ryan/mozjpeg
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

32-bit AVX2 impl. of h2v2 & h2v1 downsampling

Browse Source
This commit is contained in:
DRC
2016-07-05 15:50:50 -05:00
parent 6448a5e584
commit c6300ffe4e
4 changed files with 404 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
simd/CMakeLists.txt
View File

@@ -35,7 +35,7 @@ else()
jidctflt-sse jquant-sse jccolor-sse2 jcgray-sse2 jchuff-sse2 jcsample-sse2
jdcolor-sse2 jdmerge-sse2 jdsample-sse2 jfdctfst-sse2 jfdctint-sse2
jidctflt-sse2 jidctfst-sse2 jidctint-sse2 jidctred-sse2 jquantf-sse2
jquanti-sse2 jccolor-avx2 jcgray-avx2)
jquanti-sse2 jccolor-avx2 jcgray-avx2 jcsample-avx2)
message(STATUS "Building i386 SIMD extensions")
endif()

2
simd/Makefile.am
View File

@@ -48,7 +48,7 @@ libsimd_la_SOURCES = jsimd_i386.c jsimd.h jsimdcfg.inc.h jsimdext.inc \
jdsample-sse2.asm jfdctfst-sse2.asm jfdctint-sse2.asm \
jidctflt-sse2.asm jidctfst-sse2.asm jidctint-sse2.asm \
jidctred-sse2.asm jquantf-sse2.asm jquanti-sse2.asm \
jccolor-avx2.asm jcgray-avx2.asm
jccolor-avx2.asm jcgray-avx2.asm jcsample-avx2.asm
jccolor-mmx.lo: jccolext-mmx.asm
jcgray.-mmx.lo: jcgryext-mmx.asm

