[Clang] Add elementwise ldexp builtin function

[wenju-he]

This PR adds __builtin_elementwise_ldexp. It can be used for implementing OpenCL ldexp builtin with vector inputs.

[llvmbot]

@llvm/pr-subscribers-clang-codegen

@llvm/pr-subscribers-clang

Author: Wenju He (wenju-he)

Changes

This PR adds __builtin_elementwise_ldexp. It can be used for implementing OpenCL ldexp builtin with vector inputs.

---

Full diff: https://github.com/llvm/llvm-project/pull/166296.diff

7 Files Affected:

• (modified) clang/docs/LanguageExtensions.rst (+2)
• (modified) clang/docs/ReleaseNotes.rst (+2)
• (modified) clang/include/clang/Basic/Builtins.td (+6)
• (modified) clang/lib/CodeGen/CGBuiltin.cpp (+6)
• (modified) clang/lib/Sema/SemaChecking.cpp (+52-12)
• (modified) clang/test/CodeGen/builtins-elementwise-math.c (+31)
• (modified) clang/test/Sema/builtins-elementwise-math.c (+36)

diff --git a/clang/docs/LanguageExtensions.rst b/clang/docs/LanguageExtensions.rst index 495f2ab3926ce..ed3573012d680 100644 --- a/clang/docs/LanguageExtensions.rst +++ b/clang/docs/LanguageExtensions.rst @@ -805,6 +805,8 @@ of different sizes and signs is forbidden in binary and ternary builtins. T __builtin_elementwise_exp(T x) returns the base-e exponential, e^x, of the specified value floating point types T __builtin_elementwise_exp2(T x) returns the base-2 exponential, 2^x, of the specified value floating point types T __builtin_elementwise_exp10(T x) returns the base-10 exponential, 10^x, of the specified value floating point types + T __builtin_elementwise_ldexp(T x, IntT y) returns the product of x and 2 raised to the power y. The floating point types + number of elements in y must equal the number of elements in x. T __builtin_elementwise_sqrt(T x) return the square root of a floating-point number floating point types T __builtin_elementwise_roundeven(T x) round x to the nearest integer value in floating point format, floating point types diff --git a/clang/docs/ReleaseNotes.rst b/clang/docs/ReleaseNotes.rst index 92fc9381a5868..1d9021dfc6017 100644 --- a/clang/docs/ReleaseNotes.rst +++ b/clang/docs/ReleaseNotes.rst @@ -207,6 +207,8 @@ C23 Feature Support Non-comprehensive list of changes in this release ------------------------------------------------- +- Added ``__builtin_elementwise_ldexp``. + - Added ``__builtin_elementwise_fshl`` and ``__builtin_elementwise_fshr``. - ``__builtin_elementwise_abs`` can now be used in constant expression. diff --git a/clang/include/clang/Basic/Builtins.td b/clang/include/clang/Basic/Builtins.td index 2b400b012d6ed..9b8f7864329a9 100644 --- a/clang/include/clang/Basic/Builtins.td +++ b/clang/include/clang/Basic/Builtins.td @@ -1418,6 +1418,12 @@ def ElementwiseExp10 : Builtin { let Prototype = "void(...)"; } +def ElementwiseLdexp : Builtin { + let Spellings = ["__builtin_elementwise_ldexp"]; + let Attributes = [NoThrow, Const, CustomTypeChecking]; + let Prototype = "void(...)"; +} + def ElementwiseFloor : Builtin { let Spellings = ["__builtin_elementwise_floor"]; let Attributes = [NoThrow, Const, CustomTypeChecking]; diff --git a/clang/lib/CodeGen/CGBuiltin.cpp b/clang/lib/CodeGen/CGBuiltin.cpp index b81e0d02da2c9..bbcee34b384c0 100644 --- a/clang/lib/CodeGen/CGBuiltin.cpp +++ b/clang/lib/CodeGen/CGBuiltin.cpp @@ -3992,6 +3992,12 @@ RValue CodeGenFunction::EmitBuiltinExpr(const GlobalDecl GD, unsigned BuiltinID, case Builtin::BI__builtin_elementwise_exp10: return RValue::get(emitBuiltinWithOneOverloadedType<1>( *this, E, Intrinsic::exp10, "elt.exp10")); + case Builtin::BI__builtin_elementwise_ldexp: { + Value *Src = EmitScalarExpr(E->getArg(0)); + Value *Exp = EmitScalarExpr(E->getArg(1)); + Value *Result = Builder.CreateLdexp(Src, Exp, {}, "elt.ldexp"); + return RValue::get(Result); + } case Builtin::BI__builtin_elementwise_log: return RValue::get(emitBuiltinWithOneOverloadedType<1>( *this, E, Intrinsic::log, "elt.log")); diff --git a/clang/lib/Sema/SemaChecking.cpp b/clang/lib/Sema/SemaChecking.cpp index ad2c2e4a97bb9..a8e3fe6c07b12 