Remove AMX-TF32 support from GCC
AMX-TF32 instruction set support removed as it's deprecated and lacks hardware.
Support for the AMX-TF32 instruction set has been removed from GCC. This follows the deprecation and removal of AMX-TRANSPOSE from Intel’s Instruction Set Extensions (ISE). The removal affects various parts of the compiler, including CPU feature detection, intrinsic functions, and documentation. Parts of this change are backported to GCC 15 and 16.
In Details
AMX-TF32 support is being removed from GCC's i386 backend due to its removal from the ISA specification. This commit cleans up various internal structures, headers, and documentation related to this feature, including specific CPUID bits and option masks. The change is reflected across GCC versions to maintain consistency.
- AMX-TF32
- A specific instruction set extension related to Intel's Advanced Matrix Extensions (AMX), designed for TF32 (TensorFloat-32) operations. Its removal from hardware and specifications leads to its removal from the compiler.
- ISE
- Intel Instruction Set Extensions. A document detailing new instructions and features added to Intel processors over time.
- Intrinsics
- Special functions that allow C/C++ code to directly access specific hardware instructions, often for performance-critical operations.