Subversion Repositories Kolibri OS

Rev

Go to most recent revision | Blame | Last modification | View Log | RSS feed

  1. /*
  2.  * Copyright (C) 2002 Frederic 'dilb' Boulay
  3.  *
  4.  * Author: Frederic Boulay <dilb@handhelds.org>
  5.  *
  6.  * The function defined in this file is derived from the simple_idct function
  7.  * from the libavcodec library part of the FFmpeg project.
  8.  *
  9.  * This file is part of FFmpeg.
  10.  *
  11.  * FFmpeg is free software; you can redistribute it and/or
  12.  * modify it under the terms of the GNU Lesser General Public
  13.  * License as published by the Free Software Foundation; either
  14.  * version 2.1 of the License, or (at your option) any later version.
  15.  *
  16.  * FFmpeg is distributed in the hope that it will be useful,
  17.  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  18.  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  19.  * Lesser General Public License for more details.
  20.  *
  21.  * You should have received a copy of the GNU Lesser General Public
  22.  * License along with FFmpeg; if not, write to the Free Software
  23.  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  24.  */
  25.  
  26. #include "libavutil/arm/asm.S"
  27.  
  28. /* useful constants for the algorithm */
  29. #define W1  22725
  30. #define W2  21407
  31. #define W3  19266
  32. #define W4  16383
  33. #define W5  12873
  34. #define W6  8867
  35. #define W7  4520
  36. #define MASK_MSHW 0xFFFF0000
  37.  
  38. #define ROW_SHIFT 11
  39. #define ROW_SHIFT2MSHW (16-11)
  40. #define COL_SHIFT 20
  41. #define ROW_SHIFTED_1 1024 /* 1<< (ROW_SHIFT-1) */
  42. #define COL_SHIFTED_1 524288 /* 1<< (COL_SHIFT-1) */
  43.  
  44.  
  45. function ff_simple_idct_arm, export=1
  46.         @@ void simple_idct_arm(int16_t *block)
  47.         @@ save stack for reg needed (take all of them),
  48.         @@ R0-R3 are scratch regs, so no need to save them, but R0 contains the pointer to block
  49.         @@ so it must not be overwritten, if it is not saved!!
  50.         @@ R12 is another scratch register, so it should not be saved too
  51.         @@ save all registers
  52.         stmfd sp!, {r4-r11, r14} @ R14 is also called LR
  53.         @@ at this point, R0=block, other registers are free.
  54.         add r14, r0, #112        @ R14=&block[8*7], better start from the last row, and decrease the value until row=0, i.e. R12=block.
  55.         @@ add 2 temporary variables in the stack: R0 and R14
  56.         sub sp, sp, #8          @ allow 2 local variables
  57.         str r0, [sp, #0]        @ save block in sp[0]
  58.         @@ stack status
  59.         @@ sp+4   free
  60.         @@ sp+0   R0  (block)
  61.  
  62.  
  63.         @@ at this point, R0=block, R14=&block[56], R12=__const_ptr_, R1-R11 free
  64.  
  65.  
  66. __row_loop:
  67.         @@ read the row and check if it is null, almost null, or not, according to strongarm specs, it is not necessary to optimize ldr accesses (i.e. split 32bits in 2 16bits words), at least it gives more usable registers :)
  68.         ldr r1, [r14, #0]        @ R1=(int32)(R12)[0]=ROWr32[0] (relative row cast to a 32b pointer)
  69.         ldr r2, [r14, #4]        @ R2=(int32)(R12)[1]=ROWr32[1]
  70.         ldr r3, [r14, #8]        @ R3=ROWr32[2]
  71.         ldr r4, [r14, #12]       @ R4=ROWr32[3]
  72.         @@ check if the words are null, if all of them are null, then proceed with next row (branch __end_row_loop),
  73.         @@ if ROWr16[0] is the only one not null, then proceed with this special case (branch __almost_empty_row)
  74.         @@ else follow the complete algorithm.
