From 5eedbe24a4b588927fdc29e27619e6f053d865e1 Mon Sep 17 00:00:00 2001
From: Andreas Kloeckner <inform@tiker.net>
Date: Sun, 23 Sep 2012 10:19:30 -0500
Subject: [PATCH] Support instructions all in a single string.
---
loopy/kernel.py | 30 +++++++++++++++++++++---------
test/test_linalg.py | 14 ++++++++------
2 files changed, 29 insertions(+), 15 deletions(-)
diff --git a/loopy/kernel.py b/loopy/kernel.py
index 5add79c78..d1eb4eddd 100644
--- a/loopy/kernel.py
+++ b/loopy/kernel.py
@@ -902,13 +902,23 @@ class LoopKernel(Record):
"instance or a parseable string. got '%s' instead."
% type(insn))
- for insn in expand_defines(insn, defines, single_valued=False):
- parse_insn(insn)
+ for insn in insn.split("\n"):
+ insn = insn.strip()
+
+ if not insn:
+ continue
+ if insn.startswith("#"):
+ continue
+
+ for sub_insn in expand_defines(insn, defines, single_valued=False):
+ parse_insn(sub_insn)
parsed_instructions = []
substitutions = substitutions.copy()
+ if isinstance(instructions, str):
+ instructions = [instructions]
for insn in instructions:
# must construct list one-by-one to facilitate unique id generation
parse_if_necessary(insn)
@@ -979,13 +989,15 @@ class LoopKernel(Record):
processed_args = []
for arg in args:
- if isinstance(arg, _ShapedArg):
- if arg.shape is not None:
- arg = arg.copy(shape=expand_defines_in_expr(arg.shape, defines))
- if arg.strides is not None:
- arg = arg.copy(strides=expand_defines_in_expr(arg.strides, defines))
-
- processed_args.append(arg)
+ for name in arg.name.split(","):
+ new_arg = arg.copy(name=name)
+ if isinstance(arg, _ShapedArg):
+ if arg.shape is not None:
+ new_arg = new_arg.copy(shape=expand_defines_in_expr(arg.shape, defines))
+ if arg.strides is not None:
+ new_arg = new_arg.copy(strides=expand_defines_in_expr(arg.strides, defines))
+
+ processed_args.append(new_arg)
# }}}
diff --git a/test/test_linalg.py b/test/test_linalg.py
index 84878d972..b6087f442 100644
--- a/test/test_linalg.py
+++ b/test/test_linalg.py
@@ -273,9 +273,10 @@ def test_variable_size_matrix_mul(ctx_factory):
def test_rank_one(ctx_factory):
dtype = np.float32
ctx = ctx_factory()
- order = "C"
+ order = "F"
- n = int(get_suitable_size(ctx)**(2.7/2))
+ #n = int(get_suitable_size(ctx)**(2.7/2))
+ n = 16**3
knl = lp.make_kernel(ctx.devices[0],
"[n] -> {[i,j]: 0<=i,j<n}",
@@ -288,7 +289,8 @@ def test_rank_one(ctx_factory):
lp.GlobalArg("c", dtype, shape=("n, n"), order=order),
lp.ValueArg("n", np.int32, approximately=n),
],
- name="rank_one", assumptions="n >= 16")
+ name="rank_one",
+ assumptions="n >= 16")
def variant_1(knl):
knl = lp.add_prefetch(knl, "a")
@@ -337,15 +339,15 @@ def test_rank_one(ctx_factory):
seq_knl = knl
- for variant in [variant_1, variant_2, variant_3, variant_4]:
- #for variant in [variant_1]:
+ #for variant in [variant_1, variant_2, variant_3, variant_4]:
+ for variant in [variant_4]:
kernel_gen = lp.generate_loop_schedules(variant(knl),
loop_priority=["i", "j"])
kernel_gen = lp.check_kernels(kernel_gen, dict(n=n))
lp.auto_test_vs_ref(seq_knl, ctx, kernel_gen,
op_count=[np.dtype(dtype).itemsize*n**2/1e9], op_label=["GBytes"],
- parameters={"n": n})
+ parameters={"n": n}, edit_code=True, do_check=False)
--
GitLab