{
vp8_reset_mb_tokens_context(xd);
}
- else
+ else if (!vp8dx_bool_error(xd->current_bc))
{
eobtotal = vp8_decode_mb_tokens(pbi, xd);
}
{
vp8_build_inter_predictors_mb(xd);
}
-
/* When we have independent partitions we can apply residual even
* though other partitions within the frame are corrupt.
*/
{
vp8_reader *bool_decoder = &pbi->bc2;
int part_idx = 1;
+ int num_token_partitions;
TOKEN_PARTITION multi_token_partition =
(TOKEN_PARTITION)vp8_read_literal(&pbi->bc, 2);
+ if (!vp8dx_bool_error(&pbi->bc))
+ pbi->common.multi_token_partition = multi_token_partition;
+ num_token_partitions = 1 << pbi->common.multi_token_partition;
+ if (num_token_partitions + 1 > pbi->num_partitions)
+ vpx_internal_error(&pbi->common.error, VPX_CODEC_CORRUPT_FRAME,
+ "Partitions missing");
assert(vp8dx_bool_error(&pbi->bc) ||
multi_token_partition == pbi->common.multi_token_partition);
if (pbi->num_partitions > 2)
pc->show_frame = (data[0] >> 4) & 1;
first_partition_length_in_bytes =
(data[0] | (data[1] << 8) | (data[2] << 16)) >> 5;
- data += 3;
if (!pbi->ec_active && (data + first_partition_length_in_bytes > data_end
|| data + first_partition_length_in_bytes < data))
vpx_internal_error(&pc->error, VPX_CODEC_CORRUPT_FRAME,
"Truncated packet or corrupt partition 0 length");
+
+ data += 3;
+
vp8_setup_version(pc);
if (pc->frame_type == KEY_FRAME)
}
}
- if (pc->Width == 0 || pc->Height == 0)
+ if ((!pbi->decoded_key_frame && pc->frame_type != KEY_FRAME) ||
+ pc->Width == 0 || pc->Height == 0)
{
return -1;
}
/* Store a pointer to this partition and return. We haven't
* received the complete frame yet, so we will wait with decoding.
*/
+ assert(pbi->num_partitions < MAX_PARTITIONS);
pbi->partitions[pbi->num_partitions] = source;
pbi->partition_sizes[pbi->num_partitions] = size;
pbi->source_sz += size;
pbi->num_partitions++;
- if (pbi->num_partitions > (1<<pbi->common.multi_token_partition) + 1)
- pbi->common.multi_token_partition++;
- if (pbi->common.multi_token_partition > EIGHT_PARTITION)
+ if (pbi->num_partitions > (1 << EIGHT_PARTITION) + 1)
{
pbi->common.error.error_code = VPX_CODEC_UNSUP_BITSTREAM;
pbi->common.error.setjmp = 0;
+ pbi->num_partitions = 0;
return -1;
}
return 0;
pbi->Source = source;
pbi->source_sz = size;
}
+ else
+ {
+ assert(pbi->common.multi_token_partition <= EIGHT_PARTITION);
+ if (pbi->num_partitions == 0)
+ {
+ pbi->num_partitions = 1;
+ pbi->partitions[0] = NULL;
+ pbi->partition_sizes[0] = 0;
+ }
+ while (pbi->num_partitions < (1 << pbi->common.multi_token_partition) + 1)
+ {
+ // Reset all missing partitions
+ pbi->partitions[pbi->num_partitions] =
+ pbi->partitions[pbi->num_partitions - 1] +
+ pbi->partition_sizes[pbi->num_partitions - 1];
+ pbi->partition_sizes[pbi->num_partitions] = 0;
+ pbi->num_partitions++;
+ }
+ }
if (pbi->source_sz == 0)
{
cm->show_frame = 0;
pbi->num_partitions = 0;
- if (pbi->input_partition)
- pbi->common.multi_token_partition = 0;
/* Nothing more to do. */
return 0;
pbi->common.error.setjmp = 0;
pbi->num_partitions = 0;
- if (pbi->input_partition)
- pbi->common.multi_token_partition = 0;
/* We do not know if the missing frame(s) was supposed to update
* any of the reference buffers, but we act conservative and
#endif
pbi->common.error.error_code = VPX_CODEC_ERROR;
pbi->common.error.setjmp = 0;
+ pbi->num_partitions = 0;
if (cm->fb_idx_ref_cnt[cm->new_fb_idx] > 0)
cm->fb_idx_ref_cnt[cm->new_fb_idx]--;
return retcode;
#endif
pbi->common.error.error_code = VPX_CODEC_ERROR;
pbi->common.error.setjmp = 0;
+ pbi->num_partitions = 0;
return -1;
}
} else
#endif
pbi->common.error.error_code = VPX_CODEC_ERROR;
pbi->common.error.setjmp = 0;
+ pbi->num_partitions = 0;
return -1;
}
pbi->ready_for_new_data = 0;
pbi->last_time_stamp = time_stamp;
pbi->num_partitions = 0;
- if (pbi->input_partition)
- pbi->common.multi_token_partition = 0;
pbi->source_sz = 0;
#if 0
{
vp8_reset_mb_tokens_context(xd);
}
- else
+ else if (!vp8dx_bool_error(xd->current_bc))
{
eobtotal = vp8_decode_mb_tokens(pbi, xd);
}
#if CONFIG_ERROR_CONCEALMENT
if (pbi->ec_active &&
(mb_row * pbi->common.mb_cols + mb_col >= pbi->mvs_corrupt_from_mb ||
- throw_residual))
+ throw_residual))
{
/* MB with corrupt residuals or corrupt mode/motion vectors.
