[ONNX] Support depth_to_space op for FQ2I (apache#8966)

* WIP support per-channel quantization

* more WIP

* More WIP

* fix issue with per-channel bias_add

* Fix fake quantize tests (#4)

* Fixed fake quantize issues.

* Formatting.

* Cleanup unused imports

* Fix real int8 tests.

* Add Relu

* One more little one (#5)

* Fixed fake quantize issues.

* Formatting.

* Cleanup unused imports

* Fix real int8 tests.

* Fix requantize shape bug.

* Non-working Per-channel Dense

* Fix legalization for non spatial operators. (#6)

* Fix legalization for non spatial operators.

* Fix axis checks for end2end functionality.

* fix axis normalization

fix lint

fix lint again

* Per channel fq2i (#8)

* WIP support per-channel quantization

* more WIP

* More WIP

* fix issue with per-channel bias_add

* Fix fake quantize tests (#4)

* Fixed fake quantize issues.

* Formatting.

* Cleanup unused imports

* Fix real int8 tests.

* Add Relu

* One more little one (#5)

* Fixed fake quantize issues.

* Formatting.

* Cleanup unused imports

* Fix real int8 tests.

* Fix requantize shape bug.

* Non-working Per-channel Dense

* Fix legalization for non spatial operators. (#6)

* Fix legalization for non spatial operators.

* Fix axis checks for end2end functionality.

* fix axis normalization

fix lint

fix lint again

* Fix bug in requantize dimension expansion.

* Format.

Co-authored-by: Josh Fromm <jwfromm@octoml.ai>

* respond to review comments

* start dtos

* wip depth_to_space

* dtos ident

Co-authored-by: Matthew <mbrookhart@octoml.ai>
Co-authored-by: Josh Fromm <jwfromm@octoml.ai>

python/tvm/relay/transform/fake_quantization_to_integer.py

@@ -87,6 +87,7 @@ def identity(expr, type_map):
 register_unary_identity("expand_dims")
 register_unary_identity("nn.max_pool2d")
 register_unary_identity("nn.batch_flatten")
+register_unary_identity("nn.depth_to_space")
 
 
 @register_fake_quantization_to_integer("nn.avg_pool2d")

tests/python/relay/test_pass_fake_quantization_to_integer.py

@@ -501,3 +501,16 @@ def test_fake_quantize_pad():
     x_np = np.random.randint(-25, 25, size=[1, 383, 128], dtype="int8")
 
     compare_fq_to_int(op, [x_np])
+
+
+def test_fake_quantize_depth_to_space():
+    x = relay.var("x", shape=[1, 3, 224, 224], dtype="int8")
+
+    zero = relay.const(0)
+    x = relay.qnn.op.dequantize(x, relay.const(2.0), zero)
+    op = relay.op.nn.depth_to_space(x, 4)
+    op = relay.qnn.op.quantize(op, relay.const(2.0), zero)
+
+    x_np = np.random.randint(-128, 127, size=[1, 3, 224, 224], dtype="int8")
+
+    compare_fq_to_int(op, [x_np])