• [SDCOMP-51360]  Support has been added for the heap protection with memory tagging feature for the optional Memory Tagging Extension in Armv8.5-A.

  For more information about this feature, refer to the Choosing a heap implementation for memory allocation functions section of the Arm C and C++ Libraries and Floating-Point Support User Guide.

Defect fixes in Arm Compiler 6.12

Compiler and integrated assembler (armclang)

• [SDCOMP-52855]  In certain circumstances, when compiling at -O0, the compiler could generate incorrect code for an expression involving a bitwise AND operator followed by a bitwise left-shift operator. This has been fixed.

• [SDCOMP-52600]  In certain circumstances, when compiling at any optimization level except -O0, the compiler could generate incorrect code for a function that conditionally dereferences the same pointer more than once. This has been fixed.

• [SDCOMP-52544]  In certain circumstances, when compiling with -mno-unaligned-access at any optimization level except -O0, the compiler could generate code that incorrectly performs an unaligned access to a member M of a local struct variable, where M has an alignment requirement of less than 4 bytes and has a size that is a multiple of 4 bytes. This has been fixed.

• [SDCOMP-52274]  In certain circumstances, when compiling at any optimization level except -O0 for AArch32 state, the compiler could generate incorrect code for a comparison operation between a negative integer constant and a variable with a size smaller than 32 bits. This has been fixed.

• [SDCOMP-52206]  When compiling with -masm=auto or -masm=armasm, and -mfpu=name where name specifies an FPU with the Advanced SIMD Extension, the compiler would incorrectly specify a target without the Advanced SIMD Extension when invoking armasm. This has been fixed.

  The following -mfpu=name options specify an FPU with the Advanced SIMD Extension:

  • neon.
  • neon-fp-16.
  • neon-vfpv4.
  • neon-fp-armv8.
  • crypto-neon-fp-armv8.

• [SDCOMP-52117]  In certain circumstances, when compiling for AArch64 state, the compiler could incorrectly fail to generate a data mapping symbol for a constant global variable V. Subsequently, at link-time, the linker would treat V as Code type instead of Data type, which could result in unexpected run time behavior. This has been fixed.

• [SDCOMP-51640]  In certain circumstances, when compiling for T32 state, the compiler could incorrectly assume that the address A of a function is always aligned to a 4-byte boundary, and subsequently generate incorrect code for calculations involving A that use the bitwise AND operator. This has been fixed.

• [SDCOMP-51578]  In certain circumstances, when assembling for a target that supports a hardware floating-point unit and AArch32 state, the inline assembler and integrated assembler could incorrectly fail to report an error for a VLDM or VSTM instruction that specifies more than 16 doubleword registers in the list of extension registers to be accessed. Instead, the inline assembler and integrated assembler could generate a VLDM or VSTM instruction with an UNPREDICTABLE encoding. This has been fixed. The inline assembler and integrated assembler now report error: list of registers must be at least 1 and at most 16.

• [SDCOMP-50652]  In certain circumstances, when compiling with -mpixolib, the compiler could generate incorrect initialization code for global variables that depend on the address of position-independent data or code. This has been fixed.

Linker (armlink)

• [SDCOMP-52640]  In certain circumstances, when linking with --callgraph, --info=stack, or --info=summarystack, and at least one input object contains an unreferenced local function that does not make any function calls, the linker could report Internal fault: [0x2c9471:<ver>]. This has been fixed.

• [SDCOMP-52261]  In certain circumstances, when linking input objects that contain debug information and a symbol has multiple definitions, the linker could incorrectly report Error: L6743E: Relocation #REL:<rel> in <object>(.debug_info) with respect to <symbol> that has an alternate def. Internal consistency check failed. This has been fixed.

Libraries and system headers

• [SDCOMP-52623]  The Arm threaded C++ library implementations of the std::condition_variable::wait_for() and std::this_thread::sleep_for() functions for AArch64 state incorrectly used the unsupported long double type. This has been fixed.

• [SDCOMP-52288]  The Arm C library provides an alternative heap implementation named Heap2, also referred to as the real time heap. When compiling for AArch64 state, the Heap2 implementation is selected, the program implements the required Arm C library mutex functions, and the Arm C library memory allocation function posix_memalign() fails to allocate memory, posix_memalign() incorrectly would not release the mutex for the heap. This could result in unexpected run time behavior. This has been fixed.

• [SDCOMP-51346]  The Arm C library provides an alternative heap implementation named Heap2, also referred to as the real time heap. When compiling for AArch64 state and the Heap2 implementation is selected, the Arm C library memory allocation functions malloc(), realloc(), calloc(), and free() incorrectly specified an 8-byte alignment requirement instead of a 16-byte alignment requirement for a pointer to a memory block. Additionally, the implementation incorrectly used a 32-bit variable to store the size of a memory block and consequently did not support memory blocks larger than 4GB in size. This could result in unexpected run time behavior. This has been fixed.

• [SDCOMP-50975]  In certain circumstances, when a program raises a signal using the Arm C library raise() function, the Arm C library implementation of the __default_signal_handler() function could incorrectly fail to print an error message when handling the raised signal. This has been fixed.

Known issues in Arm Compiler 6.12

• [SDCOMP-50507]  Compiler support for the AArch64 Armv8.2-A half-precision floating-point intrinsics has the following limitations:

  • The compiler incorrectly reports fatal error: error in backend or error: cannot compile this builtin function yet for the floating-point absolute compare intrinsics and a subset of the floating-point integer conversion intrinsics.
  • The floating-point multiply extended (by element) intrinsics, vmulxh_lane_f16() and vmulxh_laneq_f16(), are not supported.

• [SDCOMP-50470]  The compiler incorrectly fails to report an error for a function call that involves invalid default argument promotion for the _Float16 data type.

Release notes for Arm Compiler 6.11

Table of Contents

1. 1. Introduction
   1. 1.1. Arm Compiler 6 Configuration
   2. 1.2. What's Supported in Arm Compiler 6.11?
2. 2. Installation Instructions
   1. 2.1. Integration into DS-5 5.20 or later
   2. 2.2. Integration into Keil MDK 5.22 or later
   3. 2.3. Use as a standalone product
   4. 2.4. Installation on Linux
   5. 2.5. Installation on Windows
3. 3. Uninstall
4. 4. Documentation
5. 5. Feedback and Support
6. 6. Release History and Changes

1. Introduction

This is the initial set of release notes provided at the time of the release. For the latest copy of the release notes, see the latest version on https://developer.arm.com.

Arm Compiler 6.11 adds:

• armclang inline assembler and integrated assembler support for the Armv8.5-A architecture.
• Support for the optional Memory Tagging Extension in Armv8.5-A.
• Support for the optional Random Number Instructions in Armv8.5-A.
• Support for branch protection features for Armv8.3-A and later targets.
• Support for half-precision floating-point fused multiply long instructions in Armv8.2-A and later targets.

Arm Compiler 6.11 is intended for use:

• In conjunction with DS-5.
• In conjunction with Keil MDK.
• As a standalone toolchain installation.

A suitable license from one of these products must be available. Contact your sales representative or visit https://developer.arm.com/products/buy-arm-products to enquire about a license.

If you are using a floating license, your license server must be running armlmd and lmgrd version 11.14.1.0 or later. Arm recommends that you always use the latest version of the license server software that is available from https://developer.arm.com/products/software-development-tools/license-management/downloads.

1.1 Arm Compiler 6 Configuration

Arm Compiler 6 is the successor to Arm Compiler 5 and includes the components listed below. See the Migration and Compatibility Guide in the product documentation for more information on migrating projects from previous versions.

• armclang
  • armclang is the successor to armcc and is based on LLVM and Clang technology.
• armlink, armasm, fromelf, armar
  • armlink, armasm, fromelf, and armar have been extended to support Armv8 and behave similarly to Arm Compiler 5.
• Arm C and C++ libraries for embedded systems
  • The standard Arm Compiler embedded libraries have been extended to support Armv8 and behave similarly to those found in Arm Compiler 5.
  • Arm Compiler 6 includes the libc++ library as the C++ Standard Template Library.

Note regarding assemblers:

• Arm Compiler 6 adopts the LLVM integrated assembler as default because it aligns more closely with GNU assembler syntax, improving portability between the GNU and Arm Compiler toolchains. The LLVM integrated assembler is called by default by armclang. A side effect is that Arm Compiler 6 will not compile C/C++ source files which contain legacy armcc inline or embedded assembler.
• The legacy assembler (armasm) is not called by default, but is included in Arm Compiler 6 for assembling source code written in armasm syntax. The legacy assembler is included to help with migration of existing projects from Arm Compiler 5 or earlier. Although the legacy assembler has been extended to support Armv8, this assembler does not include all the features of the latest Armv8 updates, and Arm has no plans to add further architectural support to armasm. For new projects, Arm recommends the armclang inline assembler and integrated assembler.

1.2 What's Supported in Arm Compiler 6.11?

The following table describes the level of support for the most common toolkits:

Architectures and Processors		Support Level	License / Configuration
			MDK-Lite	MDK-Essential	MDK-Plus	MDK-Professional	DS-5 Professional	DS-5 Ultimate
Armv8-A	Cortex-A76AE/76/75/73/72/57/55/53/35/32	Product feature	-	-	-	-	-	Yes
Armv7-A	Cortex-A17/15/12/9/8/7/5	Product feature	-	-	-	Yes	Yes	Yes
Armv8-R	Cortex-R52	Product feature	-	-	-	-	-	Yes
Armv7-R	Cortex-R8/7/5	Product feature	-	-	-	-	Yes	Yes
	Cortex-R4F/4	Product feature	-	-	Yes	Yes	Yes	Yes
Armv8-M	Cortex-M35P/33/23	Product feature	-	Non-secure only	Yes	Yes	Yes	Yes
Armv7-M	SC300	Product feature	-	-	Yes	Yes	Yes	Yes
	Cortex-M7/4/3	Product feature	32 Kbyte code limit	Yes	Yes	Yes	Yes	Yes
Armv6-M	SC000	Product feature	-	-	Yes	Yes	Yes	Yes
	Cortex-M0/M0+	Product feature	32 Kbyte code limit	Yes	Yes	Yes	Yes	Yes
Arm architectures earlier than Armv6-M		Unsupported	-	-	-	-	-	-
Non Arm architectures		Unsupported	-	-	-	-	-	-

Support Level	Description
Product features	Production quality. Highest support priority.
Beta product features	Implementation complete, but not thoroughly tested. User experimentation and feedback is welcomed.
Alpha product features	Implementation is not complete and not thoroughly tested. User experimentation and feedback is welcomed.
Community features	Additional features that are available in the open-source technology Arm Compiler 6 is built on. Arm makes no claims about the quality level or the degree of functionality of these features. User experimentation and feedback is welcomed.
Unsupported features	Features that are either not present in the toolchain or have been deprecated. These features are completely untested. Use entirely at your own risk.

