ref: e755a283ddfab1bb7577ba13e75aab415a2cc9aa
parent: d9653b4be05d377f80fadc19295149a03d22857f
parent: fce3cee8ddd8dda91553e4701c0a8081ff4bab52
author: James Bankoski <[email protected]>
date: Tue May 3 10:11:57 EDT 2016
Merge "Move vpx_add_plane from codec to vpx_dsp and dedup."
--- a/vp10/common/postproc.c
+++ b/vp10/common/postproc.c
@@ -13,6 +13,7 @@
#include <stdio.h>
#include "./vpx_config.h"
+#include "./vpx_dsp_rtcd.h"
#include "./vpx_scale_rtcd.h"
#include "./vp10_rtcd.h"
@@ -587,32 +588,6 @@
state->last_noise = a;
}
-void vp10_plane_add_noise_c(uint8_t *start, char *noise,
- char blackclamp[16],
- char whiteclamp[16],
- char bothclamp[16],
- unsigned int width, unsigned int height, int pitch) {- unsigned int i, j;
-
- // TODO(jbb): why does simd code use both but c doesn't, normalize and
- // fix..
- (void) bothclamp;
- for (i = 0; i < height; i++) {- uint8_t *pos = start + i * pitch;
- char *ref = (char *)(noise + (rand() & 0xff)); // NOLINT
-
- for (j = 0; j < width; j++) {- if (pos[j] < blackclamp[0])
- pos[j] = blackclamp[0];
-
- if (pos[j] > 255 + whiteclamp[0])
- pos[j] = 255 + whiteclamp[0];
-
- pos[j] += ref[j];
- }
- }
-}
-
static void swap_mi_and_prev_mi(VP10_COMMON *cm) {// Current mip will be the prev_mip for the next frame.
MODE_INFO *temp = cm->postproc_state.prev_mip;
@@ -727,7 +702,7 @@
fillrd(ppstate, 63 - q, noise_level);
}
- vp10_plane_add_noise(ppbuf->y_buffer, ppstate->noise, ppstate->blackclamp,
+ vpx_plane_add_noise(ppbuf->y_buffer, ppstate->noise, ppstate->blackclamp,
ppstate->whiteclamp, ppstate->bothclamp,
ppbuf->y_width, ppbuf->y_height, ppbuf->y_stride);
}
--- a/vp10/common/vp10_rtcd_defs.pl
+++ b/vp10/common/vp10_rtcd_defs.pl
@@ -70,10 +70,6 @@
specialize qw/vp10_post_proc_down_and_across sse2/;
$vp10_post_proc_down_and_across_sse2=vp10_post_proc_down_and_across_xmm;
-add_proto qw/void vp10_plane_add_noise/, "uint8_t *Start, char *noise, char blackclamp[16], char whiteclamp[16], char bothclamp[16], unsigned int Width, unsigned int Height, int Pitch";
-specialize qw/vp10_plane_add_noise sse2/;
-$vp10_plane_add_noise_sse2=vp10_plane_add_noise_wmt;
-
add_proto qw/void vp10_filter_by_weight16x16/, "const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int src_weight";
specialize qw/vp10_filter_by_weight16x16 sse2 msa/;
@@ -326,9 +322,6 @@
add_proto qw/void vp10_highbd_post_proc_down_and_across/, "const uint16_t *src_ptr, uint16_t *dst_ptr, int src_pixels_per_line, int dst_pixels_per_line, int rows, int cols, int flimit";
specialize qw/vp10_highbd_post_proc_down_and_across/;
-
- add_proto qw/void vp10_highbd_plane_add_noise/, "uint8_t *Start, char *noise, char blackclamp[16], char whiteclamp[16], char bothclamp[16], unsigned int Width, unsigned int Height, int Pitch";
- specialize qw/vp10_highbd_plane_add_noise/;
}
#
--- a/vp10/common/x86/postproc_sse2.asm
+++ b/vp10/common/x86/postproc_sse2.asm
@@ -624,68 +624,6 @@
%undef flimit4
-;void vp10_plane_add_noise_wmt (unsigned char *start, unsigned char *noise,
-; unsigned char blackclamp[16],
-; unsigned char whiteclamp[16],
-; unsigned char bothclamp[16],
-; unsigned int width, unsigned int height, int pitch)
-global sym(vp10_plane_add_noise_wmt) PRIVATE
-sym(vp10_plane_add_noise_wmt):
- push rbp
- mov rbp, rsp
- SHADOW_ARGS_TO_STACK 8
- GET_GOT rbx
- push rsi
- push rdi
- ; end prolog
-
-.addnoise_loop:
- call sym(LIBVPX_RAND) WRT_PLT
- mov rcx, arg(1) ;noise
- and rax, 0xff
- add rcx, rax
-
- ; we rely on the fact that the clamping vectors are stored contiguously
