Reland yet again: [mlir] Add FP software implementation lowering pass: arith-to-apfloat #167608
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters. Learn more about bidirectional Unicode characters
Contributor Author
| asan odr: https://lab.llvm.org/buildbot/#/buildrequests/6200991?redirect_to_build=true Both pass on our tests (asan fails somewhere else) |
makslevental force-pushed the users/makslevental/fix-again-again branch from 61fcd2a to 0ab946b Compare makslevental requested review from banach-space, dcaballe and nicolasvasilache as code owners 
makslevental requested review from matthias-springer and removed request for banach-space, dcaballe and nicolasvasilache 
Member
| @llvm/pr-subscribers-mlir-llvm @llvm/pr-subscribers-mlir-execution-engine Author: Maksim Levental (makslevental) ChangesFix both symbol visibility issue in the mlir_apfloat_wrappers lib and the linkage issue in ArithToAPFloat Patch is 30.38 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/167608.diff 17 Files Affected:
diff --git a/mlir/include/mlir/Conversion/ArithToAPFloat/ArithToAPFloat.h b/mlir/include/mlir/Conversion/ArithToAPFloat/ArithToAPFloat.h new file mode 100644 index 0000000000000..64a42a228199e --- /dev/null +++ b/mlir/include/mlir/Conversion/ArithToAPFloat/ArithToAPFloat.h @@ -0,0 +1,21 @@ +//===- ArithToAPFloat.h - Arith to APFloat impl conversion ---*- C++ ----*-===// +// +// Part of the APFloat Project, under the Apache License v2.0 with APFloat +// Exceptions. See https://llvm.org/LICENSE.txt for license information. +// SPDX-License-Identifier: Apache-2.0 WITH APFloat-exception +// +//===----------------------------------------------------------------------===// + +#ifndef MLIR_CONVERSION_ARITHTOAPFLOAT_ARITHTOAPFLOAT_H +#define MLIR_CONVERSION_ARITHTOAPFLOAT_ARITHTOAPFLOAT_H + +#include <memory> + +namespace mlir { +class Pass; + +#define GEN_PASS_DECL_ARITHTOAPFLOATCONVERSIONPASS +#include "mlir/Conversion/Passes.h.inc" +} // namespace mlir + +#endif // MLIR_CONVERSION_ARITHTOAPFLOAT_ARITHTOAPFLOAT_H diff --git a/mlir/include/mlir/Conversion/Passes.h b/mlir/include/mlir/Conversion/Passes.h index 40d866ec7bf10..82bdfd02661a6 100644 --- a/mlir/include/mlir/Conversion/Passes.h +++ b/mlir/include/mlir/Conversion/Passes.h @@ -12,6 +12,7 @@ #include "mlir/Conversion/AMDGPUToROCDL/AMDGPUToROCDL.h" #include "mlir/Conversion/AffineToStandard/AffineToStandard.h" #include "mlir/Conversion/ArithToAMDGPU/ArithToAMDGPU.h" +#include "mlir/Conversion/ArithToAPFloat/ArithToAPFloat.h" #include "mlir/Conversion/ArithToArmSME/ArithToArmSME.h" #include "mlir/Conversion/ArithToEmitC/ArithToEmitCPass.h" #include "mlir/Conversion/ArithToLLVM/ArithToLLVM.h" diff --git a/mlir/include/mlir/Conversion/Passes.td b/mlir/include/mlir/Conversion/Passes.td index 70e3e45c225db..79bc380dbcb7a 100644 --- a/mlir/include/mlir/Conversion/Passes.td +++ b/mlir/include/mlir/Conversion/Passes.td @@ -186,6 +186,21 @@ def ArithToLLVMConversionPass : Pass<"convert-arith-to-llvm"> { ]; } +//===----------------------------------------------------------------------===// +// ArithToAPFloat +//===----------------------------------------------------------------------===// + +def ArithToAPFloatConversionPass + : Pass<"convert-arith-to-apfloat", "ModuleOp"> { + let summary = "Convert Arith ops to APFloat runtime library calls"; + let description = [{ + This pass converts supported Arith ops to APFloat-based runtime library + calls (APFloatWrappers.cpp). APFloat is a software implementation of + floating-point arithmetic operations. + }]; + let dependentDialects = ["func::FuncDialect"]; +} + //===----------------------------------------------------------------------===// // ArithToSPIRV //===----------------------------------------------------------------------===// diff --git a/mlir/include/mlir/Dialect/Func/Utils/Utils.h b/mlir/include/mlir/Dialect/Func/Utils/Utils.h index 3576126a487ac..00d50874a2e8d 100644 --- a/mlir/include/mlir/Dialect/Func/Utils/Utils.h +++ b/mlir/include/mlir/Dialect/Func/Utils/Utils.h @@ -60,6 +60,13 @@ mlir::FailureOr<std::pair<mlir::func::FuncOp, mlir::func::CallOp>> deduplicateArgsOfFuncOp(mlir::RewriterBase &rewriter, mlir::func::FuncOp funcOp, mlir::ModuleOp moduleOp); +/// Look up a FuncOp with signature `resultTypes`(`paramTypes`)` and name +/// `name`. Return a failure if the FuncOp is found but with a different +/// signature. +FailureOr<FuncOp> lookupFnDecl(SymbolOpInterface symTable, StringRef name, + FunctionType funcT, + SymbolTableCollection *symbolTables = nullptr); + } // namespace func } // namespace mlir diff --git a/mlir/include/mlir/Dialect/LLVMIR/FunctionCallUtils.h b/mlir/include/mlir/Dialect/LLVMIR/FunctionCallUtils.h index 8ad9ed18acebd..b09d32022e348 100644 --- a/mlir/include/mlir/Dialect/LLVMIR/FunctionCallUtils.h +++ b/mlir/include/mlir/Dialect/LLVMIR/FunctionCallUtils.h @@ -52,6 +52,10 @@ lookupOrCreatePrintF32Fn(OpBuilder &b, Operation *moduleOp, FailureOr<LLVM::LLVMFuncOp> lookupOrCreatePrintF64Fn(OpBuilder &b, Operation *moduleOp, SymbolTableCollection *symbolTables = nullptr); +FailureOr<LLVM::LLVMFuncOp> +lookupOrCreateApFloatPrintFn(OpBuilder &b, Operation *moduleOp, + SymbolTableCollection *symbolTables = nullptr); + /// Declares a function to print a C-string. /// If a custom runtime function is defined via `runtimeFunctionName`, it must /// have the signature void(char const*). The default function is `printString`. diff --git a/mlir/lib/Conversion/ArithToAPFloat/ArithToAPFloat.cpp b/mlir/lib/Conversion/ArithToAPFloat/ArithToAPFloat.cpp new file mode 100644 index 0000000000000..699edb188a70a --- /dev/null +++ b/mlir/lib/Conversion/ArithToAPFloat/ArithToAPFloat.cpp @@ -0,0 +1,163 @@ +//===- ArithToAPFloat.cpp - Arithmetic to APFloat Conversion --------------===// +// +// 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 "mlir/Conversion/ArithToAPFloat/ArithToAPFloat.h" + +#include "mlir/Dialect/Arith/IR/Arith.h" +#include "mlir/Dialect/Arith/Transforms/Passes.h" +#include "mlir/Dialect/Func/IR/FuncOps.h" +#include "mlir/Dialect/Func/Utils/Utils.h" +#include "mlir/IR/PatternMatch.h" +#include "mlir/IR/Verifier.h" +#include "mlir/Transforms/WalkPatternRewriteDriver.h" + +namespace mlir { +#define GEN_PASS_DEF_ARITHTOAPFLOATCONVERSIONPASS +#include "mlir/Conversion/Passes.h.inc" +} // namespace mlir + +using namespace mlir; +using namespace mlir::func; + +static FuncOp createFnDecl(OpBuilder &b, SymbolOpInterface symTable, + StringRef name, FunctionType funcT, bool setPrivate, + SymbolTableCollection *symbolTables = nullptr) { + OpBuilder::InsertionGuard g(b); + assert(!symTable->getRegion(0).empty() && "expected non-empty region"); + b.setInsertionPointToStart(&symTable->getRegion(0).front()); + FuncOp funcOp = FuncOp::create(b, symTable->getLoc(), name, funcT); + if (setPrivate) + funcOp.setPrivate(); + if (symbolTables) { + SymbolTable &symbolTable = symbolTables->getSymbolTable(symTable); + symbolTable.insert(funcOp, symTable->getRegion(0).front().begin()); + } + return funcOp; +} + +/// Helper function to look up or create the symbol for a runtime library +/// function for a binary arithmetic operation. +/// +/// Parameter 1: APFloat semantics +/// Parameter 2: Left-hand side operand +/// Parameter 3: Right-hand side operand +/// +/// This function will return a failure if the function is found but has an +/// unexpected signature. +/// +static FailureOr<FuncOp> +lookupOrCreateBinaryFn(OpBuilder &b, SymbolOpInterface symTable, StringRef name, + SymbolTableCollection *symbolTables = nullptr) { + auto i32Type = IntegerType::get(symTable->getContext(), 32); + auto i64Type = IntegerType::get(symTable->getContext(), 64); + + std::string funcName = (llvm::Twine("_mlir_apfloat_") + name).str(); + FunctionType funcT = + FunctionType::get(b.getContext(), {i32Type, i64Type, i64Type}, {i64Type}); + FailureOr<FuncOp> func = + lookupFnDecl(symTable, funcName, funcT, symbolTables); + // Failed due to type mismatch. + if (failed(func)) + return func; + // Successfully matched existing decl. + if (*func) + return *func; + + return createFnDecl(b, symTable, funcName, funcT, + /*setPrivate=*/true, symbolTables); +} + +/// Rewrite a binary arithmetic operation to an APFloat function call. +template <typename OpTy> +struct BinaryArithOpToAPFloatConversion final : OpRewritePattern<OpTy> { + BinaryArithOpToAPFloatConversion(MLIRContext *context, + const char *APFloatName, + SymbolOpInterface symTable, + PatternBenefit benefit = 1) + : OpRewritePattern<OpTy>(context, benefit), symTable(symTable), + APFloatName(APFloatName) {}; + + LogicalResult matchAndRewrite(OpTy op, + PatternRewriter &rewriter) const override { + // Get APFloat function from runtime library. + FailureOr<FuncOp> fn = + lookupOrCreateBinaryFn(rewriter, symTable, APFloatName); + if (failed(fn)) + return fn; + + rewriter.setInsertionPoint(op); + // Cast operands to 64-bit integers. + Location loc = op.getLoc(); + auto floatTy = cast<FloatType>(op.getType()); + auto intWType = rewriter.getIntegerType(floatTy.getWidth()); + auto int64Type = rewriter.getI64Type(); + Value lhsBits = arith::ExtUIOp::create( + rewriter, loc, int64Type, + arith::BitcastOp::create(rewriter, loc, intWType, op.getLhs())); + Value rhsBits = arith::ExtUIOp::create( + rewriter, loc, int64Type, + arith::BitcastOp::create(rewriter, loc, intWType, op.getRhs())); + + // Call APFloat function. + int32_t sem = + llvm::APFloatBase::SemanticsToEnum(floatTy.getFloatSemantics()); + Value semValue = arith::ConstantOp::create( + rewriter, loc, rewriter.getI32Type(), + rewriter.getIntegerAttr(rewriter.getI32Type(), sem)); + SmallVector<Value> params = {semValue, lhsBits, rhsBits}; + auto resultOp = + func::CallOp::create(rewriter, loc, TypeRange(rewriter.getI64Type()), + SymbolRefAttr::get(*fn), params); + + // Truncate result to the original width. + Value truncatedBits = arith::TruncIOp::create(rewriter, loc, intWType, + resultOp->getResult(0)); + rewriter.replaceOp( + op, arith::BitcastOp::create(rewriter, loc, floatTy, truncatedBits)); + return success(); + } + + SymbolOpInterface symTable; + const char *APFloatName; +}; + +namespace { +struct ArithToAPFloatConversionPass final + : impl::ArithToAPFloatConversionPassBase<ArithToAPFloatConversionPass> { + using Base::Base; + + void runOnOperation() override; +}; + +void ArithToAPFloatConversionPass::runOnOperation() { + MLIRContext *context = &getContext(); + RewritePatternSet patterns(context); + patterns.add<BinaryArithOpToAPFloatConversion<arith::AddFOp>>(context, "add", + getOperation()); + patterns.add<BinaryArithOpToAPFloatConversion<arith::SubFOp>>( + context, "subtract", getOperation()); + patterns.add<BinaryArithOpToAPFloatConversion<arith::MulFOp>>( + context, "multiply", getOperation()); + patterns.add<BinaryArithOpToAPFloatConversion<arith::DivFOp>>( + context, "divide", getOperation()); + patterns.add<BinaryArithOpToAPFloatConversion<arith::RemFOp>>( + context, "remainder", getOperation()); + LogicalResult result = success(); + ScopedDiagnosticHandler scopedHandler(context, [&result](Diagnostic &diag) { + if (diag.getSeverity() == DiagnosticSeverity::Error) { + result = failure(); + } + // NB: if you don't return failure, no other diag handlers will fire (see + // mlir/lib/IR/Diagnostics.cpp:DiagnosticEngineImpl::emit). + return failure(); + }); + walkAndApplyPatterns(getOperation(), std::move(patterns)); + if (failed(result)) + return signalPassFailure(); +} +} // namespace diff --git a/mlir/lib/Conversion/ArithToAPFloat/CMakeLists.txt b/mlir/lib/Conversion/ArithToAPFloat/CMakeLists.txt new file mode 100644 index 0000000000000..b5ec49c087163 --- /dev/null +++ b/mlir/lib/Conversion/ArithToAPFloat/CMakeLists.txt @@ -0,0 +1,18 @@ +add_mlir_conversion_library(MLIRArithToAPFloat + ArithToAPFloat.cpp + + ADDITIONAL_HEADER_DIRS + ${MLIR_MAIN_INCLUDE_DIR}/mlir/Conversion/ArithToLLVM + + DEPENDS + MLIRConversionPassIncGen + + LINK_COMPONENTS + Core + + LINK_LIBS PUBLIC + MLIRArithDialect + MLIRArithTransforms + MLIRFuncDialect + MLIRFuncUtils + ) diff --git a/mlir/lib/Conversion/ArithToLLVM/ArithToLLVM.cpp b/mlir/lib/Conversion/ArithToLLVM/ArithToLLVM.cpp index b6099902cc337..f2bacc3399144 100644 --- a/mlir/lib/Conversion/ArithToLLVM/ArithToLLVM.cpp +++ b/mlir/lib/Conversion/ArithToLLVM/ArithToLLVM.cpp @@ -14,6 +14,7 @@ #include "mlir/Conversion/LLVMCommon/VectorPattern.h" #include "mlir/Dialect/Arith/IR/Arith.h" #include "mlir/Dialect/Arith/Transforms/Passes.h" +#include "mlir/Dialect/LLVMIR/FunctionCallUtils.h" #include "mlir/Dialect/LLVMIR/LLVMAttrs.h" #include "mlir/Dialect/LLVMIR/LLVMDialect.h" #include "mlir/IR/TypeUtilities.h" diff --git a/mlir/lib/Conversion/CMakeLists.txt b/mlir/lib/Conversion/CMakeLists.txt index bebf1b8fff3f9..613dc6d242ceb 100644 --- a/mlir/lib/Conversion/CMakeLists.txt +++ b/mlir/lib/Conversion/CMakeLists.txt @@ -2,6 +2,7 @@ add_subdirectory(AffineToStandard) add_subdirectory(AMDGPUToROCDL) add_subdirectory(ArithCommon) add_subdirectory(ArithToAMDGPU) +add_subdirectory(ArithToAPFloat) add_subdirectory(ArithToArmSME) add_subdirectory(ArithToEmitC) add_subdirectory(ArithToLLVM) diff --git a/mlir/lib/Conversion/VectorToLLVM/ConvertVectorToLLVM.cpp b/mlir/lib/Conversion/VectorToLLVM/ConvertVectorToLLVM.cpp index 69a317ecd101f..c747e1b59558a 100644 --- a/mlir/lib/Conversion/VectorToLLVM/ConvertVectorToLLVM.cpp +++ b/mlir/lib/Conversion/VectorToLLVM/ConvertVectorToLLVM.cpp @@ -1654,6 +1654,20 @@ class VectorPrintOpConversion : public ConvertOpToLLVMPattern<vector::PrintOp> { return failure(); } } + } else if (auto floatTy = dyn_cast<FloatType>(printType)) { + // Print other floating-point types using the APFloat runtime library. + int32_t sem = + llvm::APFloatBase::SemanticsToEnum(floatTy.getFloatSemantics()); + Value semValue = LLVM::ConstantOp::create( + rewriter, loc, rewriter.getI32Type(), + rewriter.getIntegerAttr(rewriter.getI32Type(), sem)); + Value floatBits = + LLVM::ZExtOp::create(rewriter, loc, rewriter.getI64Type(), value); + printer = + LLVM::lookupOrCreateApFloatPrintFn(rewriter, parent, symbolTables); + emitCall(rewriter, loc, printer.value(), + ValueRange({semValue, floatBits})); + return success(); } else { return failure(); } diff --git a/mlir/lib/Dialect/Func/Utils/Utils.cpp b/mlir/lib/Dialect/Func/Utils/Utils.cpp index b4cb0932ef631..d6dfd0229963c 100644 --- a/mlir/lib/Dialect/Func/Utils/Utils.cpp +++ b/mlir/lib/Dialect/Func/Utils/Utils.cpp @@ -254,3 +254,28 @@ func::deduplicateArgsOfFuncOp(RewriterBase &rewriter, func::FuncOp funcOp, return std::make_pair(*newFuncOpOrFailure, newCallOp); } + +FailureOr<func::FuncOp> +func::lookupFnDecl(SymbolOpInterface symTable, StringRef name, + FunctionType funcT, SymbolTableCollection *symbolTables) { + FuncOp func; + if (symbolTables) { + func = symbolTables->lookupSymbolIn<FuncOp>( + symTable, StringAttr::get(symTable->getContext(), name)); + } else { + func = llvm::dyn_cast_or_null<FuncOp>( + SymbolTable::lookupSymbolIn(symTable, name)); + } + + if (!func) + return func; + + mlir::FunctionType foundFuncT = func.getFunctionType(); + // Assert the signature of the found function is same as expected + if (funcT != foundFuncT) { + return func.emitError("matched