Index: llvm/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp
===================================================================
--- llvm/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp
+++ llvm/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp
@@ -4361,7 +4361,8 @@
 // In all other cases the function returns 'false'.
 static bool getUniformBase(const Value *Ptr, SDValue &Base, SDValue &Index,
                            ISD::MemIndexType &IndexType, SDValue &Scale,
-                           SelectionDAGBuilder *SDB, const BasicBlock *CurBB) {
+                           SelectionDAGBuilder *SDB, const BasicBlock *CurBB,
+                           uint64_t ElemSize) {
   SelectionDAG& DAG = SDB->DAG;
   const TargetLowering &TLI = DAG.getTargetLoweringInfo();
   const DataLayout &DL = DAG.getDataLayout();
@@ -4402,9 +4403,11 @@
   Index = SDB->getValue(IndexVal);
   IndexType = ISD::SIGNED_SCALED;
 
-  // MGATHER/MSCATTER only support scaling by a power-of-two.
+  // MGATHER/MSCATTER are only required to support scaling by one or by the
+  // element size. Other scales may be produced using target-specific DAG
+  // combines.
   uint64_t ScaleVal = DL.getTypeAllocSize(GEP->getResultElementType());
-  if (!isPowerOf2_64(ScaleVal))
+  if (ScaleVal != ElemSize && ScaleVal != 1)
     return false;
 
   Scale =
@@ -4430,7 +4433,7 @@
   ISD::MemIndexType IndexType;
   SDValue Scale;
   bool UniformBase = getUniformBase(Ptr, Base, Index, IndexType, Scale, this,
-                                    I.getParent());
+                                    I.getParent(), VT.getScalarStoreSize());
 
   unsigned AS = Ptr->getType()->getScalarType()->getPointerAddressSpace();
   MachineMemOperand *MMO = DAG.getMachineFunction().getMachineMemOperand(
@@ -4538,7 +4541,7 @@
   ISD::MemIndexType IndexType;
   SDValue Scale;
   bool UniformBase = getUniformBase(Ptr, Base, Index, IndexType, Scale, this,
-                                    I.getParent());
+                                    I.getParent(), VT.getScalarStoreSize());
   unsigned AS = Ptr->getType()->getScalarType()->getPointerAddressSpace();
   MachineMemOperand *MMO = DAG.getMachineFunction().getMachineMemOperand(
       MachinePointerInfo(AS), MachineMemOperand::MOLoad,
@@ -7408,7 +7411,8 @@
     SDValue Base, Index, Scale;
     ISD::MemIndexType IndexType;
     bool UniformBase = getUniformBase(PtrOperand, Base, Index, IndexType, Scale,
-                                      this, VPIntrin.getParent());
+                                      this, VPIntrin.getParent(),
+                                      VT.getScalarStoreSize());
     if (!UniformBase) {
       Base = DAG.getConstant(0, DL, TLI.getPointerTy(DAG.getDataLayout()));
       Index = getValue(PtrOperand);
@@ -7464,7 +7468,8 @@
     SDValue Base, Index, Scale;
     ISD::MemIndexType IndexType;
     bool UniformBase = getUniformBase(PtrOperand, Base, Index, IndexType, Scale,
-                                      this, VPIntrin.getParent());
+                                      this, VPIntrin.getParent(),
