image_pack 0.2.0

data/ext/image_pack/vendor/mozjpeg/simd/x86_64/jdcolext-avx2.asm

@@ -0,0 +1,496 @@
+;
+; jdcolext.asm - colorspace conversion (64-bit AVX2)
+;
+; Copyright 2009, 2012 Pierre Ossman <ossman@cendio.se> for Cendio AB
+; Copyright (C) 2009, 2012, 2016, D. R. Commander.
+; Copyright (C) 2015, Intel Corporation.
+; Copyright (C) 2018, Matthias Räncker.
+;
+; 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
+
+%include "jcolsamp.inc"
+
+; --------------------------------------------------------------------------
+;
+; Convert some rows of samples to the output colorspace.
+;
+; GLOBAL(void)
+; jsimd_ycc_rgb_convert_avx2(JDIMENSION out_width, JSAMPIMAGE input_buf,
+;                            JDIMENSION input_row, JSAMPARRAY output_buf,
+;                            int num_rows)
+;
+
+; r10d = JDIMENSION out_width
+; r11 = JSAMPIMAGE input_buf
+; r12d = JDIMENSION input_row
+; r13 = JSAMPARRAY output_buf
+; r14d = int num_rows
+
+%define wk(i)   rbp - (WK_NUM - (i)) * SIZEOF_YMMWORD  ; ymmword wk[WK_NUM]
+%define WK_NUM  2
+
+    align       32
+    GLOBAL_FUNCTION(jsimd_ycc_rgb_convert_avx2)
+
+EXTN(jsimd_ycc_rgb_convert_avx2):
+    push        rbp
+    mov         rax, rsp                     ; rax = original rbp
+    sub         rsp, byte 4
+    and         rsp, byte (-SIZEOF_YMMWORD)  ; align to 256 bits
+    mov         [rsp], rax
+    mov         rbp, rsp                     ; rbp = aligned rbp
+    lea         rsp, [wk(0)]
+    collect_args 5
+    push        rbx
+
+    mov         ecx, r10d               ; num_cols
+    test        rcx, rcx
+    jz          near .return
+
+    push        rcx
+
+    mov         rdi, r11
+    mov         ecx, r12d
+    mov         rsip, JSAMPARRAY [rdi+0*SIZEOF_JSAMPARRAY]
+    mov         rbxp, JSAMPARRAY [rdi+1*SIZEOF_JSAMPARRAY]
+    mov         rdxp, JSAMPARRAY [rdi+2*SIZEOF_JSAMPARRAY]
+    lea         rsi, [rsi+rcx*SIZEOF_JSAMPROW]
+    lea         rbx, [rbx+rcx*SIZEOF_JSAMPROW]
+    lea         rdx, [rdx+rcx*SIZEOF_JSAMPROW]
+
+    pop         rcx
+
+    mov         rdi, r13
+    mov         eax, r14d
+    test        rax, rax
+    jle         near .return
+.rowloop:
+    push        rax
+    push        rdi
+    push        rdx
+    push        rbx
+    push        rsi
+    push        rcx                     ; col
+
+    mov         rsip, JSAMPROW [rsi]    ; inptr0
+    mov         rbxp, JSAMPROW [rbx]    ; inptr1
+    mov         rdxp, JSAMPROW [rdx]    ; inptr2
+    mov         rdip, JSAMPROW [rdi]    ; outptr
+.columnloop:
+
+    vmovdqu     ymm5, YMMWORD [rbx]     ; ymm5=Cb(0123456789ABCDEFGHIJKLMNOPQRSTUV)
+    vmovdqu     ymm1, YMMWORD [rdx]     ; ymm1=Cr(0123456789ABCDEFGHIJKLMNOPQRSTUV)
+
+    vpcmpeqw    ymm0, ymm0, ymm0
+    vpcmpeqw    ymm7, ymm7, ymm7
+    vpsrlw      ymm0, ymm0, BYTE_BIT    ; ymm0={0xFF 0x00 0xFF 0x00 ..}
