compiles

.github/workflows/ci-linux.yml

@@ -49,7 +49,10 @@ jobs:
           git remote add origin $REPO_ADDRESS
           git -c protocol.version=2 fetch --depth 1 origin $BRANCH_NAME
           git reset --hard FETCH_HEAD
-          git -c submodule."third_party/torch-mlir".update=none -c submodule."third_party/stablehlo".update=none -c submodule."src/runtime_src/core/common/aiebu".update=none submodule update --init --recursive --depth 1 --single-branch -j 10
+          git -c submodule."third_party/torch-mlir".update=none \
+              -c submodule."third_party/stablehlo".update=none \
+              -c submodule."src/runtime_src/core/common/aiebu".update=none \
+              submodule update --init --recursive --depth 1 --single-branch -j 10
           
       - name: Install deps
         run: |
@@ -61,6 +64,11 @@ jobs:
         run: |
           pip install "numpy<2" pyyaml "pybind11[global]==2.10.3" nanobind pytest
 
+      - name: Peano dep
+        run: |
+          bash build_tools/download_peano.sh
+          echo "PEANO_INSTALL_DIR=$PWD/llvm-aie" >> $GITHUB_ENV
+
       - name: Run Pytest
         run: |
           pytest build_tools/ci

.github/workflows/ci-windows.yml

@@ -60,7 +60,10 @@ jobs:
           git remote add origin $REPO_ADDRESS
           git -c protocol.version=2 fetch --depth 1 origin $BRANCH_NAME
           git reset --hard FETCH_HEAD
-          git -c submodule."third_party/torch-mlir".update=none -c submodule."third_party/stablehlo".update=none -c submodule."src/runtime_src/core/common/aiebu".update=none submodule update --init --recursive --depth 1 --single-branch -j 10
+          git -c submodule."third_party/torch-mlir".update=none \
+              -c submodule."third_party/stablehlo".update=none \
+              -c submodule."src/runtime_src/core/common/aiebu".update=none \
+              submodule update --init --recursive --depth 1 --single-branch -j 10
 
       - name: Setup Cpp
         uses: aminya/setup-cpp@v1
@@ -86,6 +89,11 @@ jobs:
           key: ${{ env.CACHE_KEY }}
           restore-keys: windows-build-test-cpp-
 
+      - name: Peano dep
+        run: |
+          .\build_tools\download_peano.ps1
+          Add-Content -Path $env:GITHUB_ENV -Value "PEANO_INSTALL_DIR=$PWD\llvm-aie"
+
       - name: Build packages
         run: |
           $env:cache_dir = "${{ env.CACHE_DIR }}"

build_tools/build_test_cpp.ps1

@@ -80,10 +80,18 @@ $CMAKE_ARGS = @(
     "-DIREE_TARGET_BACKEND_DEFAULTS=OFF"
     "-DIREE_TARGET_BACKEND_LLVM_CPU=ON"
     "-DIREE_CMAKE_PLUGIN_PATHS=$repo_root"
-    "-DIREE_EXTERNAL_HAL_DRIVERS=xrt"
+    "-DIREE_EXTERNAL_HAL_DRIVERS=xrt;xrt-lite"
     "-DIREE_BUILD_PYTHON_BINDINGS=ON"
+    # iree/runtime/src/iree/hal/cts/cts_test_base.h:173:24: error: unused variable 'device_buffer' [-Werror,-Wunused-variable]
+    "-DIREE_ENABLE_WERROR_FLAG=OFF"
 )
 
+$peano_install_dir = "$env:PEANO_INSTALL_DIR"
+if ($peano_install_dir -and (Test-Path "$peano_install_dir"))
+{
+    $CMAKE_ARGS += @("-DPEANO_INSTALL_DIR=$peano_install_dir")
+}
+
 if ($llvm_install_dir -and (Test-Path "$llvm_install_dir"))
 {
     echo "using existing llvm install @ $llvm_install_dir"
@@ -121,7 +129,7 @@ echo "-----"
 # better have git-bash installed...
 $env:Path = "C:\Program Files\Git\bin;$env:Path"
 pushd $build_dir
-& bash -l -c "ctest -R amd-aie --output-on-failure -j --repeat until-pass:5"
+& bash -l -c "ctest -R amd-aie -E driver/xrt-lite --output-on-failure -j --repeat until-pass:5"
 popd
 
 if ($llvm_install_dir -and (Test-Path "$llvm_install_dir"))

build_tools/build_test_cpp.sh

@@ -31,7 +31,7 @@ mkdir -p "${cache_dir}/pip"
 python="$(which python)"
 echo "Using python: $python"
 
