[DAG] Enable ISD::SHL SimplifyMultipleUseDemandedBits handling inside…

… SimplifyDemandedBits

Pulled out of D77804 as its going to be easier to address the regressions individually.

This patch allows SimplifyDemandedBits to call SimplifyMultipleUseDemandedBits in cases where the source operand has other uses, enabling us to peek through the shifted value if we don't demand all the bits/elts.

The lost RISCV gorc2 fold shouldn't be a problem - instcombine would have already destroyed that pattern - see #50553

Differential Revision: https://reviews.llvm.org/D124839

llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp

@@ -1671,6 +1671,16 @@ bool TargetLowering::SimplifyDemandedBits(
       // low bits known zero.
       Known.Zero.setLowBits(ShAmt);
 
+      // Attempt to avoid multi-use ops if we don't need anything from them.
+      if (!InDemandedMask.isAllOnesValue() || !DemandedElts.isAllOnesValue()) {
+        SDValue DemandedOp0 = SimplifyMultipleUseDemandedBits(
+            Op0, InDemandedMask, DemandedElts, TLO.DAG, Depth + 1);
+        if (DemandedOp0) {
+          SDValue NewOp = TLO.DAG.getNode(ISD::SHL, dl, VT, DemandedOp0, Op1);
+          return TLO.CombineTo(Op, NewOp);
+        }
+      }
+
       // Try shrinking the operation as long as the shift amount will still be
       // in range.
       if ((ShAmt < DemandedBits.getActiveBits()) &&

llvm/test/CodeGen/AMDGPU/amdgpu-codegenprepare-idiv.ll

@@ -5240,77 +5240,73 @@ define amdgpu_kernel void @srem_v3i15(<3 x i15> addrspace(1)* %out, <3 x i15> %x
 ; GFX6-NEXT:    s_mov_b32 s7, 0xf000
 ; GFX6-NEXT:    s_mov_b32 s6, -1
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX6-NEXT:    s_bfe_i32 s10, s2, 0xf0000
+; GFX6-NEXT:    v_cvt_f32_i32_e32 v5, s10
+; GFX6-NEXT:    v_mov_b32_e32 v2, s0
+; GFX6-NEXT:    v_alignbit_b32 v2, s1, v2, 30
+; GFX6-NEXT:    s_bfe_i32 s1, s0, 0xf0000
+; GFX6-NEXT:    v_cvt_f32_i32_e32 v4, s1
+; GFX6-NEXT:    s_xor_b32 s1, s10, s1
+; GFX6-NEXT:    s_ashr_i32 s1, s1, 30
+; GFX6-NEXT:    s_or_b32 s1, s1, 1
+; GFX6-NEXT:    v_rcp_iflag_f32_e32 v6, v4
+; GFX6-NEXT:    v_mov_b32_e32 v7, s1
