We should more consistently use inline assembly using the LIBC wrappers.
It's much safer to mark all of these volatile as well.
Reviewed By: lntue
Differential Revision: https://reviews.llvm.org/D152294
A previous patch added general support for printing via the RPC
interface. we should consolidate this functionality and get rid of the
old opcode that was used for simple testing.
Reviewed By: lntue
Differential Revision: https://reviews.llvm.org/D152211
This patch adds the initial support required to support basic priting in
`stdio.h` via `puts` and `fputs`. This is done using the existing LLVM C
library `File` API. In this sense we can think of the RPC interface as
our system call to dump the character string to the file. We carry a
`uintptr_t` reference as our native "file descriptor" as it will be used
as an opaque reference to the host's version once functions like
`fopen` are supported.
For some unknown reason the declaration of the `StdIn` variable causes
both the AMDGPU and NVPTX backends to crash if I use the `READ` flag.
This is not used currently as we only support output now, but it needs
to be fixed
Reviewed By: sivachandra, lntue
Differential Revision: https://reviews.llvm.org/D151282
This patch adds support for the `malloc` and `free` functions. These
currently aren't implemented in-tree so we first add the interface
filies.
This patch provides the most basic support for a true `malloc` and
`free` by using the RPC interface. This is functional, but in the future
we will want to implement a more intelligent system and primarily use
the RPC interface more as a `brk()` or `sbrk()` interface only called
when absolutely necessary. We will need to design an intelligent
allocator in the future.
The semantics of these memory allocations will need to be checked. I am
somewhat iffy on the details. I've heard that HSA can allocate
asynchronously which seems to work with my tests at least. CUDA uses an
implicit synchronization scheme so we need to use an explicitly separate
stream from the one launching the kernel or the default stream. I will
need to test the NVPTX case.
I would appreciate if anyone more experienced with the implementation details
here could chime in for the HSA and CUDA cases.
Reviewed By: sivachandra
Differential Revision: https://reviews.llvm.org/D151735
This is based on ideas from @nafi to:
- use a branchless version of 'cmp' for 'uint32_t',
- completely resolve the lexicographic comparison through vector
operations when wide types are available. We also get rid of byte
reloads and serializing '__builtin_ctzll'.
I did not include the suggestion to replace comparisons of 'uint16_t'
with two 'uint8_t' as it did not seem to help the codegen. This can
be revisited in sub-sequent patches.
The code been rewritten to reduce nested function calls, making the
job of the inliner easier and preventing harmful code duplication.
Reviewed By: nafi3000
Differential Revision: https://reviews.llvm.org/D148717
The C standard asserts that the `errno` value is an l-value thread local
integer. We cannot provide a generic thread local integer on the GPU
currently without some workarounds. Previously, we worked around this by
implementing the `errno` value as a special consumer class that made all
the writes disappear. However, this is problematic for internal tests.
Currently there are build failures because of this handling and it's
only likely to cause more problems the more we do this.
This patch instead makes the internal target used for testing export the
`errno` value as a simple global integer. This allows us to use and test
the `errno` interface correctly assuming we run with a single thread.
Because this is only used for the non-exported target we still do not
provide this feature in the version that users will use so we do not
need to worrk about it being incorrect in general.
Reviewed By: lntue
Differential Revision: https://reviews.llvm.org/D152015
This patch simply moves the special handling for `linux` files to a
subdirectory. This is done to make it easier in the future to extend
this support to targets (like the GPU) that will have different
dependencies.
Reviewed By: lntue
Differential Revision: https://reviews.llvm.org/D151231
It resolves to thread_local on all platform except for the GPUs on which
it resolves to nothing. The use of thread_local in the source code has been
replaced with the new macro.
Reviewed By: jhuber6
Differential Revision: https://reviews.llvm.org/D151486
This is largely a cosmetic change done with a few goals:
1. Reduce the conditionals in picking the correct set of tables for the
platform.
2. Avoid exposing, for example Linux errors, when building for non-Linux
platforms. This also prevents build failures when Linux errors are not
defined on the target non-Linux platform.
3. Some "_table" suffixes have been removed to avoid repeated
occurance of "table" like "tables/linux_error_table.h".
Reviewed By: michaelrj
Differential Revision: https://reviews.llvm.org/D151367
Summry:
This was accidentally dropped from a previous patch following a rebase.
Fix it to where it's consistent.
Differential Revision: https://reviews.llvm.org/D151232
Currently we have the `send_n` and `recv_n` routines to stream data,
such as a string to print, to the other side. The first operation is to
send the size so the other side knows the number of bytes to recieve.
However, this wasted 56 bytes that could've been sent. This meant that
small values, like the arguments to a function to call on the host for
example, needed to perform an extra send. This patch sends the first 56
bytes in the first packet and continues if necessary.
Depends on D150992
Reviewed By: JonChesterfield
Differential Revision: https://reviews.llvm.org/D151041
We provide the `send_n` and `recv_n` utilities as a generic way to
stream data between both sides of the process. This was previously
tested and performed as expected when using a string of constant size.
However, when the size was allowed to diverge between the threads in the
warp or wavefront this could deadlock. This did not occur on NVPTX
because of the use of the explicit warp sync. However, on AMD one of the
work items in the wavefront could continue executing and hit the next
`recv` call before the other threads, then we would deadlock as we
violated the RPC invariants.
