Fix i64x2 shift on big-endian

[sertonix]

Since the i64x2.sh* functions use i32 as shift argument the access to other.i64 may not use the correct union member. On big-endian systems this causes tests in test/spec/testsuite/simd_bit_shift.wast and a few other places to fail.

Ref #2983

[kripken]

Hmm, by the wasm validation rules, the type of other must be identical. Are you seeing a situation where it is not?

[sertonix]

Not with the indirect usage of Literal::shl through Literal::shlI64x2:

binaryen/src/wasm/literal.cpp

Lines 2264 to 2266 in 2f63efb

	Literal Literal::shlI64x2(const Literal& other) const {
	return shift<2, &Literal::getLanesI64x2, &Literal::shl>(*this, other);
	}

binaryen/src/wasm/wasm-validator.cpp

Lines 1616 to 1628 in 2f63efb

	void FunctionValidator::visitSIMDShift(SIMDShift* curr) {
	shouldBeTrue(getModule()->features.hasSIMD(),
	curr,
	"SIMD operations require SIMD [--enable-simd]");
	shouldBeEqualOrFirstIsUnreachable(
	curr->type, Type(Type::v128), curr, "vector shift must have type v128");
	shouldBeEqualOrFirstIsUnreachable(
	curr->vec->type, Type(Type::v128), curr, "expected operand of type v128");
	shouldBeEqualOrFirstIsUnreachable(curr->shift->type,
	Type(Type::i32),
	curr,
	"expected shift amount to have type i32");
	}

[kripken]

I see, thanks. I guess this makes sense - wasm semantics are more strict than our Literal APIs should allow. I was confused before.