-if [[ "$OSTYPE" == "linux-gnu"* ]]; then
+if [[ "$OSTYPE" == "linux"* ]]; then
   export CMAKE_TOOLCHAIN_FILE="$this_dir/linux_default_toolchain.cmake"
   export CC=clang
   export CXX=clang++
@@ -61,54 +61,73 @@ echo '{
 }' > $iree_dir/CMakeUserPresets.json 
 
 cd $iree_dir
-CMAKE_ARGS="\
-  -GNinja \
-  -DCMAKE_BUILD_TYPE=Release \
-  -DCMAKE_INSTALL_PREFIX=$install_dir \
-  -DCMAKE_INSTALL_LIBDIR=lib \
-  -DIREE_ERROR_ON_MISSING_SUBMODULES=OFF \
-  -DIREE_ENABLE_ASSERTIONS=ON \
-  -DIREE_BUILD_SAMPLES=OFF \
-  -DIREE_BUILD_PYTHON_BINDINGS=ON \
-  -DIREE_BUILD_BINDINGS_TFLITE=OFF \
-  -DIREE_HAL_DRIVER_DEFAULTS=OFF \
-  -DIREE_HAL_DRIVER_LOCAL_SYNC=ON \
-  -DIREE_HAL_DRIVER_LOCAL_TASK=ON \
-  -DIREE_TARGET_BACKEND_DEFAULTS=OFF \
-  -DIREE_TARGET_BACKEND_LLVM_CPU=ON \
-  -DIREE_INPUT_TOSA=OFF \
-  -DIREE_INPUT_STABLEHLO=OFF \
-  -DIREE_INPUT_TORCH=OFF \
-  -DCMAKE_OBJECT_PATH_MAX=4096 \
-  -DIREE_CMAKE_PLUGIN_PATHS=$repo_root"
-
-if [ -d "${llvm_install_dir}" ]; then
-  CMAKE_ARGS="$CMAKE_ARGS \
-    -DIREE_BUILD_BUNDLED_LLVM=OFF \
-    -DClang_DIR=$llvm_install_dir/lib/cmake/clang \
-    -DLLD_DIR=$llvm_install_dir/lib/cmake/lld \
-    -DMLIR_DIR=$llvm_install_dir/lib/cmake/mlir \
-    -DLLVM_DIR=$llvm_install_dir/lib/cmake/llvm"
+CMAKE_ARGS=(
+  -GNinja
+  -DCMAKE_BUILD_TYPE=Release
+  -DCMAKE_INSTALL_PREFIX="$install_dir"
+  -DCMAKE_INSTALL_LIBDIR=lib
+  -DIREE_ERROR_ON_MISSING_SUBMODULES=OFF
+  -DIREE_ENABLE_ASSERTIONS=ON
+  -DIREE_BUILD_SAMPLES=OFF
+  -DIREE_BUILD_PYTHON_BINDINGS=ON
+  -DIREE_BUILD_BINDINGS_TFLITE=OFF
+  -DIREE_HAL_DRIVER_DEFAULTS=OFF
+  -DIREE_HAL_DRIVER_LOCAL_SYNC=ON
+  -DIREE_HAL_DRIVER_LOCAL_TASK=ON
+  -DIREE_TARGET_BACKEND_DEFAULTS=OFF
+  -DIREE_TARGET_BACKEND_LLVM_CPU=ON
+  -DIREE_INPUT_TOSA=OFF
+  -DIREE_INPUT_STABLEHLO=OFF
+  -DIREE_INPUT_TORCH=OFF
+  -DCMAKE_OBJECT_PATH_MAX=4096
+  -DIREE_CMAKE_PLUGIN_PATHS="$repo_root"
+  # iree/runtime/src/iree/hal/cts/cts_test_base.h:173:24: error: unused variable 'device_buffer' [-Werror,-Wunused-variable]
+  -DIREE_ENABLE_WERROR_FLAG=OFF
+)
+
+PEANO_INSTALL_DIR=${PEANO_INSTALL_DIR:-""}
+if [ "$PEANO_INSTALL_DIR" != "" ] && [ -d "$PEANO_INSTALL_DIR" ]; then
+  CMAKE_ARGS+=(-DPEANO_INSTALL_DIR="$PEANO_INSTALL_DIR")
 fi
 
