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)
68 #define BLAS_LOCK_DEFINED
70 static __inline unsigned long long rpcc(void){
73 __asm__ __volatile__ ("rdtsc" : "=a" (a), "=d" (d));
75 return ((unsigned long long)a + ((unsigned long long)d << 32));
79 static __inline unsigned long getstackaddr(void){
82 __asm__ __volatile__ ("mov %%esp, %0"
83 : "=r"(addr) : : "memory");
89 static __inline long double sqrt_long(long double val) {
92 __asm__ __volatile__ ("fldt %1\n"
94 "fstpt %0\n" : "=m" (result) : "m"(val));
98 #define SQRT(a) sqrt_long(a)
100 /* This is due to gcc's bug */
101 void cpuid(int op, int *eax, int *ebx, int *ecx, int *edx);
105 static inline int WhereAmI(void){
106 int eax, ebx, ecx, edx;
109 cpuid(1, &eax, &ebx, &ecx, &edx);
110 apicid = BITMASK(ebx, 24, 0xff);
115 #ifdef ENABLE_SSE_EXCEPTION
117 #define IDEBUG_START \
119 unsigned int fp_sse_mode, new_fp_mode; \
120 __asm__ __volatile__ ("stmxcsr %0" : "=m" (fp_sse_mode) : ); \
121 new_fp_mode = fp_sse_mode & ~0xd00; \
122 __asm__ __volatile__ ("ldmxcsr %0" : : "m" (new_fp_mode) );
125 __asm__ __volatile__ ("ldmxcsr %0" : : "m" (fp_sse_mode) ); \
131 #define GET_IMAGE(res) __asm__ __volatile__("fstpt %0" : "=m"(res) : : "memory")
132 #elif defined(DOUBLE)
133 #define GET_IMAGE(res) __asm__ __volatile__("fstpl %0" : "=m"(res) : : "memory")
135 #define GET_IMAGE(res) __asm__ __volatile__("fstps %0" : "=m"(res) : : "memory");
138 #define GET_IMAGE_CANCEL __asm__ __volatile__ ("ffree %st")
141 extern unsigned int blas_quick_divide_table[];
143 static __inline int blas_quickdivide(unsigned int x, unsigned int y){
147 if (y <= 1) return x;
149 y = blas_quick_divide_table[y];
151 __asm__ __volatile__ ("mull %0" :"=d" (result) :"a"(x), "0" (y));
160 #define PAGESIZE ( 4 << 10)
162 #define HUGE_PAGESIZE ( 4 << 20)
164 #define BUFFER_SIZE (16 << 20)
168 #if defined(DOUBLE) || defined(XDOUBLE)
170 #define MMXSTORE movq
173 #define MMXSTORE movd
176 #if defined(PILEDRIVER) || defined(BULLDOZER) || defined(STEAMROLLER) || defined(EXCAVATOR)
177 //Enable some optimazation for barcelona.
178 #define BARCELONA_OPTIMIZATION
181 #if defined(HAVE_3DNOW)
183 #elif defined(HAVE_MMX)
191 #if defined(CORE2) || defined(PENTIUM4)
192 #define movapd movaps
195 #define BRANCH .byte 0x3e
196 #define NOBRANCH .byte 0x2e
197 #define PADDING .byte 0x66;
202 #define LOCAL_BUFFER_SIZE QLOCAL_BUFFER_SIZE
204 #define LOCAL_BUFFER_SIZE DLOCAL_BUFFER_SIZE
206 #define LOCAL_BUFFER_SIZE SLOCAL_BUFFER_SIZE
210 #define LOCAL_BUFFER_SIZE XLOCAL_BUFFER_SIZE
212 #define LOCAL_BUFFER_SIZE ZLOCAL_BUFFER_SIZE
214 #define LOCAL_BUFFER_SIZE CLOCAL_BUFFER_SIZE
218 #if defined(OS_WINDOWS)
219 #if LOCAL_BUFFER_SIZE > 16384
220 #define STACK_TOUCHING \
221 movl $0, 4096 * 4(%esp);\
222 movl $0, 4096 * 3(%esp);\
223 movl $0, 4096 * 2(%esp);\
224 movl $0, 4096 * 1(%esp);
225 #elif LOCAL_BUFFER_SIZE > 12288
226 #define STACK_TOUCHING \
227 movl $0, 4096 * 3(%esp);\
228 movl $0, 4096 * 2(%esp);\
229 movl $0, 4096 * 1(%esp);
230 #elif LOCAL_BUFFER_SIZE > 8192
231 #define STACK_TOUCHING \
232 movl $0, 4096 * 2(%esp);\
233 movl $0, 4096 * 1(%esp);
234 #elif LOCAL_BUFFER_SIZE > 4096
235 #define STACK_TOUCHING \
236 movl $0, 4096 * 1(%esp);
238 #define STACK_TOUCHING
241 #define STACK_TOUCHING
245 #define REALNAME ASMNAME
247 #define REALNAME ASMFNAME
250 #if defined(F_INTERFACE_PATHSCALE) || defined(F_INTERFACE_OPEN64)
251 #define RETURN_BY_STRUCT
252 #elif defined(F_INTERFACE_GFORT) || defined(F_INTERFACE_G95)
253 #define RETURN_BY_COMPLEX
255 #define RETURN_BY_STACK
259 #define PROLOGUE .text;.align 5; .globl REALNAME; REALNAME:
260 #define EPILOGUE .subsections_via_symbols
264 #if defined(OS_WINNT) || defined(OS_CYGWIN_NT) || defined(OS_INTERIX)
265 #define SAVEREGISTERS \
267 movups %xmm6, 0(%esp);\
268 movups %xmm7, 16(%esp)
270 #define RESTOREREGISTERS \
271 movups 0(%esp), %xmm6;\
272 movups 16(%esp), %xmm7;\
275 #define SAVEREGISTERS
276 #define RESTOREREGISTERS
279 #if defined(OS_WINNT) || defined(OS_CYGWIN_NT) || defined(OS_INTERIX)
284 .def REALNAME;.scl 2;.type 32;.endef; \
289 #define EPILOGUE .end REALNAME
292 #if defined(OS_LINUX) || defined(OS_FREEBSD) || defined(OS_NETBSD) || defined(__ELF__)
297 .type REALNAME, @function; \
301 #define PROFCODE call mcount
307 .size REALNAME, .-REALNAME; \
308 .section .note.GNU-stack,"",@progbits
320 #elif defined(DOUBLE)
344 #define ALIGN_2 .align 2
345 #define ALIGN_3 .align 3
346 #define ALIGN_4 .align 4
347 #define ALIGN_5 .align 5
352 #define ALIGN_2 .align 4
356 #define ALIGN_3 .align 8
360 #define ALIGN_4 .align 16
364 #define ALIGN_5 .align 32
368 #define ALIGN_6 .align 64
371 // Because Clang didn't support ffreep, we directly use the opcode.
372 // Please check out http://www.sandpile.org/x86/opc_fpu.htm
374 #define ffreep .byte 0xdf, 0xc0 #