X-Git-Url: http://repo.macrolet.net/gitweb/?a=blobdiff_plain;f=src%2Fcompiler%2Fx86%2Fcall.lisp;h=920a2e3c942b85af61906d7ecce926665f030d5c;hb=293488f3b117854e12b0d7f4faeb742b707bbc9c;hp=6caa160b0638e492df9d327ab6d78814539a4757;hpb=80304981972c91c1b3f3fca75f36dacf1fecf307;p=sbcl.git diff --git a/src/compiler/x86/call.lisp b/src/compiler/x86/call.lisp index 6caa160..920a2e3 100644 --- a/src/compiler/x86/call.lisp +++ b/src/compiler/x86/call.lisp @@ -15,11 +15,11 @@ ;;; Return a wired TN describing the N'th full call argument passing ;;; location. -(!def-vm-support-routine standard-argument-location (n) +(!def-vm-support-routine 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 - (nth n *register-arg-offsets*)) + (nth n *register-arg-offsets*)) (make-wired-tn *backend-t-primitive-type* control-stack-sc-number n))) ;;; Make a passing location TN for a local call return PC. @@ -29,10 +29,10 @@ (!def-vm-support-routine 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)) + sap-stack-sc-number return-pc-save-offset)) -;;; Similar to Make-Return-PC-Passing-Location, but makes a location -;;; to pass Old-FP in. +;;; This is similar to MAKE-RETURN-PC-PASSING-LOCATION, but makes a +;;; location to pass OLD-FP in. ;;; ;;; This is wired in both the standard and the local-call conventions, ;;; because we want to be able to assume it's always there. Besides, @@ -41,25 +41,24 @@ (!def-vm-support-routine make-old-fp-passing-location (standard) (declare (ignore standard)) (make-wired-tn *fixnum-primitive-type* control-stack-sc-number - ocfp-save-offset)) + ocfp-save-offset)) -;;; Make the TNs used to hold Old-FP and Return-PC within the current +;;; Make the TNs used to hold OLD-FP and RETURN-PC within the current ;;; function. We treat these specially so that the debugger can find ;;; them at a known location. ;;; ;;; Without using a save-tn - which does not make much sense if it is -;;; wire to the stack? -(!def-vm-support-routine make-old-fp-save-location (env) +;;; wired to the stack? +(!def-vm-support-routine make-old-fp-save-location (physenv) (physenv-debug-live-tn (make-wired-tn *fixnum-primitive-type* - control-stack-sc-number - ocfp-save-offset) - env)) - -(!def-vm-support-routine make-return-pc-save-location (env) + control-stack-sc-number + ocfp-save-offset) + physenv)) +(!def-vm-support-routine 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) - env)) + sap-stack-sc-number return-pc-save-offset) + physenv)) ;;; 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 @@ -82,7 +81,7 @@ ;;; continuation within a function. (!def-vm-support-routine make-unknown-values-locations () (list (make-stack-pointer-tn) - (make-normal-tn *fixnum-primitive-type*))) + (make-normal-tn *fixnum-primitive-type*))) ;;; This function is called by the ENTRY-ANALYZE phase, allowing ;;; VM-dependent initialization of the IR2-COMPONENT structure. We @@ -102,7 +101,7 @@ ;; we'll just live with this ugliness. -- WHN 2002-01-02 (dotimes (i (1+ code-constants-offset)) (vector-push-extend nil - (ir2-component-constants (component-info component)))) + (ir2-component-constants (component-info component)))) (values)) ;;;; frame hackery @@ -120,11 +119,31 @@ (: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))))) + (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. @@ -134,7 +153,7 @@ ;; The start of the actual code. ;; Save the return-pc. - (popw ebp-tn (- (1+ return-pc-save-offset))) + (popw ebp-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 @@ -142,9 +161,10 @@ (unless copy-more-arg-follows ;; The args fit within the frame so just allocate the frame. (inst lea esp-tn - (make-ea :dword :base ebp-tn - :disp (- (* n-word-bytes - (max 3 (sb-allocated-size 'stack))))))) + (make-ea :dword :base ebp-tn + :disp (- (* n-word-bytes + (- (max 3 (sb-allocated-size 