Merge "Removing inv_txm4x4_1_add and inv_txm4x4_add function pointers."
diff --git a/test/dct16x16_test.cc b/test/dct16x16_test.cc
index 0d19aa0..028f8ff 100644
--- a/test/dct16x16_test.cc
+++ b/test/dct16x16_test.cc
@@ -21,7 +21,7 @@
extern "C" {
#include "vp9/common/vp9_entropy.h"
#include "./vp9_rtcd.h"
-void vp9_short_idct16x16_add_c(int16_t *input, uint8_t *output, int pitch);
+void vp9_idct16x16_256_add_c(int16_t *input, uint8_t *output, int pitch);
}
#include "vpx/vpx_integer.h"
@@ -496,7 +496,7 @@
INSTANTIATE_TEST_CASE_P(
C, Trans16x16DCT,
::testing::Values(
- make_tuple(&vp9_short_fdct16x16_c, &vp9_short_idct16x16_add_c, 0)));
+ make_tuple(&vp9_short_fdct16x16_c, &vp9_idct16x16_256_add_c, 0)));
INSTANTIATE_TEST_CASE_P(
C, Trans16x16HT,
::testing::Values(
@@ -510,7 +510,7 @@
SSE2, Trans16x16DCT,
::testing::Values(
make_tuple(&vp9_short_fdct16x16_sse2,
- &vp9_short_idct16x16_add_sse2, 0)));
+ &vp9_idct16x16_256_add_sse2, 0)));
INSTANTIATE_TEST_CASE_P(
SSE2, Trans16x16HT,
::testing::Values(
diff --git a/vp9/common/arm/neon/vp9_idct16x16_neon.c b/vp9/common/arm/neon/vp9_idct16x16_neon.c
index fb7b5cd..33aa4e0 100644
--- a/vp9/common/arm/neon/vp9_idct16x16_neon.c
+++ b/vp9/common/arm/neon/vp9_idct16x16_neon.c
@@ -11,19 +11,19 @@
#include "./vp9_rtcd.h"
#include "vp9/common/vp9_common.h"
-extern void vp9_short_idct16x16_add_neon_pass1(int16_t *input,
+extern void vp9_idct16x16_256_add_neon_pass1(int16_t *input,
int16_t *output,
int output_stride);
-extern void vp9_short_idct16x16_add_neon_pass2(int16_t *src,
+extern void vp9_idct16x16_256_add_neon_pass2(int16_t *src,
int16_t *output,
int16_t *pass1Output,
int16_t skip_adding,
uint8_t *dest,
int dest_stride);
-extern void vp9_short_idct16x16_10_add_neon_pass1(int16_t *input,
+extern void vp9_idct16x16_10_add_neon_pass1(int16_t *input,
int16_t *output,
int output_stride);
-extern void vp9_short_idct16x16_10_add_neon_pass2(int16_t *src,
+extern void vp9_idct16x16_10_add_neon_pass2(int16_t *src,
int16_t *output,
int16_t *pass1Output,
int16_t skip_adding,
@@ -34,7 +34,7 @@
extern void vp9_push_neon(int64_t *store);
extern void vp9_pop_neon(int64_t *store);
-void vp9_short_idct16x16_add_neon(int16_t *input,
+void vp9_idct16x16_256_add_neon(int16_t *input,
uint8_t *dest, int dest_stride) {
int64_t store_reg[8];
int16_t pass1_output[16*16] = {0};
@@ -46,12 +46,12 @@
/* Parallel idct on the upper 8 rows */
// First pass processes even elements 0, 2, 4, 6, 8, 10, 12, 14 and save the
// stage 6 result in pass1_output.
- vp9_short_idct16x16_add_neon_pass1(input, pass1_output, 8);
+ vp9_idct16x16_256_add_neon_pass1(input, pass1_output, 8);
// Second pass processes odd elements 1, 3, 5, 7, 9, 11, 13, 15 and combines
// with result in pass1(pass1_output) to calculate final result in stage 7
// which will be saved into row_idct_output.
- vp9_short_idct16x16_add_neon_pass2(input+1,
+ vp9_idct16x16_256_add_neon_pass2(input+1,
row_idct_output,
pass1_output,
0,
@@ -61,12 +61,12 @@
/* Parallel idct on the lower 8 rows */
// First pass processes even elements 0, 2, 4, 6, 8, 10, 12, 14 and save the
// stage 6 result in pass1_output.
- vp9_short_idct16x16_add_neon_pass1(input+8*16, pass1_output, 8);
+ vp9_idct16x16_256_add_neon_pass1(input+8*16, pass1_output, 8);
// Second pass processes odd elements 1, 3, 5, 7, 9, 11, 13, 15 and combines
// with result in pass1(pass1_output) to calculate final result in stage 7
// which will be saved into row_idct_output.
- vp9_short_idct16x16_add_neon_pass2(input+8*16+1,
+ vp9_idct16x16_256_add_neon_pass2(input+8*16+1,
row_idct_output+8,
pass1_output,
0,
@@ -76,12 +76,12 @@
/* Parallel idct on the left 8 columns */
// First pass processes even elements 0, 2, 4, 6, 8, 10, 12, 14 and save the
// stage 6 result in pass1_output.
- vp9_short_idct16x16_add_neon_pass1(row_idct_output, pass1_output, 8);
+ vp9_idct16x16_256_add_neon_pass1(row_idct_output, pass1_output, 8);
// Second pass processes odd elements 1, 3, 5, 7, 9, 11, 13, 15 and combines
// with result in pass1(pass1_output) to calculate final result in stage 7.
// Then add the result to the destination data.
- vp9_short_idct16x16_add_neon_pass2(row_idct_output+1,
+ vp9_idct16x16_256_add_neon_pass2(row_idct_output+1,
row_idct_output,
pass1_output,
1,
@@ -91,12 +91,12 @@
/* Parallel idct on the right 8 columns */
// First pass processes even elements 0, 2, 4, 6, 8, 10, 12, 14 and save the
// stage 6 result in pass1_output.
- vp9_short_idct16x16_add_neon_pass1(row_idct_output+8*16, pass1_output, 8);
+ vp9_idct16x16_256_add_neon_pass1(row_idct_output+8*16, pass1_output, 8);
// Second pass processes odd elements 1, 3, 5, 7, 9, 11, 13, 15 and combines
// with result in pass1(pass1_output) to calculate final result in stage 7.
// Then add the result to the destination data.
- vp9_short_idct16x16_add_neon_pass2(row_idct_output+8*16+1,
+ vp9_idct16x16_256_add_neon_pass2(row_idct_output+8*16+1,
row_idct_output+8,
pass1_output,
1,
@@ -109,7 +109,7 @@
return;
}
-void vp9_short_idct16x16_10_add_neon(int16_t *input,
+void vp9_idct16x16_10_add_neon(int16_t *input,
uint8_t *dest, int dest_stride) {
int64_t store_reg[8];
int16_t pass1_output[16*16] = {0};
@@ -121,12 +121,12 @@
/* Parallel idct on the upper 8 rows */
// First pass processes even elements 0, 2, 4, 6, 8, 10, 12, 14 and save the
// stage 6 result in pass1_output.
- vp9_short_idct16x16_10_add_neon_pass1(input, pass1_output, 8);
+ vp9_idct16x16_10_add_neon_pass1(input, pass1_output, 8);
// Second pass processes odd elements 1, 3, 5, 7, 9, 11, 13, 15 and combines
// with result in pass1(pass1_output) to calculate final result in stage 7
// which will be saved into row_idct_output.
- vp9_short_idct16x16_10_add_neon_pass2(input+1,
+ vp9_idct16x16_10_add_neon_pass2(input+1,
row_idct_output,
pass1_output,
0,
@@ -138,12 +138,12 @@
/* Parallel idct on the left 8 columns */
// First pass processes even elements 0, 2, 4, 6, 8, 10, 12, 14 and save the
// stage 6 result in pass1_output.
- vp9_short_idct16x16_add_neon_pass1(row_idct_output, pass1_output, 8);
+ vp9_idct16x16_256_add_neon_pass1(row_idct_output, pass1_output, 8);
// Second pass processes odd elements 1, 3, 5, 7, 9, 11, 13, 15 and combines
// with result in pass1(pass1_output) to calculate final result in stage 7.
// Then add the result to the destination data.
- vp9_short_idct16x16_add_neon_pass2(row_idct_output+1,
+ vp9_idct16x16_256_add_neon_pass2(row_idct_output+1,
row_idct_output,
pass1_output,
1,
@@ -153,12 +153,12 @@
/* Parallel idct on the right 8 columns */
// First pass processes even elements 0, 2, 4, 6, 8, 10, 12, 14 and save the
// stage 6 result in pass1_output.
- vp9_short_idct16x16_add_neon_pass1(row_idct_output+8*16, pass1_output, 8);
+ vp9_idct16x16_256_add_neon_pass1(row_idct_output+8*16, pass1_output, 8);
// Second pass processes odd elements 1, 3, 5, 7, 9, 11, 13, 15 and combines
// with result in pass1(pass1_output) to calculate final result in stage 7.
// Then add the result to the destination data.
- vp9_short_idct16x16_add_neon_pass2(row_idct_output+8*16+1,
+ vp9_idct16x16_256_add_neon_pass2(row_idct_output+8*16+1,
row_idct_output+8,
pass1_output,
1,
diff --git a/vp9/common/arm/neon/vp9_short_idct16x16_1_add_neon.asm b/vp9/common/arm/neon/vp9_short_idct16x16_1_add_neon.asm
index cf5c8f7..b1fd21b 100644
--- a/vp9/common/arm/neon/vp9_short_idct16x16_1_add_neon.asm
+++ b/vp9/common/arm/neon/vp9_short_idct16x16_1_add_neon.asm
@@ -8,21 +8,21 @@
;
- EXPORT |vp9_short_idct16x16_1_add_neon|
+ EXPORT |vp9_idct16x16_1_add_neon|
ARM
REQUIRE8
PRESERVE8
AREA ||.text||, CODE, READONLY, ALIGN=2
-;void vp9_short_idct16x16_1_add_neon(int16_t *input, uint8_t *dest,
+;void vp9_idct16x16_1_add_neon(int16_t *input, uint8_t *dest,
; int dest_stride)
;
; r0 int16_t input
; r1 uint8_t *dest
; r2 int dest_stride)
-|vp9_short_idct16x16_1_add_neon| PROC
+|vp9_idct16x16_1_add_neon| PROC
ldrsh r0, [r0]
; generate cospi_16_64 = 11585
@@ -193,6 +193,6 @@
vst1.64 {d31}, [r12], r2
bx lr
- ENDP ; |vp9_short_idct16x16_1_add_neon|
+ ENDP ; |vp9_idct16x16_1_add_neon|
END
diff --git a/vp9/common/arm/neon/vp9_short_idct16x16_add_neon.asm b/vp9/common/arm/neon/vp9_short_idct16x16_add_neon.asm
index df2a052..a13c0d0 100644
--- a/vp9/common/arm/neon/vp9_short_idct16x16_add_neon.asm
+++ b/vp9/common/arm/neon/vp9_short_idct16x16_add_neon.asm
@@ -8,10 +8,10 @@
; be found in the AUTHORS file in the root of the source tree.
;
- EXPORT |vp9_short_idct16x16_add_neon_pass1|
- EXPORT |vp9_short_idct16x16_add_neon_pass2|
- EXPORT |vp9_short_idct16x16_10_add_neon_pass1|
- EXPORT |vp9_short_idct16x16_10_add_neon_pass2|
+ EXPORT |vp9_idct16x16_256_add_neon_pass1|
+ EXPORT |vp9_idct16x16_256_add_neon_pass2|
+ EXPORT |vp9_idct16x16_10_add_neon_pass1|
+ EXPORT |vp9_idct16x16_10_add_neon_pass2|
ARM
REQUIRE8
PRESERVE8
@@ -36,7 +36,7 @@
MEND
AREA Block, CODE, READONLY ; name this block of code
-;void |vp9_short_idct16x16_add_neon_pass1|(int16_t *input,
+;void |vp9_idct16x16_256_add_neon_pass1|(int16_t *input,
; int16_t *output, int output_stride)
;
; r0 int16_t input
@@ -46,7 +46,7 @@
; idct16 stage1 - stage6 on all the elements loaded in q8-q15. The output
; will be stored back into q8-q15 registers. This function will touch q0-q7
; registers and use them as buffer during calculation.
-|vp9_short_idct16x16_add_neon_pass1| PROC
+|vp9_idct16x16_256_add_neon_pass1| PROC
; TODO(hkuang): Find a better way to load the elements.
; load elements of 0, 2, 4, 6, 8, 10, 12, 14 into q8 - q15
@@ -273,9 +273,9 @@
vst1.64 {d31}, [r1], r2
bx lr
- ENDP ; |vp9_short_idct16x16_add_neon_pass1|
+ ENDP ; |vp9_idct16x16_256_add_neon_pass1|
-;void vp9_short_idct16x16_add_neon_pass2(int16_t *src,
+;void vp9_idct16x16_256_add_neon_pass2(int16_t *src,
; int16_t *output,
; int16_t *pass1Output,
; int16_t skip_adding,
@@ -292,7 +292,7 @@
; idct16 stage1 - stage7 on all the elements loaded in q8-q15. The output
; will be stored back into q8-q15 registers. This function will touch q0-q7
; registers and use them as buffer during calculation.
-|vp9_short_idct16x16_add_neon_pass2| PROC
+|vp9_idct16x16_256_add_neon_pass2| PROC
push {r3-r9}
; TODO(hkuang): Find a better way to load the elements.
@@ -784,9 +784,9 @@
end_idct16x16_pass2
pop {r3-r9}
bx lr
- ENDP ; |vp9_short_idct16x16_add_neon_pass2|
+ ENDP ; |vp9_idct16x16_256_add_neon_pass2|
-;void |vp9_short_idct16x16_10_add_neon_pass1|(int16_t *input,
+;void |vp9_idct16x16_10_add_neon_pass1|(int16_t *input,
; int16_t *output, int output_stride)
;
; r0 int16_t input
@@ -796,7 +796,7 @@
; idct16 stage1 - stage6 on all the elements loaded in q8-q15. The output
; will be stored back into q8-q15 registers. This function will touch q0-q7
; registers and use them as buffer during calculation.
-|vp9_short_idct16x16_10_add_neon_pass1| PROC
+|vp9_idct16x16_10_add_neon_pass1| PROC
; TODO(hkuang): Find a better way to load the elements.
; load elements of 0, 2, 4, 6, 8, 10, 12, 14 into q8 - q15
@@ -905,9 +905,9 @@
vst1.64 {d31}, [r1], r2
bx lr
- ENDP ; |vp9_short_idct16x16_10_add_neon_pass1|
+ ENDP ; |vp9_idct16x16_10_add_neon_pass1|
-;void vp9_short_idct16x16_10_add_neon_pass2(int16_t *src,
+;void vp9_idct16x16_10_add_neon_pass2(int16_t *src,
; int16_t *output,
; int16_t *pass1Output,
; int16_t skip_adding,
@@ -924,7 +924,7 @@
; idct16 stage1 - stage7 on all the elements loaded in q8-q15. The output
; will be stored back into q8-q15 registers. This function will touch q0-q7
; registers and use them as buffer during calculation.
-|vp9_short_idct16x16_10_add_neon_pass2| PROC
+|vp9_idct16x16_10_add_neon_pass2| PROC
push {r3-r9}
; TODO(hkuang): Find a better way to load the elements.
@@ -1175,5 +1175,5 @@
end_idct10_16x16_pass2
pop {r3-r9}
bx lr
- ENDP ; |vp9_short_idct16x16_10_add_neon_pass2|
+ ENDP ; |vp9_idct16x16_10_add_neon_pass2|
END
diff --git a/vp9/common/mips/dspr2/vp9_common_dspr2.h b/vp9/common/mips/dspr2/vp9_common_dspr2.h
index d2fa4c1..dc88f16 100644
--- a/vp9/common/mips/dspr2/vp9_common_dspr2.h
+++ b/vp9/common/mips/dspr2/vp9_common_dspr2.h
@@ -81,5 +81,34 @@
);
}
+void vp9_convolve2_horiz_dspr2(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);
+
+void vp9_convolve2_avg_horiz_dspr2(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);
+
+void vp9_convolve2_avg_vert_dspr2(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);
+
+void vp9_convolve2_dspr2(const uint8_t *src, ptrdiff_t src_stride,
+ uint8_t *dst, ptrdiff_t dst_stride,
+ const int16_t *filter,
+ int w, int h);
+
+void vp9_convolve2_vert_dspr2(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);
+
#endif // #if HAVE_DSPR2
#endif // VP9_COMMON_VP9_COMMON_DSPR2_H_
diff --git a/vp9/common/mips/dspr2/vp9_convolve2_avg_dspr2.c b/vp9/common/mips/dspr2/vp9_convolve2_avg_dspr2.c
new file mode 100644
index 0000000..91d62bc
--- /dev/null
+++ b/vp9/common/mips/dspr2/vp9_convolve2_avg_dspr2.c
@@ -0,0 +1,281 @@
+/*
+ * 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.
+ */
+
+#include <assert.h>
+#include <stdio.h>
+
+#include "./vpx_config.h"
+#include "./vp9_rtcd.h"
+#include "vp9/common/vp9_common.h"
+#include "vpx/vpx_integer.h"
+#include "vpx_ports/mem.h"
+#include "vp9/common/vp9_convolve.h"
+#include "vp9/common/mips/dspr2/vp9_common_dspr2.h"
+
+#if HAVE_DSPR2
+static void convolve_bi_avg_vert_4_dspr2(const uint8_t *src,
+ int32_t src_stride,
+ uint8_t *dst,
+ int32_t dst_stride,
+ const int16_t *filter_y,
+ int32_t w,
+ int32_t h) {
+ int32_t x, y;
+ const uint8_t *src_ptr;
+ uint8_t *dst_ptr;
+ uint8_t *cm = vp9_ff_cropTbl;
+ uint32_t vector4a = 64;
+ uint32_t load1, load2;
+ uint32_t p1, p2;
+ uint32_t scratch1, scratch2;
+ uint32_t store1, store2;
+ int32_t Temp1, Temp2;
+ const int16_t *filter = &filter_y[3];
+ uint32_t filter45;
+
+ filter45 = ((const int32_t *)filter)[0];
+
+ for (y = h; y--;) {
+ /* prefetch data to cache memory */
+ vp9_prefetch_store(dst + dst_stride);
+
+ for (x = 0; x < w; x += 4) {
+ src_ptr = src + x;
+ dst_ptr = dst + x;
+
+ __asm__ __volatile__ (
+ "ulw %[load1], 0(%[src_ptr]) \n\t"
+ "add %[src_ptr], %[src_ptr], %[src_stride] \n\t"
+ "ulw %[load2], 0(%[src_ptr]) \n\t"
+
+ "mtlo %[vector4a], $ac0 \n\t"
+ "mtlo %[vector4a], $ac1 \n\t"
+ "mtlo %[vector4a], $ac2 \n\t"
+ "mtlo %[vector4a], $ac3 \n\t"
+ "mthi $zero, $ac0 \n\t"
+ "mthi $zero, $ac1 \n\t"
+ "mthi $zero, $ac2 \n\t"
+ "mthi $zero, $ac3 \n\t"
+
+ "preceu.ph.qbr %[scratch1], %[load1] \n\t"
+ "preceu.ph.qbr %[p1], %[load2] \n\t"
+ "precrq.ph.w %[p2], %[p1], %[scratch1] \n\t" /* pixel 2 */
+ "append %[p1], %[scratch1], 16 \n\t" /* pixel 1 */
+
+ "dpa.w.ph $ac0, %[p1], %[filter45] \n\t"
+ "dpa.w.ph $ac1, %[p2], %[filter45] \n\t"
+
+ "preceu.ph.qbl %[scratch1], %[load1] \n\t"
+ "preceu.ph.qbl %[p1], %[load2] \n\t"
+ "precrq.ph.w %[p2], %[p1], %[scratch1] \n\t" /* pixel 2 */
+ "append %[p1], %[scratch1], 16 \n\t" /* pixel 1 */
+
+ "dpa.w.ph $ac2, %[p1], %[filter45] \n\t"
+ "dpa.w.ph $ac3, %[p2], %[filter45] \n\t"
+
+ "extp %[Temp1], $ac0, 31 \n\t"
+ "extp %[Temp2], $ac1, 31 \n\t"
+
+ "lbu %[scratch1], 0(%[dst_ptr]) \n\t"
+ "lbu %[scratch2], 1(%[dst_ptr]) \n\t"
+
+ "lbux %[store1], %[Temp1](%[cm]) \n\t"
+ "addqh_r.w %[store1], %[store1], %[scratch1] \n\t" /* pixel 1 */
+ "extp %[Temp1], $ac2, 31 \n\t"
+
+ "lbux %[store2], %[Temp2](%[cm]) \n\t"
+ "addqh_r.w %[store2], %[store2], %[scratch2] \n\t" /* pixel 2 */
+ "extp %[Temp2], $ac3, 31 \n\t"
+ "lbu %[scratch1], 2(%[dst_ptr]) \n\t"
+
+ "sb %[store1], 0(%[dst_ptr]) \n\t"
+ "sb %[store2], 1(%[dst_ptr]) \n\t"
+ "lbu %[scratch2], 3(%[dst_ptr]) \n\t"
+
+ "lbux %[store1], %[Temp1](%[cm]) \n\t"
+ "lbux %[store2], %[Temp2](%[cm]) \n\t"
+ "addqh_r.w %[store1], %[store1], %[scratch1] \n\t" /* pixel 3 */
+ "addqh_r.w %[store2], %[store2], %[scratch2] \n\t" /* pixel 4 */
+
+ "sb %[store1], 2(%[dst_ptr]) \n\t"
+ "sb %[store2], 3(%[dst_ptr]) \n\t"
+
+ : [load1] "=&r" (load1), [load2] "=&r" (load2),
+ [p1] "=&r" (p1), [p2] "=&r" (p2),
+ [scratch1] "=&r" (scratch1), [scratch2] "=&r" (scratch2),
+ [Temp1] "=&r" (Temp1), [Temp2] "=&r" (Temp2),
+ [store1] "=&r" (store1), [store2] "=&r" (store2),
+ [src_ptr] "+r" (src_ptr)
+ : [filter45] "r" (filter45), [vector4a] "r" (vector4a),
+ [src_stride] "r" (src_stride), [cm] "r" (cm),
+ [dst_ptr] "r" (dst_ptr)
+ );
+ }
+
+ /* Next row... */
+ src += src_stride;
+ dst += dst_stride;
+ }
+}
+
+static void convolve_bi_avg_vert_64_dspr2(const uint8_t *src,
+ int32_t src_stride,
+ uint8_t *dst,
+ int32_t dst_stride,
+ const int16_t *filter_y,
+ int32_t h) {
+ int32_t x, y;
+ const uint8_t *src_ptr;
+ uint8_t *dst_ptr;
+ uint8_t *cm = vp9_ff_cropTbl;
+ uint32_t vector4a = 64;
+ uint32_t load1, load2;
+ uint32_t p1, p2;
+ uint32_t scratch1, scratch2;
+ uint32_t store1, store2;
+ int32_t Temp1, Temp2;
+ const int16_t *filter = &filter_y[3];
+ uint32_t filter45;;
+
+ filter45 = ((const int32_t *)filter)[0];
+
+ for (y = h; y--;) {
+ /* prefetch data to cache memory */
+ vp9_prefetch_store(dst + dst_stride);
+ vp9_prefetch_store(dst + dst_stride + 32);
+
+ for (x = 0; x < 64; x += 4) {
+ src_ptr = src + x;
+ dst_ptr = dst + x;
+
+ __asm__ __volatile__ (
+ "ulw %[load1], 0(%[src_ptr]) \n\t"
+ "add %[src_ptr], %[src_ptr], %[src_stride] \n\t"
+ "ulw %[load2], 0(%[src_ptr]) \n\t"
+
+ "mtlo %[vector4a], $ac0 \n\t"
+ "mtlo %[vector4a], $ac1 \n\t"
+ "mtlo %[vector4a], $ac2 \n\t"
+ "mtlo %[vector4a], $ac3 \n\t"
+ "mthi $zero, $ac0 \n\t"
+ "mthi $zero, $ac1 \n\t"
+ "mthi $zero, $ac2 \n\t"
+ "mthi $zero, $ac3 \n\t"
+
+ "preceu.ph.qbr %[scratch1], %[load1] \n\t"
+ "preceu.ph.qbr %[p1], %[load2] \n\t"
+ "precrq.ph.w %[p2], %[p1], %[scratch1] \n\t" /* pixel 2 */
+ "append %[p1], %[scratch1], 16 \n\t" /* pixel 1 */
+
+ "dpa.w.ph $ac0, %[p1], %[filter45] \n\t"
+ "dpa.w.ph $ac1, %[p2], %[filter45] \n\t"
+
+ "preceu.ph.qbl %[scratch1], %[load1] \n\t"
+ "preceu.ph.qbl %[p1], %[load2] \n\t"
+ "precrq.ph.w %[p2], %[p1], %[scratch1] \n\t" /* pixel 2 */
+ "append %[p1], %[scratch1], 16 \n\t" /* pixel 1 */
+
+ "dpa.w.ph $ac2, %[p1], %[filter45] \n\t"
+ "dpa.w.ph $ac3, %[p2], %[filter45] \n\t"
+
+ "extp %[Temp1], $ac0, 31 \n\t"
+ "extp %[Temp2], $ac1, 31 \n\t"
+
+ "lbu %[scratch1], 0(%[dst_ptr]) \n\t"
+ "lbu %[scratch2], 1(%[dst_ptr]) \n\t"
+
+ "lbux %[store1], %[Temp1](%[cm]) \n\t"
+ "addqh_r.w %[store1], %[store1], %[scratch1] \n\t" /* pixel 1 */
+ "extp %[Temp1], $ac2, 31 \n\t"
+
+ "lbux %[store2], %[Temp2](%[cm]) \n\t"
+ "addqh_r.w %[store2], %[store2], %[scratch2] \n\t" /* pixel 2 */
+ "extp %[Temp2], $ac3, 31 \n\t"
+ "lbu %[scratch1], 2(%[dst_ptr]) \n\t"
+
+ "sb %[store1], 0(%[dst_ptr]) \n\t"
+ "sb %[store2], 1(%[dst_ptr]) \n\t"
+ "lbu %[scratch2], 3(%[dst_ptr]) \n\t"
+
+ "lbux %[store1], %[Temp1](%[cm]) \n\t"
+ "lbux %[store2], %[Temp2](%[cm]) \n\t"
+ "addqh_r.w %[store1], %[store1], %[scratch1] \n\t" /* pixel 3 */
+ "addqh_r.w %[store2], %[store2], %[scratch2] \n\t" /* pixel 4 */
+
+ "sb %[store1], 2(%[dst_ptr]) \n\t"
+ "sb %[store2], 3(%[dst_ptr]) \n\t"
+
+ : [load1] "=&r" (load1), [load2] "=&r" (load2),
+ [p1] "=&r" (p1), [p2] "=&r" (p2),
+ [scratch1] "=&r" (scratch1), [scratch2] "=&r" (scratch2),
+ [Temp1] "=&r" (Temp1), [Temp2] "=&r" (Temp2),
+ [store1] "=&r" (store1), [store2] "=&r" (store2),
+ [src_ptr] "+r" (src_ptr)
+ : [filter45] "r" (filter45), [vector4a] "r" (vector4a),
+ [src_stride] "r" (src_stride), [cm] "r" (cm),
+ [dst_ptr] "r" (dst_ptr)
+ );
+ }
+
+ /* Next row... */
+ src += src_stride;
+ dst += dst_stride;
+ }
+}
+
+void vp9_convolve2_avg_vert_dspr2(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) {
+ if (16 == y_step_q4) {
+ uint32_t pos = 38;
+
+ /* bit positon for extract from acc */
+ __asm__ __volatile__ (
+ "wrdsp %[pos], 1 \n\t"
+ :
+ : [pos] "r" (pos)
+ );
+
+ vp9_prefetch_store(dst);
+
+ switch (w) {
+ case 4:
+ case 8:
+ case 16:
+ case 32:
+ convolve_bi_avg_vert_4_dspr2(src, src_stride,
+ dst, dst_stride,
+ filter_y, w, h);
+ break;
+ case 64:
+ vp9_prefetch_store(dst + 32);
+ convolve_bi_avg_vert_64_dspr2(src, src_stride,
+ dst, dst_stride,
+ filter_y, h);
+ break;
+ default:
+ vp9_convolve8_avg_vert_c(src, src_stride,
+ dst, dst_stride,
+ filter_x, x_step_q4,
+ filter_y, y_step_q4,
+ w, h);
+ break;
+ }
+ } else {
+ vp9_convolve8_avg_vert_c(src, src_stride,
+ dst, dst_stride,
+ filter_x, x_step_q4,
+ filter_y, y_step_q4,
+ w, h);
+ }
+}
+#endif
diff --git a/vp9/common/mips/dspr2/vp9_convolve2_avg_horiz_dspr2.c b/vp9/common/mips/dspr2/vp9_convolve2_avg_horiz_dspr2.c
new file mode 100644
index 0000000..148b20f
--- /dev/null
+++ b/vp9/common/mips/dspr2/vp9_convolve2_avg_horiz_dspr2.c
@@ -0,0 +1,833 @@
+/*
+ * 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.
