Specify the operand data type constraints of operations #646

inexorabletash · 2024-04-18T14:16:35Z

Introduce constraints for input operands, either directly (e.g. input's dataType can only be "float32" or "float16") or indirectly (e.g. weight's dataType must be the same as input's).

Fixes #283

Preview | Diff

Introduce constraints for input operands, either directly (e.g. input's dataType can only be "float32" or "float16") or indirectly (e.g. weight's dataType must be the same as input's). Fixes webmachinelearning#283

huningxin

LGTM with a nit, thanks much!

index.bs

huningxin

LGTM, thanks!

index.bs

huningxin

LGTM, thanks for the fix!

inexorabletash · 2024-04-25T00:34:55Z

Conflict resolved. @fdwr, good to merge?

inexorabletash · 2024-04-25T20:33:18Z

I'd missed reduceMean() somehow. Addressed in bd9b0cf

huningxin

LGTM

fdwr

Thanks JB. Tedious work there! I update GEMM C's data type check (please review). The rest LGTM, and my ideas can wait.

index.bs

fdwr · 2024-04-26T05:30:56Z

index.bs

@@ -4923,21 +4971,21 @@ partial interface MLGraphBuilder {

    <div algorithm>
    The <dfn method for=MLGraphBuilder>reduceProduct(|input|, |options|)</dfn> method steps are:
-        1. Let |output| be the result of running the [=MLGraphBuilder/reduce-op | create reduce operation=] given "reduceProduct", |input| and |options|.
+        1. Let |output| be the result of running the [=MLGraphBuilder/reduce-op | create reduce operation=] given "reduceProduct", |input|, |options|, and « {{MLOperandDataType/"float32"}}, {{MLOperandDataType/"float16"}}, {{MLOperandDataType/"int32"}}, {{MLOperandDataType/"uint32"}} ».


I like how you define them as type lists here, e.g. {{MLOperandDataType/"float32"}}, {{MLOperandDataType/"float16"}}, {{MLOperandDataType/"int32"}}, {{MLOperandDataType/"int8"}}. Other specs I've seen that list supported data types (some don't and just leave it to you as the user to try it and find out! 🎲🤞) declare them as lists, rather than explicit prose (e.g. [=MLOperand/dataType=] is not {{MLOperandDataType/"float32"}} or {{MLOperandDataType/"float16"}}). I'm not suggesting you change those above right now, but if we start adding more type lists to more places, it will be nice to have them consistent.

Yeah, I raised this over in #643 (comment) - i.e. using the list style everywhere. I'm a fan and happy to change things to this form everywhere:

If input’s dataType is not in « "float32", "float16" », then ...

If input’s dataType is not in « "float32", "float16", "int32", "int8" », then ...

... with the question of which formulation to use if there's a single type:

If input’s dataType is not "uint8", then ...

If input’s dataType is not in « "uint8" », then ...

Thoughts? (Depends on later conversation, though.)

I guess a type list of 1 type is still a type list. So for consistency imo, If input’s dataType is not « "uint8" », then ....

index.bs

fdwr · 2024-04-26T05:56:37Z

index.bs

@@ -4385,6 +4428,8 @@ partial interface MLGraphBuilder {
    The <dfn method for=MLGraphBuilder>matmul(|a|, |b|)</dfn> method steps are:
  </summary>
    1. If [=MLGraphBuilder/validating operand=] with [=this=] and any of |a| and |b| returns false, then [=exception/throw=] a {{TypeError}}.
+    1. If |a|'s [=MLOperand/dataType=] is not {{MLOperandDataType/"float32"}} or {{MLOperandDataType/"float16"}}, then [=exception/throw=] a {{TypeError}}.
+    1. If |b|'s [=MLOperand/dataType=] is not equal to |a|'s [=MLOperand/dataType=], then [=exception/throw=] a {{TypeError}}.


Hmm, yeah, increasingly the more I read this, more I think we should just add a new section of "data types at a glance", rather than burying them inside the algorithm steps. The steps can still say "if data type is not one of the blessed types", but it would be easier to see what's what. e.g.

Compute the matrix product of two input tensors.

partial interface MLGraphBuilder {
MLOperand matmul(MLOperand a, MLOperand b);
};

Arguments:

a: an MLOperand. The first input tensor which is at least 2D.‌

b: an MLOperand. The second input tensor which is at least 2D.

Returns: an MLOperand. The output tensor that contains the matrix product of two input tensors.

Types:

a: float16, float32

b: same as a

Computes the matrix product of two input tensors as follows...

Whatcha think? Later CR? This CR? Alternate idea instead?

Let me try a parallel PR to this one that tries this. I have a love/hate relationship with the Arguments/Returns preamble; since it's not really rigorous I think of it as non-normative. Listing types out-of-band will need some formal structure, and e.g. "same as a" needs to be precisely defined (i.e. the actual data type needs to match)

Okay, @fdwr I put up #657 as an experiment, updating only batchNormalization and adding a definition, and including two variations (list and table), which I think captures what you were going for. It's a clone of this PR with one additional commit if you want to look at the diff.

