llvm/test/CodeGen/AMDGPU/llvm.sponentry.ll

; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
; RUN: llc -global-isel=0 -mtriple=amdgcn--amdpal -mcpu=gfx1200 -mattr=+real-true16 < %s | FileCheck %s --check-prefixes=CHECK,DAGISEL
; RUN: llc -global-isel=0 -mtriple=amdgcn--amdpal -mcpu=gfx1200 -mattr=-real-true16 < %s | FileCheck %s --check-prefixes=CHECK,DAGISEL
; RUN: llc -global-isel=1 -mtriple=amdgcn--amdpal -mcpu=gfx1200 -mattr=+real-true16 < %s | FileCheck %s --check-prefixes=CHECK,GISEL
; RUN: llc -global-isel=1 -mtriple=amdgcn--amdpal -mcpu=gfx1200 -mattr=-real-true16 < %s | FileCheck %s --check-prefixes=CHECK,GISEL

; Test that in dynamic VGPR mode, the return of sponentry points after the area reserved for CWSR.

define amdgpu_cs ptr addrspace(5) @sponentry_cs_dvgpr_16(i32 %val) #0 {
; CHECK-LABEL: sponentry_cs_dvgpr_16:
; CHECK:       ; %bb.0:
; CHECK-NEXT:    s_getreg_b32 s33, hwreg(HW_REG_WAVE_HW_ID2, 8, 2)
; CHECK-NEXT:    s_getreg_b32 s0, hwreg(HW_REG_WAVE_HW_ID2, 8, 2)
; CHECK-NEXT:    s_cmp_lg_u32 0, s33
; CHECK-NEXT:    s_cmovk_i32 s33, 0x1c0
; CHECK-NEXT:    s_cmp_lg_u32 0, s0
; CHECK-NEXT:    scratch_store_b32 off, v0, s33 scope:SCOPE_SYS
; CHECK-NEXT:    s_wait_storecnt 0x0
; CHECK-NEXT:    s_cmovk_i32 s0, 0x1c0
; CHECK-NEXT:    ; return to shader part epilog
  %local = alloca i32, addrspace(5)
  store volatile i32 %val, ptr addrspace(5) %local
  %stack.base = call ptr addrspace(5) @llvm.sponentry()
  ret ptr addrspace(5) %stack.base
}

; CHECK: ScratchSize: 8

define amdgpu_cs ptr addrspace(5) @sponentry_cs_dvgpr_32(i32 %val) #1 {
; CHECK-LABEL: sponentry_cs_dvgpr_32:
; CHECK:       ; %bb.0:
; CHECK-NEXT:    s_getreg_b32 s33, hwreg(HW_REG_WAVE_HW_ID2, 8, 2)
; CHECK-NEXT:    s_getreg_b32 s0, hwreg(HW_REG_WAVE_HW_ID2, 8, 2)
; CHECK-NEXT:    s_cmp_lg_u32 0, s33
; CHECK-NEXT:    s_cmovk_i32 s33, 0x380
; CHECK-NEXT:    s_cmp_lg_u32 0, s0
; CHECK-NEXT:    scratch_store_b32 off, v0, s33 scope:SCOPE_SYS
; CHECK-NEXT:    s_wait_storecnt 0x0
; CHECK-NEXT:    s_cmovk_i32 s0, 0x380
; CHECK-NEXT:    ; return to shader part epilog
  %local = alloca i32, addrspace(5)
  store volatile i32 %val, ptr addrspace(5) %local
  %stack.base = call ptr addrspace(5) @llvm.sponentry()
  ret ptr addrspace(5) %stack.base
}

; CHECK: ScratchSize: 8

; If we're not in dynamic VGPR mode, then sponentry can just return 0.

define amdgpu_cs ptr addrspace(5) @sponentry_cs_no_dvgpr(i32 %val) #2 {
; CHECK-LABEL: sponentry_cs_no_dvgpr:
; CHECK:       ; %bb.0:
; CHECK-NEXT:    s_mov_b32 s0, 0
; CHECK-NEXT:    scratch_store_b32 off, v0, off scope:SCOPE_SYS
; CHECK-NEXT:    s_wait_storecnt 0x0
; CHECK-NEXT:    ; return to shader part epilog
  %local = alloca i32, addrspace(5)
  store volatile i32 %val, ptr addrspace(5) %local
  %stack.base = call ptr addrspace(5) @llvm.sponentry()
  ret ptr addrspace(5) %stack.base
}

