1 /*********************************************************************/
2 /* Copyright 2009, 2010 The University of Texas at Austin. */
3 /* All rights reserved. */
5 /* Redistribution and use in source and binary forms, with or */
6 /* without modification, are permitted provided that the following */
7 /* conditions are met: */
9 /* 1. Redistributions of source code must retain the above */
10 /* copyright notice, this list of conditions and the following */
13 /* 2. Redistributions in binary form must reproduce the above */
14 /* copyright notice, this list of conditions and the following */
15 /* disclaimer in the documentation and/or other materials */
16 /* provided with the distribution. */
18 /* THIS SOFTWARE IS PROVIDED BY THE UNIVERSITY OF TEXAS AT */
19 /* AUSTIN ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, */
20 /* INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF */
21 /* MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE */
22 /* DISCLAIMED. IN NO EVENT SHALL THE UNIVERSITY OF TEXAS AT */
23 /* AUSTIN OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, */
24 /* INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES */
25 /* (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE */
26 /* GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR */
27 /* BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF */
28 /* LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT */
29 /* (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT */
30 /* OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE */
31 /* POSSIBILITY OF SUCH DAMAGE. */
33 /* The views and conclusions contained in the software and */
34 /* documentation are those of the authors and should not be */
35 /* interpreted as representing official policies, either expressed */
36 /* or implied, of The University of Texas at Austin. */
37 /*********************************************************************/
55 #define MB __asm__ __volatile__ ("mfence");
56 #define WMB __asm__ __volatile__ ("sfence");
64 #define MB do { __asm__ __volatile__("": : :"memory"); } while (0)
65 #define WMB do { __asm__ __volatile__("": : :"memory"); } while (0)
68 #define MB do {} while (0)
69 #define WMB do {} while (0)
73 static void __inline blas_lock(volatile BLASULONG *address){
83 while (*address) {YIELDING;}
88 : "=r"(ret), "=m"(*address)
89 : "0"(1), "m"(*address)
92 ret=InterlockedExchange64((volatile LONG64 *)(address), 1);
97 #define BLAS_LOCK_DEFINED
99 static __inline BLASULONG rpcc(void){
105 __asm__ __volatile__ ("rdtsc" : "=a" (a), "=d" (d));
107 return ((BLASULONG)a + ((BLASULONG)d << 32));
115 static __inline BLASULONG getstackaddr(void){
118 __asm__ __volatile__ ("movq %%rsp, %0"
119 : "=r"(addr) : : "memory");
125 static __inline void cpuid(int op, int *eax, int *ebx, int *ecx, int *edx){
