Excluding ADC/SBB and the bit-test instructions (BTR/BTS/BTC), the observed
latency of all other RMW integer arithmetic/logic instructions is 6cy and not
5cy.
Example (ADD):
```
addb $0, (%rsp) # Latency: 6cy
addb $7, (%rsp) # Latency: 6cy
addb %sil, (%rsp) # Latency: 6cy
addw $0, (%rsp) # Latency: 6cy
addw $511, (%rsp) # Latency: 6cy
addw %si, (%rsp) # Latency: 6cy
addl $0, (%rsp) # Latency: 6cy
addl $511, (%rsp) # Latency: 6cy
addl %esi, (%rsp) # Latency: 6cy
addq $0, (%rsp) # Latency: 6cy
addq $511, (%rsp) # Latency: 6cy
addq %rsi, (%rsp) # Latency: 6cy
```
The same latency profile applies to SUB/AND/OR/XOR/INC/DEC.
The observed latency of ADC/SBB is 7-8cy. So we need a different write to model
those. Latency of BTS/BTR/BTC is not fixed by this patch (they are much slower
than what the model for btver2 currently reports).
Differential Revision: https://reviews.llvm.org/D66636
llvm-svn: 369748
ExtName should not be decorated, just like Name.
This avoids double decoration on symbols in import libraries
that use = for renaming functions. (Weak aliases, which use ==,
worked fine with respect to decoration.)
Differential Revision: https://reviews.llvm.org/D66617
llvm-svn: 369747
If the accumulator and either of the multiply operands are negatable then we can we negate the entire expression.
Differential Revision: https://reviews.llvm.org/D63141
llvm-svn: 369746
Summary:
Add support for gfx10, where all DPP operations are confined to work
within a single row of 16 lanes, and wave32.
Reviewers: arsenm, sheredom, critson, rampitec
Subscribers: kzhuravl, jvesely, wdng, nhaehnle, yaxunl, t-tye, hiraditya, jfb, dstuttard, tpr, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D65644
llvm-svn: 369745
st_other field of a symbol usually contains its visibility.
Other bits are usually 0, though some targets, like
MIPS can set them using the named bit field values.
Problem is that there is no way to set an arbitrary value now,
though that might be useful for our test cases.
In this patch I introduced a way to set st_other to any numeric
value using the new StOther field.
I added a test and simplified the existent one to show the effect/benefit
Differential revision: https://reviews.llvm.org/D66583
llvm-svn: 369742
I think --reproduce is no longer a debug-only option but a useful
option that a common user may want to use. So, this patch updates
the description of the option in the manual page.
Differential Revision: https://reviews.llvm.org/D60557
llvm-svn: 369740
Summary:
The macro are usually defined in the common/base headers which are hard
for normal users to modify it.
Reviewers: gribozavr, alexfh
Subscribers: xazax.hun, cfe-commits
Tags: #clang
Differential Revision: https://reviews.llvm.org/D66631
llvm-svn: 369739
The AVX2 check lines show two issues. An ADD that became an OR
because we knew the input was disjoint, but really it was zero
so we should have just removed the ADD/OR all together.
Relatedly we use 128-bit VPMADDWD instructions followed by
256-bit VPADDD operations. We should be able to narrow these
VPADDDs.
llvm-svn: 369736
Summary:
This removes DeclVendor's dependency on clang (and ClangASTContext).
DeclVendor has no need to know about specific TypeSystems.
Differential Revision: https://reviews.llvm.org/D66628
llvm-svn: 369735
For v2i32 we only feed 2 i8 elements into the psadbw instructions
with 0s in the other 14 bytes. The resulting psadbw instruction
will produce zeros in bits [127:16] of the output. We need to take
the result and feed it to a v2i32 add where the first element
includes bits [15:0] of the sad result. The other element should
be zero.
Prior to this patch we were using a truncate to take 0 from
bits 95:64 of the psadbw. This results in a pshufd to move those
bits to 63:32. But since we also have zeroes in bits 63:32 of
the psadbw output, we should just take those bits.
The previous code probably worked better with promoting legalization,
but now we use widening legalization. I've preserved the old
behavior if -x86-experimental-vector-widening-legalization=false
until we get that option removed.
llvm-svn: 369733
PR43085.
Recognize .mjs files as JavaScript. .mjs is the extension for ECMAScript modules.
A specific extension (and associated content type javascript/esm) is
introduced to differentiate it from CommonJS modules and solve some
interoperability problems.
Differential Revision: https://reviews.llvm.org/D66584
Patch by Fergal Daly
llvm-svn: 369732
We were computing the loop exit value, but not ensuring the addrec belonged to the loop whose exit value we were computing. I couldn't actually trip this; the test case shows the basic setup which *might* trip this, but none of the variations I've tried actually do.
llvm-svn: 369730
Our SVal hierarchy doesn't allow modeling pointer casts as no-op. The
pointer type is instead encoded into the pointer object. Defer to our
usual pointer casting facility, SValBuilder::evalBinOp().
Fixes a crash.
llvm-svn: 369729
The alignment is calculated incorrectly, thus sometimes it doesn't generate aligned mov instructions, as shown by the example below:
```
// b.cc
typedef long long index;
extern "C" index g_tid;
extern "C" index g_num;
void add3(float* __restrict__ a, float* __restrict__ b, float* __restrict__ c) {
index n = 64*1024;
index m = 16*1024;
index k = 4*1024;
index tid = g_tid;
index num = g_num;
__builtin_assume_aligned(a, 32);
__builtin_assume_aligned(b, 32);
__builtin_assume_aligned(c, 32);
for (index i0=tid*k; i0<m; i0+=num*k)
for (index i1=0; i1<n*m; i1+=m)
for (index i2=0; i2<k; i2++)
c[i1+i0+i2] = b[i0+i2] + a[i1+i0+i2];
}
```
Compile with `clang b.cc -Ofast -march=skylake -mavx2 -S`
```
vmovaps -224(%rdi,%rbx,4), %ymm0
vmovups -192(%rdi,%rbx,4), %ymm1 # should be movaps
vmovups -160(%rdi,%rbx,4), %ymm2 # should be movaps
vmovups -128(%rdi,%rbx,4), %ymm3 # should be movaps
vaddps -224(%rsi,%rbx,4), %ymm0, %ymm0
vaddps -192(%rsi,%rbx,4), %ymm1, %ymm1
vaddps -160(%rsi,%rbx,4), %ymm2, %ymm2
vaddps -128(%rsi,%rbx,4), %ymm3, %ymm3
vmovaps %ymm0, -224(%rdx,%rbx,4)
vmovups %ymm1, -192(%rdx,%rbx,4) # should be movaps
vmovups %ymm2, -160(%rdx,%rbx,4) # should be movaps
vmovups %ymm3, -128(%rdx,%rbx,4) # should be movaps
```
Differential Revision: https://reviews.llvm.org/D66575
Patch by Dun Liang
llvm-svn: 369723
One problem with untagging memory in landing pads is that it only works
correctly if the function that catches the exception is instrumented.
If the function is uninstrumented, we have no opportunity to untag the
memory.
To address this, replace landing pad instrumentation with personality function
wrapping. Each function with an instrumented stack has its personality function
replaced with a wrapper provided by the runtime. Functions that did not have
a personality function to begin with also get wrappers if they may be unwound
past. As the unwinder calls personality functions during stack unwinding,
the original personality function is called and the function's stack frame is
untagged by the wrapper if the personality function instructs the unwinder
to keep unwinding. If unwinding stops at a landing pad, the function is
still responsible for untagging its stack frame if it resumes unwinding.
The old landing pad mechanism is preserved for compatibility with old runtimes.
Differential Revision: https://reviews.llvm.org/D66377
llvm-svn: 369721
Prefer `MCFixupKind` where possible and add getTargetKind() to
convert to `unsigned` when needed rather than scattering cast
operators around the place.
Differential Revision: https://reviews.llvm.org/D59890
llvm-svn: 369720
I've been working on a new tool, llvm-ifs, for merging interface stub files
generated by clang and I've iterated on my derivative format of TBE to a newer
format. llvm-ifs will only support the new format, so I am going to drop the
older experimental interface stubs formats in this commit to make things
simpler.
Differential Revision: https://reviews.llvm.org/D66573
llvm-svn: 369719
Summary: https://reviews.llvm.org/D66620 is accepted but was based on the multi-repo setup, so I was not able to `arc patch` it. Resubmit the diff under monorepo
Committed on behalf of @sugak (Igor Sugak)
Reviewers: sugak
Subscribers: #sanitizers, llvm-commits, vitalybuka
Tags: #sanitizers, #llvm
Differential Revision: https://reviews.llvm.org/D66624
llvm-svn: 369716
After posting llvm-ifs on phabricator, I made some progress in hardening up how
I think the format for Interface Stubs should look. There are a number of
things I think the TBE format was missing (no endianness, no info about the
Object Format because it assumes ELF), so I have added those and broken off
from being as similar to the TBE schema. In a subsequent commit I can drop the
other formats.
An example of how The format will look is as follows:
--- !experimental-ifs-v1
IfsVersion: 1.0
Triple: x86_64-unknown-linux-gnu
ObjectFileFormat: ELF
Symbols:
_Z9nothiddenv: { Type: Func }
_Z10cmdVisiblev: { Type: Func }
...
The format is still marked experimental.
Differential Revision: https://reviews.llvm.org/D66446
llvm-svn: 369715
This patch is also motivated by the Swift branch and is effectively NFC for the single-TypeSystem llvm.org branch.
In multi-language projects it is extremely common to have, e.g., a
Clang type and a similarly-named rendition of that same type in
another language. When searching for a type It is much cheaper to pass
a set of supported languages to the SymbolFile than having it
materialize every result and then rejecting the materialized types
that have the wrong language.
Differential Revision: https://reviews.llvm.org/D66546
<rdar://problem/54471165>
This reapplies r369690 with a previously missing constructor for LanguageSet.
llvm-svn: 369710
It looks like running without this argument was supported
for legacy reasons, but a Xcode 11 beta made the argument
mandatory for our usecase.
llvm-svn: 369709
- Especially MemorySanitizer fails if those sysctl configs are enabled.
Reviewers: vitalybuka, emaste, dim
Reviewed By: dim
Differential Revision: https://reviews.llvm.org/D66582
llvm-svn: 369708
Summary:
It seems that CodeGen was always using ExternalLinkage when emitting a
GlobalDecl with __attribute__((alias)). This leads to symbol
redefinitions (ODR) that cause failures at link time for static aliases.
This is readily attempting to link an ARM (32b) allyesconfig Linux
kernel built with Clang.
Reported-by: nathanchance
Suggested-by: ihalip
Link: https://bugs.llvm.org/show_bug.cgi?id=42377
Link: https://github.com/ClangBuiltLinux/linux/issues/631
Reviewers: rsmith, aaron.ballman, erichkeane
Reviewed By: aaron.ballman
Subscribers: javed.absar, kristof.beyls, cfe-commits, srhines, ihalip, nathanchance
Tags: #clang
Differential Revision: https://reviews.llvm.org/D66492
llvm-svn: 369705
I noticed another instance of the issue where references to aliases were
being replaced with aliasees, this time in InstCombine. In the instance that
I saw it turned out to be only a QoI issue (a symbol ended up being missing
from the symbol table due to the last reference to the alias being removed,
preventing HWASAN from symbolizing a global reference), but it could easily
have manifested as incorrect behaviour.
Since this is the third such issue encountered (previously: D65118, D65314)
it seems to be time to address this common error/QoI issue once and for all
and make the strip* family of functions not look through aliases.
Includes a test for the specific issue that I saw, but no doubt there are
other similar bugs fixed here.
As with D65118 this has been tested to make sure that the optimization isn't
load bearing. I built Clang, Chromium for Linux, Android and Windows as well
as the test-suite and there were no size regressions.
Differential Revision: https://reviews.llvm.org/D66606
llvm-svn: 369697
Anastasia Stulova