This patch replaces the for loop with a thread ballot. This will cause
every thread in the warp or wavefront to continue executing the loop
until all of them can exit. This acts as a more explicit wavefront sync.
Reviewed By: JonChesterfield
Differential Revision: https://reviews.llvm.org/D150992
Missing in header for `pthread_exit` and `exit`.
Missing in spec file for `pthread_exit`.
Reviewed By: sivachandra
Differential Revision: https://reviews.llvm.org/D151143
1) Avoid proper function calls and referencing local variables after
the stack has been deallocated. A proper function call/return or local
variable reference that may have spilled will cause invalid memory
reads after the stack has been deallocated.
2) Mark the function as [[noreturn]] and place
`__builtin_unreachable()` after the `SYS_exit` syscalls.
Reviewed By: sivachandra
Differential Revision: https://reviews.llvm.org/D151142
Only functional for stack growsdown (same as before), but custom
`stack`, `stacksize`, `guardsize`, and `detachstate` all should be
working.
Differential Revision: https://reviews.llvm.org/D148290
The previous string to float tests didn't check correctness, but due to
the atof differential test proving unreliable the strtofloat fuzz test
has been changed to use MPFR for correctness checking. Some minor bugs
have been found and fixed as well.
Reviewed By: lntue
Differential Revision: https://reviews.llvm.org/D150905
Unit tests for the str() method have also been added.
Previously, a separate test only helper function was being used by the
test matchers which has regressed over many cleanups. Moreover, being a
test only utility, it was not tested separately (and hence the
regression).
Reviewed By: michaelrj
Differential Revision: https://reviews.llvm.org/D150906
This new functionality will help us avoid duplicated code in various
places in the testing infrastructure. Since the string representation
of the wide numbers is to be used by tests, to keep it simple, we
zero-pad the strings.
Reviewed By: lntue
Differential Revision: https://reviews.llvm.org/D150849
This patch changes the `Process` struct to only provide the functions
expected to be visible by the interface. So, now we only export the
open, reset, and size query functions. This prevents users of the
interface from messing with the internals of the process, so now the
only existing failure mode is mismatched send and recieve calls.
Reviewed By: michaelrj
Differential Revision: https://reviews.llvm.org/D150703
LIBC_INLINE was doubly defined in two headers. Define it only in
one place. Also update a few uses to make sure it's always placed
where a function attribute is valid and is used consistently on
every declaration of the same function in case the attributes used
in its definition must match on declarations and definitions.
Reviewed By: abrachet
Differential Revision: https://reviews.llvm.org/D150731
Currently we use a template parameter called `InvertInbox` to invert the
inbox when we load it. This is more easily understood as a static check
on whether or not the process running it is the server. Inverting the
inbox makes the states 1 0 and 0 1 own the buffer, so it's easier to
simply say that the server own the buffer if in != out. Also clean up some of
the comments.
Reviewed By: JonChesterfield
Differential Revision: https://reviews.llvm.org/D150365
This patch adds two versions of `bcmp` optimized for architectures where unaligned accesses are either illegal or extremely slow.
It is currently enabled for RISCV 64 and RISCV 32 but it could be used for ARM 32 architectures as well.
Here is the before / after output of `libc.benchmarks.memory_functions.opt_host --benchmark_filter=BM_memcmp` on a quad core Linux starfive RISCV 64 board running at 1.5GHz.
Before
```
Run on (4 X 1500 MHz CPU s)
CPU Caches:
L1 Instruction 32 KiB (x4)
L1 Data 32 KiB (x4)
L2 Unified 2048 KiB (x1)
----------------------------------------------------------------------
Benchmark Time CPU Iterations UserCounters...
----------------------------------------------------------------------
BM_Memcmp/0/0 110 ns 66.4 ns 10404864 bytes_per_cycle=0.107646/s bytes_per_second=153.989M/s items_per_second=15.071M/s __llvm_libc::memcmp,memcmp Google A
BM_Memcmp/1/0 318 ns 211 ns 3026944 bytes_per_cycle=0.131539/s bytes_per_second=188.167M/s items_per_second=4.73691M/s __llvm_libc::memcmp,memcmp Google B
BM_Memcmp/2/0 204 ns 115 ns 6118400 bytes_per_cycle=0.121675/s bytes_per_second=174.058M/s items_per_second=8.70241M/s __llvm_libc::memcmp,memcmp Google D
