[DAG] visitRotate - modulo non-uniform constant rotation amounts

diff --git a/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp b/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
index 596a3a4..01bc936 100644 (file)
--- a/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
@@ -7494,13 +7494,17 @@ SDValue DAGCombiner::visitRotate(SDNode *N) {
   }
 
   // fold (rot x, c) -> (rot x, c % BitSize)
-  // TODO - support non-uniform vector amounts.
-  if (ConstantSDNode *Cst = isConstOrConstSplat(N1)) {
-    if (Cst->getAPIntValue().uge(Bitsize)) {
-      uint64_t RotAmt = Cst->getAPIntValue().urem(Bitsize);
-      return DAG.getNode(N->getOpcode(), dl, VT, N0,
-                         DAG.getConstant(RotAmt, dl, N1.getValueType()));
-    }
+  bool OutOfRange = false;
+  auto MatchOutOfRange = [Bitsize, &OutOfRange](ConstantSDNode *C) {
+    OutOfRange |= C->getAPIntValue().uge(Bitsize);
+    return true;
+  };
+  if (ISD::matchUnaryPredicate(N1, MatchOutOfRange) && OutOfRange) {
+    EVT AmtVT = N1.getValueType();
+    SDValue Bits = DAG.getConstant(Bitsize, dl, AmtVT);
+    if (SDValue Amt =
+            DAG.FoldConstantArithmetic(ISD::UREM, dl, AmtVT, {N1, Bits}))
+      return DAG.getNode(N->getOpcode(), dl, VT, N0, Amt);
   }
 
   // Simplify the operands using demanded-bits information.

diff --git a/llvm/test/CodeGen/X86/vector-fshl-rot-256.ll b/llvm/test/CodeGen/X86/vector-fshl-rot-256.ll
index ba4279f..f300eee 100644 (file)
--- a/llvm/test/CodeGen/X86/vector-fshl-rot-256.ll
+++ b/llvm/test/CodeGen/X86/vector-fshl-rot-256.ll
@@ -1272,7 +1272,7 @@ define <32 x i8> @constant_funnnel_v32i8(<32 x i8> %x) nounwind {
 ; XOPAVX1-LABEL: constant_funnnel_v32i8:
 ; XOPAVX1:       # %bb.0:
 ; XOPAVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
-; XOPAVX1-NEXT:    vmovdqa {{.*#+}} xmm2 = [0,1,2,3,4,5,6,7,8,7,6,5,4,3,2,1]
+; XOPAVX1-NEXT:    vmovdqa {{.*#+}} xmm2 = [0,1,2,3,4,5,6,7,0,7,6,5,4,3,2,1]
 ; XOPAVX1-NEXT:    vprotb %xmm2, %xmm1, %xmm1
 ; XOPAVX1-NEXT:    vprotb %xmm2, %xmm0, %xmm0
 ; XOPAVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
@@ -1281,7 +1281,7 @@ define <32 x i8> @constant_funnnel_v32i8(<32 x i8> %x) nounwind {
 ; XOPAVX2-LABEL: constant_funnnel_v32i8:
 ; XOPAVX2:       # %bb.0:
 ; XOPAVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; XOPAVX2-NEXT:    vmovdqa {{.*#+}} xmm2 = [0,1,2,3,4,5,6,7,8,7,6,5,4,3,2,1]
+; XOPAVX2-NEXT:    vmovdqa {{.*#+}} xmm2 = [0,1,2,3,4,5,6,7,0,7,6,5,4,3,2,1]
 ; XOPAVX2-NEXT:    vprotb %xmm2, %xmm1, %xmm1
 ; XOPAVX2-NEXT:    vprotb %xmm2, %xmm0, %xmm0
 ; XOPAVX2-NEXT:    vinserti128 $1, %xmm1, %ymm0, %ymm0

