Index: llvm/lib/Target/ARM/ARMInstrMVE.td
===================================================================
--- llvm/lib/Target/ARM/ARMInstrMVE.td
+++ llvm/lib/Target/ARM/ARMInstrMVE.td
@@ -4382,6 +4382,10 @@
 // vector types (v4i1<>v8i1, etc.) also as part of lowering vector shuffles.
 def predicate_cast : SDNode<"ARMISD::PREDICATE_CAST", SDTUnaryOp>;
 
+def load_align4 : PatFrag<(ops node:$ptr), (load node:$ptr), [{
+  return cast<LoadSDNode>(N)->getAlignment() >= 4;
+}]>;
+
 let Predicates = [HasMVEInt] in {
   foreach VT = [ v4i1, v8i1, v16i1 ] in {
     def : Pat<(i32 (predicate_cast (VT VCCR:$src))),
@@ -4394,6 +4398,13 @@
                 (VT  (COPY_TO_REGCLASS (VT2 VCCR:$src), VCCR))>;
   }
 
+  // If we happen to be casting from a load we can convert that straight
+  // into a predicate load, so long as the load is of the correct type.
+  foreach VT = [ v4i1, v8i1, v16i1 ] in {
+    def : Pat<(VT (predicate_cast (i32 (load_align4 taddrmode_imm7<2>:$addr)))),
+              (VT (VLDR_P0_off taddrmode_imm7<2>:$addr))>;
+  }
+
   // Here we match the specific SDNode type 'ARMVectorRegCastImpl'
   // rather than the more general 'ARMVectorRegCast' which would also
   // match some bitconverts. If we use the latter in cases where the
Index: llvm/test/CodeGen/Thumb2/mve-pred-loadstore.ll
===================================================================
--- llvm/test/CodeGen/Thumb2/mve-pred-loadstore.ll
+++ llvm/test/CodeGen/Thumb2/mve-pred-loadstore.ll
@@ -386,18 +386,16 @@
 define arm_aapcs_vfpcc <4 x i32> @load_bc4(i32* %i, <4 x i32> %a) {
 ; CHECK-LE-LABEL: load_bc4:
 ; CHECK-LE:       @ %bb.0:
-; CHECK-LE-NEXT:    ldr r0, [r0]
+; CHECK-LE-NEXT:    vldr p0, [r0]
 ; CHECK-LE-NEXT:    vmov.i32 q1, #0x0
-; CHECK-LE-NEXT:    vmsr p0, r0
 ; CHECK-LE-NEXT:    vpsel q0, q0, q1
 ; CHECK-LE-NEXT:    bx lr
 ;
 ; CHECK-BE-LABEL: load_bc4:
 ; CHECK-BE:       @ %bb.0:
-; CHECK-BE-NEXT:    ldr r0, [r0]
+; CHECK-BE-NEXT:    vldr p0, [r0]
 ; CHECK-BE-NEXT:    vrev64.32 q1, q0
 ; CHECK-BE-NEXT:    vmov.i32 q0, #0x0
-; CHECK-BE-NEXT:    vmsr p0, r0
 ; CHECK-BE-NEXT:    vpsel q1, q1, q0
 ; CHECK-BE-NEXT:    vrev64.32 q0, q1
 ; CHECK-BE-NEXT:    bx lr
@@ -410,19 +408,17 @@
 define arm_aapcs_vfpcc <8 x i16> @load_predcast8(i32* %i, <8 x i16> %a) {
 ; CHECK-LE-LABEL: load_predcast8:
 ; CHECK-LE:       @ %bb.0:
-; CHECK-LE-NEXT:    ldr r0, [r0]
+; CHECK-LE-NEXT:    vldr p0, [r0]
 ; CHECK-LE-NEXT:    vmov.i32 q1, #0x0
-; CHECK-LE-NEXT:    vmsr p0, r0
 ; CHECK-LE-NEXT:    vpsel q0, q0, q1
 ; CHECK-LE-NEXT:    bx lr
 ;
 ; CHECK-BE-LABEL: load_predcast8:
 ; CHECK-BE:       @ %bb.0:
-; CHECK-BE-NEXT:    ldr r0, [r0]
 ; CHECK-BE-NEXT:    vrev64.16 q1, q0
 ; CHECK-BE-NEXT:    vmov.i32 q0, #0x0
+; CHECK-BE-NEXT:    vldr p0, [r0]
 ; CHECK-BE-NEXT:    vrev32.16 q0, q0
-; CHECK-BE-NEXT:    vmsr p0, r0
 ; CHECK-BE-NEXT:    vpsel q1, q1, q0
 ; CHECK-BE-NEXT:    vrev64.16 q0, q1
 ; CHECK-BE-NEXT:    bx lr
@@ -435,19 +431,17 @@
 define arm_aapcs_vfpcc <16 x i8> @load_predcast16(i32* %i, <16 x i8> %a) {
 ; CHECK-LE-LABEL: load_predcast16:
 ; CHECK-LE:       @ %bb.0:
-; CHECK-LE-NEXT:    ldr r0, [r0]
+; CHECK-LE-NEXT:    vldr p0, [r0]
 ; CHECK-LE-NEXT:    vmov.i32 q1, #0x0
-; CHECK-LE-NEXT:    vmsr p0, r0
 ; CHECK-LE-NEXT:    vpsel q0, q0, q1
 ; CHECK-LE-NEXT:    bx lr
 ;
 ; CHECK-BE-LABEL: load_predcast16:
 ; CHECK-BE:       @ %bb.0:
-; CHECK-BE-NEXT:    ldr r0, [r0]
 ; CHECK-BE-NEXT:    vrev64.8 q1, q0
 ; CHECK-BE-NEXT:    vmov.i32 q0, #0x0
+; CHECK-BE-NEXT:    vldr p0, [r0]
 ; CHECK-BE-NEXT:    vrev32.8 q0, q0
-; CHECK-BE-NEXT:    vmsr p0, r0
 ; CHECK-BE-NEXT:    vpsel q1, q1, q0
 ; CHECK-BE-NEXT:    vrev64.8 q0, q1
 ; CHECK-BE-NEXT:    bx lr
@@ -460,18 +454,16 @@
 define arm_aapcs_vfpcc <4 x i32> @load_bc4_align2(i32* %i, <4 x i32> %a) {
 ; CHECK-LE-LABEL: load_bc4_align2:
 ; CHECK-LE:       @ %bb.0:
-; CHECK-LE-NEXT:    ldr r0, [r0]
+; CHECK-LE-NEXT:    vldr p0, [r0]
 ; CHECK-LE-NEXT:    vmov.i32 q1, #0x0
-; CHECK-LE-NEXT:    vmsr p0, r0
 ; CHECK-LE-NEXT:    vpsel q0, q0, q1
 ; CHECK-LE-NEXT:    bx lr
 ;
 ; CHECK-BE-LABEL: load_bc4_align2:
 ; CHECK-BE:       @ %bb.0:
-; CHECK-BE-NEXT:    ldr r0, [r0]
+; CHECK-BE-NEXT:    vldr p0, [r0]
 ; CHECK-BE-NEXT:    vrev64.32 q1, q0
 ; CHECK-BE-NEXT:    vmov.i32 q0, #0x0
-; CHECK-BE-NEXT:    vmsr p0, r0
 ; CHECK-BE-NEXT:    vpsel q1, q1, q0
 ; CHECK-BE-NEXT:    vrev64.32 q0, q1
 ; CHECK-BE-NEXT:    bx lr
@@ -484,18 +476,18 @@
 define arm_aapcs_vfpcc <4 x i32> @load_bc4_offset(i16* %i, <4 x i32> %a) {
 ; CHECK-LE-LABEL: load_bc4_offset:
 ; CHECK-LE:       @ %bb.0:
-; CHECK-LE-NEXT:    ldr.w r0, [r0, #6]
+; CHECK-LE-NEXT:    adds r0, #6
 ; CHECK-LE-NEXT:    vmov.i32 q1, #0x0