+                                      VT.getScalarStoreSize());
     if (!UniformBase) {
       Base = DAG.getConstant(0, DL, TLI.getPointerTy(DAG.getDataLayout()));
       Index = getValue(PtrOperand);
Index: llvm/test/CodeGen/X86/gather-scatter-opaque-ptr-2.ll
===================================================================
--- /dev/null
+++ llvm/test/CodeGen/X86/gather-scatter-opaque-ptr-2.ll
@@ -0,0 +1,89 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc -mtriple=x86_64-linux -mcpu=skylake -mattr=+avx2 < %s | FileCheck %s
+
+define <2 x float> @gather_v2f32_scale_512(ptr %result, <2 x i64> %idx, <2 x i1> %mask) {
+; CHECK-LABEL: gather_v2f32_scale_512:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vpsllq $9, %xmm0, %xmm2
+; CHECK-NEXT:    vpshufd {{.*#+}} xmm0 = xmm1[0,2,2,3]
+; CHECK-NEXT:    vpslld $31, %xmm0, %xmm1
+; CHECK-NEXT:    vpxor %xmm0, %xmm0, %xmm0
+; CHECK-NEXT:    vgatherqps %xmm1, (%rdi,%xmm2), %xmm0
+; CHECK-NEXT:    retq
+  %gep = getelementptr inbounds [512 x i8], ptr %result, <2 x i64> %idx
+  %res = call <2 x float> @llvm.masked.gather.v2f32.v2p0(<2 x ptr> %gep, i32 0, <2 x i1> %mask, <2 x float> zeroinitializer)
+  ret <2 x float> %res
+}
+
+define <2 x float> @gather_v2f32_scale_16(ptr %result, <2 x i64> %idx, <2 x i1> %mask) {
+; CHECK-LABEL: gather_v2f32_scale_16:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vpsllq $4, %xmm0, %xmm2
+; CHECK-NEXT:    vpshufd {{.*#+}} xmm0 = xmm1[0,2,2,3]
+; CHECK-NEXT:    vpslld $31, %xmm0, %xmm1
+; CHECK-NEXT:    vpxor %xmm0, %xmm0, %xmm0
+; CHECK-NEXT:    vgatherqps %xmm1, (%rdi,%xmm2), %xmm0
+; CHECK-NEXT:    retq
+  %gep = getelementptr inbounds [16 x i8], ptr %result, <2 x i64> %idx
+  %res = call <2 x float> @llvm.masked.gather.v2f32.v2p0(<2 x ptr> %gep, i32 0, <2 x i1> %mask, <2 x float> zeroinitializer)
+  ret <2 x float> %res
+}
+
+define <2 x float> @gather_v2f32_scale_8(ptr %result, <2 x i64> %idx, <2 x i1> %mask) {
+; CHECK-LABEL: gather_v2f32_scale_8:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[0,2,2,3]
+; CHECK-NEXT:    vpslld $31, %xmm1, %xmm2
+; CHECK-NEXT:    vpxor %xmm1, %xmm1, %xmm1
+; CHECK-NEXT:    vgatherqps %xmm2, (%rdi,%xmm0,8), %xmm1
+; CHECK-NEXT:    vmovaps %xmm1, %xmm0
+; CHECK-NEXT:    retq
+  %gep = getelementptr inbounds [8 x i8], ptr %result, <2 x i64> %idx
+  %res = call <2 x float> @llvm.masked.gather.v2f32.v2p0(<2 x ptr> %gep, i32 0, <2 x i1> %mask, <2 x float> zeroinitializer)
+  ret <2 x float> %res
+}
+
+define <2 x float> @gather_v2f32_scale_4(ptr %result, <2 x i64> %idx, <2 x i1> %mask) {
+; CHECK-LABEL: gather_v2f32_scale_4:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[0,2,2,3]
+; CHECK-NEXT:    vpslld $31, %xmm1, %xmm2
+; CHECK-NEXT:    vpxor %xmm1, %xmm1, %xmm1
+; CHECK-NEXT:    vgatherqps %xmm2, (%rdi,%xmm0,4), %xmm1
+; CHECK-NEXT:    vmovaps %xmm1, %xmm0
+; CHECK-NEXT:    retq
+  %gep = getelementptr inbounds [4 x i8], ptr %result, <2 x i64> %idx
+  %res = call <2 x float> @llvm.masked.gather.v2f32.v2p0(<2 x ptr> %gep, i32 0, <2 x i1> %mask, <2 x float> zeroinitializer)
+  ret <2 x float> %res
+}
+
+define <2 x float> @gather_v2f32_scale_3(ptr %result, <2 x i64> %idx, <2 x i1> %mask) {
