Merge "Add Neon implementation of vp9_highbd_block_error_c" into main

3 files changed, 53 insertions, 1 deletions

diff --git a/vp9/common/vp9_rtcd_defs.pl b/vp9/common/vp9_rtcd_defs.pl
index d16b94711..7f77a36d6 100644
--- a/vp9/common/vp9_rtcd_defs.pl
+++ b/vp9/common/vp9_rtcd_defs.pl
@@ -139,7 +139,7 @@ if (vpx_config("CONFIG_VP9_HIGHBITDEPTH") eq "yes") {
   specialize qw/vp9_block_error neon avx2 sse2/;
 
   add_proto qw/int64_t vp9_highbd_block_error/, "const tran_low_t *coeff, const tran_low_t *dqcoeff, intptr_t block_size, int64_t *ssz, int bd";
-  specialize qw/vp9_highbd_block_error sse2/;
+  specialize qw/vp9_highbd_block_error neon sse2/;
 } else {
   specialize qw/vp9_block_error neon avx2 msa sse2/;
 }
diff --git a/vp9/encoder/arm/neon/vp9_highbd_error_neon.c b/vp9/encoder/arm/neon/vp9_highbd_error_neon.c
new file mode 100644
index 000000000..d9b183472
--- /dev/null
+++ b/vp9/encoder/arm/neon/vp9_highbd_error_neon.c
@@ -0,0 +1,49 @@
+/*
+ *  Copyright (c) 2023 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#include <arm_neon.h>
+#include <assert.h>
+
+#include "./vp9_rtcd.h"
+#include "vpx_dsp/arm/mem_neon.h"
+#include "vpx_dsp/arm/sum_neon.h"
+
+int64_t vp9_highbd_block_error_neon(const tran_low_t *coeff,
+                                    const tran_low_t *dqcoeff,
+                                    intptr_t block_size, int64_t *ssz, int bd) {
+  uint64x2_t err_u64 = vdupq_n_u64(0);
+  int64x2_t ssz_s64 = vdupq_n_s64(0);
+
+  const int shift = 2 * (bd - 8);
+  const int rounding = shift > 0 ? 1 << (shift - 1) : 0;
+
+  assert(block_size >= 16);
+  assert((block_size % 16) == 0);
+
+  do {
+    const int32x4_t c = load_tran_low_to_s32q(coeff);
+    const int32x4_t d = load_tran_low_to_s32q(dqcoeff);
+
+    const uint32x4_t diff = vreinterpretq_u32_s32(vabdq_s32(c, d));
+
+    err_u64 = vmlal_u32(err_u64, vget_low_u32(diff), vget_low_u32(diff));
+    err_u64 = vmlal_u32(err_u64, vget_high_u32(diff), vget_high_u32(diff));
+
+    ssz_s64 = vmlal_s32(ssz_s64, vget_low_s32(c), vget_low_s32(c));
+    ssz_s64 = vmlal_s32(ssz_s64, vget_high_s32(c), vget_high_s32(c));
+
+    coeff += 4;
+    dqcoeff += 4;
+    block_size -= 4;
+  } while (block_size != 0);
+
+  *ssz = (horizontal_add_int64x2(ssz_s64) + rounding) >> shift;
+  return ((int64_t)horizontal_add_uint64x2(err_u64) + rounding) >> shift;
+}
diff --git a/vp9/vp9cx.mk b/vp9/vp9cx.mk
index 882c12d2e..3e5c0e1a7 100644
--- a/vp9/vp9cx.mk
+++ b/vp9/vp9cx.mk
@@ -142,6 +142,9 @@ VP9_CX_SRCS-$(HAVE_AVX2) += encoder/x86/vp9_error_avx2.c
 VP9_CX_SRCS-$(HAVE_NEON) += encoder/arm/neon/vp9_error_neon.c
 VP9_CX_SRCS-$(HAVE_NEON) += encoder/arm/neon/vp9_frame_scale_neon.c
 VP9_CX_SRCS-$(HAVE_NEON) += encoder/arm/neon/vp9_quantize_neon.c
+ifeq ($(CONFIG_VP9_HIGHBITDEPTH),yes)
+VP9_CX_SRCS-$(HAVE_NEON) += encoder/arm/neon/vp9_highbd_error_neon.c
+endif
 
 VP9_CX_SRCS-$(HAVE_MSA) += encoder/mips/msa/vp9_error_msa.c