michael@0: ; michael@0: ; jiss2int-64.asm - accurate integer IDCT (64-bit SSE2) michael@0: ; michael@0: ; Copyright 2009 Pierre Ossman for Cendio AB michael@0: ; Copyright 2009 D. R. Commander michael@0: ; michael@0: ; Based on michael@0: ; x86 SIMD extension for IJG JPEG library michael@0: ; Copyright (C) 1999-2006, MIYASAKA Masaru. michael@0: ; For conditions of distribution and use, see copyright notice in jsimdext.inc michael@0: ; michael@0: ; This file should be assembled with NASM (Netwide Assembler), michael@0: ; can *not* be assembled with Microsoft's MASM or any compatible michael@0: ; assembler (including Borland's Turbo Assembler). michael@0: ; NASM is available from http://nasm.sourceforge.net/ or michael@0: ; http://sourceforge.net/project/showfiles.php?group_id=6208 michael@0: ; michael@0: ; This file contains a slow-but-accurate integer implementation of the michael@0: ; inverse DCT (Discrete Cosine Transform). The following code is based michael@0: ; directly on the IJG's original jidctint.c; see the jidctint.c for michael@0: ; more details. michael@0: ; michael@0: ; [TAB8] michael@0: michael@0: %include "jsimdext.inc" michael@0: %include "jdct.inc" michael@0: michael@0: ; -------------------------------------------------------------------------- michael@0: michael@0: %define CONST_BITS 13 michael@0: %define PASS1_BITS 2 michael@0: michael@0: %define DESCALE_P1 (CONST_BITS-PASS1_BITS) michael@0: %define DESCALE_P2 (CONST_BITS+PASS1_BITS+3) michael@0: michael@0: %if CONST_BITS == 13 michael@0: F_0_298 equ 2446 ; FIX(0.298631336) michael@0: F_0_390 equ 3196 ; FIX(0.390180644) michael@0: F_0_541 equ 4433 ; FIX(0.541196100) michael@0: F_0_765 equ 6270 ; FIX(0.765366865) michael@0: F_0_899 equ 7373 ; FIX(0.899976223) michael@0: F_1_175 equ 9633 ; FIX(1.175875602) michael@0: F_1_501 equ 12299 ; FIX(1.501321110) michael@0: F_1_847 equ 15137 ; FIX(1.847759065) michael@0: F_1_961 equ 16069 ; FIX(1.961570560) michael@0: F_2_053 equ 16819 ; FIX(2.053119869) michael@0: F_2_562 equ 20995 ; FIX(2.562915447) michael@0: F_3_072 equ 25172 ; FIX(3.072711026) michael@0: %else michael@0: ; NASM cannot do compile-time arithmetic on floating-point constants. michael@0: %define DESCALE(x,n) (((x)+(1<<((n)-1)))>>(n)) michael@0: F_0_298 equ DESCALE( 320652955,30-CONST_BITS) ; FIX(0.298631336) michael@0: F_0_390 equ DESCALE( 418953276,30-CONST_BITS) ; FIX(0.390180644) michael@0: F_0_541 equ DESCALE( 581104887,30-CONST_BITS) ; FIX(0.541196100) michael@0: F_0_765 equ DESCALE( 821806413,30-CONST_BITS) ; FIX(0.765366865) michael@0: F_0_899 equ DESCALE( 966342111,30-CONST_BITS) ; FIX(0.899976223) michael@0: F_1_175 equ DESCALE(1262586813,30-CONST_BITS) ; FIX(1.175875602) michael@0: F_1_501 equ DESCALE(1612031267,30-CONST_BITS) ; FIX(1.501321110) michael@0: F_1_847 equ DESCALE(1984016188,30-CONST_BITS) ; FIX(1.847759065) michael@0: F_1_961 equ DESCALE(2106220350,30-CONST_BITS) ; FIX(1.961570560) michael@0: F_2_053 equ DESCALE(2204520673,30-CONST_BITS) ; FIX(2.053119869) michael@0: F_2_562 equ DESCALE(2751909506,30-CONST_BITS) ; FIX(2.562915447) michael@0: F_3_072 equ DESCALE(3299298341,30-CONST_BITS) ; FIX(3.072711026) michael@0: %endif michael@0: michael@0: ; -------------------------------------------------------------------------- michael@0: SECTION SEG_CONST michael@0: michael@0: alignz 16 michael@0: global EXTN(jconst_idct_islow_sse2) michael@0: michael@0: EXTN(jconst_idct_islow_sse2): michael@0: michael@0: PW_F130_F054 times 4 dw (F_0_541+F_0_765), F_0_541 michael@0: PW_F054_MF130 times 4 dw F_0_541, (F_0_541-F_1_847) michael@0: PW_MF078_F117 times 4 dw (F_1_175-F_1_961), F_1_175 michael@0: PW_F117_F078 times 4 dw F_1_175, (F_1_175-F_0_390) michael@0: PW_MF060_MF089 times 4 dw (F_0_298-F_0_899),-F_0_899 michael@0: PW_MF089_F060 times 4 dw -F_0_899, (F_1_501-F_0_899) michael@0: PW_MF050_MF256 times 4 dw (F_2_053-F_2_562),-F_2_562 michael@0: PW_MF256_F050 times 4 dw -F_2_562, (F_3_072-F_2_562) michael@0: PD_DESCALE_P1 times 4 dd 1 << (DESCALE_P1-1) michael@0: PD_DESCALE_P2 times 4 dd 1 << (DESCALE_P2-1) michael@0: PB_CENTERJSAMP times 16 db CENTERJSAMPLE michael@0: michael@0: alignz 16 michael@0: michael@0: ; -------------------------------------------------------------------------- michael@0: SECTION SEG_TEXT michael@0: BITS 64 michael@0: ; michael@0: ; Perform dequantization and inverse DCT on one block of coefficients. michael@0: ; michael@0: ; GLOBAL(void) michael@0: ; jsimd_idct_islow_sse2 (void * dct_table, JCOEFPTR