Squashed commit of the following:

commit 40dffe83929973c8e205c395be5db23c360c2397 Author: Denny Sun <[email protected]> Date: Thu Mar 23 05:06:51 2023 +0000 Merged PR 3176: [Accera] split_dim op supports dynamic dims with static split size With this fix the following test case which has dynamic dims with static split size can succeed. ``` M, MN = create_dimensions() N = 16 Input = Array(role=Role.INPUT, element_type=ScalarType.float32, shape=(MN,)) Output = Array(role=Role.INPUT_OUTPUT, element_type=ScalarType.float32, shape=(M, N)) nest = Nest(shape=(M, N)) i, j = nest.get_indices() @nest.iteration_logic def _(): split_input = Input._split_dimension(0, cast(16, ScalarType.index)) Output[i, j] = split_input[i, j] ``` commit 451b67405d77ebbe1cf0722f9c7aeb191c3b4beb Author: Mason Remy <[email protected]> Date: Thu Mar 23 01:19:37 2023 +0000 Merged PR 3174: Ensure any dynamic allocations are heap allocs that get dealloced Ensure any dynamic allocations are heap allocs commit 602b068f19cdf1ff9111aacbeb5d704974521ebd Author: Kern Handa <[email protected]> Date: Wed Mar 22 20:59:43 2023 +0000 Merged PR 3171: [test] Add some tests for Dimensions commit ccd1f5c39964fbe96815672f137b5185ee2e9885 Author: Mason Remy <[email protected]> Date: Wed Mar 22 19:41:02 2023 +0000 Merged PR 3175: Support reinterpret cast of same bitwidth without changing layout Support reinterpret cast of same bitwidth without changing layout commit 270a3c8a9c1e1c06b0da3c9d61fa2f04438e3076 Author: Kern Handa <[email protected]> Date: Fri Mar 17 22:16:08 2023 +0000 Merged PR 3167: Remove hack to treat INPUT_OUTPUT Arrays with shape (1,) as Elements I don't have complete context on this, so this might break something. If it does, that should be fixed separately rather than keep this hack around, which breaks semantics in non-obvious ways. commit efcff61727c64e7f0a37f4f92c701bc47ea1c470 Author: Lisa Ong <[email protected]> Date: Fri Mar 17 08:09:07 2023 +0000 Merged PR 3165: [build] Fix clang 14 release build warnings treated as errors on macOS/Apple Errors are showing up on release builds: ``` cmake .. -DCMAKE_BUILD_TYPE=Release -G Ninja cmake --build . --config Release ``` Clang version: ``` Apple clang version 14.0.0 (clang-1400.0.29.202) Target: arm64-apple-darwin22.3.0 Thread model: posix ``` commit 43f311aa706214243ce8d7acca7d29993bb7003b Author: Lisa Ong <[email protected]> Date: Fri Mar 17 07:02:09 2023 +0000 Merged PR 3162: Bump vcpkg to latest release Last release was Sept 2022. Update to the latest tag (2023.02.24) Preparation for LLVM 15 upgrade commit 07098f502596d997bbe241e95f1130c11e318220 Author: Mason Remy <[email protected]> Date: Thu Mar 16 23:27:04 2023 +0000 Merged PR 3161: Fix cache reduce scale constant hoisting Fix cache reduce scale constant hoisting commit 696ef0df5947067f94b64255e00b7fffc4c04f9d Author: Mason Remy <[email protected]> Date: Thu Mar 16 20:54:22 2023 +0000 Merged PR 3163: Extend vector masked loads/stores to handle arbitrary bin ops and constant operands Extend vector masked loads/stores to handle arbitrary bin ops and constant operands
microsoft · Mar 24, 2023 · 1127a60 · 1127a60
1 parent f55e4d8
commit 1127a60
Show file tree

