Add _mm_stream_si64 intrinsic.

While I'm here, also fix the alignment computation for the whole family of
intrinsics.

PR17298.

llvm-svn: 191243

clang/include/clang/Basic/BuiltinsX86.def

@@ -258,6 +258,7 @@ BUILTIN(__builtin_ia32_storeupd, "vd*V2d", "")
 BUILTIN(__builtin_ia32_movmskpd, "iV2d", "")
 BUILTIN(__builtin_ia32_pmovmskb128, "iV16c", "")
 BUILTIN(__builtin_ia32_movnti, "vi*i", "")
+BUILTIN(__builtin_ia32_movnti64, "vLLi*LLi", "")
 BUILTIN(__builtin_ia32_movntpd, "vd*V2d", "")
 BUILTIN(__builtin_ia32_movntdq, "vV2LLi*V2LLi", "")
 BUILTIN(__builtin_ia32_psadbw128, "V2LLiV16cV16c", "")

clang/lib/CodeGen/CGBuiltin.cpp

@@ -3249,7 +3249,8 @@ Value *CodeGenFunction::EmitX86BuiltinExpr(unsigned BuiltinID,
   case X86::BI__builtin_ia32_movntpd256:
   case X86::BI__builtin_ia32_movntdq:
   case X86::BI__builtin_ia32_movntdq256:
-  case X86::BI__builtin_ia32_movnti: {
+  case X86::BI__builtin_ia32_movnti:
+  case X86::BI__builtin_ia32_movnti64: {
     llvm::MDNode *Node = llvm::MDNode::get(getLLVMContext(),
                                            Builder.getInt32(1));
 
@@ -3259,7 +3260,16 @@ Value *CodeGenFunction::EmitX86BuiltinExpr(unsigned BuiltinID,
                                       "cast");
     StoreInst *SI = Builder.CreateStore(Ops[1], BC);
     SI->setMetadata(CGM.getModule().getMDKindID("nontemporal"), Node);
-    SI->setAlignment(16);
+
+    // If the operand is an integer, we can't assume alignment. Otherwise,
+    // assume natural alignment.
+    QualType ArgTy = E->getArg(1)->getType();
+    unsigned Align;
+    if (ArgTy->isIntegerType())
+      Align = 1;
+    else
+      Align = getContext().getTypeSizeInChars(ArgTy).getQuantity();
+    SI->setAlignment(Align);
     return SI;
   }
   // 3DNow!

clang/lib/Headers/emmintrin.h

@@ -1214,6 +1214,14 @@ _mm_stream_si32(int *__p, int __a)
   __builtin_ia32_movnti(__p, __a);
 }
 
+#ifdef __x86_64__
+static __inline__ void __attribute__((__always_inline__, __nodebug__))
+_mm_stream_si64(long long *__p, long long __a)
+{
+  __builtin_ia32_movnti64(__p, __a);
+}
+#endif
+
 static __inline__ void __attribute__((__always_inline__, __nodebug__))
 _mm_clflush(void const *__p)
 {

clang/test/CodeGen/builtins-x86.c

@@ -55,6 +55,7 @@ void f0() {
   const float*   tmp_fCp;
   double*        tmp_dp;
   const double*  tmp_dCp;
+  long long*     tmp_LLip;
 
 #define imm_i 32
 #define imm_i_0_2 0
@@ -288,6 +289,9 @@ void f0() {
   tmp_i = __builtin_ia32_movmskpd(tmp_V2d);
   tmp_i = __builtin_ia32_pmovmskb128(tmp_V16c);
   (void) __builtin_ia32_movnti(tmp_ip, tmp_i);
+#ifdef USE_64
+  (void) __builtin_ia32_movnti64(tmp_LLip, tmp_LLi);
+#endif
   (void) __builtin_ia32_movntpd(tmp_dp, tmp_V2d);
   (void) __builtin_ia32_movntdq(tmp_V2LLip, tmp_V2LLi);
   tmp_V2LLi = __builtin_ia32_psadbw128(tmp_V16c, tmp_V16c);

clang/test/CodeGen/sse-builtins.c

@@ -1,4 +1,4 @@
-// RUN: %clang_cc1 -ffreestanding -triple i386-apple-darwin9 -target-cpu pentium4 -target-feature +sse4.1 -g -emit-llvm %s -o - | FileCheck %s
+// RUN: %clang_cc1 -ffreestanding -triple x86_64-apple-macosx10.8.0 -target-feature +sse4.1 -g -emit-llvm %s -o - | FileCheck %s
 
 #include <xmmintrin.h>
 #include <emmintrin.h>
@@ -188,3 +188,21 @@ void test_storel_epi64(__m128i x, void* y) {
   // CHECK: store {{.*}} i64* {{.*}}, align 1{{$}}
   _mm_storel_epi64(y, x);
 }
+
+void test_stream_si32(int x, void *y) {
+  // CHECK-LABEL: define void @test_stream_si32
+  // CHECK: store {{.*}} i32* {{.*}}, align 1, !nontemporal
+  _mm_stream_si32(y, x);
+}
+
+void test_stream_si64(long long x, void *y) {
+  // CHECK-LABEL: define void @test_stream_si64
+  // CHECK: store {{.*}} i64* {{.*}}, align 1, !nontemporal
+  _mm_stream_si64(y, x);
+}
+
+void test_stream_si128(__m128i x, void *y) {
+  // CHECK-LABEL: define void @test_stream_si128
+  // CHECK: store {{.*}} <2 x i64>* {{.*}}, align 16, !nontemporal
+  _mm_stream_si128(y, x);
+}