RuntimeLibcalls: Mostly move x86 configuration into tablegen #148575
Merged
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
arsenm
requested review from
DanielKristofKiss,
dpaoliello,
efriedma-quic,
nikic and
RKSimon
|
@llvm/pr-subscribers-backend-x86 @llvm/pr-subscribers-llvm-ir Author: Matt Arsenault (arsenm) ChangesThis is getting pretty ugly, but seems to be the worst of the It also may make sense to go back to transposing the architecture Full diff: https://github.com/llvm/llvm-project/pull/148575.diff 3 Files Affected:
diff --git a/llvm/include/llvm/IR/RuntimeLibcalls.h b/llvm/include/llvm/IR/RuntimeLibcalls.h
index 97a6389844439..c63c7cd64ec4d 100644
--- a/llvm/include/llvm/IR/RuntimeLibcalls.h
+++ b/llvm/include/llvm/IR/RuntimeLibcalls.h
@@ -174,6 +174,10 @@ struct RuntimeLibcallsInfo {
(TT.isAndroid() && !TT.isAndroidVersionLT(9));
}
+ static bool hasSinCos_f32_f64(const Triple &TT) {
+ return hasSinCos(TT) || TT.isPS();
+ }
+
LLVM_ABI void initDefaultLibCallImpls();
/// Generated by tablegen.
diff --git a/llvm/include/llvm/IR/RuntimeLibcalls.td b/llvm/include/llvm/IR/RuntimeLibcalls.td
index b0651a8ac9920..be2558a8b1c8a 100644
--- a/llvm/include/llvm/IR/RuntimeLibcalls.td
+++ b/llvm/include/llvm/IR/RuntimeLibcalls.td
@@ -17,12 +17,23 @@ class DuplicateLibcallImplWithPrefix<RuntimeLibcallImpl Impl, string prefix>
/// Libcall Predicates
def isOSDarwin : RuntimeLibcallPredicate<"TT.isOSDarwin()">;
+def isOSOpenBSD : RuntimeLibcallPredicate<"TT.isOSOpenBSD()">;
def isOSWindows : RuntimeLibcallPredicate<"TT.isOSWindows()">;
+def isNotOSWindows : RuntimeLibcallPredicate<"!TT.isOSWindows()">;
+def isNotOSMSVCRT : RuntimeLibcallPredicate<"!TT.isOSMSVCRT()">;
+def isPS : RuntimeLibcallPredicate<"TT.isPS()">;
+def isNotOSWindowsOrIsCygwinMinGW
+ : RuntimeLibcallPredicate<"!TT.isOSWindows() || TT.isOSCygMing()">;
+
+def isGNUEnvironment : RuntimeLibcallPredicate<"TT.isGNUEnvironment()">;
def darwinHasSinCosStret : RuntimeLibcallPredicate<"darwinHasSinCosStret(TT)">;
def darwinHasExp10 : RuntimeLibcallPredicate<"darwinHasExp10(TT)">;
def hasSinCos : RuntimeLibcallPredicate<"hasSinCos(TT)">;
+// FIXME: Way to combine predicates
+def hasSinCos_f32_f64 : RuntimeLibcallPredicate<"hasSinCos_f32_f64(TT)">;
+
//--------------------------------------------------------------------
// Declare all kinds of used libcalls
//--------------------------------------------------------------------
@@ -514,11 +525,13 @@ def __divxf3 : RuntimeLibcallImpl<DIV_F80>;
def __divtf3 : RuntimeLibcallImpl<DIV_F128>;
def __gcc_qdiv : RuntimeLibcallImpl<DIV_PPCF128>;
+defset list<RuntimeLibcallImpl> PowiLibcallImpls = {
def __powisf2 : RuntimeLibcallImpl<POWI_F32>;
def __powidf2 : RuntimeLibcallImpl<POWI_F64>;
def __powixf2 : RuntimeLibcallImpl<POWI_F80>;
def __powitf2_f128 : RuntimeLibcallImpl<POWI_F128, "__powitf2">;
def __powitf2_ppc128 : RuntimeLibcallImpl<POWI_PPCF128, "__powitf2">;
+}
// Conversion