+ */
+
+#include <assert.h>
+#include <stdio.h>
+
+#include "./vpx_config.h"
+#include "./vp9_rtcd.h"
+#include "vp9/common/vp9_common.h"
+#include "vpx/vpx_integer.h"
+#include "vpx_ports/mem.h"
+#include "vp9/common/vp9_convolve.h"
+#include "vp9/common/mips/dspr2/vp9_common_dspr2.h"
+
+#if HAVE_DSPR2
+static void convolve_bi_avg_horiz_4_dspr2(const uint8_t *src,
+ int32_t src_stride,
+ uint8_t *dst,
+ int32_t dst_stride,
+ const int16_t *filter_x0,
+ int32_t h) {
+ int32_t y;
+ uint8_t *cm = vp9_ff_cropTbl;
+ int32_t Temp1, Temp2, Temp3, Temp4;
+ uint32_t vector4a = 64;
+ uint32_t tp1, tp2;
+ uint32_t p1, p2, p3;
+ uint32_t tn1, tn2;
+ const int16_t *filter = &filter_x0[3];
+ uint32_t filter45;
+
+ filter45 = ((const int32_t *)filter)[0];
+
+ for (y = h; y--;) {
+ /* prefetch data to cache memory */
+ vp9_prefetch_load(src + src_stride);
+ vp9_prefetch_load(src + src_stride + 32);
+ vp9_prefetch_store(dst + dst_stride);
+
+ __asm__ __volatile__ (
+ "ulw %[tp1], 0(%[src]) \n\t"
+ "ulw %[tp2], 4(%[src]) \n\t"
+
+ /* even 1. pixel */
+ "mtlo %[vector4a], $ac3 \n\t"
+ "mthi $zero, $ac3 \n\t"
+ "preceu.ph.qbr %[p1], %[tp1] \n\t"
+ "preceu.ph.qbl %[p2], %[tp1] \n\t"
+ "dpa.w.ph $ac3, %[p1], %[filter45] \n\t"
+ "extp %[Temp1], $ac3, 31 \n\t"
+
+ /* even 2. pixel */
+ "mtlo %[vector4a], $ac2 \n\t"
+ "mthi $zero, $ac2 \n\t"
+ "balign %[tp2], %[tp1], 3 \n\t"
+ "dpa.w.ph $ac2, %[p2], %[filter45] \n\t"
+ "extp %[Temp3], $ac2, 31 \n\t"
+
+ "lbu %[p2], 3(%[dst]) \n\t" /* load odd 2 */
+
+ /* odd 1. pixel */
+ "lbux %[tp1], %[Temp1](%[cm]) \n\t" /* even 1 */
+ "mtlo %[vector4a], $ac3 \n\t"
+ "mthi $zero, $ac3 \n\t"
+ "lbu %[Temp1], 1(%[dst]) \n\t" /* load odd 1 */
+ "preceu.ph.qbr %[p1], %[tp2] \n\t"
+ "preceu.ph.qbl %[p3], %[tp2] \n\t"
+ "dpa.w.ph $ac3, %[p1], %[filter45] \n\t"
+ "extp %[Temp2], $ac3, 31 \n\t"
+
+ "lbu %[tn2], 0(%[dst]) \n\t" /* load even 1 */
+
+ /* odd 2. pixel */
+ "lbux %[tp2], %[Temp3](%[cm]) \n\t" /* even 2 */
+ "mtlo %[vector4a], $ac2 \n\t"
+ "mthi $zero, $ac2 \n\t"
+ "lbux %[tn1], %[Temp2](%[cm]) \n\t" /* odd 1 */
+ "addqh_r.w %[tn2], %[tn2], %[tp1] \n\t" /* average even 1 */
+ "dpa.w.ph $ac2, %[p3], %[filter45] \n\t"
+ "extp %[Temp4], $ac2, 31 \n\t"
+
+ "lbu %[tp1], 2(%[dst]) \n\t" /* load even 2 */
+ "sb %[tn2], 0(%[dst]) \n\t" /* store even 1 */
+
+ /* clamp */
+ "addqh_r.w %[Temp1], %[Temp1], %[tn1] \n\t" /* average odd 1 */
+ "lbux %[p3], %[Temp4](%[cm]) \n\t" /* odd 2 */
+ "sb %[Temp1], 1(%[dst]) \n\t" /* store odd 1 */
+
+ "addqh_r.w %[tp1], %[tp1], %[tp2] \n\t" /* average even 2 */
+ "sb %[tp1], 2(%[dst]) \n\t" /* store even 2 */
+
+ "addqh_r.w %[p2], %[p2], %[p3] \n\t" /* average odd 2 */
+ "sb %[p2], 3(%[dst]) \n\t" /* store odd 2 */
+
+ : [tp1] "=&r" (tp1), [tp2] "=&r" (tp2),
+ [tn1] "=&r" (tn1), [tn2] "=&r" (tn2),
+ [p1] "=&r" (p1), [p2] "=&r" (p2), [p3] "=&r" (p3),
+ [Temp1] "=&r" (Temp1), [Temp2] "=&r" (Temp2),
+ [Temp3] "=&r" (Temp3), [Temp4] "=&r" (Temp4)
+ : [filter45] "r" (filter45), [vector4a] "r" (vector4a),
+ [cm] "r" (cm), [dst] "r" (dst), [src] "r" (src)
+ );
+
+ /* Next row... */
+ src += src_stride;
+ dst += dst_stride;
+ }
+}
+
+static void convolve_bi_avg_horiz_8_dspr2(const uint8_t *src,
+ int32_t src_stride,
+ uint8_t *dst,
+ int32_t dst_stride,
+ const int16_t *filter_x0,
+ int32_t h) {
+ int32_t y;
+ uint8_t *cm = vp9_ff_cropTbl;
+ uint32_t vector4a = 64;
+ int32_t Temp1, Temp2, Temp3;
+ uint32_t tp1, tp2, tp3, tp4;
+ uint32_t p1, p2, p3, p4, n1;
+ uint32_t st0, st1;
+ const int16_t *filter = &filter_x0[3];
+ uint32_t filter45;;
+
+ filter45 = ((const int32_t *)filter)[0];
+
+ for (y = h; y--;) {
+ /* prefetch data to cache memory */
+ vp9_prefetch_load(src + src_stride);
+ vp9_prefetch_load(src + src_stride + 32);
+ vp9_prefetch_store(dst + dst_stride);
+
+ __asm__ __volatile__ (
+ "ulw %[tp1], 0(%[src]) \n\t"
+ "ulw %[tp2], 4(%[src]) \n\t"
+
+ /* even 1. pixel */
+ "mtlo %[vector4a], $ac3 \n\t"
+ "mthi $zero, $ac3 \n\t"
+ "mtlo %[vector4a], $ac2 \n\t"
+ "mthi $zero, $ac2 \n\t"
+ "preceu.ph.qbr %[p1], %[tp1] \n\t"
+ "preceu.ph.qbl %[p2], %[tp1] \n\t"
+ "preceu.ph.qbr %[p3], %[tp2] \n\t"
+ "preceu.ph.qbl %[p4], %[tp2] \n\t"
+ "ulw %[tp3], 8(%[src]) \n\t"
+ "dpa.w.ph $ac3, %[p1], %[filter45] \n\t"
+ "extp %[Temp1], $ac3, 31 \n\t"
+ "lbu %[Temp2], 0(%[dst]) \n\t"
+ "lbu %[tp4], 2(%[dst]) \n\t"
+
+ /* even 2. pixel */
+ "dpa.w.ph $ac2, %[p2], %[filter45] \n\t"
+ "extp %[Temp3], $ac2, 31 \n\t"
+
+ /* even 3. pixel */
+ "lbux %[st0], %[Temp1](%[cm]) \n\t"
+ "mtlo %[vector4a], $ac1 \n\t"
+ "mthi $zero, $ac1 \n\t"
+ "lbux %[st1], %[Temp3](%[cm]) \n\t"
+ "dpa.w.ph $ac1, %[p3], %[filter45] \n\t"
+ "extp %[Temp1], $ac1, 31 \n\t"
+
+ "addqh_r.w %[Temp2], %[Temp2], %[st0] \n\t"
+ "addqh_r.w %[tp4], %[tp4], %[st1] \n\t"
+ "sb %[Temp2], 0(%[dst]) \n\t"
+ "sb %[tp4], 2(%[dst]) \n\t"
+
+ /* even 4. pixel */
+ "mtlo %[vector4a], $ac2 \n\t"
+ "mthi $zero, $ac2 \n\t"
+ "mtlo %[vector4a], $ac3 \n\t"
+ "mthi $zero, $ac3 \n\t"
+
+ "balign %[tp3], %[tp2], 3 \n\t"
+ "balign %[tp2], %[tp1], 3 \n\t"
+
+ "lbux %[st0], %[Temp1](%[cm]) \n\t"
+ "lbu %[Temp2], 4(%[dst]) \n\t"
+ "addqh_r.w %[Temp2], %[Temp2], %[st0] \n\t"
+
+ "dpa.w.ph $ac2, %[p4], %[filter45] \n\t"
+ "extp %[Temp3], $ac2, 31 \n\t"
+
+ /* odd 1. pixel */
+ "mtlo %[vector4a], $ac1 \n\t"
+ "mthi $zero, $ac1 \n\t"
+ "sb %[Temp2], 4(%[dst]) \n\t"
+ "preceu.ph.qbr %[p1], %[tp2] \n\t"
+ "preceu.ph.qbl %[p2], %[tp2] \n\t"
+ "preceu.ph.qbr %[p3], %[tp3] \n\t"
+ "preceu.ph.qbl %[p4], %[tp3] \n\t"
+ "dpa.w.ph $ac3, %[p1], %[filter45] \n\t"
+ "extp %[Temp2], $ac3, 31 \n\t"
+
+ "lbu %[tp1], 6(%[dst]) \n\t"
+
+ /* odd 2. pixel */
+ "mtlo %[vector4a], $ac3 \n\t"
+ "mthi $zero, $ac3 \n\t"
+ "mtlo %[vector4a], $ac2 \n\t"
+ "mthi $zero, $ac2 \n\t"
+ "lbux %[st0], %[Temp3](%[cm]) \n\t"
+ "dpa.w.ph $ac1, %[p2], %[filter45] \n\t"
+ "extp %[Temp3], $ac1, 31 \n\t"
+
+ "lbu %[tp2], 1(%[dst]) \n\t"
+ "lbu %[tp3], 3(%[dst]) \n\t"
+ "addqh_r.w %[tp1], %[tp1], %[st0] \n\t"
+
+ /* odd 3. pixel */
+ "lbux %[st1], %[Temp2](%[cm]) \n\t"
+ "dpa.w.ph $ac3, %[p3], %[filter45] \n\t"
+ "addqh_r.w %[tp2], %[tp2], %[st1] \n\t"
+ "extp %[Temp2], $ac3, 31 \n\t"
+
+ "lbu %[tp4], 5(%[dst]) \n\t"
+
+ /* odd 4. pixel */
+ "sb %[tp2], 1(%[dst]) \n\t"
+ "sb %[tp1], 6(%[dst]) \n\t"
+ "dpa.w.ph $ac2, %[p4], %[filter45] \n\t"
+ "extp %[Temp1], $ac2, 31 \n\t"
+
+ "lbu %[tp1], 7(%[dst]) \n\t"
+
+ /* clamp */
+ "lbux %[p4], %[Temp3](%[cm]) \n\t"
+ "addqh_r.w %[tp3], %[tp3], %[p4] \n\t"
+
+ "lbux %[p2], %[Temp2](%[cm]) \n\t"
+ "addqh_r.w %[tp4], %[tp4], %[p2] \n\t"
+
+ "lbux %[p1], %[Temp1](%[cm]) \n\t"
+ "addqh_r.w %[tp1], %[tp1], %[p1] \n\t"
+
+ /* store bytes */
+ "sb %[tp3], 3(%[dst]) \n\t"
+ "sb %[tp4], 5(%[dst]) \n\t"
+ "sb %[tp1], 7(%[dst]) \n\t"
+
+ : [tp1] "=&r" (tp1), [tp2] "=&r" (tp2),
+ [tp3] "=&r" (tp3), [tp4] "=&r" (tp4),
+ [st0] "=&r" (st0), [st1] "=&r" (st1),
+ [p1] "=&r" (p1), [p2] "=&r" (p2), [p3] "=&r" (p3), [p4] "=&r" (p4),
+ [n1] "=&r" (n1),
+ [Temp1] "=&r" (Temp1), [Temp2] "=&r" (Temp2), [Temp3] "=&r" (Temp3)
+ : [filter45] "r" (filter45), [vector4a] "r" (vector4a),
+ [cm] "r" (cm), [dst] "r" (dst), [src] "r" (src)
+ );
+
+ /* Next row... */
+ src += src_stride;
+ dst += dst_stride;
+ }
+}
+
+static void convolve_bi_avg_horiz_16_dspr2(const uint8_t *src_ptr,
+ int32_t src_stride,
+ uint8_t *dst_ptr,
+ int32_t dst_stride,
+ const int16_t *filter_x0,
+ int32_t h,
+ int32_t count) {
+ int32_t y, c;
+ const uint8_t *src;
+ uint8_t *dst;
+ uint8_t *cm = vp9_ff_cropTbl;
+ uint32_t vector_64 = 64;
+ int32_t Temp1, Temp2, Temp3;
+ uint32_t qload1, qload2, qload3;
+ uint32_t p1, p2, p3, p4, p5;
+ uint32_t st1, st2, st3;
+ const int16_t *filter = &filter_x0[3];
+ uint32_t filter45;;
+
+ filter45 = ((const int32_t *)filter)[0];
+
+ for (y = h; y--;) {
+ src = src_ptr;
+ dst = dst_ptr;
+
+ /* prefetch data to cache memory */
+ vp9_prefetch_load(src_ptr + src_stride);
+ vp9_prefetch_load(src_ptr + src_stride + 32);
+ vp9_prefetch_store(dst_ptr + dst_stride);
+
+ for (c = 0; c < count; c++) {
+ __asm__ __volatile__ (
+ "ulw %[qload1], 0(%[src]) \n\t"
+ "ulw %[qload2], 4(%[src]) \n\t"
+
+ /* even 1. pixel */
+ "mtlo %[vector_64], $ac1 \n\t" /* even 1 */
+ "mthi $zero, $ac1 \n\t"
+ "mtlo %[vector_64], $ac2 \n\t" /* even 2 */
+ "mthi $zero, $ac2 \n\t"
+ "preceu.ph.qbr %[p1], %[qload1] \n\t"
+ "preceu.ph.qbl %[p2], %[qload1] \n\t"
+ "preceu.ph.qbr %[p3], %[qload2] \n\t"
+ "preceu.ph.qbl %[p4], %[qload2] \n\t"
+ "ulw %[qload3], 8(%[src]) \n\t"
+ "dpa.w.ph $ac1, %[p1], %[filter45] \n\t" /* even 1 */
+ "extp %[Temp1], $ac1, 31 \n\t" /* even 1 */
+ "lbu %[st2], 0(%[dst]) \n\t" /* load even 1 from dst */
+
+ /* even 2. pixel */
+ "mtlo %[vector_64], $ac3 \n\t" /* even 3 */
+ "mthi $zero, $ac3 \n\t"
+ "preceu.ph.qbr %[p1], %[qload3] \n\t"
+ "preceu.ph.qbl %[p5], %[qload3] \n\t"
+ "ulw %[qload1], 12(%[src]) \n\t"
+ "dpa.w.ph $ac2, %[p2], %[filter45] \n\t" /* even 1 */
+ "extp %[Temp2], $ac2, 31 \n\t" /* even 1 */
+ "lbux %[st1], %[Temp1](%[cm]) \n\t" /* even 1 */
+
+ "lbu %[qload3], 2(%[dst]) \n\t" /* load even 2 from dst */
+
+ /* even 3. pixel */
+ "mtlo %[vector_64], $ac1 \n\t" /* even 4 */
+ "mthi $zero, $ac1 \n\t"
+ "addqh_r.w %[st2], %[st2], %[st1] \n\t" /* average even 1 */
+ "preceu.ph.qbr %[p2], %[qload1] \n\t"
+ "sb %[st2], 0(%[dst]) \n\t" /* store even 1 to dst */
+ "dpa.w.ph $ac3, %[p3], %[filter45] \n\t" /* even 3 */
+ "extp %[Temp3], $ac3, 31 \n\t" /* even 3 */
+ "lbux %[st2], %[Temp2](%[cm]) \n\t" /* even 1 */
+
+ /* even 4. pixel */
+ "mtlo %[vector_64], $ac2 \n\t" /* even 5 */
+ "mthi $zero, $ac2 \n\t"
+ "addqh_r.w %[qload3], %[qload3], %[st2] \n\t" /* average even 2 */
+ "preceu.ph.qbl %[p3], %[qload1] \n\t"
+ "sb %[qload3], 2(%[dst]) \n\t" /* store even 2 to dst */
+ "lbu %[qload3], 4(%[dst]) \n\t" /* load even 3 from dst */
+ "lbu %[qload1], 6(%[dst]) \n\t" /* load even 4 from dst */
+ "dpa.w.ph $ac1, %[p4], %[filter45] \n\t" /* even 4 */
+ "extp %[Temp1], $ac1, 31 \n\t" /* even 4 */
+ "lbux %[st3], %[Temp3](%[cm]) \n\t" /* even 3 */
+
+ /* even 5. pixel */
+ "mtlo %[vector_64], $ac3 \n\t" /* even 6 */
+ "mthi $zero, $ac3 \n\t"
+ "addqh_r.w %[qload3], %[qload3], %[st3] \n\t" /* average even 3 */
+ "sb %[qload3], 4(%[dst]) \n\t" /* store even 3 to dst */
+ "dpa.w.ph $ac2, %[p1], %[filter45] \n\t" /* even 5 */
+ "extp %[Temp2], $ac2, 31 \n\t" /* even 5 */
+ "lbux %[st1], %[Temp1](%[cm]) \n\t" /* even 4 */
+
+ /* even 6. pixel */
+ "mtlo %[vector_64], $ac1 \n\t" /* even 7 */
+ "mthi $zero, $ac1 \n\t"
+ "addqh_r.w %[qload1], %[qload1], %[st1] \n\t" /* average even 4 */
+ "sb %[qload1], 6(%[dst]) \n\t" /* store even 4 to dst */
+ "dpa.w.ph $ac3, %[p5], %[filter45] \n\t" /* even 6 */
+ "lbu %[qload2], 8(%[dst]) \n\t" /* load even 5 from dst */
+ "extp %[Temp3], $ac3, 31 \n\t" /* even 6 */
+ "lbux %[st2], %[Temp2](%[cm]) \n\t" /* even 5 */
+
+ /* even 7. pixel */
+ "mtlo %[vector_64], $ac2 \n\t" /* even 8 */
+ "mthi $zero, $ac2 \n\t"
+ "addqh_r.w %[qload2], %[qload2], %[st2] \n\t" /* average even 5 */
+ "sb %[qload2], 8(%[dst]) \n\t" /* store even 5 to dst */
+ "dpa.w.ph $ac1, %[p2], %[filter45] \n\t" /* even 7 */
+ "lbu %[qload3], 10(%[dst]) \n\t" /* load even 6 from dst */
+ "extp %[Temp1], $ac1, 31 \n\t" /* even 7 */
+ "lbux %[st3], %[Temp3](%[cm]) \n\t" /* even 6 */
+
+ "lbu %[st2], 12(%[dst]) \n\t" /* load even 7 from dst */
+
+ /* even 8. pixel */
+ "mtlo %[vector_64], $ac3 \n\t" /* odd 1 */
+ "mthi $zero, $ac3 \n\t"
+ "addqh_r.w %[qload3], %[qload3], %[st3] \n\t" /* average even 6 */
+ "dpa.w.ph $ac2, %[p3], %[filter45] \n\t" /* even 8 */
+ "sb %[qload3], 10(%[dst]) \n\t" /* store even 6 to dst */
+ "extp %[Temp2], $ac2, 31 \n\t" /* even 8 */
+ "lbux %[st1], %[Temp1](%[cm]) \n\t" /* even 7 */
+
+ /* ODD pixels */
+ "ulw %[qload1], 1(%[src]) \n\t"
+ "ulw %[qload2], 5(%[src]) \n\t"
+
+ "addqh_r.w %[st2], %[st2], %[st1] \n\t" /* average even 7 */
+
+ /* odd 1. pixel */
+ "mtlo %[vector_64], $ac1 \n\t" /* odd 2 */
+ "mthi $zero, $ac1 \n\t"
+ "preceu.ph.qbr %[p1], %[qload1] \n\t"
+ "preceu.ph.qbl %[p2], %[qload1] \n\t"
+ "preceu.ph.qbr %[p3], %[qload2] \n\t"
+ "preceu.ph.qbl %[p4], %[qload2] \n\t"
+ "sb %[st2], 12(%[dst]) \n\t" /* store even 7 to dst */
+ "ulw %[qload3], 9(%[src]) \n\t"
+ "dpa.w.ph $ac3, %[p1], %[filter45] \n\t" /* odd 1 */
+ "lbu %[qload2], 14(%[dst]) \n\t" /* load even 8 from dst */
+ "extp %[Temp3], $ac3, 31 \n\t" /* odd 1 */
+ "lbux %[st2], %[Temp2](%[cm]) \n\t" /* even 8 */
+
+ "lbu %[st1], 1(%[dst]) \n\t" /* load odd 1 from dst */
+
+ /* odd 2. pixel */
+ "mtlo %[vector_64], $ac2 \n\t" /* odd 3 */
+ "mthi $zero, $ac2 \n\t"
+ "addqh_r.w %[qload2], %[qload2], %[st2] \n\t" /* average even 8 */
+ "preceu.ph.qbr %[p1], %[qload3] \n\t"
+ "preceu.ph.qbl %[p5], %[qload3] \n\t"
+ "sb %[qload2], 14(%[dst]) \n\t" /* store even 8 to dst */
+ "ulw %[qload1], 13(%[src]) \n\t"
+ "dpa.w.ph $ac1, %[p2], %[filter45] \n\t" /* odd 2 */
+ "lbu %[qload3], 3(%[dst]) \n\t" /* load odd 2 from dst */
+ "extp %[Temp1], $ac1, 31 \n\t" /* odd 2 */
+ "lbux %[st3], %[Temp3](%[cm]) \n\t" /* odd 1 */
+
+ /* odd 3. pixel */
+ "mtlo %[vector_64], $ac3 \n\t" /* odd 4 */
+ "mthi $zero, $ac3 \n\t"
+ "addqh_r.w %[st3], %[st3], %[st1] \n\t" /* average odd 1 */
+ "preceu.ph.qbr %[p2], %[qload1] \n\t"
+ "dpa.w.ph $ac2, %[p3], %[filter45] \n\t" /* odd 3 */
+ "sb %[st3], 1(%[dst]) \n\t" /* store odd 1 to dst */
+ "extp %[Temp2], $ac2, 31 \n\t" /* odd 3 */
+ "lbux %[st1], %[Temp1](%[cm]) \n\t" /* odd 2 */
+
+ /* odd 4. pixel */
+ "mtlo %[vector_64], $ac1 \n\t" /* odd 5 */
+ "mthi $zero, $ac1 \n\t"
+ "addqh_r.w %[qload3], %[qload3], %[st1] \n\t" /* average odd 2 */
+ "preceu.ph.qbl %[p3], %[qload1] \n\t"
+ "sb %[qload3], 3(%[dst]) \n\t" /* store odd 2 to dst */
+ "lbu %[qload1], 5(%[dst]) \n\t" /* load odd 3 from dst */
+ "dpa.w.ph $ac3, %[p4], %[filter45] \n\t" /* odd 4 */
+ "extp %[Temp3], $ac3, 31 \n\t" /* odd 4 */
+ "lbux %[st2], %[Temp2](%[cm]) \n\t" /* odd 3 */
+
+ "lbu %[st1], 7(%[dst]) \n\t" /* load odd 4 from dst */
+
+ /* odd 5. pixel */
+ "mtlo %[vector_64], $ac2 \n\t" /* odd 6 */
+ "mthi $zero, $ac2 \n\t"
+ "addqh_r.w %[qload1], %[qload1], %[st2] \n\t" /* average odd 3 */
+ "sb %[qload1], 5(%[dst]) \n\t" /* store odd 3 to dst */
+ "dpa.w.ph $ac1, %[p1], %[filter45] \n\t" /* odd 5 */
+ "extp %[Temp1], $ac1, 31 \n\t" /* odd 5 */
+ "lbux %[st3], %[Temp3](%[cm]) \n\t" /* odd 4 */
+
+ "lbu %[qload1], 9(%[dst]) \n\t" /* load odd 5 from dst */
+
+ /* odd 6. pixel */
+ "mtlo %[vector_64], $ac3 \n\t" /* odd 7 */
+ "mthi $zero, $ac3 \n\t"
+ "addqh_r.w %[st1], %[st1], %[st3] \n\t" /* average odd 4 */
+ "sb %[st1], 7(%[dst]) \n\t" /* store odd 4 to dst */
+ "dpa.w.ph $ac2, %[p5], %[filter45] \n\t" /* odd 6 */
+ "extp %[Temp2], $ac2, 31 \n\t" /* odd 6 */
+ "lbux %[st1], %[Temp1](%[cm]) \n\t" /* odd 5 */
+
+ /* odd 7. pixel */
+ "mtlo %[vector_64], $ac1 \n\t" /* odd 8 */
+ "mthi $zero, $ac1 \n\t"
+ "addqh_r.w %[qload1], %[qload1], %[st1] \n\t" /* average odd 5 */
+ "sb %[qload1], 9(%[dst]) \n\t" /* store odd 5 to dst */
+ "lbu %[qload2], 11(%[dst]) \n\t" /* load odd 6 from dst */
+ "dpa.w.ph $ac3, %[p2], %[filter45] \n\t" /* odd 7 */
+ "extp %[Temp3], $ac3, 31 \n\t" /* odd 7 */
+
+ "lbu %[qload3], 13(%[dst]) \n\t" /* load odd 7 from dst */
+
+ /* odd 8. pixel */
+ "dpa.w.ph $ac1, %[p3], %[filter45] \n\t" /* odd 8 */
+ "extp %[Temp1], $ac1, 31 \n\t" /* odd 8 */
+
+ "lbu %[qload1], 15(%[dst]) \n\t" /* load odd 8 from dst */
+
+ "lbux %[st2], %[Temp2](%[cm]) \n\t" /* odd 6 */
+ "addqh_r.w %[qload2], %[qload2], %[st2] \n\t" /* average odd 6 */
+
+ "lbux %[st3], %[Temp3](%[cm]) \n\t" /* odd 7 */
+ "addqh_r.w %[qload3], %[qload3], %[st3] \n\t" /* average odd 7 */
+
+ "lbux %[st1], %[Temp1](%[cm]) \n\t" /* odd 8 */
+ "addqh_r.w %[qload1], %[qload1], %[st1] \n\t" /* average odd 8 */
+
+ "sb %[qload2], 11(%[dst]) \n\t" /* store odd 6 to dst */
+ "sb %[qload3], 13(%[dst]) \n\t" /* store odd 7 to dst */
+ "sb %[qload1], 15(%[dst]) \n\t" /* store odd 8 to dst */
+
+ : [qload1] "=&r" (qload1), [qload2] "=&r" (qload2),
+ [st1] "=&r" (st1), [st2] "=&r" (st2), [st3] "=&r" (st3),
+ [p1] "=&r" (p1), [p2] "=&r" (p2), [p3] "=&r" (p3), [p4] "=&r" (p4),
+ [qload3] "=&r" (qload3), [p5] "=&r" (p5),
+ [Temp1] "=&r" (Temp1), [Temp2] "=&r" (Temp2), [Temp3] "=&r" (Temp3)
+ : [filter45] "r" (filter45), [vector_64] "r" (vector_64),
+ [cm] "r" (cm), [dst] "r" (dst), [src] "r" (src)
+ );
+
+ src += 16;
+ dst += 16;
+ }
+
+ /* Next row... */
+ src_ptr += src_stride;
+ dst_ptr += dst_stride;
+ }
+}
+
+static void convolve_bi_avg_horiz_64_dspr2(const uint8_t *src_ptr,
+ int32_t src_stride,
+ uint8_t *dst_ptr,
+ int32_t dst_stride,
+ const int16_t *filter_x0,
+ int32_t h) {
+ int32_t y, c;
+ const uint8_t *src;
+ uint8_t *dst;
+ uint8_t *cm = vp9_ff_cropTbl;
+ uint32_t vector_64 = 64;
+ int32_t Temp1, Temp2, Temp3;
+ uint32_t qload1, qload2, qload3;
+ uint32_t p1, p2, p3, p4, p5;
+ uint32_t st1, st2, st3;
+ const int16_t *filter = &filter_x0[3];
+ uint32_t filter45;;
+
+ filter45 = ((const int32_t *)filter)[0];
+
+ for (y = h; y--;) {
+ src = src_ptr;
+ dst = dst_ptr;
+
+ /* prefetch data to cache memory */
+ vp9_prefetch_load(src_ptr + src_stride);
+ vp9_prefetch_load(src_ptr + src_stride + 32);
+ vp9_prefetch_load(src_ptr + src_stride + 64);
+ vp9_prefetch_store(dst_ptr + dst_stride);
+ vp9_prefetch_store(dst_ptr + dst_stride + 32);
+
+ for (c = 0; c < 4; c++) {
+ __asm__ __volatile__ (
+ "ulw %[qload1], 0(%[src]) \n\t"
+ "ulw %[qload2], 4(%[src]) \n\t"
+
+ /* even 1. pixel */
+ "mtlo %[vector_64], $ac1 \n\t" /* even 1 */
+ "mthi $zero, $ac1 \n\t"
+ "mtlo %[vector_64], $ac2 \n\t" /* even 2 */
+ "mthi $zero, $ac2 \n\t"
+ "preceu.ph.qbr %[p1], %[qload1] \n\t"
+ "preceu.ph.qbl %[p2], %[qload1] \n\t"
+ "preceu.ph.qbr %[p3], %[qload2] \n\t"
+ "preceu.ph.qbl %[p4], %[qload2] \n\t"
+ "ulw %[qload3], 8(%[src]) \n\t"
+ "dpa.w.ph $ac1, %[p1], %[filter45] \n\t" /* even 1 */
+ "extp %[Temp1], $ac1, 31 \n\t" /* even 1 */
+ "lbu %[st2], 0(%[dst]) \n\t" /* load even 1 from dst */
+
+ /* even 2. pixel */
+ "mtlo %[vector_64], $ac3 \n\t" /* even 3 */
+ "mthi $zero, $ac3 \n\t"
+ "preceu.ph.qbr %[p1], %[qload3] \n\t"
+ "preceu.ph.qbl %[p5], %[qload3] \n\t"
+ "ulw %[qload1], 12(%[src]) \n\t"
+ "dpa.w.ph $ac2, %[p2], %[filter45] \n\t" /* even 1 */
+ "extp %[Temp2], $ac2, 31 \n\t" /* even 1 */
+ "lbux %[st1], %[Temp1](%[cm]) \n\t" /* even 1 */
+
+ "lbu %[qload3], 2(%[dst]) \n\t" /* load even 2 from dst */
+
+ /* even 3. pixel */
+ "mtlo %[vector_64], $ac1 \n\t" /* even 4 */
+ "mthi $zero, $ac1 \n\t"
+ "addqh_r.w %[st2], %[st2], %[st1] \n\t" /* average even 1 */
+ "preceu.ph.qbr %[p2], %[qload1] \n\t"
+ "sb %[st2], 0(%[dst]) \n\t" /* store even 1 to dst */
+ "dpa.w.ph $ac3, %[p3], %[filter45] \n\t" /* even 3 */
+ "extp %[Temp3], $ac3, 31 \n\t" /* even 3 */
+ "lbux %[st2], %[Temp2](%[cm]) \n\t" /* even 1 */
+
+ /* even 4. pixel */
+ "mtlo %[vector_64], $ac2 \n\t" /* even 5 */
+ "mthi $zero, $ac2 \n\t"
+ "addqh_r.w %[qload3], %[qload3], %[st2] \n\t" /* average even 2 */
+ "preceu.ph.qbl %[p3], %[qload1] \n\t"
+ "sb %[qload3], 2(%[dst]) \n\t" /* store even 2 to dst */
+ "lbu %[qload3], 4(%[dst]) \n\t" /* load even 3 from dst */
+ "lbu %[qload1], 6(%[dst]) \n\t" /* load even 4 from dst */
+ "dpa.w.ph $ac1, %[p4], %[filter45] \n\t" /* even 4 */
+ "extp %[Temp1], $ac1, 31 \n\t" /* even 4 */
+ "lbux %[st3], %[Temp3](%[cm]) \n\t" /* even 3 */
+
+ /* even 5. pixel */
+ "mtlo %[vector_64], $ac3 \n\t" /* even 6 */
+ "mthi $zero, $ac3 \n\t"
+ "addqh_r.w %[qload3], %[qload3], %[st3] \n\t" /* average even 3 */
+ "sb %[qload3], 4(%[dst]) \n\t" /* store even 3 to dst */
+ "dpa.w.ph $ac2, %[p1], %[filter45] \n\t" /* even 5 */
+ "extp %[Temp2], $ac2, 31 \n\t" /* even 5 */
+ "lbux %[st1], %[Temp1](%[cm]) \n\t" /* even 4 */
+
+ /* even 6. pixel */
+ "mtlo %[vector_64], $ac1 \n\t" /* even 7 */
+ "mthi $zero, $ac1 \n\t"
+ "addqh_r.w %[qload1], %[qload1], %[st1] \n\t" /* average even 4 */
+ "sb %[qload1], 6(%[dst]) \n\t" /* store even 4 to dst */
+ "dpa.w.ph $ac3, %[p5], %[filter45] \n\t" /* even 6 */
+ "lbu %[qload2], 8(%[dst]) \n\t" /* load even 5 from dst */
+ "extp %[Temp3], $ac3, 31 \n\t" /* even 6 */
+ "lbux %[st2], %[Temp2](%[cm]) \n\t" /* even 5 */
+
+ /* even 7. pixel */
+ "mtlo %[vector_64], $ac2 \n\t" /* even 8 */
+ "mthi $zero, $ac2 \n\t"
+ "addqh_r.w %[qload2], %[qload2], %[st2] \n\t" /* average even 5 */
+ "sb %[qload2], 8(%[dst]) \n\t" /* store even 5 to dst */
+ "dpa.w.ph $ac1, %[p2], %[filter45] \n\t" /* even 7 */
+ "lbu %[qload3], 10(%[dst]) \n\t" /* load even 6 from dst */
+ "extp %[Temp1], $ac1, 31 \n\t" /* even 7 */
+ "lbux %[st3], %[Temp3](%[cm]) \n\t" /* even 6 */
+
+ "lbu %[st2], 12(%[dst]) \n\t" /* load even 7 from dst */
+
+ /* even 8. pixel */
+ "mtlo %[vector_64], $ac3 \n\t" /* odd 1 */
+ "mthi $zero, $ac3 \n\t"
+ "addqh_r.w %[qload3], %[qload3], %[st3] \n\t" /* average even 6 */
+ "dpa.w.ph $ac2, %[p3], %[filter45] \n\t" /* even 8 */
+ "sb %[qload3], 10(%[dst]) \n\t" /* store even 6 to dst */
+ "extp %[Temp2], $ac2, 31 \n\t" /* even 8 */
+ "lbux %[st1], %[Temp1](%[cm]) \n\t" /* even 7 */
+
+ /* ODD pixels */
+ "ulw %[qload1], 1(%[src]) \n\t"
+ "ulw %[qload2], 5(%[src]) \n\t"
+
+ "addqh_r.w %[st2], %[st2], %[st1] \n\t" /* average even 7 */
+
+ /* odd 1. pixel */
+ "mtlo %[vector_64], $ac1 \n\t" /* odd 2 */
+ "mthi $zero, $ac1 \n\t"
+ "preceu.ph.qbr %[p1], %[qload1] \n\t"
+ "preceu.ph.qbl %[p2], %[qload1] \n\t"
+ "preceu.ph.qbr %[p3], %[qload2] \n\t"
+ "preceu.ph.qbl %[p4], %[qload2] \n\t"
+ "sb %[st2], 12(%[dst]) \n\t" /* store even 7 to dst */
+ "ulw %[qload3], 9(%[src]) \n\t"
+ "dpa.w.ph $ac3, %[p1], %[filter45] \n\t" /* odd 1 */
+ "lbu %[qload2], 14(%[dst]) \n\t" /* load even 8 from dst */
+ "extp %[Temp3], $ac3, 31 \n\t" /* odd 1 */
+ "lbux %[st2], %[Temp2](%[cm]) \n\t" /* even 8 */
+
+ "lbu %[st1], 1(%[dst]) \n\t" /* load odd 1 from dst */
+
+ /* odd 2. pixel */
+ "mtlo %[vector_64], $ac2 \n\t" /* odd 3 */
+ "mthi $zero, $ac2 \n\t"
+ "addqh_r.w %[qload2], %[qload2], %[st2] \n\t" /* average even 8 */
+ "preceu.ph.qbr %[p1], %[qload3] \n\t"
+ "preceu.ph.qbl %[p5], %[qload3] \n\t"
+ "sb %[qload2], 14(%[dst]) \n\t" /* store even 8 to dst */
+ "ulw %[qload1], 13(%[src]) \n\t"
+ "dpa.w.ph $ac1, %[p2], %[filter45] \n\t" /* odd 2 */
+ "lbu %[qload3], 3(%[dst]) \n\t" /* load odd 2 from dst */
+ "extp %[Temp1], $ac1, 31 \n\t" /* odd 2 */
+ "lbux %[st3], %[Temp3](%[cm]) \n\t" /* odd 1 */
+
+ /* odd 3. pixel */
+ "mtlo %[vector_64], $ac3 \n\t" /* odd 4 */
+ "mthi $zero, $ac3 \n\t"
+ "addqh_r.w %[st3], %[st3], %[st1] \n\t" /* average odd 1 */
+ "preceu.ph.qbr %[p2], %[qload1] \n\t"
+ "dpa.w.ph $ac2, %[p3], %[filter45] \n\t" /* odd 3 */
+ "sb %[st3], 1(%[dst]) \n\t" /* store odd 1 to dst */
+ "extp %[Temp2], $ac2, 31 \n\t" /* odd 3 */
+ "lbux %[st1], %[Temp1](%[cm]) \n\t" /* odd 2 */
+
+ /* odd 4. pixel */
+ "mtlo %[vector_64], $ac1 \n\t" /* odd 5 */
+ "mthi $zero, $ac1 \n\t"
+ "addqh_r.w %[qload3], %[qload3], %[st1] \n\t" /* average odd 2 */
+ "preceu.ph.qbl %[p3], %[qload1] \n\t"
+ "sb %[qload3], 3(%[dst]) \n\t" /* store odd 2 to dst */
+ "lbu %[qload1], 5(%[dst]) \n\t" /* load odd 3 from dst */
+ "dpa.w.ph $ac3, %[p4], %[filter45] \n\t" /* odd 4 */
+ "extp %[Temp3], $ac3, 31 \n\t" /* odd 4 */
+ "lbux %[st2], %[Temp2](%[cm]) \n\t" /* odd 3 */
+
+ "lbu %[st1], 7(%[dst]) \n\t" /* load odd 4 from dst */
+
+ /* odd 5. pixel */
+ "mtlo %[vector_64], $ac2 \n\t" /* odd 6 */
+ "mthi $zero, $ac2 \n\t"
+ "addqh_r.w %[qload1], %[qload1], %[st2] \n\t" /* average odd 3 */
+ "sb %[qload1], 5(%[dst]) \n\t" /* store odd 3 to dst */
+ "dpa.w.ph $ac1, %[p1], %[filter45] \n\t" /* odd 5 */
+ "extp %[Temp1], $ac1, 31 \n\t" /* odd 5 */
+ "lbux %[st3], %[Temp3](%[cm]) \n\t" /* odd 4 */
+
+ "lbu %[qload1], 9(%[dst]) \n\t" /* load odd 5 from dst */
+
+ /* odd 6. pixel */
+ "mtlo %[vector_64], $ac3 \n\t" /* odd 7 */
+ "mthi $zero, $ac3 \n\t"
+ "addqh_r.w %[st1], %[st1], %[st3] \n\t" /* average odd 4 */
+ "sb %[st1], 7(%[dst]) \n\t" /* store odd 4 to dst */
+ "dpa.w.ph $ac2, %[p5], %[filter45] \n\t" /* odd 6 */
+ "extp %[Temp2], $ac2, 31 \n\t" /* odd 6 */
+ "lbux %[st1], %[Temp1](%[cm]) \n\t" /* odd 5 */
+
+ /* odd 7. pixel */
+ "mtlo %[vector_64], $ac1 \n\t" /* odd 8 */
+ "mthi $zero, $ac1 \n\t"
+ "addqh_r.w %[qload1], %[qload1], %[st1] \n\t" /* average odd 5 */
+ "sb %[qload1], 9(%[dst]) \n\t" /* store odd 5 to dst */
+ "lbu %[qload2], 11(%[dst]) \n\t" /* load odd 6 from dst */
+ "dpa.w.ph $ac3, %[p2], %[filter45] \n\t" /* odd 7 */
+ "extp %[Temp3], $ac3, 31 \n\t" /* odd 7 */
+
+ "lbu %[qload3], 13(%[dst]) \n\t" /* load odd 7 from dst */
+
+ /* odd 8. pixel */
+ "dpa.w.ph $ac1, %[p3], %[filter45] \n\t" /* odd 8 */
+ "extp %[Temp1], $ac1, 31 \n\t" /* odd 8 */
+
+ "lbu %[qload1], 15(%[dst]) \n\t" /* load odd 8 from dst */
+
+ "lbux %[st2], %[Temp2](%[cm]) \n\t" /* odd 6 */
+ "addqh_r.w %[qload2], %[qload2], %[st2] \n\t" /* average odd 6 */
+
+ "lbux %[st3], %[Temp3](%[cm]) \n\t" /* odd 7 */
+ "addqh_r.w %[qload3], %[qload3], %[st3] \n\t" /* average odd 7 */
+
+ "lbux %[st1], %[Temp1](%[cm]) \n\t" /* odd 8 */
+ "addqh_r.w %[qload1], %[qload1], %[st1] \n\t" /* average odd 8 */
+
+ "sb %[qload2], 11(%[dst]) \n\t" /* store odd 6 to dst */
+ "sb %[qload3], 13(%[dst]) \n\t" /* store odd 7 to dst */
+ "sb %[qload1], 15(%[dst]) \n\t" /* store odd 8 to dst */
+
+ : [qload1] "=&r" (qload1), [qload2] "=&r" (qload2),
+ [st1] "=&r" (st1), [st2] "=&r" (st2), [st3] "=&r" (st3),
+ [p1] "=&r" (p1), [p2] "=&r" (p2), [p3] "=&r" (p3), [p4] "=&r" (p4),
+ [qload3] "=&r" (qload3), [p5] "=&r" (p5),
+ [Temp1] "=&r" (Temp1), [Temp2] "=&r" (Temp2), [Temp3] "=&r" (Temp3)
+ : [filter45] "r" (filter45), [vector_64] "r" (vector_64),
+ [cm] "r" (cm), [dst] "r" (dst), [src] "r" (src)
+ );
+
+ src += 16;
+ dst += 16;
+ }
+
+ /* Next row... */
+ src_ptr += src_stride;
+ dst_ptr += dst_stride;
+ }
+}
+
+void vp9_convolve2_avg_horiz_dspr2(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) {
+ if (16 == x_step_q4) {
+ uint32_t pos = 38;
+
+ /* bit positon for extract from acc */
+ __asm__ __volatile__ (
+ "wrdsp %[pos], 1 \n\t"
+ :
+ : [pos] "r" (pos)
+ );
+
+ /* prefetch data to cache memory */
+ vp9_prefetch_load(src);
+ vp9_prefetch_load(src + 32);
+ vp9_prefetch_store(dst);
+
+ switch (w) {
+ case 4:
+ convolve_bi_avg_horiz_4_dspr2(src, src_stride,
+ dst, dst_stride,
+ filter_x, h);
+ break;
+ case 8:
+ convolve_bi_avg_horiz_8_dspr2(src, src_stride,
+ dst, dst_stride,
+ filter_x, h);
+ break;
+ case 16:
+ convolve_bi_avg_horiz_16_dspr2(src, src_stride,
+ dst, dst_stride,
+ filter_x, h, 1);
+ break;
+ case 32:
+ convolve_bi_avg_horiz_16_dspr2(src, src_stride,
+ dst, dst_stride,
+ filter_x, h, 2);
+ break;
+ case 64:
+ vp9_prefetch_load(src + 64);
+ vp9_prefetch_store(dst + 32);
+
+ convolve_bi_avg_horiz_64_dspr2(src, src_stride,
+ dst, dst_stride,
+ filter_x, h);
+ break;
+ default:
+ vp9_convolve8_avg_horiz_c(src, src_stride,
+ dst, dst_stride,
+ filter_x, x_step_q4,
+ filter_y, y_step_q4,
+ w, h);
+ break;
+ }
+ } else {
+ vp9_convolve8_avg_horiz_c(src, src_stride,
+ dst, dst_stride,
+ filter_x, x_step_q4,
+ filter_y, y_step_q4,
+ w, h);
+ }
+}
+#endif
diff --git a/vp9/common/mips/dspr2/vp9_convolve2_dspr2.c b/vp9/common/mips/dspr2/vp9_convolve2_dspr2.c
new file mode 100644
index 0000000..bc422bc
--- /dev/null
+++ b/vp9/common/mips/dspr2/vp9_convolve2_dspr2.c
@@ -0,0 +1,784 @@
+/*
+ * 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.
