[clang] Emit constraint intrinsics for arc and hyperbolic trig clang builtins by farzonl · Pull Request #98949 · llvm/llvm-project

farzonl · 2024-07-15T19:48:24Z

Change(s)

Builtins.td - Add f16 support for libm arc and hyperbolic trig functions
CGBuiltin.cpp - Emit constraint intrinsics for trig clang builtins

History

This change is part of an implementation of #87367 investigation on supporting IEEE math operations as intrinsics. Which was discussed in this RFC:
https://discourse.llvm.org/t/rfc-all-the-math-intrinsics/78294

This change adds wasm lowering cases for acos, asin, atan, cosh, sinh, and tanh.

#70079
#70080
#70081
#70083
#70084
#95966

Precursor PR(s)

Note this PR needs Merge after:

…builtins - `Builtins.td` - Add f16 support for libm arc and hyperbolic trig functions - `CGBuiltin.cpp` - Emit constraint intrinsics for trig clang builtins This change is part of an implementation of llvm#87367 investigation on supporting IEEE math operations as intrinsics. Which was discussed in this RFC: https://discourse.llvm.org/t/rfc-all-the-math-intrinsics/78294 This change adds wasm lowering cases for `acos`, `asin`, `atan`, `cosh`, `sinh`, and `tanh`. llvm#70079 llvm#70080 llvm#70081 llvm#70083 llvm#70084 llvm#95966 Note this PR needs Merge after: - llvm#98937 - llvm#98755

llvmbot · 2024-07-15T19:48:54Z

@llvm/pr-subscribers-backend-x86

@llvm/pr-subscribers-clang

Author: Farzon Lotfi (farzonl)

Changes

Change(s)

Builtins.td - Add f16 support for libm arc and hyperbolic trig functions
CGBuiltin.cpp - Emit constraint intrinsics for trig clang builtins

History

This change is part of an implementation of #87367 investigation on supporting IEEE math operations as intrinsics. Which was discussed in this RFC:
https://discourse.llvm.org/t/rfc-all-the-math-intrinsics/78294

This change adds wasm lowering cases for acos, asin, atan, cosh, sinh, and tanh.

#70079
#70080
#70081
#70083
#70084
#95966

Precursor PR(s)

Note this PR needs Merge after:

#98937
#98755

Patch is 28.53 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/98949.diff

6 Files Affected:

(modified) clang/include/clang/Basic/Builtins.td (+18-18)
(modified) clang/lib/CodeGen/CGBuiltin.cpp (+66)
(modified) clang/test/CodeGen/X86/math-builtins.c (+24-24)
(modified) clang/test/CodeGen/constrained-math-builtins.c (+41)
(modified) clang/test/CodeGen/math-libcalls.c (+37-36)
(modified) clang/test/CodeGenOpenCL/builtins-f16.cl (+18)

diff --git a/clang/include/clang/Basic/Builtins.td b/clang/include/clang/Basic/Builtins.td
index f5b15cf90d1f8..4133f6ff40cf3 100644
--- a/clang/include/clang/Basic/Builtins.td
+++ b/clang/include/clang/Basic/Builtins.td
@@ -83,11 +83,11 @@ class BitInt_Long_LongLongTemplate :
 // - _Constant: Argument has to constant-fold to an integer constant expression
 
 // __fp16 and __float128 builtin variants of libc/libm functions.
-def AcosF128 : Builtin {
-  let Spellings = ["__builtin_acosf128"];
+def AcosF16F128 : Builtin, F16F128MathTemplate {
+  let Spellings = ["__builtin_acos"];
   let Attributes = [FunctionWithBuiltinPrefix, NoThrow,
                     ConstIgnoringErrnoAndExceptions];
-  let Prototype = "__float128(__float128)";
+  let Prototype = "T(T)";
 }
 
 def AcoshF128 : Builtin {
@@ -97,11 +97,11 @@ def AcoshF128 : Builtin {
   let Prototype = "__float128(__float128)";
 }
 
-def AsinF128 : Builtin {
-  let Spellings = ["__builtin_asinf128"];
+def AsinF16F128 : Builtin, F16F128MathTemplate {
+  let Spellings = ["__builtin_asin"];
   let Attributes = [FunctionWithBuiltinPrefix, NoThrow,
                     ConstIgnoringErrnoAndExceptions];
-  let Prototype = "__float128(__float128)";
+  let Prototype = "T(T)";
 }
 
 def AsinhF128 : Builtin {
@@ -111,11 +111,11 @@ def AsinhF128 : Builtin {
   let Prototype = "__float128(__float128)";
 }
 
-def AtanF128 : Builtin {
-  let Spellings = ["__builtin_atanf128"];
+def AtanF16F128 : Builtin, F16F128MathTemplate {
+  let Spellings = ["__builtin_atan"];
   let Attributes = [FunctionWithBuiltinPrefix, NoThrow,
                     ConstIgnoringErrnoAndExceptions];
-  let Prototype = "__float128(__float128)";
+  let Prototype = "T(T)";
 }
 
 def AtanhF128 : Builtin {
@@ -143,10 +143,10 @@ def CosF16F128 : Builtin, F16F128MathTemplate {
   let Prototype = "T(T)";
 }
 
