diff --git a/.github/cases/blastoise/subsystem.json b/.github/cases/blastoise/subsystem.json
new file mode 100644
index 000000000..16eadb3af
--- /dev/null
+++ b/.github/cases/blastoise/subsystem.json
@@ -0,0 +1,17 @@
+{
+  "rvv-vp-intrinsic-add-mlir": 291,
+  "rvv-vp-intrinsic-add-scalable-mlir": 448,
+  "hello-mlir": 72,
+  "stripmining-mlir": 23635,
+  "conv2d-less-m2-intrinsic": 27600,
+  "linear_normalization-intrinsic": 1,
+  "matmul-intrinsic": 1,
+  "softmax-intrinsic": 1,
+  "uarttest-intrinsic": 1,
+  "fpsmoke-asm": 1,
+  "memcpy-asm": 1,
+  "mmm-asm": 61401,
+  "smoke-asm": 7539,
+  "strlen-asm": 1,
+  "utf8-count-asm": 1
+}
diff --git a/.github/cases/v1024-l8-b2/subsystem.json b/.github/cases/v1024-l8-b2/subsystem.json
new file mode 100644
index 000000000..16eadb3af
--- /dev/null
+++ b/.github/cases/v1024-l8-b2/subsystem.json
@@ -0,0 +1,17 @@
+{
+  "rvv-vp-intrinsic-add-mlir": 291,
+  "rvv-vp-intrinsic-add-scalable-mlir": 448,
+  "hello-mlir": 72,
+  "stripmining-mlir": 23635,
+  "conv2d-less-m2-intrinsic": 27600,
+  "linear_normalization-intrinsic": 1,
+  "matmul-intrinsic": 1,
+  "softmax-intrinsic": 1,
+  "uarttest-intrinsic": 1,
+  "fpsmoke-asm": 1,
+  "memcpy-asm": 1,
+  "mmm-asm": 61401,
+  "smoke-asm": 7539,
+  "strlen-asm": 1,
+  "utf8-count-asm": 1
+}
diff --git a/.github/scripts/ci.sc b/.github/scripts/ci.sc
index d447cea36..09de76329 100755
--- a/.github/scripts/ci.sc
+++ b/.github/scripts/ci.sc
@@ -51,7 +51,8 @@ def scheduleTasks(allTasksFile: Seq[os.Path], bucketSize: Int): Seq[String] = {
         .toSeq
   })
   // Initialize a list of buckets
-  val cargo = (0 until bucketSize).map(_ => new BucketBuffer())
+  val size = bucketSize.min(allCycleData.size)
+  val cargo = (0 until size).map(_ => new BucketBuffer())
   // _2 is the cycle number
   val (unProcessedData, normalData) = allCycleData.partition(_._2 <= 0)
   // Group tests that have cycle data into subset by their cycle size
@@ -63,8 +64,8 @@ def scheduleTasks(allTasksFile: Seq[os.Path], bucketSize: Int): Seq[String] = {
     })
   // For unprocessed data, just split them into subset that have equal size
   cargo.zipWithIndex.foreach { case(buffer, i) =>
-    val startIdx = i * bucketSize
-    val endIdx = math.min((i + 1) * bucketSize, unProcessedData.length)
+    val startIdx = i * size
+    val endIdx = math.min((i + 1) * size, unProcessedData.length)
     unProcessedData.slice(startIdx, endIdx).foreach { case(name, cycle) => buffer.push_back(name, cycle) }
   }
   cargo.map(_.mkString).toSeq
@@ -80,18 +81,15 @@ def toMatrixJson(buckets: Seq[String]) =
 // Read default tests information from '.github/cases/default.txt' file, and use that information to generate GitHub CI matrix.
 // The result will be printed to stdout, and should be pipe into $GITHUB_OUTPUT
 @main
-def generateCiMatrix(
-    runnersAmount: Int,
-) = {
-  val defaultCases = os.pwd / os.RelPath(".github/cases/default.txt")
-  println(toMatrixJson(
-    scheduleTasks(
-      os.read
-        .lines(defaultCases)
-        .map(defaultCases / os.up / os.RelPath(_)),
-      runnersAmount
-    ),
-  ))
+def generateCiMatrix(defaultCases: String = "default.json", runnersAmount: Int) = {
+  println(
+    toMatrixJson(
+      scheduleTasks(
+        os.walk(os.pwd/".github"/"cases").filter(_.last == defaultCases),
+        runnersAmount
+      ),
+    )
+  )
 }
 