+ */
+
+#include <assert.h>
+#include <stdio.h>
+
+#include "./vpx_config.h"
+#include "./vp9_rtcd.h"
+#include "vp9/common/vp9_common.h"
+#include "vpx/vpx_integer.h"
+#include "vpx_ports/mem.h"
+#include "vp9/common/vp9_convolve.h"
+#include "vp9/common/mips/dspr2/vp9_common_dspr2.h"
+
+#if HAVE_DSPR2
+static void convolve_bi_horiz_4_transposed_dspr2(const uint8_t *src,
+ int32_t src_stride,
+ uint8_t *dst,
+ int32_t dst_stride,
+ const int16_t *filter_x0,
+ int32_t h) {
+ int32_t y;
+ uint8_t *cm = vp9_ff_cropTbl;
+ uint8_t *dst_ptr;
+ int32_t Temp1, Temp2;
+ uint32_t vector4a = 64;
+ uint32_t tp1, tp2;
+ uint32_t p1, p2;
+ const int16_t *filter = &filter_x0[3];
+ uint32_t filter45;
+
+ filter45 = ((const int32_t *)filter)[0];
+
+ for (y = h; y--;) {
+ dst_ptr = dst;
+ /* prefetch data to cache memory */
+ vp9_prefetch_load(src + src_stride);
+ vp9_prefetch_load(src + src_stride + 32);
+
+ __asm__ __volatile__ (
+ "ulw %[tp1], 0(%[src]) \n\t"
+ "ulw %[tp2], 4(%[src]) \n\t"
+
+ /* even 1. pixel */
+ "mtlo %[vector4a], $ac3 \n\t"
+ "mthi $zero, $ac3 \n\t"
+ "preceu.ph.qbr %[p1], %[tp1] \n\t"
+ "preceu.ph.qbl %[p2], %[tp1] \n\t"
+ "dpa.w.ph $ac3, %[p1], %[filter45] \n\t"
+ "extp %[Temp1], $ac3, 31 \n\t"
+
+ /* even 2. pixel */
+ "mtlo %[vector4a], $ac2 \n\t"
+ "mthi $zero, $ac2 \n\t"
+ "balign %[tp2], %[tp1], 3 \n\t"
+ "dpa.w.ph $ac2, %[p2], %[filter45] \n\t"
+ "extp %[Temp2], $ac2, 31 \n\t"
+
+ /* odd 1. pixel */
+ "lbux %[tp1], %[Temp1](%[cm]) \n\t"
+ "mtlo %[vector4a], $ac3 \n\t"
+ "mthi $zero, $ac3 \n\t"
+ "preceu.ph.qbr %[p1], %[tp2] \n\t"
+ "preceu.ph.qbl %[p2], %[tp2] \n\t"
+ "dpa.w.ph $ac3, %[p1], %[filter45] \n\t"
+ "extp %[Temp1], $ac3, 31 \n\t"
+
+ /* odd 2. pixel */
+ "lbux %[tp2], %[Temp2](%[cm]) \n\t"
+ "mtlo %[vector4a], $ac2 \n\t"
+ "mthi $zero, $ac2 \n\t"
+ "dpa.w.ph $ac2, %[p2], %[filter45] \n\t"
+ "extp %[Temp2], $ac2, 31 \n\t"
+
+ /* clamp */
+ "lbux %[p1], %[Temp1](%[cm]) \n\t"
+ "lbux %[p2], %[Temp2](%[cm]) \n\t"
+
+ /* store bytes */
+ "sb %[tp1], 0(%[dst_ptr]) \n\t"
+ "addu %[dst_ptr], %[dst_ptr], %[dst_stride] \n\t"
+
+ "sb %[p1], 0(%[dst_ptr]) \n\t"
+ "addu %[dst_ptr], %[dst_ptr], %[dst_stride] \n\t"
+
+ "sb %[tp2], 0(%[dst_ptr]) \n\t"
+ "addu %[dst_ptr], %[dst_ptr], %[dst_stride] \n\t"
+
+ "sb %[p2], 0(%[dst_ptr]) \n\t"
+ "addu %[dst_ptr], %[dst_ptr], %[dst_stride] \n\t"
+
+ : [tp1] "=&r" (tp1), [tp2] "=&r" (tp2),
+ [p1] "=&r" (p1), [p2] "=&r" (p2),
+ [Temp1] "=&r" (Temp1), [Temp2] "=&r" (Temp2),
+ [dst_ptr] "+r" (dst_ptr)
+ : [filter45] "r" (filter45),[vector4a] "r" (vector4a),
+ [cm] "r" (cm), [src] "r" (src), [dst_stride] "r" (dst_stride)
+ );
+
+ /* Next row... */
+ src += src_stride;
+ dst += 1;
+ }
+}
+
+static void convolve_bi_horiz_8_transposed_dspr2(const uint8_t *src,
+ int32_t src_stride,
+ uint8_t *dst,
+ int32_t dst_stride,
+ const int16_t *filter_x0,
+ int32_t h) {
+ int32_t y;
+ uint8_t *cm = vp9_ff_cropTbl;
+ uint8_t *dst_ptr;
+ uint32_t vector4a = 64;
+ int32_t Temp1, Temp2, Temp3;
+ uint32_t tp1, tp2, tp3;
+ uint32_t p1, p2, p3, p4;
+ uint8_t *odd_dst;
+ uint32_t dst_pitch_2 = (dst_stride << 1);
+ const int16_t *filter = &filter_x0[3];
+ uint32_t filter45;
+
+ filter45 = ((const int32_t *)filter)[0];
+
+ for (y = h; y--;) {
+ /* prefetch data to cache memory */
+ vp9_prefetch_load(src + src_stride);
+ vp9_prefetch_load(src + src_stride + 32);
+
+ dst_ptr = dst;
+ odd_dst = (dst_ptr + dst_stride);
+
+ __asm__ __volatile__ (
+ "ulw %[tp1], 0(%[src]) \n\t"
+ "ulw %[tp2], 4(%[src]) \n\t"
+
+ /* even 1. pixel */
+ "mtlo %[vector4a], $ac3 \n\t"
+ "mthi $zero, $ac3 \n\t"
+ "mtlo %[vector4a], $ac2 \n\t"
+ "mthi $zero, $ac2 \n\t"
+ "preceu.ph.qbr %[p1], %[tp1] \n\t"
+ "preceu.ph.qbl %[p2], %[tp1] \n\t"
+ "preceu.ph.qbr %[p3], %[tp2] \n\t"
+ "preceu.ph.qbl %[p4], %[tp2] \n\t"
+ "ulw %[tp3], 8(%[src]) \n\t"
+ "dpa.w.ph $ac3, %[p1], %[filter45] \n\t"
+ "extp %[Temp1], $ac3, 31 \n\t"
+
+ /* even 2. pixel */
+ "dpa.w.ph $ac2, %[p2], %[filter45] \n\t"
+ "extp %[Temp3], $ac2, 31 \n\t"
+
+ /* even 3. pixel */
+ "lbux %[Temp2], %[Temp1](%[cm]) \n\t"
+ "mtlo %[vector4a], $ac1 \n\t"
+ "mthi $zero, $ac1 \n\t"
+ "balign %[tp3], %[tp2], 3 \n\t"
+ "balign %[tp2], %[tp1], 3 \n\t"
+ "dpa.w.ph $ac1, %[p3], %[filter45] \n\t"
+ "lbux %[tp1], %[Temp3](%[cm]) \n\t"
+ "extp %[p3], $ac1, 31 \n\t"
+
+ /* even 4. pixel */
+ "mtlo %[vector4a], $ac2 \n\t"
+ "mthi $zero, $ac2 \n\t"
+ "mtlo %[vector4a], $ac3 \n\t"
+ "mthi $zero, $ac3 \n\t"
+ "sb %[Temp2], 0(%[dst_ptr]) \n\t"
+ "addu %[dst_ptr], %[dst_ptr], %[dst_pitch_2] \n\t"
+ "sb %[tp1], 0(%[dst_ptr]) \n\t"
+ "addu %[dst_ptr], %[dst_ptr], %[dst_pitch_2] \n\t"
+
+ "dpa.w.ph $ac2, %[p4], %[filter45] \n\t"
+ "extp %[Temp3], $ac2, 31 \n\t"
+
+ "lbux %[Temp1], %[p3](%[cm]) \n\t"
+
+ /* odd 1. pixel */
+ "mtlo %[vector4a], $ac1 \n\t"
+ "mthi $zero, $ac1 \n\t"
+ "preceu.ph.qbr %[p1], %[tp2] \n\t"
+ "preceu.ph.qbl %[p2], %[tp2] \n\t"
+ "preceu.ph.qbr %[p3], %[tp3] \n\t"
+ "preceu.ph.qbl %[p4], %[tp3] \n\t"
+ "sb %[Temp1], 0(%[dst_ptr]) \n\t"
+ "addu %[dst_ptr], %[dst_ptr], %[dst_pitch_2] \n\t"
+
+ "dpa.w.ph $ac3, %[p1], %[filter45] \n\t"
+ "extp %[Temp2], $ac3, 31 \n\t"
+
+ /* odd 2. pixel */
+ "lbux %[tp1], %[Temp3](%[cm]) \n\t"
+ "mtlo %[vector4a], $ac3 \n\t"
+ "mthi $zero, $ac3 \n\t"
+ "mtlo %[vector4a], $ac2 \n\t"
+ "mthi $zero, $ac2 \n\t"
+ "dpa.w.ph $ac1, %[p2], %[filter45] \n\t"
+ "sb %[tp1], 0(%[dst_ptr]) \n\t"
+ "addu %[dst_ptr], %[dst_ptr], %[dst_pitch_2] \n\t"
+ "extp %[Temp3], $ac1, 31 \n\t"
+
+ /* odd 3. pixel */
+ "lbux %[tp3], %[Temp2](%[cm]) \n\t"
+ "dpa.w.ph $ac3, %[p3], %[filter45] \n\t"
+ "extp %[Temp2], $ac3, 31 \n\t"
+
+ /* odd 4. pixel */
+ "sb %[tp3], 0(%[odd_dst]) \n\t"
+ "addu %[odd_dst], %[odd_dst], %[dst_pitch_2] \n\t"
+ "dpa.w.ph $ac2, %[p4], %[filter45] \n\t"
+ "extp %[Temp1], $ac2, 31 \n\t"
+
+ /* clamp */
+ "lbux %[p4], %[Temp3](%[cm]) \n\t"
+ "lbux %[p2], %[Temp2](%[cm]) \n\t"
+ "lbux %[p1], %[Temp1](%[cm]) \n\t"
+
+ /* store bytes */
+ "sb %[p4], 0(%[odd_dst]) \n\t"
+ "addu %[odd_dst], %[odd_dst], %[dst_pitch_2] \n\t"
+
+ "sb %[p2], 0(%[odd_dst]) \n\t"
+ "addu %[odd_dst], %[odd_dst], %[dst_pitch_2] \n\t"
+
+ "sb %[p1], 0(%[odd_dst]) \n\t"
+
+ : [tp1] "=&r" (tp1), [tp2] "=&r" (tp2), [tp3] "=&r" (tp3),
+ [p1] "=&r" (p1), [p2] "=&r" (p2), [p3] "=&r" (p3), [p4] "=&r" (p4),
+ [Temp1] "=&r" (Temp1), [Temp2] "=&r" (Temp2), [Temp3] "=&r" (Temp3),
+ [dst_ptr] "+r" (dst_ptr), [odd_dst] "+r" (odd_dst)
+ : [filter45] "r" (filter45),[vector4a] "r" (vector4a), [cm] "r" (cm),
+ [src] "r" (src), [dst_pitch_2] "r" (dst_pitch_2)
+ );
+
+ /* Next row... */
+ src += src_stride;
+ dst += 1;
+ }
+}
+
+static void convolve_bi_horiz_16_transposed_dspr2(const uint8_t *src_ptr,
+ int32_t src_stride,
+ uint8_t *dst_ptr,
+ int32_t dst_stride,
+ const int16_t *filter_x0,
+ int32_t h,
+ int32_t count) {
+ int32_t c, y;
+ const uint8_t *src;
+ uint8_t *dst;
+ uint8_t *cm = vp9_ff_cropTbl;
+ uint32_t vector_64 = 64;
+ int32_t Temp1, Temp2, Temp3;
+ uint32_t qload1, qload2;
+ uint32_t p1, p2, p3, p4, p5;
+ uint32_t st1, st2, st3;
+ uint32_t dst_pitch_2 = (dst_stride << 1);
+ uint8_t *odd_dst;
+ const int16_t *filter = &filter_x0[3];
+ uint32_t filter45;
+
+ filter45 = ((const int32_t *)filter)[0];
+
+ for (y = h; y--;) {
+ /* prefetch data to cache memory */
+ vp9_prefetch_load(src_ptr + src_stride);
+ vp9_prefetch_load(src_ptr + src_stride + 32);
+
+ src = src_ptr;
+ dst = dst_ptr;
+
+ odd_dst = (dst + dst_stride);
+
+ for (c = 0; c < count; c++) {
+ __asm__ __volatile__ (
+ "ulw %[qload1], 0(%[src]) \n\t"
+ "ulw %[qload2], 4(%[src]) \n\t"
+
+ /* even 1. pixel */
+ "mtlo %[vector_64], $ac1 \n\t" /* even 1 */
+ "mthi $zero, $ac1 \n\t"
+ "mtlo %[vector_64], $ac2 \n\t" /* even 2 */
+ "mthi $zero, $ac2 \n\t"
+ "preceu.ph.qbr %[p1], %[qload1] \n\t"
+ "preceu.ph.qbl %[p2], %[qload1] \n\t"
+ "preceu.ph.qbr %[p3], %[qload2] \n\t"
+ "preceu.ph.qbl %[p4], %[qload2] \n\t"
+ "ulw %[qload1], 8(%[src]) \n\t"
+ "dpa.w.ph $ac1, %[p1], %[filter45] \n\t" /* even 1 */
+ "extp %[Temp1], $ac1, 31 \n\t" /* even 1 */
+
+ /* even 2. pixel */
+ "mtlo %[vector_64], $ac3 \n\t" /* even 3 */
+ "mthi $zero, $ac3 \n\t"
+ "preceu.ph.qbr %[p1], %[qload1] \n\t"
+ "preceu.ph.qbl %[p5], %[qload1] \n\t"
+ "ulw %[qload2], 12(%[src]) \n\t"
+ "dpa.w.ph $ac2, %[p2], %[filter45] \n\t" /* even 1 */
+ "lbux %[st1], %[Temp1](%[cm]) \n\t" /* even 1 */
+ "extp %[Temp2], $ac2, 31 \n\t" /* even 1 */
+
+ /* even 3. pixel */
+ "mtlo %[vector_64], $ac1 \n\t" /* even 4 */
+ "mthi $zero, $ac1 \n\t"
+ "preceu.ph.qbr %[p2], %[qload2] \n\t"
+ "sb %[st1], 0(%[dst]) \n\t" /* even 1 */
+ "addu %[dst], %[dst], %[dst_pitch_2] \n\t"
+ "dpa.w.ph $ac3, %[p3], %[filter45] \n\t" /* even 3 */
+ "extp %[Temp3], $ac3, 31 \n\t" /* even 3 */
+ "lbux %[st2], %[Temp2](%[cm]) \n\t" /* even 1 */
+
+ /* even 4. pixel */
+ "mtlo %[vector_64], $ac2 \n\t" /* even 5 */
+ "mthi $zero, $ac2 \n\t"
+ "preceu.ph.qbl %[p3], %[qload2] \n\t"
+ "sb %[st2], 0(%[dst]) \n\t" /* even 2 */
+ "addu %[dst], %[dst], %[dst_pitch_2] \n\t"
+ "dpa.w.ph $ac1, %[p4], %[filter45] \n\t" /* even 4 */
+ "extp %[Temp1], $ac1, 31 \n\t" /* even 4 */
+ "lbux %[st3], %[Temp3](%[cm]) \n\t" /* even 3 */
+
+ /* even 5. pixel */
+ "mtlo %[vector_64], $ac3 \n\t" /* even 6 */
+ "mthi $zero, $ac3 \n\t"
+ "sb %[st3], 0(%[dst]) \n\t" /* even 3 */
+ "addu %[dst], %[dst], %[dst_pitch_2] \n\t"
+ "dpa.w.ph $ac2, %[p1], %[filter45] \n\t" /* even 5 */
+ "extp %[Temp2], $ac2, 31 \n\t" /* even 5 */
+ "lbux %[st1], %[Temp1](%[cm]) \n\t" /* even 4 */
+
+ /* even 6. pixel */
+ "mtlo %[vector_64], $ac1 \n\t" /* even 7 */
+ "mthi $zero, $ac1 \n\t"
+ "sb %[st1], 0(%[dst]) \n\t" /* even 4 */
+ "addu %[dst], %[dst], %[dst_pitch_2] \n\t"
+ "ulw %[qload1], 20(%[src]) \n\t"
+ "dpa.w.ph $ac3, %[p5], %[filter45] \n\t" /* even 6 */
+ "extp %[Temp3], $ac3, 31 \n\t" /* even 6 */
+ "lbux %[st2], %[Temp2](%[cm]) \n\t" /* even 5 */
+
+ /* even 7. pixel */
+ "mtlo %[vector_64], $ac2 \n\t" /* even 8 */
+ "mthi $zero, $ac2 \n\t"
+ "preceu.ph.qbr %[p5], %[qload1] \n\t"
+ "sb %[st2], 0(%[dst]) \n\t" /* even 5 */
+ "addu %[dst], %[dst], %[dst_pitch_2] \n\t"
+ "dpa.w.ph $ac1, %[p2], %[filter45] \n\t" /* even 7 */
+ "extp %[Temp1], $ac1, 31 \n\t" /* even 7 */
+ "lbux %[st3], %[Temp3](%[cm]) \n\t" /* even 6 */
+
+ /* even 8. pixel */
+ "mtlo %[vector_64], $ac3 \n\t" /* odd 1 */
+ "mthi $zero, $ac3 \n\t"
+ "dpa.w.ph $ac2, %[p3], %[filter45] \n\t" /* even 8 */
+ "sb %[st3], 0(%[dst]) \n\t" /* even 6 */
+ "addu %[dst], %[dst], %[dst_pitch_2] \n\t"
+ "extp %[Temp2], $ac2, 31 \n\t" /* even 8 */
+ "lbux %[st1], %[Temp1](%[cm]) \n\t" /* even 7 */
+
+ /* ODD pixels */
+ "ulw %[qload1], 1(%[src]) \n\t"
+ "ulw %[qload2], 5(%[src]) \n\t"
+
+ /* odd 1. pixel */
+ "mtlo %[vector_64], $ac1 \n\t" /* odd 2 */
+ "mthi $zero, $ac1 \n\t"
+ "preceu.ph.qbr %[p1], %[qload1] \n\t"
+ "preceu.ph.qbl %[p2], %[qload1] \n\t"
+ "preceu.ph.qbr %[p3], %[qload2] \n\t"
+ "preceu.ph.qbl %[p4], %[qload2] \n\t"
+ "sb %[st1], 0(%[dst]) \n\t" /* even 7 */
+ "addu %[dst], %[dst], %[dst_pitch_2] \n\t"
+ "ulw %[qload2], 9(%[src]) \n\t"
+ "dpa.w.ph $ac3, %[p1], %[filter45] \n\t" /* odd 1 */
+ "extp %[Temp3], $ac3, 31 \n\t" /* odd 1 */
+ "lbux %[st2], %[Temp2](%[cm]) \n\t" /* even 8 */
+
+ /* odd 2. pixel */
+ "mtlo %[vector_64], $ac2 \n\t" /* odd 3 */
+ "mthi $zero, $ac2 \n\t"
+ "preceu.ph.qbr %[p1], %[qload2] \n\t"
+ "preceu.ph.qbl %[p5], %[qload2] \n\t"
+ "sb %[st2], 0(%[dst]) \n\t" /* even 8 */
+ "ulw %[qload1], 13(%[src]) \n\t"
+ "dpa.w.ph $ac1, %[p2], %[filter45] \n\t" /* odd 2 */
+ "extp %[Temp1], $ac1, 31 \n\t" /* odd 2 */
+ "lbux %[st3], %[Temp3](%[cm]) \n\t" /* odd 1 */
+
+ /* odd 3. pixel */
+ "mtlo %[vector_64], $ac3 \n\t" /* odd 4 */
+ "mthi $zero, $ac3 \n\t"
+ "preceu.ph.qbr %[p2], %[qload1] \n\t"
+ "sb %[st3], 0(%[odd_dst]) \n\t" /* odd 1 */
+ "addu %[odd_dst], %[odd_dst], %[dst_pitch_2] \n\t"
+ "dpa.w.ph $ac2, %[p3], %[filter45] \n\t" /* odd 3 */
+ "extp %[Temp2], $ac2, 31 \n\t" /* odd 3 */
+ "lbux %[st1], %[Temp1](%[cm]) \n\t" /* odd 2 */
+
+ /* odd 4. pixel */
+ "mtlo %[vector_64], $ac1 \n\t" /* odd 5 */
+ "mthi $zero, $ac1 \n\t"
+ "preceu.ph.qbl %[p3], %[qload1] \n\t"
+ "sb %[st1], 0(%[odd_dst]) \n\t" /* odd 2 */
+ "addu %[odd_dst], %[odd_dst], %[dst_pitch_2] \n\t"
+ "dpa.w.ph $ac3, %[p4], %[filter45] \n\t" /* odd 4 */
+ "extp %[Temp3], $ac3, 31 \n\t" /* odd 4 */
+ "lbux %[st2], %[Temp2](%[cm]) \n\t" /* odd 3 */
+
+ /* odd 5. pixel */
+ "mtlo %[vector_64], $ac2 \n\t" /* odd 6 */
+ "mthi $zero, $ac2 \n\t"
+ "sb %[st2], 0(%[odd_dst]) \n\t" /* odd 3 */
+ "addu %[odd_dst], %[odd_dst], %[dst_pitch_2] \n\t"
+ "dpa.w.ph $ac1, %[p1], %[filter45] \n\t" /* odd 5 */
+ "extp %[Temp1], $ac1, 31 \n\t" /* odd 5 */
+ "lbux %[st3], %[Temp3](%[cm]) \n\t" /* odd 4 */
+
+ /* odd 6. pixel */
+ "mtlo %[vector_64], $ac3 \n\t" /* odd 7 */
+ "mthi $zero, $ac3 \n\t"
+ "sb %[st3], 0(%[odd_dst]) \n\t" /* odd 4 */
+ "addu %[odd_dst], %[odd_dst], %[dst_pitch_2] \n\t"
+ "ulw %[qload1], 21(%[src]) \n\t"
+ "dpa.w.ph $ac2, %[p5], %[filter45] \n\t" /* odd 6 */
+ "extp %[Temp2], $ac2, 31 \n\t" /* odd 6 */
+ "lbux %[st1], %[Temp1](%[cm]) \n\t" /* odd 5 */
+
+ /* odd 7. pixel */
+ "mtlo %[vector_64], $ac1 \n\t" /* odd 8 */
+ "mthi $zero, $ac1 \n\t"
+ "preceu.ph.qbr %[p5], %[qload1] \n\t"
+ "sb %[st1], 0(%[odd_dst]) \n\t" /* odd 5 */
+ "addu %[odd_dst], %[odd_dst], %[dst_pitch_2] \n\t"
+ "dpa.w.ph $ac3, %[p2], %[filter45] \n\t" /* odd 7 */
+ "extp %[Temp3], $ac3, 31 \n\t" /* odd 7 */
+
+ /* odd 8. pixel */
+ "dpa.w.ph $ac1, %[p3], %[filter45] \n\t" /* odd 8 */
+ "extp %[Temp1], $ac1, 31 \n\t" /* odd 8 */
+
+ "lbux %[st2], %[Temp2](%[cm]) \n\t" /* odd 6 */
+ "lbux %[st3], %[Temp3](%[cm]) \n\t" /* odd 7 */
+ "lbux %[st1], %[Temp1](%[cm]) \n\t" /* odd 8 */
+
+ "sb %[st2], 0(%[odd_dst]) \n\t" /* odd 6 */
+ "addu %[odd_dst], %[odd_dst], %[dst_pitch_2] \n\t"
+
+ "sb %[st3], 0(%[odd_dst]) \n\t" /* odd 7 */
+ "addu %[odd_dst], %[odd_dst], %[dst_pitch_2] \n\t"
+
+ "sb %[st1], 0(%[odd_dst]) \n\t" /* odd 8 */
+
+ : [qload1] "=&r" (qload1), [qload2] "=&r" (qload2), [p5] "=&r" (p5),
+ [st1] "=&r" (st1), [st2] "=&r" (st2), [st3] "=&r" (st3),
+ [p1] "=&r" (p1), [p2] "=&r" (p2), [p3] "=&r" (p3), [p4] "=&r" (p4),
+ [Temp1] "=&r" (Temp1), [Temp2] "=&r" (Temp2), [Temp3] "=&r" (Temp3),
+ [dst] "+r" (dst), [odd_dst] "+r" (odd_dst)
+ : [filter45] "r" (filter45), [vector_64] "r" (vector_64),
+ [cm] "r" (cm),
+ [src] "r" (src), [dst_pitch_2] "r" (dst_pitch_2)
+ );
+
+ src += 16;
+ dst = (dst_ptr + ((c + 1) * 16 * dst_stride));
+ odd_dst = (dst + dst_stride);
+ }
+
+ /* Next row... */
+ src_ptr += src_stride;
+ dst_ptr += 1;
+ }
+}
+
+static void convolve_bi_horiz_64_transposed_dspr2(const uint8_t *src_ptr,
+ int32_t src_stride,
+ uint8_t *dst_ptr,
+ int32_t dst_stride,
+ const int16_t *filter_x0,
+ int32_t h) {
+ int32_t c, y;
+ const uint8_t *src;
+ uint8_t *dst;
+ uint8_t *cm = vp9_ff_cropTbl;
+ uint32_t vector_64 = 64;
+ int32_t Temp1, Temp2, Temp3;
+ uint32_t qload1, qload2;
+ uint32_t p1, p2, p3, p4, p5;
+ uint32_t st1, st2, st3;
+ uint32_t dst_pitch_2 = (dst_stride << 1);
+ uint8_t *odd_dst;
+ const int16_t *filter = &filter_x0[3];
+ uint32_t filter45;
+
+ filter45 = ((const int32_t *)filter)[0];
+
+ for (y = h; y--;) {
+ /* prefetch data to cache memory */
+ vp9_prefetch_load(src_ptr + src_stride);
+ vp9_prefetch_load(src_ptr + src_stride + 32);
+ vp9_prefetch_load(src_ptr + src_stride + 64);
+
+ src = src_ptr;
+ dst = dst_ptr;
+
+ odd_dst = (dst + dst_stride);
+
+ for (c = 0; c < 4; c++) {
+ __asm__ __volatile__ (
+ "ulw %[qload1], 0(%[src]) \n\t"
+ "ulw %[qload2], 4(%[src]) \n\t"
+
+ /* even 1. pixel */
+ "mtlo %[vector_64], $ac1 \n\t" /* even 1 */
+ "mthi $zero, $ac1 \n\t"
+ "mtlo %[vector_64], $ac2 \n\t" /* even 2 */
+ "mthi $zero, $ac2 \n\t"
+ "preceu.ph.qbr %[p1], %[qload1] \n\t"
+ "preceu.ph.qbl %[p2], %[qload1] \n\t"
+ "preceu.ph.qbr %[p3], %[qload2] \n\t"
+ "preceu.ph.qbl %[p4], %[qload2] \n\t"
+ "ulw %[qload1], 8(%[src]) \n\t"
+ "dpa.w.ph $ac1, %[p1], %[filter45] \n\t" /* even 1 */
+ "extp %[Temp1], $ac1, 31 \n\t" /* even 1 */
+
+ /* even 2. pixel */
+ "mtlo %[vector_64], $ac3 \n\t" /* even 3 */
+ "mthi $zero, $ac3 \n\t"
+ "preceu.ph.qbr %[p1], %[qload1] \n\t"
+ "preceu.ph.qbl %[p5], %[qload1] \n\t"
+ "ulw %[qload2], 12(%[src]) \n\t"
+ "dpa.w.ph $ac2, %[p2], %[filter45] \n\t" /* even 1 */
+ "lbux %[st1], %[Temp1](%[cm]) \n\t" /* even 1 */
+ "extp %[Temp2], $ac2, 31 \n\t" /* even 1 */
+
+ /* even 3. pixel */
+ "mtlo %[vector_64], $ac1 \n\t" /* even 4 */
+ "mthi $zero, $ac1 \n\t"
+ "preceu.ph.qbr %[p2], %[qload2] \n\t"
+ "sb %[st1], 0(%[dst]) \n\t" /* even 1 */
+ "addu %[dst], %[dst], %[dst_pitch_2] \n\t"
+ "dpa.w.ph $ac3, %[p3], %[filter45] \n\t" /* even 3 */
+ "extp %[Temp3], $ac3, 31 \n\t" /* even 3 */
+ "lbux %[st2], %[Temp2](%[cm]) \n\t" /* even 1 */
+
+ /* even 4. pixel */
+ "mtlo %[vector_64], $ac2 \n\t" /* even 5 */
+ "mthi $zero, $ac2 \n\t"
+ "preceu.ph.qbl %[p3], %[qload2] \n\t"
+ "sb %[st2], 0(%[dst]) \n\t" /* even 2 */
+ "addu %[dst], %[dst], %[dst_pitch_2] \n\t"
+ "dpa.w.ph $ac1, %[p4], %[filter45] \n\t" /* even 4 */
+ "extp %[Temp1], $ac1, 31 \n\t" /* even 4 */
+ "lbux %[st3], %[Temp3](%[cm]) \n\t" /* even 3 */
+
+ /* even 5. pixel */
+ "mtlo %[vector_64], $ac3 \n\t" /* even 6 */
+ "mthi $zero, $ac3 \n\t"
+ "sb %[st3], 0(%[dst]) \n\t" /* even 3 */
+ "addu %[dst], %[dst], %[dst_pitch_2] \n\t"
+ "dpa.w.ph $ac2, %[p1], %[filter45] \n\t" /* even 5 */
+ "extp %[Temp2], $ac2, 31 \n\t" /* even 5 */
+ "lbux %[st1], %[Temp1](%[cm]) \n\t" /* even 4 */
+
+ /* even 6. pixel */
+ "mtlo %[vector_64], $ac1 \n\t" /* even 7 */
+ "mthi $zero, $ac1 \n\t"
+ "sb %[st1], 0(%[dst]) \n\t" /* even 4 */
+ "addu %[dst], %[dst], %[dst_pitch_2] \n\t"
+ "ulw %[qload1], 20(%[src]) \n\t"
+ "dpa.w.ph $ac3, %[p5], %[filter45] \n\t" /* even 6 */
+ "extp %[Temp3], $ac3, 31 \n\t" /* even 6 */
+ "lbux %[st2], %[Temp2](%[cm]) \n\t" /* even 5 */
+
+ /* even 7. pixel */
+ "mtlo %[vector_64], $ac2 \n\t" /* even 8 */
+ "mthi $zero, $ac2 \n\t"
+ "preceu.ph.qbr %[p5], %[qload1] \n\t"
+ "sb %[st2], 0(%[dst]) \n\t" /* even 5 */
+ "addu %[dst], %[dst], %[dst_pitch_2] \n\t"
+ "dpa.w.ph $ac1, %[p2], %[filter45] \n\t" /* even 7 */
+ "extp %[Temp1], $ac1, 31 \n\t" /* even 7 */
+ "lbux %[st3], %[Temp3](%[cm]) \n\t" /* even 6 */
+
+ /* even 8. pixel */
+ "mtlo %[vector_64], $ac3 \n\t" /* odd 1 */
+ "mthi $zero, $ac3 \n\t"
+ "dpa.w.ph $ac2, %[p3], %[filter45] \n\t" /* even 8 */
+ "sb %[st3], 0(%[dst]) \n\t" /* even 6 */
+ "addu %[dst], %[dst], %[dst_pitch_2] \n\t"
+ "extp %[Temp2], $ac2, 31 \n\t" /* even 8 */
+ "lbux %[st1], %[Temp1](%[cm]) \n\t" /* even 7 */
+
+ /* ODD pixels */
+ "ulw %[qload1], 1(%[src]) \n\t"
+ "ulw %[qload2], 5(%[src]) \n\t"
+
+ /* odd 1. pixel */
+ "mtlo %[vector_64], $ac1 \n\t" /* odd 2 */
+ "mthi $zero, $ac1 \n\t"
+ "preceu.ph.qbr %[p1], %[qload1] \n\t"
+ "preceu.ph.qbl %[p2], %[qload1] \n\t"
+ "preceu.ph.qbr %[p3], %[qload2] \n\t"
+ "preceu.ph.qbl %[p4], %[qload2] \n\t"
+ "sb %[st1], 0(%[dst]) \n\t" /* even 7 */
+ "addu %[dst], %[dst], %[dst_pitch_2] \n\t"
+ "ulw %[qload2], 9(%[src]) \n\t"
+ "dpa.w.ph $ac3, %[p1], %[filter45] \n\t" /* odd 1 */
+ "extp %[Temp3], $ac3, 31 \n\t" /* odd 1 */
+ "lbux %[st2], %[Temp2](%[cm]) \n\t" /* even 8 */
+
+ /* odd 2. pixel */
+ "mtlo %[vector_64], $ac2 \n\t" /* odd 3 */
+ "mthi $zero, $ac2 \n\t"
+ "preceu.ph.qbr %[p1], %[qload2] \n\t"
+ "preceu.ph.qbl %[p5], %[qload2] \n\t"
+ "sb %[st2], 0(%[dst]) \n\t" /* even 8 */
+ "ulw %[qload1], 13(%[src]) \n\t"
+ "dpa.w.ph $ac1, %[p2], %[filter45] \n\t" /* odd 2 */
+ "extp %[Temp1], $ac1, 31 \n\t" /* odd 2 */
+ "lbux %[st3], %[Temp3](%[cm]) \n\t" /* odd 1 */
+
+ /* odd 3. pixel */
+ "mtlo %[vector_64], $ac3 \n\t" /* odd 4 */
+ "mthi $zero, $ac3 \n\t"
+ "preceu.ph.qbr %[p2], %[qload1] \n\t"
+ "sb %[st3], 0(%[odd_dst]) \n\t" /* odd 1 */
+ "addu %[odd_dst], %[odd_dst], %[dst_pitch_2] \n\t"
+ "dpa.w.ph $ac2, %[p3], %[filter45] \n\t" /* odd 3 */
+ "extp %[Temp2], $ac2, 31 \n\t" /* odd 3 */
+ "lbux %[st1], %[Temp1](%[cm]) \n\t" /* odd 2 */
+
+ /* odd 4. pixel */
+ "mtlo %[vector_64], $ac1 \n\t" /* odd 5 */
+ "mthi $zero, $ac1 \n\t"
+ "preceu.ph.qbl %[p3], %[qload1] \n\t"
+ "sb %[st1], 0(%[odd_dst]) \n\t" /* odd 2 */
+ "addu %[odd_dst], %[odd_dst], %[dst_pitch_2] \n\t"
+ "dpa.w.ph $ac3, %[p4], %[filter45] \n\t" /* odd 4 */
+ "extp %[Temp3], $ac3, 31 \n\t" /* odd 4 */
+ "lbux %[st2], %[Temp2](%[cm]) \n\t" /* odd 3 */
+
+ /* odd 5. pixel */
+ "mtlo %[vector_64], $ac2 \n\t" /* odd 6 */
+ "mthi $zero, $ac2 \n\t"
+ "sb %[st2], 0(%[odd_dst]) \n\t" /* odd 3 */
+ "addu %[odd_dst], %[odd_dst], %[dst_pitch_2] \n\t"
+ "dpa.w.ph $ac1, %[p1], %[filter45] \n\t" /* odd 5 */
+ "extp %[Temp1], $ac1, 31 \n\t" /* odd 5 */
+ "lbux %[st3], %[Temp3](%[cm]) \n\t" /* odd 4 */
+
+ /* odd 6. pixel */
+ "mtlo %[vector_64], $ac3 \n\t" /* odd 7 */
+ "mthi $zero, $ac3 \n\t"
+ "sb %[st3], 0(%[odd_dst]) \n\t" /* odd 4 */
+ "addu %[odd_dst], %[odd_dst], %[dst_pitch_2] \n\t"
+ "ulw %[qload1], 21(%[src]) \n\t"
+ "dpa.w.ph $ac2, %[p5], %[filter45] \n\t" /* odd 6 */
+ "extp %[Temp2], $ac2, 31 \n\t" /* odd 6 */
+ "lbux %[st1], %[Temp1](%[cm]) \n\t" /* odd 5 */
+
+ /* odd 7. pixel */
+ "mtlo %[vector_64], $ac1 \n\t" /* odd 8 */
+ "mthi $zero, $ac1 \n\t"
+ "preceu.ph.qbr %[p5], %[qload1] \n\t"
+ "sb %[st1], 0(%[odd_dst]) \n\t" /* odd 5 */
+ "addu %[odd_dst], %[odd_dst], %[dst_pitch_2] \n\t"
+ "dpa.w.ph $ac3, %[p2], %[filter45] \n\t" /* odd 7 */
+ "extp %[Temp3], $ac3, 31 \n\t" /* odd 7 */
+
+ /* odd 8. pixel */
+ "dpa.w.ph $ac1, %[p3], %[filter45] \n\t" /* odd 8 */
+ "extp %[Temp1], $ac1, 31 \n\t" /* odd 8 */
+
+ "lbux %[st2], %[Temp2](%[cm]) \n\t" /* odd 6 */
+ "lbux %[st3], %[Temp3](%[cm]) \n\t" /* odd 7 */
+ "lbux %[st1], %[Temp1](%[cm]) \n\t" /* odd 8 */
+
+ "sb %[st2], 0(%[odd_dst]) \n\t" /* odd 6 */
+ "addu %[odd_dst], %[odd_dst], %[dst_pitch_2] \n\t"
+
+ "sb %[st3], 0(%[odd_dst]) \n\t" /* odd 7 */
+ "addu %[odd_dst], %[odd_dst], %[dst_pitch_2] \n\t"
+
+ "sb %[st1], 0(%[odd_dst]) \n\t" /* odd 8 */
+
+ : [qload1] "=&r" (qload1), [qload2] "=&r" (qload2), [p5] "=&r" (p5),
+ [st1] "=&r" (st1), [st2] "=&r" (st2), [st3] "=&r" (st3),
+ [p1] "=&r" (p1), [p2] "=&r" (p2), [p3] "=&r" (p3), [p4] "=&r" (p4),
+ [Temp1] "=&r" (Temp1), [Temp2] "=&r" (Temp2), [Temp3] "=&r" (Temp3),
+ [dst] "+r" (dst), [odd_dst] "+r" (odd_dst)
+ : [filter45] "r" (filter45), [vector_64] "r" (vector_64),
+ [cm] "r" (cm),
+ [src] "r" (src), [dst_pitch_2] "r" (dst_pitch_2)
+ );
+
+ src += 16;
+ dst = (dst_ptr + ((c + 1) * 16 * dst_stride));
+ odd_dst = (dst + dst_stride);
+ }
+
+ /* Next row... */
+ src_ptr += src_stride;
+ dst_ptr += 1;
+ }
+}
+
+void convolve_bi_horiz_transposed(const uint8_t *src, ptrdiff_t src_stride,
+ uint8_t *dst, ptrdiff_t dst_stride,
+ const int16_t *filter, int w, int h) {
+ int x, y;
+
+ for (y = 0; y < h; ++y) {
+ for (x = 0; x < w; ++x) {
+ int sum = 0;
+
+ sum += src[x] * filter[3];
+ sum += src[x + 1] * filter[4];
+
+ dst[x * dst_stride] = clip_pixel(ROUND_POWER_OF_TWO(sum, FILTER_BITS));
+ }
+
+ src += src_stride;
+ dst += 1;
+ }
+}
+
+void vp9_convolve2_dspr2(const uint8_t *src, ptrdiff_t src_stride,
+ uint8_t *dst, ptrdiff_t dst_stride,
+ const int16_t *filter,
+ int w, int h) {
+ uint32_t pos = 38;
+
+ /* bit positon for extract from acc */
+ __asm__ __volatile__ (
+ "wrdsp %[pos], 1 \n\t"
+ :
+ : [pos] "r" (pos)
+ );
+
+ /* prefetch data to cache memory */
+ vp9_prefetch_load(src);
+ vp9_prefetch_load(src + 32);
+
+ switch (w) {
+ case 4:
+ convolve_bi_horiz_4_transposed_dspr2(src, src_stride,
+ dst, dst_stride,
+ filter, h);
+ break;
+ case 8:
+ convolve_bi_horiz_8_transposed_dspr2(src, src_stride,
+ dst, dst_stride,
+ filter, h);
+ break;
+ case 16:
+ case 32:
+ convolve_bi_horiz_16_transposed_dspr2(src, src_stride,
+ dst, dst_stride,
+ filter, h,
+ (w/16));
+ break;
+ case 64:
+ vp9_prefetch_load(src + 32);
+ convolve_bi_horiz_64_transposed_dspr2(src, src_stride,
+ dst, dst_stride,
+ filter, h);
+ break;
+ default:
+ convolve_bi_horiz_transposed(src, src_stride,
+ dst, dst_stride,
+ filter, w, h);
+ break;
+ }
+}
+#endif
diff --git a/vp9/common/mips/dspr2/vp9_convolve2_horiz_dspr2.c b/vp9/common/mips/dspr2/vp9_convolve2_horiz_dspr2.c
new file mode 100644
index 0000000..1debdb4
--- /dev/null
+++ b/vp9/common/mips/dspr2/vp9_convolve2_horiz_dspr2.c
@@ -0,0 +1,713 @@
+/*
+ * 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.
