MdePkg/BaseMemoryLibOptDxe ARM AARCH64: fix thinko in SetMem##

The new InternalMemSetMem##() implementations for ARM and AARCH64 in
BaseMemoryLibOptDxe fail to take into account that the 'length' argument
is not in bytes, but in number of items to be copied. So multiply by the
item size before proceeding.

Reported-by: Laszlo Ersek <lersek@redhat.com>
Contributed-under: TianoCore Contribution Agreement 1.0
Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Reviewed-by: Liming Gao <liming.gao@intel.com>
Tested-by: Laszlo Ersek <lersek@redhat.com>

MdePkg/Library/BaseMemoryLibOptDxe/Arm/SetMem.asm

@@ -21,21 +21,33 @@
     AREA    SetMem, CODE, READONLY, CODEALIGN, ALIGN=5
     THUMB
 
-InternalMemZeroMem
-    movs    r2, #0
+InternalMemSetMem16
+    uxth    r2, r2
+    lsl     r1, r1, #1
+    orr     r2, r2, r2, lsl #16
+    b       B0
 
+InternalMemSetMem32
+    lsl     r1, r1, #2
+    b       B0
+
+InternalMemSetMem64
+    lsl     r1, r1, #3
+    b       B1
+
+    ALIGN   32
 InternalMemSetMem
     uxtb    r2, r2
     orr     r2, r2, r2, lsl #8
+    orr     r2, r2, r2, lsl #16
+    b       B0
 
-InternalMemSetMem16
-    uxth    r2, r2
-    orr     r2, r2, r2, lsr #16
-
-InternalMemSetMem32
+InternalMemZeroMem
+    movs    r2, #0
+B0
     mov     r3, r2
 
-InternalMemSetMem64
+B1
     push    {r4, lr}
     cmp     r1, #16                 ; fewer than 16 bytes of input?
     add     r1, r1, r0              ; r1 := dst + length