ryan/mozjpeg
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity
Files
19c791cdac6df84418c66eb9d67cfa9703bc2461
mozjpeg/simd/i386/jcgryext-avx2.asm
DRC 19c791cdac Improve code formatting consistency 
With rare exceptions ...
- Always separate line continuation characters by one space from
  preceding code.
- Always use two-space indentation.  Never use tabs.
- Always use K&R-style conditional blocks.
- Always surround operators with spaces, except in raw assembly code.
- Always put a space after, but not before, a comma.
- Never put a space between type casts and variables/function calls.
- Never put a space between the function name and the argument list in
  function declarations and prototypes.
- Always surround braces ('{' and '}') with spaces.
- Always surround statements (if, for, else, catch, while, do, switch)
  with spaces.
- Always attach pointer symbols ('*' and '**') to the variable or
  function name.
- Always precede pointer symbols ('*' and '**') by a space in type
  casts.
- Use the MIN() macro from jpegint.h within the libjpeg and TurboJPEG
  API libraries (using min() from tjutil.h is still necessary for
  TJBench.)
- Where it makes sense (particularly in the TurboJPEG code), put a blank
  line after variable declaration blocks.
- Always separate statements in one-liners by two spaces.

The purpose of this was to ease maintenance on my part and also to make
it easier for contributors to figure out how to format patch
submissions.  This was admittedly confusing (even to me sometimes) when
we had 3 or 4 different style conventions in the same source tree.  The
new convention is more consistent with the formatting of other OSS code
bases.

This commit corrects deviations from the chosen formatting style in the
libjpeg API code and reformats the TurboJPEG API code such that it
conforms to the same standard.

NOTES:
- Although it is no longer necessary for the function name in function
  declarations to begin in Column 1 (this was historically necessary
  because of the ansi2knr utility, which allowed libjpeg to be built
  with non-ANSI compilers), we retain that formatting for the libjpeg
  code because it improves readability when using libjpeg's function
  attribute macros (GLOBAL(), etc.)
- This reformatting project was accomplished with the help of AStyle and
  Uncrustify, although neither was completely up to the task, and thus
  a great deal of manual tweaking was required.  Note to developers of
  code formatting utilities:  the libjpeg-turbo code base is an
  excellent test bed, because AFAICT, it breaks every single one of the
  utilities that are currently available.
- The legacy (MMX, SSE, 3DNow!) assembly code for i386 has been
  formatted to match the SSE2 code (refer to
  ff5685d5344273df321eb63a005eaae19d2496e3.)  I hadn't intended to
  bother with this, but the Loongson MMI implementation demonstrated
  that there is still academic value to the MMX implementation, as an
  algorithmic model for other 64-bit vector implementations.  Thus, it
  is desirable to improve its readability in the same manner as that of
  the SSE2 implementation.
2018-03-16 02:14:34 -05:00

460 lines
20 KiB
NASM
Raw Blame History

;
; jcgryext.asm - grayscale colorspace conversion (AVX2)
;
; Copyright (C) 2011, 2016, D. R. Commander.
; Copyright (C) 2015, Intel Corporation.
;
; 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 "jcolsamp.inc"
; --------------------------------------------------------------------------
;
; Convert some rows of samples to the output colorspace.
;
; GLOBAL(void)
; jsimd_rgb_gray_convert_avx2(JDIMENSION img_width, JSAMPARRAY input_buf,
; JSAMPIMAGE output_buf, JDIMENSION output_row,
; int num_rows);
;
%define img_width(b) (b) + 8 ; JDIMENSION img_width
%define input_buf(b) (b) + 12 ; JSAMPARRAY input_buf
%define output_buf(b) (b) + 16 ; JSAMPIMAGE output_buf
%define output_row(b) (b) + 20 ; JDIMENSION output_row
%define num_rows(b) (b) + 24 ; int num_rows
%define original_ebp ebp + 0
%define wk(i) ebp - (WK_NUM - (i)) * SIZEOF_YMMWORD
; ymmword wk[WK_NUM]
%define WK_NUM 2
%define gotptr wk(0) - SIZEOF_POINTER ; void * gotptr
align 32
GLOBAL_FUNCTION(jsimd_rgb_gray_convert_avx2)
EXTN(jsimd_rgb_gray_convert_avx2):
push ebp
mov eax, esp ; eax = original ebp
sub esp, byte 4
and esp, byte (-SIZEOF_YMMWORD) ; align to 256 bits
mov [esp], eax
mov ebp, esp ; ebp = aligned ebp
lea esp, [wk(0)]
pushpic eax ; make a room for GOT address
push ebx
; push ecx ; need not be preserved
; push edx ; need not be preserved
push esi
push edi
get_GOT ebx ; get GOT address
movpic POINTER [gotptr], ebx ; save GOT address
mov ecx, JDIMENSION [img_width(eax)]
test ecx, ecx
jz near .return
push ecx
mov esi, JSAMPIMAGE [output_buf(eax)]
mov ecx, JDIMENSION [output_row(eax)]
mov edi, JSAMPARRAY [esi+0*SIZEOF_JSAMPARRAY]
lea edi, [edi+ecx*SIZEOF_JSAMPROW]
pop ecx
mov esi, JSAMPARRAY [input_buf(eax)]
mov eax, INT [num_rows(eax)]
test eax, eax
jle near .return
alignx 16, 7
.rowloop:
pushpic eax
push edi
push esi
push ecx ; col
mov esi, JSAMPROW [esi] ; inptr
mov edi, JSAMPROW [edi] ; outptr0
movpic eax, POINTER [gotptr] ; load GOT address (eax)
cmp ecx, byte SIZEOF_YMMWORD
jae near .columnloop
alignx 16, 7
%if RGB_PIXELSIZE == 3 ; ---------------
.column_ld1:
push eax
push edx
lea ecx, [ecx+ecx*2] ; imul ecx,RGB_PIXELSIZE
test cl, SIZEOF_BYTE
jz short .column_ld2
sub ecx, byte SIZEOF_BYTE
movzx eax, BYTE [esi+ecx]
.column_ld2:
test cl, SIZEOF_WORD
jz short .column_ld4
sub ecx, byte SIZEOF_WORD
movzx edx, WORD [esi+ecx]
shl eax, WORD_BIT
or eax, edx
.column_ld4:
vmovd xmmA, eax
pop edx
pop eax
test cl, SIZEOF_DWORD
jz short .column_ld8