+; GFX6-NEXT:    s_lshr_b32 s9, s0, 15
+; GFX6-NEXT:    s_bfe_i32 s1, s2, 0xf000f
+; GFX6-NEXT:    v_mul_f32_e32 v6, v5, v6
+; GFX6-NEXT:    v_trunc_f32_e32 v6, v6
+; GFX6-NEXT:    v_mad_f32 v5, -v6, v4, v5
+; GFX6-NEXT:    v_cvt_i32_f32_e32 v6, v6
+; GFX6-NEXT:    v_cmp_ge_f32_e64 vcc, |v5|, |v4|
+; GFX6-NEXT:    v_cndmask_b32_e32 v4, 0, v7, vcc
 ; GFX6-NEXT:    v_mov_b32_e32 v0, s2
+; GFX6-NEXT:    v_add_i32_e32 v4, vcc, v4, v6
+; GFX6-NEXT:    v_mul_lo_u32 v4, v4, s0
+; GFX6-NEXT:    s_bfe_i32 s0, s0, 0xf000f
+; GFX6-NEXT:    v_cvt_f32_i32_e32 v5, s0
+; GFX6-NEXT:    v_cvt_f32_i32_e32 v6, s1
 ; GFX6-NEXT:    v_alignbit_b32 v0, s3, v0, 30
 ; GFX6-NEXT:    s_movk_i32 s3, 0x7fff
-; GFX6-NEXT:    s_and_b32 s11, s0, s3
-; GFX6-NEXT:    s_bfe_i32 s11, s11, 0xf0000
-; GFX6-NEXT:    v_cvt_f32_i32_e32 v2, s11
-; GFX6-NEXT:    s_and_b32 s9, s2, s3
-; GFX6-NEXT:    s_bfe_i32 s9, s9, 0xf0000
-; GFX6-NEXT:    v_cvt_f32_i32_e32 v3, s9
-; GFX6-NEXT:    v_rcp_iflag_f32_e32 v4, v2
-; GFX6-NEXT:    s_xor_b32 s9, s9, s11
-; GFX6-NEXT:    s_ashr_i32 s9, s9, 30
-; GFX6-NEXT:    s_or_b32 s9, s9, 1
-; GFX6-NEXT:    v_mul_f32_e32 v4, v3, v4
-; GFX6-NEXT:    v_trunc_f32_e32 v4, v4
-; GFX6-NEXT:    v_mad_f32 v3, -v4, v2, v3
-; GFX6-NEXT:    v_cvt_i32_f32_e32 v4, v4
-; GFX6-NEXT:    v_mov_b32_e32 v5, s9
-; GFX6-NEXT:    v_cmp_ge_f32_e64 vcc, |v3|, |v2|
-; GFX6-NEXT:    v_cndmask_b32_e32 v2, 0, v5, vcc
-; GFX6-NEXT:    v_mov_b32_e32 v1, s0
-; GFX6-NEXT:    s_bfe_u32 s12, s0, 0xf000f
-; GFX6-NEXT:    v_add_i32_e32 v2, vcc, v2, v4
-; GFX6-NEXT:    v_alignbit_b32 v1, s1, v1, 30
-; GFX6-NEXT:    s_lshr_b32 s1, s0, 15
-; GFX6-NEXT:    v_mul_lo_u32 v2, v2, s0
-; GFX6-NEXT:    s_bfe_i32 s0, s12, 0xf0000
-; GFX6-NEXT:    v_cvt_f32_i32_e32 v3, s0
-; GFX6-NEXT:    s_bfe_u32 s10, s2, 0xf000f
-; GFX6-NEXT:    s_lshr_b32 s8, s2, 15
-; GFX6-NEXT:    v_sub_i32_e32 v2, vcc, s2, v2
-; GFX6-NEXT:    s_bfe_i32 s2, s10, 0xf0000
-; GFX6-NEXT:    v_cvt_f32_i32_e32 v4, s2
-; GFX6-NEXT:    v_rcp_iflag_f32_e32 v5, v3
-; GFX6-NEXT:    s_xor_b32 s0, s2, s0
+; GFX6-NEXT:    v_rcp_iflag_f32_e32 v7, v5
+; GFX6-NEXT:    v_and_b32_e32 v3, s3, v2
