aarch64: vp9itxfm: Use w3 instead of x3 for the int eob parameter

The clobbering tests in checkasm are only invoked when testing
correctness, so this bug didn't show up when benchmarking the
dc-only version.

Signed-off-by: Martin Storsjö <martin@martin.st>

libavcodec/aarch64/vp9itxfm_neon.S

@@ -204,7 +204,7 @@ function ff_vp9_\txfm1\()_\txfm2\()_4x4_add_neon, export=1
 
         movi            v31.8h, #0
 .ifc \txfm1\()_\txfm2,idct_idct
-        cmp             x3,  #1
+        cmp             w3,  #1
         b.ne            1f
         // DC-only for idct/idct
         ld1r            {v2.4h},  [x2]
@@ -344,7 +344,7 @@ function ff_vp9_\txfm1\()_\txfm2\()_8x8_add_neon, export=1
         movi            v5.16b, #0
 
 .ifc \txfm1\()_\txfm2,idct_idct
-        cmp             x3,  #1
+        cmp             w3,  #1
         b.ne            1f
         // DC-only for idct/idct
         ld1r            {v2.4h},  [x2]
@@ -722,7 +722,7 @@ itxfm16_1d_funcs iadst
 .macro itxfm_func16x16 txfm1, txfm2
 function ff_vp9_\txfm1\()_\txfm2\()_16x16_add_neon, export=1
 .ifc \txfm1\()_\txfm2,idct_idct
-        cmp             x3,  #1
+        cmp             w3,  #1
         b.eq            idct16x16_dc_add_neon
 .endif
         mov             x15, x30
@@ -1074,7 +1074,7 @@ function idct32_1d_8x32_pass2_neon
 endfunc
 
 function ff_vp9_idct_idct_32x32_add_neon, export=1
-        cmp             x3,  #1
+        cmp             w3,  #1
         b.eq            idct32x32_dc_add_neon
 
         movrel          x10, idct_coeffs