1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
|
/* xmmx.h
eXtended MultiMedia eXtensions GCC interface library for IA32.
To use this library, simply include this header file
and compile with GCC. You MUST have inlining enabled
in order for xmmx_ok() to work; this can be done by
simply using -O on the GCC command line.
Compiling with -DXMMX_TRACE will cause detailed trace
output to be sent to stderr for each mmx operation.
This adds lots of code, and obviously slows execution to
a crawl, but can be very useful for debugging.
THIS SOFTWARE IS PROVIDED ``AS IS'' AND WITHOUT ANY
EXPRESS OR IMPLIED WARRANTIES, INCLUDING, WITHOUT
LIMITATION, THE IMPLIED WARRANTIES OF MERCHANTABILITY
AND FITNESS FOR ANY PARTICULAR PURPOSE.
1999 by R. Fisher
Based on libmmx, 1997-99 by H. Dietz and R. Fisher
Notes:
It appears that the latest gas has the pand problem fixed, therefore
I'll undefine BROKEN_PAND by default.
*/
#ifndef _XMMX_H
#define _XMMX_H
/* Warning: at this writing, the version of GAS packaged
with most Linux distributions does not handle the
parallel AND operation mnemonic correctly. If the
symbol BROKEN_PAND is defined, a slower alternative
coding will be used. If execution of mmxtest results
in an illegal instruction fault, define this symbol.
*/
#undef BROKEN_PAND
/* The type of an value that fits in an (Extended) MMX register
(note that long long constant values MUST be suffixed
by LL and unsigned long long values by ULL, lest
they be truncated by the compiler)
*/
#ifndef _MMX_H
typedef union {
long long q; /* Quadword (64-bit) value */
unsigned long long uq; /* Unsigned Quadword */
int d[2]; /* 2 Doubleword (32-bit) values */
unsigned int ud[2]; /* 2 Unsigned Doubleword */
short w[4]; /* 4 Word (16-bit) values */
unsigned short uw[4]; /* 4 Unsigned Word */
char b[8]; /* 8 Byte (8-bit) values */
unsigned char ub[8]; /* 8 Unsigned Byte */
float s[2]; /* Single-precision (32-bit) value */
} __attribute__ ((aligned (8))) mmx_t; /* On an 8-byte (64-bit) boundary */
#endif
/* Helper functions for the instruction macros that follow...
(note that memory-to-register, m2r, instructions are nearly
as efficient as register-to-register, r2r, instructions;
however, memory-to-memory instructions are really simulated
as a convenience, and are only 1/3 as efficient)
*/
#ifdef XMMX_TRACE
/* Include the stuff for printing a trace to stderr...
*/
#include <stdio.h>
#define mmx_i2r(op, imm, reg) \
{ \
mmx_t mmx_trace; \
mmx_trace.uq = (imm); \
fprintf(stderr, #op "_i2r(" #imm "=0x%08x%08x, ", \
mmx_trace.d[1], mmx_trace.d[0]); \
__asm__ __volatile__ ("movq %%" #reg ", %0" \
: "=X" (mmx_trace) \
: /* nothing */ ); \
fprintf(stderr, #reg "=0x%08x%08x) => ", \
mmx_trace.d[1], mmx_trace.d[0]); \
__asm__ __volatile__ (#op " %0, %%" #reg \
: /* nothing */ \
: "X" (imm)); \
__asm__ __volatile__ ("movq %%" #reg ", %0" \
: "=X" (mmx_trace) \
: /* nothing */ ); \
fprintf(stderr, #reg "=0x%08x%08x\n", \
mmx_trace.d[1], mmx_trace.d[0]); \
}
#define mmx_m2r(op, mem, reg) \
{ \
mmx_t mmx_trace; \
mmx_trace = (mem); \
fprintf(stderr, #op "_m2r(" #mem "=0x%08x%08x, ", \
mmx_trace.d[1], mmx_trace.d[0]); \
__asm__ __volatile__ ("movq %%" #reg ", %0" \
: "=X" (mmx_trace) \
: /* nothing */ ); \
