diff --git a/.gitignore b/.gitignore
index 5ec93d6ef6776bfa9199408192cd3403c15c6191..611d3b46e31d4c0c5d2dff5a826d0a64aa1f1c49 100644
--- a/.gitignore
+++ b/.gitignore
@@ -4,3 +4,4 @@ build/
venv/
pystencils_benchmark.egg-info/
__pycache__
+report.xml
diff --git a/pystencils_benchmark/benchmark.py b/pystencils_benchmark/benchmark.py
index 25c90442b4a0c574af8e9377bc50f07a92a2fb38..67cc3d58a911371e9db8ce8374cf69c36b8bffdf 100644
--- a/pystencils_benchmark/benchmark.py
+++ b/pystencils_benchmark/benchmark.py
@@ -15,6 +15,10 @@ from pystencils.integer_functions import modulo_ceil
from pystencils_benchmark.enums import Compiler
+_env = Environment(loader=PackageLoader('pystencils_benchmark'), undefined=StrictUndefined, keep_trailing_newline=True,
+ trim_blocks=True, lstrip_blocks=True)
+
+
def generate_benchmark(kernel_asts: Union[KernelFunction, List[KernelFunction]],
path: Path = None,
*,
@@ -55,11 +59,10 @@ def compiler_toolchain(path: Path, compiler: Compiler) -> None:
'compiler': name,
}
- env = Environment(loader=PackageLoader('pystencils_benchmark'), undefined=StrictUndefined)
files = ['Makefile', f'{name}.mk']
for file_name in files:
with open(path / file_name, 'w+') as f:
- template = env.get_template(file_name).render(**jinja_context)
+ template = _env.get_template(file_name).render(**jinja_context)
f.write(template)
@@ -69,10 +72,9 @@ def copy_static_files(path: Path) -> None:
include_path = path / 'include'
include_path.mkdir(parents=True, exist_ok=True)
- env = Environment(loader=PackageLoader('pystencils_benchmark'), undefined=StrictUndefined)
- files = ['aligned_malloc.h', 'timing.h', 'timing.c']
+ files = ['timing.h', 'timing.c']
for file_name in files:
- template = env.get_template(file_name).render()
+ template = _env.get_template(file_name).render()
if file_name[-1] == 'h':
target_path = include_path / file_name
elif file_name[-1] == 'c':
@@ -113,6 +115,7 @@ def kernel_main(kernels_ast: List[KernelFunction], timing: bool = True):
np_dtype = get_base_type(p.symbol.dtype).numpy_dtype
size_data_type = np_dtype.itemsize
+ # TODO double check the size computation
dim0_size = field.shape[-1]
dim1_size = np.prod(field.shape[:-1])
elements = prod(field.shape)
@@ -127,17 +130,16 @@ def kernel_main(kernels_ast: List[KernelFunction], timing: bool = True):
assert align % np_dtype.itemsize == 0
offset = ((dim0_size + padding_elements + ghost_layers) % kernel.instruction_set['width']) * size_data_type
-
- fields.append((p.field_name, dtype, elements, size, offset, align))
- call_parameters.append(p.field_name)
else:
size = elements * size_data_type
- fields.append((p.field_name, dtype, elements, size, 0, 0))
- call_parameters.append(p.field_name)
+ offset = 0
+ align = 0
+ fields.append((p.field_name, dtype, elements, size, offset, align))
+ call_parameters.append(p.field_name)
kernels.append(Kernel(name=name, fields=fields, constants=constants, call_parameters=call_parameters,
call_argument_list=",".join(call_parameters)))
- includes.add(f'#include "{name}.h"\n')
+ includes.add(name)
jinja_context = {
'kernels': kernels,
@@ -145,8 +147,7 @@ def kernel_main(kernels_ast: List[KernelFunction], timing: bool = True):
'timing': timing,
}
- env = Environment(loader=PackageLoader('pystencils_benchmark'), undefined=StrictUndefined)
- main = env.get_template('main.c').render(**jinja_context)
+ main = _env.get_template('main.c').render(**jinja_context)
return main
@@ -159,8 +160,7 @@ def kernel_header(kernel_ast: KernelFunction, dialect: Backend = Backend.C) -> s
