RFR: 8298935: fix cyclic dependency bug in create_pack logic in SuperWord::find_adjacent_refs

[Emanuel Peter]

On Tue, 31 Jan 2023 18:26:52 GMT, Emanuel Peter <epeter at openjdk.org> wrote:

> Cyclic dependencies are not handled correctly in all cases. Three examples:
> 
> https://github.com/openjdk/jdk/blob/0a834cd991a2f94b784ee4abde06825486fcb97f/test/hotspot/jtreg/compiler/loopopts/superword/TestCyclicDependency.java#L270-L277
> 
> And this, compiled with `-XX:CompileCommand=option,compiler.vectorization.TestOptionVectorizeIR::test*,Vectorize`:
> https://github.com/openjdk/jdk/blob/0a834cd991a2f94b784ee4abde06825486fcb97f/test/hotspot/jtreg/compiler/vectorization/TestOptionVectorizeIR.java#L173-L180
> 
> And for `vmIntrinsics::_forEachRemaining` compile option `Vectorize` is always enabled:
> https://github.com/openjdk/jdk/blob/0a834cd991a2f94b784ee4abde06825486fcb97f/test/hotspot/jtreg/compiler/vectorization/TestForEachRem.java#L69-L73
> 
> All of these examples are vectorized, despite the cyclic dependency of distance 2. The cyclic dependency is dropped, instead the emitted vector code implements a shift by 2, instead of repeating the same 2 values.
> 
> **Analysis**
> 
> The `create_pack` logic in `SuperWord::find_adjacent_refs` is broken in two ways:
> 
> - When the compile directive `Vectorize` is on, or we compile `vmIntrinsics::_forEachRemaining` we have `_do_vector_loop == true`. When that is the case, we blindly trust that there is no cyclic dependency larger than distance 1. Distance 1 would already be detected by the `independence(s1, s2)` checks we do for all adjacent memops. But for larger distances, we rely on `memory_alignment == 0`. But the compile directive avoids these checks.
> - If `best_align_to_mem_ref` is of a different type, and we have `memory_alignment(mem_ref, best_align_to_mem_ref) == 0`, we do not check if `mem_ref` has `memory_alignment == 0` for all other refs of the same type. In the example `TestCyclicDependency::test2`, we have `best_align_to_mem_ref` as the `StoreF`. Then we assess the `StoreI`, which is not aligned with it, but it is of a different type, so we accept it too. Finally, we look at `LoadI`, which has perfect alignment with the `StoreF`, so we accept it too (even though it is in conflict with the `StoreI`).
> 
> Generally, the nested if-statements are confusing and buggy. I propose to fix and refactor the code.
> 
> I also propose to only allow the compile directive `Vectorize` only if `vectors_should_be_aligned() == false`. If all vector operations have to be `vector_width` aligned, then they also have to be mutually aligned, and we cannot have patterns like `v[i] = v[i] + v[i+1]` for which the compile directive was introduced in the first place https://github.com/openjdk/jdk/commit/c7d33de202203b6da544f2e0f9a13952381b32dd.
> 
> **Solution**
> 
> First, I implemented `SuperWord::verify_packs` which catches cyclic dependencies just before scheduling. The idea is to reassess every pack, and check if all memops in it are mutually independent. Turns out that per vector pack, it suffices to do a single BFS over the nodes in the block (see `SuperWord::find_dependence`). With this verification in place we at least get an assert instead of wrong execution.
> 
> I then refactored and fixed the `create_pack` code, and put the logic all in `SuperWord::is_mem_ref_alignment_ok`. With the added comments, I hope the logic is more straight forward and readable. If `_do_vector_loop == true`, then I filter the vector packs again in `SuperWord::combine_packs`, since we are at that point not sure that the packs are actually independent, we only know that adjacient memops are independent.
> 
> **Testing**
> 
> I added a few more regression tests, and am running tier1-3, plus some stress testing.
> 
> However, I need help from someone who can test this on **ARM32** and **SPARC**, basically machines that have `vectors_should_be_aligned() == false`. I would love to have additional testing on those machine, and some reviews.
> 
> **Discussion / Future Work**
> 
> I wonder if we should have `_do_vector_loop == true` by default, since it allows more vectorization. With the added filtering, we are sure that we do not schedule packs with cyclic dependencies. We would have to evaluate performance and other side-effects of course. What do you think?

Generally, I am wondering about this though:
Why do we force the loads / stores of the same type to be `completely overlapped` (like @fg1417  calles it), so have `memory_alignment(p1, p2) == 0` for all `p1`, `p2` of the same type? This seems to be more constrained than necessary. Why do we not just rely on packs being internally independent, ie `independent(s1, s2)` for all `s1, s2` in the same pack?

This would then allow things like this:

    private static void test() {
        for (int i = 0; i < 100; i++) {
            iArr[i] = iArr[i+2]; // read forward is ok, vectorization leads to correct results
        }
    }

(We only do this if we have `_do_vector_loop == true` -> basically we need to give C2 the hint that vectorization is ok)

But it would prevent things like this:

    private static void test() {
        for (int i = 0; i < 100; i++) {
            iArr[i+2] = iArr[i]; // write forward leads to cyclic dependencies -> should not vectorize
        }
    }

(We currently vectorize this if we have `_do_vector_loop == true`, but that is a bug, leads to wrong results. C2 trusts us blindly.)

-------------

PR: https://git.openjdk.org/jdk/pull/12350