Index: lib/CodeGen/GlobalISel/LegalizerHelper.cpp
===================================================================
--- lib/CodeGen/GlobalISel/LegalizerHelper.cpp
+++ lib/CodeGen/GlobalISel/LegalizerHelper.cpp
@@ -2538,7 +2538,8 @@
   MIRBuilder.setInstr(MI);
   unsigned Opc = MI.getOpcode();
   switch (Opc) {
-  case TargetOpcode::G_IMPLICIT_DEF: {
+  case TargetOpcode::G_IMPLICIT_DEF:
+  case TargetOpcode::G_LOAD: {
     Observer.changingInstr(MI);
     moreElementsVectorDst(MI, MoreTy, 0);
     Observer.changedInstr(MI);
Index: lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp
===================================================================
--- lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp
+++ lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp
@@ -482,6 +482,7 @@
       [](const LegalityQuery &Query) {
         return std::make_pair(0, LLT::scalar(32));
       })
+    .moreElementsIf(isSmallOddVector(0), oneMoreElement(0))
     .fewerElementsIf([=](const LegalityQuery &Query) {
         unsigned MemSize = Query.MMODescrs[0].SizeInBits;
         return (MemSize == 96) &&
Index: test/CodeGen/AMDGPU/GlobalISel/legalize-load.mir
===================================================================
--- test/CodeGen/AMDGPU/GlobalISel/legalize-load.mir
+++ test/CodeGen/AMDGPU/GlobalISel/legalize-load.mir
@@ -1,6 +1,6 @@
 # NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
-# RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=tahiti -O0 -run-pass=legalizer %s -o - | FileCheck -check-prefix=SI %s
-# RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=fiji -O0 -run-pass=legalizer %s -o - | FileCheck -check-prefix=VI %s
+# RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=tahiti -O0 -run-pass=legalizer -global-isel-abort=0 %s -o - | FileCheck -check-prefix=SI %s
+# RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=fiji -O0 -run-pass=legalizer -global-isel-abort=0  %s -o - | FileCheck -check-prefix=VI %s
 
 ---
 name: test_load_global_i32
@@ -390,3 +390,24 @@
     %1:_(<3 x s32>) = G_LOAD %0 :: (load 12, addrspace 1, align 16)
     $vgpr0_vgpr1_vgpr2 = COPY %1
 ...
+
+---
+name: test_load_global_v3s8_align4
+body: |
+  bb.0:
+    liveins: $vgpr0_vgpr1
+
+    ; SI-LABEL: name: test_load_global_v3s8_align4
+    ; SI: [[COPY:%[0-9]+]]:_(p1) = COPY $vgpr0_vgpr1
+    ; SI: [[LOAD:%[0-9]+]]:_(<4 x s8>) = G_LOAD [[COPY]](p1) :: (load 3, align 4, addrspace 1)
+    ; SI: [[EXTRACT:%[0-9]+]]:_(<3 x s8>) = G_EXTRACT [[LOAD]](<4 x s8>), 0
+    ; SI: S_NOP 0, implicit [[EXTRACT]](<3 x s8>)
+    ; VI-LABEL: name: test_load_global_v3s8_align4
+    ; VI: [[COPY:%[0-9]+]]:_(p1) = COPY $vgpr0_vgpr1
+    ; VI: [[LOAD:%[0-9]+]]:_(<4 x s8>) = G_LOAD [[COPY]](p1) :: (load 3, align 4, addrspace 1)
+    ; VI: [[EXTRACT:%[0-9]+]]:_(<3 x s8>) = G_EXTRACT [[LOAD]](<4 x s8>), 0
+    ; VI: S_NOP 0, implicit [[EXTRACT]](<3 x s8>)
+    %0:_(p1) = COPY $vgpr0_vgpr1
+    %1:_(<3 x s8>) = G_LOAD %0 :: (load 3, align 4, addrspace 1)
+    S_NOP 0, implicit %1
+...