Add per_tensor overload for quantized_conv

backends/cadence/aot/ops_registrations.py

@@ -66,6 +66,12 @@
 lib.define(
     "quantized_conv.out(Tensor input, Tensor weight, Tensor bias, int[] stride, SymInt[] padding, int[] dilation, int groups, int input_zero_point, Tensor weight_zero_point, Tensor bias_scale, float out_scale, int out_zero_point, Tensor out_multiplier, Tensor out_shift, bool channel_last=False, *, Tensor(a!) out) -> Tensor(a!)"
 )
+lib.define(
+    "quantized_conv.per_tensor(Tensor input, Tensor weight, Tensor bias, int[] stride, SymInt[] padding, int[] dilation, int groups, int input_zero_point, int weight_zero_point, float bias_scale, float out_scale, int out_zero_point, int out_multiplier, int out_shift, bool channel_last=False) -> (Tensor Z)"
+)
+lib.define(
+    "quantized_conv.per_tensor_out(Tensor input, Tensor weight, Tensor bias, int[] stride, SymInt[] padding, int[] dilation, int groups, int input_zero_point, int weight_zero_point, float bias_scale, float out_scale, int out_zero_point, int out_multiplier, int out_shift, bool channel_last=False, *, Tensor(a!) out) -> Tensor(a!)"
+)
 
 lib.define(
     "quantized_matmul(Tensor X, int X_zero_point, Tensor Y, int Y_zero_point, Tensor? bias, int out_multiplier, int out_shift, int out_zero_point, bool transposed=False) -> (Tensor Z)"
@@ -171,6 +177,54 @@ def quantized_conv_meta(
     return input.new_empty(output_size, dtype=input.dtype)
 
 
+@register_fake("cadence::quantized_conv.per_tensor")
+def quantized_conv_per_tensor_meta(
+    input: torch.Tensor,
+    weight: torch.Tensor,
+    bias: torch.Tensor,
+    stride: Tuple[int],
+    padding: Tuple[int],
+    dilation: Tuple[int],
+    groups: int,
+    in_zero_point: int,
+    weight_zero_point: int,
+    bias_scale: float,
+    output_scale: float,
+    output_zero_point: int,
+    out_multiplier: int,
+    out_shift: int,
+    channel_last: bool = False,
+) -> torch.Tensor:
+    if channel_last:
+        out_channels, *kernel_size, _ = weight.shape
+    else:
+        out_channels, _, *kernel_size = weight.shape
+
+    in_size = input.shape
+    # Assert that the input tensor has at least 3 dimensions, and at most 6
+    assert len(in_size) > 2
+    assert len(in_size) < 6
+
+    # Compute the output tensor size
+    output_size = (
+        get_conv1d_output_size(
+            in_size,
+            out_channels,
+            stride[1],
+            padding[1],
+            dilation[1],
+            kernel_size[0],
+            channel_last,
+        )
+        if len(in_size) == 3
+        else get_conv2d_output_size(
+            in_size, out_channels, stride, padding, dilation, kernel_size, channel_last
+        )
+    )
+
+    return input.new_empty(output_size, dtype=input.dtype)
+
+
 @register_fake("cadence::quantized_layer_norm")
 def quantized_layer_norm_meta(
     input: torch.Tensor,