[AMDGPU] Use V_PERM to match buildvectors when inputs are not canonicalized (i.e. can't use V_PACK)

[jrbyrnes]

Summary

If we can not prove that f16 operands of a buildvector are canonicalized, then we can not lower into a V_PACK. In this scenario, we would previously lower into some combination of and(sdwa), shr, or. This patch allows for matching into V_PERM instead -- which uses additional SGPR (or encodes the literal in the instruction itself), but has less VALU latency.

Change-Id: Ifa4a74fdb81ef44f22ba490c7fdf81ec8aebc945

[arsenm]

llvm/lib/Target/AMDGPU/SIInstructions.td

2810–2812

Should we just replace this?

2817

Also should cover the integer cases?

llvm/test/CodeGen/AMDGPU/pack.v2f16.ll

2–1

Switching to generated checks should be a separate pre-commit

[jrbyrnes]

Hey Matt, thanks for the comments. I'll address them soon, for now I'll add you as reviewer.

[rampitec]

llvm/test/CodeGen/AMDGPU/pack.v2f16.ll

2–1

GCN is misleading here. Use something like GFX7_8

[jrbyrnes]

Precommit generated tests pack.v2f16.ll, rebase

[jrbyrnes]

Fix attributes in test

[arsenm]

llvm/lib/Target/AMDGPU/SIInstructions.td

2810–2812

This is looking dead

llvm/test/CodeGen/AMDGPU/v_perm_non_canon.ll

3 ↗

(On Diff #462545)

Probably should move this in with vector_shuffle.packed.ll

4 ↗

(On Diff #462545)

Should use addrspace(1)*

36 ↗

(On Diff #462545)

Separate functions for each tested shuffle.

Also needs versions using i16.

[jrbyrnes]

Address review comments.

Added some patterns to account for the case when we are trying to concat:
v0[0]:v1[1]
v0[1]:v1[0]

Removed some seemingly dead patterns after introducing those.

Pushing as is for potential feedback, still a sort of WIP.

[arsenm]

llvm/lib/Target/AMDGPU/SIInstructions.td

2785–2794

Can use a class or foreach over the types to avoid repeating the same pattern twice

[foad]

Can't you use v_alignbit for all the cases where you need the upper 16 bits of one register and the lower 16 bits of the other? It should be smaller than v_perm because the shift amount (16) is an inline constant.

[jrbyrnes]

Precommit generated test + Rebase

Consolidate patterns into foreach

Lower to V_ALIGNBIT for D = V[1].low : V[0].hi

[jrbyrnes]

In D134463#3817469, @foad wrote:

Can't you use v_alignbit for all the cases where you need the upper 16 bits of one register and the lower 16 bits of the other? It should be smaller than v_perm because the shift amount (16) is an inline constant.

Hey, thanks for the good suggestion! I think this will only work for the case where we want V[1].low : V[0].hi

In the case where we want V[1].hi : V[0].low we can't lower to V_ALIGNBIT_B32 $V0, $V1, 16 because that would incorrectly put the bits from $V0 as the MSBs in the dest. On the other hand V_ALIGNBIT_B32 $V1, $V0, 16 correctly has the bits from $V1 as the MSBs, but they are the lower 16 (and the higher 16 from $V0).

[foad]

In the case where we want V[1].hi : V[0].low we can't lower to V_ALIGNBIT_B32 $V0, $V1, 16 because that would incorrectly put the bits from $V0 as the MSBs in the dest. On the other hand V_ALIGNBIT_B32 $V1, $V0, 16 correctly has the bits from $V1 as the MSBs, but they are the lower 16 (and the higher 16 from $V0).

Good point. You could use V_BFI_B32 but I guess that is no better or worse than V_PERM_B32.

[arsenm]

In D134463#3820197, @foad wrote:

In the case where we want V[1].hi : V[0].low we can't lower to V_ALIGNBIT_B32 $V0, $V1, 16 because that would incorrectly put the bits from $V0 as the MSBs in the dest. On the other hand V_ALIGNBIT_B32 $V1, $V0, 16 correctly has the bits from $V1 as the MSBs, but they are the lower 16 (and the higher 16 from $V0).

Good point. You could use V_BFI_B32 but I guess that is no better or worse than V_PERM_B32.

One small point in favor of BFI is the bitmask you need is more likely CSEable for unrelated uses

[jrbyrnes]

Use V_BFI for V[1].hi : V[0].low . This allows for a bitmask which is more likely to be reused by other instructions (0xffff vs 0x7060100), potentially enabling other optimizations (e.g. CSE)

[jrbyrnes]

One small point in favor of BFI is the bitmask you need is more likely CSEable for unrelated uses

Thanks, good point. Changed the pattern in favor of BFI.

[jrbyrnes]

llvm/test/CodeGen/AMDGPU/fast-unaligned-load-store.private.ll

241

This seems illegal to me -- using SGPR and literal as operands to VALU. Looking into it.

[rampitec]

llvm/test/CodeGen/AMDGPU/fast-unaligned-load-store.private.ll

241

0 is inline literal and is free.

[foad]

llvm/test/CodeGen/AMDGPU/fast-unaligned-load-store.private.ll

241

As a code quality thing, this could have been optimized to v_and_b32 v1, 0xffff0000, v0

[jrbyrnes]

llvm/test/CodeGen/AMDGPU/fast-unaligned-load-store.private.ll

241

Stas -- I see, thanks!

Jay -- Interesting, I'll look into what's going on with the literal. As a side note, CodeGen is actually not good for this particular test. It seems to me the whole test can be combined into a 32 bit load. D133584 should be extended to handle this i16s, in which case this whole test will be optimized to a load.

[arsenm]

llvm/test/CodeGen/AMDGPU/fast-unaligned-load-store.private.ll

241

This could only be a 16-bit load if unaligned access is enabled (and I think we previously decided that doing unaligned 16-bit loads was probably worse than byte loads). The load question is orthogonal to how the bit masking should have been emitted

[jrbyrnes]

Add pattern to select V_AND v1, 0xffff000 in the case where buildvector produces bits V1.hi : 0

Add test coverage for pattern.

llvm/test/CodeGen/AMDGPU/fast-unaligned-load-store.private.ll

241

Right -- good to know about decision to use byte loads. I agree it is a bit off topic for this review.

[arsenm]

LGTM

[jrbyrnes]

Rebased to trunk && made necessary test modifications. NFC

[jrbyrnes]

Landed via rGf4e6149d8217

llvm/lib/Target/AMDGPU/SIInstructions.td

(fma (f64 (VOP3NoMods f64:$src0)),
        (f64 (VOP3NoMods f64:$src1)),
        (f64 (VOP3NoMods f64:$src2))),
   (V_FMAC_F64_e64 SRCMODS.NONE, $src0, SRCMODS.NONE, $src1,
                   SRCMODS.NONE, $src2)
 >;
 
 // 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))
 >;
 
 def : GCNPat <
   (v2i16 (DivergentBinFrag<build_vector> (i16 0), (i16 SReg_32:$src1))),
   (v2i16 (V_LSHLREV_B32_e64 (i16 16), SReg_32:$src1))
def : GCNPat <
   (v2f16 (UniformBinFrag<build_vector> (f16 undef), (f16 SReg_32:$src1))),
   (S_LSHL_B32 SReg_32:$src1, (i32 16))
 >;
 
 def : GCNPat <
   (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 <
   (v2i16 (UniformBinFrag<build_vector> (i16 SReg_32:$src0), (i16 SReg_32:$src1))),
   (S_PACK_LL_B32_B16 SReg_32:$src0, SReg_32:$src1)
 >;
 
 def : GCNPat <
   (v2i16 (DivergentBinFrag<build_vector> (i16 SReg_32:$src0), (i16 SReg_32:$src1))),
   (v2i16 (V_LSHL_OR_B32_e64 $src1, (i32 16), (i32 (V_AND_B32_e64 (i32 (V_MOV_B32_e32 (i32 0xffff))), $src0))))
 >;
 
 // With multiple uses of the shift, this will duplicate the shift and
 // increase register pressure.
 def : GCNPat <
   (v2i16 (UniformBinFrag<build_vector> (i16 SReg_32:$src0), (i16 (trunc (srl_oneuse SReg_32:$src1, (i32 16)))))),
   (v2i16 (S_PACK_LH_B32_B16 SReg_32:$src0, SReg_32:$src1))
 >;
 
 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)
+>;
 
+// 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)
+  (v2f16 (DivergentBinFrag<build_vector> (f16 VGPR_32:$a), (f16 VGPR_32:$b))),
+  (V_PERM_B32_e64 VGPR_32:$b, VGPR_32:$a, (S_MOV_B32 (i32 0x05040100)))
 >;
 
+// Take the lower 16 bits from each VGPR_32 and concat them
 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))))
+  (v2i16 (DivergentBinFrag<build_vector> (i16 VGPR_32:$a), (i16 VGPR_32:$b))),
+  (V_PERM_B32_e64 VGPR_32:$b, VGPR_32:$a, (S_MOV_B32 (i32 0x05040100)))
 >;
 
+// Take the upper 16 bits from each VGPR_32 and concat them
 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)
+  (v2f16 (DivergentBinFrag<build_vector> (f16 (bitconvert (i16 (trunc (srl VGPR_32:$a, (i32 16)))))), (f16 (bitconvert (i16 (trunc (srl VGPR_32:$b, (i32 16)))))))),
+  (V_PERM_B32_e64 VGPR_32:$b, VGPR_32:$a, (S_MOV_B32 (i32 0x07060302)))
 >;
 
-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))))
+// Take the upper 16 bits from V[0] and the lower 16 bits from V[1]
+def : GCNPat < 
+  (v2f16 (DivergentBinFrag<build_vector> (f16 VGPR_32:$a), (f16 (bitconvert (i16 (trunc (srl VGPR_32:$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]
+def : GCNPat <
+  (v2i16 (DivergentBinFrag<build_vector> (i16 VGPR_32:$a), (i16 (trunc (srl VGPR_32:$b, (i32 16)))))),
+  (V_PERM_B32_e64 VGPR_32:$b, VGPR_32:$a, (S_MOV_B32 (i32 0x07060100)))
+>;
 
+// Take the lower 16 bits from V[0] and the upper 16 bits from V[1]
+def : GCNPat <
+  (v2f16 (DivergentBinFrag<build_vector> (f16 (bitconvert (i16 (trunc (srl VGPR_32:$a, (i32 16)))))), (f16 VGPR_32:$b))),
+  (V_PERM_B32_e64 VGPR_32:$b, VGPR_32:$a, (S_MOV_B32 (i32 0x05040302)))
+>;
+
+// 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 VGPR_32:$a, (i32 16)))), (i16 VGPR_32:$b))),
+  (V_PERM_B32_e64 VGPR_32:$b, VGPR_32:$a, (S_MOV_B32 (i32 0x05040302)))
+>;
+
+// Take the upper 16 bits from each VGPR_32 and concat them
+def : GCNPat <
+  (v2i16 (DivergentBinFrag<build_vector> (i16 (trunc (srl VGPR_32:$a, (i32 16)))), (i16 (trunc (srl VGPR_32:$b, (i32 16)))))),
+  (V_PERM_B32_e64 VGPR_32:$b, VGPR_32:$a, (S_MOV_B32 (i32 0x07060302)))
+>;
+
+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
 // increase register pressure.
 let SubtargetPredicate = isGFX11Plus in
 def : GCNPat <
   (v2i16 (build_vector (i16 (trunc (srl_oneuse SReg_32:$src0, (i32 16)))), (i16 SReg_32:$src1))),
   (v2i16 (S_PACK_HL_B32_B16 SReg_32:$src0, SReg_32:$src1))

llvm/test/CodeGen/AMDGPU/add.v2i16.ll

 ; VI-NEXT:    buffer_store_dwordx4 v[0:3], off, s[4:7], 0
 ; VI-NEXT:    s_endpgm
 ;
 ; GFX9-LABEL: v_test_add_v2i16_zext_to_v2i64:
 ; GFX9:       ; %bb.0:
 ; 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
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    global_load_dword v3, v0, s[2:3] glc
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    v_mov_b32_e32 v0, 0x5040302
 ; 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 v3, v2, v3
+; GFX9-NEXT:    v_perm_b32 v2, 0, v3, v0
+; GFX9-NEXT:    v_and_b32_e32 v0, 0xffff, v3
 ; 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
 ;
 ; GFX10-LABEL: v_test_add_v2i16_zext_to_v2i64:
 ; GFX10:       ; %bb.0:
 ; GFX10-NEXT:    s_clause 0x1
 ; 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)
 ; GFX10-NEXT:    global_load_dword v2, v0, s[2:3] glc dlc
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; 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_perm_b32 v2, 0, v0, 0x5040302
+; 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
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %gep.out = getelementptr inbounds <2 x i64>, <2 x i64> addrspace(1)* %out, i32 %tid
   %gep.in0 = getelementptr inbounds <2 x i16>, <2 x i16> addrspace(1)* %in0, i32 %tid
   %gep.in1 = getelementptr inbounds <2 x i16>, <2 x i16> addrspace(1)* %in1, i32 %tid
   %a = load volatile <2 x i16>, <2 x i16> addrspace(1)* %gep.in0

llvm/test/CodeGen/AMDGPU/build-vector-packed-partial-undef.ll

; GFX8-NEXT:    s_setpc_b64 s[30:31]
   call void asm sideeffect "; use $0", "v"(<4 x half> %undef.lo);
   ret void
 }
 
 define void @undef_lo2_v4i16(<2 x i16> %arg0) {
 ; 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
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX8-LABEL: undef_lo2_v4i16:
 ; GFX8:       ; %bb.0:
 ; GFX8-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX8-NEXT:    v_lshrrev_b32_e32 v1, 16, v0
 ; GFX8-NEXT:    v_alignbit_b32 v0, v1, v0, 16
 ; GFX8-NEXT:    ;;#ASMSTART
 ; GFX8-NEXT:    ; use v[0:1]
 ; GFX8-NEXT:    ;;#ASMEND
 ; GFX8-NEXT:    s_setpc_b64 s[30:31]
   %undef.lo = shufflevector <2 x i16> %arg0, <2 x i16> undef, <4 x i32> <i32 1, i32 1, i32 2, i32 3>
   call void asm sideeffect "; use $0", "v"(<4 x i16> %undef.lo);
   ret void
 }
 
 define void @undef_lo2_v4f16(<2 x half> %arg0) {
 ; 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
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX8-LABEL: undef_lo2_v4f16:
 ; GFX8:       ; %bb.0:
 ; GFX8-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)

llvm/test/CodeGen/AMDGPU/chain-hi-to-lo.ll

 ; GFX900-NEXT:    s_waitcnt vmcnt(0)
 ; GFX900-NEXT:    buffer_store_short v0, off, s[0:3], 0 offset:4
 ; GFX900-NEXT:    s_waitcnt vmcnt(0)
 ; GFX900-NEXT:    global_load_ushort v0, v2, s[4:5] offset:2
 ; GFX900-NEXT:    s_waitcnt vmcnt(0)
 ; 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)
 ; GFX900-NEXT:    buffer_load_ushort v0, off, s[0:3], 0 offset:6
 ; GFX900-NEXT:    buffer_load_ushort v3, off, s[0:3], 0 offset:4
 ; GFX900-NEXT:    s_waitcnt vmcnt(1)
 ; 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
 ;
 ; FLATSCR-LABEL: vload2_private:
 ; FLATSCR:       ; %bb.0: ; %entry
 ; FLATSCR-NEXT:    s_add_u32 flat_scratch_lo, s2, s5
 ; FLATSCR-NEXT:    s_addc_u32 flat_scratch_hi, s3, 0

 ; GFX10_DEFAULT-NEXT:    buffer_store_short v0, off, s[0:3], 0 offset:8
 ; GFX10_DEFAULT-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10_DEFAULT-NEXT:    s_clause 0x1
 ; GFX10_DEFAULT-NEXT:    buffer_load_ushort v0, off, s[0:3], 0 offset:6
 ; GFX10_DEFAULT-NEXT:    buffer_load_ushort v3, off, s[0:3], 0 offset:4
 ; 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
 ;
 ; FLATSCR_GFX10-LABEL: vload2_private:
 ; FLATSCR_GFX10:       ; %bb.0: ; %entry
 ; FLATSCR_GFX10-NEXT:    s_add_u32 s2, s2, s5
 ; FLATSCR_GFX10-NEXT:    s_addc_u32 s3, s3, 0
bb:
   %to.hi = insertelement <2 x i16> undef, i16 %load_hi, i32 1
   %op.hi = add <2 x i16> %to.hi, <i16 12, i16 12>
   %result = insertelement <2 x i16> %op.hi, i16 %load_lo, i32 0
   ret <2 x i16> %result
 }
 
 ; 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
 ; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-NEXT:    ds_read_u16 v1, v0 offset:2
 ; 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:
 ; GFX11:       ; %bb.0: ; %bb
 ; GFX11-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX11-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-NEXT:    ds_load_u16 v1, v0 offset:2
 ; GFX11-NEXT:    ds_load_u16_d16_hi v0, v0
 ; 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
   %load_lo = load volatile i16, i16 addrspace(3)* %gep_lo
   %gep_hi = getelementptr inbounds i16, i16 addrspace(3)* %ptr, i64 0
   %load_hi = load volatile i16, i16 addrspace(3)* %gep_hi
   %to.hi = insertelement <2 x i16> undef, i16 %load_hi, i32 1
   %op.hi = add <2 x i16> %to.hi, <i16 12, i16 12>
bb:
   %load_hi = load i16, i16 addrspace(5)* %gep_hi
   %to.hi = insertelement <2 x i16> undef, i16 %load_hi, i32 1
   %op.hi = add <2 x i16> %to.hi, <i16 12, i16 12>
   %result = insertelement <2 x i16> %op.hi, i16 %load_lo, i32 0
   ret <2 x i16> %result
 }
 
 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
 ; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-NEXT:    global_load_ushort v2, v[0:1], off offset:2 glc dlc
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; 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:
 ; GFX11:       ; %bb.0: ; %bb
 ; GFX11-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX11-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-NEXT:    global_load_u16 v2, v[0:1], off offset:2 glc dlc
 ; GFX11-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-NEXT:    global_load_d16_hi_b16 v0, v[0:1], off glc dlc
 ; 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
   %load_lo = load volatile i16, i16 addrspace(1)* %gep_lo
   %gep_hi = getelementptr inbounds i16, i16 addrspace(1)* %ptr, i64 0
   %load_hi = load volatile i16, i16 addrspace(1)* %gep_hi
   %to.hi = insertelement <2 x i16> undef, i16 %load_hi, i32 1
   %op.hi = add <2 x i16> %to.hi, <i16 12, i16 12>
   %result = insertelement <2 x i16> %op.hi, i16 %load_lo, i32 0
   ret <2 x i16> %result
 }
 
 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
 ; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-NEXT:    v_add_co_u32 v2, vcc_lo, v0, 2
 ; GFX10-NEXT:    v_add_co_ci_u32_e32 v3, vcc_lo, 0, v1, vcc_lo
 ; GFX10-NEXT:    flat_load_ushort v2, v[2:3] glc dlc
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; 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:
 ; GFX11:       ; %bb.0: ; %bb
 ; GFX11-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX11-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-NEXT:    flat_load_u16 v2, v[0:1] offset:2 glc dlc
 ; GFX11-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-NEXT:    flat_load_d16_hi_b16 v0, v[0:1] glc dlc
 ; 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
   %load_lo = load volatile i16, i16 addrspace(0)* %gep_lo
   %gep_hi = getelementptr inbounds i16, i16 addrspace(0)* %ptr, i64 0
   %load_hi = load volatile i16, i16 addrspace(0)* %gep_hi
   %to.hi = insertelement <2 x i16> undef, i16 %load_hi, i32 1
   %op.hi = add <2 x i16> %to.hi, <i16 12, i16 12>
   %result = insertelement <2 x i16> %op.hi, i16 %load_lo, i32 0
   ret <2 x i16> %result
 }
 
 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
 ; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-NEXT:    v_mov_b32_e32 v2, 0x7b
 ; GFX10-NEXT:    ds_read_u16 v3, v0
 ; 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:
 ; GFX11:       ; %bb.0: ; %bb
 ; GFX11-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX11-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-NEXT:    v_mov_b32_e32 v2, 0x7b
 ; GFX11-NEXT:    ds_load_u16 v3, v0
 ; 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
   %gep_hi = getelementptr inbounds i16, i16 addrspace(3)* %ptr, i64 0
   %load_hi = load i16, i16 addrspace(3)* %gep_hi
   store i16 123, i16 addrspace(3)* %may.alias
   %load_lo = load i16, i16 addrspace(3)* %gep_lo
 
   %to.hi = insertelement <2 x i16> undef, i16 %load_hi, i32 1
   %result = insertelement <2 x i16> %to.hi, i16 %load_lo, i32 0
   ret <2 x i16> %result
 }

llvm/test/CodeGen/AMDGPU/divergence-driven-buildvector.ll

 ; GCN-NEXT:    v_lshlrev_b32_e32 v1, 16, v1
 ; GCN-NEXT:    v_and_b32_e32 v0, 0xffff, v0
 ; GCN-NEXT:    v_or_b32_e32 v0, v0, v1
 ; GCN-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; 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
   %val = bitcast <2 x i16> %vec to i32
   ret i32 %val
 }
 
 define amdgpu_kernel void @uniform_vec_i16_LH(i32 addrspace(1)* %out, i16 %a, i32 %b) {

 ; GCN-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GCN-NEXT:    s_mov_b32 s4, 0xffff
 ; GCN-NEXT:    v_bfi_b32 v0, s4, v0, v1
 ; GCN-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; 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
   %tmp = insertelement <2 x i16> undef, i16 %a, i32 0
   %vec = insertelement <2 x i16> %tmp, i16 %tr, i32 1
   %val = bitcast <2 x i16> %vec to i32
   ret i32 %val
 }

 ; GCN-NEXT:    v_lshrrev_b32_e32 v0, 16, v0
 ; GCN-NEXT:    v_and_b32_e32 v1, 0xffff0000, v1
 ; GCN-NEXT:    v_or_b32_e32 v0, v0, v1
 ; GCN-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; 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
   %shift_b = lshr i32 %b, 16
   %tr_b = trunc i32 %shift_b to i16
   %tmp = insertelement <2 x i16> undef, i16 %tr_a, i32 0
   %vec = insertelement <2 x i16> %tmp, i16 %tr_b, i32 1
   %val = bitcast <2 x i16> %vec to i32

 ; GCN-NEXT:    v_cvt_f16_f32_e32 v1, v1
 ; GCN-NEXT:    v_lshlrev_b32_e32 v1, 16, v1
 ; GCN-NEXT:    v_or_b32_e32 v0, v0, v1
 ; GCN-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; 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
   %val = bitcast <2 x half> %vec to float
   ret float %val
 }
 
 define <2 x i16> @build_vec_v2i16_undeflo_divergent(i16 addrspace(3)* %in) #0 {

llvm/test/CodeGen/AMDGPU/extract-subvector-16bit.ll

 ; GFX9-NEXT:    v_mov_b32_e32 v5, s11
 ; GFX9-NEXT:  .LBB0_3: ; %T
 ; GFX9-NEXT:    global_load_dwordx4 v[2:5], v[0:1], off glc
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:  .LBB0_4: ; %exit
 ; 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
 
 T:
   %t = load volatile <8 x i16>, <8 x i16> addrspace(1) * %p0
   br label %exit
 
 F:

 ; GFX9-NEXT:    v_mov_b32_e32 v5, s11
 ; GFX9-NEXT:  .LBB1_3: ; %T
 ; GFX9-NEXT:    global_load_dwordx4 v[2:5], v[0:1], off glc
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:  .LBB1_4: ; %exit
 ; 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
 
 T:
   %t = load volatile <8 x i16>, <8 x i16> addrspace(1) * %p0
   br label %exit
 
 F:

 ; GFX9-NEXT:    v_mov_b32_e32 v2, s8
 ; GFX9-NEXT:    v_mov_b32_e32 v3, s9
 ; GFX9-NEXT:    v_mov_b32_e32 v4, s10
 ; GFX9-NEXT:    v_mov_b32_e32 v5, s11
 ; GFX9-NEXT:  .LBB2_3: ; %T
 ; 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
 ; GFX9-NEXT:    v_cmp_ge_f16_e32 vcc, 0.5, v0
 ; GFX9-NEXT:    v_cndmask_b32_e32 v5, v3, v4, vcc
 ; GFX9-NEXT:    v_cmp_nle_f16_sdwa vcc, v0, v1 src0_sel:WORD_1 src1_sel:DWORD
 ; GFX9-NEXT:    v_cndmask_b32_e32 v6, v4, v3, vcc
 ; GFX9-NEXT:    v_cmp_ge_f16_e32 vcc, 0.5, v2

 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    global_load_dwordx4 v[4:7], v[0:1], off glc
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    ; kill: killed $vgpr0 killed $vgpr1
 ; GFX9-NEXT:  .LBB3_4: ; %exit
 ; 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
 
 T:
   %t = load volatile <16 x i16>, <16 x i16> addrspace(1) * %p0
   br label %exit
 
 F:

 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    global_load_dwordx4 v[4:7], v[0:1], off glc
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    ; kill: killed $vgpr0 killed $vgpr1
 ; GFX9-NEXT:  .LBB4_4: ; %exit
 ; 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
 
 T:
   %t = load volatile <16 x i16>, <16 x i16> addrspace(1) * %p0
   br label %exit
 
 F:

 ; GFX9-NEXT:  .LBB5_3: ; %T
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    global_load_dwordx4 v[2:5], v[0:1], off offset:16 glc
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    global_load_dwordx4 v[4:7], v[0:1], off glc
 ; 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
 ; GFX9-NEXT:    v_cmp_ge_f16_e32 vcc, 0.5, v0
 ; GFX9-NEXT:    v_cndmask_b32_e32 v5, v2, v3, vcc
 ; GFX9-NEXT:    v_cmp_nle_f16_sdwa vcc, v0, v1 src0_sel:WORD_1 src1_sel:DWORD
 ; GFX9-NEXT:    v_cndmask_b32_e32 v6, v3, v2, vcc
 ; GFX9-NEXT:    v_cmp_ge_f16_e32 vcc, 0.5, v4

llvm/test/CodeGen/AMDGPU/fast-unaligned-load-store.global.ll

 ; GFX7-UNALIGNED-NEXT:    flat_load_dword v0, v[0:1]
 ; GFX7-UNALIGNED-NEXT:    s_waitcnt vmcnt(0)
 ; GFX7-UNALIGNED-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX9-LABEL: global_load_2xi16_align1:
 ; 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]
 ;
 ; GFX10-LABEL: global_load_2xi16_align1:
 ; 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 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]
 ;
 ; GFX11-LABEL: global_load_2xi16_align1:
 ; 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 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]
   %gep.p = getelementptr i16, i16 addrspace(1)* %p, i64 1
   %p.0 = load i16, i16 addrspace(1)* %p, align 1
   %p.1 = load i16, i16 addrspace(1)* %gep.p, align 1
   %zext.0 = zext i16 %p.0 to i32
   %zext.1 = zext i16 %p.1 to i32

 ; GFX7-UNALIGNED-NEXT:    flat_load_dword v0, v[0:1]
 ; GFX7-UNALIGNED-NEXT:    s_waitcnt vmcnt(0)
 ; GFX7-UNALIGNED-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX9-LABEL: global_load_2xi16_align4:
 ; 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]
 ;
 ; GFX10-LABEL: global_load_2xi16_align4:
 ; 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 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]
 ;
 ; GFX11-LABEL: global_load_2xi16_align4:
 ; 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 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]
   %gep.p = getelementptr i16, i16 addrspace(1)* %p, i64 1
   %p.0 = load i16, i16 addrspace(1)* %p, align 4
   %p.1 = load i16, i16 addrspace(1)* %gep.p, align 2
   %zext.0 = zext i16 %p.0 to i32
   %zext.1 = zext i16 %p.1 to i32

llvm/test/CodeGen/AMDGPU/fast-unaligned-load-store.private.ll

 ; GFX7-UNALIGNED-NEXT:    buffer_load_dword v0, v0, s[0:3], 0 offen
 ; GFX7-UNALIGNED-NEXT:    s_waitcnt vmcnt(0)
 ; GFX7-UNALIGNED-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX9-LABEL: private_load_2xi16_align1:
 ; 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]
 ;
 ; GFX9-FLASTSCR-LABEL: private_load_2xi16_align1:
 ; 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]
 ;
 ; GFX10-LABEL: private_load_2xi16_align1:
 ; GFX10:       ; %bb.0:
 ; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; 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]
 ;
 ; GFX10-FLASTSCR-LABEL: private_load_2xi16_align1:
 ; GFX10-FLASTSCR:       ; %bb.0:
 ; GFX10-FLASTSCR-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; 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]
 ;
 ; GFX11-LABEL: private_load_2xi16_align1:
 ; GFX11:       ; %bb.0:
 ; GFX11-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; 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]
 ;
 ; GFX11-FLASTSCR-LABEL: private_load_2xi16_align1:
 ; GFX11-FLASTSCR:       ; %bb.0:
 ; GFX11-FLASTSCR-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; 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]
   %gep.p = getelementptr i16, i16 addrspace(5)* %p, i64 1
   %p.0 = load i16, i16 addrspace(5)* %p, align 1
   %p.1 = load i16, i16 addrspace(5)* %gep.p, align 1
   %zext.0 = zext i16 %p.0 to i32
   %zext.1 = zext i16 %p.1 to i32

 ; GFX7-UNALIGNED-NEXT:    buffer_load_dword v0, v0, s[0:3], 0 offen
 ; GFX7-UNALIGNED-NEXT:    s_waitcnt vmcnt(0)
 ; GFX7-UNALIGNED-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX9-LABEL: private_load_2xi16_align4:
 ; 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]
 ;
 ; GFX9-FLASTSCR-LABEL: private_load_2xi16_align4:
 ; 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]
 ;
 ; GFX10-LABEL: private_load_2xi16_align4:
 ; GFX10:       ; %bb.0:
 ; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; 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]
 ;
 ; GFX10-FLASTSCR-LABEL: private_load_2xi16_align4:
 ; GFX10-FLASTSCR:       ; %bb.0:
 ; GFX10-FLASTSCR-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; 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]
 ;
 ; GFX11-LABEL: private_load_2xi16_align4:
 ; GFX11:       ; %bb.0:
 ; GFX11-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; 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]
 ;
 ; GFX11-FLASTSCR-LABEL: private_load_2xi16_align4:
 ; GFX11-FLASTSCR:       ; %bb.0:
 ; GFX11-FLASTSCR-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; 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]
   %gep.p = getelementptr i16, i16 addrspace(5)* %p, i64 1
   %p.0 = load i16, i16 addrspace(5)* %p, align 4
   %p.1 = load i16, i16 addrspace(5)* %gep.p, align 2
   %zext.0 = zext i16 %p.0 to i32
   %zext.1 = zext i16 %p.1 to i32

llvm/test/CodeGen/AMDGPU/fcanonicalize.f16.ll

 ; CI-NEXT:    s_endpgm
   %val = bitcast i16 %val.arg to half
   %canonicalized = call half @llvm.canonicalize.f16(half %val)
   store half %canonicalized, half addrspace(1)* %out
   ret void
 }
 
 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)
   ret <2 x half> %canonicalized
 }
 
 define amdgpu_kernel void @v_test_canonicalize_fabs_var_f16(half addrspace(1)* %out) #1 {
 ; VI-LABEL: v_test_canonicalize_fabs_var_f16:

 ; CI-NEXT:    v_mul_f32_e32 v0, 1.0, v0
 ; CI-NEXT:    s_setpc_b64 s[30:31]
   %vec = insertelement <4 x half> undef, half %val, i32 0
   %canonicalized = call <4 x half> @llvm.canonicalize.v4f16(<4 x half> %vec)
   ret <4 x half> %canonicalized
 }
 
 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)
   ret <4 x half> %canonicalized
 }
 
 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
   %canonicalized = call <4 x half> @llvm.canonicalize.v4f16(<4 x half> %vec2)
   ret <4 x half> %canonicalized
 }
 
 attributes #0 = { nounwind readnone }
 attributes #1 = { nounwind "denormal-fp-math-f32"="preserve-sign,preserve-sign" }
 attributes #2 = { nounwind "denormal-fp-math"="preserve-sign,preserve-sign" }
 attributes #3 = { nounwind "denormal-fp-math-f32"="preserve-sign,preserve-sign" }

llvm/test/CodeGen/AMDGPU/fmax_legacy.f16.ll

 ; GFX9-SAFE:       ; %bb.0:
 ; GFX9-SAFE-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX9-SAFE-NEXT:    v_lshrrev_b32_e32 v2, 16, v1
 ; GFX9-SAFE-NEXT:    v_lshrrev_b32_e32 v3, 16, v0
 ; GFX9-SAFE-NEXT:    v_cmp_nle_f16_e32 vcc, v3, v2
 ; 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:
 ; GFX9-NNAN:       ; %bb.0:
 ; GFX9-NNAN-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX9-NNAN-NEXT:    v_pk_max_f16 v0, v0, v1
 ; GFX9-NNAN-NEXT:    s_setpc_b64 s[30:31]
 ;

 ; GFX9-SAFE-NEXT:    v_lshrrev_b32_e32 v4, 16, v2
 ; GFX9-SAFE-NEXT:    v_lshrrev_b32_e32 v5, 16, v0
 ; GFX9-SAFE-NEXT:    v_cmp_nle_f16_e32 vcc, v5, v4
 ; GFX9-SAFE-NEXT:    v_cndmask_b32_e32 v4, v4, v5, vcc
 ; GFX9-SAFE-NEXT:    v_cmp_nle_f16_e32 vcc, v1, v3
 ; 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:
 ; GFX9-NNAN:       ; %bb.0:
 ; GFX9-NNAN-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX9-NNAN-NEXT:    v_pk_max_f16 v1, v1, v3
 ; GFX9-NNAN-NEXT:    v_pk_max_f16 v0, v0, v2
 ; GFX9-NNAN-NEXT:    s_setpc_b64 s[30:31]

 ; GFX9-SAFE-NEXT:    v_cmp_nle_f16_e32 vcc, v7, v6
 ; GFX9-SAFE-NEXT:    v_cndmask_b32_e32 v6, v6, v7, vcc
 ; GFX9-SAFE-NEXT:    v_cmp_nle_f16_e32 vcc, v5, v4
 ; GFX9-SAFE-NEXT:    v_cndmask_b32_e32 v4, v4, v5, vcc
 ; GFX9-SAFE-NEXT:    v_cmp_nle_f16_e32 vcc, v1, v3
 ; 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:
 ; GFX9-NNAN:       ; %bb.0:
 ; GFX9-NNAN-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX9-NNAN-NEXT:    v_pk_max_f16 v0, v0, v2
 ; GFX9-NNAN-NEXT:    v_pk_max_f16 v1, v1, v3
 ; GFX9-NNAN-NEXT:    s_setpc_b64 s[30:31]

 ; GFX9-SAFE-NEXT:    v_cmp_nle_f16_e32 vcc, v3, v7
 ; GFX9-SAFE-NEXT:    v_cndmask_b32_e32 v3, v7, v3, vcc
 ; GFX9-SAFE-NEXT:    v_cmp_nle_f16_e32 vcc, v2, v6
 ; GFX9-SAFE-NEXT:    v_cndmask_b32_e32 v2, v6, v2, vcc
 ; GFX9-SAFE-NEXT:    v_cmp_nle_f16_e32 vcc, v1, v5
 ; 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:
 ; GFX9-NNAN:       ; %bb.0:
 ; GFX9-NNAN-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX9-NNAN-NEXT:    v_pk_max_f16 v0, v0, v4
 ; GFX9-NNAN-NEXT:    v_pk_max_f16 v1, v1, v5
 ; GFX9-NNAN-NEXT:    v_pk_max_f16 v2, v2, v6

llvm/test/CodeGen/AMDGPU/fmin_legacy.f16.ll

 ; GFX9-SAFE:       ; %bb.0:
 ; GFX9-SAFE-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX9-SAFE-NEXT:    v_lshrrev_b32_e32 v2, 16, v1
 ; GFX9-SAFE-NEXT:    v_lshrrev_b32_e32 v3, 16, v0
 ; GFX9-SAFE-NEXT:    v_cmp_ngt_f16_e32 vcc, v3, v2
 ; 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:
 ; GFX9-NNAN:       ; %bb.0:
 ; GFX9-NNAN-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX9-NNAN-NEXT:    v_pk_min_f16 v0, v0, v1
 ; GFX9-NNAN-NEXT:    s_setpc_b64 s[30:31]
 ;

 ; GFX9-SAFE-NEXT:    v_lshrrev_b32_e32 v4, 16, v2
 ; GFX9-SAFE-NEXT:    v_lshrrev_b32_e32 v5, 16, v0
 ; GFX9-SAFE-NEXT:    v_cmp_ngt_f16_e32 vcc, v5, v4
 ; GFX9-SAFE-NEXT:    v_cndmask_b32_e32 v4, v4, v5, vcc
 ; GFX9-SAFE-NEXT:    v_cmp_ngt_f16_e32 vcc, v1, v3
 ; 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:
 ; GFX9-NNAN:       ; %bb.0:
 ; GFX9-NNAN-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX9-NNAN-NEXT:    v_pk_min_f16 v1, v1, v3
 ; GFX9-NNAN-NEXT:    v_pk_min_f16 v0, v0, v2
 ; GFX9-NNAN-NEXT:    s_setpc_b64 s[30:31]

 ; GFX9-SAFE-NEXT:    v_cmp_ngt_f16_e32 vcc, v7, v6
 ; GFX9-SAFE-NEXT:    v_cndmask_b32_e32 v6, v6, v7, vcc
 ; GFX9-SAFE-NEXT:    v_cmp_ngt_f16_e32 vcc, v5, v4
 ; GFX9-SAFE-NEXT:    v_cndmask_b32_e32 v4, v4, v5, vcc
 ; GFX9-SAFE-NEXT:    v_cmp_ngt_f16_e32 vcc, v1, v3
 ; 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:
 ; GFX9-NNAN:       ; %bb.0:
 ; GFX9-NNAN-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX9-NNAN-NEXT:    v_pk_min_f16 v0, v0, v2
 ; GFX9-NNAN-NEXT:    v_pk_min_f16 v1, v1, v3
 ; GFX9-NNAN-NEXT:    s_setpc_b64 s[30:31]

 ; GFX9-SAFE-NEXT:    v_cmp_ngt_f16_e32 vcc, v3, v7
 ; GFX9-SAFE-NEXT:    v_cndmask_b32_e32 v3, v7, v3, vcc
 ; GFX9-SAFE-NEXT:    v_cmp_ngt_f16_e32 vcc, v2, v6
 ; GFX9-SAFE-NEXT:    v_cndmask_b32_e32 v2, v6, v2, vcc
 ; GFX9-SAFE-NEXT:    v_cmp_ngt_f16_e32 vcc, v1, v5
 ; 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:
 ; GFX9-NNAN:       ; %bb.0:
 ; GFX9-NNAN-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX9-NNAN-NEXT:    v_pk_min_f16 v0, v0, v4
 ; GFX9-NNAN-NEXT:    v_pk_min_f16 v1, v1, v5
 ; GFX9-NNAN-NEXT:    v_pk_min_f16 v2, v2, v6

llvm/test/CodeGen/AMDGPU/fshr.ll

 ; GFX9-NEXT:    v_lshlrev_b16_e32 v1, v7, v1
 ; GFX9-NEXT:    v_lshrrev_b16_e32 v3, v5, v3
 ; GFX9-NEXT:    v_or_b32_e32 v1, v1, v3
 ; GFX9-NEXT:    v_lshlrev_b16_e32 v0, 1, v0
 ; 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_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:
 ; R600:       ; %bb.0:
 ; R600-NEXT:    CF_END
 ; R600-NEXT:    PAD
 ;
 ; GFX10-LABEL: v_fshr_v3i16:
 ; 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]
 ;
 ; GFX11-LABEL: v_fshr_v3i16:
 ; 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]
   %ret = call <3 x i16> @llvm.fshr.v3i16(<3 x i16> %src0, <3 x i16> %src1, <3 x i16> %src2)
   ret <3 x i16> %ret
 }
 
 define <4 x i16> @v_fshr_v4i16(<4 x i16> %src0, <4 x i16> %src1, <4 x i16> %src2) {

 ; GFX9-NEXT:    v_xor_b32_e32 v8, -1, v5
 ; GFX9-NEXT:    v_lshlrev_b16_e32 v1, v8, v1
 ; GFX9-NEXT:    v_lshrrev_b16_e32 v3, v5, v3
 ; GFX9-NEXT:    v_or_b32_e32 v1, v1, v3
 ; GFX9-NEXT:    v_lshlrev_b16_e32 v0, 1, v0
 ; 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:
 ; R600:       ; %bb.0:
 ; R600-NEXT:    CF_END
 ; R600-NEXT:    PAD
 ;
 ; GFX10-LABEL: v_fshr_v4i16:
 ; 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, 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:
 ; 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, 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
 }
 
 define i64 @v_fshr_i64(i64 %src0, i64 %src1, i64 %src2) {
 ; SI-LABEL: v_fshr_i64:
 ; SI:       ; %bb.0:

llvm/test/CodeGen/AMDGPU/idot4s.ll

 ; GFX8-NEXT:    flat_store_short v[0:1], v2
 ; GFX8-NEXT:    s_endpgm
 ;
 ; GFX9-NODL-LABEL: idot4_acc16_vecMul:
 ; GFX9-NODL:       ; %bb.0: ; %entry
 ; 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
 ; 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:    global_store_short v0, v1, s[2:3]
 ; GFX9-NODL-NEXT:    s_endpgm
 ;
 ; GFX9-DL-LABEL: idot4_acc16_vecMul:
 ; GFX9-DL:       ; %bb.0: ; %entry
 ; 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
 ; 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:    global_store_short v0, v1, s[2:3]
 ; GFX9-DL-NEXT:    s_endpgm
 ;
 ; GFX10-DL-LABEL: idot4_acc16_vecMul:
 ; GFX10-DL:       ; %bb.0: ; %entry
 ; 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]
 ; GFX10-DL-NEXT:    global_load_dword v2, v0, s[6:7]
 ; 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
 ; GFX10-DL-NEXT:    v_pk_mul_lo_u16 v1, v1, v2
 ; GFX10-DL-NEXT:    v_add_nc_u16 v2, v3, v5
 ; GFX10-DL-NEXT:    v_lshrrev_b32_e32 v3, 16, v1
 ; GFX10-DL-NEXT:    v_add_nc_u16 v1, v2, v1
 ; GFX10-DL-NEXT:    v_add_nc_u16 v1, v1, v3

llvm/test/CodeGen/AMDGPU/idot4u.ll

 ; GFX8-NEXT:    s_endpgm
 ;
 ; GFX9-NODL-LABEL: udot4_acc16_vecMul:
 ; GFX9-NODL:       ; %bb.0: ; %entry
 ; 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:    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
 ;
 ; GFX9-DL-LABEL: udot4_acc16_vecMul:
 ; GFX9-DL:       ; %bb.0: ; %entry
 ; 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_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
 ;
 ; GFX10-DL-LABEL: udot4_acc16_vecMul:
 ; GFX10-DL:       ; %bb.0: ; %entry
 ; 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]
 ; GFX10-DL-NEXT:    global_load_dword v2, v0, s[6:7]
 ; 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
 ; GFX10-DL-NEXT:    v_pk_mul_lo_u16 v1, v1, v2
 ; GFX10-DL-NEXT:    v_add_nc_u16 v2, v3, v5
 ; GFX10-DL-NEXT:    v_lshrrev_b32_e32 v3, 16, v1
 ; GFX10-DL-NEXT:    v_add_nc_u16 v1, v2, v1
 ; GFX10-DL-NEXT:    v_add_nc_u16 v1, v1, v3

llvm/test/CodeGen/AMDGPU/idot8s.ll

 ; 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, 12
 ; 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_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
 ; GFX9-NEXT:    v_add_u16_sdwa v1, v2, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
 ; GFX9-NEXT:    global_store_short v0, v1, s[2:3]
 ; GFX9-NEXT:    s_endpgm
 ;
 ; GFX9-DL-LABEL: idot8_acc16_vecMul:
 ; GFX9-DL:       ; %bb.0: ; %entry
 ; GFX9-DL-NEXT:    s_mov_b32 s8, SCRATCH_RSRC_DWORD0
 ; GFX9-DL-NEXT:    s_mov_b32 s9, SCRATCH_RSRC_DWORD1
 ; GFX9-DL-NEXT:    s_mov_b32 s10, -1
 ; GFX9-DL-NEXT:    s_mov_b32 s11, 0xe00000
 ; GFX9-DL-NEXT:    s_add_u32 s8, s8, s3
 ; 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, 12
 ; 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_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
 ; 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
 ;
 ; GFX10-DL-XNACK-LABEL: idot8_acc16_vecMul:

 ; GFX10-DL-XNACK-NEXT:    s_addc_u32 s9, s9, 0
 ; GFX10-DL-XNACK-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-DL-XNACK-NEXT:    s_clause 0x1
 ; GFX10-DL-XNACK-NEXT:    global_load_dword v1, v0, s[4:5]
 ; GFX10-DL-XNACK-NEXT:    global_load_dword v2, v0, s[6:7]
 ; 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
 ; GFX10-DL-XNACK-NEXT:    v_add_nc_u16 v2, v3, v5
 ; GFX10-DL-XNACK-NEXT:    v_lshrrev_b32_e32 v3, 16, v1
 ; GFX10-DL-XNACK-NEXT:    v_add_nc_u16 v1, v2, v1
 ; GFX10-DL-XNACK-NEXT:    v_add_nc_u16 v1, v1, v3
 ; GFX10-DL-XNACK-NEXT:    global_store_short v0, v1, s[0:1]

 ; GFX10-DL-NOXNACK-NEXT:    s_add_u32 s8, s8, s3
 ; GFX10-DL-NOXNACK-NEXT:    s_addc_u32 s9, s9, 0
 ; GFX10-DL-NOXNACK-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-DL-NOXNACK-NEXT:    s_clause 0x1
 ; GFX10-DL-NOXNACK-NEXT:    global_load_dword v1, v0, s[4:5]
 ; 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
 ; GFX10-DL-NOXNACK-NEXT:    v_add_nc_u16 v1, v3, v5
 ; GFX10-DL-NOXNACK-NEXT:    v_lshrrev_b32_e32 v3, 16, v0
 ; GFX10-DL-NOXNACK-NEXT:    v_add_nc_u16 v0, v1, v0
 ; GFX10-DL-NOXNACK-NEXT:    v_add_nc_u16 v0, v0, v3
 ; GFX10-DL-NOXNACK-NEXT:    global_store_short v2, v0, s[0:1]

llvm/test/CodeGen/AMDGPU/idot8u.ll

 ; GFX9-NEXT:    s_mov_b32 s8, SCRATCH_RSRC_DWORD0
 ; GFX9-NEXT:    s_mov_b32 s9, SCRATCH_RSRC_DWORD1
 ; GFX9-NEXT:    s_mov_b32 s10, -1
 ; GFX9-NEXT:    s_mov_b32 s11, 0xe00000
 ; GFX9-NEXT:    s_add_u32 s8, s8, s3
 ; 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
 ; 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, 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
 ; GFX9-NEXT:    v_add_u16_sdwa v1, v2, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
 ; GFX9-NEXT:    global_store_short v0, v1, s[2:3]
 ; GFX9-NEXT:    s_endpgm
 ;
 ; GFX9-DL-LABEL: udot8_acc16_vecMul:
 ; GFX9-DL:       ; %bb.0: ; %entry
 ; GFX9-DL-NEXT:    s_mov_b32 s8, SCRATCH_RSRC_DWORD0
 ; GFX9-DL-NEXT:    s_mov_b32 s9, SCRATCH_RSRC_DWORD1
 ; GFX9-DL-NEXT:    s_mov_b32 s10, -1
 ; GFX9-DL-NEXT:    s_mov_b32 s11, 0xe00000
 ; GFX9-DL-NEXT:    s_add_u32 s8, s8, s3
 ; 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
 ; 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, 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
 ; GFX9-DL-NEXT:    v_add_u16_sdwa v1, v2, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1

 ; GFX10-DL-NEXT:    s_addc_u32 s9, s9, 0
 ; GFX10-DL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-DL-NEXT:    s_clause 0x1
 ; GFX10-DL-NEXT:    global_load_dword v1, v0, s[4:5]
 ; GFX10-DL-NEXT:    global_load_dword v2, v0, s[6:7]
 ; 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
 ; GFX10-DL-NEXT:    v_add_nc_u16 v2, v3, v5
 ; GFX10-DL-NEXT:    v_lshrrev_b32_e32 v3, 16, v1
 ; GFX10-DL-NEXT:    v_add_nc_u16 v1, v2, v1
 ; GFX10-DL-NEXT:    v_add_nc_u16 v1, v1, v3
 ; GFX10-DL-NEXT:    global_store_short v0, v1, s[0:1]

 ; GFX9-NEXT:    s_mov_b32 s8, SCRATCH_RSRC_DWORD0
 ; GFX9-NEXT:    s_mov_b32 s9, SCRATCH_RSRC_DWORD1
 ; GFX9-NEXT:    s_mov_b32 s10, -1
 ; GFX9-NEXT:    s_mov_b32 s11, 0xe00000
 ; GFX9-NEXT:    s_add_u32 s8, s8, s3
 ; 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
 ; 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, 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
 ; GFX9-NEXT:    v_add_u16_sdwa v1, v2, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
 ; GFX9-NEXT:    v_and_b32_e32 v1, 15, v1
 ; GFX9-NEXT:    global_store_byte v0, v1, s[2:3]
 ; GFX9-NEXT:    s_endpgm
 ;
 ; GFX9-DL-LABEL: udot8_acc4_vecMul:
 ; GFX9-DL:       ; %bb.0: ; %entry
 ; GFX9-DL-NEXT:    s_mov_b32 s8, SCRATCH_RSRC_DWORD0
 ; GFX9-DL-NEXT:    s_mov_b32 s9, SCRATCH_RSRC_DWORD1
 ; GFX9-DL-NEXT:    s_mov_b32 s10, -1
 ; GFX9-DL-NEXT:    s_mov_b32 s11, 0xe00000
 ; GFX9-DL-NEXT:    s_add_u32 s8, s8, s3
 ; 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
 ; 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, 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
 ; GFX9-DL-NEXT:    v_add_u16_sdwa v1, v2, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1

 ; GFX10-DL-NEXT:    s_addc_u32 s9, s9, 0
 ; GFX10-DL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-DL-NEXT:    s_clause 0x1
 ; GFX10-DL-NEXT:    global_load_dword v1, v0, s[4:5]
 ; GFX10-DL-NEXT:    global_load_dword v2, v0, s[6:7]
 ; 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
 ; GFX10-DL-NEXT:    v_add_nc_u16 v2, v3, v5
 ; GFX10-DL-NEXT:    v_lshrrev_b32_e32 v3, 16, v1
 ; GFX10-DL-NEXT:    v_add_nc_u16 v1, v2, v1
 ; GFX10-DL-NEXT:    v_add_nc_u16 v1, v1, v3
 ; GFX10-DL-NEXT:    v_and_b32_e32 v1, 15, v1

llvm/test/CodeGen/AMDGPU/insert_vector_elt.v2i16.ll

; GFX11-NEXT:    s_endpgm
   ret void
 }
 
 define amdgpu_kernel void @v_insertelement_v2i16_0(<2 x i16> addrspace(1)* %out, <2 x i16> addrspace(1)* %in) #0 {
 ; GFX9-LABEL: v_insertelement_v2i16_0:
 ; 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
 ;
 ; VI-LABEL: v_insertelement_v2i16_0:
 ; VI:       ; %bb.0:
 ; VI-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; VI-NEXT:    v_lshlrev_b32_e32 v2, 2, v0
 ; VI-NEXT:    s_waitcnt lgkmcnt(0)

 ; GFX11-LABEL: v_insertelement_v2i16_0:
 ; GFX11:       ; %bb.0:
 ; GFX11-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; 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_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
   %tid = call i32 @llvm.amdgcn.workitem.id.x() #1
   %tid.ext = sext i32 %tid to i64
   %in.gep = getelementptr inbounds <2 x i16>, <2 x i16> addrspace(1)* %in, i64 %tid.ext
   %out.gep = getelementptr inbounds <2 x i16>, <2 x i16> addrspace(1)* %out, i64 %tid.ext
   %vec = load <2 x i16>, <2 x i16> addrspace(1)* %in.gep
   %vecins = insertelement <2 x i16> %vec, i16 999, i32 0
   store <2 x i16> %vecins, <2 x i16> addrspace(1)* %out.gep
   ret void
 }
 
 define amdgpu_kernel void @v_insertelement_v2i16_0_reghi(<2 x i16> addrspace(1)* %out, <2 x i16> addrspace(1)* %in, i32 %elt.arg) #0 {
 ; GFX9-LABEL: v_insertelement_v2i16_0_reghi:
 ; GFX9:       ; %bb.0:
 ; 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
 ;
 ; VI-LABEL: v_insertelement_v2i16_0_reghi:
 ; VI:       ; %bb.0:
 ; VI-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; VI-NEXT:    s_load_dword s4, s[4:5], 0x10
 ; VI-NEXT:    v_lshlrev_b32_e32 v2, 2, v0

 ;
 ; GFX11-LABEL: v_insertelement_v2i16_0_reghi:
 ; GFX11:       ; %bb.0:
 ; GFX11-NEXT:    s_load_b128 s[4:7], s[0:1], 0x0
 ; GFX11-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
 ; 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
   %tid = call i32 @llvm.amdgcn.workitem.id.x() #1
   %tid.ext = sext i32 %tid to i64
   %in.gep = getelementptr inbounds <2 x i16>, <2 x i16> addrspace(1)* %in, i64 %tid.ext
   %out.gep = getelementptr inbounds <2 x i16>, <2 x i16> addrspace(1)* %out, i64 %tid.ext
   %vec = load <2 x i16>, <2 x i16> addrspace(1)* %in.gep
   %elt.hi = lshr i32 %elt.arg, 16
   %elt = trunc i32 %elt.hi to i16
   %vecins = insertelement <2 x i16> %vec, i16 %elt, i32 0
   store <2 x i16> %vecins, <2 x i16> addrspace(1)* %out.gep
   ret void
 }
 
 define amdgpu_kernel void @v_insertelement_v2i16_0_inlineimm(<2 x i16> addrspace(1)* %out, <2 x i16> addrspace(1)* %in) #0 {
 ; GFX9-LABEL: v_insertelement_v2i16_0_inlineimm:
 ; 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
 ;
 ; VI-LABEL: v_insertelement_v2i16_0_inlineimm:
 ; VI:       ; %bb.0:
 ; VI-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; VI-NEXT:    v_lshlrev_b32_e32 v2, 2, v0
 ; VI-NEXT:    s_waitcnt lgkmcnt(0)

 ;
 ; GFX11-LABEL: v_insertelement_v2i16_0_inlineimm:
 ; GFX11:       ; %bb.0:
 ; GFX11-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; 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_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
   %tid = call i32 @llvm.amdgcn.workitem.id.x() #1
   %tid.ext = sext i32 %tid to i64
   %in.gep = getelementptr inbounds <2 x i16>, <2 x i16> addrspace(1)* %in, i64 %tid.ext
   %out.gep = getelementptr inbounds <2 x i16>, <2 x i16> addrspace(1)* %out, i64 %tid.ext
   %vec = load <2 x i16>, <2 x i16> addrspace(1)* %in.gep
   %vecins = insertelement <2 x i16> %vec, i16 53, i32 0
   store <2 x i16> %vecins, <2 x i16> addrspace(1)* %out.gep
   ret void
 }
 
 ; FIXME: fold lshl_or c0, c1, v0 -> or (c0 << c1), v0
 define amdgpu_kernel void @v_insertelement_v2i16_1(<2 x i16> addrspace(1)* %out, <2 x i16> addrspace(1)* %in) #0 {
 ; GFX9-LABEL: v_insertelement_v2i16_1:
 ; 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
 ;
 ; VI-LABEL: v_insertelement_v2i16_1:
 ; VI:       ; %bb.0:
 ; VI-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; VI-NEXT:    v_lshlrev_b32_e32 v2, 2, v0
 ; VI-NEXT:    s_waitcnt lgkmcnt(0)

 ; CI-NEXT:    s_endpgm
 ;
 ; GFX11-LABEL: v_insertelement_v2i16_1:
 ; GFX11:       ; %bb.0:
 ; GFX11-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; 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
   %tid = call i32 @llvm.amdgcn.workitem.id.x() #1
   %tid.ext = sext i32 %tid to i64
   %in.gep = getelementptr inbounds <2 x i16>, <2 x i16> addrspace(1)* %in, i64 %tid.ext
   %out.gep = getelementptr inbounds <2 x i16>, <2 x i16> addrspace(1)* %out, i64 %tid.ext
   %vec = load <2 x i16>, <2 x i16> addrspace(1)* %in.gep
   %vecins = insertelement <2 x i16> %vec, i16 999, i32 1
   store <2 x i16> %vecins, <2 x i16> addrspace(1)* %out.gep
   ret void
 }
 
 define amdgpu_kernel void @v_insertelement_v2i16_1_inlineimm(<2 x i16> addrspace(1)* %out, <2 x i16> addrspace(1)* %in) #0 {
 ; GFX9-LABEL: v_insertelement_v2i16_1_inlineimm:
 ; 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
 ;
 ; VI-LABEL: v_insertelement_v2i16_1_inlineimm:
 ; VI:       ; %bb.0:
 ; VI-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; VI-NEXT:    v_lshlrev_b32_e32 v2, 2, v0
 ; VI-NEXT:    s_waitcnt lgkmcnt(0)

