diff --git a/llvm/lib/Target/AMDGPU/SIInstructions.td b/llvm/lib/Target/AMDGPU/SIInstructions.td
--- a/llvm/lib/Target/AMDGPU/SIInstructions.td
+++ b/llvm/lib/Target/AMDGPU/SIInstructions.td
@@ -2684,6 +2684,7 @@
 
 // COPY is workaround tablegen bug from multiple outputs
 // from S_LSHL_B32's multiple outputs from implicit scc def.
+let AddedComplexity = 1 in {
 def : GCNPat <
   (v2i16 (UniformBinFrag<build_vector> (i16 0), (i16 SReg_32:$src1))),
   (S_LSHL_B32 SReg_32:$src1, (i16 16))
@@ -2750,6 +2751,7 @@
   (v2f16 (DivergentBinFrag<build_vector> (f16 undef), (f16 SReg_32:$src1))),
   (v2f16 (V_LSHLREV_B32_e64 (i32 16), SReg_32:$src1))
 >;
+}
 
 let SubtargetPredicate = HasVOP3PInsts in {
 def : GCNPat <
@@ -2770,39 +2772,70 @@
 >;
 
 def : GCNPat <
-  (v2i16 (DivergentBinFrag<build_vector> (i16 SReg_32:$src0), (i16 (trunc (srl_oneuse SReg_32:$src1, (i32 16)))))),
-  (v2i16 (V_BFI_B32_e64 (i32 (V_MOV_B32_e32 (i32 0xffff))), SReg_32:$src0, SReg_32:$src1))
+  (v2i16 (UniformBinFrag<build_vector> (i16 (trunc (srl_oneuse SReg_32:$src0, (i32 16)))),
+                       (i16 (trunc (srl_oneuse SReg_32:$src1, (i32 16)))))),
+  (S_PACK_HH_B32_B16 SReg_32:$src0, SReg_32:$src1)
+>;
+
+def : GCNPat <
+  (v2f16 (UniformBinFrag<build_vector> (f16 SReg_32:$src0), (f16 SReg_32:$src1))),
+  (S_PACK_LL_B32_B16 SReg_32:$src0, SReg_32:$src1)
 >;
 
+foreach TypeList = [[i16,v2i16], [f16,v2f16]] in {
 
+defvar vecTy = TypeList[1];
+defvar Ty = TypeList[0];
+
+// Take the lower 16 bits from each VGPR_32 and concat them
 def : GCNPat <
-  (v2i16 (UniformBinFrag<build_vector> (i16 (trunc (srl_oneuse SReg_32:$src0, (i32 16)))),
-                       (i16 (trunc (srl_oneuse SReg_32:$src1, (i32 16)))))),
-  (S_PACK_HH_B32_B16 SReg_32:$src0, SReg_32:$src1)
+  (vecTy (DivergentBinFrag<build_vector> (Ty VGPR_32:$a), (Ty VGPR_32:$b))),
+  (V_PERM_B32_e64 VGPR_32:$b, VGPR_32:$a, (S_MOV_B32 (i32 0x05040100)))
 >;
 
+// Take the lower 16 bits from V[0] and the upper 16 bits from V[1]
 def : GCNPat <
-  (v2i16 (DivergentBinFrag<build_vector> (i16 (trunc (srl_oneuse SReg_32:$src0, (i32 16)))),
-                       (i16 (trunc (srl_oneuse SReg_32:$src1, (i32 16)))))),
-  (v2i16 (V_AND_OR_B32_e64 SReg_32:$src1, (i32 (V_MOV_B32_e32 (i32 0xffff0000))), (i32 (V_LSHRREV_B32_e64 (i32 16), SReg_32:$src0))))
+  (vecTy (DivergentBinFrag<build_vector> (Ty VGPR_32:$a),
+    (Ty !if(!eq(Ty, i16),
+      (Ty (trunc (srl VGPR_32:$b, (i32 16)))),
+      (Ty (bitconvert (i16 (trunc (srl VGPR_32:$b, (i32 16)))))))))),
+  (V_PERM_B32_e64 VGPR_32:$b, VGPR_32:$a, (S_MOV_B32 (i32 0x07060100)))
 >;
 
+// Take the upper 16 bits from V[0] and the lower 16 bits from V[1]
+// Special case -- V_ALIGNBIT
 def : GCNPat <
-  (v2f16 (UniformBinFrag<build_vector> (f16 SReg_32:$src0), (f16 SReg_32:$src1))),
-  (S_PACK_LL_B32_B16 SReg_32:$src0, SReg_32:$src1)
+  (vecTy (DivergentBinFrag<build_vector>
+    (Ty !if(!eq(Ty, i16),
+      (Ty (trunc (srl VGPR_32:$a, (i32 16)))),
+      (Ty (bitconvert (i16 (trunc (srl VGPR_32:$a, (i32 16)))))))),
+    (Ty VGPR_32:$b))),
+    (V_ALIGNBIT_B32_e64 VGPR_32:$b, VGPR_32:$a, (i32 16))
 >;
 
+// Take the upper 16 bits from each VGPR_32 and concat them
 def : GCNPat <
-  (v2f16 (DivergentBinFrag<build_vector> (f16 SReg_32:$src0), (f16 SReg_32:$src1))),
-  (v2f16 (V_LSHL_OR_B32_e64 SReg_32:$src1, (i32 16), (i32 (V_AND_B32_e64 (i32 (V_MOV_B32_e32 (i32 0xffff))), SReg_32:$src0))))
+  (vecTy (DivergentBinFrag<build_vector>
+    (Ty !if(!eq(Ty, i16),
+      (Ty (trunc (srl VGPR_32:$a, (i32 16)))),
+      (Ty (bitconvert (i16 (trunc (srl VGPR_32:$a, (i32 16)))))))),
+    (Ty !if(!eq(Ty, i16),
+      (Ty (trunc (srl VGPR_32:$b, (i32 16)))),
+      (Ty (bitconvert (i16 (trunc (srl VGPR_32:$b, (i32 16)))))))))),
+  (V_PERM_B32_e64 VGPR_32:$b, VGPR_32:$a, (S_MOV_B32 (i32 0x07060302)))
 >;
 
 
+} // end foreach Ty
+
+
+let AddedComplexity = 5 in {
 def : GCNPat <
   (v2f16 (is_canonicalized<build_vector> (f16 (VOP3Mods (f16 VGPR_32:$src0), i32:$src0_mods)),
                                          (f16 (VOP3Mods (f16 VGPR_32:$src1), i32:$src1_mods)))),
   (V_PACK_B32_F16_e64 $src0_mods, VGPR_32:$src0, $src1_mods, VGPR_32:$src1)
 >;
+}
 } // End SubtargetPredicate = HasVOP3PInsts
 
 // With multiple uses of the shift, this will duplicate the shift and
diff --git a/llvm/test/CodeGen/AMDGPU/add.v2i16.ll b/llvm/test/CodeGen/AMDGPU/add.v2i16.ll
--- a/llvm/test/CodeGen/AMDGPU/add.v2i16.ll
+++ b/llvm/test/CodeGen/AMDGPU/add.v2i16.ll
@@ -620,7 +620,6 @@
 ; GFX9-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
 ; GFX9-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x34
 ; GFX9-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
-; GFX9-NEXT:    v_mov_b32_e32 v4, 0xffff
 ; GFX9-NEXT:    v_mov_b32_e32 v1, 0
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    global_load_dword v2, v0, s[6:7] glc
@@ -629,9 +628,9 @@
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    s_mov_b32 s7, 0xf000
 ; GFX9-NEXT:    s_mov_b32 s6, -1
-; GFX9-NEXT:    v_pk_add_u16 v2, v2, v3
-; GFX9-NEXT:    v_and_b32_e32 v0, 0xffff, v2
-; GFX9-NEXT:    v_and_b32_sdwa v2, v4, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
+; GFX9-NEXT:    v_pk_add_u16 v0, v2, v3
+; GFX9-NEXT:    v_alignbit_b32 v2, 0, v0, 16
+; GFX9-NEXT:    v_and_b32_e32 v0, 0xffff, v0
 ; GFX9-NEXT:    v_mov_b32_e32 v3, v1
 ; GFX9-NEXT:    buffer_store_dwordx4 v[0:3], off, s[4:7], 0
 ; GFX9-NEXT:    s_endpgm
@@ -642,7 +641,6 @@
 ; GFX10-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
 ; GFX10-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x34
 ; GFX10-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
-; GFX10-NEXT:    v_mov_b32_e32 v3, 0xffff
 ; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-NEXT:    global_load_dword v1, v0, s[6:7] glc dlc
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
@@ -651,10 +649,10 @@
 ; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
 ; GFX10-NEXT:    s_mov_b32 s7, 0x31016000
 ; GFX10-NEXT:    s_mov_b32 s6, -1
-; GFX10-NEXT:    v_pk_add_u16 v2, v1, v2
+; GFX10-NEXT:    v_pk_add_u16 v0, v1, v2
 ; GFX10-NEXT:    v_mov_b32_e32 v1, 0
-; GFX10-NEXT:    v_and_b32_e32 v0, 0xffff, v2
-; GFX10-NEXT:    v_and_b32_sdwa v2, v3, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
+; GFX10-NEXT:    v_alignbit_b32 v2, 0, v0, 16
+; GFX10-NEXT:    v_and_b32_e32 v0, 0xffff, v0
 ; GFX10-NEXT:    v_mov_b32_e32 v3, v1
 ; GFX10-NEXT:    buffer_store_dwordx4 v[0:3], off, s[4:7], 0
 ; GFX10-NEXT:    s_endpgm
diff --git a/llvm/test/CodeGen/AMDGPU/build-vector-packed-partial-undef.ll b/llvm/test/CodeGen/AMDGPU/build-vector-packed-partial-undef.ll
--- a/llvm/test/CodeGen/AMDGPU/build-vector-packed-partial-undef.ll
+++ b/llvm/test/CodeGen/AMDGPU/build-vector-packed-partial-undef.ll
@@ -152,9 +152,8 @@
 ; GFX9-LABEL: undef_lo2_v4i16:
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX9-NEXT:    v_lshrrev_b32_e32 v1, 16, v0
-; GFX9-NEXT:    v_mov_b32_e32 v2, 0xffff0000
-; GFX9-NEXT:    v_and_or_b32 v0, v0, v2, v1
+; GFX9-NEXT:    s_mov_b32 s4, 0x7060302
+; GFX9-NEXT:    v_perm_b32 v0, v0, v0, s4
 ; GFX9-NEXT:    ;;#ASMSTART
 ; GFX9-NEXT:    ; use v[0:1]
 ; GFX9-NEXT:    ;;#ASMEND
@@ -178,9 +177,8 @@
 ; GFX9-LABEL: undef_lo2_v4f16:
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX9-NEXT:    v_lshrrev_b32_e32 v0, 16, v0
-; GFX9-NEXT:    v_and_b32_e32 v1, 0xffff, v0
-; GFX9-NEXT:    v_lshl_or_b32 v0, v0, 16, v1
+; GFX9-NEXT:    s_mov_b32 s4, 0x7060302
+; GFX9-NEXT:    v_perm_b32 v0, v0, v0, s4
 ; GFX9-NEXT:    ;;#ASMSTART
 ; GFX9-NEXT:    ; use v[0:1]
 ; GFX9-NEXT:    ;;#ASMEND
diff --git a/llvm/test/CodeGen/AMDGPU/chain-hi-to-lo.ll b/llvm/test/CodeGen/AMDGPU/chain-hi-to-lo.ll
--- a/llvm/test/CodeGen/AMDGPU/chain-hi-to-lo.ll
+++ b/llvm/test/CodeGen/AMDGPU/chain-hi-to-lo.ll
@@ -474,6 +474,7 @@
 ; GFX900-NEXT:    buffer_store_short v0, off, s[0:3], 0 offset:6
 ; GFX900-NEXT:    s_waitcnt vmcnt(0)
 ; GFX900-NEXT:    global_load_ushort v0, v2, s[4:5] offset:4
+; GFX900-NEXT:    s_mov_b32 s4, 0x5040100
 ; GFX900-NEXT:    s_waitcnt vmcnt(0)
 ; GFX900-NEXT:    buffer_store_short v0, off, s[0:3], 0 offset:8
 ; GFX900-NEXT:    s_waitcnt vmcnt(0)
@@ -483,8 +484,7 @@
 ; GFX900-NEXT:    v_mov_b32_e32 v1, v0
 ; GFX900-NEXT:    buffer_load_short_d16_hi v1, off, s[0:3], 0 offset:8
 ; GFX900-NEXT:    s_waitcnt vmcnt(1)
-; GFX900-NEXT:    v_and_b32_e32 v3, 0xffff, v3
-; GFX900-NEXT:    v_lshl_or_b32 v0, v0, 16, v3
+; GFX900-NEXT:    v_perm_b32 v0, v0, v3, s4
 ; GFX900-NEXT:    s_waitcnt vmcnt(0)
 ; GFX900-NEXT:    global_store_dwordx2 v2, v[0:1], s[6:7]
 ; GFX900-NEXT:    s_endpgm
@@ -544,9 +544,8 @@
 ; GFX10_DEFAULT-NEXT:    s_waitcnt vmcnt(1)
 ; GFX10_DEFAULT-NEXT:    v_mov_b32_e32 v1, v0
 ; GFX10_DEFAULT-NEXT:    s_waitcnt vmcnt(0)
-; GFX10_DEFAULT-NEXT:    v_and_b32_e32 v3, 0xffff, v3
+; GFX10_DEFAULT-NEXT:    v_perm_b32 v0, v0, v3, 0x5040100
 ; GFX10_DEFAULT-NEXT:    buffer_load_short_d16_hi v1, off, s[0:3], 0 offset:8
-; GFX10_DEFAULT-NEXT:    v_lshl_or_b32 v0, v0, 16, v3
 ; GFX10_DEFAULT-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10_DEFAULT-NEXT:    global_store_dwordx2 v2, v[0:1], s[6:7]
 ; GFX10_DEFAULT-NEXT:    s_endpgm
@@ -687,16 +686,27 @@
 
 ; The volatile operations aren't put on the same chain
 define <2 x i16> @chain_hi_to_lo_group_other_dep_multi_chain(i16 addrspace(3)* %ptr) {
-; GCN-LABEL: chain_hi_to_lo_group_other_dep_multi_chain:
-; GCN:       ; %bb.0: ; %bb
-; GCN-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GCN-NEXT:    ds_read_u16 v1, v0 offset:2
-; GCN-NEXT:    ds_read_u16_d16_hi v0, v0
-; GCN-NEXT:    v_mov_b32_e32 v2, 0xffff
-; GCN-NEXT:    s_waitcnt lgkmcnt(0)
-; GCN-NEXT:    v_pk_sub_u16 v0, v0, -12 op_sel_hi:[1,0]
-; GCN-NEXT:    v_bfi_b32 v0, v2, v1, v0
-; GCN-NEXT:    s_setpc_b64 s[30:31]
+; GFX900-LABEL: chain_hi_to_lo_group_other_dep_multi_chain:
+; GFX900:       ; %bb.0: ; %bb
+; GFX900-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX900-NEXT:    ds_read_u16 v1, v0 offset:2
+; GFX900-NEXT:    ds_read_u16_d16_hi v0, v0
+; GFX900-NEXT:    s_mov_b32 s4, 0x7060100
+; GFX900-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX900-NEXT:    v_pk_sub_u16 v0, v0, -12 op_sel_hi:[1,0]
+; GFX900-NEXT:    v_perm_b32 v0, v0, v1, s4
+; GFX900-NEXT:    s_setpc_b64 s[30:31]
+;
+; FLATSCR-LABEL: chain_hi_to_lo_group_other_dep_multi_chain:
+; FLATSCR:       ; %bb.0: ; %bb
+; FLATSCR-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; FLATSCR-NEXT:    ds_read_u16 v1, v0 offset:2
+; FLATSCR-NEXT:    ds_read_u16_d16_hi v0, v0
+; FLATSCR-NEXT:    s_mov_b32 s0, 0x7060100
+; FLATSCR-NEXT:    s_waitcnt lgkmcnt(0)
+; FLATSCR-NEXT:    v_pk_sub_u16 v0, v0, -12 op_sel_hi:[1,0]
+; FLATSCR-NEXT:    v_perm_b32 v0, v0, v1, s0
+; FLATSCR-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX10-LABEL: chain_hi_to_lo_group_other_dep_multi_chain:
 ; GFX10:       ; %bb.0: ; %bb
@@ -706,7 +716,7 @@
 ; GFX10-NEXT:    ds_read_u16_d16_hi v0, v0
 ; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-NEXT:    v_pk_sub_u16 v0, v0, -12 op_sel_hi:[1,0]
-; GFX10-NEXT:    v_bfi_b32 v0, 0xffff, v1, v0
+; GFX10-NEXT:    v_perm_b32 v0, v0, v1, 0x7060100
 ; GFX10-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX11-LABEL: chain_hi_to_lo_group_other_dep_multi_chain:
@@ -718,7 +728,7 @@
 ; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-NEXT:    v_pk_sub_u16 v0, v0, -12 op_sel_hi:[1,0]
 ; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
-; GFX11-NEXT:    v_bfi_b32 v0, 0xffff, v1, v0
+; GFX11-NEXT:    v_perm_b32 v0, v0, v1, 0x7060100
 ; GFX11-NEXT:    s_setpc_b64 s[30:31]
 bb:
   %gep_lo = getelementptr inbounds i16, i16 addrspace(3)* %ptr, i64 1
@@ -801,17 +811,29 @@
 }
 
 define <2 x i16> @chain_hi_to_lo_global_other_dep(i16 addrspace(1)* %ptr) {
-; GCN-LABEL: chain_hi_to_lo_global_other_dep:
-; GCN:       ; %bb.0: ; %bb
-; GCN-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GCN-NEXT:    global_load_ushort v2, v[0:1], off offset:2 glc
-; GCN-NEXT:    s_waitcnt vmcnt(0)
-; GCN-NEXT:    global_load_short_d16_hi v0, v[0:1], off glc
-; GCN-NEXT:    s_waitcnt vmcnt(0)
-; GCN-NEXT:    v_mov_b32_e32 v1, 0xffff
-; GCN-NEXT:    v_pk_sub_u16 v0, v0, -12 op_sel_hi:[1,0]
-; GCN-NEXT:    v_bfi_b32 v0, v1, v2, v0
-; GCN-NEXT:    s_setpc_b64 s[30:31]
+; GFX900-LABEL: chain_hi_to_lo_global_other_dep:
+; GFX900:       ; %bb.0: ; %bb
+; GFX900-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX900-NEXT:    global_load_ushort v2, v[0:1], off offset:2 glc
+; GFX900-NEXT:    s_waitcnt vmcnt(0)
+; GFX900-NEXT:    global_load_short_d16_hi v0, v[0:1], off glc
+; GFX900-NEXT:    s_waitcnt vmcnt(0)
+; GFX900-NEXT:    s_mov_b32 s4, 0x7060100
+; GFX900-NEXT:    v_pk_sub_u16 v0, v0, -12 op_sel_hi:[1,0]
+; GFX900-NEXT:    v_perm_b32 v0, v0, v2, s4
+; GFX900-NEXT:    s_setpc_b64 s[30:31]
+;
+; FLATSCR-LABEL: chain_hi_to_lo_global_other_dep:
+; FLATSCR:       ; %bb.0: ; %bb
+; FLATSCR-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; FLATSCR-NEXT:    global_load_ushort v2, v[0:1], off offset:2 glc
+; FLATSCR-NEXT:    s_waitcnt vmcnt(0)
+; FLATSCR-NEXT:    global_load_short_d16_hi v0, v[0:1], off glc
+; FLATSCR-NEXT:    s_waitcnt vmcnt(0)
+; FLATSCR-NEXT:    s_mov_b32 s0, 0x7060100
+; FLATSCR-NEXT:    v_pk_sub_u16 v0, v0, -12 op_sel_hi:[1,0]
+; FLATSCR-NEXT:    v_perm_b32 v0, v0, v2, s0
+; FLATSCR-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX10-LABEL: chain_hi_to_lo_global_other_dep:
 ; GFX10:       ; %bb.0: ; %bb
@@ -822,7 +844,7 @@
 ; GFX10-NEXT:    global_load_short_d16_hi v0, v[0:1], off glc dlc
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    v_pk_sub_u16 v0, v0, -12 op_sel_hi:[1,0]
-; GFX10-NEXT:    v_bfi_b32 v0, 0xffff, v2, v0
+; GFX10-NEXT:    v_perm_b32 v0, v0, v2, 0x7060100
 ; GFX10-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX11-LABEL: chain_hi_to_lo_global_other_dep:
@@ -835,7 +857,7 @@
 ; GFX11-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-NEXT:    v_pk_sub_u16 v0, v0, -12 op_sel_hi:[1,0]
 ; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
-; GFX11-NEXT:    v_bfi_b32 v0, 0xffff, v2, v0
+; GFX11-NEXT:    v_perm_b32 v0, v0, v2, 0x7060100
 ; GFX11-NEXT:    s_setpc_b64 s[30:31]
 bb:
   %gep_lo = getelementptr inbounds i16, i16 addrspace(1)* %ptr, i64 1
@@ -849,18 +871,31 @@
 }
 
 define <2 x i16> @chain_hi_to_lo_flat_other_dep(i16 addrspace(0)* %ptr) {
-; GCN-LABEL: chain_hi_to_lo_flat_other_dep:
-; GCN:       ; %bb.0: ; %bb
-; GCN-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GCN-NEXT:    flat_load_ushort v2, v[0:1] offset:2 glc
-; GCN-NEXT:    s_waitcnt vmcnt(0)
-; GCN-NEXT:    flat_load_short_d16_hi v0, v[0:1] glc
-; GCN-NEXT:    s_waitcnt vmcnt(0)
-; GCN-NEXT:    v_mov_b32_e32 v1, 0xffff
-; GCN-NEXT:    s_waitcnt lgkmcnt(0)
-; GCN-NEXT:    v_pk_sub_u16 v0, v0, -12 op_sel_hi:[1,0]
-; GCN-NEXT:    v_bfi_b32 v0, v1, v2, v0
-; GCN-NEXT:    s_setpc_b64 s[30:31]
+; GFX900-LABEL: chain_hi_to_lo_flat_other_dep:
+; GFX900:       ; %bb.0: ; %bb
+; GFX900-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX900-NEXT:    flat_load_ushort v2, v[0:1] offset:2 glc
+; GFX900-NEXT:    s_waitcnt vmcnt(0)
+; GFX900-NEXT:    flat_load_short_d16_hi v0, v[0:1] glc
+; GFX900-NEXT:    s_waitcnt vmcnt(0)
+; GFX900-NEXT:    s_mov_b32 s4, 0x7060100
+; GFX900-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX900-NEXT:    v_pk_sub_u16 v0, v0, -12 op_sel_hi:[1,0]
+; GFX900-NEXT:    v_perm_b32 v0, v0, v2, s4
+; GFX900-NEXT:    s_setpc_b64 s[30:31]
+;
+; FLATSCR-LABEL: chain_hi_to_lo_flat_other_dep:
+; FLATSCR:       ; %bb.0: ; %bb
+; FLATSCR-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; FLATSCR-NEXT:    flat_load_ushort v2, v[0:1] offset:2 glc
+; FLATSCR-NEXT:    s_waitcnt vmcnt(0)
+; FLATSCR-NEXT:    flat_load_short_d16_hi v0, v[0:1] glc
+; FLATSCR-NEXT:    s_waitcnt vmcnt(0)
+; FLATSCR-NEXT:    s_mov_b32 s0, 0x7060100
+; FLATSCR-NEXT:    s_waitcnt lgkmcnt(0)
+; FLATSCR-NEXT:    v_pk_sub_u16 v0, v0, -12 op_sel_hi:[1,0]
+; FLATSCR-NEXT:    v_perm_b32 v0, v0, v2, s0
+; FLATSCR-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX10-LABEL: chain_hi_to_lo_flat_other_dep:
 ; GFX10:       ; %bb.0: ; %bb
@@ -873,7 +908,7 @@
 ; GFX10-NEXT:    flat_load_short_d16_hi v0, v[0:1] glc dlc
 ; GFX10-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-NEXT:    v_pk_sub_u16 v0, v0, -12 op_sel_hi:[1,0]
-; GFX10-NEXT:    v_bfi_b32 v0, 0xffff, v2, v0
+; GFX10-NEXT:    v_perm_b32 v0, v0, v2, 0x7060100
 ; GFX10-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX11-LABEL: chain_hi_to_lo_flat_other_dep:
@@ -886,7 +921,7 @@
 ; GFX11-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX11-NEXT:    v_pk_sub_u16 v0, v0, -12 op_sel_hi:[1,0]
 ; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
-; GFX11-NEXT:    v_bfi_b32 v0, 0xffff, v2, v0
+; GFX11-NEXT:    v_perm_b32 v0, v0, v2, 0x7060100
 ; GFX11-NEXT:    s_setpc_b64 s[30:31]
 bb:
   %gep_lo = getelementptr inbounds i16, i16 addrspace(0)* %ptr, i64 1
@@ -900,17 +935,29 @@
 }
 
 define <2 x i16> @chain_hi_to_lo_group_may_alias_store(i16 addrspace(3)* %ptr, i16 addrspace(3)* %may.alias) {
-; GCN-LABEL: chain_hi_to_lo_group_may_alias_store:
-; GCN:       ; %bb.0: ; %bb
-; GCN-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GCN-NEXT:    v_mov_b32_e32 v3, 0x7b
-; GCN-NEXT:    ds_read_u16 v2, v0
-; GCN-NEXT:    ds_write_b16 v1, v3
-; GCN-NEXT:    ds_read_u16 v0, v0 offset:2
-; GCN-NEXT:    s_waitcnt lgkmcnt(0)
-; GCN-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GCN-NEXT:    v_lshl_or_b32 v0, v2, 16, v0
-; GCN-NEXT:    s_setpc_b64 s[30:31]
+; GFX900-LABEL: chain_hi_to_lo_group_may_alias_store:
+; GFX900:       ; %bb.0: ; %bb
+; GFX900-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX900-NEXT:    v_mov_b32_e32 v3, 0x7b
+; GFX900-NEXT:    ds_read_u16 v2, v0
+; GFX900-NEXT:    ds_write_b16 v1, v3
+; GFX900-NEXT:    ds_read_u16 v0, v0 offset:2
+; GFX900-NEXT:    s_mov_b32 s4, 0x5040100
+; GFX900-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX900-NEXT:    v_perm_b32 v0, v2, v0, s4
+; GFX900-NEXT:    s_setpc_b64 s[30:31]
+;
+; FLATSCR-LABEL: chain_hi_to_lo_group_may_alias_store:
+; FLATSCR:       ; %bb.0: ; %bb
+; FLATSCR-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; FLATSCR-NEXT:    v_mov_b32_e32 v3, 0x7b
+; FLATSCR-NEXT:    ds_read_u16 v2, v0
+; FLATSCR-NEXT:    ds_write_b16 v1, v3
+; FLATSCR-NEXT:    ds_read_u16 v0, v0 offset:2
+; FLATSCR-NEXT:    s_mov_b32 s0, 0x5040100
+; FLATSCR-NEXT:    s_waitcnt lgkmcnt(0)
+; FLATSCR-NEXT:    v_perm_b32 v0, v2, v0, s0
+; FLATSCR-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX10-LABEL: chain_hi_to_lo_group_may_alias_store:
 ; GFX10:       ; %bb.0: ; %bb
@@ -921,8 +968,7 @@
 ; GFX10-NEXT:    ds_write_b16 v1, v2
 ; GFX10-NEXT:    ds_read_u16 v0, v0 offset:2
 ; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX10-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX10-NEXT:    v_lshl_or_b32 v0, v3, 16, v0
+; GFX10-NEXT:    v_perm_b32 v0, v3, v0, 0x5040100
 ; GFX10-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX11-LABEL: chain_hi_to_lo_group_may_alias_store:
@@ -934,9 +980,7 @@
 ; GFX11-NEXT:    ds_store_b16 v1, v2
 ; GFX11-NEXT:    ds_load_u16 v0, v0 offset:2
 ; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX11-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
-; GFX11-NEXT:    v_lshl_or_b32 v0, v3, 16, v0
+; GFX11-NEXT:    v_perm_b32 v0, v3, v0, 0x5040100
 ; GFX11-NEXT:    s_setpc_b64 s[30:31]
 bb:
   %gep_lo = getelementptr inbounds i16, i16 addrspace(3)* %ptr, i64 1
diff --git a/llvm/test/CodeGen/AMDGPU/divergence-driven-buildvector.ll b/llvm/test/CodeGen/AMDGPU/divergence-driven-buildvector.ll
--- a/llvm/test/CodeGen/AMDGPU/divergence-driven-buildvector.ll
+++ b/llvm/test/CodeGen/AMDGPU/divergence-driven-buildvector.ll
@@ -262,15 +262,15 @@
 ; GFX9-LABEL: divergent_vec_i16_LL:
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX9-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX9-NEXT:    v_lshl_or_b32 v0, v1, 16, v0
+; GFX9-NEXT:    s_mov_b32 s4, 0x5040100
+; GFX9-NEXT:    v_perm_b32 v0, v1, v0, s4
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX906-LABEL: divergent_vec_i16_LL:
 ; GFX906:       ; %bb.0:
 ; GFX906-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX906-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX906-NEXT:    v_lshl_or_b32 v0, v1, 16, v0
+; GFX906-NEXT:    s_mov_b32 s4, 0x5040100
+; GFX906-NEXT:    v_perm_b32 v0, v1, v0, s4
 ; GFX906-NEXT:    s_setpc_b64 s[30:31]
   %tmp = insertelement <2 x i16> undef, i16 %a, i32 0
   %vec = insertelement <2 x i16> %tmp, i16 %b, i32 1
@@ -333,15 +333,15 @@
 ; GFX9-LABEL: divergent_vec_i16_LH:
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX9-NEXT:    v_mov_b32_e32 v2, 0xffff
-; GFX9-NEXT:    v_bfi_b32 v0, v2, v0, v1
+; GFX9-NEXT:    s_mov_b32 s4, 0x7060100
+; GFX9-NEXT:    v_perm_b32 v0, v1, v0, s4
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX906-LABEL: divergent_vec_i16_LH:
 ; GFX906:       ; %bb.0:
 ; GFX906-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX906-NEXT:    v_mov_b32_e32 v2, 0xffff
-; GFX906-NEXT:    v_bfi_b32 v0, v2, v0, v1
+; GFX906-NEXT:    s_mov_b32 s4, 0x7060100
+; GFX906-NEXT:    v_perm_b32 v0, v1, v0, s4
 ; GFX906-NEXT:    s_setpc_b64 s[30:31]
   %shift = lshr i32 %b, 16
   %tr = trunc i32 %shift to i16
@@ -409,17 +409,15 @@
 ; GFX9-LABEL: divergent_vec_i16_HH:
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX9-NEXT:    v_lshrrev_b32_e32 v0, 16, v0
-; GFX9-NEXT:    v_mov_b32_e32 v2, 0xffff0000
-; GFX9-NEXT:    v_and_or_b32 v0, v1, v2, v0
+; GFX9-NEXT:    s_mov_b32 s4, 0x7060302
+; GFX9-NEXT:    v_perm_b32 v0, v1, v0, s4
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX906-LABEL: divergent_vec_i16_HH:
 ; GFX906:       ; %bb.0:
 ; GFX906-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX906-NEXT:    v_lshrrev_b32_e32 v0, 16, v0
-; GFX906-NEXT:    v_mov_b32_e32 v2, 0xffff0000
-; GFX906-NEXT:    v_and_or_b32 v0, v1, v2, v0
+; GFX906-NEXT:    s_mov_b32 s4, 0x7060302
+; GFX906-NEXT:    v_perm_b32 v0, v1, v0, s4
 ; GFX906-NEXT:    s_setpc_b64 s[30:31]
   %shift_a = lshr i32 %a, 16
   %tr_a = trunc i32 %shift_a to i16
@@ -499,15 +497,15 @@
 ; GFX9-LABEL: divergent_vec_f16_LL:
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX9-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX9-NEXT:    v_lshl_or_b32 v0, v1, 16, v0
+; GFX9-NEXT:    s_mov_b32 s4, 0x5040100
+; GFX9-NEXT:    v_perm_b32 v0, v1, v0, s4
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX906-LABEL: divergent_vec_f16_LL:
 ; GFX906:       ; %bb.0:
 ; GFX906-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX906-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX906-NEXT:    v_lshl_or_b32 v0, v1, 16, v0
+; GFX906-NEXT:    s_mov_b32 s4, 0x5040100
+; GFX906-NEXT:    v_perm_b32 v0, v1, v0, s4
 ; GFX906-NEXT:    s_setpc_b64 s[30:31]
   %tmp = insertelement <2 x half> undef, half %a, i32 0
   %vec = insertelement <2 x half> %tmp, half %b, i32 1
diff --git a/llvm/test/CodeGen/AMDGPU/extract-subvector-16bit.ll b/llvm/test/CodeGen/AMDGPU/extract-subvector-16bit.ll
--- a/llvm/test/CodeGen/AMDGPU/extract-subvector-16bit.ll
+++ b/llvm/test/CodeGen/AMDGPU/extract-subvector-16bit.ll
@@ -112,15 +112,14 @@
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    v_pk_ashrrev_i16 v0, 15, v3 op_sel_hi:[0,0]
 ; GFX9-NEXT:    s_movk_i32 s4, 0x8000
-; GFX9-NEXT:    v_or_b32_sdwa v1, v0, s4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
-; GFX9-NEXT:    v_or_b32_e32 v3, 0xffff8000, v0
+; GFX9-NEXT:    v_or_b32_e32 v1, 0xffff8000, v0
+; GFX9-NEXT:    v_or_b32_sdwa v3, v0, s4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
 ; GFX9-NEXT:    v_pk_ashrrev_i16 v0, 15, v2 op_sel_hi:[0,1]
-; GFX9-NEXT:    v_or_b32_sdwa v2, v0, s4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
-; GFX9-NEXT:    v_or_b32_e32 v0, 0xffff8000, v0
-; GFX9-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX9-NEXT:    v_lshl_or_b32 v0, v2, 16, v0
-; GFX9-NEXT:    v_and_b32_e32 v2, 0xffff, v3
-; GFX9-NEXT:    v_lshl_or_b32 v1, v1, 16, v2
+; GFX9-NEXT:    v_or_b32_e32 v2, 0xffff8000, v0
+; GFX9-NEXT:    v_or_b32_sdwa v0, v0, s4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX9-NEXT:    s_mov_b32 s4, 0x5040100
+; GFX9-NEXT:    v_perm_b32 v0, v0, v2, s4
+; GFX9-NEXT:    v_perm_b32 v1, v3, v1, s4
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
   br i1 undef, label %T, label %F
 
@@ -252,15 +251,14 @@
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    v_pk_ashrrev_i16 v0, 15, v5 op_sel_hi:[0,1]
 ; GFX9-NEXT:    s_movk_i32 s4, 0x8000
-; GFX9-NEXT:    v_or_b32_sdwa v1, v0, s4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
-; GFX9-NEXT:    v_or_b32_e32 v2, 0xffff8000, v0
+; GFX9-NEXT:    v_or_b32_e32 v1, 0xffff8000, v0
+; GFX9-NEXT:    v_or_b32_sdwa v2, v0, s4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
 ; GFX9-NEXT:    v_pk_ashrrev_i16 v0, 15, v4 op_sel_hi:[0,1]
-; GFX9-NEXT:    v_or_b32_sdwa v3, v0, s4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
-; GFX9-NEXT:    v_or_b32_e32 v0, 0xffff8000, v0
-; GFX9-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX9-NEXT:    v_and_b32_e32 v2, 0xffff, v2
-; GFX9-NEXT:    v_lshl_or_b32 v0, v3, 16, v0
-; GFX9-NEXT:    v_lshl_or_b32 v1, v1, 16, v2
+; GFX9-NEXT:    v_or_b32_e32 v3, 0xffff8000, v0
+; GFX9-NEXT:    v_or_b32_sdwa v0, v0, s4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX9-NEXT:    s_mov_b32 s4, 0x5040100
+; GFX9-NEXT:    v_perm_b32 v0, v0, v3, s4
+; GFX9-NEXT:    v_perm_b32 v1, v2, v1, s4
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
   br i1 undef, label %T, label %F
 
@@ -389,9 +387,9 @@
 ; GFX9-NEXT:    global_load_dwordx4 v[2:5], v[0:1], off glc
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:  .LBB2_4: ; %exit
+; GFX9-NEXT:    s_mov_b32 s4, 0x5040100
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
-; GFX9-NEXT:    v_and_b32_e32 v0, 0xffff, v3
-; GFX9-NEXT:    v_lshl_or_b32 v0, v3, 16, v0
+; GFX9-NEXT:    v_perm_b32 v0, v3, v3, s4
 ; GFX9-NEXT:    v_mov_b32_e32 v1, 0x3800
 ; GFX9-NEXT:    v_mov_b32_e32 v3, 0x3900
 ; GFX9-NEXT:    v_mov_b32_e32 v4, 0x3d00
@@ -582,15 +580,14 @@
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    v_pk_ashrrev_i16 v0, 15, v5 op_sel_hi:[0,0]
 ; GFX9-NEXT:    s_movk_i32 s4, 0x8000
-; GFX9-NEXT:    v_or_b32_sdwa v1, v0, s4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
-; GFX9-NEXT:    v_or_b32_e32 v2, 0xffff8000, v0
+; GFX9-NEXT:    v_or_b32_e32 v1, 0xffff8000, v0
+; GFX9-NEXT:    v_or_b32_sdwa v2, v0, s4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
 ; GFX9-NEXT:    v_pk_ashrrev_i16 v0, 15, v4 op_sel_hi:[0,1]
-; GFX9-NEXT:    v_or_b32_sdwa v3, v0, s4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
-; GFX9-NEXT:    v_or_b32_e32 v0, 0xffff8000, v0
-; GFX9-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX9-NEXT:    v_and_b32_e32 v2, 0xffff, v2
-; GFX9-NEXT:    v_lshl_or_b32 v0, v3, 16, v0
-; GFX9-NEXT:    v_lshl_or_b32 v1, v1, 16, v2
+; GFX9-NEXT:    v_or_b32_e32 v3, 0xffff8000, v0
+; GFX9-NEXT:    v_or_b32_sdwa v0, v0, s4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX9-NEXT:    s_mov_b32 s4, 0x5040100
+; GFX9-NEXT:    v_perm_b32 v0, v0, v3, s4
+; GFX9-NEXT:    v_perm_b32 v1, v2, v1, s4
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
   br i1 undef, label %T, label %F
 
@@ -770,15 +767,14 @@
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    v_pk_ashrrev_i16 v0, 15, v7 op_sel_hi:[0,1]
 ; GFX9-NEXT:    s_movk_i32 s4, 0x8000
-; GFX9-NEXT:    v_or_b32_sdwa v1, v0, s4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
-; GFX9-NEXT:    v_or_b32_e32 v2, 0xffff8000, v0
+; GFX9-NEXT:    v_or_b32_e32 v1, 0xffff8000, v0
+; GFX9-NEXT:    v_or_b32_sdwa v2, v0, s4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
 ; GFX9-NEXT:    v_pk_ashrrev_i16 v0, 15, v6 op_sel_hi:[0,1]
-; GFX9-NEXT:    v_or_b32_sdwa v3, v0, s4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
-; GFX9-NEXT:    v_or_b32_e32 v0, 0xffff8000, v0
-; GFX9-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX9-NEXT:    v_and_b32_e32 v2, 0xffff, v2
-; GFX9-NEXT:    v_lshl_or_b32 v0, v3, 16, v0
-; GFX9-NEXT:    v_lshl_or_b32 v1, v1, 16, v2
+; GFX9-NEXT:    v_or_b32_e32 v3, 0xffff8000, v0
+; GFX9-NEXT:    v_or_b32_sdwa v0, v0, s4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX9-NEXT:    s_mov_b32 s4, 0x5040100
+; GFX9-NEXT:    v_perm_b32 v0, v0, v3, s4
+; GFX9-NEXT:    v_perm_b32 v1, v2, v1, s4
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
   br i1 undef, label %T, label %F
 
@@ -955,9 +951,9 @@
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    ; kill: killed $vgpr0 killed $vgpr1
 ; GFX9-NEXT:  .LBB5_4: ; %exit
+; GFX9-NEXT:    s_mov_b32 s4, 0x5040100
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
-; GFX9-NEXT:    v_and_b32_e32 v0, 0xffff, v5
-; GFX9-NEXT:    v_lshl_or_b32 v0, v5, 16, v0
+; GFX9-NEXT:    v_perm_b32 v0, v5, v5, s4
 ; GFX9-NEXT:    v_mov_b32_e32 v1, 0x3800
 ; GFX9-NEXT:    v_mov_b32_e32 v2, 0x3900
 ; GFX9-NEXT:    v_mov_b32_e32 v3, 0x3d00
diff --git a/llvm/test/CodeGen/AMDGPU/fast-unaligned-load-store.global.ll b/llvm/test/CodeGen/AMDGPU/fast-unaligned-load-store.global.ll
--- a/llvm/test/CodeGen/AMDGPU/fast-unaligned-load-store.global.ll
+++ b/llvm/test/CodeGen/AMDGPU/fast-unaligned-load-store.global.ll
@@ -184,10 +184,10 @@
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX9-NEXT:    global_load_dword v0, v[0:1], off
-; GFX9-NEXT:    v_mov_b32_e32 v1, 0xffff
+; GFX9-NEXT:    v_mov_b32_e32 v1, 0x7060100
 ; GFX9-NEXT:    s_mov_b32 s4, 0xffff
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
-; GFX9-NEXT:    v_bfi_b32 v1, v1, 0, v0
+; GFX9-NEXT:    v_perm_b32 v1, v0, 0, v1
 ; GFX9-NEXT:    v_and_or_b32 v0, v0, s4, v1
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
 ;
@@ -197,7 +197,7 @@
 ; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-NEXT:    global_load_dword v0, v[0:1], off
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
-; GFX10-NEXT:    v_bfi_b32 v1, 0xffff, 0, v0
+; GFX10-NEXT:    v_perm_b32 v1, v0, 0, 0x7060100
 ; GFX10-NEXT:    v_and_or_b32 v0, 0xffff, v0, v1
 ; GFX10-NEXT:    s_setpc_b64 s[30:31]
 ;
@@ -207,7 +207,7 @@
 ; GFX11-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-NEXT:    global_load_b32 v0, v[0:1], off
 ; GFX11-NEXT:    s_waitcnt vmcnt(0)
-; GFX11-NEXT:    v_bfi_b32 v1, 0xffff, 0, v0
+; GFX11-NEXT:    v_perm_b32 v1, v0, 0, 0x7060100
 ; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
 ; GFX11-NEXT:    v_and_or_b32 v0, 0xffff, v0, v1
 ; GFX11-NEXT:    s_setpc_b64 s[30:31]
@@ -319,10 +319,10 @@
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX9-NEXT:    global_load_dword v0, v[0:1], off
-; GFX9-NEXT:    v_mov_b32_e32 v1, 0xffff
+; GFX9-NEXT:    v_mov_b32_e32 v1, 0x7060100
 ; GFX9-NEXT:    s_mov_b32 s4, 0xffff
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
-; GFX9-NEXT:    v_bfi_b32 v1, v1, 0, v0
+; GFX9-NEXT:    v_perm_b32 v1, v0, 0, v1
 ; GFX9-NEXT:    v_and_or_b32 v0, v0, s4, v1
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
 ;
@@ -332,7 +332,7 @@
 ; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-NEXT:    global_load_dword v0, v[0:1], off
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
-; GFX10-NEXT:    v_bfi_b32 v1, 0xffff, 0, v0
+; GFX10-NEXT:    v_perm_b32 v1, v0, 0, 0x7060100
 ; GFX10-NEXT:    v_and_or_b32 v0, 0xffff, v0, v1
 ; GFX10-NEXT:    s_setpc_b64 s[30:31]
 ;
@@ -342,7 +342,7 @@
 ; GFX11-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-NEXT:    global_load_b32 v0, v[0:1], off
 ; GFX11-NEXT:    s_waitcnt vmcnt(0)
-; GFX11-NEXT:    v_bfi_b32 v1, 0xffff, 0, v0
+; GFX11-NEXT:    v_perm_b32 v1, v0, 0, 0x7060100
 ; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
 ; GFX11-NEXT:    v_and_or_b32 v0, 0xffff, v0, v1
 ; GFX11-NEXT:    s_setpc_b64 s[30:31]
diff --git a/llvm/test/CodeGen/AMDGPU/fast-unaligned-load-store.private.ll b/llvm/test/CodeGen/AMDGPU/fast-unaligned-load-store.private.ll
--- a/llvm/test/CodeGen/AMDGPU/fast-unaligned-load-store.private.ll
+++ b/llvm/test/CodeGen/AMDGPU/fast-unaligned-load-store.private.ll
@@ -235,10 +235,10 @@
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX9-NEXT:    buffer_load_dword v0, v0, s[0:3], 0 offen
-; GFX9-NEXT:    v_mov_b32_e32 v1, 0xffff
+; GFX9-NEXT:    v_mov_b32_e32 v1, 0x7060100
 ; GFX9-NEXT:    s_mov_b32 s4, 0xffff
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
-; GFX9-NEXT:    v_bfi_b32 v1, v1, 0, v0
+; GFX9-NEXT:    v_perm_b32 v1, v0, 0, v1
 ; GFX9-NEXT:    v_and_or_b32 v0, v0, s4, v1
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
 ;
@@ -246,10 +246,10 @@
 ; GFX9-FLASTSCR:       ; %bb.0:
 ; GFX9-FLASTSCR-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX9-FLASTSCR-NEXT:    scratch_load_dword v0, v0, off
-; GFX9-FLASTSCR-NEXT:    v_mov_b32_e32 v1, 0xffff
+; GFX9-FLASTSCR-NEXT:    v_mov_b32_e32 v1, 0x7060100
 ; GFX9-FLASTSCR-NEXT:    s_mov_b32 s0, 0xffff
 ; GFX9-FLASTSCR-NEXT:    s_waitcnt vmcnt(0)
-; GFX9-FLASTSCR-NEXT:    v_bfi_b32 v1, v1, 0, v0
+; GFX9-FLASTSCR-NEXT:    v_perm_b32 v1, v0, 0, v1
 ; GFX9-FLASTSCR-NEXT:    v_and_or_b32 v0, v0, s0, v1
 ; GFX9-FLASTSCR-NEXT:    s_setpc_b64 s[30:31]
 ;
@@ -259,7 +259,7 @@
 ; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-NEXT:    buffer_load_dword v0, v0, s[0:3], 0 offen
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
-; GFX10-NEXT:    v_bfi_b32 v1, 0xffff, 0, v0
+; GFX10-NEXT:    v_perm_b32 v1, v0, 0, 0x7060100
 ; GFX10-NEXT:    v_and_or_b32 v0, 0xffff, v0, v1
 ; GFX10-NEXT:    s_setpc_b64 s[30:31]
 ;
@@ -269,7 +269,7 @@
 ; GFX10-FLASTSCR-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-FLASTSCR-NEXT:    scratch_load_dword v0, v0, off
 ; GFX10-FLASTSCR-NEXT:    s_waitcnt vmcnt(0)
-; GFX10-FLASTSCR-NEXT:    v_bfi_b32 v1, 0xffff, 0, v0
+; GFX10-FLASTSCR-NEXT:    v_perm_b32 v1, v0, 0, 0x7060100
 ; GFX10-FLASTSCR-NEXT:    v_and_or_b32 v0, 0xffff, v0, v1
 ; GFX10-FLASTSCR-NEXT:    s_setpc_b64 s[30:31]
 ;
@@ -279,7 +279,7 @@
 ; GFX11-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-NEXT:    scratch_load_b32 v0, v0, off
 ; GFX11-NEXT:    s_waitcnt vmcnt(0)
-; GFX11-NEXT:    v_bfi_b32 v1, 0xffff, 0, v0
+; GFX11-NEXT:    v_perm_b32 v1, v0, 0, 0x7060100
 ; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
 ; GFX11-NEXT:    v_and_or_b32 v0, 0xffff, v0, v1
 ; GFX11-NEXT:    s_setpc_b64 s[30:31]
@@ -290,7 +290,7 @@
 ; GFX11-FLASTSCR-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-FLASTSCR-NEXT:    scratch_load_b32 v0, v0, off
 ; GFX11-FLASTSCR-NEXT:    s_waitcnt vmcnt(0)
-; GFX11-FLASTSCR-NEXT:    v_bfi_b32 v1, 0xffff, 0, v0
+; GFX11-FLASTSCR-NEXT:    v_perm_b32 v1, v0, 0, 0x7060100
 ; GFX11-FLASTSCR-NEXT:    s_delay_alu instid0(VALU_DEP_1)
 ; GFX11-FLASTSCR-NEXT:    v_and_or_b32 v0, 0xffff, v0, v1
 ; GFX11-FLASTSCR-NEXT:    s_setpc_b64 s[30:31]
@@ -414,10 +414,10 @@
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX9-NEXT:    buffer_load_dword v0, v0, s[0:3], 0 offen
-; GFX9-NEXT:    v_mov_b32_e32 v1, 0xffff
+; GFX9-NEXT:    v_mov_b32_e32 v1, 0x7060100
 ; GFX9-NEXT:    s_mov_b32 s4, 0xffff
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
-; GFX9-NEXT:    v_bfi_b32 v1, v1, 0, v0
+; GFX9-NEXT:    v_perm_b32 v1, v0, 0, v1
 ; GFX9-NEXT:    v_and_or_b32 v0, v0, s4, v1
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
 ;
@@ -425,10 +425,10 @@
 ; GFX9-FLASTSCR:       ; %bb.0:
 ; GFX9-FLASTSCR-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX9-FLASTSCR-NEXT:    scratch_load_dword v0, v0, off
-; GFX9-FLASTSCR-NEXT:    v_mov_b32_e32 v1, 0xffff
+; GFX9-FLASTSCR-NEXT:    v_mov_b32_e32 v1, 0x7060100
 ; GFX9-FLASTSCR-NEXT:    s_mov_b32 s0, 0xffff
 ; GFX9-FLASTSCR-NEXT:    s_waitcnt vmcnt(0)
-; GFX9-FLASTSCR-NEXT:    v_bfi_b32 v1, v1, 0, v0
+; GFX9-FLASTSCR-NEXT:    v_perm_b32 v1, v0, 0, v1
 ; GFX9-FLASTSCR-NEXT:    v_and_or_b32 v0, v0, s0, v1
 ; GFX9-FLASTSCR-NEXT:    s_setpc_b64 s[30:31]
 ;
@@ -438,7 +438,7 @@
 ; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-NEXT:    buffer_load_dword v0, v0, s[0:3], 0 offen
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
-; GFX10-NEXT:    v_bfi_b32 v1, 0xffff, 0, v0
+; GFX10-NEXT:    v_perm_b32 v1, v0, 0, 0x7060100
 ; GFX10-NEXT:    v_and_or_b32 v0, 0xffff, v0, v1
 ; GFX10-NEXT:    s_setpc_b64 s[30:31]
 ;
@@ -448,7 +448,7 @@
 ; GFX10-FLASTSCR-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-FLASTSCR-NEXT:    scratch_load_dword v0, v0, off
 ; GFX10-FLASTSCR-NEXT:    s_waitcnt vmcnt(0)
-; GFX10-FLASTSCR-NEXT:    v_bfi_b32 v1, 0xffff, 0, v0
+; GFX10-FLASTSCR-NEXT:    v_perm_b32 v1, v0, 0, 0x7060100
 ; GFX10-FLASTSCR-NEXT:    v_and_or_b32 v0, 0xffff, v0, v1
 ; GFX10-FLASTSCR-NEXT:    s_setpc_b64 s[30:31]
 ;
@@ -458,7 +458,7 @@
 ; GFX11-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-NEXT:    scratch_load_b32 v0, v0, off
 ; GFX11-NEXT:    s_waitcnt vmcnt(0)
-; GFX11-NEXT:    v_bfi_b32 v1, 0xffff, 0, v0
+; GFX11-NEXT:    v_perm_b32 v1, v0, 0, 0x7060100
 ; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
 ; GFX11-NEXT:    v_and_or_b32 v0, 0xffff, v0, v1
 ; GFX11-NEXT:    s_setpc_b64 s[30:31]
@@ -469,7 +469,7 @@
 ; GFX11-FLASTSCR-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-FLASTSCR-NEXT:    scratch_load_b32 v0, v0, off
 ; GFX11-FLASTSCR-NEXT:    s_waitcnt vmcnt(0)
-; GFX11-FLASTSCR-NEXT:    v_bfi_b32 v1, 0xffff, 0, v0
+; GFX11-FLASTSCR-NEXT:    v_perm_b32 v1, v0, 0, 0x7060100
 ; GFX11-FLASTSCR-NEXT:    s_delay_alu instid0(VALU_DEP_1)
 ; GFX11-FLASTSCR-NEXT:    v_and_or_b32 v0, 0xffff, v0, v1
 ; GFX11-FLASTSCR-NEXT:    s_setpc_b64 s[30:31]
diff --git a/llvm/test/CodeGen/AMDGPU/fcanonicalize.f16.ll b/llvm/test/CodeGen/AMDGPU/fcanonicalize.f16.ll
--- a/llvm/test/CodeGen/AMDGPU/fcanonicalize.f16.ll
+++ b/llvm/test/CodeGen/AMDGPU/fcanonicalize.f16.ll
@@ -129,30 +129,6 @@
 }
 
 define <2 x half> @v_test_canonicalize_build_vector_v2f16(half %lo, half %hi) #1 {
-; VI-LABEL: v_test_canonicalize_build_vector_v2f16:
-; VI:       ; %bb.0:
-; VI-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; VI-NEXT:    v_max_f16_sdwa v1, v1, v1 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
-; VI-NEXT:    v_max_f16_e32 v0, v0, v0
-; VI-NEXT:    v_or_b32_e32 v0, v0, v1
-; VI-NEXT:    s_setpc_b64 s[30:31]
-;
-; GFX9-LABEL: v_test_canonicalize_build_vector_v2f16:
-; GFX9:       ; %bb.0:
-; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX9-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX9-NEXT:    v_lshl_or_b32 v0, v1, 16, v0
-; GFX9-NEXT:    v_pk_max_f16 v0, v0, v0
-; GFX9-NEXT:    s_setpc_b64 s[30:31]
-;
-; CI-LABEL: v_test_canonicalize_build_vector_v2f16:
-; CI:       ; %bb.0:
-; CI-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CI-NEXT:    v_cvt_f16_f32_e32 v0, v0
-; CI-NEXT:    v_cvt_f16_f32_e32 v1, v1
-; CI-NEXT:    v_cvt_f32_f16_e32 v0, v0
-; CI-NEXT:    v_cvt_f32_f16_e32 v1, v1
-; CI-NEXT:    s_setpc_b64 s[30:31]
   %ins0 = insertelement <2 x half> undef, half %lo, i32 0
   %ins1 = insertelement <2 x half> %ins0, half %hi, i32 1
   %canonicalized = call <2 x half> @llvm.canonicalize.v2f16(<2 x half> %ins1)
@@ -2152,36 +2128,6 @@
 }
 
 define <4 x half> @v_test_canonicalize_reg_reg_undef_undef_v4f16(half %val0, half %val1) #1 {
-; VI-LABEL: v_test_canonicalize_reg_reg_undef_undef_v4f16:
-; VI:       ; %bb.0:
-; VI-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; VI-NEXT:    v_max_f16_e32 v0, v0, v0
-; VI-NEXT:    v_max_f16_sdwa v1, v1, v1 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
-; VI-NEXT:    v_or_b32_e32 v0, v0, v1
-; VI-NEXT:    v_mov_b32_e32 v1, 0x7e007e00
-; VI-NEXT:    s_setpc_b64 s[30:31]
-;
-; GFX9-LABEL: v_test_canonicalize_reg_reg_undef_undef_v4f16:
-; GFX9:       ; %bb.0:
-; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX9-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX9-NEXT:    v_lshl_or_b32 v0, v1, 16, v0
-; GFX9-NEXT:    v_pk_max_f16 v0, v0, v0
-; GFX9-NEXT:    v_mov_b32_e32 v1, 0x7e007e00
-; GFX9-NEXT:    s_setpc_b64 s[30:31]
-;
-; CI-LABEL: v_test_canonicalize_reg_reg_undef_undef_v4f16:
-; CI:       ; %bb.0:
-; CI-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CI-NEXT:    v_cvt_f16_f32_e32 v0, v0
-; CI-NEXT:    v_cvt_f16_f32_e32 v1, v1
-; CI-NEXT:    v_mov_b32_e32 v2, 0x7fc00000
-; CI-NEXT:    v_mov_b32_e32 v3, 0x7fc00000
-; CI-NEXT:    v_cvt_f32_f16_e32 v0, v0
-; CI-NEXT:    v_cvt_f32_f16_e32 v1, v1
-; CI-NEXT:    v_mul_f32_e32 v0, 1.0, v0
-; CI-NEXT:    v_mul_f32_e32 v1, 1.0, v1
-; CI-NEXT:    s_setpc_b64 s[30:31]
   %vec0 = insertelement <4 x half> undef, half %val0, i32 0
   %vec1 = insertelement <4 x half> %vec0, half %val1, i32 1
   %canonicalized = call <4 x half> @llvm.canonicalize.v4f16(<4 x half> %vec1)
@@ -2189,40 +2135,6 @@
 }
 