  75.         @@ at this point, R0=block, R14=&block[n], R12=__const_ptr_, R1=ROWr32[0], R2=ROWr32[1],
  76.         @@                R3=ROWr32[2], R4=ROWr32[3], R5-R11 free
  77.         orr r5, r4, r3           @ R5=R4 | R3
  78.         orr r5, r5, r2           @ R5=R4 | R3 | R2
  79.         orrs r6, r5, r1          @ Test R5 | R1 (the aim is to check if everything is null)
  80.         beq __end_row_loop
  81.         mov r7, r1, asr #16      @ R7=R1>>16=ROWr16[1] (evaluate it now, as it could be useful later)
  82.         ldrsh r6, [r14, #0]      @ R6=ROWr16[0]
  83.         orrs r5, r5, r7          @ R5=R4 | R3 | R2 | R7
  84.         beq __almost_empty_row
  85.  
  86. @@ __b_evaluation:
  87.         @@ at this point, R0=block (temp),  R1(free), R2=ROWr32[1], R3=ROWr32[2], R4=ROWr32[3],
  88.         @@     R5=(temp), R6=ROWr16[0], R7=ROWr16[1], R8-R11 free,
  89.         @@     R12=__const_ptr_, R14=&block[n]
  90.         @@ to save some registers/calls, proceed with b0-b3 first, followed by a0-a3
  91.  
  92.         @@ MUL16(b0, W1, row[1]);
  93.         @@ MUL16(b1, W3, row[1]);
  94.         @@ MUL16(b2, W5, row[1]);
  95.         @@ MUL16(b3, W7, row[1]);
  96.         @@ MAC16(b0, W3, row[3]);
  97.         @@ MAC16(b1, -W7, row[3]);
  98.         @@ MAC16(b2, -W1, row[3]);
  99.         @@ MAC16(b3, -W5, row[3]);
  100.         ldr r8, =W1              @ R8=W1
  101.         mov r2, r2, asr #16      @ R2=ROWr16[3]
  102.         mul r0, r8, r7           @ R0=W1*ROWr16[1]=b0 (ROWr16[1] must be the second arg, to have the possibility to save 1 cycle)
  103.         ldr r9, =W3              @ R9=W3
  104.         ldr r10, =W5             @ R10=W5
  105.         mul r1, r9, r7           @ R1=W3*ROWr16[1]=b1 (ROWr16[1] must be the second arg, to have the possibility to save 1 cycle)
  106.         ldr r11, =W7             @ R11=W7
  107.         mul r5, r10, r7          @ R5=W5*ROWr16[1]=b2 (ROWr16[1] must be the second arg, to have the possibility to save 1 cycle)
  108.         mul r7, r11, r7          @ R7=W7*ROWr16[1]=b3 (ROWr16[1] must be the second arg, to have the possibility to save 1 cycle)
  109.         teq r2, #0               @ if null avoid muls
  110.         itttt ne
  111.         mlane r0, r9, r2, r0     @ R0+=W3*ROWr16[3]=b0 (ROWr16[3] must be the second arg, to have the possibility to save 1 cycle)
  112.         rsbne r2, r2, #0         @ R2=-ROWr16[3]
  113.         mlane r1, r11, r2, r1    @ R1-=W7*ROWr16[3]=b1 (ROWr16[3] must be the second arg, to have the possibility to save 1 cycle)
  114.         mlane r5, r8, r2, r5     @ R5-=W1*ROWr16[3]=b2 (ROWr16[3] must be the second arg, to have the possibility to save 1 cycle)
  115.         it    ne
  116.         mlane r7, r10, r2, r7    @ R7-=W5*ROWr16[3]=b3 (ROWr16[3] must be the second arg, to have the possibility to save 1 cycle)
  117.  
  118.         @@ at this point, R0=b0,  R1=b1, R2 (free), R3=ROWr32[2], R4=ROWr32[3],
  119.         @@     R5=b2, R6=ROWr16[0], R7=b3, R8=W1, R9=W3, R10=W5, R11=W7,
  120.         @@     R12=__const_ptr_, R14=&block[n]
  121.         @@ temp = ((uint32_t*)row)[2] | ((uint32_t*)row)[3];
  122.         @@ if (temp != 0) {}
  123.         orrs r2, r3, r4          @ R2=ROWr32[2] | ROWr32[3]
  124.         beq __end_b_evaluation
  125.  
