Fix generation of float4 when float3 requested

loopy/kernel/__init__.py

@@ -1337,14 +1337,11 @@ def __setstate__(self, state):
             "linearization",
             "iname_slab_increments",
             "loop_priority",
-            # applied_iname_rewrites
+            "applied_iname_rewrites",
             "index_dtype",
             "silenced_warnings",
 
             # missing:
-            # - applied_iname_rewrites
-            #   Contains pymbolic expressions, hence a (small) headache to hash.
-            #   Likely not needed for hash uniqueness => headache avoided.
 
             # - preamble_generators
             # - symbol_manglers
@@ -1355,7 +1352,7 @@ def __setstate__(self, state):
             #   resolve hash conflicts.
             ]
 
-    update_persistent_hash = update_persistent_hash
+    update_persistent_hash = update_persistent_hash  # noqa: RUF045
 
     @memoize_method
     def __hash__(self):

loopy/kernel/array.py

@@ -52,6 +52,7 @@
 
 if TYPE_CHECKING:
     from pymbolic import ArithmeticExpression
+    from pymbolic.typing import Integer
 
     from loopy.codegen import VectorizationInfo
     from loopy.kernel import LoopKernel
@@ -1103,43 +1104,67 @@ def none_pass_mapper(s: Expression | None) -> Expression | None:
         else:
             return self
 
+    def _vector_axis_index(self) -> int | None:
+        if self.dim_tags is None or self.shape is None:
+            return None
+
+        vec_axes = [
+            i for i, dim_tag in enumerate(self.dim_tags)
+            if isinstance(dim_tag, VectorArrayDimTag)
+        ]
+        if len(vec_axes) > 1:
+            raise LoopyError("more than one axis of '{self.name}' is tagged 'vec'")
+
+        if not vec_axes:
+            return None
+
+        iaxis, = vec_axes
+        return iaxis
+
+    def vector_length(self) -> Integer:
+        iaxis = self._vector_axis_index()
+        if iaxis is None:
+            return 1
+
+        assert isinstance(self.shape, tuple)
+
+        shape_i = self.shape[iaxis]
+        if not is_integer(shape_i):
+            raise LoopyError("shape of '%s' has non-constant-integer "
+                    "length for vector axis %d (0-based)" % (
+                        self.name, iaxis))
+
+        return shape_i
+
     def vector_size(self, target: TargetBase) -> int:
         """Return the size of the vector type used for the array
         divided by the basic data type.
 
         Note: For 3-vectors, this will be 4.
         """
-
-        if self.dim_tags is None or self.shape is None:
+        iaxis = self._vector_axis_index()
+        if iaxis is None:
             return 1
 
         assert isinstance(self.shape, tuple)
         assert isinstance(self.dtype, LoopyType)
 
-        saw_vec_tag = False
-
-        for i, dim_tag in enumerate(self.dim_tags):
-            if isinstance(dim_tag, VectorArrayDimTag):
-                if saw_vec_tag:
-                    raise LoopyError("more than one axis of '{self.name}' "
-                            "is tagged 'vec'")
-                saw_vec_tag = True
+        shape_i = self.shape[iaxis]
+        if not is_integer(shape_i):
+            raise LoopyError("shape of '%s' has non-constant-integer "
+                    "length for vector axis %d (0-based)" % (
+                        self.name, iaxis))
 
-                shape_i = self.shape[i]
-                if not is_integer(shape_i):
-                    raise LoopyError("shape of '%s' has non-constant-integer "
-                            "length for vector axis %d (0-based)" % (
-                                self.name, i))
-
-                vec_dtype = target.vector_dtype(self.dtype, shape_i)
-
-                return int(vec_dtype.itemsize) // int(self.dtype.itemsize)
+        if self.dim_tags is None or self.shape is None:
+            return 1
 
-        return 1
+        vec_dtype = target.vector_dtype(self.dtype, shape_i)
 
+        return int(vec_dtype.itemsize) // int(self.dtype.itemsize)
 
 # }}}
 
+
 def drop_vec_dims(
         dim_tags: tuple[ArrayDimImplementationTag, ...],
         t: tuple[T, ...]) -> tuple[T, ...]:

loopy/target/cuda.py

@@ -431,7 +431,7 @@ def wrap_global_constant(self, decl: Declarator) -> Declarator:
     def get_array_base_declarator(self, ary: ArrayBase) -> Declarator:
         dtype = ary.dtype
 
-        vec_size = ary.vector_size(self.target)
+        vec_size = ary.vector_length()
         if vec_size > 1:
             dtype = self.target.vector_dtype(dtype, vec_size)
 

loopy/target/opencl.py

@@ -751,7 +751,7 @@ def wrap_global_constant(self, decl: Declarator) -> Declarator:
     def get_array_base_declarator(self, ary: ArrayBase) -> Declarator:
         dtype = ary.dtype
 
-        vec_size = ary.vector_size(self.target)
+        vec_size = ary.vector_length()
         if vec_size > 1:
             dtype = self.target.vector_dtype(dtype, vec_size)
 

test/test_target.py

@@ -852,6 +852,29 @@ def test_cl_vectorize_ternary(ctx_factory):
     assert np.allclose(result, result_ref)
 
 
+def test_float3():
+    # https://github.com/inducer/loopy/issues/922
+    knl = lp.make_kernel(
+         "{ [i]: 0<=i<n }",
+         """
+         out[i] = a if i == 0 else b
+         """
+    )
+    vec_size = 3
+    knl = lp.split_array_axis(knl, "out", 0, vec_size)
+    knl = lp.split_iname(knl, "i", vec_size)
+    knl = lp.tag_inames(knl, {"i_inner": "vec"})
+    knl = lp.tag_array_axes(knl, "out", "c,vec")
+    knl = lp.assume(knl, f"n % {vec_size} = 0 and n>0")
+
+    knl = lp.add_and_infer_dtypes(knl,
+                    {"a": np.dtype(np.float32), "b": np.dtype(np.float32)})
+
+    device_code = lp.generate_code_v2(knl).device_code()
+
+    assert "float3" in device_code
+
+
 if __name__ == "__main__":
     import sys
     if len(sys.argv) > 1: