dev-arm: Implement LevelSensitive PPIs in GICv3

Change-Id: I7f28408eff7d502427c4486518c83506893f4a7a
Signed-off-by: Giacomo Travaglini <giacomo.travaglini@arm.com>
Reviewed-by: Nikos Nikoleris <nikos.nikoleris@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/31516
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
Maintainer: Andreas Sandberg <andreas.sandberg@arm.com>
Tested-by: kokoro <noreply+kokoro@google.com>

dev-arm: Implement LevelSensitive SPIs in GICv3

Change-Id: If918a8aea934f0037818cc64bf458076bfd0251d
Signed-off-by: Giacomo Travaglini <giacomo.travaglini@arm.com>
Reviewed-by: Nikos Nikoleris <nikos.nikoleris@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/31515
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
Maintainer: Andreas Sandberg <andreas.sandberg@arm.com>
Tested-by: kokoro <noreply+kokoro@google.com>

dev-arm: Gicv3 maintenance interrupt never cleared

The maintenance interrupt is a level sensitive interrupt though it has
been treated as edge triggered so far.

In order to be level sensitive, it needs to be cleared once the condition
which led to its generation are not valid anymore.

Change-Id: I9af9f4bf27622a7961393b00a145d6c9835d738b
Signed-off-by: Giacomo Travaglini <giacomo.travaglini@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/31614
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
Maintainer: Andreas Sandberg <andreas.sandberg@arm.com>
Tested-by: kokoro <noreply+kokoro@google.com>

arch-arm: Add System register trap check for EL1

This change adds and refactors the register trap checks
for EL1 in the same function, unifying the registry trapping

Change-Id: Ief3e0a9f70cc8cd44c1c8215515f36168927362d
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/31694
Reviewed-by: Giacomo Travaglini <giacomo.travaglini@arm.com>
Maintainer: Giacomo Travaglini <giacomo.travaglini@arm.com>
Tested-by: kokoro <noreply+kokoro@google.com>

cpu: differentiate snoop DPRINTF messages for AtomicSimpleCPU

Those three snoop messages were the same, which made interpreting logs
harder.

Change-Id: Ibff092932bc6d2ef0c5f15bf5f7ce031d1f1956b
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/30694
Reviewed-by: Jason Lowe-Power <power.jg@gmail.com>
Maintainer: Jason Lowe-Power <power.jg@gmail.com>
Tested-by: kokoro <noreply+kokoro@google.com>

systemc: Fix systemc array bound error

Currently, gcc 10.0 does not compile gem5 due to an array bound error
in sc_unsigned and sc_signed constructors.

Previous fix only ignores array-bounds warnings for sc_unsigned.cc and
sc_signed for gcc 10.1 onwards.

This commit turns the ignoring array-bounds warning flag on for all
gcc 10 onwards.

JIRA: https://gem5.atlassian.net/browse/GEM5-677.

Change-Id: Id65b6d52cef527f62917e09231ff79e2237affd8
Signed-off-by: Hoa Nguyen <hoanguyen@ucdavis.edu>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/31634
Reviewed-by: Gabe Black <gabeblack@google.com>
Maintainer: Gabe Black <gabeblack@google.com>
Tested-by: kokoro <noreply+kokoro@google.com>

base,arch: Fixed usage of `bitfield::replaceBits`

`bitfield::replaceBits` has two parameters, `first` and `last`, which
relate to the position of the MSB and the LSB of the bits to be replaced
respectively. Therefore `first` >= `last`. In some areas of the
codebase, this assumption has been flipped with `first` <= `last`. This
caused at least one known error, recorded here:
https://gem5.atlassian.net/browse/GEM5-695. These inconsistencies have
therefore been rectified.

A note has been added to the `bitfield::replaceBits` Doxygen to make
the usage of this function clearer.

Change-Id: Ie75856161d9a5684066430ecbdcc52e04e1e77bf
Issue-on: https://gem5.atlassian.net/browse/GEM5-696
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/31674
Reviewed-by: Bobby R. Bruce <bbruce@ucdavis.edu>
Maintainer: Bobby R. Bruce <bbruce@ucdavis.edu>
Tested-by: kokoro <noreply+kokoro@google.com>

arch-mips: Implement GDB XML target description for MIPS

Change-Id: Icff3b2c3e60d5989978de854247232afbb3b0dae
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/31574
Maintainer: Jason Lowe-Power <power.jg@gmail.com>
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Gabe Black <gabeblack@google.com>
Reviewed-by: Jason Lowe-Power <power.jg@gmail.com>

arch-arm: Fix Fault subsystem adding EL2Enable func

Change-Id: I7a4f0c22ac31fd56a8976ee8a1d9760cf6055d63
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/31374
Reviewed-by: Giacomo Travaglini <giacomo.travaglini@arm.com>
Maintainer: Giacomo Travaglini <giacomo.travaglini@arm.com>
Tested-by: kokoro <noreply+kokoro@google.com>

dev-arm: Check for security attribute when writing to ICFGR registers

This is matching the GICD_ICFGR read; a non secure access to a secure
interrupt should be treated as RAZ/WI

Change-Id: I9e92e03c13fe0474ed139b0ed22cebd5847b9109
Signed-off-by: Giacomo Travaglini <giacomo.travaglini@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/31615
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
Maintainer: Andreas Sandberg <andreas.sandberg@arm.com>
Tested-by: kokoro <noreply+kokoro@google.com>

dev-arm: Remove SPI/PPI range check in Gicv3 class

This is not needed since the check will already happen within

* Gicv3Distributor::sendInt
* Gicv3Redistributor::sendPPInt

Change-Id: I1355bde367127513f0501aa19e8f68d302c7a4f4
Signed-off-by: Giacomo Travaglini <giacomo.travaglini@arm.com>
Reviewed-by: Nikos Nikoleris <nikos.nikoleris@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/31514
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
Maintainer: Andreas Sandberg <andreas.sandberg@arm.com>
Tested-by: kokoro <noreply+kokoro@google.com>

util,tests: Updated compiler-tests to support paths with spaces

This script was failing if the gem5 directory path contained a space.
E.g: `/home/bobbys home/gem5`. This script has thereby been improved to
handle this.

Change-Id: Id56ff5a0e9aa31eef14de2111b040f7f2b0b286c
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/31534
Reviewed-by: Daniel Carvalho <odanrc@yahoo.com.br>
Maintainer: Jason Lowe-Power <power.jg@gmail.com>
Tested-by: kokoro <noreply+kokoro@google.com>

misc: Updated git description length policy to 72 chars

Advocating <=75 characters is non-standard. 72 characters is
recommended. Furthermore, within Gerrit, descriptions >72 characters
causes ugly line-wrapping. This commit updates this policy to
72 characters.

The corresponding change within the gem5 webpage:
https://gem5-review.googlesource.com/c/public/gem5-website/+/31134

Change-Id: Iaf542a241772f8e400d8f90624712de1a435d096
Issue-on: https://gem5.atlassian.net/browse/GEM5-665
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/31135
Reviewed-by: Daniel Carvalho <odanrc@yahoo.com.br>
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-by: Nikos Nikoleris <nikos.nikoleris@arm.com>
Reviewed-by: Giacomo Travaglini <giacomo.travaglini@arm.com>
Maintainer: Bobby R. Bruce <bbruce@ucdavis.edu>
Tested-by: kokoro <noreply+kokoro@google.com>

tests: fix .testsignore for long regression

removed 2 tests from long regression to get nightly tests to pass
fixed .testignore, -arch64 changed to -aarch64, this enables running on ARM hw

Change-Id: Ic8456fef83dfba3417d1edcf62e90873d197d3d6
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/31594
Maintainer: Jason Lowe-Power <power.jg@gmail.com>
Reviewed-by: Bobby R. Bruce <bbruce@ucdavis.edu>
Tested-by: kokoro <noreply+kokoro@google.com>

tests,arch-riscv: Migrated asmtest to testlib

Issue-on: https://gem5.atlassian.net/browse/GEM5-141
Change-Id: I8f02e53157d32e21591c51e99de803366ad8fd8f
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/27043
Maintainer: Bobby R. Bruce <bbruce@ucdavis.edu>
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Hoa Nguyen <hoanguyen@ucdavis.edu>

arch-gcn3,gpu-compute: Fix GCN3 related compiler errors

Fix all errors that were revealed using the util/compiler-test.sh
script.

Change-Id: Ie0d35568624e5e1405143593f0677bbd0b066b61
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/31154
Reviewed-by: Anthony Gutierrez <anthony.gutierrez@amd.com>
Reviewed-by: Jason Lowe-Power <power.jg@gmail.com>
Maintainer: Anthony Gutierrez <anthony.gutierrez@amd.com>
Tested-by: kokoro <noreply+kokoro@google.com>

mem: Use beats_per_clock as the DDR data rate for DRAMPower

The data rate is used by the drampower lib to estimate the power
consumption of the DRAM Core. Previously, we used the formula:

burst_cycles = divCeil(p->tBURST_MAX, p->tCK);
data_rate = p->burst_length / burst_cycles;

to derive the data_rate. However, under certain configurations this
formula computes the wrong result due to rounding errors. This patch
simplifies the way we derive the data_rate by passing the value of the
DRAM parameter beats_per_clock.

Change-Id: Ic8cd35bb4641d9c0a704675d2672a6fe4f4ec13e
Signed-off-by: Nikos Nikoleris <nikos.nikoleris@arm.com>
Reviewed-by: Wendy Elsasser <wendy.elsasser@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/30056
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Daniel Carvalho <odanrc@yahoo.com.br>

sim: Fixed error when compiling gem5 with dramsim2.

Compiling gem5 with dramsim2 included fails due to some inconsistencies in
including SimObjects. In this patch this issue is fixed along with
temporarily disabling -Werror=nonnull-compare in CCFLAGS. Also, the remote
for cloning dramsim2 has been changed.

Change-Id: Ia24095150d026d736352aaf0d735b7554ede10bb
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/31434
Reviewed-by: Jason Lowe-Power <power.jg@gmail.com>
Maintainer: Jason Lowe-Power <power.jg@gmail.com>
Tested-by: kokoro <noreply+kokoro@google.com>

gpu-compute, arch-gcn3: Change how waitcnts are implemented

Use single counters per memory operation type and increment
them upon issue, not execute.

Change-Id: I6afc0b66b21882538ef90a14a57a3ab3cc7bd6f3
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/29973
Reviewed-by: Anthony Gutierrez <anthony.gutierrez@amd.com>
Maintainer: Anthony Gutierrez <anthony.gutierrez@amd.com>
Tested-by: kokoro <noreply+kokoro@google.com>

gpu-compute: Add pipeline stage interface classes

This change separates the pipeline stage interfaces
for the GPU's compute unit into their own classes
with a well-defined interface. This helps to create
a cleaner interface for users to extend the CU
pipeline's capabilities and also helps consolidate
all the pipeline communication code in one place
in the source.

Change-Id: I569d52bce84dc1b9fbf8f0f96d53a81a2b6773c6
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/29972
Reviewed-by: Anthony Gutierrez <anthony.gutierrez@amd.com>
Maintainer: Anthony Gutierrez <anthony.gutierrez@amd.com>
Tested-by: kokoro <noreply+kokoro@google.com>

arch-gcn3: Add case to op selector when operand is vcc_hi

Change-Id: Ib8846656e18aad04ccb8c9112bc629c69078fe36
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/29971
Maintainer: Anthony Gutierrez <anthony.gutierrez@amd.com>
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Matt Sinclair <mattdsinclair@gmail.com>

gpu-compute: No RF scheduling in case of SKIP or EMPTY

In case of flat memory instructions the status for the
LM pipe execution unit is set to SKIP or EMPTY, as the bus
between the VRF and the GM and LM pipe is shared. The
destination operands should not be scheduled for the LM pipe,
event if the wave is in the dispatch list. This can lead
to deadlock in the destination cache as DCEs are reused
and the slotsAvailableForBank count gets artificially
incremented.

Change-Id: I2230c53e3bc1032d2cccbe00fab62c99ab8de6cd
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/29970
Maintainer: Anthony Gutierrez <anthony.gutierrez@amd.com>
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Matt Sinclair <mattdsinclair@gmail.com>

gpu-compute: Use refs to CU in pipe stages/mem pipes

The pipe stages and memory pipes are changed to store
a reference to their parent CU as opposed to a pointer.
These objects will never change which CU they belong to,
and they are constructed by their parent CU.

Change-Id: Ie5476e1e2e124a024c2efebceb28cb3a9baa78c1
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/29969
Reviewed-by: Anthony Gutierrez <anthony.gutierrez@amd.com>
Maintainer: Anthony Gutierrez <anthony.gutierrez@amd.com>
Tested-by: kokoro <noreply+kokoro@google.com>

arch-gcn3: Fix stride bug in buffer OOB detection logic

The out-of-range logic for buffer accesses is missing the top 4 bits of
const_stride when dealing with scratch buffers.  This can cause
perfectly valid scratch acceses to be suppressed when const_stride is
large.

Change-Id: I8f94d44c242fda26cf6dfb75db04fa3aca934b3e
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/29968
Reviewed-by: Anthony Gutierrez <anthony.gutierrez@amd.com>
Reviewed-by: Matt Sinclair <mattdsinclair@gmail.com>
Maintainer: Anthony Gutierrez <anthony.gutierrez@amd.com>
Tested-by: kokoro <noreply+kokoro@google.com>

arch-gcn3: Replace some instances of std::isnormal with std::fpclassify

Affected instructions: V_DIV_SCALE_F64, V_CMP_CLASS_F64,
V_CMPX_CLASS_F64 and their VOPC, VOP3, F32 variants.

These instances of std::isnormal were being used to check for
subnormal (denorms) values. std::isnormal is not specific enough.
It returns true for normal values but false for NaN, Inf, 0.0, and
subnormals. std::fpclassify returns macros for each category of
floating point numbers. Now we only catch subnormals.

Change-Id: I8d8f4452ff58de71e7c8e0b2b5e73467b532e196
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/29967
Maintainer: Anthony Gutierrez <anthony.gutierrez@amd.com>
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Matt Sinclair <mattdsinclair@gmail.com>

arch-gcn3: Fix VOP3 V_LDEXP_F64

Replaced !std::isnormal with std::fpclassify because std::isnormal
is not specific enough. !std::isnormal was incorrectly catching
NaN, Inf, 0.0, and subnormals (aka denormals), where as it was only
suppose to catch subnormals.

The return value and error handling spec of std::ldexp listed on
cppreference.com appears to match up in nearly all cases after
making these changes. If std::ldexp handled subnormals as described
in the GCN3 2016 guide, we could have used vdst[lane] = std::ldexp
and not need to check for any corner cases.

Change-Id: I4c77af77c3b7798f86d40442610cef1296a28441
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/29966
Maintainer: Anthony Gutierrez <anthony.gutierrez@amd.com>
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Matt Sinclair <mattdsinclair@gmail.com>

gpu-compute: Fix Y-dimension ABI decode

We currently have a bug in decoding workitem ID from the kernel
descriptor with multiple dimensions.  The enable_vgpr_workitem_id bits
are currently seperated into x and y components, when they should be
treated as a single 2 bit value, where y is enabled when it is > 0,
and z is enabled when it is > 1.  The current setup allows a kernel
launch with vgprs reserved for the z dimension and not the y dimension,
which is incorrect.

Change-Id: Iee64b207feb95bcf064898d5db33b8f201e25323
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/29965
Maintainer: Anthony Gutierrez <anthony.gutierrez@amd.com>
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Matt Sinclair <mattdsinclair@gmail.com>

arch-gcn3: Fix roundNearestEven for V_RNDNE_F64 and V_RNDNE_F32

roundNearestEven is an inst_util function that RNDNE_F64 and F32
call, including both VOP1 and VOP3 formats. IEEE 754 spec says this
function should round inputs to the nearest integer but round ties
to the nearest even integer. Prior to this patch it was rounding all
inputs to nearest even, not just the ties. It was probably implemented
this way originally because the language in the ISA manual is ambiguous
although it provided the correct logic.

Fixed roundNearestEven to use the semantics originally described in
the GCN3 ISA manual.

Change-Id: I83ecb1d516fcf5bdf17e54ddf409b447a129a9a7
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/29964
Maintainer: Anthony Gutierrez <anthony.gutierrez@amd.com>
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Matt Sinclair <mattdsinclair@gmail.com>

gpu-compute: Don't track vector store insts in CU's headTailMap

This change fixes a memory leak due to live GPUDynInstPtr references
to vector store insts being stored in the CU's headTailMap and never
released.

This happened because store insts are not supposed to have their
head-tail latencies tracked by the headTailMap; instead they use
timing information from the GPUCoalescer. When updating the
headTailLatency stat via the headTailMap, only loads were considered
and removed from the headTailMap, however when inserting into the
headTailMap loads and stores were considered, thus leading to the
memory leak.

This change fixes the issue by only adding loads to the headTailMap.

Change-Id: I8a8f5b79f55e00481ae5e82519a9ed627a7ecbd1
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/29963
Maintainer: Anthony Gutierrez <anthony.gutierrez@amd.com>
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Matt Sinclair <mattdsinclair@gmail.com>

arch-gcn3: add all s_buffer_load_dword instructions

Adds the other s_buffer_load_dword* instruction implementations to
f134a84.

Change-Id: I8d97527278900dc68c32463ea1824409ccd04e1d
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/29962
Maintainer: Anthony Gutierrez <anthony.gutierrez@amd.com>
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Matt Sinclair <mattdsinclair@gmail.com>

arch-gcn3: Add memcpy condition when writing EXEC_LO

Some compilers emit an error on the operand template class when writing
exec mask. Add a condition to explicitly set memcpy size argument to
32b or 64b based on the number of dwords.

Change-Id: I49b0e4a1680283e772d0a5a8efd687b31d4f1624
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/29961
Reviewed-by: Anthony Gutierrez <anthony.gutierrez@amd.com>
Reviewed-by: Matt Sinclair <mattdsinclair@gmail.com>
Maintainer: Anthony Gutierrez <anthony.gutierrez@amd.com>
Tested-by: kokoro <noreply+kokoro@google.com>

arch-gcn3: Remove invalid assert when reading EXEC_LO

This assert assumed all reads to EXEC_LO would be
64b, that is, we would always read the entire EXEC
mask. This is invalid as some kernels read only
the low 32b of EXEC.

The write to EXEC_LO is also updated to handle 32b
writes.

Change-Id: Ifeb167578515bf112b1eab70bbf2201a5e936358
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/29960
Maintainer: Anthony Gutierrez <anthony.gutierrez@amd.com>
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Matt Sinclair <mattdsinclair@gmail.com>

gpu_compute: Support loading BLIT kernels

The BLIT kernels used to implement DMA through the shaders don't fill
out all of the standard fields in an amd_kernel_code_t object.  This
patch modifies the code object parsing logic to support these new
kernels.

BLIT kernels are used in APUs when using ROCm memcopies for certain size
buffers, and are used for dGPUs when the SDMA engines are disabled.

Change-Id: Id4e667474d05e311097dbec443def07dfad14a79
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/29959
Maintainer: Anthony Gutierrez <anthony.gutierrez@amd.com>
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Matt Sinclair <mattdsinclair@gmail.com>

arch-gcn3: Implement ds_swizzle

Change-Id: I7d188388afa16932217ae207368666a724207c52
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/29958
Maintainer: Anthony Gutierrez <anthony.gutierrez@amd.com>
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Matt Sinclair <mattdsinclair@gmail.com>

arch-gcn3: Implement s_buffer_load_dwordx16

Change-Id: I25382dcae9bb55eaf035385fa925157f25d39c20
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/29957
Maintainer: Anthony Gutierrez <anthony.gutierrez@amd.com>
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Matt Sinclair <mattdsinclair@gmail.com>

arch-gcn3: Fixup DIV instructions

Adds support to handle the special cases
for GCN3 DIV instructions.

Change-Id: I18f91870e802407c93831f313ce76be053bc4230
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/29956
Reviewed-by: Anthony Gutierrez <anthony.gutierrez@amd.com>
Reviewed-by: Matt Sinclair <mattdsinclair@gmail.com>
Maintainer: Anthony Gutierrez <anthony.gutierrez@amd.com>
Tested-by: kokoro <noreply+kokoro@google.com>

systemc: Fix builds on Power systems

Based on the 64-bit ELF ABI for Power systems (ppc64 and
ppc64le), the data types int64_t and uint64_t are typedefs
of long and unsigned long respectively. If the SystemC
data types int64 and uint64 point to these, several errors
are observed while building the simulator on Power systems
due to ambiguity between the types when overloading some
operators and functions.

E.g.
  ...
  build/POWER/systemc/ext/channel/../dt/bit/sc_bit.hh:114:17: error: 'static bool sc_dt::sc_bit::to_value(sc_dt::int64)' cannot be overloaded with 'static bool sc_dt::sc_bit::to_value(long int)'
    114 |     static bool to_value(tp i) { return to_value((int)i); }
        |                 ^~~~~~~~
  ...
  build/POWER/systemc/ext/channel/../dt/bit/sc_bit.hh:114:17: note: previous declaration 'static bool sc_dt::sc_bit::to_value(long int)'
    114 |     static bool to_value(tp i) { return to_value((int)i); }
        |                 ^~~~~~~~
  ...

This adds a minor change to a SystemC datatype header to
ensure that the simulator can be built on Power systems.

Change-Id: Icd8bb38134bf98768cc38f9856d7d11a01ebaf21
Signed-off-by: Sandipan Das <sandipan@linux.ibm.com>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/31414
Reviewed-by: Jason Lowe-Power <power.jg@gmail.com>
Maintainer: Gabe Black <gabeblack@google.com>
Tested-by: kokoro <noreply+kokoro@google.com>

sim: Add a ProxyPtr test.

Change-Id: If71cc374030a5ef0dab62d351bc83960ff509af7
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/29401
Reviewed-by: Matthew Poremba <matthew.poremba@amd.com>
Maintainer: Gabe Black <gabeblack@google.com>
Tested-by: kokoro <noreply+kokoro@google.com>

arch-gcn3: fixed scale,fixup,fmas f64 ops

Change-Id: Ie13794554db8a958fda1f7103ec18058fda2e66d
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/29955
Maintainer: Anthony Gutierrez <anthony.gutierrez@amd.com>
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Matt Sinclair <mattdsinclair@gmail.com>

arch-gcn3: Fix s_getpc operand information

s_getpc was currently reporting only a single operand,
and was only considering the SSRC operand. However,
this instruction' source is implicitly the PC.
Because its destination register was never tracked for
dependence checking purposes, dependence violations
are possible.