-if [[ "$OSTYPE" == "linux-gnu"* ]]; then
-  cmake $CMAKE_ARGS \
-    -DCMAKE_EXE_LINKER_FLAGS_INIT="-fuse-ld=lld" \
-    -DCMAKE_SHARED_LINKER_FLAGS_INIT="-fuse-ld=lld" \
-    -DCMAKE_MODULE_LINKER_FLAGS_INIT="-fuse-ld=lld" \
-    -DCMAKE_C_COMPILER="${CC}" \
-    -DCMAKE_CXX_COMPILER="${CXX}" \
-    -DLLVM_TARGET_ARCH=X86 \
-    -DLLVM_TARGETS_TO_BUILD=X86 \
-    -DIREE_EXTERNAL_HAL_DRIVERS=xrt \
-    -S $iree_dir -B $build_dir
+if [ -d "$llvm_install_dir" ]; then
+  CMAKE_ARGS+=(
+    -DIREE_BUILD_BUNDLED_LLVM=OFF
+    -DClang_DIR="$llvm_install_dir/lib/cmake/clang"
+    -DLLD_DIR="$llvm_install_dir/lib/cmake/lld"
+    -DMLIR_DIR="$llvm_install_dir/lib/cmake/mlir"
+    -DLLVM_DIR="$llvm_install_dir/lib/cmake/llvm"
+  )
+fi
+
+if [[ "$OSTYPE" == "linux"* ]]; then
+  CMAKE_ARGS+=(
+    -DCMAKE_EXE_LINKER_FLAGS_INIT="-fuse-ld=lld"
+    -DCMAKE_SHARED_LINKER_FLAGS_INIT="-fuse-ld=lld"
+    -DCMAKE_MODULE_LINKER_FLAGS_INIT="-fuse-ld=lld"
+    -DCMAKE_C_COMPILER="${CC}"
+    -DCMAKE_CXX_COMPILER="${CXX}"
+    -DLLVM_TARGET_ARCH=X86
+    -DLLVM_TARGETS_TO_BUILD=X86
+    -DIREE_EXTERNAL_HAL_DRIVERS="xrt;xrt-lite"
+    -S
+    "$iree_dir"
+    -B
+    "$build_dir"
+  )
 elif [[ "$OSTYPE" == "darwin"* ]]; then
-  cmake $CMAKE_ARGS \
-    -DLLVM_TARGET_ARCH="X86;ARM" \
-    -DLLVM_TARGETS_TO_BUILD="X86;ARM" \
-    -S $iree_dir -B $build_dir
+  CMAKE_ARGS+=(
+    -DLLVM_TARGET_ARCH="X86;ARM"
+    -DLLVM_TARGETS_TO_BUILD="X86;ARM"
+    -S
+    "$iree_dir"
+    -B
+    "$build_dir"
+  )
 fi
 
+cmake "${CMAKE_ARGS[@]}"
+
 echo "Building all"
 echo "------------"
 cmake --build "$build_dir" -- -k 0
@@ -123,8 +142,8 @@ cmake --build "$build_dir" --target iree-install-dist
 
 echo "CTest"
 echo "-----"
-if [[ "$OSTYPE" == "linux-gnu"* ]]; then
-  ctest --test-dir "$build_dir" -R amd-aie --output-on-failure -j
+if [[ "$OSTYPE" == "linux"* ]]; then
+  ctest --test-dir "$build_dir" -R amd-aie -E "driver/xrt-lite" --output-on-failure -j
 elif [[ "$OSTYPE" == "darwin"* ]]; then
   ctest --test-dir "$build_dir" -R amd-aie -E "matmul_pack_peel_air_e2e|matmul_elementwise_pack_peel_air_e2e|conv_fill_spec_pad" --output-on-failure -j --repeat until-pass:5
 fi