Check with your supplier for information about which architectures and processors are supported by other toolkits.

For more information on the supported features and level of support for features, see the product documentation.

2. Installation Instructions

If you received Arm Compiler 6.11 as part of a toolkit, for example DS-5, the toolkit installer takes care of the installation process. Please refer to the installation instructions for the toolkit in such cases.

For all other cases, you must select an appropriate installation location depending on how you intend to use Arm Compiler 6.11:

• Integrated into DS-5 5.20 or later.
• Integrated into Keil MDK 5.22 or later.
• As a standalone product.

2.1. Integration into DS-5 5.20 or later

Arm Compiler 6.11 can be installed in any location, including the default location, providing this is outside of a DS-5 product installation.

After it is installed, you can integrate the toolchain with DS-5 5.20 or later by following the instructions in the tutorial available at https://developer.arm.com/products/software-development-tools/ds-5-development-studio/resources/tutorials/adding-new-compiler-toolchains-to-ds-5.

Arm recommends using Arm Compiler 6.11 from the DS-5 Eclipse IDE or DS-5 Command Prompt. When using the toolchain outside these environments, you might need to configure the following environment variables:

• Set ARM_PRODUCT_PATH to the path to the sw/mappings directory within your DS-5 installation.
• Set ARM_TOOL_VARIANT appropriately if you are not using a DS-5 Professional Edition license.

For further information see https://developer.arm.com/products/software-development-tools/license-management/resources/product-and-toolkit-configuration.

2.2. Integration into Keil MDK 5.22 or later

Arm Compiler 6.11 must be installed within the ARM subdirectory of the Keil MDK installation directory. For example, if your Keil MDK installation directory is in C:\Keil_v5, the recommended installation path is C:\Keil_v5\ARM\ARMCompiler6.11.

Only the 32-bit Windows variant of Arm Compiler 6.11 can be used with a Keil Single-User License or Keil Floating-User License.

After it is installed, you can integrate the toolchain into a Keil MDK project by following the instructions in the tutorial available at http://www.keil.com/support/man/docs/uv4/uv4_armcompilers.htm.

2.3. Use as a standalone product

Arm Compiler 6.11 can be installed in any location, including the default location, providing this is outside of a DS-5 product installation.

Ensure that the ARMLMD_LICENSE_FILE environment variable is pointing to your license file or license server.

Set ARM_TOOL_VARIANT appropriately if you are not using a DS-5 Professional Edition license.

For further information see https://developer.arm.com/products/software-development-tools/license-management/resources/product-and-toolkit-configuration.

2.4. Installation on Linux

Arm Compiler 6.11 has been tested on the following supported platforms:

• Red Hat Enterprise Linux 6 Workstation, 64-bit only.
• Red Hat Enterprise Linux 7 Workstation, 64-bit only.
• Ubuntu Desktop Edition 14.04 LTS, 64-bit only.
• Ubuntu Desktop Edition 16.04 LTS, 64-bit only.

Arm Compiler 6.11 is not expected to work on older platforms.

To install Arm Compiler 6.11, run (not source) install_x86_64.sh and follow the on-screen instructions. The installer unpacks Arm Compiler 6.11 into your chosen directory.

The armclang binary is dynamically linked to a copy of libstdc++ installed under your chosen directory as part of Arm Compiler 6.11.

2.5. Installation on Windows

Arm Compiler 6.11 has been tested on the following supported platforms:

• Windows Server 2012, 64-bit only.
• Windows 7 Enterprise SP1.
• Windows 7 Professional SP1.
• Windows 8.1 Enterprise, 64-bit only.
• Windows 8.1 Professional, 64-bit only.
• Windows 10 Enterprise, 64-bit only.
• Windows 10 Professional, 64-bit only.

Arm Compiler 6.11 is not expected to work on older platforms.

Arm Compiler 6.11 has been tested on the latest feature update of Windows 10 available as of October 2018. It is expected to work well on later updates.

To install Arm Compiler 6.11, run win-x86_64\setup.exe on a supported 64-bit Windows platform or win-x86_32\setup.exe on a supported 32-bit Windows platform and follow the on-screen instructions. If you have an earlier version of Arm Compiler 6 installed and you wish to perform an upgrade, it is recommended that you uninstall the previous version before installing the new version of Arm Compiler 6.

Arm Compiler 6 is built with Microsoft Visual Studio 2015 and requires the Universal C Runtime in Windows to be installed. For further information see https://support.microsoft.com/en-gb/help/2999226/update-for-universal-c-runtime-in-windows.

3. Uninstall

On Linux, delete the Arm Compiler 6.11 installation directory.

On Windows, use Programs and Features in Control Panel, select ARM Compiler 6.11, and click the Uninstall button.

4. Documentation

The following documentation is available for Arm Compiler 6.11:

• User Guide.
• armar User Guide.
• armasm User Guide.
• armclang Reference Guide.
• armlink User Guide.
• fromelf User Guide.
• Arm C and C++ Libraries and Floating-Point Support User Guide.
• Arm Instruction Set Reference Guide.
• Migration and Compatibility Guide.
• Software Development Guide.
• Errors and Warnings Reference Guide.
• Scalable Vector Extension User Guide.

For more information, please see Arm Compiler 6 documentation in developer.arm.com.

In January 2018, Arm published information about the Variant 1: bounds check bypass (CVE-2017-5753) vulnerability. Refer to https://developer.arm.com/support/security-update/compiler-support-for-mitigations and use the Migration path recommendations with this Arm Compiler release to help mitigate against the vulnerability. Note that the API of the mitigation is subject to change.

5. Feedback and Support

Your feedback is important to us, and you are welcome to send us defect reports and suggestions for improvement on any aspect of the product. Contact your supplier or visit https://support.developer.arm.com and open a case with feedback or support issues. Where appropriate, please provide the --vsn output from the tool, the complete content of any error message that the tools produce, and include any source code, other files, and command-lines necessary to reproduce the issue.

6. Release History and Changes

The following are the releases to date of the Arm Compiler 6.11 series:

• 6.11 (released October 2018)

Below is a summary of the changes in each release, including new features and defect fixes. Changes are listed since the previous release in each case unless otherwise specified. Each itemized change is accompanied by a unique SDCOMP-<NNNNN> identifier. If you need to contact Arm about a specific issue within these release notes, please quote the appropriate identifier.

Changes in Arm Compiler 6.11

Changes are listed since the previous feature release, Arm Compiler 6.10.

General changes in Arm Compiler 6.11

• [SDCOMP-51284]  Support has been added for branch protection features for Armv8.3-A and later targets. Branch protection features can be used with the following options:

  armclang:

  • -mbranch-protection=protection to specify the level or type of branch protection.

  armlink:

  • --library_security=protection to control the selection of protected libraries for branch protection.

  For more information about these options, refer to the -mbranch-protection section of the armclang Reference Guide, and the --library_security=protection section of the armlink User Guide.

• [SDCOMP-49171]  __declspec has been deprecated.

  For information on replacing uses of __declspec with __attribute__, refer to the Migration and Compatibility Guide.

• [SDCOMP-51014]  Support has been added for the Cortex-A76 processor. To target Cortex-A76, select from the following options:

  armclang:

  • --target=aarch64-arm-none-eabi -mcpu=cortex-a76 for AArch64 state.
  • --target=arm-arm-none-eabi -mcpu=cortex-a76 for AArch32 state.

  armasm, armlink, and fromelf:

  • --cpu=8.2-A.64 for AArch64 state.
  • --cpu=8.2-A.32 for AArch32 state.

• [SDCOMP-51015]  Support has been added for the Cortex-M35P processor. To target Cortex-M35P, select from the following options:

  armclang:

  • --target=arm-arm-none-eabi -mcpu=cortex-m35p for a variant with DSP and FP.
  • --target=arm-arm-none-eabi -mcpu=cortex-m35p+nodsp for a variant without DSP but with FP.
  • --target=arm-arm-none-eabi -mcpu=cortex-m35p -mfloat-abi=soft for a variant with DSP but without FP.
  • --target=arm-arm-none-eabi -mcpu=cortex-m35p+nodsp -mfloat-abi=soft for a variant without DSP and FP.

  armasm, armlink, and fromelf:

  • --cpu=Cortex-M35P for a variant with DSP and FP.
  • --cpu=Cortex-M35P.no_dsp for a variant without DSP but with FP.
  • --cpu=Cortex-M35P --fpu=SoftVFP for a variant with DSP but without FP.
  • --cpu=Cortex-M35P.no_dsp --fpu=SoftVFP for a variant without DSP and FP.

• [SDCOMP-50232]  Support for ELF sections that contain the legacy SHF_COMDEF ELF section flag has been deprecated.

  The following have been deprecated:

  • The COMDEF section attribute of the legacy armasm syntax AREA directive.
  • Linking with legacy objects that contain ELF sections with the legacy SHF_COMDEF ELF section flag.

  Use the GRP_COMDAT ELF section flag instead of the legacy SHF_COMDEF ELF section flag by:

  • Replacing the COMDEF section attribute of the legacy armasm syntax AREA directive with the COMGROUP=symbol_name section attribute.
  • Rebuilding incompatible legacy objects that were built using Arm Compiler 5 or earlier using one of the following:
    • The same version of the compiler as before, but with --dwarf3.
    • Arm Compiler 6.

  For more information about the COMGROUP=symbol_name section attribute of the legacy armasm syntax AREA directive, refer to the armasm User Guide.

  For more information about COMDAT groups and the GRP_COMDAT ELF section flag, refer to the Itanium C++ ABI.

• [SDCOMP-50292]  The legacy R-type dynamic linking model, which does not conform to the 32-bit Application Binary Interface for the Arm Architecture, has been deprecated.

  The following have been deprecated:

  • Linking with --reloc.
  • Linking without --base_platform with a scatter file that contains the RELOC load region attribute.

  Use the Base Platform and Base Platform Application Binary Interface (BPABI) linking models to generate relocatable executable files instead of using the legacy R-type dynamic linking model.

  For information about the Base Platform and BPABI linking models, refer to the armlink User Guide.

• [SDCOMP-51090]  C++14 source language modes are now fully supported. Use one of the following options to enable the compilation of C++14 source code:

  • -std=c++14
  • -std=gnu++14

  The default is -std=gnu++14. For more information about these options, refer to the armclang Reference Guide and the LLVM component versions and language compatibility section of the Software Development Guide.

