Lowbd SMOOTH_PRED intrapred ssse3 optimization

On i7-6700:
Predictor    ssse3 v. C
4x4          ~1.3x
4x8          ~1.9x
8x4          ~2.3x
8x8          ~3.4x
8x16         ~4.1x
16x8         ~4.6x
16x16        ~5.2x
16x32        ~5.6x
32x16        ~4.2x
32x32        ~4.7x

Change-Id: Ic12383cf9d4446361d6355eb8a480a3c7602060e

aom_dsp/aom_dsp.cmake

@@ -23,6 +23,7 @@ set(AOM_DSP_COMMON_SOURCES
     "${AOM_ROOT}/aom_dsp/blend_a64_mask.c"
     "${AOM_ROOT}/aom_dsp/blend_a64_vmask.c"
     "${AOM_ROOT}/aom_dsp/intrapred.c"
+    "${AOM_ROOT}/aom_dsp/intrapred_common.h"
     "${AOM_ROOT}/aom_dsp/loopfilter.c"
     "${AOM_ROOT}/aom_dsp/prob.c"
     "${AOM_ROOT}/aom_dsp/prob.h"

aom_dsp/aom_dsp.mk

@@ -64,6 +64,7 @@ endif
 
 # intra predictions
 DSP_SRCS-yes += intrapred.c
+DSP_SRCS-yes += intrapred_common.h
 
 ifneq ($(CONFIG_ANS),yes)
 DSP_SRCS-yes += entcode.c

aom_dsp/aom_dsp_rtcd_defs.pl

@@ -147,6 +147,21 @@ specialize qw/aom_paeth_predictor_16x16 ssse3 avx2/;
 specialize qw/aom_paeth_predictor_16x32 ssse3 avx2/;
 specialize qw/aom_paeth_predictor_32x16 ssse3 avx2/;
 specialize qw/aom_paeth_predictor_32x32 ssse3 avx2/;
+specialize qw/aom_paeth_predictor_16x8 ssse3/;
+specialize qw/aom_paeth_predictor_16x16 ssse3/;
+specialize qw/aom_paeth_predictor_16x32 ssse3/;
+specialize qw/aom_paeth_predictor_32x16 ssse3/;
+specialize qw/aom_paeth_predictor_32x32 ssse3/;
+specialize qw/aom_smooth_predictor_4x4 ssse3/;
+specialize qw/aom_smooth_predictor_4x8 ssse3/;
+specialize qw/aom_smooth_predictor_8x4 ssse3/;
+specialize qw/aom_smooth_predictor_8x8 ssse3/;
+specialize qw/aom_smooth_predictor_8x16 ssse3/;
+specialize qw/aom_smooth_predictor_16x8 ssse3/;
+specialize qw/aom_smooth_predictor_16x16 ssse3/;
+specialize qw/aom_smooth_predictor_16x32 ssse3/;
+specialize qw/aom_smooth_predictor_32x16 ssse3/;
+specialize qw/aom_smooth_predictor_32x32 ssse3/;
 
 specialize qw/aom_d63e_predictor_4x4 ssse3/;
 specialize qw/aom_d135_predictor_4x4 neon/;

aom_dsp/intrapred.c

@@ -16,6 +16,7 @@
 #include "./aom_dsp_rtcd.h"
 
 #include "aom_dsp/aom_dsp_common.h"
+#include "aom_dsp/intrapred_common.h"
 #include "aom_mem/aom_mem.h"
 #include "aom_ports/bitops.h"
 
@@ -207,40 +208,6 @@ static INLINE void paeth_predictor(uint8_t *dst, ptrdiff_t stride, int bw,
   }
 }
 
