Add Neon implementation of OBMC SAD functions
Add Neon implementation of aom_obmc_sad<w>x<h> functions, as well as the
corresponding tests.
Change-Id: I0691444aa0250d52ce52e912c082c7770e2903b0
diff --git a/aom_dsp/aom_dsp.cmake b/aom_dsp/aom_dsp.cmake
index f9dbc55f..6e11237 100644
--- a/aom_dsp/aom_dsp.cmake
+++ b/aom_dsp/aom_dsp.cmake
@@ -265,6 +265,7 @@
"${AOM_ROOT}/aom_dsp/arm/hadamard_neon.c"
"${AOM_ROOT}/aom_dsp/arm/avg_neon.c"
"${AOM_ROOT}/aom_dsp/arm/obmc_variance_neon.c"
+ "${AOM_ROOT}/aom_dsp/arm/obmc_sad_neon.c"
"${AOM_ROOT}/aom_dsp/arm/sse_neon.c"
"${AOM_ROOT}/aom_dsp/arm/sum_squares_neon.c")
diff --git a/aom_dsp/aom_dsp_rtcd_defs.pl b/aom_dsp/aom_dsp_rtcd_defs.pl
index e02d244..81b3153 100755
--- a/aom_dsp/aom_dsp_rtcd_defs.pl
+++ b/aom_dsp/aom_dsp_rtcd_defs.pl
@@ -968,7 +968,7 @@
($w, $h) = @$_;
add_proto qw/unsigned int/, "aom_obmc_sad${w}x${h}", "const uint8_t *pre, int pre_stride, const int32_t *wsrc, const int32_t *mask";
if (! (($w == 128 && $h == 32) || ($w == 32 && $h == 128))) {
- specialize "aom_obmc_sad${w}x${h}", qw/sse4_1 avx2/;
+ specialize "aom_obmc_sad${w}x${h}", qw/sse4_1 avx2 neon/;
}
}
diff --git a/aom_dsp/arm/obmc_sad_neon.c b/aom_dsp/arm/obmc_sad_neon.c
new file mode 100644
index 0000000..60656e5
--- /dev/null
+++ b/aom_dsp/arm/obmc_sad_neon.c
@@ -0,0 +1,250 @@
+/*
+ * Copyright (c) 2023, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#include <arm_neon.h>
+#include "config/aom_config.h"
+#include "config/aom_dsp_rtcd.h"
+#include "mem_neon.h"
+#include "sum_neon.h"
+
+static INLINE void obmc_sad_8x1_s16_neon(int16x8_t ref_s16, const int32_t *mask,
+ const int32_t *wsrc, uint32x4_t *sum) {
+ int32x4_t wsrc_lo = vld1q_s32(wsrc);
+ int32x4_t wsrc_hi = vld1q_s32(wsrc + 4);
+
+ int32x4_t mask_lo = vld1q_s32(mask);
+ int32x4_t mask_hi = vld1q_s32(mask + 4);
+
+ int16x8_t mask_s16 =
+ vuzpq_s16(vreinterpretq_s16_s32(mask_lo), vreinterpretq_s16_s32(mask_hi))
+ .val[0];
+
+ int32x4_t pre_lo = vmull_s16(vget_low_s16(ref_s16), vget_low_s16(mask_s16));
+ int32x4_t pre_hi = vmull_s16(vget_high_s16(ref_s16), vget_high_s16(mask_s16));
+
+ uint32x4_t abs_lo = vreinterpretq_u32_s32(vabdq_s32(wsrc_lo, pre_lo));
+ uint32x4_t abs_hi = vreinterpretq_u32_s32(vabdq_s32(wsrc_hi, pre_hi));
+
+ *sum = vrsraq_n_u32(*sum, abs_lo, 12);
+ *sum = vrsraq_n_u32(*sum, abs_hi, 12);
+}
+
+#if defined(__aarch64__)
+
+// Use tbl for doing a double-width zero extension from 8->32 bits since we can
+// do this in one instruction rather than two (indices out of range (255 here)
+// are set to zero by tbl).