def __extendbfsf2 : RuntimeLibcallImpl<FPEXT_BF16_F32>;
@@ -1087,6 +1100,29 @@ defvar LibmHasSinCosF80 = LibcallImpls<(add sincos_f80), hasSinCos>;
defvar LibmHasSinCosF128 = LibcallImpls<(add sincos_f128), hasSinCos>;
defvar LibmHasSinCosPPCF128 = LibcallImpls<(add sincos_ppcf128), hasSinCos>;
+defvar WindowsMathRemovals = [
+ ldexpf, ldexp_f80, ldexp_f128, ldexp_ppcf128,
+ frexpf, frexp_f80, frexp_f128, frexp_ppcf128
+];
+
+defvar MostPowI = !listremove(PowiLibcallImpls, [__powitf2_f128]);
+defvar WindowsExclusions = !listconcat(WindowsMathRemovals, MostPowI);
+
+// Targets which support windows should start with these as a base and
+// add in calls for other OSes
+defvar Win32DefaultLibcallImpls = !listremove(DefaultLibcallImpls32, WindowsExclusions);
+defvar Win64DefaultLibcallImpls = !listremove(DefaultLibcallImpls64, WindowsExclusions);
+
+defvar LibmHasFrexpF32 = LibcallImpls<(add frexpf), isNotOSWindowsOrIsCygwinMinGW>;
+defvar LibmHasLdexpF32 = LibcallImpls<(add ldexpf), isNotOSWindowsOrIsCygwinMinGW>;
+
+defvar LibmHasFrexpF80 = LibcallImpls<(add frexp_f80), isNotOSWindowsOrIsCygwinMinGW>;
+defvar LibmHasLdexpF80 = LibcallImpls<(add ldexp_f80), isNotOSWindowsOrIsCygwinMinGW>;
+
+defvar LibmHasFrexpF128 = LibcallImpls<(add frexp_f128), isNotOSWindowsOrIsCygwinMinGW>;
+defvar LibmHasLdexpF128 = LibcallImpls<(add ldexp_f128), isNotOSWindowsOrIsCygwinMinGW>;
+
+
//===----------------------------------------------------------------------===//
// Objective-C Runtime Libcalls
//===----------------------------------------------------------------------===//
@@ -1154,7 +1190,10 @@ def isAArch64_ILP64 : RuntimeLibcallPredicate<"TT.isAArch64(64)">;
def AArch64SystemLibrary : SystemRuntimeLibrary<
isAArch64_ExceptArm64EC,
- (add DefaultRuntimeLibcallImpls,
+ (add Win64DefaultLibcallImpls,
+ LibcallImpls<(add __powisf2, __powidf2), isNotOSMSVCRT>,
+ LibmHasFrexpF32, LibmHasLdexpF32,
+ LibmHasFrexpF128, LibmHasLdexpF128,
AArch64LibcallImpls,
LibcallImpls<(add Int128RTLibcalls), isAArch64_ILP64>,
LibcallImpls<(add bzero), isOSDarwin>,
@@ -1164,7 +1203,7 @@ def AArch64SystemLibrary : SystemRuntimeLibrary<
// Prepend a # to every name
defset list<RuntimeLibcallImpl> WinArm64ECDefaultRuntimeLibcallImpls = {
- foreach libcall = DefaultLibcallImpls64 in {
+ foreach libcall = Win64DefaultLibcallImpls in {
def arm64ec_#libcall : DuplicateLibcallImplWithPrefix<libcall, "#">;
}
@@ -1421,11 +1460,15 @@ def isARMOrThumb : RuntimeLibcallPredicate<"TT.isARM() || TT.isThumb()">;
def ARMSystemLibrary
: SystemRuntimeLibrary<isARMOrThumb,
- (add DefaultLibcallImpls32,
+ (add Win32DefaultLibcallImpls,
+ LibcallImpls<(add __powisf2, __powidf2), isNotOSMSVCRT>,
+ LibmHasFrexpF32, LibmHasLdexpF32,
+ LibmHasFrexpF128, LibmHasLdexpF128,
WindowARMDivRemCalls,
WindowARMFPIntCasts,
AEABIDivRemCalls,
DarwinSinCosStret, DarwinExp10,
+ LibmHasSinCosF32, LibmHasSinCosF64, LibmHasSinCosF128,
// Use divmod compiler-rt calls for iOS 5.0 and later.