 define <4 x half> @v_test_canonicalize_reg_undef_reg_reg_v4f16(half %val0, half %val1, half %val2) #1 {
-; VI-LABEL: v_test_canonicalize_reg_undef_reg_reg_v4f16:
-; VI:       ; %bb.0:
-; VI-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; VI-NEXT:    v_max_f16_e32 v1, v1, v1
-; VI-NEXT:    v_max_f16_sdwa v2, v2, v2 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
-; VI-NEXT:    v_max_f16_e32 v0, v0, v0
-; VI-NEXT:    v_or_b32_e32 v0, 0x7e000000, v0
-; VI-NEXT:    v_or_b32_e32 v1, v1, v2
-; VI-NEXT:    s_setpc_b64 s[30:31]
-;
-; GFX9-LABEL: v_test_canonicalize_reg_undef_reg_reg_v4f16:
-; GFX9:       ; %bb.0:
-; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX9-NEXT:    v_and_b32_e32 v1, 0xffff, v1
-; GFX9-NEXT:    v_max_f16_e32 v0, v0, v0
-; GFX9-NEXT:    v_lshl_or_b32 v1, v2, 16, v1
-; GFX9-NEXT:    v_pack_b32_f16 v0, v0, 0
-; GFX9-NEXT:    v_pk_max_f16 v1, v1, v1
-; GFX9-NEXT:    s_setpc_b64 s[30:31]
-;
-; CI-LABEL: v_test_canonicalize_reg_undef_reg_reg_v4f16:
-; CI:       ; %bb.0:
-; CI-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CI-NEXT:    v_cvt_f16_f32_e32 v0, v0
-; CI-NEXT:    v_cvt_f16_f32_e32 v1, v1
-; CI-NEXT:    v_cvt_f16_f32_e32 v2, v2
-; CI-NEXT:    v_cvt_f32_f16_e32 v0, v0
-; CI-NEXT:    v_cvt_f32_f16_e32 v1, v1
-; CI-NEXT:    v_cvt_f32_f16_e32 v3, v2
-; CI-NEXT:    v_mul_f32_e32 v0, 1.0, v0
-; CI-NEXT:    v_mul_f32_e32 v2, 1.0, v1
-; CI-NEXT:    v_mul_f32_e32 v3, 1.0, v3
-; CI-NEXT:    v_mov_b32_e32 v1, 0x7fc00000
-; CI-NEXT:    s_setpc_b64 s[30:31]
   %vec0 = insertelement <4 x half> undef, half %val0, i32 0
   %vec1 = insertelement <4 x half> %vec0, half %val1, i32 2
   %vec2 = insertelement <4 x half> %vec1, half %val2, i32 3
diff --git a/llvm/test/CodeGen/AMDGPU/fmax_legacy.f16.ll b/llvm/test/CodeGen/AMDGPU/fmax_legacy.f16.ll
--- a/llvm/test/CodeGen/AMDGPU/fmax_legacy.f16.ll
+++ b/llvm/test/CodeGen/AMDGPU/fmax_legacy.f16.ll
@@ -69,8 +69,8 @@
 ; GFX9-SAFE-NEXT:    v_cndmask_b32_e32 v2, v2, v3, vcc
 ; GFX9-SAFE-NEXT:    v_cmp_nle_f16_e32 vcc, v0, v1
 ; GFX9-SAFE-NEXT:    v_cndmask_b32_e32 v0, v1, v0, vcc
-; GFX9-SAFE-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX9-SAFE-NEXT:    v_lshl_or_b32 v0, v2, 16, v0
+; GFX9-SAFE-NEXT:    s_mov_b32 s4, 0x5040100
+; GFX9-SAFE-NEXT:    v_perm_b32 v0, v2, v0, s4
 ; GFX9-SAFE-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX9-NNAN-LABEL: test_fmax_legacy_ugt_v2f16:
@@ -146,8 +146,8 @@
 ; GFX9-SAFE-NEXT:    v_cndmask_b32_e32 v1, v3, v1, vcc
 ; GFX9-SAFE-NEXT:    v_cmp_nle_f16_e32 vcc, v0, v2
 ; GFX9-SAFE-NEXT:    v_cndmask_b32_e32 v0, v2, v0, vcc
-; GFX9-SAFE-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX9-SAFE-NEXT:    v_lshl_or_b32 v0, v4, 16, v0
+; GFX9-SAFE-NEXT:    s_mov_b32 s4, 0x5040100
+; GFX9-SAFE-NEXT:    v_perm_b32 v0, v4, v0, s4
 ; GFX9-SAFE-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX9-NNAN-LABEL: test_fmax_legacy_ugt_v3f16:
@@ -241,10 +241,9 @@
 ; GFX9-SAFE-NEXT:    v_cndmask_b32_e32 v1, v3, v1, vcc
 ; GFX9-SAFE-NEXT:    v_cmp_nle_f16_e32 vcc, v0, v2
 ; GFX9-SAFE-NEXT:    v_cndmask_b32_e32 v0, v2, v0, vcc
-; GFX9-SAFE-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX9-SAFE-NEXT:    v_and_b32_e32 v1, 0xffff, v1
-; GFX9-SAFE-NEXT:    v_lshl_or_b32 v0, v4, 16, v0
-; GFX9-SAFE-NEXT:    v_lshl_or_b32 v1, v6, 16, v1
+; GFX9-SAFE-NEXT:    s_mov_b32 s4, 0x5040100
+; GFX9-SAFE-NEXT:    v_perm_b32 v0, v4, v0, s4
+; GFX9-SAFE-NEXT:    v_perm_b32 v1, v6, v1, s4
 ; GFX9-SAFE-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX9-NNAN-LABEL: test_fmax_legacy_ugt_v4f16:
@@ -368,14 +367,11 @@
 ; GFX9-SAFE-NEXT:    v_cndmask_b32_e32 v1, v5, v1, vcc
 ; GFX9-SAFE-NEXT:    v_cmp_nle_f16_e32 vcc, v0, v4
 ; GFX9-SAFE-NEXT:    v_cndmask_b32_e32 v0, v4, v0, vcc
-; GFX9-SAFE-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX9-SAFE-NEXT:    v_and_b32_e32 v1, 0xffff, v1
-; GFX9-SAFE-NEXT:    v_and_b32_e32 v2, 0xffff, v2
-; GFX9-SAFE-NEXT:    v_and_b32_e32 v3, 0xffff, v3
-; GFX9-SAFE-NEXT:    v_lshl_or_b32 v0, v8, 16, v0
-; GFX9-SAFE-NEXT:    v_lshl_or_b32 v1, v10, 16, v1
-; GFX9-SAFE-NEXT:    v_lshl_or_b32 v2, v12, 16, v2
-; GFX9-SAFE-NEXT:    v_lshl_or_b32 v3, v14, 16, v3
+; GFX9-SAFE-NEXT:    s_mov_b32 s4, 0x5040100
+; GFX9-SAFE-NEXT:    v_perm_b32 v0, v8, v0, s4
+; GFX9-SAFE-NEXT:    v_perm_b32 v1, v10, v1, s4
+; GFX9-SAFE-NEXT:    v_perm_b32 v2, v12, v2, s4
+; GFX9-SAFE-NEXT:    v_perm_b32 v3, v14, v3, s4
 ; GFX9-SAFE-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX9-NNAN-LABEL: test_fmax_legacy_ugt_v8f16:
diff --git a/llvm/test/CodeGen/AMDGPU/fmin_legacy.f16.ll b/llvm/test/CodeGen/AMDGPU/fmin_legacy.f16.ll
--- a/llvm/test/CodeGen/AMDGPU/fmin_legacy.f16.ll
+++ b/llvm/test/CodeGen/AMDGPU/fmin_legacy.f16.ll
@@ -70,8 +70,8 @@
 ; GFX9-SAFE-NEXT:    v_cndmask_b32_e32 v2, v2, v3, vcc
 ; GFX9-SAFE-NEXT:    v_cmp_ngt_f16_e32 vcc, v0, v1
 ; GFX9-SAFE-NEXT:    v_cndmask_b32_e32 v0, v1, v0, vcc
-; GFX9-SAFE-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX9-SAFE-NEXT:    v_lshl_or_b32 v0, v2, 16, v0
+; GFX9-SAFE-NEXT:    s_mov_b32 s4, 0x5040100
+; GFX9-SAFE-NEXT:    v_perm_b32 v0, v2, v0, s4
 ; GFX9-SAFE-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX9-NNAN-LABEL: test_fmin_legacy_ule_v2f16:
@@ -147,8 +147,8 @@
 ; GFX9-SAFE-NEXT:    v_cndmask_b32_e32 v1, v3, v1, vcc
 ; GFX9-SAFE-NEXT:    v_cmp_ngt_f16_e32 vcc, v0, v2
 ; GFX9-SAFE-NEXT:    v_cndmask_b32_e32 v0, v2, v0, vcc
-; GFX9-SAFE-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX9-SAFE-NEXT:    v_lshl_or_b32 v0, v4, 16, v0
+; GFX9-SAFE-NEXT:    s_mov_b32 s4, 0x5040100
+; GFX9-SAFE-NEXT:    v_perm_b32 v0, v4, v0, s4
 ; GFX9-SAFE-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX9-NNAN-LABEL: test_fmin_legacy_ule_v3f16:
@@ -242,10 +242,9 @@
 ; GFX9-SAFE-NEXT:    v_cndmask_b32_e32 v1, v3, v1, vcc
 ; GFX9-SAFE-NEXT:    v_cmp_ngt_f16_e32 vcc, v0, v2
 ; GFX9-SAFE-NEXT:    v_cndmask_b32_e32 v0, v2, v0, vcc
-; GFX9-SAFE-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX9-SAFE-NEXT:    v_and_b32_e32 v1, 0xffff, v1
-; GFX9-SAFE-NEXT:    v_lshl_or_b32 v0, v4, 16, v0
-; GFX9-SAFE-NEXT:    v_lshl_or_b32 v1, v6, 16, v1
+; GFX9-SAFE-NEXT:    s_mov_b32 s4, 0x5040100
+; GFX9-SAFE-NEXT:    v_perm_b32 v0, v4, v0, s4
+; GFX9-SAFE-NEXT:    v_perm_b32 v1, v6, v1, s4
 ; GFX9-SAFE-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX9-NNAN-LABEL: test_fmin_legacy_ule_v4f16:
@@ -369,14 +368,11 @@
 ; GFX9-SAFE-NEXT:    v_cndmask_b32_e32 v1, v5, v1, vcc
 ; GFX9-SAFE-NEXT:    v_cmp_ngt_f16_e32 vcc, v0, v4
 ; GFX9-SAFE-NEXT:    v_cndmask_b32_e32 v0, v4, v0, vcc
-; GFX9-SAFE-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX9-SAFE-NEXT:    v_and_b32_e32 v1, 0xffff, v1
-; GFX9-SAFE-NEXT:    v_and_b32_e32 v2, 0xffff, v2
-; GFX9-SAFE-NEXT:    v_and_b32_e32 v3, 0xffff, v3
-; GFX9-SAFE-NEXT:    v_lshl_or_b32 v0, v8, 16, v0
-; GFX9-SAFE-NEXT:    v_lshl_or_b32 v1, v10, 16, v1
-; GFX9-SAFE-NEXT:    v_lshl_or_b32 v2, v12, 16, v2
-; GFX9-SAFE-NEXT:    v_lshl_or_b32 v3, v14, 16, v3
+; GFX9-SAFE-NEXT:    s_mov_b32 s4, 0x5040100
+; GFX9-SAFE-NEXT:    v_perm_b32 v0, v8, v0, s4
+; GFX9-SAFE-NEXT:    v_perm_b32 v1, v10, v1, s4
+; GFX9-SAFE-NEXT:    v_perm_b32 v2, v12, v2, s4
+; GFX9-SAFE-NEXT:    v_perm_b32 v3, v14, v3, s4
 ; GFX9-SAFE-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX9-NNAN-LABEL: test_fmin_legacy_ule_v8f16:
diff --git a/llvm/test/CodeGen/AMDGPU/fshr.ll b/llvm/test/CodeGen/AMDGPU/fshr.ll
--- a/llvm/test/CodeGen/AMDGPU/fshr.ll
+++ b/llvm/test/CodeGen/AMDGPU/fshr.ll
@@ -933,8 +933,8 @@
 ; GFX9-NEXT:    v_lshlrev_b16_e32 v0, v3, v0
 ; GFX9-NEXT:    v_lshrrev_b16_e32 v2, v4, v2
 ; GFX9-NEXT:    v_or_b32_e32 v0, v0, v2
-; GFX9-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX9-NEXT:    v_lshl_or_b32 v0, v6, 16, v0
+; GFX9-NEXT:    s_mov_b32 s4, 0x5040100
+; GFX9-NEXT:    v_perm_b32 v0, v6, v0, s4
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; R600-LABEL: v_fshr_v3i16:
@@ -946,25 +946,24 @@
 ; GFX10:       ; %bb.0:
 ; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
-; GFX10-NEXT:    v_lshrrev_b32_e32 v6, 16, v4
-; GFX10-NEXT:    v_lshrrev_b32_e32 v7, 16, v0
+; GFX10-NEXT:    v_lshrrev_b32_e32 v6, 16, v0
+; GFX10-NEXT:    v_lshrrev_b32_e32 v7, 16, v4
+; GFX10-NEXT:    v_lshrrev_b32_e32 v8, 16, v2
 ; GFX10-NEXT:    v_lshlrev_b16 v0, 1, v0
-; GFX10-NEXT:    v_xor_b32_e32 v8, -1, v4
-; GFX10-NEXT:    v_lshrrev_b32_e32 v9, 16, v2
-; GFX10-NEXT:    v_xor_b32_e32 v10, -1, v6
-; GFX10-NEXT:    v_lshlrev_b16 v7, 1, v7
-; GFX10-NEXT:    v_lshrrev_b16 v2, v4, v2
-; GFX10-NEXT:    v_lshlrev_b16 v0, v8, v0
-; GFX10-NEXT:    v_lshrrev_b16 v4, v6, v9
+; GFX10-NEXT:    v_xor_b32_e32 v10, -1, v4
+; GFX10-NEXT:    v_lshlrev_b16 v6, 1, v6
+; GFX10-NEXT:    v_xor_b32_e32 v9, -1, v7
 ; GFX10-NEXT:    v_lshlrev_b16 v1, 1, v1
-; GFX10-NEXT:    v_lshlrev_b16 v6, v10, v7
+; GFX10-NEXT:    v_lshrrev_b16 v7, v7, v8
+; GFX10-NEXT:    v_lshlrev_b16 v0, v10, v0
+; GFX10-NEXT:    v_lshrrev_b16 v2, v4, v2
+; GFX10-NEXT:    v_lshlrev_b16 v6, v9, v6
+; GFX10-NEXT:    v_xor_b32_e32 v4, -1, v5
 ; GFX10-NEXT:    v_lshrrev_b16 v3, v5, v3
 ; GFX10-NEXT:    v_or_b32_e32 v0, v0, v2
-; GFX10-NEXT:    v_xor_b32_e32 v2, -1, v5
-; GFX10-NEXT:    v_or_b32_e32 v4, v6, v4
-; GFX10-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX10-NEXT:    v_lshlrev_b16 v1, v2, v1
-; GFX10-NEXT:    v_lshl_or_b32 v0, v4, 16, v0
+; GFX10-NEXT:    v_or_b32_e32 v5, v6, v7
+; GFX10-NEXT:    v_lshlrev_b16 v1, v4, v1
+; GFX10-NEXT:    v_perm_b32 v0, v5, v0, 0x5040100
 ; GFX10-NEXT:    v_or_b32_e32 v1, v1, v3
 ; GFX10-NEXT:    s_setpc_b64 s[30:31]
 ;
@@ -972,27 +971,26 @@
 ; GFX11:       ; %bb.0:
 ; GFX11-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX11-NEXT:    s_waitcnt_vscnt null, 0x0
-; GFX11-NEXT:    v_lshrrev_b32_e32 v6, 16, v4
-; GFX11-NEXT:    v_lshrrev_b32_e32 v7, 16, v0
+; GFX11-NEXT:    v_lshrrev_b32_e32 v6, 16, v0
+; GFX11-NEXT:    v_lshrrev_b32_e32 v7, 16, v4
+; GFX11-NEXT:    v_lshrrev_b32_e32 v8, 16, v2
 ; GFX11-NEXT:    v_lshlrev_b16 v0, 1, v0
-; GFX11-NEXT:    v_xor_b32_e32 v8, -1, v4
-; GFX11-NEXT:    v_lshrrev_b32_e32 v9, 16, v2
-; GFX11-NEXT:    v_xor_b32_e32 v10, -1, v6
-; GFX11-NEXT:    v_lshlrev_b16 v7, 1, v7
-; GFX11-NEXT:    v_lshrrev_b16 v2, v4, v2
-; GFX11-NEXT:    v_lshlrev_b16 v0, v8, v0
-; GFX11-NEXT:    v_lshrrev_b16 v4, v6, v9
+; GFX11-NEXT:    v_xor_b32_e32 v10, -1, v4
+; GFX11-NEXT:    v_lshlrev_b16 v6, 1, v6
+; GFX11-NEXT:    v_xor_b32_e32 v9, -1, v7
 ; GFX11-NEXT:    v_lshlrev_b16 v1, 1, v1
-; GFX11-NEXT:    v_lshlrev_b16 v6, v10, v7
+; GFX11-NEXT:    v_lshrrev_b16 v7, v7, v8
+; GFX11-NEXT:    v_lshlrev_b16 v0, v10, v0
+; GFX11-NEXT:    v_lshrrev_b16 v2, v4, v2
+; GFX11-NEXT:    v_lshlrev_b16 v6, v9, v6
+; GFX11-NEXT:    v_xor_b32_e32 v4, -1, v5
 ; GFX11-NEXT:    v_lshrrev_b16 v3, v5, v3
+; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
 ; GFX11-NEXT:    v_or_b32_e32 v0, v0, v2
-; GFX11-NEXT:    v_xor_b32_e32 v2, -1, v5
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_3)
-; GFX11-NEXT:    v_or_b32_e32 v4, v6, v4
-; GFX11-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX11-NEXT:    v_lshlrev_b16 v1, v2, v1
-; GFX11-NEXT:    v_lshl_or_b32 v0, v4, 16, v0
+; GFX11-NEXT:    v_or_b32_e32 v5, v6, v7
+; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_2)
+; GFX11-NEXT:    v_lshlrev_b16 v1, v4, v1
+; GFX11-NEXT:    v_perm_b32 v0, v5, v0, 0x5040100
 ; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_2)
 ; GFX11-NEXT:    v_or_b32_e32 v1, v1, v3
 ; GFX11-NEXT:    s_setpc_b64 s[30:31]
@@ -1080,12 +1078,11 @@
 ; GFX9-NEXT:    v_xor_b32_e32 v3, -1, v4
 ; GFX9-NEXT:    v_lshlrev_b16_e32 v0, v3, v0
 ; GFX9-NEXT:    v_lshrrev_b16_e32 v2, v4, v2
-; GFX9-NEXT:    v_or_b32_e32 v0, v0, v2
 ; GFX9-NEXT:    v_or_b32_e32 v7, v7, v9
-; GFX9-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX9-NEXT:    v_and_b32_e32 v1, 0xffff, v1
-; GFX9-NEXT:    v_lshl_or_b32 v0, v7, 16, v0
-; GFX9-NEXT:    v_lshl_or_b32 v1, v6, 16, v1
+; GFX9-NEXT:    v_or_b32_e32 v0, v0, v2
+; GFX9-NEXT:    s_mov_b32 s4, 0x5040100
+; GFX9-NEXT:    v_perm_b32 v0, v7, v0, s4
+; GFX9-NEXT:    v_perm_b32 v1, v6, v1, s4
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; R600-LABEL: v_fshr_v4i16:
@@ -1100,33 +1097,31 @@
 ; GFX10-NEXT:    v_lshrrev_b32_e32 v6, 16, v3
 ; GFX10-NEXT:    v_lshrrev_b32_e32 v7, 16, v5
 ; GFX10-NEXT:    v_lshrrev_b32_e32 v8, 16, v1
-; GFX10-NEXT:    v_lshrrev_b32_e32 v9, 16, v4
-; GFX10-NEXT:    v_lshrrev_b32_e32 v10, 16, v0
-; GFX10-NEXT:    v_lshlrev_b16 v1, 1, v1
-; GFX10-NEXT:    v_xor_b32_e32 v11, -1, v5
-; GFX10-NEXT:    v_lshlrev_b16 v0, 1, v0
-; GFX10-NEXT:    v_xor_b32_e32 v12, -1, v4
+; GFX10-NEXT:    v_lshrrev_b32_e32 v9, 16, v0
+; GFX10-NEXT:    v_lshrrev_b32_e32 v10, 16, v4
+; GFX10-NEXT:    v_lshrrev_b32_e32 v11, 16, v2
 ; GFX10-NEXT:    v_lshrrev_b16 v6, v7, v6
 ; GFX10-NEXT:    v_lshlrev_b16 v8, 1, v8
 ; GFX10-NEXT:    v_xor_b32_e32 v7, -1, v7
-; GFX10-NEXT:    v_lshrrev_b32_e32 v13, 16, v2
-; GFX10-NEXT:    v_lshlrev_b16 v10, 1, v10
-; GFX10-NEXT:    v_xor_b32_e32 v14, -1, v9
-; GFX10-NEXT:    v_lshlrev_b16 v1, v11, v1
-; GFX10-NEXT:    v_lshlrev_b16 v0, v12, v0
+; GFX10-NEXT:    v_lshlrev_b16 v9, 1, v9
+; GFX10-NEXT:    v_xor_b32_e32 v12, -1, v10
+; GFX10-NEXT:    v_lshlrev_b16 v1, 1, v1
+; GFX10-NEXT:    v_xor_b32_e32 v13, -1, v5
+; GFX10-NEXT:    v_lshlrev_b16 v0, 1, v0
+; GFX10-NEXT:    v_xor_b32_e32 v14, -1, v4
+; GFX10-NEXT:    v_lshlrev_b16 v7, v7, v8
+; GFX10-NEXT:    v_lshrrev_b16 v8, v10, v11
+; GFX10-NEXT:    v_lshlrev_b16 v9, v12, v9
+; GFX10-NEXT:    v_lshlrev_b16 v1, v13, v1
+; GFX10-NEXT:    v_lshlrev_b16 v0, v14, v0
 ; GFX10-NEXT:    v_lshrrev_b16 v2, v4, v2
 ; GFX10-NEXT:    v_lshrrev_b16 v3, v5, v3
-; GFX10-NEXT:    v_lshlrev_b16 v4, v7, v8
-; GFX10-NEXT:    v_lshrrev_b16 v5, v9, v13
-; GFX10-NEXT:    v_lshlrev_b16 v7, v14, v10
+; GFX10-NEXT:    v_or_b32_e32 v4, v7, v6
+; GFX10-NEXT:    v_or_b32_e32 v5, v9, v8
 ; GFX10-NEXT:    v_or_b32_e32 v0, v0, v2
 ; GFX10-NEXT:    v_or_b32_e32 v1, v1, v3
-; GFX10-NEXT:    v_or_b32_e32 v2, v4, v6
-; GFX10-NEXT:    v_or_b32_e32 v3, v7, v5
-; GFX10-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX10-NEXT:    v_and_b32_e32 v1, 0xffff, v1
-; GFX10-NEXT:    v_lshl_or_b32 v0, v3, 16, v0
-; GFX10-NEXT:    v_lshl_or_b32 v1, v2, 16, v1
+; GFX10-NEXT:    v_perm_b32 v0, v5, v0, 0x5040100
+; GFX10-NEXT:    v_perm_b32 v1, v4, v1, 0x5040100
 ; GFX10-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX11-LABEL: v_fshr_v4i16:
@@ -1136,36 +1131,33 @@
 ; GFX11-NEXT:    v_lshrrev_b32_e32 v6, 16, v3
 ; GFX11-NEXT:    v_lshrrev_b32_e32 v7, 16, v5
 ; GFX11-NEXT:    v_lshrrev_b32_e32 v8, 16, v1
-; GFX11-NEXT:    v_lshrrev_b32_e32 v9, 16, v4
-; GFX11-NEXT:    v_lshrrev_b32_e32 v10, 16, v0
-; GFX11-NEXT:    v_lshlrev_b16 v1, 1, v1
-; GFX11-NEXT:    v_xor_b32_e32 v11, -1, v5
-; GFX11-NEXT:    v_lshlrev_b16 v0, 1, v0
-; GFX11-NEXT:    v_xor_b32_e32 v12, -1, v4
+; GFX11-NEXT:    v_lshrrev_b32_e32 v9, 16, v0
+; GFX11-NEXT:    v_lshrrev_b32_e32 v10, 16, v4
+; GFX11-NEXT:    v_lshrrev_b32_e32 v11, 16, v2
 ; GFX11-NEXT:    v_lshrrev_b16 v6, v7, v6
 ; GFX11-NEXT:    v_lshlrev_b16 v8, 1, v8
 ; GFX11-NEXT:    v_xor_b32_e32 v7, -1, v7
-; GFX11-NEXT:    v_lshrrev_b32_e32 v13, 16, v2
-; GFX11-NEXT:    v_lshlrev_b16 v10, 1, v10
-; GFX11-NEXT:    v_xor_b32_e32 v14, -1, v9
-; GFX11-NEXT:    v_lshlrev_b16 v1, v11, v1
-; GFX11-NEXT:    v_lshlrev_b16 v0, v12, v0
+; GFX11-NEXT:    v_lshlrev_b16 v9, 1, v9
+; GFX11-NEXT:    v_xor_b32_e32 v12, -1, v10
+; GFX11-NEXT:    v_lshlrev_b16 v1, 1, v1
+; GFX11-NEXT:    v_xor_b32_e32 v13, -1, v5
+; GFX11-NEXT:    v_lshlrev_b16 v0, 1, v0
+; GFX11-NEXT:    v_xor_b32_e32 v14, -1, v4
+; GFX11-NEXT:    v_lshlrev_b16 v7, v7, v8
+; GFX11-NEXT:    v_lshrrev_b16 v8, v10, v11
+; GFX11-NEXT:    v_lshlrev_b16 v9, v12, v9
+; GFX11-NEXT:    v_lshlrev_b16 v1, v13, v1
+; GFX11-NEXT:    v_lshlrev_b16 v0, v14, v0
 ; GFX11-NEXT:    v_lshrrev_b16 v2, v4, v2
 ; GFX11-NEXT:    v_lshrrev_b16 v3, v5, v3
-; GFX11-NEXT:    v_lshlrev_b16 v4, v7, v8
-; GFX11-NEXT:    v_lshrrev_b16 v5, v9, v13
-; GFX11-NEXT:    v_lshlrev_b16 v7, v14, v10
+; GFX11-NEXT:    v_or_b32_e32 v4, v7, v6
+; GFX11-NEXT:    v_or_b32_e32 v5, v9, v8
+; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
 ; GFX11-NEXT:    v_or_b32_e32 v0, v0, v2
 ; GFX11-NEXT:    v_or_b32_e32 v1, v1, v3
-; GFX11-NEXT:    v_or_b32_e32 v2, v4, v6
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
-; GFX11-NEXT:    v_or_b32_e32 v3, v7, v5
-; GFX11-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX11-NEXT:    v_and_b32_e32 v1, 0xffff, v1
-; GFX11-NEXT:    v_lshl_or_b32 v0, v3, 16, v0
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_2)
-; GFX11-NEXT:    v_lshl_or_b32 v1, v2, 16, v1
+; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
+; GFX11-NEXT:    v_perm_b32 v0, v5, v0, 0x5040100
+; GFX11-NEXT:    v_perm_b32 v1, v4, v1, 0x5040100
 ; GFX11-NEXT:    s_setpc_b64 s[30:31]
   %ret = call <4 x i16> @llvm.fshr.v4i16(<4 x i16> %src0, <4 x i16> %src1, <4 x i16> %src2)
   ret <4 x i16> %ret
diff --git a/llvm/test/CodeGen/AMDGPU/idot4s.ll b/llvm/test/CodeGen/AMDGPU/idot4s.ll
--- a/llvm/test/CodeGen/AMDGPU/idot4s.ll
+++ b/llvm/test/CodeGen/AMDGPU/idot4s.ll
@@ -1032,29 +1032,29 @@
 ; GFX9-NODL-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
 ; GFX9-NODL-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x34
 ; GFX9-NODL-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
-; GFX9-NODL-NEXT:    v_mov_b32_e32 v4, 0xffff
+; GFX9-NODL-NEXT:    s_mov_b32 s0, 0x5040100
 ; GFX9-NODL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NODL-NEXT:    global_load_dword v1, v0, s[4:5]
 ; GFX9-NODL-NEXT:    global_load_dword v2, v0, s[6:7]
 ; GFX9-NODL-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX9-NODL-NEXT:    global_load_ushort v3, v0, s[2:3]
 ; GFX9-NODL-NEXT:    s_waitcnt vmcnt(2)
-; GFX9-NODL-NEXT:    v_lshrrev_b32_e32 v5, 16, v1
+; GFX9-NODL-NEXT:    v_lshrrev_b32_e32 v4, 16, v1
 ; GFX9-NODL-NEXT:    s_waitcnt vmcnt(1)
-; GFX9-NODL-NEXT:    v_lshrrev_b32_e32 v6, 16, v2
-; GFX9-NODL-NEXT:    v_ashrrev_i16_e32 v7, 8, v1
-; GFX9-NODL-NEXT:    v_ashrrev_i16_e32 v8, 8, v2
-; GFX9-NODL-NEXT:    v_and_b32_sdwa v2, v4, sext(v2) dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
-; GFX9-NODL-NEXT:    v_and_b32_sdwa v1, v4, sext(v1) dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
-; GFX9-NODL-NEXT:    v_lshl_or_b32 v2, v8, 16, v2
-; GFX9-NODL-NEXT:    v_lshl_or_b32 v1, v7, 16, v1
+; GFX9-NODL-NEXT:    v_lshrrev_b32_e32 v5, 16, v2
+; GFX9-NODL-NEXT:    v_ashrrev_i16_e32 v6, 8, v1
+; GFX9-NODL-NEXT:    v_bfe_i32 v1, v1, 0, 8
+; GFX9-NODL-NEXT:    v_ashrrev_i16_e32 v7, 8, v2
+; GFX9-NODL-NEXT:    v_bfe_i32 v2, v2, 0, 8
+; GFX9-NODL-NEXT:    v_perm_b32 v2, v7, v2, s0
+; GFX9-NODL-NEXT:    v_perm_b32 v1, v6, v1, s0
+; GFX9-NODL-NEXT:    v_ashrrev_i16_e32 v8, 8, v4
+; GFX9-NODL-NEXT:    v_bfe_i32 v4, v4, 0, 8
 ; GFX9-NODL-NEXT:    v_ashrrev_i16_e32 v9, 8, v5
-; GFX9-NODL-NEXT:    v_ashrrev_i16_e32 v10, 8, v6
-; GFX9-NODL-NEXT:    v_and_b32_sdwa v6, v4, sext(v6) dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
-; GFX9-NODL-NEXT:    v_and_b32_sdwa v4, v4, sext(v5) dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX9-NODL-NEXT:    v_bfe_i32 v5, v5, 0, 8
 ; GFX9-NODL-NEXT:    v_pk_mul_lo_u16 v1, v1, v2
-; GFX9-NODL-NEXT:    v_lshl_or_b32 v5, v10, 16, v6
-; GFX9-NODL-NEXT:    v_lshl_or_b32 v4, v9, 16, v4
+; GFX9-NODL-NEXT:    v_perm_b32 v5, v9, v5, s0
+; GFX9-NODL-NEXT:    v_perm_b32 v4, v8, v4, s0
 ; GFX9-NODL-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NODL-NEXT:    v_add_u16_e32 v3, v1, v3
 ; GFX9-NODL-NEXT:    v_pk_mul_lo_u16 v2, v4, v5
@@ -1069,29 +1069,29 @@
 ; GFX9-DL-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
 ; GFX9-DL-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x34
 ; GFX9-DL-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
-; GFX9-DL-NEXT:    v_mov_b32_e32 v4, 0xffff
+; GFX9-DL-NEXT:    s_mov_b32 s0, 0x5040100
 ; GFX9-DL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-DL-NEXT:    global_load_dword v1, v0, s[4:5]
 ; GFX9-DL-NEXT:    global_load_dword v2, v0, s[6:7]
 ; GFX9-DL-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX9-DL-NEXT:    global_load_ushort v3, v0, s[2:3]
 ; GFX9-DL-NEXT:    s_waitcnt vmcnt(2)
-; GFX9-DL-NEXT:    v_lshrrev_b32_e32 v5, 16, v1
+; GFX9-DL-NEXT:    v_lshrrev_b32_e32 v4, 16, v1
 ; GFX9-DL-NEXT:    s_waitcnt vmcnt(1)
-; GFX9-DL-NEXT:    v_lshrrev_b32_e32 v6, 16, v2
-; GFX9-DL-NEXT:    v_ashrrev_i16_e32 v7, 8, v1
-; GFX9-DL-NEXT:    v_ashrrev_i16_e32 v8, 8, v2
-; GFX9-DL-NEXT:    v_and_b32_sdwa v2, v4, sext(v2) dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
-; GFX9-DL-NEXT:    v_and_b32_sdwa v1, v4, sext(v1) dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
-; GFX9-DL-NEXT:    v_lshl_or_b32 v2, v8, 16, v2
-; GFX9-DL-NEXT:    v_lshl_or_b32 v1, v7, 16, v1
+; GFX9-DL-NEXT:    v_lshrrev_b32_e32 v5, 16, v2
+; GFX9-DL-NEXT:    v_ashrrev_i16_e32 v6, 8, v1
+; GFX9-DL-NEXT:    v_bfe_i32 v1, v1, 0, 8
+; GFX9-DL-NEXT:    v_ashrrev_i16_e32 v7, 8, v2
+; GFX9-DL-NEXT:    v_bfe_i32 v2, v2, 0, 8
+; GFX9-DL-NEXT:    v_perm_b32 v2, v7, v2, s0
+; GFX9-DL-NEXT:    v_perm_b32 v1, v6, v1, s0
+; GFX9-DL-NEXT:    v_ashrrev_i16_e32 v8, 8, v4
+; GFX9-DL-NEXT:    v_bfe_i32 v4, v4, 0, 8
 ; GFX9-DL-NEXT:    v_ashrrev_i16_e32 v9, 8, v5
-; GFX9-DL-NEXT:    v_ashrrev_i16_e32 v10, 8, v6
-; GFX9-DL-NEXT:    v_and_b32_sdwa v6, v4, sext(v6) dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
-; GFX9-DL-NEXT:    v_and_b32_sdwa v4, v4, sext(v5) dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX9-DL-NEXT:    v_bfe_i32 v5, v5, 0, 8
 ; GFX9-DL-NEXT:    v_pk_mul_lo_u16 v1, v1, v2
-; GFX9-DL-NEXT:    v_lshl_or_b32 v5, v10, 16, v6
-; GFX9-DL-NEXT:    v_lshl_or_b32 v4, v9, 16, v4
+; GFX9-DL-NEXT:    v_perm_b32 v5, v9, v5, s0
+; GFX9-DL-NEXT:    v_perm_b32 v4, v8, v4, s0
 ; GFX9-DL-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-DL-NEXT:    v_add_u16_e32 v3, v1, v3
 ; GFX9-DL-NEXT:    v_pk_mul_lo_u16 v2, v4, v5
@@ -1106,7 +1106,6 @@
 ; GFX10-DL-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
 ; GFX10-DL-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
 ; GFX10-DL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
-; GFX10-DL-NEXT:    v_mov_b32_e32 v4, 0xffff
 ; GFX10-DL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-DL-NEXT:    s_clause 0x1
 ; GFX10-DL-NEXT:    global_load_dword v1, v0, s[4:5]
@@ -1114,22 +1113,22 @@
 ; GFX10-DL-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX10-DL-NEXT:    global_load_ushort v3, v0, s[0:1]
 ; GFX10-DL-NEXT:    s_waitcnt vmcnt(2)
-; GFX10-DL-NEXT:    v_ashrrev_i16 v5, 8, v1
+; GFX10-DL-NEXT:    v_ashrrev_i16 v4, 8, v1
 ; GFX10-DL-NEXT:    s_waitcnt vmcnt(1)
-; GFX10-DL-NEXT:    v_ashrrev_i16 v6, 8, v2
-; GFX10-DL-NEXT:    v_and_b32_sdwa v7, v4, sext(v2) dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
-; GFX10-DL-NEXT:    v_and_b32_sdwa v8, v4, sext(v1) dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX10-DL-NEXT:    v_ashrrev_i16 v5, 8, v2
+; GFX10-DL-NEXT:    v_bfe_i32 v6, v2, 0, 8
+; GFX10-DL-NEXT:    v_bfe_i32 v7, v1, 0, 8
 ; GFX10-DL-NEXT:    v_lshrrev_b32_e32 v1, 16, v1
 ; GFX10-DL-NEXT:    v_lshrrev_b32_e32 v2, 16, v2
-; GFX10-DL-NEXT:    v_lshl_or_b32 v6, v6, 16, v7
-; GFX10-DL-NEXT:    v_lshl_or_b32 v5, v5, 16, v8
-; GFX10-DL-NEXT:    v_ashrrev_i16 v7, 8, v1
-; GFX10-DL-NEXT:    v_ashrrev_i16 v8, 8, v2
-; GFX10-DL-NEXT:    v_and_b32_sdwa v2, v4, sext(v2) dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
-; GFX10-DL-NEXT:    v_and_b32_sdwa v1, v4, sext(v1) dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
-; GFX10-DL-NEXT:    v_pk_mul_lo_u16 v4, v5, v6
-; GFX10-DL-NEXT:    v_lshl_or_b32 v2, v8, 16, v2
-; GFX10-DL-NEXT:    v_lshl_or_b32 v1, v7, 16, v1
+; GFX10-DL-NEXT:    v_perm_b32 v5, v5, v6, 0x5040100
+; GFX10-DL-NEXT:    v_perm_b32 v4, v4, v7, 0x5040100
+; GFX10-DL-NEXT:    v_ashrrev_i16 v6, 8, v1
+; GFX10-DL-NEXT:    v_ashrrev_i16 v7, 8, v2
+; GFX10-DL-NEXT:    v_bfe_i32 v2, v2, 0, 8
+; GFX10-DL-NEXT:    v_bfe_i32 v1, v1, 0, 8
+; GFX10-DL-NEXT:    v_pk_mul_lo_u16 v4, v4, v5
+; GFX10-DL-NEXT:    v_perm_b32 v2, v7, v2, 0x5040100
+; GFX10-DL-NEXT:    v_perm_b32 v1, v6, v1, 0x5040100
 ; GFX10-DL-NEXT:    v_lshrrev_b32_e32 v5, 16, v4
 ; GFX10-DL-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-DL-NEXT:    v_add_nc_u16 v3, v4, v3
diff --git a/llvm/test/CodeGen/AMDGPU/idot4u.ll b/llvm/test/CodeGen/AMDGPU/idot4u.ll
--- a/llvm/test/CodeGen/AMDGPU/idot4u.ll
+++ b/llvm/test/CodeGen/AMDGPU/idot4u.ll
@@ -1913,35 +1913,33 @@
 ; GFX9-NODL-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x34
 ; GFX9-NODL-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
 ; GFX9-NODL-NEXT:    s_movk_i32 s0, 0xff
-; GFX9-NODL-NEXT:    v_mov_b32_e32 v4, 0xffff
+; GFX9-NODL-NEXT:    s_mov_b32 s1, 0x5040100
 ; GFX9-NODL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NODL-NEXT:    global_load_dword v1, v0, s[4:5]
 ; GFX9-NODL-NEXT:    global_load_dword v2, v0, s[6:7]
 ; GFX9-NODL-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX9-NODL-NEXT:    global_load_ushort v3, v0, s[2:3]
 ; GFX9-NODL-NEXT:    s_waitcnt vmcnt(2)
-; GFX9-NODL-NEXT:    v_lshrrev_b16_e32 v5, 8, v1
-; GFX9-NODL-NEXT:    v_lshrrev_b32_e32 v6, 24, v1
+; GFX9-NODL-NEXT:    v_lshrrev_b16_e32 v4, 8, v1
+; GFX9-NODL-NEXT:    v_lshrrev_b32_e32 v5, 24, v1
 ; GFX9-NODL-NEXT:    s_waitcnt vmcnt(1)
-; GFX9-NODL-NEXT:    v_lshrrev_b16_e32 v7, 8, v2
-; GFX9-NODL-NEXT:    v_lshrrev_b32_e32 v8, 24, v2
-; GFX9-NODL-NEXT:    v_and_b32_sdwa v9, v1, s0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
-; GFX9-NODL-NEXT:    v_and_b32_sdwa v10, v2, s0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
-; GFX9-NODL-NEXT:    v_and_b32_sdwa v2, v4, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
-; GFX9-NODL-NEXT:    v_and_b32_sdwa v1, v4, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
-; GFX9-NODL-NEXT:    v_lshl_or_b32 v2, v7, 16, v2
-; GFX9-NODL-NEXT:    v_lshl_or_b32 v1, v5, 16, v1
-; GFX9-NODL-NEXT:    v_and_b32_e32 v4, 0xffff, v10
-; GFX9-NODL-NEXT:    v_and_b32_e32 v9, 0xffff, v9
+; GFX9-NODL-NEXT:    v_lshrrev_b16_e32 v6, 8, v2
+; GFX9-NODL-NEXT:    v_lshrrev_b32_e32 v7, 24, v2
+; GFX9-NODL-NEXT:    v_and_b32_e32 v8, 0xff, v1
+; GFX9-NODL-NEXT:    v_and_b32_sdwa v1, v1, s0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX9-NODL-NEXT:    v_and_b32_e32 v9, 0xff, v2
+; GFX9-NODL-NEXT:    v_and_b32_sdwa v2, v2, s0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX9-NODL-NEXT:    v_perm_b32 v2, v7, v2, s1
+; GFX9-NODL-NEXT:    v_perm_b32 v1, v5, v1, s1
+; GFX9-NODL-NEXT:    v_perm_b32 v5, v6, v9, s1
+; GFX9-NODL-NEXT:    v_perm_b32 v4, v4, v8, s1
 ; GFX9-NODL-NEXT:    v_pk_mul_lo_u16 v1, v1, v2
-; GFX9-NODL-NEXT:    v_lshl_or_b32 v4, v8, 16, v4
-; GFX9-NODL-NEXT:    v_lshl_or_b32 v5, v6, 16, v9
+; GFX9-NODL-NEXT:    v_pk_mul_lo_u16 v2, v4, v5
 ; GFX9-NODL-NEXT:    s_waitcnt vmcnt(0)
-; GFX9-NODL-NEXT:    v_add_u16_e32 v3, v1, v3
-; GFX9-NODL-NEXT:    v_pk_mul_lo_u16 v2, v5, v4
-; GFX9-NODL-NEXT:    v_add_u16_sdwa v1, v3, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
-; GFX9-NODL-NEXT:    v_add_u16_e32 v1, v1, v2
-; GFX9-NODL-NEXT:    v_add_u16_sdwa v1, v1, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
+; GFX9-NODL-NEXT:    v_add_u16_e32 v3, v2, v3
+; GFX9-NODL-NEXT:    v_add_u16_sdwa v2, v3, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
+; GFX9-NODL-NEXT:    v_add_u16_e32 v2, v2, v1
+; GFX9-NODL-NEXT:    v_add_u16_sdwa v1, v2, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
 ; GFX9-NODL-NEXT:    global_store_short v0, v1, s[2:3]
 ; GFX9-NODL-NEXT:    s_endpgm
 ;
@@ -1951,35 +1949,33 @@
 ; GFX9-DL-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x34
 ; GFX9-DL-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
 ; GFX9-DL-NEXT:    s_movk_i32 s0, 0xff
-; GFX9-DL-NEXT:    v_mov_b32_e32 v4, 0xffff
+; GFX9-DL-NEXT:    s_mov_b32 s1, 0x5040100
 ; GFX9-DL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-DL-NEXT:    global_load_dword v1, v0, s[4:5]
 ; GFX9-DL-NEXT:    global_load_dword v2, v0, s[6:7]
 ; GFX9-DL-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX9-DL-NEXT:    global_load_ushort v3, v0, s[2:3]
 ; GFX9-DL-NEXT:    s_waitcnt vmcnt(2)
-; GFX9-DL-NEXT:    v_lshrrev_b16_e32 v5, 8, v1
-; GFX9-DL-NEXT:    v_lshrrev_b32_e32 v6, 24, v1
+; GFX9-DL-NEXT:    v_lshrrev_b16_e32 v4, 8, v1
+; GFX9-DL-NEXT:    v_lshrrev_b32_e32 v5, 24, v1
 ; GFX9-DL-NEXT:    s_waitcnt vmcnt(1)
-; GFX9-DL-NEXT:    v_lshrrev_b16_e32 v7, 8, v2
-; GFX9-DL-NEXT:    v_lshrrev_b32_e32 v8, 24, v2
-; GFX9-DL-NEXT:    v_and_b32_sdwa v9, v1, s0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
-; GFX9-DL-NEXT:    v_and_b32_sdwa v10, v2, s0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
-; GFX9-DL-NEXT:    v_and_b32_sdwa v2, v4, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
-; GFX9-DL-NEXT:    v_and_b32_sdwa v1, v4, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
-; GFX9-DL-NEXT:    v_lshl_or_b32 v2, v7, 16, v2
-; GFX9-DL-NEXT:    v_lshl_or_b32 v1, v5, 16, v1
-; GFX9-DL-NEXT:    v_and_b32_e32 v4, 0xffff, v10
-; GFX9-DL-NEXT:    v_and_b32_e32 v9, 0xffff, v9
+; GFX9-DL-NEXT:    v_lshrrev_b16_e32 v6, 8, v2
+; GFX9-DL-NEXT:    v_lshrrev_b32_e32 v7, 24, v2
+; GFX9-DL-NEXT:    v_and_b32_e32 v8, 0xff, v1
+; GFX9-DL-NEXT:    v_and_b32_sdwa v1, v1, s0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX9-DL-NEXT:    v_and_b32_e32 v9, 0xff, v2
+; GFX9-DL-NEXT:    v_and_b32_sdwa v2, v2, s0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX9-DL-NEXT:    v_perm_b32 v2, v7, v2, s1
+; GFX9-DL-NEXT:    v_perm_b32 v1, v5, v1, s1
+; GFX9-DL-NEXT:    v_perm_b32 v5, v6, v9, s1
+; GFX9-DL-NEXT:    v_perm_b32 v4, v4, v8, s1
 ; GFX9-DL-NEXT:    v_pk_mul_lo_u16 v1, v1, v2
-; GFX9-DL-NEXT:    v_lshl_or_b32 v4, v8, 16, v4
-; GFX9-DL-NEXT:    v_lshl_or_b32 v5, v6, 16, v9
+; GFX9-DL-NEXT:    v_pk_mul_lo_u16 v2, v4, v5
 ; GFX9-DL-NEXT:    s_waitcnt vmcnt(0)
-; GFX9-DL-NEXT:    v_add_u16_e32 v3, v1, v3
-; GFX9-DL-NEXT:    v_pk_mul_lo_u16 v2, v5, v4
-; GFX9-DL-NEXT:    v_add_u16_sdwa v1, v3, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
-; GFX9-DL-NEXT:    v_add_u16_e32 v1, v1, v2
-; GFX9-DL-NEXT:    v_add_u16_sdwa v1, v1, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
+; GFX9-DL-NEXT:    v_add_u16_e32 v3, v2, v3
+; GFX9-DL-NEXT:    v_add_u16_sdwa v2, v3, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
+; GFX9-DL-NEXT:    v_add_u16_e32 v2, v2, v1
+; GFX9-DL-NEXT:    v_add_u16_sdwa v1, v2, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
 ; GFX9-DL-NEXT:    global_store_short v0, v1, s[2:3]
 ; GFX9-DL-NEXT:    s_endpgm
 ;
@@ -1988,8 +1984,7 @@
 ; GFX10-DL-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
 ; GFX10-DL-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
 ; GFX10-DL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
-; GFX10-DL-NEXT:    v_mov_b32_e32 v4, 0xffff
-; GFX10-DL-NEXT:    v_mov_b32_e32 v5, 0xff
+; GFX10-DL-NEXT:    v_mov_b32_e32 v8, 0xff
 ; GFX10-DL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-DL-NEXT:    s_clause 0x1
 ; GFX10-DL-NEXT:    global_load_dword v1, v0, s[4:5]
@@ -1997,22 +1992,20 @@
 ; GFX10-DL-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX10-DL-NEXT:    global_load_ushort v3, v0, s[0:1]
 ; GFX10-DL-NEXT:    s_waitcnt vmcnt(2)
-; GFX10-DL-NEXT:    v_lshrrev_b16 v6, 8, v1
+; GFX10-DL-NEXT:    v_lshrrev_b16 v4, 8, v1
 ; GFX10-DL-NEXT:    s_waitcnt vmcnt(1)
-; GFX10-DL-NEXT:    v_lshrrev_b16 v7, 8, v2
-; GFX10-DL-NEXT:    v_and_b32_sdwa v8, v4, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
-; GFX10-DL-NEXT:    v_and_b32_sdwa v4, v4, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
-; GFX10-DL-NEXT:    v_and_b32_sdwa v9, v1, v5 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
-; GFX10-DL-NEXT:    v_and_b32_sdwa v5, v2, v5 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
-; GFX10-DL-NEXT:    v_lshrrev_b32_e32 v1, 24, v1
-; GFX10-DL-NEXT:    v_lshl_or_b32 v7, v7, 16, v8
-; GFX10-DL-NEXT:    v_lshl_or_b32 v4, v6, 16, v4
-; GFX10-DL-NEXT:    v_lshrrev_b32_e32 v2, 24, v2
-; GFX10-DL-NEXT:    v_and_b32_e32 v5, 0xffff, v5
-; GFX10-DL-NEXT:    v_and_b32_e32 v6, 0xffff, v9
-; GFX10-DL-NEXT:    v_pk_mul_lo_u16 v4, v4, v7
-; GFX10-DL-NEXT:    v_lshl_or_b32 v2, v2, 16, v5
-; GFX10-DL-NEXT:    v_lshl_or_b32 v1, v1, 16, v6
+; GFX10-DL-NEXT:    v_lshrrev_b16 v5, 8, v2
+; GFX10-DL-NEXT:    v_and_b32_e32 v6, 0xff, v2
+; GFX10-DL-NEXT:    v_and_b32_e32 v7, 0xff, v1
+; GFX10-DL-NEXT:    v_perm_b32 v5, v5, v6, 0x5040100
+; GFX10-DL-NEXT:    v_perm_b32 v4, v4, v7, 0x5040100
+; GFX10-DL-NEXT:    v_lshrrev_b32_e32 v6, 24, v1
+; GFX10-DL-NEXT:    v_lshrrev_b32_e32 v7, 24, v2
+; GFX10-DL-NEXT:    v_and_b32_sdwa v2, v2, v8 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX10-DL-NEXT:    v_and_b32_sdwa v1, v1, v8 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX10-DL-NEXT:    v_pk_mul_lo_u16 v4, v4, v5
+; GFX10-DL-NEXT:    v_perm_b32 v2, v7, v2, 0x5040100
+; GFX10-DL-NEXT:    v_perm_b32 v1, v6, v1, 0x5040100
 ; GFX10-DL-NEXT:    v_lshrrev_b32_e32 v5, 16, v4
 ; GFX10-DL-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-DL-NEXT:    v_add_nc_u16 v3, v4, v3
diff --git a/llvm/test/CodeGen/AMDGPU/idot8s.ll b/llvm/test/CodeGen/AMDGPU/idot8s.ll
--- a/llvm/test/CodeGen/AMDGPU/idot8s.ll
+++ b/llvm/test/CodeGen/AMDGPU/idot8s.ll
@@ -2345,68 +2345,69 @@
 ; GFX9-NEXT:    global_load_dword v2, v0, s[6:7]
 ; GFX9-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX9-NEXT:    global_load_ushort v3, v0, s[2:3]
+; GFX9-NEXT:    s_mov_b32 s0, 0x5040100
 ; GFX9-NEXT:    s_addc_u32 s9, s9, 0
 ; GFX9-NEXT:    s_waitcnt vmcnt(2)
-; GFX9-NEXT:    v_lshrrev_b32_e32 v5, 4, v1
-; GFX9-NEXT:    v_lshlrev_b16_e32 v6, 12, v1
-; GFX9-NEXT:    v_lshrrev_b32_e32 v7, 12, v1
-; GFX9-NEXT:    v_lshrrev_b32_e32 v8, 8, v1
-; GFX9-NEXT:    v_lshrrev_b32_e32 v9, 20, v1
-; GFX9-NEXT:    v_lshlrev_b16_sdwa v10, v4, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
-; GFX9-NEXT:    v_lshrrev_b32_e32 v11, 28, v1
+; GFX9-NEXT:    v_lshrrev_b32_e32 v6, 4, v1
+; GFX9-NEXT:    v_lshrrev_b32_e32 v7, 8, v1
+; GFX9-NEXT:    v_lshrrev_b32_e32 v8, 12, v1
 ; GFX9-NEXT:    s_waitcnt vmcnt(1)
-; GFX9-NEXT:    v_lshrrev_b32_e32 v12, 4, v2
-; GFX9-NEXT:    v_lshlrev_b16_sdwa v1, v4, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_3
-; GFX9-NEXT:    v_lshlrev_b16_e32 v13, 12, v2
-; GFX9-NEXT:    v_lshrrev_b32_e32 v14, 12, v2
-; GFX9-NEXT:    v_lshrrev_b32_e32 v15, 8, v2
-; GFX9-NEXT:    v_lshrrev_b32_e32 v16, 20, v2
-; GFX9-NEXT:    v_lshlrev_b16_sdwa v17, v4, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
-; GFX9-NEXT:    v_lshrrev_b32_e32 v18, 28, v2
-; GFX9-NEXT:    v_lshlrev_b16_sdwa v2, v4, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_3
-; GFX9-NEXT:    v_lshlrev_b16_e32 v4, 12, v5
-; GFX9-NEXT:    v_ashrrev_i16_e32 v5, 12, v6
-; GFX9-NEXT:    v_lshlrev_b16_e32 v6, 12, v7
-; GFX9-NEXT:    v_lshlrev_b16_e32 v7, 12, v8
-; GFX9-NEXT:    v_lshlrev_b16_e32 v8, 12, v9
-; GFX9-NEXT:    v_ashrrev_i16_e32 v9, 12, v10
-; GFX9-NEXT:    v_lshlrev_b16_e32 v10, 12, v11
-; GFX9-NEXT:    v_lshlrev_b16_e32 v11, 12, v12
-; GFX9-NEXT:    v_ashrrev_i16_e32 v12, 12, v13
-; GFX9-NEXT:    v_ashrrev_i16_e32 v4, 12, v4
+; GFX9-NEXT:    v_lshrrev_b32_e32 v13, 4, v2
+; GFX9-NEXT:    v_lshlrev_b16_e32 v5, 12, v1
+; GFX9-NEXT:    v_lshlrev_b16_e32 v12, 12, v2
+; GFX9-NEXT:    v_lshlrev_b16_e32 v6, 12, v6
+; GFX9-NEXT:    v_lshlrev_b16_e32 v7, 12, v7
+; GFX9-NEXT:    v_lshlrev_b16_e32 v8, 12, v8
+; GFX9-NEXT:    v_lshlrev_b16_e32 v13, 12, v13
+; GFX9-NEXT:    v_lshlrev_b16_sdwa v9, v4, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
+; GFX9-NEXT:    v_lshrrev_b32_e32 v10, 20, v1
+; GFX9-NEXT:    v_lshlrev_b16_sdwa v11, v4, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_3
+; GFX9-NEXT:    v_lshrrev_b32_e32 v1, 28, v1
+; GFX9-NEXT:    v_lshrrev_b32_e32 v14, 8, v2
+; GFX9-NEXT:    v_lshrrev_b32_e32 v15, 12, v2
+; GFX9-NEXT:    v_lshlrev_b16_sdwa v16, v4, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
+; GFX9-NEXT:    v_lshrrev_b32_e32 v17, 20, v2
+; GFX9-NEXT:    v_lshlrev_b16_sdwa v4, v4, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_3
+; GFX9-NEXT:    v_lshrrev_b32_e32 v2, 28, v2
+; GFX9-NEXT:    v_ashrrev_i16_e32 v5, 12, v5
+; GFX9-NEXT:    v_ashrrev_i16_e32 v12, 12, v12
 ; GFX9-NEXT:    v_ashrrev_i16_e32 v6, 12, v6
 ; GFX9-NEXT:    v_ashrrev_i16_e32 v7, 12, v7
-; GFX9-NEXT:    v_ashrrev_i16_e32 v11, 12, v11
-; GFX9-NEXT:    v_lshlrev_b16_e32 v13, 12, v14
-; GFX9-NEXT:    v_lshlrev_b16_e32 v14, 12, v15
-; GFX9-NEXT:    v_lshl_or_b32 v6, v6, 16, v7
-; GFX9-NEXT:    v_lshl_or_b32 v7, v11, 16, v12
-; GFX9-NEXT:    v_lshl_or_b32 v4, v4, 16, v5
 ; GFX9-NEXT:    v_ashrrev_i16_e32 v8, 12, v8
 ; GFX9-NEXT:    v_ashrrev_i16_e32 v13, 12, v13
-; GFX9-NEXT:    v_ashrrev_i16_e32 v14, 12, v14
-; GFX9-NEXT:    v_pk_mul_lo_u16 v4, v4, v7
-; GFX9-NEXT:    v_lshlrev_b16_e32 v15, 12, v16
-; GFX9-NEXT:    v_ashrrev_i16_e32 v16, 12, v17
-; GFX9-NEXT:    v_lshlrev_b16_e32 v17, 12, v18
-; GFX9-NEXT:    v_lshl_or_b32 v8, v8, 16, v9
-; GFX9-NEXT:    v_lshl_or_b32 v9, v13, 16, v14
-; GFX9-NEXT:    s_waitcnt vmcnt(0)
-; GFX9-NEXT:    v_add_u16_e32 v3, v4, v3
-; GFX9-NEXT:    v_ashrrev_i16_e32 v1, 12, v1
-; GFX9-NEXT:    v_ashrrev_i16_e32 v2, 12, v2
+; GFX9-NEXT:    v_lshlrev_b16_e32 v10, 12, v10
+; GFX9-NEXT:    v_lshlrev_b16_e32 v1, 12, v1
+; GFX9-NEXT:    v_lshlrev_b16_e32 v14, 12, v14
+; GFX9-NEXT:    v_lshlrev_b16_e32 v15, 12, v15
+; GFX9-NEXT:    v_lshlrev_b16_e32 v17, 12, v17
+; GFX9-NEXT:    v_lshlrev_b16_e32 v2, 12, v2
+; GFX9-NEXT:    v_perm_b32 v7, v8, v7, s0
+; GFX9-NEXT:    v_perm_b32 v8, v13, v12, s0
+; GFX9-NEXT:    v_perm_b32 v5, v6, v5, s0
+; GFX9-NEXT:    v_ashrrev_i16_e32 v9, 12, v9
+; GFX9-NEXT:    v_ashrrev_i16_e32 v11, 12, v11
+; GFX9-NEXT:    v_ashrrev_i16_e32 v16, 12, v16
+; GFX9-NEXT:    v_ashrrev_i16_e32 v4, 12, v4
 ; GFX9-NEXT:    v_ashrrev_i16_e32 v10, 12, v10
+; GFX9-NEXT:    v_ashrrev_i16_e32 v1, 12, v1
+; GFX9-NEXT:    v_ashrrev_i16_e32 v14, 12, v14
 ; GFX9-NEXT:    v_ashrrev_i16_e32 v15, 12, v15
 ; GFX9-NEXT:    v_ashrrev_i16_e32 v17, 12, v17
-; GFX9-NEXT:    v_pk_mul_lo_u16 v5, v6, v9
-; GFX9-NEXT:    v_add_u16_sdwa v3, v3, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
-; GFX9-NEXT:    v_lshl_or_b32 v2, v17, 16, v2
-; GFX9-NEXT:    v_lshl_or_b32 v1, v10, 16, v1
-; GFX9-NEXT:    v_lshl_or_b32 v10, v15, 16, v16
-; GFX9-NEXT:    v_add_u16_e32 v3, v3, v5
+; GFX9-NEXT:    v_ashrrev_i16_e32 v2, 12, v2
+; GFX9-NEXT:    v_pk_mul_lo_u16 v5, v5, v8
+; GFX9-NEXT:    v_perm_b32 v2, v2, v4, s0
+; GFX9-NEXT:    v_perm_b32 v1, v1, v11, s0
+; GFX9-NEXT:    v_perm_b32 v4, v17, v16, s0
+; GFX9-NEXT:    v_perm_b32 v9, v10, v9, s0
+; GFX9-NEXT:    v_perm_b32 v10, v15, v14, s0
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    v_add_u16_e32 v3, v5, v3
 ; GFX9-NEXT:    v_pk_mul_lo_u16 v1, v1, v2
-; GFX9-NEXT:    v_pk_mul_lo_u16 v2, v8, v10
+; GFX9-NEXT:    v_pk_mul_lo_u16 v2, v9, v4
+; GFX9-NEXT:    v_pk_mul_lo_u16 v4, v7, v10
 ; GFX9-NEXT:    v_add_u16_sdwa v3, v3, v5 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
+; GFX9-NEXT:    v_add_u16_e32 v3, v3, v4
+; GFX9-NEXT:    v_add_u16_sdwa v3, v3, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
 ; GFX9-NEXT:    v_add_u16_e32 v3, v3, v2
 ; GFX9-NEXT:    v_add_u16_sdwa v2, v3, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
 ; GFX9-NEXT:    v_add_u16_e32 v2, v2, v1
@@ -2430,68 +2431,69 @@
 ; GFX9-DL-NEXT:    global_load_dword v2, v0, s[6:7]
 ; GFX9-DL-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX9-DL-NEXT:    global_load_ushort v3, v0, s[2:3]
+; GFX9-DL-NEXT:    s_mov_b32 s0, 0x5040100
 ; GFX9-DL-NEXT:    s_addc_u32 s9, s9, 0
 ; GFX9-DL-NEXT:    s_waitcnt vmcnt(2)
-; GFX9-DL-NEXT:    v_lshrrev_b32_e32 v5, 4, v1
-; GFX9-DL-NEXT:    v_lshlrev_b16_e32 v6, 12, v1
-; GFX9-DL-NEXT:    v_lshrrev_b32_e32 v7, 12, v1
-; GFX9-DL-NEXT:    v_lshrrev_b32_e32 v8, 8, v1
-; GFX9-DL-NEXT:    v_lshrrev_b32_e32 v9, 20, v1
-; GFX9-DL-NEXT:    v_lshlrev_b16_sdwa v10, v4, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
-; GFX9-DL-NEXT:    v_lshrrev_b32_e32 v11, 28, v1
+; GFX9-DL-NEXT:    v_lshrrev_b32_e32 v6, 4, v1
+; GFX9-DL-NEXT:    v_lshrrev_b32_e32 v7, 8, v1
+; GFX9-DL-NEXT:    v_lshrrev_b32_e32 v8, 12, v1
 ; GFX9-DL-NEXT:    s_waitcnt vmcnt(1)
-; GFX9-DL-NEXT:    v_lshrrev_b32_e32 v12, 4, v2
-; GFX9-DL-NEXT:    v_lshlrev_b16_sdwa v1, v4, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_3
-; GFX9-DL-NEXT:    v_lshlrev_b16_e32 v13, 12, v2
-; GFX9-DL-NEXT:    v_lshrrev_b32_e32 v14, 12, v2
-; GFX9-DL-NEXT:    v_lshrrev_b32_e32 v15, 8, v2
-; GFX9-DL-NEXT:    v_lshrrev_b32_e32 v16, 20, v2
-; GFX9-DL-NEXT:    v_lshlrev_b16_sdwa v17, v4, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
-; GFX9-DL-NEXT:    v_lshrrev_b32_e32 v18, 28, v2
-; GFX9-DL-NEXT:    v_lshlrev_b16_sdwa v2, v4, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_3
-; GFX9-DL-NEXT:    v_lshlrev_b16_e32 v4, 12, v5
-; GFX9-DL-NEXT:    v_ashrrev_i16_e32 v5, 12, v6
-; GFX9-DL-NEXT:    v_lshlrev_b16_e32 v6, 12, v7
-; GFX9-DL-NEXT:    v_lshlrev_b16_e32 v7, 12, v8
-; GFX9-DL-NEXT:    v_lshlrev_b16_e32 v8, 12, v9
-; GFX9-DL-NEXT:    v_ashrrev_i16_e32 v9, 12, v10
-; GFX9-DL-NEXT:    v_lshlrev_b16_e32 v10, 12, v11
-; GFX9-DL-NEXT:    v_lshlrev_b16_e32 v11, 12, v12
-; GFX9-DL-NEXT:    v_ashrrev_i16_e32 v12, 12, v13
-; GFX9-DL-NEXT:    v_ashrrev_i16_e32 v4, 12, v4
+; GFX9-DL-NEXT:    v_lshrrev_b32_e32 v13, 4, v2
+; GFX9-DL-NEXT:    v_lshlrev_b16_e32 v5, 12, v1
+; GFX9-DL-NEXT:    v_lshlrev_b16_e32 v12, 12, v2
+; GFX9-DL-NEXT:    v_lshlrev_b16_e32 v6, 12, v6
+; GFX9-DL-NEXT:    v_lshlrev_b16_e32 v7, 12, v7
+; GFX9-DL-NEXT:    v_lshlrev_b16_e32 v8, 12, v8
+; GFX9-DL-NEXT:    v_lshlrev_b16_e32 v13, 12, v13
+; GFX9-DL-NEXT:    v_lshlrev_b16_sdwa v9, v4, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
+; GFX9-DL-NEXT:    v_lshrrev_b32_e32 v10, 20, v1
+; GFX9-DL-NEXT:    v_lshlrev_b16_sdwa v11, v4, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_3
+; GFX9-DL-NEXT:    v_lshrrev_b32_e32 v1, 28, v1
+; GFX9-DL-NEXT:    v_lshrrev_b32_e32 v14, 8, v2
+; GFX9-DL-NEXT:    v_lshrrev_b32_e32 v15, 12, v2
+; GFX9-DL-NEXT:    v_lshlrev_b16_sdwa v16, v4, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
+; GFX9-DL-NEXT:    v_lshrrev_b32_e32 v17, 20, v2
+; GFX9-DL-NEXT:    v_lshlrev_b16_sdwa v4, v4, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_3
+; GFX9-DL-NEXT:    v_lshrrev_b32_e32 v2, 28, v2
+; GFX9-DL-NEXT:    v_ashrrev_i16_e32 v5, 12, v5
+; GFX9-DL-NEXT:    v_ashrrev_i16_e32 v12, 12, v12
 ; GFX9-DL-NEXT:    v_ashrrev_i16_e32 v6, 12, v6
 ; GFX9-DL-NEXT:    v_ashrrev_i16_e32 v7, 12, v7
-; GFX9-DL-NEXT:    v_ashrrev_i16_e32 v11, 12, v11
-; GFX9-DL-NEXT:    v_lshlrev_b16_e32 v13, 12, v14
-; GFX9-DL-NEXT:    v_lshlrev_b16_e32 v14, 12, v15
-; GFX9-DL-NEXT:    v_lshl_or_b32 v6, v6, 16, v7
-; GFX9-DL-NEXT:    v_lshl_or_b32 v7, v11, 16, v12
-; GFX9-DL-NEXT:    v_lshl_or_b32 v4, v4, 16, v5
 ; GFX9-DL-NEXT:    v_ashrrev_i16_e32 v8, 12, v8
 ; GFX9-DL-NEXT:    v_ashrrev_i16_e32 v13, 12, v13
-; GFX9-DL-NEXT:    v_ashrrev_i16_e32 v14, 12, v14
-; GFX9-DL-NEXT:    v_pk_mul_lo_u16 v4, v4, v7
-; GFX9-DL-NEXT:    v_lshlrev_b16_e32 v15, 12, v16
-; GFX9-DL-NEXT:    v_ashrrev_i16_e32 v16, 12, v17
-; GFX9-DL-NEXT:    v_lshlrev_b16_e32 v17, 12, v18
-; GFX9-DL-NEXT:    v_lshl_or_b32 v8, v8, 16, v9
-; GFX9-DL-NEXT:    v_lshl_or_b32 v9, v13, 16, v14
-; GFX9-DL-NEXT:    s_waitcnt vmcnt(0)
-; GFX9-DL-NEXT:    v_add_u16_e32 v3, v4, v3
-; GFX9-DL-NEXT:    v_ashrrev_i16_e32 v1, 12, v1
-; GFX9-DL-NEXT:    v_ashrrev_i16_e32 v2, 12, v2
+; GFX9-DL-NEXT:    v_lshlrev_b16_e32 v10, 12, v10
+; GFX9-DL-NEXT:    v_lshlrev_b16_e32 v1, 12, v1
+; GFX9-DL-NEXT:    v_lshlrev_b16_e32 v14, 12, v14
+; GFX9-DL-NEXT:    v_lshlrev_b16_e32 v15, 12, v15
+; GFX9-DL-NEXT:    v_lshlrev_b16_e32 v17, 12, v17
+; GFX9-DL-NEXT:    v_lshlrev_b16_e32 v2, 12, v2
+; GFX9-DL-NEXT:    v_perm_b32 v7, v8, v7, s0
+; GFX9-DL-NEXT:    v_perm_b32 v8, v13, v12, s0
+; GFX9-DL-NEXT:    v_perm_b32 v5, v6, v5, s0
+; GFX9-DL-NEXT:    v_ashrrev_i16_e32 v9, 12, v9
+; GFX9-DL-NEXT:    v_ashrrev_i16_e32 v11, 12, v11
+; GFX9-DL-NEXT:    v_ashrrev_i16_e32 v16, 12, v16
+; GFX9-DL-NEXT:    v_ashrrev_i16_e32 v4, 12, v4
 ; GFX9-DL-NEXT:    v_ashrrev_i16_e32 v10, 12, v10
+; GFX9-DL-NEXT:    v_ashrrev_i16_e32 v1, 12, v1
+; GFX9-DL-NEXT:    v_ashrrev_i16_e32 v14, 12, v14
 ; GFX9-DL-NEXT:    v_ashrrev_i16_e32 v15, 12, v15
 ; GFX9-DL-NEXT:    v_ashrrev_i16_e32 v17, 12, v17
-; GFX9-DL-NEXT:    v_pk_mul_lo_u16 v5, v6, v9
-; GFX9-DL-NEXT:    v_add_u16_sdwa v3, v3, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
-; GFX9-DL-NEXT:    v_lshl_or_b32 v2, v17, 16, v2
-; GFX9-DL-NEXT:    v_lshl_or_b32 v1, v10, 16, v1
-; GFX9-DL-NEXT:    v_lshl_or_b32 v10, v15, 16, v16
-; GFX9-DL-NEXT:    v_add_u16_e32 v3, v3, v5
+; GFX9-DL-NEXT:    v_ashrrev_i16_e32 v2, 12, v2
+; GFX9-DL-NEXT:    v_pk_mul_lo_u16 v5, v5, v8
+; GFX9-DL-NEXT:    v_perm_b32 v2, v2, v4, s0
+; GFX9-DL-NEXT:    v_perm_b32 v1, v1, v11, s0
+; GFX9-DL-NEXT:    v_perm_b32 v4, v17, v16, s0
+; GFX9-DL-NEXT:    v_perm_b32 v9, v10, v9, s0
+; GFX9-DL-NEXT:    v_perm_b32 v10, v15, v14, s0
+; GFX9-DL-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-DL-NEXT:    v_add_u16_e32 v3, v5, v3
 ; GFX9-DL-NEXT:    v_pk_mul_lo_u16 v1, v1, v2
-; GFX9-DL-NEXT:    v_pk_mul_lo_u16 v2, v8, v10
+; GFX9-DL-NEXT:    v_pk_mul_lo_u16 v2, v9, v4
+; GFX9-DL-NEXT:    v_pk_mul_lo_u16 v4, v7, v10
 ; GFX9-DL-NEXT:    v_add_u16_sdwa v3, v3, v5 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
+; GFX9-DL-NEXT:    v_add_u16_e32 v3, v3, v4
+; GFX9-DL-NEXT:    v_add_u16_sdwa v3, v3, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
 ; GFX9-DL-NEXT:    v_add_u16_e32 v3, v3, v2
 ; GFX9-DL-NEXT:    v_add_u16_sdwa v2, v3, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
 ; GFX9-DL-NEXT:    v_add_u16_e32 v2, v2, v1
@@ -2517,79 +2519,71 @@
 ; GFX10-DL-XNACK-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX10-DL-XNACK-NEXT:    global_load_ushort v3, v0, s[0:1]
 ; GFX10-DL-XNACK-NEXT:    s_waitcnt vmcnt(2)
-; GFX10-DL-XNACK-NEXT:    v_lshrrev_b32_e32 v4, 4, v1
-; GFX10-DL-XNACK-NEXT:    v_lshlrev_b16 v5, 12, v1
+; GFX10-DL-XNACK-NEXT:    v_lshrrev_b32_e32 v5, 4, v1
 ; GFX10-DL-XNACK-NEXT:    s_waitcnt vmcnt(1)
-; GFX10-DL-XNACK-NEXT:    v_lshrrev_b32_e32 v11, 4, v2
-; GFX10-DL-XNACK-NEXT:    v_lshlrev_b16 v12, 12, v2
-; GFX10-DL-XNACK-NEXT:    v_lshrrev_b32_e32 v7, 8, v1
-; GFX10-DL-XNACK-NEXT:    v_lshrrev_b32_e32 v14, 8, v2
-; GFX10-DL-XNACK-NEXT:    v_lshlrev_b16 v4, 12, v4
-; GFX10-DL-XNACK-NEXT:    v_ashrrev_i16 v5, 12, v5
-; GFX10-DL-XNACK-NEXT:    v_lshlrev_b16 v11, 12, v11
-; GFX10-DL-XNACK-NEXT:    v_ashrrev_i16 v12, 12, v12
-; GFX10-DL-XNACK-NEXT:    v_lshrrev_b32_e32 v6, 12, v1
-; GFX10-DL-XNACK-NEXT:    v_lshrrev_b32_e32 v13, 12, v2
-; GFX10-DL-XNACK-NEXT:    v_lshlrev_b16 v7, 12, v7
-; GFX10-DL-XNACK-NEXT:    v_lshlrev_b16 v14, 12, v14
+; GFX10-DL-XNACK-NEXT:    v_lshrrev_b32_e32 v12, 4, v2
+; GFX10-DL-XNACK-NEXT:    v_lshlrev_b16 v4, 12, v1
+; GFX10-DL-XNACK-NEXT:    v_lshlrev_b16 v11, 12, v2
+; GFX10-DL-XNACK-NEXT:    v_lshrrev_b32_e32 v6, 8, v1
+; GFX10-DL-XNACK-NEXT:    v_lshlrev_b16 v5, 12, v5
+; GFX10-DL-XNACK-NEXT:    v_lshlrev_b16 v12, 12, v12
+; GFX10-DL-XNACK-NEXT:    v_lshrrev_b32_e32 v7, 12, v1
+; GFX10-DL-XNACK-NEXT:    v_lshrrev_b32_e32 v8, 16, v1
+; GFX10-DL-XNACK-NEXT:    v_lshrrev_b32_e32 v13, 8, v2
+; GFX10-DL-XNACK-NEXT:    v_lshrrev_b32_e32 v14, 12, v2
 ; GFX10-DL-XNACK-NEXT:    v_ashrrev_i16 v4, 12, v4
 ; GFX10-DL-XNACK-NEXT:    v_ashrrev_i16 v11, 12, v11
-; GFX10-DL-XNACK-NEXT:    v_and_b32_e32 v12, 0xffff, v12
-; GFX10-DL-XNACK-NEXT:    v_and_b32_e32 v5, 0xffff, v5
-; GFX10-DL-XNACK-NEXT:    v_lshrrev_b32_e32 v9, 16, v1
-; GFX10-DL-XNACK-NEXT:    v_lshrrev_b32_e32 v16, 16, v2
+; GFX10-DL-XNACK-NEXT:    v_ashrrev_i16 v12, 12, v12
+; GFX10-DL-XNACK-NEXT:    v_ashrrev_i16 v5, 12, v5
 ; GFX10-DL-XNACK-NEXT:    v_lshlrev_b16 v6, 12, v6
+; GFX10-DL-XNACK-NEXT:    v_lshlrev_b16 v7, 12, v7
+; GFX10-DL-XNACK-NEXT:    v_lshlrev_b16 v8, 12, v8
 ; GFX10-DL-XNACK-NEXT:    v_lshlrev_b16 v13, 12, v13
-; GFX10-DL-XNACK-NEXT:    v_ashrrev_i16 v7, 12, v7
-; GFX10-DL-XNACK-NEXT:    v_ashrrev_i16 v14, 12, v14
-; GFX10-DL-XNACK-NEXT:    v_lshl_or_b32 v11, v11, 16, v12
-; GFX10-DL-XNACK-NEXT:    v_lshl_or_b32 v4, v4, 16, v5
-; GFX10-DL-XNACK-NEXT:    v_lshrrev_b32_e32 v8, 20, v1
-; GFX10-DL-XNACK-NEXT:    v_lshrrev_b32_e32 v15, 20, v2
-; GFX10-DL-XNACK-NEXT:    v_lshlrev_b16 v9, 12, v9
-; GFX10-DL-XNACK-NEXT:    v_lshlrev_b16 v16, 12, v16
+; GFX10-DL-XNACK-NEXT:    v_lshlrev_b16 v14, 12, v14
+; GFX10-DL-XNACK-NEXT:    v_perm_b32 v11, v12, v11, 0x5040100
+; GFX10-DL-XNACK-NEXT:    v_perm_b32 v4, v5, v4, 0x5040100
+; GFX10-DL-XNACK-NEXT:    v_lshrrev_b32_e32 v9, 20, v1
+; GFX10-DL-XNACK-NEXT:    v_lshrrev_b32_e32 v15, 16, v2
+; GFX10-DL-XNACK-NEXT:    v_lshrrev_b32_e32 v16, 20, v2
 ; GFX10-DL-XNACK-NEXT:    v_ashrrev_i16 v6, 12, v6
-; GFX10-DL-XNACK-NEXT:    v_ashrrev_i16 v5, 12, v13
-; GFX10-DL-XNACK-NEXT:    v_and_b32_e32 v12, 0xffff, v14
-; GFX10-DL-XNACK-NEXT:    v_and_b32_e32 v7, 0xffff, v7
+; GFX10-DL-XNACK-NEXT:    v_ashrrev_i16 v7, 12, v7
+; GFX10-DL-XNACK-NEXT:    v_ashrrev_i16 v5, 12, v8
+; GFX10-DL-XNACK-NEXT:    v_ashrrev_i16 v8, 12, v13
+; GFX10-DL-XNACK-NEXT:    v_ashrrev_i16 v12, 12, v14
 ; GFX10-DL-XNACK-NEXT:    v_pk_mul_lo_u16 v4, v4, v11
-; GFX10-DL-XNACK-NEXT:    v_lshrrev_b32_e32 v10, 28, v1
-; GFX10-DL-XNACK-NEXT:    v_lshrrev_b32_e32 v1, 24, v1
-; GFX10-DL-XNACK-NEXT:    v_lshrrev_b32_e32 v17, 28, v2
-; GFX10-DL-XNACK-NEXT:    v_lshrrev_b32_e32 v2, 24, v2
-; GFX10-DL-XNACK-NEXT:    v_lshlrev_b16 v8, 12, v8
+; GFX10-DL-XNACK-NEXT:    v_lshlrev_b16 v9, 12, v9
 ; GFX10-DL-XNACK-NEXT:    v_lshlrev_b16 v15, 12, v15
-; GFX10-DL-XNACK-NEXT:    v_ashrrev_i16 v9, 12, v9
-; GFX10-DL-XNACK-NEXT:    v_ashrrev_i16 v13, 12, v16
-; GFX10-DL-XNACK-NEXT:    v_lshl_or_b32 v5, v5, 16, v12
-; GFX10-DL-XNACK-NEXT:    v_lshl_or_b32 v6, v6, 16, v7
+; GFX10-DL-XNACK-NEXT:    v_lshlrev_b16 v16, 12, v16
+; GFX10-DL-XNACK-NEXT:    v_perm_b32 v8, v12, v8, 0x5040100
+; GFX10-DL-XNACK-NEXT:    v_perm_b32 v6, v7, v6, 0x5040100
 ; GFX10-DL-XNACK-NEXT:    v_lshrrev_b32_e32 v7, 16, v4
 ; GFX10-DL-XNACK-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-DL-XNACK-NEXT:    v_add_nc_u16 v3, v4, v3
-; GFX10-DL-XNACK-NEXT:    v_lshlrev_b16 v1, 12, v1
-; GFX10-DL-XNACK-NEXT:    v_ashrrev_i16 v8, 12, v8
+; GFX10-DL-XNACK-NEXT:    v_lshrrev_b32_e32 v10, 24, v1
+; GFX10-DL-XNACK-NEXT:    v_lshrrev_b32_e32 v1, 28, v1
+; GFX10-DL-XNACK-NEXT:    v_lshrrev_b32_e32 v17, 24, v2
+; GFX10-DL-XNACK-NEXT:    v_ashrrev_i16 v9, 12, v9
 ; GFX10-DL-XNACK-NEXT:    v_ashrrev_i16 v11, 12, v15
-; GFX10-DL-XNACK-NEXT:    v_lshlrev_b16 v2, 12, v2
-; GFX10-DL-XNACK-NEXT:    v_and_b32_e32 v4, 0xffff, v13
-; GFX10-DL-XNACK-NEXT:    v_and_b32_e32 v9, 0xffff, v9
-; GFX10-DL-XNACK-NEXT:    v_pk_mul_lo_u16 v5, v6, v5
+; GFX10-DL-XNACK-NEXT:    v_lshrrev_b32_e32 v2, 28, v2
+; GFX10-DL-XNACK-NEXT:    v_ashrrev_i16 v4, 12, v16
+; GFX10-DL-XNACK-NEXT:    v_pk_mul_lo_u16 v6, v6, v8
 ; GFX10-DL-XNACK-NEXT:    v_add_nc_u16 v3, v3, v7
 ; GFX10-DL-XNACK-NEXT:    v_lshlrev_b16 v10, 12, v10
-; GFX10-DL-XNACK-NEXT:    v_lshlrev_b16 v17, 12, v17
-; GFX10-DL-XNACK-NEXT:    v_ashrrev_i16 v1, 12, v1
-; GFX10-DL-XNACK-NEXT:    v_ashrrev_i16 v2, 12, v2
-; GFX10-DL-XNACK-NEXT:    v_lshl_or_b32 v4, v11, 16, v4
-; GFX10-DL-XNACK-NEXT:    v_lshl_or_b32 v6, v8, 16, v9
-; GFX10-DL-XNACK-NEXT:    v_lshrrev_b32_e32 v7, 16, v5
-; GFX10-DL-XNACK-NEXT:    v_add_nc_u16 v3, v3, v5
+; GFX10-DL-XNACK-NEXT:    v_lshlrev_b16 v1, 12, v1
+; GFX10-DL-XNACK-NEXT:    v_lshlrev_b16 v12, 12, v17
+; GFX10-DL-XNACK-NEXT:    v_lshlrev_b16 v2, 12, v2
+; GFX10-DL-XNACK-NEXT:    v_perm_b32 v4, v4, v11, 0x5040100
+; GFX10-DL-XNACK-NEXT:    v_perm_b32 v5, v9, v5, 0x5040100
+; GFX10-DL-XNACK-NEXT:    v_lshrrev_b32_e32 v7, 16, v6
+; GFX10-DL-XNACK-NEXT:    v_add_nc_u16 v3, v3, v6
 ; GFX10-DL-XNACK-NEXT:    v_ashrrev_i16 v10, 12, v10
-; GFX10-DL-XNACK-NEXT:    v_ashrrev_i16 v5, 12, v17
-; GFX10-DL-XNACK-NEXT:    v_and_b32_e32 v2, 0xffff, v2
-; GFX10-DL-XNACK-NEXT:    v_and_b32_e32 v1, 0xffff, v1
-; GFX10-DL-XNACK-NEXT:    v_pk_mul_lo_u16 v4, v6, v4
+; GFX10-DL-XNACK-NEXT:    v_ashrrev_i16 v6, 12, v12
+; GFX10-DL-XNACK-NEXT:    v_ashrrev_i16 v2, 12, v2
+; GFX10-DL-XNACK-NEXT:    v_ashrrev_i16 v1, 12, v1
+; GFX10-DL-XNACK-NEXT:    v_pk_mul_lo_u16 v4, v5, v4
 ; GFX10-DL-XNACK-NEXT:    v_add_nc_u16 v3, v3, v7
-; GFX10-DL-XNACK-NEXT:    v_lshl_or_b32 v2, v5, 16, v2
-; GFX10-DL-XNACK-NEXT:    v_lshl_or_b32 v1, v10, 16, v1
+; GFX10-DL-XNACK-NEXT:    v_perm_b32 v2, v2, v6, 0x5040100
+; GFX10-DL-XNACK-NEXT:    v_perm_b32 v1, v1, v10, 0x5040100
 ; GFX10-DL-XNACK-NEXT:    v_lshrrev_b32_e32 v5, 16, v4
 ; GFX10-DL-XNACK-NEXT:    v_add_nc_u16 v3, v3, v4
 ; GFX10-DL-XNACK-NEXT:    v_pk_mul_lo_u16 v1, v1, v2
@@ -2618,79 +2612,71 @@
 ; GFX10-DL-NOXNACK-NEXT:    global_load_dword v0, v0, s[6:7]
 ; GFX10-DL-NOXNACK-NEXT:    global_load_ushort v3, v2, s[0:1]
 ; GFX10-DL-NOXNACK-NEXT:    s_waitcnt vmcnt(2)
-; GFX10-DL-NOXNACK-NEXT:    v_lshrrev_b32_e32 v4, 4, v1
-; GFX10-DL-NOXNACK-NEXT:    v_lshlrev_b16 v5, 12, v1
+; GFX10-DL-NOXNACK-NEXT:    v_lshrrev_b32_e32 v5, 4, v1
 ; GFX10-DL-NOXNACK-NEXT:    s_waitcnt vmcnt(1)
-; GFX10-DL-NOXNACK-NEXT:    v_lshrrev_b32_e32 v11, 4, v0
-; GFX10-DL-NOXNACK-NEXT:    v_lshlrev_b16 v12, 12, v0
-; GFX10-DL-NOXNACK-NEXT:    v_lshrrev_b32_e32 v7, 8, v1
-; GFX10-DL-NOXNACK-NEXT:    v_lshrrev_b32_e32 v14, 8, v0
-; GFX10-DL-NOXNACK-NEXT:    v_lshlrev_b16 v4, 12, v4
-; GFX10-DL-NOXNACK-NEXT:    v_ashrrev_i16 v5, 12, v5
-; GFX10-DL-NOXNACK-NEXT:    v_lshlrev_b16 v11, 12, v11
-; GFX10-DL-NOXNACK-NEXT:    v_ashrrev_i16 v12, 12, v12
-; GFX10-DL-NOXNACK-NEXT:    v_lshrrev_b32_e32 v6, 12, v1
-; GFX10-DL-NOXNACK-NEXT:    v_lshrrev_b32_e32 v13, 12, v0
-; GFX10-DL-NOXNACK-NEXT:    v_lshlrev_b16 v7, 12, v7
-; GFX10-DL-NOXNACK-NEXT:    v_lshlrev_b16 v14, 12, v14
+; GFX10-DL-NOXNACK-NEXT:    v_lshrrev_b32_e32 v12, 4, v0
+; GFX10-DL-NOXNACK-NEXT:    v_lshlrev_b16 v4, 12, v1
+; GFX10-DL-NOXNACK-NEXT:    v_lshlrev_b16 v11, 12, v0
+; GFX10-DL-NOXNACK-NEXT:    v_lshrrev_b32_e32 v6, 8, v1
+; GFX10-DL-NOXNACK-NEXT:    v_lshlrev_b16 v5, 12, v5
+; GFX10-DL-NOXNACK-NEXT:    v_lshlrev_b16 v12, 12, v12
+; GFX10-DL-NOXNACK-NEXT:    v_lshrrev_b32_e32 v7, 12, v1
+; GFX10-DL-NOXNACK-NEXT:    v_lshrrev_b32_e32 v8, 16, v1
+; GFX10-DL-NOXNACK-NEXT:    v_lshrrev_b32_e32 v13, 8, v0
+; GFX10-DL-NOXNACK-NEXT:    v_lshrrev_b32_e32 v14, 12, v0
 ; GFX10-DL-NOXNACK-NEXT:    v_ashrrev_i16 v4, 12, v4
 ; GFX10-DL-NOXNACK-NEXT:    v_ashrrev_i16 v11, 12, v11
-; GFX10-DL-NOXNACK-NEXT:    v_and_b32_e32 v12, 0xffff, v12
-; GFX10-DL-NOXNACK-NEXT:    v_and_b32_e32 v5, 0xffff, v5
-; GFX10-DL-NOXNACK-NEXT:    v_lshrrev_b32_e32 v9, 16, v1
-; GFX10-DL-NOXNACK-NEXT:    v_lshrrev_b32_e32 v16, 16, v0
+; GFX10-DL-NOXNACK-NEXT:    v_ashrrev_i16 v12, 12, v12
+; GFX10-DL-NOXNACK-NEXT:    v_ashrrev_i16 v5, 12, v5
 ; GFX10-DL-NOXNACK-NEXT:    v_lshlrev_b16 v6, 12, v6
+; GFX10-DL-NOXNACK-NEXT:    v_lshlrev_b16 v7, 12, v7
+; GFX10-DL-NOXNACK-NEXT:    v_lshlrev_b16 v8, 12, v8
 ; GFX10-DL-NOXNACK-NEXT:    v_lshlrev_b16 v13, 12, v13
-; GFX10-DL-NOXNACK-NEXT:    v_ashrrev_i16 v7, 12, v7
-; GFX10-DL-NOXNACK-NEXT:    v_ashrrev_i16 v14, 12, v14
-; GFX10-DL-NOXNACK-NEXT:    v_lshl_or_b32 v11, v11, 16, v12
-; GFX10-DL-NOXNACK-NEXT:    v_lshl_or_b32 v4, v4, 16, v5
-; GFX10-DL-NOXNACK-NEXT:    v_lshrrev_b32_e32 v8, 20, v1
-; GFX10-DL-NOXNACK-NEXT:    v_lshrrev_b32_e32 v15, 20, v0
-; GFX10-DL-NOXNACK-NEXT:    v_lshlrev_b16 v9, 12, v9
-; GFX10-DL-NOXNACK-NEXT:    v_lshlrev_b16 v16, 12, v16
+; GFX10-DL-NOXNACK-NEXT:    v_lshlrev_b16 v14, 12, v14
+; GFX10-DL-NOXNACK-NEXT:    v_perm_b32 v11, v12, v11, 0x5040100
+; GFX10-DL-NOXNACK-NEXT:    v_perm_b32 v4, v5, v4, 0x5040100
+; GFX10-DL-NOXNACK-NEXT:    v_lshrrev_b32_e32 v9, 20, v1
+; GFX10-DL-NOXNACK-NEXT:    v_lshrrev_b32_e32 v15, 16, v0
+; GFX10-DL-NOXNACK-NEXT:    v_lshrrev_b32_e32 v16, 20, v0
 ; GFX10-DL-NOXNACK-NEXT:    v_ashrrev_i16 v6, 12, v6
-; GFX10-DL-NOXNACK-NEXT:    v_ashrrev_i16 v5, 12, v13
-; GFX10-DL-NOXNACK-NEXT:    v_and_b32_e32 v12, 0xffff, v14
-; GFX10-DL-NOXNACK-NEXT:    v_and_b32_e32 v7, 0xffff, v7
+; GFX10-DL-NOXNACK-NEXT:    v_ashrrev_i16 v7, 12, v7
+; GFX10-DL-NOXNACK-NEXT:    v_ashrrev_i16 v5, 12, v8
+; GFX10-DL-NOXNACK-NEXT:    v_ashrrev_i16 v8, 12, v13
+; GFX10-DL-NOXNACK-NEXT:    v_ashrrev_i16 v12, 12, v14
 ; GFX10-DL-NOXNACK-NEXT:    v_pk_mul_lo_u16 v4, v4, v11
-; GFX10-DL-NOXNACK-NEXT:    v_lshrrev_b32_e32 v10, 28, v1
-; GFX10-DL-NOXNACK-NEXT:    v_lshrrev_b32_e32 v1, 24, v1
-; GFX10-DL-NOXNACK-NEXT:    v_lshrrev_b32_e32 v17, 28, v0
-; GFX10-DL-NOXNACK-NEXT:    v_lshrrev_b32_e32 v0, 24, v0
-; GFX10-DL-NOXNACK-NEXT:    v_lshlrev_b16 v8, 12, v8
+; GFX10-DL-NOXNACK-NEXT:    v_lshlrev_b16 v9, 12, v9
 ; GFX10-DL-NOXNACK-NEXT:    v_lshlrev_b16 v15, 12, v15
-; GFX10-DL-NOXNACK-NEXT:    v_ashrrev_i16 v9, 12, v9
-; GFX10-DL-NOXNACK-NEXT:    v_ashrrev_i16 v13, 12, v16
-; GFX10-DL-NOXNACK-NEXT:    v_lshl_or_b32 v5, v5, 16, v12
-; GFX10-DL-NOXNACK-NEXT:    v_lshl_or_b32 v6, v6, 16, v7
+; GFX10-DL-NOXNACK-NEXT:    v_lshlrev_b16 v16, 12, v16
+; GFX10-DL-NOXNACK-NEXT:    v_perm_b32 v8, v12, v8, 0x5040100
+; GFX10-DL-NOXNACK-NEXT:    v_perm_b32 v6, v7, v6, 0x5040100
 ; GFX10-DL-NOXNACK-NEXT:    v_lshrrev_b32_e32 v7, 16, v4
 ; GFX10-DL-NOXNACK-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-DL-NOXNACK-NEXT:    v_add_nc_u16 v3, v4, v3
-; GFX10-DL-NOXNACK-NEXT:    v_lshlrev_b16 v1, 12, v1
-; GFX10-DL-NOXNACK-NEXT:    v_ashrrev_i16 v8, 12, v8
+; GFX10-DL-NOXNACK-NEXT:    v_lshrrev_b32_e32 v10, 24, v1
+; GFX10-DL-NOXNACK-NEXT:    v_lshrrev_b32_e32 v1, 28, v1
+; GFX10-DL-NOXNACK-NEXT:    v_lshrrev_b32_e32 v17, 24, v0
+; GFX10-DL-NOXNACK-NEXT:    v_ashrrev_i16 v9, 12, v9
 ; GFX10-DL-NOXNACK-NEXT:    v_ashrrev_i16 v11, 12, v15
-; GFX10-DL-NOXNACK-NEXT:    v_lshlrev_b16 v0, 12, v0
-; GFX10-DL-NOXNACK-NEXT:    v_and_b32_e32 v4, 0xffff, v13
-; GFX10-DL-NOXNACK-NEXT:    v_and_b32_e32 v9, 0xffff, v9
-; GFX10-DL-NOXNACK-NEXT:    v_pk_mul_lo_u16 v5, v6, v5
+; GFX10-DL-NOXNACK-NEXT:    v_lshrrev_b32_e32 v0, 28, v0
+; GFX10-DL-NOXNACK-NEXT:    v_ashrrev_i16 v4, 12, v16
+; GFX10-DL-NOXNACK-NEXT:    v_pk_mul_lo_u16 v6, v6, v8
 ; GFX10-DL-NOXNACK-NEXT:    v_add_nc_u16 v3, v3, v7
 ; GFX10-DL-NOXNACK-NEXT:    v_lshlrev_b16 v10, 12, v10
-; GFX10-DL-NOXNACK-NEXT:    v_lshlrev_b16 v17, 12, v17
-; GFX10-DL-NOXNACK-NEXT:    v_ashrrev_i16 v1, 12, v1
-; GFX10-DL-NOXNACK-NEXT:    v_ashrrev_i16 v0, 12, v0
-; GFX10-DL-NOXNACK-NEXT:    v_lshl_or_b32 v4, v11, 16, v4
-; GFX10-DL-NOXNACK-NEXT:    v_lshl_or_b32 v6, v8, 16, v9
-; GFX10-DL-NOXNACK-NEXT:    v_lshrrev_b32_e32 v7, 16, v5
-; GFX10-DL-NOXNACK-NEXT:    v_add_nc_u16 v3, v3, v5
+; GFX10-DL-NOXNACK-NEXT:    v_lshlrev_b16 v1, 12, v1
+; GFX10-DL-NOXNACK-NEXT:    v_lshlrev_b16 v12, 12, v17
+; GFX10-DL-NOXNACK-NEXT:    v_lshlrev_b16 v0, 12, v0
+; GFX10-DL-NOXNACK-NEXT:    v_perm_b32 v4, v4, v11, 0x5040100
+; GFX10-DL-NOXNACK-NEXT:    v_perm_b32 v5, v9, v5, 0x5040100
+; GFX10-DL-NOXNACK-NEXT:    v_lshrrev_b32_e32 v7, 16, v6
+; GFX10-DL-NOXNACK-NEXT:    v_add_nc_u16 v3, v3, v6
 ; GFX10-DL-NOXNACK-NEXT:    v_ashrrev_i16 v10, 12, v10
-; GFX10-DL-NOXNACK-NEXT:    v_ashrrev_i16 v5, 12, v17
-; GFX10-DL-NOXNACK-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX10-DL-NOXNACK-NEXT:    v_and_b32_e32 v1, 0xffff, v1
-; GFX10-DL-NOXNACK-NEXT:    v_pk_mul_lo_u16 v4, v6, v4
+; GFX10-DL-NOXNACK-NEXT:    v_ashrrev_i16 v6, 12, v12
+; GFX10-DL-NOXNACK-NEXT:    v_ashrrev_i16 v0, 12, v0
+; GFX10-DL-NOXNACK-NEXT:    v_ashrrev_i16 v1, 12, v1
+; GFX10-DL-NOXNACK-NEXT:    v_pk_mul_lo_u16 v4, v5, v4
 ; GFX10-DL-NOXNACK-NEXT:    v_add_nc_u16 v3, v3, v7
-; GFX10-DL-NOXNACK-NEXT:    v_lshl_or_b32 v0, v5, 16, v0
-; GFX10-DL-NOXNACK-NEXT:    v_lshl_or_b32 v1, v10, 16, v1
+; GFX10-DL-NOXNACK-NEXT:    v_perm_b32 v0, v0, v6, 0x5040100
+; GFX10-DL-NOXNACK-NEXT:    v_perm_b32 v1, v1, v10, 0x5040100
 ; GFX10-DL-NOXNACK-NEXT:    v_lshrrev_b32_e32 v5, 16, v4
 ; GFX10-DL-NOXNACK-NEXT:    v_add_nc_u16 v3, v3, v4
 ; GFX10-DL-NOXNACK-NEXT:    v_pk_mul_lo_u16 v0, v1, v0
diff --git a/llvm/test/CodeGen/AMDGPU/idot8u.ll b/llvm/test/CodeGen/AMDGPU/idot8u.ll
--- a/llvm/test/CodeGen/AMDGPU/idot8u.ll
+++ b/llvm/test/CodeGen/AMDGPU/idot8u.ll
@@ -2213,51 +2213,44 @@
 ; GFX9-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
 ; GFX9-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x34
 ; GFX9-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
-; GFX9-NEXT:    s_addc_u32 s9, s9, 0
+; GFX9-NEXT:    s_mov_b32 s0, 0x5040100
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    global_load_dword v1, v0, s[4:5]
 ; GFX9-NEXT:    global_load_dword v2, v0, s[6:7]
 ; GFX9-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX9-NEXT:    global_load_ushort v3, v0, s[2:3]
+; GFX9-NEXT:    s_addc_u32 s9, s9, 0
 ; GFX9-NEXT:    s_waitcnt vmcnt(2)
-; GFX9-NEXT:    v_and_b32_e32 v5, 15, v1
-; GFX9-NEXT:    v_bfe_u32 v7, v1, 8, 4
+; GFX9-NEXT:    v_and_b32_e32 v4, 15, v1
+; GFX9-NEXT:    v_bfe_u32 v5, v1, 4, 4
+; GFX9-NEXT:    v_bfe_u32 v6, v1, 8, 4
+; GFX9-NEXT:    v_bfe_u32 v7, v1, 12, 4
 ; GFX9-NEXT:    s_waitcnt vmcnt(1)
-; GFX9-NEXT:    v_and_b32_e32 v12, 15, v2
-; GFX9-NEXT:    v_bfe_u32 v4, v1, 4, 4
-; GFX9-NEXT:    v_bfe_u32 v6, v1, 12, 4
-; GFX9-NEXT:    v_bfe_u32 v11, v2, 4, 4
-; GFX9-NEXT:    v_and_b32_e32 v7, 0xffff, v7
-; GFX9-NEXT:    v_and_b32_e32 v12, 0xffff, v12
-; GFX9-NEXT:    v_and_b32_e32 v5, 0xffff, v5
-; GFX9-NEXT:    v_bfe_u32 v9, v1, 16, 4
-; GFX9-NEXT:    v_bfe_u32 v14, v2, 8, 4
-; GFX9-NEXT:    v_lshl_or_b32 v6, v6, 16, v7
-; GFX9-NEXT:    v_lshl_or_b32 v7, v11, 16, v12
-; GFX9-NEXT:    v_lshl_or_b32 v4, v4, 16, v5
-; GFX9-NEXT:    v_bfe_u32 v8, v1, 20, 4
-; GFX9-NEXT:    v_bfe_u32 v13, v2, 12, 4
-; GFX9-NEXT:    v_and_b32_e32 v9, 0xffff, v9
-; GFX9-NEXT:    v_and_b32_e32 v14, 0xffff, v14
+; GFX9-NEXT:    v_and_b32_e32 v11, 15, v2
+; GFX9-NEXT:    v_bfe_u32 v12, v2, 4, 4
+; GFX9-NEXT:    v_perm_b32 v6, v7, v6, s0
+; GFX9-NEXT:    v_perm_b32 v7, v12, v11, s0
+; GFX9-NEXT:    v_perm_b32 v4, v5, v4, s0
+; GFX9-NEXT:    v_bfe_u32 v8, v1, 16, 4
+; GFX9-NEXT:    v_bfe_u32 v9, v1, 20, 4
+; GFX9-NEXT:    v_bfe_u32 v13, v2, 8, 4
+; GFX9-NEXT:    v_bfe_u32 v14, v2, 12, 4
 ; GFX9-NEXT:    v_pk_mul_lo_u16 v4, v4, v7
-; GFX9-NEXT:    v_lshrrev_b32_e32 v10, 28, v1
-; GFX9-NEXT:    v_bfe_u32 v1, v1, 24, 4
-; GFX9-NEXT:    v_bfe_u32 v15, v2, 20, 4
-; GFX9-NEXT:    v_bfe_u32 v16, v2, 16, 4
-; GFX9-NEXT:    v_lshrrev_b32_e32 v17, 28, v2
-; GFX9-NEXT:    v_bfe_u32 v2, v2, 24, 4
-; GFX9-NEXT:    v_lshl_or_b32 v8, v8, 16, v9
-; GFX9-NEXT:    v_lshl_or_b32 v9, v13, 16, v14
+; GFX9-NEXT:    v_perm_b32 v8, v9, v8, s0
+; GFX9-NEXT:    v_perm_b32 v9, v14, v13, s0
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    v_add_u16_e32 v3, v4, v3
-; GFX9-NEXT:    v_and_b32_e32 v2, 0xffff, v2
-; GFX9-NEXT:    v_and_b32_e32 v1, 0xffff, v1
-; GFX9-NEXT:    v_and_b32_e32 v16, 0xffff, v16
+; GFX9-NEXT:    v_bfe_u32 v10, v1, 24, 4
+; GFX9-NEXT:    v_lshrrev_b32_e32 v1, 28, v1
+; GFX9-NEXT:    v_bfe_u32 v15, v2, 16, 4
+; GFX9-NEXT:    v_bfe_u32 v16, v2, 20, 4
+; GFX9-NEXT:    v_bfe_u32 v17, v2, 24, 4
+; GFX9-NEXT:    v_lshrrev_b32_e32 v2, 28, v2
 ; GFX9-NEXT:    v_pk_mul_lo_u16 v5, v6, v9
 ; GFX9-NEXT:    v_add_u16_sdwa v3, v3, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
-; GFX9-NEXT:    v_lshl_or_b32 v2, v17, 16, v2
-; GFX9-NEXT:    v_lshl_or_b32 v1, v10, 16, v1
-; GFX9-NEXT:    v_lshl_or_b32 v10, v15, 16, v16
+; GFX9-NEXT:    v_perm_b32 v2, v2, v17, s0
+; GFX9-NEXT:    v_perm_b32 v1, v1, v10, s0
+; GFX9-NEXT:    v_perm_b32 v10, v16, v15, s0
 ; GFX9-NEXT:    v_add_u16_e32 v3, v3, v5
 ; GFX9-NEXT:    v_pk_mul_lo_u16 v1, v1, v2
 ; GFX9-NEXT:    v_pk_mul_lo_u16 v2, v8, v10
@@ -2279,51 +2272,44 @@
 ; GFX9-DL-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
 ; GFX9-DL-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x34
 ; GFX9-DL-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
-; GFX9-DL-NEXT:    s_addc_u32 s9, s9, 0
+; GFX9-DL-NEXT:    s_mov_b32 s0, 0x5040100
 ; GFX9-DL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-DL-NEXT:    global_load_dword v1, v0, s[4:5]
 ; GFX9-DL-NEXT:    global_load_dword v2, v0, s[6:7]
 ; GFX9-DL-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX9-DL-NEXT:    global_load_ushort v3, v0, s[2:3]
+; GFX9-DL-NEXT:    s_addc_u32 s9, s9, 0
 ; GFX9-DL-NEXT:    s_waitcnt vmcnt(2)
-; GFX9-DL-NEXT:    v_and_b32_e32 v5, 15, v1
-; GFX9-DL-NEXT:    v_bfe_u32 v7, v1, 8, 4
+; GFX9-DL-NEXT:    v_and_b32_e32 v4, 15, v1
+; GFX9-DL-NEXT:    v_bfe_u32 v5, v1, 4, 4
+; GFX9-DL-NEXT:    v_bfe_u32 v6, v1, 8, 4
+; GFX9-DL-NEXT:    v_bfe_u32 v7, v1, 12, 4
 ; GFX9-DL-NEXT:    s_waitcnt vmcnt(1)
-; GFX9-DL-NEXT:    v_and_b32_e32 v12, 15, v2
-; GFX9-DL-NEXT:    v_bfe_u32 v4, v1, 4, 4
-; GFX9-DL-NEXT:    v_bfe_u32 v6, v1, 12, 4
-; GFX9-DL-NEXT:    v_bfe_u32 v11, v2, 4, 4
-; GFX9-DL-NEXT:    v_and_b32_e32 v7, 0xffff, v7
-; GFX9-DL-NEXT:    v_and_b32_e32 v12, 0xffff, v12
-; GFX9-DL-NEXT:    v_and_b32_e32 v5, 0xffff, v5
-; GFX9-DL-NEXT:    v_bfe_u32 v9, v1, 16, 4
-; GFX9-DL-NEXT:    v_bfe_u32 v14, v2, 8, 4
-; GFX9-DL-NEXT:    v_lshl_or_b32 v6, v6, 16, v7
-; GFX9-DL-NEXT:    v_lshl_or_b32 v7, v11, 16, v12
-; GFX9-DL-NEXT:    v_lshl_or_b32 v4, v4, 16, v5
-; GFX9-DL-NEXT:    v_bfe_u32 v8, v1, 20, 4
-; GFX9-DL-NEXT:    v_bfe_u32 v13, v2, 12, 4
-; GFX9-DL-NEXT:    v_and_b32_e32 v9, 0xffff, v9
-; GFX9-DL-NEXT:    v_and_b32_e32 v14, 0xffff, v14
+; GFX9-DL-NEXT:    v_and_b32_e32 v11, 15, v2
+; GFX9-DL-NEXT:    v_bfe_u32 v12, v2, 4, 4
+; GFX9-DL-NEXT:    v_perm_b32 v6, v7, v6, s0
+; GFX9-DL-NEXT:    v_perm_b32 v7, v12, v11, s0
+; GFX9-DL-NEXT:    v_perm_b32 v4, v5, v4, s0
+; GFX9-DL-NEXT:    v_bfe_u32 v8, v1, 16, 4
+; GFX9-DL-NEXT:    v_bfe_u32 v9, v1, 20, 4
+; GFX9-DL-NEXT:    v_bfe_u32 v13, v2, 8, 4
+; GFX9-DL-NEXT:    v_bfe_u32 v14, v2, 12, 4
 ; GFX9-DL-NEXT:    v_pk_mul_lo_u16 v4, v4, v7
-; GFX9-DL-NEXT:    v_lshrrev_b32_e32 v10, 28, v1
-; GFX9-DL-NEXT:    v_bfe_u32 v1, v1, 24, 4
-; GFX9-DL-NEXT:    v_bfe_u32 v15, v2, 20, 4
-; GFX9-DL-NEXT:    v_bfe_u32 v16, v2, 16, 4
-; GFX9-DL-NEXT:    v_lshrrev_b32_e32 v17, 28, v2
-; GFX9-DL-NEXT:    v_bfe_u32 v2, v2, 24, 4
-; GFX9-DL-NEXT:    v_lshl_or_b32 v8, v8, 16, v9
-; GFX9-DL-NEXT:    v_lshl_or_b32 v9, v13, 16, v14
+; GFX9-DL-NEXT:    v_perm_b32 v8, v9, v8, s0
+; GFX9-DL-NEXT:    v_perm_b32 v9, v14, v13, s0
 ; GFX9-DL-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-DL-NEXT:    v_add_u16_e32 v3, v4, v3
-; GFX9-DL-NEXT:    v_and_b32_e32 v2, 0xffff, v2
-; GFX9-DL-NEXT:    v_and_b32_e32 v1, 0xffff, v1
-; GFX9-DL-NEXT:    v_and_b32_e32 v16, 0xffff, v16
+; GFX9-DL-NEXT:    v_bfe_u32 v10, v1, 24, 4
+; GFX9-DL-NEXT:    v_lshrrev_b32_e32 v1, 28, v1
+; GFX9-DL-NEXT:    v_bfe_u32 v15, v2, 16, 4
+; GFX9-DL-NEXT:    v_bfe_u32 v16, v2, 20, 4
+; GFX9-DL-NEXT:    v_bfe_u32 v17, v2, 24, 4
+; GFX9-DL-NEXT:    v_lshrrev_b32_e32 v2, 28, v2
 ; GFX9-DL-NEXT:    v_pk_mul_lo_u16 v5, v6, v9
 ; GFX9-DL-NEXT:    v_add_u16_sdwa v3, v3, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
-; GFX9-DL-NEXT:    v_lshl_or_b32 v2, v17, 16, v2
-; GFX9-DL-NEXT:    v_lshl_or_b32 v1, v10, 16, v1
-; GFX9-DL-NEXT:    v_lshl_or_b32 v10, v15, 16, v16
+; GFX9-DL-NEXT:    v_perm_b32 v2, v2, v17, s0
+; GFX9-DL-NEXT:    v_perm_b32 v1, v1, v10, s0
+; GFX9-DL-NEXT:    v_perm_b32 v10, v16, v15, s0
 ; GFX9-DL-NEXT:    v_add_u16_e32 v3, v3, v5
 ; GFX9-DL-NEXT:    v_pk_mul_lo_u16 v1, v1, v2
 ; GFX9-DL-NEXT:    v_pk_mul_lo_u16 v2, v8, v10
@@ -2353,49 +2339,41 @@
 ; GFX10-DL-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX10-DL-NEXT:    global_load_ushort v3, v0, s[0:1]
 ; GFX10-DL-NEXT:    s_waitcnt vmcnt(2)
-; GFX10-DL-NEXT:    v_and_b32_e32 v6, 15, v1
+; GFX10-DL-NEXT:    v_and_b32_e32 v4, 15, v1
 ; GFX10-DL-NEXT:    s_waitcnt vmcnt(1)
 ; GFX10-DL-NEXT:    v_and_b32_e32 v5, 15, v2
-; GFX10-DL-NEXT:    v_bfe_u32 v4, v1, 4, 4
-; GFX10-DL-NEXT:    v_bfe_u32 v9, v2, 4, 4
-; GFX10-DL-NEXT:    v_bfe_u32 v8, v1, 8, 4
-; GFX10-DL-NEXT:    v_and_b32_e32 v6, 0xffff, v6
-; GFX10-DL-NEXT:    v_and_b32_e32 v5, 0xffff, v5
-; GFX10-DL-NEXT:    v_bfe_u32 v12, v2, 8, 4
-; GFX10-DL-NEXT:    v_bfe_u32 v7, v1, 12, 4
-; GFX10-DL-NEXT:    v_and_b32_e32 v8, 0xffff, v8
-; GFX10-DL-NEXT:    v_lshl_or_b32 v4, v4, 16, v6
-; GFX10-DL-NEXT:    v_lshl_or_b32 v5, v9, 16, v5
-; GFX10-DL-NEXT:    v_bfe_u32 v9, v2, 12, 4
-; GFX10-DL-NEXT:    v_and_b32_e32 v12, 0xffff, v12
-; GFX10-DL-NEXT:    v_bfe_u32 v11, v1, 16, 4
-; GFX10-DL-NEXT:    v_lshl_or_b32 v7, v7, 16, v8
-; GFX10-DL-NEXT:    v_pk_mul_lo_u16 v4, v4, v5
-; GFX10-DL-NEXT:    v_bfe_u32 v5, v2, 16, 4
-; GFX10-DL-NEXT:    v_lshl_or_b32 v9, v9, 16, v12
+; GFX10-DL-NEXT:    v_bfe_u32 v6, v2, 4, 4
+; GFX10-DL-NEXT:    v_bfe_u32 v7, v1, 4, 4
+; GFX10-DL-NEXT:    v_bfe_u32 v8, v2, 12, 4
+; GFX10-DL-NEXT:    v_bfe_u32 v9, v1, 12, 4
 ; GFX10-DL-NEXT:    v_bfe_u32 v10, v1, 20, 4
-; GFX10-DL-NEXT:    v_and_b32_e32 v11, 0xffff, v11
+; GFX10-DL-NEXT:    v_perm_b32 v5, v6, v5, 0x5040100
+; GFX10-DL-NEXT:    v_perm_b32 v4, v7, v4, 0x5040100
+; GFX10-DL-NEXT:    v_bfe_u32 v6, v1, 8, 4
+; GFX10-DL-NEXT:    v_bfe_u32 v7, v2, 8, 4
+; GFX10-DL-NEXT:    v_pk_mul_lo_u16 v4, v4, v5
+; GFX10-DL-NEXT:    v_perm_b32 v6, v9, v6, 0x5040100
+; GFX10-DL-NEXT:    v_perm_b32 v7, v8, v7, 0x5040100
+; GFX10-DL-NEXT:    v_bfe_u32 v5, v1, 16, 4
+; GFX10-DL-NEXT:    v_bfe_u32 v9, v2, 20, 4
 ; GFX10-DL-NEXT:    v_lshrrev_b32_e32 v8, 16, v4
 ; GFX10-DL-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-DL-NEXT:    v_add_nc_u16 v3, v4, v3
-; GFX10-DL-NEXT:    v_bfe_u32 v4, v2, 20, 4
-; GFX10-DL-NEXT:    v_and_b32_e32 v5, 0xffff, v5
-; GFX10-DL-NEXT:    v_pk_mul_lo_u16 v7, v7, v9
-; GFX10-DL-NEXT:    v_lshrrev_b32_e32 v6, 28, v1
+; GFX10-DL-NEXT:    v_bfe_u32 v4, v2, 16, 4
+; GFX10-DL-NEXT:    v_pk_mul_lo_u16 v6, v6, v7
+; GFX10-DL-NEXT:    v_perm_b32 v5, v10, v5, 0x5040100
+; GFX10-DL-NEXT:    v_bfe_u32 v7, v1, 24, 4
 ; GFX10-DL-NEXT:    v_add_nc_u16 v3, v3, v8
-; GFX10-DL-NEXT:    v_bfe_u32 v1, v1, 24, 4
-; GFX10-DL-NEXT:    v_bfe_u32 v8, v2, 24, 4
-; GFX10-DL-NEXT:    v_lshl_or_b32 v4, v4, 16, v5
-; GFX10-DL-NEXT:    v_lshl_or_b32 v5, v10, 16, v11
-; GFX10-DL-NEXT:    v_lshrrev_b32_e32 v9, 16, v7
-; GFX10-DL-NEXT:    v_add_nc_u16 v3, v3, v7
+; GFX10-DL-NEXT:    v_perm_b32 v4, v9, v4, 0x5040100
+; GFX10-DL-NEXT:    v_lshrrev_b32_e32 v8, 16, v6
+; GFX10-DL-NEXT:    v_lshrrev_b32_e32 v1, 28, v1
+; GFX10-DL-NEXT:    v_add_nc_u16 v3, v3, v6
+; GFX10-DL-NEXT:    v_bfe_u32 v6, v2, 24, 4
 ; GFX10-DL-NEXT:    v_lshrrev_b32_e32 v2, 28, v2
-; GFX10-DL-NEXT:    v_and_b32_e32 v7, 0xffff, v8
-; GFX10-DL-NEXT:    v_and_b32_e32 v1, 0xffff, v1
 ; GFX10-DL-NEXT:    v_pk_mul_lo_u16 v4, v5, v4
-; GFX10-DL-NEXT:    v_add_nc_u16 v3, v3, v9
-; GFX10-DL-NEXT:    v_lshl_or_b32 v2, v2, 16, v7
-; GFX10-DL-NEXT:    v_lshl_or_b32 v1, v6, 16, v1
+; GFX10-DL-NEXT:    v_perm_b32 v1, v1, v7, 0x5040100
+; GFX10-DL-NEXT:    v_add_nc_u16 v3, v3, v8
+; GFX10-DL-NEXT:    v_perm_b32 v2, v2, v6, 0x5040100
 ; GFX10-DL-NEXT:    v_lshrrev_b32_e32 v5, 16, v4
 ; GFX10-DL-NEXT:    v_add_nc_u16 v3, v3, v4
 ; GFX10-DL-NEXT:    v_pk_mul_lo_u16 v1, v1, v2
@@ -2930,51 +2908,44 @@
 ; GFX9-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
 ; GFX9-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x34
 ; GFX9-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
-; GFX9-NEXT:    s_addc_u32 s9, s9, 0
+; GFX9-NEXT:    s_mov_b32 s0, 0x5040100
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    global_load_dword v1, v0, s[4:5]
 ; GFX9-NEXT:    global_load_dword v2, v0, s[6:7]
 ; GFX9-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX9-NEXT:    global_load_ubyte v3, v0, s[2:3]
+; GFX9-NEXT:    s_addc_u32 s9, s9, 0
 ; GFX9-NEXT:    s_waitcnt vmcnt(2)
-; GFX9-NEXT:    v_and_b32_e32 v5, 15, v1
-; GFX9-NEXT:    v_bfe_u32 v7, v1, 8, 4
+; GFX9-NEXT:    v_and_b32_e32 v4, 15, v1
+; GFX9-NEXT:    v_bfe_u32 v5, v1, 4, 4
+; GFX9-NEXT:    v_bfe_u32 v6, v1, 8, 4
+; GFX9-NEXT:    v_bfe_u32 v7, v1, 12, 4
 ; GFX9-NEXT:    s_waitcnt vmcnt(1)
-; GFX9-NEXT:    v_and_b32_e32 v12, 15, v2
-; GFX9-NEXT:    v_bfe_u32 v4, v1, 4, 4
-; GFX9-NEXT:    v_bfe_u32 v6, v1, 12, 4
-; GFX9-NEXT:    v_bfe_u32 v11, v2, 4, 4
-; GFX9-NEXT:    v_and_b32_e32 v7, 0xffff, v7
-; GFX9-NEXT:    v_and_b32_e32 v12, 0xffff, v12
-; GFX9-NEXT:    v_and_b32_e32 v5, 0xffff, v5
-; GFX9-NEXT:    v_bfe_u32 v9, v1, 16, 4
-; GFX9-NEXT:    v_bfe_u32 v14, v2, 8, 4
-; GFX9-NEXT:    v_lshl_or_b32 v6, v6, 16, v7
-; GFX9-NEXT:    v_lshl_or_b32 v7, v11, 16, v12
-; GFX9-NEXT:    v_lshl_or_b32 v4, v4, 16, v5
-; GFX9-NEXT:    v_bfe_u32 v8, v1, 20, 4
-; GFX9-NEXT:    v_bfe_u32 v13, v2, 12, 4
-; GFX9-NEXT:    v_and_b32_e32 v9, 0xffff, v9
-; GFX9-NEXT:    v_and_b32_e32 v14, 0xffff, v14
+; GFX9-NEXT:    v_and_b32_e32 v11, 15, v2
+; GFX9-NEXT:    v_bfe_u32 v12, v2, 4, 4
+; GFX9-NEXT:    v_perm_b32 v6, v7, v6, s0
+; GFX9-NEXT:    v_perm_b32 v7, v12, v11, s0
+; GFX9-NEXT:    v_perm_b32 v4, v5, v4, s0
+; GFX9-NEXT:    v_bfe_u32 v8, v1, 16, 4
+; GFX9-NEXT:    v_bfe_u32 v9, v1, 20, 4
+; GFX9-NEXT:    v_bfe_u32 v13, v2, 8, 4
+; GFX9-NEXT:    v_bfe_u32 v14, v2, 12, 4
 ; GFX9-NEXT:    v_pk_mul_lo_u16 v4, v4, v7
-; GFX9-NEXT:    v_lshrrev_b32_e32 v10, 28, v1
-; GFX9-NEXT:    v_bfe_u32 v1, v1, 24, 4
-; GFX9-NEXT:    v_bfe_u32 v15, v2, 20, 4
-; GFX9-NEXT:    v_bfe_u32 v16, v2, 16, 4
-; GFX9-NEXT:    v_lshrrev_b32_e32 v17, 28, v2
-; GFX9-NEXT:    v_bfe_u32 v2, v2, 24, 4
-; GFX9-NEXT:    v_lshl_or_b32 v8, v8, 16, v9
-; GFX9-NEXT:    v_lshl_or_b32 v9, v13, 16, v14
+; GFX9-NEXT:    v_perm_b32 v8, v9, v8, s0
+; GFX9-NEXT:    v_perm_b32 v9, v14, v13, s0
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    v_add_u16_e32 v3, v4, v3
-; GFX9-NEXT:    v_and_b32_e32 v2, 0xffff, v2
-; GFX9-NEXT:    v_and_b32_e32 v1, 0xffff, v1
-; GFX9-NEXT:    v_and_b32_e32 v16, 0xffff, v16
+; GFX9-NEXT:    v_bfe_u32 v10, v1, 24, 4
+; GFX9-NEXT:    v_lshrrev_b32_e32 v1, 28, v1
+; GFX9-NEXT:    v_bfe_u32 v15, v2, 16, 4
+; GFX9-NEXT:    v_bfe_u32 v16, v2, 20, 4
+; GFX9-NEXT:    v_bfe_u32 v17, v2, 24, 4
+; GFX9-NEXT:    v_lshrrev_b32_e32 v2, 28, v2
 ; GFX9-NEXT:    v_pk_mul_lo_u16 v5, v6, v9
 ; GFX9-NEXT:    v_add_u16_sdwa v3, v3, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
-; GFX9-NEXT:    v_lshl_or_b32 v2, v17, 16, v2
-; GFX9-NEXT:    v_lshl_or_b32 v1, v10, 16, v1
-; GFX9-NEXT:    v_lshl_or_b32 v10, v15, 16, v16
+; GFX9-NEXT:    v_perm_b32 v2, v2, v17, s0
+; GFX9-NEXT:    v_perm_b32 v1, v1, v10, s0
+; GFX9-NEXT:    v_perm_b32 v10, v16, v15, s0
 ; GFX9-NEXT:    v_add_u16_e32 v3, v3, v5
 ; GFX9-NEXT:    v_pk_mul_lo_u16 v1, v1, v2
 ; GFX9-NEXT:    v_pk_mul_lo_u16 v2, v8, v10
@@ -2997,51 +2968,44 @@
 ; GFX9-DL-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
 ; GFX9-DL-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x34
 ; GFX9-DL-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
-; GFX9-DL-NEXT:    s_addc_u32 s9, s9, 0
+; GFX9-DL-NEXT:    s_mov_b32 s0, 0x5040100
 ; GFX9-DL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-DL-NEXT:    global_load_dword v1, v0, s[4:5]
 ; GFX9-DL-NEXT:    global_load_dword v2, v0, s[6:7]
 ; GFX9-DL-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX9-DL-NEXT:    global_load_ubyte v3, v0, s[2:3]
+; GFX9-DL-NEXT:    s_addc_u32 s9, s9, 0
 ; GFX9-DL-NEXT:    s_waitcnt vmcnt(2)
-; GFX9-DL-NEXT:    v_and_b32_e32 v5, 15, v1
-; GFX9-DL-NEXT:    v_bfe_u32 v7, v1, 8, 4
+; GFX9-DL-NEXT:    v_and_b32_e32 v4, 15, v1
+; GFX9-DL-NEXT:    v_bfe_u32 v5, v1, 4, 4
+; GFX9-DL-NEXT:    v_bfe_u32 v6, v1, 8, 4
+; GFX9-DL-NEXT:    v_bfe_u32 v7, v1, 12, 4
 ; GFX9-DL-NEXT:    s_waitcnt vmcnt(1)
-; GFX9-DL-NEXT:    v_and_b32_e32 v12, 15, v2
-; GFX9-DL-NEXT:    v_bfe_u32 v4, v1, 4, 4
-; GFX9-DL-NEXT:    v_bfe_u32 v6, v1, 12, 4
-; GFX9-DL-NEXT:    v_bfe_u32 v11, v2, 4, 4
-; GFX9-DL-NEXT:    v_and_b32_e32 v7, 0xffff, v7
-; GFX9-DL-NEXT:    v_and_b32_e32 v12, 0xffff, v12
-; GFX9-DL-NEXT:    v_and_b32_e32 v5, 0xffff, v5
-; GFX9-DL-NEXT:    v_bfe_u32 v9, v1, 16, 4
-; GFX9-DL-NEXT:    v_bfe_u32 v14, v2, 8, 4
-; GFX9-DL-NEXT:    v_lshl_or_b32 v6, v6, 16, v7
-; GFX9-DL-NEXT:    v_lshl_or_b32 v7, v11, 16, v12
-; GFX9-DL-NEXT:    v_lshl_or_b32 v4, v4, 16, v5
-; GFX9-DL-NEXT:    v_bfe_u32 v8, v1, 20, 4
-; GFX9-DL-NEXT:    v_bfe_u32 v13, v2, 12, 4
-; GFX9-DL-NEXT:    v_and_b32_e32 v9, 0xffff, v9
-; GFX9-DL-NEXT:    v_and_b32_e32 v14, 0xffff, v14
+; GFX9-DL-NEXT:    v_and_b32_e32 v11, 15, v2
+; GFX9-DL-NEXT:    v_bfe_u32 v12, v2, 4, 4
+; GFX9-DL-NEXT:    v_perm_b32 v6, v7, v6, s0
+; GFX9-DL-NEXT:    v_perm_b32 v7, v12, v11, s0
+; GFX9-DL-NEXT:    v_perm_b32 v4, v5, v4, s0
+; GFX9-DL-NEXT:    v_bfe_u32 v8, v1, 16, 4
+; GFX9-DL-NEXT:    v_bfe_u32 v9, v1, 20, 4
+; GFX9-DL-NEXT:    v_bfe_u32 v13, v2, 8, 4
+; GFX9-DL-NEXT:    v_bfe_u32 v14, v2, 12, 4
 ; GFX9-DL-NEXT:    v_pk_mul_lo_u16 v4, v4, v7
-; GFX9-DL-NEXT:    v_lshrrev_b32_e32 v10, 28, v1
-; GFX9-DL-NEXT:    v_bfe_u32 v1, v1, 24, 4
-; GFX9-DL-NEXT:    v_bfe_u32 v15, v2, 20, 4
-; GFX9-DL-NEXT:    v_bfe_u32 v16, v2, 16, 4
-; GFX9-DL-NEXT:    v_lshrrev_b32_e32 v17, 28, v2
-; GFX9-DL-NEXT:    v_bfe_u32 v2, v2, 24, 4
-; GFX9-DL-NEXT:    v_lshl_or_b32 v8, v8, 16, v9
-; GFX9-DL-NEXT:    v_lshl_or_b32 v9, v13, 16, v14
+; GFX9-DL-NEXT:    v_perm_b32 v8, v9, v8, s0
+; GFX9-DL-NEXT:    v_perm_b32 v9, v14, v13, s0
 ; GFX9-DL-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-DL-NEXT:    v_add_u16_e32 v3, v4, v3
-; GFX9-DL-NEXT:    v_and_b32_e32 v2, 0xffff, v2
-; GFX9-DL-NEXT:    v_and_b32_e32 v1, 0xffff, v1
-; GFX9-DL-NEXT:    v_and_b32_e32 v16, 0xffff, v16
+; GFX9-DL-NEXT:    v_bfe_u32 v10, v1, 24, 4
+; GFX9-DL-NEXT:    v_lshrrev_b32_e32 v1, 28, v1
+; GFX9-DL-NEXT:    v_bfe_u32 v15, v2, 16, 4
+; GFX9-DL-NEXT:    v_bfe_u32 v16, v2, 20, 4
+; GFX9-DL-NEXT:    v_bfe_u32 v17, v2, 24, 4
+; GFX9-DL-NEXT:    v_lshrrev_b32_e32 v2, 28, v2
 ; GFX9-DL-NEXT:    v_pk_mul_lo_u16 v5, v6, v9
 ; GFX9-DL-NEXT:    v_add_u16_sdwa v3, v3, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
-; GFX9-DL-NEXT:    v_lshl_or_b32 v2, v17, 16, v2
-; GFX9-DL-NEXT:    v_lshl_or_b32 v1, v10, 16, v1
-; GFX9-DL-NEXT:    v_lshl_or_b32 v10, v15, 16, v16
+; GFX9-DL-NEXT:    v_perm_b32 v2, v2, v17, s0
+; GFX9-DL-NEXT:    v_perm_b32 v1, v1, v10, s0
+; GFX9-DL-NEXT:    v_perm_b32 v10, v16, v15, s0
 ; GFX9-DL-NEXT:    v_add_u16_e32 v3, v3, v5
 ; GFX9-DL-NEXT:    v_pk_mul_lo_u16 v1, v1, v2
 ; GFX9-DL-NEXT:    v_pk_mul_lo_u16 v2, v8, v10
@@ -3072,49 +3036,41 @@
 ; GFX10-DL-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX10-DL-NEXT:    global_load_ubyte v3, v0, s[0:1]
 ; GFX10-DL-NEXT:    s_waitcnt vmcnt(2)
-; GFX10-DL-NEXT:    v_and_b32_e32 v6, 15, v1
+; GFX10-DL-NEXT:    v_and_b32_e32 v4, 15, v1
 ; GFX10-DL-NEXT:    s_waitcnt vmcnt(1)
 ; GFX10-DL-NEXT:    v_and_b32_e32 v5, 15, v2
-; GFX10-DL-NEXT:    v_bfe_u32 v4, v1, 4, 4
-; GFX10-DL-NEXT:    v_bfe_u32 v9, v2, 4, 4
-; GFX10-DL-NEXT:    v_bfe_u32 v8, v1, 8, 4
-; GFX10-DL-NEXT:    v_and_b32_e32 v6, 0xffff, v6
-; GFX10-DL-NEXT:    v_and_b32_e32 v5, 0xffff, v5
-; GFX10-DL-NEXT:    v_bfe_u32 v12, v2, 8, 4
-; GFX10-DL-NEXT:    v_bfe_u32 v7, v1, 12, 4
-; GFX10-DL-NEXT:    v_and_b32_e32 v8, 0xffff, v8
-; GFX10-DL-NEXT:    v_lshl_or_b32 v4, v4, 16, v6
-; GFX10-DL-NEXT:    v_lshl_or_b32 v5, v9, 16, v5
-; GFX10-DL-NEXT:    v_bfe_u32 v9, v2, 12, 4
-; GFX10-DL-NEXT:    v_and_b32_e32 v12, 0xffff, v12
-; GFX10-DL-NEXT:    v_bfe_u32 v11, v1, 16, 4
-; GFX10-DL-NEXT:    v_lshl_or_b32 v7, v7, 16, v8
-; GFX10-DL-NEXT:    v_pk_mul_lo_u16 v4, v4, v5
-; GFX10-DL-NEXT:    v_bfe_u32 v5, v2, 16, 4
-; GFX10-DL-NEXT:    v_lshl_or_b32 v9, v9, 16, v12
+; GFX10-DL-NEXT:    v_bfe_u32 v6, v2, 4, 4
+; GFX10-DL-NEXT:    v_bfe_u32 v7, v1, 4, 4
+; GFX10-DL-NEXT:    v_bfe_u32 v8, v2, 12, 4
+; GFX10-DL-NEXT:    v_bfe_u32 v9, v1, 12, 4
 ; GFX10-DL-NEXT:    v_bfe_u32 v10, v1, 20, 4
-; GFX10-DL-NEXT:    v_and_b32_e32 v11, 0xffff, v11
+; GFX10-DL-NEXT:    v_perm_b32 v5, v6, v5, 0x5040100
+; GFX10-DL-NEXT:    v_perm_b32 v4, v7, v4, 0x5040100
+; GFX10-DL-NEXT:    v_bfe_u32 v6, v1, 8, 4
+; GFX10-DL-NEXT:    v_bfe_u32 v7, v2, 8, 4
+; GFX10-DL-NEXT:    v_pk_mul_lo_u16 v4, v4, v5
+; GFX10-DL-NEXT:    v_perm_b32 v6, v9, v6, 0x5040100
+; GFX10-DL-NEXT:    v_perm_b32 v7, v8, v7, 0x5040100
+; GFX10-DL-NEXT:    v_bfe_u32 v5, v1, 16, 4
+; GFX10-DL-NEXT:    v_bfe_u32 v9, v2, 20, 4
 ; GFX10-DL-NEXT:    v_lshrrev_b32_e32 v8, 16, v4
 ; GFX10-DL-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-DL-NEXT:    v_add_nc_u16 v3, v4, v3
-; GFX10-DL-NEXT:    v_bfe_u32 v4, v2, 20, 4
-; GFX10-DL-NEXT:    v_and_b32_e32 v5, 0xffff, v5
-; GFX10-DL-NEXT:    v_pk_mul_lo_u16 v7, v7, v9
-; GFX10-DL-NEXT:    v_lshrrev_b32_e32 v6, 28, v1
+; GFX10-DL-NEXT:    v_bfe_u32 v4, v2, 16, 4
+; GFX10-DL-NEXT:    v_pk_mul_lo_u16 v6, v6, v7
+; GFX10-DL-NEXT:    v_perm_b32 v5, v10, v5, 0x5040100
+; GFX10-DL-NEXT:    v_bfe_u32 v7, v1, 24, 4
 ; GFX10-DL-NEXT:    v_add_nc_u16 v3, v3, v8
-; GFX10-DL-NEXT:    v_bfe_u32 v1, v1, 24, 4
-; GFX10-DL-NEXT:    v_bfe_u32 v8, v2, 24, 4
-; GFX10-DL-NEXT:    v_lshl_or_b32 v4, v4, 16, v5
-; GFX10-DL-NEXT:    v_lshl_or_b32 v5, v10, 16, v11
-; GFX10-DL-NEXT:    v_lshrrev_b32_e32 v9, 16, v7
-; GFX10-DL-NEXT:    v_add_nc_u16 v3, v3, v7
+; GFX10-DL-NEXT:    v_perm_b32 v4, v9, v4, 0x5040100
+; GFX10-DL-NEXT:    v_lshrrev_b32_e32 v8, 16, v6
+; GFX10-DL-NEXT:    v_lshrrev_b32_e32 v1, 28, v1
+; GFX10-DL-NEXT:    v_add_nc_u16 v3, v3, v6
+; GFX10-DL-NEXT:    v_bfe_u32 v6, v2, 24, 4
 ; GFX10-DL-NEXT:    v_lshrrev_b32_e32 v2, 28, v2
-; GFX10-DL-NEXT:    v_and_b32_e32 v7, 0xffff, v8
-; GFX10-DL-NEXT:    v_and_b32_e32 v1, 0xffff, v1
 ; GFX10-DL-NEXT:    v_pk_mul_lo_u16 v4, v5, v4
-; GFX10-DL-NEXT:    v_add_nc_u16 v3, v3, v9
-; GFX10-DL-NEXT:    v_lshl_or_b32 v2, v2, 16, v7
-; GFX10-DL-NEXT:    v_lshl_or_b32 v1, v6, 16, v1
+; GFX10-DL-NEXT:    v_perm_b32 v1, v1, v7, 0x5040100
+; GFX10-DL-NEXT:    v_add_nc_u16 v3, v3, v8
+; GFX10-DL-NEXT:    v_perm_b32 v2, v2, v6, 0x5040100
 ; GFX10-DL-NEXT:    v_lshrrev_b32_e32 v5, 16, v4
 ; GFX10-DL-NEXT:    v_add_nc_u16 v3, v3, v4
 ; GFX10-DL-NEXT:    v_pk_mul_lo_u16 v1, v1, v2
diff --git a/llvm/test/CodeGen/AMDGPU/insert_vector_elt.v2i16.ll b/llvm/test/CodeGen/AMDGPU/insert_vector_elt.v2i16.ll
--- a/llvm/test/CodeGen/AMDGPU/insert_vector_elt.v2i16.ll
+++ b/llvm/test/CodeGen/AMDGPU/insert_vector_elt.v2i16.ll
@@ -662,12 +662,12 @@
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX9-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
-; GFX9-NEXT:    v_mov_b32_e32 v2, 0xffff
+; GFX9-NEXT:    v_mov_b32_e32 v2, 0x7060100
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    global_load_dword v1, v0, s[2:3]
 ; GFX9-NEXT:    s_movk_i32 s2, 0x3e7
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
-; GFX9-NEXT:    v_bfi_b32 v1, v2, s2, v1
+; GFX9-NEXT:    v_perm_b32 v1, v1, s2, v2
 ; GFX9-NEXT:    global_store_dword v0, v1, s[0:1]
 ; GFX9-NEXT:    s_endpgm
 ;
@@ -715,7 +715,7 @@
 ; GFX11-NEXT:    global_load_b32 v1, v0, s[2:3]
 ; GFX11-NEXT:    s_movk_i32 s2, 0x3e7
 ; GFX11-NEXT:    s_waitcnt vmcnt(0)
-; GFX11-NEXT:    v_bfi_b32 v1, 0xffff, s2, v1
+; GFX11-NEXT:    v_perm_b32 v1, v1, s2, 0x7060100
 ; GFX11-NEXT:    global_store_b32 v0, v1, s[0:1]
 ; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-NEXT:    s_endpgm
@@ -735,12 +735,11 @@
 ; GFX9-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX9-NEXT:    s_load_dword s6, s[4:5], 0x10
 ; GFX9-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
-; GFX9-NEXT:    v_mov_b32_e32 v3, 0xffff0000
+; GFX9-NEXT:    v_mov_b32_e32 v2, 0x7060302
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    global_load_dword v1, v0, s[2:3]
-; GFX9-NEXT:    v_lshrrev_b32_e64 v2, 16, s6
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
-; GFX9-NEXT:    v_and_or_b32 v1, v1, v3, v2
+; GFX9-NEXT:    v_perm_b32 v1, v1, s6, v2
 ; GFX9-NEXT:    global_store_dword v0, v1, s[0:1]
 ; GFX9-NEXT:    s_endpgm
 ;
@@ -790,10 +789,8 @@
 ; GFX11-NEXT:    s_load_b32 s0, s[0:1], 0x10
 ; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-NEXT:    global_load_b32 v1, v0, s[6:7]
-; GFX11-NEXT:    v_lshrrev_b32_e64 v2, 16, s0
 ; GFX11-NEXT:    s_waitcnt vmcnt(0)
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
-; GFX11-NEXT:    v_and_or_b32 v1, 0xffff0000, v1, v2
+; GFX11-NEXT:    v_perm_b32 v1, v1, s0, 0x7060302
 ; GFX11-NEXT:    global_store_b32 v0, v1, s[4:5]
 ; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-NEXT:    s_endpgm
@@ -814,11 +811,11 @@
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX9-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
-; GFX9-NEXT:    v_mov_b32_e32 v2, 0xffff
+; GFX9-NEXT:    v_mov_b32_e32 v2, 0x7060100
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    global_load_dword v1, v0, s[2:3]
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
-; GFX9-NEXT:    v_bfi_b32 v1, v2, 53, v1
+; GFX9-NEXT:    v_perm_b32 v1, v1, 53, v2
 ; GFX9-NEXT:    global_store_dword v0, v1, s[0:1]
 ; GFX9-NEXT:    s_endpgm
 ;
@@ -865,7 +862,7 @@
 ; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-NEXT:    global_load_b32 v1, v0, s[2:3]
 ; GFX11-NEXT:    s_waitcnt vmcnt(0)
-; GFX11-NEXT:    v_bfi_b32 v1, 0xffff, 53, v1
+; GFX11-NEXT:    v_perm_b32 v1, v1, 53, 0x7060100
 ; GFX11-NEXT:    global_store_b32 v0, v1, s[0:1]
 ; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-NEXT:    s_endpgm
@@ -885,12 +882,12 @@
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX9-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX9-NEXT:    v_mov_b32_e32 v2, 0x5040100
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    global_load_dword v1, v0, s[2:3]
 ; GFX9-NEXT:    s_movk_i32 s2, 0x3e7
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
-; GFX9-NEXT:    v_and_b32_e32 v1, 0xffff, v1
-; GFX9-NEXT:    v_lshl_or_b32 v1, s2, 16, v1
+; GFX9-NEXT:    v_perm_b32 v1, s2, v1, v2
 ; GFX9-NEXT:    global_store_dword v0, v1, s[0:1]
 ; GFX9-NEXT:    s_endpgm
 ;
@@ -936,10 +933,9 @@
 ; GFX11-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
 ; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-NEXT:    global_load_b32 v1, v0, s[2:3]
+; GFX11-NEXT:    s_movk_i32 s2, 0x3e7
 ; GFX11-NEXT:    s_waitcnt vmcnt(0)
-; GFX11-NEXT:    v_and_b32_e32 v1, 0xffff, v1
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
-; GFX11-NEXT:    v_lshl_or_b32 v1, 0x3e7, 16, v1
+; GFX11-NEXT:    v_perm_b32 v1, s2, v1, 0x5040100
 ; GFX11-NEXT:    global_store_b32 v0, v1, s[0:1]
 ; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-NEXT:    s_endpgm
@@ -958,11 +954,11 @@
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX9-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX9-NEXT:    v_mov_b32_e32 v2, 0x5040100
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    global_load_dword v1, v0, s[2:3]
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
-; GFX9-NEXT:    v_and_b32_e32 v1, 0xffff, v1
-; GFX9-NEXT:    v_lshl_or_b32 v1, -15, 16, v1
+; GFX9-NEXT:    v_perm_b32 v1, -15, v1, v2
 ; GFX9-NEXT:    global_store_dword v0, v1, s[0:1]
 ; GFX9-NEXT:    s_endpgm
 ;
@@ -1009,9 +1005,7 @@
 ; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-NEXT:    global_load_b32 v1, v0, s[2:3]
 ; GFX11-NEXT:    s_waitcnt vmcnt(0)
-; GFX11-NEXT:    v_and_b32_e32 v1, 0xffff, v1
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
-; GFX11-NEXT:    v_lshl_or_b32 v1, -15, 16, v1
+; GFX11-NEXT:    v_perm_b32 v1, -15, v1, 0x5040100
 ; GFX11-NEXT:    global_store_b32 v0, v1, s[0:1]
 ; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-NEXT:    s_endpgm
@@ -1030,12 +1024,12 @@
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX9-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
-; GFX9-NEXT:    v_mov_b32_e32 v2, 0x4500
+; GFX9-NEXT:    v_mov_b32_e32 v2, 0x7060100
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    global_load_dword v1, v0, s[2:3]
+; GFX9-NEXT:    s_movk_i32 s2, 0x4500
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
-; GFX9-NEXT:    v_lshrrev_b32_e32 v1, 16, v1
-; GFX9-NEXT:    v_lshl_or_b32 v1, v1, 16, v2
+; GFX9-NEXT:    v_perm_b32 v1, v1, s2, v2
 ; GFX9-NEXT:    global_store_dword v0, v1, s[0:1]
 ; GFX9-NEXT:    s_endpgm
 ;
@@ -1081,10 +1075,9 @@
 ; GFX11-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
 ; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-NEXT:    global_load_b32 v1, v0, s[2:3]
+; GFX11-NEXT:    s_movk_i32 s2, 0x4500
 ; GFX11-NEXT:    s_waitcnt vmcnt(0)
-; GFX11-NEXT:    v_lshrrev_b32_e32 v1, 16, v1
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
-; GFX11-NEXT:    v_lshl_or_b32 v1, v1, 16, 0x4500
+; GFX11-NEXT:    v_perm_b32 v1, v1, s2, 0x7060100
 ; GFX11-NEXT:    global_store_b32 v0, v1, s[0:1]
 ; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-NEXT:    s_endpgm
@@ -1103,11 +1096,11 @@
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX9-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX9-NEXT:    v_mov_b32_e32 v2, 0x7060100
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    global_load_dword v1, v0, s[2:3]
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
-; GFX9-NEXT:    v_lshrrev_b32_e32 v1, 16, v1
-; GFX9-NEXT:    v_lshl_or_b32 v1, v1, 16, 53
+; GFX9-NEXT:    v_perm_b32 v1, v1, 53, v2
 ; GFX9-NEXT:    global_store_dword v0, v1, s[0:1]
 ; GFX9-NEXT:    s_endpgm
 ;
@@ -1154,9 +1147,7 @@
 ; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-NEXT:    global_load_b32 v1, v0, s[2:3]
 ; GFX11-NEXT:    s_waitcnt vmcnt(0)
-; GFX11-NEXT:    v_lshrrev_b32_e32 v1, 16, v1
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
-; GFX11-NEXT:    v_lshl_or_b32 v1, v1, 16, 53
+; GFX11-NEXT:    v_perm_b32 v1, v1, 53, 0x7060100
 ; GFX11-NEXT:    global_store_b32 v0, v1, s[0:1]
 ; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-NEXT:    s_endpgm
@@ -1175,12 +1166,12 @@
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX9-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX9-NEXT:    v_mov_b32_e32 v2, 0x5040100
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    global_load_dword v1, v0, s[2:3]
 ; GFX9-NEXT:    s_movk_i32 s2, 0x4500
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
-; GFX9-NEXT:    v_and_b32_e32 v1, 0xffff, v1
-; GFX9-NEXT:    v_lshl_or_b32 v1, s2, 16, v1
+; GFX9-NEXT:    v_perm_b32 v1, s2, v1, v2
 ; GFX9-NEXT:    global_store_dword v0, v1, s[0:1]
 ; GFX9-NEXT:    s_endpgm
 ;
@@ -1226,10 +1217,9 @@
 ; GFX11-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
 ; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-NEXT:    global_load_b32 v1, v0, s[2:3]
+; GFX11-NEXT:    s_movk_i32 s2, 0x4500
 ; GFX11-NEXT:    s_waitcnt vmcnt(0)
-; GFX11-NEXT:    v_and_b32_e32 v1, 0xffff, v1
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
-; GFX11-NEXT:    v_lshl_or_b32 v1, 0x4500, 16, v1
+; GFX11-NEXT:    v_perm_b32 v1, s2, v1, 0x5040100
 ; GFX11-NEXT:    global_store_b32 v0, v1, s[0:1]
 ; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-NEXT:    s_endpgm
@@ -1248,11 +1238,11 @@
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX9-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX9-NEXT:    v_mov_b32_e32 v2, 0x5040100
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    global_load_dword v1, v0, s[2:3]
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
-; GFX9-NEXT:    v_and_b32_e32 v1, 0xffff, v1
-; GFX9-NEXT:    v_lshl_or_b32 v1, 35, 16, v1
+; GFX9-NEXT:    v_perm_b32 v1, 35, v1, v2
 ; GFX9-NEXT:    global_store_dword v0, v1, s[0:1]
 ; GFX9-NEXT:    s_endpgm
 ;
@@ -1299,9 +1289,7 @@
 ; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-NEXT:    global_load_b32 v1, v0, s[2:3]
 ; GFX11-NEXT:    s_waitcnt vmcnt(0)
-; GFX11-NEXT:    v_and_b32_e32 v1, 0xffff, v1
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
-; GFX11-NEXT:    v_lshl_or_b32 v1, 35, 16, v1
+; GFX11-NEXT:    v_perm_b32 v1, 35, v1, 0x5040100
 ; GFX11-NEXT:    global_store_b32 v0, v1, s[0:1]
 ; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-NEXT:    s_endpgm
@@ -1592,11 +1580,11 @@
 ; GFX9-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX9-NEXT:    s_load_dword s6, s[4:5], 0x30
 ; GFX9-NEXT:    v_lshlrev_b32_e32 v2, 3, v0
-; GFX9-NEXT:    v_mov_b32_e32 v3, 0xffff
+; GFX9-NEXT:    v_mov_b32_e32 v3, 0x7060100
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    global_load_dwordx2 v[0:1], v2, s[2:3]
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
-; GFX9-NEXT:    v_bfi_b32 v0, v3, s6, v0
+; GFX9-NEXT:    v_perm_b32 v0, v0, s6, v3
 ; GFX9-NEXT:    global_store_dwordx2 v2, v[0:1], s[0:1]
 ; GFX9-NEXT:    s_endpgm
 ;
@@ -1648,7 +1636,7 @@
 ; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-NEXT:    global_load_b64 v[0:1], v2, s[6:7]
 ; GFX11-NEXT:    s_waitcnt vmcnt(0)
-; GFX11-NEXT:    v_bfi_b32 v0, 0xffff, s0, v0
+; GFX11-NEXT:    v_perm_b32 v0, v0, s0, 0x7060100
 ; GFX11-NEXT:    global_store_b64 v2, v[0:1], s[4:5]
 ; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-NEXT:    s_endpgm
@@ -1670,11 +1658,11 @@
 ; GFX9-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX9-NEXT:    s_load_dword s6, s[4:5], 0x10
 ; GFX9-NEXT:    v_lshlrev_b32_e32 v2, 3, v0
+; GFX9-NEXT:    v_mov_b32_e32 v3, 0x5040100
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    global_load_dwordx2 v[0:1], v2, s[2:3]
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
-; GFX9-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX9-NEXT:    v_lshl_or_b32 v0, s6, 16, v0
+; GFX9-NEXT:    v_perm_b32 v0, s6, v0, v3
 ; GFX9-NEXT:    global_store_dwordx2 v2, v[0:1], s[0:1]
 ; GFX9-NEXT:    s_endpgm
 ;
@@ -1726,9 +1714,7 @@
 ; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-NEXT:    global_load_b64 v[0:1], v2, s[6:7]
 ; GFX11-NEXT:    s_waitcnt vmcnt(0)
-; GFX11-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
-; GFX11-NEXT:    v_lshl_or_b32 v0, s0, 16, v0
+; GFX11-NEXT:    v_perm_b32 v0, s0, v0, 0x5040100
 ; GFX11-NEXT:    global_store_b64 v2, v[0:1], s[4:5]
 ; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-NEXT:    s_endpgm
@@ -1750,11 +1736,11 @@
 ; GFX9-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX9-NEXT:    s_load_dword s6, s[4:5], 0x30
 ; GFX9-NEXT:    v_lshlrev_b32_e32 v2, 3, v0
-; GFX9-NEXT:    v_mov_b32_e32 v3, 0xffff
+; GFX9-NEXT:    v_mov_b32_e32 v3, 0x7060100
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    global_load_dwordx2 v[0:1], v2, s[2:3]
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
-; GFX9-NEXT:    v_bfi_b32 v1, v3, s6, v1
+; GFX9-NEXT:    v_perm_b32 v1, v1, s6, v3
 ; GFX9-NEXT:    global_store_dwordx2 v2, v[0:1], s[0:1]
 ; GFX9-NEXT:    s_endpgm
 ;
@@ -1806,7 +1792,7 @@
 ; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-NEXT:    global_load_b64 v[0:1], v2, s[6:7]
 ; GFX11-NEXT:    s_waitcnt vmcnt(0)
-; GFX11-NEXT:    v_bfi_b32 v1, 0xffff, s0, v1
+; GFX11-NEXT:    v_perm_b32 v1, v1, s0, 0x7060100
 ; GFX11-NEXT:    global_store_b64 v2, v[0:1], s[4:5]
 ; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-NEXT:    s_endpgm
@@ -1828,11 +1814,11 @@
 ; GFX9-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX9-NEXT:    s_load_dword s6, s[4:5], 0x10
 ; GFX9-NEXT:    v_lshlrev_b32_e32 v2, 3, v0
+; GFX9-NEXT:    v_mov_b32_e32 v3, 0x5040100
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    global_load_dwordx2 v[0:1], v2, s[2:3]
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
-; GFX9-NEXT:    v_and_b32_e32 v1, 0xffff, v1
-; GFX9-NEXT:    v_lshl_or_b32 v1, s6, 16, v1
+; GFX9-NEXT:    v_perm_b32 v1, s6, v1, v3
 ; GFX9-NEXT:    global_store_dwordx2 v2, v[0:1], s[0:1]
 ; GFX9-NEXT:    s_endpgm
 ;
@@ -1884,9 +1870,7 @@
 ; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-NEXT:    global_load_b64 v[0:1], v2, s[6:7]
 ; GFX11-NEXT:    s_waitcnt vmcnt(0)
-; GFX11-NEXT:    v_and_b32_e32 v1, 0xffff, v1
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
-; GFX11-NEXT:    v_lshl_or_b32 v1, s0, 16, v1
+; GFX11-NEXT:    v_perm_b32 v1, s0, v1, 0x5040100
 ; GFX11-NEXT:    global_store_b64 v2, v[0:1], s[4:5]
 ; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-NEXT:    s_endpgm
@@ -1908,11 +1892,11 @@
 ; GFX9-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX9-NEXT:    s_load_dword s6, s[4:5], 0x10
 ; GFX9-NEXT:    v_lshlrev_b32_e32 v2, 3, v0
-; GFX9-NEXT:    v_mov_b32_e32 v3, 0xffff
+; GFX9-NEXT:    v_mov_b32_e32 v3, 0x7060100
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    global_load_dwordx2 v[0:1], v2, s[2:3]
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
-; GFX9-NEXT:    v_bfi_b32 v1, v3, s6, v1
+; GFX9-NEXT:    v_perm_b32 v1, v1, s6, v3
 ; GFX9-NEXT:    global_store_dwordx2 v2, v[0:1], s[0:1]
 ; GFX9-NEXT:    s_endpgm
 ;
@@ -1964,7 +1948,7 @@
 ; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-NEXT:    global_load_b64 v[0:1], v2, s[6:7]
 ; GFX11-NEXT:    s_waitcnt vmcnt(0)
-; GFX11-NEXT:    v_bfi_b32 v1, 0xffff, s0, v1
+; GFX11-NEXT:    v_perm_b32 v1, v1, s0, 0x7060100
 ; GFX11-NEXT:    global_store_b64 v2, v[0:1], s[4:5]
 ; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-NEXT:    s_endpgm
@@ -2199,11 +2183,11 @@
 ; GFX9-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX9-NEXT:    s_load_dword s6, s[4:5], 0x10
 ; GFX9-NEXT:    v_lshlrev_b32_e32 v4, 4, v0
+; GFX9-NEXT:    v_mov_b32_e32 v5, 0x5040100
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    global_load_dwordx4 v[0:3], v4, s[2:3]
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
-; GFX9-NEXT:    v_and_b32_e32 v1, 0xffff, v1
-; GFX9-NEXT:    v_lshl_or_b32 v1, s6, 16, v1
+; GFX9-NEXT:    v_perm_b32 v1, s6, v1, v5
 ; GFX9-NEXT:    global_store_dwordx4 v4, v[0:3], s[0:1]
 ; GFX9-NEXT:    s_endpgm
 ;
@@ -2257,9 +2241,7 @@
 ; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-NEXT:    global_load_b128 v[0:3], v4, s[6:7]
 ; GFX11-NEXT:    s_waitcnt vmcnt(0)
-; GFX11-NEXT:    v_and_b32_e32 v1, 0xffff, v1
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
-; GFX11-NEXT:    v_lshl_or_b32 v1, s0, 16, v1
+; GFX11-NEXT:    v_perm_b32 v1, s0, v1, 0x5040100
 ; GFX11-NEXT:    global_store_b128 v4, v[0:3], s[4:5]
 ; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-NEXT:    s_endpgm
@@ -2281,11 +2263,11 @@
 ; GFX9-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX9-NEXT:    s_load_dword s6, s[4:5], 0x10
 ; GFX9-NEXT:    v_lshlrev_b32_e32 v4, 4, v0
-; GFX9-NEXT:    v_mov_b32_e32 v5, 0xffff
+; GFX9-NEXT:    v_mov_b32_e32 v5, 0x7060100
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    global_load_dwordx4 v[0:3], v4, s[2:3]
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
-; GFX9-NEXT:    v_bfi_b32 v3, v5, s6, v3
+; GFX9-NEXT:    v_perm_b32 v3, v3, s6, v5
 ; GFX9-NEXT:    global_store_dwordx4 v4, v[0:3], s[0:1]
 ; GFX9-NEXT:    s_endpgm
 ;
@@ -2338,7 +2320,7 @@
 ; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-NEXT:    global_load_b128 v[0:3], v4, s[6:7]
 ; GFX11-NEXT:    s_waitcnt vmcnt(0)
-; GFX11-NEXT:    v_bfi_b32 v3, 0xffff, s0, v3
+; GFX11-NEXT:    v_perm_b32 v3, v3, s0, 0x7060100
 ; GFX11-NEXT:    global_store_b128 v4, v[0:3], s[4:5]
 ; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-NEXT:    s_endpgm
@@ -2362,44 +2344,41 @@
 ; GFX9-NEXT:    v_lshlrev_b32_e32 v4, 4, v0
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    global_load_dwordx4 v[0:3], v4, s[2:3]
-; GFX9-NEXT:    s_cmp_eq_u32 s7, 7
+; GFX9-NEXT:    s_cmp_eq_u32 s7, 6
 ; GFX9-NEXT:    v_mov_b32_e32 v5, s6
 ; GFX9-NEXT:    s_cselect_b64 vcc, -1, 0
-; GFX9-NEXT:    s_cmp_eq_u32 s7, 6
+; GFX9-NEXT:    s_cmp_eq_u32 s7, 7
+; GFX9-NEXT:    s_mov_b32 s2, 0x5040100
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
-; GFX9-NEXT:    v_lshrrev_b32_e32 v6, 16, v3
-; GFX9-NEXT:    v_cndmask_b32_e32 v6, v6, v5, vcc
+; GFX9-NEXT:    v_cndmask_b32_e32 v6, v3, v5, vcc
+; GFX9-NEXT:    v_lshrrev_b32_e32 v3, 16, v3
+; GFX9-NEXT:    s_cselect_b64 vcc, -1, 0
+; GFX9-NEXT:    s_cmp_eq_u32 s7, 4
+; GFX9-NEXT:    v_cndmask_b32_e32 v3, v3, v5, vcc
 ; GFX9-NEXT:    s_cselect_b64 vcc, -1, 0
 ; GFX9-NEXT:    s_cmp_eq_u32 s7, 5
 ; GFX9-NEXT:    v_lshrrev_b32_e32 v7, 16, v2
-; GFX9-NEXT:    v_cndmask_b32_e32 v3, v3, v5, vcc
+; GFX9-NEXT:    v_cndmask_b32_e32 v2, v2, v5, vcc
 ; GFX9-NEXT:    s_cselect_b64 vcc, -1, 0
-; GFX9-NEXT:    s_cmp_eq_u32 s7, 4
-; GFX9-NEXT:    v_cndmask_b32_e32 v7, v7, v5, vcc
+; GFX9-NEXT:    s_cmp_eq_u32 s7, 2
+; GFX9-NEXT:    v_perm_b32 v3, v3, v6, s2
+; GFX9-NEXT:    v_cndmask_b32_e32 v6, v7, v5, vcc
 ; GFX9-NEXT:    s_cselect_b64 vcc, -1, 0
 ; GFX9-NEXT:    s_cmp_eq_u32 s7, 3
 ; GFX9-NEXT:    v_lshrrev_b32_e32 v8, 16, v1
-; GFX9-NEXT:    v_and_b32_e32 v3, 0xffff, v3
-; GFX9-NEXT:    v_cndmask_b32_e32 v2, v2, v5, vcc
+; GFX9-NEXT:    v_cndmask_b32_e32 v1, v1, v5, vcc
 ; GFX9-NEXT:    s_cselect_b64 vcc, -1, 0
-; GFX9-NEXT:    s_cmp_eq_u32 s7, 2
-; GFX9-NEXT:    v_lshl_or_b32 v3, v6, 16, v3
+; GFX9-NEXT:    s_cmp_eq_u32 s7, 0
+; GFX9-NEXT:    v_perm_b32 v2, v6, v2, s2
 ; GFX9-NEXT:    v_cndmask_b32_e32 v6, v8, v5, vcc
 ; GFX9-NEXT:    s_cselect_b64 vcc, -1, 0
 ; GFX9-NEXT:    s_cmp_eq_u32 s7, 1
 ; GFX9-NEXT:    v_lshrrev_b32_e32 v9, 16, v0
-; GFX9-NEXT:    v_and_b32_e32 v2, 0xffff, v2
-; GFX9-NEXT:    v_cndmask_b32_e32 v1, v1, v5, vcc
-; GFX9-NEXT:    s_cselect_b64 vcc, -1, 0
-; GFX9-NEXT:    s_cmp_eq_u32 s7, 0
-; GFX9-NEXT:    v_lshl_or_b32 v2, v7, 16, v2
-; GFX9-NEXT:    v_cndmask_b32_e32 v7, v9, v5, vcc
-; GFX9-NEXT:    s_cselect_b64 vcc, -1, 0
 ; GFX9-NEXT:    v_cndmask_b32_e32 v0, v0, v5, vcc
-; GFX9-NEXT:    v_and_b32_e32 v1, 0xffff, v1
-; GFX9-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX9-NEXT:    v_lshl_or_b32 v1, v6, 16, v1
-; GFX9-NEXT:    v_lshl_or_b32 v0, v7, 16, v0
+; GFX9-NEXT:    s_cselect_b64 vcc, -1, 0
+; GFX9-NEXT:    v_cndmask_b32_e32 v5, v9, v5, vcc
+; GFX9-NEXT:    v_perm_b32 v1, v6, v1, s2
+; GFX9-NEXT:    v_perm_b32 v0, v5, v0, s2
 ; GFX9-NEXT:    global_store_dwordx4 v4, v[0:3], s[0:1]
 ; GFX9-NEXT:    s_endpgm
 ;
@@ -2534,44 +2513,40 @@
 ; GFX11-NEXT:    s_load_b64 s[0:1], s[0:1], 0x10
 ; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-NEXT:    global_load_b128 v[0:3], v4, s[6:7]
+; GFX11-NEXT:    s_cmp_eq_u32 s1, 6
+; GFX11-NEXT:    s_cselect_b32 s2, -1, 0
 ; GFX11-NEXT:    s_cmp_eq_u32 s1, 7
+; GFX11-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-NEXT:    v_cndmask_b32_e64 v5, v3, s0, s2
 ; GFX11-NEXT:    s_cselect_b32 s2, -1, 0
-; GFX11-NEXT:    s_cmp_eq_u32 s1, 6
+; GFX11-NEXT:    s_cmp_eq_u32 s1, 4
+; GFX11-NEXT:    v_lshrrev_b32_e32 v3, 16, v3
 ; GFX11-NEXT:    s_cselect_b32 s3, -1, 0
 ; GFX11-NEXT:    s_cmp_eq_u32 s1, 5
-; GFX11-NEXT:    s_waitcnt vmcnt(0)
-; GFX11-NEXT:    v_lshrrev_b32_e32 v5, 16, v3
 ; GFX11-NEXT:    v_lshrrev_b32_e32 v6, 16, v2
-; GFX11-NEXT:    v_cndmask_b32_e64 v3, v3, s0, s3
+; GFX11-NEXT:    v_cndmask_b32_e64 v2, v2, s0, s3
 ; GFX11-NEXT:    s_cselect_b32 s3, -1, 0
-; GFX11-NEXT:    s_cmp_eq_u32 s1, 4
-; GFX11-NEXT:    v_cndmask_b32_e64 v5, v5, s0, s2
+; GFX11-NEXT:    s_cmp_eq_u32 s1, 2
+; GFX11-NEXT:    v_lshrrev_b32_e32 v7, 16, v1
+; GFX11-NEXT:    v_cndmask_b32_e64 v3, v3, s0, s2
 ; GFX11-NEXT:    s_cselect_b32 s2, -1, 0
 ; GFX11-NEXT:    s_cmp_eq_u32 s1, 3
-; GFX11-NEXT:    v_cndmask_b32_e64 v6, v6, s0, s3
-; GFX11-NEXT:    s_cselect_b32 s3, -1, 0
-; GFX11-NEXT:    s_cmp_eq_u32 s1, 2
-; GFX11-NEXT:    v_cndmask_b32_e64 v2, v2, s0, s2
+; GFX11-NEXT:    v_cndmask_b32_e64 v1, v1, s0, s2
 ; GFX11-NEXT:    s_cselect_b32 s2, -1, 0
-; GFX11-NEXT:    s_cmp_eq_u32 s1, 1
-; GFX11-NEXT:    v_lshrrev_b32_e32 v7, 16, v1
-; GFX11-NEXT:    s_cselect_b32 s6, -1, 0
 ; GFX11-NEXT:    s_cmp_eq_u32 s1, 0
 ; GFX11-NEXT:    v_lshrrev_b32_e32 v8, 16, v0
+; GFX11-NEXT:    v_cndmask_b32_e64 v7, v7, s0, s2
+; GFX11-NEXT:    s_cselect_b32 s2, -1, 0
+; GFX11-NEXT:    s_cmp_eq_u32 s1, 1
+; GFX11-NEXT:    v_cndmask_b32_e64 v6, v6, s0, s3
 ; GFX11-NEXT:    s_cselect_b32 s1, -1, 0
-; GFX11-NEXT:    v_cndmask_b32_e64 v1, v1, s0, s2
-; GFX11-NEXT:    v_cndmask_b32_e64 v0, v0, s0, s1
-; GFX11-NEXT:    v_and_b32_e32 v3, 0xffff, v3
-; GFX11-NEXT:    v_and_b32_e32 v2, 0xffff, v2
-; GFX11-NEXT:    v_cndmask_b32_e64 v7, v7, s0, s3
-; GFX11-NEXT:    v_and_b32_e32 v1, 0xffff, v1
-; GFX11-NEXT:    v_cndmask_b32_e64 v8, v8, s0, s6
-; GFX11-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX11-NEXT:    v_lshl_or_b32 v3, v5, 16, v3
-; GFX11-NEXT:    v_lshl_or_b32 v2, v6, 16, v2
-; GFX11-NEXT:    v_lshl_or_b32 v1, v7, 16, v1
+; GFX11-NEXT:    v_cndmask_b32_e64 v0, v0, s0, s2
+; GFX11-NEXT:    v_cndmask_b32_e64 v8, v8, s0, s1
+; GFX11-NEXT:    v_perm_b32 v3, v3, v5, 0x5040100
+; GFX11-NEXT:    v_perm_b32 v2, v6, v2, 0x5040100
+; GFX11-NEXT:    v_perm_b32 v1, v7, v1, 0x5040100
 ; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_4)
-; GFX11-NEXT:    v_lshl_or_b32 v0, v8, 16, v0
+; GFX11-NEXT:    v_perm_b32 v0, v8, v0, 0x5040100
 ; GFX11-NEXT:    global_store_b128 v4, v[0:3], s[4:5]
 ; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-NEXT:    s_endpgm
@@ -2593,12 +2568,12 @@
 ; GFX9-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX9-NEXT:    s_load_dword s6, s[4:5], 0x10
 ; GFX9-NEXT:    v_lshlrev_b32_e32 v8, 5, v0
+; GFX9-NEXT:    v_mov_b32_e32 v9, 0x5040100
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    global_load_dwordx4 v[0:3], v8, s[2:3]
 ; GFX9-NEXT:    global_load_dwordx4 v[4:7], v8, s[2:3] offset:16
 ; GFX9-NEXT:    s_waitcnt vmcnt(1)
-; GFX9-NEXT:    v_and_b32_e32 v1, 0xffff, v1
-; GFX9-NEXT:    v_lshl_or_b32 v1, s6, 16, v1
+; GFX9-NEXT:    v_perm_b32 v1, s6, v1, v9
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    global_store_dwordx4 v8, v[4:7], s[0:1] offset:16
 ; GFX9-NEXT:    global_store_dwordx4 v8, v[0:3], s[0:1]
@@ -2668,9 +2643,7 @@
 ; GFX11-NEXT:    global_load_b128 v[0:3], v8, s[6:7]
 ; GFX11-NEXT:    global_load_b128 v[4:7], v8, s[6:7] offset:16
 ; GFX11-NEXT:    s_waitcnt vmcnt(1)
-; GFX11-NEXT:    v_and_b32_e32 v1, 0xffff, v1
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
-; GFX11-NEXT:    v_lshl_or_b32 v1, s0, 16, v1
+; GFX11-NEXT:    v_perm_b32 v1, s0, v1, 0x5040100
 ; GFX11-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-NEXT:    s_clause 0x1
 ; GFX11-NEXT:    global_store_b128 v8, v[4:7], s[4:5] offset:16
@@ -2695,12 +2668,12 @@
 ; GFX9-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX9-NEXT:    s_load_dword s6, s[4:5], 0x10
 ; GFX9-NEXT:    v_lshlrev_b32_e32 v8, 5, v0
-; GFX9-NEXT:    v_mov_b32_e32 v9, 0xffff
+; GFX9-NEXT:    v_mov_b32_e32 v9, 0x7060100
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    global_load_dwordx4 v[0:3], v8, s[2:3]
 ; GFX9-NEXT:    global_load_dwordx4 v[4:7], v8, s[2:3] offset:16
 ; GFX9-NEXT:    s_waitcnt vmcnt(1)
-; GFX9-NEXT:    v_bfi_b32 v3, v9, s6, v3
+; GFX9-NEXT:    v_perm_b32 v3, v3, s6, v9
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    global_store_dwordx4 v8, v[4:7], s[0:1] offset:16
 ; GFX9-NEXT:    global_store_dwordx4 v8, v[0:3], s[0:1]
@@ -2770,7 +2743,7 @@
 ; GFX11-NEXT:    global_load_b128 v[0:3], v8, s[6:7]
 ; GFX11-NEXT:    global_load_b128 v[4:7], v8, s[6:7] offset:16
 ; GFX11-NEXT:    s_waitcnt vmcnt(1)
-; GFX11-NEXT:    v_bfi_b32 v3, 0xffff, s0, v3
+; GFX11-NEXT:    v_perm_b32 v3, v3, s0, 0x7060100
 ; GFX11-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-NEXT:    s_clause 0x1
 ; GFX11-NEXT:    global_store_b128 v8, v[4:7], s[4:5] offset:16
@@ -2798,81 +2771,74 @@
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    global_load_dwordx4 v[0:3], v8, s[2:3]
 ; GFX9-NEXT:    global_load_dwordx4 v[4:7], v8, s[2:3] offset:16
-; GFX9-NEXT:    s_cmp_eq_u32 s7, 7
+; GFX9-NEXT:    s_cmp_eq_u32 s7, 6
 ; GFX9-NEXT:    v_mov_b32_e32 v9, s6
 ; GFX9-NEXT:    s_cselect_b64 vcc, -1, 0
-; GFX9-NEXT:    s_cmp_eq_u32 s7, 6
+; GFX9-NEXT:    s_cmp_eq_u32 s7, 7
+; GFX9-NEXT:    s_mov_b32 s2, 0x5040100
 ; GFX9-NEXT:    s_waitcnt vmcnt(1)
-; GFX9-NEXT:    v_lshrrev_b32_e32 v10, 16, v3
-; GFX9-NEXT:    v_cndmask_b32_e32 v10, v10, v9, vcc
+; GFX9-NEXT:    v_cndmask_b32_e32 v10, v3, v9, vcc
+; GFX9-NEXT:    v_lshrrev_b32_e32 v3, 16, v3
+; GFX9-NEXT:    s_cselect_b64 vcc, -1, 0
+; GFX9-NEXT:    s_cmp_eq_u32 s7, 4
+; GFX9-NEXT:    v_cndmask_b32_e32 v3, v3, v9, vcc
 ; GFX9-NEXT:    s_cselect_b64 vcc, -1, 0
 ; GFX9-NEXT:    s_cmp_eq_u32 s7, 5
 ; GFX9-NEXT:    v_lshrrev_b32_e32 v11, 16, v2
-; GFX9-NEXT:    v_cndmask_b32_e32 v3, v3, v9, vcc
+; GFX9-NEXT:    v_cndmask_b32_e32 v2, v2, v9, vcc
 ; GFX9-NEXT:    s_cselect_b64 vcc, -1, 0
-; GFX9-NEXT:    s_cmp_eq_u32 s7, 4
-; GFX9-NEXT:    v_cndmask_b32_e32 v11, v11, v9, vcc
+; GFX9-NEXT:    s_cmp_eq_u32 s7, 2
+; GFX9-NEXT:    v_perm_b32 v3, v3, v10, s2
+; GFX9-NEXT:    v_cndmask_b32_e32 v10, v11, v9, vcc
 ; GFX9-NEXT:    s_cselect_b64 vcc, -1, 0
-; GFX9-NEXT:    v_cndmask_b32_e32 v2, v2, v9, vcc
 ; GFX9-NEXT:    s_cmp_eq_u32 s7, 3
 ; GFX9-NEXT:    v_lshrrev_b32_e32 v12, 16, v1
-; GFX9-NEXT:    v_and_b32_e32 v2, 0xffff, v2
+; GFX9-NEXT:    v_cndmask_b32_e32 v1, v1, v9, vcc
 ; GFX9-NEXT:    s_cselect_b64 vcc, -1, 0
-; GFX9-NEXT:    s_cmp_eq_u32 s7, 2
-; GFX9-NEXT:    v_lshl_or_b32 v2, v11, 16, v2
+; GFX9-NEXT:    s_cmp_eq_u32 s7, 0
 ; GFX9-NEXT:    v_cndmask_b32_e32 v11, v12, v9, vcc
 ; GFX9-NEXT:    s_cselect_b64 vcc, -1, 0
 ; GFX9-NEXT:    s_cmp_eq_u32 s7, 1
 ; GFX9-NEXT:    v_lshrrev_b32_e32 v13, 16, v0
-; GFX9-NEXT:    v_cndmask_b32_e32 v1, v1, v9, vcc
+; GFX9-NEXT:    v_cndmask_b32_e32 v0, v0, v9, vcc
 ; GFX9-NEXT:    s_cselect_b64 vcc, -1, 0
-; GFX9-NEXT:    s_cmp_eq_u32 s7, 0
+; GFX9-NEXT:    s_cmp_eq_u32 s7, 14
 ; GFX9-NEXT:    v_cndmask_b32_e32 v12, v13, v9, vcc
 ; GFX9-NEXT:    s_cselect_b64 vcc, -1, 0
 ; GFX9-NEXT:    s_cmp_eq_u32 s7, 15
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    v_lshrrev_b32_e32 v14, 16, v7
-; GFX9-NEXT:    v_cndmask_b32_e32 v0, v0, v9, vcc
+; GFX9-NEXT:    v_cndmask_b32_e32 v7, v7, v9, vcc
 ; GFX9-NEXT:    s_cselect_b64 vcc, -1, 0
-; GFX9-NEXT:    s_cmp_eq_u32 s7, 14
-; GFX9-NEXT:    v_cndmask_b32_e32 v13, v14, v9, vcc
+; GFX9-NEXT:    s_cmp_eq_u32 s7, 12
+; GFX9-NEXT:    v_perm_b32 v0, v12, v0, s2
+; GFX9-NEXT:    v_cndmask_b32_e32 v12, v14, v9, vcc
 ; GFX9-NEXT:    s_cselect_b64 vcc, -1, 0
 ; GFX9-NEXT:    s_cmp_eq_u32 s7, 13
 ; GFX9-NEXT:    v_lshrrev_b32_e32 v15, 16, v6
-; GFX9-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX9-NEXT:    v_cndmask_b32_e32 v7, v7, v9, vcc
+; GFX9-NEXT:    v_cndmask_b32_e32 v6, v6, v9, vcc
 ; GFX9-NEXT:    s_cselect_b64 vcc, -1, 0
-; GFX9-NEXT:    s_cmp_eq_u32 s7, 12
-; GFX9-NEXT:    v_and_b32_e32 v3, 0xffff, v3
-; GFX9-NEXT:    v_lshl_or_b32 v0, v12, 16, v0
+; GFX9-NEXT:    s_cmp_eq_u32 s7, 10
+; GFX9-NEXT:    v_perm_b32 v7, v12, v7, s2
 ; GFX9-NEXT:    v_cndmask_b32_e32 v12, v15, v9, vcc
 ; GFX9-NEXT:    s_cselect_b64 vcc, -1, 0
 ; GFX9-NEXT:    s_cmp_eq_u32 s7, 11
-; GFX9-NEXT:    v_lshl_or_b32 v3, v10, 16, v3
+; GFX9-NEXT:    v_perm_b32 v2, v10, v2, s2
 ; GFX9-NEXT:    v_lshrrev_b32_e32 v10, 16, v5
-; GFX9-NEXT:    v_cndmask_b32_e32 v6, v6, v9, vcc
+; GFX9-NEXT:    v_cndmask_b32_e32 v5, v5, v9, vcc
 ; GFX9-NEXT:    s_cselect_b64 vcc, -1, 0
-; GFX9-NEXT:    s_cmp_eq_u32 s7, 10
-; GFX9-NEXT:    v_and_b32_e32 v1, 0xffff, v1
+; GFX9-NEXT:    s_cmp_eq_u32 s7, 8
 ; GFX9-NEXT:    v_cndmask_b32_e32 v10, v10, v9, vcc
 ; GFX9-NEXT:    s_cselect_b64 vcc, -1, 0
 ; GFX9-NEXT:    s_cmp_eq_u32 s7, 9
-; GFX9-NEXT:    v_lshl_or_b32 v1, v11, 16, v1
+; GFX9-NEXT:    v_perm_b32 v1, v11, v1, s2
 ; GFX9-NEXT:    v_lshrrev_b32_e32 v11, 16, v4
-; GFX9-NEXT:    v_cndmask_b32_e32 v5, v5, v9, vcc
-; GFX9-NEXT:    s_cselect_b64 vcc, -1, 0
-; GFX9-NEXT:    s_cmp_eq_u32 s7, 8
-; GFX9-NEXT:    v_cndmask_b32_e32 v11, v11, v9, vcc
-; GFX9-NEXT:    s_cselect_b64 vcc, -1, 0
 ; GFX9-NEXT:    v_cndmask_b32_e32 v4, v4, v9, vcc
-; GFX9-NEXT:    v_and_b32_e32 v7, 0xffff, v7
-; GFX9-NEXT:    v_and_b32_e32 v6, 0xffff, v6
-; GFX9-NEXT:    v_and_b32_e32 v5, 0xffff, v5
-; GFX9-NEXT:    v_and_b32_e32 v4, 0xffff, v4
-; GFX9-NEXT:    v_lshl_or_b32 v7, v13, 16, v7
-; GFX9-NEXT:    v_lshl_or_b32 v6, v12, 16, v6
-; GFX9-NEXT:    v_lshl_or_b32 v5, v10, 16, v5
-; GFX9-NEXT:    v_lshl_or_b32 v4, v11, 16, v4
+; GFX9-NEXT:    s_cselect_b64 vcc, -1, 0
+; GFX9-NEXT:    v_cndmask_b32_e32 v9, v11, v9, vcc
+; GFX9-NEXT:    v_perm_b32 v6, v12, v6, s2
+; GFX9-NEXT:    v_perm_b32 v5, v10, v5, s2
+; GFX9-NEXT:    v_perm_b32 v4, v9, v4, s2
 ; GFX9-NEXT:    global_store_dwordx4 v8, v[4:7], s[0:1] offset:16
 ; GFX9-NEXT:    global_store_dwordx4 v8, v[0:3], s[0:1]
 ; GFX9-NEXT:    s_endpgm
@@ -3113,80 +3079,72 @@
 ; GFX11-NEXT:    s_clause 0x1
 ; GFX11-NEXT:    global_load_b128 v[0:3], v8, s[6:7]
 ; GFX11-NEXT:    global_load_b128 v[4:7], v8, s[6:7] offset:16
+; GFX11-NEXT:    s_cmp_eq_u32 s1, 6
+; GFX11-NEXT:    s_cselect_b32 s2, -1, 0
 ; GFX11-NEXT:    s_cmp_eq_u32 s1, 7
+; GFX11-NEXT:    s_waitcnt vmcnt(1)
+; GFX11-NEXT:    v_cndmask_b32_e64 v9, v3, s0, s2
 ; GFX11-NEXT:    s_cselect_b32 s2, -1, 0
-; GFX11-NEXT:    s_cmp_eq_u32 s1, 6
+; GFX11-NEXT:    s_cmp_eq_u32 s1, 4
+; GFX11-NEXT:    v_lshrrev_b32_e32 v3, 16, v3
 ; GFX11-NEXT:    s_cselect_b32 s3, -1, 0
 ; GFX11-NEXT:    s_cmp_eq_u32 s1, 5
-; GFX11-NEXT:    s_waitcnt vmcnt(1)
-; GFX11-NEXT:    v_lshrrev_b32_e32 v9, 16, v3
-; GFX11-NEXT:    v_cndmask_b32_e64 v3, v3, s0, s3
+; GFX11-NEXT:    v_lshrrev_b32_e32 v10, 16, v2
+; GFX11-NEXT:    v_cndmask_b32_e64 v2, v2, s0, s3
 ; GFX11-NEXT:    s_cselect_b32 s3, -1, 0
-; GFX11-NEXT:    s_cmp_eq_u32 s1, 4
+; GFX11-NEXT:    s_cmp_eq_u32 s1, 2
 ; GFX11-NEXT:    v_lshrrev_b32_e32 v11, 16, v1
-; GFX11-NEXT:    v_cndmask_b32_e64 v9, v9, s0, s2
+; GFX11-NEXT:    v_cndmask_b32_e64 v3, v3, s0, s2
 ; GFX11-NEXT:    s_cselect_b32 s2, -1, 0
 ; GFX11-NEXT:    s_cmp_eq_u32 s1, 3
-; GFX11-NEXT:    v_lshrrev_b32_e32 v10, 16, v2
-; GFX11-NEXT:    v_cndmask_b32_e64 v2, v2, s0, s2
+; GFX11-NEXT:    v_cndmask_b32_e64 v1, v1, s0, s2
 ; GFX11-NEXT:    s_cselect_b32 s2, -1, 0
-; GFX11-NEXT:    s_cmp_eq_u32 s1, 2
+; GFX11-NEXT:    s_cmp_eq_u32 s1, 0
 ; GFX11-NEXT:    v_lshrrev_b32_e32 v12, 16, v0
-; GFX11-NEXT:    v_and_b32_e32 v3, 0xffff, v3
 ; GFX11-NEXT:    v_cndmask_b32_e64 v11, v11, s0, s2
 ; GFX11-NEXT:    s_cselect_b32 s2, -1, 0
 ; GFX11-NEXT:    s_cmp_eq_u32 s1, 1
-; GFX11-NEXT:    v_cndmask_b32_e64 v1, v1, s0, s2
+; GFX11-NEXT:    v_cndmask_b32_e64 v0, v0, s0, s2
 ; GFX11-NEXT:    s_cselect_b32 s2, -1, 0
-; GFX11-NEXT:    s_cmp_eq_u32 s1, 0
+; GFX11-NEXT:    s_cmp_eq_u32 s1, 14
 ; GFX11-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-NEXT:    v_lshrrev_b32_e32 v13, 16, v7
 ; GFX11-NEXT:    v_cndmask_b32_e64 v10, v10, s0, s3
-; GFX11-NEXT:    v_and_b32_e32 v2, 0xffff, v2
-; GFX11-NEXT:    v_lshl_or_b32 v3, v9, 16, v3
+; GFX11-NEXT:    v_perm_b32 v3, v3, v9, 0x5040100
 ; GFX11-NEXT:    v_cndmask_b32_e64 v9, v12, s0, s2
 ; GFX11-NEXT:    s_cselect_b32 s2, -1, 0
 ; GFX11-NEXT:    s_cmp_eq_u32 s1, 15
-; GFX11-NEXT:    v_cndmask_b32_e64 v0, v0, s0, s2
+; GFX11-NEXT:    v_cndmask_b32_e64 v7, v7, s0, s2
 ; GFX11-NEXT:    s_cselect_b32 s2, -1, 0
-; GFX11-NEXT:    s_cmp_eq_u32 s1, 14
-; GFX11-NEXT:    v_lshl_or_b32 v2, v10, 16, v2
+; GFX11-NEXT:    s_cmp_eq_u32 s1, 12
+; GFX11-NEXT:    v_lshrrev_b32_e32 v14, 16, v6
+; GFX11-NEXT:    v_perm_b32 v2, v10, v2, 0x5040100
 ; GFX11-NEXT:    v_cndmask_b32_e64 v10, v13, s0, s2
 ; GFX11-NEXT:    s_cselect_b32 s2, -1, 0
 ; GFX11-NEXT:    s_cmp_eq_u32 s1, 13
-; GFX11-NEXT:    v_lshrrev_b32_e32 v14, 16, v6
-; GFX11-NEXT:    v_cndmask_b32_e64 v7, v7, s0, s2
+; GFX11-NEXT:    v_cndmask_b32_e64 v6, v6, s0, s2
 ; GFX11-NEXT:    s_cselect_b32 s2, -1, 0
-; GFX11-NEXT:    s_cmp_eq_u32 s1, 12
+; GFX11-NEXT:    s_cmp_eq_u32 s1, 10
 ; GFX11-NEXT:    v_lshrrev_b32_e32 v15, 16, v5
-; GFX11-NEXT:    s_cselect_b32 s3, -1, 0
-; GFX11-NEXT:    s_cmp_eq_u32 s1, 11
 ; GFX11-NEXT:    v_cndmask_b32_e64 v12, v14, s0, s2
 ; GFX11-NEXT:    s_cselect_b32 s2, -1, 0
-; GFX11-NEXT:    s_cmp_eq_u32 s1, 10
-; GFX11-NEXT:    v_cndmask_b32_e64 v6, v6, s0, s3
-; GFX11-NEXT:    s_cselect_b32 s3, -1, 0
-; GFX11-NEXT:    s_cmp_eq_u32 s1, 9
-; GFX11-NEXT:    v_lshrrev_b32_e32 v16, 16, v4
-; GFX11-NEXT:    s_cselect_b32 s6, -1, 0
+; GFX11-NEXT:    s_cmp_eq_u32 s1, 11
+; GFX11-NEXT:    v_cndmask_b32_e64 v5, v5, s0, s2
+; GFX11-NEXT:    s_cselect_b32 s2, -1, 0
 ; GFX11-NEXT:    s_cmp_eq_u32 s1, 8
-; GFX11-NEXT:    v_cndmask_b32_e64 v5, v5, s0, s3
-; GFX11-NEXT:    s_cselect_b32 s1, -1, 0
-; GFX11-NEXT:    v_and_b32_e32 v7, 0xffff, v7
-; GFX11-NEXT:    v_cndmask_b32_e64 v4, v4, s0, s1
+; GFX11-NEXT:    v_lshrrev_b32_e32 v16, 16, v4
 ; GFX11-NEXT:    v_cndmask_b32_e64 v13, v15, s0, s2
-; GFX11-NEXT:    v_and_b32_e32 v6, 0xffff, v6
-; GFX11-NEXT:    v_cndmask_b32_e64 v14, v16, s0, s6
-; GFX11-NEXT:    v_and_b32_e32 v5, 0xffff, v5
-; GFX11-NEXT:    v_and_b32_e32 v4, 0xffff, v4
-; GFX11-NEXT:    v_and_b32_e32 v1, 0xffff, v1
-; GFX11-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX11-NEXT:    v_lshl_or_b32 v7, v10, 16, v7
-; GFX11-NEXT:    v_lshl_or_b32 v6, v12, 16, v6
-; GFX11-NEXT:    v_lshl_or_b32 v5, v13, 16, v5
-; GFX11-NEXT:    v_lshl_or_b32 v4, v14, 16, v4
-; GFX11-NEXT:    v_lshl_or_b32 v1, v11, 16, v1
-; GFX11-NEXT:    v_lshl_or_b32 v0, v9, 16, v0
+; GFX11-NEXT:    s_cselect_b32 s2, -1, 0
+; GFX11-NEXT:    s_cmp_eq_u32 s1, 9
+; GFX11-NEXT:    v_cndmask_b32_e64 v4, v4, s0, s2
+; GFX11-NEXT:    s_cselect_b32 s1, -1, 0
+; GFX11-NEXT:    v_perm_b32 v7, v10, v7, 0x5040100
+; GFX11-NEXT:    v_cndmask_b32_e64 v14, v16, s0, s1
+; GFX11-NEXT:    v_perm_b32 v6, v12, v6, 0x5040100
+; GFX11-NEXT:    v_perm_b32 v5, v13, v5, 0x5040100
+; GFX11-NEXT:    v_perm_b32 v1, v11, v1, 0x5040100
+; GFX11-NEXT:    v_perm_b32 v0, v9, v0, 0x5040100
+; GFX11-NEXT:    v_perm_b32 v4, v14, v4, 0x5040100
 ; GFX11-NEXT:    s_clause 0x1
 ; GFX11-NEXT:    global_store_b128 v8, v[4:7], s[4:5] offset:16
 ; GFX11-NEXT:    global_store_b128 v8, v[0:3], s[4:5]
diff --git a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.image.gather4.a16.dim.ll b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.image.gather4.a16.dim.ll
--- a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.image.gather4.a16.dim.ll
+++ b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.image.gather4.a16.dim.ll
@@ -8,8 +8,8 @@
 ; GFX9:       ; %bb.0: ; %main_body
 ; GFX9-NEXT:    s_mov_b64 s[12:13], exec
 ; GFX9-NEXT:    s_wqm_b64 exec, exec
-; GFX9-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX9-NEXT:    v_lshl_or_b32 v0, v1, 16, v0
+; GFX9-NEXT:    s_mov_b32 s14, 0x5040100
+; GFX9-NEXT:    v_perm_b32 v0, v1, v0, s14
 ; GFX9-NEXT:    s_and_b64 exec, exec, s[12:13]
 ; GFX9-NEXT:    image_gather4 v[0:3], v0, s[0:7], s[8:11] dmask:0x1 a16
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
@@ -19,8 +19,7 @@
 ; GFX10:       ; %bb.0: ; %main_body
 ; GFX10-NEXT:    s_mov_b32 s12, exec_lo
 ; GFX10-NEXT:    s_wqm_b32 exec_lo, exec_lo
-; GFX10-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX10-NEXT:    v_lshl_or_b32 v0, v1, 16, v0
+; GFX10-NEXT:    v_perm_b32 v0, v1, v0, 0x5040100
 ; GFX10-NEXT:    s_and_b32 exec_lo, exec_lo, s12
 ; GFX10-NEXT:    image_gather4 v[0:3], v0, s[0:7], s[8:11] dmask:0x1 dim:SQ_RSRC_IMG_2D a16
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
@@ -35,8 +34,8 @@
 ; GFX9:       ; %bb.0: ; %main_body
 ; GFX9-NEXT:    s_mov_b64 s[12:13], exec
 ; GFX9-NEXT:    s_wqm_b64 exec, exec
-; GFX9-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX9-NEXT:    v_lshl_or_b32 v1, v1, 16, v0
+; GFX9-NEXT:    s_mov_b32 s14, 0x5040100
+; GFX9-NEXT:    v_perm_b32 v1, v1, v0, s14
 ; GFX9-NEXT:    s_and_b64 exec, exec, s[12:13]
 ; GFX9-NEXT:    image_gather4 v[0:3], v[1:2], s[0:7], s[8:11] dmask:0x1 a16 da
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
@@ -46,8 +45,7 @@
 ; GFX10:       ; %bb.0: ; %main_body
 ; GFX10-NEXT:    s_mov_b32 s12, exec_lo
 ; GFX10-NEXT:    s_wqm_b32 exec_lo, exec_lo
-; GFX10-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX10-NEXT:    v_lshl_or_b32 v1, v1, 16, v0
+; GFX10-NEXT:    v_perm_b32 v1, v1, v0, 0x5040100
 ; GFX10-NEXT:    s_and_b32 exec_lo, exec_lo, s12
 ; GFX10-NEXT:    image_gather4 v[0:3], v[1:2], s[0:7], s[8:11] dmask:0x1 dim:SQ_RSRC_IMG_CUBE a16
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
@@ -62,8 +60,8 @@
 ; GFX9:       ; %bb.0: ; %main_body
 ; GFX9-NEXT:    s_mov_b64 s[12:13], exec
 ; GFX9-NEXT:    s_wqm_b64 exec, exec
-; GFX9-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX9-NEXT:    v_lshl_or_b32 v1, v1, 16, v0
+; GFX9-NEXT:    s_mov_b32 s14, 0x5040100
+; GFX9-NEXT:    v_perm_b32 v1, v1, v0, s14
 ; GFX9-NEXT:    s_and_b64 exec, exec, s[12:13]
 ; GFX9-NEXT:    image_gather4 v[0:3], v[1:2], s[0:7], s[8:11] dmask:0x1 a16 da
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
@@ -73,8 +71,7 @@
 ; GFX10:       ; %bb.0: ; %main_body
 ; GFX10-NEXT:    s_mov_b32 s12, exec_lo
 ; GFX10-NEXT:    s_wqm_b32 exec_lo, exec_lo
-; GFX10-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX10-NEXT:    v_lshl_or_b32 v1, v1, 16, v0
+; GFX10-NEXT:    v_perm_b32 v1, v1, v0, 0x5040100
 ; GFX10-NEXT:    s_and_b32 exec_lo, exec_lo, s12
 ; GFX10-NEXT:    image_gather4 v[0:3], v[1:2], s[0:7], s[8:11] dmask:0x1 dim:SQ_RSRC_IMG_2D_ARRAY a16
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
@@ -89,8 +86,8 @@
 ; GFX9:       ; %bb.0: ; %main_body
 ; GFX9-NEXT:    s_mov_b64 s[12:13], exec
 ; GFX9-NEXT:    s_wqm_b64 exec, exec
-; GFX9-NEXT:    v_and_b32_e32 v1, 0xffff, v1
-; GFX9-NEXT:    v_lshl_or_b32 v1, v2, 16, v1
+; GFX9-NEXT:    s_mov_b32 s14, 0x5040100
+; GFX9-NEXT:    v_perm_b32 v1, v2, v1, s14
 ; GFX9-NEXT:    s_and_b64 exec, exec, s[12:13]
 ; GFX9-NEXT:    image_gather4_c v[0:3], v[0:1], s[0:7], s[8:11] dmask:0x1 a16
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
@@ -100,8 +97,7 @@
 ; GFX10:       ; %bb.0: ; %main_body
 ; GFX10-NEXT:    s_mov_b32 s12, exec_lo
 ; GFX10-NEXT:    s_wqm_b32 exec_lo, exec_lo
-; GFX10-NEXT:    v_and_b32_e32 v1, 0xffff, v1
-; GFX10-NEXT:    v_lshl_or_b32 v1, v2, 16, v1
+; GFX10-NEXT:    v_perm_b32 v1, v2, v1, 0x5040100
 ; GFX10-NEXT:    s_and_b32 exec_lo, exec_lo, s12
 ; GFX10-NEXT:    image_gather4_c v[0:3], v[0:1], s[0:7], s[8:11] dmask:0x1 dim:SQ_RSRC_IMG_2D a16
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
@@ -116,8 +112,8 @@
 ; GFX9:       ; %bb.0: ; %main_body
 ; GFX9-NEXT:    s_mov_b64 s[12:13], exec
 ; GFX9-NEXT:    s_wqm_b64 exec, exec
-; GFX9-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX9-NEXT:    v_lshl_or_b32 v1, v1, 16, v0
+; GFX9-NEXT:    s_mov_b32 s14, 0x5040100
+; GFX9-NEXT:    v_perm_b32 v1, v1, v0, s14
 ; GFX9-NEXT:    s_and_b64 exec, exec, s[12:13]
 ; GFX9-NEXT:    image_gather4_cl v[0:3], v[1:2], s[0:7], s[8:11] dmask:0x1 a16
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
@@ -127,8 +123,7 @@
 ; GFX10:       ; %bb.0: ; %main_body
 ; GFX10-NEXT:    s_mov_b32 s12, exec_lo
 ; GFX10-NEXT:    s_wqm_b32 exec_lo, exec_lo
-; GFX10-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX10-NEXT:    v_lshl_or_b32 v1, v1, 16, v0
+; GFX10-NEXT:    v_perm_b32 v1, v1, v0, 0x5040100
 ; GFX10-NEXT:    s_and_b32 exec_lo, exec_lo, s12
 ; GFX10-NEXT:    image_gather4_cl v[0:3], v[1:2], s[0:7], s[8:11] dmask:0x1 dim:SQ_RSRC_IMG_2D a16
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
@@ -143,10 +138,10 @@
 ; GFX9:       ; %bb.0: ; %main_body
 ; GFX9-NEXT:    s_mov_b64 s[12:13], exec
 ; GFX9-NEXT:    s_wqm_b64 exec, exec
+; GFX9-NEXT:    s_mov_b32 s14, 0x5040100
 ; GFX9-NEXT:    v_mov_b32_e32 v5, v3
 ; GFX9-NEXT:    v_mov_b32_e32 v3, v0
-; GFX9-NEXT:    v_and_b32_e32 v0, 0xffff, v1
-; GFX9-NEXT:    v_lshl_or_b32 v4, v2, 16, v0
+; GFX9-NEXT:    v_perm_b32 v4, v2, v1, s14
 ; GFX9-NEXT:    s_and_b64 exec, exec, s[12:13]
 ; GFX9-NEXT:    image_gather4_c_cl v[0:3], v[3:5], s[0:7], s[8:11] dmask:0x1 a16
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
@@ -156,8 +151,7 @@
 ; GFX10:       ; %bb.0: ; %main_body
 ; GFX10-NEXT:    s_mov_b32 s12, exec_lo
 ; GFX10-NEXT:    s_wqm_b32 exec_lo, exec_lo
-; GFX10-NEXT:    v_and_b32_e32 v1, 0xffff, v1
-; GFX10-NEXT:    v_lshl_or_b32 v1, v2, 16, v1
+; GFX10-NEXT:    v_perm_b32 v1, v2, v1, 0x5040100
 ; GFX10-NEXT:    s_and_b32 exec_lo, exec_lo, s12
 ; GFX10-NEXT:    image_gather4_c_cl v[0:3], [v0, v1, v3], s[0:7], s[8:11] dmask:0x1 dim:SQ_RSRC_IMG_2D a16
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
@@ -172,8 +166,8 @@
 ; GFX9:       ; %bb.0: ; %main_body
 ; GFX9-NEXT:    s_mov_b64 s[12:13], exec
 ; GFX9-NEXT:    s_wqm_b64 exec, exec
-; GFX9-NEXT:    v_and_b32_e32 v1, 0xffff, v1
-; GFX9-NEXT:    v_lshl_or_b32 v1, v2, 16, v1
+; GFX9-NEXT:    s_mov_b32 s14, 0x5040100
+; GFX9-NEXT:    v_perm_b32 v1, v2, v1, s14
 ; GFX9-NEXT:    s_and_b64 exec, exec, s[12:13]
 ; GFX9-NEXT:    image_gather4_b v[0:3], v[0:1], s[0:7], s[8:11] dmask:0x1 a16
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
@@ -183,8 +177,7 @@
 ; GFX10:       ; %bb.0: ; %main_body
 ; GFX10-NEXT:    s_mov_b32 s12, exec_lo
 ; GFX10-NEXT:    s_wqm_b32 exec_lo, exec_lo
-; GFX10-NEXT:    v_and_b32_e32 v1, 0xffff, v1
-; GFX10-NEXT:    v_lshl_or_b32 v1, v2, 16, v1
+; GFX10-NEXT:    v_perm_b32 v1, v2, v1, 0x5040100
 ; GFX10-NEXT:    s_and_b32 exec_lo, exec_lo, s12
 ; GFX10-NEXT:    image_gather4_b v[0:3], v[0:1], s[0:7], s[8:11] dmask:0x1 dim:SQ_RSRC_IMG_2D a16
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
@@ -199,8 +192,8 @@
 ; GFX9:       ; %bb.0: ; %main_body
 ; GFX9-NEXT:    s_mov_b64 s[12:13], exec
 ; GFX9-NEXT:    s_wqm_b64 exec, exec
-; GFX9-NEXT:    v_and_b32_e32 v2, 0xffff, v2
-; GFX9-NEXT:    v_lshl_or_b32 v2, v3, 16, v2
+; GFX9-NEXT:    s_mov_b32 s14, 0x5040100
+; GFX9-NEXT:    v_perm_b32 v2, v3, v2, s14
 ; GFX9-NEXT:    s_and_b64 exec, exec, s[12:13]
 ; GFX9-NEXT:    image_gather4_c_b v[0:3], v[0:2], s[0:7], s[8:11] dmask:0x1 a16
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
@@ -210,8 +203,7 @@
 ; GFX10:       ; %bb.0: ; %main_body
 ; GFX10-NEXT:    s_mov_b32 s12, exec_lo
 ; GFX10-NEXT:    s_wqm_b32 exec_lo, exec_lo
-; GFX10-NEXT:    v_and_b32_e32 v2, 0xffff, v2
-; GFX10-NEXT:    v_lshl_or_b32 v2, v3, 16, v2
+; GFX10-NEXT:    v_perm_b32 v2, v3, v2, 0x5040100
 ; GFX10-NEXT:    s_and_b32 exec_lo, exec_lo, s12
 ; GFX10-NEXT:    image_gather4_c_b v[0:3], v[0:2], s[0:7], s[8:11] dmask:0x1 dim:SQ_RSRC_IMG_2D a16
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
@@ -226,10 +218,10 @@
 ; GFX9:       ; %bb.0: ; %main_body
 ; GFX9-NEXT:    s_mov_b64 s[12:13], exec
 ; GFX9-NEXT:    s_wqm_b64 exec, exec
+; GFX9-NEXT:    s_mov_b32 s14, 0x5040100
 ; GFX9-NEXT:    v_mov_b32_e32 v5, v3
 ; GFX9-NEXT:    v_mov_b32_e32 v3, v0
-; GFX9-NEXT:    v_and_b32_e32 v0, 0xffff, v1
-; GFX9-NEXT:    v_lshl_or_b32 v4, v2, 16, v0
+; GFX9-NEXT:    v_perm_b32 v4, v2, v1, s14
 ; GFX9-NEXT:    s_and_b64 exec, exec, s[12:13]
 ; GFX9-NEXT:    image_gather4_b_cl v[0:3], v[3:5], s[0:7], s[8:11] dmask:0x1 a16
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
@@ -239,8 +231,7 @@
 ; GFX10:       ; %bb.0: ; %main_body
 ; GFX10-NEXT:    s_mov_b32 s12, exec_lo
 ; GFX10-NEXT:    s_wqm_b32 exec_lo, exec_lo
-; GFX10-NEXT:    v_and_b32_e32 v1, 0xffff, v1
-; GFX10-NEXT:    v_lshl_or_b32 v1, v2, 16, v1
+; GFX10-NEXT:    v_perm_b32 v1, v2, v1, 0x5040100
 ; GFX10-NEXT:    s_and_b32 exec_lo, exec_lo, s12
 ; GFX10-NEXT:    image_gather4_b_cl v[0:3], [v0, v1, v3], s[0:7], s[8:11] dmask:0x1 dim:SQ_RSRC_IMG_2D a16
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
@@ -255,11 +246,11 @@
 ; GFX9:       ; %bb.0: ; %main_body
 ; GFX9-NEXT:    s_mov_b64 s[12:13], exec
 ; GFX9-NEXT:    s_wqm_b64 exec, exec
+; GFX9-NEXT:    s_mov_b32 s14, 0x5040100
 ; GFX9-NEXT:    v_mov_b32_e32 v7, v4
-; GFX9-NEXT:    v_mov_b32_e32 v4, v0
-; GFX9-NEXT:    v_and_b32_e32 v0, 0xffff, v2
 ; GFX9-NEXT:    v_mov_b32_e32 v5, v1
-; GFX9-NEXT:    v_lshl_or_b32 v6, v3, 16, v0
+; GFX9-NEXT:    v_mov_b32_e32 v4, v0
+; GFX9-NEXT:    v_perm_b32 v6, v3, v2, s14
 ; GFX9-NEXT:    s_and_b64 exec, exec, s[12:13]
 ; GFX9-NEXT:    image_gather4_c_b_cl v[0:3], v[4:7], s[0:7], s[8:11] dmask:0x1 a16
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
@@ -269,8 +260,7 @@
 ; GFX10:       ; %bb.0: ; %main_body
 ; GFX10-NEXT:    s_mov_b32 s12, exec_lo
 ; GFX10-NEXT:    s_wqm_b32 exec_lo, exec_lo
-; GFX10-NEXT:    v_and_b32_e32 v2, 0xffff, v2
-; GFX10-NEXT:    v_lshl_or_b32 v2, v3, 16, v2
+; GFX10-NEXT:    v_perm_b32 v2, v3, v2, 0x5040100
 ; GFX10-NEXT:    s_and_b32 exec_lo, exec_lo, s12
 ; GFX10-NEXT:    image_gather4_c_b_cl v[0:3], [v0, v1, v2, v4], s[0:7], s[8:11] dmask:0x1 dim:SQ_RSRC_IMG_2D a16
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
@@ -283,16 +273,15 @@
 define amdgpu_ps <4 x float> @gather4_l_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %s, half %t, half %lod) {
 ; GFX9-LABEL: gather4_l_2d:
 ; GFX9:       ; %bb.0: ; %main_body
-; GFX9-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX9-NEXT:    v_lshl_or_b32 v1, v1, 16, v0
+; GFX9-NEXT:    s_mov_b32 s12, 0x5040100
+; GFX9-NEXT:    v_perm_b32 v1, v1, v0, s12
 ; GFX9-NEXT:    image_gather4_l v[0:3], v[1:2], s[0:7], s[8:11] dmask:0x1 a16
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    ; return to shader part epilog
 ;
 ; GFX10-LABEL: gather4_l_2d:
 ; GFX10:       ; %bb.0: ; %main_body
-; GFX10-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX10-NEXT:    v_lshl_or_b32 v1, v1, 16, v0
+; GFX10-NEXT:    v_perm_b32 v1, v1, v0, 0x5040100
 ; GFX10-NEXT:    image_gather4_l v[0:3], v[1:2], s[0:7], s[8:11] dmask:0x1 dim:SQ_RSRC_IMG_2D a16
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    ; return to shader part epilog
@@ -304,18 +293,17 @@
 define amdgpu_ps <4 x float> @gather4_c_l_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %zcompare, half %s, half %t, half %lod) {
 ; GFX9-LABEL: gather4_c_l_2d:
 ; GFX9:       ; %bb.0: ; %main_body
+; GFX9-NEXT:    s_mov_b32 s12, 0x5040100
 ; GFX9-NEXT:    v_mov_b32_e32 v5, v3
 ; GFX9-NEXT:    v_mov_b32_e32 v3, v0
-; GFX9-NEXT:    v_and_b32_e32 v0, 0xffff, v1
-; GFX9-NEXT:    v_lshl_or_b32 v4, v2, 16, v0
+; GFX9-NEXT:    v_perm_b32 v4, v2, v1, s12
 ; GFX9-NEXT:    image_gather4_c_l v[0:3], v[3:5], s[0:7], s[8:11] dmask:0x1 a16
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    ; return to shader part epilog
 ;
 ; GFX10-LABEL: gather4_c_l_2d:
 ; GFX10:       ; %bb.0: ; %main_body
-; GFX10-NEXT:    v_and_b32_e32 v1, 0xffff, v1
-; GFX10-NEXT:    v_lshl_or_b32 v1, v2, 16, v1
+; GFX10-NEXT:    v_perm_b32 v1, v2, v1, 0x5040100
 ; GFX10-NEXT:    image_gather4_c_l v[0:3], [v0, v1, v3], s[0:7], s[8:11] dmask:0x1 dim:SQ_RSRC_IMG_2D a16
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    ; return to shader part epilog
@@ -327,16 +315,15 @@
 define amdgpu_ps <4 x float> @gather4_lz_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %s, half %t) {
 ; GFX9-LABEL: gather4_lz_2d:
 ; GFX9:       ; %bb.0: ; %main_body
-; GFX9-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX9-NEXT:    v_lshl_or_b32 v0, v1, 16, v0
+; GFX9-NEXT:    s_mov_b32 s12, 0x5040100
+; GFX9-NEXT:    v_perm_b32 v0, v1, v0, s12
 ; GFX9-NEXT:    image_gather4_lz v[0:3], v0, s[0:7], s[8:11] dmask:0x1 a16
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    ; return to shader part epilog
 ;
 ; GFX10-LABEL: gather4_lz_2d:
 ; GFX10:       ; %bb.0: ; %main_body
