diff -Nur -x '*.trace' -x '*.rej' -x '*.orig' -x sbcl.h -x ldso-stubs.S -x Config -x target -x CVS -x 'target-*.h' -x output -x genesis -x '*.o' -x '*.nm' -x '*.core' -x '*.map' -x '*.d' -x '*~' -x '*.lisp-obj' -x '*.fasl' -x '*.lisp-temp' -x systems -x '*.tmp' -x local-target-features.lisp-expr -x a.out -x '*.so' -x foo.c -x sbcl -x grovel-headers -x customize-target-features.lisp -x external-format-test.txt -x test-status.lisp-expr sbcl-1.0.18-reserve-ebx-2/src/compiler/x86/call.lisp sbcl-1.0.18-reserve-ebx-3/src/compiler/x86/call.lisp --- sbcl-1.0.18-reserve-ebx-2/src/compiler/x86/call.lisp 2008-07-01 10:56:22.889312280 -0400 +++ sbcl-1.0.18-reserve-ebx-3/src/compiler/x86/call.lisp 2008-07-01 18:06:57.630878840 -0400 @@ -287,46 +287,48 @@ (trace-table-entry trace-table-normal))))))) (t ;; 91 bytes for this branch. - (let ((regs-defaulted (gen-label)) - (restore-edi (gen-label)) - (no-stack-args (gen-label)) - (default-stack-vals (gen-label)) - (count-okay (gen-label))) + (assemble () (note-this-location vop :unknown-return) - ;; Hack so we can test the rest. - #!+x86-reserve-ebx - (inst mov ebx-tn esi-tn) + ;; All code paths here use string instructions, and need to + ;; move down. + (inst std) ;; Branch off to the MV case. (inst jmp :c regs-defaulted) ;; Default the register args, and set up the stack as if we ;; entered the MV return point. - (inst mov ebx-tn esp-tn) + (inst mov values-start esp-tn) (inst push edx-tn) - (inst mov edi-tn nil-value) - (inst push edi-tn) + #!+x86-reserve-ebx + (inst push nil-value) + #!-x86-reserve-ebx + (progn + (inst mov edi-tn nil-value) + (inst push edi-tn)) #!-x86-two-arg-passing-regs (inst mov esi-tn edi-tn) + ;; Compute a pointer to where to put the [defaulted] stack values. - (emit-label no-stack-args) + NO-STACK-ARGS (inst lea edi-tn (make-ea :dword :base ebp-tn :disp (frame-byte-offset register-arg-count))) - ;; Load EAX with NIL so we can quickly store it, and set up - ;; stuff for the loop. - (inst mov eax-tn nil-value) - (inst std) + + ;; Set up loop count. (inst mov ecx-tn (- nvals register-arg-count)) + ;; Jump into the default loop. (inst jmp default-stack-vals) ;; The regs are defaulted. We need to copy any stack arguments, ;; and then default the remaining stack arguments. - (emit-label regs-defaulted) + REGS-DEFAULTED + ;; Save EDI. - (storew edi-tn ebx-tn (frame-word-offset 1)) + (storew edi-tn values-start (frame-word-offset 1)) + ;; Compute the number of stack arguments, and if it's zero or ;; less, don't copy any stack arguments. (inst sub ecx-tn (fixnumize register-arg-count)) @@ -336,44 +338,61 @@ (inst cmp ecx-tn (fixnumize (- nvals register-arg-count))) (inst jmp :be count-okay) (inst mov ecx-tn (fixnumize (- nvals register-arg-count))) - (emit-label count-okay) + COUNT-OKAY + ;; Save the number of stack values. (inst mov eax-tn ecx-tn) ;; Compute a pointer to where the stack args go. (inst lea edi-tn (make-ea :dword :base ebp-tn :disp (frame-byte-offset register-arg-count))) - ;; Save ESI, and compute a pointer to where the args come from. + + ;; Save ESI. #!-x86-two-arg-passing-regs (storew esi-tn ebx-tn (frame-word-offset 2)) + #!+x86-reserve-ebx + (inst push values-start) + + ;; Compute a pointer to where the args come from (inst lea esi-tn - (make-ea :dword :base ebx-tn + (make-ea :dword :base values-start :disp (frame-byte-offset register-arg-count))) + ;; Do the copy. (inst shr ecx-tn word-shift) ; make word count - (inst std) (inst rep) (inst movs :dword) + ;; Restore ESI. #!-x86-two-arg-passing-regs (loadw esi-tn ebx-tn (frame-word-offset 2)) + #!