### Description
This PR is a preview of cherry-picks for ort-web to `rel-1.17.3` based
on `rel-1.17.2`.
<details>
<summary>Changes of ort-web to cherry-pick</summary>
The following commits are from main branch.
`o` stands for pick, and `x` stands for skip.
```
o 2e0a388c36 [js/webgpu] Add HardSigmoid support (#19215)
o d226e40856 [js/webgpu] set query type in onRunStart (#19202)
o 61610ff986 [js/webgpu] Add FusedConv clip test case (#18900)
o a33b5bd1fa [JS/WebGPU] Added Uniforms to SkipLayerNorm. (#18788)
o 591f90c0b9 [js/webgpu] Fix issue of timestamp query (#19258)
o 7252c6e747 [WebNN EP] Support WebNN async API with Asyncify (#19145)
o 5b06505073 [js/webgpu] Fix Tanh explosion (#19201)
o 656ca66186 [js/webgpu] Support uniforms for conv, conv transpose, conv grouped (#18753)
o a3f0e2422b [js/webgpu] Support f16 uniform (#19098)
o 9e69606360 fix f16 for attention, enable slice and flatten for more types (#19262)
o 624b4e2063 [js/webgpu] Remove enableShapesUniforms (#19279)
o 90883a366a [js/webgpu] Add hardSigmoid activation for fusedConv (#19233)
o 85cef0af8c [js/webgpu] Support capture and replay for jsep (#18989)
o d73131cf0f [js/webgpu] Use DataType as uniform cpu type (#19281)
o dd1f6ccc45 [js/webgpu] resolve codescan alert (#19343)
o 3a2ab1963a [js/webgpu] Refactor createTensorShapeVariables (#18883)
o efc17e79de [js/webgpu] Fix the undefined push error (#19366)
x 50806a7dd5 [js/web] support external data in npm test (#19377)
o ccbe264a39 [js/webgpu] Add LeakyRelu activation for fusedConv (#19369)
o 5ff27ef02a [js/webgpu] support customop FastGelu (#19392)
x 03be65e064 [js/web] fix types exports in package.json (#19458)
o 06269a3952 [js/webgpu] allow uint8 tensors for webgpu (#19545)
o dfeda9019c [JS/WebGPU] Add MatMulNBits (#19446)
o 1b48054e1b [js/webgpu] Create Split indices helpers by rank, not by shape (#19554)
o 3fe2c137ee [js] small fix to workaround formatter (#19400)
x 70567a4b3a [js/web] use ApiTensor insteadof onnxjs Tensor in TensorResultValidator (#19358)
o 6e04e36e3f [js/common] upgrade tsc in common from 4.9.5 to 5.2.2 (#19317)
o 58f4921686 [js] changes to allow Float16Array if any polyfill is available (#19305)
o 57d6819212 [js/web] Fix fused-conv is not included in npm test (#19581)
o ebd220b073 Misspelling in README.md (#19433)
o 38c3432393 Bump ip from 1.1.8 to 1.1.9 in /js/react_native (#19582)
o fe82fccf1a [js/webgpu] Fix Conv2DTransposeMatMul f16 compilation failure (#19596)
o 76a2a487a1 Bump ip from 1.1.8 to 1.1.9 in /js/react_native/e2e (#19583)
o 29b1106033 [node] Switch to setImmediate to avoid starving the Node.js event loop (#19610)
o ae3d73c981 [JS/WebGPU] Fix Split and Where to handle corner cases. (#19613)
o aec2389ad0 [js/webgpu] allows a ProgramInfo's RunData to use zero sized output (#19614)
o bb43a0f133 [js/webgpu] minor fixes to make tinyllama work (#19564)
o 0edb035808 [js/web] fix suite test list for zero sized tensor (#19638)
o 3cb81cdde2 [js/common] move 'env.wasm.trace' to 'env.trace' (#19617)