• [SDCOMP-49953]  Support has been added for half-precision floating-point fused multiply long instructions in Armv8.2-A and later targets. To target Armv8.2-A or later with these instructions, select from the following options:

  armclang:

  • --target=aarch64-arm-none-eabi -march=armv8.<ext>-a+fp16fml for an Armv8.<ext>-A target and AArch64 state, where <ext> is 2 or later.
  • --target=arm-arm-none-eabi -march=armv8.<ext>-a+fp16fml for an Armv8.<ext>-A target and AArch32 state, where <ext> is 2 or later.

  armlink and fromelf:

  • --cpu=8.<ext>-A.64 for an Armv8.<ext>-A target and AArch64 state, where <ext> is 2 or 3.
  • --cpu=8.<ext>-A.32 for an Armv8.<ext>-A target and AArch32 state, where <ext> is 2 or 3.
  • Do not use the --cpu=name option for an Armv8.4-A or later target.

  The legacy assembler (armasm) does not support these instructions. These instructions are optional in Armv8.2-A and Armv8.3-A.

  For more information about these options, refer to the -march and -mcpu sections of the armclang Reference Guide.

• [SDCOMP-52123]  Support has been added for the Cortex-A76AE processor. To target Cortex-A76AE, select from the following options:

  armclang:

  • --target=aarch64-arm-none-eabi -mcpu=cortex-a76ae for AArch64 state.
  • --target=arm-arm-none-eabi -mcpu=cortex-a76ae for AArch32 state.

  armasm, armlink, and fromelf:

  • --cpu=8.2-A.64 for AArch64 state.
  • --cpu=8.2-A.32 for AArch32 state.

• [SDCOMP-52124]  Support has been added for the Armv8.5-A architecture. To target Armv8.5-A, select from the following options:

  armclang:

  • --target=aarch64-arm-none-eabi -march=armv8.5-a for AArch64 state.
  • --target=arm-arm-none-eabi -march=armv8.5-a for AArch32 state.

  armlink and fromelf:

  • Do not use the --cpu=name option.

  The armclang support for Armv8.5-A is limited to the inline assembler, integrated assembler, and branch protection features. In a future release of Arm Compiler, armclang will be enhanced to provide full support for Armv8.5-A.

  The legacy assembler (armasm) does not support Armv8.5-A.

• [SDCOMP-52126]  Support has been added for the optional Memory Tagging Extension in Armv8.5-A. To target Armv8.5-A with the Memory Tagging Extension, select from the following options:

  armclang:

  • --target=aarch64-arm-none-eabi -march=armv8.5-a+memtag for AArch64 state.

  armlink and fromelf:

  • Do not use the --cpu=name option.

  The legacy assembler (armasm) does not support Armv8.5-A. The Memory Tagging Extension is not supported in AArch32 state.

  For more information, refer to the armclang Reference Guide.

• [SDCOMP-52130]  Support has been added for the optional Random Number Instructions in Armv8.5-A. To target Armv8.5-A with the Random Number Instructions, select from the following options:

  armclang:

  • --target=aarch64-arm-none-eabi -march=armv8.5-a+rng for AArch64 state.

  armlink and fromelf:

  • Do not use the --cpu=name option.

  The legacy assembler (armasm) does not support Armv8.5-A. The Random Number Instructions are not supported in AArch32 state.

  For more information, refer to the armclang Reference Guide.

Enhancements in Arm Compiler 6.11

Compiler and integrated assembler (armclang)

• [SDCOMP-44910]  Support has been added for the -fident and -fno-ident options that control whether the output file contains compiler name and version information.

  For more information about these options, refer to the armclang Reference Guide.

Linker (armlink)

• [SDCOMP-44481]  The following linker options have been removed:

  • --compress_debug and --no_compress_debug.
  • --match=crossmangled.
  • --strict_enum_size and --no_strict_enum_size.
  • --strict_wchar_size and --no_strict_wchar_size.

General enhancements

• [SDCOMP-50658]  Support has been added for Position Independent eXecute Only (PIXO) library features for Armv7-M targets. PIXO libraries can be built with the following options:

  armclang:

  • -mpixolib to enable the generation of PIXO code.

  armlink:

  • --pixolib to create a PIXO library.

  For more information about these options, refer to the -mpixolib section of the armclang Reference Guide, and the --pixolib section of the armlink User Guide.

Defect fixes in Arm Compiler 6.11

Compiler and integrated assembler (armclang)

• [SDCOMP-51736]  When compiling with -mexecute-only for a Cortex-M23 target, the compiler would incorrectly fail to disable the generation of literal pools. Subsequently, at link-time, the linker would correctly report Error: L6837E: Illegal data mapping symbol found in execute-only section <object>(<section>). This has been fixed.

• [SDCOMP-51633]  When compiling for an Armv8-M target with the DSP Extension, the compiler would incorrectly fail to declare Arm C Language Extensions (ACLE) 32-bit SIMD intrinsics. This has been fixed.

• [SDCOMP-51568]  In certain circumstances, when compiling for a target that supports a hardware floating-point unit and AArch32 state, the compiler could incorrectly generate UNPREDICTABLE VLDM or VSTM instructions that specify more than 16 doubleword registers in the list of extension registers to be accessed. This has been fixed.

• [SDCOMP-51399]  In rare circumstances, when compiling with -g or -gdwarf-version at any optimization level except -O0, the compiler could incorrectly report Segmentation fault (core dumped). This has been fixed.

• [SDCOMP-51241]  In certain circumstances, when compiling on Windows, and the code contains a large number of nested preprocessor macro expansions, the compiler could incorrectly report error: clang frontend command failed due to signal. This has been fixed.

• [SDCOMP-51054]  In certain circumstances, the compiler could incorrectly fail to generate a data mapping symbol for a constant global variable V. Subsequently, at link-time, the linker would treat V as Code type instead of Data type, which could result in unexpected run time behavior. This has been fixed.

• [SDCOMP-50997]  The compiler would incorrectly place a function that is affected by a preceding #pragma clang text="A" statement and is annotated with __attribute__((section("B"))), where A and B are section names, in a section with an empty name. This has been fixed.

• [SDCOMP-50915]  In certain circumstances, when compiling for a big-endian Armv8-M target with the Main Extension, Security Extension, and a hardware floating-point unit, the compiler could generate incorrect code for a call from a secure function to a non-secure function that has a double-precision floating-point parameter. This has been fixed.

• [SDCOMP-50794]  In certain circumstances, when compiling at any optimization level except -O0 for an Armv6-M target or an Armv8-M target without the Main Extension, the compiler could incorrectly report fatal error: error in backend: Cannot select: <value>. This has been fixed.

• [SDCOMP-50662]  When compiling code that contains a variable V that has been assigned to a register R using the register keyword, the inline assembler would incorrectly fail to report an error for an inline assembly statement that uses V as an input or output variable, and contains R in its clobber list. This has been fixed. The inline assembler now reports error: asm-specifier for input or output variable conflicts with asm clobber list.

• [SDCOMP-50628]  In rare circumstances, when compiling at any optimization level except -O0 for AArch64 state, the compiler could generate incorrect code for a function that defines a local variable with an alignment greater than 16 bytes. This has been fixed.

• [SDCOMP-50565]  In certain circumstances, when compiling for AArch32 state with -mfloat-abi=hard, and the code contains a function F with a homogenous aggregate parameter P, the compiler could generate code that incorrectly overaligns elements within P when saving and restoring it from the stack. Such code does not conform to the Parameter Passing section of the Procedure Call Standard for the Arm® Architecture ABI release 2.09. This has been fixed.

• [SDCOMP-50563]  In certain circumstances, the inline assembler could incorrectly fail to report a warning for an inline assembly statement that contains a reserved register in its clobber list. This has been fixed. The inline assembler now reports warning: inline asm clobber list contains reserved registers: <registers>.

  The following registers can be reserved by the compiler:

  • SP, WZR, W19, or W29, when compiling for AArch64 state.
  • R6, R13, SP, R15, or PC, when compiling for AArch32 state.
  • R7, when compiling for T32 state.
  • R8, when compiling with -mpixolib for AArch32 state.
  • R9, when compiling with -frwpi, -frwpi-lowering, or -mpixolib for AArch32 state.
  • R11, when compiling for A32 state.

• [SDCOMP-50528]  In certain circumstances, when compiling a program that contains a function call with an overaligned parameter P of class, struct, or union type, the compiler could generate code that incorrectly misaligns P. Such code does not conform to the Parameter Passing section of the Procedure Call Standard for the Arm® Architecture ABI release 2.09. This has been fixed.

• [SDCOMP-50520]  In certain circumstances, when compiling for an Armv8-M target with the Security Extension and a hardware floating-point unit, the compiler could generate incorrect code for a non-secure function annotated with __attribute__((cmse_nonsecure_entry)). This has been fixed.

• [SDCOMP-50433]  The inline assembler and integrated assembler would incorrectly fail to report an error for a VMOV instruction that specifies a set of non-sequential single-precision floating-point registers as its floating-point register operands. Instead, it would generate an instruction that incorrectly has a different but valid set of single-precision floating-point register operands. This has been fixed. The inline assembler and integrated assembler now report error: destination operands must be sequential or error: source operands must be sequential.

• [SDCOMP-50416]  In certain circumstances, when compiling at -Oz, and the program contains functions that throw and catch C++ exceptions, the compiler could generate an incorrect ARM_EXIDX table entry for a function that uses C++ exceptions and preserves floating-point registers. Subsequently, when a C++ exception is thrown, the variables of a function that caches the exception could be corrupted. This has been fixed.

• [SDCOMP-50228]  In certain circumstances, when compiling at any optimization level except -O0, and the program contains a variable length array, the compiler could generate code that writes an incorrect value to the stack pointer immediately before the return instruction of a function. This has been fixed.

• [SDCOMP-50190]  When assembling for AArch64 state, the inline assembler and integrated assembler would incorrectly report error: expected readable system register for an MRS instruction that specifies ICH_ELRSR_EL2 as the special register to be accessed. This has been fixed.

• [SDCOMP-50186]  When compiling in a C++ source language mode, the compiler determined the resulting type incorrectly for a decltype specifier that is applied to an unparenthesized expression of the form S.M or S->M, where M is a static member of a class or struct S. The resulting type should be the same as M. Instead, it was incorrectly a reference to M. This has been fixed.

• [SDCOMP-50128]  When compiling for a big-endian target and AArch32 state, and the code contains an inline assembly statement that has an operand with the Q or R template modifier, the inline assembler would select an incorrect register for the operand. This has been fixed.

• [SDCOMP-50102]  When compiling in a C++11 or later source language mode, the compiler would incorrectly fail to report an error for the conversion of a value of enum type to a floating-point type within a list initialization. This has been fixed. The compiler now reports error: non-constant-expression cannot be narrowed from type '<enum_type>' to '<floating-point_type>' in initializer list.

• [SDCOMP-50062]  In certain circumstances, when compiling for T32 state, the compiler could incorrectly assume that the least significant bit of the address A of a function is always zero, and subsequently generate incorrect code for calculations involving A. This has been fixed.