LibcallImpls<(add __divmodsi4, __udivmodsi4),
@@ -1919,6 +1962,8 @@ def PPCSystemLibrary
__extendkftf2, __trunctfkf2,
DefaultRuntimeLibcallImpls_ppcf128,
LibmF128Libcalls, AIX32Calls, AIX64Calls,
+ LibmHasSinCosF32, LibmHasSinCosF64, LibmHasSinCosF128,
+ LibmHasSinCosPPCF128,
AvailableIf<memcpy, isNotAIX>,
LibcallImpls<(add Int128RTLibcalls), isPPC64>)>;
@@ -1989,13 +2034,13 @@ def _Q_ulltoq : RuntimeLibcallImpl<UINTTOFP_I64_F128>;
// Windows Runtime Libcalls
//===----------------------------------------------------------------------===//
-// TT.isWindowsMSVCEnvironment() || TT.isWindowsItaniumEnvironment()
-
+defset list<RuntimeLibcallImpl> WindowsDivRemMulLibcalls = {
def _alldiv : RuntimeLibcallImpl<SDIV_I64>;
def _aulldiv : RuntimeLibcallImpl<UDIV_I64>;
def _allrem : RuntimeLibcallImpl<SREM_I64>;
def _aullrem : RuntimeLibcallImpl<UREM_I64>;
def _allmul : RuntimeLibcallImpl<MUL_I64>;
+}
//===----------------------------------------------------------------------===//
// X86 Runtime Libcalls
@@ -2003,14 +2048,45 @@ def _allmul : RuntimeLibcallImpl<MUL_I64>;
def isX86_32 : RuntimeLibcallPredicate<"TT.getArch() == Triple::x86">;
def isX86_64 : RuntimeLibcallPredicate<"TT.getArch() == Triple::x86_64">;
+def isX86 : RuntimeLibcallPredicate<"TT.isX86()">;
+
+// Some darwins have an optimized __bzero/bzero function.
+def darwinHas__bzero : RuntimeLibcallPredicate<"TT.isMacOSX() && !TT.isMacOSXVersionLT(10, 6)">;
+
+// Use the f128 variants of math functions on x86
+defvar X86_F128_Libcalls = LibcallImpls<(add LibmF128Libcalls, LibmF128FiniteLibcalls), isGNUEnvironment>;
+
+defvar SinCosF32F64Libcalls = LibcallImpls<(add sincosf, sincos), hasSinCos_f32_f64>;
+
+defvar X86CommonLibcalls =
+ (add DarwinSinCosStret, DarwinExp10,
+ X86_F128_Libcalls,
+ sincos_f80,
+ LibmHasSinCosF80, // FIXME: Depends on long double
+ SinCosF32F64Libcalls,
+ LibcallImpls<(add __bzero), darwinHas__bzero>,
+ LibmHasFrexpF32, LibmHasLdexpF32,
+ LibmHasFrexpF80, LibmHasLdexpF80,
+ DefaultRuntimeLibcallImpls_f80,
+ // MSVCRT doesn't have powi
+ // FIXME: It's almost certainly correct that MSVCRT has
+ // __powitf2_f128, but at least one test is relying on it.
