; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc -mtriple=amdgcn -verify-machineinstrs < %s | FileCheck --check-prefixes=GFX6 %s
; RUN: llc -mtriple=amdgcn -mcpu=tonga -verify-machineinstrs < %s | FileCheck --check-prefixes=GFX8 %s
; RUN: llc -mtriple=amdgcn -mcpu=gfx1100 -verify-machineinstrs < %s | FileCheck --check-prefixes=GFX11 %s
; FIXME: This should be merged with sint_to_fp.ll, but s_sint_to_fp_v2i64 crashes on r600
define amdgpu_kernel void @s_sint_to_fp_i64_to_f16(ptr addrspace(1) %out, i64 %in) #0 {
; GFX6-LABEL: s_sint_to_fp_i64_to_f16:
; GFX6: ; %bb.0:
; GFX6-NEXT: s_load_dwordx4 s[0:3], s[2:3], 0x9
; GFX6-NEXT: s_mov_b32 s7, 0xf000
; GFX6-NEXT: s_mov_b32 s6, -1
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
; GFX6-NEXT: s_mov_b32 s4, s0
; GFX6-NEXT: s_mov_b32 s5, s1
; GFX6-NEXT: s_flbit_i32 s0, s3
; GFX6-NEXT: s_xor_b32 s1, s2, s3
; GFX6-NEXT: s_add_i32 s0, s0, -1
; GFX6-NEXT: s_ashr_i32 s1, s1, 31
; GFX6-NEXT: s_add_i32 s1, s1, 32
; GFX6-NEXT: s_min_u32 s8, s0, s1
; GFX6-NEXT: s_lshl_b64 s[0:1], s[2:3], s8
; GFX6-NEXT: s_min_u32 s0, s0, 1
; GFX6-NEXT: s_or_b32 s0, s1, s0
; GFX6-NEXT: v_cvt_f32_i32_e32 v0, s0
; GFX6-NEXT: s_sub_i32 s0, 32, s8
; GFX6-NEXT: v_ldexp_f32_e64 v0, v0, s0
; GFX6-NEXT: v_cvt_f16_f32_e32 v0, v0
; GFX6-NEXT: buffer_store_short v0, off, s[4:7], 0
; GFX6-NEXT: s_endpgm
;
; GFX8-LABEL: s_sint_to_fp_i64_to_f16:
; GFX8: ; %bb.0:
; GFX8-NEXT: s_load_dwordx4 s[0:3], s[2:3], 0x24
; GFX8-NEXT: s_waitcnt lgkmcnt(0)
; GFX8-NEXT: s_xor_b32 s5, s2, s3
; GFX8-NEXT: s_flbit_i32 s4, s3
; GFX8-NEXT: s_ashr_i32 s5, s5, 31
; GFX8-NEXT: s_add_i32 s4, s4, -1
; GFX8-NEXT: s_add_i32 s5, s5, 32
; GFX8-NEXT: s_min_u32 s4, s4, s5
; GFX8-NEXT: s_lshl_b64 s[2:3], s[2:3], s4
; GFX8-NEXT: s_min_u32 s2, s2, 1
; GFX8-NEXT: s_or_b32 s2, s3, s2
; GFX8-NEXT: v_cvt_f32_i32_e32 v0, s2
; GFX8-NEXT: s_sub_i32 s2, 32, s4
; GFX8-NEXT: v_mov_b32_e32 v1, s1
; GFX8-NEXT: v_ldexp_f32 v0, v0, s2
; GFX8-NEXT: v_cvt_f16_f32_e32 v2, v0
; GFX8-NEXT: v_mov_b32_e32 v0, s0
; GFX8-NEXT: flat_store_short v[0:1], v2
; GFX8-NEXT: s_endpgm
;
; GFX11-LABEL: s_sint_to_fp_i64_to_f16:
; GFX11: ; %bb.0:
; GFX11-NEXT: s_load_b128 s[0:3], s[2:3], 0x24
; GFX11-NEXT: v_mov_b32_e32 v1, 0
; GFX11-NEXT: s_waitcnt lgkmcnt(0)
; GFX11-NEXT: s_xor_b32 s4, s2, s3
; GFX11-NEXT: s_cls_i32 s5, s3
; GFX11-NEXT: s_ashr_i32 s4, s4, 31
; GFX11-NEXT: s_add_i32 s5, s5, -1
; GFX11-NEXT: s_add_i32 s4, s4, 32
; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(SALU_CYCLE_1)
; GFX11-NEXT: s_min_u32 s4, s5, s4
; GFX11-NEXT: s_lshl_b64 s[2:3], s[2:3], s4
; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(SALU_CYCLE_1)
; GFX11-NEXT: s_min_u32 s2, s2, 1
; GFX11-NEXT: s_or_b32 s2, s3, s2
; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
; GFX11-NEXT: v_cvt_f32_i32_e32 v0, s2
; GFX11-NEXT: s_sub_i32 s2, 32, s4
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instid1(SALU_CYCLE_1)
; GFX11-NEXT: v_ldexp_f32 v0, v0, s2
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX11-NEXT: v_cvt_f16_f32_e32 v0, v0
; GFX11-NEXT: global_store_b16 v1, v0, s[0:1]
; GFX11-NEXT: s_nop 0
; GFX11-NEXT: s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
; GFX11-NEXT: s_endpgm
%result = sitofp i64 %in to half
store half %result, ptr addrspace(1) %out
ret void
}
define amdgpu_kernel void @v_sint_to_fp_i64_to_f16(ptr addrspace(1) %out, ptr addrspace(1) %in) #0 {
; GFX6-LABEL: v_sint_to_fp_i64_to_f16:
; GFX6: ; %bb.0:
; GFX6-NEXT: s_load_dwordx4 s[0:3], s[2:3], 0x9
; GFX6-NEXT: s_mov_b32 s7, 0xf000
; GFX6-NEXT: s_mov_b32 s6, 0
; GFX6-NEXT: v_lshlrev_b32_e32 v1, 3, v0
; GFX6-NEXT: v_mov_b32_e32 v2, 0
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
; GFX6-NEXT: s_mov_b64 s[4:5], s[2:3]
; GFX6-NEXT: buffer_load_dwordx2 v[3:4], v[1:2], s[4:7], 0 addr64
; GFX6-NEXT: v_lshlrev_b32_e32 v1, 1, v0
; GFX6-NEXT: s_mov_b64 s[2:3], s[6:7]
; GFX6-NEXT: s_waitcnt vmcnt(0)
; GFX6-NEXT: v_xor_b32_e32 v0, v3, v4
; GFX6-NEXT: v_ffbh_i32_e32 v5, v4
; GFX6-NEXT: v_ashrrev_i32_e32 v0, 31, v0
; GFX6-NEXT: v_add_i32_e32 v5, vcc, -1, v5
; GFX6-NEXT: v_add_i32_e32 v0, vcc, 32, v0
; GFX6-NEXT: v_min_u32_e32 v0, v5, v0
; GFX6-NEXT: v_lshl_b64 v[3:4], v[3:4], v0
; GFX6-NEXT: v_min_u32_e32 v3, 1, v3
; GFX6-NEXT: v_or_b32_e32 v3, v4, v3
; GFX6-NEXT: v_cvt_f32_i32_e32 v3, v3
; GFX6-NEXT: v_sub_i32_e32 v0, vcc, 32, v0
; GFX6-NEXT: v_ldexp_f32_e32 v0, v3, v0
; GFX6-NEXT: v_cvt_f16_f32_e32 v0, v0
; GFX6-NEXT: buffer_store_short v0, v[1:2], s[0:3], 0 addr64
; GFX6-NEXT: s_endpgm
;
; GFX8-LABEL: v_sint_to_fp_i64_to_f16:
; GFX8: ; %bb.0:
; GFX8-NEXT: s_load_dwordx4 s[0:3], s[2:3], 0x24
; GFX8-NEXT: v_lshlrev_b32_e32 v1, 3, v0
; GFX8-NEXT: v_lshlrev_b32_e32 v0, 1, v0
; GFX8-NEXT: s_waitcnt lgkmcnt(0)
; GFX8-NEXT: v_mov_b32_e32 v2, s3
; GFX8-NEXT: v_add_u32_e32 v1, vcc, s2, v1
; GFX8-NEXT: v_addc_u32_e32 v2, vcc, 0, v2, vcc
; GFX8-NEXT: flat_load_dwordx2 v[1:2], v[1:2]
; GFX8-NEXT: s_waitcnt vmcnt(0)
; GFX8-NEXT: v_xor_b32_e32 v3, v1, v2
; GFX8-NEXT: v_ffbh_i32_e32 v4, v2
; GFX8-NEXT: v_ashrrev_i32_e32 v3, 31, v3
; GFX8-NEXT: v_add_u32_e32 v4, vcc, -1, v4
; GFX8-NEXT: v_add_u32_e32 v3, vcc, 32, v3