  126.         @@ at this point, R0=b0,  R1=b1, R2 (free), R3=ROWr32[2], R4=ROWr32[3],
  127.         @@     R5=b2, R6=ROWr16[0], R7=b3, R8=W1, R9=W3, R10=W5, R11=W7,
  128.         @@     R12=__const_ptr_, R14=&block[n]
  129.         @@ MAC16(b0, W5, row[5]);
  130.         @@ MAC16(b2, W7, row[5]);
  131.         @@ MAC16(b3, W3, row[5]);
  132.         @@ MAC16(b1, -W1, row[5]);
  133.         @@ MAC16(b0, W7, row[7]);
  134.         @@ MAC16(b2, W3, row[7]);
  135.         @@ MAC16(b3, -W1, row[7]);
  136.         @@ MAC16(b1, -W5, row[7]);
  137.         mov r3, r3, asr #16      @ R3=ROWr16[5]
  138.         teq r3, #0               @ if null avoid muls
  139.         it    ne
  140.         mlane r0, r10, r3, r0    @ R0+=W5*ROWr16[5]=b0
  141.         mov r4, r4, asr #16      @ R4=ROWr16[7]
  142.         itttt ne
  143.         mlane r5, r11, r3, r5    @ R5+=W7*ROWr16[5]=b2
  144.         mlane r7, r9, r3, r7     @ R7+=W3*ROWr16[5]=b3
  145.         rsbne r3, r3, #0         @ R3=-ROWr16[5]
  146.         mlane r1, r8, r3, r1     @ R7-=W1*ROWr16[5]=b1
  147.         @@ R3 is free now
  148.         teq r4, #0               @ if null avoid muls
  149.         itttt ne
  150.         mlane r0, r11, r4, r0    @ R0+=W7*ROWr16[7]=b0
  151.         mlane r5, r9, r4, r5     @ R5+=W3*ROWr16[7]=b2
  152.         rsbne r4, r4, #0         @ R4=-ROWr16[7]
  153.         mlane r7, r8, r4, r7     @ R7-=W1*ROWr16[7]=b3
  154.         it    ne
  155.         mlane r1, r10, r4, r1    @ R1-=W5*ROWr16[7]=b1
  156.         @@ R4 is free now
  157. __end_b_evaluation:
  158.         @@ at this point, R0=b0,  R1=b1, R2=ROWr32[2] | ROWr32[3] (tmp), R3 (free), R4 (free),
  159.         @@     R5=b2, R6=ROWr16[0], R7=b3, R8 (free), R9 (free), R10 (free), R11 (free),
  160.         @@     R12=__const_ptr_, R14=&block[n]
  161.  
  162. @@ __a_evaluation:
  163.         @@ a0 = (W4 * row[0]) + (1 << (ROW_SHIFT - 1));
  164.         @@ a1 = a0 + W6 * row[2];
  165.         @@ a2 = a0 - W6 * row[2];
  166.         @@ a3 = a0 - W2 * row[2];
  167.         @@ a0 = a0 + W2 * row[2];
  168.         ldr r9, =W4              @ R9=W4
  169.         mul r6, r9, r6           @ R6=W4*ROWr16[0]
  170.         ldr r10, =W6             @ R10=W6
  171.         ldrsh r4, [r14, #4]      @ R4=ROWr16[2] (a3 not defined yet)
  172.         add r6, r6, #ROW_SHIFTED_1 @ R6=W4*ROWr16[0] + 1<<(ROW_SHIFT-1) (a0)
  173.  
  174.         mul r11, r10, r4         @ R11=W6*ROWr16[2]
  175.         ldr r8, =W2              @ R8=W2
  176.         sub r3, r6, r11          @ R3=a0-W6*ROWr16[2] (a2)
  177.         @@ temp = ((uint32_t*)row)[2] | ((uint32_t*)row)[3];
  178.         @@ if (temp != 0) {}
  179.         teq r2, #0
  180.         beq __end_bef_a_evaluation
  181.  
  182.         add r2, r6, r11          @ R2=a0+W6*ROWr16[2] (a1)
  183.         mul r11, r8, r4          @ R11=W2*ROWr16[2]
  184.         sub r4, r6, r11          @ R4=a0-W2*ROWr16[2] (a3)
  185.         add r6, r6, r11          @ R6=a0+W2*ROWr16[2] (a0)
  186.  
  187.  
  188.         @@ at this point, R0=b0,  R1=b1, R2=a1, R3=a2, R4=a3,
  189.         @@     R5=b2, R6=a0, R7=b3, R8=W2, R9=W4, R10=W6, R11 (free),
  190.         @@     R12=__const_ptr_, R14=&block[n]
  191.  
  192.  
  193.         @@ a0 += W4*row[4]
  194.         @@ a1 -= W4*row[4]
  195.         @@ a2 -= W4*row[4]
  196.         @@ a3 += W4*row[4]
  197.         ldrsh r11, [r14, #8]     @ R11=ROWr16[4]
  198.         teq r11, #0              @ if null avoid muls
  199.         it    ne
  200.         mulne r11, r9, r11       @ R11=W4*ROWr16[4]
  201.         @@ R9 is free now
  202.         ldrsh r9, [r14, #12]     @ R9=ROWr16[6]
  203.         itttt ne
  204.         addne r6, r6, r11        @ R6+=W4*ROWr16[4] (a0)
  205.         subne r2, r2, r11        @ R2-=W4*ROWr16[4] (a1)
  206.         subne r3, r3, r11        @ R3-=W4*ROWr16[4] (a2)
  207.         addne r4, r4, r11        @ R4+=W4*ROWr16[4] (a3)
  208.         @@ W6 alone is no more useful, save W2*ROWr16[6] in it instead
  209.         teq r9, #0               @ if null avoid muls
  210.         itttt ne
  211.         mulne r11, r10, r9       @ R11=W6*ROWr16[6]
  212.         addne r6, r6, r11        @ R6+=W6*ROWr16[6] (a0)
  213.         mulne r10, r8, r9        @ R10=W2*ROWr16[6]
  214.         @@ a0 += W6*row[6];
  215.         @@ a3 -= W6*row[6];
  216.         @@ a1 -= W2*row[6];
  217.         @@ a2 += W2*row[6];
  218.         subne r4, r4, r11        @ R4-=W6*ROWr16[6] (a3)
  219.         itt   ne
  220.         subne r2, r2, r10        @ R2-=W2*ROWr16[6] (a1)
  221.         addne r3, r3, r10        @ R3+=W2*ROWr16[6] (a2)
  222.  
  223. __end_a_evaluation:
  224.         @@ at this point, R0=b0,  R1=b1, R2=a1, R3=a2, R4=a3,
  225.         @@     R5=b2, R6=a0, R7=b3, R8 (free), R9 (free), R10 (free), R11 (free),
  226.         @@     R12=__const_ptr_, R14=&block[n]
  227.         @@ row[0] = (a0 + b0) >> ROW_SHIFT;
  228.         @@ row[1] = (a1 + b1) >> ROW_SHIFT;
  229.         @@ row[2] = (a2 + b2) >> ROW_SHIFT;
  230.         @@ row[3] = (a3 + b3) >> ROW_SHIFT;
  231.         @@ row[4] = (a3 - b3) >> ROW_SHIFT;
  232.         @@ row[5] = (a2 - b2) >> ROW_SHIFT;
  233.         @@ row[6] = (a1 - b1) >> ROW_SHIFT;
  234.         @@ row[7] = (a0 - b0) >> ROW_SHIFT;
  235.         add r8, r6, r0           @ R8=a0+b0
  236.         add r9, r2, r1           @ R9=a1+b1
  237.         @@ put 2 16 bits half-words in a 32bits word
  238.         @@ ROWr32[0]=ROWr16[0] | (ROWr16[1]<<16) (only Little Endian compliant then!!!)
  239.         ldr r10, =MASK_MSHW      @ R10=0xFFFF0000
  240.         and r9, r10, r9, lsl #ROW_SHIFT2MSHW @ R9=0xFFFF0000 & ((a1+b1)<<5)
  241.         mvn r11, r10             @ R11= NOT R10= 0x0000FFFF
  242.         and r8, r11, r8, asr #ROW_SHIFT @ R8=0x0000FFFF & ((a0+b0)>>11)
  243.         orr r8, r8, r9
  244.         str r8, [r14, #0]
  245.  