function '") + << name << "' but with different type: " << foundFuncT + << " (expected " << funcT << ")"; + } + return func; +} diff --git a/mlir/lib/Dialect/LLVMIR/IR/FunctionCallUtils.cpp b/mlir/lib/Dialect/LLVMIR/IR/FunctionCallUtils.cpp index feaffa34897b6..160b6ae89215c 100644 --- a/mlir/lib/Dialect/LLVMIR/IR/FunctionCallUtils.cpp +++ b/mlir/lib/Dialect/LLVMIR/IR/FunctionCallUtils.cpp @@ -30,6 +30,7 @@ static constexpr llvm::StringRef kPrintF16 = "printF16"; static constexpr llvm::StringRef kPrintBF16 = "printBF16"; static constexpr llvm::StringRef kPrintF32 = "printF32"; static constexpr llvm::StringRef kPrintF64 = "printF64"; +static constexpr llvm::StringRef kPrintApFloat = "printApFloat"; static constexpr llvm::StringRef kPrintString = "printString"; static constexpr llvm::StringRef kPrintOpen = "printOpen"; static constexpr llvm::StringRef kPrintClose = "printClose"; @@ -160,6 +161,16 @@ mlir::LLVM::lookupOrCreatePrintF64Fn(OpBuilder &b, Operation *moduleOp, LLVM::LLVMVoidType::get(moduleOp->getContext()), symbolTables); } +FailureOr<LLVM::LLVMFuncOp> +mlir::LLVM::lookupOrCreateApFloatPrintFn(OpBuilder &b, Operation *moduleOp, + SymbolTableCollection *symbolTables) { + return lookupOrCreateReservedFn( + b, moduleOp, kPrintApFloat, + {IntegerType::get(moduleOp->getContext(), 32), + IntegerType::get(moduleOp->getContext(), 64)}, + LLVM::LLVMVoidType::get(moduleOp->getContext()), symbolTables); +} + static LLVM::LLVMPointerType getCharPtr(MLIRContext *context) { return LLVM::LLVMPointerType::get(context); } diff --git a/mlir/lib/ExecutionEngine/APFloatWrappers.cpp b/mlir/lib/ExecutionEngine/APFloatWrappers.cpp new file mode 100644 index 0000000000000..0a05f7369e556 --- /dev/null +++ b/mlir/lib/ExecutionEngine/APFloatWrappers.cpp @@ -0,0 +1,89 @@ +//===- APFloatWrappers.cpp - Software Implementation of FP Arithmetics --- ===// +// +// 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 +// +//===----------------------------------------------------------------------===// +// +// This file exposes the APFloat infrastructure to MLIR programs as a runtime +// library. APFloat is a software implementation of floating point arithmetics. +// +// On the MLIR side, floating-point values must be bitcasted to 64-bit integers +// before calling a runtime function. If a floating-point type has less than +// 64 bits, it must be zero-extended to 64 bits after bitcasting it to an +// integer. +// +// Runtime functions receive the floating-point operands of the arithmeic +// operation in the form of 64-bit integers, along with the APFloat semantics +// in the form of a 32-bit integer, which will be interpreted as an +// APFloatBase::Semantics enum value. +// +#include "llvm/ADT/APFloat.h" + +#ifdef _WIN32 +#ifndef MLIR_APFLOAT_WRAPPERS_EXPORT +#ifdef mlir_apfloat_wrappers_EXPORTS +// We are building this library +#define MLIR_APFLOAT_WRAPPERS_EXPORT __declspec(dllexport) +#else +// We are using this library +#define MLIR_APFLOAT_WRAPPERS_EXPORT __declspec(dllimport) +#endif // mlir_apfloat_wrappers_EXPORTS +#endif // MLIR_APFLOAT_WRAPPERS_EXPORT +#else +// Non-windows: use visibility attributes. +#define MLIR_APFLOAT_WRAPPERS_EXPORT __attribute__((visibility("default"))) +#endif // _WIN32 + +/// Binary operations without rounding mode. +#define APFLOAT_BINARY_OP(OP) \ + MLIR_APFLOAT_WRAPPERS_EXPORT int64_t _mlir_apfloat_##OP( \ + int32_t semantics, uint64_t a, uint64_t b) { \ + const llvm::fltSemantics &sem = llvm::APFloatBase::EnumToSemantics( \ + static_cast<llvm::APFloatBase::Semantics>(semantics)); \ + unsigned bitWidth = llvm::APFloatBase::semanticsSizeInBits(sem); \ + llvm::APFloat lhs(sem, llvm::APInt(bitWidth, a)); \ + llvm::APFloat rhs(sem, llvm::APInt(bitWidth, b)); \ + lhs.OP(rhs); \ + return lhs.bitcastToAPInt().getZExtValue(); \ + } + +/// Binary operations with rounding mode. +#define APFLOAT_BINARY_OP_ROUNDING_MODE(OP, ROUNDING_MODE) \ + MLIR_APFLOAT_WRAPPERS_EXPORT int64_t _mlir_apfloat_##OP( \ + int32_t semantics, uint64_t a, uint64_t b) { \ + const llvm::fltSemantics &sem = llvm::APFloatBase::EnumToSemantics( \ + static_cast<llvm::APFloatBase::Semantics>(semantics)); \ + unsigned bitWidth = llvm::APFloatBase::semanticsSizeInBits(sem); \ + llvm::APFloat lhs(sem, llvm::APInt(bitWidth, a)); \ + llvm::APFloat rhs(sem, llvm... [truncated] |
Member
| @llvm/pr-subscribers-mlir Author: Maksim Levental (makslevental) ChangesFix both symbol visibility issue in the mlir_apfloat_wrappers lib and the linkage issue in ArithToAPFloat Patch is 30.38 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/167608.diff 17 Files Affected:
diff --git a/mlir/include/mlir/Conversion/ArithToAPFloat/ArithToAPFloat.h b/mlir/include/mlir/Conversion/ArithToAPFloat/ArithToAPFloat.h new file mode 100644 index 0000000000000..64a42a228199e --- /dev/null +++ b/mlir/include/mlir/Conversion/ArithToAPFloat/ArithToAPFloat.h @@ -0,0 +1,21 @@ +//===- ArithToAPFloat.h - Arith to APFloat impl conversion ---*- C++ ----*-===// +// +// Part of the APFloat Project, under the Apache License v2.0 with APFloat +// Exceptions. See https://llvm.org/LICENSE.txt for license information. +// SPDX-License-Identifier: Apache-2.0 WITH APFloat-exception +// +//===----------------------------------------------------------------------===// + +#ifndef MLIR_CONVERSION_ARITHTOAPFLOAT_ARITHTOAPFLOAT_H +#define MLIR_CONVERSION_ARITHTOAPFLOAT_ARITHTOAPFLOAT_H + +#include <memory> + +namespace mlir { +class Pass; + +#define GEN_PASS_DECL_ARITHTOAPFLOATCONVERSIONPASS +#include "mlir/Conversion/Passes.h.inc" +} // namespace mlir + +#endif // MLIR_CONVERSION_ARITHTOAPFLOAT_ARITHTOAPFLOAT_H diff --git a/mlir/include/mlir/Conversion/Passes.h b/mlir/include/mlir/Conversion/Passes.h index 40d866ec7bf10..82bdfd02661a6 100644 --- a/mlir/include/mlir/Conversion/Passes.h +++ b/mlir/include/mlir/Conversion/Passes.h @@ -12,6 +12,7 @@ #include "mlir/Conversion/AMDGPUToROCDL/AMDGPUToROCDL.h" #include "mlir/Conversion/AffineToStandard/AffineToStandard.h" #include "mlir/Conversion/ArithToAMDGPU/ArithToAMDGPU.h" +#include "mlir/Conversion/ArithToAPFloat/ArithToAPFloat.h" #include "mlir/Conversion/ArithToArmSME/ArithToArmSME.h" #include "mlir/Conversion/ArithToEmitC/ArithToEmitCPass.h" #include "mlir/Conversion/ArithToLLVM/ArithToLLVM.h" diff --git a/mlir/include/mlir/Conversion/Passes.td b/mlir/include/mlir/Conversion/Passes.td index 70e3e45c225db..79bc380dbcb7a 100644 --- a/mlir/include/mlir/Conversion/Passes.td +++ b/mlir/include/mlir/Conversion/Passes.td @@ -186,6 +186,21 @@ def ArithToLLVMConversionPass : Pass<"convert-arith-to-llvm"> { ]; } +//===----------------------------------------------------------------------===// +// ArithToAPFloat +//===----------------------------------------------------------------------===// + +def ArithToAPFloatConversionPass + : Pass<"convert-arith-to-apfloat", "ModuleOp"> { + let summary = "Convert Arith ops to APFloat runtime library calls"; + let description = [{ + This pass converts supported Arith ops to APFloat-based runtime library + calls (APFloatWrappers.cpp). APFloat is a software implementation of + floating-point arithmetic operations. + }]; + let dependentDialects = ["func::FuncDialect"]; +} + //===----------------------------------------------------------------------===// // ArithToSPIRV //===----------------------------------------------------------------------===// diff --git a/mlir/include/mlir/Dialect/Func/Utils/Utils.h b/mlir/include/mlir/Dialect/Func/Utils/Utils.h index 3576126a487ac..00d50874a2e8d 100644 --- a/mlir/include/mlir/Dialect/Func/Utils/Utils.h +++ b/mlir/include/mlir/Dialect/Func/Utils/Utils.h @@ -60,6 +60,13 @@ mlir::FailureOr<std::pair<mlir::func::FuncOp, mlir::func::CallOp>> deduplicateArgsOfFuncOp(mlir::RewriterBase &rewriter, mlir::func::FuncOp funcOp, mlir::ModuleOp moduleOp); +/// Look up a FuncOp with signature `resultTypes`(`paramTypes`)` and name +/// `name`. Return a failure if the FuncOp is found but with a different +/// signature. +FailureOr<FuncOp> lookupFnDecl(SymbolOpInterface symTable, StringRef name, + FunctionType funcT, + SymbolTableCollection *symbolTables = nullptr); + } // namespace func } // namespace mlir diff --git a/mlir/include/mlir/Dialect/LLVMIR/FunctionCallUtils.h b/mlir/include/mlir/Dialect/LLVMIR/FunctionCallUtils.h index 8ad9ed18acebd..b09d32022e348 100644 --- a/mlir/include/mlir/Dialect/LLVMIR/FunctionCallUtils.h +++ b/mlir/include/mlir/Dialect/LLVMIR/FunctionCallUtils.h @@ -52,6 +52,10 @@ lookupOrCreatePrintF32Fn(OpBuilder &b, Operation *moduleOp, FailureOr<LLVM::LLVMFuncOp> lookupOrCreatePrintF64Fn(OpBuilder &b, Operation *moduleOp, SymbolTableCollection *symbolTables = nullptr); +FailureOr<LLVM::LLVMFuncOp> +lookupOrCreateApFloatPrintFn(OpBuilder &b, Operation *moduleOp, + SymbolTableCollection *symbolTables = nullptr); + /// Declares a function to print a C-string. /// If a custom runtime function is defined via `runtimeFunctionName`, it must /// have the signature void(char const*). The default function is `printString`. diff --git a/mlir/lib/Conversion/ArithToAPFloat/ArithToAPFloat.cpp b/mlir/lib/Conversion/ArithToAPFloat/ArithToAPFloat.cpp new file mode 100644 index 0000000000000..699edb188a70a --- /dev/null +++ b/mlir/lib/Conversion/ArithToAPFloat/ArithToAPFloat.cpp @@ -0,0 +1,163 @@ +//===- ArithToAPFloat.cpp - Arithmetic to APFloat Conversion --------------===// +// +// 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 "mlir/Conversion/ArithToAPFloat/ArithToAPFloat.h" + +#include "mlir/Dialect/Arith/IR/Arith.h" +#include "mlir/Dialect/Arith/Transforms/Passes.h" +#include "mlir/Dialect/Func/IR/FuncOps.h" +#include "mlir/Dialect/Func/Utils/Utils.h" +#include "mlir/IR/PatternMatch.h" +#include "mlir/IR/Verifier.h" +#include "mlir/Transforms/WalkPatternRewriteDriver.h" + +namespace mlir { +#define GEN_PASS_DEF_ARITHTOAPFLOATCONVERSIONPASS +#include "mlir/Conversion/Passes.h.inc" +} // namespace mlir + +using namespace mlir; +using namespace mlir::func; + +static FuncOp createFnDecl(OpBuilder &b, SymbolOpInterface symTable, + StringRef name, FunctionType funcT, bool setPrivate, + SymbolTableCollection *symbolTables = nullptr) { + OpBuilder::InsertionGuard g(b); + assert(!symTable->getRegion(0).empty() && "expected non-empty region"); + b.setInsertionPointToStart(&symTable->getRegion(0).front()); + FuncOp funcOp = FuncOp::create(b, symTable->getLoc(), name, funcT); + if (setPrivate) + funcOp.setPrivate(); + if (symbolTables) { + SymbolTable &symbolTable = symbolTables->getSymbolTable(symTable); + symbolTable.insert(funcOp, symTable->getRegion(0).front().begin()); + } + return funcOp; +} + +/// Helper function to look up or create the symbol for a runtime library +/// function for a binary arithmetic operation. +/// +/// Parameter 1: APFloat semantics +/// Parameter 2: Left-hand side operand +/// Parameter 3: Right-hand side operand +/// +/// This function will return a failure if the function is found but has an +/// unexpected signature. +/// +static FailureOr<FuncOp> +lookupOrCreateBinaryFn(OpBuilder &b, SymbolOpInterface symTable, StringRef name, + SymbolTableCollection *symbolTables = nullptr) { + auto i32Type = IntegerType::get(symTable->getContext(), 32); + auto i64Type = IntegerType::get(symTable->getContext(), 64); + + std::string funcName = (llvm::Twine("_mlir_apfloat_") + name).str(); + FunctionType funcT = + FunctionType::get(b.getContext(), {i32Type, i64Type, i64Type}, {i64Type}); + FailureOr<FuncOp> func = + lookupFnDecl(symTable, funcName, funcT, symbolTables); + // Failed due to type mismatch. + if (failed(func)) + return func; + // Successfully matched existing decl. + if (*func) + return *func; + + return createFnDecl(b, symTable, funcName, funcT, + /*setPrivate=*/true, symbolTables); +} + +/// Rewrite a binary arithmetic operation to an APFloat function call. +template <typename OpTy> +struct BinaryArithOpToAPFloatConversion final : OpRewritePattern<OpTy> { + BinaryArithOpToAPFloatConversion(MLIRContext *context, + const char *APFloatName, + SymbolOpInterface symTable, + PatternBenefit benefit = 1) + : OpRewritePattern<OpTy>(context, benefit), symTable(symTable), + APFloatName(APFloatName) {}; + + LogicalResult matchAndRewrite(OpTy op, + PatternRewriter &rewriter) const override { + // Get APFloat function from runtime library. + FailureOr<FuncOp> fn = + lookupOrCreateBinaryFn(rewriter, symTable, APFloatName); + if (failed(fn)) + return fn; + + rewriter.setInsertionPoint(op); + // Cast operands to 64-bit integers. + Location loc = op.getLoc(); + auto floatTy = cast<FloatType>(op.getType()); + auto intWType = rewriter.getIntegerType(floatTy.getWidth()); + auto int64Type = rewriter.getI64Type(); + Value lhsBits = arith::ExtUIOp::create( + rewriter, loc, int64Type, + arith::BitcastOp::create(rewriter, loc, intWType, op.getLhs())); + Value rhsBits = arith::ExtUIOp::create( + rewriter, loc, int64Type, + arith::BitcastOp::create(rewriter, loc, intWType, op.getRhs())); + + // Call APFloat function. + int32_t sem = + llvm::APFloatBase::SemanticsToEnum(floatTy.getFloatSemantics()); + Value semValue = arith::ConstantOp::create( + rewriter, loc, rewriter.getI32Type(), + rewriter.getIntegerAttr(rewriter.getI32Type(), sem)); + SmallVector<Value> params = {semValue, lhsBits, rhsBits}; + auto resultOp = + func::CallOp::create(rewriter, loc, TypeRange(rewriter.getI64Type()), + SymbolRefAttr::get(*fn), params); + + // Truncate result to the original width. + Value truncatedBits = arith::TruncIOp::create(rewriter, loc, intWType, + resultOp->getResult(0)); + rewriter.replaceOp( + op, arith::BitcastOp::create(rewriter, loc, floatTy, truncatedBits)); + return success(); + } + + SymbolOpInterface symTable; + const char *APFloatName; +}; + +namespace { +struct ArithToAPFloatConversionPass final + : impl::ArithToAPFloatConversionPassBase<ArithToAPFloatConversionPass> { + using Base::Base; + + void runOnOperation() override; +}; + +void ArithToAPFloatConversionPass::runOnOperation() { + MLIRContext *context = &getContext(); + RewritePatternSet patterns(context); + patterns.add<BinaryArithOpToAPFloatConversion<arith::AddFOp>>(context, "add", + getOperation()); + patterns.add<BinaryArithOpToAPFloatConversion<arith::SubFOp>>( + context, "subtract", getOperation()); + patterns.add<BinaryArithOpToAPFloatConversion<arith::MulFOp>>( + context, "multiply", getOperation()); + patterns.add<BinaryArithOpToAPFloatConversion<arith::DivFOp>>( + context, "divide", getOperation()); + patterns.add<BinaryArithOpToAPFloatConversion<arith::RemFOp>>( + context, "remainder", getOperation()); + LogicalResult result = success(); + ScopedDiagnosticHandler scopedHandler(context, [&result](Diagnostic &diag) { + if (diag.getSeverity() == DiagnosticSeverity::Error) { + result = failure(); + } + // NB: if you don't return failure, no other diag handlers will fire (see + // mlir/lib/IR/Diagnostics.cpp:DiagnosticEngineImpl::emit). + return failure(); + }); + walkAndApplyPatterns(getOperation(), std::move(patterns)); + if (failed(result)) + return signalPassFailure(); +} +} // namespace diff --git a/mlir/lib/Conversion/ArithToAPFloat/CMakeLists.txt b/mlir/lib/Conversion/ArithToAPFloat/CMakeLists.txt new file mode 100644 index 