+    vpsllw      ymm7, ymm7, 7           ; ymm7={0xFF80 0xFF80 0xFF80 0xFF80 ..}
+
+    vpand       ymm4, ymm0, ymm5        ; ymm4=Cb(02468ACEGIKMOQSU)=CbE
+    vpsrlw      ymm5, ymm5, BYTE_BIT    ; ymm5=Cb(13579BDFHJLNPRTV)=CbO
+    vpand       ymm0, ymm0, ymm1        ; ymm0=Cr(02468ACEGIKMOQSU)=CrE
+    vpsrlw      ymm1, ymm1, BYTE_BIT    ; ymm1=Cr(13579BDFHJLNPRTV)=CrO
+
+    vpaddw      ymm2, ymm4, ymm7
+    vpaddw      ymm3, ymm5, ymm7
+    vpaddw      ymm6, ymm0, ymm7
+    vpaddw      ymm7, ymm1, ymm7
+
+    ; (Original)
+    ; R = Y                + 1.40200 * Cr
+    ; G = Y - 0.34414 * Cb - 0.71414 * Cr
+    ; B = Y + 1.77200 * Cb
+    ;
+    ; (This implementation)
+    ; R = Y                + 0.40200 * Cr + Cr
+    ; G = Y - 0.34414 * Cb + 0.28586 * Cr - Cr
+    ; B = Y - 0.22800 * Cb + Cb + Cb
+
+    vpaddw      ymm4, ymm2, ymm2             ; ymm4=2*CbE
+    vpaddw      ymm5, ymm3, ymm3             ; ymm5=2*CbO
+    vpaddw      ymm0, ymm6, ymm6             ; ymm0=2*CrE
+    vpaddw      ymm1, ymm7, ymm7             ; ymm1=2*CrO
+
+    vpmulhw     ymm4, ymm4, [rel PW_MF0228]  ; ymm4=(2*CbE * -FIX(0.22800))
+    vpmulhw     ymm5, ymm5, [rel PW_MF0228]  ; ymm5=(2*CbO * -FIX(0.22800))
+    vpmulhw     ymm0, ymm0, [rel PW_F0402]   ; ymm0=(2*CrE * FIX(0.40200))
+    vpmulhw     ymm1, ymm1, [rel PW_F0402]   ; ymm1=(2*CrO * FIX(0.40200))
+
+    vpaddw      ymm4, ymm4, [rel PW_ONE]
+    vpaddw      ymm5, ymm5, [rel PW_ONE]
+    vpsraw      ymm4, ymm4, 1                ; ymm4=(CbE * -FIX(0.22800))
+    vpsraw      ymm5, ymm5, 1                ; ymm5=(CbO * -FIX(0.22800))
+    vpaddw      ymm0, ymm0, [rel PW_ONE]
+    vpaddw      ymm1, ymm1, [rel PW_ONE]
+    vpsraw      ymm0, ymm0, 1                ; ymm0=(CrE * FIX(0.40200))
+    vpsraw      ymm1, ymm1, 1                ; ymm1=(CrO * FIX(0.40200))
+
+    vpaddw      ymm4, ymm4, ymm2
+    vpaddw      ymm5, ymm5, ymm3
+    vpaddw      ymm4, ymm4, ymm2             ; ymm4=(CbE * FIX(1.77200))=(B-Y)E
+    vpaddw      ymm5, ymm5, ymm3             ; ymm5=(CbO * FIX(1.77200))=(B-Y)O
+    vpaddw      ymm0, ymm0, ymm6             ; ymm0=(CrE * FIX(1.40200))=(R-Y)E
+    vpaddw      ymm1, ymm1, ymm7             ; ymm1=(CrO * FIX(1.40200))=(R-Y)O
+
+    vmovdqa     YMMWORD [wk(0)], ymm4        ; wk(0)=(B-Y)E
+    vmovdqa     YMMWORD [wk(1)], ymm5        ; wk(1)=(B-Y)O
+
+    vpunpckhwd  ymm4, ymm2, ymm6
+    vpunpcklwd  ymm2, ymm2, ymm6
+    vpmaddwd    ymm2, ymm2, [rel PW_MF0344_F0285]
+    vpmaddwd    ymm4, ymm4, [rel PW_MF0344_F0285]
+    vpunpckhwd  ymm5, ymm3, ymm7
+    vpunpcklwd  ymm3, ymm3, ymm7
+    vpmaddwd    ymm3, ymm3, [rel PW_MF0344_F0285]
+    vpmaddwd    ymm5, ymm5, [rel PW_MF0344_F0285]
+
+    vpaddd      ymm2, ymm2, [rel PD_ONEHALF]
+    vpaddd      ymm4, ymm4, [rel PD_ONEHALF]
+    vpsrad      ymm2, ymm2, SCALEBITS
