Skip to content
Projects
Groups
Snippets
Help
Loading...
Help
Support
Keyboard shortcuts
?
Submit feedback
Contribute to GitLab
Sign in / Register
Toggle navigation
O
OpenXG-RAN
Project overview
Project overview
Details
Activity
Releases
Repository
Repository
Files
Commits
Branches
Tags
Contributors
Graph
Compare
Issues
0
Issues
0
List
Boards
Labels
Milestones
Merge Requests
0
Merge Requests
0
CI / CD
CI / CD
Pipelines
Jobs
Schedules
Analytics
Analytics
CI / CD
Repository
Value Stream
Wiki
Wiki
Snippets
Snippets
Members
Members
Collapse sidebar
Close sidebar
Activity
Graph
Create a new issue
Jobs
Commits
Issue Boards
Open sidebar
Michael Black
OpenXG-RAN
Commits
abc65b1d
Commit
abc65b1d
authored
5 years ago
by
Robert Schmidt
Browse files
Options
Browse Files
Download
Email Patches
Plain Diff
Reimplement UL PP
parent
8bf7a13f
Branches unavailable
2023.w22
2023.w21
2023.w20
2023.w19
2023.w18
2023.w18b
2023.w16
2023.w15
2023.w14
2023.w13
2023.w12
2023.w11
2023.w11b
2023.w10
2023.w10b
2023.w09
2023.w08
2023.w08b
2023.w07
2023.w06
2023.w05
2023.w03
2023.w02
2022.42
2022.41
2022.w51
2022.w50
2022.w49
2022.w48
2022.w47
2022.w46
2022.w45
2022.w43
2022.w42
2022.w42b
2022.w41
2022.w40
2022.w39
2022.w38
2022.w37
2022.w37b
2022.w36
2022.w35
2022.w33
2022.w32
2022.w31
2022.w31b
2022.w30
2022.w29
2022.w26
2022.w25
2022.w24
2022.w24b
2022.w23
2022.w22
2022.w21
2022.w20
2022.w19
2022.w18
2022.w17
2022.w15
2022.w15b
2022.w14a
2022.w13
2022.w13b
2022.w13a
2022.w12
2022.w10
2022.w09
2022.w09b
2022.w08
2022.w08b
2022.w07
2022.w07b
2022.w06
2022.w06a
2022.w05
2022.w05b
2022.w03_hotfix
2022.w03_b
2022.w02
2022.w01
2021.wk46
2021.wk14_a
2021.wk13_d
2021.wk13_c
2021.w51_c
2021.w51_a
2021.w50_a
2021.w49_b
2021.w49_a
2021.w48
2021.w47
2021.w46
2021.w46-powder
2021.w45
2021.w45_b
2021.w44
2021.w43
2021.w42
2021.w37
2021.w36
2021.w35
2021.w34
2021.w33
2021.w32
2021.w31
2021.w30
2021.w29
2021.w28
2021.w27
2021.w26
2021.w25
2021.w24
2021.w23
2021.w22
2021.w20
2021.w19
2021.w18_b
2021.w18_a
2021.w17_b
2021.w16
2021.w15
2021.w14
2021.w13_a
2021.w12
2021.w11
2021.w10
2021.w09
2021.w08
2021.w06
2021.w05
2021.w04
2021.w02
2020.w51_2
2020.w51
2020.w50
2020.w49
2020.w48_2
2020.w48
2020.w47
2020.w46_2
2020.w46
2020.w45_2
2020.w45
2020.w44
2020.w42_2
2020.w42
2020.w41
2020.w39
2020.w38
2020.w37
2020.w36
2020.w34
2020.w33
2020.w31
2020.w30
2020.w29
2020.w28
2020.w26
2020.w25
2020.w24
2020.w23
2020.w22
2020.w19
2020.w17
2020.w16
2020.w15
setparam
flexran-eol
benetel_phase_rotation
benetel_gnb_rel_2.0
benetel_gnb_rel_1.0
benetel_enb_rel_2.0
benetel_enb_rel_1.0
No related merge requests found
Changes
3
Hide whitespace changes
Inline
Side-by-side
Showing
3 changed files
with
352 additions
and
132 deletions
+352
-132
openair2/LAYER2/MAC/mac.h
openair2/LAYER2/MAC/mac.h
+8
-0
openair2/LAYER2/MAC/mac_proto.h
openair2/LAYER2/MAC/mac_proto.h
+0
-9
openair2/LAYER2/MAC/pre_processor.c
openair2/LAYER2/MAC/pre_processor.c
+344
-123
No files found.
openair2/LAYER2/MAC/mac.h
View file @
abc65b1d
...
