gem5.git
3 years agotests: Add Python unit tests for m5.util.convert
Andreas Sandberg [Tue, 19 Jan 2021 16:04:33 +0000 (16:04 +0000)]
tests: Add Python unit tests for m5.util.convert

Python unit tests need to be run from within gem5. This change adds a
script to run unit tests (tests/run_pyunit.py) and a unit test for
m5.util.convert.

The tests can be run as follows:

  ./build/NULL/gem5.opt tests/run_pyunit.py

Change-Id: I80d1aabbe1d87b01b48280972f9418317e648779
Signed-off-by: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/39377
Maintainer: Jason Lowe-Power <power.jg@gmail.com>
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Daniel Carvalho <odanrc@yahoo.com.br>
3 years agosim: Consistently use ISO prefixes
Andreas Sandberg [Thu, 21 Jan 2021 12:54:18 +0000 (12:54 +0000)]
sim: Consistently use ISO prefixes

We currently use the traditional SI-like prefixes to represent
binary multipliers in some contexts. This is ambiguous in many cases
since they overload the meaning of the SI prefix.

Here are some examples of commonly used in the industry:
  * Storage vendors define 1 MB as 10**6 bytes
  * Memory vendors define 1 MB as 2**20 bytes
  * Network equipment treats 1Mbit/s as 10**6 bits/s
  * Memory vendors define 1Mbit as 2**20 bits

In practice, this means that a FLASH chip on a storage bus uses
decimal prefixes, but that same flash chip on a memory bus uses binary
prefixes. It would also be reasonable to assume that the contents of a
1Mbit FLASH chip would take 0.1s to transfer over a 10Mbit Ethernet
link. That's however not the case due to different meanings of the
prefix.

The quantity 2MX is treated differently by gem5 depending on the unit
X:

  * Physical quantities (s, Hz, V, A, J, K, C, F) use decimal prefixes.
  * Interconnect and NoC bandwidths (B/s) use binary prefixes.
  * Network bandwidths (bps) use decimal prefixes.
  * Memory sizes and storage sizes (B) use binary prefixes.

Mitigate this ambiguity by consistently using the ISO/IEC/SI prefixes
for binary multipliers for parameters and comments where appropriate.

Change-Id: I797163c8690ae0092e00e371d75f5e7cebbcd1f5
Signed-off-by: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/39579
Reviewed-by: Jason Lowe-Power <power.jg@gmail.com>
Tested-by: kokoro <noreply+kokoro@google.com>
3 years agopython: Consistently use ISO prefixes
Andreas Sandberg [Thu, 21 Jan 2021 12:44:14 +0000 (12:44 +0000)]
python: Consistently use ISO prefixes

We currently use the traditional SI-like prefixes to represent
binary multipliers in some contexts. This is ambiguous in many cases
since they overload the meaning of the SI prefix.

Here are some examples of commonly used in the industry:
  * Storage vendors define 1 MB as 10**6 bytes
  * Memory vendors define 1 MB as 2**20 bytes
  * Network equipment treats 1Mbit/s as 10**6 bits/s
  * Memory vendors define 1Mbit as 2**20 bits

In practice, this means that a FLASH chip on a storage bus uses
decimal prefixes, but that same flash chip on a memory bus uses binary
prefixes. It would also be reasonable to assume that the contents of a
1Mbit FLASH chip would take 0.1s to transfer over a 10Mbit Ethernet
link. That's however not the case due to different meanings of the
prefix.

The quantity 2MX is treated differently by gem5 depending on the unit
X:

  * Physical quantities (s, Hz, V, A, J, K, C, F) use decimal prefixes.
  * Interconnect and NoC bandwidths (B/s) use binary prefixes.
  * Network bandwidths (bps) use decimal prefixes.
  * Memory sizes and storage sizes (B) use binary prefixes.

Mitigate this ambiguity by consistently using the ISO/IEC/SI prefixes
for binary multipliers for parameters and comments where appropriate.

Change-Id: I3d0bbfa00968486af8d57c36be2c8bee034bae93
Signed-off-by: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/39577
Reviewed-by: Daniel Carvalho <odanrc@yahoo.com.br>
Reviewed-by: Jason Lowe-Power <power.jg@gmail.com>
Maintainer: Jason Lowe-Power <power.jg@gmail.com>
Tested-by: kokoro <noreply+kokoro@google.com>
3 years agoarch-arm, dev-arm: Consistently use ISO prefixes
Andreas Sandberg [Thu, 21 Jan 2021 09:47:16 +0000 (09:47 +0000)]
arch-arm, dev-arm: Consistently use ISO prefixes

We currently use the traditional SI-like prefixes to represent
binary multipliers in some contexts. This is ambiguous in many cases
since they overload the meaning of the SI prefix.

Here are some examples of commonly used in the industry:
  * Storage vendors define 1 MB as 10**6 bytes
  * Memory vendors define 1 MB as 2**20 bytes
  * Network equipment treats 1Mbit/s as 10**6 bits/s
  * Memory vendors define 1Mbit as 2**20 bits

In practice, this means that a FLASH chip on a storage bus uses
decimal prefixes, but that same flash chip on a memory bus uses binary
prefixes. It would also be reasonable to assume that the contents of a
1Mbit FLASH chip would take 0.1s to transfer over a 10Mbit Ethernet
link. That's however not the case due to different meanings of the
prefix.

The quantity 2MX is treated differently by gem5 depending on the unit
X:

  * Physical quantities (s, Hz, V, A, J, K, C, F) use decimal prefixes.
  * Interconnect and NoC bandwidths (B/s) use binary prefixes.
  * Network bandwidths (bps) use decimal prefixes.
  * Memory sizes and storage sizes (B) use binary prefixes.

Mitigate this ambiguity by consistently using the ISO/IEC/SI prefixes
for binary multipliers for parameters and comments where appropriate.

Change-Id: I9b47194d26d71c8ebedda6c31a5bac54b600d3bf
Signed-off-by: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/39575
Reviewed-by: Richard Cooper <richard.cooper@arm.com>
Tested-by: kokoro <noreply+kokoro@google.com>
3 years agotests: Stop "using namespace std" in unittest/.
Gabe Black [Thu, 21 Jan 2021 12:50:11 +0000 (04:50 -0800)]
tests: Stop "using namespace std" in unittest/.

These are the historical "unit test"s, which aren't really unit tests,
they're actually complete builds of gem5 with main functions which run a
fairly specific test instead of a simulation. They test a single unit,
but they do it with all the other units in place and potentially
participating in the test.

Change-Id: Ib0ea68f26091a79992396d932627e4ce180f7825
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/39565
Reviewed-by: Giacomo Travaglini <giacomo.travaglini@arm.com>
Maintainer: Giacomo Travaglini <giacomo.travaglini@arm.com>
Tested-by: kokoro <noreply+kokoro@google.com>
3 years agomisc: Stop "using namespace std" in protoio.cc.
Gabe Black [Thu, 21 Jan 2021 12:42:05 +0000 (04:42 -0800)]
misc: Stop "using namespace std" in protoio.cc.

Change-Id: I4f27979910230860c631b63bb500f87b45c24e33
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/39563
Reviewed-by: Daniel Carvalho <odanrc@yahoo.com.br>
Maintainer: Gabe Black <gabe.black@gmail.com>
Tested-by: kokoro <noreply+kokoro@google.com>
3 years agocpu: Stop "using namespace std"
Gabe Black [Thu, 21 Jan 2021 12:40:10 +0000 (04:40 -0800)]
cpu: Stop "using namespace std"

Change-Id: I1b648914d353672076d903ed581aa61cdd7c1d0f
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/39562
Reviewed-by: Daniel Carvalho <odanrc@yahoo.com.br>
Maintainer: Gabe Black <gabe.black@gmail.com>
Tested-by: kokoro <noreply+kokoro@google.com>
3 years agoarch-arm: Stop "using namespace std"
Gabe Black [Thu, 21 Jan 2021 11:49:07 +0000 (03:49 -0800)]
arch-arm: Stop "using namespace std"

Change-Id: If0f373bdaadce81c5ebbc37b03810335c42dd10a
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/39561
Reviewed-by: Giacomo Travaglini <giacomo.travaglini@arm.com>
Maintainer: Giacomo Travaglini <giacomo.travaglini@arm.com>
Tested-by: kokoro <noreply+kokoro@google.com>
3 years agoarch-sparc: Stop "using namespace std"
Gabe Black [Thu, 21 Jan 2021 11:43:35 +0000 (03:43 -0800)]
arch-sparc: Stop "using namespace std"

Change-Id: I4a1019b5978b08b4999edfe5f65ef7eae06481c2
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/39560
Reviewed-by: Daniel Carvalho <odanrc@yahoo.com.br>
Maintainer: Gabe Black <gabe.black@gmail.com>
Tested-by: kokoro <noreply+kokoro@google.com>
3 years agoarch-mips: Stop "using namespace std"
Gabe Black [Thu, 21 Jan 2021 11:17:40 +0000 (03:17 -0800)]
arch-mips: Stop "using namespace std"

Change-Id: I0ad5ad71d8ba2d7c050d3f368341ce98d3f87a90
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/39559
Reviewed-by: Daniel Carvalho <odanrc@yahoo.com.br>
Maintainer: Gabe Black <gabe.black@gmail.com>
Tested-by: kokoro <noreply+kokoro@google.com>
3 years agoarch-riscv: Stop "using namespace std"
Gabe Black [Thu, 21 Jan 2021 11:10:40 +0000 (03:10 -0800)]
arch-riscv: Stop "using namespace std"

Change-Id: Ia18ac69797019fb853b7c07a3961840ee6b0df39
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/39558
Reviewed-by: Daniel Carvalho <odanrc@yahoo.com.br>
Maintainer: Gabe Black <gabe.black@gmail.com>
Tested-by: kokoro <noreply+kokoro@google.com>
3 years agoarch-x86: Stop "using namespace std"
Gabe Black [Thu, 21 Jan 2021 11:04:19 +0000 (03:04 -0800)]
arch-x86: Stop "using namespace std"

Change-Id: I9763177bbd54abc97b99bd54a20a750e0adb5627
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/39557
Reviewed-by: Daniel Carvalho <odanrc@yahoo.com.br>
Maintainer: Gabe Black <gabe.black@gmail.com>
Tested-by: kokoro <noreply+kokoro@google.com>
3 years agoarch-arm: Fix Compare and Swap Pair instructions
Giacomo Travaglini [Wed, 20 Jan 2021 16:13:01 +0000 (16:13 +0000)]
arch-arm: Fix Compare and Swap Pair instructions

Those instructions were broken after:

https://gem5-review.googlesource.com/c/public/gem5/+/38381/4

Which is effectively replacing the generic StaticInst src and dest
reg array with an instruction specific one.
The size of the array is evaluated by the ISA parser, which is
counting the operands when parsing the isa code.
Alas, Compare and Swap Pair instructions were augmenting the number
of destination and source registers in the C++ world, which is
invisible to the parser. This lead to an out of bounds access
of the arrays.