+    vpsrad      ymm4, ymm4, SCALEBITS
+    vpaddd      ymm3, ymm3, [rel PD_ONEHALF]
+    vpaddd      ymm5, ymm5, [rel PD_ONEHALF]
+    vpsrad      ymm3, ymm3, SCALEBITS
+    vpsrad      ymm5, ymm5, SCALEBITS
+
+    vpackssdw   ymm2, ymm2, ymm4             ; ymm2=CbE*-FIX(0.344)+CrE*FIX(0.285)
+    vpackssdw   ymm3, ymm3, ymm5             ; ymm3=CbO*-FIX(0.344)+CrO*FIX(0.285)
+    vpsubw      ymm2, ymm2, ymm6             ; ymm2=CbE*-FIX(0.344)+CrE*-FIX(0.714)=(G-Y)E
+    vpsubw      ymm3, ymm3, ymm7             ; ymm3=CbO*-FIX(0.344)+CrO*-FIX(0.714)=(G-Y)O
+
+    vmovdqu     ymm5, YMMWORD [rsi]          ; ymm5=Y(0123456789ABCDEFGHIJKLMNOPQRSTUV)
+
+    vpcmpeqw    ymm4, ymm4, ymm4
+    vpsrlw      ymm4, ymm4, BYTE_BIT         ; ymm4={0xFF 0x00 0xFF 0x00 ..}
+    vpand       ymm4, ymm4, ymm5             ; ymm4=Y(02468ACEGIKMOQSU)=YE
+    vpsrlw      ymm5, ymm5, BYTE_BIT         ; ymm5=Y(13579BDFHJLNPRTV)=YO
+
+    vpaddw      ymm0, ymm0, ymm4             ; ymm0=((R-Y)E+YE)=RE=R(02468ACEGIKMOQSU)
+    vpaddw      ymm1, ymm1, ymm5             ; ymm1=((R-Y)O+YO)=RO=R(13579BDFHJLNPRTV)
+    vpackuswb   ymm0, ymm0, ymm0             ; ymm0=R(02468ACE********GIKMOQSU********)
+    vpackuswb   ymm1, ymm1, ymm1             ; ymm1=R(13579BDF********HJLNPRTV********)
+
+    vpaddw      ymm2, ymm2, ymm4             ; ymm2=((G-Y)E+YE)=GE=G(02468ACEGIKMOQSU)
+    vpaddw      ymm3, ymm3, ymm5             ; ymm3=((G-Y)O+YO)=GO=G(13579BDFHJLNPRTV)
+    vpackuswb   ymm2, ymm2, ymm2             ; ymm2=G(02468ACE********GIKMOQSU********)
+    vpackuswb   ymm3, ymm3, ymm3             ; ymm3=G(13579BDF********HJLNPRTV********)
+
+    vpaddw      ymm4, ymm4, YMMWORD [wk(0)]  ; ymm4=(YE+(B-Y)E)=BE=B(02468ACEGIKMOQSU)
+    vpaddw      ymm5, ymm5, YMMWORD [wk(1)]  ; ymm5=(YO+(B-Y)O)=BO=B(13579BDFHJLNPRTV)
+    vpackuswb   ymm4, ymm4, ymm4             ; ymm4=B(02468ACE********GIKMOQSU********)
+    vpackuswb   ymm5, ymm5, ymm5             ; ymm5=B(13579BDF********HJLNPRTV********)
+
+%if RGB_PIXELSIZE == 3  ; ---------------
+
+    ; ymmA=(00 02 04 06 08 0A 0C 0E ** 0G 0I 0K 0M 0O 0Q 0S 0U **)
+    ; ymmB=(01 03 05 07 09 0B 0D 0F ** 0H 0J 0L 0N 0P 0R 0T 0V **)
+    ; ymmC=(10 12 14 16 18 1A 1C 1E ** 1G 1I 1K 1M 1O 1Q 1S 1U **)
+    ; ymmD=(11 13 15 17 19 1B 1D 1F ** 1H 1J 1L 1N 1P 1R 1T 1V **)
+    ; ymmE=(20 22 24 26 28 2A 2C 2E ** 2G 2I 2K 2M 2O 2Q 2S 2U **)
+    ; ymmF=(21 23 25 27 29 2B 2D 2F ** 2H 2J 2L 2N 2P 2R 2T 2V **)
+    ; ymmG=(** ** ** ** ** ** ** ** ** ** ** ** ** ** ** ** ** **)
+    ; ymmH=(** ** ** ** ** ** ** ** ** ** ** ** ** ** ** ** ** **)
+
+    vpunpcklbw  ymmA, ymmA, ymmC        ; ymmA=(00 10 02 12 04 14 06 16 08 18 0A 1A 0C 1C 0E 1E
+                                        ;       0G 1G 0I 1I 0K 1K 0M 1M 0O 1O 0Q 1Q 0S 1S 0U 1U)
