0
|
1 ;; GCC machine description for MMX and 3dNOW! instructions
|
|
2 ;; Copyright (C) 2005, 2007, 2008
|
|
3 ;; Free Software Foundation, Inc.
|
|
4 ;;
|
|
5 ;; This file is part of GCC.
|
|
6 ;;
|
|
7 ;; GCC is free software; you can redistribute it and/or modify
|
|
8 ;; it under the terms of the GNU General Public License as published by
|
|
9 ;; the Free Software Foundation; either version 3, or (at your option)
|
|
10 ;; any later version.
|
|
11 ;;
|
|
12 ;; GCC is distributed in the hope that it will be useful,
|
|
13 ;; but WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
14 ;; MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
|
15 ;; GNU General Public License for more details.
|
|
16 ;;
|
|
17 ;; You should have received a copy of the GNU General Public License
|
|
18 ;; along with GCC; see the file COPYING3. If not see
|
|
19 ;; <http://www.gnu.org/licenses/>.
|
|
20
|
|
21 ;; The MMX and 3dNOW! patterns are in the same file because they use
|
|
22 ;; the same register file, and 3dNOW! adds a number of extensions to
|
|
23 ;; the base integer MMX isa.
|
|
24
|
|
25 ;; Note! Except for the basic move instructions, *all* of these
|
|
26 ;; patterns are outside the normal optabs namespace. This is because
|
|
27 ;; use of these registers requires the insertion of emms or femms
|
|
28 ;; instructions to return to normal fpu mode. The compiler doesn't
|
|
29 ;; know how to do that itself, which means it's up to the user. Which
|
|
30 ;; means that we should never use any of these patterns except at the
|
|
31 ;; direction of the user via a builtin.
|
|
32
|
|
33 ;; 8 byte integral modes handled by MMX (and by extension, SSE)
|
|
34 (define_mode_iterator MMXMODEI [V8QI V4HI V2SI])
|
|
35 (define_mode_iterator MMXMODEI8 [V8QI V4HI V2SI V1DI])
|
|
36
|
|
37 ;; All 8-byte vector modes handled by MMX
|
|
38 (define_mode_iterator MMXMODE [V8QI V4HI V2SI V1DI V2SF])
|
|
39
|
|
40 ;; Mix-n-match
|
|
41 (define_mode_iterator MMXMODE12 [V8QI V4HI])
|
|
42 (define_mode_iterator MMXMODE24 [V4HI V2SI])
|
|
43 (define_mode_iterator MMXMODE248 [V4HI V2SI V1DI])
|
|
44
|
|
45 ;; Mapping from integer vector mode to mnemonic suffix
|
|
46 (define_mode_attr mmxvecsize [(V8QI "b") (V4HI "w") (V2SI "d") (V1DI "q")])
|
|
47
|
|
48 ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
|
|
49 ;;
|
|
50 ;; Move patterns
|
|
51 ;;
|
|
52 ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
|
|
53
|
|
54 ;; All of these patterns are enabled for MMX as well as 3dNOW.
|
|
55 ;; This is essential for maintaining stable calling conventions.
|
|
56
|
|
57 (define_expand "mov<mode>"
|
|
58 [(set (match_operand:MMXMODEI8 0 "nonimmediate_operand" "")
|
|
59 (match_operand:MMXMODEI8 1 "nonimmediate_operand" ""))]
|
|
60 "TARGET_MMX"
|
|
61 {
|
|
62 ix86_expand_vector_move (<MODE>mode, operands);
|
|
63 DONE;
|
|
64 })
|
|
65
|
|
66 (define_insn "*mov<mode>_internal_rex64"
|
|
67 [(set (match_operand:MMXMODEI8 0 "nonimmediate_operand"
|
|
68 "=rm,r,!?y,!?y ,m ,!y,*Y2,x,x ,m,r,Yi")
|
|
69 (match_operand:MMXMODEI8 1 "vector_move_operand"
|
|
70 "Cr ,m,C ,!?ym,!?y,*Y2,!y,C,xm,x,Yi,r"))]
|
|
71 "TARGET_64BIT && TARGET_MMX
|
|
72 && !(MEM_P (operands[0]) && MEM_P (operands[1]))"
|
|
73 "@
|
|
74 mov{q}\t{%1, %0|%0, %1}
|
|
75 mov{q}\t{%1, %0|%0, %1}
|
|
76 pxor\t%0, %0
|
|
77 movq\t{%1, %0|%0, %1}
|
|
78 movq\t{%1, %0|%0, %1}
|
|
79 movdq2q\t{%1, %0|%0, %1}
|
|
80 movq2dq\t{%1, %0|%0, %1}
|
|
81 %vpxor\t%0, %d0
|
|
82 %vmovq\t{%1, %0|%0, %1}
|
|
83 %vmovq\t{%1, %0|%0, %1}
|
|
84 %vmovq\t{%1, %0|%0, %1}
|
|
85 %vmovq\t{%1, %0|%0, %1}"
|
|
86 [(set_attr "type" "imov,imov,mmx,mmxmov,mmxmov,ssecvt,ssecvt,sselog1,ssemov,ssemov,ssemov,ssemov")
|
|
87 (set_attr "unit" "*,*,*,*,*,mmx,mmx,*,*,*,*,*")
|
|
88 (set (attr "prefix")
|
|
89 (if_then_else (eq_attr "alternative" "7,8,9,10,11")
|
|
90 (const_string "maybe_vex")
|
|
91 (const_string "orig")))
|
|
92 (set_attr "mode" "DI")])
|
|
93
|
|
94 (define_insn "*mov<mode>_internal_avx"
|
|
95 [(set (match_operand:MMXMODEI8 0 "nonimmediate_operand"
|
|
96 "=!?y,!?y,m ,!y ,*Y2,*Y2,*Y2 ,m ,r ,m")
|
|
97 (match_operand:MMXMODEI8 1 "vector_move_operand"
|
|
98 "C ,!ym,!?y,*Y2,!y ,C ,*Y2m,*Y2,irm,r"))]
|
|
99 "TARGET_AVX
|
|
100 && !(MEM_P (operands[0]) && MEM_P (operands[1]))"
|
|
101 "@
|
|
102 pxor\t%0, %0
|
|
103 movq\t{%1, %0|%0, %1}
|
|
104 movq\t{%1, %0|%0, %1}
|
|
105 movdq2q\t{%1, %0|%0, %1}
|
|
106 movq2dq\t{%1, %0|%0, %1}
|
|
107 vpxor\t%0, %0, %0
|
|
108 vmovq\t{%1, %0|%0, %1}
|
|
109 vmovq\t{%1, %0|%0, %1}
|
|
110 #
|
|
111 #"
|
|
112 [(set_attr "type" "mmx,mmxmov,mmxmov,ssecvt,ssecvt,sselog1,ssemov,ssemov,*,*")
|
|
113 (set_attr "unit" "*,*,*,mmx,mmx,*,*,*,*,*")
|
|
114 (set (attr "prefix")
|
|
115 (if_then_else (eq_attr "alternative" "5,6,7")
|
|
116 (const_string "vex")
|
|
117 (const_string "orig")))
|
|
118 (set_attr "mode" "DI,DI,DI,DI,DI,TI,DI,DI,DI,DI")])
|
|
119
|
|
120 (define_insn "*mov<mode>_internal"
|
|
121 [(set (match_operand:MMXMODEI8 0 "nonimmediate_operand"
|
|
122 "=!?y,!?y,m ,!y ,*Y2,*Y2,*Y2 ,m ,*x,*x,*x,m ,r ,m")
|
|
123 (match_operand:MMXMODEI8 1 "vector_move_operand"
|
|
124 "C ,!ym,!?y,*Y2,!y ,C ,*Y2m,*Y2,C ,*x,m ,*x,irm,r"))]
|
|
125 "TARGET_MMX
|
|
126 && !(MEM_P (operands[0]) && MEM_P (operands[1]))"
|
|
127 "@
|
|
128 pxor\t%0, %0
|
|
129 movq\t{%1, %0|%0, %1}
|
|
130 movq\t{%1, %0|%0, %1}
|
|
131 movdq2q\t{%1, %0|%0, %1}
|
|
132 movq2dq\t{%1, %0|%0, %1}
|
|
133 pxor\t%0, %0
|
|
134 movq\t{%1, %0|%0, %1}
|
|
135 movq\t{%1, %0|%0, %1}
|
|
136 xorps\t%0, %0
|
|
137 movaps\t{%1, %0|%0, %1}
|
|
138 movlps\t{%1, %0|%0, %1}
|
|
139 movlps\t{%1, %0|%0, %1}
|
|
140 #
|
|
141 #"
|
|
142 [(set_attr "type" "mmx,mmxmov,mmxmov,ssecvt,ssecvt,sselog1,ssemov,ssemov,sselog1,ssemov,ssemov,ssemov,*,*")
|
|
143 (set_attr "unit" "*,*,*,mmx,mmx,*,*,*,*,*,*,*,*,*")
|
|
144 (set_attr "mode" "DI,DI,DI,DI,DI,TI,DI,DI,V4SF,V4SF,V2SF,V2SF,DI,DI")])
|
|
145
|
|
146 (define_expand "movv2sf"
|
|
147 [(set (match_operand:V2SF 0 "nonimmediate_operand" "")
|
|
148 (match_operand:V2SF 1 "nonimmediate_operand" ""))]
|
|
149 "TARGET_MMX"
|
|
150 {
|
|
151 ix86_expand_vector_move (V2SFmode, operands);
|
|
152 DONE;
|
|
153 })
|
|
154
|
|
155 (define_insn "*movv2sf_internal_rex64_avx"
|
|
156 [(set (match_operand:V2SF 0 "nonimmediate_operand"
|
|
157 "=rm,r ,!?y,!?y ,m ,!y,Y2,x,x,x,m,r,x")
|
|
158 (match_operand:V2SF 1 "vector_move_operand"
|
|
159 "Cr ,m ,C ,!?ym,!y,Y2,!y,C,x,m,x,x,r"))]
|
|
160 "TARGET_64BIT && TARGET_AVX
|
|
161 && !(MEM_P (operands[0]) && MEM_P (operands[1]))"
|
|
162 "@
|
|
163 mov{q}\t{%1, %0|%0, %1}
|
|
164 mov{q}\t{%1, %0|%0, %1}
|
|
165 pxor\t%0, %0
|
|
166 movq\t{%1, %0|%0, %1}
|
|
167 movq\t{%1, %0|%0, %1}
|
|
168 movdq2q\t{%1, %0|%0, %1}
|
|
169 movq2dq\t{%1, %0|%0, %1}
|
|
170 vxorps\t%0, %0, %0
|
|
171 vmovaps\t{%1, %0|%0, %1}
|
|
172 vmovlps\t{%1, %0, %0|%0, %0, %1}
|
|
173 vmovlps\t{%1, %0|%0, %1}
|
|
174 vmovq\t{%1, %0|%0, %1}
|
|
175 vmovq\t{%1, %0|%0, %1}"
|
|
176 [(set_attr "type" "imov,imov,mmx,mmxmov,mmxmov,ssecvt,ssecvt,ssemov,sselog1,ssemov,ssemov,ssemov,ssemov")
|
|
177 (set_attr "unit" "*,*,*,*,*,mmx,mmx,*,*,*,*,*,*")
|
|
178 (set (attr "prefix")
|
|
179 (if_then_else (eq_attr "alternative" "7,8,9,10,11,12")
|
|
180 (const_string "vex")
|
|