 // Resolve all the executable test and filter out unpassed tests, appending perf testcases
@@ -220,7 +218,7 @@ def writeCycleUpdates(job: String, testRunDir: os.Path, resultDir: os.Path) = {
 // @param: resultDir output directory of the test results, default to ./test-results
 // @param: dontBail don't throw exception when test fail. Useful for postpr.
 @main
-def runTests(jobs: String, resultDir: Option[os.Path], dontBail: Boolean = false) = {
+def runTests(jobs: String, runTarget: String = "ip", resultDir: Option[os.Path], dontBail: Boolean = false) = {
   var actualResultDir = resultDir.getOrElse(os.pwd / "test-results")
   val testRunDir = os.pwd / "testrun"
   os.makeDir.all(actualResultDir / "failed-logs")
@@ -229,8 +227,11 @@ def runTests(jobs: String, resultDir: Option[os.Path], dontBail: Boolean = false
     case(failed, (job, i)) => {
       val Array(config, caseName) = job.split(",")
       System.err.println(s"\n\n\n>>>[${i+1}/${totalJobs.length}] Running test case $config,$caseName")
+      val args = Seq("scripts/run-test.py", runTarget, "-c", config, "--base-out-dir", testRunDir.toString) ++
+        { if (runTarget == "ip") Seq("--no-log") else Seq() } ++
+        Seq(caseName)
       val handle = os
-        .proc("scripts/run-test.py", "ip", "-c", config, "--no-log", "--base-out-dir", testRunDir, caseName)
+        .proc(args)
         .call(check=false)
       if (handle.exitCode != 0) {
         val outDir = testRunDir / config / caseName
@@ -238,7 +239,10 @@ def runTests(jobs: String, resultDir: Option[os.Path], dontBail: Boolean = false
         os.write(actualResultDir / "failed-logs" / s"$job.txt", handle.out.text)
         failed :+ job
       } else {
-        writeCycleUpdates(job, testRunDir, actualResultDir)
+        if (runTarget == "ip") {
+          writeCycleUpdates(job, testRunDir, actualResultDir)
+        }
+
         failed
       }
     }
@@ -263,14 +267,14 @@ def runTests(jobs: String, resultDir: Option[os.Path], dontBail: Boolean = false
 // @param: jobs A semicolon-separated list of job names of the form $config,$caseName,$runConfig
 // @param: output directory of the test results, default to ./test-results
 @main
-def runFailedTests(jobs: String) = {
+def runFailedTests(jobs: String, runTarget: String = "ip") = {
   val testRunDir = os.pwd / "testrun"
   val totalJobs = jobs.split(";")
   val failed = totalJobs.zipWithIndex.foreach { case (job, i) => {
     val Array(config, caseName) = job.split(",")
     System.err.println(s"[${i+1}/${totalJobs.length}] Running test case with trace $config,$caseName")
     val handle = os
-      .proc("scripts/run-test.py", "ip", "-c", config, "--trace", "--no-log", "--base-out-dir", testRunDir, caseName)
+      .proc("scripts/run-test.py", runTarget, "-c", config, "--trace", "--no-log", "--base-out-dir", testRunDir, caseName)
       .call(check=false)
   }}
 }
diff --git a/.github/workflows/subsystem.yml b/.github/workflows/subsystem.yml
new file mode 100644
index 000000000..2790c0e4a
--- /dev/null
+++ b/.github/workflows/subsystem.yml
@@ -0,0 +1,199 @@
+name: subsystem
+on: [pull_request]
+env:
+  USER: runner
+  JAVA_OPTS: "-Duser.home=/run/github-runner/sequencer"
+
+# Cancel the current workflow when new commit pushed
+concurrency:
+  group: ${{ github.workflow }}-${{ github.event.pull_request.number }}
+  cancel-in-progress: true
+
+jobs:
+  build-emulators:
+    name: "Build Emulators"
+    runs-on: [self-hosted, linux, nixos, AMD Ryzen 9 7940HS w/ Radeon 780M Graphics]
+    strategy:
+      matrix:
+        config:
+          - squirtle
+          - blastoise
+    steps:
+      - uses: actions/checkout@v4
+        with:
+          ref: ${{ github.event.pull_request.head.sha }}
+      - name: "Build verilator emulator"
+        run: |
+          nix build '.#t1.${{ matrix.config }}.subsystem.emu' -L --no-link --cores 64
+
+  gen-matrix:
+    name: "Prepare for running testcases"
+    needs: [build-emulators]
+    runs-on: [self-hosted, linux, nixos, "AMD Ryzen 9 7940HS w/ Radeon 780M Graphics"]
+    env:
+      RUNNERS: 70
+    outputs:
+      ci-tests: ${{ steps.gen-matrix.outputs.matrix }}
+    steps:
+      # actions/checkout will use the "event" commit to checkout repository,
+      # which will lead to an unexpected issue that the "event" commit doesn't belongs to the repository,
+      # and causing the derivation build output cannot be cache correctly.
+      - uses: actions/checkout@v4
+        with:
+          ref: ${{ github.event.pull_request.head.sha }}
+      - name: "Build all testcases"
+        run: |
+          # Build testcases with vlen 1024 and vlen 4096
+          nix build ".#t1.bulbasaur.cases.all" --max-jobs auto -L --no-link --cores 64
+          nix build ".#t1.psyduck.cases.all" --max-jobs auto -L --no-link --cores 64
+      - id: gen-matrix
+        name: "Generate test matrix"
+        run: |
+          echo -n matrix= >> "$GITHUB_OUTPUT"
+          nix shell ".#ammonite" -c .github/scripts/ci.sc generateCiMatrix subsystem.json "$RUNNERS" >> "$GITHUB_OUTPUT"
+
+  build-trace-emulators:
+    name: "Build trace emulator"
+    needs: [gen-matrix]
+    runs-on: [self-hosted, linux, nixos, AMD Ryzen 9 7940HS w/ Radeon 780M Graphics]
+    strategy:
+      fail-fast: false
+      matrix:
+        config:
+          - squirtle
+          - blastoise
+    steps:
+      - uses: actions/checkout@v4
+        with:
+          ref: ${{ github.event.pull_request.head.sha }}
+      - name: "Build verilator emulator with trace"
+        run: nix build '.#t1.${{ matrix.config }}.subsystem.emu-trace' -L --no-link --cores 64
+
+  test-emit:
+    name: "Test elaborate"
+    runs-on: [self-hosted, linux, nixos]
+    steps:
+      - uses: actions/checkout@v4
+        with:
+          ref: ${{ github.event.pull_request.head.sha }}
+      - name: "Test elaborate"
+        run: |
+          set -e
+          nix build '.#t1.configgen' -L --out-link ./config-gen
+          for cfg in $(jq -r '.[]' ./config-gen/share/all-supported-configs.json); do
+            echo "Building .#t1.${cfg}.subsystem.rtl"
+            nix build ".#t1.${cfg}.subsystem.rtl" -L
+          done
+
+  run-testcases:
+    name: "Run testcases"
+    needs: [gen-matrix]
+    strategy:
+      fail-fast: false
+      matrix: ${{ fromJSON(needs.gen-matrix.outputs.ci-tests) }}
+    runs-on: [self-hosted, linux, nixos]
+    outputs:
+      result: ${{ steps.ci-run.outputs.result }}
+    steps:
+      - uses: actions/checkout@v4
+        with:
+          ref: ${{ github.event.pull_request.head.sha }}
+      - name: "Run testcases"
+        id: ci-run
+        run: |
+          nix shell ".#ammonite" -c .github/scripts/ci.sc runTests \
+            --runTarget subsystem \
+            --jobs "${{ matrix.jobs }}" \
+            --resultDir test-results-$(head -c 10 /dev/urandom | base32)
+
+      - uses: actions/upload-artifact@v4
+        if: ${{ !cancelled() }}
+        with:
+          name: test-reports-${{ matrix.id }}
+          path: |
+            test-results-*/failed-tests.md
+
+      - uses: actions/upload-artifact@v4