386
simd/jcsample-avx2.asm Normal file
View File

@@ -0,0 +1,386 @@
;
; jcsample.asm - downsampling (AVX2)
;
; Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
; Copyright (C) 2015, Intel Corporation.
; Copyright (C) 2016, D. R. Commander.
;
; Based on the x86 SIMD extension for IJG JPEG library
; Copyright (C) 1999-2006, MIYASAKA Masaru.
; For conditions of distribution and use, see copyright notice in jsimdext.inc
;
; This file should be assembled with NASM (Netwide Assembler),
; can *not* be assembled with Microsoft's MASM or any compatible
; assembler (including Borland's Turbo Assembler).
; NASM is available from http://nasm.sourceforge.net/ or
; http://sourceforge.net/project/showfiles.php?group_id=6208
;
; [TAB8]
%include "jsimdext.inc"
; --------------------------------------------------------------------------
SECTION SEG_TEXT
BITS 32
;
; Downsample pixel values of a single component.
; This version handles the common case of 2:1 horizontal and 1:1 vertical,
; without smoothing.
;
; GLOBAL(void)
; jsimd_h2v1_downsample_avx2 (JDIMENSION image_width, int max_v_samp_factor,
; JDIMENSION v_samp_factor, JDIMENSION width_blocks,
; JSAMPARRAY input_data, JSAMPARRAY output_data);
;
%define img_width(b) (b)+8 ; JDIMENSION image_width
%define max_v_samp(b) (b)+12 ; int max_v_samp_factor
%define v_samp(b) (b)+16 ; JDIMENSION v_samp_factor
%define width_blks(b) (b)+20 ; JDIMENSION width_blocks
%define input_data(b) (b)+24 ; JSAMPARRAY input_data
%define output_data(b) (b)+28 ; JSAMPARRAY output_data
align 32
global EXTN(jsimd_h2v1_downsample_avx2)
EXTN(jsimd_h2v1_downsample_avx2):
push ebp
mov ebp, esp
; push ebx ; unused
; push ecx ; need not be preserved
; push edx ; need not be preserved
push esi
push edi
mov ecx, JDIMENSION [width_blks(ebp)]
shl ecx, 3 ; imul ecx,DCTSIZE (ecx = output_cols)
jz near .return
mov edx, JDIMENSION [img_width(ebp)]
; -- expand_right_edge
push ecx
shl ecx, 1 ; output_cols * 2
sub ecx, edx
jle short .expand_end
mov eax, INT [max_v_samp(ebp)]
test eax, eax
jle short .expand_end
cld
mov esi, JSAMPARRAY [input_data(ebp)] ; input_data
alignx 16, 7
.expandloop:
push eax
push ecx
mov edi, JSAMPROW [esi]
add edi, edx
mov al, JSAMPLE [edi-1]
rep stosb
pop ecx
pop eax
add esi, byte SIZEOF_JSAMPROW
dec eax
jg short .expandloop
.expand_end:
pop ecx ; output_cols
; -- h2v1_downsample
mov eax, JDIMENSION [v_samp(ebp)] ; rowctr
test eax, eax
jle near .return
mov edx, 0x00010000 ; bias pattern
vmovd xmm7, edx
vpshufd xmm7, xmm7, 0x00 ; xmm7={0, 1, 0, 1, 0, 1, 0, 1}
vperm2i128 ymm7, ymm7, ymm7, 0 ; ymm7={xmm7, xmm7}
vpcmpeqw ymm6, ymm6, ymm6
vpsrlw ymm6, ymm6, BYTE_BIT ; ymm6={0xFF 0x00 0xFF 0x00 ..}
mov esi, JSAMPARRAY [input_data(ebp)] ; input_data
mov edi, JSAMPARRAY [output_data(ebp)] ; output_data
alignx 16, 7
.rowloop:
push ecx
push edi
push esi
mov esi, JSAMPROW [esi] ; inptr
mov edi, JSAMPROW [edi] ; outptr
cmp ecx, byte SIZEOF_YMMWORD
jae short .columnloop
alignx 16, 7
.columnloop_r24:
; ecx can possibly be 8, 16, 24
cmp ecx, 24
jne .columnloop_r16
vmovdqu ymm0, YMMWORD [esi+0*SIZEOF_YMMWORD]
vmovdqu xmm1, XMMWORD [esi+1*SIZEOF_YMMWORD]
mov ecx, SIZEOF_YMMWORD
jmp short .downsample
.columnloop_r16:
cmp ecx, 16
jne .columnloop_r8
vmovdqu ymm0, YMMWORD [esi+0*SIZEOF_YMMWORD]
vpxor ymm1, ymm1, ymm1
mov ecx, SIZEOF_YMMWORD
jmp short .downsample
.columnloop_r8:
vmovdqu xmm0, XMMWORD[esi+0*SIZEOF_YMMWORD]
vpxor ymm1, ymm1, ymm1
mov ecx, SIZEOF_YMMWORD
jmp short .downsample
alignx 16, 7
.columnloop:
vmovdqu ymm0, YMMWORD [esi+0*SIZEOF_YMMWORD]
vmovdqu ymm1, YMMWORD [esi+1*SIZEOF_YMMWORD]
.downsample:
vpsrlw ymm2, ymm0, BYTE_BIT
vpand ymm0, ymm0, ymm6
vpsrlw ymm3, ymm1, BYTE_BIT
vpand ymm1, ymm1, ymm6
vpaddw ymm0, ymm0, ymm2
vpaddw ymm1, ymm1, ymm3
vpaddw ymm0, ymm0, ymm7
vpaddw ymm1, ymm1, ymm7
vpsrlw ymm0, ymm0, 1
vpsrlw ymm1, ymm1, 1
vpackuswb ymm0, ymm0, ymm1
vpermq ymm0, ymm0, 0xd8
vmovdqu YMMWORD [edi+0*SIZEOF_YMMWORD], ymm0
sub ecx, byte SIZEOF_YMMWORD ; outcol
add esi, byte 2*SIZEOF_YMMWORD ; inptr
add edi, byte 1*SIZEOF_YMMWORD ; outptr
cmp ecx, byte SIZEOF_YMMWORD
jae short .columnloop
test ecx, ecx
jnz near .columnloop_r24
pop esi
pop edi
pop ecx
add esi, byte SIZEOF_JSAMPROW ; input_data
add edi, byte SIZEOF_JSAMPROW ; output_data
dec eax ; rowctr
jg near .rowloop
.return:
pop edi
pop esi
; pop edx ; need not be preserved
; pop ecx ; need not be preserved
