Split big file into two
The file sad.c alone takes 35 seconds to compile.
This often happens to be on the build critical path.
Split it into two source files so they can be compiled in parallel.
Change-Id: I35636d8a3da9d67edb8dbf202fd5e7a687a6aaa9
diff --git a/aom_dsp/aom_dsp.cmake b/aom_dsp/aom_dsp.cmake
index 71a6fb1..61140ab 100644
--- a/aom_dsp/aom_dsp.cmake
+++ b/aom_dsp/aom_dsp.cmake
@@ -328,6 +328,7 @@
"${AOM_ROOT}/aom_dsp/psnr.c"
"${AOM_ROOT}/aom_dsp/psnr.h"
"${AOM_ROOT}/aom_dsp/sad.c"
+ "${AOM_ROOT}/aom_dsp/sad_av1.c"
"${AOM_ROOT}/aom_dsp/variance.c"
"${AOM_ROOT}/aom_dsp/variance.h")
diff --git a/aom_dsp/aom_dsp.mk b/aom_dsp/aom_dsp.mk
index 5b94224..a682d68 100644
--- a/aom_dsp/aom_dsp.mk
+++ b/aom_dsp/aom_dsp.mk
@@ -344,6 +344,7 @@
ifeq ($(CONFIG_AV1_ENCODER),yes)
DSP_SRCS-yes += sad.c
+DSP_SRCS-yes += sad_av1.c
DSP_SRCS-yes += subtract.c
DSP_SRCS-$(HAVE_NEON) += arm/sad4d_neon.c
diff --git a/aom_dsp/sad.c b/aom_dsp/sad.c
index 6a9a18f..e42ec1b 100644
--- a/aom_dsp/sad.c
+++ b/aom_dsp/sad.c
@@ -388,267 +388,3 @@
#endif
/* clang-format on */
#endif // CONFIG_HIGHBITDEPTH
-
-#if CONFIG_AV1
- static INLINE
- unsigned int masked_sad(const uint8_t *src, int src_stride,
- const uint8_t *a, int a_stride, const uint8_t *b,
- int b_stride, const uint8_t *m, int m_stride,
- int width, int height) {
- int y, x;
- unsigned int sad = 0;
- for (y = 0; y < height; y++) {
- for (x = 0; x < width; x++) {
- const int16_t pred = AOM_BLEND_A64(m[x], a[x], b[x]);
- sad += abs(pred - src[x]);
- }
- src += src_stride;
- a += a_stride;
- b += b_stride;
- m += m_stride;
- }
- sad = (sad + 31) >> 6;
- return sad;
-}
-
-#define MASKSADMxN(m, n) \
- unsigned int aom_masked_sad##m##x##n##_c( \
- const uint8_t *src, int src_stride, const uint8_t *ref, int ref_stride, \
- const uint8_t *second_pred, const uint8_t *msk, int msk_stride, \
- int invert_mask) { \
- if (!invert_mask) \
- return masked_sad(src, src_stride, ref, ref_stride, second_pred, m, msk, \
- msk_stride, m, n); \
- else \
- return masked_sad(src, src_stride, second_pred, m, ref, ref_stride, msk, \
- msk_stride, m, n); \
- }
-
-/* clang-format off */
-#if CONFIG_EXT_PARTITION
-MASKSADMxN(128, 128)
-MASKSADMxN(128, 64)
-MASKSADMxN(64, 128)
-#endif // CONFIG_EXT_PARTITION
-MASKSADMxN(64, 64)
-MASKSADMxN(64, 32)
-MASKSADMxN(32, 64)
-MASKSADMxN(32, 32)
-MASKSADMxN(32, 16)
-MASKSADMxN(16, 32)
-MASKSADMxN(16, 16)
-MASKSADMxN(16, 8)
-MASKSADMxN(8, 16)
-MASKSADMxN(8, 8)