...
@@ -1262,6 +1262,14 @@ typedef struct {
uint8_t
rballoc_sub
[
NFAPI_CC_MAX
][
N_RBG_MAX
];
}
slice_info_t
;
/**
* describes contiguous RBs
*/
typedef
struct
{
int
start
;
int
length
;
}
contig_rbs_t
;
/*! \brief eNB common channels */
typedef
struct
{
int
physCellId
;
...
...
This diff is collapsed.
Click to expand it.
openair2/LAYER2/MAC/mac_proto.h
View file @
abc65b1d
...
...
@@ -674,15 +674,6 @@ void ulsch_scheduler_pre_processor(module_id_t module_idP,
sub_frame_t
subframeP
,
int
sched_frameP
,
unsigned
char
sched_subframeP
);
void
store_ulsch_buffer
(
module_id_t
module_idP
,
int
frameP
,
sub_frame_t
subframeP
);
void
assign_max_mcs_min_rb
(
module_id_t
module_idP
,
int
CC_id
,
int
frameP
,
sub_frame_t
subframeP
,
int
available_rbs
);
void
adjust_bsr_info
(
int
buffer_occupancy
,
uint16_t
TBS
,
UE_TEMPLATE
*
UE_template
);
int
phy_stats_exist
(
module_id_t
Mod_id
,
int
rnti
);
...
...
This diff is collapsed.
Click to expand it.
openair2/LAYER2/MAC/pre_processor.c
View file @
abc65b1d
...
...
@@ -380,111 +380,12 @@ dlsch_scheduler_pre_processor(module_id_t Mod_id,
}
}
if
(
print
)
LOG_
I
(
MAC
,
"%4d.%d DL scheduler allocation list: %s
\n
"
,
frameP
,
subframeP
,
t
);
LOG_
D
(
MAC
,
"%4d.%d DL scheduler allocation list: %s
\n
"
,
frameP
,
subframeP
,
t
);
#endif
}
/// ULSCH PRE_PROCESSOR
void
ulsch_scheduler_pre_processor
(
module_id_t
module_idP
,
int
CC_id
,
int
frameP
,
sub_frame_t
subframeP
,
int
sched_frameP
,
unsigned
char
sched_subframeP
)
{
uint16_t
nb_allocated_rbs
[
MAX_MOBILES_PER_ENB
];
uint16_t
total_allocated_rbs
=
0
;
uint16_t
average_rbs_per_user
=
0
;
int16_t
total_remaining_rbs
=
0
;
uint16_t
total_ue_count
=
0
;
UE_info_t
*
UE_info
=
&
RC
.
mac
[
module_idP
]
->
UE_info
;
const
int
N_RB_UL
=
to_prb
(
RC
.
mac
[
module_idP
]
->
common_channels
[
CC_id
].
ul_Bandwidth
);
const
COMMON_channels_t
*
cc
=
&
RC
.
mac
[
module_idP
]
->
common_channels
[
CC_id
];
int
available_rbs
=
0
;
int
first_rb
=
-
1
;
for
(
int
i
=
0
;
i
<
N_RB_UL
;
++
i
)
{
if
(
cc
->
vrb_map_UL
[
i
]
==
0
)
{
available_rbs
++
;
if
(
first_rb
<
0
)
first_rb
=
i
;
}
}
// maximize MCS and then allocate required RB according to the buffer
// occupancy with the limit of max available UL RB
LOG_D
(
MAC
,
"In ulsch_preprocessor: assign max mcs min rb
\n
"
);
assign_max_mcs_min_rb
(
module_idP
,
CC_id
,
frameP
,
subframeP
,
available_rbs
);
for
(
int
UE_id
=
UE_info
->
list
.
head
;
UE_id
>=
0
;
UE_id
=
UE_info
->
list
.
next
[
UE_id
])
{
if
(
UE_info
->
UE_template
[
CC_id
][
UE_id
].