'stack)) + sp->fp-offset)))))) (trace-table-entry trace-table-normal))) @@ -152,12 +172,13 @@ ;;; 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)) - (nfp)) + (: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 @@ -166,23 +187,18 @@ ;;; 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 -;;; for a fixed number of values. Values is the head of the TN-Ref +;;; for a fixed number of values. Values is the head of the TN-REF ;;; list for the locations that the values are to be received into. ;;; 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.) @@ -195,170 +211,195 @@ ;;; 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) - (inst mov esp-tn ebx-tn)) - ((<= nvals register-arg-count) - (let ((regs-defaulted (gen-label))) - (note-this-location vop :unknown-return) - (inst jmp-short 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-short 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 (- (1+ 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-short 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 (* (- (1+ register-arg-count)) n-word-bytes))) - ;; 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 (- (1+ 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 (* (- (1+ register-arg-count)) n-word-bytes))) - ;; Save ESI, and compute a pointer to where the args come from. - (storew esi-tn ebx-tn (- (1+ 2))) - (inst lea esi-tn - (make-ea :dword :base ebx-tn - :disp (* (- (1+ register-arg-count)) n-word-bytes))) - ;; 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 (- (1+ 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 (- (1+ 1))) - (inst mov esp-tn ebx-tn)))) + (type unsigned-byte nvals)) + (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 @@ -380,24 +421,37 @@ ;;; 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)) - (done (gen-label))) - (inst jmp-short variable-values) - - (inst mov start esp-tn) - (inst push (first *register-arg-tns*)) - (inst mov count (fixnumize 1)) - (inst jmp done) - - (emit-label variable-values) + (let ((type (sb!c::basic-combination-derived-type node)) + (variable-values (gen-label)) + (stack-values (gen-label)) + (done (gen-label))) + (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) @@ -409,16 +463,45 @@ ;;; handles is allocation of the result temporaries. (define-vop (unknown-values-receiver) (:temporary (:sc descriptor-reg :offset ebx-offset - :from :eval :to (:result 0)) - values-start) + :from :eval :to (:result 0)) + values-start) (:temporary (:sc any-reg :offset ecx-offset - :from :eval :to (:result 1)) - nvals) + :from :eval :to (:result 1)) + nvals) (:results (start :scs (any-reg control-stack)) - (count :scs (any-reg control-stack)))) + (count :scs (any-reg control-stack)))) ;;;; 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 rbp-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. ;;; @@ -440,38 +523,21 @@ ;;; function. (define-vop (call-local) (:args (fp) - (nfp) - (args :more t)) + (nfp) + (args :more t)) (: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 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 (- (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) + (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 @@ -479,40 +545,21 @@ ;;; glob and the number of values received. (define-vop (multiple-call-local unknown-values-receiver) (:args (fp) - (nfp) - (args :more t)) + (nfp) + (args :more t)) (: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 (- (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 + (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 @@ -525,154 +572,41 @@ ;;; we use MAYBE-LOAD-STACK-TN. (define-vop (known-call-local) (:args (fp) - (nfp) - (args :more t)) + (nfp) + (args :more t)) (: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 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 (- (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 + (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) - (vals :more t)) + (return-pc) + (vals :more t)) (:move-args :known-return) (:info val-locs) (: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 (- (* (1+ ocfp-save-offset) - n-word-bytes)))) - ;; Restore the old fp from its save location on the stack, - ;; and zot the stack. - (inst pop ebp-tn)) - - (t - (cerror "Continue any-way" - "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 (- (* (1+ (tn-offset return-pc)) n-word-bytes)))) - ;; 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 @@ -707,194 +641,207 @@ ;;; the last fixed argument. If Variable is false, then the passing ;;; locations are passed as a more arg. Variable is true if there are ;;; a variable number of arguments passed on the stack. Variable -;;; cannot be specified with :Tail return. TR variable argument call +;;; cannot be specified with :TAIL return. TR variable argument call ;;; is implemented separately. ;;; ;;; In tail call with fixed arguments, the passing locations are ;;; passed as a more arg, but there is no new-FP, since the arguments ;;; have been set up in the current frame. (macrolet ((define-full-call (name named return variable) - (aver (not (and variable (eq return :tail)))) - `(define-vop (,name - ,@(when (eq return :unknown) - '(unknown-values-receiver))) - (:args - ,@(unless (eq return :tail) - '((new-fp :scs (any-reg) :to (:argument 1)))) - - (fun :scs (descriptor-reg control-stack) - :target eax :to (:argument 0)) - - ,@(when (eq return :tail) - '((old-fp) - (return-pc))) - - ,@(unless variable '((args :more t :scs (descriptor-reg))))) - - ,@(when (eq return :fixed) - '((:results (values :more t)))) - - (:save-p ,(if (eq return :tail) :compute-only t)) - - ,@(unless (or (eq return :tail) variable) - '((:move-args :full-call))) - - (:vop-var vop) - (:info - ,@(unless (or variable (eq return :tail)) '(arg-locs)) - ,@(unless variable '(nargs)) - ,@(when (eq return :fixed) '(nvals))) - - (:ignore - ,@(unless (or variable (eq return :tail)) '(arg-locs)) - ,@(unless variable '(args))) - - ;; We pass either the fdefn object (for named call) or - ;; the actual function object (for unnamed call) in - ;; EAX. With named call, closure-tramp will replace it - ;; with the real function and invoke the real function - ;; for closures. Non-closures do not need this value, - ;; so don't care what shows up in it. - (:temporary - (:sc descriptor-reg - :offset eax-offset - :from (:argument 0) - :to :eval) - eax) - - ;; We pass the number of arguments in ECX. - (:temporary (:sc unsigned-reg :offset ecx-offset :to :eval) ecx) - - ;; With variable call, we have to load the - ;; register-args out of the (new) stack frame before - ;; doing the call. Therefore, we have to tell the - ;; lifetime stuff that we need to use them. - ,@(when variable - (mapcar (lambda (name offset) - `(:temporary (:sc descriptor-reg - :offset ,offset - :from (:argument 0) - :to :eval) - ,name)) - *register-arg-names* *register-arg-offsets*)) - - ,@(when (eq return :tail) - '((:temporary (:sc unsigned-reg - :from (:argument 1) - :to (:argument 2)) - old-fp-tmp))) - - (:generator ,(+ (if named 5 0) - (if variable 19 1) - (if (eq return :tail) 0 10) - 15 - (if (eq return :unknown) 25 0)) - (trace-table-entry trace-table-call-site) - - ;; This has to be done before the frame pointer is - ;; changed! EAX stores the 'lexical environment' needed - ;; for closures. - (move eax fun) - - - ,@(if variable - ;; For variable call, compute the number of - ;; arguments and move some of the arguments to - ;; registers. - (collect ((noise)) - ;; Compute the number of arguments. - (noise '(inst mov ecx new-fp)) - (noise '(inst sub ecx esp-tn)) - ;; Move the necessary args to registers, - ;; this moves them