+    vpunpcklbw  ymmE, ymmE, ymmB        ; ymmE=(20 01 22 03 24 05 26 07 28 09 2A 0B 2C 0D 2E 0F
+                                        ;       2G 0H 2I 0J 2K 0L 2M 0N 2O 0P 2Q 0R 2S 0T 2U 0V)
+    vpunpcklbw  ymmD, ymmD, ymmF        ; ymmD=(11 21 13 23 15 25 17 27 19 29 1B 2B 1D 2D 1F 2F
+                                        ;       1H 2H 1J 2J 1L 2L 1N 2N 1P 2P 1R 2R 1T 2T 1V 2V)
+
+    vpsrldq     ymmH, ymmA, 2           ; ymmH=(02 12 04 14 06 16 08 18 0A 1A 0C 1C 0E 1E 0G 1G
+                                        ;       0I 1I 0K 1K 0M 1M 0O 1O 0Q 1Q 0S 1S 0U 1U -- --)
+    vpunpckhwd  ymmG, ymmA, ymmE        ; ymmG=(08 18 28 09 0A 1A 2A 0B 0C 1C 2C 0D 0E 1E 2E 0F
+                                        ;       0O 1O 2O 0P 0Q 1Q 2Q 0R 0S 1S 2S 0T 0U 1U 2U 0V)
+    vpunpcklwd  ymmA, ymmA, ymmE        ; ymmA=(00 10 20 01 02 12 22 03 04 14 24 05 06 16 26 07
+                                        ;       0G 1G 2G 0H 0I 1I 2I 0J 0K 1K 2K 0L 0M 1M 2M 0N)
+
+    vpsrldq     ymmE, ymmE, 2           ; ymmE=(22 03 24 05 26 07 28 09 2A 0B 2C 0D 2E 0F 2G 0H
+                                        ;       2I 0J 2K 0L 2M 0N 2O 0P 2Q 0R 2S 0T 2U 0V -- --)
+
+    vpsrldq     ymmB, ymmD, 2           ; ymmB=(13 23 15 25 17 27 19 29 1B 2B 1D 2D 1F 2F 1H 2H
+                                        ;       1J 2J 1L 2L 1N 2N 1P 2P 1R 2R 1T 2T 1V 2V -- --)
+    vpunpckhwd  ymmC, ymmD, ymmH        ; ymmC=(19 29 0A 1A 1B 2B 0C 1C 1D 2D 0E 1E 1F 2F 0G 1G
+                                        ;       1P 2P 0Q 1Q 1R 2R 0S 1S 1T 2T 0U 1U 1V 2V -- --)
+    vpunpcklwd  ymmD, ymmD, ymmH        ; ymmD=(11 21 02 12 13 23 04 14 15 25 06 16 17 27 08 18
+                                        ;       1H 2H 0I 1I 1J 2J 0K 1K 1L 2L 0M 1M 1N 2N 0O 1O)
+
+    vpunpckhwd  ymmF, ymmE, ymmB        ; ymmF=(2A 0B 1B 2B 2C 0D 1D 2D 2E 0F 1F 2F 2G 0H 1H 2H
+                                        ;       2Q 0R 1R 2R 2S 0T 1T 2T 2U 0V 1V 2V -- -- -- --)
+    vpunpcklwd  ymmE, ymmE, ymmB        ; ymmE=(22 03 13 23 24 05 15 25 26 07 17 27 28 09 19 29
+                                        ;       2I 0J 1J 2J 2K 0L 1L 2L 2M 0N 1N 2N 2O 0P 1P 2P)
+
+    vpshufd     ymmH, ymmA, 0x4E        ; ymmH=(04 14 24 05 06 16 26 07 00 10 20 01 02 12 22 03
+                                        ;       0K 1K 2K 0L 0M 1M 2M 0N 0G 1G 2G 0H 0I 1I 2I 0J)
+    vpunpckldq  ymmA, ymmA, ymmD        ; ymmA=(00 10 20 01 11 21 02 12 02 12 22 03 13 23 04 14
+                                        ;       0G 1G 2G 0H 1H 2H 0I 1I 0I 1I 2I 0J 1J 2J 0K 1K)
+    vpunpckhdq  ymmD, ymmD, ymmE        ; ymmD=(15 25 06 16 26 07 17 27 17 27 08 18 28 09 19 29
+                                        ;       1L 2L 0M 1M 2M 0N 1N 2N 1N 2N 0O 1O 2O 0P 1P 2P)
+    vpunpckldq  ymmE, ymmE, ymmH        ; ymmE=(22 03 13 23 04 14 24 05 24 05 15 25 06 16 26 07
