Work around long compile times for complex fuzzed expressions on gcc …

…11+ (gh-686)

test/test_expression.py

@@ -242,10 +242,10 @@ def assert_parse_roundtrip(expr):
     assert expr == parsed_expr
 
 
-@pytest.mark.parametrize("target", [lp.PyOpenCLTarget, lp.ExecutableCTarget])
+@pytest.mark.parametrize("target_cls", [lp.PyOpenCLTarget, lp.ExecutableCTarget])
 @pytest.mark.parametrize("random_seed", [0, 1, 2, 3, 4, 5])
 @pytest.mark.parametrize("expr_type", ["int", "int_nonneg", "real", "complex"])
-def test_fuzz_expression_code_gen(ctx_factory, expr_type, random_seed, target):
+def test_fuzz_expression_code_gen(ctx_factory, expr_type, random_seed, target_cls):
     from pymbolic import evaluate
 
     def get_numpy_type(x):
@@ -290,7 +290,7 @@ def get_numpy_type(x):
 
         var_name = "expr%d" % i
 
-        print(expr)
+        # print(expr)
         #assert_parse_roundtrip(expr)
 
         if expr_type in ["int", "int_nonneg"]:
@@ -299,7 +299,7 @@ def get_numpy_type(x):
                     var_values,
                     lbound=result_type_iinfo.min,
                     ubound=result_type_iinfo.max)
-            print(expr)
+            # print(expr)
             try:
                 ref_values[var_name] = bceval_mapper(expr)
             except BoundsCheckError:
@@ -330,8 +330,27 @@ def get_numpy_type(x):
         if expr_type == "int_nonneg":
             var_names.extend(var_values)
 
+    if issubclass(target_cls, lp.ExecutableCTarget):
+        # https://github.com/inducer/loopy/issues/686
+        # https://gcc.gnu.org/bugzilla/show_bug.cgi?id=107127
+
+        from shutil import which
+        gcc_10 = which("gcc-10")
+        if gcc_10 is not None:
+            from loopy.target.c.c_execution import CCompiler
+            target = target_cls(compiler=CCompiler(cc=gcc_10))
+        else:
+            from warnings import warn
+            warn("Using default C compiler because gcc-10 was not found. "
+                 "These tests may take a long time, because of "
+                 "https://gcc.gnu.org/bugzilla/show_bug.cgi?id=107127.")
+            target = target_cls()
+
+    else:
+        target = target_cls()
+
     knl = lp.make_kernel("{ : }", instructions, data, seq_dependencies=True,
-            target=target())
+            target=target)
 
     import islpy as isl
     knl = lp.assume(knl, isl.BasicSet(
@@ -341,14 +360,14 @@ def get_numpy_type(x):
                 " and ".join("%s >= 0" % name for name in var_names))))
 
     knl = lp.set_options(knl, return_dict=True)
-    print(knl)
+    # print(knl)
 
-    if target == lp.PyOpenCLTarget:
+    if type(target) is lp.PyOpenCLTarget:
         cl_ctx = ctx_factory()
         knl = lp.set_options(knl, write_code=True)
         with cl.CommandQueue(cl_ctx) as queue:
             evt, lp_values = knl(queue, out_host=True)
-    elif target == lp.ExecutableCTarget:
+    elif type(target) is lp.ExecutableCTarget:
         evt, lp_values = knl()
     else:
         raise NotImplementedError("unsupported target")