; CHECK: ScratchSize: 8

define amdgpu_cs ptr addrspace(5) @sponentry_cs_dvgpr_control_flow(i32 %val, ptr addrspace(5) %ptr) #0 {
; CHECK-LABEL: sponentry_cs_dvgpr_control_flow:
; CHECK:       ; %bb.0: ; %entry
; CHECK-NEXT:    s_getreg_b32 s33, hwreg(HW_REG_WAVE_HW_ID2, 8, 2)
; CHECK-NEXT:    s_mov_b32 s0, exec_lo
; CHECK-NEXT:    s_cmp_lg_u32 0, s33
; CHECK-NEXT:    s_cmovk_i32 s33, 0x1c0
; CHECK-NEXT:    scratch_store_b32 off, v0, s33 scope:SCOPE_SYS
; CHECK-NEXT:    s_wait_storecnt 0x0
; CHECK-NEXT:    v_cmpx_gt_i32_e32 0x43, v0
; CHECK-NEXT:  ; %bb.1: ; %if.then
; CHECK-NEXT:    s_getreg_b32 s1, hwreg(HW_REG_WAVE_HW_ID2, 8, 2)
; CHECK-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_1) | instid1(SALU_CYCLE_1)
; CHECK-NEXT:    s_cmp_lg_u32 0, s1
; CHECK-NEXT:    s_cmovk_i32 s1, 0x1c0
; CHECK-NEXT:    v_mov_b32_e32 v1, s1
; CHECK-NEXT:  ; %bb.2: ; %if.end
; CHECK-NEXT:    s_or_b32 exec_lo, exec_lo, s0
; CHECK-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; CHECK-NEXT:    v_readfirstlane_b32 s0, v1
; CHECK-NEXT:    s_wait_alu depctr_va_sdst(0)
; CHECK-NEXT:    ; return to shader part epilog
entry:
  %local = alloca i32, addrspace(5)
  store volatile i32 %val, ptr addrspace(5) %local
  %which = icmp slt i32 %val, 67
  br i1 %which, label %if.then, label %if.end

if.then:
  %stack.base = call ptr addrspace(5) @llvm.sponentry()
  br label %if.end

if.end:
  %ret = phi ptr addrspace(5) [ %stack.base, %if.then ], [ %ptr, %entry ]
  ret ptr addrspace(5) %ret
}

; CHECK: ScratchSize: 8

declare amdgpu_gfx i32 @callee()

define amdgpu_cs ptr addrspace(5) @sponentry_cs_dvgpr_calls(i32 %val) #0 {
; DAGISEL-LABEL: sponentry_cs_dvgpr_calls:
; DAGISEL:       ; %bb.0:
; DAGISEL-NEXT:    s_getreg_b32 s33, hwreg(HW_REG_WAVE_HW_ID2, 8, 2)
; DAGISEL-NEXT:    s_mov_b32 s1, callee@abs32@hi
; DAGISEL-NEXT:    s_cmp_lg_u32 0, s33
; DAGISEL-NEXT:    s_mov_b32 s0, callee@abs32@lo
; DAGISEL-NEXT:    s_cmovk_i32 s33, 0x1c0
; DAGISEL-NEXT:    s_cselect_b32 s32, 0x1d0, 16
; DAGISEL-NEXT:    s_swappc_b64 s[30:31], s[0:1]
; DAGISEL-NEXT:    s_getreg_b32 s0, hwreg(HW_REG_WAVE_HW_ID2, 8, 2)
; DAGISEL-NEXT:    s_wait_storecnt 0x0
; DAGISEL-NEXT:    scratch_store_b32 off, v0, s33 scope:SCOPE_SYS
; DAGISEL-NEXT:    s_wait_storecnt 0x0
; DAGISEL-NEXT:    s_wait_alu depctr_sa_sdst(0)
; DAGISEL-NEXT:    s_cmp_lg_u32 0, s0
; DAGISEL-NEXT:    s_cmovk_i32 s0, 0x1c0
; DAGISEL-NEXT:    s_wait_alu depctr_sa_sdst(0)
; DAGISEL-NEXT:    ; return to shader part epilog
;
; GISEL-LABEL: sponentry_cs_dvgpr_calls:
; GISEL:       ; %bb.0:
; GISEL-NEXT:    s_getreg_b32 s33, hwreg(HW_REG_WAVE_HW_ID2, 8, 2)
; GISEL-NEXT:    s_mov_b32 s0, callee@abs32@lo
; GISEL-NEXT:    s_cmp_lg_u32 0, s33
; GISEL-NEXT:    s_mov_b32 s1, callee@abs32@hi
; GISEL-NEXT:    s_cmovk_i32 s33, 0x1c0
; GISEL-NEXT:    s_cselect_b32 s32, 0x1d0, 16
; GISEL-NEXT:    s_swappc_b64 s[30:31], s[0:1]
; GISEL-NEXT:    s_getreg_b32 s0, hwreg(HW_REG_WAVE_HW_ID2, 8, 2)
; GISEL-NEXT:    s_wait_storecnt 0x0
; GISEL-NEXT:    scratch_store_b32 off, v0, s33 scope:SCOPE_SYS
; GISEL-NEXT:    s_wait_storecnt 0x0
; GISEL-NEXT:    s_wait_alu depctr_sa_sdst(0)
; GISEL-NEXT:    s_cmp_lg_u32 0, s0
; GISEL-NEXT:    s_cmovk_i32 s0, 0x1c0
; GISEL-NEXT:    s_wait_alu depctr_sa_sdst(0)
; GISEL-NEXT:    ; return to shader part epilog
  %local = alloca i32, addrspace(5)
  %res = call amdgpu_gfx i32 @callee()
  store volatile i32 %res, ptr addrspace(5) %local
  %stack.base = call ptr addrspace(5) @llvm.sponentry()
  ret ptr addrspace(5) %stack.base
}

