From 9b751a81b0125e582322eec16bacba078c2c4e97 Mon Sep 17 00:00:00 2001
From: Andreas Kloeckner <inform@tiker.net>
Date: Tue, 17 Apr 2012 20:22:20 -0400
Subject: [PATCH] Rename ArrayArg -> GlobalArg, ConstantArrayArg ->
ConstantArg.
---
MEMO | 6 +--
doc/reference.rst | 4 +-
examples/matrix-mul.py | 2 +-
examples/quadrature.py | 105 ------------------------------------
loopy/__init__.py | 6 +--
loopy/check.py | 4 +-
loopy/codegen/__init__.py | 11 ++--
loopy/codegen/expression.py | 2 +-
loopy/compiled.py | 14 ++---
loopy/kernel.py | 21 +++++---
test/test_fem_assembly.py | 10 ++--
test/test_linalg.py | 52 +++++++++---------
test/test_loopy.py | 12 ++---
test/test_nbody.py | 4 +-
test/test_sem.py | 68 +++++++++++------------
15 files changed, 110 insertions(+), 211 deletions(-)
delete mode 100644 examples/quadrature.py
diff --git a/MEMO b/MEMO
index a0f7f32ff..02eaf2264 100644
--- a/MEMO
+++ b/MEMO
@@ -46,8 +46,6 @@ To-do
- Add dependencies after the fact
-- bug? with fetching only g[j,*] inside j loop
-
- Expose iname-duplicate-and-rename as a primitive.
- Allow parameters to be varying during run-time, substituting values
@@ -61,7 +59,9 @@ To-do
- If finding a maximum proves troublesome, move parameters into the domain
-- : (as in, Matlab full-sclice) in prefetches
+- : (as in, Matlab full-slice) in prefetches
+
+- ScalarArg is a bad name
Future ideas
^^^^^^^^^^^^
diff --git a/doc/reference.rst b/doc/reference.rst
index 1f19611d1..6915549f6 100644
--- a/doc/reference.rst
+++ b/doc/reference.rst
@@ -121,11 +121,11 @@ Arguments
:members:
:undoc-members:
-.. autoclass:: ArrayArg
+.. autoclass:: GlobalArg
:members:
:undoc-members:
-.. autoclass:: ConstantArrayArg
+.. autoclass:: ConstantArg
:members:
:undoc-members:
diff --git a/examples/matrix-mul.py b/examples/matrix-mul.py
index 82c8c4e33..d0f11c261 100644
--- a/examples/matrix-mul.py
+++ b/examples/matrix-mul.py
@@ -43,7 +43,7 @@ def image_matrix_mul_ilp(ctx_factory=cl.create_some_context):
[
lp.ImageArg("a", dtype, 2),
lp.ImageArg("b", dtype, 2),
- lp.ArrayArg("c", dtype, shape=(n, n), order=order),
+ lp.GlobalArg("c", dtype, shape=(n, n), order=order),
],
name="matmul")
diff --git a/examples/quadrature.py b/examples/quadrature.py
deleted file mode 100644
index 5df05e031..000000000
--- a/examples/quadrature.py
+++ /dev/null
@@ -1,105 +0,0 @@
-
-import numpy as np
-import pyopencl as cl
-import pyopencl.array as cl_array
-import loopy as lp
-
-
-
-
-def make_well_conditioned_dev_matrix(queue, shape, dtype=np.float32,
- order="C", ran_factor=1, id_factor=5, inc_factor=0, od=0):
- if isinstance(shape, int):
- shape = (shape, shape)
- l = max(shape)
- eye_ish = id_factor*np.eye(l, k=od)
- if inc_factor:
- eye_ish[np.arange(l), np.arange(l)] = inc_factor*np.arange(l)
- ary = np.asarray(
- ran_factor*np.random.randn(*shape)
- + eye_ish[:shape[0], :shape[1]],
- dtype=dtype, order=order)
-
- return cl_array.to_device(queue, ary)
-
-
-
-
-def build_mass_mat_maker(ctx_factory=cl.create_some_context):
- dtype = np.float32
- ctx = ctx_factory()
- queue = cl.CommandQueue(ctx,