coef_block, michael@0: ; JSAMPARRAY output_buf, JDIMENSION output_col) michael@0: ; michael@0: michael@0: ; r10 = jpeg_component_info * compptr michael@0: ; r11 = JCOEFPTR coef_block michael@0: ; r12 = JSAMPARRAY output_buf michael@0: ; r13 = JDIMENSION output_col michael@0: michael@0: %define original_rbp rbp+0 michael@0: %define wk(i) rbp-(WK_NUM-(i))*SIZEOF_XMMWORD ; xmmword wk[WK_NUM] michael@0: %define WK_NUM 12 michael@0: michael@0: align 16 michael@0: global EXTN(jsimd_idct_islow_sse2) michael@0: michael@0: EXTN(jsimd_idct_islow_sse2): michael@0: push rbp michael@0: mov rax,rsp ; rax = original rbp michael@0: sub rsp, byte 4 michael@0: and rsp, byte (-SIZEOF_XMMWORD) ; align to 128 bits michael@0: mov [rsp],rax michael@0: mov rbp,rsp ; rbp = aligned rbp michael@0: lea rsp, [wk(0)] michael@0: collect_args michael@0: michael@0: ; ---- Pass 1: process columns from input. michael@0: michael@0: mov rdx, r10 ; quantptr michael@0: mov rsi, r11 ; inptr michael@0: michael@0: %ifndef NO_ZERO_COLUMN_TEST_ISLOW_SSE2 michael@0: mov eax, DWORD [DWBLOCK(1,0,rsi,SIZEOF_JCOEF)] michael@0: or eax, DWORD [DWBLOCK(2,0,rsi,SIZEOF_JCOEF)] michael@0: jnz near .columnDCT michael@0: michael@0: movdqa xmm0, XMMWORD [XMMBLOCK(1,0,rsi,SIZEOF_JCOEF)] michael@0: movdqa xmm1, XMMWORD [XMMBLOCK(2,0,rsi,SIZEOF_JCOEF)] michael@0: por xmm0, XMMWORD [XMMBLOCK(3,0,rsi,SIZEOF_JCOEF)] michael@0: por xmm1, XMMWORD [XMMBLOCK(4,0,rsi,SIZEOF_JCOEF)] michael@0: por xmm0, XMMWORD [XMMBLOCK(5,0,rsi,SIZEOF_JCOEF)] michael@0: por xmm1, XMMWORD [XMMBLOCK(6,0,rsi,SIZEOF_JCOEF)] michael@0: por xmm0, XMMWORD [XMMBLOCK(7,0,rsi,SIZEOF_JCOEF)] michael@0: por xmm1,xmm0 michael@0: packsswb xmm1,xmm1 michael@0: packsswb xmm1,xmm1 michael@0: movd eax,xmm1 michael@0: test rax,rax michael@0: jnz short .columnDCT michael@0: michael@0: ; -- AC terms all zero michael@0: michael@0: movdqa xmm5, XMMWORD [XMMBLOCK(0,0,rsi,SIZEOF_JCOEF)] michael@0: pmullw xmm5, XMMWORD [XMMBLOCK(0,0,rdx,SIZEOF_ISLOW_MULT_TYPE)] michael@0: michael@0: psllw xmm5,PASS1_BITS michael@0: michael@0: movdqa xmm4,xmm5 ; xmm5=in0=(00 01 02 03 04 05 06 07) michael@0: punpcklwd xmm5,xmm5 ; xmm5=(00 00 01 01 02 02 03 03) michael@0: punpckhwd xmm4,xmm4 ; xmm4=(04 04 05 05 06 06 07 07) michael@0: michael@0: pshufd xmm7,xmm5,0x00 ; xmm7=col0=(00 00 00 00 00 00 00 00) michael@0: pshufd xmm6,xmm5,0x55 ; xmm6=col1=(01 01 01 01 01 01 01 01) michael@0: pshufd xmm1,xmm5,0xAA ; xmm1=col2=(02 02 02 02 02 02 02 02) michael@0: pshufd xmm5,xmm5,0xFF ; xmm5=col3=(03 03 03 03 03 03 03 03) michael@0: pshufd xmm0,xmm4,0x00 ; xmm0=col4=(04 04 04 04 04 04 04 04) michael@0: pshufd xmm3,xmm4,0x55 ; xmm3=col5=(05 05 05 05 05 05 05 05) michael@0: pshufd xmm2,xmm4,0xAA ; xmm2=col6=(06 06 06 06 06 06 06 06) michael@0: pshufd xmm4,xmm4,0xFF ; xmm4=col7=(07 07 07 07 07 07 07 07) michael@0: michael@0: movdqa XMMWORD [wk(8)], xmm6 ; wk(8)=col1 michael@0: movdqa XMMWORD [wk(9)], xmm5 ; wk(9)=col3 michael@0: movdqa XMMWORD [wk(10)], xmm3 ; wk(10)=col5 michael@0: movdqa XMMWORD [wk(11)], xmm4 ; wk(11)=col7 michael@0: jmp near .column_end michael@0: %endif michael@0: .columnDCT: michael@0: michael@0: ; -- Even part michael@0: michael@0: movdqa xmm0, XMMWORD [XMMBLOCK(0,0,rsi,SIZEOF_JCOEF)] michael@0: movdqa xmm1, XMMWORD [XMMBLOCK(2,0,rsi,SIZEOF_JCOEF)] michael@0: pmullw xmm0, XMMWORD [XMMBLOCK(0,0,rdx,SIZEOF_ISLOW_MULT_TYPE)] michael@0: pmullw xmm1, XMMWORD [XMMBLOCK(2,0,rdx,SIZEOF_ISLOW_MULT_TYPE)] michael@0: movdqa xmm2, XMMWORD [XMMBLOCK(4,0,rsi,SIZEOF_JCOEF)] michael@0: movdqa xmm3, XMMWORD [XMMBLOCK(6,0,rsi,SIZEOF_JCOEF)] michael@0: pmullw xmm2, XMMWORD [XMMBLOCK(4,0,rdx,SIZEOF_ISLOW_MULT_TYPE)] michael@0: pmullw xmm3, XMMWORD [XMMBLOCK(6,0,rdx,SIZEOF_ISLOW_MULT_TYPE)] michael@0: michael@0: ; (Original) michael@0: ; z1 = (z2 + z3) * 0.541196100; michael@0: ; tmp2 = z1 + z3 * -1.847759065; michael@0: ; tmp3 = z1 + z2 * 0.765366865; michael@0: ; michael@0: ; (This implementation) michael@0: ; tmp2 = z2 * 0.541196100 + z3 * (0.541196100 - 1.847759065); michael@0: ; tmp3 = z2 * (0.541196100 + 0.765366865) + z3 * 0.541196100; michael@0: michael@0: movdqa xmm4,xmm1 ; xmm1=in2=z2 michael@0: movdqa xmm5,xmm1 michael@0: punpcklwd xmm4,xmm3 ; xmm3=in6=z3 michael@0: punpckhwd xmm5,xmm3 michael@0: movdqa xmm1,xmm4 michael@0: movdqa xmm3,xmm5 michael@0: pmaddwd xmm4,[rel PW_F130_F054] ; xmm4=tmp3L michael@0: pmaddwd xmm5,[rel PW_F130_F054] ; xmm5=tmp3H michael@0: pmaddwd xmm1,[rel PW_F054_MF130] ; xmm1=tmp2L michael@0: pmaddwd xmm3,[rel PW_F054_MF130] ; xmm3=tmp2H