vec))
) ; EVAL-WHEN
+;;; SSE shuffle patterns. The names end in the number of bits of the
+;;; immediate byte that are used to encode the pattern and the radix
+;;; in which to print the value.
+(macrolet ((define-sse-shuffle-arg-type (name format-string)
+ `(sb!disassem:define-arg-type ,name
+ :type 'imm-byte
+ :printer (lambda (value stream dstate)
+ (declare (type (unsigned-byte 8) value)
+ (type stream stream)
+ (ignore dstate))
+ (format stream ,format-string value)))))
+ (define-sse-shuffle-arg-type sse-shuffle-pattern-2-2 "#b~2,'0B")
+ (define-sse-shuffle-arg-type sse-shuffle-pattern-8-4 "#4r~4,4,'0R"))
+
;;; Set assembler parameters. (In CMU CL, this was done with
;;; a call to a macro DEF-ASSEMBLER-PARAMS.)
(eval-when (:compile-toplevel :load-toplevel :execute)
(op :field (byte 8 8))
(reg/mem :fields (list (byte 2 22) (byte 3 16))
:type 'xmmreg/mem)
- (reg :field (byte 3 19) :type 'xmmreg))
+ (reg :field (byte 3 19) :type 'xmmreg)
+ ;; optional fields
+ (imm))
(sb!disassem:define-instruction-format (rex-xmm-xmm/mem 32
:default-printer
(op :field (byte 8 16))
(reg/mem :fields (list (byte 2 30) (byte 3 24))
:type 'xmmreg/mem)
- (reg :field (byte 3 27) :type 'xmmreg))
+ (reg :field (byte 3 27) :type 'xmmreg)
+ (imm))
(sb!disassem:define-instruction-format (ext-xmm-xmm/mem 32
:default-printer
(op :field (byte 8 16))
(reg/mem :fields (list (byte 2 30) (byte 3 24))
:type 'xmmreg/mem)
- (reg :field (byte 3 27) :type 'xmmreg))
+ (reg :field (byte 3 27) :type 'xmmreg)
+ (imm))
(sb!disassem:define-instruction-format (ext-rex-xmm-xmm/mem 40
:default-printer
(op :field (byte 8 24))
(reg/mem :fields (list (byte 2 38) (byte 3 32))
:type 'xmmreg/mem)
- (reg :field (byte 3 35) :type 'xmmreg))
+ (reg :field (byte 3 35) :type 'xmmreg)
+ (imm))
;;; Same as xmm-xmm/mem etc., but with direction bit.
:type 'imm-byte
:printer *sse-conditions*)
-;;; XMM instructions with 8 bit immediate data
-
-(sb!disassem:define-instruction-format (xmm-xmm/mem-imm 24
- :default-printer
- '(:name
- :tab reg ", " reg/mem ", " imm))
- (x0f :field (byte 8 0) :value #x0f)
- (op :field (byte 8 8))
- (reg/mem :fields (list (byte 2 22) (byte 3 16))
- :type 'xmmreg/mem)
- (reg :field (byte 3 19) :type 'xmmreg)
- (imm :type 'imm-byte))
-
-(sb!disassem:define-instruction-format (rex-xmm-xmm/mem-imm 32
- :default-printer
- '(:name
- :tab reg ", " reg/mem ", " imm))
- (rex :field (byte 4 4) :value #b0100)
- (wrxb :field (byte 4 0) :type 'wrxb)
- (x0f :field (byte 8 8) :value #x0f)
- (op :field (byte 8 16))
- (reg/mem :fields (list (byte 2 30) (byte 3 24))
- :type 'xmmreg/mem)
- (reg :field (byte 3 27) :type 'xmmreg)
- (imm :type 'imm-byte))
-
-(sb!disassem:define-instruction-format (ext-xmm-xmm/mem-imm 32
