enjoy-digital / litex_agilex5_test

Initial Test/Support of LiteX on Intel Agilex5 FPGAs.

3 stars 0 forks source link

Resources Usage #2

Open trabucayre opened 3 months ago

trabucayre commented 3 months ago

VexRiscv SMP

Intel Agilex5E board

time ./intel_agilex5e_065b_premium_devkit.py --cpu-type=vexriscv_smp --build --with-rvc --with-fpu --with-wishbone-memory
real    7m54.943s
user    18m14.855s
sys     0m9.737s

Resource	Usage	%
Usage Report Generated after: Place

Logic utilization (ALMs needed / total ALMs on device)	6,341 / 222,400	3 %
ALMs needed [=A-B+C]	6,341
[A] ALMs used in final placement [=a+b+c+d]	7,538 / 222,400	3 %
[a] ALMs used for LUT logic and register circuitry	2,345
[b] ALMs used for LUT logic	3,752
[c] ALMs used for register circuitry	1,441
[d] ALMs used for memory (up to half of total ALMs)	0
[B] Estimate of ALMs recoverable by dense packing	1,230 / 222,400	< 1 %
[C] Estimate of ALMs unavailable [=a+b+c+d]	33 / 222,400	< 1 %
[a] Due to location constrained logic	0
[b] Due to LAB-wide signal conflicts	0
[c] Due to LAB input limits	33
[d] Due to virtual I/Os	0

Difficulty packing design	Low

Total LABs: partially or completely used	897 / 22,240	4 %
-- Logic LABs	897
-- Memory LABs (up to half of total LABs)	0

Combinational ALUT usage for logic	9,831
-- 8 input functions	296
-- 7 input functions	69
-- 6 input functions	892
-- 5 input functions	2,024
-- 4 input functions	1,556
-- <=3 input functions	4,994
Combinational ALUT usage for route-throughs	1,132

Dedicated logic registers	8,895
-- By type:
-- LAB logic registers:
-- Primary logic registers	7,571 / 444,800	2 %
-- Secondary logic registers	734 / 444,800	< 1 %
-- Hyper-Registers:	590

Register control circuitry for power estimation	0

ALMs adjustment for power estimation	730

I/O pins	5 / 624	< 1 %
-- Clock pins	0 / 44	0 %
-- Dedicated input pins	3 / 54	6 %

M20K blocks	31 / 1,611	2 %
Total MLAB memory bits	0
Total block memory bits	297,952 / 32,993,280	< 1 %
Total block memory implementation bits	634,880 / 32,993,280	2 %

DSP Blocks Needed [=A+B+C-D]	7 / 846	< 1 %
[A] Total Fixed Point DSP Blocks	13
[B] Total Floating Point DSP Blocks	0
[C] Total DSP_PRIME Blocks	0
[D] Estimate of DSP Blocks recoverable by dense merging	6

IOPLLs	0 / 15	0 %
Global signals	1
LVDS_RX blocks	0 / 192	0 %
HMC blocks	0 / 8	0 %
Maximum fan-out	9553
Highest non-global fan-out	503
Total fan-out	72506
Average fan-out	3.75

trabucayre commented 3 months ago

naxRiscv 32bits

Intel Agilex5E board

./intel_agilex5e_065b_premium_devkit.py --cpu-type=naxriscv --with-fpu --with-rvc --build
real    40m10.748s
user    113m37.792s
sys     1m12.900s

Resource	Usage	%
Usage Report Generated after: Place

Logic utilization (ALMs needed / total ALMs on device)	65,574 / 222,400	29 %
ALMs needed [=A-B+C]	65,574
[A] ALMs used in final placement [=a+b+c+d]	80,052 / 222,400	36 %
[a] ALMs used for LUT logic and register circuitry	7,572
[b] ALMs used for LUT logic	49,323
[c] ALMs used for register circuitry	23,027
[d] ALMs used for memory (up to half of total ALMs)	130
[B] Estimate of ALMs recoverable by dense packing	14,731 / 222,400	7 %
[C] Estimate of ALMs unavailable [=a+b+c+d]	253 / 222,400	< 1 %
[a] Due to location constrained logic	0
[b] Due to LAB-wide signal conflicts	0
[c] Due to LAB input limits	253
[d] Due to virtual I/Os	0

Difficulty packing design	Low

Total LABs: partially or completely used	13,842 / 22,240	62 %
-- Logic LABs	13,829
-- Memory LABs (up to half of total LABs)	13

Combinational ALUT usage for logic	70,349
-- 8 input functions	13,013
-- 7 input functions	175
-- 6 input functions	18,159
-- 5 input functions	18,685
-- 4 input functions	9,008
-- <=3 input functions	11,309
Combinational ALUT usage for route-throughs	7,884
Memory ALUT usage	73
-- 64-address deep	0
-- 32-address deep	0


Dedicated logic registers	64,853
-- By type:
-- LAB logic registers:
-- Primary logic registers	61,197 / 444,800	14 %
-- Secondary logic registers	2,441 / 444,800	< 1 %
-- Hyper-Registers:	1,215

Register control circuitry for power estimation	260

ALMs adjustment for power estimation	16,160

I/O pins	5 / 624	< 1 %
-- Clock pins	0 / 44	0 %
-- Dedicated input pins	3 / 54	6 %

M20K blocks	405 / 1,611	25 %
Total MLAB memory bits	2,160
Total block memory bits	1,718,140 / 32,993,280	5 %
Total block memory implementation bits	8,294,400 / 32,993,280	25 %

