1.0.27.8: slightly faster x86oid pseudo atomic with {e,r}bp

Trusting that within SBCL ebp is even and that it doesn't change
within a pseudo atomic section it's possible to use it to set and
clear pseudo atomic bits. It is ever so slightly faster (about 0.5%
overall on cl-bench on a P4).

diff --git a/src/compiler/x86-64/macros.lisp b/src/compiler/x86-64/macros.lisp
index 03da6d7..7711130 100644 (file)
--- a/src/compiler/x86-64/macros.lisp
+++ b/src/compiler/x86-64/macros.lisp
@@ -294,48 +294,46 @@
 (defmacro pseudo-atomic (&rest forms)
   (with-unique-names (label)
     `(let ((,label (gen-label)))
-      (inst or (make-ea :byte
-                 :base thread-base-tn
-                 :disp (* 8 thread-pseudo-atomic-bits-slot))
-            (fixnumize 1))
-      ,@forms
-      (inst xor (make-ea :byte
-                 :base thread-base-tn
-                 :disp (* 8 thread-pseudo-atomic-bits-slot))
-            (fixnumize 1))
-      (inst jmp :z ,label)
-      ;; if PAI was set, interrupts were disabled at the same
-      ;; time using the process signal mask.
-      (inst break pending-interrupt-trap)
-      (emit-label ,label))))
+       (inst mov (make-ea :qword
+                          :base thread-base-tn
+                          :disp (* 8 thread-pseudo-atomic-bits-slot))
+             rbp-tn)
+       ,@forms
+       (inst xor (make-ea :qword
+                          :base thread-base-tn
+                          :disp (* 8 thread-pseudo-atomic-bits-slot))
+             rbp-tn)
+       (inst jmp :z ,label)
+       ;; if PAI was set, interrupts were disabled at the same time
+       ;; using the process signal mask.
+       (inst break pending-interrupt-trap)
+       (emit-label ,label))))
 
 
 #!-sb-thread
 (defmacro pseudo-atomic (&rest forms)
   (with-unique-names (label)
     `(let ((,label (gen-label)))
-      ;; FIXME: The MAKE-EA noise should become a MACROLET macro or
-      ;; something. (perhaps SVLB, for static variable low byte)
-      (inst or (make-ea :byte :disp (+ nil-value
-                                       (static-symbol-offset
-                                        '*pseudo-atomic-bits*)
-                                       (ash symbol-value-slot word-shift)
-                                       (- other-pointer-lowtag)))
-            (fixnumize 1))
-      ,@forms
-      (inst xor (make-ea :byte :disp (+ nil-value
-                                        (static-symbol-offset
-                                         '*pseudo-atomic-bits*)
-                                        (ash symbol-value-slot word-shift)
-                                        (- other-pointer-lowtag)))
-            (fixnumize 1))
-      (inst jmp :z ,label)
-      ;; if PAI was set, interrupts were disabled at the same time
-      ;; using the process signal mask.
-      (inst break pending-interrupt-trap)
-      (emit-label ,label))))
-
-
+       ;; FIXME: The MAKE-EA noise should become a MACROLET macro or
+       ;; something. (perhaps SVLB, for static variable low byte)
+       (inst mov (make-ea :qword :disp (+ nil-value
+                                          (static-symbol-offset
+                                           '*pseudo-atomic-bits*)
+                                          (ash symbol-value-slot word-shift)
+                                          (- other-pointer-lowtag)))
+             rbp-tn)
+       ,@forms
+       (inst xor (make-ea :qword :disp (+ nil-value
+                                          (static-symbol-offset
+                                           '*pseudo-atomic-bits*)
+                                          (ash symbol-value-slot word-shift)
+                                          (- other-pointer-lowtag)))
+             rbp-tn)
+       (inst jmp :z ,label)
+       ;; if PAI was set, interrupts were disabled at the same time
+       ;; using the process signal mask.
+       (inst break pending-interrupt-trap)
+       (emit-label ,label))))
 \f
 ;;;; indexed references
 

diff --git a/src/compiler/x86/macros.lisp b/src/compiler/x86/macros.lisp
index 9789ec2..aa6f1e7 100644 (file)
--- a/src/compiler/x86/macros.lisp
+++ b/src/compiler/x86/macros.lisp
@@ -354,14 +354,14 @@
 (defmacro pseudo-atomic (&rest forms)
   (with-unique-names (label)
     `(let ((,label (gen-label)))
-       (inst or (make-ea :byte :disp (* 4 thread-pseudo-atomic-bits-slot))
-             (fixnumize 1) :fs)
+       (inst mov (make-ea :dword :disp (* 4 thread-pseudo-atomic-bits-slot))
+             ebp-tn :fs)
        ,@forms
-       (inst xor (make-ea :byte :disp (* 4 thread-pseudo-atomic-bits-slot))
-             (fixnumize 1) :fs)
+       (inst xor (make-ea :dword :disp (* 4 thread-pseudo-atomic-bits-slot))
+             ebp-tn :fs)
        (inst jmp :z ,label)
-       ;; if PAI was set, interrupts were disabled at the same
-       ;; time using the process signal mask.
+       ;; if PAI was set, interrupts were disabled at the same time
+       ;; using the process signal mask.
        (inst break pending-interrupt-trap)
        (emit-label ,label))))
 