pre_allocated_nb_rb_ul
>
0
)
{
total_ue_count
++
;
}
}
if
(
total_ue_count
==
0
)
average_rbs_per_user
=
0
;
else
if
(
total_ue_count
==
1
)
average_rbs_per_user
=
available_rbs
+
1
;
else
if
(
total_ue_count
<=
available_rbs
)
average_rbs_per_user
=
(
uint16_t
)
floor
(
available_rbs
/
total_ue_count
);
else
average_rbs_per_user
=
1
;
if
(
total_ue_count
>
0
)
LOG_D
(
MAC
,
"[eNB %d] Frame %d subframe %d: total ue to be scheduled %d
\n
"
,
module_idP
,
frameP
,
subframeP
,
total_ue_count
);
for
(
int
UE_id
=
UE_info
->
list
.
head
;
UE_id
>=
0
;
UE_id
=
UE_info
->
list
.
next
[
UE_id
])
{
uint8_t
harq_pid
=
subframe2harqpid
(
&
RC
.
mac
[
module_idP
]
->
common_channels
[
CC_id
],
sched_frameP
,
sched_subframeP
);
if
(
UE_info
->
UE_sched_ctrl
[
UE_id
].
round_UL
[
CC_id
][
harq_pid
]
>
0
)
nb_allocated_rbs
[
UE_id
]
=
UE_info
->
UE_template
[
CC_id
][
UE_id
].
nb_rb_ul
[
harq_pid
];
else
nb_allocated_rbs
[
UE_id
]
=
cmin
(
UE_info
->
UE_template
[
CC_id
][
UE_id
].
pre_allocated_nb_rb_ul
,
average_rbs_per_user
);
total_allocated_rbs
+=
nb_allocated_rbs
[
UE_id
];
LOG_D
(
MAC
,
"In ulsch_preprocessor: assigning %d RBs for UE %d CCid %d, harq_pid %d
\n
"
,
nb_allocated_rbs
[
UE_id
],
UE_id
,
CC_id
,
harq_pid
);
}
for
(
int
UE_id
=
UE_info
->
list
.
head
;
UE_id
>=
0
;
UE_id
=
UE_info
->
list
.
next
[
UE_id
])
{
UE_TEMPLATE
*
UE_template
=
&
UE_info
->
UE_template
[
CC_id
][
UE_id
];
total_remaining_rbs
=
available_rbs
-
total_allocated_rbs
;
/* TODO this has already been accounted for - do we need it again? */
//if (total_ue_count == 1)
// total_remaining_rbs++;
while
(
UE_template
->
pre_allocated_nb_rb_ul
>
0
&&
nb_allocated_rbs
[
UE_id
]
<
UE_template
->
pre_allocated_nb_rb_ul
&&
total_remaining_rbs
>
0
)
{
nb_allocated_rbs
[
UE_id
]
=
cmin
(
nb_allocated_rbs
[
UE_id
]
+
1
,
UE_template
->
pre_allocated_nb_rb_ul
);
total_remaining_rbs
--
;
total_allocated_rbs
++
;
}
UE_template
->
pre_first_nb_rb_ul
=
first_rb
;
UE_template
->
pre_allocated_nb_rb_ul
=
nb_allocated_rbs
[
UE_id
];
first_rb
+=
nb_allocated_rbs
[
UE_id
];
LOG_D
(
MAC
,
"******************UL Scheduling Information for UE%d CC_id %d ************************
\n
"
,
UE_id
,
CC_id
);
LOG_D
(
MAC
,
"[eNB %d] total RB allocated for UE%d CC_id %d = %d
\n
"
,
module_idP
,
UE_id
,
CC_id
,
UE_template
->
pre_allocated_nb_rb_ul
);
}
}
void
calculate_max_mcs_min_rb
(
module_id_t
mod_id
,
int
CC_id
,
int
bytes
,
...
...
@@ -530,29 +431,106 @@ void calculate_max_mcs_min_rb(module_id_t mod_id,
}
}
void
assign_max_mcs_min_rb
(
module_id_t
module_idP
,
int
CC_id
,
int
frameP
,
sub_frame_t
subframeP
,
int
available_rbs
)
{
const
int
Ncp
=
RC
.
mac
[
module_idP
]
->
common_channels
[
CC_id
].