'function_signature': function_signature,
}
- env = Environment(loader=PackageLoader('pystencils_benchmark'), undefined=StrictUndefined)
- header = env.get_template('kernel.h').render(**jinja_context)
+ header = _env.get_template('kernel.h').render(**jinja_context)
return header
@@ -176,6 +176,5 @@ def kernel_source(kernel_ast: KernelFunction, dialect: Backend = Backend.C) -> s
'timing': True,
}
- env = Environment(loader=PackageLoader('pystencils_benchmark'), undefined=StrictUndefined)
- source = env.get_template('kernel.c').render(**jinja_context)
+ source = _env.get_template('kernel.c').render(**jinja_context)
return source
diff --git a/pystencils_benchmark/templates/Clang.mk b/pystencils_benchmark/templates/Clang.mk
index be60d3812bde44525848805090dd73aaad707a1d..96296cc6d0aff6545ee8457aa601c86265b868a4 100644
--- a/pystencils_benchmark/templates/Clang.mk
+++ b/pystencils_benchmark/templates/Clang.mk
@@ -2,11 +2,11 @@ CC = clang
LINKER = $(CC)
ANSI_CFLAGS = -ansi
-ANSI_CFLAGS += -std=c99
+ANSI_CFLAGS += -std=c11
ANSI_CFLAGS += -pedantic
ANSI_CFLAGS += -Wextra
-CFLAGS = -O3 -Wno-format -Wall $(ANSI_CFLAGS) -fopenmp -march=native
+CFLAGS = -O3 -Wno-format -Wall -Werror $(ANSI_CFLAGS) -fopenmp -march=native
# More warning pls
#CFLAGS += -Wfloat-equal -Wundef -Wshadow -Wpointer-arith -Wcast-align -Wstrict-overflow=5 -Wwrite-strings -Waggregate-return
# Maybe too much warnings
diff --git a/pystencils_benchmark/templates/GCC.mk b/pystencils_benchmark/templates/GCC.mk
index 46668e6b9d61f6d936ea315c998bfd36ab548933..a65ec464c663bfe23a90e5756203ede9eeb4d8de 100644
--- a/pystencils_benchmark/templates/GCC.mk
+++ b/pystencils_benchmark/templates/GCC.mk
@@ -2,11 +2,11 @@ CC = gcc
LINKER = $(CC)
ANSI_CFLAGS = -ansi
-ANSI_CFLAGS += -std=c99
+ANSI_CFLAGS += -std=c11
ANSI_CFLAGS += -pedantic
ANSI_CFLAGS += -Wextra
-CFLAGS = -O3 -Wno-format -Wall $(ANSI_CFLAGS) -fopenmp -march=native
+CFLAGS = -O3 -Wno-format -Wall -Werror $(ANSI_CFLAGS) -fopenmp -march=native
# More warning pls
#CFLAGS += -Wfloat-equal -Wundef -Wshadow -Wpointer-arith -Wcast-align -Wstrict-overflow=5 -Wwrite-strings -Waggregate-return
# Maybe too much warnings
diff --git a/pystencils_benchmark/templates/GCCdebug.mk b/pystencils_benchmark/templates/GCCdebug.mk
index 47b79c32d75c3ecc0d015110b7cd11b280ee6918..e0d4dd000c00c06f33d156be474d8435bbf77c90 100644
--- a/pystencils_benchmark/templates/GCCdebug.mk
+++ b/pystencils_benchmark/templates/GCCdebug.mk
@@ -2,11 +2,11 @@ CC = gcc
LINKER = $(CC)
ANSI_CFLAGS = -ansi
-ANSI_CFLAGS += -std=c99
+ANSI_CFLAGS += -std=c11
ANSI_CFLAGS += -pedantic
ANSI_CFLAGS += -Wextra
-CFLAGS = -O0 -g -Wno-format -Wall $(ANSI_CFLAGS) -march=native
+CFLAGS = -O0 -g -Wno-format -Wall -Werror $(ANSI_CFLAGS) -march=native
LFLAGS =
DEFINES = -D_GNU_SOURCE
INCLUDES =
diff --git a/pystencils_benchmark/templates/aligned_malloc.h b/pystencils_benchmark/templates/aligned_malloc.h
deleted file mode 100644
index 52693f9b7fffd2a3108574088d9112a1028f7fe5..0000000000000000000000000000000000000000
--- a/pystencils_benchmark/templates/aligned_malloc.h
+++ /dev/null
@@ -1,19 +0,0 @@
-#ifndef _ALIGNED_MALLOC_H
-#define _ALIGNED_MALLOC_H
-#include <stddef.h>
-#include <stdlib.h>
-
-inline void* aligned_malloc(size_t size, size_t align) {
- // Based on http://stackoverflow.com/q/16376942
- void *result;
- #if defined(_MSC_VER)
- result = _aligned_malloc(size, align);
- #elif defined(__INTEL_COMPILER)
- result = _mm_malloc(size, align);
- #else
- if(posix_memalign(&result, align, size)) result = 0;
