Merge pull request #46 from xianyi/develop

rebase

common_arm.h

@@ -121,7 +121,7 @@ REALNAME:
 #endif
 #define HUGE_PAGESIZE   ( 4 << 20)
 
-#define BUFFER_SIZE     (16 << 20)
+#define BUFFER_SIZE     (32 << 20)
 
 
 #define BASE_ADDRESS (START_ADDRESS - BUFFER_SIZE * MAX_CPU_NUMBER)

common_arm64.h

@@ -141,12 +141,17 @@ REALNAME:
 #endif
 #define HUGE_PAGESIZE   ( 4 << 20)
 
+#ifndef BUFFERSIZE
 #if defined(CORTEXA57)
 #define BUFFER_SIZE     (20 << 20)
+#elif defined(TSV110) || defined(EMAG8180)
+#define BUFFER_SIZE     (32 << 20)
 #else
 #define BUFFER_SIZE     (16 << 20)
 #endif
-
+#else
+#define BUFFER_SIZE	(32 << BUFFERSIZE)
+#endif
 
 #define BASE_ADDRESS (START_ADDRESS - BUFFER_SIZE * MAX_CPU_NUMBER)
 

common_x86_64.h

@@ -226,7 +226,13 @@ static __inline int blas_quickdivide(unsigned int x, unsigned int y){
 #define HUGE_PAGESIZE	( 2 << 20)
 
 #ifndef BUFFERSIZE
+#if defined(SKYLAKEX) 
+#define BUFFER_SIZE	(32 << 21)
+#elif defined(HASWELL) || defined(ZEN)
+#define BUFFER_SIZE	(32 << 22)
+#else
 #define BUFFER_SIZE	(32 << 20)
+#endif
 #else
 #define BUFFER_SIZE	(32 << BUFFERSIZE)
 #endif

common_zarch.h

@@ -123,11 +123,7 @@ REALNAME:
 #endif
 #define HUGE_PAGESIZE   ( 4 << 20)
 
-#if defined(CORTEXA57)
-#define BUFFER_SIZE     (20 << 20)
-#else
-#define BUFFER_SIZE     (16 << 20)
-#endif
+#define BUFFER_SIZE     (32 << 22)
 
 
 #define BASE_ADDRESS (START_ADDRESS - BUFFER_SIZE * MAX_CPU_NUMBER)

driver/others/memory.c

@@ -87,6 +87,30 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #endif
 #endif
 
+/* Memory buffer must fit two matrix subblocks of maximal size */
+#define XSTR(x) STR(x)
+#define STR(x) #x
+#if BUFFER_SIZE < (SGEMM_DEFAULT_P * SGEMM_DEFAULT_Q * 4 * 2) || \
+    BUFFER_SIZE < (SGEMM_DEFAULT_P * SGEMM_DEFAULT_R * 4 * 2) || \
+    BUFFER_SIZE < (SGEMM_DEFAULT_R * SGEMM_DEFAULT_Q * 4 * 2)
+#warning BUFFER_SIZE is too small for P, Q, and R of SGEMM - large calculations may crash !
+#endif
+#if BUFFER_SIZE < (DGEMM_DEFAULT_P * DGEMM_DEFAULT_Q * 8 * 2) || \
+    BUFFER_SIZE < (DGEMM_DEFAULT_P * DGEMM_DEFAULT_R * 8 * 2) || \
+    BUFFER_SIZE < (DGEMM_DEFAULT_R * DGEMM_DEFAULT_Q * 8 * 2)
+#warning BUFFER_SIZE is too small for P, Q, and R of DGEMM - large calculations may crash !
+#endif
+#if BUFFER_SIZE < (CGEMM_DEFAULT_P * CGEMM_DEFAULT_Q * 8 * 2) || \
+    BUFFER_SIZE < (CGEMM_DEFAULT_P * CGEMM_DEFAULT_R * 8 * 2) || \
+    BUFFER_SIZE < (CGEMM_DEFAULT_R * CGEMM_DEFAULT_Q * 8 * 2)
+#warning BUFFER_SIZE is too small for P, Q, and R of CGEMM - large calculations may crash !
+#endif
+#if BUFFER_SIZE < (ZGEMM_DEFAULT_P * ZGEMM_DEFAULT_Q * 16 * 2) || \
+    BUFFER_SIZE < (ZGEMM_DEFAULT_P * ZGEMM_DEFAULT_R * 16 * 2) || \
+    BUFFER_SIZE < (ZGEMM_DEFAULT_R * ZGEMM_DEFAULT_Q * 16 * 2)
+#warning BUFFER_SIZE is too small for P, Q, and R of ZGEMM - large calculations may crash !
+#endif
+
 #if defined(COMPILE_TLS)
 
 #include <errno.h>
@@ -2740,7 +2764,7 @@ void *blas_memory_alloc(int procpos){
 #ifdef DEBUG
   printf("  Position -> %d\n", position);
 #endif
-WMB;
+
   memory[position].used = 1;
 #if (defined(SMP) || defined(USE_LOCKING)) && !defined(USE_OPENMP)
   UNLOCK_COMMAND(&alloc_lock);

param.h

@@ -2229,15 +2229,26 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #define ZGEMM_DEFAULT_UNROLL_M 8
 #define ZGEMM_DEFAULT_UNROLL_N 2
 
-#define SGEMM_DEFAULT_P  1280
-#define DGEMM_DEFAULT_P  640
-#define CGEMM_DEFAULT_P  640
-#define ZGEMM_DEFAULT_P  320
-
-#define SGEMM_DEFAULT_Q  640
-#define DGEMM_DEFAULT_Q  720
-#define CGEMM_DEFAULT_Q  640
-#define ZGEMM_DEFAULT_Q  640
+#define SGEMM_DEFAULT_P  1280UL
+#define DGEMM_DEFAULT_P  640UL
+#define CGEMM_DEFAULT_P  640UL
+#define ZGEMM_DEFAULT_P  320UL
+
+#define SGEMM_DEFAULT_Q  640UL
+#define DGEMM_DEFAULT_Q  720UL
+#define CGEMM_DEFAULT_Q  640UL
+#define ZGEMM_DEFAULT_Q  640UL
+
+#if 0
+#define SGEMM_DEFAULT_R SGEMM_DEFAULT_P
+#define DGEMM_DEFAULT_R DGEMM_DEFAULT_P
+#define CGEMM_DEFAULT_R CGEMM_DEFAULT_P
+#define ZGEMM_DEFAULT_R ZGEMM_DEFAULT_P
+#endif
+#define SGEMM_DEFAULT_R 4096
+#define DGEMM_DEFAULT_R 4096
+#define CGEMM_DEFAULT_R 4096
+#define ZGEMM_DEFAULT_R 512
 
 #define SYMV_P	 8