llvm/test/CodeGen/AMDGPU/fcmp.f16.ll

; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 2
; RUN:  llc -amdgpu-scalarize-global-loads=false -mtriple=amdgcn -verify-machineinstrs < %s | FileCheck -allow-deprecated-dag-overlap -enable-var-scope -check-prefixes=SI %s
; RUN:  llc -amdgpu-scalarize-global-loads=false -mtriple=amdgcn -mcpu=fiji -mattr=-flat-for-global -verify-machineinstrs < %s | FileCheck -allow-deprecated-dag-overlap -enable-var-scope -check-prefixes=VI %s
; RUN:  llc -amdgpu-scalarize-global-loads=false -mtriple=amdgcn -mcpu=gfx1100 -mattr=-flat-for-global -verify-machineinstrs < %s | FileCheck -allow-deprecated-dag-overlap -enable-var-scope -check-prefixes=GFX11 %s

define amdgpu_kernel void @fcmp_f16_lt(
; SI-LABEL: fcmp_f16_lt:
; SI:       ; %bb.0: ; %entry
; SI-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xd
; SI-NEXT:    s_mov_b32 s11, 0xf000
; SI-NEXT:    s_mov_b32 s10, -1
; SI-NEXT:    s_mov_b32 s14, s10
; SI-NEXT:    s_mov_b32 s15, s11
; SI-NEXT:    s_mov_b32 s2, s10
; SI-NEXT:    s_mov_b32 s3, s11
; SI-NEXT:    s_waitcnt lgkmcnt(0)
; SI-NEXT:    s_mov_b32 s12, s6
; SI-NEXT:    s_mov_b32 s13, s7
; SI-NEXT:    buffer_load_ushort v0, off, s[12:15], 0 glc
; SI-NEXT:    s_waitcnt vmcnt(0)
; SI-NEXT:    buffer_load_ushort v1, off, s[0:3], 0 glc
; SI-NEXT:    s_waitcnt vmcnt(0)
; SI-NEXT:    s_mov_b32 s8, s4
; SI-NEXT:    s_mov_b32 s9, s5
; SI-NEXT:    v_cvt_f32_f16_e32 v0, v0
; SI-NEXT:    v_cvt_f32_f16_e32 v1, v1
; SI-NEXT:    v_cmp_lt_f32_e32 vcc, v0, v1
; SI-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc
; SI-NEXT:    buffer_store_dword v0, off, s[8:11], 0
; SI-NEXT:    s_endpgm
;
; VI-LABEL: fcmp_f16_lt:
; VI:       ; %bb.0: ; %entry
; VI-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
; VI-NEXT:    s_load_dwordx2 s[8:9], s[0:1], 0x34
; VI-NEXT:    s_mov_b32 s3, 0xf000
; VI-NEXT:    s_mov_b32 s2, -1
; VI-NEXT:    s_mov_b32 s14, s2
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    s_mov_b32 s12, s6
; VI-NEXT:    s_mov_b32 s13, s7
; VI-NEXT:    s_mov_b32 s15, s3
; VI-NEXT:    s_mov_b32 s10, s2
; VI-NEXT:    s_mov_b32 s11, s3
; VI-NEXT:    buffer_load_ushort v0, off, s[12:15], 0 glc
; VI-NEXT:    s_waitcnt vmcnt(0)
; VI-NEXT:    buffer_load_ushort v1, off, s[8:11], 0 glc
; VI-NEXT:    s_waitcnt vmcnt(0)
; VI-NEXT:    s_mov_b32 s0, s4
; VI-NEXT:    s_mov_b32 s1, s5
; VI-NEXT:    v_cmp_lt_f16_e32 vcc, v0, v1
; VI-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc
; VI-NEXT:    buffer_store_dword v0, off, s[0:3], 0
; VI-NEXT:    s_endpgm
;
; GFX11-LABEL: fcmp_f16_lt:
; GFX11:       ; %bb.0: ; %entry
; GFX11-NEXT:    s_clause 0x1
; GFX11-NEXT:    s_load_b128 s[4:7], s[0:1], 0x24
; GFX11-NEXT:    s_load_b64 s[0:1], s[0:1], 0x34
; GFX11-NEXT:    s_mov_b32 s10, -1
; GFX11-NEXT:    s_mov_b32 s11, 0x31016000
; GFX11-NEXT:    s_mov_b32 s14, s10
; GFX11-NEXT:    s_mov_b32 s15, s11
; GFX11-NEXT:    s_mov_b32 s2, s10
; GFX11-NEXT:    s_mov_b32 s3, s11
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    s_mov_b32 s12, s6
; GFX11-NEXT:    s_mov_b32 s13, s7
; GFX11-NEXT:    buffer_load_u16 v0, off, s[12:15], 0 glc dlc
; GFX11-NEXT:    s_waitcnt vmcnt(0)
; GFX11-NEXT:    buffer_load_u16 v1, off, s[0:3], 0 glc dlc
; GFX11-NEXT:    s_waitcnt vmcnt(0)
; GFX11-NEXT:    s_mov_b32 s8, s4
; GFX11-NEXT:    s_mov_b32 s9, s5
; GFX11-NEXT:    v_cmp_lt_f16_e32 vcc_lo, v0, v1
; GFX11-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc_lo
; GFX11-NEXT:    buffer_store_b32 v0, off, s[8:11], 0
; GFX11-NEXT:    s_nop 0
; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
; GFX11-NEXT:    s_endpgm
    ptr addrspace(1) %r,
    ptr addrspace(1) %a,
    ptr addrspace(1) %b) {
entry:
  %a.val = load volatile half, ptr addrspace(1) %a
  %b.val = load volatile half, ptr addrspace(1) %b
  %r.val = fcmp olt half %a.val, %b.val
  %r.val.sext = sext i1 %r.val to i32
  store i32 %r.val.sext, ptr addrspace(1) %r
  ret void
}

define amdgpu_kernel void @fcmp_f16_lt_abs(
; SI-LABEL: fcmp_f16_lt_abs:
; SI:       ; %bb.0: ; %entry
; SI-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xd
; SI-NEXT:    s_mov_b32 s11, 0xf000
; SI-NEXT:    s_mov_b32 s10, -1
; SI-NEXT:    s_mov_b32 s14, s10
; SI-NEXT:    s_mov_b32 s15, s11
; SI-NEXT:    s_mov_b32 s2, s10
; SI-NEXT:    s_mov_b32 s3, s11
; SI-NEXT:    s_waitcnt lgkmcnt(0)
; SI-NEXT:    s_mov_b32 s12, s6
; SI-NEXT:    s_mov_b32 s13, s7
; SI-NEXT:    buffer_load_ushort v0, off, s[12:15], 0 glc
; SI-NEXT:    s_waitcnt vmcnt(0)
; SI-NEXT:    buffer_load_ushort v1, off, s[0:3], 0 glc
; SI-NEXT:    s_waitcnt vmcnt(0)
; SI-NEXT:    s_mov_b32 s8, s4
; SI-NEXT:    s_mov_b32 s9, s5
; SI-NEXT:    v_cvt_f32_f16_e64 v0, |v0|
; SI-NEXT:    v_cvt_f32_f16_e64 v1, |v1|
; SI-NEXT:    v_cmp_lt_f32_e32 vcc, v0, v1
; SI-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc
; SI-NEXT:    buffer_store_dword v0, off, s[8:11], 0
; SI-NEXT:    s_endpgm
;
; VI-LABEL: fcmp_f16_lt_abs:
; VI:       ; %bb.0: ; %entry
; VI-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
; VI-NEXT:    s_load_dwordx2 s[8:9], s[0:1], 0x34
; VI-NEXT:    s_mov_b32 s3, 0xf000
; VI-NEXT:    s_mov_b32 s2, -1
; VI-NEXT:    s_mov_b32 s14, s2
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    s_mov_b32 s12, s6
; VI-NEXT:    s_mov_b32 s13, s7
; VI-NEXT:    s_mov_b32 s15, s3
; VI-NEXT:    s_mov_b32 s10, s2
; VI-NEXT:    s_mov_b32 s11, s3
; VI-NEXT:    buffer_load_ushort v0, off, s[12:15], 0 glc
; VI-NEXT:    s_waitcnt vmcnt(0)
; VI-NEXT:    buffer_load_ushort v1, off, s[8:11], 0 glc
; VI-NEXT:    s_waitcnt vmcnt(0)
; VI-NEXT:    s_mov_b32 s0, s4
; VI-NEXT:    s_mov_b32 s1, s5
; VI-NEXT:    v_cmp_lt_f16_e64 s[4:5], |v0|, |v1|
; VI-NEXT:    v_cndmask_b32_e64 v0, 0, -1, s[4:5]
; VI-NEXT:    buffer_store_dword v0, off, s[0:3], 0
; VI-NEXT:    s_endpgm
;
; GFX11-LABEL: fcmp_f16_lt_abs:
; GFX11:       ; %bb.0: ; %entry
; GFX11-NEXT:    s_clause 0x1
; GFX11-NEXT:    s_load_b128 s[4:7], s[0:1], 0x24
; GFX11-NEXT:    s_load_b64 s[0:1], s[0:1], 0x34
; GFX11-NEXT:    s_mov_b32 s10, -1
; GFX11-NEXT:    s_mov_b32 s11, 0x31016000
; GFX11-NEXT:    s_mov_b32 s14, s10
; GFX11-NEXT:    s_mov_b32 s15, s11
; GFX11-NEXT:    s_mov_b32 s2, s10
; GFX11-NEXT:    s_mov_b32 s3, s11
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    s_mov_b32 s12, s6
; GFX11-NEXT:    s_mov_b32 s13, s7
; GFX11-NEXT:    buffer_load_u16 v0, off, s[12:15], 0 glc dlc
; GFX11-NEXT:    s_waitcnt vmcnt(0)
; GFX11-NEXT:    buffer_load_u16 v1, off, s[0:3], 0 glc dlc
; GFX11-NEXT:    s_waitcnt vmcnt(0)
; GFX11-NEXT:    s_mov_b32 s8, s4
; GFX11-NEXT:    s_mov_b32 s9, s5
; GFX11-NEXT:    v_cmp_lt_f16_e64 s0, |v0|, |v1|
; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX11-NEXT:    v_cndmask_b32_e64 v0, 0, -1, s0
; GFX11-NEXT:    buffer_store_b32 v0, off, s[8:11], 0
; GFX11-NEXT:    s_nop 0
; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
; GFX11-NEXT:    s_endpgm
    ptr addrspace(1) %r,
    ptr addrspace(1) %a,
    ptr addrspace(1) %b) {
entry:
  %a.val = load volatile half, ptr addrspace(1) %a
  %b.val = load volatile half, ptr addrspace(1) %b
  %a.abs = call half @llvm.fabs.f16(half %a.val)
  %b.abs = call half @llvm.fabs.f16(half %b.val)
  %r.val = fcmp olt half %a.abs, %b.abs
  %r.val.sext = sext i1 %r.val to i32
  store i32 %r.val.sext, ptr addrspace(1) %r
  ret void
}

define amdgpu_kernel void @fcmp_f16_eq(
; SI-LABEL: fcmp_f16_eq:
; SI:       ; %bb.0: ; %entry
; SI-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xd
; SI-NEXT:    s_mov_b32 s11, 0xf000
; SI-NEXT:    s_mov_b32 s10, -1
; SI-NEXT:    s_mov_b32 s14, s10
; SI-NEXT:    s_mov_b32 s15, s11
; SI-NEXT:    s_mov_b32 s2, s10
; SI-NEXT:    s_mov_b32 s3, s11
; SI-NEXT:    s_waitcnt lgkmcnt(0)
; SI-NEXT:    s_mov_b32 s12, s6
; SI-NEXT:    s_mov_b32 s13, s7
; SI-NEXT:    buffer_load_ushort v0, off, s[12:15], 0 glc
; SI-NEXT:    s_waitcnt vmcnt(0)
; SI-NEXT:    buffer_load_ushort v1, off, s[0:3], 0 glc
; SI-NEXT:    s_waitcnt vmcnt(0)
; SI-NEXT:    s_mov_b32 s8, s4
; SI-NEXT:    s_mov_b32 s9, s5
; SI-NEXT:    v_cvt_f32_f16_e32 v0, v0
; SI-NEXT:    v_cvt_f32_f16_e32 v1, v1
; SI-NEXT:    v_cmp_eq_f32_e32 vcc, v0, v1
; SI-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc
; SI-NEXT:    buffer_store_dword v0, off, s[8:11], 0
; SI-NEXT:    s_endpgm
;
; VI-LABEL: fcmp_f16_eq:
; VI:       ; %bb.0: ; %entry
; VI-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
; VI-NEXT:    s_load_dwordx2 s[8:9], s[0:1], 0x34
; VI-NEXT:    s_mov_b32 s3, 0xf000
; VI-NEXT:    s_mov_b32 s2, -1
; VI-NEXT:    s_mov_b32 s14, s2
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    s_mov_b32 s12, s6
; VI-NEXT:    s_mov_b32 s13, s7
; VI-NEXT:    s_mov_b32 s15, s3
; VI-NEXT:    s_mov_b32 s10, s2
; VI-NEXT:    s_mov_b32 s11, s3
; VI-NEXT:    buffer_load_ushort v0, off, s[12:15], 0 glc
; VI-NEXT:    s_waitcnt vmcnt(0)
; VI-NEXT:    buffer_load_ushort v1, off, s[8:11], 0 glc
; VI-NEXT:    s_waitcnt vmcnt(0)
; VI-NEXT:    s_mov_b32 s0, s4
; VI-NEXT:    s_mov_b32 s1, s5
; VI-NEXT:    v_cmp_eq_f16_e32 vcc, v0, v1
; VI-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc
; VI-NEXT:    buffer_store_dword v0, off, s[0:3], 0
; VI-NEXT:    s_endpgm
;
; GFX11-LABEL: fcmp_f16_eq:
; GFX11:       ; %bb.0: ; %entry
; GFX11-NEXT:    s_clause 0x1
; GFX11-NEXT:    s_load_b128 s[4:7], s[0:1], 0x24
; GFX11-NEXT:    s_load_b64 s[0:1], s[0:1], 0x34
; GFX11-NEXT:    s_mov_b32 s10, -1
; GFX11-NEXT:    s_mov_b32 s11, 0x31016000
; GFX11-NEXT:    s_mov_b32 s14, s10
; GFX11-NEXT:    s_mov_b32 s15, s11
; GFX11-NEXT:    s_mov_b32 s2, s10
; GFX11-NEXT:    s_mov_b32 s3, s11
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    s_mov_b32 s12, s6
; GFX11-NEXT:    s_mov_b32 s13, s7
; GFX11-NEXT:    buffer_load_u16 v0, off, s[12:15], 0 glc dlc
; GFX11-NEXT:    s_waitcnt vmcnt(0)
; GFX11-NEXT:    buffer_load_u16 v1, off, s[0:3], 0 glc dlc
; GFX11-NEXT:    s_waitcnt vmcnt(0)
; GFX11-NEXT:    s_mov_b32 s8, s4
; GFX11-NEXT:    s_mov_b32 s9, s5
; GFX11-NEXT:    v_cmp_eq_f16_e32 vcc_lo, v0, v1
; GFX11-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc_lo
; GFX11-NEXT:    buffer_store_b32 v0, off, s[8:11], 0
; GFX11-NEXT:    s_nop 0
; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
; GFX11-NEXT:    s_endpgm
    ptr addrspace(1) %r,
    ptr addrspace(1) %a,
    ptr addrspace(1) %b) {
entry:
  %a.val = load volatile half, ptr addrspace(1) %a
  %b.val = load volatile half, ptr addrspace(1) %b
  %r.val = fcmp oeq half %a.val, %b.val
  %r.val.sext = sext i1 %r.val to i32
  store i32 %r.val.sext, ptr addrspace(1) %r
  ret void
}

define amdgpu_kernel void @fcmp_f16_le(
; SI-LABEL: fcmp_f16_le:
; SI:       ; %bb.0: ; %entry
; SI-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xd
; SI-NEXT:    s_mov_b32 s11, 0xf000
; SI-NEXT:    s_mov_b32 s10, -1
; SI-NEXT:    s_mov_b32 s14, s10
; SI-NEXT:    s_mov_b32 s15, s11
; SI-NEXT:    s_mov_b32 s2, s10
; SI-NEXT:    s_mov_b32 s3, s11
; SI-NEXT:    s_waitcnt lgkmcnt(0)
; SI-NEXT:    s_mov_b32 s12, s6
; SI-NEXT:    s_mov_b32 s13, s7
; SI-NEXT:    buffer_load_ushort v0, off, s[12:15], 0 glc
; SI-NEXT:    s_waitcnt vmcnt(0)
; SI-NEXT:    buffer_load_ushort v1, off, s[0:3], 0 glc
; SI-NEXT:    s_waitcnt vmcnt(0)
; SI-NEXT:    s_mov_b32 s8, s4
; SI-NEXT:    s_mov_b32 s9, s5
; SI-NEXT:    v_cvt_f32_f16_e32 v0, v0
; SI-NEXT:    v_cvt_f32_f16_e32 v1, v1
; SI-NEXT:    v_cmp_le_f32_e32 vcc, v0, v1
; SI-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc
; SI-NEXT:    buffer_store_dword v0, off, s[8:11], 0
; SI-NEXT:    s_endpgm
;
; VI-LABEL: fcmp_f16_le:
; VI:       ; %bb.0: ; %entry
; VI-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
; VI-NEXT:    s_load_dwordx2 s[8:9], s[0:1], 0x34
; VI-NEXT:    s_mov_b32 s3, 0xf000
; VI-NEXT:    s_mov_b32 s2, -1
; VI-NEXT:    s_mov_b32 s14, s2
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    s_mov_b32 s12, s6
; VI-NEXT:    s_mov_b32 s13, s7
; VI-NEXT:    s_mov_b32 s15, s3
; VI-NEXT:    s_mov_b32 s10, s2
; VI-NEXT:    s_mov_b32 s11, s3
; VI-NEXT:    buffer_load_ushort v0, off, s[12:15], 0 glc
; VI-NEXT:    s_waitcnt vmcnt(0)
; VI-NEXT:    buffer_load_ushort v1, off, s[8:11], 0 glc
; VI-NEXT:    s_waitcnt vmcnt(0)
; VI-NEXT:    s_mov_b32 s0, s4
; VI-NEXT:    s_mov_b32 s1, s5
; VI-NEXT:    v_cmp_le_f16_e32 vcc, v0, v1
; VI-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc
; VI-NEXT:    buffer_store_dword v0, off, s[0:3], 0
; VI-NEXT:    s_endpgm
;
; GFX11-LABEL: fcmp_f16_le:
; GFX11:       ; %bb.0: ; %entry
; GFX11-NEXT:    s_clause 0x1
; GFX11-NEXT:    s_load_b128 s[4:7], s[0:1], 0x24
; GFX11-NEXT:    s_load_b64 s[0:1], s[0:1], 0x34
; GFX11-NEXT:    s_mov_b32 s10, -1
; GFX11-NEXT:    s_mov_b32 s11, 0x31016000
; GFX11-NEXT:    s_mov_b32 s14, s10
; GFX11-NEXT:    s_mov_b32 s15, s11
; GFX11-NEXT:    s_mov_b32 s2, s10
; GFX11-NEXT:    s_mov_b32 s3, s11
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    s_mov_b32 s12, s6
; GFX11-NEXT:    s_mov_b32 s13, s7
; GFX11-NEXT:    buffer_load_u16 v0, off, s[12:15], 0 glc dlc
; GFX11-NEXT:    s_waitcnt vmcnt(0)
; GFX11-NEXT:    buffer_load_u16 v1, off, s[0:3], 0 glc dlc
; GFX11-NEXT:    s_waitcnt vmcnt(0)
; GFX11-NEXT:    s_mov_b32 s8, s4
; GFX11-NEXT:    s_mov_b32 s9, s5
; GFX11-NEXT:    v_cmp_le_f16_e32 vcc_lo, v0, v1
; GFX11-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc_lo
; GFX11-NEXT:    buffer_store_b32 v0, off, s[8:11], 0
; GFX11-NEXT:    s_nop 0
; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
; GFX11-NEXT:    s_endpgm
    ptr addrspace(1) %r,
    ptr addrspace(1) %a,
    ptr addrspace(1) %b) {
entry:
  %a.val = load volatile half, ptr addrspace(1) %a
  %b.val = load volatile half, ptr addrspace(1) %b
  %r.val = fcmp ole half %a.val, %b.val
  %r.val.sext = sext i1 %r.val to i32
  store i32 %r.val.sext, ptr addrspace(1) %r
  ret void
}