181 (const_string "orig")))
|
|
182 (set_attr "mode" "DI,DI,DI,DI,DI,DI,DI,V4SF,V4SF,V2SF,V2SF,DI,DI")])
|
|
183
|
|
184 (define_insn "*movv2sf_internal_rex64"
|
|
185 [(set (match_operand:V2SF 0 "nonimmediate_operand"
|
|
186 "=rm,r ,!?y,!?y ,m ,!y,*Y2,x,x,x,m,r,Yi")
|
|
187 (match_operand:V2SF 1 "vector_move_operand"
|
|
188 "Cr ,m ,C ,!?ym,!y,*Y2,!y,C,x,m,x,Yi,r"))]
|
|
189 "TARGET_64BIT && TARGET_MMX
|
|
190 && !(MEM_P (operands[0]) && MEM_P (operands[1]))"
|
|
191 "@
|
|
192 mov{q}\t{%1, %0|%0, %1}
|
|
193 mov{q}\t{%1, %0|%0, %1}
|
|
194 pxor\t%0, %0
|
|
195 movq\t{%1, %0|%0, %1}
|
|
196 movq\t{%1, %0|%0, %1}
|
|
197 movdq2q\t{%1, %0|%0, %1}
|
|
198 movq2dq\t{%1, %0|%0, %1}
|
|
199 xorps\t%0, %0
|
|
200 movaps\t{%1, %0|%0, %1}
|
|
201 movlps\t{%1, %0|%0, %1}
|
|
202 movlps\t{%1, %0|%0, %1}
|
|
203 movd\t{%1, %0|%0, %1}
|
|
204 movd\t{%1, %0|%0, %1}"
|
|
205 [(set_attr "type" "imov,imov,mmx,mmxmov,mmxmov,ssecvt,ssecvt,ssemov,sselog1,ssemov,ssemov,ssemov,ssemov")
|
|
206 (set_attr "unit" "*,*,*,*,*,mmx,mmx,*,*,*,*,*,*")
|
|
207 (set_attr "mode" "DI,DI,DI,DI,DI,DI,DI,V4SF,V4SF,V2SF,V2SF,DI,DI")])
|
|
208
|
|
209 (define_insn "*movv2sf_internal_avx"
|
|
210 [(set (match_operand:V2SF 0 "nonimmediate_operand"
|
|
211 "=!?y,!?y ,m ,!y ,*Y2,*x,*x,*x,m ,r ,m")
|
|
212 (match_operand:V2SF 1 "vector_move_operand"
|
|
213 "C ,!?ym,!?y,*Y2,!y ,C ,*x,m ,*x,irm,r"))]
|
|
214 "TARGET_AVX
|
|
215 && !(MEM_P (operands[0]) && MEM_P (operands[1]))"
|
|
216 "@
|
|
217 pxor\t%0, %0
|
|
218 movq\t{%1, %0|%0, %1}
|
|
219 movq\t{%1, %0|%0, %1}
|
|
220 movdq2q\t{%1, %0|%0, %1}
|
|
221 movq2dq\t{%1, %0|%0, %1}
|
|
222 vxorps\t%0, %0, %0
|
|
223 vmovaps\t{%1, %0|%0, %1}
|
|
224 vmovlps\t{%1, %0, %0|%0, %0, %1}
|
|
225 vmovlps\t{%1, %0|%0, %1}
|
|
226 #
|
|
227 #"
|
|
228 [(set_attr "type" "mmx,mmxmov,mmxmov,ssecvt,ssecvt,sselog1,ssemov,ssemov,ssemov,*,*")
|
|
229 (set_attr "unit" "*,*,*,mmx,mmx,*,*,*,*,*,*")
|
|
230 (set (attr "prefix")
|
|
231 (if_then_else (eq_attr "alternative" "5,6,7,8")
|
|
232 (const_string "vex")
|
|
233 (const_string "orig")))
|
|
234 (set_attr "mode" "DI,DI,DI,DI,DI,V4SF,V4SF,V2SF,V2SF,DI,DI")])
|
|
235
|
|
236 (define_insn "*movv2sf_internal"
|
|
237 [(set (match_operand:V2SF 0 "nonimmediate_operand"
|
|
238 "=!?y,!?y ,m ,!y ,*Y2,*x,*x,*x,m ,r ,m")
|
|
239 (match_operand:V2SF 1 "vector_move_operand"
|
|
240 "C ,!?ym,!?y,*Y2,!y ,C ,*x,m ,*x,irm,r"))]
|
|
241 "TARGET_MMX
|
|
242 && !(MEM_P (operands[0]) && MEM_P (operands[1]))"
|
|
243 "@
|
|
244 pxor\t%0, %0
|
|
245 movq\t{%1, %0|%0, %1}
|
|
246 movq\t{%1, %0|%0, %1}
|
|
247 movdq2q\t{%1, %0|%0, %1}
|
|
248 movq2dq\t{%1, %0|%0, %1}
|
|
249 xorps\t%0, %0
|
|
250 movaps\t{%1, %0|%0, %1}
|
|
251 movlps\t{%1, %0|%0, %1}
|
|
252 movlps\t{%1, %0|%0, %1}
|
|
253 #
|
|
254 #"
|
|
255 [(set_attr "type" "mmx,mmxmov,mmxmov,ssecvt,ssecvt,sselog1,ssemov,ssemov,ssemov,*,*")
|
|
256 (set_attr "unit" "*,*,*,mmx,mmx,*,*,*,*,*,*")
|
|
257 (set_attr "mode" "DI,DI,DI,DI,DI,V4SF,V4SF,V2SF,V2SF,DI,DI")])
|
|
258
|
|
259 ;; %%% This multiword shite has got to go.
|
|
260 (define_split
|
|
261 [(set (match_operand:MMXMODE 0 "nonimmediate_operand" "")
|
|
262 (match_operand:MMXMODE 1 "general_operand" ""))]
|
|
263 "!TARGET_64BIT && reload_completed
|
|
264 && (!MMX_REG_P (operands[0]) && !SSE_REG_P (operands[0]))
|
|
265 && (!MMX_REG_P (operands[1]) && !SSE_REG_P (operands[1]))"
|
|
266 [(const_int 0)]
|
|
267 "ix86_split_long_move (operands); DONE;")
|
|
268
|
|
269 (define_expand "push<mode>1"
|
|
270 [(match_operand:MMXMODE 0 "register_operand" "")]
|
|
271 "TARGET_MMX"
|
|
272 {
|
|
273 ix86_expand_push (<MODE>mode, operands[0]);
|
|
274 DONE;
|
|
275 })
|
|
276
|
|
277 (define_expand "movmisalign<mode>"
|
|
278 [(set (match_operand:MMXMODE 0 "nonimmediate_operand" "")
|
|
279 (match_operand:MMXMODE 1 "nonimmediate_operand" ""))]
|
|
280 "TARGET_MMX"
|
|
281 {
|
|
282 ix86_expand_vector_move (<MODE>mode, operands);
|
|
283 DONE;
|
|
284 })
|
|
285
|
|
286 (define_insn "sse_movntdi"
|
|
287 [(set (match_operand:DI 0 "memory_operand" "=m")
|
|
288 (unspec:DI [(match_operand:DI 1 "register_operand" "y")]
|
|
289 UNSPEC_MOVNT))]
|
|
290 "TARGET_SSE || TARGET_3DNOW_A"
|
|
291 "movntq\t{%1, %0|%0, %1}"
|
|
292 [(set_attr "type" "mmxmov")
|
|
293 (set_attr "mode" "DI")])
|
|
294
|
|
295 ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
|
|
296 ;;
|
|
297 ;; Parallel single-precision floating point arithmetic
|
|
298 ;;
|
|
299 ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
|
|
300
|
|
301 (define_expand "mmx_addv2sf3"
|
|
302 [(set (match_operand:V2SF 0 "register_operand" "")
|
|
303 (plus:V2SF
|
|
304 (match_operand:V2SF 1 "nonimmediate_operand" "")
|
|
305 (match_operand:V2SF 2 "nonimmediate_operand" "")))]
|
|
306 "TARGET_3DNOW"
|
|
307 "ix86_fixup_binary_operands_no_copy (PLUS, V2SFmode, operands);")
|
|
308
|
|
309 (define_insn "*mmx_addv2sf3"
|
|
310 [(set (match_operand:V2SF 0 "register_operand" "=y")
|
|
311 (plus:V2SF (match_operand:V2SF 1 "nonimmediate_operand" "%0")
|
|
312 (match_operand:V2SF 2 "nonimmediate_operand" "ym")))]
|
|
313 "TARGET_3DNOW && ix86_binary_operator_ok (PLUS, V2SFmode, operands)"
|
|
314 "pfadd\t{%2, %0|%0, %2}"
|
|
315 [(set_attr "type" "mmxadd")
|
|
316 (set_attr "mode" "V2SF")])
|
|
317
|
|
318 (define_expand "mmx_subv2sf3"
|
|
319 [(set (match_operand:V2SF 0 "register_operand" "")
|
|
320 (minus:V2SF (match_operand:V2SF 1 "register_operand" "")
|
|
321 (match_operand:V2SF 2 "nonimmediate_operand" "")))]
|
|
322 "TARGET_3DNOW"
|
|
323 "")
|
|
324
|
|
325 (define_expand "mmx_subrv2sf3"
|
|
326 [(set (match_operand:V2SF 0 "register_operand" "")
|
|
327 (minus:V2SF (match_operand:V2SF 2 "register_operand" "")
|
|
328 (match_operand:V2SF 1 "nonimmediate_operand" "")))]
|
|
329 "TARGET_3DNOW"
|
|
330 "")
|
|
331
|
|
332 (define_insn "*mmx_subv2sf3"
|
|
333 [(set (match_operand:V2SF 0 "register_operand" "=y,y")
|
|
334 (minus:V2SF (match_operand:V2SF 1 "nonimmediate_operand" "0,ym")
|
|
335 (match_operand:V2SF 2 "nonimmediate_operand" "ym,0")))]
|
|
336 "TARGET_3DNOW && !(MEM_P (operands[0]) && MEM_P (operands[1]))"
|
|
337 "@
|
|
338 pfsub\t{%2, %0|%0, %2}
|
|
339 pfsubr\t{%2, %0|%0, %2}"
|
|
340 [(set_attr "type" "mmxadd")
|
|
341 (set_attr "mode" "V2SF")])
|
|
342
|
|
343 (define_expand "mmx_mulv2sf3"
|
|
344 [(set (match_operand:V2SF 0 "register_operand" "")
|
|
345 (mult:V2SF (match_operand:V2SF 1 "nonimmediate_operand" "")
|
|
346 (match_operand:V2SF 2 "nonimmediate_operand" "")))]
|
|
347 "TARGET_3DNOW"
|
|
348 "ix86_fixup_binary_operands_no_copy (MULT, V2SFmode, operands);")
|
|
349
|
|
350 (define_insn "*mmx_mulv2sf3"
|
|
351 [(set (match_operand:V2SF 0 "register_operand" "=y")
|
|
352 (mult:V2SF (match_operand:V2SF 1 "nonimmediate_operand" "%0")
|
|
353 (match_operand:V2SF 2 "nonimmediate_operand" "ym")))]
|
|
354 "TARGET_3DNOW && ix86_binary_operator_ok (MULT, V2SFmode, operands)"
|
|
355 "pfmul\t{%2, %0|%0, %2}"
|
|
356 [(set_attr "type" "mmxmul")
|
|
357 (set_attr "mode" "V2SF")])
|
|
358
|
|
359 ;; ??? For !flag_finite_math_only, the representation with SMIN/SMAX
|
|
360 ;; isn't really correct, as those rtl operators aren't defined when
|
|
361 ;; applied to NaNs. Hopefully the optimizers won't get too smart on us.
