Changes in / [a7a5ff3:c98c772] in sasmodels

Files:

• sascomp (modified) (2 diffs)
• sasmodels/kernel_template.c (modified) (3 diffs)
• sasmodels/kernelcl.py (modified) (2 diffs)
• sasmodels/kerneldll.py (modified) (3 diffs)

sascomp

@@ -4 +4 @@
 import os
 import glob
+import logging
+logging.basicConfig(level=logging.INFO)
 
 def main():
@@ -11 +13 @@
     sys.path.insert(0, os.path.join(root, 'bumps'))
     sys.path.insert(0, os.path.join(root, 'periodictable'))
+    sys.path.insert(0, os.path.join(root, 'tinycc'))
     if sasview:  # glob returns a list
         sys.path.insert(0, sasview[0])

sasmodels/kernel_template.c

@@ -12 +12 @@
 // Note: if using a C++ compiler, then define kernel as extern "C"
 #ifndef USE_OPENCL
+// Use SAS_DOUBLE to force the use of double even for float kernels
+#  define SAS_DOUBLE dou ## ble
 #  ifdef __cplusplus
       #include <cstdio>
@@ -38 +40 @@
      #if defined(__TINYC__)
          #include <math.h>
-         inline double trunc(double x) { return x>=0?floor(x):-floor(-x); }
-         inline double fmin(double x, double y) { return x>y ? y : x; }
-         inline double fmax(double x, double y) { return x<y ? y : x; }
          // TODO: test isnan
          inline double _isnan(double x) { return x != x; } // hope this doesn't optimize away!
          #undef isnan
          #define isnan(x) _isnan(x)
+         // Defeat the double->float conversion since we don't have tgmath
+         inline SAS_DOUBLE trunc(SAS_DOUBLE x) { return x>=0?floor(x):-floor(-x); }
+         inline SAS_DOUBLE fmin(SAS_DOUBLE x, SAS_DOUBLE y) { return x>y ? y : x; }
+         inline SAS_DOUBLE fmax(SAS_DOUBLE x, SAS_DOUBLE y) { return x<y ? y : x; }
          #define NEED_EXPM1
          #define NEED_TGAMMA
@@ -70 +73 @@
 
 #if defined(NEED_EXPM1)
-   static double expm1(double x) {
+   static SAS_DOUBLE expm1(SAS_DOUBLE x_in) {
+      double x = (double)x_in;  // go back to float for single precision kernels
       // Adapted from the cephes math library.
       // Copyright 1984 - 1992 by Stephen L. Moshier

sasmodels/kernelcl.py

@@ -61 +61 @@
     warnings.warn(str(exc))
     raise RuntimeError("OpenCL not available")
+
+# CRUFT: pyopencl < 2017.1  (as of June 2016 needs quotes around include path)
+def _quote_path(v):
+    """
+    Quote the path if it is not already quoted.
+
+    If v starts with '-', then assume that it is a -I option or similar
+    and do not quote it.  This is fragile:  -Ipath with space needs to
+    be quoted.
+    """
+    return '"'+v+'"' if v and ' ' in v and not v[0] in "\"'-" else v
+
+if hasattr(cl, '_DEFAULT_INCLUDE_OPTIONS'):
+    cl._DEFAULT_INCLUDE_OPTIONS = [_quote_path(v) for v in cl._DEFAULT_INCLUDE_OPTIONS]
 
 from pyopencl import mem_flags as mf
@@ -224 +238 @@
             context = self.get_context(dtype)
             logging.info("building %s for OpenCL %s"
-                         % (key, context.devices[0].name))
+                         % (key, context.devices[0].name.strip()))
             program = compile_model(context, source, np.dtype(dtype), fast)
             self.compiled[key] = program

sasmodels/kerneldll.py

@@ -60 +60 @@
 from .exception import annotate_exception
 
-# Compiler platform details
-if sys.platform == 'darwin':
-    #COMPILE = "gcc-mp-4.7 -shared -fPIC -std=c99 -fopenmp -O2 -Wall %s -o %s -lm -lgomp"
-    COMPILE = "gcc -shared -fPIC -std=c99 -O2 -Wall %(source)s -o %(output)s -lm"
-elif os.name == 'nt':
+if os.name == 'nt':
+    # Windows compiler; check if TinyCC is available
+    try:
+        from tinycc import TCC
+    except ImportError:
+        TCC = None
     # call vcvarsall.bat before compiling to set path, headers, libs, etc.
     if "VCINSTALLDIR" in os.environ:
@@ -73 +74 @@
         # TODO: remove intermediate OBJ file created in the directory
         # TODO: maybe don't use randomized name for the c file
+        # TODO: maybe ask distutils to find MSVC
         CC = "cl /nologo /Ox /MD /W3 /GS- /DNDEBUG /Tp%(source)s "
         LN = "/link /DLL /INCREMENTAL:NO /MANIFEST /OUT:%(output)s"
@@ -79 +81 @@
         else:
             COMPILE = " ".join((CC, LN))
-    elif True:
-        # If MSVC compiler is not available, try using mingw
-        # fPIC is not needed on windows
+    elif TCC:
+        # TinyCC compiler.
+        COMPILE = TCC + " -shared -rdynamic -Wall %(source)s -o %(output)s"
+    else:
+        # MinGW compiler.
         COMPILE = "gcc -shared -std=c99 -O2 -Wall %(source)s -o %(output)s -lm"
         if "SAS_OPENMP" in os.environ:
-            COMPILE = COMPILE + " -fopenmp"
-    else:
-        # If MSVC compiler is not available, try using tinycc
-        from tinycc import TCC
-        COMPILE = TCC + " -shared -rdynamic -Wall %(source)s -o %(output)s"
+            COMPILE += " -fopenmp"
 else:
-    COMPILE = "cc -shared -fPIC -fopenmp -std=c99 -O2 -Wall %(source)s -o %(output)s -lm"
+    # Generic unix compile
+    # On mac users will need the X code command line tools installed
+    COMPILE = "cc -shared -fPIC -std=c99 -O2 -Wall %(source)s -o %(output)s -lm"
+
+    # add openmp support if not running on a mac
+    if sys.platform != 'darwin':
+        #COMPILE = "gcc-mp-4.7 -shared -fPIC -std=c99 -fopenmp -O2 -Wall %s -o %s -lm -lgomp"
+        COMPILE += " -fopenmp"
 
 # Windows-specific solution