Add backend object management - #30
Open
kpouget wants to merge 3 commits into
Open
Conversation
|
[APPROVALNOTIFIER] This PR is NOT APPROVED This pull-request has been approved by: The full list of commands accepted by this bot can be found here. DetailsNeeds approval from an approver in each of these files:Approvers can indicate their approval by writing |
WalkthroughThis refactoring transitions the backend from a single global initialization model to per-device, per-backend-instance management. It introduces lifecycle commands for initialize and cleanup, adds device context tracking with per-device backend instance maps, updates initialization to return device handles and backend IDs, and modifies graph compute to require these identifiers. Changes
Sequence DiagramsequenceDiagram
participant Client as Client (Frontend)
participant RemotingBackend as Remoting Backend<br/>(ggml-backend.cpp)
participant VirtGPU as VirtGPU<br/>(virtgpu-forward)
participant DispatchBackend as Dispatch Backend<br/>(backend-dispatched)
participant DeviceContext as Device Context<br/>& Backend Instance Map
Note over Client,DeviceContext: Backend Initialization Flow
Client->>RemotingBackend: ggml_backend_remoting_device_init(ctx)
RemotingBackend->>VirtGPU: apir_backend_initialize(gpu, reg_fct, &device_handle, &backend_id)
VirtGPU->>DispatchBackend: APIR_COMMAND_TYPE_BACKEND_INITIALIZE (encodes reg_fct)
DispatchBackend->>DeviceContext: ensure_device_context(dev)
DeviceContext-->>DispatchBackend: apir_device_context*
DispatchBackend->>DeviceContext: create_backend_instance(dev)
DeviceContext-->>DispatchBackend: backend_id, instance*
DispatchBackend->>VirtGPU: response (device_handle, backend_id)
VirtGPU-->>RemotingBackend: (device_handle, backend_id)
RemotingBackend-->>Client: ggml_backend_t
Note over Client,DeviceContext: Graph Compute Flow (with Device/Backend IDs)
Client->>RemotingBackend: ggml_backend_remoting_graph_compute(backend, cgraph)
RemotingBackend->>VirtGPU: apir_backend_graph_compute(gpu, device_handle, backend_id, cgraph)
VirtGPU->>DispatchBackend: APIR_COMMAND_TYPE_BACKEND_GRAPH_COMPUTE<br/>(device_handle, backend_id, cgraph)
DispatchBackend->>DeviceContext: get_backend_instance(dev, backend_id)
DeviceContext-->>DispatchBackend: instance*
DispatchBackend->>DispatchBackend: instance->bck->iface.graph_compute(cgraph)
DispatchBackend->>VirtGPU: response
VirtGPU-->>RemotingBackend: result
RemotingBackend-->>Client: status
Note over Client,DeviceContext: Backend Cleanup Flow
Client->>RemotingBackend: ggml_backend_remoting_free(backend)
RemotingBackend->>VirtGPU: apir_backend_cleanup(gpu, device_handle, backend_id)
VirtGPU->>DispatchBackend: APIR_COMMAND_TYPE_BACKEND_CLEANUP<br/>(device_handle, backend_id)
DispatchBackend->>DeviceContext: cleanup_backend_instance(dev, backend_id)
DeviceContext->>DeviceContext: delete instance, ggml_backend_free(bck)
DispatchBackend->>VirtGPU: response
VirtGPU-->>RemotingBackend: void
RemotingBackend-->>Client: success
Estimated code review effort🎯 4 (Complex) | ⏱️ ~45 minutes Possibly related PRs
Poem
🚥 Pre-merge checks | ✅ 1 | ❌ 2❌ Failed checks (2 warnings)
✅ Passed checks (1 passed)
✏️ Tip: You can configure your own custom pre-merge checks in the settings. ✨ Finishing Touches🧪 Generate unit tests (beta)
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
There was a problem hiding this comment.
Actionable comments posted: 4
Caution
Some comments are outside the diff and can’t be posted inline due to platform limitations.