+ __powitf2_f128,
+ LibcallImpls<(add MostPowI), isNotOSMSVCRT>
+);
+
+defvar Windows32DivRemMulCalls =
+ LibcallImpls<(add WindowsDivRemMulLibcalls),
+ RuntimeLibcallPredicate<"TT.isWindowsMSVCEnvironment() || TT.isWindowsItaniumEnvironment()">>;
def X86_32SystemLibrary
: SystemRuntimeLibrary<isX86_32,
- (add DefaultLibcallImpls32, DefaultRuntimeLibcallImpls_f80)>;
+ (add X86CommonLibcalls,
+ Windows32DivRemMulCalls, Win32DefaultLibcallImpls)>;
def X86_64SystemLibrary
: SystemRuntimeLibrary<isX86_64,
- (add DefaultLibcallImpls64, DefaultRuntimeLibcallImpls_f80)>;
+ (add X86CommonLibcalls, Win64DefaultLibcallImpls, Int128RTLibcalls)>;
//===----------------------------------------------------------------------===//
// XCore Runtime Libcalls
diff --git a/llvm/lib/IR/RuntimeLibcalls.cpp b/llvm/lib/IR/RuntimeLibcalls.cpp
index 74cde5e9cee92..60da0047291a9 100644
--- a/llvm/lib/IR/RuntimeLibcalls.cpp
+++ b/llvm/lib/IR/RuntimeLibcalls.cpp
@@ -62,69 +62,6 @@ static void setARMLibcallNames(RuntimeLibcallsInfo &Info, const Triple &TT,
Info.setLibcallImplCallingConv(Impl, CallingConv::ARM_AAPCS);
}
-static void setLongDoubleIsF128Libm(RuntimeLibcallsInfo &Info,
- bool FiniteOnlyFuncs = false) {
- Info.setLibcallImpl(RTLIB::REM_F128, RTLIB::fmodf128);
- Info.setLibcallImpl(RTLIB::FMA_F128, RTLIB::fmaf128);
- Info.setLibcallImpl(RTLIB::SQRT_F128, RTLIB::sqrtf128);
- Info.setLibcallImpl(RTLIB::CBRT_F128, RTLIB::cbrtf128);
- Info.setLibcallImpl(RTLIB::LOG_F128, RTLIB::logf128);
- Info.setLibcallImpl(RTLIB::LOG2_F128, RTLIB::log2f128);
- Info.setLibcallImpl(RTLIB::LOG10_F128, RTLIB::log10f128);
- Info.setLibcallImpl(RTLIB::EXP_F128, RTLIB::expf128);
- Info.setLibcallImpl(RTLIB::EXP2_F128, RTLIB::exp2f128);
- Info.setLibcallImpl(RTLIB::EXP10_F128, RTLIB::exp10f128);
- Info.setLibcallImpl(RTLIB::SIN_F128, RTLIB::sinf128);
- Info.setLibcallImpl(RTLIB::COS_F128, RTLIB::cosf128);
- Info.setLibcallImpl(RTLIB::TAN_F128, RTLIB::tanf128);
- Info.setLibcallImpl(RTLIB::SINCOS_F128, RTLIB::sincosf128);
- Info.setLibcallImpl(RTLIB::ASIN_F128, RTLIB::asinf128);
- Info.setLibcallImpl(RTLIB::ACOS_F128, RTLIB::acosf128);
- Info.setLibcallImpl(RTLIB::ATAN_F128, RTLIB::atanf128);
- Info.setLibcallImpl(RTLIB::ATAN2_F128, RTLIB::atan2f128);
- Info.setLibcallImpl(RTLIB::SINH_F128, RTLIB::sinhf128);
- Info.setLibcallImpl(RTLIB::COSH_F128, RTLIB::coshf128);
- Info.setLibcallImpl(RTLIB::TANH_F128, RTLIB::tanhf128);
- Info.setLibcallImpl(RTLIB::POW_F128, RTLIB::powf128);
- Info.setLibcallImpl(RTLIB::CEIL_F128, RTLIB::ceilf128);
- Info.setLibcallImpl(RTLIB::TRUNC_F128, RTLIB::truncf128);
- Info.setLibcallImpl(RTLIB::RINT_F128, RTLIB::rintf128);
- Info.setLibcallImpl(RTLIB::NEARBYINT_F128, RTLIB::nearbyintf128);
- Info.setLibcallImpl(RTLIB::ROUND_F128, RTLIB::roundf128);
