From fd2f98aa6e85921cd8e91e5c0d2f1f0189360594 Mon Sep 17 00:00:00 2001
From: Jean-Baptiste Keck <Jean-Baptiste.Keck@imag.fr>
Date: Fri, 7 Apr 2017 16:56:17 +0200
Subject: [PATCH] codegen debug
---
hysop/__init__.py | 2 +-
hysop/backend/arrays/opencl_array.py | 1 +
hysop/backend/arrays/opencl_backend.py | 3 +-
hysop/backend/codegen/base/cl_extensions.py | 2 +-
hysop/backend/codegen/base/codegen.py | 84 ++---
hysop/backend/codegen/base/enum_codegen.py | 2 +-
.../backend/codegen/base/function_codegen.py | 13 +-
hysop/backend/codegen/base/kernel_codegen.py | 7 +-
hysop/backend/codegen/base/opencl_codegen.py | 5 +-
hysop/backend/codegen/base/statistics.py | 52 +--
hysop/backend/codegen/base/struct_codegen.py | 2 +-
hysop/backend/codegen/base/test.py | 22 +-
hysop/backend/codegen/base/variables.py | 12 +-
.../codegen/functions/advection_rhs.py | 10 +-
.../codegen/functions/apply_stencil.py | 66 ++--
hysop/backend/codegen/functions/cache_load.py | 11 +-
.../codegen/functions/compute_index.py | 2 +-
hysop/backend/codegen/functions/empty.py | 2 +-
hysop/backend/codegen/functions/gradient.py | 4 +-
.../backend/codegen/functions/runge_kutta.py | 7 +-
.../codegen/functions/stretching_rhs.py | 34 +-
hysop/backend/codegen/kernels/bandwidth.py | 6 +-
hysop/backend/codegen/kernels/copy_kernel.py | 12 +-
.../codegen/kernels/directional_advection.py | 33 +-
.../codegen/kernels/directional_stretching.py | 44 +--
hysop/backend/codegen/kernels/empty.py | 2 +-
hysop/backend/codegen/kernels/stretching.py | 2 +-
.../tests/test_directional_advection.py | 7 +-
.../tests/test_directional_stretching.py | 23 +-
hysop/backend/codegen/kernels/transpose.py | 2 +-
hysop/backend/codegen/structs/mesh_info.py | 49 +--
hysop/backend/opencl/cltypes.py | 333 ------------------
hysop/backend/opencl/discrete.py | 8 +-
hysop/backend/opencl/opencl_discrete.py | 6 +-
.../opencl/{clenv.py => opencl_env.py} | 4 +-
hysop/backend/opencl/tools.py | 10 +-
hysop/deps.py | 2 +-
hysop/numerics/fftw_f/fft3d.f90 | 2 +-
hysop/numerics/stencil/stencil.py | 50 +--
hysop/old/gpu.old/QtRendering.py | 10 +-
hysop/old/gpu.old/gpu_diffusion.py | 12 +-
hysop/old/gpu.old/gpu_discrete.py | 16 +-
hysop/old/gpu.old/gpu_kernel.py | 4 +-
.../gpu.old/gpu_multiphase_baroclinic_rhs.py | 8 +-
.../old/gpu.old/gpu_multiresolution_filter.py | 16 +-
hysop/old/gpu.old/gpu_particle_advection.py | 28 +-
.../old/gpu.old/gpu_particle_advection_dir.py | 12 +-
hysop/old/gpu.old/gpu_stretching.py | 8 +-
.../gpu.old/multi_gpu_particle_advection.py | 16 +-
.../static_gpu_particle_advection_dir.py | 16 +-
hysop/old/gpu.old/tests/test_copy.py | 16 +-
.../gpu.old/tests/test_opencl_environment.py | 6 +-
hysop/old/gpu.old/tests/test_transposition.py | 50 +--
hysop/old/gpu.old/tools.py | 16 +-
hysop/old/gpu.old/visu/marchingcube.py | 2 +-
hysop/tools/sympy_utils.py | 13 +-
56 files changed, 418 insertions(+), 769 deletions(-)
delete mode 100644 hysop/backend/opencl/cltypes.py
rename hysop/backend/opencl/{clenv.py => opencl_env.py} (99%)
diff --git a/hysop/__init__.py b/hysop/__init__.py
index 34130caa9..9fa51f813 100644
--- a/hysop/__init__.py
+++ b/hysop/__init__.py
@@ -24,7 +24,7 @@ __PROFILE__ = "OFF" in ["0", "1"]
__ENABLE_LONG_TESTS__ = "OFF" is "ON"
# OpenCL
-__DEFAULT_PLATFORM_ID__ = 0
+__DEFAULT_PLATFORM_ID__ = 1
__DEFAULT_DEVICE_ID__ = 0
diff --git a/hysop/backend/arrays/opencl_array.py b/hysop/backend/arrays/opencl_array.py
index 22e25bf16..d526a27c8 100644
--- a/hysop/backend/arrays/opencl_array.py
+++ b/hysop/backend/arrays/opencl_array.py
@@ -1,4 +1,5 @@
+
import numpy as np
from hysop.backend.opencl import clArray, clTools
from hysop.backend.arrays import MemoryType, MemoryOrdering
diff --git a/hysop/backend/arrays/opencl_backend.py b/hysop/backend/arrays/opencl_backend.py
index 346ab6c2e..584d55671 100644
--- a/hysop/backend/arrays/opencl_backend.py
+++ b/hysop/backend/arrays/opencl_backend.py
@@ -25,6 +25,7 @@ default_cl_context = default_cl_env.context()
default_cl_queue = default_cl_env.default_queue()
default_cl_allocator = default_cl_env.allocator()
default_cl_pool = default_cl_env.memory_pool()
+
class _ElementwiseKernel(object):
"""
@@ -2157,7 +2158,7 @@ class OpenClBackend(ArrayBackend):
if is_integer(x1):
expr = '(x1[i]==0 ? 0 : {})'.format(expr)
convert_inputs=(np.float64,None)
- if is_fp(x2):
+ if is_fp(x1) or is_fp(x2):
expr = '(isnan(x0[i])||isnan(x1[i])?NAN:(isinf(x1[i])?(isinf(x0[i])?NAN:(signbit(x0[i])^signbit(x1[i])?x1[i]:x0[i])):{}))'.format(expr)
convert_inputs='f'
expr='y0[i] = '+expr
diff --git a/hysop/backend/codegen/base/cl_extensions.py b/hysop/backend/codegen/base/cl_extensions.py
index dae06467a..75c042c99 100644
--- a/hysop/backend/codegen/base/cl_extensions.py
+++ b/hysop/backend/codegen/base/cl_extensions.py
@@ -17,7 +17,7 @@ _cl_extension_custom_declarations = {
}
from hysop.backend.codegen.base.opencl_codegen import OpenClCodeGenerator
-from hysop.backend.codegen.base.types import OpenClTypeGen
+from hysop.backend.opencl.opencl_types import OpenClTypeGen
from hysop.backend.codegen.base.test import _test_typegen
class ClExtCodeGen(OpenClCodeGenerator):
diff --git a/hysop/backend/codegen/base/codegen.py b/hysop/backend/codegen/base/codegen.py
index 952578727..a7572c3e7 100644
--- a/hysop/backend/codegen/base/codegen.py
+++ b/hysop/backend/codegen/base/codegen.py
@@ -314,10 +314,8 @@ class CodeGenerator(object):
@contextmanager
def _var_block_(self):
vb = self.VarBlock()
- try:
- yield vb
- finally:
- self.append(vb.code())
+ yield vb
+ self.append(vb.code())
class AlignBlock(object):
@@ -366,13 +364,8 @@ class CodeGenerator(object):
@contextmanager
def _align_(self,sep='$'):
ab = self.AlignBlock(sep=sep)
- try:
- yield ab
- except:
- raise
- finally:
- self.append(ab.code())
-
+ yield ab
+ self.append(ab.code())
def declare_codeblocks(self,names,initial_code=None,priorities=None,comments=None):
if not isinstance(names,list):
@@ -412,16 +405,14 @@ class CodeGenerator(object):
elif priority is not None and self.blocks[blockname][0] != priority:
raise ValueError('Priority mismatch!')
- try:
- code = self.code
- indent = self.indent_level
- self.indent_level = self.initial_indent_level
- self.code = self.blocks[blockname][1]
- yield
- finally:
- self.blocks[blockname][1] = self.code
- self.code = code
- self.indent_level = indent
+ code = self.code
+ indent = self.indent_level
+ self.indent_level = self.initial_indent_level
+ self.code = self.blocks[blockname][1]
+ yield
+ self.blocks[blockname][1] = self.code
+ self.code = code
+ self.indent_level = indent
def block_exists(self,name):
return (name in self.blocks)
@@ -451,38 +442,35 @@ class CodeGenerator(object):
def _block_(self,header_prefix='',header_postfix='',footer_postfix='',compact=False):
count = 1-int(compact)
newline = not compact
- try:
- header = header_prefix + '{' + header_postfix
- self.append(header,newline).indent(count)
- yield
- finally:
- if compact:
- self.supress_newline()
- self.code += ' '
- footer = '}' + footer_postfix
- self.dedent(count).append(footer)
+
+ header = header_prefix + '{' + header_postfix
+ self.append(header,newline).indent(count)
+ yield
+ if compact:
+ self.supress_newline()
+ self.code += ' '
+ footer = '}' + footer_postfix
+ self.dedent(count).append(footer)
#conditional facilities
@contextmanager
def _if_(self,cond,compact=False,force_spaces=False):
- try:
- if cond=='true':
- yield
- elif cond=='false':
- code = self.code
- indent = self.indent_level
- self.code = ''
- yield
- else:
- sep = self.space() if force_spaces else self.empty()
- header_prefix = 'if (' + sep + cond + sep + ') '
- with self._block_(header_prefix=header_prefix,compact=compact) as b:
- yield b
- finally:
- if cond=='false':
- self.code = code
- self.indent_level = indent
+ if cond=='true':
+ yield
+ elif cond=='false':
+ code = self.code
+ indent = self.indent_level
+ self.code = ''
+ yield
+ else:
+ sep = self.space() if force_spaces else self.empty()
+ header_prefix = 'if (' + sep + cond + sep + ') '
+ with self._block_(header_prefix=header_prefix,compact=compact) as b:
+ yield b
+ if cond=='false':
+ self.code = code
+ self.indent_level = indent
@contextmanager
def _elif_(self,cond,compact=False,force_spaces=True):
sep = self.space() if force_spaces else self.empty()
diff --git a/hysop/backend/codegen/base/enum_codegen.py b/hysop/backend/codegen/base/enum_codegen.py
index b997591fa..6741e2325 100644
--- a/hysop/backend/codegen/base/enum_codegen.py
+++ b/hysop/backend/codegen/base/enum_codegen.py
@@ -2,7 +2,7 @@
import numpy as np
import pyopencl as cl
from hysop.backend.codegen.base.codegen import CodeGenerator
-from hysop.backend.codegen.base.types import TypeGen
+from hysop.backend.opencl.opencl_types import TypeGen
from hysop.tools.enum import EnumFactory
from hysop.backend.codegen.base.variables import register_ctype_dtype
diff --git a/hysop/backend/codegen/base/function_codegen.py b/hysop/backend/codegen/base/function_codegen.py
index 654c9a4f7..3e10b411b 100644
--- a/hysop/backend/codegen/base/function_codegen.py
+++ b/hysop/backend/codegen/base/function_codegen.py
@@ -2,10 +2,11 @@
import pyopencl as cl
from contextlib import contextmanager
+from hysop.tools.types import check_instance
from hysop.backend.codegen.base.codegen import CodeGenerator
from hysop.backend.codegen.base.opencl_codegen import OpenClCodeGenerator
from hysop.backend.codegen.base.utils import ArgDict
-from hysop.backend.codegen.base.types import TypeGen, OpenClTypeGen
+from hysop.backend.opencl.opencl_types import TypeGen, OpenClTypeGen
from hysop.backend.codegen.base.variables import CodegenVariable, CodegenVectorClBuiltin, CodegenVectorClBuiltinFunc
class FunctionBase(object):
@@ -17,7 +18,7 @@ class FunctionBase(object):
super(FunctionBase,self).__init__(typegen=typegen,**kargs)
known_args = ArgDict() if (known_args is None) else known_args
- assert isinstance(fargs, ArgDict)
+ check_instance(fargs,ArgDict)
fargs.release()
for varname, varval in known_args.iteritems():
@@ -88,11 +89,11 @@ class FunctionCodeGenerator(FunctionBase, CodeGenerator):
def __init__(self,basename,typegen,output,
args=None,known_args=None,inline=False,ext='.tmp'):
- assert isinstance(typegen, TypeGen)
+ check_instance(typegen,TypeGen)
name = basename
fname = basename
if args is not None:
- assert isinstance(args, ArgDict)
+ check_instance(args,ArgDict)
fname += args.function_name_suffix(output)
name += args.codegen_name_suffix(output)
@@ -117,11 +118,11 @@ class OpenClFunctionCodeGenerator(FunctionBase, OpenClCodeGenerator):
def __init__(self,basename,typegen,output,
args=None,known_args=None,inline=False,ext='.cl'):
- assert isinstance(typegen, OpenClTypeGen)
+ check_instance(typegen,OpenClTypeGen)
name = basename
fname = basename
if args is not None:
- assert isinstance(args, ArgDict)
+ check_instance(args,ArgDict)
fname += args.function_name_suffix(output)
name += args.codegen_name_suffix(output)
diff --git a/hysop/backend/codegen/base/kernel_codegen.py b/hysop/backend/codegen/base/kernel_codegen.py
index 859ed8db1..64b17dcff 100644
--- a/hysop/backend/codegen/base/kernel_codegen.py
+++ b/hysop/backend/codegen/base/kernel_codegen.py
@@ -1,10 +1,11 @@
from contextlib import contextmanager
+from hysop.tools.types import check_instance
from hysop.backend.codegen.base.opencl_codegen import OpenClCodeGenerator
from hysop.backend.codegen.base.function_codegen import FunctionBase
from hysop.backend.codegen.base.utils import VarDict, ArgDict, WriteOnceDict
-from hysop.backend.codegen.base.types import OpenClTypeGen
+from hysop.backend.opencl.opencl_types import OpenClTypeGen
from hysop.backend.codegen.base.variables import CodegenVariable, CodegenVectorClBuiltin, \
CodegenVectorClBuiltinFunc
@@ -27,8 +28,8 @@ class KernelCodeGenerator(KernelBase, OpenClCodeGenerator):
kernel_args = ArgDict() if (kernel_args is None) else kernel_args
known_vars = WriteOnceDict() if (known_vars is None) else known_vars
- assert isinstance(typegen, OpenClTypeGen)
- assert isinstance(kernel_args, ArgDict)
+ check_instance(typegen,OpenClTypeGen)
+ check_instance(kernel_args,ArgDict)
assert work_dim>0 and work_dim<=3
if (vec_type_hint is not None):
diff --git a/hysop/backend/codegen/base/opencl_codegen.py b/hysop/backend/codegen/base/opencl_codegen.py
index 9179e7011..7891a18a1 100644
--- a/hysop/backend/codegen/base/opencl_codegen.py
+++ b/hysop/backend/codegen/base/opencl_codegen.py
@@ -2,8 +2,9 @@
import pyopencl as cl
from contextlib import contextmanager
+from hysop.tools.types import check_instance
from hysop.backend.codegen.base.codegen import CodeGenerator
-from hysop.backend.codegen.base.types import OpenClTypeGen
+from hysop.backend.opencl.opencl_types import OpenClTypeGen
from hysop.backend.codegen.base.utils import VarDict
from hysop.backend.codegen.base.variables import CodegenVariable, CodegenVectorClBuiltin, CodegenVectorClBuiltinFunc
@@ -31,7 +32,7 @@ class OpenClCodeGenerator(CodeGenerator):
known_vars=None, declare_cl_exts=True,
**kargs):
- assert isinstance(typegen, OpenClTypeGen)
+ check_instance(typegen,OpenClTypeGen)
super(OpenClCodeGenerator,self).__init__(name=name,typegen=typegen, ext=ext,
known_vars=known_vars, keywords=self.default_keywords,
escape_seqs=self.default_escape_seqs, **kargs)
diff --git a/hysop/backend/codegen/base/statistics.py b/hysop/backend/codegen/base/statistics.py
index 6115d9d0c..25de0f49d 100644
--- a/hysop/backend/codegen/base/statistics.py
+++ b/hysop/backend/codegen/base/statistics.py
@@ -2,8 +2,9 @@
import copy
import numpy as np
+from hysop.tools.types import check_instance
from hysop.tools.units import bytes2str
-from hysop.backend.codegen.base.types import vsizes, signed_base_types, unsigned_base_types, float_base_types
+from hysop.backend.opencl.opencl_types import vsizes, signed_base_types, unsigned_base_types, float_base_types
dtype_ops = {
np.int8: (0.25, 'IOPS'),
@@ -44,7 +45,7 @@ _fill_dtype_ops()
class WorkStatistics(object):
def __init__(self, stat=None):
if (stat is not None):
- assert isinstance(stat, WorkStatistics)
+ check_instance(stat,WorkStatistics)
self.global_mem_byte_reads = stat.global_mem_byte_reads
self.global_mem_byte_writes = stat.global_mem_byte_writes
self.local_mem_byte_reads = stat.local_mem_byte_reads
@@ -62,23 +63,28 @@ class WorkStatistics(object):
def global_mem_transactions(self):
return self.global_mem_byte_writes + self.global_mem_byte_reads
- def global_mem_write_ratio(self):
+ def global_mem_rw_ratio(self):
return float(self.global_mem_byte_writes)/self.global_mem_transactions()
def global_mem_read_ratio(self):
return float(self.global_mem_byte_reads)/self.global_mem_transactions()
def local_mem_transactions(self):
return self.local_mem_byte_writes + self.local_mem_byte_reads
- def local_mem_write_ratio(self):
+ def local_mem_rw_ratio(self):
return float(self.local_mem_byte_writes)/self.local_mem_transactions()
def local_mem_read_ratio(self):
return float(self.local_mem_byte_reads)/self.local_mem_transactions()
def total_mem_transactions(self):
return self.local_mem_transactions() + self.global_mem_transactions()
+
+ def has_local_mem_transactions(self):
+ return (self.local_mem_transactions() > 0)
+ def has_global_mem_transactions(self):
+ return (self.global_mem_transactions() > 0)
def __add__(self, rhs):
- assert isinstance(rhs, WorkStatistics)
+ check_instance(rhs,WorkStatistics)
stats = copy.deepcopy(self)
stats.global_mem_byte_reads += rhs.global_mem_byte_reads
stats.global_mem_byte_writes += rhs.global_mem_byte_writes
@@ -93,7 +99,7 @@ class WorkStatistics(object):
return stats
def __mul__(self, rhs):
- assert isinstance(rhs,int)
+ check_instance(rhs,int)
stats = copy.deepcopy(self)
stats.global_mem_byte_reads *= rhs
stats.global_mem_byte_writes *= rhs
@@ -104,28 +110,30 @@ class WorkStatistics(object):
return stats
def __rmul__(self, lhs):
- assert isinstance(lhs,int)
+ check_instance(lhs,int)
return self.__mul__(lhs)
def __str__(self):
op_count = [''] + ['{}: {}'.format(k,v) for (k,v) in self.ops_per_type.iteritems() ]
op_count = '\n '.join(op_count)
- ss = \
-''':: Work Statistics ::
- Global memory: load={} store={} total={} write_ratio={}
- Local memory: load={} store={} total={} write_ratio={}
- Operations count:{}
-'''.format(
- bytes2str(self.global_mem_byte_reads),
- bytes2str(self.global_mem_byte_writes),
- bytes2str(self.global_mem_transactions()),
- round(self.global_mem_write_ratio(),2),
- bytes2str(self.local_mem_byte_reads),
- bytes2str(self.local_mem_byte_writes),
- bytes2str(self.local_mem_transactions()),
- round(self.local_mem_write_ratio(),2),
- op_count)
+ ss = ':: Work Statistics ::'
+
+ if self.has_global_mem_transactions():