-// Weights are quadratic from '1' to '1 / block_size', scaled by
-// 2^sm_weight_log2_scale.
-static const int sm_weight_log2_scale = 8;
-
-#if CONFIG_TX64X64
-// max(block_size_wide[BLOCK_LARGEST], block_size_high[BLOCK_LARGEST])
-#define MAX_BLOCK_DIM 64
-#else
-#define MAX_BLOCK_DIM 32
-#endif  // CONFIG_TX64X64
-
-static const uint8_t sm_weight_arrays[2 * MAX_BLOCK_DIM] = {
-  // Unused, because we always offset by bs, which is at least 2.
-  0, 0,
-  // bs = 2
-  255, 128,
-  // bs = 4
-  255, 149, 85, 64,
-  // bs = 8
-  255, 197, 146, 105, 73, 50, 37, 32,
-  // bs = 16
-  255, 225, 196, 170, 145, 123, 102, 84, 68, 54, 43, 33, 26, 20, 17, 16,
-  // bs = 32
-  255, 240, 225, 210, 196, 182, 169, 157, 145, 133, 122, 111, 101, 92, 83, 74,
-  66, 59, 52, 45, 39, 34, 29, 25, 21, 17, 14, 12, 10, 9, 8, 8,
-#if CONFIG_TX64X64
-  // bs = 64
-  255, 248, 240, 233, 225, 218, 210, 203, 196, 189, 182, 176, 169, 163, 156,
-  150, 144, 138, 133, 127, 121, 116, 111, 106, 101, 96, 91, 86, 82, 77, 73, 69,
-  65, 61, 57, 54, 50, 47, 44, 41, 38, 35, 32, 29, 27, 25, 22, 20, 18, 16, 15,
-  13, 12, 10, 9, 8, 7, 6, 6, 5, 5, 4, 4, 4,
-#endif  // CONFIG_TX64X64
-};
-
 // Some basic checks on weights for smooth predictor.
 #define sm_weights_sanity_checks(weights_w, weights_h, weights_scale, \
                                  pred_scale)                          \

aom_dsp/intrapred_common.h

+/*
+ * Copyright (c) 2016, 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.
+ */
+
+#ifndef _AOM_DSP_INTRAPRED_COMMON_H
+#define _AOM_DSP_INTRAPRED_COMMON_H
+
+#include "./aom_config.h"
+
+// Weights are quadratic from '1' to '1 / block_size', scaled by
+// 2^sm_weight_log2_scale.
+static const int sm_weight_log2_scale = 8;
+
+#if CONFIG_TX64X64
+// max(block_size_wide[BLOCK_LARGEST], block_size_high[BLOCK_LARGEST])
+#define MAX_BLOCK_DIM 64
+#else
+#define MAX_BLOCK_DIM 32
+#endif  // CONFIG_TX64X64
+
+static const uint8_t sm_weight_arrays[2 * MAX_BLOCK_DIM] = {
+  // Unused, because we always offset by bs, which is at least 2.
+  0, 0,
+  // bs = 2
+  255, 128,
+  // bs = 4
+  255, 149, 85, 64,
+  // bs = 8
+  255, 197, 146, 105, 73, 50, 37, 32,
+  // bs = 16
+  255, 225, 196, 170, 145, 123, 102, 84, 68, 54, 43, 33, 26, 20, 17, 16,
+  // bs = 32
+  255, 240, 225, 210, 196, 182, 169, 157, 145, 133, 122, 111, 101, 92, 83, 74,
+  66, 59, 52, 45, 39, 34, 29, 25, 21, 17, 14, 12, 10, 9, 8, 8,
+#if CONFIG_TX64X64
+  // bs = 64
+  255, 248, 240, 233, 225, 218, 210, 203, 196, 189, 182, 176, 169, 163, 156,
+  150, 144, 138, 133, 127, 121, 116, 111, 106, 101, 96, 91, 86, 82, 77, 73, 69,
+  65, 61, 57, 54, 50, 47, 44, 41, 38, 35, 32, 29, 27, 25, 22, 20, 18, 16, 15,
+  13, 12, 10, 9, 8, 7, 6, 6, 5, 5, 4, 4, 4,
+#endif  // CONFIG_TX64X64
+};
+
+#endif  // _AOM_DSP_INTRAPRED_COMMON_H

test/intrapred_test.cc

@@ -246,7 +246,7 @@ INSTANTIATE_TEST_CASE_P(AVX2, LowbdIntraPredTest,
 
 #if HAVE_SSSE3
 const IntraPredFunc<IntraPred> LowbdIntraPredTestVectorSsse3[] = {
-  lowbd_intrapred(paeth, ssse3),
+  lowbd_intrapred(paeth, ssse3), lowbd_intrapred(smooth, ssse3),
 };
 
 INSTANTIATE_TEST_CASE_P(SSSE3, LowbdIntraPredTest,

test/test_intra_pred_speed.cc

@@ -440,10 +440,12 @@ INTRA_PRED_TEST(SSE2_2, TestIntraPred4, "intra4x8", aom_dc_predictor_4x8_sse2,
 INTRA_PRED_TEST(SSSE3_1, TestIntraPred4, "intra4x4", NULL, NULL, NULL, NULL,
                 NULL, NULL, NULL, NULL, NULL, aom_d153_predictor_4x4_ssse3,
                 NULL, aom_d63e_predictor_4x4_ssse3,
-                aom_paeth_predictor_4x4_ssse3, NULL, NULL, NULL)
+                aom_paeth_predictor_4x4_ssse3, aom_smooth_predictor_4x4_ssse3,
+                NULL, NULL)
 INTRA_PRED_TEST(SSSE3_2, TestIntraPred4, "intra4x8", NULL, NULL, NULL, NULL,
                 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
-                aom_paeth_predictor_4x8_ssse3, NULL, NULL, NULL)
+                aom_paeth_predictor_4x8_ssse3, aom_smooth_predictor_4x8_ssse3,
+                NULL, NULL)
 #endif  // HAVE_SSSE3
 
