X-Git-Url: http://repo.macrolet.net/gitweb/?a=blobdiff_plain;f=src%2Fcompiler%2Fx86%2Fcall.lisp;h=b0bbd96441847c135c5359453c2b51dc170076b0;hb=12b1dae1a1ed90c6ffe4d958f1281c1c04a8e89b;hp=982a3c165f40b6d19807041f55171056381d1d7a;hpb=1a19da9ff50a45962fc2cf42c7a241a2ce97a364;p=sbcl.git diff --git a/src/compiler/x86/call.lisp b/src/compiler/x86/call.lisp index 982a3c1..b0bbd96 100644 --- a/src/compiler/x86/call.lisp +++ b/src/compiler/x86/call.lisp @@ -15,7 +15,7 @@ ;;; Return a wired TN describing the N'th full call argument passing ;;; location. -(!def-vm-support-routine standard-arg-location (n) +(defun standard-arg-location (n) (declare (type unsigned-byte n)) (if (< n register-arg-count) (make-wired-tn *backend-t-primitive-type* descriptor-reg-sc-number @@ -26,7 +26,7 @@ ;;; ;;; Always wire the return PC location to the stack in its standard ;;; location. -(!def-vm-support-routine make-return-pc-passing-location (standard) +(defun make-return-pc-passing-location (standard) (declare (ignore standard)) (make-wired-tn (primitive-type-or-lose 'system-area-pointer) sap-stack-sc-number return-pc-save-offset)) @@ -38,7 +38,7 @@ ;;; because we want to be able to assume it's always there. Besides, ;;; the x86 doesn't have enough registers to really make it profitable ;;; to pass it in a register. -(!def-vm-support-routine make-old-fp-passing-location (standard) +(defun make-old-fp-passing-location (standard) (declare (ignore standard)) (make-wired-tn *fixnum-primitive-type* control-stack-sc-number ocfp-save-offset)) @@ -49,12 +49,12 @@ ;;; ;;; Without using a save-tn - which does not make much sense if it is ;;; wired to the stack? -(!def-vm-support-routine make-old-fp-save-location (physenv) +(defun make-old-fp-save-location (physenv) (physenv-debug-live-tn (make-wired-tn *fixnum-primitive-type* control-stack-sc-number ocfp-save-offset) physenv)) -(!def-vm-support-routine make-return-pc-save-location (physenv) +(defun make-return-pc-save-location (physenv) (physenv-debug-live-tn (make-wired-tn (primitive-type-or-lose 'system-area-pointer) sap-stack-sc-number return-pc-save-offset) @@ -63,23 +63,23 @@ ;;; Make a TN for the standard argument count passing location. We only ;;; need to make the standard location, since a count is never passed when we ;;; are using non-standard conventions. -(!def-vm-support-routine make-arg-count-location () +(defun make-arg-count-location () (make-wired-tn *fixnum-primitive-type* any-reg-sc-number ecx-offset)) ;;; Make a TN to hold the number-stack frame pointer. This is allocated ;;; once per component, and is component-live. -(!def-vm-support-routine make-nfp-tn () +(defun make-nfp-tn () (make-restricted-tn *fixnum-primitive-type* ignore-me-sc-number)) -(!def-vm-support-routine make-stack-pointer-tn () +(defun make-stack-pointer-tn () (make-normal-tn *fixnum-primitive-type*)) -(!def-vm-support-routine make-number-stack-pointer-tn () +(defun make-number-stack-pointer-tn () (make-restricted-tn *fixnum-primitive-type* ignore-me-sc-number)) ;;; Return a list of TNs that can be used to represent an unknown-values ;;; continuation within a function. -(!def-vm-support-routine make-unknown-values-locations () +(defun make-unknown-values-locations () (list (make-stack-pointer-tn) (make-normal-tn *fixnum-primitive-type*))) @@ -87,7 +87,7 @@ ;;; VM-dependent initialization of the IR2-COMPONENT structure. We ;;; push placeholder entries in the CONSTANTS to leave room for ;;; additional noise in the code object header. -(!def-vm-support-routine select-component-format (component) +(defun select-component-format (component) (declare (type component component)) ;; The 1+ here is because for the x86 the first constant is a ;; pointer to a list of fixups, or NIL if the code object has none. @@ -119,11 +119,162 @@ (:generator 1 nil)) +;;; Accessing a slot from an earlier stack frame is definite hackery. +(define-vop (ancestor-frame-ref) + (:args (frame-pointer :scs (descriptor-reg)) + (variable-home-tn :load-if nil)) + (:results (value :scs (descriptor-reg any-reg))) + (:policy :fast-safe) + (:generator 4 + (aver (sc-is variable-home-tn control-stack)) + (loadw value frame-pointer + (frame-word-offset (tn-offset variable-home-tn))))) +(define-vop (ancestor-frame-set) + (:args (frame-pointer :scs (descriptor-reg)) + (value :scs (descriptor-reg any-reg))) + (:results (variable-home-tn :load-if nil)) + (:policy :fast-safe) + (:generator 4 + (aver (sc-is variable-home-tn control-stack)) + (storew value frame-pointer + (frame-word-offset (tn-offset variable-home-tn))))) + +(macrolet ((define-frame-op + (suffix sc stack-sc instruction + &optional (ea + `(make-ea :dword + :base frame-pointer + :disp (frame-byte-offset + (tn-offset variable-home-tn))))) + (let ((reffer (symbolicate 'ancestor-frame-ref '/ suffix)) + (setter (symbolicate 'ancestor-frame-set '/ suffix))) + `(progn + (define-vop (,reffer ancestor-frame-ref) + (:results (value :scs (,sc))) + (:generator 4 + (aver (sc-is variable-home-tn ,stack-sc)) + (inst ,instruction value + ,ea))) + (define-vop (,setter ancestor-frame-set) + (:args (frame-pointer :scs (descriptor-reg)) + (value :scs (,sc))) + (:generator 4 + (aver (sc-is variable-home-tn ,stack-sc)) + (inst ,instruction ,ea value)))))) + (define-x87-frame-op + (suffix sc stack-sc (load set) + &optional (ea + `(make-ea :dword + :base frame-pointer + :disp (frame-byte-offset + (tn-offset variable-home-tn))))) + (let ((reffer (symbolicate 'ancestor-frame-ref '/ suffix)) + (setter (symbolicate 'ancestor-frame-set '/ suffix))) + `(progn + (define-vop (,reffer ancestor-frame-ref) + (:results (value :scs (,sc))) + (:generator 4 + (aver (sc-is variable-home-tn ,stack-sc)) + ,(if (symbolp load) + `(with-empty-tn@fp-top (value) + (inst ,load ,ea)) + load))) + (define-vop (,setter ancestor-frame-set) + (:args (frame-pointer :scs (descriptor-reg)) + (value :scs (,sc))) + (:generator 4 + (aver (sc-is variable-home-tn ,stack-sc)) + ,(if (symbolp set) + `(with-tn@fp-top (value) + (inst ,set ,ea)) + set))))))) + (define-frame-op signed-byte-32 signed-reg signed-stack mov) + (define-frame-op unsigned-byte-32 unsigned-reg unsigned-stack