blob: 4d9a7fc63331c824b0b5bf315dc04b7b6d1343ec [file]
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 4
; RUN: llc -global-isel=0 -mtriple=amdgpu9.00-amd-amdhsa < %s | FileCheck -check-prefixes=GCN,SDAG %s
; RUN: llc -global-isel=1 -mtriple=amdgpu9.00-amd-amdhsa < %s | FileCheck -check-prefixes=GCN,GISEL %s
define bfloat @sitofp_i128_to_bf16(i128 %x) {
; SDAG-LABEL: sitofp_i128_to_bf16:
; SDAG: ; %bb.0: ; %itofp-entry
; SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; SDAG-NEXT: v_or_b32_e32 v5, v1, v3
; SDAG-NEXT: v_or_b32_e32 v4, v0, v2
; SDAG-NEXT: v_cmp_ne_u64_e32 vcc, 0, v[4:5]
; SDAG-NEXT: v_mov_b32_e32 v4, 0
; SDAG-NEXT: s_and_saveexec_b64 s[6:7], vcc
; SDAG-NEXT: s_cbranch_execz .LBB0_12
; SDAG-NEXT: ; %bb.1: ; %itofp-if-end
; SDAG-NEXT: v_ashrrev_i32_e32 v5, 31, v3
; SDAG-NEXT: v_xor_b32_e32 v0, v0, v5
; SDAG-NEXT: v_xor_b32_e32 v1, v1, v5
; SDAG-NEXT: v_sub_co_u32_e32 v0, vcc, v0, v5
; SDAG-NEXT: v_subb_co_u32_e32 v1, vcc, v1, v5, vcc
; SDAG-NEXT: v_xor_b32_e32 v2, v2, v5
; SDAG-NEXT: v_xor_b32_e32 v6, v3, v5
; SDAG-NEXT: v_subb_co_u32_e32 v4, vcc, v2, v5, vcc
; SDAG-NEXT: v_subb_co_u32_e32 v5, vcc, v6, v5, vcc
; SDAG-NEXT: v_ffbh_u32_e32 v2, v4
; SDAG-NEXT: v_add_u32_e32 v2, 32, v2
; SDAG-NEXT: v_ffbh_u32_e32 v6, v5
; SDAG-NEXT: v_min_u32_e32 v2, v2, v6
; SDAG-NEXT: v_ffbh_u32_e32 v6, v0
; SDAG-NEXT: v_add_u32_e32 v6, 32, v6
; SDAG-NEXT: v_ffbh_u32_e32 v7, v1
; SDAG-NEXT: v_min_u32_e32 v6, v6, v7
; SDAG-NEXT: v_cmp_ne_u64_e32 vcc, 0, v[4:5]
; SDAG-NEXT: v_add_u32_e32 v6, 64, v6
; SDAG-NEXT: v_cndmask_b32_e32 v7, v6, v2, vcc
; SDAG-NEXT: v_sub_u32_e32 v2, 0x80, v7
; SDAG-NEXT: v_cmp_gt_i32_e32 vcc, 25, v2
; SDAG-NEXT: ; implicit-def: $vgpr8
; SDAG-NEXT: s_and_saveexec_b64 s[4:5], vcc
; SDAG-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
; SDAG-NEXT: s_cbranch_execnz .LBB0_13
; SDAG-NEXT: .LBB0_2: ; %Flow3
; SDAG-NEXT: s_or_saveexec_b64 s[8:9], s[4:5]
; SDAG-NEXT: v_sub_u32_e32 v6, 0x7f, v7
; SDAG-NEXT: s_xor_b64 exec, exec, s[8:9]
; SDAG-NEXT: s_cbranch_execz .LBB0_11
; SDAG-NEXT: ; %bb.3: ; %NodeBlock
; SDAG-NEXT: v_cmp_lt_i32_e32 vcc, 25, v2
; SDAG-NEXT: s_and_saveexec_b64 s[4:5], vcc
; SDAG-NEXT: s_xor_b64 s[10:11], exec, s[4:5]
; SDAG-NEXT: s_cbranch_execz .LBB0_7
; SDAG-NEXT: ; %bb.4: ; %LeafBlock
; SDAG-NEXT: v_cmp_ne_u32_e32 vcc, 26, v2
; SDAG-NEXT: s_and_saveexec_b64 s[12:13], vcc
; SDAG-NEXT: s_cbranch_execz .LBB0_6
; SDAG-NEXT: ; %bb.5: ; %itofp-sw-default
; SDAG-NEXT: v_sub_u32_e32 v12, 0x66, v7
; SDAG-NEXT: v_sub_u32_e32 v10, 64, v12
; SDAG-NEXT: v_lshrrev_b64 v[8:9], v12, v[0:1]
; SDAG-NEXT: v_lshlrev_b64 v[10:11], v10, v[4:5]
; SDAG-NEXT: v_sub_u32_e32 v13, 38, v7
; SDAG-NEXT: v_or_b32_e32 v11, v9, v11
; SDAG-NEXT: v_or_b32_e32 v10, v8, v10
; SDAG-NEXT: v_lshrrev_b64 v[8:9], v13, v[4:5]
; SDAG-NEXT: v_cmp_gt_u32_e32 vcc, 64, v12
; SDAG-NEXT: v_add_u32_e32 v14, 26, v7
; SDAG-NEXT: v_cndmask_b32_e32 v9, v9, v11, vcc