michael@0: michael@0: movdqa xmm6,xmm0 michael@0: paddw xmm0,xmm2 ; xmm0=in0+in4 michael@0: psubw xmm6,xmm2 ; xmm6=in0-in4 michael@0: michael@0: pxor xmm7,xmm7 michael@0: pxor xmm2,xmm2 michael@0: punpcklwd xmm7,xmm0 ; xmm7=tmp0L michael@0: punpckhwd xmm2,xmm0 ; xmm2=tmp0H michael@0: psrad xmm7,(16-CONST_BITS) ; psrad xmm7,16 & pslld xmm7,CONST_BITS michael@0: psrad xmm2,(16-CONST_BITS) ; psrad xmm2,16 & pslld xmm2,CONST_BITS michael@0: michael@0: movdqa xmm0,xmm7 michael@0: paddd xmm7,xmm4 ; xmm7=tmp10L michael@0: psubd xmm0,xmm4 ; xmm0=tmp13L michael@0: movdqa xmm4,xmm2 michael@0: paddd xmm2,xmm5 ; xmm2=tmp10H michael@0: psubd xmm4,xmm5 ; xmm4=tmp13H michael@0: michael@0: movdqa XMMWORD [wk(0)], xmm7 ; wk(0)=tmp10L michael@0: movdqa XMMWORD [wk(1)], xmm2 ; wk(1)=tmp10H michael@0: movdqa XMMWORD [wk(2)], xmm0 ; wk(2)=tmp13L michael@0: movdqa XMMWORD [wk(3)], xmm4 ; wk(3)=tmp13H michael@0: michael@0: pxor xmm5,xmm5 michael@0: pxor xmm7,xmm7 michael@0: punpcklwd xmm5,xmm6 ; xmm5=tmp1L michael@0: punpckhwd xmm7,xmm6 ; xmm7=tmp1H michael@0: psrad xmm5,(16-CONST_BITS) ; psrad xmm5,16 & pslld xmm5,CONST_BITS michael@0: psrad xmm7,(16-CONST_BITS) ; psrad xmm7,16 & pslld xmm7,CONST_BITS michael@0: michael@0: movdqa xmm2,xmm5 michael@0: paddd xmm5,xmm1 ; xmm5=tmp11L michael@0: psubd xmm2,xmm1 ; xmm2=tmp12L michael@0: movdqa xmm0,xmm7 michael@0: paddd xmm7,xmm3 ; xmm7=tmp11H michael@0: psubd xmm0,xmm3 ; xmm0=tmp12H michael@0: michael@0: movdqa XMMWORD [wk(4)], xmm5 ; wk(4)=tmp11L michael@0: movdqa XMMWORD [wk(5)], xmm7 ; wk(5)=tmp11H michael@0: movdqa XMMWORD [wk(6)], xmm2 ; wk(6)=tmp12L michael@0: movdqa XMMWORD [wk(7)], xmm0 ; wk(7)=tmp12H michael@0: michael@0: ; -- Odd part michael@0: michael@0: movdqa xmm4, XMMWORD [XMMBLOCK(1,0,rsi,SIZEOF_JCOEF)] michael@0: movdqa xmm6, XMMWORD [XMMBLOCK(3,0,rsi,SIZEOF_JCOEF)] michael@0: pmullw xmm4, XMMWORD [XMMBLOCK(1,0,rdx,SIZEOF_ISLOW_MULT_TYPE)] michael@0: pmullw xmm6, XMMWORD [XMMBLOCK(3,0,rdx,SIZEOF_ISLOW_MULT_TYPE)] michael@0: movdqa xmm1, XMMWORD [XMMBLOCK(5,0,rsi,SIZEOF_JCOEF)] michael@0: movdqa xmm3, XMMWORD [XMMBLOCK(7,0,rsi,SIZEOF_JCOEF)] michael@0: pmullw xmm1, XMMWORD [XMMBLOCK(5,0,rdx,SIZEOF_ISLOW_MULT_TYPE)] michael@0: pmullw xmm3, XMMWORD [XMMBLOCK(7,0,rdx,SIZEOF_ISLOW_MULT_TYPE)] michael@0: michael@0: movdqa xmm5,xmm6 michael@0: movdqa xmm7,xmm4 michael@0: paddw xmm5,xmm3 ; xmm5=z3 michael@0: paddw xmm7,xmm1 ; xmm7=z4 michael@0: michael@0: ; (Original) michael@0: ; z5 = (z3 + z4) * 1.175875602; michael@0: ; z3 = z3 * -1.961570560; z4 = z4 * -0.390180644; michael@0: ; z3 += z5; z4 += z5; michael@0: ; michael@0: ; (This implementation) michael@0: ; z3 = z3 * (1.175875602 - 1.961570560) + z4 * 1.175875602; michael@0: ; z4 = z3 * 1.175875602 + z4 * (1.175875602 - 0.390180644); michael@0: michael@0: movdqa xmm2,xmm5 michael@0: movdqa xmm0,xmm5 michael@0: punpcklwd xmm2,xmm7 michael@0: punpckhwd xmm0,xmm7 michael@0: movdqa xmm5,xmm2 michael@0: movdqa xmm7,xmm0 michael@0: pmaddwd xmm2,[rel PW_MF078_F117] ; xmm2=z3L michael@0: pmaddwd xmm0,[rel PW_MF078_F117] ; xmm0=z3H michael@0: pmaddwd xmm5,[rel PW_F117_F078] ; xmm5=z4L michael@0: pmaddwd xmm7,[rel PW_F117_F078] ; xmm7=z4H michael@0: michael@0: movdqa XMMWORD [wk(10)], xmm2 ; wk(10)=z3L michael@0: movdqa XMMWORD [wk(11)], xmm0 ; wk(11)=z3H michael@0: michael@0: ; (Original) michael@0: ; z1 = tmp0 + tmp3; z2 = tmp1 + tmp2; michael@0: ; tmp0 = tmp0 * 0.298631336; tmp1 = tmp1 * 2.053119869; michael@0: ; tmp2 = tmp2 * 3.072711026; tmp3 = tmp3 * 1.501321110; michael@0: ; z1 = z1 * -0.899976223; z2 = z2 * -2.562915447; michael@0: ; tmp0 += z1 + z3; tmp1 += z2 + z4; michael@0: ; tmp2 += z2 + z3; tmp3 += z1 + z4; michael@0: ; michael@0: ; (This implementation) michael@0: ; tmp0 = tmp0 * (0.298631336 - 0.899976223) + tmp3 * -0.899976223; michael@0: ; tmp1 = tmp1 * (2.053119869 - 2.562915447) + tmp2 * -2.562915447; michael@0: ; tmp2 = tmp1 * -2.562915447 + tmp2 * (3.072711026 - 2.562915447); michael@0: ; tmp3 = tmp0 * -0.899976223 + tmp3 * (1.501321110 - 0.899976223); michael@0: ; tmp0 += z3; tmp1 += z4; michael@0: ; tmp2 += z3; tmp3 += z4; michael@0: michael@0: movdqa xmm2,xmm3 michael@0: movdqa xmm0,xmm3 michael@0: punpcklwd xmm2,xmm4 michael@0: punpckhwd xmm0,xmm4 michael@0: movdqa xmm3,xmm2 michael@0: movdqa xmm4,xmm0 michael@0: pmaddwd xmm2,[rel PW_MF060_MF089] ; xmm2=tmp0L michael@0: pmaddwd xmm0,[rel PW_MF060_MF089] ; xmm0=tmp0H michael@0: pmaddwd xmm3,[rel PW_MF089_F060] ; xmm3=tmp3L michael@0: pmaddwd xmm4,[rel