all even if they are - ;; not all needed. - (loop - for name in *register-arg-names* - for index downfrom -1 - do (noise `(loadw ,name new-fp ,index))) - (noise)) - '((if (zerop nargs) - (inst xor ecx ecx) - (inst mov ecx (fixnumize nargs))))) - ,@(cond ((eq return :tail) - '(;; Python has figured out what frame we should - ;; return to so might as well use that clue. - ;; This seems really important to the - ;; implementation of things like - ;; (without-interrupts ...) - ;; - ;; dtc; Could be doing a tail call from a - ;; known-local-call etc in which the old-fp - ;; or ret-pc are in regs or in non-standard - ;; places. If the passing location were - ;; wired to the stack in standard locations - ;; then these moves will be un-necessary; - ;; this is probably best for the x86. - (sc-case old-fp - ((control-stack) - (unless (= ocfp-save-offset - (tn-offset old-fp)) - ;; FIXME: FORMAT T for stale - ;; diagnostic output (several of - ;; them around here), ick - (format t "** tail-call old-fp not S0~%") - (move old-fp-tmp old-fp) - (storew old-fp-tmp - ebp-tn - (- (1+ ocfp-save-offset))))) - ((any-reg descriptor-reg) - (format t "** tail-call old-fp in reg not S0~%") - (storew old-fp - ebp-tn - (- (1+ ocfp-save-offset))))) - - ;; For tail call, we have to push the - ;; return-pc so that it looks like we CALLed - ;; despite the fact that we are going to JMP. - (inst push return-pc) - )) - (t - ;; For non-tail call, we have to save our - ;; frame pointer and install the new frame - ;; pointer. We can't load stack tns after this - ;; point. - `(;; Python doesn't seem to allocate a frame - ;; here which doesn't leave room for the - ;; ofp/ret stuff. - - ;; The variable args are on the stack and - ;; become the frame, but there may be <3 - ;; args and 3 stack slots are assumed - ;; allocate on the call. So need to ensure - ;; there are at least 3 slots. This hack - ;; just adds 3 more. - ,(if variable - '(inst sub esp-tn (fixnumize 3))) - - ;; Save the fp - (storew ebp-tn new-fp (- (1+ ocfp-save-offset))) - - (move ebp-tn new-fp) ; NB - now on new stack frame. - ))) - - (note-this-location vop :call-site) - - (inst ,(if (eq return :tail) 'jmp 'call) - (make-ea :dword :base eax - :disp ,(if named - '(- (* fdefn-raw-addr-slot - n-word-bytes) - other-pointer-lowtag) - '(- (* closure-fun-slot n-word-bytes) - fun-pointer-lowtag)))) - ,@(ecase return - (:fixed - '((default-unknown-values vop values nvals))) - (:unknown - '((note-this-location vop :unknown-return) - (receive-unknown-values values-start nvals start count))) - (:tail)) - (trace-table-entry trace-table-normal))))) + (aver (not (and variable (eq return :tail)))) + `(define-vop (,name + ,@(when (eq return :unknown) + '(unknown-values-receiver))) + (:args + ,@(unless (eq return :tail) + '((new-fp :scs (any-reg) :to (:argument 1)))) + + (fun :scs (descriptor-reg control-stack) + :target eax :to (:argument 0)) + + ,@(when (eq return :tail) + '((old-fp) + (return-pc))) + + ,@(unless variable '((args :more t :scs (descriptor-reg))))) + + ,@(when (eq return :fixed) + '((:results (values :more t)))) + + (:save-p ,(if (eq return :tail) :compute-only t)) + + ,@(unless (or (eq return :tail) variable) + '((:move-args :full-call))) + + (:vop-var vop) + (:info + ,@(unless (or variable (eq return :tail)) '(arg-locs)) + ,@(unless variable '(nargs)) + ,@(when (eq return :fixed) '(nvals)) + step-instrumenting) + + (:ignore + ,@(unless (or variable (eq return :tail)) '(arg-locs)) + ,@(unless variable '(args))) + + ;; We pass either the fdefn object (for named call) or + ;; the actual function object (for unnamed call) in + ;; EAX. With named call, closure-tramp will replace it + ;; with the real function and invoke the real function + ;; for closures. Non-closures do not need this value, + ;; so don't care what shows