- properties=cl.command_queue_properties.PROFILING_ENABLE)
-
- Nb = 3
- Nv = 3
- Nq = 3*3
- Nc = 1600
-
- knl = lp.LoopKernel(ctx.devices[0],
- "[ncells] -> {[c,i,j,q]: 0<=c<ncells and 0 <= i < %(Nv)s "
- "and 0<=j<%(Nb)s and 0<=q<%(Nq)s}" % dict(
- Nv=Nv, Nb=Nb, Nq=Nq),
- [
- "m[c,i,j] = w[q]*det_j[c]*phi[i,q]*phi[j,q]",
- ],
- [
- lp.ArrayArg("m", dtype, shape=(Nc, Nv, Nb)),
- lp.ArrayArg("w", dtype, shape=(Nq,)),
- lp.ArrayArg("det_j", dtype, shape=(Nc,)),
- lp.ArrayArg("phi", dtype, shape=(Nv, Nq,)),
- lp.ScalarArg("ncells", np.int32, approximately=1000),
- ],
- name="mass_mat",
- iname_to_tag=dict(i="l.0", j="l.1"),
- assumptions="ncells >= 1"
- )
- knl = lp.split_dimension(knl, "c", 8, inner_tag="l.2",
- outer_slab_increments=(0,0))
- knl = lp.split_dimension(knl, "c_outer", 8, outer_tag="g.0",
- outer_slab_increments=(0,0))
-
- # fix reg prefetch
- knl = lp.add_prefetch(knl, "det_j", ["c_inner"],
- loc_fetch_axes={0: (0, 1)})
-
- #ilp = 4
- #knl = lp.split_dimension(knl, "i", 2, outer_tag="g.0", inner_tag="l.1")
- #j_inner_split = 16
- #knl = lp.split_dimension(knl, "j", ilp*j_inner_split, outer_tag="g.1")
- #knl = lp.split_dimension(knl, "j_inner", j_inner_split, outer_tag="ilp", inner_tag="l.0")
- #knl = lp.split_dimension(knl, "k", 2)
-
- #knl = lp.add_prefetch(knl, 'a', ["i_inner", "k_inner"])
- #knl = lp.add_prefetch(knl, 'b', ["j_inner_outer", "j_inner_inner", "k_inner"])
- #assert knl.get_problems({})[0] <= 2
-
- kernel_gen = (knl
- #kernel_gen = (lp.insert_register_prefetches(knl)
- for knl in lp.generate_loop_schedules(knl))
-
- if False:
- a = make_well_conditioned_dev_matrix(queue, n, dtype=dtype, order=order,
- ran_factor=1, id_factor=5)
- b = make_well_conditioned_dev_matrix(queue, n, dtype=dtype, order=order,
- ran_factor=1, id_factor=5, inc_factor=0)
- c = cl_array.empty_like(a)
- a_img = cl.image_from_array(ctx, a.get(), 1)
- b_img = cl.image_from_array(ctx, b.get(), 1)
-
- def launcher(kernel, gsize, lsize, check):
- 1/0
- evt = kernel(queue, gsize(), lsize(), a_img, b_img, c.data,
- g_times_l=True)
-
- return evt
-
- from pyopencl.characterize import get_fast_inaccurate_build_options
- lp.drive_timing_run(kernel_gen, queue, launcher, flop_count=0,
- options=get_fast_inaccurate_build_options(ctx.devices[0]))
-
-
-
-
-if __name__ == "__main__":
- build_mass_mat_maker()
diff --git a/loopy/__init__.py b/loopy/__init__.py
index d23cc6707..189aabd4c 100644
--- a/loopy/__init__.py
+++ b/loopy/__init__.py
@@ -10,7 +10,6 @@ register_mpz_with_pymbolic()
import islpy as isl
from islpy import dim_type
-import numpy as np
@@ -20,7 +19,7 @@ class LoopyAdvisory(UserWarning):
# {{{ imported user interface
-from loopy.kernel import ScalarArg, ArrayArg, ConstantArrayArg, ImageArg
+from loopy.kernel import ScalarArg, GlobalArg, ArrayArg, ConstantArg, ImageArg
from loopy.kernel import (AutoFitLocalIndexTag, get_dot_dependency_graph,
LoopKernel, Instruction)
@@ -33,7 +32,8 @@ from loopy.codegen import generate_code
from loopy.compiled import CompiledKernel, drive_timing_run, auto_test_vs_ref
from loopy.check import check_kernels