+DECLARE_ALIGNED(16, static const uint8_t, obmc_variance_permute_idx[]) = {
+ 0, 255, 255, 255, 1, 255, 255, 255, 2, 255, 255, 255, 3, 255, 255, 255,
+ 4, 255, 255, 255, 5, 255, 255, 255, 6, 255, 255, 255, 7, 255, 255, 255,
+ 8, 255, 255, 255, 9, 255, 255, 255, 10, 255, 255, 255, 11, 255, 255, 255,
+ 12, 255, 255, 255, 13, 255, 255, 255, 14, 255, 255, 255, 15, 255, 255, 255
+};
+
+static INLINE void obmc_sad_8x1_s32_neon(uint32x4_t ref_u32_lo,
+ uint32x4_t ref_u32_hi,
+ const int32_t *mask,
+ const int32_t *wsrc,
+ uint32x4_t sum[2]) {
+ int32x4_t wsrc_lo = vld1q_s32(wsrc);
+ int32x4_t wsrc_hi = vld1q_s32(wsrc + 4);
+ int32x4_t mask_lo = vld1q_s32(mask);
+ int32x4_t mask_hi = vld1q_s32(mask + 4);
+
+ int32x4_t pre_lo = vmulq_s32(vreinterpretq_s32_u32(ref_u32_lo), mask_lo);
+ int32x4_t pre_hi = vmulq_s32(vreinterpretq_s32_u32(ref_u32_hi), mask_hi);
+
+ uint32x4_t abs_lo = vreinterpretq_u32_s32(vabdq_s32(wsrc_lo, pre_lo));
+ uint32x4_t abs_hi = vreinterpretq_u32_s32(vabdq_s32(wsrc_hi, pre_hi));
+
+ sum[0] = vrsraq_n_u32(sum[0], abs_lo, 12);
+ sum[1] = vrsraq_n_u32(sum[1], abs_hi, 12);
+}
+
+static INLINE unsigned int obmc_sad_large_neon(const uint8_t *ref,
+ int ref_stride,
+ const int32_t *wsrc,
+ const int32_t *mask, int width,
+ int height) {
+ uint32x4_t sum[2] = { vdupq_n_u32(0), vdupq_n_u32(0) };
+
+ // Use tbl for doing a double-width zero extension from 8->32 bits since we
+ // can do this in one instruction rather than two.
+ uint8x16_t pre_idx0 = vld1q_u8(&obmc_variance_permute_idx[0]);
+ uint8x16_t pre_idx1 = vld1q_u8(&obmc_variance_permute_idx[16]);
+ uint8x16_t pre_idx2 = vld1q_u8(&obmc_variance_permute_idx[32]);
+ uint8x16_t pre_idx3 = vld1q_u8(&obmc_variance_permute_idx[48]);
+
+ int h = height;
+ do {
+ int w = width;
+ const uint8_t *ref_ptr = ref;
+ do {
+ uint8x16_t r = vld1q_u8(ref_ptr);
+
+ uint32x4_t ref_u32_lo = vreinterpretq_u32_u8(vqtbl1q_u8(r, pre_idx0));
+ uint32x4_t ref_u32_hi = vreinterpretq_u32_u8(vqtbl1q_u8(r, pre_idx1));
+ obmc_sad_8x1_s32_neon(ref_u32_lo, ref_u32_hi, mask, wsrc, sum);
+
+ ref_u32_lo = vreinterpretq_u32_u8(vqtbl1q_u8(r, pre_idx2));
+ ref_u32_hi = vreinterpretq_u32_u8(vqtbl1q_u8(r, pre_idx3));
+ obmc_sad_8x1_s32_neon(ref_u32_lo, ref_u32_hi, mask + 8, wsrc + 8, sum);
+
+ ref_ptr += 16;
+ wsrc += 16;
+ mask += 16;
+ w -= 16;
+ } while (w != 0);
+
+ ref += ref_stride;
+ } while (--h != 0);
+
+ return horizontal_add_u32x4(vaddq_u32(sum[0], sum[1]));
+}
+
+#else // !defined(__aarch64__)
+
+static INLINE unsigned int obmc_sad_large_neon(const uint8_t *ref,
+ int ref_stride,
