MMCR1 determines the events to be sampled by the PMU. Updating the
counters at every MMCR1 write ensures that we're not sampling more
or less events by looking only at MMCR0 and the PMCs.
It is worth noticing that both the Book3S PowerPC PMU, and this IBM
Power8+ PMU that we're modeling, also uses MMCRA, MMCR2 and MMCR3 to
control the PMU. These three registers aren't being handled in this
initial implementation, so for now we're controlling all the PMU
aspects using MMCR0, MMCR1 and the PMCs.
Reviewed-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Daniel Henrique Barboza <danielhb413@gmail.com>
Message-Id: <20211201151734.654994-5-danielhb413@gmail.com>
Signed-off-by: Cédric Le Goater <clg@kaod.org>
Calling pmu_update_cycles() on every PMC read/write operation ensures
that the values being fetched are up to date with the current PMU state.
In theory we can get away by just trapping PMCs reads, but we're going
to trap PMC writes to deal with counter overflow logic later on. Let's
put the required wiring for that and make our lives a bit easier in the
next patches.
Reviewed-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Daniel Henrique Barboza <danielhb413@gmail.com>
Message-Id: <20211201151734.654994-4-danielhb413@gmail.com>
Signed-off-by: Cédric Le Goater <clg@kaod.org>
This patch adds the barebones of the PMU logic by enabling cycle
counting. The overall logic goes as follows:
- MMCR0 reg initial value is set to 0x80000000 (MMCR0_FC set) to avoid
having to spin the PMU right at system init;
- to retrieve the events that are being profiled, pmc_get_event() will
check the current MMCR0 and MMCR1 value and return the appropriate
PMUEventType. For PMCs 1-4, event 0x2 is the implementation dependent
value of PMU_EVENT_INSTRUCTIONS and event 0x1E is the implementation
dependent value of PMU_EVENT_CYCLES. These events are supported by IBM
Power chips since Power8, at least, and the Linux Perf driver makes use
of these events until kernel v5.15. For PMC1, event 0xF0 is the
architected PowerISA event for cycles. Event 0xFE is the architected
PowerISA event for instructions;
- if the counter is frozen, either via the global MMCR0_FC bit or its
individual frozen counter bits, PMU_EVENT_INACTIVE is returned;
- pmu_update_cycles() will go through each counter and update the
values of all PMCs that are counting cycles. This function will be
called every time a MMCR0 update is done to keep counters values
up to date. Upcoming patches will use this function to allow the
counters to be properly updated during read/write of the PMCs
and MMCR1 writes.
Given that the base CPU frequency is fixed at 1Ghz for both powernv and
pseries clock, cycle calculation assumes that 1 nanosecond equals 1 CPU
cycle. Cycle value is then calculated by adding the elapsed time, in
nanoseconds, of the last cycle update done via pmu_update_cycles().
Reviewed-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Daniel Henrique Barboza <danielhb413@gmail.com>
Message-Id: <20211201151734.654994-3-danielhb413@gmail.com>
Signed-off-by: Cédric Le Goater <clg@kaod.org>
PPC instruction xsmaxcdp, xsmincdp, xsmaxjdp, and xsminjdp are using
vector registers when they should be using VSX ones. This happens
because the instructions are using GEN_VSX_HELPER_R3, which adds 32
to the register numbers, effectively making them vector registers.
This patch fixes it by changing these instructions to use
GEN_VSX_HELPER_X3.
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Victor Colombo <victor.colombo@eldorado.org.br>
Message-Id: <20211213120958.24443-2-victor.colombo@eldorado.org.br>
Signed-off-by: Cédric Le Goater <clg@kaod.org>
(Applies to 7441, 7445, 7450, 7451, 7455, 7457, 7447, 7447a and 7448)
The QEMU-side software TLB implementation for the 7450 family of CPUs
is being removed due to lack of known users in the real world. The
last users in the code were removed by the two previous commits.
A brief history:
The feature was added in QEMU by commit 7dbe11acd8 ("Handle all MMU
models in switches...") with the mention that Linux was not able to
handle the TLB miss interrupts and the MMU model would be kept
disabled.
At some point later, commit 8ca3f6c382 ("Allow selection of all
defined PowerPC 74xx (aka G4) CPUs.") enabled the model for the 7450
family without further justification.
We have since the year 2011 [1] been unable to run OpenBIOS in the
7450s and have not heard of any other software that is used with those
CPUs in QEMU. Attempts were made to find a guest OS that implemented
the TLB miss handlers and none were found among Linux 5.15, FreeBSD 13,
MacOS9, MacOSX and MorphOS 3.15.
All CPUs that registered this feature were moved to an MMU model that
replaces the software TLB with a QEMU hardware TLB
implementation. They can now run the same software as the 7400 CPUs,
including the OSes mentioned above.
References:
- https://bugs.launchpad.net/qemu/+bug/812398https://gitlab.com/qemu-project/qemu/-/issues/86
- https://lists.nongnu.org/archive/html/qemu-ppc/2021-11/msg00289.html
message id: 20211119134431.406753-1-farosas@linux.ibm.com
Signed-off-by: Fabiano Rosas <farosas@linux.ibm.com>
Reviewed-by: Cédric Le Goater <clg@kaod.org>
Message-Id: <20211130230123.781844-4-farosas@linux.ibm.com>
Signed-off-by: Cédric Le Goater <clg@kaod.org>
There is no double-rounding bug here, because the result is
merely an estimate to within 1 part in 32, but perform the
operation with float64r32_div for consistency.
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Message-Id: <20211119160502.17432-33-richard.henderson@linaro.org>
Signed-off-by: Cédric Le Goater <clg@kaod.org>
Use float64r32_mul. Fixes a double-rounding issue with performing
the compuation in float64 and then rounding afterward.
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Message-Id: <20211119160502.17432-32-richard.henderson@linaro.org>
Signed-off-by: Cédric Le Goater <clg@kaod.org>
Use float64r32_{add,sub,div}. Fixes a double-rounding issue with
performing the compuation in float64 and then rounding afterward.
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Message-Id: <20211119160502.17432-31-richard.henderson@linaro.org>
Signed-off-by: Cédric Le Goater <clg@kaod.org>
Use float64r32_sqrt. Fixes a double-rounding issue with performing
the compuation in float64 and then rounding afterward.
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Message-Id: <20211119160502.17432-30-richard.henderson@linaro.org>
Signed-off-by: Cédric Le Goater <clg@kaod.org>
Use float64r32_muladd. Fixes a double-rounding issue with performing
the compuation in float64 and then rounding afterward.
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Message-Id: <20211119160502.17432-29-richard.henderson@linaro.org>
Signed-off-by: Cédric Le Goater <clg@kaod.org>
mtfsf, mtfsfi and mtfsb1 instructions call helper_float_check_status
after updating the value of FPSCR, but helper_float_check_status
checks fp_status and fp_status isn't updated based on FPSCR and
since the value of fp_status is reset earlier in the instruction,
it's always 0.
Because of this helper_float_check_status would change the FI bit to 0
as this bit checks if the last operation was inexact and
float_flag_inexact is always 0.
These instructions also don't throw exceptions correctly since
helper_float_check_status throw exceptions based on fp_status.
This commit created a new helper, helper_fpscr_check_status that checks
FPSCR value instead of fp_status and checks for a larger variety of
exceptions than do_float_check_status.
Since fp_status isn't used, gen_reset_fpstatus() was removed.
The hardware used to compare QEMU's behavior to was a Power9.
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Lucas Mateus Castro (alqotel) <lucas.araujo@eldorado.org.br>
Message-Id: <20211201163808.440385-2-lucas.araujo@eldorado.org.br>
Signed-off-by: Cédric Le Goater <clg@kaod.org>
Implement the following PowerISA v3.1 instructions:
vextdubvlx: Vector Extract Double Unsigned Byte to VSR using
GPR-specified Left-Index
vextduhvlx: Vector Extract Double Unsigned Halfword to VSR using
GPR-specified Left-Index
vextduwvlx: Vector Extract Double Unsigned Word to VSR using
GPR-specified Left-Index
vextddvlx: Vector Extract Double Doubleword to VSR using
GPR-specified Left-Index
vextdubvrx: Vector Extract Double Unsigned Byte to VSR using
GPR-specified Right-Index
vextduhvrx: Vector Extract Double Unsigned Halfword to VSR using
GPR-specified Right-Index
vextduwvrx: Vector Extract Double Unsigned Word to VSR using
GPR-specified Right-Index
vextddvrx: Vector Extract Double Doubleword to VSR using
GPR-specified Right-Index
Suggested-by: Richard Henderson <richard.henderson@linaro.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Luis Pires <luis.pires@eldorado.org.br>
Signed-off-by: Matheus Ferst <matheus.ferst@eldorado.org.br>
Message-Id: <20211104123719.323713-10-matheus.ferst@eldorado.org.br>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Implements the following PowerISA v3.1 instructions:
vinsblx: Vector Insert Byte from GPR using GPR-specified Left-Index
vinshlx: Vector Insert Halfword from GPR using GPR-specified Left-Index
vinswlx: Vector Insert Word from GPR using GPR-specified Left-Index
vinsdlx: Vector Insert Doubleword from GPR using GPR-specified
Left-Index
vinsbrx: Vector Insert Byte from GPR using GPR-specified Right-Index
vinshrx: Vector Insert Halfword from GPR using GPR-specified
Right-Index
vinswrx: Vector Insert Word from GPR using GPR-specified Right-Index
vinsdrx: Vector Insert Doubleword from GPR using GPR-specified
Right-Index
The helpers and do_vinsx receive i64 to allow code sharing with the
future implementation of Vector Insert from VSR using GPR Index.