- :default-printer
- '(:name
- :tab reg ", " reg/mem ", " imm))
- (prefix :field (byte 8 0))
- (x0f :field (byte 8 8) :value #x0f)
- (op :field (byte 8 16))
- (reg/mem :fields (list (byte 2 30) (byte 3 24))
- :type 'xmmreg/mem)
- (reg :field (byte 3 27) :type 'xmmreg)
- (imm :type 'imm-byte))
-
-(sb!disassem:define-instruction-format (ext-rex-xmm-xmm/mem-imm 40
- :default-printer
- '(:name
- :tab reg ", " reg/mem ", " imm))
- (prefix :field (byte 8 0))
- (rex :field (byte 4 12) :value #b0100)
- (wrxb :field (byte 4 8) :type 'wrxb)
- (x0f :field (byte 8 16) :value #x0f)
- (op :field (byte 8 24))
- (reg/mem :fields (list (byte 2 38) (byte 3 32))
- :type 'xmmreg/mem)
- (reg :field (byte 3 35) :type 'xmmreg)
- (imm :type 'imm-byte))
-
(sb!disassem:define-instruction-format (string-op 8
:include 'simple
:default-printer '(:name width)))
(macrolet
((define-imm-sse-instruction (name opcode /i)
`(define-instruction ,name (segment dst/src imm)
- (:printer ext-rex-xmm-imm ((prefix #x66) (op ,opcode) (/i ,/i)))
- (:printer ext-xmm-imm ((prefix #x66) (op ,opcode) (/i ,/i)))
+ (:printer-list
+ ',(sse-inst-printer-list 'xmm-imm #x66 opcode
+ :more-fields `((/i ,/i))))
(:emitter
(emit-sse-inst-with-imm segment dst/src imm
#x66 ,opcode ,/i
:operand-size :do-not-set)))))
(define-imm-sse-instruction pslldq #x73 7)
- (define-imm-sse-instruction psllw #x71 6)
- (define-imm-sse-instruction pslld #x72 6)
- (define-imm-sse-instruction psllq #x73 6)
+ (define-imm-sse-instruction psllw-imm #x71 6)
+ (define-imm-sse-instruction pslld-imm #x72 6)
+ (define-imm-sse-instruction psllq-imm #x73 6)
(define-imm-sse-instruction psraw-imm #x71 4)
(define-imm-sse-instruction psrad-imm #x72 4)
(define-imm-sse-instruction psrldq #x73 3)
- (define-imm-sse-instruction psrlw #x71 2)
- (define-imm-sse-instruction psrld #x72 2)
- (define-imm-sse-instruction psrlq #x73 2))
+ (define-imm-sse-instruction psrlw-imm #x71 2)
+ (define-imm-sse-instruction psrld-imm #x72 2)
+ (define-imm-sse-instruction psrlq-imm #x73 2))
;;; Emit an SSE instruction that has an XMM register as the destination
;;; operand and for which the size of the operands is implicitly given
(macrolet ((define-regular-sse-inst (name prefix opcode)
`(define-instruction ,name (segment dst src)
- ,@(if prefix
- `((:printer ext-xmm-xmm/mem
- ((prefix ,prefix) (op ,opcode)))
- (:printer ext-rex-xmm-xmm/mem
- ((prefix ,prefix) (op ,opcode))))
- `((:printer xmm-xmm/mem ((op ,opcode)))
- (:printer rex-xmm-xmm/mem ((op ,opcode)))))
+ (:printer-list
+ ',(sse-inst-printer-list 'xmm-xmm/mem prefix opcode))
(:emitter
(emit-regular-sse-inst segment dst src ,prefix ,opcode)))))
;; logical
(define-regular-sse-inst mulps nil #x59)
(define-regular-sse-inst mulsd #xf2 #x59)
(define-regular-sse-inst mulss #xf3 #x59)
- (define-regular-sse-inst rccps nil #x53)
+ (define-regular-sse-inst rcpps nil #x53)
(define-regular-sse-inst rcpss #xf3 #x53)
(define-regular-sse-inst rsqrtps nil #x52)
(define-regular-sse-inst rsqrtss #xf3 #x52)
+ (define-regular-sse-inst sqrtpd #x66 #x51)
(define-regular-sse-inst sqrtps nil #x51)
(define-regular-sse-inst sqrtsd #xf2 #x51)
(define-regular-sse-inst sqrtss #xf3 #x51)
(define-regular-sse-inst paddsb #x66 #xec)
(define-regular-sse-inst paddsw #x66 #xed)
(define-regular-sse-inst paddusb #x66 #xdc)
- (define-regular-sse-inst padduwb #x66 #xdd)
+ (define-regular-sse-inst paddusw #x66 #xdd)
(define-regular-sse-inst pavgb #x66 #xe0)
(define-regular-sse-inst pavgw #x66 #xe3)
(define-regular-sse-inst pmaddwd #x66 #xf5)
(define-regular-sse-inst pmullw #x66 #xd5)
(define-regular-sse-inst pmuludq #x66 #xf4)
(define-regular-sse-inst psadbw #x66 #xf6)
+ (define-regular-sse-inst psllw #x66 #xf1)
+ (define-regular-sse-inst pslld #x66 #xf2)
+ (define-regular-sse-inst psllq #x66 #xf3)
(define-regular-sse-inst psraw #x66 #xe1)
(define-regular-sse-inst psrad #x66 #xe2)
+ (define-regular-sse-inst psrlw #x66 #xd1)
+ (define-regular-sse-inst psrld #x66 #xd2)
+ (define-regular-sse-inst psrlq #x66 #xd3)
(define-regular-sse-inst psubb #x66 #xf8)
(define-regular-sse-inst psubw #x66 #xf9)
(define-regular-sse-inst psubd #x66 #xfa)
(define-regular-sse-inst psubq #x66 #xfb)
- (define-regular-sse-inst psubsb #x66 #xd8)
- (define-regular-sse-inst psubsw #x66 #xd9)
+ (define-regular-sse-inst psubsb #x66 #xe8)
+ (define-regular-sse-inst psubsw #x66 #xe9)
+ (define-regular-sse-inst psubusb #x66 #xd8)
+ (define-regular-sse-inst psubusw #x66 #xd9)
;; conversion
(define-regular-sse-inst cvtdq2pd #xf3 #xe6)
(define-regular-sse-inst cvtdq2ps nil #x5b)
(define-regular-sse-inst cvtss2sd #xf3 #x5a)
(define-regular-sse-inst cvttpd2dq #x66 #xe6)
(define-regular-sse-inst cvttps2dq #xf3 #x5b)
- ;; moves
- (define-regular-sse-inst movntdq #x66 #xe7)
- (define-regular-sse-inst movntpd #x66 #x2b)
- (define-regular-sse-inst movntps nil #x2b)
;; integer
(define-regular-sse-inst packsswb #x66 #x63)
(define-regular-sse-inst packssdw #x66 #x6b)
+ (define-regular-sse-inst packuswb #x66 #x67)
(define-regular-sse-inst punpckhbw #x66 #x68)
(define-regular-sse-inst punpckhwd #x66 #x69)
(define-regular-sse-inst punpckhdq #x66 #x6a)
(define-regular-sse-inst punpckldq #x66 #x62)
(define-regular-sse-inst punpcklqdq #x66 #x6c))
-(macrolet ((define-xmm-shuffle-sse-inst (name prefix opcode)
- `(define-instruction ,name (segment dst src pattern)
- ,@(if prefix
- `((:printer ext-xmm-xmm/mem-imm ; suboptimal
- ((prefix ,prefix) (op ,opcode)))