+ */
+
+#include <assert.h>
+#include <stdio.h>
+
+#include "./vpx_config.h"
+#include "./vp9_rtcd.h"
+#include "vp9/common/vp9_common.h"
+#include "vpx/vpx_integer.h"
+#include "vpx_ports/mem.h"
+#include "vp9/common/vp9_convolve.h"
+#include "vp9/common/mips/dspr2/vp9_common_dspr2.h"
+
+#if HAVE_DSPR2
+static void convolve_bi_horiz_4_dspr2(const uint8_t *src,
+ int32_t src_stride,
+ uint8_t *dst,
+ int32_t dst_stride,
+ const int16_t *filter_x0,
+ int32_t h) {
+ int32_t y;
+ uint8_t *cm = vp9_ff_cropTbl;
+ int32_t Temp1, Temp2, Temp3, Temp4;
+ uint32_t vector4a = 64;
+ uint32_t tp1, tp2;
+ uint32_t p1, p2;
+ const int16_t *filter = &filter_x0[3];
+ uint32_t filter45;;
+
+ filter45 = ((const int32_t *)filter)[0];
+
+ for (y = h; y--;) {
+ /* prefetch data to cache memory */
+ vp9_prefetch_load(src + src_stride);
+ vp9_prefetch_load(src + src_stride + 32);
+ vp9_prefetch_store(dst + dst_stride);
+
+ __asm__ __volatile__ (
+ "ulw %[tp1], 0(%[src]) \n\t"
+ "ulw %[tp2], 4(%[src]) \n\t"
+
+ /* even 1. pixel */
+ "mtlo %[vector4a], $ac3 \n\t"
+ "mthi $zero, $ac3 \n\t"
+ "preceu.ph.qbr %[p1], %[tp1] \n\t"
+ "preceu.ph.qbl %[p2], %[tp1] \n\t"
+ "dpa.w.ph $ac3, %[p1], %[filter45] \n\t"
+ "extp %[Temp1], $ac3, 31 \n\t"
+
+ /* even 2. pixel */
+ "mtlo %[vector4a], $ac2 \n\t"
+ "mthi $zero, $ac2 \n\t"
+ "balign %[tp2], %[tp1], 3 \n\t"
+ "dpa.w.ph $ac2, %[p2], %[filter45] \n\t"
+ "extp %[Temp3], $ac2, 31 \n\t"
+
+ /* odd 1. pixel */
+ "lbux %[tp1], %[Temp1](%[cm]) \n\t"
+ "mtlo %[vector4a], $ac3 \n\t"
+ "mthi $zero, $ac3 \n\t"
+ "preceu.ph.qbr %[p1], %[tp2] \n\t"
+ "preceu.ph.qbl %[p2], %[tp2] \n\t"
+ "dpa.w.ph $ac3, %[p1], %[filter45] \n\t"
+ "extp %[Temp2], $ac3, 31 \n\t"
+
+ /* odd 2. pixel */
+ "lbux %[tp2], %[Temp3](%[cm]) \n\t"
+ "mtlo %[vector4a], $ac2 \n\t"
+ "mthi $zero, $ac2 \n\t"
+ "dpa.w.ph $ac2, %[p2], %[filter45] \n\t"
+ "extp %[Temp4], $ac2, 31 \n\t"
+
+ /* clamp */
+ "lbux %[p1], %[Temp2](%[cm]) \n\t"
+ "lbux %[p2], %[Temp4](%[cm]) \n\t"
+
+ /* store bytes */
+ "sb %[tp1], 0(%[dst]) \n\t"
+ "sb %[p1], 1(%[dst]) \n\t"
+ "sb %[tp2], 2(%[dst]) \n\t"
+ "sb %[p2], 3(%[dst]) \n\t"
+
+ : [tp1] "=&r" (tp1), [tp2] "=&r" (tp2),
+ [p1] "=&r" (p1), [p2] "=&r" (p2),
+ [Temp1] "=&r" (Temp1), [Temp2] "=&r" (Temp2),
+ [Temp3] "=&r" (Temp3), [Temp4] "=&r" (Temp4)
+ : [filter45] "r" (filter45), [vector4a] "r" (vector4a),
+ [cm] "r" (cm), [dst] "r" (dst), [src] "r" (src)
+ );
+
+ /* Next row... */
+ src += src_stride;
+ dst += dst_stride;
+ }
+}
+
+static void convolve_bi_horiz_8_dspr2(const uint8_t *src,
+ int32_t src_stride,
+ uint8_t *dst,
+ int32_t dst_stride,
+ const int16_t *filter_x0,
+ int32_t h) {
+ int32_t y;
+ uint8_t *cm = vp9_ff_cropTbl;
+ uint32_t vector4a = 64;
+ int32_t Temp1, Temp2, Temp3;
+ uint32_t tp1, tp2, tp3;
+ uint32_t p1, p2, p3, p4;
+ uint32_t st0, st1;
+ const int16_t *filter = &filter_x0[3];
+ uint32_t filter45;;
+
+ filter45 = ((const int32_t *)filter)[0];
+
+ for (y = h; y--;) {
+ /* prefetch data to cache memory */
+ vp9_prefetch_load(src + src_stride);
+ vp9_prefetch_load(src + src_stride + 32);
+ vp9_prefetch_store(dst + dst_stride);
+
+ __asm__ __volatile__ (
+ "ulw %[tp1], 0(%[src]) \n\t"
+ "ulw %[tp2], 4(%[src]) \n\t"
+
+ /* even 1. pixel */
+ "mtlo %[vector4a], $ac3 \n\t"
+ "mthi $zero, $ac3 \n\t"
+ "mtlo %[vector4a], $ac2 \n\t"
+ "mthi $zero, $ac2 \n\t"
+ "preceu.ph.qbr %[p1], %[tp1] \n\t"
+ "preceu.ph.qbl %[p2], %[tp1] \n\t"
+ "preceu.ph.qbr %[p3], %[tp2] \n\t"
+ "preceu.ph.qbl %[p4], %[tp2] \n\t"
+ "ulw %[tp3], 8(%[src]) \n\t"
+ "dpa.w.ph $ac3, %[p1], %[filter45] \n\t"
+ "extp %[Temp1], $ac3, 31 \n\t"
+
+ /* even 2. pixel */
+ "dpa.w.ph $ac2, %[p2], %[filter45] \n\t"
+ "extp %[Temp3], $ac2, 31 \n\t"
+
+ /* even 3. pixel */
+ "lbux %[st0], %[Temp1](%[cm]) \n\t"
+ "mtlo %[vector4a], $ac1 \n\t"
+ "mthi $zero, $ac1 \n\t"
+ "dpa.w.ph $ac1, %[p3], %[filter45] \n\t"
+ "extp %[Temp1], $ac1, 31 \n\t"
+
+ /* even 4. pixel */
+ "mtlo %[vector4a], $ac2 \n\t"
+ "mthi $zero, $ac2 \n\t"
+ "mtlo %[vector4a], $ac3 \n\t"
+ "mthi $zero, $ac3 \n\t"
+ "sb %[st0], 0(%[dst]) \n\t"
+ "lbux %[st1], %[Temp3](%[cm]) \n\t"
+
+ "balign %[tp3], %[tp2], 3 \n\t"
+ "balign %[tp2], %[tp1], 3 \n\t"
+
+ "dpa.w.ph $ac2, %[p4], %[filter45] \n\t"
+ "extp %[Temp3], $ac2, 31 \n\t"
+
+ "lbux %[st0], %[Temp1](%[cm]) \n\t"
+
+ /* odd 1. pixel */
+ "mtlo %[vector4a], $ac1 \n\t"
+ "mthi $zero, $ac1 \n\t"
+ "sb %[st1], 2(%[dst]) \n\t"
+ "preceu.ph.qbr %[p1], %[tp2] \n\t"
+ "preceu.ph.qbl %[p2], %[tp2] \n\t"
+ "preceu.ph.qbr %[p3], %[tp3] \n\t"
+ "preceu.ph.qbl %[p4], %[tp3] \n\t"
+ "sb %[st0], 4(%[dst]) \n\t"
+ "dpa.w.ph $ac3, %[p1], %[filter45] \n\t"
+ "extp %[Temp2], $ac3, 31 \n\t"
+
+ /* odd 2. pixel */
+ "mtlo %[vector4a], $ac3 \n\t"
+ "mthi $zero, $ac3 \n\t"
+ "mtlo %[vector4a], $ac2 \n\t"
+ "mthi $zero, $ac2 \n\t"
+ "lbux %[st0], %[Temp3](%[cm]) \n\t"
+ "dpa.w.ph $ac1, %[p2], %[filter45] \n\t"
+ "extp %[Temp3], $ac1, 31 \n\t"
+
+ /* odd 3. pixel */
+ "lbux %[st1], %[Temp2](%[cm]) \n\t"
+ "dpa.w.ph $ac3, %[p3], %[filter45] \n\t"
+ "extp %[Temp2], $ac3, 31 \n\t"
+
+ /* odd 4. pixel */
+ "sb %[st1], 1(%[dst]) \n\t"
+ "sb %[st0], 6(%[dst]) \n\t"
+ "dpa.w.ph $ac2, %[p4], %[filter45] \n\t"
+ "extp %[Temp1], $ac2, 31 \n\t"
+
+ /* clamp */
+ "lbux %[p4], %[Temp3](%[cm]) \n\t"
+ "lbux %[p2], %[Temp2](%[cm]) \n\t"
+ "lbux %[p1], %[Temp1](%[cm]) \n\t"
+
+ /* store bytes */
+ "sb %[p4], 3(%[dst]) \n\t"
+ "sb %[p2], 5(%[dst]) \n\t"
+ "sb %[p1], 7(%[dst]) \n\t"
+
+ : [tp1] "=&r" (tp1), [tp2] "=&r" (tp2), [tp3] "=&r" (tp3),
+ [st0] "=&r" (st0), [st1] "=&r" (st1),
+ [p1] "=&r" (p1), [p2] "=&r" (p2), [p3] "=&r" (p3), [p4] "=&r" (p4),
+ [Temp1] "=&r" (Temp1), [Temp2] "=&r" (Temp2), [Temp3] "=&r" (Temp3)
+ : [filter45] "r" (filter45), [vector4a] "r" (vector4a),
+ [cm] "r" (cm), [dst] "r" (dst), [src] "r" (src)
+ );
+
+ /* Next row... */
+ src += src_stride;
+ dst += dst_stride;
+ }
+}
+
+static void convolve_bi_horiz_16_dspr2(const uint8_t *src_ptr,
+ int32_t src_stride,
+ uint8_t *dst_ptr,
+ int32_t dst_stride,
+ const int16_t *filter_x0,
+ int32_t h,
+ int32_t count) {
+ int32_t y, c;
+ const uint8_t *src;
+ uint8_t *dst;
+ uint8_t *cm = vp9_ff_cropTbl;
+ uint32_t vector_64 = 64;
+ int32_t Temp1, Temp2, Temp3;
+ uint32_t qload1, qload2, qload3;
+ uint32_t p1, p2, p3, p4, p5;
+ uint32_t st1, st2, st3;
+ const int16_t *filter = &filter_x0[3];
+ uint32_t filter45;;
+
+ filter45 = ((const int32_t *)filter)[0];
+
+ for (y = h; y--;) {
+ src = src_ptr;
+ dst = dst_ptr;
+
+ /* prefetch data to cache memory */
+ vp9_prefetch_load(src_ptr + src_stride);
+ vp9_prefetch_load(src_ptr + src_stride + 32);
+ vp9_prefetch_store(dst_ptr + dst_stride);
+
+ for (c = 0; c < count; c++) {
+ __asm__ __volatile__ (
+ "ulw %[qload1], 0(%[src]) \n\t"
+ "ulw %[qload2], 4(%[src]) \n\t"
+
+ /* even 1. pixel */
+ "mtlo %[vector_64], $ac1 \n\t" /* even 1 */
+ "mthi $zero, $ac1 \n\t"
+ "mtlo %[vector_64], $ac2 \n\t" /* even 2 */
+ "mthi $zero, $ac2 \n\t"
+ "preceu.ph.qbr %[p1], %[qload1] \n\t"
+ "preceu.ph.qbl %[p2], %[qload1] \n\t"
+ "preceu.ph.qbr %[p3], %[qload2] \n\t"
+ "preceu.ph.qbl %[p4], %[qload2] \n\t"
+ "ulw %[qload3], 8(%[src]) \n\t"
+ "dpa.w.ph $ac1, %[p1], %[filter45] \n\t" /* even 1 */
+ "extp %[Temp1], $ac1, 31 \n\t" /* even 1 */
+
+ /* even 2. pixel */
+ "mtlo %[vector_64], $ac3 \n\t" /* even 3 */
+ "mthi $zero, $ac3 \n\t"
+ "preceu.ph.qbr %[p1], %[qload3] \n\t"
+ "preceu.ph.qbl %[p5], %[qload3] \n\t"
+ "ulw %[qload1], 12(%[src]) \n\t"
+ "dpa.w.ph $ac2, %[p2], %[filter45] \n\t" /* even 1 */
+ "extp %[Temp2], $ac2, 31 \n\t" /* even 1 */
+ "lbux %[st1], %[Temp1](%[cm]) \n\t" /* even 1 */
+
+ /* even 3. pixel */
+ "mtlo %[vector_64], $ac1 \n\t" /* even 4 */
+ "mthi $zero, $ac1 \n\t"
+ "preceu.ph.qbr %[p2], %[qload1] \n\t"
+ "sb %[st1], 0(%[dst]) \n\t" /* even 1 */
+ "dpa.w.ph $ac3, %[p3], %[filter45] \n\t" /* even 3 */
+ "extp %[Temp3], $ac3, 31 \n\t" /* even 3 */
+ "lbux %[st2], %[Temp2](%[cm]) \n\t" /* even 1 */
+
+ /* even 4. pixel */
+ "mtlo %[vector_64], $ac2 \n\t" /* even 5 */
+ "mthi $zero, $ac2 \n\t"
+ "preceu.ph.qbl %[p3], %[qload1] \n\t"
+ "sb %[st2], 2(%[dst]) \n\t" /* even 1 */
+ "dpa.w.ph $ac1, %[p4], %[filter45] \n\t" /* even 4 */
+ "extp %[Temp1], $ac1, 31 \n\t" /* even 4 */
+ "lbux %[st3], %[Temp3](%[cm]) \n\t" /* even 3 */
+
+ /* even 5. pixel */
+ "mtlo %[vector_64], $ac3 \n\t" /* even 6 */
+ "mthi $zero, $ac3 \n\t"
+ "sb %[st3], 4(%[dst]) \n\t" /* even 3 */
+ "dpa.w.ph $ac2, %[p1], %[filter45] \n\t" /* even 5 */
+ "extp %[Temp2], $ac2, 31 \n\t" /* even 5 */
+ "lbux %[st1], %[Temp1](%[cm]) \n\t" /* even 4 */
+
+ /* even 6. pixel */
+ "mtlo %[vector_64], $ac1 \n\t" /* even 7 */
+ "mthi $zero, $ac1 \n\t"
+ "sb %[st1], 6(%[dst]) \n\t" /* even 4 */
+ "dpa.w.ph $ac3, %[p5], %[filter45] \n\t" /* even 6 */
+ "extp %[Temp3], $ac3, 31 \n\t" /* even 6 */
+ "lbux %[st2], %[Temp2](%[cm]) \n\t" /* even 5 */
+
+ /* even 7. pixel */
+ "mtlo %[vector_64], $ac2 \n\t" /* even 8 */
+ "mthi $zero, $ac2 \n\t"
+ "sb %[st2], 8(%[dst]) \n\t" /* even 5 */
+ "dpa.w.ph $ac1, %[p2], %[filter45] \n\t" /* even 7 */
+ "extp %[Temp1], $ac1, 31 \n\t" /* even 7 */
+ "lbux %[st3], %[Temp3](%[cm]) \n\t" /* even 6 */
+
+ /* even 8. pixel */
+ "mtlo %[vector_64], $ac3 \n\t" /* odd 1 */
+ "mthi $zero, $ac3 \n\t"
+ "dpa.w.ph $ac2, %[p3], %[filter45] \n\t" /* even 8 */
+ "sb %[st3], 10(%[dst]) \n\t" /* even 6 */
+ "extp %[Temp2], $ac2, 31 \n\t" /* even 8 */
+ "lbux %[st1], %[Temp1](%[cm]) \n\t" /* even 7 */
+
+ /* ODD pixels */
+ "ulw %[qload1], 1(%[src]) \n\t"
+ "ulw %[qload2], 5(%[src]) \n\t"
+
+ /* odd 1. pixel */
+ "mtlo %[vector_64], $ac1 \n\t" /* odd 2 */
+ "mthi $zero, $ac1 \n\t"
+ "preceu.ph.qbr %[p1], %[qload1] \n\t"
+ "preceu.ph.qbl %[p2], %[qload1] \n\t"
+ "preceu.ph.qbr %[p3], %[qload2] \n\t"
+ "preceu.ph.qbl %[p4], %[qload2] \n\t"
+ "sb %[st1], 12(%[dst]) \n\t" /* even 7 */
+ "ulw %[qload3], 9(%[src]) \n\t"
+ "dpa.w.ph $ac3, %[p1], %[filter45] \n\t" /* odd 1 */
+ "extp %[Temp3], $ac3, 31 \n\t" /* odd 1 */
+ "lbux %[st2], %[Temp2](%[cm]) \n\t" /* even 8 */
+
+ /* odd 2. pixel */
+ "mtlo %[vector_64], $ac2 \n\t" /* odd 3 */
+ "mthi $zero, $ac2 \n\t"
+ "preceu.ph.qbr %[p1], %[qload3] \n\t"
+ "preceu.ph.qbl %[p5], %[qload3] \n\t"
+ "sb %[st2], 14(%[dst]) \n\t" /* even 8 */
+ "ulw %[qload1], 13(%[src]) \n\t"
+ "dpa.w.ph $ac1, %[p2], %[filter45] \n\t" /* odd 2 */
+ "extp %[Temp1], $ac1, 31 \n\t" /* odd 2 */
+ "lbux %[st3], %[Temp3](%[cm]) \n\t" /* odd 1 */
+
+ /* odd 3. pixel */
+ "mtlo %[vector_64], $ac3 \n\t" /* odd 4 */
+ "mthi $zero, $ac3 \n\t"
+ "preceu.ph.qbr %[p2], %[qload1] \n\t"
+ "sb %[st3], 1(%[dst]) \n\t" /* odd 1 */
+ "dpa.w.ph $ac2, %[p3], %[filter45] \n\t" /* odd 3 */
+ "extp %[Temp2], $ac2, 31 \n\t" /* odd 3 */
+ "lbux %[st1], %[Temp1](%[cm]) \n\t" /* odd 2 */
+
+ /* odd 4. pixel */
+ "mtlo %[vector_64], $ac1 \n\t" /* odd 5 */
+ "mthi $zero, $ac1 \n\t"
+ "preceu.ph.qbl %[p3], %[qload1] \n\t"
+ "sb %[st1], 3(%[dst]) \n\t" /* odd 2 */
+ "dpa.w.ph $ac3, %[p4], %[filter45] \n\t" /* odd 4 */
+ "extp %[Temp3], $ac3, 31 \n\t" /* odd 4 */
+ "lbux %[st2], %[Temp2](%[cm]) \n\t" /* odd 3 */
+
+ /* odd 5. pixel */
+ "mtlo %[vector_64], $ac2 \n\t" /* odd 6 */
+ "mthi $zero, $ac2 \n\t"
+ "sb %[st2], 5(%[dst]) \n\t" /* odd 3 */
+ "dpa.w.ph $ac1, %[p1], %[filter45] \n\t" /* odd 5 */
+ "extp %[Temp1], $ac1, 31 \n\t" /* odd 5 */
+ "lbux %[st3], %[Temp3](%[cm]) \n\t" /* odd 4 */
+
+ /* odd 6. pixel */
+ "mtlo %[vector_64], $ac3 \n\t" /* odd 7 */
+ "mthi $zero, $ac3 \n\t"
+ "sb %[st3], 7(%[dst]) \n\t" /* odd 4 */
+ "dpa.w.ph $ac2, %[p5], %[filter45] \n\t" /* odd 6 */
+ "extp %[Temp2], $ac2, 31 \n\t" /* odd 6 */
+ "lbux %[st1], %[Temp1](%[cm]) \n\t" /* odd 5 */
+
+ /* odd 7. pixel */
+ "mtlo %[vector_64], $ac1 \n\t" /* odd 8 */
+ "mthi $zero, $ac1 \n\t"
+ "sb %[st1], 9(%[dst]) \n\t" /* odd 5 */
+ "dpa.w.ph $ac3, %[p2], %[filter45] \n\t" /* odd 7 */
+ "extp %[Temp3], $ac3, 31 \n\t" /* odd 7 */
+
+ /* odd 8. pixel */
+ "dpa.w.ph $ac1, %[p3], %[filter45] \n\t" /* odd 8 */
+ "extp %[Temp1], $ac1, 31 \n\t" /* odd 8 */
+
+ "lbux %[st2], %[Temp2](%[cm]) \n\t" /* odd 6 */
+ "lbux %[st3], %[Temp3](%[cm]) \n\t" /* odd 7 */
+ "lbux %[st1], %[Temp1](%[cm]) \n\t" /* odd 8 */
+
+ "sb %[st2], 11(%[dst]) \n\t" /* odd 6 */
+ "sb %[st3], 13(%[dst]) \n\t" /* odd 7 */
+ "sb %[st1], 15(%[dst]) \n\t" /* odd 8 */
+
+ : [qload1] "=&r" (qload1), [qload2] "=&r" (qload2), [qload3] "=&r" (qload3),
+ [st1] "=&r" (st1), [st2] "=&r" (st2), [st3] "=&r" (st3),
+ [p1] "=&r" (p1), [p2] "=&r" (p2), [p3] "=&r" (p3), [p4] "=&r" (p4),
+ [p5] "=&r" (p5),
+ [Temp1] "=&r" (Temp1), [Temp2] "=&r" (Temp2), [Temp3] "=&r" (Temp3)
+ : [filter45] "r" (filter45), [vector_64] "r" (vector_64),
+ [cm] "r" (cm), [dst] "r" (dst), [src] "r" (src)
+ );
+
+ src += 16;
+ dst += 16;
+ }
+
+ /* Next row... */
+ src_ptr += src_stride;
+ dst_ptr += dst_stride;
+ }
+}
+
+static void convolve_bi_horiz_64_dspr2(const uint8_t *src_ptr,
+ int32_t src_stride,
+ uint8_t *dst_ptr,
+ int32_t dst_stride,
+ const int16_t *filter_x0,
+ int32_t h) {
+ int32_t y, c;
+ const uint8_t *src;
+ uint8_t *dst;
+ uint8_t *cm = vp9_ff_cropTbl;
+ uint32_t vector_64 = 64;
+ int32_t Temp1, Temp2, Temp3;
+ uint32_t qload1, qload2, qload3;
+ uint32_t p1, p2, p3, p4, p5;
+ uint32_t st1, st2, st3;
+ const int16_t *filter = &filter_x0[3];
+ uint32_t filter45;;
+
+ filter45 = ((const int32_t *)filter)[0];
+
+ for (y = h; y--;) {
+ src = src_ptr;
+ dst = dst_ptr;
+
+ /* prefetch data to cache memory */
+ vp9_prefetch_load(src_ptr + src_stride);
+ vp9_prefetch_load(src_ptr + src_stride + 32);
+ vp9_prefetch_load(src_ptr + src_stride + 64);
+ vp9_prefetch_store(dst_ptr + dst_stride);
+ vp9_prefetch_store(dst_ptr + dst_stride + 32);
+
+ for (c = 0; c < 4; c++) {
+ __asm__ __volatile__ (
+ "ulw %[qload1], 0(%[src]) \n\t"
+ "ulw %[qload2], 4(%[src]) \n\t"
+
+ /* even 1. pixel */
+ "mtlo %[vector_64], $ac1 \n\t" /* even 1 */
+ "mthi $zero, $ac1 \n\t"
+ "mtlo %[vector_64], $ac2 \n\t" /* even 2 */
+ "mthi $zero, $ac2 \n\t"
+ "preceu.ph.qbr %[p1], %[qload1] \n\t"
+ "preceu.ph.qbl %[p2], %[qload1] \n\t"
+ "preceu.ph.qbr %[p3], %[qload2] \n\t"
+ "preceu.ph.qbl %[p4], %[qload2] \n\t"
+ "ulw %[qload3], 8(%[src]) \n\t"
+ "dpa.w.ph $ac1, %[p1], %[filter45] \n\t" /* even 1 */
+ "extp %[Temp1], $ac1, 31 \n\t" /* even 1 */
+
+ /* even 2. pixel */
+ "mtlo %[vector_64], $ac3 \n\t" /* even 3 */
+ "mthi $zero, $ac3 \n\t"
+ "preceu.ph.qbr %[p1], %[qload3] \n\t"
+ "preceu.ph.qbl %[p5], %[qload3] \n\t"
+ "ulw %[qload1], 12(%[src]) \n\t"
+ "dpa.w.ph $ac2, %[p2], %[filter45] \n\t" /* even 1 */
+ "extp %[Temp2], $ac2, 31 \n\t" /* even 1 */
+ "lbux %[st1], %[Temp1](%[cm]) \n\t" /* even 1 */
+
+ /* even 3. pixel */
+ "mtlo %[vector_64], $ac1 \n\t" /* even 4 */
+ "mthi $zero, $ac1 \n\t"
+ "preceu.ph.qbr %[p2], %[qload1] \n\t"
+ "sb %[st1], 0(%[dst]) \n\t" /* even 1 */
+ "dpa.w.ph $ac3, %[p3], %[filter45] \n\t" /* even 3 */
+ "extp %[Temp3], $ac3, 31 \n\t" /* even 3 */
+ "lbux %[st2], %[Temp2](%[cm]) \n\t" /* even 1 */
+
+ /* even 4. pixel */
+ "mtlo %[vector_64], $ac2 \n\t" /* even 5 */
+ "mthi $zero, $ac2 \n\t"
+ "preceu.ph.qbl %[p3], %[qload1] \n\t"
+ "sb %[st2], 2(%[dst]) \n\t" /* even 1 */
+ "dpa.w.ph $ac1, %[p4], %[filter45] \n\t" /* even 4 */
+ "extp %[Temp1], $ac1, 31 \n\t" /* even 4 */
+ "lbux %[st3], %[Temp3](%[cm]) \n\t" /* even 3 */
+
+ /* even 5. pixel */
+ "mtlo %[vector_64], $ac3 \n\t" /* even 6 */
+ "mthi $zero, $ac3 \n\t"
+ "sb %[st3], 4(%[dst]) \n\t" /* even 3 */
+ "dpa.w.ph $ac2, %[p1], %[filter45] \n\t" /* even 5 */
+ "extp %[Temp2], $ac2, 31 \n\t" /* even 5 */
+ "lbux %[st1], %[Temp1](%[cm]) \n\t" /* even 4 */
+
+ /* even 6. pixel */
+ "mtlo %[vector_64], $ac1 \n\t" /* even 7 */
+ "mthi $zero, $ac1 \n\t"
+ "sb %[st1], 6(%[dst]) \n\t" /* even 4 */
+ "dpa.w.ph $ac3, %[p5], %[filter45] \n\t" /* even 6 */
+ "extp %[Temp3], $ac3, 31 \n\t" /* even 6 */
+ "lbux %[st2], %[Temp2](%[cm]) \n\t" /* even 5 */
+
+ /* even 7. pixel */
+ "mtlo %[vector_64], $ac2 \n\t" /* even 8 */
+ "mthi $zero, $ac2 \n\t"
+ "sb %[st2], 8(%[dst]) \n\t" /* even 5 */
+ "dpa.w.ph $ac1, %[p2], %[filter45] \n\t" /* even 7 */
+ "extp %[Temp1], $ac1, 31 \n\t" /* even 7 */
+ "lbux %[st3], %[Temp3](%[cm]) \n\t" /* even 6 */
+
+ /* even 8. pixel */
+ "mtlo %[vector_64], $ac3 \n\t" /* odd 1 */
+ "mthi $zero, $ac3 \n\t"
+ "dpa.w.ph $ac2, %[p3], %[filter45] \n\t" /* even 8 */
+ "sb %[st3], 10(%[dst]) \n\t" /* even 6 */
+ "extp %[Temp2], $ac2, 31 \n\t" /* even 8 */
+ "lbux %[st1], %[Temp1](%[cm]) \n\t" /* even 7 */
+
+ /* ODD pixels */
+ "ulw %[qload1], 1(%[src]) \n\t"
+ "ulw %[qload2], 5(%[src]) \n\t"
+
+ /* odd 1. pixel */
+ "mtlo %[vector_64], $ac1 \n\t" /* odd 2 */
+ "mthi $zero, $ac1 \n\t"
+ "preceu.ph.qbr %[p1], %[qload1] \n\t"
+ "preceu.ph.qbl %[p2], %[qload1] \n\t"
+ "preceu.ph.qbr %[p3], %[qload2] \n\t"
+ "preceu.ph.qbl %[p4], %[qload2] \n\t"
+ "sb %[st1], 12(%[dst]) \n\t" /* even 7 */
+ "ulw %[qload3], 9(%[src]) \n\t"
+ "dpa.w.ph $ac3, %[p1], %[filter45] \n\t" /* odd 1 */
+ "extp %[Temp3], $ac3, 31 \n\t" /* odd 1 */
+ "lbux %[st2], %[Temp2](%[cm]) \n\t" /* even 8 */
+
+ /* odd 2. pixel */
+ "mtlo %[vector_64], $ac2 \n\t" /* odd 3 */
+ "mthi $zero, $ac2 \n\t"
+ "preceu.ph.qbr %[p1], %[qload3] \n\t"
+ "preceu.ph.qbl %[p5], %[qload3] \n\t"
+ "sb %[st2], 14(%[dst]) \n\t" /* even 8 */
+ "ulw %[qload1], 13(%[src]) \n\t"
+ "dpa.w.ph $ac1, %[p2], %[filter45] \n\t" /* odd 2 */
+ "extp %[Temp1], $ac1, 31 \n\t" /* odd 2 */
+ "lbux %[st3], %[Temp3](%[cm]) \n\t" /* odd 1 */
+
+ /* odd 3. pixel */
+ "mtlo %[vector_64], $ac3 \n\t" /* odd 4 */
+ "mthi $zero, $ac3 \n\t"
+ "preceu.ph.qbr %[p2], %[qload1] \n\t"
+ "sb %[st3], 1(%[dst]) \n\t" /* odd 1 */
+ "dpa.w.ph $ac2, %[p3], %[filter45] \n\t" /* odd 3 */
+ "extp %[Temp2], $ac2, 31 \n\t" /* odd 3 */
+ "lbux %[st1], %[Temp1](%[cm]) \n\t" /* odd 2 */
+
+ /* odd 4. pixel */
+ "mtlo %[vector_64], $ac1 \n\t" /* odd 5 */
+ "mthi $zero, $ac1 \n\t"
+ "preceu.ph.qbl %[p3], %[qload1] \n\t"
+ "sb %[st1], 3(%[dst]) \n\t" /* odd 2 */
+ "dpa.w.ph $ac3, %[p4], %[filter45] \n\t" /* odd 4 */
+ "extp %[Temp3], $ac3, 31 \n\t" /* odd 4 */
+ "lbux %[st2], %[Temp2](%[cm]) \n\t" /* odd 3 */
+
+ /* odd 5. pixel */
+ "mtlo %[vector_64], $ac2 \n\t" /* odd 6 */
+ "mthi $zero, $ac2 \n\t"
+ "sb %[st2], 5(%[dst]) \n\t" /* odd 3 */
+ "dpa.w.ph $ac1, %[p1], %[filter45] \n\t" /* odd 5 */
+ "extp %[Temp1], $ac1, 31 \n\t" /* odd 5 */
+ "lbux %[st3], %[Temp3](%[cm]) \n\t" /* odd 4 */
+
+ /* odd 6. pixel */
+ "mtlo %[vector_64], $ac3 \n\t" /* odd 7 */
+ "mthi $zero, $ac3 \n\t"
+ "sb %[st3], 7(%[dst]) \n\t" /* odd 4 */
+ "dpa.w.ph $ac2, %[p5], %[filter45] \n\t" /* odd 6 */
+ "extp %[Temp2], $ac2, 31 \n\t" /* odd 6 */
+ "lbux %[st1], %[Temp1](%[cm]) \n\t" /* odd 5 */
+
+ /* odd 7. pixel */
+ "mtlo %[vector_64], $ac1 \n\t" /* odd 8 */
+ "mthi $zero, $ac1 \n\t"
+ "sb %[st1], 9(%[dst]) \n\t" /* odd 5 */
+ "dpa.w.ph $ac3, %[p2], %[filter45] \n\t" /* odd 7 */
+ "extp %[Temp3], $ac3, 31 \n\t" /* odd 7 */
+
+ /* odd 8. pixel */
+ "dpa.w.ph $ac1, %[p3], %[filter45] \n\t" /* odd 8 */
+ "extp %[Temp1], $ac1, 31 \n\t" /* odd 8 */
+
+ "lbux %[st2], %[Temp2](%[cm]) \n\t" /* odd 6 */
+ "lbux %[st3], %[Temp3](%[cm]) \n\t" /* odd 7 */
+ "lbux %[st1], %[Temp1](%[cm]) \n\t" /* odd 8 */
+
+ "sb %[st2], 11(%[dst]) \n\t" /* odd 6 */
+ "sb %[st3], 13(%[dst]) \n\t" /* odd 7 */
+ "sb %[st1], 15(%[dst]) \n\t" /* odd 8 */
+
+ : [qload1] "=&r" (qload1), [qload2] "=&r" (qload2), [qload3] "=&r" (qload3),
+ [st1] "=&r" (st1), [st2] "=&r" (st2), [st3] "=&r" (st3),
+ [p1] "=&r" (p1), [p2] "=&r" (p2), [p3] "=&r" (p3), [p4] "=&r" (p4),
+ [p5] "=&r" (p5),
+ [Temp1] "=&r" (Temp1), [Temp2] "=&r" (Temp2), [Temp3] "=&r" (Temp3)
+ : [filter45] "r" (filter45), [vector_64] "r" (vector_64),
+ [cm] "r" (cm), [dst] "r" (dst), [src] "r" (src)
+ );
+
+ src += 16;
+ dst += 16;
+ }
+
+ /* Next row... */
+ src_ptr += src_stride;
+ dst_ptr += dst_stride;
+ }
+}
+
+void vp9_convolve2_horiz_dspr2(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) {
+ if (16 == x_step_q4) {
+ uint32_t pos = 38;
+
+ vp9_prefetch_load((const uint8_t *)filter_x);
+
+ /* bit positon for extract from acc */
+ __asm__ __volatile__ (
+ "wrdsp %[pos], 1 \n\t"
+ :
+ : [pos] "r" (pos)
+ );
+
+ /* prefetch data to cache memory */
+ vp9_prefetch_load(src);
+ vp9_prefetch_load(src + 32);
+ vp9_prefetch_store(dst);
+
+ switch (w) {
+ case 4:
+ convolve_bi_horiz_4_dspr2(src, (int32_t)src_stride,
+ dst, (int32_t)dst_stride,
+ filter_x, (int32_t)h);
+ break;
+ case 8:
+ convolve_bi_horiz_8_dspr2(src, (int32_t)src_stride,
+ dst, (int32_t)dst_stride,
+ filter_x, (int32_t)h);
+ break;
+ case 16:
+ convolve_bi_horiz_16_dspr2(src, (int32_t)src_stride,
+ dst, (int32_t)dst_stride,
+ filter_x, (int32_t)h, 1);
+ break;
+ case 32:
+ convolve_bi_horiz_16_dspr2(src, (int32_t)src_stride,
+ dst, (int32_t)dst_stride,
+ filter_x, (int32_t)h, 2);
+ break;
+ case 64:
+ vp9_prefetch_load(src + 64);
+ vp9_prefetch_store(dst + 32);
+
+ convolve_bi_horiz_64_dspr2(src, (int32_t)src_stride,
+ dst, (int32_t)dst_stride,
+ filter_x, (int32_t)h);
+ break;
+ default:
+ vp9_convolve8_horiz_c(src, src_stride,
+ dst, dst_stride,
+ filter_x, x_step_q4,
+ filter_y, y_step_q4,
+ w, h);
+ break;
+ }
+ } else {
+ vp9_convolve8_horiz_c(src, src_stride,
+ dst, dst_stride,
+ filter_x, x_step_q4,
+ filter_y, y_step_q4,
+ w, h);
+ }
+}
+#endif
diff --git a/vp9/common/mips/dspr2/vp9_convolve2_vert_dspr2.c b/vp9/common/mips/dspr2/vp9_convolve2_vert_dspr2.c
new file mode 100644
index 0000000..8eb105c
--- /dev/null
+++ b/vp9/common/mips/dspr2/vp9_convolve2_vert_dspr2.c
@@ -0,0 +1,266 @@
+/*
+ * 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.