+; GFX6-NEXT:    v_sub_i32_e32 v4, vcc, s2, v4
+; GFX6-NEXT:    v_mul_f32_e32 v7, v6, v7
+; GFX6-NEXT:    v_trunc_f32_e32 v7, v7
+; GFX6-NEXT:    s_xor_b32 s0, s1, s0
+; GFX6-NEXT:    v_mad_f32 v6, -v7, v5, v6
+; GFX6-NEXT:    v_bfe_i32 v2, v2, 0, 15
 ; GFX6-NEXT:    s_ashr_i32 s0, s0, 30
+; GFX6-NEXT:    v_cvt_i32_f32_e32 v7, v7
+; GFX6-NEXT:    v_cmp_ge_f32_e64 vcc, |v6|, |v5|
+; GFX6-NEXT:    v_cvt_f32_i32_e32 v6, v2
 ; GFX6-NEXT:    s_or_b32 s0, s0, 1
-; GFX6-NEXT:    v_mul_f32_e32 v5, v4, v5
-; GFX6-NEXT:    v_trunc_f32_e32 v5, v5
-; GFX6-NEXT:    v_mad_f32 v4, -v5, v3, v4
-; GFX6-NEXT:    v_cvt_i32_f32_e32 v5, v5
-; GFX6-NEXT:    v_and_b32_e32 v1, s3, v1
-; GFX6-NEXT:    v_mov_b32_e32 v6, s0
-; GFX6-NEXT:    v_cmp_ge_f32_e64 vcc, |v4|, |v3|
-; GFX6-NEXT:    v_cndmask_b32_e32 v3, 0, v6, vcc
-; GFX6-NEXT:    v_bfe_i32 v4, v1, 0, 15
-; GFX6-NEXT:    v_add_i32_e32 v3, vcc, v3, v5
-; GFX6-NEXT:    v_cvt_f32_i32_e32 v5, v4
-; GFX6-NEXT:    v_and_b32_e32 v0, s3, v0
-; GFX6-NEXT:    v_bfe_i32 v6, v0, 0, 15
-; GFX6-NEXT:    v_cvt_f32_i32_e32 v7, v6
-; GFX6-NEXT:    v_rcp_iflag_f32_e32 v8, v5
-; GFX6-NEXT:    v_xor_b32_e32 v4, v6, v4
-; GFX6-NEXT:    v_ashrrev_i32_e32 v4, 30, v4
-; GFX6-NEXT:    v_or_b32_e32 v4, 1, v4
-; GFX6-NEXT:    v_mul_f32_e32 v6, v7, v8
-; GFX6-NEXT:    v_trunc_f32_e32 v6, v6
-; GFX6-NEXT:    v_mad_f32 v7, -v6, v5, v7
-; GFX6-NEXT:    v_cvt_i32_f32_e32 v6, v6
-; GFX6-NEXT:    v_cmp_ge_f32_e64 vcc, |v7|, |v5|
-; GFX6-NEXT:    v_cndmask_b32_e32 v4, 0, v4, vcc
-; GFX6-NEXT:    v_mul_lo_u32 v3, v3, s1
-; GFX6-NEXT:    v_add_i32_e32 v4, vcc, v4, v6
-; GFX6-NEXT:    v_mul_lo_u32 v1, v4, v1
-; GFX6-NEXT:    v_sub_i32_e32 v3, vcc, s8, v3
-; GFX6-NEXT:    v_and_b32_e32 v3, s3, v3
-; GFX6-NEXT:    v_subrev_i32_e32 v0, vcc, v1, v0
-; GFX6-NEXT:    v_lshl_b64 v[0:1], v[0:1], 30
+; GFX6-NEXT:    v_mov_b32_e32 v8, s0
+; GFX6-NEXT:    v_and_b32_e32 v1, s3, v0
+; GFX6-NEXT:    v_cndmask_b32_e32 v5, 0, v8, vcc
+; GFX6-NEXT:    v_bfe_i32 v0, v0, 0, 15
+; GFX6-NEXT:    v_add_i32_e32 v5, vcc, v5, v7
+; GFX6-NEXT:    v_cvt_f32_i32_e32 v7, v0