|
|
362
|
|
363 (define_expand "mmx_<code>v2sf3"
|
|
364 [(set (match_operand:V2SF 0 "register_operand" "")
|
|
365 (smaxmin:V2SF
|
|
366 (match_operand:V2SF 1 "nonimmediate_operand" "")
|
|
367 (match_operand:V2SF 2 "nonimmediate_operand" "")))]
|
|
368 "TARGET_3DNOW"
|
|
369 {
|
|
370 if (!flag_finite_math_only)
|
|
371 operands[1] = force_reg (V2SFmode, operands[1]);
|
|
372 ix86_fixup_binary_operands_no_copy (<CODE>, V2SFmode, operands);
|
|
373 })
|
|
374
|
|
375 (define_insn "*mmx_<code>v2sf3_finite"
|
|
376 [(set (match_operand:V2SF 0 "register_operand" "=y")
|
|
377 (smaxmin:V2SF
|
|
378 (match_operand:V2SF 1 "nonimmediate_operand" "%0")
|
|
379 (match_operand:V2SF 2 "nonimmediate_operand" "ym")))]
|
|
380 "TARGET_3DNOW && flag_finite_math_only
|
|
381 && ix86_binary_operator_ok (<CODE>, V2SFmode, operands)"
|
|
382 "pf<maxminfprefix>\t{%2, %0|%0, %2}"
|
|
383 [(set_attr "type" "mmxadd")
|
|
384 (set_attr "mode" "V2SF")])
|
|
385
|
|
386 (define_insn "*mmx_<code>v2sf3"
|
|
387 [(set (match_operand:V2SF 0 "register_operand" "=y")
|
|
388 (smaxmin:V2SF
|
|
389 (match_operand:V2SF 1 "register_operand" "0")
|
|
390 (match_operand:V2SF 2 "nonimmediate_operand" "ym")))]
|
|
391 "TARGET_3DNOW"
|
|
392 "pf<maxminfprefix>\t{%2, %0|%0, %2}"
|
|
393 [(set_attr "type" "mmxadd")
|
|
394 (set_attr "mode" "V2SF")])
|
|
395
|
|
396 (define_insn "mmx_rcpv2sf2"
|
|
397 [(set (match_operand:V2SF 0 "register_operand" "=y")
|
|
398 (unspec:V2SF [(match_operand:V2SF 1 "nonimmediate_operand" "ym")]
|
|
399 UNSPEC_PFRCP))]
|
|
400 "TARGET_3DNOW"
|
|
401 "pfrcp\t{%1, %0|%0, %1}"
|
|
402 [(set_attr "type" "mmx")
|
|
403 (set_attr "mode" "V2SF")])
|
|
404
|
|
405 (define_insn "mmx_rcpit1v2sf3"
|
|
406 [(set (match_operand:V2SF 0 "register_operand" "=y")
|
|
407 (unspec:V2SF [(match_operand:V2SF 1 "register_operand" "0")
|
|
408 (match_operand:V2SF 2 "nonimmediate_operand" "ym")]
|
|
409 UNSPEC_PFRCPIT1))]
|
|
410 "TARGET_3DNOW"
|
|
411 "pfrcpit1\t{%2, %0|%0, %2}"
|
|
412 [(set_attr "type" "mmx")
|
|
413 (set_attr "mode" "V2SF")])
|
|
414
|
|
415 (define_insn "mmx_rcpit2v2sf3"
|
|
416 [(set (match_operand:V2SF 0 "register_operand" "=y")
|
|
417 (unspec:V2SF [(match_operand:V2SF 1 "register_operand" "0")
|
|
418 (match_operand:V2SF 2 "nonimmediate_operand" "ym")]
|
|
419 UNSPEC_PFRCPIT2))]
|
|
420 "TARGET_3DNOW"
|
|
421 "pfrcpit2\t{%2, %0|%0, %2}"
|
|
422 [(set_attr "type" "mmx")
|
|
423 (set_attr "mode" "V2SF")])
|
|
424
|
|
425 (define_insn "mmx_rsqrtv2sf2"
|
|
426 [(set (match_operand:V2SF 0 "register_operand" "=y")
|
|
427 (unspec:V2SF [(match_operand:V2SF 1 "nonimmediate_operand" "ym")]
|
|
428 UNSPEC_PFRSQRT))]
|
|
429 "TARGET_3DNOW"
|
|
430 "pfrsqrt\t{%1, %0|%0, %1}"
|
|
431 [(set_attr "type" "mmx")
|
|
432 (set_attr "mode" "V2SF")])
|
|
433
|
|
434 (define_insn "mmx_rsqit1v2sf3"
|
|
435 [(set (match_operand:V2SF 0 "register_operand" "=y")
|
|
436 (unspec:V2SF [(match_operand:V2SF 1 "register_operand" "0")
|
|
437 (match_operand:V2SF 2 "nonimmediate_operand" "ym")]
|
|
438 UNSPEC_PFRSQIT1))]
|
|
439 "TARGET_3DNOW"
|
|
440 "pfrsqit1\t{%2, %0|%0, %2}"
|
|
441 [(set_attr "type" "mmx")
|
|
442 (set_attr "mode" "V2SF")])
|
|
443
|
|
444 (define_insn "mmx_haddv2sf3"
|
|
445 [(set (match_operand:V2SF 0 "register_operand" "=y")
|
|
446 (vec_concat:V2SF
|
|
447 (plus:SF
|
|
448 (vec_select:SF
|
|
449 (match_operand:V2SF 1 "register_operand" "0")
|
|
450 (parallel [(const_int 0)]))
|
|
451 (vec_select:SF (match_dup 1) (parallel [(const_int 1)])))
|
|
452 (plus:SF
|
|
453 (vec_select:SF
|
|
454 (match_operand:V2SF 2 "nonimmediate_operand" "ym")
|
|
455 (parallel [(const_int 0)]))
|
|
456 (vec_select:SF (match_dup 2) (parallel [(const_int 1)])))))]
|
|
457 "TARGET_3DNOW"
|
|
458 "pfacc\t{%2, %0|%0, %2}"
|
|
459 [(set_attr "type" "mmxadd")
|
|
460 (set_attr "mode" "V2SF")])
|
|
461
|
|
462 (define_insn "mmx_hsubv2sf3"
|
|
463 [(set (match_operand:V2SF 0 "register_operand" "=y")
|
|
464 (vec_concat:V2SF
|
|
465 (minus:SF
|
|
466 (vec_select:SF
|
|
467 (match_operand:V2SF 1 "register_operand" "0")
|
|
468 (parallel [(const_int 0)]))
|
|
469 (vec_select:SF (match_dup 1) (parallel [(const_int 1)])))
|
|
470 (minus:SF
|
|
471 (vec_select:SF
|
|
472 (match_operand:V2SF 2 "nonimmediate_operand" "ym")
|
|
473 (parallel [(const_int 0)]))
|
|
474 (vec_select:SF (match_dup 2) (parallel [(const_int 1)])))))]
|
|
475 "TARGET_3DNOW_A"
|
|
476 "pfnacc\t{%2, %0|%0, %2}"
|
|
477 [(set_attr "type" "mmxadd")
|
|
478 (set_attr "mode" "V2SF")])
|
|
479
|
|
480 (define_insn "mmx_addsubv2sf3"
|
|
481 [(set (match_operand:V2SF 0 "register_operand" "=y")
|
|
482 (vec_merge:V2SF
|
|
483 (plus:V2SF
|
|
484 (match_operand:V2SF 1 "register_operand" "0")
|
|
485 (match_operand:V2SF 2 "nonimmediate_operand" "ym"))
|
|
486 (minus:V2SF (match_dup 1) (match_dup 2))
|
|
487 (const_int 1)))]
|
|
488 "TARGET_3DNOW_A"
|
|
489 "pfpnacc\t{%2, %0|%0, %2}"
|
|
490 [(set_attr "type" "mmxadd")
|
|
491 (set_attr "mode" "V2SF")])
|
|
492
|
|
493 ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
|
|
494 ;;
|
|
495 ;; Parallel single-precision floating point comparisons
|
|
496 ;;
|
|
497 ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
|
|
498
|
|
499 (define_expand "mmx_eqv2sf3"
|
|
500 [(set (match_operand:V2SI 0 "register_operand" "")
|
|
501 (eq:V2SI (match_operand:V2SF 1 "nonimmediate_operand" "")
|
|
502 (match_operand:V2SF 2 "nonimmediate_operand" "")))]
|
|
503 "TARGET_3DNOW"
|
|
504 "ix86_fixup_binary_operands_no_copy (EQ, V2SFmode, operands);")
|
|
505
|
|
506 (define_insn "*mmx_eqv2sf3"
|
|
507 [(set (match_operand:V2SI 0 "register_operand" "=y")
|
|
508 (eq:V2SI (match_operand:V2SF 1 "nonimmediate_operand" "%0")
|
|
509 (match_operand:V2SF 2 "nonimmediate_operand" "ym")))]
|
|
510 "TARGET_3DNOW && ix86_binary_operator_ok (EQ, V2SFmode, operands)"
|
|
511 "pfcmpeq\t{%2, %0|%0, %2}"
|
|
512 [(set_attr "type" "mmxcmp")
|
|
513 (set_attr "mode" "V2SF")])
|
|
514
|
|
515 (define_insn "mmx_gtv2sf3"
|
|
516 [(set (match_operand:V2SI 0 "register_operand" "=y")
|
|
517 (gt:V2SI (match_operand:V2SF 1 "register_operand" "0")
|
|
518 (match_operand:V2SF 2 "nonimmediate_operand" "ym")))]
|
|
519 "TARGET_3DNOW"
|
|
520 "pfcmpgt\t{%2, %0|%0, %2}"
|
|
521 [(set_attr "type" "mmxcmp")
|
|
522 (set_attr "mode" "V2SF")])
|
|
523
|
|
524 (define_insn "mmx_gev2sf3"
|
|
525 [(set (match_operand:V2SI 0 "register_operand" "=y")
|
|
526 (ge:V2SI (match_operand:V2SF 1 "register_operand" "0")
|
|
527 (match_operand:V2SF 2 "nonimmediate_operand" "ym")))]
|
|
528 "TARGET_3DNOW"
|
|
529 "pfcmpge\t{%2, %0|%0, %2}"
|
|
530 [(set_attr "type" "mmxcmp")
|
|
531 (set_attr "mode" "V2SF")])
|
|
532
|
|
533 ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
|
|
534 ;;
|
|
535 ;; Parallel single-precision floating point conversion operations
|
|
536 ;;
|
|
537 ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
|
|
538
|
|
539 (define_insn "mmx_pf2id"
|
|
540 [(set (match_operand:V2SI 0 "register_operand" "=y")
|
|
541 (fix:V2SI (match_operand:V2SF 1 "nonimmediate_operand" "ym")))]
|
|
542 "TARGET_3DNOW"
|
|
543 "pf2id\t{%1, %0|%0, %1}"
|
|
544 [(set_attr "type" "mmxcvt")
|
|
545 (set_attr "mode" "V2SF")])
|
|
546
|
|
547 (define_insn "mmx_pf2iw"
|
|
548 [(set (match_operand:V2SI 0 "register_operand" "=y")
|
|
549 (sign_extend:V2SI
|
|
550 (ss_truncate:V2HI
|
|
551 (fix:V2SI
|
|
552 (match_operand:V2SF 1 "nonimmediate_operand" "ym")))))]
|
|
553 "TARGET_3DNOW_A"
|
|
554 "pf2iw\t{%1, %0|%0, %1}"
|
|
555 [(set_attr "type" "mmxcvt")
|
|
556 (set_attr "mode" "V2SF")])
|
|
557
|
|
558 (define_insn "mmx_pi2fw"
|
|
559 [(set (match_operand:V2SF 0 "register_operand" "=y")
|
|
560 (float:V2SF
|
|
561 (sign_extend:V2SI
|
|
562 (truncate:V2HI
|
|
563 (match_operand:V2SI 1 "nonimmediate_operand" "ym")))))]
|
|
564 "TARGET_3DNOW_A"
|
|
565 "pi2fw\t{%1, %0|%0, %1}"
|
|
566 [(set_attr "type" "mmxcvt")
|
|
567 (set_attr "mode" "V2SF")])
|
|
568
|
|
569 (define_insn "mmx_floatv2si2"
|
|
570 [(set (match_operand:V2SF 0 "register_operand" "=y")
|
|
571 (float:V2SF (match_operand:V2SI 1 "nonimmediate_operand" "ym")))]