sub ecx, byte SIZEOF_DWORD
vmovd xmmF, XMM_DWORD [esi+ecx]
vpslldq xmmA, xmmA, SIZEOF_DWORD
vpor xmmA, xmmA, xmmF
.column_ld8:
test cl, SIZEOF_MMWORD
jz short .column_ld16
sub ecx, byte SIZEOF_MMWORD
vmovq xmmB, XMM_MMWORD [esi+ecx]
vpslldq xmmA, xmmA, SIZEOF_MMWORD
vpor xmmA, xmmA, xmmB
.column_ld16:
test cl, SIZEOF_XMMWORD
jz short .column_ld32
sub ecx, byte SIZEOF_XMMWORD
vmovdqu xmmB, XMM_MMWORD [esi+ecx]
vperm2i128 ymmA, ymmA, ymmA, 1
vpor ymmA, ymmB
.column_ld32:
test cl, SIZEOF_YMMWORD
jz short .column_ld64
sub ecx, byte SIZEOF_YMMWORD
vmovdqa ymmF, ymmA
vmovdqu ymmA, YMMWORD [esi+0*SIZEOF_YMMWORD]
.column_ld64:
test cl, 2*SIZEOF_YMMWORD
mov ecx, SIZEOF_YMMWORD
jz short .rgb_gray_cnv
vmovdqa ymmB, ymmA
vmovdqu ymmA, YMMWORD [esi+0*SIZEOF_YMMWORD]
vmovdqu ymmF, YMMWORD [esi+1*SIZEOF_YMMWORD]
jmp short .rgb_gray_cnv
alignx 16, 7
.columnloop:
vmovdqu ymmA, YMMWORD [esi+0*SIZEOF_YMMWORD]
vmovdqu ymmF, YMMWORD [esi+1*SIZEOF_YMMWORD]
vmovdqu ymmB, YMMWORD [esi+2*SIZEOF_YMMWORD]
.rgb_gray_cnv:
; ymmA=(00 10 20 01 11 21 02 12 22 03 13 23 04 14 24 05
; 15 25 06 16 26 07 17 27 08 18 28 09 19 29 0A 1A)
; ymmF=(2A 0B 1B 2B 0C 1C 2C 0D 1D 2D 0E 1E 2E 0F 1F 2F
; 0G 1G 2G 0H 1H 2H 0I 1I 2I 0J 1J 2J 0K 1K 2K 0L)
; ymmB=(1L 2L 0M 1M 2M 0N 1N 2N 0O 1O 2O 0P 1P 2P 0Q 1Q
; 2Q 0R 1R 2R 0S 1S 2S 0T 1T 2T 0U 1U 2U 0V 1V 2V)
vmovdqu ymmC, ymmA
vinserti128 ymmA, ymmF, xmmA, 0 ; ymmA=(00 10 20 01 11 21 02 12 22 03 13 23 04 14 24 05
; 0G 1G 2G 0H 1H 2H 0I 1I 2I 0J 1J 2J 0K 1K 2K 0L)
vinserti128 ymmC, ymmC, xmmB, 0 ; ymmC=(1L 2L 0M 1M 2M 0N 1N 2N 0O 1O 2O 0P 1P 2P 0Q 1Q
; 15 25 06 16 26 07 17 27 08 18 28 09 19 29 0A 1A)
vinserti128 ymmB, ymmB, xmmF, 0 ; ymmB=(2A 0B 1B 2B 0C 1C 2C 0D 1D 2D 0E 1E 2E 0F 1F 2F
; 2Q 0R 1R 2R 0S 1S 2S 0T 1T 2T 0U 1U 2U 0V 1V 2V)
vperm2i128 ymmF, ymmC, ymmC, 1 ; ymmF=(15 25 06 16 26 07 17 27 08 18 28 09 19 29 0A 1A
; 1L 2L 0M 1M 2M 0N 1N 2N 0O 1O 2O 0P 1P 2P 0Q 1Q)
vmovdqa ymmG, ymmA
vpslldq ymmA, ymmA, 8 ; ymmA=(-- -- -- -- -- -- -- -- 00 10 20 01 11 21 02 12
; 22 03 13 23 04 14 24 05 0G 1G 2G 0H 1H 2H 0I 1I)
vpsrldq ymmG, ymmG, 8 ; ymmG=(22 03 13 23 04 14 24 05 0G 1G 2G 0H 1H 2H 0I 1I
