AMDGPU: Split flat offsets that don't fit in DAG

We handle it this way for some other address spaces.

Since r349196, SILoadStoreOptimizer has been trying to do this. This
is after SIFoldOperands runs, which can change the addressing
patterns. It's simpler to just split this earlier.

llvm-svn: 375366

1 files changed, 170 insertions, 208 deletions

diff --git a/llvm/test/CodeGen/AMDGPU/offset-split-global.ll b/llvm/test/CodeGen/AMDGPU/offset-split-global.ll
index ad32b739bdb..add4e687926 100644
--- a/llvm/test/CodeGen/AMDGPU/offset-split-global.ll
+++ b/llvm/test/CodeGen/AMDGPU/offset-split-global.ll
@@ -59,10 +59,10 @@ define i8 @global_inst_valu_offset_12bit_max(i8 addrspace(1)* %p) {
 ; GFX10:       ; %bb.0:
 ; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
-; GFX10-NEXT:    v_add_co_u32_e64 v0, vcc_lo, 0xfff, v0
+; GFX10-NEXT:    v_add_co_u32_e64 v0, vcc_lo, 0x800, v0
 ; GFX10-NEXT:    ; implicit-def: $vcc_hi
 ; GFX10-NEXT:    v_add_co_ci_u32_e32 v1, vcc_lo, 0, v1, vcc_lo
-; GFX10-NEXT:    global_load_ubyte v0, v[0:1], off
+; GFX10-NEXT:    global_load_ubyte v0, v[0:1], off offset:2047
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    s_setpc_b64 s[30:31]
   %gep = getelementptr i8, i8 addrspace(1)* %p, i64 4095
@@ -74,9 +74,9 @@ define i8 @global_inst_valu_offset_13bit_max(i8 addrspace(1)* %p) {
 ; GFX9-LABEL: global_inst_valu_offset_13bit_max:
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX9-NEXT:    v_add_co_u32_e32 v0, vcc, 0x1fff, v0
+; GFX9-NEXT:    v_add_co_u32_e32 v0, vcc, 0x1000, v0
 ; GFX9-NEXT:    v_addc_co_u32_e32 v1, vcc, 0, v1, vcc
-; GFX9-NEXT:    global_load_ubyte v0, v[0:1], off
+; GFX9-NEXT:    global_load_ubyte v0, v[0:1], off offset:4095
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
 ;
@@ -84,10 +84,10 @@ define i8 @global_inst_valu_offset_13bit_max(i8 addrspace(1)* %p) {
 ; GFX10:       ; %bb.0:
 ; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
-; GFX10-NEXT:    v_add_co_u32_e64 v0, vcc_lo, 0x1fff, v0
+; GFX10-NEXT:    v_add_co_u32_e64 v0, vcc_lo, 0x1800, v0
 ; GFX10-NEXT:    ; implicit-def: $vcc_hi
 ; GFX10-NEXT:    v_add_co_ci_u32_e32 v1, vcc_lo, 0, v1, vcc_lo
-; GFX10-NEXT:    global_load_ubyte v0, v[0:1], off
+; GFX10-NEXT:    global_load_ubyte v0, v[0:1], off offset:2047
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    s_setpc_b64 s[30:31]
   %gep = getelementptr i8, i8 addrspace(1)* %p, i64 8191
@@ -176,10 +176,10 @@ define i8 @global_inst_valu_offset_2x_11bit_max(i8 addrspace(1)* %p) {
 ; GFX10:       ; %bb.0:
 ; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
-; GFX10-NEXT:    v_add_co_u32_e64 v0, vcc_lo, 0xfff, v0
+; GFX10-NEXT:    v_add_co_u32_e64 v0, vcc_lo, 0x800, v0
 ; GFX10-NEXT:    ; implicit-def: $vcc_hi
 ; GFX10-NEXT:    v_add_co_ci_u32_e32 v1, vcc_lo, 0, v1, vcc_lo
-; GFX10-NEXT:    global_load_ubyte v0, v[0:1], off
+; GFX10-NEXT:    global_load_ubyte v0, v[0:1], off offset:2047
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    s_setpc_b64 s[30:31]
   %gep = getelementptr i8, i8 addrspace(1)* %p, i64 4095
@@ -191,9 +191,9 @@ define i8 @global_inst_valu_offset_2x_12bit_max(i8 addrspace(1)* %p) {
 ; GFX9-LABEL: global_inst_valu_offset_2x_12bit_max:
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX9-NEXT:    v_add_co_u32_e32 v0, vcc, 0x1fff, v0
+; GFX9-NEXT:    v_add_co_u32_e32 v0, vcc, 0x1000, v0
 ; GFX9-NEXT:    v_addc_co_u32_e32 v1, vcc, 0, v1, vcc
-; GFX9-NEXT:    global_load_ubyte v0, v[0:1], off
+; GFX9-NEXT:    global_load_ubyte v0, v[0:1], off offset:4095
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
 ;
@@ -201,10 +201,10 @@ define i8 @global_inst_valu_offset_2x_12bit_max(i8 addrspace(1)* %p) {
 ; GFX10:       ; %bb.0:
 ; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
-; GFX10-NEXT:    v_add_co_u32_e64 v0, vcc_lo, 0x1fff, v0
+; GFX10-NEXT:    v_add_co_u32_e64 v0, vcc_lo, 0x1800, v0
 ; GFX10-NEXT:    ; implicit-def: $vcc_hi
 ; GFX10-NEXT:    v_add_co_ci_u32_e32 v1, vcc_lo, 0, v1, vcc_lo
-; GFX10-NEXT:    global_load_ubyte v0, v[0:1], off
+; GFX10-NEXT:    global_load_ubyte v0, v[0:1], off offset:2047
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    s_setpc_b64 s[30:31]
   %gep = getelementptr i8, i8 addrspace(1)* %p, i64 8191
@@ -216,9 +216,9 @@ define i8 @global_inst_valu_offset_2x_13bit_max(i8 addrspace(1)* %p) {
 ; GFX9-LABEL: global_inst_valu_offset_2x_13bit_max:
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX9-NEXT:    v_add_co_u32_e32 v0, vcc, 0x3fff, v0
+; GFX9-NEXT:    v_add_co_u32_e32 v0, vcc, 0x3000, v0
 ; GFX9-NEXT:    v_addc_co_u32_e32 v1, vcc, 0, v1, vcc
-; GFX9-NEXT:    global_load_ubyte v0, v[0:1], off
+; GFX9-NEXT:    global_load_ubyte v0, v[0:1], off offset:4095
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
 ;
@@ -226,10 +226,10 @@ define i8 @global_inst_valu_offset_2x_13bit_max(i8 addrspace(1)* %p) {
 ; GFX10:       ; %bb.0:
 ; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
-; GFX10-NEXT:    v_add_co_u32_e64 v0, vcc_lo, 0x3fff, v0
+; GFX10-NEXT:    v_add_co_u32_e64 v0, vcc_lo, 0x3800, v0
 ; GFX10-NEXT:    ; implicit-def: $vcc_hi
 ; GFX10-NEXT:    v_add_co_ci_u32_e32 v1, vcc_lo, 0, v1, vcc_lo
-; GFX10-NEXT:    global_load_ubyte v0, v[0:1], off
+; GFX10-NEXT:    global_load_ubyte v0, v[0:1], off offset:2047
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    s_setpc_b64 s[30:31]
   %gep = getelementptr i8, i8 addrspace(1)* %p, i64 16383
@@ -315,9 +315,9 @@ define i8 @global_inst_valu_offset_64bit_11bit_split0(i8 addrspace(1)* %p) {
 ; GFX9-LABEL: global_inst_valu_offset_64bit_11bit_split0:
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX9-NEXT:    v_add_co_u32_e32 v0, vcc, 0x7ff, v0
+; GFX9-NEXT:    v_add_co_u32_e32 v0, vcc, 0, v0
 ; GFX9-NEXT:    v_addc_co_u32_e32 v1, vcc, 2, v1, vcc
-; GFX9-NEXT:    global_load_ubyte v0, v[0:1], off
+; GFX9-NEXT:    global_load_ubyte v0, v[0:1], off offset:2047
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
 ;
@@ -325,10 +325,10 @@ define i8 @global_inst_valu_offset_64bit_11bit_split0(i8 addrspace(1)* %p) {
 ; GFX10:       ; %bb.0:
 ; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
-; GFX10-NEXT:    v_add_co_u32_e64 v0, vcc_lo, 0x7ff, v0
+; GFX10-NEXT:    v_add_co_u32_e64 v0, vcc_lo, 0, v0
 ; GFX10-NEXT:    ; implicit-def: $vcc_hi
 ; GFX10-NEXT:    v_add_co_ci_u32_e32 v1, vcc_lo, 2, v1, vcc_lo
-; GFX10-NEXT:    global_load_ubyte v0, v[0:1], off
+; GFX10-NEXT:    global_load_ubyte v0, v[0:1], off offset:2047
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    s_setpc_b64 s[30:31]
   %gep = getelementptr i8, i8 addrspace(1)* %p, i64 8589936639
@@ -341,9 +341,9 @@ define i8 @global_inst_valu_offset_64bit_11bit_split1(i8 addrspace(1)* %p) {
 ; GFX9-LABEL: global_inst_valu_offset_64bit_11bit_split1:
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX9-NEXT:    v_add_co_u32_e32 v0, vcc, 0x800, v0
+; GFX9-NEXT:    v_add_co_u32_e32 v0, vcc, 0, v0
 ; GFX9-NEXT:    v_addc_co_u32_e32 v1, vcc, 2, v1, vcc
-; GFX9-NEXT:    global_load_ubyte v0, v[0:1], off
+; GFX9-NEXT:    global_load_ubyte v0, v[0:1], off offset:2048
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
 ;
@@ -367,9 +367,9 @@ define i8 @global_inst_valu_offset_64bit_12bit_split0(i8 addrspace(1)* %p) {
 ; GFX9-LABEL: global_inst_valu_offset_64bit_12bit_split0:
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX9-NEXT:    v_add_co_u32_e32 v0, vcc, 0xfff, v0
+; GFX9-NEXT:    v_add_co_u32_e32 v0, vcc, 0, v0
 ; GFX9-NEXT:    v_addc_co_u32_e32 v1, vcc, 2, v1, vcc
-; GFX9-NEXT:    global_load_ubyte v0, v[0:1], off
+; GFX9-NEXT:    global_load_ubyte v0, v[0:1], off offset:4095
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
 ;
@@ -377,10 +377,10 @@ define i8 @global_inst_valu_offset_64bit_12bit_split0(i8 addrspace(1)* %p) {
 ; GFX10:       ; %bb.0:
 ; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
-; GFX10-NEXT:    v_add_co_u32_e64 v0, vcc_lo, 0xfff, v0
+; GFX10-NEXT:    v_add_co_u32_e64 v0, vcc_lo, 0x800, v0
 ; GFX10-NEXT:    ; implicit-def: $vcc_hi
 ; GFX10-NEXT:    v_add_co_ci_u32_e32 v1, vcc_lo, 2, v1, vcc_lo
-; GFX10-NEXT:    global_load_ubyte v0, v[0:1], off
+; GFX10-NEXT:    global_load_ubyte v0, v[0:1], off offset:2047
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    s_setpc_b64 s[30:31]
   %gep = getelementptr i8, i8 addrspace(1)* %p, i64 8589938687
@@ -419,9 +419,9 @@ define i8 @global_inst_valu_offset_64bit_13bit_split0(i8 addrspace(1)* %p) {
 ; GFX9-LABEL: global_inst_valu_offset_64bit_13bit_split0:
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX9-NEXT:    v_add_co_u32_e32 v0, vcc, 0x1fff, v0
+; GFX9-NEXT:    v_add_co_u32_e32 v0, vcc, 0x1000, v0
 ; GFX9-NEXT:    v_addc_co_u32_e32 v1, vcc, 2, v1, vcc
-; GFX9-NEXT:    global_load_ubyte v0, v[0:1], off
+; GFX9-NEXT:    global_load_ubyte v0, v[0:1], off offset:4095
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
 ;
@@ -429,10 +429,10 @@ define i8 @global_inst_valu_offset_64bit_13bit_split0(i8 addrspace(1)* %p) {
 ; GFX10:       ; %bb.0:
 ; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
-; GFX10-NEXT:    v_add_co_u32_e64 v0, vcc_lo, 0x1fff, v0
+; GFX10-NEXT:    v_add_co_u32_e64 v0, vcc_lo, 0x1800, v0
 ; GFX10-NEXT:    ; implicit-def: $vcc_hi
 ; GFX10-NEXT:    v_add_co_ci_u32_e32 v1, vcc_lo, 2, v1, vcc_lo
-; GFX10-NEXT:    global_load_ubyte v0, v[0:1], off
+; GFX10-NEXT:    global_load_ubyte v0, v[0:1], off offset:2047
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    s_setpc_b64 s[30:31]
   %gep = getelementptr i8, i8 addrspace(1)* %p, i64 8589942783
@@ -471,10 +471,10 @@ define i8 @global_inst_valu_offset_64bit_11bit_neg_high_split0(i8 addrspace(1)*
 ; GFX9-LABEL: global_inst_valu_offset_64bit_11bit_neg_high_split0:
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX9-NEXT:    v_add_co_u32_e32 v0, vcc, 0x7ff, v0
 ; GFX9-NEXT:    v_bfrev_b32_e32 v2, 1
-; GFX9-NEXT:    v_addc_co_u32_e32 v1, vcc, v1, v2, vcc
-; GFX9-NEXT:    global_load_ubyte v0, v[0:1], off
+; GFX9-NEXT:    v_add_co_u32_e32 v0, vcc, 0, v0
+; GFX9-NEXT:    v_addc_co_u32_e32 v1, vcc, v2, v1, vcc
+; GFX9-NEXT:    global_load_ubyte v0, v[0:1], off offset:2047
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
 ;
@@ -482,10 +482,10 @@ define i8 @global_inst_valu_offset_64bit_11bit_neg_high_split0(i8 addrspace(1)*
 ; GFX10:       ; %bb.0:
 ; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
-; GFX10-NEXT:    v_add_co_u32_e64 v0, vcc_lo, 0x7ff, v0
+; GFX10-NEXT:    v_add_co_u32_e64 v0, vcc_lo, 0, v0
 ; GFX10-NEXT:    ; implicit-def: $vcc_hi
 ; GFX10-NEXT:    v_add_co_ci_u32_e32 v1, vcc_lo, 0x80000000, v1, vcc_lo
-; GFX10-NEXT:    global_load_ubyte v0, v[0:1], off
+; GFX10-NEXT:    global_load_ubyte v0, v[0:1], off offset:2047
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    s_setpc_b64 s[30:31]
   %gep = getelementptr i8, i8 addrspace(1)* %p, i64 -9223372036854773761
@@ -498,10 +498,10 @@ define i8 @global_inst_valu_offset_64bit_11bit_neg_high_split1(i8 addrspace(1)*
 ; GFX9-LABEL: global_inst_valu_offset_64bit_11bit_neg_high_split1:
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX9-NEXT:    v_add_co_u32_e32 v0, vcc, 0x800, v0
 ; GFX9-NEXT:    v_bfrev_b32_e32 v2, 1