+ ss += '\n Global memory: load={} store={} total={} rw_ratio={}'.format(
+ bytes2str(self.global_mem_byte_reads),
+ bytes2str(self.global_mem_byte_writes),
+ bytes2str(self.global_mem_transactions()),
+ round(self.global_mem_rw_ratio(),2))
+
+ if self.has_local_mem_transactions():
+ ss += '\n Local memory: load={} store={} total={} rw_ratio={}'.format(
+ bytes2str(self.local_mem_byte_reads),
+ bytes2str(self.local_mem_byte_writes),
+ bytes2str(self.local_mem_transactions()),
+ round(self.local_mem_rw_ratio(),2))
+
+ ss += '\n Operations count: {}'.format(op_count)
return ss
diff --git a/hysop/backend/codegen/base/struct_codegen.py b/hysop/backend/codegen/base/struct_codegen.py
index d805e7346..41a85509e 100644
--- a/hysop/backend/codegen/base/struct_codegen.py
+++ b/hysop/backend/codegen/base/struct_codegen.py
@@ -5,7 +5,7 @@ import pyopencl.tools
import numpy as np
import re
-from hysop.backend.codegen.base.types import np_dtype
+from hysop.backend.opencl.opencl_types import np_dtype
from hysop.backend.codegen.base.opencl_codegen import OpenClCodeGenerator
from hysop.backend.codegen.base.variables import VarDict, CodegenVariable, CodegenVector, CodegenStruct, CodegenVectorClBuiltin
from hysop.backend.codegen.base.variables import register_ctype_dtype
diff --git a/hysop/backend/codegen/base/test.py b/hysop/backend/codegen/base/test.py
index 79c32d764..9aacfa1cb 100644
--- a/hysop/backend/codegen/base/test.py
+++ b/hysop/backend/codegen/base/test.py
@@ -1,29 +1,18 @@
from hysop.constants import np
-from hysop.backend.opencl import pyopencl
-from hysop.constants import MeshDirection, TranspositionState
+from hysop.backend.opencl import cl
from hysop.backend.codegen.structs.mesh_info import MeshBaseStruct, MeshInfoStruct
-
-from hysop import __DEFAULT_PLATFORM_ID__, __DEFAULT_DEVICE_ID__
-from hysop.backend.codegen.base.types import OpenClTypeGen
+from hysop.backend.opencl.opencl_types import OpenClTypeGen
## Quickly get a default opencl typegen **ONLY** for testing or debugging purpose
## => do not use in production code
def _test_typegen(fbtype='float',float_dump_mode='dec'):
- try:
- default_platform = pyopencl.get_platforms()[__DEFAULT_PLATFORM_ID__]
- except:
- default_platform = pyopencl.get_platforms()[0]
- try:
- default_device = default_platform.get_devices()[__DEFAULT_DEVICE_ID__]
- except:
- default_device = default_platform.get_devices()[0]
- default_context = pyopencl.Context(devices=[default_device])
+ from hysop.backend.arrays.opencl_backend import\
+ default_cl_context, default_cl_device, default_cl_platform
return OpenClTypeGen(
- context=default_context, device=default_device, platform=default_platform,
+ context=default_cl_context, device=default_cl_device, platform=default_cl_platform,
fbtype=fbtype,float_dump_mode=float_dump_mode)
-
def _test_mesh_info(typegen,dim,ghosts,resolution,**kargs):
ghosts = [ghosts]*dim if np.isscalar(ghosts) else ghosts
ghosts = np.asarray(ghosts)[:dim]
@@ -62,7 +51,6 @@ def _test_mesh_info(typegen,dim,ghosts,resolution,**kargs):
ghosts=ghosts,
dx=dx,
local_mesh=local_mesh, global_mesh=global_mesh,
- mesh_dir=MeshDirection.X, mesh_state=TranspositionState.XYZ,
**kargs)
return (np_mis, cg_mis)
diff --git a/hysop/backend/codegen/base/variables.py b/hysop/backend/codegen/base/variables.py
index e86244d3f..9681eba4a 100644
--- a/hysop/backend/codegen/base/variables.py
+++ b/hysop/backend/codegen/base/variables.py
@@ -1,13 +1,13 @@
import re
-import hysop.backend.codegen.base.types
+import hysop.backend.opencl.opencl_types
from hysop.constants import np
from hysop.backend.codegen.base.utils import VarDict
-from hysop.backend.codegen.base.types import OpenClTypeGen
+from hysop.backend.opencl.opencl_types import OpenClTypeGen
# opencl extras
-from hysop.backend.codegen.base.types import cl_type_to_dtype
+from hysop.backend.opencl.opencl_types import cl_type_to_dtype
_ctype_to_dtype = {
'bool': np.bool_,
@@ -264,11 +264,13 @@ class CodegenVariable(object):
return self.__call__()
else:
assert self.ptr, '{} is not a pointer!'.format(self.name)
- def __str__(self):
+ def __repr__(self):
if self.is_symbolic():
return '{}({})'.format(self.name,self.ctype)
else:
return '{}({},{})'.format(self.name,self.ctype,self.value)
+ def __str__(self):
+ return self.sval()
class CodegenArray(CodegenVariable):
@staticmethod
@@ -685,7 +687,7 @@ class CodegenStruct(CodegenVariable):
var = var_overrides[fieldname]
var.struct_var = self
var.const = self.const
- elif is_struct and (ctype in hysop.backend.codegen.base.types.builtin_types):
+ elif is_struct and (ctype in hysop.backend.opencl.opencl_types.builtin_types):
tg=struct.typegen
btype = tg.basetype(ctype)
dim = tg.components(ctype)
diff --git a/hysop/backend/codegen/functions/advection_rhs.py b/hysop/backend/codegen/functions/advection_rhs.py
index fa927ce37..2e9add494 100644
--- a/hysop/backend/codegen/functions/advection_rhs.py
+++ b/hysop/backend/codegen/functions/advection_rhs.py
@@ -1,15 +1,16 @@
+from hysop.tools.types import check_instance
from hysop.backend.codegen.base.opencl_codegen import OpenClCodeGenerator
from hysop.backend.codegen.base.function_codegen import OpenClFunctionCodeGenerator
from hysop.backend.codegen.base.variables import CodegenVariable, CodegenVectorClBuiltin
-from hysop.backend.codegen.base.types import OpenClTypeGen
+from hysop.backend.opencl.opencl_types import OpenClTypeGen
from hysop.backend.codegen.base.utils import WriteOnceDict, ArgDict
from hysop.backend.codegen.base.statistics import WorkStatistics
from hysop.constants import BoundaryCondition
from hysop.backend.codegen.functions.apply_stencil import ApplyStencilFunction
-from hysop.numerics.stencil.stencil import Stencil1D, Stencil
+from hysop.numerics.stencil.stencil import Stencil
# compute alpha*grad(u).w + (1-alpha)*grad(u)^T.w
# where u may be is_cached and w is private
@@ -22,12 +23,9 @@ class DirectionalAdvectionRhsFunction(OpenClFunctionCodeGenerator):
known_args=None):
assert work_dim>=1 and work_dim<=3
- assert boundary in BoundaryCondition.entries()
+ check_instance(boundary,BoundaryCondition)
assert nparticles in [1,2,4,8,16]
- boundary = BoundaryCondition.value(boundary)
- sboundary = BoundaryCondition.svalue(boundary)
-
is_periodic = (boundary==BoundaryCondition.PERIODIC)
if is_cached:
diff --git a/hysop/backend/codegen/functions/apply_stencil.py b/hysop/backend/codegen/functions/apply_stencil.py
index 98b0e3294..cd65cb631 100644
--- a/hysop/backend/codegen/functions/apply_stencil.py
+++ b/hysop/backend/codegen/functions/apply_stencil.py
@@ -1,8 +1,10 @@
+from hysop.deps import sm
+from hysop.tools.types import check_instance
from hysop.backend.codegen.base.opencl_codegen import OpenClCodeGenerator
from hysop.backend.codegen.base.function_codegen import OpenClFunctionCodeGenerator
from hysop.backend.codegen.base.variables import CodegenVariable, CodegenVectorClBuiltin
-from hysop.backend.codegen.base.types import OpenClTypeGen
+from hysop.backend.opencl.opencl_types import OpenClTypeGen
from hysop.backend.codegen.base.utils import ArgDict
from hysop.backend.codegen.base.statistics import WorkStatistics
@@ -11,6 +13,7 @@ from hysop.numerics.stencil.stencil import Stencil
class ApplyStencilFunction(OpenClFunctionCodeGenerator):
def __init__(self,typegen,stencil,ftype,
+ symbol2var=None,
components=1, vectorize=True,
extra_inputs=[],
scalar_inputs=[],
@@ -23,7 +26,14 @@ class ApplyStencilFunction(OpenClFunctionCodeGenerator):
itype='int',
known_args=None):
- assert isinstance(stencil, Stencil)
+ check_instance(stencil,Stencil)
+ check_instance(symbol2var, dict, keys=sm.Symbol, values=CodegenVariable, allow_none=True)
+ assert set(symbol2var.keys())==stencil.variables()
+
+ extra_inputs = set(extra_inputs + symbol2var.values())
+ scalar_inputs = set(scalar_inputs)
+ vector_inputs = set(vector_inputs)
+
dim = stencil.dim
vtype = typegen.vtype(ftype,components)
@@ -37,20 +47,25 @@ class ApplyStencilFunction(OpenClFunctionCodeGenerator):
for iname in vector_inputs:
if vectorize:
name = iname
- args[name] = CodegenVariable(name, vtype, typegen, const=True, add_impl_const=True, storage=data_storage, ptr=True, restrict=restrict)
+ args[name] = CodegenVariable(name, vtype, typegen, const=True,
+ add_impl_const=True, storage=data_storage, ptr=True, restrict=restrict)
else:
for i in xrange(components):
name = '{}{}'.format(iname,vector_suffixes[i])
- args[name] = CodegenVariable(name, ftype, typegen, const=True, add_impl_const=True, storage=data_storage, ptr=True, restrict=restrict)
+ args[name] = CodegenVariable(name, ftype, typegen,
+ const=True, add_impl_const=True, storage=data_storage,
+ ptr=True, restrict=restrict)
for iname in scalar_inputs:
name = iname
- args[name] = CodegenVariable(name, ftype, typegen, const=True, add_impl_const=True, storage=data_storage, ptr=True, restrict=restrict)
+ args[name] = CodegenVariable(name, ftype, typegen, const=True,
+ add_impl_const=True, storage=data_storage, ptr=True, restrict=restrict)
for arg in extra_inputs:
args[arg.name] = arg
if not has_custom_id:
args['offset'] = CodegenVariable('offset', itype, typegen, add_impl_const=True)
- args['stride'] = CodegenVectorClBuiltin('stride', itype, dim, typegen, add_impl_const=True,nl=True)
+ args['stride'] = CodegenVectorClBuiltin('stride', itype, dim, typegen,
+ add_impl_const=True,nl=True)
for varname,vartype in multipliers.iteritems():
if vartype=='ftype':
vartype=ftype
@@ -81,6 +96,7 @@ class ApplyStencilFunction(OpenClFunctionCodeGenerator):
self.has_custom_id = has_custom_id
self.custom_id = custom_id
self.op = op
+ self.symbol2var = symbol2var
self.gencode()
@@ -101,7 +117,7 @@ class ApplyStencilFunction(OpenClFunctionCodeGenerator):
cached = self.is_cached()
ncoeffs = self.stencil.non_zero_coefficients()
- nmul = len(self.multipliers)
+ nmul = len(self.multipliers)+int(self.stencil.has_factor())
stats = WorkStatistics()
@@ -135,11 +151,8 @@ class ApplyStencilFunction(OpenClFunctionCodeGenerator):
ftype = s.ftype
vtype = s.vtype
+ stencil = s.stencil
- has_multiplier = ('multiplier' in s.vars.keys())
- if has_multiplier:
- multiplier = s.vars['multiplier']
-
res = CodegenVectorClBuiltin('res',ftype,components,tg)
with s._function_():
@@ -150,39 +163,44 @@ class ApplyStencilFunction(OpenClFunctionCodeGenerator):
if i>0:
al.jumpline()
operands = {}
- if self.vectorized:
- vector_varnames = self.vector_inputs
+ if s.vectorized:
+ vector_varnames = s.vector_inputs
_res = res()
else:
- vector_varnames = ['{}{}'.format(varname,vector_suffixes[i]) for varname in self.vector_inputs]
+ vector_varnames = ['{}{}'.format(varname,vector_suffixes[i])
+ for varname in s.vector_inputs]
_res = res[i]
for j,vn in enumerate(vector_varnames):
operands['vinput{}'.format(j)] = s.vars[vn]()
- for j,vn in enumerate(self.scalar_inputs):
+ for j,vn in enumerate(s.scalar_inputs):
operands['sinput{}'.format(j)] = s.vars[vn]()
- for (off,coeff) in self.stencil.iteritems():
+ for (off,coeff) in stencil.iteritems(include_factor=False,
+ svars=s.symbol2var):
+ if coeff=='0':
+ continue
if not has_custom_id:
offset = s.vars['offset']
stride = s.vars['stride']
strided=''
for i in xrange(dim):
- if off[i]==0:
- continue
if stride.known() and stride.value[0] == 1:
strided+='{}'.format(tg.dump(off[i]))
else:
strided+='{}*{}'.format(tg.dump(off[i]),stride[i])
operands['id'] = '{}+{}'.format(offset(),strided)
else:
- operands['id'] = self.custom_id.format(offset=tg.dump(off[0]))
- code = '{} += {} $* {};'.format(_res,tg.dump(coeff),self.op.format(**operands))
+ operands['id'] = s.custom_id.format(offset=tg.dump(off[0]))
+ code = '{} += {} $* {};'.format(_res,tg.dump(coeff),
+ s.op.format(**operands))
al.append(code)
if vectorized:
break
mul = ''
- for mult in self.multipliers:
+ for mult in s.multipliers:
mul+='{}*'.format(s.vars[mult]())
+ if stencil.has_factor():
+ mul+='{}*'.format(stencil.format_factor(s.symbol2var))
ret = 'return {}{};'.format(mul,res())
s.append(ret)
@@ -190,16 +208,16 @@ class ApplyStencilFunction(OpenClFunctionCodeGenerator):
if __name__ == '__main__':
- from hysop.backend.codegen.base.test import test_typegen
+ from hysop.backend.codegen.base.test import _test_typegen
stencil = Stencil([2.0,1.0,0.0,-1.0,-2.0], origin=2, order=2)
- tg = test_typegen('double', float_dump_mode='hex')
+ tg = _test_typegen('double', float_dump_mode='hex')
asf = ApplyStencilFunction(tg,stencil,ftype=tg.fbtype,
components=3, vectorize=False, data_storage='__local',
scalar_inputs = ['S'], vector_inputs=['A','B'], vector_suffixes=['x','y','z'],
op='{sinput0}[{id}] * ({vinput0}[{id}] + {vinput1}[{id}])',
multipliers={'a':'int','b':'float','c':'ftype','d':'vtype'})
- #print asf.per_work_statistics()
+ print asf.per_work_statistics()
asf.edit()
diff --git a/hysop/backend/codegen/functions/cache_load.py b/hysop/backend/codegen/functions/cache_load.py
index 5adeba9be..1a815b24b 100644
--- a/hysop/backend/codegen/functions/cache_load.py
+++ b/hysop/backend/codegen/functions/cache_load.py
@@ -2,10 +2,11 @@
import contextlib
from contextlib import contextmanager
+from hysop.tools.types import check_instance
from hysop.backend.codegen.base.opencl_codegen import OpenClCodeGenerator
from hysop.backend.codegen.base.function_codegen import OpenClFunctionCodeGenerator
from hysop.backend.codegen.base.variables import CodegenVariable, CodegenVectorClBuiltin
-from hysop.backend.codegen.base.types import OpenClTypeGen
+from hysop.backend.opencl.opencl_types import OpenClTypeGen
from hysop.backend.codegen.base.utils import ArgDict
from hysop.backend.codegen.base.statistics import WorkStatistics
@@ -32,18 +33,16 @@ class CacheLoadFunction(OpenClFunctionCodeGenerator):
assert work_dim>0
- assert boundary in BoundaryCondition.entries()
- boundary = BoundaryCondition.value(boundary)
- sboundary = BoundaryCondition.svalue(boundary)
+ check_instance(boundary,BoundaryCondition)
if boundary not in [BoundaryCondition.NONE, BoundaryCondition.PERIODIC]:
- raise NotImplemented('Boundary \'{}\' not implemented yet!'.format(sboundary.lower()))
+ raise NotImplemented('Boundary \'{}\' not implemented yet!'.format(str(boundary).lower()))
tg = typegen
fs = force_symbolic
vtype = tg.vtype(ftype,components)
name = 'cache_load_{}d'.format(work_dim)
if boundary != BoundaryCondition.NONE:
- name+='_{}'.format(sboundary.lower())
+ name+='_{}'.format(str(boundary).lower())
output = 'void'
args = ArgDict()
diff --git a/hysop/backend/codegen/functions/compute_index.py b/hysop/backend/codegen/functions/compute_index.py
index dd052c855..d1de3c489 100644
--- a/hysop/backend/codegen/functions/compute_index.py
+++ b/hysop/backend/codegen/functions/compute_index.py
@@ -3,7 +3,7 @@
from hysop.backend.codegen.base.opencl_codegen import OpenClCodeGenerator
from hysop.backend.codegen.base.function_codegen import OpenClFunctionCodeGenerator
from hysop.backend.codegen.base.variables import CodegenVariable, CodegenVectorClBuiltin
-from hysop.backend.codegen.base.types import OpenClTypeGen
+from hysop.backend.opencl.opencl_types import OpenClTypeGen
from hysop.backend.codegen.base.utils import ArgDict
from hysop.backend.codegen.base.statistics import WorkStatistics
diff --git a/hysop/backend/codegen/functions/empty.py b/hysop/backend/codegen/functions/empty.py
index 43832ee27..8070ef4ff 100644
--- a/hysop/backend/codegen/functions/empty.py
+++ b/hysop/backend/codegen/functions/empty.py
@@ -2,7 +2,7 @@
from hysop.backend.codegen.base.codegen import CodeGenerator
from hysop.backend.codegen.base.function_codegen import FunctionCodeGenerator
from hysop.backend.codegen.base.variables import CodegenVariable
-from hysop.backend.codegen.base.types import TypeGen
+from hysop.backend.opencl.opencl_types import TypeGen
from hysop.backend.codegen.base.utils import ArgDict
class EmptyFunction(FunctionCodeGenerator):
diff --git a/hysop/backend/codegen/functions/gradient.py b/hysop/backend/codegen/functions/gradient.py
index 6896626dd..20ff2eff2 100644
--- a/hysop/backend/codegen/functions/gradient.py
+++ b/hysop/backend/codegen/functions/gradient.py
@@ -4,10 +4,10 @@ import numpy as np
from hysop.backend.codegen.base.opencl_codegen import OpenClCodeGenerator
from hysop.backend.codegen.base.function_codegen import OpenClFunctionCodeGenerator
from hysop.backend.codegen.base.variables import CodegenVariable, CodegenVectorClBuiltin
-from hysop.backend.codegen.base.types import OpenClTypeGen
+from hysop.backend.opencl.opencl_types import OpenClTypeGen
from hysop.backend.codegen.base.utils import ArgDict
-from hysop.numerics.stencil.stencil import Stencil1D, Stencil
+from hysop.numerics.stencil.stencil import Stencil
from hysop.backend.codegen.functions.compute_index import ComputeIndexFunction