; 2I 0J 1J 2J 0K 1K 2K 0L -- -- -- -- -- -- -- --)
vpunpckhbw ymmA, ymmA, ymmF ; ymmA=(00 08 10 18 20 28 01 09 11 19 21 29 02 0A 12 1A
; 0G 0O 1G 1O 2G 2O 0H 0P 1H 1P 2H 2P 0I 0Q 1I 1Q)
vpslldq ymmF, ymmF, 8 ; ymmF=(-- -- -- -- -- -- -- -- 15 25 06 16 26 07 17 27
; 08 18 28 09 19 29 0A 1A 1L 2L 0M 1M 2M 0N 1N 2N)
vpunpcklbw ymmG, ymmG, ymmB ; ymmG=(22 2A 03 0B 13 1B 23 2B 04 0C 14 1C 24 2C 05 0D
; 2I 2Q 0J 0R 1J 1R 2J 2R 0K 0S 1K 1S 2K 2S 0L 0T)
vpunpckhbw ymmF, ymmF, ymmB ; ymmF=(15 1D 25 2D 06 0E 16 1E 26 2E 07 0F 17 1F 27 2F
; 1L 1T 2L 2T 0M 0U 1M 1U 2M 2U 0N 0V 1N 1V 2N 2V)
vmovdqa ymmD, ymmA
vpslldq ymmA, ymmA, 8 ; ymmA=(-- -- -- -- -- -- -- -- 00 08 10 18 20 28 01 09
; 11 19 21 29 02 0A 12 1A 0G 0O 1G 1O 2G 2O 0H 0P)
vpsrldq ymmD, ymmD, 8 ; ymmD=(11 19 21 29 02 0A 12 1A 0G 0O 1G 1O 2G 2O 0H 0P
; 1H 1P 2H 2P 0I 0Q 1I 1Q -- -- -- -- -- -- -- --)
vpunpckhbw ymmA, ymmA, ymmG ; ymmA=(00 04 08 0C 10 14 18 1C 20 24 28 2C 01 05 09 0D
; 0G 0K 0O 0S 1G 1K 1O 1S 2G 2K 2O 2S 0H 0L 0P 0T)
vpslldq ymmG, ymmG, 8 ; ymmG=(-- -- -- -- -- -- -- -- 22 2A 03 0B 13 1B 23 2B
; 04 0C 14 1C 24 2C 05 0D 2I 2Q 0J 0R 1J 1R 2J 2R)
vpunpcklbw ymmD, ymmD, ymmF ; ymmD=(11 15 19 1D 21 25 29 2D 02 06 0A 0E 12 16 1A 1E
; 1H 1L 1P 1T 2H 2L 2P 2T 0I 0M 0Q 0U 1I 1M 1Q 1U)
vpunpckhbw ymmG, ymmG, ymmF ; ymmG=(22 26 2A 2E 03 07 0B 0F 13 17 1B 1F 23 27 2B 2F
; 2I 2M 2Q 2U 0J 0N 0R 0V 1J 1N 1R 1V 2J 2N 2R 2V)
vmovdqa ymmE, ymmA
vpslldq ymmA, ymmA, 8 ; ymmA=(-- -- -- -- -- -- -- -- 00 04 08 0C 10 14 18 1C
; 20 24 28 2C 01 05 09 0D 0G 0K 0O 0S 1G 1K 1O 1S)
vpsrldq ymmE, ymmE, 8 ; ymmE=(20 24 28 2C 01 05 09 0D 0G 0K 0O 0S 1G 1K 1O 1S
; 2G 2K 2O 2S 0H 0L 0P 0T -- -- -- -- -- -- -- --)
vpunpckhbw ymmA, ymmA, ymmD ; ymmA=(00 02 04 06 08 0A 0C 0E 10 12 14 16 18 1A 1C 1E
; 0G 0I 0K 0M 0O 0Q 0S 0U 1G 1I 1K 1M 1O 1Q 1S 1U)
vpslldq ymmD, ymmD, 8 ; ymmD=(-- -- -- -- -- -- -- -- 11 15 19 1D 21 25 29 2D
; 02 06 0A 0E 12 16 1A 1E 1H 1L 1P 1T 2H 2L 2P 2T)
vpunpcklbw ymmE, ymmE, ymmG ; ymmE=(20 22 24 26 28 2A 2C 2E 01 03 05 07 09 0B 0D 0F
; 2G 2I 2K 2M 2O 2Q 2S 2U 0H 0J 0L 0N 0P 0R 0T 0V)