-; GFX9-NEXT:    v_addc_co_u32_e32 v1, vcc, v1, v2, vcc
-; GFX9-NEXT:    global_load_ubyte v0, v[0:1], off
+; GFX9-NEXT:    v_add_co_u32_e32 v0, vcc, 0, v0
+; GFX9-NEXT:    v_addc_co_u32_e32 v1, vcc, v2, v1, vcc
+; GFX9-NEXT:    global_load_ubyte v0, v[0:1], off offset:2048
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
 ;
@@ -509,10 +509,10 @@ define i8 @global_inst_valu_offset_64bit_11bit_neg_high_split1(i8 addrspace(1)*
 ; GFX10:       ; %bb.0:
 ; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
-; GFX10-NEXT:    v_add_co_u32_e64 v0, vcc_lo, 0x800, v0
+; GFX10-NEXT:    v_add_co_u32_e64 v0, vcc_lo, 0x1000, v0
 ; GFX10-NEXT:    ; implicit-def: $vcc_hi
 ; GFX10-NEXT:    v_add_co_ci_u32_e32 v1, vcc_lo, 0x80000000, v1, vcc_lo
-; GFX10-NEXT:    global_load_ubyte v0, v[0:1], off
+; GFX10-NEXT:    global_load_ubyte v0, v[0:1], off offset:-2048
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    s_setpc_b64 s[30:31]
   %gep = getelementptr i8, i8 addrspace(1)* %p, i64 -9223372036854773760
@@ -525,10 +525,10 @@ define i8 @global_inst_valu_offset_64bit_12bit_neg_high_split0(i8 addrspace(1)*
 ; GFX9-LABEL: global_inst_valu_offset_64bit_12bit_neg_high_split0:
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX9-NEXT:    v_add_co_u32_e32 v0, vcc, 0xfff, v0
 ; GFX9-NEXT:    v_bfrev_b32_e32 v2, 1
-; GFX9-NEXT:    v_addc_co_u32_e32 v1, vcc, v1, v2, vcc
-; GFX9-NEXT:    global_load_ubyte v0, v[0:1], off
+; GFX9-NEXT:    v_add_co_u32_e32 v0, vcc, 0, v0
+; GFX9-NEXT:    v_addc_co_u32_e32 v1, vcc, v2, v1, vcc
+; GFX9-NEXT:    global_load_ubyte v0, v[0:1], off offset:4095
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
 ;
@@ -536,10 +536,10 @@ define i8 @global_inst_valu_offset_64bit_12bit_neg_high_split0(i8 addrspace(1)*
 ; GFX10:       ; %bb.0:
 ; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
-; GFX10-NEXT:    v_add_co_u32_e64 v0, vcc_lo, 0xfff, v0
+; GFX10-NEXT:    v_add_co_u32_e64 v0, vcc_lo, 0x1000, v0
 ; GFX10-NEXT:    ; implicit-def: $vcc_hi
 ; GFX10-NEXT:    v_add_co_ci_u32_e32 v1, vcc_lo, 0x80000000, v1, vcc_lo
-; GFX10-NEXT:    global_load_ubyte v0, v[0:1], off
+; GFX10-NEXT:    global_load_ubyte v0, v[0:1], off offset:-1
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    s_setpc_b64 s[30:31]
   %gep = getelementptr i8, i8 addrspace(1)* %p, i64 -9223372036854771713
@@ -552,10 +552,10 @@ define i8 @global_inst_valu_offset_64bit_12bit_neg_high_split1(i8 addrspace(1)*
 ; GFX9-LABEL: global_inst_valu_offset_64bit_12bit_neg_high_split1:
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX9-NEXT:    v_add_co_u32_e32 v0, vcc, 0x1000, v0
+; GFX9-NEXT:    v_add_co_u32_e32 v0, vcc, 0x2000, v0
 ; GFX9-NEXT:    v_bfrev_b32_e32 v2, 1
-; GFX9-NEXT:    v_addc_co_u32_e32 v1, vcc, v1, v2, vcc
-; GFX9-NEXT:    global_load_ubyte v0, v[0:1], off
+; GFX9-NEXT:    v_addc_co_u32_e32 v1, vcc, v2, v1, vcc
+; GFX9-NEXT:    global_load_ubyte v0, v[0:1], off offset:-4096
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
 ;
@@ -579,10 +579,10 @@ define i8 @global_inst_valu_offset_64bit_13bit_neg_high_split0(i8 addrspace(1)*
 ; GFX9-LABEL: global_inst_valu_offset_64bit_13bit_neg_high_split0:
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX9-NEXT:    v_add_co_u32_e32 v0, vcc, 0x1fff, v0
+; GFX9-NEXT:    v_add_co_u32_e32 v0, vcc, 0x2000, v0
 ; GFX9-NEXT:    v_bfrev_b32_e32 v2, 1
-; GFX9-NEXT:    v_addc_co_u32_e32 v1, vcc, v1, v2, vcc
-; GFX9-NEXT:    global_load_ubyte v0, v[0:1], off
+; GFX9-NEXT:    v_addc_co_u32_e32 v1, vcc, v2, v1, vcc
+; GFX9-NEXT:    global_load_ubyte v0, v[0:1], off offset:-1
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
 ;
@@ -590,10 +590,10 @@ define i8 @global_inst_valu_offset_64bit_13bit_neg_high_split0(i8 addrspace(1)*
 ; GFX10:       ; %bb.0:
 ; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
-; GFX10-NEXT:    v_add_co_u32_e64 v0, vcc_lo, 0x1fff, v0
+; GFX10-NEXT:    v_add_co_u32_e64 v0, vcc_lo, 0x2000, v0
 ; GFX10-NEXT:    ; implicit-def: $vcc_hi
 ; GFX10-NEXT:    v_add_co_ci_u32_e32 v1, vcc_lo, 0x80000000, v1, vcc_lo
-; GFX10-NEXT:    global_load_ubyte v0, v[0:1], off
+; GFX10-NEXT:    global_load_ubyte v0, v[0:1], off offset:-1
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    s_setpc_b64 s[30:31]
   %gep = getelementptr i8, i8 addrspace(1)* %p, i64 -9223372036854767617
@@ -608,7 +608,7 @@ define i8 @global_inst_valu_offset_64bit_13bit_neg_high_split1(i8 addrspace(1)*
 ; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX9-NEXT:    v_add_co_u32_e32 v0, vcc, 0x2000, v0
 ; GFX9-NEXT:    v_bfrev_b32_e32 v2, 1
-; GFX9-NEXT:    v_addc_co_u32_e32 v1, vcc, v1, v2, vcc
+; GFX9-NEXT:    v_addc_co_u32_e32 v1, vcc, v2, v1, vcc
 ; GFX9-NEXT:    global_load_ubyte v0, v[0:1], off
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
@@ -703,11 +703,9 @@ define amdgpu_kernel void @global_inst_salu_offset_12bit_max(i8 addrspace(1)* %p
 ; GFX10-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX10-NEXT:    ; implicit-def: $vcc_hi
 ; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX10-NEXT:    s_add_u32 s0, s0, 0xfff
-; GFX10-NEXT:    s_addc_u32 s1, s1, 0
-; GFX10-NEXT:    v_mov_b32_e32 v0, s0
-; GFX10-NEXT:    v_mov_b32_e32 v1, s1
-; GFX10-NEXT:    global_load_ubyte v0, v[0:1], off
+; GFX10-NEXT:    v_add_co_u32_e64 v0, s0, 0x800, s0
+; GFX10-NEXT:    v_add_co_ci_u32_e64 v1, s0, 0, s1, s0
+; GFX10-NEXT:    global_load_ubyte v0, v[0:1], off offset:2047
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    global_store_byte v[0:1], v0, off
 ; GFX10-NEXT:    s_endpgm
@@ -722,11 +720,11 @@ define amdgpu_kernel void @global_inst_salu_offset_13bit_max(i8 addrspace(1)* %p
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX9-NEXT:    s_add_u32 s0, s0, 0x1fff
-; GFX9-NEXT:    s_addc_u32 s1, s1, 0
 ; GFX9-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX9-NEXT:    v_mov_b32_e32 v1, s1
-; GFX9-NEXT:    global_load_ubyte v0, v[0:1], off
+; GFX9-NEXT:    v_add_co_u32_e32 v0, vcc, 0x1000, v0
+; GFX9-NEXT:    v_addc_co_u32_e32 v1, vcc, 0, v1, vcc
+; GFX9-NEXT:    global_load_ubyte v0, v[0:1], off offset:4095
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    global_store_byte v[0:1], v0, off
 ; GFX9-NEXT:    s_endpgm
@@ -736,11 +734,9 @@ define amdgpu_kernel void @global_inst_salu_offset_13bit_max(i8 addrspace(1)* %p
 ; GFX10-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX10-NEXT:    ; implicit-def: $vcc_hi
 ; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX10-NEXT:    s_add_u32 s0, s0, 0x1fff
-; GFX10-NEXT:    s_addc_u32 s1, s1, 0
-; GFX10-NEXT:    v_mov_b32_e32 v0, s0
-; GFX10-NEXT:    v_mov_b32_e32 v1, s1
-; GFX10-NEXT:    global_load_ubyte v0, v[0:1], off
+; GFX10-NEXT:    v_add_co_u32_e64 v0, s0, 0x1800, s0
+; GFX10-NEXT:    v_add_co_ci_u32_e64 v1, s0, 0, s1, s0
+; GFX10-NEXT:    global_load_ubyte v0, v[0:1], off offset:2047
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    global_store_byte v[0:1], v0, off
 ; GFX10-NEXT:    s_endpgm
@@ -796,10 +792,8 @@ define amdgpu_kernel void @global_inst_salu_offset_neg_12bit_max(i8 addrspace(1)
 ; GFX10-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX10-NEXT:    ; implicit-def: $vcc_hi
 ; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX10-NEXT:    s_add_u32 s0, s0, 0xfffff000
-; GFX10-NEXT:    s_addc_u32 s1, s1, -1
-; GFX10-NEXT:    v_mov_b32_e32 v0, s0
-; GFX10-NEXT:    v_mov_b32_e32 v1, s1
+; GFX10-NEXT:    v_add_co_u32_e64 v0, s0, 0xfffff000, s0
+; GFX10-NEXT:    v_add_co_ci_u32_e64 v1, s0, -1, s1, s0
 ; GFX10-NEXT:    global_load_ubyte v0, v[0:1], off
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    global_store_byte v[0:1], v0, off
@@ -815,10 +809,10 @@ define amdgpu_kernel void @global_inst_salu_offset_neg_13bit_max(i8 addrspace(1)
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX9-NEXT:    s_add_u32 s0, s0, 0xffffe000
-; GFX9-NEXT:    s_addc_u32 s1, s1, -1
 ; GFX9-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX9-NEXT:    v_mov_b32_e32 v1, s1
+; GFX9-NEXT:    v_add_co_u32_e32 v0, vcc, 0xffffe000, v0
+; GFX9-NEXT:    v_addc_co_u32_e32 v1, vcc, -1, v1, vcc
 ; GFX9-NEXT:    global_load_ubyte v0, v[0:1], off
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    global_store_byte v[0:1], v0, off
@@ -829,10 +823,8 @@ define amdgpu_kernel void @global_inst_salu_offset_neg_13bit_max(i8 addrspace(1)
 ; GFX10-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX10-NEXT:    ; implicit-def: $vcc_hi
 ; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX10-NEXT:    s_add_u32 s0, s0, 0xffffe000
-; GFX10-NEXT:    s_addc_u32 s1, s1, -1
-; GFX10-NEXT:    v_mov_b32_e32 v0, s0
-; GFX10-NEXT:    v_mov_b32_e32 v1, s1
+; GFX10-NEXT:    v_add_co_u32_e64 v0, s0, 0xffffe000, s0
+; GFX10-NEXT:    v_add_co_ci_u32_e64 v1, s0, -1, s1, s0
 ; GFX10-NEXT:    global_load_ubyte v0, v[0:1], off
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    global_store_byte v[0:1], v0, off
@@ -860,11 +852,9 @@ define amdgpu_kernel void @global_inst_salu_offset_2x_11bit_max(i8 addrspace(1)*
 ; GFX10-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX10-NEXT:    ; implicit-def: $vcc_hi
 ; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX10-NEXT:    s_add_u32 s0, s0, 0xfff
-; GFX10-NEXT:    s_addc_u32 s1, s1, 0
-; GFX10-NEXT:    v_mov_b32_e32 v0, s0
-; GFX10-NEXT:    v_mov_b32_e32 v1, s1
-; GFX10-NEXT:    global_load_ubyte v0, v[0:1], off
+; GFX10-NEXT:    v_add_co_u32_e64 v0, s0, 0x800, s0
+; GFX10-NEXT:    v_add_co_ci_u32_e64 v1, s0, 0, s1, s0
+; GFX10-NEXT:    global_load_ubyte v0, v[0:1], off offset:2047
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    global_store_byte v[0:1], v0, off
 ; GFX10-NEXT:    s_endpgm
@@ -879,11 +869,11 @@ define amdgpu_kernel void @global_inst_salu_offset_2x_12bit_max(i8 addrspace(1)*
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX9-NEXT:    s_add_u32 s0, s0, 0x1fff
-; GFX9-NEXT:    s_addc_u32 s1, s1, 0
 ; GFX9-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX9-NEXT:    v_mov_b32_e32 v1, s1
-; GFX9-NEXT:    global_load_ubyte v0, v[0:1], off
+; GFX9-NEXT:    v_add_co_u32_e32 v0, vcc, 0x1000, v0
+; GFX9-NEXT:    v_addc_co_u32_e32 v1, vcc, 0, v1, vcc
+; GFX9-NEXT:    global_load_ubyte v0, v[0:1], off offset:4095
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    global_store_byte v[0:1], v0, off
 ; GFX9-NEXT:    s_endpgm
@@ -893,11 +883,9 @@ define amdgpu_kernel void @global_inst_salu_offset_2x_12bit_max(i8 addrspace(1)*
 ; GFX10-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX10-NEXT:    ; implicit-def: $vcc_hi
 ; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX10-NEXT:    s_add_u32 s0, s0, 0x1fff
-; GFX10-NEXT:    s_addc_u32 s1, s1, 0
-; GFX10-NEXT:    v_mov_b32_e32 v0, s0
-; GFX10-NEXT:    v_mov_b32_e32 v1, s1
-; GFX10-NEXT:    global_load_ubyte v0, v[0:1], off
+; GFX10-NEXT:    v_add_co_u32_e64 v0, s0, 0x1800, s0
+; GFX10-NEXT:    v_add_co_ci_u32_e64 v1, s0, 0, s1, s0
+; GFX10-NEXT:    global_load_ubyte v0, v[0:1], off offset:2047
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    global_store_byte v[0:1], v0, off
 ; GFX10-NEXT:    s_endpgm
@@ -912,11 +900,11 @@ define amdgpu_kernel void @global_inst_salu_offset_2x_13bit_max(i8 addrspace(1)*
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX9-NEXT:    s_add_u32 s0, s0, 0x3fff
-; GFX9-NEXT:    s_addc_u32 s1, s1, 0
 ; GFX9-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX9-NEXT:    v_mov_b32_e32 v1, s1
-; GFX9-NEXT:    global_load_ubyte v0, v[0:1], off
+; GFX9-NEXT:    v_add_co_u32_e32 v0, vcc, 0x3000, v0
+; GFX9-NEXT:    v_addc_co_u32_e32 v1, vcc, 0, v1, vcc
+; GFX9-NEXT:    global_load_ubyte v0, v[0:1], off offset:4095
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    global_store_byte v[0:1], v0, off
 ; GFX9-NEXT:    s_endpgm
@@ -926,11 +914,9 @@ define amdgpu_kernel void @global_inst_salu_offset_2x_13bit_max(i8 addrspace(1)*
 ; GFX10-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX10-NEXT:    ; implicit-def: $vcc_hi
 ; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX10-NEXT:    s_add_u32 s0, s0, 0x3fff
-; GFX10-NEXT:    s_addc_u32 s1, s1, 0
-; GFX10-NEXT:    v_mov_b32_e32 v0, s0
-; GFX10-NEXT:    v_mov_b32_e32 v1, s1
-; GFX10-NEXT:    global_load_ubyte v0, v[0:1], off
+; GFX10-NEXT:    v_add_co_u32_e64 v0, s0, 0x3800, s0
+; GFX10-NEXT:    v_add_co_ci_u32_e64 v1, s0, 0, s1, s0
+; GFX10-NEXT:    global_load_ubyte v0, v[0:1], off offset:2047
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    global_store_byte v[0:1], v0, off
 ; GFX10-NEXT:    s_endpgm
@@ -957,10 +943,8 @@ define amdgpu_kernel void @global_inst_salu_offset_2x_neg_11bit_max(i8 addrspace
 ; GFX10-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX10-NEXT:    ; implicit-def: $vcc_hi
 ; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX10-NEXT:    s_add_u32 s0, s0, 0xfffff000
-; GFX10-NEXT:    s_addc_u32 s1, s1, -1
-; GFX10-NEXT:    v_mov_b32_e32 v0, s0
-; GFX10-NEXT:    v_mov_b32_e32 v1, s1
+; GFX10-NEXT:    v_add_co_u32_e64 v0, s0, 0xfffff000, s0
+; GFX10-NEXT:    v_add_co_ci_u32_e64 v1, s0, -1, s1, s0
 ; GFX10-NEXT:    global_load_ubyte v0, v[0:1], off
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    global_store_byte v[0:1], v0, off
@@ -976,10 +960,10 @@ define amdgpu_kernel void @global_inst_salu_offset_2x_neg_12bit_max(i8 addrspace
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX9-NEXT:    s_add_u32 s0, s0, 0xffffe000
-; GFX9-NEXT:    s_addc_u32 s1, s1, -1
 ; GFX9-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX9-NEXT:    v_mov_b32_e32 v1, s1
+; GFX9-NEXT:    v_add_co_u32_e32 v0, vcc, 0xffffe000, v0
+; GFX9-NEXT:    v_addc_co_u32_e32 v1, vcc, -1, v1, vcc
 ; GFX9-NEXT:    global_load_ubyte v0, v[0:1], off
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    global_store_byte v[0:1], v0, off
@@ -990,10 +974,8 @@ define amdgpu_kernel void @global_inst_salu_offset_2x_neg_12bit_max(i8 addrspace
 ; GFX10-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX10-NEXT:    ; implicit-def: $vcc_hi
 ; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX10-NEXT:    s_add_u32 s0, s0, 0xffffe000
-; GFX10-NEXT:    s_addc_u32 s1, s1, -1
-; GFX10-NEXT:    v_mov_b32_e32 v0, s0
-; GFX10-NEXT:    v_mov_b32_e32 v1, s1
+; GFX10-NEXT:    v_add_co_u32_e64 v0, s0, 0xffffe000, s0
+; GFX10-NEXT:    v_add_co_ci_u32_e64 v1, s0, -1, s1, s0
 ; GFX10-NEXT:    global_load_ubyte v0, v[0:1], off
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    global_store_byte v[0:1], v0, off
@@ -1009,10 +991,10 @@ define amdgpu_kernel void @global_inst_salu_offset_2x_neg_13bit_max(i8 addrspace
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX9-NEXT:    s_add_u32 s0, s0, 0xffffc000
-; GFX9-NEXT:    s_addc_u32 s1, s1, -1
 ; GFX9-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX9-NEXT:    v_mov_b32_e32 v1, s1
+; GFX9-NEXT:    v_add_co_u32_e32 v0, vcc, 0xffffc000, v0
+; GFX9-NEXT:    v_addc_co_u32_e32 v1, vcc, -1, v1, vcc
 ; GFX9-NEXT:    global_load_ubyte v0, v[0:1], off
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    global_store_byte v[0:1], v0, off
@@ -1023,10 +1005,8 @@ define amdgpu_kernel void @global_inst_salu_offset_2x_neg_13bit_max(i8 addrspace
 ; GFX10-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX10-NEXT:    ; implicit-def: $vcc_hi
 ; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX10-NEXT:    s_add_u32 s0, s0, 0xffffc000
-; GFX10-NEXT:    s_addc_u32 s1, s1, -1
-; GFX10-NEXT:    v_mov_b32_e32 v0, s0
-; GFX10-NEXT:    v_mov_b32_e32 v1, s1
+; GFX10-NEXT:    v_add_co_u32_e64 v0, s0, 0xffffc000, s0
+; GFX10-NEXT:    v_add_co_ci_u32_e64 v1, s0, -1, s1, s0
 ; GFX10-NEXT:    global_load_ubyte v0, v[0:1], off
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    global_store_byte v[0:1], v0, off
@@ -1043,11 +1023,10 @@ define amdgpu_kernel void @global_inst_salu_offset_64bit_11bit_split0(i8 addrspa
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX9-NEXT:    s_add_u32 s0, s0, 0x7ff
-; GFX9-NEXT:    s_addc_u32 s1, s1, 2
-; GFX9-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX9-NEXT:    v_mov_b32_e32 v1, s1
-; GFX9-NEXT:    global_load_ubyte v0, v[0:1], off
+; GFX9-NEXT:    v_add_co_u32_e64 v0, vcc, 0, s0
+; GFX9-NEXT:    v_addc_co_u32_e32 v1, vcc, 2, v1, vcc
+; GFX9-NEXT:    global_load_ubyte v0, v[0:1], off offset:2047
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    global_store_byte v[0:1], v0, off
 ; GFX9-NEXT:    s_endpgm
@@ -1057,11 +1036,9 @@ define amdgpu_kernel void @global_inst_salu_offset_64bit_11bit_split0(i8 addrspa
 ; GFX10-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX10-NEXT:    ; implicit-def: $vcc_hi
 ; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX10-NEXT:    s_add_u32 s0, s0, 0x7ff
-; GFX10-NEXT:    s_addc_u32 s1, s1, 2
-; GFX10-NEXT:    v_mov_b32_e32 v0, s0
-; GFX10-NEXT:    v_mov_b32_e32 v1, s1
-; GFX10-NEXT:    global_load_ubyte v0, v[0:1], off
+; GFX10-NEXT:    v_add_co_u32_e64 v0, s0, 0, s0
+; GFX10-NEXT:    v_add_co_ci_u32_e64 v1, s0, 2, s1, s0
+; GFX10-NEXT:    global_load_ubyte v0, v[0:1], off offset:2047
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    global_store_byte v[0:1], v0, off
 ; GFX10-NEXT:    s_endpgm
@@ -1077,11 +1054,10 @@ define amdgpu_kernel void @global_inst_salu_offset_64bit_11bit_split1(i8 addrspa
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX9-NEXT:    s_add_u32 s0, s0, 0x800
-; GFX9-NEXT:    s_addc_u32 s1, s1, 2
-; GFX9-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX9-NEXT:    v_mov_b32_e32 v1, s1
-; GFX9-NEXT:    global_load_ubyte v0, v[0:1], off
+; GFX9-NEXT:    v_add_co_u32_e64 v0, vcc, 0, s0
+; GFX9-NEXT:    v_addc_co_u32_e32 v1, vcc, 2, v1, vcc
+; GFX9-NEXT:    global_load_ubyte v0, v[0:1], off offset:2048
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    global_store_byte v[0:1], v0, off
 ; GFX9-NEXT:    s_endpgm
@@ -1091,10 +1067,8 @@ define amdgpu_kernel void @global_inst_salu_offset_64bit_11bit_split1(i8 addrspa
 ; GFX10-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX10-NEXT:    ; implicit-def: $vcc_hi
 ; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX10-NEXT:    s_add_u32 s0, s0, 0x800
-; GFX10-NEXT:    s_addc_u32 s1, s1, 2
-; GFX10-NEXT:    v_mov_b32_e32 v0, s0
-; GFX10-NEXT:    v_mov_b32_e32 v1, s1
+; GFX10-NEXT:    v_add_co_u32_e64 v0, s0, 0x800, s0
+; GFX10-NEXT:    v_add_co_ci_u32_e64 v1, s0, 2, s1, s0
 ; GFX10-NEXT:    global_load_ubyte v0, v[0:1], off
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    global_store_byte v[0:1], v0, off
@@ -1111,11 +1085,10 @@ define amdgpu_kernel void @global_inst_salu_offset_64bit_12bit_split0(i8 addrspa
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX9-NEXT:    s_add_u32 s0, s0, 0xfff
-; GFX9-NEXT:    s_addc_u32 s1, s1, 2
-; GFX9-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX9-NEXT:    v_mov_b32_e32 v1, s1
-; GFX9-NEXT:    global_load_ubyte v0, v[0:1], off
+; GFX9-NEXT:    v_add_co_u32_e64 v0, vcc, 0, s0
+; GFX9-NEXT:    v_addc_co_u32_e32 v1, vcc, 2, v1, vcc
+; GFX9-NEXT:    global_load_ubyte v0, v[0:1], off offset:4095
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    global_store_byte v[0:1], v0, off
 ; GFX9-NEXT:    s_endpgm
@@ -1125,11 +1098,9 @@ define amdgpu_kernel void @global_inst_salu_offset_64bit_12bit_split0(i8 addrspa
 ; GFX10-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX10-NEXT:    ; implicit-def: $vcc_hi
 ; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX10-NEXT:    s_add_u32 s0, s0, 0xfff
-; GFX10-NEXT:    s_addc_u32 s1, s1, 2
-; GFX10-NEXT:    v_mov_b32_e32 v0, s0
-; GFX10-NEXT:    v_mov_b32_e32 v1, s1
-; GFX10-NEXT:    global_load_ubyte v0, v[0:1], off
+; GFX10-NEXT:    v_add_co_u32_e64 v0, s0, 0x800, s0
+; GFX10-NEXT:    v_add_co_ci_u32_e64 v1, s0, 2, s1, s0
+; GFX10-NEXT:    global_load_ubyte v0, v[0:1], off offset:2047
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    global_store_byte v[0:1], v0, off
 ; GFX10-NEXT:    s_endpgm
@@ -1145,10 +1116,10 @@ define amdgpu_kernel void @global_inst_salu_offset_64bit_12bit_split1(i8 addrspa
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX9-NEXT:    s_add_u32 s0, s0, 0x1000
-; GFX9-NEXT:    s_addc_u32 s1, s1, 2
 ; GFX9-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX9-NEXT:    v_mov_b32_e32 v1, s1
+; GFX9-NEXT:    v_add_co_u32_e32 v0, vcc, 0x1000, v0
+; GFX9-NEXT:    v_addc_co_u32_e32 v1, vcc, 2, v1, vcc
 ; GFX9-NEXT:    global_load_ubyte v0, v[0:1], off
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    global_store_byte v[0:1], v0, off
@@ -1159,10 +1130,8 @@ define amdgpu_kernel void @global_inst_salu_offset_64bit_12bit_split1(i8 addrspa
 ; GFX10-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX10-NEXT:    ; implicit-def: $vcc_hi
 ; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX10-NEXT:    s_add_u32 s0, s0, 0x1000
-; GFX10-NEXT:    s_addc_u32 s1, s1, 2
-; GFX10-NEXT:    v_mov_b32_e32 v0, s0
-; GFX10-NEXT:    v_mov_b32_e32 v1, s1
+; GFX10-NEXT:    v_add_co_u32_e64 v0, s0, 0x1000, s0
+; GFX10-NEXT:    v_add_co_ci_u32_e64 v1, s0, 2, s1, s0
 ; GFX10-NEXT:    global_load_ubyte v0, v[0:1], off
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    global_store_byte v[0:1], v0, off
@@ -1179,11 +1148,11 @@ define amdgpu_kernel void @global_inst_salu_offset_64bit_13bit_split0(i8 addrspa
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX9-NEXT:    s_add_u32 s0, s0, 0x1fff
-; GFX9-NEXT:    s_addc_u32 s1, s1, 2
 ; GFX9-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX9-NEXT:    v_mov_b32_e32 v1, s1
-; GFX9-NEXT:    global_load_ubyte v0, v[0:1], off
+; GFX9-NEXT:    v_add_co_u32_e32 v0, vcc, 0x1000, v0
+; GFX9-NEXT:    v_addc_co_u32_e32 v1, vcc, 2, v1, vcc
+; GFX9-NEXT:    global_load_ubyte v0, v[0:1], off offset:4095
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    global_store_byte v[0:1], v0, off
 ; GFX9-NEXT:    s_endpgm
@@ -1193,11 +1162,9 @@ define amdgpu_kernel void @global_inst_salu_offset_64bit_13bit_split0(i8 addrspa
 ; GFX10-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX10-NEXT:    ; implicit-def: $vcc_hi
 ; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX10-NEXT:    s_add_u32 s0, s0, 0x1fff
-; GFX10-NEXT:    s_addc_u32 s1, s1, 2
-; GFX10-NEXT:    v_mov_b32_e32 v0, s0
-; GFX10-NEXT:    v_mov_b32_e32 v1, s1
-; GFX10-NEXT:    global_load_ubyte v0, v[0:1], off
+; GFX10-NEXT:    v_add_co_u32_e64 v0, s0, 0x1800, s0
+; GFX10-NEXT:    v_add_co_ci_u32_e64 v1, s0, 2, s1, s0
+; GFX10-NEXT:    global_load_ubyte v0, v[0:1], off offset:2047
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    global_store_byte v[0:1], v0, off
 ; GFX10-NEXT:    s_endpgm
@@ -1213,10 +1180,10 @@ define amdgpu_kernel void @global_inst_salu_offset_64bit_13bit_split1(i8 addrspa
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX9-NEXT:    s_add_u32 s0, s0, 0x2000
-; GFX9-NEXT:    s_addc_u32 s1, s1, 2
 ; GFX9-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX9-NEXT:    v_mov_b32_e32 v1, s1
+; GFX9-NEXT:    v_add_co_u32_e32 v0, vcc, 0x2000, v0
+; GFX9-NEXT:    v_addc_co_u32_e32 v1, vcc, 2, v1, vcc
 ; GFX9-NEXT:    global_load_ubyte v0, v[0:1], off
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    global_store_byte v[0:1], v0, off
@@ -1227,10 +1194,8 @@ define amdgpu_kernel void @global_inst_salu_offset_64bit_13bit_split1(i8 addrspa
 ; GFX10-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX10-NEXT:    ; implicit-def: $vcc_hi
 ; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX10-NEXT:    s_add_u32 s0, s0, 0x2000
-; GFX10-NEXT:    s_addc_u32 s1, s1, 2
-; GFX10-NEXT:    v_mov_b32_e32 v0, s0
-; GFX10-NEXT:    v_mov_b32_e32 v1, s1
+; GFX10-NEXT:    v_add_co_u32_e64 v0, s0, 0x2000, s0
+; GFX10-NEXT:    v_add_co_ci_u32_e64 v1, s0, 2, s1, s0
 ; GFX10-NEXT:    global_load_ubyte v0, v[0:1], off
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    global_store_byte v[0:1], v0, off
@@ -1246,12 +1211,12 @@ define amdgpu_kernel void @global_inst_salu_offset_64bit_11bit_neg_high_split0(i
 ; GFX9-LABEL: global_inst_salu_offset_64bit_11bit_neg_high_split0:
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GFX9-NEXT:    v_bfrev_b32_e32 v1, 1
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX9-NEXT:    s_add_u32 s0, s0, 0x7ff
-; GFX9-NEXT:    s_addc_u32 s1, s1, 0x80000000
-; GFX9-NEXT:    v_mov_b32_e32 v0, s0
-; GFX9-NEXT:    v_mov_b32_e32 v1, s1
-; GFX9-NEXT:    global_load_ubyte v0, v[0:1], off
+; GFX9-NEXT:    v_mov_b32_e32 v2, s1
+; GFX9-NEXT:    v_add_co_u32_e64 v0, vcc, 0, s0
+; GFX9-NEXT:    v_addc_co_u32_e32 v1, vcc, v1, v2, vcc
+; GFX9-NEXT:    global_load_ubyte v0, v[0:1], off offset:2047
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    global_store_byte v[0:1], v0, off
 ; GFX9-NEXT:    s_endpgm
@@ -1261,11 +1226,10 @@ define amdgpu_kernel void @global_inst_salu_offset_64bit_11bit_neg_high_split0(i
 ; GFX10-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX10-NEXT:    ; implicit-def: $vcc_hi
 ; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX10-NEXT:    s_add_u32 s0, s0, 0x7ff
-; GFX10-NEXT:    s_addc_u32 s1, s1, 0x80000000
-; GFX10-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-NEXT:    v_mov_b32_e32 v1, s1
-; GFX10-NEXT:    global_load_ubyte v0, v[0:1], off
+; GFX10-NEXT:    v_add_co_u32_e64 v0, vcc_lo, 0, s0
+; GFX10-NEXT:    v_add_co_ci_u32_e32 v1, vcc_lo, 0x80000000, v1, vcc_lo
+; GFX10-NEXT:    global_load_ubyte v0, v[0:1], off offset:2047
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    global_store_byte v[0:1], v0, off
 ; GFX10-NEXT:    s_endpgm
@@ -1280,12 +1244,12 @@ define amdgpu_kernel void @global_inst_salu_offset_64bit_11bit_neg_high_split1(i
 ; GFX9-LABEL: global_inst_salu_offset_64bit_11bit_neg_high_split1:
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GFX9-NEXT:    v_bfrev_b32_e32 v1, 1
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX9-NEXT:    s_add_u32 s0, s0, 0x800
-; GFX9-NEXT:    s_addc_u32 s1, s1, 0x80000000
-; GFX9-NEXT:    v_mov_b32_e32 v0, s0
-; GFX9-NEXT:    v_mov_b32_e32 v1, s1
-; GFX9-NEXT:    global_load_ubyte v0, v[0:1], off
+; GFX9-NEXT:    v_mov_b32_e32 v2, s1
+; GFX9-NEXT:    v_add_co_u32_e64 v0, vcc, 0, s0
+; GFX9-NEXT:    v_addc_co_u32_e32 v1, vcc, v1, v2, vcc
+; GFX9-NEXT:    global_load_ubyte v0, v[0:1], off offset:2048
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    global_store_byte v[0:1], v0, off
 ; GFX9-NEXT:    s_endpgm
@@ -1295,11 +1259,10 @@ define amdgpu_kernel void @global_inst_salu_offset_64bit_11bit_neg_high_split1(i
 ; GFX10-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX10-NEXT:    ; implicit-def: $vcc_hi
 ; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX10-NEXT:    s_add_u32 s0, s0, 0x800
-; GFX10-NEXT:    s_addc_u32 s1, s1, 0x80000000
-; GFX10-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-NEXT:    v_mov_b32_e32 v1, s1
-; GFX10-NEXT:    global_load_ubyte v0, v[0:1], off
+; GFX10-NEXT:    v_add_co_u32_e64 v0, vcc_lo, 0x1000, s0
+; GFX10-NEXT:    v_add_co_ci_u32_e32 v1, vcc_lo, 0x80000000, v1, vcc_lo
+; GFX10-NEXT:    global_load_ubyte v0, v[0:1], off offset:-2048
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    global_store_byte v[0:1], v0, off
 ; GFX10-NEXT:    s_endpgm
@@ -1314,12 +1277,12 @@ define amdgpu_kernel void @global_inst_salu_offset_64bit_12bit_neg_high_split0(i
 ; GFX9-LABEL: global_inst_salu_offset_64bit_12bit_neg_high_split0:
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GFX9-NEXT:    v_bfrev_b32_e32 v1, 1
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX9-NEXT:    s_add_u32 s0, s0, 0xfff
-; GFX9-NEXT:    s_addc_u32 s1, s1, 0x80000000
-; GFX9-NEXT:    v_mov_b32_e32 v0, s0
-; GFX9-NEXT:    v_mov_b32_e32 v1, s1
-; GFX9-NEXT:    global_load_ubyte v0, v[0:1], off
+; GFX9-NEXT:    v_mov_b32_e32 v2, s1
+; GFX9-NEXT:    v_add_co_u32_e64 v0, vcc, 0, s0
+; GFX9-NEXT:    v_addc_co_u32_e32 v1, vcc, v1, v2, vcc
+; GFX9-NEXT:    global_load_ubyte v0, v[0:1], off offset:4095
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    global_store_byte v[0:1], v0, off
 ; GFX9-NEXT:    s_endpgm
@@ -1329,11 +1292,10 @@ define amdgpu_kernel void @global_inst_salu_offset_64bit_12bit_neg_high_split0(i
 ; GFX10-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX10-NEXT:    ; implicit-def: $vcc_hi
 ; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX10-NEXT:    s_add_u32 s0, s0, 0xfff
-; GFX10-NEXT:    s_addc_u32 s1, s1, 0x80000000
-; GFX10-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-NEXT:    v_mov_b32_e32 v1, s1
-; GFX10-NEXT:    global_load_ubyte v0, v[0:1], off
+; GFX10-NEXT:    v_add_co_u32_e64 v0, vcc_lo, 0x1000, s0
+; GFX10-NEXT:    v_add_co_ci_u32_e32 v1, vcc_lo, 0x80000000, v1, vcc_lo
+; GFX10-NEXT:    global_load_ubyte v0, v[0:1], off offset:-1
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    global_store_byte v[0:1], v0, off
 ; GFX10-NEXT:    s_endpgm
@@ -1348,12 +1310,13 @@ define amdgpu_kernel void @global_inst_salu_offset_64bit_12bit_neg_high_split1(i
 ; GFX9-LABEL: global_inst_salu_offset_64bit_12bit_neg_high_split1:
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GFX9-NEXT:    v_bfrev_b32_e32 v1, 1
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX9-NEXT:    s_add_u32 s0, s0, 0x1000
-; GFX9-NEXT:    s_addc_u32 s1, s1, 0x80000000
 ; GFX9-NEXT:    v_mov_b32_e32 v0, s0
-; GFX9-NEXT:    v_mov_b32_e32 v1, s1
-; GFX9-NEXT:    global_load_ubyte v0, v[0:1], off
+; GFX9-NEXT:    v_mov_b32_e32 v2, s1
+; GFX9-NEXT:    v_add_co_u32_e32 v0, vcc, 0x2000, v0
+; GFX9-NEXT:    v_addc_co_u32_e32 v1, vcc, v1, v2, vcc
+; GFX9-NEXT:    global_load_ubyte v0, v[0:1], off offset:-4096
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    global_store_byte v[0:1], v0, off
 ; GFX9-NEXT:    s_endpgm
@@ -1363,10 +1326,9 @@ define amdgpu_kernel void @global_inst_salu_offset_64bit_12bit_neg_high_split1(i
 ; GFX10-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX10-NEXT:    ; implicit-def: $vcc_hi
 ; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX10-NEXT:    s_add_u32 s0, s0, 0x1000
-; GFX10-NEXT:    s_addc_u32 s1, s1, 0x80000000
-; GFX10-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-NEXT:    v_mov_b32_e32 v1, s1
+; GFX10-NEXT:    v_add_co_u32_e64 v0, vcc_lo, 0x1000, s0
+; GFX10-NEXT:    v_add_co_ci_u32_e32 v1, vcc_lo, 0x80000000, v1, vcc_lo
 ; GFX10-NEXT:    global_load_ubyte v0, v[0:1], off
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    global_store_byte v[0:1], v0, off
@@ -1382,12 +1344,13 @@ define amdgpu_kernel void @global_inst_salu_offset_64bit_13bit_neg_high_split0(i
 ; GFX9-LABEL: global_inst_salu_offset_64bit_13bit_neg_high_split0:
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GFX9-NEXT:    v_bfrev_b32_e32 v1, 1
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX9-NEXT:    s_add_u32 s0, s0, 0x1fff
-; GFX9-NEXT:    s_addc_u32 s1, s1, 0x80000000
 ; GFX9-NEXT:    v_mov_b32_e32 v0, s0
-; GFX9-NEXT:    v_mov_b32_e32 v1, s1
-; GFX9-NEXT:    global_load_ubyte v0, v[0:1], off
+; GFX9-NEXT:    v_mov_b32_e32 v2, s1
+; GFX9-NEXT:    v_add_co_u32_e32 v0, vcc, 0x2000, v0
+; GFX9-NEXT:    v_addc_co_u32_e32 v1, vcc, v1, v2, vcc
+; GFX9-NEXT:    global_load_ubyte v0, v[0:1], off offset:-1
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    global_store_byte v[0:1], v0, off
 ; GFX9-NEXT:    s_endpgm
@@ -1397,11 +1360,10 @@ define amdgpu_kernel void @global_inst_salu_offset_64bit_13bit_neg_high_split0(i
 ; GFX10-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX10-NEXT:    ; implicit-def: $vcc_hi
 ; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX10-NEXT:    s_add_u32 s0, s0, 0x1fff
-; GFX10-NEXT:    s_addc_u32 s1, s1, 0x80000000
-; GFX10-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-NEXT:    v_mov_b32_e32 v1, s1
-; GFX10-NEXT:    global_load_ubyte v0, v[0:1], off
+; GFX10-NEXT:    v_add_co_u32_e64 v0, vcc_lo, 0x2000, s0
+; GFX10-NEXT:    v_add_co_ci_u32_e32 v1, vcc_lo, 0x80000000, v1, vcc_lo
+; GFX10-NEXT:    global_load_ubyte v0, v[0:1], off offset:-1
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    global_store_byte v[0:1], v0, off
 ; GFX10-NEXT:    s_endpgm
@@ -1416,11 +1378,12 @@ define amdgpu_kernel void @global_inst_salu_offset_64bit_13bit_neg_high_split1(i
 ; GFX9-LABEL: global_inst_salu_offset_64bit_13bit_neg_high_split1:
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GFX9-NEXT:    v_bfrev_b32_e32 v1, 1
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX9-NEXT:    s_add_u32 s0, s0, 0x2000
-; GFX9-NEXT:    s_addc_u32 s1, s1, 0x80000000
 ; GFX9-NEXT:    v_mov_b32_e32 v0, s0
-; GFX9-NEXT:    v_mov_b32_e32 v1, s1
+; GFX9-NEXT:    v_mov_b32_e32 v2, s1
+; GFX9-NEXT:    v_add_co_u32_e32 v0, vcc, 0x2000, v0
+; GFX9-NEXT:    v_addc_co_u32_e32 v1, vcc, v1, v2, vcc
 ; GFX9-NEXT:    global_load_ubyte v0, v[0:1], off
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    global_store_byte v[0:1], v0, off
@@ -1431,10 +1394,9 @@ define amdgpu_kernel void @global_inst_salu_offset_64bit_13bit_neg_high_split1(i
 ; GFX10-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX10-NEXT:    ; implicit-def: $vcc_hi
 ; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX10-NEXT:    s_add_u32 s0, s0, 0x2000
-; GFX10-NEXT:    s_addc_u32 s1, s1, 0x80000000
-; GFX10-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-NEXT:    v_mov_b32_e32 v1, s1
+; GFX10-NEXT:    v_add_co_u32_e64 v0, vcc_lo, 0x2000, s0
+; GFX10-NEXT:    v_add_co_ci_u32_e32 v1, vcc_lo, 0x80000000, v1, vcc_lo
 ; GFX10-NEXT:    global_load_ubyte v0, v[0:1], off
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    global_store_byte v[0:1], v0, off