  246.         add r8, r3, r5           @ R8=a2+b2
  247.         add r9, r4, r7           @ R9=a3+b3
  248.         and r9, r10, r9, lsl #ROW_SHIFT2MSHW @ R9=0xFFFF0000 & ((a3+b3)<<5)
  249.         and r8, r11, r8, asr #ROW_SHIFT @ R8=0x0000FFFF & ((a2+b2)>>11)
  250.         orr r8, r8, r9
  251.         str r8, [r14, #4]
  252.  
  253.         sub r8, r4, r7           @ R8=a3-b3
  254.         sub r9, r3, r5           @ R9=a2-b2
  255.         and r9, r10, r9, lsl #ROW_SHIFT2MSHW @ R9=0xFFFF0000 & ((a2-b2)<<5)
  256.         and r8, r11, r8, asr #ROW_SHIFT @ R8=0x0000FFFF & ((a3-b3)>>11)
  257.         orr r8, r8, r9
  258.         str r8, [r14, #8]
  259.  
  260.         sub r8, r2, r1           @ R8=a1-b1
  261.         sub r9, r6, r0           @ R9=a0-b0
  262.         and r9, r10, r9, lsl #ROW_SHIFT2MSHW @ R9=0xFFFF0000 & ((a0-b0)<<5)
  263.         and r8, r11, r8, asr #ROW_SHIFT @ R8=0x0000FFFF & ((a1-b1)>>11)
  264.         orr r8, r8, r9
  265.         str r8, [r14, #12]
  266.  
  267.         bal __end_row_loop
  268.  
  269. __almost_empty_row:
  270.         @@ the row was empty, except ROWr16[0], now, management of this special case
  271.         @@ at this point, R0=block, R14=&block[n], R12=__const_ptr_, R1=ROWr32[0], R2=ROWr32[1],
  272.         @@                R3=ROWr32[2], R4=ROWr32[3], R5=(temp), R6=ROWr16[0], R7=ROWr16[1],
  273.         @@                R8=0xFFFF (temp), R9-R11 free
  274.         mov r8, #0x10000         @ R8=0xFFFF (2 steps needed!) it saves a ldr call (because of delay run).
  275.         sub r8, r8, #1           @ R8 is now ready.
  276.         and r5, r8, r6, lsl #3   @ R5=R8 & (R6<<3)= (ROWr16[0]<<3) & 0xFFFF
  277.         orr r5, r5, r5, lsl #16  @ R5=R5 | (R5<<16)
  278.         str r5, [r14, #0]        @ R14[0]=ROWr32[0]=R5
  279.         str r5, [r14, #4]        @ R14[4]=ROWr32[1]=R5
  280.         str r5, [r14, #8]        @ R14[8]=ROWr32[2]=R5
  281.         str r5, [r14, #12]       @ R14[12]=ROWr32[3]=R5
  282.  
  283. __end_row_loop:
  284.         @@ at this point, R0-R11 (free)
  285.         @@     R12=__const_ptr_, R14=&block[n]
  286.         ldr r0, [sp, #0]         @ R0=block
  287.         teq r0, r14              @ compare current &block[8*n] to block, when block is reached, the loop is finished.
  288.         sub r14, r14, #16
  289.         bne __row_loop
  290.  
  291.  
  292.  
  293.         @@ at this point, R0=block, R1-R11 (free)
  294.         @@     R12=__const_ptr_, R14=&block[n]
  295.         add r14, r0, #14        @ R14=&block[7], better start from the last col, and decrease the value until col=0, i.e. R14=block.
  296. __col_loop:
  297.  
