[clang] constexpr built-in elementwise add_sat/sub_sat functions. (#119082)

Part of #51787.

This patch adds constexpr support for the built-in elementwise add_sat
and sub_sat functions.

Author: c8ef

clang/docs/LanguageExtensions.rst

@@ -648,7 +648,8 @@ elementwise to the input.
 Unless specified otherwise operation(±0) = ±0 and operation(±infinity) = ±infinity
 
 The integer elementwise intrinsics, including ``__builtin_elementwise_popcount``,
-``__builtin_elementwise_bitreverse``, can be called in a ``constexpr`` context.
+``__builtin_elementwise_bitreverse``, ``__builtin_elementwise_add_sat``,
+``__builtin_elementwise_sub_sat`` can be called in a ``constexpr`` context.
 
 ============================================== ====================================================================== =========================================
          Name                                   Operation                                                             Supported element types

clang/docs/ReleaseNotes.rst

@@ -408,12 +408,11 @@ Non-comprehensive list of changes in this release
   The flexible array member (FAM) can now be accessed immediately without causing
   issues with the sanitizer because the counter is automatically set.
 
-- ``__builtin_reduce_add`` function can now be used in constant expressions.
-- ``__builtin_reduce_mul`` function can now be used in constant expressions.
-- ``__builtin_reduce_and`` function can now be used in constant expressions.
-- ``__builtin_reduce_or`` and ``__builtin_reduce_xor`` functions can now be used in constant expressions.
-- ``__builtin_elementwise_popcount`` function can now be used in constant expressions.
-- ``__builtin_elementwise_bitreverse`` function can now be used in constant expressions.
+- The following builtins can now be used in constant expressions: ``__builtin_reduce_add``,
+  ``__builtin_reduce_mul``, ``__builtin_reduce_and``, ``__builtin_reduce_or``,
+  ``__builtin_reduce_xor``, ``__builtin_elementwise_popcount``,
+  ``__builtin_elementwise_bitreverse``, ``__builtin_elementwise_add_sat``,
+  ``__builtin_elementwise_sub_sat``.
 
 New Compiler Flags
 ------------------

clang/include/clang/Basic/Builtins.td

@@ -1450,13 +1450,13 @@ def ElementwiseFma : Builtin {
 
 def ElementwiseAddSat : Builtin {
   let Spellings = ["__builtin_elementwise_add_sat"];
-  let Attributes = [NoThrow, Const, CustomTypeChecking];
+  let Attributes = [NoThrow, Const, CustomTypeChecking, Constexpr];
   let Prototype = "void(...)";
 }
 
 def ElementwiseSubSat : Builtin {
   let Spellings = ["__builtin_elementwise_sub_sat"];
-  let Attributes = [NoThrow, Const, CustomTypeChecking];
+  let Attributes = [NoThrow, Const, CustomTypeChecking, Constexpr];
   let Prototype = "void(...)";
 }
 

clang/lib/AST/ExprConstant.cpp

@@ -11339,6 +11339,37 @@ bool VectorExprEvaluator::VisitCallExpr(const CallExpr *E) {
 
     return Success(APValue(ResultElements.data(), ResultElements.size()), E);
   }
+  case Builtin::BI__builtin_elementwise_add_sat:
+  case Builtin::BI__builtin_elementwise_sub_sat: {
+    APValue SourceLHS, SourceRHS;
+    if (!EvaluateAsRValue(Info, E->getArg(0), SourceLHS) ||
+        !EvaluateAsRValue(Info, E->getArg(1), SourceRHS))
+      return false;
+
+    QualType DestEltTy = E->getType()->castAs<VectorType>()->getElementType();
+    unsigned SourceLen = SourceLHS.getVectorLength();
+    SmallVector<APValue, 4> ResultElements;
+    ResultElements.reserve(SourceLen);
+
+    for (unsigned EltNum = 0; EltNum < SourceLen; ++EltNum) {
+      APSInt LHS = SourceLHS.getVectorElt(EltNum).getInt();
+      APSInt RHS = SourceRHS.getVectorElt(EltNum).getInt();
+      switch (E->getBuiltinCallee()) {
+      case Builtin::BI__builtin_elementwise_add_sat:
+        ResultElements.push_back(APValue(
+            APSInt(LHS.isSigned() ? LHS.sadd_sat(RHS) : RHS.uadd_sat(RHS),
+                   DestEltTy->isUnsignedIntegerOrEnumerationType())));
+        break;
+      case Builtin::BI__builtin_elementwise_sub_sat:
+        ResultElements.push_back(APValue(
+            APSInt(LHS.isSigned() ? LHS.ssub_sat(RHS) : RHS.usub_sat(RHS),
+                   DestEltTy->isUnsignedIntegerOrEnumerationType())));
+        break;
+      }
+    }
+
+    return Success(APValue(ResultElements.data(), ResultElements.size()), E);
+  }
   }
 }
 
@@ -13204,6 +13235,25 @@ bool IntExprEvaluator::VisitBuiltinCallExpr(const CallExpr *E,
     return Success(Val.rotr(Amt.urem(Val.getBitWidth())), E);
   }
 