Ncp
;
UE_info_t
*
UE_info
=
&
RC
.
mac
[
module_idP
]
->
UE_info
;
int
pp_find_rb_table_index
(
int
approximate
)
{
int
lo
=
2
;
if
(
approximate
<=
rb_table
[
lo
])
return
lo
;
int
hi
=
sizeof
(
rb_table
)
-
1
;
if
(
approximate
>=
rb_table
[
hi
])
return
hi
;
int
p
=
(
hi
+
lo
)
/
2
;
for
(;
lo
+
1
!=
hi
;
p
=
(
hi
+
lo
)
/
2
)
{
if
(
approximate
<=
rb_table
[
p
])
hi
=
p
;
else
lo
=
p
;
}
return
p
+
1
;
}
for
(
int
UE_id
=
UE_info
->
list
.
head
;
UE_id
>=
0
;
UE_id
=
UE_info
->
list
.
next
[
UE_id
])
{
int
g_start_ue_ul
=
-
1
;
int
round_robin_ul
(
module_id_t
Mod_id
,
int
CC_id
,
int
frame
,
int
subframe
,
int
sched_frame
,
int
sched_subframe
,
UE_list_t
*
UE_list
,
int
max_num_ue
,
int
num_contig_rb
,
contig_rbs_t
*
rbs
)
{
AssertFatal
(
num_contig_rb
<=
2
,
"cannot handle more than two contiguous RB regions
\n
"
);
UE_info_t
*
UE_info
=
&
RC
.
mac
[
Mod_id
]
->
UE_info
;
const
int
max_rb
=
num_contig_rb
>
1
?
MAX
(
rbs
[
0
].
length
,
rbs
[
1
].
length
)
:
rbs
[
0
].
length
;
/* for every UE: check whether we have to handle a retransmission (and
* allocate, if so). If not, compute how much RBs this UE would need */
int
rb_idx_required
[
MAX_MOBILES_PER_ENB
];
memset
(
rb_idx_required
,
0
,
sizeof
(
rb_idx_required
));
int
num_ue_req
=
0
;
for
(
int
UE_id
=
UE_list
->
head
;
UE_id
>=
0
;
UE_id
=
UE_list
->
next
[
UE_id
])
{
UE_TEMPLATE
*
UE_template
=
&
UE_info
->
UE_template
[
CC_id
][
UE_id
];
uint8_t
harq_pid
=
subframe2harqpid
(
&
RC
.
mac
[
Mod_id
]
->
common_channels
[
CC_id
],
sched_frame
,
sched_subframe
);
if
(
UE_info
->
UE_sched_ctrl
[
UE_id
].
round_UL
[
CC_id
][
harq_pid
]
>
0
)
{
/* this UE has a retransmission, allocate it right away */
const
int
nb_rb
=
UE_template
->
nb_rb_ul
[
harq_pid
];
if
(
nb_rb
==
0
)
{
LOG_E
(
MAC
,
"%4d.%d UE %d retransmission of 0 RBs in round %d, ignoring
\n
"
,
sched_frame
,
sched_subframe
,
UE_id
,
UE_info
->
UE_sched_ctrl
[
UE_id
].
round_UL
[
CC_id
][
harq_pid
]);
continue
;
}
if
(
rbs
[
0
].
length
>=
nb_rb
)
{
// fits in first contiguous region
UE_template
->
pre_first_nb_rb_ul
=
rbs
[
0
].
start
;
rbs
[
0
].
length
-=
nb_rb
;
rbs
[
0
].
start
+=
nb_rb
;
}
else
if
(
num_contig_rb
==
2
&&
rbs
[
1
].
length
>=
nb_rb
)
{
// in second
UE_template
->
pre_first_nb_rb_ul
=
rbs
[
1
].
start
;
rbs
[
1
].
length
-=
nb_rb
;
rbs
[
1
].
start
+=
nb_rb
;
}
else
if
(
num_contig_rb
==
2
&&
rbs
[
1
].
start
+
rbs
[
1
].
length
-
rbs
[
0
].
start
>=
nb_rb
)
{
// overlapping the middle
UE_template
->
pre_first_nb_rb_ul
=
rbs
[
0
].
start
;
rbs
[
0
].
length
=
0
;
int
ol
=
nb_rb
-
(
rbs
[
1
].
start
-
rbs
[
0
].
start
);
// how much overlap in second region
if
(
ol
>
0
)
{
rbs
[
1
].
length
-=
ol
;
rbs
[
1
].