+ const int32_t *wsrc,
+ const int32_t *mask, int width,
+ int height) {
+ uint32x4_t sum = vdupq_n_u32(0);
+
+ int h = height;
+ do {
+ int w = width;
+ const uint8_t *ref_ptr = ref;
+ do {
+ uint8x16_t r = vld1q_u8(ref_ptr);
+
+ int16x8_t ref_s16 = vreinterpretq_s16_u16(vmovl_u8(vget_low_u8(r)));
+ obmc_sad_8x1_s16_neon(ref_s16, mask, wsrc, &sum);
+
+ ref_s16 = vreinterpretq_s16_u16(vmovl_u8(vget_high_u8(r)));
+ obmc_sad_8x1_s16_neon(ref_s16, mask + 8, wsrc + 8, &sum);
+
+ ref_ptr += 16;
+ wsrc += 16;
+ mask += 16;
+ w -= 16;
+ } while (w != 0);
+
+ ref += ref_stride;
+ } while (--h != 0);
+
+ return horizontal_add_u32x4(sum);
+}
+
+#endif // defined(__aarch64__)
+
+static INLINE unsigned int obmc_sad_128xh_neon(const uint8_t *ref,
+ int ref_stride,
+ const int32_t *wsrc,
+ const int32_t *mask, int h) {
+ return obmc_sad_large_neon(ref, ref_stride, wsrc, mask, 128, h);
+}
+
+static INLINE unsigned int obmc_sad_64xh_neon(const uint8_t *ref,
+ int ref_stride,
+ const int32_t *wsrc,
+ const int32_t *mask, int h) {
+ return obmc_sad_large_neon(ref, ref_stride, wsrc, mask, 64, h);
+}
+
+static INLINE unsigned int obmc_sad_32xh_neon(const uint8_t *ref,
+ int ref_stride,
+ const int32_t *wsrc,
+ const int32_t *mask, int h) {
+ return obmc_sad_large_neon(ref, ref_stride, wsrc, mask, 32, h);
+}
+
+static INLINE unsigned int obmc_sad_16xh_neon(const uint8_t *ref,
+ int ref_stride,
+ const int32_t *wsrc,
+ const int32_t *mask, int h) {
+ return obmc_sad_large_neon(ref, ref_stride, wsrc, mask, 16, h);
+}
+
+static INLINE unsigned int obmc_sad_8xh_neon(const uint8_t *ref, int ref_stride,
+ const int32_t *wsrc,
+ const int32_t *mask, int height) {
+ uint32x4_t sum = vdupq_n_u32(0);
+
+ int h = height;
+ do {
+ uint8x8_t r = vld1_u8(ref);
+
+ int16x8_t ref_s16 = vreinterpretq_s16_u16(vmovl_u8(r));
+ obmc_sad_8x1_s16_neon(ref_s16, mask, wsrc, &sum);
+
+ ref += ref_stride;
+ wsrc += 8;
+ mask += 8;
+ } while (--h != 0);
+
+ return horizontal_add_u32x4(sum);
+}
+
+static INLINE unsigned int obmc_sad_4xh_neon(const uint8_t *ref, int ref_stride,
+ const int32_t *wsrc,
+ const int32_t *mask, int height) {
+ uint32x4_t sum = vdupq_n_u32(0);
+
+ int h = height / 2;
+ do {
+ uint8x8_t r = load_unaligned_u8(ref, ref_stride);
+
+ int16x8_t ref_s16 = vreinterpretq_s16_u16(vmovl_u8(r));
+ obmc_sad_8x1_s16_neon(ref_s16, mask, wsrc, &sum);
+
+ ref += 2 * ref_stride;
+ wsrc += 8;
+ mask += 8;
+ } while (--h != 0);
+
+ return horizontal_add_u32x4(sum);
+}
+
+#define OBMC_SAD_WXH_NEON(w, h) \
+ unsigned int aom_obmc_sad##w##x##h##_neon( \