 ;
 ; GFX11-LABEL: v_insertelement_v2i16_1_inlineimm:
 ; GFX11:       ; %bb.0:
 ; GFX11-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; 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_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
   %tid = call i32 @llvm.amdgcn.workitem.id.x() #1
   %tid.ext = sext i32 %tid to i64
   %in.gep = getelementptr inbounds <2 x i16>, <2 x i16> addrspace(1)* %in, i64 %tid.ext
   %out.gep = getelementptr inbounds <2 x i16>, <2 x i16> addrspace(1)* %out, i64 %tid.ext
   %vec = load <2 x i16>, <2 x i16> addrspace(1)* %in.gep
   %vecins = insertelement <2 x i16> %vec, i16 -15, i32 1
   store <2 x i16> %vecins, <2 x i16> addrspace(1)* %out.gep
   ret void
 }
 
 define amdgpu_kernel void @v_insertelement_v2f16_0(<2 x half> addrspace(1)* %out, <2 x half> addrspace(1)* %in) #0 {
 ; GFX9-LABEL: v_insertelement_v2f16_0:
 ; 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
 ;
 ; VI-LABEL: v_insertelement_v2f16_0:
 ; VI:       ; %bb.0:
 ; VI-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; VI-NEXT:    v_lshlrev_b32_e32 v2, 2, v0
 ; VI-NEXT:    s_waitcnt lgkmcnt(0)

 ; CI-NEXT:    s_endpgm
 ;
 ; GFX11-LABEL: v_insertelement_v2f16_0:
 ; GFX11:       ; %bb.0:
 ; GFX11-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; 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
   %tid = call i32 @llvm.amdgcn.workitem.id.x() #1
   %tid.ext = sext i32 %tid to i64
   %in.gep = getelementptr inbounds <2 x half>, <2 x half> addrspace(1)* %in, i64 %tid.ext
   %out.gep = getelementptr inbounds <2 x half>, <2 x half> addrspace(1)* %out, i64 %tid.ext
   %vec = load <2 x half>, <2 x half> addrspace(1)* %in.gep
   %vecins = insertelement <2 x half> %vec, half 5.000000e+00, i32 0
   store <2 x half> %vecins, <2 x half> addrspace(1)* %out.gep
   ret void
 }
 
 define amdgpu_kernel void @v_insertelement_v2f16_0_inlineimm(<2 x half> addrspace(1)* %out, <2 x half> addrspace(1)* %in) #0 {
 ; GFX9-LABEL: v_insertelement_v2f16_0_inlineimm:
 ; 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
 ;
 ; VI-LABEL: v_insertelement_v2f16_0_inlineimm:
 ; VI:       ; %bb.0:
 ; VI-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; VI-NEXT:    v_lshlrev_b32_e32 v2, 2, v0
 ; VI-NEXT:    s_waitcnt lgkmcnt(0)

 ;
 ; GFX11-LABEL: v_insertelement_v2f16_0_inlineimm:
 ; GFX11:       ; %bb.0:
 ; GFX11-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; 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_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
   %tid = call i32 @llvm.amdgcn.workitem.id.x() #1
   %tid.ext = sext i32 %tid to i64
   %in.gep = getelementptr inbounds <2 x half>, <2 x half> addrspace(1)* %in, i64 %tid.ext
   %out.gep = getelementptr inbounds <2 x half>, <2 x half> addrspace(1)* %out, i64 %tid.ext
   %vec = load <2 x half>, <2 x half> addrspace(1)* %in.gep
   %vecins = insertelement <2 x half> %vec, half 0xH0035, i32 0
   store <2 x half> %vecins, <2 x half> addrspace(1)* %out.gep
   ret void
 }
 
 define amdgpu_kernel void @v_insertelement_v2f16_1(<2 x half> addrspace(1)* %out, <2 x half> addrspace(1)* %in) #0 {
 ; GFX9-LABEL: v_insertelement_v2f16_1:
 ; 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
 ;
 ; VI-LABEL: v_insertelement_v2f16_1:
 ; VI:       ; %bb.0:
 ; VI-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; VI-NEXT:    v_lshlrev_b32_e32 v2, 2, v0
 ; VI-NEXT:    s_waitcnt lgkmcnt(0)

 ; CI-NEXT:    s_endpgm
 ;
 ; GFX11-LABEL: v_insertelement_v2f16_1:
 ; GFX11:       ; %bb.0:
 ; GFX11-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; 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
   %tid = call i32 @llvm.amdgcn.workitem.id.x() #1
   %tid.ext = sext i32 %tid to i64
   %in.gep = getelementptr inbounds <2 x half>, <2 x half> addrspace(1)* %in, i64 %tid.ext
   %out.gep = getelementptr inbounds <2 x half>, <2 x half> addrspace(1)* %out, i64 %tid.ext
   %vec = load <2 x half>, <2 x half> addrspace(1)* %in.gep
   %vecins = insertelement <2 x half> %vec, half 5.000000e+00, i32 1
   store <2 x half> %vecins, <2 x half> addrspace(1)* %out.gep
   ret void
 }
 
 define amdgpu_kernel void @v_insertelement_v2f16_1_inlineimm(<2 x half> addrspace(1)* %out, <2 x half> addrspace(1)* %in) #0 {
 ; GFX9-LABEL: v_insertelement_v2f16_1_inlineimm:
 ; 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
 ;
 ; VI-LABEL: v_insertelement_v2f16_1_inlineimm:
 ; VI:       ; %bb.0:
 ; VI-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; VI-NEXT:    v_lshlrev_b32_e32 v2, 2, v0
 ; VI-NEXT:    s_waitcnt lgkmcnt(0)

 ;
 ; GFX11-LABEL: v_insertelement_v2f16_1_inlineimm:
 ; GFX11:       ; %bb.0:
 ; GFX11-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
 ; 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_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
   %tid = call i32 @llvm.amdgcn.workitem.id.x() #1
   %tid.ext = sext i32 %tid to i64
   %in.gep = getelementptr inbounds <2 x half>, <2 x half> addrspace(1)* %in, i64 %tid.ext
   %out.gep = getelementptr inbounds <2 x half>, <2 x half> addrspace(1)* %out, i64 %tid.ext
   %vec = load <2 x half>, <2 x half> addrspace(1)* %in.gep

 }
 
 define amdgpu_kernel void @v_insertelement_v4f16_0(<4 x half> addrspace(1)* %out, <4 x half> addrspace(1)* %in, [8 x i32], i32 %val) #0 {
 ; GFX9-LABEL: v_insertelement_v4f16_0:
 ; GFX9:       ; %bb.0:
 ; 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
 ;
 ; VI-LABEL: v_insertelement_v4f16_0:
 ; VI:       ; %bb.0:
 ; VI-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; VI-NEXT:    s_load_dword s4, s[4:5], 0x30
 ; VI-NEXT:    v_lshlrev_b32_e32 v2, 3, v0

 ; GFX11-LABEL: v_insertelement_v4f16_0:
 ; GFX11:       ; %bb.0:
 ; GFX11-NEXT:    s_load_b128 s[4:7], s[0:1], 0x0
 ; GFX11-NEXT:    v_lshlrev_b32_e32 v2, 3, v0
 ; GFX11-NEXT:    s_load_b32 s0, s[0:1], 0x30
 ; 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
   %tid = call i32 @llvm.amdgcn.workitem.id.x() #1
   %tid.ext = sext i32 %tid to i64
   %in.gep = getelementptr inbounds <4 x half>, <4 x half> addrspace(1)* %in, i64 %tid.ext
   %out.gep = getelementptr inbounds <4 x half>, <4 x half> addrspace(1)* %out, i64 %tid.ext
   %vec = load <4 x half>, <4 x half> addrspace(1)* %in.gep
   %val.trunc = trunc i32 %val to i16
   %val.cvt = bitcast i16 %val.trunc to half
   %vecins = insertelement <4 x half> %vec, half %val.cvt, i32 0
   store <4 x half> %vecins, <4 x half> addrspace(1)* %out.gep
   ret void
 }
 
 define amdgpu_kernel void @v_insertelement_v4f16_1(<4 x half> addrspace(1)* %out, <4 x half> addrspace(1)* %in, i32 %val) #0 {
 ; GFX9-LABEL: v_insertelement_v4f16_1:
 ; GFX9:       ; %bb.0:
 ; 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
 ;
 ; VI-LABEL: v_insertelement_v4f16_1:
 ; VI:       ; %bb.0:
 ; VI-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; VI-NEXT:    s_load_dword s4, s[4:5], 0x10
 ; VI-NEXT:    v_lshlrev_b32_e32 v2, 3, v0

 ; GFX11-LABEL: v_insertelement_v4f16_1:
 ; GFX11:       ; %bb.0:
 ; GFX11-NEXT:    s_load_b128 s[4:7], s[0:1], 0x0
 ; GFX11-NEXT:    v_lshlrev_b32_e32 v2, 3, v0
 ; GFX11-NEXT:    s_load_b32 s0, s[0:1], 0x10
 ; 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
   %tid = call i32 @llvm.amdgcn.workitem.id.x() #1
   %tid.ext = sext i32 %tid to i64
   %in.gep = getelementptr inbounds <4 x half>, <4 x half> addrspace(1)* %in, i64 %tid.ext
   %out.gep = getelementptr inbounds <4 x half>, <4 x half> addrspace(1)* %out, i64 %tid.ext
   %vec = load <4 x half>, <4 x half> addrspace(1)* %in.gep
   %val.trunc = trunc i32 %val to i16
   %val.cvt = bitcast i16 %val.trunc to half
   %vecins = insertelement <4 x half> %vec, half %val.cvt, i32 1
   store <4 x half> %vecins, <4 x half> addrspace(1)* %out.gep
   ret void
 }
 
 define amdgpu_kernel void @v_insertelement_v4f16_2(<4 x half> addrspace(1)* %out, <4 x half> addrspace(1)* %in, [8 x i32], i32 %val) #0 {
 ; GFX9-LABEL: v_insertelement_v4f16_2:
 ; GFX9:       ; %bb.0:
 ; 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
 ;
 ; VI-LABEL: v_insertelement_v4f16_2:
 ; VI:       ; %bb.0:
 ; VI-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; VI-NEXT:    s_load_dword s4, s[4:5], 0x30
 ; VI-NEXT:    v_lshlrev_b32_e32 v2, 3, v0

 ; GFX11-LABEL: v_insertelement_v4f16_2:
 ; GFX11:       ; %bb.0:
 ; GFX11-NEXT:    s_load_b128 s[4:7], s[0:1], 0x0
 ; GFX11-NEXT:    v_lshlrev_b32_e32 v2, 3, v0
 ; GFX11-NEXT:    s_load_b32 s0, s[0:1], 0x30
 ; 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
   %tid = call i32 @llvm.amdgcn.workitem.id.x() #1
   %tid.ext = sext i32 %tid to i64
   %in.gep = getelementptr inbounds <4 x half>, <4 x half> addrspace(1)* %in, i64 %tid.ext
   %out.gep = getelementptr inbounds <4 x half>, <4 x half> addrspace(1)* %out, i64 %tid.ext
   %vec = load <4 x half>, <4 x half> addrspace(1)* %in.gep
   %val.trunc = trunc i32 %val to i16
   %val.cvt = bitcast i16 %val.trunc to half
   %vecins = insertelement <4 x half> %vec, half %val.cvt, i32 2
   store <4 x half> %vecins, <4 x half> addrspace(1)* %out.gep
   ret void
 }
 
 define amdgpu_kernel void @v_insertelement_v4f16_3(<4 x half> addrspace(1)* %out, <4 x half> addrspace(1)* %in, i32 %val) #0 {
 ; GFX9-LABEL: v_insertelement_v4f16_3:
 ; GFX9:       ; %bb.0:
 ; 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
 ;
 ; VI-LABEL: v_insertelement_v4f16_3:
 ; VI:       ; %bb.0:
 ; VI-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; VI-NEXT:    s_load_dword s4, s[4:5], 0x10
 ; VI-NEXT:    v_lshlrev_b32_e32 v2, 3, v0

 ; GFX11-LABEL: v_insertelement_v4f16_3:
 ; GFX11:       ; %bb.0:
 ; GFX11-NEXT:    s_load_b128 s[4:7], s[0:1], 0x0
 ; GFX11-NEXT:    v_lshlrev_b32_e32 v2, 3, v0
 ; GFX11-NEXT:    s_load_b32 s0, s[0:1], 0x10
 ; 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
   %tid = call i32 @llvm.amdgcn.workitem.id.x() #1
   %tid.ext = sext i32 %tid to i64
   %in.gep = getelementptr inbounds <4 x half>, <4 x half> addrspace(1)* %in, i64 %tid.ext
   %out.gep = getelementptr inbounds <4 x half>, <4 x half> addrspace(1)* %out, i64 %tid.ext
   %vec = load <4 x half>, <4 x half> addrspace(1)* %in.gep
   %val.trunc = trunc i32 %val to i16
   %val.cvt = bitcast i16 %val.trunc to half
   %vecins = insertelement <4 x half> %vec, half %val.cvt, i32 3
   store <4 x half> %vecins, <4 x half> addrspace(1)* %out.gep
   ret void
 }
 
 define amdgpu_kernel void @v_insertelement_v4i16_2(<4 x i16> addrspace(1)* %out, <4 x i16> addrspace(1)* %in, i32 %val) #0 {
 ; GFX9-LABEL: v_insertelement_v4i16_2:
 ; GFX9:       ; %bb.0:
 ; 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
 ;
 ; VI-LABEL: v_insertelement_v4i16_2:
 ; VI:       ; %bb.0:
 ; VI-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; VI-NEXT:    s_load_dword s4, s[4:5], 0x10
 ; VI-NEXT:    v_lshlrev_b32_e32 v2, 3, v0

 ; GFX11-LABEL: v_insertelement_v4i16_2:
 ; GFX11:       ; %bb.0:
 ; GFX11-NEXT:    s_load_b128 s[4:7], s[0:1], 0x0
 ; GFX11-NEXT:    v_lshlrev_b32_e32 v2, 3, v0
 ; GFX11-NEXT:    s_load_b32 s0, s[0:1], 0x10
 ; 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
   %tid = call i32 @llvm.amdgcn.workitem.id.x() #1
   %tid.ext = sext i32 %tid to i64
   %in.gep = getelementptr inbounds <4 x i16>, <4 x i16> addrspace(1)* %in, i64 %tid.ext
   %out.gep = getelementptr inbounds <4 x i16>, <4 x i16> addrspace(1)* %out, i64 %tid.ext
   %vec = load <4 x i16>, <4 x i16> addrspace(1)* %in.gep

 }
 
 define amdgpu_kernel void @v_insertelement_v8f16_3(<8 x half> addrspace(1)* %out, <8 x half> addrspace(1)* %in, i32 %val) {
 ; GFX9-LABEL: v_insertelement_v8f16_3:
 ; GFX9:       ; %bb.0:
 ; 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
 ;
 ; VI-LABEL: v_insertelement_v8f16_3:
 ; VI:       ; %bb.0:
 ; VI-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; VI-NEXT:    s_load_dword s4, s[4:5], 0x10
 ; VI-NEXT:    v_lshlrev_b32_e32 v4, 4, v0

 ; GFX11-LABEL: v_insertelement_v8f16_3:
 ; GFX11:       ; %bb.0:
 ; GFX11-NEXT:    s_load_b128 s[4:7], s[0:1], 0x0
 ; GFX11-NEXT:    v_lshlrev_b32_e32 v4, 4, v0
 ; GFX11-NEXT:    s_load_b32 s0, 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_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
   %tid = call i32 @llvm.amdgcn.workitem.id.x() #1
   %tid.ext = sext i32 %tid to i64
   %in.gep = getelementptr inbounds <8 x half>, <8 x half> addrspace(1)* %in, i64 %tid.ext
   %out.gep = getelementptr inbounds <8 x half>, <8 x half> addrspace(1)* %out, i64 %tid.ext
   %vec = load <8 x half>, <8 x half> addrspace(1)* %in.gep
   %val.trunc = trunc i32 %val to i16
   %val.cvt = bitcast i16 %val.trunc to half
   %vecins = insertelement <8 x half> %vec, half %val.cvt, i32 3
   store <8 x half> %vecins, <8 x half> addrspace(1)* %out.gep
   ret void
 }
 
 define amdgpu_kernel void @v_insertelement_v8i16_6(<8 x i16> addrspace(1)* %out, <8 x i16> addrspace(1)* %in, i32 %val) {
 ; GFX9-LABEL: v_insertelement_v8i16_6:
 ; GFX9:       ; %bb.0:
 ; 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
 ;
 ; VI-LABEL: v_insertelement_v8i16_6:
 ; VI:       ; %bb.0:
 ; VI-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; VI-NEXT:    s_load_dword s4, s[4:5], 0x10
 ; VI-NEXT:    v_lshlrev_b32_e32 v4, 4, v0

 ; GFX11-LABEL: v_insertelement_v8i16_6:
 ; GFX11:       ; %bb.0:
 ; GFX11-NEXT:    s_load_b128 s[4:7], s[0:1], 0x0
 ; GFX11-NEXT:    v_lshlrev_b32_e32 v4, 4, v0
 ; GFX11-NEXT:    s_load_b32 s0, 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_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
   %tid = call i32 @llvm.amdgcn.workitem.id.x() #1
   %tid.ext = sext i32 %tid to i64
   %in.gep = getelementptr inbounds <8 x i16>, <8 x i16> addrspace(1)* %in, i64 %tid.ext
   %out.gep = getelementptr inbounds <8 x i16>, <8 x i16> addrspace(1)* %out, i64 %tid.ext
   %vec = load <8 x i16>, <8 x i16> addrspace(1)* %in.gep
   %val.trunc = trunc i32 %val to i16
   %val.cvt = bitcast i16 %val.trunc to i16
   %vecins = insertelement <8 x i16> %vec, i16 %val.cvt, i32 6
   store <8 x i16> %vecins, <8 x i16> addrspace(1)* %out.gep
   ret void
 }
 
 define amdgpu_kernel void @v_insertelement_v8f16_dynamic(<8 x half> addrspace(1)* %out, <8 x half> addrspace(1)* %in, i32 %val, i32 %n) {
 ; GFX9-LABEL: v_insertelement_v8f16_dynamic:
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX9-NEXT:    s_load_dwordx2 s[6:7], s[4:5], 0x10
 ; 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
 ;
 ; VI-LABEL: v_insertelement_v8f16_dynamic:
 ; VI:       ; %bb.0:
 ; VI-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; VI-NEXT:    s_load_dwordx2 s[4:5], s[4:5], 0x10
 ; VI-NEXT:    v_lshlrev_b32_e32 v4, 4, v0

 ;
 ; GFX11-LABEL: v_insertelement_v8f16_dynamic:
 ; GFX11:       ; %bb.0:
 ; GFX11-NEXT:    s_load_b128 s[4:7], s[0:1], 0x0
 ; GFX11-NEXT:    v_lshlrev_b32_e32 v4, 4, v0
 ; 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
   %tid = call i32 @llvm.amdgcn.workitem.id.x() #1
   %tid.ext = sext i32 %tid to i64
   %in.gep = getelementptr inbounds <8 x half>, <8 x half> addrspace(1)* %in, i64 %tid.ext
   %out.gep = getelementptr inbounds <8 x half>, <8 x half> addrspace(1)* %out, i64 %tid.ext
   %vec = load <8 x half>, <8 x half> addrspace(1)* %in.gep
   %val.trunc = trunc i32 %val to i16
   %val.cvt = bitcast i16 %val.trunc to half
   %vecins = insertelement <8 x half> %vec, half %val.cvt, i32 %n
   store <8 x half> %vecins, <8 x half> addrspace(1)* %out.gep
   ret void
 }
 
 define amdgpu_kernel void @v_insertelement_v16f16_3(<16 x half> addrspace(1)* %out, <16 x half> addrspace(1)* %in, i32 %val) {
 ; GFX9-LABEL: v_insertelement_v16f16_3:
 ; GFX9:       ; %bb.0:
 ; 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]
 ; GFX9-NEXT:    s_endpgm
 ;
 ; VI-LABEL: v_insertelement_v16f16_3:
 ; VI:       ; %bb.0:
 ; VI-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX11-NEXT:    s_load_b128 s[4:7], s[0:1], 0x0
 ; GFX11-NEXT:    v_lshlrev_b32_e32 v8, 5, v0
 ; GFX11-NEXT:    s_load_b32 s0, s[0:1], 0x10
 ; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
 ; 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_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
 ; GFX11-NEXT:    global_store_b128 v8, v[0:3], s[4:5]
 ; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-NEXT:    s_endpgm
   %tid = call i32 @llvm.amdgcn.workitem.id.x() #1
   %tid.ext = sext i32 %tid to i64
   %in.gep = getelementptr inbounds <16 x half>, <16 x half> addrspace(1)* %in, i64 %tid.ext
   %out.gep = getelementptr inbounds <16 x half>, <16 x half> addrspace(1)* %out, i64 %tid.ext
   %vec = load <16 x half>, <16 x half> addrspace(1)* %in.gep
   %val.trunc = trunc i32 %val to i16
   %val.cvt = bitcast i16 %val.trunc to half
   %vecins = insertelement <16 x half> %vec, half %val.cvt, i32 3
   store <16 x half> %vecins, <16 x half> addrspace(1)* %out.gep
   ret void
 }
 
 define amdgpu_kernel void @v_insertelement_v16i16_6(<16 x i16> addrspace(1)* %out, <16 x i16> addrspace(1)* %in, i32 %val) {
 ; GFX9-LABEL: v_insertelement_v16i16_6:
 ; GFX9:       ; %bb.0:
 ; 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]
 ; GFX9-NEXT:    s_endpgm
 ;
 ; VI-LABEL: v_insertelement_v16i16_6:
 ; VI:       ; %bb.0:
 ; VI-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX11-NEXT:    s_load_b128 s[4:7], s[0:1], 0x0
 ; GFX11-NEXT:    v_lshlrev_b32_e32 v8, 5, v0
 ; GFX11-NEXT:    s_load_b32 s0, s[0:1], 0x10
 ; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
 ; 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_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
 ; GFX11-NEXT:    global_store_b128 v8, v[0:3], s[4:5]
 ; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-NEXT:    s_endpgm
   %tid = call i32 @llvm.amdgcn.workitem.id.x() #1
   %tid.ext = sext i32 %tid to i64

 ; GFX9-LABEL: v_insertelement_v16f16_dynamic:
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX9-NEXT:    s_load_dwordx2 s[6:7], s[4:5], 0x10
 ; GFX9-NEXT:    v_lshlrev_b32_e32 v8, 5, v0
 ; 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
 ;
 ; VI-LABEL: v_insertelement_v16f16_dynamic:
 ; VI:       ; %bb.0:
 ; VI-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; VI-NEXT:    s_load_dwordx2 s[6:7], s[4:5], 0x10

 ; GFX11:       ; %bb.0:
 ; GFX11-NEXT:    s_load_b128 s[4:7], s[0:1], 0x0
 ; GFX11-NEXT:    v_lshlrev_b32_e32 v8, 5, v0
 ; GFX11-NEXT:    s_load_b64 s[0:1], s[0:1], 0x10
 ; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
 ; 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]
 ; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-NEXT:    s_endpgm
   %tid = call i32 @llvm.amdgcn.workitem.id.x() #1
   %tid.ext = sext i32 %tid to i64
   %in.gep = getelementptr inbounds <16 x half>, <16 x half> addrspace(1)* %in, i64 %tid.ext

llvm/test/CodeGen/AMDGPU/llvm.amdgcn.image.gather4.a16.dim.ll

 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc -march=amdgcn -mcpu=gfx900 -verify-machineinstrs < %s | FileCheck -check-prefixes=GFX9 %s
 ; RUN: llc -march=amdgcn -mcpu=gfx1010 -verify-machineinstrs < %s | FileCheck -check-prefixes=GFX10 %s
 ; RUN: llc -march=amdgcn -mcpu=gfx1100 -amdgpu-enable-delay-alu=0 -verify-machineinstrs < %s | FileCheck -check-prefixes=GFX10 %s
 
 define amdgpu_ps <4 x float> @gather4_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %s, half %t) {
 ; GFX9-LABEL: gather4_2d:
 ; 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)
 ; GFX9-NEXT:    ; return to shader part epilog
 ;
 ; GFX10-LABEL: gather4_2d:
 ; 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)
 ; GFX10-NEXT:    ; return to shader part epilog
 main_body:
   %v = call <4 x float> @llvm.amdgcn.image.gather4.2d.v4f32.f16(i32 1, half %s, half %t, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
 
 define amdgpu_ps <4 x float> @gather4_cube(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %s, half %t, half %face) {
 ; GFX9-LABEL: gather4_cube:
 ; 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)
 ; GFX9-NEXT:    ; return to shader part epilog
 ;
 ; GFX10-LABEL: gather4_cube:
 ; 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)
 ; GFX10-NEXT:    ; return to shader part epilog
 main_body:
   %v = call <4 x float> @llvm.amdgcn.image.gather4.cube.v4f32.f16(i32 1, half %s, half %t, half %face, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
 
 define amdgpu_ps <4 x float> @gather4_2darray(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %s, half %t, half %slice) {
 ; GFX9-LABEL: gather4_2darray:
 ; 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)
 ; GFX9-NEXT:    ; return to shader part epilog
 ;
 ; GFX10-LABEL: gather4_2darray:
 ; 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)
 ; GFX10-NEXT:    ; return to shader part epilog
 main_body:
   %v = call <4 x float> @llvm.amdgcn.image.gather4.2darray.v4f32.f16(i32 1, half %s, half %t, half %slice, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
 
 define amdgpu_ps <4 x float> @gather4_c_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %zcompare, half %s, half %t) {
 ; GFX9-LABEL: gather4_c_2d:
 ; 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)
 ; GFX9-NEXT:    ; return to shader part epilog
 ;
 ; GFX10-LABEL: gather4_c_2d:
 ; 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)
 ; GFX10-NEXT:    ; return to shader part epilog
 main_body:
   %v = call <4 x float> @llvm.amdgcn.image.gather4.c.2d.v4f32.f32(i32 1, float %zcompare, half %s, half %t, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
 
 define amdgpu_ps <4 x float> @gather4_cl_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %s, half %t, half %clamp) {
 ; GFX9-LABEL: gather4_cl_2d:
 ; 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)
 ; GFX9-NEXT:    ; return to shader part epilog
 ;
 ; GFX10-LABEL: gather4_cl_2d:
 ; 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)
 ; GFX10-NEXT:    ; return to shader part epilog
 main_body:
   %v = call <4 x float> @llvm.amdgcn.image.gather4.cl.2d.v4f32.f16(i32 1, half %s, half %t, half %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
 
 define amdgpu_ps <4 x float> @gather4_c_cl_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %zcompare, half %s, half %t, half %clamp) {
 ; GFX9-LABEL: gather4_c_cl_2d:
 ; 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)
 ; GFX9-NEXT:    ; return to shader part epilog
 ;
 ; GFX10-LABEL: gather4_c_cl_2d:
 ; 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)
 ; GFX10-NEXT:    ; return to shader part epilog
 main_body:
   %v = call <4 x float> @llvm.amdgcn.image.gather4.c.cl.2d.v4f32.f32(i32 1, float %zcompare, half %s, half %t, half %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
 
 define amdgpu_ps <4 x float> @gather4_b_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %bias, half %s, half %t) {
 ; GFX9-LABEL: gather4_b_2d:
 ; 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)
 ; GFX9-NEXT:    ; return to shader part epilog
 ;
 ; GFX10-LABEL: gather4_b_2d:
 ; 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)
 ; GFX10-NEXT:    ; return to shader part epilog
 main_body:
   %v = call <4 x float> @llvm.amdgcn.image.gather4.b.2d.v4f32.f16.f16(i32 1, half %bias, half %s, half %t, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
 
 define amdgpu_ps <4 x float> @gather4_c_b_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %bias, float %zcompare, half %s, half %t) {
 ; GFX9-LABEL: gather4_c_b_2d:
 ; 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)
 ; GFX9-NEXT:    ; return to shader part epilog
 ;
 ; GFX10-LABEL: gather4_c_b_2d:
 ; 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)
 ; GFX10-NEXT:    ; return to shader part epilog
 main_body:
   %v = call <4 x float> @llvm.amdgcn.image.gather4.c.b.2d.v4f32.f16.f16(i32 1, half %bias, float %zcompare, half %s, half %t, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
 
 define amdgpu_ps <4 x float> @gather4_b_cl_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %bias, half %s, half %t, half %clamp) {
 ; GFX9-LABEL: gather4_b_cl_2d:
 ; 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)
 ; GFX9-NEXT:    ; return to shader part epilog
 ;
 ; GFX10-LABEL: gather4_b_cl_2d:
 ; 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)
 ; GFX10-NEXT:    ; return to shader part epilog
 main_body:
   %v = call <4 x float> @llvm.amdgcn.image.gather4.b.cl.2d.v4f32.f16.f16(i32 1, half %bias, half %s, half %t, half %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
 
 define amdgpu_ps <4 x float> @gather4_c_b_cl_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %bias, float %zcompare, half %s, half %t, half %clamp) {
 ; GFX9-LABEL: gather4_c_b_cl_2d:
 ; 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)
 ; GFX9-NEXT:    ; return to shader part epilog
 ;
 ; GFX10-LABEL: gather4_c_b_cl_2d:
 ; 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)
 ; GFX10-NEXT:    ; return to shader part epilog
 main_body:
   %v = call <4 x float> @llvm.amdgcn.image.gather4.c.b.cl.2d.v4f32.f16.f16(i32 1, half %bias, float %zcompare, half %s, half %t, half %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
 