Signed-off-by: Matheus Ferst <matheus.ferst@eldorado.org.br>
Message-Id: <20211104123719.323713-6-matheus.ferst@eldorado.org.br>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
pdepd and pextd helpers are moved out of #ifdef (TARGET_PPC64) to allow
them to be reused as GVecGen3.fni8.
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Luis Pires <luis.pires@eldorado.org.br>
Signed-off-by: Matheus Ferst <matheus.ferst@eldorado.org.br>
Message-Id: <20211104123719.323713-4-matheus.ferst@eldorado.org.br>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
There's no reason to keep vector-impl.c.inc separate from
vmx-impl.c.inc. Additionally, let GVec handle the multiple calls to
helper_cfuged for us.
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Matheus Ferst <matheus.ferst@eldorado.org.br>
Message-Id: <20211104123719.323713-2-matheus.ferst@eldorado.org.br>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Move the following instructions to decodetree:
ddedpd: DFP Decode DPD To BCD
ddedpdq: DFP Decode DPD To BCD Quad
denbcd: DFP Encode BCD To DPD
denbcdq: DFP Encode BCD To DPD Quad
dscli: DFP Shift Significand Left Immediate
dscliq: DFP Shift Significand Left Immediate Quad
dscri: DFP Shift Significand Right Immediate
dscriq: DFP Shift Significand Right Immediate Quad
Also deleted dfp-ops.c.inc, now that all PPC DFP instructions were
moved to decodetree.
Signed-off-by: Luis Pires <luis.pires@eldorado.org.br>
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Message-Id: <20211029192417.400707-16-luis.pires@eldorado.org.br>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Move the following instructions to decodetree:
dctdp: DFP Convert To DFP Long
dctqpq: DFP Convert To DFP Extended
drsp: DFP Round To DFP Short
drdpq: DFP Round To DFP Long
dcffix: DFP Convert From Fixed
dcffixq: DFP Convert From Fixed Quad
dctfix: DFP Convert To Fixed
dctfixq: DFP Convert To Fixed Quad
dxex: DFP Extract Biased Exponent
dxexq: DFP Extract Biased Exponent Quad
Signed-off-by: Luis Pires <luis.pires@eldorado.org.br>
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Message-Id: <20211029192417.400707-15-luis.pires@eldorado.org.br>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Move the following instructions to decodetree:
dquai: DFP Quantize Immediate
dquaiq: DFP Quantize Immediate Quad
drintx: DFP Round to FP Integer With Inexact
drintxq: DFP Round to FP Integer With Inexact Quad
drintn: DFP Round to FP Integer Without Inexact
drintnq: DFP Round to FP Integer Without Inexact Quad
Signed-off-by: Luis Pires <luis.pires@eldorado.org.br>
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Message-Id: <20211029192417.400707-13-luis.pires@eldorado.org.br>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Move the following instructions to decodetree:
dtstdc: DFP Test Data Class
dtstdcq: DFP Test Data Class Quad
dtstdg: DFP Test Data Group
dtstdgq: DFP Test Data Group Quad
Signed-off-by: Luis Pires <luis.pires@eldorado.org.br>
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Message-Id: <20211029192417.400707-10-luis.pires@eldorado.org.br>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Implement the following PowerISA v3.1 instruction:
dctfixqq: DFP Convert To Fixed Quadword Quad
Signed-off-by: Luis Pires <luis.pires@eldorado.org.br>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Message-Id: <20211029192417.400707-8-luis.pires@eldorado.org.br>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Implement the following PowerISA v3.1 instruction:
dcffixqq: DFP Convert From Fixed Quadword Quad
Signed-off-by: Luis Pires <luis.pires@eldorado.org.br>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Message-Id: <20211029192417.400707-5-luis.pires@eldorado.org.br>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
We will shortly be interested in distinguishing pointers
from integers in the helper's declaration, as well as a
true void return. We currently have two parallel 1 bit
fields; merge them and expand to a 3 bit field.