• [SDCOMP-49962]  In certain circumstances, when compiling for AArch32 state, the compiler could generate incorrect code for a call to a function that has a parameter of float16x4_t or float16x8_t type. This has been fixed.

• [SDCOMP-49942]  When compiling in a C++ source language mode, the compiler would mangle symbol names prefixed with $Sub$$ or $Super$$ incorrectly. Subsequently, the linker was unable to use such prefixed symbols to patch an existing symbol, which could result in unexpected run time behavior. This has been fixed.

• [SDCOMP-49695]  When assembling for AArch32 state, the inline assembler and integrated assembler would incorrectly fail to report an error for a VMOV.I32 or VMVN.I32 instruction that specifies an invalid value as the constant operand. Instead, the inline assembler and integrated assembler would generate an instruction that incorrectly has a different but valid value as the constant operand. This has been fixed. The inline assembler and integrated assembler now report error: invalid operand for instruction or error: invalid instruction.

• [SDCOMP-49694]  When compiling code that contains global data which is larger than 16 bytes in size and does not have an alignment requirement of 16 bytes, the compiler would incorrectly overalign the global data to a 16-byte boundary. This has been fixed.

Linker (armlink)

• [SDCOMP-45296]  In certain circumstances, after discarding a section from a COMDAT ELF section group, the linker could incorrectly report Warning: L6776W: The debug frame in .debug_frame(<object>) does not describe an executable section. This has been fixed.

• [SDCOMP-30157]  In certain circumstances, after discarding a section from a COMDAT ELF section group, the linker could incorrectly report Warning: L6439W: Multiply defined Global Symbol <symbol> defined in invalid_group(<object>) rejected in favour of Symbol defined in <section>(<object>). This has been fixed.

• [SDCOMP-29506]  In certain circumstances, when linking an object containing a mergeable strings section that does not end with a terminating null character, the linker would incorrectly fail to report a warning. Instead, the linker would either report ARM Linker: Execution interrupted due to an illegal storage access, or generate an image containing arbitrary data. This could result in a generated image that changes in size and content each time it is linked. This has been fixed. The linker now reports Warning: L6096W: String merge section strings.o(.conststring) is not null-terminated.

Libraries and system headers

• [SDCOMP-49935]  In certain circumstances, the Arm C++ library implementations of the complex number functions std::acos(), std::acosh(), std::asin(), and std::asinh() could return an incorrect result. The result would be incorrectly separated by more than one unit in the last place (ULP) from the exact result. This has been fixed.

• [SDCOMP-49750]  In certain circumstances, instantiations of the Arm C++ library implementation of the std::valarray<T> class template could incorrectly and unexpectedly construct and destruct objects of type T using the destructor or the default constructor of T. This has been fixed.

• [SDCOMP-29067]  In certain circumstances, when compiling in a C++ source language mode, and the code contains a dynamic_cast conversion from a pointer or reference to a polymorphic type A to a pointer or reference to a different polymorphic type B, where B has been derived from A, the compiler could incorrectly fail to return a null pointer value if the public base class of B is ambiguous. This has been fixed.

Other issues

• [SDCOMP-51100]  When Arm Compiler was used on Windows, and the ARMLMD_LICENSE_FILE or ARM_PRODUCT_PATH environment variables contained double quotes, the tools would incorrectly report error: Unable to determine the current toolkit. Check that ARM_TOOL_VARIANT is set correctly. This has been fixed.

• [SDCOMP-50941]  When Arm Compiler was used on Windows 10 version 1803 (April 2018 Update) with an MDK-ARM installation and a Keil Single-User License or Keil Floating-User License, the tools would incorrectly report error: Failed to check out a license.Keil Licensing error: No TOOLS.ini file found. This has been fixed.

• [SDCOMP-50663]  In certain circumstances, the legacy assembler, librarian, linker, and fromelf utility could incorrectly report an error for a valid command-line argument immediately after an empty string ("") command-line argument. This has been fixed. The legacy assembler, librarian, linker and fromelf utility now treat empty string command-line arguments as ordinary arguments.

Known issues in Arm Compiler 6.11

• [SDCOMP-50507]  Compiler support for the AArch64 Armv8.2-A half-precision floating-point intrinsics has the following limitations:

  • The compiler incorrectly reports fatal error: error in backend or error: cannot compile this builtin function yet for the floating-point absolute compare intrinsics and a subset of the floating-point integer conversion intrinsics.
  • The floating-point multiply extended (by element) intrinsics, vmulxh_lane_f16() and vmulxh_laneq_f16(), are not supported.

• [SDCOMP-50470]  The compiler incorrectly fails to report an error for a function call that involves invalid default argument promotion for the _Float16 data type.

Release notes for Arm Compiler 6.10.1

1. Introduction

Arm Compiler 6.10.1 is an update release to Arm Compiler 6.10 and is intended for use:

• In conjunction with DS-5.
• In conjunction with Keil MDK.
• As a standalone toolchain installation.

A suitable license from one of these products must be available. Contact your sales representative or visit https://developer.arm.com/products/buy-arm-products to enquire about a license.

If you are using a floating license, your license server must be running armlmd and lmgrd version 11.14.1.0 or later. Arm recommends that you always use the latest version of the license server software that is available from https://developer.arm.com/products/software-development-tools/license-management/downloads.

1.1 Arm Compiler 6 Configuration

Arm Compiler 6 is the successor to Arm Compiler 5 and includes the components listed below. See the Migration and Compatibility Guide in the product documentation for more information on migrating projects from previous versions.

• armclang
  • armclang is the successor to armcc and is based on LLVM and Clang technology.
• armlink, armasm, fromelf, armar
  • armlink, armasm, fromelf, and armar have been extended to support Armv8 and behave similarly to Arm Compiler 5.
• Arm C and C++ libraries for embedded systems
  • The standard Arm Compiler embedded libraries have been extended to support Armv8 and behave similarly to those found in Arm Compiler 5.
  • Arm Compiler 6 includes the libc++ library as the C++ Standard Template Library.

Note regarding assemblers:

• Arm Compiler 6 adopts the LLVM integrated assembler as default because it aligns more closely with GNU assembler syntax, improving portability between the GNU and Arm Compiler toolchains. The LLVM integrated assembler is called by default by armclang. A side effect is that Arm Compiler 6 will not compile C/C++ source files which contain legacy armcc inline or embedded assembler.
• The legacy assembler (armasm) is not called by default, but is included in Arm Compiler 6 for assembling source code written in armasm syntax. The legacy assembler is included to help with migration of existing projects from Arm Compiler 5 or earlier. Although the legacy assembler has been extended to support Armv8, this assembler does not include all the features of the latest Armv8 updates, and Arm has no plans to add further architectural support to armasm. For new projects, Arm recommends the armclang inline assembler and integrated assembler.

1.2 What's Supported in Arm Compiler 6.10.1?

The following table describes the level of support for the most common toolkits:

Architectures and Processors		Support Level	License / Configuration
			MDK-Lite	MDK-Essential	MDK-Plus	MDK-Professional	DS-5 Professional	DS-5 Ultimate
Armv8-A	Cortex-A76/75/73/72/57/55/53/35/32	Product feature	-	-	-	-	-	Yes
Armv7-A	Cortex-A17/15/12/9/8/7/5	Product feature	-	-	-	Yes	Yes	Yes
Armv8-R	Cortex-R52	Product feature	-	-	-	-	-	Yes
Armv7-R	Cortex-R8/7/5	Product feature	-	-	-	-	Yes	Yes
	Cortex-R4F/4	Product feature	-	-	Yes	Yes	Yes	Yes
Armv8-M	Cortex-M35P/33/23	Product feature	-	Non-secure only	Yes	Yes	Yes	Yes
Armv7-M	SC300	Product feature	-	-	Yes	Yes	Yes	Yes
	Cortex-M7/4/3	Product feature	32 Kbyte code limit	Yes	Yes	Yes	Yes	Yes
Armv6-M	SC000	Product feature	-	-	Yes	Yes	Yes	Yes
	Cortex-M0/M0+	Product feature	32 Kbyte code limit	Yes	Yes	Yes	Yes	Yes
Arm architectures earlier than Armv6-M		Unsupported	-	-	-	-	-	-
Non Arm architectures		Unsupported	-	-	-	-	-	-

Support Level	Description
Product features	Production quality. Highest support priority.
Beta product features	Implementation complete, but not thoroughly tested. User experimentation and feedback is welcomed.
Alpha product features	Implementation is not complete and not thoroughly tested. User experimentation and feedback is welcomed.
Community features	Additional features that are available in the open-source technology Arm Compiler 6 is built on. Arm makes no claims about the quality level or the degree of functionality of these features. User experimentation and feedback is welcomed.
Unsupported features	Features that are either not present in the toolchain or have been deprecated. These features are completely untested. Use entirely at your own risk.

Check with your supplier for information about which architectures and processors are supported by other toolkits.

For more information on the supported features and level of support for features, see the product documentation.

2. Installation Instructions

If you received Arm Compiler 6.10.1 as part of a toolkit, for example DS-5, the toolkit installer takes care of the installation process. Please refer to the installation instructions for the toolkit in such cases.

For all other cases, you must select an appropriate installation location depending on how you intend to use Arm Compiler 6.10.1:

• Integrated into DS-5 5.20 or later.
• Integrated into Keil MDK 5.22 or later.
• As a standalone product.

2.1. Integration into DS-5 5.20 or later

Arm Compiler 6.10.1 can be installed in any location, including the default location, providing this is outside of a DS-5 product installation.

After it is installed, you can integrate the toolchain with DS-5 5.20 or later by following the instructions in the tutorial available at https://developer.arm.com/products/software-development-tools/ds-5-development-studio/resources/tutorials/adding-new-compiler-toolchains-to-ds-5.

Arm recommends using Arm Compiler 6.10.1 from the DS-5 Eclipse IDE or DS-5 Command Prompt. When using the toolchain outside these environments, you might need to configure the following environment variables:

• Set ARM_PRODUCT_PATH to the path to the sw/mappings directory within your DS-5 installation. Please note this path must not contain double quotes on Windows. A path that contains spaces will still work without the quotes.
• Set ARM_TOOL_VARIANT appropriately if you are not using a DS-5 Professional Edition license.

For further information see https://developer.arm.com/products/software-development-tools/license-management/resources/product-and-toolkit-configuration.

2.2. Integration into Keil MDK 5.22 or later

Arm Compiler 6.10.1 must be installed within the ARM subdirectory of the Keil MDK installation directory. For example, if your Keil MDK installation directory is in C:\Keil_v5, the recommended installation path is C:\Keil_v5\ARM\ARMCompiler6.10.1.

After it is installed, you can integrate the toolchain into a Keil MDK project by following the instructions in the tutorial available at http://www.keil.com/support/man/docs/uv4/uv4_armcompilers.htm.