|
|
572 "TARGET_3DNOW"
|
|
573 "pi2fd\t{%1, %0|%0, %1}"
|
|
574 [(set_attr "type" "mmxcvt")
|
|
575 (set_attr "mode" "V2SF")])
|
|
576
|
|
577 ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
|
|
578 ;;
|
|
579 ;; Parallel single-precision floating point element swizzling
|
|
580 ;;
|
|
581 ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
|
|
582
|
|
583 (define_insn "mmx_pswapdv2sf2"
|
|
584 [(set (match_operand:V2SF 0 "register_operand" "=y")
|
|
585 (vec_select:V2SF (match_operand:V2SF 1 "nonimmediate_operand" "ym")
|
|
586 (parallel [(const_int 1) (const_int 0)])))]
|
|
587 "TARGET_3DNOW_A"
|
|
588 "pswapd\t{%1, %0|%0, %1}"
|
|
589 [(set_attr "type" "mmxcvt")
|
|
590 (set_attr "mode" "V2SF")])
|
|
591
|
|
592 (define_insn "*vec_dupv2sf"
|
|
593 [(set (match_operand:V2SF 0 "register_operand" "=y")
|
|
594 (vec_duplicate:V2SF
|
|
595 (match_operand:SF 1 "register_operand" "0")))]
|
|
596 "TARGET_MMX"
|
|
597 "punpckldq\t%0, %0"
|
|
598 [(set_attr "type" "mmxcvt")
|
|
599 (set_attr "mode" "DI")])
|
|
600
|
|
601 (define_insn "*mmx_concatv2sf"
|
|
602 [(set (match_operand:V2SF 0 "register_operand" "=y,y")
|
|
603 (vec_concat:V2SF
|
|
604 (match_operand:SF 1 "nonimmediate_operand" " 0,rm")
|
|
605 (match_operand:SF 2 "vector_move_operand" "ym,C")))]
|
|
606 "TARGET_MMX && !TARGET_SSE"
|
|
607 "@
|
|
608 punpckldq\t{%2, %0|%0, %2}
|
|
609 movd\t{%1, %0|%0, %1}"
|
|
610 [(set_attr "type" "mmxcvt,mmxmov")
|
|
611 (set_attr "mode" "DI")])
|
|
612
|
|
613 (define_expand "vec_setv2sf"
|
|
614 [(match_operand:V2SF 0 "register_operand" "")
|
|
615 (match_operand:SF 1 "register_operand" "")
|
|
616 (match_operand 2 "const_int_operand" "")]
|
|
617 "TARGET_MMX"
|
|
618 {
|
|
619 ix86_expand_vector_set (false, operands[0], operands[1],
|
|
620 INTVAL (operands[2]));
|
|
621 DONE;
|
|
622 })
|
|
623
|
|
624 ;; Avoid combining registers from different units in a single alternative,
|
|
625 ;; see comment above inline_secondary_memory_needed function in i386.c
|
|
626 (define_insn_and_split "*vec_extractv2sf_0"
|
|
627 [(set (match_operand:SF 0 "nonimmediate_operand" "=x, m,y ,m,f,r")
|
|
628 (vec_select:SF
|
|
629 (match_operand:V2SF 1 "nonimmediate_operand" " xm,x,ym,y,m,m")
|
|
630 (parallel [(const_int 0)])))]
|
|
631 "TARGET_MMX && !(MEM_P (operands[0]) && MEM_P (operands[1]))"
|
|
632 "#"
|
|
633 "&& reload_completed"
|
|
634 [(const_int 0)]
|
|
635 {
|
|
636 rtx op1 = operands[1];
|
|
637 if (REG_P (op1))
|
|
638 op1 = gen_rtx_REG (SFmode, REGNO (op1));
|
|
639 else
|
|
640 op1 = gen_lowpart (SFmode, op1);
|
|
641 emit_move_insn (operands[0], op1);
|
|
642 DONE;
|
|
643 })
|
|
644
|
|
645 ;; Avoid combining registers from different units in a single alternative,
|
|
646 ;; see comment above inline_secondary_memory_needed function in i386.c
|
|
647 (define_insn "*vec_extractv2sf_1"
|
|
648 [(set (match_operand:SF 0 "nonimmediate_operand" "=y,x,y,x,f,r")
|
|
649 (vec_select:SF
|
|
650 (match_operand:V2SF 1 "nonimmediate_operand" " 0,0,o,o,o,o")
|
|
651 (parallel [(const_int 1)])))]
|
|
652 "TARGET_MMX && !(MEM_P (operands[0]) && MEM_P (operands[1]))"
|
|
653 "@
|
|
654 punpckhdq\t%0, %0
|
|
655 unpckhps\t%0, %0
|
|
656 #
|
|
657 #
|
|
658 #
|
|
659 #"
|
|
660 [(set_attr "type" "mmxcvt,sselog1,mmxmov,ssemov,fmov,imov")
|
|
661 (set_attr "mode" "DI,V4SF,SF,SF,SF,SF")])
|
|
662
|
|
663 (define_split
|
|
664 [(set (match_operand:SF 0 "register_operand" "")
|
|
665 (vec_select:SF
|
|
666 (match_operand:V2SF 1 "memory_operand" "")
|
|
667 (parallel [(const_int 1)])))]
|
|
668 "TARGET_MMX && reload_completed"
|
|
669 [(const_int 0)]
|
|
670 {
|
|
671 operands[1] = adjust_address (operands[1], SFmode, 4);
|
|
672 emit_move_insn (operands[0], operands[1]);
|
|
673 DONE;
|
|
674 })
|
|
675
|
|
676 (define_expand "vec_extractv2sf"
|
|
677 [(match_operand:SF 0 "register_operand" "")
|
|
678 (match_operand:V2SF 1 "register_operand" "")
|
|
679 (match_operand 2 "const_int_operand" "")]
|
|
680 "TARGET_MMX"
|
|
681 {
|
|
682 ix86_expand_vector_extract (false, operands[0], operands[1],
|
|
683 INTVAL (operands[2]));
|
|
684 DONE;
|
|
685 })
|
|
686
|
|
687 (define_expand "vec_initv2sf"
|
|
688 [(match_operand:V2SF 0 "register_operand" "")
|
|
689 (match_operand 1 "" "")]
|
|
690 "TARGET_SSE"
|
|
691 {
|
|
692 ix86_expand_vector_init (false, operands[0], operands[1]);
|
|
693 DONE;
|
|
694 })
|
|
695
|
|
696 ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
|
|
697 ;;
|
|
698 ;; Parallel integral arithmetic
|
|
699 ;;
|
|
700 ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
|
|
701
|
|
702 (define_expand "mmx_<plusminus_insn><mode>3"
|
|
703 [(set (match_operand:MMXMODEI8 0 "register_operand" "")
|
|
704 (plusminus:MMXMODEI8
|
|
705 (match_operand:MMXMODEI8 1 "nonimmediate_operand" "")
|
|
706 (match_operand:MMXMODEI8 2 "nonimmediate_operand" "")))]
|
|
707 "TARGET_MMX || (TARGET_SSE2 && <MODE>mode == V1DImode)"
|
|
708 "ix86_fixup_binary_operands_no_copy (<CODE>, <MODE>mode, operands);")
|
|
709
|
|
710 (define_insn "*mmx_<plusminus_insn><mode>3"
|
|
711 [(set (match_operand:MMXMODEI8 0 "register_operand" "=y")
|
|
712 (plusminus:MMXMODEI8
|
|
713 (match_operand:MMXMODEI8 1 "nonimmediate_operand" "<comm>0")
|
|
714 (match_operand:MMXMODEI8 2 "nonimmediate_operand" "ym")))]
|
|
715 "(TARGET_MMX || (TARGET_SSE2 && <MODE>mode == V1DImode))
|
|
716 && ix86_binary_operator_ok (<CODE>, <MODE>mode, operands)"
|
|
717 "p<plusminus_mnemonic><mmxvecsize>\t{%2, %0|%0, %2}"
|
|
718 [(set_attr "type" "mmxadd")
|
|
719 (set_attr "mode" "DI")])
|
|
720
|
|
721 (define_expand "mmx_<plusminus_insn><mode>3"
|
|
722 [(set (match_operand:MMXMODE12 0 "register_operand" "")
|
|
723 (sat_plusminus:MMXMODE12
|
|
724 (match_operand:MMXMODE12 1 "nonimmediate_operand" "")
|
|
725 (match_operand:MMXMODE12 2 "nonimmediate_operand" "")))]
|
|
726 "TARGET_MMX"
|
|
727 "ix86_fixup_binary_operands_no_copy (<CODE>, <MODE>mode, operands);")
|
|
728
|
|
729 (define_insn "*mmx_<plusminus_insn><mode>3"
|
|
730 [(set (match_operand:MMXMODE12 0 "register_operand" "=y")
|
|
731 (sat_plusminus:MMXMODE12
|
|
732 (match_operand:MMXMODE12 1 "nonimmediate_operand" "<comm>0")
|
|
733 (match_operand:MMXMODE12 2 "nonimmediate_operand" "ym")))]
|
|
734 "TARGET_MMX && ix86_binary_operator_ok (<CODE>, <MODE>mode, operands)"
|
|
735 "p<plusminus_mnemonic><mmxvecsize>\t{%2, %0|%0, %2}"
|
|
736 [(set_attr "type" "mmxadd")
|
|
737 (set_attr "mode" "DI")])
|
|
738
|
|
739 (define_expand "mmx_mulv4hi3"
|
|
740 [(set (match_operand:V4HI 0 "register_operand" "")
|
|
741 (mult:V4HI (match_operand:V4HI 1 "nonimmediate_operand" "")
|
|
742 (match_operand:V4HI 2 "nonimmediate_operand" "")))]
|
|
743 "TARGET_MMX"
|
|
744 "ix86_fixup_binary_operands_no_copy (MULT, V4HImode, operands);")
|
|
745
|
|
746 (define_insn "*mmx_mulv4hi3"
|
|
747 [(set (match_operand:V4HI 0 "register_operand" "=y")
|
|
748 (mult:V4HI (match_operand:V4HI 1 "nonimmediate_operand" "%0")
|
|
749 (match_operand:V4HI 2 "nonimmediate_operand" "ym")))]
|
|
750 "TARGET_MMX && ix86_binary_operator_ok (MULT, V4HImode, operands)"
|
|
751 "pmullw\t{%2, %0|%0, %2}"
|
|
752 [(set_attr "type" "mmxmul")
|
|
753 (set_attr "mode" "DI")])
|
|
754
|
|
755 (define_expand "mmx_smulv4hi3_highpart"
|
|
756 [(set (match_operand:V4HI 0 "register_operand" "")
|
|
757 (truncate:V4HI
|
|
758 (lshiftrt:V4SI
|
|
759 (mult:V4SI
|
|
760 (sign_extend:V4SI
|
|
761 (match_operand:V4HI 1 "nonimmediate_operand" ""))
|
|
762 (sign_extend:V4SI
|
|
763 (match_operand:V4HI 2 "nonimmediate_operand" "")))
|
|
764 (const_int 16))))]
|
|
765 "TARGET_MMX"
|
|
766 "ix86_fixup_binary_operands_no_copy (MULT, V4HImode, operands);")
|
|
767
|
|
768 (define_insn "*mmx_smulv4hi3_highpart"
|
|
769 [(set (match_operand:V4HI 0 "register_operand" "=y")
|
|
770 (truncate:V4HI
|
|
771 (lshiftrt:V4SI
|
|
772 (mult:V4SI
|
|
773 (sign_extend:V4SI
|
|
774 (match_operand:V4HI 1 "nonimmediate_operand" "%0"))
|
|
775 (sign_extend:V4SI
|
|
776 (match_operand:V4HI 2 "nonimmediate_operand" "ym")))
|
|
777 (const_int 16))))]
|
|
778 "TARGET_MMX && ix86_binary_operator_ok (MULT, V4HImode, operands)"
|
|
779 "pmulhw\t{%2, %0|%0, %2}"
|
|
780 [(set_attr "type" "mmxmul")
|
|
781 (set_attr "mode" "DI")])
|
|
782
|
|
783 (define_expand "mmx_umulv4hi3_highpart"
|
|
784 [(set (match_operand:V4HI 0 "register_operand" "")
|
|
785 (truncate:V4HI
|
|
786 (lshiftrt:V4SI
|
|
787 (mult:V4SI
|
|
788 (zero_extend:V4SI
|
|
789 (match_operand:V4HI 1 "nonimmediate_operand" ""))
|
|
790 (zero_extend:V4SI
|
|
791 (match_operand:V4HI 2 "nonimmediate_operand" "")))
|
|
792 (const_int 16))))]
|
|
793 "TARGET_SSE || TARGET_3DNOW_A"
|
|
794 "ix86_fixup_binary_operands_no_copy (MULT, V4HImode, operands);")
|
|
795
|
|
796 (define_insn "*mmx_umulv4hi3_highpart"
|
|
797 [(set (match_operand:V4HI 0 "register_operand" "=y")
|
|
798 (truncate:V4HI
|
|
799 (lshiftrt:V4SI
|
|
800 (mult:V4SI
|
|
801 (zero_extend:V4SI
|
|
802 (match_operand:V4HI 1 "nonimmediate_operand" "%0"))
|
|
803 (zero_extend:V4SI
|
|
804 (match_operand:V4HI 2 "nonimmediate_operand" "ym")))
|
|
805 (const_int 16))))]
|
|
806 "(TARGET_SSE || TARGET_3DNOW_A)
|
|
807 && ix86_binary_operator_ok (MULT, V4HImode, operands)"
|
|
808 "pmulhuw\t{%2, %0|%0, %2}"
|
|
809 [(set_attr "type" "mmxmul")
|
|
810 (set_attr "mode" "DI")])
|
|
811
|
|
812 (define_expand "mmx_pmaddwd"
|
|
813 [(set (match_operand:V2SI 0 "register_operand" "")
|
|
814 (plus:V2SI
|
|
815 (mult:V2SI
|
|
816 (sign_extend:V2SI
|
|
817 (vec_select:V2HI
|
|
818 (match_operand:V4HI 1 "nonimmediate_operand" "")
|
|
819 (parallel [(const_int 0) (const_int 2)])))
|
|
820 (sign_extend:V2SI
|
|
821 (vec_select:V2HI
|
|
822 (match_operand:V4HI 2 "nonimmediate_operand" "")
|
|
823 (parallel [(const_int 0) (const_int 2)]))))
|
|
824 (mult:V2SI
|
|
825 (sign_extend:V2SI
|
|
826 (vec_select:V2HI (match_dup 1)
|
|
827 (parallel [(const_int 1) (const_int 3)])))
|
|
828 (sign_extend:V2SI
|
|
829 (vec_select:V2HI (match_dup 2)
|
|
830 (parallel [(const_int 1) (const_int 3)]))))))]
|
|
831 "TARGET_MMX"
|
|
832 "ix86_fixup_binary_operands_no_copy (MULT, V4HImode, operands);")
|
|
833
|
|
834 (define_insn "*mmx_pmaddwd"
|
|
835 [(set (match_operand:V2SI 0 "register_operand" "=y")
|
|
836 (plus:V2SI
|
|
837 (mult:V2SI
|
|
838 (sign_extend:V2SI
|
|
839 (vec_select:V2HI
|
|
840 (match_operand:V4HI 1 "nonimmediate_operand" "%0")
|
|
841 (parallel [(const_int 0) (const_int 2)])))
|
|
842 (sign_extend:V2SI
|
|
843 (vec_select:V2HI
|
|
844 (match_operand:V4HI 2 "nonimmediate_operand" "ym")
|
|
845 (parallel [(const_int 0) (const_int 2)]))))
|
|
846 (mult:V2SI
|
|
847 (sign_extend:V2SI
|
|
848 (vec_select:V2HI (match_dup 1)
|
|
849 (parallel [(const_int 1) (const_int 3)])))
|
|
850 (sign_extend:V2SI
|
|
851 (vec_select:V2HI (match_dup 2)
|
|
852 (parallel [(const_int 1) (const_int 3)]))))))]
|
|
853 "TARGET_MMX && ix86_binary_operator_ok (MULT, V4HImode, operands)"
|
|
854 "pmaddwd\t{%2, %0|%0, %2}"
|
|
855 [(set_attr "type" "mmxmul")
|
|
856 (set_attr "mode" "DI")])
|
|
857
|
|
858 (define_expand "mmx_pmulhrwv4hi3"
|
|
859 [(set (match_operand:V4HI 0 "register_operand" "")
|
|
860 (truncate:V4HI
|
|
861 (lshiftrt:V4SI
|
|
862 (plus:V4SI
|
|
863 (mult:V4SI
|
|
864 (sign_extend:V4SI
|
|
865 (match_operand:V4HI 1 "nonimmediate_operand" ""))
|
|
866 (sign_extend:V4SI
|
|
867 (match_operand:V4HI 2 "nonimmediate_operand" "")))
|
|
868 (const_vector:V4SI [(const_int 32768) (const_int 32768)
|
|
869 (const_int 32768) (const_int 32768)]))
|
|
870 (const_int 16))))]
|
|
871 "TARGET_3DNOW"
|
|
872 "ix86_fixup_binary_operands_no_copy (MULT, V4HImode, operands);")
|
|
873
|
|
874 (define_insn "*mmx_pmulhrwv4hi3"
|
|
875 [(set (match_operand:V4HI 0 "register_operand" "=y")
|
|
876 (truncate:V4HI
|
|
877 (lshiftrt:V4SI
|
|
878 (plus:V4SI
|
|
879 (mult:V4SI
|
|
880 (sign_extend:V4SI
|
|
881 (match_operand:V4HI 1 "nonimmediate_operand" "%0"))
|
|
882 (sign_extend:V4SI
|
|
883 (match_operand:V4HI 2 "nonimmediate_operand" "ym")))
|
|
884 (const_vector:V4SI [(const_int 32768) (const_int 32768)
|
|
885 (const_int 32768) (const_int 32768)]))
|
|
886 (const_int 16))))]
|
|
887 "TARGET_3DNOW && ix86_binary_operator_ok (MULT, V4HImode, operands)"
|
|
888 "pmulhrw\t{%2, %0|%0, %2}"
|
|
889 [(set_attr "type" "mmxmul")
|
|
890 (set_attr "mode" "DI")])
|
|
891
|
|
892 (define_expand "sse2_umulv1siv1di3"
|
|
893 [(set (match_operand:V1DI 0 "register_operand" "")
|
|
894 (mult:V1DI
|
|
895 (zero_extend:V1DI
|
|
896 (vec_select:V1SI
|
|
897 (match_operand:V2SI 1 "nonimmediate_operand" "")
|
|
898 (parallel [(const_int 0)])))
|
|
899 (zero_extend:V1DI
|
|
900 (vec_select:V1SI
|
|
901 (match_operand:V2SI 2 "nonimmediate_operand" "")
|
|
902 (parallel [(const_int 0)])))))]
|
|
903 "TARGET_SSE2"
|
|
904 "ix86_fixup_binary_operands_no_copy (MULT, V2SImode, operands);")
|
|
905
|
|
906 (define_insn "*sse2_umulv1siv1di3"
|
|
907 [(set (match_operand:V1DI 0 "register_operand" "=y")
|
|
908 (mult:V1DI
|
|
909 (zero_extend:V1DI
|
|
910 (vec_select:V1SI
|
|
911 (match_operand:V2SI 1 "nonimmediate_operand" "%0")
|
|
912 (parallel [(const_int 0)])))
|
|
913 (zero_extend:V1DI
|
|
914 (vec_select:V1SI
|
|
915 (match_operand:V2SI 2 "nonimmediate_operand" "ym")
|
|
916 (parallel [(const_int 0)])))))]
|
|
917 "TARGET_SSE2 && ix86_binary_operator_ok (MULT, V2SImode, operands)"
|
|
918 "pmuludq\t{%2, %0|%0, %2}"
|
|
919 [(set_attr "type" "mmxmul")
|
|
920 (set_attr "mode" "DI")])
|
|
921
|
|
922 (define_expand "mmx_<code>v4hi3"
|
|
923 [(set (match_operand:V4HI 0 "register_operand" "")
|
|
924 (smaxmin:V4HI
|
|
925 (match_operand:V4HI 1 "nonimmediate_operand" "")
|
|
926 (match_operand:V4HI 2 "nonimmediate_operand" "")))]
|
|
927 "TARGET_SSE || TARGET_3DNOW_A"
|
|
928 "ix86_fixup_binary_operands_no_copy (<CODE>, V4HImode, operands);")
|
|
929
|
|
930 (define_insn "*mmx_<code>v4hi3"
|
|
931 [(set (match_operand:V4HI 0 "register_operand" "=y")
|
|
932 (smaxmin:V4HI
|
|
933 (match_operand:V4HI 1 "nonimmediate_operand" "%0")
|
|
934 (match_operand:V4HI 2 "nonimmediate_operand" "ym")))]
|
|
935 "(TARGET_SSE || TARGET_3DNOW_A)
|
|
936 && ix86_binary_operator_ok (<CODE>, V4HImode, operands)"
|
|
937 "p<maxminiprefix>w\t{%2, %0|%0, %2}"
|
|
938 [(set_attr "type" "mmxadd")
|
|
939 (set_attr "mode" "DI")])
|
|
940
|
|
941 (define_expand "mmx_<code>v8qi3"
|
|
942 [(set (match_operand:V8QI 0 "register_operand" "")
|
|
943 (umaxmin:V8QI
|
|
944 (match_operand:V8QI 1 "nonimmediate_operand" "")
|
|
945 (match_operand:V8QI 2 "nonimmediate_operand" "")))]
|
|
946 "TARGET_SSE || TARGET_3DNOW_A"
|
|
947 "ix86_fixup_binary_operands_no_copy (<CODE>, V8QImode, operands);")
|
|
948
|
|
949 (define_insn "*mmx_<code>v8qi3"
|
|
950 [(set (match_operand:V8QI 0 "register_operand" "=y")
|
|
951 (umaxmin:V8QI
|
|
952 (match_operand:V8QI 1 "nonimmediate_operand" "%0")
|
|
953 (match_operand:V8QI 2 "nonimmediate_operand" "ym")))]
|
|
954 "(TARGET_SSE || TARGET_3DNOW_A)
|
|
955 && ix86_binary_operator_ok (<CODE>, V8QImode, operands)"
|
|
956 "p<maxminiprefix>b\t{%2, %0|%0, %2}"
|
|
957 [(set_attr "type" "mmxadd")
|
|
958 (set_attr "mode" "DI")])
|
|
959
|
|
960 (define_insn "mmx_ashr<mode>3"
|
|
961 [(set (match_operand:MMXMODE24 0 "register_operand" "=y")
|
|
962 (ashiftrt:MMXMODE24
|
|
963 (match_operand:MMXMODE24 1 "register_operand" "0")
|
|
964 (match_operand:SI 2 "nonmemory_operand" "yN")))]
|
|
965 "TARGET_MMX"
|
|
966 "psra<mmxvecsize>\t{%2, %0|%0, %2}"
|
|
967 [(set_attr "type" "mmxshft")
|
|
968 (set_attr "mode" "DI")])
|
|
969
|
|
970 (define_insn "mmx_lshr<mode>3"
|
|
971 [(set (match_operand:MMXMODE248 0 "register_operand" "=y")
|
|
972 (lshiftrt:MMXMODE248
|
|
973 (match_operand:MMXMODE248 1 "register_operand" "0")
|
|
974 (match_operand:SI 2 "nonmemory_operand" "yN")))]
|
|
975 "TARGET_MMX"
|
|
976 "psrl<mmxvecsize>\t{%2, %0|%0, %2}"
|
|
977 [(set_attr "type" "mmxshft")
|
|
978 (set_attr "mode" "DI")])
|
|
979
|
|
980 (define_insn "mmx_ashl<mode>3"
|
|
981 [(set (match_operand:MMXMODE248 0 "register_operand" "=y")
|
|
982 (ashift:MMXMODE248
|
|
983 (match_operand:MMXMODE248 1 "register_operand" "0")
|
|
984 (match_operand:SI 2 "nonmemory_operand" "yN")))]
|
|
985 "TARGET_MMX"
|
|
986 "psll<mmxvecsize>\t{%2, %0|%0, %2}"
|
|
987 [(set_attr "type" "mmxshft")
|
|
988 (set_attr "mode" "DI")])
|
|
989
|
|
990 ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
|
|
991 ;;
|
|
992 ;; Parallel integral comparisons
|
|
993 ;;
|
|
994 ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
|
|
995
|
|
996 (define_expand "mmx_eq<mode>3"
|
|
997 [(set (match_operand:MMXMODEI 0 "register_operand" "")
|
|
998 (eq:MMXMODEI
|
|
999 (match_operand:MMXMODEI 1 "nonimmediate_operand" "")
|
|
1000 (match_operand:MMXMODEI 2 "nonimmediate_operand" "")))]
|
|
1001 "TARGET_MMX"
|
|
1002 "ix86_fixup_binary_operands_no_copy (EQ, <MODE>mode, operands);")
|
|
1003
|
|
1004 (define_insn "*mmx_eq<mode>3"
|
|
1005 [(set (match_operand:MMXMODEI 0 "register_operand" "=y")
|
|
1006 (eq:MMXMODEI
|
|
1007 (match_operand:MMXMODEI 1 "nonimmediate_operand" "%0")
|
|
1008 (match_operand:MMXMODEI 2 "nonimmediate_operand" "ym")))]
|
|
1009 "TARGET_MMX && ix86_binary_operator_ok (EQ, <MODE>mode, operands)"
|
|
1010 "pcmpeq<mmxvecsize>\t{%2, %0|%0, %2}"
|
|
1011 [(set_attr "type" "mmxcmp")
|
|
1012 (set_attr "mode" "DI")])
|
|
1013
|
|
1014 (define_insn "mmx_gt<mode>3"
|
|
1015 [(set (match_operand:MMXMODEI 0 "register_operand" "=y")
|
|
1016 (gt:MMXMODEI
|
|
1017 (match_operand:MMXMODEI 1 "register_operand" "0")
|
|
1018 (match_operand:MMXMODEI 2 "nonimmediate_operand" "ym")))]
|
|
1019 "TARGET_MMX"
|
|
1020 "pcmpgt<mmxvecsize>\t{%2, %0|%0, %2}"
|
|
1021 [(set_attr "type" "mmxcmp")
|
|
1022 (set_attr "mode" "DI")])
|
|
1023
|
|
1024 ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
|
|
1025 ;;
|
|
1026 ;; Parallel integral logical operations
|
|
1027 ;;
|
|
1028 ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
|
|
1029
|
|
1030 (define_insn "mmx_andnot<mode>3"
|
|
1031 [(set (match_operand:MMXMODEI 0 "register_operand" "=y")
|
|
1032 (and:MMXMODEI
|
|
1033 (not:MMXMODEI (match_operand:MMXMODEI 1 "register_operand" "0"))
|
|
1034 (match_operand:MMXMODEI 2 "nonimmediate_operand" "ym")))]
|
|
1035 "TARGET_MMX"
|
|
1036 "pandn\t{%2, %0|%0, %2}"
|
|
1037 [(set_attr "type" "mmxadd")
|
|
1038 (set_attr "mode" "DI")])
|
|
1039
|
|
1040 (define_expand "mmx_<code><mode>3"
|
|
1041 [(set (match_operand:MMXMODEI 0 "register_operand" "")
|
|
1042 (plogic:MMXMODEI
|
|
1043 (match_operand:MMXMODEI 1 "nonimmediate_operand" "")
|
|
1044 (match_operand:MMXMODEI 2 "nonimmediate_operand" "")))]
|
|
1045 "TARGET_MMX"
|
|
1046 "ix86_fixup_binary_operands_no_copy (<CODE>, <MODE>mode, operands);")
|
|
1047
|
|
1048 (define_insn "*mmx_<code><mode>3"
|
|
1049 [(set (match_operand:MMXMODEI 0 "register_operand" "=y")
|
|
1050 (plogic:MMXMODEI
|
|
1051 (match_operand:MMXMODEI 1 "nonimmediate_operand" "%0")
|
|
1052 (match_operand:MMXMODEI 2 "nonimmediate_operand" "ym")))]
|
|
1053 "TARGET_MMX && ix86_binary_operator_ok (<CODE>, <MODE>mode, operands)"
|
|
1054 "p<plogicprefix>\t{%2, %0|%0, %2}"
|
|
1055 [(set_attr "type" "mmxadd")
|
|
1056 (set_attr "mode" "DI")])
|
|
1057
|
|
1058 ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
|
|
1059 ;;
|
|
1060 ;; Parallel integral element swizzling
|
|
1061 ;;
|
|
1062 ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
|
|
1063
|
|
1064 (define_insn "mmx_packsswb"
|
|
1065 [(set (match_operand:V8QI 0 "register_operand" "=y")
|
|
1066 (vec_concat:V8QI
|
|
1067 (ss_truncate:V4QI
|
|
1068 (match_operand:V4HI 1 "register_operand" "0"))
|
|
1069 (ss_truncate:V4QI
|
|
1070 (match_operand:V4HI 2 "nonimmediate_operand" "ym"))))]
|
|
1071 "TARGET_MMX"
|
|
1072 "packsswb\t{%2, %0|%0, %2}"
|
|
1073 [(set_attr "type" "mmxshft")
|
|
1074 (set_attr "mode" "DI")])
|
|
1075
|
|
1076 (define_insn "mmx_packssdw"
|
|
1077 [(set (match_operand:V4HI 0 "register_operand" "=y")
|
|
1078 (vec_concat:V4HI
|
|
1079 (ss_truncate:V2HI
|
|
1080 (match_operand:V2SI 1 "register_operand" "0"))
|
|
1081 (ss_truncate:V2HI
|
|
1082 (match_operand:V2SI 2 "nonimmediate_operand" "ym"))))]
|
|
1083 "TARGET_MMX"
|
|
1084 "packssdw\t{%2, %0|%0, %2}"
|
|
1085 [(set_attr "type" "mmxshft")
|
|
1086 (set_attr "mode" "DI")])
|
|
1087
|
|
1088 (define_insn "mmx_packuswb"
|
|
1089 [(set (match_operand:V8QI 0 "register_operand" "=y")
|
|
1090 (vec_concat:V8QI
|
|
1091 (us_truncate:V4QI
|
|
1092 (match_operand:V4HI 1 "register_operand" "0"))
|
|
1093 (us_truncate:V4QI
|
|
1094 (match_operand:V4HI 2 "nonimmediate_operand" "ym"))))]
|
|
1095 "TARGET_MMX"
|
|
1096 "packuswb\t{%2, %0|%0, %2}"
|
|
1097 [(set_attr "type" "mmxshft")
|
|
1098 (set_attr "mode" "DI")])
|
|
1099
|
|
1100 (define_insn "mmx_punpckhbw"
|
|
1101 [(set (match_operand:V8QI 0 "register_operand" "=y")
|
|
1102 (vec_select:V8QI
|
|
1103 (vec_concat:V16QI
|
|
1104 (match_operand:V8QI 1 "register_operand" "0")
|
|
1105 (match_operand:V8QI 2 "nonimmediate_operand" "ym"))
|
|
1106 (parallel [(const_int 4) (const_int 12)
|
|
1107 (const_int 5) (const_int 13)
|
|
1108 (const_int 6) (const_int 14)
|
|
1109 (const_int 7) (const_int 15)])))]
|
|
1110 "TARGET_MMX"
|
|
1111 "punpckhbw\t{%2, %0|%0, %2}"
|
|
1112 [(set_attr "type" "mmxcvt")
|
|
1113 (set_attr "mode" "DI")])
|
|
1114
|
|
1115 (define_insn "mmx_punpcklbw"
|
|
1116 [(set (match_operand:V8QI 0 "register_operand" "=y")
|
|
1117 (vec_select:V8QI
|
|
1118 (vec_concat:V16QI
|
|
1119 (match_operand:V8QI 1 "register_operand" "0")
|
|
1120 (match_operand:V8QI 2 "nonimmediate_operand" "ym"))
|
|
1121 (parallel [(const_int 0) (const_int 8)
|
|
1122 (const_int 1) (const_int 9)
|
|
1123 (const_int 2) (const_int 10)
|
|
1124 (const_int 3) (const_int 11)])))]
|
|
1125 "TARGET_MMX"
|
|
1126 "punpcklbw\t{%2, %0|%0, %2}"
|
|
1127 [(set_attr "type" "mmxcvt")
|
|
1128 (set_attr "mode" "DI")])
|
|
1129
|
|
1130 (define_insn "mmx_punpckhwd"
|
|
1131 [(set (match_operand:V4HI 0 "register_operand" "=y")
|
|
1132 (vec_select:V4HI
|
|
1133 (vec_concat:V8HI
|
|
1134 (match_operand:V4HI 1 "register_operand" "0")
|
|
1135 (match_operand:V4HI 2 "nonimmediate_operand" "ym"))
|
|
1136 (parallel [(const_int 2) (const_int 6)
|
|
1137 (const_int 3) (const_int 7)])))]
|
|
1138 "TARGET_MMX"
|
|
1139 "punpckhwd\t{%2, %0|%0, %2}"
|
|
1140 [(set_attr "type" "mmxcvt")
|
|
1141 (set_attr "mode" "DI")])
|
|
1142
|
|
1143 (define_insn "mmx_punpcklwd"
|
|
1144 [(set (match_operand:V4HI 0 "register_operand" "=y")
|
|
1145 (vec_select:V4HI
|
|
1146 (vec_concat:V8HI
|
|
1147 (match_operand:V4HI 1 "register_operand" "0")
|
|
1148 (match_operand:V4HI 2 "nonimmediate_operand" "ym"))
|
|
1149 (parallel [(const_int 0) (const_int 4)
|
|
1150 (const_int 1) (const_int 5)])))]
|
|
1151 "TARGET_MMX"
|
|
1152 "punpcklwd\t{%2, %0|%0, %2}"
|
|
1153 [(set_attr "type" "mmxcvt")
|
|
1154 (set_attr "mode" "DI")])
|
|
1155
|
|
1156 (define_insn "mmx_punpckhdq"
|
|
1157 [(set (match_operand:V2SI 0 "register_operand" "=y")
|
|
1158 (vec_select:V2SI
|
|
1159 (vec_concat:V4SI
|
|
1160 (match_operand:V2SI 1 "register_operand" "0")
|
|
1161 (match_operand:V2SI 2 "nonimmediate_operand" "ym"))
|
|
1162 (parallel [(const_int 1)
|
|
1163 (const_int 3)])))]
|
|
1164 "TARGET_MMX"
|
|
1165 "punpckhdq\t{%2, %0|%0, %2}"
|
|
1166 [(set_attr "type" "mmxcvt")
|
|
1167 (set_attr "mode" "DI")])
|
|
1168
|
|
1169 (define_insn "mmx_punpckldq"
|
|
1170 [(set (match_operand:V2SI 0 "register_operand" "=y")
|
|
1171 (vec_select:V2SI
|
|
1172 (vec_concat:V4SI
|
|
1173 (match_operand:V2SI 1 "register_operand" "0")
|
|
1174 (match_operand:V2SI 2 "nonimmediate_operand" "ym"))
|
|
1175 (parallel [(const_int 0)
|
|
1176 (const_int 2)])))]
|
|
1177 "TARGET_MMX"
|
|
1178 "punpckldq\t{%2, %0|%0, %2}"
|
|
1179 [(set_attr "type" "mmxcvt")
|
|
1180 (set_attr "mode" "DI")])
|
|
1181
|
|
1182 (define_expand "mmx_pinsrw"
|
|
1183 [(set (match_operand:V4HI 0 "register_operand" "")
|
|
1184 (vec_merge:V4HI
|
|
1185 (vec_duplicate:V4HI
|
|
1186 (match_operand:SI 2 "nonimmediate_operand" ""))
|
|
1187 (match_operand:V4HI 1 "register_operand" "")
|
|
1188 (match_operand:SI 3 "const_0_to_3_operand" "")))]
|
|
1189 "TARGET_SSE || TARGET_3DNOW_A"
|
|
1190 {
|
|
1191 operands[2] = gen_lowpart (HImode, operands[2]);
|
|
1192 operands[3] = GEN_INT (1 << INTVAL (operands[3]));
|
|
1193 })
|
|
1194
|
|
1195 (define_insn "*mmx_pinsrw"
|
|
1196 [(set (match_operand:V4HI 0 "register_operand" "=y")
|
|
1197 (vec_merge:V4HI
|
|
1198 (vec_duplicate:V4HI
|
|
1199 (match_operand:HI 2 "nonimmediate_operand" "rm"))
|
|
1200 (match_operand:V4HI 1 "register_operand" "0")
|
|
1201 (match_operand:SI 3 "const_pow2_1_to_8_operand" "n")))]
|
|
1202 "TARGET_SSE || TARGET_3DNOW_A"
|
|
1203 {
|
|
1204 operands[3] = GEN_INT (exact_log2 (INTVAL (operands[3])));
|
|
1205 return "pinsrw\t{%3, %k2, %0|%0, %k2, %3}";
|
|
1206 }
|
|
1207 [(set_attr "type" "mmxcvt")
|
|
1208 (set_attr "mode" "DI")])
|
|
1209
|
|
1210 (define_insn "mmx_pextrw"
|
|
1211 [(set (match_operand:SI 0 "register_operand" "=r")
|
|
1212 (zero_extend:SI
|
|
1213 (vec_select:HI
|
|
1214 (match_operand:V4HI 1 "register_operand" "y")
|
|
1215 (parallel [(match_operand:SI 2 "const_0_to_3_operand" "n")]))))]
|
|
1216 "TARGET_SSE || TARGET_3DNOW_A"
|
|
1217 "pextrw\t{%2, %1, %0|%0, %1, %2}"
|
|
1218 [(set_attr "type" "mmxcvt")
|
|
1219 (set_attr "mode" "DI")])
|
|
1220
|
|
1221 (define_expand "mmx_pshufw"
|
|
1222 [(match_operand:V4HI 0 "register_operand" "")
|
|
1223 (match_operand:V4HI 1 "nonimmediate_operand" "")
|
|
1224 (match_operand:SI 2 "const_int_operand" "")]
|
|
1225 "TARGET_SSE || TARGET_3DNOW_A"
|
|
1226 {
|
|
1227 int mask = INTVAL (operands[2]);
|
|
1228 emit_insn (gen_mmx_pshufw_1 (operands[0], operands[1],
|
|
1229 GEN_INT ((mask >> 0) & 3),
|
|
1230 GEN_INT ((mask >> 2) & 3),
|
|
1231 GEN_INT ((mask >> 4) & 3),
|
|
1232 GEN_INT ((mask >> 6) & 3)));
|
|
1233 DONE;
|
|
1234 })
|
|
1235
|
|
1236 (define_insn "mmx_pshufw_1"
|
|
1237 [(set (match_operand:V4HI 0 "register_operand" "=y")
|
|
1238 (vec_select:V4HI
|
|
1239 (match_operand:V4HI 1 "nonimmediate_operand" "ym")
|
|
1240 (parallel [(match_operand 2 "const_0_to_3_operand" "")
|
|
1241 (match_operand 3 "const_0_to_3_operand" "")
|
|
1242 (match_operand 4 "const_0_to_3_operand" "")
|
|
1243 (match_operand 5 "const_0_to_3_operand" "")])))]
|
|
1244 "TARGET_SSE || TARGET_3DNOW_A"
|
|
1245 {
|
|
1246 int mask = 0;
|
|
1247 mask |= INTVAL (operands[2]) << 0;
|
|
1248 mask |= INTVAL (operands[3]) << 2;
|
|
1249 mask |= INTVAL (operands[4]) << 4;
|
|
1250 mask |= INTVAL (operands[5]) << 6;
|
|
1251 operands[2] = GEN_INT (mask);
|
|
1252
|
|
1253 return "pshufw\t{%2, %1, %0|%0, %1, %2}";
|
|
1254 }
|
|
1255 [(set_attr "type" "mmxcvt")
|
|
1256 (set_attr "mode" "DI")])
|
|
1257
|
|
1258 (define_insn "mmx_pswapdv2si2"
|
|
1259 [(set (match_operand:V2SI 0 "register_operand" "=y")
|
|
1260 (vec_select:V2SI
|
|
1261 (match_operand:V2SI 1 "nonimmediate_operand" "ym")
|
|
1262 (parallel [(const_int 1) (const_int 0)])))]
|
|
1263 "TARGET_3DNOW_A"
|
|
1264 "pswapd\t{%1, %0|%0, %1}"
|
|
1265 [(set_attr "type" "mmxcvt")
|
|
1266 (set_attr "mode" "DI")])
|
|
1267
|
|
1268 (define_insn "*vec_dupv4hi"
|
|
1269 [(set (match_operand:V4HI 0 "register_operand" "=y")
|
|
1270 (vec_duplicate:V4HI
|
|
1271 (truncate:HI
|
|
1272 (match_operand:SI 1 "register_operand" "0"))))]
|
|
1273 "TARGET_SSE || TARGET_3DNOW_A"
|
|
1274 "pshufw\t{$0, %0, %0|%0, %0, 0}"
|
|
1275 [(set_attr "type" "mmxcvt")
|
|
1276 (set_attr "mode" "DI")])
|
|
1277
|
|
1278 (define_insn "*vec_dupv2si"
|
|
1279 [(set (match_operand:V2SI 0 "register_operand" "=y")
|
|
1280 (vec_duplicate:V2SI
|
|
1281 (match_operand:SI 1 "register_operand" "0")))]
|
|
1282 "TARGET_MMX"
|
|
1283 "punpckldq\t%0, %0"
|
|
1284 [(set_attr "type" "mmxcvt")
|
|
1285 (set_attr "mode" "DI")])
|
|
1286
|
|
1287 (define_insn "*mmx_concatv2si"
|
|
1288 [(set (match_operand:V2SI 0 "register_operand" "=y,y")
|
|
1289 (vec_concat:V2SI
|
|
1290 (match_operand:SI 1 "nonimmediate_operand" " 0,rm")
|
|
1291 (match_operand:SI 2 "vector_move_operand" "ym,C")))]
|
|
1292 "TARGET_MMX && !TARGET_SSE"
|
|
1293 "@
|
|
1294 punpckldq\t{%2, %0|%0, %2}
|
|
1295 movd\t{%1, %0|%0, %1}"
|
|
1296 [(set_attr "type" "mmxcvt,mmxmov")
|
|
1297 (set_attr "mode" "DI")])
|
|
1298
|
|
1299 (define_expand "vec_setv2si"
|
|
1300 [(match_operand:V2SI 0 "register_operand" "")
|
|
1301 (match_operand:SI 1 "register_operand" "")
|
|
1302 (match_operand 2 "const_int_operand" "")]
|
|
1303 "TARGET_MMX"
|
|
1304 {
|
|
1305 ix86_expand_vector_set (false, operands[0], operands[1],
|
|
1306 INTVAL (operands[2]));
|
|
1307 DONE;
|
|
1308 })
|
|
1309
|
|
1310 ;; Avoid combining registers from different units in a single alternative,
|
|
1311 ;; see comment above inline_secondary_memory_needed function in i386.c
|
|
1312 (define_insn_and_split "*vec_extractv2si_0"
|
|
1313 [(set (match_operand:SI 0 "nonimmediate_operand" "=x,m,y, m,r")
|
|
1314 (vec_select:SI
|
|
1315 (match_operand:V2SI 1 "nonimmediate_operand" "xm,x,ym,y,m")
|
|
1316 (parallel [(const_int 0)])))]
|
|
1317 "TARGET_MMX && !(MEM_P (operands[0]) && MEM_P (operands[1]))"
|
|
1318 "#"
|
|
1319 "&& reload_completed"
|
|
1320 [(const_int 0)]
|
|
1321 {
|
|
1322 rtx op1 = operands[1];
|
|
1323 if (REG_P (op1))
|
|
1324 op1 = gen_rtx_REG (SImode, REGNO (op1));
|
|
1325 else
|
|
1326 op1 = gen_lowpart (SImode, op1);
|
|
1327 emit_move_insn (operands[0], op1);
|
|
1328 DONE;
|
|
1329 })
|
|
1330
|
|
1331 ;; Avoid combining registers from different units in a single alternative,
|
|
1332 ;; see comment above inline_secondary_memory_needed function in i386.c
|
|
1333 (define_insn "*vec_extractv2si_1"
|
|
1334 [(set (match_operand:SI 0 "nonimmediate_operand" "=y,Y2,Y2,x,y,x,r")
|
|
1335 (vec_select:SI
|
|
1336 (match_operand:V2SI 1 "nonimmediate_operand" " 0,0 ,Y2,0,o,o,o")
|
|
1337 (parallel [(const_int 1)])))]
|
|
1338 "TARGET_MMX && !(MEM_P (operands[0]) && MEM_P (operands[1]))"
|
|
1339 "@
|
|
1340 punpckhdq\t%0, %0
|
|
1341 punpckhdq\t%0, %0
|
|
1342 pshufd\t{$85, %1, %0|%0, %1, 85}
|
|
1343 unpckhps\t%0, %0
|
|
1344 #
|
|
1345 #
|
|
1346 #"
|
|
1347 [(set_attr "type" "mmxcvt,sselog1,sselog1,sselog1,mmxmov,ssemov,imov")
|
|
1348 (set_attr "mode" "DI,TI,TI,V4SF,SI,SI,SI")])
|
|
1349
|
|
1350 (define_split
|
|
1351 [(set (match_operand:SI 0 "register_operand" "")
|
|
1352 (vec_select:SI
|
|
1353 (match_operand:V2SI 1 "memory_operand" "")
|
|
1354 (parallel [(const_int 1)])))]
|
|
1355 "TARGET_MMX && reload_completed"
|
|
1356 [(const_int 0)]
|
|
1357 {
|
|
1358 operands[1] = adjust_address (operands[1], SImode, 4);
|
|
1359 emit_move_insn (operands[0], operands[1]);
|
|
1360 DONE;
|
|
1361 })
|
|
1362
|
|
1363 (define_expand "vec_extractv2si"
|
|
1364 [(match_operand:SI 0 "register_operand" "")
|
|
1365 (match_operand:V2SI 1 "register_operand" "")
|
|
1366 (match_operand 2 "const_int_operand" "")]
|
|
1367 "TARGET_MMX"
|
|
1368 {
|
|
1369 ix86_expand_vector_extract (false, operands[0], operands[1],
|
|
1370 INTVAL (operands[2]));
|
|
1371 DONE;
|
|
1372 })
|
|
1373
|
|
1374 (define_expand "vec_initv2si"
|
|
1375 [(match_operand:V2SI 0 "register_operand" "")
|
|
1376 (match_operand 1 "" "")]
|
|
1377 "TARGET_SSE"
|
|
1378 {
|
|
1379 ix86_expand_vector_init (false, operands[0], operands[1]);
|
|
1380 DONE;
|
|
1381 })
|
|
1382
|
|
1383 (define_expand "vec_setv4hi"
|
|
1384 [(match_operand:V4HI 0 "register_operand" "")
|
|
1385 (match_operand:HI 1 "register_operand" "")
|
|
1386 (match_operand 2 "const_int_operand" "")]
|
|
1387 "TARGET_MMX"
|
|
1388 {
|
|
1389 ix86_expand_vector_set (false, operands[0], operands[1],
|
|
1390 INTVAL (operands[2]));
|
|
1391 DONE;
|
|
1392 })
|
|
1393
|
|
1394 (define_expand "vec_extractv4hi"
|
|
1395 [(match_operand:HI 0 "register_operand" "")
|
|
1396 (match_operand:V4HI 1 "register_operand" "")
|
|
1397 (match_operand 2 "const_int_operand" "")]
|
|
1398 "TARGET_MMX"
|
|
1399 {
|
|
1400 ix86_expand_vector_extract (false, operands[0], operands[1],
|
|
1401 INTVAL (operands[2]));
|
|
1402 DONE;
|
|
1403 })
|
|
1404
|
|
1405 (define_expand "vec_initv4hi"
|
|
1406 [(match_operand:V4HI 0 "register_operand" "")
|
|
1407 (match_operand 1 "" "")]
|
|
1408 "TARGET_SSE"
|
|
1409 {
|
|
1410 ix86_expand_vector_init (false, operands[0], operands[1]);
|
|
1411 DONE;
|
|
1412 })
|
|
1413
|
|
1414 (define_expand "vec_setv8qi"
|
|
1415 [(match_operand:V8QI 0 "register_operand" "")
|
|
1416 (match_operand:QI 1 "register_operand" "")
|
|
1417 (match_operand 2 "const_int_operand" "")]
|
|
1418 "TARGET_MMX"
|
|
1419 {
|
|
1420 ix86_expand_vector_set (false, operands[0], operands[1],
|
|
1421 INTVAL (operands[2]));
|
|
1422 DONE;
|
|
1423 })
|
|
1424
|
|
1425 (define_expand "vec_extractv8qi"
|
|
1426 [(match_operand:QI 0 "register_operand" "")
|
|
1427 (match_operand:V8QI 1 "register_operand" "")
|
|
1428 (match_operand 2 "const_int_operand" "")]
|
|
1429 "TARGET_MMX"
|
|
1430 {
|
|
1431 ix86_expand_vector_extract (false, operands[0], operands[1],
|
|
1432 INTVAL (operands[2]));
|
|
1433 DONE;
|
|
1434 })
|
|
1435
|
|
1436 (define_expand "vec_initv8qi"
|
|
1437 [(match_operand:V8QI 0 "register_operand" "")
|
|
1438 (match_operand 1 "" "")]
|
|
1439 "TARGET_SSE"
|
|
1440 {
|
|
1441 ix86_expand_vector_init (false, operands[0], operands[1]);
|
|
1442 DONE;
|
|
1443 })
|
|
1444
|
|
1445 ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
|
|
1446 ;;
|
|
1447 ;; Miscellaneous
|
|
1448 ;;
|
|
1449 ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
|
|
1450
|
|
1451 (define_expand "mmx_uavgv8qi3"
|
|
1452 [(set (match_operand:V8QI 0 "register_operand" "")
|
|
1453 (truncate:V8QI
|
|
1454 (lshiftrt:V8HI
|
|
1455 (plus:V8HI
|
|
1456 (plus:V8HI
|
|
1457 (zero_extend:V8HI
|
|
1458 (match_operand:V8QI 1 "nonimmediate_operand" ""))
|
|
1459 (zero_extend:V8HI
|
|
1460 (match_operand:V8QI 2 "nonimmediate_operand" "")))
|
|
1461 (const_vector:V8HI [(const_int 1) (const_int 1)
|
|
1462 (const_int 1) (const_int 1)
|
|
1463 (const_int 1) (const_int 1)
|
|
1464 (const_int 1) (const_int 1)]))
|
|
1465 (const_int 1))))]
|
|
1466 "TARGET_SSE || TARGET_3DNOW"
|
|
1467 "ix86_fixup_binary_operands_no_copy (PLUS, V8QImode, operands);")
|
|
1468
|
|
1469 (define_insn "*mmx_uavgv8qi3"
|
|
1470 [(set (match_operand:V8QI 0 "register_operand" "=y")
|
|
1471 (truncate:V8QI
|
|
1472 (lshiftrt:V8HI
|
|
1473 (plus:V8HI
|
|
1474 (plus:V8HI
|
|
1475 (zero_extend:V8HI
|
|
1476 (match_operand:V8QI 1 "nonimmediate_operand" "%0"))
|
|
1477 (zero_extend:V8HI
|
|
1478 (match_operand:V8QI 2 "nonimmediate_operand" "ym")))
|
|
1479 (const_vector:V8HI [(const_int 1) (const_int 1)
|
|
1480 (const_int 1) (const_int 1)
|
|
1481 (const_int 1) (const_int 1)
|
|
1482 (const_int 1) (const_int 1)]))
|
|
1483 (const_int 1))))]
|
|
1484 "(TARGET_SSE || TARGET_3DNOW)
|
|
1485 && ix86_binary_operator_ok (PLUS, V8QImode, operands)"
|
|
1486 {
|
|
1487 /* These two instructions have the same operation, but their encoding
|
|
1488 is different. Prefer the one that is de facto standard. */
|
|
1489 if (TARGET_SSE || TARGET_3DNOW_A)
|
|
1490 return "pavgb\t{%2, %0|%0, %2}";
|
|
1491 else
|
|
1492 return "pavgusb\t{%2, %0|%0, %2}";
|
|
1493 }
|
|
1494 [(set_attr "type" "mmxshft")
|
|
1495 (set_attr "mode" "DI")])
|
|
1496
|
|
1497 (define_expand "mmx_uavgv4hi3"
|
|
1498 [(set (match_operand:V4HI 0 "register_operand" "")
|
|
1499 (truncate:V4HI
|
|
1500 (lshiftrt:V4SI
|
|
1501 (plus:V4SI
|
|
1502 (plus:V4SI
|
|
1503 (zero_extend:V4SI
|
|
1504 (match_operand:V4HI 1 "nonimmediate_operand" ""))
|
|
1505 (zero_extend:V4SI
|
|
1506 (match_operand:V4HI 2 "nonimmediate_operand" "")))
|
|
1507 (const_vector:V4SI [(const_int 1) (const_int 1)
|
|
1508 (const_int 1) (const_int 1)]))
|
|
1509 (const_int 1))))]
|
|
1510 "TARGET_SSE || TARGET_3DNOW_A"
|
|
1511 "ix86_fixup_binary_operands_no_copy (PLUS, V4HImode, operands);")
|
|
1512
|
|
1513 (define_insn "*mmx_uavgv4hi3"
|
|
1514 [(set (match_operand:V4HI 0 "register_operand" "=y")
|
|
1515 (truncate:V4HI
|
|
1516 (lshiftrt:V4SI
|
|
1517 (plus:V4SI
|
|
1518 (plus:V4SI
|
|
1519 (zero_extend:V4SI
|
|
1520 (match_operand:V4HI 1 "nonimmediate_operand" "%0"))
|
|
1521 (zero_extend:V4SI
|
|
1522 (match_operand:V4HI 2 "nonimmediate_operand" "ym")))
|
|
1523 (const_vector:V4SI [(const_int 1) (const_int 1)
|
|
1524 (const_int 1) (const_int 1)]))
|
|
1525 (const_int 1))))]
|
|
1526 "(TARGET_SSE || TARGET_3DNOW_A)
|
|
1527 && ix86_binary_operator_ok (PLUS, V4HImode, operands)"
|
|
1528 "pavgw\t{%2, %0|%0, %2}"
|
|
1529 [(set_attr "type" "mmxshft")
|
|
1530 (set_attr "mode" "DI")])
|
|
1531
|
|
1532 (define_insn "mmx_psadbw"
|
|
1533 [(set (match_operand:V1DI 0 "register_operand" "=y")
|
|
1534 (unspec:V1DI [(match_operand:V8QI 1 "register_operand" "0")
|
|
1535 (match_operand:V8QI 2 "nonimmediate_operand" "ym")]
|
|
1536 UNSPEC_PSADBW))]
|
|
1537 "TARGET_SSE || TARGET_3DNOW_A"
|
|
1538 "psadbw\t{%2, %0|%0, %2}"
|
|
1539 [(set_attr "type" "mmxshft")
|
|
1540 (set_attr "mode" "DI")])
|
|
1541
|
|
1542 (define_insn "mmx_pmovmskb"
|
|
1543 [(set (match_operand:SI 0 "register_operand" "=r")
|
|
1544 (unspec:SI [(match_operand:V8QI 1 "register_operand" "y")]
|
|
1545 UNSPEC_MOVMSK))]
|
|
1546 "TARGET_SSE || TARGET_3DNOW_A"
|
|
1547 "pmovmskb\t{%1, %0|%0, %1}"
|
|
1548 [(set_attr "type" "mmxcvt")
|
|
1549 (set_attr "mode" "DI")])
|
|
1550
|
|
1551 (define_expand "mmx_maskmovq"
|
|
1552 [(set (match_operand:V8QI 0 "memory_operand" "")
|
|
1553 (unspec:V8QI [(match_operand:V8QI 1 "register_operand" "")
|
|
1554 (match_operand:V8QI 2 "register_operand" "")
|
|
1555 (match_dup 0)]
|
|
1556 UNSPEC_MASKMOV))]
|
|
1557 "TARGET_SSE || TARGET_3DNOW_A"
|
|
1558 "")
|
|
1559
|
|
1560 (define_insn "*mmx_maskmovq"
|
|
1561 [(set (mem:V8QI (match_operand:SI 0 "register_operand" "D"))
|
|
1562 (unspec:V8QI [(match_operand:V8QI 1 "register_operand" "y")
|
|
1563 (match_operand:V8QI 2 "register_operand" "y")
|
|
1564 (mem:V8QI (match_dup 0))]
|
|
1565 UNSPEC_MASKMOV))]
|
|
1566 "(TARGET_SSE || TARGET_3DNOW_A) && !TARGET_64BIT"
|
|
1567 ;; @@@ check ordering of operands in intel/nonintel syntax
|
|
1568 "maskmovq\t{%2, %1|%1, %2}"
|
|
1569 [(set_attr "type" "mmxcvt")
|
|
1570 (set_attr "mode" "DI")])
|
|
1571
|
|
1572 (define_insn "*mmx_maskmovq_rex"
|
|
1573 [(set (mem:V8QI (match_operand:DI 0 "register_operand" "D"))
|
|
1574 (unspec:V8QI [(match_operand:V8QI 1 "register_operand" "y")
|
|
1575 (match_operand:V8QI 2 "register_operand" "y")
|
|
1576 (mem:V8QI (match_dup 0))]
|
|
1577 UNSPEC_MASKMOV))]
|
|
1578 "(TARGET_SSE || TARGET_3DNOW_A) && TARGET_64BIT"
|
|
1579 ;; @@@ check ordering of operands in intel/nonintel syntax
|
|
1580 "maskmovq\t{%2, %1|%1, %2}"
|
|
1581 [(set_attr "type" "mmxcvt")
|
|
1582 (set_attr "mode" "DI")])
|
|
1583
|
|
1584 (define_insn "mmx_emms"
|
|
1585 [(unspec_volatile [(const_int 0)] UNSPECV_EMMS)
|
|
1586 (clobber (reg:XF ST0_REG))
|
|
1587 (clobber (reg:XF ST1_REG))
|
|
1588 (clobber (reg:XF ST2_REG))
|
|
1589 (clobber (reg:XF ST3_REG))
|
|
1590 (clobber (reg:XF ST4_REG))
|
|
1591 (clobber (reg:XF ST5_REG))
|
|
1592 (clobber (reg:XF ST6_REG))
|
|
1593 (clobber (reg:XF ST7_REG))
|
|
1594 (clobber (reg:DI MM0_REG))
|
|
1595 (clobber (reg:DI MM1_REG))
|
|
1596 (clobber (reg:DI MM2_REG))
|
|
1597 (clobber (reg:DI MM3_REG))
|
|
1598 (clobber (reg:DI MM4_REG))
|
|
1599 (clobber (reg:DI MM5_REG))
|
|
1600 (clobber (reg:DI MM6_REG))
|
|
1601 (clobber (reg:DI MM7_REG))]
|
|
1602 "TARGET_MMX"
|
|
1603 "emms"
|
|
1604 [(set_attr "type" "mmx")
|
|
1605 (set_attr "memory" "unknown")])
|
|
1606
|
|
1607 (define_insn "mmx_femms"
|
|
1608 [(unspec_volatile [(const_int 0)] UNSPECV_FEMMS)
|
|
1609 (clobber (reg:XF ST0_REG))
|
|
1610 (clobber (reg:XF ST1_REG))
|
|
1611 (clobber (reg:XF ST2_REG))
|
|
1612 (clobber (reg:XF ST3_REG))
|
|
1613 (clobber (reg:XF ST4_REG))
|
|
1614 (clobber (reg:XF ST5_REG))
|
|
1615 (clobber (reg:XF ST6_REG))
|
|
1616 (clobber (reg:XF ST7_REG))
|
|
1617 (clobber (reg:DI MM0_REG))
|
|
1618 (clobber (reg:DI MM1_REG))
|
|
1619 (clobber (reg:DI MM2_REG))
|
|
1620 (clobber (reg:DI MM3_REG))
|
|
1621 (clobber (reg:DI MM4_REG))
|
|
1622 (clobber (reg:DI MM5_REG))
|
|
1623 (clobber (reg:DI MM6_REG))
|
|
1624 (clobber (reg:DI MM7_REG))]
|
|
1625 "TARGET_3DNOW"
|
|
1626 "femms"
|
|
1627 [(set_attr "type" "mmx")
|
|
1628 (set_attr "memory" "none")])
|