Our current maximum is 7 helper arguments, plus the return
makes 8 * 3 = 24 bits used within the uint32_t typemask.
Tested-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Perform the test against FSCR_SCV at runtime, in the helper.
This means we can remove the incorrect set against SCV in
ppc_tr_init_disas_context and do not need to add an HFLAGS bit.
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Message-Id: <20210323184340.619757-6-richard.henderson@linaro.org>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
vmulhsd: Vector Multiply High Signed Doubleword
vmulhud: Vector Multiply High Unsigned Doubleword
Signed-off-by: Lijun Pan <ljp@linux.ibm.com>
Message-Id: <20200724045845.89976-5-ljp@linux.ibm.com>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
vmulhsw: Vector Multiply High Signed Word
vmulhuw: Vector Multiply High Unsigned Word
Signed-off-by: Lijun Pan <ljp@linux.ibm.com>
Message-Id: <20200724045845.89976-4-ljp@linux.ibm.com>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Convert the original implementation of vmuluwm to the more generic
tcg_gen_gvec_mul.
Signed-off-by: Lijun Pan <ljp@linux.ibm.com>
Message-Id: <20200701234344.91843-5-ljp@linux.ibm.com>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
POWER9 adds scv and rfscv instructions and the system call vectored
interrupt. Linux does not support this instruction yet but it has
been tested with a modified kernel that runs on real hardware.
Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Message-Id: <20200507115328.789175-1-npiggin@gmail.com>
[dwg: Corrected an overlong line]
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
The new ISA v3.0 slbia variants have not been implemented for TCG,
which can lead to crashing when a POWER9 machine boots Linux using
the hash MMU, for example ("disable_radix" kernel command line).
Add them.
Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Message-Id: <20200319064439.1020571-1-npiggin@gmail.com>
Reviewed-by: Cédric Le Goater <clg@kaod.org>
[dwg: Fixed compile error for USER_ONLY builds]
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
The Processor Control facility for POWER8 processors and later
provides a mechanism for the hypervisor to send messages to other
threads in the system (msgsnd instruction) and cause hypervisor-level
exceptions. Privileged non-hypervisor programs can also send messages
(msgsndp instruction) but are restricted to the threads of the same
subprocessor and cause privileged-level exceptions.
The Directed Privileged Doorbell Exception State (DPDES) register
reflects the state of pending privileged doorbell exceptions and can
be used to modify that state. The register can be used to read and
modify the state of privileged doorbell exceptions for all threads of
a subprocessor and thus is a shared facility for that subprocessor.
The register can be read/written by the hypervisor and read by the
supervisor if enabled in the HFSCR, otherwise a hypervisor facility
unavailable exception is generated.
The privileged message send and clear instructions (msgsndp & msgclrp)
are used to generate and clear the presence of a directed privileged
doorbell exception, respectively. The msgsndp instruction can be used
to target any thread of the current subprocessor, msgclrp acts on the
thread issuing the instruction. These instructions are privileged, but
will generate a hypervisor facility unavailable exception if not
enabled in the HFSCR and executed in privileged non-hypervisor
state. The HV facility unavailable exception will be addressed in
other patch.
Add and implement this register and instructions by reading or
modifying the pending interrupt state of the cpu.
Note that TCG only supports one thread per core and so we only need to
worry about the cpu making the access.
Signed-off-by: Suraj Jitindar Singh <sjitindarsingh@gmail.com>
Signed-off-by: Cédric Le Goater <clg@kaod.org>
Message-Id: <20200120104935.24449-2-clg@kaod.org>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
The spr TBU40 is used to set the upper 40 bits of the timebase
register, present on POWER5+ and later processors.
This register can only be written by the hypervisor, and cannot be read.
Signed-off-by: Suraj Jitindar Singh <sjitindarsingh@gmail.com>
Reviewed-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Cédric Le Goater <clg@kaod.org>
Message-Id: <20191128134700.16091-5-clg@kaod.org>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
The Processor Utilisation of Resources Register (PURR) and Scaled
Processor Utilisation of Resources Register (SPURR) provide an estimate
of the resources used by the thread, present on POWER7 and later
processors.
