Refs #478,#482, Enable stack alloc for s/dgemv_t.(revert 9798491)

interface/gemv.c

@@ -212,21 +212,17 @@ void CNAME(enum CBLAS_ORDER order,
   // make it volatile because some gemv implementation (ex: dgemv_n.S)
   // do not restore all register
   volatile int stack_alloc_size = 0;
-  if (trans == 0) {
-    //for gemv_n, try to allocate on stack
-    //for gemv_t, use malloc
+  //for gemv_n and gemv_t, try to allocate on stack
+  stack_alloc_size = m + n;
+  if(stack_alloc_size < 128)
+    //dgemv_n.S require a 128 bytes buffer
+    stack_alloc_size = 128;
 
-    stack_alloc_size = m + n;
-    if(stack_alloc_size < 128)
-      //dgemv_n.S require a 128 bytes buffer
-      stack_alloc_size = 128;
-
-    if(stack_alloc_size > MAX_STACK_ALLOC / sizeof(FLOAT))
-      stack_alloc_size = 0;
-  }
+  if(stack_alloc_size > MAX_STACK_ALLOC / sizeof(FLOAT))
+    stack_alloc_size = 0;
 
   FLOAT stack_buffer[stack_alloc_size];
-  buffer = stack_alloc_size ? stack_buffer : (FLOAT *)blas_memory_alloc_nolock(1);
+  buffer = stack_alloc_size ? stack_buffer : (FLOAT *)blas_memory_alloc(1);
   //  printf("stack_alloc_size=%d\n", stack_alloc_size);
 #else
   //Original OpenBLAS/GotoBLAS codes.
@@ -262,7 +258,7 @@ void CNAME(enum CBLAS_ORDER order,
 
 #ifdef MAX_STACK_ALLOC
   if(!stack_alloc_size){
-    blas_memory_free_nolock(buffer);
+    blas_memory_free(buffer);
   }
 #else
     blas_memory_free(buffer);