- Info.setLibcallImpl(RTLIB::ROUNDEVEN_F128, RTLIB::roundevenf128);
- Info.setLibcallImpl(RTLIB::FLOOR_F128, RTLIB::floorf128);
- Info.setLibcallImpl(RTLIB::COPYSIGN_F128, RTLIB::copysignf128);
- Info.setLibcallImpl(RTLIB::FMIN_F128, RTLIB::fminf128);
- Info.setLibcallImpl(RTLIB::FMAX_F128, RTLIB::fmaxf128);
- Info.setLibcallImpl(RTLIB::FMINIMUM_F128, RTLIB::fminimumf128);
- Info.setLibcallImpl(RTLIB::FMAXIMUM_F128, RTLIB::fmaximumf128);
- Info.setLibcallImpl(RTLIB::FMINIMUM_NUM_F128, RTLIB::fminimum_numf128);
- Info.setLibcallImpl(RTLIB::FMAXIMUM_NUM_F128, RTLIB::fmaximum_numf128);
- Info.setLibcallImpl(RTLIB::LROUND_F128, RTLIB::lroundf128);
- Info.setLibcallImpl(RTLIB::LLROUND_F128, RTLIB::llroundf128);
- Info.setLibcallImpl(RTLIB::LRINT_F128, RTLIB::lrintf128);
- Info.setLibcallImpl(RTLIB::LLRINT_F128, RTLIB::llrintf128);
- Info.setLibcallImpl(RTLIB::LDEXP_F128, RTLIB::ldexpf128);
- Info.setLibcallImpl(RTLIB::FREXP_F128, RTLIB::frexpf128);
- Info.setLibcallImpl(RTLIB::MODF_F128, RTLIB::modff128);
-
- if (FiniteOnlyFuncs) {
- Info.setLibcallImpl(RTLIB::LOG_FINITE_F128, RTLIB::__logf128_finite);
- Info.setLibcallImpl(RTLIB::LOG2_FINITE_F128, RTLIB::__log2f128_finite);
- Info.setLibcallImpl(RTLIB::LOG10_FINITE_F128, RTLIB::__log10f128_finite);
- Info.setLibcallImpl(RTLIB::EXP_FINITE_F128, RTLIB::__expf128_finite);
- Info.setLibcallImpl(RTLIB::EXP2_FINITE_F128, RTLIB::__exp2f128_finite);
- Info.setLibcallImpl(RTLIB::POW_FINITE_F128, RTLIB::__powf128_finite);
- } else {
- Info.setLibcallImpl(RTLIB::LOG_FINITE_F128, RTLIB::Unsupported);
- Info.setLibcallImpl(RTLIB::LOG2_FINITE_F128, RTLIB::Unsupported);
- Info.setLibcallImpl(RTLIB::LOG10_FINITE_F128, RTLIB::Unsupported);
- Info.setLibcallImpl(RTLIB::EXP_FINITE_F128, RTLIB::Unsupported);
- Info.setLibcallImpl(RTLIB::EXP2_FINITE_F128, RTLIB::Unsupported);
- Info.setLibcallImpl(RTLIB::POW_FINITE_F128, RTLIB::Unsupported);
- }
-}
-
void RTLIB::RuntimeLibcallsInfo::initDefaultLibCallImpls() {
std::memcpy(LibcallImpls, DefaultLibcallImpls, sizeof(LibcallImpls));
static_assert(sizeof(LibcallImpls) == sizeof(DefaultLibcallImpls),
@@ -143,10 +80,6 @@ void RuntimeLibcallsInfo::initLibcalls(const Triple &TT,
if (TT.isAMDGPU() || TT.isNVPTX() || TT.isWasm())
return;
- // Use the f128 variants of math functions on x86
- if (TT.isX86() && TT.isGNUEnvironment())
- setLongDoubleIsF128Libm(*this, /*FiniteOnlyFuncs=*/true);
-
if (TT.isX86() || TT.isVE() || TT.isARM() || TT.isThumb()) {
if (ExceptionModel == ExceptionHandling::SjLj)
setLibcallImpl(RTLIB::UNWIND_RESUME, RTLIB::_Unwind_SjLj_Resume);
@@ -160,91 +93,40 @@ void RuntimeLibcallsInfo::initLibcalls(const Triple &TT,
setLibcallImpl(RTLIB::FPEXT_F16_F32, RTLIB::__extendhfsf2);
setLibcallImpl(RTLIB::FPROUND_F32_F16, RTLIB::__truncsfhf2);
- // Some darwins have an optimized __bzero/bzero function.