o e30618d055 [js/webgpu] use Headless for webgpu test by default (#19702)
o f06164ef8b [js/web] transfer input buffer back to caller thread (#19677)
x a788514027 [js/web] dump debug logs for karma for diagnose purpose (#19785)
o 24b72d2613 [JS/WebGPU] Preserve zero size input tensor dims. (#19737)
o 4538d31a8b [js/webgpu] expose a few properties in WebGPU API (#19857)
o 53de2d8cb0 [js/webgpu] Enable GroupedConvVectorize path (#19791)
o ed250b88c3 [JS/WebGPU] Optimize MatMulNBits (#19852)
x e771a763c3 [js/test] align web test runner flags with ort.env (#19790)
o 79e50aeef3 [js/web] rewrite backend resolve to allow multiple EPs (#19735)
o acb0df2280Fix#19931 broken Get Started link of "ONNX Runtime JavaScript API" page (#19932)
o b29849a287 [js/common] fix typedoc warnings (#19933)
o afdab62f53 Bump follow-redirects from 1.15.4 to 1.15.6 in /js/web (#19949)
o 28ad6c3955 Bump follow-redirects from 1.15.4 to 1.15.6 in /js/node (#19951)
o 7e0d424934 accumulate in fp32 for Reduce* (#19868)
o 4c6a6a37f7 [js/webgpu] Fix NAN caused by un-initialized buffer in instance-norm (#19387)
o 01c7aaf6aa [js/webgpu] allow setting env.webgpu.adapter (#19940)
o c45cff60cf [js/webgpu] fix maxpool / fp16 (#19981)
```
</details>
<details>
<summary>Cherry-pick commandlines</summary>
```sh
git cherry-pick 2e0a388c36
git cherry-pick d226e40856
git cherry-pick 61610ff986
git cherry-pick a33b5bd1fa
git cherry-pick 591f90c0b9
git cherry-pick 7252c6e747
git cherry-pick 5b06505073
git cherry-pick 656ca66186
git cherry-pick a3f0e2422b
git cherry-pick 9e69606360
git cherry-pick 624b4e2063
git cherry-pick 90883a366a
git cherry-pick 85cef0af8c #<<<<< Note: conflicts
git cherry-pick d73131cf0f
git cherry-pick dd1f6ccc45
git cherry-pick 3a2ab1963a
git cherry-pick efc17e79de
git cherry-pick ccbe264a39
git cherry-pick 5ff27ef02a
git cherry-pick 06269a3952
git cherry-pick dfeda9019c
git cherry-pick 1b48054e1b
git cherry-pick 3fe2c137ee
git cherry-pick 6e04e36e3f
git cherry-pick 58f4921686
git cherry-pick 57d6819212
git cherry-pick ebd220b073
git cherry-pick 38c3432393
git cherry-pick fe82fccf1a
git cherry-pick 76a2a487a1
git cherry-pick 29b1106033
git cherry-pick ae3d73c981
git cherry-pick aec2389ad0
git cherry-pick bb43a0f133
git cherry-pick 0edb035808
git cherry-pick 3cb81cdde2
git cherry-pick e30618d055
git cherry-pick f06164ef8b
git cherry-pick 24b72d2613
git cherry-pick 4538d31a8b
git cherry-pick 53de2d8cb0
git cherry-pick ed250b88c3
git cherry-pick 79e50aeef3
git cherry-pick acb0df2280
git cherry-pick b29849a287
git cherry-pick afdab62f53
git cherry-pick 28ad6c3955
git cherry-pick 7e0d424934
git cherry-pick 4c6a6a37f7
git cherry-pick 01c7aaf6aa
git cherry-pick c45cff60cf
```
</details>
<details>
<summary>Cherry-pick conflicts</summary>
- 85cef0af8c#18989
this change is for enabling graph capture feature for JSEP, and it is
done after ROCM EP enabled graph capture feature. However, the ROCM EP
graph capture feature is not cherry-picked in rel-1.17.2.