- #endif
- return result;
-}
-
-#endif
diff --git a/pystencils_benchmark/templates/kernel.c b/pystencils_benchmark/templates/kernel.c
index 772321ac3fd822c40e68c528d652987008b49b7b..9370baac95669d61074a4236e985a5eff22cff56 100644
--- a/pystencils_benchmark/templates/kernel.c
+++ b/pystencils_benchmark/templates/kernel.c
@@ -1,4 +1,5 @@
-{% for header in headers %}#include {{header}}
+{% for header in headers %}
+#include {{header}}
{% endfor %}
#define RESTRICT __restrict__
diff --git a/pystencils_benchmark/templates/main.c b/pystencils_benchmark/templates/main.c
index 1eea7c1e92d232f68271dad5c14fb769c0985220..a4c462d598ba4c88ad0f63c0afb66d1962c07b7f 100644
--- a/pystencils_benchmark/templates/main.c
+++ b/pystencils_benchmark/templates/main.c
@@ -1,16 +1,16 @@
-#include <stdlib.h>
-#include <stdint.h>
-#include <stdbool.h>
+#include <assert.h>
#include <math.h>
+#include <stdbool.h>
+#include <stdint.h>
#include <stdio.h>
-#include <assert.h>
+#include <stdlib.h>
#include "timing.h"
-#include "aligned_malloc.h"
-{%- for include in includes %}
-{{ include }}
-{%- endfor %}
+//kernels
+{% for include in includes %}
+#include "{{ include }}.h"
+{% endfor %}
#define RESTRICT __restrict__
#define FUNC_PREFIX
@@ -22,24 +22,25 @@ int main(int argc, char **argv)
return -1;
}
int n_repeat = atoi(argv[1]);
- {%- for kernel in kernels %}
- {
- {%- for field_name, dataType, elements, size, offset, alignment in kernel.fields %}
+ {% for kernel in kernels %}
+
+ { // Kernel: {{kernel.name}}
+ {% for field_name, dataType, elements, size, offset, alignment in kernel.fields %}
// Initialization {{field_name}}
- {%- if alignment > 0 %}
- {{dataType}} * {{field_name}} = ({{dataType}} *) aligned_malloc({{size}}, {{alignment}});//, {{offset}});
- {%- else %}
- {{dataType}} * {{field_name}} = ({{dataType}} *) malloc({{size}});
- {%- endif %}
+ {% if alignment > 0 %}
+ {{dataType}}* {{field_name}} = ({{dataType}} *) aligned_alloc({{alignment}}, {{size}});//, {{offset}});
+ {% else %}
+ {{dataType}}* {{field_name}} = ({{dataType}} *) malloc({{size}});
+ {% endif %}
for (unsigned long long i = 0; i < {{elements}}; ++i)
{{field_name}}[i] = 0.23;
- {%- endfor %}
+ {% endfor %}
- {%- for constantName, dataType in kernel.constants %}
+ {% for constantName, dataType in kernel.constants %}
// Constant {{constantName}}
{{dataType}} {{constantName}};
{{constantName}} = 0.23;
- {%- endfor %}
+ {% endfor %}
for(int warmup = 1; warmup >= 0; --warmup) {
int repeat = 2;
@@ -47,25 +48,26 @@ int main(int argc, char **argv)
repeat = n_repeat;
}
- {%- if timing %}
+ {% if timing %}
double wcStartTime, cpuStartTime, wcEndTime, cpuEndTime;
timing(&wcStartTime, &cpuStartTime);
- {%- endif %}
+ {% endif %}
for (; repeat > 0; --repeat)
{
{{kernel.name}}({{kernel.call_argument_list}});
}
- {%- if timing %}
+
+ {% if timing %}
timing(&wcEndTime, &cpuEndTime);
+
if( warmup == 0)
printf("%s\t%e\n", "{{kernel.name}}",(wcEndTime - wcStartTime) / n_repeat );
- {%- endif %}
+ {% endif %}
}
-
- {%- for field_name, dataType, elements, size, offset, alignment in kernel.fields %}
+ {% for field_name, dataType, elements, size, offset, alignment in kernel.fields %}
free({{field_name}});
- {%- endfor %}
+ {% endfor %}
}
- {%- endfor %}
+ {% endfor %}
}
diff --git a/pystencils_benchmark/templates/timing.h b/pystencils_benchmark/templates/timing.h
index 6c6ff440cfb40ab0aa5271a9874de73cfd47ed70..ed34c92c454b0ef136f6dfd70b93c32c27f80df4 100644
--- a/pystencils_benchmark/templates/timing.h