define amdgpu_kernel void @fcmp_f16_gt(
; SI-LABEL: fcmp_f16_gt:
; SI:       ; %bb.0: ; %entry
; SI-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xd
; SI-NEXT:    s_mov_b32 s11, 0xf000
; SI-NEXT:    s_mov_b32 s10, -1
; SI-NEXT:    s_mov_b32 s14, s10
; SI-NEXT:    s_mov_b32 s15, s11
; SI-NEXT:    s_mov_b32 s2, s10
; SI-NEXT:    s_mov_b32 s3, s11
; SI-NEXT:    s_waitcnt lgkmcnt(0)
; SI-NEXT:    s_mov_b32 s12, s6
; SI-NEXT:    s_mov_b32 s13, s7
; SI-NEXT:    buffer_load_ushort v0, off, s[12:15], 0 glc
; SI-NEXT:    s_waitcnt vmcnt(0)
; SI-NEXT:    buffer_load_ushort v1, off, s[0:3], 0 glc
; SI-NEXT:    s_waitcnt vmcnt(0)
; SI-NEXT:    s_mov_b32 s8, s4
; SI-NEXT:    s_mov_b32 s9, s5
; SI-NEXT:    v_cvt_f32_f16_e32 v0, v0
; SI-NEXT:    v_cvt_f32_f16_e32 v1, v1
; SI-NEXT:    v_cmp_gt_f32_e32 vcc, v0, v1
; SI-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc
; SI-NEXT:    buffer_store_dword v0, off, s[8:11], 0
; SI-NEXT:    s_endpgm
;
; VI-LABEL: fcmp_f16_gt:
; VI:       ; %bb.0: ; %entry
; VI-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
; VI-NEXT:    s_load_dwordx2 s[8:9], s[0:1], 0x34
; VI-NEXT:    s_mov_b32 s3, 0xf000
; VI-NEXT:    s_mov_b32 s2, -1
; VI-NEXT:    s_mov_b32 s14, s2
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    s_mov_b32 s12, s6
; VI-NEXT:    s_mov_b32 s13, s7
; VI-NEXT:    s_mov_b32 s15, s3
; VI-NEXT:    s_mov_b32 s10, s2
; VI-NEXT:    s_mov_b32 s11, s3
; VI-NEXT:    buffer_load_ushort v0, off, s[12:15], 0 glc
; VI-NEXT:    s_waitcnt vmcnt(0)
; VI-NEXT:    buffer_load_ushort v1, off, s[8:11], 0 glc
; VI-NEXT:    s_waitcnt vmcnt(0)
; VI-NEXT:    s_mov_b32 s0, s4
; VI-NEXT:    s_mov_b32 s1, s5
; VI-NEXT:    v_cmp_gt_f16_e32 vcc, v0, v1
; VI-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc
; VI-NEXT:    buffer_store_dword v0, off, s[0:3], 0
; VI-NEXT:    s_endpgm
;
; GFX11-LABEL: fcmp_f16_gt:
; GFX11:       ; %bb.0: ; %entry
; GFX11-NEXT:    s_clause 0x1
; GFX11-NEXT:    s_load_b128 s[4:7], s[0:1], 0x24
; GFX11-NEXT:    s_load_b64 s[0:1], s[0:1], 0x34
; GFX11-NEXT:    s_mov_b32 s10, -1
; GFX11-NEXT:    s_mov_b32 s11, 0x31016000
; GFX11-NEXT:    s_mov_b32 s14, s10
; GFX11-NEXT:    s_mov_b32 s15, s11
; GFX11-NEXT:    s_mov_b32 s2, s10
; GFX11-NEXT:    s_mov_b32 s3, s11
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    s_mov_b32 s12, s6
; GFX11-NEXT:    s_mov_b32 s13, s7
; GFX11-NEXT:    buffer_load_u16 v0, off, s[12:15], 0 glc dlc
; GFX11-NEXT:    s_waitcnt vmcnt(0)
; GFX11-NEXT:    buffer_load_u16 v1, off, s[0:3], 0 glc dlc
; GFX11-NEXT:    s_waitcnt vmcnt(0)
; GFX11-NEXT:    s_mov_b32 s8, s4
; GFX11-NEXT:    s_mov_b32 s9, s5
; GFX11-NEXT:    v_cmp_gt_f16_e32 vcc_lo, v0, v1
; GFX11-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc_lo
; GFX11-NEXT:    buffer_store_b32 v0, off, s[8:11], 0
; GFX11-NEXT:    s_nop 0
; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
; GFX11-NEXT:    s_endpgm
    ptr addrspace(1) %r,
    ptr addrspace(1) %a,
    ptr addrspace(1) %b) {
entry:
  %a.val = load volatile half, ptr addrspace(1) %a
  %b.val = load volatile half, ptr addrspace(1) %b
  %r.val = fcmp ogt half %a.val, %b.val
  %r.val.sext = sext i1 %r.val to i32
  store i32 %r.val.sext, ptr addrspace(1) %r
  ret void
}

define amdgpu_kernel void @fcmp_f16_lg(
; SI-LABEL: fcmp_f16_lg:
; SI:       ; %bb.0: ; %entry
; SI-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xd
; SI-NEXT:    s_mov_b32 s11, 0xf000
; SI-NEXT:    s_mov_b32 s10, -1
; SI-NEXT:    s_mov_b32 s14, s10
; SI-NEXT:    s_mov_b32 s15, s11
; SI-NEXT:    s_mov_b32 s2, s10
; SI-NEXT:    s_mov_b32 s3, s11
; SI-NEXT:    s_waitcnt lgkmcnt(0)
; SI-NEXT:    s_mov_b32 s12, s6
; SI-NEXT:    s_mov_b32 s13, s7
; SI-NEXT:    buffer_load_ushort v0, off, s[12:15], 0 glc
; SI-NEXT:    s_waitcnt vmcnt(0)
; SI-NEXT:    buffer_load_ushort v1, off, s[0:3], 0 glc
; SI-NEXT:    s_waitcnt vmcnt(0)
; SI-NEXT:    s_mov_b32 s8, s4
; SI-NEXT:    s_mov_b32 s9, s5
; SI-NEXT:    v_cvt_f32_f16_e32 v0, v0
; SI-NEXT:    v_cvt_f32_f16_e32 v1, v1
; SI-NEXT:    v_cmp_lg_f32_e32 vcc, v0, v1
; SI-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc
; SI-NEXT:    buffer_store_dword v0, off, s[8:11], 0
; SI-NEXT:    s_endpgm
;
; VI-LABEL: fcmp_f16_lg:
; VI:       ; %bb.0: ; %entry
; VI-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
; VI-NEXT:    s_load_dwordx2 s[8:9], s[0:1], 0x34
; VI-NEXT:    s_mov_b32 s3, 0xf000
; VI-NEXT:    s_mov_b32 s2, -1
; VI-NEXT:    s_mov_b32 s14, s2
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    s_mov_b32 s12, s6
; VI-NEXT:    s_mov_b32 s13, s7
; VI-NEXT:    s_mov_b32 s15, s3
; VI-NEXT:    s_mov_b32 s10, s2
; VI-NEXT:    s_mov_b32 s11, s3
; VI-NEXT:    buffer_load_ushort v0, off, s[12:15], 0 glc
; VI-NEXT:    s_waitcnt vmcnt(0)
; VI-NEXT:    buffer_load_ushort v1, off, s[8:11], 0 glc
; VI-NEXT:    s_waitcnt vmcnt(0)
; VI-NEXT:    s_mov_b32 s0, s4
; VI-NEXT:    s_mov_b32 s1, s5
; VI-NEXT:    v_cmp_lg_f16_e32 vcc, v0, v1
; VI-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc
; VI-NEXT:    buffer_store_dword v0, off, s[0:3], 0
; VI-NEXT:    s_endpgm
;
; GFX11-LABEL: fcmp_f16_lg:
; GFX11:       ; %bb.0: ; %entry
; GFX11-NEXT:    s_clause 0x1
; GFX11-NEXT:    s_load_b128 s[4:7], s[0:1], 0x24
; GFX11-NEXT:    s_load_b64 s[0:1], s[0:1], 0x34
; GFX11-NEXT:    s_mov_b32 s10, -1
; GFX11-NEXT:    s_mov_b32 s11, 0x31016000
; GFX11-NEXT:    s_mov_b32 s14, s10
; GFX11-NEXT:    s_mov_b32 s15, s11
; GFX11-NEXT:    s_mov_b32 s2, s10
; GFX11-NEXT:    s_mov_b32 s3, s11
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    s_mov_b32 s12, s6
; GFX11-NEXT:    s_mov_b32 s13, s7
; GFX11-NEXT:    buffer_load_u16 v0, off, s[12:15], 0 glc dlc
; GFX11-NEXT:    s_waitcnt vmcnt(0)
; GFX11-NEXT:    buffer_load_u16 v1, off, s[0:3], 0 glc dlc
; GFX11-NEXT:    s_waitcnt vmcnt(0)
; GFX11-NEXT:    s_mov_b32 s8, s4
; GFX11-NEXT:    s_mov_b32 s9, s5
; GFX11-NEXT:    v_cmp_lg_f16_e32 vcc_lo, v0, v1
; GFX11-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc_lo
; GFX11-NEXT:    buffer_store_b32 v0, off, s[8:11], 0
; GFX11-NEXT:    s_nop 0
; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
; GFX11-NEXT:    s_endpgm
    ptr addrspace(1) %r,
    ptr addrspace(1) %a,
    ptr addrspace(1) %b) {
entry:
  %a.val = load volatile half, ptr addrspace(1) %a
  %b.val = load volatile half, ptr addrspace(1) %b
  %r.val = fcmp one half %a.val, %b.val
  %r.val.sext = sext i1 %r.val to i32
  store i32 %r.val.sext, ptr addrspace(1) %r
  ret void
}

define amdgpu_kernel void @fcmp_f16_ge(
; SI-LABEL: fcmp_f16_ge:
; SI:       ; %bb.0: ; %entry
; SI-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xd
; SI-NEXT:    s_mov_b32 s11, 0xf000
; SI-NEXT:    s_mov_b32 s10, -1
; SI-NEXT:    s_mov_b32 s14, s10
; SI-NEXT:    s_mov_b32 s15, s11
; SI-NEXT:    s_mov_b32 s2, s10
; SI-NEXT:    s_mov_b32 s3, s11
; SI-NEXT:    s_waitcnt lgkmcnt(0)
; SI-NEXT:    s_mov_b32 s12, s6
; SI-NEXT:    s_mov_b32 s13, s7
; SI-NEXT:    buffer_load_ushort v0, off, s[12:15], 0 glc
; SI-NEXT:    s_waitcnt vmcnt(0)
; SI-NEXT:    buffer_load_ushort v1, off, s[0:3], 0 glc
; SI-NEXT:    s_waitcnt vmcnt(0)
; SI-NEXT:    s_mov_b32 s8, s4
; SI-NEXT:    s_mov_b32 s9, s5
; SI-NEXT:    v_cvt_f32_f16_e32 v0, v0
; SI-NEXT:    v_cvt_f32_f16_e32 v1, v1
; SI-NEXT:    v_cmp_ge_f32_e32 vcc, v0, v1
; SI-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc
; SI-NEXT:    buffer_store_dword v0, off, s[8:11], 0
; SI-NEXT:    s_endpgm
;
; VI-LABEL: fcmp_f16_ge:
; VI:       ; %bb.0: ; %entry
; VI-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
; VI-NEXT:    s_load_dwordx2 s[8:9], s[0:1], 0x34
; VI-NEXT:    s_mov_b32 s3, 0xf000
; VI-NEXT:    s_mov_b32 s2, -1
; VI-NEXT:    s_mov_b32 s14, s2
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    s_mov_b32 s12, s6
; VI-NEXT:    s_mov_b32 s13, s7
; VI-NEXT:    s_mov_b32 s15, s3
; VI-NEXT:    s_mov_b32 s10, s2
; VI-NEXT:    s_mov_b32 s11, s3
; VI-NEXT:    buffer_load_ushort v0, off, s[12:15], 0 glc
; VI-NEXT:    s_waitcnt vmcnt(0)
; VI-NEXT:    buffer_load_ushort v1, off, s[8:11], 0 glc
; VI-NEXT:    s_waitcnt vmcnt(0)
; VI-NEXT:    s_mov_b32 s0, s4
; VI-NEXT:    s_mov_b32 s1, s5
; VI-NEXT:    v_cmp_ge_f16_e32 vcc, v0, v1
; VI-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc
; VI-NEXT:    buffer_store_dword v0, off, s[0:3], 0
; VI-NEXT:    s_endpgm
;
; GFX11-LABEL: fcmp_f16_ge:
; GFX11:       ; %bb.0: ; %entry
; GFX11-NEXT:    s_clause 0x1
; GFX11-NEXT:    s_load_b128 s[4:7], s[0:1], 0x24
; GFX11-NEXT:    s_load_b64 s[0:1], s[0:1], 0x34
; GFX11-NEXT:    s_mov_b32 s10, -1
; GFX11-NEXT:    s_mov_b32 s11, 0x31016000
; GFX11-NEXT:    s_mov_b32 s14, s10
; GFX11-NEXT:    s_mov_b32 s15, s11
; GFX11-NEXT:    s_mov_b32 s2, s10
; GFX11-NEXT:    s_mov_b32 s3, s11
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    s_mov_b32 s12, s6
; GFX11-NEXT:    s_mov_b32 s13, s7
; GFX11-NEXT:    buffer_load_u16 v0, off, s[12:15], 0 glc dlc
; GFX11-NEXT:    s_waitcnt vmcnt(0)
; GFX11-NEXT:    buffer_load_u16 v1, off, s[0:3], 0 glc dlc
; GFX11-NEXT:    s_waitcnt vmcnt(0)
; GFX11-NEXT:    s_mov_b32 s8, s4
; GFX11-NEXT:    s_mov_b32 s9, s5
; GFX11-NEXT:    v_cmp_ge_f16_e32 vcc_lo, v0, v1
; GFX11-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc_lo
; GFX11-NEXT:    buffer_store_b32 v0, off, s[8:11], 0
; GFX11-NEXT:    s_nop 0
; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
; GFX11-NEXT:    s_endpgm
    ptr addrspace(1) %r,
    ptr addrspace(1) %a,
    ptr addrspace(1) %b) {
entry:
  %a.val = load volatile half, ptr addrspace(1) %a
  %b.val = load volatile half, ptr addrspace(1) %b
  %r.val = fcmp oge half %a.val, %b.val
  %r.val.sext = sext i1 %r.val to i32
  store i32 %r.val.sext, ptr addrspace(1) %r
  ret void
}

define amdgpu_kernel void @fcmp_f16_o(
; SI-LABEL: fcmp_f16_o:
; SI:       ; %bb.0: ; %entry
; SI-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xd
; SI-NEXT:    s_mov_b32 s11, 0xf000
; SI-NEXT:    s_mov_b32 s10, -1
; SI-NEXT:    s_mov_b32 s14, s10
; SI-NEXT:    s_mov_b32 s15, s11
; SI-NEXT:    s_mov_b32 s2, s10
; SI-NEXT:    s_mov_b32 s3, s11
; SI-NEXT:    s_waitcnt lgkmcnt(0)
; SI-NEXT:    s_mov_b32 s12, s6
; SI-NEXT:    s_mov_b32 s13, s7
; SI-NEXT:    buffer_load_ushort v0, off, s[12:15], 0 glc
; SI-NEXT:    s_waitcnt vmcnt(0)
; SI-NEXT:    buffer_load_ushort v1, off, s[0:3], 0 glc
; SI-NEXT:    s_waitcnt vmcnt(0)
; SI-NEXT:    s_mov_b32 s8, s4
; SI-NEXT:    s_mov_b32 s9, s5
; SI-NEXT:    v_cvt_f32_f16_e32 v0, v0
; SI-NEXT:    v_cvt_f32_f16_e32 v1, v1
; SI-NEXT:    v_cmp_o_f32_e32 vcc, v0, v1
; SI-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc
; SI-NEXT:    buffer_store_dword v0, off, s[8:11], 0
; SI-NEXT:    s_endpgm
;
; VI-LABEL: fcmp_f16_o:
; VI:       ; %bb.0: ; %entry
; VI-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
; VI-NEXT:    s_load_dwordx2 s[8:9], s[0:1], 0x34
; VI-NEXT:    s_mov_b32 s3, 0xf000
; VI-NEXT:    s_mov_b32 s2, -1
; VI-NEXT:    s_mov_b32 s14, s2
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    s_mov_b32 s12, s6
; VI-NEXT:    s_mov_b32 s13, s7
; VI-NEXT:    s_mov_b32 s15, s3
; VI-NEXT:    s_mov_b32 s10, s2
; VI-NEXT:    s_mov_b32 s11, s3
; VI-NEXT:    buffer_load_ushort v0, off, s[12:15], 0 glc
; VI-NEXT:    s_waitcnt vmcnt(0)
; VI-NEXT:    buffer_load_ushort v1, off, s[8:11], 0 glc
; VI-NEXT:    s_waitcnt vmcnt(0)
; VI-NEXT:    s_mov_b32 s0, s4
; VI-NEXT:    s_mov_b32 s1, s5
; VI-NEXT:    v_cmp_o_f16_e32 vcc, v0, v1
; VI-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc
; VI-NEXT:    buffer_store_dword v0, off, s[0:3], 0
; VI-NEXT:    s_endpgm
;
; GFX11-LABEL: fcmp_f16_o:
; GFX11:       ; %bb.0: ; %entry
; GFX11-NEXT:    s_clause 0x1
; GFX11-NEXT:    s_load_b128 s[4:7], s[0:1], 0x24
; GFX11-NEXT:    s_load_b64 s[0:1], s[0:1], 0x34
; GFX11-NEXT:    s_mov_b32 s10, -1
; GFX11-NEXT:    s_mov_b32 s11, 0x31016000
; GFX11-NEXT:    s_mov_b32 s14, s10
; GFX11-NEXT:    s_mov_b32 s15, s11
; GFX11-NEXT:    s_mov_b32 s2, s10
; GFX11-NEXT:    s_mov_b32 s3, s11
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    s_mov_b32 s12, s6
; GFX11-NEXT:    s_mov_b32 s13, s7
; GFX11-NEXT:    buffer_load_u16 v0, off, s[12:15], 0 glc dlc
; GFX11-NEXT:    s_waitcnt vmcnt(0)
; GFX11-NEXT:    buffer_load_u16 v1, off, s[0:3], 0 glc dlc
; GFX11-NEXT:    s_waitcnt vmcnt(0)
; GFX11-NEXT:    s_mov_b32 s8, s4
; GFX11-NEXT:    s_mov_b32 s9, s5
; GFX11-NEXT:    v_cmp_o_f16_e32 vcc_lo, v0, v1
; GFX11-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc_lo
; GFX11-NEXT:    buffer_store_b32 v0, off, s[8:11], 0
; GFX11-NEXT:    s_nop 0
; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
; GFX11-NEXT:    s_endpgm
    ptr addrspace(1) %r,
    ptr addrspace(1) %a,
    ptr addrspace(1) %b) {
entry:
  %a.val = load volatile half, ptr addrspace(1) %a
  %b.val = load volatile half, ptr addrspace(1) %b
  %r.val = fcmp ord half %a.val, %b.val
  %r.val.sext = sext i1 %r.val to i32
  store i32 %r.val.sext, ptr addrspace(1) %r
  ret void
}

