[clang] [AArch64] Simplify Clang's description of architecture extensions (PR #79311)
https://github.com/atrosinenko created https://github.com/llvm/llvm-project/pull/79311 Core LLVM has AArch64TargetParser.h header describing the mapping from Armv8.x and Armv9.x architecture extensions to the particular list of features that are mandatory for the extension. Clang partially reimplements this in AArch64TargetInfo::setArchFeatures() function. This patch simplifies setArchFeatures() by dropping obvious cases of duplication, though it seems not all the dependencies are listed in AArch64TargetParser.h yet. Specifically, this patch drops `HasX = true` statements for the following features: * v8.1-a: CRC, LSE, RDM * v8.3-a: RCPC, NeonMode (it is enabled by the base Armv8-a and indirectly by -target-feature +fcma) * v8.4-a: DOTPROD * v8.6-a: BF16, MATMUL In Clang's AArch64TargetInfo::handleTargetFeatures() function, the string literal "+jscvt" is changed to "+jsconv" and "+cccp" to "+ccpp", so the information is conveyed from TargetParser. Additionally, this patch makes __ARM_FEATURE_BTI and __ARM_FEATURE_JCVT conditional on the particular HasX flags instead of an architecture extension as a whole, as using TargetParser makes it possible to specify the requested architecture like "armv8.5+nojscvt". >From e0d278d9fcdd01f574bbf5fa1bcbbaf7875525de Mon Sep 17 00:00:00 2001 From: Anatoly Trosinenko Date: Tue, 23 Jan 2024 12:40:00 +0300 Subject: [PATCH] [AArch64] Simplify Clang's description of architecture extensions Core LLVM has AArch64TargetParser.h header describing the mapping from Armv8.x and Armv9.x architecture extensions to the particular list of features that are mandatory for the extension. Clang partially reimplements this in AArch64TargetInfo::setArchFeatures() function. This patch simplifies setArchFeatures() by dropping obvious cases of duplication, though it seems not all the dependencies are listed in AArch64TargetParser.h yet. Specifically, this patch drops `HasX = true` statements for the following features: * v8.1-a: CRC, LSE, RDM * v8.3-a: RCPC, NeonMode (it is enabled by the base Armv8-a and indirectly by -target-feature +fcma) * v8.4-a: DOTPROD * v8.6-a: BF16, MATMUL In Clang's AArch64TargetInfo::handleTargetFeatures() function, the string literal "+jscvt" is changed to "+jsconv" and "+cccp" to "+ccpp", so the information is conveyed from TargetParser. Additionally, this patch makes __ARM_FEATURE_BTI and __ARM_FEATURE_JCVT conditional on the particular HasX flags instead of an architecture extension as a whole, as using TargetParser makes it possible to specify the requested architecture like "armv8.5+nojscvt". --- clang/lib/Basic/Targets/AArch64.cpp | 48 + 1 file changed, 15 insertions(+), 33 deletions(-) diff --git a/clang/lib/Basic/Targets/AArch64.cpp b/clang/lib/Basic/Targets/AArch64.cpp index d47181bfca4fc86..0ac890551fadb24 100644 --- a/clang/lib/Basic/Targets/AArch64.cpp +++ b/clang/lib/Basic/Targets/AArch64.cpp @@ -55,24 +55,19 @@ static constexpr Builtin::Info BuiltinInfo[] = { }; void AArch64TargetInfo::setArchFeatures() { + // FIXME Could we drop this function altogether and migrate everything + // to AArch64TargetParser in LLVM core? if (*ArchInfo == llvm::AArch64::ARMV8R) { -HasDotProd = true; HasDIT = true; HasFlagM = true; -HasRCPC = true; FPU |= NeonMode; HasCCPP = true; -HasCRC = true; HasLSE = true; -HasRDM = true; } else if (ArchInfo->Version.getMajor() == 8) { if (ArchInfo->Version.getMinor() >= 7u) { HasWFxT = true; } -if (ArchInfo->Version.getMinor() >= 6u) { - HasBFloat16 = true; - HasMatMul = true; -} +// No special cases for Armv8.6-a if (ArchInfo->Version.getMinor() >= 5u) { HasAlternativeNZCV = true; HasFRInt3264 = true; @@ -82,30 +77,19 @@ void AArch64TargetInfo::setArchFeatures() { HasBTI = true; } if (ArchInfo->Version.getMinor() >= 4u) { - HasDotProd = true; HasDIT = true; HasFlagM = true; } -if (ArchInfo->Version.getMinor() >= 3u) { - HasRCPC = true; - FPU |= NeonMode; -} +// No special cases for Armv8.3-a if (ArchInfo->Version.getMinor() >= 2u) { HasCCPP = true; } -if (ArchInfo->Version.getMinor() >= 1u) { - HasCRC = true; - HasLSE = true; - HasRDM = true; -} +// No special cases for Armv8.1-a } else if (ArchInfo->Version.getMajor() == 9) { if (ArchInfo->Version.getMinor() >= 2u) { HasWFxT = true; } -if (ArchInfo->Version.getMinor() >= 1u) { - HasBFloat16 = true; - HasMatMul = true; -} +// No special cases for Armv9.1-a FPU |= SveMode; HasSVE2 = true; HasFullFP16 = true; @@ -115,15 +99,9 @@ void AArch64TargetInfo::setArchFeatures() { HasSB = true; HasPredRes = true; HasBTI = true; -HasDotProd = true; HasDIT = true; HasFlagM = true; -HasRCPC = true; -FPU |= NeonMode; HasCC
[clang] [AArch64] Simplify Clang's description of architecture extensions (PR #79311)
https://github.com/atrosinenko updated https://github.com/llvm/llvm-project/pull/79311 >From e0d278d9fcdd01f574bbf5fa1bcbbaf7875525de Mon Sep 17 00:00:00 2001 From: Anatoly Trosinenko Date: Tue, 23 Jan 2024 12:40:00 +0300 Subject: [PATCH 1/2] [AArch64] Simplify Clang's description of architecture extensions Core LLVM has AArch64TargetParser.h header describing the mapping from Armv8.x and Armv9.x architecture extensions to the particular list of features that are mandatory for the extension. Clang partially reimplements this in AArch64TargetInfo::setArchFeatures() function. This patch simplifies setArchFeatures() by dropping obvious cases of duplication, though it seems not all the dependencies are listed in AArch64TargetParser.h yet. Specifically, this patch drops `HasX = true` statements for the following features: * v8.1-a: CRC, LSE, RDM * v8.3-a: RCPC, NeonMode (it is enabled by the base Armv8-a and indirectly by -target-feature +fcma) * v8.4-a: DOTPROD * v8.6-a: BF16, MATMUL In Clang's AArch64TargetInfo::handleTargetFeatures() function, the string literal "+jscvt" is changed to "+jsconv" and "+cccp" to "+ccpp", so the information is conveyed from TargetParser. Additionally, this patch makes __ARM_FEATURE_BTI and __ARM_FEATURE_JCVT conditional on the particular HasX flags instead of an architecture extension as a whole, as using TargetParser makes it possible to specify the requested architecture like "armv8.5+nojscvt". --- clang/lib/Basic/Targets/AArch64.cpp | 48 + 1 file changed, 15 insertions(+), 33 deletions(-) diff --git a/clang/lib/Basic/Targets/AArch64.cpp b/clang/lib/Basic/Targets/AArch64.cpp index d47181bfca4fc86..0ac890551fadb24 100644 --- a/clang/lib/Basic/Targets/AArch64.cpp +++ b/clang/lib/Basic/Targets/AArch64.cpp @@ -55,24 +55,19 @@ static constexpr Builtin::Info BuiltinInfo[] = { }; void AArch64TargetInfo::setArchFeatures() { + // FIXME Could we drop this function altogether and migrate