diff --git a/llvm/test/CodeGen/AMDGPU/mad_64_32.ll b/llvm/test/CodeGen/AMDGPU/mad_64_32.ll
--- a/llvm/test/CodeGen/AMDGPU/mad_64_32.ll
+++ b/llvm/test/CodeGen/AMDGPU/mad_64_32.ll
@@ -1,14 +1,22 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc -march=amdgcn -mcpu=hawaii -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN,CI %s
 ; RUN: llc -march=amdgcn -mcpu=tahiti -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN,SI %s
 
-; GCN-LABEL: {{^}}mad_i64_i32_sextops:
-; CI: v_mad_i64_i32 v[0:1], s{{\[[0-9]+:[0-9]+\]}}, v0, v1, v[2:3]
-
-; SI: v_mul_lo_u32
-; SI: v_mul_hi_i32
-; SI: v_add_i32
-; SI: v_addc_u32
 define i64 @mad_i64_i32_sextops(i32 %arg0, i32 %arg1, i64 %arg2) #0 {
+; CI-LABEL: mad_i64_i32_sextops:
+; CI:       ; %bb.0:
+; CI-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; CI-NEXT:    v_mad_i64_i32 v[0:1], s[4:5], v0, v1, v[2:3]
+; CI-NEXT:    s_setpc_b64 s[30:31]
+;
+; SI-LABEL: mad_i64_i32_sextops:
+; SI:       ; %bb.0:
+; SI-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SI-NEXT:    v_mul_lo_u32 v4, v0, v1
+; SI-NEXT:    v_mul_hi_i32 v1, v0, v1
+; SI-NEXT:    v_add_i32_e32 v0, vcc, v4, v2
+; SI-NEXT:    v_addc_u32_e32 v1, vcc, v1, v3, vcc
+; SI-NEXT:    s_setpc_b64 s[30:31]
   %sext0 = sext i32 %arg0 to i64
   %sext1 = sext i32 %arg1 to i64
   %mul = mul i64 %sext0, %sext1
@@ -16,14 +24,21 @@
   ret i64 %mad
 }
 
-; GCN-LABEL: {{^}}mad_i64_i32_sextops_commute:
-; CI: v_mad_i64_i32 v[0:1], s{{\[[0-9]+:[0-9]+\]}}, v0, v1, v[2:3]
-
-; SI-DAG: v_mul_lo_u32
-; SI-DAG: v_mul_hi_i32
-; SI: v_add_i32
-; SI: v_addc_u32
 define i64 @mad_i64_i32_sextops_commute(i32 %arg0, i32 %arg1, i64 %arg2) #0 {
+; CI-LABEL: mad_i64_i32_sextops_commute:
+; CI:       ; %bb.0:
+; CI-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; CI-NEXT:    v_mad_i64_i32 v[0:1], s[4:5], v0, v1, v[2:3]
+; CI-NEXT:    s_setpc_b64 s[30:31]
+;
+; SI-LABEL: mad_i64_i32_sextops_commute:
+; SI:       ; %bb.0:
+; SI-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SI-NEXT:    v_mul_lo_u32 v4, v0, v1
+; SI-NEXT:    v_mul_hi_i32 v1, v0, v1
+; SI-NEXT:    v_add_i32_e32 v0, vcc, v2, v4
+; SI-NEXT:    v_addc_u32_e32 v1, vcc, v3, v1, vcc
+; SI-NEXT:    s_setpc_b64 s[30:31]
   %sext0 = sext i32 %arg0 to i64
   %sext1 = sext i32 %arg1 to i64
   %mul = mul i64 %sext0, %sext1
@@ -31,14 +46,21 @@
   ret i64 %mad
 }
 