Currently the [S]PURR registers simply count at the rate of the
timebase.
Preserve this behaviour but rework the implementation to store an offset
like the timebase rather than doing the calculation manually. Also allow
hypervisor write access to the register along with the currently
available read access.
Signed-off-by: Suraj Jitindar Singh <sjitindarsingh@gmail.com>
Reviewed-by: David Gibson <david@gibson.dropbear.id.au>
[ clg: rebased on current ppc tree ]
Signed-off-by: Cédric Le Goater <clg@kaod.org>
Message-Id: <20191128134700.16091-3-clg@kaod.org>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
The virtual timebase register (VTB) is a 64-bit register which
increments at the same rate as the timebase register, present on POWER8
and later processors.
The register is able to be read/written by the hypervisor and read by
the supervisor. All other accesses are illegal.
Currently the VTB is just an alias for the timebase (TB) register.
Implement the VTB so that is can be read/written independent of the TB.
Make use of the existing method for accessing timebase facilities where
by the compensation is stored and used to compute the value on reads/is
updated on writes.
Signed-off-by: Suraj Jitindar Singh <sjitindarsingh@gmail.com>
[ clg: rebased on current ppc tree ]
Signed-off-by: Cédric Le Goater <clg@kaod.org>
Message-Id: <20191128134700.16091-2-clg@kaod.org>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Since commit ef96e3ae96 "target/ppc: move FP and VMX registers into aligned vsr
register array" FP registers are no longer stored consecutively in memory and so
the current method of combining FP register pairs into DFP numbers is incorrect.
Firstly update the definition of the dh_*_fprp defines in helper.h to reflect
that FP registers are now stored as part of an array of ppc_vsr_t elements
rather than plain uint64_t elements, and then introduce a new ppc_fprp_t type
which conceptually represents a DFP even-odd register pair to be consumed by the
DFP helper functions.
Finally update the new DFP {get,set}_dfp{64,128}() helper functions to convert
between DFP numbers and DFP even-odd register pairs correctly, making use of the
existing VsrD() macro to access the correct elements regardless of host endian.
Fixes: ef96e3ae96 "target/ppc: move FP and VMX registers into aligned vsr register array"
Signed-off-by: Mark Cave-Ayland <mark.cave-ayland@ilande.co.uk>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Message-Id: <20190926185801.11176-4-mark.cave-ayland@ilande.co.uk>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Optimize Altivec instruction vclzw (Vector Count Leading Zeros Word).
This instruction counts the number of leading zeros of each word element
in source register and places result in the appropriate word element of
destination register.
Counting is to be performed in four iterations of for loop(one for each
word elemnt of source register vB). Every iteration consists of loading
appropriate word element from source register, counting leading zeros
with tcg_gen_clzi_i32, and saving the result in appropriate word element
of destination register.
Signed-off-by: Stefan Brankovic <stefan.brankovic@rt-rk.com>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Message-Id: <1563200574-11098-7-git-send-email-stefan.brankovic@rt-rk.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Optimize Altivec instruction vclzd (Vector Count Leading Zeros Doubleword).
This instruction counts the number of leading zeros of each doubleword element
in source register and places result in the appropriate doubleword element of
destination register.
Using tcg-s count leading zeros instruction two times(once for each
doubleword element of source register vB) and placing result in
appropriate doubleword element of destination register vD.
Signed-off-by: Stefan Brankovic <stefan.brankovic@rt-rk.com>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Message-Id: <1563200574-11098-6-git-send-email-stefan.brankovic@rt-rk.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Optimize altivec instruction vgbbd (Vector Gather Bits by Bytes by Doubleword)
All ith bits (i in range 1 to 8) of each byte of doubleword element in
source register are concatenated and placed into ith byte of appropriate
doubleword element in destination register.
Following solution is done for both doubleword elements of source register
in parallel, in order to reduce the number of instructions needed(that's why
arrays are used):
First, both doubleword elements of source register vB are placed in
appropriate element of array avr. Bits are gathered in 2x8 iterations(2 for
loops). In first iteration bit 1 of byte 1, bit 2 of byte 2,... bit 8 of
byte 8 are in their final spots so avr[i], i={0,1} can be and-ed with
tcg_mask. For every following iteration, both avr[i] and tcg_mask variables
have to be shifted right for 7 and 8 places, respectively, in order to get
bit 1 of byte 2, bit 2 of byte 3.. bit 7 of byte 8 in their final spots so
shifted avr values(saved in tmp) can be and-ed with new value of tcg_mask...
