Changeset cde11f0f in sasmodels

Timestamp:

Dec 23, 2015 10:07:37 AM (9 years ago)

Author:

Paul Kienzle <pkienzle@…>

Branches:

master, core_shell_microgels, costrafo411, magnetic_model, release_v0.94, release_v0.95, ticket-1257-vesicle-product, ticket_1156, ticket_1265_superball, ticket_822_more_unit_tests

Children:

07e72e6

Parents:

e3a9733

Message:

make 'fast' a numeric type option as f32 with loose accuracy

Location:

sasmodels

Files:

• core.py (modified) (4 diffs)
• kernelcl.py (modified) (3 diffs)

sasmodels/core.py

@@ -65 +65 @@
     return True
 
-def load_model(model_definition, dtype="single", platform="ocl", fast=False):
+def load_model(model_definition, dtype="single", platform="ocl"):
     """
     Prepare the model for the default execution platform.
@@ -83 +83 @@
     *platform* should be "dll" to force the dll to be used for C models,
     otherwise it uses the default "ocl".
-
-    *fast* is True if fast inaccurate math is acceptable (40% speed increase).
     """
     if isstr(model_definition):
@@ -101 +99 @@
     # source = open(info['name']+'.cl','r').read()
 
-    dtype = np.dtype(dtype)
     if (platform=="dll"
             or not HAVE_OPENCL
@@ -107 +104 @@
         return kerneldll.load_dll(source, info, dtype)
     else:
-        return kernelcl.GpuModel(source, info, dtype, fast)
+        return kernelcl.GpuModel(source, info, dtype)
 
 def make_kernel(model, q_vectors):

sasmodels/kernelcl.py

@@ -183 +183 @@
 
     def has_type(self, dtype):
-        dtype = np.dtype(dtype)
+        dtype = generate.F32 if dtype == 'fast' else np.dtype(dtype)
         return all(has_type(d, dtype) for d in self.context.devices)
 
@@ -195 +195 @@
 
     def compile_program(self, name, source, dtype, fast=False):
-        if name not in self.compiled:
+        key = "%s-%s-%s"%(name, dtype, fast)
+        if key not in self.compiled:
             #print("compiling",name)
-            self.compiled[name] = compile_model(self.context, source, dtype,
-                                                fast)
-        return self.compiled[name]
+            dtype = np.dtype(dtype)
+            program = compile_model(self.context, source, dtype, fast)
+            self.compiled[key] = program
+        return self.compiled[key]
 
     def release_program(self, name):
@@ -232 +234 @@
     for single and 'd', 'float64' or 'double' for double.  Double precision
     is an optional extension which may not be available on all devices.
-
-    *fast* is True if fast inaccurate math is acceptable (40% speed increase)
-    """
-    def __init__(self, source, info, dtype=generate.F32, fast=False):
+    Half precision ('float16','half') may be available on some devices.
+    Fast precision ('fast') is a loose version of single precision, indicating
+    that the compiler is allowed to take shortcuts.
+    """
+    def __init__(self, source, info, dtype=generate.F32):
         self.info = info
         self.source = source
-        self.dtype = np.dtype(dtype)
-        self.fast = fast
+        self.dtype = generate.F32 if dtype=='fast' else np.dtype(dtype)
+        self.fast = (dtype == 'fast')
         self.program = None # delay program creation