-__all__ = ["ScalarArg", "ArrayArg", "ConstantArrayArg", "ImageArg", "LoopKernel",
+__all__ = ["ScalarArg", "GlobalArg", "ArrayArg", "ConstantArg", "ImageArg",
+ "LoopKernel",
"Instruction", "make_kernel",
"get_dot_dependency_graph",
"preprocess_kernel", "realize_reduction",
diff --git a/loopy/check.py b/loopy/check.py
index 454f3450f..148595636 100644
--- a/loopy/check.py
+++ b/loopy/check.py
@@ -293,10 +293,10 @@ def get_problems(kernel, parameters):
msg(4, "using more local memory than available--"
"possibly OK due to cache nature")
- from loopy.kernel import ArrayArg
+ from loopy.kernel import ConstantArg
const_arg_count = sum(
1 for arg in kernel.args
- if isinstance(arg, ArrayArg) and arg.constant_mem)
+ if isinstance(arg, ConstantArg))
if const_arg_count > kernel.device.max_constant_args:
msg(5, "too many constant arguments")
diff --git a/loopy/codegen/__init__.py b/loopy/codegen/__init__.py
index f2c04e2aa..4653adc53 100644
--- a/loopy/codegen/__init__.py
+++ b/loopy/codegen/__init__.py
@@ -223,19 +223,16 @@ def generate_code(kernel, with_annotation=False,
has_double = False
has_image = False
- from loopy.kernel import ArrayArg, ConstantArrayArg, ImageArg, ScalarArg
+ from loopy.kernel import GlobalArg, ConstantArg, ImageArg, ScalarArg
args = []
for arg in kernel.args:
- if isinstance(arg, (ConstantArrayArg, ArrayArg)):
+ if isinstance(arg, (ConstantArg, GlobalArg)):
arg_decl = restrict_ptr_if_not_nvidia(
POD(arg.dtype, arg.name))
if arg_decl.name not in kernel.get_written_variables():
- if arg.constant_mem:
- arg_decl = CLConstant(Const(arg_decl))
- else:
- arg_decl = Const(arg_decl)
- if isinstance(arg, ConstantArrayArg):
+ arg_decl = Const(arg_decl)
+ if isinstance(arg, ConstantArg):
arg_decl = CLConstant(arg_decl)
else:
arg_decl = CLGlobal(arg_decl)
diff --git a/loopy/codegen/expression.py b/loopy/codegen/expression.py
index 46f8aa6ed..5435a3b4f 100644
--- a/loopy/codegen/expression.py
+++ b/loopy/codegen/expression.py
@@ -166,7 +166,7 @@ class LoopyCCodeMapper(CCodeMapper):
"non-floating-point images not supported for now")
else:
- # ArrayArg
+ # GlobalArg
index_expr = expr.index
if not isinstance(expr.index, tuple):
index_expr = (index_expr,)
diff --git a/loopy/compiled.py b/loopy/compiled.py
index c23e0b4f3..977237d50 100644
--- a/loopy/compiled.py
+++ b/loopy/compiled.py
@@ -11,7 +11,7 @@ import numpy as np
def _arg_matches_spec(arg, val, other_args):
import loopy as lp
- if isinstance(arg, lp.ArrayArg):
+ if isinstance(arg, lp.GlobalArg):
from pymbolic import evaluate
shape = evaluate(arg.shape, other_args)
@@ -168,7 +168,7 @@ class CompiledKernel:
assert _arg_matches_spec(arg, val, kwargs)
# automatically transfer host-side arrays
- if isinstance(arg, lp.ArrayArg):
+ if isinstance(arg, lp.GlobalArg):
if isinstance(val, np.ndarray):
# synchronous, so nothing to worry about
val = cl_array.to_device(queue, val, allocator=allocator)
@@ -178,7 +178,7 @@ class CompiledKernel:
if is_written:
outputs.append(val)
- if isinstance(arg, lp.ArrayArg):
+ if isinstance(arg, lp.GlobalArg):
args.append(val.data)
else:
args.append(val)