-def CoshF128 : Builtin {
-  let Spellings = ["__builtin_coshf128"];
+def CoshF16F128 : Builtin, F16F128MathTemplate {
+  let Spellings = ["__builtin_cosh"];
   let Attributes = [FunctionWithBuiltinPrefix, NoThrow, ConstIgnoringErrnoAndExceptions];
-  let Prototype = "__float128(__float128)";
+  let Prototype = "T(T)";
 }
 
 def ErfF128 : Builtin {
@@ -468,11 +468,11 @@ def SinF16F128 : Builtin, F16F128MathTemplate {
   let Prototype = "T(T)";
 }
 
-def SinhF128 : Builtin {
-  let Spellings = ["__builtin_sinhf128"];
+def SinhF16F128 : Builtin, F16F128MathTemplate {
+  let Spellings = ["__builtin_sinh"];
   let Attributes = [FunctionWithBuiltinPrefix, NoThrow,
                     ConstIgnoringErrnoAndExceptions];
-  let Prototype = "__float128(__float128)";
+  let Prototype = "T(T)";
 }
 
 def SqrtF16F128 : Builtin, F16F128MathTemplate {
@@ -489,11 +489,11 @@ def TanF16F128 : Builtin, F16F128MathTemplate {
   let Prototype = "T(T)";
 }
 
-def TanhF128 : Builtin {
-  let Spellings = ["__builtin_tanhf128"];
+def TanhF16F128 : Builtin, F16F128MathTemplate {
+  let Spellings = ["__builtin_tanh"];
   let Attributes = [FunctionWithBuiltinPrefix, NoThrow,
                     ConstIgnoringErrnoAndExceptions];
-  let Prototype = "__float128(__float128)";
+  let Prototype = "T(T)";
 }
 
 def TgammaF128 : Builtin {
diff --git a/clang/lib/CodeGen/CGBuiltin.cpp b/clang/lib/CodeGen/CGBuiltin.cpp
index a54fa7bf87aad..d78efb2217931 100644
--- a/clang/lib/CodeGen/CGBuiltin.cpp
+++ b/clang/lib/CodeGen/CGBuiltin.cpp
@@ -2640,6 +2640,39 @@ RValue CodeGenFunction::EmitBuiltinExpr(const GlobalDecl GD, unsigned BuiltinID,
   }
   if (GenerateIntrinsics) {
     switch (BuiltinIDIfNoAsmLabel) {
+    case Builtin::BIacos:
+    case Builtin::BIacosf:
+    case Builtin::BIacosl:
+    case Builtin::BI__builtin_acos:
+    case Builtin::BI__builtin_acosf:
+    case Builtin::BI__builtin_acosf16:
+    case Builtin::BI__builtin_acosl:
+    case Builtin::BI__builtin_acosf128:
+      return RValue::get(emitUnaryMaybeConstrainedFPBuiltin(
+          *this, E, Intrinsic::acos, Intrinsic::experimental_constrained_acos));
+
+    case Builtin::BIasin:
+    case Builtin::BIasinf:
+    case Builtin::BIasinl:
+    case Builtin::BI__builtin_asin:
+    case Builtin::BI__builtin_asinf:
+    case Builtin::BI__builtin_asinf16:
+    case Builtin::BI__builtin_asinl:
+    case Builtin::BI__builtin_asinf128:
+      return RValue::get(emitUnaryMaybeConstrainedFPBuiltin(
+          *this, E, Intrinsic::asin, Intrinsic::experimental_constrained_asin));
+
+    case Builtin::BIatan:
+    case Builtin::BIatanf:
+    case Builtin::BIatanl:
+    case Builtin::BI__builtin_atan:
+    case Builtin::BI__builtin_atanf:
+    case Builtin::BI__builtin_atanf16:
+    case Builtin::BI__builtin_atanl:
+    case Builtin::BI__builtin_atanf128:
+      return RValue::get(emitUnaryMaybeConstrainedFPBuiltin(
+          *this, E, Intrinsic::atan, Intrinsic::experimental_constrained_atan));
+
     case Builtin::BIceil:
     case Builtin::BIceilf:
     case Builtin::BIceill:
@@ -2675,6 +2708,17 @@ RValue CodeGenFunction::EmitBuiltinExpr(const GlobalDecl GD, unsigned BuiltinID,
                                    Intrinsic::cos,
                                    Intrinsic::experimental_constrained_cos));
 
+    case Builtin::BIcosh:
+    case Builtin::BIcoshf:
+    case Builtin::BIcoshl:
+    case Builtin::BI__builtin_cosh:
+    case Builtin::BI__builtin_coshf:
+    case Builtin::BI__builtin_coshf16:
+    case Builtin::BI__builtin_coshl:
+    case Builtin::BI__builtin_coshf128:
+      return RValue::get(emitUnaryMaybeConstrainedFPBuiltin(
+          *this, E, Intrinsic::cosh, Intrinsic::experimental_constrained_cosh));
+
     case Builtin::BIexp:
     case Builtin::BIexpf:
     case Builtin::BIexpl:
@@ -2891,6 +2935,17 @@ RValue CodeGenFunction::EmitBuiltinExpr(const GlobalDecl GD, unsigned BuiltinID,
                                    Intrinsic::sin,
                                    Intrinsic::experimental_constrained_sin));
 
