1 files changed, 65 insertions, 0 deletions
diff --git a/vpx_dsp/arm/fdct_partial_neon.c b/vpx_dsp/arm/fdct_partial_neon.c
index 0a1cdca41..718dba0d9 100644
--- a/vpx_dsp/arm/fdct_partial_neon.c
+++ b/vpx_dsp/arm/fdct_partial_neon.c
@@ -101,3 +101,68 @@ void vpx_fdct32x32_1_neon(const int16_t *input, tran_low_t *output,
   output[0] = (tran_low_t)(sum >> 3);
   output[1] = 0;
 }
+
+#if CONFIG_VP9_HIGHBITDEPTH
+
+void vpx_highbd_fdct16x16_1_neon(const int16_t *input, tran_low_t *output,
+                                 int stride) {
+  int32x4_t partial_sum[4] = { vdupq_n_s32(0), vdupq_n_s32(0), vdupq_n_s32(0),
+                               vdupq_n_s32(0) };
+  int32_t sum;
+
+  int r = 0;
+  do {
+    const int16x8_t a = vld1q_s16(input);
+    const int16x8_t b = vld1q_s16(input + 8);
+    input += stride;
+    partial_sum[0] = vaddw_s16(partial_sum[0], vget_low_s16(a));
+    partial_sum[1] = vaddw_s16(partial_sum[1], vget_high_s16(a));
+    partial_sum[2] = vaddw_s16(partial_sum[2], vget_low_s16(b));
+    partial_sum[3] = vaddw_s16(partial_sum[3], vget_high_s16(b));
+    r++;
+  } while (r < 16);
+
+  partial_sum[0] = vaddq_s32(partial_sum[0], partial_sum[1]);
+  partial_sum[2] = vaddq_s32(partial_sum[2], partial_sum[3]);
+  partial_sum[0] = vaddq_s32(partial_sum[0], partial_sum[2]);
+  sum = horizontal_add_int32x4(partial_sum[0]);
+
+  output[0] = (tran_low_t)(sum >> 1);
+  output[1] = 0;
+}
+
+void vpx_highbd_fdct32x32_1_neon(const int16_t *input, tran_low_t *output,
+                                 int stride) {
+  int32x4_t partial_sum[4] = { vdupq_n_s32(0), vdupq_n_s32(0), vdupq_n_s32(0),
+                               vdupq_n_s32(0) };
+
+  int32_t sum;
+
+  int r = 0;
+  do {
+    const int16x8_t a0 = vld1q_s16(input);
+    const int16x8_t a1 = vld1q_s16(input + 8);
+    const int16x8_t a2 = vld1q_s16(input + 16);
+    const int16x8_t a3 = vld1q_s16(input + 24);
+    input += stride;
+    partial_sum[0] = vaddw_s16(partial_sum[0], vget_low_s16(a0));
+    partial_sum[0] = vaddw_s16(partial_sum[0], vget_high_s16(a0));
+    partial_sum[1] = vaddw_s16(partial_sum[1], vget_low_s16(a1));
+    partial_sum[1] = vaddw_s16(partial_sum[1], vget_high_s16(a1));
+    partial_sum[2] = vaddw_s16(partial_sum[2], vget_low_s16(a2));
+    partial_sum[2] = vaddw_s16(partial_sum[2], vget_high_s16(a2));
+    partial_sum[3] = vaddw_s16(partial_sum[3], vget_low_s16(a3));
+    partial_sum[3] = vaddw_s16(partial_sum[3], vget_high_s16(a3));
+    r++;
+  } while (r < 32);
+
+  partial_sum[0] = vaddq_s32(partial_sum[0], partial_sum[1]);
+  partial_sum[2] = vaddq_s32(partial_sum[2], partial_sum[3]);
+  partial_sum[0] = vaddq_s32(partial_sum[0], partial_sum[2]);
+  sum = horizontal_add_int32x4(partial_sum[0]);
+
+  output[0] = (tran_low_t)(sum >> 3);
+  output[1] = 0;
+}
+
+#endif  // CONFIG_VP9_HIGHBITDEPTH