Don't expand to Q register for 4x4 intrapred

The code was expanding to Q registers so that vqrshn could be used, for
vector quad round shift and narrow. If 4 values are added together,
there is a shift by 2. If 8 values, a shift by 3. Since this accounts
for any possibility of overflow, we can skip the narrowing shift.

This allows keeping the values in D registers and casting the 16 bit
value to 8 bits.

Change-Id: I8d9cfa07176271f492c116ffa6a7b351af0b8751

diff --git a/vpx_dsp/arm/intrapred_neon.c b/vpx_dsp/arm/intrapred_neon.c
index 32dd1ba14606f25efabf12195ff882f147a31556..38e79ed69dd54f8fb4d25ed2f684e533b4f79395 100644 (file)
--- a/vpx_dsp/arm/intrapred_neon.c
+++ b/vpx_dsp/arm/intrapred_neon.c
@@ -20,37 +20,35 @@
 // 'do_above' and 'do_left' facilitate branch removal when inlined.
 static INLINE void dc_4x4(uint8_t *dst, ptrdiff_t stride, const uint8_t *above,
                           const uint8_t *left, int do_above, int do_left) {
-  uint16x8_t sum_top;
-  uint16x8_t sum_left;
-  uint8x8_t dc0;
+  uint16x4_t sum_top;
+  uint16x4_t sum_left;
+  uint16x4_t dc0;
 
   if (do_above) {
     const uint8x8_t A = vld1_u8(above);  // top row
     const uint16x4_t p0 = vpaddl_u8(A);  // cascading summation of the top
-    const uint16x4_t p1 = vpadd_u16(p0, p0);
-    sum_top = vcombine_u16(p1, p1);
+    sum_top = vpadd_u16(p0, p0);
   }
 
   if (do_left) {
     const uint8x8_t L = vld1_u8(left);   // left border
     const uint16x4_t p0 = vpaddl_u8(L);  // cascading summation of the left
-    const uint16x4_t p1 = vpadd_u16(p0, p0);
-    sum_left = vcombine_u16(p1, p1);
+    sum_left = vpadd_u16(p0, p0);
   }
 
   if (do_above && do_left) {
-    const uint16x8_t sum = vaddq_u16(sum_left, sum_top);
-    dc0 = vrshrn_n_u16(sum, 3);
+    const uint16x4_t sum = vadd_u16(sum_left, sum_top);
+    dc0 = vrshr_n_u16(sum, 3);
   } else if (do_above) {
-    dc0 = vrshrn_n_u16(sum_top, 2);
+    dc0 = vrshr_n_u16(sum_top, 2);
   } else if (do_left) {
-    dc0 = vrshrn_n_u16(sum_left, 2);
+    dc0 = vrshr_n_u16(sum_left, 2);
   } else {
-    dc0 = vdup_n_u8(0x80);
+    dc0 = vdup_n_u16(0x80);
   }
 
   {
-    const uint8x8_t dc = vdup_lane_u8(dc0, 0);
+    const uint8x8_t dc = vdup_lane_u8(vreinterpret_u8_u16(dc0), 0);
     int i;
     for (i = 0; i < 4; ++i) {
       vst1_lane_u32((uint32_t *)(dst + i * stride), vreinterpret_u32_u8(dc), 0);