129 __cpuid(cpuinfo, op);
135 __asm__ __volatile__("mov $0, %%ecx;"
145 static __inline void cpuid_count(int op, int count, int *eax, int *ebx, int *ecx, int *edx)
148 int cpuInfo[4] = {-1};
149 __cpuidex(cpuInfo, op, count);
155 #if defined(__i386__) && defined(__PIC__)
159 "xchgl %%ebx, %%edi;"
160 : "=a" (*eax), "=D" (*ebx), "=c" (*ecx), "=d" (*edx) : "0" (op), "2" (count) : "cc");
163 ("cpuid": "=a" (*eax), "=b" (*ebx), "=c" (*ecx), "=d" (*edx) : "0" (op), "2" (count) : "cc");
172 static __inline int WhereAmI(void){
173 int eax, ebx, ecx, edx;
176 cpuid(1, &eax, &ebx, &ecx, &edx);
177 apicid = BITMASK(ebx, 24, 0xff);
183 #ifdef CORE_BARCELONA
184 #define IFLUSH gotoblas_iflush()
185 #define IFLUSH_HALF gotoblas_iflush_half()
188 #ifdef ENABLE_SSE_EXCEPTION
190 #define IDEBUG_START \
192 unsigned int fp_sse_mode, new_fp_mode; \
193 __asm__ __volatile__ ("stmxcsr %0" : "=m" (fp_sse_mode) : ); \
194 new_fp_mode = fp_sse_mode & ~0xd00; \
195 __asm__ __volatile__ ("ldmxcsr %0" : : "m" (new_fp_mode) );
198 __asm__ __volatile__ ("ldmxcsr %0" : : "m" (fp_sse_mode) ); \
204 #define GET_IMAGE(res) __asm__ __volatile__("fstpt %0" : "=m"(res) : : "memory")
205 #elif defined(DOUBLE)
206 #define GET_IMAGE(res) __asm__ __volatile__("movsd %%xmm1, %0" : "=m"(res) : : "memory")
208 #define GET_IMAGE(res) __asm__ __volatile__("movss %%xmm1, %0" : "=m"(res) : : "memory")
211 #define GET_IMAGE_CANCEL
214 #if defined(USE64BITINT)
215 static __inline blasint blas_quickdivide(blasint x, blasint y){
218 #elif defined (C_MSVC)
219 static __inline BLASLONG blas_quickdivide(BLASLONG x, BLASLONG y){
223 extern unsigned int blas_quick_divide_table[];
225 static __inline unsigned int blas_quickdivide(unsigned int x, unsigned int y){
227 volatile unsigned int result;
229 if (y <= 1) return x;
231 #if (MAX_CPU_NUMBER > 64)
238 y = blas_quick_divide_table[y];
240 __asm__ __volatile__ ("mull %0" :"=d" (result), "+a"(x) : "0" (y));
249 #define PAGESIZE ( 4 << 10)
251 #define HUGE_PAGESIZE ( 2 << 20)
254 #define BUFFER_SIZE (32 << 22)
256 #define BUFFER_SIZE (32 << BUFFERSIZE)
261 #ifdef F_INTERFACE_G77
262 #define RETURN_BY_STACK
266 #ifdef F_INTERFACE_G95
267 #define RETURN_BY_PACKED
270 #ifdef F_INTERFACE_GFORT
273 #define RETURN_BY_REGS
275 #define RETURN_BY_STACK
278 #define RETURN_BY_PACKED
282 #ifdef F_INTERFACE_INTEL
283 #define RETURN_BY_STACK
286 #ifdef F_INTERFACE_FUJITSU
287 #define RETURN_BY_STACK
290 #ifdef F_INTERFACE_FLANG
291 #define RETURN_BY_STACK
294 #ifdef F_INTERFACE_PGI
295 #define RETURN_BY_STACK
298 #ifdef F_INTERFACE_PATHSCALE
299 #define RETURN_BY_PACKED
302 #ifdef F_INTERFACE_SUN
303 #define RETURN_BY_PACKED
308 #if defined(PILEDRIVER) || defined(BULLDOZER) || defined(STEAMROLLER) || defined(EXCAVATOR)
309 //Enable some optimization for barcelona.
310 #define BARCELONA_OPTIMIZATION
313 #if defined(HAVE_3DNOW)
315 #elif defined(HAVE_MMX)
323 #define BRANCH .byte 0x3e
324 #define NOBRANCH .byte 0x2e
325 #define PADDING .byte 0x66
343 #define LOCAL_BUFFER_SIZE QLOCAL_BUFFER_SIZE
345 #define LOCAL_BUFFER_SIZE DLOCAL_BUFFER_SIZE
347 #define LOCAL_BUFFER_SIZE SLOCAL_BUFFER_SIZE
351 #define LOCAL_BUFFER_SIZE XLOCAL_BUFFER_SIZE
353 #define LOCAL_BUFFER_SIZE ZLOCAL_BUFFER_SIZE
355 #define LOCAL_BUFFER_SIZE CLOCAL_BUFFER_SIZE
359 #if defined(OS_WINDOWS)
360 #if LOCAL_BUFFER_SIZE > 16384
361 #define STACK_TOUCHING \
362 movl $0, 4096 * 4(%rsp);\
363 movl $0, 4096 * 3(%rsp);\
364 movl $0, 4096 * 2(%rsp);\
365 movl $0, 4096 * 1(%rsp);
366 #elif LOCAL_BUFFER_SIZE > 12288
367 #define STACK_TOUCHING \
368 movl $0, 4096 * 3(%rsp);\
369 movl $0, 4096 * 2(%rsp);\
370 movl $0, 4096 * 1(%rsp);
371 #elif LOCAL_BUFFER_SIZE > 8192
372 #define STACK_TOUCHING \
373 movl $0, 4096 * 2(%rsp);\
374 movl $0, 4096 * 1(%rsp);
375 #elif LOCAL_BUFFER_SIZE > 4096
376 #define STACK_TOUCHING \
377 movl $0, 4096 * 1(%rsp);
379 #define STACK_TOUCHING
382 #define STACK_TOUCHING
386 #define movapd movaps
388 #define movlpd movlps
389 #define movhpd movhps
393 #define REALNAME ASMNAME
395 #define REALNAME ASMFNAME
399 #define PROLOGUE .text;.align 5; .globl REALNAME; REALNAME:
400 #define EPILOGUE .subsections_via_symbols
405 #define SAVEREGISTERS \
407 movups %xmm6, 0(%rsp);\
408 movups %xmm7, 16(%rsp);\
409 movups %xmm8, 32(%rsp);\
410 movups %xmm9, 48(%rsp);\
411 movups %xmm10, 64(%rsp);\
412 movups %xmm11, 80(%rsp);\
413 movups %xmm12, 96(%rsp);\
414 movups %xmm13, 112(%rsp);\
415 movups %xmm14, 128(%rsp);\
416 movups %xmm15, 144(%rsp)
418 #define RESTOREREGISTERS \
419 movups 0(%rsp), %xmm6;\
420 movups 16(%rsp), %xmm7;\
421 movups 32(%rsp), %xmm8;\
422 movups 48(%rsp), %xmm9;\
423 movups 64(%rsp), %xmm10;\
424 movups 80(%rsp), %xmm11;\
425 movups 96(%rsp), %xmm12;\
426 movups 112(%rsp), %xmm13;\
427 movups 128(%rsp), %xmm14;\
428 movups 144(%rsp), %xmm15;\
431 #define SAVEREGISTERS
432 #define RESTOREREGISTERS
435 #if defined(OS_WINDOWS) && !defined(C_PGI)
440 .def REALNAME;.scl 2;.type 32;.endef; \
445 #define EPILOGUE .end
448 #if defined(OS_LINUX) || defined(OS_FREEBSD) || defined(OS_NETBSD) || defined(OS_OPENBSD) || defined(OS_DRAGONFLY) || defined(__ELF__) || defined(C_PGI)
453 .type REALNAME, @function; \
458 #define PROFCODE call *mcount@GOTPCREL(%rip)
464 .size REALNAME, .-REALNAME; \
465 .section .note.GNU-stack,"",@progbits
476 #elif defined(DOUBLE)
485 #define CMPEQPD cmpeqpd
486 #define COMISD comisd
494 #define MOVUPD movupd
505 #define CMPEQPD cmpeqps
506 #define COMISD comiss
514 #define MOVUPD movups
521 #define ALIGN_2 .align 2
522 #define ALIGN_3 .align 3
523 #define ALIGN_4 .align 4
524 #define ALIGN_5 .align 5
529 #define ALIGN_2 .align 4
533 #define ALIGN_3 .align 8
537 #define ALIGN_4 .align 16
541 #define ALIGN_5 .align 32
545 #define ALIGN_6 .align 64
549 // Because Clang didn't support ffreep, we directly use the opcode.
550 // Please check out http://www.sandpile.org/x86/opc_fpu.htm
552 #define ffreep .byte 0xdf, 0xc0 #