; SDAG-NEXT: v_cmp_eq_u32_e64 s[4:5], 0, v12
; SDAG-NEXT: v_cndmask_b32_e32 v8, v8, v10, vcc
; SDAG-NEXT: v_lshrrev_b64 v[10:11], v13, v[0:1]
; SDAG-NEXT: v_lshlrev_b64 v[12:13], v14, v[4:5]
; SDAG-NEXT: v_subrev_u32_e32 v7, 38, v7
; SDAG-NEXT: v_cndmask_b32_e64 v15, v8, v0, s[4:5]
; SDAG-NEXT: v_lshlrev_b64 v[7:8], v7, v[0:1]
; SDAG-NEXT: v_cndmask_b32_e64 v9, v9, v1, s[4:5]
; SDAG-NEXT: v_or_b32_e32 v11, v13, v11
; SDAG-NEXT: v_or_b32_e32 v10, v12, v10
; SDAG-NEXT: v_cmp_gt_u32_e32 vcc, 64, v14
; SDAG-NEXT: v_lshlrev_b64 v[0:1], v14, v[0:1]
; SDAG-NEXT: v_cndmask_b32_e32 v8, v8, v11, vcc
; SDAG-NEXT: v_cmp_eq_u32_e64 s[4:5], 0, v14
; SDAG-NEXT: v_cndmask_b32_e32 v7, v7, v10, vcc
; SDAG-NEXT: v_cndmask_b32_e64 v5, v8, v5, s[4:5]
; SDAG-NEXT: v_cndmask_b32_e64 v4, v7, v4, s[4:5]
; SDAG-NEXT: v_cndmask_b32_e32 v1, 0, v1, vcc
; SDAG-NEXT: v_cndmask_b32_e32 v0, 0, v0, vcc
; SDAG-NEXT: v_or_b32_e32 v1, v1, v5
; SDAG-NEXT: v_or_b32_e32 v0, v0, v4
; SDAG-NEXT: v_cmp_ne_u64_e32 vcc, 0, v[0:1]
; SDAG-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc
; SDAG-NEXT: v_or_b32_e32 v8, v15, v0
; SDAG-NEXT: v_mov_b32_e32 v0, v8
; SDAG-NEXT: v_mov_b32_e32 v1, v9
; SDAG-NEXT: .LBB0_6: ; %Flow1
; SDAG-NEXT: s_or_b64 exec, exec, s[12:13]
; SDAG-NEXT: .LBB0_7: ; %Flow2
; SDAG-NEXT: s_andn2_saveexec_b64 s[4:5], s[10:11]
; SDAG-NEXT: ; %bb.8: ; %itofp-sw-bb
; SDAG-NEXT: v_lshlrev_b64 v[0:1], 1, v[0:1]
; SDAG-NEXT: ; %bb.9: ; %itofp-sw-epilog
; SDAG-NEXT: s_or_b64 exec, exec, s[4:5]
; SDAG-NEXT: v_lshrrev_b32_e32 v4, 2, v0
; SDAG-NEXT: v_and_or_b32 v0, v4, 1, v0
; SDAG-NEXT: v_add_co_u32_e32 v0, vcc, 1, v0
; SDAG-NEXT: v_addc_co_u32_e32 v1, vcc, 0, v1, vcc
; SDAG-NEXT: v_and_b32_e32 v4, 0x4000000, v0
; SDAG-NEXT: v_cmp_ne_u32_e32 vcc, 0, v4
; SDAG-NEXT: v_alignbit_b32 v8, v1, v0, 2
; SDAG-NEXT: s_and_saveexec_b64 s[4:5], vcc
; SDAG-NEXT: s_cbranch_execnz .LBB0_14
; SDAG-NEXT: .LBB0_10: ; %Flow
; SDAG-NEXT: s_or_b64 exec, exec, s[4:5]
; SDAG-NEXT: .LBB0_11: ; %Flow4
; SDAG-NEXT: s_or_b64 exec, exec, s[8:9]
; SDAG-NEXT: v_and_b32_e32 v0, 0x80000000, v3
; SDAG-NEXT: v_lshl_add_u32 v1, v6, 23, 1.0
; SDAG-NEXT: v_and_b32_e32 v2, 0x7fffff, v8
; SDAG-NEXT: v_or3_b32 v0, v2, v0, v1
; SDAG-NEXT: v_bfe_u32 v1, v8, 16, 1
; SDAG-NEXT: s_movk_i32 s4, 0x7fff
; SDAG-NEXT: v_add3_u32 v1, v1, v0, s4
; SDAG-NEXT: v_or_b32_e32 v2, 0x400000, v0
; SDAG-NEXT: v_cmp_u_f32_e32 vcc, v0, v0
; SDAG-NEXT: v_cndmask_b32_e32 v0, v1, v2, vcc
; SDAG-NEXT: v_lshrrev_b32_e32 v4, 16, v0
; SDAG-NEXT: .LBB0_12: ; %Flow5
; SDAG-NEXT: s_or_b64 exec, exec, s[6:7]
; SDAG-NEXT: v_mov_b32_e32 v0, v4
; SDAG-NEXT: s_setpc_b64 s[30:31]
; SDAG-NEXT: .LBB0_13: ; %itofp-if-else
; SDAG-NEXT: v_add_u32_e32 v2, 0xffffff98, v7
; SDAG-NEXT: v_lshlrev_b64 v[0:1], v2, v[0:1]
; SDAG-NEXT: v_cmp_gt_u32_e32 vcc, 64, v2
; SDAG-NEXT: v_cndmask_b32_e32 v8, 0, v0, vcc
; SDAG-NEXT: ; implicit-def: $vgpr2
