mips dsp-ase r2 vp9 decoder convolve module optimizations

Change-Id: I401536778e3c68ba2b3ae3955c689d005e1f1d59

build/make/rtcd.sh

@@ -294,6 +294,10 @@ $(set_function_pointers c $ALL_ARCHS)
 void dsputil_static_init();
 dsputil_static_init();
 #endif
+#if CONFIG_VP9
+void vp9_dsputil_static_init();
+vp9_dsputil_static_init();
+#endif
 #endif
 }
 #endif

test/convolve_test.cc

@@ -642,4 +642,26 @@ INSTANTIATE_TEST_CASE_P(NEON, ConvolveTest, ::testing::Values(
     make_tuple(32, 64, &convolve8_neon),
     make_tuple(64, 64, &convolve8_neon)));
 #endif
+
+#if HAVE_DSPR2
+const ConvolveFunctions convolve8_dspr2(
+    vp9_convolve8_horiz_dspr2, vp9_convolve8_avg_horiz_dspr2,
+    vp9_convolve8_vert_dspr2, vp9_convolve8_avg_vert_dspr2,
+    vp9_convolve8_dspr2, vp9_convolve8_avg_dspr2);
+
+INSTANTIATE_TEST_CASE_P(DSPR2, ConvolveTest, ::testing::Values(
+    make_tuple(4, 4, &convolve8_dspr2),
+    make_tuple(8, 4, &convolve8_dspr2),
+    make_tuple(4, 8, &convolve8_dspr2),
+    make_tuple(8, 8, &convolve8_dspr2),
+    make_tuple(16, 8, &convolve8_dspr2),
+    make_tuple(8, 16, &convolve8_dspr2),
+    make_tuple(16, 16, &convolve8_dspr2),
+    make_tuple(32, 16, &convolve8_dspr2),
+    make_tuple(16, 32, &convolve8_dspr2),
+    make_tuple(32, 32, &convolve8_dspr2),
+    make_tuple(64, 32, &convolve8_dspr2),
+    make_tuple(32, 64, &convolve8_dspr2),
+    make_tuple(64, 64, &convolve8_dspr2)));
+#endif
 }  // namespace

vp9/common/mips/dspr2/vp9_common_dspr2.h

+/*
+ *  Copyright (c) 2013 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.
+ */
+
+#ifndef VP9_COMMON_VP9_COMMON_DSPR2_H_
+#define VP9_COMMON_VP9_COMMON_DSPR2_H_
+
+#include <assert.h>
+
+#include "./vpx_config.h"
+#include "vpx/vpx_integer.h"
+#include "vp9/common/vp9_common.h"
+
+#if HAVE_DSPR2
+#define CROP_WIDTH 512
+extern uint8_t *vp9_ff_cropTbl;
+
+#define DCT_CONST_ROUND_SHIFT_TWICE_COSPI_16_64(input)                    ({   \
+                                                                               \
+  int32_t tmp, out;                                                            \
+  int     dct_cost_rounding = DCT_CONST_ROUNDING;                              \
+  int     in = input;                                                          \
+                                                                               \
+  __asm__ __volatile__ (                                                       \
+      /* out = dct_const_round_shift(input_dc * cospi_16_64); */               \
+      "mtlo     %[dct_cost_rounding],   $ac1                              \n\t"\
+      "mthi     $zero,                  $ac1                              \n\t"\
+      "madd     $ac1,                   %[in],            %[cospi_16_64]  \n\t"\
+      "extp     %[tmp],                 $ac1,             31              \n\t"\
+                                                                               \
+      /* out = dct_const_round_shift(out * cospi_16_64); */                    \
+      "mtlo     %[dct_cost_rounding],   $ac2                              \n\t"\
+      "mthi     $zero,                  $ac2                              \n\t"\
+      "madd     $ac2,                   %[tmp],           %[cospi_16_64]  \n\t"\
+      "extp     %[out],                 $ac2,             31              \n\t"\
+                                                                               \
+      : [tmp] "=&r" (tmp), [out] "=r" (out)                                    \
+      : [in] "r" (in),                                                         \
+        [dct_cost_rounding] "r" (dct_cost_rounding),                           \
+        [cospi_16_64] "r" (cospi_16_64)                                        \
+   );                                                                          \
+  out;                                                                    })
+
+static INLINE void vp9_prefetch_load(const unsigned char *src) {
+  __asm__ __volatile__ (
+      "pref   0,  0(%[src])   \n\t"
+      :
+      : [src] "r" (src)
+  );
+}
+
+/* prefetch data for store */
+static INLINE void vp9_prefetch_store(unsigned char *dst) {
+  __asm__ __volatile__ (
+      "pref   1,  0(%[dst])   \n\t"
+      :
+      : [dst] "r" (dst)
+  );
+}
+
+static INLINE void vp9_prefetch_load_streamed(const unsigned char *src) {
+  __asm__ __volatile__ (
+      "pref   4,  0(%[src])   \n\t"
+      :
+      : [src] "r" (src)
+  );
+}
+
+/* prefetch data for store */
+static INLINE void vp9_prefetch_store_streamed(unsigned char *dst) {
+  __asm__ __volatile__ (
+      "pref   5,  0(%[dst])   \n\t"
+      :
+      : [dst] "r" (dst)
+  );
+}
+
+#endif  // #if HAVE_DSPR2
+#endif  // VP9_COMMON_VP9_COMMON_DSPR2_H_