from hysop.backend.codegen.functions.cache_load import CacheLoadFunction
diff --git a/hysop/backend/codegen/functions/runge_kutta.py b/hysop/backend/codegen/functions/runge_kutta.py
index bc57fadf9..035c12636 100644
--- a/hysop/backend/codegen/functions/runge_kutta.py
+++ b/hysop/backend/codegen/functions/runge_kutta.py
@@ -1,10 +1,11 @@
import hashlib, copy
import numpy as np
+from hysop.tools.types import check_instance
from hysop.backend.codegen.base.opencl_codegen import OpenClCodeGenerator
from hysop.backend.codegen.base.function_codegen import OpenClFunctionCodeGenerator
from hysop.backend.codegen.base.variables import CodegenVariable, CodegenVectorClBuiltin, CodegenArray
-from hysop.backend.codegen.base.types import OpenClTypeGen
+from hysop.backend.opencl.opencl_types import OpenClTypeGen
from hysop.backend.codegen.base.utils import ArgDict
from hysop.backend.codegen.base.statistics import WorkStatistics
@@ -23,8 +24,8 @@ class RungeKuttaFunction(OpenClFunctionCodeGenerator):
used_vars=_default_used_vars,
known_args=None):
- assert isinstance(method, ExplicitRungeKutta)
- assert isinstance(rhs, OpenClFunctionCodeGenerator)
+ check_instance(method,ExplicitRungeKutta)
+ check_instance(rhs,OpenClFunctionCodeGenerator)
method.dump = typegen.dump
#find out rhs function arguments
diff --git a/hysop/backend/codegen/functions/stretching_rhs.py b/hysop/backend/codegen/functions/stretching_rhs.py
index f7304c2ac..47b040e00 100644
--- a/hysop/backend/codegen/functions/stretching_rhs.py
+++ b/hysop/backend/codegen/functions/stretching_rhs.py
@@ -1,8 +1,10 @@
+from hysop.deps import sm
+from hysop.tools.types import check_instance
from hysop.backend.codegen.base.opencl_codegen import OpenClCodeGenerator
from hysop.backend.codegen.base.function_codegen import OpenClFunctionCodeGenerator
from hysop.backend.codegen.base.variables import CodegenVariable, CodegenVectorClBuiltin
-from hysop.backend.codegen.base.types import OpenClTypeGen
+from hysop.backend.opencl.opencl_types import OpenClTypeGen
from hysop.backend.codegen.base.utils import WriteOnceDict, ArgDict
from hysop.backend.codegen.base.statistics import WorkStatistics
from hysop.methods import StretchingFormulation
@@ -33,14 +35,9 @@ class DirectionalStretchingRhsFunction(OpenClFunctionCodeGenerator):
assert dim==3
assert direction<dim
assert order>1 and order%2==0
- assert formulation in StretchingFormulation.entries()
- assert boundary in BoundaryCondition.entries()
+ check_instance(formulation,StretchingFormulation)
+ check_instance(boundary,BoundaryCondition)
- formulation = StretchingFormulation.value(formulation)
- sformulation = StretchingFormulation.svalue(formulation)
- boundary = BoundaryCondition.value(boundary)
- sboundary = BoundaryCondition.svalue(boundary)
-
is_conservative = (formulation==StretchingFormulation.CONSERVATIVE)
is_periodic = (boundary==BoundaryCondition.PERIODIC)
@@ -52,7 +49,7 @@ class DirectionalStretchingRhsFunction(OpenClFunctionCodeGenerator):
vtype = typegen.vtype(ftype,dim)
(args,basename) = self.build_prototype(typegen,dim,itype,ftype,vtype,order,direction,cached,
- restrict,storage,vectorize_u,used_variables,sformulation,is_conservative,is_periodic)
+ restrict,storage,vectorize_u,used_variables,formulation,is_conservative,is_periodic)
reqs = self.build_requirements(typegen,dim,itype,ftype,vtype,order,direction,boundary,cached,
restrict,storage,vectorize_u,used_variables,is_conservative,is_periodic,args)
@@ -82,7 +79,7 @@ class DirectionalStretchingRhsFunction(OpenClFunctionCodeGenerator):
self.gencode()
def build_prototype(self,typegen,dim,itype,ftype,vtype,order,direction,cached,
- restrict,storage, vectorize_u,used_variables,sformulation,is_conservative,is_periodic):
+ restrict,storage, vectorize_u,used_variables,formulation,is_conservative,is_periodic):
U = used_variables['U']
W = used_variables['W']
@@ -115,7 +112,8 @@ class DirectionalStretchingRhsFunction(OpenClFunctionCodeGenerator):
args['Lx'] = CodegenVariable('Lx', itype, typegen, add_impl_const=True, nl=True)
args['active'] = CodegenVariable('active','bool',typegen)
- basename = 'stretching_rhs_{}_{}{}{}_fdc{}'.format(sformulation.lower(),ftype[0],dim,('v' if vectorize_u else ''),order)
+ basename = 'stretching_rhs_{}_{}{}{}_fdc{}'.format(str(formulation).lower(),
+ ftype[0],dim,('v' if vectorize_u else ''),order)
basename+='_'+XYZ[direction]
return (args,basename)
@@ -124,7 +122,7 @@ class DirectionalStretchingRhsFunction(OpenClFunctionCodeGenerator):
assert order%2==0
h = order/2
sg = StencilGenerator()
- sg.configure(derivative=2)
+ sg.configure(dim=1,derivative=1,order=order)
stencil = sg.generate_exact_stencil(origin=h)
return stencil
@@ -157,8 +155,16 @@ class DirectionalStretchingRhsFunction(OpenClFunctionCodeGenerator):
custom_id=None
known_args={'stride':1}
+ inv_dx_s = sm.Symbol('inv_dx')
+ inv_dx_var = CodegenVariable('inv_dx', ftype, typegen, add_impl_const=True, nl=True)
+
+ stencil = self.build_stencil(order)
+ stencil.replace_symbols({stencil.dx:1/inv_dx_s})
+ symbol2var = {inv_dx_s:inv_dx_var}
+
apply_stencil = ApplyStencilFunction(typegen=typegen,
- stencil=self.build_stencil(order),
+ stencil=stencil,
+ symbol2var=symbol2var,
ftype=ftype, itype=itype,
data_storage=storage,
vectorize=vectorize_u,
@@ -169,7 +175,6 @@ class DirectionalStretchingRhsFunction(OpenClFunctionCodeGenerator):
vector_suffixes = xyz,
op=op,
custom_id=custom_id,
- multipliers={'inv_dx':'ftype'},
known_args=known_args)
reqs['apply_stencil'] = apply_stencil
@@ -230,7 +235,6 @@ class DirectionalStretchingRhsFunction(OpenClFunctionCodeGenerator):
dw_dt = CodegenVectorClBuiltin('dW_dt', ftype, dim, tg)
ghosts = CodegenVariable('ghosts', itype, tg, const=True, value=order/2, symbolic_mode=True)
-
with s._function_():
s.jumpline()
diff --git a/hysop/backend/codegen/kernels/bandwidth.py b/hysop/backend/codegen/kernels/bandwidth.py
index 768be3aee..94ac571b7 100644
--- a/hysop/backend/codegen/kernels/bandwidth.py
+++ b/hysop/backend/codegen/kernels/bandwidth.py
@@ -12,12 +12,12 @@ from hysop.constants import np
from hysop.backend.codegen.base.opencl_codegen import OpenClCodeGenerator
from hysop.backend.codegen.base.kernel_codegen import KernelCodeGenerator
from hysop.backend.codegen.base.variables import CodegenVariable, CodegenVectorClBuiltin
-from hysop.backend.codegen.base.types import OpenClTypeGen
+from hysop.backend.opencl.opencl_types import OpenClTypeGen
from hysop.backend.codegen.base.utils import WriteOnceDict, ArgDict
from hysop.backend.opencl import cl, clCharacterize
-from hysop.backend.opencl.tools import OpenCLEnvironment
-from hysop.backend.opencl.gpu_kernel import KernelLauncher
+from hysop.backend.opencl.opencl_env import OpenClEnvironment
+from hysop.backend.opencl.opencl_kernel import OpenClKernelLauncher
from hysop.backend.opencl.kernel_autotuner import KernelAutotuner, AutotunerConfig
class BandwidthKernel(KernelCodeGenerator):
diff --git a/hysop/backend/codegen/kernels/copy_kernel.py b/hysop/backend/codegen/kernels/copy_kernel.py
index 9299316e4..2274e881b 100644
--- a/hysop/backend/codegen/kernels/copy_kernel.py
+++ b/hysop/backend/codegen/kernels/copy_kernel.py
@@ -12,12 +12,12 @@ from hysop.constants import np
from hysop.backend.codegen.base.opencl_codegen import OpenClCodeGenerator
from hysop.backend.codegen.base.kernel_codegen import KernelCodeGenerator
from hysop.backend.codegen.base.variables import CodegenVariable, CodegenVectorClBuiltin
-from hysop.backend.codegen.base.types import OpenClTypeGen
+from hysop.backend.opencl.opencl_types import OpenClTypeGen
from hysop.backend.codegen.base.utils import WriteOnceDict, ArgDict
from hysop.backend.opencl import cl, clCharacterize
-from hysop.backend.opencl.tools import OpenCLEnvironment
-from hysop.backend.opencl.gpu_kernel import KernelLauncher
+from hysop.backend.opencl.opencl_env import OpenClEnvironment
+from hysop.backend.opencl.opencl_kernel import OpenClKernelLauncher
from hysop.backend.opencl.kernel_autotuner import KernelAutotuner, AutotunerConfig
class CopyKernel(KernelCodeGenerator):
@@ -115,8 +115,8 @@ class CopyKernel(KernelCodeGenerator):
def autotune(cl_env, src, dst, vtype, count, restrict,
build_opts, autotuner_config):
- if not isinstance(cl_env,OpenCLEnvironment):
- raise ValueError('cl_env is not an OpenCLEnvironment.')
+ if not isinstance(cl_env,OpenClEnvironment):
+ raise ValueError('cl_env is not an OpenClEnvironment.')
if not isinstance(cl_env.typegen,OpenClTypeGen):
raise ValueError('typegen is not an OpenClTypeGen.')
@@ -211,7 +211,7 @@ class CopyKernel(KernelCodeGenerator):
kernel_args=kernel_args, vectorized=vectorized,
force_verbose=None,force_debug=None)
- kernel_launcher = KernelLauncher(kernel, queue, list(gwi), list(lwi))
+ kernel_launcher = OpenClKernelLauncher(kernel, queue, list(gwi), list(lwi))
return (stats, kernel_launcher, kernel_args, kernel_args_mapping, cached_bytes)
candidates = [i for i in typegen.vsizes if work_size%i==0]
diff --git a/hysop/backend/codegen/kernels/directional_advection.py b/hysop/backend/codegen/kernels/directional_advection.py
index ec0741484..269e6b1a1 100644
--- a/hysop/backend/codegen/kernels/directional_advection.py
+++ b/hysop/backend/codegen/kernels/directional_advection.py
@@ -8,15 +8,16 @@ from hysop import __VERBOSE__, __KERNEL_DEBUG__
from hysop.backend.opencl import cl
from hysop.tools.misc import Utils
+from hysop.tools.types import check_instance
from hysop.constants import np
-from hysop.constants import DirectionLabels, MeshDirection, BoundaryCondition
+from hysop.constants import DirectionLabels, BoundaryCondition
from hysop.backend.codegen.base.opencl_codegen import OpenClCodeGenerator
from hysop.backend.codegen.base.kernel_codegen import KernelCodeGenerator
from hysop.backend.codegen.base.variables import CodegenVariable, CodegenVectorClBuiltin, \
CodegenArray
-from hysop.backend.codegen.base.types import OpenClTypeGen
+from hysop.backend.opencl.opencl_types import OpenClTypeGen
from hysop.backend.codegen.base.utils import WriteOnceDict, ArgDict
from hysop.backend.codegen.base.statistics import WorkStatistics
@@ -31,10 +32,10 @@ from hysop.backend.codegen.functions.advection_rhs import DirectionalAdvectionR
from hysop.numerics.odesolvers.runge_kutta import ExplicitRungeKutta
from hysop.backend.opencl import cl, clCharacterize
-from hysop.backend.opencl.tools import OpenCLEnvironment
-from hysop.backend.opencl.gpu_kernel import KernelLauncher
+from hysop.backend.opencl.opencl_env import OpenClEnvironment
+from hysop.backend.opencl.opencl_kernel import OpenClKernelLauncher
from hysop.backend.opencl.kernel_autotuner import KernelAutotuner, AutotunerFlags, KernelGenerationError
-from hysop.backend.opencl.gpu_discrete import GPUDiscreteField
+from hysop.backend.opencl.opencl_discrete import OpenClDiscreteField
class DirectionalAdvectionKernel(KernelCodeGenerator):
@@ -54,10 +55,9 @@ class DirectionalAdvectionKernel(KernelCodeGenerator):
assert work_dim>0 and work_dim<=3
assert nparticles in [1,2,4,8,16]
- assert boundary in BoundaryCondition.entries()
- assert isinstance(rk_scheme, ExplicitRungeKutta)
+ check_instance(boundary,BoundaryCondition)
+ check_instance(rk_scheme, ExplicitRungeKutta)
- boundary = BoundaryCondition.value(boundary)
is_periodic = (boundary==BoundaryCondition.PERIODIC)
assert boundary in [BoundaryCondition.PERIODIC, BoundaryCondition.NONE]
assert (is_periodic and not is_cached) or min_ghosts>0
@@ -484,8 +484,8 @@ class DirectionalAdvectionKernel(KernelCodeGenerator):
build_opts,
autotuner_config):
- if not isinstance(cl_env,OpenCLEnvironment):
- raise ValueError('cl_env is not an OpenCLEnvironment.')
+ if not isinstance(cl_env,OpenClEnvironment):
+ raise ValueError('cl_env is not an OpenClEnvironment.')
if not isinstance(cl_env.typegen,OpenClTypeGen):
raise ValueError('typegen is not an OpenClTypeGen.')
@@ -517,8 +517,8 @@ class DirectionalAdvectionKernel(KernelCodeGenerator):
.format(rk_scheme.__class__.__name__)
raise ValueError(msg)
- if not isinstance(velocity,GPUDiscreteField):
- raise ValueError('velocity is not a GPUDiscreteField')
+ if not isinstance(velocity,OpenClDiscreteField):
+ raise ValueError('velocity is not a OpenClDiscreteField')
if (velocity.nb_components != dim):
raise ValueError('Velocity vector components mismatch with dim {}.'.format(dim))
if not isinstance(position,cl.Buffer):
@@ -531,13 +531,10 @@ class DirectionalAdvectionKernel(KernelCodeGenerator):
raise ValueError('compute_resolution mismatch in mesh_info.local_mesh, got {} and {}.'\
.format(compute_resolution,cr))
- if boundary not in BoundaryCondition.entries():
+ if not isinstance(boundaryn, BoundaryCondition):
msg = 'Unknown boundary condition \'{}\', valid ones are {}.'\
- .format(boundary,BoundaryCondition.fields().keys())
+ .format(boundary,BoundaryCondition.svalues())
raise ValueError(msg)
- else:
- boundary = BoundaryCondition.value(boundary)
- sboundary = BoundaryCondition.svalue(boundary)
ghosts = mesh_info['ghosts'].value[:dim]
cache_ghosts = DirectionalAdvectionKernel.cache_ghosts(max_dt, velocity_dx, max_velocity)
@@ -661,7 +658,7 @@ class DirectionalAdvectionKernel(KernelCodeGenerator):
force_verbose=None,force_debug=None,
return_codegen=True)
- kernel_launcher = KernelLauncher(kernel, queue, list(gwi), list(lwi))
+ kernel_launcher = OpenClKernelLauncher(kernel, queue, list(gwi), list(lwi))
total_work = work_size[0]*work_size[1]*work_size[2]
# per_work_statistics = codegen.per_work_statistics()
diff --git a/hysop/backend/codegen/kernels/directional_stretching.py b/hysop/backend/codegen/kernels/directional_stretching.py
index 72c2e83a6..b4b2196f2 100644
--- a/hysop/backend/codegen/kernels/directional_stretching.py
+++ b/hysop/backend/codegen/kernels/directional_stretching.py
@@ -4,6 +4,7 @@ from hysop.deps import np, operator, hashlib, contextlib
from hysop import __VERBOSE__, __KERNEL_DEBUG__
from hysop.tools.misc import Utils
+from hysop.tools.types import check_instance
from hysop.constants import DirectionLabels, BoundaryCondition
@@ -13,7 +14,7 @@ from hysop.backend.codegen.base.opencl_codegen import OpenClCodeGenerator
from hysop.backend.codegen.base.kernel_codegen import KernelCodeGenerator
from hysop.backend.codegen.base.variables import CodegenVariable, CodegenVectorClBuiltin, \
CodegenArray
-from hysop.backend.codegen.base.types import OpenClTypeGen
+from hysop.backend.opencl.opencl_types import OpenClTypeGen
from hysop.backend.codegen.base.utils import WriteOnceDict, ArgDict
from hysop.backend.codegen.base.statistics import WorkStatistics
@@ -28,7 +29,7 @@ from hysop.backend.codegen.functions.stretching_rhs import DirectionalStretching
from hysop.numerics.odesolvers.runge_kutta import ExplicitRungeKutta
from hysop.backend.opencl import cl, clCharacterize
-from hysop.backend.opencl.clenv import OpenCLEnvironment
+from hysop.backend.opencl.opencl_env import OpenClEnvironment
from hysop.backend.opencl.opencl_kernel import OpenClKernelLauncher
from hysop.backend.opencl.kernel_autotuner import KernelAutotuner, AutotunerFlags, KernelGenerationError
from hysop.backend.opencl.opencl_discrete import OpenClDiscreteField
@@ -36,9 +37,9 @@ from hysop.backend.opencl.opencl_discrete import OpenClDiscreteField
class DirectionalStretchingKernel(KernelCodeGenerator):
@staticmethod
- def codegen_name(ftype,is_cached,direction,sformulation):
+ def codegen_name(ftype,is_cached,direction,formulation):
cache = 'cached_' if is_cached else ''
- sformulation = sformulation.lower()
+ sformulation = str(formulation).lower()
return 'directional_{}stretching_{}_{}{}'.format(cache,sformulation,
ftype[0],DirectionLabels[direction])
@@ -52,19 +53,15 @@ class DirectionalStretchingKernel(KernelCodeGenerator):
assert dim==3
assert direction<dim
assert order>1 and order%2==0
- assert formulation in StretchingFormulation.entries()
- assert boundary in BoundaryCondition.entries()
- assert isinstance(rk_scheme, ExplicitRungeKutta)
+ check_instance(formulation, StretchingFormulation)
+ check_instance(boundary, BoundaryCondition)
+ check_instance(rk_scheme,ExplicitRungeKutta)
if known_vars is None:
known_vars = {}
- boundary = BoundaryCondition.value(boundary)
assert boundary in [BoundaryCondition.NONE,BoundaryCondition.PERIODIC]
- formulation = StretchingFormulation.value(formulation)
- sformulation = StretchingFormulation.svalue(formulation)
-
local_size_known = ('local_size' in known_vars)
is_conservative = (formulation==StretchingFormulation.CONSERVATIVE)
is_periodic = (boundary==BoundaryCondition.PERIODIC)
@@ -77,7 +74,7 @@ class DirectionalStretchingKernel(KernelCodeGenerator):
if is_conservative and not is_cached:
raise ValueError('Conservetive stretching requires caching.')