@@ -287,7 +287,7 @@ def fill_rand(ary):
def make_ref_args(kernel, queue, parameters,
fill_value):
- from loopy.kernel import ScalarArg, ArrayArg, ImageArg
+ from loopy.kernel import ScalarArg, GlobalArg, ImageArg
from pymbolic import evaluate
@@ -309,7 +309,7 @@ def make_ref_args(kernel, queue, parameters,
result.append(arg_value)
- elif isinstance(arg, (ArrayArg, ImageArg)):
+ elif isinstance(arg, (GlobalArg, ImageArg)):
if arg.shape is None:
raise ValueError("arrays need known shape to use automatic "
"testing")
@@ -353,7 +353,7 @@ def make_ref_args(kernel, queue, parameters,
def make_args(queue, kernel, ref_input_arrays, parameters,
fill_value):
- from loopy.kernel import ScalarArg, ArrayArg, ImageArg
+ from loopy.kernel import ScalarArg, GlobalArg, ImageArg
from pymbolic import evaluate
@@ -373,7 +373,7 @@ def make_args(queue, kernel, ref_input_arrays, parameters,
result.append(arg_value)
- elif isinstance(arg, (ArrayArg, ImageArg)):
+ elif isinstance(arg, (GlobalArg, ImageArg)):
if arg.name in kernel.get_written_variables():
if isinstance(arg, ImageArg):
raise RuntimeError("write-mode images not supported in "
diff --git a/loopy/kernel.py b/loopy/kernel.py
index 753744974..7fc58debb 100644
--- a/loopy/kernel.py
+++ b/loopy/kernel.py
@@ -113,9 +113,9 @@ def parse_tag(tag):
# {{{ arguments
-class ArrayArg:
+class _ShapedArg:
def __init__(self, name, dtype, strides=None, shape=None, order="C",
- offset=0, constant_mem=False):
+ offset=0):
"""
All of the following are optional. Specify either strides or shape.
@@ -170,18 +170,25 @@ class ArrayArg:
self.shape = shape
self.order = order
- self.constant_mem = constant_mem
-
@property
def dimensions(self):
return len(self.shape)
+class GlobalArg(_ShapedArg):
def __repr__(self):
- return "<ArrayArg '%s' of type %s and shape (%s)>" % (
+ return "<GlobalArg '%s' of type %s and shape (%s)>" % (
self.name, self.dtype, ",".join(str(i) for i in self.shape))
-class ConstantArrayArg(ArrayArg):
- pass
+class ArrayArg(GlobalArg):
+ def __init__(self, *args, **kwargs):
+ from warnings import warn
+ warn("ArrayArg is a deprecated name of GlobalArg")
+ GlobalArg.__init__(self, *args, **kwargs)
+
+class ConstantArg(_ShapedArg):
+ def __repr__(self):
+ return "<ConstantArg '%s' of type %s and shape (%s)>" % (
+ self.name, self.dtype, ",".join(str(i) for i in self.shape))
class ImageArg:
def __init__(self, name, dtype, dimensions=None, shape=None):
diff --git a/test/test_fem_assembly.py b/test/test_fem_assembly.py
index d5e439514..7d71d76c7 100644
--- a/test/test_fem_assembly.py
+++ b/test/test_fem_assembly.py
@@ -35,11 +35,11 @@ def test_laplacian_stiffness(ctx_factory):
"sum_float32(dx_axis, dPsi$one(i,dx_axis)*dPsi$two(j,dx_axis))))"
],
[
- lp.ArrayArg("jacInv", dtype, shape=(dim, dim, Nc_sym, Nq), order=order),
- lp.ConstantArrayArg("DPsi", dtype, shape=(dim, Nb, Nq), order=order),
- lp.ArrayArg("jacDet", dtype, shape=(Nc_sym, Nq), order=order),
- lp.ConstantArrayArg("w", dtype, shape=(Nq,), order=order),
- lp.ArrayArg("A", dtype, shape=(Nc_sym, Nb, Nb), order=order),
+ lp.GlobalArg("jacInv", dtype, shape=(dim, dim, Nc_sym, Nq), order=order),