Hide file tree

Showing 24 changed files with 852 additions and 245 deletions.
diff --git a/CMakeLists.txt b/CMakeLists.txt
@@ -35,12 +35,12 @@ option(STRICT_MODE "Build with 'warnings as errors'" OFF)
 option(USE_MKL "Build with Intel MKL" OFF)
 
 option(USE_LIBCXX "Build with libc++ if using the Clang compiler" OFF)
-if(CMAKE_CXX_COMPILER_ID STREQUAL Clang)
+if(CMAKE_CXX_COMPILER_ID STREQUAL Clang OR CMAKE_CXX_COMPILER_ID STREQUAL AppleClang)
   if(USE_LIBCXX OR (CMAKE_HOST_SYSTEM_NAME STREQUAL Darwin))
     add_compile_options(-stdlib=libc++)
     link_libraries(-lc++ -lc++abi)
   endif(USE_LIBCXX OR (CMAKE_HOST_SYSTEM_NAME STREQUAL Darwin))
-endif(CMAKE_CXX_COMPILER_ID STREQUAL Clang)
+endif(CMAKE_CXX_COMPILER_ID STREQUAL Clang OR CMAKE_CXX_COMPILER_ID STREQUAL AppleClang)
 
 # Try to create a compilation database, which is useful to have when working
 # with clang tooling
@@ -161,10 +161,13 @@ else()
   set(CMAKE_C_FLAGS_DEBUG "${CMAKE_C_FLAGS_DEBUG} -ggdb3 -O0")
   set(CMAKE_CXX_FLAGS_RELWITHDEBINFO "${CMAKE_CXX_FLAGS_RELWITHDEBINFO} -ggdb3")
   set(CMAKE_C_FLAGS_RELWITHDEBINFO "${CMAKE_C_FLAGS_RELWITHDEBINFO} -ggdb3")
-  if(${CMAKE_CXX_COMPILER_ID} STREQUAL Clang)
+
+  if(CMAKE_CXX_COMPILER_ID STREQUAL Clang OR CMAKE_CXX_COMPILER_ID STREQUAL AppleClang)
     if(CMAKE_BUILD_TYPE STREQUAL Debug)
-    # Set options for Control Flow Integrity
-      add_compile_options(-fsanitize=cfi)
+      # Set options for Control Flow Integrity
+      if(NOT ${OSX_NATIVE_ARCH} STREQUAL "arm64")
+        add_compile_options(-fsanitize=cfi)
+      endif()
     endif(CMAKE_BUILD_TYPE STREQUAL Debug)
 
     add_compile_options(-Wno-backslash-newline-escape)

diff --git a/accera/hat/include/HATEmitter.h b/accera/hat/include/HATEmitter.h
@@ -294,6 +294,8 @@ class Parameter : public TOMLSerializable
         return SerializeCommonParameters();
     }
 
+    virtual ~Parameter() = default;
+
 protected:
     Parameter(const LogicalParamType& logicalType, const std::string& name, const std::string& description, const UsageType usage, const std::string& declaredType, const std::string& elementType) :
         _logicalType{ logicalType },
@@ -415,6 +417,8 @@ class Function : public TOMLSerializable
         _description{ description },
         _callingConvention{ callingConvention } {}
 
+    virtual ~Function() = default;
+
     std::string Name() const { return _name; }
 
     std::string Description() const { return _description; }
@@ -1021,6 +1025,8 @@ class ExternalLibraryReference : public TOMLSerializable
         return table;
     }
 
+    virtual ~ExternalLibraryReference() = default;
+
 private:
     std::string _name;
     std::string _version;

diff --git a/accera/ir/include/value/ValueOps.td b/accera/ir/include/value/ValueOps.td
@@ -645,9 +645,13 @@ def accv_CallOp : accv_Op<"call", [CallOpInterface]> {
       $_state.addOperands(operands);
       $_state.addAttribute("callee", callee);
       $_state.addTypes(results);
-  }]>, OpBuilder<(ins "StringRef":$callee, "ArrayRef<Type>":$results, CArg<"ValueRange", "{}">:$operands), [{
-      build($_builder, $_state, StringAttr::get($_builder.getContext(), callee), results,
-            operands);
+
+  // BUGBUG: -Werror,-Winfinite-recursion, needed?
+  // }]>, OpBuilder<(ins "StringRef":$callee, "ArrayRef<Type>":$results, CArg<"ValueRange", "{}">:$operands), [{
+  //    build($_builder, $_state, StringAttr::get($_builder.getContext(), callee), results,
+  //          operands);
+  //
+
   }]>];
 
   let extraClassDeclaration = [{

diff --git a/accera/ir/src/AffineConstraintsHelper.cpp b/accera/ir/src/AffineConstraintsHelper.cpp
@@ -200,7 +200,7 @@ namespace util
         if (_cst.containsId(val))
         {
             unsigned id = 0;
-            bool found = _cst.findId(val, &id);
+            [[maybe_unused]] bool found = _cst.findId(val, &id);
             assert(found);
             return IdWrapper::FromFullId(id, _cst);
         }

