cpu-o3: ci: implement tiered CI arch for faster PR feedback

Refactor CI pipeline into three tiers to balance speed and thoroughness:

**Tier 1: PR Quick Check (5-10 minutes)**
- New workflow: `.github/workflows/pr-quick-check.yml`
- Runs on every PR push for fast feedback
- Includes: build, unit tests, difftest smoke test
- Follows DRY principle: tests not repeated in post-merge

**Tier 1.5: On-Demand Performance Testing**
- New workflow: `.github/workflows/on-demand-spec.yml`
- Triggered by PR comment: `/run-spec [benchmark_type]`
- Supports: spec06-0.8c (default), spec06-1.0c, spec17-1.0c, RVV variants
- Permission-gated: OWNER/MEMBER/COLLABORATOR only
- Simplifies maintenance: benchmark types managed in template only

**Tier 2: Post-Merge Full Testing**
- Modified workflows: `gem5.yml`, `gem5-perf.yml`, `gem5-vector.yml`, etc.
- Removed `pull_request` triggers, kept `push` to `xs-dev`
- Runs comprehensive regression tests after merge
- Removed duplicate jobs (unit_tests, difftest_check) per DRY

**Benefits**
- PR feedback time: 2-4 hours → 5-10 minutes (95%+ improvement)
- On-demand perf testing: saves CI resources, runs only when needed
- Better separation of concerns: fast checks vs. thorough validation
- Easier maintenance: centralized benchmark configuration

Related workflows updated:
- pr-quick-check.yml (new)
- on-demand-spec.yml (new)
- gem5.yml (modified)
- gem5-perf.yml (modified)
- gem5-vector.yml (modified)
- gem5-ideal-btb-perf.yml (modified)
- gem5-ideal-btb-perf-nosc.yml (modified)
- gem5-perf-template.yml (simplified)

