commit 6e7bd3bf518e44172eab50ca5f848479a8b42f98
parent 3135b1909d73d6d7aab38f326389a3d76f19f42c
Author: Miao Bin <bin.miao@intel.com>
Date: Tue, 16 Dec 2025 08:47:10 +0000
Bug 2005690 [wpt PR 56708] - WebNN: Limit int32 input for dequantizeLinear on ORT backend, a=testonly
Automatic update from web-platform-tests
WebNN: Limit int32 input for dequantizeLinear on ORT backend
The ONNX Runtime CPUExecutionProvider has a limitation for the
dequantizeLinear operator: "DequantizeLinear with type int32 or float8
should have no zero point or all zero points should be 0" [1]. And WebNN
WPT has some cases use int32 inputs and non-zero zero_points to test
this operator. These cases will result in context lost triggered by ORT
session run failure.
This CL limits int32 input for dequantizeLinear on ORT backend to avoid
the context lost issue.
[1]
https://github.com/microsoft/onnxruntime/blob/rel-1.23.0/onnxruntime/core/providers/cpu/quantization/quantize_linear.cc#L530C18-L530C120
Bug: 467185159
Change-Id: I8913d9f63a9ee168cee73e4d950febccf57552a8
Reviewed-on: https://chromium-review.googlesource.com/c/chromium/src/+/7244911
Reviewed-by: Rafael Cintron <rafael.cintron@microsoft.com>
Reviewed-by: Hu, Ningxin <ningxin.hu@intel.com>
Commit-Queue: Miao, Bin <bin.miao@intel.com>
Cr-Commit-Position: refs/heads/main@{#1557851}
--
wpt-commits: 7142ea82ca80f8836de9e4c4fbf0e58fd24d80ae
wpt-pr: 56708
Diffstat:
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/testing/web-platform/tests/webnn/resources/minimum_datatype_set.json b/testing/web-platform/tests/webnn/resources/minimum_datatype_set.json
@@ -181,7 +181,7 @@
},
"dequantizeLinear": {
"input": {
- "dataTypes": ["int32", "int8", "uint8"],
+ "dataTypes": ["int8", "uint8"],
"rankRange": {"min": 0, "max": 5}
},
"scale": {"dataTypes": ["float32"], "rankRange": {"min": 0, "max": 5}},
