[WebNN] Only limit int32 input for dequantizeLinear for ORT default CPU EP [chromium/src : main]
0 views
Skip to first unread message
Wang, Wei4 (Gerrit)
Feb 27, 2026, 8:08:44 PM (yesterday) Feb 27
to Hu, Ningxin, Miao, Bin, Chromium LUCI CQ, AyeAye, chromium...@chromium.org, Jiewei Qian, blink-...@chromium.org, ipc-securi...@chromium.org
Attention needed from Hu, Ningxin
Wang, Wei4 added 1 comment![]( "Open in Gerrit")
Patchset-level comments
Related details
Attention is currently required from:
- Hu, Ningxin
Submit Requirements:
Code-Coverage
Code-Owners
Code-Review
Review-Enforcement
| Inspect html for hidden footers to help with email filtering. To unsubscribe visit settings. |
Hu, Ningxin (Gerrit)
Feb 27, 2026, 8:40:16 PM (yesterday) Feb 27
to Wang, Wei4, Miao, Bin, Chromium LUCI CQ, AyeAye, chromium...@chromium.org, Jiewei Qian, blink-...@chromium.org, ipc-securi...@chromium.org
Attention needed from Wang, Wei4
Hu, Ningxin added 1 comment![]( "Open in Gerrit")
Commit Message
Line 14, Patchset 2 (Latest):
This CL also makes input and zeroPoint of dequantizeLinear share
the same OpSupportLimits item, because their limitations are same
for different backends after CL[2] is merged.Hu, Ningxin . unresolved
Does it make sense to have a separate CL for this change? That would make this CL focusing on ORT backend change.
Related details
Attention is currently required from:
- Wang, Wei4
Submit Requirements:
Code-Coverage
Code-Owners
Code-Review
No-Unresolved-Comments
Review-Enforcement
| Inspect html for hidden footers to help with email filtering. To unsubscribe visit settings. |
Wang, Wei4 (Gerrit)
Feb 27, 2026, 8:45:30 PM (yesterday) Feb 27
to Hu, Ningxin, Miao, Bin, Chromium LUCI CQ, AyeAye, chromium...@chromium.org, Jiewei Qian, blink-...@chromium.org, ipc-securi...@chromium.org
Attention needed from Hu, Ningxin
Wang, Wei4 added 1 comment![]( "Open in Gerrit")
Commit Message
Line 14, Patchset 2 (Latest):
This CL also makes input and zeroPoint of dequantizeLinear share
the same OpSupportLimits item, because their limitations are same
for different backends after CL[2] is merged.Hu, Ningxin . unresolved
Does it make sense to have a separate CL for this change? That would make this CL focusing on ORT backend change.
Attention is currently required from:
- Hu, Ningxin
Submit Requirements:
Code-Coverage
Code-Owners
Code-Review
No-Unresolved-Comments
Review-Enforcement
| Inspect html for hidden footers to help with email filtering. To unsubscribe visit settings. |
Wang, Wei4 (Gerrit)
Feb 27, 2026, 10:22:24 PM (22 hours ago) Feb 27
to Hu, Ningxin, Miao, Bin, Chromium LUCI CQ, AyeAye, chromium...@chromium.org, Jiewei Qian, blink-...@chromium.org, ipc-securi...@chromium.org
Attention needed from Hu, Ningxin
Wang, Wei4 added 2 comments![]( "Open in Gerrit")
Patchset-level comments
Commit Message
Line 14, Patchset 2:
This CL also makes input and zeroPoint of dequantizeLinear share
the same OpSupportLimits item, because their limitations are same
for different backends after CL[2] is merged.Hu, Ningxin . resolved
Wang, Wei4Does it make sense to have a separate CL for this change? That would make this CL focusing on ORT backend change.
sure, I will separate the chages.
