THURSDAY, JULY 2, 2026
x86
x86: Disable AMX-TF32 by default
AMX-TF32 feature support is disabled by default in binutils.
The AMX-TF32 instruction set extension is now disabled by default in the binutils assembler (gas). This change reflects its depublishing and lack of current or future hardware support. Test cases are updated to use an .arch directive for .amx_tf32 to maintain test integrity.
In Details
AMX-TF32, an advanced matrix extension feature, has been removed from the default enablement in binutils gas. This change reflects its depublishing by Intel and the absence of hardware support. The opt-in mechanism via assembler directives remains for backward compatibility and testing by explicitly enabling .arch .amx_tf32.
For Context
- AMX-TF32
- Advanced Matrix Extensions with TF32 format support. A feature for accelerating matrix operations, particularly in AI/ML workloads. TF32 is a specific floating-point format.
- binutils
- A collection of programming tools for maintaining and modifying binary files. It includes the assembler (as), linker (ld), and other utilities. GNU Binutils is the most common implementation.
- gas
- The GNU Assembler, part of the GNU Binutils package. It translates assembly language code into machine code.
- x86
- A family of instruction set architectures (ISAs) developed by Intel, widely used in desktop computers, laptops, and servers.