1
0
mirror of https://github.com/RPCS3/llvm-mirror.git synced 2024-11-22 02:33:06 +01:00
llvm-mirror/lib/Analysis/VFABIDemangling.cpp

483 lines
16 KiB
C++
Raw Permalink Normal View History

//===- VFABIDemangling.cpp - Vector Function ABI demangling utilities. ---===//
//
// 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 "llvm/ADT/SmallSet.h"
#include "llvm/ADT/SmallString.h"
#include "llvm/Analysis/VectorUtils.h"
using namespace llvm;
namespace {
/// Utilities for the Vector Function ABI name parser.
/// Return types for the parser functions.
enum class ParseRet {
OK, // Found.
None, // Not found.
Error // Syntax error.
};
/// Extracts the `<isa>` information from the mangled string, and
/// sets the `ISA` accordingly.
ParseRet tryParseISA(StringRef &MangledName, VFISAKind &ISA) {
if (MangledName.empty())
return ParseRet::Error;
if (MangledName.startswith(VFABI::_LLVM_)) {
MangledName = MangledName.drop_front(strlen(VFABI::_LLVM_));
ISA = VFISAKind::LLVM;
} else {
ISA = StringSwitch<VFISAKind>(MangledName.take_front(1))
.Case("n", VFISAKind::AdvancedSIMD)
.Case("s", VFISAKind::SVE)
.Case("b", VFISAKind::SSE)
.Case("c", VFISAKind::AVX)
.Case("d", VFISAKind::AVX2)
.Case("e", VFISAKind::AVX512)
.Default(VFISAKind::Unknown);
MangledName = MangledName.drop_front(1);
}
return ParseRet::OK;
}
/// Extracts the `<mask>` information from the mangled string, and
/// sets `IsMasked` accordingly. The input string `MangledName` is
/// left unmodified.
ParseRet tryParseMask(StringRef &MangledName, bool &IsMasked) {
if (MangledName.consume_front("M")) {
IsMasked = true;
return ParseRet::OK;
}
if (MangledName.consume_front("N")) {
IsMasked = false;
return ParseRet::OK;
}
return ParseRet::Error;
}
/// Extract the `<vlen>` information from the mangled string, and
/// sets `VF` accordingly. A `<vlen> == "x"` token is interpreted as a scalable
/// vector length. On success, the `<vlen>` token is removed from
/// the input string `ParseString`.
///
ParseRet tryParseVLEN(StringRef &ParseString, unsigned &VF, bool &IsScalable) {
if (ParseString.consume_front("x")) {
[llvm][VectorUtils] Tweak VFShape for scalable vector functions. Summary: This patch makes sure that the field VFShape.VF is greater than zero when demangling the vector function name of scalable vector functions encoded in the "vector-function-abi-variant" attribute. This change is required to be able to provide instances of VFShape that can be used to query the VFDatabase for the vectorization passes, as such passes always require a positive value for the Vectorization Factor (VF) needed by the vectorization process. It is not possible to extract the value of VFShape.VF from the mangled name of scalable vector functions, because it is encoded as `x`. Therefore, the VFABI demangling function has been modified to extract such information from the IR declaration of the vector function, under the assumption that _all_ vectors in the signature of the vector function have the same number of lanes. Such assumption is valid because it is also assumed by the Vector Function ABI specifications supported by the demangling function (x86, AArch64, and LLVM internal one). The unit tests that demangle scalable names have been modified by adding the IR module that carries the declaration of the vector function name being demangled. In particular, the demangling function fails in the following cases: 1. When the declaration of the scalable vector function is not present in the module. 2. When the value of VFSHape.VF is not greater than 0. Reviewers: jdoerfert, sdesmalen, andwar Reviewed By: jdoerfert Subscribers: mgorny, kristof.beyls, hiraditya, llvm-commits Tags: #llvm Differential Revision: https://reviews.llvm.org/D73286
2020-01-22 23:34:27 +01:00
// Set VF to 0, to be later adjusted to a value grater than zero
// by looking at the signature of the vector function with
// `getECFromSignature`.
