I386 gains function multiversioning support for AVX10.2 and APXF.
GCC now supports Function Multiversioning (FMV) for upcoming AVX10.2 and APXF instruction sets, allowing optimized code execution on newer Intel processors.
GCC now supports Function Multiversioning (FMV) for the new AVX10.2 and APXF instruction sets, enabling the compiler to generate specialized code paths that take advantage of these features on compatible hardware. This update introduces new priority levels for these instruction sets within the FMV mechanism and registers processors like Diamond Rapids and NovaLake to utilize the enhanced AVX10.2 capabilities, ensuring optimal performance for applications compiled with GCC targeting these advanced architectures.
In Details
This commit extends GCC's Function Multiversioning (FMV) framework, specifically within the i386 backend, to recognize and prioritize the new AVX10.2 and APXF instruction sets. It modifies enum feature_priority in i386-cpuinfo.h to include P_AVX10_2, P_APX_F, and P_PROC_AVX10_2, which govern the selection of specialized function versions at runtime. The processor_alias_table in i386-common.cc is updated to map certain processors (Diamond Rapids, NovaLake) to P_PROC_AVX10_2, ensuring that features are correctly matched to the CPU's capabilities. This change primarily affects co…
For Context
Function Multiversioning (FMV) is a compiler optimization technique that allows a single function in your code to have multiple versions, each optimized for different CPU features. When your program runs, the operating system or runtime library detects the specific features available on your processor and then directs the program to execute the most efficient version of that function. This is particularly useful for performance-critical tasks like scientific computing or multimedia processing, where modern CPUs offer highly specialized instruction sets (like AVX for advanced vector operations). Without FMV, a compiled program would either have to use a generic instruction set that runs on all CPUs, or ship with multiple binaries, leading to larger file sizes and more complex deployment. This particular update adds support for new, even more powerful instruction sets called AVX10.2 and APXF, meaning GCC can now generate even faster code for programs running on the very latest Intel pro…