Mirko Brkušanin 80f3b376b3
[AMDGPU][GlobalISel] Combine for breaking s64 and/or into two s32 insts (#151731)
When either one of the operands is all ones in high or low parts,
splitting these opens up other opportunities for combines. One of two
new instructions will either be removed or become a simple copy.
2025-08-20 17:32:29 +02:00

1608 lines
69 KiB
LLVM

; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 4
; RUN: llc -global-isel=0 -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 < %s | FileCheck -check-prefixes=GCN,SDAG %s
; RUN: llc -global-isel=1 -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 < %s | FileCheck -check-prefixes=GCN,GISEL %s
define float @sitofp_i128_to_f32(i128 %x) {
; SDAG-LABEL: sitofp_i128_to_f32:
; 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_14
; SDAG-NEXT: ; %bb.1: ; %itofp-if-end
; SDAG-NEXT: v_sub_co_u32_e32 v4, vcc, 0, v0
; SDAG-NEXT: v_subb_co_u32_e32 v5, vcc, 0, v1, vcc
; SDAG-NEXT: v_subb_co_u32_e32 v6, vcc, 0, v2, vcc
; SDAG-NEXT: v_subb_co_u32_e32 v7, vcc, 0, v3, vcc
; SDAG-NEXT: v_cmp_gt_i64_e32 vcc, 0, v[2:3]
; SDAG-NEXT: ; implicit-def: $vgpr8
; SDAG-NEXT: v_cndmask_b32_e32 v0, v0, v4, vcc
; SDAG-NEXT: v_cndmask_b32_e32 v4, v2, v6, vcc
; SDAG-NEXT: v_cndmask_b32_e32 v1, v1, v5, vcc
; SDAG-NEXT: v_cndmask_b32_e32 v5, v3, v7, 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: s_and_saveexec_b64 s[4:5], vcc
; SDAG-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
; SDAG-NEXT: ; %bb.2: ; %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: ; %bb.3: ; %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_13
; SDAG-NEXT: ; %bb.4: ; %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_8
; SDAG-NEXT: ; %bb.5: ; %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_7
; SDAG-NEXT: ; %bb.6: ; %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_7: ; %Flow1
; SDAG-NEXT: s_or_b64 exec, exec, s[12:13]
; SDAG-NEXT: .LBB0_8: ; %Flow2
; SDAG-NEXT: s_andn2_saveexec_b64 s[4:5], s[10:11]
; SDAG-NEXT: ; %bb.9: ; %itofp-sw-bb
; SDAG-NEXT: v_lshlrev_b64 v[0:1], 1, v[0:1]
; SDAG-NEXT: ; %bb.10: ; %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: ; %bb.11: ; %itofp-if-then20
; SDAG-NEXT: v_alignbit_b32 v8, v1, v0, 3
; SDAG-NEXT: v_mov_b32_e32 v6, v2
; SDAG-NEXT: ; %bb.12: ; %Flow
; SDAG-NEXT: s_or_b64 exec, exec, s[4:5]
; SDAG-NEXT: .LBB0_13: ; %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 v4, v2, v0, v1
; SDAG-NEXT: .LBB0_14: ; %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]
;
; GISEL-LABEL: sitofp_i128_to_f32:
; 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: v_cmp_ne_u64_e32 vcc, 0, v[4:5]
; GISEL-NEXT: s_mov_b32 s4, 0
; GISEL-NEXT: v_mov_b32_e32 v4, s4
; GISEL-NEXT: s_and_saveexec_b64 s[6:7], vcc
; GISEL-NEXT: s_cbranch_execz .LBB0_14
; 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: ; %bb.2: ; %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: ; %bb.3: ; %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_13
; GISEL-NEXT: ; %bb.4: ; %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_8
; GISEL-NEXT: ; %bb.5: ; %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_7
; GISEL-NEXT: ; %bb.6: ; %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_7: ; %Flow1
; GISEL-NEXT: s_or_b64 exec, exec, s[12:13]
; GISEL-NEXT: .LBB0_8: ; %Flow2
; GISEL-NEXT: s_andn2_saveexec_b64 s[4:5], s[10:11]
; GISEL-NEXT: ; %bb.9: ; %itofp-sw-bb
; GISEL-NEXT: v_lshlrev_b64 v[0:1], 1, v[0:1]
; GISEL-NEXT: ; %bb.10: ; %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_addc_co_u32_e32 v1, vcc, 0, v1, vcc
; GISEL-NEXT: v_and_b32_e32 v2, 0x4000000, v0
; GISEL-NEXT: v_mov_b32_e32 v3, 0
; 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: ; %bb.11: ; %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: ; %bb.12: ; %Flow
; GISEL-NEXT: s_or_b64 exec, exec, s[4:5]
; GISEL-NEXT: .LBB0_13: ; %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 v4, v2, v0, v1
; GISEL-NEXT: .LBB0_14: ; %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]
%cvt = sitofp i128 %x to float
ret float %cvt
}
define float @uitofp_i128_to_f32(i128 %x) {
; SDAG-LABEL: uitofp_i128_to_f32:
; 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_14
; 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: ; %bb.2: ; %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: $vgpr4
; SDAG-NEXT: ; implicit-def: $vgpr0_vgpr1
; SDAG-NEXT: ; implicit-def: $vgpr2_vgpr3