@@ -369,14 +369,14 @@
 (defmacro pseudo-atomic (&rest forms)
   (with-unique-names (label)
     `(let ((,label (gen-label)))
-       (inst or (make-ea-for-symbol-value *pseudo-atomic-bits* :byte)
-             (fixnumize 1))
+       (inst mov (make-ea-for-symbol-value *pseudo-atomic-bits* :dword)
+             ebp-tn)
        ,@forms
-       (inst xor (make-ea-for-symbol-value *pseudo-atomic-bits* :byte)
-             (fixnumize 1))
+       (inst xor (make-ea-for-symbol-value *pseudo-atomic-bits* :dword)
+             ebp-tn)
        (inst jmp :z ,label)
-       ;; if PAI was set, interrupts were disabled at the same
-       ;; time using the process signal mask.
+       ;; if PAI was set, interrupts were disabled at the same time
+       ;; using the process signal mask.
        (inst break pending-interrupt-trap)
        (emit-label ,label))))
 \f

diff --git a/src/runtime/pseudo-atomic.h b/src/runtime/pseudo-atomic.h
index 9636e5b..40898c7 100644 (file)
--- a/src/runtime/pseudo-atomic.h
+++ b/src/runtime/pseudo-atomic.h
@@ -34,18 +34,19 @@
 static inline int
 get_pseudo_atomic_atomic(struct thread *thread)
 {
-    return fixnum_value(SymbolValue(PSEUDO_ATOMIC_BITS, thread) &
-                        make_fixnum(1));
+    return SymbolValue(PSEUDO_ATOMIC_BITS, thread) & (~1);
 }
 
 static inline void
 set_pseudo_atomic_atomic(struct thread *thread)
 {
     lispobj *p = SymbolValueAddress(PSEUDO_ATOMIC_BITS, thread);
+    if (*p)
+        lose("set_pseudo_atomic_atomic: pseudo atomic bits is %d.", *p);
     __asm__ __volatile__
         ("or" LISPOBJ_ASM_SUFFIX " %0,%1"
          :
-         : "g" (make_fixnum(1)), "m" (*p)
+         : "g" (~1), "m" (*p)
          : "memory");
 }
 
@@ -56,36 +57,39 @@ clear_pseudo_atomic_atomic(struct thread *thread)
     __asm__ __volatile__
         ("and" LISPOBJ_ASM_SUFFIX " %0,%1"
          :
-         : "g" (~make_fixnum(1)), "m" (*p)
+         : "g" (1), "m" (*p)
          : "memory");
 }
 
 static inline int
 get_pseudo_atomic_interrupted(struct thread *thread)
 {
-    return fixnum_value(SymbolValue(PSEUDO_ATOMIC_BITS, thread) &
-                        make_fixnum(2));
+    return SymbolValue(PSEUDO_ATOMIC_BITS, thread) & 1;
 }
 
 static inline void
 set_pseudo_atomic_interrupted(struct thread *thread)
 {
+    if (!get_pseudo_atomic_atomic(thread))
+        lose("set_pseudo_atomic_interrupted not in pseudo atomic");
     lispobj *p = SymbolValueAddress(PSEUDO_ATOMIC_BITS, thread);
     __asm__ __volatile__
         ("or" LISPOBJ_ASM_SUFFIX " %0,%1"
          :
-         : "g" (make_fixnum(2)), "m" (*p)
+         : "g" (1), "m" (*p)
          : "memory");
 }
 
 static inline void
 clear_pseudo_atomic_interrupted(struct thread *thread)
 {
+    if (get_pseudo_atomic_atomic(thread))
+        lose("clear_pseudo_atomic_interrupted in pseudo atomic");
     lispobj *p = SymbolValueAddress(PSEUDO_ATOMIC_BITS, thread);
     __asm__ __volatile__
         ("and" LISPOBJ_ASM_SUFFIX " %0,%1"
          :
-         : "g" (~make_fixnum(2)), "m" (*p)
+         : "g" (~1), "m" (*p)
          : "memory");
 }
 

diff --git a/version.lisp-expr b/version.lisp-expr
index 5d1a5e3..ee218e0 100644 (file)
--- a/version.lisp-expr
+++ b/version.lisp-expr
@@ -17,4 +17,4 @@
 ;;; checkins which aren't released. (And occasionally for internal
 ;;; versions, especially for internal versions off the main CVS
 ;;; branch, it gets hairier, e.g. "0.pre7.14.flaky4.13".)
-"1.0.27.7"
+"1.0.27.8"