Another minor optimization of x86-64 pthread_cond_wait.

diff --git a/nptl/ChangeLog b/nptl/ChangeLog
index 0f5c231..48fcc0f 100644 (file)
--- a/nptl/ChangeLog
+++ b/nptl/ChangeLog
@@ -1,3 +1,9 @@
+2009-08-08  Ulrich Drepper  <drepper@redhat.com>
+
+       * sysdeps/unix/sysv/linux/x86_64/pthread_cond_wait.S
+       (__pthread_cond_wait): Optimize by avoiding use of callee-safe
+       register.
+
 2009-08-07  Ulrich Drepper  <drepper@redhat.com>
 
        * sysdeps/unix/sysv/linux/x86_64/sem_wait.S: Little optimizations

diff --git a/nptl/sysdeps/unix/sysv/linux/x86_64/pthread_cond_wait.S b/nptl/sysdeps/unix/sysv/linux/x86_64/pthread_cond_wait.S
index 45116b1..f5b929e 100644 (file)
--- a/nptl/sysdeps/unix/sysv/linux/x86_64/pthread_cond_wait.S
+++ b/nptl/sysdeps/unix/sysv/linux/x86_64/pthread_cond_wait.S
@@ -45,9 +45,6 @@ __pthread_cond_wait:
        cfi_lsda(DW_EH_PE_udata4, .LexceptSTART)
 #endif
 
-       pushq   %r13
-       cfi_adjust_cfa_offset(8)
-       cfi_rel_offset(%r13, 0)
 #define FRAME_SIZE 32
        leaq    -FRAME_SIZE(%rsp), %rsp
        cfi_adjust_cfa_offset(FRAME_SIZE)
@@ -140,7 +137,7 @@ __pthread_cond_wait:
        movl    $SYS_futex, %eax
        syscall
 
-       movl    $1, %r13d
+       movl    $1, %r8d
 #ifdef __ASSUME_REQUEUE_PI
        jmp     62f
 #else
@@ -158,7 +155,7 @@ __pthread_cond_wait:
 #else
        orl     %fs:PRIVATE_FUTEX, %esi
 #endif
-60:    xorl    %r13d, %r13d
+60:    xorl    %r8d, %r8d
        movl    $SYS_futex, %eax
        syscall
 
@@ -233,20 +230,18 @@ __pthread_cond_wait:
        /* If requeue_pi is used the kernel performs the locking of the
           mutex. */
 11:    movq    16(%rsp), %rdi
-       testl   %r13d, %r13d
+       testl   %r8d, %r8d
        jnz     18f
 
        callq   __pthread_mutex_cond_lock
 
-14:    movq    FRAME_SIZE(%rsp), %r13
-       leaq    FRAME_SIZE+8(%rsp), %rsp
-       cfi_adjust_cfa_offset(-(FRAME_SIZE + 8))
+14:    leaq    FRAME_SIZE(%rsp), %rsp
+       cfi_adjust_cfa_offset(-FRAME_SIZE)
 
        /* We return the result of the mutex_lock operation.  */
        retq
 
-       cfi_adjust_cfa_offset(8 + FRAME_SIZE)
-       cfi_rel_offset(%r13, FRAME_SIZE)
+       cfi_adjust_cfa_offset(FRAME_SIZE)
 
 18:    callq   __pthread_mutex_cond_lock_adjust
        xorl    %eax, %eax
@@ -341,9 +336,7 @@ versioned_symbol (libpthread, __pthread_cond_wait, pthread_cond_wait,
 __condvar_cleanup1:
        /* Stack frame:
 
-          rsp + 40
-                   +--------------------------+
-          rsp + 32 | %r13                     |
+          rsp + 32
                    +--------------------------+
           rsp + 24 | unused                   |
                    +--------------------------+
@@ -465,7 +458,6 @@ __condvar_cleanup1:
        callq   __pthread_mutex_cond_lock
 
        movq    24(%rsp), %rdi
-       movq    32(%rsp), %r13
 .LcallUR:
        call    _Unwind_Resume@PLT
        hlt