Added test bench generation tcl as a LLVM pass (#18)

* Added memory partitioning pragma for Vitis HLS LLVM * Added test bench generation tcl as a LLVM pass * Added pragma insertion for array partitioning * Removed unsupported func attr by vhls * Replaced tb tclgen script with llvm pass * Fixed a typo * Restored original tbgen in comments * Removed assertion for debugging * Switch off -xlnarraypartition if such flag is OFF * Fixed an issue with removing tile.size * Fixed unit tests * Resurrected the unittest workflow * Just build-phism in the workflow * Array partition unit test simple cases * Fixed an issue with the loop bounds * Rewrite modulo operations * Further simplifies GEP expressions * Added loop naming and flattening pass * Remove loop flattening which stops loop merging * An issue with GEP optimisation * Loop redistribution pass * Loop merge pass * Loop redis unit tests * Added unit tests for loop merge * Makes pb-flow stabler * Improved pb-flow and unit tests * Fixed issues with array partition * Try to restore cache and don't run test * Don't cmake if it has been built * Further reduce the chance of cmake runs * Temporarily removed the python interface * Try self-hosted * Added back the python test bits * Demystifying the python script execution in workflow * Changes in address calc and inline * Don't inline in MLIR * Fixed array partitioning issue for mvt * Fixed an issue with the array partitioning * Added cosim run command * Improved logging Co-authored-by: Ruizhe Zhao <[email protected]>
kumasento · Sep 16, 2021 · 367bd87 · 367bd87
1 parent 5ae1e77
commit 367bd87
Show file tree

Hide file tree

Showing 16 changed files with 2,038 additions and 159 deletions.
diff --git a/.github/workflows/buildAndTest.yml b/.github/workflows/buildAndTest.yml
@@ -0,0 +1,77 @@
+# This is a basic workflow to help you get started with Actions
+name: Build and Test
+# Controls when the action will run. Triggers the workflow on push or pull request
+# events but only for the master branch
+on: [push, pull_request]
+# A workflow run is made up of one or more jobs that can run sequentially or in parallel
+jobs:
+  # Build Phism and run its tests.
+  build-phism:
+    name: Build and Test Phism
+    runs-on: self-hosted
+    steps:
+      # - name: Configure Environment
+      #   run: echo "${GITHUB_WORKSPACE}/llvm/install/bin" >> $GITHUB_PATH
+      # Disabled for self-hosted
+      # - name: Get dependences
+      #   run: |
+      #     sudo apt-get update -y
+      #     sudo apt-get install -y build-essential libtool autoconf pkg-config flex bison libgmp-dev clang-9 libclang-9-dev texinfo python3
+      # - name: Update the LLVM/Clang version to 9
+      #   run: |
+      #     sudo update-alternatives --install /usr/bin/llvm-config llvm-config /usr/bin/llvm-config-9 100
+      #     sudo update-alternatives --install /usr/bin/FileCheck FileCheck /usr/bin/FileCheck-9 100
+
+
+      # Clone the Phism repo and its submodules. Do shallow clone to save clone
+      # time.
+      - name: Get Phism
+        uses: actions/checkout@v2
+        with:
+          submodules: "true"
+
+      # --------
+      # Restore LLVM from cache and build if it's not in there.
+      # --------
+      # Extract the LLVM submodule hash for use in the cache key.
+      - name: Get LLVM Hash
+        id: get-llvm-hash
+        run: echo "::set-output name=hash::$(git rev-parse @:./llvm)"
+        shell: bash
+      # Try to fetch LLVM from the cache.
+      - name: Cache LLVM
+        id: cache-llvm
+        uses: actions/cache@v2
+        with:
+          path: llvm/build
+          key: ${{ runner.os }}-llvm-${{ steps.get-llvm-hash.outputs.hash }}
+      # Build LLVM if we didn't hit in the cache. Even though we build it in
+      # the previous job, there is a low chance that it'll have been evicted by
+      # the time we get here.
+      # Need to delete the test directory to avoid caching them.
+      - name: Rebuild and Install LLVM
+        if: steps.cache-llvm.outputs.cache-hit != 'true'
+        run: |
+          ./scripts/build-llvm.sh ci
+          rm -rf ./llvm/build/test 
+      # --------
+      # Build and test Phism in both debug and release mode.
+      # --------
+      - name: Build and Test Phism (Assert)
+        run: |
+          ./scripts/build-phism.sh ci
+      
+      # Build and test Phism with pb-flow.
+      - name: Build and Test Phism (pb-flow)
+        run: |
+          python3 -m venv env
+          source env/bin/activate
+          which python3
+          python3 -m pip install -r requirements.txt
+          python3 ./scripts/pb-flow.py ./example/polybench --dataset SMALL --skip-vitis
+          python3 ./scripts/pb-flow.py ./example/polybench --dataset SMALL --polymer --skip-vitis
+          python3 ./scripts/pb-flow.py ./example/polybench --dataset SMALL --polymer --loop-transforms --skip-vitis
+          python3 ./scripts/pb-flow.py ./example/polybench --dataset SMALL --polymer --loop-transforms --array-partition --skip-vitis
+
+
+
diff --git a/include/phism/mlir/Transforms/Utils.h b/include/phism/mlir/Transforms/Utils.h
@@ -0,0 +1,10 @@
+//===- Utils.h - Utility functions ------------------ C++-===//
+
+#include "mlir/IR/BuiltinOps.h"
+
+namespace phism {
+
+/// Get the top function for the hardware design.
+mlir::FuncOp getTopFunction(mlir::ModuleOp m);
+
+} // namespace phism