define amdgpu_kernel void @fcmp_f16_u(
; SI-LABEL: fcmp_f16_u:
; SI:       ; %bb.0: ; %entry
; SI-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xd
; SI-NEXT:    s_mov_b32 s11, 0xf000
; SI-NEXT:    s_mov_b32 s10, -1
; SI-NEXT:    s_mov_b32 s14, s10
; SI-NEXT:    s_mov_b32 s15, s11
; SI-NEXT:    s_mov_b32 s2, s10
; SI-NEXT:    s_mov_b32 s3, s11
; SI-NEXT:    s_waitcnt lgkmcnt(0)
; SI-NEXT:    s_mov_b32 s12, s6
; SI-NEXT:    s_mov_b32 s13, s7
; SI-NEXT:    buffer_load_ushort v0, off, s[12:15], 0 glc
; SI-NEXT:    s_waitcnt vmcnt(0)
; SI-NEXT:    buffer_load_ushort v1, off, s[0:3], 0 glc
; SI-NEXT:    s_waitcnt vmcnt(0)
; SI-NEXT:    s_mov_b32 s8, s4
; SI-NEXT:    s_mov_b32 s9, s5
; SI-NEXT:    v_cvt_f32_f16_e32 v0, v0
; SI-NEXT:    v_cvt_f32_f16_e32 v1, v1
; SI-NEXT:    v_cmp_u_f32_e32 vcc, v0, v1
; SI-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc
; SI-NEXT:    buffer_store_dword v0, off, s[8:11], 0
; SI-NEXT:    s_endpgm
;
; VI-LABEL: fcmp_f16_u:
; VI:       ; %bb.0: ; %entry
; VI-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
; VI-NEXT:    s_load_dwordx2 s[8:9], s[0:1], 0x34
; VI-NEXT:    s_mov_b32 s3, 0xf000
; VI-NEXT:    s_mov_b32 s2, -1
; VI-NEXT:    s_mov_b32 s14, s2
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    s_mov_b32 s12, s6
; VI-NEXT:    s_mov_b32 s13, s7
; VI-NEXT:    s_mov_b32 s15, s3
; VI-NEXT:    s_mov_b32 s10, s2
; VI-NEXT:    s_mov_b32 s11, s3
; VI-NEXT:    buffer_load_ushort v0, off, s[12:15], 0 glc
; VI-NEXT:    s_waitcnt vmcnt(0)
; VI-NEXT:    buffer_load_ushort v1, off, s[8:11], 0 glc
; VI-NEXT:    s_waitcnt vmcnt(0)
; VI-NEXT:    s_mov_b32 s0, s4
; VI-NEXT:    s_mov_b32 s1, s5
; VI-NEXT:    v_cmp_u_f16_e32 vcc, v0, v1
; VI-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc
; VI-NEXT:    buffer_store_dword v0, off, s[0:3], 0
; VI-NEXT:    s_endpgm
;
; GFX11-LABEL: fcmp_f16_u:
; GFX11:       ; %bb.0: ; %entry
; GFX11-NEXT:    s_clause 0x1
; GFX11-NEXT:    s_load_b128 s[4:7], s[0:1], 0x24
; GFX11-NEXT:    s_load_b64 s[0:1], s[0:1], 0x34
; GFX11-NEXT:    s_mov_b32 s10, -1
; GFX11-NEXT:    s_mov_b32 s11, 0x31016000
; GFX11-NEXT:    s_mov_b32 s14, s10
; GFX11-NEXT:    s_mov_b32 s15, s11
; GFX11-NEXT:    s_mov_b32 s2, s10
; GFX11-NEXT:    s_mov_b32 s3, s11
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    s_mov_b32 s12, s6
; GFX11-NEXT:    s_mov_b32 s13, s7
; GFX11-NEXT:    buffer_load_u16 v0, off, s[12:15], 0 glc dlc
; GFX11-NEXT:    s_waitcnt vmcnt(0)
; GFX11-NEXT:    buffer_load_u16 v1, off, s[0:3], 0 glc dlc
; GFX11-NEXT:    s_waitcnt vmcnt(0)
; GFX11-NEXT:    s_mov_b32 s8, s4
; GFX11-NEXT:    s_mov_b32 s9, s5
; GFX11-NEXT:    v_cmp_u_f16_e32 vcc_lo, v0, v1
; GFX11-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc_lo
; GFX11-NEXT:    buffer_store_b32 v0, off, s[8:11], 0
; GFX11-NEXT:    s_nop 0
; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
; GFX11-NEXT:    s_endpgm
    ptr addrspace(1) %r,
    ptr addrspace(1) %a,
    ptr addrspace(1) %b) {
entry:
  %a.val = load volatile half, ptr addrspace(1) %a
  %b.val = load volatile half, ptr addrspace(1) %b
  %r.val = fcmp uno half %a.val, %b.val
  %r.val.sext = sext i1 %r.val to i32
  store i32 %r.val.sext, ptr addrspace(1) %r
  ret void
}

define amdgpu_kernel void @fcmp_f16_nge(
; SI-LABEL: fcmp_f16_nge:
; SI:       ; %bb.0: ; %entry
; SI-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xd
; SI-NEXT:    s_mov_b32 s11, 0xf000
; SI-NEXT:    s_mov_b32 s10, -1
; SI-NEXT:    s_mov_b32 s14, s10
; SI-NEXT:    s_mov_b32 s15, s11
; SI-NEXT:    s_mov_b32 s2, s10
; SI-NEXT:    s_mov_b32 s3, s11
; SI-NEXT:    s_waitcnt lgkmcnt(0)
; SI-NEXT:    s_mov_b32 s12, s6
; SI-NEXT:    s_mov_b32 s13, s7
; SI-NEXT:    buffer_load_ushort v0, off, s[12:15], 0 glc
; SI-NEXT:    s_waitcnt vmcnt(0)
; SI-NEXT:    buffer_load_ushort v1, off, s[0:3], 0 glc
; SI-NEXT:    s_waitcnt vmcnt(0)
; SI-NEXT:    s_mov_b32 s8, s4
; SI-NEXT:    s_mov_b32 s9, s5
; SI-NEXT:    v_cvt_f32_f16_e32 v0, v0
; SI-NEXT:    v_cvt_f32_f16_e32 v1, v1
; SI-NEXT:    v_cmp_nge_f32_e32 vcc, v0, v1
; SI-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc
; SI-NEXT:    buffer_store_dword v0, off, s[8:11], 0
; SI-NEXT:    s_endpgm
;
; VI-LABEL: fcmp_f16_nge:
; VI:       ; %bb.0: ; %entry
; VI-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
; VI-NEXT:    s_load_dwordx2 s[8:9], s[0:1], 0x34
; VI-NEXT:    s_mov_b32 s3, 0xf000
; VI-NEXT:    s_mov_b32 s2, -1
; VI-NEXT:    s_mov_b32 s14, s2
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    s_mov_b32 s12, s6
; VI-NEXT:    s_mov_b32 s13, s7
; VI-NEXT:    s_mov_b32 s15, s3
; VI-NEXT:    s_mov_b32 s10, s2
; VI-NEXT:    s_mov_b32 s11, s3
; VI-NEXT:    buffer_load_ushort v0, off, s[12:15], 0 glc
; VI-NEXT:    s_waitcnt vmcnt(0)
; VI-NEXT:    buffer_load_ushort v1, off, s[8:11], 0 glc
; VI-NEXT:    s_waitcnt vmcnt(0)
; VI-NEXT:    s_mov_b32 s0, s4
; VI-NEXT:    s_mov_b32 s1, s5
; VI-NEXT:    v_cmp_nge_f16_e32 vcc, v0, v1
; VI-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc
; VI-NEXT:    buffer_store_dword v0, off, s[0:3], 0
; VI-NEXT:    s_endpgm
;
; GFX11-LABEL: fcmp_f16_nge:
; GFX11:       ; %bb.0: ; %entry
; GFX11-NEXT:    s_clause 0x1
; GFX11-NEXT:    s_load_b128 s[4:7], s[0:1], 0x24
; GFX11-NEXT:    s_load_b64 s[0:1], s[0:1], 0x34
; GFX11-NEXT:    s_mov_b32 s10, -1
; GFX11-NEXT:    s_mov_b32 s11, 0x31016000
; GFX11-NEXT:    s_mov_b32 s14, s10
; GFX11-NEXT:    s_mov_b32 s15, s11
; GFX11-NEXT:    s_mov_b32 s2, s10
; GFX11-NEXT:    s_mov_b32 s3, s11
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    s_mov_b32 s12, s6
; GFX11-NEXT:    s_mov_b32 s13, s7
; GFX11-NEXT:    buffer_load_u16 v0, off, s[12:15], 0 glc dlc
; GFX11-NEXT:    s_waitcnt vmcnt(0)
; GFX11-NEXT:    buffer_load_u16 v1, off, s[0:3], 0 glc dlc
; GFX11-NEXT:    s_waitcnt vmcnt(0)
; GFX11-NEXT:    s_mov_b32 s8, s4
; GFX11-NEXT:    s_mov_b32 s9, s5
; GFX11-NEXT:    v_cmp_nge_f16_e32 vcc_lo, v0, v1
; GFX11-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc_lo
; GFX11-NEXT:    buffer_store_b32 v0, off, s[8:11], 0
; GFX11-NEXT:    s_nop 0
; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
; GFX11-NEXT:    s_endpgm
    ptr addrspace(1) %r,
    ptr addrspace(1) %a,
    ptr addrspace(1) %b) {
entry:
  %a.val = load volatile half, ptr addrspace(1) %a
  %b.val = load volatile half, ptr addrspace(1) %b
  %r.val = fcmp ult half %a.val, %b.val
  %r.val.sext = sext i1 %r.val to i32
  store i32 %r.val.sext, ptr addrspace(1) %r
  ret void
}

define amdgpu_kernel void @fcmp_f16_nlg(
; SI-LABEL: fcmp_f16_nlg:
; SI:       ; %bb.0: ; %entry
; SI-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xd
; SI-NEXT:    s_mov_b32 s11, 0xf000
; SI-NEXT:    s_mov_b32 s10, -1
; SI-NEXT:    s_mov_b32 s14, s10
; SI-NEXT:    s_mov_b32 s15, s11
; SI-NEXT:    s_mov_b32 s2, s10
; SI-NEXT:    s_mov_b32 s3, s11
; SI-NEXT:    s_waitcnt lgkmcnt(0)
; SI-NEXT:    s_mov_b32 s12, s6
; SI-NEXT:    s_mov_b32 s13, s7
; SI-NEXT:    buffer_load_ushort v0, off, s[12:15], 0 glc
; SI-NEXT:    s_waitcnt vmcnt(0)
; SI-NEXT:    buffer_load_ushort v1, off, s[0:3], 0 glc
; SI-NEXT:    s_waitcnt vmcnt(0)
; SI-NEXT:    s_mov_b32 s8, s4
; SI-NEXT:    s_mov_b32 s9, s5
; SI-NEXT:    v_cvt_f32_f16_e32 v0, v0
; SI-NEXT:    v_cvt_f32_f16_e32 v1, v1
; SI-NEXT:    v_cmp_nlg_f32_e32 vcc, v0, v1
; SI-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc
; SI-NEXT:    buffer_store_dword v0, off, s[8:11], 0
; SI-NEXT:    s_endpgm
;
; VI-LABEL: fcmp_f16_nlg:
; VI:       ; %bb.0: ; %entry
; VI-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
; VI-NEXT:    s_load_dwordx2 s[8:9], s[0:1], 0x34
; VI-NEXT:    s_mov_b32 s3, 0xf000
; VI-NEXT:    s_mov_b32 s2, -1
; VI-NEXT:    s_mov_b32 s14, s2
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    s_mov_b32 s12, s6
; VI-NEXT:    s_mov_b32 s13, s7
; VI-NEXT:    s_mov_b32 s15, s3
; VI-NEXT:    s_mov_b32 s10, s2
; VI-NEXT:    s_mov_b32 s11, s3
; VI-NEXT:    buffer_load_ushort v0, off, s[12:15], 0 glc
; VI-NEXT:    s_waitcnt vmcnt(0)
; VI-NEXT:    buffer_load_ushort v1, off, s[8:11], 0 glc
; VI-NEXT:    s_waitcnt vmcnt(0)
; VI-NEXT:    s_mov_b32 s0, s4
; VI-NEXT:    s_mov_b32 s1, s5
; VI-NEXT:    v_cmp_nlg_f16_e32 vcc, v0, v1
; VI-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc
; VI-NEXT:    buffer_store_dword v0, off, s[0:3], 0
; VI-NEXT:    s_endpgm
;
; GFX11-LABEL: fcmp_f16_nlg:
; GFX11:       ; %bb.0: ; %entry
; GFX11-NEXT:    s_clause 0x1
; GFX11-NEXT:    s_load_b128 s[4:7], s[0:1], 0x24
; GFX11-NEXT:    s_load_b64 s[0:1], s[0:1], 0x34
; GFX11-NEXT:    s_mov_b32 s10, -1
; GFX11-NEXT:    s_mov_b32 s11, 0x31016000
; GFX11-NEXT:    s_mov_b32 s14, s10
; GFX11-NEXT:    s_mov_b32 s15, s11
; GFX11-NEXT:    s_mov_b32 s2, s10
; GFX11-NEXT:    s_mov_b32 s3, s11
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    s_mov_b32 s12, s6
; GFX11-NEXT:    s_mov_b32 s13, s7
; GFX11-NEXT:    buffer_load_u16 v0, off, s[12:15], 0 glc dlc
; GFX11-NEXT:    s_waitcnt vmcnt(0)
; GFX11-NEXT:    buffer_load_u16 v1, off, s[0:3], 0 glc dlc
; GFX11-NEXT:    s_waitcnt vmcnt(0)
; GFX11-NEXT:    s_mov_b32 s8, s4
; GFX11-NEXT:    s_mov_b32 s9, s5
; GFX11-NEXT:    v_cmp_nlg_f16_e32 vcc_lo, v0, v1
; GFX11-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc_lo
; GFX11-NEXT:    buffer_store_b32 v0, off, s[8:11], 0
; GFX11-NEXT:    s_nop 0
; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
; GFX11-NEXT:    s_endpgm
    ptr addrspace(1) %r,
    ptr addrspace(1) %a,
    ptr addrspace(1) %b) {
entry:
  %a.val = load volatile half, ptr addrspace(1) %a
  %b.val = load volatile half, ptr addrspace(1) %b
  %r.val = fcmp ueq half %a.val, %b.val
  %r.val.sext = sext i1 %r.val to i32
  store i32 %r.val.sext, ptr addrspace(1) %r
  ret void
}

define amdgpu_kernel void @fcmp_f16_ngt(
; SI-LABEL: fcmp_f16_ngt:
; SI:       ; %bb.0: ; %entry
; SI-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xd
; SI-NEXT:    s_mov_b32 s11, 0xf000
; SI-NEXT:    s_mov_b32 s10, -1
; SI-NEXT:    s_mov_b32 s14, s10
; SI-NEXT:    s_mov_b32 s15, s11
; SI-NEXT:    s_mov_b32 s2, s10
; SI-NEXT:    s_mov_b32 s3, s11
; SI-NEXT:    s_waitcnt lgkmcnt(0)
; SI-NEXT:    s_mov_b32 s12, s6
; SI-NEXT:    s_mov_b32 s13, s7
; SI-NEXT:    buffer_load_ushort v0, off, s[12:15], 0 glc
; SI-NEXT:    s_waitcnt vmcnt(0)
; SI-NEXT:    buffer_load_ushort v1, off, s[0:3], 0 glc
; SI-NEXT:    s_waitcnt vmcnt(0)
; SI-NEXT:    s_mov_b32 s8, s4
; SI-NEXT:    s_mov_b32 s9, s5
; SI-NEXT:    v_cvt_f32_f16_e32 v0, v0
; SI-NEXT:    v_cvt_f32_f16_e32 v1, v1
; SI-NEXT:    v_cmp_ngt_f32_e32 vcc, v0, v1
; SI-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc
; SI-NEXT:    buffer_store_dword v0, off, s[8:11], 0
; SI-NEXT:    s_endpgm
;
; VI-LABEL: fcmp_f16_ngt:
; VI:       ; %bb.0: ; %entry
; VI-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
; VI-NEXT:    s_load_dwordx2 s[8:9], s[0:1], 0x34
; VI-NEXT:    s_mov_b32 s3, 0xf000
; VI-NEXT:    s_mov_b32 s2, -1
; VI-NEXT:    s_mov_b32 s14, s2
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    s_mov_b32 s12, s6
; VI-NEXT:    s_mov_b32 s13, s7
; VI-NEXT:    s_mov_b32 s15, s3
; VI-NEXT:    s_mov_b32 s10, s2
; VI-NEXT:    s_mov_b32 s11, s3
; VI-NEXT:    buffer_load_ushort v0, off, s[12:15], 0 glc
; VI-NEXT:    s_waitcnt vmcnt(0)
; VI-NEXT:    buffer_load_ushort v1, off, s[8:11], 0 glc
; VI-NEXT:    s_waitcnt vmcnt(0)
; VI-NEXT:    s_mov_b32 s0, s4
; VI-NEXT:    s_mov_b32 s1, s5
; VI-NEXT:    v_cmp_ngt_f16_e32 vcc, v0, v1
; VI-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc
; VI-NEXT:    buffer_store_dword v0, off, s[0:3], 0
; VI-NEXT:    s_endpgm
;
; GFX11-LABEL: fcmp_f16_ngt:
; GFX11:       ; %bb.0: ; %entry
; GFX11-NEXT:    s_clause 0x1
; GFX11-NEXT:    s_load_b128 s[4:7], s[0:1], 0x24
; GFX11-NEXT:    s_load_b64 s[0:1], s[0:1], 0x34
; GFX11-NEXT:    s_mov_b32 s10, -1
; GFX11-NEXT:    s_mov_b32 s11, 0x31016000
; GFX11-NEXT:    s_mov_b32 s14, s10
; GFX11-NEXT:    s_mov_b32 s15, s11
; GFX11-NEXT:    s_mov_b32 s2, s10
; GFX11-NEXT:    s_mov_b32 s3, s11
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    s_mov_b32 s12, s6
; GFX11-NEXT:    s_mov_b32 s13, s7
; GFX11-NEXT:    buffer_load_u16 v0, off, s[12:15], 0 glc dlc
; GFX11-NEXT:    s_waitcnt vmcnt(0)
; GFX11-NEXT:    buffer_load_u16 v1, off, s[0:3], 0 glc dlc
; GFX11-NEXT:    s_waitcnt vmcnt(0)
; GFX11-NEXT:    s_mov_b32 s8, s4
; GFX11-NEXT:    s_mov_b32 s9, s5
; GFX11-NEXT:    v_cmp_ngt_f16_e32 vcc_lo, v0, v1
; GFX11-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc_lo
; GFX11-NEXT:    buffer_store_b32 v0, off, s[8:11], 0
; GFX11-NEXT:    s_nop 0
; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
; GFX11-NEXT:    s_endpgm
    ptr addrspace(1) %r,
    ptr addrspace(1) %a,
    ptr addrspace(1) %b) {
entry:
  %a.val = load volatile half, ptr addrspace(1) %a
  %b.val = load volatile half, ptr addrspace(1) %b
  %r.val = fcmp ule half %a.val, %b.val
  %r.val.sext = sext i1 %r.val to i32
  store i32 %r.val.sext, ptr addrspace(1) %r
  ret void
}

