;; The start of the actual code.
;; Save the return-pc.
- (popw rbp-tn (- (1+ return-pc-save-offset)))
+ (popw rbp-tn (frame-word-offset return-pc-save-offset))
;; If copy-more-arg follows it will allocate the correct stack
;; size. The stack is not allocated first here as this may expose
(inst lea rsp-tn
(make-ea :qword :base rbp-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)))
;;; 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 rsp-tn)
+ (inst lea res (make-ea :qword :base rsp-tn
+ :disp (- (* sp->fp-offset n-word-bytes))))
(inst sub rsp-tn (* n-word-bytes (sb-allocated-size 'stack)))))
;;; Allocate a partial frame for passing stack arguments in a full
;;; 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 rsp-tn)
+ (inst lea res (make-ea :qword :base rsp-tn
+ :disp (- (* sp->fp-offset n-word-bytes))))
(inst sub rsp-tn (* (max nargs 3) n-word-bytes))))
\f
;;; Emit code needed at the return-point from an unknown-values call
;;; 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.)
;;; 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)
- (inst cmov :c rsp-tn rbx-tn))
- ((<= nvals register-arg-count)
- (let ((regs-defaulted (gen-label)))
- (note-this-location vop :unknown-return)
- (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 rbx-tn rsp-tn)
- (emit-label regs-defaulted)
- (inst mov rsp-tn rbx-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 rax-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) rax-tn))
-
- ;; Fake other registers so it looks like we returned with all the
- ;; registers filled in.
- (move rbx-tn rsp-tn)
- (inst push rdx-tn)
- (inst jmp default-stack-slots)
-
- (emit-label regs-defaulted)
-
- (inst mov rax-tn nil-value)
- (storew rdx-tn rbx-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 rcx-tn (fixnumize i))
- (inst jmp :be default-lab)
- (loadw rdx-tn rbx-tn (- (1+ i)))
- (inst mov tn rdx-tn)))
-
- (emit-label defaulting-done)
- (loadw rdx-tn rbx-tn -1)
- (move rsp-tn rbx-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) rax-tn))
- (inst jmp defaulting-done)
- (trace-table-entry trace-table-normal)))))))
- (t
- (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 rbx-tn rsp-tn)
- (inst push rdx-tn)
- (inst mov rdi-tn nil-value)
- (inst push rdi-tn)
- (inst mov rsi-tn rdi-tn)
- ;; Compute a pointer to where to put the [defaulted] stack values.
- (emit-label no-stack-args)
- (inst lea rdi-tn
- (make-ea :qword :base rbp-tn
- :disp (* (- (1+ register-arg-count)) n-word-bytes)))
- ;; Load RAX with NIL so we can quickly store it, and set up
- ;; stuff for the loop.
- (inst mov rax-tn nil-value)
- (inst std)
- (inst mov rcx-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 rdi-tn rbx-tn (- (1+ 1)))
- ;; Compute the number of stack arguments, and if it's zero or
- ;; less, don't copy any stack arguments.
- (inst sub rcx-tn (fixnumize register-arg-count))
- (inst jmp :le no-stack-args)
-
- ;; Throw away any unwanted args.
- (inst cmp rcx-tn (fixnumize (- nvals register-arg-count)))
- (inst jmp :be count-okay)
- (inst mov rcx-tn (fixnumize (- nvals register-arg-count)))
- (emit-label count-okay)
- ;; Save the number of stack values.
- (inst mov rax-tn rcx-tn)
- ;; Compute a pointer to where the stack args go.
- (inst lea rdi-tn
- (make-ea :qword :base rbp-tn
- :disp (* (- (1+ register-arg-count)) n-word-bytes)))
- ;; Save ESI, and compute a pointer to where the args come from.
- (storew rsi-tn rbx-tn (- (1+ 2)))
- (inst lea rsi-tn
- (make-ea :qword :base rbx-tn
- :disp (* (- (1+ register-arg-count)) n-word-bytes)))
- ;; Do the copy.
- (inst shr rcx-tn word-shift) ; make word count
- (inst std)
- (inst rep)
- (inst movs :qword)
- ;; Restore RSI.
- (loadw rsi-tn rbx-tn (- (1+ 2)))
- ;; Now we have to default the remaining args. Find out how many.