+  case Builtin::BI__builtin_elementwise_add_sat: {
+    APSInt LHS, RHS;
+    if (!EvaluateInteger(E->getArg(0), LHS, Info) ||
+        !EvaluateInteger(E->getArg(1), RHS, Info))
+      return false;
+
+    APInt Result = LHS.isSigned() ? LHS.sadd_sat(RHS) : LHS.uadd_sat(RHS);
+    return Success(APSInt(Result, !LHS.isSigned()), E);
+  }
+  case Builtin::BI__builtin_elementwise_sub_sat: {
+    APSInt LHS, RHS;
+    if (!EvaluateInteger(E->getArg(0), LHS, Info) ||
+        !EvaluateInteger(E->getArg(1), RHS, Info))
+      return false;
+
+    APInt Result = LHS.isSigned() ? LHS.ssub_sat(RHS) : LHS.usub_sat(RHS);
+    return Success(APSInt(Result, !LHS.isSigned()), E);
+  }
+
   case Builtin::BIstrlen:
   case Builtin::BIwcslen:
     // A call to strlen is not a constant expression.

clang/test/CodeGen/builtins-elementwise-math.c

@@ -112,7 +112,7 @@ void test_builtin_elementwise_add_sat(float f1, float f2, double d1, double d2,
   // CHECK-NEXT: call i32 @llvm.sadd.sat.i32(i32 [[IAS1]], i32 [[B]])
   int_as_one = __builtin_elementwise_add_sat(int_as_one, b);
 
-  // CHECK: call i32 @llvm.sadd.sat.i32(i32 1, i32 97)
+  // CHECK: store i64 98, ptr %i1.addr, align 8
   i1 = __builtin_elementwise_add_sat(1, 'a');
 }
 
@@ -165,7 +165,7 @@ void test_builtin_elementwise_sub_sat(float f1, float f2, double d1, double d2,
   // CHECK-NEXT: call i32 @llvm.ssub.sat.i32(i32 [[IAS1]], i32 [[B]])
   int_as_one = __builtin_elementwise_sub_sat(int_as_one, b);
 
-  // CHECK: call i32 @llvm.ssub.sat.i32(i32 1, i32 97)
+  // CHECK: store i64 -96, ptr %i1.addr, align 8
   i1 = __builtin_elementwise_sub_sat(1, 'a');
 }
 

clang/test/Sema/constant_builtins_vector.cpp

@@ -822,3 +822,19 @@ static_assert(__builtin_elementwise_bitreverse(0x12345678) == 0x1E6A2C48);
 static_assert(__builtin_elementwise_bitreverse(0x0123456789ABCDEFULL) == 0xF7B3D591E6A2C480);
 static_assert(__builtin_bit_cast(unsigned, __builtin_elementwise_bitreverse((vector4char){1, 2, 4, 8})) == (LITTLE_END ? 0x10204080 : 0x80402010));
 static_assert(__builtin_bit_cast(unsigned long long, __builtin_elementwise_bitreverse((vector4short){1, 2, 4, 8})) == (LITTLE_END ? 0x1000200040008000 : 0x8000400020001000));
+
+static_assert(__builtin_elementwise_add_sat(1, 2) == 3);
+static_assert(__builtin_elementwise_add_sat(1U, 2U) == 3U);
+static_assert(__builtin_elementwise_add_sat(~(1 << 31), 42) == ~(1 << 31));
+static_assert(__builtin_elementwise_add_sat((1 << 31), -42) == (1 << 31));
+static_assert(__builtin_elementwise_add_sat(~0U, 1U) == ~0U);
+static_assert(__builtin_bit_cast(unsigned, __builtin_elementwise_add_sat((vector4char){1, 2, 3, 4}, (vector4char){1, 2, 3, 4})) == (LITTLE_END ? 0x08060402 : 0x02040608));
+static_assert(__builtin_bit_cast(unsigned long long, __builtin_elementwise_add_sat((vector4short){(short)0x8000, (short)0x8001, (short)0x8002, (short)0x8003}, (vector4short){-7, -8, -9, -10}) == (LITTLE_END ? 0x8000800080008000 : 0x8000800080008000)));
+
+static_assert(__builtin_elementwise_sub_sat(1, 2) == -1);
+static_assert(__builtin_elementwise_sub_sat(2U, 1U) == 1U);
+static_assert(__builtin_elementwise_sub_sat(~(1 << 31), -42) == ~(1 << 31));
+static_assert(__builtin_elementwise_sub_sat((1 << 31), 42) == (1 << 31));
+static_assert(__builtin_elementwise_sub_sat(0U, 1U) == 0U);
+static_assert(__builtin_bit_cast(unsigned, __builtin_elementwise_sub_sat((vector4char){5, 4, 3, 2}, (vector4char){1, 1, 1, 1})) == (LITTLE_END ? 0x01020304 : 0x04030201));
+static_assert(__builtin_bit_cast(unsigned long long, __builtin_elementwise_sub_sat((vector4short){(short)0x8000, (short)0x8001, (short)0x8002, (short)0x8003}, (vector4short){7, 8, 9, 10}) == (LITTLE_END ? 0x8000800080008000 : 0x8000800080008000)));