DSP Blocks Needed [=A+B+C-D]	7 / 846	< 1 %
[A] Total Fixed Point DSP Blocks	13
[B] Total Floating Point DSP Blocks	0
[C] Total DSP_PRIME Blocks	0
[D] Estimate of DSP Blocks recoverable by dense merging	6

IOPLLs	0 / 15	0 %
Global signals	1
LVDS_RX blocks	0 / 192	0 %
HMC blocks	0 / 8	0 %
Maximum fan-out	68871
Highest non-global fan-out	2174
Total fan-out	622733
Average fan-out	4.48

trabucayre commented 3 months ago

naxRiscv 64bits

time ./intel_agilex5e_065b_premium_devkit.py --cpu-type=naxriscv --xlen=64 --with-fpu --with-rvc --build
real    55m39.381s
user    157m54.672s
sys     2m9.537s

Resource	Usage	%
Usage Report Generated after: Place

Logic utilization (ALMs needed / total ALMs on device)	86,509 / 222,400	39 %
ALMs needed [=A-B+C]	86,509
[A] ALMs used in final placement [=a+b+c+d]	104,414 / 222,400	47 %
[a] ALMs used for LUT logic and register circuitry	8,717
[b] ALMs used for LUT logic	67,225
[c] ALMs used for register circuitry	28,342
[d] ALMs used for memory (up to half of total ALMs)	130
[B] Estimate of ALMs recoverable by dense packing	18,239 / 222,400	8 %
[C] Estimate of ALMs unavailable [=a+b+c+d]	334 / 222,400	< 1 %
[a] Due to location constrained logic	0
[b] Due to LAB-wide signal conflicts	0
[c] Due to LAB input limits	334
[d] Due to virtual I/Os	0

Difficulty packing design	Low

Total LABs: partially or completely used	18,538 / 22,240	83 %
-- Logic LABs	18,525
-- Memory LABs (up to half of total LABs)	13

Combinational ALUT usage for logic	91,513
-- 8 input functions	18,566
-- 7 input functions	246
-- 6 input functions	25,000
-- 5 input functions	24,771
-- 4 input functions	9,873
-- <=3 input functions	13,057
Combinational ALUT usage for route-throughs	9,270
Memory ALUT usage	73
-- 64-address deep	0
-- 32-address deep	0


Dedicated logic registers	77,750
-- By type:
-- LAB logic registers:
-- Primary logic registers	74,117 / 444,800	17 %
-- Secondary logic registers	2,365 / 444,800	< 1 %
-- Hyper-Registers:	1,268

Register control circuitry for power estimation	260

ALMs adjustment for power estimation	20,844

I/O pins	5 / 624	< 1 %
-- Clock pins	0 / 44	0 %
-- Dedicated input pins	3 / 54	6 %

M20K blocks	411 / 1,611	26 %
Total MLAB memory bits	2,160
Total block memory bits	1,728,444 / 32,993,280	5 %
Total block memory implementation bits	8,417,280 / 32,993,280	26 %

DSP Blocks Needed [=A+B+C-D]	13 / 846	2 %
[A] Total Fixed Point DSP Blocks	25
[B] Total Floating Point DSP Blocks	0
[C] Total DSP_PRIME Blocks	0
[D] Estimate of DSP Blocks recoverable by dense merging	12

IOPLLs	0 / 15	0 %
Global signals	1
LVDS_RX blocks	0 / 192	0 %
HMC blocks	0 / 8	0 %
Maximum fan-out	81892
Highest non-global fan-out	3078
Total fan-out	787253
Average fan-out	4.54

enjoy-digital commented 3 months ago

NaxRiscv 32-bit on Artix7:

./sqrl_acorn.py --integrated-main-ram-size=0x100 --cpu-type=naxriscv --with-fpu --with-rvc --build

1. Slice Logic
--------------

+----------------------------+-------+-------+------------+-----------+-------+
|          Site Type         |  Used | Fixed | Prohibited | Available | Util% |
+----------------------------+-------+-------+------------+-----------+-------+
| Slice LUTs                 | 23597 |     0 |        800 |    133800 | 17.64 |
|   LUT as Logic             | 18677 |     0 |        800 |    133800 | 13.96 |
|   LUT as Memory            |  4920 |     0 |          0 |     46200 | 10.65 |
|     LUT as Distributed RAM |  4874 |     0 |            |           |       |
|     LUT as Shift Register  |    46 |     0 |            |           |       |
| Slice Registers            | 17373 |     0 |          0 |    269200 |  6.45 |
|   Register as Flip Flop    | 17373 |     0 |          0 |    269200 |  6.45 |
|   Register as Latch        |     0 |     0 |          0 |    269200 |  0.00 |
| F7 Muxes                   |   278 |     0 |        400 |     66900 |  0.42 |
| F8 Muxes                   |     2 |     0 |        200 |     33450 | <0.01 |
+----------------------------+-------+-------+------------+-----------+-------+

1.1 Summary of Registers by Type
--------------------------------

+-------+--------------+-------------+--------------+
| Total | Clock Enable | Synchronous | Asynchronous |
+-------+--------------+-------------+--------------+
| 0     |            _ |           - |            - |
| 0     |            _ |           - |          Set |
| 0     |            _ |           - |        Reset |
| 0     |            _ |         Set |            - |
| 0     |            _ |       Reset |            - |
| 0     |          Yes |           - |            - |
| 87    |          Yes |           - |          Set |
| 1198  |          Yes |           - |        Reset |
| 90    |          Yes |         Set |            - |
| 15998 |          Yes |       Reset |            - |
+-------+--------------+-------------+--------------+