* Better to use the predictor as reconstruction.
/*cpi->rtcd.variance.get4x4sse_cs = vp8_get4x4sse_cs_c;*/
- /*cpi->rtcd.fdct.short4x4 = vp8_short_fdct4x4_c;
- cpi->rtcd.fdct.short8x4 = vp8_short_fdct8x4_c;*/
- cpi->rtcd.fdct.fast4x4 = vp8_fast_fdct4x4_armv6;
- cpi->rtcd.fdct.fast8x4 = vp8_fast_fdct8x4_armv6;
+ cpi->rtcd.fdct.short4x4 = vp8_short_fdct4x4_armv6;
+ cpi->rtcd.fdct.short8x4 = vp8_short_fdct8x4_armv6;
+ cpi->rtcd.fdct.fast4x4 = vp8_short_fdct4x4_armv6;
+ cpi->rtcd.fdct.fast8x4 = vp8_short_fdct8x4_armv6;
cpi->rtcd.fdct.walsh_short4x4 = vp8_short_walsh4x4_armv6;
/*cpi->rtcd.encodemb.berr = vp8_block_error_c;
sub r7, r5, #1 ; range-1
cmp r1, #0
- mul r4, r4, r7 ; ((range-1) * probability)
+ mul r6, r4, r7 ; ((range-1) * probability)
mov r7, #1
- add r4, r7, r4, lsr #8 ; 1 + (((range-1) * probability) >> 8)
+ add r4, r7, r6, lsr #8 ; 1 + (((range-1) * probability) >> 8)
addne r2, r2, r4 ; if (bit) lowvalue += split
subne r4, r5, r4 ; if (bit) range = range-split
; off of v, so set a flag here based on this.
; This value is refered to as "bb"
lsls r12, r12, #1 ; bb = v >> n
- mul r4, r4, r7 ; ((range-1) * pp[i>>1]))
+ mul r6, r4, r7 ; ((range-1) * pp[i>>1]))
; bb can only be 0 or 1. So only execute this statement
; if bb == 1, otherwise it will act like i + 0
mov r7, #1
ldrsb lr, [r10, lr] ; i = vp8_coef_tree[i+bb]
- add r4, r7, r4, lsr #8 ; 1 + (((range-1) * pp[i>>1]) >> 8)
+ add r4, r7, r6, lsr #8 ; 1 + (((range-1) * pp[i>>1]) >> 8)
addcs r2, r2, r4 ; if (bb) lowvalue += split
subcs r4, r5, r4 ; if (bb) range = range-split
ldrb r4, [r9, lr, asr #1] ; pp[i>>1]
sub r7, r5, #1 ; range-1
lsls r12, r12, #1 ; v >> n
- mul r4, r4, r7 ; (range-1) * pp[i>>1]
+ mul r6, r4, r7 ; (range-1) * pp[i>>1]
addcs lr, lr, #1 ; i + bb
mov r7, #1
ldrsb lr, [r10, lr] ; i = b->tree[i+bb]
- add r4, r7, r4, lsr #8 ; split = 1 + (((range-1) * pp[i>>1]) >> 8)
+ add r4, r7, r6, lsr #8 ; split = 1 + (((range-1) * pp[i>>1]) >> 8)
addcs r2, r2, r4 ; if (bb) lowvalue += split
subcs r4, r5, r4 ; if (bb) range = range-split
; off of v, so set a flag here based on this.