-MASKSADMxN(8, 4)
-MASKSADMxN(4, 8)
-MASKSADMxN(4, 4)
-
-#if CONFIG_EXT_PARTITION_TYPES
-MASKSADMxN(4, 16)
-MASKSADMxN(16, 4)
-MASKSADMxN(8, 32)
-MASKSADMxN(32, 8)
-MASKSADMxN(16, 64)
-MASKSADMxN(64, 16)
-MASKSADMxN(32, 128)
-MASKSADMxN(128, 32)
-#endif
-/* clang-format on */
-
-#if CONFIG_HIGHBITDEPTH
- static INLINE
- unsigned int highbd_masked_sad(const uint8_t *src8, int src_stride,
- const uint8_t *a8, int a_stride,
- const uint8_t *b8, int b_stride,
- const uint8_t *m, int m_stride, int width,
- int height) {
- int y, x;
- unsigned int sad = 0;
- const uint16_t *src = CONVERT_TO_SHORTPTR(src8);
- const uint16_t *a = CONVERT_TO_SHORTPTR(a8);
- const uint16_t *b = CONVERT_TO_SHORTPTR(b8);
-
- for (y = 0; y < height; y++) {
- for (x = 0; x < width; x++) {
- const uint16_t pred = AOM_BLEND_A64(m[x], a[x], b[x]);
- sad += abs(pred - src[x]);
- }
-
- src += src_stride;
- a += a_stride;
- b += b_stride;
- m += m_stride;
- }
- sad = (sad + 31) >> 6;
-
- return sad;
-}
-
-#define HIGHBD_MASKSADMXN(m, n) \
- unsigned int aom_highbd_masked_sad##m##x##n##_c( \
- const uint8_t *src8, int src_stride, const uint8_t *ref8, \
- int ref_stride, const uint8_t *second_pred8, const uint8_t *msk, \
- int msk_stride, int invert_mask) { \
- if (!invert_mask) \
- return highbd_masked_sad(src8, src_stride, ref8, ref_stride, \
- second_pred8, m, msk, msk_stride, m, n); \
- else \
- return highbd_masked_sad(src8, src_stride, second_pred8, m, ref8, \
- ref_stride, msk, msk_stride, m, n); \
- }
-
-#if CONFIG_EXT_PARTITION
-HIGHBD_MASKSADMXN(128, 128)
-HIGHBD_MASKSADMXN(128, 64)
-HIGHBD_MASKSADMXN(64, 128)
-#endif // CONFIG_EXT_PARTITION
-HIGHBD_MASKSADMXN(64, 64)
-HIGHBD_MASKSADMXN(64, 32)
-HIGHBD_MASKSADMXN(32, 64)
-HIGHBD_MASKSADMXN(32, 32)
-HIGHBD_MASKSADMXN(32, 16)
-HIGHBD_MASKSADMXN(16, 32)
-HIGHBD_MASKSADMXN(16, 16)
-HIGHBD_MASKSADMXN(16, 8)
-HIGHBD_MASKSADMXN(8, 16)
-HIGHBD_MASKSADMXN(8, 8)
-HIGHBD_MASKSADMXN(8, 4)
-HIGHBD_MASKSADMXN(4, 8)
-HIGHBD_MASKSADMXN(4, 4)
-
-#if CONFIG_AV1 && CONFIG_EXT_PARTITION_TYPES
-HIGHBD_MASKSADMXN(4, 16)
-HIGHBD_MASKSADMXN(16, 4)
-HIGHBD_MASKSADMXN(8, 32)
-HIGHBD_MASKSADMXN(32, 8)
-HIGHBD_MASKSADMXN(16, 64)
-HIGHBD_MASKSADMXN(64, 16)
-HIGHBD_MASKSADMXN(32, 128)
-HIGHBD_MASKSADMXN(128, 32)
-#endif
-#endif // CONFIG_HIGHBITDEPTH
-#endif // CONFIG_AV1
-
-#if CONFIG_AV1
-// pre: predictor being evaluated
-// wsrc: target weighted prediction (has been *4096 to keep precision)
-// mask: 2d weights (scaled by 4096)