-; GCN-LABEL: {{^}}mad_u64_u32_zextops:
-; CI: v_mad_u64_u32 v[0:1], s{{\[[0-9]+:[0-9]+\]}}, v0, v1, v[2:3]
-
-; SI-DAG: v_mul_lo_u32
-; SI-DAG: v_mul_hi_u32
-; SI: v_add_i32
-; SI: v_addc_u32
 define i64 @mad_u64_u32_zextops(i32 %arg0, i32 %arg1, i64 %arg2) #0 {
+; CI-LABEL: mad_u64_u32_zextops:
+; CI:       ; %bb.0:
+; CI-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; CI-NEXT:    v_mad_u64_u32 v[0:1], s[4:5], v0, v1, v[2:3]
+; CI-NEXT:    s_setpc_b64 s[30:31]
+;
+; SI-LABEL: mad_u64_u32_zextops:
+; SI:       ; %bb.0:
+; SI-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SI-NEXT:    v_mul_lo_u32 v4, v0, v1
+; SI-NEXT:    v_mul_hi_u32 v1, v0, v1
+; SI-NEXT:    v_add_i32_e32 v0, vcc, v4, v2
+; SI-NEXT:    v_addc_u32_e32 v1, vcc, v1, v3, vcc
+; SI-NEXT:    s_setpc_b64 s[30:31]
   %sext0 = zext i32 %arg0 to i64
   %sext1 = zext i32 %arg1 to i64
   %mul = mul i64 %sext0, %sext1
@@ -46,14 +68,21 @@
   ret i64 %mad
 }
 
-; GCN-LABEL: {{^}}mad_u64_u32_zextops_commute:
-; CI: v_mad_u64_u32 v[0:1], s{{\[[0-9]+:[0-9]+\]}}, v0, v1, v[2:3]
-
-; SI-DAG: v_mul_lo_u32
-; SI-DAG: v_mul_hi_u32
-; SI: v_add_i32
-; SI: v_addc_u32
 define i64 @mad_u64_u32_zextops_commute(i32 %arg0, i32 %arg1, i64 %arg2) #0 {
+; CI-LABEL: mad_u64_u32_zextops_commute:
+; CI:       ; %bb.0:
+; CI-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; CI-NEXT:    v_mad_u64_u32 v[0:1], s[4:5], v0, v1, v[2:3]
+; CI-NEXT:    s_setpc_b64 s[30:31]
+;
+; SI-LABEL: mad_u64_u32_zextops_commute:
+; SI:       ; %bb.0:
+; SI-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SI-NEXT:    v_mul_lo_u32 v4, v0, v1
+; SI-NEXT:    v_mul_hi_u32 v1, v0, v1
+; SI-NEXT:    v_add_i32_e32 v0, vcc, v2, v4
+; SI-NEXT:    v_addc_u32_e32 v1, vcc, v3, v1, vcc
+; SI-NEXT:    s_setpc_b64 s[30:31]
   %sext0 = zext i32 %arg0 to i64
   %sext1 = zext i32 %arg1 to i64
   %mul = mul i64 %sext0, %sext1
@@ -61,20 +90,66 @@
   ret i64 %mad
 }
 