vp9/common/vp9_rtcd_defs.sh

@@ -257,28 +257,28 @@ specialize vp9_blend_b
 # Sub Pixel Filters
 #
 prototype void vp9_convolve_copy "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h"
-specialize vp9_convolve_copy $sse2_x86inc neon
+specialize vp9_convolve_copy $sse2_x86inc neon dspr2
 
 prototype void vp9_convolve_avg "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h"
-specialize vp9_convolve_avg $sse2_x86inc neon
+specialize vp9_convolve_avg $sse2_x86inc neon dspr2
 
 prototype void vp9_convolve8 "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h"
-specialize vp9_convolve8 ssse3 neon
+specialize vp9_convolve8 ssse3 neon dspr2
 
 prototype void vp9_convolve8_horiz "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h"
-specialize vp9_convolve8_horiz ssse3 neon
+specialize vp9_convolve8_horiz ssse3 neon dspr2
 
 prototype void vp9_convolve8_vert "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h"
-specialize vp9_convolve8_vert ssse3 neon
+specialize vp9_convolve8_vert ssse3 neon dspr2
 
 prototype void vp9_convolve8_avg "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h"
-specialize vp9_convolve8_avg ssse3 neon
+specialize vp9_convolve8_avg ssse3 neon dspr2
 
 prototype void vp9_convolve8_avg_horiz "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h"
-specialize vp9_convolve8_avg_horiz ssse3 neon
+specialize vp9_convolve8_avg_horiz ssse3 neon dspr2
 
 prototype void vp9_convolve8_avg_vert "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h"
-specialize vp9_convolve8_avg_vert ssse3 neon
+specialize vp9_convolve8_avg_vert ssse3 neon dspr2
 
 #
 # dct

vp9/vp9_common.mk

@@ -87,6 +87,14 @@ VP9_COMMON_SRCS-$(HAVE_SSE2) += common/x86/vp9_intrapred_sse2.asm
 VP9_COMMON_SRCS-$(HAVE_SSSE3) += common/x86/vp9_intrapred_ssse3.asm
 endif
 
+# common (c)
+VP9_COMMON_SRCS-$(HAVE_DSPR2)  += common/mips/dspr2/vp9_common_dspr2.h
+VP9_COMMON_SRCS-$(HAVE_DSPR2)  += common/mips/dspr2/vp9_convolve8_avg_dspr2.c
+VP9_COMMON_SRCS-$(HAVE_DSPR2)  += common/mips/dspr2/vp9_convolve8_avg_horiz_dspr2.c
+VP9_COMMON_SRCS-$(HAVE_DSPR2)  += common/mips/dspr2/vp9_convolve8_dspr2.c
+VP9_COMMON_SRCS-$(HAVE_DSPR2)  += common/mips/dspr2/vp9_convolve8_horiz_dspr2.c
+VP9_COMMON_SRCS-$(HAVE_DSPR2)  += common/mips/dspr2/vp9_convolve8_vert_dspr2.c
+
 VP9_COMMON_SRCS-$(HAVE_SSE2) += common/x86/vp9_idct_intrin_sse2.c
 
 VP9_COMMON_SRCS-$(HAVE_NEON) += common/arm/neon/vp9_convolve_neon.c