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 /*********************************************************************/
47 EXTERN gotoblas_t gotoblas_KATMAI;
48 EXTERN gotoblas_t gotoblas_COPPERMINE;
49 EXTERN gotoblas_t gotoblas_NORTHWOOD;
50 EXTERN gotoblas_t gotoblas_BANIAS;
51 EXTERN gotoblas_t gotoblas_ATHLON;
53 extern gotoblas_t gotoblas_PRESCOTT;
54 extern gotoblas_t gotoblas_ATOM;
55 extern gotoblas_t gotoblas_NANO;
56 extern gotoblas_t gotoblas_CORE2;
57 extern gotoblas_t gotoblas_PENRYN;
58 extern gotoblas_t gotoblas_DUNNINGTON;
59 extern gotoblas_t gotoblas_NEHALEM;
60 extern gotoblas_t gotoblas_OPTERON;
61 extern gotoblas_t gotoblas_OPTERON_SSE3;
62 extern gotoblas_t gotoblas_BARCELONA;
63 extern gotoblas_t gotoblas_BOBCAT;
65 extern gotoblas_t gotoblas_SANDYBRIDGE;
66 extern gotoblas_t gotoblas_BULLDOZER;
67 extern gotoblas_t gotoblas_PILEDRIVER;
69 //Use NEHALEM kernels for sandy bridge
70 #define gotoblas_SANDYBRIDGE gotoblas_NEHALEM
71 #define gotoblas_BULLDOZER gotoblas_BARCELONA
72 #define gotoblas_PILEDRIVER gotoblas_BARCELONA
74 //Use sandy bridge kernels for haswell.
75 #define gotoblas_HASWELL gotoblas_SANDYBRIDGE
77 #define VENDOR_INTEL 1
79 #define VENDOR_CENTAUR 3
80 #define VENDOR_UNKNOWN 99
82 #define BITMASK(a, b, c) ((((a) >> (b)) & (c)))
85 static inline void xgetbv(int op, int * eax, int * edx){
86 //Use binary code for xgetbv
88 (".byte 0x0f, 0x01, 0xd0": "=a" (*eax), "=d" (*edx) : "c" (op) : "cc");
94 int eax, ebx, ecx, edx;
97 cpuid(1, &eax, &ebx, &ecx, &edx);
98 if ((ecx & (1 << 28)) != 0 && (ecx & (1 << 27)) != 0 && (ecx & (1 << 26)) != 0){
99 xgetbv(0, &eax, &edx);
101 ret=1; //OS support AVX
110 static int get_vendor(void){
111 int eax, ebx, ecx, edx;
114 cpuid(0, &eax, &ebx, &ecx, &edx);
116 *(int *)(&vendor[0]) = ebx;
117 *(int *)(&vendor[4]) = edx;
118 *(int *)(&vendor[8]) = ecx;
119 vendor[12] = (char)0;
121 if (!strcmp(vendor, "GenuineIntel")) return VENDOR_INTEL;
122 if (!strcmp(vendor, "AuthenticAMD")) return VENDOR_AMD;
123 if (!strcmp(vendor, "CentaurHauls")) return VENDOR_CENTAUR;
125 if ((eax == 0) || ((eax & 0x500) != 0)) return VENDOR_INTEL;
127 return VENDOR_UNKNOWN;
130 static gotoblas_t *get_coretype(void){
132 int eax, ebx, ecx, edx;
133 int family, exfamily, model, vendor, exmodel;
135 cpuid(1, &eax, &ebx, &ecx, &edx);
137 family = BITMASK(eax, 8, 0x0f);
138 exfamily = BITMASK(eax, 20, 0xff);
139 model = BITMASK(eax, 4, 0x0f);
140 exmodel = BITMASK(eax, 16, 0x0f);
142 vendor = get_vendor();
144 if (vendor == VENDOR_INTEL){
149 if (model <= 0x7) return &gotoblas_KATMAI;
150 if ((model == 0x8) || (model == 0xa) || (model == 0xb)) return &gotoblas_COPPERMINE;
151 if ((model == 0x9) || (model == 0xd)) return &gotoblas_BANIAS;
152 if (model == 14) return &gotoblas_BANIAS;
153 if (model == 15) return &gotoblas_CORE2;
157 if (model == 6) return &gotoblas_CORE2;
158 if (model == 7) return &gotoblas_PENRYN;
159 if (model == 13) return &gotoblas_DUNNINGTON;
160 if ((model == 10) || (model == 11) || (model == 14) || (model == 15)) return &gotoblas_NEHALEM;
161 if (model == 12) return &gotoblas_ATOM;
165 //Intel Core (Clarkdale) / Core (Arrandale)
166 // Pentium (Clarkdale) / Pentium Mobile (Arrandale)
167 // Xeon (Clarkdale), 32nm
168 if (model == 5) return &gotoblas_NEHALEM;
170 //Intel Xeon Processor 5600 (Westmere-EP)
171 //Xeon Processor E7 (Westmere-EX)
173 if (model == 12 || model == 14 || model == 15) return &gotoblas_NEHALEM;
175 //Intel Core i5-2000 /i7-2000 (Sandy Bridge)
176 //Intel Core i7-3000 / Xeon E5
177 if (model == 10 || model == 13) {
179 return &gotoblas_SANDYBRIDGE;
181 fprintf(stderr, "OpenBLAS : Your OS does not support AVX instructions. OpenBLAS is using Nehalem kernels as a fallback, which may give poorer performance.\n");
182 return &gotoblas_NEHALEM; //OS doesn't support AVX. Use old kernels.