 #if HAVE_DSPR2
@@ -549,13 +551,16 @@ INTRA_PRED_TEST(SSE2_3, TestIntraPred8, "intra8x16", aom_dc_predictor_8x16_sse2,
 #if HAVE_SSSE3
 INTRA_PRED_TEST(SSSE3_1, TestIntraPred8, "intra8x8", NULL, NULL, NULL, NULL,
                 NULL, NULL, NULL, NULL, NULL, aom_d153_predictor_8x8_ssse3,
-                NULL, NULL, aom_paeth_predictor_8x8_ssse3, NULL, NULL, NULL)
+                NULL, NULL, aom_paeth_predictor_8x8_ssse3,
+                aom_smooth_predictor_8x8_ssse3, NULL, NULL)
 INTRA_PRED_TEST(SSSE3_2, TestIntraPred8, "intra8x4", NULL, NULL, NULL, NULL,
                 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
-                aom_paeth_predictor_8x4_ssse3, NULL, NULL, NULL)
+                aom_paeth_predictor_8x4_ssse3, aom_smooth_predictor_8x4_ssse3,
+                NULL, NULL)
 INTRA_PRED_TEST(SSSE3_3, TestIntraPred8, "intra8x16", NULL, NULL, NULL, NULL,
                 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
-                aom_paeth_predictor_8x16_ssse3, NULL, NULL, NULL)
+                aom_paeth_predictor_8x16_ssse3, aom_smooth_predictor_8x16_ssse3,
+                NULL, NULL)
 #endif  // HAVE_SSSE3
 
 #if HAVE_DSPR2
@@ -663,13 +668,16 @@ INTRA_PRED_TEST(SSE2_3, TestIntraPred16, "intra16x32",
 #if HAVE_SSSE3
 INTRA_PRED_TEST(SSSE3_1, TestIntraPred16, "intra16x16", NULL, NULL, NULL, NULL,
                 NULL, NULL, NULL, NULL, NULL, aom_d153_predictor_16x16_ssse3,
-                NULL, NULL, aom_paeth_predictor_16x16_ssse3, NULL, NULL, NULL)
+                NULL, NULL, aom_paeth_predictor_16x16_ssse3,
+                aom_smooth_predictor_16x16_ssse3, NULL, NULL)
 INTRA_PRED_TEST(SSSE3_2, TestIntraPred16, "intra16x8", NULL, NULL, NULL, NULL,
                 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
-                aom_paeth_predictor_16x8_ssse3, NULL, NULL, NULL)
+                aom_paeth_predictor_16x8_ssse3, aom_smooth_predictor_16x8_ssse3,
+                NULL, NULL)
 INTRA_PRED_TEST(SSSE3_3, TestIntraPred16, "intra16x32", NULL, NULL, NULL, NULL,
                 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
-                aom_paeth_predictor_16x32_ssse3, NULL, NULL, NULL)
+                aom_paeth_predictor_16x32_ssse3,
+                aom_smooth_predictor_16x32_ssse3, NULL, NULL)
 #endif  // HAVE_SSSE3
 
 #if HAVE_AVX2
@@ -767,10 +775,12 @@ INTRA_PRED_TEST(SSE2_2, TestIntraPred32, "intra32x16",
 #if HAVE_SSSE3
 INTRA_PRED_TEST(SSSE3_1, TestIntraPred32, "intra32x32", NULL, NULL, NULL, NULL,
                 NULL, NULL, NULL, NULL, NULL, aom_d153_predictor_32x32_ssse3,
-                NULL, NULL, aom_paeth_predictor_32x32_ssse3, NULL, NULL, NULL)
+                NULL, NULL, aom_paeth_predictor_32x32_ssse3,
+                aom_smooth_predictor_32x32_ssse3, NULL, NULL)
 INTRA_PRED_TEST(SSSE3_2, TestIntraPred32, "intra32x16", NULL, NULL, NULL, NULL,
                 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
-                aom_paeth_predictor_32x16_ssse3, NULL, NULL, NULL)
+                aom_paeth_predictor_32x16_ssse3,
+                aom_smooth_predictor_32x16_ssse3, NULL, NULL)
 #endif  // HAVE_SSSE3
 
 #if HAVE_AVX2