From 2cfc7190a2b84a5cf8da6be33da29ed4f2d1aa1d Mon Sep 17 00:00:00 2001
From: XYenChi <oriachiuan@gmail.com>
Date: Sun, 19 Apr 2026 23:32:28 +0800
Subject: [PATCH 01/16] Add RISC-V Blocklist (#1)

* Add RISC-V 64 BLOCK_LIST

* Skip long time testcase
---
 test/run_test.py    | 47 +++++++++++++++++++++++++++++++++++++++++++++
 test/test_linalg.py |  1 +
 third_party/kineto  |  2 +-
 3 files changed, 49 insertions(+), 1 deletion(-)

diff --git a/test/run_test.py b/test/run_test.py
index 9a85ecd2195cd..c753d18ab7c6c 100755
--- a/test/run_test.py
+++ b/test/run_test.py
@@ -110,6 +110,7 @@ def upload_adhoc_failure_json(*args, **kwargs):
 INDUCTOR_TEST_PREFIX = "inductor"
 IS_SLOW = "slow" in TEST_CONFIG or "slow" in BUILD_ENVIRONMENT
 IS_S390X = platform.machine() == "s390x"
+IS_RISCV64 = platform.machine() == "riscv64"
 
 
 # Note [ROCm parallel CI testing]
@@ -285,6 +286,45 @@ def __contains__(self, item):
     "test_xpu",
 ]
 
+RISCV64_BLOCKLIST = [
+    # disable distributed related test
+    "inductor/test_distributed_patterns"
+    "fx/test_dce_pass"
+    "export/test_cpp_serdes"
+    "export/test_export"
+    "export/test_export_strict"
+    "export/test_export_training_ir_to_run_decomp"
+    "export/test_retraceability"
+    "export/test_serdes"
+    "export/test_strict_export_v2"
+    "test_public_bindings"
+    # quantized engine NoQEngine is not supported
+    "test_torch"
+    "ao/sparsity/test_composability"
+    # QNNPACK is not supported
+    "export/test_converter"
+    # record_contex_cpp is not support on non-linux non-x86_64 platforms
+    "torch_np/numpy_tests/core/test_numeric"
+    # Failed to import torch.distributed.run: cannot import name 'Store' from 'torch.distributed'
+    "test_testing"
+    # TODO:L1 cache size = 0, need to fix
+    "inductor/test_cpu_select_algorithm"
+    "inductor/test_aot_inductor_arrayref"
+    "inductor/test_cpu_repro"
+    # TODO:scalar value not equal, need to fix
+    "profiler/test_profiler"
+    # TODO precision
+    "test_binary_ufuncs"
+    "test_decomp"
+    # TODO no CUDA related module
+    "quantization/core/test_workflow_module"  # TestFakeQuantize.test_fq_module_per_channel
+    "quantization/core/test_workflow_ops"
+    "quantization/core/test_quantized_op"
+    # z3-solver build fail
+    "test_proxy_tensor"
+]
+
+
 # The tests inside these files should never be run in parallel with each other
 RUN_PARALLEL_BLOCKLIST = [
     "test_extension_utils",
@@ -1875,6 +1915,13 @@ def get_selected_tests(options) -> list[str]:
             selected_tests,
             "Skip distributed tests on s390x",
         )
+    elif IS_RISCV64:
+        selected_tests = exclude_tests(RISCV64_BLOCKLIST, selected_tests, "on riscv64")
+        selected_tests = exclude_tests(
+            DISTRIBUTED_TESTS,
+            selected_tests,
+            "Skip distributed tests on riscv64",
+        )
 
     # skip all distributed tests if distributed package is not available.
     if not dist.is_available():
diff --git a/test/test_linalg.py b/test/test_linalg.py
index 0ac35e077946d..5e7018e1cf9ab 100644
--- a/test/test_linalg.py
+++ b/test/test_linalg.py
@@ -9002,6 +9002,7 @@ def test_matrix_exp_backward_input_validation(self, device, dtype):
         with self.assertRaisesRegex(RuntimeError, "must be batches of square matrices"):
             torch.ops.aten.matrix_exp_backward(non_square, grad_non_square)
 
+    @slowTest
     @skipCUDAIfNoMagmaAndNoLinalgsolver
     @skipCPUIfNoLapack
     @dtypes(torch.float, torch.double, torch.complex64, torch.complex128)
diff --git a/third_party/kineto b/third_party/kineto
index 5902263f641a8..7739225509b84 160000
--- a/third_party/kineto
+++ b/third_party/kineto
@@ -1 +1 @@
-Subproject commit 5902263f641a8634af077031fb4befc7162cefbb
+Subproject commit 7739225509b847e7b1ce7638f1ead383d15b077f

From bf9cf6d4f77057f64d6bbf721b9d31056b913f6e Mon Sep 17 00:00:00 2001
From: Bo YU <tsu.yubo@gmail.com>
Date: Wed, 22 Apr 2026 08:10:52 +0000
Subject: [PATCH 02/16] add riscv64 ci

---
 .github/workflows/ci-riscv64.yml | 99 ++++++++++++++++++++++++++++++++
 1 file changed, 99 insertions(+)
 create mode 100644 .github/workflows/ci-riscv64.yml

diff --git a/.github/workflows/ci-riscv64.yml b/.github/workflows/ci-riscv64.yml
new file mode 100644
index 0000000000000..ffc859265b3c9
--- /dev/null
+++ b/.github/workflows/ci-riscv64.yml
@@ -0,0 +1,99 @@
+# Note: this runner is provided externally, so we minimize its access to
+# secrets.
+on:
+  push:
+    branches: [riscv]
+
+  pull_request_target:
+    types: [opened, synchronize, reopened]
+  
+
+name: CI (riscv64)
+
+permissions:
+  contents: read
+  # No permissions to secrets.
+
+concurrency:
+  group: ${{ github.workflow }}-${{ github.head_ref || github.run_id }}
+  cancel-in-progress: true
+
+# FIXME: Drop this
+env:
+  RUSTFLAGS: -D warnings
+  CARGO_TERM_COLOR: always
+
+jobs:
+  build:
+    name: Build and test
+    runs-on: [self-hosted, linux, amd64]
+    # This is in its own separate environment.
+    environment: riscv64
+    steps:
+      - name: Checkout repo
+        uses: actions/checkout@v4
+        with:
+          fetch-depth: 0   # merge-base
+
+      - name: Extract PR info
+        run: |
+          echo "BASE_SHA=${{ github.event.pull_request.base.sha }}" >> $GITHUB_ENV
+          echo "HEAD_SHA=${{ github.event.pull_request.head.sha }}" >> GITHUB_ENV
+          echo "PR_NUMBER=${{ github.event.pull_request.number }}" >> GITHUB_ENV
+
+      - name: Diff base and head
+        run: |
+          if [ "${{ github.event_name }}" = "pull_request" ]; then
+          echo "Push PR build"
+            BASE_REF="${{ github.base_ref }}"
+            HEAD_REF="${{ github.head_ref }}"
+
+            echo "Base ref: $BASE_REF"
+            echo "Head ref: $HEAD_REF"
+
+            # 强约束: PR 必须基于 riscv
+            if [ "$BASE_REF" != "riscv" ]; then
+              echo "ERROR: PR must target 'riscv' branch, got '$BASE_REF'"
+              exit 1
+            fi
+
+            BASE="$BASE_SHA"
+            HEAD="$HEAD_SHA"
+          else
+            echo "Push to riscv"
+            # 统一用 riscv 作为 baseline
+            git fetch origin riscv
+
+            BASE=$(git merge-base HEAD origin/main)
+            HEAD=$(git rev-parse HEAD)
+
+          fi
+
+          echo "BASE_COMMIT=$BASE" >> $GITHUB_ENV
+          echo "HEAD_COMMIT=$HEAD" >> $GITHUB_ENV
+
+          echo "Base: $BASE"
+          echo "Head: $HEAD"
+
+      - name: Generate patch
+        run: |
+          echo "Generating patch..."
+
+          git diff $BASE_COMMIT $HEAD_COMMIT > patch.diff
+
+          echo "Patch size:"
+          wc -l patch.diff
+          cat patch.diff
+
+          # 可选：避免空 patch
+          if [ ! -s patch.diff ]; then
+            echo "Warning: empty patch"
+          fi
+
+      - name: Trigger Jenkins Job
+        run: |
+          export BASE_COMMIT=${BASE_COMMIT}
+          export PATCH_FILE=$(pwd)/patch.diff
+          export GITHUB_PR=${PR_NUMBER:-0}
+
+          #bash /home/jenkins/scripts/jenkins-run.sh

From 6c3b1f17c77613e59e630139d849d193e63ca8e9 Mon Sep 17 00:00:00 2001
From: vimer <yuzibode@126.com>
Date: Fri, 24 Apr 2026 14:58:31 +0800
Subject: [PATCH 03/16] Test ci with PR (#8)

* Add riscv64 ci with PR
---
 .github/workflows/ci-riscv64.yml | 13 ++++++++-----
 1 file changed, 8 insertions(+), 5 deletions(-)

diff --git a/.github/workflows/ci-riscv64.yml b/.github/workflows/ci-riscv64.yml
index ffc859265b3c9..c14ce1e4c3efa 100644
--- a/.github/workflows/ci-riscv64.yml
+++ b/.github/workflows/ci-riscv64.yml
@@ -6,7 +6,6 @@ on:
 
   pull_request_target:
     types: [opened, synchronize, reopened]
-  
 
 name: CI (riscv64)
 
@@ -33,7 +32,7 @@ jobs:
       - name: Checkout repo
         uses: actions/checkout@v4
         with:
-          fetch-depth: 0   # merge-base
+          fetch-depth: 3000   # shadow clone?
 
       - name: Extract PR info
         run: |
@@ -44,7 +43,7 @@ jobs:
       - name: Diff base and head
         run: |
           if [ "${{ github.event_name }}" = "pull_request" ]; then
-          echo "Push PR build"
+            echo "Push PR build"
             BASE_REF="${{ github.base_ref }}"
             HEAD_REF="${{ github.head_ref }}"
 
@@ -57,8 +56,12 @@ jobs:
               exit 1
             fi
 
-            BASE="$BASE_SHA"
-            HEAD="$HEAD_SHA"
+            // need to get contents of the PR
+            git fetch origin pull/${{ github.event.pull_request.number }}/head:pr-head
+            git fetch origin pull/${{ github.event.pull_request.number }}/base:pr-base
+
+            BASE=$(git merge-base pr-base pr-head)
+            HEAD=pr-head
           else
             echo "Push to riscv"
             # 统一用 riscv 作为 baseline

From ff108da766ac4fc65e3b1ebb7ea97dc5b3a3eb34 Mon Sep 17 00:00:00 2001
From: Bo YU <tsu.yubo@gmail.com>
Date: Fri, 24 Apr 2026 07:06:46 +0000
Subject: [PATCH 04/16] Fix no main brach issue
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

⭐ Run Main Diff base and head
Push to riscv
From https://github.com/RuyiAI-Stack/pytorch
 * branch              riscv      -> FETCH_HEAD
fatal: Not a valid object name origin/main
Error:   ❌  Failure - Main Diff base and head
Error: exit status 128
---
 .github/workflows/ci-riscv64.yml | 23 +++++++++++------------
 1 file changed, 11 insertions(+), 12 deletions(-)

diff --git a/.github/workflows/ci-riscv64.yml b/.github/workflows/ci-riscv64.yml
index c14ce1e4c3efa..3b3ddd23483e7 100644
--- a/.github/workflows/ci-riscv64.yml
+++ b/.github/workflows/ci-riscv64.yml
@@ -33,16 +33,16 @@ jobs:
         uses: actions/checkout@v4
         with:
           fetch-depth: 3000   # shadow clone?
+          ref: ${{ github.sha }} # including latest sha
 
       - name: Extract PR info
         run: |
           echo "BASE_SHA=${{ github.event.pull_request.base.sha }}" >> $GITHUB_ENV
           echo "HEAD_SHA=${{ github.event.pull_request.head.sha }}" >> GITHUB_ENV
-          echo "PR_NUMBER=${{ github.event.pull_request.number }}" >> GITHUB_ENV
-
+         
       - name: Diff base and head
         run: |
-          if [ "${{ github.event_name }}" = "pull_request" ]; then
+          if [[ "${{ github.event_name }}" = "pull_request" || "${{ github.event_name }}" == "pull_request_target" ]]; then
             echo "Push PR build"
             BASE_REF="${{ github.base_ref }}"
             HEAD_REF="${{ github.head_ref }}"
@@ -56,19 +56,19 @@ jobs:
               exit 1
             fi
 
-            // need to get contents of the PR
+            # need to get contents of the PR
             git fetch origin pull/${{ github.event.pull_request.number }}/head:pr-head
-            git fetch origin pull/${{ github.event.pull_request.number }}/base:pr-base
-
-            BASE=$(git merge-base pr-base pr-head)
+            git fetch origin main
+            BASE=$(git merge-base pr-head origin/main)
             HEAD=pr-head
           else
             echo "Push to riscv"
             # 统一用 riscv 作为 baseline
-            git fetch origin riscv
+            git fetch origin main
+            #git fetch origin riscv
 
-            BASE=$(git merge-base HEAD origin/main)
-            HEAD=$(git rev-parse HEAD)
+            BASE=$(git merge-base ${{ github.sha }} origin/main) # The latest commit
+            HEAD=${{ github.sha }}
 
           fi
 
@@ -97,6 +97,5 @@ jobs:
         run: |
           export BASE_COMMIT=${BASE_COMMIT}
           export PATCH_FILE=$(pwd)/patch.diff
-          export GITHUB_PR=${PR_NUMBER:-0}
 
-          #bash /home/jenkins/scripts/jenkins-run.sh
+          bash /home/jenkins/scripts/jenkins-run.sh

From 47c19e55efd0e575563bd56133031f18a3ab1777 Mon Sep 17 00:00:00 2001
From: Bo YU <tsu.yubo@gmail.com>
Date: Sat, 25 Apr 2026 14:21:06 +0000
Subject: [PATCH 05/16] move the patch to dest

---
 .github/workflows/ci-riscv64.yml | 18 +++++++++---------
 1 file changed, 9 insertions(+), 9 deletions(-)

diff --git a/.github/workflows/ci-riscv64.yml b/.github/workflows/ci-riscv64.yml
index 3b3ddd23483e7..dec68a6d12c16 100644
--- a/.github/workflows/ci-riscv64.yml
+++ b/.github/workflows/ci-riscv64.yml
@@ -39,7 +39,7 @@ jobs:
         run: |
           echo "BASE_SHA=${{ github.event.pull_request.base.sha }}" >> $GITHUB_ENV
           echo "HEAD_SHA=${{ github.event.pull_request.head.sha }}" >> GITHUB_ENV
-         
+
       - name: Diff base and head
         run: |
           if [[ "${{ github.event_name }}" = "pull_request" || "${{ github.event_name }}" == "pull_request_target" ]]; then
@@ -82,20 +82,20 @@ jobs:
         run: |
           echo "Generating patch..."
 
-          git diff $BASE_COMMIT $HEAD_COMMIT > patch.diff
+          SHORT_HEAD=${HEAD_COMMIT:0:7}
+          PATCH_NAME="patch_${SHORT_HEAD}.patch"
+
+          git diff $BASE_COMMIT $HEAD_COMMIT > $PATCH_NAME
 
           echo "Patch size:"
-          wc -l patch.diff
-          cat patch.diff
+          wc -l $PATCH_NAME
 
-          # 可选：避免空 patch
-          if [ ! -s patch.diff ]; then
-            echo "Warning: empty patch"
-          fi
+          cp $PATCH_NAME /home/jenkins/patch/
+          cat /home/jenkins/patch/$PATCH_NAME
 
       - name: Trigger Jenkins Job
         run: |
           export BASE_COMMIT=${BASE_COMMIT}
           export PATCH_FILE=$(pwd)/patch.diff
 
-          bash /home/jenkins/scripts/jenkins-run.sh
+          #bash /home/jenkins/scripts/jenkins-run.sh

From b984cb258fc01ec8e2c4b0da45708c9bd0d9f307 Mon Sep 17 00:00:00 2001
From: XYenChi <oriachiuan@gmail.com>
Date: Mon, 27 Apr 2026 10:59:07 +0800
Subject: [PATCH 06/16] Fix block list format and remove
 test_cpu_select_algorithm (#4)

* mklnn is unavailable on RISC-V

* Remove test_cpu_select_algorithm from block_list

* Fix block list format
---
 test/inductor/test_cpu_select_algorithm.py |  2 +
 test/run_test.py                           | 52 +++++++++++-----------
 2 files changed, 28 insertions(+), 26 deletions(-)

diff --git a/test/inductor/test_cpu_select_algorithm.py b/test/inductor/test_cpu_select_algorithm.py
index 236712f7950bb..8c064db44220e 100644
--- a/test/inductor/test_cpu_select_algorithm.py
+++ b/test/inductor/test_cpu_select_algorithm.py
@@ -1583,6 +1583,7 @@ def forward(self, x):
         vec_amx = VecAMX()
         self._check_amx_counter(vec_amx)
 
+    @unittest.skipIf(not torch._C._has_mkldnn, "MKLDNN is not enabled")
     @inductor_config.patch({"freezing": True})
     @patches
     @torch.no_grad
@@ -1700,6 +1701,7 @@ def forward(self, x, scale):
             vec_amx = VecAMX()
             self._check_amx_counter(vec_amx)
 
+    @unittest.skipIf(not torch._C._has_mkldnn, "MKLDNN is not enabled")
     @inductor_config.patch({"freezing": True, "cpp.enable_concat_linear": True})
     @patches
     @torch.no_grad
diff --git a/test/run_test.py b/test/run_test.py
index c753d18ab7c6c..4939b685e0163 100755
--- a/test/run_test.py
+++ b/test/run_test.py
@@ -288,40 +288,40 @@ def __contains__(self, item):
 
 RISCV64_BLOCKLIST = [
     # disable distributed related test
-    "inductor/test_distributed_patterns"
-    "fx/test_dce_pass"
-    "export/test_cpp_serdes"
-    "export/test_export"
-    "export/test_export_strict"
-    "export/test_export_training_ir_to_run_decomp"
-    "export/test_retraceability"
-    "export/test_serdes"
-    "export/test_strict_export_v2"
-    "test_public_bindings"
+    "inductor/test_distributed_patterns",
+    "fx/test_dce_pass",
+    "export/test_cpp_serdes",
+    "export/test_export",
+    "export/test_export_strict",
+    "export/test_export_training_ir_to_run_decomp",
+    "export/test_retraceability",
+    "export/test_serdes",
+    "export/test_strict_export_v2",
+    "test_public_bindings",
     # quantized engine NoQEngine is not supported
-    "test_torch"
-    "ao/sparsity/test_composability"
+    "test_torch",
+    "ao/sparsity/test_composability",
     # QNNPACK is not supported
-    "export/test_converter"
+    "export/test_converter",
     # record_contex_cpp is not support on non-linux non-x86_64 platforms
-    "torch_np/numpy_tests/core/test_numeric"
+    "torch_np/numpy_tests/core/test_numeric",
     # Failed to import torch.distributed.run: cannot import name 'Store' from 'torch.distributed'
-    "test_testing"
-    # TODO:L1 cache size = 0, need to fix
-    "inductor/test_cpu_select_algorithm"
-    "inductor/test_aot_inductor_arrayref"
-    "inductor/test_cpu_repro"
+    "test_testing",
+    "inductor/test_aot_inductor_arrayref",
+    "inductor/test_cpu_repro",
+    # TODO: mkldnn not available, shape guard failures on RISC-V
+    "inductor/test_cpu_select_algorithm",
     # TODO:scalar value not equal, need to fix
-    "profiler/test_profiler"
+    "profiler/test_profiler",
     # TODO precision
-    "test_binary_ufuncs"
-    "test_decomp"
+    "test_binary_ufuncs",
+    "test_decomp",
     # TODO no CUDA related module
-    "quantization/core/test_workflow_module"  # TestFakeQuantize.test_fq_module_per_channel
-    "quantization/core/test_workflow_ops"
-    "quantization/core/test_quantized_op"
+    "quantization/core/test_workflow_module",  # TestFakeQuantize.test_fq_module_per_channel
+    "quantization/core/test_workflow_ops",
+    "quantization/core/test_quantized_op",
     # z3-solver build fail
-    "test_proxy_tensor"
+    "test_proxy_tensor",
 ]
 
 

From f7e97ba3b6971ed3d66789352759333d106efcb3 Mon Sep 17 00:00:00 2001
From: Bo YU <tsu.yubo@gmail.com>
Date: Sat, 25 Apr 2026 14:21:06 +0000
Subject: [PATCH 07/16] move the patch to dest

---
 .github/workflows/ci-riscv64.yml | 13 ++++++-------
 1 file changed, 6 insertions(+), 7 deletions(-)

diff --git a/.github/workflows/ci-riscv64.yml b/.github/workflows/ci-riscv64.yml
index dec68a6d12c16..d8cd1ecdd8de3 100644
--- a/.github/workflows/ci-riscv64.yml
+++ b/.github/workflows/ci-riscv64.yml
@@ -57,14 +57,14 @@ jobs:
             fi
 
             # need to get contents of the PR
-            git fetch origin pull/${{ github.event.pull_request.number }}/head:pr-head
-            git fetch origin main
+            git fetch --quiet origin pull/${{ github.event.pull_request.number }}/head:pr-head
+            git fetch --quiet origin main
             BASE=$(git merge-base pr-head origin/main)
             HEAD=pr-head
           else
             echo "Push to riscv"
             # 统一用 riscv 作为 baseline
-            git fetch origin main
+            git fetch --quiet origin main
             #git fetch origin riscv
 
             BASE=$(git merge-base ${{ github.sha }} origin/main) # The latest commit
@@ -93,9 +93,8 @@ jobs:
           cp $PATCH_NAME /home/jenkins/patch/
           cat /home/jenkins/patch/$PATCH_NAME
 
+          echo "PATCH_FILE=$PATCH_NAME" >> $GITHUB_ENV
+
       - name: Trigger Jenkins Job
         run: |
-          export BASE_COMMIT=${BASE_COMMIT}
-          export PATCH_FILE=$(pwd)/patch.diff
-
-          #bash /home/jenkins/scripts/jenkins-run.sh
+          bash /home/jenkins/scripts/jenkins-run.sh $BASE_COMMIT  $PATCH_FILE

From 89901fb309dd78f6c8a09ccdc176f701e3653bce Mon Sep 17 00:00:00 2001
From: vimer <yuzibode@126.com>
Date: Sun, 3 May 2026 09:05:59 +0800
Subject: [PATCH 08/16] [blacklist]: update it (#11)

These cases are too slow on riscv64, adding them to here simply

Drop test_torch from the list because it is one core case
---
 test/run_test.py | 11 +++++++++--
 1 file changed, 9 insertions(+), 2 deletions(-)

diff --git a/test/run_test.py b/test/run_test.py
index 4939b685e0163..22561c879d13d 100755
--- a/test/run_test.py
+++ b/test/run_test.py
@@ -298,8 +298,6 @@ def __contains__(self, item):
     "export/test_serdes",
     "export/test_strict_export_v2",
     "test_public_bindings",
-    # quantized engine NoQEngine is not supported
-    "test_torch",
     "ao/sparsity/test_composability",
     # QNNPACK is not supported
     "export/test_converter",
@@ -322,6 +320,15 @@ def __contains__(self, item):
     "quantization/core/test_quantized_op",
     # z3-solver build fail
     "test_proxy_tensor",
+    # too slow on riscv64
+    # 53013.55 s
+    "functorch/test_aotdispatch",
+    # 25069 s
+    "functorch/test_ops",
+    # 17528 s
+    "test_transformers",
+    # 10897 s
+    "functorch/test_vmap",
 ]
 
 

From af405abae3ab26218aaf3111f1f9835dcabf7a5d Mon Sep 17 00:00:00 2001
From: vimer <yuzibode@126.com>
Date: Thu, 7 May 2026 12:32:52 +0800
Subject: [PATCH 09/16] Use commit sha on PR workflow (#12)

---
 .github/workflows/ci-riscv64.yml | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/.github/workflows/ci-riscv64.yml b/.github/workflows/ci-riscv64.yml
index d8cd1ecdd8de3..44cd575a01e01 100644
--- a/.github/workflows/ci-riscv64.yml
+++ b/.github/workflows/ci-riscv64.yml
@@ -60,7 +60,7 @@ jobs:
             git fetch --quiet origin pull/${{ github.event.pull_request.number }}/head:pr-head
             git fetch --quiet origin main
             BASE=$(git merge-base pr-head origin/main)
-            HEAD=pr-head
+            HEAD=$(git rev-parse pr-head)
           else
             echo "Push to riscv"
             # 统一用 riscv 作为 baseline

From 9fb76e6fd981bf5938600c15c6311e6a982b9a48 Mon Sep 17 00:00:00 2001
From: Yixuan Chen <chenyixuan@iscas.ac.cn>
Date: Sun, 26 Apr 2026 00:49:38 +0800
Subject: [PATCH 10/16] Fix bytes_to_scalar for float/complex on RISC-V

bytes_to_scalar previously round-tripped raw bytes through Python
float/complex values (via ctypes) before constructing the tensor. This
loses NaN bit patterns on architectures (such as RISC-V) that
canonicalize NaNs in floating-point loads/conversions, causing
test_bytes_to_scalar_cpu_{float32,float64,complex64,complex128} to
fail with mismatched storage bytes.

Construct the scalar tensor by writing the raw bytes directly into its
untyped storage so all input bit patterns (including NaN payloads) are
preserved exactly.
---
 torch/testing/_internal/common_utils.py | 37 +++++++++++--------------
 1 file changed, 16 insertions(+), 21 deletions(-)

diff --git a/torch/testing/_internal/common_utils.py b/torch/testing/_internal/common_utils.py
index 608c8ee4b84b3..e50f2f00c9bd4 100644
--- a/torch/testing/_internal/common_utils.py
+++ b/torch/testing/_internal/common_utils.py
@@ -5668,27 +5668,22 @@ def check_bytes(byte_list):
             if not (0 <= byte <= 255):
                 raise AssertionError(f"byte value out of range: expected 0 <= byte <= 255, got {byte}")
 
-    if dtype.is_complex:
-        if len(byte_list) != (num_bytes * 2):
-            raise AssertionError(
-                f"expected len(byte_list) == {num_bytes * 2} for complex dtype, got {len(byte_list)}"
-            )
-        check_bytes(byte_list)
-        real = ctype.from_buffer((ctypes.c_byte * num_bytes)(
-            *byte_list[:num_bytes])).value
-        imag = ctype.from_buffer((ctypes.c_byte * num_bytes)(
-            *byte_list[num_bytes:])).value
-        res = real + 1j * imag
-    else:
-        if len(byte_list) != num_bytes:
-            raise AssertionError(
-                f"expected len(byte_list) == {num_bytes}, got {len(byte_list)}"
-            )
-        check_bytes(byte_list)
-        res = ctype.from_buffer((ctypes.c_byte * num_bytes)(
-            *byte_list)).value
-
-    return torch.tensor(res, device=device, dtype=dtype)
+    expected_len = num_bytes * 2 if dtype.is_complex else num_bytes
+    if len(byte_list) != expected_len:
+        raise AssertionError(
+            f"expected len(byte_list) == {expected_len}"
+            f"{' for complex dtype' if dtype.is_complex else ''}, got {len(byte_list)}"
+        )
+    check_bytes(byte_list)
+
+    # Write bytes directly into storage to preserve exact bit patterns
+    # (e.g. NaN payloads, which are not preserved when round-tripping through
+    # Python float/complex, especially on architectures like RISC-V that
+    # canonicalize NaNs).
+    res = torch.empty((), dtype=dtype, device=device)
+    src = torch.tensor(byte_list, dtype=torch.uint8, device=device)
+    res.untyped_storage().copy_(src.untyped_storage())
+    return res
 
 
 def copy_func(f):

From 08d645aaefd14905891180c0b920a938bc593127 Mon Sep 17 00:00:00 2001
From: Yixuan Chen <chenyixuan@iscas.ac.cn>
Date: Wed, 22 Apr 2026 15:27:45 +0800
Subject: [PATCH 11/16] Fix test_float_to_int_conversion_nonfinite for RISC-V

RISC-V converts non-finite floats to integers by saturating:
-inf -> min, inf/nan -> max for wider int types.
Add IS_RISCV64 flag and RISC-V-specific reference values.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
---
 test/test_tensor_creation_ops.py        | 8 ++++++++
 torch/testing/_internal/common_utils.py | 1 +
 2 files changed, 9 insertions(+)

diff --git a/test/test_tensor_creation_ops.py b/test/test_tensor_creation_ops.py
index b9e8e537b53cb..99396cea12f5b 100644
--- a/test/test_tensor_creation_ops.py
+++ b/test/test_tensor_creation_ops.py
@@ -32,6 +32,7 @@
     IS_SANDCASTLE,
     IS_S390X,
     IS_ARM64,
+    IS_RISCV64,
     parametrize,
     TEST_WITH_TORCHDYNAMO,
     xfailIfTorchDynamo,
@@ -1109,6 +1110,13 @@ def test_float_to_int_conversion_nonfinite(self, device, dtype):
 
         if dtype == torch.bool:
             refs = (True, True, True)
+        elif IS_RISCV64:
+            if dtype in (torch.int32, torch.int64):
+                refs = (torch.iinfo(dtype).min, torch.iinfo(dtype).max, torch.iinfo(dtype).max)
+            elif dtype == torch.uint8:
+                refs = (0, torch.iinfo(dtype).max, torch.iinfo(dtype).max)
+            elif dtype in (torch.int8, torch.int16):
+                refs = (0, -1, -1)
         elif IS_ARM64:
             refs = (torch.iinfo(dtype).min, torch.iinfo(dtype).max, 0)
             if dtype in (torch.int8, torch.int16):
diff --git a/torch/testing/_internal/common_utils.py b/torch/testing/_internal/common_utils.py
index e50f2f00c9bd4..25e0c69804234 100644
--- a/torch/testing/_internal/common_utils.py
+++ b/torch/testing/_internal/common_utils.py
@@ -1438,6 +1438,7 @@ def printErrors(self) -> None:
 IS_PPC = platform.machine() == "ppc64le"
 IS_X86 = platform.machine() in ('x86_64', 'i386')
 IS_ARM64 = platform.machine() in ('arm64', 'aarch64', 'ARM64')
+IS_RISCV64 = platform.machine() == 'riscv64'
 IS_S390X = platform.machine() == "s390x"
 IS_AVX512_VNNI_SUPPORTED = torch.cpu.get_capabilities().get("avx512_vnni", False)
 IS_CPU_EXT_SVE_SUPPORTED = torch.cpu.get_capabilities().get("sve", False)

From 1e6be1a1af4cb7c0cde3337e2b8005adbf67c9fc Mon Sep 17 00:00:00 2001
From: Yixuan Chen <chenyixuan@iscas.ac.cn>
Date: Sat, 9 May 2026 15:59:22 +0800
Subject: [PATCH 12/16] Skip test if no qengine

---
 test/test_torch.py                      | 11 +++++++++--
 torch/testing/_internal/common_utils.py |  4 ++++
 2 files changed, 13 insertions(+), 2 deletions(-)

diff --git a/test/test_torch.py b/test/test_torch.py
index cf95adbc2c916..f93661cf386f4 100644
--- a/test/test_torch.py
+++ b/test/test_torch.py
@@ -43,7 +43,7 @@
     wrapDeterministicFlagAPITest, DeterministicGuard, CudaSyncGuard,
     bytes_to_scalar, parametrize, noncontiguous_like,
     AlwaysWarnTypedStorageRemoval, TEST_WITH_TORCHDYNAMO, xfailIfTorchDynamo,
-    xfailIfS390X, set_warn_always_context, decorateIf, isRocmArchAnyOf)
+    xfailIfS390X, xfailIfRISCV, set_warn_always_context, decorateIf, isRocmArchAnyOf)
 from multiprocessing.reduction import ForkingPickler
 from torch.testing._internal.common_device_type import (
     expectedFailureMeta,
@@ -9594,14 +9594,21 @@ def test_type(self):
 
     # FIXME: port to a quantization test suite
     @xfailIfS390X
+    @xfailIfRISCV
     def test_qengine(self):
         qengines = torch.backends.quantized.supported_engines
+        if not qengines:
+            self.skipTest("No quantized engines supported on this platform")
         original_qe = torch.backends.quantized.engine
         for qe in qengines:
             torch.backends.quantized.engine = qe
             if torch.backends.quantized.engine != qe:
                 raise AssertionError(f"qengine not set successfully: expected {qe}, got {torch.backends.quantized.engine}")
-        torch.backends.quantized.engine = original_qe
+        # On platforms where no qengine is compiled in as the default (e.g. RISC-V),
+        # the initial engine reads as "none" (NoQEngine), which is not a valid value
+        # to pass back to _set_qengine. Only restore if it was a real engine.
+        if original_qe != "none":
+            torch.backends.quantized.engine = original_qe
 
     def test_terminate_handler_on_crash(self):
         cmd = [sys.executable, '-c', "import os; os.environ[\"TORCH_CUSTOM_TERMINATE\"] ='1'; \
diff --git a/torch/testing/_internal/common_utils.py b/torch/testing/_internal/common_utils.py
index 25e0c69804234..099382a195ed0 100644
--- a/torch/testing/_internal/common_utils.py
+++ b/torch/testing/_internal/common_utils.py
@@ -1444,6 +1444,7 @@ def printErrors(self) -> None:
 IS_CPU_EXT_SVE_SUPPORTED = torch.cpu.get_capabilities().get("sve", False)
 IS_CPU_CAPABILITY_SVE = torch._C._get_cpu_capability() in ("SVE128", "SVE256")
 IS_CPU_CAPABILITY_SVE256 = torch._C._get_cpu_capability() == "SVE256"
+IS_RISCV = platform.machine() in ('riscv64', 'riscv32')
 
 if IS_WINDOWS:
     @contextmanager
@@ -2266,6 +2267,9 @@ def wrap_fn(self, *args, **kwargs):
 def xfailIfS390X(func):
     return unittest.expectedFailure(func) if IS_S390X else func
 
+def xfailIfRISCV(func):
+    return unittest.expectedFailure(func) if IS_RISCV else func
+
 def xfailIf(condition):
     def wrapper(func):
         if condition:

From a9a4cbdf36ef4f52d123acd551c2f498312f4323 Mon Sep 17 00:00:00 2001
From: XYenChi <oriachiuan@gmail.com>
Date: Wed, 20 May 2026 14:15:08 +0800
Subject: [PATCH 13/16] Replace offical cpuinfo repo (#18)