; This value is refered to as "bb"
lsls r12, r12, #1 ; bb = v >> n
- mul r4, r4, r7 ; ((range-1) * pp[i>>1]))
+ mul r6, r4, r7 ; ((range-1) * pp[i>>1]))
; bb can only be 0 or 1. So only execute this statement
; if bb == 1, otherwise it will act like i + 0
mov r7, #1
ldrsb lr, [r10, lr] ; i = vp8_coef_tree[i+bb]
- add r4, r7, r4, lsr #8 ; 1 + (((range-1) * pp[i>>1]) >> 8)
+ add r4, r7, r6, lsr #8 ; 1 + (((range-1) * pp[i>>1]) >> 8)
addcs r2, r2, r4 ; if (bb) lowvalue += split
subcs r4, r5, r4 ; if (bb) range = range-split
ldrb r4, [r9, lr, asr #1] ; pp[i>>1]
sub r7, r5, #1 ; range-1
lsls r12, r12, #1 ; v >> n
- mul r4, r4, r7 ; (range-1) * pp[i>>1]
+ mul r6, r4, r7 ; (range-1) * pp[i>>1]
addcs lr, lr, #1 ; i + bb
mov r7, #1
ldrsb lr, [r10, lr] ; i = b->tree[i+bb]
- add r4, r7, r4, lsr #8 ; split = 1 + (((range-1) * pp[i>>1]) >> 8)
+ add r4, r7, r6, lsr #8 ; split = 1 + (((range-1) * pp[i>>1]) >> 8)
addcs r2, r2, r4 ; if (bb) lowvalue += split
subcs r4, r5, r4 ; if (bb) range = range-split
; off of v, so set a flag here based on this.
; This value is refered to as "bb"
lsls r12, r12, #1 ; bb = v >> n
- mul r4, r4, r7 ; ((range-1) * pp[i>>1]))
+ mul r6, r4, r7 ; ((range-1) * pp[i>>1]))
; bb can only be 0 or 1. So only execute this statement
; if bb == 1, otherwise it will act like i + 0
mov r7, #1
ldrsb lr, [r10, lr] ; i = vp8_coef_tree[i+bb]
- add r4, r7, r4, lsr #8 ; 1 + (((range-1) * pp[i>>1]) >> 8)
+ add r4, r7, r6, lsr #8 ; 1 + (((range-1) * pp[i>>1]) >> 8)
addcs r2, r2, r4 ; if (bb) lowvalue += split
subcs r4, r5, r4 ; if (bb) range = range-split
ldrb r4, [r9, lr, asr #1] ; pp[i>>1]
sub r7, r5, #1 ; range-1
lsls r12, r12, #1 ; v >> n
- mul r4, r4, r7 ; (range-1) * pp[i>>1]
+ mul r6, r4, r7 ; (range-1) * pp[i>>1]
addcs lr, lr, #1 ; i + bb
mov r7, #1
ldrsb lr, [r10, lr] ; i = b->tree[i+bb]
- add r4, r7, r4, lsr #8 ; split = 1 + (((range-1) * pp[i>>1]) >> 8)
+ add r4, r7, r6, lsr #8 ; split = 1 + (((range-1) * pp[i>>1]) >> 8)
addcs r2, r2, r4 ; if (bb) lowvalue += split
subcs r4, r5, r4 ; if (bb) range = range-split
; be found in the AUTHORS file in the root of the source tree.
;
- EXPORT |vp8_fast_fdct4x4_armv6|
+ EXPORT |vp8_short_fdct4x4_armv6|
ARM
REQUIRE8
AREA |.text|, CODE, READONLY
; void vp8_short_fdct4x4_c(short *input, short *output, int pitch)
-|vp8_fast_fdct4x4_armv6| PROC
+|vp8_short_fdct4x4_armv6| PROC
stmfd sp!, {r4 - r12, lr}
AREA |.text|, CODE, READONLY ; name this block of code
;short vp8_short_walsh4x4_armv6(short *input, short *output, int pitch)
+; r0 short *input,
+; r1 short *output,
+; r2 int pitch
|vp8_short_walsh4x4_armv6| PROC