start
+=
ol
;
}
}
else
{
LOG_W
(
MAC
,
"cannot allocate UL retransmission for UE %d (nb_rb %d)
\n
"
,
UE_id
,
nb_rb
);
continue
;
}
LOG_D
(
MAC
,
"%4d.%d UE %d retx %d RBs at start %d
\n
"
,
sched_frame
,
sched_subframe
,
UE_id
,
UE_template
->
pre_allocated_nb_rb_ul
,
UE_template
->
pre_first_nb_rb_ul
);
UE_template
->
pre_allocated_nb_rb_ul
=
nb_rb
;
max_num_ue
--
;
if
(
max_num_ue
==
0
)
/* in this case, cannot allocate any other UE anymore */
return
rbs
[
0
].
length
+
(
num_contig_rb
>
1
?
rbs
[
1
].
length
:
0
);
continue
;
}
const
int
B
=
cmax
(
UE_template
->
estimated_ul_buffer
-
UE_template
->
scheduled_ul_bytes
,
0
);
const
int
UE_to_be_scheduled
=
UE_is_to_be_scheduled
(
module_idP
,
CC_id
,
UE_id
);
const
int
UE_to_be_scheduled
=
UE_is_to_be_scheduled
(
Mod_id
,
CC_id
,
UE_id
);
if
(
B
==
0
&&
!
UE_to_be_scheduled
)
continue
;
num_ue_req
++
;
/* if UE has pending scheduling request then pre-allocate 3 RBs */
if
(
B
==
0
&&
UE_to_be_scheduled
)
{
UE_template
->
pre_assigned_mcs_ul
=
10
;
/* use QPSK mcs only */
UE_template
->
pre_allocated_rb_table_index_ul
=
2
;
UE_template
->
pre_allocated_nb_rb_ul
=
3
;
rb_idx_required
[
UE_id
]
=
2
;
//
UE_template->pre_allocated_nb_rb_ul = 3;
continue
;
}
...
...
@@ -560,28 +538,271 @@ assign_max_mcs_min_rb(module_id_t module_idP,
int
rb_table_index
;
int
tx_power
;
calculate_max_mcs_min_rb
(
module_idP
,
Mod_id
,
CC_id
,
B
,
UE_template
->
phr_info
,
UE_info
->
UE_sched_ctrl
[
UE_id
].
phr_received
==
1
?
20
:
10
,
&
mcs
,
available_rbs
,
max_rb
,
&
rb_table_index
,
&
tx_power
);
UE_template
->
pre_assigned_mcs_ul
=
mcs
;
UE_template
->
pre_allocated_rb_table_index_ul
=
rb_table_index
;
UE_template
->
pre_allocated_nb_rb_ul
=
rb_table
[
rb_table_index
];
LOG_D
(
MAC
,
"[eNB %d] frame %d subframe %d: for UE %d CC %d: pre-assigned mcs %d, pre-allocated rb_table[%d]=%d RBs (phr %d, tx power %d)
\n
"
,
module_idP
,
rb_idx_required
[
UE_id
]
=
rb_table_index
;
//UE_template->pre_allocated_nb_rb_ul = rb_table[rb_table_index];
/* only print log when PHR changed */
static
int
phr
=
0
;
if
(
phr
!=
UE_template
->
phr_info
)
{
phr
=
UE_template
->
phr_info
;
LOG_D
(
MAC
,
"%d.%d UE %d CC %d: pre mcs %d, pre rb_table[%d]=%d RBs (phr %d, tx power %d, bytes %d)
\n
"
,
frame
,
subframe
,
UE_id
,
CC_id
,
UE_template
->
pre_assigned_mcs_ul
,
UE_template
->
pre_allocated_rb_table_index_ul
,
UE_template
->
pre_allocated_nb_rb_ul
,
UE_template
->
phr_info
,
tx_power
,
B
);
}
}
if
(
num_ue_req
==
0
)
return
rbs
[
0
].
length
+
(
num_contig_rb
>
1
?
rbs
[
1
].
length
:
0
);
// calculate how many users should be in both regions, and to maximize usage,
// go from the larger to the smaller one which at least will handle a single
// full load case better.
const
int
n
=
min
(
num_ue_req
,
max_num_ue
);
int
nr
[
2
]
=
{
n
,
0
};
int
step
=
1
;
// the order if we have two regions
int
start
=
0
;
int
end
=
1
;
if
(
num_contig_rb
>
1
)
{
// proportionally divide between both regions
int
la
=
rbs
[
0
].
length
>
0
?
rbs
[
0
].
length
:
1
;
int
lb
=
rbs
[
1
].
length
>
0
?
rbs
[
1
].
length
:
1
;
nr
[
1
]
=
min
(
max
(
n
/
(
la
/
lb
+
1
),
1
),
n
-
1
);
nr
[
0
]
=
n
-
nr
[
1
];
step
=
la
>
lb
?