VF = 0;
IsScalable = true;
return ParseRet::OK;
}
if (ParseString.consumeInteger(10, VF))
return ParseRet::Error;
[llvm][VectorUtils] Tweak VFShape for scalable vector functions. Summary: This patch makes sure that the field VFShape.VF is greater than zero when demangling the vector function name of scalable vector functions encoded in the "vector-function-abi-variant" attribute. This change is required to be able to provide instances of VFShape that can be used to query the VFDatabase for the vectorization passes, as such passes always require a positive value for the Vectorization Factor (VF) needed by the vectorization process. It is not possible to extract the value of VFShape.VF from the mangled name of scalable vector functions, because it is encoded as `x`. Therefore, the VFABI demangling function has been modified to extract such information from the IR declaration of the vector function, under the assumption that _all_ vectors in the signature of the vector function have the same number of lanes. Such assumption is valid because it is also assumed by the Vector Function ABI specifications supported by the demangling function (x86, AArch64, and LLVM internal one). The unit tests that demangle scalable names have been modified by adding the IR module that carries the declaration of the vector function name being demangled. In particular, the demangling function fails in the following cases: 1. When the declaration of the scalable vector function is not present in the module. 2. When the value of VFSHape.VF is not greater than 0. Reviewers: jdoerfert, sdesmalen, andwar Reviewed By: jdoerfert Subscribers: mgorny, kristof.beyls, hiraditya, llvm-commits Tags: #llvm Differential Revision: https://reviews.llvm.org/D73286
2020-01-22 23:34:27 +01:00
// The token `0` is invalid for VLEN.
if (VF == 0)
return ParseRet::Error;
IsScalable = false;
return ParseRet::OK;
}
/// The function looks for the following strings at the beginning of
/// the input string `ParseString`:
///
/// <token> <number>
///
/// On success, it removes the parsed parameter from `ParseString`,
/// sets `PKind` to the correspondent enum value, sets `Pos` to
/// <number>, and return success. On a syntax error, it return a
/// parsing error. If nothing is parsed, it returns None.
///
/// The function expects <token> to be one of "ls", "Rs", "Us" or
/// "Ls".
ParseRet tryParseLinearTokenWithRuntimeStep(StringRef &ParseString,
VFParamKind &PKind, int &Pos,
const StringRef Token) {
if (ParseString.consume_front(Token)) {
PKind = VFABI::getVFParamKindFromString(Token);
if (ParseString.consumeInteger(10, Pos))
return ParseRet::Error;
return ParseRet::OK;
}
return ParseRet::None;
}
/// The function looks for the following stringt at the beginning of
/// the input string `ParseString`:
///
/// <token> <number>
///
/// <token> is one of "ls", "Rs", "Us" or "Ls".
///
/// On success, it removes the parsed parameter from `ParseString`,
/// sets `PKind` to the correspondent enum value, sets `StepOrPos` to
/// <number>, and return success. On a syntax error, it return a
/// parsing error. If nothing is parsed, it returns None.
ParseRet tryParseLinearWithRuntimeStep(StringRef &ParseString,
VFParamKind &PKind, int &StepOrPos) {
ParseRet Ret;
// "ls" <RuntimeStepPos>
Ret = tryParseLinearTokenWithRuntimeStep(ParseString, PKind, StepOrPos, "ls");
if (Ret != ParseRet::None)
return Ret;
// "Rs" <RuntimeStepPos>
Ret = tryParseLinearTokenWithRuntimeStep(ParseString, PKind, StepOrPos, "Rs");
if (Ret != ParseRet::None)
return Ret;
// "Ls" <RuntimeStepPos>
Ret = tryParseLinearTokenWithRuntimeStep(ParseString, PKind, StepOrPos, "Ls");
if (Ret != ParseRet::None)
return Ret;
// "Us" <RuntimeStepPos>
Ret = tryParseLinearTokenWithRuntimeStep(ParseString, PKind, StepOrPos, "Us");
if (Ret != ParseRet::None)
return Ret;
return ParseRet::None;
}
/// The function looks for the following strings at the beginning of
/// the input string `ParseString`:
///
/// <token> {"n"} <number>
///
/// On success, it removes the parsed parameter from `ParseString`,
/// sets `PKind` to the correspondent enum value, sets `LinearStep` to
/// <number>, and return success. On a syntax error, it return a
/// parsing error. If nothing is parsed, it returns None.