-; GFX10-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX10-NEXT:    v_lshl_or_b32 v0, v1, 16, v0
+; GFX10-NEXT:    v_perm_b32 v0, v1, v0, 0x5040100
 ; GFX10-NEXT:    image_gather4_lz v[0:3], v0, s[0:7], s[8:11] dmask:0x1 dim:SQ_RSRC_IMG_2D a16
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    ; return to shader part epilog
@@ -348,16 +335,15 @@
 define amdgpu_ps <4 x float> @gather4_c_lz_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %zcompare, half %s, half %t) {
 ; GFX9-LABEL: gather4_c_lz_2d:
 ; GFX9:       ; %bb.0: ; %main_body
-; GFX9-NEXT:    v_and_b32_e32 v1, 0xffff, v1
-; GFX9-NEXT:    v_lshl_or_b32 v1, v2, 16, v1
+; GFX9-NEXT:    s_mov_b32 s12, 0x5040100
+; GFX9-NEXT:    v_perm_b32 v1, v2, v1, s12
 ; GFX9-NEXT:    image_gather4_c_lz v[0:3], v[0:1], s[0:7], s[8:11] dmask:0x1 a16
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    ; return to shader part epilog
 ;
 ; GFX10-LABEL: gather4_c_lz_2d:
 ; GFX10:       ; %bb.0: ; %main_body
-; GFX10-NEXT:    v_and_b32_e32 v1, 0xffff, v1
-; GFX10-NEXT:    v_lshl_or_b32 v1, v2, 16, v1
+; GFX10-NEXT:    v_perm_b32 v1, v2, v1, 0x5040100
 ; GFX10-NEXT:    image_gather4_c_lz v[0:3], v[0:1], s[0:7], s[8:11] dmask:0x1 dim:SQ_RSRC_IMG_2D a16
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    ; return to shader part epilog
diff --git a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.image.sample.a16.dim.ll b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.image.sample.a16.dim.ll
--- a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.image.sample.a16.dim.ll
+++ b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.image.sample.a16.dim.ll
@@ -31,8 +31,8 @@
 ; GFX9:       ; %bb.0: ; %main_body
 ; GFX9-NEXT:    s_mov_b64 s[12:13], exec
 ; GFX9-NEXT:    s_wqm_b64 exec, exec
-; GFX9-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX9-NEXT:    v_lshl_or_b32 v0, v1, 16, v0
+; GFX9-NEXT:    s_mov_b32 s14, 0x5040100
+; GFX9-NEXT:    v_perm_b32 v0, v1, v0, s14
 ; GFX9-NEXT:    s_and_b64 exec, exec, s[12:13]
 ; GFX9-NEXT:    image_sample v[0:3], v0, s[0:7], s[8:11] dmask:0xf a16
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
@@ -42,8 +42,7 @@
 ; GFX10:       ; %bb.0: ; %main_body
 ; GFX10-NEXT:    s_mov_b32 s12, exec_lo
 ; GFX10-NEXT:    s_wqm_b32 exec_lo, exec_lo
-; GFX10-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX10-NEXT:    v_lshl_or_b32 v0, v1, 16, v0
+; GFX10-NEXT:    v_perm_b32 v0, v1, v0, 0x5040100
 ; GFX10-NEXT:    s_and_b32 exec_lo, exec_lo, s12
 ; GFX10-NEXT:    image_sample v[0:3], v0, s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_2D a16
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
@@ -58,8 +57,8 @@
 ; GFX9:       ; %bb.0: ; %main_body
 ; GFX9-NEXT:    s_mov_b64 s[12:13], exec
 ; GFX9-NEXT:    s_wqm_b64 exec, exec
-; GFX9-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX9-NEXT:    v_lshl_or_b32 v1, v1, 16, v0
+; GFX9-NEXT:    s_mov_b32 s14, 0x5040100
+; GFX9-NEXT:    v_perm_b32 v1, v1, v0, s14
 ; GFX9-NEXT:    s_and_b64 exec, exec, s[12:13]
 ; GFX9-NEXT:    image_sample v[0:3], v[1:2], s[0:7], s[8:11] dmask:0xf a16
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
@@ -69,8 +68,7 @@
 ; GFX10:       ; %bb.0: ; %main_body
 ; GFX10-NEXT:    s_mov_b32 s12, exec_lo
 ; GFX10-NEXT:    s_wqm_b32 exec_lo, exec_lo
-; GFX10-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX10-NEXT:    v_lshl_or_b32 v1, v1, 16, v0
+; GFX10-NEXT:    v_perm_b32 v1, v1, v0, 0x5040100
 ; GFX10-NEXT:    s_and_b32 exec_lo, exec_lo, s12
 ; GFX10-NEXT:    image_sample v[0:3], v[1:2], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_3D a16
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
@@ -85,8 +83,8 @@
 ; GFX9:       ; %bb.0: ; %main_body
 ; GFX9-NEXT:    s_mov_b64 s[12:13], exec
 ; GFX9-NEXT:    s_wqm_b64 exec, exec
-; GFX9-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX9-NEXT:    v_lshl_or_b32 v1, v1, 16, v0
+; GFX9-NEXT:    s_mov_b32 s14, 0x5040100
+; GFX9-NEXT:    v_perm_b32 v1, v1, v0, s14
 ; GFX9-NEXT:    s_and_b64 exec, exec, s[12:13]
 ; GFX9-NEXT:    image_sample v[0:3], v[1:2], s[0:7], s[8:11] dmask:0xf a16 da
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
@@ -96,8 +94,7 @@
 ; GFX10:       ; %bb.0: ; %main_body
 ; GFX10-NEXT:    s_mov_b32 s12, exec_lo
 ; GFX10-NEXT:    s_wqm_b32 exec_lo, exec_lo
-; GFX10-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX10-NEXT:    v_lshl_or_b32 v1, v1, 16, v0
+; GFX10-NEXT:    v_perm_b32 v1, v1, v0, 0x5040100
 ; GFX10-NEXT:    s_and_b32 exec_lo, exec_lo, s12
 ; GFX10-NEXT:    image_sample v[0:3], v[1:2], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_CUBE a16
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
@@ -112,8 +109,8 @@
 ; GFX9:       ; %bb.0: ; %main_body
 ; GFX9-NEXT:    s_mov_b64 s[12:13], exec
 ; GFX9-NEXT:    s_wqm_b64 exec, exec
-; GFX9-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX9-NEXT:    v_lshl_or_b32 v0, v1, 16, v0
+; GFX9-NEXT:    s_mov_b32 s14, 0x5040100
+; GFX9-NEXT:    v_perm_b32 v0, v1, v0, s14
 ; GFX9-NEXT:    s_and_b64 exec, exec, s[12:13]
 ; GFX9-NEXT:    image_sample v[0:3], v0, s[0:7], s[8:11] dmask:0xf a16 da
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
@@ -123,8 +120,7 @@
 ; GFX10:       ; %bb.0: ; %main_body
 ; GFX10-NEXT:    s_mov_b32 s12, exec_lo
 ; GFX10-NEXT:    s_wqm_b32 exec_lo, exec_lo
-; GFX10-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX10-NEXT:    v_lshl_or_b32 v0, v1, 16, v0
+; GFX10-NEXT:    v_perm_b32 v0, v1, v0, 0x5040100
 ; GFX10-NEXT:    s_and_b32 exec_lo, exec_lo, s12
 ; GFX10-NEXT:    image_sample v[0:3], v0, s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_1D_ARRAY a16
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
@@ -139,8 +135,8 @@
 ; GFX9:       ; %bb.0: ; %main_body
 ; GFX9-NEXT:    s_mov_b64 s[12:13], exec
 ; GFX9-NEXT:    s_wqm_b64 exec, exec
-; GFX9-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX9-NEXT:    v_lshl_or_b32 v1, v1, 16, v0
+; GFX9-NEXT:    s_mov_b32 s14, 0x5040100
+; GFX9-NEXT:    v_perm_b32 v1, v1, v0, s14
 ; GFX9-NEXT:    s_and_b64 exec, exec, s[12:13]
 ; GFX9-NEXT:    image_sample v[0:3], v[1:2], s[0:7], s[8:11] dmask:0xf a16 da
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
@@ -150,8 +146,7 @@
 ; GFX10:       ; %bb.0: ; %main_body
 ; GFX10-NEXT:    s_mov_b32 s12, exec_lo
 ; GFX10-NEXT:    s_wqm_b32 exec_lo, exec_lo
-; GFX10-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX10-NEXT:    v_lshl_or_b32 v1, v1, 16, v0
+; GFX10-NEXT:    v_perm_b32 v1, v1, v0, 0x5040100
 ; GFX10-NEXT:    s_and_b32 exec_lo, exec_lo, s12
 ; GFX10-NEXT:    image_sample v[0:3], v[1:2], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_2D_ARRAY a16
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
@@ -189,8 +184,8 @@
 ; GFX9:       ; %bb.0: ; %main_body
 ; GFX9-NEXT:    s_mov_b64 s[12:13], exec
 ; GFX9-NEXT:    s_wqm_b64 exec, exec
-; GFX9-NEXT:    v_and_b32_e32 v1, 0xffff, v1
-; GFX9-NEXT:    v_lshl_or_b32 v1, v2, 16, v1
+; GFX9-NEXT:    s_mov_b32 s14, 0x5040100
+; GFX9-NEXT:    v_perm_b32 v1, v2, v1, s14
 ; GFX9-NEXT:    s_and_b64 exec, exec, s[12:13]
 ; GFX9-NEXT:    image_sample_c v[0:3], v[0:1], s[0:7], s[8:11] dmask:0xf a16
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
@@ -200,8 +195,7 @@
 ; GFX10:       ; %bb.0: ; %main_body
 ; GFX10-NEXT:    s_mov_b32 s12, exec_lo
 ; GFX10-NEXT:    s_wqm_b32 exec_lo, exec_lo
-; GFX10-NEXT:    v_and_b32_e32 v1, 0xffff, v1
-; GFX10-NEXT:    v_lshl_or_b32 v1, v2, 16, v1
+; GFX10-NEXT:    v_perm_b32 v1, v2, v1, 0x5040100
 ; GFX10-NEXT:    s_and_b32 exec_lo, exec_lo, s12
 ; GFX10-NEXT:    image_sample_c v[0:3], v[0:1], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_2D a16
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
@@ -216,8 +210,8 @@
 ; GFX9:       ; %bb.0: ; %main_body
 ; GFX9-NEXT:    s_mov_b64 s[12:13], exec
 ; GFX9-NEXT:    s_wqm_b64 exec, exec
-; GFX9-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX9-NEXT:    v_lshl_or_b32 v0, v1, 16, v0
+; GFX9-NEXT:    s_mov_b32 s14, 0x5040100
+; GFX9-NEXT:    v_perm_b32 v0, v1, v0, s14
 ; GFX9-NEXT:    s_and_b64 exec, exec, s[12:13]
 ; GFX9-NEXT:    image_sample_cl v[0:3], v0, s[0:7], s[8:11] dmask:0xf a16
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
@@ -227,8 +221,7 @@
 ; GFX10:       ; %bb.0: ; %main_body
 ; GFX10-NEXT:    s_mov_b32 s12, exec_lo
 ; GFX10-NEXT:    s_wqm_b32 exec_lo, exec_lo
-; GFX10-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX10-NEXT:    v_lshl_or_b32 v0, v1, 16, v0
+; GFX10-NEXT:    v_perm_b32 v0, v1, v0, 0x5040100
 ; GFX10-NEXT:    s_and_b32 exec_lo, exec_lo, s12
 ; GFX10-NEXT:    image_sample_cl v[0:3], v0, s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_1D a16
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
@@ -243,8 +236,8 @@
 ; GFX9:       ; %bb.0: ; %main_body
 ; GFX9-NEXT:    s_mov_b64 s[12:13], exec
 ; GFX9-NEXT:    s_wqm_b64 exec, exec
-; GFX9-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX9-NEXT:    v_lshl_or_b32 v1, v1, 16, v0
+; GFX9-NEXT:    s_mov_b32 s14, 0x5040100
+; GFX9-NEXT:    v_perm_b32 v1, v1, v0, s14
 ; GFX9-NEXT:    s_and_b64 exec, exec, s[12:13]
 ; GFX9-NEXT:    image_sample_cl v[0:3], v[1:2], s[0:7], s[8:11] dmask:0xf a16
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
@@ -254,8 +247,7 @@
 ; GFX10:       ; %bb.0: ; %main_body
 ; GFX10-NEXT:    s_mov_b32 s12, exec_lo
 ; GFX10-NEXT:    s_wqm_b32 exec_lo, exec_lo
-; GFX10-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX10-NEXT:    v_lshl_or_b32 v1, v1, 16, v0
+; GFX10-NEXT:    v_perm_b32 v1, v1, v0, 0x5040100
 ; GFX10-NEXT:    s_and_b32 exec_lo, exec_lo, s12
 ; GFX10-NEXT:    image_sample_cl v[0:3], v[1:2], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_2D a16
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
@@ -270,8 +262,8 @@
 ; GFX9:       ; %bb.0: ; %main_body
 ; GFX9-NEXT:    s_mov_b64 s[12:13], exec
 ; GFX9-NEXT:    s_wqm_b64 exec, exec
-; GFX9-NEXT:    v_and_b32_e32 v1, 0xffff, v1
-; GFX9-NEXT:    v_lshl_or_b32 v1, v2, 16, v1
+; GFX9-NEXT:    s_mov_b32 s14, 0x5040100
+; GFX9-NEXT:    v_perm_b32 v1, v2, v1, s14
 ; GFX9-NEXT:    s_and_b64 exec, exec, s[12:13]
 ; GFX9-NEXT:    image_sample_c_cl v[0:3], v[0:1], s[0:7], s[8:11] dmask:0xf a16
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
@@ -281,8 +273,7 @@
 ; GFX10:       ; %bb.0: ; %main_body
 ; GFX10-NEXT:    s_mov_b32 s12, exec_lo
 ; GFX10-NEXT:    s_wqm_b32 exec_lo, exec_lo
-; GFX10-NEXT:    v_and_b32_e32 v1, 0xffff, v1
-; GFX10-NEXT:    v_lshl_or_b32 v1, v2, 16, v1
+; GFX10-NEXT:    v_perm_b32 v1, v2, v1, 0x5040100
 ; GFX10-NEXT:    s_and_b32 exec_lo, exec_lo, s12
 ; GFX10-NEXT:    image_sample_c_cl v[0:3], v[0:1], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_1D a16
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
@@ -297,10 +288,10 @@
 ; GFX9:       ; %bb.0: ; %main_body
 ; GFX9-NEXT:    s_mov_b64 s[12:13], exec
 ; GFX9-NEXT:    s_wqm_b64 exec, exec
+; GFX9-NEXT:    s_mov_b32 s14, 0x5040100
 ; GFX9-NEXT:    v_mov_b32_e32 v5, v3
 ; GFX9-NEXT:    v_mov_b32_e32 v3, v0
-; GFX9-NEXT:    v_and_b32_e32 v0, 0xffff, v1
-; GFX9-NEXT:    v_lshl_or_b32 v4, v2, 16, v0
+; GFX9-NEXT:    v_perm_b32 v4, v2, v1, s14
 ; GFX9-NEXT:    s_and_b64 exec, exec, s[12:13]
 ; GFX9-NEXT:    image_sample_c_cl v[0:3], v[3:5], s[0:7], s[8:11] dmask:0xf a16
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
@@ -310,8 +301,7 @@
 ; GFX10:       ; %bb.0: ; %main_body
 ; GFX10-NEXT:    s_mov_b32 s12, exec_lo
 ; GFX10-NEXT:    s_wqm_b32 exec_lo, exec_lo
-; GFX10-NEXT:    v_and_b32_e32 v1, 0xffff, v1
-; GFX10-NEXT:    v_lshl_or_b32 v1, v2, 16, v1
+; GFX10-NEXT:    v_perm_b32 v1, v2, v1, 0x5040100
 ; GFX10-NEXT:    s_and_b32 exec_lo, exec_lo, s12
 ; GFX10-NEXT:    image_sample_c_cl v[0:3], [v0, v1, v3], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_2D a16
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
@@ -349,8 +339,8 @@
 ; GFX9:       ; %bb.0: ; %main_body
 ; GFX9-NEXT:    s_mov_b64 s[12:13], exec
 ; GFX9-NEXT:    s_wqm_b64 exec, exec
-; GFX9-NEXT:    v_and_b32_e32 v1, 0xffff, v1
-; GFX9-NEXT:    v_lshl_or_b32 v1, v2, 16, v1
+; GFX9-NEXT:    s_mov_b32 s14, 0x5040100
+; GFX9-NEXT:    v_perm_b32 v1, v2, v1, s14
 ; GFX9-NEXT:    s_and_b64 exec, exec, s[12:13]
 ; GFX9-NEXT:    image_sample_b v[0:3], v[0:1], s[0:7], s[8:11] dmask:0xf a16
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
@@ -360,8 +350,7 @@
 ; GFX10:       ; %bb.0: ; %main_body
 ; GFX10-NEXT:    s_mov_b32 s12, exec_lo
 ; GFX10-NEXT:    s_wqm_b32 exec_lo, exec_lo
-; GFX10-NEXT:    v_and_b32_e32 v1, 0xffff, v1
-; GFX10-NEXT:    v_lshl_or_b32 v1, v2, 16, v1
+; GFX10-NEXT:    v_perm_b32 v1, v2, v1, 0x5040100
 ; GFX10-NEXT:    s_and_b32 exec_lo, exec_lo, s12
 ; GFX10-NEXT:    image_sample_b v[0:3], v[0:1], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_2D a16
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
@@ -399,8 +388,8 @@
 ; GFX9:       ; %bb.0: ; %main_body
 ; GFX9-NEXT:    s_mov_b64 s[12:13], exec
 ; GFX9-NEXT:    s_wqm_b64 exec, exec
-; GFX9-NEXT:    v_and_b32_e32 v2, 0xffff, v2
-; GFX9-NEXT:    v_lshl_or_b32 v2, v3, 16, v2
+; GFX9-NEXT:    s_mov_b32 s14, 0x5040100
+; GFX9-NEXT:    v_perm_b32 v2, v3, v2, s14
 ; GFX9-NEXT:    s_and_b64 exec, exec, s[12:13]
 ; GFX9-NEXT:    image_sample_c_b v[0:3], v[0:2], s[0:7], s[8:11] dmask:0xf a16
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
@@ -410,8 +399,7 @@
 ; GFX10:       ; %bb.0: ; %main_body
 ; GFX10-NEXT:    s_mov_b32 s12, exec_lo
 ; GFX10-NEXT:    s_wqm_b32 exec_lo, exec_lo
-; GFX10-NEXT:    v_and_b32_e32 v2, 0xffff, v2
-; GFX10-NEXT:    v_lshl_or_b32 v2, v3, 16, v2
+; GFX10-NEXT:    v_perm_b32 v2, v3, v2, 0x5040100
 ; GFX10-NEXT:    s_and_b32 exec_lo, exec_lo, s12
 ; GFX10-NEXT:    image_sample_c_b v[0:3], v[0:2], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_2D a16
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
@@ -426,8 +414,8 @@
 ; GFX9:       ; %bb.0: ; %main_body
 ; GFX9-NEXT:    s_mov_b64 s[12:13], exec
 ; GFX9-NEXT:    s_wqm_b64 exec, exec
-; GFX9-NEXT:    v_and_b32_e32 v1, 0xffff, v1
-; GFX9-NEXT:    v_lshl_or_b32 v1, v2, 16, v1
+; GFX9-NEXT:    s_mov_b32 s14, 0x5040100
+; GFX9-NEXT:    v_perm_b32 v1, v2, v1, s14
 ; GFX9-NEXT:    s_and_b64 exec, exec, s[12:13]
 ; GFX9-NEXT:    image_sample_b_cl v[0:3], v[0:1], s[0:7], s[8:11] dmask:0xf a16
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
@@ -437,8 +425,7 @@
 ; GFX10:       ; %bb.0: ; %main_body
 ; GFX10-NEXT:    s_mov_b32 s12, exec_lo
 ; GFX10-NEXT:    s_wqm_b32 exec_lo, exec_lo
-; GFX10-NEXT:    v_and_b32_e32 v1, 0xffff, v1
-; GFX10-NEXT:    v_lshl_or_b32 v1, v2, 16, v1
+; GFX10-NEXT:    v_perm_b32 v1, v2, v1, 0x5040100
 ; GFX10-NEXT:    s_and_b32 exec_lo, exec_lo, s12
 ; GFX10-NEXT:    image_sample_b_cl v[0:3], v[0:1], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_1D a16
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
@@ -453,10 +440,10 @@
 ; GFX9:       ; %bb.0: ; %main_body
 ; GFX9-NEXT:    s_mov_b64 s[12:13], exec
 ; GFX9-NEXT:    s_wqm_b64 exec, exec
+; GFX9-NEXT:    s_mov_b32 s14, 0x5040100
 ; GFX9-NEXT:    v_mov_b32_e32 v5, v3
 ; GFX9-NEXT:    v_mov_b32_e32 v3, v0
-; GFX9-NEXT:    v_and_b32_e32 v0, 0xffff, v1
-; GFX9-NEXT:    v_lshl_or_b32 v4, v2, 16, v0
+; GFX9-NEXT:    v_perm_b32 v4, v2, v1, s14
 ; GFX9-NEXT:    s_and_b64 exec, exec, s[12:13]
 ; GFX9-NEXT:    image_sample_b_cl v[0:3], v[3:5], s[0:7], s[8:11] dmask:0xf a16
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
@@ -466,8 +453,7 @@
 ; GFX10:       ; %bb.0: ; %main_body
 ; GFX10-NEXT:    s_mov_b32 s12, exec_lo
 ; GFX10-NEXT:    s_wqm_b32 exec_lo, exec_lo
-; GFX10-NEXT:    v_and_b32_e32 v1, 0xffff, v1
-; GFX10-NEXT:    v_lshl_or_b32 v1, v2, 16, v1
+; GFX10-NEXT:    v_perm_b32 v1, v2, v1, 0x5040100
 ; GFX10-NEXT:    s_and_b32 exec_lo, exec_lo, s12
 ; GFX10-NEXT:    image_sample_b_cl v[0:3], [v0, v1, v3], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_2D a16
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
@@ -482,8 +468,8 @@
 ; GFX9:       ; %bb.0: ; %main_body
 ; GFX9-NEXT:    s_mov_b64 s[12:13], exec
 ; GFX9-NEXT:    s_wqm_b64 exec, exec
-; GFX9-NEXT:    v_and_b32_e32 v2, 0xffff, v2
-; GFX9-NEXT:    v_lshl_or_b32 v2, v3, 16, v2
+; GFX9-NEXT:    s_mov_b32 s14, 0x5040100
+; GFX9-NEXT:    v_perm_b32 v2, v3, v2, s14
 ; GFX9-NEXT:    s_and_b64 exec, exec, s[12:13]
 ; GFX9-NEXT:    image_sample_c_b_cl v[0:3], v[0:2], s[0:7], s[8:11] dmask:0xf a16
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
@@ -493,8 +479,7 @@
 ; GFX10:       ; %bb.0: ; %main_body
 ; GFX10-NEXT:    s_mov_b32 s12, exec_lo
 ; GFX10-NEXT:    s_wqm_b32 exec_lo, exec_lo
-; GFX10-NEXT:    v_and_b32_e32 v2, 0xffff, v2
-; GFX10-NEXT:    v_lshl_or_b32 v2, v3, 16, v2
+; GFX10-NEXT:    v_perm_b32 v2, v3, v2, 0x5040100
 ; GFX10-NEXT:    s_and_b32 exec_lo, exec_lo, s12
 ; GFX10-NEXT:    image_sample_c_b_cl v[0:3], v[0:2], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_1D a16
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
@@ -509,11 +494,11 @@
 ; GFX9:       ; %bb.0: ; %main_body
 ; GFX9-NEXT:    s_mov_b64 s[12:13], exec
 ; GFX9-NEXT:    s_wqm_b64 exec, exec
+; GFX9-NEXT:    s_mov_b32 s14, 0x5040100
 ; GFX9-NEXT:    v_mov_b32_e32 v7, v4
-; GFX9-NEXT:    v_mov_b32_e32 v4, v0
-; GFX9-NEXT:    v_and_b32_e32 v0, 0xffff, v2
 ; GFX9-NEXT:    v_mov_b32_e32 v5, v1
-; GFX9-NEXT:    v_lshl_or_b32 v6, v3, 16, v0
+; GFX9-NEXT:    v_mov_b32_e32 v4, v0
+; GFX9-NEXT:    v_perm_b32 v6, v3, v2, s14
 ; GFX9-NEXT:    s_and_b64 exec, exec, s[12:13]
 ; GFX9-NEXT:    image_sample_c_b_cl v[0:3], v[4:7], s[0:7], s[8:11] dmask:0xf a16
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
@@ -523,8 +508,7 @@
 ; GFX10:       ; %bb.0: ; %main_body
 ; GFX10-NEXT:    s_mov_b32 s12, exec_lo
 ; GFX10-NEXT:    s_wqm_b32 exec_lo, exec_lo
-; GFX10-NEXT:    v_and_b32_e32 v2, 0xffff, v2
-; GFX10-NEXT:    v_lshl_or_b32 v2, v3, 16, v2
+; GFX10-NEXT:    v_perm_b32 v2, v3, v2, 0x5040100
 ; GFX10-NEXT:    s_and_b32 exec_lo, exec_lo, s12
 ; GFX10-NEXT:    image_sample_c_b_cl v[0:3], [v0, v1, v2, v4], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_2D a16
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
@@ -554,24 +538,19 @@
 define amdgpu_ps <4 x float> @sample_d_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %dsdh, half %dtdh, half %dsdv, half %dtdv, half %s, half %t) {
 ; GFX9-LABEL: sample_d_2d:
 ; GFX9:       ; %bb.0: ; %main_body
-; GFX9-NEXT:    v_and_b32_e32 v4, 0xffff, v4
-; GFX9-NEXT:    v_and_b32_e32 v2, 0xffff, v2
-; GFX9-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX9-NEXT:    v_lshl_or_b32 v4, v5, 16, v4
-; GFX9-NEXT:    v_lshl_or_b32 v3, v3, 16, v2
-; GFX9-NEXT:    v_lshl_or_b32 v2, v1, 16, v0
+; GFX9-NEXT:    s_mov_b32 s12, 0x5040100
+; GFX9-NEXT:    v_perm_b32 v4, v5, v4, s12
+; GFX9-NEXT:    v_perm_b32 v3, v3, v2, s12
+; GFX9-NEXT:    v_perm_b32 v2, v1, v0, s12
 ; GFX9-NEXT:    image_sample_d v[0:3], v[2:4], s[0:7], s[8:11] dmask:0xf a16
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    ; return to shader part epilog
 ;
 ; GFX10-LABEL: sample_d_2d:
 ; GFX10:       ; %bb.0: ; %main_body
-; GFX10-NEXT:    v_and_b32_e32 v4, 0xffff, v4
-; GFX10-NEXT:    v_and_b32_e32 v2, 0xffff, v2
-; GFX10-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX10-NEXT:    v_lshl_or_b32 v4, v5, 16, v4
-; GFX10-NEXT:    v_lshl_or_b32 v3, v3, 16, v2
-; GFX10-NEXT:    v_lshl_or_b32 v2, v1, 16, v0
+; GFX10-NEXT:    v_perm_b32 v4, v5, v4, 0x5040100
+; GFX10-NEXT:    v_perm_b32 v3, v3, v2, 0x5040100
+; GFX10-NEXT:    v_perm_b32 v2, v1, v0, 0x5040100
 ; GFX10-NEXT:    image_sample_d_g16 v[0:3], v[2:4], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_2D a16
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    ; return to shader part epilog
@@ -583,15 +562,13 @@
 define amdgpu_ps <4 x float> @sample_d_3d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %dsdh, half %dtdh, half %drdh, half %dsdv, half %dtdv, half %drdv, half %s, half %t, half %r) {
 ; GFX9-LABEL: sample_d_3d:
 ; GFX9:       ; %bb.0: ; %main_body
+; GFX9-NEXT:    s_mov_b32 s12, 0x5040100
 ; GFX9-NEXT:    v_mov_b32_e32 v12, v8
-; GFX9-NEXT:    v_mov_b32_e32 v8, v2
-; GFX9-NEXT:    v_and_b32_e32 v2, 0xffff, v6
-; GFX9-NEXT:    v_lshl_or_b32 v11, v7, 16, v2
-; GFX9-NEXT:    v_and_b32_e32 v2, 0xffff, v3
-; GFX9-NEXT:    v_and_b32_e32 v0, 0xffff, v0
 ; GFX9-NEXT:    v_mov_b32_e32 v10, v5
-; GFX9-NEXT:    v_lshl_or_b32 v9, v4, 16, v2
-; GFX9-NEXT:    v_lshl_or_b32 v7, v1, 16, v0
+; GFX9-NEXT:    v_mov_b32_e32 v8, v2
+; GFX9-NEXT:    v_perm_b32 v11, v7, v6, s12
+; GFX9-NEXT:    v_perm_b32 v9, v4, v3, s12
+; GFX9-NEXT:    v_perm_b32 v7, v1, v0, s12
 ; GFX9-NEXT:    image_sample_d v[0:3], v[7:12], s[0:7], s[8:11] dmask:0xf a16
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    ; return to shader part epilog
@@ -599,14 +576,11 @@
 ; GFX10-LABEL: sample_d_3d:
 ; GFX10:       ; %bb.0: ; %main_body
 ; GFX10-NEXT:    v_mov_b32_e32 v12, v8
-; GFX10-NEXT:    v_mov_b32_e32 v8, v2
-; GFX10-NEXT:    v_and_b32_e32 v2, 0xffff, v6
-; GFX10-NEXT:    v_and_b32_e32 v3, 0xffff, v3
-; GFX10-NEXT:    v_and_b32_e32 v0, 0xffff, v0
 ; GFX10-NEXT:    v_mov_b32_e32 v10, v5
-; GFX10-NEXT:    v_lshl_or_b32 v11, v7, 16, v2
-; GFX10-NEXT:    v_lshl_or_b32 v9, v4, 16, v3
-; GFX10-NEXT:    v_lshl_or_b32 v7, v1, 16, v0
+; GFX10-NEXT:    v_mov_b32_e32 v8, v2
+; GFX10-NEXT:    v_perm_b32 v11, v7, v6, 0x5040100
+; GFX10-NEXT:    v_perm_b32 v9, v4, v3, 0x5040100
+; GFX10-NEXT:    v_perm_b32 v7, v1, v0, 0x5040100
 ; GFX10-NEXT:    image_sample_d_g16 v[0:3], v[7:12], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_3D a16
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    ; return to shader part epilog
@@ -637,24 +611,19 @@
 ; GFX9:       ; %bb.0: ; %main_body
 ; GFX9-NEXT:    v_mov_b32_e32 v7, v3
 ; GFX9-NEXT:    v_mov_b32_e32 v8, v2
-; GFX9-NEXT:    v_and_b32_e32 v2, 0xffff, v5
-; GFX9-NEXT:    v_lshl_or_b32 v3, v6, 16, v2
-; GFX9-NEXT:    v_and_b32_e32 v2, 0xffff, v7
-; GFX9-NEXT:    v_and_b32_e32 v1, 0xffff, v1
-; GFX9-NEXT:    v_lshl_or_b32 v2, v4, 16, v2
-; GFX9-NEXT:    v_lshl_or_b32 v1, v8, 16, v1
+; GFX9-NEXT:    s_mov_b32 s12, 0x5040100
+; GFX9-NEXT:    v_perm_b32 v3, v6, v5, s12
+; GFX9-NEXT:    v_perm_b32 v2, v4, v7, s12
+; GFX9-NEXT:    v_perm_b32 v1, v8, v1, s12
 ; GFX9-NEXT:    image_sample_c_d v[0:3], v[0:3], s[0:7], s[8:11] dmask:0xf a16
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    ; return to shader part epilog
 ;
 ; GFX10-LABEL: sample_c_d_2d:
 ; GFX10:       ; %bb.0: ; %main_body
-; GFX10-NEXT:    v_and_b32_e32 v5, 0xffff, v5
-; GFX10-NEXT:    v_and_b32_e32 v3, 0xffff, v3
-; GFX10-NEXT:    v_and_b32_e32 v1, 0xffff, v1
-; GFX10-NEXT:    v_lshl_or_b32 v5, v6, 16, v5
-; GFX10-NEXT:    v_lshl_or_b32 v3, v4, 16, v3
-; GFX10-NEXT:    v_lshl_or_b32 v1, v2, 16, v1
+; GFX10-NEXT:    v_perm_b32 v5, v6, v5, 0x5040100
+; GFX10-NEXT:    v_perm_b32 v3, v4, v3, 0x5040100
+; GFX10-NEXT:    v_perm_b32 v1, v2, v1, 0x5040100
 ; GFX10-NEXT:    image_sample_c_d_g16 v[0:3], [v0, v1, v3, v5], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_2D a16
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    ; return to shader part epilog
@@ -666,16 +635,15 @@
 define amdgpu_ps <4 x float> @sample_d_cl_1d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %dsdh, half %dsdv, half %s, half %clamp) {
 ; GFX9-LABEL: sample_d_cl_1d:
 ; GFX9:       ; %bb.0: ; %main_body
-; GFX9-NEXT:    v_and_b32_e32 v2, 0xffff, v2
-; GFX9-NEXT:    v_lshl_or_b32 v2, v3, 16, v2
+; GFX9-NEXT:    s_mov_b32 s12, 0x5040100
+; GFX9-NEXT:    v_perm_b32 v2, v3, v2, s12
 ; GFX9-NEXT:    image_sample_d_cl v[0:3], v[0:2], s[0:7], s[8:11] dmask:0xf a16
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    ; return to shader part epilog
 ;
 ; GFX10-LABEL: sample_d_cl_1d:
 ; GFX10:       ; %bb.0: ; %main_body
-; GFX10-NEXT:    v_and_b32_e32 v2, 0xffff, v2
-; GFX10-NEXT:    v_lshl_or_b32 v2, v3, 16, v2
+; GFX10-NEXT:    v_perm_b32 v2, v3, v2, 0x5040100
 ; GFX10-NEXT:    image_sample_d_cl_g16 v[0:3], v[0:2], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_1D a16
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    ; return to shader part epilog
@@ -687,24 +655,19 @@
 define amdgpu_ps <4 x float> @sample_d_cl_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %dsdh, half %dtdh, half %dsdv, half %dtdv, half %s, half %t, half %clamp) {
 ; GFX9-LABEL: sample_d_cl_2d:
 ; GFX9:       ; %bb.0: ; %main_body
-; GFX9-NEXT:    v_and_b32_e32 v4, 0xffff, v4
-; GFX9-NEXT:    v_and_b32_e32 v2, 0xffff, v2
-; GFX9-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX9-NEXT:    v_lshl_or_b32 v5, v5, 16, v4
-; GFX9-NEXT:    v_lshl_or_b32 v4, v3, 16, v2
-; GFX9-NEXT:    v_lshl_or_b32 v3, v1, 16, v0
+; GFX9-NEXT:    s_mov_b32 s12, 0x5040100
+; GFX9-NEXT:    v_perm_b32 v5, v5, v4, s12
+; GFX9-NEXT:    v_perm_b32 v4, v3, v2, s12
+; GFX9-NEXT:    v_perm_b32 v3, v1, v0, s12
 ; GFX9-NEXT:    image_sample_d_cl v[0:3], v[3:6], s[0:7], s[8:11] dmask:0xf a16
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    ; return to shader part epilog
 ;
 ; GFX10-LABEL: sample_d_cl_2d:
 ; GFX10:       ; %bb.0: ; %main_body
-; GFX10-NEXT:    v_and_b32_e32 v4, 0xffff, v4
-; GFX10-NEXT:    v_and_b32_e32 v2, 0xffff, v2
-; GFX10-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX10-NEXT:    v_lshl_or_b32 v4, v5, 16, v4
-; GFX10-NEXT:    v_lshl_or_b32 v2, v3, 16, v2
-; GFX10-NEXT:    v_lshl_or_b32 v0, v1, 16, v0
+; GFX10-NEXT:    v_perm_b32 v4, v5, v4, 0x5040100
+; GFX10-NEXT:    v_perm_b32 v2, v3, v2, 0x5040100
+; GFX10-NEXT:    v_perm_b32 v0, v1, v0, 0x5040100
 ; GFX10-NEXT:    image_sample_d_cl_g16 v[0:3], [v0, v2, v4, v6], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_2D a16
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    ; return to shader part epilog
@@ -716,16 +679,15 @@
 define amdgpu_ps <4 x float> @sample_c_d_cl_1d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %zcompare, half %dsdh, half %dsdv, half %s, half %clamp) {
 ; GFX9-LABEL: sample_c_d_cl_1d:
 ; GFX9:       ; %bb.0: ; %main_body
-; GFX9-NEXT:    v_and_b32_e32 v3, 0xffff, v3
-; GFX9-NEXT:    v_lshl_or_b32 v3, v4, 16, v3
+; GFX9-NEXT:    s_mov_b32 s12, 0x5040100
+; GFX9-NEXT:    v_perm_b32 v3, v4, v3, s12
 ; GFX9-NEXT:    image_sample_c_d_cl v[0:3], v[0:3], s[0:7], s[8:11] dmask:0xf a16
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    ; return to shader part epilog
 ;
 ; GFX10-LABEL: sample_c_d_cl_1d:
 ; GFX10:       ; %bb.0: ; %main_body
-; GFX10-NEXT:    v_and_b32_e32 v3, 0xffff, v3
-; GFX10-NEXT:    v_lshl_or_b32 v3, v4, 16, v3
+; GFX10-NEXT:    v_perm_b32 v3, v4, v3, 0x5040100
 ; GFX10-NEXT:    image_sample_c_d_cl_g16 v[0:3], v[0:3], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_1D a16
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    ; return to shader part epilog
@@ -737,26 +699,21 @@
 define amdgpu_ps <4 x float> @sample_c_d_cl_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %zcompare, half %dsdh, half %dtdh, half %dsdv, half %dtdv, half %s, half %t, half %clamp) {
 ; GFX9-LABEL: sample_c_d_cl_2d:
 ; GFX9:       ; %bb.0: ; %main_body
+; GFX9-NEXT:    s_mov_b32 s12, 0x5040100
 ; GFX9-NEXT:    v_mov_b32_e32 v11, v7
 ; GFX9-NEXT:    v_mov_b32_e32 v7, v0
-; GFX9-NEXT:    v_and_b32_e32 v0, 0xffff, v5
-; GFX9-NEXT:    v_lshl_or_b32 v10, v6, 16, v0
-; GFX9-NEXT:    v_and_b32_e32 v0, 0xffff, v3
-; GFX9-NEXT:    v_lshl_or_b32 v9, v4, 16, v0
-; GFX9-NEXT:    v_and_b32_e32 v0, 0xffff, v1
-; GFX9-NEXT:    v_lshl_or_b32 v8, v2, 16, v0
+; GFX9-NEXT:    v_perm_b32 v10, v6, v5, s12
+; GFX9-NEXT:    v_perm_b32 v9, v4, v3, s12
+; GFX9-NEXT:    v_perm_b32 v8, v2, v1, s12
 ; GFX9-NEXT:    image_sample_c_d_cl v[0:3], v[7:11], s[0:7], s[8:11] dmask:0xf a16
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    ; return to shader part epilog
 ;
 ; GFX10-LABEL: sample_c_d_cl_2d:
 ; GFX10:       ; %bb.0: ; %main_body
-; GFX10-NEXT:    v_and_b32_e32 v5, 0xffff, v5
-; GFX10-NEXT:    v_and_b32_e32 v3, 0xffff, v3
-; GFX10-NEXT:    v_and_b32_e32 v1, 0xffff, v1
-; GFX10-NEXT:    v_lshl_or_b32 v5, v6, 16, v5
-; GFX10-NEXT:    v_lshl_or_b32 v3, v4, 16, v3
-; GFX10-NEXT:    v_lshl_or_b32 v1, v2, 16, v1
+; GFX10-NEXT:    v_perm_b32 v5, v6, v5, 0x5040100
+; GFX10-NEXT:    v_perm_b32 v3, v4, v3, 0x5040100
+; GFX10-NEXT:    v_perm_b32 v1, v2, v1, 0x5040100
 ; GFX10-NEXT:    image_sample_c_d_cl_g16 v[0:3], [v0, v1, v3, v5, v7], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_2D a16
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    ; return to shader part epilog
@@ -768,16 +725,15 @@
 define amdgpu_ps <4 x float> @sample_l_1d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %s, half %lod) {
 ; GFX9-LABEL: sample_l_1d:
 ; GFX9:       ; %bb.0: ; %main_body
-; GFX9-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX9-NEXT:    v_lshl_or_b32 v0, v1, 16, v0
+; GFX9-NEXT:    s_mov_b32 s12, 0x5040100
+; GFX9-NEXT:    v_perm_b32 v0, v1, v0, s12
 ; GFX9-NEXT:    image_sample_l v[0:3], v0, s[0:7], s[8:11] dmask:0xf a16
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    ; return to shader part epilog
 ;
 ; GFX10-LABEL: sample_l_1d:
 ; GFX10:       ; %bb.0: ; %main_body
-; GFX10-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX10-NEXT:    v_lshl_or_b32 v0, v1, 16, v0
+; GFX10-NEXT:    v_perm_b32 v0, v1, v0, 0x5040100
 ; GFX10-NEXT:    image_sample_l v[0:3], v0, s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_1D a16
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    ; return to shader part epilog
@@ -789,16 +745,15 @@
 define amdgpu_ps <4 x float> @sample_l_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %s, half %t, half %lod) {
 ; GFX9-LABEL: sample_l_2d:
 ; GFX9:       ; %bb.0: ; %main_body
-; GFX9-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX9-NEXT:    v_lshl_or_b32 v1, v1, 16, v0
+; GFX9-NEXT:    s_mov_b32 s12, 0x5040100
+; GFX9-NEXT:    v_perm_b32 v1, v1, v0, s12
 ; GFX9-NEXT:    image_sample_l v[0:3], v[1:2], s[0:7], s[8:11] dmask:0xf a16
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    ; return to shader part epilog
 ;
 ; GFX10-LABEL: sample_l_2d:
 ; GFX10:       ; %bb.0: ; %main_body
-; GFX10-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX10-NEXT:    v_lshl_or_b32 v1, v1, 16, v0
+; GFX10-NEXT:    v_perm_b32 v1, v1, v0, 0x5040100
 ; GFX10-NEXT:    image_sample_l v[0:3], v[1:2], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_2D a16
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    ; return to shader part epilog
@@ -810,16 +765,15 @@
 define amdgpu_ps <4 x float> @sample_c_l_1d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %zcompare, half %s, half %lod) {
 ; GFX9-LABEL: sample_c_l_1d:
 ; GFX9:       ; %bb.0: ; %main_body
-; GFX9-NEXT:    v_and_b32_e32 v1, 0xffff, v1
-; GFX9-NEXT:    v_lshl_or_b32 v1, v2, 16, v1
+; GFX9-NEXT:    s_mov_b32 s12, 0x5040100
+; GFX9-NEXT:    v_perm_b32 v1, v2, v1, s12
 ; GFX9-NEXT:    image_sample_c_l v[0:3], v[0:1], s[0:7], s[8:11] dmask:0xf a16
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    ; return to shader part epilog
 ;
 ; GFX10-LABEL: sample_c_l_1d:
 ; GFX10:       ; %bb.0: ; %main_body
-; GFX10-NEXT:    v_and_b32_e32 v1, 0xffff, v1
-; GFX10-NEXT:    v_lshl_or_b32 v1, v2, 16, v1
+; GFX10-NEXT:    v_perm_b32 v1, v2, v1, 0x5040100
 ; GFX10-NEXT:    image_sample_c_l v[0:3], v[0:1], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_1D a16
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    ; return to shader part epilog
@@ -831,18 +785,17 @@
 define amdgpu_ps <4 x float> @sample_c_l_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %zcompare, half %s, half %t, half %lod) {
 ; GFX9-LABEL: sample_c_l_2d:
 ; GFX9:       ; %bb.0: ; %main_body
+; GFX9-NEXT:    s_mov_b32 s12, 0x5040100
 ; GFX9-NEXT:    v_mov_b32_e32 v5, v3
 ; GFX9-NEXT:    v_mov_b32_e32 v3, v0
-; GFX9-NEXT:    v_and_b32_e32 v0, 0xffff, v1
-; GFX9-NEXT:    v_lshl_or_b32 v4, v2, 16, v0
+; GFX9-NEXT:    v_perm_b32 v4, v2, v1, s12
 ; GFX9-NEXT:    image_sample_c_l v[0:3], v[3:5], s[0:7], s[8:11] dmask:0xf a16
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    ; return to shader part epilog
 ;
 ; GFX10-LABEL: sample_c_l_2d:
 ; GFX10:       ; %bb.0: ; %main_body
-; GFX10-NEXT:    v_and_b32_e32 v1, 0xffff, v1
-; GFX10-NEXT:    v_lshl_or_b32 v1, v2, 16, v1
+; GFX10-NEXT:    v_perm_b32 v1, v2, v1, 0x5040100
 ; GFX10-NEXT:    image_sample_c_l v[0:3], [v0, v1, v3], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_2D a16
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    ; return to shader part epilog
@@ -871,16 +824,15 @@
 define amdgpu_ps <4 x float> @sample_lz_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %s, half %t) {
 ; GFX9-LABEL: sample_lz_2d:
 ; GFX9:       ; %bb.0: ; %main_body
-; GFX9-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX9-NEXT:    v_lshl_or_b32 v0, v1, 16, v0
+; GFX9-NEXT:    s_mov_b32 s12, 0x5040100
+; GFX9-NEXT:    v_perm_b32 v0, v1, v0, s12
 ; GFX9-NEXT:    image_sample_lz v[0:3], v0, s[0:7], s[8:11] dmask:0xf a16
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    ; return to shader part epilog
 ;
 ; GFX10-LABEL: sample_lz_2d:
 ; GFX10:       ; %bb.0: ; %main_body
-; GFX10-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX10-NEXT:    v_lshl_or_b32 v0, v1, 16, v0
+; GFX10-NEXT:    v_perm_b32 v0, v1, v0, 0x5040100
 ; GFX10-NEXT:    image_sample_lz v[0:3], v0, s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_2D a16
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    ; return to shader part epilog
@@ -909,16 +861,15 @@
 define amdgpu_ps <4 x float> @sample_c_lz_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %zcompare, half %s, half %t) {
 ; GFX9-LABEL: sample_c_lz_2d:
 ; GFX9:       ; %bb.0: ; %main_body
-; GFX9-NEXT:    v_and_b32_e32 v1, 0xffff, v1
-; GFX9-NEXT:    v_lshl_or_b32 v1, v2, 16, v1
+; GFX9-NEXT:    s_mov_b32 s12, 0x5040100
+; GFX9-NEXT:    v_perm_b32 v1, v2, v1, s12
 ; GFX9-NEXT:    image_sample_c_lz v[0:3], v[0:1], s[0:7], s[8:11] dmask:0xf a16
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    ; return to shader part epilog
 ;
 ; GFX10-LABEL: sample_c_lz_2d:
 ; GFX10:       ; %bb.0: ; %main_body
-; GFX10-NEXT:    v_and_b32_e32 v1, 0xffff, v1
-; GFX10-NEXT:    v_lshl_or_b32 v1, v2, 16, v1
+; GFX10-NEXT:    v_perm_b32 v1, v2, v1, 0x5040100
 ; GFX10-NEXT:    image_sample_c_lz v[0:3], v[0:1], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_2D a16
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    ; return to shader part epilog
@@ -930,15 +881,13 @@
 define amdgpu_ps float @sample_c_d_o_2darray_V1(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, i32 %offset, float %zcompare, half %dsdh, half %dtdh, half %dsdv, half %dtdv, half %s, half %t, half %slice) {
 ; GFX9-LABEL: sample_c_d_o_2darray_V1:
 ; GFX9:       ; %bb.0: ; %main_body
+; GFX9-NEXT:    s_mov_b32 s12, 0x5040100
 ; GFX9-NEXT:    v_mov_b32_e32 v13, v8
-; GFX9-NEXT:    v_mov_b32_e32 v8, v0
-; GFX9-NEXT:    v_and_b32_e32 v0, 0xffff, v6
-; GFX9-NEXT:    v_lshl_or_b32 v12, v7, 16, v0
-; GFX9-NEXT:    v_and_b32_e32 v0, 0xffff, v4
-; GFX9-NEXT:    v_lshl_or_b32 v11, v5, 16, v0
-; GFX9-NEXT:    v_and_b32_e32 v0, 0xffff, v2
 ; GFX9-NEXT:    v_mov_b32_e32 v9, v1
-; GFX9-NEXT:    v_lshl_or_b32 v10, v3, 16, v0
+; GFX9-NEXT:    v_mov_b32_e32 v8, v0
+; GFX9-NEXT:    v_perm_b32 v12, v7, v6, s12
+; GFX9-NEXT:    v_perm_b32 v11, v5, v4, s12
+; GFX9-NEXT:    v_perm_b32 v10, v3, v2, s12
 ; GFX9-NEXT:    image_sample_c_d_o v0, v[8:13], s[0:7], s[8:11] dmask:0x4 a16 da
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    ; return to shader part epilog
@@ -948,12 +897,9 @@
 ; GFX10-NEXT:    v_mov_b32_e32 v13, v8
 ; GFX10-NEXT:    v_mov_b32_e32 v9, v1
 ; GFX10-NEXT:    v_mov_b32_e32 v8, v0
-; GFX10-NEXT:    v_and_b32_e32 v0, 0xffff, v6
-; GFX10-NEXT:    v_and_b32_e32 v1, 0xffff, v4
-; GFX10-NEXT:    v_and_b32_e32 v2, 0xffff, v2
-; GFX10-NEXT:    v_lshl_or_b32 v12, v7, 16, v0
-; GFX10-NEXT:    v_lshl_or_b32 v11, v5, 16, v1
-; GFX10-NEXT:    v_lshl_or_b32 v10, v3, 16, v2
+; GFX10-NEXT:    v_perm_b32 v12, v7, v6, 0x5040100
+; GFX10-NEXT:    v_perm_b32 v11, v5, v4, 0x5040100
+; GFX10-NEXT:    v_perm_b32 v10, v3, v2, 0x5040100
 ; GFX10-NEXT:    image_sample_c_d_o_g16 v0, v[8:13], s[0:7], s[8:11] dmask:0x4 dim:SQ_RSRC_IMG_2D_ARRAY a16
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    ; return to shader part epilog
@@ -965,15 +911,13 @@
 define amdgpu_ps <2 x float> @sample_c_d_o_2darray_V2(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, i32 %offset, float %zcompare, half %dsdh, half %dtdh, half %dsdv, half %dtdv, half %s, half %t, half %slice) {
 ; GFX9-LABEL: sample_c_d_o_2darray_V2:
 ; GFX9:       ; %bb.0: ; %main_body
+; GFX9-NEXT:    s_mov_b32 s12, 0x5040100
 ; GFX9-NEXT:    v_mov_b32_e32 v13, v8
-; GFX9-NEXT:    v_mov_b32_e32 v8, v0
-; GFX9-NEXT:    v_and_b32_e32 v0, 0xffff, v6
-; GFX9-NEXT:    v_lshl_or_b32 v12, v7, 16, v0
-; GFX9-NEXT:    v_and_b32_e32 v0, 0xffff, v4
-; GFX9-NEXT:    v_lshl_or_b32 v11, v5, 16, v0
-; GFX9-NEXT:    v_and_b32_e32 v0, 0xffff, v2
 ; GFX9-NEXT:    v_mov_b32_e32 v9, v1
-; GFX9-NEXT:    v_lshl_or_b32 v10, v3, 16, v0
+; GFX9-NEXT:    v_mov_b32_e32 v8, v0
+; GFX9-NEXT:    v_perm_b32 v12, v7, v6, s12
+; GFX9-NEXT:    v_perm_b32 v11, v5, v4, s12
+; GFX9-NEXT:    v_perm_b32 v10, v3, v2, s12
 ; GFX9-NEXT:    image_sample_c_d_o v[0:1], v[8:13], s[0:7], s[8:11] dmask:0x6 a16 da
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    ; return to shader part epilog
@@ -983,12 +927,9 @@
 ; GFX10-NEXT:    v_mov_b32_e32 v13, v8
 ; GFX10-NEXT:    v_mov_b32_e32 v9, v1
 ; GFX10-NEXT:    v_mov_b32_e32 v8, v0
-; GFX10-NEXT:    v_and_b32_e32 v0, 0xffff, v6
-; GFX10-NEXT:    v_and_b32_e32 v1, 0xffff, v4
-; GFX10-NEXT:    v_and_b32_e32 v2, 0xffff, v2
-; GFX10-NEXT:    v_lshl_or_b32 v12, v7, 16, v0
-; GFX10-NEXT:    v_lshl_or_b32 v11, v5, 16, v1
-; GFX10-NEXT:    v_lshl_or_b32 v10, v3, 16, v2
+; GFX10-NEXT:    v_perm_b32 v12, v7, v6, 0x5040100
+; GFX10-NEXT:    v_perm_b32 v11, v5, v4, 0x5040100
+; GFX10-NEXT:    v_perm_b32 v10, v3, v2, 0x5040100
 ; GFX10-NEXT:    image_sample_c_d_o_g16 v[0:1], v[8:13], s[0:7], s[8:11] dmask:0x6 dim:SQ_RSRC_IMG_2D_ARRAY a16
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    ; return to shader part epilog
diff --git a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.image.sample.cd.a16.dim.ll b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.image.sample.cd.a16.dim.ll
--- a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.image.sample.cd.a16.dim.ll
+++ b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.image.sample.cd.a16.dim.ll
@@ -22,24 +22,19 @@
 define amdgpu_ps <4 x float> @sample_cd_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %dsdh, half %dtdh, half %dsdv, half %dtdv, half %s, half %t) {
 ; GFX9-LABEL: sample_cd_2d:
 ; GFX9:       ; %bb.0: ; %main_body
-; GFX9-NEXT:    v_and_b32_e32 v4, 0xffff, v4
-; GFX9-NEXT:    v_and_b32_e32 v2, 0xffff, v2
-; GFX9-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX9-NEXT:    v_lshl_or_b32 v4, v5, 16, v4
-; GFX9-NEXT:    v_lshl_or_b32 v3, v3, 16, v2
-; GFX9-NEXT:    v_lshl_or_b32 v2, v1, 16, v0
+; GFX9-NEXT:    s_mov_b32 s12, 0x5040100
+; GFX9-NEXT:    v_perm_b32 v4, v5, v4, s12
+; GFX9-NEXT:    v_perm_b32 v3, v3, v2, s12
+; GFX9-NEXT:    v_perm_b32 v2, v1, v0, s12
 ; GFX9-NEXT:    image_sample_cd v[0:3], v[2:4], s[0:7], s[8:11] dmask:0xf a16
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    ; return to shader part epilog
 ;
 ; GFX10-LABEL: sample_cd_2d:
 ; GFX10:       ; %bb.0: ; %main_body
-; GFX10-NEXT:    v_and_b32_e32 v4, 0xffff, v4
-; GFX10-NEXT:    v_and_b32_e32 v2, 0xffff, v2
-; GFX10-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX10-NEXT:    v_lshl_or_b32 v4, v5, 16, v4
-; GFX10-NEXT:    v_lshl_or_b32 v3, v3, 16, v2
-; GFX10-NEXT:    v_lshl_or_b32 v2, v1, 16, v0
+; GFX10-NEXT:    v_perm_b32 v4, v5, v4, 0x5040100
+; GFX10-NEXT:    v_perm_b32 v3, v3, v2, 0x5040100
+; GFX10-NEXT:    v_perm_b32 v2, v1, v0, 0x5040100
 ; GFX10-NEXT:    image_sample_cd_g16 v[0:3], v[2:4], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_2D a16
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    ; return to shader part epilog
@@ -70,24 +65,19 @@
 ; GFX9:       ; %bb.0: ; %main_body
 ; GFX9-NEXT:    v_mov_b32_e32 v7, v3
 ; GFX9-NEXT:    v_mov_b32_e32 v8, v2
-; GFX9-NEXT:    v_and_b32_e32 v2, 0xffff, v5
-; GFX9-NEXT:    v_lshl_or_b32 v3, v6, 16, v2
-; GFX9-NEXT:    v_and_b32_e32 v2, 0xffff, v7
-; GFX9-NEXT:    v_and_b32_e32 v1, 0xffff, v1
-; GFX9-NEXT:    v_lshl_or_b32 v2, v4, 16, v2
-; GFX9-NEXT:    v_lshl_or_b32 v1, v8, 16, v1
+; GFX9-NEXT:    s_mov_b32 s12, 0x5040100
+; GFX9-NEXT:    v_perm_b32 v3, v6, v5, s12
+; GFX9-NEXT:    v_perm_b32 v2, v4, v7, s12
+; GFX9-NEXT:    v_perm_b32 v1, v8, v1, s12
 ; GFX9-NEXT:    image_sample_c_cd v[0:3], v[0:3], s[0:7], s[8:11] dmask:0xf a16
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    ; return to shader part epilog
 ;
 ; GFX10-LABEL: sample_c_cd_2d:
 ; GFX10:       ; %bb.0: ; %main_body
-; GFX10-NEXT:    v_and_b32_e32 v5, 0xffff, v5
-; GFX10-NEXT:    v_and_b32_e32 v3, 0xffff, v3
-; GFX10-NEXT:    v_and_b32_e32 v1, 0xffff, v1
-; GFX10-NEXT:    v_lshl_or_b32 v5, v6, 16, v5
-; GFX10-NEXT:    v_lshl_or_b32 v3, v4, 16, v3
-; GFX10-NEXT:    v_lshl_or_b32 v1, v2, 16, v1
+; GFX10-NEXT:    v_perm_b32 v5, v6, v5, 0x5040100
+; GFX10-NEXT:    v_perm_b32 v3, v4, v3, 0x5040100
+; GFX10-NEXT:    v_perm_b32 v1, v2, v1, 0x5040100
 ; GFX10-NEXT:    image_sample_c_cd_g16 v[0:3], [v0, v1, v3, v5], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_2D a16
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    ; return to shader part epilog
@@ -99,16 +89,15 @@
 define amdgpu_ps <4 x float> @sample_cd_cl_1d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %dsdh, half %dsdv, half %s, half %clamp) {
 ; GFX9-LABEL: sample_cd_cl_1d:
 ; GFX9:       ; %bb.0: ; %main_body
-; GFX9-NEXT:    v_and_b32_e32 v2, 0xffff, v2
-; GFX9-NEXT:    v_lshl_or_b32 v2, v3, 16, v2
+; GFX9-NEXT:    s_mov_b32 s12, 0x5040100
+; GFX9-NEXT:    v_perm_b32 v2, v3, v2, s12
 ; GFX9-NEXT:    image_sample_cd_cl v[0:3], v[0:2], s[0:7], s[8:11] dmask:0xf a16
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    ; return to shader part epilog
 ;
 ; GFX10-LABEL: sample_cd_cl_1d:
 ; GFX10:       ; %bb.0: ; %main_body
-; GFX10-NEXT:    v_and_b32_e32 v2, 0xffff, v2
-; GFX10-NEXT:    v_lshl_or_b32 v2, v3, 16, v2
+; GFX10-NEXT:    v_perm_b32 v2, v3, v2, 0x5040100
 ; GFX10-NEXT:    image_sample_cd_cl_g16 v[0:3], v[0:2], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_1D a16
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    ; return to shader part epilog
@@ -120,24 +109,19 @@
 define amdgpu_ps <4 x float> @sample_cd_cl_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %dsdh, half %dtdh, half %dsdv, half %dtdv, half %s, half %t, half %clamp) {
 ; GFX9-LABEL: sample_cd_cl_2d:
 ; GFX9:       ; %bb.0: ; %main_body
-; GFX9-NEXT:    v_and_b32_e32 v4, 0xffff, v4
-; GFX9-NEXT:    v_and_b32_e32 v2, 0xffff, v2
-; GFX9-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX9-NEXT:    v_lshl_or_b32 v5, v5, 16, v4
-; GFX9-NEXT:    v_lshl_or_b32 v4, v3, 16, v2
-; GFX9-NEXT:    v_lshl_or_b32 v3, v1, 16, v0
+; GFX9-NEXT:    s_mov_b32 s12, 0x5040100
+; GFX9-NEXT:    v_perm_b32 v5, v5, v4, s12
+; GFX9-NEXT:    v_perm_b32 v4, v3, v2, s12
+; GFX9-NEXT:    v_perm_b32 v3, v1, v0, s12
 ; GFX9-NEXT:    image_sample_cd_cl v[0:3], v[3:6], s[0:7], s[8:11] dmask:0xf a16
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    ; return to shader part epilog
 ;
 ; GFX10-LABEL: sample_cd_cl_2d:
 ; GFX10:       ; %bb.0: ; %main_body
-; GFX10-NEXT:    v_and_b32_e32 v4, 0xffff, v4
-; GFX10-NEXT:    v_and_b32_e32 v2, 0xffff, v2
-; GFX10-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX10-NEXT:    v_lshl_or_b32 v4, v5, 16, v4
-; GFX10-NEXT:    v_lshl_or_b32 v2, v3, 16, v2
-; GFX10-NEXT:    v_lshl_or_b32 v0, v1, 16, v0
+; GFX10-NEXT:    v_perm_b32 v4, v5, v4, 0x5040100
+; GFX10-NEXT:    v_perm_b32 v2, v3, v2, 0x5040100
+; GFX10-NEXT:    v_perm_b32 v0, v1, v0, 0x5040100
 ; GFX10-NEXT:    image_sample_cd_cl_g16 v[0:3], [v0, v2, v4, v6], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_2D a16
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    ; return to shader part epilog
@@ -149,16 +133,15 @@
 define amdgpu_ps <4 x float> @sample_c_cd_cl_1d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %zcompare, half %dsdh, half %dsdv, half %s, half %clamp) {
 ; GFX9-LABEL: sample_c_cd_cl_1d:
 ; GFX9:       ; %bb.0: ; %main_body
-; GFX9-NEXT:    v_and_b32_e32 v3, 0xffff, v3
-; GFX9-NEXT:    v_lshl_or_b32 v3, v4, 16, v3
+; GFX9-NEXT:    s_mov_b32 s12, 0x5040100
+; GFX9-NEXT:    v_perm_b32 v3, v4, v3, s12
 ; GFX9-NEXT:    image_sample_c_cd_cl v[0:3], v[0:3], s[0:7], s[8:11] dmask:0xf a16
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    ; return to shader part epilog
 ;
 ; GFX10-LABEL: sample_c_cd_cl_1d:
 ; GFX10:       ; %bb.0: ; %main_body
-; GFX10-NEXT:    v_and_b32_e32 v3, 0xffff, v3
-; GFX10-NEXT:    v_lshl_or_b32 v3, v4, 16, v3
+; GFX10-NEXT:    v_perm_b32 v3, v4, v3, 0x5040100
 ; GFX10-NEXT:    image_sample_c_cd_cl_g16 v[0:3], v[0:3], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_1D a16
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    ; return to shader part epilog
@@ -170,26 +153,21 @@
 define amdgpu_ps <4 x float> @sample_c_cd_cl_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %zcompare, half %dsdh, half %dtdh, half %dsdv, half %dtdv, half %s, half %t, half %clamp) {
 ; GFX9-LABEL: sample_c_cd_cl_2d:
 ; GFX9:       ; %bb.0: ; %main_body
+; GFX9-NEXT:    s_mov_b32 s12, 0x5040100
 ; GFX9-NEXT:    v_mov_b32_e32 v11, v7
 ; GFX9-NEXT:    v_mov_b32_e32 v7, v0
-; GFX9-NEXT:    v_and_b32_e32 v0, 0xffff, v5
-; GFX9-NEXT:    v_lshl_or_b32 v10, v6, 16, v0
-; GFX9-NEXT:    v_and_b32_e32 v0, 0xffff, v3
-; GFX9-NEXT:    v_lshl_or_b32 v9, v4, 16, v0
-; GFX9-NEXT:    v_and_b32_e32 v0, 0xffff, v1
-; GFX9-NEXT:    v_lshl_or_b32 v8, v2, 16, v0
+; GFX9-NEXT:    v_perm_b32 v10, v6, v5, s12
+; GFX9-NEXT:    v_perm_b32 v9, v4, v3, s12
+; GFX9-NEXT:    v_perm_b32 v8, v2, v1, s12
 ; GFX9-NEXT:    image_sample_c_cd_cl v[0:3], v[7:11], s[0:7], s[8:11] dmask:0xf a16
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    ; return to shader part epilog
 ;
 ; GFX10-LABEL: sample_c_cd_cl_2d:
 ; GFX10:       ; %bb.0: ; %main_body
-; GFX10-NEXT:    v_and_b32_e32 v5, 0xffff, v5
-; GFX10-NEXT:    v_and_b32_e32 v3, 0xffff, v3
-; GFX10-NEXT:    v_and_b32_e32 v1, 0xffff, v1
-; GFX10-NEXT:    v_lshl_or_b32 v5, v6, 16, v5
-; GFX10-NEXT:    v_lshl_or_b32 v3, v4, 16, v3
-; GFX10-NEXT:    v_lshl_or_b32 v1, v2, 16, v1
+; GFX10-NEXT:    v_perm_b32 v5, v6, v5, 0x5040100
+; GFX10-NEXT:    v_perm_b32 v3, v4, v3, 0x5040100
+; GFX10-NEXT:    v_perm_b32 v1, v2, v1, 0x5040100
 ; GFX10-NEXT:    image_sample_c_cd_cl_g16 v[0:3], [v0, v1, v3, v5, v7], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_2D a16
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    ; return to shader part epilog
diff --git a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.image.sample.cd.g16.encode.ll b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.image.sample.cd.g16.encode.ll
--- a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.image.sample.cd.g16.encode.ll
+++ b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.image.sample.cd.g16.encode.ll
@@ -15,10 +15,8 @@
 define amdgpu_ps <4 x float> @sample_cd_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %dsdh, half %dtdh, half %dsdv, half %dtdv, float %s, float %t) {
 ; GFX10-LABEL: sample_cd_2d:
 ; GFX10:       ; %bb.0: ; %main_body
-; GFX10-NEXT:    v_and_b32_e32 v2, 0xffff, v2 ; encoding: [0xff,0x04,0x04,0x36,0xff,0xff,0x00,0x00]
-; GFX10-NEXT:    v_and_b32_e32 v0, 0xffff, v0 ; encoding: [0xff,0x00,0x00,0x36,0xff,0xff,0x00,0x00]
-; GFX10-NEXT:    v_lshl_or_b32 v2, v3, 16, v2 ; encoding: [0x02,0x00,0x6f,0xd7,0x03,0x21,0x09,0x04]
-; GFX10-NEXT:    v_lshl_or_b32 v0, v1, 16, v0 ; encoding: [0x00,0x00,0x6f,0xd7,0x01,0x21,0x01,0x04]
+; GFX10-NEXT:    v_perm_b32 v2, v3, v2, 0x5040100 ; encoding: [0x02,0x00,0x44,0xd7,0x03,0x05,0xfe,0x03,0x00,0x01,0x04,0x05]
+; GFX10-NEXT:    v_perm_b32 v0, v1, v0, 0x5040100 ; encoding: [0x00,0x00,0x44,0xd7,0x01,0x01,0xfe,0x03,0x00,0x01,0x04,0x05]
 ; GFX10-NEXT:    image_sample_cd_g16 v[0:3], [v0, v2, v4, v5], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_2D ; encoding: [0x0b,0x0f,0xa0,0xf1,0x00,0x00,0x40,0x00,0x02,0x04,0x05,0x00]
 ; GFX10-NEXT:    s_waitcnt vmcnt(0) ; encoding: [0x70,0x3f,0x8c,0xbf]
 ; GFX10-NEXT:    ; return to shader part epilog
@@ -41,10 +39,8 @@
 define amdgpu_ps <4 x float> @sample_c_cd_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %zcompare, half %dsdh, half %dtdh, half %dsdv, half %dtdv, float %s, float %t) {
 ; GFX10-LABEL: sample_c_cd_2d:
 ; GFX10:       ; %bb.0: ; %main_body
-; GFX10-NEXT:    v_and_b32_e32 v3, 0xffff, v3 ; encoding: [0xff,0x06,0x06,0x36,0xff,0xff,0x00,0x00]
-; GFX10-NEXT:    v_and_b32_e32 v1, 0xffff, v1 ; encoding: [0xff,0x02,0x02,0x36,0xff,0xff,0x00,0x00]
-; GFX10-NEXT:    v_lshl_or_b32 v3, v4, 16, v3 ; encoding: [0x03,0x00,0x6f,0xd7,0x04,0x21,0x0d,0x04]
-; GFX10-NEXT:    v_lshl_or_b32 v1, v2, 16, v1 ; encoding: [0x01,0x00,0x6f,0xd7,0x02,0x21,0x05,0x04]
+; GFX10-NEXT:    v_perm_b32 v3, v4, v3, 0x5040100 ; encoding: [0x03,0x00,0x44,0xd7,0x04,0x07,0xfe,0x03,0x00,0x01,0x04,0x05]
+; GFX10-NEXT:    v_perm_b32 v1, v2, v1, 0x5040100 ; encoding: [0x01,0x00,0x44,0xd7,0x02,0x03,0xfe,0x03,0x00,0x01,0x04,0x05]
 ; GFX10-NEXT:    image_sample_c_cd_g16 v[0:3], [v0, v1, v3, v5, v6], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_2D ; encoding: [0x0b,0x0f,0xa8,0xf1,0x00,0x00,0x40,0x00,0x01,0x03,0x05,0x06]
 ; GFX10-NEXT:    s_waitcnt vmcnt(0) ; encoding: [0x70,0x3f,0x8c,0xbf]
 ; GFX10-NEXT:    ; return to shader part epilog
@@ -67,10 +63,8 @@
 define amdgpu_ps <4 x float> @sample_cd_cl_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %dsdh, half %dtdh, half %dsdv, half %dtdv, float %s, float %t, float %clamp) {
 ; GFX10-LABEL: sample_cd_cl_2d:
 ; GFX10:       ; %bb.0: ; %main_body
-; GFX10-NEXT:    v_and_b32_e32 v2, 0xffff, v2 ; encoding: [0xff,0x04,0x04,0x36,0xff,0xff,0x00,0x00]
-; GFX10-NEXT:    v_and_b32_e32 v0, 0xffff, v0 ; encoding: [0xff,0x00,0x00,0x36,0xff,0xff,0x00,0x00]
-; GFX10-NEXT:    v_lshl_or_b32 v2, v3, 16, v2 ; encoding: [0x02,0x00,0x6f,0xd7,0x03,0x21,0x09,0x04]
-; GFX10-NEXT:    v_lshl_or_b32 v0, v1, 16, v0 ; encoding: [0x00,0x00,0x6f,0xd7,0x01,0x21,0x01,0x04]
+; GFX10-NEXT:    v_perm_b32 v2, v3, v2, 0x5040100 ; encoding: [0x02,0x00,0x44,0xd7,0x03,0x05,0xfe,0x03,0x00,0x01,0x04,0x05]
+; GFX10-NEXT:    v_perm_b32 v0, v1, v0, 0x5040100 ; encoding: [0x00,0x00,0x44,0xd7,0x01,0x01,0xfe,0x03,0x00,0x01,0x04,0x05]
 ; GFX10-NEXT:    image_sample_cd_cl_g16 v[0:3], [v0, v2, v4, v5, v6], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_2D ; encoding: [0x0b,0x0f,0xa4,0xf1,0x00,0x00,0x40,0x00,0x02,0x04,0x05,0x06]
 ; GFX10-NEXT:    s_waitcnt vmcnt(0) ; encoding: [0x70,0x3f,0x8c,0xbf]
 ; GFX10-NEXT:    ; return to shader part epilog
@@ -95,10 +89,8 @@
 ; GFX10:       ; %bb.0: ; %main_body
 ; GFX10-NEXT:    v_mov_b32_e32 v8, v2 ; encoding: [0x02,0x03,0x10,0x7e]
 ; GFX10-NEXT:    v_mov_b32_e32 v2, v0 ; encoding: [0x00,0x03,0x04,0x7e]
-; GFX10-NEXT:    v_and_b32_e32 v0, 0xffff, v3 ; encoding: [0xff,0x06,0x00,0x36,0xff,0xff,0x00,0x00]
-; GFX10-NEXT:    v_and_b32_e32 v1, 0xffff, v1 ; encoding: [0xff,0x02,0x02,0x36,0xff,0xff,0x00,0x00]
-; GFX10-NEXT:    v_lshl_or_b32 v4, v4, 16, v0 ; encoding: [0x04,0x00,0x6f,0xd7,0x04,0x21,0x01,0x04]
-; GFX10-NEXT:    v_lshl_or_b32 v3, v8, 16, v1 ; encoding: [0x03,0x00,0x6f,0xd7,0x08,0x21,0x05,0x04]
+; GFX10-NEXT:    v_perm_b32 v4, v4, v3, 0x5040100 ; encoding: [0x04,0x00,0x44,0xd7,0x04,0x07,0xfe,0x03,0x00,0x01,0x04,0x05]
+; GFX10-NEXT:    v_perm_b32 v3, v8, v1, 0x5040100 ; encoding: [0x03,0x00,0x44,0xd7,0x08,0x03,0xfe,0x03,0x00,0x01,0x04,0x05]
 ; GFX10-NEXT:    image_sample_c_cd_cl_g16 v[0:3], v[2:7], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_2D ; encoding: [0x09,0x0f,0xac,0xf1,0x02,0x00,0x40,0x00]
 ; GFX10-NEXT:    s_waitcnt vmcnt(0) ; encoding: [0x70,0x3f,0x8c,0xbf]
 ; GFX10-NEXT:    ; return to shader part epilog
diff --git a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.image.sample.cd.g16.ll b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.image.sample.cd.g16.ll
--- a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.image.sample.cd.g16.ll
+++ b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.image.sample.cd.g16.ll
@@ -15,10 +15,8 @@
 define amdgpu_ps <4 x float> @sample_cd_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %dsdh, half %dtdh, half %dsdv, half %dtdv, float %s, float %t) {
 ; GFX10-LABEL: sample_cd_2d:
 ; GFX10:       ; %bb.0: ; %main_body
-; GFX10-NEXT:    v_and_b32_e32 v2, 0xffff, v2
-; GFX10-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX10-NEXT:    v_lshl_or_b32 v2, v3, 16, v2
-; GFX10-NEXT:    v_lshl_or_b32 v0, v1, 16, v0
+; GFX10-NEXT:    v_perm_b32 v2, v3, v2, 0x5040100
+; GFX10-NEXT:    v_perm_b32 v0, v1, v0, 0x5040100
 ; GFX10-NEXT:    image_sample_cd_g16 v[0:3], [v0, v2, v4, v5], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_2D
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    ; return to shader part epilog
@@ -41,10 +39,8 @@
 define amdgpu_ps <4 x float> @sample_c_cd_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %zcompare, half %dsdh, half %dtdh, half %dsdv, half %dtdv, float %s, float %t) {
 ; GFX10-LABEL: sample_c_cd_2d:
 ; GFX10:       ; %bb.0: ; %main_body
-; GFX10-NEXT:    v_and_b32_e32 v3, 0xffff, v3
-; GFX10-NEXT:    v_and_b32_e32 v1, 0xffff, v1
-; GFX10-NEXT:    v_lshl_or_b32 v3, v4, 16, v3
-; GFX10-NEXT:    v_lshl_or_b32 v1, v2, 16, v1
+; GFX10-NEXT:    v_perm_b32 v3, v4, v3, 0x5040100
+; GFX10-NEXT:    v_perm_b32 v1, v2, v1, 0x5040100
 ; GFX10-NEXT:    image_sample_c_cd_g16 v[0:3], [v0, v1, v3, v5, v6], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_2D
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    ; return to shader part epilog
@@ -67,10 +63,8 @@
 define amdgpu_ps <4 x float> @sample_cd_cl_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %dsdh, half %dtdh, half %dsdv, half %dtdv, float %s, float %t, float %clamp) {
 ; GFX10-LABEL: sample_cd_cl_2d:
 ; GFX10:       ; %bb.0: ; %main_body
-; GFX10-NEXT:    v_and_b32_e32 v2, 0xffff, v2
-; GFX10-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX10-NEXT:    v_lshl_or_b32 v2, v3, 16, v2
-; GFX10-NEXT:    v_lshl_or_b32 v0, v1, 16, v0
+; GFX10-NEXT:    v_perm_b32 v2, v3, v2, 0x5040100
+; GFX10-NEXT:    v_perm_b32 v0, v1, v0, 0x5040100
 ; GFX10-NEXT:    image_sample_cd_cl_g16 v[0:3], [v0, v2, v4, v5, v6], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_2D
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    ; return to shader part epilog
@@ -95,10 +89,8 @@
 ; GFX10:       ; %bb.0: ; %main_body
 ; GFX10-NEXT:    v_mov_b32_e32 v8, v2
 ; GFX10-NEXT:    v_mov_b32_e32 v2, v0
-; GFX10-NEXT:    v_and_b32_e32 v0, 0xffff, v3
-; GFX10-NEXT:    v_and_b32_e32 v1, 0xffff, v1
-; GFX10-NEXT:    v_lshl_or_b32 v4, v4, 16, v0
-; GFX10-NEXT:    v_lshl_or_b32 v3, v8, 16, v1
+; GFX10-NEXT:    v_perm_b32 v4, v4, v3, 0x5040100
+; GFX10-NEXT:    v_perm_b32 v3, v8, v1, 0x5040100
 ; GFX10-NEXT:    image_sample_c_cd_cl_g16 v[0:3], v[2:7], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_2D
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    ; return to shader part epilog
diff --git a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.image.sample.g16.a16.dim.ll b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.image.sample.g16.a16.dim.ll
--- a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.image.sample.g16.a16.dim.ll
+++ b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.image.sample.g16.a16.dim.ll
@@ -26,8 +26,7 @@
 define amdgpu_ps <4 x float> @sample_d_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %dsdh, float %dtdh, float %dsdv, float %dtdv, half %s, half %t) {
 ; GFX10-LABEL: sample_d_2d:
 ; GFX10:       ; %bb.0: ; %main_body
-; GFX10-NEXT:    v_and_b32_e32 v4, 0xffff, v4
-; GFX10-NEXT:    v_lshl_or_b32 v4, v5, 16, v4
+; GFX10-NEXT:    v_perm_b32 v4, v5, v4, 0x5040100
 ; GFX10-NEXT:    image_sample_d v[0:3], v[0:4], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_2D a16
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    ; return to shader part epilog
@@ -48,14 +47,13 @@
 ; GFX10-LABEL: sample_d_3d:
 ; GFX10:       ; %bb.0: ; %main_body
 ; GFX10-NEXT:    v_mov_b32_e32 v15, v8
-; GFX10-NEXT:    v_mov_b32_e32 v8, v0
-; GFX10-NEXT:    v_and_b32_e32 v0, 0xffff, v6
 ; GFX10-NEXT:    v_mov_b32_e32 v13, v5
 ; GFX10-NEXT:    v_mov_b32_e32 v12, v4
 ; GFX10-NEXT:    v_mov_b32_e32 v11, v3
 ; GFX10-NEXT:    v_mov_b32_e32 v10, v2
 ; GFX10-NEXT:    v_mov_b32_e32 v9, v1
-; GFX10-NEXT:    v_lshl_or_b32 v14, v7, 16, v0
+; GFX10-NEXT:    v_mov_b32_e32 v8, v0
+; GFX10-NEXT:    v_perm_b32 v14, v7, v6, 0x5040100
 ; GFX10-NEXT:    image_sample_d v[0:3], v[8:15], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_3D a16
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    ; return to shader part epilog
@@ -96,8 +94,7 @@
 define amdgpu_ps <4 x float> @sample_c_d_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %zcompare, float %dsdh, float %dtdh, float %dsdv, float %dtdv, half %s, half %t) {
 ; GFX10-LABEL: sample_c_d_2d:
 ; GFX10:       ; %bb.0: ; %main_body
-; GFX10-NEXT:    v_and_b32_e32 v5, 0xffff, v5
-; GFX10-NEXT:    v_lshl_or_b32 v5, v6, 16, v5
+; GFX10-NEXT:    v_perm_b32 v5, v6, v5, 0x5040100
 ; GFX10-NEXT:    image_sample_c_d v[0:3], v[0:5], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_2D a16
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    ; return to shader part epilog
@@ -117,8 +114,7 @@
 define amdgpu_ps <4 x float> @sample_d_cl_1d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %dsdh, float %dsdv, half %s, half %clamp) {
 ; GFX10-LABEL: sample_d_cl_1d:
 ; GFX10:       ; %bb.0: ; %main_body
-; GFX10-NEXT:    v_and_b32_e32 v2, 0xffff, v2
-; GFX10-NEXT:    v_lshl_or_b32 v2, v3, 16, v2
+; GFX10-NEXT:    v_perm_b32 v2, v3, v2, 0x5040100
 ; GFX10-NEXT:    image_sample_d_cl v[0:3], v[0:2], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_1D a16
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    ; return to shader part epilog
@@ -139,12 +135,11 @@
 ; GFX10-LABEL: sample_d_cl_2d:
 ; GFX10:       ; %bb.0: ; %main_body
 ; GFX10-NEXT:    v_mov_b32_e32 v11, v6
-; GFX10-NEXT:    v_mov_b32_e32 v6, v0
-; GFX10-NEXT:    v_and_b32_e32 v0, 0xffff, v4
 ; GFX10-NEXT:    v_mov_b32_e32 v9, v3
 ; GFX10-NEXT:    v_mov_b32_e32 v8, v2
 ; GFX10-NEXT:    v_mov_b32_e32 v7, v1
-; GFX10-NEXT:    v_lshl_or_b32 v10, v5, 16, v0
+; GFX10-NEXT:    v_mov_b32_e32 v6, v0
+; GFX10-NEXT:    v_perm_b32 v10, v5, v4, 0x5040100
 ; GFX10-NEXT:    image_sample_d_cl v[0:3], v[6:11], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_2D a16
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    ; return to shader part epilog
@@ -166,8 +161,7 @@
 define amdgpu_ps <4 x float> @sample_c_d_cl_1d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %zcompare, float %dsdh, float %dsdv, half %s, half %clamp) {
 ; GFX10-LABEL: sample_c_d_cl_1d:
 ; GFX10:       ; %bb.0: ; %main_body
-; GFX10-NEXT:    v_and_b32_e32 v3, 0xffff, v3
-; GFX10-NEXT:    v_lshl_or_b32 v3, v4, 16, v3
+; GFX10-NEXT:    v_perm_b32 v3, v4, v3, 0x5040100
 ; GFX10-NEXT:    image_sample_c_d_cl v[0:3], v[0:3], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_1D a16
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    ; return to shader part epilog
@@ -188,13 +182,12 @@
 ; GFX10-LABEL: sample_c_d_cl_2d:
 ; GFX10:       ; %bb.0: ; %main_body
 ; GFX10-NEXT:    v_mov_b32_e32 v13, v7
-; GFX10-NEXT:    v_mov_b32_e32 v7, v0
-; GFX10-NEXT:    v_and_b32_e32 v0, 0xffff, v5
 ; GFX10-NEXT:    v_mov_b32_e32 v11, v4
 ; GFX10-NEXT:    v_mov_b32_e32 v10, v3
 ; GFX10-NEXT:    v_mov_b32_e32 v9, v2
 ; GFX10-NEXT:    v_mov_b32_e32 v8, v1
-; GFX10-NEXT:    v_lshl_or_b32 v12, v6, 16, v0
+; GFX10-NEXT:    v_mov_b32_e32 v7, v0
+; GFX10-NEXT:    v_perm_b32 v12, v6, v5, 0x5040100
 ; GFX10-NEXT:    image_sample_c_d_cl v[0:3], v[7:13], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_2D a16
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    ; return to shader part epilog
@@ -235,8 +228,7 @@
 define amdgpu_ps <4 x float> @sample_cd_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %dsdh, float %dtdh, float %dsdv, float %dtdv, half %s, half %t) {
 ; GFX10-LABEL: sample_cd_2d:
 ; GFX10:       ; %bb.0: ; %main_body
-; GFX10-NEXT:    v_and_b32_e32 v4, 0xffff, v4
-; GFX10-NEXT:    v_lshl_or_b32 v4, v5, 16, v4
+; GFX10-NEXT:    v_perm_b32 v4, v5, v4, 0x5040100
 ; GFX10-NEXT:    image_sample_cd v[0:3], v[0:4], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_2D a16
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    ; return to shader part epilog
@@ -275,8 +267,7 @@
 define amdgpu_ps <4 x float> @sample_c_cd_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %zcompare, float %dsdh, float %dtdh, float %dsdv, float %dtdv, half %s, half %t) {
 ; GFX10-LABEL: sample_c_cd_2d:
 ; GFX10:       ; %bb.0: ; %main_body
-; GFX10-NEXT:    v_and_b32_e32 v5, 0xffff, v5
-; GFX10-NEXT:    v_lshl_or_b32 v5, v6, 16, v5
+; GFX10-NEXT:    v_perm_b32 v5, v6, v5, 0x5040100
 ; GFX10-NEXT:    image_sample_c_cd v[0:3], v[0:5], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_2D a16
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    ; return to shader part epilog
@@ -296,8 +287,7 @@
 define amdgpu_ps <4 x float> @sample_cd_cl_1d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %dsdh, float %dsdv, half %s, half %clamp) {
 ; GFX10-LABEL: sample_cd_cl_1d:
 ; GFX10:       ; %bb.0: ; %main_body
-; GFX10-NEXT:    v_and_b32_e32 v2, 0xffff, v2
-; GFX10-NEXT:    v_lshl_or_b32 v2, v3, 16, v2
+; GFX10-NEXT:    v_perm_b32 v2, v3, v2, 0x5040100
 ; GFX10-NEXT:    image_sample_cd_cl v[0:3], v[0:2], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_1D a16
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    ; return to shader part epilog
@@ -318,12 +308,11 @@
 ; GFX10-LABEL: sample_cd_cl_2d:
 ; GFX10:       ; %bb.0: ; %main_body
 ; GFX10-NEXT:    v_mov_b32_e32 v11, v6
-; GFX10-NEXT:    v_mov_b32_e32 v6, v0
-; GFX10-NEXT:    v_and_b32_e32 v0, 0xffff, v4
 ; GFX10-NEXT:    v_mov_b32_e32 v9, v3
 ; GFX10-NEXT:    v_mov_b32_e32 v8, v2
 ; GFX10-NEXT:    v_mov_b32_e32 v7, v1
-; GFX10-NEXT:    v_lshl_or_b32 v10, v5, 16, v0
+; GFX10-NEXT:    v_mov_b32_e32 v6, v0
+; GFX10-NEXT:    v_perm_b32 v10, v5, v4, 0x5040100
 ; GFX10-NEXT:    image_sample_cd_cl v[0:3], v[6:11], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_2D a16
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    ; return to shader part epilog
@@ -345,8 +334,7 @@
 define amdgpu_ps <4 x float> @sample_c_cd_cl_1d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %zcompare, float %dsdh, float %dsdv, half %s, half %clamp) {
 ; GFX10-LABEL: sample_c_cd_cl_1d:
 ; GFX10:       ; %bb.0: ; %main_body
-; GFX10-NEXT:    v_and_b32_e32 v3, 0xffff, v3
-; GFX10-NEXT:    v_lshl_or_b32 v3, v4, 16, v3
+; GFX10-NEXT:    v_perm_b32 v3, v4, v3, 0x5040100
 ; GFX10-NEXT:    image_sample_c_cd_cl v[0:3], v[0:3], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_1D a16
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    ; return to shader part epilog
@@ -367,13 +355,12 @@
 ; GFX10-LABEL: sample_c_cd_cl_2d:
 ; GFX10:       ; %bb.0: ; %main_body
 ; GFX10-NEXT:    v_mov_b32_e32 v13, v7
-; GFX10-NEXT:    v_mov_b32_e32 v7, v0
-; GFX10-NEXT:    v_and_b32_e32 v0, 0xffff, v5
 ; GFX10-NEXT:    v_mov_b32_e32 v11, v4
 ; GFX10-NEXT:    v_mov_b32_e32 v10, v3
 ; GFX10-NEXT:    v_mov_b32_e32 v9, v2
 ; GFX10-NEXT:    v_mov_b32_e32 v8, v1
-; GFX10-NEXT:    v_lshl_or_b32 v12, v6, 16, v0
+; GFX10-NEXT:    v_mov_b32_e32 v7, v0
+; GFX10-NEXT:    v_perm_b32 v12, v6, v5, 0x5040100
 ; GFX10-NEXT:    image_sample_c_cd_cl v[0:3], v[7:13], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_2D a16
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    ; return to shader part epilog
@@ -396,14 +383,13 @@
 ; GFX10-LABEL: sample_c_d_o_2darray_V1:
 ; GFX10:       ; %bb.0: ; %main_body
 ; GFX10-NEXT:    v_mov_b32_e32 v15, v8
-; GFX10-NEXT:    v_mov_b32_e32 v8, v0
-; GFX10-NEXT:    v_and_b32_e32 v0, 0xffff, v6
 ; GFX10-NEXT:    v_mov_b32_e32 v13, v5
 ; GFX10-NEXT:    v_mov_b32_e32 v12, v4
 ; GFX10-NEXT:    v_mov_b32_e32 v11, v3
 ; GFX10-NEXT:    v_mov_b32_e32 v10, v2
 ; GFX10-NEXT:    v_mov_b32_e32 v9, v1
-; GFX10-NEXT:    v_lshl_or_b32 v14, v7, 16, v0
+; GFX10-NEXT:    v_mov_b32_e32 v8, v0
+; GFX10-NEXT:    v_perm_b32 v14, v7, v6, 0x5040100
 ; GFX10-NEXT:    image_sample_c_d_o v0, v[8:15], s[0:7], s[8:11] dmask:0x4 dim:SQ_RSRC_IMG_2D_ARRAY a16
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    ; return to shader part epilog
@@ -426,14 +412,13 @@
 ; GFX10-LABEL: sample_c_d_o_2darray_V2:
 ; GFX10:       ; %bb.0: ; %main_body
 ; GFX10-NEXT:    v_mov_b32_e32 v15, v8
-; GFX10-NEXT:    v_mov_b32_e32 v8, v0
-; GFX10-NEXT:    v_and_b32_e32 v0, 0xffff, v6
 ; GFX10-NEXT:    v_mov_b32_e32 v13, v5
 ; GFX10-NEXT:    v_mov_b32_e32 v12, v4
 ; GFX10-NEXT:    v_mov_b32_e32 v11, v3
 ; GFX10-NEXT:    v_mov_b32_e32 v10, v2
 ; GFX10-NEXT:    v_mov_b32_e32 v9, v1
-; GFX10-NEXT:    v_lshl_or_b32 v14, v7, 16, v0
+; GFX10-NEXT:    v_mov_b32_e32 v8, v0
+; GFX10-NEXT:    v_perm_b32 v14, v7, v6, 0x5040100
 ; GFX10-NEXT:    image_sample_c_d_o v[0:1], v[8:15], s[0:7], s[8:11] dmask:0x6 dim:SQ_RSRC_IMG_2D_ARRAY a16
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    ; return to shader part epilog
@@ -497,10 +482,8 @@
 define amdgpu_ps <4 x float> @sample_g16_noa16_d_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %dsdh, half %dtdh, half %dsdv, half %dtdv, float %s, float %t) {
 ; GFX10-LABEL: sample_g16_noa16_d_2d:
 ; GFX10:       ; %bb.0: ; %main_body
-; GFX10-NEXT:    v_and_b32_e32 v2, 0xffff, v2
-; GFX10-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX10-NEXT:    v_lshl_or_b32 v2, v3, 16, v2
-; GFX10-NEXT:    v_lshl_or_b32 v0, v1, 16, v0
+; GFX10-NEXT:    v_perm_b32 v2, v3, v2, 0x5040100
+; GFX10-NEXT:    v_perm_b32 v0, v1, v0, 0x5040100
 ; GFX10-NEXT:    image_sample_d_g16 v[0:3], [v0, v2, v4, v5], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_2D
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    ; return to shader part epilog
@@ -524,10 +507,8 @@
 ; GFX10:       ; %bb.0: ; %main_body
 ; GFX10-NEXT:    v_mov_b32_e32 v9, v3
 ; GFX10-NEXT:    v_mov_b32_e32 v3, v2
-; GFX10-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX10-NEXT:    v_and_b32_e32 v2, 0xffff, v9
-; GFX10-NEXT:    v_lshl_or_b32 v4, v4, 16, v2
-; GFX10-NEXT:    v_lshl_or_b32 v2, v1, 16, v0
+; GFX10-NEXT:    v_perm_b32 v2, v1, v0, 0x5040100
+; GFX10-NEXT:    v_perm_b32 v4, v4, v9, 0x5040100
 ; GFX10-NEXT:    image_sample_d_g16 v[0:3], v[2:8], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_3D
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    ; return to shader part epilog
@@ -574,10 +555,8 @@
 define amdgpu_ps <4 x float> @sample_g16_noa16_c_d_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %zcompare, half %dsdh, half %dtdh, half %dsdv, half %dtdv, float %s, float %t) {
 ; GFX10-LABEL: sample_g16_noa16_c_d_2d:
 ; GFX10:       ; %bb.0: ; %main_body
-; GFX10-NEXT:    v_and_b32_e32 v3, 0xffff, v3
-; GFX10-NEXT:    v_and_b32_e32 v1, 0xffff, v1
-; GFX10-NEXT:    v_lshl_or_b32 v3, v4, 16, v3
-; GFX10-NEXT:    v_lshl_or_b32 v1, v2, 16, v1
+; GFX10-NEXT:    v_perm_b32 v3, v4, v3, 0x5040100
+; GFX10-NEXT:    v_perm_b32 v1, v2, v1, 0x5040100
 ; GFX10-NEXT:    image_sample_c_d_g16 v[0:3], [v0, v1, v3, v5, v6], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_2D
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    ; return to shader part epilog
@@ -619,10 +598,8 @@
 define amdgpu_ps <4 x float> @sample_g16_noa16_d_cl_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %dsdh, half %dtdh, half %dsdv, half %dtdv, float %s, float %t, float %clamp) {
 ; GFX10-LABEL: sample_g16_noa16_d_cl_2d:
 ; GFX10:       ; %bb.0: ; %main_body
-; GFX10-NEXT:    v_and_b32_e32 v2, 0xffff, v2
-; GFX10-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX10-NEXT:    v_lshl_or_b32 v2, v3, 16, v2
-; GFX10-NEXT:    v_lshl_or_b32 v0, v1, 16, v0
+; GFX10-NEXT:    v_perm_b32 v2, v3, v2, 0x5040100
+; GFX10-NEXT:    v_perm_b32 v0, v1, v0, 0x5040100
 ; GFX10-NEXT:    image_sample_d_cl_g16 v[0:3], [v0, v2, v4, v5, v6], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_2D
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    ; return to shader part epilog
@@ -666,10 +643,8 @@
 ; GFX10:       ; %bb.0: ; %main_body
 ; GFX10-NEXT:    v_mov_b32_e32 v8, v2
 ; GFX10-NEXT:    v_mov_b32_e32 v2, v0
-; GFX10-NEXT:    v_and_b32_e32 v0, 0xffff, v3
-; GFX10-NEXT:    v_and_b32_e32 v1, 0xffff, v1
-; GFX10-NEXT:    v_lshl_or_b32 v4, v4, 16, v0
-; GFX10-NEXT:    v_lshl_or_b32 v3, v8, 16, v1
+; GFX10-NEXT:    v_perm_b32 v4, v4, v3, 0x5040100
+; GFX10-NEXT:    v_perm_b32 v3, v8, v1, 0x5040100
 ; GFX10-NEXT:    image_sample_c_d_cl_g16 v[0:3], v[2:7], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_2D
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    ; return to shader part epilog
@@ -714,10 +689,8 @@
 define amdgpu_ps <4 x float> @sample_g16_noa16_cd_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %dsdh, half %dtdh, half %dsdv, half %dtdv, float %s, float %t) {
 ; GFX10-LABEL: sample_g16_noa16_cd_2d:
 ; GFX10:       ; %bb.0: ; %main_body
-; GFX10-NEXT:    v_and_b32_e32 v2, 0xffff, v2
-; GFX10-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX10-NEXT:    v_lshl_or_b32 v2, v3, 16, v2
-; GFX10-NEXT:    v_lshl_or_b32 v0, v1, 16, v0
+; GFX10-NEXT:    v_perm_b32 v2, v3, v2, 0x5040100
+; GFX10-NEXT:    v_perm_b32 v0, v1, v0, 0x5040100
 ; GFX10-NEXT:    image_sample_cd_g16 v[0:3], [v0, v2, v4, v5], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_2D
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    ; return to shader part epilog
@@ -759,10 +732,8 @@
 define amdgpu_ps <4 x float> @sample_g16_noa16_c_cd_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %zcompare, half %dsdh, half %dtdh, half %dsdv, half %dtdv, float %s, float %t) {
 ; GFX10-LABEL: sample_g16_noa16_c_cd_2d:
 ; GFX10:       ; %bb.0: ; %main_body
-; GFX10-NEXT:    v_and_b32_e32 v3, 0xffff, v3
-; GFX10-NEXT:    v_and_b32_e32 v1, 0xffff, v1
-; GFX10-NEXT:    v_lshl_or_b32 v3, v4, 16, v3
-; GFX10-NEXT:    v_lshl_or_b32 v1, v2, 16, v1
+; GFX10-NEXT:    v_perm_b32 v3, v4, v3, 0x5040100
+; GFX10-NEXT:    v_perm_b32 v1, v2, v1, 0x5040100
 ; GFX10-NEXT:    image_sample_c_cd_g16 v[0:3], [v0, v1, v3, v5, v6], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_2D
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    ; return to shader part epilog
@@ -804,10 +775,8 @@
 define amdgpu_ps <4 x float> @sample_g16_noa16_cd_cl_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %dsdh, half %dtdh, half %dsdv, half %dtdv, float %s, float %t, float %clamp) {
 ; GFX10-LABEL: sample_g16_noa16_cd_cl_2d:
 ; GFX10:       ; %bb.0: ; %main_body
-; GFX10-NEXT:    v_and_b32_e32 v2, 0xffff, v2
-; GFX10-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX10-NEXT:    v_lshl_or_b32 v2, v3, 16, v2
-; GFX10-NEXT:    v_lshl_or_b32 v0, v1, 16, v0
+; GFX10-NEXT:    v_perm_b32 v2, v3, v2, 0x5040100
+; GFX10-NEXT:    v_perm_b32 v0, v1, v0, 0x5040100
 ; GFX10-NEXT:    image_sample_cd_cl_g16 v[0:3], [v0, v2, v4, v5, v6], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_2D
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    ; return to shader part epilog
@@ -851,10 +820,8 @@
 ; GFX10:       ; %bb.0: ; %main_body
 ; GFX10-NEXT:    v_mov_b32_e32 v8, v2
 ; GFX10-NEXT:    v_mov_b32_e32 v2, v0
-; GFX10-NEXT:    v_and_b32_e32 v0, 0xffff, v3
-; GFX10-NEXT:    v_and_b32_e32 v1, 0xffff, v1
-; GFX10-NEXT:    v_lshl_or_b32 v4, v4, 16, v0
-; GFX10-NEXT:    v_lshl_or_b32 v3, v8, 16, v1
+; GFX10-NEXT:    v_perm_b32 v4, v4, v3, 0x5040100
+; GFX10-NEXT:    v_perm_b32 v3, v8, v1, 0x5040100
 ; GFX10-NEXT:    image_sample_c_cd_cl_g16 v[0:3], v[2:7], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_2D
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    ; return to shader part epilog
@@ -879,14 +846,12 @@
 define amdgpu_ps float @sample_g16_noa16_c_d_o_2darray_V1(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, i32 %offset, float %zcompare, half %dsdh, half %dtdh, half %dsdv, half %dtdv, float %s, float %t, float %slice) {
 ; GFX10-LABEL: sample_g16_noa16_c_d_o_2darray_V1:
 ; GFX10:       ; %bb.0: ; %main_body
-; GFX10-NEXT:    v_mov_b32_e32 v9, v2
-; GFX10-NEXT:    v_mov_b32_e32 v10, v3
+; GFX10-NEXT:    v_mov_b32_e32 v9, v3
+; GFX10-NEXT:    v_mov_b32_e32 v10, v2
 ; GFX10-NEXT:    v_mov_b32_e32 v3, v1
 ; GFX10-NEXT:    v_mov_b32_e32 v2, v0
-; GFX10-NEXT:    v_and_b32_e32 v0, 0xffff, v4
-; GFX10-NEXT:    v_and_b32_e32 v1, 0xffff, v9
-; GFX10-NEXT:    v_lshl_or_b32 v5, v5, 16, v0
-; GFX10-NEXT:    v_lshl_or_b32 v4, v10, 16, v1
+; GFX10-NEXT:    v_perm_b32 v5, v5, v4, 0x5040100
+; GFX10-NEXT:    v_perm_b32 v4, v9, v10, 0x5040100
 ; GFX10-NEXT:    image_sample_c_d_o_g16 v0, v[2:8], s[0:7], s[8:11] dmask:0x4 dim:SQ_RSRC_IMG_2D_ARRAY
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    ; return to shader part epilog
@@ -913,14 +878,12 @@
 define amdgpu_ps <2 x float> @sample_g16_noa16_c_d_o_2darray_V2(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, i32 %offset, float %zcompare, half %dsdh, half %dtdh, half %dsdv, half %dtdv, float %s, float %t, float %slice) {
 ; GFX10-LABEL: sample_g16_noa16_c_d_o_2darray_V2:
 ; GFX10:       ; %bb.0: ; %main_body
-; GFX10-NEXT:    v_mov_b32_e32 v9, v2
-; GFX10-NEXT:    v_mov_b32_e32 v10, v3
+; GFX10-NEXT:    v_mov_b32_e32 v9, v3
+; GFX10-NEXT:    v_mov_b32_e32 v10, v2
 ; GFX10-NEXT:    v_mov_b32_e32 v3, v1
 ; GFX10-NEXT:    v_mov_b32_e32 v2, v0
-; GFX10-NEXT:    v_and_b32_e32 v0, 0xffff, v4
-; GFX10-NEXT:    v_and_b32_e32 v1, 0xffff, v9
-; GFX10-NEXT:    v_lshl_or_b32 v5, v5, 16, v0
-; GFX10-NEXT:    v_lshl_or_b32 v4, v10, 16, v1
+; GFX10-NEXT:    v_perm_b32 v5, v5, v4, 0x5040100
+; GFX10-NEXT:    v_perm_b32 v4, v9, v10, 0x5040100
 ; GFX10-NEXT:    image_sample_c_d_o_g16 v[0:1], v[2:8], s[0:7], s[8:11] dmask:0x6 dim:SQ_RSRC_IMG_2D_ARRAY
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    ; return to shader part epilog
diff --git a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.image.sample.g16.encode.ll b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.image.sample.g16.encode.ll
--- a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.image.sample.g16.encode.ll
+++ b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.image.sample.g16.encode.ll
@@ -22,21 +22,16 @@
 define amdgpu_ps <4 x float> @sample_d_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %dsdh, half %dtdh, half %dsdv, half %dtdv, float %s, float %t) {
 ; GFX10-LABEL: sample_d_2d:
 ; GFX10:       ; %bb.0: ; %main_body
-; GFX10-NEXT:    v_and_b32_e32 v2, 0xffff, v2 ; encoding: [0xff,0x04,0x04,0x36,0xff,0xff,0x00,0x00]
-; GFX10-NEXT:    v_and_b32_e32 v0, 0xffff, v0 ; encoding: [0xff,0x00,0x00,0x36,0xff,0xff,0x00,0x00]
-; GFX10-NEXT:    v_lshl_or_b32 v2, v3, 16, v2 ; encoding: [0x02,0x00,0x6f,0xd7,0x03,0x21,0x09,0x04]
-; GFX10-NEXT:    v_lshl_or_b32 v0, v1, 16, v0 ; encoding: [0x00,0x00,0x6f,0xd7,0x01,0x21,0x01,0x04]
+; GFX10-NEXT:    v_perm_b32 v2, v3, v2, 0x5040100 ; encoding: [0x02,0x00,0x44,0xd7,0x03,0x05,0xfe,0x03,0x00,0x01,0x04,0x05]
+; GFX10-NEXT:    v_perm_b32 v0, v1, v0, 0x5040100 ; encoding: [0x00,0x00,0x44,0xd7,0x01,0x01,0xfe,0x03,0x00,0x01,0x04,0x05]
 ; GFX10-NEXT:    image_sample_d_g16 v[0:3], [v0, v2, v4, v5], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_2D ; encoding: [0x0b,0x0f,0x88,0xf0,0x00,0x00,0x40,0x00,0x02,0x04,0x05,0x00]
 ; GFX10-NEXT:    s_waitcnt vmcnt(0) ; encoding: [0x70,0x3f,0x8c,0xbf]
 ; GFX10-NEXT:    ; return to shader part epilog
 ;
 ; GFX11-LABEL: sample_d_2d:
 ; GFX11:       ; %bb.0: ; %main_body
-; GFX11-NEXT:    v_and_b32_e32 v2, 0xffff, v2 ; encoding: [0xff,0x04,0x04,0x36,0xff,0xff,0x00,0x00]
-; GFX11-NEXT:    v_and_b32_e32 v0, 0xffff, v0 ; encoding: [0xff,0x00,0x00,0x36,0xff,0xff,0x00,0x00]
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) ; encoding: [0x12,0x01,0x87,0xbf]
-; GFX11-NEXT:    v_lshl_or_b32 v2, v3, 16, v2 ; encoding: [0x02,0x00,0x56,0xd6,0x03,0x21,0x09,0x04]
-; GFX11-NEXT:    v_lshl_or_b32 v0, v1, 16, v0 ; encoding: [0x00,0x00,0x56,0xd6,0x01,0x21,0x01,0x04]
+; GFX11-NEXT:    v_perm_b32 v2, v3, v2, 0x5040100 ; encoding: [0x02,0x00,0x44,0xd6,0x03,0x05,0xfe,0x03,0x00,0x01,0x04,0x05]
+; GFX11-NEXT:    v_perm_b32 v0, v1, v0, 0x5040100 ; encoding: [0x00,0x00,0x44,0xd6,0x01,0x01,0xfe,0x03,0x00,0x01,0x04,0x05]
 ; GFX11-NEXT:    image_sample_d_g16 v[0:3], [v0, v2, v4, v5], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_2D ; encoding: [0x05,0x0f,0xe4,0xf0,0x00,0x00,0x00,0x08,0x02,0x04,0x05,0x00]
 ; GFX11-NEXT:    s_waitcnt vmcnt(0) ; encoding: [0xf7,0x03,0x89,0xbf]
 ; GFX11-NEXT:    ; return to shader part epilog
@@ -50,10 +45,8 @@
 ; GFX10:       ; %bb.0: ; %main_body
 ; GFX10-NEXT:    v_mov_b32_e32 v9, v3 ; encoding: [0x03,0x03,0x12,0x7e]
 ; GFX10-NEXT:    v_mov_b32_e32 v3, v2 ; encoding: [0x02,0x03,0x06,0x7e]
-; GFX10-NEXT:    v_and_b32_e32 v0, 0xffff, v0 ; encoding: [0xff,0x00,0x00,0x36,0xff,0xff,0x00,0x00]
-; GFX10-NEXT:    v_and_b32_e32 v2, 0xffff, v9 ; encoding: [0xff,0x12,0x04,0x36,0xff,0xff,0x00,0x00]
-; GFX10-NEXT:    v_lshl_or_b32 v4, v4, 16, v2 ; encoding: [0x04,0x00,0x6f,0xd7,0x04,0x21,0x09,0x04]
-; GFX10-NEXT:    v_lshl_or_b32 v2, v1, 16, v0 ; encoding: [0x02,0x00,0x6f,0xd7,0x01,0x21,0x01,0x04]
+; GFX10-NEXT:    v_perm_b32 v2, v1, v0, 0x5040100 ; encoding: [0x02,0x00,0x44,0xd7,0x01,0x01,0xfe,0x03,0x00,0x01,0x04,0x05]
+; GFX10-NEXT:    v_perm_b32 v4, v4, v9, 0x5040100 ; encoding: [0x04,0x00,0x44,0xd7,0x04,0x13,0xfe,0x03,0x00,0x01,0x04,0x05]
 ; GFX10-NEXT:    image_sample_d_g16 v[0:3], v[2:8], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_3D ; encoding: [0x11,0x0f,0x88,0xf0,0x02,0x00,0x40,0x00]
 ; GFX10-NEXT:    s_waitcnt vmcnt(0) ; encoding: [0x70,0x3f,0x8c,0xbf]
 ; GFX10-NEXT:    ; return to shader part epilog
@@ -62,12 +55,9 @@
 ; GFX11:       ; %bb.0: ; %main_body
 ; GFX11-NEXT:    v_mov_b32_e32 v9, v3 ; encoding: [0x03,0x03,0x12,0x7e]
 ; GFX11-NEXT:    v_mov_b32_e32 v3, v2 ; encoding: [0x02,0x03,0x06,0x7e]
-; GFX11-NEXT:    v_and_b32_e32 v0, 0xffff, v0 ; encoding: [0xff,0x00,0x00,0x36,0xff,0xff,0x00,0x00]
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) ; encoding: [0x93,0x00,0x87,0xbf]
-; GFX11-NEXT:    v_and_b32_e32 v2, 0xffff, v9 ; encoding: [0xff,0x12,0x04,0x36,0xff,0xff,0x00,0x00]
-; GFX11-NEXT:    v_lshl_or_b32 v4, v4, 16, v2 ; encoding: [0x04,0x00,0x56,0xd6,0x04,0x21,0x09,0x04]
+; GFX11-NEXT:    v_perm_b32 v2, v1, v0, 0x5040100 ; encoding: [0x02,0x00,0x44,0xd6,0x01,0x01,0xfe,0x03,0x00,0x01,0x04,0x05]
 ; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_3) ; encoding: [0x03,0x00,0x87,0xbf]
-; GFX11-NEXT:    v_lshl_or_b32 v2, v1, 16, v0 ; encoding: [0x02,0x00,0x56,0xd6,0x01,0x21,0x01,0x04]
+; GFX11-NEXT:    v_perm_b32 v4, v4, v9, 0x5040100 ; encoding: [0x04,0x00,0x44,0xd6,0x04,0x13,0xfe,0x03,0x00,0x01,0x04,0x05]
 ; GFX11-NEXT:    image_sample_d_g16 v[0:3], v[2:8], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_3D ; encoding: [0x08,0x0f,0xe4,0xf0,0x02,0x00,0x00,0x08]
 ; GFX11-NEXT:    s_waitcnt vmcnt(0) ; encoding: [0xf7,0x03,0x89,0xbf]
 ; GFX11-NEXT:    ; return to shader part epilog
@@ -96,21 +86,16 @@
 define amdgpu_ps <4 x float> @sample_c_d_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %zcompare, half %dsdh, half %dtdh, half %dsdv, half %dtdv, float %s, float %t) {
 ; GFX10-LABEL: sample_c_d_2d:
 ; GFX10:       ; %bb.0: ; %main_body
-; GFX10-NEXT:    v_and_b32_e32 v3, 0xffff, v3 ; encoding: [0xff,0x06,0x06,0x36,0xff,0xff,0x00,0x00]
-; GFX10-NEXT:    v_and_b32_e32 v1, 0xffff, v1 ; encoding: [0xff,0x02,0x02,0x36,0xff,0xff,0x00,0x00]
-; GFX10-NEXT:    v_lshl_or_b32 v3, v4, 16, v3 ; encoding: [0x03,0x00,0x6f,0xd7,0x04,0x21,0x0d,0x04]
-; GFX10-NEXT:    v_lshl_or_b32 v1, v2, 16, v1 ; encoding: [0x01,0x00,0x6f,0xd7,0x02,0x21,0x05,0x04]
+; GFX10-NEXT:    v_perm_b32 v3, v4, v3, 0x5040100 ; encoding: [0x03,0x00,0x44,0xd7,0x04,0x07,0xfe,0x03,0x00,0x01,0x04,0x05]
+; GFX10-NEXT:    v_perm_b32 v1, v2, v1, 0x5040100 ; encoding: [0x01,0x00,0x44,0xd7,0x02,0x03,0xfe,0x03,0x00,0x01,0x04,0x05]
 ; GFX10-NEXT:    image_sample_c_d_g16 v[0:3], [v0, v1, v3, v5, v6], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_2D ; encoding: [0x0b,0x0f,0xa8,0xf0,0x00,0x00,0x40,0x00,0x01,0x03,0x05,0x06]
 ; GFX10-NEXT:    s_waitcnt vmcnt(0) ; encoding: [0x70,0x3f,0x8c,0xbf]
 ; GFX10-NEXT:    ; return to shader part epilog
 ;
 ; GFX11-LABEL: sample_c_d_2d:
 ; GFX11:       ; %bb.0: ; %main_body
-; GFX11-NEXT:    v_and_b32_e32 v3, 0xffff, v3 ; encoding: [0xff,0x06,0x06,0x36,0xff,0xff,0x00,0x00]
-; GFX11-NEXT:    v_and_b32_e32 v1, 0xffff, v1 ; encoding: [0xff,0x02,0x02,0x36,0xff,0xff,0x00,0x00]
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) ; encoding: [0x12,0x01,0x87,0xbf]
-; GFX11-NEXT:    v_lshl_or_b32 v3, v4, 16, v3 ; encoding: [0x03,0x00,0x56,0xd6,0x04,0x21,0x0d,0x04]
-; GFX11-NEXT:    v_lshl_or_b32 v1, v2, 16, v1 ; encoding: [0x01,0x00,0x56,0xd6,0x02,0x21,0x05,0x04]
+; GFX11-NEXT:    v_perm_b32 v3, v4, v3, 0x5040100 ; encoding: [0x03,0x00,0x44,0xd6,0x04,0x07,0xfe,0x03,0x00,0x01,0x04,0x05]
+; GFX11-NEXT:    v_perm_b32 v1, v2, v1, 0x5040100 ; encoding: [0x01,0x00,0x44,0xd6,0x02,0x03,0xfe,0x03,0x00,0x01,0x04,0x05]
 ; GFX11-NEXT:    image_sample_c_d_g16 v[0:3], [v0, v1, v3, v5, v6], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_2D ; encoding: [0x05,0x0f,0xe8,0xf0,0x00,0x00,0x00,0x08,0x01,0x03,0x05,0x06]
 ; GFX11-NEXT:    s_waitcnt vmcnt(0) ; encoding: [0xf7,0x03,0x89,0xbf]
 ; GFX11-NEXT:    ; return to shader part epilog
@@ -139,21 +124,16 @@
 define amdgpu_ps <4 x float> @sample_d_cl_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %dsdh, half %dtdh, half %dsdv, half %dtdv, float %s, float %t, float %clamp) {
 ; GFX10-LABEL: sample_d_cl_2d:
 ; GFX10:       ; %bb.0: ; %main_body
-; GFX10-NEXT:    v_and_b32_e32 v2, 0xffff, v2 ; encoding: [0xff,0x04,0x04,0x36,0xff,0xff,0x00,0x00]
-; GFX10-NEXT:    v_and_b32_e32 v0, 0xffff, v0 ; encoding: [0xff,0x00,0x00,0x36,0xff,0xff,0x00,0x00]
-; GFX10-NEXT:    v_lshl_or_b32 v2, v3, 16, v2 ; encoding: [0x02,0x00,0x6f,0xd7,0x03,0x21,0x09,0x04]
-; GFX10-NEXT:    v_lshl_or_b32 v0, v1, 16, v0 ; encoding: [0x00,0x00,0x6f,0xd7,0x01,0x21,0x01,0x04]
+; GFX10-NEXT:    v_perm_b32 v2, v3, v2, 0x5040100 ; encoding: [0x02,0x00,0x44,0xd7,0x03,0x05,0xfe,0x03,0x00,0x01,0x04,0x05]
+; GFX10-NEXT:    v_perm_b32 v0, v1, v0, 0x5040100 ; encoding: [0x00,0x00,0x44,0xd7,0x01,0x01,0xfe,0x03,0x00,0x01,0x04,0x05]
 ; GFX10-NEXT:    image_sample_d_cl_g16 v[0:3], [v0, v2, v4, v5, v6], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_2D ; encoding: [0x0b,0x0f,0x8c,0xf0,0x00,0x00,0x40,0x00,0x02,0x04,0x05,0x06]
 ; GFX10-NEXT:    s_waitcnt vmcnt(0) ; encoding: [0x70,0x3f,0x8c,0xbf]
 ; GFX10-NEXT:    ; return to shader part epilog
 ;
 ; GFX11-LABEL: sample_d_cl_2d:
 ; GFX11:       ; %bb.0: ; %main_body
-; GFX11-NEXT:    v_and_b32_e32 v2, 0xffff, v2 ; encoding: [0xff,0x04,0x04,0x36,0xff,0xff,0x00,0x00]
-; GFX11-NEXT:    v_and_b32_e32 v0, 0xffff, v0 ; encoding: [0xff,0x00,0x00,0x36,0xff,0xff,0x00,0x00]
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) ; encoding: [0x12,0x01,0x87,0xbf]
-; GFX11-NEXT:    v_lshl_or_b32 v2, v3, 16, v2 ; encoding: [0x02,0x00,0x56,0xd6,0x03,0x21,0x09,0x04]
-; GFX11-NEXT:    v_lshl_or_b32 v0, v1, 16, v0 ; encoding: [0x00,0x00,0x56,0xd6,0x01,0x21,0x01,0x04]
+; GFX11-NEXT:    v_perm_b32 v2, v3, v2, 0x5040100 ; encoding: [0x02,0x00,0x44,0xd6,0x03,0x05,0xfe,0x03,0x00,0x01,0x04,0x05]
+; GFX11-NEXT:    v_perm_b32 v0, v1, v0, 0x5040100 ; encoding: [0x00,0x00,0x44,0xd6,0x01,0x01,0xfe,0x03,0x00,0x01,0x04,0x05]
 ; GFX11-NEXT:    image_sample_d_cl_g16 v[0:3], [v0, v2, v4, v5, v6], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_2D ; encoding: [0x05,0x0f,0x7c,0xf1,0x00,0x00,0x00,0x08,0x02,0x04,0x05,0x06]
 ; GFX11-NEXT:    s_waitcnt vmcnt(0) ; encoding: [0xf7,0x03,0x89,0xbf]
 ; GFX11-NEXT:    ; return to shader part epilog
@@ -184,10 +164,8 @@
 ; GFX10:       ; %bb.0: ; %main_body
 ; GFX10-NEXT:    v_mov_b32_e32 v8, v2 ; encoding: [0x02,0x03,0x10,0x7e]
 ; GFX10-NEXT:    v_mov_b32_e32 v2, v0 ; encoding: [0x00,0x03,0x04,0x7e]
-; GFX10-NEXT:    v_and_b32_e32 v0, 0xffff, v3 ; encoding: [0xff,0x06,0x00,0x36,0xff,0xff,0x00,0x00]
-; GFX10-NEXT:    v_and_b32_e32 v1, 0xffff, v1 ; encoding: [0xff,0x02,0x02,0x36,0xff,0xff,0x00,0x00]
-; GFX10-NEXT:    v_lshl_or_b32 v4, v4, 16, v0 ; encoding: [0x04,0x00,0x6f,0xd7,0x04,0x21,0x01,0x04]
-; GFX10-NEXT:    v_lshl_or_b32 v3, v8, 16, v1 ; encoding: [0x03,0x00,0x6f,0xd7,0x08,0x21,0x05,0x04]
+; GFX10-NEXT:    v_perm_b32 v4, v4, v3, 0x5040100 ; encoding: [0x04,0x00,0x44,0xd7,0x04,0x07,0xfe,0x03,0x00,0x01,0x04,0x05]
+; GFX10-NEXT:    v_perm_b32 v3, v8, v1, 0x5040100 ; encoding: [0x03,0x00,0x44,0xd7,0x08,0x03,0xfe,0x03,0x00,0x01,0x04,0x05]
 ; GFX10-NEXT:    image_sample_c_d_cl_g16 v[0:3], v[2:7], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_2D ; encoding: [0x09,0x0f,0xac,0xf0,0x02,0x00,0x40,0x00]
 ; GFX10-NEXT:    s_waitcnt vmcnt(0) ; encoding: [0x70,0x3f,0x8c,0xbf]
 ; GFX10-NEXT:    ; return to shader part epilog
@@ -196,11 +174,9 @@
 ; GFX11:       ; %bb.0: ; %main_body
 ; GFX11-NEXT:    v_mov_b32_e32 v8, v2 ; encoding: [0x02,0x03,0x10,0x7e]
 ; GFX11-NEXT:    v_mov_b32_e32 v2, v0 ; encoding: [0x00,0x03,0x04,0x7e]
-; GFX11-NEXT:    v_and_b32_e32 v0, 0xffff, v3 ; encoding: [0xff,0x06,0x00,0x36,0xff,0xff,0x00,0x00]
-; GFX11-NEXT:    v_and_b32_e32 v1, 0xffff, v1 ; encoding: [0xff,0x02,0x02,0x36,0xff,0xff,0x00,0x00]
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) ; encoding: [0x12,0x01,0x87,0xbf]
-; GFX11-NEXT:    v_lshl_or_b32 v4, v4, 16, v0 ; encoding: [0x04,0x00,0x56,0xd6,0x04,0x21,0x01,0x04]
-; GFX11-NEXT:    v_lshl_or_b32 v3, v8, 16, v1 ; encoding: [0x03,0x00,0x56,0xd6,0x08,0x21,0x05,0x04]
+; GFX11-NEXT:    v_perm_b32 v4, v4, v3, 0x5040100 ; encoding: [0x04,0x00,0x44,0xd6,0x04,0x07,0xfe,0x03,0x00,0x01,0x04,0x05]
+; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_3) ; encoding: [0x03,0x00,0x87,0xbf]
+; GFX11-NEXT:    v_perm_b32 v3, v8, v1, 0x5040100 ; encoding: [0x03,0x00,0x44,0xd6,0x08,0x03,0xfe,0x03,0x00,0x01,0x04,0x05]
 ; GFX11-NEXT:    image_sample_c_d_cl_g16 v[0:3], v[2:7], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_2D ; encoding: [0x04,0x0f,0x50,0xf1,0x02,0x00,0x00,0x08]
 ; GFX11-NEXT:    s_waitcnt vmcnt(0) ; encoding: [0xf7,0x03,0x89,0xbf]
 ; GFX11-NEXT:    ; return to shader part epilog
@@ -212,29 +188,25 @@
 define amdgpu_ps float @sample_c_d_o_2darray_V1(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, i32 %offset, float %zcompare, half %dsdh, half %dtdh, half %dsdv, half %dtdv, float %s, float %t, float %slice) {
 ; GFX10-LABEL: sample_c_d_o_2darray_V1:
 ; GFX10:       ; %bb.0: ; %main_body
-; GFX10-NEXT:    v_mov_b32_e32 v9, v2 ; encoding: [0x02,0x03,0x12,0x7e]
-; GFX10-NEXT:    v_mov_b32_e32 v10, v3 ; encoding: [0x03,0x03,0x14,0x7e]
+; GFX10-NEXT:    v_mov_b32_e32 v9, v3 ; encoding: [0x03,0x03,0x12,0x7e]
+; GFX10-NEXT:    v_mov_b32_e32 v10, v2 ; encoding: [0x02,0x03,0x14,0x7e]
 ; GFX10-NEXT:    v_mov_b32_e32 v3, v1 ; encoding: [0x01,0x03,0x06,0x7e]
 ; GFX10-NEXT:    v_mov_b32_e32 v2, v0 ; encoding: [0x00,0x03,0x04,0x7e]
-; GFX10-NEXT:    v_and_b32_e32 v0, 0xffff, v4 ; encoding: [0xff,0x08,0x00,0x36,0xff,0xff,0x00,0x00]
-; GFX10-NEXT:    v_and_b32_e32 v1, 0xffff, v9 ; encoding: [0xff,0x12,0x02,0x36,0xff,0xff,0x00,0x00]
-; GFX10-NEXT:    v_lshl_or_b32 v5, v5, 16, v0 ; encoding: [0x05,0x00,0x6f,0xd7,0x05,0x21,0x01,0x04]
-; GFX10-NEXT:    v_lshl_or_b32 v4, v10, 16, v1 ; encoding: [0x04,0x00,0x6f,0xd7,0x0a,0x21,0x05,0x04]
+; GFX10-NEXT:    v_perm_b32 v5, v5, v4, 0x5040100 ; encoding: [0x05,0x00,0x44,0xd7,0x05,0x09,0xfe,0x03,0x00,0x01,0x04,0x05]
+; GFX10-NEXT:    v_perm_b32 v4, v9, v10, 0x5040100 ; encoding: [0x04,0x00,0x44,0xd7,0x09,0x15,0xfe,0x03,0x00,0x01,0x04,0x05]
 ; GFX10-NEXT:    image_sample_c_d_o_g16 v0, v[2:8], s[0:7], s[8:11] dmask:0x4 dim:SQ_RSRC_IMG_2D_ARRAY ; encoding: [0x29,0x04,0xe8,0xf0,0x02,0x00,0x40,0x00]
 ; GFX10-NEXT:    s_waitcnt vmcnt(0) ; encoding: [0x70,0x3f,0x8c,0xbf]
 ; GFX10-NEXT:    ; return to shader part epilog
 ;
 ; GFX11-LABEL: sample_c_d_o_2darray_V1:
 ; GFX11:       ; %bb.0: ; %main_body
-; GFX11-NEXT:    v_mov_b32_e32 v9, v2 ; encoding: [0x02,0x03,0x12,0x7e]
-; GFX11-NEXT:    v_mov_b32_e32 v10, v3 ; encoding: [0x03,0x03,0x14,0x7e]
+; GFX11-NEXT:    v_mov_b32_e32 v9, v3 ; encoding: [0x03,0x03,0x12,0x7e]
+; GFX11-NEXT:    v_mov_b32_e32 v10, v2 ; encoding: [0x02,0x03,0x14,0x7e]
 ; GFX11-NEXT:    v_mov_b32_e32 v3, v1 ; encoding: [0x01,0x03,0x06,0x7e]
 ; GFX11-NEXT:    v_mov_b32_e32 v2, v0 ; encoding: [0x00,0x03,0x04,0x7e]
-; GFX11-NEXT:    v_and_b32_e32 v0, 0xffff, v4 ; encoding: [0xff,0x08,0x00,0x36,0xff,0xff,0x00,0x00]
-; GFX11-NEXT:    v_and_b32_e32 v1, 0xffff, v9 ; encoding: [0xff,0x12,0x02,0x36,0xff,0xff,0x00,0x00]
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) ; encoding: [0x12,0x01,0x87,0xbf]
-; GFX11-NEXT:    v_lshl_or_b32 v5, v5, 16, v0 ; encoding: [0x05,0x00,0x56,0xd6,0x05,0x21,0x01,0x04]
-; GFX11-NEXT:    v_lshl_or_b32 v4, v10, 16, v1 ; encoding: [0x04,0x00,0x56,0xd6,0x0a,0x21,0x05,0x04]
+; GFX11-NEXT:    v_perm_b32 v5, v5, v4, 0x5040100 ; encoding: [0x05,0x00,0x44,0xd6,0x05,0x09,0xfe,0x03,0x00,0x01,0x04,0x05]
+; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_4) ; encoding: [0x04,0x00,0x87,0xbf]
+; GFX11-NEXT:    v_perm_b32 v4, v9, v10, 0x5040100 ; encoding: [0x04,0x00,0x44,0xd6,0x09,0x15,0xfe,0x03,0x00,0x01,0x04,0x05]
 ; GFX11-NEXT:    image_sample_c_d_o_g16 v0, v[2:8], s[0:7], s[8:11] dmask:0x4 dim:SQ_RSRC_IMG_2D_ARRAY ; encoding: [0x14,0x04,0xf0,0xf0,0x02,0x00,0x00,0x08]
 ; GFX11-NEXT:    s_waitcnt vmcnt(0) ; encoding: [0xf7,0x03,0x89,0xbf]
 ; GFX11-NEXT:    ; return to shader part epilog
@@ -246,29 +218,25 @@
 define amdgpu_ps <2 x float> @sample_c_d_o_2darray_V2(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, i32 %offset, float %zcompare, half %dsdh, half %dtdh, half %dsdv, half %dtdv, float %s, float %t, float %slice) {
 ; GFX10-LABEL: sample_c_d_o_2darray_V2:
 ; GFX10:       ; %bb.0: ; %main_body
-; GFX10-NEXT:    v_mov_b32_e32 v9, v2 ; encoding: [0x02,0x03,0x12,0x7e]
-; GFX10-NEXT:    v_mov_b32_e32 v10, v3 ; encoding: [0x03,0x03,0x14,0x7e]
+; GFX10-NEXT:    v_mov_b32_e32 v9, v3 ; encoding: [0x03,0x03,0x12,0x7e]
+; GFX10-NEXT:    v_mov_b32_e32 v10, v2 ; encoding: [0x02,0x03,0x14,0x7e]
 ; GFX10-NEXT:    v_mov_b32_e32 v3, v1 ; encoding: [0x01,0x03,0x06,0x7e]
 ; GFX10-NEXT:    v_mov_b32_e32 v2, v0 ; encoding: [0x00,0x03,0x04,0x7e]
-; GFX10-NEXT:    v_and_b32_e32 v0, 0xffff, v4 ; encoding: [0xff,0x08,0x00,0x36,0xff,0xff,0x00,0x00]
-; GFX10-NEXT:    v_and_b32_e32 v1, 0xffff, v9 ; encoding: [0xff,0x12,0x02,0x36,0xff,0xff,0x00,0x00]
-; GFX10-NEXT:    v_lshl_or_b32 v5, v5, 16, v0 ; encoding: [0x05,0x00,0x6f,0xd7,0x05,0x21,0x01,0x04]
-; GFX10-NEXT:    v_lshl_or_b32 v4, v10, 16, v1 ; encoding: [0x04,0x00,0x6f,0xd7,0x0a,0x21,0x05,0x04]
+; GFX10-NEXT:    v_perm_b32 v5, v5, v4, 0x5040100 ; encoding: [0x05,0x00,0x44,0xd7,0x05,0x09,0xfe,0x03,0x00,0x01,0x04,0x05]
+; GFX10-NEXT:    v_perm_b32 v4, v9, v10, 0x5040100 ; encoding: [0x04,0x00,0x44,0xd7,0x09,0x15,0xfe,0x03,0x00,0x01,0x04,0x05]
 ; GFX10-NEXT:    image_sample_c_d_o_g16 v[0:1], v[2:8], s[0:7], s[8:11] dmask:0x6 dim:SQ_RSRC_IMG_2D_ARRAY ; encoding: [0x29,0x06,0xe8,0xf0,0x02,0x00,0x40,0x00]
 ; GFX10-NEXT:    s_waitcnt vmcnt(0) ; encoding: [0x70,0x3f,0x8c,0xbf]
 ; GFX10-NEXT:    ; return to shader part epilog
 ;
 ; GFX11-LABEL: sample_c_d_o_2darray_V2:
 ; GFX11:       ; %bb.0: ; %main_body
-; GFX11-NEXT:    v_mov_b32_e32 v9, v2 ; encoding: [0x02,0x03,0x12,0x7e]
-; GFX11-NEXT:    v_mov_b32_e32 v10, v3 ; encoding: [0x03,0x03,0x14,0x7e]
+; GFX11-NEXT:    v_mov_b32_e32 v9, v3 ; encoding: [0x03,0x03,0x12,0x7e]
+; GFX11-NEXT:    v_mov_b32_e32 v10, v2 ; encoding: [0x02,0x03,0x14,0x7e]
 ; GFX11-NEXT:    v_mov_b32_e32 v3, v1 ; encoding: [0x01,0x03,0x06,0x7e]
 ; GFX11-NEXT:    v_mov_b32_e32 v2, v0 ; encoding: [0x00,0x03,0x04,0x7e]
-; GFX11-NEXT:    v_and_b32_e32 v0, 0xffff, v4 ; encoding: [0xff,0x08,0x00,0x36,0xff,0xff,0x00,0x00]
-; GFX11-NEXT:    v_and_b32_e32 v1, 0xffff, v9 ; encoding: [0xff,0x12,0x02,0x36,0xff,0xff,0x00,0x00]
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) ; encoding: [0x12,0x01,0x87,0xbf]
-; GFX11-NEXT:    v_lshl_or_b32 v5, v5, 16, v0 ; encoding: [0x05,0x00,0x56,0xd6,0x05,0x21,0x01,0x04]
-; GFX11-NEXT:    v_lshl_or_b32 v4, v10, 16, v1 ; encoding: [0x04,0x00,0x56,0xd6,0x0a,0x21,0x05,0x04]
+; GFX11-NEXT:    v_perm_b32 v5, v5, v4, 0x5040100 ; encoding: [0x05,0x00,0x44,0xd6,0x05,0x09,0xfe,0x03,0x00,0x01,0x04,0x05]
+; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_4) ; encoding: [0x04,0x00,0x87,0xbf]
+; GFX11-NEXT:    v_perm_b32 v4, v9, v10, 0x5040100 ; encoding: [0x04,0x00,0x44,0xd6,0x09,0x15,0xfe,0x03,0x00,0x01,0x04,0x05]
 ; GFX11-NEXT:    image_sample_c_d_o_g16 v[0:1], v[2:8], s[0:7], s[8:11] dmask:0x6 dim:SQ_RSRC_IMG_2D_ARRAY ; encoding: [0x14,0x06,0xf0,0xf0,0x02,0x00,0x00,0x08]
 ; GFX11-NEXT:    s_waitcnt vmcnt(0) ; encoding: [0xf7,0x03,0x89,0xbf]
 ; GFX11-NEXT:    ; return to shader part epilog
diff --git a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.image.sample.g16.ll b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.image.sample.g16.ll
--- a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.image.sample.g16.ll
+++ b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.image.sample.g16.ll
@@ -16,10 +16,8 @@
 define amdgpu_ps <4 x float> @sample_d_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %dsdh, half %dtdh, half %dsdv, half %dtdv, float %s, float %t) {
 ; GFX10-LABEL: sample_d_2d:
 ; GFX10:       ; %bb.0: ; %main_body
-; GFX10-NEXT:    v_and_b32_e32 v2, 0xffff, v2
-; GFX10-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX10-NEXT:    v_lshl_or_b32 v2, v3, 16, v2
-; GFX10-NEXT:    v_lshl_or_b32 v0, v1, 16, v0
+; GFX10-NEXT:    v_perm_b32 v2, v3, v2, 0x5040100
+; GFX10-NEXT:    v_perm_b32 v0, v1, v0, 0x5040100
 ; GFX10-NEXT:    image_sample_d_g16 v[0:3], [v0, v2, v4, v5], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_2D
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    ; return to shader part epilog
@@ -33,10 +31,8 @@
 ; GFX10:       ; %bb.0: ; %main_body
 ; GFX10-NEXT:    v_mov_b32_e32 v9, v3
 ; GFX10-NEXT:    v_mov_b32_e32 v3, v2
-; GFX10-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX10-NEXT:    v_and_b32_e32 v2, 0xffff, v9
-; GFX10-NEXT:    v_lshl_or_b32 v4, v4, 16, v2
-; GFX10-NEXT:    v_lshl_or_b32 v2, v1, 16, v0
+; GFX10-NEXT:    v_perm_b32 v2, v1, v0, 0x5040100
+; GFX10-NEXT:    v_perm_b32 v4, v4, v9, 0x5040100
 ; GFX10-NEXT:    image_sample_d_g16 v[0:3], v[2:8], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_3D
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    ; return to shader part epilog
@@ -59,10 +55,8 @@
 define amdgpu_ps <4 x float> @sample_c_d_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %zcompare, half %dsdh, half %dtdh, half %dsdv, half %dtdv, float %s, float %t) {
 ; GFX10-LABEL: sample_c_d_2d:
 ; GFX10:       ; %bb.0: ; %main_body
-; GFX10-NEXT:    v_and_b32_e32 v3, 0xffff, v3
-; GFX10-NEXT:    v_and_b32_e32 v1, 0xffff, v1
-; GFX10-NEXT:    v_lshl_or_b32 v3, v4, 16, v3
-; GFX10-NEXT:    v_lshl_or_b32 v1, v2, 16, v1
+; GFX10-NEXT:    v_perm_b32 v3, v4, v3, 0x5040100
+; GFX10-NEXT:    v_perm_b32 v1, v2, v1, 0x5040100
 ; GFX10-NEXT:    image_sample_c_d_g16 v[0:3], [v0, v1, v3, v5, v6], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_2D
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    ; return to shader part epilog
@@ -85,10 +79,8 @@
 define amdgpu_ps <4 x float> @sample_d_cl_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %dsdh, half %dtdh, half %dsdv, half %dtdv, float %s, float %t, float %clamp) {
 ; GFX10-LABEL: sample_d_cl_2d:
 ; GFX10:       ; %bb.0: ; %main_body
-; GFX10-NEXT:    v_and_b32_e32 v2, 0xffff, v2
-; GFX10-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX10-NEXT:    v_lshl_or_b32 v2, v3, 16, v2
-; GFX10-NEXT:    v_lshl_or_b32 v0, v1, 16, v0
+; GFX10-NEXT:    v_perm_b32 v2, v3, v2, 0x5040100
+; GFX10-NEXT:    v_perm_b32 v0, v1, v0, 0x5040100
 ; GFX10-NEXT:    image_sample_d_cl_g16 v[0:3], [v0, v2, v4, v5, v6], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_2D
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    ; return to shader part epilog
@@ -113,10 +105,8 @@
 ; GFX10:       ; %bb.0: ; %main_body
 ; GFX10-NEXT:    v_mov_b32_e32 v8, v2
 ; GFX10-NEXT:    v_mov_b32_e32 v2, v0
-; GFX10-NEXT:    v_and_b32_e32 v0, 0xffff, v3
-; GFX10-NEXT:    v_and_b32_e32 v1, 0xffff, v1
-; GFX10-NEXT:    v_lshl_or_b32 v4, v4, 16, v0
-; GFX10-NEXT:    v_lshl_or_b32 v3, v8, 16, v1
+; GFX10-NEXT:    v_perm_b32 v4, v4, v3, 0x5040100
+; GFX10-NEXT:    v_perm_b32 v3, v8, v1, 0x5040100
 ; GFX10-NEXT:    image_sample_c_d_cl_g16 v[0:3], v[2:7], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_2D
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    ; return to shader part epilog
@@ -128,14 +118,12 @@
 define amdgpu_ps float @sample_c_d_o_2darray_V1(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, i32 %offset, float %zcompare, half %dsdh, half %dtdh, half %dsdv, half %dtdv, float %s, float %t, float %slice) {
 ; GFX10-LABEL: sample_c_d_o_2darray_V1:
 ; GFX10:       ; %bb.0: ; %main_body
-; GFX10-NEXT:    v_mov_b32_e32 v9, v2
-; GFX10-NEXT:    v_mov_b32_e32 v10, v3
+; GFX10-NEXT:    v_mov_b32_e32 v9, v3
+; GFX10-NEXT:    v_mov_b32_e32 v10, v2
 ; GFX10-NEXT:    v_mov_b32_e32 v3, v1
 ; GFX10-NEXT:    v_mov_b32_e32 v2, v0
-; GFX10-NEXT:    v_and_b32_e32 v0, 0xffff, v4
-; GFX10-NEXT:    v_and_b32_e32 v1, 0xffff, v9
-; GFX10-NEXT:    v_lshl_or_b32 v5, v5, 16, v0
-; GFX10-NEXT:    v_lshl_or_b32 v4, v10, 16, v1
+; GFX10-NEXT:    v_perm_b32 v5, v5, v4, 0x5040100
+; GFX10-NEXT:    v_perm_b32 v4, v9, v10, 0x5040100
 ; GFX10-NEXT:    image_sample_c_d_o_g16 v0, v[2:8], s[0:7], s[8:11] dmask:0x4 dim:SQ_RSRC_IMG_2D_ARRAY
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    ; return to shader part epilog
@@ -147,14 +135,12 @@
 define amdgpu_ps <2 x float> @sample_c_d_o_2darray_V2(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, i32 %offset, float %zcompare, half %dsdh, half %dtdh, half %dsdv, half %dtdv, float %s, float %t, float %slice) {
 ; GFX10-LABEL: sample_c_d_o_2darray_V2:
 ; GFX10:       ; %bb.0: ; %main_body
-; GFX10-NEXT:    v_mov_b32_e32 v9, v2
-; GFX10-NEXT:    v_mov_b32_e32 v10, v3
+; GFX10-NEXT:    v_mov_b32_e32 v9, v3
+; GFX10-NEXT:    v_mov_b32_e32 v10, v2
 ; GFX10-NEXT:    v_mov_b32_e32 v3, v1
 ; GFX10-NEXT:    v_mov_b32_e32 v2, v0
-; GFX10-NEXT:    v_and_b32_e32 v0, 0xffff, v4
-; GFX10-NEXT:    v_and_b32_e32 v1, 0xffff, v9
-; GFX10-NEXT:    v_lshl_or_b32 v5, v5, 16, v0
-; GFX10-NEXT:    v_lshl_or_b32 v4, v10, 16, v1
+; GFX10-NEXT:    v_perm_b32 v5, v5, v4, 0x5040100
+; GFX10-NEXT:    v_perm_b32 v4, v9, v10, 0x5040100
 ; GFX10-NEXT:    image_sample_c_d_o_g16 v[0:1], v[2:8], s[0:7], s[8:11] dmask:0x6 dim:SQ_RSRC_IMG_2D_ARRAY
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    ; return to shader part epilog
diff --git a/llvm/test/CodeGen/AMDGPU/load-hi16.ll b/llvm/test/CodeGen/AMDGPU/load-hi16.ll
--- a/llvm/test/CodeGen/AMDGPU/load-hi16.ll
+++ b/llvm/test/CodeGen/AMDGPU/load-hi16.ll
@@ -24,12 +24,12 @@
 ; GFX906-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX906-NEXT:    ds_read_u16 v1, v0
 ; GFX906-NEXT:    ds_read_u16 v0, v0 offset:16
+; GFX906-NEXT:    s_mov_b32 s4, 0x5040100
 ; GFX906-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX906-NEXT:    s_waitcnt lgkmcnt(1)
 ; GFX906-NEXT:    ds_write_b16 v2, v1
-; GFX906-NEXT:    v_and_b32_e32 v1, 0xffff, v1
 ; GFX906-NEXT:    s_waitcnt lgkmcnt(1)
-; GFX906-NEXT:    v_lshl_or_b32 v0, v0, 16, v1
+; GFX906-NEXT:    v_perm_b32 v0, v0, v1, s4
 ; GFX906-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX906-NEXT:    s_setpc_b64 s[30:31]
 ;
@@ -77,12 +77,12 @@
 ; GFX900-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX900-NEXT:    ds_read_u16 v1, v0 offset:16
 ; GFX900-NEXT:    ds_read_u16 v0, v0
+; GFX900-NEXT:    s_mov_b32 s4, 0x5040100
 ; GFX900-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX900-NEXT:    s_waitcnt lgkmcnt(1)
 ; GFX900-NEXT:    ds_write_b16 v2, v1
 ; GFX900-NEXT:    s_waitcnt lgkmcnt(1)
-; GFX900-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX900-NEXT:    v_lshl_or_b32 v0, v1, 16, v0
+; GFX900-NEXT:    v_perm_b32 v0, v1, v0, s4
 ; GFX900-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX900-NEXT:    s_setpc_b64 s[30:31]
 ;
@@ -91,12 +91,12 @@
 ; GFX906-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX906-NEXT:    ds_read_u16 v1, v0 offset:16
 ; GFX906-NEXT:    ds_read_u16 v0, v0
+; GFX906-NEXT:    s_mov_b32 s4, 0x5040100
 ; GFX906-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX906-NEXT:    s_waitcnt lgkmcnt(1)
 ; GFX906-NEXT:    ds_write_b16 v2, v1
 ; GFX906-NEXT:    s_waitcnt lgkmcnt(1)
-; GFX906-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX906-NEXT:    v_lshl_or_b32 v0, v1, 16, v0
+; GFX906-NEXT:    v_perm_b32 v0, v1, v0, s4
 ; GFX906-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX906-NEXT:    s_setpc_b64 s[30:31]
 ;
@@ -120,12 +120,12 @@
 ; GFX900-FLATSCR-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX900-FLATSCR-NEXT:    ds_read_u16 v1, v0 offset:16
 ; GFX900-FLATSCR-NEXT:    ds_read_u16 v0, v0
+; GFX900-FLATSCR-NEXT:    s_mov_b32 s0, 0x5040100
 ; GFX900-FLATSCR-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX900-FLATSCR-NEXT:    s_waitcnt lgkmcnt(1)
 ; GFX900-FLATSCR-NEXT:    ds_write_b16 v2, v1
 ; GFX900-FLATSCR-NEXT:    s_waitcnt lgkmcnt(1)
-; GFX900-FLATSCR-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX900-FLATSCR-NEXT:    v_lshl_or_b32 v0, v1, 16, v0
+; GFX900-FLATSCR-NEXT:    v_perm_b32 v0, v1, v0, s0
 ; GFX900-FLATSCR-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX900-FLATSCR-NEXT:    s_setpc_b64 s[30:31]
 entry:
@@ -144,12 +144,12 @@
 ; GFX900-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX900-NEXT:    ds_read_u16 v3, v0
 ; GFX900-NEXT:    ds_read_u16 v0, v0 offset:16
+; GFX900-NEXT:    s_mov_b32 s4, 0x5040100
 ; GFX900-NEXT:    s_waitcnt lgkmcnt(1)
 ; GFX900-NEXT:    ds_write_b16 v1, v3
 ; GFX900-NEXT:    s_waitcnt lgkmcnt(1)
 ; GFX900-NEXT:    ds_write_b16 v2, v0
-; GFX900-NEXT:    v_and_b32_e32 v1, 0xffff, v3
-; GFX900-NEXT:    v_lshl_or_b32 v0, v0, 16, v1
+; GFX900-NEXT:    v_perm_b32 v0, v0, v3, s4
 ; GFX900-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX900-NEXT:    s_setpc_b64 s[30:31]
 ;
@@ -158,12 +158,12 @@
 ; GFX906-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX906-NEXT:    ds_read_u16 v3, v0
 ; GFX906-NEXT:    ds_read_u16 v0, v0 offset:16
+; GFX906-NEXT:    s_mov_b32 s4, 0x5040100
 ; GFX906-NEXT:    s_waitcnt lgkmcnt(1)
 ; GFX906-NEXT:    ds_write_b16 v1, v3
 ; GFX906-NEXT:    s_waitcnt lgkmcnt(1)
 ; GFX906-NEXT:    ds_write_b16 v2, v0
-; GFX906-NEXT:    v_and_b32_e32 v1, 0xffff, v3
-; GFX906-NEXT:    v_lshl_or_b32 v0, v0, 16, v1
+; GFX906-NEXT:    v_perm_b32 v0, v0, v3, s4
 ; GFX906-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX906-NEXT:    s_setpc_b64 s[30:31]
 ;
@@ -187,12 +187,12 @@
 ; GFX900-FLATSCR-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX900-FLATSCR-NEXT:    ds_read_u16 v3, v0
 ; GFX900-FLATSCR-NEXT:    ds_read_u16 v0, v0 offset:16
+; GFX900-FLATSCR-NEXT:    s_mov_b32 s0, 0x5040100
 ; GFX900-FLATSCR-NEXT:    s_waitcnt lgkmcnt(1)
 ; GFX900-FLATSCR-NEXT:    ds_write_b16 v1, v3
 ; GFX900-FLATSCR-NEXT:    s_waitcnt lgkmcnt(1)
 ; GFX900-FLATSCR-NEXT:    ds_write_b16 v2, v0
-; GFX900-FLATSCR-NEXT:    v_and_b32_e32 v1, 0xffff, v3
-; GFX900-FLATSCR-NEXT:    v_lshl_or_b32 v0, v0, 16, v1
+; GFX900-FLATSCR-NEXT:    v_perm_b32 v0, v0, v3, s0
 ; GFX900-FLATSCR-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX900-FLATSCR-NEXT:    s_setpc_b64 s[30:31]
 entry:
@@ -256,9 +256,9 @@
 ; GFX906:       ; %bb.0: ; %entry
 ; GFX906-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX906-NEXT:    ds_read_u16 v0, v0
-; GFX906-NEXT:    v_and_b32_e32 v1, 0xffff, v1
+; GFX906-NEXT:    s_mov_b32 s4, 0x5040100
 ; GFX906-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX906-NEXT:    v_lshl_or_b32 v0, v0, 16, v1
+; GFX906-NEXT:    v_perm_b32 v0, v0, v1, s4
 ; GFX906-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX803-LABEL: load_local_hi_v2i16_reglo:
@@ -299,9 +299,9 @@
 ; GFX906:       ; %bb.0: ; %entry
 ; GFX906-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX906-NEXT:    ds_read_u16 v0, v0
-; GFX906-NEXT:    v_and_b32_e32 v1, 0xffff, v1
+; GFX906-NEXT:    s_mov_b32 s4, 0x5040100
 ; GFX906-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX906-NEXT:    v_lshl_or_b32 v0, v0, 16, v1
+; GFX906-NEXT:    v_perm_b32 v0, v0, v1, s4
 ; GFX906-NEXT:    global_store_dword v[0:1], v0, off
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
 ; GFX906-NEXT:    s_setpc_b64 s[30:31]
@@ -430,9 +430,9 @@
 ; GFX906:       ; %bb.0: ; %entry
 ; GFX906-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX906-NEXT:    ds_read_u16 v0, v0
-; GFX906-NEXT:    v_and_b32_e32 v1, 0xffff, v1
+; GFX906-NEXT:    s_mov_b32 s4, 0x5040100
 ; GFX906-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX906-NEXT:    v_lshl_or_b32 v0, v0, 16, v1
+; GFX906-NEXT:    v_perm_b32 v0, v0, v1, s4
 ; GFX906-NEXT:    global_store_dword v[0:1], v0, off
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
 ; GFX906-NEXT:    s_setpc_b64 s[30:31]
@@ -479,9 +479,9 @@
 ; GFX906:       ; %bb.0: ; %entry
 ; GFX906-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX906-NEXT:    ds_read_u8 v0, v0
-; GFX906-NEXT:    v_and_b32_e32 v1, 0xffff, v1
+; GFX906-NEXT:    s_mov_b32 s4, 0x5040100
 ; GFX906-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX906-NEXT:    v_lshl_or_b32 v0, v0, 16, v1
+; GFX906-NEXT:    v_perm_b32 v0, v0, v1, s4
 ; GFX906-NEXT:    global_store_dword v[0:1], v0, off
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
 ; GFX906-NEXT:    s_setpc_b64 s[30:31]
@@ -529,9 +529,9 @@
 ; GFX906:       ; %bb.0: ; %entry
 ; GFX906-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX906-NEXT:    ds_read_i8 v0, v0
-; GFX906-NEXT:    v_and_b32_e32 v1, 0xffff, v1
+; GFX906-NEXT:    s_mov_b32 s4, 0x5040100
 ; GFX906-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX906-NEXT:    v_lshl_or_b32 v0, v0, 16, v1
+; GFX906-NEXT:    v_perm_b32 v0, v0, v1, s4
 ; GFX906-NEXT:    global_store_dword v[0:1], v0, off
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
 ; GFX906-NEXT:    s_setpc_b64 s[30:31]
@@ -579,9 +579,9 @@
 ; GFX906:       ; %bb.0: ; %entry
 ; GFX906-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX906-NEXT:    ds_read_u8 v0, v0
-; GFX906-NEXT:    v_and_b32_e32 v1, 0xffff, v1
+; GFX906-NEXT:    s_mov_b32 s4, 0x5040100
 ; GFX906-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX906-NEXT:    v_lshl_or_b32 v0, v0, 16, v1
+; GFX906-NEXT:    v_perm_b32 v0, v0, v1, s4
 ; GFX906-NEXT:    global_store_dword v[0:1], v0, off
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
 ; GFX906-NEXT:    s_setpc_b64 s[30:31]
@@ -631,9 +631,9 @@
 ; GFX906:       ; %bb.0: ; %entry
 ; GFX906-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX906-NEXT:    ds_read_i8 v0, v0
-; GFX906-NEXT:    v_and_b32_e32 v1, 0xffff, v1
+; GFX906-NEXT:    s_mov_b32 s4, 0x5040100
 ; GFX906-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX906-NEXT:    v_lshl_or_b32 v0, v0, 16, v1
+; GFX906-NEXT:    v_perm_b32 v0, v0, v1, s4
 ; GFX906-NEXT:    global_store_dword v[0:1], v0, off
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
 ; GFX906-NEXT:    s_setpc_b64 s[30:31]
@@ -683,9 +683,9 @@
 ; GFX906:       ; %bb.0: ; %entry
 ; GFX906-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX906-NEXT:    global_load_ushort v0, v[0:1], off offset:-4094
-; GFX906-NEXT:    v_and_b32_e32 v1, 0xffff, v2
+; GFX906-NEXT:    s_mov_b32 s4, 0x5040100
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
-; GFX906-NEXT:    v_lshl_or_b32 v0, v0, 16, v1
+; GFX906-NEXT:    v_perm_b32 v0, v0, v2, s4
 ; GFX906-NEXT:    global_store_dword v[0:1], v0, off
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
 ; GFX906-NEXT:    s_setpc_b64 s[30:31]
@@ -734,9 +734,9 @@
 ; GFX906:       ; %bb.0: ; %entry
 ; GFX906-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX906-NEXT:    global_load_ushort v0, v[0:1], off offset:-4094
-; GFX906-NEXT:    v_and_b32_e32 v1, 0xffff, v2
+; GFX906-NEXT:    s_mov_b32 s4, 0x5040100
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
-; GFX906-NEXT:    v_lshl_or_b32 v0, v0, 16, v1
+; GFX906-NEXT:    v_perm_b32 v0, v0, v2, s4
 ; GFX906-NEXT:    global_store_dword v[0:1], v0, off
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
 ; GFX906-NEXT:    s_setpc_b64 s[30:31]
@@ -785,9 +785,9 @@
 ; GFX906:       ; %bb.0: ; %entry
 ; GFX906-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX906-NEXT:    global_load_ubyte v0, v[0:1], off offset:-4095
-; GFX906-NEXT:    v_and_b32_e32 v1, 0xffff, v2
+; GFX906-NEXT:    s_mov_b32 s4, 0x5040100
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
-; GFX906-NEXT:    v_lshl_or_b32 v0, v0, 16, v1
+; GFX906-NEXT:    v_perm_b32 v0, v0, v2, s4
 ; GFX906-NEXT:    global_store_dword v[0:1], v0, off
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
 ; GFX906-NEXT:    s_setpc_b64 s[30:31]
@@ -837,9 +837,9 @@
 ; GFX906:       ; %bb.0: ; %entry
 ; GFX906-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX906-NEXT:    global_load_sbyte v0, v[0:1], off offset:-4095
-; GFX906-NEXT:    v_and_b32_e32 v1, 0xffff, v2
+; GFX906-NEXT:    s_mov_b32 s4, 0x5040100
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
-; GFX906-NEXT:    v_lshl_or_b32 v0, v0, 16, v1
+; GFX906-NEXT:    v_perm_b32 v0, v0, v2, s4
 ; GFX906-NEXT:    global_store_dword v[0:1], v0, off
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
 ; GFX906-NEXT:    s_setpc_b64 s[30:31]
@@ -889,9 +889,9 @@
 ; GFX906:       ; %bb.0: ; %entry
 ; GFX906-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX906-NEXT:    global_load_sbyte v0, v[0:1], off offset:-4095
-; GFX906-NEXT:    v_and_b32_e32 v1, 0xffff, v2
+; GFX906-NEXT:    s_mov_b32 s4, 0x5040100
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
-; GFX906-NEXT:    v_lshl_or_b32 v0, v0, 16, v1
+; GFX906-NEXT:    v_perm_b32 v0, v0, v2, s4
 ; GFX906-NEXT:    global_store_dword v[0:1], v0, off
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
 ; GFX906-NEXT:    s_setpc_b64 s[30:31]
@@ -942,9 +942,9 @@
 ; GFX906:       ; %bb.0: ; %entry
 ; GFX906-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX906-NEXT:    global_load_ubyte v0, v[0:1], off offset:-4095
-; GFX906-NEXT:    v_and_b32_e32 v1, 0xffff, v2
+; GFX906-NEXT:    s_mov_b32 s4, 0x5040100
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
-; GFX906-NEXT:    v_lshl_or_b32 v0, v0, 16, v1
+; GFX906-NEXT:    v_perm_b32 v0, v0, v2, s4
 ; GFX906-NEXT:    global_store_dword v[0:1], v0, off
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
 ; GFX906-NEXT:    s_setpc_b64 s[30:31]
@@ -995,9 +995,9 @@
 ; GFX906:       ; %bb.0: ; %entry
 ; GFX906-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX906-NEXT:    flat_load_ushort v0, v[0:1]
-; GFX906-NEXT:    v_and_b32_e32 v1, 0xffff, v2
+; GFX906-NEXT:    s_mov_b32 s4, 0x5040100
 ; GFX906-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
-; GFX906-NEXT:    v_lshl_or_b32 v0, v0, 16, v1
+; GFX906-NEXT:    v_perm_b32 v0, v0, v2, s4
 ; GFX906-NEXT:    global_store_dword v[0:1], v0, off
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
 ; GFX906-NEXT:    s_setpc_b64 s[30:31]
@@ -1043,9 +1043,9 @@
 ; GFX906:       ; %bb.0: ; %entry
 ; GFX906-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX906-NEXT:    flat_load_ushort v0, v[0:1]
-; GFX906-NEXT:    v_and_b32_e32 v1, 0xffff, v2
+; GFX906-NEXT:    s_mov_b32 s4, 0x5040100
 ; GFX906-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
-; GFX906-NEXT:    v_lshl_or_b32 v0, v0, 16, v1
+; GFX906-NEXT:    v_perm_b32 v0, v0, v2, s4
 ; GFX906-NEXT:    global_store_dword v[0:1], v0, off
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
 ; GFX906-NEXT:    s_setpc_b64 s[30:31]
@@ -1091,9 +1091,9 @@
 ; GFX906:       ; %bb.0: ; %entry
 ; GFX906-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX906-NEXT:    flat_load_ubyte v0, v[0:1]
-; GFX906-NEXT:    v_and_b32_e32 v1, 0xffff, v2
+; GFX906-NEXT:    s_mov_b32 s4, 0x5040100
 ; GFX906-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
-; GFX906-NEXT:    v_lshl_or_b32 v0, v0, 16, v1
+; GFX906-NEXT:    v_perm_b32 v0, v0, v2, s4
 ; GFX906-NEXT:    global_store_dword v[0:1], v0, off
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
 ; GFX906-NEXT:    s_setpc_b64 s[30:31]
@@ -1140,9 +1140,9 @@
 ; GFX906:       ; %bb.0: ; %entry
 ; GFX906-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX906-NEXT:    flat_load_sbyte v0, v[0:1]
-; GFX906-NEXT:    v_and_b32_e32 v1, 0xffff, v2
+; GFX906-NEXT:    s_mov_b32 s4, 0x5040100
 ; GFX906-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
-; GFX906-NEXT:    v_lshl_or_b32 v0, v0, 16, v1
+; GFX906-NEXT:    v_perm_b32 v0, v0, v2, s4
 ; GFX906-NEXT:    global_store_dword v[0:1], v0, off
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
 ; GFX906-NEXT:    s_setpc_b64 s[30:31]
@@ -1189,9 +1189,9 @@
 ; GFX906:       ; %bb.0: ; %entry
 ; GFX906-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX906-NEXT:    flat_load_ubyte v0, v[0:1]
-; GFX906-NEXT:    v_and_b32_e32 v1, 0xffff, v2
+; GFX906-NEXT:    s_mov_b32 s4, 0x5040100
 ; GFX906-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
-; GFX906-NEXT:    v_lshl_or_b32 v0, v0, 16, v1
+; GFX906-NEXT:    v_perm_b32 v0, v0, v2, s4
 ; GFX906-NEXT:    global_store_dword v[0:1], v0, off
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
 ; GFX906-NEXT:    s_setpc_b64 s[30:31]
@@ -1239,9 +1239,9 @@
 ; GFX906:       ; %bb.0: ; %entry
 ; GFX906-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX906-NEXT:    flat_load_sbyte v0, v[0:1]
-; GFX906-NEXT:    v_and_b32_e32 v1, 0xffff, v2
+; GFX906-NEXT:    s_mov_b32 s4, 0x5040100
 ; GFX906-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
-; GFX906-NEXT:    v_lshl_or_b32 v0, v0, 16, v1
+; GFX906-NEXT:    v_perm_b32 v0, v0, v2, s4
 ; GFX906-NEXT:    global_store_dword v[0:1], v0, off
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
 ; GFX906-NEXT:    s_setpc_b64 s[30:31]
@@ -1289,9 +1289,9 @@
 ; GFX906:       ; %bb.0: ; %entry
 ; GFX906-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX906-NEXT:    buffer_load_ushort v1, off, s[0:3], s32 offset:4094
-; GFX906-NEXT:    v_and_b32_e32 v0, 0xffff, v0
+; GFX906-NEXT:    s_mov_b32 s4, 0x5040100
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
-; GFX906-NEXT:    v_lshl_or_b32 v0, v1, 16, v0
+; GFX906-NEXT:    v_perm_b32 v0, v1, v0, s4
 ; GFX906-NEXT:    global_store_dword v[0:1], v0, off
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
 ; GFX906-NEXT:    s_setpc_b64 s[30:31]
@@ -1338,9 +1338,9 @@
 ; GFX906:       ; %bb.0: ; %entry
 ; GFX906-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX906-NEXT:    buffer_load_ushort v1, off, s[0:3], s32 offset:4094
-; GFX906-NEXT:    v_and_b32_e32 v0, 0xffff, v0
+; GFX906-NEXT:    s_mov_b32 s4, 0x5040100
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
-; GFX906-NEXT:    v_lshl_or_b32 v0, v1, 16, v0
+; GFX906-NEXT:    v_perm_b32 v0, v1, v0, s4
 ; GFX906-NEXT:    global_store_dword v[0:1], v0, off
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
 ; GFX906-NEXT:    s_setpc_b64 s[30:31]
@@ -1388,8 +1388,8 @@
 ; GFX906-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX906-NEXT:    buffer_load_ushort v1, off, s[0:3], 0 offset:4094 glc
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
-; GFX906-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX906-NEXT:    v_lshl_or_b32 v0, v1, 16, v0
+; GFX906-NEXT:    s_mov_b32 s4, 0x5040100
+; GFX906-NEXT:    v_perm_b32 v0, v1, v0, s4
 ; GFX906-NEXT:    global_store_dword v[0:1], v0, off
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
 ; GFX906-NEXT:    s_setpc_b64 s[30:31]
@@ -1437,8 +1437,8 @@
 ; GFX906-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX906-NEXT:    buffer_load_ushort v0, off, s[0:3], 0 offset:4094 glc
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
-; GFX906-NEXT:    v_and_b32_e32 v1, 0xffff, v1
-; GFX906-NEXT:    v_lshl_or_b32 v0, v0, 16, v1
+; GFX906-NEXT:    s_mov_b32 s4, 0x5040100
+; GFX906-NEXT:    v_perm_b32 v0, v0, v1, s4
 ; GFX906-NEXT:    global_store_dword v[0:1], v0, off
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
 ; GFX906-NEXT:    s_setpc_b64 s[30:31]
@@ -1485,9 +1485,9 @@
 ; GFX906:       ; %bb.0: ; %entry
 ; GFX906-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX906-NEXT:    buffer_load_ubyte v1, off, s[0:3], s32 offset:4095
-; GFX906-NEXT:    v_and_b32_e32 v0, 0xffff, v0
+; GFX906-NEXT:    s_mov_b32 s4, 0x5040100
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
-; GFX906-NEXT:    v_lshl_or_b32 v0, v1, 16, v0
+; GFX906-NEXT:    v_perm_b32 v0, v1, v0, s4
 ; GFX906-NEXT:    global_store_dword v[0:1], v0, off
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
 ; GFX906-NEXT:    s_setpc_b64 s[30:31]
@@ -1535,9 +1535,9 @@
 ; GFX906:       ; %bb.0: ; %entry
 ; GFX906-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX906-NEXT:    buffer_load_ubyte v1, off, s[0:3], s32 offset:4095
-; GFX906-NEXT:    v_and_b32_e32 v0, 0xffff, v0
+; GFX906-NEXT:    s_mov_b32 s4, 0x5040100
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
-; GFX906-NEXT:    v_lshl_or_b32 v0, v1, 16, v0
+; GFX906-NEXT:    v_perm_b32 v0, v1, v0, s4
 ; GFX906-NEXT:    global_store_dword v[0:1], v0, off
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
 ; GFX906-NEXT:    s_setpc_b64 s[30:31]
@@ -1586,9 +1586,9 @@
 ; GFX906:       ; %bb.0: ; %entry
 ; GFX906-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX906-NEXT:    buffer_load_sbyte v1, off, s[0:3], s32 offset:4095
-; GFX906-NEXT:    v_and_b32_e32 v0, 0xffff, v0
+; GFX906-NEXT:    s_mov_b32 s4, 0x5040100
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
-; GFX906-NEXT:    v_lshl_or_b32 v0, v1, 16, v0
+; GFX906-NEXT:    v_perm_b32 v0, v1, v0, s4
 ; GFX906-NEXT:    global_store_dword v[0:1], v0, off
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
 ; GFX906-NEXT:    s_setpc_b64 s[30:31]
@@ -1637,9 +1637,9 @@
 ; GFX906:       ; %bb.0: ; %entry
 ; GFX906-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX906-NEXT:    buffer_load_sbyte v1, off, s[0:3], s32 offset:4095
-; GFX906-NEXT:    v_and_b32_e32 v0, 0xffff, v0
+; GFX906-NEXT:    s_mov_b32 s4, 0x5040100
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
-; GFX906-NEXT:    v_lshl_or_b32 v0, v1, 16, v0
+; GFX906-NEXT:    v_perm_b32 v0, v1, v0, s4
 ; GFX906-NEXT:    global_store_dword v[0:1], v0, off
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
 ; GFX906-NEXT:    s_setpc_b64 s[30:31]
@@ -1688,8 +1688,8 @@
 ; GFX906-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX906-NEXT:    buffer_load_ubyte v0, off, s[0:3], 0 offset:4094 glc
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
-; GFX906-NEXT:    v_and_b32_e32 v1, 0xffff, v1
-; GFX906-NEXT:    v_lshl_or_b32 v0, v0, 16, v1
+; GFX906-NEXT:    s_mov_b32 s4, 0x5040100
+; GFX906-NEXT:    v_perm_b32 v0, v0, v1, s4
 ; GFX906-NEXT:    global_store_dword v[0:1], v0, off
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
 ; GFX906-NEXT:    s_setpc_b64 s[30:31]
@@ -1738,8 +1738,8 @@
 ; GFX906-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX906-NEXT:    buffer_load_sbyte v0, off, s[0:3], 0 offset:4094 glc
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
-; GFX906-NEXT:    v_and_b32_e32 v1, 0xffff, v1
-; GFX906-NEXT:    v_lshl_or_b32 v0, v0, 16, v1
+; GFX906-NEXT:    s_mov_b32 s4, 0x5040100
+; GFX906-NEXT:    v_perm_b32 v0, v0, v1, s4
 ; GFX906-NEXT:    global_store_dword v[0:1], v0, off
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
 ; GFX906-NEXT:    s_setpc_b64 s[30:31]
@@ -1788,8 +1788,8 @@
 ; GFX906-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX906-NEXT:    buffer_load_ubyte v0, off, s[0:3], 0 offset:4094 glc
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
-; GFX906-NEXT:    v_and_b32_e32 v1, 0xffff, v1
-; GFX906-NEXT:    v_lshl_or_b32 v0, v0, 16, v1
+; GFX906-NEXT:    s_mov_b32 s4, 0x5040100
+; GFX906-NEXT:    v_perm_b32 v0, v0, v1, s4
 ; GFX906-NEXT:    global_store_dword v[0:1], v0, off
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
 ; GFX906-NEXT:    s_setpc_b64 s[30:31]
@@ -1838,9 +1838,9 @@
 ; GFX906:       ; %bb.0: ; %entry
 ; GFX906-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX906-NEXT:    global_load_ushort v0, v[0:1], off offset:-4094
-; GFX906-NEXT:    v_and_b32_e32 v1, 0xffff, v2
+; GFX906-NEXT:    s_mov_b32 s4, 0x5040100
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
-; GFX906-NEXT:    v_lshl_or_b32 v0, v0, 16, v1
+; GFX906-NEXT:    v_perm_b32 v0, v0, v2, s4
 ; GFX906-NEXT:    global_store_dword v[0:1], v0, off
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
 ; GFX906-NEXT:    s_setpc_b64 s[30:31]
@@ -1889,9 +1889,9 @@
 ; GFX906:       ; %bb.0: ; %entry
 ; GFX906-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX906-NEXT:    global_load_ushort v0, v[0:1], off offset:-4094
-; GFX906-NEXT:    v_and_b32_e32 v1, 0xffff, v2
+; GFX906-NEXT:    s_mov_b32 s4, 0x5040100
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
-; GFX906-NEXT:    v_lshl_or_b32 v0, v0, 16, v1
+; GFX906-NEXT:    v_perm_b32 v0, v0, v2, s4
 ; GFX906-NEXT:    global_store_dword v[0:1], v0, off
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
 ; GFX906-NEXT:    s_setpc_b64 s[30:31]
@@ -1940,9 +1940,9 @@
 ; GFX906:       ; %bb.0: ; %entry
 ; GFX906-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX906-NEXT:    global_load_sbyte v0, v[0:1], off offset:-4095
-; GFX906-NEXT:    v_and_b32_e32 v1, 0xffff, v2
+; GFX906-NEXT:    s_mov_b32 s4, 0x5040100
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
-; GFX906-NEXT:    v_lshl_or_b32 v0, v0, 16, v1
+; GFX906-NEXT:    v_perm_b32 v0, v0, v2, s4
 ; GFX906-NEXT:    global_store_dword v[0:1], v0, off
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
 ; GFX906-NEXT:    s_setpc_b64 s[30:31]
@@ -1993,9 +1993,9 @@
 ; GFX906:       ; %bb.0: ; %entry
 ; GFX906-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX906-NEXT:    global_load_ubyte v0, v[0:1], off offset:-4095
-; GFX906-NEXT:    v_and_b32_e32 v1, 0xffff, v2
+; GFX906-NEXT:    s_mov_b32 s4, 0x5040100
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
-; GFX906-NEXT:    v_lshl_or_b32 v0, v0, 16, v1
+; GFX906-NEXT:    v_perm_b32 v0, v0, v2, s4
 ; GFX906-NEXT:    global_store_dword v[0:1], v0, off
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
 ; GFX906-NEXT:    s_setpc_b64 s[30:31]
@@ -2055,9 +2055,9 @@
 ; GFX906-NEXT:    buffer_store_dword v2, v1, s[0:3], 0 offen
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
 ; GFX906-NEXT:    buffer_load_ushort v1, off, s[0:3], s32 offset:4058
-; GFX906-NEXT:    v_and_b32_e32 v0, 0xffff, v0
+; GFX906-NEXT:    s_mov_b32 s4, 0x5040100
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
-; GFX906-NEXT:    v_lshl_or_b32 v0, v1, 16, v0
+; GFX906-NEXT:    v_perm_b32 v0, v1, v0, s4
 ; GFX906-NEXT:    global_store_dword v[0:1], v0, off
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
 ; GFX906-NEXT:    s_setpc_b64 s[30:31]
@@ -2119,9 +2119,9 @@
 ; GFX906-NEXT:    buffer_store_dword v2, v1, s[0:3], 0 offen
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
 ; GFX906-NEXT:    buffer_load_sbyte v1, off, s[0:3], s32 offset:4059
-; GFX906-NEXT:    v_and_b32_e32 v0, 0xffff, v0
+; GFX906-NEXT:    s_mov_b32 s4, 0x5040100
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
-; GFX906-NEXT:    v_lshl_or_b32 v0, v1, 16, v0
+; GFX906-NEXT:    v_perm_b32 v0, v1, v0, s4
 ; GFX906-NEXT:    global_store_dword v[0:1], v0, off
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
 ; GFX906-NEXT:    s_setpc_b64 s[30:31]
@@ -2184,9 +2184,9 @@
 ; GFX906-NEXT:    buffer_store_dword v2, v1, s[0:3], 0 offen
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
 ; GFX906-NEXT:    buffer_load_ubyte v1, off, s[0:3], s32 offset:4059
-; GFX906-NEXT:    v_and_b32_e32 v0, 0xffff, v0
+; GFX906-NEXT:    s_mov_b32 s4, 0x5040100
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
-; GFX906-NEXT:    v_lshl_or_b32 v0, v1, 16, v0
+; GFX906-NEXT:    v_perm_b32 v0, v1, v0, s4
 ; GFX906-NEXT:    global_store_dword v[0:1], v0, off
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
 ; GFX906-NEXT:    s_setpc_b64 s[30:31]
@@ -2247,10 +2247,9 @@
 ; GFX906-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX906-NEXT:    ds_read_u16 v1, v0
 ; GFX906-NEXT:    ds_read_u16 v0, v0 offset:2
-; GFX906-NEXT:    s_waitcnt lgkmcnt(1)
-; GFX906-NEXT:    v_and_b32_e32 v1, 0xffff, v1
+; GFX906-NEXT:    s_mov_b32 s4, 0x5040100
 ; GFX906-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX906-NEXT:    v_lshl_or_b32 v0, v0, 16, v1
+; GFX906-NEXT:    v_perm_b32 v0, v0, v1, s4
 ; GFX906-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX803-LABEL: load_local_v2i16_split_multi_chain:
@@ -2298,10 +2297,9 @@
 ; GFX906-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX906-NEXT:    ds_read_u16 v1, v0
 ; GFX906-NEXT:    ds_read_u16 v0, v0 offset:16
-; GFX906-NEXT:    s_waitcnt lgkmcnt(1)
-; GFX906-NEXT:    v_and_b32_e32 v1, 0xffff, v1
+; GFX906-NEXT:    s_mov_b32 s4, 0x5040100
 ; GFX906-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX906-NEXT:    v_lshl_or_b32 v0, v0, 16, v1
+; GFX906-NEXT:    v_perm_b32 v0, v0, v1, s4
 ; GFX906-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX803-LABEL: load_local_lo_hi_v2i16_samechain:
@@ -2340,18 +2338,18 @@
 ; GFX900:       ; %bb.0: ; %entry
 ; GFX900-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX900-NEXT:    ds_read_u16 v0, v0
+; GFX900-NEXT:    s_mov_b32 s4, 0x5040100
 ; GFX900-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX900-NEXT:    v_and_b32_e32 v1, 0xffff, v0
-; GFX900-NEXT:    v_lshl_or_b32 v0, v0, 16, v1
+; GFX900-NEXT:    v_perm_b32 v0, v0, v0, s4
 ; GFX900-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX906-LABEL: load_local_v2i16_broadcast:
 ; GFX906:       ; %bb.0: ; %entry
 ; GFX906-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX906-NEXT:    ds_read_u16 v0, v0
+; GFX906-NEXT:    s_mov_b32 s4, 0x5040100
 ; GFX906-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX906-NEXT:    v_and_b32_e32 v1, 0xffff, v0
-; GFX906-NEXT:    v_lshl_or_b32 v0, v0, 16, v1
+; GFX906-NEXT:    v_perm_b32 v0, v0, v0, s4
 ; GFX906-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX803-LABEL: load_local_v2i16_broadcast:
@@ -2368,9 +2366,9 @@
 ; GFX900-FLATSCR:       ; %bb.0: ; %entry
 ; GFX900-FLATSCR-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX900-FLATSCR-NEXT:    ds_read_u16 v0, v0
+; GFX900-FLATSCR-NEXT:    s_mov_b32 s0, 0x5040100
 ; GFX900-FLATSCR-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX900-FLATSCR-NEXT:    v_and_b32_e32 v1, 0xffff, v0
-; GFX900-FLATSCR-NEXT:    v_lshl_or_b32 v0, v0, 16, v1
+; GFX900-FLATSCR-NEXT:    v_perm_b32 v0, v0, v0, s0
 ; GFX900-FLATSCR-NEXT:    s_setpc_b64 s[30:31]
 entry:
   %gep = getelementptr inbounds i16, i16 addrspace(3)* %in, i32 1
@@ -2400,10 +2398,9 @@
 ; GFX906-NEXT:    ds_read_u16 v2, v0
 ; GFX906-NEXT:    ds_write_b16 v1, v3
 ; GFX906-NEXT:    ds_read_u16 v0, v0 offset:16
-; GFX906-NEXT:    s_waitcnt lgkmcnt(2)
-; GFX906-NEXT:    v_and_b32_e32 v1, 0xffff, v2
+; GFX906-NEXT:    s_mov_b32 s4, 0x5040100
 ; GFX906-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX906-NEXT:    v_lshl_or_b32 v0, v0, 16, v1
+; GFX906-NEXT:    v_perm_b32 v0, v0, v2, s4
 ; GFX906-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX803-LABEL: load_local_lo_hi_v2i16_side_effect:
@@ -2459,8 +2456,8 @@
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
 ; GFX906-NEXT:    global_load_ushort v3, v[0:1], off offset:2 glc
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
-; GFX906-NEXT:    v_and_b32_e32 v0, 0xffff, v2
-; GFX906-NEXT:    v_lshl_or_b32 v0, v3, 16, v0
+; GFX906-NEXT:    s_mov_b32 s4, 0x5040100
+; GFX906-NEXT:    v_perm_b32 v0, v3, v2, s4
 ; GFX906-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX803-LABEL: load_global_v2i16_split:
@@ -2512,9 +2509,10 @@
 ; GFX906-NEXT:    flat_load_ushort v2, v[0:1] glc
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
 ; GFX906-NEXT:    flat_load_ushort v3, v[0:1] offset:2 glc
-; GFX906-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
-; GFX906-NEXT:    v_and_b32_e32 v0, 0xffff, v2
-; GFX906-NEXT:    v_lshl_or_b32 v0, v3, 16, v0
+; GFX906-NEXT:    s_waitcnt vmcnt(0)
+; GFX906-NEXT:    s_mov_b32 s4, 0x5040100
+; GFX906-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX906-NEXT:    v_perm_b32 v0, v3, v2, s4
 ; GFX906-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX803-LABEL: load_flat_v2i16_split:
@@ -2565,10 +2563,9 @@
 ; GFX906-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX906-NEXT:    global_load_ushort v2, v[0:1], off glc
 ; GFX906-NEXT:    global_load_ushort v3, v[0:1], off offset:2 glc
-; GFX906-NEXT:    s_waitcnt vmcnt(1)
-; GFX906-NEXT:    v_and_b32_e32 v0, 0xffff, v2
+; GFX906-NEXT:    s_mov_b32 s4, 0x5040100
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
-; GFX906-NEXT:    v_lshl_or_b32 v0, v3, 16, v0
+; GFX906-NEXT:    v_perm_b32 v0, v3, v2, s4
 ; GFX906-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX803-LABEL: load_constant_v2i16_split:
@@ -2620,8 +2617,8 @@
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
 ; GFX906-NEXT:    buffer_load_ushort v1, off, s[0:3], s32 offset:2 glc
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
-; GFX906-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX906-NEXT:    v_lshl_or_b32 v0, v1, 16, v0
+; GFX906-NEXT:    s_mov_b32 s4, 0x5040100
+; GFX906-NEXT:    v_perm_b32 v0, v1, v0, s4
 ; GFX906-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX803-LABEL: load_private_v2i16_split:
@@ -2671,10 +2668,10 @@
 ; GFX906:       ; %bb.0: ; %entry
 ; GFX906-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX906-NEXT:    ds_read_u16 v2, v1
-; GFX906-NEXT:    v_and_b32_e32 v3, 0xffff, v0
+; GFX906-NEXT:    s_mov_b32 s4, 0x5040100
 ; GFX906-NEXT:    ds_write_b16 v1, v0
 ; GFX906-NEXT:    s_waitcnt lgkmcnt(1)
-; GFX906-NEXT:    v_lshl_or_b32 v2, v2, 16, v3
+; GFX906-NEXT:    v_perm_b32 v2, v2, v0, s4
 ; GFX906-NEXT:    v_mov_b32_e32 v0, v2
 ; GFX906-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX906-NEXT:    s_setpc_b64 s[30:31]
@@ -2711,3 +2708,4 @@
 }
 