+ const uint8_t *ref, int ref_stride, const int32_t *wsrc, \
+ const int32_t *mask) { \
+ return obmc_sad_##w##xh_neon(ref, ref_stride, wsrc, mask, h); \
+ }
+
+OBMC_SAD_WXH_NEON(4, 4)
+OBMC_SAD_WXH_NEON(4, 8)
+OBMC_SAD_WXH_NEON(4, 16)
+
+OBMC_SAD_WXH_NEON(8, 4)
+OBMC_SAD_WXH_NEON(8, 8)
+OBMC_SAD_WXH_NEON(8, 16)
+OBMC_SAD_WXH_NEON(8, 32)
+
+OBMC_SAD_WXH_NEON(16, 4)
+OBMC_SAD_WXH_NEON(16, 8)
+OBMC_SAD_WXH_NEON(16, 16)
+OBMC_SAD_WXH_NEON(16, 32)
+OBMC_SAD_WXH_NEON(16, 64)
+
+OBMC_SAD_WXH_NEON(32, 8)
+OBMC_SAD_WXH_NEON(32, 16)
+OBMC_SAD_WXH_NEON(32, 32)
+OBMC_SAD_WXH_NEON(32, 64)
+
+OBMC_SAD_WXH_NEON(64, 16)
+OBMC_SAD_WXH_NEON(64, 32)
+OBMC_SAD_WXH_NEON(64, 64)
+OBMC_SAD_WXH_NEON(64, 128)
+
+OBMC_SAD_WXH_NEON(128, 64)
+OBMC_SAD_WXH_NEON(128, 128)
diff --git a/test/obmc_sad_test.cc b/test/obmc_sad_test.cc
index 9b70366..8d13ac1 100644
--- a/test/obmc_sad_test.cc
+++ b/test/obmc_sad_test.cc
@@ -147,6 +147,37 @@
::testing::ValuesIn(avx2_functions));
#endif // HAVE_AVX2
+#if HAVE_NEON
+const ObmcSadTest::ParamType neon_functions[] = {
+ TestFuncs(aom_obmc_sad128x128_c, aom_obmc_sad128x128_neon),
+ TestFuncs(aom_obmc_sad128x64_c, aom_obmc_sad128x64_neon),
+ TestFuncs(aom_obmc_sad64x128_c, aom_obmc_sad64x128_neon),
+ TestFuncs(aom_obmc_sad64x64_c, aom_obmc_sad64x64_neon),
+ TestFuncs(aom_obmc_sad64x32_c, aom_obmc_sad64x32_neon),
+ TestFuncs(aom_obmc_sad32x64_c, aom_obmc_sad32x64_neon),
+ TestFuncs(aom_obmc_sad32x32_c, aom_obmc_sad32x32_neon),
+ TestFuncs(aom_obmc_sad32x16_c, aom_obmc_sad32x16_neon),
+ TestFuncs(aom_obmc_sad16x32_c, aom_obmc_sad16x32_neon),
+ TestFuncs(aom_obmc_sad16x16_c, aom_obmc_sad16x16_neon),
+ TestFuncs(aom_obmc_sad16x8_c, aom_obmc_sad16x8_neon),
+ TestFuncs(aom_obmc_sad8x16_c, aom_obmc_sad8x16_neon),
+ TestFuncs(aom_obmc_sad8x8_c, aom_obmc_sad8x8_neon),
+ TestFuncs(aom_obmc_sad8x4_c, aom_obmc_sad8x4_neon),
+ TestFuncs(aom_obmc_sad4x8_c, aom_obmc_sad4x8_neon),
+ TestFuncs(aom_obmc_sad4x4_c, aom_obmc_sad4x4_neon),
+
+ TestFuncs(aom_obmc_sad64x16_c, aom_obmc_sad64x16_neon),
+ TestFuncs(aom_obmc_sad16x64_c, aom_obmc_sad16x64_neon),
+ TestFuncs(aom_obmc_sad32x8_c, aom_obmc_sad32x8_neon),
+ TestFuncs(aom_obmc_sad8x32_c, aom_obmc_sad8x32_neon),
+ TestFuncs(aom_obmc_sad16x4_c, aom_obmc_sad16x4_neon),
+ TestFuncs(aom_obmc_sad4x16_c, aom_obmc_sad4x16_neon),
+};
+
+INSTANTIATE_TEST_SUITE_P(NEON, ObmcSadTest,
+ ::testing::ValuesIn(neon_functions));
+#endif // HAVE_NEON
+
#if CONFIG_AV1_HIGHBITDEPTH
////////////////////////////////////////////////////////////////////////////////
// High bit-depth