BM_Memcmp/3/0 143 ns 99.6 ns 7013376 bytes_per_cycle=0.117974/s bytes_per_second=168.763M/s items_per_second=10.0437M/s __llvm_libc::memcmp,memcmp Google L
BM_Memcmp/4/0 81.3 ns 58.2 ns 11426816 bytes_per_cycle=0.101125/s bytes_per_second=144.661M/s items_per_second=17.1805M/s __llvm_libc::memcmp,memcmp Google M
BM_Memcmp/5/0 177 ns 118 ns 5952512 bytes_per_cycle=0.120612/s bytes_per_second=172.537M/s items_per_second=8.45549M/s __llvm_libc::memcmp,memcmp Google Q
BM_Memcmp/6/0 342 ns 220 ns 3483648 bytes_per_cycle=0.132004/s bytes_per_second=188.834M/s items_per_second=4.54739M/s __llvm_libc::memcmp,memcmp Google S
BM_Memcmp/7/0 208 ns 130 ns 5681152 bytes_per_cycle=0.12468/s bytes_per_second=178.356M/s items_per_second=7.6674M/s __llvm_libc::memcmp,memcmp Google U
BM_Memcmp/8/0 123 ns 79.1 ns 8387584 bytes_per_cycle=0.110593/s bytes_per_second=158.204M/s items_per_second=12.6439M/s __llvm_libc::memcmp,memcmp Google W
BM_Memcmp/9/0 20707 ns 10643 ns 67584 bytes_per_cycle=0.142401/s bytes_per_second=203.707M/s items_per_second=93.9559k/s __llvm_libc::memcmp,uniform 384 to 4096
```
After
```
BM_Memcmp/0/0 80.4 ns 55.8 ns 12648448 bytes_per_cycle=0.132703/s bytes_per_second=189.834M/s items_per_second=17.9256M/s __llvm_libc::memcmp,memcmp Google A
BM_Memcmp/1/0 140 ns 80.5 ns 8230912 bytes_per_cycle=0.337273/s bytes_per_second=482.474M/s items_per_second=12.4165M/s __llvm_libc::memcmp,memcmp Google B
BM_Memcmp/2/0 101 ns 66.4 ns 10571776 bytes_per_cycle=0.208539/s bytes_per_second=298.317M/s items_per_second=15.0687M/s __llvm_libc::memcmp,memcmp Google D
BM_Memcmp/3/0 118 ns 67.6 ns 10533888 bytes_per_cycle=0.176822/s bytes_per_second=252.946M/s items_per_second=14.7946M/s __llvm_libc::memcmp,memcmp Google L
BM_Memcmp/4/0 106 ns 53.0 ns 12722176 bytes_per_cycle=0.111141/s bytes_per_second=158.988M/s items_per_second=18.8591M/s __llvm_libc::memcmp,memcmp Google M
BM_Memcmp/5/0 141 ns 70.2 ns 10436608 bytes_per_cycle=0.26032/s bytes_per_second=372.39M/s items_per_second=14.2458M/s __llvm_libc::memcmp,memcmp Google Q
BM_Memcmp/6/0 144 ns 79.3 ns 8932352 bytes_per_cycle=0.353168/s bytes_per_second=505.211M/s items_per_second=12.612M/s __llvm_libc::memcmp,memcmp Google S
BM_Memcmp/7/0 123 ns 71.7 ns 9945088 bytes_per_cycle=0.22143/s bytes_per_second=316.758M/s items_per_second=13.9421M/s __llvm_libc::memcmp,memcmp Google U
BM_Memcmp/8/0 97.0 ns 56.2 ns 12509184 bytes_per_cycle=0.160526/s bytes_per_second=229.635M/s items_per_second=17.7784M/s __llvm_libc::memcmp,memcmp Google W
BM_Memcmp/9/0 1840 ns 989 ns 676864 bytes_per_cycle=1.4894/s bytes_per_second=2.08067G/s items_per_second=1010.92k/s __llvm_libc::memcmp,uniform 384 to 4096
```
glibc
```
BM_Memcmp/0/0 72.6 ns 51.7 ns 12963840 bytes_per_cycle=0.141261/s bytes_per_second=202.075M/s items_per_second=19.3246M/s glibc::memcmp,memcmp Google A
BM_Memcmp/1/0 118 ns 75.2 ns 9280512 bytes_per_cycle=0.354054/s bytes_per_second=506.478M/s items_per_second=13.3046M/s glibc::memcmp,memcmp Google B
BM_Memcmp/2/0 114 ns 62.9 ns 11152384 bytes_per_cycle=0.222675/s bytes_per_second=318.539M/s items_per_second=15.8943M/s glibc::memcmp,memcmp Google D
BM_Memcmp/3/0 84.0 ns 63.5 ns 11030528 bytes_per_cycle=0.186353/s bytes_per_second=266.581M/s items_per_second=15.7378M/s glibc::memcmp,memcmp Google L
BM_Memcmp/4/0 93.5 ns 51.2 ns 13462528 bytes_per_cycle=0.119215/s bytes_per_second=170.539M/s items_per_second=19.5384M/s glibc::memcmp,memcmp Google M
BM_Memcmp/5/0 123 ns 61.7 ns 11376640 bytes_per_cycle=0.225262/s bytes_per_second=322.239M/s items_per_second=16.1993M/s glibc::memcmp,memcmp Google Q
BM_Memcmp/6/0 122 ns 71.6 ns 9967616 bytes_per_cycle=0.380844/s bytes_per_second=544.802M/s items_per_second=13.9579M/s glibc::memcmp,memcmp Google S
BM_Memcmp/7/0 118 ns 65.6 ns 10555392 bytes_per_cycle=0.238677/s bytes_per_second=341.43M/s items_per_second=15.2334M/s glibc::memcmp,memcmp Google U
BM_Memcmp/8/0 90.4 ns 54.0 ns 12920832 bytes_per_cycle=0.161987/s bytes_per_second=231.724M/s items_per_second=18.5169M/s glibc::memcmp,memcmp Google W
BM_Memcmp/9/0 1045 ns 601 ns 1195008 bytes_per_cycle=2.53677/s bytes_per_second=3.54383G/s items_per_second=1.66423M/s glibc::memcmp,uniform 384 to 4096
```
Reviewed By: sivachandra
Differential Revision: https://reviews.llvm.org/D150663
[libc] Add optimized bcmp for RISCV
This patch adds two versions of bcmp optimized for architectures where unaligned accesses are either illegal or extremely slow.