; pop ebx ; unused
pop ebp
ret
; --------------------------------------------------------------------------
;
; Downsample pixel values of a single component.
; This version handles the standard case of 2:1 horizontal and 2:1 vertical,
; without smoothing.
;
; GLOBAL(void)
; jsimd_h2v2_downsample_avx2 (JDIMENSION image_width, int max_v_samp_factor,
; JDIMENSION v_samp_factor, JDIMENSION width_blocks,
; JSAMPARRAY input_data, JSAMPARRAY output_data);
;
%define img_width(b) (b)+8 ; JDIMENSION image_width
%define max_v_samp(b) (b)+12 ; int max_v_samp_factor
%define v_samp(b) (b)+16 ; JDIMENSION v_samp_factor
%define width_blks(b) (b)+20 ; JDIMENSION width_blocks
%define input_data(b) (b)+24 ; JSAMPARRAY input_data
%define output_data(b) (b)+28 ; JSAMPARRAY output_data
align 32
global EXTN(jsimd_h2v2_downsample_avx2)
EXTN(jsimd_h2v2_downsample_avx2):
push ebp
mov ebp, esp
; push ebx ; unused
; push ecx ; need not be preserved
; push edx ; need not be preserved
push esi
push edi
mov ecx, JDIMENSION [width_blks(ebp)]
shl ecx, 3 ; imul ecx,DCTSIZE (ecx = output_cols)
jz near .return
mov edx, JDIMENSION [img_width(ebp)]
; -- expand_right_edge
push ecx
shl ecx, 1 ; output_cols * 2
sub ecx, edx
jle short .expand_end
mov eax, INT [max_v_samp(ebp)]
test eax, eax
jle short .expand_end
cld
mov esi, JSAMPARRAY [input_data(ebp)] ; input_data
alignx 16, 7
.expandloop:
push eax
push ecx
mov edi, JSAMPROW [esi]
add edi, edx
mov al, JSAMPLE [edi-1]
rep stosb
pop ecx
pop eax
add esi, byte SIZEOF_JSAMPROW
dec eax
jg short .expandloop
.expand_end:
pop ecx ; output_cols
; -- h2v2_downsample
mov eax, JDIMENSION [v_samp(ebp)] ; rowctr
test eax, eax
jle near .return
mov edx, 0x00020001 ; bias pattern
vmovd xmm7, edx
vpcmpeqw ymm6, ymm6, ymm6
vpshufd xmm7, xmm7, 0x00 ; ymm7={1, 2, 1, 2, 1, 2, 1, 2}
vperm2i128 ymm7, ymm7, ymm7, 0
vpsrlw ymm6, ymm6, BYTE_BIT ; ymm6={0xFF 0x00 0xFF 0x00 ..}
mov esi, JSAMPARRAY [input_data(ebp)] ; input_data
mov edi, JSAMPARRAY [output_data(ebp)] ; output_data
alignx 16, 7
.rowloop:
push ecx
push edi
push esi
mov edx, JSAMPROW [esi+0*SIZEOF_JSAMPROW] ; inptr0
mov esi, JSAMPROW [esi+1*SIZEOF_JSAMPROW] ; inptr1
mov edi, JSAMPROW [edi] ; outptr
cmp ecx, byte SIZEOF_YMMWORD
jae short .columnloop
alignx 16, 7
.columnloop_r24:
cmp ecx, 24
jne .columnloop_r16
vmovdqu ymm0, YMMWORD [edx+0*SIZEOF_YMMWORD]
vmovdqu ymm1, YMMWORD [esi+0*SIZEOF_YMMWORD]
vmovdqu xmm2, XMMWORD [edx+1*SIZEOF_YMMWORD]
vmovdqu xmm3, XMMWORD [esi+1*SIZEOF_YMMWORD]
mov ecx, SIZEOF_YMMWORD
jmp short .downsample
.columnloop_r16:
cmp ecx, 16
jne .columnloop_r8
vmovdqu ymm0, YMMWORD [edx+0*SIZEOF_YMMWORD]
vmovdqu ymm1, YMMWORD [esi+0*SIZEOF_YMMWORD]
vpxor ymm2, ymm2, ymm2
vpxor ymm3, ymm3, ymm3
mov ecx, SIZEOF_YMMWORD
jmp short .downsample
.columnloop_r8:
vmovdqu xmm0, XMMWORD [edx+0*SIZEOF_XMMWORD]
vmovdqu xmm1, XMMWORD [esi+0*SIZEOF_XMMWORD]
vpxor ymm2, ymm2, ymm2
vpxor ymm3, ymm3, ymm3
mov ecx, SIZEOF_YMMWORD
jmp short .downsample
alignx 16, 7
.columnloop:
vmovdqu ymm0, YMMWORD [edx+0*SIZEOF_YMMWORD]
vmovdqu ymm1, YMMWORD [esi+0*SIZEOF_YMMWORD]
vmovdqu ymm2, YMMWORD [edx+1*SIZEOF_YMMWORD]
vmovdqu ymm3, YMMWORD [esi+1*SIZEOF_YMMWORD]
.downsample:
vpand ymm4, ymm0, ymm6
vpsrlw ymm0, ymm0, BYTE_BIT
vpand ymm5, ymm1, ymm6
vpsrlw ymm1, ymm1, BYTE_BIT
vpaddw ymm0, ymm0, ymm4
vpaddw ymm1, ymm1, ymm5
vpand ymm4, ymm2, ymm6
vpsrlw ymm2, ymm2, BYTE_BIT
vpand ymm5, ymm3, ymm6
vpsrlw ymm3, ymm3, BYTE_BIT
vpaddw ymm2, ymm2, ymm4
vpaddw ymm3, ymm3, ymm5
vpaddw ymm0, ymm0, ymm1
vpaddw ymm2, ymm2, ymm3
vpaddw ymm0, ymm0, ymm7
vpaddw ymm2, ymm2, ymm7
vpsrlw ymm0, ymm0, 2
vpsrlw ymm2, ymm2, 2
vpackuswb ymm0, ymm0, ymm2
vpermq ymm0, ymm0, 0xd8
vmovdqu YMMWORD [edi+0*SIZEOF_YMMWORD], ymm0
sub ecx, byte SIZEOF_YMMWORD ; outcol
add edx, byte 2*SIZEOF_YMMWORD ; inptr0
add esi, byte 2*SIZEOF_YMMWORD ; inptr1
add edi, byte 1*SIZEOF_YMMWORD ; outptr
cmp ecx, byte SIZEOF_YMMWORD
jae near .columnloop
test ecx, ecx
jnz near .columnloop_r24
pop esi
pop edi
pop ecx
add esi, byte 2*SIZEOF_JSAMPROW ; input_data
add edi, byte 1*SIZEOF_JSAMPROW ; output_data
dec eax ; rowctr
jg near .rowloop
.return:
pop edi
pop esi
; pop edx ; need not be preserved
; pop ecx ; need not be preserved
; pop ebx ; unused
pop ebp
ret
; For some reason, the OS X linker does not honor the request to align the
; segment unless we do this.
align 32

