[tvm] branch main updated: [fix] quantize op consistent with python description (#11872)

[ma...@apache.org]

This is an automated email from the ASF dual-hosted git repository.

masahi pushed a commit to branch main
in repository https://gitbox.apache.org/repos/asf/tvm.git


The following commit(s) were added to refs/heads/main by this push:
     new b99d93afec [fix] quantize op consistent with python description (#11872)
b99d93afec is described below

commit b99d93afec217f1dabc8b25e2e95251e50079ae8
Author: sisleyli <43...@users.noreply.github.com>
AuthorDate: Fri Jun 24 18:04:32 2022 +0800

    [fix] quantize op consistent with python description (#11872)
    
    * move round op before `add expanded_output_zero_point`
    * consistent with python description `(round(input_tensor/output_scale) + output_zero_point`
---
 src/relay/qnn/op/quantize.cc | 5 ++---
 1 file changed, 2 insertions(+), 3 deletions(-)

diff --git a/src/relay/qnn/op/quantize.cc b/src/relay/qnn/op/quantize.cc
index da33aaac81..1a16705932 100644
--- a/src/relay/qnn/op/quantize.cc
+++ b/src/relay/qnn/op/quantize.cc
@@ -142,11 +142,10 @@ Expr QuantizeLower(const Expr& input_tensor, const Expr& output_scale,
 
   const int32_t min_val = GetQmin(out_dtype);
   const int32_t max_val = GetQmax(out_dtype);
-  auto scale_data = Divide(input_tensor, expanded_output_scale);
+  auto scale_data = Round(Divide(input_tensor, expanded_output_scale));
   auto add_zero_point = Add(scale_data, Cast(expanded_output_zero_point, DataType::Float(32)));
   auto clamped_output = Clip(add_zero_point, min_val, max_val);
-  auto rounded_clamped_output = Round(clamped_output);
-  return Cast(rounded_clamped_output, out_dtype);
+  return Cast(clamped_output, out_dtype);
 }
 
 Expr QuantizeQnnCanonicalize(const Attrs& attrs, const Array<Expr>& new_args,