2. Slice Logic Distribution
---------------------------

+--------------------------------------------+-------+-------+------------+-----------+-------+
|                  Site Type                 |  Used | Fixed | Prohibited | Available | Util% |
+--------------------------------------------+-------+-------+------------+-----------+-------+
| Slice                                      |  8074 |     0 |        200 |     33450 | 24.14 |
|   SLICEL                                   |  4886 |     0 |            |           |       |
|   SLICEM                                   |  3188 |     0 |            |           |       |
| LUT as Logic                               | 18677 |     0 |        800 |    133800 | 13.96 |
|   using O5 output only                     |     3 |       |            |           |       |
|   using O6 output only                     | 15109 |       |            |           |       |
|   using O5 and O6                          |  3565 |       |            |           |       |
| LUT as Memory                              |  4920 |     0 |          0 |     46200 | 10.65 |
|   LUT as Distributed RAM                   |  4874 |     0 |            |           |       |
|     using O5 output only                   |     8 |       |            |           |       |
|     using O6 output only                   |  3206 |       |            |           |       |
|     using O5 and O6                        |  1660 |       |            |           |       |
|   LUT as Shift Register                    |    46 |     0 |            |           |       |
|     using O5 output only                   |     9 |       |            |           |       |
|     using O6 output only                   |    12 |       |            |           |       |
|     using O5 and O6                        |    25 |       |            |           |       |
| Slice Registers                            | 17373 |     0 |          0 |    269200 |  6.45 |
|   Register driven from within the Slice    |  7661 |       |            |           |       |
|   Register driven from outside the Slice   |  9712 |       |            |           |       |
|     LUT in front of the register is unused |  5072 |       |            |           |       |
|     LUT in front of the register is used   |  4640 |       |            |           |       |
| Unique Control Sets                        |   466 |       |        200 |     33450 |  1.39 |
+--------------------------------------------+-------+-------+------------+-----------+-------+
* * Note: Available Control Sets calculated as Slice * 1, Review the Control Sets Report for more information regarding control sets.

3. Memory
---------

+-------------------+------+-------+------------+-----------+-------+
|     Site Type     | Used | Fixed | Prohibited | Available | Util% |
+-------------------+------+-------+------------+-----------+-------+
| Block RAM Tile    | 63.5 |     0 |          0 |       365 | 17.40 |
|   RAMB36/FIFO*    |   57 |     0 |          0 |       365 | 15.62 |
|     RAMB36E1 only |   57 |       |            |           |       |
|   RAMB18          |   13 |     0 |          0 |       730 |  1.78 |
|     RAMB18E1 only |   13 |       |            |           |       |
+-------------------+------+-------+------------+-----------+-------+
* Note: Each Block RAM Tile only has one FIFO logic available and therefore can accommodate only one FIFO36E1 or one FIFO18E1. However, if a FIFO18E1 occupies a Block RAM Tile, that tile can still accommodate a RAMB18E1

4. DSP
------

+----------------+------+-------+------------+-----------+-------+
|    Site Type   | Used | Fixed | Prohibited | Available | Util% |
+----------------+------+-------+------------+-----------+-------+
| DSPs           |   13 |     0 |          0 |       740 |  1.76 |
|   DSP48E1 only |   13 |       |            |           |       |
+----------------+------+-------+------------+-----------+-------+

enjoy-digital commented 3 months ago

VexRiscv SMP on Artix7:

./sqrl_acorn.py --integrated-main-ram-size=0x100 --cpu-type=vexriscv_smp --build --with-rvc --with-fpu --with-wishbone-memory

1. Slice Logic
--------------

+----------------------------+------+-------+------------+-----------+-------+
|          Site Type         | Used | Fixed | Prohibited | Available | Util% |
+----------------------------+------+-------+------------+-----------+-------+
| Slice LUTs                 | 7755 |     0 |        800 |    133800 |  5.80 |
|   LUT as Logic             | 7644 |     0 |        800 |    133800 |  5.71 |
|   LUT as Memory            |  111 |     0 |          0 |     46200 |  0.24 |
|     LUT as Distributed RAM |   76 |     0 |            |           |       |
|     LUT as Shift Register  |   35 |     0 |            |           |       |
| Slice Registers            | 7840 |     0 |          0 |    269200 |  2.91 |
|   Register as Flip Flop    | 7840 |     0 |          0 |    269200 |  2.91 |
|   Register as Latch        |    0 |     0 |          0 |    269200 |  0.00 |
| F7 Muxes                   |   48 |     0 |        400 |     66900 |  0.07 |
| F8 Muxes                   |    5 |     0 |        200 |     33450 |  0.01 |
+----------------------------+------+-------+------------+-----------+-------+

1.1 Summary of Registers by Type
--------------------------------

+-------+--------------+-------------+--------------+
| Total | Clock Enable | Synchronous | Asynchronous |
+-------+--------------+-------------+--------------+
| 0     |            _ |           - |            - |
| 0     |            _ |           - |          Set |
| 0     |            _ |           - |        Reset |
| 0     |            _ |         Set |            - |
| 0     |            _ |       Reset |            - |
| 0     |          Yes |           - |            - |
| 36    |          Yes |           - |          Set |
| 511   |          Yes |           - |        Reset |
| 120   |          Yes |         Set |            - |
| 7173  |          Yes |       Reset |            - |
+-------+--------------+-------------+--------------+

