michael@0: ;
michael@0: ;  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
michael@0: ;
michael@0: ;  Use of this source code is governed by a BSD-style license
michael@0: ;  that can be found in the LICENSE file in the root of the source
michael@0: ;  tree. An additional intellectual property rights grant can be found
michael@0: ;  in the file PATENTS.  All contributing project authors may
michael@0: ;  be found in the AUTHORS file in the root of the source tree.
michael@0: ;
michael@0: 
michael@0: 
michael@0:     EXPORT |vp8_start_encode|
michael@0:     EXPORT |vp8_encode_bool|
michael@0:     EXPORT |vp8_stop_encode|
michael@0:     EXPORT |vp8_encode_value|
michael@0:     IMPORT |vp8_validate_buffer_arm|
michael@0: 
michael@0:     INCLUDE vp8_asm_enc_offsets.asm
michael@0: 
michael@0:     ARM
michael@0:     REQUIRE8
michael@0:     PRESERVE8
michael@0: 
michael@0:     AREA    |.text|, CODE, READONLY
michael@0: 
michael@0:     ; macro for validating write buffer position
michael@0:     ; needs vp8_writer in r0
michael@0:     ; start shall not be in r1
michael@0:     MACRO
michael@0:     VALIDATE_POS $start, $pos
michael@0:     push {r0-r3, r12, lr}        ; rest of regs are preserved by subroutine call
michael@0:     ldr  r2, [r0, #vp8_writer_buffer_end]
michael@0:     ldr  r3, [r0, #vp8_writer_error]
michael@0:     mov  r1, $pos
michael@0:     mov  r0, $start
michael@0:     bl   vp8_validate_buffer_arm
michael@0:     pop  {r0-r3, r12, lr}
michael@0:     MEND
michael@0: 
michael@0: ; r0 BOOL_CODER *br
michael@0: ; r1 unsigned char *source
michael@0: ; r2 unsigned char *source_end
michael@0: |vp8_start_encode| PROC
michael@0:     str     r2,  [r0, #vp8_writer_buffer_end]
michael@0:     mov     r12, #0
michael@0:     mov     r3,  #255
michael@0:     mvn     r2,  #23
michael@0:     str     r12, [r0, #vp8_writer_lowvalue]
michael@0:     str     r3,  [r0, #vp8_writer_range]
michael@0:     str     r2,  [r0, #vp8_writer_count]
michael@0:     str     r12, [r0, #vp8_writer_pos]
michael@0:     str     r1,  [r0, #vp8_writer_buffer]
michael@0:     bx      lr
michael@0:     ENDP
michael@0: 
michael@0: ; r0 BOOL_CODER *br
michael@0: ; r1 int bit
michael@0: ; r2 int probability
michael@0: |vp8_encode_bool| PROC
michael@0:     push    {r4-r10, lr}
michael@0: 
michael@0:     mov     r4, r2
michael@0: 
michael@0:     ldr     r2, [r0, #vp8_writer_lowvalue]
michael@0:     ldr     r5, [r0, #vp8_writer_range]
michael@0:     ldr     r3, [r0, #vp8_writer_count]
michael@0: 
michael@0:     sub     r7, r5, #1                  ; range-1
michael@0: 
michael@0:     cmp     r1, #0
michael@0:     mul     r6, r4, r7                  ; ((range-1) * probability)
michael@0: 
michael@0:     mov     r7, #1
michael@0:     add     r4, r7, r6, lsr #8          ; 1 + (((range-1) * probability) >> 8)
michael@0: 
michael@0:     addne   r2, r2, r4                  ; if  (bit) lowvalue += split
michael@0:     subne   r4, r5, r4                  ; if  (bit) range = range-split
michael@0: 
michael@0:     ; Counting the leading zeros is used to normalize range.