define amdgpu_kernel void @fcmp_f16_nle(
; SI-LABEL: fcmp_f16_nle:
; SI:       ; %bb.0: ; %entry
; SI-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xd
; SI-NEXT:    s_mov_b32 s11, 0xf000
; SI-NEXT:    s_mov_b32 s10, -1
; SI-NEXT:    s_mov_b32 s14, s10
; SI-NEXT:    s_mov_b32 s15, s11
; SI-NEXT:    s_mov_b32 s2, s10
; SI-NEXT:    s_mov_b32 s3, s11
; SI-NEXT:    s_waitcnt lgkmcnt(0)
; SI-NEXT:    s_mov_b32 s12, s6
; SI-NEXT:    s_mov_b32 s13, s7
; SI-NEXT:    buffer_load_ushort v0, off, s[12:15], 0 glc
; SI-NEXT:    s_waitcnt vmcnt(0)
; SI-NEXT:    buffer_load_ushort v1, off, s[0:3], 0 glc
; SI-NEXT:    s_waitcnt vmcnt(0)
; SI-NEXT:    s_mov_b32 s8, s4
; SI-NEXT:    s_mov_b32 s9, s5
; SI-NEXT:    v_cvt_f32_f16_e32 v0, v0
; SI-NEXT:    v_cvt_f32_f16_e32 v1, v1
; SI-NEXT:    v_cmp_nle_f32_e32 vcc, v0, v1
; SI-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc
; SI-NEXT:    buffer_store_dword v0, off, s[8:11], 0
; SI-NEXT:    s_endpgm
;
; VI-LABEL: fcmp_f16_nle:
; VI:       ; %bb.0: ; %entry
; VI-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
; VI-NEXT:    s_load_dwordx2 s[8:9], s[0:1], 0x34
; VI-NEXT:    s_mov_b32 s3, 0xf000
; VI-NEXT:    s_mov_b32 s2, -1
; VI-NEXT:    s_mov_b32 s14, s2
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    s_mov_b32 s12, s6
; VI-NEXT:    s_mov_b32 s13, s7
; VI-NEXT:    s_mov_b32 s15, s3
; VI-NEXT:    s_mov_b32 s10, s2
; VI-NEXT:    s_mov_b32 s11, s3
; VI-NEXT:    buffer_load_ushort v0, off, s[12:15], 0 glc
; VI-NEXT:    s_waitcnt vmcnt(0)
; VI-NEXT:    buffer_load_ushort v1, off, s[8:11], 0 glc
; VI-NEXT:    s_waitcnt vmcnt(0)
; VI-NEXT:    s_mov_b32 s0, s4
; VI-NEXT:    s_mov_b32 s1, s5
; VI-NEXT:    v_cmp_nle_f16_e32 vcc, v0, v1
; VI-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc
; VI-NEXT:    buffer_store_dword v0, off, s[0:3], 0
; VI-NEXT:    s_endpgm
;
; GFX11-LABEL: fcmp_f16_nle:
; GFX11:       ; %bb.0: ; %entry
; GFX11-NEXT:    s_clause 0x1
; GFX11-NEXT:    s_load_b128 s[4:7], s[0:1], 0x24
; GFX11-NEXT:    s_load_b64 s[0:1], s[0:1], 0x34
; GFX11-NEXT:    s_mov_b32 s10, -1
; GFX11-NEXT:    s_mov_b32 s11, 0x31016000
; GFX11-NEXT:    s_mov_b32 s14, s10
; GFX11-NEXT:    s_mov_b32 s15, s11
; GFX11-NEXT:    s_mov_b32 s2, s10
; GFX11-NEXT:    s_mov_b32 s3, s11
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    s_mov_b32 s12, s6
; GFX11-NEXT:    s_mov_b32 s13, s7
; GFX11-NEXT:    buffer_load_u16 v0, off, s[12:15], 0 glc dlc
; GFX11-NEXT:    s_waitcnt vmcnt(0)
; GFX11-NEXT:    buffer_load_u16 v1, off, s[0:3], 0 glc dlc
; GFX11-NEXT:    s_waitcnt vmcnt(0)
; GFX11-NEXT:    s_mov_b32 s8, s4
; GFX11-NEXT:    s_mov_b32 s9, s5
; GFX11-NEXT:    v_cmp_nle_f16_e32 vcc_lo, v0, v1
; GFX11-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc_lo
; GFX11-NEXT:    buffer_store_b32 v0, off, s[8:11], 0
; GFX11-NEXT:    s_nop 0
; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
; GFX11-NEXT:    s_endpgm
    ptr addrspace(1) %r,
    ptr addrspace(1) %a,
    ptr addrspace(1) %b) {
entry:
  %a.val = load volatile half, ptr addrspace(1) %a
  %b.val = load volatile half, ptr addrspace(1) %b
  %r.val = fcmp ugt half %a.val, %b.val
  %r.val.sext = sext i1 %r.val to i32
  store i32 %r.val.sext, ptr addrspace(1) %r
  ret void
}

define amdgpu_kernel void @fcmp_f16_neq(
; SI-LABEL: fcmp_f16_neq:
; SI:       ; %bb.0: ; %entry
; SI-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xd
; SI-NEXT:    s_mov_b32 s11, 0xf000
; SI-NEXT:    s_mov_b32 s10, -1
; SI-NEXT:    s_mov_b32 s14, s10
; SI-NEXT:    s_mov_b32 s15, s11
; SI-NEXT:    s_mov_b32 s2, s10
; SI-NEXT:    s_mov_b32 s3, s11
; SI-NEXT:    s_waitcnt lgkmcnt(0)
; SI-NEXT:    s_mov_b32 s12, s6
; SI-NEXT:    s_mov_b32 s13, s7
; SI-NEXT:    buffer_load_ushort v0, off, s[12:15], 0 glc
; SI-NEXT:    s_waitcnt vmcnt(0)
; SI-NEXT:    buffer_load_ushort v1, off, s[0:3], 0 glc
; SI-NEXT:    s_waitcnt vmcnt(0)
; SI-NEXT:    s_mov_b32 s8, s4
; SI-NEXT:    s_mov_b32 s9, s5
; SI-NEXT:    v_cvt_f32_f16_e32 v0, v0
; SI-NEXT:    v_cvt_f32_f16_e32 v1, v1
; SI-NEXT:    v_cmp_neq_f32_e32 vcc, v0, v1
; SI-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc
; SI-NEXT:    buffer_store_dword v0, off, s[8:11], 0
; SI-NEXT:    s_endpgm
;
; VI-LABEL: fcmp_f16_neq:
; VI:       ; %bb.0: ; %entry
; VI-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
; VI-NEXT:    s_load_dwordx2 s[8:9], s[0:1], 0x34
; VI-NEXT:    s_mov_b32 s3, 0xf000
; VI-NEXT:    s_mov_b32 s2, -1
; VI-NEXT:    s_mov_b32 s14, s2
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    s_mov_b32 s12, s6
; VI-NEXT:    s_mov_b32 s13, s7
; VI-NEXT:    s_mov_b32 s15, s3
; VI-NEXT:    s_mov_b32 s10, s2
; VI-NEXT:    s_mov_b32 s11, s3
; VI-NEXT:    buffer_load_ushort v0, off, s[12:15], 0 glc
; VI-NEXT:    s_waitcnt vmcnt(0)
; VI-NEXT:    buffer_load_ushort v1, off, s[8:11], 0 glc
; VI-NEXT:    s_waitcnt vmcnt(0)
; VI-NEXT:    s_mov_b32 s0, s4
; VI-NEXT:    s_mov_b32 s1, s5
; VI-NEXT:    v_cmp_neq_f16_e32 vcc, v0, v1
; VI-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc
; VI-NEXT:    buffer_store_dword v0, off, s[0:3], 0
; VI-NEXT:    s_endpgm
;
; GFX11-LABEL: fcmp_f16_neq:
; GFX11:       ; %bb.0: ; %entry
; GFX11-NEXT:    s_clause 0x1
; GFX11-NEXT:    s_load_b128 s[4:7], s[0:1], 0x24
; GFX11-NEXT:    s_load_b64 s[0:1], s[0:1], 0x34
; GFX11-NEXT:    s_mov_b32 s10, -1
; GFX11-NEXT:    s_mov_b32 s11, 0x31016000
; GFX11-NEXT:    s_mov_b32 s14, s10
; GFX11-NEXT:    s_mov_b32 s15, s11
; GFX11-NEXT:    s_mov_b32 s2, s10
; GFX11-NEXT:    s_mov_b32 s3, s11
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    s_mov_b32 s12, s6
; GFX11-NEXT:    s_mov_b32 s13, s7
; GFX11-NEXT:    buffer_load_u16 v0, off, s[12:15], 0 glc dlc
; GFX11-NEXT:    s_waitcnt vmcnt(0)
; GFX11-NEXT:    buffer_load_u16 v1, off, s[0:3], 0 glc dlc
; GFX11-NEXT:    s_waitcnt vmcnt(0)
; GFX11-NEXT:    s_mov_b32 s8, s4
; GFX11-NEXT:    s_mov_b32 s9, s5
; GFX11-NEXT:    v_cmp_neq_f16_e32 vcc_lo, v0, v1
; GFX11-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc_lo
; GFX11-NEXT:    buffer_store_b32 v0, off, s[8:11], 0
; GFX11-NEXT:    s_nop 0
; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
; GFX11-NEXT:    s_endpgm
    ptr addrspace(1) %r,
    ptr addrspace(1) %a,
    ptr addrspace(1) %b) {
entry:
  %a.val = load volatile half, ptr addrspace(1) %a
  %b.val = load volatile half, ptr addrspace(1) %b
  %r.val = fcmp une half %a.val, %b.val
  %r.val.sext = sext i1 %r.val to i32
  store i32 %r.val.sext, ptr addrspace(1) %r
  ret void
}

define amdgpu_kernel void @fcmp_f16_nlt(
; SI-LABEL: fcmp_f16_nlt:
; SI:       ; %bb.0: ; %entry
; SI-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xd
; SI-NEXT:    s_mov_b32 s11, 0xf000
; SI-NEXT:    s_mov_b32 s10, -1
; SI-NEXT:    s_mov_b32 s14, s10
; SI-NEXT:    s_mov_b32 s15, s11
; SI-NEXT:    s_mov_b32 s2, s10
; SI-NEXT:    s_mov_b32 s3, s11
; SI-NEXT:    s_waitcnt lgkmcnt(0)
; SI-NEXT:    s_mov_b32 s12, s6
; SI-NEXT:    s_mov_b32 s13, s7
; SI-NEXT:    buffer_load_ushort v0, off, s[12:15], 0 glc
; SI-NEXT:    s_waitcnt vmcnt(0)
; SI-NEXT:    buffer_load_ushort v1, off, s[0:3], 0 glc
; SI-NEXT:    s_waitcnt vmcnt(0)
; SI-NEXT:    s_mov_b32 s8, s4
; SI-NEXT:    s_mov_b32 s9, s5
; SI-NEXT:    v_cvt_f32_f16_e32 v0, v0
; SI-NEXT:    v_cvt_f32_f16_e32 v1, v1
; SI-NEXT:    v_cmp_nlt_f32_e32 vcc, v0, v1
; SI-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc
; SI-NEXT:    buffer_store_dword v0, off, s[8:11], 0
; SI-NEXT:    s_endpgm
;
; VI-LABEL: fcmp_f16_nlt:
; VI:       ; %bb.0: ; %entry
; VI-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
; VI-NEXT:    s_load_dwordx2 s[8:9], s[0:1], 0x34
; VI-NEXT:    s_mov_b32 s3, 0xf000
; VI-NEXT:    s_mov_b32 s2, -1
; VI-NEXT:    s_mov_b32 s14, s2
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    s_mov_b32 s12, s6
; VI-NEXT:    s_mov_b32 s13, s7
; VI-NEXT:    s_mov_b32 s15, s3
; VI-NEXT:    s_mov_b32 s10, s2
; VI-NEXT:    s_mov_b32 s11, s3
; VI-NEXT:    buffer_load_ushort v0, off, s[12:15], 0 glc
; VI-NEXT:    s_waitcnt vmcnt(0)
; VI-NEXT:    buffer_load_ushort v1, off, s[8:11], 0 glc
; VI-NEXT:    s_waitcnt vmcnt(0)
; VI-NEXT:    s_mov_b32 s0, s4
; VI-NEXT:    s_mov_b32 s1, s5
; VI-NEXT:    v_cmp_nlt_f16_e32 vcc, v0, v1
; VI-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc
; VI-NEXT:    buffer_store_dword v0, off, s[0:3], 0
; VI-NEXT:    s_endpgm
;
; GFX11-LABEL: fcmp_f16_nlt:
; GFX11:       ; %bb.0: ; %entry
; GFX11-NEXT:    s_clause 0x1
; GFX11-NEXT:    s_load_b128 s[4:7], s[0:1], 0x24
; GFX11-NEXT:    s_load_b64 s[0:1], s[0:1], 0x34
; GFX11-NEXT:    s_mov_b32 s10, -1
; GFX11-NEXT:    s_mov_b32 s11, 0x31016000
; GFX11-NEXT:    s_mov_b32 s14, s10
; GFX11-NEXT:    s_mov_b32 s15, s11
; GFX11-NEXT:    s_mov_b32 s2, s10
; GFX11-NEXT:    s_mov_b32 s3, s11
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    s_mov_b32 s12, s6
; GFX11-NEXT:    s_mov_b32 s13, s7
; GFX11-NEXT:    buffer_load_u16 v0, off, s[12:15], 0 glc dlc
; GFX11-NEXT:    s_waitcnt vmcnt(0)
; GFX11-NEXT:    buffer_load_u16 v1, off, s[0:3], 0 glc dlc
; GFX11-NEXT:    s_waitcnt vmcnt(0)
; GFX11-NEXT:    s_mov_b32 s8, s4
; GFX11-NEXT:    s_mov_b32 s9, s5
; GFX11-NEXT:    v_cmp_nlt_f16_e32 vcc_lo, v0, v1
; GFX11-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc_lo
; GFX11-NEXT:    buffer_store_b32 v0, off, s[8:11], 0
; GFX11-NEXT:    s_nop 0
; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
; GFX11-NEXT:    s_endpgm
    ptr addrspace(1) %r,
    ptr addrspace(1) %a,
    ptr addrspace(1) %b) {
entry:
  %a.val = load volatile half, ptr addrspace(1) %a
  %b.val = load volatile half, ptr addrspace(1) %b
  %r.val = fcmp uge half %a.val, %b.val
  %r.val.sext = sext i1 %r.val to i32
  store i32 %r.val.sext, ptr addrspace(1) %r
  ret void
}

define amdgpu_kernel void @fcmp_v2f16_lt(
; SI-LABEL: fcmp_v2f16_lt:
; SI:       ; %bb.0: ; %entry
; SI-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xd
; SI-NEXT:    s_mov_b32 s11, 0xf000
; SI-NEXT:    s_mov_b32 s10, -1
; SI-NEXT:    s_mov_b32 s14, s10
; SI-NEXT:    s_mov_b32 s15, s11
; SI-NEXT:    s_waitcnt lgkmcnt(0)
; SI-NEXT:    s_mov_b32 s12, s6
; SI-NEXT:    s_mov_b32 s13, s7
; SI-NEXT:    buffer_load_dword v0, off, s[12:15], 0
; SI-NEXT:    s_mov_b32 s2, s10
; SI-NEXT:    s_mov_b32 s3, s11
; SI-NEXT:    buffer_load_dword v1, off, s[0:3], 0
; SI-NEXT:    s_mov_b32 s8, s4
; SI-NEXT:    s_mov_b32 s9, s5
; SI-NEXT:    s_waitcnt vmcnt(1)
; SI-NEXT:    v_cvt_f32_f16_e32 v2, v0
; SI-NEXT:    v_lshrrev_b32_e32 v0, 16, v0
; SI-NEXT:    s_waitcnt vmcnt(0)
; SI-NEXT:    v_cvt_f32_f16_e32 v3, v1
; SI-NEXT:    v_lshrrev_b32_e32 v1, 16, v1
; SI-NEXT:    v_cvt_f32_f16_e32 v4, v0
; SI-NEXT:    v_cvt_f32_f16_e32 v1, v1
; SI-NEXT:    v_cmp_lt_f32_e32 vcc, v2, v3
; SI-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc
; SI-NEXT:    v_cmp_lt_f32_e32 vcc, v4, v1
; SI-NEXT:    v_cndmask_b32_e64 v1, 0, -1, vcc
; SI-NEXT:    buffer_store_dwordx2 v[0:1], off, s[8:11], 0
; SI-NEXT:    s_endpgm
;
; VI-LABEL: fcmp_v2f16_lt:
; VI:       ; %bb.0: ; %entry
; VI-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
; VI-NEXT:    s_load_dwordx2 s[8:9], s[0:1], 0x34
; VI-NEXT:    s_mov_b32 s3, 0xf000
; VI-NEXT:    s_mov_b32 s2, -1
; VI-NEXT:    s_mov_b32 s10, s2
; VI-NEXT:    s_mov_b32 s11, s3
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    s_mov_b32 s12, s6
; VI-NEXT:    s_mov_b32 s13, s7
; VI-NEXT:    s_mov_b32 s14, s2
; VI-NEXT:    s_mov_b32 s15, s3
; VI-NEXT:    buffer_load_dword v0, off, s[8:11], 0
; VI-NEXT:    buffer_load_dword v1, off, s[12:15], 0
; VI-NEXT:    s_mov_b32 s0, s4
; VI-NEXT:    s_mov_b32 s1, s5
; VI-NEXT:    s_waitcnt vmcnt(1)
; VI-NEXT:    v_lshrrev_b32_e32 v2, 16, v0
; VI-NEXT:    s_waitcnt vmcnt(0)
; VI-NEXT:    v_lshrrev_b32_e32 v3, 16, v1
; VI-NEXT:    v_cmp_lt_f16_e32 vcc, v1, v0
; VI-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc
; VI-NEXT:    v_cmp_lt_f16_e32 vcc, v3, v2
; VI-NEXT:    v_cndmask_b32_e64 v1, 0, -1, vcc
; VI-NEXT:    buffer_store_dwordx2 v[0:1], off, s[0:3], 0
; VI-NEXT:    s_endpgm
;
; GFX11-LABEL: fcmp_v2f16_lt:
; GFX11:       ; %bb.0: ; %entry
; GFX11-NEXT:    s_clause 0x1
; GFX11-NEXT:    s_load_b128 s[4:7], s[0:1], 0x24
; GFX11-NEXT:    s_load_b64 s[0:1], s[0:1], 0x34
; GFX11-NEXT:    s_mov_b32 s10, -1
; GFX11-NEXT:    s_mov_b32 s11, 0x31016000
; GFX11-NEXT:    s_mov_b32 s2, s10
; GFX11-NEXT:    s_mov_b32 s3, s11
; GFX11-NEXT:    s_mov_b32 s14, s10
; GFX11-NEXT:    s_mov_b32 s15, s11
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    s_mov_b32 s12, s6
; GFX11-NEXT:    s_mov_b32 s13, s7
; GFX11-NEXT:    buffer_load_b32 v0, off, s[0:3], 0
; GFX11-NEXT:    buffer_load_b32 v1, off, s[12:15], 0
; GFX11-NEXT:    s_mov_b32 s8, s4
; GFX11-NEXT:    s_mov_b32 s9, s5
; GFX11-NEXT:    s_waitcnt vmcnt(1)
; GFX11-NEXT:    v_lshrrev_b32_e32 v2, 16, v0
; GFX11-NEXT:    s_waitcnt vmcnt(0)
; GFX11-NEXT:    v_lshrrev_b32_e32 v3, 16, v1
; GFX11-NEXT:    v_cmp_lt_f16_e32 vcc_lo, v1, v0
; GFX11-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc_lo
; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_3)
; GFX11-NEXT:    v_cmp_lt_f16_e32 vcc_lo, v3, v2
; GFX11-NEXT:    v_cndmask_b32_e64 v1, 0, -1, vcc_lo
; GFX11-NEXT:    buffer_store_b64 v[0:1], off, s[8:11], 0
; GFX11-NEXT:    s_nop 0
; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
; GFX11-NEXT:    s_endpgm
    ptr addrspace(1) %r,
    ptr addrspace(1) %a,
    ptr addrspace(1) %b) {
entry:
  %a.val = load <2 x half>, ptr addrspace(1) %a
  %b.val = load <2 x half>, ptr addrspace(1) %b
  %r.val = fcmp olt <2 x half> %a.val, %b.val
  %r.val.sext = sext <2 x i1> %r.val to <2 x i32>
  store <2 x i32> %r.val.sext, ptr addrspace(1) %r
  ret void
}