; SDAG-NEXT: ; implicit-def: $vgpr0_vgpr1
; SDAG-NEXT: ; implicit-def: $vgpr4_vgpr5
; SDAG-NEXT: s_branch .LBB0_2
; SDAG-NEXT: .LBB0_14: ; %itofp-if-then20
; SDAG-NEXT: v_alignbit_b32 v8, v1, v0, 3
; SDAG-NEXT: v_mov_b32_e32 v6, v2
; SDAG-NEXT: s_branch .LBB0_10
;
; GISEL-LABEL: sitofp_i128_to_bf16:
; GISEL: ; %bb.0: ; %itofp-entry
; GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GISEL-NEXT: v_or_b32_e32 v4, v0, v2
; GISEL-NEXT: v_or_b32_e32 v5, v1, v3
; GISEL-NEXT: s_mov_b32 s4, 0
; GISEL-NEXT: v_cmp_ne_u64_e32 vcc, 0, v[4:5]
; GISEL-NEXT: v_mov_b32_e32 v4, s4
; GISEL-NEXT: s_and_saveexec_b64 s[6:7], vcc
; GISEL-NEXT: s_cbranch_execz .LBB0_12
; GISEL-NEXT: ; %bb.1: ; %itofp-if-end
; GISEL-NEXT: v_ashrrev_i32_e32 v6, 31, v3
; GISEL-NEXT: v_xor_b32_e32 v0, v6, v0
; GISEL-NEXT: v_xor_b32_e32 v1, v6, v1
; GISEL-NEXT: v_sub_co_u32_e32 v0, vcc, v0, v6
; GISEL-NEXT: v_xor_b32_e32 v2, v6, v2
; GISEL-NEXT: v_subb_co_u32_e32 v1, vcc, v1, v6, vcc
; GISEL-NEXT: v_xor_b32_e32 v3, v6, v3
; GISEL-NEXT: v_subb_co_u32_e32 v2, vcc, v2, v6, vcc
; GISEL-NEXT: v_ffbh_u32_e32 v5, v0
; GISEL-NEXT: v_subb_co_u32_e32 v3, vcc, v3, v6, vcc
; GISEL-NEXT: v_ffbh_u32_e32 v4, v1
; GISEL-NEXT: v_add_u32_e32 v5, 32, v5
; GISEL-NEXT: v_ffbh_u32_e32 v7, v2
; GISEL-NEXT: v_min_u32_e32 v4, v4, v5
; GISEL-NEXT: v_ffbh_u32_e32 v5, v3
; GISEL-NEXT: v_add_u32_e32 v7, 32, v7
; GISEL-NEXT: v_cmp_eq_u64_e32 vcc, 0, v[2:3]
; GISEL-NEXT: v_add_u32_e32 v4, 64, v4
; GISEL-NEXT: v_min_u32_e32 v5, v5, v7
; GISEL-NEXT: v_cndmask_b32_e32 v5, v5, v4, vcc
; GISEL-NEXT: v_sub_u32_e32 v7, 0x80, v5
; GISEL-NEXT: v_cmp_ge_i32_e32 vcc, 24, v7
; GISEL-NEXT: ; implicit-def: $vgpr4
; GISEL-NEXT: s_and_saveexec_b64 s[4:5], vcc
; GISEL-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
; GISEL-NEXT: s_cbranch_execnz .LBB0_13
; GISEL-NEXT: .LBB0_2: ; %Flow3
; GISEL-NEXT: s_or_saveexec_b64 s[8:9], s[4:5]
; GISEL-NEXT: v_sub_u32_e32 v8, 0x7f, v5
; GISEL-NEXT: s_xor_b64 exec, exec, s[8:9]
; GISEL-NEXT: s_cbranch_execz .LBB0_11
; GISEL-NEXT: ; %bb.3: ; %NodeBlock
; GISEL-NEXT: v_cmp_le_i32_e32 vcc, 26, v7
; GISEL-NEXT: s_and_saveexec_b64 s[4:5], vcc
; GISEL-NEXT: s_xor_b64 s[10:11], exec, s[4:5]
; GISEL-NEXT: s_cbranch_execz .LBB0_7
; GISEL-NEXT: ; %bb.4: ; %LeafBlock
; GISEL-NEXT: v_cmp_ne_u32_e32 vcc, 26, v7
; GISEL-NEXT: s_and_saveexec_b64 s[12:13], vcc
; GISEL-NEXT: s_cbranch_execz .LBB0_6
; GISEL-NEXT: ; %bb.5: ; %itofp-sw-default
; GISEL-NEXT: v_sub_u32_e32 v4, 0x66, v5
; GISEL-NEXT: v_sub_u32_e32 v11, 64, v4
; GISEL-NEXT: v_lshrrev_b64 v[9:10], v4, v[0:1]
; GISEL-NEXT: v_lshlrev_b64 v[11:12], v11, v[2:3]
; GISEL-NEXT: v_add_u32_e32 v13, 0xffffffc0, v4
; GISEL-NEXT: v_or_b32_e32 v11, v9, v11
; GISEL-NEXT: v_or_b32_e32 v12, v10, v12
; GISEL-NEXT: v_lshrrev_b64 v[9:10], v13, v[2:3]
; GISEL-NEXT: v_cmp_gt_u32_e32 vcc, 64, v4
; GISEL-NEXT: v_add_u32_e32 v14, 26, v5
; GISEL-NEXT: v_cndmask_b32_e32 v9, v9, v11, vcc
