https://github.com/Stylie777 updated https://github.com/llvm/llvm-project/pull/130623
>From 7c0b94752285f2734d79e9e8d38aa20b3f7e8e61 Mon Sep 17 00:00:00 2001 From: Jack Styles <jack.sty...@arm.com> Date: Fri, 7 Mar 2025 15:51:34 +0000 Subject: [PATCH 1/4] [NFC][ARM] Split SIMD identifier away from MVE Previously, the use of MVE or MVE.FP would be defined by using the `ARM::AEK_SIMD` identifier. SIMD relates to the Cortex-A and Cortex-R extension that enables NEON instructions, which is called MVE for Cortex-M. To enable the linking of `+simd` and `+nosimd` to `+neon` and `-neon` when using clang, MVE and MVE.FP can now be defined using a unique enum identifier, rather than using the existing identifier for SIMD. This was originally planned to be merged as part of #130296 but the changes made `+nosimd` an invalid argument, which, while not having any functionality, was allowed in previous versions of LLVM. To avoid regressions being introduced, this has been combined with the fix for `+nosimd` on AArch32. --- llvm/include/llvm/TargetParser/ARMTargetParser.def | 10 +++++----- llvm/include/llvm/TargetParser/ARMTargetParser.h | 1 + llvm/lib/Target/ARM/AsmParser/ARMAsmParser.cpp | 2 +- llvm/lib/Target/ARM/MCTargetDesc/ARMTargetStreamer.cpp | 2 +- llvm/unittests/TargetParser/TargetParserTest.cpp | 6 +++--- 5 files changed, 11 insertions(+), 10 deletions(-) diff --git a/llvm/include/llvm/TargetParser/ARMTargetParser.def b/llvm/include/llvm/TargetParser/ARMTargetParser.def index 6b96c3e83c8c4..e515ab665d117 100644 --- a/llvm/include/llvm/TargetParser/ARMTargetParser.def +++ b/llvm/include/llvm/TargetParser/ARMTargetParser.def @@ -224,8 +224,8 @@ ARM_ARCH_EXT_NAME("dotprod", ARM::AEK_DOTPROD, "+dotprod", "-dotprod") ARM_ARCH_EXT_NAME("dsp", ARM::AEK_DSP, "+dsp", "-dsp") ARM_ARCH_EXT_NAME("fp", ARM::AEK_FP, {}, {}) ARM_ARCH_EXT_NAME("fp.dp", ARM::AEK_FP_DP, {}, {}) -ARM_ARCH_EXT_NAME("mve", (ARM::AEK_DSP | ARM::AEK_SIMD), "+mve", "-mve") -ARM_ARCH_EXT_NAME("mve.fp", (ARM::AEK_DSP | ARM::AEK_SIMD | ARM::AEK_FP), +ARM_ARCH_EXT_NAME("mve", (ARM::AEK_DSP | ARM::AEK_MVE), "+mve", "-mve") +ARM_ARCH_EXT_NAME("mve.fp", (ARM::AEK_DSP | ARM::AEK_MVE | ARM::AEK_FP), "+mve.fp", "-mve.fp") ARM_ARCH_EXT_NAME("idiv", (ARM::AEK_HWDIVARM | ARM::AEK_HWDIVTHUMB), {}, {}) ARM_ARCH_EXT_NAME("mp", ARM::AEK_MP, {}, {}) @@ -345,12 +345,12 @@ ARM_CPU_NAME("cortex-m33", ARMV8MMainline, FK_FPV5_SP_D16, false, ARM::AEK_DSP) ARM_CPU_NAME("star-mc1", ARMV8MMainline, FK_FPV5_SP_D16, false, ARM::AEK_DSP) ARM_CPU_NAME("cortex-m35p", ARMV8MMainline, FK_FPV5_SP_D16, false, ARM::AEK_DSP) ARM_CPU_NAME("cortex-m55", ARMV8_1MMainline, FK_FP_ARMV8_FULLFP16_D16, false, - (ARM::AEK_DSP | ARM::AEK_SIMD | ARM::AEK_FP | ARM::AEK_FP16)) + (ARM::AEK_DSP | ARM::AEK_MVE | ARM::AEK_FP | ARM::AEK_FP16)) ARM_CPU_NAME("cortex-m85", ARMV8_1MMainline, FK_FP_ARMV8_FULLFP16_D16, false, - (ARM::AEK_DSP | ARM::AEK_SIMD | ARM::AEK_FP | ARM::AEK_FP16 | + (ARM::AEK_DSP | ARM::AEK_MVE | ARM::AEK_FP | ARM::AEK_FP16 | ARM::AEK_RAS | ARM::AEK_PACBTI)) ARM_CPU_NAME("cortex-m52", ARMV8_1MMainline, FK_FP_ARMV8_FULLFP16_D16, false, - (ARM::AEK_DSP | ARM::AEK_SIMD | ARM::AEK_FP | ARM::AEK_FP16 | + (ARM::AEK_DSP | ARM::AEK_MVE | ARM::AEK_FP | ARM::AEK_FP16 | ARM::AEK_RAS | ARM::AEK_PACBTI)) ARM_CPU_NAME("cortex-a32", ARMV8A, FK_CRYPTO_NEON_FP_ARMV8, false, ARM::AEK_CRC) ARM_CPU_NAME("cortex-a35", ARMV8A, FK_CRYPTO_NEON_FP_ARMV8, false, ARM::AEK_CRC) diff --git a/llvm/include/llvm/TargetParser/ARMTargetParser.h b/llvm/include/llvm/TargetParser/ARMTargetParser.h index 5dbcfd3d2d693..b2403f42f1b79 100644 --- a/llvm/include/llvm/TargetParser/ARMTargetParser.h +++ b/llvm/include/llvm/TargetParser/ARMTargetParser.h @@ -61,6 +61,7 @@ enum ArchExtKind : uint64_t { AEK_CDECP6 = 1 << 28, AEK_CDECP7 = 1 << 29, AEK_PACBTI = 1 << 30, + AEK_MVE = 1ULL << 31, // Unsupported extensions. AEK_OS = 1ULL << 59, AEK_IWMMXT = 1ULL << 60, diff --git a/llvm/lib/Target/ARM/AsmParser/ARMAsmParser.cpp b/llvm/lib/Target/ARM/AsmParser/ARMAsmParser.cpp index d6a586e1f247b..79b54e6b04330 100644 --- a/llvm/lib/Target/ARM/AsmParser/ARMAsmParser.cpp +++ b/llvm/lib/Target/ARM/AsmParser/ARMAsmParser.cpp @@ -12971,7 +12971,7 @@ bool ARMAsmParser::enableArchExtFeature(StringRef Name, SMLoc &ExtLoc) { {ARM::AEK_CRYPTO, {Feature_HasV8Bit}, {ARM::FeatureCrypto, ARM::FeatureNEON, ARM::FeatureFPARMv8}}, - {(ARM::AEK_DSP | ARM::AEK_SIMD | ARM::AEK_FP), + {(ARM::AEK_DSP | ARM::AEK_MVE | ARM::AEK_FP), {Feature_HasV8_1MMainlineBit}, {ARM::HasMVEFloatOps}}, {ARM::AEK_FP, diff --git a/llvm/lib/Target/ARM/MCTargetDesc/ARMTargetStreamer.cpp b/llvm/lib/Target/ARM/MCTargetDesc/ARMTargetStreamer.cpp index b0fa03a35ec04..632dbebf58f04 100644 --- a/llvm/lib/Target/ARM/MCTargetDesc/ARMTargetStreamer.cpp +++ b/llvm/lib/Target/ARM/MCTargetDesc/ARMTargetStreamer.cpp @@ -248,7 +248,7 @@ void ARMTargetStreamer::emitTargetAttributes(const MCSubtargetInfo &STI) { emitFPU(STI.hasFeature(ARM::FeatureFP64) ? ARM::FK_FPV5_D16 : ARM::FK_FPV5_SP_D16); if (STI.hasFeature(ARM::HasMVEFloatOps)) - emitArchExtension(ARM::AEK_SIMD | ARM::AEK_DSP | ARM::AEK_FP); + emitArchExtension(ARM::AEK_MVE | ARM::AEK_DSP | ARM::AEK_FP); } } else if (STI.hasFeature(ARM::FeatureVFP4_D16_SP)) emitFPU(STI.hasFeature(ARM::FeatureD32) diff --git a/llvm/unittests/TargetParser/TargetParserTest.cpp b/llvm/unittests/TargetParser/TargetParserTest.cpp index 5d771a1a153f7..574d69f847b4d 100644 --- a/llvm/unittests/TargetParser/TargetParserTest.cpp +++ b/llvm/unittests/TargetParser/TargetParserTest.cpp @@ -507,17 +507,17 @@ INSTANTIATE_TEST_SUITE_P( "8-M.Mainline"), ARMCPUTestParams<uint64_t>( "cortex-m55", "armv8.1-m.main", "fp-armv8-fullfp16-d16", - ARM::AEK_HWDIVTHUMB | ARM::AEK_DSP | ARM::AEK_SIMD | ARM::AEK_FP | + ARM::AEK_HWDIVTHUMB | ARM::AEK_DSP | ARM::AEK_MVE | ARM::AEK_FP | ARM::AEK_RAS | ARM::AEK_LOB | ARM::AEK_FP16, "8.1-M.Mainline"), ARMCPUTestParams<uint64_t>( "cortex-m85", "armv8.1-m.main", "fp-armv8-fullfp16-d16", - ARM::AEK_HWDIVTHUMB | ARM::AEK_DSP | ARM::AEK_SIMD | ARM::AEK_FP | + ARM::AEK_HWDIVTHUMB | ARM::AEK_DSP | ARM::AEK_MVE | ARM::AEK_FP | ARM::AEK_RAS | ARM::AEK_LOB | ARM::AEK_FP16 | ARM::AEK_PACBTI, "8.1-M.Mainline"), ARMCPUTestParams<uint64_t>( "cortex-m52", "armv8.1-m.main", "fp-armv8-fullfp16-d16", - ARM::AEK_HWDIVTHUMB | ARM::AEK_DSP | ARM::AEK_SIMD | ARM::AEK_FP | + ARM::AEK_HWDIVTHUMB | ARM::AEK_DSP | ARM::AEK_MVE | ARM::AEK_FP | ARM::AEK_RAS | ARM::AEK_LOB | ARM::AEK_FP16 | ARM::AEK_PACBTI, "8.1-M.Mainline"), ARMCPUTestParams<uint64_t>("iwmmxt", "iwmmxt", "none", ARM::AEK_NONE, >From d5fb6b32069a7c78540035d743fa4cf30ba40619 Mon Sep 17 00:00:00 2001 From: Jack Styles <jack.sty...@arm.com> Date: Fri, 7 Mar 2025 15:49:00 +0000 Subject: [PATCH 2/4] [Clang][ARM] Enable the use of `+simd` and `+nosimd` for AArch32 Targets `+simd` and `+nosimd` are used to enable or disable NEON Instructions when compiling for AArch32 Targets. However, up until now, using these has not been possible. To enable this, these options are mapped to the relevant LLVM backend option (`+neon` and `-neon`) so it can be both enabled and disabled successfully by the user. Tests have been added to ensure this behaviour is maintained in the future, along with updates to existing tests as behaviour has now changed relating to the use of `+simd` and `+nosimd`. As `simd` has been mapped within the ARMTargetParser.def, support for this extension is also added for the `--print-support-extensions` command when the target is AArch32. This will print the `simd` option, along with the description that relates to the Neon feature. This previously was not possible as `simd` did not have a related Feature or Negative Feature. --- .../test/Driver/print-supported-extensions-arm.c | 1 + clang/test/Preprocessor/arm-target-features.c | 8 ++++++++ llvm/docs/ReleaseNotes.md | 3 +++ .../include/llvm/TargetParser/ARMTargetParser.def | 6 +++--- llvm/lib/TargetParser/ARMTargetParser.cpp | 4 ++++ llvm/unittests/TargetParser/TargetParserTest.cpp | 15 +++++++-------- 6 files changed, 26 insertions(+), 11 deletions(-) diff --git a/clang/test/Driver/print-supported-extensions-arm.c b/clang/test/Driver/print-supported-extensions-arm.c index 0dc2e9fc69738..407adc5e9c384 100644 --- a/clang/test/Driver/print-supported-extensions-arm.c +++ b/clang/test/Driver/print-supported-extensions-arm.c @@ -12,6 +12,7 @@ // CHECK-NEXT: dsp Supports DSP instructions in ARM and/or Thumb2 // CHECK-NEXT: mve Support M-Class Vector Extension with integer ops // CHECK-NEXT: mve.fp Support M-Class Vector Extension with integer and floating ops +// CHECK-NEXT: simd Enable NEON instructions // CHECK-NEXT: fp16 Enable half-precision floating point // CHECK-NEXT: ras Enable Reliability, Availability