+                                        ;       2I 0J 1J 2J 0K 1K 2K 0L 2K 0L 1L 2L 0M 1M 2M 0N)
+
+    vpshufd     ymmH, ymmG, 0x4E        ; ymmH=(0C 1C 2C 0D 0E 1E 2E 0F 08 18 28 09 0A 1A 2A 0B
+                                        ;       0S 1S 2S 0T 0U 1U 2U 0V 0O 1O 2O 0P 0Q 1Q 2Q 0R)
+    vpunpckldq  ymmG, ymmG, ymmC        ; ymmG=(08 18 28 09 19 29 0A 1A 0A 1A 2A 0B 1B 2B 0C 1C
+                                        ;       0O 1O 2O 0P 1P 2P 0Q 1Q 0Q 1Q 2Q 0R 1R 2R 0S 1S)
+    vpunpckhdq  ymmC, ymmC, ymmF        ; ymmC=(1D 2D 0E 1E 2E 0F 1F 2F 1F 2F 0G 1G 2G 0H 1H 2H
+                                        ;       1T 2T 0U 1U 2U 0V 1V 2V 1V 2V -- -- -- -- -- --)
+    vpunpckldq  ymmF, ymmF, ymmH        ; ymmF=(2A 0B 1B 2B 0C 1C 2C 0D 2C 0D 1D 2D 0E 1E 2E 0F
+                                        ;       2Q 0R 1R 2R 0S 1S 2S 0T 2S 0T 1T 2T 0U 1U 2U 0V)
+
+    vpunpcklqdq ymmH, ymmA, ymmE        ; ymmH=(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)
+    vpunpcklqdq ymmG, ymmD, ymmG        ; ymmG=(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)
+    vpunpcklqdq ymmC, ymmF, ymmC        ; ymmC=(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  ymmA, ymmH, ymmG, 0x20  ; 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)
+    vperm2i128  ymmD, ymmC, ymmH, 0x30  ; ymmD=(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)
+    vperm2i128  ymmF, ymmG, ymmC, 0x31  ; ymmF=(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)
+
+    cmp         rcx, byte SIZEOF_YMMWORD
+    jb          short .column_st64
+
+    test        rdi, SIZEOF_YMMWORD-1
+    jnz         short .out1
+    ; --(aligned)-------------------
+    vmovntdq    YMMWORD [rdi+0*SIZEOF_YMMWORD], ymmA
+    vmovntdq    YMMWORD [rdi+1*SIZEOF_YMMWORD], ymmD
+    vmovntdq    YMMWORD [rdi+2*SIZEOF_YMMWORD], ymmF
+    jmp         short .out0
+.out1:  ; --(unaligned)-----------------
+    vmovdqu     YMMWORD [rdi+0*SIZEOF_YMMWORD], ymmA
+    vmovdqu     YMMWORD [rdi+1*SIZEOF_YMMWORD], ymmD
+    vmovdqu     YMMWORD [rdi+2*SIZEOF_YMMWORD], ymmF
+.out0:
+    add         rdi, byte RGB_PIXELSIZE*SIZEOF_YMMWORD  ; outptr
+    sub         rcx, byte SIZEOF_YMMWORD
+    jz          near .nextrow
+
+    add         rsi, byte SIZEOF_YMMWORD  ; inptr0
+    add         rbx, byte SIZEOF_YMMWORD  ; inptr1
+    add         rdx, byte SIZEOF_YMMWORD  ; inptr2
+    jmp         near .columnloop
+
+.column_st64:
+    lea         rcx, [rcx+rcx*2]            ; imul ecx, RGB_PIXELSIZE
+    cmp         rcx, byte 2*SIZEOF_YMMWORD
+    jb          short .column_st32
+    vmovdqu     YMMWORD [rdi+0*SIZEOF_YMMWORD], ymmA
+    vmovdqu     YMMWORD [rdi+1*SIZEOF_YMMWORD], ymmD
+    add         rdi, byte 2*SIZEOF_YMMWORD  ; outptr
+    vmovdqa     ymmA, ymmF
+    sub         rcx, byte 2*SIZEOF_YMMWORD
+    jmp         short .column_st31
+.column_st32:
+    cmp         rcx, byte SIZEOF_YMMWORD
+    jb          short .column_st31
+    vmovdqu     YMMWORD [rdi+0*SIZEOF_YMMWORD], ymmA
+    add         rdi, byte SIZEOF_YMMWORD    ; outptr
+    vmovdqa     ymmA, ymmD
+    sub         rcx, byte SIZEOF_YMMWORD
+    jmp         short .column_st31
+.column_st31:
+    cmp         rcx, byte SIZEOF_XMMWORD
+    jb          short .column_st15
+    vmovdqu     XMMWORD [rdi+0*SIZEOF_XMMWORD], xmmA
+    add         rdi, byte SIZEOF_XMMWORD    ; outptr
+    vperm2i128  ymmA, ymmA, ymmA, 1
+    sub         rcx, byte SIZEOF_XMMWORD
+.column_st15:
+    ; Store the lower 8 bytes of xmmA to the output when it has enough
+    ; space.
+    cmp         rcx, byte SIZEOF_MMWORD
+    jb          short .column_st7
+    vmovq       XMM_MMWORD [rdi], xmmA
+    add         rdi, byte SIZEOF_MMWORD
+    sub         rcx, byte SIZEOF_MMWORD
+    vpsrldq     xmmA, xmmA, SIZEOF_MMWORD
+.column_st7:
+    ; Store the lower 4 bytes of xmmA to the output when it has enough
+    ; space.
+    cmp         rcx, byte SIZEOF_DWORD
+    jb          short .column_st3
+    vmovd       XMM_DWORD [rdi], xmmA
+    add         rdi, byte SIZEOF_DWORD
+    sub         rcx, byte SIZEOF_DWORD
+    vpsrldq     xmmA, xmmA, SIZEOF_DWORD
+.column_st3:
+    ; Store the lower 2 bytes of rax to the output when it has enough
+    ; space.
+    vmovd       eax, xmmA
+    cmp         rcx, byte SIZEOF_WORD
+    jb          short .column_st1
+    mov         word [rdi], ax
+    add         rdi, byte SIZEOF_WORD
+    sub         rcx, byte SIZEOF_WORD
+    shr         rax, 16
+.column_st1:
+    ; Store the lower 1 byte of rax to the output when it has enough
+    ; space.
+    test        rcx, rcx
+    jz          short .nextrow
+    mov         byte [rdi], al
+
+%else  ; RGB_PIXELSIZE == 4 ; -----------
+
+%ifdef RGBX_FILLER_0XFF
+    vpcmpeqb    ymm6, ymm6, ymm6        ; ymm6=XE=X(02468ACE********GIKMOQSU********)
+    vpcmpeqb    ymm7, ymm7, ymm7        ; ymm7=XO=X(13579BDF********HJLNPRTV********)
+%else
+    vpxor       ymm6, ymm6, ymm6        ; ymm6=XE=X(02468ACE********GIKMOQSU********)
+    vpxor       ymm7, ymm7, ymm7        ; ymm7=XO=X(13579BDF********HJLNPRTV********)
+%endif
+    ; ymmA=(00 02 04 06 08 0A 0C 0E ** 0G 0I 0K 0M 0O 0Q 0S 0U **)
+    ; ymmB=(01 03 05 07 09 0B 0D 0F ** 0H 0J 0L 0N 0P 0R 0T 0V **)
+    ; ymmC=(10 12 14 16 18 1A 1C 1E ** 1G 1I 1K 1M 1O 1Q 1S 1U **)
+    ; ymmD=(11 13 15 17 19 1B 1D 1F ** 1H 1J 1L 1N 1P 1R 1T 1V **)
+    ; ymmE=(20 22 24 26 28 2A 2C 2E ** 2G 2I 2K 2M 2O 2Q 2S 2U **)
+    ; ymmF=(21 23 25 27 29 2B 2D 2F ** 2H 2J 2L 2N 2P 2R 2T 2V **)
+    ; ymmG=(30 32 34 36 38 3A 3C 3E ** 3G 3I 3K 3M 3O 3Q 3S 3U **)
+    ; ymmH=(31 33 35 37 39 3B 3D 3F ** 3H 3J 3L 3N 3P 3R 3T 3V **)
+
+    vpunpcklbw  ymmA, ymmA, ymmC        ; ymmA=(00 10 02 12 04 14 06 16 08 18 0A 1A 0C 1C 0E 1E
