Modify calling of NEON code in sub-pixel search

In vp8_find_best_sub_pixel_step_iteratively(), many times xoffset
and yoffset are specific values - (4,0) (0,4) and (4,4). Modified
code to call simplified NEON version at these specific offsets to
help with the performance.

Change-Id: Iaf896a0f7aae4697bd36a49e182525dd1ef1ab4d

diff --git a/vp8/encoder/arm/arm_csystemdependent.c b/vp8/encoder/arm/arm_csystemdependent.c
index a1f110260d81179f84bdee39448352fa20103069..4f68a9576cad98102b492bb3e18eab517893cef1 100644 (file)
--- a/vp8/encoder/arm/arm_csystemdependent.c
+++ b/vp8/encoder/arm/arm_csystemdependent.c
@@ -18,6 +18,37 @@ extern void (*vp8_yv12_copy_partial_frame_ptr)(YV12_BUFFER_CONFIG *src_ybc, YV12
 extern void vp8_yv12_copy_partial_frame(YV12_BUFFER_CONFIG *src_ybc, YV12_BUFFER_CONFIG *dst_ybc, int Fraction);
 extern void vpxyv12_copy_partial_frame_neon(YV12_BUFFER_CONFIG *src_ybc, YV12_BUFFER_CONFIG *dst_ybc, int Fraction);
 
+extern unsigned int vp8_sub_pixel_variance16x16_neon_func
+(
+    const unsigned char  *src_ptr,
+    int  src_pixels_per_line,
+    int  xoffset,
+    int  yoffset,
+    const unsigned char *dst_ptr,
+    int dst_pixels_per_line,
+    unsigned int *sse
+);
+unsigned int vp8_sub_pixel_variance16x16_neon
+(
+    const unsigned char  *src_ptr,
+    int  src_pixels_per_line,
+    int  xoffset,
+    int  yoffset,
+    const unsigned char *dst_ptr,
+    int dst_pixels_per_line,
+    unsigned int *sse
+)
+{
+  if (xoffset == 4 && yoffset == 0)
+      return vp8_variance_halfpixvar16x16_h_neon(src_ptr, src_pixels_per_line, dst_ptr, dst_pixels_per_line, sse);
+  else if (xoffset == 0 && yoffset == 4)
+      return vp8_variance_halfpixvar16x16_v_neon(src_ptr, src_pixels_per_line, dst_ptr, dst_pixels_per_line, sse);
+  else if (xoffset == 4 && yoffset == 4)
+      return vp8_variance_halfpixvar16x16_hv_neon(src_ptr, src_pixels_per_line, dst_ptr, dst_pixels_per_line, sse);
+  else
+      return vp8_sub_pixel_variance16x16_neon_func(src_ptr, src_pixels_per_line, xoffset, yoffset, dst_ptr, dst_pixels_per_line, sse);
+}
+
 void vp8_arch_arm_encoder_init(VP8_COMP *cpi)
 {
 #if CONFIG_RUNTIME_CPU_DETECT

diff --git a/vp8/encoder/arm/neon/vp8_subpixelvariance16x16_neon.asm b/vp8/encoder/arm/neon/vp8_subpixelvariance16x16_neon.asm
index 1b09cfe4cb268faa7e3a4213d507fa7e2b00a44b..1475f76df212a4fd5e77c21cb7114bb0d30b179f 100644 (file)
--- a/vp8/encoder/arm/neon/vp8_subpixelvariance16x16_neon.asm
+++ b/vp8/encoder/arm/neon/vp8_subpixelvariance16x16_neon.asm
@@ -9,7 +9,7 @@
 ;
 
 
-    EXPORT  |vp8_sub_pixel_variance16x16_neon|
+    EXPORT  |vp8_sub_pixel_variance16x16_neon_func|
     ARM
     REQUIRE8
     PRESERVE8
@@ -24,7 +24,7 @@
 ; stack(r6) unsigned int *sse
 ;note: most of the code is copied from bilinear_predict16x16_neon and vp8_variance16x16_neon.
 
-|vp8_sub_pixel_variance16x16_neon| PROC
+|vp8_sub_pixel_variance16x16_neon_func| PROC
     push            {r4-r6, lr}
 
     ldr             r12, _BilinearTaps_coeff_