and Serviceability extensions // CHECK-NEXT: fp16fml Enable full half-precision floating point fml instructions diff --git a/clang/test/Preprocessor/arm-target-features.c b/clang/test/Preprocessor/arm-target-features.c index 94dcfc2424bb1..b36dd1765b6de 100644 --- a/clang/test/Preprocessor/arm-target-features.c +++ b/clang/test/Preprocessor/arm-target-features.c @@ -1027,3 +1027,11 @@ // CHECK-R52-NEXT: #define __ARM_VFPV4__ 1 // CHECK-R52-NOT: #define __ARM_NEON 1 // CHECK-R52-NOT: #define __ARM_NEON__ + +// Check that on AArch32 appropriate targets, +nosimd correctly disables NEON instructions. +// RUN: %clang -target arm-none-eabi -march=armv8-a+nosimd -mfloat-abi=hard -x c -E -dM -o - %s | FileCheck -check-prefix=CHECK-NOSIMD %s +// RUN: %clang -target arm-none-eabi -mcpu=cortex-r52+nosimd -mfloat-abi=hard -x c -E -dM -o - %s | FileCheck -check-prefix=CHECK-NOSIMD %s +// RUN: %clang -target arm-none-eabi -mcpu=cortex-a57+nosimd -mfloat-abi=hard -x c -E -dM -o - %s | FileCheck -check-prefix=CHECK-NOSIMD %s +// CHECK-NOSIMD-NOT: #define __ARM_NEON 1 +// CHECK-NOSIMD-NOT: #define __ARM_NEON_FP 0x6 +// CHECK-NOSIMD-NOT: #define __ARM_NEON__ 1 diff --git a/llvm/docs/ReleaseNotes.md b/llvm/docs/ReleaseNotes.md index d7a80ae93aa34..43ed754183be7 100644 --- a/llvm/docs/ReleaseNotes.md +++ b/llvm/docs/ReleaseNotes.md @@ -85,6 +85,9 @@ Changes to the AMDGPU Backend Changes to the ARM Backend -------------------------- +* The `+nosimd` attribute is now fully supported. Previously, this had no effect when being used with +AArch32 targets, however will now disable NEON instructions being generated. The `simd` is also now +printed when the `--print-supported-extensions` option is used.. Changes to the AVR Backend -------------------------- diff --git a/llvm/include/llvm/TargetParser/ARMTargetParser.def b/llvm/include/llvm/TargetParser/ARMTargetParser.def index e515ab665d117..e8837cb89ed29 100644 --- a/llvm/include/llvm/TargetParser/ARMTargetParser.def +++ b/llvm/include/llvm/TargetParser/ARMTargetParser.def @@ -229,7 +229,7 @@ ARM_ARCH_EXT_NAME("mve.fp", (ARM::AEK_DSP | ARM::AEK_MVE | ARM::AEK_FP), "+mve.fp", "-mve.fp") ARM_ARCH_EXT_NAME("idiv", (ARM::AEK_HWDIVARM | ARM::AEK_HWDIVTHUMB), {}, {}) ARM_ARCH_EXT_NAME("mp", ARM::AEK_MP, {}, {}) -ARM_ARCH_EXT_NAME("simd", ARM::AEK_SIMD, {}, {}) +ARM_ARCH_EXT_NAME("simd", ARM::AEK_SIMD, "+neon", "-neon") ARM_ARCH_EXT_NAME("sec", ARM::AEK_SEC, {}, {}) ARM_ARCH_EXT_NAME("virt", ARM::AEK_VIRT, {}, {}) ARM_ARCH_EXT_NAME("fp16", ARM::AEK_FP16, "+fullfp16", "-fullfp16") @@ -334,8 +334,8 @@ ARM_CPU_NAME("cortex-r7", ARMV7R, FK_VFPV3_D16_FP16, false, (ARM::AEK_MP | ARM::AEK_HWDIVARM)) ARM_CPU_NAME("cortex-r8", ARMV7R, FK_VFPV3_D16_FP16, false, (ARM::AEK_MP | ARM::AEK_HWDIVARM)) -ARM_CPU_NAME("cortex-r52", ARMV8R, FK_NEON_FP_ARMV8, false, ARM::AEK_NONE) -ARM_CPU_NAME("cortex-r52plus", ARMV8R, FK_NEON_FP_ARMV8, false, ARM::AEK_NONE) +ARM_CPU_NAME("cortex-r52", ARMV8R, FK_NEON_FP_ARMV8, false, ARM::AEK_SIMD) +ARM_CPU_NAME("cortex-r52plus", ARMV8R, FK_NEON_FP_ARMV8, false, ARM::AEK_SIMD) ARM_CPU_NAME("sc300", ARMV7M, FK_NONE, false, ARM::AEK_NONE) ARM_CPU_NAME("cortex-m3", ARMV7M, FK_NONE, true, ARM::AEK_NONE) ARM_CPU_NAME("cortex-m4", ARMV7EM, FK_FPV4_SP_D16, true, ARM::AEK_NONE) diff --git a/llvm/lib/TargetParser/ARMTargetParser.cpp b/llvm/lib/TargetParser/ARMTargetParser.cpp index 8f9753775c204..a7a895d872668 100644 --- a/llvm/lib/TargetParser/ARMTargetParser.cpp +++ b/llvm/lib/TargetParser/ARMTargetParser.cpp @@ -657,6 +657,10 @@ void ARM::PrintSupportedExtensions(StringMap<StringRef> DescMap) { // Extensions without a feature cannot be used with -march. if (!Ext.Feature.empty()) { std::string Description = DescMap[Ext.Name].str(); + // With SIMD, this links to the NEON feature, so the description should be + // taken from here, as SIMD does not exist in TableGen. + if (Ext.Name == "simd") + Description = DescMap["neon"].str(); outs() << " " << format(Description.empty() ? "%s\n" : "%-20s%s\n", Ext.Name.str().c_str(), Description.c_str()); diff --git a/llvm/unittests/TargetParser/TargetParserTest.cpp b/llvm/unittests/TargetParser/TargetParserTest.cpp index 574d69f847b4d..2a97d1c80ce8d 100644 --- a/llvm/unittests/TargetParser/TargetParserTest.cpp +++ b/llvm/unittests/TargetParser/TargetParserTest.cpp @@ -318,14 +318,14 @@ INSTANTIATE_TEST_SUITE_P( ARM::AEK_HWDIVTHUMB | ARM::AEK_DSP, "7-R"), ARMCPUTestParams<uint64_t>("cortex-r52", "armv8-r", "neon-fp-armv8", - ARM::AEK_NONE | ARM::AEK_CRC | ARM::AEK_MP | - ARM::AEK_VIRT | ARM::AEK_HWDIVARM | - ARM::AEK_HWDIVTHUMB | ARM::AEK_DSP, + ARM::AEK_CRC | ARM::AEK_MP | ARM::AEK_VIRT | + ARM::AEK_HWDIVARM | ARM::AEK_HWDIVTHUMB | + ARM::AEK_DSP | ARM::AEK_SIMD, "8-R"), ARMCPUTestParams<uint64_t>("cortex-r52plus", "armv8-r", "neon-fp-armv8", - ARM::AEK_NONE | ARM::AEK_CRC | ARM::AEK_MP | - ARM::AEK_VIRT | ARM::AEK_HWDIVARM | - ARM::AEK_HWDIVTHUMB | ARM::AEK_DSP, + ARM::AEK_CRC | ARM::AEK_MP | ARM::AEK_VIRT | + ARM::AEK_HWDIVARM | ARM::AEK_HWDIVTHUMB | + ARM::AEK_DSP | ARM::AEK_SIMD, "8-R"), ARMCPUTestParams<uint64_t>("sc300", "armv7-m", "none", ARM::AEK_NONE | ARM::AEK_HWDIVTHUMB, "7-M"), @@ -801,7 +801,7 @@ TEST(TargetParserTest, ARMArchExtFeature) { {"fp", "nofp", nullptr, nullptr}, {"idiv", "noidiv", nullptr, nullptr}, {"mp", "nomp", nullptr, nullptr}, - {"simd", "nosimd", nullptr, nullptr}, + {"simd", "nosimd", "+neon", "-neon"}, {"sec", "nosec", nullptr, nullptr}, {"virt", "novirt", nullptr, nullptr}, {"fp16", "nofp16", "+fullfp16", "-fullfp16"}, @@ -1046,7 +1046,6 @@ TEST(TargetParserTest, ARMPrintSupportedExtensions) { EXPECT_EQ(std::string::npos, captured.find("invalid")); // Should not include anything that lacks a feature name. Checking a few here // but not all as if one is hidden correctly the rest should be. - EXPECT_EQ(std::string::npos, captured.find("simd")); EXPECT_EQ(std::string::npos, captured.find("maverick")); EXPECT_EQ(std::string::npos, captured.find("xscale")); } >From 45337928c5ef679cdc7c47d268cb9c9cd40191b4 Mon Sep 17 00:00:00 2001 From: Jack Styles <jack.sty...