  298. @@ __b_evaluation2:
  299.         @@ at this point, R0=block (temp),  R1-R11 (free)
  300.         @@     R12=__const_ptr_, R14=&block[n]
  301.         @@ proceed with b0-b3 first, followed by a0-a3
  302.         @@ MUL16(b0, W1, col[8x1]);
  303.         @@ MUL16(b1, W3, col[8x1]);
  304.         @@ MUL16(b2, W5, col[8x1]);
  305.         @@ MUL16(b3, W7, col[8x1]);
  306.         @@ MAC16(b0, W3, col[8x3]);
  307.         @@ MAC16(b1, -W7, col[8x3]);
  308.         @@ MAC16(b2, -W1, col[8x3]);
  309.         @@ MAC16(b3, -W5, col[8x3]);
  310.         ldr r8, =W1              @ R8=W1
  311.         ldrsh r7, [r14, #16]
  312.         mul r0, r8, r7           @ R0=W1*ROWr16[1]=b0 (ROWr16[1] must be the second arg, to have the possibility to save 1 cycle)
  313.         ldr r9, =W3              @ R9=W3
  314.         ldr r10, =W5             @ R10=W5
  315.         mul r1, r9, r7           @ R1=W3*ROWr16[1]=b1 (ROWr16[1] must be the second arg, to have the possibility to save 1 cycle)
  316.         ldr r11, =W7             @ R11=W7
  317.         mul r5, r10, r7          @ R5=W5*ROWr16[1]=b2 (ROWr16[1] must be the second arg, to have the possibility to save 1 cycle)
  318.         ldrsh r2, [r14, #48]
  319.         mul r7, r11, r7          @ R7=W7*ROWr16[1]=b3 (ROWr16[1] must be the second arg, to have the possibility to save 1 cycle)
  320.         teq r2, #0               @ if 0, then avoid muls
  321.         itttt ne
  322.         mlane r0, r9, r2, r0     @ R0+=W3*ROWr16[3]=b0 (ROWr16[3] must be the second arg, to have the possibility to save 1 cycle)
  323.         rsbne r2, r2, #0         @ R2=-ROWr16[3]
  324.         mlane r1, r11, r2, r1    @ R1-=W7*ROWr16[3]=b1 (ROWr16[3] must be the second arg, to have the possibility to save 1 cycle)
  325.         mlane r5, r8, r2, r5     @ R5-=W1*ROWr16[3]=b2 (ROWr16[3] must be the second arg, to have the possibility to save 1 cycle)
  326.         it    ne
  327.         mlane r7, r10, r2, r7    @ R7-=W5*ROWr16[3]=b3 (ROWr16[3] must be the second arg, to have the possibility to save 1 cycle)
  328.  
  329.         @@ at this point, R0=b0,  R1=b1, R2 (free), R3 (free), R4 (free),
  330.         @@     R5=b2, R6 (free), R7=b3, R8=W1, R9=W3, R10=W5, R11=W7,
  331.         @@     R12=__const_ptr_, R14=&block[n]
  332.         @@ MAC16(b0, W5, col[5x8]);
  333.         @@ MAC16(b2, W7, col[5x8]);
  334.         @@ MAC16(b3, W3, col[5x8]);
  335.         @@ MAC16(b1, -W1, col[5x8]);
  336.         @@ MAC16(b0, W7, col[7x8]);
  337.         @@ MAC16(b2, W3, col[7x8]);
  338.         @@ MAC16(b3, -W1, col[7x8]);
  339.         @@ MAC16(b1, -W5, col[7x8]);
  340.         ldrsh r3, [r14, #80]     @ R3=COLr16[5x8]
  341.         teq r3, #0               @ if 0 then avoid muls
  342.         itttt ne
  343.         mlane r0, r10, r3, r0    @ R0+=W5*ROWr16[5x8]=b0
  344.         mlane r5, r11, r3, r5    @ R5+=W7*ROWr16[5x8]=b2
  345.         mlane r7, r9, r3, r7     @ R7+=W3*ROWr16[5x8]=b3
  346.         rsbne r3, r3, #0         @ R3=-ROWr16[5x8]
  347.         ldrsh r4, [r14, #112]    @ R4=COLr16[7x8]
  348.         it    ne
  349.         mlane r1, r8, r3, r1     @ R7-=W1*ROWr16[5x8]=b1
  350.         @@ R3 is free now
  351.         teq r4, #0               @ if 0 then avoid muls
  352.         itttt ne
  353.         mlane r0, r11, r4, r0    @ R0+=W7*ROWr16[7x8]=b0
  354.         mlane r5, r9, r4, r5     @ R5+=W3*ROWr16[7x8]=b2
  355.         rsbne r4, r4, #0         @ R4=-ROWr16[7x8]
  356.         mlane r7, r8, r4, r7     @ R7-=W1*ROWr16[7x8]=b3
  357.         it    ne
  358.         mlane r1, r10, r4, r1    @ R1-=W5*ROWr16[7x8]=b1
  359.         @@ R4 is free now
  360. @@ __end_b_evaluation2:
  361.         @@ at this point, R0=b0,  R1=b1, R2 (free), R3 (free), R4 (free),
  362.         @@     R5=b2, R6 (free), R7=b3, R8 (free), R9 (free), R10 (free), R11 (free),
  363.         @@     R12=__const_ptr_, R14=&block[n]
  364.  