; GFX8-NEXT: v_min_u32_e32 v3, v4, v3
; GFX8-NEXT: v_lshlrev_b64 v[1:2], v3, v[1:2]
; GFX8-NEXT: v_sub_u32_e32 v3, vcc, 32, v3
; GFX8-NEXT: v_min_u32_e32 v1, 1, v1
; GFX8-NEXT: v_or_b32_e32 v1, v2, v1
; GFX8-NEXT: v_cvt_f32_i32_e32 v1, v1
; GFX8-NEXT: v_mov_b32_e32 v2, s1
; GFX8-NEXT: v_add_u32_e32 v0, vcc, s0, v0
; GFX8-NEXT: v_ldexp_f32 v1, v1, v3
; GFX8-NEXT: v_cvt_f16_f32_e32 v3, v1
; GFX8-NEXT: v_addc_u32_e32 v1, vcc, 0, v2, vcc
; GFX8-NEXT: flat_store_short v[0:1], v3
; GFX8-NEXT: s_endpgm
;
; GFX11-LABEL: v_sint_to_fp_i64_to_f16:
; GFX11: ; %bb.0:
; GFX11-NEXT: s_load_b128 s[0:3], s[2:3], 0x24
; GFX11-NEXT: v_and_b32_e32 v2, 0x3ff, v0
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX11-NEXT: v_lshlrev_b32_e32 v0, 3, v2
; GFX11-NEXT: s_waitcnt lgkmcnt(0)
; GFX11-NEXT: global_load_b64 v[0:1], v0, s[2:3]
; GFX11-NEXT: s_waitcnt vmcnt(0)
; GFX11-NEXT: v_xor_b32_e32 v3, v0, v1
; GFX11-NEXT: v_cls_i32_e32 v4, v1
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX11-NEXT: v_ashrrev_i32_e32 v3, 31, v3
; GFX11-NEXT: v_add_nc_u32_e32 v4, -1, v4
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX11-NEXT: v_add_nc_u32_e32 v3, 32, v3
; GFX11-NEXT: v_min_u32_e32 v3, v4, v3
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX11-NEXT: v_lshlrev_b64 v[0:1], v3, v[0:1]
; GFX11-NEXT: v_min_u32_e32 v0, 1, v0
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2)
; GFX11-NEXT: v_or_b32_e32 v0, v1, v0
; GFX11-NEXT: v_sub_nc_u32_e32 v1, 32, v3
; GFX11-NEXT: v_cvt_f32_i32_e32 v0, v0
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2)
; GFX11-NEXT: v_ldexp_f32 v0, v0, v1
; GFX11-NEXT: v_lshlrev_b32_e32 v1, 1, v2
; GFX11-NEXT: v_cvt_f16_f32_e32 v0, v0
; GFX11-NEXT: global_store_b16 v1, v0, s[0:1]
; GFX11-NEXT: s_nop 0
; GFX11-NEXT: s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
; GFX11-NEXT: s_endpgm
%tid = call i32 @llvm.amdgcn.workitem.id.x()
%in.gep = getelementptr i64, ptr addrspace(1) %in, i32 %tid
%out.gep = getelementptr half, ptr addrspace(1) %out, i32 %tid
%val = load i64, ptr addrspace(1) %in.gep
%result = sitofp i64 %val to half
store half %result, ptr addrspace(1) %out.gep
ret void
}
define amdgpu_kernel void @s_sint_to_fp_i64_to_f32(ptr addrspace(1) %out, i64 %in) #0 {
; GFX6-LABEL: s_sint_to_fp_i64_to_f32:
; GFX6: ; %bb.0:
; GFX6-NEXT: s_load_dwordx4 s[0:3], s[2:3], 0x9
; GFX6-NEXT: s_mov_b32 s7, 0xf000
; GFX6-NEXT: s_mov_b32 s6, -1
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
; GFX6-NEXT: s_mov_b32 s4, s0
; GFX6-NEXT: s_mov_b32 s5, s1
; GFX6-NEXT: s_flbit_i32 s0, s3
; GFX6-NEXT: s_xor_b32 s1, s2, s3
; GFX6-NEXT: s_add_i32 s0, s0, -1
; GFX6-NEXT: s_ashr_i32 s1, s1, 31
; GFX6-NEXT: s_add_i32 s1, s1, 32
; GFX6-NEXT: s_min_u32 s8, s0, s1
; GFX6-NEXT: s_lshl_b64 s[0:1], s[2:3], s8
; GFX6-NEXT: s_min_u32 s0, s0, 1
; GFX6-NEXT: s_or_b32 s0, s1, s0
; GFX6-NEXT: v_cvt_f32_i32_e32 v0, s0
; GFX6-NEXT: s_sub_i32 s0, 32, s8
; GFX6-NEXT: v_ldexp_f32_e64 v0, v0, s0
; GFX6-NEXT: buffer_store_dword v0, off, s[4:7], 0
; GFX6-NEXT: s_endpgm
;
; GFX8-LABEL: s_sint_to_fp_i64_to_f32:
; GFX8: ; %bb.0:
; GFX8-NEXT: s_load_dwordx4 s[0:3], s[2:3], 0x24
; GFX8-NEXT: s_waitcnt lgkmcnt(0)
; GFX8-NEXT: s_xor_b32 s5, s2, s3
; GFX8-NEXT: s_flbit_i32 s4, s3
; GFX8-NEXT: s_ashr_i32 s5, s5, 31
; GFX8-NEXT: s_add_i32 s4, s4, -1
; GFX8-NEXT: s_add_i32 s5, s5, 32
; GFX8-NEXT: s_min_u32 s4, s4, s5
; GFX8-NEXT: s_lshl_b64 s[2:3], s[2:3], s4
; GFX8-NEXT: s_min_u32 s2, s2, 1
; GFX8-NEXT: s_or_b32 s2, s3, s2
; GFX8-NEXT: v_cvt_f32_i32_e32 v2, s2
; GFX8-NEXT: v_mov_b32_e32 v0, s0
; GFX8-NEXT: s_sub_i32 s0, 32, s4
; GFX8-NEXT: v_mov_b32_e32 v1, s1
; GFX8-NEXT: v_ldexp_f32 v2, v2, s0
; GFX8-NEXT: flat_store_dword v[0:1], v2
; GFX8-NEXT: s_endpgm
;
; GFX11-LABEL: s_sint_to_fp_i64_to_f32:
; GFX11: ; %bb.0:
; GFX11-NEXT: s_load_b128 s[0:3], s[2:3], 0x24
; GFX11-NEXT: v_mov_b32_e32 v1, 0
; GFX11-NEXT: s_waitcnt lgkmcnt(0)
; GFX11-NEXT: s_xor_b32 s4, s2, s3
; GFX11-NEXT: s_cls_i32 s5, s3
; GFX11-NEXT: s_ashr_i32 s4, s4, 31
; GFX11-NEXT: s_add_i32 s5, s5, -1
; GFX11-NEXT: s_add_i32 s4, s4, 32
; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(SALU_CYCLE_1)
; GFX11-NEXT: s_min_u32 s4, s5, s4
; GFX11-NEXT: s_lshl_b64 s[2:3], s[2:3], s4
; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(SALU_CYCLE_1)
; GFX11-NEXT: s_min_u32 s2, s2, 1
; GFX11-NEXT: s_or_b32 s2, s3, s2
; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
; GFX11-NEXT: v_cvt_f32_i32_e32 v0, s2
; GFX11-NEXT: s_sub_i32 s2, 32, s4
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instid1(SALU_CYCLE_1)
; GFX11-NEXT: v_ldexp_f32 v0, v0, s2
; GFX11-NEXT: global_store_b32 v1, v0, s[0:1]
; GFX11-NEXT: s_nop 0
; GFX11-NEXT: s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
; GFX11-NEXT: s_endpgm
%result = sitofp i64 %in to float
store float %result, ptr addrspace(1) %out
ret void
}
define amdgpu_kernel void @v_sint_to_fp_i64_to_f32(ptr addrspace(1) %out, ptr addrspace(1) %in) #0 {
; GFX6-LABEL: v_sint_to_fp_i64_to_f32:
; GFX6: ; %bb.0:
; GFX6-NEXT: s_load_dwordx4 s[0:3], s[2:3], 0x9
; GFX6-NEXT: s_mov_b32 s7, 0xf000
; GFX6-NEXT: s_mov_b32 s6, 0