stmdb sp!, {r4 - r11, lr}
- mov r12, r2 ; ugh. not clean
- ldr r2, [r0] ; [1 | 0]
- ldr r3, [r0, #4] ; [3 | 2]
- ldr r4, [r0, r12]! ; [5 | 4]
- ldr r5, [r0, #4] ; [7 | 6]
- ldr r6, [r0, r12]! ; [9 | 8]
- ldr r7, [r0, #4] ; [11 | 10]
- ldr r8, [r0, r12]! ; [13 | 12]
- ldr r9, [r0, #4] ; [15 | 14]
-
- qsubaddx r10, r2, r3 ; [c1|a1] [1-2 | 0+3]
- qaddsubx r11, r2, r3 ; [b1|d1] [1+2 | 0-3]
- qsubaddx r12, r4, r5 ; [c1|a1] [5-6 | 4+7]
- qaddsubx lr, r4, r5 ; [b1|d1] [5+6 | 4-7]
-
- qaddsubx r2, r10, r11 ; [1 | 2] [c1+d1 | a1-b1]
- qaddsubx r3, r11, r10 ; [0 | 3] [b1+a1 | d1-c1]
- qaddsubx r4, r12, lr ; [5 | 6] [c1+d1 | a1-b1]
- qaddsubx r5, lr, r12 ; [4 | 7] [b1+a1 | d1-c1]
-
- qsubaddx r10, r6, r7 ; [c1|a1] [9-10 | 8+11]
- qaddsubx r11, r6, r7 ; [b1|d1] [9+10 | 8-11]
- qsubaddx r12, r8, r9 ; [c1|a1] [13-14 | 12+15]
- qaddsubx lr, r8, r9 ; [b1|d1] [13+14 | 12-15]
-
- qaddsubx r6, r10, r11 ; [9 |10] [c1+d1 | a1-b1]
- qaddsubx r7, r11, r10 ; [8 |11] [b1+a1 | d1-c1]
- qaddsubx r8, r12, lr ; [13|14] [c1+d1 | a1-b1]
- qaddsubx r9, lr, r12 ; [12|15] [b1+a1 | d1-c1]
-
- ; first transform complete
-
- qadd16 r10, r3, r9 ; a1 [0+12 | 3+15]
- qadd16 r11, r5, r7 ; b1 [4+8 | 7+11]
- qsub16 r12, r5, r7 ; c1 [4-8 | 7-11]
- qsub16 lr, r3, r9 ; d1 [0-12 | 3-15]
-
- qadd16 r3, r10, r11 ; a2 [a1+b1] [0 | 3]
- qadd16 r5, r12, lr ; b2 [c1+d1] [4 | 7]
- qsub16 r7, r10, r11 ; c2 [a1-b1] [8 |11]
- qsub16 r9, lr, r12 ; d2 [d1-c1] [12|15]
-
- qadd16 r10, r2, r8 ; a1 [1+13 | 2+14]
- qadd16 r11, r4, r6 ; b1 [5+9 | 6+10]
- qsub16 r12, r4, r6 ; c1 [5-9 | 6-10]
- qsub16 lr, r2, r8 ; d1 [1-13 | 2-14]
-
- qadd16 r2, r10, r11 ; a2 [a1+b1] [1 | 2]
- qadd16 r4, r12, lr ; b2 [c1+d1] [5 | 6]
- qsub16 r6, r10, r11 ; c2 [a1-b1] [9 |10]
- qsub16 r8, lr, r12 ; d2 [d1-c1] [13|14]
-
- ; [a-d]2 += ([a-d]2 > 0)
-
- asrs r10, r3, #16
- addpl r10, r10, #1 ; [~0]
- asrs r11, r2, #16
- addpl r11, r11, #1 ; [~1]
- lsl r11, r11, #15 ; [1 | x]
- pkhtb r10, r11, r10, asr #1; [1 | 0]
- str r10, [r1], #4
-
- lsls r11, r2, #16
- addpl r11, r11, #0x10000 ; [~2]
- lsls r12, r3, #16
- addpl r12, r12, #0x10000 ; [~3]
- asr r12, r12, #1 ; [3 | x]
- pkhtb r11, r12, r11, asr #17; [3 | 2]
- str r11, [r1], #4
-
- asrs r2, r5, #16
- addpl r2, r2, #1 ; [~4]
- asrs r3, r4, #16
- addpl r3, r3, #1 ; [~5]
- lsl r3, r3, #15 ; [5 | x]
- pkhtb r2, r3, r2, asr #1 ; [5 | 4]
- str r2, [r1], #4
-
- lsls r2, r4, #16
- addpl r2, r2, #0x10000 ; [~6]
- lsls r3, r5, #16
- addpl r3, r3, #0x10000 ; [~7]
- asr r3, r3, #1 ; [7 | x]
- pkhtb r2, r3, r2, asr #17 ; [7 | 6]
- str r2, [r1], #4
-
- asrs r2, r7, #16
- addpl r2, r2, #1 ; [~8]
- asrs r3, r6, #16