- if (TT.isX86()) {
- if (TT.isMacOSX() && !TT.isMacOSXVersionLT(10, 6))
- setLibcallImpl(RTLIB::BZERO, RTLIB::__bzero);
- }
-
- if (darwinHasSinCosStret(TT)) {
- setLibcallImpl(RTLIB::SINCOS_STRET_F32, RTLIB::__sincosf_stret);
- setLibcallImpl(RTLIB::SINCOS_STRET_F64, RTLIB::__sincos_stret);
- if (TT.isWatchABI()) {
- setLibcallImplCallingConv(RTLIB::__sincosf_stret,
- CallingConv::ARM_AAPCS_VFP);
- setLibcallImplCallingConv(RTLIB::__sincos_stret,
- CallingConv::ARM_AAPCS_VFP);
- }
+ if (TT.isWatchABI()) {
+ setLibcallImplCallingConv(RTLIB::__sincosf_stret,
+ CallingConv::ARM_AAPCS_VFP);
+ setLibcallImplCallingConv(RTLIB::__sincos_stret,
+ CallingConv::ARM_AAPCS_VFP);
}
- if (darwinHasExp10(TT)) {
- setLibcallImpl(RTLIB::EXP10_F32, RTLIB::__exp10f);
- setLibcallImpl(RTLIB::EXP10_F64, RTLIB::__exp10);
- } else {
+ if (!darwinHasExp10(TT)) {
+ // FIXME: Remove exp10 from default set
setLibcallImpl(RTLIB::EXP10_F32, RTLIB::Unsupported);
setLibcallImpl(RTLIB::EXP10_F64, RTLIB::Unsupported);
}
}
- if (hasSinCos(TT)) {
- setLibcallImpl(RTLIB::SINCOS_F32, RTLIB::sincosf);
- setLibcallImpl(RTLIB::SINCOS_F64, RTLIB::sincos);
- setLibcallImpl(RTLIB::SINCOS_F80, RTLIB::sincos_f80);
- setLibcallImpl(RTLIB::SINCOS_F128, RTLIB::sincos_f128);
- setLibcallImpl(RTLIB::SINCOS_PPCF128, RTLIB::sincos_ppcf128);
- }
-
- if (TT.isPS()) {
- setLibcallImpl(RTLIB::SINCOS_F32, RTLIB::sincosf);
- setLibcallImpl(RTLIB::SINCOS_F64, RTLIB::sincos);
- }
-
if (TT.isOSOpenBSD()) {
setLibcallImpl(RTLIB::STACKPROTECTOR_CHECK_FAIL, RTLIB::Unsupported);
}
- if (TT.isOSWindows() && !TT.isOSCygMing()) {
- setLibcallImpl(RTLIB::LDEXP_F32, RTLIB::Unsupported);
- setLibcallImpl(RTLIB::LDEXP_F80, RTLIB::Unsupported);
- setLibcallImpl(RTLIB::LDEXP_F128, RTLIB::Unsupported);
- setLibcallImpl(RTLIB::LDEXP_PPCF128, RTLIB::Unsupported);
-
- setLibcallImpl(RTLIB::FREXP_F32, RTLIB::Unsupported);
- setLibcallImpl(RTLIB::FREXP_F80, RTLIB::Unsupported);
- setLibcallImpl(RTLIB::FREXP_F128, RTLIB::Unsupported);
- setLibcallImpl(RTLIB::FREXP_PPCF128, RTLIB::Unsupported);
- }
+ // Skip default manual processing for targets that have been fully ported to
+ // tablegen for now. Eventually the rest of this should be deleted.
