[bsp] faster startup for cortex-a

If the application defines dozens of global variables,
the speed of clearing the bss segment will be slower.

Because icache can be enabled before the mmu enabled.
Therefore, in order to speed up the process of clearing the BSS segment,
enable icache needs to be put ahead.

libcpu/arm/cortex-a/start_gcc.S

@@ -122,6 +122,13 @@ secondary_loop:
     b secondary_loop
 
 normal_setup:
+
+    /* enable I cache + branch prediction */
+    mrc p15, 0, r0, c1, c0, 0
+    orr     r0, r0, #(1<<12)
+    orr     r0, r0, #(1<<11)
+    mcr p15, 0, r0, c1, c0, 0
+
     /* setup stack */
     bl      stack_setup
 
@@ -142,10 +149,6 @@ bss_loop:
     mcr p15, 0, r1, c1, c0, 1 //enable smp
 #endif
 
-    /* enable branch prediction */
-    mrc p15, 0, r0, c1, c0, 0
-    orr     r0, r0, #(1<<11)
-    mcr p15, 0, r0, c1, c0, 0
 
     /* initialize the mmu table and enable mmu */
     ldr r0, =platform_mem_desc