- name = DirectionalStretchingKernel.codegen_name(ftype,is_cached,direction,sformulation)
+ name = DirectionalStretchingKernel.codegen_name(ftype,is_cached,direction,formulation)
kernel_reqs = self.build_requirements(typegen, dim, ftype, order, is_cached,
rk_scheme, direction, boundary, symbolic_mode, formulation, storage,
@@ -104,7 +101,6 @@ class DirectionalStretchingKernel(KernelCodeGenerator):
self.boundary = boundary
self.rk_scheme = rk_scheme
self.formulation = formulation
- self.sformulation = sformulation
self.storage = storage
self.local_size_known = local_size_known
self.is_conservative = is_conservative
@@ -623,8 +619,8 @@ class DirectionalStretchingKernel(KernelCodeGenerator):
mesh_info, dt,
build_opts, autotuner_config):
- if not isinstance(cl_env,OpenCLEnvironment):
- raise ValueError('cl_env is not an OpenCLEnvironment.')
+ if not isinstance(cl_env,OpenClEnvironment):
+ raise ValueError('cl_env is not an OpenClEnvironment.')
if not isinstance(cl_env.typegen,OpenClTypeGen):
raise ValueError('typegen is not an OpenClTypeGen.')
@@ -657,13 +653,10 @@ class DirectionalStretchingKernel(KernelCodeGenerator):
msg = 'Given Runge-Kutta scheme is not an instance of ExplicitRungeKutta, got a \'{}\'.'\
.format(rk_scheme.__class__.__name__)
raise TypeError(msg)
- if formulation not in StretchingFormulation.entries():
+ if not isinstance(formulation, StretchingFormulation):
msg = 'unknown stretching formulation \'{}\', valid ones are {}.'\
- .format(formulation,StretchingFormulation.svalues())
+ .format(formulation.svalues())
raise TypeError(msg)
- else:
- formulation = StretchingFormulation.value(formulation)
- sformulation = StretchingFormulation.svalue(formulation)
if not isinstance(velocity,OpenClDiscreteField):
raise ValueError('velocity is not a OpenClDiscreteField')
@@ -678,13 +671,10 @@ class DirectionalStretchingKernel(KernelCodeGenerator):
raise ValueError('compute_resolution mismatch in mesh_info.local_mesh, got {} and {}.'\
.format(compute_resolution,cr))
- if boundary not in BoundaryCondition.entries():
+ if not isinstance(boundary, BoundaryCondition):
msg = 'Unknown boundary condition \'{}\', valid ones are {}.'\
- .format(boundary,BoundaryCondition.fields().keys())
+ .format(boundary,BoundaryCondition.svalues())
raise TypeError(msg)
- else:
- boundary = BoundaryCondition.value(boundary)
- sboundary = BoundaryCondition.svalue(boundary)
ghosts = mesh_info['ghosts'].value[:dim]
min_ghosts = DirectionalStretchingKernel.min_ghosts(boundary, formulation,
@@ -742,7 +732,7 @@ class DirectionalStretchingKernel(KernelCodeGenerator):
## CodeGenerator
cached=True
- name = DirectionalStretchingKernel.codegen_name(ftype, cached, direction, sformulation)
+ name = DirectionalStretchingKernel.codegen_name(ftype, cached, direction, formulation)
codegen = DirectionalStretchingKernel(typegen=typegen,
order=order, dim=dim, direction=direction, boundary=boundary,
formulation=formulation, rk_scheme=rk_scheme,
@@ -771,7 +761,7 @@ class DirectionalStretchingKernel(KernelCodeGenerator):
return (kernel, kernel_args, src_hash, global_size)
## Kernel Autotuner
- name = DirectionalStretchingKernel.codegen_name(ftype, False, direction, sformulation)
+ name = DirectionalStretchingKernel.codegen_name(ftype, False, direction, formulation)
autotuner = KernelAutotuner(name=name, work_dim=dim, local_work_dim=1,
build_opts=build_opts, autotuner_config=autotuner_config)
diff --git a/hysop/backend/codegen/kernels/empty.py b/hysop/backend/codegen/kernels/empty.py
index 0825ec983..c9a95056b 100644
--- a/hysop/backend/codegen/kernels/empty.py
+++ b/hysop/backend/codegen/kernels/empty.py
@@ -1,7 +1,7 @@
from hysop.backend.codegen.base.variables import CodegenVectorClBuiltin
from hysop.backend.codegen.base.kernel_codegen import KernelCodeGenerator
-from hysop.backend.codegen.base.types import OpenClTypeGen
+from hysop.backend.opencl.opencl_types import OpenClTypeGen
from hysop.backend.codegen.base.utils import ArgDict
class EmptyKernel(KernelCodeGenerator):
diff --git a/hysop/backend/codegen/kernels/stretching.py b/hysop/backend/codegen/kernels/stretching.py
index f54ff811b..f370e5281 100644
--- a/hysop/backend/codegen/kernels/stretching.py
+++ b/hysop/backend/codegen/kernels/stretching.py
@@ -5,7 +5,7 @@ import numpy as np
from hysop.backend.codegen.base.opencl_codegen import OpenClCodeGenerator
from hysop.backend.codegen.base.kernel_codegen import KernelCodeGenerator
from hysop.backend.codegen.base.variables import CodegenVariable, CodegenVectorClBuiltin
-from hysop.backend.codegen.base.types import OpenClTypeGen
+from hysop.backend.opencl.opencl_types import OpenClTypeGen
from hysop.backend.codegen.base.utils import WriteOnceDict, ArgDict
from hysop.backend.codegen.structs.mesh_info import MeshBaseStruct, MeshInfoStruct
diff --git a/hysop/backend/codegen/kernels/tests/test_directional_advection.py b/hysop/backend/codegen/kernels/tests/test_directional_advection.py
index 62929b0cd..9e75de89f 100644
--- a/hysop/backend/codegen/kernels/tests/test_directional_advection.py
+++ b/hysop/backend/codegen/kernels/tests/test_directional_advection.py
@@ -3,9 +3,10 @@ import copy, math
from hysop import __ENABLE_LONG_TESTS__
from hysop.backend.opencl import cl
+from hysop.tools.types import check_instance
from hysop.constants import np, BoundaryCondition
from hysop.backend.codegen.base.test import _test_mesh_info , _test_typegen
-from hysop.methods import ExplicitRungeKutta
+from hysop.numerics.odesolvers.runge_kutta import ExplicitRungeKutta
from hysop.backend.codegen.kernels.directional_advection import DirectionalAdvectionKernel
class TestDirectionalAdvection(object):
@@ -220,7 +221,7 @@ class TestDirectionalAdvection(object):
.format('cached ' if cached else '',
work_dim,
rk_scheme.name(),
- BoundaryCondition.svalue(boundary).lower(),
+ str(boundary).lower(),
nparticles)
print msg
@@ -393,7 +394,7 @@ class TestDirectionalAdvection(object):
def _check_kernels(self, rk_scheme):
- assert isinstance(rk_scheme, ExplicitRungeKutta)
+ check_instance(rk_scheme,ExplicitRungeKutta)
cached=[False,True]
boundaries=[BoundaryCondition.PERIODIC]
diff --git a/hysop/backend/codegen/kernels/tests/test_directional_stretching.py b/hysop/backend/codegen/kernels/tests/test_directional_stretching.py
index 15fde42b8..235e864cc 100644
--- a/hysop/backend/codegen/kernels/tests/test_directional_stretching.py
+++ b/hysop/backend/codegen/kernels/tests/test_directional_stretching.py
@@ -3,10 +3,12 @@ import copy
from hysop import __ENABLE_LONG_TESTS__
from hysop.backend.opencl import cl
+from hysop.tools.types import check_instance
from hysop.constants import np, BoundaryCondition
from hysop.backend.codegen.base.test import _test_mesh_info , _test_typegen
from hysop.backend.codegen.kernels.directional_stretching import DirectionalStretchingKernel
-from hysop.methods import StretchingFormulation, ExplicitRungeKutta
+from hysop.methods import StretchingFormulation
+from hysop.numerics.odesolvers.runge_kutta import ExplicitRungeKutta
class TestDirectionalStretching(object):
@@ -201,8 +203,7 @@ class TestDirectionalStretching(object):
W1 = [W0[i] + dt*K0[i] for i in xrange(3)]
W = W1
else:
- msg = 'Unknown stretching formulation scheme {}.'.format(
- StretchingFormulation.svalue(formulation))
+ msg = 'Unknown stretching formulation scheme {}.'.format(formulation)
raise ValueError(msg)
elif rk_scheme.name() == 'RK2':
Wc = [ w.copy() for w in W ]
@@ -238,8 +239,7 @@ class TestDirectionalStretching(object):
W2 = [W0[i] + 1.0*dt*K1[i] for i in xrange(3)]
W = W2
else:
- msg = 'Unknown stretching formulation scheme {}.'.format(
- StretchingFormulation.svalue(formulation))
+ msg = 'Unknown stretching formulation scheme {}.'.format(formulation)
raise ValueError(msg)
elif rk_scheme.name() == 'RK4':
Wc = [ w.copy() for w in W ]
@@ -297,8 +297,7 @@ class TestDirectionalStretching(object):
W4 = [W0[i] + dt*K[i] for i in xrange(3)]
W = W4
else:
- msg = 'Unknown stretching formulation scheme {}.'.format(
- StretchingFormulation.svalue(formulation))
+ msg = 'Unknown stretching formulation scheme {}.'.format(formulation)
raise ValueError(msg)
else:
msg = 'Unknown Runge-Kutta scheme {}.'.format(rk_scheme)
@@ -312,11 +311,11 @@ class TestDirectionalStretching(object):
msg = '\nTesting {}{} with order {} and scheme {} in direction {} with {} boundaries.'\
.format('cached ' if cached else '',
- StretchingFormulation.svalue(formulation).lower(),
+ str(formulation).lower(),
order,
rk_scheme.name(),
direction,
- BoundaryCondition.svalue(boundary).lower())
+ str(boundary).lower())
print msg
dt = self.dt
@@ -463,8 +462,8 @@ class TestDirectionalStretching(object):
def _check_kernels(self, formulation, rk_scheme):
- assert formulation in StretchingFormulation.entries()
- assert isinstance(rk_scheme, ExplicitRungeKutta)
+ check_instance(formulation,StretchingFormulation)
+ check_instance(rk_scheme,ExplicitRungeKutta)
cached=[False,True]
boundaries=[BoundaryCondition.NONE, BoundaryCondition.PERIODIC]
@@ -558,7 +557,7 @@ class TestDirectionalStretching(object):
if __name__ == '__main__':
- TestDirectionalStretching.setup_class(do_extra_tests=False, enable_error_plots=True)
+ TestDirectionalStretching.setup_class(do_extra_tests=True, enable_error_plots=False)
test = TestDirectionalStretching()
test.test_stretching_gradUW_Euler()
diff --git a/hysop/backend/codegen/kernels/transpose.py b/hysop/backend/codegen/kernels/transpose.py
index d3723de2f..43d94839d 100644
--- a/hysop/backend/codegen/kernels/transpose.py
+++ b/hysop/backend/codegen/kernels/transpose.py
@@ -4,7 +4,7 @@ import numpy as np
from hysop.backend.codegen.base.opencl_codegen import OpenClCodeGenerator
from hysop.backend.codegen.base.kernel_codegen import KernelCodeGenerator
from hysop.backend.codegen.base.variables import CodegenVariable, CodegenVectorClBuiltin, CodegenArray
-from hysop.backend.codegen.base.types import OpenClTypeGen
+from hysop.backend.opencl.opencl_types import OpenClTypeGen
from hysop.backend.codegen.base.utils import WriteOnceDict, ArgDict
from hysop.backend.codegen.functions.compute_index import ComputeIndexFunction
diff --git a/hysop/backend/codegen/structs/mesh_info.py b/hysop/backend/codegen/structs/mesh_info.py
index f73a063ef..3512c5307 100644
--- a/hysop/backend/codegen/structs/mesh_info.py
+++ b/hysop/backend/codegen/structs/mesh_info.py
@@ -1,16 +1,9 @@
-from hysop.constants import np
+from hysop.deps import np
+from hysop.tools.types import check_instance
from hysop.backend.codegen.base.enum_codegen import EnumCodeGenerator
from hysop.backend.codegen.base.struct_codegen import StructCodeGenerator
-from hysop.backend.codegen.base.types import OpenClTypeGen
-
-# class MeshDirectionEnum(EnumCodeGenerator):
- # def __init__(self):
- # super(MeshDirectionEnum,self).__init__(MeshDirection)
-
-# class TranspositionStateEnum(EnumCodeGenerator):
- # def __init__(self):
- # super(TranspositionStateEnum,self).__init__(TranspositionState)
+from hysop.backend.opencl.opencl_types import OpenClTypeGen
class MeshBaseStruct(StructCodeGenerator):
def __init__(self, typegen, typedef=None):
@@ -119,21 +112,6 @@ class MeshInfoStruct(StructCodeGenerator):
_append( ('global_mesh', mesh_base.dtype) )
i+=2
- # mesh_dir = MeshDirectionEnum()
- # mesh_dir_field = i
- # _append( ('direction', mesh_dir.dtype) )
- # i+=1
-
- # mesh_state = TranspositionStateEnum()
- # mesh_state_field = i
- # _append( ('state', mesh_state.dtype) )
- # i+=1
-
- # ctype_overrides = {
- # mesh_dir_field:mesh_dir.ctype,
- # mesh_state_field:mesh_state.ctype
- # }
-
comments = [
"Dimension of the mesh",
"Index of the first local compute point in the global grid",
@@ -143,11 +121,9 @@ class MeshInfoStruct(StructCodeGenerator):
"1/dx",
"Local mesh",
"Global mesh",
- "Current memory aligned component",
- "Current components reordering state"
]
- return dtypes, comments, ctype_overrides, [mesh_base, mesh_dir, mesh_state]
+ return dtypes, comments, None, [mesh_base]
def create(self,
name,
@@ -155,7 +131,6 @@ class MeshInfoStruct(StructCodeGenerator):
start, stop, ghosts,
dx,
local_mesh, global_mesh,
- mesh_dir, mesh_state,
**kargs):
if dim>4:
@@ -166,9 +141,6 @@ class MeshInfoStruct(StructCodeGenerator):
dtype,_,_,_ = MeshInfoStruct.build_dtype(tg)
- # direction = MeshDirection.value(mesh_dir)
- # state = TranspositionState.value(mesh_state)
-
dx = np.asarray(dx)
mesh_info_vals = {
'dim' : tg.make_intn(dim,1),
@@ -178,9 +150,7 @@ class MeshInfoStruct(StructCodeGenerator):
'dx' : tg.make_floatn(dx,4),
'inv_dx' : tg.make_floatn(1.0/dx,4),
'local_mesh' : local_mesh[0],
- 'global_mesh': global_mesh[0],
- # 'direction' : direction,
- # 'state' : state
+ 'global_mesh': global_mesh[0]
}
def extend(var,d=0):
@@ -204,8 +174,6 @@ class MeshInfoStruct(StructCodeGenerator):
var_overrides = dict(
local_mesh=local_mesh[1],
global_mesh=global_mesh[1],
- # direction = MeshDirection.variable('direction',tg,mesh_dir),
- # state = TranspositionState.variable('state',tg,mesh_state)
)
cg_var = self.build_codegen_variable(name=name,
value=value,var_overrides=var_overrides,
@@ -218,12 +186,9 @@ class MeshInfoStruct(StructCodeGenerator):
@staticmethod
def create_from_mesh(name,cl_env, mesh,
- # mesh_dir=MeshDirection.X, mesh_state=TranspositionState.XYZ,
**kargs):
from hysop.domain.mesh import Mesh
- assert isinstance(mesh,Mesh)
- # assert mesh_dir in MeshDirection.entries()
- # assert mesh_state in TranspositionState.entries()
+ check_instance(mesh,Mesh)
tg = cl_env.typegen
@@ -273,8 +238,6 @@ class MeshInfoStruct(StructCodeGenerator):
var_overrides = dict(
local_mesh = self.mesh_base.build_codegen_variable('local_mesh',const=const),
global_mesh = self.mesh_base.build_codegen_variable('global_mesh',const=const),
- # direction = MeshDirection.variable('direction',tg),
- # state = TranspositionState.variable('state',tg)
)
return super(MeshInfoStruct,self).build_codegen_variable(name=name,var_overrides=var_overrides,**kargs)
diff --git a/hysop/backend/opencl/cltypes.py b/hysop/backend/opencl/cltypes.py
deleted file mode 100644
index cbb8640a2..000000000
--- a/hysop/backend/opencl/cltypes.py
+++ /dev/null
@@ -1,333 +0,0 @@
-
-import string
-
-from hysop import __KERNEL_DEBUG__
-from hysop.constants import np, it
-from hysop.backend.opencl import cl, clArray
-from hysop.tools.numerics import MPZ, MPZ, MPFR, F2Q
-
-vsizes = [1,2,3,4,8,16]
-base_types = ['float','signed','unsigned']
-float_base_types = ['half','float','double']
-signed_base_types = ['char','short','int','long']
-unsigned_base_types = ['uchar','ushort','uint','ulong']
-
-float_types = []
-signed_types = []
-unsigned_types = []
-for b in base_types:
- b_base_types = eval(b+'_base_types')
- b_types = eval(b+'_types')
- for f,c in it.product(b_base_types,vsizes):
- if c==1:
- if f=='half': continue
- else: ftype = f
- else:
- ftype = f+str(c)
- b_types.append(ftype)
-integer_types = signed_types + unsigned_types
-builtin_types = integer_types + float_types
-
-
-float_base_type_require = {
- 'half' : 'cl_khr_fp16',
- 'float' : None,
- 'double': 'cl_khr_fp64'
-}
-
-FLT_DIG = {
- 'half' : 3, # = HALF_DIG
- 'float' : 6, # = FLT_DIG
- 'double': 15 # = DBL_DIG
-}
-FLT_MANT_DIG = {
- 'half' : 11, # = HALF_MANT_DIG
- 'float' : 24, # = FLT_MANT_DIG
- 'double': 53 # = DBL_MANT_DIG
-}
-FLT_LITERAL = {
- 'half' : 'h',
- 'float' : 'f',
- 'double': ''
-}
-FLT_BYTES = {
- 'half' : 2,
- 'float' : 4,
- 'double': 8
-}
-
-def basetype(fulltype):
- return fulltype.translate(None,string.digits)
-def components(fulltype):
- comp = fulltype.translate(None,string.ascii_letters)
- return 1 if comp == '' else int(comp)
-def mangle_vtype(fulltype):
- return basetype(fulltype)[0]+str(components(fulltype))
-
-def vtype(basetype,N):
- return basetype + ('' if N==1 else str(N))
-def itype(fulltype):
- N = components(fulltype)
- return 'int' + ('' if N==1 else str(N))
-def uitype(fulltype):
- N = components(fulltype)
- return 'uint' + ('' if N==1 else str(N))
-def np_dtype(fulltype):
- return cl.tools.get_or_register_dtype(fulltype)
-
-def vtype_component_adressing(i,mode='hex'):
- if mode=='hex': return '0123456789abcdef'[i]
- elif mode=='HEX': return '0123456789ABCDEF'[i]
- elif mode=='pos': return 'xyzw'[i]
- else: raise ValueError('Bad vtype component adressing mode!')
