Add vp9_tm_predictor_4x4 neon implementation

Change-Id: I10c423bde7ea5a3bac9f14f35c73b6bc31c8f3e3

diff --git a/vp9/common/arm/neon/vp9_reconintra_neon.asm b/vp9/common/arm/neon/vp9_reconintra_neon.asm
index f106bc78e24df8de3209591378e2d0cabec4cec9..98619bb30582f9042844cfa611bc4b3c2c923641 100644 (file)
--- a/vp9/common/arm/neon/vp9_reconintra_neon.asm
+++ b/vp9/common/arm/neon/vp9_reconintra_neon.asm
@@ -16,6 +16,7 @@
     EXPORT  |vp9_h_predictor_8x8_neon|
     EXPORT  |vp9_h_predictor_16x16_neon|
     EXPORT  |vp9_h_predictor_32x32_neon|
+    EXPORT  |vp9_tm_predictor_4x4_neon|
     ARM
     REQUIRE8
     PRESERVE8
@@ -283,4 +284,52 @@ loop_h
     bx                  lr
     ENDP                ; |vp9_h_predictor_32x32_neon|
 
+;void vp9_tm_predictor_4x4_neon (uint8_t *dst, ptrdiff_t y_stride,
+;                                const uint8_t *above,
+;                                const uint8_t *left)
+; r0  uint8_t *dst
+; r1  ptrdiff_t y_stride
+; r2  const uint8_t *above
+; r3  const uint8_t *left
+
+|vp9_tm_predictor_4x4_neon| PROC
+    ; Load ytop_left = above[-1];
+    sub                 r12, r2, #1
+    ldrb                r12, [r12]
+    vdup.u8             d0, r12
+
+    ; Load above 4 pixels
+    vld1.32             {d2[0]}, [r2]
+
+    ; Compute above - ytop_left
+    vsubl.u8            q3, d2, d0
+
+    ; Load left row by row and compute left + (above - ytop_left)
+    ; 1st row and 2nd row
+    ldrb                r12, [r3], #1
+    ldrb                r2, [r3], #1
+    vdup.u16            q1, r12
+    vdup.u16            q2, r2
+    vadd.s16            q1, q1, q3
+    vadd.s16            q2, q2, q3
+    vqshrun.s16         d0, q1, #0
+    vqshrun.s16         d1, q2, #0
+    vst1.32             {d0[0]}, [r0], r1
+    vst1.32             {d1[0]}, [r0], r1
+
+    ; 3rd row and 4th row
+    ldrb                r12, [r3], #1
+    ldrb                r2, [r3], #1
+    vdup.u16            q1, r12
+    vdup.u16            q2, r2
+    vadd.s16            q1, q1, q3
+    vadd.s16            q2, q2, q3
+    vqshrun.s16         d0, q1, #0
+    vqshrun.s16         d1, q2, #0
+    vst1.32             {d0[0]}, [r0], r1
+    vst1.32             {d1[0]}, [r0], r1
+
+    bx                  lr
+    ENDP                ; |vp9_tm_predictor_4x4_neon|
+
     END

diff --git a/vp9/common/vp9_rtcd_defs.sh b/vp9/common/vp9_rtcd_defs.sh
index cc571072fcf67ae8871da54a5dcd525ae9994049..a32f98aa50ee6182e8af92484702f11d6d1ad933 100644 (file)
--- a/vp9/common/vp9_rtcd_defs.sh
+++ b/vp9/common/vp9_rtcd_defs.sh
@@ -57,7 +57,7 @@ prototype void vp9_v_predictor_4x4 "uint8_t *dst, ptrdiff_t y_stride, const uint
 specialize vp9_v_predictor_4x4 $sse_x86inc neon
 
 prototype void vp9_tm_predictor_4x4 "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left"
-specialize vp9_tm_predictor_4x4 $sse_x86inc dspr2
+specialize vp9_tm_predictor_4x4 $sse_x86inc neon dspr2
 
 prototype void vp9_dc_predictor_4x4 "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left"
 specialize vp9_dc_predictor_4x4 $sse_x86inc dspr2