[CIR] Upstream codegen for X86 SSE builtins

clang/include/clang/CIR/Dialect/IR/CIROps.td

@@ -413,6 +413,12 @@ def CIR_ConstantOp : CIR_Op<"const", [
 
     template <typename T>
     T getValueAttr() { return mlir::dyn_cast<T>(getValue()); }
+
+    llvm::APInt getIntValue() {
+      if (const auto intAttr = getValueAttr<cir::IntAttr>())
+        return intAttr.getValue();
+      llvm_unreachable("Expected an IntAttr in ConstantOp");
+    }
   }];
 
   let hasFolder = 1;

clang/lib/CIR/CodeGen/CIRGenBuiltin.cpp

@@ -625,6 +625,22 @@ CIRGenFunction::emitTargetBuiltinExpr(unsigned builtinID, const CallExpr *e,
                                    getTarget().getTriple().getArch());
 }
 
+mlir::Value CIRGenFunction::emitScalarOrConstFoldImmArg(
+    const unsigned iceArguments, const unsigned idx, const Expr *argExpr) {
+  mlir::Value arg = {};
+  if ((iceArguments & (1 << idx)) == 0) {
+    arg = emitScalarExpr(argExpr);
+  } else {
+    // If this is required to be a constant, constant fold it so that we
+    // know that the generated intrinsic gets a ConstantInt.
+    const std::optional<llvm::APSInt> result =
+        argExpr->getIntegerConstantExpr(getContext());
+    assert(result && "Expected argument to be a constant");
+    arg = builder.getConstInt(getLoc(argExpr->getSourceRange()), *result);
+  }
+  return arg;
+}
+
 /// Given a builtin id for a function like "__builtin_fabsf", return a Function*
 /// for "fabsf".
 cir::FuncOp CIRGenModule::getBuiltinLibFunction(const FunctionDecl *fd,

clang/lib/CIR/CodeGen/CIRGenBuiltinX86.cpp

@@ -16,7 +16,6 @@
 #include "clang/Basic/Builtins.h"
 #include "clang/Basic/TargetBuiltins.h"
 #include "clang/CIR/MissingFeatures.h"
-#include "llvm/IR/IntrinsicsX86.h"
 
 using namespace clang;
 using namespace clang::CIRGen;
@@ -43,6 +42,18 @@ mlir::Value CIRGenFunction::emitX86BuiltinExpr(unsigned builtinID,
   // Find out if any arguments are required to be integer constant expressions.
   assert(!cir::MissingFeatures::handleBuiltinICEArguments());
 
+  llvm::SmallVector<mlir::Value> ops;
+
+  // Find out if any arguments are required to be integer constant expressions.
+  unsigned iceArguments = 0;
+  ASTContext::GetBuiltinTypeError error;
+  getContext().GetBuiltinType(builtinID, error, &iceArguments);
+  assert(error == ASTContext::GE_None && "Should not codegen an error");
+
+  for (auto [idx, arg] : llvm::enumerate(e->arguments())) {
+    ops.push_back(emitScalarOrConstFoldImmArg(iceArguments, idx, arg));
+  }
+
   switch (builtinID) {
   default:
     return {};
@@ -63,6 +74,10 @@ mlir::Value CIRGenFunction::emitX86BuiltinExpr(unsigned builtinID,
   case X86::BI__builtin_ia32_undef128:
   case X86::BI__builtin_ia32_undef256:
   case X86::BI__builtin_ia32_undef512:
+    cgm.errorNYI(e->getSourceRange(),
+                 std::string("unimplemented X86 builtin call: ") +
+                     getContext().BuiltinInfo.getName(builtinID));
+    return {};
   case X86::BI__builtin_ia32_vec_ext_v4hi:
   case X86::BI__builtin_ia32_vec_ext_v16qi:
   case X86::BI__builtin_ia32_vec_ext_v8hi:
@@ -72,7 +87,24 @@ mlir::Value CIRGenFunction::emitX86BuiltinExpr(unsigned builtinID,
   case X86::BI__builtin_ia32_vec_ext_v32qi:
   case X86::BI__builtin_ia32_vec_ext_v16hi:
   case X86::BI__builtin_ia32_vec_ext_v8si:
-  case X86::BI__builtin_ia32_vec_ext_v4di:
+  case X86::BI__builtin_ia32_vec_ext_v4di: {
+    unsigned NumElts = cast<cir::VectorType>(ops[0].getType()).getSize();
+
+    uint64_t index =
+        ops[1].getDefiningOp<cir::ConstantOp>().getIntValue().getZExtValue();
+
+    index &= NumElts - 1;
+
+    auto indexAttr = cir::IntAttr::get(
+        cir::IntType::get(&getMLIRContext(), 64, false), index);
+    auto indexVal =
+        cir::ConstantOp::create(builder, getLoc(e->getExprLoc()), indexAttr);
+
+    // These builtins exist so we can ensure the index is an ICE and in range.
+    // Otherwise we could just do this in the header file.
+    return cir::VecExtractOp::create(builder, getLoc(e->getExprLoc()), ops[0],
+                                     indexVal);
+  }
   case X86::BI__builtin_ia32_vec_set_v4hi:
   case X86::BI__builtin_ia32_vec_set_v16qi:
   case X86::BI__builtin_ia32_vec_set_v8hi:

clang/lib/CIR/CodeGen/CIRGenFunction.h

@@ -1699,6 +1699,9 @@ class CIRGenFunction : public CIRGenTypeCache {
   void emitScalarInit(const clang::Expr *init, mlir::Location loc,
                       LValue lvalue, bool capturedByInit = false);
 
+  mlir::Value emitScalarOrConstFoldImmArg(unsigned iceArguments, unsigned idx,
+                                          const Expr *argExpr);
+
   void emitStaticVarDecl(const VarDecl &d, cir::GlobalLinkageKind linkage);
 
   void emitStoreOfComplex(mlir::Location loc, mlir::Value v, LValue dest,

clang/test/CIR/CodeGen/X86/sse2-builtins.c

@@ -0,0 +1,27 @@
+// RUN: %clang_cc1 -x c -flax-vector-conversions=none -ffreestanding %s -triple=x86_64-unknown-linux -target-feature +sse2 -fclangir -emit-cir -o %t.cir -Wall -Werror
+// RUN: FileCheck --check-prefixes=CIR-CHECK --input-file=%t.cir %s
+// RUN: %clang_cc1 -x c -flax-vector-conversions=none -ffreestanding %s -triple=x86_64-unknown-linux -target-feature +sse2 -fno-signed-char -fclangir -emit-cir -o %t.cir -Wall -Werror
+// RUN: FileCheck --check-prefixes=CIR-CHECK --input-file=%t.cir %s
+
+// RUN: %clang_cc1 -x c++ -flax-vector-conversions=none -ffreestanding %s -triple=x86_64-unknown-linux -target-feature +sse2 -fclangir -emit-llvm -o %t.ll -Wall -Werror
+// RUN: FileCheck --check-prefixes=LLVM-CHECK --input-file=%t.ll %s
+// RUN: %clang_cc1 -x c++ -flax-vector-conversions=none -ffreestanding %s -triple=x86_64-unknown-linux -target-feature +sse2 -fno-signed-char -fclangir -emit-llvm -o %t.ll -Wall -Werror
+// RUN: FileCheck --check-prefixes=LLVM-CHECK --input-file=%t.ll %s
+
+// This test mimics clang/test/CodeGen/X86/sse2-builtins.c, which eventually
+// CIR shall be able to support fully.
+
+#include <immintrin.h>
+
+// Lowering to pextrw requires optimization.
+int test_mm_extract_epi16(__m128i A) {
+
+  // CIR-CHECK-LABEL: test_mm_extract_epi16
+  // CIR-CHECK %{{.*}} = cir.vec.extract %{{.*}}[%{{.*}} : {{!u32i|!u64i}}] : !cir.vector<!s16i x 8>
+  // CIR-CHECK %{{.*}} = cir.cast integral %{{.*}} : !u16i -> !s32i
+
+  // LLVM-CHECK-LABEL: test_mm_extract_epi16
+  // LLVM-CHECK: extractelement <8 x i16> %{{.*}}, {{i32|i64}} 1
+  // LLVM-CHECK: zext i16 %{{.*}} to i32
+  return _mm_extract_epi16(A, 1);
+}