; CHECK: ScratchSize: 16

define amdgpu_cs ptr addrspace(5) @sponentry_cs_dvgpr_realign(i32 %val) #0 {
; CHECK-LABEL: sponentry_cs_dvgpr_realign:
; CHECK:       ; %bb.0:
; CHECK-NEXT:    s_getreg_b32 s33, hwreg(HW_REG_WAVE_HW_ID2, 8, 2)
; CHECK-NEXT:    s_getreg_b32 s0, hwreg(HW_REG_WAVE_HW_ID2, 8, 2)
; CHECK-NEXT:    s_cmp_lg_u32 0, s33
; CHECK-NEXT:    s_cmovk_i32 s33, 0x200
; CHECK-NEXT:    s_cmp_lg_u32 0, s0
; CHECK-NEXT:    scratch_store_b32 off, v0, s33 scope:SCOPE_SYS
; CHECK-NEXT:    s_wait_storecnt 0x0
; CHECK-NEXT:    s_cmovk_i32 s0, 0x200
; CHECK-NEXT:    ; return to shader part epilog
  %boop = alloca i32, addrspace(5)
  %local = alloca i32, align 128, addrspace(5)
  store volatile i32 %val, ptr addrspace(5) %local
  %stack.base = call ptr addrspace(5) @llvm.sponentry()
  ret ptr addrspace(5) %stack.base
}

; CHECK: ScratchSize: 128

define amdgpu_gfx ptr addrspace(5) @sponentry_gfx(i32 %val, ptr addrspace(5) %ptr) #0 {
; DAGISEL-LABEL: sponentry_gfx:
; DAGISEL:       ; %bb.0: ; %entry
; DAGISEL-NEXT:    s_wait_loadcnt_dscnt 0x0
; DAGISEL-NEXT:    s_wait_expcnt 0x0
; DAGISEL-NEXT:    s_wait_samplecnt 0x0
; DAGISEL-NEXT:    s_wait_bvhcnt 0x0
; DAGISEL-NEXT:    s_wait_kmcnt 0x0
; DAGISEL-NEXT:    s_mov_b32 s0, exec_lo
; DAGISEL-NEXT:    s_wait_storecnt 0x0
; DAGISEL-NEXT:    scratch_store_b32 off, v0, s32 offset:4 scope:SCOPE_SYS
; DAGISEL-NEXT:    s_wait_storecnt 0x0
; DAGISEL-NEXT:    v_cmpx_gt_i32_e32 0x43, v0
; DAGISEL-NEXT:  ; %bb.1: ; %if.then
; DAGISEL-NEXT:    v_mov_b32_e32 v1, s32
; DAGISEL-NEXT:  ; %bb.2: ; %if.end
; DAGISEL-NEXT:    s_wait_alu depctr_sa_sdst(0)
; DAGISEL-NEXT:    s_or_b32 exec_lo, exec_lo, s0
; DAGISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; DAGISEL-NEXT:    v_mov_b32_e32 v0, v1
; DAGISEL-NEXT:    s_setpc_b64 s[30:31]
;
; GISEL-LABEL: sponentry_gfx:
; GISEL:       ; %bb.0: ; %entry
; GISEL-NEXT:    s_wait_loadcnt_dscnt 0x0
; GISEL-NEXT:    s_wait_expcnt 0x0
; GISEL-NEXT:    s_wait_samplecnt 0x0
; GISEL-NEXT:    s_wait_bvhcnt 0x0
; GISEL-NEXT:    s_wait_kmcnt 0x0
; GISEL-NEXT:    v_mov_b32_e32 v2, v0
; GISEL-NEXT:    v_mov_b32_e32 v0, v1
; GISEL-NEXT:    s_mov_b32 s0, exec_lo
; GISEL-NEXT:    s_wait_storecnt 0x0
; GISEL-NEXT:    scratch_store_b32 off, v2, s32 offset:4 scope:SCOPE_SYS
; GISEL-NEXT:    s_wait_storecnt 0x0
; GISEL-NEXT:    v_cmpx_gt_i32_e32 0x43, v2
; GISEL-NEXT:  ; %bb.1: ; %if.then
; GISEL-NEXT:    s_mov_b32 s1, s32
; GISEL-NEXT:    s_wait_alu depctr_sa_sdst(0)
; GISEL-NEXT:    v_mov_b32_e32 v0, s1
; GISEL-NEXT:  ; %bb.2: ; %if.end
; GISEL-NEXT:    s_or_b32 exec_lo, exec_lo, s0
; GISEL-NEXT:    s_setpc_b64 s[30:31]
entry:
  %local = alloca i32, addrspace(5)
  store volatile i32 %val, ptr addrspace(5) %local
  %which = icmp slt i32 %val, 67
  br i1 %which, label %if.then, label %if.end

if.then:
  %stack.base = call ptr addrspace(5) @llvm.sponentry()
  br label %if.end

if.end:
  %ret = phi ptr addrspace(5) [ %stack.base, %if.then ], [ %ptr, %entry ]
  ret ptr addrspace(5) %ret
}

