[mlir] mlir-cpu-runner test's cblas_interface should export functions on Windows

[kernhanda]

Summary

This change fixes the build on Windows, so that cblas_interface.dll exports functions correctly and an implib is created and installed correctly.

Currently, LLVM cannot be consumed on Windows after it has been installed in a location because cblas_interface.lib is not created/installed, thus failing the import check in LLVMExports.cmake.

[kernhanda]

I don't have commit access, so someone else will have to submit this.

[ftynse]

mlir/test/mlir-cpu-runner/include/cblas_interface.h

2

I'm not certain cblas_interface is the right name here, it doesn't seem to contain anything specific to BLAS. Maybe linalg_interface would have been a better name. May be a separate clean up commit after syncing with @nicolasvasilache

14–26

This is a third occurrence of a similar macro definition. Would it be possible to generalize it somehow and have only one?

22

Please add a comment that indicates which #if this is closing

[kernhanda]

Added comments on #endif statements to indicate block they're closing.

[kernhanda]

mlir/test/mlir-cpu-runner/include/cblas_interface.h

2

Yes, I would prefer to make that as a separate change, after consensus has been reached on a new name.

14–26

AFAICT, no. Three separate shared modules are created, each with their own exports. Furthermore, the three headers are all used within cblas_interface, so the macro can't be shared, because defining it would be telling the Windows linker that all the functions should be exported, instead of just the ones that are defined in cblas_interface.

The only other way is to have a shared preprocessor symbol between the three modules, but set it to different numerical values, which would be the way the different usages are identified.

Thoughts? I would suggest leaving it as is.

[ftynse]

mlir/test/mlir-cpu-runner/include/cblas_interface.h

14–26

Works for me. This seems to be the last "library" file for the runner, so the number of macros should not grow uncontrollably.

[kernhanda]

Not sure if we're waiting for more approvals, but in case we're not, just a reminder that someone else will have to commit this on my behalf. :) Thanks!

mlir/test/mlir-cpu-runner/CMakeLists.txt

 set(LLVM_OPTIONAL_SOURCES
   cblas.cpp
   cblas_interface.cpp
   mlir_runner_utils.cpp
   )
 
 add_llvm_library(cblas SHARED cblas.cpp)
 target_compile_definitions(cblas PRIVATE cblas_EXPORTS)
 
 add_llvm_library(cblas_interface SHARED cblas_interface.cpp)
 target_link_libraries(cblas_interface PRIVATE cblas)
+target_compile_definitions(cblas_interface PRIVATE cblas_interface_EXPORTS)
 
 add_llvm_library(mlir_runner_utils SHARED mlir_runner_utils.cpp)
 target_compile_definitions(mlir_runner_utils PRIVATE mlir_runner_utils_EXPORTS)
-

mlir/test/mlir-cpu-runner/cblas_interface.cpp

 //===- cblas_interface.cpp - Simple Blas subset interface -----------------===//
 //
 // Part of the MLIR Project, under the Apache License v2.0 with LLVM Exceptions.
 // See https://llvm.org/LICENSE.txt for license information.
 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 //
 //===----------------------------------------------------------------------===//
 //
 // Simple Blas subset interface implementation.
 //
 //===----------------------------------------------------------------------===//
 
+#include "include/cblas_interface.h"
 #include "include/cblas.h"
 #include <assert.h>
 #include <iostream>
 
 extern "C" void linalg_fill_viewf32_f32(StridedMemRefType<float, 0> *X,
                                         float f) {
   X->data[X->offset] = f;
 }

mlir/test/mlir-cpu-runner/include/cblas.h

 #ifdef _WIN32
 #ifndef MLIR_CBLAS_EXPORT
 #ifdef cblas_EXPORTS
 /* We are building this library */
 #define MLIR_CBLAS_EXPORT __declspec(dllexport)
 #else
 /* We are using this library */
 #define MLIR_CBLAS_EXPORT __declspec(dllimport)
-#endif
-#endif
+#endif // cblas_EXPORTS
+#endif // MLIR_CBLAS_EXPORT
 #else
 #define MLIR_CBLAS_EXPORT
-#endif
+#endif // _WIN32
 