- addpl r3, r3, #1 ; [~9]
- lsl r3, r3, #15 ; [9 | x]
- pkhtb r2, r3, r2, asr #1 ; [9 | 8]
- str r2, [r1], #4
-
- lsls r2, r6, #16
- addpl r2, r2, #0x10000 ; [~10]
- lsls r3, r7, #16
- addpl r3, r3, #0x10000 ; [~11]
- asr r3, r3, #1 ; [11 | x]
- pkhtb r2, r3, r2, asr #17 ; [11 | 10]
- str r2, [r1], #4
-
- asrs r2, r9, #16
- addpl r2, r2, #1 ; [~12]
- asrs r3, r8, #16
- addpl r3, r3, #1 ; [~13]
- lsl r3, r3, #15 ; [13 | x]
- pkhtb r2, r3, r2, asr #1 ; [13 | 12]
- str r2, [r1], #4
-
- lsls r2, r8, #16
- addpl r2, r2, #0x10000 ; [~14]
- lsls r3, r9, #16
- addpl r3, r3, #0x10000 ; [~15]
- asr r3, r3, #1 ; [15 | x]
- pkhtb r2, r3, r2, asr #17 ; [15 | 14]
- str r2, [r1]
+ ldrd r4, r5, [r0], r2
+ ldr lr, c00040004
+ ldrd r6, r7, [r0], r2
+
+ ; 0-3
+ qadd16 r3, r4, r5 ; [d1|a1] [1+3 | 0+2]
+ qsub16 r4, r4, r5 ; [c1|b1] [1-3 | 0-2]
+
+ ldrd r8, r9, [r0], r2
+ ; 4-7
+ qadd16 r5, r6, r7 ; [d1|a1] [5+7 | 4+6]
+ qsub16 r6, r6, r7 ; [c1|b1] [5-7 | 4-6]
+
+ ldrd r10, r11, [r0]
+ ; 8-11
+ qadd16 r7, r8, r9 ; [d1|a1] [9+11 | 8+10]
+ qsub16 r8, r8, r9 ; [c1|b1] [9-11 | 8-10]
+
+ ; 12-15
+ qadd16 r9, r10, r11 ; [d1|a1] [13+15 | 12+14]
+ qsub16 r10, r10, r11 ; [c1|b1] [13-15 | 12-14]
+
+
+ lsls r2, r3, #16
+ smuad r11, r3, lr ; A0 = a1<<2 + d1<<2
+ addne r11, r11, #1 ; A0 += (a1!=0)
+
+ lsls r2, r7, #16
+ smuad r12, r7, lr ; C0 = a1<<2 + d1<<2
+ addne r12, r12, #1 ; C0 += (a1!=0)
+
+ add r0, r11, r12 ; a1_0 = A0 + C0
+ sub r11, r11, r12 ; b1_0 = A0 - C0
+
+ lsls r2, r5, #16
+ smuad r12, r5, lr ; B0 = a1<<2 + d1<<2
+ addne r12, r12, #1 ; B0 += (a1!=0)
+
+ lsls r2, r9, #16
+ smuad r2, r9, lr ; D0 = a1<<2 + d1<<2
+ addne r2, r2, #1 ; D0 += (a1!=0)
+
+ add lr, r12, r2 ; d1_0 = B0 + D0
+ sub r12, r12, r2 ; c1_0 = B0 - D0
+
+ ; op[0,4,8,12]
+ adds r2, r0, lr ; a2 = a1_0 + d1_0
+ addmi r2, r2, #1 ; += a2 < 0
+ add r2, r2, #3 ; += 3
+ subs r0, r0, lr ; d2 = a1_0 - d1_0
+ mov r2, r2, asr #3 ; >> 3
+ strh r2, [r1] ; op[0]
+
+ addmi r0, r0, #1 ; += a2 < 0
+ add r0, r0, #3 ; += 3
+ ldr lr, c00040004
+ mov r0, r0, asr #3 ; >> 3
+ strh r0, [r1, #24] ; op[12]
+
+ adds r2, r11, r12 ; b2 = b1_0 + c1_0
+ addmi r2, r2, #1 ; += a2 < 0
+ add r2, r2, #3 ; += 3
+ subs r0, r11, r12 ; c2 = b1_0 - c1_0
+ mov r2, r2, asr #3 ; >> 3
+ strh r2, [r1, #8] ; op[4]
+
+ addmi r0, r0, #1 ; += a2 < 0
+ add r0, r0, #3 ; += 3
+ smusd r3, r3, lr ; A3 = a1<<2 - d1<<2
+ smusd r7, r7, lr ; C3 = a1<<2 - d1<<2
+ mov r0, r0, asr #3 ; >> 3
+ strh r0, [r1, #16] ; op[8]
+
+
+ ; op[3,7,11,15]
+ add r0, r3, r7 ; a1_3 = A3 + C3
+ sub r3, r3, r7 ; b1_3 = A3 - C3
+
+ smusd r5, r5, lr ; B3 = a1<<2 - d1<<2
+ smusd r9, r9, lr ; D3 = a1<<2 - d1<<2