- (inst sub rax-tn (fixnumize (- nvals register-arg-count)))
- (inst neg rax-tn)
- ;; If none, then just blow out of here.
- (inst jmp :le restore-edi)
- (inst mov rcx-tn rax-tn)
- (inst shr rcx-tn word-shift) ; word count
- ;; Load RAX with NIL for fast storing.
- (inst mov rax-tn nil-value)
- ;; Do the store.
- (emit-label default-stack-vals)
- (inst rep)
- (inst stos rax-tn)
- ;; Restore EDI, and reset the stack.
- (emit-label restore-edi)
- (loadw rdi-tn rbx-tn (- (1+ 1)))
- (inst mov rsp-tn rbx-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 rsp-tn rbx-tn))
+ (t
+ (inst cmov :c rsp-tn rbx-tn))))
+ ((<= 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 rbx-tn rsp-tn)
+ (emit-label regs-defaulted)))
+ (when (< register-arg-count
+ (sb!kernel:values-type-max-value-count type))
+ (inst mov rsp-tn rbx-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 rax-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) rax-tn))
+ ;; Fake other registers so it looks like we returned with all the
+ ;; registers filled in.
+ (move rbx-tn rsp-tn)
+ (inst jmp default-stack-slots)
+ (emit-label regs-defaulted)
+ (inst mov rax-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 rcx-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 RDX in its first slot temporalily.
+ (storew rdx-tn rbx-tn (frame-word-offset sp->fp-offset)))
+ (loadw rdx-tn rbx-tn (frame-word-offset (+ sp->fp-offset i)))
+ (inst mov tn rdx-tn)))
+ (emit-label defaulting-done)
+ (loadw rdx-tn rbx-tn (frame-word-offset sp->fp-offset))
+ (move rsp-tn rbx-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 RDX by
+ ;; pushing it to the stack, it will end up at same
+ ;; place as in the (STOREW RDX-TN RBX-TN -1) case
+ ;; above.
+ (inst push rdx-tn))
+ (inst mov (second default) rax-tn))
+ (inst jmp defaulting-done)
+ (trace-table-entry trace-table-normal)))))))
+ (t
+ (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 rbx-tn rsp-tn)
+ (inst mov rdi-tn nil-value)
+ (inst mov rsi-tn rdi-tn)
+ ;; Compute a pointer to where to put the [defaulted] stack values.
+ (emit-label no-stack-args)
+ (inst push rdx-tn)
+ (inst push rdi-tn)
+ (inst lea rdi-tn
+ (make-ea :qword :base rbp-tn
+ :disp (frame-byte-offset register-arg-count)))
+ ;; Load RAX with NIL so we can quickly store it, and set up
+ ;; stuff for the loop.
+ (inst mov rax-tn nil-value)
+ (inst std)
+ (inst mov rcx-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 rcx-tn (fixnumize register-arg-count))
+ (inst jmp :le no-stack-args)
+ ;; Save EDI.
+ (storew rdi-tn rbx-tn (frame-word-offset (+ sp->fp-offset 1)))
+ ;; Throw away any unwanted args.
+ (inst cmp rcx-tn (fixnumize (- nvals register-arg-count)))
+ (inst jmp :be count-okay)
+ (inst mov rcx-tn (fixnumize (- nvals register-arg-count)))
+ (emit-label count-okay)
+ ;; Save the number of stack values.
+ (inst mov rax-tn rcx-tn)
+ ;; Compute a pointer to where the stack args go.
+ (inst lea rdi-tn
+ (make-ea :qword :base rbp-tn
+ :disp (frame-byte-offset register-arg-count)))
+ ;; Save ESI, and compute a pointer to where the args come from.
+ (storew rsi-tn rbx-tn (frame-word-offset (+ sp->fp-offset 2)))
+ (inst lea rsi-tn
+ (make-ea :qword :base rbx-tn
+ :disp (frame-byte-offset
+ (+ sp->fp-offset register-arg-count))))
+ ;; Do the copy.
+ (inst shr rcx-tn word-shift) ; make word count
+ (inst std)
+ (inst rep)
+ (inst movs :qword)
+ ;; Restore RSI.
+ (loadw rsi-tn rbx-tn (frame-word-offset (+ sp->fp-offset 2)))
+ ;; Now we have to default the remaining args. Find out how many.