187 //Intel Sandy Bridge 22nm (Ivy Bridge?)
190 return &gotoblas_SANDYBRIDGE;
192 fprintf(stderr, "OpenBLAS : Your OS does not support AVX instructions. OpenBLAS is using Nehalem kernels as a fallback, which may give poorer performance.\n");
193 return &gotoblas_NEHALEM; //OS doesn't support AVX. Use old kernels.
199 return &gotoblas_HASWELL;
201 fprintf(stderr, "OpenBLAS : Your OS does not support AVX instructions. OpenBLAS is using Nehalem kernels as a fallback, which may give poorer performance.\n");
202 return &gotoblas_NEHALEM; //OS doesn't support AVX. Use old kernels.
210 return &gotoblas_HASWELL;
212 fprintf(stderr, "OpenBLAS : Your OS does not support AVX instructions. OpenBLAS is using Nehalem kernels as a fallback, which may give poorer performance.\n");
213 return &gotoblas_NEHALEM; //OS doesn't support AVX. Use old kernels.
219 if (model <= 0x2) return &gotoblas_NORTHWOOD;
220 return &gotoblas_PRESCOTT;
224 if (vendor == VENDOR_AMD){
225 if (family <= 0xe) return &gotoblas_ATHLON;
227 if ((exfamily == 0) || (exfamily == 2)) {
228 if (ecx & (1 << 0)) return &gotoblas_OPTERON_SSE3;
229 else return &gotoblas_OPTERON;
230 } else if (exfamily == 5) {
231 return &gotoblas_BOBCAT;
232 } else if (exfamily == 6) {
234 //AMD Bulldozer Opteron 6200 / Opteron 4200 / AMD FX-Series
236 return &gotoblas_BULLDOZER;
238 fprintf(stderr, "OpenBLAS : Your OS does not support AVX instructions. OpenBLAS is using Barcelona kernels as a fallback, which may give poorer performance.\n");
239 return &gotoblas_BARCELONA; //OS doesn't support AVX. Use old kernels.
241 }else if(model == 2){
242 //AMD Bulldozer Opteron 6300 / Opteron 4300 / Opteron 3300
244 return &gotoblas_PILEDRIVER;
246 fprintf(stderr, "OpenBLAS : Your OS does not support AVX instructions. OpenBLAS is using Barcelona kernels as a fallback, which may give poorer performance.\n");
247 return &gotoblas_BARCELONA; //OS doesn't support AVX. Use old kernels.
251 return &gotoblas_BARCELONA;
256 if (vendor == VENDOR_CENTAUR) {
259 return &gotoblas_NANO;
267 static char *corename[] = {
290 char *gotoblas_corename(void) {
292 if (gotoblas == &gotoblas_KATMAI) return corename[ 1];
293 if (gotoblas == &gotoblas_COPPERMINE) return corename[ 2];
294 if (gotoblas == &gotoblas_NORTHWOOD) return corename[ 3];
295 if (gotoblas == &gotoblas_PRESCOTT) return corename[ 4];
296 if (gotoblas == &gotoblas_BANIAS) return corename[ 5];
297 if (gotoblas == &gotoblas_ATOM) return corename[ 6];
298 if (gotoblas == &gotoblas_CORE2) return corename[ 7];
299 if (gotoblas == &gotoblas_PENRYN) return corename[ 8];
300 if (gotoblas == &gotoblas_DUNNINGTON) return corename[ 9];
301 if (gotoblas == &gotoblas_NEHALEM) return corename[10];
302 if (gotoblas == &gotoblas_ATHLON) return corename[11];
303 if (gotoblas == &gotoblas_OPTERON_SSE3) return corename[12];
304 if (gotoblas == &gotoblas_OPTERON) return corename[13];
305 if (gotoblas == &gotoblas_BARCELONA) return corename[14];
306 if (gotoblas == &gotoblas_NANO) return corename[15];
307 if (gotoblas == &gotoblas_SANDYBRIDGE) return corename[16];
308 if (gotoblas == &gotoblas_BOBCAT) return corename[17];
309 if (gotoblas == &gotoblas_BULLDOZER) return corename[18];
310 if (gotoblas == &gotoblas_PILEDRIVER) return corename[19];
315 void gotoblas_dynamic_init(void) {
317 if (gotoblas) return;
319 gotoblas = get_coretype();
322 if (gotoblas == NULL) gotoblas = &gotoblas_KATMAI;
324 if (gotoblas == NULL) gotoblas = &gotoblas_PRESCOTT;
325 /* sanity check, if 64bit pointer we can't have a 32 bit cpu */
326 if (sizeof(void*) == 8) {
327 if (gotoblas == &gotoblas_KATMAI ||
328 gotoblas == &gotoblas_COPPERMINE ||
329 gotoblas == &gotoblas_NORTHWOOD ||
330 gotoblas == &gotoblas_BANIAS ||
331 gotoblas == &gotoblas_ATHLON)
332 gotoblas = &gotoblas_PRESCOTT;
336 if (gotoblas && gotoblas -> init) {
339 fprintf(stderr, "OpenBLAS : Architecture Initialization failed. No initialization function found.\n");
345 void gotoblas_dynamic_quit(void) {