The use of "allowed data types" links in the steps means the reader/developer needs to understand that there's a nearby table giving the constraint details - I didn't want to add explicit dfns/links for each table entry, but maybe we should. I didn't find exact precedent but this is vaguely similar to CSS property definitions.

~~The bikeshed for linking the argument is unfortunately very verbose.~~ I figured out a shorter syntax!

I've gotten some feedback from @a-sully and @reillyeon about the approaches, and what this could enable (e.g. merge w/ Arguments, make it define shape as well, ...).

I might suggest that we merge this PR, either as-is or with the everything is a list change suggested above, and then iterate on table approach. Having this PR together has already identified some implementation gaps so start with this then iterate?

I might suggest that we merge this PR, either as-is

Yep yep. I'll do that now.

make it define shape as well

That would be interesting. Some other specs I've read were hand-wavy about how the output shape calculations occur, leaving a lot of room for ~~bugs~~ interpretation in implementations. So, seeing that clearly conveyed, without needing to refer to some separate Javascript codebase, would be nice (but having the WPT reference implementation is also very valuable).

As specified in webmachinelearning/webnn#646 Bug: 328567884 Change-Id: I620653f574303e6f5b18d3b540b647f07e0cf64c

SHA: b5677b6 Reason: push, by fdwr Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>

As specified in webmachinelearning/webnn#646 Bug: 328567884 Change-Id: I620653f574303e6f5b18d3b540b647f07e0cf64c Reviewed-on: https://chromium-review.googlesource.com/c/chromium/src/+/5492307 Reviewed-by: ningxin hu <[email protected]> Commit-Queue: Austin Sullivan <[email protected]> Cr-Commit-Position: refs/heads/main@{#1293407}

As specified in webmachinelearning/webnn#646 Bug: 328567884 Change-Id: Ia55a214e7ad281ec3c8911e9116f388fac209d05

This reverts commit ef29e11bfb1f1f4eae0935fb58f0e4129620837b. Reason for revert: LUCI Bisection has identified this change as the cause of a test failure. See the analysis: https://ci.chromium.org/ui/p/chromium/bisection/test-analysis/b/5728899092709376 Sample build with failed test: https://ci.chromium.org/b/8748452022241283393 Affected test(s): [ninja://services:services_unittests/WebNNGraphImplBackendTest.BuildAndComputeSingleOperatorGather](https://ci.chromium.org/ui/test/chromium/ninja:%2F%2Fservices:services_unittests%2FWebNNGraphImplBackendTest.BuildAndComputeSingleOperatorGather?q=VHash%3A3db3852ce13cc70a) If this is a false positive, please report it at http://b.corp.google.com/createIssue?component=1199205&description=Analysis%3A+https%3A%2F%2Fci.chromium.org%2Fui%2Fp%2Fchromium%2Fbisection%2Ftest-analysis%2Fb%2F5728899092709376&format=PLAIN&priority=P3&title=Wrongly+blamed+https%3A%2F%2Fchromium-review.googlesource.com%2Fc%2Fchromium%2Fsrc%2F%2B%2F5520313&type=BUG Original change's description: > webnn: Enforce input data type constraints for gather indices > > As specified in webmachinelearning/webnn#646 > > Bug: 328567884 > Change-Id: I33eba7e1def430b1cb94e3e7a4868e82c5bbd9a3 > Cq-Include-Trybots: luci.chromium.try:mac14-blink-rel,mac14.arm64-blink-rel > Reviewed-on: https://chromium-review.googlesource.com/c/chromium/src/+/5520313 > Reviewed-by: Austin Sullivan <[email protected]> > Commit-Queue: Lisha Guo <[email protected]> > Reviewed-by: ningxin hu <[email protected]> > Cr-Commit-Position: refs/heads/main@{#1298443} > Bug: 328567884 Change-Id: Iab874e915f5603861929cdf07ba4c09698b76503 No-Presubmit: true No-Tree-Checks: true No-Try: true Reviewed-on: https://chromium-review.googlesource.com/c/chromium/src/+/5528819 Reviewed-by: Austin Sullivan <[email protected]> Reviewed-by: Phillis Tang <[email protected]> Commit-Queue: Austin Sullivan <[email protected]> Cr-Commit-Position: refs/heads/main@{#1298718}

As specified in webmachinelearning/webnn#646 Bug: 328567884 Change-Id: I5e600bdc791ecd4530408b65c15dd20611c211a3 Cq-Include-Trybots: luci.chromium.try:mac14-blink-rel,mac14.arm64-blink-rel