+ (inst sub rax-tn (fixnumize (- nvals register-arg-count)))
+ (inst neg rax-tn)
+ ;; If none, then just blow out of here.
+ (inst jmp :le restore-edi)
+ (inst mov rcx-tn rax-tn)
+ (inst shr rcx-tn word-shift) ; word count
+ ;; Load RAX with NIL for fast storing.
+ (inst mov rax-tn nil-value)
+ ;; Do the store.
+ (emit-label default-stack-vals)
+ (inst rep)
+ (inst stos rax-tn)
+ ;; Restore EDI, and reset the stack.
+ (emit-label restore-edi)
+ (loadw rdi-tn rbx-tn (frame-word-offset (+ sp->fp-offset 1)))
+ (inst mov rsp-tn rbx-tn)
+ (inst cld)))))
(values))
\f
;;;; unknown values receiving
;;; 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 :qword :base rsp-tn :disp 8)))
- (t (inst mov start rsp-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 :qword :base rsp-tn :disp n-word-bytes)))
+ (t (inst mov start rsp-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 rsp-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 rcx 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)
\f
;;;; 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?")))
+
+;;; Instead of JMPing to TARGET, CALL a trampoline that saves the
+;;; return pc and jumps. Although this is an incredibly stupid trick
+;;; the paired CALL/RET instructions are a big win.
+(defun make-local-call (target)
+ (let ((tramp (gen-label)))
+ (inst call tramp)
+ (assemble (*elsewhere*)
+ (emit-label tramp)
+ (popw rbp-tn (frame-word-offset return-pc-save-offset))
+ (inst jmp target))))
+
;;; Non-TR local call for a fixed number of values passed according to
;;; the unknown values convention.
;;;
(:args (fp)
(nfp)
(args :more t))
- (:temporary (:sc unsigned-reg) return-label)
(:results (values :more t))
(:save-p t)
(: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 rbp-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))
- (inst lea return-label (make-fixup nil :code-object RETURN))
- (storew return-label rbp-tn (- (1+ (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)
+ (make-local-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
(:args (fp)
(nfp)
(args :more t))
- (:temporary (:sc unsigned-reg) return-label)
(: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 rbp-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
- (inst lea return-label (make-fixup nil :code-object RETURN))
- (storew return-label rbp-tn (- (1+ (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
+ (make-local-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)))
\f
;;;; local call with known values return
(:args (fp)
(nfp)
(args :more t))
- (:temporary (:sc unsigned-reg) return-label)
(:results (res :more t))
(: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 rbp-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
- (inst lea return-label (make-fixup nil :code-object RETURN))
- (storew return-label rbp-tn (- (1+ (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
+ (make-local-call target)
(note-this-location vop :known-return)
(trace-table-entry trace-table-normal)))
\f
-;;; 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 rsp-tn rbp-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 rbp-tn old-fp)
- (inst jmp rpc)
- (trace-table-entry trace-table-normal)))
-\f
;;; 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)
(: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)
- ;; return-pc may be either in a register or on the stack.
- (sc-case return-pc
- ((sap-reg)
- (sc-case old-fp
- ((control-stack)
- (cond ((zerop (tn-offset old-fp))
- ;; Zot all of the stack except for the old-fp.
- (inst lea rsp-tn (make-ea :qword :base rbp-tn
- :disp (- (* (1+ ocfp-save-offset)
- n-word-bytes))))
- ;; Restore the old fp from its save location on the stack,
- ;; and zot the stack.
- (inst pop rbp-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 rsp-tn rbp-tn)
- ;; Restore the old-fp.
- (move rbp-tn old-fp)))
-
- ;; Return; return-pc is in a register.
- (inst jmp return-pc))
-
- ((sap-stack)
- (inst lea rsp-tn
- (make-ea :qword :base rbp-tn
- :disp (- (* (1+ (tn-offset return-pc)) n-word-bytes))))
- (move rbp-tn 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 rsp-tn rbp-tn)
+ (inst pop rbp-tn)
+ (inst ret)
(trace-table-entry trace-table-normal)))
\f
;;;; full call
: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)
;; 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
rbp-tn
- (- (1+ ocfp-save-offset)))))
+ (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
rbp-tn
- (- (1+ ocfp-save-offset)))))
+ (frame-word-offset ocfp-save-offset))))
;; For tail call, we have to push the
;; return-pc so that it looks like we CALLed
- ;; drspite the fact that we are going to JMP.
