Ref #79 Added GEMM_MULTITHREAD_THRESHOLD flag to use single thread in gemm function with small matrices.

Makefile.rule

@@ -88,6 +88,11 @@ VERSION = 0.1alpha2.5
 # If you need to synchronize FP CSR between threads (for x86/x86_64 only).
 # CONSISTENT_FPCSR = 1
 
+# If any gemm arguement m, n or k is less or equal this threshold, gemm will be execute 
+# with single thread. You can use this flag to avoid the overhead of multi-threading 
+# in small matrix sizes. The default value is 4.
+# GEMM_MULTITHREAD_THRESHOLD = 4
+
 # If you need santy check by comparing reference BLAS. It'll be very
 # slow (Not implemented yet).
 # SANITY_CHECK = 1

Makefile.system

@@ -40,6 +40,11 @@ ifdef INTERFACE64
 GETARCH_FLAGS	+= -DUSE64BITINT
 endif
 
+ifndef GEMM_MULTITHREAD_THRESHOLD
+GEMM_MULTITHREAD_THRESHOLD=4
+endif
+GETARCH_FLAGS	+= -DGEMM_MULTITHREAD_THRESHOLD=$(GEMM_MULTITHREAD_THRESHOLD) 
+
 # This operation is expensive, so execution should be once.
 ifndef GOTOBLAS_MAKEFILE
 export GOTOBLAS_MAKEFILE = 1

getarch_2nd.c

@@ -34,6 +34,7 @@ int main(int argc, char **argv) {
 #ifdef USE64BITINT
 	printf("#define USE64BITINT\n");
 #endif
+	printf("#define GEMM_MULTITHREAD_THRESHOLD\t%ld\n", GEMM_MULTITHREAD_THRESHOLD);
   }
 
   return 0;

interface/gemm.c

@@ -397,8 +397,13 @@ void CNAME(enum CBLAS_ORDER order, enum CBLAS_TRANSPOSE TransA, enum CBLAS_TRANS
   mode |= (transb << BLAS_TRANSB_SHIFT);
 
   args.common = NULL;
-  args.nthreads = num_cpu_avail(3);
 
+  if(args.m <= GEMM_MULTITHREAD_THRESHOLD || args.n <= GEMM_MULTITHREAD_THRESHOLD 
+     || args.k <=GEMM_MULTITHREAD_THRESHOLD){
+    args.nthreads = 1;
+  }else{
+    args.nthreads = num_cpu_avail(3);
+  }
  if (args.nthreads == 1) {
 #endif