+; CHECK-LABEL: gather_v2f32_scale_3:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vpaddq %xmm0, %xmm0, %xmm2
+; CHECK-NEXT:    vpaddq %xmm0, %xmm2, %xmm2
+; CHECK-NEXT:    vpshufd {{.*#+}} xmm0 = xmm1[0,2,2,3]
+; CHECK-NEXT:    vpslld $31, %xmm0, %xmm1
+; CHECK-NEXT:    vpxor %xmm0, %xmm0, %xmm0
+; CHECK-NEXT:    vgatherqps %xmm1, (%rdi,%xmm2), %xmm0
+; CHECK-NEXT:    retq
+  %gep = getelementptr inbounds [3 x i8], ptr %result, <2 x i64> %idx
+  %res = call <2 x float> @llvm.masked.gather.v2f32.v2p0(<2 x ptr> %gep, i32 0, <2 x i1> %mask, <2 x float> zeroinitializer)
+  ret <2 x float> %res
+}
+
+define <2 x float> @gather_v2f32_scale_1(ptr %result, <2 x i64> %idx, <2 x i1> %mask) {
+; CHECK-LABEL: gather_v2f32_scale_1:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[0,2,2,3]
+; CHECK-NEXT:    vpslld $31, %xmm1, %xmm2
+; CHECK-NEXT:    vpxor %xmm1, %xmm1, %xmm1
+; CHECK-NEXT:    vgatherqps %xmm2, (%rdi,%xmm0), %xmm1
+; CHECK-NEXT:    vmovaps %xmm1, %xmm0
+; CHECK-NEXT:    retq
+  %gep = getelementptr inbounds [1 x i8], ptr %result, <2 x i64> %idx
+  %res = call <2 x float> @llvm.masked.gather.v2f32.v2p0(<2 x ptr> %gep, i32 0, <2 x i1> %mask, <2 x float> zeroinitializer)
+  ret <2 x float> %res
+}
+
+declare <2 x float> @llvm.masked.gather.v2f32.v2p0(<2 x ptr>, i32 immarg, <2 x i1>, <2 x float>)
Index: llvm/test/CodeGen/X86/gather-scatter-opaque-ptr.ll
===================================================================
--- /dev/null
+++ llvm/test/CodeGen/X86/gather-scatter-opaque-ptr.ll
@@ -0,0 +1,178 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc -mtriple=x86_64-linux -mcpu=skylake-avx512 < %s | FileCheck %s
+
+define void @scatter_scale_512(ptr %result, <4 x i64> %idx, <4 x i1> %mask) {
+; CHECK-LABEL: scatter_scale_512:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vpslld $31, %xmm1, %xmm1
+; CHECK-NEXT:    vpmovd2m %xmm1, %k1
+; CHECK-NEXT:    vpsllq $9, %ymm0, %ymm0
+; CHECK-NEXT:    vpxor %xmm1, %xmm1, %xmm1
+; CHECK-NEXT:    vscatterqpd %ymm1, (%rdi,%ymm0) {%k1}
+; CHECK-NEXT:    vzeroupper
+; CHECK-NEXT:    retq
+  %gep = getelementptr inbounds [512 x i8], ptr %result, <4 x i64> %idx
+  call void @llvm.masked.scatter.v4f64.v4p0(<4 x double> zeroinitializer, <4 x ptr> %gep, i32 0, <4 x i1> %mask)
+  ret void
+}
+
+define void @scatter_scale_16(ptr %result, <4 x i64> %idx, <4 x i1> %mask) {
+; CHECK-LABEL: scatter_scale_16:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vpslld $31, %xmm1, %xmm1
+; CHECK-NEXT:    vpmovd2m %xmm1, %k1
+; CHECK-NEXT:    vpsllq $4, %ymm0, %ymm0
+; CHECK-NEXT:    vpxor %xmm1, %xmm1, %xmm1
+; CHECK-NEXT:    vscatterqpd %ymm1, (%rdi,%ymm0) {%k1}
+; CHECK-NEXT:    vzeroupper
+; CHECK-NEXT:    retq
+  %gep = getelementptr inbounds [16 x i8], ptr %result, <4 x i64> %idx
+  call void @llvm.masked.scatter.v4f64.v4p0(<4 x double> zeroinitializer, <4 x ptr> %gep, i32 0, <4 x i1> %mask)
+  ret void
+}
+
+define void @scatter_scale_8(ptr %result, <4 x i64> %idx, <4 x i1> %mask) {
+; CHECK-LABEL: scatter_scale_8:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vpslld $31, %xmm1, %xmm1