up in it. + (:temporary + (:sc descriptor-reg + :offset eax-offset + :from (:argument 0) + :to :eval) + eax) + + ;; We pass the number of arguments in ECX. + (:temporary (:sc unsigned-reg :offset ecx-offset :to :eval) ecx) + + ;; With variable call, we have to load the + ;; register-args out of the (new) stack frame before + ;; doing the call. Therefore, we have to tell the + ;; lifetime stuff that we need to use them. + ,@(when variable + (mapcar (lambda (name offset) + `(:temporary (:sc descriptor-reg + :offset ,offset + :from (:argument 0) + :to :eval) + ,name)) + *register-arg-names* *register-arg-offsets*)) + + ,@(when (eq return :tail) + '((:temporary (:sc unsigned-reg + :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) + (if (eq return :tail) 0 10) + 15 + (if (eq return :unknown) 25 0)) + (trace-table-entry trace-table-call-site) + + ;; This has to be done before the frame pointer is + ;; changed! EAX stores the 'lexical environment' needed + ;; for closures. + (move eax fun) + + + ,@(if variable + ;; For variable call, compute the number of + ;; arguments and move some of the arguments to + ;; registers. + (collect ((noise)) + ;; Compute the number of arguments. + (noise '(inst mov ecx new-fp)) + (noise '(inst sub ecx esp-tn)) + ;; Move the necessary args to registers, + ;; this moves them all even if they are + ;; not all needed. + (loop + for name in *register-arg-names* + for index downfrom -1 + do (noise `(loadw ,name new-fp ,index))) + (noise)) + '((if (zerop nargs) + (inst xor ecx ecx) + (inst mov ecx (fixnumize nargs))))) + ,@(cond ((eq return :tail) + '(;; Python has figured out what frame we should + ;; return to so might as well use that clue. + ;; This seems really important to the + ;; implementation of things like + ;; (without-interrupts ...) + ;; + ;; dtc; Could be doing a tail call from a + ;; known-local-call etc in which the old-fp + ;; or ret-pc are in regs or in non-standard + ;; places. If the passing location were + ;; wired to the stack in standard locations + ;; then these moves will be un-necessary; + ;; this is probably best for the x86. + (sc-case old-fp + ((control-stack) + (unless (= ocfp-save-offset + (tn-offset old-fp)) + ;; FIXME: FORMAT T for stale + ;; diagnostic output (several of + ;; them around here), ick + (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) + (error "** tail-call old-fp in reg not S0~%") + (storew old-fp + ebp-tn + (frame-word-offset ocfp-save-offset)))) + + ;; For tail call, we have to push the + ;; return-pc so that it looks like we CALLed + ;; despite the fact that we are going to JMP. + (inst push return-pc) + )) + (t + ;; For non-tail call, we have to save our + ;; frame pointer and install the new frame + ;; pointer. We can't load stack tns after this + ;; point. + `(;; Python doesn't seem to allocate a frame + ;; here which doesn't leave room for the + ;; ofp/ret stuff. + + ;; The variable args are on the stack and + ;; become the frame, but there may be <3 + ;; args and 3 stack slots are assumed + ;; allocate on the call. So need to ensure + ;; there are at least 3 slots. This hack + ;; just adds 3 more. + ,(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)) + + (move ebp-tn new-fp) ; NB - now on new stack frame. + ))) + + (when step-instrumenting + (emit-single-step-test) + (inst jmp :eq DONE) + (inst break single-step-around-trap)) + DONE + + (note-this-location vop :call-site) + + (inst ,(if (eq return :tail) 'jmp 'call) + ,(if named + '(make-ea-for-object-slot eax fdefn-raw-addr-slot + other-pointer-lowtag) + '(make-ea-for-object-slot eax closure-fun-slot + fun-pointer-lowtag))) + ,@(ecase return + (:fixed + '((default-unknown-values vop values nvals node))) + (:unknown + '((note-this-location vop :unknown-return) + (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) @@ -911,60 +858,46 @@ ;;; routine. We just set things up so that it can find what it needs. (define-vop (tail-call-variable) (:args (args :scs (any-reg control-stack) :target esi) - (function :scs (descriptor-reg control-stack) :target eax) - (old-fp) - (ret-addr)) + (function :scs (descriptor-reg control-stack) :target eax) + (old-fp) + (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. +;;; +;;; 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)) - (:temporary (:sc unsigned-reg) ofp) - (:temporary (:sc unsigned-reg) ret) + (return-pc) + (value)) (:ignore value) (:generator 6 + (check-ocfp-and-return-pc old-fp return-pc) (trace-table-entry trace-table-fun-epilogue) - (move ret return-pc) - ;; Clear the control stack - (move ofp old-fp) - ;; Adjust the return address for the single value return. - (inst add ret 2) - ;; Restore the frame pointer. - (move esp-tn ebp-tn) - (move ebp-tn ofp) - ;; Out of here. - (inst jmp ret))) + ;; 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 @@ -975,60 +908,68 @@ ;;; the values, and jump directly to return-pc. (define-vop (return) (:args (old-fp) - (return-pc :to (:eval 1)) - (values :more t)) + (return-pc :to (:eval 1)) + (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. (:temporary (:sc unsigned-reg :offset (first *register-arg-offsets*) - :from :eval) a0) + :from :eval) a0) (:temporary (:sc unsigned-reg :offset (second *register-arg-offsets*) - :from :eval) a1) + :from :eval) a1) (:temporary (:sc unsigned-reg :offset (third *register-arg-offsets*) - :from :eval) a2) + :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))) - (first (first arg-tns))) - (inst mov first nil-value) - (dolist (tn (cdr arg-tns)) - (inst mov tn first)))) + (first (first arg-tns))) + (inst mov first nil-value) + (dolist (tn (cdr arg-tns)) + (inst mov tn first)))) + ;; 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? - (inst ret)) - (t - (inst jmp (make-ea :dword :base ebx - :disp (- (* (1+ (tn-offset return-pc)) - n-word-bytes)))))) + (cond ((<= nvals register-arg-count) + (inst mov esp-tn ebp-tn) + (inst pop ebp-tn) + (inst ret)) + (t + ;; 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))) @@ -1039,53 +980,39 @@ ;;; 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) - (vals :scs (any-reg) :target esi) - (nvals :scs (any-reg) :target ecx)) - - (:temporary (:sc unsigned-reg :offset eax-offset :from (:argument 1)) eax) + (:args (old-fp) + (return-pc) + (vals :scs (any-reg) :target esi) + (nvals :scs (any-reg) :target ecx)) (: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) + :from (:eval 0)) a0) (: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) - ;; Fix the return-pc to point at the single-value entry point. - (inst add eax 2) - ;; Out of here. - (inst jmp eax) - - ;; Nope, not the single case. Jump to the assembly routine. - (emit-label not-single))) + (inst cmp nvals (fixnumize 1)) + (inst jmp :ne not-single) + ;; Return with one value. + (loadw a0 vals -1) + ;; Clear the stack until ocfp. + (inst mov esp-tn ebp-tn) + ;; clear the multiple-value return flag + (inst clc) + ;; Out of here. + (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))) @@ -1118,7 +1045,7 @@ ;;; ;;; 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. @@ -1133,18 +1060,18 @@ (:generator 20 ;; Avoid the copy if there are no more args. (cond ((zerop fixed) - (inst jecxz just-alloc-frame)) - (t - (inst cmp ecx-tn (fixnumize fixed)) - (inst jmp :be just-alloc-frame))) + (inst jecxz JUST-ALLOC-FRAME)) + (t + (inst cmp ecx-tn (fixnumize fixed)) + (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) (inst lea ebx-tn - (make-ea :dword :base ebp-tn - :disp (- (fixnumize fixed) - (* n-word-bytes - (max 3 (sb-allocated-size 'stack)))))) + (make-ea :dword :base ebp-tn + :disp (* n-word-bytes + (- (+ sp->fp-offset fixed) + (max 3 (sb-allocated-size 'stack)))))) (inst sub ebx-tn ecx-tn) ; Got the new stack in ebx (inst mov esp-tn ebx-tn) @@ -1154,37 +1081,47 @@ (inst mov ebx-tn ecx-tn) (cond ((< fixed register-arg-count) - ;; We must stop when we run out of stack args, not when we - ;; run out of more args. - ;; Number to copy = nargs-3 - (inst sub ecx-tn (fixnumize register-arg-count)) - ;; Everything of interest in registers. - (inst jmp :be do-regs)) - (t - ;; Number to copy = nargs-fixed - (inst sub ecx-tn (fixnumize fixed)))) + ;; We must stop when we run out of stack args, not when we + ;; run out of more args. + ;; Number to copy = nargs-3 + (inst sub ecx-tn (fixnumize register-arg-count)) + ;; Everything of interest in registers. + (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 dst to be end of stack; skiping the values pushed - ;; above. - (inst lea edi-tn (make-ea :dword :base esp-tn :disp 8)) - ;; Initialize src to be end of args. - (inst mov esi-tn ebp-tn) + (inst lea esi-tn (make-ea :dword :base ebp-tn + :disp (* sp->fp-offset n-word-bytes))) (inst sub esi-tn ebx-tn) - (inst shr ecx-tn word-shift) ; make word count - ;; And copy the args. - (inst cld) ; auto-inc ESI and EDI. - (inst rep) - (inst movs :dword) + ;; 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) @@ -1195,73 +1132,80 @@ ;; 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 - (max 3 (sb-allocated-size 'stack)))))) + (make-ea :dword :base ebp-tn + :disp (* n-word-bytes + (- sp->fp-offset + (max 3 (sb-allocated-size 'stack)))))) DONE)) -;;; &MORE args are stored contiguously on the stack, starting -;;; immediately at the context pointer. The context pointer is not -;;; typed, so the lowtag is 0. -(define-vop (more-arg) - (:translate %more-arg) +(define-vop (more-kw-arg) + (:translate sb!c::%more-kw-arg) (:policy :fast-safe) - (:args (object :scs (descriptor-reg) :to :result) - (index :scs (any-reg) :target temp)) + (:args (object :scs (descriptor-reg) :to (:result 1)) + (index :scs (any-reg immediate) :to (:result 1) :target keyword)) (:arg-types * tagged-num) - (:temporary (:sc unsigned-reg :from (:argument 1) :to :result) temp) - (:results (value :scs (any-reg descriptor-reg))) - (:result-types *) - (:generator 5 - (move temp index) - (inst neg temp) - (inst mov value (make-ea :dword :base object :index temp)))) + (:results (value :scs (descriptor-reg any-reg)) + (keyword :scs (descriptor-reg any-reg))) + (:result-types * *) + (:generator 4 + (sc-case index + (immediate + (inst mov value (make-ea :dword :base object :disp (tn-value index))) + (inst mov keyword (make-ea :dword :base object + :disp (+ (tn-value index) n-word-bytes)))) + (t + (inst mov value (make-ea :dword :base object :index index)) + (inst mov keyword (make-ea :dword :base object :index index + :disp n-word-bytes)))))) -(define-vop (more-arg-c) - (:translate %more-arg) +(define-vop (more-arg) + (:translate sb!c::%more-arg) (:policy :fast-safe) - (:args (object :scs (descriptor-reg))) - (:info index) - (:arg-types * (:constant (signed-byte 30))) - (:results (value :scs (any-reg descriptor-reg))) + (:args (object :scs (descriptor-reg) :to (:result 1)) + (index :scs (any-reg) :to (:result 1) :target value)) + (:arg-types * tagged-num) + (:results (value :scs (descriptor-reg any-reg))) (:result-types *) (:generator 4 - (inst mov value - (make-ea :dword :base object :disp (- (* index n-word-bytes)))))) - + (move value index) + (inst neg value) + (inst mov value (make-ea :dword :base object :index value)))) ;;; Turn more arg (context, count) into a list. (define-vop (listify-rest-args) (:translate %listify-rest-args) (:policy :safe) (:args (context :scs (descriptor-reg) :target src) - (count :scs (any-reg) :target ecx)) + (count :scs (any-reg) :target ecx)) (:arg-types * tagged-num) (:temporary (:sc unsigned-reg :offset esi-offset :from (:argument 0)) src) (:temporary (:sc unsigned-reg :offset ecx-offset :from (:argument 1)) ecx) @@ -1271,25 +1215,22 @@ (:node-var node) (:generator 20 (let ((enter (gen-label)) - (loop (gen-label)) - (done (gen-label))) + (loop (gen-label)) + (done (gen-label)) + (stack-allocate-p (node-stack-allocate-p node))) (move src context) (move ecx count) ;; Check to see whether there are no args, and just return NIL if so. (inst mov result nil-value) (inst jecxz done) - (inst lea dst (make-ea :dword :index ecx :scale 2)) - (pseudo-atomic - (allocation dst dst node) - (inst lea dst (make-ea :byte :base dst :disp list-pointer-lowtag)) - ;; Convert the count into a raw value, so that we can use the - ;; LOOP instruction. - (inst shr ecx 2) + (inst lea dst (make-ea :dword :base ecx :index ecx)) + (maybe-pseudo-atomic stack-allocate-p + (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) @@ -1302,9 +1243,11 @@ (inst lods eax) (storew eax dst 0 list-pointer-lowtag) ;; Go back for more. - (inst loop loop) + (inst sub ecx n-word-bytes) + (inst jmp :nz loop) ;; NIL out the last cons. - (storew nil-value dst 1 list-pointer-lowtag)) + (storew nil-value dst 1 list-pointer-lowtag) + (inst cld)) (emit-label done)))) ;;; Return the location and size of the &MORE arg glob created by @@ -1324,7 +1267,7 @@ (:arg-types positive-fixnum (:constant fixnum)) (:info fixed) (:results (context :scs (descriptor-reg)) - (count :scs (any-reg))) + (count :scs (any-reg))) (:result-types t tagged-num) (:note "more-arg-context") (:generator 5 @@ -1332,8 +1275,8 @@ ;; SP at this point points at the last arg pushed. ;; 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)))) + :index count :scale 1 + :disp (- (+ (fixnumize fixed) n-word-bytes)))) (unless (zerop fixed) (inst sub count (fixnumize fixed))))) @@ -1348,25 +1291,25 @@ (: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))) + (inst test nargs nargs) ; smaller instruction + (inst cmp nargs (fixnumize count))) (inst jmp :ne err-lab)))) ;;; Various other error signallers. (macrolet ((def (name error translate &rest args) - `(define-vop (,name) - ,@(when translate - `((:policy :fast-safe) - (:translate ,translate))) - (:args ,@(mapcar (lambda (arg) - `(,arg :scs (any-reg descriptor-reg))) - args)) - (:vop-var vop) - (:save-p :compute-only) - (:generator 1000 - (error-call vop ,error ,@args))))) + `(define-vop (,name) + ,@(when translate + `((:policy :fast-safe) + (:translate ,translate))) + (:args ,@(mapcar (lambda (arg) + `(,arg :scs (any-reg descriptor-reg))) + args)) + (:vop-var vop) + (:save-p :compute-only) + (:generator 1000 + (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 @@ -1378,3 +1321,32 @@ (def unknown-key-arg-error unknown-key-arg-error sb!c::%unknown-key-arg-error key) (def nil-fun-returned-error nil-fun-returned-error nil fun)) + +;;; Single-stepping + +(defun emit-single-step-test () + ;; We use different ways of representing whether stepping is on on + ;; +SB-THREAD / -SB-THREAD: on +SB-THREAD, we use a slot in the + ;; thread structure. On -SB-THREAD we use the value of a static + ;; symbol. Things are done this way, since reading a thread-local + ;; slot from a symbol would require an extra register on +SB-THREAD, + ;; and reading a slot from a thread structure would require an extra + ;; register on -SB-THREAD. + #!+sb-thread + (progn + (inst cmp (make-ea :dword + :disp (* thread-stepping-slot n-word-bytes)) + nil-value :fs)) + #!-sb-thread + (inst cmp (make-ea-for-symbol-value sb!impl::*stepping*) + nil-value)) + +(define-vop (step-instrument-before-vop) + (:policy :fast-safe) + (:vop-var vop) + (:generator 3 + (emit-single-step-test) + (inst jmp :eq DONE) + (inst break single-step-before-trap) + DONE + (note-this-location vop :step-before-vop)))