AMDGPU: Sign extend immediates for 32-bit subregister extracts (#154870)

extractSubregFromImm previously would sign extend the 16-bit subregister
extracts, but not the 32-bit. We try to consistently store immediates
as sign extended, since not doing it can result in misreported
isInlineImmediate checks.

Author: arsenm

llvm/lib/Target/AMDGPU/SIInstrInfo.cpp

@@ -3475,9 +3475,9 @@ std::optional<int64_t> SIInstrInfo::extractSubregFromImm(int64_t Imm,
   case AMDGPU::NoSubRegister:
     return Imm;
   case AMDGPU::sub0:
-    return Lo_32(Imm);
+    return SignExtend64<32>(Imm);
   case AMDGPU::sub1:
-    return Hi_32(Imm);
+    return SignExtend64<32>(Imm >> 32);
   case AMDGPU::lo16:
     return SignExtend64<16>(Imm);
   case AMDGPU::hi16:

llvm/test/CodeGen/AMDGPU/flat-atomicrmw-fadd.ll

@@ -503,9 +503,10 @@ define float @flat_agent_atomic_fadd_ret_f32__offset12b_neg__amdgpu_no_fine_grai
 ; GFX908:       ; %bb.0:
 ; GFX908-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX908-NEXT:    v_add_co_u32_e32 v3, vcc, 0xfffff800, v0
-; GFX908-NEXT:    v_addc_co_u32_e32 v4, vcc, -1, v1, vcc
-; GFX908-NEXT:    v_add_co_u32_e32 v0, vcc, 0xfffff800, v0
+; GFX908-NEXT:    s_mov_b64 s[4:5], vcc
+; GFX908-NEXT:    v_addc_co_u32_e64 v4, s[4:5], -1, v1, s[4:5]
 ; GFX908-NEXT:    v_addc_co_u32_e32 v1, vcc, -1, v1, vcc
+; GFX908-NEXT:    v_mov_b32_e32 v0, v3
 ; GFX908-NEXT:    flat_load_dword v0, v[0:1]
 ; GFX908-NEXT:    s_mov_b64 s[4:5], 0
 ; GFX908-NEXT:  .LBB2_1: ; %atomicrmw.start
@@ -2827,9 +2828,10 @@ define float @flat_agent_atomic_fadd_ret_f32__offset12b_neg__ftz__amdgpu_no_fine
 ; GFX908:       ; %bb.0:
 ; GFX908-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX908-NEXT:    v_add_co_u32_e32 v3, vcc, 0xfffff800, v0
-; GFX908-NEXT:    v_addc_co_u32_e32 v4, vcc, -1, v1, vcc
-; GFX908-NEXT:    v_add_co_u32_e32 v0, vcc, 0xfffff800, v0
+; GFX908-NEXT:    s_mov_b64 s[4:5], vcc
+; GFX908-NEXT:    v_addc_co_u32_e64 v4, s[4:5], -1, v1, s[4:5]
 ; GFX908-NEXT:    v_addc_co_u32_e32 v1, vcc, -1, v1, vcc
+; GFX908-NEXT:    v_mov_b32_e32 v0, v3
 ; GFX908-NEXT:    flat_load_dword v0, v[0:1]
 ; GFX908-NEXT:    s_mov_b64 s[4:5], 0
 ; GFX908-NEXT:  .LBB14_1: ; %atomicrmw.start
@@ -16759,14 +16761,11 @@ define <2 x half> @flat_agent_atomic_fadd_ret_v2f16__offset12b_neg__amdgpu_no_fi
 ; GFX11-LABEL: flat_agent_atomic_fadd_ret_v2f16__offset12b_neg__amdgpu_no_fine_grained_memory:
 ; GFX11:       ; %bb.0:
 ; GFX11-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX11-NEXT:    v_mov_b32_e32 v3, v0
-; GFX11-NEXT:    s_mov_b32 s0, 0
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX11-NEXT:    v_add_co_u32 v4, vcc_lo, 0xfffff800, v3
-; GFX11-NEXT:    v_add_co_ci_u32_e64 v5, null, -1, v1, vcc_lo
-; GFX11-NEXT:    v_add_co_u32 v3, vcc_lo, 0xfffff800, v3
-; GFX11-NEXT:    flat_load_b32 v0, v[4:5]
+; GFX11-NEXT:    v_add_co_u32 v3, vcc_lo, 0xfffff800, v0
+; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
 ; GFX11-NEXT:    v_add_co_ci_u32_e64 v4, null, -1, v1, vcc_lo
+; GFX11-NEXT:    s_mov_b32 s0, 0
+; GFX11-NEXT:    flat_load_b32 v0, v[3:4]
 ; GFX11-NEXT:  .LBB58_1: ; %atomicrmw.start
 ; GFX11-NEXT:    ; =>This Inner Loop Header: Depth=1
 ; GFX11-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
@@ -16816,9 +16815,10 @@ define <2 x half> @flat_agent_atomic_fadd_ret_v2f16__offset12b_neg__amdgpu_no_fi
 ; GFX90A:       ; %bb.0:
 ; GFX90A-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX90A-NEXT:    v_add_co_u32_e32 v4, vcc, 0xfffff800, v0
-; GFX90A-NEXT:    v_addc_co_u32_e32 v5, vcc, -1, v1, vcc
-; GFX90A-NEXT:    v_add_co_u32_e32 v0, vcc, 0xfffff800, v0
+; GFX90A-NEXT:    s_mov_b64 s[4:5], vcc
+; GFX90A-NEXT:    v_addc_co_u32_e64 v5, s[4:5], -1, v1, s[4:5]
 ; GFX90A-NEXT:    v_addc_co_u32_e32 v1, vcc, -1, v1, vcc
+; GFX90A-NEXT:    v_mov_b32_e32 v0, v4
 ; GFX90A-NEXT:    flat_load_dword v0, v[0:1]
 ; GFX90A-NEXT:    s_mov_b64 s[4:5], 0
 ; GFX90A-NEXT:  .LBB58_1: ; %atomicrmw.start
@@ -16841,9 +16841,10 @@ define <2 x half> @flat_agent_atomic_fadd_ret_v2f16__offset12b_neg__amdgpu_no_fi
 ; GFX908:       ; %bb.0:
 ; GFX908-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX908-NEXT:    v_add_co_u32_e32 v3, vcc, 0xfffff800, v0
-; GFX908-NEXT:    v_addc_co_u32_e32 v4, vcc, -1, v1, vcc
-; GFX908-NEXT:    v_add_co_u32_e32 v0, vcc, 0xfffff800, v0
+; GFX908-NEXT:    s_mov_b64 s[4:5], vcc
+; GFX908-NEXT:    v_addc_co_u32_e64 v4, s[4:5], -1, v1, s[4:5]
 ; GFX908-NEXT:    v_addc_co_u32_e32 v1, vcc, -1, v1, vcc
+; GFX908-NEXT:    v_mov_b32_e32 v0, v3
 ; GFX908-NEXT:    flat_load_dword v0, v[0:1]
 ; GFX908-NEXT:    s_mov_b64 s[4:5], 0
 ; GFX908-NEXT:  .LBB58_1: ; %atomicrmw.start
@@ -17331,13 +17332,11 @@ define void @flat_agent_atomic_fadd_noret_v2f16__offset12b_neg__amdgpu_no_fine_g
 ; GFX11-LABEL: flat_agent_atomic_fadd_noret_v2f16__offset12b_neg__amdgpu_no_fine_grained_memory:
 ; GFX11:       ; %bb.0:
 ; GFX11-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX11-NEXT:    v_add_co_u32 v3, vcc_lo, 0xfffff800, v0
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1)
-; GFX11-NEXT:    v_add_co_ci_u32_e64 v4, null, -1, v1, vcc_lo
 ; GFX11-NEXT:    v_add_co_u32 v0, vcc_lo, 0xfffff800, v0
+; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
 ; GFX11-NEXT:    v_add_co_ci_u32_e64 v1, null, -1, v1, vcc_lo
-; GFX11-NEXT:    flat_load_b32 v4, v[3:4]
 ; GFX11-NEXT:    s_mov_b32 s0, 0
+; GFX11-NEXT:    flat_load_b32 v4, v[0:1]
 ; GFX11-NEXT:  .LBB61_1: ; %atomicrmw.start
 ; GFX11-NEXT:    ; =>This Inner Loop Header: Depth=1
 ; GFX11-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
@@ -17386,9 +17385,10 @@ define void @flat_agent_atomic_fadd_noret_v2f16__offset12b_neg__amdgpu_no_fine_g
 ; GFX90A:       ; %bb.0:
 ; GFX90A-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX90A-NEXT:    v_add_co_u32_e32 v4, vcc, 0xfffff800, v0
-; GFX90A-NEXT:    v_addc_co_u32_e32 v5, vcc, -1, v1, vcc
-; GFX90A-NEXT:    v_add_co_u32_e32 v0, vcc, 0xfffff800, v0
+; GFX90A-NEXT:    s_mov_b64 s[4:5], vcc
+; GFX90A-NEXT:    v_addc_co_u32_e64 v5, s[4:5], -1, v1, s[4:5]
 ; GFX90A-NEXT:    v_addc_co_u32_e32 v1, vcc, -1, v1, vcc
+; GFX90A-NEXT:    v_mov_b32_e32 v0, v4
 ; GFX90A-NEXT:    flat_load_dword v1, v[0:1]
 ; GFX90A-NEXT:    s_mov_b64 s[4:5], 0
 ; GFX90A-NEXT:  .LBB61_1: ; %atomicrmw.start
@@ -17411,9 +17411,10 @@ define void @flat_agent_atomic_fadd_noret_v2f16__offset12b_neg__amdgpu_no_fine_g
 ; GFX908:       ; %bb.0:
 ; GFX908-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX908-NEXT:    v_add_co_u32_e32 v3, vcc, 0xfffff800, v0
-; GFX908-NEXT:    v_addc_co_u32_e32 v4, vcc, -1, v1, vcc
-; GFX908-NEXT:    v_add_co_u32_e32 v0, vcc, 0xfffff800, v0
+; GFX908-NEXT:    s_mov_b64 s[4:5], vcc
+; GFX908-NEXT:    v_addc_co_u32_e64 v4, s[4:5], -1, v1, s[4:5]
 ; GFX908-NEXT:    v_addc_co_u32_e32 v1, vcc, -1, v1, vcc
+; GFX908-NEXT:    v_mov_b32_e32 v0, v3
 ; GFX908-NEXT:    flat_load_dword v1, v[0:1]
 ; GFX908-NEXT:    s_mov_b64 s[4:5], 0
 ; GFX908-NEXT:  .LBB61_1: ; %atomicrmw.start
@@ -19313,16 +19314,13 @@ define <2 x bfloat> @flat_agent_atomic_fadd_ret_v2bf16__offset12b_neg__amdgpu_no
 ; GFX11-TRUE16-LABEL: flat_agent_atomic_fadd_ret_v2bf16__offset12b_neg__amdgpu_no_fine_grained_memory:
 ; GFX11-TRUE16:       ; %bb.0:
 ; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX11-TRUE16-NEXT:    v_mov_b32_e32 v3, v0
-; GFX11-TRUE16-NEXT:    s_mov_b32 s0, 0
-; GFX11-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX11-TRUE16-NEXT:    v_add_co_u32 v4, vcc_lo, 0xfffff800, v3
-; GFX11-TRUE16-NEXT:    v_add_co_ci_u32_e64 v5, null, -1, v1, vcc_lo
-; GFX11-TRUE16-NEXT:    v_add_co_u32 v3, vcc_lo, 0xfffff800, v3
-; GFX11-TRUE16-NEXT:    flat_load_b32 v0, v[4:5]
+; GFX11-TRUE16-NEXT:    v_add_co_u32 v3, vcc_lo, 0xfffff800, v0
+; GFX11-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_1)
 ; GFX11-TRUE16-NEXT:    v_add_co_ci_u32_e64 v4, null, -1, v1, vcc_lo
 ; GFX11-TRUE16-NEXT:    v_and_b32_e32 v1, 0xffff0000, v2
 ; GFX11-TRUE16-NEXT:    v_lshlrev_b32_e32 v2, 16, v2
+; GFX11-TRUE16-NEXT:    s_mov_b32 s0, 0
+; GFX11-TRUE16-NEXT:    flat_load_b32 v0, v[3:4]
 ; GFX11-TRUE16-NEXT:    s_set_inst_prefetch_distance 0x1
 ; GFX11-TRUE16-NEXT:    .p2align 6
 ; GFX11-TRUE16-NEXT:  .LBB70_1: ; %atomicrmw.start
@@ -19367,16 +19365,13 @@ define <2 x bfloat> @flat_agent_atomic_fadd_ret_v2bf16__offset12b_neg__amdgpu_no
 ; GFX11-FAKE16-LABEL: flat_agent_atomic_fadd_ret_v2bf16__offset12b_neg__amdgpu_no_fine_grained_memory:
 ; GFX11-FAKE16:       ; %bb.0:
 ; GFX11-FAKE16-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX11-FAKE16-NEXT:    v_mov_b32_e32 v3, v0
-; GFX11-FAKE16-NEXT:    s_mov_b32 s1, 0
-; GFX11-FAKE16-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX11-FAKE16-NEXT:    v_add_co_u32 v4, vcc_lo, 0xfffff800, v3
-; GFX11-FAKE16-NEXT:    v_add_co_ci_u32_e64 v5, null, -1, v1, vcc_lo
-; GFX11-FAKE16-NEXT:    v_add_co_u32 v3, vcc_lo, 0xfffff800, v3
-; GFX11-FAKE16-NEXT:    flat_load_b32 v0, v[4:5]
+; GFX11-FAKE16-NEXT:    v_add_co_u32 v3, vcc_lo, 0xfffff800, v0
+; GFX11-FAKE16-NEXT:    s_delay_alu instid0(VALU_DEP_1)
 ; GFX11-FAKE16-NEXT:    v_add_co_ci_u32_e64 v4, null, -1, v1, vcc_lo
 ; GFX11-FAKE16-NEXT:    v_lshlrev_b32_e32 v1, 16, v2
 ; GFX11-FAKE16-NEXT:    v_and_b32_e32 v2, 0xffff0000, v2
+; GFX11-FAKE16-NEXT:    s_mov_b32 s1, 0
+; GFX11-FAKE16-NEXT:    flat_load_b32 v0, v[3:4]
 ; GFX11-FAKE16-NEXT:    s_set_inst_prefetch_distance 0x1
 ; GFX11-FAKE16-NEXT:    .p2align 6
 ; GFX11-FAKE16-NEXT:  .LBB70_1: ; %atomicrmw.start
@@ -19463,9 +19458,10 @@ define <2 x bfloat> @flat_agent_atomic_fadd_ret_v2bf16__offset12b_neg__amdgpu_no
 ; GFX90A:       ; %bb.0:
 ; GFX90A-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX90A-NEXT:    v_add_co_u32_e32 v4, vcc, 0xfffff800, v0