mov) + (define-frame-op system-area-pointer sap-reg sap-stack mov) + + (define-x87-frame-op double-float double-reg double-stack + (fldd fstd) (make-ea :dword + :base frame-pointer + :disp (frame-byte-offset + (1+ (tn-offset variable-home-tn))))) + (define-x87-frame-op single-float single-reg single-stack + (fld fst)) + + (define-x87-frame-op complex-double-float complex-double-reg + complex-double-stack + ((let ((real (complex-double-reg-real-tn value)) + (imag (complex-double-reg-imag-tn value))) + (with-empty-tn@fp-top (real) + (inst fldd (ea-for-cdf-real-stack variable-home-tn frame-pointer))) + (with-empty-tn@fp-top (imag) + (inst fldd (ea-for-cdf-imag-stack variable-home-tn frame-pointer)))) + (let ((real (complex-double-reg-real-tn value)) + (imag (complex-double-reg-imag-tn value))) + (with-tn@fp-top (real) + (inst fstd (ea-for-cdf-real-stack variable-home-tn frame-pointer))) + (with-tn@fp-top (imag) + (inst fstd (ea-for-cdf-imag-stack variable-home-tn frame-pointer)))))) + (define-x87-frame-op complex-single-float complex-single-reg + complex-single-stack + ((let ((real (complex-single-reg-real-tn value)) + (imag (complex-single-reg-imag-tn value))) + (with-empty-tn@fp-top (real) + (inst fld (ea-for-csf-real-stack variable-home-tn frame-pointer))) + (with-empty-tn@fp-top (imag) + (inst fld (ea-for-csf-imag-stack variable-home-tn frame-pointer)))) + (let ((real (complex-single-reg-real-tn value)) + (imag (complex-single-reg-imag-tn value))) + (with-tn@fp-top (real) + (inst fst (ea-for-csf-real-stack variable-home-tn frame-pointer))) + (with-tn@fp-top (imag) + (inst fst (ea-for-csf-imag-stack variable-home-tn frame-pointer))))))) + +(defun primitive-type-indirect-cell-type (ptype) + (declare (type primitive-type ptype)) + (macrolet ((foo (&body data) + `(case (primitive-type-name ptype) + ,@(loop for (name stack-sc ref set) in data + collect + `(,name + (load-time-value + (list (primitive-type-or-lose ',name) + (sc-or-lose ',stack-sc) + (lambda (node block fp value res) + (sb!c::vop ,ref node block + fp value res)) + (lambda (node block fp new-val value) + (sb!c::vop ,set node block + fp new-val value))))))))) + (foo (double-float double-stack + ancestor-frame-ref/double-float + ancestor-frame-set/double-float) + (single-float single-stack + ancestor-frame-ref/single-float + ancestor-frame-set/single-float) + (complex-double-float complex-double-stack + ancestor-frame-ref/complex-double-float + ancestor-frame-set/complex-double-float) + (complex-single-float complex-single-stack + ancestor-frame-ref/complex-single-float + ancestor-frame-set/complex-single-float) + (signed-byte-32 signed-stack + ancestor-frame-ref/signed-byte-32 + ancestor-frame-set/signed-byte-32) + (unsigned-byte-32 unsigned-stack + ancestor-frame-ref/unsigned-byte-32 + ancestor-frame-set/unsigned-byte-32) + (unsigned-byte-31 unsigned-stack + ancestor-frame-ref/unsigned-byte-32 + ancestor-frame-set/unsigned-byte-32) + (system-area-pointer sap-stack + ancestor-frame-ref/system-area-pointer + ancestor-frame-set/system-area-pointer)))) + (define-vop (xep-allocate-frame) (:info start-lab copy-more-arg-follows) (:vop-var vop) (:generator 1 - (align n-lowtag-bits) + (emit-alignment n-lowtag-bits) (trace-table-entry trace-table-fun-prologue) (emit-label start-lab) ;; Skip space for the function header. @@ -143,7 +294,8 @@ (inst lea esp-tn (make-ea :dword :base ebp-tn :disp (- (* n-word-bytes - (max 3 (sb-allocated-size 'stack))))))) + (- (max 3 (sb-allocated-size 'stack)) + sp->fp-offset)))))) (trace-table-entry trace-table-normal))) @@ -151,12 +303,13 @@ ;;; or a multiple-call-local. All it does is allocate stack space for the ;;; callee (who has the same size stack as us). (define-vop (allocate-frame) - (:results (res :scs (any-reg control-stack)) + (:results (res :scs (any-reg)) (nfp)) (:info callee) (:ignore nfp callee) (:generator 2 - (move res esp-tn) + (inst lea res (make-ea :dword :base esp-tn + :disp (- (* sp->fp-offset n-word-bytes)))) (inst sub esp-tn (* n-word-bytes (sb-allocated-size 'stack))))) ;;; Allocate a partial frame for passing stack arguments in a full @@ -165,9 +318,10 @@ ;;; before it can extend the stack. (define-vop (allocate-full-call-frame) (:info nargs) - (:results (res :scs (any-reg control-stack))) + (:results (res :scs (any-reg))) (:generator 2 - (move res esp-tn) + (inst lea res (make-ea :dword :base esp-tn + :disp (- (* sp->fp-offset n-word-bytes)))) (inst sub esp-tn (* (max nargs 3) n-word-bytes)))) ;;; Emit code needed at the return-point from an unknown-values call @@ -176,12 +330,6 @@ ;;; Nvals is the number of values that are to be received (should ;;; equal the length of Values). ;;; -;;; MOVE-TEMP is a DESCRIPTOR-REG TN used as a temporary. -;;; -;;; This code exploits the fact that in the unknown-values convention, -;;; a single value return returns at the return PC + 2, whereas a -;;; return of other than one value returns directly at the return PC. -;;; ;;; If 0 or 1 values are expected, then we just emit an instruction to ;;; reset the SP (which will only be executed when other than 1 value ;;; is returned.) @@ -194,178 +342,195 @@ ;;; there are stack values. ;;; -- Reset SP. This must be done whenever other than 1 value is ;;; returned, regardless of the number of values desired. -(defun default-unknown-values (vop values nvals) +(defun default-unknown-values (vop values nvals node) (declare (type (or tn-ref null) values) (type unsigned-byte nvals)) - (cond - ((<= nvals 1) - (note-this-location vop :single-value-return) - (let ((single-value (gen-label))) - (cond - ((member :cmov *backend-subfeatures*) - (inst cmov :c esp-tn ebx-tn)) - (t - (inst jmp :nc single-value) - (inst mov esp-tn ebx-tn) - (emit-label single-value))))) - ((<= nvals register-arg-count) - (let ((regs-defaulted (gen-label))) - (note-this-location vop :unknown-return) - (inst jmp :c regs-defaulted) - ;; Default the unsuppled registers. - (let* ((2nd-tn-ref (tn-ref-across values)) - (2nd-tn (tn-ref-tn 2nd-tn-ref))) - (inst mov 2nd-tn nil-value) - (when (> nvals 2) - (loop - for tn-ref = (tn-ref-across 2nd-tn-ref) - then (tn-ref-across tn-ref) - for count from 2 below register-arg-count - do (inst mov (tn-ref-tn tn-ref) 2nd-tn)))) - (inst mov ebx-tn esp-tn) - (emit-label regs-defaulted) - (inst mov esp-tn ebx-tn))) - ((<= nvals 7) - ;; The number of bytes depends on the relative jump instructions. - ;; Best case is 31+(n-3)*14, worst case is 35+(n-3)*18. For - ;; NVALS=6 that is 73/89 bytes, and for NVALS=7 that is 87/107 - ;; bytes which is likely better than using the blt below. - (let ((regs-defaulted (gen-label)) - (defaulting-done (gen-label)) - (default-stack-slots (gen-label))) - (note-this-location vop :unknown-return) - ;; Branch off to the MV case. - (inst jmp :c regs-defaulted) - ;; Do the single value case. - ;; Default the register args - (inst mov eax-tn nil-value) - (do ((i 1 (1+ i)) - (val (tn-ref-across values) (tn-ref-across val))) - ((= i (min nvals register-arg-count))) - (inst mov (tn-ref-tn val) eax-tn)) - - ;; Fake other registers so it looks like we returned with all the - ;; registers filled in. - (move ebx-tn esp-tn) - (inst push edx-tn) - (inst jmp default-stack-slots) - - (emit-label regs-defaulted) - - (inst mov eax-tn nil-value) - (storew edx-tn ebx-tn -1) - (collect ((defaults)) - (do ((i register-arg-count (1+ i)) - (val (do ((i 0 (1+ i)) - (val values (tn-ref-across val))) - ((= i register-arg-count) val)) - (tn-ref-across val))) - ((null val)) - (let ((default-lab (gen-label)) - (tn (tn-ref-tn val))) - (defaults (cons default-lab tn)) - - (inst cmp ecx-tn (fixnumize i)) - (inst jmp :be default-lab) - (loadw edx-tn ebx-tn (frame-word-offset i)) - (inst mov tn edx-tn))) - - (emit-label defaulting-done) - (loadw edx-tn ebx-tn -1) - (move esp-tn ebx-tn) - - (let ((defaults (defaults))) - (when defaults - (assemble (*elsewhere*) - (trace-table-entry trace-table-fun-prologue) - (emit-label default-stack-slots) - (dolist (default defaults) - (emit-label (car default)) - (inst mov (cdr default) eax-tn)) - (inst jmp defaulting-done) - (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))) - (note-this-location vop :unknown-return) - ;; 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 push edx-tn) - (inst mov edi-tn nil-value) - (inst push edi-tn) - (inst mov esi-tn edi-tn) - ;; Compute a pointer to where to put the [defaulted] stack values. - (emit-label 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) - (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) - ;; Save EDI. - (storew edi-tn ebx-tn (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)) - (inst jmp :le no-stack-args) - - ;; Throw away any unwanted args. - (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) - ;; 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. - (storew esi-tn ebx-tn (frame-word-offset 2)) - (inst lea esi-tn - (make-ea :dword :base ebx-tn - :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. - (loadw esi-tn ebx-tn (frame-word-offset 2)) - ;; 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 - ;; 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) - (inst cld)))) + (let ((type (sb!c::basic-combination-derived-type node))) + (cond + ((<= nvals 1) + (note-this-location vop :single-value-return) + (cond + ((<= (sb!kernel:values-type-max-value-count type) + register-arg-count) + (when (and (named-type-p type) + (eq nil (named-type-name type))) + ;; The function never returns, it may happen that the code + ;; ends right here leavig the :SINGLE-VALUE-RETURN note + ;; dangling. Let's emit a NOP. + (inst nop))) + ((not (sb!kernel:values-type-may-be-single-value-p type)) + (inst mov esp-tn ebx-tn)) + ((member :cmov *backend-subfeatures*) + (inst cmov :c esp-tn ebx-tn)) + (t + (let ((single-value (gen-label))) + (inst jmp :nc single-value) + (inst mov esp-tn ebx-tn) + (emit-label single-value))))) + ((<= nvals register-arg-count) + (note-this-location vop :unknown-return) + (when (sb!kernel:values-type-may-be-single-value-p type) + (let ((regs-defaulted (gen-label))) + (inst jmp :c regs-defaulted) + ;; Default the unsupplied registers. + (let* ((2nd-tn-ref (tn-ref-across values)) + (2nd-tn (tn-ref-tn 2nd-tn-ref))) + (inst mov 2nd-tn nil-value) + (when (> nvals 2) + (loop + for tn-ref = (tn-ref-across 2nd-tn-ref) + then (tn-ref-across tn-ref) + for count from 2 below register-arg-count + do (inst mov (tn-ref-tn tn-ref) 2nd-tn)))) + (inst mov ebx-tn esp-tn) + (emit-label regs-defaulted))) + (when (< register-arg-count + (sb!kernel:values-type-max-value-count type)) + (inst mov esp-tn ebx-tn))) + ((<= nvals 7) + ;; The number of bytes depends on the relative jump instructions. + ;; Best case is 31+(n-3)*14, worst case is 35+(n-3)*18. For + ;; NVALS=6 that is 73/89 bytes, and for NVALS=7 that is 87/107 + ;; bytes which is likely better than using the blt below. + (let ((regs-defaulted (gen-label)) + (defaulting-done (gen-label)) + (default-stack-slots (gen-label))) + (note-this-location vop :unknown-return) + ;; Branch off to the MV case. + (inst jmp :c regs-defaulted) + ;; Do the single value case. + ;; Default the register args + (inst mov eax-tn nil-value) + (do ((i 1 (1+ i)) + (val (tn-ref-across values) (tn-ref-across val))) + ((= i (min nvals register-arg-count))) + (inst mov (tn-ref-tn val) eax-tn)) + ;; Fake other registers so it looks like we returned with all the + ;; registers filled in. + (move ebx-tn esp-tn) + (inst jmp default-stack-slots) + (emit-label regs-defaulted) + (inst mov eax-tn nil-value) + (collect ((defaults)) + (do ((i register-arg-count (1+ i)) + (val (do ((i 0 (1+ i)) + (val values (tn-ref-across val))) + ((= i register-arg-count) val)) + (tn-ref-across val))) + ((null val)) + (let ((default-lab (gen-label)) + (tn (tn-ref-tn val)) + (first-stack-arg-p (= i register-arg-count))) + (defaults (cons default-lab + (cons tn first-stack-arg-p))) + (inst cmp ecx-tn (fixnumize i)) + (inst jmp :be default-lab) + (when first-stack-arg-p + ;; There are stack args so the frame of the callee is + ;; still there, save EDX in its first slot temporalily. + (storew edx-tn ebx-tn (frame-word-offset sp->fp-offset))) + (loadw edx-tn ebx-tn (frame-word-offset (+ sp->fp-offset i))) + (inst mov tn edx-tn))) + (emit-label