+; CHECK-NEXT:    vpmovd2m %xmm1, %k1
+; CHECK-NEXT:    vpxor %xmm1, %xmm1, %xmm1
+; CHECK-NEXT:    vscatterqpd %ymm1, (%rdi,%ymm0,8) {%k1}
+; CHECK-NEXT:    vzeroupper
+; CHECK-NEXT:    retq
+  %gep = getelementptr inbounds [8 x i8], ptr %result, <4 x i64> %idx
+  call void @llvm.masked.scatter.v4f64.v4p0(<4 x double> zeroinitializer, <4 x ptr> %gep, i32 0, <4 x i1> %mask)
+  ret void
+}
+
+define void @scatter_scale_4(ptr %result, <4 x i64> %idx, <4 x i1> %mask) {
+; CHECK-LABEL: scatter_scale_4:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vpslld $31, %xmm1, %xmm1
+; CHECK-NEXT:    vpmovd2m %xmm1, %k1
+; CHECK-NEXT:    vpxor %xmm1, %xmm1, %xmm1
+; CHECK-NEXT:    vscatterqpd %ymm1, (%rdi,%ymm0,4) {%k1}
+; CHECK-NEXT:    vzeroupper
+; CHECK-NEXT:    retq
+  %gep = getelementptr inbounds [4 x i8], ptr %result, <4 x i64> %idx
+  call void @llvm.masked.scatter.v4f64.v4p0(<4 x double> zeroinitializer, <4 x ptr> %gep, i32 0, <4 x i1> %mask)
+  ret void
+}
+
+define void @scatter_scale_3(ptr %result, <4 x i64> %idx, <4 x i1> %mask) {
+; CHECK-LABEL: scatter_scale_3:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vpslld $31, %xmm1, %xmm1
+; CHECK-NEXT:    vpmovd2m %xmm1, %k1
+; CHECK-NEXT:    vpaddq %ymm0, %ymm0, %ymm1
+; CHECK-NEXT:    vpaddq %ymm0, %ymm1, %ymm0
+; CHECK-NEXT:    vpxor %xmm1, %xmm1, %xmm1
+; CHECK-NEXT:    vscatterqpd %ymm1, (%rdi,%ymm0) {%k1}
+; CHECK-NEXT:    vzeroupper
+; CHECK-NEXT:    retq
+  %gep = getelementptr inbounds [3 x i8], ptr %result, <4 x i64> %idx
+  call void @llvm.masked.scatter.v4f64.v4p0(<4 x double> zeroinitializer, <4 x ptr> %gep, i32 0, <4 x i1> %mask)
+  ret void
+}
+
+define void @scatter_scale_1(ptr %result, <4 x i64> %idx, <4 x i1> %mask) {
+; CHECK-LABEL: scatter_scale_1:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vpslld $31, %xmm1, %xmm1
+; CHECK-NEXT:    vpmovd2m %xmm1, %k1
+; CHECK-NEXT:    vpxor %xmm1, %xmm1, %xmm1
+; CHECK-NEXT:    vscatterqpd %ymm1, (%rdi,%ymm0) {%k1}
+; CHECK-NEXT:    vzeroupper
+; CHECK-NEXT:    retq
+  %gep = getelementptr inbounds [1 x i8], ptr %result, <4 x i64> %idx
+  call void @llvm.masked.scatter.v4f64.v4p0(<4 x double> zeroinitializer, <4 x ptr> %gep, i32 0, <4 x i1> %mask)
+  ret void
+}
+
+define <4 x double> @gather_scale_512(ptr %result, <4 x i64> %idx, <4 x i1> %mask) {
+; CHECK-LABEL: gather_scale_512:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vpslld $31, %xmm1, %xmm1
+; CHECK-NEXT:    vpmovd2m %xmm1, %k1
+; CHECK-NEXT:    vpsllq $9, %ymm0, %ymm1
+; CHECK-NEXT:    vpxor %xmm0, %xmm0, %xmm0
+; CHECK-NEXT:    vgatherqpd (%rdi,%ymm1), %ymm0 {%k1}
+; CHECK-NEXT:    retq
+  %gep = getelementptr inbounds [512 x i8], ptr %result, <4 x i64> %idx
+  %res = call <4 x double> @llvm.masked.gather.v4f64.v4p0(<4 x ptr> %gep, i32 0, <4 x i1> %mask, <4 x double> zeroinitializer)
+  ret <4 x double> %res
+}
+
+define <4 x double> @gather_scale_16(ptr %result, <4 x i64> %idx, <4 x i1> %mask) {
+; CHECK-LABEL: gather_scale_16:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vpslld $31, %xmm1, %xmm1
+; CHECK-NEXT:    vpmovd2m %xmm1, %k1
