diff options
| author | Parag Salasakar <img.mips1@gmail.com> | 2013-10-09 18:05:27 +0530 |
|---|---|---|
| committer | Parag Salasakar <img.mips1@gmail.com> | 2013-10-09 18:05:27 +0530 |
| commit | eeb5b62dc19aef73c0fb37655253a46b0c4dc602 (patch) | |
| tree | 00d1140ff0258e56eeaa81124a40842a34178190 /vp9/common/mips/dspr2 | |
| parent | b4148c3a03a1f91d3b6d931d22d9cd5c44879bfe (diff) | |
| download | libvpx-eeb5b62dc19aef73c0fb37655253a46b0c4dc602.tar libvpx-eeb5b62dc19aef73c0fb37655253a46b0c4dc602.tar.gz libvpx-eeb5b62dc19aef73c0fb37655253a46b0c4dc602.tar.bz2 libvpx-eeb5b62dc19aef73c0fb37655253a46b0c4dc602.zip | |
mips dsp-ase r2 vp9 decoder bilinear convolve optimizations
Change-Id: Ic31b4ef85e65070b4f8b9f26e068ccfaae00c4f0
Diffstat (limited to 'vp9/common/mips/dspr2')
| -rw-r--r-- | vp9/common/mips/dspr2/vp9_common_dspr2.h | 29 | ||||
| -rw-r--r-- | vp9/common/mips/dspr2/vp9_convolve2_avg_dspr2.c | 281 | ||||
| -rw-r--r-- | vp9/common/mips/dspr2/vp9_convolve2_avg_horiz_dspr2.c | 833 | ||||
| -rw-r--r-- | vp9/common/mips/dspr2/vp9_convolve2_dspr2.c | 784 | ||||
| -rw-r--r-- | vp9/common/mips/dspr2/vp9_convolve2_horiz_dspr2.c | 713 | ||||
| -rw-r--r-- | vp9/common/mips/dspr2/vp9_convolve2_vert_dspr2.c | 266 | ||||
| -rw-r--r-- | vp9/common/mips/dspr2/vp9_convolve8_avg_dspr2.c | 6 | ||||
| -rw-r--r-- | vp9/common/mips/dspr2/vp9_convolve8_avg_horiz_dspr2.c | 6 | ||||
| -rw-r--r-- | vp9/common/mips/dspr2/vp9_convolve8_dspr2.c | 59 | ||||
| -rw-r--r-- | vp9/common/mips/dspr2/vp9_convolve8_horiz_dspr2.c | 6 | ||||
| -rw-r--r-- | vp9/common/mips/dspr2/vp9_convolve8_vert_dspr2.c | 6 |
11 files changed, 2961 insertions, 28 deletions
diff --git a/vp9/common/mips/dspr2/vp9_common_dspr2.h b/vp9/common/mips/dspr2/vp9_common_dspr2.h index d2fa4c1dc..dc88f1603 100644 --- a/vp9/common/mips/dspr2/vp9_common_dspr2.h +++ b/vp9/common/mips/dspr2/vp9_common_dspr2.h @@ -81,5 +81,34 @@ static INLINE void vp9_prefetch_store_streamed(unsigned char *dst) { ); } +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 000000000..91d62bc49 --- /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 000000000..148b20fba --- /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 000000000..bc422bcf1 --- /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 000000000..1debdb4c0 --- /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 000000000..8eb105c0c --- /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 0930ad123..da7f0fdbb 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 @@ void vp9_convolve8_avg_vert_dspr2(const uint8_t *src, ptrdiff_t src_stride, 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 37c665be9..69da1cfd6 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 @@ void vp9_convolve8_avg_horiz_dspr2(const uint8_t *src, ptrdiff_t src_stride, 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 2c48bd038..126e05a67 100644 --- a/vp9/common/mips/dspr2/vp9_convolve8_dspr2.c +++ b/vp9/common/mips/dspr2/vp9_convolve8_dspr2.c @@ -930,6 +930,21 @@ void convolve_horiz_transposed(const uint8_t *src, ptrdiff_t src_stride, } } +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 @@ void vp9_convolve8_dspr2(const uint8_t *src, ptrdiff_t src_stride, /* 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 @@ void vp9_convolve8_dspr2(const uint8_t *src, ptrdiff_t src_stride, /* 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 743d64116..030389615 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 @@ void vp9_convolve8_horiz_dspr2(const uint8_t *src, ptrdiff_t src_stride, 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 bdc7930b7..0930bb3d8 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 @@ void vp9_convolve8_vert_dspr2(const uint8_t *src, ptrdiff_t src_stride, 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; |