vpunpckhbw ymmD, ymmD, ymmG ; ymmD=(11 13 15 17 19 1B 1D 1F 21 23 25 27 29 2B 2D 2F
; 1H 1J 1L 1N 1P 1R 1T 1V 2H 2J 2L 2N 2P 2R 2T 2V)
vpxor ymmH, ymmH, ymmH
vmovdqa ymmC, ymmA
vpunpcklbw ymmA, ymmA, ymmH ; ymmA=(00 02 04 06 08 0A 0C 0E 0G 0I 0K 0M 0O 0Q 0S 0U)
vpunpckhbw ymmC, ymmC, ymmH ; ymmC=(10 12 14 16 18 1A 1C 1E 1G 1I 1K 1M 1O 1Q 1S 1U)
vmovdqa ymmB, ymmE
vpunpcklbw ymmE, ymmE, ymmH ; ymmE=(20 22 24 26 28 2A 2C 2E 2G 2I 2K 2M 2O 2Q 2S 2U)
vpunpckhbw ymmB, ymmB, ymmH ; ymmB=(01 03 05 07 09 0B 0D 0F 0H 0J 0L 0N 0P 0R 0T 0V)
vmovdqa ymmF, ymmD
vpunpcklbw ymmD, ymmD, ymmH ; ymmD=(11 13 15 17 19 1B 1D 1F 1H 1J 1L 1N 1P 1R 1T 1V)
vpunpckhbw ymmF, ymmF, ymmH ; ymmF=(21 23 25 27 29 2B 2D 2F 2H 2J 2L 2N 2P 2R 2T 2V)
%else ; RGB_PIXELSIZE == 4 ; -----------
.column_ld1:
test cl, SIZEOF_XMMWORD/16
jz short .column_ld2
sub ecx, byte SIZEOF_XMMWORD/16
vmovd xmmA, XMM_DWORD [esi+ecx*RGB_PIXELSIZE]
.column_ld2:
test cl, SIZEOF_XMMWORD/8
jz short .column_ld4
sub ecx, byte SIZEOF_XMMWORD/8
vmovq xmmF, XMM_MMWORD [esi+ecx*RGB_PIXELSIZE]
vpslldq xmmA, xmmA, SIZEOF_MMWORD
vpor xmmA, xmmA, xmmF
.column_ld4:
test cl, SIZEOF_XMMWORD/4
jz short .column_ld8
sub ecx, byte SIZEOF_XMMWORD/4
vmovdqa xmmF, xmmA
vperm2i128 ymmF, ymmF, ymmF, 1
vmovdqu xmmA, XMMWORD [esi+ecx*RGB_PIXELSIZE]
vpor ymmA, ymmA, ymmF
.column_ld8:
test cl, SIZEOF_XMMWORD/2
jz short .column_ld16
sub ecx, byte SIZEOF_XMMWORD/2
vmovdqa ymmF, ymmA
vmovdqu ymmA, YMMWORD [esi+ecx*RGB_PIXELSIZE]
.column_ld16:
test cl, SIZEOF_XMMWORD
mov ecx, SIZEOF_YMMWORD
jz short .rgb_gray_cnv
vmovdqa ymmE, ymmA
vmovdqa ymmH, ymmF
vmovdqu ymmA, YMMWORD [esi+0*SIZEOF_YMMWORD]
vmovdqu ymmF, YMMWORD [esi+1*SIZEOF_YMMWORD]
jmp short .rgb_gray_cnv
alignx 16, 7
.columnloop:
vmovdqu ymmA, YMMWORD [esi+0*SIZEOF_YMMWORD]
vmovdqu ymmF, YMMWORD [esi+1*SIZEOF_YMMWORD]
vmovdqu ymmE, YMMWORD [esi+2*SIZEOF_YMMWORD]
vmovdqu ymmH, YMMWORD [esi+3*SIZEOF_YMMWORD]
.rgb_gray_cnv:
; ymmA=(00 10 20 30 01 11 21 31 02 12 22 32 03 13 23 33
; 04 14 24 34 05 15 25 35 06 16 26 36 07 17 27 37)
; ymmF=(08 18 28 38 09 19 29 39 0A 1A 2A 3A 0B 1B 2B 3B
