Add support for Bipolar and Binary FINN datatype for Quant op. (#41)

* Add support for Bipolar and Binary FINN datatype for Quant op. * [Quant] custom qnt execution for bipolar until Brevitas bug resolved Co-authored-by: Yaman Umuroglu <[email protected]>
Xilinx · Aug 26, 2021 · 1429935 · 1429935
1 parent 3aaaff4
commit 1429935
Showing 1 changed file with 23 additions and 12 deletions.
diff --git a/src/finn/custom_op/general/quant.py b/src/finn/custom_op/general/quant.py
@@ -27,8 +27,8 @@
 # OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 import numpy as np
-import onnx.helper as helper
 
+import onnx.helper as helper
 from finn.core.datatype import DataType
 from finn.custom_op.base import CustomOp
 
@@ -97,14 +97,19 @@ def quant(inp_tensor, scale, zeropt, bitwidth, signed, narrow):
     # Scaling
     y_int = inp_tensor / scale
     y_int = y_int + zeropt
-    # Clamping
-    min_int_val = min_int(signed, narrow, bitwidth)
-    max_int_val = max_int(signed, narrow, bitwidth)
-    y_int = np.where(y_int > max_int_val, max_int_val.astype(y_int.dtype), y_int)
-    y_int = np.where(y_int < min_int_val, min_int_val.astype(y_int.dtype), y_int)
-    # Rounding
-    y_int = np.round(y_int)
-
+    if bitwidth == 1 and signed:
+        # BUG: 1-bit Quant ops currently not exported correctly
+        # manually convert to bipolar values
+        y_ones = np.ones(y_int.shape, dtype=y_int.dtype)
+        y_int = np.where(y_int >= 0.0, y_ones, -y_ones)
+    else:
+        # Clamping
+        min_int_val = min_int(signed, narrow, bitwidth)
+        max_int_val = max_int(signed, narrow, bitwidth)
+        y_int = np.where(y_int > max_int_val, max_int_val.astype(y_int.dtype), y_int)
+        y_int = np.where(y_int < min_int_val, min_int_val.astype(y_int.dtype), y_int)
+        # Rounding
+        y_int = np.round(y_int)
     # Re-scaling
     out_tensor = y_int - zeropt
     out_tensor = out_tensor * scale
@@ -165,10 +170,16 @@ def get_quant_config(self, model):
         zero_zeropt = np.all(zeropt == 0.0)
         assert zero_zeropt, "Only zero_point=0 Quant nodes supported for now"
         if unit_scale and zero_zeropt:
-            if signed:
-                finn_dt = DataType["INT" + str(bitwidth)]
+            if bitwidth == 1:
+                if signed:
+                    finn_dt = DataType["BIPOLAR"]
+                else:
+                    finn_dt = DataType["BINARY"]
             else:
-                finn_dt = DataType["UINT" + str(bitwidth)]
+                if signed:
+                    finn_dt = DataType["INT" + str(bitwidth)]
+                else:
+                    finn_dt = DataType["UINT" + str(bitwidth)]
         else:
             if signed:
                 finn_dt = DataType["SCALEDINT" + str(bitwidth)]