 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
 main_body:
   %v = call <4 x float> @llvm.amdgcn.image.gather4.l.2d.v4f32.f16(i32 1, half %s, half %t, half %lod, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
 
 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
 main_body:
   %v = call <4 x float> @llvm.amdgcn.image.gather4.c.l.2d.v4f32.f32(i32 1, float %zcompare, half %s, half %t, half %lod, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
 
 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
 main_body:
   %v = call <4 x float> @llvm.amdgcn.image.gather4.lz.2d.v4f32.f16(i32 1, half %s, half %t, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
 
 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
 main_body:
   %v = call <4 x float> @llvm.amdgcn.image.gather4.c.lz.2d.v4f32.f32(i32 1, float %zcompare, half %s, half %t, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
 

llvm/test/CodeGen/AMDGPU/llvm.amdgcn.image.sample.a16.dim.ll

main_body:
   ret <4 x float> %v
 }
 
 define amdgpu_ps <4 x float> @sample_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %s, half %t) {
 ; GFX9-LABEL: sample_2d:
 ; 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)
 ; GFX9-NEXT:    ; return to shader part epilog
 ;
 ; GFX10-LABEL: sample_2d:
 ; 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)
 ; GFX10-NEXT:    ; return to shader part epilog
 main_body:
   %v = call <4 x float> @llvm.amdgcn.image.sample.2d.v4f32.f16(i32 15, half %s, half %t, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
 
 define amdgpu_ps <4 x float> @sample_3d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %s, half %t, half %r) {
 ; GFX9-LABEL: sample_3d:
 ; 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)
 ; GFX9-NEXT:    ; return to shader part epilog
 ;
 ; GFX10-LABEL: sample_3d:
 ; 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)
 ; GFX10-NEXT:    ; return to shader part epilog
 main_body:
   %v = call <4 x float> @llvm.amdgcn.image.sample.3d.v4f32.f16(i32 15, half %s, half %t, half %r, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
 
 define amdgpu_ps <4 x float> @sample_cube(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %s, half %t, half %face) {
 ; GFX9-LABEL: sample_cube:
 ; 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)
 ; GFX9-NEXT:    ; return to shader part epilog
 ;
 ; GFX10-LABEL: sample_cube:
 ; 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)
 ; GFX10-NEXT:    ; return to shader part epilog
 main_body:
   %v = call <4 x float> @llvm.amdgcn.image.sample.cube.v4f32.f16(i32 15, half %s, half %t, half %face, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
 
 define amdgpu_ps <4 x float> @sample_1darray(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %s, half %slice) {
 ; GFX9-LABEL: sample_1darray:
 ; 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)
 ; GFX9-NEXT:    ; return to shader part epilog
 ;
 ; GFX10-LABEL: sample_1darray:
 ; 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)
 ; GFX10-NEXT:    ; return to shader part epilog
 main_body:
   %v = call <4 x float> @llvm.amdgcn.image.sample.1darray.v4f32.f16(i32 15, half %s, half %slice, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
 
 define amdgpu_ps <4 x float> @sample_2darray(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %s, half %t, half %slice) {
 ; GFX9-LABEL: sample_2darray:
 ; 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)
 ; GFX9-NEXT:    ; return to shader part epilog
 ;
 ; GFX10-LABEL: sample_2darray:
 ; 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)
 ; GFX10-NEXT:    ; return to shader part epilog
 main_body:
   %v = call <4 x float> @llvm.amdgcn.image.sample.2darray.v4f32.f16(i32 15, half %s, half %t, half %slice, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
main_body:
   ret <4 x float> %v
 }
 
 define amdgpu_ps <4 x float> @sample_c_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %zcompare, half %s, half %t) {
 ; GFX9-LABEL: sample_c_2d:
 ; 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)
 ; GFX9-NEXT:    ; return to shader part epilog
 ;
 ; GFX10-LABEL: sample_c_2d:
 ; 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)
 ; GFX10-NEXT:    ; return to shader part epilog
 main_body:
   %v = call <4 x float> @llvm.amdgcn.image.sample.c.2d.v4f32.f16(i32 15, float %zcompare, half %s, half %t, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
 
 define amdgpu_ps <4 x float> @sample_cl_1d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %s, half %clamp) {
 ; GFX9-LABEL: sample_cl_1d:
 ; 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)
 ; GFX9-NEXT:    ; return to shader part epilog
 ;
 ; GFX10-LABEL: sample_cl_1d:
 ; 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)
 ; GFX10-NEXT:    ; return to shader part epilog
 main_body:
   %v = call <4 x float> @llvm.amdgcn.image.sample.cl.1d.v4f32.f16(i32 15, half %s, half %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
 
 define amdgpu_ps <4 x float> @sample_cl_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %s, half %t, half %clamp) {
 ; GFX9-LABEL: sample_cl_2d:
 ; 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)
 ; GFX9-NEXT:    ; return to shader part epilog
 ;
 ; GFX10-LABEL: sample_cl_2d:
 ; 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)
 ; GFX10-NEXT:    ; return to shader part epilog
 main_body:
   %v = call <4 x float> @llvm.amdgcn.image.sample.cl.2d.v4f32.f16(i32 15, half %s, half %t, half %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
 
 define amdgpu_ps <4 x float> @sample_c_cl_1d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %zcompare, half %s, half %clamp) {
 ; GFX9-LABEL: sample_c_cl_1d:
 ; 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)
 ; GFX9-NEXT:    ; return to shader part epilog
 ;
 ; GFX10-LABEL: sample_c_cl_1d:
 ; 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)
 ; GFX10-NEXT:    ; return to shader part epilog
 main_body:
   %v = call <4 x float> @llvm.amdgcn.image.sample.c.cl.1d.v4f32.f16(i32 15, float %zcompare, half %s, half %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
 
 define amdgpu_ps <4 x float> @sample_c_cl_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %zcompare, half %s, half %t, half %clamp) {
 ; GFX9-LABEL: sample_c_cl_2d:
 ; 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)
 ; GFX9-NEXT:    ; return to shader part epilog
 ;
 ; GFX10-LABEL: sample_c_cl_2d:
 ; 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)
 ; GFX10-NEXT:    ; return to shader part epilog
 main_body:
   %v = call <4 x float> @llvm.amdgcn.image.sample.c.cl.2d.v4f32.f16(i32 15, float %zcompare, half %s, half %t, half %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
main_body:
   ret <4 x float> %v
 }
 
 define amdgpu_ps <4 x float> @sample_b_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %bias, half %s, half %t) {
 ; GFX9-LABEL: sample_b_2d:
 ; 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)
 ; GFX9-NEXT:    ; return to shader part epilog
 ;
 ; GFX10-LABEL: sample_b_2d:
 ; 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)
 ; GFX10-NEXT:    ; return to shader part epilog
 main_body:
   %v = call <4 x float> @llvm.amdgcn.image.sample.b.2d.v4f32.f16.f16(i32 15, half %bias, half %s, half %t, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
main_body:
   ret <4 x float> %v
 }
 
 define amdgpu_ps <4 x float> @sample_c_b_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %bias, float %zcompare, half %s, half %t) {
 ; GFX9-LABEL: sample_c_b_2d:
 ; 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)
 ; GFX9-NEXT:    ; return to shader part epilog
 ;
 ; GFX10-LABEL: sample_c_b_2d:
 ; 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)
 ; GFX10-NEXT:    ; return to shader part epilog
 main_body:
   %v = call <4 x float> @llvm.amdgcn.image.sample.c.b.2d.v4f32.f16.f16(i32 15, half %bias, float %zcompare, half %s, half %t, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
 
 define amdgpu_ps <4 x float> @sample_b_cl_1d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %bias, half %s, half %clamp) {
 ; GFX9-LABEL: sample_b_cl_1d:
 ; 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)
 ; GFX9-NEXT:    ; return to shader part epilog
 ;
 ; GFX10-LABEL: sample_b_cl_1d:
 ; 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)
 ; GFX10-NEXT:    ; return to shader part epilog
 main_body:
   %v = call <4 x float> @llvm.amdgcn.image.sample.b.cl.1d.v4f32.f16.f16(i32 15, half %bias, half %s, half %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
 
 define amdgpu_ps <4 x float> @sample_b_cl_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %bias, half %s, half %t, half %clamp) {
 ; GFX9-LABEL: sample_b_cl_2d:
 ; 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)
 ; GFX9-NEXT:    ; return to shader part epilog
 ;
 ; GFX10-LABEL: sample_b_cl_2d:
 ; 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)
 ; GFX10-NEXT:    ; return to shader part epilog
 main_body:
   %v = call <4 x float> @llvm.amdgcn.image.sample.b.cl.2d.v4f32.f16.f16(i32 15, half %bias, half %s, half %t, half %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
 
 define amdgpu_ps <4 x float> @sample_c_b_cl_1d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %bias, float %zcompare, half %s, half %clamp) {
 ; GFX9-LABEL: sample_c_b_cl_1d:
 ; 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)
 ; GFX9-NEXT:    ; return to shader part epilog
 ;
 ; GFX10-LABEL: sample_c_b_cl_1d:
 ; 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)
 ; GFX10-NEXT:    ; return to shader part epilog
 main_body:
   %v = call <4 x float> @llvm.amdgcn.image.sample.c.b.cl.1d.v4f32.f16.f16(i32 15, half %bias, float %zcompare, half %s, half %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
 
 define amdgpu_ps <4 x float> @sample_c_b_cl_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %bias, float %zcompare, half %s, half %t, half %clamp) {
 ; GFX9-LABEL: sample_c_b_cl_2d:
 ; 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)
 ; GFX9-NEXT:    ; return to shader part epilog
 ;
 ; GFX10-LABEL: sample_c_b_cl_2d:
 ; 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)
 ; GFX10-NEXT:    ; return to shader part epilog
 main_body:
   %v = call <4 x float> @llvm.amdgcn.image.sample.c.b.cl.2d.v4f32.f16.f16(i32 15, half %bias, float %zcompare, half %s, half %t, half %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }

 main_body:
   %v = call <4 x float> @llvm.amdgcn.image.sample.d.1d.v4f32.f16.f16(i32 15, half %dsdh, half %dsdv, half %s, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
 
 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
 main_body:
   %v = call <4 x float> @llvm.amdgcn.image.sample.d.2d.v4f32.f16.f16(i32 15, half %dsdh, half %dtdh, half %dsdv, half %dtdv, half %s, half %t, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
 
 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
 ;
 ; 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
 main_body:
   %v = call <4 x float> @llvm.amdgcn.image.sample.d.3d.v4f32.f16.f16(i32 15, half %dsdh, half %dtdh, half %drdh, half %dsdv, half %dtdv, half %drdv, half %s, half %t, half %r, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
 
main_body:
   ret <4 x float> %v
 }
 
 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, half %s, half %t) {
 ; GFX9-LABEL: sample_c_d_2d:
 ; 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
 main_body:
   %v = call <4 x float> @llvm.amdgcn.image.sample.c.d.2d.v4f32.f32.f16(i32 15, float %zcompare, half %dsdh, half %dtdh, half %dsdv, half %dtdv, half %s, half %t, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
 
 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
 main_body:
   %v = call <4 x float> @llvm.amdgcn.image.sample.d.cl.1d.v4f32.f16.f16(i32 15, half %dsdh, half %dsdv, half %s, half %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
 
 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
 main_body:
   %v = call <4 x float> @llvm.amdgcn.image.sample.d.cl.2d.v4f32.f16.f16(i32 15, half %dsdh, half %dtdh, half %dsdv, half %dtdv, half %s, half %t, half %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
 
 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
 main_body:
   %v = call <4 x float> @llvm.amdgcn.image.sample.c.d.cl.1d.v4f32.f32.f16(i32 15, float %zcompare, half %dsdh, half %dsdv, half %s, half %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
 
 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
 main_body:
   %v = call <4 x float> @llvm.amdgcn.image.sample.c.d.cl.2d.v4f32.f32.f16(i32 15, float %zcompare, half %dsdh, half %dtdh, half %dsdv, half %dtdv, half %s, half %t, half %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
 
 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
 main_body:
   %v = call <4 x float> @llvm.amdgcn.image.sample.l.1d.v4f32.f16(i32 15, half %s, half %lod, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
 
 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
 main_body:
   %v = call <4 x float> @llvm.amdgcn.image.sample.l.2d.v4f32.f16(i32 15, half %s, half %t, half %lod, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
 
 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
 main_body:
   %v = call <4 x float> @llvm.amdgcn.image.sample.c.l.1d.v4f32.f16(i32 15, float %zcompare, half %s, half %lod, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
 
 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
 main_body:
   %v = call <4 x float> @llvm.amdgcn.image.sample.c.l.2d.v4f32.f16(i32 15, float %zcompare, half %s, half %t, half %lod, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
 

 main_body:
   %v = call <4 x float> @llvm.amdgcn.image.sample.lz.1d.v4f32.f16(i32 15, half %s, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
 
 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
 main_body:
   %v = call <4 x float> @llvm.amdgcn.image.sample.lz.2d.v4f32.f16(i32 15, half %s, half %t, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
 

 main_body:
   %v = call <4 x float> @llvm.amdgcn.image.sample.c.lz.1d.v4f32.f16(i32 15, float %zcompare, half %s, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
 
 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
 main_body:
   %v = call <4 x float> @llvm.amdgcn.image.sample.c.lz.2d.v4f32.f16(i32 15, float %zcompare, half %s, half %t, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
 
 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
 ;
 ; GFX10-LABEL: sample_c_d_o_2darray_V1:
 ; GFX10:       ; %bb.0: ; %main_body
 ; 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
 main_body:
   %v = call float @llvm.amdgcn.image.sample.c.d.o.2darray.f32.f16.f16(i32 4, i32 %offset, float %zcompare, half %dsdh, half %dtdh, half %dsdv, half %dtdv, half %s, half %t, half %slice, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret float %v
 }
 
 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
 ;
 ; GFX10-LABEL: sample_c_d_o_2darray_V2:
 ; GFX10:       ; %bb.0: ; %main_body
 ; 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
 main_body:
   %v = call <2 x float> @llvm.amdgcn.image.sample.c.d.o.2darray.v2f32.f32.f16(i32 6, i32 %offset, float %zcompare, half %dsdh, half %dtdh, half %dsdv, half %dtdv, half %s, half %t, half %slice, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <2 x float> %v
 }
 

llvm/test/CodeGen/AMDGPU/llvm.amdgcn.image.sample.cd.a16.dim.ll

 main_body:
   %v = call <4 x float> @llvm.amdgcn.image.sample.cd.1d.v4f32.f16.f16(i32 15, half %dsdh, half %dsdv, half %s, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
 
 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
 main_body:
   %v = call <4 x float> @llvm.amdgcn.image.sample.cd.2d.v4f32.f16.f16(i32 15, half %dsdh, half %dtdh, half %dsdv, half %dtdv, half %s, half %t, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
 
main_body:
   ret <4 x float> %v
 }
 
 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, half %s, half %t) {
 ; GFX9-LABEL: sample_c_cd_2d:
 ; 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
 main_body:
   %v = call <4 x float> @llvm.amdgcn.image.sample.c.cd.2d.v4f32.f32.f16(i32 15, float %zcompare, half %dsdh, half %dtdh, half %dsdv, half %dtdv, half %s, half %t, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
 
 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
 main_body:
   %v = call <4 x float> @llvm.amdgcn.image.sample.cd.cl.1d.v4f32.f16.f16(i32 15, half %dsdh, half %dsdv, half %s, half %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
 
 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
 main_body:
   %v = call <4 x float> @llvm.amdgcn.image.sample.cd.cl.2d.v4f32.f16.f16(i32 15, half %dsdh, half %dtdh, half %dsdv, half %dtdv, half %s, half %t, half %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
 
 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
 main_body:
   %v = call <4 x float> @llvm.amdgcn.image.sample.c.cd.cl.1d.v4f32.f32.f16(i32 15, float %zcompare, half %dsdh, half %dsdv, half %s, half %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
 
 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
 main_body:
   %v = call <4 x float> @llvm.amdgcn.image.sample.c.cd.cl.2d.v4f32.f32.f16(i32 15, float %zcompare, half %dsdh, half %dtdh, half %dsdv, half %dtdv, half %s, half %t, half %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
 

llvm/test/CodeGen/AMDGPU/llvm.amdgcn.image.sample.cd.g16.encode.ll

 main_body:
   %v = call <4 x float> @llvm.amdgcn.image.sample.cd.1d.v4f32.f16.f32(i32 15, half %dsdh, half %dsdv, float %s, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
 
 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
 main_body:
   %v = call <4 x float> @llvm.amdgcn.image.sample.cd.2d.v4f32.f16.f32(i32 15, half %dsdh, half %dtdh, half %dsdv, half %dtdv, float %s, float %t, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
 
 define amdgpu_ps <4 x float> @sample_c_cd_1d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %zcompare, half %dsdh, half %dsdv, float %s) {
 ; GFX10-LABEL: sample_c_cd_1d:
 ; GFX10:       ; %bb.0: ; %main_body
 ; GFX10-NEXT:    image_sample_c_cd_g16 v[0:3], v[0:3], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_1D ; encoding: [0x01,0x0f,0xa8,0xf1,0x00,0x00,0x40,0x00]
 ; GFX10-NEXT:    s_waitcnt vmcnt(0) ; encoding: [0x70,0x3f,0x8c,0xbf]
 ; GFX10-NEXT:    ; return to shader part epilog
 main_body:
   %v = call <4 x float> @llvm.amdgcn.image.sample.c.cd.1d.v4f32.f16.f32(i32 15, float %zcompare, half %dsdh, half %dsdv, float %s, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
 
 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
 main_body:
   %v = call <4 x float> @llvm.amdgcn.image.sample.c.cd.2d.v4f32.f16.f32(i32 15, float %zcompare, half %dsdh, half %dtdh, half %dsdv, half %dtdv, float %s, float %t, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
 
 define amdgpu_ps <4 x float> @sample_cd_cl_1d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %dsdh, half %dsdv, float %s, float %clamp) {
 ; GFX10-LABEL: sample_cd_cl_1d:
 ; GFX10:       ; %bb.0: ; %main_body
 ; GFX10-NEXT:    image_sample_cd_cl_g16 v[0:3], v[0:3], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_1D ; encoding: [0x01,0x0f,0xa4,0xf1,0x00,0x00,0x40,0x00]
 ; GFX10-NEXT:    s_waitcnt vmcnt(0) ; encoding: [0x70,0x3f,0x8c,0xbf]
 ; GFX10-NEXT:    ; return to shader part epilog
 main_body:
   %v = call <4 x float> @llvm.amdgcn.image.sample.cd.cl.1d.v4f32.f16.f32(i32 15, half %dsdh, half %dsdv, float %s, float %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
 
 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
 main_body:
   %v = call <4 x float> @llvm.amdgcn.image.sample.cd.cl.2d.v4f32.f16.f32(i32 15, half %dsdh, half %dtdh, half %dsdv, half %dtdv, float %s, float %t, float %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
 
 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, float %s, float %clamp) {
 ; GFX10-LABEL: sample_c_cd_cl_1d:
 ; GFX10:       ; %bb.0: ; %main_body
 ; GFX10-NEXT:    image_sample_c_cd_cl_g16 v[0:3], v[0:4], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_1D ; encoding: [0x01,0x0f,0xac,0xf1,0x00,0x00,0x40,0x00]
 ; GFX10-NEXT:    s_waitcnt vmcnt(0) ; encoding: [0x70,0x3f,0x8c,0xbf]
 ; GFX10-NEXT:    ; return to shader part epilog
 main_body:
   %v = call <4 x float> @llvm.amdgcn.image.sample.c.cd.cl.1d.v4f32.f16.f32(i32 15, float %zcompare, half %dsdh, half %dsdv, float %s, float %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
 
 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, float %s, float %t, float %clamp) {
 ; GFX10-LABEL: sample_c_cd_cl_2d:
 ; 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
 main_body:
   %v = call <4 x float> @llvm.amdgcn.image.sample.c.cd.cl.2d.v4f32.f16.f32(i32 15, float %zcompare, half %dsdh, half %dtdh, half %dsdv, half %dtdv, float %s, float %t, float %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
 

llvm/test/CodeGen/AMDGPU/llvm.amdgcn.image.sample.cd.g16.ll

 main_body:
   %v = call <4 x float> @llvm.amdgcn.image.sample.cd.1d.v4f32.f16.f32(i32 15, half %dsdh, half %dsdv, float %s, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
 
 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
 main_body:
   %v = call <4 x float> @llvm.amdgcn.image.sample.cd.2d.v4f32.f16.f32(i32 15, half %dsdh, half %dtdh, half %dsdv, half %dtdv, float %s, float %t, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
 
 define amdgpu_ps <4 x float> @sample_c_cd_1d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %zcompare, half %dsdh, half %dsdv, float %s) {
 ; GFX10-LABEL: sample_c_cd_1d:
 ; GFX10:       ; %bb.0: ; %main_body
 ; GFX10-NEXT:    image_sample_c_cd_g16 v[0:3], v[0:3], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_1D
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    ; return to shader part epilog
 main_body:
   %v = call <4 x float> @llvm.amdgcn.image.sample.c.cd.1d.v4f32.f16.f32(i32 15, float %zcompare, half %dsdh, half %dsdv, float %s, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
 
 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
 main_body:
   %v = call <4 x float> @llvm.amdgcn.image.sample.c.cd.2d.v4f32.f16.f32(i32 15, float %zcompare, half %dsdh, half %dtdh, half %dsdv, half %dtdv, float %s, float %t, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
 
 define amdgpu_ps <4 x float> @sample_cd_cl_1d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %dsdh, half %dsdv, float %s, float %clamp) {
 ; GFX10-LABEL: sample_cd_cl_1d:
 ; GFX10:       ; %bb.0: ; %main_body
 ; GFX10-NEXT:    image_sample_cd_cl_g16 v[0:3], v[0:3], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_1D
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    ; return to shader part epilog
 main_body:
   %v = call <4 x float> @llvm.amdgcn.image.sample.cd.cl.1d.v4f32.f16.f32(i32 15, half %dsdh, half %dsdv, float %s, float %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
 
 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
 main_body:
   %v = call <4 x float> @llvm.amdgcn.image.sample.cd.cl.2d.v4f32.f16.f32(i32 15, half %dsdh, half %dtdh, half %dsdv, half %dtdv, float %s, float %t, float %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
 
 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, float %s, float %clamp) {
 ; GFX10-LABEL: sample_c_cd_cl_1d:
 ; GFX10:       ; %bb.0: ; %main_body
 ; GFX10-NEXT:    image_sample_c_cd_cl_g16 v[0:3], v[0:4], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_1D
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    ; return to shader part epilog
 main_body:
   %v = call <4 x float> @llvm.amdgcn.image.sample.c.cd.cl.1d.v4f32.f16.f32(i32 15, float %zcompare, half %dsdh, half %dsdv, float %s, float %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
 
 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, float %s, float %t, float %clamp) {
 ; GFX10-LABEL: sample_c_cd_cl_2d:
 ; 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
 main_body:
   %v = call <4 x float> @llvm.amdgcn.image.sample.c.cd.cl.2d.v4f32.f16.f32(i32 15, float %zcompare, half %dsdh, half %dtdh, half %dsdv, half %dtdv, float %s, float %t, float %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
 

llvm/test/CodeGen/AMDGPU/llvm.amdgcn.image.sample.g16.a16.dim.ll

 main_body:
   %v = call <4 x float> @llvm.amdgcn.image.sample.d.1d.v4f32.f32.f16(i32 15, float %dsdh, float %dsdv, half %s, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
 
 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
 ;
 ; GFX10GISEL-LABEL: sample_d_2d:
 ; GFX10GISEL:       ; %bb.0: ; %main_body
 ; GFX10GISEL-NEXT:    v_lshlrev_b32_e32 v5, 16, v5
 ; GFX10GISEL-NEXT:    v_and_or_b32 v4, 0xffff, v4, v5
 ; GFX10GISEL-NEXT:    image_sample_d v[0:3], v[0:4], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_2D a16
 ; GFX10GISEL-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10GISEL-NEXT:    ; return to shader part epilog
 main_body:
   %v = call <4 x float> @llvm.amdgcn.image.sample.d.2d.v4f32.f32.f16(i32 15, float %dsdh, float %dtdh, float %dsdv, float %dtdv, half %s, half %t, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
 
 define amdgpu_ps <4 x float> @sample_d_3d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %dsdh, float %dtdh, float %drdh, float %dsdv, float %dtdv, float %drdv, half %s, half %t, half %r) {
 ; 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
 ;
 ; GFX10GISEL-LABEL: sample_d_3d:
 ; GFX10GISEL:       ; %bb.0: ; %main_body
 ; GFX10GISEL-NEXT:    v_lshlrev_b32_e32 v7, 16, v7
 ; GFX10GISEL-NEXT:    s_lshl_b32 s12, s0, 16

 main_body:
   %v = call <4 x float> @llvm.amdgcn.image.sample.c.d.1d.v4f32.f32.f16(i32 15, float %zcompare, float %dsdh, float %dsdv, half %s, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
 
 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
 ;
 ; GFX10GISEL-LABEL: sample_c_d_2d:
 ; GFX10GISEL:       ; %bb.0: ; %main_body
 ; GFX10GISEL-NEXT:    v_lshlrev_b32_e32 v6, 16, v6
 ; GFX10GISEL-NEXT:    v_and_or_b32 v5, 0xffff, v5, v6
 ; GFX10GISEL-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
 ; GFX10GISEL-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10GISEL-NEXT:    ; return to shader part epilog
 main_body:
   %v = call <4 x float> @llvm.amdgcn.image.sample.c.d.2d.v4f32.f32.f16(i32 15, float %zcompare, float %dsdh, float %dtdh, float %dsdv, float %dtdv, half %s, half %t, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
 
 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
 ;
 ; GFX10GISEL-LABEL: sample_d_cl_1d:
 ; GFX10GISEL:       ; %bb.0: ; %main_body
 ; GFX10GISEL-NEXT:    v_lshlrev_b32_e32 v3, 16, v3
 ; GFX10GISEL-NEXT:    v_and_or_b32 v2, 0xffff, v2, v3
 ; GFX10GISEL-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
 ; GFX10GISEL-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10GISEL-NEXT:    ; return to shader part epilog
 main_body:
   %v = call <4 x float> @llvm.amdgcn.image.sample.d.cl.1d.v4f32.f32.f16(i32 15, float %dsdh, float %dsdv, half %s, half %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
 
 define amdgpu_ps <4 x float> @sample_d_cl_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %dsdh, float %dtdh, float %dsdv, float %dtdv, half %s, half %t, half %clamp) {
 ; 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
 ;
 ; GFX10GISEL-LABEL: sample_d_cl_2d:
 ; GFX10GISEL:       ; %bb.0: ; %main_body
 ; GFX10GISEL-NEXT:    v_lshlrev_b32_e32 v5, 16, v5
 ; GFX10GISEL-NEXT:    s_lshl_b32 s12, s0, 16
 ; GFX10GISEL-NEXT:    v_and_or_b32 v4, 0xffff, v4, v5
 ; GFX10GISEL-NEXT:    v_and_or_b32 v5, 0xffff, v6, s12
 ; GFX10GISEL-NEXT:    image_sample_d_cl v[0:3], v[0:5], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_2D a16
 ; GFX10GISEL-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10GISEL-NEXT:    ; return to shader part epilog
 main_body:
   %v = call <4 x float> @llvm.amdgcn.image.sample.d.cl.2d.v4f32.f32.f16(i32 15, float %dsdh, float %dtdh, float %dsdv, float %dtdv, half %s, half %t, half %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
 