This patch is fixing this behaviour by defining XResult2, which
is the second compare/result register for a paired CAS

Change-Id: Ie35c26256f42459805e007847896ac58b178fd42
Signed-off-by: Giacomo Travaglini <giacomo.travaglini@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/39456
Reviewed-by: Richard Cooper <richard.cooper@arm.com>
Reviewed-by: Gabe Black <gabe.black@gmail.com>
Tested-by: kokoro <noreply+kokoro@google.com>
3 years agomem: Stop "using namespace std"
Gabe Black [Thu, 21 Jan 2021 10:43:46 +0000 (02:43 -0800)]
mem: Stop "using namespace std"

Change-Id: I26fd73f1b7d38e1e00eece12459f7a96227900ed
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/39555
Tested-by: kokoro <noreply+kokoro@google.com>
Maintainer: Gabe Black <gabe.black@gmail.com>
Reviewed-by: Matthew Poremba <matthew.poremba@amd.com>
3 years agodev: Stop "using namespace std"
Gabe Black [Thu, 21 Jan 2021 08:00:53 +0000 (00:00 -0800)]
dev: Stop "using namespace std"

Change-Id: I317df9b566936445c3a02c28ed37146a16610454
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/39538
Tested-by: kokoro <noreply+kokoro@google.com>
Maintainer: Gabe Black <gabe.black@gmail.com>
Reviewed-by: Daniel Carvalho <odanrc@yahoo.com.br>
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
3 years agosim: Stop "using namespace std"
Gabe Black [Thu, 21 Jan 2021 07:14:24 +0000 (23:14 -0800)]
sim: Stop "using namespace std"

Change-Id: Ic641cb82a069ccb2b185d74a3b49a96b27111035
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/39537
Reviewed-by: Gabe Black <gabe.black@gmail.com>
Maintainer: Gabe Black <gabe.black@gmail.com>
Tested-by: kokoro <noreply+kokoro@google.com>
3 years agobase: Stop "using namespace std".
Gabe Black [Thu, 21 Jan 2021 06:41:42 +0000 (22:41 -0800)]
base: Stop "using namespace std".

As the std namespace expands, it becomes more and more likely that
blanketly importing all its symbols will cause a collision. Also, when
it was imported, the std:: was used or left off arbitrarily, sometimes
inconsistently even in the same function signature.

Change-Id: Ie30cbab154b00c60433908a206c229230d2b109f
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/39536
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Gabe Black <gabe.black@gmail.com>
Maintainer: Gabe Black <gabe.black@gmail.com>

3 years agobase: Style fixes in cprintf related files.
Gabe Black [Thu, 21 Jan 2021 05:38:47 +0000 (21:38 -0800)]
base: Style fixes in cprintf related files.

Fix braces, capitalization, missing line breaks, and remove mostly
useless "using namespace std"s.

Change-Id: I1210dd3b918e7cba8e576cbbdf47aa986d9278e6
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/39535
Reviewed-by: Gabe Black <gabe.black@gmail.com>
Maintainer: Gabe Black <gabe.black@gmail.com>
Tested-by: kokoro <noreply+kokoro@google.com>
3 years agotests: Fix building of unittest/stattest.cc.
Gabe Black [Thu, 21 Jan 2021 12:48:58 +0000 (04:48 -0800)]
tests: Fix building of unittest/stattest.cc.

This file wasn't including eventq.hh which it happened to be getting
transitively before.

Change-Id: I8ce5572651ddd59160d84794bf5efc90c82d83e6
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/39564
Reviewed-by: Daniel Carvalho <odanrc@yahoo.com.br>
Maintainer: Jason Lowe-Power <power.jg@gmail.com>
Tested-by: kokoro <noreply+kokoro@google.com>
3 years agoarch-x86: implement PSHUFB SSE instruction.
Tong Shen [Thu, 21 Jan 2021 00:11:01 +0000 (16:11 -0800)]
arch-x86: implement PSHUFB SSE instruction.

Change-Id: I9398f9ecb26b6aabf4015e0e285fdc2f4c2487dd
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/39495
Reviewed-by: Tong Shen <endlessroad@google.com>
Reviewed-by: Jason Lowe-Power <power.jg@gmail.com>
Maintainer: Jason Lowe-Power <power.jg@gmail.com>
Tested-by: kokoro <noreply+kokoro@google.com>
3 years agomem: Consistently use ISO prefixes
Andreas Sandberg [Thu, 21 Jan 2021 12:33:52 +0000 (12:33 +0000)]
mem: Consistently use ISO prefixes

We currently use the traditional SI-like prefixes for to represent
binary multipliers in some contexts. This is ambiguous in many cases
since they overload the meaning of the SI prefix.

Here are some examples of commonly used in the industry:
  * Storage vendors define 1 MB as 10**6 bytes
  * Memory vendors define 1 MB as 2**20 bytes
  * Network equipment treats 1Mbit/s as 10**6 bits/s
  * Memory vendors define 1Mbit as 2**20 bits

In practice, this means that a FLASH chip on a storage bus uses
decimal prefixes, but that same flash chip on a memory bus uses binary
prefixes. It would also be reasonable to assume that the contents of a
1Mbit FLASH chip would take 0.1s to transfer over a 10Mbit Ethernet
link. That's however not the case due to different meanings of the
prefix.

The quantity 2MX is treated differently by gem5 depending on the unit
X:

  * Physical quantities (s, Hz, V, A, J, K, C, F) use decimal prefixes.
  * Interconnect and NoC bandwidths (B/s) use binary prefixes.
  * Network bandwidths (bps) use decimal prefixes.
  * Memory sizes and storage sizes (B) use binary prefixes.

Mitigate this ambiguity by consistently using the ISO/IEC/SI prefixes
for binary multipliers for parameters and comments where appropriate.

Change-Id: I2d24682d207830f3b7b0ad2ff82b55e082cccb32
Signed-off-by: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/39576
Reviewed-by: Richard Cooper <richard.cooper@arm.com>
Reviewed-by: Daniel Carvalho <odanrc@yahoo.com.br>
Reviewed-by: Nikos Nikoleris <nikos.nikoleris@arm.com>
Maintainer: Nikos Nikoleris <nikos.nikoleris@arm.com>
Tested-by: kokoro <noreply+kokoro@google.com>
3 years agosim-se: Handle simultaneous page faults in SE-mode multithreading
Jiayi Huang [Thu, 21 Jan 2021 01:16:49 +0000 (17:16 -0800)]
sim-se: Handle simultaneous page faults in SE-mode multithreading

When running multithreaded programs in SE-mode with DerivO3CPU model,
there are cases that two or more cores have page faults on the same page
in nearby ticks (can be at the same tick) when fetching instructions
(more likely) or accessing data. When these cores try come to the commit
stage in nearby ticks/cycles, they will try to handle the faults
(without clobbering). Then the first core will ask for a physical page
frame to map with the virtual page. In the previous version, the right
next core that tries to handle the fault will hit a panic condition in
the EmulationPageTable::map(...) as the page has been mapped and this
page fault is not to clobber the existing mapping.

In this changeset, if it is found that the page has been mapped and it
is not to clobber the existing mapping, it will return without further
mapping activities as the page fault has been handled previously.

Jira Issue: https://gem5.atlassian.net/browse/GEM5-798

Change-Id: I9bb1163f9d1379c6fed9725101e4400fefdc8079
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/39515
Reviewed-by: Jason Lowe-Power <power.jg@gmail.com>
Maintainer: Jason Lowe-Power <power.jg@gmail.com>
Tested-by: kokoro <noreply+kokoro@google.com>
3 years agoscons: Remove Python 2.7 compatibility code
Andreas Sandberg [Thu, 21 Jan 2021 17:18:21 +0000 (17:18 +0000)]
scons: Remove Python 2.7 compatibility code

Remove the dependency on six and most 'import x from __future__'. A
few instances of imports from the future have been left in place to
ensure that Python 2.7 users still get an error message when invoking
scons.

Change-Id: I366275a6040f0084e91198b5b5c2a648bffbf2d2
Signed-off-by: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/39585
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Gabe Black <gabe.black@gmail.com>
Maintainer: Gabe Black <gabe.black@gmail.com>

3 years agopython: Remove Python 2.7 compatibility code
Andreas Sandberg [Thu, 21 Jan 2021 17:09:38 +0000 (17:09 +0000)]
python: Remove Python 2.7 compatibility code

We don't support Python 2.7 anymore. Remove glue code like the six
dependency and "from __future__" imports from gem5's standard library.

Change-Id: I71834c325f86ff0329b222be87794ead96081f05
Signed-off-by: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/39584
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Gabe Black <gabe.black@gmail.com>
Maintainer: Jason Lowe-Power <power.jg@gmail.com>

3 years agodev: Consistently use ISO prefixes
Andreas Sandberg [Thu, 21 Jan 2021 12:45:24 +0000 (12:45 +0000)]
dev: Consistently use ISO prefixes

We currently use the traditional SI-like prefixes for to represent
binary multipliers in some contexts. This is ambiguous in many cases
since they overload the meaning of the SI prefix.

Here are some examples of commonly used in the industry:
  * Storage vendors define 1 MB as 10**6 bytes
  * Memory vendors define 1 MB as 2**20 bytes
  * Network equipment treats 1Mbit/s as 10**6 bits/s
  * Memory vendors define 1Mbit as 2**20 bits

In practice, this means that a FLASH chip on a storage bus uses
decimal prefixes, but that same flash chip on a memory bus uses binary
prefixes. It would also be reasonable to assume that the contents of a
1Mbit FLASH chip would take 0.1s to transfer over a 10Mbit Ethernet
link. That's however not the case due to different meanings of the
prefix.

The quantity 2MX is treated differently by gem5 depending on the unit
X:

  * Physical quantities (s, Hz, V, A, J, K, C, F) use decimal prefixes.
  * Interconnect and NoC bandwidths (B/s) use binary prefixes.
  * Network bandwidths (bps) use decimal prefixes.
  * Memory sizes and storage sizes (B) use binary prefixes.

Mitigate this ambiguity by consistently using the ISO/IEC/SI prefixes
for binary multipliers for parameters and comments where appropriate.

Change-Id: I6ab03934af850494d95a37dcda5c2000794b4d3a
Signed-off-by: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/39578
Reviewed-by: Richard Cooper <richard.cooper@arm.com>
Reviewed-by: Daniel Carvalho <odanrc@yahoo.com.br>
Reviewed-by: Gabe Black <gabe.black@gmail.com>
Maintainer: Gabe Black <gabe.black@gmail.com>
Tested-by: kokoro <noreply+kokoro@google.com>
3 years agosim,base: make checkpointMapIn warn if an unknown key is found
Ciro Santilli [Tue, 20 Oct 2020 15:28:26 +0000 (16:28 +0100)]
sim,base: make checkpointMapIn warn if an unknown key is found

The warning happens when a key is present in the checkpoint but not in the
values that gem5 source code knows about.

To do this, we must expose iteration over IniFile section keys. To not
have to make those classes public, a visitor method is implemented.

Change-Id: I23340a953f3e604642b97690a7328b10fdd740a8
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/37575
Reviewed-by: Daniel Carvalho <odanrc@yahoo.com.br>
Maintainer: Bobby R. Bruce <bbruce@ucdavis.edu>
Tested-by: kokoro <noreply+kokoro@google.com>
3 years agoconfigs: Weed out old port terminology in Arm examples
Andreas Sandberg [Thu, 21 Jan 2021 17:03:17 +0000 (17:03 +0000)]
configs: Weed out old port terminology in Arm examples

Stop using the deprecated port names in Arm example scripts.

Change-Id: I11fea3e0df945ac64075b647766570604b70cad8
Signed-off-by: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/39582
Reviewed-by: Gabe Black <gabe.black@gmail.com>
Reviewed-by: Jason Lowe-Power <power.jg@gmail.com>
Maintainer: Jason Lowe-Power <power.jg@gmail.com>
Tested-by: kokoro <noreply+kokoro@google.com>
3 years agoconfigs: Remove Python 2 compatibility code in Arm configs
Andreas Sandberg [Thu, 21 Jan 2021 16:45:59 +0000 (16:45 +0000)]
configs: Remove Python 2 compatibility code in Arm configs

Remove uses of six and imports from __future__ and use native Python 3
functionality instead.

Change-Id: If37718ba99def2d6f176604e20d4ebeda75474ad
Signed-off-by: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/39581
Reviewed-by: Giacomo Travaglini <giacomo.travaglini@arm.com>
Reviewed-by: Richard Cooper <richard.cooper@arm.com>
Maintainer: Giacomo Travaglini <giacomo.travaglini@arm.com>
Tested-by: kokoro <noreply+kokoro@google.com>
3 years agoarch-arm, dev-arm: Remove Python 2 compatibility code
Andreas Sandberg [Thu, 21 Jan 2021 16:44:16 +0000 (16:44 +0000)]
arch-arm, dev-arm: Remove Python 2 compatibility code

Remove uses of six and imports from __future__ and use native Python 3
functionality instead.

Change-Id: Ifeb925c0b802f8186dd148e382aefe1c32fc8176
Signed-off-by: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/39580
Reviewed-by: Giacomo Travaglini <giacomo.travaglini@arm.com>
Maintainer: Giacomo Travaglini <giacomo.travaglini@arm.com>
Tested-by: kokoro <noreply+kokoro@google.com>
3 years agogpu-compute: Simplify LGKM decrementing for Flat instructions
Kyle Roarty [Tue, 19 Jan 2021 20:00:58 +0000 (14:00 -0600)]
gpu-compute: Simplify LGKM decrementing for Flat instructions

This commit makes it so LGKM count is decremented in a single place
(after completeAcc), which fixes a couple of potential bugs

1. Data is only written by completeAcc, not after initiateAcc. LGKM
count is supposed to be decremented after data is written.
2. LGKM count is now properly decremented for atomics without return

Change-Id: Ic791af3b42e04f7baaa0ce50cb2a2c6286c54f5a
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/39396
Reviewed-by: Matt Sinclair <mattdsinclair@gmail.com>
Reviewed-by: Matthew Poremba <matthew.poremba@amd.com>
Maintainer: Matt Sinclair <mattdsinclair@gmail.com>
Tested-by: kokoro <noreply+kokoro@google.com>
3 years agobase: Fix unsigned underflow mishandling.
Philip Metzler [Wed, 20 Jan 2021 20:59:48 +0000 (12:59 -0800)]
base: Fix unsigned underflow mishandling.

The second argument in the std::max call is treated as an unsigned value
as all variables are unsigned as well. This will result in an
unsigned underflow, and as such the std::max is a no-op and will result
in the underflowed value.

The `start` and `used` value get corrupted after that, and checks for
`empty` and other stuff downstream break.

Change-Id: I00017e22ba84e65f6b1c596f47d348f342fbc304
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/39496
Reviewed-by: Gabe Black <gabe.black@gmail.com>
Maintainer: Gabe Black <gabe.black@gmail.com>
Tested-by: kokoro <noreply+kokoro@google.com>
3 years agodev: Fixing EtherDevice stats initialization order
Hoa Nguyen [Tue, 19 Jan 2021 18:05:47 +0000 (10:05 -0800)]
dev: Fixing EtherDevice stats initialization order

Previously, the stat `totalBandwidth` is initialized before
`txBandwidth` and `rxBandwidth`. However, `totalBandwith` is of
 type Stats::Formula and `totalBandwidth = txBandwidth + rxBandwidth`.
Therefore, `totalBandwidth` should be initialized after the other two.

This change fixes the variable and stats initialization order accordingly.

The bug was reported here:  https://github.com/gem5/gem5/commit/3db48cbbc6e475592e6608b52a870d92ac2214aa#commitcomment-46094633.

Jira: https://gem5.atlassian.net/browse/GEM5-894

Change-Id: I2c7cc4120df672edf15b9a3ab6becc0bbebb778b
Signed-off-by: Hoa Nguyen <hoanguyen@ucdavis.edu>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/39395
Reviewed-by: Jason Lowe-Power <power.jg@gmail.com>
Maintainer: Jason Lowe-Power <power.jg@gmail.com>
Tested-by: kokoro <noreply+kokoro@google.com>
3 years agoriscv: Get rid of some unused constants.
Gabe Black [Mon, 18 Jan 2021 04:43:38 +0000 (20:43 -0800)]
riscv: Get rid of some unused constants.

Change-Id: I464e86dc6bfcd333a0bee32e56d9dcaa6fdf682d
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/39317
Maintainer: Gabe Black <gabe.black@gmail.com>
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Ayaz Akram <yazakram@ucdavis.edu>
Reviewed-by: Jason Lowe-Power <power.jg@gmail.com>
3 years agoarm: Use the "reg" ABI for gem5 ops.
Gabe Black [Mon, 18 Jan 2021 01:48:07 +0000 (17:48 -0800)]
arm: Use the "reg" ABI for gem5 ops.

The generic PseudoInstABI just calls back into the ISA specific
getArgument function, and that adds a lot of handling for cases that
aren't used and, besides those, basically just boils down to the "reg"
ABI anyway.

Change-Id: I57e738631dbccbf89cba3a6ca62b1f954b39e959
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/39316
Reviewed-by: Giacomo Travaglini <giacomo.travaglini@arm.com>
Maintainer: Giacomo Travaglini <giacomo.travaglini@arm.com>
Tested-by: kokoro <noreply+kokoro@google.com>
3 years agopython: Fix incorrect prefixes is m5.utils.convert
Andreas Sandberg [Tue, 19 Jan 2021 16:05:08 +0000 (16:05 +0000)]
python: Fix incorrect prefixes is m5.utils.convert

The conversion functions incorrectly assumed that kibibytes are 'kiB'
rather than 'KiB' (correct).

Change-Id: I7ef9e54546fdb3379435b40af6d9f619ad9b37a5
Signed-off-by: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/39375
Reviewed-by: Daniel Carvalho <odanrc@yahoo.com.br>
Reviewed-by: Jason Lowe-Power <power.jg@gmail.com>
Maintainer: Jason Lowe-Power <power.jg@gmail.com>
Tested-by: kokoro <noreply+kokoro@google.com>
3 years agosim: Don't serialise params in thermal models
Andreas Sandberg [Tue, 19 Jan 2021 10:36:11 +0000 (10:36 +0000)]
sim: Don't serialise params in thermal models

ThermalDomain and ThermalReference shouldn't serialise their params.

Change-Id: Idc4438b68c0db1fe312d37888c901f2ea87b1d60
Signed-off-by: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/39221
Reviewed-by: Gabe Black <gabe.black@gmail.com>
Reviewed-by: Jason Lowe-Power <power.jg@gmail.com>
Maintainer: Jason Lowe-Power <power.jg@gmail.com>
Tested-by: kokoro <noreply+kokoro@google.com>
3 years agosim: Thermal model style fixes
Andreas Sandberg [Tue, 19 Jan 2021 10:29:27 +0000 (10:29 +0000)]
sim: Thermal model style fixes

Fix various style issues in the thermal model implementation.

Change-Id: Ie31c862a23885f32f2931e927d7f87b7992bd099
Signed-off-by: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/39220
Reviewed-by: Gabe Black <gabe.black@gmail.com>
Reviewed-by: Daniel Carvalho <odanrc@yahoo.com.br>
Reviewed-by: Jason Lowe-Power <power.jg@gmail.com>
Maintainer: Jason Lowe-Power <power.jg@gmail.com>
Tested-by: kokoro <noreply+kokoro@google.com>
3 years agoarch-riscv,util: Add m5op.S for riscv to enable pseudo inst use
Ayaz Akram [Tue, 15 Dec 2020 09:33:52 +0000 (01:33 -0800)]
arch-riscv,util: Add m5op.S for riscv to enable pseudo inst use

This change adds assembly code for riscv pseudo instructions so
that they can be used with riscv benchmarks.

Change-Id: Ic979fd375e7750e92f52b900bf39e351f629fe2c
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/38515
Reviewed-by: Jason Lowe-Power <power.jg@gmail.com>
Reviewed-by: Gabe Black <gabe.black@gmail.com>
Maintainer: Jason Lowe-Power <power.jg@gmail.com>
Maintainer: Gabe Black <gabe.black@gmail.com>
Tested-by: kokoro <noreply+kokoro@google.com>
3 years agoarm: Export the mostly generic syscall ABI.
Gabe Black [Mon, 18 Jan 2021 01:40:57 +0000 (17:40 -0800)]
arm: Export the mostly generic syscall ABI.

This ABI is also applicable for gem5 ops. Rather than have the gem5 ops
use the syscall ABI, this change exports the syscall ABI and renames it
the "reg" ABI, or in other words an ABI which only uses registers. The
SE workload class then just creates a local name for the "reg" ABI so it
can continue to use it for system calls.

Change-Id: Ifaa38a94d6f0d49b8a2e515e02ce94472a499a00
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/39315
Reviewed-by: Giacomo Travaglini <giacomo.travaglini@arm.com>
Maintainer: Giacomo Travaglini <giacomo.travaglini@arm.com>
Tested-by: kokoro <noreply+kokoro@google.com>
3 years agotests: Fix syntax error in cpu_tests/test.py
Giacomo Travaglini [Mon, 18 Jan 2021 09:50:00 +0000 (09:50 +0000)]
tests: Fix syntax error in cpu_tests/test.py

The testsuite was not loaded with the following error:

Exception thrown while loading
<gem5>/tests/gem5/cpu_tests/test.py

Signed-off-by: Giacomo Travaglini <giacomo.travaglini@arm.com>
Change-Id: I1e88b8957bb24471e1bb6113ffc7c78886b6ed70
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/39295
Reviewed-by: Hoa Nguyen <hoanguyen@ucdavis.edu>
Maintainer: Jason Lowe-Power <power.jg@gmail.com>
Tested-by: kokoro <noreply+kokoro@google.com>
3 years agoconfigs: MemConfig, add QoSMemSinkCtrl support
Adrian Herrera [Fri, 2 Oct 2020 07:48:23 +0000 (08:48 +0100)]
configs: MemConfig, add QoSMemSinkCtrl support

QoSMemSinkInterface is a special case of memory interface type, similar
to SimpleMemory. It requires a QoSMemSinkCtrl where most model parameters
are exposed. By adding support in "config_mem", we allow configurations
with multiple QoSMemSinkCtrls to be centrally configured, particularly
interleaving parameters.

Change-Id: I46462b55d587acd2c861963bc0279bce92d5f450
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/35797
Reviewed-by: Jason Lowe-Power <power.jg@gmail.com>
Maintainer: Jason Lowe-Power <power.jg@gmail.com>
Tested-by: kokoro <noreply+kokoro@google.com>
3 years agotests: Stop using memcmp in the circlebuf test.
Gabe Black [Sat, 5 Dec 2020 07:44:59 +0000 (23:44 -0800)]
tests: Stop using memcmp in the circlebuf test.

Comparing arrays with memcmp is fairly easy to do and will correctly
identify when a value is incorrect, but gtest doesn't know what
comparison actually happened and can't print any diagnostic information
to help the person running the test determine what went wrong.

Unfortunately, gtest is also slightly too smart and also not smart
enough to make it easy to compare the contents of sub-arrays with each
other. The thing you're checking the value of *must* be an array with a
well defined size (not a pointer), and the size *must* exactly match the
number of elements it expects to find.

One fairly clean way to get around this problem would be to use the new
std::span type introduced in c++20 which lets you refer to a sub-section
of another container in place, adjusting indexing, sizing, etc as
needed. Unfortunately since we only require up to c++-14 currently we
can't use that type.

Instead, we can create vectors which hold copies of the required data.
This is suboptimal since it means we're copying around data which
doesn't really need to be copied, but it means that the templates in
gtest will get a type they can handle, and the sizes will match like it
expects them to. Since the number of checks/copies is still small, the
overhead should be trivial in practice.

A helper function, subArr, has been added to help keep things fairly
clutter free.

Change-Id: I9f88c583a6a742346b177dba7cae791824b65942
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/38895
Maintainer: Gabe Black <gabe.black@gmail.com>
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Daniel Carvalho <odanrc@yahoo.com.br>
3 years agobase: Rename Flags::update as Flags::replace
Daniel R. Carvalho [Sat, 26 Dec 2020 14:38:29 +0000 (11:38 -0300)]
base: Rename Flags::update as Flags::replace

The function name `update` is too generic. Given that
the expected functionality is to replace the selected
flag bits with the bits provided as parameters, rename
it as `replace`.

Change-Id: Ic7613ae09ecf9b92e31103b4e928192c07e9b640
Signed-off-by: Daniel R. Carvalho <odanrc@yahoo.com.br>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/38737
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Jason Lowe-Power <power.jg@gmail.com>
Maintainer: Bobby R. Bruce <bbruce@ucdavis.edu>

3 years agobase: Remove dubious/unused Flags functions
Daniel R. Carvalho [Sat, 26 Dec 2020 14:31:03 +0000 (11:31 -0300)]
base: Remove dubious/unused Flags functions

The functions isSet(), noneSet(), and allSet() assume that
all bits of the underlying container have a corresponding
flag. This is typically not true, and the likelihood of
incorrectly using these functions is high.

Fortunately these functions are not being used anywhere,
and can be safely removed.

Alternatively, a mask could be provided on construction to
determine which bits of the underlying container correspond
to a flag bit, so that the proper bits are checked in these
functions.

Change-Id: Ia7cbfd0726943506a3f04dc417e67a0b57cdbf95
Signed-off-by: Daniel R. Carvalho <odanrc@yahoo.com.br>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/38736
Maintainer: Bobby R. Bruce <bbruce@ucdavis.edu>
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Jason Lowe-Power <power.jg@gmail.com>
Reviewed-by: Bobby R. Bruce <bbruce@ucdavis.edu>
3 years agobase: Add unit tests for flags.hh
Daniel R. Carvalho [Sat, 26 Dec 2020 02:31:17 +0000 (23:31 -0300)]
base: Add unit tests for flags.hh

Add a unit test for flags.hh.

Change-Id: I58135b5f2fc016c30e1b4535f3daf46a77e99aff
Signed-off-by: Daniel R. Carvalho <odanrc@yahoo.com.br>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/38714
Reviewed-by: Bobby R. Bruce <bbruce@ucdavis.edu>
Tested-by: kokoro <noreply+kokoro@google.com>
Maintainer: Bobby R. Bruce <bbruce@ucdavis.edu>

3 years agoarch: Fix the code that computes MaxMiscDestReg.
Gabe Black [Mon, 7 Dec 2020 12:23:39 +0000 (04:23 -0800)]
arch: Fix the code that computes MaxMiscDestReg.

This code was simplified a little while ago, and the wrong variable name
was used in that computation accidentally. Fortunately the "wrong" value
would be too large, and so nothing bad would happen except a pair of
arrays would be overly large in the O3 instruction class.

Change-Id: I9694f1a8c79a62a172ef63bdd2f98fa0ace06acd
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/38383
Reviewed-by: Giacomo Travaglini <giacomo.travaglini@arm.com>
Maintainer: Giacomo Travaglini <giacomo.travaglini@arm.com>
Tested-by: kokoro <noreply+kokoro@google.com>
3 years agocpu: Remove the default, globally sized index arrays from StaticInst.
Gabe Black [Mon, 7 Dec 2020 12:19:52 +0000 (04:19 -0800)]
cpu: Remove the default, globally sized index arrays from StaticInst.

All ISAs now allocate their own arrays in all their instructions, making
these arrays unnecessary.

Change-Id: Ie2bc5d7a2903e07703dddd809505cdaaf6c578f5
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/38382
Reviewed-by: Giacomo Travaglini <giacomo.travaglini@arm.com>
Maintainer: Giacomo Travaglini <giacomo.travaglini@arm.com>
Tested-by: kokoro <noreply+kokoro@google.com>
3 years agoarm: Use local src and dest reg index arrays.
Gabe Black [Mon, 7 Dec 2020 12:19:08 +0000 (04:19 -0800)]
arm: Use local src and dest reg index arrays.

Change-Id: I7512c0abfaa2148b31b51fa43c3789bdca179105
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/38381
Reviewed-by: Giacomo Travaglini <giacomo.travaglini@arm.com>
Maintainer: Giacomo Travaglini <giacomo.travaglini@arm.com>
Tested-by: kokoro <noreply+kokoro@google.com>
3 years agoarch: Wrap InstObjParams with a class and not a function.
Gabe Black [Sun, 17 Jan 2021 09:28:07 +0000 (01:28 -0800)]
arch: Wrap InstObjParams with a class and not a function.

When parsing an ISA description, the InstObjParams class needs to have a
reference to the current parser. It does that by exposing a wrapper to
the description rather than the actual InstObjParams class. That wrapper
injects an additional argument into the InstObjParams constructor.

Originally, the wrapper which injectect the additional argument was a
function which masqueraded as a class. That made it impossible to
subclass InstObjParams.

Instead, this change replaces that function wrapper with a class
wrapper, and injects the extra argument in the __init__ method. This
preserves the fact that the InstObjParams name refers to a class, and
allows any sort of interaction that's normally allowed with a class like
subclassing.

Change-Id: I550ea2e60eadac3c7c0b9afa7d71f4607b49a5d2
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/39275
Reviewed-by: Giacomo Travaglini <giacomo.travaglini@arm.com>
Maintainer: Giacomo Travaglini <giacomo.travaglini@arm.com>
Tested-by: kokoro <noreply+kokoro@google.com>
3 years agoarch-arm: dtb_addr is already encoding the loadAddrOffset
Giacomo Travaglini [Mon, 18 Jan 2021 16:24:22 +0000 (16:24 +0000)]
arch-arm: dtb_addr is already encoding the loadAddrOffset

This fixes a bug in AArch32 where the dtb_address is
adding the loadAddrOffset twice to the dtb base address
after

https://gem5-review.googlesource.com/c/public/gem5/+/35076

Change-Id: Ia8bd35a02d998c54fbc3a889739c9abbeb506d96
Signed-off-by: Giacomo Travaglini <giacomo.travaglini@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/39216
Reviewed-by: Ciro Santilli <ciro.santilli@arm.com>
Tested-by: kokoro <noreply+kokoro@google.com>
3 years agoarch-gcn3,gpu-compute: Update stats style for GPU
Matthew Poremba [Thu, 14 Jan 2021 16:29:37 +0000 (10:29 -0600)]
arch-gcn3,gpu-compute: Update stats style for GPU

Convert all gpu-compute stats to Stats::Group style.

Change-Id: I29116f1de53ae379210c6cfb5bed3fc74f50cca5
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/39135
Reviewed-by: Matthew Poremba <matthew.poremba@amd.com>
Maintainer: Matthew Poremba <matthew.poremba@amd.com>
Tested-by: kokoro <noreply+kokoro@google.com>
3 years agoarch-riscv: fix incorrect interrupt checking logic
Cui Jin [Wed, 13 Jan 2021 02:36:56 +0000 (10:36 +0800)]
arch-riscv: fix incorrect interrupt checking logic

Whether global interrupt enabling or not is not simply decided by
xIE bit in mstatus, it also depends on current privilige level.
All level lower/higher than current should be disabled/enabled
regardless of the xIE bit. xIE bit is only control the enabling
of interrupt in current privilige level.

The fix is verified in FS.

Jira Issue: https://gem5.atlassian.net/browse/GEM5-883

Change-Id: I37f83ab77af2efbf1da9b81845828d322e49bf5f
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/39035
Reviewed-by: Jason Lowe-Power <power.jg@gmail.com>
Reviewed-by: Ayaz Akram <yazakram@ucdavis.edu>
Maintainer: Jason Lowe-Power <power.jg@gmail.com>
Tested-by: kokoro <noreply+kokoro@google.com>
3 years agomisc: Fix some includes
Daniel R. Carvalho [Sun, 27 Dec 2020 16:35:48 +0000 (13:35 -0300)]
misc: Fix some includes

Fix some missing and extra includes around the codebase.

Change-Id: Ibf314b43a966943a8096958f68382e1e245f29e3
Signed-off-by: Daniel R. Carvalho <odanrc@yahoo.com.br>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/38738
Reviewed-by: Bobby R. Bruce <bbruce@ucdavis.edu>
Maintainer: Bobby R. Bruce <bbruce@ucdavis.edu>
Tested-by: kokoro <noreply+kokoro@google.com>
3 years agoutil: Fix gem5_within_systemc and tlm coupling
Lukas Steiner [Fri, 15 Jan 2021 08:21:59 +0000 (09:21 +0100)]
util: Fix gem5_within_systemc and tlm coupling

Adapt the code to changes introduced with commits 3fb9139 and 392c1ce.

Jira Issue: https://gem5.atlassian.net/browse/GEM5-874

Change-Id: I0023157f1a2e5935f624c0a12e61f6b3e0d98b97
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/39215
Reviewed-by: Gabe Black <gabe.black@gmail.com>
Maintainer: Gabe Black <gabe.black@gmail.com>
Tested-by: kokoro <noreply+kokoro@google.com>
3 years agomem-ruby: Fix race related to atomics in VIPER
Matthew Poremba [Fri, 15 Jan 2021 01:03:00 +0000 (19:03 -0600)]
mem-ruby: Fix race related to atomics in VIPER

There is a race condition in VIPER where an atomic issued to the same
address can occur resulting in multiple trigger messages signalling the
compleition of the atomic operation. The first message was deallocating
the TBE causing the second message to dereference a nullptr when looking
up the TBE.

A counter is added to track the number of in flight AtomicDone trigger
messages. The AtomicDone is not called until the last in flight message
arrives at the trigger queue. The remaining messages call AtomicNotDone
which simply pops the message from the queue and keeps the TBE
allocated.

Change-Id: Ie1de0436861a7c393ad6d2fb2faceb83c18d4cc3
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/39175
Reviewed-by: Matt Sinclair <mattdsinclair@gmail.com>
Reviewed-by: Jason Lowe-Power <power.jg@gmail.com>
Maintainer: Matt Sinclair <mattdsinclair@gmail.com>
Tested-by: kokoro <noreply+kokoro@google.com>
3 years agoutil-docker: Add gdb to gcn-gpu Dockerfile
Matthew Poremba [Thu, 14 Jan 2021 16:31:41 +0000 (10:31 -0600)]
util-docker: Add gdb to gcn-gpu Dockerfile

gdb is generally useful.

Change-Id: Ic73822a5f61914cafba0699949dccabc81c03d2a
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/39096
Reviewed-by: Matt Sinclair <mattdsinclair@gmail.com>
Reviewed-by: Jason Lowe-Power <power.jg@gmail.com>
Reviewed-by: Bobby R. Bruce <bbruce@ucdavis.edu>
Reviewed-by: Kyle Roarty <kyleroarty1716@gmail.com>
Maintainer: Matt Sinclair <mattdsinclair@gmail.com>
Tested-by: kokoro <noreply+kokoro@google.com>
3 years agobase: Rename Flag status to enabled
Daniel R. Carvalho [Fri, 25 Dec 2020 20:49:09 +0000 (17:49 -0300)]
base: Rename Flag status to enabled

The semantic of "The status of a flag is true/false" may be
confusing. Rename `status` to `enabled`, so that queries
become "Is this flag enabled?".

Change-Id: I8cdd76766d80d65007a9f204abcf71b18211ab42
Signed-off-by: Daniel R. Carvalho <odanrc@yahoo.com.br>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/38711
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Jason Lowe-Power <power.jg@gmail.com>
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
Maintainer: Jason Lowe-Power <power.jg@gmail.com>

3 years agobase: Add unit test for debug.hh
Daniel R. Carvalho [Fri, 25 Dec 2020 18:56:18 +0000 (15:56 -0300)]
base: Add unit test for debug.hh

Add unit tests for the functions and classes in debug.hh.

breakpoint() is not being tested.

Change-Id: I1ae7318a043aa9dc479d4f970c1fd6c9946f4357
Signed-off-by: Daniel R. Carvalho <odanrc@yahoo.com.br>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/38710
Reviewed-by: Jason Lowe-Power <power.jg@gmail.com>
Maintainer: Jason Lowe-Power <power.jg@gmail.com>
Tested-by: kokoro <noreply+kokoro@google.com>
3 years agogpu-compute: Fix LGKM decrementing for flat atomic insts
Kyle Roarty [Fri, 15 Jan 2021 00:06:30 +0000 (18:06 -0600)]
gpu-compute: Fix LGKM decrementing for flat atomic insts

A prior commit (f6ec145fc0) fixed early LGKM decrementing for flat loads
and stores, but failed to address flat atomics.

Per the GCN3 ISA, LGKM count is decremented on flat atomics with return
when the data has been returned. This patch checks if the flat
instruction is an atomic with return, and decrements LGKM count if so.

Change-Id: I5c0c2c205a8b21327d4c42ba71c59842c15bd63b
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/39155
Reviewed-by: Matthew Poremba <matthew.poremba@amd.com>
Reviewed-by: Matt Sinclair <mattdsinclair@gmail.com>
Maintainer: Matthew Poremba <matthew.poremba@amd.com>
Tested-by: kokoro <noreply+kokoro@google.com>
3 years agosim: Rename the root stats group to RootStats
Hoa Nguyen [Mon, 11 Jan 2021 10:29:40 +0000 (02:29 -0800)]
sim: Rename the root stats group to RootStats

Currently, the name of the stats group of thr Root object is
Stats, which is likely to be confused with the Stats namespace.

This commit renames the struct to RootStats. This allows the
Stats namespace to be expressed as `Stats::`, which is
consistent with how the namespace is accessed in other part of
gem5.

Change-Id: Ieb425c3df1f5c0d5f11b1a467a36b2e0e07b2771
Signed-off-by: Hoa Nguyen <hoanguyen@ucdavis.edu>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/38915
Reviewed-by: Daniel Carvalho <odanrc@yahoo.com.br>
Maintainer: Jason Lowe-Power <power.jg@gmail.com>
Tested-by: kokoro <noreply+kokoro@google.com>
3 years agobase: Remove the curTick prototype from base/statistics.hh.
Gabe Black [Tue, 12 Jan 2021 07:54:43 +0000 (23:54 -0800)]
base: Remove the curTick prototype from base/statistics.hh.

This prototype might convince the compiler that it should refer to
curTick indirectly through the linker, but curTick is inline (and making
it not has very high overhead), so there's a decent chance no non-inline
version will be emitted.

Change-Id: Iab5aacb145d4a974bc1bc0abdf7275c40fbb9c38
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/38997
Reviewed-by: Daniel Carvalho <odanrc@yahoo.com.br>
Reviewed-by: Jason Lowe-Power <power.jg@gmail.com>
Maintainer: Jason Lowe-Power <power.jg@gmail.com>
Tested-by: kokoro <noreply+kokoro@google.com>
3 years agosim: Break the eventq.hh dependency in core.hh.
Gabe Black [Tue, 12 Jan 2021 07:44:10 +0000 (23:44 -0800)]
sim: Break the eventq.hh dependency in core.hh.

The original implementation of curTick used a thread local variable,
_curEventQueue, and its getCurTick() method, to retrieve the curTick for
the currently active event queue. That meant that core.hh needed to
include eventq.hh so that the EventQueue type was available, which also
indirectly brought in a lot of other dependencies.

Unfortunately this couldn't easily be fixed by making curTick()
non-inline since this added a significant amount of overhead when
tested.

Instead, this change makes the code in core.hh/core.cc keep a pointer
directly to a Tick. The code which sets _curEventQueue now also sets
that pointer when _curEventQueue changes.

The way curTick() now reaches into the guts of the current EventQueue
directly is not great from a modularity perspective, but if curTick is
considered an extension of the EventQueue, then it's just odd that this
part is broken out into a different file.

Change-Id: I8341b40fe75e90672eb1d70e1a368975fcbfe926
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/38996
Maintainer: Gabe Black <gabe.black@gmail.com>
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Daniel Carvalho <odanrc@yahoo.com.br>
Reviewed-by: Jason Lowe-Power <power.jg@gmail.com>
3 years agogpu-compute: Support for dynamic register alloc
gauravjain14 [Sat, 19 Oct 2019 23:51:41 +0000 (18:51 -0500)]
gpu-compute: Support for dynamic register alloc

SimplePoolManager doesn't allow mapping of two WGs
simultaneously on the same Compute Unit (provided
the previous WG has been mapped to all the SIMDs)
even if there is sufficient VRF and SRF space
available.

DynPoolManager takes care of that by dynamically
allocating and deallocating register file space
to wavefronts

Change-Id: I2255c68d4b421615d7b231edc05d3ebb27cbd66c
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/32034
Maintainer: Matt Sinclair <mattdsinclair@gmail.com>
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Jason Lowe-Power <power.jg@gmail.com>
Reviewed-by: Alexandru Duțu <alexandru.dutu@amd.com>
3 years agocpu-o3: Add missing tid in Rename's debug message
Daniel R. Carvalho [Thu, 14 Jan 2021 02:55:40 +0000 (23:55 -0300)]
cpu-o3: Add missing tid in Rename's debug message

These arguments were missing.

Change-Id: I8a76e46b2bcfc57f299145954fe72196f5969f29
Signed-off-by: Daniel R. Carvalho <odanrc@yahoo.com.br>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/39075
Reviewed-by: Giacomo Travaglini <giacomo.travaglini@arm.com>
Maintainer: Giacomo Travaglini <giacomo.travaglini@arm.com>
Tested-by: kokoro <noreply+kokoro@google.com>
3 years agoarch-riscv: CSR registers support in RISC-V remote GDB.
Peter Yuen [Mon, 11 Jan 2021 16:38:23 +0000 (00:38 +0800)]
arch-riscv: CSR registers support in RISC-V remote GDB.

Note:
Some less frequently needed CSR registers (e.g. hpm and pmp registers)
are commented out on purpose. Instructions to add them back are
described in remote_gdb.hh comments. This is to avoid spamming the
remote GDB log when using `info reg all`.

Changes:
1. Added GDB XML files to the ext/ directory (mostly from QEMU)
2. Modified RiscvGdbRegCache
- struct r: added CSR registers
- getRegs, setRegs: reading / setting CSR registers
3. Modified RemoteGDB
- availableFeatures: indicate support for XML registers
- getXferFeaturesRead: return XML blobs

Change-Id: Ica03b63edb3f0c9b6a7789228b995891dbfb26b2
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/38955
Reviewed-by: Jason Lowe-Power <power.jg@gmail.com>
Maintainer: Jason Lowe-Power <power.jg@gmail.com>
Tested-by: kokoro <noreply+kokoro@google.com>
3 years agodev: Let the pixel pump bypass the DMA FIFO in non-caching mode.
Gabe Black [Thu, 24 Dec 2020 15:53:57 +0000 (07:53 -0800)]
dev: Let the pixel pump bypass the DMA FIFO in non-caching mode.

When in non-caching mode, performance metrics are not meaningful, and
we're just interested in functional level behavior. Going through the
DMA FIFO in the HDLCD controller is very inefficient, and prevents
reading a batch of pixels from memory all in one go.

Change-Id: I3fb6d4d06730b5a94b5399f01aa02186baa5c9b3
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/38721
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
Maintainer: Andreas Sandberg <andreas.sandberg@arm.com>
Tested-by: kokoro <noreply+kokoro@google.com>
3 years agotests, base: Added GTests for base/amo.hh
Neil Natekar [Fri, 20 Mar 2020 05:18:26 +0000 (22:18 -0700)]
tests, base: Added GTests for base/amo.hh

Issue-on: https://gem5.atlassian.net/browse/GEM5-231

Change-Id: Ia915f9c8bd0732c6c918e8056253bd2fdcdf6b5d
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/26944
Reviewed-by: Daniel Carvalho <odanrc@yahoo.com.br>
Maintainer: Jason Lowe-Power <power.jg@gmail.com>
Tested-by: kokoro <noreply+kokoro@google.com>
3 years agoarch-arm: inform bootloader of kernel position with a register
Ciro Santilli [Mon, 21 Sep 2020 16:11:59 +0000 (17:11 +0100)]
arch-arm: inform bootloader of kernel position with a register

Before the commit, the bootloader had a hardcoded entry point that it
would jump to.

However, the Linux kernel arm64 v5.8 forced us to change the kernel
entry point because the required memory alignment has changed at:
https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/
commit/?h=v5.8&id=cfa7ede20f133cc81cef01dc3a516dda3a9721ee

Therefore the only way to have a single bootloader that boots both
pre-v5.8 and post-v5.8 kernels is to pass that information from gem5
to the bootloader, which we do in this patch via registers.

This approach was already used by the 32-bit bootloader, which passed
that value via r3, and we try to use the same register x3 in 64-bit.

Since we are now passing this information, the this patch also removes
the hardcoding of DTB and cpu-release-addr, and also passes those
values via registers.

We store the cpu-release-addr in x5 as that value appears to have a
function similar to flags_addr, which is used only in 32-bit arm and
gets stored in r5.

This commit renames atags_addr to dtb_addr, since both are mutually
exclusive, and serve a similar purpose, DTB being the newer recommended
approach.

Similarly, flags_addr is renamed to cpu_release_addr, and it is moved
from ArmSystem into ArmFsWorkload, since it is not an intrinsic system
property, and should be together with dtb_addr instead.

Before this commit, flags_addr was being set from FSConfig.py and
configs/example/arm/devices.py to self.realview.realview_io.pio_addr
+ 0x30. This commit moves that logic into RealView.py instead, and
sets the flags address 8 bytes before the start of the DTB address.

JIRA: https://gem5.atlassian.net/browse/GEM5-787
Change-Id: If70bea9690be04b84e6040e256a9b03e46710e10
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/35076
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
Maintainer: Andreas Sandberg <andreas.sandberg@arm.com>
Tested-by: kokoro <noreply+kokoro@google.com>
3 years agobase: Add documentation to flags.hh
Daniel R. Carvalho [Sat, 26 Dec 2020 14:13:02 +0000 (11:13 -0300)]
base: Add documentation to flags.hh

Add documentation to the Flags class. Use this opportunity
to rename some arguments to make their intention clearer.

Finally, the constructors have been merged using a default
value of 0.

Change-Id: I924b1d5c20a3e2066be64ab124ae1a5d96d4b3bf
Signed-off-by: Daniel R. Carvalho <odanrc@yahoo.com.br>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/38735
Reviewed-by: Jason Lowe-Power <power.jg@gmail.com>
Reviewed-by: Bobby R. Bruce <bbruce@ucdavis.edu>
Maintainer: Bobby R. Bruce <bbruce@ucdavis.edu>
Tested-by: kokoro <noreply+kokoro@google.com>
3 years agobase: Remove Flags<U> assignment
Daniel R. Carvalho [Sat, 26 Dec 2020 01:52:19 +0000 (22:52 -0300)]
base: Remove Flags<U> assignment

Currently unused and broken. Since these are templated classes,
and _flags is private, the assignment is a compilation error.
Furthermore, assignment of flags of different types is likely
undefined behavior.

Change-Id: I8430045c42c003efc74e343cc5b4a4350bc2ad92
Signed-off-by: Daniel R. Carvalho <odanrc@yahoo.com.br>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/38713
Reviewed-by: Jason Lowe-Power <power.jg@gmail.com>
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-by: Bobby R. Bruce <bbruce@ucdavis.edu>
Maintainer: Jason Lowe-Power <power.jg@gmail.com>
Tested-by: kokoro <noreply+kokoro@google.com>
3 years agobase: Assert Flags' type is unsigned
Daniel R. Carvalho [Sat, 26 Dec 2020 00:52:00 +0000 (21:52 -0300)]
base: Assert Flags' type is unsigned

Operations rely on the use of unsigned integers.

Change-Id: I825a88f81b54577585976d6558b1409870897721
Signed-off-by: Daniel R. Carvalho <odanrc@yahoo.com.br>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/38712
Reviewed-by: Jason Lowe-Power <power.jg@gmail.com>
Maintainer: Bobby R. Bruce <bbruce@ucdavis.edu>
Tested-by: kokoro <noreply+kokoro@google.com>
3 years agobase: Remove flag from allFlags on destruction
Daniel R. Carvalho [Fri, 25 Dec 2020 17:31:02 +0000 (14:31 -0300)]
base: Remove flag from allFlags on destruction

When a flag is destroyed it must be removed from the list
containing all flags.

Use this opportunity to remove "using namespace std" since
it is barely used.

Change-Id: I201371a770c56e11b92532e146d577c6ecb29d34
Signed-off-by: Daniel R. Carvalho <odanrc@yahoo.com.br>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/38709
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-by: Gabe Black <gabe.black@gmail.com>
Maintainer: Bobby R. Bruce <bbruce@ucdavis.edu>
Tested-by: kokoro <noreply+kokoro@google.com>
3 years agobase: Remove negation operator in Flag
Daniel R. Carvalho [Fri, 25 Dec 2020 20:44:14 +0000 (17:44 -0300)]
base: Remove negation operator in Flag

There is already a bool conversion operator, so there
is no need to provide a negation operator.

Change-Id: If5f99f8a0bb1707c115d139417aedd47bd162963
Signed-off-by: Daniel R. Carvalho <odanrc@yahoo.com.br>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/38708
Reviewed-by: Gabe Black <gabe.black@gmail.com>
Maintainer: Bobby R. Bruce <bbruce@ucdavis.edu>
Tested-by: kokoro <noreply+kokoro@google.com>
3 years agobase: Fix uninitialized variable in Flag
Daniel R. Carvalho [Fri, 25 Dec 2020 20:43:11 +0000 (17:43 -0300)]
base: Fix uninitialized variable in Flag

This was uninitialized, and was breaking expected values
under certain situations.

Change-Id: If51ab6ae038c7c397bc83de1c73af348c1db4ef8
Signed-off-by: Daniel R. Carvalho <odanrc@yahoo.com.br>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/38707
Reviewed-by: Bobby R. Bruce <bbruce@ucdavis.edu>
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
Maintainer: Bobby R. Bruce <bbruce@ucdavis.edu>
Tested-by: kokoro <noreply+kokoro@google.com>
3 years agomem-cache: Fix update of useful prefetches
Daniel R. Carvalho [Thu, 3 Dec 2020 11:06:25 +0000 (12:06 +0100)]
mem-cache: Fix update of useful prefetches

The probe notification must be parsed on every hit, even if
the prefetcher is set not to generate prefetches on accesses.
This fixes the calculation of useful prefetches.

Change-Id: Iff298f7bea11013f411f4ba39dba705fd81a0cd4
Signed-off-by: Daniel R. Carvalho <odanrc@yahoo.com.br>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/38177
Reviewed-by: Bobby R. Bruce <bbruce@ucdavis.edu>
Maintainer: Bobby R. Bruce <bbruce@ucdavis.edu>
Tested-by: kokoro <noreply+kokoro@google.com>
3 years agomisc: Fix missing includes.
Gabe Black [Tue, 12 Jan 2021 06:40:19 +0000 (22:40 -0800)]
misc: Fix missing includes.

Change-Id: I545ff03041e8fe66dc489c6aa95c009e54df0970
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/38995
Reviewed-by: Gabe Black <gabe.black@gmail.com>
Maintainer: Gabe Black <gabe.black@gmail.com>
Tested-by: kokoro <noreply+kokoro@google.com>
3 years agobase: Remove begin() and end() from CircleBuf.
Gabe Black [Tue, 12 Jan 2021 09:02:49 +0000 (01:02 -0800)]
base: Remove begin() and end() from CircleBuf.

These functions return iterators which are inconsistent with the usage
model for this type. It should be accessed using the peek, push, and pop
methods and not iterators. If you need a class with iterators which is
oriented around accessing individual elements at a time, the
CircularQueue type is likely a better choice.

Change-Id: I9f37eab12e490b63d870d378a91f601dad353f25
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/38998
Reviewed-by: Daniel Carvalho <odanrc@yahoo.com.br>
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
Maintainer: Gabe Black <gabe.black@gmail.com>
Maintainer: Andreas Sandberg <andreas.sandberg@arm.com>
Tested-by: kokoro <noreply+kokoro@google.com>
3 years agodev: Use regular atomic accesses for DMA in bypass mode.
Gabe Black [Thu, 24 Dec 2020 14:48:21 +0000 (06:48 -0800)]
dev: Use regular atomic accesses for DMA in bypass mode.

These are now accelerated with backdoor accesses and should be at least
as fast as functional accesses. This removes a dependency on port
proxies, and also stops the HDLCD from using functional accesses.

Change-Id: I5e959288eb533d09cffa7b79938aa2f61e4aff7d
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/38720
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
Maintainer: Andreas Sandberg <andreas.sandberg@arm.com>
Tested-by: kokoro <noreply+kokoro@google.com>
3 years agodev: Teach the DmaPort to use atomic memory backdoors.
Gabe Black [Thu, 24 Dec 2020 14:06:17 +0000 (06:06 -0800)]
dev: Teach the DmaPort to use atomic memory backdoors.

This is implemented similary to the NonCachingSimpleCPU, except that
both the normal atomic and noncaching atomic behaviors are implemented
by the same class. The sendDma function now dispatches to a method which
implements one or the other behavior since that function was getting too
big and complex.

Change-Id: I7972971ef41d1373424e587cf67c8444d50de748
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/38719
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
Maintainer: Andreas Sandberg <andreas.sandberg@arm.com>
Tested-by: kokoro <noreply+kokoro@google.com>
3 years agodev: Generate DMA packets as needed.
Gabe Black [Thu, 24 Dec 2020 12:15:56 +0000 (04:15 -0800)]
dev: Generate DMA packets as needed.

Instead of generating all of the DMA packets when a request is
initiated, keep track of the necessary properties and generate them as
needed.

The primary benefit of this approach is that if we don't actually need
packets, for instance if we can satisfy the request using a memory
backdoor, we can just skip them. Otherwise we'll have wasted time
creating them, and then have to clean them up.

Change-Id: I04d399fb7bce1ff9a44979c311be356baf2db555
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/38717
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
Maintainer: Andreas Sandberg <andreas.sandberg@arm.com>
Tested-by: kokoro <noreply+kokoro@google.com>
3 years agobase: Re-implement CircleBuf without using CircularQueue.
Gabe Black [Mon, 11 Jan 2021 05:56:49 +0000 (21:56 -0800)]
base: Re-implement CircleBuf without using CircularQueue.

CircularQueue provides iterators which make it easier for users to
interact with it and helps abstract its internal state, but at the same
time it prevents standard algorithms like std::copy from recognizing
opportunities to use bulk copies to speed up execution. It also hides
the seams when wrapping around the buffer happens which std::copy
wouldn't know how to handle.

CircleBuf seems to be intended as a simpler type which doesn't hold
complex entries like the CircularQueue does, and instead just acts as a
wrap around buffer, like the name suggests. This change reimplements it
to not use CircularQueue, and to instead use an underlying vector.

The way internal indexing is handled CircularQueue was simplified
recently, and using the same scheme here means that this code is
actually not much more verbose than it was before. It also intrinsically
handles indexing and bulk accesses, and uses std::copy_n in a way that
lets it recognize and take advantage of contiguous storage and bulk
copies.

Change-Id: I78e7cfe174c52f60c95c81e5cd3d71c6052d4d41
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/38896
Reviewed-by: Daniel Carvalho <odanrc@yahoo.com.br>
Maintainer: Gabe Black <gabe.black@gmail.com>
Maintainer: Andreas Sandberg <andreas.sandberg@arm.com>
Tested-by: kokoro <noreply+kokoro@google.com>
3 years agobase: Add a setNext method to the ChunkGenerator.
Gabe Black [Thu, 24 Dec 2020 14:05:14 +0000 (06:05 -0800)]
base: Add a setNext method to the ChunkGenerator.

This method lets you stretch the current chunk, if you want to skip over
some of the upcoming chunks since you've already handled their bytes.

For instance, if you were iterating over pages in a range of virtual
addresses, you might be able to handle multiple page sized chunks at
once if they were represented by a single large sized page table entry.
This mechanism would let you move past all the pages you had just
handled without having to walk through them all one by one.

Change-Id: I7d962f548947b77f0aa1b725036dbcc9e1b28659
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/38718
Reviewed-by: Gabe Black <gabe.black@gmail.com>
Maintainer: Gabe Black <gabe.black@gmail.com>
Tested-by: kokoro <noreply+kokoro@google.com>
3 years agogpu-compute: Fix FLAT insts decrementing lgkm count early
Kyle Roarty [Thu, 24 Dec 2020 01:30:37 +0000 (19:30 -0600)]
gpu-compute: Fix FLAT insts decrementing lgkm count early

FLAT instructions used to decrement lgkm count on execute, while the
GCN3 ISA specifies that lgkm count should be decremented on data being
returned or data being written.

This patch changes it so that lgkm is decremented after initiateAcc (for
stores) and after completeAcc (for loads) to better reflect the ISA
definition.

This fixes a bug where waitcnts would be satisfied even though the
memory access wasn't completed, which lead to instructions using the
wrong data.

Change-Id: I596cb031af9cda8d47a1b5e146e4a4ffd793d36c
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/38696
Reviewed-by: Matt Sinclair <mattdsinclair@gmail.com>
Reviewed-by: Matthew Poremba <matthew.poremba@amd.com>
Maintainer: Matt Sinclair <mattdsinclair@gmail.com>
Tested-by: kokoro <noreply+kokoro@google.com>
3 years agoconfigs: Remove default bootscript option for fs_bigLITTLE.py
Giacomo Travaglini [Wed, 6 Jan 2021 16:00:42 +0000 (16:00 +0000)]
configs: Remove default bootscript option for fs_bigLITTLE.py

Since the beginning fs_bigLITTLE has been pointing to a default

default_rcs = 'bootscript.rcS'

as a System.readfile parameter. That script is not present in
the gem5 repo and all the other fs scripts (starter_fs.py, fs.py
through Options.py) are using an emptry string as default
readfile param value.

We are hence aligning to the other scripts by removing this
default value

Change-Id: I20dc7714deae890d61706459c8d13bd8f5aac7a0
Signed-off-by: Giacomo Travaglini <giacomo.travaglini@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/38815
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
Maintainer: Andreas Sandberg <andreas.sandberg@arm.com>
Tested-by: kokoro <noreply+kokoro@google.com>
3 years agodev-arm: SMMUv3, enable interrupt interface
Adrian Herrera [Thu, 10 Dec 2020 18:07:21 +0000 (18:07 +0000)]
dev-arm: SMMUv3, enable interrupt interface

Users can set "irq_interface_enable" to allow software to program
SMMU_IRQ_CTRL and SMMU_IRQ_CTRLACK. This is required to boot Linux v5.4+
in a reasonable time. Notice the model does not implement architectural
interrupt sources, so no assertions will happen.

Change-Id: Ie138befdf5a204fe8fce961081c575c2166e22b9
Signed-off-by: Adrian Herrera <adrian.herrera@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/38555
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
Maintainer: Andreas Sandberg <andreas.sandberg@arm.com>

3 years agodev: Remove the return type from DmaPort::dmaAction.
Gabe Black [Thu, 24 Dec 2020 11:17:29 +0000 (03:17 -0800)]
dev: Remove the return type from DmaPort::dmaAction.

This function had a comment claiming that returning an arbitrary request
from the call was necessary for page table walker statistics, but
looking at the actual code, the return type was never used. Also
returning whatever the last request happens to be seems arbitrary, and a
bad boundary for modularization. The page table walker should not depend
on the internal implementation of the DMA port.

Change-Id: I00281fbaf6aeb85b15baf54f3d4a23ca1ac72b8b
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/38716
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
Maintainer: Andreas Sandberg <andreas.sandberg@arm.com>
Tested-by: kokoro <noreply+kokoro@google.com>
3 years agodev: Fix style in the pixel pump base class.
Gabe Black [Thu, 24 Dec 2020 08:24:27 +0000 (00:24 -0800)]
dev: Fix style in the pixel pump base class.

Change-Id: I8aa25911b367d36d6862780b39781f13724e79dc
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/38715
Reviewed-by: Daniel Carvalho <odanrc@yahoo.com.br>
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
Maintainer: Andreas Sandberg <andreas.sandberg@arm.com>
Tested-by: kokoro <noreply+kokoro@google.com>
3 years agodev: Style fixes in the ARM HDLCD device.
Gabe Black [Sat, 5 Dec 2020 08:30:02 +0000 (00:30 -0800)]
dev: Style fixes in the ARM HDLCD device.

Change-Id: I230e0e0db879a56bc23c3ed439b9722e76cdd8e4
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/38484
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-by: Daniel Carvalho <odanrc@yahoo.com.br>
Maintainer: Andreas Sandberg <andreas.sandberg@arm.com>
Tested-by: kokoro <noreply+kokoro@google.com>
3 years agoarch-riscv: fix the wrong cause register setting
Cui Jin [Thu, 31 Dec 2020 08:17:35 +0000 (16:17 +0800)]
arch-riscv: fix the wrong cause register setting

The most significant bit should be set based on interrupt or
exception. I assume in current RV64 implementation the bit should
be 63rd, rather than 31st. This causes interrupt handler to get
invalid cause code.

Minor bug is for the mpie is suppossed to be set to the value of
old mie.

The fix is verified in FS.

Jira Issue: https://gem5.atlassian.net/browse/GEM5-858

Change-Id: I1cc166c254b35f5c1acb3f5774c43149c61cc37a
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/38755
Reviewed-by: Jason Lowe-Power <power.jg@gmail.com>
Maintainer: Jason Lowe-Power <power.jg@gmail.com>
Tested-by: kokoro <noreply+kokoro@google.com>
3 years agoarch: Add a mechanism to pad the src or dest reg index arrays.
Gabe Black [Mon, 7 Dec 2020 12:16:09 +0000 (04:16 -0800)]
arch: Add a mechanism to pad the src or dest reg index arrays.

ARM reaches in and pads out the source register index list behind the
parser's back to force dest regs to also be sources in case an
instruction fails predication and needs to forward the original register
values. It shouldn't be hacking up these values in that way, but since
it is, this will let it continue to do so while still fitting in the new
system where each instruction allocates its src/dest reg index arrays to
size.

Change-Id: Ia296be9f63123f18f6cdc0d3bb1314d33e759b3a
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/38380
Reviewed-by: Giacomo Travaglini <giacomo.travaglini@arm.com>
Maintainer: Giacomo Travaglini <giacomo.travaglini@arm.com>
Tested-by: kokoro <noreply+kokoro@google.com>
3 years agodev: Cache the cacheLineSize in the DMA read FIFO.
Gabe Black [Sat, 5 Dec 2020 01:43:19 +0000 (17:43 -0800)]
dev: Cache the cacheLineSize in the DMA read FIFO.

This is a minor simplification which decouples the FIFO from the system
object at run time, although it does need to read the cache line size
out at construction time.

Change-Id: I57d96a676b9604663b6c9ed7c662640f507c5305
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/38482
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-by: Giacomo Travaglini <giacomo.travaglini@arm.com>
Reviewed-by: Daniel Carvalho <odanrc@yahoo.com.br>
Maintainer: Andreas Sandberg <andreas.sandberg@arm.com>
Tested-by: kokoro <noreply+kokoro@google.com>
3 years agodev: Make DMA devices use their own ports for functional accesses.
Gabe Black [Fri, 4 Dec 2020 09:49:48 +0000 (01:49 -0800)]
dev: Make DMA devices use their own ports for functional accesses.

DMA devices already have ports they use for non-functional accesses. We
can just attach a port proxy to that instead of getting one from the
system object.

Change-Id: I5e9adee43c7fe07b4c90978dbb7ec71468caadbb
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/38481
Maintainer: Gabe Black <gabe.black@gmail.com>
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Alexandru Duțu <alexandru.dutu@amd.com>
3 years agodev: Style fixes in src/dev/dma_device.(cc|hh).
Gabe Black [Fri, 4 Dec 2020 08:57:48 +0000 (00:57 -0800)]
dev: Style fixes in src/dev/dma_device.(cc|hh).

Change-Id: Ie72f30d95e7f889f9a440d0fed57a5940747b40d
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/38480
Reviewed-by: Daniel Carvalho <odanrc@yahoo.com.br>
Maintainer: Gabe Black <gabe.black@gmail.com>
Tested-by: kokoro <noreply+kokoro@google.com>
3 years agobase: Style fixes in the CircleBuf and Fifo classes.
Gabe Black [Fri, 4 Dec 2020 08:22:34 +0000 (00:22 -0800)]
base: Style fixes in the CircleBuf and Fifo classes.

Change-Id: Ia08548027973e2b18e09bc3f05a6498855bdd7f7
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/38479
Maintainer: Gabe Black <gabe.black@gmail.com>
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Daniel Carvalho <odanrc@yahoo.com.br>
3 years agobase,cpu: Simplify the CircularQueue class significantly.
Gabe Black [Fri, 4 Dec 2020 06:37:37 +0000 (22:37 -0800)]
base,cpu: Simplify the CircularQueue class significantly.

This class had been trying to keep all indices within the modulus of the
queue size, and to use all elements in the underlying storage by making
the empty and full conditions alias, differentiated by a bool. To keep
track of the difference between a storage location on one trip around
the queue vs other times around, ie an alias once the indices had
wrapped, it also keep track of a "round" value in both the queue itself,
and any iterators it created.

All this bookkeeping significantly complicated the data structure.
Instead, this change modifies it to keep track of a monotonically
increasing index which is wrapped at the time it's used. Only the head
index and current size need to be tracked in the queue itself, and only
a pointer to the queue and an index need to be tracked in the iterators.

Theoretically, it's possible that this value could overflow eventually
since it increases forever, unlike before where the index wrapped and
was never larger than the queue's capacity. In practice, the type of the
index was changed from a uint32_t to a size_t, probably a 64 bit value
in modern systems, which will hold much larger values. Also, the round
counter and the index values together acted like a smaller than 64 bit
value anyway, since the round counter would overflow after 2^32 times
around a less than 2^32 entry queue.

One minor interface difference is that the head() and tail() values
returned by the queue are no longer pre-wrapped to be modulo the queue's
capacity. As long as consumers don't try to be overly clever and feed in
fixed values, do their own bounds checking, etc., something that would
be cumbersome considering the wrapping nature of the structure, this
shouldn't be an issue.

Also, since external consumers no longer need to worry about wrapping,
since only one of them was used in only one place, and because they
weren't even marked as part of the interface, the modulo helper
functions have been eliminated from the queue. If other code wants to
perform modulo arithmetic for some reason (which the queue no longer
requires) they can accomplish basically the same thing in basically the
same amount of code using normal math.

Also, rather than inherit from std::vector, this change makes the vector
internal to the queue. That prevents methods of the vector that aren't
aware of the circular nature of the structure from leaking out if
they're not overridden or otherwise proactively blocked.

On top of simplifying the implementation, this also makes it perform
*slightly* better. To measure that, I ran the following command:

$ time build/ARM/base/circular_queue.test.opt --gtest_repeat=100000 > /dev/null

and found a few percent improvement in total run time. While this
difference was small and not measured against realistic usage of the
data structure, it was still measurable, and at minimum doesn't seem to
have hurt performance.

Change-Id: Ic2baa28de135be7086fa94579bbec451d69b3b15
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/38478
Reviewed-by: Daniel Carvalho <odanrc@yahoo.com.br>
Maintainer: Gabe Black <gabe.black@gmail.com>
Tested-by: kokoro <noreply+kokoro@google.com>
3 years agobase: Fix style issues in the circular queue.
Gabe Black [Fri, 4 Dec 2020 02:46:06 +0000 (18:46 -0800)]
base: Fix style issues in the circular queue.

Change-Id: I61da587d760019a338522f098745f375a5ce429e
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/38477
Reviewed-by: Daniel Carvalho <odanrc@yahoo.com.br>
Maintainer: Gabe Black <gabe.black@gmail.com>
Tested-by: kokoro <noreply+kokoro@google.com>
3 years agoarch-riscv: fix MIE csr register setting bugs
Cui Jin [Tue, 22 Dec 2020 08:46:46 +0000 (16:46 +0800)]
arch-riscv: fix MIE csr register setting bugs

Any changes on xIE bits changes should trigger the updating
of CSR register. The old condition is wrongly reversed.

The fix is verified in FS.

Jira Issue: https://gem5.atlassian.net/browse/GEM5-855

Change-Id: Ia2c6d3fbfd24d7f9d23f7cfa6f25f893544f4157
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/38578
Reviewed-by: Jason Lowe-Power <power.jg@gmail.com>
Reviewed-by: Ayaz Akram <yazakram@ucdavis.edu>
Maintainer: Jason Lowe-Power <power.jg@gmail.com>
Tested-by: kokoro <noreply+kokoro@google.com>
3 years agoext: googletest: upgrade to googletest 1.10
Yu-hsin Wang [Tue, 22 Dec 2020 02:48:50 +0000 (10:48 +0800)]
ext: googletest: upgrade to googletest 1.10

Upgrade googletest to 1.10
upstream commit: 703bd9caab50b139428cea1aaff9974ebee5742e
tag: release-1.10.0

sha1sum c4bb446ee3f8a730168097f44c31c90b62c63287 generated by command:
find . -type f ! -name SConscript ! -path "./.git/*" ! -name ".*" -print0 \
| sort -z | xargs -0 sha1sum | sha1sum

Change-Id: I6500e13e9c094efd99ebc72b3ae564861e67c159
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/38577
Reviewed-by: Gabe Black <gabe.black@gmail.com>
Reviewed-by: Bobby R. Bruce <bbruce@ucdavis.edu>
Maintainer: Bobby R. Bruce <bbruce@ucdavis.edu>
Tested-by: kokoro <noreply+kokoro@google.com>
4 years agoscons: Fix the scons script failing to install git hooks
Hoa Nguyen [Tue, 22 Dec 2020 12:39:29 +0000 (04:39 -0800)]
scons: Fix the scons script failing to install git hooks

A recent change,
https://gem5-review.googlesource.com/c/public/gem5/+/38616,
removed the 'root' environment variable, while the git hooks
installation script relies on the variable. This commit
addresses this problem.

Change-Id: I2d27861794021a6a60ef8668461e4127cb1db261
Signed-off-by: Hoa Nguyen <hoanguyen@ucdavis.edu>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/38655
Reviewed-by: Jason Lowe-Power <power.jg@gmail.com>
Maintainer: Jason Lowe-Power <power.jg@gmail.com>
Tested-by: kokoro <noreply+kokoro@google.com>
4 years agomem-cache: Generate error on compression misconfiguration
Daniel R. Carvalho [Mon, 21 Dec 2020 01:15:23 +0000 (22:15 -0300)]
mem-cache: Generate error on compression misconfiguration

Compressed caches must use the compressed tags, otherwise a
seg fault will be generated. Besides, if no compressor is
assigned; yet compressed tags are used, data is not compressed.

Generate an error for the first case, and a warning for the
second.

Change-Id: Iac5474ed919163ce38a8c4e8efd9727e5b3d8417
Signed-off-by: Daniel R. Carvalho <odanrc@yahoo.com.br>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/38635
Reviewed-by: Nikos Nikoleris <nikos.nikoleris@arm.com>
Maintainer: Nikos Nikoleris <nikos.nikoleris@arm.com>
Tested-by: kokoro <noreply+kokoro@google.com>