[DAGCombiner] Add vector support for (srl (trunc (srl x, c1)), c2) combine.

llvm-svn: 301305

diff --git a/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp b/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
index f7a15ba..c67bca4 100644 (file)
--- a/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
@@ -5615,23 +5615,24 @@ SDValue DAGCombiner::visitSRL(SDNode *N) {
 
   // fold (srl (trunc (srl x, c1)), c2) -> 0 or (trunc (srl x, (add c1, c2)))
   if (N1C && N0.getOpcode() == ISD::TRUNCATE &&
-      N0.getOperand(0).getOpcode() == ISD::SRL &&
-      isa<ConstantSDNode>(N0.getOperand(0)->getOperand(1))) {
-    uint64_t c1 = N0.getOperand(0).getConstantOperandVal(1);
-    uint64_t c2 = N1C->getZExtValue();
-    EVT InnerShiftVT = N0.getOperand(0).getValueType();
-    EVT ShiftCountVT = N0.getOperand(0)->getOperand(1).getValueType();
-    uint64_t InnerShiftSize = InnerShiftVT.getScalarSizeInBits();
-    // This is only valid if the OpSizeInBits + c1 = size of inner shift.
-    if (c1 + OpSizeInBits == InnerShiftSize) {
-      SDLoc DL(N0);
-      if (c1 + c2 >= InnerShiftSize)
-        return DAG.getConstant(0, DL, VT);
-      return DAG.getNode(ISD::TRUNCATE, DL, VT,
-                         DAG.getNode(ISD::SRL, DL, InnerShiftVT,
-                                     N0.getOperand(0)->getOperand(0),
-                                     DAG.getConstant(c1 + c2, DL,
-                                                     ShiftCountVT)));
+      N0.getOperand(0).getOpcode() == ISD::SRL) {
+    if (auto N001C = isConstOrConstSplat(N0.getOperand(0).getOperand(1))) {
+      uint64_t c1 = N001C->getZExtValue();
+      uint64_t c2 = N1C->getZExtValue();
+      EVT InnerShiftVT = N0.getOperand(0).getValueType();
+      EVT ShiftCountVT = N0.getOperand(0).getOperand(1).getValueType();
+      uint64_t InnerShiftSize = InnerShiftVT.getScalarSizeInBits();
+      // This is only valid if the OpSizeInBits + c1 = size of inner shift.
+      if (c1 + OpSizeInBits == InnerShiftSize) {
+        SDLoc DL(N0);
+        if (c1 + c2 >= InnerShiftSize)
+          return DAG.getConstant(0, DL, VT);
+        return DAG.getNode(ISD::TRUNCATE, DL, VT,
+                           DAG.getNode(ISD::SRL, DL, InnerShiftVT,
+                                       N0.getOperand(0).getOperand(0),
+                                       DAG.getConstant(c1 + c2, DL,
+                                                       ShiftCountVT)));
+      }
     }
   }
 

diff --git a/llvm/test/CodeGen/X86/combine-srl.ll b/llvm/test/CodeGen/X86/combine-srl.ll
index 44c4510..706e890 100644 (file)
--- a/llvm/test/CodeGen/X86/combine-srl.ll
+++ b/llvm/test/CodeGen/X86/combine-srl.ll
@@ -223,18 +223,17 @@ define <4 x i32> @combine_vec_lshr_lshr_zero1(<4 x i32> %x) {
 define <4 x i32> @combine_vec_lshr_trunc_lshr0(<4 x i64> %x) {
 ; SSE-LABEL: combine_vec_lshr_trunc_lshr0:
 ; SSE:       # BB#0:
-; SSE-NEXT:    psrlq $32, %xmm1
-; SSE-NEXT:    psrlq $32, %xmm0
+; SSE-NEXT:    psrlq $48, %xmm1
+; SSE-NEXT:    psrlq $48, %xmm0
 ; SSE-NEXT:    shufps {{.*#+}} xmm0 = xmm0[0,2],xmm1[0,2]
-; SSE-NEXT:    psrld $16, %xmm0
 ; SSE-NEXT:    retq
 ;
 ; AVX-LABEL: combine_vec_lshr_trunc_lshr0:
 ; AVX:       # BB#0:
-; AVX-NEXT:    vpsrlq $32, %ymm0, %ymm0
+; AVX-NEXT:    vpsrlq $48, %ymm0, %ymm0
 ; AVX-NEXT:    vpshufd {{.*#+}} ymm0 = ymm0[0,2,2,3,4,6,6,7]
 ; AVX-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,2,3]
-; AVX-NEXT:    vpsrld $16, %xmm0, %xmm0
+; AVX-NEXT:    # kill: %XMM0<def> %XMM0<kill> %YMM0<kill>
 ; AVX-NEXT:    vzeroupper
 ; AVX-NEXT:    retq
   %1 = lshr <4 x i64> %x, <i64 32, i64 32, i64 32, i64 32>