; GFX6-NEXT: v_lshlrev_b32_e32 v1, 3, v0
; GFX6-NEXT: v_mov_b32_e32 v2, 0
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
; GFX6-NEXT: s_mov_b64 s[4:5], s[2:3]
; GFX6-NEXT: buffer_load_dwordx2 v[3:4], v[1:2], s[4:7], 0 addr64
; GFX6-NEXT: v_lshlrev_b32_e32 v1, 2, v0
; GFX6-NEXT: s_mov_b64 s[2:3], s[6:7]
; GFX6-NEXT: s_waitcnt vmcnt(0)
; GFX6-NEXT: v_xor_b32_e32 v0, v3, v4
; GFX6-NEXT: v_ffbh_i32_e32 v5, v4
; GFX6-NEXT: v_ashrrev_i32_e32 v0, 31, v0
; GFX6-NEXT: v_add_i32_e32 v5, vcc, -1, v5
; GFX6-NEXT: v_add_i32_e32 v0, vcc, 32, v0
; GFX6-NEXT: v_min_u32_e32 v0, v5, v0
; GFX6-NEXT: v_lshl_b64 v[3:4], v[3:4], v0
; GFX6-NEXT: v_min_u32_e32 v3, 1, v3
; GFX6-NEXT: v_or_b32_e32 v3, v4, v3
; GFX6-NEXT: v_cvt_f32_i32_e32 v3, v3
; GFX6-NEXT: v_sub_i32_e32 v0, vcc, 32, v0
; GFX6-NEXT: v_ldexp_f32_e32 v0, v3, v0
; GFX6-NEXT: buffer_store_dword v0, v[1:2], s[0:3], 0 addr64
; GFX6-NEXT: s_endpgm
;
; GFX8-LABEL: v_sint_to_fp_i64_to_f32:
; GFX8: ; %bb.0:
; GFX8-NEXT: s_load_dwordx4 s[0:3], s[2:3], 0x24
; GFX8-NEXT: v_lshlrev_b32_e32 v1, 3, v0
; GFX8-NEXT: v_lshlrev_b32_e32 v3, 2, v0
; GFX8-NEXT: s_waitcnt lgkmcnt(0)
; GFX8-NEXT: v_mov_b32_e32 v2, s3
; GFX8-NEXT: v_add_u32_e32 v1, vcc, s2, v1
; GFX8-NEXT: v_addc_u32_e32 v2, vcc, 0, v2, vcc
; GFX8-NEXT: flat_load_dwordx2 v[1:2], v[1:2]
; GFX8-NEXT: s_waitcnt vmcnt(0)
; GFX8-NEXT: v_xor_b32_e32 v0, v1, v2
; GFX8-NEXT: v_ffbh_i32_e32 v4, v2
; GFX8-NEXT: v_ashrrev_i32_e32 v0, 31, v0
; GFX8-NEXT: v_add_u32_e32 v4, vcc, -1, v4
; GFX8-NEXT: v_add_u32_e32 v0, vcc, 32, v0
; GFX8-NEXT: v_min_u32_e32 v4, v4, v0
; GFX8-NEXT: v_lshlrev_b64 v[0:1], v4, v[1:2]
; GFX8-NEXT: v_mov_b32_e32 v2, s1
; GFX8-NEXT: v_min_u32_e32 v0, 1, v0
; GFX8-NEXT: v_or_b32_e32 v0, v1, v0
; GFX8-NEXT: v_cvt_f32_i32_e32 v5, v0
; GFX8-NEXT: v_add_u32_e32 v0, vcc, s0, v3
; GFX8-NEXT: v_addc_u32_e32 v1, vcc, 0, v2, vcc
; GFX8-NEXT: v_sub_u32_e32 v2, vcc, 32, v4
; GFX8-NEXT: v_ldexp_f32 v2, v5, v2
; GFX8-NEXT: flat_store_dword v[0:1], v2
; GFX8-NEXT: s_endpgm
;
; GFX11-LABEL: v_sint_to_fp_i64_to_f32:
; GFX11: ; %bb.0:
; GFX11-NEXT: s_load_b128 s[0:3], s[2:3], 0x24
; GFX11-NEXT: v_and_b32_e32 v2, 0x3ff, v0
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX11-NEXT: v_lshlrev_b32_e32 v0, 3, v2
; GFX11-NEXT: v_lshlrev_b32_e32 v2, 2, v2
; GFX11-NEXT: s_waitcnt lgkmcnt(0)
; GFX11-NEXT: global_load_b64 v[0:1], v0, s[2:3]
; GFX11-NEXT: s_waitcnt vmcnt(0)
; GFX11-NEXT: v_xor_b32_e32 v3, v0, v1
; GFX11-NEXT: v_cls_i32_e32 v4, v1
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX11-NEXT: v_ashrrev_i32_e32 v3, 31, v3
; GFX11-NEXT: v_add_nc_u32_e32 v4, -1, v4
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX11-NEXT: v_add_nc_u32_e32 v3, 32, v3
; GFX11-NEXT: v_min_u32_e32 v3, v4, v3
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX11-NEXT: v_lshlrev_b64 v[0:1], v3, v[0:1]
; GFX11-NEXT: v_min_u32_e32 v0, 1, v0
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2)
; GFX11-NEXT: v_or_b32_e32 v0, v1, v0
; GFX11-NEXT: v_sub_nc_u32_e32 v1, 32, v3
; GFX11-NEXT: v_cvt_f32_i32_e32 v0, v0
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX11-NEXT: v_ldexp_f32 v0, v0, v1
; GFX11-NEXT: global_store_b32 v2, v0, s[0:1]
; GFX11-NEXT: s_nop 0
; GFX11-NEXT: s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
; GFX11-NEXT: s_endpgm
%tid = call i32 @llvm.amdgcn.workitem.id.x()
%in.gep = getelementptr i64, ptr addrspace(1) %in, i32 %tid
%out.gep = getelementptr float, ptr addrspace(1) %out, i32 %tid
%val = load i64, ptr addrspace(1) %in.gep
%result = sitofp i64 %val to float
store float %result, ptr addrspace(1) %out.gep
ret void
}
define amdgpu_kernel void @s_sint_to_fp_v2i64_to_v2f32(ptr addrspace(1) %out, <2 x i64> %in) #0{
; GFX6-LABEL: s_sint_to_fp_v2i64_to_v2f32:
; GFX6: ; %bb.0:
; GFX6-NEXT: s_load_dwordx4 s[4:7], s[2:3], 0xd
; GFX6-NEXT: s_load_dwordx2 s[0:1], s[2:3], 0x9
; GFX6-NEXT: s_mov_b32 s3, 0xf000
; GFX6-NEXT: s_mov_b32 s2, -1
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
; GFX6-NEXT: s_flbit_i32 s8, s7
; GFX6-NEXT: s_xor_b32 s9, s6, s7
; GFX6-NEXT: s_flbit_i32 s10, s5
; GFX6-NEXT: s_xor_b32 s11, s4, s5
; GFX6-NEXT: s_add_i32 s8, s8, -1
; GFX6-NEXT: s_ashr_i32 s9, s9, 31
; GFX6-NEXT: s_add_i32 s10, s10, -1
; GFX6-NEXT: s_ashr_i32 s11, s11, 31
; GFX6-NEXT: s_add_i32 s9, s9, 32
; GFX6-NEXT: s_add_i32 s11, s11, 32
; GFX6-NEXT: s_min_u32 s8, s8, s9
; GFX6-NEXT: s_min_u32 s9, s10, s11
; GFX6-NEXT: s_lshl_b64 s[6:7], s[6:7], s8
; GFX6-NEXT: s_sub_i32 s8, 32, s8
; GFX6-NEXT: s_lshl_b64 s[4:5], s[4:5], s9
; GFX6-NEXT: s_sub_i32 s9, 32, s9
; GFX6-NEXT: s_min_u32 s6, s6, 1
; GFX6-NEXT: s_min_u32 s4, s4, 1
; GFX6-NEXT: s_or_b32 s6, s7, s6
; GFX6-NEXT: s_or_b32 s4, s5, s4
; GFX6-NEXT: v_cvt_f32_i32_e32 v0, s6
; GFX6-NEXT: v_cvt_f32_i32_e32 v2, s4
; GFX6-NEXT: v_ldexp_f32_e64 v1, v0, s8
; GFX6-NEXT: v_ldexp_f32_e64 v0, v2, s9
; GFX6-NEXT: buffer_store_dwordx2 v[0:1], off, s[0:3], 0
; GFX6-NEXT: s_endpgm
;
; GFX8-LABEL: s_sint_to_fp_v2i64_to_v2f32:
; GFX8: ; %bb.0:
; GFX8-NEXT: s_load_dwordx4 s[4:7], s[2:3], 0x34
; GFX8-NEXT: s_load_dwordx2 s[0:1], s[2:3], 0x24
; GFX8-NEXT: s_waitcnt lgkmcnt(0)
; GFX8-NEXT: s_xor_b32 s3, s6, s7