</details>
---------
Signed-off-by: dependabot[bot] <support@github.com>
Co-authored-by: Jiajia Qin <jiajia.qin@intel.com>
Co-authored-by: Xu Xing <xing.xu@intel.com>
Co-authored-by: satyajandhyala <satya.k.jandhyala@gmail.com>
Co-authored-by: Yang Gu <yang.gu@intel.com>
Co-authored-by: Wanming Lin <wanming.lin@intel.com>
Co-authored-by: Jiajie Hu <jiajie.hu@intel.com>
Co-authored-by: Guenther Schmuelling <guschmue@microsoft.com>
Co-authored-by: Matttttt <18152455+martholomew@users.noreply.github.com>
Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com>
Co-authored-by: Segev Finer <segev208@gmail.com>
Co-authored-by: Belem Zhang <belem.zhang@intel.com>
### Description
This PR provides a vectorized algorithm for NHWC GroupedConv to improve
performance.
The aggregate time of GroupedConv in mobilenetv2-12 becomes ~1ms from
~4ms on Intel Alder Lake machine. About 20% improvement for the whole
model.
### Description
Also update the op test suite.
### Motivation and Context
Previously the *total* size in case `Expand - last dim is not divisible
by 4` was a multiple of 4, even though the *last dimension* was not, so
the bug has never been caught.
### Description
<!-- Describe your changes. -->
### Motivation and Context
<!-- - Why is this change required? What problem does it solve?
- If it fixes an open issue, please link to the issue here. -->
### Description
Add trilinear interpolation to Resize and changed activation_params attribute as optional for FuseConv.
### Motivation and Context
<!-- - Why is this change required? What problem does it solve?
- If it fixes an open issue, please link to the issue here. -->
### Description
Add uinforms to Einsum
### Motivation and Context
<!-- - Why is this change required? What problem does it solve?
- If it fixes an open issue, please link to the issue here. -->
Improve performance.
### Description
Currently, all conv2dMatmul with inChannels = 3 and outChannels % 4 = 0
will report compilation errors. Models, which include this kind of shape
will be impacted, like mobilenetv2-12, resnet50 .
The errors is introduced by #18452https://github.com/microsoft/onnxruntime/pull/18452/files#diff-8b24ea43aa11b1346c0c9e327f9bce6b37a93bd8f2bf8a6392b2b263972b7ea2R200,
which accidentally pass `components` to `x`. But `x`'s components is
`innerElementSize` not `components `. And when `innerElementSize` is 3,
we should use `1` in current design.
### Description
This PR adds `BatchNormalization` with `float` support.
Some Todos:
1. all inputs don't have same data type. For example, x/y is float16,
but bias/scale is float32 or double.
2. training mode support.
We see many models are using `BatchNormalization` ops. However, due to
the missing in jsep, all of them run on cpu, which result very poor
performance. With this PR's support, densenet-9 model becomes 20.29 ms
from 250.69 ms.
### Description
<!-- Describe your changes. -->
Added Uniforms to Expand operator kernel
### Motivation and Context
<!-- - Why is this change required? What problem does it solve?
- If it fixes an open issue, please link to the issue here. -->
Improve performance
### Description
This is a narrow implementation of Attention/MultiHeadAttention as it
does not support:
a. inputs 5-7 for MHA
b. packed QKV/KV
c. past/present
d. attention mask
But it works well for StableDiffusion and can be extended later. It
reduces VRAM usage as it combines many ops into few
I've updated demo here https://islamov.ai/stable-diffusion-webgpu/ it
takes ~13sec for 1 image with 20 steps on RTX3090Ti and about 25s on M1
Pro
VRAM usage is about 8gb if you don't use img2img
Going to focus on SDXL now
---------
Co-authored-by: Guenther Schmuelling <guschmue@microsoft.com>
Co-authored-by: Yulong Wang <7679871+fs-eire@users.noreply.github.com>
### Description
Support uniforms in Slice op
### Motivation and Context
<!-- - Why is this change required? What problem does it solve?
- If it fixes an open issue, please link to the issue here. -->
Improve ferformance
### Description
Added FusedConv and FusedConvTranspose
### Motivation and Context
<!-- - Why is this change required? What problem does it solve?
- If it fixes an open issue, please link to the issue here. -->
Improve performance
### Description
<!-- Describe your changes. -->
Currently, the uniform support has bugs when dims rank is larger than 4.
See https://github.com/microsoft/onnxruntime/issues/17860 item 1.
