Backport some PRs from main to release/15.x

.github/workflows/pip.yml

@@ -29,7 +29,7 @@ jobs:
         arch: [ x86_64, aarch64 ]
 
     steps:
-      - uses: actions/checkout@v3
+      - uses: actions/checkout@v3.1.0
 
       - name: Log in to GitHub Container Registry
         uses: docker/login-action@v2.0.0
@@ -67,9 +67,6 @@ jobs:
 
     runs-on: ${{ matrix.runner }}
 
-    env:
-      CMAKE_PREFIX_PATH: ${{ github.workspace }}/local
-
     strategy:
       fail-fast: false
       matrix:
@@ -83,28 +80,29 @@ jobs:
             arch: x86_64;arm64
 
     steps:
-      - uses: actions/checkout@v3
+      - uses: actions/checkout@v3.1.0
 
       - name: Cache LLVM build folder
         id: cache-llvm
-        uses: actions/cache@v3
+        uses: actions/cache@v3.0.11
         with:
-          path: local
+          path: local-llvm
           key: llvmorg-${{ env.LLVM_VER }}-${{ runner.os }}
 
       - uses: ilammy/msvc-dev-cmd@v1
       - uses: lukka/get-cmake@latest
 
-      - uses: actions/checkout@v3
+      - uses: actions/checkout@v3.1.0
         if: steps.cache-llvm.outputs.cache-hit != 'true'
         with:
+          path: llvm-src
           repository: llvm/llvm-project
           ref: llvmorg-${{ env.LLVM_VER }}
 
       - name: Configure LLVM
         if: steps.cache-llvm.outputs.cache-hit != 'true'
         run: >
-          cmake -G Ninja -S llvm -B build
+          cmake -G Ninja -S llvm-src/llvm -B llvm-build
           -DCMAKE_BUILD_TYPE=Release
           "-DCMAKE_OSX_ARCHITECTURES=arm64;x86_64"
           "-DLLVM_TARGETS_TO_BUILD=X86;ARM;NVPTX;AArch64;Mips;Hexagon;WebAssembly"
@@ -116,46 +114,59 @@ jobs:
           -DLLVM_ENABLE_TERMINFO=OFF
           -DLLVM_ENABLE_ZSTD=OFF
           -DLLVM_ENABLE_ZLIB=OFF
+          -DLLVM_ENABLE_OCAMLDOC=OFF
+          -DLLVM_ENABLE_BINDINGS=OFF
+          -DLLVM_ENABLE_IDE=OFF
 
       - name: Build LLVM
         if: steps.cache-llvm.outputs.cache-hit != 'true'
-        run: cmake --build build
+        run: cmake --build llvm-build
 
       - name: Install LLVM
         if: steps.cache-llvm.outputs.cache-hit != 'true'
-        run: cmake --install build --prefix local
+        run: cmake --install llvm-build --prefix local-llvm
+
+      # Remove the LLVM source tree after building it, otherwise we can
+      # run out of local space while building halide
+      - name: Clean LLVM Source
+        if: steps.cache-llvm.outputs.cache-hit != 'true'
+        shell: bash
+        run: rm -rf llvm-src
 
       - name: Configure Halide
         if: runner.os == 'Windows'
         run: >
-          cmake -G "Visual Studio 17 2022" -T ClangCL -A "${{ matrix.arch }}" -S . -B build
+          cmake -G "Visual Studio 17 2022" -T ClangCL -A "${{ matrix.arch }}" -S . -B halide-build
           -DWITH_DOCS=NO
           -DWITH_PYTHON_BINDINGS=NO
           -DWITH_TESTS=NO
           -DWITH_TUTORIALS=NO
           -DWITH_UTILS=NO
+          -DLLVM_DIR=${{ github.workspace }}/local-llvm/lib/cmake/llvm
 
       - name: Configure Halide
         if: runner.os != 'Windows'
         run: >
-          cmake -G Ninja -S . -B build
+          cmake -G Ninja -S . -B halide-build
           -DCMAKE_BUILD_TYPE=Release
           "-DCMAKE_OSX_ARCHITECTURES=${{ matrix.arch }}"
           -DWITH_DOCS=NO
           -DWITH_PYTHON_BINDINGS=NO
           -DWITH_TESTS=NO
           -DWITH_TUTORIALS=NO
           -DWITH_UTILS=NO
+          -DLLVM_DIR=${{ github.workspace }}/local-llvm/lib/cmake/llvm
 
       - name: Build Halide
-        run: cmake --build build --config Release
+        run: cmake --build halide-build --config Release
 
       - name: Install Halide
-        run: cmake --install build --config Release --prefix local
+        run: cmake --install halide-build --config Release --prefix local-halide
 
       - name: Build wheels
         uses: pypa/cibuildwheel@v2.10.2
         env:
+          CMAKE_PREFIX_PATH: ${{ github.workspace }}/local-halide
           CIBW_BUILD: "cp38-${{ matrix.pytag }} cp39-${{ matrix.pytag }} cp310-${{ matrix.pytag }}"
           CIBW_ARCHS_MACOS: "universal2"
 
@@ -168,7 +179,7 @@ jobs:
     name: Make SDist
     runs-on: ubuntu-latest
     steps:
-      - uses: actions/checkout@v3
+      - uses: actions/checkout@v3.1.0
       - run: pipx run build --sdist
       - uses: actions/upload-artifact@v3
         with:

CMakeLists.txt

@@ -90,6 +90,28 @@ option(TARGET_SPIRV "Include SPIR-V target" OFF)
 
 add_subdirectory(dependencies)
 
+##
+# Declare options
+##
+
+# Declare these options after we include dependencies (since it declares Halide_ENABLE_RTTI etc)
+# but before we add any subdirectories, since any option you test before it is defined is
+# implicitly false the *first* time that the build file is processed, and there are some
+# out-of-order dependencies here (e.g, code in src/ eventually checks WITH_UTILS).
+# This is especially subtle since it means that some options can end up with different
+# values if you build a target as part of the initial CMake run, so (e.g.) a `make install`
+# from as totally clean build might neglect to install some pieces.
+
+option(WITH_TESTS "Build tests" "${PROJECT_IS_TOP_LEVEL}")
+option(WITH_TUTORIALS "Build tutorials" "${PROJECT_IS_TOP_LEVEL}")
+option(WITH_DOCS "Build documentation" OFF)
+option(WITH_UTILS "Build utils" "${PROJECT_IS_TOP_LEVEL}")
+cmake_dependent_option(
+    WITH_PYTHON_BINDINGS "Build Python bindings" "${PROJECT_IS_TOP_LEVEL}"
+    "Halide_ENABLE_RTTI AND Halide_ENABLE_EXCEPTIONS" OFF
+)
+
+
 ##
 # Add source directories
 ##
@@ -101,42 +123,34 @@ add_subdirectory(tools)
 # Add tests, tutorials, etc. if we're not being imported into another CMake project.
 ##
 
-option(WITH_TESTS "Build tests" "${PROJECT_IS_TOP_LEVEL}")
 if (WITH_TESTS)
     message(STATUS "Building tests enabled")
     add_subdirectory(test)
 else ()
     message(STATUS "Building tests disabled")
 endif ()
 
-cmake_dependent_option(
-    WITH_PYTHON_BINDINGS "Build Python bindings" "${PROJECT_IS_TOP_LEVEL}"
-    "Halide_ENABLE_RTTI AND Halide_ENABLE_EXCEPTIONS" OFF
-)
 if (WITH_PYTHON_BINDINGS)
     message(STATUS "Building Python bindings enabled")
     add_subdirectory(python_bindings)
 else ()
     message(STATUS "Building Python bindings disabled")
 endif ()
 