+                                        ;       0G 1G 0I 1I 0K 1K 0M 1M 0O 1O 0Q 1Q 0S 1S 0U 1U)
+    vpunpcklbw  ymmE, ymmE, ymmG        ; ymmE=(20 30 22 32 24 34 26 36 28 38 2A 3A 2C 3C 2E 3E
+                                        ;       2G 3G 2I 3I 2K 3K 2M 3M 2O 3O 2Q 3Q 2S 3S 2U 3U)
+    vpunpcklbw  ymmB, ymmB, ymmD        ; ymmB=(01 11 03 13 05 15 07 17 09 19 0B 1B 0D 1D 0F 1F
+                                        ;       0H 1H 0J 1J 0L 1L 0N 1N 0P 1P 0R 1R 0T 1T 0V 1V)
+    vpunpcklbw  ymmF, ymmF, ymmH        ; ymmF=(21 31 23 33 25 35 27 37 29 39 2B 3B 2D 3D 2F 3F
+                                        ;       2H 3H 2J 3J 2L 3L 2N 3N 2P 3P 2R 3R 2T 3T 2V 3V)
+
+    vpunpckhwd  ymmC, ymmA, ymmE        ; ymmC=(08 18 28 38 0A 1A 2A 3A 0C 1C 2C 3C 0E 1E 2E 3E
+                                        ;       0O 1O 2O 3O 0Q 1Q 2Q 3Q 0S 1S 2S 3S 0U 1U 2U 3U)
+    vpunpcklwd  ymmA, ymmA, ymmE        ; ymmA=(00 10 20 30 02 12 22 32 04 14 24 34 06 16 26 36
+                                        ;       0G 1G 2G 3G 0I 1I 2I 3I 0K 1K 2K 3K 0M 1M 2M 3M)
+    vpunpckhwd  ymmG, ymmB, ymmF        ; ymmG=(09 19 29 39 0B 1B 2B 3B 0D 1D 2D 3D 0F 1F 2F 3F
+                                        ;       0P 1P 2P 3P 0R 1R 2R 3R 0T 1T 2T 3T 0V 1V 2V 3V)
+    vpunpcklwd  ymmB, ymmB, ymmF        ; ymmB=(01 11 21 31 03 13 23 33 05 15 25 35 07 17 27 37
+                                        ;       0H 1H 2H 3H 0J 1J 2J 3J 0L 1L 2L 3L 0N 1N 2N 3N)
+
+    vpunpckhdq  ymmE, ymmA, ymmB        ; 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)
+    vpunpckldq  ymmB, ymmA, ymmB        ; ymmB=(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)
+    vpunpckhdq  ymmF, ymmC, ymmG        ; ymmF=(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)
+    vpunpckldq  ymmG, ymmC, ymmG        ; ymmG=(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  ymmA, ymmB, ymmE, 0x20  ; 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)
+    vperm2i128  ymmD, ymmG, ymmF, 0x20  ; ymmD=(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)
+    vperm2i128  ymmC, ymmB, ymmE, 0x31  ; ymmC=(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)
+    vperm2i128  ymmH, ymmG, ymmF, 0x31  ; 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)
+
+    cmp         rcx, byte SIZEOF_YMMWORD
+    jb          short .column_st64
+
+    test        rdi, SIZEOF_YMMWORD-1
+    jnz         short .out1
+    ; --(aligned)-------------------
+    vmovntdq    YMMWORD [rdi+0*SIZEOF_YMMWORD], ymmA
+    vmovntdq    YMMWORD [rdi+1*SIZEOF_YMMWORD], ymmD
+    vmovntdq    YMMWORD [rdi+2*SIZEOF_YMMWORD], ymmC
+    vmovntdq    YMMWORD [rdi+3*SIZEOF_YMMWORD], ymmH
+    jmp         short .out0
+.out1:  ; --(unaligned)-----------------
+    vmovdqu     YMMWORD [rdi+0*SIZEOF_YMMWORD], ymmA