; GFX8-NEXT: s_flbit_i32 s2, s7
; GFX8-NEXT: s_ashr_i32 s3, s3, 31
; GFX8-NEXT: s_add_i32 s2, s2, -1
; GFX8-NEXT: s_add_i32 s3, s3, 32
; GFX8-NEXT: s_min_u32 s9, s2, s3
; GFX8-NEXT: s_lshl_b64 s[2:3], s[6:7], s9
; GFX8-NEXT: s_min_u32 s2, s2, 1
; GFX8-NEXT: s_or_b32 s2, s3, s2
; GFX8-NEXT: v_cvt_f32_i32_e32 v0, s2
; GFX8-NEXT: s_xor_b32 s2, s4, s5
; GFX8-NEXT: s_flbit_i32 s8, s5
; GFX8-NEXT: s_ashr_i32 s2, s2, 31
; GFX8-NEXT: s_add_i32 s8, s8, -1
; GFX8-NEXT: s_add_i32 s2, s2, 32
; GFX8-NEXT: s_min_u32 s6, s8, s2
; GFX8-NEXT: s_lshl_b64 s[2:3], s[4:5], s6
; GFX8-NEXT: s_min_u32 s2, s2, 1
; GFX8-NEXT: s_or_b32 s2, s3, s2
; GFX8-NEXT: v_cvt_f32_i32_e32 v2, s2
; GFX8-NEXT: s_sub_i32 s2, 32, s9
; GFX8-NEXT: v_ldexp_f32 v1, v0, s2
; GFX8-NEXT: s_sub_i32 s2, 32, s6
; GFX8-NEXT: v_ldexp_f32 v0, v2, s2
; GFX8-NEXT: v_mov_b32_e32 v3, s1
; GFX8-NEXT: v_mov_b32_e32 v2, s0
; GFX8-NEXT: flat_store_dwordx2 v[2:3], v[0:1]
; GFX8-NEXT: s_endpgm
;
; GFX11-LABEL: s_sint_to_fp_v2i64_to_v2f32:
; GFX11: ; %bb.0:
; GFX11-NEXT: s_clause 0x1
; GFX11-NEXT: s_load_b128 s[4:7], s[2:3], 0x34
; GFX11-NEXT: s_load_b64 s[0:1], s[2:3], 0x24
; GFX11-NEXT: v_mov_b32_e32 v3, 0
; GFX11-NEXT: s_waitcnt lgkmcnt(0)
; GFX11-NEXT: s_xor_b32 s3, s6, s7
; GFX11-NEXT: s_xor_b32 s9, s4, s5
; GFX11-NEXT: s_cls_i32 s2, s7
; GFX11-NEXT: s_cls_i32 s8, s5
; GFX11-NEXT: s_ashr_i32 s3, s3, 31
; GFX11-NEXT: s_ashr_i32 s9, s9, 31
; GFX11-NEXT: s_add_i32 s2, s2, -1
; GFX11-NEXT: s_add_i32 s8, s8, -1
; GFX11-NEXT: s_add_i32 s3, s3, 32
; GFX11-NEXT: s_add_i32 s9, s9, 32
; GFX11-NEXT: s_min_u32 s10, s2, s3
; GFX11-NEXT: s_min_u32 s8, s8, s9
; GFX11-NEXT: s_lshl_b64 s[2:3], s[6:7], s10
; GFX11-NEXT: s_lshl_b64 s[4:5], s[4:5], s8
; GFX11-NEXT: s_min_u32 s2, s2, 1
; GFX11-NEXT: s_min_u32 s4, s4, 1
; GFX11-NEXT: s_or_b32 s2, s3, s2
; GFX11-NEXT: s_or_b32 s3, s5, s4
; GFX11-NEXT: v_cvt_f32_i32_e32 v0, s2
; GFX11-NEXT: v_cvt_f32_i32_e32 v2, s3
; GFX11-NEXT: s_sub_i32 s2, 32, s10
; GFX11-NEXT: s_sub_i32 s3, 32, s8
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX11-NEXT: v_ldexp_f32 v1, v0, s2
; GFX11-NEXT: v_ldexp_f32 v0, v2, s3
; GFX11-NEXT: global_store_b64 v3, v[0:1], s[0:1]
; GFX11-NEXT: s_nop 0
; GFX11-NEXT: s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
; GFX11-NEXT: s_endpgm
%result = sitofp <2 x i64> %in to <2 x float>
store <2 x float> %result, ptr addrspace(1) %out
ret void
}
define amdgpu_kernel void @v_sint_to_fp_v4i64_to_v4f32(ptr addrspace(1) %out, ptr addrspace(1) %in) #0 {
; GFX6-LABEL: v_sint_to_fp_v4i64_to_v4f32:
; GFX6: ; %bb.0:
; GFX6-NEXT: s_load_dwordx4 s[0:3], s[2:3], 0x9
; GFX6-NEXT: s_mov_b32 s7, 0xf000
; GFX6-NEXT: s_mov_b32 s6, 0
; GFX6-NEXT: v_lshlrev_b32_e32 v8, 5, v0
; GFX6-NEXT: v_mov_b32_e32 v9, 0
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
; GFX6-NEXT: s_mov_b64 s[4:5], s[2:3]
; GFX6-NEXT: buffer_load_dwordx4 v[1:4], v[8:9], s[4:7], 0 addr64 offset:16
; GFX6-NEXT: buffer_load_dwordx4 v[5:8], v[8:9], s[4:7], 0 addr64
; GFX6-NEXT: v_lshlrev_b32_e32 v10, 4, v0
; GFX6-NEXT: v_mov_b32_e32 v11, v9
; GFX6-NEXT: s_mov_b64 s[2:3], s[6:7]
; GFX6-NEXT: s_waitcnt vmcnt(1)
; GFX6-NEXT: v_xor_b32_e32 v0, v3, v4
; GFX6-NEXT: v_ffbh_i32_e32 v9, v4
; GFX6-NEXT: v_xor_b32_e32 v12, v1, v2
; GFX6-NEXT: v_ffbh_i32_e32 v13, v2
; GFX6-NEXT: s_waitcnt vmcnt(0)
; GFX6-NEXT: v_xor_b32_e32 v14, v7, v8
; GFX6-NEXT: v_ffbh_i32_e32 v15, v8
; GFX6-NEXT: v_xor_b32_e32 v16, v5, v6
; GFX6-NEXT: v_ffbh_i32_e32 v17, v6
; GFX6-NEXT: v_ashrrev_i32_e32 v0, 31, v0
; GFX6-NEXT: v_add_i32_e32 v9, vcc, -1, v9
; GFX6-NEXT: v_ashrrev_i32_e32 v12, 31, v12
; GFX6-NEXT: v_add_i32_e32 v13, vcc, -1, v13
; GFX6-NEXT: v_ashrrev_i32_e32 v14, 31, v14
; GFX6-NEXT: v_add_i32_e32 v15, vcc, -1, v15
; GFX6-NEXT: v_ashrrev_i32_e32 v16, 31, v16
; GFX6-NEXT: v_add_i32_e32 v17, vcc, -1, v17
; GFX6-NEXT: v_add_i32_e32 v0, vcc, 32, v0
; GFX6-NEXT: v_add_i32_e32 v12, vcc, 32, v12
; GFX6-NEXT: v_add_i32_e32 v14, vcc, 32, v14
; GFX6-NEXT: v_add_i32_e32 v16, vcc, 32, v16
; GFX6-NEXT: v_min_u32_e32 v0, v9, v0
; GFX6-NEXT: v_min_u32_e32 v9, v13, v12
; GFX6-NEXT: v_min_u32_e32 v12, v15, v14
; GFX6-NEXT: v_min_u32_e32 v13, v17, v16
; GFX6-NEXT: v_lshl_b64 v[3:4], v[3:4], v0
; GFX6-NEXT: v_sub_i32_e32 v14, vcc, 32, v0
; GFX6-NEXT: v_lshl_b64 v[0:1], v[1:2], v9
; GFX6-NEXT: v_sub_i32_e32 v2, vcc, 32, v9
; GFX6-NEXT: v_lshl_b64 v[7:8], v[7:8], v12
; GFX6-NEXT: v_sub_i32_e32 v9, vcc, 32, v12
; GFX6-NEXT: v_lshl_b64 v[5:6], v[5:6], v13
; GFX6-NEXT: v_sub_i32_e32 v12, vcc, 32, v13
; GFX6-NEXT: v_min_u32_e32 v3, 1, v3
; GFX6-NEXT: v_min_u32_e32 v0, 1, v0
; GFX6-NEXT: v_min_u32_e32 v7, 1, v7
; GFX6-NEXT: v_min_u32_e32 v5, 1, v5
; GFX6-NEXT: v_or_b32_e32 v3, v4, v3
; GFX6-NEXT: v_or_b32_e32 v0, v1, v0
; GFX6-NEXT: v_or_b32_e32 v1, v8, v7
; GFX6-NEXT: v_or_b32_e32 v4, v6, v5
; GFX6-NEXT: v_cvt_f32_i32_e32 v3, v3
; GFX6-NEXT: v_cvt_f32_i32_e32 v0, v0
; GFX6-NEXT: v_cvt_f32_i32_e32 v1, v1
; GFX6-NEXT: v_cvt_f32_i32_e32 v4, v4
; GFX6-NEXT: v_ldexp_f32_e32 v3, v3, v14
; GFX6-NEXT: v_ldexp_f32_e32 v2, v0, v2
; GFX6-NEXT: v_ldexp_f32_e32 v1, v1, v9
; GFX6-NEXT: v_ldexp_f32_e32 v0, v4, v12
; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[10:11], s[0:3], 0 addr64
; GFX6-NEXT: s_endpgm
;
; GFX8-LABEL: v_sint_to_fp_v4i64_to_v4f32:
; GFX8: ; %bb.0:
; GFX8-NEXT: s_load_dwordx4 s[0:3], s[2:3], 0x24
; GFX8-NEXT: v_lshlrev_b32_e32 v1, 5, v0
; GFX8-NEXT: v_lshlrev_b32_e32 v0, 4, v0
; GFX8-NEXT: s_waitcnt lgkmcnt(0)
; GFX8-NEXT: v_mov_b32_e32 v2, s3
; GFX8-NEXT: v_add_u32_e32 v5, vcc, s2, v1
; GFX8-NEXT: v_addc_u32_e32 v6, vcc, 0, v2, vcc
; GFX8-NEXT: flat_load_dwordx4 v[1:4], v[5:6]
; GFX8-NEXT: v_add_u32_e32 v5, vcc, 16, v5
; GFX8-NEXT: v_addc_u32_e32 v6, vcc, 0, v6, vcc
; GFX8-NEXT: flat_load_dwordx4 v[5:8], v[5:6]
; GFX8-NEXT: v_add_u32_e32 v9, vcc, s0, v0
; GFX8-NEXT: v_mov_b32_e32 v10, s1
; GFX8-NEXT: v_addc_u32_e32 v10, vcc, 0, v10, vcc
; GFX8-NEXT: s_waitcnt vmcnt(1)
; GFX8-NEXT: v_xor_b32_e32 v0, v3, v4
; GFX8-NEXT: v_xor_b32_e32 v12, v1, v2
; GFX8-NEXT: v_ffbh_i32_e32 v11, v4
; GFX8-NEXT: v_ffbh_i32_e32 v13, v2
; GFX8-NEXT: v_ashrrev_i32_e32 v0, 31, v0
; GFX8-NEXT: s_waitcnt vmcnt(0)
; GFX8-NEXT: v_xor_b32_e32 v14, v7, v8
; GFX8-NEXT: v_xor_b32_e32 v16, v5, v6
; GFX8-NEXT: v_ffbh_i32_e32 v15, v8
; GFX8-NEXT: v_ffbh_i32_e32 v17, v6
; GFX8-NEXT: v_ashrrev_i32_e32 v12, 31, v12
; GFX8-NEXT: v_ashrrev_i32_e32 v14, 31, v14
; GFX8-NEXT: v_ashrrev_i32_e32 v16, 31, v16
; GFX8-NEXT: v_add_u32_e32 v11, vcc, -1, v11
; GFX8-NEXT: v_add_u32_e32 v13, vcc, -1, v13
; GFX8-NEXT: v_add_u32_e32 v15, vcc, -1, v15
; GFX8-NEXT: v_add_u32_e32 v17, vcc, -1, v17
; GFX8-NEXT: v_add_u32_e32 v0, vcc, 32, v0
; GFX8-NEXT: v_add_u32_e32 v12, vcc, 32, v12
; GFX8-NEXT: v_add_u32_e32 v14, vcc, 32, v14
; GFX8-NEXT: v_add_u32_e32 v16, vcc, 32, v16
; GFX8-NEXT: v_min_u32_e32 v0, v11, v0
; GFX8-NEXT: v_min_u32_e32 v11, v13, v12
; GFX8-NEXT: v_min_u32_e32 v12, v15, v14
; GFX8-NEXT: v_min_u32_e32 v13, v17, v16
; GFX8-NEXT: v_lshlrev_b64 v[3:4], v0, v[3:4]
; GFX8-NEXT: v_sub_u32_e32 v14, vcc, 32, v0
; GFX8-NEXT: v_lshlrev_b64 v[0:1], v11, v[1:2]
; GFX8-NEXT: v_lshlrev_b64 v[7:8], v12, v[7:8]
; GFX8-NEXT: v_lshlrev_b64 v[5:6], v13, v[5:6]
; GFX8-NEXT: v_min_u32_e32 v3, 1, v3
; GFX8-NEXT: v_min_u32_e32 v0, 1, v0
; GFX8-NEXT: v_min_u32_e32 v7, 1, v7
; GFX8-NEXT: v_min_u32_e32 v5, 1, v5
; GFX8-NEXT: v_or_b32_e32 v3, v4, v3
; GFX8-NEXT: v_or_b32_e32 v0, v1, v0
; GFX8-NEXT: v_or_b32_e32 v1, v8, v7
; GFX8-NEXT: v_or_b32_e32 v4, v6, v5
; GFX8-NEXT: v_cvt_f32_i32_e32 v3, v3
; GFX8-NEXT: v_cvt_f32_i32_e32 v0, v0
; GFX8-NEXT: v_cvt_f32_i32_e32 v5, v1
; GFX8-NEXT: v_cvt_f32_i32_e32 v4, v4
; GFX8-NEXT: v_sub_u32_e32 v2, vcc, 32, v11
; GFX8-NEXT: v_sub_u32_e32 v11, vcc, 32, v12
; GFX8-NEXT: v_sub_u32_e32 v12, vcc, 32, v13
; GFX8-NEXT: v_ldexp_f32 v1, v3, v14
; GFX8-NEXT: v_ldexp_f32 v0, v0, v2
; GFX8-NEXT: v_ldexp_f32 v3, v5, v11
; GFX8-NEXT: v_ldexp_f32 v2, v4, v12
; GFX8-NEXT: flat_store_dwordx4 v[9:10], v[0:3]
; GFX8-NEXT: s_endpgm
;
; GFX11-LABEL: v_sint_to_fp_v4i64_to_v4f32:
; GFX11: ; %bb.0:
; GFX11-NEXT: s_load_b128 s[0:3], s[2:3], 0x24
; GFX11-NEXT: v_and_b32_e32 v8, 0x3ff, v0
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX11-NEXT: v_lshlrev_b32_e32 v4, 5, v8
; GFX11-NEXT: s_waitcnt lgkmcnt(0)
; GFX11-NEXT: s_clause 0x1
; GFX11-NEXT: global_load_b128 v[0:3], v4, s[2:3] offset:16
; GFX11-NEXT: global_load_b128 v[4:7], v4, s[2:3]
; GFX11-NEXT: s_waitcnt vmcnt(1)
; GFX11-NEXT: v_xor_b32_e32 v9, v2, v3
; GFX11-NEXT: v_xor_b32_e32 v11, v0, v1
; GFX11-NEXT: s_waitcnt vmcnt(0)
; GFX11-NEXT: v_xor_b32_e32 v13, v6, v7
; GFX11-NEXT: v_xor_b32_e32 v15, v4, v5
; GFX11-NEXT: v_cls_i32_e32 v10, v3
; GFX11-NEXT: v_cls_i32_e32 v12, v1
; GFX11-NEXT: v_cls_i32_e32 v14, v7
; GFX11-NEXT: v_cls_i32_e32 v16, v5
; GFX11-NEXT: v_ashrrev_i32_e32 v9, 31, v9
; GFX11-NEXT: v_ashrrev_i32_e32 v11, 31, v11
; GFX11-NEXT: v_ashrrev_i32_e32 v13, 31, v13
; GFX11-NEXT: v_ashrrev_i32_e32 v15, 31, v15
; GFX11-NEXT: v_add_nc_u32_e32 v10, -1, v10
; GFX11-NEXT: v_add_nc_u32_e32 v12, -1, v12
; GFX11-NEXT: v_add_nc_u32_e32 v14, -1, v14
; GFX11-NEXT: v_add_nc_u32_e32 v16, -1, v16
; GFX11-NEXT: v_add_nc_u32_e32 v9, 32, v9
; GFX11-NEXT: v_add_nc_u32_e32 v11, 32, v11
; GFX11-NEXT: v_add_nc_u32_e32 v13, 32, v13
; GFX11-NEXT: v_add_nc_u32_e32 v15, 32, v15
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
; GFX11-NEXT: v_min_u32_e32 v9, v10, v9
; GFX11-NEXT: v_min_u32_e32 v10, v12, v11
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
; GFX11-NEXT: v_min_u32_e32 v11, v14, v13
; GFX11-NEXT: v_min_u32_e32 v12, v16, v15
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
; GFX11-NEXT: v_lshlrev_b64 v[2:3], v9, v[2:3]
; GFX11-NEXT: v_lshlrev_b64 v[0:1], v10, v[0:1]
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
; GFX11-NEXT: v_lshlrev_b64 v[6:7], v11, v[6:7]
; GFX11-NEXT: v_lshlrev_b64 v[4:5], v12, v[4:5]
; GFX11-NEXT: v_sub_nc_u32_e32 v9, 32, v9
; GFX11-NEXT: v_sub_nc_u32_e32 v10, 32, v10
; GFX11-NEXT: v_min_u32_e32 v2, 1, v2
; GFX11-NEXT: v_min_u32_e32 v0, 1, v0
; GFX11-NEXT: v_min_u32_e32 v6, 1, v6
; GFX11-NEXT: v_min_u32_e32 v4, 1, v4
; GFX11-NEXT: v_sub_nc_u32_e32 v11, 32, v11
; GFX11-NEXT: v_or_b32_e32 v2, v3, v2
; GFX11-NEXT: v_or_b32_e32 v0, v1, v0
; GFX11-NEXT: v_or_b32_e32 v1, v7, v6
