arm: dsputil: Add a bunch of missing #includes

[ffmpeg] / libavcodec / arm / simple_idct_neon.S

diff --git a/libavcodec/arm/simple_idct_neon.S b/libavcodec/arm/simple_idct_neon.S
index cbed9eefe45ede8507c2a9be2375cb7d94d60ef9..a1cde8d80a13a369ed72c5459006d748f9ffaa6a 100644 (file)
--- a/libavcodec/arm/simple_idct_neon.S
+++ b/libavcodec/arm/simple_idct_neon.S
@@ -23,7 +23,7 @@
  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  */
 
-#include "asm.S"
+#include "libavutil/arm/asm.S"
 
 #define W1  22725  //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
 #define W2  21407  //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
@@ -159,8 +159,8 @@ function idct_col4_neon
         vmull.s16       q15, d30, w4 /* q15 = W4*(col[0]+(1<<COL_SHIFT-1)/W4)*/
         vld1.64         {d8}, [r2,:64], ip /* d5 = col[3] */
 
-        ldrd            r4,  [r2]
-        ldrd            r6,  [r2, #16]
+        ldrd            r4,  r5,  [r2]
+        ldrd            r6,  r7,  [r2, #16]
         orrs            r4,  r4,  r5
 
         idct_col4_top
@@ -176,7 +176,7 @@ function idct_col4_neon
         vadd.i32        q14, q14, q7
 
 1:      orrs            r6,  r6,  r7
-        ldrd            r4,  [r2, #16]
+        ldrd            r4,  r5,  [r2, #16]
         it              eq
         addeq           r2,  r2,  #16
         beq             2f
@@ -188,7 +188,7 @@ function idct_col4_neon
         vmlal.s16       q6,  d5,  w3    /* q6  += W3 * col[5] */
 
 2:      orrs            r4,  r4,  r5
-        ldrd            r4,  [r2, #16]
+        ldrd            r4,  r5,  [r2, #16]
         it              eq
         addeq           r2,  r2,  #16
         beq             3f
@@ -243,10 +243,9 @@ function idct_col4_st8_neon
         bx              lr
 endfunc
 
-        .section .rodata
-        .align 4
-idct_coeff_neon:
+const   idct_coeff_neon, align=4
         .short W1, W2, W3, W4, W5, W6, W7, W4c
+endconst
 
         .macro idct_start data
         push            {r4-r7, lr}
@@ -262,7 +261,7 @@ idct_coeff_neon:
         pop             {r4-r7, pc}
         .endm
 
-/* void ff_simple_idct_put_neon(uint8_t *dst, int line_size, DCTELEM *data); */
+/* void ff_simple_idct_put_neon(uint8_t *dst, int line_size, int16_t *data); */
 function ff_simple_idct_put_neon, export=1
         idct_start      r2
 
@@ -317,7 +316,7 @@ function idct_col4_add8_neon
         bx              lr
 endfunc
 
-/* void ff_simple_idct_add_neon(uint8_t *dst, int line_size, DCTELEM *data); */
+/* void ff_simple_idct_add_neon(uint8_t *dst, int line_size, int16_t *data); */
 function ff_simple_idct_add_neon, export=1
         idct_start      r2
 
@@ -356,7 +355,7 @@ function idct_col4_st16_neon
         bx              lr
 endfunc
 
-/* void ff_simple_idct_neon(DCTELEM *data); */
+/* void ff_simple_idct_neon(int16_t *data); */
 function ff_simple_idct_neon, export=1
         idct_start      r0