defaulting-done) + (loadw edx-tn ebx-tn (frame-word-offset sp->fp-offset)) + (move esp-tn ebx-tn) + (let ((defaults (defaults))) + (when defaults + (assemble (*elsewhere*) + (trace-table-entry trace-table-fun-prologue) + (emit-label default-stack-slots) + (dolist (default defaults) + (emit-label (car default)) + (when (cddr default) + ;; We are setting the first stack argument to NIL. + ;; The callee's stack frame is dead, save EDX by + ;; pushing it to the stack, it will end up at same + ;; place as in the (STOREW EDX-TN EBX-TN -1) case + ;; above. + (inst push edx-tn)) + (inst mov (second default) eax-tn)) + (inst jmp defaulting-done) + (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))) + (note-this-location vop :unknown-return) + ;; 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 edi-tn nil-value) + (inst mov esi-tn edi-tn) + ;; Compute a pointer to where to put the [defaulted] stack values. + (emit-label no-stack-args) + (inst push edx-tn) + (inst push edi-tn) + (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) + (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) + ;; 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)) + (inst jmp :le no-stack-args) + ;; Save EDI. + (storew edi-tn ebx-tn (frame-word-offset (+ sp->fp-offset 1))) + ;; Throw away any unwanted args. + (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) + ;; 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. + (storew esi-tn ebx-tn (frame-word-offset (+ sp->fp-offset 2))) + (inst lea esi-tn + (make-ea :dword :base ebx-tn + :disp (frame-byte-offset + (+ sp->fp-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. + (loadw esi-tn ebx-tn (frame-word-offset (+ sp->fp-offset 2))) + ;; 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 + ;; 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 (+ sp->fp-offset 1))) + (inst mov esp-tn ebx-tn) + (inst cld))))) (values)) ;;;; unknown values receiving @@ -387,27 +552,37 @@ ;;; explicitly allocate these TNs, since their lifetimes overlap with ;;; the results start and count. (Also, it's nice to be able to target ;;; them.) -(defun receive-unknown-values (args nargs start count) +(defun receive-unknown-values (args nargs start count node) (declare (type tn args nargs start count)) - (let ((variable-values (gen-label)) + (let ((type (sb!c::basic-combination-derived-type node)) + (variable-values (gen-label)) + (stack-values (gen-label)) (done (gen-label))) - (inst jmp :c variable-values) - - (cond ((location= start (first *register-arg-tns*)) - (inst push (first *register-arg-tns*)) - (inst lea start (make-ea :dword :base esp-tn :disp 4))) - (t (inst mov start esp-tn) - (inst push (first *register-arg-tns*)))) - (inst mov count (fixnumize 1)) - (inst jmp done) - - (emit-label variable-values) + (when (sb!kernel:values-type-may-be-single-value-p type) + (inst jmp :c variable-values) + (cond ((location= start (first *register-arg-tns*)) + (inst push (first *register-arg-tns*)) + (inst lea start (make-ea :dword :base esp-tn :disp n-word-bytes))) + (t (inst mov start esp-tn) + (inst push (first *register-arg-tns*)))) + (inst mov count (fixnumize 1)) + (inst jmp done) + (emit-label variable-values)) + ;; The stack frame is burnt and RETurned from if there are no + ;; stack values. In this case quickly reallocate sufficient space. + (when (<= (sb!kernel:values-type-min-value-count type) + register-arg-count) + (inst cmp nargs (fixnumize register-arg-count)) + (inst jmp :g stack-values) + (inst sub esp-tn nargs) + (emit-label stack-values)) ;; dtc: this writes the registers onto the stack even if they are ;; not needed, only the number specified in ecx are used and have ;; stack allocated to them. No harm is done. (loop for arg in *register-arg-tns* for i downfrom -1 + for j below (sb!kernel:values-type-max-value-count type) do (storew arg args i)) (move start args) (move count nargs) @@ -429,6 +604,35 @@ ;;;; local call with unknown values convention return +(defun check-ocfp-and-return-pc (old-fp return-pc) + #+nil + (format t "*known-return: old-fp ~S, tn-kind ~S; ~S ~S~%" + old-fp (sb!c::tn-kind old-fp) (sb!c::tn-save-tn old-fp) + (sb!c::tn-kind (sb!c::tn-save-tn old-fp))) + #+nil + (format t "*known-return: return-pc ~S, tn-kind ~S; ~S ~S~%" + return-pc (sb!c::tn-kind return-pc) + (sb!c::tn-save-tn return-pc) + (sb!c::tn-kind (sb!c::tn-save-tn return-pc))) + (unless (and (sc-is old-fp control-stack) + (= (tn-offset old-fp) ocfp-save-offset)) + (error "ocfp not on stack in standard save location?")) + (unless (and (sc-is return-pc sap-stack) + (= (tn-offset return-pc) return-pc-save-offset)) + (error "return-pc not on stack in standard save location?"))) + +;;; The local call convention doesn't fit that well with x86-style +;;; calls. Emit a header for local calls to pop the return address +;;; in the right place. +(defun emit-block-header (start-label trampoline-label fall-thru-p alignp) + (declare (ignore alignp)) + (when trampoline-label + (when fall-thru-p + (inst jmp start-label)) + (emit-label trampoline-label) + (popw ebp-tn (frame-word-offset return-pc-save-offset))) + (emit-label start-label)) + ;;; Non-TR local call for a fixed number of values passed according to ;;; the unknown values convention. ;;; @@ -457,31 +661,14 @@ (:move-args :local-call) (:info arg-locs callee target nvals) (:vop-var vop) - (:ignore nfp arg-locs args #+nil callee) + (:ignore nfp arg-locs args callee) + (:node-var node) (:generator 5 (trace-table-entry trace-table-call-site) (move ebp-tn fp) - - (let ((ret-tn (callee-return-pc-tn callee))) - #+nil - (format t "*call-local ~S; tn-kind ~S; tn-save-tn ~S; its tn-kind ~S~%" - ret-tn (sb!c::tn-kind ret-tn) (sb!c::tn-save-tn ret-tn) - (sb!c::tn-kind (sb!c::tn-save-tn ret-tn))) - - ;; Is the return-pc on the stack or in a register? - (sc-case ret-tn - ((sap-stack) - #+nil (format t "*call-local: ret-tn on stack; offset=~S~%" - (tn-offset ret-tn)) - (storew (make-fixup nil :code-object return) - ebp-tn (frame-word-offset (tn-offset ret-tn)))) - ((sap-reg) - (inst lea ret-tn (make-fixup nil :code-object return))))) - (note-this-location vop :call-site) - (inst jmp