Avoid undefined behaviour with signed integer multiplication

[rathann]

This is the minimal patch that makes tests pass with clang 20.1.x on Fedora rawhide x86_64 and aarch64.

Fixes #1295 .

[aqrit]

Fixes #1295 .

simde_vmul_laneq_s32 was for an example.
The "many intrinsics" statement is key.

related:

Undefined behavior stemming from signed overflow seems to be everywhere.
At this time, I'd say that SIMDE should never be used without -fwrapv.
#901 (comment)

[rathann]

It looks like this is now failing when compiled as C++:

../test/arm/neon/../../../simde/arm/neon/mul_lane.h:389:32: error: use of old-style cast [-Werror,-Wold-style-cast]

What is the recommended way of keeping the code compatible with both C and C++? I'm not very good with C++.

[mr-c]

It looks like this is now failing when compiled as C++:

../test/arm/neon/../../../simde/arm/neon/mul_lane.h:389:32: error: use of old-style cast [-Werror,-Wold-style-cast]

What is the recommended way of keeping the code compatible with both C and C++? I'm not very good with C++.

My mistake, sorry

Try one of HEDLEY_STATIC_CAST or HEDLEY_REINTERPRET_CAST

simde/simde/hedley.h

Lines 842 to 858 in 51743e7

	#if defined(HEDLEY_REINTERPRET_CAST)
	# undef HEDLEY_REINTERPRET_CAST
	#endif
	#if defined(__cplusplus)
	# define HEDLEY_REINTERPRET_CAST(T, expr) (reinterpret_cast<T>(expr))
	#else
	# define HEDLEY_REINTERPRET_CAST(T, expr) ((T) (expr))
	#endif
	#if defined(HEDLEY_STATIC_CAST)
	# undef HEDLEY_STATIC_CAST
	#endif
	#if defined(__cplusplus)
	# define HEDLEY_STATIC_CAST(T, expr) (static_cast<T>(expr))
	#else
	# define HEDLEY_STATIC_CAST(T, expr) ((T) (expr))
	#endif