; GFX11-NEXT: v_or_b32_e32 v3, v5, v4
; GFX11-NEXT: v_sub_nc_u32_e32 v4, 32, v12
; GFX11-NEXT: v_cvt_f32_i32_e32 v2, v2
; GFX11-NEXT: v_cvt_f32_i32_e32 v0, v0
; GFX11-NEXT: v_cvt_f32_i32_e32 v1, v1
; GFX11-NEXT: v_cvt_f32_i32_e32 v5, v3
; GFX11-NEXT: v_lshlrev_b32_e32 v6, 4, v8
; GFX11-NEXT: v_ldexp_f32 v3, v2, v9
; GFX11-NEXT: v_ldexp_f32 v2, v0, v10
; GFX11-NEXT: v_ldexp_f32 v1, v1, v11
; GFX11-NEXT: v_ldexp_f32 v0, v5, v4
; GFX11-NEXT: global_store_b128 v6, v[0:3], s[0:1]
; GFX11-NEXT: s_nop 0
; GFX11-NEXT: s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
; GFX11-NEXT: s_endpgm
%tid = call i32 @llvm.amdgcn.workitem.id.x()
%in.gep = getelementptr <4 x i64>, ptr addrspace(1) %in, i32 %tid
%out.gep = getelementptr <4 x float>, ptr addrspace(1) %out, i32 %tid
%value = load <4 x i64>, ptr addrspace(1) %in.gep
%result = sitofp <4 x i64> %value to <4 x float>
store <4 x float> %result, ptr addrspace(1) %out.gep
ret void
}
define amdgpu_kernel void @s_sint_to_fp_v2i64_to_v2f16(ptr addrspace(1) %out, <2 x i64> %in) #0{
; GFX6-LABEL: s_sint_to_fp_v2i64_to_v2f16:
; GFX6: ; %bb.0:
; GFX6-NEXT: s_load_dwordx4 s[4:7], s[2:3], 0xd
; GFX6-NEXT: s_load_dwordx2 s[0:1], s[2:3], 0x9
; GFX6-NEXT: s_mov_b32 s3, 0xf000
; GFX6-NEXT: s_mov_b32 s2, -1
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
; GFX6-NEXT: s_flbit_i32 s8, s7
; GFX6-NEXT: s_xor_b32 s9, s6, s7
; GFX6-NEXT: s_flbit_i32 s10, s5
; GFX6-NEXT: s_xor_b32 s11, s4, s5
; GFX6-NEXT: s_add_i32 s8, s8, -1
; GFX6-NEXT: s_ashr_i32 s9, s9, 31
; GFX6-NEXT: s_add_i32 s10, s10, -1
; GFX6-NEXT: s_ashr_i32 s11, s11, 31
; GFX6-NEXT: s_add_i32 s9, s9, 32
; GFX6-NEXT: s_add_i32 s11, s11, 32
; GFX6-NEXT: s_min_u32 s8, s8, s9
; GFX6-NEXT: s_min_u32 s9, s10, s11
; GFX6-NEXT: s_lshl_b64 s[6:7], s[6:7], s8
; GFX6-NEXT: s_sub_i32 s8, 32, s8
; GFX6-NEXT: s_lshl_b64 s[4:5], s[4:5], s9
; GFX6-NEXT: s_sub_i32 s9, 32, s9
; GFX6-NEXT: s_min_u32 s6, s6, 1
; GFX6-NEXT: s_min_u32 s4, s4, 1
; GFX6-NEXT: s_or_b32 s6, s7, s6
; GFX6-NEXT: s_or_b32 s4, s5, s4
; GFX6-NEXT: v_cvt_f32_i32_e32 v0, s6
; GFX6-NEXT: v_cvt_f32_i32_e32 v1, s4
; GFX6-NEXT: v_ldexp_f32_e64 v0, v0, s8
; GFX6-NEXT: v_ldexp_f32_e64 v1, v1, s9
; GFX6-NEXT: v_cvt_f16_f32_e32 v0, v0
; GFX6-NEXT: v_lshlrev_b32_e32 v0, 16, v0
; GFX6-NEXT: v_cvt_f16_f32_e32 v1, v1
; GFX6-NEXT: v_or_b32_e32 v0, v1, v0
; GFX6-NEXT: buffer_store_dword v0, off, s[0:3], 0
; GFX6-NEXT: s_endpgm
;
; GFX8-LABEL: s_sint_to_fp_v2i64_to_v2f16:
; GFX8: ; %bb.0:
; GFX8-NEXT: s_load_dwordx4 s[4:7], s[2:3], 0x34
; GFX8-NEXT: s_load_dwordx2 s[0:1], s[2:3], 0x24
; GFX8-NEXT: s_waitcnt lgkmcnt(0)
; GFX8-NEXT: s_xor_b32 s3, s6, s7
; GFX8-NEXT: s_flbit_i32 s2, s7
; GFX8-NEXT: s_ashr_i32 s3, s3, 31
; GFX8-NEXT: s_add_i32 s2, s2, -1
; GFX8-NEXT: s_add_i32 s3, s3, 32
; GFX8-NEXT: s_min_u32 s8, s2, s3
; GFX8-NEXT: s_lshl_b64 s[2:3], s[6:7], s8
; GFX8-NEXT: s_min_u32 s2, s2, 1
; GFX8-NEXT: s_or_b32 s2, s3, s2
; GFX8-NEXT: s_xor_b32 s3, s4, s5
; GFX8-NEXT: v_cvt_f32_i32_e32 v0, s2
; GFX8-NEXT: s_flbit_i32 s2, s5
; GFX8-NEXT: s_ashr_i32 s3, s3, 31
; GFX8-NEXT: s_add_i32 s2, s2, -1
; GFX8-NEXT: s_add_i32 s3, s3, 32
; GFX8-NEXT: s_min_u32 s7, s2, s3
; GFX8-NEXT: s_lshl_b64 s[2:3], s[4:5], s7
; GFX8-NEXT: s_min_u32 s2, s2, 1
; GFX8-NEXT: s_or_b32 s2, s3, s2
; GFX8-NEXT: v_cvt_f32_i32_e32 v1, s2
; GFX8-NEXT: s_sub_i32 s6, 32, s8
; GFX8-NEXT: s_sub_i32 s2, 32, s7
; GFX8-NEXT: v_ldexp_f32 v0, v0, s6
; GFX8-NEXT: v_ldexp_f32 v1, v1, s2
; GFX8-NEXT: v_cvt_f16_f32_sdwa v0, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD
; GFX8-NEXT: v_cvt_f16_f32_e32 v1, v1
; GFX8-NEXT: v_or_b32_e32 v2, v1, v0
; GFX8-NEXT: v_mov_b32_e32 v0, s0
; GFX8-NEXT: v_mov_b32_e32 v1, s1
; GFX8-NEXT: flat_store_dword v[0:1], v2
; GFX8-NEXT: s_endpgm
;
; GFX11-LABEL: s_sint_to_fp_v2i64_to_v2f16:
; GFX11: ; %bb.0:
; GFX11-NEXT: s_clause 0x1
; GFX11-NEXT: s_load_b128 s[4:7], s[2:3], 0x34
; GFX11-NEXT: s_load_b64 s[0:1], s[2:3], 0x24
; GFX11-NEXT: v_mov_b32_e32 v2, 0
; GFX11-NEXT: s_waitcnt lgkmcnt(0)
; GFX11-NEXT: s_xor_b32 s3, s6, s7
; GFX11-NEXT: s_xor_b32 s9, s4, s5
; GFX11-NEXT: s_cls_i32 s2, s7
; GFX11-NEXT: s_cls_i32 s8, s5
; GFX11-NEXT: s_ashr_i32 s3, s3, 31
; GFX11-NEXT: s_ashr_i32 s9, s9, 31
; GFX11-NEXT: s_add_i32 s2, s2, -1
; GFX11-NEXT: s_add_i32 s8, s8, -1
; GFX11-NEXT: s_add_i32 s3, s3, 32
; GFX11-NEXT: s_add_i32 s9, s9, 32
; GFX11-NEXT: s_min_u32 s10, s2, s3
; GFX11-NEXT: s_min_u32 s8, s8, s9
; GFX11-NEXT: s_lshl_b64 s[2:3], s[6:7], s10
; GFX11-NEXT: s_lshl_b64 s[4:5], s[4:5], s8
; GFX11-NEXT: s_min_u32 s2, s2, 1
; GFX11-NEXT: s_min_u32 s4, s4, 1
; GFX11-NEXT: s_or_b32 s2, s3, s2
; GFX11-NEXT: s_or_b32 s3, s5, s4
; GFX11-NEXT: v_cvt_f32_i32_e32 v0, s2
; GFX11-NEXT: v_cvt_f32_i32_e32 v1, s3
; GFX11-NEXT: s_sub_i32 s2, 32, s10
; GFX11-NEXT: s_sub_i32 s3, 32, s8
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX11-NEXT: v_ldexp_f32 v0, v0, s2
; GFX11-NEXT: v_ldexp_f32 v1, v1, s3
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX11-NEXT: v_cvt_f16_f32_e32 v0, v0
; GFX11-NEXT: v_cvt_f16_f32_e32 v1, v1
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX11-NEXT: v_pack_b32_f16 v0, v1, v0