⚠️ Outside diff range comments (1)
ggml/src/ggml-virtgpu/ggml-backend.cpp (1)
66-83:⚠️ Potential issue | 🔴 CriticalKeep backend-instance IDs off the shared device context.
dev->contextis shared by every backend created from that device, but this path writesdevice_handleandbackend_idinto that shared object and then stores the samectxon eachggml_backend. Initializing a second backend on the same device will overwrite the identifiers used by the first one, so latergraph_compute()orfree()calls can hit or destroy the wrong remote backend instance.🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@ggml/src/ggml-virtgpu/ggml-backend.cpp` around lines 66 - 83, The shared dev->context (ggml_backend_remoting_device_context) is being used to store per-backend identifiers (device_handle and backend_id), causing races/overwrites when multiple ggml_backend instances are created for the same device; instead allocate or clone a new per-backend context structure for each ggml_backend (do not write into the shared dev->context), call apir_backend_initialize with that per-backend context to populate its device_handle/backend_id, and set ggml_backend::context to this new per-backend context; locate uses in ggml_backend_remoting_device_context, apir_backend_initialize, ggml_backend_t allocation, ggml_backend_remoting_interface, ggml_backend_remoting_guid and ggml_backend_reg_dev_get to implement the per-backend context allocation and initialization and ensure teardown/free uses the per-backend context.
🧹 Nitpick comments (2)
ggml/src/ggml-virtgpu/backend/backend-dispatched.h (1)
38-39: Type mismatch:next_backend_idisuintptr_tbut API usesuint32_t.
next_backend_idis declared asuintptr_t(line 39), butbackend_dispatch_initializeoutputsuint32_t * out_backend_id(line 59), andget_backend_instancetakesuintptr_t backend_id(line 56). This could cause truncation on 64-bit systems if many backends are created.Consider using a consistent type throughout—either
uint32_t(sufficient for practical use) oruintptr_teverywhere.🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@ggml/src/ggml-virtgpu/backend/backend-dispatched.h` around lines 38 - 39, The current mix of uintptr_t and uint32_t for backend IDs can truncate on 64-bit systems; make the ID type consistent by switching the internal storage and APIs to uint32_t: change backend_instances key type from uintptr_t to uint32_t, change next_backend_id from uintptr_t to uint32_t, and update get_backend_instance's parameter to uint32_t (and any internal uses/casts) so backend_dispatch_initialize's uint32_t *out_backend_id matches the internal types everywhere.ggml/src/ggml-virtgpu/virtgpu-forward-backend.cpp (1)
3-7: Misleading function name and potential dead code.The function is named
current_time_msbut returns nanoseconds (ts.tv_sec * 1000000000LL + ts.tv_nsec). If this is used elsewhere for timing, the calculation is correct for nanoseconds; otherwise, consider removing if unused.🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@ggml/src/ggml-virtgpu/virtgpu-forward-backend.cpp` around lines 3 - 7, The function current_time_ms returns nanoseconds while its name implies milliseconds; update the implementation or name to match intent: either rename current_time_ms to current_time_ns (and adjust any callers that expect nanoseconds) or change the calculation in current_time_ms to return milliseconds by dividing the nanosecond value by 1'000'000; also consider switching clock_gettime to CLOCK_MONOTONIC for elapsed timing if used for intervals and remove the function entirely if it is unused. Ensure to update all references to current_time_ms/current_time_ns accordingly.
🤖 Prompt for all review comments with AI agents
Verify each finding against the current code and only fix it if needed.
Inline comments:
In `@ggml/src/ggml-virtgpu/backend/backend-dispatched.h`:
- Around line 29-47: The comment for the magic field in apir_backend_instance is
wrong—update the inline comment next to the backend instance's uint32_t magic to
match APIR_BACKEND_INSTANCE_MAGIC (0xCD4321BA); also verify the comment for
apir_device_context::magic matches APIR_DEVICE_EXTENSION_MAGIC (0xAB1234CD) so
the two struct comments correctly reference APIR_BACKEND_INSTANCE_MAGIC and
APIR_DEVICE_EXTENSION_MAGIC respectively (use the symbols apir_backend_instance,
apir_device_context, APIR_BACKEND_INSTANCE_MAGIC, APIR_DEVICE_EXTENSION_MAGIC to
locate and fix the comments).
In `@ggml/src/ggml-virtgpu/backend/backend.cpp`:
- Around line 107-117: The code currently always calls
ggml_backend_reg_dev_get(reg, 0) and stores that single dev for all backends
(via backend_reg_fct and dev), which pins every instance to device 0; instead,
propagate the selected device index through the new initialize flow and call
ggml_backend_reg_dev_get(reg, device_index) per backend instance (do not reuse
the single dev variable). Update the initialize/creation functions that call
backend_reg_fct/ggml_backend_reg_dev_get to accept/forward a device parameter
and obtain a device handle for each instance (refer to backend_reg_fct,
ggml_backend_reg_dev_get, and the dev variable) so backends requested for device
n use device n.
