Use ConcreteTemplate for intrinsics and AbstractTemplate for operators

sklam · Sep 13, 2022 · Feb 16, 2022 · Feb 16, 2022 · Feb 17, 2022 · Feb 17, 2022
commit 87b8c2cd18375a83991922cadc3749f261107ef6
@@ -338,98 +338,66 @@ def generic(self, args, kws):
         return signature(a, test, a, a)
 
 
-#def _genfp16_unary(l_key):
-#    @register
-#    class Cuda_fp16_unary(ConcreteTemplate):
-#        key = l_key
-#        cases = [signature(types.float16, types.float16)]
-#
-#    return Cuda_fp16_unary
-
 def _genfp16_unary(l_key):
+    @register
+    class Cuda_fp16_unary(ConcreteTemplate):
+        key = l_key
+        cases = [signature(types.float16, types.float16)]
+
+    return Cuda_fp16_unary
+
+
+def _genfp16_unary_operator(l_key):
     @register
     class Cuda_fp16_unary(AbstractTemplate):
         key = l_key
+
         def generic(self, args, kws):
             assert not kws
             if len(args) == 1 and args[0] == types.float16:
                 return signature(types.float16, types.float16)
 
     return Cuda_fp16_unary
 
-#def _genfp16_binary(l_key):
-#    @register
-#    class Cuda_fp16_binary(ConcreteTemplate):
-#        key = l_key
-#        cases = [signature(types.float16, types.float16, types.float16)]
-#
-#    return Cuda_fp16_binary
 
 def _genfp16_binary(l_key):
+    @register
+    class Cuda_fp16_binary(ConcreteTemplate):
+        key = l_key
+        cases = [signature(types.float16, types.float16, types.float16)]
+
+    return Cuda_fp16_binary
+
+
+def _genfp16_binary_operator(l_key):
     @register
     class Cuda_fp16_binary(AbstractTemplate):
         key = l_key
+
         def generic(self, args, kws):
             assert not kws
-            if len(args) == 2 and args[0] == types.float16 and args[1] == types.float16:
+            if len(args) == 2 and \
+                    (args[0] == types.float16 or args[1] == types.float16):
                 return signature(types.float16, types.float16, types.float16)
 
     return Cuda_fp16_binary
 
 
 Cuda_hadd = _genfp16_binary(cuda.fp16.hadd)
-Cuda_add = _genfp16_binary(operator.add)
-Cuda_iadd = _genfp16_binary(operator.iadd)
-
+Cuda_add = _genfp16_binary_operator(operator.add)
+Cuda_iadd = _genfp16_binary_operator(operator.iadd)
 Cuda_hsub = _genfp16_binary(cuda.fp16.hsub)
-Cuda_sub = _genfp16_binary(operator.sub)
-Cuda_isub = _genfp16_binary(operator.isub)
-
-
+Cuda_sub = _genfp16_binary_operator(operator.sub)
+Cuda_isub = _genfp16_binary_operator(operator.isub)
 Cuda_hmul = _genfp16_binary(cuda.fp16.hmul)
-Cuda_mul = _genfp16_binary(operator.mul)
-Cuda_imul = _genfp16_binary(operator.imul)
-
-
-Cuda_hneg = _genfp16_unary(cuda.fp16.hneg)
+Cuda_mul = _genfp16_binary_operator(operator.mul)
+Cuda_imul = _genfp16_binary_operator(operator.imul)
+Cuda_hneg = _genfp16_unary_operator(cuda.fp16.hneg)
 Cuda_neg = _genfp16_unary(operator.neg)
-Cuda_habs = _genfp16_unary(cuda.fp16.habs)
+Cuda_habs = _genfp16_unary_operator(cuda.fp16.habs)
 Cuda_abs = _genfp16_unary(operator.abs)
 
 
-class FP16BinOp(ConcreteTemplate):
-    cases = [signature(types.float16, types.float16, types.float16)]
-
-#@register_global(operator.iadd)
-#@register_global(operator.add)
-#class BinOpFP16Add(FP16BinOp):
-#    pass
-
-#@register_global(operator.isub)
-#@register_global(operator.sub)
-#class BinOpFP16Sub(FP16BinOp):
-#    pass
-
-#@register_global(operator.imul)
-#@register_global(operator.mul)
-#class BinOpFP16Mul(FP16BinOp):
-#    pass
-
-
-class FP16UnaryOp(ConcreteTemplate):
-    cases = [signature(types.float16, types.float16)]
-
-
-#@register_global(operator.neg)
-#class UnaryOpFP16Neg(FP16UnaryOp):
-#    pass
-
-
-#@register_global(operator.abs)
-#class UnaryOpFP16Abs(FP16UnaryOp):
-#    pass
-
-
 # generate atomic operations
 def _gen(l_key, supported_types):
     @register