diff --git a/accera/ir/src/TranslateToHeader.cpp b/accera/ir/src/TranslateToHeader.cpp
@@ -750,14 +750,6 @@ namespace ir
                     }
                     std::vector<size_t> shapeVec;
                     std::transform(shape.begin(), shape.end(), std::back_inserter(shapeVec), [](int64_t val) { return static_cast<size_t>(val); });
-                    if (usage != hat::UsageType::Input && shapeVec.size() == 1 && shapeVec[0] == 1)
-                    {
-                        // TODO: This is currently a hack since output Dimension does not work. So in the DSL we use Array
-                        // instead and here we emulate an ElementParameter instead. Remove this when output Dimension are working.
-                        assert(declaredType.back() == '*');
-                        return std::make_unique<hat::ElementParameter>(name, description, usage, declaredType.substr(0, declaredType.length() - 1), elementType);
-                    }
-
                     return std::make_unique<hat::AffineArrayParameter>(name, description, usage, declaredType, elementType, shapeVec, affineMap, affineOffset);
                 }
 

diff --git a/accera/ir/src/value/ValueDialect.cpp b/accera/ir/src/value/ValueDialect.cpp
@@ -511,9 +511,9 @@ MemRefType ViewOp::computeMemRefType(Value source, ValueRange sizes, ValueRange
     auto context = source.getContext();
     auto sourceMemRefType = source.getType().cast<mlir::MemRefType>();
     int64_t sourceRank = sourceMemRefType.getRank();
-    int64_t numOffsets = static_cast<int64_t>(offsets.size());
-    int64_t numSizes = static_cast<int64_t>(sizes.size());
-    int64_t numStrides = static_cast<int64_t>(strides.size());
+    [[maybe_unused]] int64_t numOffsets = static_cast<int64_t>(offsets.size());
+    [[maybe_unused]] int64_t numSizes = static_cast<int64_t>(sizes.size());
+    [[maybe_unused]] int64_t numStrides = static_cast<int64_t>(strides.size());
     assert(sourceRank == numOffsets);
     assert(sourceRank == numSizes);
     assert(sourceRank == numStrides);

diff --git a/accera/python/accera/lang/Array.py b/accera/python/accera/lang/Array.py
@@ -11,7 +11,7 @@
 from operator import mul
 from typing import *
 
-from .._lang_python import ScalarType, _MemoryLayout, AllocateFlags, Role
+from .._lang_python import ScalarType, _MemoryLayout, AllocateFlags, Role, type_size_bytes
 from .._lang_python._lang import Array as NativeArray, Dimension
 from .Layout import Layout, MemoryMapLayout
 from ..Parameter import DelayedParameter
@@ -177,22 +177,28 @@ def _value(self):
             return None
 
     def _reinterpret_cast_internal(self, element_type):
-        if any(map(lambda d: isinstance(d, Dimension), self.shape)):
-            expected_layout = [-1]
+        src_element_size = type_size_bytes(self.element_type)
+        dst_element_size = type_size_bytes(element_type)
+        if src_element_size == dst_element_size:
+            expected_layout = self.shape
         else:
-            src_element_size = np.dtype(SCALAR_TYPE_TO_DTYPE_STR[self.element_type]).itemsize
-            dst_element_size = np.dtype(SCALAR_TYPE_TO_DTYPE_STR[element_type]).itemsize
-            expected_layout = [int(self._num_elements * (src_element_size / dst_element_size))]
+            if any(map(lambda d: isinstance(d, Dimension), self.shape)):
+                expected_layout = [-1]
+            else:
+                expected_layout = [int(self._num_elements * (src_element_size / dst_element_size))]
         reinterpreted = Array(role=self.role, element_type=element_type, shape=expected_layout)
         return reinterpreted
 
     def _get_memory_buffer(self):
         return self._reinterpret_cast_internal(ScalarType.uint8)
 
     def _reinterpret_cast(self, element_type):
-        if self.element_type != ScalarType.uint8 or len(self.shape) != 1:
+        src_bytewidth = type_size_bytes(self.element_type)
+        dst_bytewidth = type_size_bytes(element_type)
+        if src_bytewidth != dst_bytewidth and \
+            (self.element_type != ScalarType.uint8 or len(self.shape) != 1):
             raise RuntimeError(
-                "Can only call reinterpret cast on flat uint8 memory buffers. Call _get_memory_buffer first?"
+                "Can only call reinterpret cast such that the bitwidth doesn't change, or on flat uint8 memory buffers. Call _get_memory_buffer first?"
             )
 
         return self._reinterpret_cast_internal(element_type)

diff --git a/accera/python/accera/test/dsl_tests.py b/accera/python/accera/test/dsl_tests.py
@@ -640,6 +640,22 @@ def test_reinterpret_cast(self) -> None:
             shape=(256, 256),
         )
 
