Use memset in new_heap_segment; remove home-made zero_dnodes

It turns out that memset is faster.
Clozure · Jul 12, 2024 · 678d152 · 678d152
1 parent f7ac284
commit 678d152
Show file tree

Hide file tree

Showing 6 changed files with 2 additions and 65 deletions.
diff --git a/lisp-kernel/arm-asmutils.s b/lisp-kernel/arm-asmutils.s
@@ -252,27 +252,6 @@ _exportfn(call_handler_on_main_stack)
         __(bx ip)
 _endfn                
 
-/* zero N (r1) dnodes, starting at the dnode-aligned address in r0 */
-_exportfn(C(zero_dnodes))
-        __(adr r2,9f)
-        __(fldmiad r2,{d0-d7})
-        __(ands r2,r1,#7)
-        __(sub r1,r1,r2)
-        __(b 1f)
-0:      __(subs r2,r2,#1)        
-        __(fstd d0,[r0])
-        __(add r0,r0,#dnode_size)
-1:      __(bne 0b)
-        __(cmp r1,#0)
-        __(b 3f)
-2:      __(subs r1,r1,#8)
-        __(fstmiad r0!,{d0-d7})
-3:      __(bne 2b)
-        __(bx lr)
-        .align 3
-9:      .long 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0
-_endfn             
-
 _exportfn(C(dmb))
         __(dmb)
         __(bx lr)

diff --git a/lisp-kernel/gc-common.c b/lisp-kernel/gc-common.c
@@ -2033,9 +2033,9 @@ new_heap_segment(ExceptionInformation *xp, natural need, Boolean extend, TCR *tc
   platform_new_heap_segment(xp, tcr, (BytePtr)oldlimit, (BytePtr)newlimit);
   if ((BytePtr)oldlimit < heap_dirty_limit) {
     if ((BytePtr)newlimit < heap_dirty_limit) {
-      zero_dnodes((void *)oldlimit,area_dnode(newlimit,oldlimit)); 
+      memset((void *)oldlimit, 0, newlimit - oldlimit);
     } else {
-      zero_dnodes((void *)oldlimit,area_dnode(heap_dirty_limit,oldlimit));
+      memset((void *)oldlimit, 0, (size_t)heap_dirty_limit - oldlimit);
     }
   }
   if ((BytePtr)newlimit > heap_dirty_limit) {

diff --git a/lisp-kernel/gc.h b/lisp-kernel/gc.h
@@ -251,7 +251,5 @@ extern Boolean
 did_gc_notification_since_last_full_gc;
 
 extern BytePtr heap_dirty_limit;
-extern void zero_dnodes(void *,natural);
-
 
 #endif                          /* __GC_H__ */
diff --git a/lisp-kernel/ppc-asmutils.s b/lisp-kernel/ppc-asmutils.s
@@ -426,19 +426,4 @@ _exportfn(C(get_vector_registers))
 	__(lvx v0,r3,r4)
 	__(blr)
 _endfn
-
-/* zero N (r4) dnodes, starting at the dnode-aligned address in r3 */
-_exportfn(C(zero_dnodes))
-        __(cmpri(r4,0))
-        __(li r5,0)
-        __(li r6,0)
-        __(b 1f)
-0:      __(subi r4,r4,1)
-        __(str(r5,0(r3)))
-        __(cmpri(r4,0))
-        __(str(r6,node_size(r3)))
-        __(la r3,dnode_size(r3))
-1:      __(bne 0b)
-        __(blr)
-_endfn                
 	_endfile
diff --git a/lisp-kernel/x86-asmutils32.s b/lisp-kernel/x86-asmutils32.s
@@ -279,19 +279,5 @@ _exportfn(C(ensure_safe_for_string_operations))
 _endfn                                       
         __endif
 
-/* zero arg1 dnodes,starting at the dnode-aligned address in arg0 */
-_exportfn(C(zero_dnodes))  
-        __(xorl %eax,%eax)
-        __(mov 4(%esp),%edx)
-        __(mov 8(%esp),%ecx)
-        __(testl %ecx,%ecx)
-        __(jmp 1f)
-0:      __(mov %eax,0(%edx))
-        __(mov %eax,4(%edx))
-        __(lea dnode_size(%edx),%edx)
-        __(subl $1,%ecx)
-1:      __(jne 0b)
-        __(ret)
-_endfn        
         _endfile
 
diff --git a/lisp-kernel/x86-asmutils64.s b/lisp-kernel/x86-asmutils64.s
@@ -279,15 +279,4 @@ _endfn
 
         __endif
 
-/* zero N (%carg1) dnodes, starting at the dnode-aligned address in %carg0 */
-_exportfn(C(zero_dnodes))
-        __(pxor %xmm0,%xmm0)
-        __(cmpq $0,%carg1)
-        __(jmp 1f)
-0:      __(movdqa %xmm0,(%carg0))
-        __(lea 16(%carg0),%carg0)
-        __(subq $1,%carg1)
-1:      __(jne 0b)
-        __(ret)
-_endfn        
 	_endfile