[mlir] move _mlir_alloc and friends to CRunnerUtils

[ftynse]

Summary

These functions don't depend on the C++ runtime and therefore belong to
CRunnerUtils. Clean up the macros on the way as _MSC_VER indicates the
compiler, not the platform, which is indicated by _WIN32 and will be
present when, e.g., compiling with minGW.

[mscuttari]

I was unsure where to put those functions inside the runtime library, so it's perfectly fine for me to move them.
I'm unsure about the _WIN32 macro though, the usage of _MSC_VER was caused by the fact that MSVC's implementation of the standard library doesn't provide the aligned_alloc function, despite it being part of the standard. Instead, it provides an _aligned_malloc function. Moreover, MSVC also requires the pointers obtained through _aligned_malloc to be deallocated through _aligned_free, and not free (it doesn't impact the previous reasoning, but better to make it known)

[ftynse]

In D130025#3660205, @mscuttari wrote:

I was unsure where to put those functions inside the runtime library, so it's perfectly fine for me to move them.
I'm unsure about the _WIN32 macro though, the usage of _MSC_VER was caused by the fact that MSVC's implementation of the standard library doesn't provide the aligned_alloc function, despite it being part of the standard. Instead, it provides an _aligned_malloc function. Moreover, MSVC also requires the pointers obtained through _aligned_malloc to be deallocated through _aligned_free, and not free (it doesn't impact the previous reasoning, but better to make it known)

Hmm, should we then guard by a conjunction of both macros?

[mscuttari]

Yes, that wouldn't hurt. I feel like the _WIN32 macro alone, though, would cause problems.
I understand that the MSVC macro is not the best in term of maintenability, but unfortunately this is the situation. Here is a link to the official docs for what regards _aligned_malloc and _aligned_free: https://docs.microsoft.com/en-us/cpp/c-runtime-library/reference/aligned-malloc

Side question, as these are my first times dealing with Phabricator: where do I find the option to accept the revision when I'm among the reviewers?

[mstorsjo]

FWIW, from the mingw point of view, this patch seems fine - mingw doesn’t yet provide the standard aligned_alloc either, but _aligned_malloc can be used, just like with MSVC. So therefore, using _WIN32 for the ifdef seems fine.

mlir/lib/ExecutionEngine/CRunnerUtils.cpp

 #endif
 #include <sys/time.h>
 #else
 #include "malloc.h"
 #endif // _WIN32
 
 #include <cinttypes>
 #include <cstdio>
+#include <cstdlib>
 #include <string.h>
 
 #ifdef MLIR_CRUNNERUTILS_DEFINE_FUNCTIONS
 
 // Small runtime support "lib" for vector.print lowering.
 // By providing elementary printing methods only, this
 // library can remain fully unaware of low-level implementation
 // details of our vectors. Also useful for direct LLVM IR output.
if (stat != 0)
     fprintf(stderr, "Error returning time from gettimeofday: %d\n", stat);
   return (tp.tv_sec + tp.tv_usec * 1.0e-6);
 #else
   fprintf(stderr, "Timing utility not implemented on Windows\n");
   return 0.0;
 #endif // _WIN32
 }
 
+extern "C" void *_mlir_alloc(uint64_t size) { return malloc(size); }
+
+extern "C" void *_mlir_aligned_alloc(uint64_t alignment, uint64_t size) {
+#ifdef _WIN32
+  return _aligned_malloc(size, alignment);
+#else
+  return aligned_alloc(alignment, size);
+#endif
+}
+
+extern "C" void _mlir_free(void *ptr) { free(ptr); }
+
+extern "C" void _mlir_aligned_free(void *ptr) {
+#ifdef _WIN32
+  _aligned_free(ptr);
+#else
+  free(ptr);
+#endif
+}
+
 #endif // MLIR_CRUNNERUTILS_DEFINE_FUNCTIONS