define amdgpu_kernel void @fcmp_v2f16_eq(
; SI-LABEL: fcmp_v2f16_eq:
; SI:       ; %bb.0: ; %entry
; SI-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xd
; SI-NEXT:    s_mov_b32 s11, 0xf000
; SI-NEXT:    s_mov_b32 s10, -1
; SI-NEXT:    s_mov_b32 s14, s10
; SI-NEXT:    s_mov_b32 s15, s11
; SI-NEXT:    s_waitcnt lgkmcnt(0)
; SI-NEXT:    s_mov_b32 s12, s6
; SI-NEXT:    s_mov_b32 s13, s7
; SI-NEXT:    buffer_load_dword v0, off, s[12:15], 0
; SI-NEXT:    s_mov_b32 s2, s10
; SI-NEXT:    s_mov_b32 s3, s11
; SI-NEXT:    buffer_load_dword v1, off, s[0:3], 0
; SI-NEXT:    s_mov_b32 s8, s4
; SI-NEXT:    s_mov_b32 s9, s5
; SI-NEXT:    s_waitcnt vmcnt(1)
; SI-NEXT:    v_cvt_f32_f16_e32 v2, v0
; SI-NEXT:    v_lshrrev_b32_e32 v0, 16, v0
; SI-NEXT:    s_waitcnt vmcnt(0)
; SI-NEXT:    v_cvt_f32_f16_e32 v3, v1
; SI-NEXT:    v_lshrrev_b32_e32 v1, 16, v1
; SI-NEXT:    v_cvt_f32_f16_e32 v4, v0
; SI-NEXT:    v_cvt_f32_f16_e32 v1, v1
; SI-NEXT:    v_cmp_eq_f32_e32 vcc, v2, v3
; SI-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc
; SI-NEXT:    v_cmp_eq_f32_e32 vcc, v4, v1
; SI-NEXT:    v_cndmask_b32_e64 v1, 0, -1, vcc
; SI-NEXT:    buffer_store_dwordx2 v[0:1], off, s[8:11], 0
; SI-NEXT:    s_endpgm
;
; VI-LABEL: fcmp_v2f16_eq:
; VI:       ; %bb.0: ; %entry
; VI-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
; VI-NEXT:    s_load_dwordx2 s[8:9], s[0:1], 0x34
; VI-NEXT:    s_mov_b32 s3, 0xf000
; VI-NEXT:    s_mov_b32 s2, -1
; VI-NEXT:    s_mov_b32 s10, s2
; VI-NEXT:    s_mov_b32 s11, s3
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    s_mov_b32 s12, s6
; VI-NEXT:    s_mov_b32 s13, s7
; VI-NEXT:    s_mov_b32 s14, s2
; VI-NEXT:    s_mov_b32 s15, s3
; VI-NEXT:    buffer_load_dword v0, off, s[8:11], 0
; VI-NEXT:    buffer_load_dword v1, off, s[12:15], 0
; VI-NEXT:    s_mov_b32 s0, s4
; VI-NEXT:    s_mov_b32 s1, s5
; VI-NEXT:    s_waitcnt vmcnt(1)
; VI-NEXT:    v_lshrrev_b32_e32 v2, 16, v0
; VI-NEXT:    s_waitcnt vmcnt(0)
; VI-NEXT:    v_lshrrev_b32_e32 v3, 16, v1
; VI-NEXT:    v_cmp_eq_f16_e32 vcc, v1, v0
; VI-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc
; VI-NEXT:    v_cmp_eq_f16_e32 vcc, v3, v2
; VI-NEXT:    v_cndmask_b32_e64 v1, 0, -1, vcc
; VI-NEXT:    buffer_store_dwordx2 v[0:1], off, s[0:3], 0
; VI-NEXT:    s_endpgm
;
; GFX11-LABEL: fcmp_v2f16_eq:
; GFX11:       ; %bb.0: ; %entry
; GFX11-NEXT:    s_clause 0x1
; GFX11-NEXT:    s_load_b128 s[4:7], s[0:1], 0x24
; GFX11-NEXT:    s_load_b64 s[0:1], s[0:1], 0x34
; GFX11-NEXT:    s_mov_b32 s10, -1
; GFX11-NEXT:    s_mov_b32 s11, 0x31016000
; GFX11-NEXT:    s_mov_b32 s2, s10
; GFX11-NEXT:    s_mov_b32 s3, s11
; GFX11-NEXT:    s_mov_b32 s14, s10
; GFX11-NEXT:    s_mov_b32 s15, s11
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    s_mov_b32 s12, s6
; GFX11-NEXT:    s_mov_b32 s13, s7
; GFX11-NEXT:    buffer_load_b32 v0, off, s[0:3], 0
; GFX11-NEXT:    buffer_load_b32 v1, off, s[12:15], 0
; GFX11-NEXT:    s_mov_b32 s8, s4
; GFX11-NEXT:    s_mov_b32 s9, s5
; GFX11-NEXT:    s_waitcnt vmcnt(1)
; GFX11-NEXT:    v_lshrrev_b32_e32 v2, 16, v0
; GFX11-NEXT:    s_waitcnt vmcnt(0)
; GFX11-NEXT:    v_lshrrev_b32_e32 v3, 16, v1
; GFX11-NEXT:    v_cmp_eq_f16_e32 vcc_lo, v1, v0
; GFX11-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc_lo
; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_3)
; GFX11-NEXT:    v_cmp_eq_f16_e32 vcc_lo, v3, v2
; GFX11-NEXT:    v_cndmask_b32_e64 v1, 0, -1, vcc_lo
; GFX11-NEXT:    buffer_store_b64 v[0:1], off, s[8:11], 0
; GFX11-NEXT:    s_nop 0
; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
; GFX11-NEXT:    s_endpgm
    ptr addrspace(1) %r,
    ptr addrspace(1) %a,
    ptr addrspace(1) %b) {
entry:
  %a.val = load <2 x half>, ptr addrspace(1) %a
  %b.val = load <2 x half>, ptr addrspace(1) %b
  %r.val = fcmp oeq <2 x half> %a.val, %b.val
  %r.val.sext = sext <2 x i1> %r.val to <2 x i32>
  store <2 x i32> %r.val.sext, ptr addrspace(1) %r
  ret void
}

define amdgpu_kernel void @fcmp_v2f16_le(
; SI-LABEL: fcmp_v2f16_le:
; SI:       ; %bb.0: ; %entry
; SI-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xd
; SI-NEXT:    s_mov_b32 s11, 0xf000
; SI-NEXT:    s_mov_b32 s10, -1
; SI-NEXT:    s_mov_b32 s14, s10
; SI-NEXT:    s_mov_b32 s15, s11
; SI-NEXT:    s_waitcnt lgkmcnt(0)
; SI-NEXT:    s_mov_b32 s12, s6
; SI-NEXT:    s_mov_b32 s13, s7
; SI-NEXT:    buffer_load_dword v0, off, s[12:15], 0
; SI-NEXT:    s_mov_b32 s2, s10
; SI-NEXT:    s_mov_b32 s3, s11
; SI-NEXT:    buffer_load_dword v1, off, s[0:3], 0
; SI-NEXT:    s_mov_b32 s8, s4
; SI-NEXT:    s_mov_b32 s9, s5
; SI-NEXT:    s_waitcnt vmcnt(1)
; SI-NEXT:    v_cvt_f32_f16_e32 v2, v0
; SI-NEXT:    v_lshrrev_b32_e32 v0, 16, v0
; SI-NEXT:    s_waitcnt vmcnt(0)
; SI-NEXT:    v_cvt_f32_f16_e32 v3, v1
; SI-NEXT:    v_lshrrev_b32_e32 v1, 16, v1
; SI-NEXT:    v_cvt_f32_f16_e32 v4, v0
; SI-NEXT:    v_cvt_f32_f16_e32 v1, v1
; SI-NEXT:    v_cmp_le_f32_e32 vcc, v2, v3
; SI-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc
; SI-NEXT:    v_cmp_le_f32_e32 vcc, v4, v1
; SI-NEXT:    v_cndmask_b32_e64 v1, 0, -1, vcc
; SI-NEXT:    buffer_store_dwordx2 v[0:1], off, s[8:11], 0
; SI-NEXT:    s_endpgm
;
; VI-LABEL: fcmp_v2f16_le:
; VI:       ; %bb.0: ; %entry
; VI-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
; VI-NEXT:    s_load_dwordx2 s[8:9], s[0:1], 0x34
; VI-NEXT:    s_mov_b32 s3, 0xf000
; VI-NEXT:    s_mov_b32 s2, -1
; VI-NEXT:    s_mov_b32 s10, s2
; VI-NEXT:    s_mov_b32 s11, s3
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    s_mov_b32 s12, s6
; VI-NEXT:    s_mov_b32 s13, s7
; VI-NEXT:    s_mov_b32 s14, s2
; VI-NEXT:    s_mov_b32 s15, s3
; VI-NEXT:    buffer_load_dword v0, off, s[8:11], 0
; VI-NEXT:    buffer_load_dword v1, off, s[12:15], 0
; VI-NEXT:    s_mov_b32 s0, s4
; VI-NEXT:    s_mov_b32 s1, s5
; VI-NEXT:    s_waitcnt vmcnt(1)
; VI-NEXT:    v_lshrrev_b32_e32 v2, 16, v0
; VI-NEXT:    s_waitcnt vmcnt(0)
; VI-NEXT:    v_lshrrev_b32_e32 v3, 16, v1
; VI-NEXT:    v_cmp_le_f16_e32 vcc, v1, v0
; VI-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc
; VI-NEXT:    v_cmp_le_f16_e32 vcc, v3, v2
; VI-NEXT:    v_cndmask_b32_e64 v1, 0, -1, vcc
; VI-NEXT:    buffer_store_dwordx2 v[0:1], off, s[0:3], 0
; VI-NEXT:    s_endpgm
;
; GFX11-LABEL: fcmp_v2f16_le:
; GFX11:       ; %bb.0: ; %entry
; GFX11-NEXT:    s_clause 0x1
; GFX11-NEXT:    s_load_b128 s[4:7], s[0:1], 0x24
; GFX11-NEXT:    s_load_b64 s[0:1], s[0:1], 0x34
; GFX11-NEXT:    s_mov_b32 s10, -1
; GFX11-NEXT:    s_mov_b32 s11, 0x31016000
; GFX11-NEXT:    s_mov_b32 s2, s10
; GFX11-NEXT:    s_mov_b32 s3, s11
; GFX11-NEXT:    s_mov_b32 s14, s10
; GFX11-NEXT:    s_mov_b32 s15, s11
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    s_mov_b32 s12, s6
; GFX11-NEXT:    s_mov_b32 s13, s7
; GFX11-NEXT:    buffer_load_b32 v0, off, s[0:3], 0
; GFX11-NEXT:    buffer_load_b32 v1, off, s[12:15], 0
; GFX11-NEXT:    s_mov_b32 s8, s4
; GFX11-NEXT:    s_mov_b32 s9, s5
; GFX11-NEXT:    s_waitcnt vmcnt(1)
; GFX11-NEXT:    v_lshrrev_b32_e32 v2, 16, v0
; GFX11-NEXT:    s_waitcnt vmcnt(0)
; GFX11-NEXT:    v_lshrrev_b32_e32 v3, 16, v1
; GFX11-NEXT:    v_cmp_le_f16_e32 vcc_lo, v1, v0
; GFX11-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc_lo
; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_3)
; GFX11-NEXT:    v_cmp_le_f16_e32 vcc_lo, v3, v2
; GFX11-NEXT:    v_cndmask_b32_e64 v1, 0, -1, vcc_lo
; GFX11-NEXT:    buffer_store_b64 v[0:1], off, s[8:11], 0
; GFX11-NEXT:    s_nop 0
; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
; GFX11-NEXT:    s_endpgm
    ptr addrspace(1) %r,
    ptr addrspace(1) %a,
    ptr addrspace(1) %b) {
entry:
  %a.val = load <2 x half>, ptr addrspace(1) %a
  %b.val = load <2 x half>, ptr addrspace(1) %b
  %r.val = fcmp ole <2 x half> %a.val, %b.val
  %r.val.sext = sext <2 x i1> %r.val to <2 x i32>
  store <2 x i32> %r.val.sext, ptr addrspace(1) %r
  ret void
}

define amdgpu_kernel void @fcmp_v2f16_gt(
; SI-LABEL: fcmp_v2f16_gt:
; SI:       ; %bb.0: ; %entry
; SI-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xd
; SI-NEXT:    s_mov_b32 s11, 0xf000
; SI-NEXT:    s_mov_b32 s10, -1
; SI-NEXT:    s_mov_b32 s14, s10
; SI-NEXT:    s_mov_b32 s15, s11
; SI-NEXT:    s_waitcnt lgkmcnt(0)
; SI-NEXT:    s_mov_b32 s12, s6
; SI-NEXT:    s_mov_b32 s13, s7
; SI-NEXT:    buffer_load_dword v0, off, s[12:15], 0
; SI-NEXT:    s_mov_b32 s2, s10
; SI-NEXT:    s_mov_b32 s3, s11
; SI-NEXT:    buffer_load_dword v1, off, s[0:3], 0
; SI-NEXT:    s_mov_b32 s8, s4
; SI-NEXT:    s_mov_b32 s9, s5
; SI-NEXT:    s_waitcnt vmcnt(1)
; SI-NEXT:    v_cvt_f32_f16_e32 v2, v0
; SI-NEXT:    v_lshrrev_b32_e32 v0, 16, v0
; SI-NEXT:    s_waitcnt vmcnt(0)
; SI-NEXT:    v_cvt_f32_f16_e32 v3, v1
; SI-NEXT:    v_lshrrev_b32_e32 v1, 16, v1
; SI-NEXT:    v_cvt_f32_f16_e32 v4, v0
; SI-NEXT:    v_cvt_f32_f16_e32 v1, v1
; SI-NEXT:    v_cmp_gt_f32_e32 vcc, v2, v3
; SI-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc
; SI-NEXT:    v_cmp_gt_f32_e32 vcc, v4, v1
; SI-NEXT:    v_cndmask_b32_e64 v1, 0, -1, vcc
; SI-NEXT:    buffer_store_dwordx2 v[0:1], off, s[8:11], 0
; SI-NEXT:    s_endpgm
;
; VI-LABEL: fcmp_v2f16_gt:
; VI:       ; %bb.0: ; %entry
; VI-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
; VI-NEXT:    s_load_dwordx2 s[8:9], s[0:1], 0x34
; VI-NEXT:    s_mov_b32 s3, 0xf000
; VI-NEXT:    s_mov_b32 s2, -1
; VI-NEXT:    s_mov_b32 s10, s2
; VI-NEXT:    s_mov_b32 s11, s3
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    s_mov_b32 s12, s6
; VI-NEXT:    s_mov_b32 s13, s7
; VI-NEXT:    s_mov_b32 s14, s2
; VI-NEXT:    s_mov_b32 s15, s3
; VI-NEXT:    buffer_load_dword v0, off, s[8:11], 0
; VI-NEXT:    buffer_load_dword v1, off, s[12:15], 0
; VI-NEXT:    s_mov_b32 s0, s4
; VI-NEXT:    s_mov_b32 s1, s5
; VI-NEXT:    s_waitcnt vmcnt(1)
; VI-NEXT:    v_lshrrev_b32_e32 v2, 16, v0
; VI-NEXT:    s_waitcnt vmcnt(0)
; VI-NEXT:    v_lshrrev_b32_e32 v3, 16, v1
; VI-NEXT:    v_cmp_gt_f16_e32 vcc, v1, v0
; VI-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc
; VI-NEXT:    v_cmp_gt_f16_e32 vcc, v3, v2
; VI-NEXT:    v_cndmask_b32_e64 v1, 0, -1, vcc
; VI-NEXT:    buffer_store_dwordx2 v[0:1], off, s[0:3], 0
; VI-NEXT:    s_endpgm
;
; GFX11-LABEL: fcmp_v2f16_gt:
; GFX11:       ; %bb.0: ; %entry
; GFX11-NEXT:    s_clause 0x1
; GFX11-NEXT:    s_load_b128 s[4:7], s[0:1], 0x24
; GFX11-NEXT:    s_load_b64 s[0:1], s[0:1], 0x34
; GFX11-NEXT:    s_mov_b32 s10, -1
; GFX11-NEXT:    s_mov_b32 s11, 0x31016000
; GFX11-NEXT:    s_mov_b32 s2, s10
; GFX11-NEXT:    s_mov_b32 s3, s11
; GFX11-NEXT:    s_mov_b32 s14, s10
; GFX11-NEXT:    s_mov_b32 s15, s11
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    s_mov_b32 s12, s6
; GFX11-NEXT:    s_mov_b32 s13, s7
; GFX11-NEXT:    buffer_load_b32 v0, off, s[0:3], 0
; GFX11-NEXT:    buffer_load_b32 v1, off, s[12:15], 0
; GFX11-NEXT:    s_mov_b32 s8, s4
; GFX11-NEXT:    s_mov_b32 s9, s5
; GFX11-NEXT:    s_waitcnt vmcnt(1)
; GFX11-NEXT:    v_lshrrev_b32_e32 v2, 16, v0
; GFX11-NEXT:    s_waitcnt vmcnt(0)
; GFX11-NEXT:    v_lshrrev_b32_e32 v3, 16, v1
; GFX11-NEXT:    v_cmp_gt_f16_e32 vcc_lo, v1, v0
; GFX11-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc_lo
; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_3)
; GFX11-NEXT:    v_cmp_gt_f16_e32 vcc_lo, v3, v2
; GFX11-NEXT:    v_cndmask_b32_e64 v1, 0, -1, vcc_lo
; GFX11-NEXT:    buffer_store_b64 v[0:1], off, s[8:11], 0
; GFX11-NEXT:    s_nop 0
; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
; GFX11-NEXT:    s_endpgm
    ptr addrspace(1) %r,
    ptr addrspace(1) %a,
    ptr addrspace(1) %b) {
entry:
  %a.val = load <2 x half>, ptr addrspace(1) %a
  %b.val = load <2 x half>, ptr addrspace(1) %b
  %r.val = fcmp ogt <2 x half> %a.val, %b.val
  %r.val.sext = sext <2 x i1> %r.val to <2 x i32>
  store <2 x i32> %r.val.sext, ptr addrspace(1) %r
  ret void
}