+ lp.ConstantArg("DPsi", dtype, shape=(dim, Nb, Nq), order=order),
+ lp.GlobalArg("jacDet", dtype, shape=(Nc_sym, Nq), order=order),
+ lp.ConstantArg("w", dtype, shape=(Nq,), order=order),
+ lp.GlobalArg("A", dtype, shape=(Nc_sym, Nb, Nb), order=order),
lp.ScalarArg("Nc", np.int32, approximately=1000),
],
name="lapquad", assumptions="Nc>=1")
diff --git a/test/test_linalg.py b/test/test_linalg.py
index ffd99a6c7..a81094336 100644
--- a/test/test_linalg.py
+++ b/test/test_linalg.py
@@ -111,10 +111,10 @@ def test_axpy(ctx_factory):
],
[
lp.ScalarArg("a", dtype),
- lp.ArrayArg("x", dtype, shape="n,"),
+ lp.GlobalArg("x", dtype, shape="n,"),
lp.ScalarArg("b", dtype),
- lp.ArrayArg("y", dtype, shape="n,"),
- lp.ArrayArg("z", dtype, shape="n,"),
+ lp.GlobalArg("y", dtype, shape="n,"),
+ lp.GlobalArg("z", dtype, shape="n,"),
lp.ScalarArg("n", np.int32, approximately=n),
],
name="axpy", assumptions="n>=1",
@@ -163,8 +163,8 @@ def test_transpose(ctx_factory):
"b[i, j] = a[j, i]"
],
[
- lp.ArrayArg("a", dtype, shape=(n, n), order=order),
- lp.ArrayArg("b", dtype, shape=(n, n), order=order),
+ lp.GlobalArg("a", dtype, shape=(n, n), order=order),
+ lp.GlobalArg("b", dtype, shape=(n, n), order=order),
],
name="transpose")
@@ -202,9 +202,9 @@ def test_plain_matrix_mul(ctx_factory):
"c[i, j] = %s(k, a[i, k]*b[k, j])" % reduction_func
],
[
- lp.ArrayArg("a", dtype, shape=(n, n), order=order),
- lp.ArrayArg("b", dtype, shape=(n, n), order=order),
- lp.ArrayArg("c", dtype, shape=(n, n), order=order),
+ lp.GlobalArg("a", dtype, shape=(n, n), order=order),
+ lp.GlobalArg("b", dtype, shape=(n, n), order=order),
+ lp.GlobalArg("c", dtype, shape=(n, n), order=order),
],
name="matmul")
@@ -244,9 +244,9 @@ def test_variable_size_matrix_mul(ctx_factory):
"label: c[i, j] = sum_float32(k, a[i, k]*b[k, j])"
],
[
- lp.ArrayArg("a", dtype, shape=(n, n), order=order),
- lp.ArrayArg("b", dtype, shape=(n, n), order=order),
- lp.ArrayArg("c", dtype, shape=(n, n), order=order),
+ lp.GlobalArg("a", dtype, shape=(n, n), order=order),
+ lp.GlobalArg("b", dtype, shape=(n, n), order=order),
+ lp.GlobalArg("c", dtype, shape=(n, n), order=order),
lp.ScalarArg("n", np.int32, approximately=n),
],
name="matmul", assumptions="n >= 16")
@@ -295,9 +295,9 @@ def test_rank_one(ctx_factory):
"label: c[i, j] = a[i]*b[j]"
],
[
- lp.ArrayArg("a", dtype, shape=(n,), order=order),
- lp.ArrayArg("b", dtype, shape=(n,), order=order),
- lp.ArrayArg("c", dtype, shape=(n, n), order=order),
+ lp.GlobalArg("a", dtype, shape=(n,), order=order),
+ lp.GlobalArg("b", dtype, shape=(n,), order=order),
+ lp.GlobalArg("c", dtype, shape=(n, n), order=order),
lp.ScalarArg("n", np.int32, approximately=n),
],
name="rank_one", assumptions="n >= 16")
@@ -376,9 +376,9 @@ def test_troublesome_premagma_fermi_matrix_mul(ctx_factory):
"c[i, j] = sum_float32(k, a[i, k]*b[k, j])"
],
[
- lp.ArrayArg("a", dtype, shape=(n, n), order=order),
- lp.ArrayArg("b", dtype, shape=(n, n), order=order),
- lp.ArrayArg("c", dtype, shape=(n, n), order=order),
+ lp.GlobalArg("a", dtype, shape=(n, n), order=order),
