Enhance symbolic onnx parsing via sym_eval by shivadbhavsar · Pull Request #5148 · ROCm/AMDMIGraphX

shivadbhavsar · 2026-08-18T05:48:22Z

No description provided.

Bring the PR #5112-based parser prerequisite branch onto the current development baseline before applying its remaining changes. Co-authored-by: Cursor <cursoragent@cursor.com>

Keep the merged resolver regression compatible with the dynamic-slice interface introduced by PR #5112. Co-authored-by: Cursor <cursoragent@cursor.com>

Track exact integral shape values through parser operations so dynamic consumers retain symbolic output relationships without changing runtime dataflow. Co-authored-by: Cursor <cursoragent@cursor.com>

Keep the symbolic-value change focused by restoring existing resolver diagnostics and simplifying the signed Gather size declaration. Co-authored-by: Cursor <cursoragent@cursor.com>

Copilot

Pull request overview

This PR extends MIGraphX’s symbolic shape machinery by introducing symbolic tensor evaluation (sym_eval / symbolic_compute) and using it in ONNX parsing and attribute normalization so shape-dependent subgraphs can preserve symbolic output shapes instead of falling back to range-dynamic shapes.

Changes:

Add sym_argument + operation::symbolic_compute() and instruction::sym_eval() to enable symbolic evaluation of int64/bool “metadata tensors” (e.g., shapes, indices).
Enhance ONNX parsers (e.g., Slice/Reshape/Expand/Range/Trilu/Where/Shape) to leverage sym_eval for improved symbolic output-shape propagation.
Add the new dyn_slice operator and symbolic attribute normalization, with broad test coverage across ref/gpu/onnx.

Reviewed changes

Copilot reviewed 68 out of 71 changed files in this pull request and generated 2 comments.

