Add SSE4A MOVNTSS/MOVNTSD instructions.

llvm-svn: 156281

3 files changed, 43 insertions, 0 deletions

diff --git a/llvm/include/llvm/IntrinsicsX86.td b/llvm/include/llvm/IntrinsicsX86.td
index cb7b3eadc87..1abeb9e5559 100644
--- a/llvm/include/llvm/IntrinsicsX86.td
+++ b/llvm/include/llvm/IntrinsicsX86.td
@@ -1005,6 +1005,16 @@ let TargetPrefix = "x86" in { // All intrinsics start with "llvm.x86.".
 }
 
 //===----------------------------------------------------------------------===//
+// SSE4A
+
+let TargetPrefix = "x86" in {  // All intrinsics start with "llvm.x86.".
+  def int_x86_sse4a_movnt_ss : GCCBuiltin<"__builtin_ia32_movntss">,
+    Intrinsic<[], [llvm_ptr_ty, llvm_v4f32_ty], []>;
+  def int_x86_sse4a_movnt_sd : GCCBuiltin<"__builtin_ia32_movntsd">,
+    Intrinsic<[], [llvm_ptr_ty, llvm_v2f64_ty], []>;
+}
+
+//===----------------------------------------------------------------------===//
 // AVX
 
 // Arithmetic ops
diff --git a/llvm/lib/Target/X86/X86InstrSSE.td b/llvm/lib/Target/X86/X86InstrSSE.td
index aa3e08bd4dc..0a940552453 100644
--- a/llvm/lib/Target/X86/X86InstrSSE.td
+++ b/llvm/lib/Target/X86/X86InstrSSE.td
@@ -7271,6 +7271,20 @@ defm : pclmul_alias<"lqhq", 0x10>;
 defm : pclmul_alias<"lqlq", 0x00>;
 
 //===----------------------------------------------------------------------===//
+// SSE4A Instructions
+//===----------------------------------------------------------------------===//
+
+let Predicates = [HasSSE4A] in {
+def MOVNTSS : I<0x2B, MRMDestMem, (outs), (ins f32mem:$dst, VR128:$src),
+                "movntss\t{$src, $dst|$dst, $src}",
+                [(int_x86_sse4a_movnt_ss addr:$dst, VR128:$src)]>, XS;
+
+def MOVNTSD : I<0x2B, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
+                "movntsd\t{$src, $dst|$dst, $src}",
+                [(int_x86_sse4a_movnt_sd addr:$dst, VR128:$src)]>, XD;
+}
+
+//===----------------------------------------------------------------------===//
 // AVX Instructions
 //===----------------------------------------------------------------------===//
 
diff --git a/llvm/test/CodeGen/X86/sse4a.ll b/llvm/test/CodeGen/X86/sse4a.ll
new file mode 100644
index 00000000000..0732353fce4
--- /dev/null
+++ b/llvm/test/CodeGen/X86/sse4a.ll
@@ -0,0 +1,19 @@
+; RUN: llc < %s -mtriple=i686-apple-darwin9 -mattr=sse4a | FileCheck %s
+
+define void @test1(float* %p, <4 x float> %a) nounwind optsize ssp {
+; CHECK: movntss
+entry:
+  tail call void @llvm.x86.sse4a.movnt.ss(float* %p, <4 x float> %a) nounwind
+  ret void
+}
+
+declare void @llvm.x86.sse4a.movnt.ss(float*, <4 x float>)
+
+define void @test2(double* %p, <2 x double> %a) nounwind optsize ssp {
+; CHECK: movntsd
+entry:
+  tail call void @llvm.x86.sse4a.movnt.sd(double* %p, <2 x double> %a) nounwind
+  ret void
+}
+
+declare void @llvm.x86.sse4a.movnt.sd(double*, <2 x double>)