define amdgpu_kernel void @fcmp_v2f16_lg(
; SI-LABEL: fcmp_v2f16_lg:
; SI:       ; %bb.0: ; %entry
; SI-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xd
; SI-NEXT:    s_mov_b32 s11, 0xf000
; SI-NEXT:    s_mov_b32 s10, -1
; SI-NEXT:    s_mov_b32 s14, s10
; SI-NEXT:    s_mov_b32 s15, s11
; SI-NEXT:    s_waitcnt lgkmcnt(0)
; SI-NEXT:    s_mov_b32 s12, s6
; SI-NEXT:    s_mov_b32 s13, s7
; SI-NEXT:    buffer_load_dword v0, off, s[12:15], 0
; SI-NEXT:    s_mov_b32 s2, s10
; SI-NEXT:    s_mov_b32 s3, s11
; SI-NEXT:    buffer_load_dword v1, off, s[0:3], 0
; SI-NEXT:    s_mov_b32 s8, s4
; SI-NEXT:    s_mov_b32 s9, s5
; SI-NEXT:    s_waitcnt vmcnt(1)
; SI-NEXT:    v_cvt_f32_f16_e32 v2, v0
; SI-NEXT:    v_lshrrev_b32_e32 v0, 16, v0
; SI-NEXT:    s_waitcnt vmcnt(0)
; SI-NEXT:    v_cvt_f32_f16_e32 v3, v1
; SI-NEXT:    v_lshrrev_b32_e32 v1, 16, v1
; SI-NEXT:    v_cvt_f32_f16_e32 v4, v0
; SI-NEXT:    v_cvt_f32_f16_e32 v1, v1
; SI-NEXT:    v_cmp_lg_f32_e32 vcc, v2, v3
; SI-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc
; SI-NEXT:    v_cmp_lg_f32_e32 vcc, v4, v1
; SI-NEXT:    v_cndmask_b32_e64 v1, 0, -1, vcc
; SI-NEXT:    buffer_store_dwordx2 v[0:1], off, s[8:11], 0
; SI-NEXT:    s_endpgm
;
; VI-LABEL: fcmp_v2f16_lg:
; VI:       ; %bb.0: ; %entry
; VI-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
; VI-NEXT:    s_load_dwordx2 s[8:9], s[0:1], 0x34
; VI-NEXT:    s_mov_b32 s3, 0xf000
; VI-NEXT:    s_mov_b32 s2, -1
; VI-NEXT:    s_mov_b32 s10, s2
; VI-NEXT:    s_mov_b32 s11, s3
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    s_mov_b32 s12, s6
; VI-NEXT:    s_mov_b32 s13, s7
; VI-NEXT:    s_mov_b32 s14, s2
; VI-NEXT:    s_mov_b32 s15, s3
; VI-NEXT:    buffer_load_dword v0, off, s[8:11], 0
; VI-NEXT:    buffer_load_dword v1, off, s[12:15], 0
; VI-NEXT:    s_mov_b32 s0, s4
; VI-NEXT:    s_mov_b32 s1, s5
; VI-NEXT:    s_waitcnt vmcnt(1)
; VI-NEXT:    v_lshrrev_b32_e32 v2, 16, v0
; VI-NEXT:    s_waitcnt vmcnt(0)
; VI-NEXT:    v_lshrrev_b32_e32 v3, 16, v1
; VI-NEXT:    v_cmp_lg_f16_e32 vcc, v1, v0
; VI-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc
; VI-NEXT:    v_cmp_lg_f16_e32 vcc, v3, v2
; VI-NEXT:    v_cndmask_b32_e64 v1, 0, -1, vcc
; VI-NEXT:    buffer_store_dwordx2 v[0:1], off, s[0:3], 0
; VI-NEXT:    s_endpgm
;
; GFX11-LABEL: fcmp_v2f16_lg:
; GFX11:       ; %bb.0: ; %entry
; GFX11-NEXT:    s_clause 0x1
; GFX11-NEXT:    s_load_b128 s[4:7], s[0:1], 0x24
; GFX11-NEXT:    s_load_b64 s[0:1], s[0:1], 0x34
; GFX11-NEXT:    s_mov_b32 s10, -1
; GFX11-NEXT:    s_mov_b32 s11, 0x31016000
; GFX11-NEXT:    s_mov_b32 s2, s10
; GFX11-NEXT:    s_mov_b32 s3, s11
; GFX11-NEXT:    s_mov_b32 s14, s10
; GFX11-NEXT:    s_mov_b32 s15, s11
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    s_mov_b32 s12, s6
; GFX11-NEXT:    s_mov_b32 s13, s7
; GFX11-NEXT:    buffer_load_b32 v0, off, s[0:3], 0
; GFX11-NEXT:    buffer_load_b32 v1, off, s[12:15], 0
; GFX11-NEXT:    s_mov_b32 s8, s4
; GFX11-NEXT:    s_mov_b32 s9, s5
; GFX11-NEXT:    s_waitcnt vmcnt(1)
; GFX11-NEXT:    v_lshrrev_b32_e32 v2, 16, v0
; GFX11-NEXT:    s_waitcnt vmcnt(0)
; GFX11-NEXT:    v_lshrrev_b32_e32 v3, 16, v1
; GFX11-NEXT:    v_cmp_lg_f16_e32 vcc_lo, v1, v0
; GFX11-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc_lo
; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_3)
; GFX11-NEXT:    v_cmp_lg_f16_e32 vcc_lo, v3, v2
; GFX11-NEXT:    v_cndmask_b32_e64 v1, 0, -1, vcc_lo
; GFX11-NEXT:    buffer_store_b64 v[0:1], off, s[8:11], 0
; GFX11-NEXT:    s_nop 0
; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
; GFX11-NEXT:    s_endpgm
    ptr addrspace(1) %r,
    ptr addrspace(1) %a,
    ptr addrspace(1) %b) {
entry:
  %a.val = load <2 x half>, ptr addrspace(1) %a
  %b.val = load <2 x half>, ptr addrspace(1) %b
  %r.val = fcmp one <2 x half> %a.val, %b.val
  %r.val.sext = sext <2 x i1> %r.val to <2 x i32>
  store <2 x i32> %r.val.sext, ptr addrspace(1) %r
  ret void
}


define amdgpu_kernel void @fcmp_v2f16_ge(
; SI-LABEL: fcmp_v2f16_ge:
; SI:       ; %bb.0: ; %entry
; SI-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xd
; SI-NEXT:    s_mov_b32 s11, 0xf000
; SI-NEXT:    s_mov_b32 s10, -1
; SI-NEXT:    s_mov_b32 s14, s10
; SI-NEXT:    s_mov_b32 s15, s11
; SI-NEXT:    s_waitcnt lgkmcnt(0)
; SI-NEXT:    s_mov_b32 s12, s6
; SI-NEXT:    s_mov_b32 s13, s7
; SI-NEXT:    buffer_load_dword v0, off, s[12:15], 0
; SI-NEXT:    s_mov_b32 s2, s10
; SI-NEXT:    s_mov_b32 s3, s11
; SI-NEXT:    buffer_load_dword v1, off, s[0:3], 0
; SI-NEXT:    s_mov_b32 s8, s4
; SI-NEXT:    s_mov_b32 s9, s5
; SI-NEXT:    s_waitcnt vmcnt(1)
; SI-NEXT:    v_cvt_f32_f16_e32 v2, v0
; SI-NEXT:    v_lshrrev_b32_e32 v0, 16, v0
; SI-NEXT:    s_waitcnt vmcnt(0)
; SI-NEXT:    v_cvt_f32_f16_e32 v3, v1
; SI-NEXT:    v_lshrrev_b32_e32 v1, 16, v1
; SI-NEXT:    v_cvt_f32_f16_e32 v4, v0
; SI-NEXT:    v_cvt_f32_f16_e32 v1, v1
; SI-NEXT:    v_cmp_ge_f32_e32 vcc, v2, v3
; SI-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc
; SI-NEXT:    v_cmp_ge_f32_e32 vcc, v4, v1
; SI-NEXT:    v_cndmask_b32_e64 v1, 0, -1, vcc
; SI-NEXT:    buffer_store_dwordx2 v[0:1], off, s[8:11], 0
; SI-NEXT:    s_endpgm
;
; VI-LABEL: fcmp_v2f16_ge:
; VI:       ; %bb.0: ; %entry
; VI-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
; VI-NEXT:    s_load_dwordx2 s[8:9], s[0:1], 0x34
; VI-NEXT:    s_mov_b32 s3, 0xf000
; VI-NEXT:    s_mov_b32 s2, -1
; VI-NEXT:    s_mov_b32 s10, s2
; VI-NEXT:    s_mov_b32 s11, s3
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    s_mov_b32 s12, s6
; VI-NEXT:    s_mov_b32 s13, s7
; VI-NEXT:    s_mov_b32 s14, s2
; VI-NEXT:    s_mov_b32 s15, s3
; VI-NEXT:    buffer_load_dword v0, off, s[8:11], 0
; VI-NEXT:    buffer_load_dword v1, off, s[12:15], 0
; VI-NEXT:    s_mov_b32 s0, s4
; VI-NEXT:    s_mov_b32 s1, s5
; VI-NEXT:    s_waitcnt vmcnt(1)
; VI-NEXT:    v_lshrrev_b32_e32 v2, 16, v0
; VI-NEXT:    s_waitcnt vmcnt(0)
; VI-NEXT:    v_lshrrev_b32_e32 v3, 16, v1
; VI-NEXT:    v_cmp_ge_f16_e32 vcc, v1, v0
; VI-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc
; VI-NEXT:    v_cmp_ge_f16_e32 vcc, v3, v2
; VI-NEXT:    v_cndmask_b32_e64 v1, 0, -1, vcc
; VI-NEXT:    buffer_store_dwordx2 v[0:1], off, s[0:3], 0
; VI-NEXT:    s_endpgm
;
; GFX11-LABEL: fcmp_v2f16_ge:
; GFX11:       ; %bb.0: ; %entry
; GFX11-NEXT:    s_clause 0x1
; GFX11-NEXT:    s_load_b128 s[4:7], s[0:1], 0x24
; GFX11-NEXT:    s_load_b64 s[0:1], s[0:1], 0x34
; GFX11-NEXT:    s_mov_b32 s10, -1
; GFX11-NEXT:    s_mov_b32 s11, 0x31016000
; GFX11-NEXT:    s_mov_b32 s2, s10
; GFX11-NEXT:    s_mov_b32 s3, s11
; GFX11-NEXT:    s_mov_b32 s14, s10
; GFX11-NEXT:    s_mov_b32 s15, s11
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    s_mov_b32 s12, s6
; GFX11-NEXT:    s_mov_b32 s13, s7
; GFX11-NEXT:    buffer_load_b32 v0, off, s[0:3], 0
; GFX11-NEXT:    buffer_load_b32 v1, off, s[12:15], 0
; GFX11-NEXT:    s_mov_b32 s8, s4
; GFX11-NEXT:    s_mov_b32 s9, s5
; GFX11-NEXT:    s_waitcnt vmcnt(1)
; GFX11-NEXT:    v_lshrrev_b32_e32 v2, 16, v0
; GFX11-NEXT:    s_waitcnt vmcnt(0)
; GFX11-NEXT:    v_lshrrev_b32_e32 v3, 16, v1
; GFX11-NEXT:    v_cmp_ge_f16_e32 vcc_lo, v1, v0
; GFX11-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc_lo
; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_3)
; GFX11-NEXT:    v_cmp_ge_f16_e32 vcc_lo, v3, v2
; GFX11-NEXT:    v_cndmask_b32_e64 v1, 0, -1, vcc_lo
; GFX11-NEXT:    buffer_store_b64 v[0:1], off, s[8:11], 0
; GFX11-NEXT:    s_nop 0
; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
; GFX11-NEXT:    s_endpgm
    ptr addrspace(1) %r,
    ptr addrspace(1) %a,
    ptr addrspace(1) %b) {
entry:
  %a.val = load <2 x half>, ptr addrspace(1) %a
  %b.val = load <2 x half>, ptr addrspace(1) %b
  %r.val = fcmp oge <2 x half> %a.val, %b.val
  %r.val.sext = sext <2 x i1> %r.val to <2 x i32>
  store <2 x i32> %r.val.sext, ptr addrspace(1) %r
  ret void
}


define amdgpu_kernel void @fcmp_v2f16_o(
; SI-LABEL: fcmp_v2f16_o:
; SI:       ; %bb.0: ; %entry
; SI-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xd
; SI-NEXT:    s_mov_b32 s11, 0xf000
; SI-NEXT:    s_mov_b32 s10, -1
; SI-NEXT:    s_mov_b32 s14, s10
; SI-NEXT:    s_mov_b32 s15, s11
; SI-NEXT:    s_waitcnt lgkmcnt(0)
; SI-NEXT:    s_mov_b32 s12, s6
; SI-NEXT:    s_mov_b32 s13, s7
; SI-NEXT:    buffer_load_dword v0, off, s[12:15], 0
; SI-NEXT:    s_mov_b32 s2, s10
; SI-NEXT:    s_mov_b32 s3, s11
; SI-NEXT:    buffer_load_dword v1, off, s[0:3], 0
; SI-NEXT:    s_mov_b32 s8, s4
; SI-NEXT:    s_mov_b32 s9, s5
; SI-NEXT:    s_waitcnt vmcnt(1)
; SI-NEXT:    v_cvt_f32_f16_e32 v2, v0
; SI-NEXT:    v_lshrrev_b32_e32 v0, 16, v0
; SI-NEXT:    s_waitcnt vmcnt(0)
; SI-NEXT:    v_cvt_f32_f16_e32 v3, v1
; SI-NEXT:    v_lshrrev_b32_e32 v1, 16, v1
; SI-NEXT:    v_cvt_f32_f16_e32 v4, v0
; SI-NEXT:    v_cvt_f32_f16_e32 v1, v1
; SI-NEXT:    v_cmp_o_f32_e32 vcc, v2, v3
; SI-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc
; SI-NEXT:    v_cmp_o_f32_e32 vcc, v4, v1
; SI-NEXT:    v_cndmask_b32_e64 v1, 0, -1, vcc
; SI-NEXT:    buffer_store_dwordx2 v[0:1], off, s[8:11], 0
; SI-NEXT:    s_endpgm
;
; VI-LABEL: fcmp_v2f16_o:
; VI:       ; %bb.0: ; %entry
; VI-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
; VI-NEXT:    s_load_dwordx2 s[8:9], s[0:1], 0x34
; VI-NEXT:    s_mov_b32 s3, 0xf000
; VI-NEXT:    s_mov_b32 s2, -1
; VI-NEXT:    s_mov_b32 s10, s2
; VI-NEXT:    s_mov_b32 s11, s3
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    s_mov_b32 s12, s6
; VI-NEXT:    s_mov_b32 s13, s7
; VI-NEXT:    s_mov_b32 s14, s2
; VI-NEXT:    s_mov_b32 s15, s3
; VI-NEXT:    buffer_load_dword v0, off, s[8:11], 0
; VI-NEXT:    buffer_load_dword v1, off, s[12:15], 0
; VI-NEXT:    s_mov_b32 s0, s4
; VI-NEXT:    s_mov_b32 s1, s5
; VI-NEXT:    s_waitcnt vmcnt(1)
; VI-NEXT:    v_lshrrev_b32_e32 v2, 16, v0
; VI-NEXT:    s_waitcnt vmcnt(0)
; VI-NEXT:    v_lshrrev_b32_e32 v3, 16, v1
; VI-NEXT:    v_cmp_o_f16_e32 vcc, v1, v0
; VI-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc
; VI-NEXT:    v_cmp_o_f16_e32 vcc, v3, v2
; VI-NEXT:    v_cndmask_b32_e64 v1, 0, -1, vcc
; VI-NEXT:    buffer_store_dwordx2 v[0:1], off, s[0:3], 0
; VI-NEXT:    s_endpgm
;
; GFX11-LABEL: fcmp_v2f16_o:
; GFX11:       ; %bb.0: ; %entry
; GFX11-NEXT:    s_clause 0x1
; GFX11-NEXT:    s_load_b128 s[4:7], s[0:1], 0x24
; GFX11-NEXT:    s_load_b64 s[0:1], s[0:1], 0x34
; GFX11-NEXT:    s_mov_b32 s10, -1
; GFX11-NEXT:    s_mov_b32 s11, 0x31016000
; GFX11-NEXT:    s_mov_b32 s2, s10
; GFX11-NEXT:    s_mov_b32 s3, s11
; GFX11-NEXT:    s_mov_b32 s14, s10
; GFX11-NEXT:    s_mov_b32 s15, s11
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    s_mov_b32 s12, s6
; GFX11-NEXT:    s_mov_b32 s13, s7
; GFX11-NEXT:    buffer_load_b32 v0, off, s[0:3], 0
; GFX11-NEXT:    buffer_load_b32 v1, off, s[12:15], 0
; GFX11-NEXT:    s_mov_b32 s8, s4
; GFX11-NEXT:    s_mov_b32 s9, s5
; GFX11-NEXT:    s_waitcnt vmcnt(1)
; GFX11-NEXT:    v_lshrrev_b32_e32 v2, 16, v0
; GFX11-NEXT:    s_waitcnt vmcnt(0)
; GFX11-NEXT:    v_lshrrev_b32_e32 v3, 16, v1
; GFX11-NEXT:    v_cmp_o_f16_e32 vcc_lo, v1, v0
; GFX11-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc_lo
; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_3)
; GFX11-NEXT:    v_cmp_o_f16_e32 vcc_lo, v3, v2
; GFX11-NEXT:    v_cndmask_b32_e64 v1, 0, -1, vcc_lo
; GFX11-NEXT:    buffer_store_b64 v[0:1], off, s[8:11], 0
; GFX11-NEXT:    s_nop 0
; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
; GFX11-NEXT:    s_endpgm
    ptr addrspace(1) %r,
    ptr addrspace(1) %a,
    ptr addrspace(1) %b) {
entry:
  %a.val = load <2 x half>, ptr addrspace(1) %a
  %b.val = load <2 x half>, ptr addrspace(1) %b
  %r.val = fcmp ord <2 x half> %a.val, %b.val
  %r.val.sext = sext <2 x i1> %r.val to <2 x i32>
  store <2 x i32> %r.val.sext, ptr addrspace(1) %r
  ret void
}