It is currently enabled for RISCV 64 and RISCV 32 but it could be used for ARM 32 architectures as well.
Here is the before / after output of libc.benchmarks.memory_functions.opt_host --benchmark_filter=BM_Bcmp on a quad core Linux starfive RISCV 64 board running at 1.5GHz.
Before
```
Run on (4 X 1500 MHz CPU s)
CPU Caches:
L1 Instruction 32 KiB (x4)
L1 Data 32 KiB (x4)
L2 Unified 2048 KiB (x1)
Load Average: 7.03, 5.98, 3.71
----------------------------------------------------------------------
Benchmark Time CPU Iterations UserCounters...
----------------------------------------------------------------------
BM_Bcmp/0/0 102 ns 60.5 ns 11662336 bytes_per_cycle=0.122696/s bytes_per_second=175.518M/s items_per_second=16.5258M/s __llvm_libc::bcmp,memcmp Google A
BM_Bcmp/1/0 328 ns 172 ns 3737600 bytes_per_cycle=0.15256/s bytes_per_second=218.238M/s items_per_second=5.80575M/s __llvm_libc::bcmp,memcmp Google B
BM_Bcmp/2/0 199 ns 99.7 ns 7019520 bytes_per_cycle=0.141897/s bytes_per_second=202.986M/s items_per_second=10.032M/s __llvm_libc::bcmp,memcmp Google D
BM_Bcmp/3/0 173 ns 86.5 ns 8361984 bytes_per_cycle=0.13863/s bytes_per_second=198.312M/s items_per_second=11.5669M/s __llvm_libc::bcmp,memcmp Google L
BM_Bcmp/4/0 105 ns 51.8 ns 13213696 bytes_per_cycle=0.116399/s bytes_per_second=166.51M/s items_per_second=19.2931M/s __llvm_libc::bcmp,memcmp Google M
BM_Bcmp/5/0 167 ns 93.9 ns 7853056 bytes_per_cycle=0.139432/s bytes_per_second=199.459M/s items_per_second=10.6503M/s __llvm_libc::bcmp,memcmp Google Q
BM_Bcmp/6/0 262 ns 165 ns 3931136 bytes_per_cycle=0.151516/s bytes_per_second=216.745M/s items_per_second=6.07091M/s __llvm_libc::bcmp,memcmp Google S
BM_Bcmp/7/0 168 ns 105 ns 6665216 bytes_per_cycle=0.143159/s bytes_per_second=204.791M/s items_per_second=9.52163M/s __llvm_libc::bcmp,memcmp Google U
BM_Bcmp/8/0 108 ns 68.0 ns 10175488 bytes_per_cycle=0.125504/s bytes_per_second=179.535M/s items_per_second=14.701M/s __llvm_libc::bcmp,memcmp Google W
BM_Bcmp/9/0 15371 ns 9007 ns 78848 bytes_per_cycle=0.166128/s bytes_per_second=237.648M/s items_per_second=111.031k/s __llvm_libc::bcmp,uniform 384 to 4096
```
After
```
BM_Bcmp/0/0 74.2 ns 49.7 ns 14306304 bytes_per_cycle=0.148927/s bytes_per_second=213.042M/s items_per_second=20.1101M/s __llvm_libc::bcmp,memcmp Google A
BM_Bcmp/1/0 108 ns 68.1 ns 10350592 bytes_per_cycle=0.411197/s bytes_per_second=588.222M/s items_per_second=14.6849M/s __llvm_libc::bcmp,memcmp Google B
BM_Bcmp/2/0 80.2 ns 56.0 ns 12386304 bytes_per_cycle=0.258588/s bytes_per_second=369.912M/s items_per_second=17.8585M/s __llvm_libc::bcmp,memcmp Google D
BM_Bcmp/3/0 92.4 ns 55.7 ns 12555264 bytes_per_cycle=0.206835/s bytes_per_second=295.88M/s items_per_second=17.943M/s __llvm_libc::bcmp,memcmp Google L
BM_Bcmp/4/0 79.3 ns 46.8 ns 14288896 bytes_per_cycle=0.125872/s bytes_per_second=180.061M/s items_per_second=21.3611M/s __llvm_libc::bcmp,memcmp Google M
BM_Bcmp/5/0 98.0 ns 57.9 ns 12232704 bytes_per_cycle=0.268815/s bytes_per_second=384.543M/s items_per_second=17.2711M/s __llvm_libc::bcmp,memcmp Google Q
BM_Bcmp/6/0 132 ns 65.5 ns 10474496 bytes_per_cycle=0.417246/s bytes_per_second=596.875M/s items_per_second=15.2673M/s __llvm_libc::bcmp,memcmp Google S
BM_Bcmp/7/0 101 ns 60.9 ns 11505664 bytes_per_cycle=0.253733/s bytes_per_second=362.968M/s items_per_second=16.4202M/s __llvm_libc::bcmp,memcmp Google U
BM_Bcmp/8/0 72.5 ns 50.2 ns 14082048 bytes_per_cycle=0.183262/s bytes_per_second=262.158M/s items_per_second=19.9271M/s __llvm_libc::bcmp,memcmp Google W