1
:
-
1
;
// 1: from 0 to 1, -1: from 1 to 0
start
=
la
>
lb
?
0
:
1
;
end
=
la
>
lb
?
2
:
-
1
;
}
if
(
g_start_ue_ul
==
-
1
)
g_start_ue_ul
=
UE_list
->
head
;
int
sUE_id
=
g_start_ue_ul
;
int
rb_idx_given
[
MAX_MOBILES_PER_ENB
];
memset
(
rb_idx_given
,
0
,
sizeof
(
rb_idx_given
));
for
(
int
r
=
start
;
r
!=
end
;
r
+=
step
)
{
// don't allocate if we have too little RBs
if
(
rbs
[
r
].
length
<
3
)
continue
;
if
(
nr
[
r
]
<=
0
)
continue
;
UE_list_t
UE_sched
;
// average RB index: just below the index that fits all UEs
int
start_idx
=
pp_find_rb_table_index
(
rbs
[
r
].
length
/
nr
[
r
])
-
1
;
int
num_ue_sched
=
0
;
int
rb_required_add
=
0
;
int
*
cur_UE
=
&
UE_sched
.
head
;
while
(
num_ue_sched
<
nr
[
r
])
{
while
(
rb_idx_required
[
sUE_id
]
==
0
)
sUE_id
=
next_ue_list_looped
(
UE_list
,
sUE_id
);
/* TODO: check that CCE allocated is feasible. If it is not, reduce
* nr[r] by one as this would been one opportunity */
*
cur_UE
=
sUE_id
;
cur_UE
=
&
UE_sched
.
next
[
sUE_id
];
rb_idx_given
[
sUE_id
]
=
min
(
start_idx
,
rb_idx_required
[
sUE_id
]);
rb_required_add
+=
rb_table
[
rb_idx_required
[
sUE_id
]]
-
rb_table
[
rb_idx_given
[
sUE_id
]];
rbs
[
r
].
length
-=
rb_table
[
rb_idx_given
[
sUE_id
]];
num_ue_sched
++
;
sUE_id
=
next_ue_list_looped
(
UE_list
,
sUE_id
);
}
*
cur_UE
=
-
1
;
/* give remaining RBs in RR fashion. Since we don't know in advance the
* amount of RBs we can give (the "step size" in rb_table is non-linear), go
* through all UEs and try to give a bit more. Continue until no UE can be
* given a higher index because the remaining RBs do not suffice to increase */
int
UE_id
=
UE_sched
.
head
;
int
rb_required_add_old
;
do
{
rb_required_add_old
=
rb_required_add
;
for
(
int
UE_id
=
UE_sched
.
head
;
UE_id
>=
0
;
UE_id
=
UE_sched
.
next
[
UE_id
])
{
if
(
rb_idx_given
[
UE_id
]
>=
rb_idx_required
[
UE_id
])
continue
;
// this UE does not need more
const
int
new_idx
=
rb_idx_given
[
UE_id
]
+
1
;
const
int
rb_inc
=
rb_table
[
new_idx
]
-
rb_table
[
rb_idx_given
[
UE_id
]];
if
(
rbs
[
r
].
length
<
rb_inc
)
continue
;
rb_idx_given
[
UE_id
]
=
new_idx
;
rbs
[
r
].
length
-=
rb_inc
;
rb_required_add
-=
rb_inc
;
}
}
while
(
rb_required_add
!=
rb_required_add_old
);
for
(
UE_id
=
UE_sched
.
head
;
UE_id
>=
0
;
UE_id
=
UE_sched
.
next
[
UE_id
])
{
UE_TEMPLATE
*
UE_template
=
&
UE_info
->
UE_template
[
CC_id
][
UE_id
];
/* MCS has been allocated previously */
UE_template
->
pre_first_nb_rb_ul
=
rbs
[
r
].
start
;
UE_template
->
pre_allocated_rb_table_index_ul
=
rb_idx_given
[
UE_id
];
UE_template
->
pre_allocated_nb_rb_ul
=
rb_table
[
rb_idx_given
[
UE_id
]];
rbs
[
r
].
