From 7f29d20573af85eac723944b426adf734ce96710 Mon Sep 17 00:00:00 2001
From: Andreas Kloeckner <inform@tiker.net>
Date: Sat, 8 Jun 2013 14:51:47 -0400
Subject: [PATCH] Add workarounds for
http://llvm.org/bugs/show_bug.cgi?id=16149
---
pyopencl/clrandom.py | 19 +++++++++++--------
pyopencl/elementwise.py | 18 ++++++++++++------
2 files changed, 23 insertions(+), 14 deletions(-)
diff --git a/pyopencl/clrandom.py b/pyopencl/clrandom.py
index 04ebe99e..bae170f5 100644
--- a/pyopencl/clrandom.py
+++ b/pyopencl/clrandom.py
@@ -187,7 +187,7 @@ class RanluxGenerator(object):
return "\n".join(lines)
@memoize_method
- def get_gen_kernel(self, dtype, flavor=""):
+ def get_gen_kernel(self, dtype, distribution="uniform"):
size_multiplier = 1
arg_dtype = dtype
@@ -212,16 +212,19 @@ class RanluxGenerator(object):
size_multiplier = 4
arg_dtype = np.float32
elif dtype == np.int32:
- assert flavor == ""
+ assert distribution == "uniform"
bits = 32
c_type = "int"
rng_expr = ("(shift "
- "+ convert_int4(scale * gen) "
- "+ convert_int4((scale / (1<<24)) * gen))")
+ "+ convert_int4((float) scale * gen) "
+ "+ convert_int4((float) (scale / (1<<24)) * gen))")
else:
raise TypeError("unsupported RNG data type '%s'" % dtype)
- rl_flavor = "%d%s" % (bits, flavor)
+ rl_flavor = "%d%s" % (bits, {
+ "uniform": "",
+ "normal": "norm"
+ }[distribution])
src = """//CL//
%(defines)s
@@ -231,7 +234,7 @@ class RanluxGenerator(object):
typedef %(output_t)s output_t;
typedef %(output_t)s4 output_vec_t;
#define NUM_WORKITEMS %(num_work_items)d
- #define RANLUX_FUNC ranluxcl##%(rlflavor)s
+ #define RANLUX_FUNC ranluxcl%(rlflavor)s
#define GET_RANDOM_NUM(gen) %(rng_expr)s
kernel void generate(
@@ -290,7 +293,7 @@ class RanluxGenerator(object):
if queue is None:
queue = ary.queue
- knl, size_multiplier = self.get_gen_kernel(ary.dtype, "")
+ knl, size_multiplier = self.get_gen_kernel(ary.dtype, "uniform")
knl(queue,
(self.num_work_items,), None,
self.state.data, ary.data, ary.size*size_multiplier,
@@ -315,7 +318,7 @@ class RanluxGenerator(object):
if queue is None:
queue = ary.queue
- knl, size_multiplier = self.get_gen_kernel(ary.dtype, "norm")
+ knl, size_multiplier = self.get_gen_kernel(ary.dtype, "normal")
knl(queue,
(self.num_work_items,), self.wg_size,
self.state.data, ary.data, ary.size*size_multiplier, sigma, mu)
diff --git a/pyopencl/elementwise.py b/pyopencl/elementwise.py
index 319ffb6b..3e63f4e0 100644
--- a/pyopencl/elementwise.py
+++ b/pyopencl/elementwise.py
@@ -721,12 +721,18 @@ def get_reverse_kernel(context, dtype):
@context_dependent_memoize
def get_arange_kernel(context, dtype):
- return get_elwise_kernel(context,
- "%(tp)s *z, %(tp)s start, %(tp)s step" % {
- "tp": dtype_to_ctype(dtype),
- },
- "z[i] = start + i*step",
- name="arange")
+ if dtype.kind == "c":
+ i = "%s_fromreal(i)" % complex_dtype_to_name(dtype)
+ else:
+ i = "(%s) i" % dtype_to_ctype(dtype)
+
+ return get_elwise_kernel(context, [
+ VectorArg(dtype, "z", with_offset=True),
+ ScalarArg(dtype, "start"),
+ ScalarArg(dtype, "step"),
+ ],
+ "z[i] = start + %s*step" % i,
+ name="arange")
@context_dependent_memoize
--
GitLab