BM_Bcmp/9/0 852 ns 803 ns 854016 bytes_per_cycle=1.85028/s bytes_per_second=2.58481G/s items_per_second=1.24597M/s __llvm_libc::bcmp,uniform 384 to 4096
```
For comparison with glibc
```
BM_Bcmp/0/0 106 ns 52.6 ns 12906496 bytes_per_cycle=0.142072/s bytes_per_second=203.235M/s items_per_second=19.0271M/s glibc::bcmp,memcmp Google A
BM_Bcmp/1/0 132 ns 77.1 ns 8905728 bytes_per_cycle=0.365072/s bytes_per_second=522.239M/s items_per_second=12.9782M/s glibc::bcmp,memcmp Google B
BM_Bcmp/2/0 122 ns 62.3 ns 10909696 bytes_per_cycle=0.222667/s bytes_per_second=318.527M/s items_per_second=16.0563M/s glibc::bcmp,memcmp Google D
BM_Bcmp/3/0 99.5 ns 64.2 ns 11074560 bytes_per_cycle=0.185126/s bytes_per_second=264.825M/s items_per_second=15.5674M/s glibc::bcmp,memcmp Google L
BM_Bcmp/4/0 86.6 ns 50.2 ns 13488128 bytes_per_cycle=0.117941/s bytes_per_second=168.717M/s items_per_second=19.9053M/s glibc::bcmp,memcmp Google M
BM_Bcmp/5/0 106 ns 61.4 ns 11344896 bytes_per_cycle=0.248968/s bytes_per_second=356.151M/s items_per_second=16.284M/s glibc::bcmp,memcmp Google Q
BM_Bcmp/6/0 145 ns 71.9 ns 10046464 bytes_per_cycle=0.389814/s bytes_per_second=557.633M/s items_per_second=13.9019M/s glibc::bcmp,memcmp Google S
BM_Bcmp/7/0 119 ns 65.6 ns 10718208 bytes_per_cycle=0.243756/s bytes_per_second=348.696M/s items_per_second=15.2329M/s glibc::bcmp,memcmp Google U
BM_Bcmp/8/0 86.4 ns 54.5 ns 13250560 bytes_per_cycle=0.154831/s bytes_per_second=221.488M/s items_per_second=18.3532M/s glibc::bcmp,memcmp Google W
BM_Bcmp/9/0 1090 ns 604 ns 1186816 bytes_per_cycle=2.53848/s bytes_per_second=3.54622G/s items_per_second=1.65598M/s glibc::bcmp,uniform 384 to 4096
```
Reviewed By: sivachandra
Differential Revision: https://reviews.llvm.org/D150567
This patch is similar to 86fe88c8d9 and adds several explicit
constructor calls (bool(...), uint64_t(...), uint8_t(...)) that are
needed when we use UInt<T> (in my case UInt<128> in riscv32).
This patch also adds two operators to UInt<T>:
* operator/= required by printf_core/float_hex_converter.h:148
* operator-- required by FPUtil/ManipulationFunctions.h:166
Reviewed By: sivachandra, lntue
Differential Revision: https://reviews.llvm.org/D149594
We do a lot of arithmetic on void pointers here, so include a helper and
make some more consistent names. Changes no functionality.
Reviewed By: JonChesterfield
Differential Revision: https://reviews.llvm.org/D150576
This patch adds another variable to cache cases where we know that we
own the buffer. This allows us to skip the atomic load on the inbox
because we already know its state. This is legal immediately after
opening a port, or when sending immediately after a recieve. This
caching nets a significant (~17%) speedup for the basic open, send,
recieve combination.
Reviewed By: JonChesterfield
Differential Revision: https://reviews.llvm.org/D150516
This patch adds two versions of `memset` optimized for architectures where unaligned accesses are either illegal or extremely slow.
It is currently enabled for RISCV 64 and RISCV 32 but it could be used for ARM 32 architectures as well.
Here is the before / after output of libc.benchmarks.memory_functions.opt_host --benchmark_filter=BM_Memset on a quad core Linux starfive RISCV 64 board running at 1.5GHz.
Before
```
Run on (4 X 1500 MHz CPU s)
CPU Caches:
L1 Instruction 32 KiB (x4)
L1 Data 32 KiB (x4)
L2 Unified 2048 KiB (x1)
------------------------------------------------------------------------
Benchmark Time CPU Iterations UserCounters...