+        if: failure()
+        with:
+          name: failed-logs-${{ matrix.id }}
+          path: test-results-*/failed-logs
+
+  gen-fail-wave-matrix:
+    name: "Generate matrix for re-testing failing tests"
+    if: ${{ !cancelled() }}
+    needs: [run-testcases]
+    runs-on: [self-hosted, linux, nixos]
+    outputs:
+      generate_wave: ${{ steps.generate-matrix.outputs.generate_wave }}
+      retry_tasks: ${{ steps.generate-matrix.outputs.retry_tasks }}
+    steps:
+      - uses: actions/download-artifact@v4
+        with:
+          pattern: test-reports-*
+          merge-multiple: true
+      - id: generate-matrix
+        name: "Generate matrix"
+        run: |
+          touch all-fail-tests.txt
+          shopt -s nullglob
+          cat test-results-*/failed-tests.md > all-failed-tests.txt
+          retry=$(sed 's/\* //' all-failed-tests.txt | shuf -n3)  # only retry last three failed
+          echo "build wave for: $retry"
+          if [ -n "$retry" ]; then
+            echo "generate_wave=true" >> "$GITHUB_OUTPUT"
+            echo -n "retry_tasks=" >> "$GITHUB_OUTPUT"
+            echo "$retry" | \
+              jq -nR --indent 0 '{"include": [inputs | {"job": ., "id": (input_line_number)}]}' >> "$GITHUB_OUTPUT"
+          fi
+
+  build-fail-wave:
+    name: "Generate wave for failing tests"
+    needs: [build-emulators, gen-fail-wave-matrix]
+    if: ${{ !cancelled() && needs.gen-fail-wave-matrix.outputs.generate_wave == 'true'  }}
+    strategy:
+      fail-fast: false
+      matrix: ${{ fromJSON(needs.gen-fail-wave-matrix.outputs.retry_tasks) }}
+    runs-on: [self-hosted, linux, nixos]
+    steps:
+      - uses: actions/checkout@v4
+        with:
+          ref: ${{ github.event.pull_request.head.sha }}
+      - name: "Run failed testcases"
+        run: |
+          nix shell ".#ammonite" -c .github/scripts/ci.sc runFailedTests --runTarget subsystem --jobs "${{ matrix.job }}"
+          waveFile=$(find testrun -name 'wave.fst')
+          hierFile=$(find testrun -name 'wave.hier')
+          if [[ -z "$waveFile" || -n "$hierFile" ]]; then  # when emulator failed, .hier is not merged into .fst
+            echo "Verilator doesn't generate wave correctly"
+            exit 1
+          fi
+          mv "$waveFile" ./wave-${{ matrix.job }}.fst
+      - uses: actions/upload-artifact@v4
+        with:
+          name: failed-tests-wave-${{ matrix.id }}
+          path: 'wave-*.fst'
+      - run: |
+          echo "Test ${{ matrix.job }} run fail"
+          # We are just running post action for failing test, so we need to avoid the workflow finishing successfully
+          exit 1
+
+  report:
+    name: "Report CI result"
+    if: ${{ !cancelled() }}
+    needs: [run-testcases]
+    runs-on: [self-hosted, linux, nixos]
+    steps:
+      - uses: actions/checkout@v4
+        with:
+          fetch-depth: 0
+          ref: ${{ github.head_ref }}
+      - uses: actions/download-artifact@v4
+        with:
+          pattern: test-reports-*
+          merge-multiple: true
+      - name: "Print step summary"
+        run: |
+          echo -e "\n## Failed tests\n" >> $GITHUB_STEP_SUMMARY
+          shopt -s nullglob
+          cat test-results-*/failed-tests.md >> $GITHUB_STEP_SUMMARY
diff --git a/nix/t1/testcases/make-intrinsic-case.nix b/nix/t1/testcases/make-intrinsic-case.nix
index 34a28fa6e..7b7b2917f 100644
--- a/nix/t1/testcases/make-intrinsic-case.nix
+++ b/nix/t1/testcases/make-intrinsic-case.nix
@@ -6,6 +6,8 @@ stdenv.mkDerivation (rec {
   name = "${caseName}-intrinsic";
 