-; GFX90A-NEXT:    v_addc_co_u32_e32 v5, vcc, -1, v1, vcc
-; GFX90A-NEXT:    v_add_co_u32_e32 v0, vcc, 0xfffff800, v0
+; GFX90A-NEXT:    s_mov_b64 s[4:5], vcc
+; GFX90A-NEXT:    v_addc_co_u32_e64 v5, s[4:5], -1, v1, s[4:5]
 ; GFX90A-NEXT:    v_addc_co_u32_e32 v1, vcc, -1, v1, vcc
+; GFX90A-NEXT:    v_mov_b32_e32 v0, v4
 ; GFX90A-NEXT:    flat_load_dword v0, v[0:1]
 ; GFX90A-NEXT:    s_mov_b64 s[6:7], 0
 ; GFX90A-NEXT:    v_lshlrev_b32_e32 v1, 16, v2
@@ -19506,9 +19502,10 @@ define <2 x bfloat> @flat_agent_atomic_fadd_ret_v2bf16__offset12b_neg__amdgpu_no
 ; GFX908:       ; %bb.0:
 ; GFX908-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX908-NEXT:    v_add_co_u32_e32 v3, vcc, 0xfffff800, v0
-; GFX908-NEXT:    v_addc_co_u32_e32 v4, vcc, -1, v1, vcc
-; GFX908-NEXT:    v_add_co_u32_e32 v0, vcc, 0xfffff800, v0
+; GFX908-NEXT:    s_mov_b64 s[4:5], vcc
+; GFX908-NEXT:    v_addc_co_u32_e64 v4, s[4:5], -1, v1, s[4:5]
 ; GFX908-NEXT:    v_addc_co_u32_e32 v1, vcc, -1, v1, vcc
+; GFX908-NEXT:    v_mov_b32_e32 v0, v3
 ; GFX908-NEXT:    flat_load_dword v0, v[0:1]
 ; GFX908-NEXT:    s_mov_b64 s[6:7], 0
 ; GFX908-NEXT:    v_lshlrev_b32_e32 v1, 16, v2
@@ -20299,15 +20296,13 @@ define void @flat_agent_atomic_fadd_noret_v2bf16__offset12b_neg__amdgpu_no_fine_
 ; GFX11-TRUE16-LABEL: flat_agent_atomic_fadd_noret_v2bf16__offset12b_neg__amdgpu_no_fine_grained_memory:
 ; GFX11-TRUE16:       ; %bb.0:
 ; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX11-TRUE16-NEXT:    v_add_co_u32 v3, vcc_lo, 0xfffff800, v0
-; GFX11-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1)
-; GFX11-TRUE16-NEXT:    v_add_co_ci_u32_e64 v4, null, -1, v1, vcc_lo
 ; GFX11-TRUE16-NEXT:    v_add_co_u32 v0, vcc_lo, 0xfffff800, v0
+; GFX11-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_1)
 ; GFX11-TRUE16-NEXT:    v_add_co_ci_u32_e64 v1, null, -1, v1, vcc_lo
-; GFX11-TRUE16-NEXT:    flat_load_b32 v3, v[3:4]
 ; GFX11-TRUE16-NEXT:    v_and_b32_e32 v4, 0xffff0000, v2
 ; GFX11-TRUE16-NEXT:    v_lshlrev_b32_e32 v5, 16, v2
 ; GFX11-TRUE16-NEXT:    s_mov_b32 s0, 0
+; GFX11-TRUE16-NEXT:    flat_load_b32 v3, v[0:1]
 ; GFX11-TRUE16-NEXT:    s_set_inst_prefetch_distance 0x1
 ; GFX11-TRUE16-NEXT:    .p2align 6
 ; GFX11-TRUE16-NEXT:  .LBB73_1: ; %atomicrmw.start
@@ -20352,15 +20347,13 @@ define void @flat_agent_atomic_fadd_noret_v2bf16__offset12b_neg__amdgpu_no_fine_