------------------------------------------------------------------------
BM_Memset/0/0 506 ns 252 ns 2883584 bytes_per_cycle=0.238312/s bytes_per_second=340.908M/s items_per_second=3.96043M/s __llvm_libc::memset,memset Google A
BM_Memset/1/0 296 ns 189 ns 2900992 bytes_per_cycle=0.234589/s bytes_per_second=335.583M/s items_per_second=5.29382M/s __llvm_libc::memset,memset Google B
BM_Memset/2/0 2110 ns 1049 ns 678912 bytes_per_cycle=0.24687/s bytes_per_second=353.151M/s items_per_second=953.527k/s __llvm_libc::memset,memset Google D
BM_Memset/3/0 397 ns 254 ns 3055616 bytes_per_cycle=0.238479/s bytes_per_second=341.147M/s items_per_second=3.93224M/s __llvm_libc::memset,memset Google L
BM_Memset/4/0 1119 ns 621 ns 1079296 bytes_per_cycle=0.244925/s bytes_per_second=350.368M/s items_per_second=1.61047M/s __llvm_libc::memset,memset Google M
BM_Memset/5/0 605 ns 349 ns 1644544 bytes_per_cycle=0.241364/s bytes_per_second=345.274M/s items_per_second=2.8614M/s __llvm_libc::memset,memset Google Q
BM_Memset/6/0 472 ns 271 ns 2310144 bytes_per_cycle=0.238615/s bytes_per_second=341.341M/s items_per_second=3.68799M/s __llvm_libc::memset,memset Google S
BM_Memset/7/0 262 ns 143 ns 3956736 bytes_per_cycle=0.225812/s bytes_per_second=323.026M/s items_per_second=7.0087M/s __llvm_libc::memset,memset Google U
BM_Memset/8/0 454 ns 261 ns 2940928 bytes_per_cycle=0.238883/s bytes_per_second=341.725M/s items_per_second=3.82716M/s __llvm_libc::memset,memset Google W
BM_Memset/9/0 8768 ns 5998 ns 115712 bytes_per_cycle=0.249196/s bytes_per_second=356.478M/s items_per_second=166.724k/s __llvm_libc::memset,uniform 384 to 4096
```
After
```
BM_Memset/0/0 117 ns 69.5 ns 9761792 bytes_per_cycle=0.935152/s bytes_per_second=1.30639G/s items_per_second=14.3834M/s __llvm_libc::memset,memset Google A
BM_Memset/1/0 97.8 ns 58.5 ns 13002752 bytes_per_cycle=0.892814/s bytes_per_second=1.24725G/s items_per_second=17.0848M/s __llvm_libc::memset,memset Google B
BM_Memset/2/0 326 ns 163 ns 5156864 bytes_per_cycle=1.54408/s bytes_per_second=2.15706G/s items_per_second=6.1192M/s __llvm_libc::memset,memset Google D
BM_Memset/3/0 132 ns 65.4 ns 11455488 bytes_per_cycle=0.876411/s bytes_per_second=1.22433G/s items_per_second=15.2803M/s __llvm_libc::memset,memset Google L
BM_Memset/4/0 222 ns 120 ns 6405120 bytes_per_cycle=1.44398/s bytes_per_second=2.01722G/s items_per_second=8.30758M/s __llvm_libc::memset,memset Google M
BM_Memset/5/0 119 ns 79.2 ns 8930304 bytes_per_cycle=1.13327/s bytes_per_second=1.58317G/s items_per_second=12.6189M/s __llvm_libc::memset,memset Google Q
BM_Memset/6/0 123 ns 64.0 ns 11609088 bytes_per_cycle=1.008/s bytes_per_second=1.40817G/s items_per_second=15.6365M/s __llvm_libc::memset,memset Google S
BM_Memset/7/0 85.9 ns 52.1 ns 12423168 bytes_per_cycle=0.641164/s bytes_per_second=917.192M/s items_per_second=19.1937M/s __llvm_libc::memset,memset Google U
BM_Memset/8/0 114 ns 67.1 ns 10347520 bytes_per_cycle=0.911968/s bytes_per_second=1.274G/s items_per_second=14.9015M/s __llvm_libc::memset,memset Google W
BM_Memset/9/0 1326 ns 785 ns 907264 bytes_per_cycle=1.89716/s bytes_per_second=2.6503G/s items_per_second=1.27348M/s __llvm_libc::memset,uniform 384 to 4096
```
Again not as good as current glibc but it's a first step in the right direction.
```
BM_Memset/0/0 108 ns 53.6 ns 12894208 bytes_per_cycle=1.02858/s bytes_per_second=1.4369G/s items_per_second=18.668M/s glibc::memset,memset Google A
BM_Memset/1/0 84.6 ns 47.6 ns 14284800 bytes_per_cycle=1.00197/s bytes_per_second=1.39974G/s items_per_second=21.0256M/s glibc::memset,memset Google B
BM_Memset/2/0 160 ns 85.8 ns 8927232 bytes_per_cycle=3.30805/s bytes_per_second=4.62129G/s items_per_second=11.6596M/s glibc::memset,memset Google D
BM_Memset/3/0 78.9 ns 53.6 ns 13326336 bytes_per_cycle=1.14058/s bytes_per_second=1.59338G/s items_per_second=18.674M/s glibc::memset,memset Google L
BM_Memset/4/0 99.2 ns 60.8 ns 11460608 bytes_per_cycle=2.54751/s bytes_per_second=3.55884G/s items_per_second=16.4587M/s glibc::memset,memset Google M
BM_Memset/5/0 93.0 ns 56.1 ns 12219392 bytes_per_cycle=1.73379/s bytes_per_second=2.42207G/s items_per_second=17.8157M/s glibc::memset,memset Google Q
BM_Memset/6/0 89.4 ns 47.2 ns 14692352 bytes_per_cycle=1.34846/s bytes_per_second=1.88377G/s items_per_second=21.1795M/s glibc::memset,memset Google S
BM_Memset/7/0 84.0 ns 50.0 ns 14468096 bytes_per_cycle=0.911198/s bytes_per_second=1.27293G/s items_per_second=19.994M/s glibc::memset,memset Google U
BM_Memset/8/0 93.4 ns 52.8 ns 13063168 bytes_per_cycle=1.06642/s bytes_per_second=1.48977G/s items_per_second=18.9524M/s glibc::memset,memset Google W
BM_Memset/9/0 438 ns 241 ns 2853888 bytes_per_cycle=6.1185/s bytes_per_second=8.54744G/s items_per_second=4.15064M/s glibc::memset,uniform 384 to 4096
```
Reviewed By: sivachandra
Differential Revision: https://reviews.llvm.org/D150433
Currently we provide the `send_n` and `recv_n` functions. These were
somewhat divergent and not tested on the GPU. This patch changes the
support to be more common. We do this my making the CPU provide an array
equal the to at least the lane size while the GPU can rely on the
private memory address of its stack variables. This allows us to send
data back and forth generically.