everything + // to AArch64TargetParser in LLVM core? if (*ArchInfo == llvm::AArch64::ARMV8R) { -HasDotProd = true; HasDIT = true; HasFlagM = true; -HasRCPC = true; FPU |= NeonMode; HasCCPP = true; -HasCRC = true; HasLSE = true; -HasRDM = true; } else if (ArchInfo->Version.getMajor() == 8) { if (ArchInfo->Version.getMinor() >= 7u) { HasWFxT = true; } -if (ArchInfo->Version.getMinor() >= 6u) { - HasBFloat16 = true; - HasMatMul = true; -} +// No special cases for Armv8.6-a if (ArchInfo->Version.getMinor() >= 5u) { HasAlternativeNZCV = true; HasFRInt3264 = true; @@ -82,30 +77,19 @@ void AArch64TargetInfo::setArchFeatures() { HasBTI = true; } if (ArchInfo->Version.getMinor() >= 4u) { - HasDotProd = true; HasDIT = true; HasFlagM = true; } -if (ArchInfo->Version.getMinor() >= 3u) { - HasRCPC = true; - FPU |= NeonMode; -} +// No special cases for Armv8.3-a if (ArchInfo->Version.getMinor() >= 2u) { HasCCPP = true; } -if (ArchInfo->Version.getMinor() >= 1u) { - HasCRC = true; - HasLSE = true; - HasRDM = true; -} +// No special cases for Armv8.1-a } else if (ArchInfo->Version.getMajor() == 9) { if (ArchInfo->Version.getMinor() >= 2u) { HasWFxT = true; } -if (ArchInfo->Version.getMinor() >= 1u) { - HasBFloat16 = true; - HasMatMul = true; -} +// No special cases for Armv9.1-a FPU |= SveMode; HasSVE2 = true; HasFullFP16 = true; @@ -115,15 +99,9 @@ void AArch64TargetInfo::setArchFeatures() { HasSB = true; HasPredRes = true; HasBTI = true; -HasDotProd = true; HasDIT = true; HasFlagM = true; -HasRCPC = true; -FPU |= NeonMode; HasCCPP = true; -HasCRC = true; -HasLSE = true; -HasRDM = true; } } @@ -257,7 +235,6 @@ void AArch64TargetInfo::getTargetDefinesARMV82A(const LangOptions &Opts, void AArch64TargetInfo::getTargetDefinesARMV83A(const LangOptions &Opts, MacroBuilder &Builder) const { Builder.defineMacro("__ARM_FEATURE_COMPLEX", "1"); - Builder.defineMacro("__ARM_FEATURE_JCVT", "1"); Builder.defineMacro("__ARM_FEATURE_PAUTH", "1"); // Also include the Armv8.2 defines getTargetDefinesARMV82A(Opts, Builder); @@ -272,7 +249,6 @@ void AArch64TargetInfo::getTargetDefinesARMV84A(const LangOptions &Opts, void AArch64TargetInfo::getTargetDefinesARMV85A(const LangOptions &Opts, MacroBuilder &Builder) const { Builder.defineMacro("__ARM_FEATURE_FRINT", "1"); - Builder.defineMacro("__ARM_FEATURE_BTI", "1"); // Also include the Armv8.4 defines getTargetDefinesARMV84A(Opts, Builder); } @@ -472,7 +448,10 @@ void AArch64TargetInfo::getTargetDefines(const LangOptions &Opts, if ((FPU & NeonMode) && HasFullFP16) Builder.defineMa
[clang] [AArch64] Simplify Clang's description of architecture extensions (PR #79311)
llvmbot wrote: @llvm/pr-subscribers-clang Author: Anatoly Trosinenko (atrosinenko) Changes Core LLVM has AArch64TargetParser.h header describing the mapping from Armv8.x and Armv9.x architecture extensions to the particular list of features that are mandatory for the extension. Clang partially reimplements this in AArch64TargetInfo::setArchFeatures() function. This patch simplifies setArchFeatures() by dropping obvious cases of duplication, though it seems not all the dependencies are listed in AArch64TargetParser.h yet. Specifically, this patch drops `HasX = true` statements for the following features: * v8.1-a: CRC, LSE, RDM * v8.3-a: RCPC, NeonMode (it is enabled by the base Armv8-a and indirectly by -target-feature +fcma) * v8.4-a: DOTPROD * v8.6-a: BF16, MATMUL In Clang's AArch64TargetInfo::handleTargetFeatures() function, the string literal "+jscvt" is changed to "+jsconv" and "+cccp" to "+ccpp", so the information is conveyed from TargetParser. Additionally, this patch makes __ARM_FEATURE_BTI and __ARM_FEATURE_JCVT conditional on the particular HasX flags instead of an architecture extension as a whole, as using TargetParser makes it possible to specify the requested architecture like "armv8.5+nojscvt". --- Full diff: https://github.com/llvm/llvm-project/pull/79311.diff 1 Files Affected: - (modified) clang/lib/Basic/Targets/AArch64.cpp (+15-34) ``diff diff --git a/clang/lib/Basic/Targets/AArch64.cpp b/clang/lib/Basic/Targets/AArch64.cpp index d47181bfca4fc8..76bb60db41099e 100644 --- a/clang/lib/Basic/Targets/AArch64.cpp +++ b/clang/lib/Basic/Targets/AArch64.cpp @@ -55,24 +55,18 @@ static constexpr Builtin::Info BuiltinInfo[] = { }; void AArch64TargetInfo::setArchFeatures() { + // FIXME Could we drop this function altogether and migrate everything + // to AArch64TargetParser in LLVM core? if (*ArchInfo == llvm::AArch64::ARMV8R) { -HasDotProd = true; HasDIT = true; HasFlagM = true; -HasRCPC = true; -FPU |= NeonMode; HasCCPP = true; -HasCRC = true; HasLSE = true; -HasRDM = true; } else if (ArchInfo->Version.getMajor() == 8) { if (ArchInfo->Version.getMinor() >= 7u) { HasWFxT = true; } -if (ArchInfo->Version.getMinor() >= 6u) { - HasBFloat16 = true; - HasMatMul = true; -} +// No special cases for Armv8.6-a if (ArchInfo->Version.getMinor() >= 5u) { HasAlternativeNZCV = true; HasFRInt3264 = true; @@ -82,30 +76,19 @@ void AArch64TargetInfo::setArchFeatures() { HasBTI = true; } if (ArchInfo->Version.getMinor() >= 4u) { - HasDotProd = true; HasDIT = true; HasFlagM = true; } -if (ArchInfo->Version.getMinor() >= 3u) { - HasRCPC = true; - FPU |= NeonMode; -} +// No special cases for Armv8.3-a if (ArchInfo->Version.getMinor() >= 2u) { HasCCPP = true; } -if (ArchInfo->Version.getMinor() >= 1u) { - HasCRC = true; - HasLSE = true; - HasRDM = true; -} +// No special cases for Armv8.1-a } else if (ArchInfo->Version.getMajor() == 9) { if (ArchInfo->Version.getMinor() >= 2u) { HasWFxT = true; } -if (ArchInfo->Version.getMinor() >= 1u) { - HasBFloat16 = true; - HasMatMul = true; -} +// No special cases for Armv9.1-a FPU |= SveMode; HasSVE2 = true; HasFullFP16 = true; @@ -115,15 +98,9 @@ void AArch64TargetInfo::setArchFeatures() { HasSB = true; HasPredRes = true; HasBTI = true; -HasDotProd = true; HasDIT = true; HasFlagM = true; -HasRCPC = true; -FPU |= NeonMode; HasCCPP = true; -HasCRC = true; -HasLSE = true; -HasRDM = true; } } @@ -257,7 +234,6 @@ void AArch64TargetInfo::getTargetDefinesARMV82A(const LangOptions &Opts, void AArch64TargetInfo::getTargetDefinesARMV83A(const LangOptions &Opts, MacroBuilder &Builder) const { Builder.defineMacro("__ARM_FEATURE_COMPLEX", "1"); - Builder.defineMacro("__ARM_FEATURE_JCVT", "1"); Builder.defineMacro("__ARM_FEATURE_PAUTH", "1"); // Also include the Armv8.2 defines getTargetDefinesARMV82A(Opts, Builder); @@ -272,7 +248,6 @@ void AArch64TargetInfo::getTargetDefinesARMV84A(const LangOptions &Opts, void AArch64TargetInfo::getTargetDefinesARMV85A(const LangOptions &Opts, MacroBuilder &Builder) const { Builder.defineMacro("__ARM_FEATURE_FRINT", "1"); - Builder.defineMacro("__ARM_FEATURE_BTI", "1"); // Also include the Armv8.4 defines getTargetDefinesARMV84A(Opts, Builder); } @@ -472,7 +447,10 @@ void AArch64TargetInfo::getTargetDefines(const LangOptions &Opts, if ((FPU & NeonMode) && HasFullFP16) Builder.defineMacro("__ARM_FEATURE_FP16_VECTOR_ARITHMETIC", "1"); if (HasFullFP16) - Builder.defineMacro("__ARM_FEATURE_FP16_SCALAR_ARITHMETIC", "1"); +Builder.defineMacro("__A