This is a reland of commit ef29e11bfb1f1f4eae0935fb58f0e4129620837b Original change's description: > webnn: Enforce input data type constraints for gather indices > > As specified in webmachinelearning/webnn#646 > > Bug: 328567884 > Change-Id: I33eba7e1def430b1cb94e3e7a4868e82c5bbd9a3 > Cq-Include-Trybots: luci.chromium.try:mac14-blink-rel,mac14.arm64-blink-rel > Reviewed-on: https://chromium-review.googlesource.com/c/chromium/src/+/5520313 > Reviewed-by: Austin Sullivan <[email protected]> > Commit-Queue: Lisha Guo <[email protected]> > Reviewed-by: ningxin hu <[email protected]> > Cr-Commit-Position: refs/heads/main@{#1298443} Bug: 328567884 Change-Id: Ide6a0e1bc6d3ebea7dce6bf37da4318e7b6e1bd1 Cq-Include-Trybots: luci.chromium.try:mac14-blink-rel,mac14.arm64-blink-rel

… abs, neg and relu, a=testonly Automatic update from web-platform-tests webnn: Enforce data type constraints for abs, neg and relu As specified in webmachinelearning/webnn#646 Bug: 328567884 Change-Id: Id377fcf423f72e2dd6d7851fa979fdfa8caeefdc Reviewed-on: https://chromium-review.googlesource.com/c/chromium/src/+/5517439 Commit-Queue: Shiyi Zou <[email protected]> Reviewed-by: Austin Sullivan <[email protected]> Reviewed-by: ningxin hu <[email protected]> Cr-Commit-Position: refs/heads/main@{#1297321} -- wpt-commits: 602dae0eda79af364575e1f99ad39967bdd7fbc9 wpt-pr: 46075

…ts for resample2d, a=testonly Automatic update from web-platform-tests webnn: Enforce input data type constraints for resample2d As specified in webmachinelearning/webnn#646 Bug: 328567884 Change-Id: Icea0e4a586aee33d7993d6e72393412b981d26ab Reviewed-on: https://chromium-review.googlesource.com/c/chromium/src/+/5495876 Reviewed-by: Austin Sullivan <[email protected]> Reviewed-by: ningxin hu <[email protected]> Commit-Queue: Lisha Guo <[email protected]> Cr-Commit-Position: refs/heads/main@{#1297332} -- wpt-commits: b5ab7ede813bbbcdb0506ab2468e0e5e44c89ee5 wpt-pr: 45946

…ts for prelu, a=testonly Automatic update from web-platform-tests webnn: Enforce input data type constraints for prelu As specified in webmachinelearning/webnn#646. Besides, this CL also fixes the throwing TypeError and migrates the unit tests to WPT for prelu. Bug: 328567884, 327337526, 328026885 Change-Id: I2c6c0097b8d4fdc8c92bd66d21abd2cbda91e030 Cq-Include-Trybots: luci.chromium.try:win11-blink-rel Reviewed-on: https://chromium-review.googlesource.com/c/chromium/src/+/5495874 Reviewed-by: Austin Sullivan <[email protected]> Reviewed-by: ningxin hu <[email protected]> Commit-Queue: Lisha Guo <[email protected]> Cr-Commit-Position: refs/heads/main@{#1297344} -- wpt-commits: f409c99393325cfd014ff2002d84b6bb8f5afcbb wpt-pr: 45947

… averagePool2d and l2Pool2d, a=testonly Automatic update from web-platform-tests webnn: Enforce data type constraints for averagePool2d and l2Pool2d As specified in webmachinelearning/webnn#646 Bug: 328567884 Change-Id: If0ea256805b5cfebdf15e0727fbc91acab918bbb Reviewed-on: https://chromium-review.googlesource.com/c/chromium/src/+/5500424 Reviewed-by: Austin Sullivan <[email protected]> Auto-Submit: Shiyi Zou <[email protected]> Commit-Queue: Shiyi Zou <[email protected]> Reviewed-by: ningxin hu <[email protected]> Cr-Commit-Position: refs/heads/main@{#1297393} -- wpt-commits: 8420f5224b6c558c67a32052f1148eca46245003 wpt-pr: 45974

… abs, neg and relu, a=testonly Automatic update from web-platform-tests webnn: Enforce data type constraints for abs, neg and relu As specified in webmachinelearning/webnn#646 Bug: 328567884 Change-Id: Id377fcf423f72e2dd6d7851fa979fdfa8caeefdc Reviewed-on: https://chromium-review.googlesource.com/c/chromium/src/+/5517439 Commit-Queue: Shiyi Zou <[email protected]> Reviewed-by: Austin Sullivan <[email protected]> Reviewed-by: ningxin hu <[email protected]> Cr-Commit-Position: refs/heads/main@{#1297321} -- wpt-commits: 602dae0eda79af364575e1f99ad39967bdd7fbc9 wpt-pr: 46075

…ts for resample2d, a=testonly Automatic update from web-platform-tests webnn: Enforce input data type constraints for resample2d As specified in webmachinelearning/webnn#646 Bug: 328567884 Change-Id: Icea0e4a586aee33d7993d6e72393412b981d26ab Reviewed-on: https://chromium-review.googlesource.com/c/chromium/src/+/5495876 Reviewed-by: Austin Sullivan <[email protected]> Reviewed-by: ningxin hu <[email protected]> Commit-Queue: Lisha Guo <[email protected]> Cr-Commit-Position: refs/heads/main@{#1297332} -- wpt-commits: b5ab7ede813bbbcdb0506ab2468e0e5e44c89ee5 wpt-pr: 45946

…ts for prelu, a=testonly Automatic update from web-platform-tests webnn: Enforce input data type constraints for prelu As specified in webmachinelearning/webnn#646. Besides, this CL also fixes the throwing TypeError and migrates the unit tests to WPT for prelu. Bug: 328567884, 327337526, 328026885 Change-Id: I2c6c0097b8d4fdc8c92bd66d21abd2cbda91e030 Cq-Include-Trybots: luci.chromium.try:win11-blink-rel Reviewed-on: https://chromium-review.googlesource.com/c/chromium/src/+/5495874 Reviewed-by: Austin Sullivan <[email protected]> Reviewed-by: ningxin hu <[email protected]> Commit-Queue: Lisha Guo <[email protected]> Cr-Commit-Position: refs/heads/main@{#1297344} -- wpt-commits: f409c99393325cfd014ff2002d84b6bb8f5afcbb wpt-pr: 45947

… averagePool2d and l2Pool2d, a=testonly Automatic update from web-platform-tests webnn: Enforce data type constraints for averagePool2d and l2Pool2d As specified in webmachinelearning/webnn#646 Bug: 328567884 Change-Id: If0ea256805b5cfebdf15e0727fbc91acab918bbb Reviewed-on: https://chromium-review.googlesource.com/c/chromium/src/+/5500424 Reviewed-by: Austin Sullivan <[email protected]> Auto-Submit: Shiyi Zou <[email protected]> Commit-Queue: Shiyi Zou <[email protected]> Reviewed-by: ningxin hu <[email protected]> Cr-Commit-Position: refs/heads/main@{#1297393} -- wpt-commits: 8420f5224b6c558c67a32052f1148eca46245003 wpt-pr: 45974

As specified in webmachinelearning/webnn#646 Bug: 328567884 Change-Id: I5e600bdc791ecd4530408b65c15dd20611c211a3 Cq-Include-Trybots: luci.chromium.try:mac14-blink-rel,mac14.arm64-blink-rel

This is a reland of commit ef29e11bfb1f1f4eae0935fb58f0e4129620837b Original change's description: > webnn: Enforce input data type constraints for gather indices > > As specified in webmachinelearning/webnn#646 > > Bug: 328567884 > Change-Id: I33eba7e1def430b1cb94e3e7a4868e82c5bbd9a3 > Cq-Include-Trybots: luci.chromium.try:mac14-blink-rel,mac14.arm64-blink-rel > Reviewed-on: https://chromium-review.googlesource.com/c/chromium/src/+/5520313 > Reviewed-by: Austin Sullivan <[email protected]> > Commit-Queue: Lisha Guo <[email protected]> > Reviewed-by: ningxin hu <[email protected]> > Cr-Commit-Position: refs/heads/main@{#1298443} Bug: 328567884 Change-Id: Ide6a0e1bc6d3ebea7dce6bf37da4318e7b6e1bd1 Cq-Include-Trybots: luci.chromium.try:mac14-blink-rel,mac14.arm64-blink-rel

As specified in webmachinelearning/webnn#646 Bug: 328567884 Change-Id: I5e600bdc791ecd4530408b65c15dd20611c211a3 Cq-Include-Trybots: luci.chromium.try:mac14-blink-rel,mac14.arm64-blink-rel Reviewed-on: https://chromium-review.googlesource.com/c/chromium/src/+/5517811 Reviewed-by: Austin Sullivan <[email protected]> Reviewed-by: ningxin hu <[email protected]> Commit-Queue: Lisha Guo <[email protected]> Cr-Commit-Position: refs/heads/main@{#1303757}

…ts for gather indices, a=testonly Automatic update from web-platform-tests webnn: Enforce input data type constraints for gather indices As specified in webmachinelearning/webnn#646 Bug: 328567884 Change-Id: I33eba7e1def430b1cb94e3e7a4868e82c5bbd9a3 Cq-Include-Trybots: luci.chromium.try:mac14-blink-rel,mac14.arm64-blink-rel Reviewed-on: https://chromium-review.googlesource.com/c/chromium/src/+/5520313 Reviewed-by: Austin Sullivan <[email protected]> Commit-Queue: Lisha Guo <[email protected]> Reviewed-by: ningxin hu <[email protected]> Cr-Commit-Position: refs/heads/main@{#1298443} -- wpt-commits: d32b68ce2a00a378db571901573b2ded4d8ab349 wpt-pr: 46129