Reviewed By: JonChesterfield
Differential Revision: https://reviews.llvm.org/D150379
Small cleanup of the server code and fixes a constant name not following
the naming convention.
Differential Revision: https://reviews.llvm.org/D150361
Allows moving the pointer swap between server and client into reset.
Single allocation simplifies whatever allocates the client/server, currently
the libc loaders.
Reviewed By: jhuber6
Differential Revision: https://reviews.llvm.org/D150337
The interface exported by the RPC library allows users to simply send
and recieve fixed sized packets without worrying about the data motion
underneath. However, this was broken in the current implementation. We
can think of the send and recieve implementations in terms of waiting
for ownership of the buffer, using the buffer, and posting ownership to
the other side. Our implementation of `recv` was incorrect in the
following scenarios.
recv -> send // we still own the buffer and should give away ownership
recv -> close // The other side is not waiting for data, this will
result in multiple openings of the same port
This patch attempts to fix this with an admittedly hacky fix where we
track if the previous implementation was a recv and post conditionally.
Reviewed By: JonChesterfield
Differential Revision: https://reviews.llvm.org/D150327
Replaces the globals currently used. Worth changing to a bitmap
before allowing runtime number of ports >> 64. One bit per port is likely
to be cheap enough that sizing for the worst case is always fine, otherwise
in the future we can change to dynamically allocating it.
Reviewed By: jhuber6
Differential Revision: https://reviews.llvm.org/D150309
The Port type has stipuations that the same exact mask used to open it
needs to close it. This can currently be violated by calling its move
constructor to put it somewhere else. We still need the move constructor
to handle the open and closing functions. So, we simply make these
constructors private and only allow a few classes to have move
priviledges on it.
Reviewed By: JonChesterfield, lntue
Differential Revision: https://reviews.llvm.org/D150118
This patch adds two versions of memcpy optimized for architectures where unaligned accesses are either illegal or extremely slow.
It is currently enabled for RISCV 64 and RISCV 32 but it could be used for ARM 32 architectures as well.
Here is the before / after output of `libc.benchmarks.memory_functions.opt_host --benchmark_filter=BM_Memcpy` on a quad core Linux starfive RISCV 64 board running at 1.5GHz.
Before:
```
Run on (4 X 1500 MHz CPU s)
CPU Caches:
L1 Instruction 32 KiB (x4)
L1 Data 32 KiB (x4)
L2 Unified 2048 KiB (x1)
------------------------------------------------------------------------
Benchmark Time CPU Iterations UserCounters...
------------------------------------------------------------------------
BM_Memcpy/0/0 474 ns 474 ns 1483776 bytes_per_cycle=0.243492/s bytes_per_second=348.318M/s items_per_second=2.11097M/s __llvm_libc::memcpy,memcpy Google A
BM_Memcpy/1/0 210 ns 209 ns 3649536 bytes_per_cycle=0.233819/s bytes_per_second=334.481M/s items_per_second=4.77519M/s __llvm_libc::memcpy,memcpy Google B
BM_Memcpy/2/0 1814 ns 1814 ns 396288 bytes_per_cycle=0.247899/s bytes_per_second=354.622M/s items_per_second=551.402k/s __llvm_libc::memcpy,memcpy Google D
BM_Memcpy/3/0 89.3 ns 89.2 ns 7459840 bytes_per_cycle=0.217415/s bytes_per_second=311.014M/s items_per_second=11.2071M/s __llvm_libc::memcpy,memcpy Google L
BM_Memcpy/4/0 134 ns 134 ns 3815424 bytes_per_cycle=0.226584/s bytes_per_second=324.131M/s items_per_second=7.44567M/s __llvm_libc::memcpy,memcpy Google M
BM_Memcpy/5/0 52.8 ns 52.6 ns 11001856 bytes_per_cycle=0.194893/s bytes_per_second=278.797M/s items_per_second=19.0284M/s __llvm_libc::memcpy,memcpy Google Q