-static INLINE unsigned int obmc_sad(const uint8_t *pre, int pre_stride,
- const int32_t *wsrc, const int32_t *mask,
- int width, int height) {
- int y, x;
- unsigned int sad = 0;
-
- for (y = 0; y < height; y++) {
- for (x = 0; x < width; x++)
- sad += ROUND_POWER_OF_TWO(abs(wsrc[x] - pre[x] * mask[x]), 12);
-
- pre += pre_stride;
- wsrc += width;
- mask += width;
- }
-
- return sad;
-}
-
-#define OBMCSADMxN(m, n) \
- unsigned int aom_obmc_sad##m##x##n##_c(const uint8_t *ref, int ref_stride, \
- const int32_t *wsrc, \
- const int32_t *mask) { \
- return obmc_sad(ref, ref_stride, wsrc, mask, m, n); \
- }
-
-/* clang-format off */
-#if CONFIG_EXT_PARTITION
-OBMCSADMxN(128, 128)
-OBMCSADMxN(128, 64)
-OBMCSADMxN(64, 128)
-#endif // CONFIG_EXT_PARTITION
-OBMCSADMxN(64, 64)
-OBMCSADMxN(64, 32)
-OBMCSADMxN(32, 64)
-OBMCSADMxN(32, 32)
-OBMCSADMxN(32, 16)
-OBMCSADMxN(16, 32)
-OBMCSADMxN(16, 16)
-OBMCSADMxN(16, 8)
-OBMCSADMxN(8, 16)
-OBMCSADMxN(8, 8)
-OBMCSADMxN(8, 4)
-OBMCSADMxN(4, 8)
-OBMCSADMxN(4, 4)
-
-#if CONFIG_AV1 && CONFIG_EXT_PARTITION_TYPES
-OBMCSADMxN(4, 16)
-OBMCSADMxN(16, 4)
-OBMCSADMxN(8, 32)
-OBMCSADMxN(32, 8)
-OBMCSADMxN(16, 64)
-OBMCSADMxN(64, 16)
-OBMCSADMxN(32, 128)
-OBMCSADMxN(128, 32)
-#endif
-/* clang-format on */
-
-#if CONFIG_HIGHBITDEPTH
- static INLINE
- unsigned int highbd_obmc_sad(const uint8_t *pre8, int pre_stride,
- const int32_t *wsrc, const int32_t *mask,
- int width, int height) {
- int y, x;
- unsigned int sad = 0;
- const uint16_t *pre = CONVERT_TO_SHORTPTR(pre8);
-
- for (y = 0; y < height; y++) {
- for (x = 0; x < width; x++)
- sad += ROUND_POWER_OF_TWO(abs(wsrc[x] - pre[x] * mask[x]), 12);
-
- pre += pre_stride;
- wsrc += width;
- mask += width;
- }
-
- return sad;
-}
-
-#define HIGHBD_OBMCSADMXN(m, n) \
- unsigned int aom_highbd_obmc_sad##m##x##n##_c( \
- const uint8_t *ref, int ref_stride, const int32_t *wsrc, \
- const int32_t *mask) { \
- return highbd_obmc_sad(ref, ref_stride, wsrc, mask, m, n); \
- }
-
-/* clang-format off */
-#if CONFIG_EXT_PARTITION
-HIGHBD_OBMCSADMXN(128, 128)
-HIGHBD_OBMCSADMXN(128, 64)
-HIGHBD_OBMCSADMXN(64, 128)
-#endif // CONFIG_EXT_PARTITION
-HIGHBD_OBMCSADMXN(64, 64)
-HIGHBD_OBMCSADMXN(64, 32)
-HIGHBD_OBMCSADMXN(32, 64)
-HIGHBD_OBMCSADMXN(32, 32)
-HIGHBD_OBMCSADMXN(32, 16)
-HIGHBD_OBMCSADMXN(16, 32)
-HIGHBD_OBMCSADMXN(16, 16)
-HIGHBD_OBMCSADMXN(16, 8)
-HIGHBD_OBMCSADMXN(8, 16)
-HIGHBD_OBMCSADMXN(8, 8)
-HIGHBD_OBMCSADMXN(8, 4)
-HIGHBD_OBMCSADMXN(4, 8)
