Now that we are translating the fetch effective address to real one
cycle earlier, we can use the real address to index the icache array.
This has the benefit that the set size can be larger than a page,
enabling us to configure the icache to be larger without having to
increase its associativity. Previously the set size was limited to
the page size to avoid aliasing problems. Thus for example a 32kB
icache would need to be 8-way associative, resulting in large numbers
of LUTs being used for tag comparisons in FPGA implementations, and
poor timing. With this change, a 32kB icache can be 1 or 2-way
associative, which means deeper and narrower tag and data RAMs and
fewer tag comparators.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
This moves the address translation step for instruction fetches one
cycle earlier, so that it now happens in the fetch1 stage. There is
now a 2-entry mini translation cache ("ERAT", or effective to real
address translation cache) which operates on the output of the
multiplexer that selects the instruction address for the next cycle.
The ERAT consists of two effective address registers and two
corresponding real address registers. They store the page number part
of the addresses for a 4kB page size, which is the smallest page size
supported by the architecture.
If the effective address doesn't match either of the EA registers, and
address translation is enabled, then i_out.req goes low for two cycles
while the iTLB is looked up. Experimentally, this delay results in a
0.1% drop in coremark performance; allowing two cycles for the lookup
results in better timing. The result from the iTLB is placed into the
least recently used ERAT entry and then used to translate the address
as normal. If address translation is not enabled then the EA is used
directly as the real address.
The iTLB structure is the same as it was before; direct mapped,
indexed using a hashed EA.
The "fetch failed" signal, which indicates a TLB miss or protection
violation, is now generated in fetch1 and passed through icache.
When it is asserted, fetch1 goes into a stalled state until a PTE
arrives from the MMU (which gets put into both the iTLB and the ERAT),
or an interrupt or redirect occurs.
Any TLB invalidations from the MMU invalidate the whole ERAT.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
This reduces the number of possible sources for the next NIA from 4
down to 3, by routing interrupt vector addresses through the
r_int.next_nia register, as is already done for reset. This adds one
extra cycle of latency when taking interrupts. During this extra cycle,
i_out.req is 0.
Writeback now no longer combines redirects (branches, rfid, isync)
with interrupts; they are presented separately to fetch1.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
This adds a next_nia field to the Fetch1ToIcacheType record, which
provides an indication of what will be in the nia field on the next
non-stalled cycle. This is intended to be as fast as possible, being
a selection from two redirect addresses (from writeback and decode1)
or an internal register (r_int.next_nia). Reset addresses and
predicted branch targets come through this internal register.
The rearrangement here has the side effect that we can now use the BTC
on the first instruction after a taken branch, whereas previously the
BTC was only active starting with the second instruction after a taken
branch. This provides a slight improvement in performance.
This also fixes a buglet in icache where it would assert its stall
output when i_in.req was false.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
This gets rid of the adder in writeback that computes redirect_nia.
Instead, the main adder in the ALU is used to compute the branch
target for relative branches. We now decode b and bc differently
depending on the AA field, generating INSN_brel, INSN_babs, INSN_bcrel
or INSN_bcabs as appropriate. Each one has a separate entry in the
decode table in decode1; the *rel versions use CIA as the A input.
The bclr/bcctr/bctar and rfid instructions now select ramspr_result
for the main result mux to get the redirect address into
ex1.e.write_data.
For branches which are predicted taken but not actually taken, we need
to redirect to the following instruction. We also need to do that for
isync. We do this in the execute2 stage since whether or not to do it
depends on the branch result. The next_nia computation is moved to
the execute2 stage and comes in via a new leg on the secondary result
multiplexer, making next_nia available ultimately in ex2.e.write_data.
This also means that the next_nia leg of the primary result
multiplexer is gone. Incrementing last_nia by 4 for sc (so that SRR0
points to the following instruction) is also moved to execute2.
Writing CIA+4 to LR was previously done through the main result
multiplexer. Now it comes in explicitly in the ramspr write logic.
Overall this removes the br_offset and abs_br fields and the logic to
add br_offset and next_nia, and one leg of the primary result
multiplexer, at the cost of a few extra control signals between
execute1 and execute2 and some multiplexing for the ramspr write side
and an extra input on the secondary result multiplexer.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
This arranges to generate an illegal instruction type program
interrupt for illegal prefixed instructions, that is, those where the
suffix is not a legal value given the prefix, or the prefix has a
reserved value in the subtype field. This implementation doesn't
generate an interrupt for the invalid 8LS:D and MLS:D instruction
forms where R = 1 and RA != 0. (In those cases it uses (RA) as the
addend, i.e. it ignores the R bit.)