; FIXME: Optimize away the 4 bytes for the sponentry frame index.
; CHECK: ScratchSize: 12

define amdgpu_gfx ptr addrspace(5) @sponentry_gfx_dvgpr_realign(i32 %val) #0 {
; CHECK-LABEL: sponentry_gfx_dvgpr_realign:
; CHECK:       ; %bb.0:
; CHECK-NEXT:    s_wait_loadcnt_dscnt 0x0
; CHECK-NEXT:    s_wait_expcnt 0x0
; CHECK-NEXT:    s_wait_samplecnt 0x0
; CHECK-NEXT:    s_wait_bvhcnt 0x0
; CHECK-NEXT:    s_wait_kmcnt 0x0
; CHECK-NEXT:    s_mov_b32 s0, s33
; CHECK-NEXT:    s_add_co_i32 s33, s32, 0x7f
; CHECK-NEXT:    s_mov_b32 s1, s34
; CHECK-NEXT:    s_wait_alu depctr_sa_sdst(0)
; CHECK-NEXT:    s_and_b32 s33, s33, 0xffffff80
; CHECK-NEXT:    s_mov_b32 s34, s32
; CHECK-NEXT:    s_wait_storecnt 0x0
; CHECK-NEXT:    scratch_store_b32 off, v0, s33 offset:128 scope:SCOPE_SYS
; CHECK-NEXT:    s_wait_storecnt 0x0
; CHECK-NEXT:    s_wait_alu depctr_sa_sdst(0)
; CHECK-NEXT:    v_mov_b32_e32 v0, s34
; CHECK-NEXT:    s_addk_co_i32 s32, 0x180
; CHECK-NEXT:    s_mov_b32 s32, s34
; CHECK-NEXT:    s_mov_b32 s34, s1
; CHECK-NEXT:    s_mov_b32 s33, s0
; CHECK-NEXT:    s_wait_alu depctr_sa_sdst(0)
; CHECK-NEXT:    s_setpc_b64 s[30:31]
  %boop = alloca i32, addrspace(5)
  %local = alloca i32, align 128, addrspace(5)
  store volatile i32 %val, ptr addrspace(5) %local
  %stack.base = call ptr addrspace(5) @llvm.sponentry()
  ret ptr addrspace(5) %stack.base
}

; FIXME: Optimize away the sponentry fixed object.
; CHECK: ScratchSize: 384

define amdgpu_gfx ptr addrspace(5) @sponentry_gfx_stack_args(<32 x i32> %fill.vgprs, i32 %val, ptr addrspace(5) %ptr) #0 {
; DAGISEL-LABEL: sponentry_gfx_stack_args:
; DAGISEL:       ; %bb.0:
; DAGISEL-NEXT:    s_wait_loadcnt_dscnt 0x0
; DAGISEL-NEXT:    s_wait_expcnt 0x0
; DAGISEL-NEXT:    s_wait_samplecnt 0x0
; DAGISEL-NEXT:    s_wait_bvhcnt 0x0
; DAGISEL-NEXT:    s_wait_kmcnt 0x0
; DAGISEL-NEXT:    s_clause 0x1
; DAGISEL-NEXT:    scratch_load_b32 v0, off, s32 offset:4
; DAGISEL-NEXT:    scratch_load_b32 v1, off, s32
; DAGISEL-NEXT:    s_wait_loadcnt 0x0
; DAGISEL-NEXT:    s_wait_storecnt 0x0
; DAGISEL-NEXT:    scratch_store_b32 v0, v1, off scope:SCOPE_SYS
; DAGISEL-NEXT:    s_wait_storecnt 0x0
; DAGISEL-NEXT:    v_mov_b32_e32 v0, s32
; DAGISEL-NEXT:    s_setpc_b64 s[30:31]
;
; GISEL-LABEL: sponentry_gfx_stack_args:
; GISEL:       ; %bb.0:
; GISEL-NEXT:    s_wait_loadcnt_dscnt 0x0
; GISEL-NEXT:    s_wait_expcnt 0x0
; GISEL-NEXT:    s_wait_samplecnt 0x0
; GISEL-NEXT:    s_wait_bvhcnt 0x0
; GISEL-NEXT:    s_wait_kmcnt 0x0
; GISEL-NEXT:    s_clause 0x1
; GISEL-NEXT:    scratch_load_b32 v0, off, s32
; GISEL-NEXT:    scratch_load_b32 v1, off, s32 offset:4
; GISEL-NEXT:    s_wait_loadcnt 0x0
; GISEL-NEXT:    s_wait_storecnt 0x0
; GISEL-NEXT:    scratch_store_b32 v1, v0, off scope:SCOPE_SYS
; GISEL-NEXT:    s_wait_storecnt 0x0
; GISEL-NEXT:    v_mov_b32_e32 v0, s32
; GISEL-NEXT:    s_setpc_b64 s[30:31]
  store volatile i32 %val, ptr addrspace(5) %ptr
  %stack.base = call ptr addrspace(5) @llvm.sponentry()
  ret ptr addrspace(5) %stack.base
}