build_tools/download_peano.ps1

@@ -9,4 +9,7 @@ $ErrorActionPreference = 'Stop'
 $this_dir = Split-Path -Path $MyInvocation.MyCommand.Path -Parent
 $RELEASE = (Get-Content -Path "$this_dir/peano_commit.txt")
 pip download llvm_aie==$RELEASE -f https://github.com/Xilinx/llvm-aie/releases/expanded_assets/nightly
-Expand-Archive (Get-ChildItem -Filter llvm*.whl).FullName -DestinationPath $PWD.Path
+$peano = (Get-ChildItem -Filter llvm*.whl)
+$new_name = ($peano.Basename + ".zip")
+Rename-Item -Path $peano.Name -NewName $new_name
+Expand-Archive $new_name -DestinationPath $PWD.Path

runtime/src/iree-amd-aie/driver/xrt-lite/shim/linux/kmq/amdxdna_accel.h

@@ -132,11 +132,11 @@ struct amdxdna_cu_config {
  * @pad: MBZ
  * @cu_configs: Array of CU configurations of struct amdxdna_cu_config
  */
-struct amdxdna_hwctx_param_config_cu {
-  __u16 num_cus;
-  __u16 pad[3];
-  struct amdxdna_cu_config cu_configs[] __counted_by(num_cus);
-};
+// struct amdxdna_hwctx_param_config_cu {
+//   __u16 num_cus;
+//   __u16 pad[3];
+//   struct amdxdna_cu_config cu_configs[] __counted_by(num_cus);
+// };
 
 enum amdxdna_drm_config_hwctx_param {
   DRM_AMDXDNA_HWCTX_CONFIG_CU,

runtime/src/iree-amd-aie/driver/xrt-lite/shim/linux/kmq/bo.cpp

@@ -9,14 +9,15 @@
 #include <unistd.h>
 #include <x86intrin.h>
 
+#include "pcidev.h"
 #include "shim_debug.h"
 
 namespace {
 
 uint32_t alloc_drm_bo(const shim_xdna::pdev& dev, amdxdna_bo_type type,
                       void* buf, size_t size) {
   amdxdna_drm_create_bo cbo = {
-      .type = type,
+      .type = static_cast<uint32_t>(type),
       .vaddr = reinterpret_cast<uintptr_t>(buf),
       .size = size,
   };
@@ -38,7 +39,7 @@ void get_drm_bo_info(const shim_xdna::pdev& dev, uint32_t boh,
 void* map_parent_range(size_t size) {
   auto p = ::mmap(0, size, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS,
                   -1, 0);
-  if (!p) shim_err(errno, "mmap(len=%ld) failed", size);
+  if (!p) shim_xdna::shim_err(errno, "mmap(len=%ld) failed", size);
 
   return p;
 }
@@ -101,7 +102,7 @@ bool is_power_of_two(size_t x) { return (x > 0) && ((x & (x - 1)) == 0); }
 
 void* addr_align(void* p, size_t align) {
   if (!is_power_of_two(align))
-    shim_err(EINVAL, "Alignment 0x%lx is not power of two", align);
+    shim_xdna::shim_err(EINVAL, "Alignment 0x%lx is not power of two", align);
 
   return (void*)(((uintptr_t)p + align) & ~(align - 1));
 }
@@ -129,7 +130,8 @@ inline void clflush_data(const void* base, size_t offset, size_t len) {
 
   if (!cacheline_size) {
     long sz = sysconf(_SC_LEVEL1_DCACHE_LINESIZE);
-    if (sz <= 0) shim_err(EINVAL, "Invalid cache line size: %ld", sz);
+    if (sz <= 0)
+      shim_xdna::shim_err(EINVAL, "Invalid cache line size: %ld", sz);
     cacheline_size = sz;
   }
 
@@ -313,8 +315,7 @@ std::unique_ptr<shared_handle> bo::share() const {
 
 amdxdna_bo_type bo::get_type() const { return m_type; }
 
-bo::bo(const device& device, hw_ctx::slot_id ctx_id, size_t size,
-       uint64_t flags)
+bo::bo(const device& device, uint32_t ctx_id, size_t size, uint64_t flags)
     : bo(device, ctx_id, size, flags, flag_to_type(flags)) {
   if (m_type == AMDXDNA_BO_INVALID)
     shim_err(EINVAL, "Invalid BO flags: 0x%lx", flags);
@@ -323,8 +324,8 @@ bo::bo(const device& device, hw_ctx::slot_id ctx_id, size_t size,
 bo::bo(const device& device, size_t size, amdxdna_bo_type type)
     : bo(device, AMDXDNA_INVALID_CTX_HANDLE, size, 0, type) {}
 