-option(WITH_TUTORIALS "Build tutorials" "${PROJECT_IS_TOP_LEVEL}")
 if (WITH_TUTORIALS)
     message(STATUS "Building tutorials enabled")
     add_subdirectory(tutorial)
 else ()
     message(STATUS "Building tutorials disabled")
 endif ()
 
-option(WITH_DOCS "Build documentation" OFF)
 if (WITH_DOCS)
     message(STATUS "Building docs enabled")
     add_subdirectory(doc)
 else ()
     message(STATUS "Building docs disabled")
 endif ()
 
-option(WITH_UTILS "Build utils" "${PROJECT_IS_TOP_LEVEL}")
 if (WITH_UTILS)
     message(STATUS "Building utils enabled")
     add_subdirectory(util)

README_cmake.md

@@ -935,8 +935,8 @@ generators were imported (and hence won't be built). Otherwise, it will be set
 to false. This variable may be used to conditionally set properties on
 `<target>`.
 
-Please see [test/integration/xc](https://github.com/halide/Halide/tree/master/test/integration/xc) for a simple example
-and [apps/hannk](https://github.com/halide/Halide/tree/master/apps/hannk) for a complete app that uses it extensively.
+Please see [test/integration/xc](https://github.com/halide/Halide/tree/main/test/integration/xc) for a simple example
+and [apps/hannk](https://github.com/halide/Halide/tree/main/apps/hannk) for a complete app that uses it extensively.
 
 If `PYSTUB` is specified, then a Python Extension will be built that
 wraps the Generator with CPython glue to allow use of the Generator
@@ -1190,7 +1190,7 @@ without broader approval. Confine dependencies to the `dependencies/` subtree.
 Any variables that are specific to languages that are not enabled should, of
 course, be avoided. But of greater concern are variables that are easy to misuse
 or should not be overridden for our end-users. The following (non-exhaustive)
-list of variables shall not be used in code merged into master.
+list of variables shall not be used in code merged into main.
 
 | Variable                        | Reason                                        | Alternative                                                                                             |
 |---------------------------------|-----------------------------------------------|---------------------------------------------------------------------------------------------------------|

README_python.md

@@ -772,7 +772,7 @@ in future releases.
 ## License
 
 The Python bindings use the same
-[MIT license](https://github.com/halide/Halide/blob/master/LICENSE.txt) as
+[MIT license](https://github.com/halide/Halide/blob/main/LICENSE.txt) as
 Halide.
 
 Python bindings provided by Connelly Barnes (2012-2013), Fred Rotbart (2014),

apps/bgu/bgu_generator.cpp

@@ -1,6 +1,6 @@
 // A Halide implementation of bilateral-guided upsampling.
 
-// Adapted from https://github.com/google/bgu/blob/master/src/halide/bgu.cpp
+// Adapted from https://github.com/google/bgu/tree/master/src/halide
 
 // Copyright 2016 Google Inc.
 //

python_bindings/src/halide/halide_/PyBuffer.cpp

@@ -195,8 +195,6 @@ Type format_descriptor_to_type(const std::string &fd) {
     return Type();
 }
 
-namespace {
-
 py::object buffer_getitem_operator(Buffer<> &buf, const std::vector<int> &pos) {
     if ((size_t)pos.size() != (size_t)buf.dimensions()) {
         throw py::value_error("Incorrect number of dimensions.");
@@ -228,6 +226,8 @@ py::object buffer_getitem_operator(Buffer<> &buf, const std::vector<int> &pos) {
     return py::object();
 }
 
+namespace {
+
 py::object buffer_setitem_operator(Buffer<> &buf, const std::vector<int> &pos, const py::object &value) {
     if ((size_t)pos.size() != (size_t)buf.dimensions()) {
         throw py::value_error("Incorrect number of dimensions.");

python_bindings/src/halide/halide_/PyBuffer.h

@@ -10,6 +10,8 @@ void define_buffer(py::module &m);
 
 Type format_descriptor_to_type(const std::string &fd);
 
