AArch64 gains new floating-point to integer conversion intrinsics.
GCC now exposes new AArch64 NEON intrinsics for float-to-int conversions, expanding the available rounding and conversion options.
This commit adds several new AArch64 NEON intrinsics for converting floating-point values to integers. These new intrinsics cover specific conversion paths, such as float32_t to uint64_t and float64_t to int32_t, with various rounding modes. Developers using NEON can now directly access these specialized conversions, which can lead to more efficient and precise floating-point to integer handling in their applications.
In Details
This change introduces new ARM NEON intrinsics within aarch64-simd-builtins.def and arm_neon.h. These intrinsics provide direct access to specific floating-point to integer conversion operations, such as vcvtd_s32_f64 (convert double to signed 32-bit integer) and vcvtas_u64_f32 (convert float to unsigned 64-bit integer, rounding to nearest, ties away from zero). These are essentially wrappers around underlying AArch64 instructions, simplifying their use for developers. This expands the set of available SIMD conversion operations, which can be crucial for numerical precision and perfor…
For Context
In computer programming, 'intrinsics' are special functions that allow you to use very specific, optimized instructions of the computer's processor. This update to GCC, a popular compiler, adds new intrinsic functions for AArch64 processors. Specifically, these new functions help convert numbers with decimal points (floating-point numbers) into whole numbers (integers) in various ways, such as converting a 32-bit float to a 64-bit unsigned integer, or a 64-bit float to a 32-bit signed integer. These conversions are important in many applications, like scientific calculations or graphics, where you need to change how numbers are represented without losing too much precision or to fit them into different data sizes. By providing direct access to these highly optimized processor instructions, developers can write faster and more efficient code.