diff --git a/llvm/test/CodeGen/X86/vector-fshr-rot-256.ll b/llvm/test/CodeGen/X86/vector-fshr-rot-256.ll
index 7d6a5ee..bc898ad 100644 (file)
--- a/llvm/test/CodeGen/X86/vector-fshr-rot-256.ll
+++ b/llvm/test/CodeGen/X86/vector-fshr-rot-256.ll
@@ -1350,7 +1350,7 @@ define <32 x i8> @constant_funnnel_v32i8(<32 x i8> %x) nounwind {
 ; XOPAVX1-LABEL: constant_funnnel_v32i8:
 ; XOPAVX1:       # %bb.0:
 ; XOPAVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
-; XOPAVX1-NEXT:    vmovdqa {{.*#+}} xmm2 = [0,255,254,253,252,251,250,249,248,249,250,251,252,253,254,255]
+; XOPAVX1-NEXT:    vmovdqa {{.*#+}} xmm2 = [0,7,6,5,4,3,2,1,0,1,2,3,4,5,6,7]
 ; XOPAVX1-NEXT:    vprotb %xmm2, %xmm1, %xmm1
 ; XOPAVX1-NEXT:    vprotb %xmm2, %xmm0, %xmm0
 ; XOPAVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
@@ -1359,7 +1359,7 @@ define <32 x i8> @constant_funnnel_v32i8(<32 x i8> %x) nounwind {
 ; XOPAVX2-LABEL: constant_funnnel_v32i8:
 ; XOPAVX2:       # %bb.0:
 ; XOPAVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; XOPAVX2-NEXT:    vmovdqa {{.*#+}} xmm2 = [0,255,254,253,252,251,250,249,248,249,250,251,252,253,254,255]
+; XOPAVX2-NEXT:    vmovdqa {{.*#+}} xmm2 = [0,7,6,5,4,3,2,1,0,1,2,3,4,5,6,7]
 ; XOPAVX2-NEXT:    vprotb %xmm2, %xmm1, %xmm1
 ; XOPAVX2-NEXT:    vprotb %xmm2, %xmm0, %xmm0
 ; XOPAVX2-NEXT:    vinserti128 $1, %xmm1, %ymm0, %ymm0

diff --git a/llvm/test/CodeGen/X86/vector-rotate-256.ll b/llvm/test/CodeGen/X86/vector-rotate-256.ll
index 33f14d7..c0e5945 100644 (file)
--- a/llvm/test/CodeGen/X86/vector-rotate-256.ll
+++ b/llvm/test/CodeGen/X86/vector-rotate-256.ll
@@ -1265,7 +1265,7 @@ define <32 x i8> @constant_rotate_v32i8(<32 x i8> %a) nounwind {
 ; XOPAVX1-LABEL: constant_rotate_v32i8:
 ; XOPAVX1:       # %bb.0:
 ; XOPAVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
-; XOPAVX1-NEXT:    vmovdqa {{.*#+}} xmm2 = [0,1,2,3,4,5,6,7,8,7,6,5,4,3,2,1]
+; XOPAVX1-NEXT:    vmovdqa {{.*#+}} xmm2 = [0,1,2,3,4,5,6,7,0,7,6,5,4,3,2,1]
 ; XOPAVX1-NEXT:    vprotb %xmm2, %xmm1, %xmm1
 ; XOPAVX1-NEXT:    vprotb %xmm2, %xmm0, %xmm0
 ; XOPAVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
@@ -1274,7 +1274,7 @@ define <32 x i8> @constant_rotate_v32i8(<32 x i8> %a) nounwind {
 ; XOPAVX2-LABEL: constant_rotate_v32i8:
 ; XOPAVX2:       # %bb.0:
 ; XOPAVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; XOPAVX2-NEXT:    vmovdqa {{.*#+}} xmm2 = [0,1,2,3,4,5,6,7,8,7,6,5,4,3,2,1]
+; XOPAVX2-NEXT:    vmovdqa {{.*#+}} xmm2 = [0,1,2,3,4,5,6,7,0,7,6,5,4,3,2,1]
 ; XOPAVX2-NEXT:    vprotb %xmm2, %xmm1, %xmm1
 ; XOPAVX2-NEXT:    vprotb %xmm2, %xmm0, %xmm0
 ; XOPAVX2-NEXT:    vinserti128 $1, %xmm1, %ymm0, %ymm0