 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
 ;
 ; GFX10GISEL-LABEL: sample_c_d_cl_1d:
 ; GFX10GISEL:       ; %bb.0: ; %main_body
 ; GFX10GISEL-NEXT:    v_lshlrev_b32_e32 v4, 16, v4
 ; GFX10GISEL-NEXT:    v_and_or_b32 v3, 0xffff, v3, v4
 ; GFX10GISEL-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
 ; GFX10GISEL-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10GISEL-NEXT:    ; return to shader part epilog
 main_body:
   %v = call <4 x float> @llvm.amdgcn.image.sample.c.d.cl.1d.v4f32.f32.f16(i32 15, float %zcompare, float %dsdh, float %dsdv, half %s, half %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
 
 define amdgpu_ps <4 x float> @sample_c_d_cl_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, half %clamp) {
 ; 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
 ;
 ; GFX10GISEL-LABEL: sample_c_d_cl_2d:
 ; GFX10GISEL:       ; %bb.0: ; %main_body
 ; GFX10GISEL-NEXT:    v_lshlrev_b32_e32 v6, 16, v6
 ; GFX10GISEL-NEXT:    s_lshl_b32 s12, s0, 16

 main_body:
   %v = call <4 x float> @llvm.amdgcn.image.sample.cd.1d.v4f32.f32.f16(i32 15, float %dsdh, float %dsdv, half %s, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
 
 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
 ;
 ; GFX10GISEL-LABEL: sample_cd_2d:
 ; GFX10GISEL:       ; %bb.0: ; %main_body
 ; GFX10GISEL-NEXT:    v_lshlrev_b32_e32 v5, 16, v5
 ; GFX10GISEL-NEXT:    v_and_or_b32 v4, 0xffff, v4, v5

 main_body:
   %v = call <4 x float> @llvm.amdgcn.image.sample.c.cd.1d.v4f32.f32.f16(i32 15, float %zcompare, float %dsdh, float %dsdv, half %s, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
 
 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
 ;
 ; GFX10GISEL-LABEL: sample_c_cd_2d:
 ; GFX10GISEL:       ; %bb.0: ; %main_body
 ; GFX10GISEL-NEXT:    v_lshlrev_b32_e32 v6, 16, v6
 ; GFX10GISEL-NEXT:    v_and_or_b32 v5, 0xffff, v5, v6
 ; GFX10GISEL-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
 ; GFX10GISEL-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10GISEL-NEXT:    ; return to shader part epilog
 main_body:
   %v = call <4 x float> @llvm.amdgcn.image.sample.c.cd.2d.v4f32.f32.f16(i32 15, float %zcompare, float %dsdh, float %dtdh, float %dsdv, float %dtdv, half %s, half %t, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
 
 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
 ;
 ; GFX10GISEL-LABEL: sample_cd_cl_1d:
 ; GFX10GISEL:       ; %bb.0: ; %main_body
 ; GFX10GISEL-NEXT:    v_lshlrev_b32_e32 v3, 16, v3
 ; GFX10GISEL-NEXT:    v_and_or_b32 v2, 0xffff, v2, v3
 ; GFX10GISEL-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
 ; GFX10GISEL-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10GISEL-NEXT:    ; return to shader part epilog
 main_body:
   %v = call <4 x float> @llvm.amdgcn.image.sample.cd.cl.1d.v4f32.f32.f16(i32 15, float %dsdh, float %dsdv, half %s, half %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
 
 define amdgpu_ps <4 x float> @sample_cd_cl_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %dsdh, float %dtdh, float %dsdv, float %dtdv, half %s, half %t, half %clamp) {
 ; 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
 ;
 ; GFX10GISEL-LABEL: sample_cd_cl_2d:
 ; GFX10GISEL:       ; %bb.0: ; %main_body
 ; GFX10GISEL-NEXT:    v_lshlrev_b32_e32 v5, 16, v5
 ; GFX10GISEL-NEXT:    s_lshl_b32 s12, s0, 16
 ; GFX10GISEL-NEXT:    v_and_or_b32 v4, 0xffff, v4, v5
 ; GFX10GISEL-NEXT:    v_and_or_b32 v5, 0xffff, v6, s12
 ; GFX10GISEL-NEXT:    image_sample_cd_cl v[0:3], v[0:5], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_2D a16
 ; GFX10GISEL-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10GISEL-NEXT:    ; return to shader part epilog
 main_body:
   %v = call <4 x float> @llvm.amdgcn.image.sample.cd.cl.2d.v4f32.f32.f16(i32 15, float %dsdh, float %dtdh, float %dsdv, float %dtdv, half %s, half %t, half %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
 
 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
 ;
 ; GFX10GISEL-LABEL: sample_c_cd_cl_1d:
 ; GFX10GISEL:       ; %bb.0: ; %main_body
 ; GFX10GISEL-NEXT:    v_lshlrev_b32_e32 v4, 16, v4
 ; GFX10GISEL-NEXT:    v_and_or_b32 v3, 0xffff, v3, v4
 ; GFX10GISEL-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
 ; GFX10GISEL-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10GISEL-NEXT:    ; return to shader part epilog
 main_body:
   %v = call <4 x float> @llvm.amdgcn.image.sample.c.cd.cl.1d.v4f32.f32.f16(i32 15, float %zcompare, float %dsdh, float %dsdv, half %s, half %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
 
 define amdgpu_ps <4 x float> @sample_c_cd_cl_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, half %clamp) {
 ; 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
 ;
 ; GFX10GISEL-LABEL: sample_c_cd_cl_2d:
 ; GFX10GISEL:       ; %bb.0: ; %main_body
 ; GFX10GISEL-NEXT:    v_lshlrev_b32_e32 v6, 16, v6
 ; GFX10GISEL-NEXT:    s_lshl_b32 s12, s0, 16
 ; GFX10GISEL-NEXT:    v_and_or_b32 v5, 0xffff, v5, v6
 ; GFX10GISEL-NEXT:    v_and_or_b32 v6, 0xffff, v7, s12
 ; GFX10GISEL-NEXT:    image_sample_c_cd_cl v[0:3], v[0:6], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_2D a16
 ; GFX10GISEL-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10GISEL-NEXT:    ; return to shader part epilog
 main_body:
   %v = call <4 x float> @llvm.amdgcn.image.sample.c.cd.cl.2d.v4f32.f32.f16(i32 15, float %zcompare, float %dsdh, float %dtdh, float %dsdv, float %dtdv, half %s, half %t, half %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
 
 define amdgpu_ps float @sample_c_d_o_2darray_V1(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, i32 %offset, float %zcompare, float %dsdh, float %dtdh, float %dsdv, float %dtdv, half %s, half %t, half %slice) {
 ; 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
 ;
 ; GFX10GISEL-LABEL: sample_c_d_o_2darray_V1:
 ; GFX10GISEL:       ; %bb.0: ; %main_body
 ; GFX10GISEL-NEXT:    v_lshlrev_b32_e32 v7, 16, v7
 ; GFX10GISEL-NEXT:    s_lshl_b32 s12, s0, 16
 ; GFX10GISEL-NEXT:    v_and_or_b32 v6, 0xffff, v6, v7
 ; GFX10GISEL-NEXT:    v_and_or_b32 v7, 0xffff, v8, s12
 ; GFX10GISEL-NEXT:    image_sample_c_d_o v0, v[0:7], s[0:7], s[8:11] dmask:0x4 dim:SQ_RSRC_IMG_2D_ARRAY a16
 ; GFX10GISEL-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10GISEL-NEXT:    ; return to shader part epilog
 main_body:
   %v = call float @llvm.amdgcn.image.sample.c.d.o.2darray.f32.f32.f16(i32 4, i32 %offset, float %zcompare, float %dsdh, float %dtdh, float %dsdv, float %dtdv, half %s, half %t, half %slice, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret float %v
 }
 
 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, float %dsdh, float %dtdh, float %dsdv, float %dtdv, half %s, half %t, half %slice) {
 ; 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
 ;
 ; GFX10GISEL-LABEL: sample_c_d_o_2darray_V2:
 ; GFX10GISEL:       ; %bb.0: ; %main_body
 ; GFX10GISEL-NEXT:    v_lshlrev_b32_e32 v7, 16, v7
 ; GFX10GISEL-NEXT:    s_lshl_b32 s12, s0, 16

 main_body:
   %v = call <4 x float> @llvm.amdgcn.image.sample.d.1d.v4f32.f16.f32(i32 15, half %dsdh, half %dsdv, float %s, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
 
 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
 ;
 ; GFX10GISEL-LABEL: sample_g16_noa16_d_2d:
 ; GFX10GISEL:       ; %bb.0: ; %main_body
 ; GFX10GISEL-NEXT:    v_lshlrev_b32_e32 v1, 16, v1
 ; GFX10GISEL-NEXT:    v_lshlrev_b32_e32 v3, 16, v3
 ; GFX10GISEL-NEXT:    v_and_or_b32 v0, 0xffff, v0, v1
 ; GFX10GISEL-NEXT:    v_and_or_b32 v1, 0xffff, v2, v3
 ; GFX10GISEL-NEXT:    image_sample_d_g16 v[0:3], [v0, v1, v4, v5], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_2D
 ; GFX10GISEL-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10GISEL-NEXT:    ; return to shader part epilog
 main_body:
   %v = call <4 x float> @llvm.amdgcn.image.sample.d.2d.v4f32.f16.f32(i32 15, half %dsdh, half %dtdh, half %dsdv, half %dtdv, float %s, float %t, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
 
 define amdgpu_ps <4 x float> @sample_g16_noa16_d_3d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %dsdh, half %dtdh, half %drdh, half %dsdv, half %dtdv, half %drdv, float %s, float %t, float %r) {
 ; GFX10-LABEL: sample_g16_noa16_d_3d:
 ; 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
 ;
 ; GFX10GISEL-LABEL: sample_g16_noa16_d_3d:
 ; GFX10GISEL:       ; %bb.0: ; %main_body
 ; GFX10GISEL-NEXT:    v_mov_b32_e32 v9, v2
 ; GFX10GISEL-NEXT:    v_mov_b32_e32 v10, v3

 main_body:
   %v = call <4 x float> @llvm.amdgcn.image.sample.c.d.1d.v4f32.f16.f32(i32 15, float %zcompare, half %dsdh, half %dsdv, float %s, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
 
 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
 ;
 ; GFX10GISEL-LABEL: sample_g16_noa16_c_d_2d:
 ; GFX10GISEL:       ; %bb.0: ; %main_body
 ; GFX10GISEL-NEXT:    v_lshlrev_b32_e32 v2, 16, v2
 ; GFX10GISEL-NEXT:    v_lshlrev_b32_e32 v4, 16, v4

 main_body:
   %v = call <4 x float> @llvm.amdgcn.image.sample.d.cl.1d.v4f32.f16.f32(i32 15, half %dsdh, half %dsdv, float %s, float %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
 
 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
 ;
 ; GFX10GISEL-LABEL: sample_g16_noa16_d_cl_2d:
 ; GFX10GISEL:       ; %bb.0: ; %main_body
 ; GFX10GISEL-NEXT:    v_lshlrev_b32_e32 v1, 16, v1
 ; GFX10GISEL-NEXT:    v_lshlrev_b32_e32 v3, 16, v3
main_body:
   ret <4 x float> %v
 }
 
 define amdgpu_ps <4 x float> @sample_g16_noa16_c_d_cl_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, float %clamp) {
 ; GFX10-LABEL: sample_g16_noa16_c_d_cl_2d:
 ; 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
 ;
 ; GFX10GISEL-LABEL: sample_g16_noa16_c_d_cl_2d:
 ; GFX10GISEL:       ; %bb.0: ; %main_body
 ; GFX10GISEL-NEXT:    v_mov_b32_e32 v8, v2
 ; GFX10GISEL-NEXT:    v_mov_b32_e32 v9, v3

 main_body:
   %v = call <4 x float> @llvm.amdgcn.image.sample.cd.1d.v4f32.f16.f32(i32 15, half %dsdh, half %dsdv, float %s, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
 
 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
 ;
 ; GFX10GISEL-LABEL: sample_g16_noa16_cd_2d:
 ; GFX10GISEL:       ; %bb.0: ; %main_body
 ; GFX10GISEL-NEXT:    v_lshlrev_b32_e32 v1, 16, v1
 ; GFX10GISEL-NEXT:    v_lshlrev_b32_e32 v3, 16, v3

 main_body:
   %v = call <4 x float> @llvm.amdgcn.image.sample.c.cd.1d.v4f32.f16.f32(i32 15, float %zcompare, half %dsdh, half %dsdv, float %s, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
 
 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
 ;
 ; GFX10GISEL-LABEL: sample_g16_noa16_c_cd_2d:
 ; GFX10GISEL:       ; %bb.0: ; %main_body
 ; GFX10GISEL-NEXT:    v_lshlrev_b32_e32 v2, 16, v2
 ; GFX10GISEL-NEXT:    v_lshlrev_b32_e32 v4, 16, v4

 main_body:
   %v = call <4 x float> @llvm.amdgcn.image.sample.cd.cl.1d.v4f32.f16.f32(i32 15, half %dsdh, half %dsdv, float %s, float %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
 
 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
 ;
 ; GFX10GISEL-LABEL: sample_g16_noa16_cd_cl_2d:
 ; GFX10GISEL:       ; %bb.0: ; %main_body
 ; GFX10GISEL-NEXT:    v_lshlrev_b32_e32 v1, 16, v1
 ; GFX10GISEL-NEXT:    v_lshlrev_b32_e32 v3, 16, v3
main_body:
   ret <4 x float> %v
 }
 
 define amdgpu_ps <4 x float> @sample_g16_noa16_c_cd_cl_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, float %clamp) {
 ; GFX10-LABEL: sample_g16_noa16_c_cd_cl_2d:
 ; 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
 ;
 ; GFX10GISEL-LABEL: sample_g16_noa16_c_cd_cl_2d:
 ; GFX10GISEL:       ; %bb.0: ; %main_body
 ; GFX10GISEL-NEXT:    v_mov_b32_e32 v8, v2
 ; GFX10GISEL-NEXT:    v_mov_b32_e32 v9, v3
 ; GFX10GISEL-NEXT:    v_mov_b32_e32 v2, v0
 ; GFX10GISEL-NEXT:    v_lshlrev_b32_e32 v4, 16, v4
 ; GFX10GISEL-NEXT:    v_lshlrev_b32_e32 v0, 16, v8
 ; GFX10GISEL-NEXT:    v_and_or_b32 v4, 0xffff, v9, v4
 ; GFX10GISEL-NEXT:    v_and_or_b32 v3, 0xffff, v1, v0
 ; GFX10GISEL-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
 ; GFX10GISEL-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10GISEL-NEXT:    ; return to shader part epilog
 main_body:
   %v = call <4 x float> @llvm.amdgcn.image.sample.c.cd.cl.2d.v4f32.f16.f32(i32 15, float %zcompare, half %dsdh, half %dtdh, half %dsdv, half %dtdv, float %s, float %t, float %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
 
 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
 ;
 ; GFX10GISEL-LABEL: sample_g16_noa16_c_d_o_2darray_V1:
 ; GFX10GISEL:       ; %bb.0: ; %main_body
 ; GFX10GISEL-NEXT:    v_mov_b32_e32 v9, v3
 ; GFX10GISEL-NEXT:    v_mov_b32_e32 v10, v2

 main_body:
   %v = call float @llvm.amdgcn.image.sample.c.d.o.2darray.f32.f16.f32(i32 4, i32 %offset, float %zcompare, half %dsdh, half %dtdh, half %dsdv, half %dtdv, float %s, float %t, float %slice, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret float %v
 }
 
 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
 ;
 ; GFX10GISEL-LABEL: sample_g16_noa16_c_d_o_2darray_V2:
 ; GFX10GISEL:       ; %bb.0: ; %main_body
 ; GFX10GISEL-NEXT:    v_mov_b32_e32 v9, v3
 ; GFX10GISEL-NEXT:    v_mov_b32_e32 v10, v2

llvm/test/CodeGen/AMDGPU/llvm.amdgcn.image.sample.g16.encode.ll

 main_body:
   %v = call <4 x float> @llvm.amdgcn.image.sample.d.1d.v4f32.f16.f32(i32 15, half %dsdh, half %dsdv, float %s, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
 
 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
 main_body:
   %v = call <4 x float> @llvm.amdgcn.image.sample.d.2d.v4f32.f16.f32(i32 15, half %dsdh, half %dtdh, half %dsdv, half %dtdv, float %s, float %t, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
 
 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, float %s, float %t, float %r) {
 ; GFX10-LABEL: sample_d_3d:
 ; 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
 ;
 ; GFX11-LABEL: sample_d_3d:
 ; 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
 main_body:
   %v = call <4 x float> @llvm.amdgcn.image.sample.d.3d.v4f32.f16.f32(i32 15, half %dsdh, half %dtdh, half %drdh, half %dsdv, half %dtdv, half %drdv, float %s, float %t, float %r, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
 

 main_body:
   %v = call <4 x float> @llvm.amdgcn.image.sample.c.d.1d.v4f32.f16.f32(i32 15, float %zcompare, half %dsdh, half %dsdv, float %s, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
 
 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
 main_body:
   %v = call <4 x float> @llvm.amdgcn.image.sample.c.d.2d.v4f32.f16.f32(i32 15, float %zcompare, half %dsdh, half %dtdh, half %dsdv, half %dtdv, float %s, float %t, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
 

 main_body:
   %v = call <4 x float> @llvm.amdgcn.image.sample.d.cl.1d.v4f32.f16.f32(i32 15, half %dsdh, half %dsdv, float %s, float %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
 
 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
 main_body:
   %v = call <4 x float> @llvm.amdgcn.image.sample.d.cl.2d.v4f32.f16.f32(i32 15, half %dsdh, half %dtdh, half %dsdv, half %dtdv, float %s, float %t, float %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
 
main_body:
   ret <4 x float> %v
 }
 
 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, float %s, float %t, float %clamp) {
 ; GFX10-LABEL: sample_c_d_cl_2d:
 ; 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
 ;
 ; GFX11-LABEL: sample_c_d_cl_2d:
 ; 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
 main_body:
   %v = call <4 x float> @llvm.amdgcn.image.sample.c.d.cl.2d.v4f32.f16.f32(i32 15, float %zcompare, half %dsdh, half %dtdh, half %dsdv, half %dtdv, float %s, float %t, float %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
 
 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
 main_body:
   %v = call float @llvm.amdgcn.image.sample.c.d.o.2darray.f16.f32.f32(i32 4, i32 %offset, float %zcompare, half %dsdh, half %dtdh, half %dsdv, half %dtdv, float %s, float %t, float %slice, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret float %v
 }
 
 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
 main_body:
   %v = call <2 x float> @llvm.amdgcn.image.sample.c.d.o.2darray.v2f32.f16.f32(i32 6, i32 %offset, float %zcompare, half %dsdh, half %dtdh, half %dsdv, half %dtdv, float %s, float %t, float %slice, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <2 x float> %v
 }
 

llvm/test/CodeGen/AMDGPU/llvm.amdgcn.image.sample.g16.ll

 main_body:
   %v = call <4 x float> @llvm.amdgcn.image.sample.d.1d.v4f32.f16.f32(i32 15, half %dsdh, half %dsdv, float %s, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
 
 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
 main_body:
   %v = call <4 x float> @llvm.amdgcn.image.sample.d.2d.v4f32.f16.f32(i32 15, half %dsdh, half %dtdh, half %dsdv, half %dtdv, float %s, float %t, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
 
 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, float %s, float %t, float %r) {
 ; GFX10-LABEL: sample_d_3d:
 ; 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
 main_body:
   %v = call <4 x float> @llvm.amdgcn.image.sample.d.3d.v4f32.f16.f32(i32 15, half %dsdh, half %dtdh, half %drdh, half %dsdv, half %dtdv, half %drdv, float %s, float %t, float %r, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
 
 define amdgpu_ps <4 x float> @sample_c_d_1d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %zcompare, half %dsdh, half %dsdv, float %s) {
 ; GFX10-LABEL: sample_c_d_1d:
 ; GFX10:       ; %bb.0: ; %main_body
 ; GFX10-NEXT:    image_sample_c_d_g16 v[0:3], v[0:3], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_1D
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    ; return to shader part epilog
 main_body:
   %v = call <4 x float> @llvm.amdgcn.image.sample.c.d.1d.v4f32.f16.f32(i32 15, float %zcompare, half %dsdh, half %dsdv, float %s, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
 
 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
 main_body:
   %v = call <4 x float> @llvm.amdgcn.image.sample.c.d.2d.v4f32.f16.f32(i32 15, float %zcompare, half %dsdh, half %dtdh, half %dsdv, half %dtdv, float %s, float %t, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
 
 define amdgpu_ps <4 x float> @sample_d_cl_1d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %dsdh, half %dsdv, float %s, float %clamp) {
 ; GFX10-LABEL: sample_d_cl_1d:
 ; GFX10:       ; %bb.0: ; %main_body
 ; GFX10-NEXT:    image_sample_d_cl_g16 v[0:3], v[0:3], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_1D
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    ; return to shader part epilog
 main_body:
   %v = call <4 x float> @llvm.amdgcn.image.sample.d.cl.1d.v4f32.f16.f32(i32 15, half %dsdh, half %dsdv, float %s, float %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
 
 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
 main_body:
   %v = call <4 x float> @llvm.amdgcn.image.sample.d.cl.2d.v4f32.f16.f32(i32 15, half %dsdh, half %dtdh, half %dsdv, half %dtdv, float %s, float %t, float %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
 
 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, float %s, float %clamp) {
 ; GFX10-LABEL: sample_c_d_cl_1d:
 ; GFX10:       ; %bb.0: ; %main_body
 ; GFX10-NEXT:    image_sample_c_d_cl_g16 v[0:3], v[0:4], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_1D
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    ; return to shader part epilog
 main_body:
   %v = call <4 x float> @llvm.amdgcn.image.sample.c.d.cl.1d.v4f32.f16.f32(i32 15, float %zcompare, half %dsdh, half %dsdv, float %s, float %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
 
 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, float %s, float %t, float %clamp) {
 ; GFX10-LABEL: sample_c_d_cl_2d:
 ; 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
 main_body:
   %v = call <4 x float> @llvm.amdgcn.image.sample.c.d.cl.2d.v4f32.f16.f32(i32 15, float %zcompare, half %dsdh, half %dtdh, half %dsdv, half %dtdv, float %s, float %t, float %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v
 }
 
 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
 main_body:
   %v = call float @llvm.amdgcn.image.sample.c.d.o.2darray.f16.f32.f32(i32 4, i32 %offset, float %zcompare, half %dsdh, half %dtdh, half %dsdv, half %dtdv, float %s, float %t, float %slice, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret float %v
 }
 
 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
 main_body:
   %v = call <2 x float> @llvm.amdgcn.image.sample.c.d.o.2darray.v2f32.f16.f32(i32 6, i32 %offset, float %zcompare, half %dsdh, half %dtdh, half %dsdv, half %dtdv, float %s, float %t, float %slice, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <2 x float> %v
 }
 

llvm/test/CodeGen/AMDGPU/load-hi16.ll

 ; GFX900-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX900-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX906-LABEL: load_local_lo_hi_v2i16_multi_use_lo:
 ; GFX906:       ; %bb.0: ; %entry
 ; 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]
 ;
 ; GFX803-LABEL: load_local_lo_hi_v2i16_multi_use_lo:
 ; GFX803:       ; %bb.0: ; %entry
 ; GFX803-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX803-NEXT:    s_mov_b32 m0, -1
 ; GFX803-NEXT:    ds_read_u16 v1, v0

 }
 
 define <2 x i16> @load_local_lo_hi_v2i16_multi_use_hi(i16 addrspace(3)* noalias %in) #0 {
 ; GFX900-LABEL: load_local_lo_hi_v2i16_multi_use_hi:
 ; GFX900:       ; %bb.0: ; %entry
 ; 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]
 ;
 ; GFX906-LABEL: load_local_lo_hi_v2i16_multi_use_hi:
 ; GFX906:       ; %bb.0: ; %entry
 ; 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]
 ;
 ; GFX803-LABEL: load_local_lo_hi_v2i16_multi_use_hi:
 ; GFX803:       ; %bb.0: ; %entry
 ; GFX803-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX803-NEXT:    s_mov_b32 m0, -1
 ; GFX803-NEXT:    ds_read_u16 v1, v0 offset:16
 ; GFX803-NEXT:    ds_read_u16 v0, v0
 ; GFX803-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX803-NEXT:    s_waitcnt lgkmcnt(1)
 ; GFX803-NEXT:    ds_write_b16 v2, v1
 ; GFX803-NEXT:    v_lshlrev_b32_e32 v1, 16, v1
 ; GFX803-NEXT:    s_waitcnt lgkmcnt(1)
 ; GFX803-NEXT:    v_or_b32_e32 v0, v0, v1
 ; GFX803-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX803-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX900-FLATSCR-LABEL: load_local_lo_hi_v2i16_multi_use_hi:
 ; GFX900-FLATSCR:       ; %bb.0: ; %entry
 ; 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:
   %gep = getelementptr inbounds i16, i16 addrspace(3)* %in, i32 8
   %load.lo = load i16, i16 addrspace(3)* %in
   %load.hi = load i16, i16 addrspace(3)* %gep
   store i16 %load.hi, i16 addrspace(3)* null
   %build0 = insertelement <2 x i16> undef, i16 %load.lo, i32 0
   %build1 = insertelement <2 x i16> %build0, i16 %load.hi, i32 1
   ret <2 x i16> %build1
 }
 
 define <2 x i16> @load_local_lo_hi_v2i16_multi_use_lohi(i16 addrspace(3)* noalias %in, i16 addrspace(3)* noalias %out0, i16 addrspace(3)* noalias %out1) #0 {
 ; GFX900-LABEL: load_local_lo_hi_v2i16_multi_use_lohi:
 ; GFX900:       ; %bb.0: ; %entry
 ; 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]
 ;
 ; GFX906-LABEL: load_local_lo_hi_v2i16_multi_use_lohi:
 ; GFX906:       ; %bb.0: ; %entry
 ; 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]
 ;
 ; GFX803-LABEL: load_local_lo_hi_v2i16_multi_use_lohi:
 ; GFX803:       ; %bb.0: ; %entry
 ; GFX803-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX803-NEXT:    s_mov_b32 m0, -1
 ; GFX803-NEXT:    ds_read_u16 v3, v0
 ; GFX803-NEXT:    ds_read_u16 v0, v0 offset:16
 ; GFX803-NEXT:    s_waitcnt lgkmcnt(1)
 ; GFX803-NEXT:    ds_write_b16 v1, v3
 ; GFX803-NEXT:    s_waitcnt lgkmcnt(1)
 ; GFX803-NEXT:    ds_write_b16 v2, v0
 ; GFX803-NEXT:    v_lshlrev_b32_e32 v0, 16, v0
 ; GFX803-NEXT:    v_or_b32_sdwa v0, v3, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD
 ; GFX803-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX803-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX900-FLATSCR-LABEL: load_local_lo_hi_v2i16_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 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:
   %gep = getelementptr inbounds i16, i16 addrspace(3)* %in, i32 8
   %load.lo = load i16, i16 addrspace(3)* %in
   %load.hi = load i16, i16 addrspace(3)* %gep
   store i16 %load.lo, i16 addrspace(3)* %out0
   store i16 %load.hi, i16 addrspace(3)* %out1

 ; GFX900-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX900-NEXT:    v_mov_b32_e32 v0, v1
 ; GFX900-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX906-LABEL: load_local_hi_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:    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:
 ; GFX803:       ; %bb.0: ; %entry
 ; GFX803-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX803-NEXT:    s_mov_b32 m0, -1
 ; GFX803-NEXT:    ds_read_u16 v0, v0
 ; GFX803-NEXT:    s_waitcnt lgkmcnt(0)

 ; GFX900-NEXT:    global_store_dword v[0:1], v1, off
 ; GFX900-NEXT:    s_waitcnt vmcnt(0)
 ; GFX900-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX906-LABEL: load_local_hi_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:    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]
 ;
 ; GFX803-LABEL: load_local_hi_v2i16_reglo_vreg:
 ; GFX803:       ; %bb.0: ; %entry
 ; GFX803-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX803-NEXT:    s_mov_b32 m0, -1

 ; GFX900-NEXT:    global_store_dword v[0:1], v1, off
 ; GFX900-NEXT:    s_waitcnt vmcnt(0)
 ; GFX900-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX906-LABEL: load_local_hi_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:    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]
 ;
 ; GFX803-LABEL: load_local_hi_v2f16_reglo_vreg:
 ; GFX803:       ; %bb.0: ; %entry
 ; GFX803-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX803-NEXT:    s_mov_b32 m0, -1

 ; GFX900-NEXT:    global_store_dword v[0:1], v1, off
 ; GFX900-NEXT:    s_waitcnt vmcnt(0)
 ; GFX900-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX906-LABEL: load_local_hi_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:    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]
 ;
 ; GFX803-LABEL: load_local_hi_v2i16_reglo_vreg_zexti8:
 ; GFX803:       ; %bb.0: ; %entry
 ; GFX803-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX803-NEXT:    s_mov_b32 m0, -1

 ; GFX900-NEXT:    global_store_dword v[0:1], v1, off
 ; GFX900-NEXT:    s_waitcnt vmcnt(0)
 ; GFX900-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX906-LABEL: load_local_hi_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:    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]
 ;
 ; GFX803-LABEL: load_local_hi_v2i16_reglo_vreg_sexti8:
 ; GFX803:       ; %bb.0: ; %entry
 ; GFX803-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX803-NEXT:    s_mov_b32 m0, -1