-
-
-
-
-
-; GCN-LABEL: {{^}}mad_i64_i32_sextops_i32_i128:
-; CI: v_mad_u64_u32
-; CI: v_mad_u64_u32
-; CI: v_mad_i64_i32
-; CI: v_mad_u64_u32
-
-
-; SI-NOT: v_mad_
 define i128 @mad_i64_i32_sextops_i32_i128(i32 %arg0, i32 %arg1, i128 %arg2) #0 {
+; CI-LABEL: mad_i64_i32_sextops_i32_i128:
+; CI:       ; %bb.0:
+; CI-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; CI-NEXT:    v_mul_hi_u32 v6, v0, v1
+; CI-NEXT:    v_ashrrev_i32_e32 v12, 31, v0
+; CI-NEXT:    v_mov_b32_e32 v7, 0
+; CI-NEXT:    v_ashrrev_i32_e32 v13, 31, v1
+; CI-NEXT:    v_mad_u64_u32 v[8:9], s[4:5], v12, v1, v[6:7]
+; CI-NEXT:    v_mul_hi_i32 v11, v1, v12
+; CI-NEXT:    v_mul_lo_u32 v10, v1, v12
+; CI-NEXT:    v_mov_b32_e32 v6, v9
+; CI-NEXT:    v_mov_b32_e32 v9, v7
+; CI-NEXT:    v_mad_u64_u32 v[7:8], s[4:5], v0, v13, v[8:9]
+; CI-NEXT:    v_mad_i64_i32 v[10:11], s[4:5], v13, v0, v[10:11]
+; CI-NEXT:    v_add_i32_e32 v8, vcc, v6, v8
+; CI-NEXT:    v_addc_u32_e64 v9, s[4:5], 0, 0, vcc
+; CI-NEXT:    v_mad_u64_u32 v[8:9], s[4:5], v12, v13, v[8:9]
+; CI-NEXT:    v_mul_lo_u32 v0, v0, v1
+; CI-NEXT:    v_mov_b32_e32 v1, v7
+; CI-NEXT:    v_add_i32_e32 v6, vcc, v8, v10
+; CI-NEXT:    v_addc_u32_e32 v8, vcc, v9, v11, vcc
+; CI-NEXT:    v_add_i32_e32 v0, vcc, v0, v2
+; CI-NEXT:    v_addc_u32_e32 v1, vcc, v1, v3, vcc
+; CI-NEXT:    v_addc_u32_e32 v2, vcc, v6, v4, vcc
+; CI-NEXT:    v_addc_u32_e32 v3, vcc, v8, v5, vcc
+; CI-NEXT:    s_setpc_b64 s[30:31]
+;
+; SI-LABEL: mad_i64_i32_sextops_i32_i128:
+; SI:       ; %bb.0:
+; SI-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SI-NEXT:    v_ashrrev_i32_e32 v6, 31, v0
+; SI-NEXT:    v_mul_lo_u32 v11, v6, v1
+; SI-NEXT:    v_mul_hi_u32 v12, v0, v1
+; SI-NEXT:    v_ashrrev_i32_e32 v7, 31, v1
+; SI-NEXT:    v_mul_hi_u32 v14, v6, v1
+; SI-NEXT:    v_mul_lo_u32 v13, v0, v7
+; SI-NEXT:    v_mul_hi_u32 v10, v0, v7
+; SI-NEXT:    v_add_i32_e32 v12, vcc, v11, v12
+; SI-NEXT:    v_addc_u32_e32 v14, vcc, 0, v14, vcc
+; SI-NEXT:    v_mul_hi_u32 v8, v6, v7
+; SI-NEXT:    v_add_i32_e32 v12, vcc, v13, v12
+; SI-NEXT:    v_addc_u32_e32 v10, vcc, 0, v10, vcc
+; SI-NEXT:    v_mul_i32_i24_e32 v9, v6, v7
+; SI-NEXT:    v_add_i32_e32 v10, vcc, v14, v10
+; SI-NEXT:    v_mul_hi_i32 v6, v1, v6
+; SI-NEXT:    v_mul_hi_i32 v7, v7, v0
+; SI-NEXT:    v_addc_u32_e64 v14, s[4:5], 0, 0, vcc
+; SI-NEXT:    v_add_i32_e32 v9, vcc, v9, v10
+; SI-NEXT:    v_addc_u32_e32 v8, vcc, v8, v14, vcc
+; SI-NEXT:    v_add_i32_e32 v10, vcc, v13, v11
+; SI-NEXT:    v_mul_lo_u32 v0, v0, v1
+; SI-NEXT:    v_addc_u32_e32 v6, vcc, v7, v6, vcc
+; SI-NEXT:    v_add_i32_e32 v7, vcc, v9, v10
+; SI-NEXT:    v_addc_u32_e32 v6, vcc, v8, v6, vcc
+; SI-NEXT:    v_add_i32_e32 v0, vcc, v0, v2
+; SI-NEXT:    v_addc_u32_e32 v1, vcc, v12, v3, vcc
+; SI-NEXT:    v_addc_u32_e32 v2, vcc, v7, v4, vcc
+; SI-NEXT:    v_addc_u32_e32 v3, vcc, v6, v5, vcc
+; SI-NEXT:    s_setpc_b64 s[30:31]
   %sext0 = sext i32 %arg0 to i128
   %sext1 = sext i32 %arg1 to i128
   %mul = mul i128 %sext0, %sext1
@@ -82,13 +157,23 @@
   ret i128 %mad
 }
 