; CHECK: ScratchSize: 12

define amdgpu_gfx ptr addrspace(5) @sponentry_gfx_dyn_alloc(i32 %val) #0 {
; DAGISEL-LABEL: sponentry_gfx_dyn_alloc:
; DAGISEL:       ; %bb.0:
; DAGISEL-NEXT:    s_wait_loadcnt_dscnt 0x0
; DAGISEL-NEXT:    s_wait_expcnt 0x0
; DAGISEL-NEXT:    s_wait_samplecnt 0x0
; DAGISEL-NEXT:    s_wait_bvhcnt 0x0
; DAGISEL-NEXT:    s_wait_kmcnt 0x0
; DAGISEL-NEXT:    s_mov_b32 s2, s33
; DAGISEL-NEXT:    s_mov_b32 s33, s32
; DAGISEL-NEXT:    s_xor_saveexec_b32 s0, -1
; DAGISEL-NEXT:    s_clause 0x1 ; 8-byte Folded Spill
; DAGISEL-NEXT:    scratch_store_b32 off, v1, s33 offset:4
; DAGISEL-NEXT:    scratch_store_b32 off, v2, s33 offset:8
; DAGISEL-NEXT:    s_wait_alu depctr_sa_sdst(0)
; DAGISEL-NEXT:    s_mov_b32 exec_lo, s0
; DAGISEL-NEXT:    v_lshl_add_u32 v3, v0, 2, 15
; DAGISEL-NEXT:    s_add_co_i32 s32, s32, 16
; DAGISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_1)
; DAGISEL-NEXT:    v_and_b32_e32 v3, -16, v3
; DAGISEL-NEXT:    s_or_saveexec_b32 s0, -1
; DAGISEL-NEXT:    s_wait_alu depctr_sa_sdst(0)
; DAGISEL-NEXT:    v_cndmask_b32_e64 v1, 0, v3, s0
; DAGISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; DAGISEL-NEXT:    v_max_u32_dpp v1, v1, v1 row_shr:1 row_mask:0xf bank_mask:0xf
; DAGISEL-NEXT:    v_max_u32_dpp v1, v1, v1 row_shr:2 row_mask:0xf bank_mask:0xf
; DAGISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; DAGISEL-NEXT:    v_max_u32_dpp v1, v1, v1 row_shr:4 row_mask:0xf bank_mask:0xf
; DAGISEL-NEXT:    v_max_u32_dpp v1, v1, v1 row_shr:8 row_mask:0xf bank_mask:0xf
; DAGISEL-NEXT:    ds_swizzle_b32 v2, v1 offset:swizzle(BROADCAST,32,15)
; DAGISEL-NEXT:    s_wait_dscnt 0x0
; DAGISEL-NEXT:    v_max_u32_e32 v1, v1, v2
; DAGISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; DAGISEL-NEXT:    v_readlane_b32 s1, v1, 31
; DAGISEL-NEXT:    s_mov_b32 exec_lo, s0
; DAGISEL-NEXT:    s_mov_b32 s0, s32
; DAGISEL-NEXT:    s_wait_alu depctr_sa_sdst(0)
; DAGISEL-NEXT:    v_add_nc_u32_e64 v3, s0, s1
; DAGISEL-NEXT:    s_wait_storecnt 0x0
; DAGISEL-NEXT:    scratch_store_b32 off, v0, s0 scope:SCOPE_SYS
; DAGISEL-NEXT:    s_wait_storecnt 0x0
; DAGISEL-NEXT:    v_mov_b32_e32 v0, s33
; DAGISEL-NEXT:    v_readfirstlane_b32 s32, v3
; DAGISEL-NEXT:    s_mov_b32 s32, s33
; DAGISEL-NEXT:    s_xor_saveexec_b32 s0, -1
; DAGISEL-NEXT:    s_clause 0x1 ; 8-byte Folded Reload
; DAGISEL-NEXT:    scratch_load_b32 v1, off, s33 offset:4
; DAGISEL-NEXT:    scratch_load_b32 v2, off, s33 offset:8
; DAGISEL-NEXT:    s_wait_alu depctr_sa_sdst(0)
; DAGISEL-NEXT:    s_mov_b32 exec_lo, s0
; DAGISEL-NEXT:    s_mov_b32 s33, s2
; DAGISEL-NEXT:    s_wait_loadcnt 0x0
; DAGISEL-NEXT:    s_wait_alu depctr_sa_sdst(0)
; DAGISEL-NEXT:    s_setpc_b64 s[30:31]
;
; GISEL-LABEL: sponentry_gfx_dyn_alloc:
; GISEL:       ; %bb.0:
; GISEL-NEXT:    s_wait_loadcnt_dscnt 0x0
; GISEL-NEXT:    s_wait_expcnt 0x0
; GISEL-NEXT:    s_wait_samplecnt 0x0
; GISEL-NEXT:    s_wait_bvhcnt 0x0
; GISEL-NEXT:    s_wait_kmcnt 0x0
; GISEL-NEXT:    s_mov_b32 s3, s33
; GISEL-NEXT:    s_mov_b32 s33, s32
; GISEL-NEXT:    s_xor_saveexec_b32 s0, -1
; GISEL-NEXT:    s_clause 0x1 ; 8-byte Folded Spill
; GISEL-NEXT:    scratch_store_b32 off, v1, s33 offset:4
; GISEL-NEXT:    scratch_store_b32 off, v2, s33 offset:8
; GISEL-NEXT:    s_wait_alu depctr_sa_sdst(0)
; GISEL-NEXT:    s_mov_b32 exec_lo, s0
; GISEL-NEXT:    v_lshl_add_u32 v3, v0, 2, 15
; GISEL-NEXT:    s_add_co_i32 s32, s32, 16
; GISEL-NEXT:    s_wait_alu depctr_sa_sdst(0)
; GISEL-NEXT:    s_mov_b32 s0, s32
; GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_1)
; GISEL-NEXT:    v_and_b32_e32 v3, -16, v3
; GISEL-NEXT:    s_or_saveexec_b32 s1, -1
; GISEL-NEXT:    s_wait_alu depctr_sa_sdst(0)
; GISEL-NEXT:    v_cndmask_b32_e64 v1, 0, v3, s1
; GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GISEL-NEXT:    v_max_u32_dpp v1, v1, v1 row_shr:1 row_mask:0xf bank_mask:0xf
; GISEL-NEXT:    v_max_u32_dpp v1, v1, v1 row_shr:2 row_mask:0xf bank_mask:0xf
; GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GISEL-NEXT:    v_max_u32_dpp v1, v1, v1 row_shr:4 row_mask:0xf bank_mask:0xf
; GISEL-NEXT:    v_max_u32_dpp v1, v1, v1 row_shr:8 row_mask:0xf bank_mask:0xf
; GISEL-NEXT:    ds_swizzle_b32 v2, v1 offset:swizzle(BROADCAST,32,15)
; GISEL-NEXT:    s_wait_dscnt 0x0
; GISEL-NEXT:    v_max_u32_e32 v1, v1, v2
; GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GISEL-NEXT:    v_readlane_b32 s2, v1, 31
; GISEL-NEXT:    s_mov_b32 exec_lo, s1
; GISEL-NEXT:    s_add_co_u32 s32, s0, s2
; GISEL-NEXT:    s_wait_storecnt 0x0
; GISEL-NEXT:    scratch_store_b32 off, v0, s0 scope:SCOPE_SYS
; GISEL-NEXT:    s_wait_storecnt 0x0
; GISEL-NEXT:    v_mov_b32_e32 v0, s33
; GISEL-NEXT:    s_mov_b32 s32, s33
; GISEL-NEXT:    s_xor_saveexec_b32 s0, -1
; GISEL-NEXT:    s_clause 0x1 ; 8-byte Folded Reload
; GISEL-NEXT:    scratch_load_b32 v1, off, s33 offset:4
; GISEL-NEXT:    scratch_load_b32 v2, off, s33 offset:8
; GISEL-NEXT:    s_wait_alu depctr_sa_sdst(0)
; GISEL-NEXT:    s_mov_b32 exec_lo, s0
; GISEL-NEXT:    s_mov_b32 s33, s3
; GISEL-NEXT:    s_wait_loadcnt 0x0
; GISEL-NEXT:    s_wait_alu depctr_sa_sdst(0)
; GISEL-NEXT:    s_setpc_b64 s[30:31]
  %local = alloca i32, i32 %val, addrspace(5)
  store volatile i32 %val, ptr addrspace(5) %local
  %stack.base = call ptr addrspace(5) @llvm.sponentry()
  ret ptr addrspace(5) %stack.base
}