2.3. Use as a standalone product

Arm Compiler 6.10.1 can be installed in any location, including the default location, providing this is outside of a DS-5 product installation.

Ensure that the ARMLMD_LICENSE_FILE environment variable is pointing to your license file or license server. Please note this path must not contain double quotes on Windows. A path that contains spaces will still work without the quotes.

Set ARM_TOOL_VARIANT appropriately if you are not using a DS-5 Professional Edition license.

For further information see https://developer.arm.com/products/software-development-tools/license-management/resources/product-and-toolkit-configuration.

2.4. Installation on Linux

Arm Compiler 6.10.1 has been tested on the following supported platforms:

• Red Hat Enterprise Linux 6 Workstation, 64-bit only.
• Red Hat Enterprise Linux 7 Workstation, 64-bit only.
• Ubuntu Desktop Edition 14.04 LTS, 64-bit only.
• Ubuntu Desktop Edition 16.04 LTS, 64-bit only.

Arm Compiler 6.10.1 is not expected to work on older platforms.

To install Arm Compiler 6.10.1, run (not source) install_x86_64.sh and follow the on-screen instructions. The installer unpacks Arm Compiler 6.10.1 into your chosen directory.

The armclang binary is dynamically linked to a copy of libstdc++ installed under your chosen directory as part of Arm Compiler 6.10.1.

2.5. Installation on Windows

Arm Compiler 6.10.1 has been tested on the following supported platforms:

• Windows Server 2012, 64-bit only.
• Windows 7 Enterprise SP1.
• Windows 7 Professional SP1.
• Windows 8.1 Enterprise, 64-bit only.
• Windows 8.1 Professional, 64-bit only.
• Windows 10 Enterprise, 64-bit only.
• Windows 10 Professional, 64-bit only.

Arm Compiler 6.10.1 is not expected to work on older platforms.

Arm Compiler 6.10.1 has been tested on the latest feature update of Windows 10 available as of June 2018. It is expected to work well on later updates.

To install Arm Compiler 6.10.1, run win-x86_64\setup.exe on a supported 64-bit Windows platform or win-x86_32\setup.exe on a supported 32-bit Windows platform and follow the on-screen instructions. If you have an earlier version of Arm Compiler 6 installed and you wish to perform an upgrade, it is recommended that you uninstall the previous version before installing the new version of Arm Compiler 6.

Arm Compiler 6 is built with Microsoft Visual Studio 2015 and requires the Universal C Runtime in Windows to be installed. For further information see https://support.microsoft.com/en-gb/help/2999226/update-for-universal-c-runtime-in-windows.

3. Uninstall

On Linux, delete the Arm Compiler 6.10.1 installation directory.

On Windows, use Programs and Features in Control Panel, select ARM Compiler 6.10.1, and click the Uninstall button.

4. Documentation

The following documentation is available for Arm Compiler 6.10.1:

• User Guide.
• armar User Guide.
• armasm User Guide.
• armclang Reference Guide.
• armlink User Guide.
• fromelf User Guide.
• Arm C and C++ Libraries and Floating-Point Support User Guide.
• Migration and Compatibility Guide.
• Software Development Guide.
• Errors and Warnings Reference Guide.
• Scalable Vector Extension User Guide.

For more information, please see Arm Compiler 6 documentation in developer.arm.com.

In January 2018, Arm published information about the Variant 1: bounds check bypass (CVE-2017-5753) vulnerability. Refer to https://developer.arm.com/support/security-update/compiler-support-for-mitigations and use the Migration path recommendations with this Arm Compiler release to help mitigate against the vulnerability. Note that the API of the mitigation is subject to change.

5. Feedback and Support

Your feedback is important to us, and you are welcome to send us defect reports and suggestions for improvement on any aspect of the product. Contact your supplier or visit https://support.developer.arm.com and open a case with feedback or support issues. Where appropriate, please provide the --vsn output from the tool, the complete content of any error message that the tools produce, and include any source code, other files, and command-lines necessary to reproduce the issue.

6. Release History and Changes

The following are the releases to date of the Arm Compiler 6.10 series:

• 6.10.1 (released June 2018)
• 6.10 (released March 2018)

Below is a summary of the changes in each release, including new features and defect fixes. Changes are listed since the previous release in each case unless otherwise specified. Each itemized change is accompanied by a unique SDCOMP-<NNNNN> identifier. If you need to contact Arm about a specific issue within these release notes, please quote the appropriate identifier.

Changes in Arm Compiler 6.10.1

General changes in Arm Compiler 6.10.1

• [SDCOMP-51014]  Support has been added for the Cortex-A76 processor. To target Cortex-A76, select from the following options:

  armclang:

  • --target=aarch64-arm-none-eabi -mcpu=cortex-a76 for AArch64 state.
  • --target=arm-arm-none-eabi -mcpu=cortex-a76 for AArch32 state.

  armasm, armlink, and fromelf:

  • --cpu=8.2-A.64 for AArch64 state.
  • --cpu=8.2-A.32 for AArch32 state.

• [SDCOMP-51015]  Support has been added for the Cortex-M35P processor. To target Cortex-M35P, select from the following options:

  armclang:

  • --target=arm-arm-none-eabi -mcpu=cortex-m35p for a variant with DSP and FP.
  • --target=arm-arm-none-eabi -mcpu=cortex-m35p+nodsp for a variant without DSP but with FP.
  • --target=arm-arm-none-eabi -mcpu=cortex-m35p -mfloat-abi=soft for a variant with DSP but without FP.
  • --target=arm-arm-none-eabi -mcpu=cortex-m35p+nodsp -mfloat-abi=soft for a variant without DSP and FP.

  armasm, armlink, and fromelf:

  • --cpu=Cortex-M35P for a variant with DSP and FP.
  • --cpu=Cortex-M35P.no_dsp for a variant without DSP but with FP.
  • --cpu=Cortex-M35P --fpu=SoftVFP for a variant with DSP but without FP.
  • --cpu=Cortex-M35P.no_dsp --fpu=SoftVFP for a variant without DSP and FP.

Defect fixes in Arm Compiler 6.10.1

Other issues

• [SDCOMP-50941]  When Arm Compiler was used on Windows 10 version 1803 (April 2018 Update) with an MDK-ARM installation and a Keil Single-User or Keil Floating license, the tools would incorrectly report error: Failed to check out a license.Keil Licensing error: No TOOLS.ini file found. This has been fixed.

Known issues in Arm Compiler 6.10.1

• [SDCOMP-50507]  Compiler support for the AArch64 Armv8.2-A half-precision floating-point intrinsics has the following limitations:

  • The compiler incorrectly reports fatal error: error in backend or error: cannot compile this builtin function yet for the floating-point absolute compare intrinsics and a subset of the floating-point integer conversion intrinsics.
  • The floating-point multiply extended (by element) intrinsics, vmulxh_lane_f16() and vmulxh_laneq_f16(), are not supported.

• [SDCOMP-50470]  The compiler incorrectly fails to report an error for a function call that involves invalid default argument promotion for the _Float16 data type.

Changes in Arm Compiler 6.10

Changes are listed since the previous feature release, Arm Compiler 6.9.

General changes in Arm Compiler 6.10

• [SDCOMP-50185]  Previously, when assembling for AArch32 state and for a target that supports the A32 and T32 instruction sets, the legacy assembler option --apcs=/nointerwork was selected by default. This behavior has been changed. In these circumstances, --apcs=/interwork is now selected by default.

  For more information about these options, refer to the armasm User Guide.

• [SDCOMP-49700]  In certain circumstances, when a legacy assembler or linker process invoked the compiler as a sub-process to preprocess a file but all suitable licenses were already in use, the processes could deadlock. This has been fixed.

• [SDCOMP-47596]  The default C++ source language mode has changed from gnu++98 to gnu++14.

  gnu++14 language and library features are a [BETA] product feature. Arm recommends compiling with -std=c++11 to restrict Arm Compiler to using only C++11 language and library features, which are fully supported.

  For more information, refer to the -std section of the armclang Reference Guide.

• [SDCOMP-30915]  Previously, functions or variables annotated with __attribute__((used)) could be removed by the unused section elimination feature of the linker, unless they were retained using the --keep=section_id or --no_remove linker options. This behavior has been changed. Functions or variables annotated with __attribute__((used)) are no longer removed by the unused section elimination feature of the linker.

Enhancements in Arm Compiler 6.10

Compiler and integrated assembler (armclang)

• [SDCOMP-49242]  Support has been added to the compiler for targeting the Armv8.1-A atomics architecture extension. This includes generating the CASP instruction for C++11 std::atomic compare and swap operations on 128-bit types.

• [SDCOMP-49008]  Support has been added to the inline assembler, integrated assembler, legacy assembler, and the fromelf utility for the CSDB barrier instruction that can be used to mitigate the Variant 1: bounds check bypass (CVE-2017-5753) vulnerability.

• [SDCOMP-48997]  Support has been added for the -fno-builtin option that can be used to prevent the compiler from optimizing calls to certain standard C library functions, such as printf(). When compiling without -fno-builtin, the compiler can replace such calls with inline code or with calls to other library functions.
  
  For more information about this option, refer to the armclang Reference Guide.

• [SDCOMP-45236]  Support has been added for the _Float16 data type in all C and C++ source language modes.

  The _Float16 data type is defined in ISO/IEC TS 18661.

Defect fixes in Arm Compiler 6.10

Compiler and integrated assembler (armclang)

• [SDCOMP-50315]  In certain circumstances, when compiling C++ code containing a lambda with a parameter of array type, the compiler could incorrectly report Segmentation fault (core dumped). This has been fixed.

• [SDCOMP-50226]  When compiling with -S or --save-temps, and the program contains an inline assembly statement with a .type directive that specifies a %notype or STT_NOTYPE type, the assembly output incorrectly contained an invalid type. This has been fixed.

• [SDCOMP-49972]  In certain circumstances, when compiling for an Armv7-M target or an Armv8-M target without the DSP Extension, the compiler could generate code that incorrectly contained the architecturally UNDEFINED instruction SMMLS. This has been fixed.

• [SDCOMP-49927]  When assembling for AArch32 state, the inline assembler and integrated assembler incorrectly failed to report an error for a memory access instruction that specifies an invalid shift operator for the shifted second register operand. Instead, the inline assembler and integrated assembler incorrectly ignored the instruction. This has been fixed. The inline assembler and integrated assembler now report error: illegal shift operator.

• [SDCOMP-49843]  When assembling for a big-endian target and T32 state, and the program contains an inst.w directive that specifies an instruction I, the inline assembler and integrated assembler would encode I with the bytes in an incorrect order. This has been fixed.

• [SDCOMP-49817]  In certain circumstances, when compiling at -O3, -Ofast, or -Omax, the compiler could generate code that incorrectly fails to pass pointer arguments to a function that is annotated with __attribute__((naked)). This has been fixed.