start
+=
rb_table
[
rb_idx_given
[
UE_id
]];
LOG_D
(
MAC
,
"%4d.%d UE %d allocated %d RBs start %d new start %d
\n
"
,
sched_frame
,
sched_subframe
,
UE_id
,
UE_template
->
pre_allocated_nb_rb_ul
,
UE_template
->
pre_first_nb_rb_ul
,
rbs
[
r
].
start
);
}
}
/* if not all UEs could be allocated in this round */
if
(
num_ue_req
>
max_num_ue
)
{
/* go to the first one we missed */
for
(
int
i
=
0
;
i
<
max_num_ue
;
++
i
)
g_start_ue_ul
=
next_ue_list_looped
(
UE_list
,
g_start_ue_ul
);
}
else
{
/* else, just start with the next UE next time */
g_start_ue_ul
=
next_ue_list_looped
(
UE_list
,
g_start_ue_ul
);
}
return
rbs
[
0
].
length
+
(
num_contig_rb
>
1
?
rbs
[
1
].
length
:
0
);
}
void
ulsch_scheduler_pre_processor
(
module_id_t
Mod_id
,
int
CC_id
,
int
frameP
,
sub_frame_t
subframeP
,
int
sched_frameP
,
unsigned
char
sched_subframeP
)
{
UE_info_t
*
UE_info
=
&
RC
.
mac
[
Mod_id
]
->
UE_info
;
const
int
N_RB_UL
=
to_prb
(
RC
.
mac
[
Mod_id
]
->
common_channels
[
CC_id
].
ul_Bandwidth
);
COMMON_channels_t
*
cc
=
&
RC
.
mac
[
Mod_id
]
->
common_channels
[
CC_id
];
UE_list_t
UE_to_sched
;
UE_to_sched
.
head
=
-
1
;
for
(
int
i
=
0
;
i
<
MAX_MOBILES_PER_ENB
;
++
i
)
UE_to_sched
.
next
[
i
]
=
-
1
;
int
last_UE_id
=
-
1
;
for
(
int
UE_id
=
UE_info
->
list
.
head
;
UE_id
>=
0
;
UE_id
=
UE_info
->
list
.
next
[
UE_id
])
{
UE_TEMPLATE
*
UE_template
=
&
UE_info
->
UE_template
[
CC_id
][
UE_id
];
UE_sched_ctrl_t
*
ue_sched_ctrl
=
&
UE_info
->
UE_sched_ctrl
[
UE_id
];
/* initialize per-UE scheduling information */
UE_template
->
pre_assigned_mcs_ul
=
0
;
UE_template
->
pre_allocated_nb_rb_ul
=
0
;
UE_template
->
pre_allocated_rb_table_index_ul
=
-
1
;
UE_template
->
pre_first_nb_rb_ul
=
0
;
const
rnti_t
rnti
=
UE_RNTI
(
Mod_id
,
UE_id
);
if
(
rnti
==
NOT_A_RNTI
)
{
LOG_E
(
MAC
,
"UE %d has RNTI NOT_A_RNTI!
\n
"
,
UE_id
);
continue
;
}
if
(
ue_sched_ctrl
->
cdrx_configured
&&
!
ue_sched_ctrl
->
in_active_time
)
continue
;
if
(
UE_info
->
UE_template
[
CC_id
][
UE_id
].
rach_resource_type
>
0
)
continue
;
/* define UEs to schedule */
if
(
UE_to_sched
.
head
<
0
)
UE_to_sched
.
head
=
UE_id
;
else
UE_to_sched
.
next
[
last_UE_id
]
=
UE_id
;
UE_to_sched
.
next
[
UE_id
]
=
-
1
;
last_UE_id
=
UE_id
;
}
if
(
UE_to_sched
.
head
<
0
)
return
;
int
last_rb_blocked
=
1
;
int
n_contig
=
0
;
contig_rbs_t
rbs
[
2
];
// up to two contig RBs for PRACH in between
for
(
int
i
=
0
;
i
<
N_RB_UL
;
++
i
)
{
if
(
cc
->
vrb_map_UL
[
i
]
==
0
&&
last_rb_blocked
==
1
)
{
last_rb_blocked
=
0
;
n_contig
++
;
AssertFatal
(
n_contig
<=
2
,
"cannot handle more than two contiguous RB regions
\n
"
);
rbs
[
n_contig
-
1
].