Change-Id: Ia80b8b3e24d5885f646a9ee41212a2cb35b9ffe6
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/29954
Maintainer: Anthony Gutierrez <anthony.gutierrez@amd.com>
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Matt Sinclair <mattdsinclair@gmail.com>

arch-gcn3: Add handling for Inf/overflow in CVT insts

Change-Id: I0fddffdeaebd9f45fe89f44d536f80a43de63ff5
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/29953
Maintainer: Anthony Gutierrez <anthony.gutierrez@amd.com>
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Matt Sinclair <mattdsinclair@gmail.com>

arch-gcn3: Add ds_bpermute and ds_permute insts

The implementation of these insts provided by this
change is based on the description provided here:

https://gpuopen.com/amd-gcn-assembly-cross-lane-operations/

Change-Id: Id63b6c34c9fdc6e0dbd445d859e7b209023f2874
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/29952
Maintainer: Anthony Gutierrez <anthony.gutierrez@amd.com>
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Matt Sinclair <mattdsinclair@gmail.com>

arch-gcn3: ds_read_u8 and ds_read_u16 fix

This changeset zero extends the destination register
for ds_read_u8 and ds_read_u16 instructions.

Change-Id: I193adadd68adf2572b59743b1504f18ad225f506
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/29951
Maintainer: Anthony Gutierrez <anthony.gutierrez@amd.com>
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Matt Sinclair <mattdsinclair@gmail.com>

arch-gcn3: convert vALU instruction counters from 32 to 64-bit

The vALU instruction counters were previously 32 bits, but for some
workloads this value wraps around and triggers an assert failure
because the max vALU operations are reached.  To resolve this, this
commit increases the counter size to 64 bits.

Change-Id: I90ed4514669485cfea7ccc37ba9d69665277bccb
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/29950
Maintainer: Anthony Gutierrez <anthony.gutierrez@amd.com>
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Matt Sinclair <mattdsinclair@gmail.com>

arch-gcn3: implement instruction s_setreg_b32

Instruction s_setreg_b32 was unimplemented, but is used by hipified
rodinia 'srad'. The instruction sets values of hardware internal
registers. If the instruction is writing into MODE to control
single-precision FP round and denorm modes, a simple warn will be
printed; for all other cases (non-MODE hw register or other
precisions), panic will happen.

Change-Id: Idb1cd5f60548a146bc980f1a27faff30259e74ce
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/29949
Maintainer: Anthony Gutierrez <anthony.gutierrez@amd.com>
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Xianwei Zhang <xianwei.zhang@amd.com>

arch-gcn3: add support for v_mbcnt_hi and v_mbcnt_lo

Change-Id: I1c70fe693c904f1abd7d5a2b99220c74a075eae5
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/29948
Maintainer: Anthony Gutierrez <anthony.gutierrez@amd.com>
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Matt Sinclair <mattdsinclair@gmail.com>

arch-gcn3: fix bug with DPP support

Instructions that use the DPP field need to use the extra SRC0
register associated with the DPP instruction instead of the
"default" SRC0 register, since the default SRC0 register contains
the DPP information when DPP is being used.  This commit fixes
2735c3bb88 to take this into account.  Additionally, this commit
removes write of the src register from the DPP helper functions,
to avoid overwriting any changes made to the destination register.
Finally, this change modifies the instructions that use DPP to
simplify the flow through the execute() functions.

Change-Id: I80fd0af1f131f287f18ff73b3c1c9122d8c60823
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/29947
Maintainer: Anthony Gutierrez <anthony.gutierrez@amd.com>
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Matt Sinclair <mattdsinclair@gmail.com>

arch-gcn3: implement multi-dword buffer loads and stores

Add support for all multi-dword buffer loads and stores:
buffer_load_dword x2, x3, and x4 and buffer_store_dword x2, x3, and x4

Change-Id: I4017b6b4f625fc92002ce8ade695ae29700fa55e
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/29946
Maintainer: Anthony Gutierrez <anthony.gutierrez@amd.com>
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Matt Sinclair <mattdsinclair@gmail.com>

gpu-compute: Init CU object for pipe stages in their ctors

This change updates the constructors of the CU's pipe
stages/memory pipelines to accept a pointer to their
parent CU. Because the CU creates these objects, and
can pass a pointer to itself to these object via their
constructors, this is the safer way to initalize these
classes.

Change-Id: I0b3732ce7c03781ee15332dac7a21c097ad387a4
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/29945
Reviewed-by: Anthony Gutierrez <anthony.gutierrez@amd.com>
Maintainer: Anthony Gutierrez <anthony.gutierrez@amd.com>
Tested-by: kokoro <noreply+kokoro@google.com>

arch-gcn3: Add support for rd/wr EXEC_HI to operand class

Change-Id: Ib22dd604f88ea56801964235082835002deffca1
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/29944
Maintainer: Anthony Gutierrez <anthony.gutierrez@amd.com>
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Matt Sinclair <mattdsinclair@gmail.com>

gpu-compute, arch-gcn3: refactor barriers

Barriers were not modeled properly. Firstly, barriers were
allocated to each WG that was launched, which is not
correct, and the CU would provide an infinite number
of barrier slots. There are a limited number of barrier slots
per CU in reality. In addition, the CU will not allocate
barrier slots to WGs with a single WF (nothing to sync if
only one WF).

Beyond modeling problems, there also the issue of deadlock.
The barrier could deadlock because not all WFs are freed
from the barrier once it has been satisfied. Instead, we
relied on the scoreboard stage to release them lazily,
one-by-one.

Under this implementation the scoreboard may not fully release
all WFs participating in a barrier; this happens because the
first WF to be freed from the barrier could reach an s_barrier
instruction again, forever causing the barrier counts across
WFs to be out-of-sync.

This change refactors the barrier logic to:

1) Create a proper barrier slot implementation

2) Enforce (via a parameter) the number of barrier
   slots on the CU.

3) Simplify the logic and cleanup the code (i.e., we
   no longer iterate through the entire WF list each
   time we check if a barrier is satisfied).

4) Fix deadlock issues.

Change-Id: If53955b54931886baaae322640a7b9da7a1595e0
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/29943
Reviewed-by: Anthony Gutierrez <anthony.gutierrez@amd.com>
Maintainer: Anthony Gutierrez <anthony.gutierrez@amd.com>
Tested-by: kokoro <noreply+kokoro@google.com>

arch-gcn3: add support of 64-bit SOPK instruction

s_setreg_imm32_b32 is a 64-bit instruction, using a 32-bit literal
constant. Related functions are added to support decoding the second
dword.

Change-Id: I290f8578f726885c137dbfac3773035f814e0a3a
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/29942
Maintainer: Anthony Gutierrez <anthony.gutierrez@amd.com>
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Xianwei Zhang <xianwei.zhang@amd.com>

arch-gcn3: ensure that atomics follow HSA conventions

Add asserts to make sure atomics are following the HSA conventions
that atomics should be word aligned (i.e., can't be byte aligned)
and should not be misaligned such that a given lane's access
spans multiple cache lines.

Change-Id: Ia48758b9ed96764864234dc607f337e30e287d1c
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/29941
Maintainer: Anthony Gutierrez <anthony.gutierrez@amd.com>
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Matt Sinclair <mattdsinclair@gmail.com>

gpu-compute: Fix LDS out-of-bounds behavior

The LDS is capable of handling out-of-bounds accesses,
that is, accesses that are outside the bounds of the
chunk allocated to a WG. Currently, the simulator asserts
on these accesses. This patch changes the behavior of the
LDS to return 0 for reads and dropping writes that are
out-of-bounds.