In `@ggml/src/ggml-virtgpu/ggmlremoting_functions.yaml`:
- Around line 143-148: The wire contract for initialize currently encodes three
values (int result, uintptr_t device_handle, uint32_t backend_id) but one
implementation decodes only two, causing shifted/corrupted ids; update the
decoder in the virtgpu frontend implementation of initialize to read the values
in the exact order and types declared by the contract (first read an int named
result, then a uintptr_t device_handle, then a uint32_t backend_id), handle
non-zero result appropriately (return/fail early), and ensure the
encoder/decoder ordering and types match exactly across implementations (result,
device_handle, backend_id).
In `@ggml/src/ggml-virtgpu/virtgpu-forward-backend.cpp`:
- Around line 114-128: The cleanup currently captures but ignores the
REMOTE_CALL return code in apir_backend_cleanup; update apir_backend_cleanup to
check the ApirForwardReturnCode ret after REMOTE_CALL (e.g., if ret != success)
and emit a diagnostic log including ret and context (device_handle, backend_id)
before calling remote_call_finish; reference apir_backend_cleanup, REMOTE_CALL,
ret, and remote_call_finish when making the change and use the existing logging
facility available on virtgpu (or stderr/fprintf if none) so cleanup failures
are visible for debugging.
---
Outside diff comments:
In `@ggml/src/ggml-virtgpu/ggml-backend.cpp`:
- Around line 66-83: The shared dev->context
(ggml_backend_remoting_device_context) is being used to store per-backend
identifiers (device_handle and backend_id), causing races/overwrites when
multiple ggml_backend instances are created for the same device; instead
allocate or clone a new per-backend context structure for each ggml_backend (do
not write into the shared dev->context), call apir_backend_initialize with that
per-backend context to populate its device_handle/backend_id, and set
ggml_backend::context to this new per-backend context; locate uses in
ggml_backend_remoting_device_context, apir_backend_initialize, ggml_backend_t
allocation, ggml_backend_remoting_interface, ggml_backend_remoting_guid and
ggml_backend_reg_dev_get to implement the per-backend context allocation and
initialization and ensure teardown/free uses the per-backend context.
---
Nitpick comments:
In `@ggml/src/ggml-virtgpu/backend/backend-dispatched.h`:
- Around line 38-39: The current mix of uintptr_t and uint32_t for backend IDs
can truncate on 64-bit systems; make the ID type consistent by switching the
internal storage and APIs to uint32_t: change backend_instances key type from
uintptr_t to uint32_t, change next_backend_id from uintptr_t to uint32_t, and
update get_backend_instance's parameter to uint32_t (and any internal
uses/casts) so backend_dispatch_initialize's uint32_t *out_backend_id matches
the internal types everywhere.
In `@ggml/src/ggml-virtgpu/virtgpu-forward-backend.cpp`:
- Around line 3-7: The function current_time_ms returns nanoseconds while its
name implies milliseconds; update the implementation or name to match intent:
either rename current_time_ms to current_time_ns (and adjust any callers that
expect nanoseconds) or change the calculation in current_time_ms to return
milliseconds by dividing the nanosecond value by 1'000'000; also consider
switching clock_gettime to CLOCK_MONOTONIC for elapsed timing if used for
intervals and remove the function entirely if it is unused. Ensure to update all
references to current_time_ms/current_time_ns accordingly.