+ if (TT.isX86() || TT.isAArch64() || TT.isWasm())
+ return;
- if (TT.isOSMSVCRT()) {
- // MSVCRT doesn't have powi; fall back to pow
- setLibcallImpl(RTLIB::POWI_F32, RTLIB::Unsupported);
- setLibcallImpl(RTLIB::POWI_F64, RTLIB::Unsupported);
+ if (TT.isARM() || TT.isThumb()) {
+ setARMLibcallNames(*this, TT, FloatABI, EABIVersion);
+ return;
}
- // Setup Windows compiler runtime calls.
- if (TT.getArch() == Triple::x86 &&
- (TT.isWindowsMSVCEnvironment() || TT.isWindowsItaniumEnvironment())) {
- static const struct {
- const RTLIB::Libcall Op;
- const RTLIB::LibcallImpl Impl;
- const CallingConv::ID CC;
- } LibraryCalls[] = {
- {RTLIB::SDIV_I64, RTLIB::_alldiv, CallingConv::X86_StdCall},
- {RTLIB::UDIV_I64, RTLIB::_aulldiv, CallingConv::X86_StdCall},
- {RTLIB::SREM_I64, RTLIB::_allrem, CallingConv::X86_StdCall},
- {RTLIB::UREM_I64, RTLIB::_aullrem, CallingConv::X86_StdCall},
- {RTLIB::MUL_I64, RTLIB::_allmul, CallingConv::X86_StdCall},
- };
-
- for (const auto &LC : LibraryCalls) {
- setLibcallImpl(LC.Op, LC.Impl);
- setLibcallImplCallingConv(LC.Impl, LC.CC);
- }
+ if (hasSinCos(TT)) {
+ setLibcallImpl(RTLIB::SINCOS_F32, RTLIB::sincosf);
+ setLibcallImpl(RTLIB::SINCOS_F64, RTLIB::sincos);
+ setLibcallImpl(RTLIB::SINCOS_F128, RTLIB::sincos_f128);
}
- if (TT.isARM() || TT.isThumb())
- setARMLibcallNames(*this, TT, FloatABI, EABIVersion);
-
// These libcalls are only available in compiler-rt, not libgcc.
if (TT.isArch64Bit()) {
setLibcallImpl(RTLIB::SHL_I128, RTLIB::__ashlti3);
|
arsenm
force-pushed
the
users/arsenm/runtime-libcalls/x86-mostly-move-libcalls-tablegen
branch
from
bbd359e to
1a5ee21
Compare
arsenm
force-pushed
the
users/arsenm/runtime-libcalls/switch-compiler-rt-64-libcalls-opt-in
branch
from
2f195de to
e933483
Compare
arsenm
force-pushed
the
users/arsenm/runtime-libcalls/switch-compiler-rt-64-libcalls-opt-in
branch
from
e933483 to
b1cc318
Compare
Base automatically changed from
users/arsenm/runtime-libcalls/switch-compiler-rt-64-libcalls-opt-in
to
main
July 14, 2025 15:59
dpaoliello
approved these changes
Jul 14, 2025
| LibmHasFrexpF80, LibmHasLdexpF80, | ||
| DefaultRuntimeLibcallImpls_f80, | ||
| // MSVCRT doesn't have powi | ||
| // FIXME: It's almost certainly correct that MSVCRT has |
There was a problem hiding this comment.