///
/// The function expects <token> to be one of "l", "R", "U" or
/// "L".
ParseRet tryParseCompileTimeLinearToken(StringRef &ParseString,
VFParamKind &PKind, int &LinearStep,
const StringRef Token) {
if (ParseString.consume_front(Token)) {
PKind = VFABI::getVFParamKindFromString(Token);
const bool Negate = ParseString.consume_front("n");
if (ParseString.consumeInteger(10, LinearStep))
LinearStep = 1;
if (Negate)
LinearStep *= -1;
return ParseRet::OK;
}
return ParseRet::None;
}
/// The function looks for the following strings at the beginning of
/// the input string `ParseString`:
///
/// ["l" | "R" | "U" | "L"] {"n"} <number>
///
/// On success, it removes the parsed parameter from `ParseString`,
/// sets `PKind` to the correspondent enum value, sets `LinearStep` to
/// <number>, and return success. On a syntax error, it return a
/// parsing error. If nothing is parsed, it returns None.
ParseRet tryParseLinearWithCompileTimeStep(StringRef &ParseString,
VFParamKind &PKind, int &StepOrPos) {
// "l" {"n"} <CompileTimeStep>
if (tryParseCompileTimeLinearToken(ParseString, PKind, StepOrPos, "l") ==
ParseRet::OK)
return ParseRet::OK;
// "R" {"n"} <CompileTimeStep>
if (tryParseCompileTimeLinearToken(ParseString, PKind, StepOrPos, "R") ==
ParseRet::OK)
return ParseRet::OK;
// "L" {"n"} <CompileTimeStep>
if (tryParseCompileTimeLinearToken(ParseString, PKind, StepOrPos, "L") ==
ParseRet::OK)
return ParseRet::OK;
// "U" {"n"} <CompileTimeStep>
if (tryParseCompileTimeLinearToken(ParseString, PKind, StepOrPos, "U") ==
ParseRet::OK)
return ParseRet::OK;
return ParseRet::None;
}
/// Looks into the <parameters> part of the mangled name in search
/// for valid paramaters at the beginning of the string
/// `ParseString`.
///
/// On success, it removes the parsed parameter from `ParseString`,
/// sets `PKind` to the correspondent enum value, sets `StepOrPos`
/// accordingly, and return success. On a syntax error, it return a
/// parsing error. If nothing is parsed, it returns None.
ParseRet tryParseParameter(StringRef &ParseString, VFParamKind &PKind,
int &StepOrPos) {
if (ParseString.consume_front("v")) {
PKind = VFParamKind::Vector;
StepOrPos = 0;
return ParseRet::OK;
}
if (ParseString.consume_front("u")) {
PKind = VFParamKind::OMP_Uniform;
StepOrPos = 0;
return ParseRet::OK;
}
const ParseRet HasLinearRuntime =
tryParseLinearWithRuntimeStep(ParseString, PKind, StepOrPos);
if (HasLinearRuntime != ParseRet::None)
return HasLinearRuntime;
const ParseRet HasLinearCompileTime =
tryParseLinearWithCompileTimeStep(ParseString, PKind, StepOrPos);
if (HasLinearCompileTime != ParseRet::None)
return HasLinearCompileTime;
return ParseRet::None;
}
/// Looks into the <parameters> part of the mangled name in search
/// of a valid 'aligned' clause. The function should be invoked
/// after parsing a parameter via `tryParseParameter`.
///
/// On success, it removes the parsed parameter from `ParseString`,
/// sets `PKind` to the correspondent enum value, sets `StepOrPos`
/// accordingly, and return success. On a syntax error, it return a
/// parsing error. If nothing is parsed, it returns None.