; CHECK: ScratchSize: 16

define amdgpu_cs_chain void @sponentry_cs_chain(i32 %val, ptr addrspace(5) %ptr) #0 {
; DAGISEL-LABEL: sponentry_cs_chain:
; DAGISEL:       ; %bb.0: ; %entry
; DAGISEL-NEXT:    s_wait_loadcnt_dscnt 0x0
; DAGISEL-NEXT:    s_wait_expcnt 0x0
; DAGISEL-NEXT:    s_wait_samplecnt 0x0
; DAGISEL-NEXT:    s_wait_bvhcnt 0x0
; DAGISEL-NEXT:    s_wait_kmcnt 0x0
; DAGISEL-NEXT:    v_mov_b32_e32 v0, v9
; DAGISEL-NEXT:    s_mov_b32 s0, exec_lo
; DAGISEL-NEXT:    s_wait_storecnt 0x0
; DAGISEL-NEXT:    scratch_store_b32 off, v8, s32 offset:4 scope:SCOPE_SYS
; DAGISEL-NEXT:    s_wait_storecnt 0x0
; DAGISEL-NEXT:    v_cmpx_gt_i32_e32 0x43, v8
; DAGISEL-NEXT:  ; %bb.1: ; %if.then
; DAGISEL-NEXT:    v_mov_b32_e32 v0, s32
; DAGISEL-NEXT:  ; %bb.2: ; %if.end
; DAGISEL-NEXT:    s_wait_alu depctr_sa_sdst(0)
; DAGISEL-NEXT:    s_or_b32 exec_lo, exec_lo, s0
; DAGISEL-NEXT:    scratch_store_b32 v9, v0, off scope:SCOPE_SYS
; DAGISEL-NEXT:    s_wait_storecnt 0x0
; DAGISEL-NEXT:    s_alloc_vgpr 0
; DAGISEL-NEXT:    s_endpgm
;
; GISEL-LABEL: sponentry_cs_chain:
; GISEL:       ; %bb.0: ; %entry
; GISEL-NEXT:    s_wait_loadcnt_dscnt 0x0
; GISEL-NEXT:    s_wait_expcnt 0x0
; GISEL-NEXT:    s_wait_samplecnt 0x0
; GISEL-NEXT:    s_wait_bvhcnt 0x0
; GISEL-NEXT:    s_wait_kmcnt 0x0
; GISEL-NEXT:    v_mov_b32_e32 v0, v9
; GISEL-NEXT:    s_mov_b32 s0, exec_lo
; GISEL-NEXT:    s_wait_storecnt 0x0
; GISEL-NEXT:    scratch_store_b32 off, v8, s32 offset:4 scope:SCOPE_SYS
; GISEL-NEXT:    s_wait_storecnt 0x0
; GISEL-NEXT:    v_cmpx_gt_i32_e32 0x43, v8
; GISEL-NEXT:  ; %bb.1: ; %if.then
; GISEL-NEXT:    s_mov_b32 s1, s32
; GISEL-NEXT:    s_wait_alu depctr_sa_sdst(0)
; GISEL-NEXT:    v_mov_b32_e32 v0, s1
; GISEL-NEXT:  ; %bb.2: ; %if.end
; GISEL-NEXT:    s_or_b32 exec_lo, exec_lo, s0
; GISEL-NEXT:    scratch_store_b32 v9, v0, off scope:SCOPE_SYS
; GISEL-NEXT:    s_wait_storecnt 0x0
; GISEL-NEXT:    s_alloc_vgpr 0
; GISEL-NEXT:    s_endpgm
entry:
  %local = alloca i32, addrspace(5)
  store volatile i32 %val, ptr addrspace(5) %local
  %which = icmp slt i32 %val, 67
  br i1 %which, label %if.then, label %if.end