…onstraints for gather indices", a=testonly Automatic update from web-platform-tests Revert "webnn: Enforce input data type constraints for gather indices" This reverts commit ef29e11bfb1f1f4eae0935fb58f0e4129620837b. Reason for revert: LUCI Bisection has identified this change as the cause of a test failure. See the analysis: https://ci.chromium.org/ui/p/chromium/bisection/test-analysis/b/5728899092709376 Sample build with failed test: https://ci.chromium.org/b/8748452022241283393 Affected test(s): [ninja://services:services_unittests/WebNNGraphImplBackendTest.BuildAndComputeSingleOperatorGather](https://ci.chromium.org/ui/test/chromium/ninja:%2F%2Fservices:services_unittests%2FWebNNGraphImplBackendTest.BuildAndComputeSingleOperatorGather?q=VHash%3A3db3852ce13cc70a) If this is a false positive, please report it at http://b.corp.google.com/createIssue?component=1199205&description=Analysis%3A+https%3A%2F%2Fci.chromium.org%2Fui%2Fp%2Fchromium%2Fbisection%2Ftest-analysis%2Fb%2F5728899092709376&format=PLAIN&priority=P3&title=Wrongly+blamed+https%3A%2F%2Fchromium-review.googlesource.com%2Fc%2Fchromium%2Fsrc%2F%2B%2F5520313&type=BUG Original change's description: > webnn: Enforce input data type constraints for gather indices > > As specified in webmachinelearning/webnn#646 > > Bug: 328567884 > Change-Id: I33eba7e1def430b1cb94e3e7a4868e82c5bbd9a3 > Cq-Include-Trybots: luci.chromium.try:mac14-blink-rel,mac14.arm64-blink-rel > Reviewed-on: https://chromium-review.googlesource.com/c/chromium/src/+/5520313 > Reviewed-by: Austin Sullivan <[email protected]> > Commit-Queue: Lisha Guo <[email protected]> > Reviewed-by: ningxin hu <[email protected]> > Cr-Commit-Position: refs/heads/main@{#1298443} > Bug: 328567884 Change-Id: Iab874e915f5603861929cdf07ba4c09698b76503 No-Presubmit: true No-Tree-Checks: true No-Try: true Reviewed-on: https://chromium-review.googlesource.com/c/chromium/src/+/5528819 Reviewed-by: Austin Sullivan <[email protected]> Reviewed-by: Phillis Tang <[email protected]> Commit-Queue: Austin Sullivan <[email protected]> Cr-Commit-Position: refs/heads/main@{#1298718} -- wpt-commits: 61d40184e4b3d7aa0cca8c16cf3488e248c3ff04 wpt-pr: 46183

This is a reland of commit ef29e11bfb1f1f4eae0935fb58f0e4129620837b The failure of services_unittests/WebNNGraphImplBackendTest. BuildAndComputeSingleOperatorGather has been fixed by another CL: https://chromium-review.googlesource.com/c/chromium/src/+/5544945. This CL adds int32 dataType support check besides uint32 and int64 for gather indices compared with original CL. Original change's description: > webnn: Enforce input data type constraints for gather indices > > As specified in webmachinelearning/webnn#646 > > Bug: 328567884 > Change-Id: I33eba7e1def430b1cb94e3e7a4868e82c5bbd9a3 > Cq-Include-Trybots: luci.chromium.try:mac14-blink-rel,mac14.arm64-blink-rel > Reviewed-on: https://chromium-review.googlesource.com/c/chromium/src/+/5520313 > Reviewed-by: Austin Sullivan <[email protected]> > Commit-Queue: Lisha Guo <[email protected]> > Reviewed-by: ningxin hu <[email protected]> > Cr-Commit-Position: refs/heads/main@{#1298443} Bug: 328567884 Change-Id: Ide6a0e1bc6d3ebea7dce6bf37da4318e7b6e1bd1 Cq-Include-Trybots: luci.chromium.try:mac14-blink-rel,mac14.arm64-blink-rel,gpu-fyi-try-win11-qualcomm-rel-64