This detects the case where the address of an instruction prefix is
equal mod 64 to 60, and generates an alignment interrupt in that case.
This also arranges to set bit 34 of SRR1 when an interrupt occurs due
to a prefixed instruction, for those interrupts where that is required
(i.e. trace, alignment, floating-point unavailable, data storage, data
segment, and most cases of program interrupt).
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
This adds logic to do basic decoding of the prefixed instructions
defined in PowerISA v3.1B which are in the SFFS (Scalar Fixed plus
Floating-Point Subset) compliancy subset. In PowerISA v3.1B SFFS,
there are 14 prefixed load/store instructions plus the prefixed no-op
instruction (pnop). The prefixed load/store instructions all use an
extended version of D-form, which has an extra 18 bits of displacement
in the prefix, plus an 'R' bit which enables PC-relative addressing.
When decode1 sees an instruction word where the insn_code is
INSN_prefix (i.e. the primary opcode was 1), it stores the prefix word
and sends nothing down to decode2 in that cycle. When the next valid
instruction word arrives, it is interpreted as a suffix, meaning that
its insn_code gets modified before being used to look up the decode
table.
The insn_code values are rearranged so that the values for
instructions which are the suffix of a valid prefixed instruction are
all at even indexes, and the corresponding prefixed instructions
follow immediately, so that an insn_code value can be converted to the
corresponding prefixed value by setting the LSB of the insn_code
value. There are two prefixed instructions, pld and pstd, for which
the suffix is not a valid SFFS instruction by itself, so these have
been given dummy insn_code values which decode as illegal (INSN_op57
and INSN_op61).
For a prefixed instruction, decode1 examines the type and subtype
fields of the prefix and checks that the suffix is valid for the type
and subtype. This check doesn't affect which entry of the decode
table is used; the result is passed down to decode2, and will in
future be acted upon in execute1.
The instruction address passed down to decode2 is the address of the
prefix. To enable this, part of the instruction address is saved when
the prefix is seen, and then the instruction address received from
icache is partly overlaid by the saved prefix address. Because
prefixed instructions are not permitted to cross 64-byte boundaries,
we only need to save bits 5:2 of the instruction to do this. If the
alignment restriction ever gets relaxed, we will then need to save
more bits of the address.
Decode2 has been extended to handle the R bit of the prefix (in 8LS
and MLS forms) and to be able to generate the 34-bit immediate value
from the prefix and suffix.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
This moves the insn_code values for mcrfs, mtfsb0/1 and mtfsfi into
the region used for floating-point instructions. This means that in
no-FPU implementations, they will get turned into illegal instructions
in predecode. We then don't need the code in execute1 that makes FP
instructions illegal in no-FPU implementations.
We also remove the NONE value for unit_t, since it was only ever used
with insn_type = OP_ILLEGAL, and the check for unit = NONE was
redundant with the check for insn_type = OP_ILLEGAL. Thus the check
for unit = NONE is no longer needed and is removed here.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
This adds an 'is_signed' signal to MultiplyInputType to indicate
whether the data1 and data2 fields are to be interpreted as signed or
unsigned numbers.
The 'not_result' field is replaced by a 'subtract' field which
provides a more intuitive interface for requesting that the product be
subtracted from the addend rather than added, i.e. subtract = 1 gives
C - A * B, vs. subtract = 0 giving C + A * B. (Previously the users
of the multipliers got the same effect by complementing the addend and
setting not_result = 1.)
The is_32bit field is removed because it is no longer used now that we
have a separate 32-bit multiplier.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
This splits out the decoding done in the decode0 step into a separate
predecoder, used when writing instructions into the icache. The
icache now holds 36 bits per instruction rather than 32. For valid
instructions, those 36 bits comprise the bottom 26 bits of the
instruction word, a 9-bit insn_code value (which uniquely identifies
the instruction), and a zero in the MSB. For illegal instructions,
the MSB is one and the full instruction word is in the bottom 32 bits.