+    case Builtin::BIsinh:
+    case Builtin::BIsinhf:
+    case Builtin::BIsinhl:
+    case Builtin::BI__builtin_sinh:
+    case Builtin::BI__builtin_sinhf:
+    case Builtin::BI__builtin_sinhf16:
+    case Builtin::BI__builtin_sinhl:
+    case Builtin::BI__builtin_sinhf128:
+      return RValue::get(emitUnaryMaybeConstrainedFPBuiltin(
+          *this, E, Intrinsic::sinh, Intrinsic::experimental_constrained_sinh));
+
     case Builtin::BIsqrt:
     case Builtin::BIsqrtf:
     case Builtin::BIsqrtl:
@@ -2917,6 +2972,17 @@ RValue CodeGenFunction::EmitBuiltinExpr(const GlobalDecl GD, unsigned BuiltinID,
       return RValue::get(emitUnaryMaybeConstrainedFPBuiltin(
           *this, E, Intrinsic::tan, Intrinsic::experimental_constrained_tan));
 
+    case Builtin::BItanh:
+    case Builtin::BItanhf:
+    case Builtin::BItanhl:
+    case Builtin::BI__builtin_tanh:
+    case Builtin::BI__builtin_tanhf:
+    case Builtin::BI__builtin_tanhf16:
+    case Builtin::BI__builtin_tanhl:
+    case Builtin::BI__builtin_tanhf128:
+      return RValue::get(emitUnaryMaybeConstrainedFPBuiltin(
+          *this, E, Intrinsic::tanh, Intrinsic::experimental_constrained_tanh));
+
     case Builtin::BItrunc:
     case Builtin::BItruncf:
     case Builtin::BItruncl:
diff --git a/clang/test/CodeGen/X86/math-builtins.c b/clang/test/CodeGen/X86/math-builtins.c
index d26db19574051..48465df21cca1 100644
--- a/clang/test/CodeGen/X86/math-builtins.c
+++ b/clang/test/CodeGen/X86/math-builtins.c
@@ -168,10 +168,10 @@ void foo(double *d, float f, float *fp, long double *l, int *i, const char *c) {
   /* math */
   __builtin_acos(f);       __builtin_acosf(f);      __builtin_acosl(f); __builtin_acosf128(f);
 
-// NO__ERRNO: declare double @acos(double noundef) [[READNONE]]
-// NO__ERRNO: declare float @acosf(float noundef) [[READNONE]]
-// NO__ERRNO: declare x86_fp80 @acosl(x86_fp80 noundef) [[READNONE]]
-// NO__ERRNO: declare fp128 @acosf128(fp128 noundef) [[READNONE]]
+// NO__ERRNO: declare double @llvm.acos.f64(double) [[READNONE_INTRINSIC]]
+// NO__ERRNO: declare float @llvm.acos.f32(float) [[READNONE_INTRINSIC]]
+// NO__ERRNO: declare x86_fp80 @llvm.acos.f80(x86_fp80) [[READNONE_INTRINSIC]]
+// NO__ERRNO: declare fp128 @llvm.acos.f128(fp128) [[READNONE_INTRINSIC]]
 // HAS_ERRNO: declare double @acos(double noundef) [[NOT_READNONE]]
 // HAS_ERRNO: declare float @acosf(float noundef) [[NOT_READNONE]]
 // HAS_ERRNO: declare x86_fp80 @acosl(x86_fp80 noundef) [[NOT_READNONE]]
@@ -190,10 +190,10 @@ void foo(double *d, float f, float *fp, long double *l, int *i, const char *c) {
 
   __builtin_asin(f);       __builtin_asinf(f);      __builtin_asinl(f); __builtin_asinf128(f);
 
-// NO__ERRNO: declare double @asin(double noundef) [[READNONE]]
-// NO__ERRNO: declare float @asinf(float noundef) [[READNONE]]
-// NO__ERRNO: declare x86_fp80 @asinl(x86_fp80 noundef) [[READNONE]]
-// NO__ERRNO: declare fp128 @asinf128(fp128 noundef) [[READNONE]]
+// NO__ERRNO: declare double @llvm.asin.f64(double) [[READNONE_INTRINSIC]]
+// NO__ERRNO: declare float @llvm.asin.f32(float) [[READNONE_INTRINSIC]]
+// NO__ERRNO: declare x86_fp80 @llvm.asin.f80(x86_fp80) [[READNONE_INTRINSIC]]
+// NO__ERRNO: declare fp128 @llvm.asin.f128(fp128) [[READNONE_INTRINSIC]]
 // HAS_ERRNO: declare double @asin(double noundef) [[NOT_READNONE]]
 // HAS_ERRNO: declare float @asinf(float noundef) [[NOT_READNONE]]
 // HAS_ERRNO: declare x86_fp80 @asinl(x86_fp80 noundef) [[NOT_READNONE]]
@@ -212,10 +212,10 @@ void foo(double *d, float f, float *fp, long double *l, int *i, const char *c) {
 