; GISEL-NEXT: v_cndmask_b32_e32 v10, v10, v12, vcc
; GISEL-NEXT: v_cmp_eq_u32_e32 vcc, 0, v4
; GISEL-NEXT: v_sub_u32_e32 v11, 64, v14
; GISEL-NEXT: v_cndmask_b32_e32 v13, v9, v0, vcc
; GISEL-NEXT: v_cndmask_b32_e32 v4, v10, v1, vcc
; GISEL-NEXT: v_lshrrev_b64 v[9:10], v14, -1
; GISEL-NEXT: v_lshlrev_b64 v[11:12], v11, -1
; GISEL-NEXT: v_add_u32_e32 v5, 0xffffffda, v5
; GISEL-NEXT: v_or_b32_e32 v15, v9, v11
; GISEL-NEXT: v_or_b32_e32 v16, v10, v12
; GISEL-NEXT: v_lshrrev_b64 v[11:12], v5, -1
; GISEL-NEXT: v_cmp_gt_u32_e32 vcc, 64, v14
; GISEL-NEXT: v_cndmask_b32_e32 v5, v11, v15, vcc
; GISEL-NEXT: v_cndmask_b32_e32 v11, v12, v16, vcc
; GISEL-NEXT: v_cmp_eq_u32_e64 s[4:5], 0, v14
; GISEL-NEXT: v_cndmask_b32_e32 v9, 0, v9, vcc
; GISEL-NEXT: v_cndmask_b32_e32 v10, 0, v10, vcc
; GISEL-NEXT: v_cndmask_b32_e64 v5, v5, -1, s[4:5]
; GISEL-NEXT: v_cndmask_b32_e64 v11, v11, -1, s[4:5]
; GISEL-NEXT: v_and_b32_e32 v2, v9, v2
; GISEL-NEXT: v_and_b32_e32 v3, v10, v3
; GISEL-NEXT: v_and_or_b32 v0, v5, v0, v2
; GISEL-NEXT: v_and_or_b32 v1, v11, v1, v3
; GISEL-NEXT: v_cmp_ne_u64_e32 vcc, 0, v[0:1]
; GISEL-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc
; GISEL-NEXT: v_or_b32_e32 v3, v13, v0
; GISEL-NEXT: v_mov_b32_e32 v0, v3
; GISEL-NEXT: v_mov_b32_e32 v1, v4
; GISEL-NEXT: v_mov_b32_e32 v2, v5
; GISEL-NEXT: v_mov_b32_e32 v3, v6
; GISEL-NEXT: .LBB0_6: ; %Flow1
; GISEL-NEXT: s_or_b64 exec, exec, s[12:13]
; GISEL-NEXT: .LBB0_7: ; %Flow2
; GISEL-NEXT: s_andn2_saveexec_b64 s[4:5], s[10:11]
; GISEL-NEXT: ; %bb.8: ; %itofp-sw-bb
; GISEL-NEXT: v_lshlrev_b64 v[0:1], 1, v[0:1]
; GISEL-NEXT: ; %bb.9: ; %itofp-sw-epilog
; GISEL-NEXT: s_or_b64 exec, exec, s[4:5]
; GISEL-NEXT: v_bfe_u32 v2, v0, 2, 1
; GISEL-NEXT: v_or_b32_e32 v0, v0, v2
; GISEL-NEXT: v_add_co_u32_e32 v0, vcc, 1, v0
; GISEL-NEXT: v_bfrev_b32_e32 v2, 32
; GISEL-NEXT: v_addc_co_u32_e32 v1, vcc, 0, v1, vcc
; GISEL-NEXT: v_mov_b32_e32 v3, 0
; GISEL-NEXT: v_and_b32_e32 v2, 0x4000000, v0
; GISEL-NEXT: v_lshrrev_b64 v[4:5], 2, v[0:1]
; GISEL-NEXT: v_cmp_ne_u64_e32 vcc, 0, v[2:3]
; GISEL-NEXT: s_and_saveexec_b64 s[4:5], vcc
; GISEL-NEXT: s_cbranch_execnz .LBB0_14
; GISEL-NEXT: .LBB0_10: ; %Flow
; GISEL-NEXT: s_or_b64 exec, exec, s[4:5]
; GISEL-NEXT: .LBB0_11: ; %Flow4
; GISEL-NEXT: s_or_b64 exec, exec, s[8:9]
; GISEL-NEXT: v_and_b32_e32 v0, 0x80000000, v6
; GISEL-NEXT: v_lshl_add_u32 v1, v8, 23, 1.0
; GISEL-NEXT: v_and_b32_e32 v2, 0x7fffff, v4
; GISEL-NEXT: v_or3_b32 v0, v2, v0, v1
; GISEL-NEXT: v_bfe_u32 v2, v0, 16, 1
; GISEL-NEXT: v_mov_b32_e32 v3, 0x7fff
; GISEL-NEXT: v_or_b32_e32 v1, 0x400000, v0
; GISEL-NEXT: v_add3_u32 v2, v2, v0, v3
; GISEL-NEXT: v_cmp_u_f32_e32 vcc, 0, v0
; GISEL-NEXT: v_cndmask_b32_e32 v0, v2, v1, vcc
; GISEL-NEXT: v_lshrrev_b32_e32 v4, 16, v0
; GISEL-NEXT: .LBB0_12: ; %Flow5
; GISEL-NEXT: s_or_b64 exec, exec, s[6:7]
; GISEL-NEXT: v_mov_b32_e32 v0, v4
; GISEL-NEXT: s_setpc_b64 s[30:31]