if.then:
  %stack.base = call ptr addrspace(5) @llvm.sponentry()
  br label %if.end

if.end:
  %ret = phi ptr addrspace(5) [ %stack.base, %if.then ], [ %ptr, %entry ]
  store volatile ptr addrspace(5) %ret, ptr addrspace(5) %ptr
  ret void
}

; FIXME: Optimize away the 4 bytes for the sponentry frame index.
; CHECK: ScratchSize: 12

define amdgpu_cs_chain void @sponentry_cs_chain_dvgpr_realign(i32 %val) #0 {
; CHECK-LABEL: sponentry_cs_chain_dvgpr_realign:
; CHECK:       ; %bb.0:
; CHECK-NEXT:    s_wait_loadcnt_dscnt 0x0
; CHECK-NEXT:    s_wait_expcnt 0x0
; CHECK-NEXT:    s_wait_samplecnt 0x0
; CHECK-NEXT:    s_wait_bvhcnt 0x0
; CHECK-NEXT:    s_wait_kmcnt 0x0
; CHECK-NEXT:    s_mov_b32 s34, s32
; CHECK-NEXT:    s_add_co_i32 s33, s32, 0x7f
; CHECK-NEXT:    s_wait_alu depctr_sa_sdst(0)
; CHECK-NEXT:    v_mov_b32_e32 v0, s34
; CHECK-NEXT:    s_and_b32 s33, s33, 0xffffff80
; CHECK-NEXT:    s_addk_co_i32 s32, 0x180
; CHECK-NEXT:    s_wait_storecnt 0x0
; CHECK-NEXT:    scratch_store_b32 off, v8, s33 offset:128 scope:SCOPE_SYS
; CHECK-NEXT:    s_wait_storecnt 0x0
; CHECK-NEXT:    scratch_store_b32 off, v0, s33 offset:128 scope:SCOPE_SYS
; CHECK-NEXT:    s_wait_storecnt 0x0
; CHECK-NEXT:    s_alloc_vgpr 0
; CHECK-NEXT:    s_endpgm
  %boop = alloca i32, addrspace(5)
  %local = alloca i32, align 128, addrspace(5)
  store volatile i32 %val, ptr addrspace(5) %local
  %stack.base = call ptr addrspace(5) @llvm.sponentry()
  store volatile ptr addrspace(5) %stack.base, ptr addrspace(5) %local
  ret void
}

