aarch64: vp9itxfm: Use w3 instead of x3 for the int eob parameter
authorMartin Storsjö <martin@martin.st>
Mon, 9 Jan 2017 22:15:10 +0000 (00:15 +0200)
committerMichael Niedermayer <michael@niedermayer.cc>
Sat, 14 Jan 2017 20:13:16 +0000 (21:13 +0100)
The clobbering tests in checkasm are only invoked when testing
correctness, so this bug didn't show up when benchmarking the
dc-only version.

This is cherrypicked from libav commit
4d960a11855f4212eb3a4e470ce890db7f01df29.

Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
libavcodec/aarch64/vp9itxfm_neon.S

index 7ce3116..3535c7b 100644 (file)
@@ -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