// COM: Direct HotSwap entry displacement is supported across gfx125x.

// RUN: %clang -target amdgcn-amd-amdhsa -mcpu=gfx1251 -nostdlib %s -o %t.gfx1251.elf
// RUN: hotswap-rewrite %t.gfx1251.elf \
// RUN:   amdgcn-amd-amdhsa--gfx1251 amdgcn-amd-amdhsa--gfx1251 \
// RUN:   --entry-trampolines --output %t.gfx1251.out.elf \
// RUN:   | %FileCheck --check-prefix=API %s
// RUN: %llvm-objdump -d %t.gfx1251.out.elf | %FileCheck --check-prefix=DISASM %s

// COM: Direct displacement does not need scratch SGPRs, so it remains valid at
// COM: the architectural SGPR limit.
// RUN: sed 's/.sgpr_count: 1/.sgpr_count: 105/' %s > %t.highsgpr.s
// RUN: %clang -target amdgcn-amd-amdhsa -mcpu=gfx1251 -nostdlib \
// RUN:   %t.highsgpr.s -o %t.highsgpr.elf
// RUN: hotswap-rewrite %t.highsgpr.elf \
// RUN:   amdgcn-amd-amdhsa--gfx1251 amdgcn-amd-amdhsa--gfx1251 \
// RUN:   --entry-trampolines --output %t.highsgpr.out.elf \
// RUN:   | %FileCheck --check-prefix=API %s
// RUN: %llvm-objdump -d %t.highsgpr.out.elf \
// RUN:   | %FileCheck --check-prefix=DISASM %s

// COM: gfx12-5-generic omitted: its entry-trampoline path needs newer
// COM: gfx12-5-generic assembler support than this branch's LLVM base.

// API: RESULT: SUCCESS

// DISASM-LABEL: <entry_tramp_family_kernel>:
// DISASM-NEXT: global_prefetch_b8 v0, s[0:1] scope:SCOPE_SE
// DISASM-NEXT: v_nop
// DISASM-NEXT: v_mov_b32_e32 v0, 0
// DISASM-NEXT: s_endpgm
// DISASM-NOT: s_get_pc_i64

.amdgcn_target "amdgcn-amd-amdhsa--gfx1251"
.text
.globl entry_tramp_family_kernel
.p2align 8
.type entry_tramp_family_kernel,@function
entry_tramp_family_kernel:
  v_mov_b32_e32 v0, 0
  s_endpgm
.Lentry_tramp_family_kernel_end:
.size entry_tramp_family_kernel, .Lentry_tramp_family_kernel_end-entry_tramp_family_kernel

.rodata
.p2align 8
.amdhsa_kernel entry_tramp_family_kernel
  .amdhsa_next_free_vgpr 1
  .amdhsa_next_free_sgpr 1
  .amdhsa_inst_pref_size 7
.end_amdhsa_kernel

.amdgpu_metadata
  amdhsa.version:
    - 3
    - 0
  amdhsa.kernels:
    - .name: entry_tramp_family_kernel
      .symbol: entry_tramp_family_kernel.kd
      .sgpr_count: 1
      .vgpr_count: 1
      .kernarg_segment_size: 0
      .group_segment_fixed_size: 0
      .private_segment_fixed_size: 0
      .kernarg_segment_align: 8
      .wavefront_size: 64
      .max_flat_workgroup_size: 256
.end_amdgpu_metadata