PW_MF089_F060] ; xmm4=tmp3H michael@0: michael@0: paddd xmm2, XMMWORD [wk(10)] ; xmm2=tmp0L michael@0: paddd xmm0, XMMWORD [wk(11)] ; xmm0=tmp0H michael@0: paddd xmm3,xmm5 ; xmm3=tmp3L michael@0: paddd xmm4,xmm7 ; xmm4=tmp3H michael@0: michael@0: movdqa XMMWORD [wk(8)], xmm2 ; wk(8)=tmp0L michael@0: movdqa XMMWORD [wk(9)], xmm0 ; wk(9)=tmp0H michael@0: michael@0: movdqa xmm2,xmm1 michael@0: movdqa xmm0,xmm1 michael@0: punpcklwd xmm2,xmm6 michael@0: punpckhwd xmm0,xmm6 michael@0: movdqa xmm1,xmm2 michael@0: movdqa xmm6,xmm0 michael@0: pmaddwd xmm2,[rel PW_MF050_MF256] ; xmm2=tmp1L michael@0: pmaddwd xmm0,[rel PW_MF050_MF256] ; xmm0=tmp1H michael@0: pmaddwd xmm1,[rel PW_MF256_F050] ; xmm1=tmp2L michael@0: pmaddwd xmm6,[rel PW_MF256_F050] ; xmm6=tmp2H michael@0: michael@0: paddd xmm2,xmm5 ; xmm2=tmp1L michael@0: paddd xmm0,xmm7 ; xmm0=tmp1H michael@0: paddd xmm1, XMMWORD [wk(10)] ; xmm1=tmp2L michael@0: paddd xmm6, XMMWORD [wk(11)] ; xmm6=tmp2H michael@0: michael@0: movdqa XMMWORD [wk(10)], xmm2 ; wk(10)=tmp1L michael@0: movdqa XMMWORD [wk(11)], xmm0 ; wk(11)=tmp1H michael@0: michael@0: ; -- Final output stage michael@0: michael@0: movdqa xmm5, XMMWORD [wk(0)] ; xmm5=tmp10L michael@0: movdqa xmm7, XMMWORD [wk(1)] ; xmm7=tmp10H michael@0: michael@0: movdqa xmm2,xmm5 michael@0: movdqa xmm0,xmm7 michael@0: paddd xmm5,xmm3 ; xmm5=data0L michael@0: paddd xmm7,xmm4 ; xmm7=data0H michael@0: psubd xmm2,xmm3 ; xmm2=data7L michael@0: psubd xmm0,xmm4 ; xmm0=data7H michael@0: michael@0: movdqa xmm3,[rel PD_DESCALE_P1] ; xmm3=[rel PD_DESCALE_P1] michael@0: michael@0: paddd xmm5,xmm3 michael@0: paddd xmm7,xmm3 michael@0: psrad xmm5,DESCALE_P1 michael@0: psrad xmm7,DESCALE_P1 michael@0: paddd xmm2,xmm3 michael@0: paddd xmm0,xmm3 michael@0: psrad xmm2,DESCALE_P1 michael@0: psrad xmm0,DESCALE_P1 michael@0: michael@0: packssdw xmm5,xmm7 ; xmm5=data0=(00 01 02 03 04 05 06 07) michael@0: packssdw xmm2,xmm0 ; xmm2=data7=(70 71 72 73 74 75 76 77) michael@0: michael@0: movdqa xmm4, XMMWORD [wk(4)] ; xmm4=tmp11L michael@0: movdqa xmm3, XMMWORD [wk(5)] ; xmm3=tmp11H michael@0: michael@0: movdqa xmm7,xmm4 michael@0: movdqa xmm0,xmm3 michael@0: paddd xmm4,xmm1 ; xmm4=data1L michael@0: paddd xmm3,xmm6 ; xmm3=data1H michael@0: psubd xmm7,xmm1 ; xmm7=data6L michael@0: psubd xmm0,xmm6 ; xmm0=data6H michael@0: michael@0: movdqa xmm1,[rel PD_DESCALE_P1] ; xmm1=[rel PD_DESCALE_P1] michael@0: michael@0: paddd xmm4,xmm1 michael@0: paddd xmm3,xmm1 michael@0: psrad xmm4,DESCALE_P1 michael@0: psrad xmm3,DESCALE_P1 michael@0: paddd xmm7,xmm1 michael@0: paddd xmm0,xmm1 michael@0: psrad xmm7,DESCALE_P1 michael@0: psrad xmm0,DESCALE_P1 michael@0: michael@0: packssdw xmm4,xmm3 ; xmm4=data1=(10 11 12 13 14 15 16 17) michael@0: packssdw xmm7,xmm0 ; xmm7=data6=(60 61 62 63 64 65 66 67) michael@0: michael@0: movdqa xmm6,xmm5 ; transpose coefficients(phase 1) michael@0: punpcklwd xmm5,xmm4 ; xmm5=(00 10 01 11 02 12 03 13) michael@0: punpckhwd xmm6,xmm4 ; xmm6=(04 14 05 15 06 16 07 17) michael@0: movdqa xmm1,xmm7 ; transpose coefficients(phase 1) michael@0: punpcklwd xmm7,xmm2 ; xmm7=(60 70 61 71 62 72 63 73) michael@0: punpckhwd xmm1,xmm2 ; xmm1=(64 74 65 75 66 76 67 77) michael@0: michael@0: movdqa xmm3, XMMWORD [wk(6)] ; xmm3=tmp12L michael@0: movdqa xmm0, XMMWORD [wk(7)] ; xmm0=tmp12H michael@0: movdqa xmm4, XMMWORD [wk(10)] ; xmm4=tmp1L michael@0: movdqa xmm2, XMMWORD [wk(11)] ; xmm2=tmp1H michael@0: michael@0: movdqa XMMWORD [wk(0)], xmm5 ; wk(0)=(00 10 01 11 02 12 03 13) michael@0: movdqa XMMWORD [wk(1)], xmm6 ; wk(1)=(04 14 05 15 06 16 07 17) michael@0: movdqa XMMWORD [wk(4)], xmm7 ; wk(4)=(60 70 61 71 62 72 63 73) michael@0: movdqa XMMWORD [wk(5)], xmm1 ; wk(5)=(64 74 65 75 66 76 67 77) michael@0: michael@0: movdqa xmm5,xmm3 michael@0: movdqa xmm6,xmm0 michael@0: paddd xmm3,xmm4 ; xmm3=data2L michael@0: paddd xmm0,xmm2 ; xmm0=data2H michael@0: psubd xmm5,xmm4 ; xmm5=data5L michael@0: psubd xmm6,xmm2 ; xmm6=data5H michael@0: michael@0: movdqa xmm7,[rel PD_DESCALE_P1] ; xmm7=[rel PD_DESCALE_P1] michael@0: michael@0: paddd xmm3,xmm7 michael@0: paddd xmm0,xmm7 michael@0: psrad xmm3,DESCALE_P1 michael@0: psrad xmm0,DESCALE_P1 michael@0: paddd xmm5,xmm7 michael@0: paddd xmm6,xmm7 michael@0: psrad xmm5,DESCALE_P1 michael@0: psrad xmm6,DESCALE_P1 michael@0: michael@0: packssdw xmm3,xmm0 ; xmm3=data2=(20 21 22 23 24 25 26 27) michael@0: packssdw xmm5,xmm6 ; xmm5=data5=(50 51 52 53 54 55 56 57) michael@0: michael@0: movdqa xmm1, XMMWORD [wk(2)] ; xmm1=tmp13L michael@0: movdqa