+        def reinterpret_arr_as_int16(array: Array):
+            # Assumes array is f32
+
+            num_elements = reduce(lambda x, y: x*y, array.shape, 1)
+            arr_mb = array._get_memory_buffer()
+            self.assertEqual(arr_mb.shape, [num_elements * 4])
+            self.assertEqual(arr_mb.element_type, ScalarType.uint8)
+            print(arr_mb.layout)
+
+            arr_as_int16 = arr_mb._reinterpret_cast(ScalarType.int16)
+            self.assertEqual(arr_as_int16.shape, [num_elements * 2])
+            self.assertEqual(arr_as_int16.element_type, ScalarType.int16)
+            print(arr_as_int16.layout)
+
+            return arr_as_int16
+
         def reinterpret_arr_as_int32(array: Array):
             # Assumes array is f32
 
@@ -656,9 +672,19 @@ def reinterpret_arr_as_int32(array: Array):
 
             return arr_as_int32
 
+        def simple_reinterpret_arr_as_int32(array: Array):
+            # Assumes array is f32
+
+            num_elements = reduce(lambda x, y: x*y, array.shape, 1)
+            arr_as_int32 = array._reinterpret_cast(ScalarType.int32)
+            self.assertEqual(arr_as_int32.shape, array.shape)
+            self.assertEqual(arr_as_int32.element_type, ScalarType.int32)
+            print(arr_as_int32.layout)
+
+            return arr_as_int32
+
         # add a function that utilizes a subarray layout
-        def make_reinterpreted_fn(array):
-            reinterpreted = reinterpret_arr_as_int32(array)
+        def make_reinterpreted_fn(array, reinterpreted):
             nest = Nest(shape=reinterpreted.shape)
             i = nest.get_indices()
 
@@ -668,12 +694,23 @@ def _():
 
             return package.add(nest, args=(reinterpreted, ))
 
-        reinterpreted_fn = make_reinterpreted_fn(arr)
+        reinterpreted_i32 = reinterpret_arr_as_int32(arr)
+        reinterpreted_i32_fn = make_reinterpreted_fn(arr, reinterpreted_i32)
+
+        reinterpreted_i16 = reinterpret_arr_as_int16(arr)
+        reinterpreted_i16_fn = make_reinterpreted_fn(arr, reinterpreted_i16)
+
+        simple_reinterpreted_i32 = simple_reinterpret_arr_as_int32(arr)
+        simple_reinterpreted_i32_fn = make_reinterpreted_fn(arr, simple_reinterpreted_i32)
 
         # add a function that instantiates a subarray of the input array and calls the function above
         def main(array):
-            reinterpreted_array = reinterpret_arr_as_int32(array)
-            reinterpreted_fn(reinterpreted_array)
+            reinterpreted_array_i32 = reinterpret_arr_as_int32(array)
+            reinterpreted_array_i16 = reinterpret_arr_as_int16(array)
+            simple_reinterpreted_array_i32 = simple_reinterpret_arr_as_int32(array)
+            reinterpreted_i32_fn(reinterpreted_array_i32)
+            reinterpreted_i16_fn(reinterpreted_array_i16)
+            simple_reinterpreted_i32_fn(simple_reinterpreted_array_i32)
 
         package.add(main, args=(arr, ))
 
@@ -1432,7 +1469,7 @@ def _():
         }
 
         # TODO: Disabling this verification for now, re-enable it when undoing this change.
-        # self._verify_helper(package, get_size_fn_name, get_size_fn.name, correctness_check_values)
+        self._verify_helper(package, get_size_fn_name, get_size_fn.name, correctness_check_values)
 
         correctness_check_values = {
             "pre": [size_test, x_ref, start_array_pre_test, delta_test],