+ add r7, r5, r9 ; d1_3 = B3 + D3
+ sub r5, r5, r9 ; c1_3 = B3 - D3
+
+ adds r2, r0, r7 ; a2 = a1_3 + d1_3
+ addmi r2, r2, #1 ; += a2 < 0
+ add r2, r2, #3 ; += 3
+ adds r9, r3, r5 ; b2 = b1_3 + c1_3
+ mov r2, r2, asr #3 ; >> 3
+ strh r2, [r1, #6] ; op[3]
+
+ addmi r9, r9, #1 ; += a2 < 0
+ add r9, r9, #3 ; += 3
+ subs r2, r3, r5 ; c2 = b1_3 - c1_3
+ mov r9, r9, asr #3 ; >> 3
+ strh r9, [r1, #14] ; op[7]
+
+ addmi r2, r2, #1 ; += a2 < 0
+ add r2, r2, #3 ; += 3
+ subs r9, r0, r7 ; d2 = a1_3 - d1_3
+ mov r2, r2, asr #3 ; >> 3
+ strh r2, [r1, #22] ; op[11]
+
+ addmi r9, r9, #1 ; += a2 < 0
+ add r9, r9, #3 ; += 3
+ smuad r3, r4, lr ; A1 = b1<<2 + c1<<2
+ smuad r5, r8, lr ; C1 = b1<<2 + c1<<2
+ mov r9, r9, asr #3 ; >> 3
+ strh r9, [r1, #30] ; op[15]
+
+ ; op[1,5,9,13]
+ add r0, r3, r5 ; a1_1 = A1 + C1
+ sub r3, r3, r5 ; b1_1 = A1 - C1
+
+ smuad r7, r6, lr ; B1 = b1<<2 + c1<<2
+ smuad r9, r10, lr ; D1 = b1<<2 + c1<<2
+ add r5, r7, r9 ; d1_1 = B1 + D1
+ sub r7, r7, r9 ; c1_1 = B1 - D1
+
+ adds r2, r0, r5 ; a2 = a1_1 + d1_1
+ addmi r2, r2, #1 ; += a2 < 0
+ add r2, r2, #3 ; += 3
+ adds r9, r3, r7 ; b2 = b1_1 + c1_1
+ mov r2, r2, asr #3 ; >> 3
+ strh r2, [r1, #2] ; op[1]
+
+ addmi r9, r9, #1 ; += a2 < 0
+ add r9, r9, #3 ; += 3
+ subs r2, r3, r7 ; c2 = b1_1 - c1_1
+ mov r9, r9, asr #3 ; >> 3
+ strh r9, [r1, #10] ; op[5]
+
+ addmi r2, r2, #1 ; += a2 < 0
+ add r2, r2, #3 ; += 3
+ subs r9, r0, r5 ; d2 = a1_1 - d1_1
+ mov r2, r2, asr #3 ; >> 3
+ strh r2, [r1, #18] ; op[9]
+
+ addmi r9, r9, #1 ; += a2 < 0
+ add r9, r9, #3 ; += 3
+ smusd r4, r4, lr ; A2 = b1<<2 - c1<<2
+ smusd r8, r8, lr ; C2 = b1<<2 - c1<<2
+ mov r9, r9, asr #3 ; >> 3
+ strh r9, [r1, #26] ; op[13]
+
+
+ ; op[2,6,10,14]
+ add r11, r4, r8 ; a1_2 = A2 + C2
+ sub r12, r4, r8 ; b1_2 = A2 - C2
+
+ smusd r6, r6, lr ; B2 = b1<<2 - c1<<2
+ smusd r10, r10, lr ; D2 = b1<<2 - c1<<2
+ add r4, r6, r10 ; d1_2 = B2 + D2
+ sub r8, r6, r10 ; c1_2 = B2 - D2
+
+ adds r2, r11, r4 ; a2 = a1_2 + d1_2
+ addmi r2, r2, #1 ; += a2 < 0
+ add r2, r2, #3 ; += 3
+ adds r9, r12, r8 ; b2 = b1_2 + c1_2
+ mov r2, r2, asr #3 ; >> 3
+ strh r2, [r1, #4] ; op[2]
+
+ addmi r9, r9, #1 ; += a2 < 0
+ add r9, r9, #3 ; += 3
+ subs r2, r12, r8 ; c2 = b1_2 - c1_2
+ mov r9, r9, asr #3 ; >> 3
+ strh r9, [r1, #12] ; op[6]
+
+ addmi r2, r2, #1 ; += a2 < 0
+ add r2, r2, #3 ; += 3
+ subs r9, r11, r4 ; d2 = a1_2 - d1_2
+ mov r2, r2, asr #3 ; >> 3
+ strh r2, [r1, #20] ; op[10]
+
+ addmi r9, r9, #1 ; += a2 < 0
+ add r9, r9, #3 ; += 3
+ mov r9, r9, asr #3 ; >> 3
+ strh r9, [r1, #28] ; op[14]
+
ldmia sp!, {r4 - r11, pc}
ENDP ; |vp8_short_walsh4x4_armv6|