This is a reland of commit ef29e11bfb1f1f4eae0935fb58f0e4129620837b The failure of services_unittests/WebNNGraphImplBackendTest. BuildAndComputeSingleOperatorGather has been fixed by another CL: https://chromium-review.googlesource.com/c/chromium/src/+/5544945. This CL adds int32 dataType support check besides uint32 and int64 for gather indices compared with original CL. Original change's description: > webnn: Enforce input data type constraints for gather indices > > As specified in webmachinelearning/webnn#646 > > Bug: 328567884 > Change-Id: I33eba7e1def430b1cb94e3e7a4868e82c5bbd9a3 > Cq-Include-Trybots: luci.chromium.try:mac14-blink-rel,mac14.arm64-blink-rel > Reviewed-on: https://chromium-review.googlesource.com/c/chromium/src/+/5520313 > Reviewed-by: Austin Sullivan <[email protected]> > Commit-Queue: Lisha Guo <[email protected]> > Reviewed-by: ningxin hu <[email protected]> > Cr-Commit-Position: refs/heads/main@{#1298443} Bug: 328567884 Change-Id: Ide6a0e1bc6d3ebea7dce6bf37da4318e7b6e1bd1 Cq-Include-Trybots: luci.chromium.try:mac14-blink-rel,mac14.arm64-blink-rel,gpu-fyi-try-win11-qualcomm-rel-64 Reviewed-on: https://chromium-review.googlesource.com/c/chromium/src/+/5532907 Reviewed-by: ningxin hu <[email protected]> Commit-Queue: Lisha Guo <[email protected]> Reviewed-by: Austin Sullivan <[email protected]> Cr-Commit-Position: refs/heads/main@{#1304870}

…ts for gather indices, a=testonly Automatic update from web-platform-tests webnn: Enforce input data type constraints for gather indices As specified in webmachinelearning/webnn#646 Bug: 328567884 Change-Id: I33eba7e1def430b1cb94e3e7a4868e82c5bbd9a3 Cq-Include-Trybots: luci.chromium.try:mac14-blink-rel,mac14.arm64-blink-rel Reviewed-on: https://chromium-review.googlesource.com/c/chromium/src/+/5520313 Reviewed-by: Austin Sullivan <[email protected]> Commit-Queue: Lisha Guo <[email protected]> Reviewed-by: ningxin hu <[email protected]> Cr-Commit-Position: refs/heads/main@{#1298443} -- wpt-commits: d32b68ce2a00a378db571901573b2ded4d8ab349 wpt-pr: 46129

…onstraints for gather indices", a=testonly Automatic update from web-platform-tests Revert "webnn: Enforce input data type constraints for gather indices" This reverts commit ef29e11bfb1f1f4eae0935fb58f0e4129620837b. Reason for revert: LUCI Bisection has identified this change as the cause of a test failure. See the analysis: https://ci.chromium.org/ui/p/chromium/bisection/test-analysis/b/5728899092709376 Sample build with failed test: https://ci.chromium.org/b/8748452022241283393 Affected test(s): [ninja://services:services_unittests/WebNNGraphImplBackendTest.BuildAndComputeSingleOperatorGather](https://ci.chromium.org/ui/test/chromium/ninja:%2F%2Fservices:services_unittests%2FWebNNGraphImplBackendTest.BuildAndComputeSingleOperatorGather?q=VHash%3A3db3852ce13cc70a) If this is a false positive, please report it at http://b.corp.google.com/createIssue?component=1199205&description=Analysis%3A+https%3A%2F%2Fci.chromium.org%2Fui%2Fp%2Fchromium%2Fbisection%2Ftest-analysis%2Fb%2F5728899092709376&format=PLAIN&priority=P3&title=Wrongly+blamed+https%3A%2F%2Fchromium-review.googlesource.com%2Fc%2Fchromium%2Fsrc%2F%2B%2F5520313&type=BUG Original change's description: > webnn: Enforce input data type constraints for gather indices > > As specified in webmachinelearning/webnn#646 > > Bug: 328567884 > Change-Id: I33eba7e1def430b1cb94e3e7a4868e82c5bbd9a3 > Cq-Include-Trybots: luci.chromium.try:mac14-blink-rel,mac14.arm64-blink-rel > Reviewed-on: https://chromium-review.googlesource.com/c/chromium/src/+/5520313 > Reviewed-by: Austin Sullivan <[email protected]> > Commit-Queue: Lisha Guo <[email protected]> > Reviewed-by: ningxin hu <[email protected]> > Cr-Commit-Position: refs/heads/main@{#1298443} > Bug: 328567884 Change-Id: Iab874e915f5603861929cdf07ba4c09698b76503 No-Presubmit: true No-Tree-Checks: true No-Try: true Reviewed-on: https://chromium-review.googlesource.com/c/chromium/src/+/5528819 Reviewed-by: Austin Sullivan <[email protected]> Reviewed-by: Phillis Tang <[email protected]> Commit-Queue: Austin Sullivan <[email protected]> Cr-Commit-Position: refs/heads/main@{#1298718} -- wpt-commits: 61d40184e4b3d7aa0cca8c16cf3488e248c3ff04 wpt-pr: 46183