+; GFX6-NEXT:    v_rcp_iflag_f32_e32 v8, v6
+; GFX6-NEXT:    v_xor_b32_e32 v0, v0, v2
+; GFX6-NEXT:    v_ashrrev_i32_e32 v0, 30, v0
+; GFX6-NEXT:    v_or_b32_e32 v0, 1, v0
+; GFX6-NEXT:    v_mul_f32_e32 v2, v7, v8
+; GFX6-NEXT:    v_trunc_f32_e32 v2, v2
+; GFX6-NEXT:    v_mad_f32 v7, -v2, v6, v7
+; GFX6-NEXT:    v_cvt_i32_f32_e32 v2, v2
+; GFX6-NEXT:    v_cmp_ge_f32_e64 vcc, |v7|, |v6|
+; GFX6-NEXT:    v_cndmask_b32_e32 v0, 0, v0, vcc
+; GFX6-NEXT:    v_mul_lo_u32 v5, v5, s9
+; GFX6-NEXT:    v_add_i32_e32 v0, vcc, v0, v2
+; GFX6-NEXT:    v_mul_lo_u32 v0, v0, v3
+; GFX6-NEXT:    s_lshr_b32 s8, s2, 15
+; GFX6-NEXT:    v_sub_i32_e32 v2, vcc, s8, v5
+; GFX6-NEXT:    v_subrev_i32_e32 v0, vcc, v0, v1
 ; GFX6-NEXT:    v_and_b32_e32 v2, s3, v2
-; GFX6-NEXT:    v_lshlrev_b32_e32 v3, 15, v3
-; GFX6-NEXT:    v_or_b32_e32 v2, v2, v3
+; GFX6-NEXT:    v_lshl_b64 v[0:1], v[0:1], 30
+; GFX6-NEXT:    v_and_b32_e32 v3, s3, v4
+; GFX6-NEXT:    v_lshlrev_b32_e32 v2, 15, v2
+; GFX6-NEXT:    v_or_b32_e32 v2, v3, v2
 ; GFX6-NEXT:    v_or_b32_e32 v0, v2, v0
 ; GFX6-NEXT:    buffer_store_dword v0, off, s[4:7], 0
 ; GFX6-NEXT:    s_waitcnt expcnt(0)
@@ -5324,82 +5320,78 @@ define amdgpu_kernel void @srem_v3i15(<3 x i15> addrspace(1)* %out, <3 x i15> %x
 ; GFX9-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x24
 ; GFX9-NEXT:    s_movk_i32 s8, 0x7fff
 ; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
+; GFX9-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    s_bfe_i32 s6, s2, 0xf0000
+; GFX9-NEXT:    v_cvt_f32_i32_e32 v5, s6
 ; GFX9-NEXT:    v_mov_b32_e32 v0, s2
-; GFX9-NEXT:    v_alignbit_b32 v0, s3, v0, 30
-; GFX9-NEXT:    s_and_b32 s3, s2, s8
 ; GFX9-NEXT:    v_mov_b32_e32 v1, s0
 ; GFX9-NEXT:    v_alignbit_b32 v1, s1, v1, 30
-; GFX9-NEXT:    s_and_b32 s1, s0, s8
-; GFX9-NEXT:    s_bfe_i32 s1, s1, 0xf0000
-; GFX9-NEXT:    v_cvt_f32_i32_e32 v2, s1
-; GFX9-NEXT:    s_bfe_i32 s3, s3, 0xf0000
-; GFX9-NEXT:    v_cvt_f32_i32_e32 v3, s3
-; GFX9-NEXT:    s_xor_b32 s1, s3, s1
-; GFX9-NEXT:    v_rcp_iflag_f32_e32 v4, v2
+; GFX9-NEXT:    s_bfe_i32 s1, s0, 0xf0000
+; GFX9-NEXT:    v_cvt_f32_i32_e32 v4, s1