2. Slice Logic Distribution
---------------------------

+--------------------------------------------+------+-------+------------+-----------+-------+
|                  Site Type                 | Used | Fixed | Prohibited | Available | Util% |
+--------------------------------------------+------+-------+------------+-----------+-------+
| Slice                                      | 3023 |     0 |        200 |     33450 |  9.04 |
|   SLICEL                                   | 1986 |     0 |            |           |       |
|   SLICEM                                   | 1037 |     0 |            |           |       |
| LUT as Logic                               | 7644 |     0 |        800 |    133800 |  5.71 |
|   using O5 output only                     |    8 |       |            |           |       |
|   using O6 output only                     | 5779 |       |            |           |       |
|   using O5 and O6                          | 1857 |       |            |           |       |
| LUT as Memory                              |  111 |     0 |          0 |     46200 |  0.24 |
|   LUT as Distributed RAM                   |   76 |     0 |            |           |       |
|     using O5 output only                   |    0 |       |            |           |       |
|     using O6 output only                   |   20 |       |            |           |       |
|     using O5 and O6                        |   56 |       |            |           |       |
|   LUT as Shift Register                    |   35 |     0 |            |           |       |
|     using O5 output only                   |    3 |       |            |           |       |
|     using O6 output only                   |    0 |       |            |           |       |
|     using O5 and O6                        |   32 |       |            |           |       |
| Slice Registers                            | 7840 |     0 |          0 |    269200 |  2.91 |
|   Register driven from within the Slice    | 3411 |       |            |           |       |
|   Register driven from outside the Slice   | 4429 |       |            |           |       |
|     LUT in front of the register is unused | 2520 |       |            |           |       |
|     LUT in front of the register is used   | 1909 |       |            |           |       |
| Unique Control Sets                        |  172 |       |        200 |     33450 |  0.51 |
+--------------------------------------------+------+-------+------------+-----------+-------+
* * Note: Available Control Sets calculated as Slice * 1, Review the Control Sets Report for more information regarding control sets.

3. Memory
---------

+-------------------+------+-------+------------+-----------+-------+
|     Site Type     | Used | Fixed | Prohibited | Available | Util% |
+-------------------+------+-------+------------+-----------+-------+
| Block RAM Tile    | 16.5 |     0 |          0 |       365 |  4.52 |
|   RAMB36/FIFO*    |   15 |     0 |          0 |       365 |  4.11 |
|     RAMB36E1 only |   15 |       |            |           |       |
|   RAMB18          |    3 |     0 |          0 |       730 |  0.41 |
|     RAMB18E1 only |    3 |       |            |           |       |
+-------------------+------+-------+------------+-----------+-------+
* Note: Each Block RAM Tile only has one FIFO logic available and therefore can accommodate only one FIFO36E1 or one FIFO18E1. However, if a FIFO18E1 occupies a Block RAM Tile, that tile can still accommodate a RAMB18E1

4. DSP
------

+----------------+------+-------+------------+-----------+-------+
|    Site Type   | Used | Fixed | Prohibited | Available | Util% |
+----------------+------+-------+------------+-----------+-------+
| DSPs           |   13 |     0 |          0 |       740 |  1.76 |
|   DSP48E1 only |   13 |       |            |           |       |
+----------------+------+-------+------------+-----------+-------+

trabucayre commented 3 months ago

VexiiRiscv 32bits (rv32imafc) on Intel Agilex5E board

time ./intel_agilex5e_065b_premium_devkit.py  --bus-standard=axi-lite \
    --cpu-type=vexiiriscv --cpu-variant=linux --vexii-args="--with-rvc --with-rvf" \
    --with-coherent-dma --build

Resource	Usage	%
Usage Report Generated after: Place

Logic utilization (ALMs needed / total ALMs on device)	9,609 / 222,400	4 %
ALMs needed [=A-B+C]	9,609
[A] ALMs used in final placement [=a+b+c+d]	12,073 / 222,400	5 %
[a] ALMs used for LUT logic and register circuitry	2,494
[b] ALMs used for LUT logic	6,137
[c] ALMs used for register circuitry	3,442
[d] ALMs used for memory (up to half of total ALMs)	0
[B] Estimate of ALMs recoverable by dense packing	2,501 / 222,400	1 %
[C] Estimate of ALMs unavailable [=a+b+c+d]	37 / 222,400	< 1 %
[a] Due to location constrained logic	0
[b] Due to LAB-wide signal conflicts	0
[c] Due to LAB input limits	37
[d] Due to virtual I/Os	0

Difficulty packing design	Low

Total LABs: partially or completely used	1,553 / 22,240	7 %
-- Logic LABs	1,553
-- Memory LABs (up to half of total LABs)	0

Combinational ALUT usage for logic	13,086
-- 8 input functions	607
-- 7 input functions	107
-- 6 input functions	1,710
-- 5 input functions	3,141
-- 4 input functions	2,550
-- <=3 input functions	4,971
Combinational ALUT usage for route-throughs	1,488

Dedicated logic registers	12,671
-- By type:
-- LAB logic registers:
-- Primary logic registers	11,872 / 444,800	3 %
-- Secondary logic registers	740 / 444,800	< 1 %
-- Hyper-Registers:	59

Register control circuitry for power estimation	0

ALMs adjustment for power estimation	1,947

I/O pins	5 / 624	< 1 %
-- Clock pins	0 / 44	0 %
-- Dedicated input pins	3 / 54	6 %