• [SDCOMP-49718]  When compiling for an Armv6-M target or an Armv8-M target without the Main Extension, and the code contains a function that has a parameter P and a local variable that has an alignment requirement of more than 8 bytes, where P is a variable argument list or must be passed using the stack, the compiler would generate code that accessed P incorrectly. This has been fixed.

• [SDCOMP-49707]  In certain circumstances, when compiling at -Oz, and the program contains functions that throw and catch C++ exceptions, the compiler could generate an incorrect ARM_EXIDX table entry for a function F that catches C++ exceptions. Subsequently, when a C++ exception was thrown, the variables in F could be corrupted. This has been fixed.

• [SDCOMP-49698]  When compiling in a C++ source language mode, the compiler incorrectly failed to report an error for an invalid declaration specifier in a template non-type parameter. Instead, it incorrectly ignored the invalid declaration specifier. This has been fixed. The compiler now reports error: invalid declaration specifier in template non-type parameter.

• [SDCOMP-49645]  In certain circumstances, when compiling for a target that supports the Advanced SIMD Extension (NEON) and AArch32 state, and the program contains a comparison operation that is promoted to an integral type, the compiler could generate incorrect code. This has been fixed.

• [SDCOMP-49456]  In certain circumstances, when compiling at any optimization level except -O0 and -O1 for a big-endian target and AArch32 state, the compiler could incorrectly reorder memory accesses. This has been fixed.

• [SDCOMP-49246]  In certain circumstances, when compiling for AArch32 state with -mfloat-abi=soft and without -mfpu=none, the compiler could generate an object that incorrectly contained the build attribute Tag_FP_arch. Subsequently, the linker could incorrectly select implementations of library functions containing hardware floating-point instructions. This has been fixed.

• [SDCOMP-49132]  In certain circumstances, when compiling for an Armv8-M target without the Main Extension, and the code contains a call through a function pointer, the compiler could incorrectly report fatal error: error in backend: ran out of registers during register allocation. This has been fixed.

• [SDCOMP-48979]  In certain circumstances, when compiling at any optimization level except -O0 and -O1 for a big-endian target and AArch64 state, the compiler could generate incorrect code for vectorizable memory access operations. This has been fixed.

• [SDCOMP-48864]  In certain circumstances, when assembling for an Armv6-M, Armv7, or Armv8-M target and T32 state, the inline assembler and integrated assembler could incorrectly fail to report an error for a VMRS or VMSR instruction that specifies SP as the general-purpose register operand. Instead, the inline assembler and integrated assembler could generate a VMRS or VMSR instruction with the UNPREDICTABLE general-purpose register encoding value 13. This has been fixed. The inline assembler and integrated assembler now report error: invalid operand for instruction.

• [SDCOMP-48429]  When assembling for AArch32 state, the inline assembler and integrated assembler would incorrectly fail to report an error for an instruction that specifies an invalid register for the base register operand or the optionally shifted second register operand. Instead, the inline assembler and integrated assembler would generate an instruction that incorrectly has a different but valid general-purpose register as the specified operand. This has been fixed. The inline assembler and integrated assembler now report error: invalid operand for instruction.

• [SDCOMP-47542]  In certain circumstances, when compiling with -g or -gdwarf-version, the compiler could generate incorrect debug information for enumerator values. This has been fixed.

• [SDCOMP-46968]  In certain circumstances, when compiling a static variable that is annotated with __attribute__((used)), the compiler could generate code that incorrectly used global linkage instead of local linkage for the variable. This has been fixed.

• [SDCOMP-25212]  When compiling C or C++ language source code, the compiler is required to define the __GNUC__ preprocessor macro, but incorrectly failed to do so for source language modes that do not support additional GNU extensions. This has been fixed.

Legacy assembler (armasm)

• [SDCOMP-49312]  When assembling a source file that contains a data block B followed by an exported symbol S that is associated with an instruction I, and padding must be added after B to ensure that I is correctly aligned, the assembler would associate S with an incorrect address. This has been fixed.

Libraries and system headers

• [SDCOMP-49975]  In certain circumstances, the Arm Compiler library implementations of the following functions could incorrectly fail to set errno to ERANGE when the return value underflows to zero:

  • atan2().
  • atan2f().
  • erfc().

  This has been fixed.

• [SDCOMP-49764]  The microlib implementation of the strrchr() function incorrectly returned a null pointer when used to locate the terminating null character '\0' of a string. This has been fixed.

• [SDCOMP-49759]  The default constructor of the Arm C++ library implementation of the class template std::basic_stringbuf incorrectly failed to initialize all its sequence pointers to null pointers. This has been fixed.

• [SDCOMP-49721]  When compiling code that contains an == or != operator that involves std::istream_iterator, and the operator is referenced using its qualified name, the compiler would incorrectly report error: no matching function for call to 'operator<operator>'. This has been fixed.

• [SDCOMP-49323]  The arm_compat.h system header incorrectly failed to include the arm_acle.h system header when the macro __ARM_ACLE was defined. This has been fixed.

Fromelf

• [SDCOMP-49646]  When processing an ELF file with --elf --localize=option[,option,…], the fromelf utility would incorrectly localize external references identified by the arguments to --localize, resulting in an output object file that could result in the linker reporting Error: L6283E: Object <objname> contains illegal local reference to symbol <symbolname>. This has been fixed. The --localize option now only localizes definitions, and not references.

• [SDCOMP-49582]  In certain circumstances, when disassembling an ELF file that contains an instruction modified by a relocation, the fromelf utility could produce disassembly that incorrectly failed to indicate the presence of the relocation. This has been fixed.

• [SDCOMP-49313]  When disassembling an ELF file that contains a PC-relative load instruction modified by a relocation that has an addend, the fromelf utility would produce disassembly that contained an incorrect load value for the instruction. This has been fixed.

• [SDCOMP-48274]  In certain circumstances, when disassembling an ELF file that contains an A64 load or store instruction modified by a relocation, the fromelf utility could produce disassembly that incorrectly contained AArch32 registers instead of AArch64 registers. This has been fixed.

• [SDCOMP-30908]  In certain circumstances, when processing an AArch64 ELF file with --text -g, the fromelf utility would report offsets that were incorrectly 8 bytes greater than their actual values for entries in the .debug_loc section. This has been fixed.

Known issues in Arm Compiler 6.10

• [SDCOMP-50941]  When Arm Compiler is used on Windows 10 version 1803 (April 2018 Update) with an MDK-ARM installation and a Keil Single-User or Keil Floating license, the tools would incorrectly report error: Failed to check out a license.Keil Licensing error: No TOOLS.ini file found. For further information, see http://www.keil.com/support/docs/4043.htm.

• [SDCOMP-50507]  Compiler support for the AArch64 Armv8.2-A half-precision floating-point intrinsics has the following limitations:

  • The compiler incorrectly reports fatal error: error in backend or error: cannot compile this builtin function yet for the floating-point absolute compare intrinsics and a subset of the floating-point integer conversion intrinsics.
  • The floating-point multiply extended (by element) intrinsics, vmulxh_lane_f16() and vmulxh_laneq_f16(), are not supported.

• [SDCOMP-50470]  The compiler incorrectly fails to report an error for a function call that involves invalid default argument promotion for the _Float16 data type.

Changes in Future Releases

• The legacy R-type dynamic linking model, which does not conform to the 32-bit Application Binary Interface for the Arm Architecture, will be deprecated in a future release of Arm Compiler 6.

  The following will be deprecated:

  • Linking with --reloc.
  • Linking without --base_platform with a scatter file that contains the RELOC load region attribute.

  Use the Base Platform and Base Platform Application Binary Interface (BPABI) linking models to generate relocatable executable files instead of using the legacy R-type dynamic linking model.

  For information about the Base Platform and BPABI linking models, refer to the armlink User Guide.

• Support for ELF sections that contain the legacy SHF_COMDEF ELF section flag will be deprecated in a future release of Arm Compiler 6.

  The following will be deprecated:

  • The COMDEF section attribute of the legacy armasm syntax AREA directive.
  • Linking with legacy objects that contain ELF sections with the legacy SHF_COMDEF ELF section flag.

  Use the GRP_COMDAT ELF section flag instead of the legacy SHF_COMDEF ELF section flag by:

  • Replacing the COMDEF section attribute of the legacy armasm syntax AREA directive with the COMGROUP=symbol_name section attribute.
  • Rebuilding incompatible legacy objects that were built using Arm Compiler 5 or earlier using one of the following:
    • The same version of the compiler as before, but with --dwarf3.
    • Arm Compiler 6.

  For more information about the COMGROUP=symbol_name section attribute of the legacy armasm syntax AREA directive, refer to the armasm User Guide.

  For more information about COMDAT groups and the GRP_COMDAT ELF section flag, refer to the Itanium C++ ABI.

• __declspec will be deprecated in a future release of Arm Compiler 6.

  For information on replacing uses of __declspec, refer to the Migration and Compatibility Guide.

Release notes for Arm Compiler 6.10

1. Introduction

Arm Compiler 6.10 adds:

• armclang support for targeting the Armv8.2-A FP16 architecture extension.
• armclang support for targeting the Armv8.1-A atomics architecture extension.

Arm Compiler 6.10 is intended for use:

• In conjunction with DS-5.
• In conjunction with Keil MDK.
• As a standalone toolchain installation.

A suitable license from one of these products must be available. Contact your sales representative or visit https://developer.arm.com/products/buy-arm-products to enquire about a license.

If you are using a floating license, your license server must be running armlmd and lmgrd version 11.14.1.0 or later. Arm recommends that you always use the latest version of the license server software that is available from https://developer.arm.com/products/software-development-tools/license-management/downloads.

1.1 Arm Compiler 6 Configuration

Arm Compiler 6 is the successor to Arm Compiler 5 and includes the components listed below. See the Migration and Compatibility Guide in the product documentation for more information on migrating projects from previous versions.

• armclang
  • armclang is the successor to armcc and is based on LLVM and Clang technology.
• armlink, armasm, fromelf, armar
  • armlink, armasm, fromelf, and armar have been extended to support Armv8 and behave similarly to Arm Compiler 5.
• Arm C and C++ libraries for embedded systems
  • The standard Arm Compiler embedded libraries have been extended to support Armv8 and behave similarly to those found in Arm Compiler 5.
  • Arm Compiler 6 includes the libc++ library as the C++ Standard Template Library.

