[LegalizeTypes] When promoting the result of EXTRACT_SUBVECTOR, also check if the input needs to be promoted. Use that to determine the element type to extract.
Otherwise we end up creating extract_vector_elts that then each need to have their input promoted. This can lead to truncates needing to be emitted for each of those.
But we already emitted any_extends when we legalized the extract_subvector. So now we have pairs of any_extend+trunc that partially cancel. But depending on how DAGCombiner visits them we can get weird results.
By promoting the input at the same time we can create only a single any_extend or truncate.
There's one regression in the vector-narrow-binop.ll case, but that looks easy to fix with a follow up patch.
llvm-svn: 354647
diff --git a/llvm/lib/CodeGen/SelectionDAG/LegalizeIntegerTypes.cpp b/llvm/lib/CodeGen/SelectionDAG/LegalizeIntegerTypes.cpp
index e664e06..fa0e7b8 100644
--- a/llvm/lib/CodeGen/SelectionDAG/LegalizeIntegerTypes.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/LegalizeIntegerTypes.cpp
@@ -3667,8 +3667,6 @@
SDValue DAGTypeLegalizer::PromoteIntRes_EXTRACT_SUBVECTOR(SDNode *N) {
- SDValue InOp0 = N->getOperand(0);
- EVT InVT = InOp0.getValueType();
EVT OutVT = N->getValueType(0);
EVT NOutVT = TLI.getTypeToTransformTo(*DAG.getContext(), OutVT);
@@ -3679,6 +3677,12 @@
SDLoc dl(N);
SDValue BaseIdx = N->getOperand(1);
+ SDValue InOp0 = N->getOperand(0);
+ if (getTypeAction(InOp0.getValueType()) == TargetLowering::TypePromoteInteger)
+ InOp0 = GetPromotedInteger(N->getOperand(0));
+
+ EVT InVT = InOp0.getValueType();
+
SmallVector<SDValue, 8> Ops;
Ops.reserve(OutNumElems);
for (unsigned i = 0; i != OutNumElems; ++i) {
@@ -3689,7 +3693,7 @@
SDValue Ext = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, dl,
InVT.getVectorElementType(), N->getOperand(0), Index);
- SDValue Op = DAG.getNode(ISD::ANY_EXTEND, dl, NOutVTElem, Ext);
+ SDValue Op = DAG.getAnyExtOrTrunc(Ext, dl, NOutVTElem);
// Insert the converted element to the new vector.
Ops.push_back(Op);
}
diff --git a/llvm/test/CodeGen/X86/2011-12-28-vselecti8.ll b/llvm/test/CodeGen/X86/2011-12-28-vselecti8.ll
index d564e9f..bb3b4ed 100644
--- a/llvm/test/CodeGen/X86/2011-12-28-vselecti8.ll
+++ b/llvm/test/CodeGen/X86/2011-12-28-vselecti8.ll
@@ -18,10 +18,8 @@
define void @foo8(float* nocapture %RET) nounwind {
; CHECK-LABEL: foo8:
; CHECK: ## %bb.0: ## %allocas
-; CHECK-NEXT: pmovzxbd {{.*#+}} xmm0 = mem[0],zero,zero,zero,mem[1],zero,zero,zero,mem[2],zero,zero,zero,mem[3],zero,zero,zero
-; CHECK-NEXT: pmovzxbd {{.*#+}} xmm1 = mem[0],zero,zero,zero,mem[1],zero,zero,zero,mem[2],zero,zero,zero,mem[3],zero,zero,zero
-; CHECK-NEXT: cvtdq2ps %xmm0, %xmm0
-; CHECK-NEXT: cvtdq2ps %xmm1, %xmm1
+; CHECK-NEXT: movaps {{.*#+}} xmm0 = [1.0E+2,2.0E+0,1.0E+2,4.0E+0]
+; CHECK-NEXT: movaps {{.*#+}} xmm1 = [1.0E+2,6.0E+0,1.0E+2,8.0E+0]
; CHECK-NEXT: movups %xmm1, 16(%rdi)
; CHECK-NEXT: movups %xmm0, (%rdi)
; CHECK-NEXT: retq
diff --git a/llvm/test/CodeGen/X86/pr35918.ll b/llvm/test/CodeGen/X86/pr35918.ll
index 6ce9706..5b9e7a6 100644
--- a/llvm/test/CodeGen/X86/pr35918.ll
+++ b/llvm/test/CodeGen/X86/pr35918.ll
@@ -69,7 +69,7 @@
; X64-SKX-NEXT: vpmaxsd %xmm1, %xmm0, %xmm0
; X64-SKX-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0],xmm1[1],xmm0[2],xmm1[3],xmm0[4],xmm1[5],xmm0[6],xmm1[7]
; X64-SKX-NEXT: vpsrld $7, %xmm0, %xmm0
-; X64-SKX-NEXT: vpmovzxwd {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero
+; X64-SKX-NEXT: vpmovzxdq {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero
; X64-SKX-NEXT: vpmovqw %xmm0, -{{[0-9]+}}(%rsp)
; X64-SKX-NEXT: vmovd {{.*#+}} xmm0 = mem[0],zero,zero,zero
; X64-SKX-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[0],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero
diff --git a/llvm/test/CodeGen/X86/vector-narrow-binop.ll b/llvm/test/CodeGen/X86/vector-narrow-binop.ll
index 4836b49..a452ca44 100644
--- a/llvm/test/CodeGen/X86/vector-narrow-binop.ll
+++ b/llvm/test/CodeGen/X86/vector-narrow-binop.ll
@@ -107,18 +107,16 @@
; SSE: # %bb.0:
; SSE-NEXT: pxor %xmm2, %xmm2
; SSE-NEXT: psubd %xmm0, %xmm2
-; SSE-NEXT: pshufd {{.*#+}} xmm0 = xmm2[0,2,2,3]
-; SSE-NEXT: psrld $16, %xmm0
-; SSE-NEXT: movsd {{.*#+}} xmm1 = xmm0[0],xmm1[1]
-; SSE-NEXT: movapd %xmm1, %xmm0
+; SSE-NEXT: punpcklbw {{.*#+}} xmm2 = xmm2[0],xmm0[0],xmm2[1],xmm0[1],xmm2[2],xmm0[2],xmm2[3],xmm0[3],xmm2[4],xmm0[4],xmm2[5],xmm0[5],xmm2[6],xmm0[6],xmm2[7],xmm0[7]
+; SSE-NEXT: shufps {{.*#+}} xmm2 = xmm2[1,1],xmm1[2,3]
+; SSE-NEXT: movaps %xmm2, %xmm0
; SSE-NEXT: retq
;
; AVX1-LABEL: PR39893:
; AVX1: # %bb.0:
; AVX1-NEXT: vpxor %xmm2, %xmm2, %xmm2
; AVX1-NEXT: vpsubd %xmm0, %xmm2, %xmm0
-; AVX1-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
-; AVX1-NEXT: vpsrld $16, %xmm0, %xmm0
+; AVX1-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[2],zero,xmm0[3],zero,xmm0[2],zero,xmm0[3],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero
; AVX1-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0,1,2,3],xmm1[4,5,6,7]
; AVX1-NEXT: retq
;
@@ -126,8 +124,7 @@
; AVX2: # %bb.0:
; AVX2-NEXT: vpxor %xmm2, %xmm2, %xmm2
; AVX2-NEXT: vpsubd %xmm0, %xmm2, %xmm0
-; AVX2-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
-; AVX2-NEXT: vpsrld $16, %xmm0, %xmm0
+; AVX2-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[2],zero,xmm0[3],zero,xmm0[2],zero,xmm0[3],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero
; AVX2-NEXT: vpblendd {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,3]
; AVX2-NEXT: retq
;
@@ -135,8 +132,7 @@
; AVX512: # %bb.0:
; AVX512-NEXT: vpxor %xmm2, %xmm2, %xmm2
; AVX512-NEXT: vpsubd %xmm0, %xmm2, %xmm0
-; AVX512-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
-; AVX512-NEXT: vpsrld $16, %xmm0, %xmm0
+; AVX512-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[2],zero,xmm0[3],zero,xmm0[2],zero,xmm0[3],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero
; AVX512-NEXT: vpblendd {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,3]
; AVX512-NEXT: retq
%sub = sub <2 x i32> <i32 0, i32 undef>, %x
@@ -151,15 +147,16 @@
; SSE-NEXT: xorps %xmm1, %xmm1
; SSE-NEXT: subps %xmm0, %xmm1
; SSE-NEXT: punpcklbw {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3],xmm1[4],xmm0[4],xmm1[5],xmm0[5],xmm1[6],xmm0[6],xmm1[7],xmm0[7]
-; SSE-NEXT: punpcklwd {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3]
-; SSE-NEXT: pshufd {{.*#+}} xmm0 = xmm1[0,1,1,3]
+; SSE-NEXT: pshufd {{.*#+}} xmm0 = xmm1[0,1,0,3]
+; SSE-NEXT: pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,5,5,6,7]
; SSE-NEXT: retq
;
; AVX-LABEL: PR39893_2:
; AVX: # %bb.0:
; AVX-NEXT: vxorps %xmm1, %xmm1, %xmm1
; AVX-NEXT: vsubps %xmm0, %xmm1, %xmm0
-; AVX-NEXT: vpmovzxbq {{.*#+}} xmm0 = xmm0[0],zero,zero,zero,zero,zero,zero,zero,xmm0[1],zero,zero,zero,zero,zero,zero,zero
+; AVX-NEXT: vpmovzxbw {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
+; AVX-NEXT: vpmovzxwq {{.*#+}} xmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero
; AVX-NEXT: retq
%fsub = fsub <2 x float> zeroinitializer, %x
%bc = bitcast <2 x float> %fsub to <8 x i8>