start
=
i
;
}
if
(
cc
->
vrb_map_UL
[
i
]
==
1
&&
last_rb_blocked
==
0
)
{
last_rb_blocked
=
1
;
rbs
[
n_contig
-
1
].
length
=
i
-
rbs
[
n_contig
-
1
].
start
;
}
}
round_robin_ul
(
Mod_id
,
CC_id
,
frameP
,
subframeP
,
sched_frameP
,
sched_subframeP
,
&
UE_to_sched
,
3
,
// max_num_ue
n_contig
,
rbs
);
// the following block is meant for validation of the pre-processor to check
// whether all UE allocations are non-overlapping and is not necessary for
// scheduling functionality
#ifdef DEBUG_eNB_SCHEDULER
char
t
[
101
]
=
"__________________________________________________"
"__________________________________________________"
;
t
[
N_RB_UL
]
=
0
;
for
(
int
j
=
0
;
j
<
N_RB_UL
;
j
++
)
if
(
cc
->
vrb_map_UL
[
j
]
!=
0
)
t
[
j
]
=
'x'
;
int
print
=
0
;
for
(
int
UE_id
=
UE_info
->
list
.
head
;
UE_id
>=
0
;
UE_id
=
UE_info
->
list
.
next
[
UE_id
])
{
UE_TEMPLATE
*
UE_template
=
&
UE_info
->
UE_template
[
CC_id
][
UE_id
];
if
(
UE_template
->
pre_allocated_nb_rb_ul
==
0
)
continue
;
print
=
1
;
uint8_t
harq_pid
=
subframe2harqpid
(
&
RC
.
mac
[
Mod_id
]
->
common_channels
[
CC_id
],
sched_frameP
,
sched_subframeP
);
LOG_D
(
MAC
,
"%4d.%d UE%d %d RBs (index %d) at start %d, pre MCS %d %s
\n
"
,
frameP
,
subframeP
,
UE_id
,
CC_id
,
UE_template
->
pre_assigned_mcs_ul
,
UE_template
->
pre_allocated_rb_table_index_ul
,
UE_template
->
pre_allocated_nb_rb_ul
,
UE_template
->
phr_info
,
tx_power
);
UE_template
->
pre_allocated_rb_table_index_ul
,
UE_template
->
pre_first_nb_rb_ul
,
UE_template
->
pre_assigned_mcs_ul
,
UE_info
->
UE_sched_ctrl
[
UE_id
].
round_UL
[
CC_id
][
harq_pid
]
>
0
?
"(retx)"
:
""
);
for
(
int
i
=
0
;
i
<
UE_template
->
pre_allocated_nb_rb_ul
;
++
i
)
{
/* only check if this is not a retransmission */
if
(
UE_info
->
UE_sched_ctrl
[
UE_id
].
round_UL
[
CC_id
][
harq_pid
]
==
0
&&
cc
->
vrb_map_UL
[
UE_template
->
pre_first_nb_rb_ul
+
i
]
==
1
)
{
LOG_I
(
MAC
,
"%4d.%d UL scheduler allocation list: %s
\n
"
,
frameP
,
subframeP
,
t
);
LOG_E
(
MAC
,
"%4d.%d: UE %d allocated at locked RB %d (is: allocated start "
"%d/length %d)
\n
"
,
frameP
,
subframeP
,
UE_id
,
UE_template
->
pre_first_nb_rb_ul
+
i
,
UE_template
->
pre_first_nb_rb_ul
,
UE_template
->
pre_allocated_nb_rb_ul
);
}
cc
->
vrb_map_UL
[
UE_template
->
pre_first_nb_rb_ul
+
i
]
=
1
;
t
[
UE_template
->
pre_first_nb_rb_ul
+
i
]
=
UE_id
+
'0'
;
}
}
if
(
print
)
LOG_D
(
MAC
,
"%4d.%d UL scheduler allocation list: %s
\n
"
,
sched_frameP
,
sched_subframeP
,
t
);
#endif
}
This diff is collapsed.
Click to expand it.
Write
Preview
Markdown
is supported
0%
Try again
or
attach a new file
Attach a file
Cancel
You are about to add
0
people
to the discussion. Proceed with caution.
Finish editing this message first!
Cancel
Please
register
or
sign in
to comment