ParseRet tryParseAlign(StringRef &ParseString, Align &Alignment) {
uint64_t Val;
// "a" <number>
if (ParseString.consume_front("a")) {
if (ParseString.consumeInteger(10, Val))
return ParseRet::Error;
if (!isPowerOf2_64(Val))
return ParseRet::Error;
Alignment = Align(Val);
return ParseRet::OK;
}
return ParseRet::None;
}
[llvm][VectorUtils] Tweak VFShape for scalable vector functions. Summary: This patch makes sure that the field VFShape.VF is greater than zero when demangling the vector function name of scalable vector functions encoded in the "vector-function-abi-variant" attribute. This change is required to be able to provide instances of VFShape that can be used to query the VFDatabase for the vectorization passes, as such passes always require a positive value for the Vectorization Factor (VF) needed by the vectorization process. It is not possible to extract the value of VFShape.VF from the mangled name of scalable vector functions, because it is encoded as `x`. Therefore, the VFABI demangling function has been modified to extract such information from the IR declaration of the vector function, under the assumption that _all_ vectors in the signature of the vector function have the same number of lanes. Such assumption is valid because it is also assumed by the Vector Function ABI specifications supported by the demangling function (x86, AArch64, and LLVM internal one). The unit tests that demangle scalable names have been modified by adding the IR module that carries the declaration of the vector function name being demangled. In particular, the demangling function fails in the following cases: 1. When the declaration of the scalable vector function is not present in the module. 2. When the value of VFSHape.VF is not greater than 0. Reviewers: jdoerfert, sdesmalen, andwar Reviewed By: jdoerfert Subscribers: mgorny, kristof.beyls, hiraditya, llvm-commits Tags: #llvm Differential Revision: https://reviews.llvm.org/D73286
2020-01-22 23:34:27 +01:00
#ifndef NDEBUG
// Verify the assumtion that all vectors in the signature of a vector
// function have the same number of elements.
bool verifyAllVectorsHaveSameWidth(FunctionType *Signature) {
SmallVector<VectorType *, 2> VecTys;
if (auto *RetTy = dyn_cast<VectorType>(Signature->getReturnType()))
VecTys.push_back(RetTy);
for (auto *Ty : Signature->params())
if (auto *VTy = dyn_cast<VectorType>(Ty))
VecTys.push_back(VTy);
if (VecTys.size() <= 1)
return true;
assert(VecTys.size() > 1 && "Invalid number of elements.");
const ElementCount EC = VecTys[0]->getElementCount();
return llvm::all_of(llvm::drop_begin(VecTys), [&EC](VectorType *VTy) {
2021-01-15 05:30:32 +01:00
return (EC == VTy->getElementCount());
});
[llvm][VectorUtils] Tweak VFShape for scalable vector functions. Summary: This patch makes sure that the field VFShape.VF is greater than zero when demangling the vector function name of scalable vector functions encoded in the "vector-function-abi-variant" attribute. This change is required to be able to provide instances of VFShape that can be used to query the VFDatabase for the vectorization passes, as such passes always require a positive value for the Vectorization Factor (VF) needed by the vectorization process. It is not possible to extract the value of VFShape.VF from the mangled name of scalable vector functions, because it is encoded as `x`. Therefore, the VFABI demangling function has been modified to extract such information from the IR declaration of the vector function, under the assumption that _all_ vectors in the signature of the vector function have the same number of lanes. Such assumption is valid because it is also assumed by the Vector Function ABI specifications supported by the demangling function (x86, AArch64, and LLVM internal one). The unit tests that demangle scalable names have been modified by adding the IR module that carries the declaration of the vector function name being demangled. In particular, the demangling function fails in the following cases: 1. When the declaration of the scalable vector function is not present in the module. 2. When the value of VFSHape.VF is not greater than 0. Reviewers: jdoerfert, sdesmalen, andwar Reviewed By: jdoerfert Subscribers: mgorny, kristof.beyls, hiraditya, llvm-commits Tags: #llvm Differential Revision: https://reviews.llvm.org/D73286
2020-01-22 23:34:27 +01:00
}
#endif // NDEBUG
// Extract the VectorizationFactor from a given function signature,
// under the assumtion that all vectors have the same number of
// elements, i.e. same ElementCount.Min.