100644 --- a/clang/lib/Sema/SemaChecking.cpp +++ b/clang/lib/Sema/SemaChecking.cpp @@ -2609,6 +2609,18 @@ static ExprResult BuiltinInvoke(Sema &S, CallExpr *TheCall) { Args.drop_front(), TheCall->getRParenLoc()); } +// Performs a similar job to Sema::UsualUnaryConversions, but without any +// implicit promotion of integral/enumeration types. +static ExprResult BuiltinVectorMathConversions(Sema &S, Expr *E) { + // First, convert to an r-value. + ExprResult Res = S.DefaultFunctionArrayLvalueConversion(E); + if (Res.isInvalid()) + return ExprError(); + + // Promote floating-point types. + return S.UsualUnaryFPConversions(Res.get()); +} + ExprResult Sema::CheckBuiltinFunctionCall(FunctionDecl *FDecl, unsigned BuiltinID, CallExpr *TheCall) { @@ -3273,6 +3285,46 @@ Sema::CheckBuiltinFunctionCall(FunctionDecl *FDecl, unsigned BuiltinID, return ExprError(); break; + case Builtin::BI__builtin_elementwise_ldexp: { + if (checkArgCount(TheCall, 2)) + return ExprError(); + + ExprResult A = BuiltinVectorMathConversions(*this, TheCall->getArg(0)); + if (A.isInvalid()) + return ExprError(); + QualType TyA = A.get()->getType(); + if (checkMathBuiltinElementType(*this, A.get()->getBeginLoc(), TyA, + EltwiseBuiltinArgTyRestriction::FloatTy, 1)) + return ExprError(); + + ExprResult Exp = UsualUnaryConversions(TheCall->getArg(1)); + if (Exp.isInvalid()) + return ExprError(); + QualType TyExp = Exp.get()->getType(); + if (checkMathBuiltinElementType(*this, Exp.get()->getBeginLoc(), TyExp, + EltwiseBuiltinArgTyRestriction::IntegerTy, + 2)) + return ExprError(); + + // Check the two arguments are either scalars or vectors of equal length. + const auto *Vec0 = TyA->getAs<VectorType>(); + const auto *Vec1 = TyExp->getAs<VectorType>(); + unsigned Arg0Length = Vec0 ? Vec0->getNumElements() : 0; + unsigned Arg1Length = Vec1 ? Vec1->getNumElements() : 0; + if (Arg0Length != Arg1Length) { + Diag(Exp.get()->getBeginLoc(), + diag::err_typecheck_vector_lengths_not_equal) + << TyA << TyExp << A.get()->getSourceRange() + << Exp.get()->getSourceRange(); + return ExprError(); + } + + TheCall->setArg(0, A.get()); + TheCall->setArg(1, Exp.get()); + TheCall->setType(TyA); + break; + } + // These builtins restrict the element type to floating point // types only, and take in two arguments. case Builtin::BI__builtin_elementwise_minnum: @@ -15992,18 +16044,6 @@ void Sema::CheckAddressOfPackedMember(Expr *rhs) { _2, _3, _4)); } -// Performs a similar job to Sema::UsualUnaryConversions, but without any -// implicit promotion of integral/enumeration types. -static ExprResult BuiltinVectorMathConversions(Sema &S, Expr *E) { - // First, convert to an r-value. - ExprResult Res = S.DefaultFunctionArrayLvalueConversion(E); - if (Res.isInvalid()) - return ExprError(); - - // Promote floating-point types. - return S.UsualUnaryFPConversions(Res.get()); -} - bool Sema::PrepareBuiltinElementwiseMathOneArgCall( CallExpr *TheCall, EltwiseBuiltinArgTyRestriction ArgTyRestr) { if (checkArgCount(TheCall, 1)) diff --git a/clang/test/CodeGen/builtins-elementwise-math.c b/clang/test/CodeGen/builtins-elementwise-math.c index e9344d8fe0b8b..2df485f0155c3 100644 --- a/clang/test/CodeGen/builtins-elementwise-math.c +++ b/clang/test/CodeGen/builtins-elementwise-math.c @@ -6,6 +6,7 @@ typedef half half2 __attribute__((ext_vector_type(2))); typedef float float2 __attribute__((ext_vector_type(2))); typedef float float4 __attribute__((ext_vector_type(4))); typedef short int si8 __attribute__((ext_vector_type(8))); +typedef int int4 __attribute__((ext_vector_type(4))); typedef unsigned int u4 __attribute__((ext_vector_type(4))); typedef double double2 __attribute__((ext_vector_type(2))); typedef double double3 __attribute__((ext_vector_type(3))); @@ -729,6 +730,36 @@ void test_builtin_elementwise_exp10(float f1, float f2, double d1, double d2, vf2 = __builtin_elementwise_exp10(vf1); } +void test_builtin_elementwise_ldexp(float f1, float f2, double d1, double d2, + float4 vf1, float4 vf2, int i1, int4 vi1, short s1, long l1) { + // CHECK-LABEL: define void @test_builtin_elementwise_ldexp( + // CHECK: [[F1:%.