This comment is confusing: does MSVC NOT have __powitf2_f128 but there's a test relying on it, or does MSVC have __powitf2_f128 and there's a test relying on it NOT having it.
There was a problem hiding this comment.
It doesn't have it, I assume (none of this stuff is documented but it would be weird if it had the f128 version and not the f32/f64). There's a test that tests all the types for the constrained powi intrinsic that happens to use an mcvc run line which currently compiles, this is avoiding turning that into an error
arsenm
force-pushed
the
users/arsenm/runtime-libcalls/x86-mostly-move-libcalls-tablegen
branch
from
1a5ee21 to
5915e12
Compare
This is getting pretty ugly, but seems to be the worst of the cases. Opting out of the base set of calls for the various windows cases is really ugly. We need to apply that to the arm cases as well. It also may make sense to go back to transposing the architecture and operating system (i.e. make isOSWindows the SystemLibrary and then modify based on architecture).
tgross35
reviewed
Jul 15, 2025
Comment on lines
1188
to
1194
| def AArch64SystemLibrary : SystemRuntimeLibrary< | ||
| isAArch64_ExceptArm64EC, | ||
| (add DefaultRuntimeLibcallImpls, | ||
| (add Win64DefaultLibcallImpls, | ||
| LibcallImpls<(add __powisf2, __powidf2), isNotOSMSVCRT>, | ||
| LibmHasFrexpF32, LibmHasLdexpF32, | ||
| LibmHasFrexpF128, LibmHasLdexpF128, | ||
| AArch64LibcallImpls, |
There was a problem hiding this comment.
Might be missing something here, but it doesn't look like AArch64SystemLibrary is only for Windows. Why did it change to Win64DefaultLibcallImpls rather than DefaultRuntimeLibcallImpls?
There was a problem hiding this comment.
Because Win64DefaultLibcallImpls is the subset supported by windows. Everything else is additive for different not windows
There was a problem hiding this comment.
That makes sense. Just surprising that the libcalls on most other targets tends to be a proper superset of Windows libcalls.
There was a problem hiding this comment.
It's not, there are some isOSWindows in there.
This whole thing is also messier than it should be, it should improve once the list of calls is fully separated from the lowering decision
arsenm
force-pushed
the
users/arsenm/runtime-libcalls/x86-mostly-move-libcalls-tablegen
branch
from
5915e12 to
27fd017
Compare
This was referenced Jul 15, 2025
arsenm
deleted the
users/arsenm/runtime-libcalls/x86-mostly-move-libcalls-tablegen
branch
|
This miscompiles for 32 bit x86 for MSVC targets. This is reproducible with https://martin.st/temp/avcodec-preproc.c as input. $ clang --target=i686-windows-msvc -S avcodec-preproc.c -o out.s -O2This change causes the following diff in the generated code: --- out-good.s 2025-07-16 16:02:51.333780818 +0300
+++ out-bad.s 2025-07-16 16:03:00.851790884 +0300
@@ -790,6 +790,7 @@
pushl $2147483647 # imm = 0x7FFFFFFF
pushl $-1
calll __alldiv
+ addl $16, %esp
movl %eax, (%esp) # 4-byte Spill
movl %edx, %ecx
movl %ebx, %eax(Plus a couple more similar cases.) |
|
Also, whatever fix is applied, also needs to be cherry-picked to the 21.x release branch. |
|
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.
This is getting pretty ugly, but seems to be the worst of the
cases. Opting out of the base set of calls for the various windows
cases is really ugly. We need to apply that to the arm cases as well.
It also may make sense to go back to transposing the architecture
and operating system (i.e. make isOSWindows the SystemLibrary
and then modify based on architecture).