---
 .gitmodules | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/.gitmodules b/.gitmodules
index 076ce38ac7938..35feb14cec9ae 100644
--- a/.gitmodules
+++ b/.gitmodules
@@ -41,7 +41,7 @@
 [submodule "third_party/cpuinfo"]
     ignore = dirty
     path = third_party/cpuinfo
-    url = https://github.com/pytorch/cpuinfo.git
+    url = https://github.com/RuyiAI-Stack/cpuinfo.git
 [submodule "third_party/python-peachpy"]
     ignore = dirty
     path = third_party/python-peachpy

From d3d51ee2704a8b8d940c54fe7261fd1cbef3f39d Mon Sep 17 00:00:00 2001
From: vimer <yuzibode@126.com>
Date: Sun, 24 May 2026 23:34:18 +0800
Subject: [PATCH 14/16] [ci] split core and full ci (#20)

---
 .github/workflows/ci-riscv64.yml | 79 +++++++++++++++++++++++++++++---
 1 file changed, 72 insertions(+), 7 deletions(-)

diff --git a/.github/workflows/ci-riscv64.yml b/.github/workflows/ci-riscv64.yml
index 44cd575a01e01..127e0a01e36eb 100644
--- a/.github/workflows/ci-riscv64.yml
+++ b/.github/workflows/ci-riscv64.yml
@@ -1,5 +1,8 @@
 # Note: this runner is provided externally, so we minimize its access to
 # secrets.
+
+name: CI (riscv64)
+
 on:
   push:
     branches: [riscv]
@@ -7,8 +10,6 @@ on:
   pull_request_target:
     types: [opened, synchronize, reopened]
 
-name: CI (riscv64)
-
 permissions:
   contents: read
   # No permissions to secrets.
@@ -23,9 +24,17 @@ env:
   CARGO_TERM_COLOR: always
 
 jobs:
-  build:
-    name: Build and test
+  core-ci:
+    name: pytorch-riscv64-core-ci
     runs-on: [self-hosted, linux, amd64]
+
+    outputs:
+      base_commit: ${{ steps.meta.outputs.base_commit }}
+      head_commit: ${{ steps.meta.outputs.head_commit }}
+      patch_file: ${{ steps.patch.outputs.patch_file }}
+      ci_result_base_url: ${{ steps.jenkins.outputs.ci_result_base_url }}
+      ci_stat_url: ${{ steps.jenkins.outputs.ci_stat_url }}
+
     # This is in its own separate environment.
     environment: riscv64
     steps:
@@ -38,7 +47,7 @@ jobs:
       - name: Extract PR info
         run: |
           echo "BASE_SHA=${{ github.event.pull_request.base.sha }}" >> $GITHUB_ENV
-          echo "HEAD_SHA=${{ github.event.pull_request.head.sha }}" >> GITHUB_ENV
+          echo "HEAD_SHA=${{ github.event.pull_request.head.sha }}" >> $GITHUB_ENV
 
       - name: Diff base and head
         run: |
@@ -50,7 +59,7 @@ jobs:
             echo "Base ref: $BASE_REF"
             echo "Head ref: $HEAD_REF"
 
-            # 强约束: PR 必须基于 riscv
+            # must based on riscv
             if [ "$BASE_REF" != "riscv" ]; then
               echo "ERROR: PR must target 'riscv' branch, got '$BASE_REF'"
               exit 1
@@ -97,4 +106,60 @@ jobs:
 
       - name: Trigger Jenkins Job
         run: |
-          bash /home/jenkins/scripts/jenkins-run.sh $BASE_COMMIT  $PATCH_FILE
+          set -euo pipefail
+
+          BASE=${{ steps.meta.outputs.base_commit }}
+          PATCH=${{ steps.patch.outputs.patch_file }}
+
+          bash /home/jenkins/scripts/jenkins-run.sh $BASE_COMMIT  $PATCH_FILE | tee jenkins.log
+
+          CI_STAT_URL=$(grep -oE 'https://[^ ]+/pytorch-ci-stat\.json' jenkins.log | tail -n1)
+
+          if [[ -z "$CI_STAT_URL" ]]; then
+            echo "ERROR: cannot find pytorch-ci-stat.json URL from Jenkins log"
+            exit 1
+          fi
+
+          CI_RESULT_BASE_URL="${CI_STAT_URL%/pytorch-ci-stat.json}"
+
+          echo "ci_stat_url=$CI_STAT_URL" >> "$GITHUB_OUTPUT"
+          echo "ci_result_base_url=$CI_RESULT_BASE_URL" >> "$GITHUB_OUTPUT"
+
+          echo "CI_STAT_URL=$CI_STAT_URL"
+          echo "CI_RESULT_BASE_URL=$CI_RESULT_BASE_URL"
+
+full-ci:
+  name: pytorch-riscv64-full-ci
+  runs-on: [self-hosted, linux, amd64]
+  needs: core-ci
+  if: always()
+  continue-on-error: true
+
+  steps:
+    - name: Query existing full test result
+      shell: bash
+      run: |
+        set -euo pipefail
+
+        BASE_URL="${{ needs.core-ci.outputs.ci_result_base_url }}"
+        STAT_URL="${BASE_URL}/pytorch-ci-stat.json"
+
+        echo "STAT_URL=$STAT_URL"
+
+         curl -fsSL "$STAT_URL" -o pytorch-ci-stat.json
+
+        echo "==== FULL TEST STAT ===="
+        cat pytorch-ci-stat.json
+        echo
+
+        FAILED=$(jq '.failed | length' pytorch-ci-stat.json)
+
+        if [[ "$FAILED" != "0" ]]; then
+          echo "==== FULL TEST FAILED ===="
+          echo "failed cases: $FAILED"
+          exit 1
+        fi
+
+        echo "==== FULL TEST PASSED ===="
+        echo "full test no failures"
+