ElementCount getECFromSignature(FunctionType *Signature) {
assert(verifyAllVectorsHaveSameWidth(Signature) &&
"Invalid vector signature.");
if (auto *RetTy = dyn_cast<VectorType>(Signature->getReturnType()))
return RetTy->getElementCount();
for (auto *Ty : Signature->params())
if (auto *VTy = dyn_cast<VectorType>(Ty))
return VTy->getElementCount();
return ElementCount::getFixed(/*Min=*/1);
[llvm][VectorUtils] Tweak VFShape for scalable vector functions. Summary: This patch makes sure that the field VFShape.VF is greater than zero when demangling the vector function name of scalable vector functions encoded in the "vector-function-abi-variant" attribute. This change is required to be able to provide instances of VFShape that can be used to query the VFDatabase for the vectorization passes, as such passes always require a positive value for the Vectorization Factor (VF) needed by the vectorization process. It is not possible to extract the value of VFShape.VF from the mangled name of scalable vector functions, because it is encoded as `x`. Therefore, the VFABI demangling function has been modified to extract such information from the IR declaration of the vector function, under the assumption that _all_ vectors in the signature of the vector function have the same number of lanes. Such assumption is valid because it is also assumed by the Vector Function ABI specifications supported by the demangling function (x86, AArch64, and LLVM internal one). The unit tests that demangle scalable names have been modified by adding the IR module that carries the declaration of the vector function name being demangled. In particular, the demangling function fails in the following cases: 1. When the declaration of the scalable vector function is not present in the module. 2. When the value of VFSHape.VF is not greater than 0. Reviewers: jdoerfert, sdesmalen, andwar Reviewed By: jdoerfert Subscribers: mgorny, kristof.beyls, hiraditya, llvm-commits Tags: #llvm Differential Revision: https://reviews.llvm.org/D73286
2020-01-22 23:34:27 +01:00
}
} // namespace
// Format of the ABI name:
// _ZGV<isa><mask><vlen><parameters>_<scalarname>[(<redirection>)]
[llvm][VectorUtils] Tweak VFShape for scalable vector functions. Summary: This patch makes sure that the field VFShape.VF is greater than zero when demangling the vector function name of scalable vector functions encoded in the "vector-function-abi-variant" attribute. This change is required to be able to provide instances of VFShape that can be used to query the VFDatabase for the vectorization passes, as such passes always require a positive value for the Vectorization Factor (VF) needed by the vectorization process. It is not possible to extract the value of VFShape.VF from the mangled name of scalable vector functions, because it is encoded as `x`. Therefore, the VFABI demangling function has been modified to extract such information from the IR declaration of the vector function, under the assumption that _all_ vectors in the signature of the vector function have the same number of lanes. Such assumption is valid because it is also assumed by the Vector Function ABI specifications supported by the demangling function (x86, AArch64, and LLVM internal one). The unit tests that demangle scalable names have been modified by adding the IR module that carries the declaration of the vector function name being demangled. In particular, the demangling function fails in the following cases: 1. When the declaration of the scalable vector function is not present in the module. 2. When the value of VFSHape.VF is not greater than 0. Reviewers: jdoerfert, sdesmalen, andwar Reviewed By: jdoerfert Subscribers: mgorny, kristof.beyls, hiraditya, llvm-commits Tags: #llvm Differential Revision: https://reviews.llvm.org/D73286
2020-01-22 23:34:27 +01:00
Optional<VFInfo> VFABI::tryDemangleForVFABI(StringRef MangledName,
const Module &M) {
const StringRef OriginalName = MangledName;
// Assume there is no custom name <redirection>, and therefore the
// vector name consists of
// _ZGV<isa><mask><vlen><parameters>_<scalarname>.
StringRef VectorName = MangledName;
// Parse the fixed size part of the manled name
if (!MangledName.consume_front("_ZGV"))
return None;
// Extract ISA. An unknow ISA is also supported, so we accept all
// values.