; 0C 1C 2C 3C 0D 1D 2D 3D 0E 1E 2E 3E 0F 1F 2F 3F)
; ymmE=(0G 1G 2G 3G 0H 1H 2H 3H 0I 1I 2I 3I 0J 1J 2J 3J
; 0K 1K 2K 3K 0L 1L 2L 3L 0M 1M 2M 3M 0N 1N 2N 3N)
; ymmH=(0O 1O 2O 3O 0P 1P 2P 3P 0Q 1Q 2Q 3Q 0R 1R 2R 3R
; 0S 1S 2S 3S 0T 1T 2T 3T 0U 1U 2U 3U 0V 1V 2V 3V)
vmovdqa ymmB, ymmA
vinserti128 ymmA, ymmA, xmmE, 1 ; ymmA=(00 10 20 30 01 11 21 31 02 12 22 32 03 13 23 33
; 0G 1G 2G 3G 0H 1H 2H 3H 0I 1I 2I 3I 0J 1J 2J 3J)
vperm2i128 ymmE, ymmB, ymmE, 0x31 ; ymmE=(04 14 24 34 05 15 25 35 06 16 26 36 07 17 27 37
; 0K 1K 2K 3K 0L 1L 2L 3L 0M 1M 2M 3M 0N 1N 2N 3N)
vmovdqa ymmB, ymmF
vinserti128 ymmF, ymmF, xmmH, 1 ; ymmF=(08 18 28 38 09 19 29 39 0A 1A 2A 3A 0B 1B 2B 3B
; 0O 1O 2O 3O 0P 1P 2P 3P 0Q 1Q 2Q 3Q 0R 1R 2R 3R)
vperm2i128 ymmH, ymmB, ymmH, 0x31 ; ymmH=(0C 1C 2C 3C 0D 1D 2D 3D 0E 1E 2E 3E 0F 1F 2F 3F
; 0S 1S 2S 3S 0T 1T 2T 3T 0U 1U 2U 3U 0V 1V 2V 3V)
vmovdqa ymmD, ymmA
vpunpcklbw ymmA, ymmA, ymmE ; ymmA=(00 04 10 14 20 24 30 34 01 05 11 15 21 25 31 35
; 0G 0K 1G 1K 2G 2K 3G 3K 0H 0L 1H 1L 2H 2L 3H 3L)
vpunpckhbw ymmD, ymmD, ymmE ; ymmD=(02 06 12 16 22 26 32 36 03 07 13 17 23 27 33 37
; 0I 0M 1I 1M 2I 2M 3I 3M 0J 0N 1J 1N 2J 2N 3J 3N)
vmovdqa ymmC, ymmF
vpunpcklbw ymmF, ymmF, ymmH ; ymmF=(08 0C 18 1C 28 2C 38 3C 09 0D 19 1D 29 2D 39 3D
; 0O 0S 1O 1S 2O 2S 3O 3S 0P 0T 1P 1T 2P 2T 3P 3T)
vpunpckhbw ymmC, ymmC, ymmH ; ymmC=(0A 0E 1A 1E 2A 2E 3A 3E 0B 0F 1B 1F 2B 2F 3B 3F
; 0Q 0U 1Q 1U 2Q 2U 3Q 3U 0R 0V 1R 1V 2R 2V 3R 3V)
vmovdqa ymmB, ymmA
vpunpcklwd ymmA, ymmA, ymmF ; ymmA=(00 04 08 0C 10 14 18 1C 20 24 28 2C 30 34 38 3C
; 0G 0K 0O 0S 1G 1K 1O 1S 2G 2K 2O 2S 3G 3K 3O 3S)
vpunpckhwd ymmB, ymmB, ymmF ; ymmB=(01 05 09 0D 11 15 19 1D 21 25 29 2D 31 35 39 3D
; 0H 0L 0P 0T 1H 1L 1P 1T 2H 2L 2P 2T 3H 3L 3P 3T)
vmovdqa ymmG, ymmD
vpunpcklwd ymmD, ymmD, ymmC ; ymmD=(02 06 0A 0E 12 16 1A 1E 22 26 2A 2E 32 36 3A 3E
; 0I 0M 0Q 0U 1I 1M 1Q 1U 2I 2M 2Q 2U 3I 3M 3Q 3U)
vpunpckhwd ymmG, ymmG, ymmC ; ymmG=(03 07 0B 0F 13 17 1B 1F 23 27 2B 2F 33 37 3B 3F
; 0J 0N 0R 0V 1J 1N 1R 1V 2J 2N 2R 2V 3J 3N 3R 3V)
vmovdqa ymmE, ymmA
vpunpcklbw ymmA, ymmA, ymmD ; ymmA=(00 02 04 06 08 0A 0C 0E 10 12 14 16 18 1A 1C 1E