 attributes #0 = { nounwind }
+
diff --git a/llvm/test/CodeGen/AMDGPU/load-lo16.ll b/llvm/test/CodeGen/AMDGPU/load-lo16.ll
--- a/llvm/test/CodeGen/AMDGPU/load-lo16.ll
+++ b/llvm/test/CodeGen/AMDGPU/load-lo16.ll
@@ -33,22 +33,22 @@
 }
 
 define <2 x i16> @load_local_lo_v2i16_reglo(i16 addrspace(3)* %in, i16 %reg) #0 {
-; GFX900-LABEL: load_local_lo_v2i16_reglo:
-; GFX900:       ; %bb.0: ; %entry
-; GFX900-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX900-NEXT:    ds_read_u16 v0, v0
-; GFX900-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX900-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX900-NEXT:    v_lshl_or_b32 v0, v1, 16, v0
-; GFX900-NEXT:    s_setpc_b64 s[30:31]
+; GFX900-MUBUF-LABEL: load_local_lo_v2i16_reglo:
+; GFX900-MUBUF:       ; %bb.0: ; %entry
+; GFX900-MUBUF-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX900-MUBUF-NEXT:    ds_read_u16 v0, v0
+; GFX900-MUBUF-NEXT:    s_mov_b32 s4, 0x5040100
+; GFX900-MUBUF-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX900-MUBUF-NEXT:    v_perm_b32 v0, v1, v0, s4
+; GFX900-MUBUF-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX906-LABEL: load_local_lo_v2i16_reglo:
 ; GFX906:       ; %bb.0: ; %entry
 ; GFX906-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX906-NEXT:    ds_read_u16 v0, v0
+; GFX906-NEXT:    s_mov_b32 s4, 0x5040100
 ; GFX906-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX906-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX906-NEXT:    v_lshl_or_b32 v0, v1, 16, v0
+; GFX906-NEXT:    v_perm_b32 v0, v1, v0, s4
 ; GFX906-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX803-LABEL: load_local_lo_v2i16_reglo:
@@ -60,6 +60,15 @@
 ; GFX803-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX803-NEXT:    v_or_b32_e32 v0, v0, v1
 ; GFX803-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX900-FLATSCR-LABEL: load_local_lo_v2i16_reglo:
+; GFX900-FLATSCR:       ; %bb.0: ; %entry
+; GFX900-FLATSCR-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX900-FLATSCR-NEXT:    ds_read_u16 v0, v0
+; GFX900-FLATSCR-NEXT:    s_mov_b32 s0, 0x5040100
+; GFX900-FLATSCR-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX900-FLATSCR-NEXT:    v_perm_b32 v0, v1, v0, s0
+; GFX900-FLATSCR-NEXT:    s_setpc_b64 s[30:31]
 entry:
   %load = load i16, i16 addrspace(3)* %in
   %build0 = insertelement <2 x i16> undef, i16 %reg, i32 1
@@ -69,24 +78,24 @@
 
