support high priority stream

[Chao1Han]

Support high priority stream for xccl, test case add in #2049
We need merge this pr first and upstream op register pytorch/pytorch#163049 and then test case could be pass

[Copilot]

Pull Request Overview

This pull request adds support for high priority streams in the XCCL process group. Key changes include adding a new Options struct with high priority and group name parameters, introducing a new groupRanks() accessor, and updating constructor and logging logic to reflect high priority stream usage.

Reviewed Changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 1 comment.

File	Description
src/xccl/ProcessGroupXCCL.hpp	Added a high priority stream option and new Options struct for configuration.
src/xccl/ProcessGroupXCCL.cpp	Updated constructor initialization, logging, and introduced groupRanks().

Comments suppressed due to low confidence (1)

src/xccl/ProcessGroupXCCL.hpp:25

• The constant TORCH_XCCL_HIGH_PRIORITY is defined as a non-const vector. Consider renaming and declaring it as a const container (or using constexpr) to clearly indicate its immutability.

static std::vector<std::string> TORCH_XCCL_HIGH_PRIORITY = {

[pytorchxpubot]

@sys_pytorchxpubot triage result for run 15864761625

Triage bot UT analaysis result for reference only, please note unique error message only report once:

1. third_party.torch-xpu-ops.test.xpu.test_modules_xpu.TestModuleXPU test_cpu_gpu_parity_nn_CrossEntropyLoss_xpu_float64 got failed with error message

 AssertionError: Scalars are not close! 

Triage bot response:

{
  "similar_issue_id": 645,
  "similar_issue_state": "closed",
  "issue_owner": "daisyden",
  "issue_description": "UT got failed with FP64 emulation feature. The reporter is mengfei25, and the assignee is daisyden. The issue is closed.",
  "root_causes": [
    "Issues related to tensor operations and reductions leading to precision mismatches.",
    "Potential differences in computation between CPU and XPU implementations.",
    "Possible issues with the CrossEntropyLoss implementation on XPU."
  ],
  "suggested_solutions": [
    "Investigate the precision handling in CrossEntropyLoss on XPU.",
    "Check for any implementation differences causing scalar mismatches.",
    "Consider allowing a small tolerance in scalar comparisons for CPU-GPU parity tests.",
    "Review and update test cases to handle potential precision discrepancies."
  ]
}