fprintf(stderr, #reg "=0x%08x%08x) => ", \
mmx_trace.d[1], mmx_trace.d[0]); \
__asm__ __volatile__ (#op " %0, %%" #reg \
: /* nothing */ \
: "X" (mem)); \
__asm__ __volatile__ ("movq %%" #reg ", %0" \
: "=X" (mmx_trace) \
: /* nothing */ ); \
fprintf(stderr, #reg "=0x%08x%08x\n", \
mmx_trace.d[1], mmx_trace.d[0]); \
}
#define mmx_r2m(op, reg, mem) \
{ \
mmx_t mmx_trace; \
__asm__ __volatile__ ("movq %%" #reg ", %0" \
: "=X" (mmx_trace) \
: /* nothing */ ); \
fprintf(stderr, #op "_r2m(" #reg "=0x%08x%08x, ", \
mmx_trace.d[1], mmx_trace.d[0]); \
mmx_trace = (mem); \
fprintf(stderr, #mem "=0x%08x%08x) => ", \
mmx_trace.d[1], mmx_trace.d[0]); \
__asm__ __volatile__ (#op " %%" #reg ", %0" \
: "=X" (mem) \
: /* nothing */ ); \
mmx_trace = (mem); \
fprintf(stderr, #mem "=0x%08x%08x\n", \
mmx_trace.d[1], mmx_trace.d[0]); \
}
#define mmx_r2r(op, regs, regd) \
{ \
mmx_t mmx_trace; \
__asm__ __volatile__ ("movq %%" #regs ", %0" \
: "=X" (mmx_trace) \
: /* nothing */ ); \
fprintf(stderr, #op "_r2r(" #regs "=0x%08x%08x, ", \
mmx_trace.d[1], mmx_trace.d[0]); \
__asm__ __volatile__ ("movq %%" #regd ", %0" \
: "=X" (mmx_trace) \
: /* nothing */ ); \
fprintf(stderr, #regd "=0x%08x%08x) => ", \
mmx_trace.d[1], mmx_trace.d[0]); \
__asm__ __volatile__ (#op " %" #regs ", %" #regd); \
__asm__ __volatile__ ("movq %%" #regd ", %0" \
: "=X" (mmx_trace) \
: /* nothing */ ); \
fprintf(stderr, #regd "=0x%08x%08x\n", \
mmx_trace.d[1], mmx_trace.d[0]); \
}
#define mmx_m2m(op, mems, memd) \
{ \
mmx_t mmx_trace; \
mmx_trace = (mems); \
fprintf(stderr, #op "_m2m(" #mems "=0x%08x%08x, ", \
mmx_trace.d[1], mmx_trace.d[0]); \
mmx_trace = (memd); \
fprintf(stderr, #memd "=0x%08x%08x) => ", \
mmx_trace.d[1], mmx_trace.d[0]); \
__asm__ __volatile__ ("movq %0, %%mm0\n\t" \
#op " %1, %%mm0\n\t" \
"movq %%mm0, %0" \
: "=X" (memd) \
: "X" (mems)); \
mmx_trace = (memd); \
fprintf(stderr, #memd "=0x%08x%08x\n", \
mmx_trace.d[1], mmx_trace.d[0]); \
}
#else
/* These macros are a lot simpler without the tracing...
*/
#define mmx_i2r(op, imm, reg) \
__asm__ __volatile__ (#op " %0, %%" #reg \
: /* nothing */ \
: "X" (imm) )
#define mmx_m2r(op, mem, reg) \
__asm__ __volatile__ (#op " %0, %%" #reg \
: /* nothing */ \
: "X" (mem))
#define mmx_m2ir(op, mem, rs) \
__asm__ __volatile__ (#op " %0, %%" #rs \
: /* nothing */ \
: "X" (mem) )
#define mmx_r2m(op, reg, mem) \
__asm__ __volatile__ (#op " %%" #reg ", %0" \
: "=X" (mem) \
: /* nothing */ )
#define mmx_r2r(op, regs, regd) \
__asm__ __volatile__ (#op " %" #regs ", %" #regd)
#define mmx_r2ir(op, rs1, rs2) \
__asm__ __volatile__ (#op " %%" #rs1 ", %%" #rs2 \
: /* nothing */ \
: /* nothing */ )
#define mmx_m2m(op, mems, memd) \
__asm__ __volatile__ ("movq %0, %%mm0\n\t" \
#op " %1, %%mm0\n\t" \
"movq %%mm0, %0" \
: "=X" (memd) \
: "X" (mems))
#endif
/* 1x64 MOVe Quadword
(this is both a load and a store...
in fact, it is the only way to store)
*/
#define movq_m2r(var, reg) mmx_m2r(movq, var, reg)
#define movq_r2m(reg, var) mmx_r2m(movq, reg, var)
#define movq_r2r(regs, regd) mmx_r2r(movq, regs, regd)
#define movq(vars, vard) \
__asm__ __volatile__ ("movq %1, %%mm0\n\t" \
"movq %%mm0, %0" \
: "=X" (vard) \
: "X" (vars))
/* 1x32 MOVe Doubleword
(like movq, this is both load and store...
but is most useful for moving things between
mmx registers and ordinary registers)
*/
#define movd_m2r(var, reg) mmx_m2r(movd, var, reg)
#define movd_r2m(reg, var) mmx_r2m(movd, reg, var)
#define movd_r2r(regs, regd) mmx_r2r(movd, regs, regd)
#define movd(vars, vard) \
__asm__ __volatile__ ("movd %1, %%mm0\n\t" \
"movd %%mm0, %0" \
: "=X" (vard) \
: "X" (vars))
/* 4x16 Parallel MAGnitude
*/
#define pmagw_m2r(var, reg) mmx_m2r(pmagw, var, reg)
#define pmagw_r2r(regs, regd) mmx_r2r(pmagw, regs, regd)
#define pmagw(vars, vard) mmx_m2m(pmagw, vars, vard)
/* 4x16 Parallel ADDs using Saturation arithmetic
and Implied destination
*/
#define paddsiw_m2ir(var, rs) mmx_m2ir(paddsiw, var, rs)
#define paddsiw_r2ir(rs1, rs2) mmx_r2ir(paddsiw, rs1, rs2)
#define paddsiw(vars, vard) mmx_m2m(paddsiw, vars, vard)
/* 4x16 Parallel SUBs using Saturation arithmetic
and Implied destination
*/
#define psubsiw_m2ir(var, rs) mmx_m2ir(psubsiw, var, rs)
#define psubsiw_r2ir(rs1, rs2) mmx_r2ir(psubsiw, rs1, rs2)
#define psubsiw(vars, vard) mmx_m2m(psubsiw, vars, vard)
/* 4x16 Parallel MULs giving High 4x16 portions of results
Rounded with 1/2 bit 15.
*/
#define pmulhrw_m2r(var, reg) mmx_m2r(pmulhrw, var, reg)
#define pmulhrw_r2r(regs, regd) mmx_r2r(pmulhrw, regs, regd)
#define pmulhrw(vars, vard) mmx_m2m(pmulhrw, vars, vard)
/* 4x16 Parallel MULs giving High 4x16 portions of results
Rounded with 1/2 bit 15, storing to Implied register
*/
#define pmulhriw_m2ir(var, rs) mmx_m2ir(pmulhriw, var, rs)
#define pmulhriw_r2ir(rs1, rs2) mmx_r2ir(pmulhriw, rs1, rs2)
#define pmulhriw(vars, vard) mmx_m2m(pmulhriw, vars, vard)
/* 4x16 Parallel Muls (and ACcumulate) giving High 4x16 portions
of results Rounded with 1/2 bit 15, accumulating with Implied register
*/
#define pmachriw_m2ir(var, rs) mmx_m2ir(pmachriw, var, rs)
#define pmachriw_r2ir(rs1, rs2) mmx_r2ir(pmachriw, rs1, rs2)
#define pmachriw(vars, vard) mmx_m2m(pmachriw, vars, vard)
/* 8x8u Parallel AVErage
*/
#define paveb_m2r(var, reg) mmx_m2r(paveb, var, reg)
#define paveb_r2r(regs, regd) mmx_r2r(paveb, regs, regd)
#define paveb(vars, vard) mmx_m2m(paveb, vars, vard)
/* 8x8u Parallel DISTance and accumulate with
unsigned saturation to Implied register
*/
#define pdistib_m2ir(var, rs) mmx_m2ir(pdistib, var, rs)
#define pdistib(vars, vard) mmx_m2m(pdistib, vars, vard)
/* 8x8 Parallel conditional MoVe
if implied register field is Zero
*/
#define pmvzb_m2ir(var, rs) mmx_m2ir(pmvzb, var, rs)
/* 8x8 Parallel conditional MoVe
if implied register field is Not Zero
*/
#define pmvnzb_m2ir(var, rs) mmx_m2ir(pmvnzb, var, rs)
/* 8x8 Parallel conditional MoVe
if implied register field is Less than Zero
*/
#define pmvlzb_m2ir(var, rs) mmx_m2ir(pmvlzb, var, rs)
/* 8x8 Parallel conditional MoVe
if implied register field is Greater than or Equal to Zero
*/
#define pmvgezb_m2ir(var, rs) mmx_m2ir(pmvgezb, var, rs)
/* Fast Empty MMx State
(used to clean-up when going from mmx to float use
of the registers that are shared by both; note that
there is no float-to-xmmx operation needed, because
only the float tag word info is corruptible)
*/
#ifdef XMMX_TRACE
#define femms() \
{ \
fprintf(stderr, "femms()\n"); \
__asm__ __volatile__ ("femms"); \
}
#else
#define femms() __asm__ __volatile__ ("femms")
#endif
#endif
|