+ lp.GlobalArg("b", dtype, shape=(n, n), order=order),
+ lp.GlobalArg("c", dtype, shape=(n, n), order=order),
],
name="matmul")
@@ -430,9 +430,9 @@ def test_intel_matrix_mul(ctx_factory):
"c[i, j] = sum_float32(k, a[i, k]*b[k, j])"
],
[
- lp.ArrayArg("a", dtype, shape=(n, n), order=order),
- lp.ArrayArg("b", dtype, shape=(n, n), order=order),
- lp.ArrayArg("c", dtype, shape=(n, n), order=order),
+ lp.GlobalArg("a", dtype, shape=(n, n), order=order),
+ lp.GlobalArg("b", dtype, shape=(n, n), order=order),
+ lp.GlobalArg("c", dtype, shape=(n, n), order=order),
],
name="matmul")
@@ -497,7 +497,7 @@ def test_magma_fermi_matrix_mul(ctx_factory):
[
lp.ImageArg("a", dtype, 2),
lp.ImageArg("b", dtype, 2),
- lp.ArrayArg("c", dtype, shape=(n, n), order=order),
+ lp.GlobalArg("c", dtype, shape=(n, n), order=order),
],
name="matmul")
@@ -557,7 +557,7 @@ def test_image_matrix_mul(ctx_factory):
[
lp.ImageArg("a", dtype, 2),
lp.ImageArg("b", dtype, 2),
- lp.ArrayArg("c", dtype, shape=(n, n), order=order),
+ lp.GlobalArg("c", dtype, shape=(n, n), order=order),
],
name="matmul")
@@ -607,7 +607,7 @@ def test_image_matrix_mul_ilp(ctx_factory):
[
lp.ImageArg("a", dtype, shape=(n, n)),
lp.ImageArg("b", dtype, shape=(n, n)),
- lp.ArrayArg("c", dtype, shape=(n, n), order=order),
+ lp.GlobalArg("c", dtype, shape=(n, n), order=order),
],
name="matmul")
@@ -650,9 +650,9 @@ def test_fancy_matrix_mul(ctx_factory):
"c[i, j] = sum_float32(k, a[i, k]*b[k, j])"
],
[
- lp.ArrayArg("a", dtype, shape="(n, n)", order=order),
- lp.ArrayArg("b", dtype, shape="(n, n)", order=order),
- lp.ArrayArg("c", dtype, shape="(n, n)", order=order),
+ lp.GlobalArg("a", dtype, shape="(n, n)", order=order),
+ lp.GlobalArg("b", dtype, shape="(n, n)", order=order),
+ lp.GlobalArg("c", dtype, shape="(n, n)", order=order),
lp.ScalarArg("n", np.int32, approximately=1000),
], name="fancy_matmul", assumptions="n>=1")
diff --git a/test/test_loopy.py b/test/test_loopy.py
index 2eeea576f..d412b4861 100644
--- a/test/test_loopy.py
+++ b/test/test_loopy.py
@@ -22,7 +22,7 @@ def test_owed_barriers(ctx_factory):
[
"[i:l.0] <float32> z[i] = a[i]"
],
- [lp.ArrayArg("a", np.float32, shape=(100,))]
+ [lp.GlobalArg("a", np.float32, shape=(100,))]
)
kernel_gen = lp.generate_loop_schedules(knl)
@@ -43,7 +43,7 @@ def test_wg_too_small(ctx_factory):
[
"[i:l.0] <float32> z[i] = a[i]"
],
- [lp.ArrayArg("a", np.float32, shape=(100,))],
+ [lp.GlobalArg("a", np.float32, shape=(100,))],
local_sizes={0: 16})
kernel_gen = lp.generate_loop_schedules(knl)
@@ -69,7 +69,7 @@ def test_multi_cse(ctx_factory):
[
"[i] <float32> z[i] = a[i] + a[i]**2"
],
- [lp.ArrayArg("a", np.float32, shape=(100,))],
+ [lp.GlobalArg("a", np.float32, shape=(100,))],
local_sizes={0: 16})
knl = lp.split_dimension(knl, "i", 16, inner_tag="l.0")
@@ -99,7 +99,7 @@ def test_stencil(ctx_factory):
" + a[i+1,j]"
],
[
- lp.ArrayArg("a", np.float32, shape=(32,32,))
+ lp.GlobalArg("a", np.float32, shape=(32,32,))
])
@@ -134,8 +134,8 @@ def test_eq_constraint(ctx_factory):
"a[i] = b[i]"
],
[
- lp.ArrayArg("a", np.float32, shape=(1000,)),