+ ;; despite the fact that we are going to JMP.
(inst push return-pc)
))
(t
,(if variable
'(inst sub rsp-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 rbp-tn new-fp (- (1+ ocfp-save-offset)))
+ (storew rbp-tn new-fp
+ (frame-word-offset ocfp-save-offset))
(move rbp-tn new-fp) ; NB - now on new stack frame.
)))
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)))))
(define-full-call call nil :fixed nil)
- (define-full-call call-named t :fixed nil)
+ (define-full-call call-named t :fixed nil)
(define-full-call multiple-call nil :unknown nil)
(define-full-call multiple-call-named t :unknown nil)
(define-full-call tail-call nil :tail nil)
(:args (args :scs (any-reg control-stack) :target rsi)
(function :scs (descriptor-reg control-stack) :target rax)
(old-fp)
- (ret-addr))
+ (return-pc))
(:temporary (:sc unsigned-reg :offset rsi-offset :from (:argument 0)) rsi)
(:temporary (:sc unsigned-reg :offset rax-offset :from (:argument 1)) rax)
(:temporary (:sc unsigned-reg) call-target)
-; (: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 rsi args)
(move rax 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 lea call-target
(make-ea :qword
:disp (make-fixup 'tail-call-variable :assembly-routine)))
- ;; And jump to the assembly routine.
(inst jmp call-target)))
\f
;;;; unknown values return
-;;; Return a single-value using the Unknown-Values convention. Specifically,
-;;; we jump to clear the stack and jump to return-pc+3.
-;;;
-;;; We require old-fp to be in a register, because we want to reset RSP before
-;;; restoring RBP. 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.
+;;;
+;;; First off, modifying the return-pc defeats the branch-prediction
+;;; optimizations on modern CPUs quite handily. Second, we can do all
+;;; this without needing a temp register. Fixed the latter, at least.
+;;; -- AB 2006/Feb/04
(define-vop (return-single)
(:args (old-fp)
(return-pc)
(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 rsp-tn (make-ea :dword :base rbp-tn
- :disp (- (* (1+ ocfp-save-offset)
- n-word-bytes))))
- (inst pop rbp-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 rsp 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 rsp-tn rbp-tn)
- (move rbp-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 rsp-tn (make-ea :dword :base rbp-tn
- :disp (- (* (1+ (tn-offset return-pc))
- n-word-bytes))))
- ;; Set single-value return flag
- (inst clc)
- ;; Restore the old frame pointer
- (move rbp-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 rsp-tn rbp-tn)
+ ;; Clear the multiple-value return flag
+ (inst clc)
+ ;; Restore the old frame pointer
+ (inst pop rbp-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
(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 rbx-offset) rbx)
(:temporary (:sc unsigned-reg :offset rcx-offset) rcx)
-
;; 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.
: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 rbx rbp-tn)
+ (inst lea rbx (make-ea :qword :base rbp-tn
+ :disp (* sp->fp-offset n-word-bytes)))
(if (zerop nvals)
(zeroize rcx) ; smaller
- (inst mov rcx (fixnumize nvals)))
- ;; Restore the frame pointer.
- (move rbp-tn old-fp)
- ;; Clear as much of the stack as possible, but not past the return
- ;; address.
- (inst lea rsp-tn (make-ea :qword :base rbx
- :disp (- (* (max nvals 2) n-word-bytes))))
+ (inst mov rcx (fixnumize nvals)))
;; Pre-default any argument register that need it.
(when (< nvals register-arg-count)
(let* ((arg-tns (nthcdr nvals (list a0 a1 a2)))
;; 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 RBX instead of RBP.
- (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 rsp-tn rbp-tn)
+ (inst pop rbp-tn)
(inst ret))
(t
- (inst jmp (make-ea :qword :base rbx
- :disp (- (* (1+ (tn-offset return-pc))
- n-word-bytes))))))
+ ;; 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 rsp-tn
+ (make-ea :qword :base rbp-tn
+ :disp (frame-byte-offset (1- nvals))))
+ (move rbp-tn old-fp)
+ (inst push (make-ea :qword :base rbx
+ :disp (frame-byte-offset
+ (+ sp->fp-offset
+ (tn-offset return-pc)))))
+ (inst ret)))
(trace-table-entry trace-table-normal)))
;;; assembly-routine.
;;;
;;; The assembly routine takes the following args:
-;;; RAX -- the return-pc to finally jump to.
-;;; RBX -- pointer to where to put the values.
;;; RCX -- number of values to find there.
;;; RSI -- pointer to where to find the values.
(define-vop (return-multiple)
- (:args (old-fp :to (:eval 1) :target old-fp-temp)
- (return-pc :target rax)
+ (:args (old-fp)
+ (return-pc)
(vals :scs (any-reg) :target rsi)
(nvals :scs (any-reg) :target rcx))
-
- (:temporary (:sc unsigned-reg :offset rax-offset :from (:argument 1)) rax)
(:temporary (:sc unsigned-reg :offset rsi-offset :from (:argument 2)) rsi)
(:temporary (:sc unsigned-reg :offset rcx-offset :from (:argument 3)) rcx)
- (:temporary (:sc unsigned-reg :offset rbx-offset :from (:eval 0)) rbx)
(:temporary (:sc unsigned-reg) return-asm)
(: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 rax 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 rsp-tn rbp-tn)
- (move rbp-tn old-fp-temp)
+ ;; Clear the stack until ocfp.
+ (inst mov rsp-tn rbp-tn)
;; clear the multiple-value return flag
(inst clc)
;; Out of here.
- (inst jmp rax)
-
+ (inst pop rbp-tn)
+ (inst ret)
;; Nope, not the single case. Jump to the assembly routine.
(emit-label not-single)))
(move rsi vals)
(move rcx nvals)
- (move rbx rbp-tn)
- (move rbp-tn old-fp)
(inst lea return-asm
(make-ea :qword :disp (make-fixup 'return-multiple
:assembly-routine)))
(inst jmp :be JUST-ALLOC-FRAME)))
;; Allocate the space on the stack.
- ;; stack = rbp - (max 3 frame-size) - (nargs - fixed)
+ ;; stack = rbp + sp->fp-offset - (max 3 frame-size) - (nargs - fixed)
(inst lea rbx-tn
(make-ea :qword :base rbp-tn
- :disp (- (fixnumize fixed)
- (* n-word-bytes
+ :disp (* n-word-bytes
+ (- (+ sp->fp-offset fixed)
(max 3 (sb-allocated-size 'stack))))))
(inst sub rbx-tn rcx-tn) ; Got the new stack in rbx
(inst mov rsp-tn rbx-tn)
(inst sub rcx-tn (fixnumize fixed))))
;; Initialize R8 to be the end of args.
- (inst mov source rbp-tn)
+ (inst lea source (make-ea :qword :base rbp-tn
+ :disp (* sp->fp-offset n-word-bytes)))
(inst sub source rbx-tn)
;; We need to copy from downwards up to avoid overwriting some of
( nil )
;; Store it relative to rbp
(inst mov (make-ea :qword :base rbp-tn
- :disp (- (* n-word-bytes
- (+ 1 (- i fixed)
- (max 3 (sb-allocated-size 'stack))))))
+ :disp (* n-word-bytes
+ (- sp->fp-offset
+ (+ 1
+ (- i fixed)
+ (max 3 (sb-allocated-size
+ 'stack))))))
(nth i *register-arg-tns*))
(incf i)
JUST-ALLOC-FRAME
(inst lea rsp-tn
(make-ea :qword :base rbp-tn
- :disp (- (* n-word-bytes
+ :disp (* n-word-bytes
+ (- sp->fp-offset
(max 3 (sb-allocated-size 'stack))))))
DONE))
(inst lea dst (make-ea :qword :base rcx :index rcx))
(maybe-pseudo-atomic stack-allocate-p
(allocation dst dst node stack-allocate-p list-pointer-lowtag)
- (inst shr rcx (1- n-lowtag-bits))
;; Set decrement mode (successive args at lower addresses)
(inst std)
;; Set up the result.
(inst lods rax)
(storew rax dst 0 list-pointer-lowtag)
;; Go back for more.
- (inst sub rcx 1)
+ (inst sub rcx n-word-bytes)
(inst jmp :nz loop)
;; NIL out the last cons.
(storew nil-value dst 1 list-pointer-lowtag)