michael@0:     clz     r6, r4
michael@0:     sub     r6, r6, #24                 ; shift
michael@0: 
michael@0:     ; Flag is set on the sum of count.  This flag is used later
michael@0:     ; to determine if count >= 0
michael@0:     adds    r3, r3, r6                  ; count += shift
michael@0:     lsl     r5, r4, r6                  ; range <<= shift
michael@0:     bmi     token_count_lt_zero         ; if(count >= 0)
michael@0: 
michael@0:     sub     r6, r6, r3                  ; offset = shift - count
michael@0:     sub     r4, r6, #1                  ; offset-1
michael@0:     lsls    r4, r2, r4                  ; if((lowvalue<<(offset-1)) & 0x80000000 )
michael@0:     bpl     token_high_bit_not_set
michael@0: 
michael@0:     ldr     r4, [r0, #vp8_writer_pos]   ; x
michael@0:     sub     r4, r4, #1                  ; x = w->pos-1
michael@0:     b       token_zero_while_start
michael@0: token_zero_while_loop
michael@0:     mov     r9, #0
michael@0:     strb    r9, [r7, r4]                ; w->buffer[x] =(unsigned char)0
michael@0:     sub     r4, r4, #1                  ; x--
michael@0: token_zero_while_start
michael@0:     cmp     r4, #0
michael@0:     ldrge   r7, [r0, #vp8_writer_buffer]
michael@0:     ldrb    r1, [r7, r4]
michael@0:     cmpge   r1, #0xff
michael@0:     beq     token_zero_while_loop
michael@0: 
michael@0:     ldr     r7, [r0, #vp8_writer_buffer]
michael@0:     ldrb    r9, [r7, r4]                ; w->buffer[x]
michael@0:     add     r9, r9, #1
michael@0:     strb    r9, [r7, r4]                ; w->buffer[x] + 1
michael@0: token_high_bit_not_set
michael@0:     rsb     r4, r6, #24                 ; 24-offset
michael@0:     ldr     r9, [r0, #vp8_writer_buffer]
michael@0:     lsr     r7, r2, r4                  ; lowvalue >> (24-offset)
michael@0:     ldr     r4, [r0, #vp8_writer_pos]   ; w->pos
michael@0:     lsl     r2, r2, r6                  ; lowvalue <<= offset
michael@0:     mov     r6, r3                      ; shift = count
michael@0:     add     r1, r4, #1                  ; w->pos++
michael@0:     bic     r2, r2, #0xff000000         ; lowvalue &= 0xffffff
michael@0:     str     r1, [r0, #vp8_writer_pos]
michael@0:     sub     r3, r3, #8                  ; count -= 8
michael@0: 
michael@0:     VALIDATE_POS r9, r1                 ; validate_buffer at pos
michael@0: 
michael@0:     strb    r7, [r9, r4]                ; w->buffer[w->pos++]
michael@0: 
michael@0: token_count_lt_zero
michael@0:     lsl     r2, r2, r6                  ; lowvalue <<= shift
michael@0: 
michael@0:     str     r2, [r0, #vp8_writer_lowvalue]
michael@0:     str     r5, [r0, #vp8_writer_range]
michael@0:     str     r3, [r0, #vp8_writer_count]
michael@0:     pop     {r4-r10, pc}
michael@0:     ENDP
michael@0: 
michael@0: ; r0 BOOL_CODER *br
michael@0: |vp8_stop_encode| PROC
michael@0:     push    {r4-r10, lr}
michael@0: 
michael@0:     ldr     r2, [r0, #vp8_writer_lowvalue]
michael@0:     ldr     r5, [r0, #vp8_writer_range]
michael@0:     ldr     r3, [r0, #vp8_writer_count]
michael@0: 
michael@0:     mov     r10, #32
michael@0: 
michael@0: stop_encode_loop
michael@0:     sub     r7, r5, #1                  ; range-1
michael@0: 
michael@0:     mov     r4, r7, lsl #7              ; ((range-1) * 128)
michael@0: 
michael@0:     mov     r7, #1
michael@0:     add     r4, r7, r4, lsr #8          ; 1 + (((range-1) * 128) >> 8)
michael@0: 
michael@0:     ; Counting the leading zeros is used to normalize range.
michael@0:     clz     r6, r4
michael@0:     sub     r6, r6, #24                 ; shift
michael@0: 
michael@0:     ; Flag is set on the sum of count.  This flag is used later
michael@0:     ; to determine if count >= 0
michael@0:     adds    r3, r3, r6                  ; count += shift
michael@0:     lsl     r5, r4, r6                  ; range <<= shift
michael@0:     bmi     token_count_lt_zero_se      ; if(count >= 0)
michael@0: 
michael@0:     sub     r6, r6, r3                  ; offset = shift - count
michael@0:     sub     r4, r6, #1                  ; offset-1
michael@0:     lsls    r4, r2, r4                  ; if((lowvalue<<(offset-1)) & 0x80000000 )
michael@0:     bpl     token_high_bit_not_set_se
michael@0: 
michael@0:     ldr     r4, [r0, #vp8_writer_pos]   ; x
michael@0:     sub     r4, r4, #1                  ; x = w->pos-1
michael@0:     b       token_zero_while_start_se
michael@0: token_zero_while_loop_se
michael@0:     mov     r9, #0
michael@0:     strb    r9, [r7, r4]                ; w->buffer[x] =(unsigned char)0
michael@0:     sub     r4, r4, #1                  ; x--
michael@0: token_zero_while_start_se
michael@0:     cmp     r4, #0
michael@0:     ldrge   r7, [r0, #vp8_writer_buffer]
michael@0:     ldrb    r1, [r7, r4]
michael@0:     cmpge   r1, #0xff
michael@0:     beq     token_zero_while_loop_se
michael@0: 
michael@0:     ldr     r7, [r0, #vp8_writer_buffer]
michael@0:     ldrb    r9, [r7, r4]                ; w->buffer[x]
michael@0:     add     r9, r9, #1
michael@0:     strb    r9, [r7, r4]                ; w->buffer[x] + 1
michael@0: token_high_bit_not_set_se
michael@0:     rsb     r4, r6, #24                 ; 24-offset
michael@0:     ldr     r9, [r0, #vp8_writer_buffer]
michael@0:     lsr     r7, r2, r4                  ; lowvalue >> (24-offset)
michael@0:     ldr     r4, [r0, #vp8_writer_pos]   ; w->pos
michael@0:     lsl     r2, r2, r6                  ; lowvalue <<= offset
michael@0:     mov     r6, r3                      ; shift = count
michael@0:     add     r1, r4, #1                  ; w->pos++
michael@0:     bic     r2, r2, #0xff000000         ; lowvalue &= 0xffffff
michael@0:     str     r1, [r0, #vp8_writer_pos]
michael@0:     sub     r3, r3, #8                  ; count -= 8
michael@0: 
michael@0:     VALIDATE_POS r9, r1                 ; validate_buffer at pos
michael@0: 
michael@0:     strb    r7, [r9, r4]                ; w->buffer[w->pos++]
michael@0: 
michael@0: token_count_lt_zero_se
michael@0:     lsl     r2, r2, r6                  ; lowvalue <<= shift
michael@0: 
michael@0:     subs    r10, r10, #1
michael@0:     bne     stop_encode_loop
michael@0: 
michael@0:     str     r2, [r0, #vp8_writer_lowvalue]
michael@0:     str     r5, [r0, #vp8_writer_range]
michael@0:     str     r3, [r0, #vp8_writer_count]
michael@0:     pop     {r4-r10, pc}
michael@0: 
michael@0:     ENDP
michael@0: 
michael@0: ; r0 BOOL_CODER *br
michael@0: ; r1 int data
michael@0: ; r2 int bits
michael@0: |vp8_encode_value| PROC
michael@0:     push    {r4-r12, lr}
michael@0: 
michael@0:     mov     r10, r2
michael@0: 
michael@0:     ldr     r2, [r0, #vp8_writer_lowvalue]
michael@0:     ldr     r5, [r0, #vp8_writer_range]
michael@0:     ldr     r3, [r0, #vp8_writer_count]
michael@0: 
michael@0:     rsb     r4, r10, #32                 ; 32-n
michael@0: 
michael@0:     ; v is kept in r1 during the token pack loop
michael@0:     lsl     r1, r1, r4                  ; r1 = v << 32 - n
michael@0: 
michael@0: encode_value_loop
michael@0:     sub     r7, r5, #1                  ; range-1
michael@0: 
michael@0:     ; Decisions are made based on the bit value shifted
michael@0:     ; off of v, so set a flag here based on this.
michael@0:     ; This value is refered to as "bb"
michael@0:     lsls    r1, r1, #1                  ; bit = v >> n
michael@0:     mov     r4, r7, lsl #7              ; ((range-1) * 128)
michael@0: 
michael@0:     mov     r7, #1
michael@0:     add     r4, r7, r4, lsr #8          ; 1 + (((range-1) * 128) >> 8)
michael@0: 
michael@0:     addcs   r2, r2, r4                  ; if  (bit) lowvalue += split
michael@0:     subcs   r4, r5, r4                  ; if  (bit) range = range-split
michael@0: 
michael@0:     ; Counting the leading zeros is used to normalize range.
michael@0:     clz     r6, r4
michael@0:     sub     r6, r6, #24                 ; shift
michael@0: 
michael@0:     ; Flag is set on the sum of count.  This flag is used later
michael@0:     ; to determine if count >= 0
michael@0:     adds    r3, r3, r6                  ; count += shift
michael@0:     lsl     r5, r4, r6                  ; range <<= shift
michael@0:     bmi     token_count_lt_zero_ev      ; if(count >= 0)
michael@0: 
michael@0:     sub     r6, r6, r3                  ; offset = shift - count
michael@0:     sub     r4, r6, #1                  ; offset-1
michael@0:     lsls    r4, r2, r4                  ; if((lowvalue<<(offset-1)) & 0x80000000 )
michael@0:     bpl     token_high_bit_not_set_ev
michael@0: 
michael@0:     ldr     r4, [r0, #vp8_writer_pos]   ; x
michael@0:     sub     r4, r4, #1                  ; x = w->pos-1
michael@0:     b       token_zero_while_start_ev
michael@0: token_zero_while_loop_ev
michael@0:     mov     r9, #0
michael@0:     strb    r9, [r7, r4]                ; w->buffer[x] =(unsigned char)0
michael@0:     sub     r4, r4, #1                  ; x--
michael@0: token_zero_while_start_ev
michael@0:     cmp     r4, #0
michael@0:     ldrge   r7, [r0, #vp8_writer_buffer]
michael@0:     ldrb    r11, [r7, r4]
michael@0:     cmpge   r11, #0xff
michael@0:     beq     token_zero_while_loop_ev
michael@0: 
michael@0:     ldr     r7, [r0, #vp8_writer_buffer]
michael@0:     ldrb    r9, [r7, r4]                ; w->buffer[x]
michael@0:     add     r9, r9, #1
michael@0:     strb    r9, [r7, r4]                ; w->buffer[x] + 1
michael@0: token_high_bit_not_set_ev
michael@0:     rsb     r4, r6, #24                 ; 24-offset
michael@0:     ldr     r9, [r0, #vp8_writer_buffer]
michael@0:     lsr     r7, r2, r4                  ; lowvalue >> (24-offset)
michael@0:     ldr     r4, [r0, #vp8_writer_pos]   ; w->pos
michael@0:     lsl     r2, r2, r6                  ; lowvalue <<= offset
michael@0:     mov     r6, r3                      ; shift = count
michael@0:     add     r11, r4, #1                 ; w->pos++
michael@0:     bic     r2, r2, #0xff000000         ; lowvalue &= 0xffffff
michael@0:     str     r11, [r0, #vp8_writer_pos]
michael@0:     sub     r3, r3, #8                  ; count -= 8
michael@0: 
michael@0:     VALIDATE_POS r9, r11                ; validate_buffer at pos
michael@0: 
michael@0:     strb    r7, [r9, r4]                ; w->buffer[w->pos++]
michael@0: 
michael@0: token_count_lt_zero_ev
michael@0:     lsl     r2, r2, r6                  ; lowvalue <<= shift
michael@0: 
michael@0:     subs    r10, r10, #1
michael@0:     bne     encode_value_loop
michael@0: 
michael@0:     str     r2, [r0, #vp8_writer_lowvalue]
michael@0:     str     r5, [r0, #vp8_writer_range]
michael@0:     str     r3, [r0, #vp8_writer_count]
michael@0:     pop     {r4-r12, pc}
michael@0:     ENDP
michael@0: 
michael@0:     END