+c00040004
+ DCD 0x00040004
+
END
#if HAVE_ARMV6
-void vp8_fast_fdct8x4_armv6(short *input, short *output, int pitch)
+void vp8_short_fdct8x4_armv6(short *input, short *output, int pitch)
{
- vp8_fast_fdct4x4_armv6(input, output, pitch);
- vp8_fast_fdct4x4_armv6(input + 4, output + 16, pitch);
+ vp8_short_fdct4x4_armv6(input, output, pitch);
+ vp8_short_fdct4x4_armv6(input + 4, output + 16, pitch);
}
#endif /* HAVE_ARMV6 */
-
-
#if HAVE_ARMV6
extern prototype_fdct(vp8_short_walsh4x4_armv6);
-extern prototype_fdct(vp8_fast_fdct4x4_armv6);
-extern prototype_fdct(vp8_fast_fdct8x4_armv6);
+extern prototype_fdct(vp8_short_fdct4x4_armv6);
+extern prototype_fdct(vp8_short_fdct8x4_armv6);
#if !CONFIG_RUNTIME_CPU_DETECT
#undef vp8_fdct_walsh_short4x4
#define vp8_fdct_walsh_short4x4 vp8_short_walsh4x4_armv6
+#undef vp8_fdct_short4x4
+#define vp8_fdct_short4x4 vp8_short_fdct4x4_armv6
+
+#undef vp8_fdct_short8x4
+#define vp8_fdct_short8x4 vp8_short_fdct8x4_armv6
+
#undef vp8_fdct_fast4x4
-#define vp8_fdct_fast4x4 vp8_fast_fdct4x4_armv6
+#define vp8_fdct_fast4x4 vp8_short_fdct4x4_armv6
#undef vp8_fdct_fast8x4
-#define vp8_fdct_fast8x4 vp8_fast_fdct8x4_armv6
+#define vp8_fdct_fast8x4 vp8_short_fdct8x4_armv6
#endif
#endif /* HAVE_ARMV6 */
PRESERVE8
AREA ||.text||, CODE, READONLY, ALIGN=2
-;void vp8_short_walsh4x4_c(short *input, short *output, int pitch)
-
+;void vp8_short_walsh4x4_neon(short *input, short *output, int pitch)
+; r0 short *input,
+; r1 short *output,
+; r2 int pitch
|vp8_short_walsh4x4_neon| PROC
- vld1.16 {d2}, [r0], r2 ;load input
- vld1.16 {d3}, [r0], r2
- vld1.16 {d4}, [r0], r2
- vld1.16 {d5}, [r0], r2
- ;First for-loop
- ;transpose d2, d3, d4, d5. Then, d2=ip[0], d3=ip[1], d4=ip[2], d5=ip[3]
- vtrn.32 d2, d4
- vtrn.32 d3, d5
- vtrn.16 d2, d3
- vtrn.16 d4, d5
+ vld1.16 {d0}, [r0@64], r2 ; load input
+ vld1.16 {d1}, [r0@64], r2
+ vld1.16 {d2}, [r0@64], r2
+ vld1.16 {d3}, [r0@64]
- vadd.s16 d6, d2, d5 ;a1 = ip[0]+ip[3]
- vadd.s16 d7, d3, d4 ;b1 = ip[1]+ip[2]
- vsub.s16 d8, d3, d4 ;c1 = ip[1]-ip[2]
- vsub.s16 d9, d2, d5 ;d1 = ip[0]-ip[3]
+ ;First for-loop
+ ;transpose d0, d1, d2, d3. Then, d0=ip[0], d1=ip[1], d2=ip[2], d3=ip[3]
+ vtrn.32 d0, d2
+ vtrn.32 d1, d3
- vadd.s16 d2, d6, d7 ;op[0] = a1 + b1
- vsub.s16 d4, d6, d7 ;op[2] = a1 - b1
- vadd.s16 d3, d8, d9 ;op[1] = c1 + d1
- vsub.s16 d5, d9, d8 ;op[3] = d1 - c1
+ vmov.s32 q15, #3 ; add 3 to all values
- ;Second for-loop
- ;transpose d2, d3, d4, d5. Then, d2=ip[0], d3=ip[4], d4=ip[8], d5=ip[12]
- vtrn.32 d2, d4
- vtrn.32 d3, d5
+ vtrn.16 d0, d1
vtrn.16 d2, d3
- vtrn.16 d4, d5
- vadd.s16 d6, d2, d5 ;a1 = ip[0]+ip[12]
