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)
80 #define BLAS_LOCK_DEFINED
82 static __inline unsigned long long rpcc(void){
83 #if defined(_MSC_VER) && !defined(__clang__)
84 return __rdtsc(); // use MSVC intrinsic
88 __asm__ __volatile__ ("rdtsc" : "=a" (a), "=d" (d));
90 return ((unsigned long long)a + ((unsigned long long)d << 32));
95 static __inline unsigned long getstackaddr(void){
96 #if defined(_MSC_VER) && !defined(__clang__)
97 return (unsigned long)_ReturnAddress(); // use MSVC intrinsic
101 __asm__ __volatile__ ("mov %%esp, %0"
102 : "=r"(addr) : : "memory");
109 static __inline long double sqrt_long(long double val) {
110 #if defined(_MSC_VER) && !defined(__clang__)
111 return sqrt(val); // not sure if this will use fsqrt
115 __asm__ __volatile__ ("fldt %1\n"
117 "fstpt %0\n" : "=m" (result) : "m"(val));
122 #define SQRT(a) sqrt_long(a)
124 /* This is due to gcc's bug */
125 void cpuid(int op, int *eax, int *ebx, int *ecx, int *edx);
129 static __inline int WhereAmI(void){
130 int eax, ebx, ecx, edx;
133 cpuid(1, &eax, &ebx, &ecx, &edx);
134 apicid = BITMASK(ebx, 24, 0xff);
139 #ifdef ENABLE_SSE_EXCEPTION
141 #define IDEBUG_START \
143 unsigned int fp_sse_mode, new_fp_mode; \
144 __asm__ __volatile__ ("stmxcsr %0" : "=m" (fp_sse_mode) : ); \
145 new_fp_mode = fp_sse_mode & ~0xd00; \
146 __asm__ __volatile__ ("ldmxcsr %0" : : "m" (new_fp_mode) );
149 __asm__ __volatile__ ("ldmxcsr %0" : : "m" (fp_sse_mode) ); \
155 #define GET_IMAGE(res) __asm__ __volatile__("fstpt %0" : "=m"(res) : : "memory")
156 #elif defined(DOUBLE)
157 #define GET_IMAGE(res) __asm__ __volatile__("fstpl %0" : "=m"(res) : : "memory")
159 #define GET_IMAGE(res) __asm__ __volatile__("fstps %0" : "=m"(res) : : "memory");
162 #define GET_IMAGE_CANCEL __asm__ __volatile__ ("ffree %st")
165 extern unsigned int blas_quick_divide_table[];
167 static __inline int blas_quickdivide(unsigned int x, unsigned int y){
171 if (y <= 1) return x;
173 y = blas_quick_divide_table[y];
175 #if defined(_MSC_VER) && !defined(__clang__)
179 __asm__ __volatile__ ("mull %0" :"=d" (result) :"a"(x), "0" (y));
189 #define PAGESIZE ( 4 << 10)
191 #define HUGE_PAGESIZE ( 4 << 20)
193 #define BUFFER_SIZE (16 << 20)
197 #if defined(DOUBLE) || defined(XDOUBLE)
199 #define MMXSTORE movq
202 #define MMXSTORE movd
205 #if defined(PILEDRIVER) || defined(BULLDOZER) || defined(STEAMROLLER) || defined(EXCAVATOR)
206 //Enable some optimazation for barcelona.
207 #define BARCELONA_OPTIMIZATION
210 #if defined(HAVE_3DNOW)
212 #elif defined(HAVE_MMX)
220 #if defined(CORE2) || defined(PENTIUM4)
221 #define movapd movaps
224 #define BRANCH .byte 0x3e
225 #define NOBRANCH .byte 0x2e
226 #define PADDING .byte 0x66;
231 #define LOCAL_BUFFER_SIZE QLOCAL_BUFFER_SIZE
233 #define LOCAL_BUFFER_SIZE DLOCAL_BUFFER_SIZE
235 #define LOCAL_BUFFER_SIZE SLOCAL_BUFFER_SIZE
239 #define LOCAL_BUFFER_SIZE XLOCAL_BUFFER_SIZE
241 #define LOCAL_BUFFER_SIZE ZLOCAL_BUFFER_SIZE
243 #define LOCAL_BUFFER_SIZE CLOCAL_BUFFER_SIZE
247 #if defined(OS_WINDOWS)
248 #if LOCAL_BUFFER_SIZE > 16384
249 #define STACK_TOUCHING \
250 movl $0, 4096 * 4(%esp);\
251 movl $0, 4096 * 3(%esp);\
252 movl $0, 4096 * 2(%esp);\
253 movl $0, 4096 * 1(%esp);
254 #elif LOCAL_BUFFER_SIZE > 12288
255 #define STACK_TOUCHING \
256 movl $0, 4096 * 3(%esp);\
257 movl $0, 4096 * 2(%esp);\
258 movl $0, 4096 * 1(%esp);
259 #elif LOCAL_BUFFER_SIZE > 8192
260 #define STACK_TOUCHING \
261 movl $0, 4096 * 2(%esp);\
262 movl $0, 4096 * 1(%esp);
263 #elif LOCAL_BUFFER_SIZE > 4096
264 #define STACK_TOUCHING \
265 movl $0, 4096 * 1(%esp);
267 #define STACK_TOUCHING
270 #define STACK_TOUCHING
274 #define REALNAME ASMNAME
276 #define REALNAME ASMFNAME
279 #if defined(F_INTERFACE_PATHSCALE) || defined(F_INTERFACE_OPEN64)
280 #define RETURN_BY_STRUCT
281 #elif defined(F_INTERFACE_GFORT) || defined(F_INTERFACE_G95)
282 #define RETURN_BY_COMPLEX
284 #define RETURN_BY_STACK
288 #define PROLOGUE .text;.align 5; .globl REALNAME; REALNAME:
289 #define EPILOGUE .subsections_via_symbols
293 #if defined(OS_WINNT) || defined(OS_CYGWIN_NT) || defined(OS_INTERIX)
294 #define SAVEREGISTERS \
296 movups %xmm6, 0(%esp);\
297 movups %xmm7, 16(%esp)
299 #define RESTOREREGISTERS \
300 movups 0(%esp), %xmm6;\
301 movups 16(%esp), %xmm7;\
304 #define SAVEREGISTERS
305 #define RESTOREREGISTERS
308 #if defined(OS_WINNT) || defined(OS_CYGWIN_NT) || defined(OS_INTERIX)
313 .def REALNAME;.scl 2;.type 32;.endef; \
319 #define EPILOGUE .end
321 #define EPILOGUE .end REALNAME
325 #if defined(OS_LINUX) || defined(OS_FREEBSD) || defined(OS_NETBSD) || defined(__ELF__)
330 .type REALNAME, @function; \
334 #define PROFCODE call mcount
340 .size REALNAME, .-REALNAME; \
341 .section .note.GNU-stack,"",@progbits
353 #elif defined(DOUBLE)
377 #define ALIGN_2 .align 2
378 #define ALIGN_3 .align 3
379 #define ALIGN_4 .align 4
380 #define ALIGN_5 .align 5
385 #define ALIGN_2 .align 4
389 #define ALIGN_3 .align 8
393 #define ALIGN_4 .align 16
397 #define ALIGN_5 .align 32
401 #define ALIGN_6 .align 64
404 // Because Clang didn't support ffreep, we directly use the opcode.
405 // Please check out http://www.sandpile.org/x86/opc_fpu.htm
407 #define ffreep .byte 0xdf, 0xc0 #