   __builtin_atan(f);       __builtin_atanf(f);      __builtin_atanl(f); __builtin_atanf128(f);
 
-// NO__ERRNO: declare double @atan(double noundef) [[READNONE]]
-// NO__ERRNO: declare float @atanf(float noundef) [[READNONE]]
-// NO__ERRNO: declare x86_fp80 @atanl(x86_fp80 noundef) [[READNONE]]
-// NO__ERRNO: declare fp128 @atanf128(fp128 noundef) [[READNONE]]
+// NO__ERRNO: declare double @llvm.atan.f64(double) [[READNONE_INTRINSIC]]
+// NO__ERRNO: declare float @llvm.atan.f32(float) [[READNONE_INTRINSIC]]
+// NO__ERRNO: declare x86_fp80 @llvm.atan.f80(x86_fp80) [[READNONE_INTRINSIC]]
+// NO__ERRNO: declare fp128 @llvm.atan.f128(fp128) [[READNONE_INTRINSIC]]
 // HAS_ERRNO: declare double @atan(double noundef) [[NOT_READNONE]]
 // HAS_ERRNO: declare float @atanf(float noundef) [[NOT_READNONE]]
 // HAS_ERRNO: declare x86_fp80 @atanl(x86_fp80 noundef) [[NOT_READNONE]]
@@ -267,10 +267,10 @@ void foo(double *d, float f, float *fp, long double *l, int *i, const char *c) {
 
   __builtin_cosh(f);       __builtin_coshf(f);      __builtin_coshl(f); __builtin_coshf128(f);
 
-// NO__ERRNO: declare double @cosh(double noundef) [[READNONE]]
-// NO__ERRNO: declare float @coshf(float noundef) [[READNONE]]
-// NO__ERRNO: declare x86_fp80 @coshl(x86_fp80 noundef) [[READNONE]]
-// NO__ERRNO: declare fp128 @coshf128(fp128 noundef) [[READNONE]]
+// NO__ERRNO: declare double @llvm.cosh.f64(double) [[READNONE_INTRINSIC]]
+// NO__ERRNO: declare float @llvm.cosh.f32(float) [[READNONE_INTRINSIC]]
+// NO__ERRNO: declare x86_fp80 @llvm.cosh.f80(x86_fp80) [[READNONE_INTRINSIC]]
+// NO__ERRNO: declare fp128 @llvm.cosh.f128(fp128) [[READNONE_INTRINSIC]]
 // HAS_ERRNO: declare double @cosh(double noundef) [[NOT_READNONE]]
 // HAS_ERRNO: declare float @coshf(float noundef) [[NOT_READNONE]]
 // HAS_ERRNO: declare x86_fp80 @coshl(x86_fp80 noundef) [[NOT_READNONE]]
@@ -656,10 +656,10 @@ __builtin_sin(f);        __builtin_sinf(f);       __builtin_sinl(f); __builtin_s
 
 __builtin_sinh(f);       __builtin_sinhf(f);      __builtin_sinhl(f); __builtin_sinhf128(f);
 
-// NO__ERRNO: declare double @sinh(double noundef) [[READNONE]]
-// NO__ERRNO: declare float @sinhf(float noundef) [[READNONE]]
-// NO__ERRNO: declare x86_fp80 @sinhl(x86_fp80 noundef) [[READNONE]]
-// NO__ERRNO: declare fp128 @sinhf128(fp128 noundef) [[READNONE]]
+// NO__ERRNO: declare double @llvm.sinh.f64(double) [[READNONE_INTRINSIC]]
+// NO__ERRNO: declare float @llvm.sinh.f32(float) [[READNONE_INTRINSIC]]
+// NO__ERRNO: declare x86_fp80 @llvm.sinh.f80(x86_fp80) [[READNONE_INTRINSIC]]
+// NO__ERRNO: declare fp128 @llvm.sinh.f128(fp128) [[READNONE_INTRINSIC]]
 // HAS_ERRNO: declare double @sinh(double noundef) [[NOT_READNONE]]
 // HAS_ERRNO: declare float @sinhf(float noundef) [[NOT_READNONE]]
 // HAS_ERRNO: declare x86_fp80 @sinhl(x86_fp80 noundef) [[NOT_READNONE]]
@@ -689,10 +689,10 @@ __builtin_tan(f);        __builtin_tanf(f);       __builtin_tanl(f); __builtin_t
 
