diff --git a/MEMO b/MEMO index 4bbf9177621cc60840357c64a3b2be0c72305778..751f9769df87ed194078f4dc4e6c206c7b2cd198 100644 --- a/MEMO +++ b/MEMO @@ -42,6 +42,8 @@ Things to consider To-do ^^^^^ +- variable shuffle detection + - Deal with equality constraints. (These arise, e.g., when partitioning a loop of length 16 into 16s.) @@ -74,8 +76,6 @@ Future ideas - Try, fix indirect addressing -- variable shuffle detection - - Use gists (why do disjoint sets arise?) - Nested slab decomposition (in conjunction with conditional hoisting) could diff --git a/test/test_loopy.py b/test/test_loopy.py index 24446b0b9a3f9bdad0e67417a10255088ac812f7..f5eef7318657e3a0b3317dd1bfa05627442822cf 100644 --- a/test/test_loopy.py +++ b/test/test_loopy.py @@ -1,9 +1,6 @@ from __future__ import division import numpy as np -import pyopencl as cl -import pyopencl.array as cl_array -import pyopencl.clrandom as cl_random import loopy as lp from pyopencl.tools import pytest_generate_tests_for_pyopencl \ @@ -15,18 +12,14 @@ from pyopencl.tools import pytest_generate_tests_for_pyopencl \ def test_owed_barriers(ctx_factory): dtype = np.float32 ctx = ctx_factory() - order = "C" - queue = cl.CommandQueue(ctx, - properties=cl.command_queue_properties.PROFILING_ENABLE) knl = lp.make_kernel(ctx.devices[0], "{[i]: 0<=i<100}", [ "[i:l.0] <float32> z[i] = a[i]" ], - [ - lp.ArrayArg("a", dtype, shape=(100,)), - ]) + [lp.ArrayArg("a", np.float32, shape=(100,))] + ) kernel_gen = lp.generate_loop_schedules(knl) kernel_gen = lp.check_kernels(kernel_gen) @@ -39,20 +32,14 @@ def test_owed_barriers(ctx_factory): def test_wg_too_small(ctx_factory): - dtype = np.float32 ctx = ctx_factory() - order = "C" - queue = cl.CommandQueue(ctx, - properties=cl.command_queue_properties.PROFILING_ENABLE) knl = lp.make_kernel(ctx.devices[0], "{[i]: 0<=i<100}", [ "[i:l.0] <float32> z[i] = a[i]" ], - [ - lp.ArrayArg("a", dtype, shape=(100,)), - ], + [lp.ArrayArg("a", np.float32, shape=(100,))], local_sizes={0: 16}) kernel_gen = lp.generate_loop_schedules(knl)