; SDAG-NEXT: ; %bb.3: ; %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_13
; SDAG-NEXT: ; %bb.4: ; %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_8
; SDAG-NEXT: ; %bb.5: ; %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_7
; SDAG-NEXT: ; %bb.6: ; %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_7: ; %Flow1
; SDAG-NEXT: s_or_b64 exec, exec, s[12:13]
; SDAG-NEXT: .LBB1_8: ; %Flow2
; SDAG-NEXT: s_andn2_saveexec_b64 s[4:5], s[10:11]
; SDAG-NEXT: ; %bb.9: ; %itofp-sw-bb
; SDAG-NEXT: v_lshlrev_b64 v[0:1], 1, v[0:1]
; SDAG-NEXT: ; %bb.10: ; %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: ; %bb.11: ; %itofp-if-then20
; SDAG-NEXT: v_alignbit_b32 v7, v1, v0, 3
; SDAG-NEXT: v_mov_b32_e32 v5, v4
; SDAG-NEXT: ; %bb.12: ; %Flow
; SDAG-NEXT: s_or_b64 exec, exec, s[4:5]
; SDAG-NEXT: .LBB1_13: ; %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 v4, 1.0, v0
; SDAG-NEXT: .LBB1_14: ; %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]
;
; GISEL-LABEL: uitofp_i128_to_f32:
; 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: v_cmp_ne_u64_e32 vcc, 0, v[4:5]
; GISEL-NEXT: s_mov_b32 s4, 0
; GISEL-NEXT: v_mov_b32_e32 v4, s4
; GISEL-NEXT: s_and_saveexec_b64 s[6:7], vcc
; GISEL-NEXT: s_cbranch_execz .LBB1_14
; 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: ; %bb.2: ; %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
; GISEL-NEXT: ; implicit-def: $vgpr2
; GISEL-NEXT: ; %bb.3: ; %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_13
; GISEL-NEXT: ; %bb.4: ; %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_8
; GISEL-NEXT: ; %bb.5: ; %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_7
; GISEL-NEXT: ; %bb.6: ; %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_7: ; %Flow1
; GISEL-NEXT: s_or_b64 exec, exec, s[12:13]
; GISEL-NEXT: .LBB1_8: ; %Flow2
; GISEL-NEXT: s_andn2_saveexec_b64 s[4:5], s[10:11]
; GISEL-NEXT: ; %bb.9: ; %itofp-sw-bb
; GISEL-NEXT: v_lshlrev_b64 v[0:1], 1, v[0:1]
; GISEL-NEXT: ; %bb.10: ; %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_addc_co_u32_e32 v1, vcc, 0, v1, vcc
; GISEL-NEXT: v_and_b32_e32 v2, 0x4000000, v0
; GISEL-NEXT: v_mov_b32_e32 v3, 0
; 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: ; %bb.11: ; %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: ; %bb.12: ; %Flow
; GISEL-NEXT: s_or_b64 exec, exec, s[4:5]
; GISEL-NEXT: .LBB1_13: ; %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 v4, v4, v1, v0
; GISEL-NEXT: .LBB1_14: ; %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]
%cvt = uitofp i128 %x to float
ret float %cvt
}
define double @sitofp_i128_to_f64(i128 %x) {
; SDAG-LABEL: sitofp_i128_to_f64:
; SDAG: ; %bb.0: ; %itofp-entry
; SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; SDAG-NEXT: v_mov_b32_e32 v5, v1
; SDAG-NEXT: v_mov_b32_e32 v4, v0
; SDAG-NEXT: v_or_b32_e32 v1, v5, v3
; SDAG-NEXT: v_or_b32_e32 v0, v4, v2
; SDAG-NEXT: v_cmp_ne_u64_e32 vcc, 0, v[0:1]
; SDAG-NEXT: v_mov_b32_e32 v0, 0
; SDAG-NEXT: v_mov_b32_e32 v1, 0
; SDAG-NEXT: s_and_saveexec_b64 s[6:7], vcc
; SDAG-NEXT: s_cbranch_execz .LBB2_14
; SDAG-NEXT: ; %bb.1: ; %itofp-if-end
; SDAG-NEXT: v_sub_co_u32_e32 v0, vcc, 0, v4
; SDAG-NEXT: v_subb_co_u32_e32 v1, vcc, 0, v5, vcc
; SDAG-NEXT: v_subb_co_u32_e32 v6, vcc, 0, v2, vcc
; SDAG-NEXT: v_subb_co_u32_e32 v7, vcc, 0, v3, vcc
; SDAG-NEXT: v_cmp_gt_i64_e32 vcc, 0, v[2:3]
; SDAG-NEXT: ; implicit-def: $vgpr10
; SDAG-NEXT: v_cndmask_b32_e32 v6, v2, v6, vcc
; SDAG-NEXT: v_cndmask_b32_e32 v4, v4, v0, vcc
; SDAG-NEXT: v_cndmask_b32_e32 v7, v3, v7, vcc
; SDAG-NEXT: v_ffbh_u32_e32 v0, v6
; SDAG-NEXT: v_cndmask_b32_e32 v5, v5, v1, vcc
; SDAG-NEXT: v_add_u32_e32 v0, 32, v0
; SDAG-NEXT: v_ffbh_u32_e32 v1, v7
; SDAG-NEXT: v_min_u32_e32 v0, v0, v1
; SDAG-NEXT: v_ffbh_u32_e32 v1, v4
; SDAG-NEXT: v_add_u32_e32 v1, 32, v1
; SDAG-NEXT: v_ffbh_u32_e32 v2, v5
; SDAG-NEXT: v_min_u32_e32 v1, v1, v2
; SDAG-NEXT: v_cmp_ne_u64_e32 vcc, 0, v[6:7]
; SDAG-NEXT: v_add_u32_e32 v1, 64, v1
; SDAG-NEXT: v_cndmask_b32_e32 v9, v1, v0, vcc
; SDAG-NEXT: v_sub_u32_e32 v2, 0x80, v9
; SDAG-NEXT: v_cmp_gt_i32_e32 vcc, 54, v2
; SDAG-NEXT: ; implicit-def: $vgpr0_vgpr1
; SDAG-NEXT: s_and_saveexec_b64 s[4:5], vcc
; SDAG-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
; SDAG-NEXT: ; %bb.2: ; %itofp-if-else
; SDAG-NEXT: v_add_u32_e32 v2, 0xffffffb5, v9
; SDAG-NEXT: v_lshlrev_b64 v[0:1], v2, v[4:5]
; SDAG-NEXT: v_cmp_gt_u32_e32 vcc, 64, v2
; SDAG-NEXT: v_cndmask_b32_e32 v10, 0, v1, vcc