Change-Id: I5f467d0f52113e8565e1a3029e82fb89cc6f07ea
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/29940
Maintainer: Anthony Gutierrez <anthony.gutierrez@amd.com>
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Matt Sinclair <mattdsinclair@gmail.com>

base,scons: Fixed `stats/hdf5.cc` CXXFlags to `-Wno-deprecated`

`Wno-deprecated-copy` was added to disable a warning in hdf5.cc:
https://gem5-review.googlesource.com/c/public/gem5/+/26325.

This works with GCC but does not work with clang. Clang returns
`error: unknown warning option '-Wno-deprecated-copy'; did you mean
'-Wno-deprecated'? [-Werror,-Wunknown-warning-option]` when this flag
is enabled. This flag has therefore been changed to `Wno-deprecated`.
This works in both GCC and Clang.

Change-Id: I38dd58f3007975ccb60b2eec936c3b200b3df3ca
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/31216
Reviewed-by: Jason Lowe-Power <power.jg@gmail.com>
Reviewed-by: Gabe Black <gabeblack@google.com>
Maintainer: Gabe Black <gabeblack@google.com>
Tested-by: kokoro <noreply+kokoro@google.com>

arch-arm: Added missing `break` to miscregs.cc

Later GCC compilers >=9 fail with a `this statement may fall through`
error due to this missing break.

Change-Id: I44b3386930a0b71b842a3a9b4837e4d6ad588f9d
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/31215
Reviewed-by: Giacomo Travaglini <giacomo.travaglini@arm.com>
Maintainer: Giacomo Travaglini <giacomo.travaglini@arm.com>
Tested-by: kokoro <noreply+kokoro@google.com>

sim: Added M5_VAR_USED to unused `cpu` var

The `BaseCPU cpu` variable unused when compiling gem5.fast. This
causes the compilation to fail. Adding the M5_VAR_USED resolves this
issue.

Change-Id: I62588563e9cde384e30755742d6bc754e819d7f4
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/31214
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>

util: Added cloudbuild_create_images.yaml

This cloudbuild file is used to build and upload our Docker images
to our Google Cloud infrastructure. To run:

```
gcloud builds submit --config \
util/cloudbuild/cloudbuild_create_images.yaml
```

Change-Id: I812584284a3cf79aac244a3b04a0f316f4281c49
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/30516
Reviewed-by: Hoa Nguyen <hoanguyen@ucdavis.edu>
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>

util,tests: Added gcc version 10 to compiler tests

In addition, gcc version 9 has been added, which was previously
served by "ubuntu-20.04_all-dependencies".

Change-Id: I57aeac2aa75b7751f0d4010efee7780e23d447d4
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/30515
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>

util: Added Dockerfile for GCC versions on ubuntu 20.04

Ubuntu 20.04 contains GCC verisons 9 and 10, which are not easily
obtainable (at least through APT) on Ubuntu 18.04. Therefore, a
Dockerfile for obtaining GCC versions in Ubuntu 20.04 has been added.
The orignal GCC version Dockerfile (Ubuntu 18.04) has been kept as
GCC versions 4.8, 5, and 6 are not obtainable, via APT, on Ubuntu
20.04. A complete migration to the 20.04 Dockerfile is not possible
until these earlier GCC versions are dropped.

The Docker images for GCC Versions 9 and 10 can be found here:
https://gcr.io/gem5-test/gcc-version-10
https://gcr.io/gem5-test/gcc-version-9

The other Dockerfile directories have been renamed for consistency.

Change-Id: I569249331095ee62d1be5be479c7ba7da0077422
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/30514
Reviewed-by: Daniel Carvalho <odanrc@yahoo.com.br>
Reviewed-by: Bobby R. Bruce <bbruce@ucdavis.edu>
Maintainer: Bobby R. Bruce <bbruce@ucdavis.edu>
Tested-by: kokoro <noreply+kokoro@google.com>

util,tests: Added compiler test script

This script runs a series of compilations on gem5. The following
compilers are tested:

clang-9
clang-8
clang-7
clang-6
clang-5
clang-4
clang-3.9
gcc-9
gcc-8
gcc-7
gcc-6
gcc-5
gcc-4.8 (to be dropped soon:
         https://gem5.atlassian.net/browse/GEM5-218)

They are tested by building the following build targets:

ARM
ARM_MESI_Three_Level
Garnet_standalone
GCN3_X86
MIPS
NULL_MESI_Two_Level
NULL_MOESI_CMP_directory
NULL_MOESI_CMP_token
NULL_MOESI_hammer
POWER
RISCV
SPARC
X86
X86_MOESI_AMD_BASE

For each, ".opt" and ".fast" compiler build settings are tested.

clang-9 and gcc-9 are tested against all targets with each build
setting. For the remaining compilers, a random build target is
chosen. After the script has run, the output of the tests can be
found in "compile-test-out".

Docker is required to run this script.

Change-Id: Id3bf4c89b9d424c87e9409930ee2aceaef72cb29
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/30395
Reviewed-by: Jason Lowe-Power <power.jg@gmail.com>
Reviewed-by: Daniel Carvalho <odanrc@yahoo.com.br>
Maintainer: Jason Lowe-Power <power.jg@gmail.com>
Tested-by: kokoro <noreply+kokoro@google.com>

mem-ruby: Add support for MemSync reqs in VIPER

Change-Id: Ib129e82be5348c641a8ae18093324bcedfb38abe
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/29939
Reviewed-by: Jason Lowe-Power <power.jg@gmail.com>
Maintainer: Anthony Gutierrez <anthony.gutierrez@amd.com>
Tested-by: kokoro <noreply+kokoro@google.com>

mem-ruby: Fix type casting in makeNextStrideAddress

The RubyPrefetcher uses makeNextStrideAddress() with
a negative stride to find prefetched address.
The type of this expression is:
uint64_t + uint32_t * int;
This gives wrong result due to implicit conversion.
Fix this with static cast and it works correctly:
uint64_t + int * int;

Change-Id: I36e17e00d5c66c3699fe1d5b287971225a162d04
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/31314
Reviewed-by: Jason Lowe-Power <power.jg@gmail.com>
Maintainer: Jason Lowe-Power <power.jg@gmail.com>
Tested-by: kokoro <noreply+kokoro@google.com>

arch-arm: AddressSize check on translateMmuOff for AArch64 only

Motivation:
An AddressSizeFault on AArch32 can only happen during a table walk
since the register used as a base by LD/ST is always 32 bit wide.
On AArch64 on the other hand, addresses can be 64bit wide;
when MMU is off (no virtual memory) an invalid physical address
can be specified

Change-Id: Id3ef170e99202c6b0b511fa7205c754956861720
Signed-off-by: Giacomo Travaglini <giacomo.travaglini@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/31274
Reviewed-by: Nikos Nikoleris <nikos.nikoleris@arm.com>
Tested-by: kokoro <noreply+kokoro@google.com>

misc: Add a code of conduct

This file codifies our current unofficial policies. This doesn't change
the community in any way except for clarifying our policies.

This was modeled off of https://www.contributor-covenant.org/.

Change-Id: Ib976636b490bbe4d46bf79260e6a345b46c02e2c
Signed-off-by: Jason Lowe-Power <jason@lowepower.com>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/30954
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
Maintainer: Jason Lowe-Power <power.jg@gmail.com>
Tested-by: kokoro <noreply+kokoro@google.com>

gpu-compute: enable kernel-end WB functionality

Change-Id: Ib17e1d700586d1aa04d408e7b924270f0de82efe
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/29938
Maintainer: Anthony Gutierrez <anthony.gutierrez@amd.com>
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Xianwei Zhang <xianwei.zhang@amd.com>

arch-gcn3: Implementation of flat atomic swap instruction

Change-Id: I9b9042899e65e8c9848b31c509eb2e3b13293e52
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/29937
Maintainer: Anthony Gutierrez <anthony.gutierrez@amd.com>
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Matt Sinclair <mattdsinclair@gmail.com>

misc: Remove support for checking out as a mercurial repo.

This will still be technically possible with the right converters, but
this removes the tags, ignore file, and style checking hooks related to
mercurial. We no longer maintain a mercurial mirror of the main git
repository, and this support adds clutter and could diverge from the git
style hooks, etc, over time.

Change-Id: Icf4833c4f0fda51ea98989d1d741432ae3ddc6dd
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/31174
Reviewed-by: Jason Lowe-Power <power.jg@gmail.com>
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
Maintainer: Gabe Black <gabeblack@google.com>
Maintainer: Andreas Sandberg <andreas.sandberg@arm.com>
Tested-by: kokoro <noreply+kokoro@google.com>

arch-gcn3: Fix VOP2 dissasembly prints

VOP2 prints VSRC1 register index as hex instead of decimal if the
instruction contains a literal operand.  This patch resets the
format specifiers in the stream to print the register correctly.

Change-Id: Icc7e6588b3c5af545be6590ce412460e72df253f
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/29936
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Anthony Gutierrez <anthony.gutierrez@amd.com>
Maintainer: Anthony Gutierrez <anthony.gutierrez@amd.com>

arch-gcn3, gpu-compute: Implement out-of-range accesses

Certain buffer out-of-range memory accesses should be special
cased and not generate memory accesses. This patch implements
those special cases and supresses lanes from accessing memory
when the calculated address falls in an ISA-specified out-of-range
condition.

Change-Id: I8298f861c6b59587789853a01e503ba7d98cb13d
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/29935
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Anthony Gutierrez <anthony.gutierrez@amd.com>
Reviewed-by: Matt Sinclair <mattdsinclair@gmail.com>
Maintainer: Anthony Gutierrez <anthony.gutierrez@amd.com>

arch-gcn3: Fix writelane src0,src1 usage

Src1 should only be used for lane select.  The data should come
from src0.

Change-Id: Ibe960df2e56d351a3819b40194104d2972a5cd4c
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/29933
Tested-by: kokoro <noreply+kokoro@google.com>
Maintainer: Anthony Gutierrez <anthony.gutierrez@amd.com>
Reviewed-by: Anthony Gutierrez <anthony.gutierrez@amd.com>
Reviewed-by: Matt Sinclair <mattdsinclair@gmail.com>

gpu-compute: Dropping fetchs when no entry is reserved in the buffer

This changeset drops fetches if there is no entry reserved in the
fetch buffer for that instruction. This can happen due to a fetch
attempted to be issued in the same cycle where a branch instruction
flushed the fetch buffer, while an ITLB or I-cache request is still
pending.

Change-Id: I3b80dbd71af27ccf790b543bd5c034bb9b02624a
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/29932
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Anthony Gutierrez <anthony.gutierrez@amd.com>
Reviewed-by: Onur Kayıran <onur.kayiran@amd.com>
Maintainer: Anthony Gutierrez <anthony.gutierrez@amd.com>

arch-gcn3: fix bits that SDWA selects

This commit fixes a bug in 200f2408 where the SDWA support was selecting bits
backwards.  As part of this commit, to help resolve this problem in the
future, I have added asserts in the helper functions in bitfield.hh to ensure
that the number of bits aren't negative.

Change-Id: I4b0ecb0e7c110600c0b5063101b75f9adcc512ac
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/29931
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Anthony Gutierrez <anthony.gutierrez@amd.com>
Reviewed-by: Matt Sinclair <mattdsinclair@gmail.com>
Maintainer: Anthony Gutierrez <anthony.gutierrez@amd.com>

arch-arm: Fix coding style in self_debug.[cc, hh]

Change-Id: I67be98af412b745ea9e16d4e8c6d422c9fbb29fc
Signed-off-by: Giacomo Travaglini <giacomo.travaglini@arm.com>
Reviewed-by: Nikos Nikoleris <nikos.nikoleris@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/31082
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
Maintainer: Andreas Sandberg <andreas.sandberg@arm.com>
Tested-by: kokoro <noreply+kokoro@google.com>

arch-arm: Remove getters/setters from SelfDebug class

Change-Id: I63e5ed25e453cb8fcb2c39ba0728cc81c499c166
Signed-off-by: Giacomo Travaglini <giacomo.travaglini@arm.com>
Reviewed-by: Nikos Nikoleris <nikos.nikoleris@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/31081
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
Maintainer: Andreas Sandberg <andreas.sandberg@arm.com>
Tested-by: kokoro <noreply+kokoro@google.com>

arch-arm: Fix pmc == on SelfDebug

The Assignment operator was used instead of the Equal-To

Change-Id: Ibf5a0006bce79b67d662fd1f8942699582956d58
Signed-off-by: Giacomo Travaglini <giacomo.travaglini@arm.com>
Reviewed-by: Nikos Nikoleris <nikos.nikoleris@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/31080
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
Maintainer: Andreas Sandberg <andreas.sandberg@arm.com>
Tested-by: kokoro <noreply+kokoro@google.com>

arch-arm: Move breakpoint/watchpoint check out of the TLB

The breakpoint, watchpoint, vector catch and software step checks
have been moved from the TLB to the SelfDebug class.

This is cleaningup the TLB model which is simply asking the SelfDebug
class if there is a pending debug fault

Change-Id: I1724896b24e4728b32a6b46c5cd51cc6ef279fd7
Signed-off-by: Giacomo Travaglini <giacomo.travaglini@arm.com>
Reviewed-by: Nikos Nikoleris <nikos.nikoleris@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/31079
Maintainer: Andreas Sandberg <andreas.sandberg@arm.com>
Tested-by: kokoro <noreply+kokoro@google.com>

dev-arm: Style fixes for src/dev/arm/gic_v2.hh

JIRA: https://gem5.atlassian.net/browse/GEM5-667

Change-Id: I80dce7b72775beabafa3b54e915a369571f2e4c9
Signed-off-by: Giacomo Travaglini <giacomo.travaglini@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/31057
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
Maintainer: Andreas Sandberg <andreas.sandberg@arm.com>
Tested-by: kokoro <noreply+kokoro@google.com>

dev-arm: Implement Level Sensitive PPIs in GICv2

JIRA: https://gem5.atlassian.net/browse/GEM5-667

Change-Id: I9ae411110f08f4a1de95469ff5ed6788354abafc
Signed-off-by: Giacomo Travaglini <giacomo.travaglini@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/31056
Reviewed-by: Hsuan Hsu <kugwa2000@gmail.com>
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
Maintainer: Andreas Sandberg <andreas.sandberg@arm.com>
Tested-by: kokoro <noreply+kokoro@google.com>

dev-arm: Use getIntConfig when reading/writing GICD_ICFGR

This patch is changing the getIntConfig helper (which has been
used so far by isLevelSensitive only) to make it usable by the
read/writes of the GICD_ICFGR register.

While the helper was previously returning the irq config bits
provided a single irq as an input, this new version is returning
the entire GICD_ICFGR word (read/writable)

JIRA: https://gem5.atlassian.net/browse/GEM5-667

Change-Id: I07e455a9e2819fed1f97a0e372d9d9a2e5ad4801
Signed-off-by: Giacomo Travaglini <giacomo.travaglini@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/31055
Reviewed-by: Hsuan Hsu <kugwa2000@gmail.com>
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
Maintainer: Andreas Sandberg <andreas.sandberg@arm.com>
Tested-by: kokoro <noreply+kokoro@google.com>

dev-arm: Move GICv2 intConfig for consistency

Every other helper is placed below the respective array storage

JIRA: https://gem5.atlassian.net/browse/GEM5-667

Change-Id: I398ac23eb68d84a8e0ed856550bfac8e403a86b3
Signed-off-by: Giacomo Travaglini <giacomo.travaglini@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/31054
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
Maintainer: Andreas Sandberg <andreas.sandberg@arm.com>
Tested-by: kokoro <noreply+kokoro@google.com>

arch,cpu: Consolidate most of the StackTrace classes into a base class.

These classes are all basically empty now that Alpha has been deleted,
except in cases where the arch versions had copied versions of the Alpha
code.

This change pulls all the generic logic out of the arch versions, making
the arch versions much simpler and making it clearer what the core
functionality of the class is, and what parts are architecture specific
details.

In the future, the way the StackTrace class is instantiated should be
delegated to the Workload class so that ISA agnostic code doesn't need
to know about a particular ISA's StackTrace class, and so that
StackTrace logic can, at least theoretically, be specialized for a
particular workload. The way a stack trace is collected could vary from
OS to OS, for example.

Change-Id: Id8108f94e9fe8baf9b4056f2b6404571e9fa52f1
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/30961
Reviewed-by: Gabe Black <gabeblack@google.com>
Maintainer: Gabe Black <gabeblack@google.com>
Tested-by: kokoro <noreply+kokoro@google.com>

cpu: Slightly modernize and simplify code in cpu/profile.(hh|cc).

Change-Id: Ideb104d20b333305ead2356cbfff2aac2e0173b5
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/30960
Maintainer: Gabe Black <gabeblack@google.com>
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>

arch-arm: Initialized some variables

Some of the variables in pauth_helpers.cc
are uninitialized in certain control paths
which causes a compiler warning. We initialize
these to false since they should be updated
to the correct value in all valid code paths.

Change-Id: If34d7daaf2404c2cf014c7b4c0c2f979580f36b9
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/31094
Reviewed-by: Giacomo Travaglini <giacomo.travaglini@arm.com>
Maintainer: Giacomo Travaglini <giacomo.travaglini@arm.com>
Tested-by: kokoro <noreply+kokoro@google.com>

arch-arm: Add basic support for KVM_CAP_ARM_USER_IRQ

KVM_CAP_ARM_USER_IRQ is a KVM extension introduced in newer versions of
Linux (>= 4.12). It supports delivering interrupt from the kernel-space
timer to the user-space GIC, which means that it will be unnecessary to
use the memory-mapped timer and emulate it in gem5 anymore.

Using the option provided by this change, Linux is able to boot with 1
CPU successfully, and the speed is slightly faster then the memory-
mapped timer option. However, multicore seems to hang during boot and
still needs more investigation to be enabled.

JIRA: https://gem5.atlassian.net/browse/GEM5-663

Change-Id: I146bbcce3cf66f8f5ebee04ea5f1b9f54868721a
Signed-off-by: Hsuan Hsu <hsuan.hsu@mediatek.com>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/30921
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
Maintainer: Andreas Sandberg <andreas.sandberg@arm.com>

cpu-kvm,arch-arm: Improve KvmCPU tick event scheduling

The memory-mapped timer emulated by gem5 is driven by the underlying
gem5 tick, which means that we must align the tick with the host time
to make the timer interrupt fire at a nearly native rate.

In each KVM execution round, the number of ticks incremented is
directly calculated from the number of instructions executed. However,
when a guest CPU switches to idle state, KVM seems to stay in kernel-
space until the POSIX timer set up in user-space raises an expiration
signal, instead of trapping to user-space immediately; and somehow the
instruction count is just too low to match the elapsed host time. This
makes the gem5 tick increment very slowly when the guest is idle and
drastically slow down workloads being sensitive to the guest time which
is driven by timer interrupt.

Before switching to KVM to execute the guest code, gem5 programs the
POSIX timer to expire according to the remaining ticks before the next
event in the event queue. Based on this, we can come up with the
following solution: If KVM returns to user-space due to POSIX timer
expiration, it must be time to process the next gem5 event, so we just
fast-forward the tick (by scheduling the next CPU tick event) to that
event directly without calculating from the instruction count.

There is one more related issue needed to be solved. The KVM exit
reason, KVM_EXIT_INTR, was treated as the case where the KVM execution
was disturbed by POSIX timer expiration. However, there exists a case
where the exit reason is KVM_EXIT_INTR but the POSIX timer has not
expired. Its cause is still unknown, but it can be observed via the
"old_value" argument returned by timer_settime() when disarming the
POSIX timer. In addition, it seems to happen often when a guest CPU is
not in idle state. When this happens, the above tick event scheduling
incorrectly treats KVM_EXIT_INTR as POSIX timer expiration and fast-
forwards the tick to process the next event too early. This makes the
guest feel external events come too fast, and will sometimes cause
trouble. One example is the VSYNC interrupt from HDLCD. The guest seems
to get stuck in VSYNC handling if the KVM CPU is not given enough time
between each VSYNC interrupt to complete a service. (Honestly I did not
dig in to see how the guest handled the VSYNC interrupt and how the
above situation became trouble. I just observed from the debug trace of
GIC & HDLCD & timer, and made this conclusion.) This change also uses
a workaround to detect POSIX timer expiration correctly to make the
guest work with HDLCD.

JIRA: https://gem5.atlassian.net/browse/GEM5-663

Change-Id: I6159238a36fc18c0c881d177a742d8a7745a23ca
Signed-off-by: Hsuan Hsu <hsuan.hsu@mediatek.com>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/30919
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
Maintainer: Andreas Sandberg <andreas.sandberg@arm.com>
Tested-by: kokoro <noreply+kokoro@google.com>

dev-arm: Fix handling of writing timer control registers

We should also deal with change of the imask bit, or we will lose timer
interrupt if the timer expires before the guest kernel unmasks the bit.
More precisely, consider the following common pattern in timer interrupt
handling:

    1. Set the interrupt mask bit (CNTV_CTL.IMASK)
    2. Reprogram the downcounter (CNTV_TVAL) for the next interrupt
    3. Clear the interrupt mask bit (CNTV_CTL.IMASK)

The timer can expires between step 2 & 3 if the value programmed in step
2 is small enough, and this seems very likely to happen in KVM mode. If
we don't check for timer expiration right after unmasking, we will miss
the only chance to inject the interrupt.

JIRA: https://gem5.atlassian.net/browse/GEM5-663

Change-Id: I75e8253bb78d15ae72cb985ed132f896d8e92ca6
Signed-off-by: Hsuan Hsu <hsuan.hsu@mediatek.com>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/30918
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
Maintainer: Andreas Sandberg <andreas.sandberg@arm.com>
Tested-by: kokoro <noreply+kokoro@google.com>

cpu-kvm: Initialize _hasKernelIRQChip in the constructor

This class member was only correctly set to true when using an in-kernel
interrupt controller, but was un-initialized when trying to use a user-
space one and would cause trouble.

JIRA: https://gem5.atlassian.net/browse/GEM5-663

Change-Id: I71b052c6da7e8790b05a15c07e7933bc4f912785
Signed-off-by: Hsuan Hsu <hsuan.hsu@mediatek.com>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/30917
Reviewed-by: Giacomo Travaglini <giacomo.travaglini@arm.com>
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
Maintainer: Giacomo Travaglini <giacomo.travaglini@arm.com>
Maintainer: Andreas Sandberg <andreas.sandberg@arm.com>
Tested-by: kokoro <noreply+kokoro@google.com>

dev-arm: Make generic timer work with level-sensitive support

Support for level-sensitive PPIs and SPIs has been added to GICv2 now.
It is therefore the timer's responsibility to notify GICv2 to clear its
interrupt pending state. Without doing this, the guest will get stuck
in just a single round of the interrupt handler because GICv2 does not
clear the pending state, and eventually make the guest treat this
interrupt as problematic and then just disable it.

JIRA: https://gem5.atlassian.net/browse/GEM5-663

Change-Id: Ia8fd96bf00b28e91aa440274e6f8bb000446fbe3
Signed-off-by: Hsuan Hsu <hsuan.hsu@mediatek.com>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/30916
Reviewed-by: Giacomo Travaglini <giacomo.travaglini@arm.com>
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
Maintainer: Giacomo Travaglini <giacomo.travaglini@arm.com>
Maintainer: Andreas Sandberg <andreas.sandberg@arm.com>
Tested-by: kokoro <noreply+kokoro@google.com>

configs: Add earlycon to default kernel_cmd.

The earlyprintk kernel command line argument does not take a value on Arm.
Rather pass early console name using the earlycon command line argument.

Change-Id: Ie14fc425e87c50a0b59fa4270a3743ed4fe97589
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/31074
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
Maintainer: Andreas Sandberg <andreas.sandberg@arm.com>
Tested-by: kokoro <noreply+kokoro@google.com>

mem: Optionally share the backing store

This patch adds the ability for a host-OS process external to gem5
to access the backing store via POSIX shared memory.
The new param shared_backstore of the System object is the filename
of the shared memory (i.e., the first argument to shm_open()).

Change-Id: I98c948a32a15049a4515e6c02a14595fb5fe379f
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/30994
Reviewed-by: Jason Lowe-Power <power.jg@gmail.com>
Maintainer: Jason Lowe-Power <power.jg@gmail.com>
Tested-by: kokoro <noreply+kokoro@google.com>

arm: Don't use pseudo instructions to implement regular instructions.

Some ARM instructions were using quiesce and quiesceSkip pseudo
instruction bodies instead of implementing the one line of each of those
functions themselves. This creates two problems. First, it adds an
artificial depedence on the pseudo instruction implementations. Second,
it would confusing cause pseudo instruction DPRINTFs to fire when normal
instructions were executing.

This change simply replaces the calls with their targets one line
implementation, with some very minor duplication from multiple call
sights factored out into a local variable.

Change-Id: I596eafd8714227fa7f69edd542108598c9809b11
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/27790
Reviewed-by: Giacomo Travaglini <giacomo.travaglini@arm.com>
Reviewed-by: Jason Lowe-Power <power.jg@gmail.com>
Maintainer: Giacomo Travaglini <giacomo.travaglini@arm.com>
Tested-by: kokoro <noreply+kokoro@google.com>

dev-arm: Verify number of CPUs when restoring Generic Timer Cpts.

When restoring a checkpoint containing a generic timer, the checkpoint
expects to connect the timer to the same number of CPUs that were
present when the checkpoint was taken. If the number of CPUs in the
new simulation is different, deserialization will fail. In the case
that the number of CPUs expected by the checkpoint is greater than the
number of CPUs present, this will cause a segmentation fault caused by
reading off the end of the list of Thread Contexts.

This commit fixes the problem by checking the number of CPUs present
in the simulation matches the number of CPUs expected by the generic
timer checkpoint. If there is a mismatch, a fatal error is triggered
with an informative message to the user.

Change-Id: Iff9ad68d64e67b3df51682b7e4e272e5f355bcd6
Reviewed-by: Giacomo Travaglini <giacomo.travaglini@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/30576
Tested-by: kokoro <noreply+kokoro@google.com>
Maintainer: Giacomo Travaglini <giacomo.travaglini@arm.com>

arch-arm: mark ID_AA64ISAR1_EL1.JSCVT implemented

The feature was implemented at: I1b24839daef775bbb1eb9da5f32c4bb3843e0b28

Change-Id: I0c0f55e55a1ca3ca6bf40206a989ef0bb353ee84
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/30934
Reviewed-by: Giacomo Travaglini <giacomo.travaglini@arm.com>
Maintainer: Giacomo Travaglini <giacomo.travaglini@arm.com>
Tested-by: kokoro <noreply+kokoro@google.com>

cpu-o3: Avoid passing ReExec 'faults' on CPU tracing interface

The O3 model uses ReExec faults to flush the pipeline and restart
after a memory ordering violation, e.g. due to an incoming snoop.

These, just like branch mispredict flushes, are not architectural
faults but micro-architectural events, and should therefore not
show up on the instruction tracing interface.

This adds a check on faulting instructions in commit, to verify
if the instruction faulted due to ReExec, to avoid tracing it.

Change-Id: I1d3eaffb0ff22411e0e16a69ef07961924c88c10
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/30554
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Jason Lowe-Power <power.jg@gmail.com>
Maintainer: Jason Lowe-Power <power.jg@gmail.com>

arch-arm: Fix routeToHyp conditions for Excp Type

Change-Id: I8eadd8e1f8c53d5e61969b492d9f2cbd12110188
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/30620
Reviewed-by: Giacomo Travaglini <giacomo.travaglini@arm.com>
Maintainer: Giacomo Travaglini <giacomo.travaglini@arm.com>
Tested-by: kokoro <noreply+kokoro@google.com>

x86: Fix the indentation in arch/x86/stacktrace.hh.

The namespace shouldn't be indented.

Change-Id: I2bd5fa5adbbad62a35cfb54a9509c48d45076539
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/30959
Reviewed-by: Jason Lowe-Power <power.jg@gmail.com>
Maintainer: Jason Lowe-Power <power.jg@gmail.com>
Tested-by: kokoro <noreply+kokoro@google.com>

arch: Delete the unused ProcessInfo class.

Change-Id: Ie67f696005fa60e117e1e4e4e985aee5e767ccec
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/30958
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.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>

util: Delete util/batch, util/pbs, and util qdo.

The util/pbs directory has a set of python scripts which were written to
submit jobs to the PBS pool at the University of Michigan. They aren't
incredibly specialized for that environment, but they do have a little
bit of hard coding which, for instance, uses paths which are only
meaningful there.

The util/batch directory was added alongside a seemingly unrelated
change (perhaps by accident?) and is a slightly updated copy of util/pbs
which also (or instead?) supports OAR.

The qdo script seems to be a script for managing job queues on PBS
and/or OAR, and is also tuned to the UofM environment, for instance
insisting that a path starts with /n/poolfs so that files are available
on an NFS volume shared with the pool.

All three of these scripts could potentially be useful with modification
in a similar environment, but also all three are unmaintained. The
environment in UofM may no longer actually match the expectations of
these scripts, and even if it does/did, gem5 may no longer be 100%
compatible with them.

If these scripts sit in util not being used by anyone, they add clutter
and complexity without adding any value. If someone really needs to know
what was once in them, they can be recovered from revision control.

Change-Id: I0192bd119893f7a41fcb820f4cf408609b03cd27
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/30957
Reviewed-by: Steve Reinhardt <stever@gmail.com>
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.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>

sim: Add M5_VAR_USED to var used in dprint

Change-Id: I8f8654b8546ee8df3d4acd1ccbc5080ad38764c1
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/30896
Reviewed-by: Jason Lowe-Power <power.jg@gmail.com>
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
Maintainer: Gabe Black <gabeblack@google.com>
Maintainer: Andreas Sandberg <andreas.sandberg@arm.com>
Tested-by: kokoro <noreply+kokoro@google.com>