+++ b/pystencils_benchmark/templates/timing.h
@@ -9,4 +9,4 @@
void timing(double* wcTime, double* cpuTime);
-#endif
\ No newline at end of file
+#endif
diff --git a/tests/test_benchmark.py b/tests/test_benchmark.py
index ea2b3bc656f7b298249966a6b040fe2fa140d0f6..31e115e80e95325784bcf281ae848cf98c9f870e 100755
--- a/tests/test_benchmark.py
+++ b/tests/test_benchmark.py
@@ -2,30 +2,37 @@ import subprocess
import numpy as np
import sympy as sp
import tempfile
+
+import pytest
import pystencils as ps
from pathlib import Path
from pystencils_benchmark import generate_benchmark, Compiler
-def test_generate():
+compilers = (Compiler.GCC, Compiler.GCCdebug, Compiler.Clang)
+config_kwargs = ({}, {'cpu_vectorize_info': {'instruction_set': 'best'}})
+
+
+@pytest.mark.parametrize('compiler', compilers)
+@pytest.mark.parametrize('config_kwarg', config_kwargs)
+def test_generate(compiler, config_kwarg):
a, b, c = ps.fields(a=np.ones(4000000), b=np.ones(4000000), c=np.ones(4000000))
alpha = sp.symbols('alpha')
- @ps.kernel_config(ps.CreateKernelConfig())
+ @ps.kernel_config(ps.CreateKernelConfig(**config_kwarg))
def vadd():
a[0] @= b[0] + c[0]
kernel_vadd = ps.create_kernel(**vadd)
- @ps.kernel_config(ps.CreateKernelConfig())
+ @ps.kernel_config(ps.CreateKernelConfig(**config_kwarg))
def daxpy():
b[0] @= alpha * a[0] + b[0]
kernel_daxpy = ps.create_kernel(**daxpy)
- for compiler in [Compiler.GCC, Compiler.GCCdebug, Compiler.Clang]:
- with tempfile.TemporaryDirectory(dir=Path.cwd()) as temp_dir:
- temp_dir = Path(temp_dir)
- generate_benchmark([kernel_vadd, kernel_daxpy], temp_dir, compiler=compiler)
- subprocess.run(['make', '-C', f'{temp_dir}'], check=True)
- subprocess.run([f'{temp_dir}/benchmark-{compiler.name}', '10'], check=True)
+ with tempfile.TemporaryDirectory(dir=Path.cwd()) as temp_dir:
+ temp_dir = Path(temp_dir)
+ generate_benchmark([kernel_vadd, kernel_daxpy], temp_dir, compiler=compiler)
+ subprocess.run(['make', '-C', f'{temp_dir}'], check=True)
+ subprocess.run([f'{temp_dir}/benchmark-{compiler.name}', '10'], check=True)
diff --git a/tests/test_benchmark_vector.py b/tests/test_benchmark_vector.py
deleted file mode 100644
index 6fe2a6ebb1c7a4dc66912dda8a275689a0261f05..0000000000000000000000000000000000000000
--- a/tests/test_benchmark_vector.py
+++ /dev/null
@@ -1,31 +0,0 @@
-import subprocess
-import numpy as np
-import sympy as sp
-import tempfile
-import pystencils as ps
-from pathlib import Path
-from pystencils_benchmark import generate_benchmark, Compiler
-
-
-def test_generate():
- a, b, c = ps.fields(a=np.ones(4000000), b=np.ones(4000000), c=np.ones(4000000))
- alpha = sp.symbols('alpha')
-
- @ps.kernel_config(ps.CreateKernelConfig(cpu_vectorize_info={'instruction_set': 'best'}))
- def vadd():
- a[0] @= b[0] + c[0]
- kernel_vadd = ps.create_kernel(**vadd)
-
- @ps.kernel_config(ps.CreateKernelConfig(cpu_vectorize_info={'instruction_set': 'best'}))
- def daxpy():
- b[0] @= alpha * a[0] + b[0]
- kernel_daxpy = ps.create_kernel(**daxpy)
-
- for compiler in [Compiler.GCC, Compiler.GCCdebug, Compiler.Clang]:
- with tempfile.TemporaryDirectory(dir=Path.cwd()) as temp_dir:
- temp_dir = Path(temp_dir)
- generate_benchmark([kernel_vadd, kernel_daxpy], temp_dir, compiler=compiler)
- subprocess.run(['make', '-C', f'{temp_dir}'], check=True)
- subprocess.run([f'{temp_dir}/benchmark-{compiler.name}', '10'], check=True)
-
-
diff --git a/examle/test.py b/ve_example/test.py
similarity index 100%
rename from examle/test.py
rename to ve_example/test.py