 ; Show that we get reasonable regalloc without physreg constraints.
 define void @load_local_lo_v2i16_reglo_vreg(i16 addrspace(3)* %in, i16 %reg) #0 {
-; GFX900-LABEL: load_local_lo_v2i16_reglo_vreg:
-; GFX900:       ; %bb.0: ; %entry
-; GFX900-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX900-NEXT:    ds_read_u16 v0, v0
-; GFX900-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX900-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX900-NEXT:    v_lshl_or_b32 v0, v1, 16, v0
-; GFX900-NEXT:    global_store_dword v[0:1], v0, off
-; GFX900-NEXT:    s_waitcnt vmcnt(0)
-; GFX900-NEXT:    s_setpc_b64 s[30:31]
+; GFX900-MUBUF-LABEL: load_local_lo_v2i16_reglo_vreg:
+; GFX900-MUBUF:       ; %bb.0: ; %entry
+; GFX900-MUBUF-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX900-MUBUF-NEXT:    ds_read_u16 v0, v0
+; GFX900-MUBUF-NEXT:    s_mov_b32 s4, 0x5040100
+; GFX900-MUBUF-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX900-MUBUF-NEXT:    v_perm_b32 v0, v1, v0, s4
+; GFX900-MUBUF-NEXT:    global_store_dword v[0:1], v0, off
+; GFX900-MUBUF-NEXT:    s_waitcnt vmcnt(0)
+; GFX900-MUBUF-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX906-LABEL: load_local_lo_v2i16_reglo_vreg:
 ; GFX906:       ; %bb.0: ; %entry
 ; GFX906-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX906-NEXT:    ds_read_u16 v0, v0
+; GFX906-NEXT:    s_mov_b32 s4, 0x5040100
 ; GFX906-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX906-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX906-NEXT:    v_lshl_or_b32 v0, v1, 16, v0
+; GFX906-NEXT:    v_perm_b32 v0, v1, v0, s4
 ; GFX906-NEXT:    global_store_dword v[0:1], v0, off
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
 ; GFX906-NEXT:    s_setpc_b64 s[30:31]
@@ -102,6 +111,17 @@
 ; GFX803-NEXT:    flat_store_dword v[0:1], v0
 ; GFX803-NEXT:    s_waitcnt vmcnt(0)
 ; GFX803-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX900-FLATSCR-LABEL: load_local_lo_v2i16_reglo_vreg:
+; GFX900-FLATSCR:       ; %bb.0: ; %entry
+; GFX900-FLATSCR-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX900-FLATSCR-NEXT:    ds_read_u16 v0, v0
+; GFX900-FLATSCR-NEXT:    s_mov_b32 s0, 0x5040100
+; GFX900-FLATSCR-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX900-FLATSCR-NEXT:    v_perm_b32 v0, v1, v0, s0
+; GFX900-FLATSCR-NEXT:    global_store_dword v[0:1], v0, off
+; GFX900-FLATSCR-NEXT:    s_waitcnt vmcnt(0)
+; GFX900-FLATSCR-NEXT:    s_setpc_b64 s[30:31]
 entry:
   %load = load i16, i16 addrspace(3)* %in
   %build0 = insertelement <2 x i16> undef, i16 %reg, i32 1
@@ -156,9 +176,9 @@
 ; GFX906-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX906-NEXT:    ds_read_u16 v0, v0
 ; GFX906-NEXT:    s_movk_i32 s4, 0x4000
+; GFX906-NEXT:    v_mov_b32_e32 v1, 0x5040100
 ; GFX906-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX906-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX906-NEXT:    v_lshl_or_b32 v0, s4, 16, v0
+; GFX906-NEXT:    v_perm_b32 v0, s4, v0, v1
 ; GFX906-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX803-LABEL: load_local_lo_v2f16_fpimm:
@@ -189,10 +209,9 @@
 ; GFX906:       ; %bb.0: ; %entry
 ; GFX906-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX906-NEXT:    ds_read_u16 v0, v0
-; GFX906-NEXT:    v_lshrrev_b32_e32 v1, 16, v1
+; GFX906-NEXT:    s_mov_b32 s4, 0x7060100
 ; GFX906-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX906-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX906-NEXT:    v_lshl_or_b32 v0, v1, 16, v0
+; GFX906-NEXT:    v_perm_b32 v0, v1, v0, s4
 ; GFX906-NEXT:    global_store_dword v[0:1], v0, off
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
 ; GFX906-NEXT:    s_setpc_b64 s[30:31]
@@ -217,24 +236,24 @@
 }
 