; GISEL-NEXT: .LBB0_13: ; %itofp-if-else
; GISEL-NEXT: v_add_u32_e32 v2, 0xffffff98, v5
; GISEL-NEXT: v_lshlrev_b64 v[0:1], v2, v[0:1]
; GISEL-NEXT: v_cmp_gt_u32_e32 vcc, 64, v2
; GISEL-NEXT: v_cndmask_b32_e32 v4, 0, v0, vcc
; GISEL-NEXT: ; implicit-def: $vgpr7
; GISEL-NEXT: ; implicit-def: $vgpr0
; GISEL-NEXT: ; implicit-def: $vgpr2
; GISEL-NEXT: s_branch .LBB0_2
; GISEL-NEXT: .LBB0_14: ; %itofp-if-then20
; GISEL-NEXT: v_lshrrev_b64 v[4:5], 3, v[0:1]
; GISEL-NEXT: v_mov_b32_e32 v8, v7
; GISEL-NEXT: s_branch .LBB0_10
%cvt = sitofp i128 %x to bfloat
ret bfloat %cvt
}
define bfloat @uitofp_i128_to_bf16(i128 %x) {
; SDAG-LABEL: uitofp_i128_to_bf16:
; SDAG: ; %bb.0: ; %itofp-entry
; SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; SDAG-NEXT: v_or_b32_e32 v5, v1, v3
; SDAG-NEXT: v_or_b32_e32 v4, v0, v2
; SDAG-NEXT: v_cmp_ne_u64_e32 vcc, 0, v[4:5]
; SDAG-NEXT: v_mov_b32_e32 v4, 0
; SDAG-NEXT: s_and_saveexec_b64 s[6:7], vcc
; SDAG-NEXT: s_cbranch_execz .LBB1_12
; SDAG-NEXT: ; %bb.1: ; %itofp-if-end
; SDAG-NEXT: v_ffbh_u32_e32 v4, v2
; SDAG-NEXT: v_add_u32_e32 v4, 32, v4
; SDAG-NEXT: v_ffbh_u32_e32 v5, v3
; SDAG-NEXT: v_min_u32_e32 v4, v4, v5
; SDAG-NEXT: v_ffbh_u32_e32 v5, v0
; SDAG-NEXT: v_add_u32_e32 v5, 32, v5
; SDAG-NEXT: v_ffbh_u32_e32 v6, v1
; SDAG-NEXT: v_min_u32_e32 v5, v5, v6
; SDAG-NEXT: v_cmp_ne_u64_e32 vcc, 0, v[2:3]
; SDAG-NEXT: v_add_u32_e32 v5, 64, v5
; SDAG-NEXT: v_cndmask_b32_e32 v6, v5, v4, vcc
; SDAG-NEXT: v_sub_u32_e32 v4, 0x80, v6
; SDAG-NEXT: v_cmp_gt_i32_e32 vcc, 25, v4
; SDAG-NEXT: ; implicit-def: $vgpr7
; SDAG-NEXT: s_and_saveexec_b64 s[4:5], vcc
; SDAG-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
; SDAG-NEXT: s_cbranch_execnz .LBB1_13
; SDAG-NEXT: .LBB1_2: ; %Flow3
; SDAG-NEXT: s_or_saveexec_b64 s[8:9], s[4:5]
; SDAG-NEXT: v_sub_u32_e32 v5, 0x7f, v6
; SDAG-NEXT: s_xor_b64 exec, exec, s[8:9]
; SDAG-NEXT: s_cbranch_execz .LBB1_11
; SDAG-NEXT: ; %bb.3: ; %NodeBlock
; SDAG-NEXT: v_cmp_lt_i32_e32 vcc, 25, v4
; SDAG-NEXT: s_and_saveexec_b64 s[4:5], vcc
; SDAG-NEXT: s_xor_b64 s[10:11], exec, s[4:5]
; SDAG-NEXT: s_cbranch_execz .LBB1_7
; SDAG-NEXT: ; %bb.4: ; %LeafBlock
; SDAG-NEXT: v_cmp_ne_u32_e32 vcc, 26, v4
; SDAG-NEXT: s_and_saveexec_b64 s[12:13], vcc
; SDAG-NEXT: s_cbranch_execz .LBB1_6
; SDAG-NEXT: ; %bb.5: ; %itofp-sw-default
; SDAG-NEXT: v_sub_u32_e32 v11, 0x66, v6
; SDAG-NEXT: v_sub_u32_e32 v9, 64, v11
; SDAG-NEXT: v_lshrrev_b64 v[7:8], v11, v[0:1]
; SDAG-NEXT: v_lshlrev_b64 v[9:10], v9, v[2:3]
; SDAG-NEXT: v_sub_u32_e32 v12, 38, v6
; SDAG-NEXT: v_or_b32_e32 v10, v8, v10
; SDAG-NEXT: v_or_b32_e32 v9, v7, v9
; SDAG-NEXT: v_lshrrev_b64 v[7:8], v12, v[2:3]
; SDAG-NEXT: v_cmp_gt_u32_e32 vcc, 64, v11
; SDAG-NEXT: v_add_u32_e32 v13, 26, v6
; SDAG-NEXT: v_cndmask_b32_e32 v8, v8, v10, vcc
; SDAG-NEXT: v_cmp_eq_u32_e64 s[4:5], 0, v11
; SDAG-NEXT: v_cndmask_b32_e32 v7, v7, v9, vcc
; SDAG-NEXT: v_lshrrev_b64 v[9:10], v12, v[0:1]