- (:printer ext-rex-xmm-xmm/mem-imm
- ((prefix ,prefix) (op ,opcode))))
- `((:printer xmm-xmm/mem-imm ((op ,opcode)))
- (:printer rex-xmm-xmm/mem-imm ((op ,opcode)))))
- (:emitter
- (aver (typep pattern '(unsigned-byte 8)))
- (emit-regular-sse-inst segment dst src ,prefix ,opcode
- :remaining-bytes 1)
- (emit-byte segment pattern)))))
- (define-xmm-shuffle-sse-inst pshufd #x66 #x70)
- (define-xmm-shuffle-sse-inst pshufhw #xf3 #x70)
- (define-xmm-shuffle-sse-inst pshuflw #xf2 #x70)
- (define-xmm-shuffle-sse-inst shufpd #x66 #xc6)
- (define-xmm-shuffle-sse-inst shufps nil #xc6))
+(macrolet ((define-xmm-shuffle-sse-inst (name prefix opcode n-bits radix)
+ (let ((shuffle-pattern
+ (intern (format nil "SSE-SHUFFLE-PATTERN-~D-~D"
+ n-bits radix))))
+ `(define-instruction ,name (segment dst src pattern)
+ (:printer-list
+ ',(sse-inst-printer-list
+ 'xmm-xmm/mem prefix opcode
+ :more-fields `((imm nil :type ,shuffle-pattern))
+ :printer '(:name :tab reg ", " reg/mem ", " imm)))
+
+ (:emitter
+ (aver (typep pattern '(unsigned-byte ,n-bits)))
+ (emit-regular-sse-inst segment dst src ,prefix ,opcode
+ :remaining-bytes 1)
+ (emit-byte segment pattern))))))
+ (define-xmm-shuffle-sse-inst pshufd #x66 #x70 8 4)
+ (define-xmm-shuffle-sse-inst pshufhw #xf3 #x70 8 4)
+ (define-xmm-shuffle-sse-inst pshuflw #xf2 #x70 8 4)
+ (define-xmm-shuffle-sse-inst shufpd #x66 #xc6 2 2)
+ (define-xmm-shuffle-sse-inst shufps nil #xc6 8 4))
;; MASKMOVDQU (dst is DS:RDI)
(define-instruction maskmovdqu (segment src mask)
- (:printer ext-xmm-xmm/mem
- ((prefix #x66) (op #xf7)))
- (:printer ext-rex-xmm-xmm/mem
- ((prefix #x66) (op #xf7)))
+ (:printer-list
+ (sse-inst-printer-list 'xmm-xmm/mem #x66 #xf7))
(:emitter
(aver (xmm-register-p src))
(aver (xmm-register-p mask))
`(define-instruction ,name (segment op x y)
(:printer-list
',(sse-inst-printer-list
- 'xmm-xmm/mem-imm prefix opcode
+ 'xmm-xmm/mem prefix opcode
:more-fields '((imm nil :type sse-condition-code))
:printer `(,name-prefix imm ,name-suffix
:tab reg ", " reg/mem)))
;;; MOVSD, MOVSS
(macrolet ((define-movsd/ss-sse-inst (name prefix)
`(define-instruction ,name (segment dst src)
- (:printer ext-xmm-xmm/mem-dir ((prefix ,prefix)
- (op #b0001000)))
- (:printer ext-rex-xmm-xmm/mem-dir ((prefix ,prefix)
- (op #b0001000)))
+ (:printer-list
+ ',(sse-inst-printer-list 'xmm-xmm/mem-dir
+ prefix #b0001000))
(:emitter
(cond ((xmm-register-p dst)
(emit-sse-inst segment dst src ,prefix #x10
(:emitter
(aver (xmm-register-p dst))
(aver (xmm-register-p src))
- (emit-regular-sse-inst segment dst src ,prefix ,opcode-from))))
+ (emit-regular-sse-inst segment dst src
+ ,prefix ,opcode-from))))
(define-instruction ,name (segment dst src)
- ,@(if prefix
