Clean up and micro-optimize list checking in some x86-64 VOPs.

[sbcl.git] / src / compiler / x86-64 / values.lisp

diff --git a/src/compiler/x86-64/values.lisp b/src/compiler/x86-64/values.lisp
index 4f5f5ae..624483b 100644 (file)
--- a/src/compiler/x86-64/values.lisp
+++ b/src/compiler/x86-64/values.lisp
@@ -38,6 +38,7 @@
     (inst movs :qword)
     (inst cmp rsp-tn rsi)
     (inst jmp :be LOOP)
+    (inst cld)
     DONE
     (inst lea rsp-tn (make-ea :qword :base rdi :disp n-word-bytes))
     (inst sub rdi rsi)
@@ -74,29 +75,27 @@
   (:results (start :scs (any-reg))
             (count :scs (any-reg)))
   (:temporary (:sc descriptor-reg :from (:argument 0) :to (:result 1)) list)
-  (:temporary (:sc descriptor-reg :to (:result 1)) nil-temp)
-  (:temporary (:sc unsigned-reg :offset rax-offset :to (:result 1)) rax)
+  (:temporary (:sc dword-reg :offset eax-offset :to (:result 1)) eax)
+  (:ignore eax)
   (:vop-var vop)
   (:save-p :compute-only)
   (:generator 0
     (move list arg)
     (move start rsp-tn)                 ; WARN pointing 1 below
-    (inst mov nil-temp nil-value)
 
     LOOP
-    (inst cmp list nil-temp)
+    (inst cmp list nil-value)
     (inst jmp :e DONE)
     (pushw list cons-car-slot list-pointer-lowtag)
     (loadw list list cons-cdr-slot list-pointer-lowtag)
-    (inst mov rax list)
-    (inst and al-tn lowtag-mask)
-    (inst cmp al-tn list-pointer-lowtag)
-    (inst jmp :e LOOP)
-    (error-call vop bogus-arg-to-values-list-error list)
+    (%test-lowtag list LOOP nil list-pointer-lowtag)
+    (error-call vop 'bogus-arg-to-values-list-error list)
 
     DONE
     (inst mov count start)              ; start is high address
-    (inst sub count rsp-tn)))           ; stackp is low address
+    (inst sub count rsp-tn)             ; stackp is low address
+    #!-#.(cl:if (cl:= sb!vm:word-shift sb!vm:n-fixnum-tag-bits) '(and) '(or))
+    (inst shr count (- word-shift n-fixnum-tag-bits))))
 
 ;;; Copy the more arg block to the top of the stack so we can use them
 ;;; as function arguments.
@@ -113,7 +112,7 @@
   (:arg-types * positive-fixnum positive-fixnum)
   (:temporary (:sc any-reg :offset rsi-offset :from (:argument 0)) src)
   (:temporary (:sc descriptor-reg :offset rax-offset) temp)
-  (:temporary (:sc unsigned-reg :offset rcx-offset) temp1)
+  (:temporary (:sc unsigned-reg :offset rcx-offset) loop-index)
   (:results (start :scs (any-reg))
             (count :scs (any-reg)))
   (:generator 20
@@ -131,21 +130,31 @@
 
       (any-reg
        (move src context)
+       #!+#.(cl:if (cl:= sb!vm:word-shift sb!vm:n-fixnum-tag-bits) '(and) '(or))
        (inst sub src skip)
+       #!-#.(cl:if (cl:= sb!vm:word-shift sb!vm:n-fixnum-tag-bits) '(and) '(or))
+       (progn
+         ;; FIXME: This can't be efficient, but LEA (my first choice)
+         ;; doesn't do subtraction.
+         (inst shl skip (- word-shift n-fixnum-tag-bits))
+         (inst sub src skip)
+         (inst shr skip (- word-shift n-fixnum-tag-bits)))
        (move count num)
        (inst sub count skip)))
 
-    (move temp1 count)
+    (inst lea loop-index (make-ea :byte :index count
+                                  :scale (ash 1 (- word-shift n-fixnum-tag-bits))))
     (inst mov start rsp-tn)
-    (inst jecxz DONE)  ; check for 0 count?
+    (inst jrcxz DONE)  ; check for 0 count?
 
-    (inst shr temp1 word-shift) ; convert the fixnum to a count.
+    (inst sub rsp-tn loop-index)
+    (inst sub src loop-index)
 
-    (inst std) ; move down the stack as more value are copied to the bottom.
     LOOP
-    (inst lods temp)
-    (inst push temp)
-    (inst loop LOOP)
+    (inst mov temp (make-ea :qword :base src :index loop-index))
+    (inst sub loop-index n-word-bytes)
+    (inst mov (make-ea :qword :base rsp-tn :index loop-index) temp)
+    (inst jmp :nz LOOP)
 
     DONE))