Wang, Wei4
Done
Related details
Attention is currently required from:
- Hu, Ningxin
Submit Requirements:
Code-Coverage
Code-Owners
Code-Review
Review-Enforcement
| Inspect html for hidden footers to help with email filtering. To unsubscribe visit settings. |
Hu, Ningxin (Gerrit)
3:48 AM (17 hours ago) 3:48 AM
to Wang, Wei4, Rafael Cintron, Dwayne Robinson, Miao, Bin, Chromium LUCI CQ, AyeAye, chromium...@chromium.org, Jiewei Qian, blink-...@chromium.org, ipc-securi...@chromium.org
Attention needed from Rafael Cintron and Wang, Wei4
Hu, Ningxin added 5 comments![]( "Open in Gerrit")
Patchset-level comments
File services/webnn/ort/context_impl_ort.h
Line 68, Patchset 4 (Latest):
bool dequantizeLinear_support_int32);Hu, Ningxin . unresolved
```suggestion
bool dequantize_linear_input_support_int32);
```
Line 47, Patchset 4 (Latest):
bool dequantizeLinear_support_int32,Hu, Ningxin . unresolved
```suggestion
bool dequantize_linear_input_support_int32,
```
File services/webnn/ort/context_impl_ort.cc
Line 78, Patchset 4 (Latest):
bool is_default_CPU_EP_first_selected = Environment::IsDefaultCpuEpDevice(Hu, Ningxin . unresolved
```suggestion
bool dequantize_linear_input_support_int32= !Environment::IsDefaultCpuEpDevice(
```
Line 209, Patchset 4 (Latest):
{dequantizeLinear_support_int32 ? kInts4To8Int32
: DataTypeConstraint::kInts4ToInts8,Hu, Ningxin . unresolved
This change seems to be unnecessary.
Related details
Attention is currently required from:
- Rafael Cintron
- Wang, Wei4
Submit Requirements:
Code-Coverage
Code-Owners
Code-Review
No-Unresolved-Comments
Review-Enforcement
| Inspect html for hidden footers to help with email filtering. To unsubscribe visit settings. |
Wang, Wei4 (Gerrit)
4:00 AM (17 hours ago) 4:00 AM
to Rafael Cintron, Dwayne Robinson, Hu, Ningxin, Miao, Bin, Chromium LUCI CQ, AyeAye, chromium...@chromium.org, Jiewei Qian, blink-...@chromium.org, ipc-securi...@chromium.org
Attention needed from Hu, Ningxin and Rafael Cintron
Wang, Wei4 added 1 comment![]( "Open in Gerrit")
File services/webnn/ort/context_impl_ort.cc
Line 209, Patchset 4 (Latest):
{dequantizeLinear_support_int32 ? kInts4To8Int32
: DataTypeConstraint::kInts4ToInts8,Hu, Ningxin . unresolved
This change seems to be unnecessary.
Wang, Wei4
According to WebNN spec, zeroPoint has same "allowed data types" with input, so is it better to keep them same?
Related details
Attention is currently required from:
- Hu, Ningxin
- Rafael Cintron
Wang, Wei4 (Gerrit)
5:14 AM (16 hours ago) 5:14 AM
to Rafael Cintron, Dwayne Robinson, Hu, Ningxin, Miao, Bin, Chromium LUCI CQ, AyeAye, chromium...@chromium.org, Jiewei Qian, blink-...@chromium.org, ipc-securi...@chromium.org
Attention needed from Hu, Ningxin and Rafael Cintron
Wang, Wei4 added 1 comment![]( "Open in Gerrit")
File services/webnn/ort/context_impl_ort.cc
Line 209, Patchset 4 (Latest):
{dequantizeLinear_support_int32 ? kInts4To8Int32
: DataTypeConstraint::kInts4ToInts8,Hu, Ningxin . unresolved
Wang, Wei4This change seems to be unnecessary.
According to WebNN spec, zeroPoint has same "allowed data types" with input, so is it better to keep them same?
Wang, Wei4
From the implementation perspective, they can share the same OpSupportLimits item if they are same, just like the changes in CL: https://chromium-review.googlesource.com/c/chromium/src/+/7618977
Reply all
Reply to author
Forward
0 new messages