Improve _Py_yield to use light weight cpu instruction

[corona10]

Proposal:

_Py_yield() currently relies on sched_yield()/SwitchToThread(), which are OS-level syscalls. We can replace these with lightweight CPU pause instructions (x86 PAUSE, AArch64 WFE, etc.) as CPython's bundled mimalloc already does in mi_atomic_yield():

cpython/Include/internal/mimalloc/mimalloc/atomic.h

Lines 323 to 389 in e682141

	// Yield
	#if defined(__cplusplus)
	#include <thread>
	static inline void mi_atomic_yield(void) {
	std::this_thread::yield();
	}
	#elif defined(_WIN32)
	#define WIN32_LEAN_AND_MEAN
	#include <windows.h>
	static inline void mi_atomic_yield(void) {
	YieldProcessor();
	}
	#elif defined(__SSE2__)
	#include <emmintrin.h>
	static inline void mi_atomic_yield(void) {
	_mm_pause();
	}
	#elif (defined(__GNUC__) || defined(__clang__)) && \
	(defined(__x86_64__) || defined(__i386__) || \
	defined(__aarch64__) || defined(__arm__) || \
	defined(__powerpc__) || defined(__ppc__) || defined(__PPC__) || defined(__POWERPC__))
	#if defined(__x86_64__) || defined(__i386__)
	static inline void mi_atomic_yield(void) {
	__asm__ volatile ("pause" ::: "memory");
	}
	#elif defined(__aarch64__)
	static inline void mi_atomic_yield(void) {
	__asm__ volatile("wfe");
	}
	#elif defined(__arm__)
	#if __ARM_ARCH >= 7
	static inline void mi_atomic_yield(void) {
	__asm__ volatile("yield" ::: "memory");
	}
	#else
	static inline void mi_atomic_yield(void) {
	__asm__ volatile ("nop" ::: "memory");
	}
	#endif
	#elif defined(__powerpc__) || defined(__ppc__) || defined(__PPC__) || defined(__POWERPC__)
	#ifdef __APPLE__
	static inline void mi_atomic_yield(void) {
	__asm__ volatile ("or r27,r27,r27" ::: "memory");
	}
	#else
	static inline void mi_atomic_yield(void) {
	__asm__ __volatile__ ("or 27,27,27" ::: "memory");
	}
	#endif
	#endif
	#elif defined(__sun)
	// Fallback for other archs
	#include <synch.h>
	static inline void mi_atomic_yield(void) {
	smt_pause();
	}
	#elif defined(__wasi__)
	#include <sched.h>
	static inline void mi_atomic_yield(void) {
	sched_yield();
	}
	#else
	#include <unistd.h>
	static inline void mi_atomic_yield(void) {
	sleep(0);
	}
	#endif

Has this already been discussed elsewhere?

No response given

Links to previous discussion of this feature:

No response

Linked PRs

• gh-144586: Improve _Py_yield to improve light weight cpu instruction #144587

[corona10]

I also plan to fix our mimalloc downstream, if microsoft/mimalloc#1225 is merged.