@arm.com> Date: Tue, 11 Mar 2025 09:35:15 +0000 Subject: [PATCH 3/4] Improve Release Notes --- llvm/docs/ReleaseNotes.md | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/llvm/docs/ReleaseNotes.md b/llvm/docs/ReleaseNotes.md index 43ed754183be7..575a20288e685 100644 --- a/llvm/docs/ReleaseNotes.md +++ b/llvm/docs/ReleaseNotes.md @@ -86,8 +86,8 @@ Changes to the AMDGPU Backend Changes to the ARM Backend -------------------------- * The `+nosimd` attribute is now fully supported. Previously, this had no effect when being used with -AArch32 targets, however will now disable NEON instructions being generated. The `simd` is also now -printed when the `--print-supported-extensions` option is used.. +AArch32 targets, however this will now disable NEON instructions being generated. The `simd` option is +also now printed when the `--print-supported-extensions` option is used. Changes to the AVR Backend -------------------------- >From ece428714cfd9199e697f122ef8c1dec2307a837 Mon Sep 17 00:00:00 2001 From: Jack Styles <jack.sty...@arm.com> Date: Wed, 12 Mar 2025 08:44:12 +0000 Subject: [PATCH 4/4] Move release notes to clang --- clang/docs/ReleaseNotes.rst | 4 ++++ llvm/docs/ReleaseNotes.md | 3 --- 2 files changed, 4 insertions(+), 3 deletions(-) diff --git a/clang/docs/ReleaseNotes.rst b/clang/docs/ReleaseNotes.rst index 004f78f22ac36..e6a8d920dfb4c 100644 --- a/clang/docs/ReleaseNotes.rst +++ b/clang/docs/ReleaseNotes.rst @@ -290,6 +290,10 @@ Bug Fixes to Attribute Support or ``__attribute__((malloc(deallocator, ptr-index)))`` (`#51607 <https://github.com/llvm/llvm-project/issues/51607>`_). +- The ``+nosimd`` attribute is now fully supported. Previously, this had no effect when being used with +AArch32 targets, however this will now disable NEON instructions being generated. The ``simd`` option is +also now printed when the ``--print-supported-extensions`` option is used. + Bug Fixes to C++ Support ^^^^^^^^^^^^^^^^^^^^^^^^ diff --git a/llvm/docs/ReleaseNotes.md b/llvm/docs/ReleaseNotes.md index 575a20288e685..d7a80ae93aa34 100644 --- a/llvm/docs/ReleaseNotes.md +++ b/llvm/docs/ReleaseNotes.md @@ -85,9 +85,6 @@ Changes to the AMDGPU Backend Changes to the ARM Backend -------------------------- -* The `+nosimd` attribute is now fully supported. Previously, this had no effect when being used with -AArch32 targets, however this will now disable NEON instructions being generated. The `simd` option is -also now printed when the `--print-supported-extensions` option is used. Changes to the AVR Backend -------------------------- _______________________________________________ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits