; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx900 < %s | FileCheck -check-prefixes=GCN,GFX9 %s
; RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1010 -mattr=+wavefrontsize64 < %s | FileCheck -check-prefixes=GCN,GFX10 %s

; Test using saddr addressing mode of global_*load_* flat instructions.

; --------------------------------------------------------------------------------
; No vgpr offset, constants
; --------------------------------------------------------------------------------

; SGPR base only
define amdgpu_ps float @global_load_saddr_i8_offset_0(i8 addrspace(1)* inreg %sbase) {
; GCN-LABEL: global_load_saddr_i8_offset_0:
; GCN:       ; %bb.0:
; GCN-NEXT:    v_mov_b32_e32 v0, 0
; GCN-NEXT:    global_load_ubyte v0, v0, s[2:3]
; GCN-NEXT:    s_waitcnt vmcnt(0)
; GCN-NEXT:    ; return to shader part epilog
  %load = load i8, i8 addrspace(1)* %sbase
  %zext = zext i8 %load to i32
  %to.vgpr = bitcast i32 %zext to float
  ret float %to.vgpr
}

; SGPR base with maximum gfx9 immediate offset
define amdgpu_ps float @global_load_saddr_i8_offset_4095(i8 addrspace(1)* inreg %sbase) {
; GFX9-LABEL: global_load_saddr_i8_offset_4095:
; GFX9:       ; %bb.0:
; GFX9-NEXT:    v_mov_b32_e32 v0, 0
; GFX9-NEXT:    global_load_ubyte v0, v0, s[2:3] offset:4095
; GFX9-NEXT:    s_waitcnt vmcnt(0)
; GFX9-NEXT:    ; return to shader part epilog
;
; GFX10-LABEL: global_load_saddr_i8_offset_4095:
; GFX10:       ; %bb.0:
; GFX10-NEXT:    v_mov_b32_e32 v0, 0x800
; GFX10-NEXT:    global_load_ubyte v0, v0, s[2:3] offset:2047
; GFX10-NEXT:    s_waitcnt vmcnt(0)
; GFX10-NEXT:    ; return to shader part epilog
  %gep0 = getelementptr inbounds i8, i8 addrspace(1)* %sbase, i64 4095
  %load = load i8, i8 addrspace(1)* %gep0
  %zext = zext i8 %load to i32
  %to.vgpr = bitcast i32 %zext to float
  ret float %to.vgpr
}

; SGPR base with maximum gfx9 immediate offset + 1
define amdgpu_ps float @global_load_saddr_i8_offset_4096(i8 addrspace(1)* inreg %sbase) {
; GCN-LABEL: global_load_saddr_i8_offset_4096:
; GCN:       ; %bb.0:
; GCN-NEXT:    v_mov_b32_e32 v0, 0x1000
; GCN-NEXT:    global_load_ubyte v0, v0, s[2:3]
; GCN-NEXT:    s_waitcnt vmcnt(0)
; GCN-NEXT:    ; return to shader part epilog
  %gep0 = getelementptr inbounds i8, i8 addrspace(1)* %sbase, i64 4096
  %load = load i8, i8 addrspace(1)* %gep0
  %zext = zext i8 %load to i32
  %to.vgpr = bitcast i32 %zext to float
  ret float %to.vgpr
}

; SGPR base with maximum gfx9 immediate offset + 2
define amdgpu_ps float @global_load_saddr_i8_offset_4097(i8 addrspace(1)* inreg %sbase) {
; GCN-LABEL: global_load_saddr_i8_offset_4097:
; GCN:       ; %bb.0:
; GCN-NEXT:    v_mov_b32_e32 v0, 0x1000
; GCN-NEXT:    global_load_ubyte v0, v0, s[2:3] offset:1
; GCN-NEXT:    s_waitcnt vmcnt(0)
; GCN-NEXT:    ; return to shader part epilog
  %gep0 = getelementptr inbounds i8, i8 addrspace(1)* %sbase, i64 4097
  %load = load i8, i8 addrspace(1)* %gep0
  %zext = zext i8 %load to i32
  %to.vgpr = bitcast i32 %zext to float
  ret float %to.vgpr
}

; SGPR base with maximum negative gfx9 immediate offset
define amdgpu_ps float @global_load_saddr_i8_offset_neg4096(i8 addrspace(1)* inreg %sbase) {
; GFX9-LABEL: global_load_saddr_i8_offset_neg4096:
; GFX9:       ; %bb.0:
; GFX9-NEXT:    v_mov_b32_e32 v0, 0
; GFX9-NEXT:    global_load_ubyte v0, v0, s[2:3] offset:-4096
; GFX9-NEXT:    s_waitcnt vmcnt(0)
; GFX9-NEXT:    ; return to shader part epilog
;
; GFX10-LABEL: global_load_saddr_i8_offset_neg4096:
; GFX10:       ; %bb.0:
; GFX10-NEXT:    v_add_co_u32 v0, s[0:1], 0xfffff000, s2
; GFX10-NEXT:    v_add_co_ci_u32_e64 v1, s[0:1], -1, s3, s[0:1]
; GFX10-NEXT:    global_load_ubyte v0, v[0:1], off
; GFX10-NEXT:    s_waitcnt vmcnt(0)
; GFX10-NEXT:    ; return to shader part epilog
  %gep0 = getelementptr inbounds i8, i8 addrspace(1)* %sbase, i64 -4096
  %load = load i8, i8 addrspace(1)* %gep0
  %zext = zext i8 %load to i32
  %to.vgpr = bitcast i32 %zext to float
  ret float %to.vgpr
}

; SGPR base with maximum negative gfx9 immediate offset -1
define amdgpu_ps float @global_load_saddr_i8_offset_neg4097(i8 addrspace(1)* inreg %sbase) {
; GFX9-LABEL: global_load_saddr_i8_offset_neg4097:
; GFX9:       ; %bb.0:
; GFX9-NEXT:    s_add_u32 s0, s2, 0xffffefff
; GFX9-NEXT:    s_addc_u32 s1, s3, -1
; GFX9-NEXT:    v_mov_b32_e32 v0, 0
; GFX9-NEXT:    global_load_ubyte v0, v0, s[0:1]
; GFX9-NEXT:    s_waitcnt vmcnt(0)
; GFX9-NEXT:    ; return to shader part epilog
;
; GFX10-LABEL: global_load_saddr_i8_offset_neg4097:
; GFX10:       ; %bb.0:
; GFX10-NEXT:    v_add_co_u32 v0, s[0:1], 0xfffff000, s2
; GFX10-NEXT:    v_add_co_ci_u32_e64 v1, s[0:1], -1, s3, s[0:1]
; GFX10-NEXT:    global_load_ubyte v0, v[0:1], off offset:-1
; GFX10-NEXT:    s_waitcnt vmcnt(0)
; GFX10-NEXT:    ; return to shader part epilog
  %gep0 = getelementptr inbounds i8, i8 addrspace(1)* %sbase, i64 -4097
  %load = load i8, i8 addrspace(1)* %gep0
  %zext = zext i8 %load to i32
  %to.vgpr = bitcast i32 %zext to float
  ret float %to.vgpr
}

; SGPR base with maximum negative gfx9 immediate offset -2
define amdgpu_ps float @global_load_saddr_i8_offset_neg4098(i8 addrspace(1)* inreg %sbase) {
; GFX9-LABEL: global_load_saddr_i8_offset_neg4098:
; GFX9:       ; %bb.0:
; GFX9-NEXT:    s_add_u32 s0, s2, 0xffffeffe
; GFX9-NEXT:    s_addc_u32 s1, s3, -1
; GFX9-NEXT:    v_mov_b32_e32 v0, 0
; GFX9-NEXT:    global_load_ubyte v0, v0, s[0:1]
; GFX9-NEXT:    s_waitcnt vmcnt(0)
; GFX9-NEXT:    ; return to shader part epilog
;
; GFX10-LABEL: global_load_saddr_i8_offset_neg4098:
; GFX10:       ; %bb.0:
; GFX10-NEXT:    v_add_co_u32 v0, s[0:1], 0xfffff000, s2
; GFX10-NEXT:    v_add_co_ci_u32_e64 v1, s[0:1], -1, s3, s[0:1]
; GFX10-NEXT:    global_load_ubyte v0, v[0:1], off offset:-2
; GFX10-NEXT:    s_waitcnt vmcnt(0)
; GFX10-NEXT:    ; return to shader part epilog
  %gep0 = getelementptr inbounds i8, i8 addrspace(1)* %sbase, i64 -4098
  %load = load i8, i8 addrspace(1)* %gep0
  %zext = zext i8 %load to i32
  %to.vgpr = bitcast i32 %zext to float
  ret float %to.vgpr
}

; SGPR base with maximum gfx10 immediate offset
define amdgpu_ps float @global_load_saddr_i8_offset_2048(i8 addrspace(1)* inreg %sbase) {
; GFX9-LABEL: global_load_saddr_i8_offset_2048:
; GFX9:       ; %bb.0:
; GFX9-NEXT:    v_mov_b32_e32 v0, 0
; GFX9-NEXT:    global_load_ubyte v0, v0, s[2:3] offset:2048
; GFX9-NEXT:    s_waitcnt vmcnt(0)
; GFX9-NEXT:    ; return to shader part epilog
;
; GFX10-LABEL: global_load_saddr_i8_offset_2048:
; GFX10:       ; %bb.0:
; GFX10-NEXT:    v_mov_b32_e32 v0, 0x800
; GFX10-NEXT:    global_load_ubyte v0, v0, s[2:3]
; GFX10-NEXT:    s_waitcnt vmcnt(0)
; GFX10-NEXT:    ; return to shader part epilog
  %gep0 = getelementptr inbounds i8, i8 addrspace(1)* %sbase, i64 2048
  %load = load i8, i8 addrspace(1)* %gep0
  %zext = zext i8 %load to i32
  %to.vgpr = bitcast i32 %zext to float
  ret float %to.vgpr
}

; SGPR base with maximum gfx10 immediate offset + 1
define amdgpu_ps float @global_load_saddr_i8_offset_2049(i8 addrspace(1)* inreg %sbase) {
; GFX9-LABEL: global_load_saddr_i8_offset_2049:
; GFX9:       ; %bb.0:
; GFX9-NEXT:    v_mov_b32_e32 v0, 0
; GFX9-NEXT:    global_load_ubyte v0, v0, s[2:3] offset:2049
; GFX9-NEXT:    s_waitcnt vmcnt(0)
; GFX9-NEXT:    ; return to shader part epilog
;
; GFX10-LABEL: global_load_saddr_i8_offset_2049:
; GFX10:       ; %bb.0:
; GFX10-NEXT:    v_mov_b32_e32 v0, 0x800
; GFX10-NEXT:    global_load_ubyte v0, v0, s[2:3] offset:1
; GFX10-NEXT:    s_waitcnt vmcnt(0)
; GFX10-NEXT:    ; return to shader part epilog
  %gep0 = getelementptr inbounds i8, i8 addrspace(1)* %sbase, i64 2049
  %load = load i8, i8 addrspace(1)* %gep0
  %zext = zext i8 %load to i32
  %to.vgpr = bitcast i32 %zext to float
  ret float %to.vgpr
}

; SGPR base with maximum gfx10 immediate offset + 2
define amdgpu_ps float @global_load_saddr_i8_offset_2050(i8 addrspace(1)* inreg %sbase) {
; GFX9-LABEL: global_load_saddr_i8_offset_2050:
; GFX9:       ; %bb.0:
; GFX9-NEXT:    v_mov_b32_e32 v0, 0
; GFX9-NEXT:    global_load_ubyte v0, v0, s[2:3] offset:2050
; GFX9-NEXT:    s_waitcnt vmcnt(0)
; GFX9-NEXT:    ; return to shader part epilog
;
; GFX10-LABEL: global_load_saddr_i8_offset_2050:
; GFX10:       ; %bb.0:
; GFX10-NEXT:    v_mov_b32_e32 v0, 0x800
; GFX10-NEXT:    global_load_ubyte v0, v0, s[2:3] offset:2
; GFX10-NEXT:    s_waitcnt vmcnt(0)
; GFX10-NEXT:    ; return to shader part epilog
  %gep0 = getelementptr inbounds i8, i8 addrspace(1)* %sbase, i64 2050
  %load = load i8, i8 addrspace(1)* %gep0
  %zext = zext i8 %load to i32
  %to.vgpr = bitcast i32 %zext to float
  ret float %to.vgpr
}

; SGPR base with maximum negative gfx10 immediate offset
define amdgpu_ps float @global_load_saddr_i8_offset_neg2048(i8 addrspace(1)* inreg %sbase) {
; GCN-LABEL: global_load_saddr_i8_offset_neg2048:
; GCN:       ; %bb.0:
; GCN-NEXT:    v_mov_b32_e32 v0, 0
; GCN-NEXT:    global_load_ubyte v0, v0, s[2:3] offset:-2048
; GCN-NEXT:    s_waitcnt vmcnt(0)
; GCN-NEXT:    ; return to shader part epilog
  %gep0 = getelementptr inbounds i8, i8 addrspace(1)* %sbase, i64 -2048
  %load = load i8, i8 addrspace(1)* %gep0
  %zext = zext i8 %load to i32
  %to.vgpr = bitcast i32 %zext to float
  ret float %to.vgpr
}

; SGPR base with maximum negative gfx10 immediate offset - 1
define amdgpu_ps float @global_load_saddr_i8_offset_neg2049(i8 addrspace(1)* inreg %sbase) {
; GFX9-LABEL: global_load_saddr_i8_offset_neg2049:
; GFX9:       ; %bb.0:
; GFX9-NEXT:    v_mov_b32_e32 v0, 0
; GFX9-NEXT:    global_load_ubyte v0, v0, s[2:3] offset:-2049
; GFX9-NEXT:    s_waitcnt vmcnt(0)
; GFX9-NEXT:    ; return to shader part epilog
;
; GFX10-LABEL: global_load_saddr_i8_offset_neg2049:
; GFX10:       ; %bb.0:
; GFX10-NEXT:    v_add_co_u32 v0, s[0:1], 0xfffff800, s2
; GFX10-NEXT:    v_add_co_ci_u32_e64 v1, s[0:1], -1, s3, s[0:1]
; GFX10-NEXT:    global_load_ubyte v0, v[0:1], off offset:-1
; GFX10-NEXT:    s_waitcnt vmcnt(0)
; GFX10-NEXT:    ; return to shader part epilog
  %gep0 = getelementptr inbounds i8, i8 addrspace(1)* %sbase, i64 -2049
  %load = load i8, i8 addrspace(1)* %gep0
  %zext = zext i8 %load to i32
  %to.vgpr = bitcast i32 %zext to float
  ret float %to.vgpr
}

; SGPR base with maximum negative gfx10 immediate offset - 1
define amdgpu_ps float @global_load_saddr_i8_offset_neg2050(i8 addrspace(1)* inreg %sbase) {
; GFX9-LABEL: global_load_saddr_i8_offset_neg2050:
; GFX9:       ; %bb.0:
; GFX9-NEXT:    v_mov_b32_e32 v0, 0
; GFX9-NEXT:    global_load_ubyte v0, v0, s[2:3] offset:-2050
; GFX9-NEXT:    s_waitcnt vmcnt(0)
; GFX9-NEXT:    ; return to shader part epilog
;
; GFX10-LABEL: global_load_saddr_i8_offset_neg2050:
; GFX10:       ; %bb.0:
; GFX10-NEXT:    v_add_co_u32 v0, s[0:1], 0xfffff800, s2
; GFX10-NEXT:    v_add_co_ci_u32_e64 v1, s[0:1], -1, s3, s[0:1]
; GFX10-NEXT:    global_load_ubyte v0, v[0:1], off offset:-2
; GFX10-NEXT:    s_waitcnt vmcnt(0)
; GFX10-NEXT:    ; return to shader part epilog
  %gep0 = getelementptr inbounds i8, i8 addrspace(1)* %sbase, i64 -2050
  %load = load i8, i8 addrspace(1)* %gep0
  %zext = zext i8 %load to i32
  %to.vgpr = bitcast i32 %zext to float
  ret float %to.vgpr
}

define amdgpu_ps float @global_load_saddr_i8_offset_4294967295(i8 addrspace(1)* inreg %sbase) {
; GFX9-LABEL: global_load_saddr_i8_offset_4294967295:
; GFX9:       ; %bb.0:
; GFX9-NEXT:    v_mov_b32_e32 v0, 0xfffff000
; GFX9-NEXT:    global_load_ubyte v0, v0, s[2:3] offset:4095
; GFX9-NEXT:    s_waitcnt vmcnt(0)
; GFX9-NEXT:    ; return to shader part epilog
;
; GFX10-LABEL: global_load_saddr_i8_offset_4294967295:
; GFX10:       ; %bb.0:
; GFX10-NEXT:    v_mov_b32_e32 v0, 0xfffff800
; GFX10-NEXT:    global_load_ubyte v0, v0, s[2:3] offset:2047
; GFX10-NEXT:    s_waitcnt vmcnt(0)
; GFX10-NEXT:    ; return to shader part epilog
  %gep0 = getelementptr inbounds i8, i8 addrspace(1)* %sbase, i64 4294967295
  %load = load i8, i8 addrspace(1)* %gep0
  %zext = zext i8 %load to i32
  %to.vgpr = bitcast i32 %zext to float
  ret float %to.vgpr
}

define amdgpu_ps float @global_load_saddr_i8_offset_4294967296(i8 addrspace(1)* inreg %sbase) {
; GFX9-LABEL: global_load_saddr_i8_offset_4294967296:
; GFX9:       ; %bb.0:
; GFX9-NEXT:    v_mov_b32_e32 v1, s3
; GFX9-NEXT:    v_add_co_u32_e64 v0, vcc, 0, s2
; 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:    ; return to shader part epilog
;
; GFX10-LABEL: global_load_saddr_i8_offset_4294967296:
; GFX10:       ; %bb.0:
; GFX10-NEXT:    v_add_co_u32 v0, s[0:1], 0, s2
; GFX10-NEXT:    v_add_co_ci_u32_e64 v1, s[0:1], 1, s3, s[0:1]
; GFX10-NEXT:    global_load_ubyte v0, v[0:1], off
; GFX10-NEXT:    s_waitcnt vmcnt(0)
; GFX10-NEXT:    ; return to shader part epilog
  %gep0 = getelementptr inbounds i8, i8 addrspace(1)* %sbase, i64 4294967296
  %load = load i8, i8 addrspace(1)* %gep0
  %zext = zext i8 %load to i32
  %to.vgpr = bitcast i32 %zext to float
  ret float %to.vgpr
}

define amdgpu_ps float @global_load_saddr_i8_offset_4294967297(i8 addrspace(1)* inreg %sbase) {
; GFX9-LABEL: global_load_saddr_i8_offset_4294967297:
; GFX9:       ; %bb.0:
; GFX9-NEXT:    v_mov_b32_e32 v1, s3
; GFX9-NEXT:    v_add_co_u32_e64 v0, vcc, 0, s2
; GFX9-NEXT:    v_addc_co_u32_e32 v1, vcc, 1, v1, vcc
; GFX9-NEXT:    global_load_ubyte v0, v[0:1], off offset:1
; GFX9-NEXT:    s_waitcnt vmcnt(0)
; GFX9-NEXT:    ; return to shader part epilog
;
; GFX10-LABEL: global_load_saddr_i8_offset_4294967297:
; GFX10:       ; %bb.0:
; GFX10-NEXT:    v_add_co_u32 v0, s[0:1], 0, s2
; GFX10-NEXT:    v_add_co_ci_u32_e64 v1, s[0:1], 1, s3, s[0:1]
; GFX10-NEXT:    global_load_ubyte v0, v[0:1], off offset:1
; GFX10-NEXT:    s_waitcnt vmcnt(0)
; GFX10-NEXT:    ; return to shader part epilog
  %gep0 = getelementptr inbounds i8, i8 addrspace(1)* %sbase, i64 4294967297
  %load = load i8, i8 addrspace(1)* %gep0
  %zext = zext i8 %load to i32
  %to.vgpr = bitcast i32 %zext to float
  ret float %to.vgpr
}

define amdgpu_ps float @global_load_saddr_i8_offset_4294971391(i8 addrspace(1)* inreg %sbase) {
; GFX9-LABEL: global_load_saddr_i8_offset_4294971391:
; GFX9:       ; %bb.0:
; GFX9-NEXT:    s_add_u32 s0, s2, 0xfff
; GFX9-NEXT:    s_addc_u32 s1, s3, 1
; GFX9-NEXT:    v_mov_b32_e32 v0, 0
; GFX9-NEXT:    global_load_ubyte v0, v0, s[0:1]
; GFX9-NEXT:    s_waitcnt vmcnt(0)
; GFX9-NEXT:    ; return to shader part epilog
;
; GFX10-LABEL: global_load_saddr_i8_offset_4294971391:
; GFX10:       ; %bb.0:
; GFX10-NEXT:    v_add_co_u32 v0, s[0:1], 0x800, s2
; GFX10-NEXT:    v_add_co_ci_u32_e64 v1, s[0:1], 1, s3, s[0:1]
; GFX10-NEXT:    global_load_ubyte v0, v[0:1], off offset:2047
; GFX10-NEXT:    s_waitcnt vmcnt(0)
; GFX10-NEXT:    ; return to shader part epilog
  %gep0 = getelementptr inbounds i8, i8 addrspace(1)* %sbase, i64 4294971391
  %load = load i8, i8 addrspace(1)* %gep0
  %zext = zext i8 %load to i32
  %to.vgpr = bitcast i32 %zext to float
  ret float %to.vgpr
}

define amdgpu_ps float @global_load_saddr_i8_offset_4294971392(i8 addrspace(1)* inreg %sbase) {
; GFX9-LABEL: global_load_saddr_i8_offset_4294971392:
; GFX9:       ; %bb.0:
; GFX9-NEXT:    s_add_u32 s0, s2, 0x1000
; GFX9-NEXT:    s_addc_u32 s1, s3, 1
; GFX9-NEXT:    v_mov_b32_e32 v0, 0
; GFX9-NEXT:    global_load_ubyte v0, v0, s[0:1]
; GFX9-NEXT:    s_waitcnt vmcnt(0)
; GFX9-NEXT:    ; return to shader part epilog
;
; GFX10-LABEL: global_load_saddr_i8_offset_4294971392:
; GFX10:       ; %bb.0:
; GFX10-NEXT:    v_add_co_u32 v0, s[0:1], 0x1000, s2
; GFX10-NEXT:    v_add_co_ci_u32_e64 v1, s[0:1], 1, s3, s[0:1]
; GFX10-NEXT:    global_load_ubyte v0, v[0:1], off
; GFX10-NEXT:    s_waitcnt vmcnt(0)
; GFX10-NEXT:    ; return to shader part epilog
  %gep0 = getelementptr inbounds i8, i8 addrspace(1)* %sbase, i64 4294971392
  %load = load i8, i8 addrspace(1)* %gep0
  %zext = zext i8 %load to i32
  %to.vgpr = bitcast i32 %zext to float
  ret float %to.vgpr
}

define amdgpu_ps float @global_load_saddr_i8_offset_neg4294967295(i8 addrspace(1)* inreg %sbase) {
; GFX9-LABEL: global_load_saddr_i8_offset_neg4294967295:
; GFX9:       ; %bb.0:
; GFX9-NEXT:    v_mov_b32_e32 v0, s2
; GFX9-NEXT:    v_add_co_u32_e32 v0, vcc, 0x1000, v0
; GFX9-NEXT:    v_mov_b32_e32 v1, s3
; GFX9-NEXT:    v_addc_co_u32_e32 v1, vcc, -1, v1, vcc
; GFX9-NEXT:    global_load_ubyte v0, v[0:1], off offset:-4095
; GFX9-NEXT:    s_waitcnt vmcnt(0)
; GFX9-NEXT:    ; return to shader part epilog
;
; GFX10-LABEL: global_load_saddr_i8_offset_neg4294967295:
; GFX10:       ; %bb.0:
; GFX10-NEXT:    v_add_co_u32 v0, s[0:1], 0x800, s2
; GFX10-NEXT:    v_add_co_ci_u32_e64 v1, s[0:1], -1, s3, s[0:1]
; GFX10-NEXT:    global_load_ubyte v0, v[0:1], off offset:-2047
; GFX10-NEXT:    s_waitcnt vmcnt(0)
; GFX10-NEXT:    ; return to shader part epilog
  %gep0 = getelementptr inbounds i8, i8 addrspace(1)* %sbase, i64 -4294967295
  %load = load i8, i8 addrspace(1)* %gep0
  %zext = zext i8 %load to i32
  %to.vgpr = bitcast i32 %zext to float
  ret float %to.vgpr
}

define amdgpu_ps float @global_load_saddr_i8_offset_neg4294967296(i8 addrspace(1)* inreg %sbase) {
; GFX9-LABEL: global_load_saddr_i8_offset_neg4294967296:
; GFX9:       ; %bb.0:
; GFX9-NEXT:    v_mov_b32_e32 v1, s3
; GFX9-NEXT:    v_add_co_u32_e64 v0, vcc, 0, s2
; 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:    ; return to shader part epilog
;
; GFX10-LABEL: global_load_saddr_i8_offset_neg4294967296:
; GFX10:       ; %bb.0:
; GFX10-NEXT:    v_add_co_u32 v0, s[0:1], 0, s2
; GFX10-NEXT:    v_add_co_ci_u32_e64 v1, s[0:1], -1, s3, s[0:1]
; GFX10-NEXT:    global_load_ubyte v0, v[0:1], off
; GFX10-NEXT:    s_waitcnt vmcnt(0)
; GFX10-NEXT:    ; return to shader part epilog
  %gep0 = getelementptr inbounds i8, i8 addrspace(1)* %sbase, i64 -4294967296
  %load = load i8, i8 addrspace(1)* %gep0
  %zext = zext i8 %load to i32
  %to.vgpr = bitcast i32 %zext to float
  ret float %to.vgpr
}

define amdgpu_ps float @global_load_saddr_i8_offset_neg4294967297(i8 addrspace(1)* inreg %sbase) {
; GFX9-LABEL: global_load_saddr_i8_offset_neg4294967297:
; GFX9:       ; %bb.0:
; GFX9-NEXT:    v_mov_b32_e32 v1, s3
; GFX9-NEXT:    v_add_co_u32_e64 v0, vcc, 0, s2
; GFX9-NEXT:    v_addc_co_u32_e32 v1, vcc, -1, v1, vcc
; GFX9-NEXT:    global_load_ubyte v0, v[0:1], off offset:-1
; GFX9-NEXT:    s_waitcnt vmcnt(0)
; GFX9-NEXT:    ; return to shader part epilog
;
; GFX10-LABEL: global_load_saddr_i8_offset_neg4294967297:
; GFX10:       ; %bb.0:
; GFX10-NEXT:    v_add_co_u32 v0, s[0:1], 0, s2
; GFX10-NEXT:    v_add_co_ci_u32_e64 v1, s[0:1], -1, s3, s[0:1]
; GFX10-NEXT:    global_load_ubyte v0, v[0:1], off offset:-1
; GFX10-NEXT:    s_waitcnt vmcnt(0)
; GFX10-NEXT:    ; return to shader part epilog
  %gep0 = getelementptr inbounds i8, i8 addrspace(1)* %sbase, i64 -4294967297
  %load = load i8, i8 addrspace(1)* %gep0
  %zext = zext i8 %load to i32
  %to.vgpr = bitcast i32 %zext to float
  ret float %to.vgpr
}

; --------------------------------------------------------------------------------
; Basic addressing patterns
; --------------------------------------------------------------------------------

; Basic pattern, no immediate offset.
define amdgpu_ps float @global_load_saddr_i8_zext_vgpr(i8 addrspace(1)* inreg %sbase, i32 %voffset) {
; GCN-LABEL: global_load_saddr_i8_zext_vgpr:
; GCN:       ; %bb.0:
; GCN-NEXT:    global_load_ubyte v0, v0, s[2:3]
; GCN-NEXT:    s_waitcnt vmcnt(0)
; GCN-NEXT:    ; return to shader part epilog
  %zext.offset = zext i32 %voffset to i64
  %gep0 = getelementptr inbounds i8, i8 addrspace(1)* %sbase, i64 %zext.offset
  %load = load i8, i8 addrspace(1)* %gep0
  %zext = zext i8 %load to i32
  %to.vgpr = bitcast i32 %zext to float
  ret float %to.vgpr
}

; Maximum positive offset on gfx9
define amdgpu_ps float @global_load_saddr_i8_zext_vgpr_offset_4095(i8 addrspace(1)* inreg %sbase, i32 %voffset) {
; GFX9-LABEL: global_load_saddr_i8_zext_vgpr_offset_4095:
; GFX9:       ; %bb.0:
; GFX9-NEXT:    global_load_ubyte v0, v0, s[2:3] offset:4095
; GFX9-NEXT:    s_waitcnt vmcnt(0)
; GFX9-NEXT:    ; return to shader part epilog
;
; GFX10-LABEL: global_load_saddr_i8_zext_vgpr_offset_4095:
; GFX10:       ; %bb.0:
; GFX10-NEXT:    v_add_co_u32 v0, s[0:1], s2, v0
; GFX10-NEXT:    v_add_co_ci_u32_e64 v1, s[0:1], s3, 0, s[0:1]
; GFX10-NEXT:    v_add_co_u32 v0, vcc, 0x800, v0
; GFX10-NEXT:    v_add_co_ci_u32_e32 v1, vcc, 0, v1, vcc
; GFX10-NEXT:    global_load_ubyte v0, v[0:1], off offset:2047
; GFX10-NEXT:    s_waitcnt vmcnt(0)
; GFX10-NEXT:    ; return to shader part epilog
  %zext.offset = zext i32 %voffset to i64
  %gep0 = getelementptr inbounds i8, i8 addrspace(1)* %sbase, i64 %zext.offset
  %gep1 = getelementptr inbounds i8, i8 addrspace(1)* %gep0, i64 4095
  %load = load i8, i8 addrspace(1)* %gep1
  %zext = zext i8 %load to i32
  %to.vgpr = bitcast i32 %zext to float
  ret float %to.vgpr
}

; Maximum positive offset on gfx9 + 1
define amdgpu_ps float @global_load_saddr_i8_zext_vgpr_offset_4096(i8 addrspace(1)* inreg %sbase, i32 %voffset) {
; GFX9-LABEL: global_load_saddr_i8_zext_vgpr_offset_4096:
; GFX9:       ; %bb.0:
; GFX9-NEXT:    v_mov_b32_e32 v1, s3
; GFX9-NEXT:    v_add_co_u32_e32 v0, vcc, s2, v0
; GFX9-NEXT:    v_addc_co_u32_e32 v1, vcc, 0, v1, vcc
; 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:    s_waitcnt vmcnt(0)
; GFX9-NEXT:    ; return to shader part epilog
;
; GFX10-LABEL: global_load_saddr_i8_zext_vgpr_offset_4096:
; GFX10:       ; %bb.0:
; GFX10-NEXT:    v_add_co_u32 v0, s[0:1], s2, v0
; GFX10-NEXT:    v_add_co_ci_u32_e64 v1, s[0:1], s3, 0, s[0:1]
; GFX10-NEXT:    v_add_co_u32 v0, vcc, 0x1000, v0
; GFX10-NEXT:    v_add_co_ci_u32_e32 v1, vcc, 0, v1, vcc
; GFX10-NEXT:    global_load_ubyte v0, v[0:1], off
; GFX10-NEXT:    s_waitcnt vmcnt(0)
; GFX10-NEXT:    ; return to shader part epilog
  %zext.offset = zext i32 %voffset to i64
  %gep0 = getelementptr inbounds i8, i8 addrspace(1)* %sbase, i64 %zext.offset
  %gep1 = getelementptr inbounds i8, i8 addrspace(1)* %gep0, i64 4096
  %load = load i8, i8 addrspace(1)* %gep1
  %zext = zext i8 %load to i32
  %to.vgpr = bitcast i32 %zext to float
  ret float %to.vgpr
}

; Maximum negative offset on gfx9
define amdgpu_ps float @global_load_saddr_i8_zext_vgpr_offset_neg4096(i8 addrspace(1)* inreg %sbase, i32 %voffset) {
; GFX9-LABEL: global_load_saddr_i8_zext_vgpr_offset_neg4096:
; GFX9:       ; %bb.0:
; GFX9-NEXT:    global_load_ubyte v0, v0, s[2:3] offset:-4096
; GFX9-NEXT:    s_waitcnt vmcnt(0)
; GFX9-NEXT:    ; return to shader part epilog
;
; GFX10-LABEL: global_load_saddr_i8_zext_vgpr_offset_neg4096:
; GFX10:       ; %bb.0:
; GFX10-NEXT:    v_add_co_u32 v0, s[0:1], s2, v0
; GFX10-NEXT:    v_add_co_ci_u32_e64 v1, s[0:1], s3, 0, s[0:1]
; GFX10-NEXT:    v_add_co_u32 v0, vcc, 0xfffff000, v0
; GFX10-NEXT:    v_add_co_ci_u32_e32 v1, vcc, -1, v1, vcc
; GFX10-NEXT:    global_load_ubyte v0, v[0:1], off
; GFX10-NEXT:    s_waitcnt vmcnt(0)
; GFX10-NEXT:    ; return to shader part epilog
  %zext.offset = zext i32 %voffset to i64
  %gep0 = getelementptr inbounds i8, i8 addrspace(1)* %sbase, i64 %zext.offset
  %gep1 = getelementptr inbounds i8, i8 addrspace(1)* %gep0, i64 -4096
  %load = load i8, i8 addrspace(1)* %gep1
  %zext = zext i8 %load to i32
  %to.vgpr = bitcast i32 %zext to float
  ret float %to.vgpr
}

; Maximum negative offset on gfx9 - 1
define amdgpu_ps float @global_load_saddr_i8_zext_vgpr_offset_neg4097(i8 addrspace(1)* inreg %sbase, i32 %voffset) {
; GFX9-LABEL: global_load_saddr_i8_zext_vgpr_offset_neg4097:
; GFX9:       ; %bb.0:
; GFX9-NEXT:    v_mov_b32_e32 v1, s3
; GFX9-NEXT:    v_add_co_u32_e32 v0, vcc, s2, v0
; GFX9-NEXT:    v_addc_co_u32_e32 v1, vcc, 0, v1, vcc
; GFX9-NEXT:    v_add_co_u32_e32 v0, vcc, 0xfffff000, v0
; GFX9-NEXT:    v_addc_co_u32_e32 v1, vcc, -1, v1, vcc
; GFX9-NEXT:    global_load_ubyte v0, v[0:1], off offset:-1
; GFX9-NEXT:    s_waitcnt vmcnt(0)
; GFX9-NEXT:    ; return to shader part epilog
;
; GFX10-LABEL: global_load_saddr_i8_zext_vgpr_offset_neg4097:
; GFX10:       ; %bb.0:
; GFX10-NEXT:    v_add_co_u32 v0, s[0:1], s2, v0
; GFX10-NEXT:    v_add_co_ci_u32_e64 v1, s[0:1], s3, 0, s[0:1]
; GFX10-NEXT:    v_add_co_u32 v0, vcc, 0xfffff000, v0
; GFX10-NEXT:    v_add_co_ci_u32_e32 v1, vcc, -1, v1, vcc
; GFX10-NEXT:    global_load_ubyte v0, v[0:1], off offset:-1
; GFX10-NEXT:    s_waitcnt vmcnt(0)
; GFX10-NEXT:    ; return to shader part epilog
  %zext.offset = zext i32 %voffset to i64
  %gep0 = getelementptr inbounds i8, i8 addrspace(1)* %sbase, i64 %zext.offset
  %gep1 = getelementptr inbounds i8, i8 addrspace(1)* %gep0, i64 -4097
  %load = load i8, i8 addrspace(1)* %gep1
  %zext = zext i8 %load to i32
  %to.vgpr = bitcast i32 %zext to float
  ret float %to.vgpr
}

; Maximum positive offset on gfx10
define amdgpu_ps float @global_load_saddr_i8_zext_vgpr_offset_2047(i8 addrspace(1)* inreg %sbase, i32 %voffset) {
; GCN-LABEL: global_load_saddr_i8_zext_vgpr_offset_2047:
; GCN:       ; %bb.0:
; GCN-NEXT:    global_load_ubyte v0, v0, s[2:3] offset:2047
; GCN-NEXT:    s_waitcnt vmcnt(0)
; GCN-NEXT:    ; return to shader part epilog
  %zext.offset = zext i32 %voffset to i64
  %gep0 = getelementptr inbounds i8, i8 addrspace(1)* %sbase, i64 %zext.offset
  %gep1 = getelementptr inbounds i8, i8 addrspace(1)* %gep0, i64 2047
  %load = load i8, i8 addrspace(1)* %gep1
  %zext = zext i8 %load to i32
  %to.vgpr = bitcast i32 %zext to float
  ret float %to.vgpr
}

; Maximum positive offset on gfx10 + 1
define amdgpu_ps float @global_load_saddr_i8_zext_vgpr_offset_2048(i8 addrspace(1)* inreg %sbase, i32 %voffset) {
; GFX9-LABEL: global_load_saddr_i8_zext_vgpr_offset_2048:
; GFX9:       ; %bb.0:
; GFX9-NEXT:    global_load_ubyte v0, v0, s[2:3] offset:2048
; GFX9-NEXT:    s_waitcnt vmcnt(0)
; GFX9-NEXT:    ; return to shader part epilog
;
; GFX10-LABEL: global_load_saddr_i8_zext_vgpr_offset_2048:
; GFX10:       ; %bb.0:
; GFX10-NEXT:    v_add_co_u32 v0, s[0:1], s2, v0
; GFX10-NEXT:    v_add_co_ci_u32_e64 v1, s[0:1], s3, 0, s[0:1]
; GFX10-NEXT:    v_add_co_u32 v0, vcc, 0x800, v0
; GFX10-NEXT:    v_add_co_ci_u32_e32 v1, vcc, 0, v1, vcc
; GFX10-NEXT:    global_load_ubyte v0, v[0:1], off
; GFX10-NEXT:    s_waitcnt vmcnt(0)
; GFX10-NEXT:    ; return to shader part epilog
  %zext.offset = zext i32 %voffset to i64
  %gep0 = getelementptr inbounds i8, i8 addrspace(1)* %sbase, i64 %zext.offset
  %gep1 = getelementptr inbounds i8, i8 addrspace(1)* %gep0, i64 2048
  %load = load i8, i8 addrspace(1)* %gep1
  %zext = zext i8 %load to i32
  %to.vgpr = bitcast i32 %zext to float
  ret float %to.vgpr
}

; Maximum negative offset on gfx10
define amdgpu_ps float @global_load_saddr_i8_zext_vgpr_offset_neg2048(i8 addrspace(1)* inreg %sbase, i32 %voffset) {
; GCN-LABEL: global_load_saddr_i8_zext_vgpr_offset_neg2048:
; GCN:       ; %bb.0:
; GCN-NEXT:    global_load_ubyte v0, v0, s[2:3] offset:-2048
; GCN-NEXT:    s_waitcnt vmcnt(0)
; GCN-NEXT:    ; return to shader part epilog
  %zext.offset = zext i32 %voffset to i64
  %gep0 = getelementptr inbounds i8, i8 addrspace(1)* %sbase, i64 %zext.offset
  %gep1 = getelementptr inbounds i8, i8 addrspace(1)* %gep0, i64 -2048
  %load = load i8, i8 addrspace(1)* %gep1
  %zext = zext i8 %load to i32
  %to.vgpr = bitcast i32 %zext to float
  ret float %to.vgpr
}

; Maximum negative offset on gfx10 - 1
define amdgpu_ps float @global_load_saddr_i8_zext_vgpr_offset_neg2049(i8 addrspace(1)* inreg %sbase, i32 %voffset) {
; GFX9-LABEL: global_load_saddr_i8_zext_vgpr_offset_neg2049:
; GFX9:       ; %bb.0:
; GFX9-NEXT:    global_load_ubyte v0, v0, s[2:3] offset:-2049
; GFX9-NEXT:    s_waitcnt vmcnt(0)
; GFX9-NEXT:    ; return to shader part epilog
;
; GFX10-LABEL: global_load_saddr_i8_zext_vgpr_offset_neg2049:
; GFX10:       ; %bb.0:
; GFX10-NEXT:    v_add_co_u32 v0, s[0:1], s2, v0
; GFX10-NEXT:    v_add_co_ci_u32_e64 v1, s[0:1], s3, 0, s[0:1]
; GFX10-NEXT:    v_add_co_u32 v0, vcc, 0xfffff800, v0
; GFX10-NEXT:    v_add_co_ci_u32_e32 v1, vcc, -1, v1, vcc
; GFX10-NEXT:    global_load_ubyte v0, v[0:1], off offset:-1
; GFX10-NEXT:    s_waitcnt vmcnt(0)
; GFX10-NEXT:    ; return to shader part epilog
  %zext.offset = zext i32 %voffset to i64
  %gep0 = getelementptr inbounds i8, i8 addrspace(1)* %sbase, i64 %zext.offset
  %gep1 = getelementptr inbounds i8, i8 addrspace(1)* %gep0, i64 -2049
  %load = load i8, i8 addrspace(1)* %gep1
  %zext = zext i8 %load to i32
  %to.vgpr = bitcast i32 %zext to float
  ret float %to.vgpr
}

; Maximum positive offset on gfx9, and immediate needs to be moved lower.
define amdgpu_ps float @global_load_saddr_i8_zext_vgpr_offset_4095_gep_order(i8 addrspace(1)* inreg %sbase, i32 %voffset) {
; GFX9-LABEL: global_load_saddr_i8_zext_vgpr_offset_4095_gep_order:
; GFX9:       ; %bb.0:
; GFX9-NEXT:    global_load_ubyte v0, v0, s[2:3] offset:4095
; GFX9-NEXT:    s_waitcnt vmcnt(0)
; GFX9-NEXT:    ; return to shader part epilog
;
; GFX10-LABEL: global_load_saddr_i8_zext_vgpr_offset_4095_gep_order:
; GFX10:       ; %bb.0:
; GFX10-NEXT:    v_add_co_u32 v0, s[0:1], s2, v0
; GFX10-NEXT:    v_add_co_ci_u32_e64 v1, s[0:1], s3, 0, s[0:1]
; GFX10-NEXT:    v_add_co_u32 v0, vcc, 0x800, v0
; GFX10-NEXT:    v_add_co_ci_u32_e32 v1, vcc, 0, v1, vcc
; GFX10-NEXT:    global_load_ubyte v0, v[0:1], off offset:2047
; GFX10-NEXT:    s_waitcnt vmcnt(0)
; GFX10-NEXT:    ; return to shader part epilog
  %zext.offset = zext i32 %voffset to i64
  %gep0 = getelementptr inbounds i8, i8 addrspace(1)* %sbase, i64 4095
  %gep1 = getelementptr inbounds i8, i8 addrspace(1)* %gep0, i64 %zext.offset
  %load = load i8, i8 addrspace(1)* %gep1
  %zext = zext i8 %load to i32
  %to.vgpr = bitcast i32 %zext to float
  ret float %to.vgpr
}

; pointer addressing done in integers
define amdgpu_ps float @global_load_saddr_i8_zext_vgpr_ptrtoint(i8 addrspace(1)* inreg %sbase, i32 %voffset) {
; GCN-LABEL: global_load_saddr_i8_zext_vgpr_ptrtoint:
; GCN:       ; %bb.0:
; GCN-NEXT:    global_load_ubyte v0, v0, s[2:3]
; GCN-NEXT:    s_waitcnt vmcnt(0)
; GCN-NEXT:    ; return to shader part epilog
  %zext.offset = zext i32 %voffset to i64
  %sbase.as.int = ptrtoint i8 addrspace(1)* %sbase to i64
  %add = add i64 %sbase.as.int, %zext.offset
  %dirty.gep = inttoptr i64 %add to i8 addrspace(1)*
  %load = load i8, i8 addrspace(1)* %dirty.gep
  %zext = zext i8 %load to i32
  %to.vgpr = bitcast i32 %zext to float
  ret float %to.vgpr
}

; zext forced to LHS of addressing expression
define amdgpu_ps float @global_load_saddr_i8_zext_vgpr_ptrtoint_commute_add(i8 addrspace(1)* inreg %sbase, i32 %voffset) {
; GCN-LABEL: global_load_saddr_i8_zext_vgpr_ptrtoint_commute_add:
; GCN:       ; %bb.0:
; GCN-NEXT:    global_load_ubyte v0, v0, s[2:3]
; GCN-NEXT:    s_waitcnt vmcnt(0)
; GCN-NEXT:    ; return to shader part epilog
  %zext.offset = zext i32 %voffset to i64
  %sbase.as.int = ptrtoint i8 addrspace(1)* %sbase to i64
  %add = add i64 %zext.offset, %sbase.as.int
  %dirty.gep = inttoptr i64 %add to i8 addrspace(1)*
  %load = load i8, i8 addrspace(1)* %dirty.gep
  %zext = zext i8 %load to i32
  %to.vgpr = bitcast i32 %zext to float
  ret float %to.vgpr
}

; zext forced to LHS of addressing expression, with immediate offset
define amdgpu_ps float @global_load_saddr_i8_zext_vgpr_ptrtoint_commute_add_imm_offset0(i8 addrspace(1)* inreg %sbase, i32 %voffset) {
; GCN-LABEL: global_load_saddr_i8_zext_vgpr_ptrtoint_commute_add_imm_offset0:
; GCN:       ; %bb.0:
; GCN-NEXT:    global_load_ubyte v0, v0, s[2:3] offset:128
; GCN-NEXT:    s_waitcnt vmcnt(0)
; GCN-NEXT:    ; return to shader part epilog
  %zext.offset = zext i32 %voffset to i64
  %sbase.as.int = ptrtoint i8 addrspace(1)* %sbase to i64
  %add = add i64 %zext.offset, %sbase.as.int
  %add.immoffset = add i64 %add, 128
  %dirty.gep = inttoptr i64 %add.immoffset to i8 addrspace(1)*
  %load = load i8, i8 addrspace(1)* %dirty.gep
  %zext = zext i8 %load to i32
  %to.vgpr = bitcast i32 %zext to float
  ret float %to.vgpr
}

; zext forced to LHS of addressing expression, with immediate offset in non-canonical position
define amdgpu_ps float @global_load_saddr_i8_zext_vgpr_ptrtoint_commute_add_imm_offset1(i8 addrspace(1)* inreg %sbase, i32 %voffset) {
; GCN-LABEL: global_load_saddr_i8_zext_vgpr_ptrtoint_commute_add_imm_offset1:
; GCN:       ; %bb.0:
; GCN-NEXT:    global_load_ubyte v0, v0, s[2:3] offset:128
; GCN-NEXT:    s_waitcnt vmcnt(0)
; GCN-NEXT:    ; return to shader part epilog
  %zext.offset = zext i32 %voffset to i64
  %sbase.as.int = ptrtoint i8 addrspace(1)* %sbase to i64
  %add.immoffset = add i64 %sbase.as.int, 128
  %add = add i64 %zext.offset, %add.immoffset
  %dirty.gep = inttoptr i64 %add to i8 addrspace(1)*
  %load = load i8, i8 addrspace(1)* %dirty.gep
  %zext = zext i8 %load to i32
  %to.vgpr = bitcast i32 %zext to float
  ret float %to.vgpr
}

; --------------------------------------------------------------------------------
; Uniformity edge cases
; --------------------------------------------------------------------------------

@ptr.in.lds = internal addrspace(3) global i8 addrspace(1)* undef

; Base pointer is uniform, but also in VGPRs
define amdgpu_ps float @global_load_saddr_uniform_ptr_in_vgprs(i32 %voffset) {
; GFX9-LABEL: global_load_saddr_uniform_ptr_in_vgprs:
; GFX9:       ; %bb.0:
; GFX9-NEXT:    v_mov_b32_e32 v1, 0
; GFX9-NEXT:    ds_read_b64 v[1:2], v1
; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-NEXT:    v_readfirstlane_b32 s0, v1
; GFX9-NEXT:    v_readfirstlane_b32 s1, v2
; GFX9-NEXT:    s_nop 4
; GFX9-NEXT:    global_load_ubyte v0, v0, s[0:1]
; GFX9-NEXT:    s_waitcnt vmcnt(0)
; GFX9-NEXT:    ; return to shader part epilog
;
; GFX10-LABEL: global_load_saddr_uniform_ptr_in_vgprs:
; GFX10:       ; %bb.0:
; GFX10-NEXT:    v_mov_b32_e32 v1, 0
; GFX10-NEXT:    ds_read_b64 v[1:2], v1
; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-NEXT:    v_readfirstlane_b32 s0, v1
; GFX10-NEXT:    v_readfirstlane_b32 s1, v2
; GFX10-NEXT:    global_load_ubyte v0, v0, s[0:1]
; GFX10-NEXT:    s_waitcnt vmcnt(0)
; GFX10-NEXT:    ; return to shader part epilog
  %sbase = load i8 addrspace(1)*, i8 addrspace(1)* addrspace(3)* @ptr.in.lds
  %zext.offset = zext i32 %voffset to i64
  %gep0 = getelementptr inbounds i8, i8 addrspace(1)* %sbase, i64 %zext.offset
  %load = load i8, i8 addrspace(1)* %gep0
  %zext = zext i8 %load to i32
  %to.vgpr = bitcast i32 %zext to float
  ret float %to.vgpr
}

; Base pointer is uniform, but also in VGPRs, with imm offset
define amdgpu_ps float @global_load_saddr_uniform_ptr_in_vgprs_immoffset(i32 %voffset) {
; GFX9-LABEL: global_load_saddr_uniform_ptr_in_vgprs_immoffset:
; GFX9:       ; %bb.0:
; GFX9-NEXT:    v_mov_b32_e32 v1, 0
; GFX9-NEXT:    ds_read_b64 v[1:2], v1
; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-NEXT:    v_readfirstlane_b32 s0, v1
; GFX9-NEXT:    v_readfirstlane_b32 s1, v2
; GFX9-NEXT:    s_nop 4
; GFX9-NEXT:    global_load_ubyte v0, v0, s[0:1] offset:42
; GFX9-NEXT:    s_waitcnt vmcnt(0)
; GFX9-NEXT:    ; return to shader part epilog
;
; GFX10-LABEL: global_load_saddr_uniform_ptr_in_vgprs_immoffset:
; GFX10:       ; %bb.0:
; GFX10-NEXT:    v_mov_b32_e32 v1, 0
; GFX10-NEXT:    ds_read_b64 v[1:2], v1
; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-NEXT:    v_readfirstlane_b32 s0, v1
; GFX10-NEXT:    v_readfirstlane_b32 s1, v2
; GFX10-NEXT:    global_load_ubyte v0, v0, s[0:1] offset:42
; GFX10-NEXT:    s_waitcnt vmcnt(0)
; GFX10-NEXT:    ; return to shader part epilog
  %sbase = load i8 addrspace(1)*, i8 addrspace(1)* addrspace(3)* @ptr.in.lds
  %zext.offset = zext i32 %voffset to i64
  %gep0 = getelementptr inbounds i8, i8 addrspace(1)* %sbase, i64 %zext.offset
  %gep1 = getelementptr inbounds i8, i8 addrspace(1)* %gep0, i64 42
  %load = load i8, i8 addrspace(1)* %gep1
  %zext = zext i8 %load to i32
  %to.vgpr = bitcast i32 %zext to float
  ret float %to.vgpr
}

; Both 64-bit base and 32-bit offset are scalar
define amdgpu_ps float @global_load_saddr_i8_zext_uniform_offset(i8 addrspace(1)* inreg %sbase, i32 inreg %soffset) {
; GCN-LABEL: global_load_saddr_i8_zext_uniform_offset:
; GCN:       ; %bb.0:
; GCN-NEXT:    v_mov_b32_e32 v0, s4
; GCN-NEXT:    global_load_ubyte v0, v0, s[2:3]
; GCN-NEXT:    s_waitcnt vmcnt(0)
; GCN-NEXT:    ; return to shader part epilog
  %zext.offset = zext i32 %soffset to i64
  %gep0 = getelementptr inbounds i8, i8 addrspace(1)* %sbase, i64 %zext.offset
  %load = load i8, i8 addrspace(1)* %gep0
  %zext = zext i8 %load to i32
  %to.vgpr = bitcast i32 %zext to float
  ret float %to.vgpr
}

; Both 64-bit base and 32-bit offset are scalar, with immediate offset.
define amdgpu_ps float @global_load_saddr_i8_zext_uniform_offset_immoffset(i8 addrspace(1)* inreg %sbase, i32 inreg %soffset) {
; GCN-LABEL: global_load_saddr_i8_zext_uniform_offset_immoffset:
; GCN:       ; %bb.0:
; GCN-NEXT:    v_mov_b32_e32 v0, s4
; GCN-NEXT:    global_load_ubyte v0, v0, s[2:3] offset:-24
; GCN-NEXT:    s_waitcnt vmcnt(0)
; GCN-NEXT:    ; return to shader part epilog
  %zext.offset = zext i32 %soffset to i64
  %gep0 = getelementptr inbounds i8, i8 addrspace(1)* %sbase, i64 %zext.offset
  %gep1 = getelementptr inbounds i8, i8 addrspace(1)* %gep0, i64 -24
  %load = load i8, i8 addrspace(1)* %gep1
  %zext = zext i8 %load to i32
  %to.vgpr = bitcast i32 %zext to float
  ret float %to.vgpr
}

; Both components uniform, zext forced to LHS of addressing expression
define amdgpu_ps float @global_load_saddr_i8_zext_sgpr_ptrtoint_commute_add(i8 addrspace(1)* inreg %sbase, i32 inreg %soffset) {
; GCN-LABEL: global_load_saddr_i8_zext_sgpr_ptrtoint_commute_add:
; GCN:       ; %bb.0:
; GCN-NEXT:    v_mov_b32_e32 v0, s4
; GCN-NEXT:    global_load_ubyte v0, v0, s[2:3]
; GCN-NEXT:    s_waitcnt vmcnt(0)
; GCN-NEXT:    ; return to shader part epilog
  %zext.offset = zext i32 %soffset to i64
  %sbase.as.int = ptrtoint i8 addrspace(1)* %sbase to i64
  %add = add i64 %zext.offset, %sbase.as.int
  %dirty.gep = inttoptr i64 %add to i8 addrspace(1)*
  %load = load i8, i8 addrspace(1)* %dirty.gep
  %zext = zext i8 %load to i32
  %to.vgpr = bitcast i32 %zext to float
  ret float %to.vgpr
}

; Both components uniform, zext forced to LHS of addressing expression, with immediate offset
define amdgpu_ps float @global_load_saddr_i8_zext_sgpr_ptrtoint_commute_add_imm_offset0(i8 addrspace(1)* inreg %sbase, i32 inreg %soffset) {
; GCN-LABEL: global_load_saddr_i8_zext_sgpr_ptrtoint_commute_add_imm_offset0:
; GCN:       ; %bb.0:
; GCN-NEXT:    v_mov_b32_e32 v0, s4
; GCN-NEXT:    global_load_ubyte v0, v0, s[2:3] offset:128
; GCN-NEXT:    s_waitcnt vmcnt(0)
; GCN-NEXT:    ; return to shader part epilog
  %zext.offset = zext i32 %soffset to i64
  %sbase.as.int = ptrtoint i8 addrspace(1)* %sbase to i64
  %add = add i64 %zext.offset, %sbase.as.int
  %add.immoffset = add i64 %add, 128
  %dirty.gep = inttoptr i64 %add.immoffset to i8 addrspace(1)*
  %load = load i8, i8 addrspace(1)* %dirty.gep
  %zext = zext i8 %load to i32
  %to.vgpr = bitcast i32 %zext to float
  ret float %to.vgpr
}

; divergent 64-bit base, 32-bit scalar offset.
define amdgpu_ps float @global_load_i8_vgpr64_sgpr32(i8 addrspace(1)* %vbase, i32 inreg %soffset) {
; GFX9-LABEL: global_load_i8_vgpr64_sgpr32:
; GFX9:       ; %bb.0:
; GFX9-NEXT:    v_add_co_u32_e32 v0, vcc, s2, 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:    s_waitcnt vmcnt(0)
; GFX9-NEXT:    ; return to shader part epilog
;
; GFX10-LABEL: global_load_i8_vgpr64_sgpr32:
; GFX10:       ; %bb.0:
; GFX10-NEXT:    v_add_co_u32 v0, vcc, v0, s2
; GFX10-NEXT:    v_add_co_ci_u32_e32 v1, vcc, 0, v1, vcc
; GFX10-NEXT:    global_load_ubyte v0, v[0:1], off
; GFX10-NEXT:    s_waitcnt vmcnt(0)
; GFX10-NEXT:    ; return to shader part epilog
  %zext.offset = zext i32 %soffset to i64
  %gep0 = getelementptr inbounds i8, i8 addrspace(1)* %vbase, i64 %zext.offset
  %load = load i8, i8 addrspace(1)* %gep0
  %zext = zext i8 %load to i32
  %to.vgpr = bitcast i32 %zext to float
  ret float %to.vgpr
}

; divergent 64-bit base, 32-bit scalar offset, with imm offset
define amdgpu_ps float @global_load_i8_vgpr64_sgpr32_offset_4095(i8 addrspace(1)* %vbase, i32 inreg %soffset) {
; GFX9-LABEL: global_load_i8_vgpr64_sgpr32_offset_4095:
; GFX9:       ; %bb.0:
; GFX9-NEXT:    v_add_co_u32_e32 v0, vcc, s2, 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:    ; return to shader part epilog
;
; GFX10-LABEL: global_load_i8_vgpr64_sgpr32_offset_4095:
; GFX10:       ; %bb.0:
; GFX10-NEXT:    v_add_co_u32 v0, vcc, v0, s2
; GFX10-NEXT:    v_add_co_ci_u32_e32 v1, vcc, 0, v1, vcc
; GFX10-NEXT:    v_add_co_u32 v0, vcc, 0x800, v0
; GFX10-NEXT:    v_add_co_ci_u32_e32 v1, vcc, 0, v1, vcc
; GFX10-NEXT:    global_load_ubyte v0, v[0:1], off offset:2047
; GFX10-NEXT:    s_waitcnt vmcnt(0)
; GFX10-NEXT:    ; return to shader part epilog
  %zext.offset = zext i32 %soffset to i64
  %gep0 = getelementptr inbounds i8, i8 addrspace(1)* %vbase, i64 %zext.offset
  %gep1 = getelementptr inbounds i8, i8 addrspace(1)* %gep0, i64 4095
  %load = load i8, i8 addrspace(1)* %gep1
  %zext = zext i8 %load to i32
  %to.vgpr = bitcast i32 %zext to float
  ret float %to.vgpr
}

; --------------------------------------------------------------------------------
; Natural addressing shifts with restricted range
; --------------------------------------------------------------------------------

; Cannot push the shift into 32-bits, and cannot match.
define amdgpu_ps float @global_load_saddr_f32_natural_addressing(float addrspace(1)* inreg %sbase, i32 addrspace(1)* %voffset.ptr) {
; GFX9-LABEL: global_load_saddr_f32_natural_addressing:
; GFX9:       ; %bb.0:
; GFX9-NEXT:    global_load_dword v0, v[0:1], off
; GFX9-NEXT:    v_mov_b32_e32 v1, 0
; GFX9-NEXT:    v_mov_b32_e32 v2, s3
; GFX9-NEXT:    s_waitcnt vmcnt(0)
; GFX9-NEXT:    v_lshlrev_b64 v[0:1], 2, v[0:1]
; GFX9-NEXT:    v_add_co_u32_e32 v0, vcc, s2, v0
; GFX9-NEXT:    v_addc_co_u32_e32 v1, vcc, v2, v1, vcc
; GFX9-NEXT:    global_load_dword v0, v[0:1], off
; GFX9-NEXT:    s_waitcnt vmcnt(0)
; GFX9-NEXT:    ; return to shader part epilog
;
; GFX10-LABEL: global_load_saddr_f32_natural_addressing:
; GFX10:       ; %bb.0:
; GFX10-NEXT:    global_load_dword v0, v[0:1], off
; GFX10-NEXT:    v_mov_b32_e32 v1, 0
; GFX10-NEXT:    s_waitcnt vmcnt(0)
; GFX10-NEXT:    v_lshlrev_b64 v[0:1], 2, v[0:1]
; GFX10-NEXT:    v_add_co_u32 v0, vcc, s2, v0
; GFX10-NEXT:    v_add_co_ci_u32_e32 v1, vcc, s3, v1, vcc
; GFX10-NEXT:    global_load_dword v0, v[0:1], off
; GFX10-NEXT:    s_waitcnt vmcnt(0)
; GFX10-NEXT:    ; return to shader part epilog
  %voffset = load i32, i32 addrspace(1)* %voffset.ptr
  %zext.offset = zext i32 %voffset to i64
  %gep = getelementptr inbounds float, float addrspace(1)* %sbase, i64 %zext.offset
  %load = load float, float addrspace(1)* %gep
  ret float %load
}

; Cannot push the shift into 32-bits, with an immediate offset.
define amdgpu_ps float @global_load_saddr_f32_natural_addressing_immoffset(i8 addrspace(1)* inreg %sbase, i32 addrspace(1)* %voffset.ptr) {
; GCN-LABEL: global_load_saddr_f32_natural_addressing_immoffset:
; GCN:       ; %bb.0:
; GCN-NEXT:    global_load_dword v0, v[0:1], off
; GCN-NEXT:    s_waitcnt vmcnt(0)
; GCN-NEXT:    global_load_dword v0, v0, s[2:3] offset:128
; GCN-NEXT:    s_waitcnt vmcnt(0)
; GCN-NEXT:    ; return to shader part epilog
  %voffset = load i32, i32 addrspace(1)* %voffset.ptr
  %zext.offset = zext i32 %voffset to i64
  %gep0 = getelementptr inbounds i8, i8 addrspace(1)* %sbase, i64 %zext.offset
  %gep1 = getelementptr inbounds i8, i8 addrspace(1)* %gep0, i64 128
  %gep1.cast = bitcast i8 addrspace(1)* %gep1 to float addrspace(1)*
  %load = load float, float addrspace(1)* %gep1.cast
  ret float %load
}

; Range is sufficiently restricted to push the shift into 32-bits.
define amdgpu_ps float @global_load_f32_saddr_zext_vgpr_range(float addrspace(1)* inreg %sbase, i32 addrspace(1)* %voffset.ptr) {
; GCN-LABEL: global_load_f32_saddr_zext_vgpr_range:
; GCN:       ; %bb.0:
; GCN-NEXT:    global_load_dword v0, v[0:1], off
; GCN-NEXT:    s_waitcnt vmcnt(0)
; GCN-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
; GCN-NEXT:    global_load_dword v0, v0, s[2:3]
; GCN-NEXT:    s_waitcnt vmcnt(0)
; GCN-NEXT:    ; return to shader part epilog
  %voffset = load i32, i32 addrspace(1)* %voffset.ptr, !range !0
  %zext.offset = zext i32 %voffset to i64
  %gep = getelementptr inbounds float, float addrspace(1)* %sbase, i64 %zext.offset
  %load = load float, float addrspace(1)* %gep
  ret float %load
}

; Range is sufficiently restricted to push the shift into 32-bits, with an imm offset
define amdgpu_ps float @global_load_f32_saddr_zext_vgpr_range_imm_offset(float addrspace(1)* inreg %sbase, i32 addrspace(1)* %voffset.ptr) {
; GCN-LABEL: global_load_f32_saddr_zext_vgpr_range_imm_offset:
; GCN:       ; %bb.0:
; GCN-NEXT:    global_load_dword v0, v[0:1], off
; GCN-NEXT:    s_waitcnt vmcnt(0)
; GCN-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
; GCN-NEXT:    global_load_dword v0, v0, s[2:3] offset:400
; GCN-NEXT:    s_waitcnt vmcnt(0)
; GCN-NEXT:    ; return to shader part epilog
  %voffset = load i32, i32 addrspace(1)* %voffset.ptr, !range !0
  %zext.offset = zext i32 %voffset to i64
  %gep0 = getelementptr inbounds float, float addrspace(1)* %sbase, i64 %zext.offset
  %gep1 = getelementptr inbounds float, float addrspace(1)* %gep0, i64 100
  %load = load float, float addrspace(1)* %gep1
  ret float %load
}

; Range is 1 beyond the limit where we can move the shift into 32-bits.
define amdgpu_ps float @global_load_f32_saddr_zext_vgpr_range_too_large(float addrspace(1)* inreg %sbase, i32 addrspace(1)* %voffset.ptr) {
; GFX9-LABEL: global_load_f32_saddr_zext_vgpr_range_too_large:
; GFX9:       ; %bb.0:
; GFX9-NEXT:    global_load_dword v0, v[0:1], off
; GFX9-NEXT:    v_mov_b32_e32 v1, 0
; GFX9-NEXT:    v_mov_b32_e32 v2, s3
; GFX9-NEXT:    s_waitcnt vmcnt(0)
; GFX9-NEXT:    v_lshlrev_b64 v[0:1], 2, v[0:1]
; GFX9-NEXT:    v_add_co_u32_e32 v0, vcc, s2, v0
; GFX9-NEXT:    v_addc_co_u32_e32 v1, vcc, v2, v1, vcc
; GFX9-NEXT:    global_load_dword v0, v[0:1], off
; GFX9-NEXT:    s_waitcnt vmcnt(0)
; GFX9-NEXT:    ; return to shader part epilog
;
; GFX10-LABEL: global_load_f32_saddr_zext_vgpr_range_too_large:
; GFX10:       ; %bb.0:
; GFX10-NEXT:    global_load_dword v0, v[0:1], off
; GFX10-NEXT:    v_mov_b32_e32 v1, 0
; GFX10-NEXT:    s_waitcnt vmcnt(0)
; GFX10-NEXT:    v_lshlrev_b64 v[0:1], 2, v[0:1]
; GFX10-NEXT:    v_add_co_u32 v0, vcc, s2, v0
; GFX10-NEXT:    v_add_co_ci_u32_e32 v1, vcc, s3, v1, vcc
; GFX10-NEXT:    global_load_dword v0, v[0:1], off
; GFX10-NEXT:    s_waitcnt vmcnt(0)
; GFX10-NEXT:    ; return to shader part epilog
  %voffset = load i32, i32 addrspace(1)* %voffset.ptr, !range !1
  %zext.offset = zext i32 %voffset to i64
  %gep = getelementptr inbounds float, float addrspace(1)* %sbase, i64 %zext.offset
  %load = load float, float addrspace(1)* %gep
  ret float %load
}

; --------------------------------------------------------------------------------
; Stress various type loads
; --------------------------------------------------------------------------------

define amdgpu_ps half @global_load_saddr_i16(i8 addrspace(1)* inreg %sbase, i32 %voffset) {
; GCN-LABEL: global_load_saddr_i16:
; GCN:       ; %bb.0:
; GCN-NEXT:    global_load_ushort v0, v0, s[2:3]
; GCN-NEXT:    s_waitcnt vmcnt(0)
; GCN-NEXT:    ; return to shader part epilog
  %zext.offset = zext i32 %voffset to i64
  %gep0 = getelementptr inbounds i8, i8 addrspace(1)* %sbase, i64 %zext.offset
  %gep0.cast = bitcast i8 addrspace(1)* %gep0 to i16 addrspace(1)*
  %load = load i16, i16 addrspace(1)* %gep0.cast
  %cast.load = bitcast i16 %load to half
  ret half %cast.load
}

define amdgpu_ps half @global_load_saddr_i16_immneg128(i8 addrspace(1)* inreg %sbase, i32 %voffset) {
; GCN-LABEL: global_load_saddr_i16_immneg128:
; GCN:       ; %bb.0:
; GCN-NEXT:    global_load_ushort v0, v0, s[2:3] offset:-128
; GCN-NEXT:    s_waitcnt vmcnt(0)
; GCN-NEXT:    ; return to shader part epilog
  %zext.offset = zext i32 %voffset to i64
  %gep0 = getelementptr inbounds i8, i8 addrspace(1)* %sbase, i64 %zext.offset
  %gep1 = getelementptr inbounds i8, i8 addrspace(1)* %gep0, i64 -128
  %gep1.cast = bitcast i8 addrspace(1)* %gep1 to i16 addrspace(1)*
  %load = load i16, i16 addrspace(1)* %gep1.cast
  %cast.load = bitcast i16 %load to half
  ret half %cast.load
}

define amdgpu_ps half @global_load_saddr_f16(i8 addrspace(1)* inreg %sbase, i32 %voffset) {
; GCN-LABEL: global_load_saddr_f16:
; GCN:       ; %bb.0:
; GCN-NEXT:    global_load_ushort v0, v0, s[2:3]
; GCN-NEXT:    s_waitcnt vmcnt(0)
; GCN-NEXT:    ; return to shader part epilog
  %zext.offset = zext i32 %voffset to i64
  %gep0 = getelementptr inbounds i8, i8 addrspace(1)* %sbase, i64 %zext.offset
  %gep0.cast = bitcast i8 addrspace(1)* %gep0 to half addrspace(1)*
  %load = load half, half addrspace(1)* %gep0.cast
  ret half %load
}

define amdgpu_ps half @global_load_saddr_f16_immneg128(i8 addrspace(1)* inreg %sbase, i32 %voffset) {
; GCN-LABEL: global_load_saddr_f16_immneg128:
; GCN:       ; %bb.0:
; GCN-NEXT:    global_load_ushort v0, v0, s[2:3] offset:-128
; GCN-NEXT:    s_waitcnt vmcnt(0)
; GCN-NEXT:    ; return to shader part epilog
  %zext.offset = zext i32 %voffset to i64
  %gep0 = getelementptr inbounds i8, i8 addrspace(1)* %sbase, i64 %zext.offset
  %gep1 = getelementptr inbounds i8, i8 addrspace(1)* %gep0, i64 -128
  %gep1.cast = bitcast i8 addrspace(1)* %gep1 to half addrspace(1)*
  %load = load half, half addrspace(1)* %gep1.cast
  ret half %load
}

define amdgpu_ps float @global_load_saddr_i32(i8 addrspace(1)* inreg %sbase, i32 %voffset) {
; GCN-LABEL: global_load_saddr_i32:
; GCN:       ; %bb.0:
; GCN-NEXT:    global_load_dword v0, v0, s[2:3]
; GCN-NEXT:    s_waitcnt vmcnt(0)
; GCN-NEXT:    ; return to shader part epilog
  %zext.offset = zext i32 %voffset to i64
  %gep0 = getelementptr inbounds i8, i8 addrspace(1)* %sbase, i64 %zext.offset
  %gep0.cast = bitcast i8 addrspace(1)* %gep0 to i32 addrspace(1)*
  %load = load i32, i32 addrspace(1)* %gep0.cast
  %cast.load = bitcast i32 %load to float
  ret float %cast.load
}

define amdgpu_ps float @global_load_saddr_i32_immneg128(i8 addrspace(1)* inreg %sbase, i32 %voffset) {
; GCN-LABEL: global_load_saddr_i32_immneg128:
; GCN:       ; %bb.0:
; GCN-NEXT:    global_load_dword v0, v0, s[2:3] offset:-128
; GCN-NEXT:    s_waitcnt vmcnt(0)
; GCN-NEXT:    ; return to shader part epilog
  %zext.offset = zext i32 %voffset to i64
  %gep0 = getelementptr inbounds i8, i8 addrspace(1)* %sbase, i64 %zext.offset
  %gep1 = getelementptr inbounds i8, i8 addrspace(1)* %gep0, i64 -128
  %gep1.cast = bitcast i8 addrspace(1)* %gep1 to i32 addrspace(1)*
  %load = load i32, i32 addrspace(1)* %gep1.cast
  %cast.load = bitcast i32 %load to float
  ret float %cast.load
}

define amdgpu_ps float @global_load_saddr_f32(i8 addrspace(1)* inreg %sbase, i32 %voffset) {
; GCN-LABEL: global_load_saddr_f32:
; GCN:       ; %bb.0:
; GCN-NEXT:    global_load_dword v0, v0, s[2:3]
; GCN-NEXT:    s_waitcnt vmcnt(0)
; GCN-NEXT:    ; return to shader part epilog
  %zext.offset = zext i32 %voffset to i64
  %gep0 = getelementptr inbounds i8, i8 addrspace(1)* %sbase, i64 %zext.offset
  %gep0.cast = bitcast i8 addrspace(1)* %gep0 to float addrspace(1)*
  %load = load float, float addrspace(1)* %gep0.cast
  ret float %load
}

define amdgpu_ps float @global_load_saddr_f32_immneg128(i8 addrspace(1)* inreg %sbase, i32 %voffset) {
; GCN-LABEL: global_load_saddr_f32_immneg128:
; GCN:       ; %bb.0:
; GCN-NEXT:    global_load_dword v0, v0, s[2:3] offset:-128
; GCN-NEXT:    s_waitcnt vmcnt(0)
; GCN-NEXT:    ; return to shader part epilog
  %zext.offset = zext i32 %voffset to i64
  %gep0 = getelementptr inbounds i8, i8 addrspace(1)* %sbase, i64 %zext.offset
  %gep1 = getelementptr inbounds i8, i8 addrspace(1)* %gep0, i64 -128
  %gep1.cast = bitcast i8 addrspace(1)* %gep1 to float addrspace(1)*
  %load = load float, float addrspace(1)* %gep1.cast
  ret float %load
}

define amdgpu_ps <2 x half> @global_load_saddr_v2i16(i8 addrspace(1)* inreg %sbase, i32 %voffset) {
; GCN-LABEL: global_load_saddr_v2i16:
; GCN:       ; %bb.0:
; GCN-NEXT:    global_load_dword v0, v0, s[2:3]
; GCN-NEXT:    s_waitcnt vmcnt(0)
; GCN-NEXT:    ; return to shader part epilog
  %zext.offset = zext i32 %voffset to i64
  %gep0 = getelementptr inbounds i8, i8 addrspace(1)* %sbase, i64 %zext.offset
  %gep0.cast = bitcast i8 addrspace(1)* %gep0 to <2 x i16> addrspace(1)*
  %load = load <2 x i16>, <2 x i16> addrspace(1)* %gep0.cast
  %cast.load = bitcast <2 x i16> %load to <2 x half>
  ret <2 x half> %cast.load
}

define amdgpu_ps <2 x half> @global_load_saddr_v2i16_immneg128(i8 addrspace(1)* inreg %sbase, i32 %voffset) {
; GCN-LABEL: global_load_saddr_v2i16_immneg128:
; GCN:       ; %bb.0:
; GCN-NEXT:    global_load_dword v0, v0, s[2:3] offset:-128
; GCN-NEXT:    s_waitcnt vmcnt(0)
; GCN-NEXT:    ; return to shader part epilog
  %zext.offset = zext i32 %voffset to i64
  %gep0 = getelementptr inbounds i8, i8 addrspace(1)* %sbase, i64 %zext.offset
  %gep1 = getelementptr inbounds i8, i8 addrspace(1)* %gep0, i64 -128
  %gep1.cast = bitcast i8 addrspace(1)* %gep1 to <2 x i16> addrspace(1)*
  %load = load <2 x i16>, <2 x i16> addrspace(1)* %gep1.cast
  %cast.load = bitcast <2 x i16> %load to <2 x half>
  ret <2 x half> %cast.load
}

define amdgpu_ps <2 x half> @global_load_saddr_v2f16(i8 addrspace(1)* inreg %sbase, i32 %voffset) {
; GCN-LABEL: global_load_saddr_v2f16:
; GCN:       ; %bb.0:
; GCN-NEXT:    global_load_dword v0, v0, s[2:3]
; GCN-NEXT:    s_waitcnt vmcnt(0)
; GCN-NEXT:    ; return to shader part epilog
  %zext.offset = zext i32 %voffset to i64
  %gep0 = getelementptr inbounds i8, i8 addrspace(1)* %sbase, i64 %zext.offset
  %gep0.cast = bitcast i8 addrspace(1)* %gep0 to <2 x half> addrspace(1)*
  %load = load <2 x half>, <2 x half> addrspace(1)* %gep0.cast
  ret <2 x half> %load
}

define amdgpu_ps <2 x half> @global_load_saddr_v2f16_immneg128(i8 addrspace(1)* inreg %sbase, i32 %voffset) {
; GCN-LABEL: global_load_saddr_v2f16_immneg128:
; GCN:       ; %bb.0:
; GCN-NEXT:    global_load_dword v0, v0, s[2:3] offset:-128
; GCN-NEXT:    s_waitcnt vmcnt(0)
; GCN-NEXT:    ; return to shader part epilog
  %zext.offset = zext i32 %voffset to i64
  %gep0 = getelementptr inbounds i8, i8 addrspace(1)* %sbase, i64 %zext.offset
  %gep1 = getelementptr inbounds i8, i8 addrspace(1)* %gep0, i64 -128
  %gep1.cast = bitcast i8 addrspace(1)* %gep1 to <2 x half> addrspace(1)*
  %load = load <2 x half>, <2 x half> addrspace(1)* %gep1.cast
  ret <2 x half> %load
}

define amdgpu_ps <2 x half> @global_load_saddr_p3(i8 addrspace(1)* inreg %sbase, i32 %voffset) {
; GCN-LABEL: global_load_saddr_p3:
; GCN:       ; %bb.0:
; GCN-NEXT:    global_load_dword v0, v0, s[2:3]
; GCN-NEXT:    s_waitcnt vmcnt(0)
; GCN-NEXT:    ; return to shader part epilog
  %zext.offset = zext i32 %voffset to i64
  %gep0 = getelementptr inbounds i8, i8 addrspace(1)* %sbase, i64 %zext.offset
  %gep0.cast = bitcast i8 addrspace(1)* %gep0 to i8 addrspace(3)* addrspace(1)*
  %load = load i8 addrspace(3)*, i8 addrspace(3)* addrspace(1)* %gep0.cast
  %cast.load0 = ptrtoint i8 addrspace(3)* %load to i32
  %cast.load1 = bitcast i32 %cast.load0 to <2 x half>
  ret <2 x half> %cast.load1
}

define amdgpu_ps <2 x half> @global_load_saddr_p3_immneg128(i8 addrspace(1)* inreg %sbase, i32 %voffset) {
; GCN-LABEL: global_load_saddr_p3_immneg128:
; GCN:       ; %bb.0:
; GCN-NEXT:    global_load_dword v0, v0, s[2:3] offset:-128
; GCN-NEXT:    s_waitcnt vmcnt(0)
; GCN-NEXT:    ; return to shader part epilog
  %zext.offset = zext i32 %voffset to i64
  %gep0 = getelementptr inbounds i8, i8 addrspace(1)* %sbase, i64 %zext.offset
  %gep1 = getelementptr inbounds i8, i8 addrspace(1)* %gep0, i64 -128
  %gep1.cast = bitcast i8 addrspace(1)* %gep1 to i8 addrspace(3)* addrspace(1)*
  %load = load i8 addrspace(3)*, i8 addrspace(3)* addrspace(1)* %gep1.cast
  %cast.load0 = ptrtoint i8 addrspace(3)* %load to i32
  %cast.load1 = bitcast i32 %cast.load0 to <2 x half>
  ret <2 x half> %cast.load1
}

define amdgpu_ps <2 x float> @global_load_saddr_f64(i8 addrspace(1)* inreg %sbase, i32 %voffset) {
; GCN-LABEL: global_load_saddr_f64:
; GCN:       ; %bb.0:
; GCN-NEXT:    global_load_dwordx2 v[0:1], v0, s[2:3]
; GCN-NEXT:    s_waitcnt vmcnt(0)
; GCN-NEXT:    ; return to shader part epilog
  %zext.offset = zext i32 %voffset to i64
  %gep0 = getelementptr inbounds i8, i8 addrspace(1)* %sbase, i64 %zext.offset
  %gep0.cast = bitcast i8 addrspace(1)* %gep0 to double addrspace(1)*
  %load = load double, double addrspace(1)* %gep0.cast
  %cast.load = bitcast double %load to <2 x float>
  ret <2 x float> %cast.load
}

define amdgpu_ps <2 x float> @global_load_saddr_f64_immneg128(i8 addrspace(1)* inreg %sbase, i32 %voffset) {
; GCN-LABEL: global_load_saddr_f64_immneg128:
; GCN:       ; %bb.0:
; GCN-NEXT:    global_load_dwordx2 v[0:1], v0, s[2:3] offset:-128
; GCN-NEXT:    s_waitcnt vmcnt(0)
; GCN-NEXT:    ; return to shader part epilog
  %zext.offset = zext i32 %voffset to i64
  %gep0 = getelementptr inbounds i8, i8 addrspace(1)* %sbase, i64 %zext.offset
  %gep1 = getelementptr inbounds i8, i8 addrspace(1)* %gep0, i64 -128
  %gep1.cast = bitcast i8 addrspace(1)* %gep1 to double addrspace(1)*
  %load = load double, double addrspace(1)* %gep1.cast
  %cast.load = bitcast double %load to <2 x float>
  ret <2 x float> %cast.load
}

define amdgpu_ps <2 x float> @global_load_saddr_i64(i8 addrspace(1)* inreg %sbase, i32 %voffset) {
; GCN-LABEL: global_load_saddr_i64:
; GCN:       ; %bb.0:
; GCN-NEXT:    global_load_dwordx2 v[0:1], v0, s[2:3]
; GCN-NEXT:    s_waitcnt vmcnt(0)
; GCN-NEXT:    ; return to shader part epilog
  %zext.offset = zext i32 %voffset to i64
  %gep0 = getelementptr inbounds i8, i8 addrspace(1)* %sbase, i64 %zext.offset
  %gep0.cast = bitcast i8 addrspace(1)* %gep0 to i64 addrspace(1)*
  %load = load i64, i64 addrspace(1)* %gep0.cast
  %cast.load = bitcast i64 %load to <2 x float>
  ret <2 x float> %cast.load
}

define amdgpu_ps <2 x float> @global_load_saddr_i64_immneg128(i8 addrspace(1)* inreg %sbase, i32 %voffset) {
; GCN-LABEL: global_load_saddr_i64_immneg128:
; GCN:       ; %bb.0:
; GCN-NEXT:    global_load_dwordx2 v[0:1], v0, s[2:3] offset:-128
; GCN-NEXT:    s_waitcnt vmcnt(0)
; GCN-NEXT:    ; return to shader part epilog
  %zext.offset = zext i32 %voffset to i64
  %gep0 = getelementptr inbounds i8, i8 addrspace(1)* %sbase, i64 %zext.offset
  %gep1 = getelementptr inbounds i8, i8 addrspace(1)* %gep0, i64 -128
  %gep1.cast = bitcast i8 addrspace(1)* %gep1 to i64 addrspace(1)*
  %load = load i64, i64 addrspace(1)* %gep1.cast
  %cast.load = bitcast i64 %load to <2 x float>
  ret <2 x float> %cast.load
}

define amdgpu_ps <2 x float> @global_load_saddr_v2f32(i8 addrspace(1)* inreg %sbase, i32 %voffset) {
; GCN-LABEL: global_load_saddr_v2f32:
; GCN:       ; %bb.0:
; GCN-NEXT:    global_load_dwordx2 v[0:1], v0, s[2:3]
; GCN-NEXT:    s_waitcnt vmcnt(0)
; GCN-NEXT:    ; return to shader part epilog
  %zext.offset = zext i32 %voffset to i64
  %gep0 = getelementptr inbounds i8, i8 addrspace(1)* %sbase, i64 %zext.offset
  %gep0.cast = bitcast i8 addrspace(1)* %gep0 to <2 x float> addrspace(1)*
  %load = load <2 x float>, <2 x float> addrspace(1)* %gep0.cast
  ret <2 x float> %load
}

define amdgpu_ps <2 x float> @global_load_saddr_v2f32_immneg128(i8 addrspace(1)* inreg %sbase, i32 %voffset) {
; GCN-LABEL: global_load_saddr_v2f32_immneg128:
; GCN:       ; %bb.0:
; GCN-NEXT:    global_load_dwordx2 v[0:1], v0, s[2:3] offset:-128
; GCN-NEXT:    s_waitcnt vmcnt(0)
; GCN-NEXT:    ; return to shader part epilog
  %zext.offset = zext i32 %voffset to i64
  %gep0 = getelementptr inbounds i8, i8 addrspace(1)* %sbase, i64 %zext.offset
  %gep1 = getelementptr inbounds i8, i8 addrspace(1)* %gep0, i64 -128
  %gep1.cast = bitcast i8 addrspace(1)* %gep1 to <2 x float> addrspace(1)*
  %load = load <2 x float>, <2 x float> addrspace(1)* %gep1.cast
  ret <2 x float> %load
}

define amdgpu_ps <2 x float> @global_load_saddr_v2i32(i8 addrspace(1)* inreg %sbase, i32 %voffset) {
; GCN-LABEL: global_load_saddr_v2i32:
; GCN:       ; %bb.0:
; GCN-NEXT:    global_load_dwordx2 v[0:1], v0, s[2:3]
; GCN-NEXT:    s_waitcnt vmcnt(0)
; GCN-NEXT:    ; return to shader part epilog
  %zext.offset = zext i32 %voffset to i64
  %gep0 = getelementptr inbounds i8, i8 addrspace(1)* %sbase, i64 %zext.offset
  %gep0.cast = bitcast i8 addrspace(1)* %gep0 to <2 x i32> addrspace(1)*
  %load = load <2 x i32>, <2 x i32> addrspace(1)* %gep0.cast
  %cast.load = bitcast <2 x i32> %load to <2 x float>
  ret <2 x float> %cast.load
}

define amdgpu_ps <2 x float> @global_load_saddr_v2i32_immneg128(i8 addrspace(1)* inreg %sbase, i32 %voffset) {
; GCN-LABEL: global_load_saddr_v2i32_immneg128:
; GCN:       ; %bb.0:
; GCN-NEXT:    global_load_dwordx2 v[0:1], v0, s[2:3] offset:-128
; GCN-NEXT:    s_waitcnt vmcnt(0)
; GCN-NEXT:    ; return to shader part epilog
  %zext.offset = zext i32 %voffset to i64
  %gep0 = getelementptr inbounds i8, i8 addrspace(1)* %sbase, i64 %zext.offset
  %gep1 = getelementptr inbounds i8, i8 addrspace(1)* %gep0, i64 -128
  %gep1.cast = bitcast i8 addrspace(1)* %gep1 to <2 x i32> addrspace(1)*
  %load = load <2 x i32>, <2 x i32> addrspace(1)* %gep1.cast
  %cast.load = bitcast <2 x i32> %load to <2 x float>
  ret <2 x float> %cast.load
}

define amdgpu_ps <2 x float> @global_load_saddr_v4i16(i8 addrspace(1)* inreg %sbase, i32 %voffset) {
; GCN-LABEL: global_load_saddr_v4i16:
; GCN:       ; %bb.0:
; GCN-NEXT:    global_load_dwordx2 v[0:1], v0, s[2:3]
; GCN-NEXT:    s_waitcnt vmcnt(0)
; GCN-NEXT:    ; return to shader part epilog
  %zext.offset = zext i32 %voffset to i64
  %gep0 = getelementptr inbounds i8, i8 addrspace(1)* %sbase, i64 %zext.offset
  %gep0.cast = bitcast i8 addrspace(1)* %gep0 to <4 x i16> addrspace(1)*
  %load = load <4 x i16>, <4 x i16> addrspace(1)* %gep0.cast
  %cast.load = bitcast <4 x i16> %load to <2 x float>
  ret <2 x float> %cast.load
}

define amdgpu_ps <2 x float> @global_load_saddr_v4i16_immneg128(i8 addrspace(1)* inreg %sbase, i32 %voffset) {
; GCN-LABEL: global_load_saddr_v4i16_immneg128:
; GCN:       ; %bb.0:
; GCN-NEXT:    global_load_dwordx2 v[0:1], v0, s[2:3] offset:-128
; GCN-NEXT:    s_waitcnt vmcnt(0)
; GCN-NEXT:    ; return to shader part epilog
  %zext.offset = zext i32 %voffset to i64
  %gep0 = getelementptr inbounds i8, i8 addrspace(1)* %sbase, i64 %zext.offset
  %gep1 = getelementptr inbounds i8, i8 addrspace(1)* %gep0, i64 -128
  %gep1.cast = bitcast i8 addrspace(1)* %gep1 to <4 x i16> addrspace(1)*
  %load = load <4 x i16>, <4 x i16> addrspace(1)* %gep1.cast
  %cast.load = bitcast <4 x i16> %load to <2 x float>
  ret <2 x float> %cast.load
}

define amdgpu_ps <2 x float> @global_load_saddr_v4f16(i8 addrspace(1)* inreg %sbase, i32 %voffset) {
; GCN-LABEL: global_load_saddr_v4f16:
; GCN:       ; %bb.0:
; GCN-NEXT:    global_load_dwordx2 v[0:1], v0, s[2:3]
; GCN-NEXT:    s_waitcnt vmcnt(0)
; GCN-NEXT:    ; return to shader part epilog
  %zext.offset = zext i32 %voffset to i64
  %gep0 = getelementptr inbounds i8, i8 addrspace(1)* %sbase, i64 %zext.offset
  %gep0.cast = bitcast i8 addrspace(1)* %gep0 to <4 x half> addrspace(1)*
  %load = load <4 x half>, <4 x half> addrspace(1)* %gep0.cast
  %cast.load = bitcast <4 x half> %load to <2 x float>
  ret <2 x float> %cast.load
}

define amdgpu_ps <2 x float> @global_load_saddr_v4f16_immneg128(i8 addrspace(1)* inreg %sbase, i32 %voffset) {
; GCN-LABEL: global_load_saddr_v4f16_immneg128:
; GCN:       ; %bb.0:
; GCN-NEXT:    global_load_dwordx2 v[0:1], v0, s[2:3] offset:-128
; GCN-NEXT:    s_waitcnt vmcnt(0)
; GCN-NEXT:    ; return to shader part epilog
  %zext.offset = zext i32 %voffset to i64
  %gep0 = getelementptr inbounds i8, i8 addrspace(1)* %sbase, i64 %zext.offset
  %gep1 = getelementptr inbounds i8, i8 addrspace(1)* %gep0, i64 -128
  %gep1.cast = bitcast i8 addrspace(1)* %gep1 to <4 x half> addrspace(1)*
  %load = load <4 x half>, <4 x half> addrspace(1)* %gep1.cast
  %cast.load = bitcast <4 x half> %load to <2 x float>
  ret <2 x float> %cast.load
}

define amdgpu_ps <2 x float> @global_load_saddr_p1(i8 addrspace(1)* inreg %sbase, i32 %voffset) {
; GCN-LABEL: global_load_saddr_p1:
; GCN:       ; %bb.0:
; GCN-NEXT:    global_load_dwordx2 v[0:1], v0, s[2:3]
; GCN-NEXT:    s_waitcnt vmcnt(0)
; GCN-NEXT:    ; return to shader part epilog
  %zext.offset = zext i32 %voffset to i64
  %gep0 = getelementptr inbounds i8, i8 addrspace(1)* %sbase, i64 %zext.offset
  %gep0.cast = bitcast i8 addrspace(1)* %gep0 to i8 addrspace(1)* addrspace(1)*
  %load = load i8 addrspace(1)*, i8 addrspace(1)* addrspace(1)* %gep0.cast
  %cast.load0 = ptrtoint i8 addrspace(1)* %load to i64
  %cast.load1 = bitcast i64 %cast.load0 to <2 x float>
  ret <2 x float> %cast.load1
}

define amdgpu_ps <2 x float> @global_load_saddr_p1_immneg128(i8 addrspace(1)* inreg %sbase, i32 %voffset) {
; GCN-LABEL: global_load_saddr_p1_immneg128:
; GCN:       ; %bb.0:
; GCN-NEXT:    global_load_dwordx2 v[0:1], v0, s[2:3] offset:-128
; GCN-NEXT:    s_waitcnt vmcnt(0)
; GCN-NEXT:    ; return to shader part epilog
  %zext.offset = zext i32 %voffset to i64
  %gep0 = getelementptr inbounds i8, i8 addrspace(1)* %sbase, i64 %zext.offset
  %gep1 = getelementptr inbounds i8, i8 addrspace(1)* %gep0, i64 -128
  %gep1.cast = bitcast i8 addrspace(1)* %gep1 to i8 addrspace(1)* addrspace(1)*
  %load = load i8 addrspace(1)*, i8 addrspace(1)* addrspace(1)* %gep1.cast
  %cast.load0 = ptrtoint i8 addrspace(1)* %load to i64
  %cast.load1 = bitcast i64 %cast.load0 to <2 x float>
  ret <2 x float> %cast.load1
}

define amdgpu_ps <3 x float> @global_load_saddr_v3f32(i8 addrspace(1)* inreg %sbase, i32 %voffset) {
; GCN-LABEL: global_load_saddr_v3f32:
; GCN:       ; %bb.0:
; GCN-NEXT:    global_load_dwordx3 v[0:2], v0, s[2:3]
; GCN-NEXT:    s_waitcnt vmcnt(0)
; GCN-NEXT:    ; return to shader part epilog
  %zext.offset = zext i32 %voffset to i64
  %gep0 = getelementptr inbounds i8, i8 addrspace(1)* %sbase, i64 %zext.offset
  %gep0.cast = bitcast i8 addrspace(1)* %gep0 to <3 x float> addrspace(1)*
  %load = load <3 x float>, <3 x float> addrspace(1)* %gep0.cast
  ret <3 x float> %load
}

define amdgpu_ps <3 x float> @global_load_saddr_v3f32_immneg128(i8 addrspace(1)* inreg %sbase, i32 %voffset) {
; GCN-LABEL: global_load_saddr_v3f32_immneg128:
; GCN:       ; %bb.0:
; GCN-NEXT:    global_load_dwordx3 v[0:2], v0, s[2:3] offset:-128
; GCN-NEXT:    s_waitcnt vmcnt(0)
; GCN-NEXT:    ; return to shader part epilog
  %zext.offset = zext i32 %voffset to i64
  %gep0 = getelementptr inbounds i8, i8 addrspace(1)* %sbase, i64 %zext.offset
  %gep1 = getelementptr inbounds i8, i8 addrspace(1)* %gep0, i64 -128
  %gep1.cast = bitcast i8 addrspace(1)* %gep1 to <3 x float> addrspace(1)*
  %load = load <3 x float>, <3 x float> addrspace(1)* %gep1.cast
  ret <3 x float> %load
}

define amdgpu_ps <3 x float> @global_load_saddr_v3i32(i8 addrspace(1)* inreg %sbase, i32 %voffset) {
; GCN-LABEL: global_load_saddr_v3i32:
; GCN:       ; %bb.0:
; GCN-NEXT:    global_load_dwordx3 v[0:2], v0, s[2:3]
; GCN-NEXT:    s_waitcnt vmcnt(0)
; GCN-NEXT:    ; return to shader part epilog
  %zext.offset = zext i32 %voffset to i64
  %gep0 = getelementptr inbounds i8, i8 addrspace(1)* %sbase, i64 %zext.offset
  %gep0.cast = bitcast i8 addrspace(1)* %gep0 to <3 x i32> addrspace(1)*
  %load = load <3 x i32>, <3 x i32> addrspace(1)* %gep0.cast
  %cast.load = bitcast <3 x i32> %load to <3 x float>
  ret <3 x float> %cast.load
}

define amdgpu_ps <3 x float> @global_load_saddr_v3i32_immneg128(i8 addrspace(1)* inreg %sbase, i32 %voffset) {
; GCN-LABEL: global_load_saddr_v3i32_immneg128:
; GCN:       ; %bb.0:
; GCN-NEXT:    global_load_dwordx3 v[0:2], v0, s[2:3] offset:-128
; GCN-NEXT:    s_waitcnt vmcnt(0)
; GCN-NEXT:    ; return to shader part epilog
  %zext.offset = zext i32 %voffset to i64
  %gep0 = getelementptr inbounds i8, i8 addrspace(1)* %sbase, i64 %zext.offset
  %gep1 = getelementptr inbounds i8, i8 addrspace(1)* %gep0, i64 -128
  %gep1.cast = bitcast i8 addrspace(1)* %gep1 to <3 x i32> addrspace(1)*
  %load = load <3 x i32>, <3 x i32> addrspace(1)* %gep1.cast
  %cast.load = bitcast <3 x i32> %load to <3 x float>
  ret <3 x float> %cast.load
}

define amdgpu_ps <6 x half> @global_load_saddr_v6f16(i8 addrspace(1)* inreg %sbase, i32 %voffset) {
; GCN-LABEL: global_load_saddr_v6f16:
; GCN:       ; %bb.0:
; GCN-NEXT:    global_load_dwordx3 v[0:2], v0, s[2:3]
; GCN-NEXT:    s_waitcnt vmcnt(0)
; GCN-NEXT:    ; return to shader part epilog
  %zext.offset = zext i32 %voffset to i64
  %gep0 = getelementptr inbounds i8, i8 addrspace(1)* %sbase, i64 %zext.offset
  %gep0.cast = bitcast i8 addrspace(1)* %gep0 to <6 x half> addrspace(1)*
  %load = load <6 x half>, <6 x half> addrspace(1)* %gep0.cast
  ret <6 x half> %load
}

define amdgpu_ps <6 x half> @global_load_saddr_v6f16_immneg128(i8 addrspace(1)* inreg %sbase, i32 %voffset) {
; GCN-LABEL: global_load_saddr_v6f16_immneg128:
; GCN:       ; %bb.0:
; GCN-NEXT:    global_load_dwordx3 v[0:2], v0, s[2:3] offset:-128
; GCN-NEXT:    s_waitcnt vmcnt(0)
; GCN-NEXT:    ; return to shader part epilog
  %zext.offset = zext i32 %voffset to i64
  %gep0 = getelementptr inbounds i8, i8 addrspace(1)* %sbase, i64 %zext.offset
  %gep1 = getelementptr inbounds i8, i8 addrspace(1)* %gep0, i64 -128
  %gep1.cast = bitcast i8 addrspace(1)* %gep1 to <6 x half> addrspace(1)*
  %load = load <6 x half>, <6 x half> addrspace(1)* %gep1.cast
  ret <6 x half> %load
}

define amdgpu_ps <4 x float> @global_load_saddr_v4f32(i8 addrspace(1)* inreg %sbase, i32 %voffset) {
; GCN-LABEL: global_load_saddr_v4f32:
; GCN:       ; %bb.0:
; GCN-NEXT:    global_load_dwordx4 v[0:3], v0, s[2:3]
; GCN-NEXT:    s_waitcnt vmcnt(0)
; GCN-NEXT:    ; return to shader part epilog
  %zext.offset = zext i32 %voffset to i64
  %gep0 = getelementptr inbounds i8, i8 addrspace(1)* %sbase, i64 %zext.offset
  %gep0.cast = bitcast i8 addrspace(1)* %gep0 to <4 x float> addrspace(1)*
  %load = load <4 x float>, <4 x float> addrspace(1)* %gep0.cast
  ret <4 x float> %load
}

define amdgpu_ps <4 x float> @global_load_saddr_v4f32_immneg128(i8 addrspace(1)* inreg %sbase, i32 %voffset) {
; GCN-LABEL: global_load_saddr_v4f32_immneg128:
; GCN:       ; %bb.0:
; GCN-NEXT:    global_load_dwordx4 v[0:3], v0, s[2:3] offset:-128
; GCN-NEXT:    s_waitcnt vmcnt(0)
; GCN-NEXT:    ; return to shader part epilog
  %zext.offset = zext i32 %voffset to i64
  %gep0 = getelementptr inbounds i8, i8 addrspace(1)* %sbase, i64 %zext.offset
  %gep1 = getelementptr inbounds i8, i8 addrspace(1)* %gep0, i64 -128
  %gep1.cast = bitcast i8 addrspace(1)* %gep1 to <4 x float> addrspace(1)*
  %load = load <4 x float>, <4 x float> addrspace(1)* %gep1.cast
  ret <4 x float> %load
}

define amdgpu_ps <4 x float> @global_load_saddr_v4i32(i8 addrspace(1)* inreg %sbase, i32 %voffset) {
; GCN-LABEL: global_load_saddr_v4i32:
; GCN:       ; %bb.0:
; GCN-NEXT:    global_load_dwordx4 v[0:3], v0, s[2:3]
; GCN-NEXT:    s_waitcnt vmcnt(0)
; GCN-NEXT:    ; return to shader part epilog
  %zext.offset = zext i32 %voffset to i64
  %gep0 = getelementptr inbounds i8, i8 addrspace(1)* %sbase, i64 %zext.offset
  %gep0.cast = bitcast i8 addrspace(1)* %gep0 to <4 x i32> addrspace(1)*
  %load = load <4 x i32>, <4 x i32> addrspace(1)* %gep0.cast
  %cast.load = bitcast <4 x i32> %load to <4 x float>
  ret <4 x float> %cast.load
}

define amdgpu_ps <4 x float> @global_load_saddr_v4i32_immneg128(i8 addrspace(1)* inreg %sbase, i32 %voffset) {
; GCN-LABEL: global_load_saddr_v4i32_immneg128:
; GCN:       ; %bb.0:
; GCN-NEXT:    global_load_dwordx4 v[0:3], v0, s[2:3] offset:-128
; GCN-NEXT:    s_waitcnt vmcnt(0)
; GCN-NEXT:    ; return to shader part epilog
  %zext.offset = zext i32 %voffset to i64
  %gep0 = getelementptr inbounds i8, i8 addrspace(1)* %sbase, i64 %zext.offset
  %gep1 = getelementptr inbounds i8, i8 addrspace(1)* %gep0, i64 -128
  %gep1.cast = bitcast i8 addrspace(1)* %gep1 to <4 x i32> addrspace(1)*
  %load = load <4 x i32>, <4 x i32> addrspace(1)* %gep1.cast
  %cast.load = bitcast <4 x i32> %load to <4 x float>
  ret <4 x float> %cast.load
}

define amdgpu_ps <4 x float> @global_load_saddr_v2i64(i8 addrspace(1)* inreg %sbase, i32 %voffset) {
; GCN-LABEL: global_load_saddr_v2i64:
; GCN:       ; %bb.0:
; GCN-NEXT:    global_load_dwordx4 v[0:3], v0, s[2:3]
; GCN-NEXT:    s_waitcnt vmcnt(0)
; GCN-NEXT:    ; return to shader part epilog
  %zext.offset = zext i32 %voffset to i64
  %gep0 = getelementptr inbounds i8, i8 addrspace(1)* %sbase, i64 %zext.offset
  %gep0.cast = bitcast i8 addrspace(1)* %gep0 to <2 x i64> addrspace(1)*
  %load = load <2 x i64>, <2 x i64> addrspace(1)* %gep0.cast
  %cast.load = bitcast <2 x i64> %load to <4 x float>
  ret <4 x float> %cast.load
}

define amdgpu_ps <4 x float> @global_load_saddr_v2i64_immneg128(i8 addrspace(1)* inreg %sbase, i32 %voffset) {
; GCN-LABEL: global_load_saddr_v2i64_immneg128:
; GCN:       ; %bb.0:
; GCN-NEXT:    global_load_dwordx4 v[0:3], v0, s[2:3] offset:-128
; GCN-NEXT:    s_waitcnt vmcnt(0)
; GCN-NEXT:    ; return to shader part epilog
  %zext.offset = zext i32 %voffset to i64
  %gep0 = getelementptr inbounds i8, i8 addrspace(1)* %sbase, i64 %zext.offset
  %gep1 = getelementptr inbounds i8, i8 addrspace(1)* %gep0, i64 -128
  %gep1.cast = bitcast i8 addrspace(1)* %gep1 to <2 x i64> addrspace(1)*
  %load = load <2 x i64>, <2 x i64> addrspace(1)* %gep1.cast
  %cast.load = bitcast <2 x i64> %load to <4 x float>
  ret <4 x float> %cast.load
}

define amdgpu_ps <4 x float> @global_load_saddr_i128(i8 addrspace(1)* inreg %sbase, i32 %voffset) {
; GCN-LABEL: global_load_saddr_i128:
; GCN:       ; %bb.0:
; GCN-NEXT:    global_load_dwordx4 v[0:3], v0, s[2:3]
; GCN-NEXT:    s_waitcnt vmcnt(0)
; GCN-NEXT:    ; return to shader part epilog
  %zext.offset = zext i32 %voffset to i64
  %gep0 = getelementptr inbounds i8, i8 addrspace(1)* %sbase, i64 %zext.offset
  %gep0.cast = bitcast i8 addrspace(1)* %gep0 to i128 addrspace(1)*
  %load = load i128, i128 addrspace(1)* %gep0.cast
  %cast.load = bitcast i128 %load to <4 x float>
  ret <4 x float> %cast.load
}

define amdgpu_ps <4 x float> @global_load_saddr_i128_immneg128(i8 addrspace(1)* inreg %sbase, i32 %voffset) {
; GCN-LABEL: global_load_saddr_i128_immneg128:
; GCN:       ; %bb.0:
; GCN-NEXT:    global_load_dwordx4 v[0:3], v0, s[2:3] offset:-128
; GCN-NEXT:    s_waitcnt vmcnt(0)
; GCN-NEXT:    ; return to shader part epilog
  %zext.offset = zext i32 %voffset to i64
  %gep0 = getelementptr inbounds i8, i8 addrspace(1)* %sbase, i64 %zext.offset
  %gep1 = getelementptr inbounds i8, i8 addrspace(1)* %gep0, i64 -128
  %gep1.cast = bitcast i8 addrspace(1)* %gep1 to i128 addrspace(1)*
  %load = load i128, i128 addrspace(1)* %gep1.cast
  %cast.load = bitcast i128 %load to <4 x float>
  ret <4 x float> %cast.load
}

define amdgpu_ps <4 x float> @global_load_saddr_v2p1(i8 addrspace(1)* inreg %sbase, i32 %voffset) {
; GCN-LABEL: global_load_saddr_v2p1:
; GCN:       ; %bb.0:
; GCN-NEXT:    global_load_dwordx4 v[0:3], v0, s[2:3]
; GCN-NEXT:    s_waitcnt vmcnt(0)
; GCN-NEXT:    ; return to shader part epilog
  %zext.offset = zext i32 %voffset to i64
  %gep0 = getelementptr inbounds i8, i8 addrspace(1)* %sbase, i64 %zext.offset
  %gep0.cast = bitcast i8 addrspace(1)* %gep0 to <2 x i8 addrspace(1)*> addrspace(1)*
  %load = load <2 x i8 addrspace(1)*>, <2 x i8 addrspace(1)*> addrspace(1)* %gep0.cast
  %cast.load0 = ptrtoint <2 x i8 addrspace(1)*> %load to <2 x i64>
  %cast.load1 = bitcast <2 x i64> %cast.load0 to <4 x float>
  ret <4 x float> %cast.load1
}

define amdgpu_ps <4 x float> @global_load_saddr_v2p1_immneg128(i8 addrspace(1)* inreg %sbase, i32 %voffset) {
; GCN-LABEL: global_load_saddr_v2p1_immneg128:
; GCN:       ; %bb.0:
; GCN-NEXT:    global_load_dwordx4 v[0:3], v0, s[2:3] offset:-128
; GCN-NEXT:    s_waitcnt vmcnt(0)
; GCN-NEXT:    ; return to shader part epilog
  %zext.offset = zext i32 %voffset to i64
  %gep0 = getelementptr inbounds i8, i8 addrspace(1)* %sbase, i64 %zext.offset
  %gep1 = getelementptr inbounds i8, i8 addrspace(1)* %gep0, i64 -128
  %gep1.cast = bitcast i8 addrspace(1)* %gep1 to <2 x i8 addrspace(1)*> addrspace(1)*
  %load = load <2 x i8 addrspace(1)*>, <2 x i8 addrspace(1)*> addrspace(1)* %gep1.cast
  %cast.load0 = ptrtoint <2 x i8 addrspace(1)*> %load to <2 x i64>
  %cast.load1 = bitcast <2 x i64> %cast.load0 to <4 x float>
  ret <4 x float> %cast.load1
}

define amdgpu_ps <4 x float> @global_load_saddr_v4p3(i8 addrspace(1)* inreg %sbase, i32 %voffset) {
; GCN-LABEL: global_load_saddr_v4p3:
; GCN:       ; %bb.0:
; GCN-NEXT:    global_load_dwordx4 v[0:3], v0, s[2:3]
; GCN-NEXT:    s_waitcnt vmcnt(0)
; GCN-NEXT:    ; return to shader part epilog
  %zext.offset = zext i32 %voffset to i64
  %gep0 = getelementptr inbounds i8, i8 addrspace(1)* %sbase, i64 %zext.offset
  %gep0.cast = bitcast i8 addrspace(1)* %gep0 to <4 x i8 addrspace(3)*> addrspace(1)*
  %load = load <4 x i8 addrspace(3)*>, <4 x i8 addrspace(3)*> addrspace(1)* %gep0.cast
  %cast.load0 = ptrtoint <4 x i8 addrspace(3)*> %load to <4 x i32>
  %cast.load1 = bitcast <4 x i32> %cast.load0 to <4 x float>
  ret <4 x float> %cast.load1
}

define amdgpu_ps <4 x float> @global_load_saddr_v4p3_immneg128(i8 addrspace(1)* inreg %sbase, i32 %voffset) {
; GCN-LABEL: global_load_saddr_v4p3_immneg128:
; GCN:       ; %bb.0:
; GCN-NEXT:    global_load_dwordx4 v[0:3], v0, s[2:3] offset:-128
; GCN-NEXT:    s_waitcnt vmcnt(0)
; GCN-NEXT:    ; return to shader part epilog
  %zext.offset = zext i32 %voffset to i64
  %gep0 = getelementptr inbounds i8, i8 addrspace(1)* %sbase, i64 %zext.offset
  %gep1 = getelementptr inbounds i8, i8 addrspace(1)* %gep0, i64 -128
  %gep1.cast = bitcast i8 addrspace(1)* %gep1 to <4 x i8 addrspace(3)*> addrspace(1)*
  %load = load <4 x i8 addrspace(3)*>, <4 x i8 addrspace(3)*> addrspace(1)* %gep1.cast
  %cast.load0 = ptrtoint <4 x i8 addrspace(3)*> %load to <4 x i32>
  %cast.load1 = bitcast <4 x i32> %cast.load0 to <4 x float>
  ret <4 x float> %cast.load1
}

; --------------------------------------------------------------------------------
; Extending loads
; --------------------------------------------------------------------------------

define amdgpu_ps float @global_sextload_saddr_i8(i8 addrspace(1)* inreg %sbase, i32 %voffset) {
; GCN-LABEL: global_sextload_saddr_i8:
; GCN:       ; %bb.0:
; GCN-NEXT:    global_load_sbyte v0, v0, s[2:3]
; GCN-NEXT:    s_waitcnt vmcnt(0)
; GCN-NEXT:    ; return to shader part epilog
  %zext.offset = zext i32 %voffset to i64
  %gep0 = getelementptr inbounds i8, i8 addrspace(1)* %sbase, i64 %zext.offset
  %load = load i8, i8 addrspace(1)* %gep0
  %sextload = sext i8 %load to i32
  %cast.load = bitcast i32 %sextload to float
  ret float %cast.load
}

define amdgpu_ps float @global_sextload_saddr_i8_immneg128(i8 addrspace(1)* inreg %sbase, i32 %voffset) {
; GCN-LABEL: global_sextload_saddr_i8_immneg128:
; GCN:       ; %bb.0:
; GCN-NEXT:    global_load_sbyte v0, v0, s[2:3] offset:-128
; GCN-NEXT:    s_waitcnt vmcnt(0)
; GCN-NEXT:    ; return to shader part epilog
  %zext.offset = zext i32 %voffset to i64
  %gep0 = getelementptr inbounds i8, i8 addrspace(1)* %sbase, i64 %zext.offset
  %gep1 = getelementptr inbounds i8, i8 addrspace(1)* %gep0, i64 -128
  %load = load i8, i8 addrspace(1)* %gep1
  %sextload = sext i8 %load to i32
  %cast.load = bitcast i32 %sextload to float
  ret float %cast.load
}

define amdgpu_ps float @global_sextload_saddr_i16(i8 addrspace(1)* inreg %sbase, i32 %voffset) {
; GCN-LABEL: global_sextload_saddr_i16:
; GCN:       ; %bb.0:
; GCN-NEXT:    global_load_sshort v0, v0, s[2:3]
; GCN-NEXT:    s_waitcnt vmcnt(0)
; GCN-NEXT:    ; return to shader part epilog
  %zext.offset = zext i32 %voffset to i64
  %gep0 = getelementptr inbounds i8, i8 addrspace(1)* %sbase, i64 %zext.offset
  %gep0.cast = bitcast i8 addrspace(1)* %gep0 to i16 addrspace(1)*
  %load = load i16, i16 addrspace(1)* %gep0.cast
  %sextload = sext i16 %load to i32
  %cast.load = bitcast i32 %sextload to float
  ret float %cast.load
}

define amdgpu_ps float @global_sextload_saddr_i16_immneg128(i8 addrspace(1)* inreg %sbase, i32 %voffset) {
; GCN-LABEL: global_sextload_saddr_i16_immneg128:
; GCN:       ; %bb.0:
; GCN-NEXT:    global_load_sshort v0, v0, s[2:3] offset:-128
; GCN-NEXT:    s_waitcnt vmcnt(0)
; GCN-NEXT:    ; return to shader part epilog
  %zext.offset = zext i32 %voffset to i64
  %gep0 = getelementptr inbounds i8, i8 addrspace(1)* %sbase, i64 %zext.offset
  %gep1 = getelementptr inbounds i8, i8 addrspace(1)* %gep0, i64 -128
  %gep1.cast = bitcast i8 addrspace(1)* %gep1 to i16 addrspace(1)*
  %load = load i16, i16 addrspace(1)* %gep1.cast
  %sextload = sext i16 %load to i32
  %cast.load = bitcast i32 %sextload to float
  ret float %cast.load
}

define amdgpu_ps float @global_zextload_saddr_i8(i8 addrspace(1)* inreg %sbase, i32 %voffset) {
; GCN-LABEL: global_zextload_saddr_i8:
; GCN:       ; %bb.0:
; GCN-NEXT:    global_load_ubyte v0, v0, s[2:3]
; GCN-NEXT:    s_waitcnt vmcnt(0)
; GCN-NEXT:    ; return to shader part epilog
  %zext.offset = zext i32 %voffset to i64
  %gep0 = getelementptr inbounds i8, i8 addrspace(1)* %sbase, i64 %zext.offset
  %load = load i8, i8 addrspace(1)* %gep0
  %zextload = zext i8 %load to i32
  %cast.load = bitcast i32 %zextload to float
  ret float %cast.load
}

define amdgpu_ps float @global_zextload_saddr_i8_immneg128(i8 addrspace(1)* inreg %sbase, i32 %voffset) {
; GCN-LABEL: global_zextload_saddr_i8_immneg128:
; GCN:       ; %bb.0:
; GCN-NEXT:    global_load_ubyte v0, v0, s[2:3] offset:-128
; GCN-NEXT:    s_waitcnt vmcnt(0)
; GCN-NEXT:    ; return to shader part epilog
  %zext.offset = zext i32 %voffset to i64
  %gep0 = getelementptr inbounds i8, i8 addrspace(1)* %sbase, i64 %zext.offset
  %gep1 = getelementptr inbounds i8, i8 addrspace(1)* %gep0, i64 -128
  %load = load i8, i8 addrspace(1)* %gep1
  %zextload = zext i8 %load to i32
  %cast.load = bitcast i32 %zextload to float
  ret float %cast.load
}

define amdgpu_ps float @global_zextload_saddr_i16(i8 addrspace(1)* inreg %sbase, i32 %voffset) {
; GCN-LABEL: global_zextload_saddr_i16:
; GCN:       ; %bb.0:
; GCN-NEXT:    global_load_ushort v0, v0, s[2:3]
; GCN-NEXT:    s_waitcnt vmcnt(0)
; GCN-NEXT:    ; return to shader part epilog
  %zext.offset = zext i32 %voffset to i64
  %gep0 = getelementptr inbounds i8, i8 addrspace(1)* %sbase, i64 %zext.offset
  %gep0.cast = bitcast i8 addrspace(1)* %gep0 to i16 addrspace(1)*
  %load = load i16, i16 addrspace(1)* %gep0.cast
  %zextload = zext i16 %load to i32
  %cast.load = bitcast i32 %zextload to float
  ret float %cast.load
}

define amdgpu_ps float @global_zextload_saddr_i16_immneg128(i8 addrspace(1)* inreg %sbase, i32 %voffset) {
; GCN-LABEL: global_zextload_saddr_i16_immneg128:
; GCN:       ; %bb.0:
; GCN-NEXT:    global_load_ushort v0, v0, s[2:3] offset:-128
; GCN-NEXT:    s_waitcnt vmcnt(0)
; GCN-NEXT:    ; return to shader part epilog
  %zext.offset = zext i32 %voffset to i64
  %gep0 = getelementptr inbounds i8, i8 addrspace(1)* %sbase, i64 %zext.offset
  %gep1 = getelementptr inbounds i8, i8 addrspace(1)* %gep0, i64 -128
  %gep1.cast = bitcast i8 addrspace(1)* %gep1 to i16 addrspace(1)*
  %load = load i16, i16 addrspace(1)* %gep1.cast
  %zextload = zext i16 %load to i32
  %cast.load = bitcast i32 %zextload to float
  ret float %cast.load
}

; --------------------------------------------------------------------------------
; Atomic load
; --------------------------------------------------------------------------------

define amdgpu_ps float @atomic_global_load_saddr_i32(i8 addrspace(1)* inreg %sbase, i32 %voffset) {
; GFX9-LABEL: atomic_global_load_saddr_i32:
; GFX9:       ; %bb.0:
; GFX9-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GFX9-NEXT:    global_load_dword v0, v0, s[2:3] glc
; GFX9-NEXT:    s_waitcnt vmcnt(0)
; GFX9-NEXT:    buffer_wbinvl1
; GFX9-NEXT:    ; return to shader part epilog
;
; GFX10-LABEL: atomic_global_load_saddr_i32:
; GFX10:       ; %bb.0:
; GFX10-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
; GFX10-NEXT:    global_load_dword v0, v0, s[2:3] glc dlc
; GFX10-NEXT:    s_waitcnt vmcnt(0)
; GFX10-NEXT:    buffer_gl0_inv
; GFX10-NEXT:    buffer_gl1_inv
; GFX10-NEXT:    ; return to shader part epilog
  %zext.offset = zext i32 %voffset to i64
  %gep0 = getelementptr inbounds i8, i8 addrspace(1)* %sbase, i64 %zext.offset
  %gep0.cast = bitcast i8 addrspace(1)* %gep0 to i32 addrspace(1)*
  %load = load atomic i32, i32 addrspace(1)* %gep0.cast seq_cst, align 4
  %cast.load = bitcast i32 %load to float
  ret float %cast.load
}

define amdgpu_ps float @atomic_global_load_saddr_i32_immneg128(i8 addrspace(1)* inreg %sbase, i32 %voffset) {
; GFX9-LABEL: atomic_global_load_saddr_i32_immneg128:
; GFX9:       ; %bb.0:
; GFX9-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GFX9-NEXT:    global_load_dword v0, v0, s[2:3] offset:-128 glc
; GFX9-NEXT:    s_waitcnt vmcnt(0)
; GFX9-NEXT:    buffer_wbinvl1
; GFX9-NEXT:    ; return to shader part epilog
;
; GFX10-LABEL: atomic_global_load_saddr_i32_immneg128:
; GFX10:       ; %bb.0:
; GFX10-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
; GFX10-NEXT:    global_load_dword v0, v0, s[2:3] offset:-128 glc dlc
; GFX10-NEXT:    s_waitcnt vmcnt(0)
; GFX10-NEXT:    buffer_gl0_inv
; GFX10-NEXT:    buffer_gl1_inv
; GFX10-NEXT:    ; return to shader part epilog
  %zext.offset = zext i32 %voffset to i64
  %gep0 = getelementptr inbounds i8, i8 addrspace(1)* %sbase, i64 %zext.offset
  %gep1 = getelementptr inbounds i8, i8 addrspace(1)* %gep0, i64 -128
  %gep1.cast = bitcast i8 addrspace(1)* %gep1 to i32 addrspace(1)*
  %load = load atomic i32, i32 addrspace(1)* %gep1.cast seq_cst, align 4
  %cast.load = bitcast i32 %load to float
  ret float %cast.load
}

define amdgpu_ps <2 x float> @atomic_global_load_saddr_i64(i8 addrspace(1)* inreg %sbase, i32 %voffset) {
; GFX9-LABEL: atomic_global_load_saddr_i64:
; GFX9:       ; %bb.0:
; GFX9-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GFX9-NEXT:    global_load_dwordx2 v[0:1], v0, s[2:3] glc
; GFX9-NEXT:    s_waitcnt vmcnt(0)
; GFX9-NEXT:    buffer_wbinvl1
; GFX9-NEXT:    ; return to shader part epilog
;
; GFX10-LABEL: atomic_global_load_saddr_i64:
; GFX10:       ; %bb.0:
; GFX10-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
; GFX10-NEXT:    global_load_dwordx2 v[0:1], v0, s[2:3] glc dlc
; GFX10-NEXT:    s_waitcnt vmcnt(0)
; GFX10-NEXT:    buffer_gl0_inv
; GFX10-NEXT:    buffer_gl1_inv
; GFX10-NEXT:    ; return to shader part epilog
  %zext.offset = zext i32 %voffset to i64
  %gep0 = getelementptr inbounds i8, i8 addrspace(1)* %sbase, i64 %zext.offset
  %gep0.cast = bitcast i8 addrspace(1)* %gep0 to i64 addrspace(1)*
  %load = load atomic i64, i64 addrspace(1)* %gep0.cast seq_cst, align 8
  %cast.load = bitcast i64 %load to <2 x float>
  ret <2 x float> %cast.load
}

define amdgpu_ps <2 x float> @atomic_global_load_saddr_i64_immneg128(i8 addrspace(1)* inreg %sbase, i32 %voffset) {
; GFX9-LABEL: atomic_global_load_saddr_i64_immneg128:
; GFX9:       ; %bb.0:
; GFX9-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GFX9-NEXT:    global_load_dwordx2 v[0:1], v0, s[2:3] offset:-128 glc
; GFX9-NEXT:    s_waitcnt vmcnt(0)
; GFX9-NEXT:    buffer_wbinvl1
; GFX9-NEXT:    ; return to shader part epilog
;
; GFX10-LABEL: atomic_global_load_saddr_i64_immneg128:
; GFX10:       ; %bb.0:
; GFX10-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
; GFX10-NEXT:    global_load_dwordx2 v[0:1], v0, s[2:3] offset:-128 glc dlc
; GFX10-NEXT:    s_waitcnt vmcnt(0)
; GFX10-NEXT:    buffer_gl0_inv
; GFX10-NEXT:    buffer_gl1_inv
; GFX10-NEXT:    ; return to shader part epilog
  %zext.offset = zext i32 %voffset to i64
  %gep0 = getelementptr inbounds i8, i8 addrspace(1)* %sbase, i64 %zext.offset
  %gep1 = getelementptr inbounds i8, i8 addrspace(1)* %gep0, i64 -128
  %gep1.cast = bitcast i8 addrspace(1)* %gep1 to i64 addrspace(1)*
  %load = load atomic i64, i64 addrspace(1)* %gep1.cast seq_cst, align 8
  %cast.load = bitcast i64 %load to <2 x float>
  ret <2 x float> %cast.load
}

; --------------------------------------------------------------------------------
; D16 load (low 16)
; --------------------------------------------------------------------------------

define amdgpu_ps <2 x half> @global_load_saddr_i16_d16lo_undef_hi(i8 addrspace(1)* inreg %sbase, i32 %voffset) {
; GCN-LABEL: global_load_saddr_i16_d16lo_undef_hi:
; GCN:       ; %bb.0:
; GCN-NEXT:    global_load_short_d16 v0, v0, s[2:3]
; GCN-NEXT:    s_waitcnt vmcnt(0)
; GCN-NEXT:    ; return to shader part epilog
  %zext.offset = zext i32 %voffset to i64
  %gep0 = getelementptr inbounds i8, i8 addrspace(1)* %sbase, i64 %zext.offset
  %gep0.cast = bitcast i8 addrspace(1)* %gep0 to i16 addrspace(1)*
  %load = load i16, i16 addrspace(1)* %gep0.cast
  %build = insertelement <2 x i16> undef, i16 %load, i32 0
  %cast = bitcast <2 x i16> %build to <2 x half>
  ret <2 x half> %cast
}

define amdgpu_ps <2 x half> @global_load_saddr_i16_d16lo_undef_hi_immneg128(i8 addrspace(1)* inreg %sbase, i32 %voffset) {
; GCN-LABEL: global_load_saddr_i16_d16lo_undef_hi_immneg128:
; GCN:       ; %bb.0:
; GCN-NEXT:    global_load_short_d16 v0, v0, s[2:3] offset:-128
; GCN-NEXT:    s_waitcnt vmcnt(0)
; GCN-NEXT:    ; return to shader part epilog
  %zext.offset = zext i32 %voffset to i64
  %gep0 = getelementptr inbounds i8, i8 addrspace(1)* %sbase, i64 %zext.offset
  %gep1 = getelementptr inbounds i8, i8 addrspace(1)* %gep0, i64 -128
  %gep1.cast = bitcast i8 addrspace(1)* %gep1 to i16 addrspace(1)*
  %load = load i16, i16 addrspace(1)* %gep1.cast
  %build = insertelement <2 x i16> undef, i16 %load, i32 0
  %cast = bitcast <2 x i16> %build to <2 x half>
  ret <2 x half> %cast
}

define amdgpu_ps <2 x half> @global_load_saddr_i16_d16lo_zero_hi(i8 addrspace(1)* inreg %sbase, i32 %voffset) {
; GCN-LABEL: global_load_saddr_i16_d16lo_zero_hi:
; GCN:       ; %bb.0:
; GCN-NEXT:    v_mov_b32_e32 v1, 0
; GCN-NEXT:    global_load_short_d16 v1, v0, s[2:3]
; GCN-NEXT:    s_waitcnt vmcnt(0)
; GCN-NEXT:    v_mov_b32_e32 v0, v1
; GCN-NEXT:    ; return to shader part epilog
  %zext.offset = zext i32 %voffset to i64
  %gep0 = getelementptr inbounds i8, i8 addrspace(1)* %sbase, i64 %zext.offset
  %gep0.cast = bitcast i8 addrspace(1)* %gep0 to i16 addrspace(1)*
  %load = load i16, i16 addrspace(1)* %gep0.cast
  %build = insertelement <2 x i16> zeroinitializer, i16 %load, i32 0
  %cast = bitcast <2 x i16> %build to <2 x half>
  ret <2 x half> %cast
}

define amdgpu_ps <2 x half> @global_load_saddr_i16_d16lo_zero_hi_immneg128(i8 addrspace(1)* inreg %sbase, i32 %voffset) {
; GCN-LABEL: global_load_saddr_i16_d16lo_zero_hi_immneg128:
; GCN:       ; %bb.0:
; GCN-NEXT:    v_mov_b32_e32 v1, 0
; GCN-NEXT:    global_load_short_d16 v1, v0, s[2:3] offset:-128
; GCN-NEXT:    s_waitcnt vmcnt(0)
; GCN-NEXT:    v_mov_b32_e32 v0, v1
; GCN-NEXT:    ; return to shader part epilog
  %zext.offset = zext i32 %voffset to i64
  %gep0 = getelementptr inbounds i8, i8 addrspace(1)* %sbase, i64 %zext.offset
  %gep1 = getelementptr inbounds i8, i8 addrspace(1)* %gep0, i64 -128
  %gep1.cast = bitcast i8 addrspace(1)* %gep1 to i16 addrspace(1)*
  %load = load i16, i16 addrspace(1)* %gep1.cast
  %build = insertelement <2 x i16> zeroinitializer, i16 %load, i32 0
  %cast = bitcast <2 x i16> %build to <2 x half>
  ret <2 x half> %cast
}

define amdgpu_ps <2 x half> @global_load_saddr_i16_d16lo_reg_hi(i8 addrspace(1)* inreg %sbase, i32 %voffset, <2 x i16> %reg) {
; GCN-LABEL: global_load_saddr_i16_d16lo_reg_hi:
; GCN:       ; %bb.0:
; GCN-NEXT:    global_load_short_d16 v1, v0, s[2:3]
; GCN-NEXT:    s_waitcnt vmcnt(0)
; GCN-NEXT:    v_mov_b32_e32 v0, v1
; GCN-NEXT:    ; return to shader part epilog
  %zext.offset = zext i32 %voffset to i64
  %gep0 = getelementptr inbounds i8, i8 addrspace(1)* %sbase, i64 %zext.offset
  %gep0.cast = bitcast i8 addrspace(1)* %gep0 to i16 addrspace(1)*
  %load = load i16, i16 addrspace(1)* %gep0.cast
  %build = insertelement <2 x i16> %reg, i16 %load, i32 0
  %cast = bitcast <2 x i16> %build to <2 x half>
  ret <2 x half> %cast
}

define amdgpu_ps <2 x half> @global_load_saddr_i16_d16lo_reg_hi_immneg128(i8 addrspace(1)* inreg %sbase, i32 %voffset, <2 x i16> %reg) {
; GCN-LABEL: global_load_saddr_i16_d16lo_reg_hi_immneg128:
; GCN:       ; %bb.0:
; GCN-NEXT:    global_load_short_d16 v1, v0, s[2:3] offset:-128
; GCN-NEXT:    s_waitcnt vmcnt(0)
; GCN-NEXT:    v_mov_b32_e32 v0, v1
; GCN-NEXT:    ; return to shader part epilog
  %zext.offset = zext i32 %voffset to i64
  %gep0 = getelementptr inbounds i8, i8 addrspace(1)* %sbase, i64 %zext.offset
  %gep1 = getelementptr inbounds i8, i8 addrspace(1)* %gep0, i64 -128
  %gep1.cast = bitcast i8 addrspace(1)* %gep1 to i16 addrspace(1)*
  %load = load i16, i16 addrspace(1)* %gep1.cast
  %build = insertelement <2 x i16> %reg, i16 %load, i32 0
  %cast = bitcast <2 x i16> %build to <2 x half>
  ret <2 x half> %cast
}

define amdgpu_ps <2 x half> @global_load_saddr_i16_d16lo_zexti8_reg_hi(i8 addrspace(1)* inreg %sbase, i32 %voffset, <2 x i16> %reg) {
; GCN-LABEL: global_load_saddr_i16_d16lo_zexti8_reg_hi:
; GCN:       ; %bb.0:
; GCN-NEXT:    global_load_ubyte_d16 v1, v0, s[2:3]
; GCN-NEXT:    s_waitcnt vmcnt(0)
; GCN-NEXT:    v_mov_b32_e32 v0, v1
; GCN-NEXT:    ; return to shader part epilog
  %zext.offset = zext i32 %voffset to i64
  %gep0 = getelementptr inbounds i8, i8 addrspace(1)* %sbase, i64 %zext.offset
  %gep0.cast = bitcast i8 addrspace(1)* %gep0 to i8 addrspace(1)*
  %load = load i8, i8 addrspace(1)* %gep0.cast
  %zext.load = zext i8 %load to i16
  %build = insertelement <2 x i16> %reg, i16 %zext.load, i32 0
  %cast = bitcast <2 x i16> %build to <2 x half>
  ret <2 x half> %cast
}

define amdgpu_ps <2 x half> @global_load_saddr_i16_d16lo_zexti8_reg_hi_immneg128(i8 addrspace(1)* inreg %sbase, i32 %voffset, <2 x i16> %reg) {
; GCN-LABEL: global_load_saddr_i16_d16lo_zexti8_reg_hi_immneg128:
; GCN:       ; %bb.0:
; GCN-NEXT:    global_load_ubyte_d16 v1, v0, s[2:3] offset:-128
; GCN-NEXT:    s_waitcnt vmcnt(0)
; GCN-NEXT:    v_mov_b32_e32 v0, v1
; GCN-NEXT:    ; return to shader part epilog
  %zext.offset = zext i32 %voffset to i64
  %gep0 = getelementptr inbounds i8, i8 addrspace(1)* %sbase, i64 %zext.offset
  %gep1 = getelementptr inbounds i8, i8 addrspace(1)* %gep0, i64 -128
  %gep1.cast = bitcast i8 addrspace(1)* %gep1 to i8 addrspace(1)*
  %load = load i8, i8 addrspace(1)* %gep1.cast
  %zext.load = zext i8 %load to i16
  %build = insertelement <2 x i16> %reg, i16 %zext.load, i32 0
  %cast = bitcast <2 x i16> %build to <2 x half>
  ret <2 x half> %cast
}

define amdgpu_ps <2 x half> @global_load_saddr_i16_d16lo_sexti8_reg_hi(i8 addrspace(1)* inreg %sbase, i32 %voffset, <2 x i16> %reg) {
; GCN-LABEL: global_load_saddr_i16_d16lo_sexti8_reg_hi:
; GCN:       ; %bb.0:
; GCN-NEXT:    global_load_sbyte_d16 v1, v0, s[2:3]
; GCN-NEXT:    s_waitcnt vmcnt(0)
; GCN-NEXT:    v_mov_b32_e32 v0, v1
; GCN-NEXT:    ; return to shader part epilog
  %zext.offset = zext i32 %voffset to i64
  %gep0 = getelementptr inbounds i8, i8 addrspace(1)* %sbase, i64 %zext.offset
  %gep0.cast = bitcast i8 addrspace(1)* %gep0 to i8 addrspace(1)*
  %load = load i8, i8 addrspace(1)* %gep0.cast
  %sext.load = sext i8 %load to i16
  %build = insertelement <2 x i16> %reg, i16 %sext.load, i32 0
  %cast = bitcast <2 x i16> %build to <2 x half>
  ret <2 x half> %cast
}

define amdgpu_ps <2 x half> @global_load_saddr_i16_d16lo_sexti8_reg_hi_immneg128(i8 addrspace(1)* inreg %sbase, i32 %voffset, <2 x i16> %reg) {
; GCN-LABEL: global_load_saddr_i16_d16lo_sexti8_reg_hi_immneg128:
; GCN:       ; %bb.0:
; GCN-NEXT:    global_load_sbyte_d16 v1, v0, s[2:3] offset:-128
; GCN-NEXT:    s_waitcnt vmcnt(0)
; GCN-NEXT:    v_mov_b32_e32 v0, v1
; GCN-NEXT:    ; return to shader part epilog
  %zext.offset = zext i32 %voffset to i64
  %gep0 = getelementptr inbounds i8, i8 addrspace(1)* %sbase, i64 %zext.offset
  %gep1 = getelementptr inbounds i8, i8 addrspace(1)* %gep0, i64 -128
  %gep1.cast = bitcast i8 addrspace(1)* %gep1 to i8 addrspace(1)*
  %load = load i8, i8 addrspace(1)* %gep1.cast
  %sext.load = sext i8 %load to i16
  %build = insertelement <2 x i16> %reg, i16 %sext.load, i32 0
  %cast = bitcast <2 x i16> %build to <2 x half>
  ret <2 x half> %cast
}

; --------------------------------------------------------------------------------
; D16 hi load (hi16)
; --------------------------------------------------------------------------------

define amdgpu_ps <2 x half> @global_load_saddr_i16_d16hi_undef_hi(i8 addrspace(1)* inreg %sbase, i32 %voffset) {
; GCN-LABEL: global_load_saddr_i16_d16hi_undef_hi:
; GCN:       ; %bb.0:
; GCN-NEXT:    global_load_short_d16_hi v0, v0, s[2:3]
; GCN-NEXT:    s_waitcnt vmcnt(0)
; GCN-NEXT:    ; return to shader part epilog
  %zext.offset = zext i32 %voffset to i64
  %gep0 = getelementptr inbounds i8, i8 addrspace(1)* %sbase, i64 %zext.offset
  %gep0.cast = bitcast i8 addrspace(1)* %gep0 to i16 addrspace(1)*
  %load = load i16, i16 addrspace(1)* %gep0.cast
  %build = insertelement <2 x i16> undef, i16 %load, i32 1
  %cast = bitcast <2 x i16> %build to <2 x half>
  ret <2 x half> %cast
}

define amdgpu_ps <2 x half> @global_load_saddr_i16_d16hi_undef_hi_immneg128(i8 addrspace(1)* inreg %sbase, i32 %voffset) {
; GCN-LABEL: global_load_saddr_i16_d16hi_undef_hi_immneg128:
; GCN:       ; %bb.0:
; GCN-NEXT:    global_load_short_d16_hi v0, v0, s[2:3] offset:-128
; GCN-NEXT:    s_waitcnt vmcnt(0)
; GCN-NEXT:    ; return to shader part epilog
  %zext.offset = zext i32 %voffset to i64
  %gep0 = getelementptr inbounds i8, i8 addrspace(1)* %sbase, i64 %zext.offset
  %gep1 = getelementptr inbounds i8, i8 addrspace(1)* %gep0, i64 -128
  %gep1.cast = bitcast i8 addrspace(1)* %gep1 to i16 addrspace(1)*
  %load = load i16, i16 addrspace(1)* %gep1.cast
  %build = insertelement <2 x i16> undef, i16 %load, i32 1
  %cast = bitcast <2 x i16> %build to <2 x half>
  ret <2 x half> %cast
}

define amdgpu_ps <2 x half> @global_load_saddr_i16_d16hi_zero_hi(i8 addrspace(1)* inreg %sbase, i32 %voffset) {
; GCN-LABEL: global_load_saddr_i16_d16hi_zero_hi:
; GCN:       ; %bb.0:
; GCN-NEXT:    v_mov_b32_e32 v1, 0
; GCN-NEXT:    global_load_short_d16_hi v1, v0, s[2:3]
; GCN-NEXT:    s_waitcnt vmcnt(0)
; GCN-NEXT:    v_mov_b32_e32 v0, v1
; GCN-NEXT:    ; return to shader part epilog
  %zext.offset = zext i32 %voffset to i64
  %gep0 = getelementptr inbounds i8, i8 addrspace(1)* %sbase, i64 %zext.offset
  %gep0.cast = bitcast i8 addrspace(1)* %gep0 to i16 addrspace(1)*
  %load = load i16, i16 addrspace(1)* %gep0.cast
  %build = insertelement <2 x i16> zeroinitializer, i16 %load, i32 1
  %cast = bitcast <2 x i16> %build to <2 x half>
  ret <2 x half> %cast
}

define amdgpu_ps <2 x half> @global_load_saddr_i16_d16hi_zero_hi_immneg128(i8 addrspace(1)* inreg %sbase, i32 %voffset) {
; GCN-LABEL: global_load_saddr_i16_d16hi_zero_hi_immneg128:
; GCN:       ; %bb.0:
; GCN-NEXT:    v_mov_b32_e32 v1, 0
; GCN-NEXT:    global_load_short_d16_hi v1, v0, s[2:3] offset:-128
; GCN-NEXT:    s_waitcnt vmcnt(0)
; GCN-NEXT:    v_mov_b32_e32 v0, v1
; GCN-NEXT:    ; return to shader part epilog
  %zext.offset = zext i32 %voffset to i64
  %gep0 = getelementptr inbounds i8, i8 addrspace(1)* %sbase, i64 %zext.offset
  %gep1 = getelementptr inbounds i8, i8 addrspace(1)* %gep0, i64 -128
  %gep1.cast = bitcast i8 addrspace(1)* %gep1 to i16 addrspace(1)*
  %load = load i16, i16 addrspace(1)* %gep1.cast
  %build = insertelement <2 x i16> zeroinitializer, i16 %load, i32 1
  %cast = bitcast <2 x i16> %build to <2 x half>
  ret <2 x half> %cast
}

define amdgpu_ps <2 x half> @global_load_saddr_i16_d16hi_reg_hi(i8 addrspace(1)* inreg %sbase, i32 %voffset, <2 x i16> %reg) {
; GCN-LABEL: global_load_saddr_i16_d16hi_reg_hi:
; GCN:       ; %bb.0:
; GCN-NEXT:    global_load_short_d16_hi v1, v0, s[2:3]
; GCN-NEXT:    s_waitcnt vmcnt(0)
; GCN-NEXT:    v_mov_b32_e32 v0, v1
; GCN-NEXT:    ; return to shader part epilog
  %zext.offset = zext i32 %voffset to i64
  %gep0 = getelementptr inbounds i8, i8 addrspace(1)* %sbase, i64 %zext.offset
  %gep0.cast = bitcast i8 addrspace(1)* %gep0 to i16 addrspace(1)*
  %load = load i16, i16 addrspace(1)* %gep0.cast
  %build = insertelement <2 x i16> %reg, i16 %load, i32 1
  %cast = bitcast <2 x i16> %build to <2 x half>
  ret <2 x half> %cast
}

define amdgpu_ps <2 x half> @global_load_saddr_i16_d16hi_reg_hi_immneg128(i8 addrspace(1)* inreg %sbase, i32 %voffset, <2 x i16> %reg) {
; GCN-LABEL: global_load_saddr_i16_d16hi_reg_hi_immneg128:
; GCN:       ; %bb.0:
; GCN-NEXT:    global_load_short_d16_hi v1, v0, s[2:3] offset:-128
; GCN-NEXT:    s_waitcnt vmcnt(0)
; GCN-NEXT:    v_mov_b32_e32 v0, v1
; GCN-NEXT:    ; return to shader part epilog
  %zext.offset = zext i32 %voffset to i64
  %gep0 = getelementptr inbounds i8, i8 addrspace(1)* %sbase, i64 %zext.offset
  %gep1 = getelementptr inbounds i8, i8 addrspace(1)* %gep0, i64 -128
  %gep1.cast = bitcast i8 addrspace(1)* %gep1 to i16 addrspace(1)*
  %load = load i16, i16 addrspace(1)* %gep1.cast
  %build = insertelement <2 x i16> %reg, i16 %load, i32 1
  %cast = bitcast <2 x i16> %build to <2 x half>
  ret <2 x half> %cast
}

define amdgpu_ps <2 x half> @global_load_saddr_i16_d16hi_zexti8_reg_hi(i8 addrspace(1)* inreg %sbase, i32 %voffset, <2 x i16> %reg) {
; GCN-LABEL: global_load_saddr_i16_d16hi_zexti8_reg_hi:
; GCN:       ; %bb.0:
; GCN-NEXT:    global_load_ubyte_d16_hi v1, v0, s[2:3]
; GCN-NEXT:    s_waitcnt vmcnt(0)
; GCN-NEXT:    v_mov_b32_e32 v0, v1
; GCN-NEXT:    ; return to shader part epilog
  %zext.offset = zext i32 %voffset to i64
  %gep0 = getelementptr inbounds i8, i8 addrspace(1)* %sbase, i64 %zext.offset
  %gep0.cast = bitcast i8 addrspace(1)* %gep0 to i8 addrspace(1)*
  %load = load i8, i8 addrspace(1)* %gep0.cast
  %zext.load = zext i8 %load to i16
  %build = insertelement <2 x i16> %reg, i16 %zext.load, i32 1
  %cast = bitcast <2 x i16> %build to <2 x half>
  ret <2 x half> %cast
}

define amdgpu_ps <2 x half> @global_load_saddr_i16_d16hi_zexti8_reg_hi_immneg128(i8 addrspace(1)* inreg %sbase, i32 %voffset, <2 x i16> %reg) {
; GCN-LABEL: global_load_saddr_i16_d16hi_zexti8_reg_hi_immneg128:
; GCN:       ; %bb.0:
; GCN-NEXT:    global_load_ubyte_d16_hi v1, v0, s[2:3] offset:-128
; GCN-NEXT:    s_waitcnt vmcnt(0)
; GCN-NEXT:    v_mov_b32_e32 v0, v1
; GCN-NEXT:    ; return to shader part epilog
  %zext.offset = zext i32 %voffset to i64
  %gep0 = getelementptr inbounds i8, i8 addrspace(1)* %sbase, i64 %zext.offset
  %gep1 = getelementptr inbounds i8, i8 addrspace(1)* %gep0, i64 -128
  %gep1.cast = bitcast i8 addrspace(1)* %gep1 to i8 addrspace(1)*
  %load = load i8, i8 addrspace(1)* %gep1.cast
  %zext.load = zext i8 %load to i16
  %build = insertelement <2 x i16> %reg, i16 %zext.load, i32 1
  %cast = bitcast <2 x i16> %build to <2 x half>
  ret <2 x half> %cast
}

define amdgpu_ps <2 x half> @global_load_saddr_i16_d16hi_sexti8_reg_hi(i8 addrspace(1)* inreg %sbase, i32 %voffset, <2 x i16> %reg) {
; GCN-LABEL: global_load_saddr_i16_d16hi_sexti8_reg_hi:
; GCN:       ; %bb.0:
; GCN-NEXT:    global_load_sbyte_d16_hi v1, v0, s[2:3]
; GCN-NEXT:    s_waitcnt vmcnt(0)
; GCN-NEXT:    v_mov_b32_e32 v0, v1
; GCN-NEXT:    ; return to shader part epilog
  %zext.offset = zext i32 %voffset to i64
  %gep0 = getelementptr inbounds i8, i8 addrspace(1)* %sbase, i64 %zext.offset
  %gep0.cast = bitcast i8 addrspace(1)* %gep0 to i8 addrspace(1)*
  %load = load i8, i8 addrspace(1)* %gep0.cast
  %sext.load = sext i8 %load to i16
  %build = insertelement <2 x i16> %reg, i16 %sext.load, i32 1
  %cast = bitcast <2 x i16> %build to <2 x half>
  ret <2 x half> %cast
}

define amdgpu_ps <2 x half> @global_load_saddr_i16_d16hi_sexti8_reg_hi_immneg128(i8 addrspace(1)* inreg %sbase, i32 %voffset, <2 x i16> %reg) {
; GCN-LABEL: global_load_saddr_i16_d16hi_sexti8_reg_hi_immneg128:
; GCN:       ; %bb.0:
; GCN-NEXT:    global_load_sbyte_d16_hi v1, v0, s[2:3] offset:-128
; GCN-NEXT:    s_waitcnt vmcnt(0)
; GCN-NEXT:    v_mov_b32_e32 v0, v1
; GCN-NEXT:    ; return to shader part epilog
  %zext.offset = zext i32 %voffset to i64
  %gep0 = getelementptr inbounds i8, i8 addrspace(1)* %sbase, i64 %zext.offset
  %gep1 = getelementptr inbounds i8, i8 addrspace(1)* %gep0, i64 -128
  %gep1.cast = bitcast i8 addrspace(1)* %gep1 to i8 addrspace(1)*
  %load = load i8, i8 addrspace(1)* %gep1.cast
  %sext.load = sext i8 %load to i16
  %build = insertelement <2 x i16> %reg, i16 %sext.load, i32 1
  %cast = bitcast <2 x i16> %build to <2 x half>
  ret <2 x half> %cast
}

; --------------------------------------------------------------------------------
; or-with-constant as add
; --------------------------------------------------------------------------------

; Check add-as-or with split 64-bit or.
define amdgpu_ps float @global_load_saddr_i8_offset_or_i64_imm_offset_16(i8 addrspace(6)* inreg %sbase, i32 %idx) {
; GCN-LABEL: global_load_saddr_i8_offset_or_i64_imm_offset_16:
; GCN:       ; %bb.0:
; GCN-NEXT:    v_or_b32_e32 v0, 16, v0
; GCN-NEXT:    v_mov_b32_e32 v1, 0
; GCN-NEXT:    global_load_ubyte v0, v[0:1], off
; GCN-NEXT:    s_waitcnt vmcnt(0)
; GCN-NEXT:    ; return to shader part epilog
  %zext.idx = zext i32 %idx to i64
  %or = or i64 %zext.idx, 16
  %addr = inttoptr i64 %or to i8 addrspace(1)*
  %load = load i8, i8 addrspace(1)* %addr
  %zext = zext i8 %load to i32
  %to.vgpr = bitcast i32 %zext to float
  ret float %to.vgpr
}

define amdgpu_ps float @global_load_saddr_i8_offset_or_i64_imm_offset_4160(i8 addrspace(6)* inreg %sbase, i32 %idx) {
; GCN-LABEL: global_load_saddr_i8_offset_or_i64_imm_offset_4160:
; GCN:       ; %bb.0:
; GCN-NEXT:    v_or_b32_e32 v0, 0x1040, v0
; GCN-NEXT:    v_mov_b32_e32 v1, 0
; GCN-NEXT:    global_load_ubyte v0, v[0:1], off
; GCN-NEXT:    s_waitcnt vmcnt(0)
; GCN-NEXT:    ; return to shader part epilog
  %zext.idx = zext i32 %idx to i64
  %or = or i64 %zext.idx, 4160
  %addr = inttoptr i64 %or to i8 addrspace(1)*
  %load = load i8, i8 addrspace(1)* %addr
  %zext = zext i8 %load to i32
  %to.vgpr = bitcast i32 %zext to float
  ret float %to.vgpr
}

; --------------------------------------------------------------------------------
; Full 64-bit scalar add.
; --------------------------------------------------------------------------------

define amdgpu_ps void @global_addr_64bit_lsr_iv(float addrspace(1)* inreg %arg) {
; GFX9-LABEL: global_addr_64bit_lsr_iv:
; GFX9:       ; %bb.0: ; %bb
; GFX9-NEXT:    s_mov_b64 s[0:1], 0
; GFX9-NEXT:    v_mov_b32_e32 v0, 0
; GFX9-NEXT:  .LBB128_1: ; %bb3
; GFX9-NEXT:    ; =>This Inner Loop Header: Depth=1
; GFX9-NEXT:    s_add_u32 s4, s2, s0
; GFX9-NEXT:    s_addc_u32 s5, s3, s1
; GFX9-NEXT:    global_load_dword v1, v0, s[4:5] glc
; GFX9-NEXT:    s_waitcnt vmcnt(0)
; GFX9-NEXT:    s_add_u32 s0, s0, 4
; GFX9-NEXT:    s_addc_u32 s1, s1, 0
; GFX9-NEXT:    s_cmpk_eq_i32 s0, 0x400
; GFX9-NEXT:    s_cbranch_scc0 .LBB128_1
; GFX9-NEXT:  ; %bb.2: ; %bb2
; GFX9-NEXT:    s_endpgm
;
; GFX10-LABEL: global_addr_64bit_lsr_iv:
; GFX10:       ; %bb.0: ; %bb
; GFX10-NEXT:    v_mov_b32_e32 v0, 0
; GFX10-NEXT:    s_mov_b64 s[0:1], 0
; GFX10-NEXT:  .LBB128_1: ; %bb3
; GFX10-NEXT:    ; =>This Inner Loop Header: Depth=1
; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
; GFX10-NEXT:    s_add_u32 s4, s2, s0
; GFX10-NEXT:    s_addc_u32 s5, s3, s1
; GFX10-NEXT:    s_add_u32 s0, s0, 4
; GFX10-NEXT:    global_load_dword v1, v0, s[4:5] glc dlc
; GFX10-NEXT:    s_waitcnt vmcnt(0)
; GFX10-NEXT:    s_addc_u32 s1, s1, 0
; GFX10-NEXT:    s_cmpk_eq_i32 s0, 0x400
; GFX10-NEXT:    s_cbranch_scc0 .LBB128_1
; GFX10-NEXT:  ; %bb.2: ; %bb2
; GFX10-NEXT:    s_endpgm
bb:
  br label %bb3

bb2:                                              ; preds = %bb3
  ret void

bb3:                                              ; preds = %bb3, %bb
  %i = phi i32 [ 0, %bb ], [ %i8, %bb3 ]
  %i4 = zext i32 %i to i64
  %i5 = getelementptr inbounds float, float addrspace(1)* %arg, i64 %i4
  %i6 = load volatile float, float addrspace(1)* %i5, align 4
  %i8 = add nuw nsw i32 %i, 1
  %i9 = icmp eq i32 %i8, 256
  br i1 %i9, label %bb2, label %bb3
}

; Make sure we only have a single zero vaddr initialization.

define amdgpu_ps void @global_addr_64bit_lsr_iv_multiload(float addrspace(1)* inreg %arg, float addrspace(1)* inreg %arg.1) {
; GFX9-LABEL: global_addr_64bit_lsr_iv_multiload:
; GFX9:       ; %bb.0: ; %bb
; GFX9-NEXT:    s_mov_b64 s[0:1], 0
; GFX9-NEXT:    v_mov_b32_e32 v0, 0
; GFX9-NEXT:  .LBB129_1: ; %bb3
; GFX9-NEXT:    ; =>This Inner Loop Header: Depth=1
; GFX9-NEXT:    s_add_u32 s4, s2, s0
; GFX9-NEXT:    s_addc_u32 s5, s3, s1
; GFX9-NEXT:    global_load_dword v1, v0, s[4:5] glc
; GFX9-NEXT:    s_waitcnt vmcnt(0)
; GFX9-NEXT:    global_load_dword v1, v0, s[4:5] glc
; GFX9-NEXT:    s_waitcnt vmcnt(0)
; GFX9-NEXT:    s_add_u32 s0, s0, 4
; GFX9-NEXT:    s_addc_u32 s1, s1, 0
; GFX9-NEXT:    s_cmpk_eq_i32 s0, 0x400
; GFX9-NEXT:    ; kill: killed $sgpr4 killed $sgpr5
; GFX9-NEXT:    s_cbranch_scc0 .LBB129_1
; GFX9-NEXT:  ; %bb.2: ; %bb2
; GFX9-NEXT:    s_endpgm
;
; GFX10-LABEL: global_addr_64bit_lsr_iv_multiload:
; GFX10:       ; %bb.0: ; %bb
; GFX10-NEXT:    v_mov_b32_e32 v0, 0
; GFX10-NEXT:    s_mov_b64 s[0:1], 0
; GFX10-NEXT:  .LBB129_1: ; %bb3
; GFX10-NEXT:    ; =>This Inner Loop Header: Depth=1
; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
; GFX10-NEXT:    s_add_u32 s4, s2, s0
; GFX10-NEXT:    s_addc_u32 s5, s3, s1
; GFX10-NEXT:    s_add_u32 s0, s0, 4
; GFX10-NEXT:    global_load_dword v1, v0, s[4:5] glc dlc
; GFX10-NEXT:    s_waitcnt vmcnt(0)
; GFX10-NEXT:    global_load_dword v1, v0, s[4:5] glc dlc
; GFX10-NEXT:    s_waitcnt vmcnt(0)
; GFX10-NEXT:    s_addc_u32 s1, s1, 0
; GFX10-NEXT:    s_cmpk_eq_i32 s0, 0x400
; GFX10-NEXT:    ; kill: killed $sgpr4 killed $sgpr5
; GFX10-NEXT:    s_cbranch_scc0 .LBB129_1
; GFX10-NEXT:  ; %bb.2: ; %bb2
; GFX10-NEXT:    s_endpgm
bb:
  br label %bb3

bb2:                                              ; preds = %bb3
  ret void

bb3:                                              ; preds = %bb3, %bb
  %i = phi i32 [ 0, %bb ], [ %i8, %bb3 ]
  %i4 = zext i32 %i to i64
  %i5 = getelementptr inbounds float, float addrspace(1)* %arg, i64 %i4
  %i6 = load volatile float, float addrspace(1)* %i5, align 4
  %i5.1 = getelementptr inbounds float, float addrspace(1)* %arg.1, i64 %i4
  %i6.1 = load volatile float, float addrspace(1)* %i5, align 4
  %i8 = add nuw nsw i32 %i, 1
  %i9 = icmp eq i32 %i8, 256
  br i1 %i9, label %bb2, label %bb3
}

!0 = !{i32 0, i32 1073741824} ; (1 << 30)
!1 = !{i32 0, i32 1073741825} ; (1 << 30) + 1