- `((:printer ext-xmm-xmm/mem
- ((prefix ,prefix) (op ,opcode-from)))
- (:printer ext-rex-xmm-xmm/mem
- ((prefix ,prefix) (op ,opcode-from)))
- (:printer ext-xmm-xmm/mem
- ((prefix ,prefix) (op ,opcode-to))
- '(:name :tab reg/mem ", " reg))
- (:printer ext-rex-xmm-xmm/mem
- ((prefix ,prefix) (op ,opcode-to))
- '(:name :tab reg/mem ", " reg)))
- `((:printer xmm-xmm/mem
- ((op ,opcode-from)))
- (:printer rex-xmm-xmm/mem
- ((op ,opcode-from)))
- (:printer xmm-xmm/mem
- ((op ,opcode-to))
- '(:name :tab reg/mem ", " reg))
- (:printer rex-xmm-xmm/mem
- ((op ,opcode-to))
- '(:name :tab reg/mem ", " reg))))
+ (:printer-list
+ '(,@(when opcode-from
+ (sse-inst-printer-list
+ 'xmm-xmm/mem prefix opcode-from))
+ ,@(sse-inst-printer-list
+ 'xmm-xmm/mem prefix opcode-to
+ :printer '(:name :tab reg/mem ", " reg))))
(:emitter
- (cond ((xmm-register-p dst)
- ,(when force-to-mem
- `(aver (not (or (register-p src)
- (xmm-register-p src)))))
- (emit-regular-sse-inst segment dst src ,prefix ,opcode-from))
+ (cond ,@(when opcode-from
+ `(((xmm-register-p dst)
+ ,(when force-to-mem
+ `(aver (not (or (register-p src)
+ (xmm-register-p src)))))
+ (emit-regular-sse-inst
+ segment dst src ,prefix ,opcode-from))))
(t
(aver (xmm-register-p src))
,(when force-to-mem
`(aver (not (or (register-p dst)
(xmm-register-p dst)))))
- (emit-regular-sse-inst segment src dst ,prefix ,opcode-to))))))))
+ (emit-regular-sse-inst segment src dst
+ ,prefix ,opcode-to))))))))
;; direction bit?
(define-mov-sse-inst movapd #x66 #x28 #x29)
(define-mov-sse-inst movaps nil #x28 #x29)
(define-mov-sse-inst movdqa #x66 #x6f #x7f)
(define-mov-sse-inst movdqu #xf3 #x6f #x7f)
+ ;; streaming
+ (define-mov-sse-inst movntdq #x66 nil #xe7 :force-to-mem t)
+ (define-mov-sse-inst movntpd #x66 nil #x2b :force-to-mem t)
+ (define-mov-sse-inst movntps nil nil #x2b :force-to-mem t)
+
;; use movhps for movlhps and movlps for movhlps
(define-mov-sse-inst movhpd #x66 #x16 #x17 :force-to-mem t)
(define-mov-sse-inst movhps nil #x16 #x17 :reg-reg-name movlhps)
;;; MOVQ
(define-instruction movq (segment dst src)
- (:printer ext-xmm-xmm/mem ((prefix #xf3) (op #x7e)))
- (:printer ext-rex-xmm-xmm/mem ((prefix #xf3) (op #x7e)))
- (:printer ext-xmm-xmm/mem ((prefix #x66) (op #xd6))
- '(:name :tab reg/mem ", " reg))
- (:printer ext-rex-xmm-xmm/mem ((prefix #x66) (op #xd6))
- '(:name :tab reg/mem ", " reg))
+ (:printer-list
+ (append
+ (sse-inst-printer-list 'xmm-xmm/mem #xf3 #x7e)
+ (sse-inst-printer-list 'xmm-xmm/mem #x66 #xd6
+ :printer '(:name :tab reg/mem ", " reg))))
(:emitter
(cond ((xmm-register-p dst)
(emit-sse-inst segment dst src #xf3 #x7e
;;; with zero extension or vice versa.
;;; We do not support the MMX version of this instruction.