 __builtin_tanh(f);       __builtin_tanhf(f);      __builtin_tanhl(f); __builtin_tanhf128(f);
 
-// NO__ERRNO: declare double @tanh(double noundef) [[READNONE]]
-// NO__ERRNO: declare float @tanhf(float noundef) [[READNONE]]
-// NO__ERRNO: declare x86_fp80 @tanhl(x86_fp80 noundef) [[READNONE]]
-// NO__ERRNO: declare fp128 @tanhf128(fp128 noundef) [[READNONE]]
+// NO__ERRNO: declare double @llvm.tanh.f64(double) [[READNONE_INTRINSIC]]
+// NO__ERRNO: declare float @llvm.tanh.f32(float) [[READNONE_INTRINSIC]]
+// NO__ERRNO: declare x86_fp80 @llvm.tanh.f80(x86_fp80) [[READNONE_INTRINSIC]]
+// NO__ERRNO: declare fp128 @llvm.tanh.f128(fp128) [[READNONE_INTRINSIC]]
 // HAS_ERRNO: declare double @tanh(double noundef) [[NOT_READNONE]]
 // HAS_ERRNO: declare float @tanhf(float noundef) [[NOT_READNONE]]
 // HAS_ERRNO: declare x86_fp80 @tanhl(x86_fp80 noundef) [[NOT_READNONE]]
diff --git a/clang/test/CodeGen/constrained-math-builtins.c b/clang/test/CodeGen/constrained-math-builtins.c
index 42c9e3c5008a3..aa77620b44535 100644
--- a/clang/test/CodeGen/constrained-math-builtins.c
+++ b/clang/test/CodeGen/constrained-math-builtins.c
@@ -36,6 +36,27 @@ void foo(double *d, float f, float *fp, long double *l, int *i, const char *c, _
 // CHECK: call float @llvm.experimental.constrained.ldexp.f32.i32(float %{{.*}}, i32 %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
 // CHECK: call x86_fp80 @llvm.experimental.constrained.ldexp.f80.i32(x86_fp80 %{{.*}}, i32 %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
 
+  __builtin_acos(f);        __builtin_acosf(f);       __builtin_acosl(f); __builtin_acosf128(f);
+
+// CHECK: call double @llvm.experimental.constrained.acos.f64(double %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+// CHECK: call float @llvm.experimental.constrained.acos.f32(float %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+// CHECK: call x86_fp80 @llvm.experimental.constrained.acos.f80(x86_fp80 %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+// CHECK: call fp128 @llvm.experimental.constrained.acos.f128(fp128 %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+
+__builtin_asin(f);        __builtin_asinf(f);       __builtin_asinl(f); __builtin_asinf128(f);
+
+// CHECK: call double @llvm.experimental.constrained.asin.f64(double %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+// CHECK: call float @llvm.experimental.constrained.asin.f32(float %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+// CHECK: call x86_fp80 @llvm.experimental.constrained.asin.f80(x86_fp80 %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+// CHECK: call fp128 @llvm.experimental.constrained.asin.f128(fp128 %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+
+__builtin_atan(f);        __builtin_atanf(f);       __builtin_atanl(f); __builtin_atanf128(f);
+
+// CHECK: call double @llvm.experimental.constrained.atan.f64(double %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+// CHECK: call float @llvm.experimental.constrained.atan.f32(float %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+// CHECK: call x86_fp80 @llvm.experimental.constrained.atan.f80(x86_fp80 %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+// CHECK: call fp128 @llvm.experimental.constrained.atan.f128(fp128 %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+
   __builtin_ceil(f);       __builtin_ceilf(f);      __builtin_ceill(f); __builtin_ceilf128(f);
 
 // CHECK: call double @llvm.experimental.constrained.ceil.f64(double %{{.*}}, metadata !"fpexcept.strict")
@@ -50,6 +71,13 @@ void foo(double *d, float f, float *fp, long double *l, int *i, const char *c, _
 // CHECK: call x86_fp80 @llvm.experimental.constrained.cos.f80(x86_fp80 %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
 // CHECK: call fp128 @llvm.experimental.constrained.cos.f128(fp128 %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
 
+  __builtin_cosh(f);        __builtin_coshf(f);       __builtin_coshl(f); __builtin_coshf128(f);
+
+// CHECK: call double @llvm.experimental.constrained.cosh.f64(double %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+// CHECK: call float @llvm.experimental.constrained.cosh.f32(float %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+// CHECK: call x86_fp80 @llvm.experimental.constrained.cosh.f80(x86_fp80 %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+// CHECK: call fp128 @llvm.experimental.constrained.cosh.f128(fp128 %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+
   __builtin_exp(f);        __builtin_expf(f);       __builtin_expl(f); __builtin_expf128(f);
 
 // CHECK: call double @llvm.experimental.constrained.exp.f64(double %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
@@ -177,6 +205,13 @@ void foo(double *d, float f, float *fp, long double *l, int *i, const char *c, _
 // CHECK: call x86_fp80 @llvm.experimental.constrained.sin.f80(x86_fp80 %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
 // CHECK: call fp128 @llvm.experimental.constrained.sin.f128(fp128 %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
 