-; GCN-LABEL: {{^}}mad_i64_i32_sextops_i32_i63:
-; CI: v_lshl_b64
-; CI: v_ashr
-; CI: v_mad_i64_i32 v[0:1], s{{\[[0-9]+:[0-9]+\]}}, v0, v1, v[2:3]
-
-; SI-NOT: v_mad_u64_u32
 define i63 @mad_i64_i32_sextops_i32_i63(i32 %arg0, i32 %arg1, i63 %arg2) #0 {
+; CI-LABEL: mad_i64_i32_sextops_i32_i63:
+; CI:       ; %bb.0:
+; CI-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; CI-NEXT:    v_lshl_b64 v[2:3], v[2:3], 1
+; CI-NEXT:    v_ashr_i64 v[2:3], v[2:3], 1
+; CI-NEXT:    v_mad_i64_i32 v[0:1], s[4:5], v0, v1, v[2:3]
+; CI-NEXT:    s_setpc_b64 s[30:31]
+;
+; SI-LABEL: mad_i64_i32_sextops_i32_i63:
+; SI:       ; %bb.0:
+; SI-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SI-NEXT:    v_mul_lo_u32 v4, v0, v1
+; SI-NEXT:    v_mul_hi_i32 v1, v0, v1
+; SI-NEXT:    v_add_i32_e32 v0, vcc, v4, v2
+; SI-NEXT:    v_addc_u32_e32 v1, vcc, v1, v3, vcc
+; SI-NEXT:    s_setpc_b64 s[30:31]
   %sext0 = sext i32 %arg0 to i63
   %sext1 = sext i32 %arg1 to i63
   %mul = mul i63 %sext0, %sext1
@@ -96,13 +181,29 @@
   ret i63 %mad
 }
 
-; GCN-LABEL: {{^}}mad_i64_i32_sextops_i31_i63:
-; CI: v_lshl_b64
-; CI: v_bfe_i32 v[[B1:[0-9]+]], v1, 0, 31
-; CI: v_ashr_i64
-; CI: v_bfe_i32 v[[B2:[0-9]+]], v0, 0, 31
-; CI: v_mad_i64_i32 v[0:1], s{{\[[0-9]+:[0-9]+\]}}, v[[B2]], v[[B1]], v{{\[[0-9]+:[0-9]+\]}}
 define i63 @mad_i64_i32_sextops_i31_i63(i31 %arg0, i31 %arg1, i63 %arg2) #0 {
+; CI-LABEL: mad_i64_i32_sextops_i31_i63:
+; CI:       ; %bb.0:
+; CI-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; CI-NEXT:    v_lshl_b64 v[2:3], v[2:3], 1
+; CI-NEXT:    v_bfe_i32 v1, v1, 0, 31
+; CI-NEXT:    v_ashr_i64 v[2:3], v[2:3], 1
+; CI-NEXT:    v_bfe_i32 v0, v0, 0, 31
+; CI-NEXT:    v_mad_i64_i32 v[0:1], s[4:5], v0, v1, v[2:3]
+; CI-NEXT:    s_setpc_b64 s[30:31]
+;
+; SI-LABEL: mad_i64_i32_sextops_i31_i63:
+; SI:       ; %bb.0:
+; SI-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SI-NEXT:    v_lshlrev_b32_e32 v4, 1, v0
+; SI-NEXT:    v_lshlrev_b32_e32 v1, 1, v1
+; SI-NEXT:    v_ashr_i64 v[4:5], v[3:4], 33
+; SI-NEXT:    v_ashr_i64 v[0:1], v[0:1], 33
+; SI-NEXT:    v_mul_lo_u32 v1, v4, v0
+; SI-NEXT:    v_mul_hi_i32 v4, v4, v0
+; SI-NEXT:    v_add_i32_e32 v0, vcc, v1, v2
+; SI-NEXT:    v_addc_u32_e32 v1, vcc, v4, v3, vcc
+; SI-NEXT:    s_setpc_b64 s[30:31]
   %sext0 = sext i31 %arg0 to i63
   %sext1 = sext i31 %arg1 to i63
   %mul = mul i63 %sext0, %sext1
@@ -110,9 +211,21 @@
   ret i63 %mad
 }
 
-; GCN-LABEL: {{^}}mad_u64_u32_bitops:
-; CI: v_mad_u64_u32 v[0:1], s{{\[[0-9]+:[0-9]+\]}}, v0, v2, v[4:5]
 define i64 @mad_u64_u32_bitops(i64 %arg0, i64 %arg1, i64 %arg2) #0 {
+; CI-LABEL: mad_u64_u32_bitops:
+; CI:       ; %bb.0:
+; CI-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; CI-NEXT:    v_mad_u64_u32 v[0:1], s[4:5], v0, v2, v[4:5]
+; CI-NEXT:    s_setpc_b64 s[30:31]
+;
+; SI-LABEL: mad_u64_u32_bitops:
+; SI:       ; %bb.0:
+; SI-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SI-NEXT:    v_mul_lo_u32 v1, v0, v2
+; SI-NEXT:    v_mul_hi_u32 v2, v0, v2
+; SI-NEXT:    v_add_i32_e32 v0, vcc, v1, v4
+; SI-NEXT:    v_addc_u32_e32 v1, vcc, v2, v5, vcc
+; SI-NEXT:    s_setpc_b64 s[30:31]
   %trunc.lhs = and i64 %arg0, 4294967295
   %trunc.rhs = and i64 %arg1, 4294967295
   %mul = mul i64 %trunc.lhs, %trunc.rhs
@@ -120,9 +233,18 @@
   ret i64 %add
 }
 
-; GCN-LABEL: {{^}}mad_u64_u32_bitops_lhs_mask_small:
-; GCN-NOT: v_mad_
 define i64 @mad_u64_u32_bitops_lhs_mask_small(i64 %arg0, i64 %arg1, i64 %arg2) #0 {
+; GCN-LABEL: mad_u64_u32_bitops_lhs_mask_small:
+; GCN:       ; %bb.0:
+; GCN-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT:    v_and_b32_e32 v1, 1, v1
+; GCN-NEXT:    v_mul_hi_u32 v3, v0, v2
+; GCN-NEXT:    v_mul_lo_u32 v1, v1, v2
+; GCN-NEXT:    v_mul_lo_u32 v0, v0, v2
+; GCN-NEXT:    v_add_i32_e32 v1, vcc, v3, v1
+; GCN-NEXT:    v_add_i32_e32 v0, vcc, v0, v4
+; GCN-NEXT:    v_addc_u32_e32 v1, vcc, v1, v5, vcc
+; GCN-NEXT:    s_setpc_b64 s[30:31]
   %trunc.lhs = and i64 %arg0, 8589934591
   %trunc.rhs = and i64 %arg1, 4294967295
   %mul = mul i64 %trunc.lhs, %trunc.rhs
@@ -130,9 +252,18 @@
   ret i64 %add
 }
 