  365. @@ __a_evaluation2:
  366.         @@ a0 = (W4 * col[8x0]) + (1 << (COL_SHIFT - 1));
  367.         @@ a1 = a0 + W6 * row[2];
  368.         @@ a2 = a0 - W6 * row[2];
  369.         @@ a3 = a0 - W2 * row[2];
  370.         @@ a0 = a0 + W2 * row[2];
  371.         ldrsh r6, [r14, #0]
  372.         ldr r9, =W4              @ R9=W4
  373.         mul r6, r9, r6           @ R6=W4*ROWr16[0]
  374.         ldr r10, =W6             @ R10=W6
  375.         ldrsh r4, [r14, #32]     @ R4=ROWr16[2] (a3 not defined yet)
  376.         add r6, r6, #COL_SHIFTED_1 @ R6=W4*ROWr16[0] + 1<<(COL_SHIFT-1) (a0)
  377.         mul r11, r10, r4         @ R11=W6*ROWr16[2]
  378.         ldr r8, =W2              @ R8=W2
  379.         add r2, r6, r11          @ R2=a0+W6*ROWr16[2] (a1)
  380.         sub r3, r6, r11          @ R3=a0-W6*ROWr16[2] (a2)
  381.         mul r11, r8, r4          @ R11=W2*ROWr16[2]
  382.         sub r4, r6, r11          @ R4=a0-W2*ROWr16[2] (a3)
  383.         add r6, r6, r11          @ R6=a0+W2*ROWr16[2] (a0)
  384.  
  385.         @@ at this point, R0=b0,  R1=b1, R2=a1, R3=a2, R4=a3,
  386.         @@     R5=b2, R6=a0, R7=b3, R8=W2, R9=W4, R10=W6, R11 (free),
  387.         @@     R12=__const_ptr_, R14=&block[n]
  388.         @@ a0 += W4*row[4]
  389.         @@ a1 -= W4*row[4]
  390.         @@ a2 -= W4*row[4]
  391.         @@ a3 += W4*row[4]
  392.         ldrsh r11, [r14, #64]    @ R11=ROWr16[4]
  393.         teq r11, #0              @ if null avoid muls
  394.         itttt ne
  395.         mulne r11, r9, r11       @ R11=W4*ROWr16[4]
  396.         @@ R9 is free now
  397.         addne r6, r6, r11        @ R6+=W4*ROWr16[4] (a0)
  398.         subne r2, r2, r11        @ R2-=W4*ROWr16[4] (a1)
  399.         subne r3, r3, r11        @ R3-=W4*ROWr16[4] (a2)
  400.         ldrsh r9, [r14, #96]     @ R9=ROWr16[6]
  401.         it    ne
  402.         addne r4, r4, r11        @ R4+=W4*ROWr16[4] (a3)
  403.         @@ W6 alone is no more useful, save W2*ROWr16[6] in it instead
  404.         teq r9, #0               @ if null avoid muls
  405.         itttt ne
  406.         mulne r11, r10, r9       @ R11=W6*ROWr16[6]
  407.         addne r6, r6, r11        @ R6+=W6*ROWr16[6] (a0)
  408.         mulne r10, r8, r9        @ R10=W2*ROWr16[6]
  409.         @@ a0 += W6*row[6];
  410.         @@ a3 -= W6*row[6];
  411.         @@ a1 -= W2*row[6];
  412.         @@ a2 += W2*row[6];
  413.         subne r4, r4, r11        @ R4-=W6*ROWr16[6] (a3)
  414.         itt   ne
  415.         subne r2, r2, r10        @ R2-=W2*ROWr16[6] (a1)
  416.         addne r3, r3, r10        @ R3+=W2*ROWr16[6] (a2)
  417. @@ __end_a_evaluation2:
  418.         @@ at this point, R0=b0,  R1=b1, R2=a1, R3=a2, R4=a3,