Having the full instruction word available for illegal instructions
means that it can be printed in the log when simulating, or in future
could be placed in the HEIR register.
If we don't have an FPU, then the floating-point instructions are
regarded as illegal. In that case, the insn_code values would fit
into 8 bits, which could be used in future to reduce the size of
decode_rom from 512 to 256 entries.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
This reduces the block RAM requirements for instruction decoding by
splitting it into two steps. The first, in a new pipeline stage
called decode0 (implemented by code in decode1.vhdl) maps the
instruction to a 9-bit instruction code using major and row decode
ROMs. The second maps the 9-bit code to the final decode_rom_t (about
44 bits wide). Branch prediction done in decode is now done in
decode0 rather than decode1.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
Do more decoding of the instruction ahead of the IDLE state
processing so that the IDLE state code becomes much simpler.
To make the decoding easier, we now use four insn_type_t codes for
floating-point operations rather than two. This also rearranges the
insn_type_t values a little to get the 4 FP opcode values to differ
only in the bottom 2 bits, and put OP_DIV, OP_DIVE and OP_MOD next to
them.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
Instead of having a multiplexer in loadstore1 in order to be able to
put the instruction address into v.addr, we now set decode.input_reg_a
to CIA in the decode table entry for OP_FETCH_FAILED. That means that
the operand selection machinery in decode2 will supply the instruction
address to loadstore1 on the lv.addr1 input and no special case is
needed in loadstore1. This saves a few LUTs (~40 on the Artix-7).
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
This removes some logic that was previously added for the 16-byte
loads and stores (lq, lqarx, stq, stqcx.) and not completely removed
in commit c9e838b656 ("Remove support for lq, stq, lqarx and
stqcx.", 2022-06-04).
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
This improves timing a little because the register addresses now come
directly from a latch instead of being calculated by
decode_input_reg_*. The asserts that check that the two are the same
are now in decode2 rather than register_file.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
With this, the register RAM is read synchronously using the addresses
supplied by decode1. That means the register RAM can now be block RAM
rather than LUT RAM.
Debug accesses are done via the B port on cycles when decode1
indicates that there is no valid instruction or the instruction
doesn't use a [F]RB operand.
We latch the addresses being read in each cycle and use the same
address next cycle if stalled. Data that is being written is latched
and a multiplexer on each read port then supplies the latched write
data if the read address for that port equals the write address.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
This adds some relatively simple logic to decode1 to compute the
GPR/FPR addresses that an instruction will access. It always computes
three addresses regardless of whether the instruction will actually
use all of them. The main things it computes are whether the
instruction uses the RS field or the RC field for the 3rd operand, and
whether the operands are FPRs or GPRs (it is possible for RS to be an
FPR but RA and RB to be GPRs, as for example with stfdx).
At the moment all we do with these computed register addresses is to
assert that they are identical to the ones coming from decode2 one
cycle later.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
This provides access to the SPRs via the JTAG DMI interface. For now
they are still accessed as if they were GPR/FPRs using the same
numbering as before (GPRs at 0 - 0x1f, SPRs at 0x20 - 0x2d, FPRs at
0x40 - 0x5f).
For XER, debug reads now report the full value, not just the bits that
were previously stored in the register file. The "slow" SPR mux is
not used for debug reads.
Decode2 determines on each cycle whether a debug SPR access will
happen next cycle, based on whether there is a request and whether the
current instruction accesses the SPR RAM.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
With this, the register file now contains 64 entries, for 32 GPRs and
32 FPRs, rather than the 128 it had previously. Several things get
simplified - decode1 no longer has to work out the ispr{1,2,o} values,
decode_input_reg_{a,b,c} no longer have the t = SPR case, etc.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
By putting CTR on the odd side and LR and TAR on the even side, we can
read and write CTR for bdnz-style instructions in parallel with
reading LR or TAR for indirect branches and writing LR for branches
with LK=1. Thus we don't need to double up any of these instructions,
giving a simplification in decode2.
We now have logic for printing LR and CTR at the end of a simulation
in execute1, in addition to the similar logic in register_file and
cr_file.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
This starts the process of removing SPRs from the register file by
moving SRR0/1, SPRG0-3, HSRR0/1 and HSPRG0/1 out of the register file
and putting them into execute1. They are stored in a pair of small
RAM arrays, referred to as "even" and "odd". The reason for having
two arrays is so that two values can be read and written in each
cycle. For example, SRR0 and SRR1 can be written in parallel by an
interrupt and read in parallel by the rfid instruction.
