* [libre-riscv-dev] Understanding the LibreSOC core
From: Sanjay Menon @ 2020-06-08 18:03 UTC (permalink / raw)
To: libre-riscv-dev
Hi,
Currently making a little progress on the nMigen tool.
Got to see some interesting things been discussed on the Alexa IOT pitch
for LibreSOC.
I have a few questions for you guys
1.Where do I start if I where to check into the LibreSoC source code?
2. Is this SoC in any way similar to the Xilinx ZYNQ architecture?.
Where the ARM embedded processor would be replaced by PowerPC and the
reconfigurable logic is converted to a GPU like entity, together forming a
heterogeneous SoC.
3. Do you plan to run the SoC on an FPGA for prototyping and testing
purposes?, or are you using some other methodologies?
Thanks and regards,
Sanjay
_______________________________________________
libre-riscv-dev mailing list
libre-riscv-dev@lists.libre-riscv.org
http://lists.libre-riscv.org/mailman/listinfo/libre-riscv-dev
^ permalink raw reply
* [libre-riscv-dev] [Bug 368] Need one example unit test of how to run some assembly code "qemu vs simulator" rather than "qemu vs hardware"
From: bugzilla-daemon @ 2020-06-08 18:00 UTC (permalink / raw)
To: libre-riscv-dev
In-Reply-To: <bug-368-13@https.bugs.libre-soc.org/>
https://bugs.libre-soc.org/show_bug.cgi?id=368
--- Comment #7 from Luke Kenneth Casson Leighton <lkcl@lkcl.net> ---
(In reply to Michael Nolan from comment #6)
> Ok, I think I've got something working. In soc/simulator/test_sim.py, I
> replaced the code dealing with my handwritten simulator with what's needed
> to use the pseudocode simulator. All the tests, except for multiplication
> now work
fantastic. now we're cooking. ultimately it should be possible to morph
this into classes that have a "set inputs" method, a "run_program" method,
and a "get results" method.
then we can construct unit tests that can be passed in one class instance
that does qemu, and another that does hardware, another that does simulator,
and compare anything-against-anything.
--
You are receiving this mail because:
You are on the CC list for the bug.
_______________________________________________
libre-riscv-dev mailing list
libre-riscv-dev@lists.libre-riscv.org
http://lists.libre-riscv.org/mailman/listinfo/libre-riscv-dev
^ permalink raw reply
* [libre-riscv-dev] [Bug 369] missing XER SO/OV/32 check in test_pipe_caller.py
From: bugzilla-daemon @ 2020-06-08 17:57 UTC (permalink / raw)
To: libre-riscv-dev
In-Reply-To: <bug-369-13@https.bugs.libre-soc.org/>
https://bugs.libre-soc.org/show_bug.cgi?id=369
--- Comment #4 from Luke Kenneth Casson Leighton <lkcl@lkcl.net> ---
found it.
--- a/src/soc/decoder/isa/caller.py
+++ b/src/soc/decoder/isa/caller.py
@@ -380,9 +380,8 @@ class ISACaller:
carry_en = yield self.dec2.e.output_carry
if carry_en:
yield from self.handle_carry_(inputs, results, already_done)
- ov_en = yield self.dec2.e.oe.oe
- ov_ok = yield self.dec2.e.oe.ok
- if ov_en & ov_ok:
+ ov_en = yield self.dec2.e.oe
+ if ov_en:
--
You are receiving this mail because:
You are on the CC list for the bug.
_______________________________________________
libre-riscv-dev mailing list
libre-riscv-dev@lists.libre-riscv.org
http://lists.libre-riscv.org/mailman/listinfo/libre-riscv-dev
^ permalink raw reply
* [libre-riscv-dev] [Bug 368] Need one example unit test of how to run some assembly code "qemu vs simulator" rather than "qemu vs hardware"
From: bugzilla-daemon @ 2020-06-08 17:48 UTC (permalink / raw)
To: libre-riscv-dev
In-Reply-To: <bug-368-13@https.bugs.libre-soc.org/>
https://bugs.libre-soc.org/show_bug.cgi?id=368
--- Comment #6 from Michael Nolan <mtnolan2640@gmail.com> ---
Ok, I think I've got something working. In soc/simulator/test_sim.py, I
replaced the code dealing with my handwritten simulator with what's needed to
use the pseudocode simulator. All the tests, except for multiplication now work
--
You are receiving this mail because:
You are on the CC list for the bug.
_______________________________________________
libre-riscv-dev mailing list
libre-riscv-dev@lists.libre-riscv.org
http://lists.libre-riscv.org/mailman/listinfo/libre-riscv-dev
^ permalink raw reply
* Re: [libre-riscv-dev] Request for input and technical expertise for Systèmes Libres Amazon Alexa IOT Pitch 10-JUN-2020
From: Luke Kenneth Casson Leighton @ 2020-06-08 15:27 UTC (permalink / raw)
To: Libre-RISCV General Development
In-Reply-To: <20200608135554.54lyyd5wd3xlzi4m@topoi.pooq.com>
(btw thank you staf for the insights)
On Mon, Jun 8, 2020 at 2:56 PM Hendrik Boom <hendrik@topoi.pooq.com> wrote:
>
> On Mon, Jun 08, 2020 at 12:13:48PM +0100, Luke Kenneth Casson Leighton wrote:
> >
> > all of these things are at the architectural level. we are not doing
> > anything fancy at the gate level. it is a matter of making
> > *architectural* decisions that reduce power consumption, operating
> > within the exact same gate-level power consumption constraints as
> > every other ASIC out there.
>
> You point out that the CPU and GPU share cache, being the same processor.
yes. or: the CPU instructions and GPU instructions, by being in the
same ISA, the GPU *workload* will push CPU workload(s) out of the
(same) L1 Cache.
>
> But we are designing a four-core chip?
yes. therefore there will be 1x L1 Data and 1x L1 Instruction Cache
per each of those four cores.
> To what extent to the four cores share cache?
L1? not at all - ever.
> And on avoiding data copying between CPU ad GPU:
>
> I believe the OpenGL API involves copying data from CPU buffers to GPU
> buffers, with the understanding that the CPU copies can be discarded
> while the GPU goes on with its copy.
... because the assumption is that the GPU is a completely and utterly
separate processor, the "command" to perform that copy is expected to
involve the excruciatingly-painful process previously mentioned, from
which i excluded the userspace-kernelspace context-switching so as not
to have people run away screaming in terror.
in our case: it would simply be... a memcpy.
> Having the same storage for both sets of buffers could obviously obviate
> these copies, except that software that uses this API will likely rely
> on being able to overwrite the CPU-side buffers with impunity. So the
> copy will still have to be done.
sounds reasonable to me. actually now that i think about it, if the
buffer is placed into a shmem segment with copy-on-write semantics,
the memcpy will not be needed, and the "overwriting", because of the
CoW semantics, would only be done on-demand.
this however would be an optimisation.
the *only reason* that we can even remotely consider such an
optimisation is precisely because of the hybrid architecture.
> Do I misunderstand OpenGL? Is Vulcan different?
don't know.
> Will users want to bypass these libraries and use the graphics instructios directly?
only if they want to become an assembly-level expert, with all the
inherent implications and performance-complexity tradeoffs that always
come with doing assembly programming.
> Or is there some other sublety I'm missing?
no idea :)
l.
_______________________________________________
libre-riscv-dev mailing list
libre-riscv-dev@lists.libre-riscv.org
http://lists.libre-riscv.org/mailman/listinfo/libre-riscv-dev
^ permalink raw reply
* [libre-riscv-dev] [Bug 369] missing XER SO/OV/32 check in test_pipe_caller.py
From: bugzilla-daemon @ 2020-06-08 14:28 UTC (permalink / raw)
To: libre-riscv-dev
In-Reply-To: <bug-369-13@https.bugs.libre-soc.org/>
https://bugs.libre-soc.org/show_bug.cgi?id=369
Luke Kenneth Casson Leighton <lkcl@lkcl.net> changed:
What |Removed |Added
----------------------------------------------------------------------------
CC| |mtnolan2640@gmail.com
--- Comment #3 from Luke Kenneth Casson Leighton <lkcl@lkcl.net> ---
in decoder/isa/caller.py does this look reasonable? it does seem to match
the hardware version (i am however seeing subf fail in test_core.py - am
assuming for now that this is a separate issue)
+ # OV (64-bit)
input_sgn = [exts(x.value, x.bits) < 0 for x in inputs]
output_sgn = exts(output.value, output.bits) < 0
ov = 1 if input_sgn[0] == input_sgn[1] and \
output_sgn != input_sgn[0] else 0
+ # OV (32-bit)
+ input32_sgn = [exts(x.value, 32) < 0 for x in inputs]
+ output32_sgn = exts(output.value, 32) < 0
+ ov32 = 1 if input32_sgn[0] == input32_sgn[1] and \
+ output32_sgn != input32_sgn[0] else 0
+
self.spr['XER'][XER_bits['OV']] = ov
+ self.spr['XER'][XER_bits['OV32']] = ov32
--
You are receiving this mail because:
You are on the CC list for the bug.
_______________________________________________
libre-riscv-dev mailing list
libre-riscv-dev@lists.libre-riscv.org
http://lists.libre-riscv.org/mailman/listinfo/libre-riscv-dev
^ permalink raw reply
* Re: [libre-riscv-dev] Request for input and technical expertise for Systèmes Libres Amazon Alexa IOT Pitch 10-JUN-2020
From: Lauri Kasanen @ 2020-06-08 14:15 UTC (permalink / raw)
To: libre-riscv-dev
In-Reply-To: <20200608135554.54lyyd5wd3xlzi4m@topoi.pooq.com>
On Mon, 8 Jun 2020 09:55:54 -0400
Hendrik Boom <hendrik@topoi.pooq.com> wrote:
> And on avoiding data copying between CPU ad GPU:
>
> I believe the OpenGL API involves copying data from CPU buffers to GPU
> buffers, with the understanding that the CPU copies can be discarded
> while the GPU goes on with its copy.
>
> Having the same storage for both sets of buffers could obviously obviate
> these copies, except that software that uses this API will likely rely
> on being able to overwrite the CPU-side buffers with impunity. So the
> copy will still have to be done.
>
> Do I misunderstand OpenGL? Is Vulcan different? Will users want to
> bypass these libraries and use the graphics instructios directly? Or
> is there some other sublety I'm missing?
Your understanding is correct at a high level, one way of using OpenGL
will have to do that memcpy even on the Libresoc. Vulkan is explicit and
there are other ways in OpenGL too.
However you're missing the copy after that. On a traditional system,
using OpenGL that way, there will be a copy after the memcpy, to
transfer everything over PCI-E to the GPU. That's one of the copies that
Libresoc eliminates.
Intel and AMD APUs may also eliminate that, sharing the main RAM, but
integrating the GPU in the same package does not save them from
marshalling the commands. Etc.
- Lauri
_______________________________________________
libre-riscv-dev mailing list
libre-riscv-dev@lists.libre-riscv.org
http://lists.libre-riscv.org/mailman/listinfo/libre-riscv-dev
^ permalink raw reply
* [libre-riscv-dev] [Bug 369] missing XER SO/OV/32 check in test_pipe_caller.py
From: bugzilla-daemon @ 2020-06-08 14:10 UTC (permalink / raw)
To: libre-riscv-dev
In-Reply-To: <bug-369-13@https.bugs.libre-soc.org/>
https://bugs.libre-soc.org/show_bug.cgi?id=369
--- Comment #2 from Luke Kenneth Casson Leighton <lkcl@lkcl.net> ---
- comb += ov_o.data[0].eq((add_o[-2] != a[-1]) & (a[-1] ==
b[-1]))
- comb += ov_o.data[1].eq((add_o[32] != a[31]) & (a[31] ==
b[31]))
+ # 32-bit (ov[1]) and 64-bit (ov[0]) overflow
+ ov = Signal(2, reset_less=True)
+ comb += ov[0].eq(calc_ov(a[-1], b[-1], ca[0], add_o[-2]))
+ comb += ov[1].eq(calc_ov(a[31], b[31], ca[1], add_o[32]))
+ comb += ov_o.data.eq(ov)
michael i believe the former code may have been incorrect, one of the tests
should have been using the carry (32/64) flags rather than the MSB of a?
also the OV32 handling in the simulator is missing, i will try to add
something.
--
You are receiving this mail because:
You are on the CC list for the bug.
_______________________________________________
libre-riscv-dev mailing list
libre-riscv-dev@lists.libre-riscv.org
http://lists.libre-riscv.org/mailman/listinfo/libre-riscv-dev
^ permalink raw reply
* Re: [libre-riscv-dev] Request for input and technical expertise for Systèmes Libres Amazon Alexa IOT Pitch 10-JUN-2020
From: Hendrik Boom @ 2020-06-08 13:55 UTC (permalink / raw)
To: libre-riscv-dev
In-Reply-To: <CAPweEDys8JLW8RvQSz837q_=W-JqGihjL_vpk9f5WJofa6NmhQ@mail.gmail.com>
On Mon, Jun 08, 2020 at 12:13:48PM +0100, Luke Kenneth Casson Leighton wrote:
>
> all of these things are at the architectural level. we are not doing
> anything fancy at the gate level. it is a matter of making
> *architectural* decisions that reduce power consumption, operating
> within the exact same gate-level power consumption constraints as
> every other ASIC out there.
You point out that the CPU and GPU share cache, being the same processor.
But we are designing a four-core chip? To what extent to the four
cores share cache?
And on avoiding data copying between CPU ad GPU:
I believe the OpenGL API involves copying data from CPU buffers to GPU
buffers, with the understanding that the CPU copies can be discarded
while the GPU goes on with its copy.
Having the same storage for both sets of buffers could obviously obviate
these copies, except that software that uses this API will likely rely
on being able to overwrite the CPU-side buffers with impunity. So the
copy will still have to be done.
Do I misunderstand OpenGL? Is Vulcan different? Will users want to
bypass these libraries and use the graphics instructios directly? Or
is there some other sublety I'm missing?
-- hendrik
_______________________________________________
libre-riscv-dev mailing list
libre-riscv-dev@lists.libre-riscv.org
http://lists.libre-riscv.org/mailman/listinfo/libre-riscv-dev
^ permalink raw reply
* [libre-riscv-dev] [Bug 369] missing XER SO/OV/32 check in test_pipe_caller.py
From: bugzilla-daemon @ 2020-06-08 13:50 UTC (permalink / raw)
To: libre-riscv-dev
In-Reply-To: <bug-369-13@https.bugs.libre-soc.org/>
https://bugs.libre-soc.org/show_bug.cgi?id=369
--- Comment #1 from Luke Kenneth Casson Leighton <lkcl@lkcl.net> ---
commit c822c90668e0fcac8d1717650408bce2c2dda353 (HEAD -> master)
Author: Luke Kenneth Casson Leighton <lkcl@lkcl.net>
Date: Mon Jun 8 14:50:35 2020 +0100
added check which shows that OV32 in "adde." is not correct
--
You are receiving this mail because:
You are on the CC list for the bug.
_______________________________________________
libre-riscv-dev mailing list
libre-riscv-dev@lists.libre-riscv.org
http://lists.libre-riscv.org/mailman/listinfo/libre-riscv-dev
^ permalink raw reply
* [libre-riscv-dev] [Bug 325] create POWER9 TRAP pipeline
From: bugzilla-daemon @ 2020-06-08 13:42 UTC (permalink / raw)
To: libre-riscv-dev
In-Reply-To: <bug-325-13@https.bugs.libre-soc.org/>
https://bugs.libre-soc.org/show_bug.cgi?id=325
--- Comment #87 from Luke Kenneth Casson Leighton <lkcl@lkcl.net> ---
(In reply to Luke Kenneth Casson Leighton from comment #86)
> hi michael,
cole! gaah :) names... massive headache (again) at the moment.
--
You are receiving this mail because:
You are on the CC list for the bug.
_______________________________________________
libre-riscv-dev mailing list
libre-riscv-dev@lists.libre-riscv.org
http://lists.libre-riscv.org/mailman/listinfo/libre-riscv-dev
^ permalink raw reply
* [libre-riscv-dev] [Bug 325] create POWER9 TRAP pipeline
From: bugzilla-daemon @ 2020-06-08 12:59 UTC (permalink / raw)
To: libre-riscv-dev
In-Reply-To: <bug-325-13@https.bugs.libre-soc.org/>
https://bugs.libre-soc.org/show_bug.cgi?id=325
--- Comment #86 from Luke Kenneth Casson Leighton <lkcl@lkcl.net> ---
hi michael, apologies i couldn't help it, i did some of the TODO
comments on trap main_stage:
- # TODO: MSR (into srr1)
+ # take a copy of the current MSR in SRR1
+ comb += msr_copy(srr1_o.data, msr_i) # old MSR
+ comb += srr1_o.ok.eq(1)
also i wandered through the 3.0B PDF and found the section that
covers "Program Interrupts" - Book III 7.5.9
i therefore added these defines:
PI_TRAP = (63 - 46) # 1 if exception is "trap" type
and changed the setting of the relevant SRR1 bit to use that definition:
self.trap(0x700, cia_i)
comb += srr1_o.data[PI_TRAP].eq(1)
now we have words instead of numbers when it comes to reading through the
code. we can see, in two lines of code:
1. "it's a trap" (self.trap - duh).
2. execution continues at address 0x700
3. the trap handler (in software) can read SRR1 and know exactly what to do
--
You are receiving this mail because:
You are on the CC list for the bug.
_______________________________________________
libre-riscv-dev mailing list
libre-riscv-dev@lists.libre-riscv.org
http://lists.libre-riscv.org/mailman/listinfo/libre-riscv-dev
^ permalink raw reply
* Re: [libre-riscv-dev] Request for input and technical expertise for Systèmes Libres Amazon Alexa IOT Pitch 10-JUN-2020
From: Staf Verhaegen @ 2020-06-08 12:37 UTC (permalink / raw)
To: libre-riscv-dev
In-Reply-To: <CAPweEDys8JLW8RvQSz837q_=W-JqGihjL_vpk9f5WJofa6NmhQ@mail.gmail.com>
Luke Kenneth Casson Leighton schreef op ma 08-06-2020 om 12:13 [+0100]:
>
> how exactly this would be communicated - in summary form - to
> Executives, i have no idea.
>
I think it should be doable in a few slides showing the two alternative
block diagrams and how data moves through it.
greets,
Staf.
_______________________________________________
libre-riscv-dev mailing list
libre-riscv-dev@lists.libre-riscv.org
http://lists.libre-riscv.org/mailman/listinfo/libre-riscv-dev
^ permalink raw reply
* Re: [libre-riscv-dev] Request for input and technical expertise for Systèmes Libres Amazon Alexa IOT Pitch 10-JUN-2020
From: Luke Kenneth Casson Leighton @ 2020-06-08 11:13 UTC (permalink / raw)
To: Libre-RISCV General Development
In-Reply-To: <a408cbcf44e43d6aefb2e772d671d671a7e2e3f6.camel@fibraservi.eu>
On Mon, Jun 8, 2020 at 11:17 AM Staf Verhaegen <staf@fibraservi.eu> wrote:
> Given that you currently are doing power instruction set I expect you
> will have more power consumption in decoding the power instruction set,
> so even if the GPU part would cause less overhead would you still be
> better than ARM + MALI or RISC-V + some GPU ?
the architecture of a CPU + GPU arrangement is as follows:
* CPU L1 I-Cache, D-Cache
* GPU L1 I-Cache, D-Cache
* inter-connect on-die Shared Memory Bus between CPU and GPU.
therefore, to perform even the simplest GPU-based instruction - even
just the one instruction, the following must be actioned:
* CPU must run a compiler that turns (Vulkan?) command into GPU instructions
* CPU must "package" GPU instructions into a serial buffer.
* CPU must send the serial buffer through the CPU L1 D-Cache, onto the
Shared Memory Bus, into the GPU L1 I-Cache
* GPU must execute the instructions.
* GPU must respond indicating completion of the commands, back through
the L1 Caches, over the Shared Memory Bus.
in a hybrid architecture, steps 2 thru 5 are replaced with *DIRECT*
execution of the GPU instruction.
it should be clear that a complex detailed gate-level power analysis
of this scenario is not needed, because even at a high level it should
be obvious that there is significant power consumption just by having
this type of dual-processor architecture compared to a hybrid one.
where that does *NOT* work is if design mistakes are made in the
hybrid CPU-GPU architecture such as having insufficient registers.
Jeff Bush's Nyuzi work showed very clearly that the highest power
consumption of a GPU is getting the data through the L1 cache barrier.
GPU workloads are almost exclusively:
* LOAD
* compute
* STORE
with negligeable or zero cross-references between each cycle. compare
this with general-purpose CPU workloads:
* LOAD
* partial compute
* STORE
* LOAD partial results (from multiple *different* previously
partially-computed prior sequences)
* compute
* STORE
i vaguely recall somewhere someone saying that the cross-referencing
of prior data, in general-purpose CPU workloads, is somewhere around
30% back-referencing of prior data. in GPU workloads, due to the
parallelism, this is **ZERO** for the majority of data (hundreds of
megabytes per second)
therefore if *any* of the GPU data has to go back through the L1 Cache
(because there are not enough registers), it can result in a whopping
50% increase in power consumption when compared to a design that does
have enough registers to keep the computation from spilling over.
again: this does not need gate-level analysis, it's just a design
architectural fact.
another key critical aspect of what we are doing is down to the fact
that the CPU and GPU - being now one and the same - effectively have
the exact same I-Cache.
normally in a split CPU-GPU architecture, CPU execution would be
unaffected by GPU workload execution, because it is totally
*different* I-Caches. because they are different, the designers of
the GPU ISA completely ignore instruction compactness, often using
VLIW (128-bit instructions) or 64-bit instructions, on the basis that
each of these instructions is executing 4 parallel tracks (4x FP
operations), and that therefore the high bang-per-buck ratio easily
justifies the large instruction rate.
with a shared I-Cache, when executing GPU instructions these require
context-switching the CPU instructions out. such massively-large
instructions become a serious problem for a hybrid architecture.
therefore, to that end, we effectively designed Simple-V to be a
"hardware compression algorithm" - to compress a *Vectorised* ISA at
the *hardware* level.
instead of 64 or 128 bit Vector instructions, we have 32-bit
Vector-capable instructions and in some cases even 16-bit
Vector-capable instructions.
this *drastically* reduces I-cache utilisation, which is a major
factor in power consumption due to context-switching.
all of these things are at the architectural level. we are not doing
anything fancy at the gate level. it is a matter of making
*architectural* decisions that reduce power consumption, operating
within the exact same gate-level power consumption constraints as
every other ASIC out there.
how exactly this would be communicated - in summary form - to
Executives, i have no idea.
> I don't understand the BOM advantage when CPU and GPU are integrated in one chip, e.g. ARM Cortex + MALI GPU.
i also commented that this is incorrect.
l.
_______________________________________________
libre-riscv-dev mailing list
libre-riscv-dev@lists.libre-riscv.org
http://lists.libre-riscv.org/mailman/listinfo/libre-riscv-dev
^ permalink raw reply
* Re: [libre-riscv-dev] Request for input and technical expertise for Systèmes Libres Amazon Alexa IOT Pitch 10-JUN-2020
From: Staf Verhaegen @ 2020-06-08 10:17 UTC (permalink / raw)
To: Libre-RISCV General Development
In-Reply-To: <CAC2bXD4PmsSRZrxMAJv88+UrfYV1YzABuwRAi4+hu=0wJnK-dQ@mail.gmail.com>
Jacob Lifshay schreef op zo 07-06-2020 om 17:29 [-0700]:
> On Sun, Jun 7, 2020, 16:56 Luke Kenneth Casson Leighton <lkcl@lkcl.net>wrote:
> > On Mon, Jun 8, 2020 at 12:42 AM Cole Poirier <colepoirier@gmail.com>wrote:
> > > Hi Libre-SOC Developers,
> > > I have been working on a first draft of the script/contents of apresentation pitching Libre-SOC to Amazon Alexa's IOT division. I haveput what I've been able to come up with on my own on a wiki page here:
> >
> > https://libre-soc.org/Systemes_Libres_Amazon_Alexa_IOT_Pitch_10-JUN-2020/.
> > (i moved it to here, sorry!
> > https://libre-soc.org/systemes_libre/Systemes_Libres_Amazon_Alexa_IOT_Pitch_10-JUN-2020/
> > )
> > > If you can, please review what I have written and edit, rearrange, andadd relevant points as you see fit. I would appriciate help specificallyin filling in or correcting technical details or false claims that Ihave made.
>
> I added some notes.
OK, some comment from a guy coming out of the hardware development world. I may sound skeptical but this is to help you and prepare for a possible skeptical audience.
I get the advantage of RTOS and RPC needed between CPU and GPU; we will have to see how important your audience will find this.
Playing devil's advocate I would even wonder if IoT devices typically need a display (and thus GPU) functionality anyway ? Or if they do, can't they just live with a simple framebuffer; possibly with some 2D acceleration ?
Will the libre-SOC GPU instructions have advantage for machine learning/artificial intelligence ?
Big part of your pitch is based on power advantages, as also indicated by Jacob I would be very prudent in doing power claims without having real numbers to back it up.
Modern design techniques and recent (proprietary) EDA tools are good at having virtually no power consumption when a circuit is not doing anything. So I don't see a reason why your combined CPU-GPU would be so much better in power consumption then a chip with a CPU and GPU. Given that you currently are doing power instruction set I expect you will have more power consumption in decoding the power instruction set, so even if the GPU part would cause less overhead would you still be better than ARM + MALI or RISC-V + some GPU ?
As always I like to be proven wrong by real numbers.
Also to get there several man-years of development of the open source EDA tool chain will be needed: proper parasitic extraction of interconnects, timing driven P&R, support for DVFS (dynamic frequency voltage scaling), clock gating, power gating, ...
If there will be ASIC hardware developers in your audience they will be very skeptical when you present power consumption numbers based on simulation and will likely want to see the details how you derived the numbers; they will just dismiss power claims made on principle. Reason is that in smaller nodes power consumption is mainly driven by the capacitive load of the interconnects so doing power claims without an actual implementation in the targeted node is considered just guessing.
I don't understand the BOM advantage when CPU and GPU are integrated in one chip, e.g. ARM Cortex + MALI GPU.
greets,
Staf.
_______________________________________________
libre-riscv-dev mailing list
libre-riscv-dev@lists.libre-riscv.org
http://lists.libre-riscv.org/mailman/listinfo/libre-riscv-dev
^ permalink raw reply
* Re: [libre-riscv-dev] Using formal to expose bugs in scoreboard
From: Luke Kenneth Casson Leighton @ 2020-06-08 9:40 UTC (permalink / raw)
To: Libre-RISCV General Development
In-Reply-To: <E2E475D2-7421-4970-92BE-146FC5514AA5@gatech.edu>
On Mon, Jun 8, 2020 at 4:11 AM Yehowshua <yimmanuel3@gatech.edu> wrote:
>
> This error highlights the need for formal on the scoreboard.
this exact need was taken into consideration when the application for
NLNet Funding for the Formal Correctnes Proofs was made, a year ago.
https://bugs.libre-soc.org/show_bug.cgi?id=81
as it is critically important a budget of EUR 5,000 was allocated just
to this *one* task:
https://bugs.libre-soc.org/show_bug.cgi?id=197
michael began a Formal Correctness Proof for the MultiCompUnit on the
24th of May:
https://bugs.libre-soc.org/show_bug.cgi?id=342
that Correctness Proof is designed to complement the bugreport that
you referenced (see further below)
https://bugs.libre-soc.org/show_bug.cgi?id=336#c54
Samuel Falvo's work was what originally inspired us to use a hybrid
combination of "simple, clear, well-documented" unit tests plus
extensive Formal Correctness Proofs:
http://lists.libre-riscv.org/pipermail/libre-riscv-dev/2019-August/002311.html
although i cannot find the original discussion, it is elsewhere on the
internet several months prior to the Aug 2019 discussion, Samuel's
insights showed us that writing comprehensive Unit Tests simply does
not catch everything, and even if efforts are made to try to do so the
sheer quantity of code involved masks errors, distracts from the
actual code being tested, and doesn't catch all errors anyway.
the IEEE754 FPU tests are a good example. this is the basic
infrastructure that allows hundreds of thousands of random and
regression tests to be called:
https://git.libre-soc.org/?p=ieee754fpu.git;a=blob;f=src/ieee754/fpcommon/test/case_gen.py;hb=HEAD
https://git.libre-soc.org/?p=ieee754fpu.git;a=blob;f=src/ieee754/fpcommon/test/fpmux.py;hb=HEAD
and each pipeline then contains sometimes up to 10 separate unit tests
to cover different aspects:
https://git.libre-soc.org/?p=ieee754fpu.git;a=tree;f=src/ieee754/fpdiv/test;hb=HEAD
https://git.libre-soc.org/?p=ieee754fpu.git;a=blob;f=src/ieee754/fpmax/test/test_fpmax_pipe.py;hb=HEAD
https://git.libre-soc.org/?p=ieee754fpu.git;a=tree;f=src/ieee754/fpadd/test;hb=HEAD
https://git.libre-soc.org/?p=ieee754fpu.git;a=tree;f=src/ieee754/fpcmp/test;hb=HEAD
none of which *actually* gives us full confidence in the code, despite
taking a week to run even on the fastest modern machines available.
many of the errors during the 6+ month development of the IEEE754 FPU
code were only caught by running well over 100,000 random tests on the
code when parameterised down to FP16.
with only 65536 possible values in FP16, this gave a high probability
that corner-cases would be caught, and in some cases (single parameter
rather than two 2^16 parameters) allows for full coverage.
therefore, yes: the idea has always been to follow this step-by-step
process, for *all* code:
1) write the code and a basic "understanding" unit test infrastructure
sufficient to get the code functional
2) write Formal Correctness Proofs that catch errors
3) write *SIMPLE* unit tests that in no way cover the entirety of the
code: their purpose is as a *tutorial* and example of how to use the
code.
good examples of (3) when it comes to the pipeline API are here:
https://git.libre-soc.org/?p=nmutil.git;a=blob;f=src/nmutil/test/test_buf_pipe.py;hb=HEAD
https://git.libre-soc.org/?p=nmutil.git;a=blob;f=src/nmutil/test/test_inout_mux_pipe.py;hb=HEAD
l.
_______________________________________________
libre-riscv-dev mailing list
libre-riscv-dev@lists.libre-riscv.org
http://lists.libre-riscv.org/mailman/listinfo/libre-riscv-dev
^ permalink raw reply
* [libre-riscv-dev] [Bug 81] implement 6600-style "precise" out-of-order scoreboard
From: bugzilla-daemon @ 2020-06-08 9:13 UTC (permalink / raw)
To: libre-riscv-dev
In-Reply-To: <bug-81-13@https.bugs.libre-soc.org/>
https://bugs.libre-soc.org/show_bug.cgi?id=81
Luke Kenneth Casson Leighton <lkcl@lkcl.net> changed:
What |Removed |Added
----------------------------------------------------------------------------
See Also| |https://bugs.libre-soc.org/
| |show_bug.cgi?id=197
--
You are receiving this mail because:
You are on the CC list for the bug.
_______________________________________________
libre-riscv-dev mailing list
libre-riscv-dev@lists.libre-riscv.org
http://lists.libre-riscv.org/mailman/listinfo/libre-riscv-dev
^ permalink raw reply
* [libre-riscv-dev] [Bug 197] Formal correctness proof needed of the 6600-style Out-of-Order execution engine
From: bugzilla-daemon @ 2020-06-08 9:13 UTC (permalink / raw)
To: libre-riscv-dev
In-Reply-To: <bug-197-13@https.bugs.libre-soc.org/>
https://bugs.libre-soc.org/show_bug.cgi?id=197
Luke Kenneth Casson Leighton <lkcl@lkcl.net> changed:
What |Removed |Added
----------------------------------------------------------------------------
See Also| |https://bugs.libre-soc.org/
| |show_bug.cgi?id=81
--
You are receiving this mail because:
You are on the CC list for the bug.
_______________________________________________
libre-riscv-dev mailing list
libre-riscv-dev@lists.libre-riscv.org
http://lists.libre-riscv.org/mailman/listinfo/libre-riscv-dev
^ permalink raw reply
* Re: [libre-riscv-dev] Using formal to expose bugs in scoreboard
From: Yehowshua @ 2020-06-08 3:18 UTC (permalink / raw)
To: Libre-RISCV General Development
In-Reply-To: <E2E475D2-7421-4970-92BE-146FC5514AA5@gatech.edu>
> On Jun 7, 2020, at 11:11 PM, Yehowshua <yimmanuel3@gatech.edu> wrote:
>
> This error highlights the need for formal on the scoreboard.
> https://bugs.libre-soc.org/show_bug.cgi?id=336#c54
I believe I read the bug report too quickly and that link from SymbioticEDA doesn’t apply directly to the linked bug.
Yehowshua
_______________________________________________
libre-riscv-dev mailing list
libre-riscv-dev@lists.libre-riscv.org
http://lists.libre-riscv.org/mailman/listinfo/libre-riscv-dev
^ permalink raw reply
* [libre-riscv-dev] Using formal to expose bugs in scoreboard
From: Yehowshua @ 2020-06-08 3:11 UTC (permalink / raw)
To: Libre-RISCV General Development
This error highlights the need for formal on the scoreboard.
https://bugs.libre-soc.org/show_bug.cgi?id=336#c54
Symbiotic EDA has a wonderful tutorial on how to specify
hazard dependencies as formal constraints.
https://www.youtube.com/watch?v=tgTt-8UqSMU
I imagine it shouldn’t be too hard to turn that tutorial into
something that can fit the scoreboard.
You would keep repairing until the asserts pass basically.
Cheers,
Yehowshua
_______________________________________________
libre-riscv-dev mailing list
libre-riscv-dev@lists.libre-riscv.org
http://lists.libre-riscv.org/mailman/listinfo/libre-riscv-dev
^ permalink raw reply
* [libre-riscv-dev] [Bug 336] ALU CompUnit needs to recognise that RA (src1) can be zero
From: bugzilla-daemon @ 2020-06-08 1:28 UTC (permalink / raw)
To: libre-riscv-dev
In-Reply-To: <bug-336-13@https.bugs.libre-soc.org/>
https://bugs.libre-soc.org/show_bug.cgi?id=336
--- Comment #56 from Luke Kenneth Casson Leighton <lkcl@lkcl.net> ---
(In reply to Cesar Strauss from comment #55)
> (In reply to Luke Kenneth Casson Leighton from comment #54)
> > oh wait... that's a *combinatorial* ALU? iinteresting. i wonder
> > what that could be used for.
>
> Well, the ALU output is already latched in MultiCompUnit. Maybe having
> another output register, inside the ALU itself, is redundant?
ah right yes i understand, now.
yes, absolutely. the only reason for the latch, inside the test ALU, was to
simulate the concept of a pipeline stage.
however you are absolutely right: with both the incoming src regs being
latched, and the outgoing result being latched also by MultiCompUnit, it is not
necessary ro latch inside alu_hier.ALU as well.
>
> > hmmm i tried putting OP_NOP in there, and both compunit1 and parallel
> > locked up. this tends to indicate a combinatorial loop. any clues?
>
> I see. I'm looking into it.
it may be because "now" is set at the default action. i have yet to work out
how to detect combinatorial loops. there may be something that can be done
using yosys synth, michael showed me something a couple weeks ago. ltp
command?
--
You are receiving this mail because:
You are on the CC list for the bug.
_______________________________________________
libre-riscv-dev mailing list
libre-riscv-dev@lists.libre-riscv.org
http://lists.libre-riscv.org/mailman/listinfo/libre-riscv-dev
^ permalink raw reply
* [libre-riscv-dev] [Bug 336] ALU CompUnit needs to recognise that RA (src1) can be zero
From: bugzilla-daemon @ 2020-06-08 1:20 UTC (permalink / raw)
To: libre-riscv-dev
In-Reply-To: <bug-336-13@https.bugs.libre-soc.org/>
https://bugs.libre-soc.org/show_bug.cgi?id=336
--- Comment #55 from Cesar Strauss <cestrauss@gmail.com> ---
(In reply to Luke Kenneth Casson Leighton from comment #54)
> oh wait... that's a *combinatorial* ALU? iinteresting. i wonder
> what that could be used for.
Well, the ALU output is already latched in MultiCompUnit. Maybe having another
output register, inside the ALU itself, is redundant?
> hmmm i tried putting OP_NOP in there, and both compunit1 and parallel
> locked up. this tends to indicate a combinatorial loop. any clues?
I see. I'm looking into it.
--
You are receiving this mail because:
You are on the CC list for the bug.
_______________________________________________
libre-riscv-dev mailing list
libre-riscv-dev@lists.libre-riscv.org
http://lists.libre-riscv.org/mailman/listinfo/libre-riscv-dev
^ permalink raw reply
* [libre-riscv-dev] [Bug 369] New: missing XER SO/OV/32 check in test_pipe_caller.py
From: bugzilla-daemon @ 2020-06-08 1:09 UTC (permalink / raw)
To: libre-riscv-dev
https://bugs.libre-soc.org/show_bug.cgi?id=369
Bug ID: 369
Summary: missing XER SO/OV/32 check in test_pipe_caller.py
Product: Libre-SOC's first SoC
Version: unspecified
Hardware: Other
OS: Linux
Status: CONFIRMED
Severity: enhancement
Priority: ---
Component: Source Code
Assignee: lkcl@lkcl.net
Reporter: lkcl@lkcl.net
CC: libre-riscv-dev@lists.libre-riscv.org
NLnet milestone: ---
https://git.libre-soc.org/?p=soc.git;a=blob;f=src/soc/fu/alu/test/test_pipe_caller.py;hb=HEAD
at least ALU and ShiftRot are misding so/ov/32 checking.
test_core.py shows some changes to the XER regfiles that appear not to be
correct, on execution of "adde."
--
You are receiving this mail because:
You are on the CC list for the bug.
_______________________________________________
libre-riscv-dev mailing list
libre-riscv-dev@lists.libre-riscv.org
http://lists.libre-riscv.org/mailman/listinfo/libre-riscv-dev
^ permalink raw reply
* Re: [libre-riscv-dev] "simple" core
From: Luke Kenneth Casson Leighton @ 2020-06-08 0:35 UTC (permalink / raw)
To: Libre-RISCV General Development
In-Reply-To: <86EDE30A-FA1C-4248-8434-10FC28E7520E@gatech.edu>
On Monday, June 8, 2020, Yehowshua <yimmanuel3@gatech.edu> wrote:
>
> > core.py really is *very* simple. its execution model is as follows:
> This is great to hear.
> Keep up the great work.
very slow without michael's input.
really need help and speeding up.
l.
--
---
crowd-funded eco-conscious hardware: https://www.crowdsupply.com/eoma68
_______________________________________________
libre-riscv-dev mailing list
libre-riscv-dev@lists.libre-riscv.org
http://lists.libre-riscv.org/mailman/listinfo/libre-riscv-dev
^ permalink raw reply
* Re: [libre-riscv-dev] Request for input and technical expertise for Systèmes Libres Amazon Alexa IOT Pitch 10-JUN-2020
From: Jacob Lifshay @ 2020-06-08 0:29 UTC (permalink / raw)
To: Libre-RISCV General Development
In-Reply-To: <CAPweEDwDMFuONT=LtAkE7cJx-=ZPkmKAhb-7fYOr+LoQDDkGkQ@mail.gmail.com>
On Sun, Jun 7, 2020, 16:56 Luke Kenneth Casson Leighton <lkcl@lkcl.net>
wrote:
> On Mon, Jun 8, 2020 at 12:42 AM Cole Poirier <colepoirier@gmail.com>
> wrote:
> >
> > Hi Libre-SOC Developers,
> >
> > I have been working on a first draft of the script/contents of a
> > presentation pitching Libre-SOC to Amazon Alexa's IOT division. I have
> > put what I've been able to come up with on my own on a wiki page here:
> https://libre-soc.org/Systemes_Libres_Amazon_Alexa_IOT_Pitch_10-JUN-2020/.
>
> (i moved it to here, sorry!
>
> https://libre-soc.org/systemes_libre/Systemes_Libres_Amazon_Alexa_IOT_Pitch_10-JUN-2020/
> )
>
> > If you can, please review what I have written and edit, rearrange, and
> > add relevant points as you see fit. I would appriciate help specifically
> > in filling in or correcting technical details or false claims that I
> > have made.
>
I added some notes.
Jacob
>
_______________________________________________
libre-riscv-dev mailing list
libre-riscv-dev@lists.libre-riscv.org
http://lists.libre-riscv.org/mailman/listinfo/libre-riscv-dev
^ permalink raw reply
page: next prev latest
Libre RISC-V Mailing list Archives (accessible via public-inbox)
Archives are clonable: git clone --mirror https://inbox.libre-soc.org/libre-riscv-dev
Example config snippet for mirrors
AGPL code for this site: git clone https://public-inbox.org/public-inbox.git