paweld/fpc
1
0
Fork 0
mirror of https://gitlab.com/freepascal.org/fpc/source.git synced 2025-10-31 02:31:49 +01:00
Code Issues Packages Projects Releases Wiki Activity

+ implemented other 32-bit rol/ror by const methods for i8086 and added a

Browse Source 
(hopefully) reasonable selection for them for all consts between 1 and 31

git-svn-id: trunk@35744 -
This commit is contained in:
nickysn 2017-04-06 15:41:39 +00:00
parent 0fd860d9e7
commit 1560f20e7f
2 changed files with 102 additions and 24 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

110
compiler/i8086/cgcpu.pas
View File

@ -583,11 +583,40 @@ unit cgcpu;
ror_amount:=a;
end;
case rol_amount of
1: rox32method:=rm_unrolledleftloop;
15: rox32method:=rm_unrolledrightloop;
16: rox32method:=rm_unrolledleftloop;
17: rox32method:=rm_unrolledrightloop;
31: rox32method:=rm_unrolledrightloop;
1,17:
rox32method:=rm_unrolledleftloop;
2,18:
if current_settings.cputype>=cpu_386 then
rox32method:=rm_fast_386
else if not (cs_opt_size in current_settings.optimizerswitches) then
rox32method:=rm_unrolledleftloop
else
rox32method:=rm_loopleft;
3..8,19..24:
if current_settings.cputype>=cpu_386 then
rox32method:=rm_fast_386
else
rox32method:=rm_loopleft;
15,31:
rox32method:=rm_unrolledrightloop;
14,30:
if current_settings.cputype>=cpu_386 then
rox32method:=rm_fast_386
else if not (cs_opt_size in current_settings.optimizerswitches) then
rox32method:=rm_unrolledrightloop
else
{ the left loop has a smaller size }
rox32method:=rm_loopleft;
9..13,25..29:
if current_settings.cputype>=cpu_386 then
rox32method:=rm_fast_386
else if not (cs_opt_size in current_settings.optimizerswitches) then
rox32method:=rm_loopright
else
{ the left loop has a smaller size }
rox32method:=rm_loopleft;
16:
rox32method:=rm_unrolledleftloop;
else
internalerror(2017040601);
end;
@ -622,9 +651,74 @@ unit cgcpu;
list.Concat(taicpu.op_const_reg(A_RCR,S_W,1,reg));
end;
end;
rm_loopleft,rm_loopright,rm_fast_386:
{todo: implement the other methods}
internalerror(2017040603);
rm_loopleft:
begin
if (rol_amount>=16) and not (cs_opt_size in current_settings.optimizerswitches) then
begin
list.Concat(taicpu.op_reg_reg(A_XCHG,S_W,reg,GetNextReg(reg)));
dec(rol_amount,16);
if rol_amount=0 then
exit;
end;
getcpuregister(list,NR_CX);
a_load_const_reg(list,OS_16,rol_amount,NR_CX);
current_asmdata.getjumplabel(hl_loop_start);
a_label(list,hl_loop_start);
list.Concat(taicpu.op_const_reg(A_SHL,S_W,1,GetNextReg(reg)));
list.Concat(taicpu.op_const_reg(A_RCL,S_W,1,reg));
list.Concat(taicpu.op_const_reg(A_ADC,S_W,0,GetNextReg(reg)));
ai:=Taicpu.Op_Sym(A_LOOP,S_W,hl_loop_start);
ai.is_jmp:=true;
list.concat(ai);
ungetcpuregister(list,NR_CX);
end;
rm_loopright:
begin
if (ror_amount>=16) and not (cs_opt_size in current_settings.optimizerswitches) then
begin
list.Concat(taicpu.op_reg_reg(A_XCHG,S_W,reg,GetNextReg(reg)));
dec(ror_amount,16);
if ror_amount=0 then
exit;
end;
getcpuregister(list,NR_CX);
a_load_const_reg(list,OS_16,ror_amount,NR_CX);
current_asmdata.getjumplabel(hl_loop_start);
a_label(list,hl_loop_start);
tmpreg:=getintregister(list,OS_16);
a_load_reg_reg(list,OS_16,OS_16,reg,tmpreg);
list.Concat(taicpu.op_const_reg(A_SHR,S_W,1,tmpreg));
list.Concat(taicpu.op_const_reg(A_RCR,S_W,1,GetNextReg(reg)));
list.Concat(taicpu.op_const_reg(A_RCR,S_W,1,reg));
ai:=Taicpu.Op_Sym(A_LOOP,S_W,hl_loop_start);
ai.is_jmp:=true;
list.concat(ai);
ungetcpuregister(list,NR_CX);
end;
rm_fast_386:
begin
tmpreg:=getintregister(list,OS_16);
if op=OP_ROL then
begin
list.Concat(taicpu.op_const_reg_reg(A_SHLD,S_W,rol_amount,reg,GetNextReg(reg)));
list.Concat(taicpu.op_const_reg_reg(A_SHLD,S_W,rol_amount,tmpreg,reg));
end
else
begin
list.Concat(taicpu.op_const_reg_reg(A_SHRD,S_W,ror_amount,reg,GetNextReg(reg)));
list.Concat(taicpu.op_const_reg_reg(A_SHRD,S_W,ror_amount,tmpreg,reg));
end;
end;
else
internalerror(2017040602);
end;

16
compiler/nadd.pas
View File

@ -1049,10 +1049,6 @@ implementation
begin
if (tordconstnode(tshlshrnode(left).right).value=
tshlshrnode(left).left.resultdef.size*8-tordconstnode(tshlshrnode(right).right).value)
{$ifdef i8086}
and (not(torddef(left.resultdef).ordtype in [s32bit,u32bit]) or
(tordconstnode(tshlshrnode(left).right).value.svalue in [1,15,16,17,31]))
{$endif i8086}
then
begin
result:=cinlinenode.create(in_ror_x_y,false,
@ -1064,10 +1060,6 @@ implementation
end
else if (tordconstnode(tshlshrnode(right).right).value=
tshlshrnode(left).left.resultdef.size*8-tordconstnode(tshlshrnode(left).right).value)
{$ifdef i8086}
and (not(torddef(left.resultdef).ordtype in [s32bit,u32bit]) or
(tordconstnode(tshlshrnode(right).right).value.svalue in [1,15,16,17,31]))
{$endif i8086}
then
begin
result:=cinlinenode.create(in_rol_x_y,false,
@ -1088,10 +1080,6 @@ implementation
begin
if (tordconstnode(tshlshrnode(left).right).value=
tshlshrnode(left).left.resultdef.size*8-tordconstnode(tshlshrnode(right).right).value)
{$ifdef i8086}
and (not(torddef(left.resultdef).ordtype in [s32bit,u32bit]) or
(tordconstnode(tshlshrnode(left).right).value.svalue in [1,15,16,17,31]))
{$endif i8086}
then
begin
result:=cinlinenode.create(in_rol_x_y,false,
@ -1103,10 +1091,6 @@ implementation
end
else if (tordconstnode(tshlshrnode(right).right).value=
tshlshrnode(left).left.resultdef.size*8-tordconstnode(tshlshrnode(left).right).value)
{$ifdef i8086}
and (not(torddef(left.resultdef).ordtype in [s32bit,u32bit]) or
(tordconstnode(tshlshrnode(right).right).value.svalue in [1,15,16,17,31]))
{$endif i8086}
then
begin
result:=cinlinenode.create(in_ror_x_y,false,