-; CHECK-LE-NEXT:    vmsr p0, r0
+; CHECK-LE-NEXT:    vldr p0, [r0]
 ; CHECK-LE-NEXT:    vpsel q0, q0, q1
 ; CHECK-LE-NEXT:    bx lr
 ;
 ; CHECK-BE-LABEL: load_bc4_offset:
 ; CHECK-BE:       @ %bb.0:
-; CHECK-BE-NEXT:    ldr.w r0, [r0, #6]
+; CHECK-BE-NEXT:    adds r0, #6
 ; CHECK-BE-NEXT:    vrev64.32 q1, q0
+; CHECK-BE-NEXT:    vldr p0, [r0]
 ; CHECK-BE-NEXT:    vmov.i32 q0, #0x0
-; CHECK-BE-NEXT:    vmsr p0, r0
 ; CHECK-BE-NEXT:    vpsel q1, q1, q0
 ; CHECK-BE-NEXT:    vrev64.32 q0, q1
 ; CHECK-BE-NEXT:    bx lr
@@ -510,18 +502,16 @@
 define arm_aapcs_vfpcc <4 x i32> @load_bc4_range4(i32* %i, <4 x i32> %a) {
 ; CHECK-LE-LABEL: load_bc4_range4:
 ; CHECK-LE:       @ %bb.0:
-; CHECK-LE-NEXT:    ldr r0, [r0, #4]
+; CHECK-LE-NEXT:    vldr p0, [r0, #4]
 ; CHECK-LE-NEXT:    vmov.i32 q1, #0x0
-; CHECK-LE-NEXT:    vmsr p0, r0
 ; CHECK-LE-NEXT:    vpsel q0, q0, q1
 ; CHECK-LE-NEXT:    bx lr
 ;
 ; CHECK-BE-LABEL: load_bc4_range4:
 ; CHECK-BE:       @ %bb.0:
-; CHECK-BE-NEXT:    ldr r0, [r0, #4]
+; CHECK-BE-NEXT:    vldr p0, [r0, #4]
 ; CHECK-BE-NEXT:    vrev64.32 q1, q0
 ; CHECK-BE-NEXT:    vmov.i32 q0, #0x0
-; CHECK-BE-NEXT:    vmsr p0, r0
 ; CHECK-BE-NEXT:    vpsel q1, q1, q0
 ; CHECK-BE-NEXT:    vrev64.32 q0, q1
 ; CHECK-BE-NEXT:    bx lr
@@ -535,18 +525,16 @@
 define arm_aapcs_vfpcc <4 x i32> @load_bc4_range(i32* %i, <4 x i32> %a) {
 ; CHECK-LE-LABEL: load_bc4_range:
 ; CHECK-LE:       @ %bb.0:
-; CHECK-LE-NEXT:    ldr.w r0, [r0, #508]
+; CHECK-LE-NEXT:    vldr p0, [r0, #508]
 ; CHECK-LE-NEXT:    vmov.i32 q1, #0x0
-; CHECK-LE-NEXT:    vmsr p0, r0
 ; CHECK-LE-NEXT:    vpsel q0, q0, q1
 ; CHECK-LE-NEXT:    bx lr
 ;
 ; CHECK-BE-LABEL: load_bc4_range:
 ; CHECK-BE:       @ %bb.0:
-; CHECK-BE-NEXT:    ldr.w r0, [r0, #508]
+; CHECK-BE-NEXT:    vldr p0, [r0, #508]
 ; CHECK-BE-NEXT:    vrev64.32 q1, q0
 ; CHECK-BE-NEXT:    vmov.i32 q0, #0x0
-; CHECK-BE-NEXT:    vmsr p0, r0
 ; CHECK-BE-NEXT:    vpsel q1, q1, q0
 ; CHECK-BE-NEXT:    vrev64.32 q0, q1
 ; CHECK-BE-NEXT:    bx lr
@@ -560,22 +548,16 @@
 define arm_aapcs_vfpcc <4 x i32> @load_bc4_range2(i32* %i, <4 x i32> %a) {
 ; CHECK-LE-LABEL: load_bc4_range2:
 ; CHECK-LE:       @ %bb.0:
-; CHECK-LE-NEXT:    movw r1, #65028
+; CHECK-LE-NEXT:    vldr p0, [r0, #-508]
 ; CHECK-LE-NEXT:    vmov.i32 q1, #0x0
-; CHECK-LE-NEXT:    movt r1, #65535