-
-def vtype_access(i,N,mode='hex'):
- assert i<N
- if N==1: return ''
- else: return ('s' if mode.lower()=='hex' else '') + vtype_component_adressing(i,mode)
-
-def float_to_hex_str(f,fbtype):
- sf = float(f).hex().split('0x') + ['']
- buf = sf[1].split('p')
-
- mantissa = buf[0]
- exponent = buf[1]
-
- mant_dig = FLT_MANT_DIG[fbtype]
- literal = FLT_LITERAL [fbtype]
-
- nhex = (mant_dig-1+3)//4 + 2
- # +2= leading one or zero and decimal point characters (1.abde... or 0.abcde...)
-
- sf[0] = ('+' if sf[0] == '' else sf[0])+'0x'
- sf[1] = mantissa[:nhex]
- sf[2] = 'p'+exponent+literal
- return ''.join(sf)
-
-def float_to_dec_str(f,fbtype):
- sf = float(f).__repr__().split('.')
- offset = (1 if sf[0][0] in ['-','+'] else 0)
- sf[1] = sf[1][:FLT_DIG[fbtype]-len(sf[0])+offset+1]
- return ('+' if f>0 else '') + '.'.join(sf)+FLT_LITERAL[fbtype]
-
-
-
-#pyopencl specific
-vec = clArray.vec
-
-def npmake(dtype):
- return lambda scalar: np.array([scalar], dtype=dtype)
-
-vtype_int = [np.int32, vec.int2, vec.int3, vec.int4, vec.int8, vec.int16 ]
-vtype_uint = [np.uint32, vec.uint2, vec.uint3, vec.uint4, vec.uint8, vec.uint16 ]
-vtype_simple = [np.float32, vec.float2, vec.float3, vec.float4, vec.float8, vec.float16 ]
-vtype_double = [np.float64, vec.double2, vec.double3, vec.double4, vec.double8, vec.double16 ]
-
-make_int = [npmake(np.int32), vec.make_int2, vec.make_int3,
- vec.make_int4, vec.make_int8,
- vec.make_int16 ]
-make_uint = [npmake(np.uint32), vec.make_uint2, vec.make_uint3,
- vec.make_uint4, vec.make_uint8,
- vec.make_uint16 ]
-make_simple = [npmake(np.float32), vec.make_float2, vec.make_float3,
- vec.make_float4, vec.make_float8,
- vec.make_float16 ]
-make_double = [npmake(np.float64), vec.make_double2, vec.make_double3,
- vec.make_double4, vec.make_double8,
- vec.make_double16 ]
-
-def simplen(n):
- if n==1: return np.float32
- i = vsizes.index(n)
- return vtype_simple[i]
-def doublen(n):
- if n==1: return np.float64
- i = vsizes.index(n)
- return vtype_double[i]
-def intn(n):
- if n==1: return np.int32
- i = vsizes.index(n)
- return vtype_int[i]
-def uintn(n):
- if n==1: return np.uint32
- i = vsizes.index(n)
- return vtype_uint[i]
-
-_typen = {
- 'float' : simplen,
- 'simple': simplen,
- 'double': doublen,
- 'int' : intn,
- 'uint' : uintn
-}
-
-
-def typen(btype,n):
- return _typen[btype](n)
-
-def cl_type_to_dtype(cl_type):
- btype = basetype(cl_type)
- N = components(cl_type)
- return typen(btype,N)
-
-def make_simplen(vals,n,dval=0):
- vals = (vals,) if np.isscalar(vals) else tuple(vals)
- vals += (dval,)*(n-len(vals))
- i = vsizes.index(n)
- return make_simple[i](*vals)
-def make_doublen(vals,n,dval=0):
- vals = (vals,) if np.isscalar(vals) else tuple(vals)
- vals += (dval,)*(n-len(vals))
- i = vsizes.index(n)
- return make_double[i](*vals)
-def make_intn(vals,n,dval=0):
- vals = (vals,) if np.isscalar(vals) else tuple(vals)
- vals += (dval,)*(n-len(vals))
- i = vsizes.index(n)
- return make_int[i](*vals)
-def make_uintn(vals,n,dval=0):
- vals = (vals,) if np.isscalar(vals) else tuple(vals)
- vals += (dval,)*(n-len(vals))
- i = vsizes.index(n)
- return make_uint[i](*vals)
-
-
-class TypeGen(object):
- def __init__(self, fbtype='float', float_dump_mode='dec'):
-
- self.float_base_types = float_base_types
- self.FLT_BYTES = FLT_BYTES
- self.FLT_DIG = FLT_DIG
- self.FLT_MANT_DIG = FLT_MANT_DIG
- self.FLT_LITERAL = FLT_LITERAL
-
- self.np_dtype = np_dtype
-
- self.float_to_dec_str = float_to_dec_str
- self.float_to_hex_str = float_to_hex_str
-
- self.fbtype = fbtype
-
- self.float_dump_mode = float_dump_mode
- if float_dump_mode in ['hex', 'hexadecimal']:
- self.float_to_str = float_to_hex_str
- elif float_dump_mode in ['dec','decimal']:
- self.float_to_str = float_to_dec_str
- else:
- raise ValueError('Unknown float_dump_mode \'{}\''.format(float_dump_mode))
-
- def dump(self, val):
- if val.__class__ in [list,tuple,dict,np.ndarray]:
- raise ValueError('Value is not a scalar, got {}.'.format(val))
- if val.__class__ in [float,np.float16,np.float32,np.float64]:
- sval = self.float_to_str(val, self.fbtype)
- return '({})'.format(sval)
- elif val.__class__ in [int,np.int8,np.int16,np.int32,np.int64,MPZ]:
- sign = ('' if val==0 else ('+' if val>0 else '-'))
- sval = str(val)
- if val<0:
- sval=sval[1:]
- if val!=0:
- sval = '({}{})'.format(sign,sval)
- else:
- sval = '0'
- return sval
- elif val.__class__ in [bool,np.bool_]:
- return 'true' if val else 'false'
- elif isinstance(val, MPQ):
- if __KERNEL_DEBUG__:
- return '({}.0{f}/{}.0{f})'.format(val.numerator,val.denominator,
- f=FLT_LITERAL[self.fbtype])
- else:
- return self.dump(float(val))
- else:
- return val.__str__()
-
-
-# struct type generation (type size and struct field offsets) is different for each device
-# depending on architecture and compiler implementation and features.
-# /!\ do not use the same opencl typegen instance for two different devices that are
-# not equivalent.
-class OpenClTypeGen(TypeGen):
- @staticmethod
- def devicelessTypegen():
- """
- Sometimes we do not need structs and code generation is device independent.
- """
- return OpenClTypeGen(device=None,context=None,platform=None);
-
- def __init__(self, device, context, platform,
- fbtype='float', float_dump_mode='dec'):
- super(OpenClTypeGen,self).__init__(fbtype,float_dump_mode)
-
- self.device = device
- self.context = context
- self.platform = platform
-
- self.vsizes = vsizes
- self.signed_base_types = signed_base_types
- self.unsigned_base_types = unsigned_base_types
- self.integer_base_types = signed_base_types + unsigned_base_types
-
- self.float_types = float_types
- self.signed_types = signed_types
- self.unsigned_types = unsigned_types
- self.integer_types = integer_types
- self.builtin_types = builtin_types
-
- self.float_base_type_require = float_base_type_require
-
- self.basetype = basetype
- self.components = components
- self.vtype = vtype
- self.itype = itype
- self.uitype = uitype
- self.np_dtype = np_dtype
-
- self.vtype_component_adressing = vtype_component_adressing
- self.vtype_access = vtype_access
- self.mangle_vtype = mangle_vtype
- self.float_to_dec_str = float_to_dec_str
- self.float_to_hex_str = float_to_hex_str
-
- #pyopencl specifics
- self.intn = intn
- self.uintn = uintn
- self.simplen = simplen
- self.doublen = doublen
- self.typen = typen
-
- self.make_intn = make_intn
- self.make_uintn = make_uintn
- self.make_simplen = make_simplen
- self.make_doublen = make_doublen
-
- if fbtype == 'float':
- self.floatn = simplen
- self.make_floatn = make_simplen
- elif fbtype == 'double':
- self.floatn = doublen
- self.make_floatn = make_doublen
- # elif fbtype == 'half':
- # self.floatn = halfn
- # self.make_floatn = make_halfn
- else:
- raise ValueError('Unknown fbtype \'{}\''.format(fbtype))
-
- def device_has_ftype(self,device):
- dev_exts = device.extensions.split(' ')
- req = self.float_base_type_require[self.fbtype]
- return (req is None) or (req[0] in dev_exts)
- def cl_requirements(self):
- return [self.float_base_type_require[self.fbtype]];
-
- def dtype_from_str(self,stype):
- stype = stype.replace('ftype', self.fbtype).replace('fbtype',self.fbtype)
- btype = basetype(stype)
- N = components(stype)
- return typen(btype,N)
-
- def __repr__(self):
- return '{}_{}_{}_{}'.format(self.platform.name,self.device.name,
- self.fbtype,self.float_dump_mode)
-
diff --git a/hysop/backend/opencl/discrete.py b/hysop/backend/opencl/discrete.py
index b58f92b09..a60a54135 100644
--- a/hysop/backend/opencl/discrete.py
+++ b/hysop/backend/opencl/discrete.py
@@ -4,7 +4,7 @@ from hysop import __VERBOSE__
from hysop.constants import HYSOP_ORDER, HYSOP_REAL, DirectionLabels
from hysop.fields.discrete import DiscreteField
from hysop.backend.opencl import cl
-from hysop.backend.opencl.gpu_kernel import KernelLauncher, KernelListLauncher
+from hysop.backend.opencl.opencl_kernel import OpenClKernelLauncher, KernelListLauncher
from hysop.backend.arrays.numpy_backend import NumpyBackend as npw
from hysop.backend.arrays.numpy_backend import OpenClBackend as npcl
@@ -30,7 +30,7 @@ class OpenClDiscreteField(DiscreteField):
Field name
"""
# init base class
- super(GPUDiscreteField, self).__init__(topology, is_vector, name)
+ super(OpenClDiscreteField, self).__init__(topology, is_vector, name)
# OpenCL environment
self.cl_env = cl_env
@@ -54,9 +54,9 @@ class OpenClDiscreteField(DiscreteField):
@classmethod
def from_field(cls, cl_env, vfield):
- if not isinstance(vfield, GPUDiscreteField):
+ if not isinstance(vfield, OpenClDiscreteField):
vfield.__class__ = cls
- GPUDiscreteField.__init__(
+ OpenClDiscreteField.__init__(
vfield, cl_env,
vfield.topology, vfield.nb_components > 1, vfield.name)
diff --git a/hysop/backend/opencl/opencl_discrete.py b/hysop/backend/opencl/opencl_discrete.py
index 9f0b63df2..639366bb4 100644
--- a/hysop/backend/opencl/opencl_discrete.py
+++ b/hysop/backend/opencl/opencl_discrete.py
@@ -48,7 +48,7 @@ class OpenClDiscreteField(DiscreteField):
self.precision = precision
# Memory used
self.mem_size = 0
- ## Initialization OpenCL kernel as KernelLauncher
+ ## Initialization OpenCL kernel as OpenClKernelLauncher
self.init_kernel = None
self._isReleased = False
## OpenCL Buffer pointer
@@ -157,7 +157,7 @@ class OpenClDiscreteField(DiscreteField):
def setInitializationKernel(self, kernel):
"""
Set the initialization kernel
- @param kernel : KernelLauncher to use for initialize field.
+ @param kernel : OpenClKernelLauncher to use for initialize field.
"""
self.init_kernel = kernel
@@ -194,7 +194,7 @@ class OpenClDiscreteField(DiscreteField):
t = self.precision(time)
if __VERBOSE__:
print "{" + str(self._rank) + "}", "Initialize", self.name
- isGPUKernel = isinstance(formula, KernelLauncher) \
+ isGPUKernel = isinstance(formula, OpenClKernelLauncher) \
or isinstance(formula, KernelListLauncher)
if not isGPUKernel and self.init_kernel is None:
DiscreteField.initialize(self, formula, False, time, *args)
diff --git a/hysop/backend/opencl/clenv.py b/hysop/backend/opencl/opencl_env.py
similarity index 99%
rename from hysop/backend/opencl/clenv.py
rename to hysop/backend/opencl/opencl_env.py
index ce6f84f3e..cc6510085 100644
--- a/hysop/backend/opencl/clenv.py
+++ b/hysop/backend/opencl/opencl_env.py
@@ -9,10 +9,10 @@ from hysop.backend.opencl import cl, clTools, __OPENCL_PROFILE__
from hysop.backend.opencl.tools import convert_device_type, convert_precision
from hysop.backend.opencl.tools import get_platform, get_context, get_device, \
create_queue, parse_opencl_file
-from hysop.backend.opencl.cltypes import OpenClTypeGen
+from hysop.backend.opencl.opencl_types import OpenClTypeGen
-class OpenCLEnvironment(object):
+class OpenClEnvironment(object):
"""OpenCL environment informations and useful functions.
"""
def __init__(self, platform_id = __DEFAULT_PLATFORM_ID__,
diff --git a/hysop/backend/opencl/tools.py b/hysop/backend/opencl/tools.py
index 6f0161795..470ea2621 100644
--- a/hysop/backend/opencl/tools.py
+++ b/hysop/backend/opencl/tools.py
@@ -1,6 +1,6 @@
"""Classes and tools used to handle the OpenCL backend.
-* :class:`~hysop.gpu.tools.OpenCLEnvironment`:
+* :class:`~hysop.gpu.tools.OpenClEnvironment`:
object handling opencl platform, device ... info.
* :func:`~hysop.gpu.tools.get_opengl_shared_environment`:
build or get an OpenCL environment with openGL properties.
@@ -129,17 +129,17 @@ def get_or_create_opencl_env(
precision = Precision.DEFAULT_PRECISION,
gl_sharing=False, comm=None):
"""
- Create or an OpenCLEnvironment from given parameters if it does not already exists.
+ Create or an OpenClEnvironment from given parameters if it does not already exists.
All environements are kept alive (cached) in a dictionary local to this
- function (ie. all opencl operators can share the same OpenCLEnvironment).
+ function (ie. all opencl operators can share the same OpenClEnvironment).
"""
key = (platform_id,device_id, device_type, gl_sharing, comm)
if key in get_or_create_opencl_env.opencl_environments:
return get_or_create_opencl_env.opencl_environments[key]
- from hysop.backend.opencl.clenv import OpenCLEnvironment
- env = OpenCLEnvironment(platform_id=platform_id, device_id=device_id,
+ from hysop.backend.opencl.opencl_env import OpenClEnvironment
+ env = OpenClEnvironment(platform_id=platform_id, device_id=device_id,
device_type=device_type, gl_sharing=gl_sharing, comm=comm)
get_or_create_opencl_env.opencl_environments[key] = env
diff --git a/hysop/deps.py b/hysop/deps.py
index fe6938772..29dd22101 100644
--- a/hysop/deps.py
+++ b/hysop/deps.py
@@ -25,7 +25,7 @@ from abc import ABCMeta, abstractmethod
import sys, os, subprocess, platform
import inspect, functools, operator
import hashlib, gzip, copy, types
-import math, re
+import math, re, contextlib
import itertools as it
import numpy as np
diff --git a/hysop/numerics/fftw_f/fft3d.f90 b/hysop/numerics/fftw_f/fft3d.f90
index 23580aca7..b3da0f41d 100755
--- a/hysop/numerics/fftw_f/fft3d.f90
+++ b/hysop/numerics/fftw_f/fft3d.f90
@@ -494,7 +494,7 @@ contains
call c_f_pointer(cbuffer1, rdatain_many, [howmany,2*halfLength,fft_resolution(c_Y),local_resolution(c_Z)])
call c_f_pointer(cbuffer1, dataout_many, [howmany,halfLength, fft_resolution(c_Z), local_resolution(c_Y)])
- ! create MPI plans for in-place forward/backward DFT (note dimension reversal)
+ ! create MPI plans for in-place forward/backward DFT (note dimension reversal)
n(3) = fft_resolution(c_X)
plan_forward1 = fftw_mpi_plan_many_dft_r2c(3,n,howmany,blocksize,blocksize, rdatain_many, dataout_many, &
diff --git a/hysop/numerics/stencil/stencil.py b/hysop/numerics/stencil/stencil.py
index f1844bcc7..95af6627d 100644
--- a/hysop/numerics/stencil/stencil.py
+++ b/hysop/numerics/stencil/stencil.py
@@ -8,8 +8,8 @@
"""
-from hysop.deps import sm, sp, it, np
-from hysop.tools.sympy_utils import recurse_expression_tree
+from hysop.deps import sm, sp, it, np, hashlib
+from hysop.tools.sympy_utils import recurse_expression_tree, expr2str
class Stencil(object):
"""
@@ -73,31 +73,42 @@ class Stencil(object):
:class:`StencilGenerator`: Generate Stencil objects.
