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;};
61 : "=r"(ret), "=m"(*address)
62 : "0"(1), "m"(*address)
69 static __inline unsigned long long rpcc(void){
72 __asm__ __volatile__ ("rdtsc" : "=a" (a), "=d" (d));
74 return ((unsigned long long)a + ((unsigned long long)d << 32));
77 static __inline unsigned long getstackaddr(void){
80 __asm__ __volatile__ ("mov %%esp, %0"
81 : "=r"(addr) : : "memory");
87 static __inline long double sqrt_long(long double val) {
90 __asm__ __volatile__ ("fldt %1\n"
92 "fstpt %0\n" : "=m" (result) : "m"(val));
96 #define SQRT(a) sqrt_long(a)
98 /* This is due to gcc's bug */
99 void cpuid(int op, int *eax, int *ebx, int *ecx, int *edx);
103 static inline int WhereAmI(void){
104 int eax, ebx, ecx, edx;
107 cpuid(1, &eax, &ebx, &ecx, &edx);
108 apicid = BITMASK(ebx, 24, 0xff);
113 #ifdef ENABLE_SSE_EXCEPTION
115 #define IDEBUG_START \
117 unsigned int fp_sse_mode, new_fp_mode; \
118 __asm__ __volatile__ ("stmxcsr %0" : "=m" (fp_sse_mode) : ); \
119 new_fp_mode = fp_sse_mode & ~0xd00; \
120 __asm__ __volatile__ ("ldmxcsr %0" : : "m" (new_fp_mode) );
123 __asm__ __volatile__ ("ldmxcsr %0" : : "m" (fp_sse_mode) ); \
129 #define GET_IMAGE(res) __asm__ __volatile__("fstpt %0" : "=m"(res) : : "memory")
130 #elif defined(DOUBLE)
131 #define GET_IMAGE(res) __asm__ __volatile__("fstpl %0" : "=m"(res) : : "memory")
133 #define GET_IMAGE(res) __asm__ __volatile__("fstps %0" : "=m"(res) : : "memory");
136 #define GET_IMAGE_CANCEL __asm__ __volatile__ ("ffree %st")
139 extern unsigned int blas_quick_divide_table[];
141 static __inline int blas_quickdivide(unsigned int x, unsigned int y){
145 if (y <= 1) return x;
147 y = blas_quick_divide_table[y];
149 __asm__ __volatile__ ("mull %0" :"=d" (result) :"a"(x), "0" (y));
158 #define PAGESIZE ( 4 << 10)
160 #define HUGE_PAGESIZE ( 4 << 20)
162 #define BUFFER_SIZE (16 << 20)
166 #if defined(DOUBLE) || defined(XDOUBLE)
168 #define MMXSTORE movq
171 #define MMXSTORE movd
174 #if defined(PILEDRIVER) || defined(BULLDOZER)
175 //Enable some optimazation for barcelona.
176 #define BARCELONA_OPTIMIZATION
179 #if defined(HAVE_3DNOW)
181 #elif defined(HAVE_MMX)
189 #if defined(CORE2) || defined(PENTIUM4)
190 #define movapd movaps
193 #define BRANCH .byte 0x3e
194 #define NOBRANCH .byte 0x2e
195 #define PADDING .byte 0x66;
200 #define LOCAL_BUFFER_SIZE QLOCAL_BUFFER_SIZE
202 #define LOCAL_BUFFER_SIZE DLOCAL_BUFFER_SIZE
204 #define LOCAL_BUFFER_SIZE SLOCAL_BUFFER_SIZE
208 #define LOCAL_BUFFER_SIZE XLOCAL_BUFFER_SIZE
210 #define LOCAL_BUFFER_SIZE ZLOCAL_BUFFER_SIZE
212 #define LOCAL_BUFFER_SIZE CLOCAL_BUFFER_SIZE
216 #if defined(OS_WINDOWS)
217 #if LOCAL_BUFFER_SIZE > 16384
218 #define STACK_TOUCHING \
219 movl $0, 4096 * 4(%esp);\
220 movl $0, 4096 * 3(%esp);\
221 movl $0, 4096 * 2(%esp);\
222 movl $0, 4096 * 1(%esp);
223 #elif LOCAL_BUFFER_SIZE > 12288
224 #define STACK_TOUCHING \
225 movl $0, 4096 * 3(%esp);\
226 movl $0, 4096 * 2(%esp);\
227 movl $0, 4096 * 1(%esp);
228 #elif LOCAL_BUFFER_SIZE > 8192
229 #define STACK_TOUCHING \
230 movl $0, 4096 * 2(%esp);\
231 movl $0, 4096 * 1(%esp);
232 #elif LOCAL_BUFFER_SIZE > 4096
233 #define STACK_TOUCHING \
234 movl $0, 4096 * 1(%esp);
236 #define STACK_TOUCHING
239 #define STACK_TOUCHING
243 #define REALNAME ASMNAME
245 #define REALNAME ASMFNAME
248 #if defined(F_INTERFACE_PATHSCALE) || defined(F_INTERFACE_OPEN64)
249 #define RETURN_BY_STRUCT
250 #elif defined(F_INTERFACE_GFORT) || defined(F_INTERFACE_G95)
251 #define RETURN_BY_COMPLEX
253 #define RETURN_BY_STACK
257 #define PROLOGUE .text;.align 5; .globl REALNAME; REALNAME:
258 #define EPILOGUE .subsections_via_symbols
262 #if defined(OS_WINNT) || defined(OS_CYGWIN_NT) || defined(OS_INTERIX)
263 #define SAVEREGISTERS \
265 movups %xmm6, 0(%esp);\
266 movups %xmm7, 16(%esp)
268 #define RESTOREREGISTERS \
269 movups 0(%esp), %xmm6;\
270 movups 16(%esp), %xmm7;\
273 #define SAVEREGISTERS
274 #define RESTOREREGISTERS
277 #if defined(OS_WINNT) || defined(OS_CYGWIN_NT) || defined(OS_INTERIX)
282 .def REALNAME;.scl 2;.type 32;.endef; \
287 #define EPILOGUE .end REALNAME
290 #if defined(OS_LINUX) || defined(OS_FREEBSD) || defined(OS_NETBSD) || defined(__ELF__)
295 .type REALNAME, @function; \
299 #define PROFCODE call mcount
305 .size REALNAME, .-REALNAME; \
306 .section .note.GNU-stack,"",%progbits
318 #elif defined(DOUBLE)
342 #define ALIGN_2 .align 2
343 #define ALIGN_3 .align 3
344 #define ALIGN_4 .align 4
345 #define ALIGN_5 .align 5
350 #define ALIGN_2 .align 4
354 #define ALIGN_3 .align 8
358 #define ALIGN_4 .align 16
362 #define ALIGN_5 .align 32
366 #define ALIGN_6 .align 64
369 // Because Clang didn't support ffreep, we directly use the opcode.
370 // Please check out http://www.sandpile.org/x86/opc_fpu.htm
372 #define ffreep .byte 0xdf, 0xc0 #