paweld/fpc
1
0
Fork 0
mirror of https://gitlab.com/freepascal.org/fpc/source.git synced 2025-08-20 12:49:08 +02:00
Code Issues Packages Projects Releases Wiki Activity

* cleanup

Browse Source
This commit is contained in:
florian 2024-10-26 20:32:07 +02:00
parent b416418134
commit 54dcfa78f8
2 changed files with 0 additions and 58 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

33
rtl/i386/set.inc
View File

@ -415,32 +415,6 @@ asm
sub $16, %ecx sub $16, %ecx
jl .LFallback { probably dead branch... } jl .LFallback { probably dead branch... }
{$if false}
{ Scans 16 bytes at a time left to right with early exits.
Would be better for large enough sets (maybe around 64 bytes or even more) — if they existed, but worse for actually existing 32.
Kept for the future. }
push %ebx
pxor %xmm2, %xmm2 { xmm2 = 0 }
add %ecx, %eax { set1 += size }
add %ecx, %edx { set2 += size }
neg %ecx { Now "size" (ecx) = -(orig.size - 16), "set1" (eax) points to orig.set1 + orig.size - 16, "set2" (edx) points to orig.set2 + orig.size - 16.
Loop ends on "size" >= 0, leaving up to 16 tail bytes. }
.L16x_Loop:
movdqu (%eax,%ecx), %xmm1
movdqu (%edx,%ecx), %xmm0
pandn %xmm1, %xmm0
pcmpeqb %xmm2, %xmm0
pmovmskb %xmm0, %ebx
inc %bx
jnz .LNo
add $16, %ecx
js .L16x_Loop
pop %ebx
movdqu (%eax), %xmm1
movdqu (%edx), %xmm0
pandn %xmm1, %xmm0
{$else}
{ Folds all 16-byte "set1 and not set2" chunks with OR and checks the final result for zero. Better for small enough sets. } { Folds all 16-byte "set1 and not set2" chunks with OR and checks the final result for zero. Better for small enough sets. }
movdqu (%eax), %xmm1 movdqu (%eax), %xmm1
movdqu (%edx), %xmm2 movdqu (%edx), %xmm2
@ -455,7 +429,6 @@ asm
ja .L16x_Loop ja .L16x_Loop
pxor %xmm0, %xmm0 pxor %xmm0, %xmm0
{$endif}
pcmpeqb %xmm2,%xmm0 pcmpeqb %xmm2,%xmm0
pmovmskb %xmm0, %ecx pmovmskb %xmm0, %ecx
xor %eax, %eax xor %eax, %eax
@ -466,12 +439,6 @@ asm
.LFallback: .LFallback:
add $16, %ecx add $16, %ecx
jmp fpc_varset_contains_sets_plain jmp fpc_varset_contains_sets_plain
{$if false}
.LNo:
xor %eax, %eax
pop %ebx
{$endif}
end; end;
function fpc_varset_contains_sets_dispatch(const set1,set2;size : ptrint):boolean; forward; function fpc_varset_contains_sets_dispatch(const set1,set2;size : ptrint):boolean; forward;

25
rtl/x86_64/set.inc
View File

@ -163,30 +163,6 @@ asm
sub size, 16 sub size, 16
jl @Bytewise_Prepare { probably dead branch... } jl @Bytewise_Prepare { probably dead branch... }
{$if false}
{ Scans 16 bytes at a time left to right with early exits.
Would be better for large enough sets (maybe around 64 bytes or even more) — if they existed, but worse for actually existing 32.
Kept for the future. }
pxor xmm2, xmm2 { xmm2 = 0 }
add set1, size
add set2, size
neg size { Now "size" = -(orig.size - 16), "set1" points to orig.set1 + orig.size - 16, "set2" points to orig.set2 + orig.size - 16.
Loop ends on "size" >= 0, leaving up to 16 tail bytes. }
@16x_Loop:
movdqu xmm1, xmmword ptr [set1 + size]
movdqu xmm0, xmmword ptr [set2 + size]
pandn xmm0, xmm1
pcmpeqb xmm0, xmm2
pmovmskb eax, xmm0
inc ax
jnz @No
add size, 16
js @16x_Loop
movdqu xmm1, xmmword ptr [set1]
movdqu xmm0, xmmword ptr [set2]
pandn xmm0, xmm1
{$else}
{ Folds all 16-byte "set1 and not set2" chunks with OR and checks the final result for zero. Better for small enough sets. } { Folds all 16-byte "set1 and not set2" chunks with OR and checks the final result for zero. Better for small enough sets. }
movdqu xmm1, xmmword ptr [set1] movdqu xmm1, xmmword ptr [set1]
movdqu xmm2, xmmword ptr [set2] movdqu xmm2, xmmword ptr [set2]
@ -201,7 +177,6 @@ asm
ja @16x_Loop ja @16x_Loop
pxor xmm0, xmm0 pxor xmm0, xmm0
{$endif}
pcmpeqb xmm0, xmm2 pcmpeqb xmm0, xmm2
pmovmskb ecx, xmm0 pmovmskb ecx, xmm0
xor eax, eax xor eax, eax