 define void @load_local_lo_v2f16_reglo_vreg(half addrspace(3)* %in, half %reg) #0 {
-; GFX900-LABEL: load_local_lo_v2f16_reglo_vreg:
-; GFX900:       ; %bb.0: ; %entry
-; GFX900-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX900-NEXT:    ds_read_u16 v0, v0
-; GFX900-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX900-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX900-NEXT:    v_lshl_or_b32 v0, v1, 16, v0
-; GFX900-NEXT:    global_store_dword v[0:1], v0, off
-; GFX900-NEXT:    s_waitcnt vmcnt(0)
-; GFX900-NEXT:    s_setpc_b64 s[30:31]
+; GFX900-MUBUF-LABEL: load_local_lo_v2f16_reglo_vreg:
+; GFX900-MUBUF:       ; %bb.0: ; %entry
+; GFX900-MUBUF-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX900-MUBUF-NEXT:    ds_read_u16 v0, v0
+; GFX900-MUBUF-NEXT:    s_mov_b32 s4, 0x5040100
+; GFX900-MUBUF-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX900-MUBUF-NEXT:    v_perm_b32 v0, v1, v0, s4
+; GFX900-MUBUF-NEXT:    global_store_dword v[0:1], v0, off
+; GFX900-MUBUF-NEXT:    s_waitcnt vmcnt(0)
+; GFX900-MUBUF-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX906-LABEL: load_local_lo_v2f16_reglo_vreg:
 ; GFX906:       ; %bb.0: ; %entry
 ; GFX906-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX906-NEXT:    ds_read_u16 v0, v0
+; GFX906-NEXT:    s_mov_b32 s4, 0x5040100
 ; GFX906-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX906-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX906-NEXT:    v_lshl_or_b32 v0, v1, 16, v0
+; GFX906-NEXT:    v_perm_b32 v0, v1, v0, s4
 ; GFX906-NEXT:    global_store_dword v[0:1], v0, off
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
 ; GFX906-NEXT:    s_setpc_b64 s[30:31]
@@ -250,6 +269,17 @@
 ; GFX803-NEXT:    flat_store_dword v[0:1], v0
 ; GFX803-NEXT:    s_waitcnt vmcnt(0)
 ; GFX803-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX900-FLATSCR-LABEL: load_local_lo_v2f16_reglo_vreg:
+; GFX900-FLATSCR:       ; %bb.0: ; %entry
+; GFX900-FLATSCR-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX900-FLATSCR-NEXT:    ds_read_u16 v0, v0
+; GFX900-FLATSCR-NEXT:    s_mov_b32 s0, 0x5040100
+; GFX900-FLATSCR-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX900-FLATSCR-NEXT:    v_perm_b32 v0, v1, v0, s0
+; GFX900-FLATSCR-NEXT:    global_store_dword v[0:1], v0, off
+; GFX900-FLATSCR-NEXT:    s_waitcnt vmcnt(0)
+; GFX900-FLATSCR-NEXT:    s_setpc_b64 s[30:31]
 entry:
   %load = load half, half addrspace(3)* %in
   %build0 = insertelement <2 x half> undef, half %reg, i32 1
@@ -272,9 +302,9 @@
 ; GFX906:       ; %bb.0: ; %entry
 ; GFX906-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX906-NEXT:    ds_read_u8 v0, v0
-; GFX906-NEXT:    v_mov_b32_e32 v2, 0xffff
+; GFX906-NEXT:    s_mov_b32 s4, 0x7060100
 ; GFX906-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX906-NEXT:    v_bfi_b32 v0, v2, v0, v1
+; GFX906-NEXT:    v_perm_b32 v0, v1, v0, s4
 ; GFX906-NEXT:    global_store_dword v[0:1], v0, off
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
 ; GFX906-NEXT:    s_setpc_b64 s[30:31]
@@ -300,24 +330,24 @@
 }
 