- lp.ArrayArg("b", np.float32, shape=(1000,))
+ lp.GlobalArg("a", np.float32, shape=(1000,)),
+ lp.GlobalArg("b", np.float32, shape=(1000,))
])
knl = lp.split_dimension(knl, "i", 16, outer_tag="g.0")
diff --git a/test/test_nbody.py b/test/test_nbody.py
index f1641256f..13cfc02ba 100644
--- a/test/test_nbody.py
+++ b/test/test_nbody.py
@@ -21,8 +21,8 @@ def test_nbody(ctx_factory):
"pot[i] = sum_float32(j, if(i != j, invdist, 0))",
],
[
- lp.ArrayArg("x", dtype, shape="N,3", order="C"),
- lp.ArrayArg("pot", dtype, shape="N", order="C"),
+ lp.GlobalArg("x", dtype, shape="N,3", order="C"),
+ lp.GlobalArg("pot", dtype, shape="N", order="C"),
lp.ScalarArg("N", np.int32),
],
name="nbody", assumptions="N>=1")
diff --git a/test/test_sem.py b/test/test_sem.py
index 8a93d11c0..6ee04953d 100644
--- a/test/test_sem.py
+++ b/test/test_sem.py
@@ -52,10 +52,10 @@ def test_laplacian(ctx_factory):
"+ sum_float32(m, D[m,k]*Gw(i,j,m))"
],
[
- lp.ArrayArg("u", dtype, shape=field_shape, order=order),
- lp.ArrayArg("lap", dtype, shape=field_shape, order=order),
- lp.ArrayArg("G", dtype, shape=(6,)+field_shape, order=order),
- lp.ArrayArg("D", dtype, shape=(n, n), order=order),
+ lp.GlobalArg("u", dtype, shape=field_shape, order=order),
+ lp.GlobalArg("lap", dtype, shape=field_shape, order=order),
+ lp.GlobalArg("G", dtype, shape=(6,)+field_shape, order=order),
+ lp.GlobalArg("D", dtype, shape=(n, n), order=order),
lp.ScalarArg("K", np.int32, approximately=1000),
],
name="semlap", assumptions="K>=1")
@@ -140,10 +140,10 @@ def test_laplacian_lmem(ctx_factory):
"+ sum_float32(m, D[m,k]*(G[2,e,i,j,m]*ur(i,j,m) + G[4,e,i,j,m]*us(i,j,m) + G[5,e,i,j,m]*ut(i,j,m)))"
],
[
- lp.ArrayArg("u", dtype, shape=field_shape, order=order),
- lp.ArrayArg("lap", dtype, shape=field_shape, order=order),
- lp.ArrayArg("G", dtype, shape=(6,)+field_shape, order=order),
- lp.ArrayArg("D", dtype, shape=(n, n), order=order),
+ lp.GlobalArg("u", dtype, shape=field_shape, order=order),
+ lp.GlobalArg("lap", dtype, shape=field_shape, order=order),
+ lp.GlobalArg("G", dtype, shape=(6,)+field_shape, order=order),
+ lp.GlobalArg("D", dtype, shape=(n, n), order=order),
lp.ScalarArg("K", np.int32, approximately=1000),
],
name="semlap", assumptions="K>=1")
@@ -215,10 +215,10 @@ def test_laplacian_lmem_ilp(ctx_factory):
"+ sum_float32(m, D[m,k]*(G[2,e,i,j,m]*ur(i,j,m) + G[4,e,i,j,m]*us(i,j,m) + G[5,e,i,j,m]*ut(i,j,m)))"
],
[
- lp.ArrayArg("u", dtype, shape=field_shape, order=order),
- lp.ArrayArg("lap", dtype, shape=field_shape, order=order),
- lp.ArrayArg("G", dtype, shape=(6,)+field_shape, order=order),
- lp.ArrayArg("D", dtype, shape=(n, n), order=order),
+ lp.GlobalArg("u", dtype, shape=field_shape, order=order),
+ lp.GlobalArg("lap", dtype, shape=field_shape, order=order),
+ lp.GlobalArg("G", dtype, shape=(6,)+field_shape, order=order),
+ lp.GlobalArg("D", dtype, shape=(n, n), order=order),
lp.ScalarArg("K", np.int32, approximately=1000),
],
name="semlap", assumptions="K>=1")
@@ -313,14 +313,14 @@ def test_advect(ctx_factory):
"Nw[e,i,j,k] = Vr(i,j,k)*wr(i,j,k)+Vs(i,j,k)*ws(i,j,k)+Vt(i,j,k)*wt(i,j,k)",