xmm4, XMMWORD [wk(3)] ; xmm4=tmp13H michael@0: movdqa xmm2, XMMWORD [wk(8)] ; xmm2=tmp0L michael@0: movdqa xmm7, XMMWORD [wk(9)] ; xmm7=tmp0H michael@0: michael@0: movdqa xmm0,xmm1 michael@0: movdqa xmm6,xmm4 michael@0: paddd xmm1,xmm2 ; xmm1=data3L michael@0: paddd xmm4,xmm7 ; xmm4=data3H michael@0: psubd xmm0,xmm2 ; xmm0=data4L michael@0: psubd xmm6,xmm7 ; xmm6=data4H michael@0: michael@0: movdqa xmm2,[rel PD_DESCALE_P1] ; xmm2=[rel PD_DESCALE_P1] michael@0: michael@0: paddd xmm1,xmm2 michael@0: paddd xmm4,xmm2 michael@0: psrad xmm1,DESCALE_P1 michael@0: psrad xmm4,DESCALE_P1 michael@0: paddd xmm0,xmm2 michael@0: paddd xmm6,xmm2 michael@0: psrad xmm0,DESCALE_P1 michael@0: psrad xmm6,DESCALE_P1 michael@0: michael@0: packssdw xmm1,xmm4 ; xmm1=data3=(30 31 32 33 34 35 36 37) michael@0: packssdw xmm0,xmm6 ; xmm0=data4=(40 41 42 43 44 45 46 47) michael@0: michael@0: movdqa xmm7, XMMWORD [wk(0)] ; xmm7=(00 10 01 11 02 12 03 13) michael@0: movdqa xmm2, XMMWORD [wk(1)] ; xmm2=(04 14 05 15 06 16 07 17) michael@0: michael@0: movdqa xmm4,xmm3 ; transpose coefficients(phase 1) michael@0: punpcklwd xmm3,xmm1 ; xmm3=(20 30 21 31 22 32 23 33) michael@0: punpckhwd xmm4,xmm1 ; xmm4=(24 34 25 35 26 36 27 37) michael@0: movdqa xmm6,xmm0 ; transpose coefficients(phase 1) michael@0: punpcklwd xmm0,xmm5 ; xmm0=(40 50 41 51 42 52 43 53) michael@0: punpckhwd xmm6,xmm5 ; xmm6=(44 54 45 55 46 56 47 57) michael@0: michael@0: movdqa xmm1,xmm7 ; transpose coefficients(phase 2) michael@0: punpckldq xmm7,xmm3 ; xmm7=(00 10 20 30 01 11 21 31) michael@0: punpckhdq xmm1,xmm3 ; xmm1=(02 12 22 32 03 13 23 33) michael@0: movdqa xmm5,xmm2 ; transpose coefficients(phase 2) michael@0: punpckldq xmm2,xmm4 ; xmm2=(04 14 24 34 05 15 25 35) michael@0: punpckhdq xmm5,xmm4 ; xmm5=(06 16 26 36 07 17 27 37) michael@0: michael@0: movdqa xmm3, XMMWORD [wk(4)] ; xmm3=(60 70 61 71 62 72 63 73) michael@0: movdqa xmm4, XMMWORD [wk(5)] ; xmm4=(64 74 65 75 66 76 67 77) michael@0: michael@0: movdqa XMMWORD [wk(6)], xmm2 ; wk(6)=(04 14 24 34 05 15 25 35) michael@0: movdqa XMMWORD [wk(7)], xmm5 ; wk(7)=(06 16 26 36 07 17 27 37) michael@0: michael@0: movdqa xmm2,xmm0 ; transpose coefficients(phase 2) michael@0: punpckldq xmm0,xmm3 ; xmm0=(40 50 60 70 41 51 61 71) michael@0: punpckhdq xmm2,xmm3 ; xmm2=(42 52 62 72 43 53 63 73) michael@0: movdqa xmm5,xmm6 ; transpose coefficients(phase 2) michael@0: punpckldq xmm6,xmm4 ; xmm6=(44 54 64 74 45 55 65 75) michael@0: punpckhdq xmm5,xmm4 ; xmm5=(46 56 66 76 47 57 67 77) michael@0: michael@0: movdqa xmm3,xmm7 ; transpose coefficients(phase 3) michael@0: punpcklqdq xmm7,xmm0 ; xmm7=col0=(00 10 20 30 40 50 60 70) michael@0: punpckhqdq xmm3,xmm0 ; xmm3=col1=(01 11 21 31 41 51 61 71) michael@0: movdqa xmm4,xmm1 ; transpose coefficients(phase 3) michael@0: punpcklqdq xmm1,xmm2 ; xmm1=col2=(02 12 22 32 42 52 62 72) michael@0: punpckhqdq xmm4,xmm2 ; xmm4=col3=(03 13 23 33 43 53 63 73) michael@0: michael@0: movdqa xmm0, XMMWORD [wk(6)] ; xmm0=(04 14 24 34 05 15 25 35) michael@0: movdqa xmm2, XMMWORD [wk(7)] ; xmm2=(06 16 26 36 07 17 27 37) michael@0: michael@0: movdqa XMMWORD [wk(8)], xmm3 ; wk(8)=col1 michael@0: movdqa XMMWORD [wk(9)], xmm4 ; wk(9)=col3 michael@0: michael@0: movdqa xmm3,xmm0 ; transpose coefficients(phase 3) michael@0: punpcklqdq xmm0,xmm6 ; xmm0=col4=(04 14 24 34 44 54 64 74) michael@0: punpckhqdq xmm3,xmm6 ; xmm3=col5=(05 15 25 35 45 55 65 75) michael@0: movdqa xmm4,xmm2 ; transpose coefficients(phase 3) michael@0: punpcklqdq xmm2,xmm5 ; xmm2=col6=(06 16 26 36 46 56 66 76) michael@0: punpckhqdq xmm4,xmm5 ; xmm4=col7=(07 17 27 37 47 57 67 77) michael@0: michael@0: movdqa XMMWORD [wk(10)], xmm3 ; wk(10)=col5 michael@0: movdqa XMMWORD [wk(11)], xmm4 ; wk(11)=col7 michael@0: .column_end: michael@0: michael@0: ; -- Prefetch the next coefficient block michael@0: michael@0: prefetchnta [rsi + DCTSIZE2*SIZEOF_JCOEF + 0*32] michael@0: prefetchnta [rsi + DCTSIZE2*SIZEOF_JCOEF + 1*32] michael@0: prefetchnta [rsi + DCTSIZE2*SIZEOF_JCOEF + 2*32] michael@0: prefetchnta [rsi + DCTSIZE2*SIZEOF_JCOEF + 3*32] michael@0: michael@0: ; ---- Pass 2: process rows from work array, store into output array. michael@0: michael@0: mov rax, [original_rbp] michael@0: mov rdi, r12 ; (JSAMPROW *) michael@0: mov rax, r13 michael@0: michael@0: ; -- Even part michael@0: michael@0: ; xmm7=col0, xmm1=col2, xmm0=col4, xmm2=col6 michael@0: michael@0: ; (Original) michael@0: ; z1 = (z2 + z3) * 0.541196100; michael@0: ; tmp2 = z1 + z3 * -1.847759065; michael@0: ; tmp3 = z1 + z2 * 0.765366865; michael@0: ; michael@0: ; (This implementation) michael@0: ; tmp2 = z2 * 0.541196100 + z3 * (0.541196100 - 1.847759065); michael@0: ; tmp3 = z2 * (0.541196100 + 0.765366865) + z3 * 0.541196100; michael@0: michael@0: movdqa xmm6,xmm1 ; xmm1=in2=z2 michael@0: movdqa xmm5,xmm1 michael@0: punpcklwd xmm6,xmm2 ; xmm2=in6=z3 michael@0: punpckhwd xmm5,xmm2 michael@0: movdqa xmm1,xmm6 michael@0: movdqa xmm2,xmm5 michael@0: pmaddwd xmm6,[rel PW_F130_F054] ; xmm6=tmp3L michael@0: pmaddwd xmm5,[rel PW_F130_F054] ; xmm5=tmp3H michael@0: pmaddwd xmm1,[rel PW_F054_MF130] ; xmm1=tmp2L michael@0: pmaddwd xmm2,[rel PW_F054_MF130] ; xmm2=tmp2H michael@0: michael@0: movdqa xmm3,xmm7 michael@0: paddw xmm7,xmm0 ; xmm7=in0+in4 michael@0: psubw xmm3,xmm0 ; xmm3=in0-in4 michael@0: michael@0: pxor xmm4,xmm4 michael@0: pxor xmm0,xmm0 michael@0: punpcklwd xmm4,xmm7 ; xmm4=tmp0L michael@0: punpckhwd xmm0,xmm7 ; xmm0=tmp0H michael@0: psrad xmm4,(16-CONST_BITS) ; psrad xmm4,16 & pslld xmm4,CONST_BITS michael@0: psrad xmm0,(16-CONST_BITS) ; psrad xmm0,16 & pslld xmm0,CONST_BITS michael@0: michael@0: movdqa xmm7,xmm4 michael@0: paddd xmm4,xmm6 ; xmm4=tmp10L michael@0: psubd xmm7,xmm6 ; xmm7=tmp13L michael@0: movdqa xmm6,xmm0 michael@0: paddd xmm0,xmm5 ; xmm0=tmp10H michael@0: psubd xmm6,xmm5 ; xmm6=tmp13H michael@0: michael@0: movdqa XMMWORD [wk(0)], xmm4 ; wk(0)=tmp10L michael@0: movdqa XMMWORD [wk(1)], xmm0 ; wk(1)=tmp10H michael@0: movdqa XMMWORD [wk(2)], xmm7 ; wk(2)=tmp13L michael@0: movdqa XMMWORD [wk(3)], xmm6 ; wk(3)=tmp13H michael@0: michael@0: pxor xmm5,xmm5 michael@0: pxor xmm4,xmm4 michael@0: punpcklwd xmm5,xmm3 ; xmm5=tmp1L michael@0: punpckhwd xmm4,xmm3 ; xmm4=tmp1H michael@0: psrad xmm5,(16-CONST_BITS) ; psrad xmm5,16 & pslld xmm5,CONST_BITS michael@0: psrad xmm4,(16-CONST_BITS) ; psrad xmm4,16 & pslld xmm4,CONST_BITS michael@0: michael@0: movdqa xmm0,xmm5 michael@0: paddd xmm5,xmm1 ; xmm5=tmp11L michael@0: psubd xmm0,xmm1 ; xmm0=tmp12L michael@0: movdqa xmm7,xmm4 michael@0: paddd xmm4,xmm2 ; xmm4=tmp11H michael@0: psubd xmm7,xmm2 ; xmm7=tmp12H michael@0: michael@0: movdqa XMMWORD [wk(4)], xmm5 ; wk(4)=tmp11L michael@0: movdqa XMMWORD [wk(5)], xmm4 ; wk(5)=tmp11H michael@0: movdqa XMMWORD [wk(6)], xmm0 ; wk(6)=tmp12L michael@0: movdqa XMMWORD [wk(7)], xmm7 ; wk(7)=tmp12H michael@0: michael@0: ; -- Odd part michael@0: michael@0: movdqa xmm6, XMMWORD [wk(9)] ; xmm6=col3 michael@0: movdqa xmm3, XMMWORD [wk(8)] ; xmm3=col1 michael@0: movdqa xmm1, XMMWORD [wk(11)] ; xmm1=col7 michael@0: movdqa xmm2, XMMWORD [wk(10)] ; xmm2=col5 michael@0: michael@0: movdqa xmm5,xmm6 michael@0: movdqa xmm4,xmm3 michael@0: paddw xmm5,xmm1 ; xmm5=z3 michael@0: paddw xmm4,xmm2 ; xmm4=z4 michael@0: michael@0: ; (Original) michael@0: ; z5 = (z3 + z4) * 1.175875602; michael@0: ; z3 = z3 * -1.961570560; z4 = z4 * -0.390180644; michael@0: ; z3 += z5; z4 += z5; michael@0: ; michael@0: ; (This implementation) michael@0: ; z3 = z3 * (1.175875602 - 1.961570560) + z4 * 1.175875602; michael@0: ; z4 = z3 * 1.175875602 + z4 * (1.175875602 - 0.390180644); michael@0: michael@0: movdqa xmm0,xmm5 michael@0: movdqa xmm7,xmm5 michael@0: punpcklwd xmm0,xmm4 michael@0: punpckhwd xmm7,xmm4 michael@0: movdqa xmm5,xmm0 michael@0: movdqa xmm4,xmm7 michael@0: pmaddwd xmm0,[rel PW_MF078_F117] ; xmm0=z3L michael@0: pmaddwd xmm7,[rel PW_MF078_F117] ; xmm7=z3H michael@0: pmaddwd xmm5,[rel PW_F117_F078] ; xmm5=z4L michael@0: pmaddwd xmm4,[rel PW_F117_F078] ; xmm4=z4H michael@0: michael@0: movdqa XMMWORD [wk(10)], xmm0 ; wk(10)=z3L michael@0: movdqa XMMWORD [wk(11)], xmm7 ; wk(11)=z3H michael@0: michael@0: ; (Original) michael@0: ; z1 = tmp0 + tmp3; z2 = tmp1 + tmp2; michael@0: ; tmp0 = tmp0 * 0.298631336; tmp1 = tmp1 * 2.053119869; michael@0: ; tmp2 = tmp2 * 3.072711026; tmp3 = tmp3 * 1.501321110; michael@0: ; z1 = z1 * -0.899976223; z2 = z2 * -2.562915447; michael@0: ; tmp0 += z1 + z3; tmp1 += z2 + z4; michael@0: ; tmp2 += z2 + z3; tmp3 += z1 + z4; michael@0: ; michael@0: ; (This implementation) michael@0: ; tmp0 = tmp0 * (0.298631336 - 0.899976223) + tmp3 * -0.899976223; michael@0: ; tmp1 = tmp1 * (2.053119869 - 2.562915447) + tmp2 * -2.562915447; michael@0: ; tmp2 = tmp1 * -2.562915447 + tmp2 * (3.072711026 - 2.562915447); michael@0: ; tmp3 = tmp0 * -0.899976223 + tmp3 * (1.501321110 - 0.899976223); michael@0: ; tmp0 += z3; tmp1 += z4; michael@0: ; tmp2 += z3; tmp3 += z4; michael@0: michael@0: movdqa xmm0,xmm1 michael@0: movdqa xmm7,xmm1 michael@0: punpcklwd