…ts for some reduce operators, a=testonly Automatic update from web-platform-tests webnn: Enforce input data type constraints for some reduce operators As specified in webmachinelearning/webnn#646 Bug: 328567884 Change-Id: I5e600bdc791ecd4530408b65c15dd20611c211a3 Cq-Include-Trybots: luci.chromium.try:mac14-blink-rel,mac14.arm64-blink-rel Reviewed-on: https://chromium-review.googlesource.com/c/chromium/src/+/5517811 Reviewed-by: Austin Sullivan <[email protected]> Reviewed-by: ningxin hu <[email protected]> Commit-Queue: Lisha Guo <[email protected]> Cr-Commit-Position: refs/heads/main@{#1303757} -- wpt-commits: 0fa5e3bc5d097d84c074bc138271d8999277a50c wpt-pr: 46180

…onstraints for gather indices", a=testonly Automatic update from web-platform-tests Reland "webnn: Enforce input data type constraints for gather indices" This is a reland of commit ef29e11bfb1f1f4eae0935fb58f0e4129620837b The failure of services_unittests/WebNNGraphImplBackendTest. BuildAndComputeSingleOperatorGather has been fixed by another CL: https://chromium-review.googlesource.com/c/chromium/src/+/5544945. This CL adds int32 dataType support check besides uint32 and int64 for gather indices compared with original CL. Original change's description: > webnn: Enforce input data type constraints for gather indices > > As specified in webmachinelearning/webnn#646 > > Bug: 328567884 > Change-Id: I33eba7e1def430b1cb94e3e7a4868e82c5bbd9a3 > Cq-Include-Trybots: luci.chromium.try:mac14-blink-rel,mac14.arm64-blink-rel > Reviewed-on: https://chromium-review.googlesource.com/c/chromium/src/+/5520313 > Reviewed-by: Austin Sullivan <[email protected]> > Commit-Queue: Lisha Guo <[email protected]> > Reviewed-by: ningxin hu <[email protected]> > Cr-Commit-Position: refs/heads/main@{#1298443} Bug: 328567884 Change-Id: Ide6a0e1bc6d3ebea7dce6bf37da4318e7b6e1bd1 Cq-Include-Trybots: luci.chromium.try:mac14-blink-rel,mac14.arm64-blink-rel,gpu-fyi-try-win11-qualcomm-rel-64 Reviewed-on: https://chromium-review.googlesource.com/c/chromium/src/+/5532907 Reviewed-by: ningxin hu <[email protected]> Commit-Queue: Lisha Guo <[email protected]> Reviewed-by: Austin Sullivan <[email protected]> Cr-Commit-Position: refs/heads/main@{#1304870} -- wpt-commits: dde81ead14885761dbf9b0671bcd24a483b12b24 wpt-pr: 46217

…ts for conv2d, a=testonly Automatic update from web-platform-tests webnn: Enforce input data type constraints for conv2d As specified in webmachinelearning/webnn#646 Bug: 328567884 Change-Id: I620653f574303e6f5b18d3b540b647f07e0cf64c Reviewed-on: https://chromium-review.googlesource.com/c/chromium/src/+/5492307 Reviewed-by: ningxin hu <[email protected]> Commit-Queue: Austin Sullivan <[email protected]> Cr-Commit-Position: refs/heads/main@{#1293407} -- wpt-commits: 839bdcfea290fd869732772d922ed910efbf052c wpt-pr: 45922

…ts for gemm and matmul, a=testonly Automatic update from web-platform-tests webnn: Enforce input data type constraints for gemm and matmul As specified in webmachinelearning/webnn#646 Bug: 328567884 Change-Id: Ia55a214e7ad281ec3c8911e9116f388fac209d05 Reviewed-on: https://chromium-review.googlesource.com/c/chromium/src/+/5495161 Auto-Submit: Shiyi Zou <[email protected]> Commit-Queue: Shiyi Zou <[email protected]> Reviewed-by: Austin Sullivan <[email protected]> Reviewed-by: ningxin hu <[email protected]> Cr-Commit-Position: refs/heads/main@{#1294123} -- wpt-commits: 0664d8153b085b9b3a94ecc52bc5b04d37e7d6fa wpt-pr: 45945

Specify the operand data type constraints of operations

a6afe10

Introduce constraints for input operands, either directly (e.g. input's dataType can only be "float32" or "float16") or indirectly (e.g. weight's dataType must be the same as input's). Fixes webmachinelearning#283