+  __builtin_sinh(f);        __builtin_sinhf(f);       __builtin_sinhl(f); __builtin_sinhf128(f);
+
+// CHECK: call double @llvm.experimental.constrained.sinh.f64(double %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+// CHECK: call float @llvm.experimental.constrained.sinh.f32(float %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+// CHECK: call x86_fp80 @llvm.experimental.constrained.sinh.f80(x86_fp80 %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+// CHECK: call fp128 @llvm.experimental.constrained.sinh.f128(fp128 %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+
   __builtin_sqrt(f);       __builtin_sqrtf(f);      __builtin_sqrtl(f); __builtin_sqrtf128(f);
 
 // CHECK: call double @llvm.experimental.constrained.sqrt.f64(double %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
@@ -191,6 +226,12 @@ void foo(double *d, float f, float *fp, long double *l, int *i, const char *c, _
 // CHECK: call x86_fp80 @llvm.experimental.constrained.tan.f80(x86_fp80 %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
 // CHECK: call fp128 @llvm.experimental.constrained.tan.f128(fp128 %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
 
+  __builtin_tanh(f);        __builtin_tanhf(f);       __builtin_tanhl(f); __builtin_tanhf128(f);
+
+// CHECK: call double @llvm.experimental.constrained.tanh.f64(double %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+// CHECK: call float @llvm.experimental.constrained.tanh.f32(float %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+// CHECK: call x86_fp80 @llvm.experimental.constrained.tanh.f80(x86_fp80 %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+// CHECK: call fp128 @llvm.experimental.constrained.tanh.f128(fp128 %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
 
   __builtin_trunc(f);      __builtin_truncf(f);     __builtin_truncl(f); __builtin_truncf128(f);
 
diff --git a/clang/test/CodeGen/math-libcalls.c b/clang/test/CodeGen/math-libcalls.c
index a249182692762..5b23a4a3faef3 100644
--- a/clang/test/CodeGen/math-libcalls.c
+++ b/clang/test/CodeGen/math-libcalls.c
@@ -121,15 +121,15 @@ void foo(double *d, float f, float *fp, long double *l, int *i, const char *c) {
   /* math */
   acos(f);       acosf(f);      acosl(f);
 
-// NO__ERRNO: declare double @acos(double noundef) [[READNONE]]
-// NO__ERRNO: declare float @acosf(float noundef) [[READNONE]]
-// NO__ERRNO: declare x86_fp80 @acosl(x86_fp80 noundef) [[READNONE]]
+// NO__ERRNO: declare double @llvm.acos.f64(double) [[READNONE_INTRINSIC]]
+// NO__ERRNO: declare float @llvm.acos.f32(float) [[READNONE_INTRINSIC]...
[truncated]

efriedma-quic

LGTM

…builtins (#98949) Summary: ## Change(s) - `Builtins.td` - Add f16 support for libm arc and hyperbolic trig functions - `CGBuiltin.cpp` - Emit constraint intrinsics for trig clang builtins ## History This change is part of an implementation of #87367 investigation on supporting IEEE math operations as intrinsics. Which was discussed in this RFC: https://discourse.llvm.org/t/rfc-all-the-math-intrinsics/78294 This change adds wasm lowering cases for `acos`, `asin`, `atan`, `cosh`, `sinh`, and `tanh`. #70079 #70080 #70081 #70083 #70084 #95966 ## Precursor PR(s) Note this PR needs Merge after: - #98937 - #98755 Test Plan: Reviewers: Subscribers: Tasks: Tags: Differential Revision: https://phabricator.intern.facebook.com/D60251574

alanzhao1 · 2024-07-30T22:57:26Z

We (Chrome) are seeing linker failures on our x86 Windows builds due to this PR: https://crbug.com/355455678

[12702/86218] LINK glmark2_wgl.exe glmark2_wgl.exe.pdb
..\..\third_party\llvm-build\Release+Asserts\bin\lld-link.exe /OUT:./glmark2_wgl.exe /nologo -libpath:..\..\third_party\llvm-build\Release+Asserts\lib\clang\20\lib\windows /winsysroot:../../third_party/depot_tools/win_toolchain/vs_files/7393122652 /MACHINE:X86 /PDB:./glmark2_wgl.exe.pdb @./glmark2_wgl.exe.rsp
lld-link: error: undefined symbol: _asinf
>>> referenced by .\..\..\third_party\angle\third_party\glmark2\src\src\model.cpp:266
>>>               obj/third_party/angle/third_party/glmark2/glmark2_common_gl/model.obj:(public: void __thiscall Model::calculate_texcoords(void))
>>> referenced by .\..\..\third_party\angle\third_party\glmark2\src\src\model.cpp:267
>>>               obj/third_party/angle/third_party/glmark2/glmark2_common_gl/model.obj:(public: void __thiscall Model::calculate_texcoords(void))