+ */
+
+#include <assert.h>
+#include <stdio.h>
+
+#include "./vpx_config.h"
+#include "./vp9_rtcd.h"
+#include "vp9/common/vp9_common.h"
+#include "vpx/vpx_integer.h"
+#include "vpx_ports/mem.h"
+#include "vp9/common/vp9_convolve.h"
+#include "vp9/common/mips/dspr2/vp9_common_dspr2.h"
+
+#if HAVE_DSPR2
+static void convolve_bi_vert_4_dspr2(const uint8_t *src,
+ int32_t src_stride,
+ uint8_t *dst,
+ int32_t dst_stride,
+ const int16_t *filter_y,
+ int32_t w,
+ int32_t h) {
+ int32_t x, y;
+ const uint8_t *src_ptr;
+ uint8_t *dst_ptr;
+ uint8_t *cm = vp9_ff_cropTbl;
+ uint32_t vector4a = 64;
+ uint32_t load1, load2;
+ uint32_t p1, p2;
+ uint32_t scratch1;
+ uint32_t store1, store2;
+ int32_t Temp1, Temp2;
+ const int16_t *filter = &filter_y[3];
+ uint32_t filter45;
+
+ filter45 = ((const int32_t *)filter)[0];
+
+ for (y = h; y--;) {
+ /* prefetch data to cache memory */
+ vp9_prefetch_store(dst + dst_stride);
+
+ for (x = 0; x < w; x += 4) {
+ src_ptr = src + x;
+ dst_ptr = dst + x;
+
+ __asm__ __volatile__ (
+ "ulw %[load1], 0(%[src_ptr]) \n\t"
+ "add %[src_ptr], %[src_ptr], %[src_stride] \n\t"
+ "ulw %[load2], 0(%[src_ptr]) \n\t"
+
+ "mtlo %[vector4a], $ac0 \n\t"
+ "mtlo %[vector4a], $ac1 \n\t"
+ "mtlo %[vector4a], $ac2 \n\t"
+ "mtlo %[vector4a], $ac3 \n\t"
+ "mthi $zero, $ac0 \n\t"
+ "mthi $zero, $ac1 \n\t"
+ "mthi $zero, $ac2 \n\t"
+ "mthi $zero, $ac3 \n\t"
+
+ "preceu.ph.qbr %[scratch1], %[load1] \n\t"
+ "preceu.ph.qbr %[p1], %[load2] \n\t"
+
+ "precrq.ph.w %[p2], %[p1], %[scratch1] \n\t" /* pixel 2 */
+ "append %[p1], %[scratch1], 16 \n\t" /* pixel 1 */
+
+ "dpa.w.ph $ac0, %[p1], %[filter45] \n\t"
+ "dpa.w.ph $ac1, %[p2], %[filter45] \n\t"
+
+ "preceu.ph.qbl %[scratch1], %[load1] \n\t"
+ "preceu.ph.qbl %[p1], %[load2] \n\t"
+
+ "precrq.ph.w %[p2], %[p1], %[scratch1] \n\t" /* pixel 2 */
+ "append %[p1], %[scratch1], 16 \n\t" /* pixel 1 */
+
+ "dpa.w.ph $ac2, %[p1], %[filter45] \n\t"
+ "dpa.w.ph $ac3, %[p2], %[filter45] \n\t"
+
+ "extp %[Temp1], $ac0, 31 \n\t"
+ "extp %[Temp2], $ac1, 31 \n\t"
+
+ "lbux %[store1], %[Temp1](%[cm]) \n\t"
+ "extp %[Temp1], $ac2, 31 \n\t"
+
+ "lbux %[store2], %[Temp2](%[cm]) \n\t"
+ "extp %[Temp2], $ac3, 31 \n\t"
+
+ "sb %[store1], 0(%[dst_ptr]) \n\t"
+ "sb %[store2], 1(%[dst_ptr]) \n\t"
+
+ "lbux %[store1], %[Temp1](%[cm]) \n\t"
+ "lbux %[store2], %[Temp2](%[cm]) \n\t"
+
+ "sb %[store1], 2(%[dst_ptr]) \n\t"
+ "sb %[store2], 3(%[dst_ptr]) \n\t"
+
+ : [load1] "=&r" (load1), [load2] "=&r" (load2),
+ [p1] "=&r" (p1), [p2] "=&r" (p2),
+ [scratch1] "=&r" (scratch1),
+ [Temp1] "=&r" (Temp1), [Temp2] "=&r" (Temp2),
+ [store1] "=&r" (store1), [store2] "=&r" (store2),
+ [src_ptr] "+r" (src_ptr)
+ : [filter45] "r" (filter45),[vector4a] "r" (vector4a),
+ [src_stride] "r" (src_stride),
+ [cm] "r" (cm), [dst_ptr] "r" (dst_ptr)
+ );
+ }
+
+ /* Next row... */
+ src += src_stride;
+ dst += dst_stride;
+ }
+}
+
+static void convolve_bi_vert_64_dspr2(const uint8_t *src,
+ int32_t src_stride,
+ uint8_t *dst,
+ int32_t dst_stride,
+ const int16_t *filter_y,
+ int32_t h) {
+ int32_t x, y;
+ const uint8_t *src_ptr;
+ uint8_t *dst_ptr;
+ uint8_t *cm = vp9_ff_cropTbl;
+ uint32_t vector4a = 64;
+ uint32_t load1, load2;
+ uint32_t p1, p2;
+ uint32_t scratch1;
+ uint32_t store1, store2;
+ int32_t Temp1, Temp2;
+ const int16_t *filter = &filter_y[3];
+ uint32_t filter45;
+
+ filter45 = ((const int32_t *)filter)[0];
+
+ for (y = h; y--;) {
+ /* prefetch data to cache memory */
+ vp9_prefetch_store(dst + dst_stride);
+
+ for (x = 0; x < 64; x += 4) {
+ src_ptr = src + x;
+ dst_ptr = dst + x;
+
+ __asm__ __volatile__ (
+ "ulw %[load1], 0(%[src_ptr]) \n\t"
+ "add %[src_ptr], %[src_ptr], %[src_stride] \n\t"
+ "ulw %[load2], 0(%[src_ptr]) \n\t"
+
+ "mtlo %[vector4a], $ac0 \n\t"
+ "mtlo %[vector4a], $ac1 \n\t"
+ "mtlo %[vector4a], $ac2 \n\t"
+ "mtlo %[vector4a], $ac3 \n\t"
+ "mthi $zero, $ac0 \n\t"
+ "mthi $zero, $ac1 \n\t"
+ "mthi $zero, $ac2 \n\t"
+ "mthi $zero, $ac3 \n\t"
+
+ "preceu.ph.qbr %[scratch1], %[load1] \n\t"
+ "preceu.ph.qbr %[p1], %[load2] \n\t"
+
+ "precrq.ph.w %[p2], %[p1], %[scratch1] \n\t" /* pixel 2 */
+ "append %[p1], %[scratch1], 16 \n\t" /* pixel 1 */
+
+ "dpa.w.ph $ac0, %[p1], %[filter45] \n\t"
+ "dpa.w.ph $ac1, %[p2], %[filter45] \n\t"
+
+ "preceu.ph.qbl %[scratch1], %[load1] \n\t"
+ "preceu.ph.qbl %[p1], %[load2] \n\t"
+
+ "precrq.ph.w %[p2], %[p1], %[scratch1] \n\t" /* pixel 2 */
+ "append %[p1], %[scratch1], 16 \n\t" /* pixel 1 */
+
+ "dpa.w.ph $ac2, %[p1], %[filter45] \n\t"
+ "dpa.w.ph $ac3, %[p2], %[filter45] \n\t"
+
+ "extp %[Temp1], $ac0, 31 \n\t"
+ "extp %[Temp2], $ac1, 31 \n\t"
+
+ "lbux %[store1], %[Temp1](%[cm]) \n\t"
+ "extp %[Temp1], $ac2, 31 \n\t"
+
+ "lbux %[store2], %[Temp2](%[cm]) \n\t"
+ "extp %[Temp2], $ac3, 31 \n\t"
+
+ "sb %[store1], 0(%[dst_ptr]) \n\t"
+ "sb %[store2], 1(%[dst_ptr]) \n\t"
+
+ "lbux %[store1], %[Temp1](%[cm]) \n\t"
+ "lbux %[store2], %[Temp2](%[cm]) \n\t"
+
+ "sb %[store1], 2(%[dst_ptr]) \n\t"
+ "sb %[store2], 3(%[dst_ptr]) \n\t"
+
+ : [load1] "=&r" (load1), [load2] "=&r" (load2),
+ [p1] "=&r" (p1), [p2] "=&r" (p2),
+ [scratch1] "=&r" (scratch1),
+ [Temp1] "=&r" (Temp1), [Temp2] "=&r" (Temp2),
+ [store1] "=&r" (store1), [store2] "=&r" (store2),
+ [src_ptr] "+r" (src_ptr)
+ : [filter45] "r" (filter45),[vector4a] "r" (vector4a),
+ [src_stride] "r" (src_stride),
+ [cm] "r" (cm), [dst_ptr] "r" (dst_ptr)
+ );
+ }
+
+ /* Next row... */
+ src += src_stride;
+ dst += dst_stride;
+ }
+}
+
+void vp9_convolve2_vert_dspr2(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) {
+ if (16 == y_step_q4) {
+ uint32_t pos = 38;
+
+ /* bit positon for extract from acc */
+ __asm__ __volatile__ (
+ "wrdsp %[pos], 1 \n\t"
+ :
+ : [pos] "r" (pos)
+ );
+
+ vp9_prefetch_store(dst);
+
+ switch (w) {
+ case 4 :
+ case 8 :
+ case 16 :
+ case 32 :
+ convolve_bi_vert_4_dspr2(src, src_stride,
+ dst, dst_stride,
+ filter_y, w, h);
+ break;
+ case 64 :
+ vp9_prefetch_store(dst + 32);
+ convolve_bi_vert_64_dspr2(src, src_stride,
+ dst, dst_stride,
+ filter_y, h);
+ break;
+ default:
+ vp9_convolve8_vert_c(src, src_stride,
+ dst, dst_stride,
+ filter_x, x_step_q4,
+ filter_y, y_step_q4,
+ w, h);
+ break;
+ }
+ } else {
+ vp9_convolve8_vert_c(src, src_stride,
+ dst, dst_stride,
+ filter_x, x_step_q4,
+ filter_y, y_step_q4,
+ w, h);
+ }
+}
+#endif
diff --git a/vp9/common/mips/dspr2/vp9_convolve8_avg_dspr2.c b/vp9/common/mips/dspr2/vp9_convolve8_avg_dspr2.c
index 0930ad1..da7f0fd 100644
--- a/vp9/common/mips/dspr2/vp9_convolve8_avg_dspr2.c
+++ b/vp9/common/mips/dspr2/vp9_convolve8_avg_dspr2.c
@@ -355,6 +355,12 @@
filter_x, x_step_q4,
filter_y, y_step_q4,
w, h);
+ } else if (((const int32_t *)filter_y)[0] == 0) {
+ vp9_convolve2_avg_vert_dspr2(src, src_stride,
+ dst, dst_stride,
+ filter_x, x_step_q4,
+ filter_y, y_step_q4,
+ w, h);
} else {
if (16 == y_step_q4) {
uint32_t pos = 38;
diff --git a/vp9/common/mips/dspr2/vp9_convolve8_avg_horiz_dspr2.c b/vp9/common/mips/dspr2/vp9_convolve8_avg_horiz_dspr2.c
index 37c665b..69da1cf 100644
--- a/vp9/common/mips/dspr2/vp9_convolve8_avg_horiz_dspr2.c
+++ b/vp9/common/mips/dspr2/vp9_convolve8_avg_horiz_dspr2.c
@@ -965,6 +965,12 @@
filter_x, x_step_q4,
filter_y, y_step_q4,
w, h);
+ } else if (((const int32_t *)filter_x)[0] == 0) {
+ vp9_convolve2_avg_horiz_dspr2(src, src_stride,
+ dst, dst_stride,
+ filter_x, x_step_q4,
+ filter_y, y_step_q4,
+ w, h);
} else {
if (16 == x_step_q4) {
uint32_t pos = 38;
diff --git a/vp9/common/mips/dspr2/vp9_convolve8_dspr2.c b/vp9/common/mips/dspr2/vp9_convolve8_dspr2.c
index 2c48bd0..126e05a 100644
--- a/vp9/common/mips/dspr2/vp9_convolve8_dspr2.c
+++ b/vp9/common/mips/dspr2/vp9_convolve8_dspr2.c
@@ -930,6 +930,21 @@
}
}
+void copy_horiz_transposed(const uint8_t *src, ptrdiff_t src_stride,
+ uint8_t *dst, ptrdiff_t dst_stride,
+ int w, int h) {
+ int x, y;
+
+ for (y = 0; y < h; ++y) {
+ for (x = 0; x < w; ++x) {
+ dst[x * dst_stride] = src[x];
+ }
+
+ src += src_stride;
+ dst += 1;
+ }
+}
+
void vp9_convolve8_dspr2(const uint8_t *src, ptrdiff_t src_stride,
uint8_t *dst, ptrdiff_t dst_stride,
const int16_t *filter_x, int x_step_q4,
@@ -966,20 +981,14 @@
/* copy the src to dst */
if (filter_x[3] == 0x80) {
- int32_t y;
- int32_t c;
- const uint8_t *src_ptr = src - src_stride * 3;
- uint8_t *dst_ptr = temp;
-
- for (y = intermediate_height; y--;) {
- for (c = 0; c < w; c++) {
- dst_ptr[c * intermediate_height] = src_ptr[c];
- }
-
- /* next row... */
- src_ptr += src_stride;
- dst_ptr += 1;
- }
+ copy_horiz_transposed(src - src_stride * 3, src_stride,
+ temp, intermediate_height,
+ w, intermediate_height);
+ } else if (((const int32_t *)filter_x)[0] == 0) {
+ vp9_convolve2_dspr2(src - src_stride * 3, src_stride,
+ temp, intermediate_height,
+ filter_x,
+ w, intermediate_height);
} else {
src -= (src_stride * 3 + 3);
@@ -1021,20 +1030,14 @@
/* copy the src to dst */
if (filter_y[3] == 0x80) {
- int32_t y;
- int32_t c;
- uint8_t *src_ptr = temp + 3;
- uint8_t *dst_ptr = dst;
-
- for (y = w; y--;) {
- for (c = 0; c < h; c++) {
- dst_ptr[c * dst_stride] = src_ptr[c];
- }
-
- /* next row... */
- src_ptr += intermediate_height;
- dst_ptr += 1;
- }
+ copy_horiz_transposed(temp + 3, intermediate_height,
+ dst, dst_stride,
+ h, w);
+ } else if (((const int32_t *)filter_y)[0] == 0) {
+ vp9_convolve2_dspr2(temp + 3, intermediate_height,
+ dst, dst_stride,
+ filter_y,
+ h, w);
} else {
switch (h) {
case 4:
diff --git a/vp9/common/mips/dspr2/vp9_convolve8_horiz_dspr2.c b/vp9/common/mips/dspr2/vp9_convolve8_horiz_dspr2.c
index 743d641..0303896 100644
--- a/vp9/common/mips/dspr2/vp9_convolve8_horiz_dspr2.c
+++ b/vp9/common/mips/dspr2/vp9_convolve8_horiz_dspr2.c
@@ -849,6 +849,12 @@
filter_x, x_step_q4,
filter_y, y_step_q4,
w, h);
+ } else if (((const int32_t *)filter_x)[0] == 0) {
+ vp9_convolve2_horiz_dspr2(src, src_stride,
+ dst, dst_stride,
+ filter_x, x_step_q4,
+ filter_y, y_step_q4,
+ w, h);
} else {
if (16 == x_step_q4) {
uint32_t pos = 38;
diff --git a/vp9/common/mips/dspr2/vp9_convolve8_vert_dspr2.c b/vp9/common/mips/dspr2/vp9_convolve8_vert_dspr2.c
index bdc7930..0930bb3 100644
--- a/vp9/common/mips/dspr2/vp9_convolve8_vert_dspr2.c
+++ b/vp9/common/mips/dspr2/vp9_convolve8_vert_dspr2.c
@@ -341,6 +341,12 @@
filter_x, x_step_q4,
filter_y, y_step_q4,
w, h);
+ } else if (((const int32_t *)filter_y)[0] == 0) {
+ vp9_convolve2_vert_dspr2(src, src_stride,
+ dst, dst_stride,
+ filter_x, x_step_q4,
+ filter_y, y_step_q4,
+ w, h);
} else {
if (16 == y_step_q4) {
uint32_t pos = 38;
diff --git a/vp9/common/vp9_alloccommon.c b/vp9/common/vp9_alloccommon.c
index f0c653f..0f50f37 100644
--- a/vp9/common/vp9_alloccommon.c
+++ b/vp9/common/vp9_alloccommon.c
@@ -179,6 +179,7 @@
}
void vp9_initialize_common() {
+ vp9_init_neighbors();
vp9_coef_tree_initialize();
vp9_entropy_mode_init();
vp9_entropy_mv_init();
diff --git a/vp9/common/vp9_blockd.h b/vp9/common/vp9_blockd.h
index d0df966..b1af138 100644
--- a/vp9/common/vp9_blockd.h
+++ b/vp9/common/vp9_blockd.h
@@ -576,7 +576,7 @@
}
}
-static int get_tx_eob(struct segmentation *seg, int segment_id,
+static int get_tx_eob(const struct segmentation *seg, int segment_id,
TX_SIZE tx_size) {
const int eob_max = 16 << (tx_size << 1);
return vp9_segfeature_active(seg, segment_id, SEG_LVL_SKIP) ? 0 : eob_max;
diff --git a/vp9/common/vp9_entropy.c b/vp9/common/vp9_entropy.c
index 72ea72e..8ebe0e5 100644
--- a/vp9/common/vp9_entropy.c
+++ b/vp9/common/vp9_entropy.c
@@ -52,222 +52,7 @@
0, 1, 2, 3, 3, 4, 4, 5, 5, 5, 5, 5
};
-DECLARE_ALIGNED(16, const int16_t, vp9_default_scan_4x4[16]) = {
- 0, 4, 1, 5,
- 8, 2, 12, 9,
- 3, 6, 13, 10,
- 7, 14, 11, 15,
-};
-DECLARE_ALIGNED(16, const int16_t, vp9_col_scan_4x4[16]) = {
- 0, 4, 8, 1,
- 12, 5, 9, 2,
- 13, 6, 10, 3,
- 7, 14, 11, 15,
-};
-
-DECLARE_ALIGNED(16, const int16_t, vp9_row_scan_4x4[16]) = {
- 0, 1, 4, 2,
- 5, 3, 6, 8,
- 9, 7, 12, 10,
- 13, 11, 14, 15,
-};
-
-DECLARE_ALIGNED(16, const int16_t, vp9_default_scan_8x8[64]) = {
- 0, 8, 1, 16, 9, 2, 17, 24,
- 10, 3, 18, 25, 32, 11, 4, 26,
- 33, 19, 40, 12, 34, 27, 5, 41,
- 20, 48, 13, 35, 42, 28, 21, 6,
- 49, 56, 36, 43, 29, 7, 14, 50,
- 57, 44, 22, 37, 15, 51, 58, 30,
- 45, 23, 52, 59, 38, 31, 60, 53,
- 46, 39, 61, 54, 47, 62, 55, 63,
-};
-
-DECLARE_ALIGNED(16, const int16_t, vp9_col_scan_8x8[64]) = {
- 0, 8, 16, 1, 24, 9, 32, 17,
- 2, 40, 25, 10, 33, 18, 48, 3,
- 26, 41, 11, 56, 19, 34, 4, 49,
- 27, 42, 12, 35, 20, 57, 50, 28,
- 5, 43, 13, 36, 58, 51, 21, 44,
- 6, 29, 59, 37, 14, 52, 22, 7,
- 45, 60, 30, 15, 38, 53, 23, 46,
- 31, 61, 39, 54, 47, 62, 55, 63,
-};
-
-DECLARE_ALIGNED(16, const int16_t, vp9_row_scan_8x8[64]) = {
- 0, 1, 2, 8, 9, 3, 16, 10,
- 4, 17, 11, 24, 5, 18, 25, 12,
- 19, 26, 32, 6, 13, 20, 33, 27,
- 7, 34, 40, 21, 28, 41, 14, 35,
- 48, 42, 29, 36, 49, 22, 43, 15,
- 56, 37, 50, 44, 30, 57, 23, 51,
- 58, 45, 38, 52, 31, 59, 53, 46,
- 60, 39, 61, 47, 54, 55, 62, 63,
-};
-
-DECLARE_ALIGNED(16, const int16_t, vp9_default_scan_16x16[256]) = {
- 0, 16, 1, 32, 17, 2, 48, 33, 18, 3, 64, 34, 49, 19, 65, 80,
- 50, 4, 35, 66, 20, 81, 96, 51, 5, 36, 82, 97, 67, 112, 21, 52,
- 98, 37, 83, 113, 6, 68, 128, 53, 22, 99, 114, 84, 7, 129, 38, 69,
- 100, 115, 144, 130, 85, 54, 23, 8, 145, 39, 70, 116, 101, 131, 160, 146,
- 55, 86, 24, 71, 132, 117, 161, 40, 9, 102, 147, 176, 162, 87, 56, 25,
- 133, 118, 177, 148, 72, 103, 41, 163, 10, 192, 178, 88, 57, 134, 149, 119,
- 26, 164, 73, 104, 193, 42, 179, 208, 11, 135, 89, 165, 120, 150, 58, 194,
- 180, 27, 74, 209, 105, 151, 136, 43, 90, 224, 166, 195, 181, 121, 210, 59,
- 12, 152, 106, 167, 196, 75, 137, 225, 211, 240, 182, 122, 91, 28, 197, 13,
- 226, 168, 183, 153, 44, 212, 138, 107, 241, 60, 29, 123, 198, 184, 227, 169,
- 242, 76, 213, 154, 45, 92, 14, 199, 139, 61, 228, 214, 170, 185, 243, 108,
- 77, 155, 30, 15, 200, 229, 124, 215, 244, 93, 46, 186, 171, 201, 109, 140,
- 230, 62, 216, 245, 31, 125, 78, 156, 231, 47, 187, 202, 217, 94, 246, 141,
- 63, 232, 172, 110, 247, 157, 79, 218, 203, 126, 233, 188, 248, 95, 173, 142,
- 219, 111, 249, 234, 158, 127, 189, 204, 250, 235, 143, 174, 220, 205, 159,
- 251,
- 190, 221, 175, 236, 237, 191, 206, 252, 222, 253, 207, 238, 223, 254, 239,
- 255,
-};
-
-DECLARE_ALIGNED(16, const int16_t, vp9_col_scan_16x16[256]) = {
- 0, 16, 32, 48, 1, 64, 17, 80, 33, 96, 49, 2, 65, 112, 18, 81,
- 34, 128, 50, 97, 3, 66, 144, 19, 113, 35, 82, 160, 98, 51, 129, 4,
- 67, 176, 20, 114, 145, 83, 36, 99, 130, 52, 192, 5, 161, 68, 115, 21,
- 146, 84, 208, 177, 37, 131, 100, 53, 162, 224, 69, 6, 116, 193, 147, 85,
- 22, 240, 132, 38, 178, 101, 163, 54, 209, 117, 70, 7, 148, 194, 86, 179,
- 225, 23, 133, 39, 164, 8, 102, 210, 241, 55, 195, 118, 149, 71, 180, 24,
- 87, 226, 134, 165, 211, 40, 103, 56, 72, 150, 196, 242, 119, 9, 181, 227,
- 88, 166, 25, 135, 41, 104, 212, 57, 151, 197, 120, 73, 243, 182, 136, 167,
- 213, 89, 10, 228, 105, 152, 198, 26, 42, 121, 183, 244, 168, 58, 137, 229,
- 74, 214, 90, 153, 199, 184, 11, 106, 245, 27, 122, 230, 169, 43, 215, 59,
- 200, 138, 185, 246, 75, 12, 91, 154, 216, 231, 107, 28, 44, 201, 123, 170,
- 60, 247, 232, 76, 139, 13, 92, 217, 186, 248, 155, 108, 29, 124, 45, 202,
- 233, 171, 61, 14, 77, 140, 15, 249, 93, 30, 187, 156, 218, 46, 109, 125,
- 62, 172, 78, 203, 31, 141, 234, 94, 47, 188, 63, 157, 110, 250, 219, 79,
- 126, 204, 173, 142, 95, 189, 111, 235, 158, 220, 251, 127, 174, 143, 205,
- 236,
- 159, 190, 221, 252, 175, 206, 237, 191, 253, 222, 238, 207, 254, 223, 239,
- 255,
-};
-
-DECLARE_ALIGNED(16, const int16_t, vp9_row_scan_16x16[256]) = {
- 0, 1, 2, 16, 3, 17, 4, 18, 32, 5, 33, 19, 6, 34, 48, 20,
- 49, 7, 35, 21, 50, 64, 8, 36, 65, 22, 51, 37, 80, 9, 66, 52,
- 23, 38, 81, 67, 10, 53, 24, 82, 68, 96, 39, 11, 54, 83, 97, 69,
- 25, 98, 84, 40, 112, 55, 12, 70, 99, 113, 85, 26, 41, 56, 114, 100,
- 13, 71, 128, 86, 27, 115, 101, 129, 42, 57, 72, 116, 14, 87, 130, 102,
- 144, 73, 131, 117, 28, 58, 15, 88, 43, 145, 103, 132, 146, 118, 74, 160,
- 89, 133, 104, 29, 59, 147, 119, 44, 161, 148, 90, 105, 134, 162, 120, 176,
- 75, 135, 149, 30, 60, 163, 177, 45, 121, 91, 106, 164, 178, 150, 192, 136,
- 165, 179, 31, 151, 193, 76, 122, 61, 137, 194, 107, 152, 180, 208, 46, 166,
- 167, 195, 92, 181, 138, 209, 123, 153, 224, 196, 77, 168, 210, 182, 240, 108,
- 197, 62, 154, 225, 183, 169, 211, 47, 139, 93, 184, 226, 212, 241, 198, 170,
- 124, 155, 199, 78, 213, 185, 109, 227, 200, 63, 228, 242, 140, 214, 171, 186,
- 156, 229, 243, 125, 94, 201, 244, 215, 216, 230, 141, 187, 202, 79, 172, 110,
- 157, 245, 217, 231, 95, 246, 232, 126, 203, 247, 233, 173, 218, 142, 111,
- 158,
- 188, 248, 127, 234, 219, 249, 189, 204, 143, 174, 159, 250, 235, 205, 220,
- 175,
- 190, 251, 221, 191, 206, 236, 207, 237, 252, 222, 253, 223, 238, 239, 254,
- 255,
-};
-
-DECLARE_ALIGNED(16, const int16_t, vp9_default_scan_32x32[1024]) = {
- 0, 32, 1, 64, 33, 2, 96, 65, 34, 128, 3, 97, 66, 160,
- 129, 35, 98, 4, 67, 130, 161, 192, 36, 99, 224, 5, 162, 193,
- 68, 131, 37, 100,
- 225, 194, 256, 163, 69, 132, 6, 226, 257, 288, 195, 101, 164, 38,
- 258, 7, 227, 289, 133, 320, 70, 196, 165, 290, 259, 228, 39, 321,
- 102, 352, 8, 197,
- 71, 134, 322, 291, 260, 353, 384, 229, 166, 103, 40, 354, 323, 292,
- 135, 385, 198, 261, 72, 9, 416, 167, 386, 355, 230, 324, 104, 293,
- 41, 417, 199, 136,
- 262, 387, 448, 325, 356, 10, 73, 418, 231, 168, 449, 294, 388, 105,
- 419, 263, 42, 200, 357, 450, 137, 480, 74, 326, 232, 11, 389, 169,
- 295, 420, 106, 451,
- 481, 358, 264, 327, 201, 43, 138, 512, 482, 390, 296, 233, 170, 421,
- 75, 452, 359, 12, 513, 265, 483, 328, 107, 202, 514, 544, 422, 391,
- 453, 139, 44, 234,
- 484, 297, 360, 171, 76, 515, 545, 266, 329, 454, 13, 423, 203, 108,
- 546, 485, 576, 298, 235, 140, 361, 330, 172, 547, 45, 455, 267, 577,
- 486, 77, 204, 362,
- 608, 14, 299, 578, 109, 236, 487, 609, 331, 141, 579, 46, 15, 173,
- 610, 363, 78, 205, 16, 110, 237, 611, 142, 47, 174, 79, 206, 17,
- 111, 238, 48, 143,
- 80, 175, 112, 207, 49, 18, 239, 81, 113, 19, 50, 82, 114, 51,
- 83, 115, 640, 516, 392, 268, 144, 20, 672, 641, 548, 517, 424,
- 393, 300, 269, 176, 145,
- 52, 21, 704, 673, 642, 580, 549, 518, 456, 425, 394, 332, 301,
- 270, 208, 177, 146, 84, 53, 22, 736, 705, 674, 643, 612, 581,
- 550, 519, 488, 457, 426, 395,
- 364, 333, 302, 271, 240, 209, 178, 147, 116, 85, 54, 23, 737,
- 706, 675, 613, 582, 551, 489, 458, 427, 365, 334, 303, 241,
- 210, 179, 117, 86, 55, 738, 707,
- 614, 583, 490, 459, 366, 335, 242, 211, 118, 87, 739, 615, 491,
- 367, 243, 119, 768, 644, 520, 396, 272, 148, 24, 800, 769, 676,
- 645, 552, 521, 428, 397, 304,
- 273, 180, 149, 56, 25, 832, 801, 770, 708, 677, 646, 584, 553,
- 522, 460, 429, 398, 336, 305, 274, 212, 181, 150, 88, 57, 26,
- 864, 833, 802, 771, 740, 709,
- 678, 647, 616, 585, 554, 523, 492, 461, 430, 399, 368, 337, 306,
- 275, 244, 213, 182, 151, 120, 89, 58, 27, 865, 834, 803, 741,
- 710, 679, 617, 586, 555, 493,
- 462, 431, 369, 338, 307, 245, 214, 183, 121, 90, 59, 866, 835,
- 742, 711, 618, 587, 494, 463, 370, 339, 246, 215, 122, 91, 867,
- 743, 619, 495, 371, 247, 123,
- 896, 772, 648, 524, 400, 276, 152, 28, 928, 897, 804, 773, 680,
- 649, 556, 525, 432, 401, 308, 277, 184, 153, 60, 29, 960, 929,
- 898, 836, 805, 774, 712, 681,
- 650, 588, 557, 526, 464, 433, 402, 340, 309, 278, 216, 185, 154,
- 92, 61, 30, 992, 961, 930, 899, 868, 837, 806, 775, 744, 713, 682,
- 651, 620, 589, 558, 527,
- 496, 465, 434, 403, 372, 341, 310, 279, 248, 217, 186, 155, 124,
- 93, 62, 31, 993, 962, 931, 869, 838, 807, 745, 714, 683, 621, 590,
- 559, 497, 466, 435, 373,
- 342, 311, 249, 218, 187, 125, 94, 63, 994, 963, 870, 839, 746, 715,
- 622, 591, 498, 467, 374, 343, 250, 219, 126, 95, 995, 871, 747, 623,
- 499, 375, 251, 127,
- 900, 776, 652, 528, 404, 280, 156, 932, 901, 808, 777, 684, 653, 560,
- 529, 436, 405, 312, 281, 188, 157, 964, 933, 902, 840, 809, 778, 716,
- 685, 654, 592, 561,
- 530, 468, 437, 406, 344, 313, 282, 220, 189, 158, 996, 965, 934, 903,
- 872, 841, 810, 779, 748, 717, 686, 655, 624, 593, 562, 531, 500, 469,
- 438, 407, 376, 345,
- 314, 283, 252, 221, 190, 159, 997, 966, 935, 873, 842, 811, 749, 718,
- 687, 625, 594, 563, 501, 470, 439, 377, 346, 315, 253, 222, 191, 998,
- 967, 874, 843, 750,
- 719, 626, 595, 502, 471, 378, 347, 254, 223, 999, 875, 751, 627, 503,
- 379, 255, 904, 780, 656, 532, 408, 284, 936, 905, 812, 781, 688, 657,
- 564, 533, 440, 409,
- 316, 285, 968, 937, 906, 844, 813, 782, 720, 689, 658, 596, 565, 534,
- 472, 441, 410, 348, 317, 286, 1000, 969, 938, 907, 876, 845, 814, 783,
- 752, 721, 690, 659,
- 628, 597, 566, 535, 504, 473, 442, 411, 380, 349, 318, 287, 1001, 970,
- 939, 877, 846, 815, 753, 722, 691, 629, 598, 567, 505, 474, 443, 381,
- 350, 319, 1002, 971,
- 878, 847, 754, 723, 630, 599, 506, 475, 382, 351, 1003, 879, 755, 631,
- 507, 383, 908, 784, 660, 536, 412, 940, 909, 816, 785, 692, 661, 568,
- 537, 444, 413, 972,
- 941, 910, 848, 817, 786, 724, 693, 662, 600, 569, 538, 476, 445, 414,
- 1004, 973, 942, 911, 880, 849, 818, 787, 756, 725, 694, 663, 632, 601,
- 570, 539, 508, 477,
- 446, 415, 1005, 974, 943, 881, 850, 819, 757, 726, 695, 633, 602, 571,
- 509, 478, 447, 1006, 975, 882, 851, 758, 727, 634, 603, 510, 479,
- 1007, 883, 759, 635, 511,
- 912, 788, 664, 540, 944, 913, 820, 789, 696, 665, 572, 541, 976, 945,
- 914, 852, 821, 790, 728, 697, 666, 604, 573, 542, 1008, 977, 946, 915,
- 884, 853, 822, 791,
- 760, 729, 698, 667, 636, 605, 574, 543, 1009, 978, 947, 885, 854, 823,
- 761, 730, 699, 637, 606, 575, 1010, 979, 886, 855, 762, 731, 638, 607,
- 1011, 887, 763, 639,
- 916, 792, 668, 948, 917, 824, 793, 700, 669, 980, 949, 918, 856, 825,
- 794, 732, 701, 670, 1012, 981, 950, 919, 888, 857, 826, 795, 764, 733,
- 702, 671, 1013, 982,
- 951, 889, 858, 827, 765, 734, 703, 1014, 983, 890, 859, 766, 735, 1015,
- 891, 767, 920, 796, 952, 921, 828, 797, 984, 953, 922, 860, 829, 798,
- 1016, 985, 954, 923,
- 892, 861, 830, 799, 1017, 986, 955, 893, 862, 831, 1018, 987, 894, 863,
- 1019, 895, 924, 956, 925, 988, 957, 926, 1020, 989, 958, 927, 1021,
- 990, 959, 1022, 991, 1023,
-};
/* Array indices are identical to previously-existing CONTEXT_NODE indices */
@@ -513,134 +298,7 @@
vp9_copy(cm->fc.coef_probs[TX_32X32], default_coef_probs_32x32);
}
-// Neighborhood 5-tuples for various scans and blocksizes,
-// in {top, left, topleft, topright, bottomleft} order
-// for each position in raster scan order.