; SDAG-NEXT: v_cndmask_b32_e32 v0, 0, v0, vcc
; SDAG-NEXT: ; implicit-def: $vgpr2
; SDAG-NEXT: ; implicit-def: $vgpr6_vgpr7
; SDAG-NEXT: ; implicit-def: $vgpr4_vgpr5
; SDAG-NEXT: ; %bb.3: ; %Flow3
; SDAG-NEXT: s_or_saveexec_b64 s[8:9], s[4:5]
; SDAG-NEXT: v_sub_u32_e32 v8, 0x7f, v9
; SDAG-NEXT: s_xor_b64 exec, exec, s[8:9]
; SDAG-NEXT: s_cbranch_execz .LBB2_13
; SDAG-NEXT: ; %bb.4: ; %NodeBlock
; SDAG-NEXT: v_cmp_lt_i32_e32 vcc, 54, 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 .LBB2_8
; SDAG-NEXT: ; %bb.5: ; %LeafBlock
; SDAG-NEXT: v_cmp_ne_u32_e32 vcc, 55, v2
; SDAG-NEXT: s_and_saveexec_b64 s[12:13], vcc
; SDAG-NEXT: s_cbranch_execz .LBB2_7
; SDAG-NEXT: ; %bb.6: ; %itofp-sw-default
; SDAG-NEXT: v_sub_u32_e32 v12, 0x49, v9
; SDAG-NEXT: v_sub_u32_e32 v10, 64, v12
; SDAG-NEXT: v_lshrrev_b64 v[0:1], v12, v[4:5]
; SDAG-NEXT: v_lshlrev_b64 v[10:11], v10, v[6:7]
; SDAG-NEXT: v_sub_u32_e32 v13, 9, v9
; SDAG-NEXT: v_or_b32_e32 v11, v1, v11
; SDAG-NEXT: v_or_b32_e32 v10, v0, v10
; SDAG-NEXT: v_lshrrev_b64 v[0:1], v13, v[6:7]
; SDAG-NEXT: v_cmp_gt_u32_e32 vcc, 64, v12
; SDAG-NEXT: v_add_u32_e32 v16, 55, v9
; SDAG-NEXT: v_cndmask_b32_e32 v1, v1, v11, vcc
; SDAG-NEXT: v_cmp_eq_u32_e64 s[4:5], 0, v12
; SDAG-NEXT: v_cndmask_b32_e32 v0, v0, v10, vcc
; SDAG-NEXT: v_lshrrev_b64 v[10:11], v12, v[6:7]
; SDAG-NEXT: v_lshrrev_b64 v[12:13], v13, v[4:5]
; SDAG-NEXT: v_lshlrev_b64 v[14:15], v16, v[6:7]
; SDAG-NEXT: v_add_u32_e32 v9, -9, v9
; SDAG-NEXT: v_or_b32_e32 v15, v15, v13
; SDAG-NEXT: v_or_b32_e32 v14, v14, v12
; SDAG-NEXT: v_lshlrev_b64 v[12:13], v9, v[4:5]
; SDAG-NEXT: v_cndmask_b32_e32 v11, 0, v11, vcc
; SDAG-NEXT: v_cndmask_b32_e32 v10, 0, v10, vcc
; SDAG-NEXT: v_cmp_gt_u32_e32 vcc, 64, v16
; SDAG-NEXT: v_cndmask_b32_e64 v1, v1, v5, s[4:5]
; SDAG-NEXT: v_cndmask_b32_e64 v0, v0, v4, s[4:5]
; SDAG-NEXT: v_cndmask_b32_e32 v9, v13, v15, vcc
; SDAG-NEXT: v_cmp_eq_u32_e64 s[4:5], 0, v16
; SDAG-NEXT: v_lshlrev_b64 v[4:5], v16, v[4:5]
; SDAG-NEXT: v_cndmask_b32_e64 v7, v9, v7, s[4:5]
; SDAG-NEXT: v_cndmask_b32_e32 v9, v12, v14, vcc
; SDAG-NEXT: v_cndmask_b32_e64 v6, v9, v6, s[4:5]
; SDAG-NEXT: v_cndmask_b32_e32 v5, 0, v5, vcc
; SDAG-NEXT: v_cndmask_b32_e32 v4, 0, v4, vcc
; SDAG-NEXT: v_or_b32_e32 v5, v5, v7
; SDAG-NEXT: v_or_b32_e32 v4, v4, v6
; SDAG-NEXT: v_cmp_ne_u64_e32 vcc, 0, v[4:5]
; SDAG-NEXT: v_mov_b32_e32 v6, v10
; SDAG-NEXT: v_cndmask_b32_e64 v4, 0, 1, vcc
; SDAG-NEXT: v_or_b32_e32 v0, v0, v4
; SDAG-NEXT: v_mov_b32_e32 v5, v1
; SDAG-NEXT: v_mov_b32_e32 v4, v0
; SDAG-NEXT: v_mov_b32_e32 v7, v11
; SDAG-NEXT: .LBB2_7: ; %Flow1
; SDAG-NEXT: s_or_b64 exec, exec, s[12:13]
; SDAG-NEXT: .LBB2_8: ; %Flow2
; SDAG-NEXT: s_andn2_saveexec_b64 s[4:5], s[10:11]
; SDAG-NEXT: ; %bb.9: ; %itofp-sw-bb
; SDAG-NEXT: v_lshlrev_b64 v[6:7], 1, v[6:7]
; SDAG-NEXT: v_lshrrev_b32_e32 v0, 31, v5
; SDAG-NEXT: v_lshlrev_b64 v[4:5], 1, v[4:5]
; SDAG-NEXT: v_or_b32_e32 v6, v6, v0
; SDAG-NEXT: ; %bb.10: ; %itofp-sw-epilog
; SDAG-NEXT: s_or_b64 exec, exec, s[4:5]
; SDAG-NEXT: v_lshrrev_b32_e32 v0, 2, v4
; SDAG-NEXT: v_and_or_b32 v0, v0, 1, v4
; SDAG-NEXT: v_add_co_u32_e32 v4, vcc, 1, v0
; SDAG-NEXT: v_addc_co_u32_e32 v5, vcc, 0, v5, vcc
; SDAG-NEXT: v_addc_co_u32_e32 v6, vcc, 0, v6, vcc
; SDAG-NEXT: v_lshrrev_b64 v[0:1], 2, v[4:5]
; SDAG-NEXT: v_lshlrev_b32_e32 v7, 30, v6
; SDAG-NEXT: v_or_b32_e32 v10, v1, v7
; SDAG-NEXT: v_and_b32_e32 v1, 0x800000, v5
; SDAG-NEXT: v_cmp_ne_u32_e32 vcc, 0, v1
; SDAG-NEXT: s_and_saveexec_b64 s[4:5], vcc
; SDAG-NEXT: ; %bb.11: ; %itofp-if-then20
; SDAG-NEXT: v_lshrrev_b64 v[0:1], 3, v[4:5]
; SDAG-NEXT: v_lshlrev_b32_e32 v4, 29, v6
; SDAG-NEXT: v_or_b32_e32 v10, v1, v4
; SDAG-NEXT: v_mov_b32_e32 v8, v2
; SDAG-NEXT: ; %bb.12: ; %Flow
; SDAG-NEXT: s_or_b64 exec, exec, s[4:5]
; SDAG-NEXT: .LBB2_13: ; %Flow4
; SDAG-NEXT: s_or_b64 exec, exec, s[8:9]
; SDAG-NEXT: v_mov_b32_e32 v2, 0x3ff00000
; SDAG-NEXT: v_and_b32_e32 v1, 0x80000000, v3
; SDAG-NEXT: v_lshl_add_u32 v2, v8, 20, v2
; SDAG-NEXT: v_and_b32_e32 v3, 0xfffff, v10
; SDAG-NEXT: v_or3_b32 v1, v3, v1, v2
; SDAG-NEXT: .LBB2_14: ; %Flow5
; SDAG-NEXT: s_or_b64 exec, exec, s[6:7]
; SDAG-NEXT: s_setpc_b64 s[30:31]
;
; GISEL-LABEL: sitofp_i128_to_f64:
; GISEL: ; %bb.0: ; %itofp-entry
; GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GISEL-NEXT: v_mov_b32_e32 v4, v0
; GISEL-NEXT: v_mov_b32_e32 v5, v1
; GISEL-NEXT: s_mov_b64 s[4:5], 0
; GISEL-NEXT: v_or_b32_e32 v0, v4, v2
; GISEL-NEXT: v_or_b32_e32 v1, v5, v3