After first 8 iteration(first loop), all the first bits are in their final
places, all second bits but second bit from eight byte are in their places...
only 1 eight bit from eight byte is in it's place). In second loop we do all
operations symmetrically, in order to get other half of bits in their final
spots. Results for first and second doubleword elements are saved in
result[0] and result[1] respectively. In the end those results are saved in
appropriate doubleword element of destination register vD.
Signed-off-by: Stefan Brankovic <stefan.brankovic@rt-rk.com>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Message-Id: <1563200574-11098-5-git-send-email-stefan.brankovic@rt-rk.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Optimization of altivec instructions vsl and vsr(Vector Shift Left/Rigt).
Perform shift operation (left and right respectively) on 128 bit value of
register vA by value specified in bits 125-127 of register vB. Lowest 3
bits in each byte element of register vB must be identical or result is
undefined.
For vsl instruction, the first step is bits 125-127 of register vB have
to be saved in variable sh. Then, the highest sh bits of the lower
doubleword element of register vA are saved in variable shifted,
in order not to lose those bits when shift operation is performed on
the lower doubleword element of register vA, which is the next
step. After shifting the lower doubleword element shift operation
is performed on higher doubleword element of vA, with replacement of
the lowest sh bits(that are now 0) with bits saved in shifted.
For vsr instruction, firstly, the bits 125-127 of register vB have
to be saved in variable sh. Then, the lowest sh bits of the higher
doubleword element of register vA are saved in variable shifted,
in odred not to lose those bits when the shift operation is
performed on the higher doubleword element of register vA, which is
the next step. After shifting higher doubleword element, shift operation
is performed on lower doubleword element of vA, with replacement of
highest sh bits(that are now 0) with bits saved in shifted.
Signed-off-by: Stefan Brankovic <stefan.brankovic@rt-rk.com>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Message-Id: <1563200574-11098-3-git-send-email-stefan.brankovic@rt-rk.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Adding simple macro that is calling tcg implementation of appropriate
instruction if altivec support is active.
Optimization of altivec instruction lvsl (Load Vector for Shift Left).
Place bytes sh:sh+15 of value 0x00 || 0x01 || 0x02 || ... || 0x1E || 0x1F
in destination register. Sh is calculated by adding 2 source registers and
getting bits 60-63 of result.
First, the bits [28-31] are placed from EA to variable sh. After that,
the bytes are created in the following way:
sh:(sh+7) of X(from description) by multiplying sh with 0x0101010101010101
followed by addition of the result with 0x0001020304050607. Value obtained
is placed in higher doubleword element of vD.
(sh+8):(sh+15) by adding the result of previous multiplication with
0x08090a0b0c0d0e0f. Value obtained is placed in lower doubleword element
of vD.
Optimization of altivec instruction lvsr (Load Vector for Shift Right).
Place bytes 16-sh:31-sh of value 0x00 || 0x01 || 0x02 || ... || 0x1E ||
0x1F in destination register. Sh is calculated by adding 2 source
registers and getting bits 60-63 of result.
First, the bits [28-31] are placed from EA to variable sh. After that,
the bytes are created in the following way:
sh:(sh+7) of X(from description) by multiplying sh with 0x0101010101010101
followed by substraction of the result from 0x1011121314151617. Value
obtained is placed in higher doubleword element of vD.
(sh+8):(sh+15) by substracting the result of previous multiplication from
0x18191a1b1c1d1e1f. Value obtained is placed in lower doubleword element
of vD.
Signed-off-by: Stefan Brankovic <stefan.brankovic@rt-rk.com>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Message-Id: <1563200574-11098-2-git-send-email-stefan.brankovic@rt-rk.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Introduce a new GEN_VSX_HELPER_VSX_MADD macro for the generator function which
enables the source and destination registers to be decoded at translation time.
This enables the determination of a or m form to be made at translation time so
that a single helper function can now be used for both variants.
Signed-off-by: Mark Cave-Ayland <mark.cave-ayland@ilande.co.uk>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Message-Id: <20190616123751.781-16-mark.cave-ayland@ilande.co.uk>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Daniel Henrique Barboza
Matheus Ferst
Victor Colombo
Fabiano Rosas
Richard Henderson
Lucas Mateus Castro (alqotel)
Luis Pires
Lijun Pan
Nicholas Piggin
Cédric Le Goater
Suraj Jitindar Singh
Mark Cave-Ayland
Stefan Brankovic