Note regarding assemblers:

• Arm Compiler 6 adopts the LLVM integrated assembler as default because it aligns more closely with GNU assembler syntax, improving portability between the GNU and Arm Compiler toolchains. The LLVM integrated assembler is called by default by armclang. A side effect is that Arm Compiler 6 will not compile C/C++ source files which contain legacy armcc inline or embedded assembler.
• The legacy assembler (armasm) is not called by default, but is included in Arm Compiler 6 for assembling source code written in armasm syntax. The legacy assembler is included to help with migration of existing projects from Arm Compiler 5 or earlier. Although the legacy assembler has been extended to support Armv8, this assembler does not include all the features of the latest Armv8 updates, and Arm has no plans to add further architectural support to armasm. For new projects, Arm recommends the armclang inline assembler and integrated assembler.

1.2 What's Supported in Arm Compiler 6.10?

The following table describes the level of support for the most common toolkits:

Architectures and Processors		Support Level	License / Configuration
			MDK-Lite	MDK-Essential	MDK-Plus	MDK-Professional	DS-5 Professional	DS-5 Ultimate
Armv8-A	Cortex-A75/73/72/57/55/53/35/32	Product feature	-	-	-	-	-	Yes
Armv7-A	Cortex-A17/15/12/9/8/7/5	Product feature	-	-	-	Yes	Yes	Yes
Armv8-R	Cortex-R52	Product feature	-	-	-	-	-	Yes
Armv7-R	Cortex-R8/7/5	Product feature	-	-	-	-	Yes	Yes
	Cortex-R4F/4	Product feature	-	-	Yes	Yes	Yes	Yes
Armv8-M	Cortex-M33/23	Product feature	-	Non-secure only	Yes	Yes	Yes	Yes
Armv7-M	SC300	Product feature	-	-	Yes	Yes	Yes	Yes
	Cortex-M7/4/3	Product feature	32 Kbyte code limit	Yes	Yes	Yes	Yes	Yes
Armv6-M	SC000	Product feature	-	-	Yes	Yes	Yes	Yes
	Cortex-M0/M0+	Product feature	32 Kbyte code limit	Yes	Yes	Yes	Yes	Yes
Arm architectures earlier than Armv6-M		Unsupported	-	-	-	-	-	-
Non Arm architectures		Unsupported	-	-	-	-	-	-

Support Level	Description
Product features	Production quality. Highest support priority.
Beta product features	Implementation complete, but not thoroughly tested. User experimentation and feedback is welcomed.
Alpha product features	Implementation is not complete and not thoroughly tested. User experimentation and feedback is welcomed.
Community features	Additional features that are available in the open-source technology Arm Compiler 6 is built on. Arm makes no claims about the quality level or the degree of functionality of these features. User experimentation and feedback is welcomed.
Unsupported features	Features that are either not present in the toolchain or have been deprecated. These features are completely untested. Use entirely at your own risk.

Check with your supplier for information about which architectures and processors are supported by other toolkits.

For more information on the supported features and level of support for features, see the product documentation.

2. Installation Instructions

If you received Arm Compiler 6.10 as part of a toolkit, for example DS-5, the toolkit installer takes care of the installation process. Please refer to the installation instructions for the toolkit in such cases.

For all other cases, you must select an appropriate installation location depending on how you intend to use Arm Compiler 6.10:

• Integrated into DS-5 5.20 or later.
• Integrated into Keil MDK 5.22 or later.
• As a standalone product.

2.1. Integration into DS-5 5.20 or later

Arm Compiler 6.10 can be installed in any location, including the default location, providing this is outside of a DS-5 product installation.

After it is installed, you can integrate the toolchain with DS-5 5.20 or later by following the instructions in the tutorial available at https://developer.arm.com/products/software-development-tools/ds-5-development-studio/resources/tutorials/adding-new-compiler-toolchains-to-ds-5.

Arm recommends using Arm Compiler 6.10 from the DS-5 Eclipse IDE or DS-5 Command Prompt. When using the toolchain outside these environments, you might need to configure the following environment variables:

• Set ARM_PRODUCT_PATH to the path to the sw/mappings directory within your DS-5 installation. Please note this path must not contain double quotes on Windows. A path that contains spaces will still work without the quotes.
• Set ARM_TOOL_VARIANT appropriately if you are not using a DS-5 Professional Edition license.

For further information see https://developer.arm.com/products/software-development-tools/license-management/resources/product-and-toolkit-configuration.

2.2. Integration into Keil MDK 5.22 or later

Arm Compiler 6.10 must be installed within the ARM subdirectory of the Keil MDK installation directory. For example, if your Keil MDK installation directory is in C:\Keil_v5, the recommended installation path is C:\Keil_v5\ARM\ARMCompiler6.10.

After it is installed, you can integrate the toolchain into a Keil MDK project by following the instructions in the tutorial available at http://www.keil.com/support/man/docs/uv4/uv4_armcompilers.htm.

2.3. Use as a standalone product

Arm Compiler 6.10 can be installed in any location, including the default location, providing this is outside of a DS-5 product installation.

Ensure that the ARMLMD_LICENSE_FILE environment variable is pointing to your license file or license server. Please note this path must not contain double quotes on Windows. A path that contains spaces will still work without the quotes.

Set ARM_TOOL_VARIANT appropriately if you are not using a DS-5 Professional Edition license.

For further information see https://developer.arm.com/products/software-development-tools/license-management/resources/product-and-toolkit-configuration.

2.4. Installation on Linux

Arm Compiler 6.10 has been tested on the following supported platforms:

• Red Hat Enterprise Linux 6 Workstation, 64-bit only.
• Red Hat Enterprise Linux 7 Workstation, 64-bit only.
• Ubuntu Desktop Edition 14.04 LTS, 64-bit only.
• Ubuntu Desktop Edition 16.04 LTS, 64-bit only.

Arm Compiler 6.10 is not expected to work on older platforms.

To install Arm Compiler 6.10, run (not source) install_x86_64.sh and follow the on-screen instructions. The installer unpacks Arm Compiler 6.10 into your chosen directory.

The armclang binary is dynamically linked to a copy of libstdc++ installed under your chosen directory as part of Arm Compiler 6.10.

2.5. Installation on Windows

Arm Compiler 6.10 has been tested on the following supported platforms:

• Windows Server 2012, 64-bit only.
• Windows 7 Enterprise SP1.
• Windows 7 Professional SP1.
• Windows 8.1 Enterprise, 64-bit only.
• Windows 8.1 Professional, 64-bit only.
• Windows 10 Enterprise, 64-bit only.
• Windows 10 Professional, 64-bit only.

Arm Compiler 6.10 is not expected to work on older platforms.

Arm Compiler 6.10 has been tested on the latest feature update of Windows 10 available as of March 2018. It is expected to work well on later updates.

To install Arm Compiler 6.10, run win-x86_64\setup.exe on a supported 64-bit Windows platform or win-x86_32\setup.exe on a supported 32-bit Windows platform and follow the on-screen instructions. If you have an earlier version of Arm Compiler 6 installed and you wish to perform an upgrade, it is recommended that you uninstall the previous version before installing the new version of Arm Compiler 6.

Arm Compiler 6 is built with Microsoft Visual Studio 2015 and requires the Universal C Runtime in Windows to be installed. For further information see https://support.microsoft.com/en-gb/help/2999226/update-for-universal-c-runtime-in-windows.

3. Uninstall

On Linux, delete the Arm Compiler 6.10 installation directory.

On Windows, use Programs and Features in Control Panel, select ARM Compiler 6.10, and click the Uninstall button.

4. Documentation

The following documentation is available for Arm Compiler 6.10:

• User Guide.
• armar User Guide.
• armasm User Guide.
• armclang Reference Guide.
• armlink User Guide.
• fromelf User Guide.
• Arm C and C++ Libraries and Floating-Point Support User Guide.
• Migration and Compatibility Guide.
• Software Development Guide.
• Errors and Warnings Reference Guide.
• Scalable Vector Extension User Guide.

For more information, please see Arm Compiler 6 documentation in developer.arm.com.

In January 2018, Arm published information about the Variant 1: bounds check bypass (CVE-2017-5753) vulnerability. Refer to https://developer.arm.com/support/security-update/compiler-support-for-mitigations and use the Migration path recommendations with this Arm Compiler release to help mitigate against the vulnerability. Note that the API of the mitigation is subject to change.

5. Feedback and Support

Your feedback is important to us, and you are welcome to send us defect reports and suggestions for improvement on any aspect of the product. Contact your supplier or visit https://support.developer.arm.com and open a case with feedback or support issues. Where appropriate, please provide the --vsn output from the tool, the complete content of any error message that the tools produce, and include any source code, other files, and command-lines necessary to reproduce the issue.

6. Release History and Changes

The following are the releases to date of the Arm Compiler 6.10 series:

• 6.10 (released March 2018)

Below is a summary of the changes in each release, including new features and defect fixes. Changes are listed since the previous release in each case unless otherwise specified. Each itemized change is accompanied by a unique SDCOMP-<NNNNN> identifier. If you need to contact Arm about a specific issue within these release notes, please quote the appropriate identifier.

Changes in Arm Compiler 6.10

Changes are listed since the previous feature release, Arm Compiler 6.9.

General changes in Arm Compiler 6.10

• [SDCOMP-50185]  Previously, when assembling for AArch32 state and for a target that supports the A32 and T32 instruction sets, the legacy assembler option --apcs=/nointerwork was selected by default. This behavior has been changed. In these circumstances, --apcs=/interwork is now selected by default.

  For more information about these options, refer to the armasm User Guide.

• [SDCOMP-49700]  In certain circumstances, when a legacy assembler or linker process invoked the compiler as a sub-process to preprocess a file but all suitable licenses were already in use, the processes could deadlock. This has been fixed.

• [SDCOMP-47596]  The default C++ source language mode has changed from gnu++98 to gnu++14.

  gnu++14 language and library features are a [BETA] product feature. Arm recommends compiling with -std=c++11 to restrict Arm Compiler to using only C++11 language and library features, which are fully supported.

  For more information, refer to the -std section of the armclang Reference Guide.

• [SDCOMP-30915]  Previously, functions or variables annotated with __attribute__((used)) could be removed by the unused section elimination feature of the linker, unless they were retained using the --keep=section_id or --no_remove linker options. This behavior has been changed. Functions or variables annotated with __attribute__((used)) are no longer removed by the unused section elimination feature of the linker.

Enhancements in Arm Compiler 6.10

Compiler and integrated assembler (armclang)

• [SDCOMP-49242]  Support has been added to the compiler for targeting the Armv8.1-A atomics architecture extension. This includes generating the CASP instruction for C++11 std::atomic compare and swap operations on 128-bit types.

• [SDCOMP-49008]  Support has been added to the inline assembler, integrated assembler, legacy assembler, and the fromelf utility for the CSDB barrier instruction that can be used to mitigate the Variant 1: bounds check bypass (CVE-2017-5753) vulnerability.

• [SDCOMP-48997]  Support has been added for the -fno-builtin option that can be used to prevent the compiler from optimizing calls to certain standard C library functions, such as printf(). When compiling without -fno-builtin, the compiler can replace such calls with inline code or with calls to other library functions.
  