-// -1 indicates the neighbor does not exist.
-DECLARE_ALIGNED(16, int16_t,
- vp9_default_scan_4x4_neighbors[17 * MAX_NEIGHBORS]);
-DECLARE_ALIGNED(16, int16_t,
- vp9_col_scan_4x4_neighbors[17 * MAX_NEIGHBORS]);
-DECLARE_ALIGNED(16, int16_t,
- vp9_row_scan_4x4_neighbors[17 * MAX_NEIGHBORS]);
-DECLARE_ALIGNED(16, int16_t,
- vp9_col_scan_8x8_neighbors[65 * MAX_NEIGHBORS]);
-DECLARE_ALIGNED(16, int16_t,
- vp9_row_scan_8x8_neighbors[65 * MAX_NEIGHBORS]);
-DECLARE_ALIGNED(16, int16_t,
- vp9_default_scan_8x8_neighbors[65 * MAX_NEIGHBORS]);
-DECLARE_ALIGNED(16, int16_t,
- vp9_col_scan_16x16_neighbors[257 * MAX_NEIGHBORS]);
-DECLARE_ALIGNED(16, int16_t,
- vp9_row_scan_16x16_neighbors[257 * MAX_NEIGHBORS]);
-DECLARE_ALIGNED(16, int16_t,
- vp9_default_scan_16x16_neighbors[257 * MAX_NEIGHBORS]);
-DECLARE_ALIGNED(16, int16_t,
- vp9_default_scan_32x32_neighbors[1025 * MAX_NEIGHBORS]);
-
-DECLARE_ALIGNED(16, int16_t, vp9_default_iscan_4x4[16]);
-DECLARE_ALIGNED(16, int16_t, vp9_col_iscan_4x4[16]);
-DECLARE_ALIGNED(16, int16_t, vp9_row_iscan_4x4[16]);
-DECLARE_ALIGNED(16, int16_t, vp9_col_iscan_8x8[64]);
-DECLARE_ALIGNED(16, int16_t, vp9_row_iscan_8x8[64]);
-DECLARE_ALIGNED(16, int16_t, vp9_default_iscan_8x8[64]);
-DECLARE_ALIGNED(16, int16_t, vp9_col_iscan_16x16[256]);
-DECLARE_ALIGNED(16, int16_t, vp9_row_iscan_16x16[256]);
-DECLARE_ALIGNED(16, int16_t, vp9_default_iscan_16x16[256]);
-DECLARE_ALIGNED(16, int16_t, vp9_default_iscan_32x32[1024]);
-
-static int find_in_scan(const int16_t *scan, int l, int idx) {
- int n, l2 = l * l;
- for (n = 0; n < l2; n++) {
- int rc = scan[n];
- if (rc == idx)
- return n;
- }
- assert(0);
- return -1;
-}
-static void init_scan_neighbors(const int16_t *scan,
- int16_t *iscan,
- int l, int16_t *neighbors) {
- int l2 = l * l;
- int n, i, j;
-
- // dc doesn't use this type of prediction
- neighbors[MAX_NEIGHBORS * 0 + 0] = 0;
- neighbors[MAX_NEIGHBORS * 0 + 1] = 0;
- iscan[0] = find_in_scan(scan, l, 0);
- for (n = 1; n < l2; n++) {
- int rc = scan[n];
- iscan[n] = find_in_scan(scan, l, n);
- i = rc / l;
- j = rc % l;
- if (i > 0 && j > 0) {
- // col/row scan is used for adst/dct, and generally means that
- // energy decreases to zero much faster in the dimension in
- // which ADST is used compared to the direction in which DCT
- // is used. Likewise, we find much higher correlation between
- // coefficients within the direction in which DCT is used.
- // Therefore, if we use ADST/DCT, prefer the DCT neighbor coeff
- // as a context. If ADST or DCT is used in both directions, we
- // use the combination of the two as a context.
- int a = (i - 1) * l + j;
- int b = i * l + j - 1;
- if (scan == vp9_col_scan_4x4 || scan == vp9_col_scan_8x8 ||
- scan == vp9_col_scan_16x16) {
- // in the col/row scan cases (as well as left/top edge cases), we set
- // both contexts to the same value, so we can branchlessly do a+b+1>>1
- // which automatically becomes a if a == b
- neighbors[MAX_NEIGHBORS * n + 0] =
- neighbors[MAX_NEIGHBORS * n + 1] = a;
- } else if (scan == vp9_row_scan_4x4 || scan == vp9_row_scan_8x8 ||
- scan == vp9_row_scan_16x16) {
- neighbors[MAX_NEIGHBORS * n + 0] =
- neighbors[MAX_NEIGHBORS * n + 1] = b;
- } else {
- neighbors[MAX_NEIGHBORS * n + 0] = a;
- neighbors[MAX_NEIGHBORS * n + 1] = b;
- }
- } else if (i > 0) {
- neighbors[MAX_NEIGHBORS * n + 0] =
- neighbors[MAX_NEIGHBORS * n + 1] = (i - 1) * l + j;
- } else {
- assert(j > 0);
- neighbors[MAX_NEIGHBORS * n + 0] =
- neighbors[MAX_NEIGHBORS * n + 1] = i * l + j - 1;
- }
- assert(iscan[neighbors[MAX_NEIGHBORS * n + 0]] < n);
- }
- // one padding item so we don't have to add branches in code to handle
- // calls to get_coef_context() for the token after the final dc token
- neighbors[MAX_NEIGHBORS * l2 + 0] = 0;
- neighbors[MAX_NEIGHBORS * l2 + 1] = 0;
-}
-
-void vp9_init_neighbors() {
- init_scan_neighbors(vp9_default_scan_4x4, vp9_default_iscan_4x4, 4,
- vp9_default_scan_4x4_neighbors);
- init_scan_neighbors(vp9_row_scan_4x4, vp9_row_iscan_4x4, 4,
- vp9_row_scan_4x4_neighbors);
- init_scan_neighbors(vp9_col_scan_4x4, vp9_col_iscan_4x4, 4,
- vp9_col_scan_4x4_neighbors);
- init_scan_neighbors(vp9_default_scan_8x8, vp9_default_iscan_8x8, 8,
- vp9_default_scan_8x8_neighbors);
- init_scan_neighbors(vp9_row_scan_8x8, vp9_row_iscan_8x8, 8,
- vp9_row_scan_8x8_neighbors);
- init_scan_neighbors(vp9_col_scan_8x8, vp9_col_iscan_8x8, 8,
- vp9_col_scan_8x8_neighbors);
- init_scan_neighbors(vp9_default_scan_16x16, vp9_default_iscan_16x16, 16,
- vp9_default_scan_16x16_neighbors);
- init_scan_neighbors(vp9_row_scan_16x16, vp9_row_iscan_16x16, 16,
- vp9_row_scan_16x16_neighbors);
- init_scan_neighbors(vp9_col_scan_16x16, vp9_col_iscan_16x16, 16,
- vp9_col_scan_16x16_neighbors);
- init_scan_neighbors(vp9_default_scan_32x32, vp9_default_iscan_32x32, 32,
- vp9_default_scan_32x32_neighbors);
-}
-
void vp9_coef_tree_initialize() {
- vp9_init_neighbors();
init_bit_trees();
vp9_tokens_from_tree(vp9_coef_encodings, vp9_coef_tree);
}
@@ -657,10 +315,10 @@
static void adapt_coef_probs(VP9_COMMON *cm, TX_SIZE tx_size,
unsigned int count_sat,
unsigned int update_factor) {
- FRAME_CONTEXT *pre_fc = &cm->frame_contexts[cm->frame_context_idx];
+ const FRAME_CONTEXT *pre_fc = &cm->frame_contexts[cm->frame_context_idx];
vp9_coeff_probs_model *dst_coef_probs = cm->fc.coef_probs[tx_size];
- vp9_coeff_probs_model *pre_coef_probs = pre_fc->coef_probs[tx_size];
+ const vp9_coeff_probs_model *pre_coef_probs = pre_fc->coef_probs[tx_size];
vp9_coeff_count_model *coef_counts = cm->counts.coef[tx_size];
unsigned int (*eob_branch_count)[REF_TYPES][COEF_BANDS][PREV_COEF_CONTEXTS] =
cm->counts.eob_branch[tx_size];
diff --git a/vp9/common/vp9_entropy.h b/vp9/common/vp9_entropy.h
index d412781..6d8edec 100644
--- a/vp9/common/vp9_entropy.h
+++ b/vp9/common/vp9_entropy.h
@@ -12,9 +12,11 @@
#define VP9_COMMON_VP9_ENTROPY_H_
#include "vpx/vpx_integer.h"
-#include "vp9/common/vp9_treecoder.h"
+
#include "vp9/common/vp9_blockd.h"
#include "vp9/common/vp9_common.h"
+#include "vp9/common/vp9_scan.h"
+#include "vp9/common/vp9_treecoder.h"
/* Coefficient token alphabet */
@@ -99,64 +101,8 @@
struct VP9Common;
void vp9_default_coef_probs(struct VP9Common *cm);
-extern DECLARE_ALIGNED(16, const int16_t, vp9_default_scan_4x4[16]);
-extern DECLARE_ALIGNED(16, const int16_t, vp9_col_scan_4x4[16]);
-extern DECLARE_ALIGNED(16, const int16_t, vp9_row_scan_4x4[16]);
-
-extern DECLARE_ALIGNED(16, const int16_t, vp9_default_scan_8x8[64]);
-
-extern DECLARE_ALIGNED(16, const int16_t, vp9_col_scan_8x8[64]);
-extern DECLARE_ALIGNED(16, const int16_t, vp9_row_scan_8x8[64]);
-
-extern DECLARE_ALIGNED(16, const int16_t, vp9_default_scan_16x16[256]);
-
-extern DECLARE_ALIGNED(16, const int16_t, vp9_col_scan_16x16[256]);
-extern DECLARE_ALIGNED(16, const int16_t, vp9_row_scan_16x16[256]);
-
-extern DECLARE_ALIGNED(16, const int16_t, vp9_default_scan_32x32[1024]);
-
-extern DECLARE_ALIGNED(16, int16_t, vp9_default_iscan_4x4[16]);
-
-extern DECLARE_ALIGNED(16, int16_t, vp9_col_iscan_4x4[16]);
-extern DECLARE_ALIGNED(16, int16_t, vp9_row_iscan_4x4[16]);
-
-extern DECLARE_ALIGNED(16, int16_t, vp9_default_iscan_8x8[64]);
-
-extern DECLARE_ALIGNED(16, int16_t, vp9_col_iscan_8x8[64]);
-extern DECLARE_ALIGNED(16, int16_t, vp9_row_iscan_8x8[64]);
-
-extern DECLARE_ALIGNED(16, int16_t, vp9_default_iscan_16x16[256]);
-
-extern DECLARE_ALIGNED(16, int16_t, vp9_col_iscan_16x16[256]);
-extern DECLARE_ALIGNED(16, int16_t, vp9_row_iscan_16x16[256]);
-
-extern DECLARE_ALIGNED(16, int16_t, vp9_default_iscan_32x32[1024]);
-
-#define MAX_NEIGHBORS 2
-
-extern DECLARE_ALIGNED(16, int16_t,
- vp9_default_scan_4x4_neighbors[17 * MAX_NEIGHBORS]);
-extern DECLARE_ALIGNED(16, int16_t,
- vp9_col_scan_4x4_neighbors[17 * MAX_NEIGHBORS]);
-extern DECLARE_ALIGNED(16, int16_t,
- vp9_row_scan_4x4_neighbors[17 * MAX_NEIGHBORS]);
-extern DECLARE_ALIGNED(16, int16_t,
- vp9_col_scan_8x8_neighbors[65 * MAX_NEIGHBORS]);
-extern DECLARE_ALIGNED(16, int16_t,
- vp9_row_scan_8x8_neighbors[65 * MAX_NEIGHBORS]);
-extern DECLARE_ALIGNED(16, int16_t,
- vp9_default_scan_8x8_neighbors[65 * MAX_NEIGHBORS]);
-extern DECLARE_ALIGNED(16, int16_t,
- vp9_col_scan_16x16_neighbors[257 * MAX_NEIGHBORS]);
-extern DECLARE_ALIGNED(16, int16_t,
- vp9_row_scan_16x16_neighbors[257 * MAX_NEIGHBORS]);
-extern DECLARE_ALIGNED(16, int16_t,
- vp9_default_scan_16x16_neighbors[257 * MAX_NEIGHBORS]);
-extern DECLARE_ALIGNED(16, int16_t,
- vp9_default_scan_32x32_neighbors[1025 * MAX_NEIGHBORS]);
-
-void vp9_coef_tree_initialize(void);
+void vp9_coef_tree_initialize();
void vp9_adapt_coef_probs(struct VP9Common *cm);
static INLINE void reset_skip_context(MACROBLOCKD *xd, BLOCK_SIZE bsize) {
@@ -186,13 +132,6 @@
? (COEF_BANDS-1) : band_translate[coef_index];
}
-static INLINE int get_coef_context(const int16_t *neighbors,
- uint8_t *token_cache,
- int c) {
- return (1 + token_cache[neighbors[MAX_NEIGHBORS * c + 0]] +
- token_cache[neighbors[MAX_NEIGHBORS * c + 1]]) >> 1;
-}
-
// 128 lists of probabilities are stored for the following ONE node probs:
// 1, 3, 5, 7, ..., 253, 255
// In between probabilities are interpolated linearly
@@ -213,126 +152,6 @@
void vp9_model_to_full_probs(const vp9_prob *model, vp9_prob *full);
-static INLINE const int16_t* get_scan_4x4(TX_TYPE tx_type) {
- switch (tx_type) {
- case ADST_DCT:
- return vp9_row_scan_4x4;
- case DCT_ADST:
- return vp9_col_scan_4x4;
- default:
- return vp9_default_scan_4x4;
- }
-}
-
-static INLINE void get_scan_nb_4x4(TX_TYPE tx_type,
- const int16_t **scan, const int16_t **nb) {
- switch (tx_type) {
- case ADST_DCT:
- *scan = vp9_row_scan_4x4;
- *nb = vp9_row_scan_4x4_neighbors;
- break;
- case DCT_ADST:
- *scan = vp9_col_scan_4x4;
- *nb = vp9_col_scan_4x4_neighbors;
- break;
- default:
- *scan = vp9_default_scan_4x4;
- *nb = vp9_default_scan_4x4_neighbors;
- break;
- }
-}
-
-static INLINE const int16_t* get_iscan_4x4(TX_TYPE tx_type) {
- switch (tx_type) {
- case ADST_DCT:
- return vp9_row_iscan_4x4;
- case DCT_ADST:
- return vp9_col_iscan_4x4;
- default:
- return vp9_default_iscan_4x4;
- }
-}
-
-static INLINE const int16_t* get_scan_8x8(TX_TYPE tx_type) {
- switch (tx_type) {
- case ADST_DCT:
- return vp9_row_scan_8x8;
- case DCT_ADST:
- return vp9_col_scan_8x8;
- default:
- return vp9_default_scan_8x8;
- }
-}
-
-static INLINE void get_scan_nb_8x8(TX_TYPE tx_type,
- const int16_t **scan, const int16_t **nb) {
- switch (tx_type) {
- case ADST_DCT:
- *scan = vp9_row_scan_8x8;
- *nb = vp9_row_scan_8x8_neighbors;
- break;
- case DCT_ADST:
- *scan = vp9_col_scan_8x8;
- *nb = vp9_col_scan_8x8_neighbors;
- break;
- default:
- *scan = vp9_default_scan_8x8;
- *nb = vp9_default_scan_8x8_neighbors;
- break;
- }
-}
-
-static INLINE const int16_t* get_iscan_8x8(TX_TYPE tx_type) {
- switch (tx_type) {
- case ADST_DCT:
- return vp9_row_iscan_8x8;
- case DCT_ADST:
- return vp9_col_iscan_8x8;
- default:
- return vp9_default_iscan_8x8;
- }
-}
-
-static INLINE const int16_t* get_scan_16x16(TX_TYPE tx_type) {
- switch (tx_type) {
- case ADST_DCT:
- return vp9_row_scan_16x16;
- case DCT_ADST:
- return vp9_col_scan_16x16;
- default:
- return vp9_default_scan_16x16;
- }
-}
-
-static INLINE void get_scan_nb_16x16(TX_TYPE tx_type,
- const int16_t **scan, const int16_t **nb) {
- switch (tx_type) {
- case ADST_DCT:
- *scan = vp9_row_scan_16x16;
- *nb = vp9_row_scan_16x16_neighbors;
- break;
- case DCT_ADST:
- *scan = vp9_col_scan_16x16;
- *nb = vp9_col_scan_16x16_neighbors;
- break;
- default:
- *scan = vp9_default_scan_16x16;
- *nb = vp9_default_scan_16x16_neighbors;
- break;
- }
-}
-
-static INLINE const int16_t* get_iscan_16x16(TX_TYPE tx_type) {
- switch (tx_type) {
- case ADST_DCT:
- return vp9_row_iscan_16x16;
- case DCT_ADST:
- return vp9_col_iscan_16x16;
- default:
- return vp9_default_iscan_16x16;
- }
-}
-
static int get_entropy_context(TX_SIZE tx_size,
ENTROPY_CONTEXT *a, ENTROPY_CONTEXT *l) {
ENTROPY_CONTEXT above_ec = 0, left_ec = 0;
diff --git a/vp9/common/vp9_entropymv.c b/vp9/common/vp9_entropymv.c
index a9e25b7..baff637 100644
--- a/vp9/common/vp9_entropymv.c
+++ b/vp9/common/vp9_entropymv.c
@@ -214,11 +214,11 @@
void vp9_adapt_mv_probs(VP9_COMMON *cm, int allow_hp) {
int i, j;
- FRAME_CONTEXT *pre_fc = &cm->frame_contexts[cm->frame_context_idx];
+ const FRAME_CONTEXT *pre_fc = &cm->frame_contexts[cm->frame_context_idx];
nmv_context *ctx = &cm->fc.nmvc;
- nmv_context *pre_ctx = &pre_fc->nmvc;
- nmv_context_counts *cts = &cm->counts.mv;
+ const nmv_context *pre_ctx = &pre_fc->nmvc;
+ const nmv_context_counts *cts = &cm->counts.mv;
adapt_probs(0, vp9_mv_joint_tree, ctx->joints, pre_ctx->joints, cts->joints);
diff --git a/vp9/common/vp9_idct.c b/vp9/common/vp9_idct.c
index dea9237..5846504 100644
--- a/vp9/common/vp9_idct.c
+++ b/vp9/common/vp9_idct.c
@@ -611,7 +611,7 @@
output[15] = step2[0] - step2[15];
}
-void vp9_short_idct16x16_add_c(int16_t *input, uint8_t *dest, int dest_stride) {
+void vp9_idct16x16_256_add_c(int16_t *input, uint8_t *dest, int dest_stride) {
int16_t out[16 * 16];
int16_t *outptr = out;
int i, j;
@@ -838,7 +838,7 @@
+ dest[j * dest_stride + i]); }
}
-void vp9_short_idct16x16_10_add_c(int16_t *input, uint8_t *dest,
+void vp9_idct16x16_10_add_c(int16_t *input, uint8_t *dest,
int dest_stride) {
int16_t out[16 * 16] = { 0 };
int16_t *outptr = out;
@@ -864,7 +864,7 @@
}
}
-void vp9_short_idct16x16_1_add_c(int16_t *input, uint8_t *dest,
+void vp9_idct16x16_1_add_c(int16_t *input, uint8_t *dest,
int dest_stride) {
int i, j;
int a1;
@@ -1253,7 +1253,20 @@
// Rows
for (i = 0; i < 32; ++i) {
- idct32_1d(input, outptr);
+ int16_t zero_coeff[16];
+ for (j = 0; j < 16; ++j)
+ zero_coeff[j] = input[2 * j] | input[2 * j + 1];
+ for (j = 0; j < 8; ++j)
+ zero_coeff[j] = zero_coeff[2 * j] | zero_coeff[2 * j + 1];
+ for (j = 0; j < 4; ++j)
+ zero_coeff[j] = zero_coeff[2 * j] | zero_coeff[2 * j + 1];
+ for (j = 0; j < 2; ++j)
+ zero_coeff[j] = zero_coeff[2 * j] | zero_coeff[2 * j + 1];
+
+ if (zero_coeff[0] | zero_coeff[1])
+ idct32_1d(input, outptr);
+ else
+ vpx_memset(outptr, 0, sizeof(int16_t) * 32);
input += 32;
outptr += 32;
}
@@ -1320,17 +1333,17 @@
}
}
-void vp9_idct_add_16x16(int16_t *input, uint8_t *dest, int stride, int eob) {
+void vp9_idct16x16_add(int16_t *input, uint8_t *dest, int stride, int eob) {
/* The calculation can be simplified if there are not many non-zero dct
* coefficients. Use eobs to separate different cases. */
if (eob) {
if (eob == 1)
/* DC only DCT coefficient. */
- vp9_short_idct16x16_1_add(input, dest, stride);
+ vp9_idct16x16_1_add(input, dest, stride);
else if (eob <= 10)
- vp9_short_idct16x16_10_add(input, dest, stride);
+ vp9_idct16x16_10_add(input, dest, stride);
else
- vp9_short_idct16x16_add(input, dest, stride);
+ vp9_idct16x16_256_add(input, dest, stride);
}
}
@@ -1366,7 +1379,7 @@
void vp9_iht_add_16x16(TX_TYPE tx_type, int16_t *input, uint8_t *dest,
int stride, int eob) {
if (tx_type == DCT_DCT) {
- vp9_idct_add_16x16(input, dest, stride, eob);
+ vp9_idct16x16_add(input, dest, stride, eob);
} else {
if (eob > 0) {
vp9_short_iht16x16_add(input, dest, stride, tx_type);
diff --git a/vp9/common/vp9_idct.h b/vp9/common/vp9_idct.h
index 41519ce..e01c663 100644
--- a/vp9/common/vp9_idct.h
+++ b/vp9/common/vp9_idct.h
@@ -91,7 +91,7 @@
void vp9_idct4x4_add(int16_t *input, uint8_t *dest, int stride, int eob);
void vp9_iwht4x4_add(int16_t *input, uint8_t *dest, int stride, int eob);
void vp9_idct8x8_add(int16_t *input, uint8_t *dest, int stride, int eob);
-void vp9_idct_add_16x16(int16_t *input, uint8_t *dest, int stride, int eob);
+void vp9_idct16x16_add(int16_t *input, uint8_t *dest, int stride, int eob);
void vp9_idct_add_32x32(int16_t *input, uint8_t *dest, int stride, int eob);
void vp9_iht_add(TX_TYPE tx_type, int16_t *input, uint8_t *dest,
diff --git a/vp9/common/vp9_rtcd_defs.sh b/vp9/common/vp9_rtcd_defs.sh
index 244064f..9aaf848 100644
--- a/vp9/common/vp9_rtcd_defs.sh
+++ b/vp9/common/vp9_rtcd_defs.sh
@@ -282,14 +282,14 @@
prototype void vp9_idct8x8_10_add "int16_t *input, uint8_t *dest, int dest_stride"
specialize vp9_idct8x8_10_add sse2 neon
-prototype void vp9_short_idct16x16_1_add "int16_t *input, uint8_t *dest, int dest_stride"
-specialize vp9_short_idct16x16_1_add sse2 neon
+prototype void vp9_idct16x16_1_add "int16_t *input, uint8_t *dest, int dest_stride"
+specialize vp9_idct16x16_1_add sse2 neon
-prototype void vp9_short_idct16x16_add "int16_t *input, uint8_t *dest, int dest_stride"
-specialize vp9_short_idct16x16_add sse2 neon
+prototype void vp9_idct16x16_256_add "int16_t *input, uint8_t *dest, int dest_stride"
+specialize vp9_idct16x16_256_add sse2 neon
-prototype void vp9_short_idct16x16_10_add "int16_t *input, uint8_t *dest, int dest_stride"
-specialize vp9_short_idct16x16_10_add sse2 neon
+prototype void vp9_idct16x16_10_add "int16_t *input, uint8_t *dest, int dest_stride"
+specialize vp9_idct16x16_10_add sse2 neon
prototype void vp9_short_idct32x32_add "int16_t *input, uint8_t *dest, int dest_stride"
specialize vp9_short_idct32x32_add sse2 neon
diff --git a/vp9/common/vp9_scan.c b/vp9/common/vp9_scan.c
new file mode 100644
index 0000000..f17da91
--- /dev/null
+++ b/vp9/common/vp9_scan.c
@@ -0,0 +1,357 @@
+/*
+ * Copyright (c) 2010 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 <assert.h>
+
+#include "vp9/common/vp9_scan.h"
+
+DECLARE_ALIGNED(16, const int16_t, vp9_default_scan_4x4[16]) = {
+ 0, 4, 1, 5,
+ 8, 2, 12, 9,
+ 3, 6, 13, 10,
+ 7, 14, 11, 15,
+};
+
+DECLARE_ALIGNED(16, const int16_t, vp9_col_scan_4x4[16]) = {
+ 0, 4, 8, 1,
+ 12, 5, 9, 2,
+ 13, 6, 10, 3,
+ 7, 14, 11, 15,
+};
+
+DECLARE_ALIGNED(16, const int16_t, vp9_row_scan_4x4[16]) = {
+ 0, 1, 4, 2,
+ 5, 3, 6, 8,
+ 9, 7, 12, 10,
+ 13, 11, 14, 15,
+};
+
+DECLARE_ALIGNED(16, const int16_t, vp9_default_scan_8x8[64]) = {
+ 0, 8, 1, 16, 9, 2, 17, 24,
+ 10, 3, 18, 25, 32, 11, 4, 26,
+ 33, 19, 40, 12, 34, 27, 5, 41,
+ 20, 48, 13, 35, 42, 28, 21, 6,
+ 49, 56, 36, 43, 29, 7, 14, 50,
+ 57, 44, 22, 37, 15, 51, 58, 30,
+ 45, 23, 52, 59, 38, 31, 60, 53,
+ 46, 39, 61, 54, 47, 62, 55, 63,
+};
+
+DECLARE_ALIGNED(16, const int16_t, vp9_col_scan_8x8[64]) = {
+ 0, 8, 16, 1, 24, 9, 32, 17,
+ 2, 40, 25, 10, 33, 18, 48, 3,
+ 26, 41, 11, 56, 19, 34, 4, 49,
+ 27, 42, 12, 35, 20, 57, 50, 28,
+ 5, 43, 13, 36, 58, 51, 21, 44,
+ 6, 29, 59, 37, 14, 52, 22, 7,
+ 45, 60, 30, 15, 38, 53, 23, 46,
+ 31, 61, 39, 54, 47, 62, 55, 63,
+};
+
+DECLARE_ALIGNED(16, const int16_t, vp9_row_scan_8x8[64]) = {
+ 0, 1, 2, 8, 9, 3, 16, 10,
+ 4, 17, 11, 24, 5, 18, 25, 12,
+ 19, 26, 32, 6, 13, 20, 33, 27,
+ 7, 34, 40, 21, 28, 41, 14, 35,
+ 48, 42, 29, 36, 49, 22, 43, 15,
+ 56, 37, 50, 44, 30, 57, 23, 51,
+ 58, 45, 38, 52, 31, 59, 53, 46,
+ 60, 39, 61, 47, 54, 55, 62, 63,
+};
+
+DECLARE_ALIGNED(16, const int16_t, vp9_default_scan_16x16[256]) = {
+ 0, 16, 1, 32, 17, 2, 48, 33, 18, 3, 64, 34, 49, 19, 65, 80,
+ 50, 4, 35, 66, 20, 81, 96, 51, 5, 36, 82, 97, 67, 112, 21, 52,
+ 98, 37, 83, 113, 6, 68, 128, 53, 22, 99, 114, 84, 7, 129, 38, 69,
+ 100, 115, 144, 130, 85, 54, 23, 8, 145, 39, 70, 116, 101, 131, 160, 146,
+ 55, 86, 24, 71, 132, 117, 161, 40, 9, 102, 147, 176, 162, 87, 56, 25,
+ 133, 118, 177, 148, 72, 103, 41, 163, 10, 192, 178, 88, 57, 134, 149, 119,
+ 26, 164, 73, 104, 193, 42, 179, 208, 11, 135, 89, 165, 120, 150, 58, 194,
+ 180, 27, 74, 209, 105, 151, 136, 43, 90, 224, 166, 195, 181, 121, 210, 59,
+ 12, 152, 106, 167, 196, 75, 137, 225, 211, 240, 182, 122, 91, 28, 197, 13,
+ 226, 168, 183, 153, 44, 212, 138, 107, 241, 60, 29, 123, 198, 184, 227, 169,
+ 242, 76, 213, 154, 45, 92, 14, 199, 139, 61, 228, 214, 170, 185, 243, 108,
+ 77, 155, 30, 15, 200, 229, 124, 215, 244, 93, 46, 186, 171, 201, 109, 140,
+ 230, 62, 216, 245, 31, 125, 78, 156, 231, 47, 187, 202, 217, 94, 246, 141,
+ 63, 232, 172, 110, 247, 157, 79, 218, 203, 126, 233, 188, 248, 95, 173, 142,
+ 219, 111, 249, 234, 158, 127, 189, 204, 250, 235, 143, 174, 220, 205, 159,
+ 251,
+ 190, 221, 175, 236, 237, 191, 206, 252, 222, 253, 207, 238, 223, 254, 239,
+ 255,
+};
+
+DECLARE_ALIGNED(16, const int16_t, vp9_col_scan_16x16[256]) = {
+ 0, 16, 32, 48, 1, 64, 17, 80, 33, 96, 49, 2, 65, 112, 18, 81,
+ 34, 128, 50, 97, 3, 66, 144, 19, 113, 35, 82, 160, 98, 51, 129, 4,
+ 67, 176, 20, 114, 145, 83, 36, 99, 130, 52, 192, 5, 161, 68, 115, 21,
+ 146, 84, 208, 177, 37, 131, 100, 53, 162, 224, 69, 6, 116, 193, 147, 85,
+ 22, 240, 132, 38, 178, 101, 163, 54, 209, 117, 70, 7, 148, 194, 86, 179,
+ 225, 23, 133, 39, 164, 8, 102, 210, 241, 55, 195, 118, 149, 71, 180, 24,
+ 87, 226, 134, 165, 211, 40, 103, 56, 72, 150, 196, 242, 119, 9, 181, 227,
+ 88, 166, 25, 135, 41, 104, 212, 57, 151, 197, 120, 73, 243, 182, 136, 167,
+ 213, 89, 10, 228, 105, 152, 198, 26, 42, 121, 183, 244, 168, 58, 137, 229,
+ 74, 214, 90, 153, 199, 184, 11, 106, 245, 27, 122, 230, 169, 43, 215, 59,
+ 200, 138, 185, 246, 75, 12, 91, 154, 216, 231, 107, 28, 44, 201, 123, 170,
+ 60, 247, 232, 76, 139, 13, 92, 217, 186, 248, 155, 108, 29, 124, 45, 202,
+ 233, 171, 61, 14, 77, 140, 15, 249, 93, 30, 187, 156, 218, 46, 109, 125,
+ 62, 172, 78, 203, 31, 141, 234, 94, 47, 188, 63, 157, 110, 250, 219, 79,
+ 126, 204, 173, 142, 95, 189, 111, 235, 158, 220, 251, 127, 174, 143, 205,
+ 236,
+ 159, 190, 221, 252, 175, 206, 237, 191, 253, 222, 238, 207, 254, 223, 239,
+ 255,
+};
+
+DECLARE_ALIGNED(16, const int16_t, vp9_row_scan_16x16[256]) = {
+ 0, 1, 2, 16, 3, 17, 4, 18, 32, 5, 33, 19, 6, 34, 48, 20,
+ 49, 7, 35, 21, 50, 64, 8, 36, 65, 22, 51, 37, 80, 9, 66, 52,
+ 23, 38, 81, 67, 10, 53, 24, 82, 68, 96, 39, 11, 54, 83, 97, 69,
+ 25, 98, 84, 40, 112, 55, 12, 70, 99, 113, 85, 26, 41, 56, 114, 100,
+ 13, 71, 128, 86, 27, 115, 101, 129, 42, 57, 72, 116, 14, 87, 130, 102,
+ 144, 73, 131, 117, 28, 58, 15, 88, 43, 145, 103, 132, 146, 118, 74, 160,
+ 89, 133, 104, 29, 59, 147, 119, 44, 161, 148, 90, 105, 134, 162, 120, 176,
+ 75, 135, 149, 30, 60, 163, 177, 45, 121, 91, 106, 164, 178, 150, 192, 136,
+ 165, 179, 31, 151, 193, 76, 122, 61, 137, 194, 107, 152, 180, 208, 46, 166,
+ 167, 195, 92, 181, 138, 209, 123, 153, 224, 196, 77, 168, 210, 182, 240, 108,
+ 197, 62, 154, 225, 183, 169, 211, 47, 139, 93, 184, 226, 212, 241, 198, 170,
+ 124, 155, 199, 78, 213, 185, 109, 227, 200, 63, 228, 242, 140, 214, 171, 186,
+ 156, 229, 243, 125, 94, 201, 244, 215, 216, 230, 141, 187, 202, 79, 172, 110,
+ 157, 245, 217, 231, 95, 246, 232, 126, 203, 247, 233, 173, 218, 142, 111,
+ 158,
+ 188, 248, 127, 234, 219, 249, 189, 204, 143, 174, 159, 250, 235, 205, 220,
+ 175,
+ 190, 251, 221, 191, 206, 236, 207, 237, 252, 222, 253, 223, 238, 239, 254,
+ 255,
+};
+
+DECLARE_ALIGNED(16, const int16_t, vp9_default_scan_32x32[1024]) = {
+ 0, 32, 1, 64, 33, 2, 96, 65, 34, 128, 3, 97, 66, 160,
+ 129, 35, 98, 4, 67, 130, 161, 192, 36, 99, 224, 5, 162, 193,
+ 68, 131, 37, 100,
+ 225, 194, 256, 163, 69, 132, 6, 226, 257, 288, 195, 101, 164, 38,
+ 258, 7, 227, 289, 133, 320, 70, 196, 165, 290, 259, 228, 39, 321,
+ 102, 352, 8, 197,
+ 71, 134, 322, 291, 260, 353, 384, 229, 166, 103, 40, 354, 323, 292,
+ 135, 385, 198, 261, 72, 9, 416, 167, 386, 355, 230, 324, 104, 293,
+ 41, 417, 199, 136,
+ 262, 387, 448, 325, 356, 10, 73, 418, 231, 168, 449, 294, 388, 105,
+ 419, 263, 42, 200, 357, 450, 137, 480, 74, 326, 232, 11, 389, 169,
+ 295, 420, 106, 451,
+ 481, 358, 264, 327, 201, 43, 138, 512, 482, 390, 296, 233, 170, 421,
+ 75, 452, 359, 12, 513, 265, 483, 328, 107, 202, 514, 544, 422, 391,
+ 453, 139, 44, 234,
+ 484, 297, 360, 171, 76, 515, 545, 266, 329, 454, 13, 423, 203, 108,
+ 546, 485, 576, 298, 235, 140, 361, 330, 172, 547, 45, 455, 267, 577,
+ 486, 77, 204, 362,
+ 608, 14, 299, 578, 109, 236, 487, 609, 331, 141, 579, 46, 15, 173,
+ 610, 363, 78, 205, 16, 110, 237, 611, 142, 47, 174, 79, 206, 17,
+ 111, 238, 48, 143,
+ 80, 175, 112, 207, 49, 18, 239, 81, 113, 19, 50, 82, 114, 51,
+ 83, 115, 640, 516, 392, 268, 144, 20, 672, 641, 548, 517, 424,
+ 393, 300, 269, 176, 145,
+ 52, 21, 704, 673, 642, 580, 549, 518, 456, 425, 394, 332, 301,
+ 270, 208, 177, 146, 84, 53, 22, 736, 705, 674, 643, 612, 581,
+ 550, 519, 488, 457, 426, 395,
+ 364, 333, 302, 271, 240, 209, 178, 147, 116, 85, 54, 23, 737,
+ 706, 675, 613, 582, 551, 489, 458, 427, 365, 334, 303, 241,
+ 210, 179, 117, 86, 55, 738, 707,
+ 614, 583, 490, 459, 366, 335, 242, 211, 118, 87, 739, 615, 491,
+ 367, 243, 119, 768, 644, 520, 396, 272, 148, 24, 800, 769, 676,
+ 645, 552, 521, 428, 397, 304,
+ 273, 180, 149, 56, 25, 832, 801, 770, 708, 677, 646, 584, 553,
+ 522, 460, 429, 398, 336, 305, 274, 212, 181, 150, 88, 57, 26,
+ 864, 833, 802, 771, 740, 709,
+ 678, 647, 616, 585, 554, 523, 492, 461, 430, 399, 368, 337, 306,
+ 275, 244, 213, 182, 151, 120, 89, 58, 27, 865, 834, 803, 741,
+ 710, 679, 617, 586, 555, 493,
+ 462, 431, 369, 338, 307, 245, 214, 183, 121, 90, 59, 866, 835,
+ 742, 711, 618, 587, 494, 463, 370, 339, 246, 215, 122, 91, 867,
+ 743, 619, 495, 371, 247, 123,
+ 896, 772, 648, 524, 400, 276, 152, 28, 928, 897, 804, 773, 680,
+ 649, 556, 525, 432, 401, 308, 277, 184, 153, 60, 29, 960, 929,
+ 898, 836, 805, 774, 712, 681,
+ 650, 588, 557, 526, 464, 433, 402, 340, 309, 278, 216, 185, 154,
+ 92, 61, 30, 992, 961, 930, 899, 868, 837, 806, 775, 744, 713, 682,
+ 651, 620, 589, 558, 527,
+ 496, 465, 434, 403, 372, 341, 310, 279, 248, 217, 186, 155, 124,
+ 93, 62, 31, 993, 962, 931, 869, 838, 807, 745, 714, 683, 621, 590,
+ 559, 497, 466, 435, 373,
+ 342, 311, 249, 218, 187, 125, 94, 63, 994, 963, 870, 839, 746, 715,
+ 622, 591, 498, 467, 374, 343, 250, 219, 126, 95, 995, 871, 747, 623,
+ 499, 375, 251, 127,
+ 900, 776, 652, 528, 404, 280, 156, 932, 901, 808, 777, 684, 653, 560,
+ 529, 436, 405, 312, 281, 188, 157, 964, 933, 902, 840, 809, 778, 716,
+ 685, 654, 592, 561,
+ 530, 468, 437, 406, 344, 313, 282, 220, 189, 158, 996, 965, 934, 903,
+ 872, 841, 810, 779, 748, 717, 686, 655, 624, 593, 562, 531, 500, 469,
+ 438, 407, 376, 345,
+ 314, 283, 252, 221, 190, 159, 997, 966, 935, 873, 842, 811, 749, 718,
+ 687, 625, 594, 563, 501, 470, 439, 377, 346, 315, 253, 222, 191, 998,
+ 967, 874, 843, 750,
+ 719, 626, 595, 502, 471, 378, 347, 254, 223, 999, 875, 751, 627, 503,
+ 379, 255, 904, 780, 656, 532, 408, 284, 936, 905, 812, 781, 688, 657,
+ 564, 533, 440, 409,
+ 316, 285, 968, 937, 906, 844, 813, 782, 720, 689, 658, 596, 565, 534,
+ 472, 441, 410, 348, 317, 286, 1000, 969, 938, 907, 876, 845, 814, 783,
+ 752, 721, 690, 659,
+ 628, 597, 566, 535, 504, 473, 442, 411, 380, 349, 318, 287, 1001, 970,
+ 939, 877, 846, 815, 753, 722, 691, 629, 598, 567, 505, 474, 443, 381,
+ 350, 319, 1002, 971,
+ 878, 847, 754, 723, 630, 599, 506, 475, 382, 351, 1003, 879, 755, 631,
+ 507, 383, 908, 784, 660, 536, 412, 940, 909, 816, 785, 692, 661, 568,
+ 537, 444, 413, 972,
+ 941, 910, 848, 817, 786, 724, 693, 662, 600, 569, 538, 476, 445, 414,
+ 1004, 973, 942, 911, 880, 849, 818, 787, 756, 725, 694, 663, 632, 601,
+ 570, 539, 508, 477,
+ 446, 415, 1005, 974, 943, 881, 850, 819, 757, 726, 695, 633, 602, 571,
+ 509, 478, 447, 1006, 975, 882, 851, 758, 727, 634, 603, 510, 479,
+ 1007, 883, 759, 635, 511,
+ 912, 788, 664, 540, 944, 913, 820, 789, 696, 665, 572, 541, 976, 945,
+ 914, 852, 821, 790, 728, 697, 666, 604, 573, 542, 1008, 977, 946, 915,
+ 884, 853, 822, 791,
+ 760, 729, 698, 667, 636, 605, 574, 543, 1009, 978, 947, 885, 854, 823,
+ 761, 730, 699, 637, 606, 575, 1010, 979, 886, 855, 762, 731, 638, 607,
+ 1011, 887, 763, 639,
+ 916, 792, 668, 948, 917, 824, 793, 700, 669, 980, 949, 918, 856, 825,
+ 794, 732, 701, 670, 1012, 981, 950, 919, 888, 857, 826, 795, 764, 733,
+ 702, 671, 1013, 982,
+ 951, 889, 858, 827, 765, 734, 703, 1014, 983, 890, 859, 766, 735, 1015,
+ 891, 767, 920, 796, 952, 921, 828, 797, 984, 953, 922, 860, 829, 798,
+ 1016, 985, 954, 923,
+ 892, 861, 830, 799, 1017, 986, 955, 893, 862, 831, 1018, 987, 894, 863,
+ 1019, 895, 924, 956, 925, 988, 957, 926, 1020, 989, 958, 927, 1021,
+ 990, 959, 1022, 991, 1023,
+};
+
+// Neighborhood 5-tuples for various scans and blocksizes,
+// in {top, left, topleft, topright, bottomleft} order
+// for each position in raster scan order.
