[fir] Add fir transformational intrinsic builder

[clementval]

Summary

This patch adds the builder to generate transformational
intrinsic runtime API calls.

This patch is part of the upstreaming effort from fir-dev branch.

Co-authored-by: Jean Perier <[e-mail]>
Co-authored-by: mleair <[e-mail]>
Co-authored-by: Kiran Chandramohan <[e-mail]>
Co-authored-by: Peter Steinfeld <[e-mail]>

[clementval]

Rebase on D114460

[clementval]

Rebase on top of D114557

[rovka]

LGTM with nit.

flang/lib/Optimizer/Builder/Runtime/Transformational.cpp

90

Should we have one for MatmultDirect as well?

[clementval]

flang/lib/Optimizer/Builder/Runtime/Transformational.cpp

90

It is not needed by lowering in fir-dev so far so I would prefer to not add it until it is needed.

flang/include/flang/Optimizer/Builder/Runtime/Transformational.h

+//===-- Transformational.h --------------------------------------*- C++ -*-===//
+// Generate transformational intrinsic runtime API calls.
+//
+// Part of the LLVM 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 FORTRAN_OPTIMIZER_BUILDER_RUNTIME_TRANSFORMATIONAL_H
+#define FORTRAN_OPTIMIZER_BUILDER_RUNTIME_TRANSFORMATIONAL_H
+
+#include "mlir/Dialect/StandardOps/IR/Ops.h"
+
+namespace fir {
+class ExtendedValue;
+class FirOpBuilder;
+} // namespace fir
+
+namespace fir::runtime {
+
+void genCshift(fir::FirOpBuilder &builder, mlir::Location loc,
+               mlir::Value resultBox, mlir::Value arrayBox,
+               mlir::Value shiftBox, mlir::Value dimBox);
+
+void genCshiftVector(fir::FirOpBuilder &builder, mlir::Location loc,
+                     mlir::Value resultBox, mlir::Value arrayBox,
+                     mlir::Value shiftBox);
+
+void genEoshift(fir::FirOpBuilder &builder, mlir::Location loc,
+                mlir::Value resultBox, mlir::Value arrayBox,
+                mlir::Value shiftBox, mlir::Value boundBox, mlir::Value dimBox);
+
+void genEoshiftVector(fir::FirOpBuilder &builder, mlir::Location loc,
+                      mlir::Value resultBox, mlir::Value arrayBox,
+                      mlir::Value shiftBox, mlir::Value boundBox);
+
+void genMatmul(fir::FirOpBuilder &builder, mlir::Location loc,
+               mlir::Value matrixABox, mlir::Value matrixBBox,
+               mlir::Value resultBox);
+
+void genPack(fir::FirOpBuilder &builder, mlir::Location loc,
+             mlir::Value resultBox, mlir::Value arrayBox, mlir::Value maskBox,
+             mlir::Value vectorBox);
+
+void genReshape(fir::FirOpBuilder &builder, mlir::Location loc,
+                mlir::Value resultBox, mlir::Value sourceBox,
+                mlir::Value shapeBox, mlir::Value padBox, mlir::Value orderBox);
+
+void genSpread(fir::FirOpBuilder &builder, mlir::Location loc,
+               mlir::Value resultBox, mlir::Value sourceBox, mlir::Value dim,
+               mlir::Value ncopies);
+
+void genTranspose(fir::FirOpBuilder &builder, mlir::Location loc,
+                  mlir::Value resultBox, mlir::Value sourceBox);
+
+void genUnpack(fir::FirOpBuilder &builder, mlir::Location loc,
+               mlir::Value resultBox, mlir::Value vectorBox,
+               mlir::Value maskBox, mlir::Value fieldBox);
+
+} // namespace fir::runtime
+
+#endif // FORTRAN_OPTIMIZER_BUILDER_RUNTIME_TRANSFORMATIONAL_H

flang/lib/Optimizer/Builder/CMakeLists.txt

 get_property(dialect_libs GLOBAL PROPERTY MLIR_DIALECT_LIBS)
 
 add_flang_library(FIRBuilder
   BoxValue.cpp
   Character.cpp
   Complex.cpp
   DoLoopHelper.cpp
   FIRBuilder.cpp
   MutableBox.cpp
+  Runtime/Transformational.cpp
 