+    vmovdqu     YMMWORD [rdi+1*SIZEOF_YMMWORD], ymmD
+    vmovdqu     YMMWORD [rdi+2*SIZEOF_YMMWORD], ymmC
+    vmovdqu     YMMWORD [rdi+3*SIZEOF_YMMWORD], ymmH
+.out0:
+    add         rdi, RGB_PIXELSIZE*SIZEOF_YMMWORD  ; outptr
+    sub         rcx, byte SIZEOF_YMMWORD
+    jz          near .nextrow
+
+    add         rsi, byte SIZEOF_YMMWORD  ; inptr0
+    add         rbx, byte SIZEOF_YMMWORD  ; inptr1
+    add         rdx, byte SIZEOF_YMMWORD  ; inptr2
+    jmp         near .columnloop
+
+.column_st64:
+    cmp         rcx, byte SIZEOF_YMMWORD/2
+    jb          short .column_st32
+    vmovdqu     YMMWORD [rdi+0*SIZEOF_YMMWORD], ymmA
+    vmovdqu     YMMWORD [rdi+1*SIZEOF_YMMWORD], ymmD
+    add         rdi, byte 2*SIZEOF_YMMWORD  ; outptr
+    vmovdqa     ymmA, ymmC
+    vmovdqa     ymmD, ymmH
+    sub         rcx, byte SIZEOF_YMMWORD/2
+.column_st32:
+    cmp         rcx, byte SIZEOF_YMMWORD/4
+    jb          short .column_st16
+    vmovdqu     YMMWORD [rdi+0*SIZEOF_YMMWORD], ymmA
+    add         rdi, byte SIZEOF_YMMWORD    ; outptr
+    vmovdqa     ymmA, ymmD
+    sub         rcx, byte SIZEOF_YMMWORD/4
+.column_st16:
+    cmp         rcx, byte SIZEOF_YMMWORD/8
+    jb          short .column_st15
+    vmovdqu     XMMWORD [rdi+0*SIZEOF_XMMWORD], xmmA
+    vperm2i128  ymmA, ymmA, ymmA, 1
+    add         rdi, byte SIZEOF_XMMWORD    ; outptr
+    sub         rcx, byte SIZEOF_YMMWORD/8
+.column_st15:
+    ; Store two pixels (8 bytes) of ymmA to the output when it has enough
+    ; space.
+    cmp         rcx, byte SIZEOF_YMMWORD/16
+    jb          short .column_st7
+    vmovq       MMWORD [rdi], xmmA
+    add         rdi, byte SIZEOF_YMMWORD/16*4
+    sub         rcx, byte SIZEOF_YMMWORD/16
+    vpsrldq     xmmA, SIZEOF_YMMWORD/16*4
+.column_st7:
+    ; Store one pixel (4 bytes) of ymmA to the output when it has enough
+    ; space.
+    test        rcx, rcx
+    jz          short .nextrow
+    vmovd       XMM_DWORD [rdi], xmmA
+
+%endif  ; RGB_PIXELSIZE ; ---------------
+
+.nextrow:
+    pop         rcx
+    pop         rsi
+    pop         rbx
+    pop         rdx
+    pop         rdi
+    pop         rax
+
+    add         rsi, byte SIZEOF_JSAMPROW
+    add         rbx, byte SIZEOF_JSAMPROW
+    add         rdx, byte SIZEOF_JSAMPROW
+    add         rdi, byte SIZEOF_JSAMPROW  ; output_buf
+    dec         rax                        ; num_rows
+    jg          near .rowloop
+
+    sfence                              ; flush the write buffer
+
+.return:
+    pop         rbx
+    vzeroupper
+    uncollect_args 5
+    mov         rsp, rbp                ; rsp <- aligned rbp
+    pop         rsp                     ; rsp <- original rbp
+    pop         rbp
+    ret
+
+; For some reason, the OS X linker does not honor the request to align the
+; segment unless we do this.
+    align       32