[CIR][CIRGen] handle vectors of size 3 like size 4 for better performance (#1363)

This PR removes XFAIL test_vec3

Author: FantasqueX

clang/lib/CIR/CodeGen/CIRGenExpr.cpp

@@ -39,6 +39,8 @@
 #include "mlir/IR/Operation.h"
 #include "mlir/IR/Value.h"
 
+#include <numeric>
+
 using namespace clang;
 using namespace clang::CIRGen;
 using namespace cir;
@@ -2936,7 +2938,13 @@ mlir::Value CIRGenFunction::emitLoadOfScalar(Address addr, bool isVolatile,
         CGM.getABIInfo().getOptimalVectorMemoryType(vTy, getLangOpts());
 
     if (vTy != newVecTy) {
-      llvm_unreachable("NYI");
+      const Address cast = addr.withElementType(newVecTy);
+      mlir::Value v = builder.createLoad(loc, cast, isVolatile);
+      const uint64_t oldNumElements = vTy.getSize();
+      SmallVector<int64_t, 16> mask(oldNumElements);
+      std::iota(mask.begin(), mask.end(), 0);
+      v = builder.createVecShuffle(loc, v, mask);
+      return emitFromMemory(v, ty);
     }
   }
 

clang/lib/CIR/CodeGen/TargetInfo.cpp

@@ -378,7 +378,7 @@ cir::VectorType
 ABIInfo::getOptimalVectorMemoryType(cir::VectorType T,
                                     const clang::LangOptions &Opt) const {
   if (T.getSize() == 3 && !Opt.PreserveVec3Type) {
-    llvm_unreachable("NYI");
+    return cir::VectorType::get(&CGT.getMLIRContext(), T.getEltType(), 4);
   }
   return T;
 }

clang/test/CIR/CodeGen/vectype-ext.cpp

@@ -2,7 +2,6 @@
 // RUN: FileCheck --input-file=%t.cir %s -check-prefix=CIR
 // RUN: %clang_cc1 -std=c++17 -fclangir -emit-llvm -triple x86_64-unknown-linux-gnu %s -o %t.ll
 // RUN: FileCheck --input-file=%t.ll %s -check-prefix=LLVM
-// XFAIL: *
 
 typedef int vi4 __attribute__((ext_vector_type(4)));
 typedef int vi3 __attribute__((ext_vector_type(3)));