   DEPENDS
   FIRDialect
   FIRSupport
   ${dialect_libs}
 
   LINK_LIBS
   FIRDialect
   FIRSupport
   ${dialect_libs}
 )

flang/lib/Optimizer/Builder/Runtime/Transformational.cpp

+//===-- Transformational.cpp ------------------------------------*- C++ -*-===//
+// Generate transformational intrinsic runtime API calls.
+//
+// Part of the LLVM 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
+//
+//===----------------------------------------------------------------------===//
+
+#include "flang/Optimizer/Builder/Runtime/Transformational.h"
+#include "flang/Lower/Todo.h"
+#include "flang/Optimizer/Builder/BoxValue.h"
+#include "flang/Optimizer/Builder/Character.h"
+#include "flang/Optimizer/Builder/FIRBuilder.h"
+#include "flang/Optimizer/Builder/Runtime/RTBuilder.h"
+#include "flang/Runtime/matmul.h"
+#include "flang/Runtime/transformational.h"
+#include "mlir/Dialect/StandardOps/IR/Ops.h"
+
+using namespace Fortran::runtime;
+
+/// Generate call to Cshift intrinsic
+void fir::runtime::genCshift(fir::FirOpBuilder &builder, mlir::Location loc,
+                             mlir::Value resultBox, mlir::Value arrayBox,
+                             mlir::Value shiftBox, mlir::Value dimBox) {
+  auto cshiftFunc = fir::runtime::getRuntimeFunc<mkRTKey(Cshift)>(loc, builder);
+  auto fTy = cshiftFunc.getType();
+  auto sourceFile = fir::factory::locationToFilename(builder, loc);
+  auto sourceLine =
+      fir::factory::locationToLineNo(builder, loc, fTy.getInput(5));
+  auto args =
+      fir::runtime::createArguments(builder, loc, fTy, resultBox, arrayBox,
+                                    shiftBox, dimBox, sourceFile, sourceLine);
+  builder.create<fir::CallOp>(loc, cshiftFunc, args);
+}
+
+/// Generate call to the vector version of the Cshift intrinsic
+void fir::runtime::genCshiftVector(fir::FirOpBuilder &builder,
+                                   mlir::Location loc, mlir::Value resultBox,
+                                   mlir::Value arrayBox, mlir::Value shiftBox) {
+  auto cshiftFunc =
+      fir::runtime::getRuntimeFunc<mkRTKey(CshiftVector)>(loc, builder);
+  auto fTy = cshiftFunc.getType();
+
+  auto sourceFile = fir::factory::locationToFilename(builder, loc);
+  auto sourceLine =
+      fir::factory::locationToLineNo(builder, loc, fTy.getInput(4));
+  auto args = fir::runtime::createArguments(
+      builder, loc, fTy, resultBox, arrayBox, shiftBox, sourceFile, sourceLine);
+  builder.create<fir::CallOp>(loc, cshiftFunc, args);
+}
+
+/// Generate call to Eoshift intrinsic
+void fir::runtime::genEoshift(fir::FirOpBuilder &builder, mlir::Location loc,
+                              mlir::Value resultBox, mlir::Value arrayBox,
+                              mlir::Value shiftBox, mlir::Value boundBox,
+                              mlir::Value dimBox) {
+  auto eoshiftFunc =
+      fir::runtime::getRuntimeFunc<mkRTKey(Eoshift)>(loc, builder);
+  auto fTy = eoshiftFunc.getType();
+  auto sourceFile = fir::factory::locationToFilename(builder, loc);
+  auto sourceLine =
+      fir::factory::locationToLineNo(builder, loc, fTy.getInput(6));
+  auto args = fir::runtime::createArguments(builder, loc, fTy, resultBox,
+                                            arrayBox, shiftBox, boundBox,
+                                            dimBox, sourceFile, sourceLine);
+  builder.create<fir::CallOp>(loc, eoshiftFunc, args);
+}
+
+/// Generate call to the vector version of the Eoshift intrinsic
+void fir::runtime::genEoshiftVector(fir::FirOpBuilder &builder,
+                                    mlir::Location loc, mlir::Value resultBox,
+                                    mlir::Value arrayBox, mlir::Value shiftBox,
+                                    mlir::Value boundBox) {
+  auto eoshiftFunc =
+      fir::runtime::getRuntimeFunc<mkRTKey(EoshiftVector)>(loc, builder);
+  auto fTy = eoshiftFunc.getType();
+
+  auto sourceFile = fir::factory::locationToFilename(builder, loc);
+  auto sourceLine =
+      fir::factory::locationToLineNo(builder, loc, fTy.getInput(5));
+
+  auto args =
+      fir::runtime::createArguments(builder, loc, fTy, resultBox, arrayBox,
+                                    shiftBox, boundBox, sourceFile, sourceLine);
+  builder.create<fir::CallOp>(loc, eoshiftFunc, args);
+}
+
+/// Generate call to Matmul intrinsic runtime routine.
+void fir::runtime::genMatmul(fir::FirOpBuilder &builder, mlir::Location loc,
+                             mlir::Value resultBox, mlir::Value matrixABox,
+                             mlir::Value matrixBBox) {
+  auto func = fir::runtime::getRuntimeFunc<mkRTKey(Matmul)>(loc, builder);
+  auto fTy = func.getType();
+  auto sourceFile = fir::factory::locationToFilename(builder, loc);
+  auto sourceLine =
+      fir::factory::locationToLineNo(builder, loc, fTy.getInput(4));
+  auto args =
+      fir::runtime::createArguments(builder, loc, fTy, resultBox, matrixABox,
+                                    matrixBBox, sourceFile, sourceLine);
+  builder.create<fir::CallOp>(loc, func, args);
+}
+
+/// Generate call to Pack intrinsic runtime routine.
+void fir::runtime::genPack(fir::FirOpBuilder &builder, mlir::Location loc,
+                           mlir::Value resultBox, mlir::Value arrayBox,
+                           mlir::Value maskBox, mlir::Value vectorBox) {
+  auto packFunc = fir::runtime::getRuntimeFunc<mkRTKey(Pack)>(loc, builder);
+  auto fTy = packFunc.getType();
+  auto sourceFile = fir::factory::locationToFilename(builder, loc);
+  auto sourceLine =
+      fir::factory::locationToLineNo(builder, loc, fTy.getInput(5));
+  auto args =
+      fir::runtime::createArguments(builder, loc, fTy, resultBox, arrayBox,
+                                    maskBox, vectorBox, sourceFile, sourceLine);
+  builder.create<fir::CallOp>(loc, packFunc, args);
+}
+
+/// Generate call to Reshape intrinsic runtime routine.
+void fir::runtime::genReshape(fir::FirOpBuilder &builder, mlir::Location loc,
+                              mlir::Value resultBox, mlir::Value sourceBox,
+                              mlir::Value shapeBox, mlir::Value padBox,
+                              mlir::Value orderBox) {
+  auto func = fir::runtime::getRuntimeFunc<mkRTKey(Reshape)>(loc, builder);
+  auto fTy = func.getType();
+  auto sourceFile = fir::factory::locationToFilename(builder, loc);
+  auto sourceLine =
+      fir::factory::locationToLineNo(builder, loc, fTy.getInput(6));
+  auto args = fir::runtime::createArguments(builder, loc, fTy, resultBox,
+                                            sourceBox, shapeBox, padBox,
+                                            orderBox, sourceFile, sourceLine);
+  builder.create<fir::CallOp>(loc, func, args);
+}
+
+/// Generate call to Spread intrinsic runtime routine.
+void fir::runtime::genSpread(fir::FirOpBuilder &builder, mlir::Location loc,
+                             mlir::Value resultBox, mlir::Value sourceBox,
+                             mlir::Value dim, mlir::Value ncopies) {
+  auto func = fir::runtime::getRuntimeFunc<mkRTKey(Spread)>(loc, builder);
+  auto fTy = func.getType();
+  auto sourceFile = fir::factory::locationToFilename(builder, loc);
+  auto sourceLine =
+      fir::factory::locationToLineNo(builder, loc, fTy.getInput(5));
+  auto args =
+      fir::runtime::createArguments(builder, loc, fTy, resultBox, sourceBox,
+                                    dim, ncopies, sourceFile, sourceLine);
+  builder.create<fir::CallOp>(loc, func, args);
+}
+
+/// Generate call to Transpose intrinsic runtime routine.
+void fir::runtime::genTranspose(fir::FirOpBuilder &builder, mlir::Location loc,
+                                mlir::Value resultBox, mlir::Value sourceBox) {
+  auto func = fir::runtime::getRuntimeFunc<mkRTKey(Transpose)>(loc, builder);
+  auto fTy = func.getType();
+  auto sourceFile = fir::factory::locationToFilename(builder, loc);
+  auto sourceLine =
+      fir::factory::locationToLineNo(builder, loc, fTy.getInput(3));
+  auto args = fir::runtime::createArguments(builder, loc, fTy, resultBox,
+                                            sourceBox, sourceFile, sourceLine);
+  builder.create<fir::CallOp>(loc, func, args);
+}
+
+/// Generate call to Unpack intrinsic runtime routine.
+void fir::runtime::genUnpack(fir::FirOpBuilder &builder, mlir::Location loc,
+                             mlir::Value resultBox, mlir::Value vectorBox,
+                             mlir::Value maskBox, mlir::Value fieldBox) {
+  auto unpackFunc = fir::runtime::getRuntimeFunc<mkRTKey(Unpack)>(loc, builder);
+  auto fTy = unpackFunc.getType();
+  auto sourceFile = fir::factory::locationToFilename(builder, loc);
+  auto sourceLine =
+      fir::factory::locationToLineNo(builder, loc, fTy.getInput(5));
+  auto args =
+      fir::runtime::createArguments(builder, loc, fTy, resultBox, vectorBox,
+                                    maskBox, fieldBox, sourceFile, sourceLine);
+  builder.create<fir::CallOp>(loc, unpackFunc, args);
+}

flang/unittests/Optimizer/Builder/Runtime/TransformationalTest.cpp

+//===- TransformationalTest.cpp -- Transformational intrinsic generation --===//
+//
+// Part of the LLVM 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
+//
+//===----------------------------------------------------------------------===//
+
+#include "flang/Optimizer/Builder/Runtime/Transformational.h"
+#include "RuntimeCallTestBase.h"
+#include "gtest/gtest.h"
+
+TEST_F(RuntimeCallTest, genCshiftTest) {
+  auto loc = firBuilder->getUnknownLoc();
+  mlir::Type seqTy =
+      fir::SequenceType::get(fir::SequenceType::Shape(1, 10), i32Ty);
+  mlir::Value result = firBuilder->create<fir::UndefOp>(loc, seqTy);
+  mlir::Value array = firBuilder->create<fir::UndefOp>(loc, seqTy);
+  mlir::Value shift = firBuilder->create<fir::UndefOp>(loc, seqTy);
+  mlir::Value dim = firBuilder->create<fir::UndefOp>(loc, seqTy);
+  fir::runtime::genCshift(*firBuilder, loc, result, array, shift, dim);
+  checkCallOpFromResultBox(result, "_FortranACshift", 4);
+}
+
+TEST_F(RuntimeCallTest, genCshiftVectorTest) {
+  auto loc = firBuilder->getUnknownLoc();
+  mlir::Type seqTy =
+      fir::SequenceType::get(fir::SequenceType::Shape(1, 10), i32Ty);
+  mlir::Value result = firBuilder->create<fir::UndefOp>(loc, seqTy);
+  mlir::Value array = firBuilder->create<fir::UndefOp>(loc, seqTy);
+  mlir::Value shift = firBuilder->create<fir::UndefOp>(loc, seqTy);
+  fir::runtime::genCshiftVector(*firBuilder, loc, result, array, shift);
+  checkCallOpFromResultBox(result, "_FortranACshiftVector", 3);
+}
+
+TEST_F(RuntimeCallTest, genEoshiftTest) {
+  auto loc = firBuilder->getUnknownLoc();
+  mlir::Type seqTy =
+      fir::SequenceType::get(fir::SequenceType::Shape(1, 10), i32Ty);
+  mlir::Value result = firBuilder->create<fir::UndefOp>(loc, seqTy);
+  mlir::Value array = firBuilder->create<fir::UndefOp>(loc, seqTy);
+  mlir::Value shift = firBuilder->create<fir::UndefOp>(loc, seqTy);
+  mlir::Value bound = firBuilder->create<fir::UndefOp>(loc, seqTy);
+  mlir::Value dim = firBuilder->create<fir::UndefOp>(loc, seqTy);
+  fir::runtime::genEoshift(*firBuilder, loc, result, array, shift, bound, dim);
+  checkCallOpFromResultBox(result, "_FortranAEoshift", 5);
+}
+
+TEST_F(RuntimeCallTest, genEoshiftVectorTest) {
+  auto loc = firBuilder->getUnknownLoc();
+  mlir::Type seqTy =
+      fir::SequenceType::get(fir::SequenceType::Shape(1, 10), i32Ty);
+  mlir::Value result = firBuilder->create<fir::UndefOp>(loc, seqTy);
+  mlir::Value array = firBuilder->create<fir::UndefOp>(loc, seqTy);
+  mlir::Value shift = firBuilder->create<fir::UndefOp>(loc, seqTy);
+  mlir::Value bound = firBuilder->create<fir::UndefOp>(loc, seqTy);
+  fir::runtime::genEoshiftVector(*firBuilder, loc, result, array, shift, bound);
+  checkCallOpFromResultBox(result, "_FortranAEoshiftVector", 4);
+}
+
+TEST_F(RuntimeCallTest, genMatmulTest) {
+  auto loc = firBuilder->getUnknownLoc();
+  mlir::Type seqTy =
+      fir::SequenceType::get(fir::SequenceType::Shape(1, 10), i32Ty);
+  mlir::Value result = firBuilder->create<fir::UndefOp>(loc, seqTy);
+  mlir::Value matrixA = firBuilder->create<fir::UndefOp>(loc, seqTy);
+  mlir::Value matrixB = firBuilder->create<fir::UndefOp>(loc, seqTy);
+  fir::runtime::genMatmul(*firBuilder, loc, matrixA, matrixB, result);
+  checkCallOpFromResultBox(result, "_FortranAMatmul", 3);
+}
+
+TEST_F(RuntimeCallTest, genPackTest) {