 define void @load_local_lo_v2i16_reglo_vreg_zexti8(i8 addrspace(3)* %in, i16 %reg) #0 {
-; GFX900-LABEL: load_local_lo_v2i16_reglo_vreg_zexti8:
-; GFX900:       ; %bb.0: ; %entry
-; GFX900-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX900-NEXT:    ds_read_u8 v0, v0
-; GFX900-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX900-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX900-NEXT:    v_lshl_or_b32 v0, v1, 16, v0
-; GFX900-NEXT:    global_store_dword v[0:1], v0, off
-; GFX900-NEXT:    s_waitcnt vmcnt(0)
-; GFX900-NEXT:    s_setpc_b64 s[30:31]
+; GFX900-MUBUF-LABEL: load_local_lo_v2i16_reglo_vreg_zexti8:
+; GFX900-MUBUF:       ; %bb.0: ; %entry
+; GFX900-MUBUF-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX900-MUBUF-NEXT:    ds_read_u8 v0, v0
+; GFX900-MUBUF-NEXT:    s_mov_b32 s4, 0x5040100
+; GFX900-MUBUF-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX900-MUBUF-NEXT:    v_perm_b32 v0, v1, v0, s4
+; GFX900-MUBUF-NEXT:    global_store_dword v[0:1], v0, off
+; GFX900-MUBUF-NEXT:    s_waitcnt vmcnt(0)
+; GFX900-MUBUF-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX906-LABEL: load_local_lo_v2i16_reglo_vreg_zexti8:
 ; GFX906:       ; %bb.0: ; %entry
 ; GFX906-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX906-NEXT:    ds_read_u8 v0, v0
+; GFX906-NEXT:    s_mov_b32 s4, 0x5040100
 ; GFX906-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX906-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX906-NEXT:    v_lshl_or_b32 v0, v1, 16, v0
+; GFX906-NEXT:    v_perm_b32 v0, v1, v0, s4
 ; GFX906-NEXT:    global_store_dword v[0:1], v0, off
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
 ; GFX906-NEXT:    s_setpc_b64 s[30:31]
@@ -333,6 +363,17 @@
 ; GFX803-NEXT:    flat_store_dword v[0:1], v0
 ; GFX803-NEXT:    s_waitcnt vmcnt(0)
 ; GFX803-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX900-FLATSCR-LABEL: load_local_lo_v2i16_reglo_vreg_zexti8:
+; GFX900-FLATSCR:       ; %bb.0: ; %entry
+; GFX900-FLATSCR-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX900-FLATSCR-NEXT:    ds_read_u8 v0, v0
+; GFX900-FLATSCR-NEXT:    s_mov_b32 s0, 0x5040100
+; GFX900-FLATSCR-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX900-FLATSCR-NEXT:    v_perm_b32 v0, v1, v0, s0
+; GFX900-FLATSCR-NEXT:    global_store_dword v[0:1], v0, off
+; GFX900-FLATSCR-NEXT:    s_waitcnt vmcnt(0)
+; GFX900-FLATSCR-NEXT:    s_setpc_b64 s[30:31]
 entry:
   %load = load i8, i8 addrspace(3)* %in
   %ext = zext i8 %load to i16
@@ -356,9 +397,9 @@
 ; GFX906:       ; %bb.0: ; %entry
 ; GFX906-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX906-NEXT:    ds_read_i8 v0, v0
-; GFX906-NEXT:    v_mov_b32_e32 v2, 0xffff
+; GFX906-NEXT:    s_mov_b32 s4, 0x7060100
 ; GFX906-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX906-NEXT:    v_bfi_b32 v0, v2, v0, v1
+; GFX906-NEXT:    v_perm_b32 v0, v1, v0, s4
 ; GFX906-NEXT:    global_store_dword v[0:1], v0, off
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
 ; GFX906-NEXT:    s_setpc_b64 s[30:31]
@@ -384,24 +425,24 @@
 }
 
 define void @load_local_lo_v2i16_reglo_vreg_sexti8(i8 addrspace(3)* %in, i16 %reg) #0 {
-; GFX900-LABEL: load_local_lo_v2i16_reglo_vreg_sexti8:
-; GFX900:       ; %bb.0: ; %entry
-; GFX900-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX900-NEXT:    ds_read_i8 v0, v0
-; GFX900-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX900-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX900-NEXT:    v_lshl_or_b32 v0, v1, 16, v0
-; GFX900-NEXT:    global_store_dword v[0:1], v0, off
-; GFX900-NEXT:    s_waitcnt vmcnt(0)
-; GFX900-NEXT:    s_setpc_b64 s[30:31]
+; GFX900-MUBUF-LABEL: load_local_lo_v2i16_reglo_vreg_sexti8:
+; GFX900-MUBUF:       ; %bb.0: ; %entry
+; GFX900-MUBUF-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX900-MUBUF-NEXT:    ds_read_i8 v0, v0
+; GFX900-MUBUF-NEXT:    s_mov_b32 s4, 0x5040100
+; GFX900-MUBUF-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX900-MUBUF-NEXT:    v_perm_b32 v0, v1, v0, s4
+; GFX900-MUBUF-NEXT:    global_store_dword v[0:1], v0, off
+; GFX900-MUBUF-NEXT:    s_waitcnt vmcnt(0)
+; GFX900-MUBUF-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX906-LABEL: load_local_lo_v2i16_reglo_vreg_sexti8:
 ; GFX906:       ; %bb.0: ; %entry
 ; GFX906-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX906-NEXT:    ds_read_i8 v0, v0
+; GFX906-NEXT:    s_mov_b32 s4, 0x5040100
 ; GFX906-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX906-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX906-NEXT:    v_lshl_or_b32 v0, v1, 16, v0
+; GFX906-NEXT:    v_perm_b32 v0, v1, v0, s4
 ; GFX906-NEXT:    global_store_dword v[0:1], v0, off
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
 ; GFX906-NEXT:    s_setpc_b64 s[30:31]
@@ -417,6 +458,17 @@
 ; GFX803-NEXT:    flat_store_dword v[0:1], v0
 ; GFX803-NEXT:    s_waitcnt vmcnt(0)
 ; GFX803-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX900-FLATSCR-LABEL: load_local_lo_v2i16_reglo_vreg_sexti8:
+; GFX900-FLATSCR:       ; %bb.0: ; %entry
+; GFX900-FLATSCR-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX900-FLATSCR-NEXT:    ds_read_i8 v0, v0
+; GFX900-FLATSCR-NEXT:    s_mov_b32 s0, 0x5040100
+; GFX900-FLATSCR-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX900-FLATSCR-NEXT:    v_perm_b32 v0, v1, v0, s0
+; GFX900-FLATSCR-NEXT:    global_store_dword v[0:1], v0, off
+; GFX900-FLATSCR-NEXT:    s_waitcnt vmcnt(0)
+; GFX900-FLATSCR-NEXT:    s_setpc_b64 s[30:31]
 entry:
   %load = load i8, i8 addrspace(3)* %in
   %ext = sext i8 %load to i16
@@ -427,24 +479,24 @@
 }
 
 define void @load_local_lo_v2f16_reglo_vreg_zexti8(i8 addrspace(3)* %in, half %reg) #0 {
-; GFX900-LABEL: load_local_lo_v2f16_reglo_vreg_zexti8:
-; GFX900:       ; %bb.0: ; %entry
-; GFX900-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX900-NEXT:    ds_read_u8 v0, v0
-; GFX900-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX900-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX900-NEXT:    v_lshl_or_b32 v0, v1, 16, v0
-; GFX900-NEXT:    global_store_dword v[0:1], v0, off
-; GFX900-NEXT:    s_waitcnt vmcnt(0)
-; GFX900-NEXT:    s_setpc_b64 s[30:31]
+; GFX900-MUBUF-LABEL: load_local_lo_v2f16_reglo_vreg_zexti8:
+; GFX900-MUBUF:       ; %bb.0: ; %entry
+; GFX900-MUBUF-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX900-MUBUF-NEXT:    ds_read_u8 v0, v0
+; GFX900-MUBUF-NEXT:    s_mov_b32 s4, 0x5040100
+; GFX900-MUBUF-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX900-MUBUF-NEXT:    v_perm_b32 v0, v1, v0, s4
+; GFX900-MUBUF-NEXT:    global_store_dword v[0:1], v0, off
+; GFX900-MUBUF-NEXT:    s_waitcnt vmcnt(0)
+; GFX900-MUBUF-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX906-LABEL: load_local_lo_v2f16_reglo_vreg_zexti8:
 ; GFX906:       ; %bb.0: ; %entry
 ; GFX906-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX906-NEXT:    ds_read_u8 v0, v0
+; GFX906-NEXT:    s_mov_b32 s4, 0x5040100
 ; GFX906-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX906-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX906-NEXT:    v_lshl_or_b32 v0, v1, 16, v0
+; GFX906-NEXT:    v_perm_b32 v0, v1, v0, s4
 ; GFX906-NEXT:    global_store_dword v[0:1], v0, off
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
 ; GFX906-NEXT:    s_setpc_b64 s[30:31]
@@ -460,6 +512,17 @@
 ; GFX803-NEXT:    flat_store_dword v[0:1], v0
 ; GFX803-NEXT:    s_waitcnt vmcnt(0)
 ; GFX803-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX900-FLATSCR-LABEL: load_local_lo_v2f16_reglo_vreg_zexti8:
+; GFX900-FLATSCR:       ; %bb.0: ; %entry
+; GFX900-FLATSCR-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX900-FLATSCR-NEXT:    ds_read_u8 v0, v0
+; GFX900-FLATSCR-NEXT:    s_mov_b32 s0, 0x5040100
+; GFX900-FLATSCR-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX900-FLATSCR-NEXT:    v_perm_b32 v0, v1, v0, s0
+; GFX900-FLATSCR-NEXT:    global_store_dword v[0:1], v0, off
+; GFX900-FLATSCR-NEXT:    s_waitcnt vmcnt(0)
+; GFX900-FLATSCR-NEXT:    s_setpc_b64 s[30:31]
 entry:
   %load = load i8, i8 addrspace(3)* %in
   %ext = zext i8 %load to i16
@@ -471,24 +534,24 @@
 }
 
 define void @load_local_lo_v2f16_reglo_vreg_sexti8(i8 addrspace(3)* %in, half %reg) #0 {
-; GFX900-LABEL: load_local_lo_v2f16_reglo_vreg_sexti8:
-; GFX900:       ; %bb.0: ; %entry
-; GFX900-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX900-NEXT:    ds_read_i8 v0, v0
-; GFX900-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX900-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX900-NEXT:    v_lshl_or_b32 v0, v1, 16, v0
-; GFX900-NEXT:    global_store_dword v[0:1], v0, off
-; GFX900-NEXT:    s_waitcnt vmcnt(0)
-; GFX900-NEXT:    s_setpc_b64 s[30:31]
+; GFX900-MUBUF-LABEL: load_local_lo_v2f16_reglo_vreg_sexti8:
+; GFX900-MUBUF:       ; %bb.0: ; %entry
+; GFX900-MUBUF-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX900-MUBUF-NEXT:    ds_read_i8 v0, v0
+; GFX900-MUBUF-NEXT:    s_mov_b32 s4, 0x5040100
+; GFX900-MUBUF-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX900-MUBUF-NEXT:    v_perm_b32 v0, v1, v0, s4
+; GFX900-MUBUF-NEXT:    global_store_dword v[0:1], v0, off
+; GFX900-MUBUF-NEXT:    s_waitcnt vmcnt(0)
+; GFX900-MUBUF-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX906-LABEL: load_local_lo_v2f16_reglo_vreg_sexti8:
 ; GFX906:       ; %bb.0: ; %entry
 ; GFX906-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX906-NEXT:    ds_read_i8 v0, v0
+; GFX906-NEXT:    s_mov_b32 s4, 0x5040100
 ; GFX906-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX906-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX906-NEXT:    v_lshl_or_b32 v0, v1, 16, v0
+; GFX906-NEXT:    v_perm_b32 v0, v1, v0, s4
 ; GFX906-NEXT:    global_store_dword v[0:1], v0, off
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
 ; GFX906-NEXT:    s_setpc_b64 s[30:31]
@@ -504,6 +567,17 @@
 ; GFX803-NEXT:    flat_store_dword v[0:1], v0
 ; GFX803-NEXT:    s_waitcnt vmcnt(0)
 ; GFX803-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX900-FLATSCR-LABEL: load_local_lo_v2f16_reglo_vreg_sexti8:
+; GFX900-FLATSCR:       ; %bb.0: ; %entry
+; GFX900-FLATSCR-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX900-FLATSCR-NEXT:    ds_read_i8 v0, v0
+; GFX900-FLATSCR-NEXT:    s_mov_b32 s0, 0x5040100
+; GFX900-FLATSCR-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX900-FLATSCR-NEXT:    v_perm_b32 v0, v1, v0, s0
+; GFX900-FLATSCR-NEXT:    global_store_dword v[0:1], v0, off
+; GFX900-FLATSCR-NEXT:    s_waitcnt vmcnt(0)
+; GFX900-FLATSCR-NEXT:    s_setpc_b64 s[30:31]
 entry:
   %load = load i8, i8 addrspace(3)* %in
   %ext = sext i8 %load to i16
@@ -515,28 +589,28 @@
 }
 
 define void @load_local_lo_v2i16_reghi_vreg_multi_use_lo(i16 addrspace(3)* %in, <2 x i16> %reg) #0 {
-; GFX900-LABEL: load_local_lo_v2i16_reghi_vreg_multi_use_lo:
-; GFX900:       ; %bb.0: ; %entry
-; GFX900-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX900-NEXT:    ds_read_u16 v0, v0
-; GFX900-NEXT:    v_mov_b32_e32 v2, 0
-; GFX900-NEXT:    v_mov_b32_e32 v3, 0xffff
-; GFX900-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX900-NEXT:    ds_write_b16 v2, v0
-; GFX900-NEXT:    v_bfi_b32 v0, v3, v0, v1
-; GFX900-NEXT:    global_store_dword v[0:1], v0, off
-; GFX900-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
-; GFX900-NEXT:    s_setpc_b64 s[30:31]
+; GFX900-MUBUF-LABEL: load_local_lo_v2i16_reghi_vreg_multi_use_lo:
+; GFX900-MUBUF:       ; %bb.0: ; %entry
+; GFX900-MUBUF-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX900-MUBUF-NEXT:    ds_read_u16 v0, v0
+; GFX900-MUBUF-NEXT:    v_mov_b32_e32 v2, 0
+; GFX900-MUBUF-NEXT:    s_mov_b32 s4, 0x7060100
+; GFX900-MUBUF-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX900-MUBUF-NEXT:    ds_write_b16 v2, v0
+; GFX900-MUBUF-NEXT:    v_perm_b32 v0, v1, v0, s4
+; GFX900-MUBUF-NEXT:    global_store_dword v[0:1], v0, off
+; GFX900-MUBUF-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; GFX900-MUBUF-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX906-LABEL: load_local_lo_v2i16_reghi_vreg_multi_use_lo:
 ; GFX906:       ; %bb.0: ; %entry
 ; GFX906-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX906-NEXT:    ds_read_u16 v0, v0
 ; GFX906-NEXT:    v_mov_b32_e32 v2, 0
-; GFX906-NEXT:    v_mov_b32_e32 v3, 0xffff
+; GFX906-NEXT:    s_mov_b32 s4, 0x7060100
 ; GFX906-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX906-NEXT:    ds_write_b16 v2, v0
-; GFX906-NEXT:    v_bfi_b32 v0, v3, v0, v1
+; GFX906-NEXT:    v_perm_b32 v0, v1, v0, s4
 ; GFX906-NEXT:    global_store_dword v[0:1], v0, off
 ; GFX906-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX906-NEXT:    s_setpc_b64 s[30:31]
@@ -554,6 +628,19 @@
 ; GFX803-NEXT:    flat_store_dword v[0:1], v0
 ; GFX803-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX803-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX900-FLATSCR-LABEL: load_local_lo_v2i16_reghi_vreg_multi_use_lo:
+; GFX900-FLATSCR:       ; %bb.0: ; %entry
+; GFX900-FLATSCR-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX900-FLATSCR-NEXT:    ds_read_u16 v0, v0
+; GFX900-FLATSCR-NEXT:    v_mov_b32_e32 v2, 0
+; GFX900-FLATSCR-NEXT:    s_mov_b32 s0, 0x7060100
+; GFX900-FLATSCR-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX900-FLATSCR-NEXT:    ds_write_b16 v2, v0
+; GFX900-FLATSCR-NEXT:    v_perm_b32 v0, v1, v0, s0
+; GFX900-FLATSCR-NEXT:    global_store_dword v[0:1], v0, off
+; GFX900-FLATSCR-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; GFX900-FLATSCR-NEXT:    s_setpc_b64 s[30:31]
 entry:
   %load = load i16, i16 addrspace(3)* %in
   %elt1 = extractelement <2 x i16> %reg, i32 1
@@ -580,12 +667,12 @@
 ; GFX906:       ; %bb.0: ; %entry
 ; GFX906-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX906-NEXT:    ds_read_u16 v0, v0
+; GFX906-NEXT:    s_mov_b32 s4, 0x7060100
 ; GFX906-NEXT:    v_lshrrev_b32_e32 v2, 16, v1
 ; GFX906-NEXT:    v_mov_b32_e32 v3, 0
 ; GFX906-NEXT:    ds_write_b16 v3, v2
-; GFX906-NEXT:    v_mov_b32_e32 v2, 0xffff
 ; GFX906-NEXT:    s_waitcnt lgkmcnt(1)
-; GFX906-NEXT:    v_bfi_b32 v0, v2, v0, v1
+; GFX906-NEXT:    v_perm_b32 v0, v1, v0, s4
 ; GFX906-NEXT:    global_store_dword v[0:1], v0, off
 ; GFX906-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX906-NEXT:    s_setpc_b64 s[30:31]
@@ -614,30 +701,30 @@
 }
 
 define void @load_local_lo_v2i16_reghi_vreg_multi_use_lohi(i16 addrspace(3)* noalias %in, <2 x i16> %reg, i16 addrspace(3)* noalias %out0, i16 addrspace(3)* noalias %out1) #0 {
-; GFX900-LABEL: load_local_lo_v2i16_reghi_vreg_multi_use_lohi:
-; GFX900:       ; %bb.0: ; %entry
-; GFX900-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX900-NEXT:    ds_read_u16 v0, v0
-; GFX900-NEXT:    v_lshrrev_b32_e32 v4, 16, v1
-; GFX900-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX900-NEXT:    ds_write_b16 v2, v0
-; GFX900-NEXT:    ds_write_b16 v3, v4
-; GFX900-NEXT:    v_mov_b32_e32 v2, 0xffff
-; GFX900-NEXT:    v_bfi_b32 v0, v2, v0, v1
-; GFX900-NEXT:    global_store_dword v[0:1], v0, off
-; GFX900-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
-; GFX900-NEXT:    s_setpc_b64 s[30:31]
+; GFX900-MUBUF-LABEL: load_local_lo_v2i16_reghi_vreg_multi_use_lohi:
+; GFX900-MUBUF:       ; %bb.0: ; %entry
+; GFX900-MUBUF-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX900-MUBUF-NEXT:    ds_read_u16 v0, v0
+; GFX900-MUBUF-NEXT:    s_mov_b32 s4, 0x7060100
+; GFX900-MUBUF-NEXT:    v_lshrrev_b32_e32 v4, 16, v1
+; GFX900-MUBUF-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX900-MUBUF-NEXT:    ds_write_b16 v2, v0
+; GFX900-MUBUF-NEXT:    ds_write_b16 v3, v4
+; GFX900-MUBUF-NEXT:    v_perm_b32 v0, v1, v0, s4
+; GFX900-MUBUF-NEXT:    global_store_dword v[0:1], v0, off
+; GFX900-MUBUF-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; GFX900-MUBUF-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX906-LABEL: load_local_lo_v2i16_reghi_vreg_multi_use_lohi:
 ; GFX906:       ; %bb.0: ; %entry
 ; GFX906-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX906-NEXT:    ds_read_u16 v0, v0
+; GFX906-NEXT:    s_mov_b32 s4, 0x7060100
 ; GFX906-NEXT:    v_lshrrev_b32_e32 v4, 16, v1
 ; GFX906-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX906-NEXT:    ds_write_b16 v2, v0
 ; GFX906-NEXT:    ds_write_b16 v3, v4
-; GFX906-NEXT:    v_mov_b32_e32 v2, 0xffff
-; GFX906-NEXT:    v_bfi_b32 v0, v2, v0, v1
+; GFX906-NEXT:    v_perm_b32 v0, v1, v0, s4
 ; GFX906-NEXT:    global_store_dword v[0:1], v0, off
 ; GFX906-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX906-NEXT:    s_setpc_b64 s[30:31]
@@ -656,6 +743,20 @@
 ; GFX803-NEXT:    flat_store_dword v[0:1], v0
 ; GFX803-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX803-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX900-FLATSCR-LABEL: load_local_lo_v2i16_reghi_vreg_multi_use_lohi:
+; GFX900-FLATSCR:       ; %bb.0: ; %entry
+; GFX900-FLATSCR-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX900-FLATSCR-NEXT:    ds_read_u16 v0, v0
+; GFX900-FLATSCR-NEXT:    s_mov_b32 s0, 0x7060100
+; GFX900-FLATSCR-NEXT:    v_lshrrev_b32_e32 v4, 16, v1
+; GFX900-FLATSCR-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX900-FLATSCR-NEXT:    ds_write_b16 v2, v0
+; GFX900-FLATSCR-NEXT:    ds_write_b16 v3, v4
+; GFX900-FLATSCR-NEXT:    v_perm_b32 v0, v1, v0, s0
+; GFX900-FLATSCR-NEXT:    global_store_dword v[0:1], v0, off
+; GFX900-FLATSCR-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; GFX900-FLATSCR-NEXT:    s_setpc_b64 s[30:31]
 entry:
   %load = load i16, i16 addrspace(3)* %in
   %elt1 = extractelement <2 x i16> %reg, i32 1
@@ -680,9 +781,9 @@
 ; GFX906:       ; %bb.0: ; %entry
 ; GFX906-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX906-NEXT:    global_load_ushort v0, v[0:1], off offset:-4094
-; GFX906-NEXT:    v_mov_b32_e32 v1, 0xffff
+; GFX906-NEXT:    s_mov_b32 s4, 0x7060100
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
-; GFX906-NEXT:    v_bfi_b32 v0, v1, v0, v2
+; GFX906-NEXT:    v_perm_b32 v0, v2, v0, s4
 ; GFX906-NEXT:    global_store_dword v[0:1], v0, off
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
 ; GFX906-NEXT:    s_setpc_b64 s[30:31]
@@ -722,10 +823,9 @@
 ; GFX906:       ; %bb.0: ; %entry
 ; GFX906-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX906-NEXT:    global_load_ushort v0, v[0:1], off offset:-4094
-; GFX906-NEXT:    v_lshrrev_b32_e32 v1, 16, v2
+; GFX906-NEXT:    s_mov_b32 s4, 0x7060100
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
-; GFX906-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX906-NEXT:    v_lshl_or_b32 v0, v1, 16, v0
+; GFX906-NEXT:    v_perm_b32 v0, v2, v0, s4
 ; GFX906-NEXT:    global_store_dword v[0:1], v0, off
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
 ; GFX906-NEXT:    s_setpc_b64 s[30:31]
@@ -765,9 +865,9 @@
 ; GFX906:       ; %bb.0: ; %entry
 ; GFX906-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX906-NEXT:    global_load_ubyte v0, v[0:1], off offset:-4095
-; GFX906-NEXT:    v_mov_b32_e32 v1, 0xffff
+; GFX906-NEXT:    s_mov_b32 s4, 0x7060100
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
-; GFX906-NEXT:    v_bfi_b32 v0, v1, v0, v2
+; GFX906-NEXT:    v_perm_b32 v0, v2, v0, s4
 ; GFX906-NEXT:    global_store_dword v[0:1], v0, off
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
 ; GFX906-NEXT:    s_setpc_b64 s[30:31]
@@ -808,9 +908,9 @@
 ; GFX906:       ; %bb.0: ; %entry
 ; GFX906-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX906-NEXT:    global_load_sbyte v0, v[0:1], off offset:-4095
-; GFX906-NEXT:    v_mov_b32_e32 v1, 0xffff
+; GFX906-NEXT:    s_mov_b32 s4, 0x7060100
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
-; GFX906-NEXT:    v_bfi_b32 v0, v1, v0, v2
+; GFX906-NEXT:    v_perm_b32 v0, v2, v0, s4
 ; GFX906-NEXT:    global_store_dword v[0:1], v0, off
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
 ; GFX906-NEXT:    s_setpc_b64 s[30:31]
@@ -851,10 +951,9 @@
 ; GFX906:       ; %bb.0: ; %entry
 ; GFX906-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX906-NEXT:    global_load_ubyte v0, v[0:1], off offset:-4095
-; GFX906-NEXT:    v_lshrrev_b32_e32 v1, 16, v2
+; GFX906-NEXT:    s_mov_b32 s4, 0x7060100
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
-; GFX906-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX906-NEXT:    v_lshl_or_b32 v0, v1, 16, v0
+; GFX906-NEXT:    v_perm_b32 v0, v2, v0, s4
 ; GFX906-NEXT:    global_store_dword v[0:1], v0, off
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
 ; GFX906-NEXT:    s_setpc_b64 s[30:31]
@@ -896,10 +995,9 @@
 ; GFX906:       ; %bb.0: ; %entry
 ; GFX906-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX906-NEXT:    global_load_sbyte v0, v[0:1], off offset:-4095
-; GFX906-NEXT:    v_lshrrev_b32_e32 v1, 16, v2
+; GFX906-NEXT:    s_mov_b32 s4, 0x7060100
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
-; GFX906-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX906-NEXT:    v_lshl_or_b32 v0, v1, 16, v0
+; GFX906-NEXT:    v_perm_b32 v0, v2, v0, s4
 ; GFX906-NEXT:    global_store_dword v[0:1], v0, off
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
 ; GFX906-NEXT:    s_setpc_b64 s[30:31]
@@ -941,9 +1039,9 @@
 ; GFX906:       ; %bb.0: ; %entry
 ; GFX906-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX906-NEXT:    flat_load_ushort v0, v[0:1]
-; GFX906-NEXT:    v_mov_b32_e32 v1, 0xffff
+; GFX906-NEXT:    s_mov_b32 s4, 0x7060100
 ; GFX906-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
-; GFX906-NEXT:    v_bfi_b32 v0, v1, v0, v2
+; GFX906-NEXT:    v_perm_b32 v0, v2, v0, s4
 ; GFX906-NEXT:    global_store_dword v[0:1], v0, off
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
 ; GFX906-NEXT:    s_setpc_b64 s[30:31]
@@ -980,10 +1078,9 @@
 ; GFX906:       ; %bb.0: ; %entry
 ; GFX906-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX906-NEXT:    flat_load_ushort v0, v[0:1]
-; GFX906-NEXT:    v_lshrrev_b32_e32 v1, 16, v2
+; GFX906-NEXT:    s_mov_b32 s4, 0x7060100
 ; GFX906-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
-; GFX906-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX906-NEXT:    v_lshl_or_b32 v0, v1, 16, v0
+; GFX906-NEXT:    v_perm_b32 v0, v2, v0, s4
 ; GFX906-NEXT:    global_store_dword v[0:1], v0, off
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
 ; GFX906-NEXT:    s_setpc_b64 s[30:31]
@@ -1022,9 +1119,9 @@
 ; GFX906:       ; %bb.0: ; %entry
 ; GFX906-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX906-NEXT:    flat_load_ubyte v0, v[0:1]
-; GFX906-NEXT:    v_mov_b32_e32 v1, 0xffff
+; GFX906-NEXT:    s_mov_b32 s4, 0x7060100
 ; GFX906-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
-; GFX906-NEXT:    v_bfi_b32 v0, v1, v0, v2
+; GFX906-NEXT:    v_perm_b32 v0, v2, v0, s4
 ; GFX906-NEXT:    global_store_dword v[0:1], v0, off
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
 ; GFX906-NEXT:    s_setpc_b64 s[30:31]
@@ -1062,9 +1159,9 @@
 ; GFX906:       ; %bb.0: ; %entry
 ; GFX906-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX906-NEXT:    flat_load_sbyte v0, v[0:1]
-; GFX906-NEXT:    v_mov_b32_e32 v1, 0xffff
+; GFX906-NEXT:    s_mov_b32 s4, 0x7060100
 ; GFX906-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
-; GFX906-NEXT:    v_bfi_b32 v0, v1, v0, v2
+; GFX906-NEXT:    v_perm_b32 v0, v2, v0, s4
 ; GFX906-NEXT:    global_store_dword v[0:1], v0, off
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
 ; GFX906-NEXT:    s_setpc_b64 s[30:31]
@@ -1102,10 +1199,9 @@
 ; GFX906:       ; %bb.0: ; %entry
 ; GFX906-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX906-NEXT:    flat_load_ubyte v0, v[0:1]
-; GFX906-NEXT:    v_lshrrev_b32_e32 v1, 16, v2
+; GFX906-NEXT:    s_mov_b32 s4, 0x7060100
 ; GFX906-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
-; GFX906-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX906-NEXT:    v_lshl_or_b32 v0, v1, 16, v0
+; GFX906-NEXT:    v_perm_b32 v0, v2, v0, s4
 ; GFX906-NEXT:    global_store_dword v[0:1], v0, off
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
 ; GFX906-NEXT:    s_setpc_b64 s[30:31]
@@ -1144,10 +1240,9 @@
 ; GFX906:       ; %bb.0: ; %entry
 ; GFX906-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX906-NEXT:    flat_load_sbyte v0, v[0:1]
-; GFX906-NEXT:    v_lshrrev_b32_e32 v1, 16, v2
+; GFX906-NEXT:    s_mov_b32 s4, 0x7060100
 ; GFX906-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
-; GFX906-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX906-NEXT:    v_lshl_or_b32 v0, v1, 16, v0
+; GFX906-NEXT:    v_perm_b32 v0, v2, v0, s4
 ; GFX906-NEXT:    global_store_dword v[0:1], v0, off
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
 ; GFX906-NEXT:    s_setpc_b64 s[30:31]
@@ -1186,9 +1281,9 @@
 ; GFX906:       ; %bb.0: ; %entry
 ; GFX906-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX906-NEXT:    buffer_load_ushort v1, off, s[0:3], s32 offset:4094
-; GFX906-NEXT:    v_mov_b32_e32 v2, 0xffff
+; GFX906-NEXT:    s_mov_b32 s4, 0x7060100
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
-; GFX906-NEXT:    v_bfi_b32 v0, v2, v1, v0
+; GFX906-NEXT:    v_perm_b32 v0, v0, v1, s4
 ; GFX906-NEXT:    global_store_dword v[0:1], v0, off
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
 ; GFX906-NEXT:    s_setpc_b64 s[30:31]
@@ -1226,9 +1321,9 @@
 ; GFX900-MUBUF:       ; %bb.0: ; %entry
 ; GFX900-MUBUF-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX900-MUBUF-NEXT:    buffer_load_ushort v1, off, s[0:3], s32 offset:4094
+; GFX900-MUBUF-NEXT:    s_mov_b32 s4, 0x5040100
 ; GFX900-MUBUF-NEXT:    s_waitcnt vmcnt(0)
-; GFX900-MUBUF-NEXT:    v_and_b32_e32 v1, 0xffff, v1
-; GFX900-MUBUF-NEXT:    v_lshl_or_b32 v0, v0, 16, v1
+; GFX900-MUBUF-NEXT:    v_perm_b32 v0, v0, v1, s4
 ; GFX900-MUBUF-NEXT:    global_store_dword v[0:1], v0, off
 ; GFX900-MUBUF-NEXT:    s_waitcnt vmcnt(0)
 ; GFX900-MUBUF-NEXT:    s_setpc_b64 s[30:31]
@@ -1237,9 +1332,9 @@
 ; GFX906:       ; %bb.0: ; %entry
 ; GFX906-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX906-NEXT:    buffer_load_ushort v1, off, s[0:3], s32 offset:4094
+; GFX906-NEXT:    s_mov_b32 s4, 0x5040100
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
-; GFX906-NEXT:    v_and_b32_e32 v1, 0xffff, v1
-; GFX906-NEXT:    v_lshl_or_b32 v0, v0, 16, v1
+; GFX906-NEXT:    v_perm_b32 v0, v0, v1, s4
 ; GFX906-NEXT:    global_store_dword v[0:1], v0, off
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
 ; GFX906-NEXT:    s_setpc_b64 s[30:31]
@@ -1259,9 +1354,9 @@
 ; GFX900-FLATSCR:       ; %bb.0: ; %entry
 ; GFX900-FLATSCR-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX900-FLATSCR-NEXT:    scratch_load_ushort v1, off, s32 offset:4094
+; GFX900-FLATSCR-NEXT:    s_mov_b32 s0, 0x5040100
 ; GFX900-FLATSCR-NEXT:    s_waitcnt vmcnt(0)
-; GFX900-FLATSCR-NEXT:    v_and_b32_e32 v1, 0xffff, v1
-; GFX900-FLATSCR-NEXT:    v_lshl_or_b32 v0, v0, 16, v1
+; GFX900-FLATSCR-NEXT:    v_perm_b32 v0, v0, v1, s0
 ; GFX900-FLATSCR-NEXT:    global_store_dword v[0:1], v0, off
 ; GFX900-FLATSCR-NEXT:    s_waitcnt vmcnt(0)
 ; GFX900-FLATSCR-NEXT:    s_setpc_b64 s[30:31]
@@ -1288,10 +1383,9 @@
 ; GFX906:       ; %bb.0: ; %entry
 ; GFX906-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX906-NEXT:    buffer_load_ushort v1, off, s[0:3], s32 offset:4094
-; GFX906-NEXT:    v_lshrrev_b32_e32 v0, 16, v0
+; GFX906-NEXT:    s_mov_b32 s4, 0x7060100
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
-; GFX906-NEXT:    v_and_b32_e32 v1, 0xffff, v1
-; GFX906-NEXT:    v_lshl_or_b32 v0, v0, 16, v1
+; GFX906-NEXT:    v_perm_b32 v0, v0, v1, s4
 ; GFX906-NEXT:    global_store_dword v[0:1], v0, off
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
 ; GFX906-NEXT:    s_setpc_b64 s[30:31]
@@ -1339,8 +1433,8 @@
 ; GFX906-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX906-NEXT:    buffer_load_ushort v0, off, s[0:3], 0 offset:4094 glc
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
-; GFX906-NEXT:    v_mov_b32_e32 v2, 0xffff
-; GFX906-NEXT:    v_bfi_b32 v0, v2, v0, v1
+; GFX906-NEXT:    s_mov_b32 s4, 0x7060100
+; GFX906-NEXT:    v_perm_b32 v0, v1, v0, s4
 ; GFX906-NEXT:    global_store_dword v[0:1], v0, off
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
 ; GFX906-NEXT:    s_setpc_b64 s[30:31]
@@ -1388,8 +1482,8 @@
 ; GFX906-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX906-NEXT:    buffer_load_ushort v0, off, s[0:3], 0 offset:4094 glc
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
-; GFX906-NEXT:    v_mov_b32_e32 v2, 0xffff
-; GFX906-NEXT:    v_bfi_b32 v0, v2, v0, v1
+; GFX906-NEXT:    s_mov_b32 s4, 0x7060100
+; GFX906-NEXT:    v_perm_b32 v0, v1, v0, s4
 ; GFX906-NEXT:    global_store_dword v[0:1], v0, off
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
 ; GFX906-NEXT:    s_setpc_b64 s[30:31]
@@ -1437,9 +1531,8 @@
 ; GFX906-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX906-NEXT:    buffer_load_ushort v0, off, s[0:3], 0 offset:4094 glc
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
-; GFX906-NEXT:    v_lshrrev_b32_e32 v1, 16, v1
-; GFX906-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX906-NEXT:    v_lshl_or_b32 v0, v1, 16, v0
+; GFX906-NEXT:    s_mov_b32 s4, 0x7060100
+; GFX906-NEXT:    v_perm_b32 v0, v1, v0, s4
 ; GFX906-NEXT:    global_store_dword v[0:1], v0, off
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
 ; GFX906-NEXT:    s_setpc_b64 s[30:31]
@@ -1486,9 +1579,9 @@
 ; GFX906:       ; %bb.0: ; %entry
 ; GFX906-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX906-NEXT:    buffer_load_ubyte v1, off, s[0:3], s32 offset:4095
-; GFX906-NEXT:    v_mov_b32_e32 v2, 0xffff
+; GFX906-NEXT:    s_mov_b32 s4, 0x7060100
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
-; GFX906-NEXT:    v_bfi_b32 v0, v2, v1, v0
+; GFX906-NEXT:    v_perm_b32 v0, v0, v1, s4
 ; GFX906-NEXT:    global_store_dword v[0:1], v0, off
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
 ; GFX906-NEXT:    s_setpc_b64 s[30:31]
@@ -1536,9 +1629,9 @@
 ; GFX906:       ; %bb.0: ; %entry
 ; GFX906-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX906-NEXT:    buffer_load_sbyte v1, off, s[0:3], s32 offset:4095
-; GFX906-NEXT:    v_mov_b32_e32 v2, 0xffff
+; GFX906-NEXT:    s_mov_b32 s4, 0x7060100
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
-; GFX906-NEXT:    v_bfi_b32 v0, v2, v1, v0
+; GFX906-NEXT:    v_perm_b32 v0, v0, v1, s4
 ; GFX906-NEXT:    global_store_dword v[0:1], v0, off
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
 ; GFX906-NEXT:    s_setpc_b64 s[30:31]
@@ -1587,8 +1680,8 @@
 ; GFX906-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX906-NEXT:    buffer_load_ubyte v0, off, s[0:3], 0 offset:4094 glc
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
-; GFX906-NEXT:    v_mov_b32_e32 v2, 0xffff
-; GFX906-NEXT:    v_bfi_b32 v0, v2, v0, v1
+; GFX906-NEXT:    s_mov_b32 s4, 0x7060100
+; GFX906-NEXT:    v_perm_b32 v0, v1, v0, s4
 ; GFX906-NEXT:    global_store_dword v[0:1], v0, off
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
 ; GFX906-NEXT:    s_setpc_b64 s[30:31]
@@ -1637,8 +1730,8 @@
 ; GFX906-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX906-NEXT:    buffer_load_sbyte v0, off, s[0:3], 0 offset:4094 glc
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
-; GFX906-NEXT:    v_mov_b32_e32 v2, 0xffff
-; GFX906-NEXT:    v_bfi_b32 v0, v2, v0, v1
+; GFX906-NEXT:    s_mov_b32 s4, 0x7060100
+; GFX906-NEXT:    v_perm_b32 v0, v1, v0, s4
 ; GFX906-NEXT:    global_store_dword v[0:1], v0, off
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
 ; GFX906-NEXT:    s_setpc_b64 s[30:31]
@@ -1687,9 +1780,8 @@
 ; GFX906-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX906-NEXT:    buffer_load_ubyte v0, off, s[0:3], 0 offset:4094 glc
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
-; GFX906-NEXT:    v_lshrrev_b32_e32 v1, 16, v1
-; GFX906-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX906-NEXT:    v_lshl_or_b32 v0, v1, 16, v0
+; GFX906-NEXT:    s_mov_b32 s4, 0x7060100
+; GFX906-NEXT:    v_perm_b32 v0, v1, v0, s4
 ; GFX906-NEXT:    global_store_dword v[0:1], v0, off
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
 ; GFX906-NEXT:    s_setpc_b64 s[30:31]
@@ -1738,9 +1830,9 @@
 ; GFX906:       ; %bb.0: ; %entry
 ; GFX906-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX906-NEXT:    global_load_ushort v0, v[0:1], off offset:-4094
-; GFX906-NEXT:    v_mov_b32_e32 v1, 0xffff
+; GFX906-NEXT:    s_mov_b32 s4, 0x7060100
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
-; GFX906-NEXT:    v_bfi_b32 v0, v1, v0, v2
+; GFX906-NEXT:    v_perm_b32 v0, v2, v0, s4
 ; GFX906-NEXT:    global_store_dword v[0:1], v0, off
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
 ; GFX906-NEXT:    s_setpc_b64 s[30:31]
@@ -1780,10 +1872,9 @@
 ; GFX906:       ; %bb.0: ; %entry
 ; GFX906-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX906-NEXT:    global_load_ushort v0, v[0:1], off offset:-4094
-; GFX906-NEXT:    v_lshrrev_b32_e32 v1, 16, v2
+; GFX906-NEXT:    s_mov_b32 s4, 0x7060100
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
-; GFX906-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX906-NEXT:    v_lshl_or_b32 v0, v1, 16, v0
+; GFX906-NEXT:    v_perm_b32 v0, v2, v0, s4
 ; GFX906-NEXT:    global_store_dword v[0:1], v0, off
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
 ; GFX906-NEXT:    s_setpc_b64 s[30:31]
@@ -1823,10 +1914,9 @@
 ; GFX906:       ; %bb.0: ; %entry
 ; GFX906-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX906-NEXT:    global_load_ubyte v0, v[0:1], off offset:-4095
-; GFX906-NEXT:    v_lshrrev_b32_e32 v1, 16, v2
+; GFX906-NEXT:    s_mov_b32 s4, 0x7060100
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
-; GFX906-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX906-NEXT:    v_lshl_or_b32 v0, v1, 16, v0
+; GFX906-NEXT:    v_perm_b32 v0, v2, v0, s4
 ; GFX906-NEXT:    global_store_dword v[0:1], v0, off
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
 ; GFX906-NEXT:    s_setpc_b64 s[30:31]
@@ -1868,10 +1958,9 @@
 ; GFX906:       ; %bb.0: ; %entry
 ; GFX906-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX906-NEXT:    global_load_sbyte v0, v[0:1], off offset:-4095
-; GFX906-NEXT:    v_lshrrev_b32_e32 v1, 16, v2
+; GFX906-NEXT:    s_mov_b32 s4, 0x7060100
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
-; GFX906-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX906-NEXT:    v_lshl_or_b32 v0, v1, 16, v0
+; GFX906-NEXT:    v_perm_b32 v0, v2, v0, s4
 ; GFX906-NEXT:    global_store_dword v[0:1], v0, off
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
 ; GFX906-NEXT:    s_setpc_b64 s[30:31]
@@ -1919,11 +2008,11 @@
 ; GFX906-NEXT:    v_mov_b32_e32 v1, 0x7b
 ; GFX906-NEXT:    buffer_store_dword v1, off, s[0:3], s32 offset:4
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
-; GFX906-NEXT:    v_mov_b32_e32 v3, 44
-; GFX906-NEXT:    buffer_load_ushort v1, v3, s[0:3], s32 offen offset:4054 glc
+; GFX906-NEXT:    v_mov_b32_e32 v2, 44
+; GFX906-NEXT:    buffer_load_ushort v1, v2, s[0:3], s32 offen offset:4054 glc
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
-; GFX906-NEXT:    v_mov_b32_e32 v2, 0xffff
-; GFX906-NEXT:    v_bfi_b32 v0, v2, v1, v0
+; GFX906-NEXT:    s_mov_b32 s4, 0x7060100
+; GFX906-NEXT:    v_perm_b32 v0, v0, v1, s4
 ; GFX906-NEXT:    global_store_dword v[0:1], v0, off
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
 ; GFX906-NEXT:    s_setpc_b64 s[30:31]
@@ -1988,11 +2077,11 @@
 ; GFX906-NEXT:    v_mov_b32_e32 v1, 0x7b
 ; GFX906-NEXT:    buffer_store_dword v1, off, s[0:3], s32 offset:4
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
-; GFX906-NEXT:    v_mov_b32_e32 v3, 44
-; GFX906-NEXT:    buffer_load_sbyte v1, v3, s[0:3], s32 offen offset:4055 glc
+; GFX906-NEXT:    v_mov_b32_e32 v2, 44
+; GFX906-NEXT:    buffer_load_sbyte v1, v2, s[0:3], s32 offen offset:4055 glc
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
-; GFX906-NEXT:    v_mov_b32_e32 v2, 0xffff
-; GFX906-NEXT:    v_bfi_b32 v0, v2, v1, v0
+; GFX906-NEXT:    s_mov_b32 s4, 0x7060100
+; GFX906-NEXT:    v_perm_b32 v0, v0, v1, s4
 ; GFX906-NEXT:    global_store_dword v[0:1], v0, off
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
 ; GFX906-NEXT:    s_setpc_b64 s[30:31]
@@ -2058,11 +2147,11 @@
 ; GFX906-NEXT:    v_mov_b32_e32 v1, 0x7b
 ; GFX906-NEXT:    buffer_store_dword v1, off, s[0:3], s32 offset:4
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
-; GFX906-NEXT:    v_mov_b32_e32 v3, 44
-; GFX906-NEXT:    buffer_load_ubyte v1, v3, s[0:3], s32 offen offset:4055 glc
+; GFX906-NEXT:    v_mov_b32_e32 v2, 44
+; GFX906-NEXT:    buffer_load_ubyte v1, v2, s[0:3], s32 offen offset:4055 glc
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
-; GFX906-NEXT:    v_mov_b32_e32 v2, 0xffff
-; GFX906-NEXT:    v_bfi_b32 v0, v2, v1, v0
+; GFX906-NEXT:    s_mov_b32 s4, 0x7060100
+; GFX906-NEXT:    v_perm_b32 v0, v0, v1, s4
 ; GFX906-NEXT:    global_store_dword v[0:1], v0, off
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
 ; GFX906-NEXT:    s_setpc_b64 s[30:31]
@@ -2131,9 +2220,8 @@
 ; GFX906-NEXT:    v_mov_b32_e32 v2, 44
 ; GFX906-NEXT:    buffer_load_sbyte v1, v2, s[0:3], s32 offen offset:4055 glc
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
-; GFX906-NEXT:    v_lshrrev_b32_e32 v0, 16, v0
-; GFX906-NEXT:    v_and_b32_e32 v1, 0xffff, v1
-; GFX906-NEXT:    v_lshl_or_b32 v0, v0, 16, v1
+; GFX906-NEXT:    s_mov_b32 s4, 0x7060100
+; GFX906-NEXT:    v_perm_b32 v0, v0, v1, s4
 ; GFX906-NEXT:    global_store_dword v[0:1], v0, off
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
 ; GFX906-NEXT:    s_setpc_b64 s[30:31]
@@ -2203,9 +2291,8 @@
 ; GFX906-NEXT:    v_mov_b32_e32 v2, 44
 ; GFX906-NEXT:    buffer_load_ubyte v1, v2, s[0:3], s32 offen offset:4055 glc
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
-; GFX906-NEXT:    v_lshrrev_b32_e32 v0, 16, v0
-; GFX906-NEXT:    v_and_b32_e32 v1, 0xffff, v1
-; GFX906-NEXT:    v_lshl_or_b32 v0, v0, 16, v1
+; GFX906-NEXT:    s_mov_b32 s4, 0x7060100
+; GFX906-NEXT:    v_perm_b32 v0, v0, v1, s4
 ; GFX906-NEXT:    global_store_dword v[0:1], v0, off
 ; GFX906-NEXT:    s_waitcnt vmcnt(0)
 ; GFX906-NEXT:    s_setpc_b64 s[30:31]
diff --git a/llvm/test/CodeGen/AMDGPU/pack.v2f16.ll b/llvm/test/CodeGen/AMDGPU/pack.v2f16.ll
--- a/llvm/test/CodeGen/AMDGPU/pack.v2f16.ll
+++ b/llvm/test/CodeGen/AMDGPU/pack.v2f16.ll
@@ -169,8 +169,8 @@
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    global_load_dword v2, v0, s[2:3] glc
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
-; GFX9-NEXT:    v_and_b32_e32 v0, 0xffff, v1
-; GFX9-NEXT:    v_lshl_or_b32 v0, v2, 16, v0
+; GFX9-NEXT:    s_mov_b32 s0, 0x5040100
+; GFX9-NEXT:    v_perm_b32 v0, v2, v1, s0
 ; GFX9-NEXT:    ;;#ASMSTART
 ; GFX9-NEXT:    ; use v0
 ; GFX9-NEXT:    ;;#ASMEND
@@ -247,10 +247,10 @@
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    global_load_dword v2, v0, s[2:3] glc
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    s_mov_b32 s0, 0x5040100
 ; GFX9-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX9-NEXT:    s_mov_b32 s2, -1
-; GFX9-NEXT:    v_and_b32_e32 v0, 0xffff, v1
-; GFX9-NEXT:    v_lshl_or_b32 v0, v2, 16, v0
+; GFX9-NEXT:    v_perm_b32 v0, v2, v1, s0
 ; GFX9-NEXT:    v_add_u32_e32 v0, 9, v0
 ; GFX9-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
@@ -326,11 +326,12 @@
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX9-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
-; GFX9-NEXT:    v_mov_b32_e32 v1, 0x1234
+; GFX9-NEXT:    v_mov_b32_e32 v1, 0x5040100
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    global_load_dword v0, v0, s[0:1] glc
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
-; GFX9-NEXT:    v_lshl_or_b32 v0, v0, 16, v1
+; GFX9-NEXT:    s_movk_i32 s0, 0x1234
+; GFX9-NEXT:    v_perm_b32 v0, v0, s0, v1
 ; GFX9-NEXT:    ;;#ASMSTART
 ; GFX9-NEXT:    ; use v0
 ; GFX9-NEXT:    ;;#ASMEND
@@ -387,11 +388,12 @@
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX9-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
-; GFX9-NEXT:    v_mov_b32_e32 v1, 0x4400
+; GFX9-NEXT:    v_mov_b32_e32 v1, 0x5040100
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    global_load_dword v0, v0, s[0:1] glc
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
-; GFX9-NEXT:    v_lshl_or_b32 v0, v0, 16, v1
+; GFX9-NEXT:    s_movk_i32 s0, 0x4400
+; GFX9-NEXT:    v_perm_b32 v0, v0, s0, v1
 ; GFX9-NEXT:    ;;#ASMSTART
 ; GFX9-NEXT:    ; use v0
 ; GFX9-NEXT:    ;;#ASMEND
@@ -448,12 +450,12 @@
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX9-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX9-NEXT:    v_mov_b32_e32 v1, 0x5040100
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    global_load_dword v0, v0, s[0:1] glc
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    s_movk_i32 s0, 0x1234
-; GFX9-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX9-NEXT:    v_lshl_or_b32 v0, s0, 16, v0
+; GFX9-NEXT:    v_perm_b32 v0, s0, v0, v1
 ; GFX9-NEXT:    ;;#ASMSTART
 ; GFX9-NEXT:    ; use v0
 ; GFX9-NEXT:    ;;#ASMEND
@@ -510,12 +512,12 @@
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX9-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX9-NEXT:    v_mov_b32_e32 v1, 0x5040100
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    global_load_dword v0, v0, s[0:1] glc
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    s_movk_i32 s0, 0x3c00
-; GFX9-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX9-NEXT:    v_lshl_or_b32 v0, s0, 16, v0
+; GFX9-NEXT:    v_perm_b32 v0, s0, v0, v1
 ; GFX9-NEXT:    ;;#ASMSTART
 ; GFX9-NEXT:    ; use v0
 ; GFX9-NEXT:    ;;#ASMEND
@@ -572,11 +574,11 @@
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX9-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX9-NEXT:    v_mov_b32_e32 v1, 0x5040100
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    global_load_dword v0, v0, s[0:1] glc
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
-; GFX9-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX9-NEXT:    v_lshl_or_b32 v0, 64, 16, v0
+; GFX9-NEXT:    v_perm_b32 v0, 64, v0, v1
 ; GFX9-NEXT:    ;;#ASMSTART
 ; GFX9-NEXT:    ; use v0
 ; GFX9-NEXT:    ;;#ASMEND
diff --git a/llvm/test/CodeGen/AMDGPU/pack.v2i16.ll b/llvm/test/CodeGen/AMDGPU/pack.v2i16.ll
--- a/llvm/test/CodeGen/AMDGPU/pack.v2i16.ll
+++ b/llvm/test/CodeGen/AMDGPU/pack.v2i16.ll
@@ -165,8 +165,8 @@
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    global_load_dword v2, v0, s[2:3] glc
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
-; GFX9-NEXT:    v_and_b32_e32 v0, 0xffff, v1
-; GFX9-NEXT:    v_lshl_or_b32 v0, v2, 16, v0
+; GFX9-NEXT:    s_mov_b32 s0, 0x5040100
+; GFX9-NEXT:    v_perm_b32 v0, v2, v1, s0
 ; GFX9-NEXT:    ;;#ASMSTART
 ; GFX9-NEXT:    ; use v0
 ; GFX9-NEXT:    ;;#ASMEND
@@ -241,10 +241,10 @@
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    global_load_dword v2, v0, s[2:3] glc
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    s_mov_b32 s0, 0x5040100
 ; GFX9-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX9-NEXT:    s_mov_b32 s2, -1
-; GFX9-NEXT:    v_and_b32_e32 v0, 0xffff, v1
-; GFX9-NEXT:    v_lshl_or_b32 v0, v2, 16, v0
+; GFX9-NEXT:    v_perm_b32 v0, v2, v1, s0
 ; GFX9-NEXT:    v_add_u32_e32 v0, 9, v0
 ; GFX9-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
@@ -318,11 +318,12 @@
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX9-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
-; GFX9-NEXT:    v_mov_b32_e32 v1, 0x7b
+; GFX9-NEXT:    v_mov_b32_e32 v1, 0x5040100
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    global_load_dword v0, v0, s[0:1] glc
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
-; GFX9-NEXT:    v_lshl_or_b32 v0, v0, 16, v1
+; GFX9-NEXT:    s_movk_i32 s0, 0x7b
+; GFX9-NEXT:    v_perm_b32 v0, v0, s0, v1
 ; GFX9-NEXT:    ;;#ASMSTART
 ; GFX9-NEXT:    ; use v0
 ; GFX9-NEXT:    ;;#ASMEND
@@ -378,10 +379,11 @@
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX9-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX9-NEXT:    v_mov_b32_e32 v1, 0x5040100
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    global_load_dword v0, v0, s[0:1] glc
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
-; GFX9-NEXT:    v_lshl_or_b32 v0, v0, 16, 64
+; GFX9-NEXT:    v_perm_b32 v0, v0, 64, v1
 ; GFX9-NEXT:    ;;#ASMSTART
 ; GFX9-NEXT:    ; use v0
 ; GFX9-NEXT:    ;;#ASMEND
@@ -437,12 +439,12 @@
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX9-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX9-NEXT:    v_mov_b32_e32 v1, 0x5040100
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    global_load_dword v0, v0, s[0:1] glc
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    s_movk_i32 s0, 0x7b
-; GFX9-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX9-NEXT:    v_lshl_or_b32 v0, s0, 16, v0
+; GFX9-NEXT:    v_perm_b32 v0, s0, v0, v1
 ; GFX9-NEXT:    ;;#ASMSTART
 ; GFX9-NEXT:    ; use v0
 ; GFX9-NEXT:    ;;#ASMEND
@@ -498,11 +500,11 @@
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX9-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX9-NEXT:    v_mov_b32_e32 v1, 0x5040100
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    global_load_dword v0, v0, s[0:1] glc
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
-; GFX9-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX9-NEXT:    v_lshl_or_b32 v0, 7, 16, v0
+; GFX9-NEXT:    v_perm_b32 v0, 7, v0, v1
 ; GFX9-NEXT:    ;;#ASMSTART
 ; GFX9-NEXT:    ; use v0
 ; GFX9-NEXT:    ;;#ASMEND
@@ -557,3 +559,4 @@
 