(define-instruction movd (segment dst src)
- (:printer ext-xmm-reg/mem ((prefix #x66) (op #x6e)))
- (:printer ext-rex-xmm-reg/mem ((prefix #x66) (op #x6e)))
- (:printer ext-xmm-reg/mem ((prefix #x66) (op #x7e))
- '(:name :tab reg/mem ", " reg))
- (:printer ext-rex-xmm-reg/mem ((prefix #x66) (op #x7e))
- '(:name :tab reg/mem ", " reg))
+ (:printer-list
+ (append
+ (sse-inst-printer-list 'xmm-reg/mem #x66 #x6e)
+ (sse-inst-printer-list 'xmm-reg/mem #x66 #x7e
+ :printer '(:name :tab reg/mem ", " reg))))
(:emitter
(cond ((xmm-register-p dst)
(emit-sse-inst segment dst src #x66 #x6e))
(aver (xmm-register-p src))
(emit-sse-inst segment src dst #x66 #x7e)))))
+(define-instruction pinsrw (segment dst src imm)
+ (:printer-list
+ (sse-inst-printer-list
+ 'xmm-reg/mem #x66 #xc4
+ :more-fields '((imm nil :type imm-byte))
+ :printer '(:name :tab reg ", " reg/mem ", " imm)))
+ (:emitter
+ (aver (xmm-register-p dst))
+ (let ((src-size (operand-size src)))
+ (aver (or (not (register-p src))
+ (eq src-size :qword) (eq src-size :dword)))
+ (emit-sse-inst segment dst src #x66 #xc4
+ :operand-size (if (register-p src) src-size :do-not-set)
+ :remaining-bytes 1))
+ (emit-byte segment imm)))
+
+(define-instruction pextrw (segment dst src imm)
+ (:printer-list
+ (sse-inst-printer-list
+ 'reg-xmm/mem #x66 #xc5
+ :more-fields '((imm nil :type imm-byte))
+ :printer '(:name :tab reg ", " reg/mem ", " imm)))
+ (:emitter
+ (aver (xmm-register-p src))
+ (aver (register-p dst))
+ (let ((dst-size (operand-size dst)))
+ (aver (or (eq dst-size :qword) (eq dst-size :dword)))
+ (emit-sse-inst segment dst src #x66 #xc5
+ :operand-size dst-size
+ :remaining-bytes 1))
+ (emit-byte segment imm)))
+
(macrolet ((define-integer-source-sse-inst (name prefix opcode &key mem-only)
`(define-instruction ,name (segment dst src)
- ,@(if prefix
- `((:printer ext-xmm-reg/mem ((prefix ,prefix) (op ,opcode)))
- (:printer ext-rex-xmm-reg/mem ((prefix ,prefix) (op ,opcode))))
- `((:printer xmm-reg/mem ((op ,opcode)))
- (:printer rex-xmm-reg/mem ((op ,opcode)))))
-
+ (:printer-list
+ ',(sse-inst-printer-list 'xmm-reg/mem prefix opcode))
(:emitter
(aver (xmm-register-p dst))
,(when mem-only
(macrolet ((define-gpr-destination-sse-inst (name prefix opcode &key reg-only)
`(define-instruction ,name (segment dst src)
- ,@(if prefix
- `((:printer ext-reg-xmm/mem ((prefix ,prefix) (op ,opcode)))
- (:printer ext-rex-reg-xmm/mem ((prefix ,prefix) (op ,opcode))))
- `((:printer reg-xmm/mem ((op ,opcode)))
- (:printer rex-reg-xmm/mem ((op ,opcode)))))
+ (:printer-list
+ ',(sse-inst-printer-list 'reg-xmm/mem prefix opcode))
(:emitter
(aver (register-p dst))
,(when reg-only
(aver (eq (operand-size src) :byte))
(maybe-emit-rex-for-ea segment src nil)
(emit-byte segment #x0f)
- (emit-byte segment #x18)
+ (emit-byte segment #xae)
(emit-ea segment src 7)))
(macrolet ((define-fence-instruction (name last-byte)
projects / sbcl.git / blobdiff