-HIGHBD_OBMCSADMXN(4, 4)
-
-#if CONFIG_AV1 && CONFIG_EXT_PARTITION_TYPES
-HIGHBD_OBMCSADMXN(4, 16)
-HIGHBD_OBMCSADMXN(16, 4)
-HIGHBD_OBMCSADMXN(8, 32)
-HIGHBD_OBMCSADMXN(32, 8)
-HIGHBD_OBMCSADMXN(16, 64)
-HIGHBD_OBMCSADMXN(64, 16)
-HIGHBD_OBMCSADMXN(32, 128)
-HIGHBD_OBMCSADMXN(128, 32)
-#endif
-/* clang-format on */
-#endif // CONFIG_HIGHBITDEPTH
-#endif // CONFIG_AV1
diff --git a/aom_dsp/sad_av1.c b/aom_dsp/sad_av1.c
new file mode 100644
index 0000000..7e6ecf1
--- /dev/null
+++ b/aom_dsp/sad_av1.c
@@ -0,0 +1,281 @@
+/*
+ * Copyright (c) 2017, 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 <stdlib.h>
+
+#include "./aom_config.h"
+#include "./aom_dsp_rtcd.h"
+
+#include "aom/aom_integer.h"
+#include "aom_ports/mem.h"
+#include "aom_dsp/blend.h"
+
+#if CONFIG_AV1
+static INLINE unsigned int masked_sad(const uint8_t *src, int src_stride,
+ const uint8_t *a, int a_stride,
+ const uint8_t *b, int b_stride,
+ const uint8_t *m, int m_stride, int width,
+ int height) {
+ int y, x;
+ unsigned int sad = 0;
+ for (y = 0; y < height; y++) {
+ for (x = 0; x < width; x++) {
+ const int16_t pred = AOM_BLEND_A64(m[x], a[x], b[x]);
+ sad += abs(pred - src[x]);
+ }
+ src += src_stride;
+ a += a_stride;
+ b += b_stride;
+ m += m_stride;
+ }
+ sad = (sad + 31) >> 6;
+ return sad;
+}
+
+#define MASKSADMxN(m, n) \
+ unsigned int aom_masked_sad##m##x##n##_c( \
+ const uint8_t *src, int src_stride, const uint8_t *ref, int ref_stride, \
+ const uint8_t *second_pred, const uint8_t *msk, int msk_stride, \
+ int invert_mask) { \
+ if (!invert_mask) \
+ return masked_sad(src, src_stride, ref, ref_stride, second_pred, m, msk, \
+ msk_stride, m, n); \
+ else \
+ return masked_sad(src, src_stride, second_pred, m, ref, ref_stride, msk, \
+ msk_stride, m, n); \
+ }
+
+/* clang-format off */
+#if CONFIG_EXT_PARTITION
+MASKSADMxN(128, 128)
+MASKSADMxN(128, 64)
+MASKSADMxN(64, 128)
+#endif // CONFIG_EXT_PARTITION
+MASKSADMxN(64, 64)
+MASKSADMxN(64, 32)
+MASKSADMxN(32, 64)
+MASKSADMxN(32, 32)
+MASKSADMxN(32, 16)
+MASKSADMxN(16, 32)
+MASKSADMxN(16, 16)
+MASKSADMxN(16, 8)
+MASKSADMxN(8, 16)
+MASKSADMxN(8, 8)
+MASKSADMxN(8, 4)
+MASKSADMxN(4, 8)
+MASKSADMxN(4, 4)
+
+#if CONFIG_EXT_PARTITION_TYPES
+MASKSADMxN(4, 16)
+MASKSADMxN(16, 4)
+MASKSADMxN(8, 32)
+MASKSADMxN(32, 8)
+MASKSADMxN(16, 64)
+MASKSADMxN(64, 16)
+MASKSADMxN(32, 128)
+MASKSADMxN(128, 32)
+#endif
+/* clang-format on */
+
+#if CONFIG_HIGHBITDEPTH
+ static INLINE