-; GCN-LABEL: {{^}}mad_u64_u32_bitops_rhs_mask_small:
-; GCN-NOT: v_mad_
 define i64 @mad_u64_u32_bitops_rhs_mask_small(i64 %arg0, i64 %arg1, i64 %arg2) #0 {
+; GCN-LABEL: mad_u64_u32_bitops_rhs_mask_small:
+; GCN:       ; %bb.0:
+; GCN-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT:    v_and_b32_e32 v1, 1, v3
+; GCN-NEXT:    v_mul_hi_u32 v3, v0, v2
+; GCN-NEXT:    v_mul_lo_u32 v1, v0, v1
+; GCN-NEXT:    v_mul_lo_u32 v0, v0, v2
+; GCN-NEXT:    v_add_i32_e32 v1, vcc, v3, v1
+; GCN-NEXT:    v_add_i32_e32 v0, vcc, v0, v4
+; GCN-NEXT:    v_addc_u32_e32 v1, vcc, v1, v5, vcc
+; GCN-NEXT:    s_setpc_b64 s[30:31]
   %trunc.lhs = and i64 %arg0, 4294967295
   %trunc.rhs = and i64 %arg1, 8589934591
   %mul = mul i64 %trunc.lhs, %trunc.rhs
@@ -140,10 +271,21 @@
   ret i64 %add
 }
 
-; GCN-LABEL: {{^}}mad_i64_i32_bitops:
-; CI: v_mad_i64_i32 v[0:1], s{{\[[0-9]+:[0-9]+\]}}, v0, v2, v[4:5]
-; SI-NOT: v_mad_
 define i64 @mad_i64_i32_bitops(i64 %arg0, i64 %arg1, i64 %arg2) #0 {
+; CI-LABEL: mad_i64_i32_bitops:
+; CI:       ; %bb.0:
+; CI-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; CI-NEXT:    v_mad_i64_i32 v[0:1], s[4:5], v0, v2, v[4:5]
+; CI-NEXT:    s_setpc_b64 s[30:31]
+;
+; SI-LABEL: mad_i64_i32_bitops:
+; SI:       ; %bb.0:
+; SI-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SI-NEXT:    v_mul_lo_u32 v1, v0, v2
+; SI-NEXT:    v_mul_hi_i32 v2, v0, v2
+; SI-NEXT:    v_add_i32_e32 v0, vcc, v1, v4
+; SI-NEXT:    v_addc_u32_e32 v1, vcc, v2, v5, vcc
+; SI-NEXT:    s_setpc_b64 s[30:31]
   %shl.lhs = shl i64 %arg0, 32
   %trunc.lhs = ashr i64 %shl.lhs, 32
   %shl.rhs = shl i64 %arg1, 32
@@ -154,10 +296,21 @@
 }
 
 ; Example from bug report
-; GCN-LABEL: {{^}}mad_i64_i32_unpack_i64ops:
-; CI: v_mad_u64_u32 v[0:1], s{{\[[0-9]+:[0-9]+\]}}, v1, v0, v[0:1]
-; SI-NOT: v_mad_u64_u32
 define i64 @mad_i64_i32_unpack_i64ops(i64 %arg0) #0 {
+; CI-LABEL: mad_i64_i32_unpack_i64ops:
+; CI:       ; %bb.0:
+; CI-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; CI-NEXT:    v_mad_u64_u32 v[0:1], s[4:5], v1, v0, v[0:1]
+; CI-NEXT:    s_setpc_b64 s[30:31]
+;
+; SI-LABEL: mad_i64_i32_unpack_i64ops:
+; SI:       ; %bb.0:
+; SI-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SI-NEXT:    v_mul_lo_u32 v2, v1, v0
+; SI-NEXT:    v_mul_hi_u32 v3, v1, v0
+; SI-NEXT:    v_add_i32_e32 v0, vcc, v2, v0
+; SI-NEXT:    v_addc_u32_e32 v1, vcc, v3, v1, vcc
+; SI-NEXT:    s_setpc_b64 s[30:31]
   %tmp4 = lshr i64 %arg0, 32
   %tmp5 = and i64 %arg0, 4294967295
   %mul = mul nuw i64 %tmp4, %tmp5