"""
+ coeffs = np.atleast_1d(coeffs)
+ origin = np.atleast_1d(origin)
+ order = np.atleast_1d(order)
+ dx = np.atleast_1d(dx)
+
if (origin<0).any():
raise ValueError('Origin component < 0!\norigin={}'.format(origin))
- if coeffs.ndim==1:
- origin = origin if np.isscalar(origin) else origin[0]
- order = order if np.isscalar(order) else order[0]
- dx = dx if isinstance(dx,sm.Symbol) else dx[0]
- else:
- coeffs = np.asarray(coeffs)
- origin = np.asarray(origin)
- order = np.asarray(order)
-
- if isinstance(dx, list):
- dx = np.asarray(dx)
-
- self.dx = dx
+ self.dx = dx[0] if dx.size==1 else dx
self.error = error
- self.origin = origin
- self.order = order
+ self.origin = origin[0] if origin.size==1 else origin
+ self.order = order[0] if order.size==1 else order
self.factor = factor
self.coeffs = self._delete_zeros(coeffs)
self._update_attributes()
+ def format_factor(self, svars):
+ return expr2str(self.factor,svars)
+
+ def has_factor(self):
+ return (self.factor!=1)
+
+ def non_zero_coefficients(self):
+ return np.sum(self.coeffs!=0)
+
+ def replace_symbols(self, dic):
+ if isinstance(self.factor,sm.Basic):
+ self.factor = self.factor.xreplace(dic)
+ coeffs = self.coeffs.ravel()
+ for i,coeff in enumerate(coeffs):
+ if isinstance(coeff,sm.Basic):
+ coeffs[i] = coeff.xreplace(dic)
+
+
def _update_attributes(self):
self.dim = self.coeffs.ndim
self.shape = self.coeffs.shape
@@ -201,11 +212,10 @@ class Stencil(object):
Zipped offset and coefficient iterator.
"""
factor = self.factor if include_factor else 1
+ svars = dict(zip(svars.keys(),[str(v) for v in svars.values()]))
def mapfun(x):
offset = x-self.origin
- value = factor*self.coeffs[x]
- if isinstance(value, sm.Expr):
- value = value.xreplace(svars)
+ value = expr2str(factor*self.coeffs[x],svars)
return (offset,value)
iterator = np.ndindex(self.shape)
iterator = it.imap(mapfun, iterator)
diff --git a/hysop/old/gpu.old/QtRendering.py b/hysop/old/gpu.old/QtRendering.py
index fa769dcab..3092f4b52 100644
--- a/hysop/old/gpu.old/QtRendering.py
+++ b/hysop/old/gpu.old/QtRendering.py
@@ -10,8 +10,8 @@ from PyQt4.QtOpenGL import QGLWidget
import OpenGL.GL as gl
from hysop.backend.opencl.tools import get_opengl_shared_environment
from hysop.backend.opencl import cl
-from hysop.backend.opencl.gpu_discrete import GPUDiscreteField
-from hysop.backend.opencl.gpu_kernel import KernelLauncher
+from hysop.backend.opencl.opencl_discrete import OpenClDiscreteField
+from hysop.backend.opencl.opencl_kernel import OpenClKernelLauncher
from hysop.core.mpi import main_rank
from hysop.operator.computational import Computational
from hysop.tools.numpywrappers import npw
@@ -74,7 +74,7 @@ class QtOpenGLRendering(Computational):
"""
## GPU scalar field
for df in self.variables[0].discrete_fields.values():
- if isinstance(df, GPUDiscreteField):
+ if isinstance(df, OpenClDiscreteField):
self.gpu_field = df
# Create OpenGL VBOs
## VBO for coordinates
@@ -138,11 +138,11 @@ class QtOpenGLRendering(Computational):
gwi = self.gpu_field.data[self.component].shape
else:
gwi = self.gpu_field.data[0].shape
- self.initCoordinates = KernelLauncher(
+ self.initCoordinates = OpenClKernelLauncher(
self.prg.initPointCoordinates, self.window.widget.cl_env.queue,
gwi, None)
## OpenCL kernel for computing colors
- self.numMethod = KernelLauncher(
+ self.numMethod = OpenClKernelLauncher(
self.prg.colorize, self.window.widget.cl_env.queue,
gwi, None)
diff --git a/hysop/old/gpu.old/gpu_diffusion.py b/hysop/old/gpu.old/gpu_diffusion.py
index bc69254f9..b6fe69583 100644
--- a/hysop/old/gpu.old/gpu_diffusion.py
+++ b/hysop/old/gpu.old/gpu_diffusion.py
@@ -10,8 +10,8 @@ from hysop.operator.discrete.discrete import DiscreteOperator
from hysop.operator.discrete.discrete import get_extra_args_from_method
from hysop.backend.opencl import cl
from hysop.backend.opencl.gpu_operator import GPUOperator
-from hysop.backend.opencl.gpu_kernel import KernelLauncher
-from hysop.backend.opencl.gpu_discrete import GPUDiscreteField
+from hysop.backend.opencl.opencl_kernel import OpenClKernelLauncher
+from hysop.backend.opencl.opencl_discrete import OpenClDiscreteField
from hysop.tools.profiler import FProfiler
from hysop.core.mpi import Wtime
@@ -38,8 +38,8 @@ class GPUDiffusion(DiscreteOperator, GPUOperator):
**kwds)
## GPU allocation.
- alloc = not isinstance(self.field, GPUDiscreteField)
- GPUDiscreteField.fromField(self.cl_env, self.field,
+ alloc = not isinstance(self.field, OpenClDiscreteField)
+ OpenClDiscreteField.fromField(self.cl_env, self.field,
self.gpu_precision, simple_layout=False)
if not self.field.gpu_allocated:
self.field.allocate()
@@ -159,9 +159,9 @@ class GPUDiffusion(DiscreteOperator, GPUOperator):
build_options += " -D NB_GROUPS_I={0}".format(blocs_nb[0])
build_options += " -D NB_GROUPS_II={0}".format(blocs_nb[1])
prg = self.cl_env.build_src(src, build_options, vec)
- self.num_diffusion = KernelLauncher(
+ self.num_diffusion = OpenClKernelLauncher(
prg.diffusion, self.cl_env.queue, gwi, lwi)
- self.copy = KernelLauncher(cl.enqueue_copy,
+ self.copy = OpenClKernelLauncher(cl.enqueue_copy,
self.cl_env.queue)
def _compute_diffusion(self, simulation):
diff --git a/hysop/old/gpu.old/gpu_discrete.py b/hysop/old/gpu.old/gpu_discrete.py
index 0b5e18343..c38245cb2 100644
--- a/hysop/old/gpu.old/gpu_discrete.py
+++ b/hysop/old/gpu.old/gpu_discrete.py
@@ -5,7 +5,7 @@ from hysop.constants import ORDER, np,\
debug, HYSOP_REAL, DirectionLabels
from hysop.fields.discrete import DiscreteField
from hysop.backend.opencl import cl, CL_PROFILE
-from hysop.backend.opencl.gpu_kernel import KernelLauncher, KernelListLauncher
+from hysop.backend.opencl.opencl_kernel import OpenClKernelLauncher, KernelListLauncher
from hysop.tools.profiler import FProfiler
fromLayoutMgrFunc_3D_seq = [
@@ -65,7 +65,7 @@ toLayoutMgrFunc_1D = [
]
-class GPUDiscreteField(DiscreteField):
+class OpenClDiscreteField(DiscreteField):
"""GPU Discrete vector field implementation.
Allocates OpenCL device memory for the field.
"""
@@ -94,14 +94,14 @@ class GPUDiscreteField(DiscreteField):
indicates if in the Z direction, layout is ZYX (simple) or ZXY.
"""
# init base class
- super(GPUDiscreteField, self).__init__(topology, is_vector, name)
+ super(OpenClDiscreteField, self).__init__(topology, is_vector, name)
# OpenCL environment
self.cl_env = cl_env
# Precision for the field
self.precision = precision
# Memory used
self.mem_size = 0
- ## Initialization OpenCL kernel as KernelLauncher
+ ## Initialization OpenCL kernel as OpenClKernelLauncher
self.init_kernel = None
self._isReleased = False
## OpenCL Buffer pointer
@@ -200,9 +200,9 @@ class GPUDiscreteField(DiscreteField):
@param layout : Boolean indicating if components are arranged in memory
@param simple_layout : Boolean indicating if in the Z direction,
"""
- if not isinstance(vfield, GPUDiscreteField):
+ if not isinstance(vfield, OpenClDiscreteField):
vfield.__class__ = cls
- GPUDiscreteField.__init__(
+ OpenClDiscreteField.__init__(
vfield, cl_env,
vfield.topology, vfield.nb_components > 1, vfield.name,
precision, layout, simple_layout)
@@ -210,7 +210,7 @@ class GPUDiscreteField(DiscreteField):
def setInitializationKernel(self, kernel):
"""
Set the initialization kernel
- @param kernel : KernelLauncher to use for initialize field.
+ @param kernel : OpenClKernelLauncher to use for initialize field.
"""
self.init_kernel = kernel
@@ -247,7 +247,7 @@ class GPUDiscreteField(DiscreteField):
t = self.precision(time)
if __VERBOSE__:
print "{" + str(self._rank) + "}", "Initialize", self.name
- isGPUKernel = isinstance(formula, KernelLauncher) \
+ isGPUKernel = isinstance(formula, OpenClKernelLauncher) \
or isinstance(formula, KernelListLauncher)
if not isGPUKernel and self.init_kernel is None:
DiscreteField.initialize(self, formula, False, time, *args)
diff --git a/hysop/old/gpu.old/gpu_kernel.py b/hysop/old/gpu.old/gpu_kernel.py
index 53526a467..bcbd43030 100644
--- a/hysop/old/gpu.old/gpu_kernel.py
+++ b/hysop/old/gpu.old/gpu_kernel.py
@@ -123,7 +123,7 @@ class KernelListLauncher(object):
for d in xrange(len(self.kernel))]
-class KernelLauncher(KernelListLauncher):
+class OpenClKernelLauncher(KernelListLauncher):
"""
OpenCL kernel launcher.
@@ -133,7 +133,7 @@ class KernelLauncher(KernelListLauncher):
@debug
def __init__(self, kernel, queue, gsize=None, lsize=None):
"""
- Create a KernelLauncher.
+ Create a OpenClKernelLauncher.
Create a KernelListLauncher with a list of one kernel.
diff --git a/hysop/old/gpu.old/gpu_multiphase_baroclinic_rhs.py b/hysop/old/gpu.old/gpu_multiphase_baroclinic_rhs.py
index 0b2142ce7..84e970eb8 100644
--- a/hysop/old/gpu.old/gpu_multiphase_baroclinic_rhs.py
+++ b/hysop/old/gpu.old/gpu_multiphase_baroclinic_rhs.py
@@ -10,8 +10,8 @@ from hysop.operator.discrete.discrete import DiscreteOperator
from hysop.operator.discrete.discrete import get_extra_args_from_method
from hysop.backend.opencl import cl
from hysop.backend.opencl.gpu_operator import GPUOperator
-from hysop.backend.opencl.gpu_kernel import KernelListLauncher
-from hysop.backend.opencl.gpu_discrete import GPUDiscreteField
+from hysop.backend.opencl.opencl_kernel import KernelListLauncher
+from hysop.backend.opencl.opencl_discrete import OpenClDiscreteField
from hysop.tools.profiler import FProfiler
from hysop.core.mpi import Wtime
from hysop.methods import SpaceDiscretisation
@@ -74,8 +74,8 @@ class BaroclinicRHS(DiscreteOperator, GPUOperator):
# GPU allocation.
for field in self.variables:
- alloc = not isinstance(field, GPUDiscreteField)
- GPUDiscreteField.fromField(self.cl_env, field,
+ alloc = not isinstance(field, OpenClDiscreteField)
+ OpenClDiscreteField.fromField(self.cl_env, field,
self.gpu_precision, layout=False)
if not field.gpu_allocated:
field.allocate()
diff --git a/hysop/old/gpu.old/gpu_multiresolution_filter.py b/hysop/old/gpu.old/gpu_multiresolution_filter.py
index cabfa87ff..9d57dce43 100644
--- a/hysop/old/gpu.old/gpu_multiresolution_filter.py
+++ b/hysop/old/gpu.old/gpu_multiresolution_filter.py
@@ -8,8 +8,8 @@ from hysop.tools.numpywrappers import npw
from hysop.operator.discrete.multiresolution_filter import FilterFineToCoarse
from hysop.backend.opencl.gpu_operator import GPUOperator
from hysop.operator.discrete.discrete import get_extra_args_from_method
-from hysop.backend.opencl.gpu_discrete import GPUDiscreteField
-from hysop.backend.opencl.gpu_kernel import KernelLauncher
+from hysop.backend.opencl.opencl_discrete import OpenClDiscreteField
+from hysop.backend.opencl.opencl_kernel import OpenClKernelLauncher
from hysop.methods import Remesh
@@ -38,16 +38,16 @@ class GPUFilterFineToCoarse(FilterFineToCoarse, GPUOperator):
**kwds)
#GPU allocations
- alloc = not isinstance(self.field_in[0], GPUDiscreteField)
- GPUDiscreteField.fromField(self.cl_env, self.field_in[0],
+ alloc = not isinstance(self.field_in[0], OpenClDiscreteField)
+ OpenClDiscreteField.fromField(self.cl_env, self.field_in[0],
self.gpu_precision, layout=False)
if not self.field_in[0].gpu_allocated:
self.field_in[0].allocate()
if alloc:
self.size_global_alloc += self.field_in[0].mem_size
- alloc = not isinstance(self.field_out[0], GPUDiscreteField)
- GPUDiscreteField.fromField(self.cl_env, self.field_out[0],
+ alloc = not isinstance(self.field_out[0], OpenClDiscreteField)
+ OpenClDiscreteField.fromField(self.cl_env, self.field_out[0],
self.gpu_precision, layout=False)
if not self.field_out[0].gpu_allocated:
self.field_out[0].allocate()
@@ -122,9 +122,9 @@ class GPUFilterFineToCoarse(FilterFineToCoarse, GPUOperator):
build_options += " -D WG=" + str(lwi[0])
build_options += " -D FORMULA=" + self.method[Remesh].__name__.upper()
prg = self.cl_env.build_src(src, build_options, vec)
- self.fine_to_coarse = KernelLauncher(
+ self.fine_to_coarse = OpenClKernelLauncher(
prg.coarse_to_fine_filter, self.cl_env.queue, gwi, lwi)
- self.initialize = KernelLauncher(
+ self.initialize = OpenClKernelLauncher(
prg.initialize_output, self.cl_env.queue,
self.field_out[0].data[0].shape, None)
self._evts = [None, ] * self.field_in[0].dimension
diff --git a/hysop/old/gpu.old/gpu_particle_advection.py b/hysop/old/gpu.old/gpu_particle_advection.py
index 9a132ce71..305e2667b 100644
--- a/hysop/old/gpu.old/gpu_particle_advection.py
+++ b/hysop/old/gpu.old/gpu_particle_advection.py
@@ -9,10 +9,10 @@ from hysop.numerics.odesolvers import Euler
from hysop.operator.discrete.particle_advection import ParticleAdvection
from hysop.operator.discrete.discrete import get_extra_args_from_method
from hysop.backend.opencl import cl
-from hysop.backend.opencl.gpu_kernel import KernelLauncher
+from hysop.backend.opencl.opencl_kernel import OpenClKernelLauncher
import hysop.default_methods as default
from hysop.tools.numpywrappers import npw
-from hysop.backend.opencl.gpu_discrete import GPUDiscreteField
+from hysop.backend.opencl.opencl_discrete import OpenClDiscreteField
from hysop.backend.opencl.gpu_operator import GPUOperator
from hysop.tools.profiler import profile
from hysop.numerics.update_ghosts import UpdateGhostsFull
@@ -284,15 +284,15 @@ class GPUParticleAdvection(ParticleAdvection, GPUOperator):
"""Allocate OpenCL buffers for velocity and advected field.
"""
# Velocity.
- alloc = not isinstance(self.velocity, GPUDiscreteField)
- GPUDiscreteField.fromField(self.cl_env, self.velocity,
+ alloc = not isinstance(self.velocity, OpenClDiscreteField)
+ OpenClDiscreteField.fromField(self.cl_env, self.velocity,
self.gpu_precision, simple_layout=False)
if alloc:
self.size_global_alloc += self.velocity.mem_size
# Transported field.
- alloc = not isinstance(self.fields_on_grid[0], GPUDiscreteField)
- GPUDiscreteField.fromField(self.cl_env,
+ alloc = not isinstance(self.fields_on_grid[0], OpenClDiscreteField)
+ OpenClDiscreteField.fromField(self.cl_env,
self.fields_on_grid[0],
self.gpu_precision,
layout=False)
@@ -338,7 +338,7 @@ class GPUParticleAdvection(ParticleAdvection, GPUOperator):
else:
workItemNumber, gwi, lwi = \
self.cl_env.get_work_items(self.resol_dir)
- gpudf.setInitializationKernel(KernelLauncher(
+ gpudf.setInitializationKernel(OpenClKernelLauncher(
cl.Kernel(self.prg, k_name), self.cl_env.queue,
gwi, lwi))
@@ -360,9 +360,9 @@ class GPUParticleAdvection(ParticleAdvection, GPUOperator):
# src,
# build_options,
# vec)
- # self.copy = KernelLauncher(prg.copy,
+ # self.copy = OpenClKernelLauncher(prg.copy,
# self.cl_env.queue, gwi, lwi)
- return KernelLauncher(cl.enqueue_copy, self.cl_env.queue)
+ return OpenClKernelLauncher(cl.enqueue_copy, self.cl_env.queue)
def _collect_kernels_cl_src_transpositions_xy(self):
"""Compile OpenCL sources for transpositions kernel.
@@ -418,7 +418,7 @@ class GPUParticleAdvection(ParticleAdvection, GPUOperator):
build_options += " -D NB_GROUPS_II={0}".format(blocs_nb[1])
build_options += ocl_cte
prg = self.cl_env.build_src(src, build_options, vec)
- return KernelLauncher(prg.transpose_xy, self.cl_env.queue, gwi, lwi)
+ return OpenClKernelLauncher(prg.transpose_xy, self.cl_env.queue, gwi, lwi)
def _collect_kernels_cl_src_transpositions_xz(self):
resol = self.fields_topo.mesh.resolution
@@ -477,7 +477,7 @@ class GPUParticleAdvection(ParticleAdvection, GPUOperator):
src,
build_options,
vec)
- return KernelLauncher(prg.transpose_xz, self.cl_env.queue, gwi, lwi)
+ return OpenClKernelLauncher(prg.transpose_xz, self.cl_env.queue, gwi, lwi)
def _collect_usr_cl_src(self):
"""Build user sources.