; GISEL-NEXT: v_cmp_ne_u64_e32 vcc, 0, v[0:1]
; GISEL-NEXT: v_mov_b32_e32 v0, s4
; GISEL-NEXT: v_mov_b32_e32 v1, s5
; GISEL-NEXT: s_and_saveexec_b64 s[6:7], vcc
; GISEL-NEXT: s_cbranch_execz .LBB2_14
; GISEL-NEXT: ; %bb.1: ; %itofp-if-end
; GISEL-NEXT: v_ashrrev_i32_e32 v6, 31, v3
; GISEL-NEXT: v_xor_b32_e32 v0, v6, v4
; GISEL-NEXT: v_xor_b32_e32 v1, v6, v5
; GISEL-NEXT: v_xor_b32_e32 v4, v6, v2
; GISEL-NEXT: v_sub_co_u32_e32 v2, vcc, v0, v6
; GISEL-NEXT: v_xor_b32_e32 v5, v6, v3
; GISEL-NEXT: v_subb_co_u32_e32 v3, vcc, v1, v6, vcc
; GISEL-NEXT: v_subb_co_u32_e32 v4, vcc, v4, v6, vcc
; GISEL-NEXT: v_ffbh_u32_e32 v1, v2
; GISEL-NEXT: v_subb_co_u32_e32 v5, vcc, v5, v6, vcc
; GISEL-NEXT: v_ffbh_u32_e32 v0, v3
; GISEL-NEXT: v_add_u32_e32 v1, 32, v1
; GISEL-NEXT: v_ffbh_u32_e32 v7, v4
; GISEL-NEXT: v_min_u32_e32 v0, v0, v1
; GISEL-NEXT: v_ffbh_u32_e32 v1, v5
; GISEL-NEXT: v_add_u32_e32 v7, 32, v7
; GISEL-NEXT: v_cmp_eq_u64_e32 vcc, 0, v[4:5]
; GISEL-NEXT: v_add_u32_e32 v0, 64, v0
; GISEL-NEXT: v_min_u32_e32 v1, v1, v7
; GISEL-NEXT: v_cndmask_b32_e32 v9, v1, v0, vcc
; GISEL-NEXT: v_sub_u32_e32 v7, 0x80, v9
; GISEL-NEXT: v_cmp_ge_i32_e32 vcc, 53, v7
; GISEL-NEXT: ; implicit-def: $vgpr10
; GISEL-NEXT: ; implicit-def: $vgpr0_vgpr1
; GISEL-NEXT: s_and_saveexec_b64 s[4:5], vcc
; GISEL-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
; GISEL-NEXT: ; %bb.2: ; %itofp-if-else
; GISEL-NEXT: v_add_u32_e32 v4, 0xffffffb5, v9
; GISEL-NEXT: v_lshlrev_b64 v[0:1], v4, v[2:3]
; GISEL-NEXT: v_cmp_gt_u32_e32 vcc, 64, v4
; GISEL-NEXT: v_cndmask_b32_e32 v0, 0, v0, vcc
; GISEL-NEXT: v_cndmask_b32_e32 v10, 0, v1, vcc
; GISEL-NEXT: ; implicit-def: $vgpr7
; GISEL-NEXT: ; implicit-def: $vgpr2
; GISEL-NEXT: ; %bb.3: ; %Flow3
; GISEL-NEXT: s_or_saveexec_b64 s[8:9], s[4:5]
; GISEL-NEXT: v_sub_u32_e32 v8, 0x7f, v9
; GISEL-NEXT: s_xor_b64 exec, exec, s[8:9]
; GISEL-NEXT: s_cbranch_execz .LBB2_13
; GISEL-NEXT: ; %bb.4: ; %NodeBlock
; GISEL-NEXT: v_cmp_le_i32_e32 vcc, 55, 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 .LBB2_8
; GISEL-NEXT: ; %bb.5: ; %LeafBlock
; GISEL-NEXT: v_cmp_ne_u32_e32 vcc, 55, v7
; GISEL-NEXT: s_and_saveexec_b64 s[12:13], vcc
; GISEL-NEXT: s_cbranch_execz .LBB2_7
; GISEL-NEXT: ; %bb.6: ; %itofp-sw-default
; GISEL-NEXT: v_sub_u32_e32 v14, 0x49, v9
; GISEL-NEXT: v_sub_u32_e32 v10, 64, v14
; GISEL-NEXT: v_lshrrev_b64 v[0:1], v14, v[2:3]
; GISEL-NEXT: v_lshlrev_b64 v[10:11], v10, v[4:5]
; GISEL-NEXT: v_add_u32_e32 v15, 0xffffffc0, v14
; GISEL-NEXT: v_lshrrev_b64 v[12:13], v14, v[4:5]
; GISEL-NEXT: v_or_b32_e32 v10, v0, v10
; GISEL-NEXT: v_or_b32_e32 v11, v1, v11
; GISEL-NEXT: v_lshrrev_b64 v[0:1], v15, v[4:5]
; GISEL-NEXT: v_cmp_gt_u32_e32 vcc, 64, v14
; GISEL-NEXT: v_add_u32_e32 v15, 55, v9
; GISEL-NEXT: v_cndmask_b32_e32 v0, v0, v10, vcc
; GISEL-NEXT: v_cndmask_b32_e32 v1, v1, v11, vcc
; GISEL-NEXT: v_cmp_eq_u32_e64 s[4:5], 0, v14
; GISEL-NEXT: v_cndmask_b32_e32 v11, 0, v12, vcc
; GISEL-NEXT: v_sub_u32_e32 v12, 64, v15
; GISEL-NEXT: v_cndmask_b32_e64 v14, v0, v2, s[4:5]
; GISEL-NEXT: v_cndmask_b32_e64 v10, v1, v3, s[4:5]
; GISEL-NEXT: v_lshrrev_b64 v[0:1], v15, -1
; GISEL-NEXT: v_lshlrev_b64 v[12:13], v12, -1
; GISEL-NEXT: v_add_u32_e32 v9, -9, v9
; GISEL-NEXT: v_or_b32_e32 v16, v0, v12
; GISEL-NEXT: v_or_b32_e32 v17, v1, v13
; GISEL-NEXT: v_lshrrev_b64 v[12:13], v9, -1
; GISEL-NEXT: v_cmp_gt_u32_e32 vcc, 64, v15
; GISEL-NEXT: v_cndmask_b32_e32 v9, v12, v16, vcc
; GISEL-NEXT: v_cndmask_b32_e32 v12, v13, v17, vcc
; GISEL-NEXT: v_cmp_eq_u32_e64 s[4:5], 0, v15
; GISEL-NEXT: v_cndmask_b32_e32 v0, 0, v0, vcc
; GISEL-NEXT: v_cndmask_b32_e32 v1, 0, v1, vcc
; GISEL-NEXT: v_cndmask_b32_e64 v9, v9, -1, s[4:5]
; GISEL-NEXT: v_cndmask_b32_e64 v12, v12, -1, s[4:5]
; GISEL-NEXT: v_and_b32_e32 v0, v0, v4
; GISEL-NEXT: v_and_b32_e32 v1, v1, v5
; GISEL-NEXT: v_and_or_b32 v0, v9, v2, v0
; GISEL-NEXT: v_and_or_b32 v1, v12, v3, v1
; 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 v9, v14, v0
; GISEL-NEXT: v_mov_b32_e32 v2, v9
; GISEL-NEXT: v_mov_b32_e32 v3, v10
; GISEL-NEXT: v_mov_b32_e32 v4, v11
; GISEL-NEXT: v_mov_b32_e32 v5, v12
; GISEL-NEXT: .LBB2_7: ; %Flow1
; GISEL-NEXT: s_or_b64 exec, exec, s[12:13]
; GISEL-NEXT: .LBB2_8: ; %Flow2
; GISEL-NEXT: s_andn2_saveexec_b64 s[4:5], s[10:11]
; GISEL-NEXT: s_cbranch_execz .LBB2_10
; GISEL-NEXT: ; %bb.9: ; %itofp-sw-bb
; GISEL-NEXT: v_lshlrev_b64 v[4:5], 1, v[4:5]
; GISEL-NEXT: v_lshlrev_b64 v[0:1], 1, v[2:3]
