diff --git a/loopy/auto_test.py b/loopy/auto_test.py
index e7e3eb6d2ac3792f18fbc4d193c6d7ad1df82aa1..607d091b98b7a1919dacbe10f0280ea35c666a43 100644
--- a/loopy/auto_test.py
+++ b/loopy/auto_test.py
@@ -311,19 +311,23 @@ def _default_check_result(result, ref_result):
 
 # {{{ find device for reference test
 
-def _enumerate_cl_devices_for_ref_test():
+def _enumerate_cl_devices_for_ref_test(blacklist_ref_vendors):
     import pyopencl as cl
 
     noncpu_devs = []
     cpu_devs = []
 
+    if isinstance(blacklist_ref_vendors, str):
+        blacklist_ref_vendors = blacklist_ref_vendors.split(",")
+
     for pf in cl.get_platforms():
         for dev in pf.get_devices():
+            if any(bl in dev.platform.vendor
+                    for bl in blacklist_ref_vendors):
+                print("blacklist", dev, blacklist_ref_vendors)
+                continue
+
             if dev.type & cl.device_type.CPU:
-                if "Advanced Micro" in dev.platform.vendor:
-                    # Sorry, AMD, your CPU CL has gotten too crashy of late.
-                    # (Feb 2016)
-                    continue
                 if "Intel" in dev.platform.vendor:
                     # Sorry, Intel, your CPU CL has gotten too crashy of late.
                     # (Feb 2016)
@@ -358,7 +362,7 @@ def auto_test_vs_ref(
         dump_binary=False,
         fills_entire_output=None, do_check=True, check_result=None,
         max_test_kernel_count=1,
-        quiet=False):
+        quiet=False, blacklist_ref_vendors=[]):
     """Compare results of `ref_knl` to the kernels generated by
     scheduling *test_knl*.
 
@@ -413,7 +417,7 @@ def auto_test_vs_ref(
 
     ref_errors = []
 
-    for dev in _enumerate_cl_devices_for_ref_test():
+    for dev in _enumerate_cl_devices_for_ref_test(blacklist_ref_vendors):
         ref_ctx = cl.Context([dev])
         ref_queue = cl.CommandQueue(ref_ctx,
                 properties=cl.command_queue_properties.PROFILING_ENABLE)
@@ -424,6 +428,9 @@ def auto_test_vs_ref(
             ref_sched_kernel = knl
             break
 
+        logger.info("%s (ref): trying %s for the reference calculation" % (
+            ref_knl.name, dev))
+
         ref_compiled = CompiledKernel(ref_ctx, ref_sched_kernel)
         if not quiet and print_ref_code:
             print(75*"-")
diff --git a/loopy/codegen/loop.py b/loopy/codegen/loop.py
index cc45f95e7548945d83f38ae0e7bc9c08c5d4f013..70530d3a90cbdd5a48395034dd05000f6781b473 100644
--- a/loopy/codegen/loop.py
+++ b/loopy/codegen/loop.py
@@ -431,7 +431,8 @@ def generate_sequential_loop_dim_code(kernel, sched_index, codegen_state):
         if (static_ubound - static_lbound).plain_is_zero():
             # single-trip, generate just a variable assignment, not a loop
             result.append(gen_code_block([
-                Initializer(Const(POD(kernel.index_dtype, loop_iname)),
+                Initializer(
+                    Const(POD(kernel.target, kernel.index_dtype, loop_iname)),
                     ecm(aff_to_expr(static_lbound), PREC_NONE, "i")),
                 Line(),
                 inner,
diff --git a/test/test_linalg.py b/test/test_linalg.py
index 9c6803e93f2e53a0c071e0372bf71256854de38a..6aeec63c49a72c784ad5cccf1ee7acc1fcae0f2a 100644
--- a/test/test_linalg.py
+++ b/test/test_linalg.py
@@ -30,6 +30,8 @@ import pyopencl as cl
 import pyopencl.array as cl_array
 import loopy as lp
 
+import logging
+
 from pyopencl.tools import (  # noqa
         pytest_generate_tests_for_pyopencl
         as pytest_generate_tests)
@@ -61,12 +63,16 @@ def check_float4(result, ref_result):
 
 
 def test_axpy(ctx_factory):
+    logging.basicConfig(level="INFO")
     ctx = ctx_factory()
 
     n = 3145182
 
     vec = cl_array.vec
 
+    if ctx.devices[0].platform.vendor.startswith("Advanced Micro"):
+        pytest.skip("crashes on AMD 15.12")
+
     for dtype, check, a, b in [
             (np.complex64, None, 5, 7),
             (vec.float4, check_float4,
@@ -111,7 +117,8 @@ def test_axpy(ctx_factory):
             lp.auto_test_vs_ref(seq_knl, ctx, variant(knl),
                     op_count=[np.dtype(dtype).itemsize*n*3/1e9],
                     op_label=["GBytes"],
-                    parameters={"a": a, "b": b, "n": n}, check_result=check)
+                    parameters={"a": a, "b": b, "n": n}, check_result=check,
+                    blacklist_ref_vendors=["Advanced Micro"])
 
 
 def test_transpose(ctx_factory):
@@ -463,7 +470,7 @@ def test_magma_fermi_matrix_mul(ctx_factory):
 
     lp.auto_test_vs_ref(seq_knl, ctx, knl,
             op_count=[2*n**3/1e9], op_label=["GFlops"],
-            parameters={})
+            parameters={}, blacklist_ref_vendors="pocl")
 
 
 def test_image_matrix_mul(ctx_factory):
diff --git a/test/test_loopy.py b/test/test_loopy.py
index 667d7365d412d9be6d9a51c8d63c9470ffede2f9..1fed3289aac1c184b2267e3425aed2d8023f9a03 100644
--- a/test/test_loopy.py
+++ b/test/test_loopy.py
@@ -529,6 +529,9 @@ def test_fuzz_code_generator(ctx_factory):
     ctx = ctx_factory()
     queue = cl.CommandQueue(ctx)
 
+    if ctx.devices[0].platform.vendor.startswith("Advanced Micro"):
+        pytest.skip("crashes on AMD 15.12")
+
     #from expr_fuzz import get_fuzz_examples
     #for expr, var_values in get_fuzz_examples():
     for expr, var_values in generate_random_fuzz_examples(50):