@@ -523,7 +523,7 @@ class GPUParticleAdvection(ParticleAdvection, GPUOperator):
src, build_options, vec,
nb_remesh_components=self.fields_on_grid[0].nb_components)
- self.num_advec_and_remesh = KernelLauncher(
+ self.num_advec_and_remesh = OpenClKernelLauncher(
prg.advection_and_remeshing, self.cl_env.queue, gwi, lwi)
def _collect_kernels_cl_src_2k(self):
@@ -565,7 +565,7 @@ class GPUParticleAdvection(ParticleAdvection, GPUOperator):
vec,
nb_remesh_components=self.fields_on_grid[0].nb_components)
- self.num_advec = KernelLauncher(
+ self.num_advec = OpenClKernelLauncher(
prg.advection_kernel, self.cl_env.queue, gwi, lwi)
# remeshing
@@ -584,7 +584,7 @@ class GPUParticleAdvection(ParticleAdvection, GPUOperator):
prg = self.cl_env.build_src(
src, build_options, vec,
nb_remesh_components=self.fields_on_grid[0].nb_components)
- self.num_remesh = KernelLauncher(
+ self.num_remesh = OpenClKernelLauncher(
prg.remeshing_kernel, self.cl_env.queue, gwi, lwi)
@debug
diff --git a/hysop/old/gpu.old/gpu_particle_advection_dir.py b/hysop/old/gpu.old/gpu_particle_advection_dir.py
index e3ab7ee88..abbb887a0 100644
--- a/hysop/old/gpu.old/gpu_particle_advection_dir.py
+++ b/hysop/old/gpu.old/gpu_particle_advection_dir.py
@@ -9,9 +9,9 @@ from hysop.operator.discrete.particle_advection_dir import ParticleAdvectionDir
from hysop.numerics.update_ghosts import UpdateGhostsFull
from hysop.backend.opencl import cl
-from hysop.backend.opencl.gpu_discrete import GPUDiscreteField
+from hysop.backend.opencl.opencl_discrete import OpenClDiscreteField
from hysop.backend.opencl.gpu_operator import GPUOperator
-from hysop.backend.opencl.gpu_kernel import KernelLauncher
+from hysop.backend.opencl.opencl_kernel import OpenClKernelLauncher
from hysop.backend.opencl.kernel_autotuner import AutotunerConfig
from hysop.backend.codegen.structs.mesh_info import MeshInfoStruct, TranspositionState, MeshDirection
@@ -262,7 +262,7 @@ class GPUParticleAdvectionDir(ParticleAdvectionDir, GPUOperator):
self.v_resol_dir = v_resol_dir
def _collect_copy_kernels(self):
- self.copy = KernelLauncher(cl.enqueue_copy,
+ self.copy = OpenClKernelLauncher(cl.enqueue_copy,
self.cl_env.queue)
def _collect_transposition_kernels(self):
@@ -384,14 +384,14 @@ class GPUParticleAdvectionDir(ParticleAdvectionDir, GPUOperator):
raise RuntimeError('_set_work_arrays has not been implemented properly.')
# Velocity
- GPUDiscreteField.fromField(self.cl_env, self.velocity,
+ OpenClDiscreteField.fromField(self.cl_env, self.velocity,
self.gpu_precision, simple_layout=False)
if self.velocity.allocate():
self.size_global_alloc += self.velocity.mem_size
# Fields on grids
for fg in self.fields_on_grid:
- GPUDiscreteField.fromField(self.cl_env,
+ OpenClDiscreteField.fromField(self.cl_env,
fg,
self.gpu_precision,
layout=False)
@@ -448,7 +448,7 @@ class GPUParticleAdvectionDir(ParticleAdvectionDir, GPUOperator):
init_field_kernel = cl.Kernel(self.prg, k_name)
gpudf.setInitializationKernel(
- KernelLauncher(init_field_kernel,self.cl_env.queue,gwi,lwi)
+ OpenClKernelLauncher(init_field_kernel,self.cl_env.queue,gwi,lwi)
)
diff --git a/hysop/old/gpu.old/gpu_stretching.py b/hysop/old/gpu.old/gpu_stretching.py
index 1ae2c555f..8b7d191be 100644
--- a/hysop/old/gpu.old/gpu_stretching.py
+++ b/hysop/old/gpu.old/gpu_stretching.py
@@ -13,8 +13,8 @@ from hysop.operator.discrete.discrete import DiscreteOperator, get_extra_args_fr
from hysop.core.mpi import Wtime
from hysop.backend.opencl import cl
from hysop.backend.opencl.gpu_operator import GPUOperator
-from hysop.backend.opencl.gpu_kernel import KernelLauncher
-from hysop.backend.opencl.gpu_discrete import GPUDiscreteField
+from hysop.backend.opencl.opencl_kernel import OpenClKernelLauncher
+from hysop.backend.opencl.opencl_discrete import OpenClDiscreteField
from hysop.tools.profiler import FProfiler
from hysop.methods import TimeIntegrator, SpaceDiscretisation, Formulation, Support
@@ -73,7 +73,7 @@ class GPUStretching(DiscreteOperator, GPUOperator):
## GPU allocations
for field in variables:
- GPUDiscreteField.fromField(self.cl_env, field,
+ OpenClDiscreteField.fromField(self.cl_env, field,
self.gpu_precision, simple_layout=False)
if field.allocate():
self.size_global_alloc += field.mem_size
@@ -168,7 +168,7 @@ class GPUStretching(DiscreteOperator, GPUOperator):
self.size_local_alloc += cached_bytes
kernels = {}
- kernels['stretching'] = KernelLauncher(kernel, cl_env.queue, gwi, lwi)
+ kernels['stretching'] = OpenClKernelLauncher(kernel, cl_env.queue, gwi, lwi)
self.kernels = kernels
def _gen_and_build_kernel(self, local_size, global_size, kernel_args,
diff --git a/hysop/old/gpu.old/multi_gpu_particle_advection.py b/hysop/old/gpu.old/multi_gpu_particle_advection.py
index fc8b24553..d3848365f 100644
--- a/hysop/old/gpu.old/multi_gpu_particle_advection.py
+++ b/hysop/old/gpu.old/multi_gpu_particle_advection.py
@@ -11,7 +11,7 @@ from hysop.operator.discrete.discrete import get_extra_args_from_method
from hysop.methods import TimeIntegrator, MultiScale, Remesh
from hysop.numerics.odesolvers import RK2
from hysop.numerics.remeshing import Linear as Linear_rmsh
-from hysop.backend.opencl.gpu_kernel import KernelLauncher
+from hysop.backend.opencl.opencl_kernel import OpenClKernelLauncher
from hysop.tools.profiler import FProfiler
from hysop.backend.opencl import cl, CL_PROFILE
from hysop.core.mpi import Wtime
@@ -375,13 +375,13 @@ class MultiGPUParticleAdvection(GPUParticleAdvection):
build_options += " -D BUFF_WIDTH=" + str(self._s_buff_width)
prg = self.cl_env.build_src(
src, build_options, 1)
- self.num_advec_and_remesh_comm_l = KernelLauncher(
+ self.num_advec_and_remesh_comm_l = OpenClKernelLauncher(
prg.buff_advec_and_remesh_l, self.cl_env.queue,
(gwi[1], gwi[2]), (32, 1))
- self.num_advec_and_remesh_comm_r = KernelLauncher(
+ self.num_advec_and_remesh_comm_r = OpenClKernelLauncher(
prg.buff_advec_and_remesh_r, self.cl_env.queue,
(gwi[1], gwi[2]), (32, 1))
- self.num_advec_and_remesh = KernelLauncher(
+ self.num_advec_and_remesh = OpenClKernelLauncher(
prg.buff_advec_and_remesh, self.cl_env.queue,
gwi, lwi)
@@ -409,7 +409,7 @@ class MultiGPUParticleAdvection(GPUParticleAdvection):
build_options += " -D V_BUFF_WIDTH=" + str(self._v_buff_width)
prg = self.cl_env.build_src(
src, build_options, 1)
- self.num_advec = KernelLauncher(
+ self.num_advec = OpenClKernelLauncher(
prg.buff_advec, self.cl_env.queue,
gwi, lwi)
@@ -429,13 +429,13 @@ class MultiGPUParticleAdvection(GPUParticleAdvection):
build_options += " -D BUFF_WIDTH=" + str(self._s_buff_width)
prg = self.cl_env.build_src(
src, build_options, 1)
- self.num_remesh_comm_l = KernelLauncher(
+ self.num_remesh_comm_l = OpenClKernelLauncher(
prg.buff_remesh_l, self._queue_comm_m,
(gwi[1], gwi[2]), (32, 1))
- self.num_remesh_comm_r = KernelLauncher(
+ self.num_remesh_comm_r = OpenClKernelLauncher(
prg.buff_remesh_r, self._queue_comm_p,
(gwi[1], gwi[2]), (32, 1))
- self.num_remesh = KernelLauncher(
+ self.num_remesh = OpenClKernelLauncher(
prg.remesh, self.cl_env.queue,
gwi, lwi)
diff --git a/hysop/old/gpu.old/static_gpu_particle_advection_dir.py b/hysop/old/gpu.old/static_gpu_particle_advection_dir.py
index 45c78e9a3..636b22dd5 100644
--- a/hysop/old/gpu.old/static_gpu_particle_advection_dir.py
+++ b/hysop/old/gpu.old/static_gpu_particle_advection_dir.py
@@ -2,7 +2,7 @@
from hysop import __VERBOSE__
from hysop.backend.opencl import cl
-from hysop.backend.opencl.gpu_kernel import KernelLauncher
+from hysop.backend.opencl.opencl_kernel import OpenClKernelLauncher
from hysop.backend.opencl.gpu_particle_advection_dir import GPUParticleAdvectionDir, MeshDirection
from hysop.methods import TimeIntegrator, Remesh, ExtraArgs, \
@@ -89,8 +89,8 @@ class StaticGPUParticleAdvectionDir(GPUParticleAdvectionDir):
if order<0:
msg='Stretching order < 0'
raise ValueError(msg)
- if formulation not in StretchingFormulation.entries():
- msg='Stretching formulation is not one of {}.'.format(StretchingFormulation.fields().keys())
+ if not isinstance(formulation,StretchingFormulation):
+ msg='Stretching formulation is not one of {}.'.format(StretchingFormulation.svalues())
raise ValueError(msg)
def _initialize_cl_env(self):
@@ -213,7 +213,7 @@ class StaticGPUParticleAdvectionDir(GPUParticleAdvectionDir):
build_options += defines
prg = self.cl_env.build_src(src, build_options, vec)
- return KernelLauncher(prg.transpose_xy, self.cl_env.queue, gwi, lwi)
+ return OpenClKernelLauncher(prg.transpose_xy, self.cl_env.queue, gwi, lwi)
def _build_transpose_kernel_xz(self, resolution, defines):
build_options = self._build_options + self._size_constants
@@ -237,7 +237,7 @@ class StaticGPUParticleAdvectionDir(GPUParticleAdvectionDir):
build_options += defines
prg = self.cl_env.build_src(src,build_options,vec)
- return KernelLauncher(prg.transpose_xz, self.cl_env.queue, gwi, lwi)
+ return OpenClKernelLauncher(prg.transpose_xz, self.cl_env.queue, gwi, lwi)
# def _collect_advec_kernel(self):
# """
@@ -280,7 +280,7 @@ class StaticGPUParticleAdvectionDir(GPUParticleAdvectionDir):
# nb_remesh_components=self.velocity.nb_components)
# callback_profiler.register_tasks('advection')
- # self._advec = KernelLauncher(
+ # self._advec = OpenClKernelLauncher(
# prg.advection_kernel, self.cl_env.queue, gwi, lwi)
def _collect_remesh_kernel(self):
@@ -311,7 +311,7 @@ class StaticGPUParticleAdvectionDir(GPUParticleAdvectionDir):
prg = self.cl_env.build_src(
src, build_options, vec,
nb_remesh_components=rc)
- self._remesh[rc] = KernelLauncher(
+ self._remesh[rc] = OpenClKernelLauncher(
prg.remeshing_kernel, self.cl_env.queue, gwi, lwi)
cname = 'remesh' if len(self.required_components)==1 \
@@ -359,7 +359,7 @@ class StaticGPUParticleAdvectionDir(GPUParticleAdvectionDir):
prg = self.cl_env.build_src(
src, build_options, vec,
nb_remesh_components=rc)
- self._advec_and_remesh[rc] = KernelLauncher(
+ self._advec_and_remesh[rc] = OpenClKernelLauncher(
prg.advection_and_remeshing, self.cl_env.queue, gwi, lwi)
cname = 'advec_remesh' if len(self.required_components)==1 \
diff --git a/hysop/old/gpu.old/tests/test_copy.py b/hysop/old/gpu.old/tests/test_copy.py
index 54f2381ea..a71bdd8e7 100644
--- a/hysop/old/gpu.old/tests/test_copy.py
+++ b/hysop/old/gpu.old/tests/test_copy.py
@@ -5,7 +5,7 @@ Testing copy kernels.
from hysop.backend.opencl import cl
from hysop.constants import np
from hysop.backend.opencl.tools import get_opencl_environment
-from hysop.backend.opencl.gpu_kernel import KernelLauncher
+from hysop.backend.opencl.opencl_kernel import OpenClKernelLauncher
from hysop.tools.numpywrappers import npw
@@ -22,7 +22,7 @@ def test_copy2D():
int(resolution[1] / 2))
lwi = (8, 8)
prg = cl_env.build_src(src_copy, build_options, vec)
- copy = KernelLauncher(prg.copy, cl_env.queue, gwi, lwi)
+ copy = OpenClKernelLauncher(prg.copy, cl_env.queue, gwi, lwi)
data_in = npw.asrealarray(np.random.random(resolution))
data_out = npw.empty_like(data_in)
@@ -59,7 +59,7 @@ def test_copy2D_rect():
int(resolution[1] / 2))
lwi = (8, 8)
prg = cl_env.build_src(src_copy, build_options, vec)
- copy_x = KernelLauncher(prg.copy, cl_env.queue, gwi, lwi)
+ copy_x = OpenClKernelLauncher(prg.copy, cl_env.queue, gwi, lwi)
build_options = ""
build_options += " -D NB_I=512 -D NB_II=256"
@@ -69,7 +69,7 @@ def test_copy2D_rect():
int(resolution[0] / 2))
lwi = (8, 8)
prg = cl_env.build_src(src_copy, build_options, vec)
- copy_y = KernelLauncher(prg.copy, cl_env.queue, gwi, lwi)
+ copy_y = OpenClKernelLauncher(prg.copy, cl_env.queue, gwi, lwi)
data_in = npw.asrealarray(np.random.random(resolution))
data_out = npw.empty_like(data_in)
@@ -121,7 +121,7 @@ def test_copy3D():
# Build code
prg = cl_env.build_src(src_copy, build_options, vec)
- init_copy = KernelLauncher(prg.copy, cl_env.queue, gwi, lwi)
+ init_copy = OpenClKernelLauncher(prg.copy, cl_env.queue, gwi, lwi)
data_in = npw.asrealarray(np.random.random(resolution))
data_out = npw.empty_like(data_in)
@@ -161,7 +161,7 @@ def test_copy3D_rect():
int(resolution_x[2]))
lwi = (4, 8, 1)
prg = cl_env.build_src(src_copy, build_options, vec)
- init_copy_x = KernelLauncher(prg.copy, cl_env.queue, gwi, lwi)
+ init_copy_x = OpenClKernelLauncher(prg.copy, cl_env.queue, gwi, lwi)
build_options = ""
build_options += " -D NB_I=32 -D NB_II=16 -D NB_III=64"
@@ -172,7 +172,7 @@ def test_copy3D_rect():
int(resolution_x[2]))
lwi = (4, 8, 1)
prg = cl_env.build_src(src_copy, build_options, vec)
- init_copy_y = KernelLauncher(prg.copy, cl_env.queue, gwi, lwi)
+ init_copy_y = OpenClKernelLauncher(prg.copy, cl_env.queue, gwi, lwi)
build_options = ""
build_options += " -D NB_I=64 -D NB_II=16 -D NB_III=32"
@@ -183,7 +183,7 @@ def test_copy3D_rect():
int(resolution_x[1]))
lwi = (4, 8, 1)
prg = cl_env.build_src(src_copy, build_options, vec)
- init_copy_z = KernelLauncher(prg.copy, cl_env.queue, gwi, lwi)
+ init_copy_z = OpenClKernelLauncher(prg.copy, cl_env.queue, gwi, lwi)
data_in = npw.asrealarray(np.random.random(resolution_x))
data_out = np.empty_like(data_in)
diff --git a/hysop/old/gpu.old/tests/test_opencl_environment.py b/hysop/old/gpu.old/tests/test_opencl_environment.py
index 9e7732fc1..70b893758 100644
--- a/hysop/old/gpu.old/tests/test_opencl_environment.py
+++ b/hysop/old/gpu.old/tests/test_opencl_environment.py
@@ -1,6 +1,6 @@
"""Test hysop implementation of OpenCL basic functionnalities"""
import numpy as np
-from hysop.backend.opencl.tools import get_opencl_environment, explore, OpenCLEnvironment
+from hysop.backend.opencl.tools import get_opencl_environment, explore, OpenClEnvironment
from hysop.constants import HYSOP_REAL
FLOAT_GPU = np.float32
from hysop.core.mpi import main_comm
@@ -14,7 +14,7 @@ def test_opencl_env_default():
explore()
# Create default opencl env.
cl_env = get_opencl_environment()
- assert isinstance(cl_env, OpenCLEnvironment)
+ assert isinstance(cl_env, OpenClEnvironment)
assert cl_env.device is not None
assert cl_env.ctx is not None
assert cl_env.queue is not None
@@ -35,7 +35,7 @@ def test_opencl_env():
cl_env = get_opencl_environment(platform_id=nb_platforms - 1,
device_id=nb_devices - 1,
precision=FLOAT_GPU, comm=comm)
- assert isinstance(cl_env, OpenCLEnvironment)
+ assert isinstance(cl_env, OpenClEnvironment)
assert cl_env.platform == plt
assert cl_env.device == device
assert cl_env.ctx is not None
diff --git a/hysop/old/gpu.old/tests/test_transposition.py b/hysop/old/gpu.old/tests/test_transposition.py
index f944d3c5e..11c14821c 100644
--- a/hysop/old/gpu.old/tests/test_transposition.py
+++ b/hysop/old/gpu.old/tests/test_transposition.py
@@ -5,7 +5,7 @@ Testing copy kernels.