; SDAG-NEXT: v_lshlrev_b64 v[11:12], v13, v[2:3]
; SDAG-NEXT: v_subrev_u32_e32 v6, 38, v6
; SDAG-NEXT: v_cndmask_b32_e64 v14, v7, v0, s[4:5]
; SDAG-NEXT: v_lshlrev_b64 v[6:7], v6, v[0:1]
; SDAG-NEXT: v_cndmask_b32_e64 v8, v8, v1, s[4:5]
; SDAG-NEXT: v_or_b32_e32 v10, v12, v10
; SDAG-NEXT: v_or_b32_e32 v9, v11, v9
; SDAG-NEXT: v_cmp_gt_u32_e32 vcc, 64, v13
; SDAG-NEXT: v_lshlrev_b64 v[0:1], v13, v[0:1]
; SDAG-NEXT: v_cndmask_b32_e32 v7, v7, v10, vcc
; SDAG-NEXT: v_cmp_eq_u32_e64 s[4:5], 0, v13
; SDAG-NEXT: v_cndmask_b32_e32 v6, v6, v9, vcc
; SDAG-NEXT: v_cndmask_b32_e64 v3, v7, v3, s[4:5]
; SDAG-NEXT: v_cndmask_b32_e64 v2, v6, v2, s[4:5]
; SDAG-NEXT: v_cndmask_b32_e32 v1, 0, v1, vcc
; SDAG-NEXT: v_cndmask_b32_e32 v0, 0, v0, vcc
; SDAG-NEXT: v_or_b32_e32 v1, v1, v3
; SDAG-NEXT: v_or_b32_e32 v0, v0, v2
; SDAG-NEXT: v_cmp_ne_u64_e32 vcc, 0, v[0:1]
; SDAG-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc
; SDAG-NEXT: v_or_b32_e32 v7, v14, v0
; SDAG-NEXT: v_mov_b32_e32 v0, v7
; SDAG-NEXT: v_mov_b32_e32 v1, v8
; SDAG-NEXT: .LBB1_6: ; %Flow1
; SDAG-NEXT: s_or_b64 exec, exec, s[12:13]
; SDAG-NEXT: .LBB1_7: ; %Flow2
; SDAG-NEXT: s_andn2_saveexec_b64 s[4:5], s[10:11]
; SDAG-NEXT: ; %bb.8: ; %itofp-sw-bb
; SDAG-NEXT: v_lshlrev_b64 v[0:1], 1, v[0:1]
; SDAG-NEXT: ; %bb.9: ; %itofp-sw-epilog
; SDAG-NEXT: s_or_b64 exec, exec, s[4:5]
; SDAG-NEXT: v_lshrrev_b32_e32 v2, 2, v0
; SDAG-NEXT: v_and_or_b32 v0, v2, 1, v0
; SDAG-NEXT: v_add_co_u32_e32 v0, vcc, 1, v0
; SDAG-NEXT: v_addc_co_u32_e32 v1, vcc, 0, v1, vcc
; SDAG-NEXT: v_and_b32_e32 v2, 0x4000000, v0
; SDAG-NEXT: v_cmp_ne_u32_e32 vcc, 0, v2
; SDAG-NEXT: v_alignbit_b32 v7, v1, v0, 2
; SDAG-NEXT: s_and_saveexec_b64 s[4:5], vcc
; SDAG-NEXT: s_cbranch_execnz .LBB1_14
; SDAG-NEXT: .LBB1_10: ; %Flow
; SDAG-NEXT: s_or_b64 exec, exec, s[4:5]
; SDAG-NEXT: .LBB1_11: ; %Flow4
; SDAG-NEXT: s_or_b64 exec, exec, s[8:9]
; SDAG-NEXT: v_and_b32_e32 v0, 0x7fffff, v7
; SDAG-NEXT: v_lshl_or_b32 v0, v5, 23, v0
; SDAG-NEXT: v_add_u32_e32 v0, 1.0, v0
; SDAG-NEXT: v_bfe_u32 v1, v0, 16, 1
; SDAG-NEXT: s_movk_i32 s4, 0x7fff
; SDAG-NEXT: v_add3_u32 v1, v1, v0, s4
; SDAG-NEXT: v_or_b32_e32 v2, 0x400000, v0
; SDAG-NEXT: v_cmp_u_f32_e32 vcc, v0, v0
; SDAG-NEXT: v_cndmask_b32_e32 v0, v1, v2, vcc
; SDAG-NEXT: v_lshrrev_b32_e32 v4, 16, v0
; SDAG-NEXT: .LBB1_12: ; %Flow5
; SDAG-NEXT: s_or_b64 exec, exec, s[6:7]
; SDAG-NEXT: v_mov_b32_e32 v0, v4
; SDAG-NEXT: s_setpc_b64 s[30:31]
; SDAG-NEXT: .LBB1_13: ; %itofp-if-else
; SDAG-NEXT: v_add_u32_e32 v2, 0xffffff98, v6
; SDAG-NEXT: v_lshlrev_b64 v[0:1], v2, v[0:1]
; SDAG-NEXT: v_cmp_gt_u32_e32 vcc, 64, v2
; SDAG-NEXT: v_cndmask_b32_e32 v7, 0, v0, vcc
; SDAG-NEXT: ; implicit-def: $vgpr0_vgpr1
; SDAG-NEXT: ; implicit-def: $vgpr4
; SDAG-NEXT: ; implicit-def: $vgpr2
; SDAG-NEXT: s_branch .LBB1_2
; SDAG-NEXT: .LBB1_14: ; %itofp-if-then20
; SDAG-NEXT: v_alignbit_b32 v7, v1, v0, 3