 ; GFX900-NEXT:    global_store_dword v[0:1], v1, off
 ; GFX900-NEXT:    s_waitcnt vmcnt(0)
 ; GFX900-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX906-LABEL: load_local_hi_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:    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]
 ;
 ; GFX803-LABEL: load_local_hi_v2f16_reglo_vreg_zexti8:
 ; GFX803:       ; %bb.0: ; %entry
 ; GFX803-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX803-NEXT:    s_mov_b32 m0, -1

 ; GFX900-NEXT:    global_store_dword v[0:1], v1, off
 ; GFX900-NEXT:    s_waitcnt vmcnt(0)
 ; GFX900-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX906-LABEL: load_local_hi_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:    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]
 ;
 ; GFX803-LABEL: load_local_hi_v2f16_reglo_vreg_sexti8:
 ; GFX803:       ; %bb.0: ; %entry
 ; GFX803-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX803-NEXT:    s_mov_b32 m0, -1

 ; GFX900-NEXT:    global_store_dword v[0:1], v2, off
 ; GFX900-NEXT:    s_waitcnt vmcnt(0)
 ; GFX900-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX906-LABEL: load_global_hi_v2i16_reglo_vreg:
 ; 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]
 ;
 ; GFX803-LABEL: load_global_hi_v2i16_reglo_vreg:
 ; GFX803:       ; %bb.0: ; %entry
 ; GFX803-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX803-NEXT:    v_add_u32_e32 v0, vcc, 0xfffff002, v0

 ; GFX900-NEXT:    global_store_dword v[0:1], v2, off
 ; GFX900-NEXT:    s_waitcnt vmcnt(0)
 ; GFX900-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX906-LABEL: load_global_hi_v2f16_reglo_vreg:
 ; 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]
 ;
 ; GFX803-LABEL: load_global_hi_v2f16_reglo_vreg:
 ; GFX803:       ; %bb.0: ; %entry
 ; GFX803-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX803-NEXT:    v_add_u32_e32 v0, vcc, 0xfffff002, v0

 ; GFX900-NEXT:    global_store_dword v[0:1], v2, off
 ; GFX900-NEXT:    s_waitcnt vmcnt(0)
 ; GFX900-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX906-LABEL: load_global_hi_v2i16_reglo_vreg_zexti8:
 ; 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]
 ;
 ; GFX803-LABEL: load_global_hi_v2i16_reglo_vreg_zexti8:
 ; GFX803:       ; %bb.0: ; %entry
 ; GFX803-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX803-NEXT:    v_add_u32_e32 v0, vcc, 0xfffff001, v0

 ; GFX900-NEXT:    global_store_dword v[0:1], v2, off
 ; GFX900-NEXT:    s_waitcnt vmcnt(0)
 ; GFX900-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX906-LABEL: load_global_hi_v2i16_reglo_vreg_sexti8:
 ; 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]
 ;
 ; GFX803-LABEL: load_global_hi_v2i16_reglo_vreg_sexti8:
 ; GFX803:       ; %bb.0: ; %entry
 ; GFX803-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX803-NEXT:    v_add_u32_e32 v0, vcc, 0xfffff001, v0

 ; GFX900-NEXT:    global_store_dword v[0:1], v2, off
 ; GFX900-NEXT:    s_waitcnt vmcnt(0)
 ; GFX900-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX906-LABEL: load_global_hi_v2f16_reglo_vreg_sexti8:
 ; 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]
 ;
 ; GFX803-LABEL: load_global_hi_v2f16_reglo_vreg_sexti8:
 ; GFX803:       ; %bb.0: ; %entry
 ; GFX803-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX803-NEXT:    v_add_u32_e32 v0, vcc, 0xfffff001, v0

 ; GFX900-NEXT:    global_store_dword v[0:1], v2, off
 ; GFX900-NEXT:    s_waitcnt vmcnt(0)
 ; GFX900-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX906-LABEL: load_global_hi_v2f16_reglo_vreg_zexti8:
 ; 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]
 ;
 ; GFX803-LABEL: load_global_hi_v2f16_reglo_vreg_zexti8:
 ; GFX803:       ; %bb.0: ; %entry
 ; GFX803-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX803-NEXT:    v_add_u32_e32 v0, vcc, 0xfffff001, v0

 ; GFX900-NEXT:    global_store_dword v[0:1], v2, off
 ; GFX900-NEXT:    s_waitcnt vmcnt(0)
 ; GFX900-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX906-LABEL: load_flat_hi_v2i16_reglo_vreg:
 ; 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]
 ;
 ; GFX803-LABEL: load_flat_hi_v2i16_reglo_vreg:
 ; GFX803:       ; %bb.0: ; %entry
 ; GFX803-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX803-NEXT:    flat_load_ushort v0, v[0:1]

 ; GFX900-NEXT:    global_store_dword v[0:1], v2, off
 ; GFX900-NEXT:    s_waitcnt vmcnt(0)
 ; GFX900-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX906-LABEL: load_flat_hi_v2f16_reglo_vreg:
 ; 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]
 ;
 ; GFX803-LABEL: load_flat_hi_v2f16_reglo_vreg:
 ; GFX803:       ; %bb.0: ; %entry
 ; GFX803-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX803-NEXT:    flat_load_ushort v0, v[0:1]

 ; GFX900-NEXT:    global_store_dword v[0:1], v2, off
 ; GFX900-NEXT:    s_waitcnt vmcnt(0)
 ; GFX900-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX906-LABEL: load_flat_hi_v2i16_reglo_vreg_zexti8:
 ; 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]
 ;
 ; GFX803-LABEL: load_flat_hi_v2i16_reglo_vreg_zexti8:
 ; GFX803:       ; %bb.0: ; %entry
 ; GFX803-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX803-NEXT:    flat_load_ubyte v0, v[0:1]

 ; GFX900-NEXT:    global_store_dword v[0:1], v2, off
 ; GFX900-NEXT:    s_waitcnt vmcnt(0)
 ; GFX900-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX906-LABEL: load_flat_hi_v2i16_reglo_vreg_sexti8:
 ; 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]
 ;
 ; GFX803-LABEL: load_flat_hi_v2i16_reglo_vreg_sexti8:
 ; GFX803:       ; %bb.0: ; %entry
 ; GFX803-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX803-NEXT:    flat_load_sbyte v0, v[0:1]

 ; GFX900-NEXT:    global_store_dword v[0:1], v2, off
 ; GFX900-NEXT:    s_waitcnt vmcnt(0)
 ; GFX900-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX906-LABEL: load_flat_hi_v2f16_reglo_vreg_zexti8:
 ; 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]
 ;
 ; GFX803-LABEL: load_flat_hi_v2f16_reglo_vreg_zexti8:
 ; GFX803:       ; %bb.0: ; %entry
 ; GFX803-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX803-NEXT:    flat_load_ubyte v0, v[0:1]

 ; GFX900-NEXT:    global_store_dword v[0:1], v2, off
 ; GFX900-NEXT:    s_waitcnt vmcnt(0)
 ; GFX900-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX906-LABEL: load_flat_hi_v2f16_reglo_vreg_sexti8:
 ; 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]
 ;
 ; GFX803-LABEL: load_flat_hi_v2f16_reglo_vreg_sexti8:
 ; GFX803:       ; %bb.0: ; %entry
 ; GFX803-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX803-NEXT:    flat_load_sbyte v0, v[0:1]

 ; GFX900-NEXT:    global_store_dword v[0:1], v0, off
 ; GFX900-NEXT:    s_waitcnt vmcnt(0)
 ; GFX900-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX906-LABEL: load_private_hi_v2i16_reglo_vreg:
 ; 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]
 ;
 ; GFX803-LABEL: load_private_hi_v2i16_reglo_vreg:
 ; GFX803:       ; %bb.0: ; %entry
 ; GFX803-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX803-NEXT:    buffer_load_ushort v1, off, s[0:3], s32 offset:4094

 ; GFX900-NEXT:    global_store_dword v[0:1], v0, off
 ; GFX900-NEXT:    s_waitcnt vmcnt(0)
 ; GFX900-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX906-LABEL: load_private_hi_v2f16_reglo_vreg:
 ; 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]
 ;
 ; GFX803-LABEL: load_private_hi_v2f16_reglo_vreg:
 ; GFX803:       ; %bb.0: ; %entry
 ; GFX803-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX803-NEXT:    buffer_load_ushort v1, off, s[0:3], s32 offset:4094

 ; GFX900-NEXT:    s_waitcnt vmcnt(0)
 ; GFX900-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX906-LABEL: load_private_hi_v2i16_reglo_vreg_nooff:
 ; GFX906:       ; %bb.0: ; %entry
 ; 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]
 ;
 ; GFX803-LABEL: load_private_hi_v2i16_reglo_vreg_nooff:
 ; GFX803:       ; %bb.0: ; %entry
 ; GFX803-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX803-NEXT:    buffer_load_ushort v1, off, s[0:3], 0 offset:4094 glc

 ; GFX900-NEXT:    s_waitcnt vmcnt(0)
 ; GFX900-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX906-LABEL: load_private_hi_v2f16_reglo_vreg_nooff:
 ; GFX906:       ; %bb.0: ; %entry
 ; 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]
 ;
 ; GFX803-LABEL: load_private_hi_v2f16_reglo_vreg_nooff:
 ; GFX803:       ; %bb.0: ; %entry
 ; GFX803-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX803-NEXT:    buffer_load_ushort v0, off, s[0:3], 0 offset:4094 glc

 ; GFX900-NEXT:    global_store_dword v[0:1], v0, off
 ; GFX900-NEXT:    s_waitcnt vmcnt(0)
 ; GFX900-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX906-LABEL: load_private_hi_v2i16_reglo_vreg_zexti8:
 ; 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]
 ;
 ; GFX803-LABEL: load_private_hi_v2i16_reglo_vreg_zexti8:
 ; GFX803:       ; %bb.0: ; %entry
 ; GFX803-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX803-NEXT:    buffer_load_ubyte v1, off, s[0:3], s32 offset:4095

 ; GFX900-NEXT:    global_store_dword v[0:1], v0, off
 ; GFX900-NEXT:    s_waitcnt vmcnt(0)
 ; GFX900-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX906-LABEL: load_private_hi_v2f16_reglo_vreg_zexti8:
 ; 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]
 ;
 ; GFX803-LABEL: load_private_hi_v2f16_reglo_vreg_zexti8:
 ; GFX803:       ; %bb.0: ; %entry
 ; GFX803-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX803-NEXT:    buffer_load_ubyte v1, off, s[0:3], s32 offset:4095

 ; GFX900-NEXT:    global_store_dword v[0:1], v0, off
 ; GFX900-NEXT:    s_waitcnt vmcnt(0)
 ; GFX900-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX906-LABEL: load_private_hi_v2f16_reglo_vreg_sexti8:
 ; 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]
 ;
 ; GFX803-LABEL: load_private_hi_v2f16_reglo_vreg_sexti8:
 ; GFX803:       ; %bb.0: ; %entry
 ; GFX803-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX803-NEXT:    buffer_load_sbyte v1, off, s[0:3], s32 offset:4095

 ; GFX900-NEXT:    global_store_dword v[0:1], v0, off
 ; GFX900-NEXT:    s_waitcnt vmcnt(0)
 ; GFX900-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX906-LABEL: load_private_hi_v2i16_reglo_vreg_sexti8:
 ; 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]
 ;
 ; GFX803-LABEL: load_private_hi_v2i16_reglo_vreg_sexti8:
 ; GFX803:       ; %bb.0: ; %entry
 ; GFX803-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX803-NEXT:    buffer_load_sbyte v1, off, s[0:3], s32 offset:4095

 ; GFX900-NEXT:    s_waitcnt vmcnt(0)
 ; GFX900-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX906-LABEL: load_private_hi_v2i16_reglo_vreg_nooff_zexti8:
 ; GFX906:       ; %bb.0: ; %entry
 ; 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]
 ;
 ; GFX803-LABEL: load_private_hi_v2i16_reglo_vreg_nooff_zexti8:
 ; GFX803:       ; %bb.0: ; %entry
 ; GFX803-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX803-NEXT:    buffer_load_ubyte v0, off, s[0:3], 0 offset:4094 glc

 ; GFX900-NEXT:    s_waitcnt vmcnt(0)
 ; GFX900-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX906-LABEL: load_private_hi_v2i16_reglo_vreg_nooff_sexti8:
 ; GFX906:       ; %bb.0: ; %entry
 ; 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]
 ;
 ; GFX803-LABEL: load_private_hi_v2i16_reglo_vreg_nooff_sexti8:
 ; GFX803:       ; %bb.0: ; %entry
 ; GFX803-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX803-NEXT:    buffer_load_sbyte v0, off, s[0:3], 0 offset:4094 glc

 ; GFX900-NEXT:    s_waitcnt vmcnt(0)
 ; GFX900-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX906-LABEL: load_private_hi_v2f16_reglo_vreg_nooff_zexti8:
 ; GFX906:       ; %bb.0: ; %entry
 ; 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]
 ;
 ; GFX803-LABEL: load_private_hi_v2f16_reglo_vreg_nooff_zexti8:
 ; GFX803:       ; %bb.0: ; %entry
 ; GFX803-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX803-NEXT:    buffer_load_ubyte v0, off, s[0:3], 0 offset:4094 glc

 ; GFX900-NEXT:    global_store_dword v[0:1], v2, off
 ; GFX900-NEXT:    s_waitcnt vmcnt(0)
 ; GFX900-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX906-LABEL: load_constant_hi_v2i16_reglo_vreg:
 ; 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]
 ;
 ; GFX803-LABEL: load_constant_hi_v2i16_reglo_vreg:
 ; GFX803:       ; %bb.0: ; %entry
 ; GFX803-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX803-NEXT:    v_add_u32_e32 v0, vcc, 0xfffff002, v0

 ; GFX900-NEXT:    global_store_dword v[0:1], v2, off
 ; GFX900-NEXT:    s_waitcnt vmcnt(0)
 ; GFX900-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX906-LABEL: load_constant_hi_v2f16_reglo_vreg:
 ; 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]
 ;
 ; GFX803-LABEL: load_constant_hi_v2f16_reglo_vreg:
 ; GFX803:       ; %bb.0: ; %entry
 ; GFX803-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX803-NEXT:    v_add_u32_e32 v0, vcc, 0xfffff002, v0

 ; GFX900-NEXT:    global_store_dword v[0:1], v2, off
 ; GFX900-NEXT:    s_waitcnt vmcnt(0)
 ; GFX900-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX906-LABEL: load_constant_hi_v2f16_reglo_vreg_sexti8:
 ; 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]
 ;
 ; GFX803-LABEL: load_constant_hi_v2f16_reglo_vreg_sexti8:
 ; GFX803:       ; %bb.0: ; %entry
 ; GFX803-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX803-NEXT:    v_add_u32_e32 v0, vcc, 0xfffff001, v0

 ; GFX900-NEXT:    global_store_dword v[0:1], v2, off
 ; GFX900-NEXT:    s_waitcnt vmcnt(0)
 ; GFX900-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX906-LABEL: load_constant_hi_v2f16_reglo_vreg_zexti8:
 ; 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]
 ;
 ; GFX803-LABEL: load_constant_hi_v2f16_reglo_vreg_zexti8:
 ; GFX803:       ; %bb.0: ; %entry
 ; GFX803-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX803-NEXT:    v_add_u32_e32 v0, vcc, 0xfffff001, v0

 ;
 ; GFX906-LABEL: load_private_hi_v2i16_reglo_vreg_to_offset:
 ; GFX906:       ; %bb.0: ; %entry
 ; GFX906-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX906-NEXT:    v_mov_b32_e32 v2, 0x7b
 ; 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]
 ;
 ; GFX803-LABEL: load_private_hi_v2i16_reglo_vreg_to_offset:
 ; GFX803:       ; %bb.0: ; %entry
 ; GFX803-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX803-NEXT:    v_mov_b32_e32 v2, 0x7b

 ;
 ; GFX906-LABEL: load_private_hi_v2i16_reglo_vreg_sexti8_to_offset:
 ; GFX906:       ; %bb.0: ; %entry
 ; GFX906-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX906-NEXT:    v_mov_b32_e32 v2, 0x7b
 ; 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]
 ;
 ; GFX803-LABEL: load_private_hi_v2i16_reglo_vreg_sexti8_to_offset:
 ; GFX803:       ; %bb.0: ; %entry
 ; GFX803-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX803-NEXT:    v_mov_b32_e32 v2, 0x7b

 ;
 ; GFX906-LABEL: load_private_hi_v2i16_reglo_vreg_zexti8_to_offset:
 ; GFX906:       ; %bb.0: ; %entry
 ; GFX906-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX906-NEXT:    v_mov_b32_e32 v2, 0x7b
 ; 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]
 ;
 ; GFX803-LABEL: load_private_hi_v2i16_reglo_vreg_zexti8_to_offset:
 ; GFX803:       ; %bb.0: ; %entry
 ; GFX803-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX803-NEXT:    v_mov_b32_e32 v2, 0x7b

 ; GFX900-NEXT:    v_mov_b32_e32 v0, v1
 ; GFX900-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX906-LABEL: load_local_v2i16_split_multi_chain:
 ; GFX906:       ; %bb.0: ; %entry
 ; 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:
 ; GFX803:       ; %bb.0: ; %entry
 ; GFX803-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX803-NEXT:    s_mov_b32 m0, -1
 ; GFX803-NEXT:    ds_read_u16 v1, v0
 ; GFX803-NEXT:    ds_read_u16 v0, v0 offset:2

 ; GFX900-NEXT:    v_mov_b32_e32 v0, v1
 ; GFX900-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX906-LABEL: load_local_lo_hi_v2i16_samechain:
 ; GFX906:       ; %bb.0: ; %entry
 ; 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:
 ; GFX803:       ; %bb.0: ; %entry
 ; GFX803-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX803-NEXT:    s_mov_b32 m0, -1
 ; GFX803-NEXT:    ds_read_u16 v1, v0 offset:16
 ; GFX803-NEXT:    ds_read_u16 v0, v0

 }
 
 ; FIXME: Remove and
 define <2 x i16> @load_local_v2i16_broadcast(i16 addrspace(3)* %in) #0 {
 ; GFX900-LABEL: load_local_v2i16_broadcast:
 ; 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:
 ; GFX803:       ; %bb.0: ; %entry
 ; GFX803-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX803-NEXT:    s_mov_b32 m0, -1
 ; GFX803-NEXT:    ds_read_u16 v0, v0
 ; GFX803-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX803-NEXT:    v_lshlrev_b32_e32 v1, 16, v0
 ; GFX803-NEXT:    v_or_b32_e32 v0, v0, v1
 ; GFX803-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX900-FLATSCR-LABEL: load_local_v2i16_broadcast:
 ; 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
   %load0 = load i16, i16 addrspace(3)* %in
   %build0 = insertelement <2 x i16> undef, i16 %load0, i32 0
   %build1 = insertelement <2 x i16> %build0, i16 %load0, i32 1
   ret <2 x i16> %build1
 }

 ;
 ; GFX906-LABEL: load_local_lo_hi_v2i16_side_effect:
 ; GFX906:       ; %bb.0: ; %entry
 ; GFX906-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX906-NEXT:    v_mov_b32_e32 v3, 0x7b
 ; 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:
 ; GFX803:       ; %bb.0: ; %entry
 ; GFX803-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX803-NEXT:    s_mov_b32 m0, -1
 ; GFX803-NEXT:    v_mov_b32_e32 v3, 0x7b
 ; GFX803-NEXT:    ds_read_u16 v2, v0

 ;
 ; GFX906-LABEL: load_global_v2i16_split:
 ; GFX906:       ; %bb.0: ; %entry
 ; GFX906-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX906-NEXT:    global_load_ushort v2, v[0:1], off glc
 ; 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:
 ; GFX803:       ; %bb.0: ; %entry
 ; GFX803-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX803-NEXT:    v_add_u32_e32 v2, vcc, 2, v0
 ; GFX803-NEXT:    v_addc_u32_e32 v3, vcc, 0, v1, vcc
 ; GFX803-NEXT:    flat_load_ushort v0, v[0:1] glc

 ; GFX900-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX906-LABEL: load_flat_v2i16_split:
 ; GFX906:       ; %bb.0: ; %entry
 ; GFX906-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; 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:
 ; GFX803:       ; %bb.0: ; %entry
 ; GFX803-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX803-NEXT:    v_add_u32_e32 v2, vcc, 2, v0
 ; GFX803-NEXT:    v_addc_u32_e32 v3, vcc, 0, v1, vcc
 ; GFX803-NEXT:    flat_load_ushort v0, v[0:1] glc

 ; GFX900-NEXT:    v_mov_b32_e32 v0, v2
 ; GFX900-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX906-LABEL: load_constant_v2i16_split:
 ; GFX906:       ; %bb.0: ; %entry
 ; 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:
 ; GFX803:       ; %bb.0: ; %entry
 ; GFX803-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX803-NEXT:    v_add_u32_e32 v2, vcc, 2, v0
 ; GFX803-NEXT:    v_addc_u32_e32 v3, vcc, 0, v1, vcc
 ; GFX803-NEXT:    flat_load_ushort v0, v[0:1] glc

 ;
 ; GFX906-LABEL: load_private_v2i16_split:
 ; GFX906:       ; %bb.0: ; %entry
 ; GFX906-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX906-NEXT:    buffer_load_ushort v0, off, s[0:3], s32 glc
 ; 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:
 ; GFX803:       ; %bb.0: ; %entry
 ; GFX803-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX803-NEXT:    buffer_load_ushort v0, off, s[0:3], s32 glc
 ; GFX803-NEXT:    s_waitcnt vmcnt(0)
 ; GFX803-NEXT:    buffer_load_ushort v1, off, s[0:3], s32 offset:2 glc

 ; GFX900-NEXT:    v_mov_b32_e32 v0, v2
 ; GFX900-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX900-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX906-LABEL: load_local_hi_v2i16_store_local_lo:
 ; 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]
 ;
 ; GFX803-LABEL: load_local_hi_v2i16_store_local_lo:
 ; GFX803:       ; %bb.0: ; %entry
 ; GFX803-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX803-NEXT:    s_mov_b32 m0, -1
entry:
   %load = load i16, i16 addrspace(3)* %in
   %build0 = insertelement <2 x i16> undef, i16 %reg, i32 0
   %build1 = insertelement <2 x i16> %build0, i16 %load, i32 1
   store volatile i16 %reg, i16 addrspace(3)* %in
   ret <2 x i16> %build1
 }
 
 attributes #0 = { nounwind }
+

llvm/test/CodeGen/AMDGPU/load-lo16.ll

 ; GFX803-NEXT:    s_setpc_b64 s[30:31]
 entry:
   %load = load i16, i16 addrspace(3)* %in
   %build = insertelement <2 x i16> undef, i16 %load, i32 0
   ret <2 x i16> %build
 }
 
 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:
 ; GFX803:       ; %bb.0: ; %entry
 ; GFX803-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX803-NEXT:    s_mov_b32 m0, -1
 ; GFX803-NEXT:    ds_read_u16 v0, v0
 ; GFX803-NEXT:    v_lshlrev_b32_e32 v1, 16, v1
 ; 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
   %build1 = insertelement <2 x i16> %build0, i16 %load, i32 0
   ret <2 x i16> %build1
 }
 
 ; 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]
 ;
 ; GFX803-LABEL: load_local_lo_v2i16_reglo_vreg:
 ; GFX803:       ; %bb.0: ; %entry
 ; GFX803-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX803-NEXT:    s_mov_b32 m0, -1
 ; GFX803-NEXT:    ds_read_u16 v0, v0
 ; GFX803-NEXT:    v_lshlrev_b32_e32 v1, 16, v1
 ; GFX803-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX803-NEXT:    v_or_b32_e32 v0, v0, v1
 ; 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
   %build1 = insertelement <2 x i16> %build0, i16 %load, i32 0
   store <2 x i16> %build1, <2 x i16> addrspace(1)* undef
   ret void
 }
 

 ; GFX900-NEXT:    v_mov_b32_e32 v0, v1
 ; GFX900-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX906-LABEL: load_local_lo_v2f16_fpimm:
 ; GFX906:       ; %bb.0: ; %entry
 ; 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:
 ; GFX803:       ; %bb.0: ; %entry
 ; GFX803-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX803-NEXT:    s_mov_b32 m0, -1
 ; GFX803-NEXT:    ds_read_u16 v0, v0
 ; GFX803-NEXT:    s_waitcnt lgkmcnt(0)

 ; GFX900-NEXT:    global_store_dword v[0:1], v1, off
 ; GFX900-NEXT:    s_waitcnt vmcnt(0)
 ; GFX900-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX906-LABEL: load_local_lo_v2f16_reghi_vreg:
 ; 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]
 ;
 ; GFX803-LABEL: load_local_lo_v2f16_reghi_vreg:
 ; GFX803:       ; %bb.0: ; %entry
 ; GFX803-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX803-NEXT:    s_mov_b32 m0, -1
 ; GFX803-NEXT:    ds_read_u16 v0, v0
 ; GFX803-NEXT:    v_and_b32_e32 v1, 0xffff0000, v1
 ; GFX803-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX803-NEXT:    v_or_b32_e32 v0, v0, v1
 ; GFX803-NEXT:    flat_store_dword v[0:1], v0
 ; GFX803-NEXT:    s_waitcnt vmcnt(0)
 ; GFX803-NEXT:    s_setpc_b64 s[30:31]
 entry:
   %reg.bc = bitcast i32 %reg to <2 x half>
   %load = load half, half addrspace(3)* %in
   %build1 = insertelement <2 x half> %reg.bc, half %load, i32 0
   store <2 x half> %build1, <2 x half> addrspace(1)* undef
   ret void
 }
 
 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]
 ;
 ; GFX803-LABEL: load_local_lo_v2f16_reglo_vreg:
 ; GFX803:       ; %bb.0: ; %entry
 ; GFX803-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX803-NEXT:    s_mov_b32 m0, -1
 ; GFX803-NEXT:    ds_read_u16 v0, v0
 ; GFX803-NEXT:    v_lshlrev_b32_e32 v1, 16, v1
 ; GFX803-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX803-NEXT:    v_or_b32_e32 v0, v0, v1
 ; 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
   %build1 = insertelement <2 x half> %build0, half %load, i32 0
   store <2 x half> %build1, <2 x half> addrspace(1)* undef
   ret void
 }
 
 define void @load_local_lo_v2i16_reghi_vreg_zexti8(i8 addrspace(3)* %in, i32 %reg) #0 {
 ; GFX900-LABEL: load_local_lo_v2i16_reghi_vreg_zexti8:
 ; GFX900:       ; %bb.0: ; %entry
 ; GFX900-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX900-NEXT:    ds_read_u8_d16 v1, v0
 ; GFX900-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX900-NEXT:    global_store_dword v[0:1], v1, off
 ; GFX900-NEXT:    s_waitcnt vmcnt(0)
 ; GFX900-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX906-LABEL: load_local_lo_v2i16_reghi_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:    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]
 ;
 ; GFX803-LABEL: load_local_lo_v2i16_reghi_vreg_zexti8:
 ; GFX803:       ; %bb.0: ; %entry
 ; GFX803-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX803-NEXT:    s_mov_b32 m0, -1
entry:
   %load = load i8, i8 addrspace(3)* %in
   %ext = zext i8 %load to i16
   %build1 = insertelement <2 x i16> %reg.bc, i16 %ext, i32 0
   store <2 x i16> %build1, <2 x i16> addrspace(1)* undef
   ret void
 }
 
 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]
 ;
 ; GFX803-LABEL: load_local_lo_v2i16_reglo_vreg_zexti8:
 ; GFX803:       ; %bb.0: ; %entry
 ; GFX803-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX803-NEXT:    s_mov_b32 m0, -1
 ; GFX803-NEXT:    ds_read_u8 v0, v0
 ; GFX803-NEXT:    v_lshlrev_b32_e32 v1, 16, v1
 ; GFX803-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX803-NEXT:    v_or_b32_e32 v0, v0, v1
 ; 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
   %build0 = insertelement <2 x i16> undef, i16 %reg, i32 1
   %build1 = insertelement <2 x i16> %build0, i16 %ext, i32 0
   store <2 x i16> %build1, <2 x i16> addrspace(1)* undef
   ret void
 }
 
 define void @load_local_lo_v2i16_reghi_vreg_sexti8(i8 addrspace(3)* %in, i32 %reg) #0 {
 ; GFX900-LABEL: load_local_lo_v2i16_reghi_vreg_sexti8:
 ; GFX900:       ; %bb.0: ; %entry
 ; GFX900-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX900-NEXT:    ds_read_i8_d16 v1, v0
 ; GFX900-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX900-NEXT:    global_store_dword v[0:1], v1, off
 ; GFX900-NEXT:    s_waitcnt vmcnt(0)
 ; GFX900-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX906-LABEL: load_local_lo_v2i16_reghi_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:    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]
 ;
 ; GFX803-LABEL: load_local_lo_v2i16_reghi_vreg_sexti8:
 ; GFX803:       ; %bb.0: ; %entry
 ; GFX803-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX803-NEXT:    s_mov_b32 m0, -1
entry:
   %load = load i8, i8 addrspace(3)* %in
   %ext = sext i8 %load to i16
   %build1 = insertelement <2 x i16> %reg.bc, i16 %ext, i32 0
   store <2 x i16> %build1, <2 x i16> addrspace(1)* undef
   ret void
 }
 