- ; in black/white/both order. Note that we have to reload this here because
- ; rdx could be trashed by rand()
- mov rdx, arg(2) ; blackclamp
-
-
- mov rdi, rcx
- movsxd rcx, dword arg(5) ;[Width]
- mov rsi, arg(0) ;Pos
- xor rax,rax
-
-.addnoise_nextset:
- movdqu xmm1,[rsi+rax] ; get the source
-
- psubusb xmm1, [rdx] ;blackclamp ; clamp both sides so we don't outrange adding noise
- paddusb xmm1, [rdx+32] ;bothclamp
- psubusb xmm1, [rdx+16] ;whiteclamp
-
- movdqu xmm2,[rdi+rax] ; get the noise for this line
- paddb xmm1,xmm2 ; add it in
- movdqu [rsi+rax],xmm1 ; store the result
-
- add rax,16 ; move to the next line
-
- cmp rax, rcx
- jl .addnoise_nextset
-
- movsxd rax, dword arg(7) ; Pitch
- add arg(0), rax ; Start += Pitch
- sub dword arg(6), 1 ; Height -= 1
- jg .addnoise_loop
-
- ; begin epilog
- pop rdi
- pop rsi
- RESTORE_GOT
- UNSHADOW_ARGS
- pop rbp
- ret
-
-
SECTION_RODATA
align 16
rd42:
--- a/vp8/common/mips/msa/postproc_msa.c
+++ b/vp8/common/mips/msa/postproc_msa.c
@@ -10,6 +10,7 @@
#include <stdlib.h>
#include "./vp8_rtcd.h"
+#include "./vpx_dsp_rtcd.h"
#include "vp8/common/mips/msa/vp8_macros_msa.h"
static const int16_t vp8_rv_msa[] =
@@ -795,57 +796,6 @@
}
dst_tmp += pitch;
- }
- }
-}
-
-void vp8_plane_add_noise_msa(uint8_t *start_ptr, char *noise,
- char blackclamp[16], char whiteclamp[16],
- char bothclamp[16],
- uint32_t width, uint32_t height,
- int32_t pitch)
-{- uint32_t i, j;
-
- for (i = 0; i < height / 2; ++i)
- {- uint8_t *pos0_ptr = start_ptr + (2 * i) * pitch;
- int8_t *ref0_ptr = (int8_t *) (noise + (rand() & 0xff));
- uint8_t *pos1_ptr = start_ptr + (2 * i + 1) * pitch;
- int8_t *ref1_ptr = (int8_t *) (noise + (rand() & 0xff));
- for (j = width / 16; j--;)
- {- v16i8 temp00_s, temp01_s;
- v16u8 temp00, temp01, black_clamp, white_clamp;
- v16u8 pos0, ref0, pos1, ref1;
- v16i8 const127 = __msa_ldi_b(127);
-
- pos0 = LD_UB(pos0_ptr);
- ref0 = LD_UB(ref0_ptr);
- pos1 = LD_UB(pos1_ptr);
- ref1 = LD_UB(ref1_ptr);
- black_clamp = (v16u8)__msa_fill_b(blackclamp[0]);
- white_clamp = (v16u8)__msa_fill_b(whiteclamp[0]);
- temp00 = (pos0 < black_clamp);
- pos0 = __msa_bmnz_v(pos0, black_clamp, temp00);
- temp01 = (pos1 < black_clamp);
- pos1 = __msa_bmnz_v(pos1, black_clamp, temp01);
- XORI_B2_128_UB(pos0, pos1);
- temp00_s = __msa_adds_s_b((v16i8)white_clamp, const127);
- temp00 = (v16u8)(temp00_s < pos0);
- pos0 = (v16u8)__msa_bmnz_v((v16u8)pos0, (v16u8)temp00_s, temp00);
- temp01_s = __msa_adds_s_b((v16i8)white_clamp, const127);
- temp01 = (temp01_s < pos1);
- pos1 = (v16u8)__msa_bmnz_v((v16u8)pos1, (v16u8)temp01_s, temp01);
- XORI_B2_128_UB(pos0, pos1);
- pos0 += ref0;
- ST_UB(pos0, pos0_ptr);
- pos1 += ref1;
- ST_UB(pos1, pos1_ptr);
- pos0_ptr += 16;
- pos1_ptr += 16;
- ref0_ptr += 16;
- ref1_ptr += 16;
}
}
}
--- a/vp8/common/postproc.c
+++ b/vp8/common/postproc.c
@@ -10,6 +10,7 @@
#include "vpx_config.h"
+#include "vpx_dsp_rtcd.h"
#include "vp8_rtcd.h"
#include "vpx_scale_rtcd.h"
#include "vpx_scale/yv12config.h"
@@ -490,54 +491,6 @@
state->last_noise = a;
}
-/****************************************************************************
- *
- * ROUTINE : plane_add_noise_c
- *
- * INPUTS : unsigned char *Start starting address of buffer to add gaussian
- * noise to
- * unsigned int Width width of plane
- * unsigned int Height height of plane
- * int Pitch distance between subsequent lines of frame
- * int q quantizer used to determine amount of noise
- * to add
- *
- * OUTPUTS : None.
- *
- * RETURNS : void.
- *
- * FUNCTION : adds gaussian noise to a plane of pixels
- *
- * SPECIAL NOTES : None.
- *
- ****************************************************************************/
-void vp8_plane_add_noise_c(unsigned char *Start, char *noise,
- char blackclamp[16],
- char whiteclamp[16],
- char bothclamp[16],
- unsigned int Width, unsigned int Height, int Pitch)
-{- unsigned int i, j;
- (void)bothclamp;
-
- for (i = 0; i < Height; i++)
- {- unsigned char *Pos = Start + i * Pitch;
- char *Ref = (char *)(noise + (rand() & 0xff));
-
- for (j = 0; j < Width; j++)
- {- if (Pos[j] < blackclamp[0])
- Pos[j] = blackclamp[0];
-
- if (Pos[j] > 255 + whiteclamp[0])
- Pos[j] = 255 + whiteclamp[0];
-
- Pos[j] += Ref[j];
- }
- }
-}
-
/* Blend the macro block with a solid colored square. Leave the
* edges unblended to give distinction to macro blocks in areas
* filled with the same color block.
@@ -828,7 +781,7 @@
fillrd(&oci->postproc_state, 63 - q, noise_level);
}
- vp8_plane_add_noise
+ vpx_plane_add_noise
(oci->post_proc_buffer.y_buffer,
oci->postproc_state.noise,
oci->postproc_state.blackclamp,
--- a/vp8/common/rtcd_defs.pl
+++ b/vp8/common/rtcd_defs.pl
@@ -167,10 +167,6 @@
add_proto qw/void vp8_post_proc_down_and_across_mb_row/, "unsigned char *src, unsigned char *dst, int src_pitch, int dst_pitch, int cols, unsigned char *flimits, int size";
specialize qw/vp8_post_proc_down_and_across_mb_row sse2 msa/;
- add_proto qw/void vp8_plane_add_noise/, "unsigned char *s, char *noise, char blackclamp[16], char whiteclamp[16], char bothclamp[16], unsigned int w, unsigned int h, int pitch";
- specialize qw/vp8_plane_add_noise mmx sse2 msa/;
- $vp8_plane_add_noise_sse2=vp8_plane_add_noise_wmt;
-
add_proto qw/void vp8_blend_mb_inner/, "unsigned char *y, unsigned char *u, unsigned char *v, int y1, int u1, int v1, int alpha, int stride";
# no asm yet
--- a/vp8/common/x86/postproc_mmx.asm
+++ b/vp8/common/x86/postproc_mmx.asm
@@ -241,68 +241,6 @@
%undef flimit2
-;void vp8_plane_add_noise_mmx (unsigned char *Start, unsigned char *noise,
-; unsigned char blackclamp[16],
-; unsigned char whiteclamp[16],
-; unsigned char bothclamp[16],
-; unsigned int Width, unsigned int Height, int Pitch)
-global sym(vp8_plane_add_noise_mmx) PRIVATE
-sym(vp8_plane_add_noise_mmx):
- push rbp
- mov rbp, rsp
- SHADOW_ARGS_TO_STACK 8
- GET_GOT rbx
- push rsi
- push rdi
- ; end prolog
-
-.addnoise_loop:
- call sym(LIBVPX_RAND) WRT_PLT
- mov rcx, arg(1) ;noise
- and rax, 0xff
- add rcx, rax
-
- ; we rely on the fact that the clamping vectors are stored contiguously
- ; in black/white/both order. Note that we have to reload this here because
- ; rdx could be trashed by rand()
- mov rdx, arg(2) ; blackclamp
-
-
- mov rdi, rcx
- movsxd rcx, dword arg(5) ;[Width]
- mov rsi, arg(0) ;Pos
- xor rax,rax
-
-.addnoise_nextset:
- movq mm1,[rsi+rax] ; get the source
-
- psubusb mm1, [rdx] ;blackclamp ; clamp both sides so we don't outrange adding noise
- paddusb mm1, [rdx+32] ;bothclamp
- psubusb mm1, [rdx+16] ;whiteclamp
-
- movq mm2,[rdi+rax] ; get the noise for this line
- paddb mm1,mm2 ; add it in
- movq [rsi+rax],mm1 ; store the result
-
- add rax,8 ; move to the next line
-
- cmp rax, rcx
- jl .addnoise_nextset
-
- movsxd rax, dword arg(7) ; Pitch
- add arg(0), rax ; Start += Pitch
- sub dword arg(6), 1 ; Height -= 1
- jg .addnoise_loop
-
- ; begin epilog
- pop rdi
- pop rsi
- RESTORE_GOT
- UNSHADOW_ARGS
- pop rbp
- ret
-
-
SECTION_RODATA
align 16
Blur:
--- a/vp8/common/x86/postproc_sse2.asm
+++ b/vp8/common/x86/postproc_sse2.asm
@@ -655,68 +655,6 @@
%undef flimit4
-;void vp8_plane_add_noise_wmt (unsigned char *Start, unsigned char *noise,
-; unsigned char blackclamp[16],
-; unsigned char whiteclamp[16],
-; unsigned char bothclamp[16],
-; unsigned int Width, unsigned int Height, int Pitch)
-global sym(vp8_plane_add_noise_wmt) PRIVATE
-sym(vp8_plane_add_noise_wmt):
- push rbp
- mov rbp, rsp
- SHADOW_ARGS_TO_STACK 8
- GET_GOT rbx
- push rsi
- push rdi
- ; end prolog
-
-.addnoise_loop:
- call sym(LIBVPX_RAND) WRT_PLT
- mov rcx, arg(1) ;noise
- and rax, 0xff
- add rcx, rax
-
- ; we rely on the fact that the clamping vectors are stored contiguously
- ; in black/white/both order. Note that we have to reload this here because
- ; rdx could be trashed by rand()
- mov rdx, arg(2) ; blackclamp
-
-
- mov rdi, rcx
- movsxd rcx, dword arg(5) ;[Width]
- mov rsi, arg(0) ;Pos
- xor rax,rax
-
-.addnoise_nextset:
- movdqu xmm1,[rsi+rax] ; get the source
-
- psubusb xmm1, [rdx] ;blackclamp ; clamp both sides so we don't outrange adding noise
- paddusb xmm1, [rdx+32] ;bothclamp
- psubusb xmm1, [rdx+16] ;whiteclamp
-
- movdqu xmm2,[rdi+rax] ; get the noise for this line
- paddb xmm1,xmm2 ; add it in
- movdqu [rsi+rax],xmm1 ; store the result
-
- add rax,16 ; move to the next line
-
- cmp rax, rcx
- jl .addnoise_nextset
-
- movsxd rax, dword arg(7) ; Pitch
- add arg(0), rax ; Start += Pitch
- sub dword arg(6), 1 ; Height -= 1
- jg .addnoise_loop
-
- ; begin epilog
- pop rdi
- pop rsi
- RESTORE_GOT
- UNSHADOW_ARGS
- pop rbp
- ret
-
-
SECTION_RODATA
align 16
four8s:
--- a/vp9/common/vp9_postproc.c
+++ b/vp9/common/vp9_postproc.c
@@ -12,6 +12,7 @@
#include <stdlib.h>
#include <stdio.h>
+#include "./vpx_dsp_rtcd.h"
#include "./vpx_config.h"
#include "./vpx_scale_rtcd.h"
#include "./vp9_rtcd.h"
@@ -587,32 +588,6 @@
state->last_noise = a;
}
-void vp9_plane_add_noise_c(uint8_t *start, char *noise,
- char blackclamp[16],
- char whiteclamp[16],
- char bothclamp[16],
- unsigned int width, unsigned int height, int pitch) {- unsigned int i, j;
-
- // TODO(jbb): why does simd code use both but c doesn't, normalize and
- // fix..
- (void) bothclamp;
- for (i = 0; i < height; i++) {- uint8_t *pos = start + i * pitch;
- char *ref = (char *)(noise + (rand() & 0xff)); // NOLINT
-
- for (j = 0; j < width; j++) {- if (pos[j] < blackclamp[0])
- pos[j] = blackclamp[0];
-
- if (pos[j] > 255 + whiteclamp[0])
- pos[j] = 255 + whiteclamp[0];
-
- pos[j] += ref[j];
- }
- }
-}
-
static void swap_mi_and_prev_mi(VP9_COMMON *cm) {// Current mip will be the prev_mip for the next frame.
MODE_INFO *temp = cm->postproc_state.prev_mip;
@@ -726,8 +701,7 @@
ppstate->last_noise != noise_level) {fillrd(ppstate, 63 - q, noise_level);
}
-
- vp9_plane_add_noise(ppbuf->y_buffer, ppstate->noise, ppstate->blackclamp,
+ vpx_plane_add_noise(ppbuf->y_buffer, ppstate->noise, ppstate->blackclamp,
ppstate->whiteclamp, ppstate->bothclamp,
ppbuf->y_width, ppbuf->y_height, ppbuf->y_stride);
}
--- a/vp9/common/vp9_rtcd_defs.pl
+++ b/vp9/common/vp9_rtcd_defs.pl
@@ -70,10 +70,6 @@
specialize qw/vp9_post_proc_down_and_across sse2/;
$vp9_post_proc_down_and_across_sse2=vp9_post_proc_down_and_across_xmm;
-add_proto qw/void vp9_plane_add_noise/, "uint8_t *Start, char *noise, char blackclamp[16], char whiteclamp[16], char bothclamp[16], unsigned int Width, unsigned int Height, int Pitch";
-specialize qw/vp9_plane_add_noise sse2/;
-$vp9_plane_add_noise_sse2=vp9_plane_add_noise_wmt;
-
add_proto qw/void vp9_filter_by_weight16x16/, "const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int src_weight";
specialize qw/vp9_filter_by_weight16x16 sse2 msa/;
@@ -169,9 +165,6 @@
add_proto qw/void vp9_highbd_post_proc_down_and_across/, "const uint16_t *src_ptr, uint16_t *dst_ptr, int src_pixels_per_line, int dst_pixels_per_line, int rows, int cols, int flimit";
specialize qw/vp9_highbd_post_proc_down_and_across/;
-
- add_proto qw/void vp9_highbd_plane_add_noise/, "uint8_t *Start, char *noise, char blackclamp[16], char whiteclamp[16], char bothclamp[16], unsigned int Width, unsigned int Height, int Pitch";
- specialize qw/vp9_highbd_plane_add_noise/;
}
#
--- a/vp9/common/x86/vp9_postproc_sse2.asm
+++ b/vp9/common/x86/vp9_postproc_sse2.asm
@@ -624,68 +624,6 @@
%undef flimit4
-;void vp9_plane_add_noise_wmt (unsigned char *start, unsigned char *noise,
-; unsigned char blackclamp[16],
-; unsigned char whiteclamp[16],
-; unsigned char bothclamp[16],
-; unsigned int width, unsigned int height, int pitch)
-global sym(vp9_plane_add_noise_wmt) PRIVATE
-sym(vp9_plane_add_noise_wmt):
- push rbp
- mov rbp, rsp
- SHADOW_ARGS_TO_STACK 8
- GET_GOT rbx
- push rsi
- push rdi
- ; end prolog
-
-.addnoise_loop:
- call sym(LIBVPX_RAND) WRT_PLT
- mov rcx, arg(1) ;noise
- and rax, 0xff
- add rcx, rax
-
- ; we rely on the fact that the clamping vectors are stored contiguously
- ; in black/white/both order. Note that we have to reload this here because
- ; rdx could be trashed by rand()
- mov rdx, arg(2) ; blackclamp
-
-
- mov rdi, rcx
- movsxd rcx, dword arg(5) ;[Width]
- mov rsi, arg(0) ;Pos
- xor rax,rax
-
-.addnoise_nextset:
- movdqu xmm1,[rsi+rax] ; get the source
-
- psubusb xmm1, [rdx] ;blackclamp ; clamp both sides so we don't outrange adding noise
- paddusb xmm1, [rdx+32] ;bothclamp
- psubusb xmm1, [rdx+16] ;whiteclamp
-
- movdqu xmm2,[rdi+rax] ; get the noise for this line
- paddb xmm1,xmm2 ; add it in
- movdqu [rsi+rax],xmm1 ; store the result
-
- add rax,16 ; move to the next line
-
- cmp rax, rcx
- jl .addnoise_nextset
-
- movsxd rax, dword arg(7) ; Pitch
- add arg(0), rax ; Start += Pitch
- sub dword arg(6), 1 ; Height -= 1
- jg .addnoise_loop
-
- ; begin epilog
- pop rdi
- pop rsi
- RESTORE_GOT
- UNSHADOW_ARGS
- pop rbp
- ret
-
-
SECTION_RODATA
align 16
rd42:
--- /dev/null
+++ b/vpx_dsp/mips/postproc_msa.c
@@ -1,0 +1,59 @@
+/*
+ * Copyright (c) 2015 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 <stdlib.h>
+#include "./macros_msa.h"
+
+void vpx_plane_add_noise_msa(uint8_t *start_ptr, char *noise,
+ char blackclamp[16], char whiteclamp[16],
+ char bothclamp[16], uint32_t width,
+ uint32_t height, int32_t pitch) {+ uint32_t i, j;
+
+ for (i = 0; i < height / 2; ++i) {+ uint8_t *pos0_ptr = start_ptr + (2 * i) * pitch;
+ int8_t *ref0_ptr = (int8_t *)(noise + (rand() & 0xff));
+ uint8_t *pos1_ptr = start_ptr + (2 * i + 1) * pitch;
+ int8_t *ref1_ptr = (int8_t *)(noise + (rand() & 0xff));
+ for (j = width / 16; j--;) {+ v16i8 temp00_s, temp01_s;
+ v16u8 temp00, temp01, black_clamp, white_clamp;
+ v16u8 pos0, ref0, pos1, ref1;
+ v16i8 const127 = __msa_ldi_b(127);
+
+ pos0 = LD_UB(pos0_ptr);
+ ref0 = LD_UB(ref0_ptr);
+ pos1 = LD_UB(pos1_ptr);
+ ref1 = LD_UB(ref1_ptr);
+ black_clamp = (v16u8)__msa_fill_b(blackclamp[0]);
+ white_clamp = (v16u8)__msa_fill_b(whiteclamp[0]);
+ temp00 = (pos0 < black_clamp);
+ pos0 = __msa_bmnz_v(pos0, black_clamp, temp00);
+ temp01 = (pos1 < black_clamp);
+ pos1 = __msa_bmnz_v(pos1, black_clamp, temp01);
+ XORI_B2_128_UB(pos0, pos1);
+ temp00_s = __msa_adds_s_b((v16i8)white_clamp, const127);
+ temp00 = (v16u8)(temp00_s < pos0);
+ pos0 = (v16u8)__msa_bmnz_v((v16u8)pos0, (v16u8)temp00_s, temp00);
+ temp01_s = __msa_adds_s_b((v16i8)white_clamp, const127);
+ temp01 = (temp01_s < pos1);
+ pos1 = (v16u8)__msa_bmnz_v((v16u8)pos1, (v16u8)temp01_s, temp01);
+ XORI_B2_128_UB(pos0, pos1);
+ pos0 += ref0;
+ ST_UB(pos0, pos0_ptr);
+ pos1 += ref1;
+ ST_UB(pos1, pos1_ptr);
+ pos0_ptr += 16;
+ pos1_ptr += 16;
+ ref0_ptr += 16;
+ ref1_ptr += 16;
+ }
+ }
+}
--- /dev/null
+++ b/vpx_dsp/postproc.c
@@ -1,0 +1,43 @@
+/*
+ * Copyright (c) 2015 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 <stdlib.h>
+
+#include "./vpx_config.h"
+#include "./vpx_dsp_rtcd.h"
+
+#include "vpx/vpx_integer.h"
+#include "vpx_ports/mem.h"
+
+void vpx_plane_add_noise_c(uint8_t *start, char *noise,
+ char blackclamp[16],
+ char whiteclamp[16],
+ char bothclamp[16],
+ unsigned int width, unsigned int height, int pitch) {+ unsigned int i, j;
+
+ // TODO(jbb): why does simd code use both but c doesn't, normalize and
+ // fix..
+ (void) bothclamp;
+ for (i = 0; i < height; i++) {+ uint8_t *pos = start + i * pitch;
+ char *ref = (char *)(noise + (rand() & 0xff)); // NOLINT
+
+ for (j = 0; j < width; j++) {+ if (pos[j] < blackclamp[0])
+ pos[j] = blackclamp[0];
+
+ if (pos[j] > 255 + whiteclamp[0])
+ pos[j] = 255 + whiteclamp[0];
+
+ pos[j] += ref[j];
+ }
+ }
+}
--- a/vpx_dsp/vpx_dsp.mk
+++ b/vpx_dsp/vpx_dsp.mk
@@ -53,6 +53,13 @@
endif # CONFIG_USE_X86INC
endif # CONFIG_VP9_HIGHBITDEPTH
+ifneq ($(filter yes,$(CONFIG_POSTPROC) $(CONFIG_VP9_POSTPROC)),)
+DSP_SRCS-yes += postproc.c
+DSP_SRCS-$(HAVE_MSA) += mips/postproc_msa.c
+DSP_SRCS-$(HAVE_MMX) += x86/postproc_mmx.asm
+DSP_SRCS-$(HAVE_SSE2) += x86/postproc_sse2.asm
+endif # CONFIG_POSTPROC
+
DSP_SRCS-$(HAVE_NEON_ASM) += arm/intrapred_neon_asm$(ASM)
DSP_SRCS-$(HAVE_NEON) += arm/intrapred_neon.c
DSP_SRCS-$(HAVE_MSA) += mips/intrapred_msa.c
--- a/vpx_dsp/vpx_dsp_rtcd_defs.pl
+++ b/vpx_dsp/vpx_dsp_rtcd_defs.pl
@@ -1907,6 +1907,15 @@
add_proto qw/uint32_t vpx_highbd_8_sub_pixel_avg_variance4x4/, "const uint8_t *src_ptr, int source_stride, int xoffset, int yoffset, const uint8_t *ref_ptr, int ref_stride, uint32_t *sse, const uint8_t *second_pred";
} # CONFIG_VP9_HIGHBITDEPTH
+
+#
+# Post Processing
+#
+if (vpx_config("CONFIG_POSTPROC") eq "yes" || vpx_config("CONFIG_VP9_POSTPROC") eq "yes") {+ add_proto qw/void vpx_plane_add_noise/, "uint8_t *Start, char *noise, char blackclamp[16], char whiteclamp[16], char bothclamp[16], unsigned int Width, unsigned int Height, int Pitch";
+ specialize qw/vpx_plane_add_noise mmx sse2 msa/;
+}
+
} # CONFIG_ENCODERS || CONFIG_POSTPROC || CONFIG_VP9_POSTPROC
1;
--- /dev/null
+++ b/vpx_dsp/x86/postproc_mmx.asm
@@ -1,0 +1,84 @@
+;
+; Copyright (c) 2015 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 "vpx_ports/x86_abi_support.asm"
+
+;void vpx_plane_add_noise_mmx (unsigned char *Start, unsigned char *noise,
+; unsigned char blackclamp[16],
+; unsigned char whiteclamp[16],
+; unsigned char bothclamp[16],
+; unsigned int Width, unsigned int Height, int Pitch)
+global sym(vpx_plane_add_noise_mmx) PRIVATE
+sym(vpx_plane_add_noise_mmx):
+ push rbp
+ mov rbp, rsp
+ SHADOW_ARGS_TO_STACK 8
+ GET_GOT rbx
+ push rsi
+ push rdi
+ ; end prolog
+
+.addnoise_loop:
+ call sym(LIBVPX_RAND) WRT_PLT
+ mov rcx, arg(1) ;noise
+ and rax, 0xff
+ add rcx, rax
+
+ ; we rely on the fact that the clamping vectors are stored contiguously
+ ; in black/white/both order. Note that we have to reload this here because
+ ; rdx could be trashed by rand()
+ mov rdx, arg(2) ; blackclamp
+
+
+ mov rdi, rcx
+ movsxd rcx, dword arg(5) ;[Width]
+ mov rsi, arg(0) ;Pos
+ xor rax,rax
+
+.addnoise_nextset:
+ movq mm1,[rsi+rax] ; get the source
+
+ psubusb mm1, [rdx] ;blackclamp ; clamp both sides so we don't outrange adding noise
+ paddusb mm1, [rdx+32] ;bothclamp
+ psubusb mm1, [rdx+16] ;whiteclamp
+
+ movq mm2,[rdi+rax] ; get the noise for this line
+ paddb mm1,mm2 ; add it in
+ movq [rsi+rax],mm1 ; store the result
+
+ add rax,8 ; move to the next line
+
+ cmp rax, rcx
+ jl .addnoise_nextset
+
+ movsxd rax, dword arg(7) ; Pitch
+ add arg(0), rax ; Start += Pitch
+ sub dword arg(6), 1 ; Height -= 1
+ jg .addnoise_loop
+
+ ; begin epilog
+ pop rdi
+ pop rsi
+ RESTORE_GOT
+ UNSHADOW_ARGS
+ pop rbp
+ ret
+
+
+SECTION_RODATA
+align 16
+Blur:
+ times 16 dw 16
+ times 8 dw 64
+ times 16 dw 16
+ times 8 dw 0
+
+rd:
+ times 4 dw 0x40
--- /dev/null
+++ b/vpx_dsp/x86/postproc_sse2.asm
@@ -1,0 +1,82 @@
+;
+; Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+;
+; Use of this source code is governed by a BSD-style license
+; that can be found in the LICENSE file in the root of the source
+; tree. An additional intellectual property rights grant can be found
+; in the file PATENTS. All contributing project authors may
+; be found in the AUTHORS file in the root of the source tree.
+;
+
+
+%include "vpx_ports/x86_abi_support.asm"
+
+;void vpx_plane_add_noise_sse2(unsigned char *start, unsigned char *noise,
+; unsigned char blackclamp[16],
+; unsigned char whiteclamp[16],
+; unsigned char bothclamp[16],
+; unsigned int width, unsigned int height,
+; int pitch)
+global sym(vpx_plane_add_noise_sse2) PRIVATE
+sym(vpx_plane_add_noise_sse2):
+ push rbp
+ mov rbp, rsp
+ SHADOW_ARGS_TO_STACK 8
+ GET_GOT rbx
+ push rsi
+ push rdi
+ ; end prolog
+
+.addnoise_loop:
+ call sym(LIBVPX_RAND) WRT_PLT
+ mov rcx, arg(1) ;noise
+ and rax, 0xff
+ add rcx, rax
+
+ ; we rely on the fact that the clamping vectors are stored contiguously
+ ; in black/white/both order. Note that we have to reload this here because
+ ; rdx could be trashed by rand()
+ mov rdx, arg(2) ; blackclamp
+
+
+ mov rdi, rcx
+ movsxd rcx, dword arg(5) ;[Width]
+ mov rsi, arg(0) ;Pos
+ xor rax,rax
+
+.addnoise_nextset:
+ movdqu xmm1,[rsi+rax] ; get the source
+
+ psubusb xmm1, [rdx] ;blackclamp ; clamp both sides so we don't outrange adding noise
+ paddusb xmm1, [rdx+32] ;bothclamp
+ psubusb xmm1, [rdx+16] ;whiteclamp
+
+ movdqu xmm2,[rdi+rax] ; get the noise for this line
+ paddb xmm1,xmm2 ; add it in
+ movdqu [rsi+rax],xmm1 ; store the result
+
+ add rax,16 ; move to the next line
+
+ cmp rax, rcx
+ jl .addnoise_nextset
+
+ movsxd rax, dword arg(7) ; Pitch
+ add arg(0), rax ; Start += Pitch
+ sub dword arg(6), 1 ; Height -= 1
+ jg .addnoise_loop
+
+ ; begin epilog
+ pop rdi
+ pop rsi
+ RESTORE_GOT
+ UNSHADOW_ARGS
+ pop rbp
+ ret
+
+
+SECTION_RODATA
+align 16
+rd42:
+ times 8 dw 0x04
+four8s:
+ times 4 dd 8