[llvm-branch-commits] [llvm] ARM: Mark more generic libgcc functions as available (PR #210961)
Matt Arsenault via llvm-branch-commits
llvm-branch-commits at lists.llvm.org
Tue Jul 21 05:26:18 PDT 2026
https://github.com/arsenm created https://github.com/llvm/llvm-project/pull/210961
Generic libgcc/compiler-rt functions coexist with aeabi variants
(e.g., __divsi3 and __aeabi_idiv) according to my reading of the
build. At least in compiler-rt, they are aliases (such that I'm not sure
what the point of ever emitting the __aeabi name is).
They were previously removed from the available set on AEABI+AAPCS targets
to force selection of the preferred __aeabi_* variants, back when
only one implementation per libcall could be recorded.
Now that multiple implementations can be available per libcall, stop hiding
the generics and select the __aeabi_* variant explicitly as the preferred
implemntation. This reduces the number of special cases to consider for
future libcalls info improvements.
Co-authored-by: Claude (Claude-Opus-4.8) <noreply at anthropic.com>
>From 9a85c264daab05b7226e58a0b9a564d519eed14e Mon Sep 17 00:00:00 2001
From: Matt Arsenault <Matthew.Arsenault at amd.com>
Date: Tue, 21 Jul 2026 13:20:25 +0200
Subject: [PATCH] ARM: Mark more generic libgcc functions as available
Generic libgcc/compiler-rt functions coexist with aeabi variants
(e.g., __divsi3 and __aeabi_idiv) according to my reading of the
build. At least in compiler-rt, they are aliases (such that I'm not sure
what the point of ever emitting the __aeabi name is).
They were previously removed from the available set on AEABI+AAPCS targets
to force selection of the preferred __aeabi_* variants, back when
only one implementation per libcall could be recorded.
Now that multiple implementations can be available per libcall, stop hiding
the generics and select the __aeabi_* variant explicitly as the preferred
implemntation. This reduces the number of special cases to consider for
future libcalls info improvements.
Co-authored-by: Claude (Claude-Opus-4.8) <noreply at anthropic.com>
---
llvm/include/llvm/IR/RuntimeLibcalls.td | 7 +--
llvm/lib/Target/ARM/ARMSubtarget.cpp | 64 +++++++++++++++++++++++++
2 files changed, 65 insertions(+), 6 deletions(-)
diff --git a/llvm/include/llvm/IR/RuntimeLibcalls.td b/llvm/include/llvm/IR/RuntimeLibcalls.td
index 802188be31d09..e6a57c57c0adb 100644
--- a/llvm/include/llvm/IR/RuntimeLibcalls.td
+++ b/llvm/include/llvm/IR/RuntimeLibcalls.td
@@ -2520,12 +2520,7 @@ def ARMSystemLibrary
EABIHalfConvertCalls,
GNUEABIHalfConvertCalls,
ARMDoubleToHalfCalls,
-
- LibcallImpls<(add AEABIOverrides),
- RuntimeLibcallAvailability<[{
- (!hasAEABILibcalls(TT) || !isAAPCS_ABI(TT, ABIName)) &&
- !TT.isOSWindows()
- }]>>,
+ LibcallImpls<(add AEABIOverrides), isNotOSWindows>,
// Use divmod compiler-rt calls for iOS 5.0 and later.
LibcallImpls<(add __divmodsi4, __udivmodsi4),
RuntimeLibcallAvailability<[{TT.isOSBinFormatMachO() &&
diff --git a/llvm/lib/Target/ARM/ARMSubtarget.cpp b/llvm/lib/Target/ARM/ARMSubtarget.cpp
index 137f5e15af0cb..2f0dcf7a11a69 100644
--- a/llvm/lib/Target/ARM/ARMSubtarget.cpp
+++ b/llvm/lib/Target/ARM/ARMSubtarget.cpp
@@ -197,6 +197,70 @@ void ARMSubtarget::initLibcallLoweringInfo(LibcallLoweringInfo &Info) const {
Info.setLibcallImpl(LC.Op, LC.Impl);
}
}
+
+ static const struct {
+ const RTLIB::Libcall Op;
+ const RTLIB::LibcallImpl Impl;
+ } AEABISelected[] = {
+ // Double-precision arithmetic.
+ {RTLIB::ADD_F64, RTLIB::impl___aeabi_dadd},
+ {RTLIB::DIV_F64, RTLIB::impl___aeabi_ddiv},
+ {RTLIB::MUL_F64, RTLIB::impl___aeabi_dmul},
+ {RTLIB::SUB_F64, RTLIB::impl___aeabi_dsub},
+ // Double-precision comparisons.
