Fix TEXTRELs in the ARM asm.

Besides imposing a performance penalty at startup in most
 configurations, these relocations break the dynamic linker for
 native Fennec, since it does not support them at all.

Change-Id: Id5dc768609354ebb4379966eb61a7313e6fd18de

vp8/common/arm/neon/vp8_subpixelvariance16x16_neon.asm

@@ -9,6 +9,11 @@
 ;
 
 
+bilinear_taps_coeff
+    DCD     128, 0, 112, 16, 96, 32, 80, 48, 64, 64, 48, 80, 32, 96, 16, 112
+
+;-----------------
+
     EXPORT  |vp8_sub_pixel_variance16x16_neon_func|
     ARM
     REQUIRE8
@@ -27,7 +32,7 @@
 |vp8_sub_pixel_variance16x16_neon_func| PROC
     push            {r4-r6, lr}
 
-    ldr             r12, _BilinearTaps_coeff_
+    adr             r12, bilinear_taps_coeff
     ldr             r4, [sp, #16]           ;load *dst_ptr from stack
     ldr             r5, [sp, #20]           ;load dst_pixels_per_line from stack
     ldr             r6, [sp, #24]           ;load *sse from stack
@@ -415,11 +420,4 @@ sub_pixel_variance16x16_neon_loop
 
     ENDP
 
-;-----------------
-
-_BilinearTaps_coeff_
-    DCD     bilinear_taps_coeff
-bilinear_taps_coeff
-    DCD     128, 0, 112, 16, 96, 32, 80, 48, 64, 64, 48, 80, 32, 96, 16, 112
-
     END

vp8/common/arm/neon/vp8_subpixelvariance8x8_neon.asm

@@ -27,7 +27,7 @@
 |vp8_sub_pixel_variance8x8_neon| PROC
     push            {r4-r5, lr}
 
-    ldr             r12, _BilinearTaps_coeff_
+    adr             r12, bilinear_taps_coeff
     ldr             r4, [sp, #12]           ;load *dst_ptr from stack
     ldr             r5, [sp, #16]           ;load dst_pixels_per_line from stack
     ldr             lr, [sp, #20]           ;load *sse from stack
@@ -216,8 +216,6 @@ sub_pixel_variance8x8_neon_loop
 
 ;-----------------
 
-_BilinearTaps_coeff_
-    DCD     bilinear_taps_coeff
 bilinear_taps_coeff
     DCD     128, 0, 112, 16, 96, 32, 80, 48, 64, 64, 48, 80, 32, 96, 16, 112
 

vp8/encoder/arm/neon/fastquantizeb_neon.asm

@@ -98,7 +98,7 @@
     vmul.s16        q2, q6, q4          ; x * Dequant
     vmul.s16        q3, q7, q5
 
-    ldr             r0, _inv_zig_zag_   ; load ptr of inverse zigzag table
+    adr             r0, inv_zig_zag     ; load ptr of inverse zigzag table
 
     vceq.s16        q8, q8              ; set q8 to all 1
 
@@ -181,7 +181,7 @@
     vadd.s16        q12, q14            ; x + Round
     vadd.s16        q13, q15
 
-    ldr             r0, _inv_zig_zag_   ; load ptr of inverse zigzag table
+    adr             r0, inv_zig_zag     ; load ptr of inverse zigzag table
 
     vqdmulh.s16     q12, q8             ; y = ((Round+abs(z)) * Quant) >> 16
     vqdmulh.s16     q13, q9
@@ -247,9 +247,6 @@ zero_output
     ENDP
 
 ; default inverse zigzag table is defined in vp8/common/entropy.c
-_inv_zig_zag_
-    DCD inv_zig_zag
-
     ALIGN 16    ; enable use of @128 bit aligned loads
 inv_zig_zag
     DCW 0x0001, 0x0002, 0x0006, 0x0007