paweld/fpc
1
0
Fork 0
mirror of https://gitlab.com/freepascal.org/fpc/source.git synced 2025-08-05 13:46:11 +02:00
Code Issues Packages Projects Releases Wiki Activity

+ make use of mulx instruction, if available

Browse Source 
git-svn-id: trunk@43706 -
This commit is contained in:
florian 2019-12-21 16:20:10 +00:00
parent 30927039a6
commit 8465710329
1 changed files with 48 additions and 19 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

67
compiler/i386/n386add.pas
View File

@ -50,6 +50,7 @@ interface
aasmbase,aasmtai,aasmdata,aasmcpu, aasmbase,aasmtai,aasmdata,aasmcpu,
cgbase,procinfo, cgbase,procinfo,
ncon,nset,cgutils,tgobj, ncon,nset,cgutils,tgobj,
cpuinfo,
cga,ncgutil,cgobj,cg64f32,cgx86, cga,ncgutil,cgobj,cg64f32,cgx86,
hlcgobj; hlcgobj;
@ -446,7 +447,7 @@ interface
procedure ti386addnode.second_mul(unsigned: boolean); procedure ti386addnode.second_mul(unsigned: boolean);
var reg:Tregister; var reg,reghi,reglo:Tregister;
ref:Treference; ref:Treference;
use_ref:boolean; use_ref:boolean;
hl4 : tasmlabel; hl4 : tasmlabel;
@ -473,30 +474,58 @@ interface
end end
else else
begin begin
{LOC_CONSTANT for example.} { LOC_CONSTANT for example.}
reg:=cg.getintregister(current_asmdata.CurrAsmList,OS_INT); reg:=cg.getintregister(current_asmdata.CurrAsmList,OS_INT);
hlcg.a_load_loc_reg(current_asmdata.CurrAsmList,left.resultdef,osuinttype,left.location,reg); hlcg.a_load_loc_reg(current_asmdata.CurrAsmList,left.resultdef,osuinttype,left.location,reg);
end; end;
{Allocate EAX.}
cg.getcpuregister(current_asmdata.CurrAsmList,NR_EAX); if (CPUX86_HAS_BMI2 in cpu_capabilities[current_settings.cputype]) and
{Load the right value.} (not(needoverflowcheck) or
hlcg.a_load_loc_reg(current_asmdata.CurrAsmList,right.resultdef,osuinttype,right.location,NR_EAX);
{Also allocate EDX, since it is also modified by a mul (JM).}
cg.getcpuregister(current_asmdata.CurrAsmList,NR_EDX);
if use_ref then
emit_ref(asmops[unsigned],S_L,ref)
else
emit_reg(asmops[unsigned],S_L,reg);
if needoverflowcheck and
{ 32->64 bit cannot overflow } { 32->64 bit cannot overflow }
(not is_64bit(resultdef)) then is_64bit(resultdef)) then
begin begin
current_asmdata.getjumplabel(hl4); cg.getcpuregister(current_asmdata.CurrAsmList,NR_EDX);
cg.a_jmp_flags(current_asmdata.CurrAsmList,F_AE,hl4); hlcg.a_load_loc_reg(current_asmdata.CurrAsmList,right.resultdef,osuinttype,right.location,NR_EDX);
cg.a_call_name(current_asmdata.CurrAsmList,'FPC_OVERFLOW',false); cg.ungetcpuregister(current_asmdata.CurrAsmList,NR_EDX);
cg.a_label(current_asmdata.CurrAsmList,hl4); reglo:=cg.getintregister(current_asmdata.CurrAsmList,OS_INT);
reghi:=cg.getintregister(current_asmdata.CurrAsmList,OS_INT);
if use_ref then
current_asmdata.CurrAsmList.concat(Taicpu.Op_ref_reg_reg(A_MULX,S_L,ref,reglo,reghi))
else
emit_reg_reg_reg(A_MULX,S_L,reg,reglo,reghi);
location_reset(location,LOC_REGISTER,def_cgsize(resultdef));
location.register64.reglo:=reglo;
if is_64bit(resultdef) then
location.register64.reghi:=reghi;
location_freetemp(current_asmdata.CurrAsmList,left.location);
location_freetemp(current_asmdata.CurrAsmList,right.location);
end
else
begin
{ Allocate EAX. }
cg.getcpuregister(current_asmdata.CurrAsmList,NR_EAX);
{ Load the right value. }
hlcg.a_load_loc_reg(current_asmdata.CurrAsmList,right.resultdef,osuinttype,right.location,NR_EAX);
{ Also allocate EDX, since it is also modified by a mul (JM). }
cg.getcpuregister(current_asmdata.CurrAsmList,NR_EDX);
if use_ref then
emit_ref(asmops[unsigned],S_L,ref)
else
emit_reg(asmops[unsigned],S_L,reg);
if needoverflowcheck and
{ 32->64 bit cannot overflow }
(not is_64bit(resultdef)) then
begin
current_asmdata.getjumplabel(hl4);
cg.a_jmp_flags(current_asmdata.CurrAsmList,F_AE,hl4);
cg.a_call_name(current_asmdata.CurrAsmList,'FPC_OVERFLOW',false);
cg.a_label(current_asmdata.CurrAsmList,hl4);
end;
set_mul_result_location;
end; end;
set_mul_result_location;
end; end;