[CIR] Upstream CIR codegen for vec_ext x86 builtins

clang/lib/CIR/CodeGen/CIRGenBuilder.h

@@ -315,8 +315,10 @@ class CIRGenBuilderTy : public cir::CIRBaseBuilderTy {
     return getConstantInt(loc, getUInt32Ty(), c);
   }
   cir::ConstantOp getSInt64(uint64_t c, mlir::Location loc) {
-    cir::IntType sInt64Ty = getSInt64Ty();
-    return cir::ConstantOp::create(*this, loc, cir::IntAttr::get(sInt64Ty, c));
+    return getConstantInt(loc, getSInt64Ty(), c);
+  }
+  cir::ConstantOp getUInt64(uint64_t c, mlir::Location loc) {
+    return getConstantInt(loc, getUInt64Ty(), c);
   }
 
   mlir::Value createNeg(mlir::Value value) {

clang/lib/CIR/CodeGen/CIRGenBuiltin.cpp

@@ -630,6 +630,22 @@ 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;
@@ -66,9 +65,8 @@ mlir::Value CIRGenFunction::emitX86BuiltinExpr(unsigned builtinID,
   getContext().GetBuiltinType(builtinID, error, &iceArguments);
   assert(error == ASTContext::GE_None && "Error while getting builtin type.");
 
-  for (auto [idx, arg] : llvm::enumerate(e->arguments())) {
+  for (auto [idx, arg] : llvm::enumerate(e->arguments()))
     ops.push_back(emitScalarOrConstFoldImmArg(iceArguments, idx, arg));
-  }
 
   CIRGenBuilderTy &builder = getBuilder();
   mlir::Type voidTy = builder.getVoidTy();
@@ -98,6 +96,10 @@ 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:
@@ -107,7 +109,22 @@ 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;
+
+    cir::ConstantOp indexVal =
+        builder.getUInt64(index, getLoc(e->getExprLoc()));
+
+    // 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

@@ -1721,6 +1721,9 @@ 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

@@ -16,6 +16,21 @@
 
 #include <immintrin.h>
 
+// Lowering to pextrw requires optimization.
+int test_mm_extract_epi16(__m128i A) {
+  // CIR-LABEL: test_mm_extract_epi16
+  // CIR %{{.*}} = cir.vec.extract %{{.*}}[%{{.*}} : {{!u32i|!u64i}}] : !cir.vector<!s16i x 8>
+  // CIR %{{.*}} = cir.cast integral %{{.*}} : !u16i -> !s32i
+
+  // LLVM-LABEL: test_mm_extract_epi16
+  // LLVM: extractelement <8 x i16> %{{.*}}, {{i32|i64}} 1
+  // LLVM: zext i16 %{{.*}} to i32
+
+  // OGCG-LABEL: test_mm_extract_epi16
+  // OGCG: extractelement <8 x i16> %{{.*}}, {{i32|i64}} 1
+  // OGCG: zext i16 %{{.*}} to i32
+  return _mm_extract_epi16(A, 1);
+}
 
 void test_mm_clflush(void* A) {
   // CIR-LABEL: test_mm_clflush