; 0G 0I 0K 0M 0O 0Q 0S 0U 1G 1I 1K 1M 1O 1Q 1S 1U)
vpunpckhbw ymmE, ymmE, ymmD ; ymmE=(20 22 24 26 28 2A 2C 2E 30 32 34 36 38 3A 3C 3E
; 2G 2I 2K 2M 2O 2Q 2S 2U 3G 3I 3K 3M 3O 3Q 3S 3U)
vmovdqa ymmH, ymmB
vpunpcklbw ymmB, ymmB, ymmG ; ymmB=(01 03 05 07 09 0B 0D 0F 11 13 15 17 19 1B 1D 1F
; 0H 0J 0L 0N 0P 0R 0T 0V 1H 1J 1L 1N 1P 1R 1T 1V)
vpunpckhbw ymmH, ymmH, ymmG ; ymmH=(21 23 25 27 29 2B 2D 2F 31 33 35 37 39 3B 3D 3F
; 2H 2J 2L 2N 2P 2R 2T 2V 3H 3J 3L 3N 3P 3R 3T 3V)
vpxor ymmF, ymmF, ymmF
vmovdqa ymmC, ymmA
vpunpcklbw ymmA, ymmA, ymmF ; ymmA=(00 02 04 06 08 0A 0C 0E 0G 0I 0K 0M 0O 0Q 0S 0U)
vpunpckhbw ymmC, ymmC, ymmF ; ymmC=(10 12 14 16 18 1A 1C 1E 1G 1I 1K 1M 1O 1Q 1S 1U)
vmovdqa ymmD, ymmB
vpunpcklbw ymmB, ymmB, ymmF ; ymmB=(01 03 05 07 09 0B 0D 0F 0H 0J 0L 0N 0P 0R 0T 0V)
vpunpckhbw ymmD, ymmD, ymmF ; ymmD=(11 13 15 17 19 1B 1D 1F 1H 1J 1L 1N 1P 1R 1T 1V)
vmovdqa ymmG, ymmE
vpunpcklbw ymmE, ymmE, ymmF ; ymmE=(20 22 24 26 28 2A 2C 2E 2G 2I 2K 2M 2O 2Q 2S 2U)
vpunpckhbw ymmG, ymmG, ymmF ; ymmG=(30 32 34 36 38 3A 3C 3E 3G 3I 3K 3M 3O 3Q 3S 3U)
vpunpcklbw ymmF, ymmF, ymmH
vpunpckhbw ymmH, ymmH, ymmH
vpsrlw ymmF, ymmF, BYTE_BIT ; ymmF=(21 23 25 27 29 2B 2D 2F 2H 2J 2L 2N 2P 2R 2T 2V)
vpsrlw ymmH, ymmH, BYTE_BIT ; ymmH=(31 33 35 37 39 3B 3D 3F 3H 3J 3L 3N 3P 3R 3T 3V)
%endif ; RGB_PIXELSIZE ; ---------------
; ymm0=R(02468ACEGIKMOQSU)=RE, ymm2=G(02468ACEGIKMOQSU)=GE, ymm4=B(02468ACEGIKMOQSU)=BE
; ymm1=R(13579BDFHJLNPRTV)=RO, ymm3=G(13579BDFHJLNPRTV)=GO, ymm5=B(13579BDFHJLNPRTV)=BO
; (Original)
; Y = 0.29900 * R + 0.58700 * G + 0.11400 * B
;
; (This implementation)
; Y = 0.29900 * R + 0.33700 * G + 0.11400 * B + 0.25000 * G
vmovdqa ymm6, ymm1
vpunpcklwd ymm1, ymm1, ymm3
vpunpckhwd ymm6, ymm6, ymm3
vpmaddwd ymm1, ymm1, [GOTOFF(eax,PW_F0299_F0337)] ; ymm1=ROL*FIX(0.299)+GOL*FIX(0.337)
vpmaddwd ymm6, ymm6, [GOTOFF(eax,PW_F0299_F0337)] ; ymm6=ROH*FIX(0.299)+GOH*FIX(0.337)
vmovdqa ymm7, ymm6 ; ymm7=ROH*FIX(0.299)+GOH*FIX(0.337)
vmovdqa ymm6, ymm0
vpunpcklwd ymm0, ymm0, ymm2
vpunpckhwd ymm6, ymm6, ymm2
vpmaddwd ymm0, ymm0, [GOTOFF(eax,PW_F0299_F0337)] ; ymm0=REL*FIX(0.299)+GEL*FIX(0.337)