define amdgpu_kernel void @fcmp_v2f16_u(
; SI-LABEL: fcmp_v2f16_u:
; SI:       ; %bb.0: ; %entry
; SI-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xd
; SI-NEXT:    s_mov_b32 s11, 0xf000
; SI-NEXT:    s_mov_b32 s10, -1
; SI-NEXT:    s_mov_b32 s14, s10
; SI-NEXT:    s_mov_b32 s15, s11
; SI-NEXT:    s_waitcnt lgkmcnt(0)
; SI-NEXT:    s_mov_b32 s12, s6
; SI-NEXT:    s_mov_b32 s13, s7
; SI-NEXT:    buffer_load_dword v0, off, s[12:15], 0
; SI-NEXT:    s_mov_b32 s2, s10
; SI-NEXT:    s_mov_b32 s3, s11
; SI-NEXT:    buffer_load_dword v1, off, s[0:3], 0
; SI-NEXT:    s_mov_b32 s8, s4
; SI-NEXT:    s_mov_b32 s9, s5
; SI-NEXT:    s_waitcnt vmcnt(1)
; SI-NEXT:    v_cvt_f32_f16_e32 v2, v0
; SI-NEXT:    v_lshrrev_b32_e32 v0, 16, v0
; SI-NEXT:    s_waitcnt vmcnt(0)
; SI-NEXT:    v_cvt_f32_f16_e32 v3, v1
; SI-NEXT:    v_lshrrev_b32_e32 v1, 16, v1
; SI-NEXT:    v_cvt_f32_f16_e32 v4, v0
; SI-NEXT:    v_cvt_f32_f16_e32 v1, v1
; SI-NEXT:    v_cmp_u_f32_e32 vcc, v2, v3
; SI-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc
; SI-NEXT:    v_cmp_u_f32_e32 vcc, v4, v1
; SI-NEXT:    v_cndmask_b32_e64 v1, 0, -1, vcc
; SI-NEXT:    buffer_store_dwordx2 v[0:1], off, s[8:11], 0
; SI-NEXT:    s_endpgm
;
; VI-LABEL: fcmp_v2f16_u:
; VI:       ; %bb.0: ; %entry
; VI-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
; VI-NEXT:    s_load_dwordx2 s[8:9], s[0:1], 0x34
; VI-NEXT:    s_mov_b32 s3, 0xf000
; VI-NEXT:    s_mov_b32 s2, -1
; VI-NEXT:    s_mov_b32 s10, s2
; VI-NEXT:    s_mov_b32 s11, s3
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    s_mov_b32 s12, s6
; VI-NEXT:    s_mov_b32 s13, s7
; VI-NEXT:    s_mov_b32 s14, s2
; VI-NEXT:    s_mov_b32 s15, s3
; VI-NEXT:    buffer_load_dword v0, off, s[8:11], 0
; VI-NEXT:    buffer_load_dword v1, off, s[12:15], 0
; VI-NEXT:    s_mov_b32 s0, s4
; VI-NEXT:    s_mov_b32 s1, s5
; VI-NEXT:    s_waitcnt vmcnt(1)
; VI-NEXT:    v_lshrrev_b32_e32 v2, 16, v0
; VI-NEXT:    s_waitcnt vmcnt(0)
; VI-NEXT:    v_lshrrev_b32_e32 v3, 16, v1
; VI-NEXT:    v_cmp_u_f16_e32 vcc, v1, v0
; VI-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc
; VI-NEXT:    v_cmp_u_f16_e32 vcc, v3, v2
; VI-NEXT:    v_cndmask_b32_e64 v1, 0, -1, vcc
; VI-NEXT:    buffer_store_dwordx2 v[0:1], off, s[0:3], 0
; VI-NEXT:    s_endpgm
;
; GFX11-LABEL: fcmp_v2f16_u:
; GFX11:       ; %bb.0: ; %entry
; GFX11-NEXT:    s_clause 0x1
; GFX11-NEXT:    s_load_b128 s[4:7], s[0:1], 0x24
; GFX11-NEXT:    s_load_b64 s[0:1], s[0:1], 0x34
; GFX11-NEXT:    s_mov_b32 s10, -1
; GFX11-NEXT:    s_mov_b32 s11, 0x31016000
; GFX11-NEXT:    s_mov_b32 s2, s10
; GFX11-NEXT:    s_mov_b32 s3, s11
; GFX11-NEXT:    s_mov_b32 s14, s10
; GFX11-NEXT:    s_mov_b32 s15, s11
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    s_mov_b32 s12, s6
; GFX11-NEXT:    s_mov_b32 s13, s7
; GFX11-NEXT:    buffer_load_b32 v0, off, s[0:3], 0
; GFX11-NEXT:    buffer_load_b32 v1, off, s[12:15], 0
; GFX11-NEXT:    s_mov_b32 s8, s4
; GFX11-NEXT:    s_mov_b32 s9, s5
; GFX11-NEXT:    s_waitcnt vmcnt(1)
; GFX11-NEXT:    v_lshrrev_b32_e32 v2, 16, v0
; GFX11-NEXT:    s_waitcnt vmcnt(0)
; GFX11-NEXT:    v_lshrrev_b32_e32 v3, 16, v1
; GFX11-NEXT:    v_cmp_u_f16_e32 vcc_lo, v1, v0
; GFX11-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc_lo
; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_3)
; GFX11-NEXT:    v_cmp_u_f16_e32 vcc_lo, v3, v2
; GFX11-NEXT:    v_cndmask_b32_e64 v1, 0, -1, vcc_lo
; GFX11-NEXT:    buffer_store_b64 v[0:1], off, s[8:11], 0
; GFX11-NEXT:    s_nop 0
; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
; GFX11-NEXT:    s_endpgm
    ptr addrspace(1) %r,
    ptr addrspace(1) %a,
    ptr addrspace(1) %b) {
entry:
  %a.val = load <2 x half>, ptr addrspace(1) %a
  %b.val = load <2 x half>, ptr addrspace(1) %b
  %r.val = fcmp uno <2 x half> %a.val, %b.val
  %r.val.sext = sext <2 x i1> %r.val to <2 x i32>
  store <2 x i32> %r.val.sext, ptr addrspace(1) %r
  ret void
}

define amdgpu_kernel void @fcmp_v2f16_nge(
; SI-LABEL: fcmp_v2f16_nge:
; SI:       ; %bb.0: ; %entry
; SI-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xd
; SI-NEXT:    s_mov_b32 s11, 0xf000
; SI-NEXT:    s_mov_b32 s10, -1
; SI-NEXT:    s_mov_b32 s14, s10
; SI-NEXT:    s_mov_b32 s15, s11
; SI-NEXT:    s_waitcnt lgkmcnt(0)
; SI-NEXT:    s_mov_b32 s12, s6
; SI-NEXT:    s_mov_b32 s13, s7
; SI-NEXT:    buffer_load_dword v0, off, s[12:15], 0
; SI-NEXT:    s_mov_b32 s2, s10
; SI-NEXT:    s_mov_b32 s3, s11
; SI-NEXT:    buffer_load_dword v1, off, s[0:3], 0
; SI-NEXT:    s_mov_b32 s8, s4
; SI-NEXT:    s_mov_b32 s9, s5
; SI-NEXT:    s_waitcnt vmcnt(1)
; SI-NEXT:    v_cvt_f32_f16_e32 v2, v0
; SI-NEXT:    v_lshrrev_b32_e32 v0, 16, v0
; SI-NEXT:    s_waitcnt vmcnt(0)
; SI-NEXT:    v_cvt_f32_f16_e32 v3, v1
; SI-NEXT:    v_lshrrev_b32_e32 v1, 16, v1
; SI-NEXT:    v_cvt_f32_f16_e32 v4, v0
; SI-NEXT:    v_cvt_f32_f16_e32 v1, v1
; SI-NEXT:    v_cmp_nge_f32_e32 vcc, v2, v3
; SI-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc
; SI-NEXT:    v_cmp_nge_f32_e32 vcc, v4, v1
; SI-NEXT:    v_cndmask_b32_e64 v1, 0, -1, vcc
; SI-NEXT:    buffer_store_dwordx2 v[0:1], off, s[8:11], 0
; SI-NEXT:    s_endpgm
;
; VI-LABEL: fcmp_v2f16_nge:
; VI:       ; %bb.0: ; %entry
; VI-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
; VI-NEXT:    s_load_dwordx2 s[8:9], s[0:1], 0x34
; VI-NEXT:    s_mov_b32 s3, 0xf000
; VI-NEXT:    s_mov_b32 s2, -1
; VI-NEXT:    s_mov_b32 s10, s2
; VI-NEXT:    s_mov_b32 s11, s3
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    s_mov_b32 s12, s6
; VI-NEXT:    s_mov_b32 s13, s7
; VI-NEXT:    s_mov_b32 s14, s2
; VI-NEXT:    s_mov_b32 s15, s3
; VI-NEXT:    buffer_load_dword v0, off, s[8:11], 0
; VI-NEXT:    buffer_load_dword v1, off, s[12:15], 0
; VI-NEXT:    s_mov_b32 s0, s4
; VI-NEXT:    s_mov_b32 s1, s5
; VI-NEXT:    s_waitcnt vmcnt(1)
; VI-NEXT:    v_lshrrev_b32_e32 v2, 16, v0
; VI-NEXT:    s_waitcnt vmcnt(0)
; VI-NEXT:    v_lshrrev_b32_e32 v3, 16, v1
; VI-NEXT:    v_cmp_nge_f16_e32 vcc, v1, v0
; VI-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc
; VI-NEXT:    v_cmp_nge_f16_e32 vcc, v3, v2
; VI-NEXT:    v_cndmask_b32_e64 v1, 0, -1, vcc
; VI-NEXT:    buffer_store_dwordx2 v[0:1], off, s[0:3], 0
; VI-NEXT:    s_endpgm
;
; GFX11-LABEL: fcmp_v2f16_nge:
; GFX11:       ; %bb.0: ; %entry
; GFX11-NEXT:    s_clause 0x1
; GFX11-NEXT:    s_load_b128 s[4:7], s[0:1], 0x24
; GFX11-NEXT:    s_load_b64 s[0:1], s[0:1], 0x34
; GFX11-NEXT:    s_mov_b32 s10, -1
; GFX11-NEXT:    s_mov_b32 s11, 0x31016000
; GFX11-NEXT:    s_mov_b32 s2, s10
; GFX11-NEXT:    s_mov_b32 s3, s11
; GFX11-NEXT:    s_mov_b32 s14, s10
; GFX11-NEXT:    s_mov_b32 s15, s11
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    s_mov_b32 s12, s6
; GFX11-NEXT:    s_mov_b32 s13, s7
; GFX11-NEXT:    buffer_load_b32 v0, off, s[0:3], 0
; GFX11-NEXT:    buffer_load_b32 v1, off, s[12:15], 0
; GFX11-NEXT:    s_mov_b32 s8, s4
; GFX11-NEXT:    s_mov_b32 s9, s5
; GFX11-NEXT:    s_waitcnt vmcnt(1)
; GFX11-NEXT:    v_lshrrev_b32_e32 v2, 16, v0
; GFX11-NEXT:    s_waitcnt vmcnt(0)
; GFX11-NEXT:    v_lshrrev_b32_e32 v3, 16, v1
; GFX11-NEXT:    v_cmp_nge_f16_e32 vcc_lo, v1, v0
; GFX11-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc_lo
; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_3)
; GFX11-NEXT:    v_cmp_nge_f16_e32 vcc_lo, v3, v2
; GFX11-NEXT:    v_cndmask_b32_e64 v1, 0, -1, vcc_lo
; GFX11-NEXT:    buffer_store_b64 v[0:1], off, s[8:11], 0
; GFX11-NEXT:    s_nop 0
; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
; GFX11-NEXT:    s_endpgm
    ptr addrspace(1) %r,
    ptr addrspace(1) %a,
    ptr addrspace(1) %b) {
entry:
  %a.val = load <2 x half>, ptr addrspace(1) %a
  %b.val = load <2 x half>, ptr addrspace(1) %b
  %r.val = fcmp ult <2 x half> %a.val, %b.val
  %r.val.sext = sext <2 x i1> %r.val to <2 x i32>
  store <2 x i32> %r.val.sext, ptr addrspace(1) %r
  ret void
}

define amdgpu_kernel void @fcmp_v2f16_nlg(
; SI-LABEL: fcmp_v2f16_nlg:
; SI:       ; %bb.0: ; %entry
; SI-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xd
; SI-NEXT:    s_mov_b32 s11, 0xf000
; SI-NEXT:    s_mov_b32 s10, -1
; SI-NEXT:    s_mov_b32 s14, s10
; SI-NEXT:    s_mov_b32 s15, s11
; SI-NEXT:    s_waitcnt lgkmcnt(0)
; SI-NEXT:    s_mov_b32 s12, s6
; SI-NEXT:    s_mov_b32 s13, s7
; SI-NEXT:    buffer_load_dword v0, off, s[12:15], 0
; SI-NEXT:    s_mov_b32 s2, s10
; SI-NEXT:    s_mov_b32 s3, s11
; SI-NEXT:    buffer_load_dword v1, off, s[0:3], 0
; SI-NEXT:    s_mov_b32 s8, s4
; SI-NEXT:    s_mov_b32 s9, s5
; SI-NEXT:    s_waitcnt vmcnt(1)
; SI-NEXT:    v_cvt_f32_f16_e32 v2, v0
; SI-NEXT:    v_lshrrev_b32_e32 v0, 16, v0
; SI-NEXT:    s_waitcnt vmcnt(0)
; SI-NEXT:    v_cvt_f32_f16_e32 v3, v1
; SI-NEXT:    v_lshrrev_b32_e32 v1, 16, v1
; SI-NEXT:    v_cvt_f32_f16_e32 v4, v0
; SI-NEXT:    v_cvt_f32_f16_e32 v1, v1
; SI-NEXT:    v_cmp_nlg_f32_e32 vcc, v2, v3
; SI-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc
; SI-NEXT:    v_cmp_nlg_f32_e32 vcc, v4, v1
; SI-NEXT:    v_cndmask_b32_e64 v1, 0, -1, vcc
; SI-NEXT:    buffer_store_dwordx2 v[0:1], off, s[8:11], 0
; SI-NEXT:    s_endpgm
;
; VI-LABEL: fcmp_v2f16_nlg:
; VI:       ; %bb.0: ; %entry
; VI-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
; VI-NEXT:    s_load_dwordx2 s[8:9], s[0:1], 0x34
; VI-NEXT:    s_mov_b32 s3, 0xf000
; VI-NEXT:    s_mov_b32 s2, -1
; VI-NEXT:    s_mov_b32 s10, s2
; VI-NEXT:    s_mov_b32 s11, s3
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    s_mov_b32 s12, s6
; VI-NEXT:    s_mov_b32 s13, s7
; VI-NEXT:    s_mov_b32 s14, s2
; VI-NEXT:    s_mov_b32 s15, s3
; VI-NEXT:    buffer_load_dword v0, off, s[8:11], 0
; VI-NEXT:    buffer_load_dword v1, off, s[12:15], 0
; VI-NEXT:    s_mov_b32 s0, s4
; VI-NEXT:    s_mov_b32 s1, s5
; VI-NEXT:    s_waitcnt vmcnt(1)
; VI-NEXT:    v_lshrrev_b32_e32 v2, 16, v0
; VI-NEXT:    s_waitcnt vmcnt(0)
; VI-NEXT:    v_lshrrev_b32_e32 v3, 16, v1
; VI-NEXT:    v_cmp_nlg_f16_e32 vcc, v1, v0
; VI-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc
; VI-NEXT:    v_cmp_nlg_f16_e32 vcc, v3, v2
; VI-NEXT:    v_cndmask_b32_e64 v1, 0, -1, vcc
; VI-NEXT:    buffer_store_dwordx2 v[0:1], off, s[0:3], 0
; VI-NEXT:    s_endpgm
;
; GFX11-LABEL: fcmp_v2f16_nlg:
; GFX11:       ; %bb.0: ; %entry
; GFX11-NEXT:    s_clause 0x1
; GFX11-NEXT:    s_load_b128 s[4:7], s[0:1], 0x24
; GFX11-NEXT:    s_load_b64 s[0:1], s[0:1], 0x34
; GFX11-NEXT:    s_mov_b32 s10, -1
; GFX11-NEXT:    s_mov_b32 s11, 0x31016000
; GFX11-NEXT:    s_mov_b32 s2, s10
; GFX11-NEXT:    s_mov_b32 s3, s11
; GFX11-NEXT:    s_mov_b32 s14, s10
; GFX11-NEXT:    s_mov_b32 s15, s11
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    s_mov_b32 s12, s6
; GFX11-NEXT:    s_mov_b32 s13, s7
; GFX11-NEXT:    buffer_load_b32 v0, off, s[0:3], 0
; GFX11-NEXT:    buffer_load_b32 v1, off, s[12:15], 0
; GFX11-NEXT:    s_mov_b32 s8, s4
; GFX11-NEXT:    s_mov_b32 s9, s5
; GFX11-NEXT:    s_waitcnt vmcnt(1)
; GFX11-NEXT:    v_lshrrev_b32_e32 v2, 16, v0
; GFX11-NEXT:    s_waitcnt vmcnt(0)
; GFX11-NEXT:    v_lshrrev_b32_e32 v3, 16, v1
; GFX11-NEXT:    v_cmp_nlg_f16_e32 vcc_lo, v1, v0
; GFX11-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc_lo
; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_3)
; GFX11-NEXT:    v_cmp_nlg_f16_e32 vcc_lo, v3, v2
; GFX11-NEXT:    v_cndmask_b32_e64 v1, 0, -1, vcc_lo
; GFX11-NEXT:    buffer_store_b64 v[0:1], off, s[8:11], 0
; GFX11-NEXT:    s_nop 0
; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
; GFX11-NEXT:    s_endpgm
    ptr addrspace(1) %r,
    ptr addrspace(1) %a,
    ptr addrspace(1) %b) {
entry:
  %a.val = load <2 x half>, ptr addrspace(1) %a
  %b.val = load <2 x half>, ptr addrspace(1) %b
  %r.val = fcmp ueq <2 x half> %a.val, %b.val
  %r.val.sext = sext <2 x i1> %r.val to <2 x i32>
  store <2 x i32> %r.val.sext, ptr addrspace(1) %r
  ret void
}


define amdgpu_kernel void @fcmp_v2f16_ngt(
; SI-LABEL: fcmp_v2f16_ngt:
; SI:       ; %bb.0: ; %entry
; SI-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xd
; SI-NEXT:    s_mov_b32 s11, 0xf000
; SI-NEXT:    s_mov_b32 s10, -1
; SI-NEXT:    s_mov_b32 s14, s10
; SI-NEXT:    s_mov_b32 s15, s11
; SI-NEXT:    s_waitcnt lgkmcnt(0)
; SI-NEXT:    s_mov_b32 s12, s6
; SI-NEXT:    s_mov_b32 s13, s7
; SI-NEXT:    buffer_load_dword v0, off, s[12:15], 0
; SI-NEXT:    s_mov_b32 s2, s10
; SI-NEXT:    s_mov_b32 s3, s11
; SI-NEXT:    buffer_load_dword v1, off, s[0:3], 0
; SI-NEXT:    s_mov_b32 s8, s4
; SI-NEXT:    s_mov_b32 s9, s5
; SI-NEXT:    s_waitcnt vmcnt(1)
; SI-NEXT:    v_cvt_f32_f16_e32 v2, v0
; SI-NEXT:    v_lshrrev_b32_e32 v0, 16, v0
; SI-NEXT:    s_waitcnt vmcnt(0)
; SI-NEXT:    v_cvt_f32_f16_e32 v3, v1
; SI-NEXT:    v_lshrrev_b32_e32 v1, 16, v1
; SI-NEXT:    v_cvt_f32_f16_e32 v4, v0
; SI-NEXT:    v_cvt_f32_f16_e32 v1, v1
; SI-NEXT:    v_cmp_ngt_f32_e32 vcc, v2, v3
; SI-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc
; SI-NEXT:    v_cmp_ngt_f32_e32 vcc, v4, v1
; SI-NEXT:    v_cndmask_b32_e64 v1, 0, -1, vcc
; SI-NEXT:    buffer_store_dwordx2 v[0:1], off, s[8:11], 0
; SI-NEXT:    s_endpgm
;
; VI-LABEL: fcmp_v2f16_ngt:
; VI:       ; %bb.0: ; %entry
; VI-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
; VI-NEXT:    s_load_dwordx2 s[8:9], s[0:1], 0x34
; VI-NEXT:    s_mov_b32 s3, 0xf000
; VI-NEXT:    s_mov_b32 s2, -1
; VI-NEXT:    s_mov_b32 s10, s2
; VI-NEXT:    s_mov_b32 s11, s3
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    s_mov_b32 s12, s6
; VI-NEXT:    s_mov_b32 s13, s7
; VI-NEXT:    s_mov_b32 s14, s2
; VI-NEXT:    s_mov_b32 s15, s3
; VI-NEXT:    buffer_load_dword v0, off, s[8:11], 0
; VI-NEXT:    buffer_load_dword v1, off, s[12:15], 0
; VI-NEXT:    s_mov_b32 s0, s4
; VI-NEXT:    s_mov_b32 s1, s5
; VI-NEXT:    s_waitcnt vmcnt(1)
; VI-NEXT:    v_lshrrev_b32_e32 v2, 16, v0
; VI-NEXT:    s_waitcnt vmcnt(0)
; VI-NEXT:    v_lshrrev_b32_e32 v3, 16, v1
; VI-NEXT:    v_cmp_ngt_f16_e32 vcc, v1, v0
; VI-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc
; VI-NEXT:    v_cmp_ngt_f16_e32 vcc, v3, v2
; VI-NEXT:    v_cndmask_b32_e64 v1, 0, -1, vcc
; VI-NEXT:    buffer_store_dwordx2 v[0:1], off, s[0:3], 0
; VI-NEXT:    s_endpgm
;
; GFX11-LABEL: fcmp_v2f16_ngt:
; GFX11:       ; %bb.0: ; %entry
; GFX11-NEXT:    s_clause 0x1
; GFX11-NEXT:    s_load_b128 s[4:7], s[0:1], 0x24
; GFX11-NEXT:    s_load_b64 s[0:1], s[0:1], 0x34
; GFX11-NEXT:    s_mov_b32 s10, -1
; GFX11-NEXT:    s_mov_b32 s11, 0x31016000
; GFX11-NEXT:    s_mov_b32 s2, s10
; GFX11-NEXT:    s_mov_b32 s3, s11
; GFX11-NEXT:    s_mov_b32 s14, s10
; GFX11-NEXT:    s_mov_b32 s15, s11
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    s_mov_b32 s12, s6
; GFX11-NEXT:    s_mov_b32 s13, s7
; GFX11-NEXT:    buffer_load_b32 v0, off, s[0:3], 0
; GFX11-NEXT:    buffer_load_b32 v1, off, s[12:15], 0
; GFX11-NEXT:    s_mov_b32 s8, s4
; GFX11-NEXT:    s_mov_b32 s9, s5
; GFX11-NEXT:    s_waitcnt vmcnt(1)
; GFX11-NEXT:    v_lshrrev_b32_e32 v2, 16, v0
; GFX11-NEXT:    s_waitcnt vmcnt(0)
; GFX11-NEXT:    v_lshrrev_b32_e32 v3, 16, v1
; GFX11-NEXT:    v_cmp_ngt_f16_e32 vcc_lo, v1, v0
; GFX11-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc_lo
; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_3)
; GFX11-NEXT:    v_cmp_ngt_f16_e32 vcc_lo, v3, v2
; GFX11-NEXT:    v_cndmask_b32_e64 v1, 0, -1, vcc_lo
; GFX11-NEXT:    buffer_store_b64 v[0:1], off, s[8:11], 0
; GFX11-NEXT:    s_nop 0
; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
; GFX11-NEXT:    s_endpgm
    ptr addrspace(1) %r,
    ptr addrspace(1) %a,
    ptr addrspace(1) %b) {
entry:
  %a.val = load <2 x half>, ptr addrspace(1) %a
  %b.val = load <2 x half>, ptr addrspace(1) %b
  %r.val = fcmp ule <2 x half> %a.val, %b.val
  %r.val.sext = sext <2 x i1> %r.val to <2 x i32>
  store <2 x i32> %r.val.sext, ptr addrspace(1) %r
  ret void
}