 /// This reproduces a minimal subset of cblas to allow integration testing
 /// without explicitly requiring a dependence on an external library.
 /// Without loss of generality, various cblas implementations may be swapped in
 /// by including the proper headers and linking with the proper library.
 enum CBLAS_ORDER { CblasRowMajor = 101, CblasColMajor = 102 };
 enum CBLAS_TRANSPOSE {
   CblasNoTrans = 111,

mlir/test/mlir-cpu-runner/include/cblas_interface.h

+//===- cblas_interface.h - Simple Blas subset interface -------------------===//
+//
+// Part of the MLIR Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+#ifndef MLIR_CPU_RUNNER_CBLAS_INTERFACE_H_
+#define MLIR_CPU_RUNNER_CBLAS_INTERFACE_H_
+
+#include "mlir_runner_utils.h"
+
+#ifdef _WIN32
+#ifndef MLIR_CBLAS_INTERFACE_EXPORT
+#ifdef cblas_interface_EXPORTS
+/* We are building this library */
+#define MLIR_CBLAS_INTERFACE_EXPORT __declspec(dllexport)
+#else
+/* We are using this library */
+#define MLIR_CBLAS_INTERFACE_EXPORT __declspec(dllimport)
+#endif // cblas_interface_EXPORTS
+#endif // MLIR_CBLAS_INTERFACE_EXPORT
+#else
+#define MLIR_CBLAS_INTERFACE_EXPORT
+#endif // _WIN32
+
+extern "C" MLIR_CBLAS_INTERFACE_EXPORT void
+linalg_fill_viewf32_f32(StridedMemRefType<float, 0> *X, float f);
+
+extern "C" MLIR_CBLAS_INTERFACE_EXPORT void
+linalg_fill_viewsxf32_f32(StridedMemRefType<float, 1> *X, float f);
+
+extern "C" MLIR_CBLAS_INTERFACE_EXPORT void
+linalg_fill_viewsxsxf32_f32(StridedMemRefType<float, 2> *X, float f);
+
+extern "C" MLIR_CBLAS_INTERFACE_EXPORT void
+linalg_copy_viewf32_viewf32(StridedMemRefType<float, 0> *I,
+                            StridedMemRefType<float, 0> *O);
+
+extern "C" MLIR_CBLAS_INTERFACE_EXPORT void
+linalg_copy_viewsxf32_viewsxf32(StridedMemRefType<float, 1> *I,
+                                StridedMemRefType<float, 1> *O);
+
+extern "C" MLIR_CBLAS_INTERFACE_EXPORT void
+linalg_copy_viewsxsxf32_viewsxsxf32(StridedMemRefType<float, 2> *I,
+                                    StridedMemRefType<float, 2> *O);
+
+extern "C" MLIR_CBLAS_INTERFACE_EXPORT void
+linalg_dot_viewsxf32_viewsxf32_viewf32(StridedMemRefType<float, 1> *X,
+                                       StridedMemRefType<float, 1> *Y,
+                                       StridedMemRefType<float, 0> *Z);
+
+extern "C" MLIR_CBLAS_INTERFACE_EXPORT void
+linalg_matmul_viewsxsxf32_viewsxsxf32_viewsxsxf32(
+    StridedMemRefType<float, 2> *A, StridedMemRefType<float, 2> *B,
+    StridedMemRefType<float, 2> *C);
+
+#endif // MLIR_CPU_RUNNER_CBLAS_INTERFACE_H_

mlir/test/mlir-cpu-runner/include/mlir_runner_utils.h

 #ifdef _WIN32
 #ifndef MLIR_RUNNER_UTILS_EXPORT
 #ifdef mlir_runner_utils_EXPORTS
 /* We are building this library */
 #define MLIR_RUNNER_UTILS_EXPORT __declspec(dllexport)
 #else
 /* We are using this library */
 #define MLIR_RUNNER_UTILS_EXPORT __declspec(dllimport)
-#endif
-#endif
+#endif // mlir_runner_utils_EXPORTS
+#endif // MLIR_RUNNER_UTILS_EXPORT
 #else
 #define MLIR_RUNNER_UTILS_EXPORT
-#endif
+#endif // _WIN32
 
 template <typename T, int N> struct StridedMemRefType;
 template <typename StreamType, typename T, int N>
 void printMemRefMetaData(StreamType &os, StridedMemRefType<T, N> &V);
 
 template <int N> void dropFront(int64_t arr[N], int64_t *res) {
   for (unsigned i = 1; i < N; ++i)
     *(res + i - 1) = arr[i];