 ; GFX11-FAKE16-LABEL: flat_agent_atomic_fadd_noret_v2bf16__offset12b_neg__amdgpu_no_fine_grained_memory:
 ; GFX11-FAKE16:       ; %bb.0:
 ; GFX11-FAKE16-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX11-FAKE16-NEXT:    v_add_co_u32 v3, vcc_lo, 0xfffff800, v0
-; GFX11-FAKE16-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1)
-; GFX11-FAKE16-NEXT:    v_add_co_ci_u32_e64 v4, null, -1, v1, vcc_lo
 ; GFX11-FAKE16-NEXT:    v_add_co_u32 v0, vcc_lo, 0xfffff800, v0
+; GFX11-FAKE16-NEXT:    s_delay_alu instid0(VALU_DEP_1)
 ; GFX11-FAKE16-NEXT:    v_add_co_ci_u32_e64 v1, null, -1, v1, vcc_lo
-; GFX11-FAKE16-NEXT:    flat_load_b32 v3, v[3:4]
 ; GFX11-FAKE16-NEXT:    v_lshlrev_b32_e32 v4, 16, v2
 ; GFX11-FAKE16-NEXT:    v_and_b32_e32 v5, 0xffff0000, v2
 ; GFX11-FAKE16-NEXT:    s_mov_b32 s1, 0
+; GFX11-FAKE16-NEXT:    flat_load_b32 v3, v[0:1]
 ; GFX11-FAKE16-NEXT:    s_set_inst_prefetch_distance 0x1
 ; GFX11-FAKE16-NEXT:    .p2align 6
 ; GFX11-FAKE16-NEXT:  .LBB73_1: ; %atomicrmw.start
@@ -20446,9 +20439,10 @@ define void @flat_agent_atomic_fadd_noret_v2bf16__offset12b_neg__amdgpu_no_fine_
 ; GFX90A:       ; %bb.0:
 ; GFX90A-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX90A-NEXT:    v_add_co_u32_e32 v4, vcc, 0xfffff800, v0
-; GFX90A-NEXT:    v_addc_co_u32_e32 v5, vcc, -1, v1, vcc
-; GFX90A-NEXT:    v_add_co_u32_e32 v0, vcc, 0xfffff800, v0
+; GFX90A-NEXT:    s_mov_b64 s[4:5], vcc
+; GFX90A-NEXT:    v_addc_co_u32_e64 v5, s[4:5], -1, v1, s[4:5]
 ; GFX90A-NEXT:    v_addc_co_u32_e32 v1, vcc, -1, v1, vcc
+; GFX90A-NEXT:    v_mov_b32_e32 v0, v4
 ; GFX90A-NEXT:    flat_load_dword v1, v[0:1]
 ; GFX90A-NEXT:    s_mov_b64 s[6:7], 0
 ; GFX90A-NEXT:    v_lshlrev_b32_e32 v3, 16, v2
@@ -20489,9 +20483,10 @@ define void @flat_agent_atomic_fadd_noret_v2bf16__offset12b_neg__amdgpu_no_fine_
 ; GFX908:       ; %bb.0:
 ; GFX908-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX908-NEXT:    v_add_co_u32_e32 v3, vcc, 0xfffff800, v0
-; GFX908-NEXT:    v_addc_co_u32_e32 v4, vcc, -1, v1, vcc
-; GFX908-NEXT:    v_add_co_u32_e32 v0, vcc, 0xfffff800, v0
+; GFX908-NEXT:    s_mov_b64 s[4:5], vcc
+; GFX908-NEXT:    v_addc_co_u32_e64 v4, s[4:5], -1, v1, s[4:5]
 ; GFX908-NEXT:    v_addc_co_u32_e32 v1, vcc, -1, v1, vcc
+; GFX908-NEXT:    v_mov_b32_e32 v0, v3
 ; GFX908-NEXT:    flat_load_dword v1, v[0:1]
 ; GFX908-NEXT:    s_mov_b64 s[6:7], 0
 ; GFX908-NEXT:    v_lshlrev_b32_e32 v5, 16, v2