Show a summary per file

File	Description
tools/include/operation.hpp	Add `sym_argument` include + type-erased `symbolic_compute` plumbing in generator template.
test/sym.cpp	Add rewrite + interval/fixed-value/provable-equality tests for symbolic reasoning improvements.
test/sym_test.cpp	Extend sym serialization tests (bare numbers → literals) and reformat a few cases.
test/sym_eval_test.cpp	New unit tests for `instruction::sym_eval()` and symbolic compute fallback behavior.
test/ref/slice.cpp	Add symbolic slice coverage and update dynamic slicing test to use `dyn_slice`.
test/ref/dynamic_range.cpp	Add symbolic-output coverage for `dynamic_range`.
test/ref/dyn_slice.cpp	New reference backend tests for `dyn_slice` semantics and symbolic shapes.
test/onnx/verify/trilu_dynamic_test.cpp	New ONNX verify tests for symbolic `Trilu` output shapes.
test/onnx/verify/symbolic_shape_values_test.cpp	New ONNX verify test ensuring symbolic shape metadata is preserved and evaluated at runtime.
test/onnx/verify/symbolic_shape_arithmetic_test.cpp	New ONNX verify test for symbolic arithmetic propagation through shape subgraphs.
test/onnx/parse/symbolic_reshape_markers_test.cpp	New parse test to assert specific marker pattern for symbolic reshape handling.
test/onnx/gen_onnx.py	Add ONNX model generators for new symbolic shape tests.
test/normalize_ops_test.cpp	Add tests for symbolic normalization behavior (notably `dyn_slice` bounds).
test/normalize_attributes_test.cpp	New tests that characterize `normalize_attributes` integer normalization surface.
test/gpu/eval_expr_from_shape_lowering.cpp	New GPU lowering tests for keeping slice metadata on host and copying results when needed.
test/gpu/dyn_slice_lowering.cpp	New GPU lowering tests for `dyn_slice` runtime metadata movement/synchronization.
src/targets/gpu/lowering.cpp	Extend lowering to support `dyn_slice` and host-only `eval_expr_from_shape` metadata usage.
src/sym.cpp	Add `interval::contains`, `fixed_value`, `provable_equal`, folding helpers, and bare-number expr deserialization.
src/onnx/parse_where.cpp	Simplify `Where` parsing and rely on `add_common_op` for broadcasting.
src/onnx/parse_trilu.cpp	Implement symbolic-path `Trilu` parsing using dynamic-range + broadcasted mask construction.
src/onnx/parse_squeeze.cpp	Update copyright year.
src/onnx/parse_slice.cpp	Convert certain symbolic Slice cases into `dyn_slice` using `sym_eval` of bounds.
src/onnx/parse_shape.cpp	Refactor start/end normalization into a helper and tighten types.
src/onnx/parse_reshape.cpp	Use `sym_eval` to resolve reshape dims symbolically when possible (via `eval_expr_from_shape`).
src/onnx/parse_range.cpp	When possible, convert symbolic Range into `dynamic_range` with symbolic output dimension.
src/onnx/parse_generic_op.cpp	Add includes needed for new generic parsing behavior.
src/onnx/parse_expand.cpp	Use `sym_eval` to compute symbolic broadcast output dims and pass them to `broadcast_with_dims`.
src/onnx/parse_constant_of_shape.cpp	Use `sym_eval` to allocate with a symbolic output shape when possible.
src/onnx/parse_compare_op.cpp	Update copyright year.
src/onnx/parse_cast.cpp	Minor refactor + include fix.
src/onnx/include/migraphx/onnx/op_parser.hpp	Update copyright year; remove unused include.
src/onnx/include/migraphx/onnx/onnx_parser.hpp	Add missing includes (`optional`, `limits`, `unordered_set`) used by parser implementation.
src/normalize_attributes.cpp	Add symbolic attribute normalization for `sym::expr`-backed attributes and refactor integer path.
src/instruction.cpp	Implement `instruction::sym_eval()` with memoization and `symbolic_compute` fallback.
src/include/migraphx/tensor_view.hpp	Fix bounds assertions and make `to_vector()` usable for `const T` views.
src/include/migraphx/sym.hpp	Expose `interval::contains`, `provable_equal`, `fixed_value`, and `fold_min/fold_max`.
src/include/migraphx/sym_argument.hpp	New public header defining `sym_argument` and helpers for symbolic tensor evaluation.
src/include/migraphx/operators.hpp	Register `dyn_slice` operator header.
src/include/migraphx/operation.hpp	Add `symbolic_compute` to the public `operation` interface and type-erased dispatch.
src/include/migraphx/op/where.hpp	Add `where::symbolic_compute` to enable symbolic evaluation where condition is fixed.
src/include/migraphx/op/unsqueeze.hpp	Add `unsqueeze::symbolic_compute` pass-through behavior.
src/include/migraphx/op/unary.hpp	Add optional symbolic compute path to unary ops via CRTP flag.
src/include/migraphx/op/sub.hpp	Enable symbolic compute for `sub`.
src/include/migraphx/op/squeeze.hpp	Add `squeeze::symbolic_compute` pass-through behavior.
src/include/migraphx/op/slice.hpp	Clarify symbolic-shape constraints and add `slice::symbolic_compute` for 1D symbolic tensors.
src/include/migraphx/op/reshape.hpp	Add `reshape::symbolic_compute` pass-through behavior.
src/include/migraphx/op/normalize_attribute.hpp	Document symbolic normalization behavior for expression attributes.
src/include/migraphx/op/multibroadcast.hpp	Add `multibroadcast::symbolic_compute` broadcast behavior.
src/include/migraphx/op/mul.hpp	Enable symbolic compute for `mul`.
src/include/migraphx/op/identity.hpp	Add `identity::symbolic_compute` pass-through behavior.
src/include/migraphx/op/gather.hpp	Add limited `gather::symbolic_compute` (axis 0, fixed indices).
src/include/migraphx/op/equal.hpp	Enable symbolic compute for `equal` using `provable_equal`.
src/include/migraphx/op/dynamic_range.hpp	Add optional symbolic `output_dim` to produce symbolic output shape for Range.
src/include/migraphx/op/dyn_slice.hpp	New `dyn_slice` operator with symbolic-bound attributes + runtime-bound inputs.
src/include/migraphx/op/div.hpp	Enable symbolic compute for `div` with conservative div-by-zero rejection.
src/include/migraphx/op/dimensions_of.hpp	Add `dimensions_of::symbolic_compute` to return symbolic dimension expressions.
src/include/migraphx/op/convert.hpp	Enable symbolic compute for safe `int64`↔`bool` conversions.
src/include/migraphx/op/concat.hpp	Add limited `concat::symbolic_compute` for axis-0 concatenation of symbolic tensors.
src/include/migraphx/op/broadcast.hpp	Add `broadcast::symbolic_compute` broadcast behavior.
src/include/migraphx/op/broadcast_with_dims.hpp	Add optional `out_dyn_dims` attribute and validation for fully symbolic broadcast outputs.
src/include/migraphx/op/binary.hpp	Add optional symbolic compute path to binary ops via CRTP flag.
src/include/migraphx/op/add.hpp	Enable symbolic compute for `add`.
src/include/migraphx/instruction.hpp	Add `instruction::sym_eval()` declaration.
src/include/migraphx/dim_like.hpp	Add helpers for dimension lists (`all_ints`, `to_ints`) and include updates.
src/CMakeLists.txt	Register `dyn_slice` operator for build.
CHANGELOG.md	Document new `dyn_slice` operator and symbolic attribute normalization; clarify slice symbolic support.