-; CHECK-LE-NEXT:    ldr r0, [r0, r1]
-; CHECK-LE-NEXT:    vmsr p0, r0
 ; CHECK-LE-NEXT:    vpsel q0, q0, q1
 ; CHECK-LE-NEXT:    bx lr
 ;
 ; CHECK-BE-LABEL: load_bc4_range2:
 ; CHECK-BE:       @ %bb.0:
-; CHECK-BE-NEXT:    movw r1, #65028
+; CHECK-BE-NEXT:    vldr p0, [r0, #-508]
 ; CHECK-BE-NEXT:    vrev64.32 q1, q0
-; CHECK-BE-NEXT:    movt r1, #65535
 ; CHECK-BE-NEXT:    vmov.i32 q0, #0x0
-; CHECK-BE-NEXT:    ldr r0, [r0, r1]
-; CHECK-BE-NEXT:    vmsr p0, r0
 ; CHECK-BE-NEXT:    vpsel q1, q1, q0
 ; CHECK-BE-NEXT:    vrev64.32 q0, q1
 ; CHECK-BE-NEXT:    bx lr
@@ -589,18 +571,18 @@
 define arm_aapcs_vfpcc <4 x i32> @load_bc4_range3(i32* %i, <4 x i32> %a) {
 ; CHECK-LE-LABEL: load_bc4_range3:
 ; CHECK-LE:       @ %bb.0:
-; CHECK-LE-NEXT:    ldr.w r0, [r0, #512]
+; CHECK-LE-NEXT:    add.w r0, r0, #512
 ; CHECK-LE-NEXT:    vmov.i32 q1, #0x0
-; CHECK-LE-NEXT:    vmsr p0, r0
+; CHECK-LE-NEXT:    vldr p0, [r0]
 ; CHECK-LE-NEXT:    vpsel q0, q0, q1
 ; CHECK-LE-NEXT:    bx lr
 ;
 ; CHECK-BE-LABEL: load_bc4_range3:
 ; CHECK-BE:       @ %bb.0:
-; CHECK-BE-NEXT:    ldr.w r0, [r0, #512]
+; CHECK-BE-NEXT:    add.w r0, r0, #512
 ; CHECK-BE-NEXT:    vrev64.32 q1, q0
+; CHECK-BE-NEXT:    vldr p0, [r0]
 ; CHECK-BE-NEXT:    vmov.i32 q0, #0x0
-; CHECK-BE-NEXT:    vmsr p0, r0
 ; CHECK-BE-NEXT:    vpsel q1, q1, q0
 ; CHECK-BE-NEXT:    vrev64.32 q0, q1
 ; CHECK-BE-NEXT:    bx lr
@@ -614,22 +596,18 @@
 define arm_aapcs_vfpcc <4 x i32> @load_bc4_range5(i32* %i, <4 x i32> %a) {
 ; CHECK-LE-LABEL: load_bc4_range5:
 ; CHECK-LE:       @ %bb.0:
-; CHECK-LE-NEXT:    movw r1, #65024
+; CHECK-LE-NEXT:    sub.w r0, r0, #512
 ; CHECK-LE-NEXT:    vmov.i32 q1, #0x0
-; CHECK-LE-NEXT:    movt r1, #65535
-; CHECK-LE-NEXT:    ldr r0, [r0, r1]
-; CHECK-LE-NEXT:    vmsr p0, r0
+; CHECK-LE-NEXT:    vldr p0, [r0]
 ; CHECK-LE-NEXT:    vpsel q0, q0, q1
 ; CHECK-LE-NEXT:    bx lr
 ;
 ; CHECK-BE-LABEL: load_bc4_range5:
 ; CHECK-BE:       @ %bb.0:
-; CHECK-BE-NEXT:    movw r1, #65024
+; CHECK-BE-NEXT:    sub.w r0, r0, #512
 ; CHECK-BE-NEXT:    vrev64.32 q1, q0
-; CHECK-BE-NEXT:    movt r1, #65535
+; CHECK-BE-NEXT:    vldr p0, [r0]
 ; CHECK-BE-NEXT:    vmov.i32 q0, #0x0
-; CHECK-BE-NEXT:    ldr r0, [r0, r1]
-; CHECK-BE-NEXT:    vmsr p0, r0
 ; CHECK-BE-NEXT:    vpsel q1, q1, q0
 ; CHECK-BE-NEXT:    vrev64.32 q0, q1
 ; CHECK-BE-NEXT:    bx lr