Change-Id: I6798375b787cec2f2e078e7a17f4a21cfc894dc8

Author: jensen-yan

.github/workflows/README.md

@@ -0,0 +1,209 @@
+# GEM5 分层CI架构 (Tiered CI)
+
+解决当前CI在PR阶段运行过久、拖慢开发效率的问题。
+
+---
+
+## 📊 核心改进
+
+| 阶段 | 之前 | 现在 | 改进 |
+|-----|------|------|------|
+| PR快速反馈 | 2-4小时 | **5-10分钟** | ⚡ 95%+ |
+| 按需性能测试 | 每个PR强制 | 需要时触发 | 🎯 按需 |
+| 完整测试 | 每个PR重复 | 只在合入后 | ✅ DRY |
+
+---
+
+## 层次一：PR快速检查 (Tier 1) ⚡
+
+**文件**: `.github/workflows/pr-quick-check.yml`
+
+**目标**: 5-10分钟内给出快速反馈
+
+**触发**: 每次 push 到 PR 分支
+
+**内容**:
+- ✅ 编译 GEM5 opt 版本
+- ✅ 单元测试 (Unit Tests)
+- ✅ 冒烟测试 (Difftest Check)
+
+**说明**: 
+- 遵循 DRY 原则，这些测试不会在 Post-Merge 阶段重复运行
+- 使用本地 DRAMSim3 缓存，避免网络IO
+
+---
+
+## 层次 1.5：按需性能测试 (Tier 1.5) 🎯
+
+**文件**: `.github/workflows/on-demand-spec.yml`
+
+**目标**: 在合入前，按需检查有性能风险的 PR
+
+**触发**: 在 PR 评论中输入命令
+
+### 支持的命令
+
+```bash
+/run-spec                    # 默认：SPEC06 INT 80%覆盖率 (~500 checkpoints)
+/run-spec spec06-1.0c        # SPEC06 100%覆盖率
+/run-spec spec17-1.0c        # SPEC17 100%覆盖率
+/run-spec spec06-rvv-1.0c    # SPEC06 RVV扩展 100%
+/run-spec spec06int-rvv-0.8c # SPEC06 INT RVV 80%
+```
+
+### 权限控制
+
+仅以下角色可触发：OWNER / MEMBER / COLLABORATOR
+
+### 性能结果
+
+由现有的性能评论机器人 (`actions_gem5.py`) 自动处理：
+- 📊 与主分支性能对比
+- 📊 与PR上一个commit对比
+- �� 详细的性能指标表格
+
+### 优势
+
+- 只在需要时运行，节省资源
+- 支持多种 benchmark 类型
+- 添加新 benchmark 类型只需修改 template
+
+---
+
+## 层次二：主线完整测试 (Tier 2) 🛡️
+
+**目标**: 确保 `xs-dev` 分支永远健康、可发布
+
+**触发**: PR 合入 `xs-dev` 分支后自动运行
+
+### 包含的测试 Workflows
+
+#### 1. `gem5.yml` - 功能回归测试
+8个并行 jobs（遵循DRY原则，排除已在 Tier 1 运行的测试）
+
+**已移除**（避免重复）:
+- ~~`unit_tests`~~ → 在 `pr-quick-check.yml`
+- ~~`difftest_check`~~ → 在 `pr-quick-check.yml`
+
+#### 2. `gem5-perf.yml` - 标准性能测试
+SPEC06 80%覆盖率性能基线
+
+#### 3. `gem5-ideal-btb-perf.yml` - BTB性能测试
+BTB 配置下的 SPEC06 性能测试
+
+#### 4. 其他测试
+- `gem5-vector.yml` - RVV 扩展测试
+- `gem5-ideal-btb-perf-nosc.yml` - 无SC的BTB测试
+- `gem5-ideal-btb-perf-weekly.yml` - 定时任务（每周四）
+
+---
+
+## 🔑 配套策略
+
+### 1. "主线红了" 怎么办：立即回滚 (Revert)
+
+**原则**: 不允许主线 (`xs-dev`) 保持红色状态
+
+**动作**:
+```bash
+git revert <merge-commit-sha> -m 1
+git push origin xs-dev
+```
+或者直接在github 网页端，找到已经被关闭的PR, 在最下方有revert 按钮，来直接revert 这个PR.
+
+**后续**: 原 PR 作者修复 Bug 后，重新提交新的 PR
+
+### 2. 合并策略：必须支持回滚
+
+**推荐**: ✅ "Create a merge commit"
+- 保留 PR 完整提交历史
+- 回滚简单
+
+**禁用**: ❌ "Rebase and Merge"
+- 难以回滚
+- 回滚操作危险
+
+---
+
+## 📖 使用指南
+
+### PR 作者
+
+```bash
+# 场景1: 小改动（文档/注释）
+# 只需要通过 Tier 1 快速检查即可
+
+# 场景2: 性能相关改动
+/run-spec                    # 标准性能测试
+/run-spec spec06-1.0c        # 完整覆盖率测试
+
+# 或者把当前分支改名为*-perf, 这样每次push 会自动运行v3 的性能。
+```
+
+### 维护者
+
+1. 检查 Tier 1 快速检查结果
+2. 对于性能敏感的 PR，评论 `/run-spec`
+3. 审查代码和性能影响
+4. 合入后监控 Tier 2 测试
+5. 如发现失败，立即回滚
+
+---
+
+## 🤖 性能评论机器人
+
+**位置**: `https://github.com/OpenXiangShan/env-scripts/blob/main/github/actions_gem5.py`
+
+**运行**:
+```bash
+python actions_gem5.py --token <github-token> --always-on
+
+# 可以联系yanyue 来重新触发机器人
+```
+
+**兼容性**: 完全兼容新的分层 CI
+
+---
+
+## 🎯 设计原则
+
+- **DRY**: 测试不重复，配置单一来源
+- **KISS**: 简化 workflow，最小化复杂度
+- **Fail Fast**: PR 阶段快速发现问题
+- **Separation of Concerns**: 快速检查 vs 完整验证
+
+---
+
+## 📚 相关文件
+
+- `.github/workflows/pr-quick-check.yml` - Tier 1
+- `.github/workflows/on-demand-spec.yml` - Tier 1.5
+- `.github/workflows/gem5-perf-template.yml` - 性能测试模板
+- `.github/workflows/gem5.yml` - Tier 2 功能测试
+- `env-scripts/github/actions_gem5.py` - 性能评论机器人
+
+---
+
+## 💡 常见问题
+
+**Q: 为什么 PR 不再自动运行性能测试？**
+A: 性能测试耗时长，会拖慢 PR 审查。现在改为按需触发，既节省资源，又保持灵活性。
+
+**Q: 如何触发性能测试？**
+A: 在 PR 评论中输入 `/run-spec [可选benchmark类型]`
+
+**Q: 新增 benchmark 类型需要修改哪些文件？**
+A: 只需修改 `gem5-perf-template.yml`
+
+---
+
+## 🎉 总结
+
+分层 CI 架构核心价值：
+
+1. **开发效率提升 95%+**：PR 反馈从 2-4 小时降至 5-10 分钟
+2. **资源优化**：性能测试按需运行
+3. **灵活性**：支持多种 benchmark 类型
+4. **主线稳定**：Post-Merge 完整测试确保质量
+5. **易于维护**：集中管理配置，遵循 DRY 和 KISS 原则
+6. **易于回滚**：保留 merge commit，回滚简单安全

.github/workflows/gem5-ideal-btb-perf-nosc.yml

@@ -1,10 +1,7 @@
-name: gem5 Ideal BTB Performance Test (no SC)
+name: gem5 Ideal BTB Performance Test (no SC) (Tier 2 - Post-Merge)
 
 on:
-  push:
-    branches: [ xs-dev]
-  pull_request:
-    branches: [ xs-dev ]
+  # only can be triggered manually
   workflow_dispatch:
     inputs:
       branch_name:

.github/workflows/gem5-ideal-btb-perf.yml

@@ -1,10 +1,12 @@
 name: gem5 Ideal BTB Performance Test
 
 on:
+  # Tier 2: Post-Merge testing
   push:
-    branches: [ xs-dev, '*-perf' ]  # xs-dev for normal CI, *-perf for BTB-only performance testing
-  pull_request:
-    branches: [ xs-dev ]
+    branches: [ xs-dev, '*-perf' ]  # xs-dev for post-merge CI, *-perf for auto performance testing
+  # Removed pull_request trigger - use /run-spec for on-demand testing
+  
+  # Support manual trigger
   workflow_dispatch:
     inputs:
       branch_name:

.github/workflows/gem5-perf-template.yml

@@ -211,9 +211,20 @@ jobs:
           
           echo "Performance data archived successfully"
           echo "Archive size: $(du -sh "$TARGET_DIR/spec_all.tar.gz" | cut -f1)"
-      - name: Upload score
+      - name: Upload score artifact
         if: always()
+        continue-on-error: true
         uses: actions/upload-artifact@v4
         with:
           name: ${{ steps.config.outputs.artifact_name }}
-          path: ${{ github.workspace }}/score.txt 
+          path: score.txt
+          retention-days: 30
+      
+      - name: Fallback - Save score to shared storage
+        if: always()
+        run: |
+          # Fallback: 如果 artifact upload 失败，保存到共享存储
+          ARTIFACT_DIR="/nfs/home/share/gem5_ci/artifacts/${{ github.run_id }}"
+          mkdir -p "$ARTIFACT_DIR"
+          cp score.txt "$ARTIFACT_DIR/${{ steps.config.outputs.artifact_name }}.txt" 2>/dev/null || true
+          echo "✅ Score also saved to: $ARTIFACT_DIR"

.github/workflows/gem5-perf.yml

@@ -1,10 +1,9 @@
-name: gem5 Performance Test
+name: gem5 Performance Test (Tier 2 - Post-Merge)
 
 on:
   push:
     branches: [ xs-dev ]
-  pull_request:
-    branches: [ xs-dev ]
+  # Removed pull_request trigger - use /run-spec for on-demand testing
 
 jobs:
   perf_test:

.github/workflows/gem5-vector.yml

@@ -1,10 +1,9 @@
-name: gem5 vector Test
+name: gem5 Vector Test (Tier 2 - Post-Merge)
 
 on:
   push:
     branches: [ xs-dev ]
-  pull_request:
-    branches: [ xs-dev ]
+  # Removed pull_request trigger
 
 jobs:
   vector-test:

.github/workflows/gem5.yml

@@ -1,10 +1,9 @@
-name: gem5 Test
+name: gem5 Full Test (Tier 2 - Post-Merge)
 
 on:
   push:
     branches: [ xs-dev ]
-  pull_request:
-    branches: [ xs-dev ]
+  # Removed pull_request trigger - moved to pr-quick-check.yml (Tier 1)
 
 jobs:
   paralel_cpt_test:
@@ -149,35 +148,6 @@ jobs:
           cd $GEM5_HOME/util/xs_scripts/test_multi_core
           bash ../kmh-ruby-dual.sh /nfs/home/share/gem5_ci/checkpoints/multi_core_test.gz
 
-  difftest_check:
-    runs-on: [self-hosted, open]
-    continue-on-error: false
-    name: XS-GEM5 - Check difftest
-    steps:
-      - uses: actions/checkout@v2
-      - name: Build DRAMSim
-        run: |
-          export GEM5_HOME=$(pwd)
-          cd ext/dramsim3
-          git clone https://github.com/umd-memsys/DRAMsim3.git DRAMsim3
-          cd DRAMsim3 && mkdir -p build
-          cd build
-          cmake ..
-          make -j 48
-          cd $GEM5_HOME
-      - name: Build GEM5 debug
-        run: CC=clang CXX=clang++ scons build/RISCV/gem5.opt -j 48 --gold-linker
-      - name: difftest check
-        run: |
-          export GCBV_REF_SO="/nfs/home/share/gem5_ci/ref/error/riscv64-nemu-interpreter-so"
-          export GCB_RESTORER="/nfs/home/share/gem5_ci/tools/normal-gcb-restorer.bin"
-          export GEM5_HOME=$(pwd)
-          mkdir -p $GEM5_HOME/util/xs_scripts/test
-          cd $GEM5_HOME/util/xs_scripts/test
-          bash ../kmh_6wide.sh /nfs/home/share/gem5_ci/checkpoints/gcb_test.zstd 2>log.txt || exit_code=$?
-          if [ ${exit_code} -eq 0 ]; then echo "Difftest is broken, it should report error!" exit 1; fi
-          match=$(grep ".*Difftest failed!.*" log.txt -c)
-          if [ ${match} -eq 0 ]; then echo "Difftest is broken, it should report at least one agnostic related difference!" exit 1; fi
 
   test_fix_l2tlb_bugs:
     runs-on: [self-hosted, open]
@@ -233,13 +203,4 @@ jobs:
           cd $GEM5_HOME/util/xs_scripts/test_h
           bash ../kmh_6wide_h.sh /nfs/home/share/gem5_ci/checkpoints/gcbh_test.zstd
 
-  unit_tests:
-    runs-on: [self-hosted, open]
-    continue-on-error: false
-    name: XS-GEM5 - Unit Tests
-    steps:
-      - uses: actions/checkout@v2
-      - name: Build unit tests and run them
-        run: |
-          CC=gcc CXX=g++ scons build/RISCV/unittests.opt -j100 --unit-test