from hysop.backend.opencl import cl
from hysop.constants import np
from hysop.backend.opencl.tools import get_opencl_environment
-from hysop.backend.opencl.gpu_kernel import KernelLauncher
+from hysop.backend.opencl.opencl_kernel import OpenClKernelLauncher
from hysop.tools.numpywrappers import npw
@@ -65,7 +65,7 @@ def test_transposition_xy2D():
# Build code
prg = cl_env.build_src(src_transpose_xy, build_options, vec)
- init_transpose_xy = KernelLauncher(
+ init_transpose_xy = OpenClKernelLauncher(
prg.transpose_xy, cl_env.queue, gwi, lwi)
_comparison(resolution, resolution,
init_transpose_xy, init_transpose_xy,
@@ -87,7 +87,7 @@ def test_transposition_xy2D_noVec():
# Build code
prg = cl_env.build_src(src_transpose_xy, build_options)
- init_transpose_xy = KernelLauncher(
+ init_transpose_xy = OpenClKernelLauncher(
prg.transpose_xy, cl_env.queue, gwi, lwi)
_comparison(resolution, resolution,
init_transpose_xy, init_transpose_xy,
@@ -111,7 +111,7 @@ def test_transposition_xy2D_rect():
build_options += " -D NB_GROUPS_I=" + str((resolution[0] / 4) / lwi[0])
build_options += " -D NB_GROUPS_II=" + str((resolution[1] / 4) / lwi[1])
prg = cl_env.build_src(src_transpose_xy, build_options, vec)
- init_transpose_xy_x = KernelLauncher(prg.transpose_xy,
+ init_transpose_xy_x = OpenClKernelLauncher(prg.transpose_xy,
cl_env.queue,
gwi, lwi)
@@ -124,7 +124,7 @@ def test_transposition_xy2D_rect():
build_options += " -D NB_GROUPS_I=" + str((resolution[1] / 4) / lwi[0])
build_options += " -D NB_GROUPS_II=" + str((resolution[0] / 4) / lwi[1])
prg = cl_env.build_src(src_transpose_xy, build_options, vec)
- init_transpose_xy_y = KernelLauncher(prg.transpose_xy,
+ init_transpose_xy_y = OpenClKernelLauncher(prg.transpose_xy,
cl_env.queue,
gwi, lwi)
_comparison(resolution, resolutionT,
@@ -149,7 +149,7 @@ def test_transposition_xy2D_noVec_rect():
build_options += " -D NB_GROUPS_I=" + str((resolution[0]) / lwi[0])
build_options += " -D NB_GROUPS_II=" + str((resolution[1] / 4) / lwi[1])
prg = cl_env.build_src(src_transpose_xy, build_options, vec)
- init_transpose_xy_x = KernelLauncher(prg.transpose_xy,
+ init_transpose_xy_x = OpenClKernelLauncher(prg.transpose_xy,
cl_env.queue,
gwi, lwi)
@@ -162,7 +162,7 @@ def test_transposition_xy2D_noVec_rect():
build_options += " -D NB_GROUPS_I=" + str((resolution[1]) / lwi[0])
build_options += " -D NB_GROUPS_II=" + str((resolution[0] / 4) / lwi[1])
prg = cl_env.build_src(src_transpose_xy, build_options, vec)
- init_transpose_xy_y = KernelLauncher(prg.transpose_xy,
+ init_transpose_xy_y = OpenClKernelLauncher(prg.transpose_xy,
cl_env.queue,
gwi, lwi)
_comparison(resolution, resolutionT,
@@ -185,7 +185,7 @@ def test_transposition_xy3D():
build_options += " -D NB_GROUPS_I=" + str((resolution[0] / 2) / lwi[0])
build_options += " -D NB_GROUPS_II=" + str((resolution[1] / 2) / lwi[1])
prg = cl_env.build_src(src_transpose_xy, build_options, vec)
- init_transpose_xy = KernelLauncher(
+ init_transpose_xy = OpenClKernelLauncher(
prg.transpose_xy, cl_env.queue, gwi, lwi)
_comparison(resolution, resolution,
init_transpose_xy, init_transpose_xy,
@@ -207,7 +207,7 @@ def test_transposition_xy3D_noVec():
build_options += " -D NB_GROUPS_I=" + str((resolution[0]) / lwi[0])
build_options += " -D NB_GROUPS_II=" + str((resolution[1] / 2) / lwi[1])
prg = cl_env.build_src(src_transpose_xy, build_options, vec)
- init_transpose_xy = KernelLauncher(
+ init_transpose_xy = OpenClKernelLauncher(
prg.transpose_xy, cl_env.queue, gwi, lwi)
_comparison(resolution, resolution,
init_transpose_xy, init_transpose_xy,
@@ -232,7 +232,7 @@ def test_transposition_xy3D_rect():
build_options += " -D NB_GROUPS_I=" + str((resolution[0] / 2) / lwi[0])
build_options += " -D NB_GROUPS_II=" + str((resolution[1] / 2) / lwi[1])
prg = cl_env.build_src(src_transpose_xy, build_options, vec)
- init_transpose_xy_x = KernelLauncher(
+ init_transpose_xy_x = OpenClKernelLauncher(
prg.transpose_xy, cl_env.queue, gwi, lwi)
build_options = ""
@@ -245,7 +245,7 @@ def test_transposition_xy3D_rect():
build_options += " -D NB_GROUPS_I=" + str((resolution[1] / 2) / lwi[0])
build_options += " -D NB_GROUPS_II=" + str((resolution[0] / 2) / lwi[1])
prg = cl_env.build_src(src_transpose_xy, build_options, vec)
- init_transpose_xy_y = KernelLauncher(
+ init_transpose_xy_y = OpenClKernelLauncher(
prg.transpose_xy, cl_env.queue, gwi, lwi)
_comparison(resolution, resolutionT,
init_transpose_xy_x, init_transpose_xy_y,
@@ -270,7 +270,7 @@ def test_transposition_xy3D_noVec_rect():
build_options += " -D NB_GROUPS_I=" + str((resolution[0]) / lwi[0])
build_options += " -D NB_GROUPS_II=" + str((resolution[1] / 2) / lwi[1])
prg = cl_env.build_src(src_transpose_xy, build_options, vec)
- init_transpose_xy_x = KernelLauncher(
+ init_transpose_xy_x = OpenClKernelLauncher(
prg.transpose_xy, cl_env.queue, gwi, lwi)
build_options = ""
@@ -283,7 +283,7 @@ def test_transposition_xy3D_noVec_rect():
build_options += " -D NB_GROUPS_I=" + str((resolution[1]) / lwi[0])
build_options += " -D NB_GROUPS_II=" + str((resolution[0] / 2) / lwi[1])
prg = cl_env.build_src(src_transpose_xy, build_options, vec)
- init_transpose_xy_y = KernelLauncher(
+ init_transpose_xy_y = OpenClKernelLauncher(
prg.transpose_xy, cl_env.queue, gwi, lwi)
_comparison(resolution, resolutionT,
init_transpose_xy_x, init_transpose_xy_y,
@@ -307,7 +307,7 @@ def test_transposition_xz3D():
build_options += " -D NB_GROUPS_II=" + str((resolution[1] / 4) / lwi[1])
build_options += " -D NB_GROUPS_III=" + str((resolution[2] / 4) / lwi[2])
prg = cl_env.build_src(src_transpose_xz, build_options, vec)
- init_transpose_xz = KernelLauncher(
+ init_transpose_xz = OpenClKernelLauncher(
prg.transpose_xz, cl_env.queue, gwi, lwi)
_comparison(resolution, resolution,
init_transpose_xz, init_transpose_xz,
@@ -331,7 +331,7 @@ def test_transposition_xz3D_noVec():
build_options += " -D NB_GROUPS_II=" + str((resolution[1] / 4) / lwi[1])
build_options += " -D NB_GROUPS_III=" + str((resolution[2] / 4) / lwi[2])
prg = cl_env.build_src(src_transpose_xz, build_options, vec)
- init_transpose_xz = KernelLauncher(
+ init_transpose_xz = OpenClKernelLauncher(
prg.transpose_xz, cl_env.queue, gwi, lwi)
_comparison(resolution, resolution,
init_transpose_xz, init_transpose_xz,
@@ -358,7 +358,7 @@ def test_transposition_xz3D_rect():
build_options += " -D NB_GROUPS_II=" + str((resolution[1] / 4) / lwi[1])
build_options += " -D NB_GROUPS_III=" + str((resolution[2] / 4) / lwi[2])
prg = cl_env.build_src(src_transpose_xz, build_options, vec)
- init_transpose_xz_x = KernelLauncher(
+ init_transpose_xz_x = OpenClKernelLauncher(
prg.transpose_xz, cl_env.queue, gwi, lwi)
build_options = ""
@@ -373,7 +373,7 @@ def test_transposition_xz3D_rect():
build_options += " -D NB_GROUPS_II=" + str((resolution[1] / 4) / lwi[1])
build_options += " -D NB_GROUPS_III=" + str((resolution[0] / 4) / lwi[2])
prg = cl_env.build_src(src_transpose_xz, build_options, vec)
- init_transpose_xz_z = KernelLauncher(
+ init_transpose_xz_z = OpenClKernelLauncher(
prg.transpose_xz, cl_env.queue, gwi, lwi)
_comparison(resolution, resolutionT,
init_transpose_xz_x, init_transpose_xz_z,
@@ -400,7 +400,7 @@ def test_transposition_xz3D_noVec_rect():
build_options += " -D NB_GROUPS_II=" + str((resolution[1] / 4) / lwi[1])
build_options += " -D NB_GROUPS_III=" + str((resolution[2] / 4) / lwi[2])
prg = cl_env.build_src(src_transpose_xz, build_options, vec)
- init_transpose_xz_x = KernelLauncher(
+ init_transpose_xz_x = OpenClKernelLauncher(
prg.transpose_xz, cl_env.queue, gwi, lwi)
build_options = ""
@@ -415,7 +415,7 @@ def test_transposition_xz3D_noVec_rect():
build_options += " -D NB_GROUPS_II=" + str((resolution[1] / 4) / lwi[1])
build_options += " -D NB_GROUPS_III=" + str((resolution[0] / 4) / lwi[2])
prg = cl_env.build_src(src_transpose_xz, build_options, vec)
- init_transpose_xz_z = KernelLauncher(
+ init_transpose_xz_z = OpenClKernelLauncher(
prg.transpose_xz, cl_env.queue, gwi, lwi)
_comparison(resolution, resolutionT,
init_transpose_xz_x, init_transpose_xz_z,
@@ -439,7 +439,7 @@ def test_transposition_xz3Dslice():
build_options += " -D NB_GROUPS_III=" + str((resolution[2] / 4) / lwi[2])
prg = cl_env.build_src(src_transpose_xz, build_options, vec)
- init_transpose_xz = KernelLauncher(
+ init_transpose_xz = OpenClKernelLauncher(
prg.transpose_xz, cl_env.queue, gwi, lwi)
_comparison(resolution, resolution,
init_transpose_xz, init_transpose_xz,
@@ -462,7 +462,7 @@ def test_transposition_xz3Dslice_noVec():
build_options += " -D NB_GROUPS_III=" + str((resolution[2] / 4) / lwi[2])
prg = cl_env.build_src(src_transpose_xz, build_options, vec)
- init_transpose_xz = KernelLauncher(
+ init_transpose_xz = OpenClKernelLauncher(
prg.transpose_xz, cl_env.queue, gwi, lwi)
_comparison(resolution, resolution,
init_transpose_xz, init_transpose_xz,
@@ -488,7 +488,7 @@ def test_transposition_xz3Dslice_rect():
build_options += " -D NB_GROUPS_I=" + str((resolution[0] / 2) / lwi[0])
build_options += " -D NB_GROUPS_III=" + str((resolution[2] / 4) / lwi[2])
prg = cl_env.build_src(src_transpose_xz, build_options, vec)
- init_transpose_xz_x = KernelLauncher(
+ init_transpose_xz_x = OpenClKernelLauncher(
prg.transpose_xz, cl_env.queue, gwi, lwi)
build_options = ""
@@ -502,7 +502,7 @@ def test_transposition_xz3Dslice_rect():
build_options += " -D NB_GROUPS_I=" + str((resolution[2] / 2) / lwi[0])
build_options += " -D NB_GROUPS_III=" + str((resolution[0] / 4) / lwi[2])
prg = cl_env.build_src(src_transpose_xz, build_options, vec)
- init_transpose_xz_z = KernelLauncher(
+ init_transpose_xz_z = OpenClKernelLauncher(
prg.transpose_xz, cl_env.queue, gwi, lwi)
_comparison(resolution, resolutionT,
init_transpose_xz_x, init_transpose_xz_z,
@@ -527,7 +527,7 @@ def test_transposition_xz3Dslice_noVec_rect():
build_options += " -D NB_GROUPS_I=" + str(resolution[0] / lwi[0])
build_options += " -D NB_GROUPS_III=" + str((resolution[2] / 4) / lwi[2])
prg = cl_env.build_src(src_transpose_xz, build_options, vec)
- init_transpose_xz_x = KernelLauncher(
+ init_transpose_xz_x = OpenClKernelLauncher(
prg.transpose_xz, cl_env.queue, gwi, lwi)
build_options = ""
@@ -541,7 +541,7 @@ def test_transposition_xz3Dslice_noVec_rect():
build_options += " -D NB_GROUPS_I=" + str(resolution[2] / lwi[0])
build_options += " -D NB_GROUPS_III=" + str((resolution[0] / 4) / lwi[2])
prg = cl_env.build_src(src_transpose_xz, build_options, vec)
- init_transpose_xz_z = KernelLauncher(
+ init_transpose_xz_z = OpenClKernelLauncher(
prg.transpose_xz, cl_env.queue, gwi, lwi)
_comparison(resolution, resolutionT,
init_transpose_xz_x, init_transpose_xz_z,
diff --git a/hysop/old/gpu.old/tools.py b/hysop/old/gpu.old/tools.py
index eac2194b2..014681b31 100644
--- a/hysop/old/gpu.old/tools.py
+++ b/hysop/old/gpu.old/tools.py
@@ -1,7 +1,7 @@
"""Classes and tools used to handle openCL interface.
-* :class:`~hysop.gpu.tools.OpenCLEnvironment`:
+* :class:`~hysop.gpu.tools.OpenClEnvironment`:
object handling opencl platform, device ... info.
* :func:`~hysop.gpu.tools.get_opengl_shared_environment`:
build or get an OpenCL environment with openGL properties.
@@ -43,7 +43,7 @@ class KernelError(Exception):
return self.err + ': ' + self.msg
-class OpenCLEnvironment(object):
+class OpenClEnvironment(object):
"""OpenCL environment informations and useful functions.
"""
@@ -103,7 +103,7 @@ class OpenCLEnvironment(object):
key=main_comm.Get_rank())
# Floating point codegeneration mode
- from hysop.backend.codegen.base.types import OpenClTypeGen
+ from hysop.backend.opencl.opencl_types import OpenClTypeGen
_kargs = {'device':self.device, 'context':self.ctx, 'platform':self.platform }
if __KERNEL_DEBUG__:
_kargs['float_dump_mode'] = 'dec'
@@ -196,7 +196,7 @@ class OpenCLEnvironment(object):
## update opencl typegen
# Floating point codegeneration mode
- from hysop.backend.codegen.base.types import OpenClTypeGen
+ from hysop.backend.opencl.opencl_types import OpenClTypeGen
_kargs = {'device':self.device, 'context':self.ctx, 'platform':self.platform }
if __KERNEL_DEBUG__:
_kargs['float_dump_mode'] = 'dec'
@@ -910,7 +910,7 @@ def get_opengl_shared_environment(platform_id=None,
Returns
-------
- :class:`~hysop.gpu.tools.OpenCLEnvironment`
+ :class:`~hysop.gpu.tools.OpenClEnvironment`
object handling OpenCL platform, device, context and queue
The context is obtained with gl-shared properties depending on the OS.
@@ -921,7 +921,7 @@ def get_opengl_shared_environment(platform_id=None,
device_id = __DEFAULT_DEVICE_ID__
global __cl_env
if __cl_env is None:
- __cl_env = OpenCLEnvironment(platform_id, device_id, device_type,
+ __cl_env = OpenClEnvironment(platform_id, device_id, device_type,
precision, gl_sharing=True, comm=comm)
else:
__cl_env.modify(platform_id, device_id, device_type,
@@ -953,7 +953,7 @@ def get_opencl_environment(platform_id=None,
Returns
-------
- :class:`~hysop.gpu.tools.OpenCLEnvironment`
+ :class:`~hysop.gpu.tools.OpenClEnvironment`
object handling OpenCL platform, device, context and queue
"""
@@ -963,7 +963,7 @@ def get_opencl_environment(platform_id=None,
device_id = __DEFAULT_DEVICE_ID__
global __cl_env
if __cl_env is None:
- __cl_env = OpenCLEnvironment(platform_id, device_id, device_type,
+ __cl_env = OpenClEnvironment(platform_id, device_id, device_type,
precision, comm=comm)
else:
__cl_env.modify(platform_id, device_id, device_type,
diff --git a/hysop/old/gpu.old/visu/marchingcube.py b/hysop/old/gpu.old/visu/marchingcube.py
index 4434c774f..691292979 100644
--- a/hysop/old/gpu.old/visu/marchingcube.py
+++ b/hysop/old/gpu.old/visu/marchingcube.py
@@ -82,7 +82,7 @@ class Marching_Cube(object):
self._cl_env.macros['**HP_SIZE**'] = self._size_
self.prg = self._cl_env.build_src(self.usr_src, options)
kernel_name = 'constructHPLevel' + self.field.name.split('_D')[0]
- self.numMethod = KernelLauncher(eval('self.prg.' + kernel_name),
+ self.numMethod = OpenClKernelLauncher(eval('self.prg.' + kernel_name),
self.queue,
self.gwi,
self.lwi)
diff --git a/hysop/tools/sympy_utils.py b/hysop/tools/sympy_utils.py
index 5ea732a0c..f005e9a52 100644
--- a/hysop/tools/sympy_utils.py
+++ b/hysop/tools/sympy_utils.py
@@ -1,5 +1,5 @@
-from hysop.deps import np, sm
+from hysop.deps import np, sm, copy
# unicode subscripts for decimal numbers, signs and parenthesis
dec = ['\u208{}'.format(i).decode('unicode-escape') for i in xrange(10)]
@@ -179,3 +179,14 @@ def recurse_expression_tree(op, expr):
if isinstance(expr, sm.Expr):
for arg in expr.args:
recurse_expression_tree(op, arg)
+
+def expr2str(expr, svars, dumper=str):
+ svars = dict(zip(svars.keys(), [dumper(v) for v in svars.values()]))
+ expr = copy.deepcopy(expr)
+ def op(expr):
+ if isinstance(expr,sm.Symbol) and (expr in svars):
+ expr.name = svars[expr]
+ print expr.__class__
+ recurse_expression_tree(op,expr)
+ expr = remove_pows(expr)
+ return str(expr)
--
GitLab