   unpackPhase = ''
+    runHook preUnpack
+
     if [ -z "''${srcs:-}" ]; then
         if [ -z "''${src:-}" ]; then
             echo 'variable $src or $srcs should point to the source'
@@ -13,6 +15,8 @@ stdenv.mkDerivation (rec {
         fi
         srcs="$src"
     fi
+
+    runHook postUnpack
   '';
 
   NIX_CFLAGS_COMPILE = [
diff --git a/scripts/run-test.py b/scripts/run-test.py
index 2917049d6..d44fed177 100755
--- a/scripts/run-test.py
+++ b/scripts/run-test.py
@@ -179,11 +179,11 @@ def load_elf_from_dir(config, cases_dir, case_name, use_individual_drv, force_x8
                 )
             else:
                 nix_args.append(f".#t1.{config}.{cases_attr_name}.all")
-            logger.info(f'Run "{" ".join(nix_args)}"')
+            logger.info(f'Get ELFs dir by cmd: "{" ".join(nix_args)}"')
             cases_dir = subprocess.check_output(nix_args).strip().decode("UTF-8")
 
     cases_dir = Path(cases_dir)
-    logger.info(f"Running cases in {cases_dir}")
+    logger.info(f"Reading ELFs in {cases_dir}")
 
     case_config_path = (
         cases_dir / f"{case_name}.json"
@@ -244,7 +244,7 @@ def run_test(args):
         "-t",
         f"{args.out_dir}",
     ]
-    logger.info(f'Run "{" ".join(configgen_args)}"')
+    logger.info(f'Get emulator config by cmd: "{" ".join(configgen_args)}"')
     subprocess.Popen(configgen_args).wait()
     assert (
         elaborate_config_path.exists()
@@ -319,7 +319,7 @@ def optionals(cond, items):
         ]
     ) + emu_args
 
-    logger.info(f'Run "{" ".join(process_args)}"')
+    logger.info(f'Run {emu_type} emulator with cmd: "{" ".join(process_args)}"')
     return_code = subprocess.Popen(process_args).wait()
 
     if return_code != 0:
diff --git a/subsystememu/csrc/dpic.cc b/subsystememu/csrc/dpic.cc
index 593358281..b8b410905 100644
--- a/subsystememu/csrc/dpic.cc
+++ b/subsystememu/csrc/dpic.cc
@@ -184,11 +184,13 @@ extern "C" void AXI4MMIODPI(
 
     while (uart.exist_tx()) {
         char c = uart.getc();
-        printf("%c",c);
-        fflush(stdout);
         if (c == -1) {
             exit(0);
         }
+        else {
+            printf("%c",c);
+            fflush(stdout);
+        }
     }
     // CTRL  END  }
     
diff --git a/tests/intrinsic/main.S b/tests/intrinsic/main.S
index d8a4c7e94..ec151bd97 100644
--- a/tests/intrinsic/main.S
+++ b/tests/intrinsic/main.S
@@ -17,7 +17,7 @@ exit:
     .p2align 2
 
 heap_start:
-    .zero 1073741824
+    .zero 2048
 
 heap_end:
 stack_start:
diff --git a/tests/intrinsic/uarttest/default.nix b/tests/intrinsic/uarttest/default.nix
new file mode 100644
index 000000000..df62ac6df
--- /dev/null
+++ b/tests/intrinsic/uarttest/default.nix
@@ -0,0 +1,19 @@
+{ testcase-env }:
+testcase-env.mkIntrinsicCase {
+  caseName = "uarttest";
+
+  srcs = [
+    ./uarttest.c
+    ../main.S
+  ];
+
+  postUnpack = ''
+    mkdir -p inc
+    cp ${./uart.h} ./inc/uart.h
+  '';
+
+  preBuild = ''
+    NIX_CFLAGS_COMPILE="-Iinc $NIX_CFLAGS_COMPILE"
+  '';
+}
+
diff --git a/tests/intrinsic/uart.h b/tests/intrinsic/uarttest/uart.h
similarity index 100%
rename from tests/intrinsic/uart.h
rename to tests/intrinsic/uarttest/uart.h
diff --git a/tests/intrinsic/uarttest.c b/tests/intrinsic/uarttest/uarttest.c
similarity index 100%
rename from tests/intrinsic/uarttest.c
rename to tests/intrinsic/uarttest/uarttest.c