+; GFX9-NEXT:    s_xor_b32 s1, s6, s1
 ; GFX9-NEXT:    s_ashr_i32 s1, s1, 30
-; GFX9-NEXT:    s_lshr_b32 s9, s2, 15
-; GFX9-NEXT:    s_bfe_u32 s10, s2, 0xf000f
-; GFX9-NEXT:    v_mul_f32_e32 v4, v3, v4
-; GFX9-NEXT:    v_trunc_f32_e32 v4, v4
-; GFX9-NEXT:    v_mad_f32 v3, -v4, v2, v3
-; GFX9-NEXT:    v_cvt_i32_f32_e32 v4, v4
-; GFX9-NEXT:    s_lshr_b32 s11, s0, 15
-; GFX9-NEXT:    s_bfe_u32 s12, s0, 0xf000f
+; GFX9-NEXT:    v_alignbit_b32 v0, s3, v0, 30
+; GFX9-NEXT:    v_rcp_iflag_f32_e32 v6, v4
+; GFX9-NEXT:    s_lshr_b32 s3, s2, 15
+; GFX9-NEXT:    s_lshr_b32 s9, s0, 15
 ; GFX9-NEXT:    s_or_b32 s1, s1, 1
-; GFX9-NEXT:    v_cmp_ge_f32_e64 s[6:7], |v3|, |v2|
+; GFX9-NEXT:    v_mul_f32_e32 v6, v5, v6
+; GFX9-NEXT:    v_trunc_f32_e32 v6, v6
+; GFX9-NEXT:    v_mad_f32 v5, -v6, v4, v5
+; GFX9-NEXT:    v_cvt_i32_f32_e32 v6, v6
+; GFX9-NEXT:    v_cmp_ge_f32_e64 s[6:7], |v5|, |v4|
 ; GFX9-NEXT:    s_and_b64 s[6:7], s[6:7], exec
 ; GFX9-NEXT:    s_cselect_b32 s1, s1, 0
-; GFX9-NEXT:    v_add_u32_e32 v2, s1, v4
-; GFX9-NEXT:    s_bfe_i32 s1, s12, 0xf0000
-; GFX9-NEXT:    v_cvt_f32_i32_e32 v3, s1
-; GFX9-NEXT:    v_mul_lo_u32 v2, v2, s0
-; GFX9-NEXT:    s_bfe_i32 s0, s10, 0xf0000
-; GFX9-NEXT:    v_cvt_f32_i32_e32 v4, s0
-; GFX9-NEXT:    v_rcp_iflag_f32_e32 v5, v3
+; GFX9-NEXT:    v_add_u32_e32 v4, s1, v6
+; GFX9-NEXT:    s_bfe_i32 s1, s0, 0xf000f
+; GFX9-NEXT:    v_cvt_f32_i32_e32 v5, s1
+; GFX9-NEXT:    v_mul_lo_u32 v4, v4, s0
+; GFX9-NEXT:    s_bfe_i32 s0, s2, 0xf000f
+; GFX9-NEXT:    v_cvt_f32_i32_e32 v6, s0
+; GFX9-NEXT:    v_rcp_iflag_f32_e32 v7, v5
 ; GFX9-NEXT:    s_xor_b32 s0, s0, s1
+; GFX9-NEXT:    v_and_b32_e32 v3, s8, v1
 ; GFX9-NEXT:    s_ashr_i32 s0, s0, 30
-; GFX9-NEXT:    s_or_b32 s3, s0, 1
-; GFX9-NEXT:    v_mul_f32_e32 v5, v4, v5
-; GFX9-NEXT:    v_trunc_f32_e32 v5, v5
-; GFX9-NEXT:    v_mad_f32 v4, -v5, v3, v4
-; GFX9-NEXT:    v_cvt_i32_f32_e32 v5, v5
-; GFX9-NEXT:    v_cmp_ge_f32_e64 s[0:1], |v4|, |v3|
-; GFX9-NEXT:    v_and_b32_e32 v1, s8, v1
+; GFX9-NEXT:    v_mul_f32_e32 v7, v6, v7
+; GFX9-NEXT:    v_trunc_f32_e32 v7, v7
+; GFX9-NEXT:    v_mad_f32 v6, -v7, v5, v6