xmm0,xmm3 michael@0: punpckhwd xmm7,xmm3 michael@0: movdqa xmm1,xmm0 michael@0: movdqa xmm3,xmm7 michael@0: pmaddwd xmm0,[rel PW_MF060_MF089] ; xmm0=tmp0L michael@0: pmaddwd xmm7,[rel PW_MF060_MF089] ; xmm7=tmp0H michael@0: pmaddwd xmm1,[rel PW_MF089_F060] ; xmm1=tmp3L michael@0: pmaddwd xmm3,[rel PW_MF089_F060] ; xmm3=tmp3H michael@0: michael@0: paddd xmm0, XMMWORD [wk(10)] ; xmm0=tmp0L michael@0: paddd xmm7, XMMWORD [wk(11)] ; xmm7=tmp0H michael@0: paddd xmm1,xmm5 ; xmm1=tmp3L michael@0: paddd xmm3,xmm4 ; xmm3=tmp3H michael@0: michael@0: movdqa XMMWORD [wk(8)], xmm0 ; wk(8)=tmp0L michael@0: movdqa XMMWORD [wk(9)], xmm7 ; wk(9)=tmp0H michael@0: michael@0: movdqa xmm0,xmm2 michael@0: movdqa xmm7,xmm2 michael@0: punpcklwd xmm0,xmm6 michael@0: punpckhwd xmm7,xmm6 michael@0: movdqa xmm2,xmm0 michael@0: movdqa xmm6,xmm7 michael@0: pmaddwd xmm0,[rel PW_MF050_MF256] ; xmm0=tmp1L michael@0: pmaddwd xmm7,[rel PW_MF050_MF256] ; xmm7=tmp1H michael@0: pmaddwd xmm2,[rel PW_MF256_F050] ; xmm2=tmp2L michael@0: pmaddwd xmm6,[rel PW_MF256_F050] ; xmm6=tmp2H michael@0: michael@0: paddd xmm0,xmm5 ; xmm0=tmp1L michael@0: paddd xmm7,xmm4 ; xmm7=tmp1H michael@0: paddd xmm2, XMMWORD [wk(10)] ; xmm2=tmp2L michael@0: paddd xmm6, XMMWORD [wk(11)] ; xmm6=tmp2H michael@0: michael@0: movdqa XMMWORD [wk(10)], xmm0 ; wk(10)=tmp1L michael@0: movdqa XMMWORD [wk(11)], xmm7 ; wk(11)=tmp1H michael@0: michael@0: ; -- Final output stage michael@0: michael@0: movdqa xmm5, XMMWORD [wk(0)] ; xmm5=tmp10L michael@0: movdqa xmm4, XMMWORD [wk(1)] ; xmm4=tmp10H michael@0: michael@0: movdqa xmm0,xmm5 michael@0: movdqa xmm7,xmm4 michael@0: paddd xmm5,xmm1 ; xmm5=data0L michael@0: paddd xmm4,xmm3 ; xmm4=data0H michael@0: psubd xmm0,xmm1 ; xmm0=data7L michael@0: psubd xmm7,xmm3 ; xmm7=data7H michael@0: michael@0: movdqa xmm1,[rel PD_DESCALE_P2] ; xmm1=[rel PD_DESCALE_P2] michael@0: michael@0: paddd xmm5,xmm1 michael@0: paddd xmm4,xmm1 michael@0: psrad xmm5,DESCALE_P2 michael@0: psrad xmm4,DESCALE_P2 michael@0: paddd xmm0,xmm1 michael@0: paddd xmm7,xmm1 michael@0: psrad xmm0,DESCALE_P2 michael@0: psrad xmm7,DESCALE_P2 michael@0: michael@0: packssdw xmm5,xmm4 ; xmm5=data0=(00 10 20 30 40 50 60 70) michael@0: packssdw xmm0,xmm7 ; xmm0=data7=(07 17 27 37 47 57 67 77) michael@0: michael@0: movdqa xmm3, XMMWORD [wk(4)] ; xmm3=tmp11L michael@0: movdqa xmm1, XMMWORD [wk(5)] ; xmm1=tmp11H michael@0: michael@0: movdqa xmm4,xmm3 michael@0: movdqa xmm7,xmm1 michael@0: paddd xmm3,xmm2 ; xmm3=data1L michael@0: paddd xmm1,xmm6 ; xmm1=data1H michael@0: psubd xmm4,xmm2 ; xmm4=data6L michael@0: psubd xmm7,xmm6 ; xmm7=data6H michael@0: michael@0: movdqa xmm2,[rel PD_DESCALE_P2] ; xmm2=[rel PD_DESCALE_P2] michael@0: michael@0: paddd xmm3,xmm2 michael@0: paddd xmm1,xmm2 michael@0: psrad xmm3,DESCALE_P2 michael@0: psrad xmm1,DESCALE_P2 michael@0: paddd xmm4,xmm2 michael@0: paddd xmm7,xmm2 michael@0: psrad xmm4,DESCALE_P2 michael@0: psrad xmm7,DESCALE_P2 michael@0: michael@0: packssdw xmm3,xmm1 ; xmm3=data1=(01 11 21 31 41 51 61 71) michael@0: packssdw xmm4,xmm7 ; xmm4=data6=(06 16 26 36 46 56 66 76) michael@0: michael@0: packsswb xmm5,xmm4 ; xmm5=(00 10 20 30 40 50 60 70 06 16 26 36 46 56 66 76) michael@0: packsswb xmm3,xmm0 ; xmm3=(01 11 21 31 41 51 61 71 07 17 27 37 47 57 67 77) michael@0: michael@0: movdqa xmm6, XMMWORD [wk(6)] ; xmm6=tmp12L michael@0: movdqa xmm2, XMMWORD [wk(7)] ; xmm2=tmp12H michael@0: movdqa xmm1, XMMWORD [wk(10)] ; xmm1=tmp1L michael@0: movdqa xmm7, XMMWORD [wk(11)] ; xmm7=tmp1H michael@0: michael@0: movdqa XMMWORD [wk(0)], xmm5 ; wk(0)=(00 10 20 30 40 50 60 70 06 16 26 36 46 56 66 76) michael@0: movdqa XMMWORD [wk(1)], xmm3 ; wk(1)=(01 11 21 31 41 51 61 71 07 17 27 37 47 57 67 77) michael@0: michael@0: movdqa xmm4,xmm6 michael@0: movdqa xmm0,xmm2 michael@0: paddd xmm6,xmm1 ; xmm6=data2L michael@0: paddd xmm2,xmm7 ; xmm2=data2H michael@0: psubd xmm4,xmm1 ; xmm4=data5L michael@0: psubd xmm0,xmm7 ; xmm0=data5H michael@0: michael@0: movdqa xmm5,[rel PD_DESCALE_P2] ; xmm5=[rel PD_DESCALE_P2] michael@0: michael@0: paddd xmm6,xmm5 michael@0: paddd xmm2,xmm5 michael@0: psrad xmm6,DESCALE_P2 michael@0: psrad xmm2,DESCALE_P2 michael@0: paddd xmm4,xmm5 michael@0: paddd xmm0,xmm5 michael@0: psrad xmm4,DESCALE_P2 michael@0: psrad xmm0,DESCALE_P2 michael@0: michael@0: packssdw xmm6,xmm2 ; xmm6=data2=(02 12 22 32 42 52 62 72) michael@0: packssdw xmm4,xmm0 ; xmm4=data5=(05 15 25 35 45 55 65 75) michael@0: michael@0: movdqa xmm3, XMMWORD [wk(2)] ; xmm3=tmp13L michael@0: movdqa