+]] = load float, ptr %f1.addr, align 4 + // CHECK: [[I1:%.+]] = load i32, ptr %i1.addr, align 4 + // CHECK-NEXT: call float @llvm.ldexp.f32.i32(float [[F1]], i32 [[I1]]) + f2 = __builtin_elementwise_ldexp(f1, i1); + + // CHECK: [[F2:%.+]] = load float, ptr %f1.addr, align 4 + // CHECK: [[S1:%.+]] = load i16, ptr %s1.addr, align 2 + // CHECK: [[Ext1:%.+]] = sext i16 [[S1]] to i32 + // CHECK-NEXT: call float @llvm.ldexp.f32.i32(float [[F2]], i32 [[Ext1]]) + f2 = __builtin_elementwise_ldexp(f1, s1); + + // CHECK: [[F3:%.+]] = load float, ptr %f1.addr, align 4 + // CHECK: [[L1:%.+]] = load i64, ptr %l1.addr, align 8 + // CHECK-NEXT: call float @llvm.ldexp.f32.i64(float [[F3]], i64 [[L1]]) + f2 = __builtin_elementwise_ldexp(f1, l1); + + // CHECK: [[D1:%.+]] = load double, ptr %d1.addr, align 8 + // CHECK: [[I2:%.+]] = load i32, ptr %i1.addr, align 4 + // CHECK-NEXT: call double @llvm.ldexp.f64.i32(double [[D1]], i32 [[I2]]) + d2 = __builtin_elementwise_ldexp(d1, i1); + + // CHECK: [[VF1:%.+]] = load <4 x float>, ptr %vf1.addr, align 16 + // CHECK: [[VI1:%.+]] = load <4 x i32>, ptr %vi1.addr, align 16 + // CHECK-NEXT: call <4 x float> @llvm.ldexp.v4f32.v4i32(<4 x float> [[VF1]], <4 x i32> [[VI1]]) + vf2 = __builtin_elementwise_ldexp(vf1, vi1); +} + void test_builtin_elementwise_floor(float f1, float f2, double d1, double d2, float4 vf1, float4 vf2) { // CHECK-LABEL: define void @test_builtin_elementwise_floor( diff --git a/clang/test/Sema/builtins-elementwise-math.c b/clang/test/Sema/builtins-elementwise-math.c index f9df4a6f93e05..37be0e4ebbd28 100644 --- a/clang/test/Sema/builtins-elementwise-math.c +++ b/clang/test/Sema/builtins-elementwise-math.c @@ -645,6 +645,42 @@ void test_builtin_elementwise_exp10(int i, float f, double d, float4 v, int3 iv, // expected-error@-1 {{1st argument must be a scalar or vector of floating-point types (was 'unsigned4' (vector of 4 'unsigned int' values))}} } +void test_builtin_elementwise_ldexp(int i, float f, double d, float4 v, int3 iv, unsigned u, unsigned4 uv) { + + struct Foo s = __builtin_elementwise_ldexp(f, i); + // expected-error@-1 {{initializing 'struct Foo' with an expression of incompatible type 'float'}} + + f = __builtin_elementwise_ldexp(); + // expected-error@-1 {{too few arguments to function call, expected 2, have 0}} + + f = __builtin_elementwise_ldexp(f); + // expected-error@-1 {{too few arguments to function call, expected 2, have 1}} + + f = __builtin_elementwise_ldexp(f, i, i); + // expected-error@-1 {{too many arguments to function call, expected 2, have 3}} + + f = __builtin_elementwise_ldexp(i, i); + // expected-error@-1 {{1st argument must be a scalar or vector of floating-point types (was 'int')}} + + f = __builtin_elementwise_ldexp(f, f); + // expected-error@-1 {{2nd argument must be a scalar or vector of integer types (was 'float')}} + + f = __builtin_elementwise_ldexp(v, iv); + // expected-error@-1 {{vector operands do not have the same number of elements ('float4' (vector of 4 'float' values) and 'int3' (vector of 3 'int' values))}} + + v = __builtin_elementwise_ldexp(v, i); + // expected-error@-1 {{vector operands do not have the same number of elements ('float4' (vector of 4 'float' values) and 'int')}} + + v = __builtin_elementwise_ldexp(f, iv); + // expected-error@-1 {{vector operands do not have the same number of elements ('float' and 'int3' (vector of 3 'int' values))}} + + f = __builtin_elementwise_ldexp(u, i); + // expected-error@-1 {{1st argument must be a scalar or vector of floating-point types (was 'unsigned int')}} + + f = __builtin_elementwise_ldexp(uv, i); + // expected-error@-1 {{1st argument must be a scalar or vector of floating-point types (was 'unsigned4' (vector of 4 'unsigned int' values))}} +} + void test_builtin_elementwise_floor(int i, float f, double d, float4 v, int3 iv, unsigned u, unsigned4 uv) { struct Foo s = __builtin_elementwise_floor(f); 

[AaronBallman]

LGTM!