🪄 Autofix (Beta)
Fix all unresolved CodeRabbit comments on this PR:
- Push a commit to this branch (recommended)
- Create a new PR with the fixes
ℹ️ Review info
⚙️ Run configuration
Configuration used: Organization UI
Review profile: CHILL
Plan: Pro
Run ID: 0917dd7b-cabe-46c4-ba2e-f8c90adc2388
📒 Files selected for processing (13)
ggml/src/ggml-virtgpu/backend/backend-dispatched-backend.cppggml/src/ggml-virtgpu/backend/backend-dispatched.cppggml/src/ggml-virtgpu/backend/backend-dispatched.gen.hggml/src/ggml-virtgpu/backend/backend-dispatched.hggml/src/ggml-virtgpu/backend/backend-virgl-apir.hggml/src/ggml-virtgpu/backend/backend.cppggml/src/ggml-virtgpu/backend/shared/apir_backend.gen.hggml/src/ggml-virtgpu/ggml-backend-reg.cppggml/src/ggml-virtgpu/ggml-backend.cppggml/src/ggml-virtgpu/ggml-remoting.hggml/src/ggml-virtgpu/ggmlremoting_functions.yamlggml/src/ggml-virtgpu/virtgpu-forward-backend.cppggml/src/ggml-virtgpu/virtgpu-forward.gen.h
💤 Files with no reviewable changes (1)
- ggml/src/ggml-virtgpu/backend/backend-virgl-apir.h
Comment on lines
+29
to
+47
| // Backend instance structure - one backend per instance | ||
| struct apir_backend_instance { | ||
| ggml_backend_t bck; // The actual backend | ||
| uint32_t magic; // For validation: 0xAB1234CD | ||
| }; | ||
|
|
||
| // Device context structure - can have multiple backend instances | ||
| struct apir_device_context { | ||
| std::mutex backends_mutex; | ||
| std::unordered_map<uintptr_t, apir_backend_instance *> backend_instances; | ||
| uintptr_t next_backend_id; | ||
|
|
||
| bool async_backend; // Whether the backend supports async operations | ||
|
|
||
| uint32_t magic; // For validation: 0xAB1234CD | ||
| }; | ||
|
|
||
| #define APIR_DEVICE_EXTENSION_MAGIC 0xAB1234CD | ||
| #define APIR_BACKEND_INSTANCE_MAGIC 0xCD4321BA |
There was a problem hiding this comment.
🧩 Analysis chain
🏁 Script executed:
#!/bin/bash
# Verify async_backend naming consistency
rg -n 'async_backend' --type=cpp ggml/src/ggml-virtgpu/backend/Repository: crc-org/llama.cpp
Length of output: 422
Magic number comment is incorrect.
Line 32 comment says 0xAB1234CD but APIR_BACKEND_INSTANCE_MAGIC is defined as 0xCD4321BA on line 47. The comment should reference the correct magic value:
🔧 Fix comment to match actual magic value
// Backend instance structure - one backend per instance
struct apir_backend_instance {
ggml_backend_t bck; // The actual backend
- uint32_t magic; // For validation: 0xAB1234CD
+ uint32_t magic; // For validation: 0xCD4321BA
};📝 Committable suggestion
‼️ IMPORTANT
Carefully review the code before committing. Ensure that it accurately replaces the highlighted code, contains no missing lines, and has no issues with indentation. Thoroughly test & benchmark the code to ensure it meets the requirements.
Suggested change
| // Backend instance structure - one backend per instance | |
| struct apir_backend_instance { | |
| ggml_backend_t bck; // The actual backend | |
| uint32_t magic; // For validation: 0xAB1234CD | |
| }; | |
| // Device context structure - can have multiple backend instances | |
| struct apir_device_context { | |
| std::mutex backends_mutex; | |
| std::unordered_map<uintptr_t, apir_backend_instance *> backend_instances; | |
| uintptr_t next_backend_id; | |
| bool async_backend; // Whether the backend supports async operations | |
| uint32_t magic; // For validation: 0xAB1234CD | |
| }; | |
| #define APIR_DEVICE_EXTENSION_MAGIC 0xAB1234CD | |
| #define APIR_BACKEND_INSTANCE_MAGIC 0xCD4321BA | |
| // Backend instance structure - one backend per instance | |
| struct apir_backend_instance { | |
| ggml_backend_t bck; // The actual backend | |
| uint32_t magic; // For validation: 0xCD4321BA | |
| }; | |
| // Device context structure - can have multiple backend instances | |
| struct apir_device_context { | |
| std::mutex backends_mutex; | |
| std::unordered_map<uintptr_t, apir_backend_instance *> backend_instances; | |
| uintptr_t next_backend_id; | |
| bool async_backend; // Whether the backend supports async operations | |
| uint32_t magic; // For validation: 0xAB1234CD | |
| }; | |
| `#define` APIR_DEVICE_EXTENSION_MAGIC 0xAB1234CD | |
| `#define` APIR_BACKEND_INSTANCE_MAGIC 0xCD4321BA |
🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed.
In `@ggml/src/ggml-virtgpu/backend/backend-dispatched.h` around lines 29 - 47, The
comment for the magic field in apir_backend_instance is wrong—update the inline
comment next to the backend instance's uint32_t magic to match
APIR_BACKEND_INSTANCE_MAGIC (0xCD4321BA); also verify the comment for
apir_device_context::magic matches APIR_DEVICE_EXTENSION_MAGIC (0xAB1234CD) so
the two struct comments correctly reference APIR_BACKEND_INSTANCE_MAGIC and
APIR_DEVICE_EXTENSION_MAGIC respectively (use the symbols apir_backend_instance,
apir_device_context, APIR_BACKEND_INSTANCE_MAGIC, APIR_DEVICE_EXTENSION_MAGIC to
locate and fix the comments).
Comment on lines
+107
to
+117
| reg = backend_reg_fct(); | ||
| if (reg == NULL) { | ||
| return (ApirLoadLibraryReturnCode) (APIR_LOAD_LIBRARY_INIT_BASE_INDEX + | ||
| APIR_BACKEND_INITIALIZE_BACKEND_INIT_FAILED); | ||
| } | ||
|
|
||
| dev = ggml_backend_reg_dev_get(reg, 0); | ||
| if (dev == NULL) { | ||
| return (ApirLoadLibraryReturnCode) (APIR_LOAD_LIBRARY_INIT_BASE_INDEX + | ||
| APIR_BACKEND_INITIALIZE_BACKEND_INIT_FAILED); | ||
| } |
There was a problem hiding this comment.
Don't pin every initialized backend to device 0.
This stores a single dev = ggml_backend_reg_dev_get(reg, 0) for all later backend-instance creation. The frontend init path is device-specific, but nothing in the new initialize flow carries that selection through, so a backend requested for device n > 0 will still be instantiated on the first remote device.
🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed.
In `@ggml/src/ggml-virtgpu/backend/backend.cpp` around lines 107 - 117, The code
currently always calls ggml_backend_reg_dev_get(reg, 0) and stores that single
dev for all backends (via backend_reg_fct and dev), which pins every instance to
device 0; instead, propagate the selected device index through the new
initialize flow and call ggml_backend_reg_dev_get(reg, device_index) per backend
instance (do not reuse the single dev variable). Update the initialize/creation
functions that call backend_reg_fct/ggml_backend_reg_dev_get to accept/forward a
device parameter and obtain a device handle for each instance (refer to
backend_reg_fct, ggml_backend_reg_dev_get, and the dev variable) so backends
requested for device n use device n.
Comment on lines
+143
to
+148
| initialize: | ||
| frontend_return: "int" | ||
| frontend_extra_params: | ||
| - "void *ggml_backend_reg_fct_p" | ||
| - "uintptr_t* out_device_handle" | ||
| - "uint32_t* out_backend_id" |
There was a problem hiding this comment.
The new initialize wire contract is out of sync with its implementations.
This API now declares an int return plus two output parameters, but the current backend/frontend pair disagree on the success payload: ggml/src/ggml-virtgpu/backend/backend-dispatched-backend.cpp:12-49 encodes result, device_handle, backend_id, while ggml/src/ggml-virtgpu/virtgpu-forward-backend.cpp:9-54 decodes only device_handle and backend_id. On the first successful init, both returned identifiers will be shifted/corrupted. Make both sides agree on one exact response shape.
🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed.
In `@ggml/src/ggml-virtgpu/ggmlremoting_functions.yaml` around lines 143 - 148,
The wire contract for initialize currently encodes three values (int result,
uintptr_t device_handle, uint32_t backend_id) but one implementation decodes
only two, causing shifted/corrupted ids; update the decoder in the virtgpu
frontend implementation of initialize to read the values in the exact order and
types declared by the contract (first read an int named result, then a uintptr_t
device_handle, then a uint32_t backend_id), handle non-zero result appropriately
(return/fail early), and ensure the encoder/decoder ordering and types match
exactly across implementations (result, device_handle, backend_id).
Comment on lines
+114
to
+128
| void apir_backend_cleanup(virtgpu * gpu, uintptr_t device_handle, uint32_t backend_id) { | ||
| apir_encoder * encoder; | ||
| apir_decoder * decoder; | ||
| ApirForwardReturnCode ret; | ||
|
|
||
| REMOTE_CALL_PREPARE(gpu, encoder, APIR_COMMAND_TYPE_BACKEND_CLEANUP); | ||
|
|
||
| // Send device handle and backend ID separately | ||
| apir_encode_uintptr_t(encoder, &device_handle); | ||
| apir_encode_uint32_t(encoder, &backend_id); | ||
|
|
||
| REMOTE_CALL(gpu, encoder, decoder, ret); | ||
|
|
||
| remote_call_finish(gpu, encoder, decoder); | ||
| } |
There was a problem hiding this comment.
Unused ret variable - cleanup errors are silently ignored.
The ret return code from REMOTE_CALL is captured but never used. While cleanup functions often tolerate failures (since there's little recourse), logging the error would aid debugging shutdown issues.
🔧 Suggested fix to log cleanup failures
REMOTE_CALL(gpu, encoder, decoder, ret);
+ if (ret != 0) {
+ GGML_LOG_WARN(GGML_VIRTGPU "%s: Backend cleanup returned: %d\n", __func__, ret);
+ }
+
remote_call_finish(gpu, encoder, decoder);
}📝 Committable suggestion
‼️ IMPORTANT
Carefully review the code before committing. Ensure that it accurately replaces the highlighted code, contains no missing lines, and has no issues with indentation. Thoroughly test & benchmark the code to ensure it meets the requirements.
Suggested change
| void apir_backend_cleanup(virtgpu * gpu, uintptr_t device_handle, uint32_t backend_id) { | |
| apir_encoder * encoder; | |
| apir_decoder * decoder; | |
| ApirForwardReturnCode ret; | |
| REMOTE_CALL_PREPARE(gpu, encoder, APIR_COMMAND_TYPE_BACKEND_CLEANUP); | |
| // Send device handle and backend ID separately | |
| apir_encode_uintptr_t(encoder, &device_handle); | |
| apir_encode_uint32_t(encoder, &backend_id); | |
| REMOTE_CALL(gpu, encoder, decoder, ret); | |
| remote_call_finish(gpu, encoder, decoder); | |
| } | |
| void apir_backend_cleanup(virtgpu * gpu, uintptr_t device_handle, uint32_t backend_id) { | |
| apir_encoder * encoder; | |
| apir_decoder * decoder; | |
| ApirForwardReturnCode ret; | |
| REMOTE_CALL_PREPARE(gpu, encoder, APIR_COMMAND_TYPE_BACKEND_CLEANUP); | |
| // Send device handle and backend ID separately | |
| apir_encode_uintptr_t(encoder, &device_handle); | |
| apir_encode_uint32_t(encoder, &backend_id); | |
| REMOTE_CALL(gpu, encoder, decoder, ret); | |
| if (ret != 0) { | |
| GGML_LOG_WARN(GGML_VIRTGPU "%s: Backend cleanup returned: %d\n", __func__, ret); | |
| } | |
| remote_call_finish(gpu, encoder, decoder); | |
| } |
🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed.
In `@ggml/src/ggml-virtgpu/virtgpu-forward-backend.cpp` around lines 114 - 128,
The cleanup currently captures but ignores the REMOTE_CALL return code in
apir_backend_cleanup; update apir_backend_cleanup to check the
ApirForwardReturnCode ret after REMOTE_CALL (e.g., if ret != success) and emit a
diagnostic log including ret and context (device_handle, backend_id) before
calling remote_call_finish; reference apir_backend_cleanup, REMOTE_CALL, ret,
and remote_call_finish when making the change and use the existing logging
facility available on virtgpu (or stderr/fprintf if none) so cleanup failures
are visible for debugging.
Collaborator
Author
|
/test topsail remoting_mac |
|
🔴 Test of 'mac_ai test prepare_ci' failed after 00 hours 19 minutes 33 seconds. 🔴 • Link to the test results. • No reports index generated... Test configuration: Failure indicator: Empty. (See run.log) |
Collaborator
Author
|
/test topsail remoting_mac |
|
🔴 Test of 'mac_ai test prepare_ci' failed after 00 hours 09 minutes 09 seconds. 🔴 • Link to the test results. • No reports index generated... Test configuration: Failure indicator: Empty. (See run.log) |
|
@kpouget: The following test failed, say
Full PR test history. Your PR dashboard. DetailsInstructions for interacting with me using PR comments are available here. If you have questions or suggestions related to my behavior, please file an issue against the kubernetes-sigs/prow repository. I understand the commands that are listed here. |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary by CodeRabbit
New Features
Improvements