- vadd.s16 d7, d3, d4 ;b1 = ip[4]+ip[8]
- vsub.s16 d8, d3, d4 ;c1 = ip[4]-ip[8]
- vsub.s16 d9, d2, d5 ;d1 = ip[0]-ip[12]
+ vadd.s16 d4, d0, d2 ; ip[0] + ip[2]
+ vadd.s16 d5, d1, d3 ; ip[1] + ip[3]
+ vsub.s16 d6, d1, d3 ; ip[1] - ip[3]
+ vsub.s16 d7, d0, d2 ; ip[0] - ip[2]
- vadd.s16 d2, d6, d7 ;a2 = a1 + b1;
- vsub.s16 d4, d6, d7 ;c2 = a1 - b1;
- vadd.s16 d3, d8, d9 ;b2 = c1 + d1;
- vsub.s16 d5, d9, d8 ;d2 = d1 - c1;
+ vshl.s16 d4, d4, #2 ; a1 = (ip[0] + ip[2]) << 2
+ vshl.s16 d5, d5, #2 ; d1 = (ip[1] + ip[3]) << 2
+ vshl.s16 d6, d6, #2 ; c1 = (ip[1] - ip[3]) << 2
+ vceq.s16 d16, d4, #0 ; a1 == 0
+ vshl.s16 d7, d7, #2 ; b1 = (ip[0] - ip[2]) << 2
- vcgt.s16 q3, q1, #0
- vcgt.s16 q4, q2, #0
+ vadd.s16 d0, d4, d5 ; a1 + d1
+ vmvn d16, d16 ; a1 != 0
+ vsub.s16 d3, d4, d5 ; op[3] = a1 - d1
+ vadd.s16 d1, d7, d6 ; op[1] = b1 + c1
+ vsub.s16 d2, d7, d6 ; op[2] = b1 - c1
+ vsub.s16 d0, d0, d16 ; op[0] = a1 + d1 + (a1 != 0)
- vsub.s16 q1, q1, q3
- vsub.s16 q2, q2, q4
-
- vshr.s16 q1, q1, #1
- vshr.s16 q2, q2, #1
-
- vst1.16 {q1, q2}, [r1]
+ ;Second for-loop
+ ;transpose d0, d1, d2, d3, Then, d0=ip[0], d1=ip[4], d2=ip[8], d3=ip[12]
+ vtrn.32 d1, d3
+ vtrn.32 d0, d2
+ vtrn.16 d2, d3
+ vtrn.16 d0, d1
+
+ vaddl.s16 q8, d0, d2 ; a1 = ip[0]+ip[8]
+ vaddl.s16 q9, d1, d3 ; d1 = ip[4]+ip[12]
+ vsubl.s16 q10, d1, d3 ; c1 = ip[4]-ip[12]
+ vsubl.s16 q11, d0, d2 ; b1 = ip[0]-ip[8]
+
+ vadd.s32 q0, q8, q9 ; a2 = a1 + d1
+ vadd.s32 q1, q11, q10 ; b2 = b1 + c1
+ vsub.s32 q2, q11, q10 ; c2 = b1 - c1
+ vsub.s32 q3, q8, q9 ; d2 = a1 - d1
+
+ vclt.s32 q8, q0, #0
+ vclt.s32 q9, q1, #0
+ vclt.s32 q10, q2, #0
+ vclt.s32 q11, q3, #0
+
+ ; subtract -1 (or 0)
+ vsub.s32 q0, q0, q8 ; a2 += a2 < 0
+ vsub.s32 q1, q1, q9 ; b2 += b2 < 0
+ vsub.s32 q2, q2, q10 ; c2 += c2 < 0
+ vsub.s32 q3, q3, q11 ; d2 += d2 < 0
+
+ vadd.s32 q8, q0, q15 ; a2 + 3
+ vadd.s32 q9, q1, q15 ; b2 + 3
+ vadd.s32 q10, q2, q15 ; c2 + 3
+ vadd.s32 q11, q3, q15 ; d2 + 3
+
+ ; vrshrn? would add 1 << 3-1 = 2
+ vshrn.s32 d0, q8, #3
+ vshrn.s32 d1, q9, #3
+ vshrn.s32 d2, q10, #3
+ vshrn.s32 d3, q11, #3
+
+ vst1.16 {q0, q1}, [r1@128]
bx lr
#File list for armv6
# encoder
VP8_CX_SRCS-$(HAVE_ARMV6) += encoder/arm/armv6/vp8_subtract_armv6$(ASM)
-VP8_CX_SRCS-$(HAVE_ARMV6) += encoder/arm/armv6/vp8_fast_fdct4x4_armv6$(ASM)
+VP8_CX_SRCS-$(HAVE_ARMV6) += encoder/arm/armv6/vp8_short_fdct4x4_armv6$(ASM)
VP8_CX_SRCS-$(HAVE_ARMV6) += encoder/arm/armv6/vp8_fast_quantize_b_armv6$(ASM)
VP8_CX_SRCS-$(HAVE_ARMV6) += encoder/arm/armv6/vp8_sad16x16_armv6$(ASM)
VP8_CX_SRCS-$(HAVE_ARMV6) += encoder/arm/armv6/vp8_variance16x16_armv6$(ASM)