; SDAG-NEXT: v_mov_b32_e32 v5, v4
; SDAG-NEXT: s_branch .LBB1_10
;
; GISEL-LABEL: uitofp_i128_to_bf16:
; GISEL: ; %bb.0: ; %itofp-entry
; GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GISEL-NEXT: v_or_b32_e32 v4, v0, v2
; GISEL-NEXT: v_or_b32_e32 v5, v1, v3
; GISEL-NEXT: s_mov_b32 s4, 0
; GISEL-NEXT: v_cmp_ne_u64_e32 vcc, 0, v[4:5]
; GISEL-NEXT: v_mov_b32_e32 v4, s4
; GISEL-NEXT: s_and_saveexec_b64 s[6:7], vcc
; GISEL-NEXT: s_cbranch_execz .LBB1_12
; GISEL-NEXT: ; %bb.1: ; %itofp-if-end
; GISEL-NEXT: v_ffbh_u32_e32 v5, v0
; GISEL-NEXT: v_ffbh_u32_e32 v4, v1
; GISEL-NEXT: v_add_u32_e32 v5, 32, v5
; GISEL-NEXT: v_ffbh_u32_e32 v6, v2
; GISEL-NEXT: v_min_u32_e32 v4, v4, v5
; GISEL-NEXT: v_ffbh_u32_e32 v5, v3
; GISEL-NEXT: v_add_u32_e32 v6, 32, v6
; GISEL-NEXT: v_cmp_eq_u64_e32 vcc, 0, v[2:3]
; GISEL-NEXT: v_add_u32_e32 v4, 64, v4
; GISEL-NEXT: v_min_u32_e32 v5, v5, v6
; GISEL-NEXT: v_cndmask_b32_e32 v5, v5, v4, vcc
; GISEL-NEXT: v_sub_u32_e32 v6, 0x80, v5
; GISEL-NEXT: v_cmp_ge_i32_e32 vcc, 24, v6
; GISEL-NEXT: ; implicit-def: $vgpr4
; GISEL-NEXT: s_and_saveexec_b64 s[4:5], vcc
; GISEL-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
; GISEL-NEXT: s_cbranch_execnz .LBB1_13
; GISEL-NEXT: .LBB1_2: ; %Flow3
; GISEL-NEXT: s_or_saveexec_b64 s[8:9], s[4:5]
; GISEL-NEXT: v_sub_u32_e32 v7, 0x7f, v5
; GISEL-NEXT: s_xor_b64 exec, exec, s[8:9]
; GISEL-NEXT: s_cbranch_execz .LBB1_11
; GISEL-NEXT: ; %bb.3: ; %NodeBlock
; GISEL-NEXT: v_cmp_le_i32_e32 vcc, 26, v6
; GISEL-NEXT: s_and_saveexec_b64 s[4:5], vcc
; GISEL-NEXT: s_xor_b64 s[10:11], exec, s[4:5]
; GISEL-NEXT: s_cbranch_execz .LBB1_7
; GISEL-NEXT: ; %bb.4: ; %LeafBlock
; GISEL-NEXT: v_cmp_ne_u32_e32 vcc, 26, v6
; GISEL-NEXT: s_and_saveexec_b64 s[12:13], vcc
; GISEL-NEXT: s_cbranch_execz .LBB1_6
; GISEL-NEXT: ; %bb.5: ; %itofp-sw-default
; GISEL-NEXT: v_sub_u32_e32 v4, 0x66, v5
; GISEL-NEXT: v_sub_u32_e32 v10, 64, v4
; GISEL-NEXT: v_lshrrev_b64 v[8:9], v4, v[0:1]
; GISEL-NEXT: v_lshlrev_b64 v[10:11], v10, v[2:3]
; GISEL-NEXT: v_add_u32_e32 v12, 0xffffffc0, v4
; GISEL-NEXT: v_or_b32_e32 v10, v8, v10
; GISEL-NEXT: v_or_b32_e32 v11, v9, v11
; GISEL-NEXT: v_lshrrev_b64 v[8:9], v12, v[2:3]
; GISEL-NEXT: v_cmp_gt_u32_e32 vcc, 64, v4
; GISEL-NEXT: v_add_u32_e32 v13, 26, v5
; GISEL-NEXT: v_cndmask_b32_e32 v8, v8, v10, vcc
; GISEL-NEXT: v_cndmask_b32_e32 v9, v9, v11, vcc
; GISEL-NEXT: v_cmp_eq_u32_e32 vcc, 0, v4
; GISEL-NEXT: v_sub_u32_e32 v10, 64, v13
; GISEL-NEXT: v_cndmask_b32_e32 v12, v8, v0, vcc
; GISEL-NEXT: v_cndmask_b32_e32 v4, v9, v1, vcc
; GISEL-NEXT: v_lshrrev_b64 v[8:9], v13, -1
; GISEL-NEXT: v_lshlrev_b64 v[10:11], v10, -1
; GISEL-NEXT: v_add_u32_e32 v5, 0xffffffda, v5
; GISEL-NEXT: v_or_b32_e32 v14, v8, v10
; GISEL-NEXT: v_or_b32_e32 v15, v9, v11
; GISEL-NEXT: v_lshrrev_b64 v[10:11], v5, -1
; GISEL-NEXT: v_cmp_gt_u32_e32 vcc, 64, v13
; GISEL-NEXT: v_cndmask_b32_e32 v5, v10, v14, vcc
