Move large categories of tests to their own folders (#1913)

The `tests` folder contains a lot of tests for various things. In
particular, there are a lot of tests for the interpreter and a lot of
tests related to VHLO, so IMO it makes sense to give these their own
folders. I find that this helps navigate the tests.

I considered creating folders for CHLO, printing related tests, and
verification related tests, but IMO there aren't enough of those to
really warrant it.

build_tools/github_actions/lint_version.sh

@@ -32,7 +32,7 @@ set_version_var() {
 set_version_var
 
 ## Check if compatibility tests exist for version `X_Y_0`
-COMPAT_TEST_BASE="stablehlo/tests/stablehlo_legalize_to_vhlo"
+COMPAT_TEST_BASE="stablehlo/tests/vhlo/stablehlo_legalize_to_vhlo"
 COMPAT_TEST="$COMPAT_TEST_BASE.mlir"
 TEST_VERSION="${VERSION[0]}_${VERSION[1]}_0"
 VERSIONED_COMPAT_TEST="$COMPAT_TEST_BASE.$TEST_VERSION.mlir"

docs/compatibility.md

@@ -125,8 +125,8 @@ for roundtrip examples of using the Python Serialization APIs.
 
 ## Tests
 
-We have a compatibility suite in [stablehlo/tests](../stablehlo/tests)
-that involves [a comprehensive compendium of StableHLO ops](../stablehlo/tests/stablehlo_legalize_to_vhlo.mlir)
+We have a compatibility suite in [stablehlo/tests/vhlo](../stablehlo/tests/vhlo)
+that involves [a comprehensive compendium of StableHLO ops](../stablehlo/tests/vhlo/stablehlo_legalize_to_vhlo.mlir)
 serialized for all supported StableHLO versions. For every pull request, we are
 testing both backward and forward compatibility - i.e. that the suite can be
 deserialized targeting HEAD (backward compatibility), that the compendium

docs/interpreter_status.md

@@ -147,11 +147,11 @@ for op semantics and  example usage).
 <!-- markdownlint-disable line-length -->
 We use LLVM's [lit](https://llvm.org/docs/CommandGuide/lit.html) tool to run and
 compare against generated file to diff against the output of the interpreter
-(see [stablehlo/tests](https://github.com/openxla/stablehlo/tree/main/stablehlo/tests)
+(see [stablehlo/tests/interpret](https://github.com/openxla/stablehlo/tree/main/stablehlo/tests/interpret)
 for example tests).
 
 Testing `AddOp` (sample from
-[interpret_add.mlir](https://github.com/openxla/stablehlo/blob/main/stablehlo/tests/interpret_add.mlir)):
+[interpret_add.mlir](https://github.com/openxla/stablehlo/blob/main/stablehlo/tests/interpret/add.mlir)):
 <!-- markdownlint-enable line-length -->
 
 ```mlir
@@ -171,7 +171,7 @@ Testing ops in the Distribution category requires running it via the
 
 <!-- markdownlint-disable line-length -->
 Testing `AllReduceOp` (sample from
-[interpret_all_reduce.mlir](https://github.com/openxla/stablehlo/blob/main/stablehlo/tests/interpret_all_reduce.mlir)):
+[all_reduce.mlir](https://github.com/openxla/stablehlo/blob/main/stablehlo/tests/interpret/all_reduce.mlir)):
 <!-- markdownlint-enable line-length -->
 
 ```mlir
@@ -209,7 +209,7 @@ Following the StableHLO build steps, the StableHLO binaries for tools in
 GDB can be used to step through the code:
 
 ```sh
-gdb --args ./build/bin/stablehlo-translate -allow-unregistered-dialect --interpret ./stablehlo/tests/<test>.mlir
+gdb --args ./build/bin/stablehlo-translate -allow-unregistered-dialect --interpret ./stablehlo/tests/interpret/<test>.mlir
 ```
 
 ## Appendix

docs/reference.md

@@ -129,8 +129,7 @@ A test utility `stablehlo-translate --interpret`
 is responsible for parsing the program, interpreting each function including the
 operations constituting the function. We have a dedicated test-suite, consisting
 of several tests exercising various runtime behaviors, for each StableHLO Op.
-The tests can be found [here](https://github.com/openxla/stablehlo/tree/main/stablehlo/tests/)
-(e.g. interpret\_\*.mlir).
+The tests can be found [here](https://github.com/openxla/stablehlo/tree/main/stablehlo/tests/interpret).
 
 ### Testing guidelines
 

docs/reference_checklist.md

@@ -61,8 +61,8 @@ on those fronts alongside the interpreter implementation.
         1. In case there is a mismatch between the constraints in the
            implementation VS and those in the specification, make sure there is
            an open issue reflecting that discrepancy.
-1. In [interpreter tests](https://github.com/openxla/stablehlo/tree/main/stablehlo/tests):
-    1. Add a file called `interpret_<op_mnemonic>.mlir`.
+1. In [interpreter tests](https://github.com/openxla/stablehlo/tree/main/stablehlo/tests/interpret):
+    1. Add a file called `<op_mnemonic>.mlir`.
     1. Write tests following the [testing guidelines](https://github.com/openxla/stablehlo/blob/main/docs/reference.md#testing-guidelines).
 1. In the [testdata directory](https://github.com/openxla/stablehlo/tree/main/stablehlo/testdata):
     1. Run any disabled tests that are covered by the newly added operation.