; GISEL-NEXT: v_lshrrev_b32_e32 v2, 31, v3
; GISEL-NEXT: v_or_b32_e32 v2, v4, v2
; GISEL-NEXT: v_mov_b32_e32 v5, v3
; GISEL-NEXT: v_mov_b32_e32 v4, v2
; GISEL-NEXT: v_mov_b32_e32 v3, v1
; GISEL-NEXT: v_mov_b32_e32 v2, v0
; GISEL-NEXT: .LBB2_10: ; %itofp-sw-epilog
; GISEL-NEXT: s_or_b64 exec, exec, s[4:5]
; GISEL-NEXT: v_bfe_u32 v0, v2, 2, 1
; GISEL-NEXT: v_or_b32_e32 v0, v2, v0
; GISEL-NEXT: v_add_co_u32_e32 v2, vcc, 1, v0
; GISEL-NEXT: v_addc_co_u32_e32 v3, vcc, 0, v3, vcc
; GISEL-NEXT: v_addc_co_u32_e32 v4, vcc, 0, v4, vcc
; GISEL-NEXT: v_lshrrev_b64 v[0:1], 2, v[2:3]
; GISEL-NEXT: v_mov_b32_e32 v9, 0
; GISEL-NEXT: v_and_b32_e32 v10, 0x800000, v3
; GISEL-NEXT: v_cmp_ne_u64_e32 vcc, 0, v[9:10]
; GISEL-NEXT: v_lshl_or_b32 v10, v4, 30, v1
; GISEL-NEXT: s_and_saveexec_b64 s[4:5], vcc
; GISEL-NEXT: ; %bb.11: ; %itofp-if-then20
; GISEL-NEXT: v_lshrrev_b64 v[0:1], 3, v[2:3]
; GISEL-NEXT: v_mov_b32_e32 v8, v7
; GISEL-NEXT: v_lshl_or_b32 v10, v4, 29, v1
; GISEL-NEXT: ; %bb.12: ; %Flow
; GISEL-NEXT: s_or_b64 exec, exec, s[4:5]
; GISEL-NEXT: .LBB2_13: ; %Flow4
; GISEL-NEXT: s_or_b64 exec, exec, s[8:9]
; GISEL-NEXT: v_mov_b32_e32 v2, 0x3ff00000
; GISEL-NEXT: v_and_b32_e32 v1, 0x80000000, v6
; GISEL-NEXT: v_lshl_add_u32 v2, v8, 20, v2
; GISEL-NEXT: v_and_b32_e32 v3, 0xfffff, v10
; GISEL-NEXT: v_or3_b32 v1, v3, v1, v2
; GISEL-NEXT: .LBB2_14: ; %Flow5
; GISEL-NEXT: s_or_b64 exec, exec, s[6:7]
; GISEL-NEXT: s_setpc_b64 s[30:31]
%cvt = sitofp i128 %x to double
ret double %cvt
}
define double @uitofp_i128_to_f64(i128 %x) {
; SDAG-LABEL: uitofp_i128_to_f64:
; 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: v_mov_b32_e32 v5, 0
; SDAG-NEXT: s_and_saveexec_b64 s[6:7], vcc
; SDAG-NEXT: s_cbranch_execz .LBB3_14
; 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 v8, v5, v4, vcc
; SDAG-NEXT: v_sub_u32_e32 v6, 0x80, v8
; SDAG-NEXT: v_cmp_gt_i32_e32 vcc, 54, v6
; SDAG-NEXT: ; implicit-def: $vgpr9
; SDAG-NEXT: ; implicit-def: $vgpr4_vgpr5
; SDAG-NEXT: s_and_saveexec_b64 s[4:5], vcc
; SDAG-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
; SDAG-NEXT: ; %bb.2: ; %itofp-if-else
; SDAG-NEXT: v_add_u32_e32 v2, 0xffffffb5, v8
; 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 v9, 0, v1, vcc
; SDAG-NEXT: v_cndmask_b32_e32 v4, 0, v0, vcc
; SDAG-NEXT: ; implicit-def: $vgpr6
; SDAG-NEXT: ; implicit-def: $vgpr2_vgpr3
; SDAG-NEXT: ; implicit-def: $vgpr0_vgpr1
; SDAG-NEXT: ; %bb.3: ; %Flow3
; SDAG-NEXT: s_or_saveexec_b64 s[8:9], s[4:5]
; SDAG-NEXT: v_sub_u32_e32 v7, 0x7f, v8
; SDAG-NEXT: s_xor_b64 exec, exec, s[8:9]
; SDAG-NEXT: s_cbranch_execz .LBB3_13
; SDAG-NEXT: ; %bb.4: ; %NodeBlock
; SDAG-NEXT: v_cmp_lt_i32_e32 vcc, 54, v6
; 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 .LBB3_8
; SDAG-NEXT: ; %bb.5: ; %LeafBlock
; SDAG-NEXT: v_cmp_ne_u32_e32 vcc, 55, v6
; SDAG-NEXT: s_and_saveexec_b64 s[12:13], vcc
; SDAG-NEXT: s_cbranch_execz .LBB3_7
; SDAG-NEXT: ; %bb.6: ; %itofp-sw-default
; SDAG-NEXT: v_sub_u32_e32 v11, 0x49, v8
; SDAG-NEXT: v_sub_u32_e32 v9, 64, v11
; SDAG-NEXT: v_lshrrev_b64 v[4:5], v11, v[0:1]
; SDAG-NEXT: v_lshlrev_b64 v[9:10], v9, v[2:3]
; SDAG-NEXT: v_sub_u32_e32 v12, 9, v8
; SDAG-NEXT: v_or_b32_e32 v10, v5, v10
; SDAG-NEXT: v_or_b32_e32 v9, v4, v9
; SDAG-NEXT: v_lshrrev_b64 v[4:5], v12, v[2:3]
; SDAG-NEXT: v_cmp_gt_u32_e32 vcc, 64, v11
; SDAG-NEXT: v_add_u32_e32 v15, 55, v8
; SDAG-NEXT: v_cndmask_b32_e32 v5, v5, v10, vcc
; SDAG-NEXT: v_cmp_eq_u32_e64 s[4:5], 0, v11
; SDAG-NEXT: v_cndmask_b32_e32 v4, v4, v9, vcc
; SDAG-NEXT: v_lshrrev_b64 v[9:10], v11, v[2:3]
; SDAG-NEXT: v_lshrrev_b64 v[11:12], v12, v[0:1]
; SDAG-NEXT: v_lshlrev_b64 v[13:14], v15, v[2:3]
; SDAG-NEXT: v_add_u32_e32 v8, -9, v8
; SDAG-NEXT: v_or_b32_e32 v14, v14, v12
; SDAG-NEXT: v_or_b32_e32 v13, v13, v11
; SDAG-NEXT: v_lshlrev_b64 v[11:12], v8, v[0:1]
; SDAG-NEXT: v_cndmask_b32_e32 v10, 0, v10, vcc
; SDAG-NEXT: v_cndmask_b32_e32 v9, 0, v9, vcc
; SDAG-NEXT: v_cmp_gt_u32_e32 vcc, 64, v15
; SDAG-NEXT: v_cndmask_b32_e64 v5, v5, v1, s[4:5]
; SDAG-NEXT: v_cndmask_b32_e64 v4, v4, v0, s[4:5]
; SDAG-NEXT: v_cndmask_b32_e32 v8, v12, v14, vcc
; SDAG-NEXT: v_cmp_eq_u32_e64 s[4:5], 0, v15
; SDAG-NEXT: v_lshlrev_b64 v[0:1], v15, v[0:1]
; SDAG-NEXT: v_cndmask_b32_e64 v3, v8, v3, s[4:5]
; SDAG-NEXT: v_cndmask_b32_e32 v8, v11, v13, vcc
; SDAG-NEXT: v_cndmask_b32_e64 v2, v8, 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_mov_b32_e32 v2, v9
; SDAG-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc
; SDAG-NEXT: v_or_b32_e32 v4, v4, v0
; SDAG-NEXT: v_mov_b32_e32 v0, v4
; SDAG-NEXT: v_mov_b32_e32 v1, v5
; SDAG-NEXT: v_mov_b32_e32 v3, v10
; SDAG-NEXT: .LBB3_7: ; %Flow1
; SDAG-NEXT: s_or_b64 exec, exec, s[12:13]
; SDAG-NEXT: .LBB3_8: ; %Flow2
; SDAG-NEXT: s_andn2_saveexec_b64 s[4:5], s[10:11]
; SDAG-NEXT: ; %bb.9: ; %itofp-sw-bb
; SDAG-NEXT: v_lshlrev_b64 v[2:3], 1, v[2:3]
; SDAG-NEXT: v_lshrrev_b32_e32 v3, 31, v1
; SDAG-NEXT: v_lshlrev_b64 v[0:1], 1, v[0:1]
; SDAG-NEXT: v_or_b32_e32 v2, v2, v3
; SDAG-NEXT: ; %bb.10: ; %itofp-sw-epilog
; SDAG-NEXT: s_or_b64 exec, exec, s[4:5]
; SDAG-NEXT: v_lshrrev_b32_e32 v3, 2, v0
; SDAG-NEXT: v_and_or_b32 v0, v3, 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_addc_co_u32_e32 v2, vcc, 0, v2, vcc
; SDAG-NEXT: v_lshrrev_b64 v[4:5], 2, v[0:1]
; SDAG-NEXT: v_and_b32_e32 v3, 0x800000, v1
; SDAG-NEXT: v_cmp_ne_u32_e32 vcc, 0, v3
; SDAG-NEXT: v_alignbit_b32 v9, v2, v1, 2
; SDAG-NEXT: s_and_saveexec_b64 s[4:5], vcc
; SDAG-NEXT: ; %bb.11: ; %itofp-if-then20
; SDAG-NEXT: v_lshrrev_b64 v[4:5], 3, v[0:1]
; SDAG-NEXT: v_alignbit_b32 v9, v2, v1, 3
; SDAG-NEXT: v_mov_b32_e32 v7, v6
; SDAG-NEXT: ; %bb.12: ; %Flow
; SDAG-NEXT: s_or_b64 exec, exec, s[4:5]
; SDAG-NEXT: .LBB3_13: ; %Flow4
; SDAG-NEXT: s_or_b64 exec, exec, s[8:9]
; SDAG-NEXT: v_and_b32_e32 v0, 0xfffff, v9
; SDAG-NEXT: v_lshl_or_b32 v0, v7, 20, v0
; SDAG-NEXT: v_add_u32_e32 v5, 0x3ff00000, v0
; SDAG-NEXT: .LBB3_14: ; %Flow5
; SDAG-NEXT: s_or_b64 exec, exec, s[6:7]
; SDAG-NEXT: v_mov_b32_e32 v0, v4
; SDAG-NEXT: v_mov_b32_e32 v1, v5
; SDAG-NEXT: s_setpc_b64 s[30:31]
;
; GISEL-LABEL: uitofp_i128_to_f64:
; GISEL: ; %bb.0: ; %itofp-entry
; GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GISEL-NEXT: s_mov_b64 s[4:5], 0
; GISEL-NEXT: v_or_b32_e32 v4, v0, v2
; GISEL-NEXT: v_or_b32_e32 v5, v1, v3
; GISEL-NEXT: v_cmp_ne_u64_e32 vcc, 0, v[4:5]
; GISEL-NEXT: v_mov_b32_e32 v4, s4
; GISEL-NEXT: v_mov_b32_e32 v5, s5
; GISEL-NEXT: s_and_saveexec_b64 s[6:7], vcc
; GISEL-NEXT: s_cbranch_execz .LBB3_14
; 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 v8, v5, v4, vcc
; GISEL-NEXT: v_sub_u32_e32 v6, 0x80, v8
; GISEL-NEXT: v_cmp_ge_i32_e32 vcc, 53, v6
; GISEL-NEXT: ; implicit-def: $vgpr9
; GISEL-NEXT: ; implicit-def: $vgpr4_vgpr5
; GISEL-NEXT: s_and_saveexec_b64 s[4:5], vcc
; GISEL-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
; GISEL-NEXT: ; %bb.2: ; %itofp-if-else
; GISEL-NEXT: v_add_u32_e32 v2, 0xffffffb5, v8
; 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: v_cndmask_b32_e32 v9, 0, v1, vcc
; GISEL-NEXT: ; implicit-def: $vgpr6
; GISEL-NEXT: ; implicit-def: $vgpr0
; GISEL-NEXT: ; %bb.3: ; %Flow3
; GISEL-NEXT: s_or_saveexec_b64 s[8:9], s[4:5]
; GISEL-NEXT: v_sub_u32_e32 v7, 0x7f, v8
; GISEL-NEXT: s_xor_b64 exec, exec, s[8:9]
; GISEL-NEXT: s_cbranch_execz .LBB3_13
; GISEL-NEXT: ; %bb.4: ; %NodeBlock
; GISEL-NEXT: v_cmp_le_i32_e32 vcc, 55, 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 .LBB3_8
; GISEL-NEXT: ; %bb.5: ; %LeafBlock
; GISEL-NEXT: v_cmp_ne_u32_e32 vcc, 55, v6
; GISEL-NEXT: s_and_saveexec_b64 s[12:13], vcc
; GISEL-NEXT: s_cbranch_execz .LBB3_7
; GISEL-NEXT: ; %bb.6: ; %itofp-sw-default
; GISEL-NEXT: v_sub_u32_e32 v13, 0x49, v8
; GISEL-NEXT: v_sub_u32_e32 v9, 64, v13
; GISEL-NEXT: v_lshrrev_b64 v[4:5], v13, v[0:1]
; GISEL-NEXT: v_lshlrev_b64 v[9:10], v9, v[2:3]
; GISEL-NEXT: v_add_u32_e32 v14, 0xffffffc0, v13
; GISEL-NEXT: v_lshrrev_b64 v[11:12], v13, v[2:3]
; GISEL-NEXT: v_or_b32_e32 v9, v4, v9
; GISEL-NEXT: v_or_b32_e32 v10, v5, v10
; GISEL-NEXT: v_lshrrev_b64 v[4:5], v14, v[2:3]
; GISEL-NEXT: v_cmp_gt_u32_e32 vcc, 64, v13
; GISEL-NEXT: v_add_u32_e32 v15, 55, v8
; GISEL-NEXT: v_cndmask_b32_e32 v4, v4, v9, vcc
; GISEL-NEXT: v_cndmask_b32_e32 v5, v5, v10, vcc
; GISEL-NEXT: v_cmp_eq_u32_e64 s[4:5], 0, v13
; GISEL-NEXT: v_cndmask_b32_e32 v10, 0, v11, vcc
; GISEL-NEXT: v_cndmask_b32_e32 v11, 0, v12, vcc
; GISEL-NEXT: v_sub_u32_e32 v12, 64, v15
; GISEL-NEXT: v_cndmask_b32_e64 v14, v4, v0, s[4:5]
; GISEL-NEXT: v_cndmask_b32_e64 v9, v5, v1, s[4:5]
; GISEL-NEXT: v_lshrrev_b64 v[4:5], v15, -1
; GISEL-NEXT: v_lshlrev_b64 v[12:13], v12, -1
; GISEL-NEXT: v_add_u32_e32 v8, -9, v8
; GISEL-NEXT: v_or_b32_e32 v16, v4, v12
; GISEL-NEXT: v_or_b32_e32 v17, v5, v13
; GISEL-NEXT: v_lshrrev_b64 v[12:13], v8, -1
; GISEL-NEXT: v_cmp_gt_u32_e32 vcc, 64, v15
; GISEL-NEXT: v_cndmask_b32_e32 v8, v12, v16, vcc
; GISEL-NEXT: v_cndmask_b32_e32 v12, v13, v17, vcc