+; GFX9-NEXT:    v_cvt_i32_f32_e32 v7, v7
+; GFX9-NEXT:    v_bfe_i32 v1, v1, 0, 15
+; GFX9-NEXT:    s_or_b32 s6, s0, 1
+; GFX9-NEXT:    v_cmp_ge_f32_e64 s[0:1], |v6|, |v5|
+; GFX9-NEXT:    v_cvt_f32_i32_e32 v6, v1
 ; GFX9-NEXT:    s_and_b64 s[0:1], s[0:1], exec
-; GFX9-NEXT:    s_cselect_b32 s0, s3, 0
-; GFX9-NEXT:    v_bfe_i32 v4, v1, 0, 15
-; GFX9-NEXT:    v_add_u32_e32 v3, s0, v5
-; GFX9-NEXT:    v_cvt_f32_i32_e32 v5, v4
+; GFX9-NEXT:    s_cselect_b32 s0, s6, 0
+; GFX9-NEXT:    v_add_u32_e32 v5, s0, v7
+; GFX9-NEXT:    v_bfe_i32 v7, v0, 0, 15
+; GFX9-NEXT:    v_cvt_f32_i32_e32 v8, v7
+; GFX9-NEXT:    v_rcp_iflag_f32_e32 v9, v6
+; GFX9-NEXT:    v_xor_b32_e32 v1, v7, v1
+; GFX9-NEXT:    v_ashrrev_i32_e32 v1, 30, v1
+; GFX9-NEXT:    v_or_b32_e32 v1, 1, v1
+; GFX9-NEXT:    v_mul_f32_e32 v7, v8, v9
+; GFX9-NEXT:    v_trunc_f32_e32 v7, v7
+; GFX9-NEXT:    v_cvt_i32_f32_e32 v9, v7
+; GFX9-NEXT:    v_mad_f32 v7, -v7, v6, v8
+; GFX9-NEXT:    v_cmp_ge_f32_e64 vcc, |v7|, |v6|
+; GFX9-NEXT:    v_cndmask_b32_e32 v1, 0, v1, vcc
+; GFX9-NEXT:    v_mul_lo_u32 v5, v5, s9
+; GFX9-NEXT:    v_add_u32_e32 v1, v9, v1
+; GFX9-NEXT:    v_mul_lo_u32 v1, v1, v3
 ; GFX9-NEXT:    v_and_b32_e32 v0, s8, v0
-; GFX9-NEXT:    v_bfe_i32 v6, v0, 0, 15
-; GFX9-NEXT:    v_cvt_f32_i32_e32 v7, v6
-; GFX9-NEXT:    v_rcp_iflag_f32_e32 v8, v5
-; GFX9-NEXT:    v_xor_b32_e32 v4, v6, v4
-; GFX9-NEXT:    v_ashrrev_i32_e32 v4, 30, v4
-; GFX9-NEXT:    v_or_b32_e32 v4, 1, v4
-; GFX9-NEXT:    v_mul_f32_e32 v6, v7, v8
-; GFX9-NEXT:    v_trunc_f32_e32 v6, v6
-; GFX9-NEXT:    v_cvt_i32_f32_e32 v8, v6
-; GFX9-NEXT:    v_mad_f32 v6, -v6, v5, v7
-; GFX9-NEXT:    v_cmp_ge_f32_e64 vcc, |v6|, |v5|
-; GFX9-NEXT:    v_cndmask_b32_e32 v4, 0, v4, vcc
-; GFX9-NEXT:    v_mul_lo_u32 v3, v3, s11
-; GFX9-NEXT:    v_add_u32_e32 v4, v8, v4
-; GFX9-NEXT:    v_mul_lo_u32 v1, v4, v1
-; GFX9-NEXT:    v_sub_u32_e32 v2, s2, v2
-; GFX9-NEXT:    v_sub_u32_e32 v3, s9, v3
-; GFX9-NEXT:    v_and_b32_e32 v3, s8, v3
+; GFX9-NEXT:    v_sub_u32_e32 v3, s2, v4
+; GFX9-NEXT:    v_sub_u32_e32 v4, s3, v5
 ; GFX9-NEXT:    v_sub_u32_e32 v0, v0, v1
+; GFX9-NEXT:    v_and_b32_e32 v4, s8, v4