+ unsigned int highbd_masked_sad(const uint8_t *src8, int src_stride,
+ const uint8_t *a8, int a_stride,
+ const uint8_t *b8, int b_stride,
+ const uint8_t *m, int m_stride, int width,
+ int height) {
+ int y, x;
+ unsigned int sad = 0;
+ const uint16_t *src = CONVERT_TO_SHORTPTR(src8);
+ const uint16_t *a = CONVERT_TO_SHORTPTR(a8);
+ const uint16_t *b = CONVERT_TO_SHORTPTR(b8);
+
+ for (y = 0; y < height; y++) {
+ for (x = 0; x < width; x++) {
+ const uint16_t pred = AOM_BLEND_A64(m[x], a[x], b[x]);
+ sad += abs(pred - src[x]);
+ }
+
+ src += src_stride;
+ a += a_stride;
+ b += b_stride;
+ m += m_stride;
+ }
+ sad = (sad + 31) >> 6;
+
+ return sad;
+}
+
+#define HIGHBD_MASKSADMXN(m, n) \
+ unsigned int aom_highbd_masked_sad##m##x##n##_c( \
+ const uint8_t *src8, int src_stride, const uint8_t *ref8, \
+ int ref_stride, const uint8_t *second_pred8, const uint8_t *msk, \
+ int msk_stride, int invert_mask) { \
+ if (!invert_mask) \
+ return highbd_masked_sad(src8, src_stride, ref8, ref_stride, \
+ second_pred8, m, msk, msk_stride, m, n); \
+ else \
+ return highbd_masked_sad(src8, src_stride, second_pred8, m, ref8, \
+ ref_stride, msk, msk_stride, m, n); \
+ }
+
+#if CONFIG_EXT_PARTITION
+HIGHBD_MASKSADMXN(128, 128)
+HIGHBD_MASKSADMXN(128, 64)
+HIGHBD_MASKSADMXN(64, 128)
+#endif // CONFIG_EXT_PARTITION
+HIGHBD_MASKSADMXN(64, 64)
+HIGHBD_MASKSADMXN(64, 32)
+HIGHBD_MASKSADMXN(32, 64)
+HIGHBD_MASKSADMXN(32, 32)
+HIGHBD_MASKSADMXN(32, 16)
+HIGHBD_MASKSADMXN(16, 32)
+HIGHBD_MASKSADMXN(16, 16)
+HIGHBD_MASKSADMXN(16, 8)
+HIGHBD_MASKSADMXN(8, 16)
+HIGHBD_MASKSADMXN(8, 8)
+HIGHBD_MASKSADMXN(8, 4)
+HIGHBD_MASKSADMXN(4, 8)
+HIGHBD_MASKSADMXN(4, 4)
+
+#if CONFIG_AV1 && CONFIG_EXT_PARTITION_TYPES
+HIGHBD_MASKSADMXN(4, 16)
+HIGHBD_MASKSADMXN(16, 4)
+HIGHBD_MASKSADMXN(8, 32)
+HIGHBD_MASKSADMXN(32, 8)
+HIGHBD_MASKSADMXN(16, 64)
+HIGHBD_MASKSADMXN(64, 16)
+HIGHBD_MASKSADMXN(32, 128)
+HIGHBD_MASKSADMXN(128, 32)
+#endif
+#endif // CONFIG_HIGHBITDEPTH
+
+// pre: predictor being evaluated
+// wsrc: target weighted prediction (has been *4096 to keep precision)
+// mask: 2d weights (scaled by 4096)
+static INLINE unsigned int obmc_sad(const uint8_t *pre, int pre_stride,
+ const int32_t *wsrc, const int32_t *mask,
+ int width, int height) {
+ int y, x;
+ unsigned int sad = 0;
+
+ for (y = 0; y < height; y++) {
+ for (x = 0; x < width; x++)
+ sad += ROUND_POWER_OF_TWO(abs(wsrc[x] - pre[x] * mask[x]), 12);
+
+ pre += pre_stride;
+ wsrc += width;
+ mask += width;