huningxin approved these changes Apr 19, 2024

View reviewed changes

index.bs Outdated Show resolved Hide resolved

huningxin requested a review from fdwr April 19, 2024 07:37

inexorabletash added 2 commits April 19, 2024 09:18

Merge remote-tracking branch 'origin/main' into data-type-constraints

b10bb85

gruCell: bundle hiddenState in with other type validations

1d91d80

a-sully mentioned this pull request Apr 19, 2024

Introduce MLNumber for specifying numeric inputs of any type #647

Merged

huningxin approved these changes Apr 24, 2024

View reviewed changes

Merge remote-tracking branch 'origin/main' into data-type-constraints

38f7755

fdwr reviewed Apr 24, 2024

View reviewed changes

index.bs Outdated Show resolved Hide resolved

Identity should accept all types

16e424c

huningxin approved these changes Apr 24, 2024

View reviewed changes

Merge remote-tracking branch 'origin/main' into data-type-constraints

64f9bbc

inexorabletash added 2 commits April 25, 2024 09:14

Merge remote-tracking branch 'origin/main' into data-type-constraints

26ea2c6

Add reduceMean restriction

bd9b0cf

huningxin approved these changes Apr 26, 2024

View reviewed changes

Update gemm to check c data type too

ad239bc

fdwr approved these changes Apr 26, 2024

View reviewed changes

chromium-wpt-export-bot pushed a commit to web-platform-tests/wpt that referenced this pull request Apr 26, 2024

webnn: Enforce input data type constraints for conv2d

0a52960

As specified in webmachinelearning/webnn#646 Bug: 328567884 Change-Id: I620653f574303e6f5b18d3b540b647f07e0cf64c

chromium-wpt-export-bot mentioned this pull request Apr 25, 2024

webnn: Enforce input data type constraints for conv2d web-platform-tests/wpt#45922

Merged

chromium-wpt-export-bot pushed a commit to web-platform-tests/wpt that referenced this pull request Apr 26, 2024

webnn: Enforce input data type constraints for conv2d

79baf4c

As specified in webmachinelearning/webnn#646 Bug: 328567884 Change-Id: I620653f574303e6f5b18d3b540b647f07e0cf64c

inexorabletash mentioned this pull request Apr 26, 2024

Make operand data type and rank validation table-driven #657

Draft

fdwr merged commit b5677b6 into webmachinelearning:main Apr 26, 2024
2 checks passed

github-actions bot added a commit that referenced this pull request Apr 26, 2024

Specify the operand data type constraints of operations (#646)

da4aadc

SHA: b5677b6 Reason: push, by fdwr Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>

chromium-wpt-export-bot mentioned this pull request Apr 28, 2024

webnn: Enforce input data type constraints for gemm and matmul web-platform-tests/wpt#45945

Merged

chromium-wpt-export-bot mentioned this pull request May 11, 2024

Reland "webnn: Enforce input data type constraints for gather indices" web-platform-tests/wpt#46217

Merged

huningxin mentioned this pull request Jun 11, 2024

Bugfix: Add missing 64-bit integers support for some reduction operators #695

Open

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Specify the operand data type constraints of operations #646

Specify the operand data type constraints of operations #646

inexorabletash commented Apr 18, 2024 •

edited by pr-preview bot

Loading

huningxin left a comment

huningxin left a comment

huningxin left a comment

inexorabletash commented Apr 25, 2024

inexorabletash commented Apr 25, 2024

huningxin left a comment

fdwr left a comment

fdwr Apr 26, 2024

inexorabletash Apr 26, 2024 •

edited

Loading

fdwr Apr 26, 2024

fdwr Apr 26, 2024

inexorabletash Apr 26, 2024 •

edited

Loading

inexorabletash Apr 26, 2024 •

edited

Loading

inexorabletash Apr 26, 2024

fdwr Apr 26, 2024 •

edited

Loading

Specify the operand data type constraints of operations #646

Specify the operand data type constraints of operations #646

Conversation

inexorabletash commented Apr 18, 2024 • edited by pr-preview bot Loading

huningxin left a comment

Choose a reason for hiding this comment

huningxin left a comment

Choose a reason for hiding this comment

huningxin left a comment

Choose a reason for hiding this comment

inexorabletash commented Apr 25, 2024

inexorabletash commented Apr 25, 2024

huningxin left a comment

Choose a reason for hiding this comment

fdwr left a comment

Choose a reason for hiding this comment

fdwr Apr 26, 2024

Choose a reason for hiding this comment

inexorabletash Apr 26, 2024 • edited Loading

Choose a reason for hiding this comment

fdwr Apr 26, 2024

Choose a reason for hiding this comment

fdwr Apr 26, 2024

Choose a reason for hiding this comment

inexorabletash Apr 26, 2024 • edited Loading

Choose a reason for hiding this comment

inexorabletash Apr 26, 2024 • edited Loading

Choose a reason for hiding this comment

inexorabletash Apr 26, 2024

Choose a reason for hiding this comment

fdwr Apr 26, 2024 • edited Loading

Choose a reason for hiding this comment

inexorabletash commented Apr 18, 2024 •

edited by pr-preview bot

Loading

inexorabletash Apr 26, 2024 •

edited

Loading

inexorabletash Apr 26, 2024 •

edited

Loading

inexorabletash Apr 26, 2024 •

edited

Loading

fdwr Apr 26, 2024 •

edited

Loading