[clang] [AArch64] Simplify Clang's description of architecture extensions (PR #79311)
@@ -741,7 +722,7 @@ bool AArch64TargetInfo::handleTargetFeatures(std::vector &Features, if (Feature == "+neon" || Feature == "+fp-armv8") FPU |= NeonMode; -if (Feature == "+jscvt") { +if (Feature == "+jsconv") { ostannard wrote: This looks like bug fix which would make a functional difference, does it need a test? https://github.com/llvm/llvm-project/pull/79311 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [AArch64] Simplify Clang's description of architecture extensions (PR #79311)
@@ -860,7 +841,7 @@ bool AArch64TargetInfo::handleTargetFeatures(std::vector &Features, } if (Feature == "+dit") HasDIT = true; -if (Feature == "+cccp") +if (Feature == "+ccpp") ostannard wrote: Again, does this need a test? https://github.com/llvm/llvm-project/pull/79311 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [AArch64] Simplify Clang's description of architecture extensions (PR #79311)
https://github.com/atrosinenko updated https://github.com/llvm/llvm-project/pull/79311 >From e0d278d9fcdd01f574bbf5fa1bcbbaf7875525de Mon Sep 17 00:00:00 2001 From: Anatoly Trosinenko Date: Tue, 23 Jan 2024 12:40:00 +0300 Subject: [PATCH 1/3] [AArch64] Simplify Clang's description of architecture extensions Core LLVM has AArch64TargetParser.h header describing the mapping from Armv8.x and Armv9.x architecture extensions to the particular list of features that are mandatory for the extension. Clang partially reimplements this in AArch64TargetInfo::setArchFeatures() function. This patch simplifies setArchFeatures() by dropping obvious cases of duplication, though it seems not all the dependencies are listed in AArch64TargetParser.h yet. Specifically, this patch drops `HasX = true` statements for the following features: * v8.1-a: CRC, LSE, RDM * v8.3-a: RCPC, NeonMode (it is enabled by the base Armv8-a and indirectly by -target-feature +fcma) * v8.4-a: DOTPROD * v8.6-a: BF16, MATMUL In Clang's AArch64TargetInfo::handleTargetFeatures() function, the string literal "+jscvt" is changed to "+jsconv" and "+cccp" to "+ccpp", so the information is conveyed from TargetParser. Additionally, this patch makes __ARM_FEATURE_BTI and __ARM_FEATURE_JCVT conditional on the particular HasX flags instead of an architecture extension as a whole, as using TargetParser makes it possible to specify the requested architecture like "armv8.5+nojscvt". --- clang/lib/Basic/Targets/AArch64.cpp | 48 + 1 file changed, 15 insertions(+), 33 deletions(-) diff --git a/clang/lib/Basic/Targets/AArch64.cpp b/clang/lib/Basic/Targets/AArch64.cpp index d47181bfca4fc86..0ac890551fadb24 100644 --- a/clang/lib/Basic/Targets/AArch64.cpp +++ b/clang/lib/Basic/Targets/AArch64.cpp @@ -55,24 +55,19 @@ static constexpr Builtin::Info BuiltinInfo[] = { }; void AArch64TargetInfo::setArchFeatures() { + // FIXME Could we drop this function altogether and migrate everything + // to AArch64TargetParser in LLVM core? if (*ArchInfo == llvm::AArch64::ARMV8R) { -HasDotProd = true; HasDIT = true; HasFlagM = true; -HasRCPC = true; FPU |= NeonMode; HasCCPP = true; -HasCRC = true; HasLSE = true; -HasRDM = true; } else if (ArchInfo->Version.getMajor() == 8) { if (ArchInfo->Version.getMinor() >= 7u) { HasWFxT = true; } -if (ArchInfo->Version.getMinor() >= 6u) { - HasBFloat16 = true; - HasMatMul = true; -} +// No special cases for Armv8.6-a if (ArchInfo->Version.getMinor() >= 5u) { HasAlternativeNZCV = true; HasFRInt3264 = true; @@ -82,30 +77,19 @@ void AArch64TargetInfo::setArchFeatures() { HasBTI = true; } if (ArchInfo->Version.getMinor() >= 4u) { - HasDotProd = true; HasDIT = true; HasFlagM = true; } -if (ArchInfo->Version.getMinor() >= 3u) { - HasRCPC = true; - FPU |= NeonMode; -} +// No special cases for Armv8.3-a if (ArchInfo->Version.getMinor() >= 2u) { HasCCPP = true; } -if (ArchInfo->Version.getMinor() >= 1u) { - HasCRC = true; - HasLSE = true; - HasRDM = true; -} +// No special cases for Armv8.1-a } else if (ArchInfo->Version.getMajor() == 9) { if (ArchInfo->Version.getMinor() >= 2u) { HasWFxT = true; } -if (ArchInfo->Version.getMinor() >= 1u) { - HasBFloat16 = true; - HasMatMul = true; -} +// No special cases for Armv9.1-a FPU |= SveMode; HasSVE2 = true; HasFullFP16 = true; @@ -115,15 +99,9 @@ void AArch64TargetInfo::setArchFeatures() { HasSB = true; HasPredRes = true; HasBTI = true; -HasDotProd = true; HasDIT = true; HasFlagM = true; -HasRCPC = true; -FPU |= NeonMode; HasCCPP = true; -HasCRC = true; -HasLSE = true; -HasRDM = true; } } @@ -257,7 +235,6 @@ void AArch64TargetInfo::getTargetDefinesARMV82A(const LangOptions &Opts, void AArch64TargetInfo::getTargetDefinesARMV83A(const LangOptions &Opts, MacroBuilder &Builder) const { Builder.defineMacro("__ARM_FEATURE_COMPLEX", "1"); - Builder.defineMacro("__ARM_FEATURE_JCVT", "1"); Builder.defineMacro("__ARM_FEATURE_PAUTH", "1"); // Also include the Armv8.2 defines getTargetDefinesARMV82A(Opts, Builder); @@ -272,7 +249,6 @@ void AArch64TargetInfo::getTargetDefinesARMV84A(const LangOptions &Opts, void AArch64TargetInfo::getTargetDefinesARMV85A(const LangOptions &Opts, MacroBuilder &Builder) const { Builder.defineMacro("__ARM_FEATURE_FRINT", "1"); - Builder.defineMacro("__ARM_FEATURE_BTI", "1"); // Also include the Armv8.4 defines getTargetDefinesARMV84A(Opts, Builder); } @@ -472,7 +448,10 @@ void AArch64TargetInfo::getTargetDefines(const LangOptions &Opts, if ((FPU & NeonMode) && HasFullFP16) Builder.defineMa
[clang] [AArch64] Simplify Clang's description of architecture extensions (PR #79311)
atrosinenko wrote: I tried adding a few more lines to `clang/test/Preprocessor/aarch64-target-features.c` and turned out, clang accepts `-march=armv8-a+jscvt` but not `-march=armv8-a+dpb`. Brief debugging suggests this is because in the ```cpp {"dpb", AArch64::AEK_NONE, {}, {}, FEAT_DPB, "+ccpp", 190}, ``` line, `Feature` and `NegFeature` fields are empty. I wonder if there is an overview of all the "feature-related" AArch64 stuff anywhere. As far as I understand, there are several kinds of identifiers: * feature names defined in the chapter "A2: Armv8-A Architecture Extensions" of the Arm Architecture Reference Manual (such as `FEAT_JSCVT` mentioned in "A2.5: The Armv8.3 architecture extension") * Function Multi-Versioning CPU features defined in `enum CPUFeatures` (such as `FEAT_JSCVT`) * `AEK_*` constants defined in `enum ArchExtKind` (such as `AEK_JSCVT`) * `Name`s defined in the `Extensions` array in AArch64TargetParser.h * `Feature` / `NegFeature` defined in the same array * `DependentFeatures` defined in the same array I guess, each kind of identifiers belongs to one of logical "namespaces". Here is what I see: * the names of FMV features looks identical to feature names defined in Arm ARM, but the _sets_ of identifiers are different - the former is a strict subset of the latter at best. Obviously, there is a number of features hardly related to compiler development (such as various properties of hardware debug interfaces) and some other are not related to runtime function selection by FMV machinery (or at least the rules are not yet written in ACLE for them). It is just a bit misleading that Arm ARM defines FEAT_PAUTH and this feature is mentioned as AEK_PAUTH, "+pauth" literal, etc. but not as FEAT_PAUTH in the enumeration with constants named `FEAT_*` :) * after `AEK_*` constants were recently made not restricted by the bitsize of `uint64_t` anymore, these look like the best mirror of `FEAT_*` identifiers defined in Arm ARM (and there is an explicit [naming guideline](https://github.com/llvm/llvm-project/blob/7f409cd82b322038f08a984a07377758e76b0e4c/llvm/include/llvm/TargetParser/AArch64TargetParser.h#L106)) * `Name` fields are used when parsing `-march=...+feature` arguments of Clang. These are kind of user-visible interface * `Feature`/`NegFeature` and `DependentFeatures` are used when parsing `-target-feature`s passed to `clang -cc1` and are likely to contain string identifiers from the same namespace, according to seemingly the [only in-tree user](https://github.com/llvm/llvm-project/blob/7fdb932c4e78706cec3468a1f149e5a54a865d36/clang/lib/Basic/Targets/AArch64.cpp#L1013) of `DependentFeatures` contents, but duplication like `{"jscvt", AArch64::AEK_JSCVT, "+jsconv", "-jsconv", FEAT_JSCVT, "+fp-armv8,+neon,+jsconv", 210}` (E.Feature additionally listed in E.DependentFeatures) looks redundant https://github.com/llvm/llvm-project/pull/79311 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [AArch64] Simplify Clang's description of architecture extensions (PR #79311)
@@ -741,7 +722,7 @@ bool AArch64TargetInfo::handleTargetFeatures(std::vector &Features, if (Feature == "+neon" || Feature == "+fp-armv8") FPU |= NeonMode; -if (Feature == "+jscvt") { +if (Feature == "+jsconv") { atrosinenko wrote: Agree, added a test case to `aarch64-target-features.c` https://github.com/llvm/llvm-project/pull/79311 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [AArch64] Simplify Clang's description of architecture extensions (PR #79311)
@@ -860,7 +841,7 @@ bool AArch64TargetInfo::handleTargetFeatures(std::vector &Features, } if (Feature == "+dit") HasDIT = true; -if (Feature == "+cccp") +if (Feature == "+ccpp") atrosinenko wrote: Agree, but not yet sure how it is better to test this https://github.com/llvm/llvm-project/pull/79311 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [AArch64] Simplify Clang's description of architecture extensions (PR #79311)
atrosinenko wrote: Meanwhile, which is current policy on enabling/disabling mandatory features via command line options? https://github.com/llvm/llvm-project/pull/79311 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits