2 ; Copyright (c) 2011 The WebM project authors. All Rights Reserved.
4 ; Use of this source code is governed by a BSD-style license
5 ; that can be found in the LICENSE file in the root of the source
6 ; tree. An additional intellectual property rights grant can be found
7 ; in the file PATENTS. All contributing project authors may
8 ; be found in the AUTHORS file in the root of the source tree.
12 EXPORT |vp8_fast_quantize_b_armv6|
14 INCLUDE vp8_asm_enc_offsets.asm
20 AREA ||.text||, CODE, READONLY, ALIGN=2
24 |vp8_fast_quantize_b_armv6| PROC
25 stmfd sp!, {r1, r4-r11, lr}
27 ldr r3, [r0, #vp8_block_coeff] ; coeff
28 ldr r4, [r0, #vp8_block_quant_fast] ; quant_fast
29 ldr r5, [r0, #vp8_block_round] ; round
30 ldr r6, [r1, #vp8_blockd_qcoeff] ; qcoeff
31 ldr r7, [r1, #vp8_blockd_dqcoeff] ; dqcoeff
32 ldr r8, [r1, #vp8_blockd_dequant] ; dequant
34 ldr r2, loop_count ; loop_count=0x1000000. 'lsls' instruction
35 ; is used to update the counter so that
36 ; it can be used to mark nonzero
37 ; quantized coefficient pairs.
39 mov r1, #0 ; flags for quantized coeffs
41 ; PART 1: quantization and dequantization loop
43 ldr r9, [r3], #4 ; [z1 | z0]
44 ldr r10, [r5], #4 ; [r1 | r0]
45 ldr r11, [r4], #4 ; [q1 | q0]
47 ssat16 lr, #1, r9 ; [sz1 | sz0]
48 eor r9, r9, lr ; [z1 ^ sz1 | z0 ^ sz0]
49 ssub16 r9, r9, lr ; x = (z ^ sz) - sz
50 sadd16 r9, r9, r10 ; [x1+r1 | x0+r0]
52 ldr r12, [r3], #4 ; [z3 | z2]
54 smulbb r0, r9, r11 ; [(x0+r0)*q0]
55 smultt r9, r9, r11 ; [(x1+r1)*q1]
57 ldr r10, [r5], #4 ; [r3 | r2]
59 ssat16 r11, #1, r12 ; [sz3 | sz2]
60 eor r12, r12, r11 ; [z3 ^ sz3 | z2 ^ sz2]
61 pkhtb r0, r9, r0, asr #16 ; [y1 | y0]
62 ldr r9, [r4], #4 ; [q3 | q2]
63 ssub16 r12, r12, r11 ; x = (z ^ sz) - sz
65 sadd16 r12, r12, r10 ; [x3+r3 | x2+r2]
67 eor r0, r0, lr ; [(y1 ^ sz1) | (y0 ^ sz0)]
69 smulbb r10, r12, r9 ; [(x2+r2)*q2]
70 smultt r12, r12, r9 ; [(x3+r3)*q3]
72 ssub16 r0, r0, lr ; x = (y ^ sz) - sz
74 cmp r0, #0 ; check if zero
75 orrne r1, r1, r2, lsr #24 ; add flag for nonzero coeffs
77 str r0, [r6], #4 ; *qcoeff++ = x
78 ldr r9, [r8], #4 ; [dq1 | dq0]
80 pkhtb r10, r12, r10, asr #16 ; [y3 | y2]
81 eor r10, r10, r11 ; [(y3 ^ sz3) | (y2 ^ sz2)]
82 ssub16 r10, r10, r11 ; x = (y ^ sz) - sz
84 cmp r10, #0 ; check if zero
85 orrne r1, r1, r2, lsr #23 ; add flag for nonzero coeffs
87 str r10, [r6], #4 ; *qcoeff++ = x
88 ldr r11, [r8], #4 ; [dq3 | dq2]
90 smulbb r12, r0, r9 ; [x0*dq0]
91 smultt r0, r0, r9 ; [x1*dq1]
93 smulbb r9, r10, r11 ; [x2*dq2]
94 smultt r10, r10, r11 ; [x3*dq3]
96 lsls r2, r2, #2 ; update loop counter
97 strh r12, [r7, #0] ; dqcoeff[0] = [x0*dq0]
98 strh r0, [r7, #2] ; dqcoeff[1] = [x1*dq1]
99 strh r9, [r7, #4] ; dqcoeff[2] = [x2*dq2]
100 strh r10, [r7, #6] ; dqcoeff[3] = [x3*dq3]
101 add r7, r7, #8 ; dqcoeff += 8
104 ; PART 2: check position for eob...
105 ldr r11, [sp, #0] ; restore BLOCKD pointer
106 mov lr, #0 ; init eob
107 cmp r1, #0 ; coeffs after quantization?
108 ldr r12, [r11, #vp8_blockd_eob]
109 beq end ; skip eob calculations if all zero
111 ldr r0, [r11, #vp8_blockd_qcoeff]
113 ; check shortcut for nonzero qcoeffs
115 bne quant_coeff_15_14
117 bne quant_coeff_13_11
131 ldrh r2, [r0, #30] ; rc=15, i=15
136 ldrh r3, [r0, #28] ; rc=14, i=14
142 ldrh r2, [r0, #22] ; rc=11, i=13
148 ldrh r3, [r0, #14] ; rc=7, i=12
153 ldrh r2, [r0, #20] ; rc=10, i=11
159 ldrh r3, [r0, #26] ; rc=13, i=10
164 ldrh r2, [r0, #24] ; rc=12, i=9
170 ldrh r3, [r0, #18] ; rc=9, i=8
175 ldrh r2, [r0, #12] ; rc=6, i=7
181 ldrh r3, [r0, #6] ; rc=3, i=6
186 ldrh r2, [r0, #4] ; rc=2, i=5
192 ldrh r3, [r0, #10] ; rc=5, i=4
197 ldrh r2, [r0, #16] ; rc=8, i=3
202 ldrh r3, [r0, #8] ; rc=4, i=2
208 ldrh r2, [r0, #2] ; rc=1, i=1
213 mov lr, #1 ; rc=0, i=0
217 ldmfd sp!, {r1, r4-r11, pc}