; GISEL-NEXT: v_cndmask_b32_e32 v10, v11, v15, vcc
; GISEL-NEXT: v_cmp_eq_u32_e64 s[4:5], 0, v13
; GISEL-NEXT: v_cndmask_b32_e32 v8, 0, v8, vcc
; GISEL-NEXT: v_cndmask_b32_e32 v9, 0, v9, vcc
; GISEL-NEXT: v_cndmask_b32_e64 v5, v5, -1, s[4:5]
; GISEL-NEXT: v_cndmask_b32_e64 v10, v10, -1, s[4:5]
; GISEL-NEXT: v_and_b32_e32 v2, v8, v2
; GISEL-NEXT: v_and_b32_e32 v3, v9, v3
; GISEL-NEXT: v_and_or_b32 v0, v5, v0, v2
; GISEL-NEXT: v_and_or_b32 v1, v10, v1, v3
; GISEL-NEXT: v_cmp_ne_u64_e32 vcc, 0, v[0:1]
; GISEL-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc
; GISEL-NEXT: v_or_b32_e32 v3, v12, v0
; GISEL-NEXT: v_mov_b32_e32 v0, v3
; GISEL-NEXT: v_mov_b32_e32 v1, v4
; GISEL-NEXT: v_mov_b32_e32 v2, v5
; GISEL-NEXT: v_mov_b32_e32 v3, v6
; GISEL-NEXT: .LBB1_6: ; %Flow1
; GISEL-NEXT: s_or_b64 exec, exec, s[12:13]
; GISEL-NEXT: .LBB1_7: ; %Flow2
; GISEL-NEXT: s_andn2_saveexec_b64 s[4:5], s[10:11]
; GISEL-NEXT: ; %bb.8: ; %itofp-sw-bb
; GISEL-NEXT: v_lshlrev_b64 v[0:1], 1, v[0:1]
; GISEL-NEXT: ; %bb.9: ; %itofp-sw-epilog
; GISEL-NEXT: s_or_b64 exec, exec, s[4:5]
; GISEL-NEXT: v_bfe_u32 v2, v0, 2, 1
; GISEL-NEXT: v_or_b32_e32 v0, v0, v2
; GISEL-NEXT: v_add_co_u32_e32 v0, vcc, 1, v0
; GISEL-NEXT: v_bfrev_b32_e32 v2, 32
; GISEL-NEXT: v_addc_co_u32_e32 v1, vcc, 0, v1, vcc
; GISEL-NEXT: v_mov_b32_e32 v3, 0
; GISEL-NEXT: v_and_b32_e32 v2, 0x4000000, v0
; GISEL-NEXT: v_lshrrev_b64 v[4:5], 2, v[0:1]
; GISEL-NEXT: v_cmp_ne_u64_e32 vcc, 0, v[2:3]
; GISEL-NEXT: s_and_saveexec_b64 s[4:5], vcc
; GISEL-NEXT: s_cbranch_execnz .LBB1_14
; GISEL-NEXT: .LBB1_10: ; %Flow
; GISEL-NEXT: s_or_b64 exec, exec, s[4:5]
; GISEL-NEXT: .LBB1_11: ; %Flow4
; GISEL-NEXT: s_or_b64 exec, exec, s[8:9]
; GISEL-NEXT: v_lshl_add_u32 v0, v7, 23, 1.0
; GISEL-NEXT: v_mov_b32_e32 v1, 0x7fffff
; GISEL-NEXT: v_and_or_b32 v0, v4, v1, v0
; GISEL-NEXT: v_bfe_u32 v2, v0, 16, 1
; GISEL-NEXT: v_mov_b32_e32 v3, 0x7fff
; GISEL-NEXT: v_or_b32_e32 v1, 0x400000, v0
; GISEL-NEXT: v_add3_u32 v2, v2, v0, v3
; GISEL-NEXT: v_cmp_u_f32_e32 vcc, 0, v0
; GISEL-NEXT: v_cndmask_b32_e32 v0, v2, v1, vcc
; GISEL-NEXT: v_lshrrev_b32_e32 v4, 16, v0
; GISEL-NEXT: .LBB1_12: ; %Flow5
; GISEL-NEXT: s_or_b64 exec, exec, s[6:7]
; GISEL-NEXT: v_mov_b32_e32 v0, v4
; GISEL-NEXT: s_setpc_b64 s[30:31]
; GISEL-NEXT: .LBB1_13: ; %itofp-if-else
; GISEL-NEXT: v_add_u32_e32 v2, 0xffffff98, v5
; GISEL-NEXT: v_lshlrev_b64 v[0:1], v2, v[0:1]
; GISEL-NEXT: v_cmp_gt_u32_e32 vcc, 64, v2
; GISEL-NEXT: v_cndmask_b32_e32 v4, 0, v0, vcc
; GISEL-NEXT: ; implicit-def: $vgpr6
; GISEL-NEXT: ; implicit-def: $vgpr0_vgpr1_vgpr2_vgpr3
; GISEL-NEXT: s_branch .LBB1_2
; GISEL-NEXT: .LBB1_14: ; %itofp-if-then20
; GISEL-NEXT: v_lshrrev_b64 v[4:5], 3, v[0:1]
; GISEL-NEXT: v_mov_b32_e32 v7, v6
; GISEL-NEXT: s_branch .LBB1_10
%cvt = uitofp i128 %x to bfloat
ret bfloat %cvt
}
;; NOTE: These prefixes are unused and the list is autogenerated. Do not add tests below this line:
; GCN: {{.*}}