define amdgpu_kernel void @fcmp_v2f16_nle(
; SI-LABEL: fcmp_v2f16_nle:
; SI:       ; %bb.0: ; %entry
; SI-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xd
; SI-NEXT:    s_mov_b32 s11, 0xf000
; SI-NEXT:    s_mov_b32 s10, -1
; SI-NEXT:    s_mov_b32 s14, s10
; SI-NEXT:    s_mov_b32 s15, s11
; SI-NEXT:    s_waitcnt lgkmcnt(0)
; SI-NEXT:    s_mov_b32 s12, s6
; SI-NEXT:    s_mov_b32 s13, s7
; SI-NEXT:    buffer_load_dword v0, off, s[12:15], 0
; SI-NEXT:    s_mov_b32 s2, s10
; SI-NEXT:    s_mov_b32 s3, s11
; SI-NEXT:    buffer_load_dword v1, off, s[0:3], 0
; SI-NEXT:    s_mov_b32 s8, s4
; SI-NEXT:    s_mov_b32 s9, s5
; SI-NEXT:    s_waitcnt vmcnt(1)
; SI-NEXT:    v_cvt_f32_f16_e32 v2, v0
; SI-NEXT:    v_lshrrev_b32_e32 v0, 16, v0
; SI-NEXT:    s_waitcnt vmcnt(0)
; SI-NEXT:    v_cvt_f32_f16_e32 v3, v1
; SI-NEXT:    v_lshrrev_b32_e32 v1, 16, v1
; SI-NEXT:    v_cvt_f32_f16_e32 v4, v0
; SI-NEXT:    v_cvt_f32_f16_e32 v1, v1
; SI-NEXT:    v_cmp_nle_f32_e32 vcc, v2, v3
; SI-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc
; SI-NEXT:    v_cmp_nle_f32_e32 vcc, v4, v1
; SI-NEXT:    v_cndmask_b32_e64 v1, 0, -1, vcc
; SI-NEXT:    buffer_store_dwordx2 v[0:1], off, s[8:11], 0
; SI-NEXT:    s_endpgm
;
; VI-LABEL: fcmp_v2f16_nle:
; VI:       ; %bb.0: ; %entry
; VI-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
; VI-NEXT:    s_load_dwordx2 s[8:9], s[0:1], 0x34
; VI-NEXT:    s_mov_b32 s3, 0xf000
; VI-NEXT:    s_mov_b32 s2, -1
; VI-NEXT:    s_mov_b32 s10, s2
; VI-NEXT:    s_mov_b32 s11, s3
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    s_mov_b32 s12, s6
; VI-NEXT:    s_mov_b32 s13, s7
; VI-NEXT:    s_mov_b32 s14, s2
; VI-NEXT:    s_mov_b32 s15, s3
; VI-NEXT:    buffer_load_dword v0, off, s[8:11], 0
; VI-NEXT:    buffer_load_dword v1, off, s[12:15], 0
; VI-NEXT:    s_mov_b32 s0, s4
; VI-NEXT:    s_mov_b32 s1, s5
; VI-NEXT:    s_waitcnt vmcnt(1)
; VI-NEXT:    v_lshrrev_b32_e32 v2, 16, v0
; VI-NEXT:    s_waitcnt vmcnt(0)
; VI-NEXT:    v_lshrrev_b32_e32 v3, 16, v1
; VI-NEXT:    v_cmp_nle_f16_e32 vcc, v1, v0
; VI-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc
; VI-NEXT:    v_cmp_nle_f16_e32 vcc, v3, v2
; VI-NEXT:    v_cndmask_b32_e64 v1, 0, -1, vcc
; VI-NEXT:    buffer_store_dwordx2 v[0:1], off, s[0:3], 0
; VI-NEXT:    s_endpgm
;
; GFX11-LABEL: fcmp_v2f16_nle:
; GFX11:       ; %bb.0: ; %entry
; GFX11-NEXT:    s_clause 0x1
; GFX11-NEXT:    s_load_b128 s[4:7], s[0:1], 0x24
; GFX11-NEXT:    s_load_b64 s[0:1], s[0:1], 0x34
; GFX11-NEXT:    s_mov_b32 s10, -1
; GFX11-NEXT:    s_mov_b32 s11, 0x31016000
; GFX11-NEXT:    s_mov_b32 s2, s10
; GFX11-NEXT:    s_mov_b32 s3, s11
; GFX11-NEXT:    s_mov_b32 s14, s10
; GFX11-NEXT:    s_mov_b32 s15, s11
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    s_mov_b32 s12, s6
; GFX11-NEXT:    s_mov_b32 s13, s7
; GFX11-NEXT:    buffer_load_b32 v0, off, s[0:3], 0
; GFX11-NEXT:    buffer_load_b32 v1, off, s[12:15], 0
; GFX11-NEXT:    s_mov_b32 s8, s4
; GFX11-NEXT:    s_mov_b32 s9, s5
; GFX11-NEXT:    s_waitcnt vmcnt(1)
; GFX11-NEXT:    v_lshrrev_b32_e32 v2, 16, v0
; GFX11-NEXT:    s_waitcnt vmcnt(0)
; GFX11-NEXT:    v_lshrrev_b32_e32 v3, 16, v1
; GFX11-NEXT:    v_cmp_nle_f16_e32 vcc_lo, v1, v0
; GFX11-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc_lo
; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_3)
; GFX11-NEXT:    v_cmp_nle_f16_e32 vcc_lo, v3, v2
; GFX11-NEXT:    v_cndmask_b32_e64 v1, 0, -1, vcc_lo
; GFX11-NEXT:    buffer_store_b64 v[0:1], off, s[8:11], 0
; GFX11-NEXT:    s_nop 0
; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
; GFX11-NEXT:    s_endpgm
    ptr addrspace(1) %r,
    ptr addrspace(1) %a,
    ptr addrspace(1) %b) {
entry:
  %a.val = load <2 x half>, ptr addrspace(1) %a
  %b.val = load <2 x half>, ptr addrspace(1) %b
  %r.val = fcmp ugt <2 x half> %a.val, %b.val
  %r.val.sext = sext <2 x i1> %r.val to <2 x i32>
  store <2 x i32> %r.val.sext, ptr addrspace(1) %r
  ret void
}

define amdgpu_kernel void @fcmp_v2f16_neq(
; SI-LABEL: fcmp_v2f16_neq:
; SI:       ; %bb.0: ; %entry
; SI-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xd
; SI-NEXT:    s_mov_b32 s11, 0xf000
; SI-NEXT:    s_mov_b32 s10, -1
; SI-NEXT:    s_mov_b32 s14, s10
; SI-NEXT:    s_mov_b32 s15, s11
; SI-NEXT:    s_waitcnt lgkmcnt(0)
; SI-NEXT:    s_mov_b32 s12, s6
; SI-NEXT:    s_mov_b32 s13, s7
; SI-NEXT:    buffer_load_dword v0, off, s[12:15], 0
; SI-NEXT:    s_mov_b32 s2, s10
; SI-NEXT:    s_mov_b32 s3, s11
; SI-NEXT:    buffer_load_dword v1, off, s[0:3], 0
; SI-NEXT:    s_mov_b32 s8, s4
; SI-NEXT:    s_mov_b32 s9, s5
; SI-NEXT:    s_waitcnt vmcnt(1)
; SI-NEXT:    v_cvt_f32_f16_e32 v2, v0
; SI-NEXT:    v_lshrrev_b32_e32 v0, 16, v0
; SI-NEXT:    s_waitcnt vmcnt(0)
; SI-NEXT:    v_cvt_f32_f16_e32 v3, v1
; SI-NEXT:    v_lshrrev_b32_e32 v1, 16, v1
; SI-NEXT:    v_cvt_f32_f16_e32 v4, v0
; SI-NEXT:    v_cvt_f32_f16_e32 v1, v1
; SI-NEXT:    v_cmp_neq_f32_e32 vcc, v2, v3
; SI-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc
; SI-NEXT:    v_cmp_neq_f32_e32 vcc, v4, v1
; SI-NEXT:    v_cndmask_b32_e64 v1, 0, -1, vcc
; SI-NEXT:    buffer_store_dwordx2 v[0:1], off, s[8:11], 0
; SI-NEXT:    s_endpgm
;
; VI-LABEL: fcmp_v2f16_neq:
; VI:       ; %bb.0: ; %entry
; VI-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
; VI-NEXT:    s_load_dwordx2 s[8:9], s[0:1], 0x34
; VI-NEXT:    s_mov_b32 s3, 0xf000
; VI-NEXT:    s_mov_b32 s2, -1
; VI-NEXT:    s_mov_b32 s10, s2
; VI-NEXT:    s_mov_b32 s11, s3
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    s_mov_b32 s12, s6
; VI-NEXT:    s_mov_b32 s13, s7
; VI-NEXT:    s_mov_b32 s14, s2
; VI-NEXT:    s_mov_b32 s15, s3
; VI-NEXT:    buffer_load_dword v0, off, s[8:11], 0
; VI-NEXT:    buffer_load_dword v1, off, s[12:15], 0
; VI-NEXT:    s_mov_b32 s0, s4
; VI-NEXT:    s_mov_b32 s1, s5
; VI-NEXT:    s_waitcnt vmcnt(1)
; VI-NEXT:    v_lshrrev_b32_e32 v2, 16, v0
; VI-NEXT:    s_waitcnt vmcnt(0)
; VI-NEXT:    v_lshrrev_b32_e32 v3, 16, v1
; VI-NEXT:    v_cmp_neq_f16_e32 vcc, v1, v0
; VI-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc
; VI-NEXT:    v_cmp_neq_f16_e32 vcc, v3, v2
; VI-NEXT:    v_cndmask_b32_e64 v1, 0, -1, vcc
; VI-NEXT:    buffer_store_dwordx2 v[0:1], off, s[0:3], 0
; VI-NEXT:    s_endpgm
;
; GFX11-LABEL: fcmp_v2f16_neq:
; GFX11:       ; %bb.0: ; %entry
; GFX11-NEXT:    s_clause 0x1
; GFX11-NEXT:    s_load_b128 s[4:7], s[0:1], 0x24
; GFX11-NEXT:    s_load_b64 s[0:1], s[0:1], 0x34
; GFX11-NEXT:    s_mov_b32 s10, -1
; GFX11-NEXT:    s_mov_b32 s11, 0x31016000
; GFX11-NEXT:    s_mov_b32 s2, s10
; GFX11-NEXT:    s_mov_b32 s3, s11
; GFX11-NEXT:    s_mov_b32 s14, s10
; GFX11-NEXT:    s_mov_b32 s15, s11
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    s_mov_b32 s12, s6
; GFX11-NEXT:    s_mov_b32 s13, s7
; GFX11-NEXT:    buffer_load_b32 v0, off, s[0:3], 0
; GFX11-NEXT:    buffer_load_b32 v1, off, s[12:15], 0
; GFX11-NEXT:    s_mov_b32 s8, s4
; GFX11-NEXT:    s_mov_b32 s9, s5
; GFX11-NEXT:    s_waitcnt vmcnt(1)
; GFX11-NEXT:    v_lshrrev_b32_e32 v2, 16, v0
; GFX11-NEXT:    s_waitcnt vmcnt(0)
; GFX11-NEXT:    v_lshrrev_b32_e32 v3, 16, v1
; GFX11-NEXT:    v_cmp_neq_f16_e32 vcc_lo, v1, v0
; GFX11-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc_lo
; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_3)
; GFX11-NEXT:    v_cmp_neq_f16_e32 vcc_lo, v3, v2
; GFX11-NEXT:    v_cndmask_b32_e64 v1, 0, -1, vcc_lo
; GFX11-NEXT:    buffer_store_b64 v[0:1], off, s[8:11], 0
; GFX11-NEXT:    s_nop 0
; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
; GFX11-NEXT:    s_endpgm
    ptr addrspace(1) %r,
    ptr addrspace(1) %a,
    ptr addrspace(1) %b) {
entry:
  %a.val = load <2 x half>, ptr addrspace(1) %a
  %b.val = load <2 x half>, ptr addrspace(1) %b
  %r.val = fcmp une <2 x half> %a.val, %b.val
  %r.val.sext = sext <2 x i1> %r.val to <2 x i32>
  store <2 x i32> %r.val.sext, ptr addrspace(1) %r
  ret void
}

define amdgpu_kernel void @fcmp_v2f16_nlt(
; SI-LABEL: fcmp_v2f16_nlt:
; SI:       ; %bb.0: ; %entry
; SI-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xd
; SI-NEXT:    s_mov_b32 s11, 0xf000
; SI-NEXT:    s_mov_b32 s10, -1
; SI-NEXT:    s_mov_b32 s14, s10
; SI-NEXT:    s_mov_b32 s15, s11
; SI-NEXT:    s_waitcnt lgkmcnt(0)
; SI-NEXT:    s_mov_b32 s12, s6
; SI-NEXT:    s_mov_b32 s13, s7
; SI-NEXT:    buffer_load_dword v0, off, s[12:15], 0
; SI-NEXT:    s_mov_b32 s2, s10
; SI-NEXT:    s_mov_b32 s3, s11
; SI-NEXT:    buffer_load_dword v1, off, s[0:3], 0
; SI-NEXT:    s_mov_b32 s8, s4
; SI-NEXT:    s_mov_b32 s9, s5
; SI-NEXT:    s_waitcnt vmcnt(1)
; SI-NEXT:    v_cvt_f32_f16_e32 v2, v0
; SI-NEXT:    v_lshrrev_b32_e32 v0, 16, v0
; SI-NEXT:    s_waitcnt vmcnt(0)
; SI-NEXT:    v_cvt_f32_f16_e32 v3, v1
; SI-NEXT:    v_lshrrev_b32_e32 v1, 16, v1
; SI-NEXT:    v_cvt_f32_f16_e32 v4, v0
; SI-NEXT:    v_cvt_f32_f16_e32 v1, v1
; SI-NEXT:    v_cmp_nlt_f32_e32 vcc, v2, v3
; SI-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc
; SI-NEXT:    v_cmp_nlt_f32_e32 vcc, v4, v1
; SI-NEXT:    v_cndmask_b32_e64 v1, 0, -1, vcc
; SI-NEXT:    buffer_store_dwordx2 v[0:1], off, s[8:11], 0
; SI-NEXT:    s_endpgm
;
; VI-LABEL: fcmp_v2f16_nlt:
; VI:       ; %bb.0: ; %entry
; VI-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
; VI-NEXT:    s_load_dwordx2 s[8:9], s[0:1], 0x34
; VI-NEXT:    s_mov_b32 s3, 0xf000
; VI-NEXT:    s_mov_b32 s2, -1
; VI-NEXT:    s_mov_b32 s10, s2
; VI-NEXT:    s_mov_b32 s11, s3
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    s_mov_b32 s12, s6
; VI-NEXT:    s_mov_b32 s13, s7
; VI-NEXT:    s_mov_b32 s14, s2
; VI-NEXT:    s_mov_b32 s15, s3
; VI-NEXT:    buffer_load_dword v0, off, s[8:11], 0
; VI-NEXT:    buffer_load_dword v1, off, s[12:15], 0
; VI-NEXT:    s_mov_b32 s0, s4
; VI-NEXT:    s_mov_b32 s1, s5
; VI-NEXT:    s_waitcnt vmcnt(1)
; VI-NEXT:    v_lshrrev_b32_e32 v2, 16, v0
; VI-NEXT:    s_waitcnt vmcnt(0)
; VI-NEXT:    v_lshrrev_b32_e32 v3, 16, v1
; VI-NEXT:    v_cmp_nlt_f16_e32 vcc, v1, v0
; VI-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc
; VI-NEXT:    v_cmp_nlt_f16_e32 vcc, v3, v2
; VI-NEXT:    v_cndmask_b32_e64 v1, 0, -1, vcc
; VI-NEXT:    buffer_store_dwordx2 v[0:1], off, s[0:3], 0
; VI-NEXT:    s_endpgm
;
; GFX11-LABEL: fcmp_v2f16_nlt:
; GFX11:       ; %bb.0: ; %entry
; GFX11-NEXT:    s_clause 0x1
; GFX11-NEXT:    s_load_b128 s[4:7], s[0:1], 0x24
; GFX11-NEXT:    s_load_b64 s[0:1], s[0:1], 0x34
; GFX11-NEXT:    s_mov_b32 s10, -1
; GFX11-NEXT:    s_mov_b32 s11, 0x31016000
; GFX11-NEXT:    s_mov_b32 s2, s10
; GFX11-NEXT:    s_mov_b32 s3, s11
; GFX11-NEXT:    s_mov_b32 s14, s10
; GFX11-NEXT:    s_mov_b32 s15, s11
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    s_mov_b32 s12, s6
; GFX11-NEXT:    s_mov_b32 s13, s7
; GFX11-NEXT:    buffer_load_b32 v0, off, s[0:3], 0
; GFX11-NEXT:    buffer_load_b32 v1, off, s[12:15], 0
; GFX11-NEXT:    s_mov_b32 s8, s4
; GFX11-NEXT:    s_mov_b32 s9, s5
; GFX11-NEXT:    s_waitcnt vmcnt(1)
; GFX11-NEXT:    v_lshrrev_b32_e32 v2, 16, v0
; GFX11-NEXT:    s_waitcnt vmcnt(0)
; GFX11-NEXT:    v_lshrrev_b32_e32 v3, 16, v1
; GFX11-NEXT:    v_cmp_nlt_f16_e32 vcc_lo, v1, v0
; GFX11-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc_lo
; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_3)
; GFX11-NEXT:    v_cmp_nlt_f16_e32 vcc_lo, v3, v2
; GFX11-NEXT:    v_cndmask_b32_e64 v1, 0, -1, vcc_lo
; GFX11-NEXT:    buffer_store_b64 v[0:1], off, s[8:11], 0
; GFX11-NEXT:    s_nop 0
; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
; GFX11-NEXT:    s_endpgm
    ptr addrspace(1) %r,
    ptr addrspace(1) %a,
    ptr addrspace(1) %b) {
entry:
  %a.val = load <2 x half>, ptr addrspace(1) %a
  %b.val = load <2 x half>, ptr addrspace(1) %b
  %r.val = fcmp uge <2 x half> %a.val, %b.val
  %r.val.sext = sext <2 x i1> %r.val to <2 x i32>
  store <2 x i32> %r.val.sext, ptr addrspace(1) %r
  ret void
}

declare half @llvm.fabs.f16(half) #1

attributes #0 = { nounwind }
attributes #1 = { nounwind readnone }