 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]
 ;
 ; GFX803-LABEL: load_local_lo_v2i16_reglo_vreg_sexti8:
 ; GFX803:       ; %bb.0: ; %entry
 ; GFX803-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX803-NEXT:    s_mov_b32 m0, -1
 ; GFX803-NEXT:    ds_read_i8 v0, v0
 ; GFX803-NEXT:    v_lshlrev_b32_e32 v1, 16, v1
 ; GFX803-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX803-NEXT:    v_or_b32_sdwa v0, v0, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD
 ; 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
   %build0 = insertelement <2 x i16> undef, i16 %reg, i32 1
   %build1 = insertelement <2 x i16> %build0, i16 %ext, i32 0
   store <2 x i16> %build1, <2 x i16> addrspace(1)* undef
   ret void
 }
 
 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]
 ;
 ; GFX803-LABEL: load_local_lo_v2f16_reglo_vreg_zexti8:
 ; GFX803:       ; %bb.0: ; %entry
 ; GFX803-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX803-NEXT:    s_mov_b32 m0, -1
 ; GFX803-NEXT:    ds_read_u8 v0, v0
 ; GFX803-NEXT:    v_lshlrev_b32_e32 v1, 16, v1
 ; GFX803-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX803-NEXT:    v_or_b32_e32 v0, v0, v1
 ; 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
   %bitcast = bitcast i16 %ext to half
   %build0 = insertelement <2 x half> undef, half %reg, i32 1
   %build1 = insertelement <2 x half> %build0, half %bitcast, i32 0
   store <2 x half> %build1, <2 x half> addrspace(1)* undef
   ret void
 }
 
 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]
 ;
 ; GFX803-LABEL: load_local_lo_v2f16_reglo_vreg_sexti8:
 ; GFX803:       ; %bb.0: ; %entry
 ; GFX803-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX803-NEXT:    s_mov_b32 m0, -1
 ; GFX803-NEXT:    ds_read_i8 v0, v0
 ; GFX803-NEXT:    v_lshlrev_b32_e32 v1, 16, v1
 ; GFX803-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX803-NEXT:    v_or_b32_sdwa v0, v0, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD
 ; 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
   %bitcast = bitcast i16 %ext to half
   %build0 = insertelement <2 x half> undef, half %reg, i32 1
   %build1 = insertelement <2 x half> %build0, half %bitcast, i32 0
   store <2 x half> %build1, <2 x half> addrspace(1)* undef
   ret void
 }
 
 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]
 ;
 ; GFX803-LABEL: load_local_lo_v2i16_reghi_vreg_multi_use_lo:
 ; GFX803:       ; %bb.0: ; %entry
 ; GFX803-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX803-NEXT:    s_mov_b32 m0, -1
 ; GFX803-NEXT:    ds_read_u16 v0, v0
 ; GFX803-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX803-NEXT:    v_and_b32_e32 v1, 0xffff0000, v1
 ; GFX803-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX803-NEXT:    ds_write_b16 v2, v0
 ; GFX803-NEXT:    v_or_b32_sdwa v0, v0, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD
 ; 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
   store i16 %load, i16 addrspace(3)* null
   %build1 = insertelement <2 x i16> %reg, i16 %load, i32 0
   store <2 x i16> %build1, <2 x i16> addrspace(1)* undef
   ret void
 }

 ; GFX900-NEXT:    global_store_dword v[0:1], v1, off
 ; GFX900-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX900-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX906-LABEL: load_local_lo_v2i16_reghi_vreg_multi_use_hi:
 ; 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]
 ;
 ; GFX803-LABEL: load_local_lo_v2i16_reghi_vreg_multi_use_hi:
 ; GFX803:       ; %bb.0: ; %entry
 ; GFX803-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX803-NEXT:    s_mov_b32 m0, -1
entry:
   %elt1 = extractelement <2 x i16> %reg, i32 1
   store i16 %elt1, i16 addrspace(3)* null
   %build1 = insertelement <2 x i16> %reg, i16 %load, i32 0
   store <2 x i16> %build1, <2 x i16> addrspace(1)* undef
   ret void
 }
 
 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]
 ;
 ; GFX803-LABEL: load_local_lo_v2i16_reghi_vreg_multi_use_lohi:
 ; GFX803:       ; %bb.0: ; %entry
 ; GFX803-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX803-NEXT:    s_mov_b32 m0, -1
 ; GFX803-NEXT:    ds_read_u16 v0, v0
 ; GFX803-NEXT:    v_lshrrev_b32_e32 v4, 16, v1
 ; GFX803-NEXT:    v_and_b32_e32 v1, 0xffff0000, v1
 ; GFX803-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX803-NEXT:    ds_write_b16 v2, v0
 ; GFX803-NEXT:    ds_write_b16 v3, v4
 ; GFX803-NEXT:    v_or_b32_sdwa v0, v0, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD
 ; 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
   store i16 %load, i16 addrspace(3)* %out0
   store i16 %elt1, i16 addrspace(3)* %out1
   %build1 = insertelement <2 x i16> %reg, i16 %load, i32 0
   store <2 x i16> %build1, <2 x i16> addrspace(1)* undef
   ret void
 }
 
 define void @load_global_lo_v2i16_reglo_vreg(i16 addrspace(1)* %in, i32 %reg) #0 {
 ; GFX900-LABEL: load_global_lo_v2i16_reglo_vreg:
 ; GFX900:       ; %bb.0: ; %entry
 ; GFX900-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX900-NEXT:    global_load_short_d16 v2, v[0:1], off offset:-4094
 ; GFX900-NEXT:    s_waitcnt vmcnt(0)
 ; GFX900-NEXT:    global_store_dword v[0:1], v2, off
 ; GFX900-NEXT:    s_waitcnt vmcnt(0)
 ; GFX900-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX906-LABEL: load_global_lo_v2i16_reglo_vreg:
 ; 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]
 ;
 ; GFX803-LABEL: load_global_lo_v2i16_reglo_vreg:
 ; GFX803:       ; %bb.0: ; %entry
 ; GFX803-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX803-NEXT:    v_add_u32_e32 v0, vcc, 0xfffff002, v0

 ; GFX900-NEXT:    global_store_dword v[0:1], v2, off
 ; GFX900-NEXT:    s_waitcnt vmcnt(0)
 ; GFX900-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX906-LABEL: load_global_lo_v2f16_reglo_vreg:
 ; 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]
 ;
 ; GFX803-LABEL: load_global_lo_v2f16_reglo_vreg:
 ; GFX803:       ; %bb.0: ; %entry
 ; GFX803-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX803-NEXT:    v_add_u32_e32 v0, vcc, 0xfffff002, v0

 ; GFX900-NEXT:    global_store_dword v[0:1], v2, off
 ; GFX900-NEXT:    s_waitcnt vmcnt(0)
 ; GFX900-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX906-LABEL: load_global_lo_v2i16_reglo_vreg_zexti8:
 ; 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]
 ;
 ; GFX803-LABEL: load_global_lo_v2i16_reglo_vreg_zexti8:
 ; GFX803:       ; %bb.0: ; %entry
 ; GFX803-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX803-NEXT:    v_add_u32_e32 v0, vcc, 0xfffff001, v0

 ; GFX900-NEXT:    global_store_dword v[0:1], v2, off
 ; GFX900-NEXT:    s_waitcnt vmcnt(0)
 ; GFX900-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX906-LABEL: load_global_lo_v2i16_reglo_vreg_sexti8:
 ; 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]
 ;
 ; GFX803-LABEL: load_global_lo_v2i16_reglo_vreg_sexti8:
 ; GFX803:       ; %bb.0: ; %entry
 ; GFX803-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX803-NEXT:    v_add_u32_e32 v0, vcc, 0xfffff001, v0

 ; GFX900-NEXT:    global_store_dword v[0:1], v2, off
 ; GFX900-NEXT:    s_waitcnt vmcnt(0)
 ; GFX900-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX906-LABEL: load_global_lo_v2f16_reglo_vreg_zexti8:
 ; 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]
 ;
 ; GFX803-LABEL: load_global_lo_v2f16_reglo_vreg_zexti8:
 ; GFX803:       ; %bb.0: ; %entry
 ; GFX803-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX803-NEXT:    v_add_u32_e32 v0, vcc, 0xfffff001, v0

 ; GFX900-NEXT:    global_store_dword v[0:1], v2, off
 ; GFX900-NEXT:    s_waitcnt vmcnt(0)
 ; GFX900-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX906-LABEL: load_global_lo_v2f16_reglo_vreg_sexti8:
 ; 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]
 ;
 ; GFX803-LABEL: load_global_lo_v2f16_reglo_vreg_sexti8:
 ; GFX803:       ; %bb.0: ; %entry
 ; GFX803-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX803-NEXT:    v_add_u32_e32 v0, vcc, 0xfffff001, v0

 ; GFX900-NEXT:    global_store_dword v[0:1], v2, off
 ; GFX900-NEXT:    s_waitcnt vmcnt(0)
 ; GFX900-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX906-LABEL: load_flat_lo_v2i16_reghi_vreg:
 ; 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]
 ;
 ; GFX803-LABEL: load_flat_lo_v2i16_reghi_vreg:
 ; GFX803:       ; %bb.0: ; %entry
 ; GFX803-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX803-NEXT:    flat_load_ushort v0, v[0:1]

 ; GFX900-NEXT:    global_store_dword v[0:1], v2, off
 ; GFX900-NEXT:    s_waitcnt vmcnt(0)
 ; GFX900-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX906-LABEL: load_flat_lo_v2f16_reghi_vreg:
 ; 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]
 ;
 ; GFX803-LABEL: load_flat_lo_v2f16_reghi_vreg:
 ; GFX803:       ; %bb.0: ; %entry
 ; GFX803-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX803-NEXT:    flat_load_ushort v0, v[0:1]

 ; GFX900-NEXT:    global_store_dword v[0:1], v2, off
 ; GFX900-NEXT:    s_waitcnt vmcnt(0)
 ; GFX900-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX906-LABEL: load_flat_lo_v2i16_reglo_vreg_zexti8:
 ; 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]
 ;
 ; GFX803-LABEL: load_flat_lo_v2i16_reglo_vreg_zexti8:
 ; GFX803:       ; %bb.0: ; %entry
 ; GFX803-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX803-NEXT:    flat_load_ubyte v0, v[0:1]

 ; GFX900-NEXT:    global_store_dword v[0:1], v2, off
 ; GFX900-NEXT:    s_waitcnt vmcnt(0)
 ; GFX900-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX906-LABEL: load_flat_lo_v2i16_reglo_vreg_sexti8:
 ; 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]
 ;
 ; GFX803-LABEL: load_flat_lo_v2i16_reglo_vreg_sexti8:
 ; GFX803:       ; %bb.0: ; %entry
 ; GFX803-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX803-NEXT:    flat_load_sbyte v0, v[0:1]

 ; GFX900-NEXT:    global_store_dword v[0:1], v2, off
 ; GFX900-NEXT:    s_waitcnt vmcnt(0)
 ; GFX900-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX906-LABEL: load_flat_lo_v2f16_reglo_vreg_zexti8:
 ; 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]
 ;
 ; GFX803-LABEL: load_flat_lo_v2f16_reglo_vreg_zexti8:
 ; GFX803:       ; %bb.0: ; %entry
 ; GFX803-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX803-NEXT:    flat_load_ubyte v0, v[0:1]

 ; GFX900-NEXT:    global_store_dword v[0:1], v2, off
 ; GFX900-NEXT:    s_waitcnt vmcnt(0)
 ; GFX900-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX906-LABEL: load_flat_lo_v2f16_reglo_vreg_sexti8:
 ; 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]
 ;
 ; GFX803-LABEL: load_flat_lo_v2f16_reglo_vreg_sexti8:
 ; GFX803:       ; %bb.0: ; %entry
 ; GFX803-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX803-NEXT:    flat_load_sbyte v0, v[0:1]

 ; 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_private_lo_v2i16_reglo_vreg:
 ; 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]
 ;
 ; GFX803-LABEL: load_private_lo_v2i16_reglo_vreg:
 ; GFX803:       ; %bb.0: ; %entry
 ; GFX803-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX803-NEXT:    buffer_load_ushort v1, off, s[0:3], s32 offset:4094
entry:
   ret void
 }
 
 define void @load_private_lo_v2i16_reghi_vreg(i16 addrspace(5)* byval(i16) %in, i16 %reg) #0 {
 ; GFX900-MUBUF-LABEL: load_private_lo_v2i16_reghi_vreg:
 ; 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]
 ;
 ; GFX906-LABEL: load_private_lo_v2i16_reghi_vreg:
 ; 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]
 ;
 ; GFX803-LABEL: load_private_lo_v2i16_reghi_vreg:
 ; GFX803:       ; %bb.0: ; %entry
 ; GFX803-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX803-NEXT:    buffer_load_ushort v1, off, s[0:3], s32 offset:4094
 ; GFX803-NEXT:    v_lshlrev_b32_e32 v0, 16, v0
 ; GFX803-NEXT:    s_waitcnt vmcnt(0)
 ; GFX803-NEXT:    v_or_b32_e32 v0, v1, v0
 ; 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_private_lo_v2i16_reghi_vreg:
 ; 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]
 entry:
   %gep = getelementptr inbounds i16, i16 addrspace(5)* %in, i64 2047
   %load = load i16, i16 addrspace(5)* %gep
   %build0 = insertelement <2 x i16> undef, i16 %reg, i32 1
   %build1 = insertelement <2 x i16> %build0, i16 %load, i32 0

 ; 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_private_lo_v2f16_reglo_vreg:
 ; 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]
 ;
 ; GFX803-LABEL: load_private_lo_v2f16_reglo_vreg:
 ; GFX803:       ; %bb.0: ; %entry
 ; GFX803-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX803-NEXT:    buffer_load_ushort v1, off, s[0:3], s32 offset:4094

 ; GFX900-MUBUF-NEXT:    s_waitcnt vmcnt(0)
 ; GFX900-MUBUF-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX906-LABEL: load_private_lo_v2i16_reglo_vreg_nooff:
 ; GFX906:       ; %bb.0: ; %entry
 ; 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]
 ;
 ; GFX803-LABEL: load_private_lo_v2i16_reglo_vreg_nooff:
 ; GFX803:       ; %bb.0: ; %entry
 ; GFX803-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX803-NEXT:    buffer_load_ushort v0, off, s[0:3], 0 offset:4094 glc

 ; GFX900-MUBUF-NEXT:    s_waitcnt vmcnt(0)
 ; GFX900-MUBUF-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX906-LABEL: load_private_lo_v2i16_reghi_vreg_nooff:
 ; GFX906:       ; %bb.0: ; %entry
 ; 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]
 ;
 ; GFX803-LABEL: load_private_lo_v2i16_reghi_vreg_nooff:
 ; GFX803:       ; %bb.0: ; %entry
 ; GFX803-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX803-NEXT:    buffer_load_ushort v0, off, s[0:3], 0 offset:4094 glc

 ; GFX900-MUBUF-NEXT:    s_waitcnt vmcnt(0)
 ; GFX900-MUBUF-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX906-LABEL: load_private_lo_v2f16_reglo_vreg_nooff:
 ; GFX906:       ; %bb.0: ; %entry
 ; 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]
 ;
 ; GFX803-LABEL: load_private_lo_v2f16_reglo_vreg_nooff:
 ; GFX803:       ; %bb.0: ; %entry
 ; GFX803-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX803-NEXT:    buffer_load_ushort v0, off, s[0:3], 0 offset:4094 glc

 ; 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_private_lo_v2i16_reglo_vreg_zexti8:
 ; 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]
 ;
 ; GFX803-LABEL: load_private_lo_v2i16_reglo_vreg_zexti8:
 ; GFX803:       ; %bb.0: ; %entry
 ; GFX803-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX803-NEXT:    buffer_load_ubyte v1, off, s[0:3], s32 offset:4095

 ; 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_private_lo_v2i16_reglo_vreg_sexti8:
 ; 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]
 ;
 ; GFX803-LABEL: load_private_lo_v2i16_reglo_vreg_sexti8:
 ; GFX803:       ; %bb.0: ; %entry
 ; GFX803-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX803-NEXT:    buffer_load_sbyte v1, off, s[0:3], s32 offset:4095

 ; GFX900-MUBUF-NEXT:    s_waitcnt vmcnt(0)
 ; GFX900-MUBUF-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX906-LABEL: load_private_lo_v2i16_reglo_vreg_nooff_zexti8:
 ; GFX906:       ; %bb.0: ; %entry
 ; 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]
 ;
 ; GFX803-LABEL: load_private_lo_v2i16_reglo_vreg_nooff_zexti8:
 ; GFX803:       ; %bb.0: ; %entry
 ; GFX803-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX803-NEXT:    buffer_load_ubyte v0, off, s[0:3], 0 offset:4094 glc

 ; GFX900-MUBUF-NEXT:    s_waitcnt vmcnt(0)
 ; GFX900-MUBUF-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX906-LABEL: load_private_lo_v2i16_reglo_vreg_nooff_sexti8:
 ; GFX906:       ; %bb.0: ; %entry
 ; 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]
 ;
 ; GFX803-LABEL: load_private_lo_v2i16_reglo_vreg_nooff_sexti8:
 ; GFX803:       ; %bb.0: ; %entry
 ; GFX803-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX803-NEXT:    buffer_load_sbyte v0, off, s[0:3], 0 offset:4094 glc

 ; GFX900-MUBUF-NEXT:    s_waitcnt vmcnt(0)
 ; GFX900-MUBUF-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX906-LABEL: load_private_lo_v2f16_reglo_vreg_nooff_zexti8:
 ; GFX906:       ; %bb.0: ; %entry
 ; 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]
 ;
 ; GFX803-LABEL: load_private_lo_v2f16_reglo_vreg_nooff_zexti8:
 ; GFX803:       ; %bb.0: ; %entry
 ; GFX803-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX803-NEXT:    buffer_load_ubyte v0, off, s[0:3], 0 offset:4094 glc

 ; GFX900-NEXT:    global_store_dword v[0:1], v2, off
 ; GFX900-NEXT:    s_waitcnt vmcnt(0)
 ; GFX900-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX906-LABEL: load_constant_lo_v2i16_reglo_vreg:
 ; 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]
 ;
 ; GFX803-LABEL: load_constant_lo_v2i16_reglo_vreg:
 ; GFX803:       ; %bb.0: ; %entry
 ; GFX803-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX803-NEXT:    v_add_u32_e32 v0, vcc, 0xfffff002, v0

 ; GFX900-NEXT:    global_store_dword v[0:1], v2, off
 ; GFX900-NEXT:    s_waitcnt vmcnt(0)
 ; GFX900-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX906-LABEL: load_constant_lo_v2f16_reglo_vreg:
 ; 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]
 ;
 ; GFX803-LABEL: load_constant_lo_v2f16_reglo_vreg:
 ; GFX803:       ; %bb.0: ; %entry
 ; GFX803-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX803-NEXT:    v_add_u32_e32 v0, vcc, 0xfffff002, v0

 ; GFX900-NEXT:    global_store_dword v[0:1], v2, off
 ; GFX900-NEXT:    s_waitcnt vmcnt(0)
 ; GFX900-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX906-LABEL: load_constant_lo_v2f16_reglo_vreg_zexti8:
 ; 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]
 ;
 ; GFX803-LABEL: load_constant_lo_v2f16_reglo_vreg_zexti8:
 ; GFX803:       ; %bb.0: ; %entry
 ; GFX803-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX803-NEXT:    v_add_u32_e32 v0, vcc, 0xfffff001, v0

 ; GFX900-NEXT:    global_store_dword v[0:1], v2, off
 ; GFX900-NEXT:    s_waitcnt vmcnt(0)
 ; GFX900-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX906-LABEL: load_constant_lo_v2f16_reglo_vreg_sexti8:
 ; 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]
 ;
 ; GFX803-LABEL: load_constant_lo_v2f16_reglo_vreg_sexti8:
 ; GFX803:       ; %bb.0: ; %entry
 ; GFX803-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX803-NEXT:    v_add_u32_e32 v0, vcc, 0xfffff001, v0

 ; GFX900-MUBUF-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX906-LABEL: load_private_lo_v2i16_reglo_vreg_to_offset:
 ; GFX906:       ; %bb.0: ; %entry
 ; GFX906-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; 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]
 ;
 ; GFX803-LABEL: load_private_lo_v2i16_reglo_vreg_to_offset:
 ; GFX803:       ; %bb.0: ; %entry
 ; GFX803-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX803-NEXT:    v_mov_b32_e32 v1, 0x7b

 ; GFX900-MUBUF-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX906-LABEL: load_private_lo_v2i16_reglo_vreg_sexti8_to_offset:
 ; GFX906:       ; %bb.0: ; %entry
 ; GFX906-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; 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]
 ;
 ; GFX803-LABEL: load_private_lo_v2i16_reglo_vreg_sexti8_to_offset:
 ; GFX803:       ; %bb.0: ; %entry
 ; GFX803-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX803-NEXT:    v_mov_b32_e32 v1, 0x7b

 ; GFX900-MUBUF-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX906-LABEL: load_private_lo_v2i16_reglo_vreg_zexti8_to_offset:
 ; GFX906:       ; %bb.0: ; %entry
 ; GFX906-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; 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]
 ;
 ; GFX803-LABEL: load_private_lo_v2i16_reglo_vreg_zexti8_to_offset:
 ; GFX803:       ; %bb.0: ; %entry
 ; GFX803-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX803-NEXT:    v_mov_b32_e32 v1, 0x7b

 ; GFX906:       ; %bb.0: ; %entry
 ; GFX906-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; 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 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]
 ;
 ; GFX803-LABEL: load_private_lo_v2f16_reglo_vreg_sexti8_to_offset:
 ; GFX803:       ; %bb.0: ; %entry
 ; GFX803-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX803-NEXT:    v_mov_b32_e32 v1, 0x7b

 ; GFX906:       ; %bb.0: ; %entry
 ; GFX906-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; 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 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]
 ;
 ; GFX803-LABEL: load_private_lo_v2f16_reglo_vreg_zexti8_to_offset:
 ; GFX803:       ; %bb.0: ; %entry
 ; GFX803-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX803-NEXT:    v_mov_b32_e32 v1, 0x7b

llvm/test/CodeGen/AMDGPU/pack.v2f16.ll

 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc -mtriple=amdgcn--amdhsa -mcpu=gfx900 -mattr=-flat-for-global -denormal-fp-math=preserve-sign -verify-machineinstrs < %s | FileCheck --check-prefixes=GFX9 %s
 ; RUN: llc -mtriple=amdgcn--amdhsa -mcpu=fiji -mattr=-flat-for-global -verify-machineinstrs < %s | FileCheck -check-prefix=GFX8 %s
 ; RUN: llc -mtriple=amdgcn--amdhsa -mcpu=kaveri -mattr=-flat-for-global -verify-machineinstrs < %s | FileCheck -check-prefix=GFX7 %s
 
 
 define amdgpu_kernel void @s_pack_v2f16(i32 addrspace(4)* %in0, i32 addrspace(4)* %in1) #0 {
 ; GFX9-LABEL: s_pack_v2f16:
 ; GFX9:       ; %bb.0:

 ; 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:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    global_load_dword v1, v0, s[0:1] glc
 ; 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
 ; GFX9-NEXT:    s_endpgm
 ;
 ; GFX8-LABEL: v_pack_v2f16:
 ; GFX8:       ; %bb.0:
 ; GFX8-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; 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:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    global_load_dword v1, v0, s[0:1] glc
 ; 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)
 ; GFX9-NEXT:    s_endpgm
 ;
 ; GFX8-LABEL: v_pack_v2f16_user:
 ; GFX8:       ; %bb.0:
 ; GFX8-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
; GFX7-NEXT:    s_endpgm
   ret void
 }
 
 define amdgpu_kernel void @v_pack_v2f16_imm_lo(i32 addrspace(1)* %in1) #0 {
 ; GFX9-LABEL: v_pack_v2f16_imm_lo:
 ; 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
 ; GFX9-NEXT:    s_endpgm
 ;
 ; GFX8-LABEL: v_pack_v2f16_imm_lo:
 ; GFX8:       ; %bb.0:
 ; GFX8-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
; GFX7-NEXT:    s_endpgm
   ret void
 }
 
 define amdgpu_kernel void @v_pack_v2f16_inline_imm_lo(i32 addrspace(1)* %in1) #0 {
 ; GFX9-LABEL: v_pack_v2f16_inline_imm_lo:
 ; 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
 ; GFX9-NEXT:    s_endpgm
 ;
 ; GFX8-LABEL: v_pack_v2f16_inline_imm_lo:
 ; GFX8:       ; %bb.0:
 ; GFX8-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
; GFX7-NEXT:    s_endpgm
   ret void
 }
 
 define amdgpu_kernel void @v_pack_v2f16_imm_hi(i32 addrspace(1)* %in0) #0 {
 ; GFX9-LABEL: v_pack_v2f16_imm_hi:
 ; 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
 ; GFX9-NEXT:    s_endpgm
 ;
 ; GFX8-LABEL: v_pack_v2f16_imm_hi:
 ; GFX8:       ; %bb.0:
 ; GFX8-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
; GFX7-NEXT:    s_endpgm
   ret void
 }
 
 define amdgpu_kernel void @v_pack_v2f16_inline_f16imm_hi(i32 addrspace(1)* %in0) #0 {
 ; GFX9-LABEL: v_pack_v2f16_inline_f16imm_hi:
 ; 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
 ; GFX9-NEXT:    s_endpgm
 ;
 ; GFX8-LABEL: v_pack_v2f16_inline_f16imm_hi:
 ; GFX8:       ; %bb.0:
 ; GFX8-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
; GFX7-NEXT:    s_endpgm
   ret void
 }
 
 define amdgpu_kernel void @v_pack_v2f16_inline_imm_hi(i32 addrspace(1)* %in0) #0 {
 ; GFX9-LABEL: v_pack_v2f16_inline_imm_hi:
 ; 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
 ; GFX9-NEXT:    s_endpgm
 ;
 ; GFX8-LABEL: v_pack_v2f16_inline_imm_hi:
 ; GFX8:       ; %bb.0:
 ; GFX8-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0

llvm/test/CodeGen/AMDGPU/pack.v2i16.ll

 ; 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:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    global_load_dword v1, v0, s[0:1] glc
 ; 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
 ; GFX9-NEXT:    s_endpgm
 ;
 ; GFX803-LABEL: v_pack_v2i16:
 ; GFX803:       ; %bb.0:
 ; GFX803-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; 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:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    global_load_dword v1, v0, s[0:1] glc
 ; 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)
 ; GFX9-NEXT:    s_endpgm
 ;
 ; GFX803-LABEL: v_pack_v2i16_user:
 ; GFX803:       ; %bb.0:
 ; GFX803-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
; GFX7-NEXT:    s_endpgm
   ret void
 }
 
 define amdgpu_kernel void @v_pack_v2i16_imm_lo(i32 addrspace(1)* %in1) #0 {
 ; GFX9-LABEL: v_pack_v2i16_imm_lo:
 ; 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
 ; GFX9-NEXT:    s_endpgm
 ;
 ; GFX803-LABEL: v_pack_v2i16_imm_lo:
 ; GFX803:       ; %bb.0:
 ; GFX803-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
; GFX7-NEXT:    s_endpgm
   ret void
 }
 
 define amdgpu_kernel void @v_pack_v2i16_inline_imm_lo(i32 addrspace(1)* %in1) #0 {
 ; GFX9-LABEL: v_pack_v2i16_inline_imm_lo:
 ; 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
 ; GFX9-NEXT:    s_endpgm
 ;
 ; GFX803-LABEL: v_pack_v2i16_inline_imm_lo:
 ; GFX803:       ; %bb.0:
 ; GFX803-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
; GFX7-NEXT:    s_endpgm
   ret void
 }
 
 define amdgpu_kernel void @v_pack_v2i16_imm_hi(i32 addrspace(1)* %in0) #0 {
 ; GFX9-LABEL: v_pack_v2i16_imm_hi:
 ; 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
 ; GFX9-NEXT:    s_endpgm
 ;
 ; GFX803-LABEL: v_pack_v2i16_imm_hi:
 ; GFX803:       ; %bb.0:
 ; GFX803-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
; GFX7-NEXT:    s_endpgm
   ret void
 }
 
 define amdgpu_kernel void @v_pack_v2i16_inline_imm_hi(i32 addrspace(1)* %in0) #0 {
 ; GFX9-LABEL: v_pack_v2i16_inline_imm_hi:
 ; 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
 ; GFX9-NEXT:    s_endpgm
 ;
 ; GFX803-LABEL: v_pack_v2i16_inline_imm_hi:
 ; GFX803:       ; %bb.0:
 ; GFX803-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
; GFX7-NEXT:    s_endpgm
   call void asm sideeffect "; use $0", "v"(i32 %vec.i32) #0
   ret void
 }
 
 declare i32 @llvm.amdgcn.workitem.id.x() #1
 
 attributes #0 = { nounwind }
 attributes #1 = { nounwind readnone }
+

llvm/test/CodeGen/AMDGPU/strict_fadd.f16.ll

 define <4 x half> @v_constained_fadd_v4f16_fpexcept_strict(<4 x half> %x, <4 x half> %y) #0 {
 ; GFX9-LABEL: v_constained_fadd_v4f16_fpexcept_strict:
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX9-NEXT:    v_add_f16_sdwa v4, v1, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
 ; 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:
 ; GFX8:       ; %bb.0:
 ; GFX8-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX8-NEXT:    v_add_f16_sdwa v4, v1, v3 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
 ; GFX8-NEXT:    v_add_f16_sdwa v5, v0, v2 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
 ; GFX8-NEXT:    v_add_f16_e32 v1, v1, v3
 ; GFX8-NEXT:    v_add_f16_e32 v0, v0, v2
 ; GFX8-NEXT:    v_or_b32_e32 v0, v0, v5
 ; GFX8-NEXT:    v_or_b32_e32 v1, v1, v4
 ; GFX8-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX10-LABEL: v_constained_fadd_v4f16_fpexcept_strict:
 ; 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:
 ; 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 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
 }
 
 define amdgpu_ps half @s_constained_fadd_f16_fpexcept_strict(half inreg %x, half inreg %y) #0 {
 ; GCN-LABEL: s_constained_fadd_f16_fpexcept_strict:
 ; GCN:       ; %bb.0:

llvm/test/CodeGen/AMDGPU/strict_fma.f16.ll

 ; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX9-NEXT:    v_lshrrev_b32_e32 v6, 16, v5
 ; GFX9-NEXT:    v_lshrrev_b32_e32 v7, 16, v3
 ; GFX9-NEXT:    v_lshrrev_b32_e32 v8, 16, v1
 ; GFX9-NEXT:    v_fma_f16 v6, v8, v7, v6
 ; 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:
 ; GFX8:       ; %bb.0:
 ; GFX8-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX8-NEXT:    v_lshrrev_b32_e32 v6, 16, v5
 ; GFX8-NEXT:    v_lshrrev_b32_e32 v7, 16, v3
 ; GFX8-NEXT:    v_lshrrev_b32_e32 v8, 16, v1

 ; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-NEXT:    v_lshrrev_b32_e32 v6, 16, v5
 ; GFX10-NEXT:    v_lshrrev_b32_e32 v7, 16, v3
 ; 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, 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:
 ; 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, v5
 ; GFX11-NEXT:    v_lshrrev_b32_e32 v7, 16, v3
 ; 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, 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
 }
 
 define half @v_constained_fma_f16_fpexcept_strict_fneg(half %x, half %y, half %z) #0 {
 ; GCN-LABEL: v_constained_fma_f16_fpexcept_strict_fneg:
 ; GCN:       ; %bb.0:

llvm/test/CodeGen/AMDGPU/strict_fmul.f16.ll

 define <4 x half> @v_constained_fmul_v4f16_fpexcept_strict(<4 x half> %x, <4 x half> %y) #0 {
 ; GFX9-LABEL: v_constained_fmul_v4f16_fpexcept_strict:
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX9-NEXT:    v_mul_f16_sdwa v4, v1, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
 ; 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:
 ; GFX8:       ; %bb.0:
 ; GFX8-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX8-NEXT:    v_mul_f16_sdwa v4, v1, v3 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
 ; GFX8-NEXT:    v_mul_f16_sdwa v5, v0, v2 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
 ; GFX8-NEXT:    v_mul_f16_e32 v1, v1, v3
 ; GFX8-NEXT:    v_mul_f16_e32 v0, v0, v2
 ; GFX8-NEXT:    v_or_b32_e32 v0, v0, v5
 ; GFX8-NEXT:    v_or_b32_e32 v1, v1, v4
 ; GFX8-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX10-LABEL: v_constained_fmul_v4f16_fpexcept_strict:
 ; 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:
 ; 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 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
 }
 
 define amdgpu_ps half @s_constained_fmul_f16_fpexcept_strict(half inreg %x, half inreg %y) #0 {
 ; GCN-LABEL: s_constained_fmul_f16_fpexcept_strict:
 ; GCN:       ; %bb.0:

llvm/test/CodeGen/AMDGPU/strict_fsub.f16.ll

 }
 
 define <2 x half> @v_constained_fsub_v2f16_fpexcept_strict(<2 x half> %x, <2 x half> %y) #0 {
 ; GFX9-LABEL: v_constained_fsub_v2f16_fpexcept_strict:
 ; GFX9:       ; %bb.0:
 ; 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:
 ; GFX8:       ; %bb.0:
 ; GFX8-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX8-NEXT:    v_sub_f16_sdwa v2, v0, v1 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
 ; GFX8-NEXT:    v_sub_f16_e32 v0, v0, v1
 ; GFX8-NEXT:    v_or_b32_e32 v0, v0, v2
 ; GFX8-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX10-LABEL: v_constained_fsub_v2f16_fpexcept_strict:
 ; 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:
 ; 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 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
 }
 
 define <2 x half> @v_constained_fsub_v2f16_fpexcept_ignore(<2 x half> %x, <2 x half> %y) #0 {
 ; GFX9-LABEL: v_constained_fsub_v2f16_fpexcept_ignore:
 ; GFX9:       ; %bb.0:
 ; 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:
 ; GFX8:       ; %bb.0:
 ; GFX8-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX8-NEXT:    v_sub_f16_sdwa v2, v0, v1 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
 ; GFX8-NEXT:    v_sub_f16_e32 v0, v0, v1
 ; GFX8-NEXT:    v_or_b32_e32 v0, v0, v2
 ; GFX8-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX10-LABEL: v_constained_fsub_v2f16_fpexcept_ignore:
 ; 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:
 ; 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 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
 }
 
 define <2 x half> @v_constained_fsub_v2f16_fpexcept_maytrap(<2 x half> %x, <2 x half> %y) #0 {
 ; GFX9-LABEL: v_constained_fsub_v2f16_fpexcept_maytrap:
 ; GFX9:       ; %bb.0:
 ; 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:
 ; GFX8:       ; %bb.0:
 ; GFX8-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX8-NEXT:    v_sub_f16_sdwa v2, v0, v1 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
 ; GFX8-NEXT:    v_sub_f16_e32 v0, v0, v1
 ; GFX8-NEXT:    v_or_b32_e32 v0, v0, v2
 ; GFX8-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX10-LABEL: v_constained_fsub_v2f16_fpexcept_maytrap:
 ; 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:
 ; 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 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
 }
 
 define <3 x half> @v_constained_fsub_v3f16_fpexcept_strict(<3 x half> %x, <3 x half> %y) #0 {
 ; GFX9-LABEL: v_constained_fsub_v3f16_fpexcept_strict:
 ; GFX9:       ; %bb.0:
 ; 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]
 ;
 ; GFX8-LABEL: v_constained_fsub_v3f16_fpexcept_strict:
 ; GFX8:       ; %bb.0:
 ; GFX8-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX8-NEXT:    v_sub_f16_sdwa v4, v0, v2 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
 ; GFX8-NEXT:    v_sub_f16_e32 v0, v0, v2
 ; GFX8-NEXT:    v_or_b32_e32 v0, v0, v4
 ; GFX8-NEXT:    v_sub_f16_e32 v1, v1, v3
 ; GFX8-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX10-LABEL: v_constained_fsub_v3f16_fpexcept_strict:
 ; 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:
 ; 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 v4, 16, v2
 ; GFX11-NEXT:    v_lshrrev_b32_e32 v5, 16, v0
 ; 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
 }
 
 ; FIXME: Scalarized
 define <4 x half> @v_constained_fsub_v4f16_fpexcept_strict(<4 x half> %x, <4 x half> %y) #0 {
 ; GFX9-LABEL: v_constained_fsub_v4f16_fpexcept_strict:
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX9-NEXT:    v_sub_f16_sdwa v4, v1, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
 ; 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:
 ; GFX8:       ; %bb.0:
 ; GFX8-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX8-NEXT:    v_sub_f16_sdwa v4, v1, v3 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
 ; GFX8-NEXT:    v_sub_f16_sdwa v5, v0, v2 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
 ; GFX8-NEXT:    v_sub_f16_e32 v1, v1, v3
 ; GFX8-NEXT:    v_sub_f16_e32 v0, v0, v2
 ; GFX8-NEXT:    v_or_b32_e32 v0, v0, v5
 ; GFX8-NEXT:    v_or_b32_e32 v1, v1, v4
 ; GFX8-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX10-LABEL: v_constained_fsub_v4f16_fpexcept_strict:
 ; 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:
 ; 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 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
 }
 
 define amdgpu_ps half @s_constained_fsub_f16_fpexcept_strict(half inreg %x, half inreg %y) #0 {
 ; GCN-LABEL: s_constained_fsub_f16_fpexcept_strict:
 ; GCN:       ; %bb.0:

llvm/test/CodeGen/AMDGPU/sub.v2i16.ll

 
 ; FIXME: Need to handle non-uniform case for function below (load without gep).
 define amdgpu_kernel void @v_test_sub_v2i16_zext_to_v2i64(<2 x i64> addrspace(1)* %out, <2 x i16> addrspace(1)* %in0, <2 x i16> addrspace(1)* %in1) #1 {
 ; GFX9-LABEL: v_test_sub_v2i16_zext_to_v2i64:
 ; GFX9:       ; %bb.0:
 ; 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
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    global_load_dword v3, v0, s[2:3] glc
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    v_mov_b32_e32 v0, 0x5040302
 ; 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 v3, v2, v3
+; GFX9-NEXT:    v_perm_b32 v2, 0, v3, v0
+; GFX9-NEXT:    v_and_b32_e32 v0, 0xffff, v3
 ; 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
 ;
 ; VI-LABEL: v_test_sub_v2i16_zext_to_v2i64:
 ; VI:       ; %bb.0:
 ; VI-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
 ; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34

 ; VI-NEXT:    s_endpgm
 ;
 ; GFX10-LABEL: v_test_sub_v2i16_zext_to_v2i64:
 ; GFX10:       ; %bb.0:
 ; GFX10-NEXT:    s_clause 0x1
 ; 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)
 ; GFX10-NEXT:    global_load_dword v2, v0, s[2:3] glc dlc
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; 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_perm_b32 v2, 0, v0, 0x5040302
+; 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
 ;
 ; GFX11-LABEL: v_test_sub_v2i16_zext_to_v2i64:
 ; GFX11:       ; %bb.0:
 ; GFX11-NEXT:    s_clause 0x1
 ; GFX11-NEXT:    s_load_b128 s[4:7], s[0:1], 0x24

llvm/test/CodeGen/AMDGPU/vector_shuffle.packed.ll

 }
 
 define <4 x half> @shuffle_v4f16_3u6u(<4 x half> addrspace(1)* %arg0, <4 x half> addrspace(1)* %arg1) {
 ; GFX9-LABEL: shuffle_v4f16_3u6u:
 ; GFX9:       ; %bb.0:
 ; 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 offset:4
+; GFX9-NEXT:    v_mov_b32_e32 v0, 0x5040302
 ; GFX9-NEXT:    s_waitcnt vmcnt(1)
-; GFX9-NEXT:    v_lshrrev_b32_e32 v0, 16, v5
+; GFX9-NEXT:    v_perm_b32 v0, s4, v5, v0
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    v_mov_b32_e32 v1, v4
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX10-LABEL: shuffle_v4f16_3u6u:
 ; 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 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_perm_b32 v0, s4, v5, 0x5040302
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    v_mov_b32_e32 v1, v4
 ; GFX10-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX11-LABEL: shuffle_v4f16_3u6u:
 ; 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 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_perm_b32 v0, s0, v0, 0x5040302
 ; 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
   %val1 = load <4 x half>, <4 x half> addrspace(1)* %arg1
   %shuffle = shufflevector <4 x half> %val0, <4 x half> %val1, <4 x i32> <i32 3, i32 undef, i32 6, i32 undef>
   ret <4 x half> %shuffle
 }
 
 define <4 x half> @shuffle_v4f16_3uu7(<4 x half> addrspace(1)* %arg0, <4 x half> addrspace(1)* %arg1) {
 ; GFX9-LABEL: shuffle_v4f16_3uu7:
 ; GFX9:       ; %bb.0:
 ; 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 offset:4
+; GFX9-NEXT:    v_mov_b32_e32 v0, 0x5040302
 ; GFX9-NEXT:    s_waitcnt vmcnt(1)
-; GFX9-NEXT:    v_lshrrev_b32_e32 v0, 16, v5
+; GFX9-NEXT:    v_perm_b32 v0, s4, v5, v0
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    v_mov_b32_e32 v1, v4
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX10-LABEL: shuffle_v4f16_3uu7:
 ; 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 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_perm_b32 v0, s4, v5, 0x5040302
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    v_mov_b32_e32 v1, v4
 ; GFX10-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX11-LABEL: shuffle_v4f16_3uu7:
 ; 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 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_perm_b32 v0, s0, v0, 0x5040302
 ; 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
   %val1 = load <4 x half>, <4 x half> addrspace(1)* %arg1
   %shuffle = shufflevector <4 x half> %val0, <4 x half> %val1, <4 x i32> <i32 3, i32 undef, i32 undef, i32 7>
   ret <4 x half> %shuffle
 }
 
 define <4 x half> @shuffle_v4f16_35u5(<4 x half> addrspace(1)* %arg0, <4 x half> addrspace(1)* %arg1) {
 ; GFX9-LABEL: shuffle_v4f16_35u5:
 ; GFX9:       ; %bb.0:
 ; 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]
 ;
 ; GFX10-LABEL: shuffle_v4f16_35u5:
 ; 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 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]
 ;
 ; GFX11-LABEL: shuffle_v4f16_35u5:
 ; 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 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
   %shuffle = shufflevector <4 x half> %val0, <4 x half> %val1, <4 x i32> <i32 3, i32 5, i32 undef, i32 5>
   ret <4 x half> %shuffle
 }
 
 define <4 x half> @shuffle_v4f16_357u(<4 x half> addrspace(1)* %arg0, <4 x half> addrspace(1)* %arg1) {
 ; 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:    v_mov_b32_e32 v1, 0x5040302
 ; 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_perm_b32 v1, s4, v5, v1
 ; 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_perm_b32 v1, s4, v5, 0x5040302
 ; 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_perm_b32 v1, s0, v3, 0x5040302
 ; 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
   %shuffle = shufflevector <4 x half> %val0, <4 x half> %val1, <4 x i32> <i32 3, i32 5, i32 7, i32 undef>
   ret <4 x half> %shuffle
 }
 
 define <4 x half> @shuffle_v4f16_0101(<4 x half> addrspace(1)* %arg0, <4 x half> addrspace(1)* %arg1) {

 }
 
 define <4 x half> @shuffle_v4f16_2356(<4 x half> addrspace(1)* %arg0, <4 x half> addrspace(1)* %arg1) {
 ; GFX9-LABEL: shuffle_v4f16_2356:
 ; GFX9:       ; %bb.0:
 ; 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_mov_b32 s4, 0x5040302
 ; 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_perm_b32 v1, v6, v5, s4
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    v_mov_b32_e32 v0, v4
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX10-LABEL: shuffle_v4f16_2356:
 ; GFX10:       ; %bb.0:
 ; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; 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_perm_b32 v1, v6, v5, 0x5040302
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    v_mov_b32_e32 v0, v4
 ; GFX10-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX11-LABEL: shuffle_v4f16_2356:
 ; 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_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_perm_b32 v1, v3, v2, 0x5040302
 ; 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
   %val1 = load <4 x half>, <4 x half> addrspace(1)* %arg1
   %shuffle = shufflevector <4 x half> %val0, <4 x half> %val1, <4 x i32> <i32 2, i32 3, i32 5, i32 6>
   ret <4 x half> %shuffle
 }
 
 define <4 x half> @shuffle_v4f16_5623(<4 x half> addrspace(1)* %arg0, <4 x half> addrspace(1)* %arg1) {
 ; GFX9-LABEL: shuffle_v4f16_5623:
 ; GFX9:       ; %bb.0:
 ; 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_mov_b32 s4, 0x5040302
 ; 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_perm_b32 v0, v6, v5, s4
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    v_mov_b32_e32 v1, v4
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX10-LABEL: shuffle_v4f16_5623:
 ; GFX10:       ; %bb.0:
 ; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; 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_perm_b32 v0, v6, v5, 0x5040302
 ; 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:
 ; 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_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_perm_b32 v0, v3, v2, 0x5040302
 ; 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
   %val1 = load <4 x half>, <4 x half> addrspace(1)* %arg1
   %shuffle = shufflevector <4 x half> %val0, <4 x half> %val1, <4 x i32> <i32 5, i32 6, i32 2, i32 3>
   ret <4 x half> %shuffle
 }
 
 define <4 x half> @shuffle_v4f16_3456(<4 x half> addrspace(1)* %arg0, <4 x half> addrspace(1)* %arg1) {
 ; 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_mov_b32 s4, 0x5040302
 ; 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 v1, v5, v4, s4
 ; 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_perm_b32 v0, v4, v6, s4
 ; 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_perm_b32 v1, v5, v4, 0x5040302
 ; 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_perm_b32 v0, v4, v6, 0x5040302
 ; 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_perm_b32 v1, v3, v2, 0x5040302
 ; 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_perm_b32 v0, v2, v0, 0x5040302
 ; 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
   %shuffle = shufflevector <4 x half> %val0, <4 x half> %val1, <4 x i32> <i32 3, i32 4, i32 5, i32 6>
   ret <4 x half> %shuffle
 }
 
 define <4 x half> @shuffle_v4f16_5634(<4 x half> addrspace(1)* %arg0, <4 x half> addrspace(1)* %arg1) {
 ; 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_mov_b32 s4, 0x5040302
 ; 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_lshl_or_b32 v1, v4, 16, v1
-; GFX9-NEXT:    v_lshl_or_b32 v0, v5, 16, v0
+; GFX9-NEXT:    v_perm_b32 v1, v4, v6, s4
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX10-LABEL: shuffle_v4f16_5634:
 ; GFX10:       ; %bb.0:
 ; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; 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_perm_b32 v0, v5, v4, 0x5040302
 ; 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_perm_b32 v1, v4, v6, 0x5040302
 ; GFX10-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX11-LABEL: shuffle_v4f16_5634:
 ; 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_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, 0x5040302
 ; 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_perm_b32 v1, v2, v1, 0x5040302
 ; 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
   %shuffle = shufflevector <4 x half> %val0, <4 x half> %val1, <4 x i32> <i32 5, i32 6, i32 3, i32 4>
   ret <4 x half> %shuffle
 }
 
 define <4 x half> @shuffle_v4f16_5734(<4 x half> addrspace(1)* %arg0, <4 x half> addrspace(1)* %arg1) {
 ; 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_mov_b32 s5, 0x5040302
 ; 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_perm_b32 v1, v4, v6, s5
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX10-LABEL: shuffle_v4f16_5734:
 ; GFX10:       ; %bb.0:
 ; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; 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_perm_b32 v1, v4, v6, 0x5040302
 ; GFX10-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX11-LABEL: shuffle_v4f16_5734:
 ; 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_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_perm_b32 v1, v2, v1, 0x5040302
 ; 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
   %shuffle = shufflevector <4 x half> %val0, <4 x half> %val1, <4 x i32> <i32 5, i32 7, i32 3, i32 4>
   ret <4 x half> %shuffle
 }
 
 define <4 x i16> @shuffle_v4i16_2356(<4 x i16> addrspace(1)* %arg0, <4 x i16> addrspace(1)* %arg1) {
 ; GFX9-LABEL: shuffle_v4i16_2356:
 ; GFX9:       ; %bb.0:
 ; 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_mov_b32 s4, 0x5040302
 ; 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_perm_b32 v1, v6, v5, s4
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    v_mov_b32_e32 v0, v4
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX10-LABEL: shuffle_v4i16_2356:
 ; GFX10:       ; %bb.0:
 ; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; 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_perm_b32 v1, v6, v5, 0x5040302
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    v_mov_b32_e32 v0, v4
 ; GFX10-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX11-LABEL: shuffle_v4i16_2356:
 ; GFX11:       ; %bb.0:
 ; GFX11-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX11-NEXT:    s_waitcnt_vscnt null, 0x0
; GFX11-NEXT:    s_setpc_b64 s[30:31]
   ret <4 x i16> %shuffle
 }
 
 define <4 x half> @shuffle_v4f16_0000(<4 x half> addrspace(1)* %arg0, <4 x half> addrspace(1)* %arg1) {
 ; GFX9-LABEL: shuffle_v4f16_0000:
 ; 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]
 ;
 ; GFX10-LABEL: shuffle_v4f16_0000:
 ; GFX10:       ; %bb.0:
 ; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; 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]
 ;
 ; GFX11-LABEL: shuffle_v4f16_0000:
 ; 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[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
   %val1 = load <4 x half>, <4 x half> addrspace(1)* %arg1
   %shuffle = shufflevector <4 x half> %val0, <4 x half> %val1, <4 x i32> zeroinitializer
   ret <4 x half> %shuffle
 }
 
 define <4 x half> @shuffle_v4f16_1010(<4 x half> addrspace(1)* %arg0, <4 x half> addrspace(1)* %arg1) {
 ; GFX9-LABEL: shuffle_v4f16_1010:
 ; 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, 0x5040302
 ; 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_perm_b32 v0, v0, v0, s4
 ; GFX9-NEXT:    v_mov_b32_e32 v1, v0
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX10-LABEL: shuffle_v4f16_1010:
 ; GFX10:       ; %bb.0:
 ; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; 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_perm_b32 v0, v0, v0, 0x5040302
 ; GFX10-NEXT:    v_mov_b32_e32 v1, v0
 ; GFX10-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX11-LABEL: shuffle_v4f16_1010:
 ; 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[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_perm_b32 v0, v0, v0, 0x5040302
 ; 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
   %val1 = load <4 x half>, <4 x half> addrspace(1)* %arg1
   %shuffle = shufflevector <4 x half> %val0, <4 x half> %val1, <4 x i32> <i32 1, i32 0, i32 1, i32 0>
   ret <4 x half> %shuffle
 }
 
 define <4 x half> @shuffle_v4f16_1100(<4 x half> addrspace(1)* %arg0, <4 x half> addrspace(1)* %arg1) {
 ; 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:
 ; GFX10:       ; %bb.0:
 ; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; 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:
 ; 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[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
   %shuffle = shufflevector <4 x half> %val0, <4 x half> %val1, <4 x i32> <i32 1, i32 1, i32 0, i32 0>
   ret <4 x half> %shuffle
 }
 
 define <4 x half> @shuffle_v4f16_6161(<4 x half> addrspace(1)* %arg0, <4 x half> addrspace(1)* %arg1) {
 ; 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]
 ;
 ; GFX10-LABEL: shuffle_v4f16_6161:
 ; 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]
 ;
 ; GFX11-LABEL: shuffle_v4f16_6161:
 ; 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
   %val1 = load <4 x half>, <4 x half> addrspace(1)* %arg1
   %shuffle = shufflevector <4 x half> %val0, <4 x half> %val1, <4 x i32> <i32 6, i32 1, i32 6, i32 1>
   ret <4 x half> %shuffle
 }
 
 define <4 x half> @shuffle_v4f16_2333(<4 x half> addrspace(1)* %arg0, <4 x half> addrspace(1)* %arg1) {
 ; GFX9-LABEL: shuffle_v4f16_2333:
 ; 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:
 ; 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 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:
 ; 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 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
   %shuffle = shufflevector <4 x half> %val0, <4 x half> %val1, <4 x i32> <i32 2, i32 3, i32 3, i32 3>
   ret <4 x half> %shuffle
 }
 
 define <4 x half> @shuffle_v4f16_6667(<4 x half> addrspace(1)* %arg0, <4 x half> addrspace(1)* %arg1) {
 ; GFX9-LABEL: shuffle_v4f16_6667:
 ; 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:
 ; 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 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:
 ; 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 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
   %shuffle = shufflevector <4 x half> %val0, <4 x half> %val1, <4 x i32> <i32 2, i32 3, i32 3, i32 3>
   ret <4 x half> %shuffle
 }
 
 define <4 x half> @shuffle_v8f16_0101(<8 x half> addrspace(1)* %arg0, <8 x half> addrspace(1)* %arg1) {

 }
 
 define <4 x half> @shuffle_v8f16_13_14_2_3(<8 x half> addrspace(1)* %arg0, <8 x half> addrspace(1)* %arg1) {
 ; GFX9-LABEL: shuffle_v8f16_13_14_2_3:
 ; GFX9:       ; %bb.0:
 ; 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_mov_b32 s4, 0x5040302
 ; 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_perm_b32 v0, v6, v5, s4
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    v_mov_b32_e32 v1, v4
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX10-LABEL: shuffle_v8f16_13_14_2_3:
 ; GFX10:       ; %bb.0:
 ; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; 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_perm_b32 v0, v6, v5, 0x5040302
 ; 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:
 ; 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 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_perm_b32 v0, v3, v2, 0x5040302
 ; 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
   %val1 = load <8 x half>, <8 x half> addrspace(1)* %arg1
   %shuffle = shufflevector <8 x half> %val0, <8 x half> %val1, <4 x i32> <i32 13, i32 14, i32 2, i32 3>
   ret <4 x half> %shuffle
 }
 
 define <4 x half> @shuffle_v3f16_0122(<3 x half> addrspace(1)* %arg0, <3 x half> addrspace(1)* %arg1) {
 ; GFX9-LABEL: shuffle_v3f16_0122:
 ; 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:
 ; GFX10:       ; %bb.0:
 ; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; 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:
 ; 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[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
   %shuffle = shufflevector <3 x half> %val0, <3 x half> %val1, <4 x i32> <i32 0, i32 1, i32 2, i32 2>
   ret <4 x half> %shuffle
 }
 
 define <4 x half> @shuffle_v2f16_0122(<2 x half> addrspace(1)* %arg0, <2 x half> addrspace(1)* %arg1) {
 ; GFX9-LABEL: shuffle_v2f16_0122:
 ; 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_mov_b32 s4, 0x5040302
 ; 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_perm_b32 v1, v0, v0, s4
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX10-LABEL: shuffle_v2f16_0122:
 ; 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 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_perm_b32 v1, v0, v0, 0x5040302
 ; GFX10-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX11-LABEL: shuffle_v2f16_0122:
 ; 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 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_perm_b32 v1, v0, v0, 0x5040302
 ; 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
   %shuffle = shufflevector <2 x half> %val0, <2 x half> %val1, <4 x i32> <i32 0, i32 1, i32 1, i32 0>
   ret <4 x half> %shuffle
 }
 
 define <6 x half> @shuffle_v6f16_452367(<6 x half> addrspace(1)* %arg0, <6 x half> addrspace(1)* %arg1) {

 
 define <4 x half> @shuffle_v4f16_0456(<4 x half> addrspace(1)* %arg0, <4 x half> addrspace(1)* %arg1) {
 ; GFX9-LABEL: shuffle_v4f16_0456:
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; 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:    s_mov_b32 s5, 0x5040302
 ; 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_perm_b32 v1, v6, v5, s5
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX10-LABEL: shuffle_v4f16_0456:
 ; GFX10:       ; %bb.0:
 ; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-NEXT:    global_load_dwordx2 v[4:5], v[0:1], off
 ; 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_perm_b32 v1, v6, v5, 0x5040302
 ; 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_perm_b32 v1, v2, v1, 0x5040302
 ; 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
   %shuffle = shufflevector <4 x half> %val0, <4 x half> %val1, <4 x i32> <i32 0, i32 4, i32 5, i32 6>
   ret <4 x half> %shuffle
 }
 
 define amdgpu_kernel void @shuffle_scalar_load_v8i32_0123(<8 x i32> addrspace(4)* %in, <4 x i32> addrspace(1)* %out)  {
; GFX11-NEXT:    s_endpgm
   store <4 x i32> %id, <4 x i32> addrspace(1)* %out, align 8
   ret void
 }
 
 declare <2 x half> @llvm.fma.v2f16(<2 x half>, <2 x half>, <2 x half>) #0
 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_mov_b32 s4, 0x5040302
+; 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: 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_perm_b32 v0, v5, v4, 0x5040302
+; 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_perm_b32 v0, v1, v0, 0x5040302
+; 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_mov_b32 s4, 0x5040302
+; 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_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_perm_b32 v0, v5, v4, 0x5040302
+; 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(1)
+; GFX11-NEXT:    v_lshrrev_b32_e32 v0, 16, v0
+; GFX11-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
+; GFX11-NEXT:    v_lshl_or_b32 v0, v1, 16, v0
+; 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
+}