The addresses in the RAM which will be accessed are determined in the
decode2 stage. We have one write address for both sides, but two read
addresses, since in future we will want to be able to read CTR at the
same time as either LR or TAR.
We now have a connection from writeback to execute1 which carries the
partial SRR1 value for an interrupt. SRR0 comes from the execute
pipeline; we no longer need to carry instruction addresses along the
LSU and FPU pipelines. Since SRR0 and SRR1 can be written in the same
cycle now, we don't need the little state machine in writeback any
more.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
- Arrange for XER to be written for OE=1 forms
- Arrange for condition codes to be set for RC=1 forms
(including correct handling for 32-bit mode)
- Don't instantiate the divider if we have an FPU.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
This adds logic to the FPU to accomplish 64-bit integer divisions.
No instruction actually uses this yet.
The algorithm used is to obtain an estimate of the reciprocal of the
divisor using the lookup table and refine it by one to three
iterations of the Newton-Raphson algorithm (the number of iterations
depends on the number of significant bits in the dividend). Then the
reciprocal is multiplied by the dividend to get the quotient estimate.
The remainder is calculated as dividend - quotient * divisor. If the
remainder is greater than or equal to the divisor, the quotient is
incremented, or if a modulo operation is being done, the divisor is
subtracted from the remainder. The inverse estimate after refinement
is good enough that the quotient estimate is always equal to or one
less than the true quotient.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
At present the busy/stall signal going to decode1 depends on whether
control thinks it can issue the current instruction, and that depends
on completion and bypass signals coming from execute1 and writeback.
To improve the timing of stall_out, this rearranges decode2 so that
stall_out is asserted when we have a valid instruction that couldn't
be issued in the previous cycle. This means that decode1 could give
us a new instruction when we haven't issued the previous instruction.
This in turn means that we can only use d_in in the first cycle of
processing an instruction. After the first cycle, we get register
addresses etc. from dc2 rather than d_in.
Then, to avoid the need to read register operands from register_file
in each cycle until the instruction issues, we bring the bypass path
for data being written to the register file into decode2 explicitly
rather than having it in register_file.
A new process called decode2_addrs does the process of calling
decode_input_reg_* and decode_output_reg and sets up the register file
addresses. This was split out (and decode_input_reg_* reworked) to
try to reduce the number of passes through the decode2_1 process that
need to be done in simulation.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
This makes the FPU able to stall other units at execute stage 2 and be
stalled by other units (specifically the LSU).
This means that the completion and writeback for an instruction can
now end up being deferred until the second cycle of a following
instruction, i.e. the cycle when the state machine has gone through
IDLE state into one of the DO_* states, which means we need to latch
the destination FPR number, CR mask, etc. from the previous
instruction so that we present the correct information to writeback.
The advantage of this is that we can get rid of the in_progress signal
from the LSU.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
Execute1 and loadstore1 now send each other stall signals that
indicate that a valid instruction in stage 2 can't complete in this
cycle, and hence any valid instruction in stage 1 in the other unit
can't move to stage 2. With this in place, an ALU instruction can
move into stage 1 while a LSU instruction is in stage 2.
Since the FPU doesn't yet have a way to stall completion, we can't yet
start FPU instructions while any LSU or ALU instruction is in
progress.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
This adds a second execute stage to the pipeline, in order to match up
the length of the pipeline through loadstore and dcache with the
length through execute1. This will ultimately enable us to get rid of
the 1-cycle bubble that we currently have when issuing ALU
instructions after one or more LSU instructions.
Most ALU instructions execute in the first stage, except for
count-zeroes and popcount instructions (which take two cycles and do
some of their work in the second stage) and mfspr/mtspr to "slow" SPRs
(TB, DEC, PVR, LOGA/LOGD, CFAR). Multiply and divide/mod instructions
take several cycles but the instruction stays in the first stage (ex1)
and ex1.busy is asserted until the operation is complete.
There is currently a bypass from the first stage but not the second
stage. Performance is down somewhat because of that and because this
doesn't yet eliminate the bubble between LSU and ALU instructions.