vpmaddwd ymm6, ymm6, [GOTOFF(eax,PW_F0299_F0337)] ; ymm6=REH*FIX(0.299)+GEH*FIX(0.337)
vmovdqa YMMWORD [wk(0)], ymm0 ; wk(0)=REL*FIX(0.299)+GEL*FIX(0.337)
vmovdqa YMMWORD [wk(1)], ymm6 ; wk(1)=REH*FIX(0.299)+GEH*FIX(0.337)
vmovdqa ymm0, ymm5 ; ymm0=BO
vmovdqa ymm6, ymm4 ; ymm6=BE
vmovdqa ymm4, ymm0
vpunpcklwd ymm0, ymm0, ymm3
vpunpckhwd ymm4, ymm4, ymm3
vpmaddwd ymm0, ymm0, [GOTOFF(eax,PW_F0114_F0250)] ; ymm0=BOL*FIX(0.114)+GOL*FIX(0.250)
vpmaddwd ymm4, ymm4, [GOTOFF(eax,PW_F0114_F0250)] ; ymm4=BOH*FIX(0.114)+GOH*FIX(0.250)
vmovdqa ymm3, [GOTOFF(eax,PD_ONEHALF)] ; ymm3=[PD_ONEHALF]
vpaddd ymm0, ymm0, ymm1
vpaddd ymm4, ymm4, ymm7
vpaddd ymm0, ymm0, ymm3
vpaddd ymm4, ymm4, ymm3
vpsrld ymm0, ymm0, SCALEBITS ; ymm0=YOL
vpsrld ymm4, ymm4, SCALEBITS ; ymm4=YOH
vpackssdw ymm0, ymm0, ymm4 ; ymm0=YO
vmovdqa ymm4, ymm6
vpunpcklwd ymm6, ymm6, ymm2
vpunpckhwd ymm4, ymm4, ymm2
vpmaddwd ymm6, ymm6, [GOTOFF(eax,PW_F0114_F0250)] ; ymm6=BEL*FIX(0.114)+GEL*FIX(0.250)
vpmaddwd ymm4, ymm4, [GOTOFF(eax,PW_F0114_F0250)] ; ymm4=BEH*FIX(0.114)+GEH*FIX(0.250)
vmovdqa ymm2, [GOTOFF(eax,PD_ONEHALF)] ; ymm2=[PD_ONEHALF]
vpaddd ymm6, ymm6, YMMWORD [wk(0)]
vpaddd ymm4, ymm4, YMMWORD [wk(1)]
vpaddd ymm6, ymm6, ymm2
vpaddd ymm4, ymm4, ymm2
vpsrld ymm6, ymm6, SCALEBITS ; ymm6=YEL
vpsrld ymm4, ymm4, SCALEBITS ; ymm4=YEH
vpackssdw ymm6, ymm6, ymm4 ; ymm6=YE
vpsllw ymm0, ymm0, BYTE_BIT
vpor ymm6, ymm6, ymm0 ; ymm6=Y
vmovdqu YMMWORD [edi], ymm6 ; Save Y
sub ecx, byte SIZEOF_YMMWORD
add esi, RGB_PIXELSIZE*SIZEOF_YMMWORD ; inptr
add edi, byte SIZEOF_YMMWORD ; outptr0
cmp ecx, byte SIZEOF_YMMWORD
jae near .columnloop
test ecx, ecx
jnz near .column_ld1
pop ecx ; col
pop esi
pop edi
poppic eax
add esi, byte SIZEOF_JSAMPROW ; input_buf
add edi, byte SIZEOF_JSAMPROW
dec eax ; num_rows
jg near .rowloop
.return:
vzeroupper
pop edi
pop esi
; pop edx ; need not be preserved
; pop ecx ; need not be preserved
pop ebx
mov esp, ebp ; esp <- aligned ebp
pop esp ; esp <- original ebp
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
Reference in New Issue View Git Blame Copy Permalink