AVX-512: Fixed a bug in compress and expand intrinsics.

By Igor Breger (igor.breger@intel.com)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@238724 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Target/X86/X86ISelLowering.cpp b/lib/Target/X86/X86ISelLowering.cpp
index 3822dc0a732404ffb75324086710caa541785995..aab0087ed7a5c097c0d5319680bfde3b07c2a696 100644 (file)
--- a/lib/Target/X86/X86ISelLowering.cpp
+++ b/lib/Target/X86/X86ISelLowering.cpp
@@ -15670,11 +15670,12 @@ static SDValue LowerINTRINSIC_W_CHAIN(SDValue Op, const X86Subtarget *Subtarget,
     SDValue Addr = Op.getOperand(2);
     SDValue Chain = Op.getOperand(0);
 
+    EVT VT = DataToCompress.getValueType();
     if (isAllOnes(Mask)) // return just a store
       return DAG.getStore(Chain, dl, DataToCompress, Addr,
-                          MachinePointerInfo(), false, false, 0);
+                          MachinePointerInfo(), false, false,
+                          VT.getScalarSizeInBits()/8);
 
-    EVT VT = DataToCompress.getValueType();
     EVT MaskVT = EVT::getVectorVT(*DAG.getContext(), MVT::i1,
                                   VT.getVectorNumElements());
     EVT BitcastVT = EVT::getVectorVT(*DAG.getContext(), MVT::i1,
@@ -15686,7 +15687,8 @@ static SDValue LowerINTRINSIC_W_CHAIN(SDValue Op, const X86Subtarget *Subtarget,
     SDValue Compressed =  DAG.getNode(IntrData->Opc0, dl, VT, VMask,
                                       DataToCompress, DAG.getUNDEF(VT));
     return DAG.getStore(Chain, dl, Compressed, Addr,
-                        MachinePointerInfo(), false, false, 0);
+                        MachinePointerInfo(), false, false,
+                        VT.getScalarSizeInBits()/8);
   }
   case EXPAND_FROM_MEM: {
     SDLoc dl(Op);
@@ -15698,7 +15700,7 @@ static SDValue LowerINTRINSIC_W_CHAIN(SDValue Op, const X86Subtarget *Subtarget,
 
     if (isAllOnes(Mask)) // return just a load
       return DAG.getLoad(VT, dl, Chain, Addr, MachinePointerInfo(), false, false,
-                         false, 0);
+                         false, VT.getScalarSizeInBits()/8);
     EVT MaskVT = EVT::getVectorVT(*DAG.getContext(), MVT::i1,
                                   VT.getVectorNumElements());
     EVT BitcastVT = EVT::getVectorVT(*DAG.getContext(), MVT::i1,
@@ -15708,7 +15710,8 @@ static SDValue LowerINTRINSIC_W_CHAIN(SDValue Op, const X86Subtarget *Subtarget,
                                 DAG.getIntPtrConstant(0, dl));
 
     SDValue DataToExpand = DAG.getLoad(VT, dl, Chain, Addr, MachinePointerInfo(),
-                                   false, false, false, 0);
+                                       false, false, false,
+                                       VT.getScalarSizeInBits()/8);
 
     SDValue Results[] = {
         DAG.getNode(IntrData->Opc0, dl, VT, VMask, DataToExpand, PathThru),

diff --git a/test/CodeGen/X86/avx512vl-intrinsics.ll b/test/CodeGen/X86/avx512vl-intrinsics.ll
index fd76ed5d0dbdb7bf043a69fea4f6a6cc7c894e64..b93b1d04d80d67fe0041e86ad42069bff6889f04 100644 (file)
--- a/test/CodeGen/X86/avx512vl-intrinsics.ll
+++ b/test/CodeGen/X86/avx512vl-intrinsics.ll
@@ -668,7 +668,7 @@ declare <4 x float> @llvm.x86.avx512.mask.compress.ps.128(<4 x float> %data, <4
 
 ; CHECK-LABEL: compr7
 ; CHECK-NOT: vcompress
-; CHECK: vmovapd
+; CHECK: vmovupd
 define void @compr7(i8* %addr, <8 x double> %data) {
   call void @llvm.x86.avx512.mask.compress.store.pd.512(i8* %addr, <8 x double> %data, i8 -1)
   ret void
@@ -757,7 +757,7 @@ declare <4 x float> @llvm.x86.avx512.mask.expand.ps.128(<4 x float> %data, <4 x
 
 ; CHECK-LABEL: expand7
 ; CHECK-NOT: vexpand
-; CHECK: vmovapd
+; CHECK: vmovupd
 define <8 x double> @expand7(i8* %addr, <8 x double> %data) {
   %res = call <8 x double> @llvm.x86.avx512.mask.expand.load.pd.512(i8* %addr, <8 x double> %data, i8 -1)
   ret <8 x double> %res
@@ -2552,4 +2552,4 @@ define <4 x float> @test_mm512_min_ps_128(<4 x float> %a0, <4 x float> %a1, i8 %
   %res = call <4 x float> @llvm.x86.avx512.mask.min.ps.128(<4 x float> %a0, <4 x float> %a1, <4 x float>zeroinitializer, i8 -1)
   ret <4 x float> %res
 }
-declare <4 x float> @llvm.x86.avx512.mask.min.ps.128(<4 x float>, <4 x float>, <4 x float>, i8)
\ No newline at end of file
+declare <4 x float> @llvm.x86.avx512.mask.min.ps.128(<4 x float>, <4 x float>, <4 x float>, i8)