The forwarding of XER common bits has been changed somewhat because
now there is another pipeline stage between ex1 and the committed
state in cr_file. The simplest thing for now is to record the last
value written and use that, unless there has been a flush, in which
case the committed state (obtained via e_in.xerc) is used.
Note that this fixes what was previously a benign bug in control.vhdl,
where it was possible for control to forget an instructions dependency
on a value from a previous instruction (a GPR or the CR) if this
instruction writes the value and the instruction gets to the point
where it could issue but is blocked by the busy signal from execute1.
In that situation, control may incorrectly not indicate that a bypass
should be used. That didn't matter previously because, for ALU and
FPU instructions, there was only one previous instruction in flight
and once the current instruction could issue, the previous instruction
was completing and the correct value would be obtained from
register_file or cr_file. For loadstore instructions there could be
two being executed, but because there are no bypass paths, failing to
indicate use of a bypass path is fine.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
We now have a record that represents the actions taken in executing an
instruction, and a process that computes that for the incoming
instruction. We no longer have 'current' or 'r.cur_instr', instead
things like the destination register are put into r.e in the first
cycle of an instruction and not reinitialized in subsequent busy
cycles.
For mfspr and mtspr, we now decode "slow" SPR numbers (those SPRs that
are not stored in the register file) to a new "spr_selector" record
in decode1 (excluding those in the loadstore unit). With this, the
result for mfspr is determined in the data path.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
Besides the overflow and status carry bits, XER has 18 bits which need
to retain the value written by mtxer (in case software wants to
emulate the move-assist instructions (lswi, lswx, stswi, stswx).
Until now these bits (and others) have been stored in the GPR file as
a "fast" SPR, but this causes complications because XER is not really
a fast SPR.
Instead, we now store these 18 bits in the 'ctrl' signal, which exists
in execute1. This will enable us to simplify the data path in future,
and has the added bonus that with a little bit of plumbing, we can get
the full XER value printed when dumping registers at the end of a
simulation.
Therefore this changes scripts/run_test.sh to remove the greps which
exclude XER from the comparison of actual and expected register
results.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
This moves REAL_ADDR_BITS out of the caches and defines a real_addr_t
type for a real address, along with a addr_to_real() conversion helper.
It makes the vhdl a bit more readable
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
This adds a bit to the BTC to store whether the corresponding branch
instruction was taken last time it was encountered. That lets us pass
a not-taken prediction down to decode1, which for backwards direct
branches inhibits it from redirecting fetch to the target of the
branch. This increases coremark by about 2%.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
This implements most of the architected PMU events. The ones missing
are mostly the ones that depend on which level of the cache hierarchy
data is fetched from. The events implemented here, and their raw
event codes, are:
Floating-point operation completed (100f4)
Load completed (100fc)
Store completed (200f0)
Icache miss (200fc)
ITLB miss (100f6)
ITLB miss resolved (400fc)
Dcache load miss (400f0)
Dcache load miss resolved (300f8)
Dcache store miss (300f0)
DTLB miss (300fc)
DTLB miss resolved (200f6)
No instruction available and none being executed (100f8)
Instruction dispatched (200f2, 300f2, 400f2)
Taken branch instruction completed (200fa)
Branch mispredicted (400f6)
External interrupt taken (200f8)
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
This is the start of an implementation of a PMU according to PowerISA
v3.0B. Things not implemented yet include most architected events,
the BHRB, event-based branches, thresholding, MMCR0[TBCC] field, etc.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
At present the logic prevents any interrupts from being handled while
there is a load/store instruction (one that has unit=LDST) being
executed. However, load/store instructions can still get sent to
loadstore1. Thus an instruction which should generate an interrupt
such as a floating-point unavailable interrupt will instead get
executed.
To fix this, when we detect that an interrupt should be generated but
loadstore1 is still executing a previous instruction, we don't execute
any new instructions, and set a new r.intr_pending flag. That results
in busy_out being asserted (meaning that no further instructions will
come in from decode2). When loadstore1 has finished the instructions
it has, the interrupt gets sent to writeback. If one of the
instructions in loadstore1 generates an interrupt in the meantime, the
l_in.interrupt signal gets asserted and that clears r.intr_pending, so
the interrupt we detected gets discarded.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
This implements a 1-entry partition table, so that instead of getting
the process table base address from the PRTBL SPR, the MMU now reads
the doubleword pointed to by the PTCR register plus 8 to get the
process table base address. The partition table entry is cached.