So this PR only enables shapes uniforms when shape rank is <= 4 for
transpose. Otherwise, below compilation errors are thrown:
```
1 error(s) generated while compiling the shader:
:3:50 error: uniform storage requires that array elements are aligned to 16 bytes, but array element of type 'u32' has a stride of 4 bytes. Consider using a vector or struct as the element type instead.
struct Uniforms { output_size:u32, a_shape:array<u32, 5>, a_strides:array<u32, 5>, output_shape:array<u32, 5>, output_strides:array<u32, 5> };
^^^^^^^^^^^^^
:3:7 note: see layout of struct:
/* align(4) size(84) */ struct Uniforms {
/* offset( 0) align(4) size( 4) */ output_size : u32;
/* offset( 4) align(4) size(20) */ a_shape : array<u32, 5>;
/* offset(24) align(4) size(20) */ a_strides : array<u32, 5>;
/* offset(44) align(4) size(20) */ output_shape : array<u32, 5>;
/* offset(64) align(4) size(20) */ output_strides : array<u32, 5>;
/* */ };
struct Uniforms { output_size:u32, a_shape:array<u32, 5>, a_strides:array<u32, 5>, output_shape:array<u32, 5>, output_strides:array<u32, 5> };
^^^^^^
:4:42 note: 'Uniforms' used in address space 'uniform' here
@group(0) @binding(2) var<uniform> uniforms: Uniforms;
^^^^^^^^
```
Supported type: float. int32_t, uint32_t, bool.
Case where_broadcast.jsonc is not enabled due to
https://github.com/microsoft/onnxruntime/issues/17405.
### Description
<!-- Describe your changes. -->
### Motivation and Context
<!-- - Why is this change required? What problem does it solve?
- If it fixes an open issue, please link to the issue here. -->
---------
Co-authored-by: Yulong Wang <7679871+fs-eire@users.noreply.github.com>
### Description
Two contrib kernels that supposed to speed-up StableDiffusion according
to this doc
https://github.com/microsoft/onnxruntime/blob/main/onnxruntime/python/tools/transformers/models/stable_diffusion/README.md
However, there is no noticable effect in speed or memory consumption. So
i guess the only way to make it faster is to implement
MultiHeadAttention but i'm not capable of doing that right now. So i'll
focus on existing PRs and finding the JSEP kernel that produces
incorrect results. It should be one of the old ones (i suspect Conv or
ConvTranspose), as SD was not generating images correctly on webgpu
since i started working on it. I hoped someone else would fix that by
the time i finish with kernels/optimizations 😅
---------
Co-authored-by: Guenther Schmuelling <guschmue@microsoft.com>
Co-authored-by: Yulong Wang <7679871+fs-eire@users.noreply.github.com>
### Description
Add ConvTranspose implementation using MatMul to increase perf.
### Motivation and Context
<!-- - Why is this change required? What problem does it solve?
- If it fixes an open issue, please link to the issue here. -->
### Description
<!-- Describe your changes. -->
In previous implementation, there are two loops to iterate H * W
elements to calculate the `mean` and `squaredNorm` value in one thread,
meanwhile it outputs H * W elements in one thread. That results it's
very very slow when H * W is a large value. And usually, H * W does be a
large value in a model. For example, in the `candy-8` model, the shapes
of [H, W] are [224,224], [112,112], [56,56] for `InstanceNormalization`
op. And in my ADL, `[1,224,224,32]` consumes 17 ms. See below:
```
[profiling] kernel "23848328|[InstanceNormalization] 23848328" input[0]: [1,224,224,32] | float32, input[1]: [32] | float32, input[2]: [32] | float32, output[0]: [1,224,224,32] | float32, execution time: 17007914 ns
```
In this PR, it uses workgroup memory to optimize the original algorithm.
The advantage is that it can parallelly utilize the 64 (workgroupSize)
threads in one workgroup to calculate `mean` and `squaredNorm` value.
