Use bfcvt instruction on Arm to convert to __bf16

[DesWurstes]

Documentation for bfcvt: https://developer.arm.com/documentation/ddi0602/2025-03/SVE-Instructions/BFCVT--Single-precision-down-convert-to-BFloat16-format--predicated--

This code when compiled with -march=armv9-a+fp16 -O3 could use the instruction for better performance:

void convert2(float * __restrict a, __bf16 * __restrict x) {
    x[0] = (__bf16)a[0];
    x[1] = (__bf16)a[1];
    x[2] = (__bf16)a[2];
    x[3] = (__bf16)a[3];
}

[DesWurstes]

Missed flag +bf16

[llvmbot]

@llvm/issue-subscribers-backend-aarch64

Author: None (DesWurstes)

Documentation for `bfcvt`: https://developer.arm.com/documentation/ddi0602/2025-03/SVE-Instructions/BFCVT--Single-precision-down-convert-to-BFloat16-format--predicated--

This code when compiled with -march=armv9-a+fp16 -O3 could use the instruction for better performance:

void convert2(float * __restrict a, __bf16 * __restrict x) {
    x[0] = (__bf16)a[0];
    x[1] = (__bf16)a[1];
    x[2] = (__bf16)a[2];
    x[3] = (__bf16)a[3];
}