xmm1, XMMWORD [wk(3)] ; xmm1=tmp13H michael@0: movdqa xmm7, XMMWORD [wk(8)] ; xmm7=tmp0L michael@0: movdqa xmm5, XMMWORD [wk(9)] ; xmm5=tmp0H michael@0: michael@0: movdqa xmm2,xmm3 michael@0: movdqa xmm0,xmm1 michael@0: paddd xmm3,xmm7 ; xmm3=data3L michael@0: paddd xmm1,xmm5 ; xmm1=data3H michael@0: psubd xmm2,xmm7 ; xmm2=data4L michael@0: psubd xmm0,xmm5 ; xmm0=data4H michael@0: michael@0: movdqa xmm7,[rel PD_DESCALE_P2] ; xmm7=[rel PD_DESCALE_P2] michael@0: michael@0: paddd xmm3,xmm7 michael@0: paddd xmm1,xmm7 michael@0: psrad xmm3,DESCALE_P2 michael@0: psrad xmm1,DESCALE_P2 michael@0: paddd xmm2,xmm7 michael@0: paddd xmm0,xmm7 michael@0: psrad xmm2,DESCALE_P2 michael@0: psrad xmm0,DESCALE_P2 michael@0: michael@0: movdqa xmm5,[rel PB_CENTERJSAMP] ; xmm5=[rel PB_CENTERJSAMP] michael@0: michael@0: packssdw xmm3,xmm1 ; xmm3=data3=(03 13 23 33 43 53 63 73) michael@0: packssdw xmm2,xmm0 ; xmm2=data4=(04 14 24 34 44 54 64 74) michael@0: michael@0: movdqa xmm7, XMMWORD [wk(0)] ; xmm7=(00 10 20 30 40 50 60 70 06 16 26 36 46 56 66 76) michael@0: movdqa xmm1, XMMWORD [wk(1)] ; xmm1=(01 11 21 31 41 51 61 71 07 17 27 37 47 57 67 77) michael@0: michael@0: packsswb xmm6,xmm2 ; xmm6=(02 12 22 32 42 52 62 72 04 14 24 34 44 54 64 74) michael@0: packsswb xmm3,xmm4 ; xmm3=(03 13 23 33 43 53 63 73 05 15 25 35 45 55 65 75) michael@0: michael@0: paddb xmm7,xmm5 michael@0: paddb xmm1,xmm5 michael@0: paddb xmm6,xmm5 michael@0: paddb xmm3,xmm5 michael@0: michael@0: movdqa xmm0,xmm7 ; transpose coefficients(phase 1) michael@0: punpcklbw xmm7,xmm1 ; xmm7=(00 01 10 11 20 21 30 31 40 41 50 51 60 61 70 71) michael@0: punpckhbw xmm0,xmm1 ; xmm0=(06 07 16 17 26 27 36 37 46 47 56 57 66 67 76 77) michael@0: movdqa xmm2,xmm6 ; transpose coefficients(phase 1) michael@0: punpcklbw xmm6,xmm3 ; xmm6=(02 03 12 13 22 23 32 33 42 43 52 53 62 63 72 73) michael@0: punpckhbw xmm2,xmm3 ; xmm2=(04 05 14 15 24 25 34 35 44 45 54 55 64 65 74 75) michael@0: michael@0: movdqa xmm4,xmm7 ; transpose coefficients(phase 2) michael@0: punpcklwd xmm7,xmm6 ; xmm7=(00 01 02 03 10 11 12 13 20 21 22 23 30 31 32 33) michael@0: punpckhwd xmm4,xmm6 ; xmm4=(40 41 42 43 50 51 52 53 60 61 62 63 70 71 72 73) michael@0: movdqa xmm5,xmm2 ; transpose coefficients(phase 2) michael@0: punpcklwd xmm2,xmm0 ; xmm2=(04 05 06 07 14 15 16 17 24 25 26 27 34 35 36 37) michael@0: punpckhwd xmm5,xmm0 ; xmm5=(44 45 46 47 54 55 56 57 64 65 66 67 74 75 76 77) michael@0: michael@0: movdqa xmm1,xmm7 ; transpose coefficients(phase 3) michael@0: punpckldq xmm7,xmm2 ; xmm7=(00 01 02 03 04 05 06 07 10 11 12 13 14 15 16 17) michael@0: punpckhdq xmm1,xmm2 ; xmm1=(20 21 22 23 24 25 26 27 30 31 32 33 34 35 36 37) michael@0: movdqa xmm3,xmm4 ; transpose coefficients(phase 3) michael@0: punpckldq xmm4,xmm5 ; xmm4=(40 41 42 43 44 45 46 47 50 51 52 53 54 55 56 57) michael@0: punpckhdq xmm3,xmm5 ; xmm3=(60 61 62 63 64 65 66 67 70 71 72 73 74 75 76 77) michael@0: michael@0: pshufd xmm6,xmm7,0x4E ; xmm6=(10 11 12 13 14 15 16 17 00 01 02 03 04 05 06 07) michael@0: pshufd xmm0,xmm1,0x4E ; xmm0=(30 31 32 33 34 35 36 37 20 21 22 23 24 25 26 27) michael@0: pshufd xmm2,xmm4,0x4E ; xmm2=(50 51 52 53 54 55 56 57 40 41 42 43 44 45 46 47) michael@0: pshufd xmm5,xmm3,0x4E ; xmm5=(70 71 72 73 74 75 76 77 60 61 62 63 64 65 66 67) michael@0: michael@0: mov rdx, JSAMPROW [rdi+0*SIZEOF_JSAMPROW] michael@0: mov rsi, JSAMPROW [rdi+2*SIZEOF_JSAMPROW] michael@0: movq XMM_MMWORD [rdx+rax*SIZEOF_JSAMPLE], xmm7 michael@0: movq XMM_MMWORD [rsi+rax*SIZEOF_JSAMPLE], xmm1 michael@0: mov rdx, JSAMPROW [rdi+4*SIZEOF_JSAMPROW] michael@0: mov rsi, JSAMPROW [rdi+6*SIZEOF_JSAMPROW] michael@0: movq XMM_MMWORD [rdx+rax*SIZEOF_JSAMPLE], xmm4 michael@0: movq XMM_MMWORD [rsi+rax*SIZEOF_JSAMPLE], xmm3 michael@0: michael@0: mov rdx, JSAMPROW [rdi+1*SIZEOF_JSAMPROW] michael@0: mov rsi, JSAMPROW [rdi+3*SIZEOF_JSAMPROW] michael@0: movq XMM_MMWORD [rdx+rax*SIZEOF_JSAMPLE], xmm6 michael@0: movq XMM_MMWORD [rsi+rax*SIZEOF_JSAMPLE], xmm0 michael@0: mov rdx, JSAMPROW [rdi+5*SIZEOF_JSAMPROW] michael@0: mov rsi, JSAMPROW [rdi+7*SIZEOF_JSAMPROW] michael@0: movq XMM_MMWORD [rdx+rax*SIZEOF_JSAMPLE], xmm2 michael@0: movq XMM_MMWORD [rsi+rax*SIZEOF_JSAMPLE], xmm5 michael@0: michael@0: uncollect_args michael@0: mov rsp,rbp ; rsp <- aligned rbp michael@0: pop rsp ; rsp <- original rbp michael@0: pop rbp michael@0: ret michael@0: michael@0: ; For some reason, the OS X linker does not honor the request to align the michael@0: ; segment unless we do this. michael@0: align 16