[REVIEW] Check alignment before binaryOp dispatch

[tfeher]

This PR adds pointer alignment check for the binaryOp. Without this, the added unit test would trigger cudaErrorMisalignedAddress.

The alignment check is done the same way as for the unaryOp:

raft/cpp/include/raft/linalg/unary_op.cuh

Lines 79 to 94 in ca7465e

	uint64_t inAddr = uint64_t(in);
	uint64_t outAddr = uint64_t(out);
	if (16 / maxSize && bytes % 16 == 0 && inAddr % 16 == 0 &&
	outAddr % 16 == 0) {
	unaryOpImpl<InType, 16 / maxSize, Lambda, OutType, IdxType, TPB>(
	out, in, len, op, stream);
	} else if (8 / maxSize && bytes % 8 == 0 && inAddr % 8 == 0 &&
	outAddr % 8 == 0) {
	unaryOpImpl<InType, 8 / maxSize, Lambda, OutType, IdxType, TPB>(
	out, in, len, op, stream);
	} else if (4 / maxSize && bytes % 4 == 0 && inAddr % 4 == 0 &&
	outAddr % 4 == 0) {
	unaryOpImpl<InType, 4 / maxSize, Lambda, OutType, IdxType, TPB>(
	out, in, len, op, stream);
	} else if (2 / maxSize && bytes % 2 == 0 && inAddr % 2 == 0 &&
	outAddr % 2 == 0) {

[GPUtester]

Please update the changelog in order to start CI tests.

View the gpuCI docs here.

[teju85]

Changes LGTM. (Thanks for adding an explicit mis-alignment unit-test!)

[tfeher]

rerun tests