; GISEL-NEXT: v_cmp_eq_u32_e64 s[4:5], 0, v15
; GISEL-NEXT: v_cndmask_b32_e32 v4, 0, v4, vcc
; GISEL-NEXT: v_cndmask_b32_e32 v5, 0, v5, vcc
; GISEL-NEXT: v_cndmask_b32_e64 v8, v8, -1, s[4:5]
; GISEL-NEXT: v_cndmask_b32_e64 v12, v12, -1, s[4:5]
; GISEL-NEXT: v_and_b32_e32 v2, v4, v2
; GISEL-NEXT: v_and_b32_e32 v3, v5, v3
; GISEL-NEXT: v_and_or_b32 v0, v8, v0, v2
; GISEL-NEXT: v_and_or_b32 v1, v12, 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 v8, v14, v0
; GISEL-NEXT: v_mov_b32_e32 v0, v8
; GISEL-NEXT: v_mov_b32_e32 v1, v9
; GISEL-NEXT: v_mov_b32_e32 v2, v10
; GISEL-NEXT: v_mov_b32_e32 v3, v11
; GISEL-NEXT: .LBB3_7: ; %Flow1
; GISEL-NEXT: s_or_b64 exec, exec, s[12:13]
; GISEL-NEXT: .LBB3_8: ; %Flow2
; GISEL-NEXT: s_andn2_saveexec_b64 s[4:5], s[10:11]
; GISEL-NEXT: s_cbranch_execz .LBB3_10
; GISEL-NEXT: ; %bb.9: ; %itofp-sw-bb
; GISEL-NEXT: v_lshlrev_b64 v[8:9], 1, v[0:1]
; GISEL-NEXT: v_lshlrev_b64 v[10:11], 1, v[2:3]
; GISEL-NEXT: v_lshrrev_b32_e32 v0, 31, v1
; GISEL-NEXT: v_or_b32_e32 v10, v10, v0
; GISEL-NEXT: v_mov_b32_e32 v0, v8
; GISEL-NEXT: v_mov_b32_e32 v1, v9
; GISEL-NEXT: v_mov_b32_e32 v2, v10
; GISEL-NEXT: v_mov_b32_e32 v3, v11
; GISEL-NEXT: .LBB3_10: ; %itofp-sw-epilog
; GISEL-NEXT: s_or_b64 exec, exec, s[4:5]
; GISEL-NEXT: v_bfe_u32 v4, v0, 2, 1
; GISEL-NEXT: v_or_b32_e32 v0, v0, v4
; GISEL-NEXT: v_add_co_u32_e32 v0, vcc, 1, v0
; GISEL-NEXT: v_addc_co_u32_e32 v1, vcc, 0, v1, vcc
; GISEL-NEXT: v_addc_co_u32_e32 v2, vcc, 0, v2, vcc
; GISEL-NEXT: v_addc_co_u32_e32 v3, vcc, 0, v3, vcc
; GISEL-NEXT: v_mov_b32_e32 v8, 0
; GISEL-NEXT: v_and_b32_e32 v9, 0x800000, v1
; GISEL-NEXT: v_lshrrev_b64 v[4:5], 2, v[0:1]
; GISEL-NEXT: v_cmp_ne_u64_e32 vcc, 0, v[8:9]
; GISEL-NEXT: v_lshlrev_b64 v[8:9], 30, v[2:3]
; GISEL-NEXT: v_lshrrev_b32_e32 v5, 2, v1
; GISEL-NEXT: v_or_b32_e32 v9, v5, v8
; GISEL-NEXT: s_and_saveexec_b64 s[4:5], vcc
; GISEL-NEXT: ; %bb.11: ; %itofp-if-then20
; GISEL-NEXT: v_lshlrev_b64 v[2:3], 29, v[2:3]
; GISEL-NEXT: v_lshrrev_b64 v[4:5], 3, v[0:1]
; GISEL-NEXT: v_lshrrev_b32_e32 v0, 3, v1
; GISEL-NEXT: v_or_b32_e32 v9, v0, v2
; GISEL-NEXT: v_mov_b32_e32 v7, v6
; GISEL-NEXT: ; %bb.12: ; %Flow
; GISEL-NEXT: s_or_b64 exec, exec, s[4:5]
; GISEL-NEXT: .LBB3_13: ; %Flow4
; GISEL-NEXT: s_or_b64 exec, exec, s[8:9]
; GISEL-NEXT: v_mov_b32_e32 v0, 0x3ff00000
; GISEL-NEXT: v_lshl_add_u32 v0, v7, 20, v0
; GISEL-NEXT: v_mov_b32_e32 v1, 0xfffff
; GISEL-NEXT: v_and_or_b32 v5, v9, v1, v0
; GISEL-NEXT: .LBB3_14: ; %Flow5
; GISEL-NEXT: s_or_b64 exec, exec, s[6:7]
; GISEL-NEXT: v_mov_b32_e32 v0, v4
; GISEL-NEXT: v_mov_b32_e32 v1, v5
; GISEL-NEXT: s_setpc_b64 s[30:31]
%cvt = uitofp i128 %x to double
ret double %cvt
}
define half @sitofp_i128_to_f16(i128 %x) {
; SDAG-LABEL: sitofp_i128_to_f16:
; 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 .LBB4_14
; SDAG-NEXT: ; %bb.1: ; %itofp-if-end
; SDAG-NEXT: v_sub_co_u32_e32 v4, vcc, 0, v0
; SDAG-NEXT: v_subb_co_u32_e32 v5, vcc, 0, v1, vcc
; SDAG-NEXT: v_subb_co_u32_e32 v6, vcc, 0, v2, vcc
; SDAG-NEXT: v_subb_co_u32_e32 v7, vcc, 0, v3, vcc
; SDAG-NEXT: v_cmp_gt_i64_e32 vcc, 0, v[2:3]
; SDAG-NEXT: ; implicit-def: $vgpr8
; SDAG-NEXT: v_cndmask_b32_e32 v0, v0, v4, vcc
; SDAG-NEXT: v_cndmask_b32_e32 v4, v2, v6, vcc
; SDAG-NEXT: v_cndmask_b32_e32 v1, v1, v5, vcc
; SDAG-NEXT: v_cndmask_b32_e32 v5, v3, v7, 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: s_and_saveexec_b64 s[4:5], vcc
; SDAG-NEXT: s_xor_b64 s[4:5], exec, s[4:5]
; SDAG-NEXT: ; %bb.2: ; %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: ; %bb.3: ; %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 .LBB4_13
; SDAG-NEXT: ; %bb.4: ; %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 .LBB4_8
; SDAG-NEXT: ; %bb.5: ; %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 .LBB4_7
; SDAG-NEXT: ; %bb.6: ; %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: .LBB4_7: ; %Flow1
; SDAG-NEXT: s_or_b64 exec, exec, s[12:13]
; SDAG-NEXT: .LBB4_8: ; %Flow2
; SDAG-NEXT: s_andn2_saveexec_b64 s[4:5], s[10:11]
; SDAG-NEXT: ; %bb.9: ; %itofp-sw-bb
; SDAG-NEXT: v_lshlrev_b64 v[0:1], 1, v[0:1]
; SDAG-NEXT: ; %bb.10: ; %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: ; %bb.11: ; %itofp-if-then20
; SDAG-NEXT: v_alignbit_b32 v8, v1, v0, 3
; SDAG-NEXT: v_mov_b32_e32 v6, v2
; SDAG-NEXT: ; %bb.12: ; %Flow
; SDAG-NEXT: s_or_b64 exec, exec, s[4:5]
; SDAG-NEXT: .LBB4_13: ; %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_cvt_f16_f32_e32 v4, v0
