From 0a34fa88bc63965697983817f6619339b146cc34 Mon Sep 17 00:00:00 2001
From: Markus Holzer <markus.holzer@fau.de>
Date: Thu, 10 Feb 2022 14:26:33 +0100
Subject: [PATCH] Debug checking
---
pystencils/cpu/vectorization.py | 2 ++
pystencils_tests/test_vectorization_specific.py | 5 +++--
2 files changed, 5 insertions(+), 2 deletions(-)
diff --git a/pystencils/cpu/vectorization.py b/pystencils/cpu/vectorization.py
index 6d59be9b0..9f8e381c3 100644
--- a/pystencils/cpu/vectorization.py
+++ b/pystencils/cpu/vectorization.py
@@ -77,6 +77,8 @@ class CachelineSize(ast.Node):
def vectorize(kernel_ast: ast.KernelFunction, instruction_set: str = 'best',
assume_aligned: bool = False, nontemporal: Union[bool, Container[Union[str, Field]]] = False,
assume_inner_stride_one: bool = False, assume_sufficient_line_padding: bool = True):
+ # TODO we first introduce the remainder loop and then check if we can even vectorise. Maybe first copy the ast
+ # and return the copied version on failure
"""Explicit vectorization using SIMD vectorization via intrinsics.
Args:
diff --git a/pystencils_tests/test_vectorization_specific.py b/pystencils_tests/test_vectorization_specific.py
index 58a1183d3..c6deb2de5 100644
--- a/pystencils_tests/test_vectorization_specific.py
+++ b/pystencils_tests/test_vectorization_specific.py
@@ -81,9 +81,10 @@ def test_strided(instruction_set, dtype):
ps.show_code(ast)
func = ast.compile()
- ref_func = ps.create_kernel(update_rule).compile()
+ ref_config = pystencils.config.CreateKernelConfig(default_number_float=type_string)
+ ref_func = ps.create_kernel(update_rule, config=ref_config).compile()
- arr = np.random.random((23 + 2, 17 + 2)).astype(npdtype)
+ arr = np.full((23 + 2, 17 + 2), 0.13).astype(npdtype)
print("sum arr: ", np.sum(arr))
# print("arr type: ", arr.dtype)
dst = np.zeros_like(arr, dtype=npdtype)
--
GitLab