[clang][x86] Add constexpr support for PDEP/PEXT intrinsics (llvm#110535)

RKSimon · Sterling-Augustine · commit aee59be1c8b1 · 2024-10-02T18:13:35.000-07:00
diff --git a/clang/docs/ReleaseNotes.rst b/clang/docs/ReleaseNotes.rst
@@ -523,6 +523,8 @@ X86 Support
   all bextr intrinsics in bmiintrin.h
   all tzcnt intrinsics in bmiintrin.h
   all bzhi intrinsics in bmi2intrin.h
+  all pdep intrinsics in bmi2intrin.h
+  all pext intrinsics in bmi2intrin.h
   all intrinsics in tbmintrin.h
 
 Arm and AArch64 Support
diff --git a/clang/include/clang/Basic/BuiltinsX86.def b/clang/include/clang/Basic/BuiltinsX86.def
@@ -561,8 +561,8 @@ TARGET_BUILTIN(__builtin_ia32_tzcnt_u32, "UiUi", "ncE", "")
 
 // BMI2
 TARGET_BUILTIN(__builtin_ia32_bzhi_si, "UiUiUi", "ncE", "bmi2")
-TARGET_BUILTIN(__builtin_ia32_pdep_si, "UiUiUi", "nc", "bmi2")
-TARGET_BUILTIN(__builtin_ia32_pext_si, "UiUiUi", "nc", "bmi2")
+TARGET_BUILTIN(__builtin_ia32_pdep_si, "UiUiUi", "ncE", "bmi2")
+TARGET_BUILTIN(__builtin_ia32_pext_si, "UiUiUi", "ncE", "bmi2")
 
 // TBM
 TARGET_BUILTIN(__builtin_ia32_bextri_u32, "UiUiIUi", "ncE", "tbm")
diff --git a/clang/include/clang/Basic/BuiltinsX86_64.def b/clang/include/clang/Basic/BuiltinsX86_64.def
@@ -74,8 +74,8 @@ TARGET_BUILTIN(__builtin_ia32_lzcnt_u64, "UOiUOi", "ncE", "lzcnt")
 TARGET_BUILTIN(__builtin_ia32_bextr_u64, "UOiUOiUOi", "ncE", "bmi")
 TARGET_BUILTIN(__builtin_ia32_tzcnt_u64, "UOiUOi", "ncE", "")
 TARGET_BUILTIN(__builtin_ia32_bzhi_di, "UOiUOiUOi", "ncE", "bmi2")
-TARGET_BUILTIN(__builtin_ia32_pdep_di, "UOiUOiUOi", "nc", "bmi2")
-TARGET_BUILTIN(__builtin_ia32_pext_di, "UOiUOiUOi", "nc", "bmi2")
+TARGET_BUILTIN(__builtin_ia32_pdep_di, "UOiUOiUOi", "ncE", "bmi2")
+TARGET_BUILTIN(__builtin_ia32_pext_di, "UOiUOiUOi", "ncE", "bmi2")
 TARGET_BUILTIN(__builtin_ia32_bextri_u64, "UOiUOiIUOi", "ncE", "tbm")
 TARGET_BUILTIN(__builtin_ia32_lwpins64, "UcUOiUiIUi", "n", "lwp")
 TARGET_BUILTIN(__builtin_ia32_lwpval64, "vUOiUiIUi", "n", "lwp")
diff --git a/clang/lib/AST/ExprConstant.cpp b/clang/lib/AST/ExprConstant.cpp
@@ -13518,6 +13518,36 @@ bool IntExprEvaluator::VisitBuiltinCallExpr(const CallExpr *E,
       return false;
     return Success(Val.countTrailingZeros(), E);
   }
+
+  case clang::X86::BI__builtin_ia32_pdep_si:
+  case clang::X86::BI__builtin_ia32_pdep_di: {
+    APSInt Val, Msk;
+    if (!EvaluateInteger(E->getArg(0), Val, Info) ||
+        !EvaluateInteger(E->getArg(1), Msk, Info))
+      return false;
+
+    unsigned BitWidth = Val.getBitWidth();
+    APInt Result = APInt::getZero(BitWidth);
+    for (unsigned I = 0, P = 0; I != BitWidth; ++I)
+      if (Msk[I])
+        Result.setBitVal(I, Val[P++]);
+    return Success(Result, E);
+  }
+
+  case clang::X86::BI__builtin_ia32_pext_si:
+  case clang::X86::BI__builtin_ia32_pext_di: {
+    APSInt Val, Msk;
+    if (!EvaluateInteger(E->getArg(0), Val, Info) ||
+        !EvaluateInteger(E->getArg(1), Msk, Info))
+      return false;
+
+    unsigned BitWidth = Val.getBitWidth();
+    APInt Result = APInt::getZero(BitWidth);
+    for (unsigned I = 0, P = 0; I != BitWidth; ++I)
+      if (Msk[I])
+        Result.setBitVal(P++, Val[I]);
+    return Success(Result, E);
+  }
   }
 }
 
diff --git a/clang/lib/Headers/bmi2intrin.h b/clang/lib/Headers/bmi2intrin.h
@@ -73,7 +73,7 @@ _bzhi_u32(unsigned int __X, unsigned int __Y)
 /// \param __Y
 ///    The 32-bit mask specifying where to deposit source bits.
 /// \returns The 32-bit result.
-static __inline__ unsigned int __DEFAULT_FN_ATTRS
+static __inline__ unsigned int __DEFAULT_FN_ATTRS_CONSTEXPR
 _pdep_u32(unsigned int __X, unsigned int __Y)
 {
   return __builtin_ia32_pdep_si(__X, __Y);
@@ -103,7 +103,7 @@ _pdep_u32(unsigned int __X, unsigned int __Y)
 /// \param __Y
 ///    The 32-bit mask specifying which source bits to extract.
 /// \returns The 32-bit result.
-static __inline__ unsigned int __DEFAULT_FN_ATTRS
+static __inline__ unsigned int __DEFAULT_FN_ATTRS_CONSTEXPR
 _pext_u32(unsigned int __X, unsigned int __Y)
 {
   return __builtin_ia32_pext_si(__X, __Y);
@@ -189,7 +189,7 @@ _bzhi_u64(unsigned long long __X, unsigned long long __Y)
 /// \param __Y
 ///    The 64-bit mask specifying where to deposit source bits.
 /// \returns The 64-bit result.
-static __inline__ unsigned long long __DEFAULT_FN_ATTRS
+static __inline__ unsigned long long __DEFAULT_FN_ATTRS_CONSTEXPR
 _pdep_u64(unsigned long long __X, unsigned long long __Y)
 {
   return __builtin_ia32_pdep_di(__X, __Y);
@@ -219,7 +219,7 @@ _pdep_u64(unsigned long long __X, unsigned long long __Y)
 /// \param __Y
 ///    The 64-bit mask specifying which source bits to extract.
 /// \returns The 64-bit result.
-static __inline__ unsigned long long __DEFAULT_FN_ATTRS
+static __inline__ unsigned long long __DEFAULT_FN_ATTRS_CONSTEXPR
 _pext_u64(unsigned long long __X, unsigned long long __Y)
 {
   return __builtin_ia32_pext_di(__X, __Y);
diff --git a/clang/test/CodeGen/X86/bmi2-builtins.c b/clang/test/CodeGen/X86/bmi2-builtins.c
@@ -61,10 +61,30 @@ char bzhi32_3[_bzhi_u32(0x89ABCDEF,  32) == 0x89ABCDEF ? 1 : -1];
 char bzhi32_4[_bzhi_u32(0x89ABCDEF,  99) == 0x89ABCDEF ? 1 : -1];
 char bzhi32_5[_bzhi_u32(0x89ABCDEF, 260) == 0x0000000F ? 1 : -1];
 
+char pdep32_0[_pdep_u32(0x89ABCDEF,  0x00000000) == 0x00000000 ? 1 : -1];
+char pdep32_1[_pdep_u32(0x89ABCDEF,  0x000000F0) == 0x000000F0 ? 1 : -1];
+char pdep32_2[_pdep_u32(0x89ABCDEF,  0xF00000F0) == 0xE00000F0 ? 1 : -1];
+char pdep32_3[_pdep_u32(0x89ABCDEF,  0xFFFFFFFF) == 0x89ABCDEF ? 1 : -1];
+
+char pext32_0[_pext_u32(0x89ABCDEF,  0x00000000) == 0x00000000 ? 1 : -1];
+char pext32_1[_pext_u32(0x89ABCDEF,  0x000000F0) == 0x0000000E ? 1 : -1];
+char pext32_2[_pext_u32(0x89ABCDEF,  0xF00000F0) == 0x0000008E ? 1 : -1];
+char pext32_3[_pext_u32(0x89ABCDEF,  0xFFFFFFFF) == 0x89ABCDEF ? 1 : -1];
+
 #ifdef __x86_64__
 char bzhi64_0[_bzhi_u64(0x0123456789ABCDEFULL,   0) == 0x0000000000000000ULL ? 1 : -1];
 char bzhi64_1[_bzhi_u64(0x0123456789ABCDEFULL,  32) == 0x0000000089ABCDEFULL ? 1 : -1];
 char bzhi64_2[_bzhi_u64(0x0123456789ABCDEFULL,  99) == 0x0123456789ABCDEFULL ? 1 : -1];
 char bzhi64_3[_bzhi_u64(0x0123456789ABCDEFULL, 520) == 0x00000000000000EFULL ? 1 : -1];
+
+char pdep64_0[_pdep_u64(0x0123456789ABCDEFULL, 0x0000000000000000ULL) == 0x0000000000000000ULL ? 1 : -1];
+char pdep64_1[_pdep_u64(0x0123456789ABCDEFULL, 0x00000000000000F0ULL) == 0x00000000000000F0ULL ? 1 : -1];
+char pdep64_2[_pdep_u64(0x0123456789ABCDEFULL, 0xF00000F0F00000F0ULL) == 0xC00000D0E00000F0ULL ? 1 : -1];
+char pdep64_3[_pdep_u64(0x0123456789ABCDEFULL, 0xFFFFFFFFFFFFFFFFULL) == 0x0123456789ABCDEFULL ? 1 : -1];
+
+char pext64_0[_pext_u64(0x0123456789ABCDEFULL, 0x0000000000000000ULL) == 0x0000000000000000ULL ? 1 : -1];
+char pext64_1[_pext_u64(0x0123456789ABCDEFULL, 0x00000000000000F0ULL) == 0x000000000000000EULL ? 1 : -1];
+char pext64_2[_pext_u64(0x0123456789ABCDEFULL, 0xF00000F0F00000F0ULL) == 0x000000000000068EULL ? 1 : -1];
+char pext64_3[_pext_u64(0x0123456789ABCDEFULL, 0xFFFFFFFFFFFFFFFFULL) == 0x0123456789ABCDEFULL ? 1 : -1];
 #endif
 #endif