+// -1 indicates the neighbor does not exist.
+DECLARE_ALIGNED(16, int16_t,
+ vp9_default_scan_4x4_neighbors[17 * MAX_NEIGHBORS]);
+DECLARE_ALIGNED(16, int16_t,
+ vp9_col_scan_4x4_neighbors[17 * MAX_NEIGHBORS]);
+DECLARE_ALIGNED(16, int16_t,
+ vp9_row_scan_4x4_neighbors[17 * MAX_NEIGHBORS]);
+DECLARE_ALIGNED(16, int16_t,
+ vp9_col_scan_8x8_neighbors[65 * MAX_NEIGHBORS]);
+DECLARE_ALIGNED(16, int16_t,
+ vp9_row_scan_8x8_neighbors[65 * MAX_NEIGHBORS]);
+DECLARE_ALIGNED(16, int16_t,
+ vp9_default_scan_8x8_neighbors[65 * MAX_NEIGHBORS]);
+DECLARE_ALIGNED(16, int16_t,
+ vp9_col_scan_16x16_neighbors[257 * MAX_NEIGHBORS]);
+DECLARE_ALIGNED(16, int16_t,
+ vp9_row_scan_16x16_neighbors[257 * MAX_NEIGHBORS]);
+DECLARE_ALIGNED(16, int16_t,
+ vp9_default_scan_16x16_neighbors[257 * MAX_NEIGHBORS]);
+DECLARE_ALIGNED(16, int16_t,
+ vp9_default_scan_32x32_neighbors[1025 * MAX_NEIGHBORS]);
+
+
+DECLARE_ALIGNED(16, int16_t, vp9_default_iscan_4x4[16]);
+DECLARE_ALIGNED(16, int16_t, vp9_col_iscan_4x4[16]);
+DECLARE_ALIGNED(16, int16_t, vp9_row_iscan_4x4[16]);
+DECLARE_ALIGNED(16, int16_t, vp9_col_iscan_8x8[64]);
+DECLARE_ALIGNED(16, int16_t, vp9_row_iscan_8x8[64]);
+DECLARE_ALIGNED(16, int16_t, vp9_default_iscan_8x8[64]);
+DECLARE_ALIGNED(16, int16_t, vp9_col_iscan_16x16[256]);
+DECLARE_ALIGNED(16, int16_t, vp9_row_iscan_16x16[256]);
+DECLARE_ALIGNED(16, int16_t, vp9_default_iscan_16x16[256]);
+DECLARE_ALIGNED(16, int16_t, vp9_default_iscan_32x32[1024]);
+
+static int find_in_scan(const int16_t *scan, int l, int idx) {
+ int n, l2 = l * l;
+ for (n = 0; n < l2; n++) {
+ int rc = scan[n];
+ if (rc == idx)
+ return n;
+ }
+ assert(0);
+ return -1;
+}
+static void init_scan_neighbors(const int16_t *scan,
+ int16_t *iscan,
+ int l, int16_t *neighbors) {
+ int l2 = l * l;
+ int n, i, j;
+
+ // dc doesn't use this type of prediction
+ neighbors[MAX_NEIGHBORS * 0 + 0] = 0;
+ neighbors[MAX_NEIGHBORS * 0 + 1] = 0;
+ iscan[0] = find_in_scan(scan, l, 0);
+ for (n = 1; n < l2; n++) {
+ int rc = scan[n];
+ iscan[n] = find_in_scan(scan, l, n);
+ i = rc / l;
+ j = rc % l;
+ if (i > 0 && j > 0) {
+ // col/row scan is used for adst/dct, and generally means that
+ // energy decreases to zero much faster in the dimension in
+ // which ADST is used compared to the direction in which DCT
+ // is used. Likewise, we find much higher correlation between
+ // coefficients within the direction in which DCT is used.
+ // Therefore, if we use ADST/DCT, prefer the DCT neighbor coeff
+ // as a context. If ADST or DCT is used in both directions, we
+ // use the combination of the two as a context.
+ int a = (i - 1) * l + j;
+ int b = i * l + j - 1;
+ if (scan == vp9_col_scan_4x4 || scan == vp9_col_scan_8x8 ||
+ scan == vp9_col_scan_16x16) {
+ // in the col/row scan cases (as well as left/top edge cases), we set
+ // both contexts to the same value, so we can branchlessly do a+b+1>>1
+ // which automatically becomes a if a == b
+ neighbors[MAX_NEIGHBORS * n + 0] =
+ neighbors[MAX_NEIGHBORS * n + 1] = a;
+ } else if (scan == vp9_row_scan_4x4 || scan == vp9_row_scan_8x8 ||
+ scan == vp9_row_scan_16x16) {
+ neighbors[MAX_NEIGHBORS * n + 0] =
+ neighbors[MAX_NEIGHBORS * n + 1] = b;
+ } else {
+ neighbors[MAX_NEIGHBORS * n + 0] = a;
+ neighbors[MAX_NEIGHBORS * n + 1] = b;
+ }
+ } else if (i > 0) {
+ neighbors[MAX_NEIGHBORS * n + 0] =
+ neighbors[MAX_NEIGHBORS * n + 1] = (i - 1) * l + j;
+ } else {
+ assert(j > 0);
+ neighbors[MAX_NEIGHBORS * n + 0] =
+ neighbors[MAX_NEIGHBORS * n + 1] = i * l + j - 1;
+ }
+ assert(iscan[neighbors[MAX_NEIGHBORS * n + 0]] < n);
+ }
+ // one padding item so we don't have to add branches in code to handle
+ // calls to get_coef_context() for the token after the final dc token
+ neighbors[MAX_NEIGHBORS * l2 + 0] = 0;
+ neighbors[MAX_NEIGHBORS * l2 + 1] = 0;
+}
+
+void vp9_init_neighbors() {
+ init_scan_neighbors(vp9_default_scan_4x4, vp9_default_iscan_4x4, 4,
+ vp9_default_scan_4x4_neighbors);
+ init_scan_neighbors(vp9_row_scan_4x4, vp9_row_iscan_4x4, 4,
+ vp9_row_scan_4x4_neighbors);
+ init_scan_neighbors(vp9_col_scan_4x4, vp9_col_iscan_4x4, 4,
+ vp9_col_scan_4x4_neighbors);
+ init_scan_neighbors(vp9_default_scan_8x8, vp9_default_iscan_8x8, 8,
+ vp9_default_scan_8x8_neighbors);
+ init_scan_neighbors(vp9_row_scan_8x8, vp9_row_iscan_8x8, 8,
+ vp9_row_scan_8x8_neighbors);
+ init_scan_neighbors(vp9_col_scan_8x8, vp9_col_iscan_8x8, 8,
+ vp9_col_scan_8x8_neighbors);
+ init_scan_neighbors(vp9_default_scan_16x16, vp9_default_iscan_16x16, 16,
+ vp9_default_scan_16x16_neighbors);
+ init_scan_neighbors(vp9_row_scan_16x16, vp9_row_iscan_16x16, 16,
+ vp9_row_scan_16x16_neighbors);
+ init_scan_neighbors(vp9_col_scan_16x16, vp9_col_iscan_16x16, 16,
+ vp9_col_scan_16x16_neighbors);
+ init_scan_neighbors(vp9_default_scan_32x32, vp9_default_iscan_32x32, 32,
+ vp9_default_scan_32x32_neighbors);
+}
diff --git a/vp9/common/vp9_scan.h b/vp9/common/vp9_scan.h
new file mode 100644
index 0000000..a5c8463
--- /dev/null
+++ b/vp9/common/vp9_scan.h
@@ -0,0 +1,200 @@
+/*
+ * 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_SCAN_H_
+#define VP9_COMMON_VP9_SCAN_H_
+
+#include "vpx/vpx_integer.h"
+#include "vpx_ports/mem.h"
+
+#include "vp9/common/vp9_enums.h"
+
+#define MAX_NEIGHBORS 2
+
+extern DECLARE_ALIGNED(16, const int16_t, vp9_default_scan_4x4[16]);
+extern DECLARE_ALIGNED(16, const int16_t, vp9_col_scan_4x4[16]);
+extern DECLARE_ALIGNED(16, const int16_t, vp9_row_scan_4x4[16]);
+
+extern DECLARE_ALIGNED(16, const int16_t, vp9_default_scan_8x8[64]);
+extern DECLARE_ALIGNED(16, const int16_t, vp9_col_scan_8x8[64]);
+extern DECLARE_ALIGNED(16, const int16_t, vp9_row_scan_8x8[64]);
+
+extern DECLARE_ALIGNED(16, const int16_t, vp9_default_scan_16x16[256]);
+extern DECLARE_ALIGNED(16, const int16_t, vp9_col_scan_16x16[256]);
+extern DECLARE_ALIGNED(16, const int16_t, vp9_row_scan_16x16[256]);
+
+extern DECLARE_ALIGNED(16, const int16_t, vp9_default_scan_32x32[1024]);
+
+extern DECLARE_ALIGNED(16, int16_t, vp9_default_iscan_4x4[16]);
+extern DECLARE_ALIGNED(16, int16_t, vp9_col_iscan_4x4[16]);
+extern DECLARE_ALIGNED(16, int16_t, vp9_row_iscan_4x4[16]);
+
+extern DECLARE_ALIGNED(16, int16_t, vp9_default_iscan_8x8[64]);
+extern DECLARE_ALIGNED(16, int16_t, vp9_col_iscan_8x8[64]);
+extern DECLARE_ALIGNED(16, int16_t, vp9_row_iscan_8x8[64]);
+
+extern DECLARE_ALIGNED(16, int16_t, vp9_default_iscan_16x16[256]);
+extern DECLARE_ALIGNED(16, int16_t, vp9_col_iscan_16x16[256]);
+extern DECLARE_ALIGNED(16, int16_t, vp9_row_iscan_16x16[256]);
+
+extern DECLARE_ALIGNED(16, int16_t, vp9_default_iscan_32x32[1024]);
+
+extern DECLARE_ALIGNED(16, int16_t,
+ vp9_default_scan_4x4_neighbors[17 * MAX_NEIGHBORS]);
+extern DECLARE_ALIGNED(16, int16_t,
+ vp9_col_scan_4x4_neighbors[17 * MAX_NEIGHBORS]);
+extern DECLARE_ALIGNED(16, int16_t,
+ vp9_row_scan_4x4_neighbors[17 * MAX_NEIGHBORS]);
+extern DECLARE_ALIGNED(16, int16_t,
+ vp9_col_scan_8x8_neighbors[65 * MAX_NEIGHBORS]);
+extern DECLARE_ALIGNED(16, int16_t,
+ vp9_row_scan_8x8_neighbors[65 * MAX_NEIGHBORS]);
+extern DECLARE_ALIGNED(16, int16_t,
+ vp9_default_scan_8x8_neighbors[65 * MAX_NEIGHBORS]);
+extern DECLARE_ALIGNED(16, int16_t,
+ vp9_col_scan_16x16_neighbors[257 * MAX_NEIGHBORS]);
+extern DECLARE_ALIGNED(16, int16_t,
+ vp9_row_scan_16x16_neighbors[257 * MAX_NEIGHBORS]);
+extern DECLARE_ALIGNED(16, int16_t,
+ vp9_default_scan_16x16_neighbors[257 * MAX_NEIGHBORS]);
+extern DECLARE_ALIGNED(16, int16_t,
+ vp9_default_scan_32x32_neighbors[1025 * MAX_NEIGHBORS]);
+
+
+void vp9_init_neighbors();
+
+static INLINE const int16_t* get_scan_4x4(TX_TYPE tx_type) {
+ switch (tx_type) {
+ case ADST_DCT:
+ return vp9_row_scan_4x4;
+ case DCT_ADST:
+ return vp9_col_scan_4x4;
+ default:
+ return vp9_default_scan_4x4;
+ }
+}
+
+static INLINE void get_scan_nb_4x4(TX_TYPE tx_type,
+ const int16_t **scan, const int16_t **nb) {
+ switch (tx_type) {
+ case ADST_DCT:
+ *scan = vp9_row_scan_4x4;
+ *nb = vp9_row_scan_4x4_neighbors;
+ break;
+ case DCT_ADST:
+ *scan = vp9_col_scan_4x4;
+ *nb = vp9_col_scan_4x4_neighbors;
+ break;
+ default:
+ *scan = vp9_default_scan_4x4;
+ *nb = vp9_default_scan_4x4_neighbors;
+ break;
+ }
+}
+
+static INLINE const int16_t* get_iscan_4x4(TX_TYPE tx_type) {
+ switch (tx_type) {
+ case ADST_DCT:
+ return vp9_row_iscan_4x4;
+ case DCT_ADST:
+ return vp9_col_iscan_4x4;
+ default:
+ return vp9_default_iscan_4x4;
+ }
+}
+
+static INLINE const int16_t* get_scan_8x8(TX_TYPE tx_type) {
+ switch (tx_type) {
+ case ADST_DCT:
+ return vp9_row_scan_8x8;
+ case DCT_ADST:
+ return vp9_col_scan_8x8;
+ default:
+ return vp9_default_scan_8x8;
+ }
+}
+
+static INLINE void get_scan_nb_8x8(TX_TYPE tx_type,
+ const int16_t **scan, const int16_t **nb) {
+ switch (tx_type) {
+ case ADST_DCT:
+ *scan = vp9_row_scan_8x8;
+ *nb = vp9_row_scan_8x8_neighbors;
+ break;
+ case DCT_ADST:
+ *scan = vp9_col_scan_8x8;
+ *nb = vp9_col_scan_8x8_neighbors;
+ break;
+ default:
+ *scan = vp9_default_scan_8x8;
+ *nb = vp9_default_scan_8x8_neighbors;
+ break;
+ }
+}
+
+static INLINE const int16_t* get_iscan_8x8(TX_TYPE tx_type) {
+ switch (tx_type) {
+ case ADST_DCT:
+ return vp9_row_iscan_8x8;
+ case DCT_ADST:
+ return vp9_col_iscan_8x8;
+ default:
+ return vp9_default_iscan_8x8;
+ }
+}
+
+static INLINE const int16_t* get_scan_16x16(TX_TYPE tx_type) {
+ switch (tx_type) {
+ case ADST_DCT:
+ return vp9_row_scan_16x16;
+ case DCT_ADST:
+ return vp9_col_scan_16x16;
+ default:
+ return vp9_default_scan_16x16;
+ }
+}
+
+static INLINE void get_scan_nb_16x16(TX_TYPE tx_type,
+ const int16_t **scan, const int16_t **nb) {
+ switch (tx_type) {
+ case ADST_DCT:
+ *scan = vp9_row_scan_16x16;
+ *nb = vp9_row_scan_16x16_neighbors;
+ break;
+ case DCT_ADST:
+ *scan = vp9_col_scan_16x16;
+ *nb = vp9_col_scan_16x16_neighbors;
+ break;
+ default:
+ *scan = vp9_default_scan_16x16;
+ *nb = vp9_default_scan_16x16_neighbors;
+ break;
+ }
+}
+
+static INLINE const int16_t* get_iscan_16x16(TX_TYPE tx_type) {
+ switch (tx_type) {
+ case ADST_DCT:
+ return vp9_row_iscan_16x16;
+ case DCT_ADST:
+ return vp9_col_iscan_16x16;
+ default:
+ return vp9_default_iscan_16x16;
+ }
+}
+
+static INLINE int get_coef_context(const int16_t *neighbors,
+ uint8_t *token_cache,
+ int c) {
+ return (1 + token_cache[neighbors[MAX_NEIGHBORS * c + 0]] +
+ token_cache[neighbors[MAX_NEIGHBORS * c + 1]]) >> 1;
+}
+
+#endif // VP9_COMMON_VP9_SCAN_H_
diff --git a/vp9/common/vp9_systemdependent.h b/vp9/common/vp9_systemdependent.h
index cc909e2..b8d161d 100644
--- a/vp9/common/vp9_systemdependent.h
+++ b/vp9/common/vp9_systemdependent.h
@@ -13,6 +13,7 @@
#ifdef _MSC_VER
#include <math.h>
+#define snprintf _snprintf
#endif
#include "./vpx_config.h"
diff --git a/vp9/common/x86/vp9_idct_intrin_sse2.c b/vp9/common/x86/vp9_idct_intrin_sse2.c
index 8433805..c9d0d09 100644
--- a/vp9/common/x86/vp9_idct_intrin_sse2.c
+++ b/vp9/common/x86/vp9_idct_intrin_sse2.c
@@ -1263,7 +1263,7 @@
stp2_10, stp2_13, stp2_11, stp2_12) \
}
-void vp9_short_idct16x16_add_sse2(int16_t *input, uint8_t *dest, int stride) {
+void vp9_idct16x16_256_add_sse2(int16_t *input, uint8_t *dest, int stride) {
const __m128i rounding = _mm_set1_epi32(DCT_CONST_ROUNDING);
const __m128i final_rounding = _mm_set1_epi16(1<<5);
const __m128i zero = _mm_setzero_si128();
@@ -1470,7 +1470,7 @@
}
}
-void vp9_short_idct16x16_1_add_sse2(int16_t *input, uint8_t *dest, int stride) {
+void vp9_idct16x16_1_add_sse2(int16_t *input, uint8_t *dest, int stride) {
__m128i dc_value;
const __m128i zero = _mm_setzero_si128();
int a, i;
@@ -2456,7 +2456,7 @@
write_buffer_8x16(dest, in1, stride);
}
-void vp9_short_idct16x16_10_add_sse2(int16_t *input, uint8_t *dest,
+void vp9_idct16x16_10_add_sse2(int16_t *input, uint8_t *dest,
int stride) {
const __m128i rounding = _mm_set1_epi32(DCT_CONST_ROUNDING);
const __m128i final_rounding = _mm_set1_epi16(1<<5);
diff --git a/vp9/decoder/vp9_detokenize.c b/vp9/decoder/vp9_detokenize.c
index e835b89..a67945c 100644
--- a/vp9/decoder/vp9_detokenize.c
+++ b/vp9/decoder/vp9_detokenize.c
@@ -203,7 +203,6 @@
if (c < seg_eob)
coef_counts[type][ref][band][pt][DCT_EOB_MODEL_TOKEN]++;
-
return c;
}
diff --git a/vp9/encoder/vp9_bitstream.c b/vp9/encoder/vp9_bitstream.c
index 1087536..7619d76 100644
--- a/vp9/encoder/vp9_bitstream.c
+++ b/vp9/encoder/vp9_bitstream.c
@@ -406,8 +406,6 @@
const BLOCK_SIZE bsize = mi->sb_type;
const int allow_hp = xd->allow_high_precision_mv;
- x->partition_info = x->pi + (m - cm->mi);
-
#ifdef ENTROPY_STATS
active_section = 9;
#endif
@@ -489,7 +487,7 @@
for (idy = 0; idy < 2; idy += num_4x4_blocks_high) {
for (idx = 0; idx < 2; idx += num_4x4_blocks_wide) {
const int j = idy * 2 + idx;
- const MB_PREDICTION_MODE blockmode = x->partition_info->bmi[j].mode;
+ const MB_PREDICTION_MODE blockmode = m->bmi[j].as_mode;
write_sb_mv_ref(bc, blockmode, mv_ref_p);
++cm->counts.inter_mode[mi->mode_context[rf]]
[inter_mode_offset(blockmode)];
diff --git a/vp9/encoder/vp9_block.h b/vp9/encoder/vp9_block.h
index 2e28a2e..6314b60 100644
--- a/vp9/encoder/vp9_block.h
+++ b/vp9/encoder/vp9_block.h
@@ -23,16 +23,9 @@
int offset;
} search_site;
-typedef struct {
- struct {
- MB_PREDICTION_MODE mode;
- } bmi[4];
-} PARTITION_INFO;
-
// Structure to hold snapshot of coding context during the mode picking process
typedef struct {
MODE_INFO mic;
- PARTITION_INFO partition_info;
unsigned char zcoeff_blk[256];
int skip;
int_mv best_ref_mv;
@@ -87,9 +80,6 @@
MACROBLOCKD e_mbd;
int skip_block;
- PARTITION_INFO *partition_info; /* work pointer */
- PARTITION_INFO *pi; /* Corresponds to upper left visible macroblock */
- PARTITION_INFO *pip; /* Base of allocated array */
search_site *ss;
int ss_count;
diff --git a/vp9/encoder/vp9_encodeframe.c b/vp9/encoder/vp9_encodeframe.c
index d7fb8a2..631a276 100644
--- a/vp9/encoder/vp9_encodeframe.c
+++ b/vp9/encoder/vp9_encodeframe.c
@@ -382,7 +382,6 @@
}
if (is_inter_block(mbmi) && mbmi->sb_type < BLOCK_8X8) {
- *x->partition_info = ctx->partition_info;
mbmi->mv[0].as_int = mi->bmi[3].as_mv[0].as_int;
mbmi->mv[1].as_int = mi->bmi[3].as_mv[1].as_int;
}
@@ -493,9 +492,6 @@
x->mb_activity_ptr = &cpi->mb_activity_map[idx_map];
x->active_ptr = cpi->active_map + idx_map;
- /* pointers to mode info contexts */
- x->partition_info = x->pi + idx_str;
-
xd->mi_8x8 = cm->mi_grid_visible + idx_str;
xd->prev_mi_8x8 = cm->prev_mi_grid_visible + idx_str;
diff --git a/vp9/encoder/vp9_encodemb.c b/vp9/encoder/vp9_encodemb.c
index 1e9176e..5c3977d 100644
--- a/vp9/encoder/vp9_encodemb.c
+++ b/vp9/encoder/vp9_encodemb.c
@@ -445,7 +445,7 @@
vp9_short_idct32x32_add(dqcoeff, dst, pd->dst.stride);
break;
case TX_16X16:
- vp9_idct_add_16x16(dqcoeff, dst, pd->dst.stride, pd->eobs[block]);
+ vp9_idct16x16_add(dqcoeff, dst, pd->dst.stride, pd->eobs[block]);
break;
case TX_8X8:
vp9_idct8x8_add(dqcoeff, dst, pd->dst.stride, pd->eobs[block]);
diff --git a/vp9/encoder/vp9_encodemv.c b/vp9/encoder/vp9_encodemv.c
index 04a4172..9ebcc49 100644
--- a/vp9/encoder/vp9_encodemv.c
+++ b/vp9/encoder/vp9_encodemv.c
@@ -337,7 +337,7 @@
for (idy = 0; idy < 2; idy += num_4x4_h) {
for (idx = 0; idx < 2; idx += num_4x4_w) {
const int i = idy * 2 + idx;
- if (x->partition_info->bmi[i].mode == NEWMV)
+ if (mi->bmi[i].as_mode == NEWMV)
inc_mvs(mi->bmi[i].as_mv, best_ref_mv, is_compound, &cpi->NMVcount);
}
}
diff --git a/vp9/encoder/vp9_firstpass.c b/vp9/encoder/vp9_firstpass.c
index fee0f15..b2becbb 100644
--- a/vp9/encoder/vp9_firstpass.c
+++ b/vp9/encoder/vp9_firstpass.c
@@ -515,7 +515,6 @@
setup_pre_planes(xd, 0, lst_yv12, 0, 0, NULL);
setup_dst_planes(xd, new_yv12, 0, 0);
- x->partition_info = x->pi;
xd->mi_8x8 = cm->mi_grid_visible;
// required for vp9_frame_init_quantizer
xd->this_mi =
diff --git a/vp9/encoder/vp9_onyx_if.c b/vp9/encoder/vp9_onyx_if.c
index 0430751..342ff84 100644
--- a/vp9/encoder/vp9_onyx_if.c
+++ b/vp9/encoder/vp9_onyx_if.c
@@ -310,9 +310,6 @@
cpi->mb_activity_map = 0;
vpx_free(cpi->mb_norm_activity_map);
cpi->mb_norm_activity_map = 0;
-
- vpx_free(cpi->mb.pip);
- cpi->mb.pip = 0;
}
// Computes a q delta (in "q index" terms) to get from a starting q value
@@ -780,7 +777,8 @@
? USE_FULL_RD : USE_LARGESTALL;
if (MIN(cpi->common.width, cpi->common.height) >= 720)
- sf->disable_split_mask = DISABLE_ALL_SPLIT;
+ sf->disable_split_mask = cpi->common.show_frame ?
+ DISABLE_ALL_SPLIT : DISABLE_ALL_INTER_SPLIT;
else
sf->disable_split_mask = DISABLE_COMPOUND_SPLIT;
@@ -798,10 +796,12 @@
? USE_FULL_RD : USE_LARGESTALL);
if (MIN(cpi->common.width, cpi->common.height) >= 720)
- sf->disable_split_mask = DISABLE_ALL_SPLIT;
+ sf->disable_split_mask = cpi->common.show_frame ?
+ DISABLE_ALL_SPLIT : DISABLE_ALL_INTER_SPLIT;
else
sf->disable_split_mask = LAST_AND_INTRA_SPLIT_ONLY;
+
sf->mode_search_skip_flags = FLAG_SKIP_INTRA_DIRMISMATCH |
FLAG_SKIP_INTRA_BESTINTER |
FLAG_SKIP_COMP_BESTINTRA |
@@ -984,20 +984,6 @@
"Failed to allocate altref buffer");
}
-static int alloc_partition_data(VP9_COMP *cpi) {
- vpx_free(cpi->mb.pip);
-
- cpi->mb.pip = vpx_calloc(cpi->common.mode_info_stride *
- (cpi->common.mi_rows + MI_BLOCK_SIZE),
- sizeof(PARTITION_INFO));
- if (!cpi->mb.pip)
- return 1;
-
- cpi->mb.pi = cpi->mb.pip + cpi->common.mode_info_stride + 1;
-
- return 0;
-}
-
void vp9_alloc_compressor_data(VP9_COMP *cpi) {
VP9_COMMON *cm = &cpi->common;
@@ -1005,10 +991,6 @@
vpx_internal_error(&cpi->common.error, VPX_CODEC_MEM_ERROR,
"Failed to allocate frame buffers");
- if (alloc_partition_data(cpi))
- vpx_internal_error(&cpi->common.error, VPX_CODEC_MEM_ERROR,
- "Failed to allocate partition data");
-
if (vp9_alloc_frame_buffer(&cpi->last_frame_uf,
cm->width, cm->height,
cm->subsampling_x, cm->subsampling_y,
diff --git a/vp9/encoder/vp9_onyx_int.h b/vp9/encoder/vp9_onyx_int.h
index 2652929..b271f55 100644
--- a/vp9/encoder/vp9_onyx_int.h
+++ b/vp9/encoder/vp9_onyx_int.h
@@ -36,7 +36,7 @@
#define DISABLE_RC_LONG_TERM_MEM 0
#endif
-#define MODE_TEST_HIT_STATS
+// #define MODE_TEST_HIT_STATS
// #define SPEEDSTATS 1
#if CONFIG_MULTIPLE_ARF
diff --git a/vp9/encoder/vp9_rdopt.c b/vp9/encoder/vp9_rdopt.c
index 6ed0e5d..dce73ee 100644
--- a/vp9/encoder/vp9_rdopt.c
+++ b/vp9/encoder/vp9_rdopt.c
@@ -107,8 +107,12 @@
static int rd_thresh_block_size_factor[BLOCK_SIZES] =
{2, 3, 3, 4, 6, 6, 8, 12, 12, 16, 24, 24, 32};
-#define MAX_RD_THRESH_FACT 64
-#define RD_THRESH_INC 1
+#define RD_THRESH_MAX_FACT 64
+#define RD_THRESH_INC 1
+#define RD_THRESH_POW 1.25
+
+#define MV_COST_WEIGHT 108
+#define MV_COST_WEIGHT_SUB 120
static void fill_token_costs(vp9_coeff_cost *c,
vp9_coeff_probs_model (*p)[BLOCK_TYPES]) {
@@ -166,33 +170,9 @@
cpi->mb.sadperbit4 = sad_per_bit4lut[qindex];
}
-void vp9_initialize_rd_consts(VP9_COMP *cpi, int qindex) {
+static void set_block_thresholds(VP9_COMP *cpi, int qindex) {
int q, i, bsize;
-
- vp9_clear_system_state(); // __asm emms;
-
- // Further tests required to see if optimum is different
- // for key frames, golden frames and arf frames.
