From e7ccefc67b47401f98dc90639bc0e4a1884112e4 Mon Sep 17 00:00:00 2001
From: Andreas Kloeckner <inform@tiker.net>
Date: Thu, 4 Oct 2012 19:59:48 -0400
Subject: [PATCH] Make type inference of 'float-float' not come out to double.

---
 MEMO                        |  2 ++
 loopy/codegen/expression.py | 21 ++++++++++++++++++++-
 test/test_loopy.py          | 24 ++++++++++++++++++++++++
 3 files changed, 46 insertions(+), 1 deletion(-)

diff --git a/MEMO b/MEMO
index 15d4ae95b..62bfa7f35 100644
--- a/MEMO
+++ b/MEMO
@@ -46,6 +46,8 @@ To-do
 
 - Fix timer / call code
 
+- use memory pools for arrays
+
 Fixes:
 
 - Group instructions by dependency/inames for scheduling, to
diff --git a/loopy/codegen/expression.py b/loopy/codegen/expression.py
index deb20098f..73ed05b43 100644
--- a/loopy/codegen/expression.py
+++ b/loopy/codegen/expression.py
@@ -49,7 +49,26 @@ class TypeInferenceMapper(CombineMapper):
         return result
 
     def map_constant(self, expr):
-        return np.asarray(expr).dtype
+        if isinstance(expr, int):
+            for tp in [np.int8, np.int16, np.int32, np.int64]:
+                iinfo = np.iinfo(tp)
+                if iinfo.min <= expr <= iinfo.max:
+                    return np.dtype(tp)
+
+            else:
+                raise TypeInferenceFailure("integer constant '%s' too large" % expr)
+
+        dt = np.asarray(expr).dtype
+        if dt.kind == "f":
+            # deduce the smaller type by default
+            return np.dtype(np.float32)
+        elif dt.kind == "f":
+            # deduce the smaller type by default
+            return np.dtype(np.complex64)
+        elif hasattr(expr, "dtype"):
+            return expr.dtype
+        else:
+            raise TypeInferenceFailure("cannot deduce type of constant '%s'" % expr)
 
     def map_subscript(self, expr):
         return self.rec(expr.aggregate)
diff --git a/test/test_loopy.py b/test/test_loopy.py
index 6e73e82ef..595a263e7 100644
--- a/test/test_loopy.py
+++ b/test/test_loopy.py
@@ -14,6 +14,30 @@ __all__ = ["pytest_generate_tests",
 
 
 
+def test_type_inference_no_artificial_doubles(ctx_factory):
+    ctx = ctx_factory()
+
+    knl = lp.make_kernel(ctx.devices[0],
+            "{[i]: 0<=i<n}",
+            """
+                <> bb = a[i] - b[i]
+                c[i] = bb
+                """,
+            [
+                lp.GlobalArg("a", np.float32, shape=("n",)),
+                lp.GlobalArg("b", np.float32, shape=("n",)),
+                lp.GlobalArg("c", np.float32, shape=("n",)),
+                lp.ValueArg("n", np.int32),
+                ],
+            assumptions="n>=1")
+
+    for k in lp.generate_loop_schedules(knl):
+        code = lp.generate_code(k)
+        assert "double" not in code
+
+
+
+
 def test_simple_side_effect(ctx_factory):
     ctx = ctx_factory()
 
-- 
GitLab