VFISAKind ISA;
if (tryParseISA(MangledName, ISA) != ParseRet::OK)
return None;
// Extract <mask>.
bool IsMasked;
if (tryParseMask(MangledName, IsMasked) != ParseRet::OK)
return None;
// Parse the variable size, starting from <vlen>.
unsigned VF;
bool IsScalable;
if (tryParseVLEN(MangledName, VF, IsScalable) != ParseRet::OK)
return None;
// Parse the <parameters>.
ParseRet ParamFound;
SmallVector<VFParameter, 8> Parameters;
do {
const unsigned ParameterPos = Parameters.size();
VFParamKind PKind;
int StepOrPos;
ParamFound = tryParseParameter(MangledName, PKind, StepOrPos);
// Bail off if there is a parsing error in the parsing of the parameter.
if (ParamFound == ParseRet::Error)
return None;
if (ParamFound == ParseRet::OK) {
Align Alignment;
// Look for the alignment token "a <number>".
const ParseRet AlignFound = tryParseAlign(MangledName, Alignment);
// Bail off if there is a syntax error in the align token.
if (AlignFound == ParseRet::Error)
return None;
// Add the parameter.
Parameters.push_back({ParameterPos, PKind, StepOrPos, Alignment});
}
} while (ParamFound == ParseRet::OK);
// A valid MangledName must have at least one valid entry in the
// <parameters>.
if (Parameters.empty())
return None;
// Check for the <scalarname> and the optional <redirection>, which
// are separated from the prefix with "_"
if (!MangledName.consume_front("_"))
return None;
// The rest of the string must be in the format:
// <scalarname>[(<redirection>)]
const StringRef ScalarName =
MangledName.take_while([](char In) { return In != '('; });
if (ScalarName.empty())
return None;
// Reduce MangledName to [(<redirection>)].
MangledName = MangledName.ltrim(ScalarName);
// Find the optional custom name redirection.
if (MangledName.consume_front("(")) {
if (!MangledName.consume_back(")"))
return None;
// Update the vector variant with the one specified by the user.
VectorName = MangledName;
// If the vector name is missing, bail out.
if (VectorName.empty())
return None;
}
// LLVM internal mapping via the TargetLibraryInfo (TLI) must be
// redirected to an existing name.
if (ISA == VFISAKind::LLVM && VectorName == OriginalName)
return None;
// When <mask> is "M", we need to add a parameter that is used as
// global predicate for the function.
if (IsMasked) {
const unsigned Pos = Parameters.size();
Parameters.push_back({Pos, VFParamKind::GlobalPredicate});
}
// Asserts for parameters of type `VFParamKind::GlobalPredicate`, as
// prescribed by the Vector Function ABI specifications supported by
// this parser:
// 1. Uniqueness.
// 2. Must be the last in the parameter list.
const auto NGlobalPreds = std::count_if(
Parameters.begin(), Parameters.end(), [](const VFParameter PK) {
return PK.ParamKind == VFParamKind::GlobalPredicate;
});
assert(NGlobalPreds < 2 && "Cannot have more than one global predicate.");
if (NGlobalPreds)
assert(Parameters.back().ParamKind == VFParamKind::GlobalPredicate &&
"The global predicate must be the last parameter");
[llvm][VectorUtils] Tweak VFShape for scalable vector functions. Summary: This patch makes sure that the field VFShape.VF is greater than zero when demangling the vector function name of scalable vector functions encoded in the "vector-function-abi-variant" attribute. This change is required to be able to provide instances of VFShape that can be used to query the VFDatabase for the vectorization passes, as such passes always require a positive value for the Vectorization Factor (VF) needed by the vectorization process. It is not possible to extract the value of VFShape.VF from the mangled name of scalable vector functions, because it is encoded as `x`. Therefore, the VFABI demangling function has been modified to extract such information from the IR declaration of the vector function, under the assumption that _all_ vectors in the signature of the vector function have the same number of lanes. Such assumption is valid because it is also assumed by the Vector Function ABI specifications supported by the demangling function (x86, AArch64, and LLVM internal one). The unit tests that demangle scalable names have been modified by adding the IR module that carries the declaration of the vector function name being demangled. In particular, the demangling function fails in the following cases: 1. When the declaration of the scalable vector function is not present in the module. 2. When the value of VFSHape.VF is not greater than 0. Reviewers: jdoerfert, sdesmalen, andwar Reviewed By: jdoerfert Subscribers: mgorny, kristof.beyls, hiraditya, llvm-commits Tags: #llvm Differential Revision: https://reviews.llvm.org/D73286
2020-01-22 23:34:27 +01:00
// Adjust the VF for scalable signatures. The EC.Min is not encoded
// in the name of the function, but it is encoded in the IR
// signature of the function. We need to extract this information
// because it is needed by the loop vectorizer, which reasons in
// terms of VectorizationFactor or ElementCount. In particular, we
// need to make sure that the VF field of the VFShape class is never
// set to 0.