; SDAG-NEXT: .LBB4_14: ; %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]
;
; GISEL-LABEL: sitofp_i128_to_f16:
; 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 .LBB4_14
; 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: ; %bb.2: ; %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: ; %bb.3: ; %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 .LBB4_13
; GISEL-NEXT: ; %bb.4: ; %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 .LBB4_8
; GISEL-NEXT: ; %bb.5: ; %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 .LBB4_7
; GISEL-NEXT: ; %bb.6: ; %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: .LBB4_7: ; %Flow1
; GISEL-NEXT: s_or_b64 exec, exec, s[12:13]
; GISEL-NEXT: .LBB4_8: ; %Flow2
; GISEL-NEXT: s_andn2_saveexec_b64 s[4:5], s[10:11]
; GISEL-NEXT: ; %bb.9: ; %itofp-sw-bb
; GISEL-NEXT: v_lshlrev_b64 v[0:1], 1, v[0:1]
; GISEL-NEXT: ; %bb.10: ; %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_addc_co_u32_e32 v1, vcc, 0, v1, vcc
; GISEL-NEXT: v_and_b32_e32 v2, 0x4000000, v0
; GISEL-NEXT: v_mov_b32_e32 v3, 0
; 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: ; %bb.11: ; %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: ; %bb.12: ; %Flow
; GISEL-NEXT: s_or_b64 exec, exec, s[4:5]
; GISEL-NEXT: .LBB4_13: ; %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_cvt_f16_f32_e32 v4, v0
; GISEL-NEXT: .LBB4_14: ; %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]
%cvt = sitofp i128 %x to half
ret half %cvt
}
define half @uitofp_i128_to_f16(i128 %x) {
; SDAG-LABEL: uitofp_i128_to_f16:
; 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 .LBB5_14
; 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: ; %bb.2: ; %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: $vgpr4
; SDAG-NEXT: ; implicit-def: $vgpr0_vgpr1
; SDAG-NEXT: ; implicit-def: $vgpr2_vgpr3
; SDAG-NEXT: ; %bb.3: ; %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 .LBB5_13
; SDAG-NEXT: ; %bb.4: ; %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 .LBB5_8
; SDAG-NEXT: ; %bb.5: ; %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 .LBB5_7
; SDAG-NEXT: ; %bb.6: ; %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: .LBB5_7: ; %Flow1
; SDAG-NEXT: s_or_b64 exec, exec, s[12:13]
; SDAG-NEXT: .LBB5_8: ; %Flow2
; SDAG-NEXT: s_andn2_saveexec_b64 s[4:5], s[10:11]
; SDAG-NEXT: ; %bb.9: ; %itofp-sw-bb
; SDAG-NEXT: v_lshlrev_b64 v[0:1], 1, v[0:1]
; SDAG-NEXT: ; %bb.10: ; %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: ; %bb.11: ; %itofp-if-then20
; SDAG-NEXT: v_alignbit_b32 v7, v1, v0, 3
; SDAG-NEXT: v_mov_b32_e32 v5, v4
; SDAG-NEXT: ; %bb.12: ; %Flow
; SDAG-NEXT: s_or_b64 exec, exec, s[4:5]
; SDAG-NEXT: .LBB5_13: ; %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_cvt_f16_f32_e32 v4, v0
; SDAG-NEXT: .LBB5_14: ; %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]
;
; GISEL-LABEL: uitofp_i128_to_f16:
; 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 .LBB5_14
; 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: ; %bb.2: ; %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
; GISEL-NEXT: ; implicit-def: $vgpr2
; GISEL-NEXT: ; %bb.3: ; %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 .LBB5_13
; GISEL-NEXT: ; %bb.4: ; %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 .LBB5_8
; GISEL-NEXT: ; %bb.5: ; %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 .LBB5_7
; GISEL-NEXT: ; %bb.6: ; %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: .LBB5_7: ; %Flow1
; GISEL-NEXT: s_or_b64 exec, exec, s[12:13]
; GISEL-NEXT: .LBB5_8: ; %Flow2
; GISEL-NEXT: s_andn2_saveexec_b64 s[4:5], s[10:11]
; GISEL-NEXT: ; %bb.9: ; %itofp-sw-bb
; GISEL-NEXT: v_lshlrev_b64 v[0:1], 1, v[0:1]
; GISEL-NEXT: ; %bb.10: ; %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_addc_co_u32_e32 v1, vcc, 0, v1, vcc
; GISEL-NEXT: v_and_b32_e32 v2, 0x4000000, v0
; GISEL-NEXT: v_mov_b32_e32 v3, 0
; 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: ; %bb.11: ; %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: ; %bb.12: ; %Flow
; GISEL-NEXT: s_or_b64 exec, exec, s[4:5]
; GISEL-NEXT: .LBB5_13: ; %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_cvt_f16_f32_e32 v4, v0
; GISEL-NEXT: .LBB5_14: ; %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]
%cvt = uitofp i128 %x to half
ret half %cvt
}
;; NOTE: These prefixes are unused and the list is autogenerated. Do not add tests below this line:
; GCN: {{.*}}