  For more information about this option, refer to the armclang Reference Guide.

• [SDCOMP-45236]  Support has been added for the _Float16 data type in all C and C++ source language modes.

  The _Float16 data type is defined in ISO/IEC TS 18661.

Defect fixes in Arm Compiler 6.10

Compiler and integrated assembler (armclang)

• [SDCOMP-50315]  In certain circumstances, when compiling C++ code containing a lambda with a parameter of array type, the compiler could incorrectly report Segmentation fault (core dumped). This has been fixed.

• [SDCOMP-50226]  When compiling with -S or --save-temps, and the program contains an inline assembly statement with a .type directive that specifies a %notype or STT_NOTYPE type, the assembly output incorrectly contained an invalid type. This has been fixed.

• [SDCOMP-49972]  In certain circumstances, when compiling for an Armv7-M target or an Armv8-M target without the DSP Extension, the compiler could generate code that incorrectly contained the architecturally UNDEFINED instruction SMMLS. This has been fixed.

• [SDCOMP-49927]  When assembling for AArch32 state, the inline assembler and integrated assembler incorrectly failed to report an error for a memory access instruction that specifies an invalid shift operator for the shifted second register operand. Instead, the inline assembler and integrated assembler incorrectly ignored the instruction. This has been fixed. The inline assembler and integrated assembler now report error: illegal shift operator.

• [SDCOMP-49843]  When assembling for a big-endian target and T32 state, and the program contains an inst.w directive that specifies an instruction I, the inline assembler and integrated assembler would encode I with the bytes in an incorrect order. This has been fixed.

• [SDCOMP-49817]  In certain circumstances, when compiling at -O3, -Ofast, or -Omax, the compiler could generate code that incorrectly fails to pass pointer arguments to a function that is annotated with __attribute__((naked)). This has been fixed.

• [SDCOMP-49718]  When compiling for an Armv6-M target or an Armv8-M target without the Main Extension, and the code contains a function that has a parameter P and a local variable that has an alignment requirement of more than 8 bytes, where P is a variable argument list or must be passed using the stack, the compiler would generate code that accessed P incorrectly. This has been fixed.

• [SDCOMP-49707]  In certain circumstances, when compiling at -Oz, and the program contains functions that throw and catch C++ exceptions, the compiler could generate an incorrect ARM_EXIDX table entry for a function F that catches C++ exceptions. Subsequently, when a C++ exception was thrown, the variables in F could be corrupted. This has been fixed.

• [SDCOMP-49698]  When compiling in a C++ source language mode, the compiler incorrectly failed to report an error for an invalid declaration specifier in a template non-type parameter. Instead, it incorrectly ignored the invalid declaration specifier. This has been fixed. The compiler now reports error: invalid declaration specifier in template non-type parameter.

• [SDCOMP-49645]  In certain circumstances, when compiling for a target that supports the Advanced SIMD Extension (NEON) and AArch32 state, and the program contains a comparison operation that is promoted to an integral type, the compiler could generate incorrect code. This has been fixed.

• [SDCOMP-49456]  In certain circumstances, when compiling at any optimization level except -O0 and -O1 for a big-endian target and AArch32 state, the compiler could incorrectly reorder memory accesses. This has been fixed.

• [SDCOMP-49246]  In certain circumstances, when compiling for AArch32 state with -mfloat-abi=soft and without -mfpu=none, the compiler could generate an object that incorrectly contained the build attribute Tag_FP_arch. Subsequently, the linker could incorrectly select implementations of library functions containing hardware floating-point instructions. This has been fixed.

• [SDCOMP-49132]  In certain circumstances, when compiling for an Armv8-M target without the Main Extension, and the code contains a call through a function pointer, the compiler could incorrectly report fatal error: error in backend: ran out of registers during register allocation. This has been fixed.

• [SDCOMP-48979]  In certain circumstances, when compiling at any optimization level except -O0 and -O1 for a big-endian target and AArch64 state, the compiler could generate incorrect code for vectorizable memory access operations. This has been fixed.

• [SDCOMP-48864]  In certain circumstances, when assembling for an Armv6-M, Armv7, or Armv8-M target and T32 state, the inline assembler and integrated assembler could incorrectly fail to report an error for a VMRS or VMSR instruction that specifies SP as the general-purpose register operand. Instead, the inline assembler and integrated assembler could generate a VMRS or VMSR instruction with the UNPREDICTABLE general-purpose register encoding value 13. This has been fixed. The inline assembler and integrated assembler now report error: invalid operand for instruction.

• [SDCOMP-48429]  When assembling for AArch32 state, the inline assembler and integrated assembler would incorrectly fail to report an error for an instruction that specifies an invalid register for the base register operand or the optionally shifted second register operand. Instead, the inline assembler and integrated assembler would generate an instruction that incorrectly has a different but valid general-purpose register as the specified operand. This has been fixed. The inline assembler and integrated assembler now report error: invalid operand for instruction.

• [SDCOMP-47542]  In certain circumstances, when compiling with -g or -gdwarf-version, the compiler could generate incorrect debug information for enumerator values. This has been fixed.

• [SDCOMP-46968]  In certain circumstances, when compiling a static variable that is annotated with __attribute__((used)), the compiler could generate code that incorrectly used global linkage instead of local linkage for the variable. This has been fixed.

• [SDCOMP-25212]  When compiling C or C++ language source code, the compiler is required to define the __GNUC__ preprocessor macro, but incorrectly failed to do so for source language modes that do not support additional GNU extensions. This has been fixed.

Legacy assembler (armasm)

• [SDCOMP-49312]  When assembling a source file that contains a data block B followed by an exported symbol S that is associated with an instruction I, and padding must be added after B to ensure that I is correctly aligned, the assembler would associate S with an incorrect address. This has been fixed.

Libraries and system headers

• [SDCOMP-49975]  In certain circumstances, the Arm Compiler library implementations of the following functions could incorrectly fail to set errno to ERANGE when the return value underflows to zero:

  • atan2().
  • atan2f().
  • erfc().

  This has been fixed.

• [SDCOMP-49764]  The microlib implementation of the strrchr() function incorrectly returned a null pointer when used to locate the terminating null character '\0' of a string. This has been fixed.

• [SDCOMP-49759]  The default constructor of the Arm C++ library implementation of the class template std::basic_stringbuf incorrectly failed to initialize all its sequence pointers to null pointers. This has been fixed.

• [SDCOMP-49721]  When compiling code that contains an == or != operator that involves std::istream_iterator, and the operator is referenced using its qualified name, the compiler would incorrectly report error: no matching function for call to 'operator<operator>'. This has been fixed.

• [SDCOMP-49323]  The arm_compat.h system header incorrectly failed to include the arm_acle.h system header when the macro __ARM_ACLE was defined. This has been fixed.

Fromelf

• [SDCOMP-49646]  When processing an ELF file with --elf --localize=option[,option,…], the fromelf utility would incorrectly localize external references identified by the arguments to --localize, resulting in an output object file that could result in the linker reporting Error: L6283E: Object <objname> contains illegal local reference to symbol <symbolname>. This has been fixed. The --localize option now only localizes definitions, and not references.

• [SDCOMP-49582]  In certain circumstances, when disassembling an ELF file that contains an instruction modified by a relocation, the fromelf utility could produce disassembly that incorrectly failed to indicate the presence of the relocation. This has been fixed.

• [SDCOMP-49313]  When disassembling an ELF file that contains a PC-relative load instruction modified by a relocation that has an addend, the fromelf utility would produce disassembly that contained an incorrect load value for the instruction. This has been fixed.

• [SDCOMP-48274]  In certain circumstances, when disassembling an ELF file that contains an A64 load or store instruction modified by a relocation, the fromelf utility could produce disassembly that incorrectly contained AArch32 registers instead of AArch64 registers. This has been fixed.

• [SDCOMP-30908]  In certain circumstances, when processing an AArch64 ELF file with --text -g, the fromelf utility would report offsets that were incorrectly 8 bytes greater than their actual values for entries in the .debug_loc section. This has been fixed.

Known issues in Arm Compiler 6.10

• [SDCOMP-50941]  When Arm Compiler is used on Windows 10 version 1803 (April 2018 Update) with an MDK-ARM installation and a Keil Single-User or Keil Floating license, the tools would incorrectly report error: Failed to check out a license.Keil Licensing error: No TOOLS.ini file found. For further information, see http://www.keil.com/support/docs/4043.htm.

• [SDCOMP-50507]  Compiler support for the AArch64 Armv8.2-A half-precision floating-point intrinsics has the following limitations:

  • The compiler incorrectly reports fatal error: error in backend or error: cannot compile this builtin function yet for the floating-point absolute compare intrinsics and a subset of the floating-point integer conversion intrinsics.
  • The floating-point multiply extended (by element) intrinsics, vmulxh_lane_f16() and vmulxh_laneq_f16(), are not supported.

• [SDCOMP-50470]  The compiler incorrectly fails to report an error for a function call that involves invalid default argument promotion for the _Float16 data type.

Changes in Future Releases

• The legacy R-type dynamic linking model, which does not conform to the 32-bit Application Binary Interface for the Arm Architecture, will be deprecated in a future release of Arm Compiler 6.

  The following will be deprecated:

  • Linking with --reloc.
  • Linking without --base_platform with a scatter file that contains the RELOC load region attribute.

  Use the Base Platform and Base Platform Application Binary Interface (BPABI) linking models to generate relocatable executable files instead of using the legacy R-type dynamic linking model.

  For information about the Base Platform and BPABI linking models, refer to the armlink User Guide.

• Support for ELF sections that contain the legacy SHF_COMDEF ELF section flag will be deprecated in a future release of Arm Compiler 6.

  The following will be deprecated:

  • The COMDEF section attribute of the legacy armasm syntax AREA directive.
  • Linking with legacy objects that contain ELF sections with the legacy SHF_COMDEF ELF section flag.

  Use the GRP_COMDAT ELF section flag instead of the legacy SHF_COMDEF ELF section flag by:

  • Replacing the COMDEF section attribute of the legacy armasm syntax AREA directive with the COMGROUP=symbol_name section attribute.
  • Rebuilding incompatible legacy objects that were built using Arm Compiler 5 or earlier using one of the following:
    • The same version of the compiler as before, but with --dwarf3.
    • Arm Compiler 6.

  For more information about the COMGROUP=symbol_name section attribute of the legacy armasm syntax AREA directive, refer to the armasm User Guide.

  For more information about COMDAT groups and the GRP_COMDAT ELF section flag, refer to the Itanium C++ ABI.

• __declspec will be deprecated in a future release of Arm Compiler 6.

  For information on replacing uses of __declspec, refer to the Migration and Compatibility Guide.