  419.         @@     R5=b2, R6=a0, R7=b3, R8 (free), R9 (free), R10 (free), R11 (free),
  420.         @@     R12=__const_ptr_, R14=&block[n]
  421.         @@ col[0 ] = ((a0 + b0) >> COL_SHIFT);
  422.         @@ col[8 ] = ((a1 + b1) >> COL_SHIFT);
  423.         @@ col[16] = ((a2 + b2) >> COL_SHIFT);
  424.         @@ col[24] = ((a3 + b3) >> COL_SHIFT);
  425.         @@ col[32] = ((a3 - b3) >> COL_SHIFT);
  426.         @@ col[40] = ((a2 - b2) >> COL_SHIFT);
  427.         @@ col[48] = ((a1 - b1) >> COL_SHIFT);
  428.         @@ col[56] = ((a0 - b0) >> COL_SHIFT);
  429.         @@@@@ no optimization here @@@@@
  430.         add r8, r6, r0           @ R8=a0+b0
  431.         add r9, r2, r1           @ R9=a1+b1
  432.         mov r8, r8, asr #COL_SHIFT
  433.         mov r9, r9, asr #COL_SHIFT
  434.         strh r8, [r14, #0]
  435.         strh r9, [r14, #16]
  436.         add r8, r3, r5           @ R8=a2+b2
  437.         add r9, r4, r7           @ R9=a3+b3
  438.         mov r8, r8, asr #COL_SHIFT
  439.         mov r9, r9, asr #COL_SHIFT
  440.         strh r8, [r14, #32]
  441.         strh r9, [r14, #48]
  442.         sub r8, r4, r7           @ R8=a3-b3
  443.         sub r9, r3, r5           @ R9=a2-b2
  444.         mov r8, r8, asr #COL_SHIFT
  445.         mov r9, r9, asr #COL_SHIFT
  446.         strh r8, [r14, #64]
  447.         strh r9, [r14, #80]
  448.         sub r8, r2, r1           @ R8=a1-b1
  449.         sub r9, r6, r0           @ R9=a0-b0
  450.         mov r8, r8, asr #COL_SHIFT
  451.         mov r9, r9, asr #COL_SHIFT
  452.         strh r8, [r14, #96]
  453.         strh r9, [r14, #112]
  454.  
  455. @@ __end_col_loop:
  456.         @@ at this point, R0-R11 (free)
  457.         @@     R12=__const_ptr_, R14=&block[n]
  458.         ldr r0, [sp, #0]         @ R0=block
  459.         teq r0, r14              @ compare current &block[n] to block, when block is reached, the loop is finished.
  460.         sub r14, r14, #2
  461.         bne __col_loop
  462.  
  463.  
  464.  
  465.  
  466. @@ __end_simple_idct_arm:
  467.         @@ restore registers to previous status!
  468.         add sp, sp, #8 @@ the local variables!
  469.         ldmfd sp!, {r4-r11, r15} @@ update PC with LR content.
  470.  
  471.  
  472.  
  473. @@ kind of sub-function, here not to overload the common case.
  474. __end_bef_a_evaluation:
  475.         add r2, r6, r11          @ R2=a0+W6*ROWr16[2] (a1)
  476.         mul r11, r8, r4          @ R11=W2*ROWr16[2]
  477.         sub r4, r6, r11          @ R4=a0-W2*ROWr16[2] (a3)
  478.         add r6, r6, r11          @ R6=a0+W2*ROWr16[2] (a0)
  479.         bal __end_a_evaluation
  480.