-bo::bo(const device& device, hw_ctx::slot_id ctx_id, size_t size,
-       uint64_t flags, amdxdna_bo_type type)
+bo::bo(const device& device, uint32_t ctx_id, size_t size, uint64_t flags,
+       amdxdna_bo_type type)
     : m_pdev(device.get_pdev()),
       m_aligned_size(size),
       m_flags(flags),

runtime/src/iree-amd-aie/driver/xrt-lite/shim/linux/kmq/bo.h

@@ -5,12 +5,11 @@
 #define _BO_XDNA_H_
 
 #include <atomic>
+#include <map>
+#include <mutex>
 #include <string>
 
 #include "amdxdna_accel.h"
-#include "device.h"
-#include "hwctx.h"
-#include "pcidev.h"
 #include "shared.h"
 #include "shim_debug.h"
 
@@ -62,6 +61,9 @@ struct xcl_bo_flags {
   };
 };
 
+struct device;
+struct pdev;
+
 struct bo {
   // map_type - determines how a buffer is mapped
   enum class map_type { read, write };
@@ -87,10 +89,11 @@ struct bo {
     uint64_t kmhdl;  // kernel mode handle
   };
 
-  bo(const device& device, hw_ctx::slot_id ctx_id, size_t size, uint64_t flags,
+  using uint32_t = uint32_t;
+  bo(const device& device, uint32_t ctx_id, size_t size, uint64_t flags,
      amdxdna_bo_type type);
 
-  bo(const device& device, hw_ctx::slot_id ctx_id, size_t size, uint64_t flags);
+  bo(const device& device, uint32_t ctx_id, size_t size, uint64_t flags);
 
   bo(const device& device, shared_handle::export_handle ehdl);
 
@@ -171,7 +174,7 @@ struct bo {
 
   // Used when exclusively assigned to a HW context. By default, BO is shared
   // among all HW contexts.
-  hw_ctx::slot_id m_owner_ctx_id = AMDXDNA_INVALID_CTX_HANDLE;
+  uint32_t m_owner_ctx_id = AMDXDNA_INVALID_CTX_HANDLE;
 
   void bind_at(size_t pos, const bo* bh, size_t offset, size_t size);
 

runtime/src/iree-amd-aie/driver/xrt-lite/shim/linux/kmq/device.cpp

@@ -3,18 +3,15 @@
 
 #include "device.h"
 
-#include <sys/syscall.h>
-#include <unistd.h>
-
-#include <any>
 #include <filesystem>
 
 #include "bo.h"
 #include "hwctx.h"
+#include "pcidev.h"
 
 namespace shim_xdna {
 
-device::device(const pdev& pdev, handle_type shim_handle)
+device::device(const pdev& pdev, void* shim_handle)
     : m_pdev(pdev), m_handle(shim_handle) {
   shim_debug("Created KMQ device (%s) ...", get_pdev().m_sysfs_name.c_str());
 }
@@ -24,16 +21,16 @@ device::~device() {
   m_pdev.close();
 }
 
-std::unique_ptr<hw_ctx> device::create_hw_context(
-    const device& dev, const hw_ctx::qos_type& qos) const {
-  return std::make_unique<hw_ctx>(dev, qos);
-}
+// std::unique_ptr<hw_ctx> device::create_hw_context(
+//     const device& dev, const hw_ctx::qos_type& qos) const {
+//   return std::make_unique<hw_ctx>(dev, qos);
+// }
 
-std::unique_ptr<bo> device::alloc_bo(void* userptr, hw_ctx::slot_id ctx_id,
+std::unique_ptr<bo> device::alloc_bo(void* userptr, uint32_t ctx_id,
                                      size_t size, uint64_t flags) {
   if (userptr) shim_not_supported_err("User ptr BO");
 
-  auto b = bo(this->m_pdev, ctx_id, size, flags);
+  auto b = bo(*this, ctx_id, size, flags);
   return std::make_unique<bo>(*this, ctx_id, size, flags);
 }