target) - RETURN - (default-unknown-values vop values nvals) + (inst call target) + (default-unknown-values vop values nvals node) (trace-table-entry trace-table-normal))) ;;; Non-TR local call for a variable number of return values passed according @@ -494,35 +681,16 @@ (:save-p t) (:move-args :local-call) (:info save callee target) - (:ignore args save nfp #+nil callee) + (:ignore args save nfp callee) (:vop-var vop) + (:node-var node) (:generator 20 (trace-table-entry trace-table-call-site) (move ebp-tn fp) - - (let ((ret-tn (callee-return-pc-tn callee))) - #+nil - (format t "*multiple-call-local ~S; tn-kind ~S; tn-save-tn ~S; its tn-kind ~S~%" - ret-tn (sb!c::tn-kind ret-tn) (sb!c::tn-save-tn ret-tn) - (sb!c::tn-kind (sb!c::tn-save-tn ret-tn))) - - ;; Is the return-pc on the stack or in a register? - (sc-case ret-tn - ((sap-stack) - #+nil (format t "*multiple-call-local: ret-tn on stack; offset=~S~%" - (tn-offset ret-tn)) - ;; Stack - (storew (make-fixup nil :code-object return) - ebp-tn (frame-word-offset (tn-offset ret-tn)))) - ((sap-reg) - ;; Register - (inst lea ret-tn (make-fixup nil :code-object return))))) - (note-this-location vop :call-site) - (inst jmp target) - RETURN + (inst call target) (note-this-location vop :unknown-return) - (receive-unknown-values values-start nvals start count) + (receive-unknown-values values-start nvals start count node) (trace-table-entry trace-table-normal))) ;;;; local call with known values return @@ -541,79 +709,20 @@ (:move-args :local-call) (:save-p t) (:info save callee target) - (:ignore args res save nfp #+nil callee) + (:ignore args res save nfp callee) (:vop-var vop) (:generator 5 (trace-table-entry trace-table-call-site) (move ebp-tn fp) - - (let ((ret-tn (callee-return-pc-tn callee))) - - #+nil - (format t "*known-call-local ~S; tn-kind ~S; tn-save-tn ~S; its tn-kind ~S~%" - ret-tn (sb!c::tn-kind ret-tn) (sb!c::tn-save-tn ret-tn) - (sb!c::tn-kind (sb!c::tn-save-tn ret-tn))) - - ;; Is the return-pc on the stack or in a register? - (sc-case ret-tn - ((sap-stack) - #+nil (format t "*known-call-local: ret-tn on stack; offset=~S~%" - (tn-offset ret-tn)) - ;; Stack - (storew (make-fixup nil :code-object return) - ebp-tn (frame-word-offset (tn-offset ret-tn)))) - ((sap-reg) - ;; Register - (inst lea ret-tn (make-fixup nil :code-object return))))) - (note-this-location vop :call-site) - (inst jmp target) - RETURN + (inst call target) (note-this-location vop :known-return) (trace-table-entry trace-table-normal))) -;;; Return from known values call. We receive the return locations as -;;; arguments to terminate their lifetimes in the returning function. We -;;; restore FP and CSP and jump to the Return-PC. -;;; -;;; We can assume we know exactly where old-fp and return-pc are because -;;; make-old-fp-save-location and make-return-pc-save-location always -;;; return the same place. -#+nil -(define-vop (known-return) - (:args (old-fp) - (return-pc :scs (any-reg immediate-stack) :target rpc) - (vals :more t)) - (:move-args :known-return) - (:info val-locs) - (:temporary (:sc unsigned-reg :from (:argument 1)) rpc) - (:ignore val-locs vals) - (:vop-var vop) - (:generator 6 - (trace-table-entry trace-table-fun-epilogue) - ;; Save the return-pc in a register 'cause the frame-pointer is - ;; going away. Note this not in the usual stack location so we - ;; can't use RET - (move rpc return-pc) - ;; Restore the stack. - (move esp-tn ebp-tn) - ;; Restore the old fp. We know OLD-FP is going to be in its stack - ;; save slot, which is a different frame that than this one, - ;; so we don't have to worry about having just cleared - ;; most of the stack. - (move ebp-tn old-fp) - (inst jmp rpc) - (trace-table-entry trace-table-normal))) - ;;; From Douglas Crosher ;;; Return from known values call. We receive the return locations as ;;; arguments to terminate their lifetimes in the returning function. We ;;; restore FP and CSP and jump to the Return-PC. -;;; -;;; The old-fp may be either in a register or on the stack in its -;;; standard save locations - slot 0. -;;; -;;; The return-pc may be in a register or on the stack in any slot. (define-vop (known-return) (:args (old-fp) (return-pc) @@ -623,66 +732,12 @@ (:ignore val-locs vals) (:vop-var vop) (:generator 6 + (check-ocfp-and-return-pc old-fp return-pc) (trace-table-entry trace-table-fun-epilogue) - - #+nil (format t "*known-return: old-fp ~S, tn-kind ~S; ~S ~S~%" - old-fp (sb!c::tn-kind old-fp) (sb!c::tn-save-tn old-fp) - (sb!c::tn-kind (sb!c::tn-save-tn old-fp))) - - #+nil (format t "*known-return: return-pc ~S, tn-kind ~S; ~S ~S~%" - return-pc (sb!c::tn-kind return-pc) - (sb!c::tn-save-tn return-pc) - (sb!c::tn-kind (sb!c::tn-save-tn return-pc))) - - ;; return-pc may be either in a register or on the stack. - (sc-case return-pc - ((sap-reg) - (sc-case old-fp - ((control-stack) - - #+nil (format t "*known-return: old-fp ~S on stack; offset=~S~%" - old-fp (tn-offset old-fp)) - - (cond ((zerop (tn-offset old-fp)) - ;; Zot all of the stack except for the old-fp. - (inst lea esp-tn (make-ea :dword :base ebp-tn - :disp (frame-byte-offset ocfp-save-offset))) - ;; Restore the old fp from its save location on the stack, - ;; and zot the stack. - (inst pop ebp-tn)) - - (t - (cerror "Continue anyway" - "VOP return-local doesn't work if old-fp (in slot ~ - ~S) is not in slot 0" - (tn-offset old-fp))))) - - ((any-reg descriptor-reg) - ;; Zot all the stack. - (move esp-tn ebp-tn) - ;; Restore the old-fp. - (move ebp-tn old-fp))) - - ;; Return; return-pc is in a register. - (inst jmp return-pc)) - - ((sap-stack) - - #+nil (format t "*known-return: return-pc ~S on stack; offset=~S~%" - return-pc (tn-offset return-pc)) - - ;; Zot all of the stack except for the old-fp and return-pc. - (inst lea esp-tn - (make-ea :dword :base ebp-tn - :disp (frame-byte-offset (tn-offset return-pc)))) - ;; Restore the old fp. old-fp may be either on the stack in its - ;; save location or in a register, in either case this restores it. - (move ebp-tn old-fp) - ;; The return pops the return address (4 bytes), then we need - ;; to pop all the slots before the return-pc which includes the - ;; 4 bytes for the old-fp. - (inst ret (* (tn-offset return-pc) n-word-bytes)))) - + ;; Zot all of the stack except for the old-fp and return-pc. + (inst mov esp-tn ebp-tn) + (inst pop ebp-tn) + (inst ret) (trace-table-entry trace-table-normal))) ;;;; full call @@ -794,6 +849,8 @@ :from (:argument 1) :to (:argument 2)) old-fp-tmp))) + ,@(unless (eq return :tail) + '((:node-var node))) (:generator ,(+ (if named 5 0) (if variable 19 1) @@ -848,13 +905,13 @@ ;; FIXME: FORMAT T for stale ;; diagnostic output (several of ;; them around here), ick - (format t "** tail-call old-fp not S0~%") + (error "** tail-call old-fp not S0~%") (move old-fp-tmp old-fp) (storew old-fp-tmp ebp-tn (frame-word-offset ocfp-save-offset)))) ((any-reg descriptor-reg) - (format t "** tail-call old-fp in reg not S0~%") + (error "** tail-call old-fp in reg not S0~%") (storew old-fp ebp-tn (frame-word-offset ocfp-save-offset)))) @@ -882,8 +939,13 @@ ,(if variable '(inst sub esp-tn (fixnumize 3))) + ;; Bias the new-fp for use as an fp + ,(if variable + '(inst sub new-fp (fixnumize sp->fp-offset))) + ;; Save the fp - (storew ebp-tn new-fp (frame-word-offset ocfp-save-offset)) + (storew ebp-tn new-fp + (frame-word-offset ocfp-save-offset)) (move ebp-tn new-fp) ; NB - now on new stack frame. ))) @@ -904,10 +966,11 @@ fun-pointer-lowtag))) ,@(ecase return (:fixed - '((default-unknown-values vop values nvals))) + '((default-unknown-values vop values nvals node))) (:unknown '((note-this-location vop :unknown-return) - (receive-unknown-values values-start nvals start count))) + (receive-unknown-values values-start nvals start count + node))) (:tail)) (trace-table-entry trace-table-normal))))) @@ -928,36 +991,20 @@ (:args (args :scs (any-reg control-stack) :target esi) (function :scs (descriptor-reg control-stack) :target eax) (old-fp) - (ret-addr)) + (return-pc)) (:temporary (:sc unsigned-reg :offset esi-offset :from (:argument 0)) esi) (:temporary (:sc unsigned-reg :offset eax-offset :from (:argument 1)) eax) -; (:ignore ret-addr old-fp) (:generator 75 + (check-ocfp-and-return-pc old-fp return-pc) ;; Move these into the passing locations if they are not already there. (move esi args) (move eax function) - - ;; The following assumes that the return-pc and old-fp are on the - ;; stack in their standard save locations - Check this. - (unless (and (sc-is old-fp control-stack) - (= (tn-offset old-fp) ocfp-save-offset)) - (error "tail-call-variable: ocfp not on stack in standard save location?")) - (unless (and (sc-is ret-addr sap-stack) - (= (tn-offset ret-addr) return-pc-save-offset)) - (error "tail-call-variable: ret-addr not on stack in standard save location?")) - - ;; And jump to the assembly routine. (inst jmp (make-fixup 'tail-call-variable :assembly-routine)))) ;;;; unknown values return -;;; Return a single-value using the Unknown-Values convention. Specifically, -;;; we jump to clear the stack and jump to return-pc+2. -;;; -;;; We require old-fp to be in a register, because we want to reset ESP before -;;; restoring EBP. If old-fp were still on the stack, it could get clobbered -;;; by a signal. +;;; Return a single-value using the Unknown-Values convention. ;;; ;;; pfw--get wired-tn conflicts sometimes if register sc specd for args ;;; having problems targeting args to regs -- using temps instead. @@ -972,56 +1019,16 @@ (value)) (:ignore value) (:generator 6 + (check-ocfp-and-return-pc old-fp return-pc) (trace-table-entry trace-table-fun-epilogue) - ;; Code structure lifted from known-return. - (sc-case return-pc - ((sap-reg) - ;; return PC in register for some reason (local call?) - ;; we jmp to the return pc after fixing the stack and frame. - (sc-case old-fp - ((control-stack) - ;; ofp on stack must be in slot 0 (the traditional storage place). - ;; Drop the stack above it and pop it off. - (cond ((zerop (tn-offset old-fp)) - (inst lea esp-tn (make-ea :dword :base ebp-tn - :disp (frame-byte-offset ocfp-save-offset))) - (inst pop ebp-tn)) - (t - ;; Should this ever happen, we do the same as above, but - ;; using (tn-offset old-fp) instead of ocfp-save-offset - ;; (which is 0 anyway, see src/compiler/x86/vm.lisp) and - ;; then lea esp again against itself with a displacement - ;; of (* (tn-offset old-fp) n-word-bytes) to clear the - ;; rest of the stack. - (cerror "Continue anyway" - "VOP return-single doesn't work if old-fp (in slot ~S) is not in slot 0" (tn-offset old-fp))))) - ((any-reg descriptor-reg) - ;; ofp in reg, drop the stack and load the real fp. - (move esp-tn ebp-tn) - (move ebp-tn old-fp))) - - ;; Set single-value-return flag - (inst clc) - ;; And return - (inst jmp return-pc)) - - ((sap-stack) - ;; Note that this will only work right if, when old-fp is on - ;; the stack, it has a lower tn-offset than return-pc. One of - ;; the comments in known-return indicate that this is the case - ;; (in that it will be in its save location), but we may wish - ;; to assert that (in either the weaker or stronger forms). - ;; Should this ever not be the case, we should load old-fp - ;; into a temp reg while we fix the stack. - ;; Drop stack above return-pc - (inst lea esp-tn (make-ea :dword :base ebp-tn - :disp (frame-byte-offset (tn-offset return-pc)))) - ;; Set single-value return flag - (inst clc) - ;; Restore the old frame pointer - (move ebp-tn old-fp) - ;; And return, dropping the rest of the stack as we go. - (inst ret (* (tn-offset return-pc) n-word-bytes)))))) + ;; Drop stack above old-fp + (inst mov esp-tn ebp-tn) + ;; Clear the multiple-value return flag + (inst clc) + ;; Restore the old frame pointer + (inst pop ebp-tn) + ;; And return. + (inst ret))) ;;; Do unknown-values return of a fixed (other than 1) number of ;;; values. The VALUES are required to be set up in the standard @@ -1036,12 +1043,10 @@ (values :more t)) (:ignore values) (:info nvals) - ;; In the case of other than one value, we need these registers to ;; tell the caller where they are and how many there are. (:temporary (:sc unsigned-reg :offset ebx-offset) ebx) (:temporary (:sc unsigned-reg :offset ecx-offset) ecx) - ;; We need to stretch the lifetime of return-pc past the