 attributes #0 = { nounwind }
 attributes #1 = { nounwind readnone }
+
diff --git a/llvm/test/CodeGen/AMDGPU/partial-shift-shrink.ll b/llvm/test/CodeGen/AMDGPU/partial-shift-shrink.ll
--- a/llvm/test/CodeGen/AMDGPU/partial-shift-shrink.ll
+++ b/llvm/test/CodeGen/AMDGPU/partial-shift-shrink.ll
@@ -86,9 +86,8 @@
 ; GCN-LABEL: trunc_srl_v2i64_16_to_v2i16:
 ; GCN:       ; %bb.0:
 ; GCN-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GCN-NEXT:    v_lshrrev_b32_e32 v0, 16, v0
-; GCN-NEXT:    v_mov_b32_e32 v1, 0xffff0000
-; GCN-NEXT:    v_and_or_b32 v0, v2, v1, v0
+; GCN-NEXT:    s_mov_b32 s4, 0x7060302
+; GCN-NEXT:    v_perm_b32 v0, v2, v0, s4
 ; GCN-NEXT:    s_setpc_b64 s[30:31]
   %shift = lshr <2 x i64> %x, <i64 16, i64 16>
   %trunc = trunc <2 x i64> %shift to <2 x i16>
diff --git a/llvm/test/CodeGen/AMDGPU/strict_fadd.f16.ll b/llvm/test/CodeGen/AMDGPU/strict_fadd.f16.ll
--- a/llvm/test/CodeGen/AMDGPU/strict_fadd.f16.ll
+++ b/llvm/test/CodeGen/AMDGPU/strict_fadd.f16.ll
@@ -168,8 +168,9 @@
 ; GFX9-NEXT:    v_add_f16_sdwa v5, v0, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
 ; GFX9-NEXT:    v_add_f16_e32 v1, v1, v3
 ; GFX9-NEXT:    v_add_f16_e32 v0, v0, v2
-; GFX9-NEXT:    v_lshl_or_b32 v0, v5, 16, v0
-; GFX9-NEXT:    v_lshl_or_b32 v1, v4, 16, v1
+; GFX9-NEXT:    s_mov_b32 s4, 0x5040100
+; GFX9-NEXT:    v_perm_b32 v0, v5, v0, s4
+; GFX9-NEXT:    v_perm_b32 v1, v4, v1, s4
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX8-LABEL: v_constained_fadd_v4f16_fpexcept_strict:
@@ -187,14 +188,12 @@
 ; GFX10:       ; %bb.0:
 ; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
-; GFX10-NEXT:    v_add_f16_e32 v4, v0, v2
-; GFX10-NEXT:    v_add_f16_e32 v5, v1, v3
-; GFX10-NEXT:    v_add_f16_sdwa v1, v1, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
-; GFX10-NEXT:    v_add_f16_sdwa v0, v0, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
-; GFX10-NEXT:    v_and_b32_e32 v2, 0xffff, v4
-; GFX10-NEXT:    v_and_b32_e32 v3, 0xffff, v5
-; GFX10-NEXT:    v_lshl_or_b32 v0, v0, 16, v2
-; GFX10-NEXT:    v_lshl_or_b32 v1, v1, 16, v3
+; GFX10-NEXT:    v_add_f16_sdwa v4, v1, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
+; GFX10-NEXT:    v_add_f16_sdwa v5, v0, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
+; GFX10-NEXT:    v_add_f16_e32 v0, v0, v2
+; GFX10-NEXT:    v_add_f16_e32 v1, v1, v3
+; GFX10-NEXT:    v_perm_b32 v0, v5, v0, 0x5040100
+; GFX10-NEXT:    v_perm_b32 v1, v4, v1, 0x5040100
 ; GFX10-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX11-LABEL: v_constained_fadd_v4f16_fpexcept_strict:
@@ -202,17 +201,15 @@
 ; GFX11-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX11-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-NEXT:    v_lshrrev_b32_e32 v4, 16, v3
-; GFX11-NEXT:    v_lshrrev_b32_e32 v5, 16, v1
-; GFX11-NEXT:    v_lshrrev_b32_e32 v6, 16, v2
-; GFX11-NEXT:    v_lshrrev_b32_e32 v7, 16, v0
-; GFX11-NEXT:    v_add_f16_e32 v0, v0, v2
+; GFX11-NEXT:    v_lshrrev_b32_e32 v5, 16, v2
+; GFX11-NEXT:    v_lshrrev_b32_e32 v6, 16, v0
+; GFX11-NEXT:    v_lshrrev_b32_e32 v7, 16, v1
 ; GFX11-NEXT:    v_add_f16_e32 v1, v1, v3
-; GFX11-NEXT:    v_add_f16_e32 v2, v5, v4
-; GFX11-NEXT:    v_add_f16_e32 v3, v7, v6
-; GFX11-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX11-NEXT:    v_and_b32_e32 v1, 0xffff, v1
-; GFX11-NEXT:    v_lshl_or_b32 v0, v3, 16, v0
-; GFX11-NEXT:    v_lshl_or_b32 v1, v2, 16, v1
+; GFX11-NEXT:    v_add_f16_e32 v0, v0, v2
+; GFX11-NEXT:    v_add_f16_e32 v2, v6, v5
+; GFX11-NEXT:    v_add_f16_e32 v3, v7, v4
+; GFX11-NEXT:    v_perm_b32 v0, v2, v0, 0x5040100
+; GFX11-NEXT:    v_perm_b32 v1, v3, v1, 0x5040100
 ; GFX11-NEXT:    s_setpc_b64 s[30:31]
   %val = call <4 x half> @llvm.experimental.constrained.fadd.v4f16(<4 x half> %x, <4 x half> %y, metadata !"round.tonearest", metadata !"fpexcept.strict")
   ret <4 x half> %val
diff --git a/llvm/test/CodeGen/AMDGPU/strict_fma.f16.ll b/llvm/test/CodeGen/AMDGPU/strict_fma.f16.ll
--- a/llvm/test/CodeGen/AMDGPU/strict_fma.f16.ll
+++ b/llvm/test/CodeGen/AMDGPU/strict_fma.f16.ll
@@ -115,13 +115,12 @@
 ; GFX9-NEXT:    v_lshrrev_b32_e32 v7, 16, v4
 ; GFX9-NEXT:    v_lshrrev_b32_e32 v8, 16, v2
 ; GFX9-NEXT:    v_lshrrev_b32_e32 v9, 16, v0
+; GFX9-NEXT:    v_fma_f16 v7, v9, v8, v7
 ; GFX9-NEXT:    v_fma_f16 v1, v1, v3, v5
 ; GFX9-NEXT:    v_fma_f16 v0, v0, v2, v4
-; GFX9-NEXT:    v_fma_f16 v7, v9, v8, v7
-; GFX9-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX9-NEXT:    v_and_b32_e32 v1, 0xffff, v1
-; GFX9-NEXT:    v_lshl_or_b32 v0, v7, 16, v0
-; GFX9-NEXT:    v_lshl_or_b32 v1, v6, 16, v1
+; GFX9-NEXT:    s_mov_b32 s4, 0x5040100
+; GFX9-NEXT:    v_perm_b32 v0, v7, v0, s4
+; GFX9-NEXT:    v_perm_b32 v1, v6, v1, s4
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX8-LABEL: v_constained_fma_v4f16_fpexcept_strict:
@@ -154,13 +153,11 @@
 ; GFX10-NEXT:    v_lshrrev_b32_e32 v10, 16, v2
 ; GFX10-NEXT:    v_lshrrev_b32_e32 v11, 16, v0
 ; GFX10-NEXT:    v_fmac_f16_e32 v4, v0, v2
-; GFX10-NEXT:    v_fmac_f16_e32 v5, v1, v3
 ; GFX10-NEXT:    v_fmac_f16_e32 v6, v8, v7
+; GFX10-NEXT:    v_fmac_f16_e32 v5, v1, v3
 ; GFX10-NEXT:    v_fmac_f16_e32 v9, v11, v10
-; GFX10-NEXT:    v_and_b32_e32 v0, 0xffff, v4
-; GFX10-NEXT:    v_and_b32_e32 v1, 0xffff, v5
-; GFX10-NEXT:    v_lshl_or_b32 v0, v9, 16, v0
-; GFX10-NEXT:    v_lshl_or_b32 v1, v6, 16, v1
+; GFX10-NEXT:    v_perm_b32 v1, v6, v5, 0x5040100
+; GFX10-NEXT:    v_perm_b32 v0, v9, v4, 0x5040100
 ; GFX10-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX11-LABEL: v_constained_fma_v4f16_fpexcept_strict:
@@ -174,13 +171,11 @@
 ; GFX11-NEXT:    v_lshrrev_b32_e32 v10, 16, v2
 ; GFX11-NEXT:    v_lshrrev_b32_e32 v11, 16, v0
 ; GFX11-NEXT:    v_fmac_f16_e32 v4, v0, v2
-; GFX11-NEXT:    v_fmac_f16_e32 v5, v1, v3
 ; GFX11-NEXT:    v_fmac_f16_e32 v6, v8, v7
+; GFX11-NEXT:    v_fmac_f16_e32 v5, v1, v3
 ; GFX11-NEXT:    v_fmac_f16_e32 v9, v11, v10
-; GFX11-NEXT:    v_and_b32_e32 v0, 0xffff, v4
-; GFX11-NEXT:    v_and_b32_e32 v1, 0xffff, v5
-; GFX11-NEXT:    v_lshl_or_b32 v0, v9, 16, v0
-; GFX11-NEXT:    v_lshl_or_b32 v1, v6, 16, v1
+; GFX11-NEXT:    v_perm_b32 v1, v6, v5, 0x5040100
+; GFX11-NEXT:    v_perm_b32 v0, v9, v4, 0x5040100
 ; GFX11-NEXT:    s_setpc_b64 s[30:31]
   %val = call <4 x half> @llvm.experimental.constrained.fma.v4f16(<4 x half> %x, <4 x half> %y, <4 x half> %z, metadata !"round.tonearest", metadata !"fpexcept.strict")
   ret <4 x half> %val
diff --git a/llvm/test/CodeGen/AMDGPU/strict_fmul.f16.ll b/llvm/test/CodeGen/AMDGPU/strict_fmul.f16.ll
--- a/llvm/test/CodeGen/AMDGPU/strict_fmul.f16.ll
+++ b/llvm/test/CodeGen/AMDGPU/strict_fmul.f16.ll
@@ -168,8 +168,9 @@
 ; GFX9-NEXT:    v_mul_f16_sdwa v5, v0, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
 ; GFX9-NEXT:    v_mul_f16_e32 v1, v1, v3
 ; GFX9-NEXT:    v_mul_f16_e32 v0, v0, v2
-; GFX9-NEXT:    v_lshl_or_b32 v0, v5, 16, v0
-; GFX9-NEXT:    v_lshl_or_b32 v1, v4, 16, v1
+; GFX9-NEXT:    s_mov_b32 s4, 0x5040100
+; GFX9-NEXT:    v_perm_b32 v0, v5, v0, s4
+; GFX9-NEXT:    v_perm_b32 v1, v4, v1, s4
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX8-LABEL: v_constained_fmul_v4f16_fpexcept_strict:
@@ -187,14 +188,12 @@
 ; GFX10:       ; %bb.0:
 ; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
-; GFX10-NEXT:    v_mul_f16_e32 v4, v0, v2
-; GFX10-NEXT:    v_mul_f16_e32 v5, v1, v3
-; GFX10-NEXT:    v_mul_f16_sdwa v1, v1, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
-; GFX10-NEXT:    v_mul_f16_sdwa v0, v0, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
-; GFX10-NEXT:    v_and_b32_e32 v2, 0xffff, v4
-; GFX10-NEXT:    v_and_b32_e32 v3, 0xffff, v5
-; GFX10-NEXT:    v_lshl_or_b32 v0, v0, 16, v2
-; GFX10-NEXT:    v_lshl_or_b32 v1, v1, 16, v3
+; GFX10-NEXT:    v_mul_f16_sdwa v4, v1, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
+; GFX10-NEXT:    v_mul_f16_sdwa v5, v0, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
+; GFX10-NEXT:    v_mul_f16_e32 v0, v0, v2
+; GFX10-NEXT:    v_mul_f16_e32 v1, v1, v3
+; GFX10-NEXT:    v_perm_b32 v0, v5, v0, 0x5040100
+; GFX10-NEXT:    v_perm_b32 v1, v4, v1, 0x5040100
 ; GFX10-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX11-LABEL: v_constained_fmul_v4f16_fpexcept_strict:
@@ -202,17 +201,15 @@
 ; GFX11-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX11-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-NEXT:    v_lshrrev_b32_e32 v4, 16, v3
-; GFX11-NEXT:    v_lshrrev_b32_e32 v5, 16, v1
-; GFX11-NEXT:    v_lshrrev_b32_e32 v6, 16, v2
-; GFX11-NEXT:    v_lshrrev_b32_e32 v7, 16, v0
-; GFX11-NEXT:    v_mul_f16_e32 v0, v0, v2
+; GFX11-NEXT:    v_lshrrev_b32_e32 v5, 16, v2
+; GFX11-NEXT:    v_lshrrev_b32_e32 v6, 16, v0
+; GFX11-NEXT:    v_lshrrev_b32_e32 v7, 16, v1
 ; GFX11-NEXT:    v_mul_f16_e32 v1, v1, v3
-; GFX11-NEXT:    v_mul_f16_e32 v2, v5, v4
-; GFX11-NEXT:    v_mul_f16_e32 v3, v7, v6
-; GFX11-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX11-NEXT:    v_and_b32_e32 v1, 0xffff, v1
-; GFX11-NEXT:    v_lshl_or_b32 v0, v3, 16, v0
-; GFX11-NEXT:    v_lshl_or_b32 v1, v2, 16, v1
+; GFX11-NEXT:    v_mul_f16_e32 v0, v0, v2
+; GFX11-NEXT:    v_mul_f16_e32 v2, v6, v5
+; GFX11-NEXT:    v_mul_f16_e32 v3, v7, v4
+; GFX11-NEXT:    v_perm_b32 v0, v2, v0, 0x5040100
+; GFX11-NEXT:    v_perm_b32 v1, v3, v1, 0x5040100
 ; GFX11-NEXT:    s_setpc_b64 s[30:31]
   %val = call <4 x half> @llvm.experimental.constrained.fmul.v4f16(<4 x half> %x, <4 x half> %y, metadata !"round.tonearest", metadata !"fpexcept.strict")
   ret <4 x half> %val
diff --git a/llvm/test/CodeGen/AMDGPU/strict_fsub.f16.ll b/llvm/test/CodeGen/AMDGPU/strict_fsub.f16.ll
--- a/llvm/test/CodeGen/AMDGPU/strict_fsub.f16.ll
+++ b/llvm/test/CodeGen/AMDGPU/strict_fsub.f16.ll
@@ -62,7 +62,8 @@
 ; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX9-NEXT:    v_sub_f16_sdwa v2, v0, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
 ; GFX9-NEXT:    v_sub_f16_e32 v0, v0, v1
-; GFX9-NEXT:    v_lshl_or_b32 v0, v2, 16, v0
+; GFX9-NEXT:    s_mov_b32 s4, 0x5040100
+; GFX9-NEXT:    v_perm_b32 v0, v2, v0, s4
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX8-LABEL: v_constained_fsub_v2f16_fpexcept_strict:
@@ -77,10 +78,9 @@
 ; GFX10:       ; %bb.0:
 ; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
-; GFX10-NEXT:    v_sub_f16_e32 v2, v0, v1
-; GFX10-NEXT:    v_sub_f16_sdwa v0, v0, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
-; GFX10-NEXT:    v_and_b32_e32 v1, 0xffff, v2
-; GFX10-NEXT:    v_lshl_or_b32 v0, v0, 16, v1
+; GFX10-NEXT:    v_sub_f16_sdwa v2, v0, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
+; GFX10-NEXT:    v_sub_f16_e32 v0, v0, v1
+; GFX10-NEXT:    v_perm_b32 v0, v2, v0, 0x5040100
 ; GFX10-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX11-LABEL: v_constained_fsub_v2f16_fpexcept_strict:
@@ -90,9 +90,8 @@
 ; GFX11-NEXT:    v_lshrrev_b32_e32 v2, 16, v1
 ; GFX11-NEXT:    v_lshrrev_b32_e32 v3, 16, v0
 ; GFX11-NEXT:    v_sub_f16_e32 v0, v0, v1
-; GFX11-NEXT:    v_sub_f16_e32 v1, v3, v2
-; GFX11-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX11-NEXT:    v_lshl_or_b32 v0, v1, 16, v0
+; GFX11-NEXT:    v_sub_f16_e32 v2, v3, v2
+; GFX11-NEXT:    v_perm_b32 v0, v2, v0, 0x5040100
 ; GFX11-NEXT:    s_setpc_b64 s[30:31]
   %val = call <2 x half> @llvm.experimental.constrained.fsub.v2f16(<2 x half> %x, <2 x half> %y, metadata !"round.tonearest", metadata !"fpexcept.strict")
   ret <2 x half> %val
@@ -104,7 +103,8 @@
 ; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX9-NEXT:    v_sub_f16_sdwa v2, v0, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
 ; GFX9-NEXT:    v_sub_f16_e32 v0, v0, v1
-; GFX9-NEXT:    v_lshl_or_b32 v0, v2, 16, v0
+; GFX9-NEXT:    s_mov_b32 s4, 0x5040100
+; GFX9-NEXT:    v_perm_b32 v0, v2, v0, s4
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX8-LABEL: v_constained_fsub_v2f16_fpexcept_ignore:
@@ -119,10 +119,9 @@
 ; GFX10:       ; %bb.0:
 ; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
-; GFX10-NEXT:    v_sub_f16_e32 v2, v0, v1
-; GFX10-NEXT:    v_sub_f16_sdwa v0, v0, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
-; GFX10-NEXT:    v_and_b32_e32 v1, 0xffff, v2
-; GFX10-NEXT:    v_lshl_or_b32 v0, v0, 16, v1
+; GFX10-NEXT:    v_sub_f16_sdwa v2, v0, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
+; GFX10-NEXT:    v_sub_f16_e32 v0, v0, v1
+; GFX10-NEXT:    v_perm_b32 v0, v2, v0, 0x5040100
 ; GFX10-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX11-LABEL: v_constained_fsub_v2f16_fpexcept_ignore:
@@ -132,9 +131,8 @@
 ; GFX11-NEXT:    v_lshrrev_b32_e32 v2, 16, v1
 ; GFX11-NEXT:    v_lshrrev_b32_e32 v3, 16, v0
 ; GFX11-NEXT:    v_sub_f16_e32 v0, v0, v1
-; GFX11-NEXT:    v_sub_f16_e32 v1, v3, v2
-; GFX11-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX11-NEXT:    v_lshl_or_b32 v0, v1, 16, v0
+; GFX11-NEXT:    v_sub_f16_e32 v2, v3, v2
+; GFX11-NEXT:    v_perm_b32 v0, v2, v0, 0x5040100
 ; GFX11-NEXT:    s_setpc_b64 s[30:31]
   %val = call <2 x half> @llvm.experimental.constrained.fsub.v2f16(<2 x half> %x, <2 x half> %y, metadata !"round.tonearest", metadata !"fpexcept.ignore")
   ret <2 x half> %val
@@ -146,7 +144,8 @@
 ; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX9-NEXT:    v_sub_f16_sdwa v2, v0, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
 ; GFX9-NEXT:    v_sub_f16_e32 v0, v0, v1
-; GFX9-NEXT:    v_lshl_or_b32 v0, v2, 16, v0
+; GFX9-NEXT:    s_mov_b32 s4, 0x5040100
+; GFX9-NEXT:    v_perm_b32 v0, v2, v0, s4
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX8-LABEL: v_constained_fsub_v2f16_fpexcept_maytrap:
@@ -161,10 +160,9 @@
 ; GFX10:       ; %bb.0:
 ; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
-; GFX10-NEXT:    v_sub_f16_e32 v2, v0, v1
-; GFX10-NEXT:    v_sub_f16_sdwa v0, v0, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
-; GFX10-NEXT:    v_and_b32_e32 v1, 0xffff, v2
-; GFX10-NEXT:    v_lshl_or_b32 v0, v0, 16, v1
+; GFX10-NEXT:    v_sub_f16_sdwa v2, v0, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
+; GFX10-NEXT:    v_sub_f16_e32 v0, v0, v1
+; GFX10-NEXT:    v_perm_b32 v0, v2, v0, 0x5040100
 ; GFX10-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX11-LABEL: v_constained_fsub_v2f16_fpexcept_maytrap:
@@ -174,9 +172,8 @@
 ; GFX11-NEXT:    v_lshrrev_b32_e32 v2, 16, v1
 ; GFX11-NEXT:    v_lshrrev_b32_e32 v3, 16, v0
 ; GFX11-NEXT:    v_sub_f16_e32 v0, v0, v1
-; GFX11-NEXT:    v_sub_f16_e32 v1, v3, v2
-; GFX11-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX11-NEXT:    v_lshl_or_b32 v0, v1, 16, v0
+; GFX11-NEXT:    v_sub_f16_e32 v2, v3, v2
+; GFX11-NEXT:    v_perm_b32 v0, v2, v0, 0x5040100
 ; GFX11-NEXT:    s_setpc_b64 s[30:31]
   %val = call <2 x half> @llvm.experimental.constrained.fsub.v2f16(<2 x half> %x, <2 x half> %y, metadata !"round.tonearest", metadata !"fpexcept.maytrap")
   ret <2 x half> %val
@@ -188,7 +185,8 @@
 ; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX9-NEXT:    v_sub_f16_sdwa v4, v0, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
 ; GFX9-NEXT:    v_sub_f16_e32 v0, v0, v2
-; GFX9-NEXT:    v_lshl_or_b32 v0, v4, 16, v0
+; GFX9-NEXT:    s_mov_b32 s4, 0x5040100
+; GFX9-NEXT:    v_perm_b32 v0, v4, v0, s4
 ; GFX9-NEXT:    v_sub_f16_e32 v1, v1, v3
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
 ;
@@ -205,11 +203,10 @@
 ; GFX10:       ; %bb.0:
 ; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
-; GFX10-NEXT:    v_sub_f16_e32 v4, v0, v2
-; GFX10-NEXT:    v_sub_f16_sdwa v0, v0, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
+; GFX10-NEXT:    v_sub_f16_sdwa v4, v0, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
+; GFX10-NEXT:    v_sub_f16_e32 v0, v0, v2
 ; GFX10-NEXT:    v_sub_f16_e32 v1, v1, v3
-; GFX10-NEXT:    v_and_b32_e32 v2, 0xffff, v4
-; GFX10-NEXT:    v_lshl_or_b32 v0, v0, 16, v2
+; GFX10-NEXT:    v_perm_b32 v0, v4, v0, 0x5040100
 ; GFX10-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX11-LABEL: v_constained_fsub_v3f16_fpexcept_strict:
@@ -221,8 +218,7 @@
 ; GFX11-NEXT:    v_sub_f16_e32 v0, v0, v2
 ; GFX11-NEXT:    v_sub_f16_e32 v1, v1, v3
 ; GFX11-NEXT:    v_sub_f16_e32 v2, v5, v4
-; GFX11-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX11-NEXT:    v_lshl_or_b32 v0, v2, 16, v0
+; GFX11-NEXT:    v_perm_b32 v0, v2, v0, 0x5040100
 ; GFX11-NEXT:    s_setpc_b64 s[30:31]
   %val = call <3 x half> @llvm.experimental.constrained.fsub.v3f16(<3 x half> %x, <3 x half> %y, metadata !"round.tonearest", metadata !"fpexcept.strict")
   ret <3 x half> %val
@@ -237,8 +233,9 @@
 ; GFX9-NEXT:    v_sub_f16_sdwa v5, v0, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
 ; GFX9-NEXT:    v_sub_f16_e32 v1, v1, v3
 ; GFX9-NEXT:    v_sub_f16_e32 v0, v0, v2
-; GFX9-NEXT:    v_lshl_or_b32 v0, v5, 16, v0
-; GFX9-NEXT:    v_lshl_or_b32 v1, v4, 16, v1
+; GFX9-NEXT:    s_mov_b32 s4, 0x5040100
+; GFX9-NEXT:    v_perm_b32 v0, v5, v0, s4
+; GFX9-NEXT:    v_perm_b32 v1, v4, v1, s4
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX8-LABEL: v_constained_fsub_v4f16_fpexcept_strict:
@@ -256,14 +253,12 @@
 ; GFX10:       ; %bb.0:
 ; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
-; GFX10-NEXT:    v_sub_f16_e32 v4, v0, v2
-; GFX10-NEXT:    v_sub_f16_e32 v5, v1, v3
-; GFX10-NEXT:    v_sub_f16_sdwa v1, v1, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
-; GFX10-NEXT:    v_sub_f16_sdwa v0, v0, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
-; GFX10-NEXT:    v_and_b32_e32 v2, 0xffff, v4
-; GFX10-NEXT:    v_and_b32_e32 v3, 0xffff, v5
-; GFX10-NEXT:    v_lshl_or_b32 v0, v0, 16, v2
-; GFX10-NEXT:    v_lshl_or_b32 v1, v1, 16, v3
+; GFX10-NEXT:    v_sub_f16_sdwa v4, v1, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
+; GFX10-NEXT:    v_sub_f16_sdwa v5, v0, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
+; GFX10-NEXT:    v_sub_f16_e32 v0, v0, v2
+; GFX10-NEXT:    v_sub_f16_e32 v1, v1, v3
+; GFX10-NEXT:    v_perm_b32 v0, v5, v0, 0x5040100
+; GFX10-NEXT:    v_perm_b32 v1, v4, v1, 0x5040100
 ; GFX10-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX11-LABEL: v_constained_fsub_v4f16_fpexcept_strict:
@@ -271,17 +266,15 @@
 ; GFX11-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX11-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-NEXT:    v_lshrrev_b32_e32 v4, 16, v3
-; GFX11-NEXT:    v_lshrrev_b32_e32 v5, 16, v1
-; GFX11-NEXT:    v_lshrrev_b32_e32 v6, 16, v2
-; GFX11-NEXT:    v_lshrrev_b32_e32 v7, 16, v0
-; GFX11-NEXT:    v_sub_f16_e32 v0, v0, v2
+; GFX11-NEXT:    v_lshrrev_b32_e32 v5, 16, v2
+; GFX11-NEXT:    v_lshrrev_b32_e32 v6, 16, v0
+; GFX11-NEXT:    v_lshrrev_b32_e32 v7, 16, v1
 ; GFX11-NEXT:    v_sub_f16_e32 v1, v1, v3
-; GFX11-NEXT:    v_sub_f16_e32 v2, v5, v4
-; GFX11-NEXT:    v_sub_f16_e32 v3, v7, v6
-; GFX11-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX11-NEXT:    v_and_b32_e32 v1, 0xffff, v1
-; GFX11-NEXT:    v_lshl_or_b32 v0, v3, 16, v0
-; GFX11-NEXT:    v_lshl_or_b32 v1, v2, 16, v1
+; GFX11-NEXT:    v_sub_f16_e32 v0, v0, v2
+; GFX11-NEXT:    v_sub_f16_e32 v2, v6, v5
+; GFX11-NEXT:    v_sub_f16_e32 v3, v7, v4
+; GFX11-NEXT:    v_perm_b32 v0, v2, v0, 0x5040100
+; GFX11-NEXT:    v_perm_b32 v1, v3, v1, 0x5040100
 ; GFX11-NEXT:    s_setpc_b64 s[30:31]
   %val = call <4 x half> @llvm.experimental.constrained.fsub.v4f16(<4 x half> %x, <4 x half> %y, metadata !"round.tonearest", metadata !"fpexcept.strict")
   ret <4 x half> %val
diff --git a/llvm/test/CodeGen/AMDGPU/sub.v2i16.ll b/llvm/test/CodeGen/AMDGPU/sub.v2i16.ll
--- a/llvm/test/CodeGen/AMDGPU/sub.v2i16.ll
+++ b/llvm/test/CodeGen/AMDGPU/sub.v2i16.ll
@@ -720,7 +720,6 @@
 ; GFX9-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
 ; GFX9-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x34
 ; GFX9-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
-; GFX9-NEXT:    v_mov_b32_e32 v4, 0xffff
 ; GFX9-NEXT:    v_mov_b32_e32 v1, 0
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    global_load_dword v2, v0, s[6:7] glc
@@ -729,9 +728,9 @@
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    s_mov_b32 s7, 0xf000
 ; GFX9-NEXT:    s_mov_b32 s6, -1
-; GFX9-NEXT:    v_pk_sub_i16 v2, v2, v3
-; GFX9-NEXT:    v_and_b32_e32 v0, 0xffff, v2
-; GFX9-NEXT:    v_and_b32_sdwa v2, v4, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
+; GFX9-NEXT:    v_pk_sub_i16 v0, v2, v3
+; GFX9-NEXT:    v_alignbit_b32 v2, 0, v0, 16
+; GFX9-NEXT:    v_and_b32_e32 v0, 0xffff, v0
 ; GFX9-NEXT:    v_mov_b32_e32 v3, v1
 ; GFX9-NEXT:    buffer_store_dwordx4 v[0:3], off, s[4:7], 0
 ; GFX9-NEXT:    s_endpgm
@@ -767,7 +766,6 @@
 ; GFX10-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
 ; GFX10-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x34
 ; GFX10-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
-; GFX10-NEXT:    v_mov_b32_e32 v3, 0xffff
 ; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-NEXT:    global_load_dword v1, v0, s[6:7] glc dlc
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
@@ -776,10 +774,10 @@
 ; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
 ; GFX10-NEXT:    s_mov_b32 s7, 0x31016000
 ; GFX10-NEXT:    s_mov_b32 s6, -1
-; GFX10-NEXT:    v_pk_sub_i16 v2, v1, v2
+; GFX10-NEXT:    v_pk_sub_i16 v0, v1, v2
 ; GFX10-NEXT:    v_mov_b32_e32 v1, 0
-; GFX10-NEXT:    v_and_b32_e32 v0, 0xffff, v2
-; GFX10-NEXT:    v_and_b32_sdwa v2, v3, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
+; GFX10-NEXT:    v_alignbit_b32 v2, 0, v0, 16
+; GFX10-NEXT:    v_and_b32_e32 v0, 0xffff, v0
 ; GFX10-NEXT:    v_mov_b32_e32 v3, v1
 ; GFX10-NEXT:    buffer_store_dwordx4 v[0:3], off, s[4:7], 0
 ; GFX10-NEXT:    s_endpgm
@@ -800,10 +798,8 @@
 ; GFX11-NEXT:    v_pk_sub_i16 v0, v1, v0
 ; GFX11-NEXT:    v_mov_b32_e32 v1, 0
 ; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX11-NEXT:    v_lshrrev_b32_e32 v2, 16, v0
+; GFX11-NEXT:    v_alignbit_b32 v2, 0, v0, 16
 ; GFX11-NEXT:    v_dual_mov_b32 v3, v1 :: v_dual_and_b32 v0, 0xffff, v0
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_2)
-; GFX11-NEXT:    v_lshl_or_b32 v2, 0, 16, v2
 ; GFX11-NEXT:    buffer_store_b128 v[0:3], off, s[4:7], 0
 ; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-NEXT:    s_endpgm
diff --git a/llvm/test/CodeGen/AMDGPU/vector_shuffle.packed.ll b/llvm/test/CodeGen/AMDGPU/vector_shuffle.packed.ll
--- a/llvm/test/CodeGen/AMDGPU/vector_shuffle.packed.ll
+++ b/llvm/test/CodeGen/AMDGPU/vector_shuffle.packed.ll
@@ -167,7 +167,7 @@
 ; GFX9-NEXT:    global_load_dword v5, v[0:1], off offset:4
 ; GFX9-NEXT:    global_load_dword v4, v[2:3], off offset:4
 ; GFX9-NEXT:    s_waitcnt vmcnt(1)
-; GFX9-NEXT:    v_lshrrev_b32_e32 v0, 16, v5
+; GFX9-NEXT:    v_alignbit_b32 v0, s4, v5, 16
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    v_mov_b32_e32 v1, v4
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
@@ -179,7 +179,7 @@
 ; GFX10-NEXT:    global_load_dword v5, v[0:1], off offset:4
 ; GFX10-NEXT:    global_load_dword v4, v[2:3], off offset:4
 ; GFX10-NEXT:    s_waitcnt vmcnt(1)
-; GFX10-NEXT:    v_lshrrev_b32_e32 v0, 16, v5
+; GFX10-NEXT:    v_alignbit_b32 v0, s4, v5, 16
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    v_mov_b32_e32 v1, v4
 ; GFX10-NEXT:    s_setpc_b64 s[30:31]
@@ -191,7 +191,7 @@
 ; GFX11-NEXT:    global_load_b32 v0, v[0:1], off offset:4
 ; GFX11-NEXT:    global_load_b32 v1, v[2:3], off offset:4
 ; GFX11-NEXT:    s_waitcnt vmcnt(1)
-; GFX11-NEXT:    v_lshrrev_b32_e32 v0, 16, v0
+; GFX11-NEXT:    v_alignbit_b32 v0, s0, v0, 16
 ; GFX11-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-NEXT:    s_setpc_b64 s[30:31]
   %val0 = load <4 x half>, <4 x half> addrspace(1)* %arg0
@@ -207,7 +207,7 @@
 ; GFX9-NEXT:    global_load_dword v5, v[0:1], off offset:4
 ; GFX9-NEXT:    global_load_dword v4, v[2:3], off offset:4
 ; GFX9-NEXT:    s_waitcnt vmcnt(1)
-; GFX9-NEXT:    v_lshrrev_b32_e32 v0, 16, v5
+; GFX9-NEXT:    v_alignbit_b32 v0, s4, v5, 16
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    v_mov_b32_e32 v1, v4
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
@@ -219,7 +219,7 @@
 ; GFX10-NEXT:    global_load_dword v5, v[0:1], off offset:4
 ; GFX10-NEXT:    global_load_dword v4, v[2:3], off offset:4
 ; GFX10-NEXT:    s_waitcnt vmcnt(1)
-; GFX10-NEXT:    v_lshrrev_b32_e32 v0, 16, v5
+; GFX10-NEXT:    v_alignbit_b32 v0, s4, v5, 16
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    v_mov_b32_e32 v1, v4
 ; GFX10-NEXT:    s_setpc_b64 s[30:31]
@@ -231,7 +231,7 @@
 ; GFX11-NEXT:    global_load_b32 v0, v[0:1], off offset:4
 ; GFX11-NEXT:    global_load_b32 v1, v[2:3], off offset:4
 ; GFX11-NEXT:    s_waitcnt vmcnt(1)
-; GFX11-NEXT:    v_lshrrev_b32_e32 v0, 16, v0
+; GFX11-NEXT:    v_alignbit_b32 v0, s0, v0, 16
 ; GFX11-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-NEXT:    s_setpc_b64 s[30:31]
   %val0 = load <4 x half>, <4 x half> addrspace(1)* %arg0
@@ -246,12 +246,9 @@
 ; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX9-NEXT:    global_load_dword v5, v[0:1], off offset:4
 ; GFX9-NEXT:    global_load_dword v4, v[2:3], off
-; GFX9-NEXT:    v_mov_b32_e32 v0, 0xffff
-; GFX9-NEXT:    s_waitcnt vmcnt(1)
-; GFX9-NEXT:    v_and_b32_sdwa v0, v0, v5 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
+; GFX9-NEXT:    s_mov_b32 s4, 0x7060302
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
-; GFX9-NEXT:    v_lshrrev_b32_e32 v1, 16, v4
-; GFX9-NEXT:    v_lshl_or_b32 v0, v1, 16, v0
+; GFX9-NEXT:    v_perm_b32 v0, v4, v5, s4
 ; GFX9-NEXT:    v_mov_b32_e32 v1, v4
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
 ;
@@ -261,12 +258,8 @@
 ; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-NEXT:    global_load_dword v5, v[0:1], off offset:4
 ; GFX10-NEXT:    global_load_dword v4, v[2:3], off
-; GFX10-NEXT:    v_mov_b32_e32 v0, 0xffff
-; GFX10-NEXT:    s_waitcnt vmcnt(1)
-; GFX10-NEXT:    v_and_b32_sdwa v0, v0, v5 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
-; GFX10-NEXT:    v_lshrrev_b32_e32 v1, 16, v4
-; GFX10-NEXT:    v_lshl_or_b32 v0, v1, 16, v0
+; GFX10-NEXT:    v_perm_b32 v0, v4, v5, 0x7060302
 ; GFX10-NEXT:    v_mov_b32_e32 v1, v4
 ; GFX10-NEXT:    s_setpc_b64 s[30:31]
 ;
@@ -276,13 +269,8 @@
 ; GFX11-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-NEXT:    global_load_b32 v0, v[0:1], off offset:4
 ; GFX11-NEXT:    global_load_b32 v1, v[2:3], off
-; GFX11-NEXT:    s_waitcnt vmcnt(1)
-; GFX11-NEXT:    v_lshrrev_b32_e32 v0, 16, v0
 ; GFX11-NEXT:    s_waitcnt vmcnt(0)
-; GFX11-NEXT:    v_lshrrev_b32_e32 v2, 16, v1
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX11-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX11-NEXT:    v_lshl_or_b32 v0, v2, 16, v0
+; GFX11-NEXT:    v_perm_b32 v0, v1, v0, 0x7060302
 ; GFX11-NEXT:    s_setpc_b64 s[30:31]
   %val0 = load <4 x half>, <4 x half> addrspace(1)* %arg0
   %val1 = load <4 x half>, <4 x half> addrspace(1)* %arg1
@@ -294,46 +282,37 @@
 ; GFX9-LABEL: shuffle_v4f16_357u:
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX9-NEXT:    global_load_dword v6, v[0:1], off offset:4
 ; GFX9-NEXT:    global_load_dwordx2 v[4:5], v[2:3], off
-; GFX9-NEXT:    v_mov_b32_e32 v0, 0xffff
+; GFX9-NEXT:    global_load_dword v6, v[0:1], off offset:4
+; GFX9-NEXT:    s_mov_b32 s4, 0x7060302
 ; GFX9-NEXT:    s_waitcnt vmcnt(1)
-; GFX9-NEXT:    v_and_b32_sdwa v0, v0, v6 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
+; GFX9-NEXT:    v_alignbit_b32 v1, s4, v5, 16
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
-; GFX9-NEXT:    v_lshrrev_b32_e32 v2, 16, v4
-; GFX9-NEXT:    v_lshrrev_b32_e32 v1, 16, v5
-; GFX9-NEXT:    v_lshl_or_b32 v0, v2, 16, v0
+; GFX9-NEXT:    v_perm_b32 v0, v4, v6, s4
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX10-LABEL: shuffle_v4f16_357u:
 ; GFX10:       ; %bb.0:
 ; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
-; GFX10-NEXT:    global_load_dword v6, v[0:1], off offset:4
 ; GFX10-NEXT:    global_load_dwordx2 v[4:5], v[2:3], off
-; GFX10-NEXT:    v_mov_b32_e32 v0, 0xffff
+; GFX10-NEXT:    global_load_dword v6, v[0:1], off offset:4
 ; GFX10-NEXT:    s_waitcnt vmcnt(1)
-; GFX10-NEXT:    v_and_b32_sdwa v0, v0, v6 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
+; GFX10-NEXT:    v_alignbit_b32 v1, s4, v5, 16
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
-; GFX10-NEXT:    v_lshrrev_b32_e32 v1, 16, v4
-; GFX10-NEXT:    v_lshl_or_b32 v0, v1, 16, v0
-; GFX10-NEXT:    v_lshrrev_b32_e32 v1, 16, v5
+; GFX10-NEXT:    v_perm_b32 v0, v4, v6, 0x7060302
 ; GFX10-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX11-LABEL: shuffle_v4f16_357u:
 ; GFX11:       ; %bb.0:
 ; GFX11-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX11-NEXT:    s_waitcnt_vscnt null, 0x0
-; GFX11-NEXT:    global_load_b32 v4, v[0:1], off offset:4
-; GFX11-NEXT:    global_load_b64 v[0:1], v[2:3], off
+; GFX11-NEXT:    global_load_b64 v[2:3], v[2:3], off
+; GFX11-NEXT:    global_load_b32 v0, v[0:1], off offset:4
 ; GFX11-NEXT:    s_waitcnt vmcnt(1)
-; GFX11-NEXT:    v_lshrrev_b32_e32 v2, 16, v4
+; GFX11-NEXT:    v_alignbit_b32 v1, s0, v3, 16
 ; GFX11-NEXT:    s_waitcnt vmcnt(0)
-; GFX11-NEXT:    v_lshrrev_b32_e32 v0, 16, v0
-; GFX11-NEXT:    v_lshrrev_b32_e32 v1, 16, v1
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX11-NEXT:    v_and_b32_e32 v2, 0xffff, v2
-; GFX11-NEXT:    v_lshl_or_b32 v0, v0, 16, v2
+; GFX11-NEXT:    v_perm_b32 v0, v2, v0, 0x7060302
 ; GFX11-NEXT:    s_setpc_b64 s[30:31]
   %val0 = load <4 x half>, <4 x half> addrspace(1)* %arg0
   %val1 = load <4 x half>, <4 x half> addrspace(1)* %arg1
@@ -909,10 +888,8 @@
 ; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX9-NEXT:    global_load_dwordx2 v[5:6], v[2:3], off
 ; GFX9-NEXT:    global_load_dword v4, v[0:1], off offset:4
-; GFX9-NEXT:    v_mov_b32_e32 v0, 0xffff
 ; GFX9-NEXT:    s_waitcnt vmcnt(1)
-; GFX9-NEXT:    v_and_b32_sdwa v0, v0, v5 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
-; GFX9-NEXT:    v_lshl_or_b32 v1, v6, 16, v0
+; GFX9-NEXT:    v_alignbit_b32 v1, v6, v5, 16
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    v_mov_b32_e32 v0, v4
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
@@ -923,10 +900,8 @@
 ; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-NEXT:    global_load_dwordx2 v[5:6], v[2:3], off
 ; GFX10-NEXT:    global_load_dword v4, v[0:1], off offset:4
-; GFX10-NEXT:    v_mov_b32_e32 v0, 0xffff
 ; GFX10-NEXT:    s_waitcnt vmcnt(1)
-; GFX10-NEXT:    v_and_b32_sdwa v0, v0, v5 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
-; GFX10-NEXT:    v_lshl_or_b32 v1, v6, 16, v0
+; GFX10-NEXT:    v_alignbit_b32 v1, v6, v5, 16
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    v_mov_b32_e32 v0, v4
 ; GFX10-NEXT:    s_setpc_b64 s[30:31]
@@ -938,10 +913,7 @@
 ; GFX11-NEXT:    global_load_b64 v[2:3], v[2:3], off
 ; GFX11-NEXT:    global_load_b32 v0, v[0:1], off offset:4
 ; GFX11-NEXT:    s_waitcnt vmcnt(1)
-; GFX11-NEXT:    v_lshrrev_b32_e32 v1, 16, v2
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX11-NEXT:    v_and_b32_e32 v1, 0xffff, v1
-; GFX11-NEXT:    v_lshl_or_b32 v1, v3, 16, v1
+; GFX11-NEXT:    v_alignbit_b32 v1, v3, v2, 16
 ; GFX11-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-NEXT:    s_setpc_b64 s[30:31]
   %val0 = load <4 x half>, <4 x half> addrspace(1)* %arg0
@@ -956,10 +928,8 @@
 ; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX9-NEXT:    global_load_dwordx2 v[5:6], v[2:3], off
 ; GFX9-NEXT:    global_load_dword v4, v[0:1], off offset:4
-; GFX9-NEXT:    v_mov_b32_e32 v0, 0xffff
 ; GFX9-NEXT:    s_waitcnt vmcnt(1)
-; GFX9-NEXT:    v_and_b32_sdwa v0, v0, v5 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
-; GFX9-NEXT:    v_lshl_or_b32 v0, v6, 16, v0
+; GFX9-NEXT:    v_alignbit_b32 v0, v6, v5, 16
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    v_mov_b32_e32 v1, v4
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
@@ -970,12 +940,10 @@
 ; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-NEXT:    global_load_dwordx2 v[5:6], v[2:3], off
 ; GFX10-NEXT:    global_load_dword v4, v[0:1], off offset:4
-; GFX10-NEXT:    v_mov_b32_e32 v0, 0xffff
 ; GFX10-NEXT:    s_waitcnt vmcnt(1)
-; GFX10-NEXT:    v_and_b32_sdwa v0, v0, v5 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
+; GFX10-NEXT:    v_alignbit_b32 v0, v6, v5, 16
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    v_mov_b32_e32 v1, v4
-; GFX10-NEXT:    v_lshl_or_b32 v0, v6, 16, v0
 ; GFX10-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX11-LABEL: shuffle_v4f16_5623:
@@ -985,10 +953,7 @@
 ; GFX11-NEXT:    global_load_b64 v[2:3], v[2:3], off
 ; GFX11-NEXT:    global_load_b32 v1, v[0:1], off offset:4
 ; GFX11-NEXT:    s_waitcnt vmcnt(1)
-; GFX11-NEXT:    v_lshrrev_b32_e32 v0, 16, v2
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX11-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX11-NEXT:    v_lshl_or_b32 v0, v3, 16, v0
+; GFX11-NEXT:    v_alignbit_b32 v0, v3, v2, 16
 ; GFX11-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-NEXT:    s_setpc_b64 s[30:31]
   %val0 = load <4 x half>, <4 x half> addrspace(1)* %arg0
@@ -1001,48 +966,36 @@
 ; GFX9-LABEL: shuffle_v4f16_3456:
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX9-NEXT:    global_load_dword v6, v[0:1], off offset:4
 ; GFX9-NEXT:    global_load_dwordx2 v[4:5], v[2:3], off
-; GFX9-NEXT:    v_mov_b32_e32 v0, 0xffff
+; GFX9-NEXT:    global_load_dword v6, v[0:1], off offset:4
 ; GFX9-NEXT:    s_waitcnt vmcnt(1)
-; GFX9-NEXT:    v_and_b32_sdwa v1, v0, v6 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
+; GFX9-NEXT:    v_alignbit_b32 v1, v5, v4, 16
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
-; GFX9-NEXT:    v_and_b32_sdwa v2, v0, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
-; GFX9-NEXT:    v_lshl_or_b32 v0, v4, 16, v1
-; GFX9-NEXT:    v_lshl_or_b32 v1, v5, 16, v2
+; GFX9-NEXT:    v_alignbit_b32 v0, v4, v6, 16
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX10-LABEL: shuffle_v4f16_3456:
 ; GFX10:       ; %bb.0:
 ; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
-; GFX10-NEXT:    global_load_dword v6, v[0:1], off offset:4
 ; GFX10-NEXT:    global_load_dwordx2 v[4:5], v[2:3], off
-; GFX10-NEXT:    v_mov_b32_e32 v0, 0xffff
+; GFX10-NEXT:    global_load_dword v6, v[0:1], off offset:4
 ; GFX10-NEXT:    s_waitcnt vmcnt(1)
-; GFX10-NEXT:    v_and_b32_sdwa v1, v0, v6 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
+; GFX10-NEXT:    v_alignbit_b32 v1, v5, v4, 16
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
-; GFX10-NEXT:    v_and_b32_sdwa v2, v0, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
-; GFX10-NEXT:    v_lshl_or_b32 v0, v4, 16, v1
-; GFX10-NEXT:    v_lshl_or_b32 v1, v5, 16, v2
+; GFX10-NEXT:    v_alignbit_b32 v0, v4, v6, 16
 ; GFX10-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX11-LABEL: shuffle_v4f16_3456:
 ; GFX11:       ; %bb.0:
 ; GFX11-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX11-NEXT:    s_waitcnt_vscnt null, 0x0
-; GFX11-NEXT:    global_load_b32 v4, v[0:1], off offset:4
-; GFX11-NEXT:    global_load_b64 v[0:1], v[2:3], off
+; GFX11-NEXT:    global_load_b64 v[2:3], v[2:3], off
+; GFX11-NEXT:    global_load_b32 v0, v[0:1], off offset:4
 ; GFX11-NEXT:    s_waitcnt vmcnt(1)
-; GFX11-NEXT:    v_lshrrev_b32_e32 v2, 16, v4
+; GFX11-NEXT:    v_alignbit_b32 v1, v3, v2, 16
 ; GFX11-NEXT:    s_waitcnt vmcnt(0)
-; GFX11-NEXT:    v_lshrrev_b32_e32 v3, 16, v0
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX11-NEXT:    v_and_b32_e32 v2, 0xffff, v2
-; GFX11-NEXT:    v_and_b32_e32 v3, 0xffff, v3
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX11-NEXT:    v_lshl_or_b32 v0, v0, 16, v2
-; GFX11-NEXT:    v_lshl_or_b32 v1, v1, 16, v3
+; GFX11-NEXT:    v_alignbit_b32 v0, v2, v0, 16
 ; GFX11-NEXT:    s_setpc_b64 s[30:31]
   %val0 = load <4 x half>, <4 x half> addrspace(1)* %arg0
   %val1 = load <4 x half>, <4 x half> addrspace(1)* %arg1
@@ -1054,15 +1007,12 @@
 ; GFX9-LABEL: shuffle_v4f16_5634:
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX9-NEXT:    global_load_dword v6, v[0:1], off offset:4
 ; GFX9-NEXT:    global_load_dwordx2 v[4:5], v[2:3], off
-; GFX9-NEXT:    v_mov_b32_e32 v0, 0xffff
+; GFX9-NEXT:    global_load_dword v6, v[0:1], off offset:4
 ; GFX9-NEXT:    s_waitcnt vmcnt(1)
-; GFX9-NEXT:    v_and_b32_sdwa v1, v0, v6 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
+; GFX9-NEXT:    v_alignbit_b32 v0, v5, v4, 16
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
-; GFX9-NEXT:    v_and_b32_sdwa v0, v0, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
-; GFX9-NEXT:    v_lshl_or_b32 v1, v4, 16, v1
-; GFX9-NEXT:    v_lshl_or_b32 v0, v5, 16, v0
+; GFX9-NEXT:    v_alignbit_b32 v1, v4, v6, 16
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX10-LABEL: shuffle_v4f16_5634:
@@ -1071,13 +1021,10 @@
 ; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-NEXT:    global_load_dwordx2 v[4:5], v[2:3], off
 ; GFX10-NEXT:    global_load_dword v6, v[0:1], off offset:4
-; GFX10-NEXT:    v_mov_b32_e32 v0, 0xffff
 ; GFX10-NEXT:    s_waitcnt vmcnt(1)
-; GFX10-NEXT:    v_and_b32_sdwa v1, v0, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
+; GFX10-NEXT:    v_alignbit_b32 v0, v5, v4, 16
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
-; GFX10-NEXT:    v_and_b32_sdwa v2, v0, v6 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
-; GFX10-NEXT:    v_lshl_or_b32 v0, v5, 16, v1
-; GFX10-NEXT:    v_lshl_or_b32 v1, v4, 16, v2
+; GFX10-NEXT:    v_alignbit_b32 v1, v4, v6, 16
 ; GFX10-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX11-LABEL: shuffle_v4f16_5634:
@@ -1085,17 +1032,11 @@
 ; GFX11-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX11-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-NEXT:    global_load_b64 v[2:3], v[2:3], off
-; GFX11-NEXT:    global_load_b32 v0, v[0:1], off offset:4
+; GFX11-NEXT:    global_load_b32 v1, v[0:1], off offset:4
 ; GFX11-NEXT:    s_waitcnt vmcnt(1)
-; GFX11-NEXT:    v_lshrrev_b32_e32 v1, 16, v2
+; GFX11-NEXT:    v_alignbit_b32 v0, v3, v2, 16
 ; GFX11-NEXT:    s_waitcnt vmcnt(0)
-; GFX11-NEXT:    v_lshrrev_b32_e32 v0, 16, v0
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX11-NEXT:    v_and_b32_e32 v1, 0xffff, v1
-; GFX11-NEXT:    v_and_b32_e32 v4, 0xffff, v0
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX11-NEXT:    v_lshl_or_b32 v0, v3, 16, v1
-; GFX11-NEXT:    v_lshl_or_b32 v1, v2, 16, v4
+; GFX11-NEXT:    v_alignbit_b32 v1, v2, v1, 16
 ; GFX11-NEXT:    s_setpc_b64 s[30:31]
   %val0 = load <4 x half>, <4 x half> addrspace(1)* %arg0
   %val1 = load <4 x half>, <4 x half> addrspace(1)* %arg1
@@ -1107,16 +1048,13 @@
 ; GFX9-LABEL: shuffle_v4f16_5734:
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX9-NEXT:    global_load_dword v6, v[0:1], off offset:4
 ; GFX9-NEXT:    global_load_dwordx2 v[4:5], v[2:3], off
-; GFX9-NEXT:    v_mov_b32_e32 v0, 0xffff
+; GFX9-NEXT:    global_load_dword v6, v[0:1], off offset:4
+; GFX9-NEXT:    s_mov_b32 s4, 0x7060302
 ; GFX9-NEXT:    s_waitcnt vmcnt(1)
-; GFX9-NEXT:    v_and_b32_sdwa v1, v0, v6 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
+; GFX9-NEXT:    v_perm_b32 v0, v5, v4, s4
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
-; GFX9-NEXT:    v_and_b32_sdwa v0, v0, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
-; GFX9-NEXT:    v_lshrrev_b32_e32 v2, 16, v5
-; GFX9-NEXT:    v_lshl_or_b32 v1, v4, 16, v1
-; GFX9-NEXT:    v_lshl_or_b32 v0, v2, 16, v0
+; GFX9-NEXT:    v_alignbit_b32 v1, v4, v6, 16
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX10-LABEL: shuffle_v4f16_5734:
@@ -1125,14 +1063,10 @@
 ; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-NEXT:    global_load_dwordx2 v[4:5], v[2:3], off
 ; GFX10-NEXT:    global_load_dword v6, v[0:1], off offset:4
-; GFX10-NEXT:    v_mov_b32_e32 v0, 0xffff
 ; GFX10-NEXT:    s_waitcnt vmcnt(1)
-; GFX10-NEXT:    v_and_b32_sdwa v1, v0, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
-; GFX10-NEXT:    v_lshrrev_b32_e32 v2, 16, v5
+; GFX10-NEXT:    v_perm_b32 v0, v5, v4, 0x7060302
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
-; GFX10-NEXT:    v_and_b32_sdwa v3, v0, v6 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
-; GFX10-NEXT:    v_lshl_or_b32 v0, v2, 16, v1
-; GFX10-NEXT:    v_lshl_or_b32 v1, v4, 16, v3
+; GFX10-NEXT:    v_alignbit_b32 v1, v4, v6, 16
 ; GFX10-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX11-LABEL: shuffle_v4f16_5734:
@@ -1140,18 +1074,11 @@
 ; GFX11-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX11-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-NEXT:    global_load_b64 v[2:3], v[2:3], off
-; GFX11-NEXT:    global_load_b32 v0, v[0:1], off offset:4
+; GFX11-NEXT:    global_load_b32 v1, v[0:1], off offset:4
 ; GFX11-NEXT:    s_waitcnt vmcnt(1)
-; GFX11-NEXT:    v_lshrrev_b32_e32 v1, 16, v2
+; GFX11-NEXT:    v_perm_b32 v0, v3, v2, 0x7060302
 ; GFX11-NEXT:    s_waitcnt vmcnt(0)
-; GFX11-NEXT:    v_lshrrev_b32_e32 v0, 16, v0
-; GFX11-NEXT:    v_lshrrev_b32_e32 v3, 16, v3
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3)
-; GFX11-NEXT:    v_and_b32_e32 v1, 0xffff, v1
-; GFX11-NEXT:    v_and_b32_e32 v4, 0xffff, v0
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX11-NEXT:    v_lshl_or_b32 v0, v3, 16, v1
-; GFX11-NEXT:    v_lshl_or_b32 v1, v2, 16, v4
+; GFX11-NEXT:    v_alignbit_b32 v1, v2, v1, 16
 ; GFX11-NEXT:    s_setpc_b64 s[30:31]
   %val0 = load <4 x half>, <4 x half> addrspace(1)* %arg0
   %val1 = load <4 x half>, <4 x half> addrspace(1)* %arg1
@@ -1165,10 +1092,8 @@
 ; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX9-NEXT:    global_load_dwordx2 v[5:6], v[2:3], off
 ; GFX9-NEXT:    global_load_dword v4, v[0:1], off offset:4
-; GFX9-NEXT:    v_mov_b32_e32 v0, 0xffff
 ; GFX9-NEXT:    s_waitcnt vmcnt(1)
-; GFX9-NEXT:    v_and_b32_sdwa v0, v0, v5 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
-; GFX9-NEXT:    v_lshl_or_b32 v1, v6, 16, v0
+; GFX9-NEXT:    v_alignbit_b32 v1, v6, v5, 16
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    v_mov_b32_e32 v0, v4
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
@@ -1179,10 +1104,8 @@
 ; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-NEXT:    global_load_dwordx2 v[5:6], v[2:3], off
 ; GFX10-NEXT:    global_load_dword v4, v[0:1], off offset:4
-; GFX10-NEXT:    v_mov_b32_e32 v0, 0xffff
 ; GFX10-NEXT:    s_waitcnt vmcnt(1)
-; GFX10-NEXT:    v_and_b32_sdwa v0, v0, v5 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
-; GFX10-NEXT:    v_lshl_or_b32 v1, v6, 16, v0
+; GFX10-NEXT:    v_alignbit_b32 v1, v6, v5, 16
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    v_mov_b32_e32 v0, v4
 ; GFX10-NEXT:    s_setpc_b64 s[30:31]
@@ -1194,9 +1117,7 @@
 ; GFX11-NEXT:    global_load_b64 v[2:3], v[2:3], off
 ; GFX11-NEXT:    global_load_b32 v0, v[0:1], off offset:4
 ; GFX11-NEXT:    s_waitcnt vmcnt(1)
-; GFX11-NEXT:    v_lshrrev_b32_e32 v1, 16, v2
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
-; GFX11-NEXT:    v_lshl_or_b32 v1, v3, 16, v1
+; GFX11-NEXT:    v_alignbit_b32 v1, v3, v2, 16
 ; GFX11-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-NEXT:    s_setpc_b64 s[30:31]
   %val0 = load <4 x i16>, <4 x i16> addrspace(1)* %arg0
@@ -1248,9 +1169,9 @@
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX9-NEXT:    global_load_dwordx2 v[0:1], v[0:1], off
+; GFX9-NEXT:    s_mov_b32 s4, 0x5040100
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
-; GFX9-NEXT:    v_and_b32_e32 v1, 0xffff, v0
-; GFX9-NEXT:    v_lshl_or_b32 v0, v0, 16, v1
+; GFX9-NEXT:    v_perm_b32 v0, v0, v0, s4
 ; GFX9-NEXT:    v_mov_b32_e32 v1, v0
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
 ;
@@ -1260,8 +1181,7 @@
 ; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-NEXT:    global_load_dwordx2 v[0:1], v[0:1], off
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
-; GFX10-NEXT:    v_and_b32_e32 v1, 0xffff, v0
-; GFX10-NEXT:    v_lshl_or_b32 v0, v0, 16, v1
+; GFX10-NEXT:    v_perm_b32 v0, v0, v0, 0x5040100
 ; GFX10-NEXT:    v_mov_b32_e32 v1, v0
 ; GFX10-NEXT:    s_setpc_b64 s[30:31]
 ;
@@ -1271,9 +1191,8 @@
 ; GFX11-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-NEXT:    global_load_b64 v[0:1], v[0:1], off
 ; GFX11-NEXT:    s_waitcnt vmcnt(0)
-; GFX11-NEXT:    v_and_b32_e32 v1, 0xffff, v0
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX11-NEXT:    v_lshl_or_b32 v0, v0, 16, v1
+; GFX11-NEXT:    v_perm_b32 v0, v0, v0, 0x5040100
+; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
 ; GFX11-NEXT:    v_mov_b32_e32 v1, v0
 ; GFX11-NEXT:    s_setpc_b64 s[30:31]
   %val0 = load <4 x half>, <4 x half> addrspace(1)* %arg0
@@ -1288,9 +1207,7 @@
 ; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX9-NEXT:    global_load_dwordx2 v[0:1], v[0:1], off
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
-; GFX9-NEXT:    v_mov_b32_e32 v1, 0xffff
-; GFX9-NEXT:    v_and_b32_sdwa v1, v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
-; GFX9-NEXT:    v_lshl_or_b32 v0, v0, 16, v1
+; GFX9-NEXT:    v_alignbit_b32 v0, v0, v0, 16
 ; GFX9-NEXT:    v_mov_b32_e32 v1, v0
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
 ;
@@ -1300,9 +1217,7 @@
 ; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-NEXT:    global_load_dwordx2 v[0:1], v[0:1], off
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
-; GFX10-NEXT:    v_mov_b32_e32 v1, 0xffff
-; GFX10-NEXT:    v_and_b32_sdwa v1, v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
-; GFX10-NEXT:    v_lshl_or_b32 v0, v0, 16, v1
+; GFX10-NEXT:    v_alignbit_b32 v0, v0, v0, 16
 ; GFX10-NEXT:    v_mov_b32_e32 v1, v0
 ; GFX10-NEXT:    s_setpc_b64 s[30:31]
 ;
@@ -1312,10 +1227,7 @@
 ; GFX11-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-NEXT:    global_load_b64 v[0:1], v[0:1], off
 ; GFX11-NEXT:    s_waitcnt vmcnt(0)
-; GFX11-NEXT:    v_lshrrev_b32_e32 v1, 16, v0
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX11-NEXT:    v_and_b32_e32 v1, 0xffff, v1
-; GFX11-NEXT:    v_lshl_or_b32 v0, v0, 16, v1
+; GFX11-NEXT:    v_alignbit_b32 v0, v0, v0, 16
 ; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
 ; GFX11-NEXT:    v_mov_b32_e32 v1, v0
 ; GFX11-NEXT:    s_setpc_b64 s[30:31]
@@ -1329,13 +1241,12 @@
 ; GFX9-LABEL: shuffle_v4f16_1100:
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX9-NEXT:    global_load_dwordx2 v[0:1], v[0:1], off
+; GFX9-NEXT:    global_load_dwordx2 v[1:2], v[0:1], off
+; GFX9-NEXT:    s_mov_b32 s4, 0x7060302
+; GFX9-NEXT:    s_mov_b32 s5, 0x5040100
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
-; GFX9-NEXT:    v_and_b32_e32 v1, 0xffff, v0
-; GFX9-NEXT:    v_lshrrev_b32_e32 v2, 16, v0
-; GFX9-NEXT:    v_lshl_or_b32 v1, v0, 16, v1
-; GFX9-NEXT:    v_and_b32_e32 v0, 0xffff, v2
-; GFX9-NEXT:    v_lshl_or_b32 v0, v2, 16, v0
+; GFX9-NEXT:    v_perm_b32 v0, v1, v1, s4
+; GFX9-NEXT:    v_perm_b32 v1, v1, v1, s5
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX10-LABEL: shuffle_v4f16_1100:
@@ -1344,11 +1255,8 @@
 ; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-NEXT:    global_load_dwordx2 v[1:2], v[0:1], off
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
-; GFX10-NEXT:    v_lshrrev_b32_e32 v0, 16, v1
-; GFX10-NEXT:    v_and_b32_e32 v3, 0xffff, v1
-; GFX10-NEXT:    v_and_b32_e32 v2, 0xffff, v0
-; GFX10-NEXT:    v_lshl_or_b32 v1, v1, 16, v3
-; GFX10-NEXT:    v_lshl_or_b32 v0, v0, 16, v2
+; GFX10-NEXT:    v_perm_b32 v0, v1, v1, 0x7060302
+; GFX10-NEXT:    v_perm_b32 v1, v1, v1, 0x5040100
 ; GFX10-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX11-LABEL: shuffle_v4f16_1100:
@@ -1357,13 +1265,8 @@
 ; GFX11-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-NEXT:    global_load_b64 v[1:2], v[0:1], off
 ; GFX11-NEXT:    s_waitcnt vmcnt(0)
-; GFX11-NEXT:    v_lshrrev_b32_e32 v0, 16, v1
-; GFX11-NEXT:    v_and_b32_e32 v3, 0xffff, v1
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX11-NEXT:    v_and_b32_e32 v2, 0xffff, v0
-; GFX11-NEXT:    v_lshl_or_b32 v1, v1, 16, v3
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_2)
-; GFX11-NEXT:    v_lshl_or_b32 v0, v0, 16, v2
+; GFX11-NEXT:    v_perm_b32 v0, v1, v1, 0x7060302
+; GFX11-NEXT:    v_perm_b32 v1, v1, v1, 0x5040100
 ; GFX11-NEXT:    s_setpc_b64 s[30:31]
   %val0 = load <4 x half>, <4 x half> addrspace(1)* %arg0
   %val1 = load <4 x half>, <4 x half> addrspace(1)* %arg1
@@ -1375,13 +1278,11 @@
 ; GFX9-LABEL: shuffle_v4f16_6161:
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX9-NEXT:    global_load_dword v4, v[2:3], off offset:4
-; GFX9-NEXT:    global_load_dword v5, v[0:1], off
-; GFX9-NEXT:    s_waitcnt vmcnt(1)
-; GFX9-NEXT:    v_and_b32_e32 v0, 0xffff, v4
+; GFX9-NEXT:    global_load_dword v4, v[0:1], off
+; GFX9-NEXT:    global_load_dword v5, v[2:3], off offset:4
+; GFX9-NEXT:    s_mov_b32 s4, 0x7060100
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
-; GFX9-NEXT:    v_lshrrev_b32_e32 v1, 16, v5
-; GFX9-NEXT:    v_lshl_or_b32 v0, v1, 16, v0
+; GFX9-NEXT:    v_perm_b32 v0, v4, v5, s4
 ; GFX9-NEXT:    v_mov_b32_e32 v1, v0
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
 ;
@@ -1389,13 +1290,10 @@
 ; GFX10:       ; %bb.0:
 ; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
-; GFX10-NEXT:    global_load_dword v4, v[2:3], off offset:4
-; GFX10-NEXT:    global_load_dword v5, v[0:1], off
-; GFX10-NEXT:    s_waitcnt vmcnt(1)
-; GFX10-NEXT:    v_and_b32_e32 v0, 0xffff, v4
+; GFX10-NEXT:    global_load_dword v4, v[0:1], off
+; GFX10-NEXT:    global_load_dword v5, v[2:3], off offset:4
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
-; GFX10-NEXT:    v_lshrrev_b32_e32 v1, 16, v5
-; GFX10-NEXT:    v_lshl_or_b32 v0, v1, 16, v0
+; GFX10-NEXT:    v_perm_b32 v0, v4, v5, 0x7060100
 ; GFX10-NEXT:    v_mov_b32_e32 v1, v0
 ; GFX10-NEXT:    s_setpc_b64 s[30:31]
 ;
@@ -1403,14 +1301,11 @@
 ; GFX11:       ; %bb.0:
 ; GFX11-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX11-NEXT:    s_waitcnt_vscnt null, 0x0
-; GFX11-NEXT:    global_load_b32 v2, v[2:3], off offset:4
 ; GFX11-NEXT:    global_load_b32 v0, v[0:1], off
-; GFX11-NEXT:    s_waitcnt vmcnt(1)
-; GFX11-NEXT:    v_and_b32_e32 v1, 0xffff, v2
+; GFX11-NEXT:    global_load_b32 v1, v[2:3], off offset:4
 ; GFX11-NEXT:    s_waitcnt vmcnt(0)
-; GFX11-NEXT:    v_lshrrev_b32_e32 v0, 16, v0
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX11-NEXT:    v_lshl_or_b32 v0, v0, 16, v1
+; GFX11-NEXT:    v_perm_b32 v0, v0, v1, 0x7060100
+; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
 ; GFX11-NEXT:    v_mov_b32_e32 v1, v0
 ; GFX11-NEXT:    s_setpc_b64 s[30:31]
   %val0 = load <4 x half>, <4 x half> addrspace(1)* %arg0
@@ -1424,10 +1319,9 @@
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX9-NEXT:    global_load_dword v0, v[0:1], off offset:4
+; GFX9-NEXT:    s_mov_b32 s4, 0x7060302
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
-; GFX9-NEXT:    v_lshrrev_b32_e32 v1, 16, v0
-; GFX9-NEXT:    v_and_b32_e32 v2, 0xffff, v1
-; GFX9-NEXT:    v_lshl_or_b32 v1, v1, 16, v2
+; GFX9-NEXT:    v_perm_b32 v1, v0, v0, s4
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX10-LABEL: shuffle_v4f16_2333:
@@ -1436,9 +1330,7 @@
 ; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-NEXT:    global_load_dword v0, v[0:1], off offset:4
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
-; GFX10-NEXT:    v_lshrrev_b32_e32 v1, 16, v0
-; GFX10-NEXT:    v_and_b32_e32 v2, 0xffff, v1
-; GFX10-NEXT:    v_lshl_or_b32 v1, v1, 16, v2
+; GFX10-NEXT:    v_perm_b32 v1, v0, v0, 0x7060302
 ; GFX10-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX11-LABEL: shuffle_v4f16_2333:
@@ -1447,10 +1339,7 @@
 ; GFX11-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-NEXT:    global_load_b32 v0, v[0:1], off offset:4
 ; GFX11-NEXT:    s_waitcnt vmcnt(0)
-; GFX11-NEXT:    v_lshrrev_b32_e32 v1, 16, v0
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX11-NEXT:    v_and_b32_e32 v2, 0xffff, v1
-; GFX11-NEXT:    v_lshl_or_b32 v1, v1, 16, v2
+; GFX11-NEXT:    v_perm_b32 v1, v0, v0, 0x7060302
 ; GFX11-NEXT:    s_setpc_b64 s[30:31]
   %val0 = load <4 x half>, <4 x half> addrspace(1)* %arg0
   %val1 = load <4 x half>, <4 x half> addrspace(1)* %arg1
@@ -1463,10 +1352,9 @@
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX9-NEXT:    global_load_dword v0, v[0:1], off offset:4
+; GFX9-NEXT:    s_mov_b32 s4, 0x7060302
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
-; GFX9-NEXT:    v_lshrrev_b32_e32 v1, 16, v0
-; GFX9-NEXT:    v_and_b32_e32 v2, 0xffff, v1
-; GFX9-NEXT:    v_lshl_or_b32 v1, v1, 16, v2
+; GFX9-NEXT:    v_perm_b32 v1, v0, v0, s4
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX10-LABEL: shuffle_v4f16_6667:
@@ -1475,9 +1363,7 @@
 ; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-NEXT:    global_load_dword v0, v[0:1], off offset:4
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
-; GFX10-NEXT:    v_lshrrev_b32_e32 v1, 16, v0
-; GFX10-NEXT:    v_and_b32_e32 v2, 0xffff, v1
-; GFX10-NEXT:    v_lshl_or_b32 v1, v1, 16, v2
+; GFX10-NEXT:    v_perm_b32 v1, v0, v0, 0x7060302
 ; GFX10-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX11-LABEL: shuffle_v4f16_6667:
@@ -1486,10 +1372,7 @@
 ; GFX11-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-NEXT:    global_load_b32 v0, v[0:1], off offset:4
 ; GFX11-NEXT:    s_waitcnt vmcnt(0)
-; GFX11-NEXT:    v_lshrrev_b32_e32 v1, 16, v0
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX11-NEXT:    v_and_b32_e32 v2, 0xffff, v1
-; GFX11-NEXT:    v_lshl_or_b32 v1, v1, 16, v2
+; GFX11-NEXT:    v_perm_b32 v1, v0, v0, 0x7060302
 ; GFX11-NEXT:    s_setpc_b64 s[30:31]
   %val0 = load <4 x half>, <4 x half> addrspace(1)* %arg0
   %val1 = load <4 x half>, <4 x half> addrspace(1)* %arg1
@@ -1642,10 +1525,8 @@
 ; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX9-NEXT:    global_load_dwordx2 v[5:6], v[2:3], off offset:8
 ; GFX9-NEXT:    global_load_dword v4, v[0:1], off offset:4
-; GFX9-NEXT:    v_mov_b32_e32 v0, 0xffff
 ; GFX9-NEXT:    s_waitcnt vmcnt(1)
-; GFX9-NEXT:    v_and_b32_sdwa v0, v0, v5 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
-; GFX9-NEXT:    v_lshl_or_b32 v0, v6, 16, v0
+; GFX9-NEXT:    v_alignbit_b32 v0, v6, v5, 16
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    v_mov_b32_e32 v1, v4
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
@@ -1656,12 +1537,10 @@
 ; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-NEXT:    global_load_dwordx2 v[5:6], v[2:3], off offset:8
 ; GFX10-NEXT:    global_load_dword v4, v[0:1], off offset:4
-; GFX10-NEXT:    v_mov_b32_e32 v0, 0xffff
 ; GFX10-NEXT:    s_waitcnt vmcnt(1)
-; GFX10-NEXT:    v_and_b32_sdwa v0, v0, v5 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
+; GFX10-NEXT:    v_alignbit_b32 v0, v6, v5, 16
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    v_mov_b32_e32 v1, v4
-; GFX10-NEXT:    v_lshl_or_b32 v0, v6, 16, v0
 ; GFX10-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX11-LABEL: shuffle_v8f16_13_14_2_3:
@@ -1671,10 +1550,7 @@
 ; GFX11-NEXT:    global_load_b64 v[2:3], v[2:3], off offset:8
 ; GFX11-NEXT:    global_load_b32 v1, v[0:1], off offset:4
 ; GFX11-NEXT:    s_waitcnt vmcnt(1)
-; GFX11-NEXT:    v_lshrrev_b32_e32 v0, 16, v2
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX11-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX11-NEXT:    v_lshl_or_b32 v0, v3, 16, v0
+; GFX11-NEXT:    v_alignbit_b32 v0, v3, v2, 16
 ; GFX11-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-NEXT:    s_setpc_b64 s[30:31]
   %val0 = load <8 x half>, <8 x half> addrspace(1)* %arg0
@@ -1688,9 +1564,9 @@
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX9-NEXT:    global_load_dwordx2 v[0:1], v[0:1], off
+; GFX9-NEXT:    s_mov_b32 s4, 0x5040100
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
-; GFX9-NEXT:    v_and_b32_e32 v2, 0xffff, v1
-; GFX9-NEXT:    v_lshl_or_b32 v1, v1, 16, v2
+; GFX9-NEXT:    v_perm_b32 v1, v1, v1, s4
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX10-LABEL: shuffle_v3f16_0122:
@@ -1699,8 +1575,7 @@
 ; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-NEXT:    global_load_dwordx2 v[0:1], v[0:1], off
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
-; GFX10-NEXT:    v_and_b32_e32 v2, 0xffff, v1
-; GFX10-NEXT:    v_lshl_or_b32 v1, v1, 16, v2
+; GFX10-NEXT:    v_perm_b32 v1, v1, v1, 0x5040100
 ; GFX10-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX11-LABEL: shuffle_v3f16_0122:
@@ -1709,9 +1584,7 @@
 ; GFX11-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-NEXT:    global_load_b64 v[0:1], v[0:1], off
 ; GFX11-NEXT:    s_waitcnt vmcnt(0)
-; GFX11-NEXT:    v_and_b32_e32 v2, 0xffff, v1
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
-; GFX11-NEXT:    v_lshl_or_b32 v1, v1, 16, v2
+; GFX11-NEXT:    v_perm_b32 v1, v1, v1, 0x5040100
 ; GFX11-NEXT:    s_setpc_b64 s[30:31]
   %val0 = load <3 x half>, <3 x half> addrspace(1)* %arg0
   %val1 = load <3 x half>, <3 x half> addrspace(1)* %arg1
@@ -1724,10 +1597,8 @@
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX9-NEXT:    global_load_dword v0, v[0:1], off
-; GFX9-NEXT:    v_mov_b32_e32 v1, 0xffff
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
-; GFX9-NEXT:    v_and_b32_sdwa v1, v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
-; GFX9-NEXT:    v_lshl_or_b32 v1, v0, 16, v1
+; GFX9-NEXT:    v_alignbit_b32 v1, v0, v0, 16
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX10-LABEL: shuffle_v2f16_0122:
@@ -1735,10 +1606,8 @@
 ; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-NEXT:    global_load_dword v0, v[0:1], off
-; GFX10-NEXT:    v_mov_b32_e32 v1, 0xffff
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
-; GFX10-NEXT:    v_and_b32_sdwa v1, v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
-; GFX10-NEXT:    v_lshl_or_b32 v1, v0, 16, v1
+; GFX10-NEXT:    v_alignbit_b32 v1, v0, v0, 16
 ; GFX10-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX11-LABEL: shuffle_v2f16_0122:
@@ -1747,10 +1616,7 @@
 ; GFX11-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-NEXT:    global_load_b32 v0, v[0:1], off
 ; GFX11-NEXT:    s_waitcnt vmcnt(0)
-; GFX11-NEXT:    v_lshrrev_b32_e32 v1, 16, v0
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX11-NEXT:    v_and_b32_e32 v1, 0xffff, v1
-; GFX11-NEXT:    v_lshl_or_b32 v1, v0, 16, v1
+; GFX11-NEXT:    v_alignbit_b32 v1, v0, v0, 16
 ; GFX11-NEXT:    s_setpc_b64 s[30:31]
   %val0 = load <2 x half>, <2 x half> addrspace(1)* %arg0
   %val1 = load <2 x half>, <2 x half> addrspace(1)* %arg1
@@ -1907,14 +1773,12 @@
 ; GFX9-NEXT:    global_load_dwordx2 v[4:5], v[0:1], off
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    global_load_dwordx2 v[5:6], v[2:3], off
+; GFX9-NEXT:    s_mov_b32 s4, 0x5040100
 ; GFX9-NEXT:    ; kill: killed $vgpr0 killed $vgpr1
-; GFX9-NEXT:    v_mov_b32_e32 v0, 0xffff
 ; GFX9-NEXT:    ; kill: killed $vgpr2 killed $vgpr3
-; GFX9-NEXT:    v_and_b32_e32 v1, 0xffff, v4
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
-; GFX9-NEXT:    v_and_b32_sdwa v2, v0, v5 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
-; GFX9-NEXT:    v_lshl_or_b32 v0, v5, 16, v1
-; GFX9-NEXT:    v_lshl_or_b32 v1, v6, 16, v2
+; GFX9-NEXT:    v_perm_b32 v0, v5, v4, s4
+; GFX9-NEXT:    v_alignbit_b32 v1, v6, v5, 16
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX10-LABEL: shuffle_v4f16_0456:
@@ -1925,29 +1789,22 @@
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    global_load_dwordx2 v[5:6], v[2:3], off
 ; GFX10-NEXT:    ; kill: killed $vgpr0 killed $vgpr1
-; GFX10-NEXT:    v_mov_b32_e32 v0, 0xffff
 ; GFX10-NEXT:    ; kill: killed $vgpr2 killed $vgpr3
-; GFX10-NEXT:    v_and_b32_e32 v1, 0xffff, v4
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
-; GFX10-NEXT:    v_and_b32_sdwa v2, v0, v5 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
-; GFX10-NEXT:    v_lshl_or_b32 v0, v5, 16, v1
-; GFX10-NEXT:    v_lshl_or_b32 v1, v6, 16, v2
+; GFX10-NEXT:    v_perm_b32 v0, v5, v4, 0x5040100
+; GFX10-NEXT:    v_alignbit_b32 v1, v6, v5, 16
 ; GFX10-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX11-LABEL: shuffle_v4f16_0456:
 ; GFX11:       ; %bb.0:
 ; GFX11-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX11-NEXT:    s_waitcnt_vscnt null, 0x0
-; GFX11-NEXT:    global_load_b64 v[2:3], v[2:3], off
 ; GFX11-NEXT:    global_load_b64 v[0:1], v[0:1], off
 ; GFX11-NEXT:    s_waitcnt vmcnt(0)
-; GFX11-NEXT:    v_lshrrev_b32_e32 v1, 16, v2
-; GFX11-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX11-NEXT:    v_and_b32_e32 v1, 0xffff, v1
-; GFX11-NEXT:    v_lshl_or_b32 v0, v2, 16, v0
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_2)
-; GFX11-NEXT:    v_lshl_or_b32 v1, v3, 16, v1
+; GFX11-NEXT:    global_load_b64 v[1:2], v[2:3], off
+; GFX11-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-NEXT:    v_perm_b32 v0, v1, v0, 0x5040100
+; GFX11-NEXT:    v_alignbit_b32 v1, v2, v1, 16
 ; GFX11-NEXT:    s_setpc_b64 s[30:31]
   %val0 = load <4 x half>, <4 x half> addrspace(1)* %arg0
   %val1 = load <4 x half>, <4 x half> addrspace(1)* %arg1
@@ -2006,3 +1863,304 @@
 declare i32 @llvm.amdgcn.workitem.id.x() #0
 
 attributes #0 = { nounwind readnone speculatable }
+define <2 x half> @low16bits(<2 x half> addrspace(1)* %x0, <2 x half> addrspace(1)* %x1) {
+; GFX9-LABEL: low16bits:
+; GFX9:       ; %bb.0: ; %entry
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    global_load_dword v4, v[0:1], off
+; GFX9-NEXT:    global_load_dword v5, v[2:3], off
+; GFX9-NEXT:    s_mov_b32 s4, 0x5040100
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    v_perm_b32 v0, v5, v4, s4
+; GFX9-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX10-LABEL: low16bits:
+; GFX10:       ; %bb.0: ; %entry
+; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    global_load_dword v4, v[0:1], off
+; GFX10-NEXT:    global_load_dword v5, v[2:3], off
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    v_perm_b32 v0, v5, v4, 0x5040100
+; GFX10-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX11-LABEL: low16bits:
+; GFX11:       ; %bb.0: ; %entry
+; GFX11-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX11-NEXT:    global_load_b32 v0, v[0:1], off
+; GFX11-NEXT:    global_load_b32 v1, v[2:3], off
+; GFX11-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-NEXT:    v_perm_b32 v0, v1, v0, 0x5040100
+; GFX11-NEXT:    s_setpc_b64 s[30:31]
+entry:
+  %0 = load <2 x half>, <2 x half> addrspace(1)* %x0, align 4
+  %1 = load <2 x half>, <2 x half> addrspace(1)* %x1, align 4
+  %vy1.0.vec.insert = shufflevector <2 x half> %0, <2 x half> poison, <2 x i32> <i32 0, i32 undef>
+  %vy1.2.vec.insert = shufflevector <2 x half> %vy1.0.vec.insert, <2 x half> %1, <2 x i32> <i32 0, i32 2>
+  ret <2 x half> %vy1.2.vec.insert
+}
+
+define <2 x half> @hi16bits(<2 x half> addrspace(1)* %x0, <2 x half> addrspace(1)* %x1) {
+; GFX9-LABEL: hi16bits:
+; GFX9:       ; %bb.0: ; %entry
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    global_load_dword v4, v[0:1], off
+; GFX9-NEXT:    global_load_dword v5, v[2:3], off
+; GFX9-NEXT:    s_mov_b32 s4, 0x7060302
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    v_perm_b32 v0, v5, v4, s4
+; GFX9-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX10-LABEL: hi16bits:
+; GFX10:       ; %bb.0: ; %entry
+; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    global_load_dword v4, v[0:1], off
+; GFX10-NEXT:    global_load_dword v5, v[2:3], off
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    v_perm_b32 v0, v5, v4, 0x7060302
+; GFX10-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX11-LABEL: hi16bits:
+; GFX11:       ; %bb.0: ; %entry
+; GFX11-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX11-NEXT:    global_load_b32 v0, v[0:1], off
+; GFX11-NEXT:    global_load_b32 v1, v[2:3], off
+; GFX11-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-NEXT:    v_perm_b32 v0, v1, v0, 0x7060302
+; GFX11-NEXT:    s_setpc_b64 s[30:31]
+entry:
+  %0 = load <2 x half>, <2 x half> addrspace(1)* %x0, align 4
+  %1 = load <2 x half>, <2 x half> addrspace(1)* %x1, align 4
+  %vy1.0.vec.insert = shufflevector <2 x half> %0, <2 x half> poison, <2 x i32> <i32 1, i32 undef>
+  %vy1.2.vec.insert = shufflevector <2 x half> %vy1.0.vec.insert, <2 x half> %1, <2 x i32> <i32 0, i32 3>
+  ret <2 x half> %vy1.2.vec.insert
+}
+
+define <2 x half> @low16hi16bits(<2 x half> addrspace(1)* %x0, <2 x half> addrspace(1)* %x1) {
+; GFX9-LABEL: low16hi16bits:
+; GFX9:       ; %bb.0: ; %entry
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    global_load_dword v4, v[0:1], off
+; GFX9-NEXT:    global_load_dword v5, v[2:3], off
+; GFX9-NEXT:    s_mov_b32 s4, 0x7060100
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    v_perm_b32 v0, v5, v4, s4
+; GFX9-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX10-LABEL: low16hi16bits:
+; GFX10:       ; %bb.0: ; %entry
+; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    global_load_dword v4, v[0:1], off
+; GFX10-NEXT:    global_load_dword v5, v[2:3], off
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    v_perm_b32 v0, v5, v4, 0x7060100
+; GFX10-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX11-LABEL: low16hi16bits:
+; GFX11:       ; %bb.0: ; %entry
+; GFX11-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX11-NEXT:    global_load_b32 v0, v[0:1], off
+; GFX11-NEXT:    global_load_b32 v1, v[2:3], off
+; GFX11-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-NEXT:    v_perm_b32 v0, v1, v0, 0x7060100
+; GFX11-NEXT:    s_setpc_b64 s[30:31]
+entry:
+  %0 = load <2 x half>, <2 x half> addrspace(1)* %x0, align 4
+  %1 = load <2 x half>, <2 x half> addrspace(1)* %x1, align 4
+  %vy1.0.vec.insert = shufflevector <2 x half> %0, <2 x half> poison, <2 x i32> <i32 0, i32 undef>
+  %vy1.2.vec.insert = shufflevector <2 x half> %vy1.0.vec.insert, <2 x half> %1, <2 x i32> <i32 0, i32 3>
+  ret <2 x half> %vy1.2.vec.insert
+}
+
+define <2 x half> @hi16low16bits(<2 x half> addrspace(1)* %x0, <2 x half> addrspace(1)* %x1) {
+; GFX9-LABEL: hi16low16bits:
+; GFX9:       ; %bb.0: ; %entry
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    global_load_dword v4, v[0:1], off
+; GFX9-NEXT:    global_load_dword v5, v[2:3], off
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    v_alignbit_b32 v0, v5, v4, 16
+; GFX9-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX10-LABEL: hi16low16bits:
+; GFX10:       ; %bb.0: ; %entry
+; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    global_load_dword v4, v[0:1], off
+; GFX10-NEXT:    global_load_dword v5, v[2:3], off
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    v_alignbit_b32 v0, v5, v4, 16
+; GFX10-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX11-LABEL: hi16low16bits:
+; GFX11:       ; %bb.0: ; %entry
+; GFX11-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX11-NEXT:    global_load_b32 v0, v[0:1], off
+; GFX11-NEXT:    global_load_b32 v1, v[2:3], off
+; GFX11-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-NEXT:    v_alignbit_b32 v0, v1, v0, 16
+; GFX11-NEXT:    s_setpc_b64 s[30:31]
+entry:
+  %0 = load <2 x half>, <2 x half> addrspace(1)* %x0, align 4
+  %1 = load <2 x half>, <2 x half> addrspace(1)* %x1, align 4
+  %vy1.0.vec.insert = shufflevector <2 x half> %0, <2 x half> poison, <2 x i32> <i32 1, i32 undef>
+  %vy1.2.vec.insert = shufflevector <2 x half> %vy1.0.vec.insert, <2 x half> %1, <2 x i32> <i32 0, i32 2>
+  ret <2 x half> %vy1.2.vec.insert
+}
+
+define <2 x i16> @i16_low16bits(<2 x i16> addrspace(1)* %x0, <2 x i16> addrspace(1)* %x1) {
+; GFX9-LABEL: i16_low16bits:
+; GFX9:       ; %bb.0: ; %entry
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    global_load_dword v4, v[0:1], off
+; GFX9-NEXT:    global_load_dword v5, v[2:3], off
+; GFX9-NEXT:    s_mov_b32 s4, 0x5040100
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    v_perm_b32 v0, v5, v4, s4
+; GFX9-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX10-LABEL: i16_low16bits:
+; GFX10:       ; %bb.0: ; %entry
+; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    global_load_dword v4, v[0:1], off
+; GFX10-NEXT:    global_load_dword v5, v[2:3], off
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    v_perm_b32 v0, v5, v4, 0x5040100
+; GFX10-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX11-LABEL: i16_low16bits:
+; GFX11:       ; %bb.0: ; %entry
+; GFX11-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX11-NEXT:    global_load_b32 v0, v[0:1], off
+; GFX11-NEXT:    global_load_b32 v1, v[2:3], off
+; GFX11-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-NEXT:    v_perm_b32 v0, v1, v0, 0x5040100
+; GFX11-NEXT:    s_setpc_b64 s[30:31]
+entry:
+  %0 = load <2 x i16>, <2 x i16> addrspace(1)* %x0, align 4
+  %1 = load <2 x i16>, <2 x i16> addrspace(1)* %x1, align 4
+  %vy1.0.vec.insert = shufflevector <2 x i16> %0, <2 x i16> poison, <2 x i32> <i32 0, i32 undef>
+  %vy1.2.vec.insert = shufflevector <2 x i16> %vy1.0.vec.insert, <2 x i16> %1, <2 x i32> <i32 0, i32 2>
+  ret <2 x i16> %vy1.2.vec.insert
+}
+
+define <2 x i16> @i16_low16hi16bits(<2 x i16> addrspace(1)* %x0, <2 x i16> addrspace(1)* %x1) {
+; GFX9-LABEL: i16_low16hi16bits:
+; GFX9:       ; %bb.0: ; %entry
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    global_load_dword v4, v[0:1], off
+; GFX9-NEXT:    global_load_dword v5, v[2:3], off
+; GFX9-NEXT:    s_mov_b32 s4, 0x7060100
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    v_perm_b32 v0, v5, v4, s4
+; GFX9-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX10-LABEL: i16_low16hi16bits:
+; GFX10:       ; %bb.0: ; %entry
+; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    global_load_dword v4, v[0:1], off
+; GFX10-NEXT:    global_load_dword v5, v[2:3], off
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    v_perm_b32 v0, v5, v4, 0x7060100
+; GFX10-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX11-LABEL: i16_low16hi16bits:
+; GFX11:       ; %bb.0: ; %entry
+; GFX11-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX11-NEXT:    global_load_b32 v0, v[0:1], off
+; GFX11-NEXT:    global_load_b32 v1, v[2:3], off
+; GFX11-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-NEXT:    v_perm_b32 v0, v1, v0, 0x7060100
+; GFX11-NEXT:    s_setpc_b64 s[30:31]
+entry:
+  %0 = load <2 x i16>, <2 x i16> addrspace(1)* %x0, align 4
+  %1 = load <2 x i16>, <2 x i16> addrspace(1)* %x1, align 4
+  %vy1.0.vec.insert = shufflevector <2 x i16> %0, <2 x i16> poison, <2 x i32> <i32 0, i32 undef>
+  %vy1.2.vec.insert = shufflevector <2 x i16> %vy1.0.vec.insert, <2 x i16> %1, <2 x i32> <i32 0, i32 3>
+  ret <2 x i16> %vy1.2.vec.insert
+}
+
+define <2 x i16> @i16_hi16low16bits(<2 x i16> addrspace(1)* %x0, <2 x i16> addrspace(1)* %x1) {
+; GFX9-LABEL: i16_hi16low16bits:
+; GFX9:       ; %bb.0: ; %entry
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    global_load_dword v4, v[0:1], off
+; GFX9-NEXT:    global_load_dword v5, v[2:3], off
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    v_alignbit_b32 v0, v5, v4, 16
+; GFX9-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX10-LABEL: i16_hi16low16bits:
+; GFX10:       ; %bb.0: ; %entry
+; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    global_load_dword v4, v[0:1], off
+; GFX10-NEXT:    global_load_dword v5, v[2:3], off
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    v_alignbit_b32 v0, v5, v4, 16
+; GFX10-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX11-LABEL: i16_hi16low16bits:
+; GFX11:       ; %bb.0: ; %entry
+; GFX11-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX11-NEXT:    global_load_b32 v0, v[0:1], off
+; GFX11-NEXT:    global_load_b32 v1, v[2:3], off
+; GFX11-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-NEXT:    v_alignbit_b32 v0, v1, v0, 16
+; GFX11-NEXT:    s_setpc_b64 s[30:31]
+entry:
+  %0 = load <2 x i16>, <2 x i16> addrspace(1)* %x0, align 4
+  %1 = load <2 x i16>, <2 x i16> addrspace(1)* %x1, align 4
+  %vy1.0.vec.insert = shufflevector <2 x i16> %0, <2 x i16> poison, <2 x i32> <i32 1, i32 undef>
+  %vy1.2.vec.insert = shufflevector <2 x i16> %vy1.0.vec.insert, <2 x i16> %1, <2 x i32> <i32 0, i32 2>
+  ret <2 x i16> %vy1.2.vec.insert
+}
+
+define <2 x i16> @i16_hi16bits(<2 x i16> addrspace(1)* %x0, <2 x i16> addrspace(1)* %x1) {
+; GFX9-LABEL: i16_hi16bits:
+; GFX9:       ; %bb.0: ; %entry
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    global_load_dword v4, v[0:1], off
+; GFX9-NEXT:    global_load_dword v5, v[2:3], off
+; GFX9-NEXT:    s_mov_b32 s4, 0x7060302
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    v_perm_b32 v0, v5, v4, s4
+; GFX9-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX10-LABEL: i16_hi16bits:
+; GFX10:       ; %bb.0: ; %entry
+; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    global_load_dword v4, v[0:1], off
+; GFX10-NEXT:    global_load_dword v5, v[2:3], off
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    v_perm_b32 v0, v5, v4, 0x7060302
+; GFX10-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX11-LABEL: i16_hi16bits:
+; GFX11:       ; %bb.0: ; %entry
+; GFX11-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX11-NEXT:    global_load_b32 v0, v[0:1], off
+; GFX11-NEXT:    global_load_b32 v1, v[2:3], off
+; GFX11-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-NEXT:    v_perm_b32 v0, v1, v0, 0x7060302
+; GFX11-NEXT:    s_setpc_b64 s[30:31]
+entry:
+  %0 = load <2 x i16>, <2 x i16> addrspace(1)* %x0, align 4
+  %1 = load <2 x i16>, <2 x i16> addrspace(1)* %x1, align 4
+  %vy1.0.vec.insert = shufflevector <2 x i16> %0, <2 x i16> poison, <2 x i32> <i32 1, i32 undef>
+  %vy1.2.vec.insert = shufflevector <2 x i16> %vy1.0.vec.insert, <2 x i16> %1, <2 x i32> <i32 0, i32 3>
+  ret <2 x i16> %vy1.2.vec.insert
+}