BM_Memcpy/6/0 180 ns 180 ns 4101120 bytes_per_cycle=0.231884/s bytes_per_second=331.713M/s items_per_second=5.55957M/s __llvm_libc::memcpy,memcpy Google S
BM_Memcpy/7/0 195 ns 195 ns 3906560 bytes_per_cycle=0.232951/s bytes_per_second=333.239M/s items_per_second=5.1217M/s __llvm_libc::memcpy,memcpy Google U
BM_Memcpy/8/0 152 ns 152 ns 4789248 bytes_per_cycle=0.227507/s bytes_per_second=325.452M/s items_per_second=6.58187M/s __llvm_libc::memcpy,memcpy Google W
BM_Memcpy/9/0 6036 ns 6033 ns 118784 bytes_per_cycle=0.249158/s bytes_per_second=356.423M/s items_per_second=165.75k/s __llvm_libc::memcpy,uniform 384 to 4096
```
After:
```
BM_Memcpy/0/0 126 ns 126 ns 5770240 bytes_per_cycle=1.04707/s bytes_per_second=1.46273G/s items_per_second=7.9385M/s __llvm_libc::memcpy,memcpy Google A
BM_Memcpy/1/0 75.1 ns 75.0 ns 10204160 bytes_per_cycle=0.691143/s bytes_per_second=988.687M/s items_per_second=13.3289M/s __llvm_libc::memcpy,memcpy Google B
BM_Memcpy/2/0 333 ns 333 ns 2174976 bytes_per_cycle=1.39297/s bytes_per_second=1.94596G/s items_per_second=3.00002M/s __llvm_libc::memcpy,memcpy Google D
BM_Memcpy/3/0 49.6 ns 49.5 ns 16092160 bytes_per_cycle=0.710161/s bytes_per_second=1015.89M/s items_per_second=20.1844M/s __llvm_libc::memcpy,memcpy Google L
BM_Memcpy/4/0 57.7 ns 57.7 ns 11213824 bytes_per_cycle=0.561557/s bytes_per_second=803.314M/s items_per_second=17.3228M/s __llvm_libc::memcpy,memcpy Google M
BM_Memcpy/5/0 48.0 ns 47.9 ns 16437248 bytes_per_cycle=0.346708/s bytes_per_second=495.97M/s items_per_second=20.8571M/s __llvm_libc::memcpy,memcpy Google Q
BM_Memcpy/6/0 67.5 ns 67.5 ns 10616832 bytes_per_cycle=0.614173/s bytes_per_second=878.582M/s items_per_second=14.8142M/s __llvm_libc::memcpy,memcpy Google S
BM_Memcpy/7/0 84.7 ns 84.6 ns 10480640 bytes_per_cycle=0.819077/s bytes_per_second=1.14424G/s items_per_second=11.8174M/s __llvm_libc::memcpy,memcpy Google U
BM_Memcpy/8/0 61.7 ns 61.6 ns 11191296 bytes_per_cycle=0.550078/s bytes_per_second=786.893M/s items_per_second=16.2279M/s __llvm_libc::memcpy,memcpy Google W
BM_Memcpy/9/0 981 ns 981 ns 703488 bytes_per_cycle=1.52333/s bytes_per_second=2.12807G/s items_per_second=1019.81k/s __llvm_libc::memcpy,uniform 384 to 4096
```
It is not as good as glibc for now so there's room for improvement. I suspect a path pumping 16 bytes at once given the doubled numbers for large copies.
```
BM_Memcpy/0/1 146 ns 82.5 ns 8576000 bytes_per_cycle=1.35236/s bytes_per_second=1.88922G/s items_per_second=12.1169M/s glibc memcpy,memcpy Google A
BM_Memcpy/1/1 112 ns 63.7 ns 10634240 bytes_per_cycle=0.628018/s bytes_per_second=898.387M/s items_per_second=15.702M/s glibc memcpy,memcpy Google B
BM_Memcpy/2/1 315 ns 180 ns 4079616 bytes_per_cycle=2.65229/s bytes_per_second=3.7052G/s items_per_second=5.54764M/s glibc memcpy,memcpy Google D
BM_Memcpy/3/1 85.3 ns 43.1 ns 15854592 bytes_per_cycle=0.774164/s bytes_per_second=1107.45M/s items_per_second=23.2249M/s glibc memcpy,memcpy Google L
BM_Memcpy/4/1 105 ns 54.3 ns 13427712 bytes_per_cycle=0.7793/s bytes_per_second=1114.8M/s items_per_second=18.4109M/s glibc memcpy,memcpy Google M
BM_Memcpy/5/1 77.1 ns 43.2 ns 16476160 bytes_per_cycle=0.279808/s bytes_per_second=400.269M/s items_per_second=23.1428M/s glibc memcpy,memcpy Google Q
BM_Memcpy/6/1 112 ns 62.7 ns 11236352 bytes_per_cycle=0.676078/s bytes_per_second=967.137M/s items_per_second=15.9387M/s glibc memcpy,memcpy Google S
BM_Memcpy/7/1 131 ns 65.5 ns 11751424 bytes_per_cycle=0.965616/s bytes_per_second=1.34895G/s items_per_second=15.2762M/s glibc memcpy,memcpy Google U
BM_Memcpy/8/1 104 ns 55.0 ns 12314624 bytes_per_cycle=0.583336/s bytes_per_second=834.468M/s items_per_second=18.1937M/s glibc memcpy,memcpy Google W
BM_Memcpy/9/1 932 ns 466 ns 1480704 bytes_per_cycle=3.17342/s bytes_per_second=4.43321G/s items_per_second=2.14679M/s glibc memcpy,uniform 384 to 4096
```
Reviewed By: sivachandra
Differential Revision: https://reviews.llvm.org/D150202