- // if (cpi->common.refresh_golden_frame ||
- // cpi->common.refresh_alt_ref_frame)
- qindex = clamp(qindex, 0, MAXQ);
-
- cpi->RDDIV = 100;
- cpi->RDMULT = compute_rd_mult(qindex);
- if (cpi->pass == 2 && (cpi->common.frame_type != KEY_FRAME)) {
- if (cpi->twopass.next_iiratio > 31)
- cpi->RDMULT += (cpi->RDMULT * rd_iifactor[31]) >> 4;
- else
- cpi->RDMULT +=
- (cpi->RDMULT * rd_iifactor[cpi->twopass.next_iiratio]) >> 4;
- }
- cpi->mb.errorperbit = cpi->RDMULT >> 6;
- cpi->mb.errorperbit += (cpi->mb.errorperbit == 0);
-
- vp9_set_speed_features(cpi);
-
- q = (int)pow(vp9_dc_quant(qindex, 0) >> 2, 1.25);
- q <<= 2;
+ q = ((int)pow(vp9_dc_quant(qindex, 0) >> 2, RD_THRESH_POW)) << 2;
if (q < 8)
q = 8;
@@ -223,6 +203,34 @@
}
}
}
+}
+
+void vp9_initialize_rd_consts(VP9_COMP *cpi, int qindex) {
+ int i;
+
+ vp9_clear_system_state(); // __asm emms;
+
+ // Further tests required to see if optimum is different
+ // for key frames, golden frames and arf frames.
+ // if (cpi->common.refresh_golden_frame ||
+ // cpi->common.refresh_alt_ref_frame)
+ qindex = clamp(qindex, 0, MAXQ);
+
+ cpi->RDDIV = 100;
+ cpi->RDMULT = compute_rd_mult(qindex);
+ if (cpi->pass == 2 && (cpi->common.frame_type != KEY_FRAME)) {
+ if (cpi->twopass.next_iiratio > 31)
+ cpi->RDMULT += (cpi->RDMULT * rd_iifactor[31]) >> 4;
+ else
+ cpi->RDMULT +=
+ (cpi->RDMULT * rd_iifactor[cpi->twopass.next_iiratio]) >> 4;
+ }
+ cpi->mb.errorperbit = cpi->RDMULT >> 6;
+ cpi->mb.errorperbit += (cpi->mb.errorperbit == 0);
+
+ vp9_set_speed_features(cpi);
+
+ set_block_thresholds(cpi, qindex);
fill_token_costs(cpi->mb.token_costs, cpi->common.fc.coef_probs);
@@ -565,9 +573,9 @@
xd->this_mi->mbmi.ref_frame[0] == INTRA_FRAME) {
// TODO(jingning): tune the model to better capture the distortion.
int64_t p = (pd->dequant[1] * pd->dequant[1] *
- (1 << ss_txfrm_size)) >> shift;
- args->dist[block] = p;
- args->sse[block] = p;
+ (1 << ss_txfrm_size)) >> (shift + 2);
+ args->dist[block] += (p >> 4);
+ args->sse[block] += p;
}
}
@@ -1463,12 +1471,12 @@
case NEWMV:
this_mv->as_int = seg_mvs[mbmi->ref_frame[0]].as_int;
thismvcost = vp9_mv_bit_cost(&this_mv->as_mv, &best_ref_mv->as_mv,
- mvjcost, mvcost, 102);
+ mvjcost, mvcost, MV_COST_WEIGHT_SUB);
if (has_second_rf) {
this_second_mv->as_int = seg_mvs[mbmi->ref_frame[1]].as_int;
thismvcost += vp9_mv_bit_cost(&this_second_mv->as_mv,
&second_best_ref_mv->as_mv,
- mvjcost, mvcost, 102);
+ mvjcost, mvcost, MV_COST_WEIGHT_SUB);
}
break;
case NEARESTMV:
@@ -1499,7 +1507,8 @@
if (has_second_rf)
mic->bmi[i].as_mv[1].as_int = this_second_mv->as_int;
- x->partition_info->bmi[i].mode = m;
+ mic->bmi[i].as_mode = m;
+
for (idy = 0; idy < num_4x4_blocks_high; ++idy)
for (idx = 0; idx < num_4x4_blocks_wide; ++idx)
vpx_memcpy(&mic->bmi[i + idy * 2 + idx],
@@ -1646,7 +1655,7 @@
BEST_SEG_INFO *bsi_buf, int filter_idx,
int_mv seg_mvs[4][MAX_REF_FRAMES],
int mi_row, int mi_col) {
- int i, j, br = 0, idx, idy;
+ int i, br = 0, idx, idy;
int64_t bd = 0, block_sse = 0;
MB_PREDICTION_MODE this_mode;
MODE_INFO *mi = x->e_mbd.mi_8x8[0];
@@ -2008,15 +2017,6 @@
bsi->segment_rd = INT64_MAX;
return;
}
-
- for (j = 1; j < num_4x4_blocks_high; ++j)
- vpx_memcpy(&x->partition_info->bmi[i + j * 2],
- &x->partition_info->bmi[i],
- sizeof(x->partition_info->bmi[i]));
- for (j = 1; j < num_4x4_blocks_wide; ++j)
- vpx_memcpy(&x->partition_info->bmi[i + j],
- &x->partition_info->bmi[i],
- sizeof(x->partition_info->bmi[i]));
}
} /* for each label */
@@ -2028,7 +2028,7 @@
// update the coding decisions
for (i = 0; i < 4; ++i)
- bsi->modes[i] = x->partition_info->bmi[i].mode;
+ bsi->modes[i] = mi->bmi[i].as_mode;
}
static int64_t rd_pick_best_mbsegmentation(VP9_COMP *cpi, MACROBLOCK *x,
@@ -2073,7 +2073,7 @@
if (has_second_ref(mbmi))
mi->bmi[i].as_mv[1].as_int = bsi->rdstat[i][mode_idx].mvs[1].as_int;
xd->plane[0].eobs[i] = bsi->rdstat[i][mode_idx].eobs;
- x->partition_info->bmi[i].mode = bsi->modes[i];
+ mi->bmi[i].as_mode = bsi->modes[i];
}
/*
@@ -2206,7 +2206,6 @@
static void store_coding_context(MACROBLOCK *x, PICK_MODE_CONTEXT *ctx,
int mode_index,
- PARTITION_INFO *partition,
int_mv *ref_mv,
int_mv *second_ref_mv,
int64_t comp_pred_diff[NB_PREDICTION_TYPES],
@@ -2220,9 +2219,6 @@
ctx->best_mode_index = mode_index;
ctx->mic = *xd->this_mi;
- if (partition)
- ctx->partition_info = *partition;
-
ctx->best_ref_mv.as_int = ref_mv->as_int;
ctx->second_best_ref_mv.as_int = second_ref_mv->as_int;
@@ -2454,7 +2450,7 @@
&dis, &sse);
}
*rate_mv = vp9_mv_bit_cost(&tmp_mv->as_mv, &ref_mv.as_mv,
- x->nmvjointcost, x->mvcost, 96);
+ x->nmvjointcost, x->mvcost, MV_COST_WEIGHT);
if (cpi->sf.adaptive_motion_search && cpi->common.show_frame)
x->pred_mv[ref].as_int = tmp_mv->as_int;
@@ -2615,10 +2611,10 @@
}
*rate_mv = vp9_mv_bit_cost(&frame_mv[refs[0]].as_mv,
&mbmi->ref_mvs[refs[0]][0].as_mv,
- x->nmvjointcost, x->mvcost, 96);
+ x->nmvjointcost, x->mvcost, MV_COST_WEIGHT);
*rate_mv += vp9_mv_bit_cost(&frame_mv[refs[1]].as_mv,
&mbmi->ref_mvs[refs[1]][0].as_mv,
- x->nmvjointcost, x->mvcost, 96);
+ x->nmvjointcost, x->mvcost, MV_COST_WEIGHT);
vpx_free(second_pred);
}
@@ -2671,10 +2667,10 @@
} else {
rate_mv = vp9_mv_bit_cost(&frame_mv[refs[0]].as_mv,
&mbmi->ref_mvs[refs[0]][0].as_mv,
- x->nmvjointcost, x->mvcost, 96);
+ x->nmvjointcost, x->mvcost, MV_COST_WEIGHT);
rate_mv += vp9_mv_bit_cost(&frame_mv[refs[1]].as_mv,
&mbmi->ref_mvs[refs[1]][0].as_mv,
- x->nmvjointcost, x->mvcost, 96);
+ x->nmvjointcost, x->mvcost, MV_COST_WEIGHT);
}
if (frame_mv[refs[0]].as_int == INVALID_MV ||
frame_mv[refs[1]].as_int == INVALID_MV)
@@ -3731,9 +3727,9 @@
} else {
cpi->rd_thresh_freq_fact[bsize][mode_index] += RD_THRESH_INC;
if (cpi->rd_thresh_freq_fact[bsize][mode_index] >
- (cpi->sf.adaptive_rd_thresh * MAX_RD_THRESH_FACT)) {
+ (cpi->sf.adaptive_rd_thresh * RD_THRESH_MAX_FACT)) {
cpi->rd_thresh_freq_fact[bsize][mode_index] =
- cpi->sf.adaptive_rd_thresh * MAX_RD_THRESH_FACT;
+ cpi->sf.adaptive_rd_thresh * RD_THRESH_MAX_FACT;
}
}
}
@@ -3780,7 +3776,6 @@
set_scale_factors(xd, mbmi->ref_frame[0], mbmi->ref_frame[1],
scale_factor);
store_coding_context(x, ctx, best_mode_index,
- NULL,
&mbmi->ref_mvs[mbmi->ref_frame[0]][0],
&mbmi->ref_mvs[mbmi->ref_frame[1] < 0 ? 0 :
mbmi->ref_frame[1]][0],
@@ -3839,7 +3834,6 @@
cpi->common.y_dc_delta_q);
int_mv seg_mvs[4][MAX_REF_FRAMES];
b_mode_info best_bmodes[4];
- PARTITION_INFO best_partition;
int best_skip2 = 0;
unsigned char best_zcoeff_blk[256] = { 0 };
@@ -4055,7 +4049,6 @@
&mbmi->ref_mvs[second_ref_frame][0] : NULL;
b_mode_info tmp_best_bmodes[16];
MB_MODE_INFO tmp_best_mbmode;
- PARTITION_INFO tmp_best_partition;
BEST_SEG_INFO bsi[SWITCHABLE_FILTERS];
int pred_exists = 0;
int uv_skippable;
@@ -4119,7 +4112,6 @@
tmp_best_sse = total_sse;
tmp_best_skippable = skippable;
tmp_best_mbmode = *mbmi;
- tmp_best_partition = *x->partition_info;
for (i = 0; i < 4; i++)
tmp_best_bmodes[i] = xd->this_mi->bmi[i];
pred_exists = 1;
@@ -4171,7 +4163,6 @@
distortion = tmp_best_distortion;
skippable = tmp_best_skippable;
*mbmi = tmp_best_mbmode;
- *x->partition_info = tmp_best_partition;
for (i = 0; i < 4; i++)
xd->this_mi->bmi[i] = tmp_best_bmodes[i];
}
@@ -4299,7 +4290,6 @@
RDCOST(x->rdmult, x->rddiv, rate_uv, distortion_uv);
best_mbmode = *mbmi;
best_skip2 = this_skip2;
- best_partition = *x->partition_info;
vpx_memcpy(best_zcoeff_blk, x->zcoeff_blk[mbmi->tx_size],
sizeof(best_zcoeff_blk));
@@ -4442,9 +4432,9 @@
} else {
cpi->rd_thresh_freq_sub8x8[bsize][mode_index] += RD_THRESH_INC;
if (cpi->rd_thresh_freq_sub8x8[bsize][mode_index] >
- (cpi->sf.adaptive_rd_thresh * MAX_RD_THRESH_FACT)) {
+ (cpi->sf.adaptive_rd_thresh * RD_THRESH_MAX_FACT)) {
cpi->rd_thresh_freq_sub8x8[bsize][mode_index] =
- cpi->sf.adaptive_rd_thresh * MAX_RD_THRESH_FACT;
+ cpi->sf.adaptive_rd_thresh * RD_THRESH_MAX_FACT;
}
}
}
@@ -4457,15 +4447,8 @@
for (i = 0; i < 4; i++)
xd->this_mi->bmi[i].as_mode = best_bmodes[i].as_mode;
} else {
- for (i = 0; i < 4; i++)
- xd->this_mi->bmi[i].as_mv[0].as_int =
- best_bmodes[i].as_mv[0].as_int;
-
- if (has_second_ref(mbmi))
- for (i = 0; i < 4; i++)
- xd->this_mi->bmi[i].as_mv[1].as_int = best_bmodes[i].as_mv[1].as_int;
-
- *x->partition_info = best_partition;
+ for (i = 0; i < 4; ++i)
+ vpx_memcpy(&xd->this_mi->bmi[i], &best_bmodes[i], sizeof(b_mode_info));
mbmi->mv[0].as_int = xd->this_mi->bmi[3].as_mv[0].as_int;
mbmi->mv[1].as_int = xd->this_mi->bmi[3].as_mv[1].as_int;
@@ -4508,7 +4491,6 @@
set_scale_factors(xd, mbmi->ref_frame[0], mbmi->ref_frame[1],
scale_factor);
store_coding_context(x, ctx, best_mode_index,
- &best_partition,
&mbmi->ref_mvs[mbmi->ref_frame[0]][0],
&mbmi->ref_mvs[mbmi->ref_frame[1] < 0 ? 0 :
mbmi->ref_frame[1]][0],
diff --git a/vp9/encoder/x86/vp9_dct32x32_sse2.c b/vp9/encoder/x86/vp9_dct32x32_sse2.c
index 2d5c0b4..11eec7f 100644
--- a/vp9/encoder/x86/vp9_dct32x32_sse2.c
+++ b/vp9/encoder/x86/vp9_dct32x32_sse2.c
@@ -27,24 +27,6 @@
__m128i buf1 = _mm_shuffle_epi32(b, _MM_SHUFFLE(0, 0, 2, 0));
return _mm_unpacklo_epi64(buf0, buf1);
}
-
-static INLINE __m128i k_cvtlo_epi16(__m128i a, __m128i mask16, __m128i kZero) {
- // convert the lower 4 signed 16-bit integers into 4 signed 32-bit integers
- __m128i sign_bit = _mm_and_si128(a, mask16);
- __m128i b = _mm_unpacklo_epi16(a, kZero);
- sign_bit = _mm_cmplt_epi16(sign_bit, kZero);
- sign_bit = _mm_unpacklo_epi16(kZero, sign_bit);
- return _mm_or_si128(sign_bit, b);
-}
-
-static INLINE __m128i k_cvthi_epi16(__m128i a, __m128i mask16, __m128i kZero) {
- // convert the lower 4 signed 16-bit integers into 4 signed 32-bit integers
- __m128i sign_bit = _mm_and_si128(a, mask16);
- __m128i b = _mm_unpackhi_epi16(a, kZero);
- sign_bit = _mm_cmplt_epi16(sign_bit, kZero);
- sign_bit = _mm_unpackhi_epi16(kZero, sign_bit);
- return _mm_or_si128(sign_bit, b);
-}
#endif
void FDCT32x32_2D(int16_t *input,
@@ -1159,28 +1141,43 @@
} else {
__m128i lstep1[64], lstep2[64], lstep3[64];
__m128i u[32], v[32], sign[16];
- const __m128i mask16 = _mm_set1_epi32(0x80008000);
const __m128i K32One = _mm_set_epi32(1, 1, 1, 1);
// start using 32-bit operations
// stage 3
{
// expanding to 32-bit length priori to addition operations
- lstep2[ 0] = k_cvtlo_epi16(step2[ 0], mask16, kZero);
- lstep2[ 1] = k_cvthi_epi16(step2[ 0], mask16, kZero);
- lstep2[ 2] = k_cvtlo_epi16(step2[ 1], mask16, kZero);
- lstep2[ 3] = k_cvthi_epi16(step2[ 1], mask16, kZero);
- lstep2[ 4] = k_cvtlo_epi16(step2[ 2], mask16, kZero);
- lstep2[ 5] = k_cvthi_epi16(step2[ 2], mask16, kZero);
- lstep2[ 6] = k_cvtlo_epi16(step2[ 3], mask16, kZero);
- lstep2[ 7] = k_cvthi_epi16(step2[ 3], mask16, kZero);
- lstep2[ 8] = k_cvtlo_epi16(step2[ 4], mask16, kZero);
- lstep2[ 9] = k_cvthi_epi16(step2[ 4], mask16, kZero);
- lstep2[10] = k_cvtlo_epi16(step2[ 5], mask16, kZero);
- lstep2[11] = k_cvthi_epi16(step2[ 5], mask16, kZero);
- lstep2[12] = k_cvtlo_epi16(step2[ 6], mask16, kZero);
- lstep2[13] = k_cvthi_epi16(step2[ 6], mask16, kZero);
- lstep2[14] = k_cvtlo_epi16(step2[ 7], mask16, kZero);
- lstep2[15] = k_cvthi_epi16(step2[ 7], mask16, kZero);
+ lstep2[ 0] = _mm_unpacklo_epi16(step2[ 0], kZero);
+ lstep2[ 1] = _mm_unpackhi_epi16(step2[ 0], kZero);
+ lstep2[ 2] = _mm_unpacklo_epi16(step2[ 1], kZero);
+ lstep2[ 3] = _mm_unpackhi_epi16(step2[ 1], kZero);
+ lstep2[ 4] = _mm_unpacklo_epi16(step2[ 2], kZero);
+ lstep2[ 5] = _mm_unpackhi_epi16(step2[ 2], kZero);
+ lstep2[ 6] = _mm_unpacklo_epi16(step2[ 3], kZero);
+ lstep2[ 7] = _mm_unpackhi_epi16(step2[ 3], kZero);
+ lstep2[ 8] = _mm_unpacklo_epi16(step2[ 4], kZero);
+ lstep2[ 9] = _mm_unpackhi_epi16(step2[ 4], kZero);
+ lstep2[10] = _mm_unpacklo_epi16(step2[ 5], kZero);
+ lstep2[11] = _mm_unpackhi_epi16(step2[ 5], kZero);
+ lstep2[12] = _mm_unpacklo_epi16(step2[ 6], kZero);
+ lstep2[13] = _mm_unpackhi_epi16(step2[ 6], kZero);
+ lstep2[14] = _mm_unpacklo_epi16(step2[ 7], kZero);
+ lstep2[15] = _mm_unpackhi_epi16(step2[ 7], kZero);
+ lstep2[ 0] = _mm_madd_epi16(lstep2[ 0], kOne);
+ lstep2[ 1] = _mm_madd_epi16(lstep2[ 1], kOne);
+ lstep2[ 2] = _mm_madd_epi16(lstep2[ 2], kOne);
+ lstep2[ 3] = _mm_madd_epi16(lstep2[ 3], kOne);
+ lstep2[ 4] = _mm_madd_epi16(lstep2[ 4], kOne);
+ lstep2[ 5] = _mm_madd_epi16(lstep2[ 5], kOne);
+ lstep2[ 6] = _mm_madd_epi16(lstep2[ 6], kOne);
+ lstep2[ 7] = _mm_madd_epi16(lstep2[ 7], kOne);
+ lstep2[ 8] = _mm_madd_epi16(lstep2[ 8], kOne);
+ lstep2[ 9] = _mm_madd_epi16(lstep2[ 9], kOne);
+ lstep2[10] = _mm_madd_epi16(lstep2[10], kOne);
+ lstep2[11] = _mm_madd_epi16(lstep2[11], kOne);
+ lstep2[12] = _mm_madd_epi16(lstep2[12], kOne);
+ lstep2[13] = _mm_madd_epi16(lstep2[13], kOne);
+ lstep2[14] = _mm_madd_epi16(lstep2[14], kOne);
+ lstep2[15] = _mm_madd_epi16(lstep2[15], kOne);
lstep3[ 0] = _mm_add_epi32(lstep2[14], lstep2[ 0]);
lstep3[ 1] = _mm_add_epi32(lstep2[15], lstep2[ 1]);
@@ -1231,42 +1228,75 @@
lstep3[27] = _mm_srai_epi32(s3_13_5, DCT_CONST_BITS);
}
{
- lstep2[40] = k_cvtlo_epi16(step2[20], mask16, kZero);
- lstep2[41] = k_cvthi_epi16(step2[20], mask16, kZero);
- lstep2[42] = k_cvtlo_epi16(step2[21], mask16, kZero);
- lstep2[43] = k_cvthi_epi16(step2[21], mask16, kZero);
- lstep2[44] = k_cvtlo_epi16(step2[22], mask16, kZero);
- lstep2[45] = k_cvthi_epi16(step2[22], mask16, kZero);
- lstep2[46] = k_cvtlo_epi16(step2[23], mask16, kZero);
- lstep2[47] = k_cvthi_epi16(step2[23], mask16, kZero);
- lstep2[48] = k_cvtlo_epi16(step2[24], mask16, kZero);
- lstep2[49] = k_cvthi_epi16(step2[24], mask16, kZero);
- lstep2[50] = k_cvtlo_epi16(step2[25], mask16, kZero);
- lstep2[51] = k_cvthi_epi16(step2[25], mask16, kZero);
- lstep2[52] = k_cvtlo_epi16(step2[26], mask16, kZero);
- lstep2[53] = k_cvthi_epi16(step2[26], mask16, kZero);
- lstep2[54] = k_cvtlo_epi16(step2[27], mask16, kZero);
- lstep2[55] = k_cvthi_epi16(step2[27], mask16, kZero);
+ lstep2[40] = _mm_unpacklo_epi16(step2[20], kZero);
+ lstep2[41] = _mm_unpackhi_epi16(step2[20], kZero);
+ lstep2[42] = _mm_unpacklo_epi16(step2[21], kZero);
+ lstep2[43] = _mm_unpackhi_epi16(step2[21], kZero);
+ lstep2[44] = _mm_unpacklo_epi16(step2[22], kZero);
+ lstep2[45] = _mm_unpackhi_epi16(step2[22], kZero);
+ lstep2[46] = _mm_unpacklo_epi16(step2[23], kZero);
+ lstep2[47] = _mm_unpackhi_epi16(step2[23], kZero);
+ lstep2[48] = _mm_unpacklo_epi16(step2[24], kZero);
+ lstep2[49] = _mm_unpackhi_epi16(step2[24], kZero);
+ lstep2[50] = _mm_unpacklo_epi16(step2[25], kZero);
+ lstep2[51] = _mm_unpackhi_epi16(step2[25], kZero);
+ lstep2[52] = _mm_unpacklo_epi16(step2[26], kZero);
+ lstep2[53] = _mm_unpackhi_epi16(step2[26], kZero);
+ lstep2[54] = _mm_unpacklo_epi16(step2[27], kZero);
+ lstep2[55] = _mm_unpackhi_epi16(step2[27], kZero);
+ lstep2[40] = _mm_madd_epi16(lstep2[40], kOne);
+ lstep2[41] = _mm_madd_epi16(lstep2[41], kOne);
+ lstep2[42] = _mm_madd_epi16(lstep2[42], kOne);
+ lstep2[43] = _mm_madd_epi16(lstep2[43], kOne);
+ lstep2[44] = _mm_madd_epi16(lstep2[44], kOne);
+ lstep2[45] = _mm_madd_epi16(lstep2[45], kOne);
+ lstep2[46] = _mm_madd_epi16(lstep2[46], kOne);
+ lstep2[47] = _mm_madd_epi16(lstep2[47], kOne);
+ lstep2[48] = _mm_madd_epi16(lstep2[48], kOne);
+ lstep2[49] = _mm_madd_epi16(lstep2[49], kOne);
+ lstep2[50] = _mm_madd_epi16(lstep2[50], kOne);
+ lstep2[51] = _mm_madd_epi16(lstep2[51], kOne);
+ lstep2[52] = _mm_madd_epi16(lstep2[52], kOne);
+ lstep2[53] = _mm_madd_epi16(lstep2[53], kOne);
+ lstep2[54] = _mm_madd_epi16(lstep2[54], kOne);
+ lstep2[55] = _mm_madd_epi16(lstep2[55], kOne);
- lstep1[32] = k_cvtlo_epi16(step1[16], mask16, kZero);
- lstep1[33] = k_cvthi_epi16(step1[16], mask16, kZero);
- lstep1[34] = k_cvtlo_epi16(step1[17], mask16, kZero);
- lstep1[35] = k_cvthi_epi16(step1[17], mask16, kZero);
- lstep1[36] = k_cvtlo_epi16(step1[18], mask16, kZero);
- lstep1[37] = k_cvthi_epi16(step1[18], mask16, kZero);
- lstep1[38] = k_cvtlo_epi16(step1[19], mask16, kZero);
- lstep1[39] = k_cvthi_epi16(step1[19], mask16, kZero);
- lstep1[56] = k_cvtlo_epi16(step1[28], mask16, kZero);
- lstep1[57] = k_cvthi_epi16(step1[28], mask16, kZero);
- lstep1[58] = k_cvtlo_epi16(step1[29], mask16, kZero);
- lstep1[59] = k_cvthi_epi16(step1[29], mask16, kZero);
- lstep1[60] = k_cvtlo_epi16(step1[30], mask16, kZero);
- lstep1[61] = k_cvthi_epi16(step1[30], mask16, kZero);
- lstep1[62] = k_cvtlo_epi16(step1[31], mask16, kZero);
- lstep1[63] = k_cvthi_epi16(step1[31], mask16, kZero);
+ lstep1[32] = _mm_unpacklo_epi16(step1[16], kZero);
+ lstep1[33] = _mm_unpackhi_epi16(step1[16], kZero);
+ lstep1[34] = _mm_unpacklo_epi16(step1[17], kZero);
+ lstep1[35] = _mm_unpackhi_epi16(step1[17], kZero);
+ lstep1[36] = _mm_unpacklo_epi16(step1[18], kZero);
+ lstep1[37] = _mm_unpackhi_epi16(step1[18], kZero);
+ lstep1[38] = _mm_unpacklo_epi16(step1[19], kZero);
+ lstep1[39] = _mm_unpackhi_epi16(step1[19], kZero);
+ lstep1[56] = _mm_unpacklo_epi16(step1[28], kZero);
+ lstep1[57] = _mm_unpackhi_epi16(step1[28], kZero);
+ lstep1[58] = _mm_unpacklo_epi16(step1[29], kZero);
+ lstep1[59] = _mm_unpackhi_epi16(step1[29], kZero);
+ lstep1[60] = _mm_unpacklo_epi16(step1[30], kZero);
+ lstep1[61] = _mm_unpackhi_epi16(step1[30], kZero);
+ lstep1[62] = _mm_unpacklo_epi16(step1[31], kZero);
+ lstep1[63] = _mm_unpackhi_epi16(step1[31], kZero);
+ lstep1[32] = _mm_madd_epi16(lstep1[32], kOne);
+ lstep1[33] = _mm_madd_epi16(lstep1[33], kOne);
+ lstep1[34] = _mm_madd_epi16(lstep1[34], kOne);
+ lstep1[35] = _mm_madd_epi16(lstep1[35], kOne);
+ lstep1[36] = _mm_madd_epi16(lstep1[36], kOne);
+ lstep1[37] = _mm_madd_epi16(lstep1[37], kOne);
+ lstep1[38] = _mm_madd_epi16(lstep1[38], kOne);
+ lstep1[39] = _mm_madd_epi16(lstep1[39], kOne);
+ lstep1[56] = _mm_madd_epi16(lstep1[56], kOne);
+ lstep1[57] = _mm_madd_epi16(lstep1[57], kOne);
+ lstep1[58] = _mm_madd_epi16(lstep1[58], kOne);
+ lstep1[59] = _mm_madd_epi16(lstep1[59], kOne);
+ lstep1[60] = _mm_madd_epi16(lstep1[60], kOne);
+ lstep1[61] = _mm_madd_epi16(lstep1[61], kOne);
+ lstep1[62] = _mm_madd_epi16(lstep1[62], kOne);
+ lstep1[63] = _mm_madd_epi16(lstep1[63], kOne);
lstep3[32] = _mm_add_epi32(lstep2[46], lstep1[32]);
lstep3[33] = _mm_add_epi32(lstep2[47], lstep1[33]);
+
lstep3[34] = _mm_add_epi32(lstep2[44], lstep1[34]);
lstep3[35] = _mm_add_epi32(lstep2[45], lstep1[35]);
lstep3[36] = _mm_add_epi32(lstep2[42], lstep1[36]);
@@ -1302,14 +1332,22 @@
// stage 4
{
// expanding to 32-bit length priori to addition operations
- lstep2[16] = k_cvtlo_epi16(step2[ 8], mask16, kZero);
- lstep2[17] = k_cvthi_epi16(step2[ 8], mask16, kZero);
- lstep2[18] = k_cvtlo_epi16(step2[ 9], mask16, kZero);
- lstep2[19] = k_cvthi_epi16(step2[ 9], mask16, kZero);
- lstep2[28] = k_cvtlo_epi16(step2[14], mask16, kZero);
- lstep2[29] = k_cvthi_epi16(step2[14], mask16, kZero);
- lstep2[30] = k_cvtlo_epi16(step2[15], mask16, kZero);
- lstep2[31] = k_cvthi_epi16(step2[15], mask16, kZero);
+ lstep2[16] = _mm_unpacklo_epi16(step2[ 8], kZero);
+ lstep2[17] = _mm_unpackhi_epi16(step2[ 8], kZero);
+ lstep2[18] = _mm_unpacklo_epi16(step2[ 9], kZero);
+ lstep2[19] = _mm_unpackhi_epi16(step2[ 9], kZero);
+ lstep2[28] = _mm_unpacklo_epi16(step2[14], kZero);
+ lstep2[29] = _mm_unpackhi_epi16(step2[14], kZero);
+ lstep2[30] = _mm_unpacklo_epi16(step2[15], kZero);
+ lstep2[31] = _mm_unpackhi_epi16(step2[15], kZero);
+ lstep2[16] = _mm_madd_epi16(lstep2[16], kOne);
+ lstep2[17] = _mm_madd_epi16(lstep2[17], kOne);
+ lstep2[18] = _mm_madd_epi16(lstep2[18], kOne);
+ lstep2[19] = _mm_madd_epi16(lstep2[19], kOne);
+ lstep2[28] = _mm_madd_epi16(lstep2[28], kOne);
+ lstep2[29] = _mm_madd_epi16(lstep2[29], kOne);
+ lstep2[30] = _mm_madd_epi16(lstep2[30], kOne);
+ lstep2[31] = _mm_madd_epi16(lstep2[31], kOne);
lstep1[ 0] = _mm_add_epi32(lstep3[ 6], lstep3[ 0]);
lstep1[ 1] = _mm_add_epi32(lstep3[ 7], lstep3[ 1]);
@@ -1337,41 +1375,41 @@
lstep1[31] = _mm_add_epi32(lstep3[25], lstep2[31]);
}
{
- // to be continued...
- //
- const __m128i k32_p16_p16 = pair_set_epi32(cospi_16_64, cospi_16_64);
- const __m128i k32_p16_m16 = pair_set_epi32(cospi_16_64, -cospi_16_64);
+ // to be continued...
+ //
+ const __m128i k32_p16_p16 = pair_set_epi32(cospi_16_64, cospi_16_64);
+ const __m128i k32_p16_m16 = pair_set_epi32(cospi_16_64, -cospi_16_64);
- u[0] = _mm_unpacklo_epi32(lstep3[12], lstep3[10]);
- u[1] = _mm_unpackhi_epi32(lstep3[12], lstep3[10]);
- u[2] = _mm_unpacklo_epi32(lstep3[13], lstep3[11]);
- u[3] = _mm_unpackhi_epi32(lstep3[13], lstep3[11]);
+ u[0] = _mm_unpacklo_epi32(lstep3[12], lstep3[10]);
+ u[1] = _mm_unpackhi_epi32(lstep3[12], lstep3[10]);
+ u[2] = _mm_unpacklo_epi32(lstep3[13], lstep3[11]);
+ u[3] = _mm_unpackhi_epi32(lstep3[13], lstep3[11]);
- // TODO(jingning): manually inline k_madd_epi32_ to further hide
- // instruction latency.
- v[ 0] = k_madd_epi32(u[0], k32_p16_m16);
- v[ 1] = k_madd_epi32(u[1], k32_p16_m16);
- v[ 2] = k_madd_epi32(u[2], k32_p16_m16);
- v[ 3] = k_madd_epi32(u[3], k32_p16_m16);
- v[ 4] = k_madd_epi32(u[0], k32_p16_p16);
- v[ 5] = k_madd_epi32(u[1], k32_p16_p16);
- v[ 6] = k_madd_epi32(u[2], k32_p16_p16);
- v[ 7] = k_madd_epi32(u[3], k32_p16_p16);
+ // TODO(jingning): manually inline k_madd_epi32_ to further hide
+ // instruction latency.
+ v[ 0] = k_madd_epi32(u[0], k32_p16_m16);
+ v[ 1] = k_madd_epi32(u[1], k32_p16_m16);
+ v[ 2] = k_madd_epi32(u[2], k32_p16_m16);
+ v[ 3] = k_madd_epi32(u[3], k32_p16_m16);
+ v[ 4] = k_madd_epi32(u[0], k32_p16_p16);
+ v[ 5] = k_madd_epi32(u[1], k32_p16_p16);
+ v[ 6] = k_madd_epi32(u[2], k32_p16_p16);
+ v[ 7] = k_madd_epi32(u[3], k32_p16_p16);
- u[0] = k_packs_epi64(v[0], v[1]);
- u[1] = k_packs_epi64(v[2], v[3]);
- u[2] = k_packs_epi64(v[4], v[5]);
- u[3] = k_packs_epi64(v[6], v[7]);
+ u[0] = k_packs_epi64(v[0], v[1]);
+ u[1] = k_packs_epi64(v[2], v[3]);
+ u[2] = k_packs_epi64(v[4], v[5]);
+ u[3] = k_packs_epi64(v[6], v[7]);
- v[0] = _mm_add_epi32(u[0], k__DCT_CONST_ROUNDING);
- v[1] = _mm_add_epi32(u[1], k__DCT_CONST_ROUNDING);
- v[2] = _mm_add_epi32(u[2], k__DCT_CONST_ROUNDING);
- v[3] = _mm_add_epi32(u[3], k__DCT_CONST_ROUNDING);
+ v[0] = _mm_add_epi32(u[0], k__DCT_CONST_ROUNDING);
+ v[1] = _mm_add_epi32(u[1], k__DCT_CONST_ROUNDING);
+ v[2] = _mm_add_epi32(u[2], k__DCT_CONST_ROUNDING);
+ v[3] = _mm_add_epi32(u[3], k__DCT_CONST_ROUNDING);
- lstep1[10] = _mm_srai_epi32(v[0], DCT_CONST_BITS);
- lstep1[11] = _mm_srai_epi32(v[1], DCT_CONST_BITS);
- lstep1[12] = _mm_srai_epi32(v[2], DCT_CONST_BITS);
- lstep1[13] = _mm_srai_epi32(v[3], DCT_CONST_BITS);
+ lstep1[10] = _mm_srai_epi32(v[0], DCT_CONST_BITS);
+ lstep1[11] = _mm_srai_epi32(v[1], DCT_CONST_BITS);
+ lstep1[12] = _mm_srai_epi32(v[2], DCT_CONST_BITS);
+ lstep1[13] = _mm_srai_epi32(v[3], DCT_CONST_BITS);
}
{
const __m128i k32_m08_p24 = pair_set_epi32(-cospi_8_64, cospi_24_64);
diff --git a/vp9/vp9_common.mk b/vp9/vp9_common.mk
index 10fa461..94b37b3 100644
--- a/vp9/vp9_common.mk
+++ b/vp9/vp9_common.mk
@@ -68,6 +68,8 @@
VP9_COMMON_SRCS-yes += common/vp9_treecoder.c
VP9_COMMON_SRCS-yes += common/vp9_common_data.c
VP9_COMMON_SRCS-yes += common/vp9_common_data.h
+VP9_COMMON_SRCS-yes += common/vp9_scan.c
+VP9_COMMON_SRCS-yes += common/vp9_scan.h
VP9_COMMON_SRCS-$(ARCH_X86)$(ARCH_X86_64) += common/x86/vp9_postproc_x86.h
VP9_COMMON_SRCS-$(ARCH_X86)$(ARCH_X86_64) += common/x86/vp9_asm_stubs.c
@@ -89,6 +91,11 @@
# common (c)
VP9_COMMON_SRCS-$(HAVE_DSPR2) += common/mips/dspr2/vp9_common_dspr2.h
+VP9_COMMON_SRCS-$(HAVE_DSPR2) += common/mips/dspr2/vp9_convolve2_avg_dspr2.c
+VP9_COMMON_SRCS-$(HAVE_DSPR2) += common/mips/dspr2/vp9_convolve2_avg_horiz_dspr2.c
+VP9_COMMON_SRCS-$(HAVE_DSPR2) += common/mips/dspr2/vp9_convolve2_dspr2.c
+VP9_COMMON_SRCS-$(HAVE_DSPR2) += common/mips/dspr2/vp9_convolve2_horiz_dspr2.c
+VP9_COMMON_SRCS-$(HAVE_DSPR2) += common/mips/dspr2/vp9_convolve2_vert_dspr2.c
VP9_COMMON_SRCS-$(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
