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 /*********************************************************************/
51 #define MB __asm__ __volatile__ ("mfence");
52 #define WMB __asm__ __volatile__ ("sfence");
62 static void __inline blas_lock(volatile BLASULONG *address){
67 while (*address) {YIELDING;};
71 : "=r"(ret), "=m"(*address)
72 : "0"(1), "m"(*address)
78 static __inline BLASULONG rpcc(void){
81 __asm__ __volatile__ ("rdtsc" : "=a" (a), "=d" (d));
83 return ((BLASULONG)a + ((BLASULONG)d << 32));
88 static __inline BLASULONG getstackaddr(void){
91 __asm__ __volatile__ ("movq %%rsp, %0"
92 : "=r"(addr) : : "memory");
97 static __inline void cpuid(int op, int *eax, int *ebx, int *ecx, int *edx){
99 __asm__ __volatile__("cpuid"
111 static inline int WhereAmI(void){
112 int eax, ebx, ecx, edx;
115 cpuid(1, &eax, &ebx, &ecx, &edx);
116 apicid = BITMASK(ebx, 24, 0xff);
122 #ifdef CORE_BARCELONA
123 #define IFLUSH gotoblas_iflush()
124 #define IFLUSH_HALF gotoblas_iflush_half()
127 #ifdef ENABLE_SSE_EXCEPTION
129 #define IDEBUG_START \
131 unsigned int fp_sse_mode, new_fp_mode; \
132 __asm__ __volatile__ ("stmxcsr %0" : "=m" (fp_sse_mode) : ); \
133 new_fp_mode = fp_sse_mode & ~0xd00; \
134 __asm__ __volatile__ ("ldmxcsr %0" : : "m" (new_fp_mode) );
137 __asm__ __volatile__ ("ldmxcsr %0" : : "m" (fp_sse_mode) ); \
143 #define GET_IMAGE(res) __asm__ __volatile__("fstpt %0" : "=m"(res) : : "memory")
144 #elif defined(DOUBLE)
145 #define GET_IMAGE(res) __asm__ __volatile__("movsd %%xmm1, %0" : "=m"(res) : : "memory")
147 #define GET_IMAGE(res) __asm__ __volatile__("movss %%xmm1, %0" : "=m"(res) : : "memory")
150 #define GET_IMAGE_CANCEL
154 static __inline blasint blas_quickdivide(blasint x, blasint y){
158 extern unsigned int blas_quick_divide_table[];
160 static __inline int blas_quickdivide(unsigned int x, unsigned int y){
164 if (y <= 1) return x;
166 y = blas_quick_divide_table[y];
168 __asm__ __volatile__ ("mull %0" :"=d" (result) :"a"(x), "0" (y));
178 #define PAGESIZE ( 4 << 10)
180 #define HUGE_PAGESIZE ( 2 << 20)
182 #define BUFFER_SIZE (32 << 20)
186 #ifdef F_INTERFACE_G77
187 #define RETURN_BY_STACK
191 #ifdef F_INTERFACE_G95
192 #define RETURN_BY_PACKED
195 #ifdef F_INTERFACE_GFORT
198 #define RETURN_BY_REGS
200 #define RETURN_BY_STACK
203 #define RETURN_BY_PACKED
207 #ifdef F_INTERFACE_INTEL
208 #define RETURN_BY_STACK
211 #ifdef F_INTERFACE_FUJITSU
212 #define RETURN_BY_STACK
215 #ifdef F_INTERFACE_PGI
216 #define RETURN_BY_STACK
219 #ifdef F_INTERFACE_PATHSCALE
220 #define RETURN_BY_PACKED
223 #ifdef F_INTERFACE_SUN
224 #define RETURN_BY_PACKED
229 #if defined(PILEDRIVER) || defined(BULLDOZER) || defined(STEAMROLLER) || defined(EXCAVATOR)
230 //Enable some optimazation for barcelona.
231 #define BARCELONA_OPTIMIZATION
234 #if defined(HAVE_3DNOW)
236 #elif defined(HAVE_MMX)
244 #define BRANCH .byte 0x3e
245 #define NOBRANCH .byte 0x2e
246 #define PADDING .byte 0x66
264 #define LOCAL_BUFFER_SIZE QLOCAL_BUFFER_SIZE
266 #define LOCAL_BUFFER_SIZE DLOCAL_BUFFER_SIZE
268 #define LOCAL_BUFFER_SIZE SLOCAL_BUFFER_SIZE
272 #define LOCAL_BUFFER_SIZE XLOCAL_BUFFER_SIZE
274 #define LOCAL_BUFFER_SIZE ZLOCAL_BUFFER_SIZE
276 #define LOCAL_BUFFER_SIZE CLOCAL_BUFFER_SIZE
280 #if defined(OS_WINDOWS)
281 #if LOCAL_BUFFER_SIZE > 16384
282 #define STACK_TOUCHING \
283 movl $0, 4096 * 4(%rsp);\
284 movl $0, 4096 * 3(%rsp);\
285 movl $0, 4096 * 2(%rsp);\
286 movl $0, 4096 * 1(%rsp);
287 #elif LOCAL_BUFFER_SIZE > 12288
288 #define STACK_TOUCHING \
289 movl $0, 4096 * 3(%rsp);\
290 movl $0, 4096 * 2(%rsp);\
291 movl $0, 4096 * 1(%rsp);
292 #elif LOCAL_BUFFER_SIZE > 8192
293 #define STACK_TOUCHING \
294 movl $0, 4096 * 2(%rsp);\
295 movl $0, 4096 * 1(%rsp);
296 #elif LOCAL_BUFFER_SIZE > 4096
297 #define STACK_TOUCHING \
298 movl $0, 4096 * 1(%rsp);
300 #define STACK_TOUCHING
303 #define STACK_TOUCHING
307 #define movapd movaps
309 #define movlpd movlps
310 #define movhpd movhps
314 #define REALNAME ASMNAME
316 #define REALNAME ASMFNAME
320 #define PROLOGUE .text;.align 5; .globl REALNAME; REALNAME:
321 #define EPILOGUE .subsections_via_symbols
326 #define SAVEREGISTERS \
328 movups %xmm6, 0(%rsp);\
329 movups %xmm7, 16(%rsp);\
330 movups %xmm8, 32(%rsp);\
331 movups %xmm9, 48(%rsp);\
332 movups %xmm10, 64(%rsp);\
333 movups %xmm11, 80(%rsp);\
334 movups %xmm12, 96(%rsp);\
335 movups %xmm13, 112(%rsp);\
336 movups %xmm14, 128(%rsp);\
337 movups %xmm15, 144(%rsp)
339 #define RESTOREREGISTERS \
340 movups 0(%rsp), %xmm6;\
341 movups 16(%rsp), %xmm7;\
342 movups 32(%rsp), %xmm8;\
343 movups 48(%rsp), %xmm9;\
344 movups 64(%rsp), %xmm10;\
345 movups 80(%rsp), %xmm11;\
346 movups 96(%rsp), %xmm12;\
347 movups 112(%rsp), %xmm13;\
348 movups 128(%rsp), %xmm14;\
349 movups 144(%rsp), %xmm15;\
352 #define SAVEREGISTERS
353 #define RESTOREREGISTERS
356 #if defined(OS_WINDOWS) && !defined(C_PGI)
361 .def REALNAME;.scl 2;.type 32;.endef; \
366 #define EPILOGUE .end REALNAME
369 #if defined(OS_LINUX) || defined(OS_FREEBSD) || defined(OS_NETBSD) || defined(__ELF__) || defined(C_PGI)
374 .type REALNAME, @function; \
378 #define PROFCODE call *mcount@GOTPCREL(%rip)
384 .size REALNAME, .-REALNAME; \
385 .section .note.GNU-stack,"",@progbits
396 #elif defined(DOUBLE)
405 #define CMPEQPD cmpeqpd
406 #define COMISD comisd
414 #define MOVUPD movupd
425 #define CMPEQPD cmpeqps
426 #define COMISD comiss
434 #define MOVUPD movups
441 #define ALIGN_2 .align 2
442 #define ALIGN_3 .align 3
443 #define ALIGN_4 .align 4
444 #define ALIGN_5 .align 5
449 #define ALIGN_2 .align 4
453 #define ALIGN_3 .align 8
457 #define ALIGN_4 .align 16
461 #define ALIGN_5 .align 32
465 #define ALIGN_6 .align 64
469 // Because Clang didn't support ffreep, we directly use the opcode.
470 // Please check out http://www.sandpile.org/x86/opc_fpu.htm
472 #define ffreep .byte 0xdf, 0xc0 #