M20K blocks	67 / 1,611	4 %
Total MLAB memory bits	0
Total block memory bits	567,952 / 32,993,280	2 %
Total block memory implementation bits	1,372,160 / 32,993,280	4 %

DSP Blocks Needed [=A+B+C-D]	2 / 846	< 1 %
[A] Total Fixed Point DSP Blocks	4
[B] Total Floating Point DSP Blocks	0
[C] Total DSP_PRIME Blocks	0
[D] Estimate of DSP Blocks recoverable by dense merging	2

IOPLLs	0 / 15	0 %
Global signals	1
LVDS_RX blocks	0 / 192	0 %
HMC blocks	0 / 8	0 %
Maximum fan-out	14107
Highest non-global fan-out	3342
Total fan-out	116227
Average fan-out	4.27

Dolu1990 commented 3 months ago

Hi ^^

Logic utilization (ALMs needed / total ALMs on device) 65,574 / 222,400

So yes, now i'm sure there is some memory not being properly infered into mlab / blockram.

Especialy when comparing the number of registers : Dedicated logic registers 64,853 vs Slice Registers | 17373

I will take a look at this.

trabucayre commented 3 months ago

VexiiRiscv 32bits (rv32imafc) on Artix7

./litex_acorn_baseboard_mini.py --sys-clk-freq 100e6 --bus-standard=axi-lite --cpu-type=vexiiriscv --cpu-variant=linux --vexii-args="--with-rvc --with-rvf" --with-coherent-dma --build --integrated-main-ram-size=0x100

1. Slice Logic
--------------

+----------------------------+------+-------+------------+-----------+-------+
|          Site Type         | Used | Fixed | Prohibited | Available | Util% |
+----------------------------+------+-------+------------+-----------+-------+
| Slice LUTs                 | 8844 |     0 |        800 |    133800 |  6.61 |
|   LUT as Logic             | 8546 |     0 |        800 |    133800 |  6.39 |
|   LUT as Memory            |  298 |     0 |          0 |     46200 |  0.65 |
|     LUT as Distributed RAM |  226 |     0 |            |           |       |
|     LUT as Shift Register  |   72 |     0 |            |           |       |
| Slice Registers            | 7705 |     0 |          0 |    269200 |  2.86 |
|   Register as Flip Flop    | 7705 |     0 |          0 |    269200 |  2.86 |
|   Register as Latch        |    0 |     0 |          0 |    269200 |  0.00 |
| F7 Muxes                   |    0 |     0 |        400 |     66900 |  0.00 |
| F8 Muxes                   |    0 |     0 |        200 |     33450 |  0.00 |
+----------------------------+------+-------+------------+-----------+-------+

1.1 Summary of Registers by Type
--------------------------------

+-------+--------------+-------------+--------------+
| Total | Clock Enable | Synchronous | Asynchronous |
+-------+--------------+-------------+--------------+
| 0     |            _ |           - |            - |
| 0     |            _ |           - |          Set |
| 0     |            _ |           - |        Reset |
| 0     |            _ |         Set |            - |
| 0     |            _ |       Reset |            - |
| 0     |          Yes |           - |            - |
| 47    |          Yes |           - |          Set |
| 626   |          Yes |           - |        Reset |
| 81    |          Yes |         Set |            - |
| 6951  |          Yes |       Reset |            - |
+-------+--------------+-------------+--------------+

2. Slice Logic Distribution
---------------------------

+--------------------------------------------+------+-------+------------+-----------+-------+
|                  Site Type                 | Used | Fixed | Prohibited | Available | Util% |
+--------------------------------------------+------+-------+------------+-----------+-------+
| Slice                                      | 3001 |     0 |        200 |     33450 |  8.97 |
|   SLICEL                                   | 1980 |     0 |            |           |       |
|   SLICEM                                   | 1021 |     0 |            |           |       |
| LUT as Logic                               | 8546 |     0 |        800 |    133800 |  6.39 |
|   using O5 output only                     |    2 |       |            |           |       |
|   using O6 output only                     | 7105 |       |            |           |       |
|   using O5 and O6                          | 1439 |       |            |           |       |
| LUT as Memory                              |  298 |     0 |          0 |     46200 |  0.65 |
|   LUT as Distributed RAM                   |  226 |     0 |            |           |       |
|     using O5 output only                   |    0 |       |            |           |       |
|     using O6 output only                   |    2 |       |            |           |       |
|     using O5 and O6                        |  224 |       |            |           |       |
|   LUT as Shift Register                    |   72 |     0 |            |           |       |
|     using O5 output only                   |   29 |       |            |           |       |
|     using O6 output only                   |   19 |       |            |           |       |
|     using O5 and O6                        |   24 |       |            |           |       |
| Slice Registers                            | 7705 |     0 |          0 |    269200 |  2.86 |
|   Register driven from within the Slice    | 3877 |       |            |           |       |
|   Register driven from outside the Slice   | 3828 |       |            |           |       |
|     LUT in front of the register is unused | 1921 |       |            |           |       |
|     LUT in front of the register is used   | 1907 |       |            |           |       |
| Unique Control Sets                        |  172 |       |        200 |     33450 |  0.51 |
+--------------------------------------------+------+-------+------------+-----------+-------+
* * Note: Available Control Sets calculated as Slice * 1, Review the Control Sets Report for more information regarding control sets.