mlir/test/mlir-cpu-runner/utils.mlir

-// RUN: mlir-opt %s -pass-pipeline="func.func(convert-linalg-to-loops,convert-scf-to-cf,convert-arith-to-llvm),convert-linalg-to-llvm,convert-memref-to-llvm,convert-func-to-llvm,reconcile-unrealized-casts" | mlir-cpu-runner -e print_0d -entry-point-result=void -shared-libs=%linalg_test_lib_dir/libmlir_runner_utils%shlibext | FileCheck %s --check-prefix=PRINT-0D
-// RUN: mlir-opt %s -pass-pipeline="func.func(convert-linalg-to-loops,convert-scf-to-cf,convert-arith-to-llvm),convert-linalg-to-llvm,convert-memref-to-llvm,convert-func-to-llvm,reconcile-unrealized-casts" | mlir-cpu-runner -e print_1d -entry-point-result=void -shared-libs=%linalg_test_lib_dir/libmlir_runner_utils%shlibext | FileCheck %s --check-prefix=PRINT-1D
-// RUN: mlir-opt %s -pass-pipeline="func.func(convert-linalg-to-loops,convert-scf-to-cf,convert-arith-to-llvm),convert-linalg-to-llvm,convert-memref-to-llvm,convert-func-to-llvm,reconcile-unrealized-casts" | mlir-cpu-runner -e print_3d -entry-point-result=void -shared-libs=%linalg_test_lib_dir/libmlir_runner_utils%shlibext | FileCheck %s --check-prefix=PRINT-3D
-// RUN: mlir-opt %s -pass-pipeline="func.func(convert-linalg-to-loops,convert-scf-to-cf,convert-arith-to-llvm),convert-linalg-to-llvm,convert-memref-to-llvm,convert-func-to-llvm,reconcile-unrealized-casts" | mlir-cpu-runner -e vector_splat_2d -entry-point-result=void -shared-libs=%linalg_test_lib_dir/libmlir_runner_utils%shlibext | FileCheck %s --check-prefix=PRINT-VECTOR-SPLAT-2D
+// RUN: mlir-opt %s -pass-pipeline="func.func(convert-linalg-to-loops,convert-scf-to-cf,convert-arith-to-llvm),convert-linalg-to-llvm,convert-memref-to-llvm,convert-func-to-llvm,reconcile-unrealized-casts" | mlir-cpu-runner -e print_0d -entry-point-result=void -shared-libs=%linalg_test_lib_dir/libmlir_runner_utils%shlibext -shared-libs=%linalg_test_lib_dir/libmlir_c_runner_utils%shlibext | FileCheck %s --check-prefix=PRINT-0D
+// RUN: mlir-opt %s -pass-pipeline="func.func(convert-linalg-to-loops,convert-scf-to-cf,convert-arith-to-llvm),convert-linalg-to-llvm,convert-memref-to-llvm,convert-func-to-llvm,reconcile-unrealized-casts" | mlir-cpu-runner -e print_1d -entry-point-result=void -shared-libs=%linalg_test_lib_dir/libmlir_runner_utils%shlibext -shared-libs=%linalg_test_lib_dir/libmlir_c_runner_utils%shlibext | FileCheck %s --check-prefix=PRINT-1D
+// RUN: mlir-opt %s -pass-pipeline="func.func(convert-linalg-to-loops,convert-scf-to-cf,convert-arith-to-llvm),convert-linalg-to-llvm,convert-memref-to-llvm,convert-func-to-llvm,reconcile-unrealized-casts" | mlir-cpu-runner -e print_3d -entry-point-result=void -shared-libs=%linalg_test_lib_dir/libmlir_runner_utils%shlibext -shared-libs=%linalg_test_lib_dir/libmlir_c_runner_utils%shlibext | FileCheck %s --check-prefix=PRINT-3D
+// RUN: mlir-opt %s -pass-pipeline="func.func(convert-linalg-to-loops,convert-scf-to-cf,convert-arith-to-llvm),convert-linalg-to-llvm,convert-memref-to-llvm,convert-func-to-llvm,reconcile-unrealized-casts" | mlir-cpu-runner -e vector_splat_2d -entry-point-result=void -shared-libs=%linalg_test_lib_dir/libmlir_runner_utils%shlibext -shared-libs=%linalg_test_lib_dir/libmlir_c_runner_utils%shlibext | FileCheck %s --check-prefix=PRINT-VECTOR-SPLAT-2D
 
 func.func @print_0d() {
   %f = arith.constant 2.00000e+00 : f32
   %A = memref.alloc() : memref<f32>
   memref.store %f, %A[]: memref<f32>
   %U = memref.cast %A :  memref<f32> to memref<*xf32>
   call @printMemrefF32(%U): (memref<*xf32>) -> ()
   memref.dealloc %A : memref<f32>