; GFX11-NEXT: global_store_b32 v2, v0, s[0:1]
; GFX11-NEXT: s_nop 0
; GFX11-NEXT: s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
; GFX11-NEXT: s_endpgm
%result = sitofp <2 x i64> %in to <2 x half>
store <2 x half> %result, ptr addrspace(1) %out
ret void
}
define amdgpu_kernel void @v_sint_to_fp_v4i64_to_v4f16(ptr addrspace(1) %out, ptr addrspace(1) %in) #0 {
; GFX6-LABEL: v_sint_to_fp_v4i64_to_v4f16:
; GFX6: ; %bb.0:
; GFX6-NEXT: s_load_dwordx4 s[0:3], s[2:3], 0x9
; GFX6-NEXT: s_mov_b32 s7, 0xf000
; GFX6-NEXT: s_mov_b32 s6, 0
; GFX6-NEXT: v_lshlrev_b32_e32 v8, 5, v0
; GFX6-NEXT: v_mov_b32_e32 v9, 0
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
; GFX6-NEXT: s_mov_b64 s[4:5], s[2:3]
; GFX6-NEXT: buffer_load_dwordx4 v[1:4], v[8:9], s[4:7], 0 addr64 offset:16
; GFX6-NEXT: buffer_load_dwordx4 v[5:8], v[8:9], s[4:7], 0 addr64
; GFX6-NEXT: v_lshlrev_b32_e32 v10, 3, v0
; GFX6-NEXT: v_mov_b32_e32 v11, v9
; GFX6-NEXT: s_mov_b64 s[2:3], s[6:7]
; GFX6-NEXT: s_waitcnt vmcnt(1)
; GFX6-NEXT: v_xor_b32_e32 v0, v3, v4
; GFX6-NEXT: v_ffbh_i32_e32 v9, v4
; GFX6-NEXT: v_xor_b32_e32 v12, v1, v2
; GFX6-NEXT: v_ffbh_i32_e32 v13, v2
; GFX6-NEXT: s_waitcnt vmcnt(0)
; GFX6-NEXT: v_xor_b32_e32 v14, v7, v8
; GFX6-NEXT: v_ffbh_i32_e32 v15, v8
; GFX6-NEXT: v_xor_b32_e32 v16, v5, v6
; GFX6-NEXT: v_ffbh_i32_e32 v17, v6
; GFX6-NEXT: v_ashrrev_i32_e32 v0, 31, v0
; GFX6-NEXT: v_add_i32_e32 v9, vcc, -1, v9
; GFX6-NEXT: v_ashrrev_i32_e32 v12, 31, v12
; GFX6-NEXT: v_add_i32_e32 v13, vcc, -1, v13
; GFX6-NEXT: v_ashrrev_i32_e32 v14, 31, v14
; GFX6-NEXT: v_add_i32_e32 v15, vcc, -1, v15
; GFX6-NEXT: v_ashrrev_i32_e32 v16, 31, v16
; GFX6-NEXT: v_add_i32_e32 v17, vcc, -1, v17
; GFX6-NEXT: v_add_i32_e32 v0, vcc, 32, v0
; GFX6-NEXT: v_add_i32_e32 v12, vcc, 32, v12
; GFX6-NEXT: v_add_i32_e32 v14, vcc, 32, v14
; GFX6-NEXT: v_add_i32_e32 v16, vcc, 32, v16
; GFX6-NEXT: v_min_u32_e32 v0, v9, v0
; GFX6-NEXT: v_min_u32_e32 v9, v13, v12
; GFX6-NEXT: v_min_u32_e32 v12, v15, v14
; GFX6-NEXT: v_min_u32_e32 v13, v17, v16
; GFX6-NEXT: v_lshl_b64 v[3:4], v[3:4], v0
; GFX6-NEXT: v_sub_i32_e32 v14, vcc, 32, v0
; GFX6-NEXT: v_lshl_b64 v[0:1], v[1:2], v9
; GFX6-NEXT: v_sub_i32_e32 v2, vcc, 32, v9
; GFX6-NEXT: v_lshl_b64 v[7:8], v[7:8], v12
; GFX6-NEXT: v_sub_i32_e32 v9, vcc, 32, v12
; GFX6-NEXT: v_lshl_b64 v[5:6], v[5:6], v13
; GFX6-NEXT: v_sub_i32_e32 v12, vcc, 32, v13
; GFX6-NEXT: v_min_u32_e32 v3, 1, v3
; GFX6-NEXT: v_min_u32_e32 v0, 1, v0
; GFX6-NEXT: v_min_u32_e32 v7, 1, v7
; GFX6-NEXT: v_min_u32_e32 v5, 1, v5
; GFX6-NEXT: v_or_b32_e32 v3, v4, v3
; GFX6-NEXT: v_or_b32_e32 v0, v1, v0
; GFX6-NEXT: v_or_b32_e32 v1, v8, v7
; GFX6-NEXT: v_or_b32_e32 v4, v6, v5
; GFX6-NEXT: v_cvt_f32_i32_e32 v3, v3
; GFX6-NEXT: v_cvt_f32_i32_e32 v0, v0
; GFX6-NEXT: v_cvt_f32_i32_e32 v1, v1
; GFX6-NEXT: v_cvt_f32_i32_e32 v4, v4
; GFX6-NEXT: v_ldexp_f32_e32 v3, v3, v14
; GFX6-NEXT: v_ldexp_f32_e32 v0, v0, v2
; GFX6-NEXT: v_ldexp_f32_e32 v1, v1, v9
; GFX6-NEXT: v_ldexp_f32_e32 v2, v4, v12
; GFX6-NEXT: v_cvt_f16_f32_e32 v3, v3
; GFX6-NEXT: v_cvt_f16_f32_e32 v0, v0
; GFX6-NEXT: v_cvt_f16_f32_e32 v1, v1
; GFX6-NEXT: v_cvt_f16_f32_e32 v2, v2
; GFX6-NEXT: v_lshlrev_b32_e32 v3, 16, v3
; GFX6-NEXT: v_lshlrev_b32_e32 v4, 16, v1
; GFX6-NEXT: v_or_b32_e32 v1, v0, v3
; GFX6-NEXT: v_or_b32_e32 v0, v2, v4
; GFX6-NEXT: buffer_store_dwordx2 v[0:1], v[10:11], s[0:3], 0 addr64
; GFX6-NEXT: s_endpgm
;
; GFX8-LABEL: v_sint_to_fp_v4i64_to_v4f16:
; GFX8: ; %bb.0:
; GFX8-NEXT: s_load_dwordx4 s[0:3], s[2:3], 0x24
; GFX8-NEXT: v_lshlrev_b32_e32 v1, 5, v0
; GFX8-NEXT: v_lshlrev_b32_e32 v9, 3, v0
; GFX8-NEXT: s_waitcnt lgkmcnt(0)
; GFX8-NEXT: v_mov_b32_e32 v2, s3
; GFX8-NEXT: v_add_u32_e32 v5, vcc, s2, v1
; GFX8-NEXT: v_addc_u32_e32 v6, vcc, 0, v2, vcc
; GFX8-NEXT: flat_load_dwordx4 v[1:4], v[5:6]
; GFX8-NEXT: v_add_u32_e32 v5, vcc, 16, v5
; GFX8-NEXT: v_addc_u32_e32 v6, vcc, 0, v6, vcc
; GFX8-NEXT: flat_load_dwordx4 v[5:8], v[5:6]
; GFX8-NEXT: v_mov_b32_e32 v10, s1
; GFX8-NEXT: s_waitcnt vmcnt(1)
; GFX8-NEXT: v_xor_b32_e32 v0, v3, v4
; GFX8-NEXT: v_xor_b32_e32 v12, v1, v2
; GFX8-NEXT: v_ffbh_i32_e32 v11, v4
; GFX8-NEXT: v_ffbh_i32_e32 v13, v2
; GFX8-NEXT: v_ashrrev_i32_e32 v0, 31, v0
; GFX8-NEXT: s_waitcnt vmcnt(0)
; GFX8-NEXT: v_xor_b32_e32 v14, v7, v8
; GFX8-NEXT: v_xor_b32_e32 v16, v5, v6
; GFX8-NEXT: v_ffbh_i32_e32 v15, v8
; GFX8-NEXT: v_ffbh_i32_e32 v17, v6
; GFX8-NEXT: v_ashrrev_i32_e32 v12, 31, v12
; GFX8-NEXT: v_ashrrev_i32_e32 v14, 31, v14
; GFX8-NEXT: v_ashrrev_i32_e32 v16, 31, v16
; GFX8-NEXT: v_add_u32_e32 v11, vcc, -1, v11
; GFX8-NEXT: v_add_u32_e32 v13, vcc, -1, v13
; GFX8-NEXT: v_add_u32_e32 v15, vcc, -1, v15
; GFX8-NEXT: v_add_u32_e32 v17, vcc, -1, v17
; GFX8-NEXT: v_add_u32_e32 v0, vcc, 32, v0
; GFX8-NEXT: v_add_u32_e32 v12, vcc, 32, v12
; GFX8-NEXT: v_add_u32_e32 v14, vcc, 32, v14
; GFX8-NEXT: v_add_u32_e32 v16, vcc, 32, v16
; GFX8-NEXT: v_min_u32_e32 v0, v11, v0
; GFX8-NEXT: v_min_u32_e32 v11, v13, v12
; GFX8-NEXT: v_min_u32_e32 v12, v15, v14
; GFX8-NEXT: v_min_u32_e32 v13, v17, v16
; GFX8-NEXT: v_lshlrev_b64 v[3:4], v0, v[3:4]
; GFX8-NEXT: v_sub_u32_e32 v14, vcc, 32, v0