if (IsScalable) {
const Function *F = M.getFunction(VectorName);
// The declaration of the function must be present in the module
// to be able to retrieve its signature.
if (!F)
return None;
const ElementCount EC = getECFromSignature(F->getFunctionType());
VF = EC.getKnownMinValue();
[llvm][VectorUtils] Tweak VFShape for scalable vector functions. Summary: This patch makes sure that the field VFShape.VF is greater than zero when demangling the vector function name of scalable vector functions encoded in the "vector-function-abi-variant" attribute. This change is required to be able to provide instances of VFShape that can be used to query the VFDatabase for the vectorization passes, as such passes always require a positive value for the Vectorization Factor (VF) needed by the vectorization process. It is not possible to extract the value of VFShape.VF from the mangled name of scalable vector functions, because it is encoded as `x`. Therefore, the VFABI demangling function has been modified to extract such information from the IR declaration of the vector function, under the assumption that _all_ vectors in the signature of the vector function have the same number of lanes. Such assumption is valid because it is also assumed by the Vector Function ABI specifications supported by the demangling function (x86, AArch64, and LLVM internal one). The unit tests that demangle scalable names have been modified by adding the IR module that carries the declaration of the vector function name being demangled. In particular, the demangling function fails in the following cases: 1. When the declaration of the scalable vector function is not present in the module. 2. When the value of VFSHape.VF is not greater than 0. Reviewers: jdoerfert, sdesmalen, andwar Reviewed By: jdoerfert Subscribers: mgorny, kristof.beyls, hiraditya, llvm-commits Tags: #llvm Differential Revision: https://reviews.llvm.org/D73286
2020-01-22 23:34:27 +01:00
}
// Sanity checks.
// 1. We don't accept a zero lanes vectorization factor.
// 2. We don't accept the demangling if the vector function is not
// present in the module.
if (VF == 0)
return None;
if (!M.getFunction(VectorName))
return None;
const VFShape Shape({ElementCount::get(VF, IsScalable), Parameters});
return VFInfo({Shape, std::string(ScalarName), std::string(VectorName), ISA});
}
VFParamKind VFABI::getVFParamKindFromString(const StringRef Token) {
const VFParamKind ParamKind = StringSwitch<VFParamKind>(Token)
.Case("v", VFParamKind::Vector)
.Case("l", VFParamKind::OMP_Linear)
.Case("R", VFParamKind::OMP_LinearRef)
.Case("L", VFParamKind::OMP_LinearVal)
.Case("U", VFParamKind::OMP_LinearUVal)
.Case("ls", VFParamKind::OMP_LinearPos)
.Case("Ls", VFParamKind::OMP_LinearValPos)
.Case("Rs", VFParamKind::OMP_LinearRefPos)
.Case("Us", VFParamKind::OMP_LinearUValPos)
.Case("u", VFParamKind::OMP_Uniform)
.Default(VFParamKind::Unknown);
if (ParamKind != VFParamKind::Unknown)
return ParamKind;
// This function should never be invoked with an invalid input.
llvm_unreachable("This fuction should be invoken only on parameters"
" that have a textual representation in the mangled name"
" of the Vector Function ABI");
}