Intel® C++ Compiler 18.0 Developer Guide and Reference
Lets you specify the target architecture to use when offloading code. This option only applies when targeting Intel® Xeon Phi™ products and Intel® Graphics Technology. For Intel® Graphics Technology, you can also specify a virtual ISA (vISA).
Only available when targeting Intel® Xeon Phi™ products, on Intel® 64 architecture targeting Intel® Graphics Technology, or when targeting Intel® Graphics Technology running on IA-32 architecture
Linux: | -qoffload-arch=arch[:visa] |
macOS: | None |
Windows: | /Qoffload-arch:arch[:visa] |
arch |
Specifies the target architecture. On Intel® 64 architecture targeting Intel® MIC Architecture, arch can be:
On IA-32 and Intel® 64 architectures targeting Intel® Graphics Technology, arch can be haswell, broadwell, or skylake. |
visa |
Specifies the virtual ISA (vISA) version to be used. This optional argument is only allowed on IA-32 and Intel® 64 architectures targeting Intel® Graphics Technology. You can only specify visa3.1 or visa3.2. |
varies |
On Intel® 64 architecture targeting Intel® MIC Architecture, the default arch is mic-avx512. On IA-32 and Intel® 64 architectures targeting Intel® Graphics Technology, the default arch is haswell and the default visa is visa3.1 |
This option lets you specify the target architecture to use when offloading code. For Intel® Graphics Technology, you can also specify a virtual ISA (vISA).
On IA-32 and Intel® 64 architectures targeting Intel® Graphics Technology, the compiler will check that the values specified for arch and visa are compatible. For architectures where multiple vISA versions are supported, the vISA version selected must be <= the vISA version supported by the driver installed on the system where the application is to be executed.
The following table shows the vISA version that each arch supports:
arch value |
Supported vISA |
---|---|
skylake |
vISA 3.2 |
broadwell |
vISA 3.1, vISA 3.2 |
haswell |
vISA 3.1 |
The following are various combinations and settings of command line options for offloading:
Offload: Intel® 64 architecture host and Intel® Xeon Phi™ product family x200 (formerly code name Knights Landing) target code generation:
icc … -qoffload-arch=mic-avx512 ! Linux icl … /Qoffload-arch=mic-avx512 ! Windows
Offload : Intel® Xeon Phi™ product family x200 (formerly code name Knights Landing) host and target code generation (for this host, target defaults to Intel® Xeon Phi™ product family x200 (formerly code name Knights Landing)):
icc … -xmic-avx512 ! Linux icl … /Qxmic-avx512 ! Windows
Native : Code Generation for Intel® Xeon Phi™ product family x200 (formerly code name Knights Landing):
icc … -xmic-avx512 –qno-offload ! Linux icl … /Qxmic-avx512 /Qno-offload ! Windows
Visual Studio: Code Generation [Intel C++] > Offload Target Architecture
Eclipse: Code Generation > Offload Target Architecture
Xcode: None
None