; FIXME: Optimize away the sponentry fixed object.
; CHECK: ScratchSize: 384

define amdgpu_cs_chain void @sponentry_cs_chain_dyn_alloc(i32 %val) #0 {
; DAGISEL-LABEL: sponentry_cs_chain_dyn_alloc:
; DAGISEL:       ; %bb.0:
; DAGISEL-NEXT:    s_wait_loadcnt_dscnt 0x0
; DAGISEL-NEXT:    s_wait_expcnt 0x0
; DAGISEL-NEXT:    s_wait_samplecnt 0x0
; DAGISEL-NEXT:    s_wait_bvhcnt 0x0
; DAGISEL-NEXT:    s_wait_kmcnt 0x0
; DAGISEL-NEXT:    v_lshl_add_u32 v2, v8, 2, 15
; DAGISEL-NEXT:    s_mov_b32 s33, s32
; DAGISEL-NEXT:    s_add_co_i32 s32, s32, 16
; DAGISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_1)
; DAGISEL-NEXT:    v_and_b32_e32 v2, -16, v2
; DAGISEL-NEXT:    s_or_saveexec_b32 s0, -1
; DAGISEL-NEXT:    s_wait_alu depctr_sa_sdst(0)
; DAGISEL-NEXT:    v_cndmask_b32_e64 v0, 0, v2, s0
; DAGISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; DAGISEL-NEXT:    v_max_u32_dpp v0, v0, v0 row_shr:1 row_mask:0xf bank_mask:0xf
; DAGISEL-NEXT:    v_max_u32_dpp v0, v0, v0 row_shr:2 row_mask:0xf bank_mask:0xf
; DAGISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; DAGISEL-NEXT:    v_max_u32_dpp v0, v0, v0 row_shr:4 row_mask:0xf bank_mask:0xf
; DAGISEL-NEXT:    v_max_u32_dpp v0, v0, v0 row_shr:8 row_mask:0xf bank_mask:0xf
; DAGISEL-NEXT:    ds_swizzle_b32 v1, v0 offset:swizzle(BROADCAST,32,15)
; DAGISEL-NEXT:    s_wait_dscnt 0x0
; DAGISEL-NEXT:    v_max_u32_e32 v0, v0, v1
; DAGISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; DAGISEL-NEXT:    v_readlane_b32 s1, v0, 31
; DAGISEL-NEXT:    s_mov_b32 exec_lo, s0
; DAGISEL-NEXT:    s_mov_b32 s0, s32
; DAGISEL-NEXT:    v_mov_b32_e32 v3, s33
; DAGISEL-NEXT:    s_wait_alu depctr_sa_sdst(0)
; DAGISEL-NEXT:    v_add_nc_u32_e64 v2, s0, s1
; DAGISEL-NEXT:    s_wait_storecnt 0x0
; DAGISEL-NEXT:    scratch_store_b32 off, v8, s0 scope:SCOPE_SYS
; DAGISEL-NEXT:    s_wait_storecnt 0x0
; DAGISEL-NEXT:    scratch_store_b32 off, v3, s0 scope:SCOPE_SYS
; DAGISEL-NEXT:    s_wait_storecnt 0x0
; DAGISEL-NEXT:    v_readfirstlane_b32 s32, v2
; DAGISEL-NEXT:    s_alloc_vgpr 0
; DAGISEL-NEXT:    s_endpgm
;
; GISEL-LABEL: sponentry_cs_chain_dyn_alloc:
; GISEL:       ; %bb.0:
; GISEL-NEXT:    s_wait_loadcnt_dscnt 0x0
; GISEL-NEXT:    s_wait_expcnt 0x0
; GISEL-NEXT:    s_wait_samplecnt 0x0
; GISEL-NEXT:    s_wait_bvhcnt 0x0
; GISEL-NEXT:    s_wait_kmcnt 0x0
; GISEL-NEXT:    v_lshl_add_u32 v2, v8, 2, 15
; GISEL-NEXT:    s_mov_b32 s33, s32
; GISEL-NEXT:    s_add_co_i32 s32, s32, 16
; GISEL-NEXT:    s_wait_alu depctr_sa_sdst(0)
; GISEL-NEXT:    s_mov_b32 s0, s32
; GISEL-NEXT:    v_and_b32_e32 v2, -16, v2
; GISEL-NEXT:    s_or_saveexec_b32 s1, -1
; GISEL-NEXT:    s_wait_alu depctr_sa_sdst(0)
; GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GISEL-NEXT:    v_cndmask_b32_e64 v0, 0, v2, s1
; GISEL-NEXT:    v_max_u32_dpp v0, v0, v0 row_shr:1 row_mask:0xf bank_mask:0xf
; GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GISEL-NEXT:    v_max_u32_dpp v0, v0, v0 row_shr:2 row_mask:0xf bank_mask:0xf
; GISEL-NEXT:    v_max_u32_dpp v0, v0, v0 row_shr:4 row_mask:0xf bank_mask:0xf
; GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_3) | instid1(VALU_DEP_1)
; GISEL-NEXT:    v_max_u32_dpp v0, v0, v0 row_shr:8 row_mask:0xf bank_mask:0xf
; GISEL-NEXT:    ds_swizzle_b32 v1, v0 offset:swizzle(BROADCAST,32,15)
; GISEL-NEXT:    s_wait_dscnt 0x0
; GISEL-NEXT:    v_max_u32_e32 v0, v0, v1
; GISEL-NEXT:    v_readlane_b32 s2, v0, 31
; GISEL-NEXT:    s_mov_b32 exec_lo, s1
; GISEL-NEXT:    v_mov_b32_e32 v2, s33
; GISEL-NEXT:    s_wait_storecnt 0x0
; GISEL-NEXT:    scratch_store_b32 off, v8, s0 scope:SCOPE_SYS
; GISEL-NEXT:    s_wait_storecnt 0x0
; GISEL-NEXT:    s_add_co_u32 s32, s0, s2
; GISEL-NEXT:    scratch_store_b32 off, v2, s0 scope:SCOPE_SYS
; GISEL-NEXT:    s_wait_storecnt 0x0
; GISEL-NEXT:    s_alloc_vgpr 0
; GISEL-NEXT:    s_endpgm
  %local = alloca i32, i32 %val, addrspace(5)
  store volatile i32 %val, ptr addrspace(5) %local
  %stack.base = call ptr addrspace(5) @llvm.sponentry()
  store volatile ptr addrspace(5) %stack.base, ptr addrspace(5) %local
  ret void
}

; CHECK: ScratchSize: 16

attributes #0 = { nounwind "amdgpu-dynamic-vgpr-block-size"="16" }
attributes #1 = { nounwind "amdgpu-dynamic-vgpr-block-size"="32" }
attributes #2 = { nounwind "amdgpu-dynamic-vgpr-block-size"="0" }