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 /*********************************************************************/
52 static void __inline blas_lock(volatile BLASULONG *address){
57 while (*address) {YIELDING;};
59 #if defined(_MSC_VER) && !defined(__clang__)
60 // use intrinsic instead of inline assembly
61 ret = _InterlockedExchange(address, 1);
72 : "=r"(ret), "=m"(*address)
73 : "0"(1), "m"(*address)
81 static __inline unsigned long long rpcc(void){
82 #if defined(_MSC_VER) && !defined(__clang__)
83 return __rdtsc(); // use MSVC intrinsic
87 __asm__ __volatile__ ("rdtsc" : "=a" (a), "=d" (d));
89 return ((unsigned long long)a + ((unsigned long long)d << 32));
93 static __inline unsigned long getstackaddr(void){
94 #if defined(_MSC_VER) && !defined(__clang__)
95 return (unsigned long)_ReturnAddress(); // use MSVC intrinsic
99 __asm__ __volatile__ ("mov %%esp, %0"
100 : "=r"(addr) : : "memory");
107 static __inline long double sqrt_long(long double val) {
108 #if defined(_MSC_VER) && !defined(__clang__)
109 return sqrt(val); // not sure if this will use fsqrt
113 __asm__ __volatile__ ("fldt %1\n"
115 "fstpt %0\n" : "=m" (result) : "m"(val));
120 #define SQRT(a) sqrt_long(a)
122 /* This is due to gcc's bug */
123 void cpuid(int op, int *eax, int *ebx, int *ecx, int *edx);
127 static __inline int WhereAmI(void){
128 int eax, ebx, ecx, edx;
131 cpuid(1, &eax, &ebx, &ecx, &edx);
132 apicid = BITMASK(ebx, 24, 0xff);
137 #ifdef ENABLE_SSE_EXCEPTION
139 #define IDEBUG_START \
141 unsigned int fp_sse_mode, new_fp_mode; \
142 __asm__ __volatile__ ("stmxcsr %0" : "=m" (fp_sse_mode) : ); \
143 new_fp_mode = fp_sse_mode & ~0xd00; \
144 __asm__ __volatile__ ("ldmxcsr %0" : : "m" (new_fp_mode) );
147 __asm__ __volatile__ ("ldmxcsr %0" : : "m" (fp_sse_mode) ); \
153 #define GET_IMAGE(res) __asm__ __volatile__("fstpt %0" : "=m"(res) : : "memory")
154 #elif defined(DOUBLE)
155 #define GET_IMAGE(res) __asm__ __volatile__("fstpl %0" : "=m"(res) : : "memory")
157 #define GET_IMAGE(res) __asm__ __volatile__("fstps %0" : "=m"(res) : : "memory");
160 #define GET_IMAGE_CANCEL __asm__ __volatile__ ("ffree %st")
163 extern unsigned int blas_quick_divide_table[];
165 static __inline int blas_quickdivide(unsigned int x, unsigned int y){
169 if (y <= 1) return x;
171 y = blas_quick_divide_table[y];
173 #if defined(_MSC_VER) && !defined(__clang__)
177 __asm__ __volatile__ ("mull %0" :"=d" (result) :"a"(x), "0" (y));
187 #define PAGESIZE ( 4 << 10)
189 #define HUGE_PAGESIZE ( 4 << 20)
191 #define BUFFER_SIZE (16 << 20)
195 #if defined(DOUBLE) || defined(XDOUBLE)
197 #define MMXSTORE movq
200 #define MMXSTORE movd
203 #if defined(PILEDRIVER) || defined(BULLDOZER) || defined(STEAMROLLER) || defined(EXCAVATOR)
204 //Enable some optimazation for barcelona.
205 #define BARCELONA_OPTIMIZATION
208 #if defined(HAVE_3DNOW)
210 #elif defined(HAVE_MMX)
218 #if defined(CORE2) || defined(PENTIUM4)
219 #define movapd movaps
222 #define BRANCH .byte 0x3e
223 #define NOBRANCH .byte 0x2e
224 #define PADDING .byte 0x66;
229 #define LOCAL_BUFFER_SIZE QLOCAL_BUFFER_SIZE
231 #define LOCAL_BUFFER_SIZE DLOCAL_BUFFER_SIZE
233 #define LOCAL_BUFFER_SIZE SLOCAL_BUFFER_SIZE
237 #define LOCAL_BUFFER_SIZE XLOCAL_BUFFER_SIZE
239 #define LOCAL_BUFFER_SIZE ZLOCAL_BUFFER_SIZE
241 #define LOCAL_BUFFER_SIZE CLOCAL_BUFFER_SIZE
245 #if defined(OS_WINDOWS)
246 #if LOCAL_BUFFER_SIZE > 16384
247 #define STACK_TOUCHING \
248 movl $0, 4096 * 4(%esp);\
249 movl $0, 4096 * 3(%esp);\
250 movl $0, 4096 * 2(%esp);\
251 movl $0, 4096 * 1(%esp);
252 #elif LOCAL_BUFFER_SIZE > 12288
253 #define STACK_TOUCHING \
254 movl $0, 4096 * 3(%esp);\
255 movl $0, 4096 * 2(%esp);\
256 movl $0, 4096 * 1(%esp);
257 #elif LOCAL_BUFFER_SIZE > 8192
258 #define STACK_TOUCHING \
259 movl $0, 4096 * 2(%esp);\
260 movl $0, 4096 * 1(%esp);
261 #elif LOCAL_BUFFER_SIZE > 4096
262 #define STACK_TOUCHING \
263 movl $0, 4096 * 1(%esp);
265 #define STACK_TOUCHING
268 #define STACK_TOUCHING
272 #define REALNAME ASMNAME
274 #define REALNAME ASMFNAME
277 #if defined(F_INTERFACE_PATHSCALE) || defined(F_INTERFACE_OPEN64)
278 #define RETURN_BY_STRUCT
279 #elif defined(F_INTERFACE_GFORT) || defined(F_INTERFACE_G95)
280 #define RETURN_BY_COMPLEX
282 #define RETURN_BY_STACK
286 #define PROLOGUE .text;.align 5; .globl REALNAME; REALNAME:
287 #define EPILOGUE .subsections_via_symbols
291 #if defined(OS_WINNT) || defined(OS_CYGWIN_NT) || defined(OS_INTERIX)
292 #define SAVEREGISTERS \
294 movups %xmm6, 0(%esp);\
295 movups %xmm7, 16(%esp)
297 #define RESTOREREGISTERS \
298 movups 0(%esp), %xmm6;\
299 movups 16(%esp), %xmm7;\
302 #define SAVEREGISTERS
303 #define RESTOREREGISTERS
306 #if defined(OS_WINNT) || defined(OS_CYGWIN_NT) || defined(OS_INTERIX)
311 .def REALNAME;.scl 2;.type 32;.endef; \
317 #define EPILOGUE .end
319 #define EPILOGUE .end REALNAME
323 #if defined(OS_LINUX) || defined(OS_FREEBSD) || defined(OS_NETBSD) || defined(__ELF__)
328 .type REALNAME, @function; \
332 #define PROFCODE call mcount
338 .size REALNAME, .-REALNAME; \
339 .section .note.GNU-stack,"",@progbits
351 #elif defined(DOUBLE)
375 #define ALIGN_2 .align 2
376 #define ALIGN_3 .align 3
377 #define ALIGN_4 .align 4
378 #define ALIGN_5 .align 5
383 #define ALIGN_2 .align 4
387 #define ALIGN_3 .align 8
391 #define ALIGN_4 .align 16
395 #define ALIGN_5 .align 32
399 #define ALIGN_6 .align 64
402 // Because Clang didn't support ffreep, we directly use the opcode.
403 // Please check out http://www.sandpile.org/x86/opc_fpu.htm
405 #define ffreep .byte 0xdf, 0xc0 #