+ {RTLIB::OEQ_F64, RTLIB::impl___aeabi_dcmpeq__oeq},
+ {RTLIB::UNE_F64, RTLIB::impl___aeabi_dcmpeq__une},
+ {RTLIB::OLT_F64, RTLIB::impl___aeabi_dcmplt},
+ {RTLIB::OLE_F64, RTLIB::impl___aeabi_dcmple},
+ {RTLIB::OGE_F64, RTLIB::impl___aeabi_dcmpge},
+ {RTLIB::OGT_F64, RTLIB::impl___aeabi_dcmpgt},
+ {RTLIB::UO_F64, RTLIB::impl___aeabi_dcmpun},
+ // Single-precision arithmetic.
+ {RTLIB::ADD_F32, RTLIB::impl___aeabi_fadd},
+ {RTLIB::DIV_F32, RTLIB::impl___aeabi_fdiv},
+ {RTLIB::MUL_F32, RTLIB::impl___aeabi_fmul},
+ {RTLIB::SUB_F32, RTLIB::impl___aeabi_fsub},
+ // Single-precision comparisons.
+ {RTLIB::OEQ_F32, RTLIB::impl___aeabi_fcmpeq__oeq},
+ {RTLIB::UNE_F32, RTLIB::impl___aeabi_fcmpeq__une},
+ {RTLIB::OLT_F32, RTLIB::impl___aeabi_fcmplt},
+ {RTLIB::OLE_F32, RTLIB::impl___aeabi_fcmple},
+ {RTLIB::OGE_F32, RTLIB::impl___aeabi_fcmpge},
+ {RTLIB::OGT_F32, RTLIB::impl___aeabi_fcmpgt},
+ {RTLIB::UO_F32, RTLIB::impl___aeabi_fcmpun},
+ // Floating-point to integer conversions.
+ {RTLIB::FPTOSINT_F64_I32, RTLIB::impl___aeabi_d2iz},
+ {RTLIB::FPTOUINT_F64_I32, RTLIB::impl___aeabi_d2uiz},
+ {RTLIB::FPTOSINT_F64_I64, RTLIB::impl___aeabi_d2lz},
+ {RTLIB::FPTOUINT_F64_I64, RTLIB::impl___aeabi_d2ulz},
+ {RTLIB::FPTOSINT_F32_I32, RTLIB::impl___aeabi_f2iz},
+ {RTLIB::FPTOUINT_F32_I32, RTLIB::impl___aeabi_f2uiz},
+ {RTLIB::FPTOSINT_F32_I64, RTLIB::impl___aeabi_f2lz},
+ {RTLIB::FPTOUINT_F32_I64, RTLIB::impl___aeabi_f2ulz},
+ // Integer to floating-point conversions.
+ {RTLIB::SINTTOFP_I32_F64, RTLIB::impl___aeabi_i2d},
+ {RTLIB::UINTTOFP_I32_F64, RTLIB::impl___aeabi_ui2d},
+ {RTLIB::SINTTOFP_I64_F64, RTLIB::impl___aeabi_l2d},
+ {RTLIB::UINTTOFP_I64_F64, RTLIB::impl___aeabi_ul2d},
+ {RTLIB::SINTTOFP_I32_F32, RTLIB::impl___aeabi_i2f},
+ {RTLIB::UINTTOFP_I32_F32, RTLIB::impl___aeabi_ui2f},
+ {RTLIB::SINTTOFP_I64_F32, RTLIB::impl___aeabi_l2f},
+ {RTLIB::UINTTOFP_I64_F32, RTLIB::impl___aeabi_ul2f},
+ // Long long helpers.
+ {RTLIB::MUL_I64, RTLIB::impl___aeabi_lmul},
+ {RTLIB::SHL_I64, RTLIB::impl___aeabi_llsl},
+ {RTLIB::SRL_I64, RTLIB::impl___aeabi_llsr},
+ {RTLIB::SRA_I64, RTLIB::impl___aeabi_lasr},
+ // Integer division.
+ {RTLIB::SDIV_I32, RTLIB::impl___aeabi_idiv},
+ {RTLIB::UDIV_I32, RTLIB::impl___aeabi_uidiv},
+ };
+
+ const RTLIB::RuntimeLibcallsInfo &RTLCI = Info.getRuntimeLibcallsInfo();
+ for (const auto &LC : AEABISelected) {
+ if (RTLCI.isAvailable(LC.Impl))
+ Info.setLibcallImpl(LC.Op, LC.Impl);
+ }
}
bool ARMSubtarget::isXRaySupported() const {
More information about the llvm-branch-commits
mailing list