0000000000000..b5ec49c087163 --- /dev/null +++ b/mlir/lib/Conversion/ArithToAPFloat/CMakeLists.txt @@ -0,0 +1,18 @@ +add_mlir_conversion_library(MLIRArithToAPFloat + ArithToAPFloat.cpp + + ADDITIONAL_HEADER_DIRS + ${MLIR_MAIN_INCLUDE_DIR}/mlir/Conversion/ArithToLLVM + + DEPENDS + MLIRConversionPassIncGen + + LINK_COMPONENTS + Core + + LINK_LIBS PUBLIC + MLIRArithDialect + MLIRArithTransforms + MLIRFuncDialect + MLIRFuncUtils + ) diff --git a/mlir/lib/Conversion/ArithToLLVM/ArithToLLVM.cpp b/mlir/lib/Conversion/ArithToLLVM/ArithToLLVM.cpp index b6099902cc337..f2bacc3399144 100644 --- a/mlir/lib/Conversion/ArithToLLVM/ArithToLLVM.cpp +++ b/mlir/lib/Conversion/ArithToLLVM/ArithToLLVM.cpp @@ -14,6 +14,7 @@ #include "mlir/Conversion/LLVMCommon/VectorPattern.h" #include "mlir/Dialect/Arith/IR/Arith.h" #include "mlir/Dialect/Arith/Transforms/Passes.h" +#include "mlir/Dialect/LLVMIR/FunctionCallUtils.h" #include "mlir/Dialect/LLVMIR/LLVMAttrs.h" #include "mlir/Dialect/LLVMIR/LLVMDialect.h" #include "mlir/IR/TypeUtilities.h" diff --git a/mlir/lib/Conversion/CMakeLists.txt b/mlir/lib/Conversion/CMakeLists.txt index bebf1b8fff3f9..613dc6d242ceb 100644 --- a/mlir/lib/Conversion/CMakeLists.txt +++ b/mlir/lib/Conversion/CMakeLists.txt @@ -2,6 +2,7 @@ add_subdirectory(AffineToStandard) add_subdirectory(AMDGPUToROCDL) add_subdirectory(ArithCommon) add_subdirectory(ArithToAMDGPU) +add_subdirectory(ArithToAPFloat) add_subdirectory(ArithToArmSME) add_subdirectory(ArithToEmitC) add_subdirectory(ArithToLLVM) diff --git a/mlir/lib/Conversion/VectorToLLVM/ConvertVectorToLLVM.cpp b/mlir/lib/Conversion/VectorToLLVM/ConvertVectorToLLVM.cpp index 69a317ecd101f..c747e1b59558a 100644 --- a/mlir/lib/Conversion/VectorToLLVM/ConvertVectorToLLVM.cpp +++ b/mlir/lib/Conversion/VectorToLLVM/ConvertVectorToLLVM.cpp @@ -1654,6 +1654,20 @@ class VectorPrintOpConversion : public ConvertOpToLLVMPattern<vector::PrintOp> { return failure(); } } + } else if (auto floatTy = dyn_cast<FloatType>(printType)) { + // Print other floating-point types using the APFloat runtime library. + int32_t sem = + llvm::APFloatBase::SemanticsToEnum(floatTy.getFloatSemantics()); + Value semValue = LLVM::ConstantOp::create( + rewriter, loc, rewriter.getI32Type(), + rewriter.getIntegerAttr(rewriter.getI32Type(), sem)); + Value floatBits = + LLVM::ZExtOp::create(rewriter, loc, rewriter.getI64Type(), value); + printer = + LLVM::lookupOrCreateApFloatPrintFn(rewriter, parent, symbolTables); + emitCall(rewriter, loc, printer.value(), + ValueRange({semValue, floatBits})); + return success(); } else { return failure(); } diff --git a/mlir/lib/Dialect/Func/Utils/Utils.cpp b/mlir/lib/Dialect/Func/Utils/Utils.cpp index b4cb0932ef631..d6dfd0229963c 100644 --- a/mlir/lib/Dialect/Func/Utils/Utils.cpp +++ b/mlir/lib/Dialect/Func/Utils/Utils.cpp @@ -254,3 +254,28 @@ func::deduplicateArgsOfFuncOp(RewriterBase &rewriter, func::FuncOp funcOp, return std::make_pair(*newFuncOpOrFailure, newCallOp); } + +FailureOr<func::FuncOp> +func::lookupFnDecl(SymbolOpInterface symTable, StringRef name, + FunctionType funcT, SymbolTableCollection *symbolTables) { + FuncOp func; + if (symbolTables) { + func = symbolTables->lookupSymbolIn<FuncOp>( + symTable, StringAttr::get(symTable->getContext(), name)); + } else { + func = llvm::dyn_cast_or_null<FuncOp>( + SymbolTable::lookupSymbolIn(symTable, name)); + } + + if (!func) + return func; + + mlir::FunctionType foundFuncT = func.getFunctionType(); + // Assert the signature of the found function is same as expected + if (funcT != foundFuncT) { + return func.emitError("matched function '") + << name << "' but with different type: " << foundFuncT + << " (expected " << funcT << ")"; + } + return func; +} diff --git a/mlir/lib/Dialect/LLVMIR/IR/FunctionCallUtils.cpp b/mlir/lib/Dialect/LLVMIR/IR/FunctionCallUtils.cpp index feaffa34897b6..160b6ae89215c 100644 --- a/mlir/lib/Dialect/LLVMIR/IR/FunctionCallUtils.cpp +++ b/mlir/lib/Dialect/LLVMIR/IR/FunctionCallUtils.cpp @@ -30,6 +30,7 @@ static constexpr llvm::StringRef kPrintF16 = "printF16"; static constexpr llvm::StringRef kPrintBF16 = "printBF16"; static constexpr llvm::StringRef kPrintF32 = "printF32"; static constexpr llvm::StringRef kPrintF64 = "printF64"; +static constexpr llvm::StringRef kPrintApFloat = "printApFloat"; static constexpr llvm::StringRef kPrintString = "printString"; static constexpr llvm::StringRef kPrintOpen = "printOpen"; static constexpr llvm::StringRef kPrintClose = "printClose"; @@ -160,6 +161,16 @@ mlir::LLVM::lookupOrCreatePrintF64Fn(OpBuilder &b, Operation *moduleOp, LLVM::LLVMVoidType::get(moduleOp->getContext()), symbolTables); } +FailureOr<LLVM::LLVMFuncOp> +mlir::LLVM::lookupOrCreateApFloatPrintFn(OpBuilder &b, Operation *moduleOp, + SymbolTableCollection *symbolTables) { + return lookupOrCreateReservedFn( + b, moduleOp, kPrintApFloat, + {IntegerType::get(moduleOp->getContext(), 32), + IntegerType::get(moduleOp->getContext(), 64)}, + LLVM::LLVMVoidType::get(moduleOp->getContext()), symbolTables); +} + static LLVM::LLVMPointerType getCharPtr(MLIRContext *context) { return LLVM::LLVMPointerType::get(context); } diff --git a/mlir/lib/ExecutionEngine/APFloatWrappers.cpp b/mlir/lib/ExecutionEngine/APFloatWrappers.cpp new file mode 100644 index 0000000000000..0a05f7369e556 --- /dev/null +++ b/mlir/lib/ExecutionEngine/APFloatWrappers.cpp @@ -0,0 +1,89 @@ +//===- APFloatWrappers.cpp - Software Implementation of FP Arithmetics --- ===// +// +// 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 +// +//===----------------------------------------------------------------------===// +// +// This file exposes the APFloat infrastructure to MLIR programs as a runtime +// library. APFloat is a software implementation of floating point arithmetics. +// +// On the MLIR side, floating-point values must be bitcasted to 64-bit integers +// before calling a runtime function. If a floating-point type has less than +// 64 bits, it must be zero-extended to 64 bits after bitcasting it to an +// integer. +// +// Runtime functions receive the floating-point operands of the arithmeic +// operation in the form of 64-bit integers, along with the APFloat semantics +// in the form of a 32-bit integer, which will be interpreted as an +// APFloatBase::Semantics enum value. +// +#include "llvm/ADT/APFloat.h" + +#ifdef _WIN32 +#ifndef MLIR_APFLOAT_WRAPPERS_EXPORT +#ifdef mlir_apfloat_wrappers_EXPORTS +// We are building this library +#define MLIR_APFLOAT_WRAPPERS_EXPORT __declspec(dllexport) +#else +// We are using this library +#define MLIR_APFLOAT_WRAPPERS_EXPORT __declspec(dllimport) +#endif // mlir_apfloat_wrappers_EXPORTS +#endif // MLIR_APFLOAT_WRAPPERS_EXPORT +#else +// Non-windows: use visibility attributes. +#define MLIR_APFLOAT_WRAPPERS_EXPORT __attribute__((visibility("default"))) +#endif // _WIN32 + +/// Binary operations without rounding mode. +#define APFLOAT_BINARY_OP(OP) \ + MLIR_APFLOAT_WRAPPERS_EXPORT int64_t _mlir_apfloat_##OP( \ + int32_t semantics, uint64_t a, uint64_t b) { \ + const llvm::fltSemantics &sem = llvm::APFloatBase::EnumToSemantics( \ + static_cast<llvm::APFloatBase::Semantics>(semantics)); \ + unsigned bitWidth = llvm::APFloatBase::semanticsSizeInBits(sem); \ + llvm::APFloat lhs(sem, llvm::APInt(bitWidth, a)); \ + llvm::APFloat rhs(sem, llvm::APInt(bitWidth, b)); \ + lhs.OP(rhs); \ + return lhs.bitcastToAPInt().getZExtValue(); \ + } + +/// Binary operations with rounding mode. +#define APFLOAT_BINARY_OP_ROUNDING_MODE(OP, ROUNDING_MODE) \ + MLIR_APFLOAT_WRAPPERS_EXPORT int64_t _mlir_apfloat_##OP( \ + int32_t semantics, uint64_t a, uint64_t b) { \ + const llvm::fltSemantics &sem = llvm::APFloatBase::EnumToSemantics( \ + static_cast<llvm::APFloatBase::Semantics>(semantics)); \ + unsigned bitWidth = llvm::APFloatBase::semanticsSizeInBits(sem); \ + llvm::APFloat lhs(sem, llvm::APInt(bitWidth, a)); \ + llvm::APFloat rhs(sem, llvm... [truncated] |
matthias-springer approved these changes Nov 13, 2025
git-crd pushed a commit to git-crd/crd-llvm-project that referenced this pull request Nov 13, 2025
…: `arith-to-apfloat` (llvm#167608) Fix both symbol visibility issue in the mlir_apfloat_wrappers lib and the linkage issue in ArithToAPFloat.
makslevental added a commit that referenced this pull request Nov 13, 2025
…ing pass: `arith-to-apfloat`" (#167834) Reverts #167608 Broken builder https://lab.llvm.org/buildbot/#/builders/52/builds/12781
llvm-sync bot pushed a commit to arm/arm-toolchain that referenced this pull request Nov 13, 2025
…ation lowering pass: `arith-to-apfloat`" (#167834) Reverts llvm/llvm-project#167608 Broken builder https://lab.llvm.org/buildbot/#/builders/52/builds/12781
matthias-springer pushed a commit that referenced this pull request Nov 13, 2025
…: `arith-to-apfloat` (#167608) Fix both symbol visibility issue in the mlir_apfloat_wrappers lib and the linkage issue in ArithToAPFloat.
matthias-springer pushed a commit that referenced this pull request Nov 13, 2025
…: `arith-to-apfloat` (#167608) Fix both symbol visibility issue in the mlir_apfloat_wrappers lib and the linkage issue in ArithToAPFloat.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Add this suggestion to a batch that can be applied as a single commit. This suggestion is invalid because no changes were made to the code. Suggestions cannot be applied while the pull request is closed. Suggestions cannot be applied while viewing a subset of changes. Only one suggestion per line can be applied in a batch. Add this suggestion to a batch that can be applied as a single commit. Applying suggestions on deleted lines is not supported. You must change the existing code in this line in order to create a valid suggestion. Outdated suggestions cannot be applied. This suggestion has been applied or marked resolved. Suggestions cannot be applied from pending reviews. Suggestions cannot be applied on multi-line comments. Suggestions cannot be applied while the pull request is queued to merge. Suggestion cannot be applied right now. Please check back later.
Fix both symbol visibility issue in the mlir_apfloat_wrappers lib and the linkage issue in ArithToAPFloat