From 3a5cb48e42ee82437c871b0b0b5f118a8daeb8b5 Mon Sep 17 00:00:00 2001
From: vimer <yuzibode@126.com>
Date: Sun, 24 May 2026 23:51:12 +0800
Subject: [PATCH 15/16] [ci] fix ci workflow and other issue (#21)

for action, these yaml must be merged first then take effect, so merge
it skipping ci
---
 .github/workflows/ci-riscv64.yml | 63 ++++++++++++++++++--------------
 1 file changed, 35 insertions(+), 28 deletions(-)

diff --git a/.github/workflows/ci-riscv64.yml b/.github/workflows/ci-riscv64.yml
index 127e0a01e36eb..8c53debb93a8e 100644
--- a/.github/workflows/ci-riscv64.yml
+++ b/.github/workflows/ci-riscv64.yml
@@ -50,6 +50,7 @@ jobs:
           echo "HEAD_SHA=${{ github.event.pull_request.head.sha }}" >> $GITHUB_ENV
 
       - name: Diff base and head
+        id: meta
         run: |
           if [[ "${{ github.event_name }}" = "pull_request" || "${{ github.event_name }}" == "pull_request_target" ]]; then
             echo "Push PR build"
@@ -84,10 +85,14 @@ jobs:
           echo "BASE_COMMIT=$BASE" >> $GITHUB_ENV
           echo "HEAD_COMMIT=$HEAD" >> $GITHUB_ENV
 
+          echo "base_commit=$BASE" >> "$GITHUB_OUTPUT"
+          echo "head_commit=$HEAD" >> "$GITHUB_OUTPUT"
+
           echo "Base: $BASE"
           echo "Head: $HEAD"
 
       - name: Generate patch
+        id: patch
         run: |
           echo "Generating patch..."
 
@@ -102,16 +107,18 @@ jobs:
           cp $PATCH_NAME /home/jenkins/patch/
           cat /home/jenkins/patch/$PATCH_NAME
 
-          echo "PATCH_FILE=$PATCH_NAME" >> $GITHUB_ENV
+          echo "PATCH_FILE=$PATCH_NAME" >> "$GITHUB_ENV"
+          echo "patch_file=$PATCH_NAME" >> "$GITHUB_OUTPUT"
 
       - name: Trigger Jenkins Job
+        id: jenkins
         run: |
           set -euo pipefail
 
           BASE=${{ steps.meta.outputs.base_commit }}
           PATCH=${{ steps.patch.outputs.patch_file }}
 
-          bash /home/jenkins/scripts/jenkins-run.sh $BASE_COMMIT  $PATCH_FILE | tee jenkins.log
+          bash /home/jenkins/scripts/jenkins-run.sh $BASE  $PATCH | tee jenkins.log
 
           CI_STAT_URL=$(grep -oE 'https://[^ ]+/pytorch-ci-stat\.json' jenkins.log | tail -n1)
 
@@ -128,38 +135,38 @@ jobs:
           echo "CI_STAT_URL=$CI_STAT_URL"
           echo "CI_RESULT_BASE_URL=$CI_RESULT_BASE_URL"
 
-full-ci:
-  name: pytorch-riscv64-full-ci
-  runs-on: [self-hosted, linux, amd64]
-  needs: core-ci
-  if: always()
-  continue-on-error: true
+  full-ci:
+    name: pytorch-riscv64-full-ci
+    runs-on: [self-hosted, linux, amd64]
+    needs: core-ci
+    if: always()
+    continue-on-error: true
 
-  steps:
-    - name: Query existing full test result
-      shell: bash
-      run: |
-        set -euo pipefail
+    steps:
+      - name: Query existing full test result
+        shell: bash
+        run: |
+          set -euo pipefail
 
-        BASE_URL="${{ needs.core-ci.outputs.ci_result_base_url }}"
-        STAT_URL="${BASE_URL}/pytorch-ci-stat.json"
+          BASE_URL="${{ needs.core-ci.outputs.ci_result_base_url }}"
+          STAT_URL="${BASE_URL}/pytorch-ci-stat.json"
 
-        echo "STAT_URL=$STAT_URL"
+          echo "STAT_URL=$STAT_URL"
 
-         curl -fsSL "$STAT_URL" -o pytorch-ci-stat.json
+          curl -fsSL "$STAT_URL" -o pytorch-ci-stat.json
 
-        echo "==== FULL TEST STAT ===="
-        cat pytorch-ci-stat.json
-        echo
+          echo "==== FULL TEST STAT ===="
+          cat pytorch-ci-stat.json
+          echo
 
-        FAILED=$(jq '.failed | length' pytorch-ci-stat.json)
+          FAILED=$(jq '.failed | length' pytorch-ci-stat.json)
 
-        if [[ "$FAILED" != "0" ]]; then
-          echo "==== FULL TEST FAILED ===="
-          echo "failed cases: $FAILED"
-          exit 1
-        fi
+          if [[ "$FAILED" != "0" ]]; then
+            echo "==== FULL TEST FAILED ===="
+            echo "failed cases: $FAILED"
+            exit 1
+          fi
 
-        echo "==== FULL TEST PASSED ===="
-        echo "full test no failures"
+          echo "==== FULL TEST PASSED ===="
+          echo "full test no failures"
 

From 96cf21946af9dc3b4739aaa1b83bac4e91384c76 Mon Sep 17 00:00:00 2001
From: Bo YU <tsu.yubo@gmail.com>
Date: Thu, 21 May 2026 04:04:17 +0000
Subject: [PATCH 16/16] [ci] clear unused flags

But the most important here is, to only run core test cases
---
 .github/workflows/ci-riscv64.yml | 5 -----
 1 file changed, 5 deletions(-)

diff --git a/.github/workflows/ci-riscv64.yml b/.github/workflows/ci-riscv64.yml
index 8c53debb93a8e..273034f514d79 100644
--- a/.github/workflows/ci-riscv64.yml
+++ b/.github/workflows/ci-riscv64.yml
@@ -18,11 +18,6 @@ concurrency:
   group: ${{ github.workflow }}-${{ github.head_ref || github.run_id }}
   cancel-in-progress: true
 
-# FIXME: Drop this
-env:
-  RUSTFLAGS: -D warnings
-  CARGO_TERM_COLOR: always
-
 jobs:
   core-ci:
     name: pytorch-riscv64-core-ci