+ }
+
+ return sad;
+}
+
+#define OBMCSADMxN(m, n) \
+ unsigned int aom_obmc_sad##m##x##n##_c(const uint8_t *ref, int ref_stride, \
+ const int32_t *wsrc, \
+ const int32_t *mask) { \
+ return obmc_sad(ref, ref_stride, wsrc, mask, m, n); \
+ }
+
+/* clang-format off */
+#if CONFIG_EXT_PARTITION
+OBMCSADMxN(128, 128)
+OBMCSADMxN(128, 64)
+OBMCSADMxN(64, 128)
+#endif // CONFIG_EXT_PARTITION
+OBMCSADMxN(64, 64)
+OBMCSADMxN(64, 32)
+OBMCSADMxN(32, 64)
+OBMCSADMxN(32, 32)
+OBMCSADMxN(32, 16)
+OBMCSADMxN(16, 32)
+OBMCSADMxN(16, 16)
+OBMCSADMxN(16, 8)
+OBMCSADMxN(8, 16)
+OBMCSADMxN(8, 8)
+OBMCSADMxN(8, 4)
+OBMCSADMxN(4, 8)
+OBMCSADMxN(4, 4)
+
+#if CONFIG_AV1 && CONFIG_EXT_PARTITION_TYPES
+OBMCSADMxN(4, 16)
+OBMCSADMxN(16, 4)
+OBMCSADMxN(8, 32)
+OBMCSADMxN(32, 8)
+OBMCSADMxN(16, 64)
+OBMCSADMxN(64, 16)
+OBMCSADMxN(32, 128)
+OBMCSADMxN(128, 32)
+#endif
+/* clang-format on */
+
+#if CONFIG_HIGHBITDEPTH
+ static INLINE
+ unsigned int highbd_obmc_sad(const uint8_t *pre8, int pre_stride,
+ const int32_t *wsrc, const int32_t *mask,
+ int width, int height) {
+ int y, x;
+ unsigned int sad = 0;
+ const uint16_t *pre = CONVERT_TO_SHORTPTR(pre8);
+
+ for (y = 0; y < height; y++) {
+ for (x = 0; x < width; x++)
+ sad += ROUND_POWER_OF_TWO(abs(wsrc[x] - pre[x] * mask[x]), 12);
+
+ pre += pre_stride;
+ wsrc += width;
+ mask += width;
+ }
+
+ return sad;
+}
+
+#define HIGHBD_OBMCSADMXN(m, n) \
+ unsigned int aom_highbd_obmc_sad##m##x##n##_c( \
+ const uint8_t *ref, int ref_stride, const int32_t *wsrc, \
+ const int32_t *mask) { \
+ return highbd_obmc_sad(ref, ref_stride, wsrc, mask, m, n); \
+ }
+
+/* clang-format off */
+#if CONFIG_EXT_PARTITION
+HIGHBD_OBMCSADMXN(128, 128)
+HIGHBD_OBMCSADMXN(128, 64)
+HIGHBD_OBMCSADMXN(64, 128)
+#endif // CONFIG_EXT_PARTITION
+HIGHBD_OBMCSADMXN(64, 64)
+HIGHBD_OBMCSADMXN(64, 32)
+HIGHBD_OBMCSADMXN(32, 64)
+HIGHBD_OBMCSADMXN(32, 32)
+HIGHBD_OBMCSADMXN(32, 16)
+HIGHBD_OBMCSADMXN(16, 32)
+HIGHBD_OBMCSADMXN(16, 16)
+HIGHBD_OBMCSADMXN(16, 8)
+HIGHBD_OBMCSADMXN(8, 16)
+HIGHBD_OBMCSADMXN(8, 8)
+HIGHBD_OBMCSADMXN(8, 4)
+HIGHBD_OBMCSADMXN(4, 8)
+HIGHBD_OBMCSADMXN(4, 4)
+
+#if CONFIG_AV1 && CONFIG_EXT_PARTITION_TYPES
+HIGHBD_OBMCSADMXN(4, 16)
+HIGHBD_OBMCSADMXN(16, 4)
+HIGHBD_OBMCSADMXN(8, 32)
+HIGHBD_OBMCSADMXN(32, 8)
+HIGHBD_OBMCSADMXN(16, 64)
+HIGHBD_OBMCSADMXN(64, 16)
+HIGHBD_OBMCSADMXN(32, 128)
+HIGHBD_OBMCSADMXN(128, 32)
+#endif
+/* clang-format on */
+#endif // CONFIG_HIGHBITDEPTH
+#endif // CONFIG_AV1