3. Memory
---------

+-------------------+------+-------+------------+-----------+-------+
|     Site Type     | Used | Fixed | Prohibited | Available | Util% |
+-------------------+------+-------+------------+-----------+-------+
| Block RAM Tile    | 35.5 |     0 |          0 |       365 |  9.73 |
|   RAMB36/FIFO*    |   23 |     0 |          0 |       365 |  6.30 |
|     RAMB36E1 only |   23 |       |            |           |       |
|   RAMB18          |   25 |     0 |          0 |       730 |  3.42 |
|     RAMB18E1 only |   25 |       |            |           |       |
+-------------------+------+-------+------------+-----------+-------+
* Note: Each Block RAM Tile only has one FIFO logic available and therefore can accommodate only one FIFO36E1 or one FIFO18E1. However, if a FIFO18E1 occupies a Block RAM Tile, that tile can still accommodate a RAMB18E1

4. DSP
------

+----------------+------+-------+------------+-----------+-------+
|    Site Type   | Used | Fixed | Prohibited | Available | Util% |
+----------------+------+-------+------------+-----------+-------+
| DSPs           |    4 |     0 |          0 |       740 |  0.54 |
|   DSP48E1 only |    4 |       |            |           |       |
+----------------+------+-------+------------+-----------+-------+

Dolu1990 commented 3 months ago

Hmm same story for vexiiRiscv, it is less showing than for naxriscv, but for sure, some memory aren't being infered as memory. I will take a look at that aswell.

Dolu1990 commented 3 months ago

Found a few issues :

Memories with asyncronous read aren't being infered as mlab (ramstyle = "MLAB, no_rw_check" fixes it)
By default Quartus seems to assume that read during write on MLAB will not happen (to improve timing) at the risk of creating metastable design !? Which seems crazy to me, if i understand well. There is an option to enable "MLAB Add Timing Constraints For Mixed-Port Feed-Through Mode Setting Don't Care" to avoid this (advanced fitter settings)
Memories with more than 2 asyncronous read ports aren't reconized by quartus (Fixed by automatic blackboxification in SpinalHDL + Ram_1w_1ra_Generic.v)

This reduce the gap quite a lot. So, still, with a debian capable dual core, with very very relaxed timings and retiming disable, i still get 4k more register used (16% more) than on Artix 7. Which is weird as register usage is kinda a clean metric to compare things I'm working on that.

enjoy-digital commented 3 months ago

Thanks @Dolu1990 for the first analysis!

Dolu1990 commented 3 months ago

I pushed the current WIP (https://github.com/enjoy-digital/litex/pull/2011)

enjoy-digital commented 3 months ago

Good, thanks!

Dolu1990 commented 3 months ago

./intel_agilex5e_065b_premium_devkit.py --cpu-type=naxriscv --with-fpu --with-rvc --build
real    40m10.748s
user    113m37.792s
sys     1m12.900s

+----------------------------------------------------------------------------------------------+
; Fitter Resource Usage Summary                                                                ;
+-------------------------------------------------------------+------------------------+-------+
; Resource                                                    ; Usage                  ; %     ;
+-------------------------------------------------------------+------------------------+-------+
; Usage Report Generated after: Place                         ;                        ;       ;
;                                                             ;                        ;       ;
; Logic utilization (ALMs needed / total ALMs on device)      ; 30,153 / 222,400       ; 14 %  ;
; ALMs needed [=A-B+C]                                        ; 30,153                 ;       ;
;     [A] ALMs used in final placement [=a+b+c+d]             ; 34,334 / 222,400       ; 15 %  ;
;         [a] ALMs used for LUT logic and register circuitry  ; 6,155                  ;       ;
;         [b] ALMs used for LUT logic                         ; 12,787                 ;       ;
;         [c] ALMs used for register circuitry                ; 4,752                  ;       ;
;         [d] ALMs used for memory (up to half of total ALMs) ; 10,640                 ;       ;
;     [B] Estimate of ALMs recoverable by dense packing       ; 4,342 / 222,400        ; 2 %   ;
;     [C] Estimate of ALMs unavailable [=a+b+c+d]             ; 161 / 222,400          ; < 1 % ;
;         [a] Due to location constrained logic               ; 0                      ;       ;
;         [b] Due to LAB-wide signal conflicts                ; 30                     ;       ;
;         [c] Due to LAB input limits                         ; 131                    ;       ;
;         [d] Due to virtual I/Os                             ; 0                      ;       ;
;                                                             ;                        ;       ;
; Difficulty packing design                                   ; Low                    ;       ;
;                                                             ;                        ;       ;
; Total LABs:  partially or completely used                   ; 4,073 / 22,240         ; 18 %  ;
;     -- Logic LABs                                           ; 3,009                  ;       ;
;     -- Memory LABs (up to half of total LABs)               ; 1,064                  ;       ;
;                                                             ;                        ;       ;
; Combinational ALUT usage for logic                          ; 30,003                 ;       ;
;     -- 8 input functions                                    ; 1,299                  ;       ;
;     -- 7 input functions                                    ; 131                    ;       ;
;     -- 6 input functions                                    ; 2,429                  ;       ;
;     -- 5 input functions                                    ; 6,582                  ;       ;
;     -- 4 input functions                                    ; 6,888                  ;       ;
;     -- <=3 input functions                                  ; 12,674                 ;       ;
; Combinational ALUT usage for route-throughs                 ; 2,582                  ;       ;
; Memory ALUT usage                                           ; 6,675                  ;       ;
;     -- 64-address deep                                      ; 0                      ;       ;
;     -- 32-address deep                                      ; 0                      ;       ;
;                                                             ;                        ;       ;
;                                                             ;                        ;       ;
; Dedicated logic registers                                   ; 24,900                 ;       ;
;     -- By type:                                             ;                        ;       ;
;         -- LAB logic registers:                             ;                        ;       ;
;             -- Primary logic registers                      ; 21,813 / 444,800       ; 5 %   ;
;             -- Secondary logic registers                    ; 1,361 / 444,800        ; < 1 % ;
;         -- Hyper-Registers:                                 ; 1,726                  ;       ;
;                                                             ;                        ;       ;
; Register control circuitry for power estimation             ; 20,935                 ;       ;
;                                                             ;                        ;       ;
; ALMs adjustment for power estimation                        ; 3,081                  ;       ;
;                                                             ;                        ;       ;
; I/O pins                                                    ; 63 / 624               ; 10 %  ;
;     -- Clock pins                                           ; 1 / 44                 ; 2 %   ;
;     -- Dedicated input pins                                 ; 0 / 54                 ; 0 %   ;
;                                                             ;                        ;       ;
; M20K blocks                                                 ; 129 / 1,611            ; 8 %   ;
; Total MLAB memory bits                                      ; 197,064                ;       ;
; Total block memory bits                                     ; 1,663,332 / 32,993,280 ; 5 %   ;
; Total block memory implementation bits                      ; 2,641,920 / 32,993,280 ; 8 %   ;
;                                                             ;                        ;       ;
; DSP Blocks Needed [=A+B+C-D]                                ; 7 / 846                ; < 1 % ;
;     [A] Total Fixed Point DSP Blocks                        ; 13                     ;       ;
;     [B] Total Floating Point DSP Blocks                     ; 0                      ;       ;
;     [C] Total DSP_PRIME Blocks                              ; 0                      ;       ;
;     [D] Estimate of DSP Blocks recoverable by dense merging ; 6                      ;       ;
;                                                             ;                        ;       ;
; IOPLLs                                                      ; 1 / 15                 ; 7 %   ;
; Global signals                                              ; 2                      ;       ;
; LVDS_RX blocks                                              ; 0 / 192                ; 0 %   ;
; HMC blocks                                                  ; 1 / 8                  ; 13 %  ;
; Maximum fan-out                                             ; 34130                  ;       ;
; Highest non-global fan-out                                  ; 1259                   ;       ;
; Total fan-out                                               ; 298517                 ;       ;
; Average fan-out                                             ; 4.98                   ;       ;
+-------------------------------------------------------------+------------------------+-------+

So, ALM usage reduced by 53%. But there is still something wrong. I will take a look.

Dolu1990 commented 3 months ago

Updated https://github.com/enjoy-digital/litex/pull/2011

NaxRiscv lost some weight :


+----------------------------------------------------------------------------------------------+
; Fitter Resource Usage Summary                                                                ;
+-------------------------------------------------------------+------------------------+-------+
; Resource                                                    ; Usage                  ; %     ;
+-------------------------------------------------------------+------------------------+-------+
; Usage Report Generated after: Place                         ;                        ;       ;
;                                                             ;                        ;       ;
; Logic utilization (ALMs needed / total ALMs on device)      ; 25,814 / 222,400       ; 12 %  ;
; ALMs needed [=A-B+C]                                        ; 25,814                 ;       ;
;     [A] ALMs used in final placement [=a+b+c+d]             ; 29,880 / 222,400       ; 13 %  ;
;         [a] ALMs used for LUT logic and register circuitry  ; 6,330                  ;       ;
;         [b] ALMs used for LUT logic                         ; 13,393                 ;       ;
;         [c] ALMs used for register circuitry                ; 4,187                  ;       ;
;         [d] ALMs used for memory (up to half of total ALMs) ; 5,970                  ;       ;
;     [B] Estimate of ALMs recoverable by dense packing       ; 4,222 / 222,400        ; 2 %   ;
;     [C] Estimate of ALMs unavailable [=a+b+c+d]             ; 156 / 222,400          ; < 1 % ;
;         [a] Due to location constrained logic               ; 0                      ;       ;
;         [b] Due to LAB-wide signal conflicts                ; 26                     ;       ;
;         [c] Due to LAB input limits                         ; 130                    ;       ;
;         [d] Due to virtual I/Os                             ; 0                      ;       ;
;                                                             ;                        ;       ;
; Difficulty packing design                                   ; Low                    ;       ;
;                                                             ;                        ;       ;
; Total LABs:  partially or completely used                   ; 3,520 / 22,240         ; 16 %  ;
;     -- Logic LABs                                           ; 2,923                  ;       ;
;     -- Memory LABs (up to half of total LABs)               ; 597                    ;       ;
;                                                             ;                        ;       ;
; Combinational ALUT usage for logic                          ; 31,218                 ;       ;
;     -- 8 input functions                                    ; 1,333                  ;       ;
;     -- 7 input functions                                    ; 131                    ;       ;
;     -- 6 input functions                                    ; 2,401                  ;       ;
;     -- 5 input functions                                    ; 7,167                  ;       ;
;     -- 4 input functions                                    ; 7,191                  ;       ;
;     -- <=3 input functions                                  ; 12,995                 ;       ;
; Combinational ALUT usage for route-throughs                 ; 2,520                  ;       ;
; Memory ALUT usage                                           ; 6,208                  ;       ;
;     -- 64-address deep                                      ; 0                      ;       ;
;     -- 32-address deep                                      ; 0                      ;       ;
;                                                             ;                        ;       ;
;                                                             ;                        ;       ;
; Dedicated logic registers                                   ; 24,104                 ;       ;
;     -- By type:                                             ;                        ;       ;
;         -- LAB logic registers:                             ;                        ;       ;
;             -- Primary logic registers                      ; 21,034 / 444,800       ; 5 %   ;
;             -- Secondary logic registers                    ; 1,354 / 444,800        ; < 1 % ;
;         -- Hyper-Registers:                                 ; 1,716                  ;       ;
;                                                             ;                        ;       ;
; Register control circuitry for power estimation             ; 11,595                 ;       ;
;                                                             ;                        ;       ;
; ALMs adjustment for power estimation                        ; 2,797                  ;       ;
;                                                             ;                        ;       ;
; I/O pins                                                    ; 63 / 624               ; 10 %  ;
;     -- Clock pins                                           ; 1 / 44                 ; 2 %   ;
;     -- Dedicated input pins                                 ; 0 / 54                 ; 0 %   ;
;                                                             ;                        ;       ;
; M20K blocks                                                 ; 129 / 1,611            ; 8 %   ;
; Total MLAB memory bits                                      ; 182,092                ;       ;
; Total block memory bits                                     ; 1,663,336 / 32,993,280 ; 5 %   ;
; Total block memory implementation bits                      ; 2,641,920 / 32,993,280 ; 8 %   ;
;                                                             ;                        ;       ;
; DSP Blocks Needed [=A+B+C-D]                                ; 7 / 846                ; < 1 % ;
;     [A] Total Fixed Point DSP Blocks                        ; 13                     ;       ;
;     [B] Total Floating Point DSP Blocks                     ; 0                      ;       ;
;     [C] Total DSP_PRIME Blocks                              ; 0                      ;       ;
;     [D] Estimate of DSP Blocks recoverable by dense merging ; 6                      ;       ;
;                                                             ;                        ;       ;
; IOPLLs                                                      ; 1 / 15                 ; 7 %   ;
; Global signals                                              ; 2                      ;       ;
; LVDS_RX blocks                                              ; 0 / 192                ; 0 %   ;
; HMC blocks                                                  ; 1 / 8                  ; 13 %  ;
; Maximum fan-out                                             ; 32869                  ;       ;
; Highest non-global fan-out                                  ; 1259                   ;       ;
; Total fan-out                                               ; 296698                 ;       ;
; Average fan-out                                             ; 4.97                   ;       ;
+-------------------------------------------------------------+------------------------+-------+

gsteiert commented 2 months ago

Found a few issues :

Memories with asyncronous read aren't being infered as mlab (ramstyle = "MLAB, no_rw_check" fixes it)

By default Quartus seems to assume that read during write on MLAB will not happen (to improve timing) at the risk of creating metastable design !? Which seems crazy to me, if i understand well. There is an option to enable "MLAB Add Timing Constraints For Mixed-Port Feed-Through Mode Setting Don't Care" to avoid this (advanced fitter settings)

Memories with more than 2 asyncronous read ports aren't reconized by quartus (Fixed by automatic blackboxification in SpinalHDL + Ram_1w_1ra_Generic.v)

This reduce the gap quite a lot. So, still, with a debian capable dual core, with very very relaxed timings and retiming disable, i still get 4k more register used (16% more) than on Artix 7. Which is weird as register usage is kinda a clean metric to compare things I'm working on that.

@Dolu1990 This is great insight. Can you summarize what needed to change, or point to the changes you made in the repo for this? I would like to share this with our tools team.

Dolu1990 commented 2 months ago

@gsteiert Related to the lack of MLAB inferation I had to :

Enable SpinalHDL blackboxing of memories which have asyncronous read https://github.com/SpinalHDL/NaxRiscv/commit/43195dd1f11338da2b22c90be0b2ceaaa296f8d8#diff-2193d1fc41a2107106e208c58f322dd903e2c904e0b261d85520e2f62d035767R97
Add the blackbox implementation in litex https://github.com/litex-hub/pythondata-cpu-naxriscv/blob/main/pythondata_cpu_naxriscv/verilog/Ram_1w_1ra_Generic.v

That was mostly it. Another "big" change i did is to rework how the naxriscv register renaming design : https://github.com/SpinalHDL/NaxRiscv/commit/ba63ee6dfb063e0e6b1c8da51071a85dea9f934b

The "MLAB Add Timing Constraints For Mixed-Port Feed-Through Mode Setting Don't Care" isn't fixed as far as i know, that would be a thing to handle in the quartus project itself.

Dolu1990 commented 2 months ago

Note that the SpinalHDL blackboxification of memories will decompose the 1 write + 3 async read into 3 * blackbox(1 write + 1 async read)

Dolu1990 commented 2 months ago

More generaly, when i have this verilog :

    always @ (posedge clk) begin
        if(wr_en) begin
           ram_block[wr_addr] <= wr_data;
        end
    end

    assign rd_data = ram_block[rd_addr];

I do assume that the synthesis tools would infer things into lut based RAM and that :

if i write at cycle 1, then i can read that written value at cycle 2 (same address)
if i write and read at cycle 1, then the readed value is the old value

I did had some issue with xilinx tool in the past, where they were infering it as a block ram (by merging the register which drive rd_addr) and violating the rule :

if i write at cycle 1, then i can read that written value at cycle 2 (same address)

I have to say, it is nice that NaxRiscv worked on Altera FPGA without having to debug things ^.^

Dolu1990 commented 2 months ago

I may have missed the reasons why without the ( ramstyle = "MLAB ) tag, quartus refused to infer things as mlab. Fondamentaly, that was the main thing which was making the area crazy high.