lld-link: error: undefined symbol: _atanf
>>> referenced by .\..\..\third_party\angle\third_party\glmark2\src\src\scene-build.cpp:179
>>>               obj/third_party/angle/third_party/glmark2/glmark2_common_gl/scene-build.obj:(public: virtual bool __thiscall SceneBuild::setup(void))
>>> referenced by .\..\..\third_party\angle\third_party\glmark2\src\src\scene-refract.cpp:382
>>>               obj/third_party/angle/third_party/glmark2/glmark2_common_gl/scene-refract.obj:(public: bool __thiscall RefractPrivate::setup(class std::__Cr::map<class std::__Cr::basic_string<char, struct std::__Cr::char_traits<char>, class std::__Cr::allocator<char>>, struct Scene::Option, struct std::__Cr::less<class std::__Cr::basic_string<char, struct std::__Cr::char_traits<char>, class std::__Cr::allocator<char>>>, class std::__Cr::allocator<struct std::__Cr::pair<class std::__Cr::basic_string<char, struct std::__Cr::char_traits<char>, class std::__Cr::allocator<char>> const, struct Scene::Option>>> &))
>>> referenced by .\..\..\third_party\angle\third_party\glmark2\src\src\scene-shading.cpp:250
>>>               obj/third_party/angle/third_party/glmark2/glmark2_common_gl/scene-shading.obj:(public: virtual bool __thiscall SceneShading::setup(void))
>>> referenced 2 more times
exit=1
error: 1 steps failed

efriedma-quic · 2024-07-30T23:04:51Z

Right, asinf doesn't exist on Windows (see llvm/lib/Analysis/TargetLibraryInfo.cpp). We probably need code to promote calls to llvm.asin.f32 to 64-bit. Not sure how hard it would be to revert this... probably easier to fix forward if we can do that quickly.

We'll need a fix for the regression for the 19 branch.

farzonl · 2024-07-30T23:26:18Z

Right, asinf doesn't exist on Windows (see llvm/lib/Analysis/TargetLibraryInfo.cpp). We probably need code to promote calls to llvm.asin.f32 to 64-bit. Not sure how hard it would be to revert this... probably easier to fix forward if we can do that quickly.

We'll need a fix for the regression for the 19 branch.

I see that promoting seems like it could work, but it is also disabled for f80\f128:
https://github.com/llvm/llvm-project/blob/39e192b379362e9e645427631c35450d55ed517d/llvm/lib/Analysis/TargetLibraryInfo.cpp#L322C5-L324C39

How should we handle that case?

farzonl · 2024-07-30T23:39:02Z

We (Chrome) are seeing linker failures on our x86 Windows builds due to this PR: https://crbug.com/355455678

[12702/86218] LINK glmark2_wgl.exe glmark2_wgl.exe.pdb
..\..\third_party\llvm-build\Release+Asserts\bin\lld-link.exe /OUT:./glmark2_wgl.exe /nologo -libpath:..\..\third_party\llvm-build\Release+Asserts\lib\clang\20\lib\windows /winsysroot:../../third_party/depot_tools/win_toolchain/vs_files/7393122652 /MACHINE:X86 /PDB:./glmark2_wgl.exe.pdb @./glmark2_wgl.exe.rsp
lld-link: error: undefined symbol: _asinf
>>> referenced by .\..\..\third_party\angle\third_party\glmark2\src\src\model.cpp:266
>>>               obj/third_party/angle/third_party/glmark2/glmark2_common_gl/model.obj:(public: void __thiscall Model::calculate_texcoords(void))
>>> referenced by .\..\..\third_party\angle\third_party\glmark2\src\src\model.cpp:267
>>>               obj/third_party/angle/third_party/glmark2/glmark2_common_gl/model.obj:(public: void __thiscall Model::calculate_texcoords(void))

lld-link: error: undefined symbol: _atanf
>>> referenced by .\..\..\third_party\angle\third_party\glmark2\src\src\scene-build.cpp:179
>>>               obj/third_party/angle/third_party/glmark2/glmark2_common_gl/scene-build.obj:(public: virtual bool __thiscall SceneBuild::setup(void))
>>> referenced by .\..\..\third_party\angle\third_party\glmark2\src\src\scene-refract.cpp:382
>>>               obj/third_party/angle/third_party/glmark2/glmark2_common_gl/scene-refract.obj:(public: bool __thiscall RefractPrivate::setup(class std::__Cr::map<class std::__Cr::basic_string<char, struct std::__Cr::char_traits<char>, class std::__Cr::allocator<char>>, struct Scene::Option, struct std::__Cr::less<class std::__Cr::basic_string<char, struct std::__Cr::char_traits<char>, class std::__Cr::allocator<char>>>, class std::__Cr::allocator<struct std::__Cr::pair<class std::__Cr::basic_string<char, struct std::__Cr::char_traits<char>, class std::__Cr::allocator<char>> const, struct Scene::Option>>> &))
>>> referenced by .\..\..\third_party\angle\third_party\glmark2\src\src\scene-shading.cpp:250
>>>               obj/third_party/angle/third_party/glmark2/glmark2_common_gl/scene-shading.obj:(public: virtual bool __thiscall SceneShading::setup(void))
>>> referenced 2 more times
exit=1
error: 1 steps failed

I'm preparing a fix, building it now should be up in the next 20-30 minutes. Thanks for reporting this issue!