argument ;; registers so that we can default the argument registers without ;; trashing return-pc. @@ -1053,18 +1058,17 @@ :from :eval) a2) (:generator 6 + (check-ocfp-and-return-pc old-fp return-pc) + (when (= nvals 1) + ;; This is handled in RETURN-SINGLE. + (error "nvalues is 1")) (trace-table-entry trace-table-fun-epilogue) ;; Establish the values pointer and values count. - (move ebx ebp-tn) + (inst lea ebx (make-ea :dword :base ebp-tn + :disp (* sp->fp-offset n-word-bytes))) (if (zerop nvals) - (inst xor ecx ecx) ; smaller - (inst mov ecx (fixnumize nvals))) - ;; Restore the frame pointer. - (move ebp-tn old-fp) - ;; Clear as much of the stack as possible, but not past the return - ;; address. - (inst lea esp-tn (make-ea :dword :base ebx - :disp (- (* (max nvals 2) n-word-bytes)))) + (inst xor ecx ecx) ; smaller + (inst mov ecx (fixnumize nvals))) ;; Pre-default any argument register that need it. (when (< nvals register-arg-count) (let* ((arg-tns (nthcdr nvals (list a0 a1 a2))) @@ -1072,21 +1076,31 @@ (inst mov first nil-value) (dolist (tn (cdr arg-tns)) (inst mov tn first)))) - ;; Set multi-value return flag. + ;; Set the multiple value return flag. (inst stc) ;; And away we go. Except that return-pc is still on the ;; stack and we've changed the stack pointer. So we have to ;; tell it to index off of EBX instead of EBP. - (cond ((zerop nvals) - ;; Return popping the return address and the OCFP. - (inst ret n-word-bytes)) - ((= nvals 1) - ;; Return popping the return, leaving 1 slot. Can this - ;; happen, or is a single value return handled elsewhere? + (cond ((<= nvals register-arg-count) + (inst mov esp-tn ebp-tn) + (inst pop ebp-tn) (inst ret)) (t - (inst jmp (make-ea :dword :base ebx - :disp (frame-byte-offset (tn-offset return-pc)))))) + ;; Some values are on the stack after RETURN-PC and OLD-FP, + ;; can't return normally and some slots of the frame will + ;; be used as temporaries by the receiver. + ;; + ;; Clear as much of the stack as possible, but not past the + ;; old frame address. + (inst lea esp-tn + (make-ea :dword :base ebp-tn + :disp (frame-byte-offset (1- nvals)))) + (move ebp-tn old-fp) + (inst push (make-ea :dword :base ebx + :disp (frame-byte-offset + (+ sp->fp-offset + (tn-offset return-pc))))) + (inst ret))) (trace-table-entry trace-table-normal))) @@ -1097,53 +1111,39 @@ ;;; assembly-routine. ;;; ;;; The assembly routine takes the following args: -;;; EAX -- the return-pc to finally jump to. -;;; EBX -- pointer to where to put the values. ;;; ECX -- number of values to find there. ;;; ESI -- pointer to where to find the values. (define-vop (return-multiple) - (:args (old-fp :to (:eval 1) :target old-fp-temp) - (return-pc :target eax) + (:args (old-fp) + (return-pc) (vals :scs (any-reg) :target esi) (nvals :scs (any-reg) :target ecx)) - - (:temporary (:sc unsigned-reg :offset eax-offset :from (:argument 1)) eax) (:temporary (:sc unsigned-reg :offset esi-offset :from (:argument 2)) esi) (:temporary (:sc unsigned-reg :offset ecx-offset :from (:argument 3)) ecx) - (:temporary (:sc unsigned-reg :offset ebx-offset :from (:eval 0)) ebx) (:temporary (:sc descriptor-reg :offset (first *register-arg-offsets*) :from (:eval 0)) a0) - (:temporary (:sc unsigned-reg :from (:eval 1)) old-fp-temp) (:node-var node) - (:generator 13 + (check-ocfp-and-return-pc old-fp return-pc) (trace-table-entry trace-table-fun-epilogue) - ;; Load the return-pc. - (move eax return-pc) (unless (policy node (> space speed)) ;; Check for the single case. (let ((not-single (gen-label))) (inst cmp nvals (fixnumize 1)) (inst jmp :ne not-single) - ;; Return with one value. (loadw a0 vals -1) - ;; Clear the stack. We load old-fp into a register before clearing - ;; the stack. - (move old-fp-temp old-fp) - (move esp-tn ebp-tn) - (move ebp-tn old-fp-temp) - ;; Set the single-value return flag. + ;; Clear the stack until ocfp. + (inst mov esp-tn ebp-tn) + ;; clear the multiple-value return flag (inst clc) ;; Out of here. - (inst jmp eax) - + (inst pop ebp-tn) + (inst ret) ;; Nope, not the single case. Jump to the assembly routine. (emit-label not-single))) (move esi vals) (move ecx nvals) - (move ebx ebp-tn) - (move ebp-tn old-fp) (inst jmp (make-fixup 'return-multiple :assembly-routine)) (trace-table-entry trace-table-normal))) @@ -1176,7 +1176,7 @@ ;;; ;;; EAX -- The lexenv. ;;; EBX -- Available. -;;; ECX -- The total number of arguments. +;;; ECX -- The total number of arguments * N-WORD-BYTES. ;;; EDX -- The first arg. ;;; EDI -- The second arg. ;;; ESI -- The third arg. @@ -1191,19 +1191,29 @@ (:generator 20 ;; Avoid the copy if there are no more args. (cond ((zerop fixed) - (inst jecxz just-alloc-frame)) + (inst jecxz JUST-ALLOC-FRAME)) (t (inst cmp ecx-tn (fixnumize fixed)) - (inst jmp :be just-alloc-frame))) + (inst jmp :be JUST-ALLOC-FRAME))) ;; Allocate the space on the stack. - ;; stack = ebp - (max 3 frame-size) - (nargs - fixed) + ;; stack = ebp + sp->fp-offset - (max 3 frame-size) - (nargs - fixed) + ;; + ;; Problem: this might leave some &more args outside esp, so + ;; clamp the movement for now. If fixed > frame-size, reset + ;; esp to the end of the current &more args (which *should* + ;; be a noop?), and only set esp to its final value after the + ;; stack-stack memmove loop. Otherwise, an unlucky signal + ;; could end up overwriting the &more arguments before they're + ;; moved in their final place. (inst lea ebx-tn (make-ea :dword :base ebp-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 + :disp (* n-word-bytes + (- sp->fp-offset + (max 0 + (- (max 3 (sb-allocated-size 'stack)) + fixed)))))) + (inst sub ebx-tn ecx-tn) ; Got the new stack in ebx (inst mov esp-tn ebx-tn) ;; Now: nargs>=1 && nargs>fixed @@ -1217,78 +1227,107 @@ ;; Number to copy = nargs-3 (inst sub ecx-tn (fixnumize register-arg-count)) ;; Everything of interest in registers. - (inst jmp :be do-regs)) + (inst jmp :be DO-REGS)) (t ;; Number to copy = nargs-fixed (inst sub ecx-tn (fixnumize fixed)))) - ;; Save edi and esi register args. - (inst push edi-tn) - (inst push esi-tn) - (inst push ebx-tn) - ;; Okay, we have pushed the register args. We can trash