Drop insline asm in favor of std::arch intrinsics whenever it is ready (avx512dq, avx512vl)

[martinothamar]

avx512dq and avx512vl has not yet been implemented into std::arch

related PR: rust-lang/stdarch#954
issue: rust-lang/stdarch#1437

code in question:

simd-rand/src/specific/avx2/simdrand.rs

Lines 52 to 69 in 002173c

	if is_x86_feature_detected!("avx512dq") && is_x86_feature_detected!("avx512vl") {
	// With AVX512 DQ/VL we can use the below instruction
	// with both 512bit and 256bit vectors
	// see https://www.felixcloutier.com/x86/vcvtuqq2pd
	let mut dst: __m256d;
	asm!(
	"vcvtuqq2pd {1}, {0}",
	in(ymm_reg) v,
	out(ymm_reg) dst,
	// PERF: 'nostack' tells the Rust compiler that our asm won't touch the stack.
	// If we don't include this, the compiler might inject additional
	// instructions to make the stack pointer 16byte aligned in accordance to x64 ABI.
	// If we were to 'call' in our inline asm, it would have to push the 8byte return address
	// onto the stack, so the stack would have to be 16byte aligned before this happened
	options(nostack),
	);
	dst
	} else {

and

simd-rand/src/specific/avx512/simdrand.rs

Lines 51 to 69 in 002173c

	unsafe fn m512i_to_m512d(src: __m512i) -> __m512d {
	// this should be exposed through the '_mm512_cvtepu64_pd' C/C++ intrinsic,
	// but since I can't find this exposed in Rust anywhere,
	// we're doing it the inline asm way here
	// TODO: find out what happened in std::arch
	let mut dst: __m512d;
	asm!(
	"vcvtuqq2pd {1}, {0}",
	in(zmm_reg) src,
	out(zmm_reg) dst,
	// PERF: 'nostack' tells the Rust compiler that our asm won't touch the stack.
	// If we don't include this, the compiler might inject additional
	// instructions to make the stack pointer 16byte aligned in accordance to x64 ABI.
	// If we were to 'call' in our inline asm, it would have to push the 8byte return address
	// onto the stack, so the stack would have to be 16byte aligned before this happened
	options(nostack),
	);
	dst
	}