; GFX8-NEXT: v_lshlrev_b64 v[0:1], v11, v[1:2]
; GFX8-NEXT: v_lshlrev_b64 v[7:8], v12, v[7:8]
; GFX8-NEXT: v_lshlrev_b64 v[5:6], v13, v[5:6]
; GFX8-NEXT: v_min_u32_e32 v3, 1, v3
; GFX8-NEXT: v_min_u32_e32 v0, 1, v0
; GFX8-NEXT: v_min_u32_e32 v7, 1, v7
; GFX8-NEXT: v_min_u32_e32 v5, 1, v5
; GFX8-NEXT: v_or_b32_e32 v3, v4, v3
; GFX8-NEXT: v_or_b32_e32 v0, v1, v0
; GFX8-NEXT: v_or_b32_e32 v1, v8, v7
; GFX8-NEXT: v_or_b32_e32 v4, v6, v5
; GFX8-NEXT: v_cvt_f32_i32_e32 v3, v3
; GFX8-NEXT: v_cvt_f32_i32_e32 v0, v0
; GFX8-NEXT: v_cvt_f32_i32_e32 v1, v1
; GFX8-NEXT: v_cvt_f32_i32_e32 v4, v4
; GFX8-NEXT: v_sub_u32_e32 v2, vcc, 32, v11
; GFX8-NEXT: v_sub_u32_e32 v11, vcc, 32, v12
; GFX8-NEXT: v_sub_u32_e32 v12, vcc, 32, v13
; GFX8-NEXT: v_ldexp_f32 v3, v3, v14
; GFX8-NEXT: v_ldexp_f32 v0, v0, v2
; GFX8-NEXT: v_ldexp_f32 v1, v1, v11
; GFX8-NEXT: v_ldexp_f32 v2, v4, v12
; GFX8-NEXT: v_cvt_f16_f32_sdwa v3, v3 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD
; GFX8-NEXT: v_cvt_f16_f32_e32 v4, v0
; GFX8-NEXT: v_cvt_f16_f32_sdwa v5, v1 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD
; GFX8-NEXT: v_cvt_f16_f32_e32 v6, v2
; GFX8-NEXT: v_add_u32_e32 v0, vcc, s0, v9
; GFX8-NEXT: v_addc_u32_e32 v1, vcc, 0, v10, vcc
; GFX8-NEXT: v_or_b32_e32 v2, v4, v3
; GFX8-NEXT: v_or_b32_e32 v3, v6, v5
; GFX8-NEXT: flat_store_dwordx2 v[0:1], v[2:3]
; GFX8-NEXT: s_endpgm
;
; GFX11-LABEL: v_sint_to_fp_v4i64_to_v4f16:
; GFX11: ; %bb.0:
; GFX11-NEXT: s_load_b128 s[0:3], s[2:3], 0x24
; GFX11-NEXT: v_and_b32_e32 v8, 0x3ff, v0
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX11-NEXT: v_lshlrev_b32_e32 v4, 5, v8
; GFX11-NEXT: s_waitcnt lgkmcnt(0)
; GFX11-NEXT: s_clause 0x1
; GFX11-NEXT: global_load_b128 v[0:3], v4, s[2:3] offset:16
; GFX11-NEXT: global_load_b128 v[4:7], v4, s[2:3]
; GFX11-NEXT: s_waitcnt vmcnt(1)
; GFX11-NEXT: v_xor_b32_e32 v9, v2, v3
; GFX11-NEXT: v_xor_b32_e32 v11, v0, v1
; GFX11-NEXT: s_waitcnt vmcnt(0)
; GFX11-NEXT: v_xor_b32_e32 v13, v6, v7
; GFX11-NEXT: v_xor_b32_e32 v15, v4, v5
; GFX11-NEXT: v_cls_i32_e32 v10, v3
; GFX11-NEXT: v_cls_i32_e32 v12, v1
; GFX11-NEXT: v_cls_i32_e32 v14, v7
; GFX11-NEXT: v_cls_i32_e32 v16, v5
; GFX11-NEXT: v_ashrrev_i32_e32 v9, 31, v9
; GFX11-NEXT: v_ashrrev_i32_e32 v11, 31, v11
; GFX11-NEXT: v_ashrrev_i32_e32 v13, 31, v13
; GFX11-NEXT: v_ashrrev_i32_e32 v15, 31, v15
; GFX11-NEXT: v_add_nc_u32_e32 v10, -1, v10
; GFX11-NEXT: v_add_nc_u32_e32 v12, -1, v12
; GFX11-NEXT: v_add_nc_u32_e32 v14, -1, v14
; GFX11-NEXT: v_add_nc_u32_e32 v16, -1, v16
; GFX11-NEXT: v_add_nc_u32_e32 v9, 32, v9
; GFX11-NEXT: v_add_nc_u32_e32 v11, 32, v11
; GFX11-NEXT: v_add_nc_u32_e32 v13, 32, v13
; GFX11-NEXT: v_add_nc_u32_e32 v15, 32, v15
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
; GFX11-NEXT: v_min_u32_e32 v9, v10, v9
; GFX11-NEXT: v_min_u32_e32 v10, v12, v11
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
; GFX11-NEXT: v_min_u32_e32 v11, v14, v13
; GFX11-NEXT: v_min_u32_e32 v12, v16, v15
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
; GFX11-NEXT: v_lshlrev_b64 v[2:3], v9, v[2:3]
; GFX11-NEXT: v_lshlrev_b64 v[0:1], v10, v[0:1]
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
; GFX11-NEXT: v_lshlrev_b64 v[6:7], v11, v[6:7]
; GFX11-NEXT: v_lshlrev_b64 v[4:5], v12, v[4:5]
; GFX11-NEXT: v_sub_nc_u32_e32 v9, 32, v9
; GFX11-NEXT: v_sub_nc_u32_e32 v10, 32, v10
; GFX11-NEXT: v_min_u32_e32 v2, 1, v2
; GFX11-NEXT: v_min_u32_e32 v0, 1, v0
; GFX11-NEXT: v_min_u32_e32 v6, 1, v6
; GFX11-NEXT: v_min_u32_e32 v4, 1, v4
; GFX11-NEXT: v_sub_nc_u32_e32 v11, 32, v11
; GFX11-NEXT: v_or_b32_e32 v2, v3, v2
; GFX11-NEXT: v_or_b32_e32 v0, v1, v0
; GFX11-NEXT: v_or_b32_e32 v1, v7, v6
; GFX11-NEXT: v_or_b32_e32 v3, v5, v4
; GFX11-NEXT: v_sub_nc_u32_e32 v4, 32, v12
; GFX11-NEXT: v_cvt_f32_i32_e32 v2, v2
; GFX11-NEXT: v_cvt_f32_i32_e32 v0, v0
; GFX11-NEXT: v_cvt_f32_i32_e32 v1, v1
; GFX11-NEXT: v_cvt_f32_i32_e32 v3, v3
; GFX11-NEXT: v_lshlrev_b32_e32 v5, 3, v8
; GFX11-NEXT: v_ldexp_f32 v2, v2, v9
; GFX11-NEXT: v_ldexp_f32 v0, v0, v10
; GFX11-NEXT: v_ldexp_f32 v1, v1, v11
; GFX11-NEXT: v_ldexp_f32 v3, v3, v4
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
; GFX11-NEXT: v_cvt_f16_f32_e32 v2, v2
; GFX11-NEXT: v_cvt_f16_f32_e32 v0, v0
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
; GFX11-NEXT: v_cvt_f16_f32_e32 v4, v1
; GFX11-NEXT: v_cvt_f16_f32_e32 v3, v3
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX11-NEXT: v_pack_b32_f16 v1, v0, v2
; GFX11-NEXT: v_pack_b32_f16 v0, v3, v4
; GFX11-NEXT: global_store_b64 v5, v[0:1], s[0:1]
; GFX11-NEXT: s_nop 0
; GFX11-NEXT: s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
; GFX11-NEXT: s_endpgm
%tid = call i32 @llvm.amdgcn.workitem.id.x()
%in.gep = getelementptr <4 x i64>, ptr addrspace(1) %in, i32 %tid
%out.gep = getelementptr <4 x half>, ptr addrspace(1) %out, i32 %tid
%value = load <4 x i64>, ptr addrspace(1) %in.gep
%result = sitofp <4 x i64> %value to <4 x half>
store <4 x half> %result, ptr addrspace(1) %out.gep
ret void
}
declare i32 @llvm.amdgcn.workitem.id.x() #1
attributes #0 = { nounwind }
attributes #1 = { nounwind readnone }
Codebase Browser
llvm