them - ;; now. - - ;; Initialize src to be end of args. - (inst mov esi-tn ebp-tn) - (inst sub esi-tn ebx-tn) - - ;; 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 - ;; and ECX as the loop counter, rather than using ECX for both. - (inst xor ebx-tn ebx-tn) - - ;; We used to use REP MOVS here, but on modern x86 it performs - ;; much worse than an explicit loop for small blocks. - 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) - :index ebx-tn) - edi-tn) - (inst add ebx-tn n-word-bytes) - (inst sub ecx-tn n-word-bytes) - (inst jmp :nz COPY-LOOP) - - ;; So now we need to restore EDI and ESI. - (inst pop ebx-tn) - (inst pop esi-tn) - (inst pop edi-tn) - + (let ((delta (* n-word-bytes + (- (max 3 (sb-allocated-size 'stack)) + fixed))) + (LOOP (gen-label))) + (cond ((zerop delta) + ;; nothing to move! + ) + ((minusp delta) + ;; stack frame smaller than fixed; moving args to higher + ;; addresses (stack grows downard), so copy from the + ;; end. Moreover, because we'd have to shrink the frame, + ;; esp currently points at the end of the source args. + (inst push ebx-tn) + + (emit-label LOOP) + (inst sub ecx-tn n-word-bytes) + (inst mov ebx-tn (make-ea :dword + :base esp-tn :index ecx-tn + ;; compensate for PUSH above + :disp n-word-bytes)) + (inst mov (make-ea :dword + :base esp-tn :index ecx-tn + ;; compensate for PUSH, and + ;; add (abs delta) + :disp (- n-word-bytes delta)) + ebx-tn) + (inst jmp :nz LOOP) + + (inst pop ebx-tn)) + ((plusp delta) + ;; stack frame larger than fixed. Moving args to lower + ;; addresses, so copy from the lowest address. esp + ;; already points to the lowest address of the destination. + (inst push ebx-tn) + (inst push esi-tn) + + (inst xor ebx-tn ebx-tn) + (emit-label LOOP) + (inst mov esi-tn (make-ea :dword + :base esp-tn :index ebx-tn + ;; PUSHed 2 words + :disp (+ (* 2 n-word-bytes) + delta))) + (inst mov (make-ea :dword + :base esp-tn :index ebx-tn + :disp (* 2 n-word-bytes)) + esi-tn) + (inst add ebx-tn n-word-bytes) + (inst sub ecx-tn n-word-bytes) + (inst jmp :nz LOOP) + + (inst pop esi-tn) + (inst pop ebx-tn)))) DO-REGS + ;; stack can now be set to its final size + (when (< (max 3 (sb-allocated-size 'stack)) fixed) + (inst add esp-tn (* n-word-bytes + (- fixed + (max 3 (sb-allocated-size 'stack)))))) ;; Restore ECX (inst mov ecx-tn ebx-tn) ;; Here: nargs>=1 && nargs>fixed (when (< fixed register-arg-count) - ;; Now we have to deposit any more args that showed up in - ;; registers. - (do ((i fixed)) - ( nil ) - ;; Store it relative to ebp - (inst mov (make-ea :dword :base ebp-tn - :disp (- (* 4 - (+ 1 (- i fixed) - (max 3 (sb-allocated-size 'stack)))))) - (nth i *register-arg-tns*)) - - (incf i) - (when (>= i register-arg-count) - (return)) - - ;; Don't deposit any more than there are. - (if (zerop i) - (inst test ecx-tn ecx-tn) - (inst cmp ecx-tn (fixnumize i))) - (inst jmp :eq done))) - - (inst jmp done) + ;; Now we have to deposit any more args that showed up in + ;; registers. + (do ((i fixed)) + ( nil ) + ;; Store it relative to ebp + (inst mov (make-ea :dword :base ebp-tn + :disp (* n-word-bytes + (- sp->fp-offset + (+ 1 + (- i fixed) + (max 3 (sb-allocated-size + 'stack)))))) + (nth i *register-arg-tns*)) + + (incf i) + (when (>= i register-arg-count) + (return)) + + ;; Don't deposit any more than there are. + (if (zerop i) + (inst test ecx-tn ecx-tn) + (inst cmp ecx-tn (fixnumize i))) + (inst jmp :eq DONE))) + + (inst jmp DONE) JUST-ALLOC-FRAME (inst lea esp-tn (make-ea :dword :base ebp-tn - :disp (- (* n-word-bytes + :disp (* n-word-bytes + (- sp->fp-offset (max 3 (sb-allocated-size 'stack)))))) DONE)) @@ -1327,9 +1366,6 @@ (inst mov value (make-ea :dword :base object :index value)))) ;;; Turn more arg (context, count) into a list. -(defoptimizer (%listify-rest-args stack-allocate-result) ((&rest args)) - t) - (define-vop (listify-rest-args) (:translate %listify-rest-args) (:policy :safe) @@ -1354,14 +1390,12 @@ (inst jecxz done) (inst lea dst (make-ea :dword :base ecx :index ecx)) (maybe-pseudo-atomic stack-allocate-p - (allocation dst dst node stack-allocate-p) - (inst lea dst (make-ea :byte :base dst :disp list-pointer-lowtag)) - (inst shr ecx 2) + (allocation dst dst node stack-allocate-p list-pointer-lowtag) ;; Set decrement mode (successive args at lower addresses) (inst std) ;; Set up the result. (move result dst) - ;; Jump into the middle of the loop, 'cause that's were we want + ;; Jump into the middle of the loop, 'cause that's where we want ;; to start. (inst jmp enter) (emit-label loop) @@ -1374,7 +1408,7 @@ (inst lods eax) (storew eax dst 0 list-pointer-lowtag) ;; Go back for more. - (inst sub ecx 1) + (inst sub ecx n-word-bytes) (inst jmp :nz loop) ;; NIL out the last cons. (storew nil-value dst 1 list-pointer-lowtag) @@ -1407,7 +1441,7 @@ ;; Point to the first more-arg, not above it. (inst lea context (make-ea :dword :base esp-tn :index count :scale 1 - :disp (- (+ (fixnumize fixed) 4)))) + :disp (- (+ (fixnumize fixed) n-word-bytes)))) (unless (zerop fixed) (inst sub count (fixnumize fixed))))) @@ -1422,7 +1456,7 @@ (:save-p :compute-only) (:generator 3 (let ((err-lab - (generate-error-code vop invalid-arg-count-error nargs))) + (generate-error-code vop 'invalid-arg-count-error nargs))) (if (zerop count) (inst test nargs nargs) ; smaller instruction (inst cmp nargs (fixnumize count))) @@ -1440,7 +1474,7 @@ (:vop-var vop) (:save-p :compute-only) (:generator 1000 - (error-call vop ,error ,@args))))) + (error-call vop ',error ,@args))))) (def arg-count-error invalid-arg-count-error sb!c::%arg-count-error nargs) (def type-check-error object-not-type-error sb!c::%type-check-error @@ -1465,10 +1499,12 @@ ;; register on -SB-THREAD. #!+sb-thread (progn - (inst fs-segment-prefix) - (inst cmp (make-ea :dword - :disp (* thread-stepping-slot n-word-bytes)) - nil-value)) + #!+win32 (inst push eax-tn) + (with-tls-ea (EA :base #!+win32 eax-tn #!-win32 :unused + :disp-type :constant + :disp (* thread-stepping-slot n-word-bytes)) + (inst cmp EA nil-value :maybe-fs)) + #!+win32 (inst pop eax-tn)) #!-sb-thread (inst cmp (make-ea-for-symbol-value sb!impl::*stepping*) nil-value))