+; CHECK-NEXT:    vpsllq $4, %ymm0, %ymm1
+; CHECK-NEXT:    vpxor %xmm0, %xmm0, %xmm0
+; CHECK-NEXT:    vgatherqpd (%rdi,%ymm1), %ymm0 {%k1}
+; CHECK-NEXT:    retq
+  %gep = getelementptr inbounds [16 x i8], ptr %result, <4 x i64> %idx
+  %res = call <4 x double> @llvm.masked.gather.v4f64.v4p0(<4 x ptr> %gep, i32 0, <4 x i1> %mask, <4 x double> zeroinitializer)
+  ret <4 x double> %res
+}
+
+define <4 x double> @gather_scale_8(ptr %result, <4 x i64> %idx, <4 x i1> %mask) {
+; CHECK-LABEL: gather_scale_8:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vpslld $31, %xmm1, %xmm1
+; CHECK-NEXT:    vpmovd2m %xmm1, %k1
+; CHECK-NEXT:    vpxor %xmm1, %xmm1, %xmm1
+; CHECK-NEXT:    vgatherqpd (%rdi,%ymm0,8), %ymm1 {%k1}
+; CHECK-NEXT:    vmovapd %ymm1, %ymm0
+; CHECK-NEXT:    retq
+  %gep = getelementptr inbounds [8 x i8], ptr %result, <4 x i64> %idx
+  %res = call <4 x double> @llvm.masked.gather.v4f64.v4p0(<4 x ptr> %gep, i32 0, <4 x i1> %mask, <4 x double> zeroinitializer)
+  ret <4 x double> %res
+}
+
+define <4 x double> @gather_scale_4(ptr %result, <4 x i64> %idx, <4 x i1> %mask) {
+; CHECK-LABEL: gather_scale_4:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vpslld $31, %xmm1, %xmm1
+; CHECK-NEXT:    vpmovd2m %xmm1, %k1
+; CHECK-NEXT:    vpxor %xmm1, %xmm1, %xmm1
+; CHECK-NEXT:    vgatherqpd (%rdi,%ymm0,4), %ymm1 {%k1}
+; CHECK-NEXT:    vmovapd %ymm1, %ymm0
+; CHECK-NEXT:    retq
+  %gep = getelementptr inbounds [4 x i8], ptr %result, <4 x i64> %idx
+  %res = call <4 x double> @llvm.masked.gather.v4f64.v4p0(<4 x ptr> %gep, i32 0, <4 x i1> %mask, <4 x double> zeroinitializer)
+  ret <4 x double> %res
+}
+
+define <4 x double> @gather_scale_3(ptr %result, <4 x i64> %idx, <4 x i1> %mask) {
+; CHECK-LABEL: gather_scale_3:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vpslld $31, %xmm1, %xmm1
+; CHECK-NEXT:    vpmovd2m %xmm1, %k1
+; CHECK-NEXT:    vpaddq %ymm0, %ymm0, %ymm1
+; CHECK-NEXT:    vpaddq %ymm0, %ymm1, %ymm1
+; CHECK-NEXT:    vpxor %xmm0, %xmm0, %xmm0
+; CHECK-NEXT:    vgatherqpd (%rdi,%ymm1), %ymm0 {%k1}
+; CHECK-NEXT:    retq
+  %gep = getelementptr inbounds [3 x i8], ptr %result, <4 x i64> %idx
+  %res = call <4 x double> @llvm.masked.gather.v4f64.v4p0(<4 x ptr> %gep, i32 0, <4 x i1> %mask, <4 x double> zeroinitializer)
+  ret <4 x double> %res
+}
+
+define <4 x double> @gather_scale_1(ptr %result, <4 x i64> %idx, <4 x i1> %mask) {
+; CHECK-LABEL: gather_scale_1:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vpslld $31, %xmm1, %xmm1
+; CHECK-NEXT:    vpmovd2m %xmm1, %k1
+; CHECK-NEXT:    vpxor %xmm1, %xmm1, %xmm1
+; CHECK-NEXT:    vgatherqpd (%rdi,%ymm0), %ymm1 {%k1}
+; CHECK-NEXT:    vmovapd %ymm1, %ymm0
+; CHECK-NEXT:    retq
+  %gep = getelementptr inbounds [1 x i8], ptr %result, <4 x i64> %idx
+  %res = call <4 x double> @llvm.masked.gather.v4f64.v4p0(<4 x ptr> %gep, i32 0, <4 x i1> %mask, <4 x double> zeroinitializer)
+  ret <4 x double> %res
+}
+
+declare void @llvm.masked.scatter.v4f64.v4p0(<4 x double>, <4 x ptr>, i32 immarg, <4 x i1>)
+declare <4 x double> @llvm.masked.gather.v4f64.v4p0(<4 x ptr>, i32 immarg, <4 x i1>, <4 x double>)