 ; GFX9-NEXT:    v_lshlrev_b64 v[0:1], 30, v[0:1]
-; GFX9-NEXT:    v_and_b32_e32 v2, s8, v2
-; GFX9-NEXT:    v_lshlrev_b32_e32 v3, 15, v3
-; GFX9-NEXT:    v_or_b32_e32 v2, v2, v3
-; GFX9-NEXT:    v_mov_b32_e32 v4, 0
-; GFX9-NEXT:    v_or_b32_e32 v0, v2, v0
-; GFX9-NEXT:    global_store_dword v4, v0, s[4:5]
+; GFX9-NEXT:    v_and_b32_e32 v3, s8, v3
+; GFX9-NEXT:    v_lshlrev_b32_e32 v4, 15, v4
+; GFX9-NEXT:    v_or_b32_e32 v3, v3, v4
+; GFX9-NEXT:    v_or_b32_e32 v0, v3, v0
+; GFX9-NEXT:    global_store_dword v2, v0, s[4:5]
 ; GFX9-NEXT:    v_and_b32_e32 v0, 0x1fff, v1
-; GFX9-NEXT:    global_store_short v4, v0, s[4:5] offset:4
+; GFX9-NEXT:    global_store_short v2, v0, s[4:5] offset:4
 ; GFX9-NEXT:    s_endpgm
   %r = srem <3 x i15> %x, %y
   store <3 x i15> %r, <3 x i15> addrspace(1)* %out

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

@@ -1605,10 +1605,10 @@ define amdgpu_kernel void @v_insertelement_v4i16_dynamic_vgpr(<4 x i16> addrspac
 ; VI-NEXT:    flat_load_dwordx2 v[0:1], v[0:1]
 ; VI-NEXT:    s_mov_b64 s[2:3], 0xffff
 ; VI-NEXT:    v_mov_b32_e32 v3, s1
-; VI-NEXT:    s_and_b32 s1, s4, s2
+; VI-NEXT:    s_lshl_b32 s1, s4, 16
+; VI-NEXT:    s_and_b32 s4, s4, s2
 ; VI-NEXT:    v_add_u32_e32 v2, vcc, s0, v2
-; VI-NEXT:    s_lshl_b32 s0, s1, 16
-; VI-NEXT:    s_or_b32 s0, s1, s0
+; VI-NEXT:    s_or_b32 s0, s4, s1
 ; VI-NEXT:    v_addc_u32_e32 v3, vcc, 0, v3, vcc
 ; VI-NEXT:    v_lshlrev_b32_e32 v4, 4, v4
 ; VI-NEXT:    v_lshlrev_b64 v[4:5], v4, s[2:3]
@@ -1690,11 +1690,11 @@ define amdgpu_kernel void @v_insertelement_v4f16_dynamic_sgpr(<4 x half> addrspa
 ; VI-NEXT:    s_mov_b64 s[2:3], 0xffff
 ; VI-NEXT:    v_mov_b32_e32 v3, s1
 ; VI-NEXT:    s_lshl_b32 s1, s5, 4
+; VI-NEXT:    s_lshl_b32 s5, s4, 16
 ; VI-NEXT:    s_and_b32 s4, s4, s2
 ; VI-NEXT:    v_add_u32_e32 v2, vcc, s0, v2
 ; VI-NEXT:    s_lshl_b64 s[0:1], s[2:3], s1
-; VI-NEXT:    s_lshl_b32 s2, s4, 16
-; VI-NEXT:    s_or_b32 s2, s4, s2
+; VI-NEXT:    s_or_b32 s2, s4, s5
 ; VI-NEXT:    v_mov_b32_e32 v4, s2
 ; VI-NEXT:    v_mov_b32_e32 v5, s2
 ; VI-NEXT:    v_addc_u32_e32 v3, vcc, 0, v3, vcc