18
simd/jsimd_i386.c
View File

@@ -337,6 +337,8 @@ jsimd_can_h2v2_downsample (void)
if (sizeof(JDIMENSION) != 4)
return 0;
if (simd_support & JSIMD_AVX2)
return 1;
if (simd_support & JSIMD_SSE2)
return 1;
if (simd_support & JSIMD_MMX)
@@ -356,6 +358,8 @@ jsimd_can_h2v1_downsample (void)
if (sizeof(JDIMENSION) != 4)
return 0;
if (simd_support & JSIMD_AVX2)
return 1;
if (simd_support & JSIMD_SSE2)
return 1;
if (simd_support & JSIMD_MMX)
@@ -368,7 +372,12 @@ GLOBAL(void)
jsimd_h2v2_downsample (j_compress_ptr cinfo, jpeg_component_info *compptr,
JSAMPARRAY input_data, JSAMPARRAY output_data)
{
if (simd_support & JSIMD_SSE2)
if (simd_support & JSIMD_AVX2)
jsimd_h2v2_downsample_avx2(cinfo->image_width, cinfo->max_v_samp_factor,
compptr->v_samp_factor,
compptr->width_in_blocks, input_data,
output_data);
else if (simd_support & JSIMD_SSE2)
jsimd_h2v2_downsample_sse2(cinfo->image_width, cinfo->max_v_samp_factor,
compptr->v_samp_factor,
compptr->width_in_blocks, input_data,
@@ -383,7 +392,12 @@ GLOBAL(void)
jsimd_h2v1_downsample (j_compress_ptr cinfo, jpeg_component_info *compptr,
JSAMPARRAY input_data, JSAMPARRAY output_data)
{
if (simd_support & JSIMD_SSE2)
if (simd_support & JSIMD_AVX2)
jsimd_h2v1_downsample_avx2(cinfo->image_width, cinfo->max_v_samp_factor,
compptr->v_samp_factor,
compptr->width_in_blocks, input_data,
output_data);
else if (simd_support & JSIMD_SSE2)
jsimd_h2v1_downsample_sse2(cinfo->image_width, cinfo->max_v_samp_factor,
compptr->v_samp_factor,
compptr->width_in_blocks, input_data,