Windows does not support float C89 math functions like: - acosf - asinf - atanf - coshf - sinhf - tanhf These 6 libfuncs need to be type promoted. This PR fixes the bug introduced by llvm#98949

Windows does not support float C89 math functions like: - acosf - asinf - atanf - coshf - sinhf - tanhf All 6 are disabled in [TargetLibraryInfo.cpp:293](https://github.com/llvm/llvm-project/blob/39e192b379362e9e645427631c35450d55ed517d/llvm/lib/Analysis/TargetLibraryInfo.cpp#L293C4-L296C41) These 6 libfuncs need to be type promoted. This PR fixes the bug introduced by #98949

farzonl · 2024-07-31T00:49:11Z

@alanzhao1 @efriedma-quic It should be fixed now by #101268 if you role the clang build to latest.

alanzhao1 · 2024-07-31T01:59:46Z

@alanzhao1 @efriedma-quic It should be fixed now by #101268 if you role the clang build to latest.

Confirmed. Thanks!

This change due to llvm/llvm-project#98949.

Windows does not support float C89 math functions like: - acosf - asinf - atanf - coshf - sinhf - tanhf These 6 libfuncs need to be type promoted. This PR fixes the bug introduced by llvm#98949

Windows does not support float C89 math functions like: - acosf - asinf - atanf - coshf - sinhf - tanhf These 6 libfuncs need to be type promoted. This PR fixes the bug introduced by llvm/llvm-project#98949

Windows does not support float C89 math functions like: - acosf - asinf - atanf - coshf - sinhf - tanhf All 6 are disabled in [TargetLibraryInfo.cpp:293](https://github.com/llvm/llvm-project/blob/43abf7b77f2a91a1b151ab6db518a0c6f830be20/llvm/lib/Analysis/TargetLibraryInfo.cpp#L293C4-L296C41) These 6 libfuncs need to be type promoted. This PR fixes the bug introduced by llvm/llvm-project#98949

Windows does not support float C89 math functions like: - acosf - asinf - atanf - coshf - sinhf - tanhf These 6 libfuncs need to be type promoted. This PR fixes the bug introduced by llvm/llvm-project#98949

farzonl requested review from davemgreen and efriedma-quic July 15, 2024 19:48

llvmbot added clang Clang issues not falling into any other category backend:X86 clang:frontend Language frontend issues, e.g. anything involving "Sema" clang:codegen IR generation bugs: mangling, exceptions, etc. labels Jul 15, 2024

fix test failure

db0ee81

efriedma-quic approved these changes Jul 15, 2024

View reviewed changes

farzonl self-assigned this Jul 16, 2024

farzonl merged commit a14baec into llvm:main Jul 19, 2024

farzonl mentioned this pull request Jul 31, 2024

[x86][Windows] Fix chromium build break #101268

Merged

farzonl deleted the arc-hyp-trig-constraint-intrinsics branch July 31, 2024 00:47

zahiraam mentioned this pull request Aug 12, 2024

Add missing fpbuiltin math functions. intel/llvm#15039

Merged

jsji pushed a commit to intel/llvm that referenced this pull request Aug 12, 2024

Add missing fpbuiltin math functions. (#15039)

746f3d9

This change due to llvm/llvm-project#98949.

jsji pushed a commit to intel/llvm that referenced this pull request Aug 16, 2024

Add missing fpbuiltin math functions. (#15039)

22f9f89

This change due to llvm/llvm-project#98949.

farzonl mentioned this pull request Oct 4, 2024

[x86][Windows] Fix chromium build break #111218

Merged

Navigation Menu

Uh oh!

[clang] Emit constraint intrinsics for arc and hyperbolic trig clang builtins - #98949

[clang] Emit constraint intrinsics for arc and hyperbolic trig clang builtins#98949
farzonl merged 2 commits into
llvm:mainfrom
farzonl:arc-hyp-trig-constraint-intrinsics

farzonl commented Jul 15, 2024

Uh oh!

llvmbot commented Jul 15, 2024 •

edited

Loading

Change(s)

History

Precursor PR(s)

Uh oh!

efriedma-quic left a comment

Uh oh!

alanzhao1 commented Jul 30, 2024

Uh oh!

efriedma-quic commented Jul 30, 2024

Uh oh!

farzonl commented Jul 30, 2024

Uh oh!

farzonl commented Jul 30, 2024

Uh oh!

farzonl commented Jul 31, 2024

Uh oh!

alanzhao1 commented Jul 31, 2024

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

Uh oh!

Conversation

farzonl commented Jul 15, 2024

Change(s)

History

Precursor PR(s)

Uh oh!

llvmbot commented Jul 15, 2024 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Change(s)

History

Precursor PR(s)

Uh oh!

efriedma-quic left a comment

Choose a reason for hiding this comment

Uh oh!

alanzhao1 commented Jul 30, 2024

Uh oh!

efriedma-quic commented Jul 30, 2024

Uh oh!

farzonl commented Jul 30, 2024

Uh oh!

farzonl commented Jul 30, 2024

Uh oh!

farzonl commented Jul 31, 2024

Uh oh!

alanzhao1 commented Jul 31, 2024

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

llvmbot commented Jul 15, 2024 •

edited

Loading