ref: da33728f488059193671af031f49c9581eb4be4a
parent: 04246a60d781ac5450b5099f73befb3293415e02
author: Jim Bankoski <[email protected]>
date: Tue May 10 02:52:58 EDT 2016
vpx_dsp: Rename postproc.c add_noise. Change-Id: I4906d1b79a2951e659995202b9fa97e2ea5cfba0
--- /dev/null
+++ b/vpx_dsp/add_noise.c
@@ -1,0 +1,40 @@
+/*
+ * 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;
+
+ 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++) {+ int v = pos[j];
+
+ v = clamp(v - blackclamp[0], 0, 255);
+ v = clamp(v + bothclamp[0], 0, 255);
+ v = clamp(v - whiteclamp[0], 0, 255);
+
+ pos[j] = v + ref[j];
+ }
+ }
+}
--- /dev/null
+++ b/vpx_dsp/mips/add_noise_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;
+ }
+ }
+}
--- a/vpx_dsp/mips/postproc_msa.c
+++ /dev/null
@@ -1,59 +1,0 @@
-/*
- * 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;
- }
- }
-}
--- a/vpx_dsp/postproc.c
+++ /dev/null
@@ -1,40 +1,0 @@
-/*
- * 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;
-
- 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++) {- int v = pos[j];
-
- v = clamp(v - blackclamp[0], 0, 255);
- v = clamp(v + bothclamp[0], 0, 255);
- v = clamp(v - whiteclamp[0], 0, 255);
-
- pos[j] = v + ref[j];
- }
- }
-}
--- a/vpx_dsp/vpx_dsp.mk
+++ b/vpx_dsp/vpx_dsp.mk
@@ -54,10 +54,10 @@
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
+DSP_SRCS-yes += add_noise.c
+DSP_SRCS-$(HAVE_MSA) += mips/add_noise_msa.c
+DSP_SRCS-$(HAVE_MMX) += x86/add_noise_mmx.asm
+DSP_SRCS-$(HAVE_SSE2) += x86/add_noise_sse2.asm
endif # CONFIG_POSTPROC
DSP_SRCS-$(HAVE_NEON_ASM) += arm/intrapred_neon_asm$(ASM)
--- /dev/null
+++ b/vpx_dsp/x86/add_noise_mmx.asm
@@ -1,0 +1,86 @@
+;
+; 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
+
+ ; get the clamps in registers
+ mov rdx, arg(2) ; blackclamp
+ movq mm3, [rdx]
+ mov rdx, arg(3) ; whiteclamp
+ movq mm4, [rdx]
+ mov rdx, arg(4) ; bothclamp
+ movq mm5, [rdx]
+
+.addnoise_loop:
+ call sym(LIBVPX_RAND) WRT_PLT
+ mov rcx, arg(1) ;noise
+ and rax, 0xff
+ add rcx, rax
+
+ 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, mm3 ; subtract black clamp
+ paddusb mm1, mm5 ; add both clamp
+ psubusb mm1, mm4 ; subtract 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/add_noise_sse2.asm
@@ -1,0 +1,83 @@
+;
+; 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
+
+ ; get the clamps in registers
+ mov rdx, arg(2) ; blackclamp
+ movdqu xmm3, [rdx]
+ mov rdx, arg(3) ; whiteclamp
+ movdqu xmm4, [rdx]
+ mov rdx, arg(4) ; bothclamp
+ movdqu xmm5, [rdx]
+
+.addnoise_loop:
+ call sym(LIBVPX_RAND) WRT_PLT
+ mov rcx, arg(1) ;noise
+ and rax, 0xff
+ add rcx, rax
+
+ 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, xmm3 ; subtract black clamp
+ paddusb xmm1, xmm5 ; add both clamp
+ psubusb xmm1, xmm4 ; subtract 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
--- a/vpx_dsp/x86/postproc_mmx.asm
+++ /dev/null
@@ -1,86 +1,0 @@
-;
-; 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
-
- ; get the clamps in registers
- mov rdx, arg(2) ; blackclamp
- movq mm3, [rdx]
- mov rdx, arg(3) ; whiteclamp
- movq mm4, [rdx]
- mov rdx, arg(4) ; bothclamp
- movq mm5, [rdx]
-
-.addnoise_loop:
- call sym(LIBVPX_RAND) WRT_PLT
- mov rcx, arg(1) ;noise
- and rax, 0xff
- add rcx, rax
-
- 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, mm3 ; subtract black clamp
- paddusb mm1, mm5 ; add both clamp
- psubusb mm1, mm4 ; subtract 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
--- a/vpx_dsp/x86/postproc_sse2.asm
+++ /dev/null
@@ -1,83 +1,0 @@
-;
-; 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
-
- ; get the clamps in registers
- mov rdx, arg(2) ; blackclamp
- movdqu xmm3, [rdx]
- mov rdx, arg(3) ; whiteclamp
- movdqu xmm4, [rdx]
- mov rdx, arg(4) ; bothclamp
- movdqu xmm5, [rdx]
-
-.addnoise_loop:
- call sym(LIBVPX_RAND) WRT_PLT
- mov rcx, arg(1) ;noise
- and rax, 0xff
- add rcx, rax
-
- 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, xmm3 ; subtract black clamp
- paddusb xmm1, xmm5 ; add both clamp
- psubusb xmm1, xmm4 ; subtract 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