Meanwhile, it only outputs `H * W / workgroupSize` outputs for one
thread, which greatly reduces the overhead for one thread. With this
optimization, `[1,224,224,32]` becomes 3 ms and the main overhead is the
extra two `transpose`. The `createInstanceNormProgramInfo` only needs
`0.64` ms. See below:
```
[profiling] kernel "23003600|[InstanceNormalization] 23003600" input[0]: [1,224,224,32] | float32, output[0]: [1,32,224,224] | float32, execution time: 1543792 ns
program-manager.ts:115
[profiling] kernel "23003600|[InstanceNormalization] 23003600" input[0]: [1,32,224,224] | float32, input[1]: [32] | float32, input[2]: [32] | float32, output[0]: [1,32,224,224] | float32, execution time: 642652 ns
program-manager.ts:115
[profiling] kernel "23003600|[InstanceNormalization] 23003600" input[0]: [1,32,224,224] | float32, output[0]: [1,224,224,32] | float32, execution time: 991608 ns
```
This PR currently only applies the new algorithm to NCHW format. For
NHWC format, one way is to transpose the input so that it can use the
new algorithm. But the disadvantage is that 2 extra transpose are added.
@dakenf also gives another way to optimize NHWC. Details see
[here](d45a96616d/js/web/lib/wasm/jsep/webgpu/ops/instance-norm.ts).
I checked @dakenf's method. The perf is similar with transpose +
optimized NCHW. But on different GPUs, one is a little better than
another or vice versa. So I prefer this PR only does the NCHW part.
@dakenf can submit his optimization on NHWC.
### Description
<!-- Describe your changes. -->
### Motivation and Context
<!-- - Why is this change required? What problem does it solve?
- If it fixes an open issue, please link to the issue here. -->
### Description
Added Einsum operator support to JSEP.
### Motivation and Context
<!-- - Why is this change required? What problem does it solve?
- If it fixes an open issue, please link to the issue here. -->
### Description
Include Support for neg.int32
### Motivation and Context
<!-- - Why is this change required? What problem does it solve?
- If it fixes an open issue, please link to the issue here. -->
### Description
Commit fffefb1c22 (#16969) optimized
matmul and also fixes broadcasting. So #17191 is no longer needed.
However, the newly added operator test file from the PR by @dakenf is
helpful so pick and add it to enhance the tests.
### Description
<!-- Describe your changes. -->
### Motivation and Context
<!-- - Why is this change required? What problem does it solve?
- If it fixes an open issue, please link to the issue here. -->
### Description
Changes in this PR:
1) use the optimized version `makeMatMulPacked[Vec4]Source` to support
matmul.
2) enable the conv2dByMatMul path.
3) support broadcast
4) use IndicesHelper.
MatMul with M = 512, K = 512, N = 512 becomes 2ms from 15ms when
enabling profilingMode on my ADL.
### Description
Fix JSEP ConvTranspose shader code errors.
### Motivation and Context
<!-- - Why is this change required? What problem does it solve?
- If it fixes an open issue, please link to the issue here. -->
### Description
Enable typed binary and support int32 type for binary.
Co-authored-by: Xing Xu <xing.xu@intel.com>
---------
Co-authored-by: Xing Xu <xing.xu@intel.com>
### Description
Add SkipLayerNormalization operator to JSEP.
### Motivation and Context
<!-- - Why is this change required? What problem does it solve?
- If it fixes an open issue, please link to the issue here. -->
### Description
Fix a typo. LayerNormalization takes 2 or 3 inputs. The third input,
bias, is optional.
### Motivation and Context
<!-- - Why is this change required? What problem does it solve?
- If it fixes an open issue, please link to the issue here. -->
### Description
update op test schema.
This changes fixes several problems for operator tests for web:
- `opsets` -> `opset`: an operator uses exactly one opset instead of
multiple
- `condition` -> `platformCondition`: make it less confusing
- `inputShapeDefinitions`: allows to test ORT behaviors when it get
no/partial/full shape info.
Added a JSON schema file and also an example file
### Description
Added Gelu operator to JSEP
### Motivation and Context
<!-- - Why is this change required? What problem does it solve?
- If it fixes an open issue, please link to the issue here. -->
* migrated changes to support running super resolution model using ortweb
* reverted benchmarking tool related changes which will be in a separate pr
* added kernel tests to op and node tests
* minor change to the order of variables
* added one more unit test for packed matmul