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

+bool interval::contains(const scalar& value) const
+{
+    return not scalar_less(value, min) and not scalar_less(max, value);
 }


Copilot

Pull request overview

Copilot reviewed 70 out of 73 changed files in this pull request and generated no new comments.

…DMIGraphX into core_symbolic_evaluation

gh-app-migraphx-bot-pr-write · 2026-08-23T04:35:07Z

Test	Batch	New Rate (938454)	Old Rate (3a503c)*	Diff	Status
torchvision-resnet50	64	3,298.68	3,264.92	1.03%	✅
torchvision-resnet50_fp16	64	7,882.54	7,548.67	4.42%	✅
torchvision-densenet121	32	2,491.57	2,483.99	0.31%	✅
torchvision-densenet121_fp16	32	5,017.44	5,004.24	0.26%	✅
torchvision-inceptionv3	32	2,058.98	2,058.51	0.02%	✅
torchvision-inceptionv3_fp16	32	4,436.83	4,416.99	0.45%	✅
cadene-inceptionv4	16	819.06	820.61	-0.19%	✅
cadene-resnext64x4	16	785.60	782.78	0.36%	✅
slim-mobilenet	64	8,367.54	8,386.36	-0.22%	✅
slim-nasnetalarge	64	229.25	228.86	0.17%	✅
slim-resnet50v2	64	3,229.00	3,180.91	1.51%	✅
bert-mrpc-onnx	8	1,170.52	1,168.84	0.14%	✅
bert-mrpc-tf	1	504.31	498.63	1.14%	✅
pytorch-examples-wlang-gru	1	376.07	473.35	-20.55%	🔴
pytorch-examples-wlang-lstm	1	547.79	384.83	42.35%	🔆
torchvision-resnet50_1	1	1,050.19	1,046.63	0.34%	✅
cadene-dpn92_1	1	454.84	437.32	4.01%	✅
cadene-resnext101_1	1	364.38	365.89	-0.41%	✅
onnx-taau-downsample	1	841.29	844.09	-0.33%	✅
dlrm-criteoterabyte	1	32.27	32.42	-0.46%	✅
dlrm-criteoterabyte_fp16	1	51.54	51.80	-0.51%	✅
agentmodel	1	15,064.55	9,209.12	63.58%	🔆
unet_fp16	2	58.29	58.80	-0.86%	✅
resnet50v1_fp16	1	1,449.97	1,366.11	6.14%	🔆
resnet50v1_int8	1	1,841.86	1,883.96	-2.23%	✅
bert_base_cased_fp16	64	1,100.24	1,098.16	0.19%	✅
bert_large_uncased_fp16	32	347.22	345.59	0.47%	✅
bert_large_fp16	1	207.33	206.59	0.36%	✅
distilgpt2_fp16	16	2,092.73	2,092.89	-0.01%	✅
yolov5s	1	563.24	558.33	0.88%	✅
tinyllama	1	45.80	45.83	-0.07%	✅
vicuna-fastchat	1	44.25	44.20	0.12%	✅
whisper-tiny-encoder	1	413.38	411.87	0.37%	✅
whisper-tiny-decoder	1	409.18	408.48	0.17%	✅
llama2_7b	1	20.87	20.84	0.14%	✅
qwen1.5-7b	1	23.69	23.58	0.46%	✅
phi3-3.8b	1	28.36	26.72	6.16%	🔆
llama3-8b	1	22.76	21.80	4.40%	✅
whisper-large-encoder	1	10.17	10.18	-0.08%	✅
whisper-large-decoder	1	107.56	105.30	2.15%	✅
mistral-7b	1	23.82	23.78	0.20%	✅
FLUX.1-schnell	1	789.70	755.22	4.57%	✅

Regressions detected 🔴

* No develop baseline was found for this PR's branch point; compared against the latest available develop run instead.

gh-app-migraphx-bot-pr-write · 2026-08-23T04:35:08Z

Test	Status	Result
bert-mrpc-onnx	✅	PASSED: MIGraphX meets tolerance
bert-mrpc-tf	❌	ERROR - check error output traceback Traceback (most recent call last): File "/src/AMDMIGraphX/tools/accuracy/accuracy_checker.py", line 377, in main() File "/src/AMDMIGraphX/tools/accuracy/accuracy_checker.py", line 313, in main import tensorflow as tf File "/usr/local/lib/python3.12/dist-packages/tensorflow/init.py", line 40, in from tensorflow.python import pywrap_tensorflow as _pywrap_tensorflow # pylint: disable=unused-import ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/usr/local/lib/python3.12/dist-packages/tensorflow/python/pywrap_tensorflow.py", line 37, in self_check.preload_check() File "/usr/local/lib/python3.12/dist-packages/tensorflow/python/platform/self_check.py", line 63, in preload_check from tensorflow.python.platform import _pywrap_cpu_feature_guard ImportError: libnuma.so.1: cannot open shared object file: No such file or directory
pytorch-examples-wlang-gru	🔴	FAILED: MIGraphX is not within tolerance - check verbose output
pytorch-examples-wlang-lstm	🔴	FAILED: MIGraphX is not within tolerance - check verbose output
dlrm-criteoterabyte	✅	PASSED: MIGraphX meets tolerance
agentmodel	✅	PASSED: MIGraphX meets tolerance
unet	✅	PASSED: MIGraphX meets tolerance
resnet50v1	✅	PASSED: MIGraphX meets tolerance
bert_base_cased_fp16	✅	PASSED: MIGraphX meets tolerance
bert_large_uncased_fp16	🔴	FAILED: MIGraphX is not within tolerance - check verbose output
bert_large	✅	PASSED: MIGraphX meets tolerance
yolov5s	✅	PASSED: MIGraphX meets tolerance
tinyllama	✅	PASSED: MIGraphX meets tolerance
vicuna-fastchat	✅	PASSED: MIGraphX meets tolerance
whisper-tiny-encoder	✅	PASSED: MIGraphX meets tolerance
whisper-tiny-decoder	✅	PASSED: MIGraphX meets tolerance
distilgpt2_fp16	🔴	FAILED: MIGraphX is not within tolerance - check verbose output
llama2_7b	✅	PASSED: MIGraphX meets tolerance
qwen1.5-7b	✅	PASSED: MIGraphX meets tolerance
phi3-3.8b	✅	PASSED: MIGraphX meets tolerance
llama3-8b	✅	PASSED: MIGraphX meets tolerance
whisper-large-encoder	✅	PASSED: MIGraphX meets tolerance
whisper-large-decoder	✅	PASSED: MIGraphX meets tolerance
mistral-7b	✅	PASSED: MIGraphX meets tolerance
FLUX.1-schnell	✅	PASSED: MIGraphX meets tolerance

…uation

CharlieL7 · 2026-08-24T19:55:04Z

+    static constexpr bool enable_symbolic_compute = false;
+


Why do we need this? Just for ops that inherit another op?

yeah, not every derived op can support symbolic compute

CharlieL7 · 2026-08-24T23:48:58Z

+    sym_argument symbolic_compute(const shape& output_shape,
+                                  const std::vector<sym_argument>& args) const
+    {
+        if(axis != 0 or args.empty() or any_of(args, [](const auto& arg) { return arg.empty(); }))


Why does axis have to be 0? Shouldn't the condition be instead that the args are 1 dimensional?

CharlieL7 · 2026-08-24T23:54:05Z

+    sym_argument symbolic_compute(const shape& output_shape,
+                                  const std::vector<sym_argument>& args) const
+    {
+        if(args.size() != 1 or


Shouldn't there being 1 argument already be handled by compute_shape()?

CharlieL7 · 2026-08-24T23:55:42Z

+        return [](auto x, auto y) {
+            if constexpr(std::is_same<std::decay_t<decltype(y)>, sym::expr>{})
+            {
+                if(y.eval_interval().contains(0))


Is this for handling an empty interval?

no its handling the case where there could be a division by 0

CharlieL7 · 2026-08-25T00:01:16Z

+        if(args.empty() or args[0].get_shape().dynamic() or args[0].get_shape().ndim() != 1 or
+           args[0].empty())
+            return {};
+


This is for handling slice after something like dimensions_of, right? It's somewhat unfortunate that all this handling code for normalizing the attributes needs to be redone here again.

yeah it uses the same normalization code but i dont see how we can avoid having to call normalization altogether

CharlieL7 · 2026-08-25T00:06:58Z

+    return std::nullopt;
+}
+
+std::optional<scalar> fixed_value(const expr& expression)


Is this to check that an expression has a single possible value using the interval analysis? If so, please add a short explanation comment for the function.

ya, adding comment

CharlieL7 · 2026-08-25T00:08:55Z

+    return result;
+}
+
+sym_argument instruction::sym_eval() const


Would be good to have a short comment for this function.

CharlieL7 · 2026-08-25T00:13:15Z

 {
    std::vector<op_desc> operators() const { return {{"Shape"}}; }

+    static std::pair<std::size_t, std::size_t>


Was there a functional change here?

no Im going to revert this. There was previously a functional change which is why I rewrote it, but thats not necessary now.

CharlieL7 · 2026-08-25T00:19:09Z

+
+#include <onnx_test.hpp>
+
+TEST_CASE(symbolic_reshape_markers)


What is this trying to test?

this is for when the reshape dims contain 0 and -1. But this is messy, I am rewriting it as separate and more focused tests

…DMIGraphX into core_symbolic_evaluation

CharlieL7 and others added 15 commits August 4, 2026 18:32

Dynamic slice from symbolic slice in other branch

86195d3

Test and comment cleanup

37ce966

Cleanup normalize_attributes

d58785c

Fix licensing

d4784d4

Simplify comment

fe49783

Add fold_min and fold_max to dyn_slice compute_shape()

205efd9

Add throw for if extent is always negative

72a96fe

use dim_like and 1_arg slice hold

2313bab

Add normalize_attributes tests

eb8bcfa

Merge PR #5112 dynamic slice design

cbc79d0

Merge latest develop into ONNX symbolic values

511f5af

Bring the PR #5112-based parser prerequisite branch onto the current development baseline before applying its remaining changes. Co-authored-by: Cursor <cursoragent@cursor.com>

Adapt expression resolver coverage to dyn_slice

33e2e09

Keep the merged resolver regression compatible with the dynamic-slice interface introduced by PR #5112. Co-authored-by: Cursor <cursoragent@cursor.com>

Preserve symbolic shape tensor values during ONNX parsing

9c73853

Track exact integral shape values through parser operations so dynamic consumers retain symbolic output relationships without changing runtime dataflow. Co-authored-by: Cursor <cursoragent@cursor.com>

Remove unrelated parser style changes

3a2f48c

Keep the symbolic-value change focused by restoring existing resolver diagnostics and simplifying the signed Gather size declaration. Co-authored-by: Cursor <cursoragent@cursor.com>

symbolic parsing via sym_eval

8388abf

shivadbhavsar requested review from CharlieL7, pfultz2 and turneram August 18, 2026 05:48

pfultz2 reviewed Aug 19, 2026

View reviewed changes

shivadbhavsar and others added 2 commits August 20, 2026 18:02

define sym_argument and clean up

5a53136

Merge branch 'develop' into core_symbolic_evaluation

c644515

shivadbhavsar requested a review from pfultz2 August 21, 2026 01:03

shivadbhavsar marked this pull request as ready for review August 21, 2026 01:03

shivadbhavsar requested review from a team and causten as code owners August 21, 2026 01:03

Copilot AI lite review requested due to automatic review settings August 21, 2026 01:03

Copilot started reviewing on behalf of shivadbhavsar August 21, 2026 01:05 View session

Copilot AI reviewed Aug 21, 2026

View reviewed changes

Comment thread src/sym.cpp

Comment on lines +204 to 207

bool interval::contains(const scalar& value) const

{

return not scalar_less(value, min) and not scalar_less(max, value);

}

Comment thread src/onnx/parse_where.cpp Outdated

pfultz2 reviewed Aug 21, 2026

View reviewed changes

review comments and clean up

859cb68

shivadbhavsar requested review from pfultz2 and a lite review from Copilot August 21, 2026 03:58

Copilot started reviewing on behalf of shivadbhavsar August 21, 2026 03:59 View session

Copilot AI reviewed Aug 21, 2026

View reviewed changes

shivadbhavsar and others added 5 commits August 21, 2026 09:30

fix cppcheck and other ci

3fcb86b

Merge branch 'develop' into core_symbolic_evaluation

bc0a2b7

move definitions to cpp

f3dd316

Merge branch 'core_symbolic_evaluation' of https://github.com/ROCm/AM…

1b64cd4

…DMIGraphX into core_symbolic_evaluation

Merge branch 'develop' into core_symbolic_evaluation

27e0d2a

pfultz2 approved these changes Aug 21, 2026

View reviewed changes

Comment thread src/include/migraphx/sym_argument.hpp Outdated

shivadbhavsar added 3 commits August 21, 2026 13:17

define equality as friend funcs

1c1fcc2

fix warning

0bd16a9

fix dangling ref

a69e68a

shivadbhavsar and others added 2 commits August 24, 2026 13:33

Merge remote-tracking branch 'origin/develop' into core_symbolic_eval…

77d671c

…uation

Merge branch 'develop' into core_symbolic_evaluation

938454c

CharlieL7 requested changes Aug 25, 2026

View reviewed changes

CharlieL7 reviewed Aug 25, 2026

View reviewed changes

pfultz2 and others added 4 commits August 25, 2026 14:06

Merge branch 'develop' into core_symbolic_evaluation

16ab989

review comments and format

14d8225

remove obsolete test

b7f987e

Merge branch 'core_symbolic_evaluation' of https://github.com/ROCm/AM…

d1994d1

…DMIGraphX into core_symbolic_evaluation

shivadbhavsar requested a review from CharlieL7 August 25, 2026 21:26

Merge branch 'develop' into core_symbolic_evaluation

66e8d12


		#include <onnx_test.hpp>

		TEST_CASE(symbolic_reshape_markers)

Uh oh!

Conversation

shivadbhavsar commented Aug 18, 2026

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Uh oh!

Uh oh!

gh-app-migraphx-bot-pr-write Bot commented Aug 23, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

gh-app-migraphx-bot-pr-write Bot commented Aug 23, 2026

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

gh-app-migraphx-bot-pr-write Bot commented Aug 23, 2026 •

edited

Loading