],
[
- lp.ArrayArg("u", dtype, shape=field_shape, order=order),
- lp.ArrayArg("v", dtype, shape=field_shape, order=order),
- lp.ArrayArg("w", dtype, shape=field_shape, order=order),
- lp.ArrayArg("Nu", dtype, shape=field_shape, order=order),
- lp.ArrayArg("Nv", dtype, shape=field_shape, order=order),
- lp.ArrayArg("Nw", dtype, shape=field_shape, order=order),
- lp.ArrayArg("G", dtype, shape=(9,)+field_shape, order=order),
- lp.ArrayArg("D", dtype, shape=(N, N), order=order),
+ lp.GlobalArg("u", dtype, shape=field_shape, order=order),
+ lp.GlobalArg("v", dtype, shape=field_shape, order=order),
+ lp.GlobalArg("w", dtype, shape=field_shape, order=order),
+ lp.GlobalArg("Nu", dtype, shape=field_shape, order=order),
+ lp.GlobalArg("Nv", dtype, shape=field_shape, order=order),
+ lp.GlobalArg("Nw", dtype, shape=field_shape, order=order),
+ lp.GlobalArg("G", dtype, shape=(9,)+field_shape, order=order),
+ lp.GlobalArg("D", dtype, shape=(N, N), order=order),
lp.ScalarArg("K", np.int32, approximately=1000),
],
name="sem_advect", assumptions="K>=1")
@@ -432,15 +432,15 @@ def test_advect_dealias(ctx_factory):
],
[
- lp.ArrayArg("u", dtype, shape=field_shape, order=order),
- lp.ArrayArg("v", dtype, shape=field_shape, order=order),
- lp.ArrayArg("w", dtype, shape=field_shape, order=order),
- lp.ArrayArg("INu", dtype, shape=field_shape, order=order),
- lp.ArrayArg("INv", dtype, shape=field_shape, order=order),
- lp.ArrayArg("INw", dtype, shape=field_shape, order=order),
- lp.ArrayArg("D", dtype, shape=(M,M), order=order),
- lp.ArrayArg("I", dtype, shape=(M, N), order=order),
- lp.ArrayArg("V", dtype, shape=(N, M), order=order),
+ lp.GlobalArg("u", dtype, shape=field_shape, order=order),
+ lp.GlobalArg("v", dtype, shape=field_shape, order=order),
+ lp.GlobalArg("w", dtype, shape=field_shape, order=order),
+ lp.GlobalArg("INu", dtype, shape=field_shape, order=order),
+ lp.GlobalArg("INv", dtype, shape=field_shape, order=order),
+ lp.GlobalArg("INw", dtype, shape=field_shape, order=order),
+ lp.GlobalArg("D", dtype, shape=(M,M), order=order),
+ lp.GlobalArg("I", dtype, shape=(M, N), order=order),
+ lp.GlobalArg("V", dtype, shape=(N, M), order=order),
lp.ScalarArg("K", np.int32, approximately=1000),
],
name="sem_advect", assumptions="K>=1")
@@ -500,11 +500,11 @@ def test_interp_diff(ctx_factory):
"Pu[ip,jp,kp,e] = sum_float32(i, V[ip,i]*Pu[i ,jp,kp,e])",
],
[
- lp.ArrayArg("u", dtype, shape=field_shape, order=order),
- lp.ArrayArg("P", dtype, shape=interim_field_shape, order=order),
- lp.ArrayArg("I", dtype, shape=(M, N), order=order),
- lp.ArrayArg("V", dtype, shape=(N, M), order=order),
- lp.ArrayArg("Pu", dtype, shape=field_shape, order=order),
+ lp.GlobalArg("u", dtype, shape=field_shape, order=order),
+ lp.GlobalArg("P", dtype, shape=interim_field_shape, order=order),
+ lp.GlobalArg("I", dtype, shape=(M, N), order=order),
+ lp.GlobalArg("V", dtype, shape=(N, M), order=order),
+ lp.GlobalArg("Pu", dtype, shape=field_shape, order=order),
lp.ScalarArg("K", np.int32, approximately=1000),
],
name="sem_lap_precon", assumptions="K>=1")
--
GitLab