+py::object buffer_getitem_operator(Buffer<> &buf, const std::vector<int> &pos);
+
 template<typename T = void,
          int Dims = AnyDims,
          int InClassDimStorage = (Dims == AnyDims ? 4 : std::max(Dims, 1))>

python_bindings/src/halide/halide_/PyFunc.cpp

@@ -80,15 +80,38 @@ py::object realization_to_object(const Realization &r) {
     return to_python_tuple(r);
 }
 
+py::object evaluate_impl(const py::object &expr, bool may_gpu) {
+    Tuple t = to_halide_tuple(expr);
+    Func f("evaluate_func_" + std::to_string(t.size()));
+    f() = t;
+    if (may_gpu) {
+        Internal::schedule_scalar(f);
+    }
+
+    std::optional<Realization> r;
+    {
+        py::gil_scoped_release release;
+
+        r = f.realize();
+    }
+    if (r->size() == 1) {
+        return buffer_getitem_operator((*r)[0], {});
+    } else {
+        py::tuple result(r->size());
+        for (size_t i = 0; i < r->size(); i++) {
+            result[i] = buffer_getitem_operator((*r)[i], {});
+        }
+        return result;
+    }
+}
+
 }  // namespace
 
 void define_func(py::module &m) {
     define_func_ref(m);
     define_var_or_rvar(m);
     define_loop_level(m);
 
-    // TODO: add ParamMap support.
-
     // Deliberately not supported, because they don't seem to make sense for Python:
     // - set_custom_allocator()
     // - set_custom_do_task()
@@ -380,6 +403,18 @@ void define_func(py::module &m) {
     add_schedule_methods(func_class);
 
     define_stage(m);
+
+    m.def(
+        "evaluate", [](const py::object &expr) -> py::object {
+            return evaluate_impl(expr, false);
+        },
+        py::arg("expr"));
+
+    m.def(
+        "evaluate_may_gpu", [](const py::object &expr) -> py::object {
+            return evaluate_impl(expr, true);
+        },
+        py::arg("expr"));
 }
 
 }  // namespace PythonBindings

python_bindings/src/halide/halide_/PyTuple.cpp

@@ -3,6 +3,31 @@
 namespace Halide {
 namespace PythonBindings {
 
+Tuple to_halide_tuple(const py::object &o) {
+    try {
+        Expr e = o.cast<Expr>();
+        return Tuple(e);
+    } catch (...) {
+        // fall thru
+    }
+
+    try {
+        py::tuple t = o.cast<py::tuple>();
+        if (t.empty()) {
+            throw py::value_error("Cannot use a zero-length tuple-of-Expr");
+        }
+        std::vector<Expr> v(t.size());
+        for (size_t i = 0; i < t.size(); i++) {
+            v[i] = t[i].cast<Expr>();
+        }
+        return Tuple(v);
+    } catch (...) {
+        // fall thru
+    }
+
+    throw py::value_error("Expected an Expr or tuple-of-Expr.");
+}
+
 void define_tuple(py::module &m) {
     // Halide::Tuple isn't surfaced to the user in Python;
     // we define it here to allow PyBind to do some automatic
@@ -42,6 +67,7 @@ void define_tuple(py::module &m) {
                 o << "<halide.Tuple of size " << t.size() << ">";
                 return o.str();
             });
+
     py::implicitly_convertible<py::tuple, Tuple>();
 
     // If we autoconvert from vector<Expr>, we must also special-case FuncRef, alas

python_bindings/src/halide/halide_/PyTuple.h

@@ -18,6 +18,11 @@ inline py::tuple to_python_tuple(const T &ht) {
     return pt;
 }
 
+// in: convertible-to-Expr, or tuple-of-convertible-to-Expr
+// out: Halide::Tuple
+// throws exception if not convertible
+Tuple to_halide_tuple(const py::object &o);
+
 }  // namespace PythonBindings
 }  // namespace Halide