Having the PTCR and the vestigial partition table reduces the amount
of software change required in Linux for Microwatt support.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
The idea here is that we can have multiple instructions in progress at
the same time as long as they all go to the same unit, because that
unit will keep them in order. If we get an instruction for a
different unit, we wait for all the previous instructions to finish
before executing it. Since the loadstore unit is the only one that is
currently pipelined, this boils down to saying that loadstore
instructions can go ahead while l_in.in_progress = 1 but other
instructions have to wait until it is 0.
This gives a 2% increase on coremark performance on the Arty A7-100
(from ~190 to ~194).
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
This fixes two bugs which show up when multiple operations are in
flight in the dcache, and adds a 'hold' input which will be needed
when loadstore1 is pipelined.
The first bug is that dcache needs to sample the data for a store on
the cycle after the store request comes in even if the store request
is held up because of a previous request (e.g. if the previous request
is a load miss or a dcbz).
The second bug is that a load request coming in for a cache line being
refilled needs to be handled immediately in the case where it is for
the row whose data arrives on the same cycle. If it is not, then it
will be handled as a separate cache miss and the cache line will be
refilled again into a different way, leading to two ways both being
valid for the same tag. This can lead to data corruption, in the
scenario where subsequent writes go to one of the ways and then that
way gets displaced but the other way doesn't. This bug could in
principle show up even without having multiple operations in flight in
the dcache.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
This moves the logic for redirecting fetching and writing SRR0 and
SRR1 to writeback. The aim is that ultimately units other than
execute1 can send their interrupts to writeback along with their
instruction completions, so that there can be multiple instructions
in flight without needing execute1 to keep track of the address
of each outstanding instruction.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
This changes the bypass path. Previously it went from after
execute1's output to after decode2's output. Now it goes from before
execute1's output register to before decode2's output register. The
reason is that the new path will be simpler to manage when there are
possibly multiple instructions in flight. This means that the
bypassing can be managed inside decode2 and control.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
This changes the way GPR hazards are detected and tracked. Instead of
having a model of the pipeline in gpr_hazard.vhdl, which has to mirror
the behaviour of the real pipeline exactly, we now assign a 2-bit tag
to each instruction and record which GSPR the instruction writes.
Subsequent instructions that need to use the GSPR get the tag number
and stall until the value with that tag is being written back to the
register file.
For now, the forwarding paths are disabled. That gives about a 8%
reduction in coremark performance.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
This uses the instruction doubling machinery to convert conditional
branch instructions that update both CTR and LR (e.g., bdnzl, bdnzlrl)
into two instructions, of which the first updates CTR and determines
whether the branch is taken, and the second updates LR and does the
redirect if necessary.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
This uses the instruction-doubling machinery to send load with update
instructions down to loadstore1 as two separate ops, rather than
one op with two destinations. This will help to simplify the value
tracking mechanisms.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
This implements a cache in fetch1, where each entry stores the address
of a simple branch instruction (b or bc) and the target of the branch.
When fetching sequentially, if the address being fetched matches the
cache entry, then fetching will be redirected to the branch target.
The cache has 1024 entries and is direct-mapped, i.e. indexed by bits
11..2 of the NIA.
The bus from execute1 now carries information about taken and
not-taken simple branches, which fetch1 uses to update the cache.
The cache entry is updated for both taken and not-taken branches, with
the valid bit being set if the branch was taken and cleared if the
branch was not taken.
If fetching is redirected to the branch target then that goes down the
pipe as a predicted-taken branch, and decode1 does not do any static
branch prediction. If fetching is not redirected, then the next
instruction goes down the pipe as normal and decode1 does its static
branch prediction.
In order to make timing, the lookup of the cache is pipelined, so on
each cycle the cache entry for the current NIA + 8 is read. This
means that after a redirect (from decode1 or execute1), only the third
and subsequent sequentially-fetched instructions will be able to be
predicted.
This improves the coremark value on the Arty A7-100 from about 180 to
about 190 (more than 5%).
The BTC is optional. Builds for the Artix 7 35-T part have it off by
default because the extra ~1420 LUTs it takes mean that the design
doesn't fit on the Arty A7-35 board.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
Anton Blanchard