+x86-reserve-ebx + (inst pop values-start) + ;; Now we have to default the remaining args. Find out how many. (inst sub eax-tn (fixnumize (- nvals register-arg-count))) (inst neg eax-tn) + ;; If none, then just blow out of here. (inst jmp :le restore-edi) (inst mov ecx-tn eax-tn) (inst shr ecx-tn word-shift) ; word count + + DEFAULT-STACK-VALS + ;; Load EAX with NIL for fast storing. (inst mov eax-tn nil-value) + ;; Do the store. - (emit-label default-stack-vals) (inst rep) (inst stos eax-tn) + ;; Restore EDI, and reset the stack. - (emit-label restore-edi) - (loadw edi-tn ebx-tn (frame-word-offset 1)) - (inst mov esp-tn ebx-tn) + RESTORE-EDI + (loadw edi-tn values-start (frame-word-offset 1)) + (inst mov esp-tn values-start) + + ;; Restore D flag to its normal state. (inst cld))))) (values)) @@ -1209,18 +1228,19 @@ ;; Allocate the space on the stack. ;; stack = ebp - (max 3 frame-size) - (nargs - fixed) - (inst lea ebx-tn + (inst neg ecx-tn) + (inst lea esp-tn (make-ea :dword :base ebp-tn + :index ecx-tn :disp (- (fixnumize fixed) (* n-word-bytes (max 3 (sb-allocated-size 'stack)))))) - (inst sub ebx-tn ecx-tn) ; Got the new stack in ebx - (inst mov esp-tn ebx-tn) + (inst neg ecx-tn) ;; Now: nargs>=1 && nargs>fixed ;; Save the original count of args. - (inst mov ebx-tn ecx-tn) + (inst push ecx-tn) (cond ((< fixed register-arg-count) ;; We must stop when we run out of stack args, not when we @@ -1242,7 +1262,7 @@ ;; Initialize src to be end of args. (inst mov esi-tn ebp-tn) - (inst sub esi-tn ebx-tn) + (inst sub esi-tn (make-ea :dword :base esp-tn :disp (* 3 n-word-bytes))) ;; We need to copy from downwards up to avoid overwriting some of ;; the yet uncopied args. So we need to use EBX as the copy index @@ -1254,7 +1274,7 @@ COPY-LOOP (inst mov edi-tn (make-ea :dword :base esi-tn :index ebx-tn)) ;; The :DISP is to account for the registers saved on the stack - (inst mov (make-ea :dword :base esp-tn :disp (* 3 n-word-bytes) + (inst mov (make-ea :dword :base esp-tn :disp (* 4 n-word-bytes) :index ebx-tn) edi-tn) (inst add ebx-tn n-word-bytes) @@ -1269,7 +1289,7 @@ DO-REGS ;; Restore ECX - (inst mov ecx-tn ebx-tn) + (inst pop ecx-tn) ;; Here: nargs>=1 && nargs>fixed (when (< fixed register-arg-count) diff -Nur -x '*.trace' -x '*.rej' -x '*.orig' -x sbcl.h -x ldso-stubs.S -x Config -x target -x CVS -x 'target-*.h' -x output -x genesis -x '*.o' -x '*.nm' -x '*.core' -x '*.map' -x '*.d' -x '*~' -x '*.lisp-obj' -x '*.fasl' -x '*.lisp-temp' -x systems -x '*.tmp' -x local-target-features.lisp-expr -x a.out -x '*.so' -x foo.c -x sbcl -x grovel-headers -x customize-target-features.lisp -x external-format-test.txt -x test-status.lisp-expr sbcl-1.0.18-reserve-ebx-2/src/compiler/x86/static-fn.lisp sbcl-1.0.18-reserve-ebx-3/src/compiler/x86/static-fn.lisp --- sbcl-1.0.18-reserve-ebx-2/src/compiler/x86/static-fn.lisp 2008-06-29 17:56:13.000000000 -0400 +++ sbcl-1.0.18-reserve-ebx-3/src/compiler/x86/static-fn.lisp 2008-07-01 11:33:37.369930320 -0400 @@ -17,6 +17,7 @@ (:variant-vars function) (:vop-var vop) (:node-var node) + #!-x86-reserve-ebx (:temporary (:sc unsigned-reg :offset ebx-offset :from (:eval 0) :to (:eval 2)) ebx) (:temporary (:sc unsigned-reg :offset ecx-offset @@ -78,7 +79,8 @@ ;; If speed not more important than size, duplicate the ;; effect of the ENTER with discrete instructions. Takes ;; 2+1+3+2=8 bytes as opposed to 4+3=7 bytes. - (cond ((policy node (>= speed space)) + (cond #!-x86-reserve-ebx + ((policy node (>= speed space)) (inst mov ebx esp-tn) ;; Save the old-fp (inst push ebp-tn) @@ -86,6 +88,17 @@ ;; above, two more needed. (inst sub esp-tn (fixnumize 2)) (inst mov ebp-tn ebx)) + #!+x86-reserve-ebx + ((policy node (>= speed space)) + ;; 1+1+1+4=7 bytes. Not sure the policy makes sense anymore. + ;; Save the old-fp + (inst push ebp-tn) + ;; Ensure that at least three slots are available; one + ;; above, two more needed. + (inst push ebp-tn) + (inst push ebp-tn) + ;; And compute our new fp. + (inst lea ebp-tn (make-ea :dword :base esp-tn :disp 12))) (t (inst enter (fixnumize 2)) ;; The enter instruction pushes EBP and then copies