+  auto loc = firBuilder->getUnknownLoc();
+  mlir::Type seqTy =
+      fir::SequenceType::get(fir::SequenceType::Shape(1, 10), i32Ty);
+  mlir::Value result = firBuilder->create<fir::UndefOp>(loc, seqTy);
+  mlir::Value array = firBuilder->create<fir::UndefOp>(loc, seqTy);
+  mlir::Value mask = firBuilder->create<fir::UndefOp>(loc, seqTy);
+  mlir::Value vector = firBuilder->create<fir::UndefOp>(loc, seqTy);
+  fir::runtime::genPack(*firBuilder, loc, result, array, mask, vector);
+  checkCallOpFromResultBox(result, "_FortranAPack", 4);
+}
+
+TEST_F(RuntimeCallTest, genReshapeTest) {
+  auto loc = firBuilder->getUnknownLoc();
+  mlir::Type seqTy =
+      fir::SequenceType::get(fir::SequenceType::Shape(1, 10), i32Ty);
+  mlir::Value result = firBuilder->create<fir::UndefOp>(loc, seqTy);
+  mlir::Value source = firBuilder->create<fir::UndefOp>(loc, seqTy);
+  mlir::Value shape = firBuilder->create<fir::UndefOp>(loc, seqTy);
+  mlir::Value pad = firBuilder->create<fir::UndefOp>(loc, seqTy);
+  mlir::Value order = firBuilder->create<fir::UndefOp>(loc, seqTy);
+  fir::runtime::genReshape(*firBuilder, loc, result, source, shape, pad, order);
+  checkCallOpFromResultBox(result, "_FortranAReshape", 5);
+}
+
+TEST_F(RuntimeCallTest, genSpreadTest) {
+  auto loc = firBuilder->getUnknownLoc();
+  mlir::Type seqTy =
+      fir::SequenceType::get(fir::SequenceType::Shape(1, 10), i32Ty);
+  mlir::Value result = firBuilder->create<fir::UndefOp>(loc, seqTy);
+  mlir::Value source = firBuilder->create<fir::UndefOp>(loc, seqTy);
+  mlir::Value dim = firBuilder->create<fir::UndefOp>(loc, seqTy);
+  mlir::Value ncopies = firBuilder->create<fir::UndefOp>(loc, seqTy);
+  fir::runtime::genSpread(*firBuilder, loc, result, source, dim, ncopies);
+  checkCallOpFromResultBox(result, "_FortranASpread", 4);
+}
+
+TEST_F(RuntimeCallTest, genTransposeTest) {
+  auto loc = firBuilder->getUnknownLoc();
+  mlir::Type seqTy =
+      fir::SequenceType::get(fir::SequenceType::Shape(1, 10), i32Ty);
+  mlir::Value result = firBuilder->create<fir::UndefOp>(loc, seqTy);
+  mlir::Value source = firBuilder->create<fir::UndefOp>(loc, seqTy);
+  fir::runtime::genTranspose(*firBuilder, loc, result, source);
+  checkCallOpFromResultBox(result, "_FortranATranspose", 2);
+}
+
+TEST_F(RuntimeCallTest, genUnpack) {
+  auto loc = firBuilder->getUnknownLoc();
+  mlir::Type seqTy =
+      fir::SequenceType::get(fir::SequenceType::Shape(1, 10), i32Ty);
+  mlir::Value result = firBuilder->create<fir::UndefOp>(loc, seqTy);
+  mlir::Value vector = firBuilder->create<fir::UndefOp>(loc, seqTy);
+  mlir::Value mask = firBuilder->create<fir::UndefOp>(loc, seqTy);
+  mlir::Value field = firBuilder->create<fir::UndefOp>(loc, seqTy);
+  fir::runtime::genUnpack(*firBuilder, loc, result, vector, mask, field);
+  checkCallOpFromResultBox(result, "_FortranAUnpack", 4);
+}

flang/unittests/Optimizer/CMakeLists.txt

 get_property(dialect_libs GLOBAL PROPERTY MLIR_DIALECT_LIBS)
 
 set(LIBS
   FIRBuilder
   FIRCodeGen
   FIRDialect
   FIRSupport
   ${dialect_libs}
 )
 
 add_flang_unittest(FlangOptimizerTests
   Builder/CharacterTest.cpp
   Builder/ComplexTest.cpp
   Builder/DoLoopHelperTest.cpp
   Builder/FIRBuilderTest.cpp
+  Builder/Runtime/TransformationalTest.cpp
   FIRContextTest.cpp
   InternalNamesTest.cpp
   KindMappingTest.cpp
   RTBuilder.cpp
 )
 target_link_libraries(FlangOptimizerTests
   PRIVATE
   ${LIBS})