When truncating to a format with fewer mantissa bits, APFloat::convert
will perform a right shift of the mantissa by the difference of the
precision of the two formats. Usually, this will result in just the
mantissa bits needed for the target format.
One special situation is if the input number is denormal. In this case,
the right shift may discard significant bits. This is usually not a
problem, since truncating a denormal usually results in zero (underflow)
after normalization anyway, since the result format's exponent range is
usually smaller than the target format's.
However, there is one case where the latter property does not hold:
when truncating from ppc_fp128 to double. In particular, truncating
a ppc_fp128 whose first double of the pair is denormal should result
in just that first double, not zero. The current code however
performs an excessive right shift, resulting in lost result bits.
This is then caught in the APFloat::normalize call performed by
APFloat::convert and causes an assertion failure.
This patch checks for the scenario of truncating a denormal, and
attempts to (possibly partially) replace the initial mantissa
right shift by decrementing the exponent, if doing so will still
result in a valid *target format* exponent.
Index: test/CodeGen/PowerPC/pr16573.ll
===================================================================
--- test/CodeGen/PowerPC/pr16573.ll (revision 0)
+++ test/CodeGen/PowerPC/pr16573.ll (revision 0)
@@ -0,0 +1,11 @@
+; RUN: llc < %s | FileCheck %s
+
+target triple = "powerpc64-unknown-linux-gnu"
+
+define double @test() {
+ %1 = fptrunc ppc_fp128 0xM818F2887B9295809800000000032D000 to double
+ ret double %1
+}
+
+; CHECK: .quad -9111018957755033591
+
Index: lib/Support/APFloat.cpp
===================================================================
--- lib/Support/APFloat.cpp (revision 185817)
+++ lib/Support/APFloat.cpp (working copy)
@@ -1956,6 +1956,23 @@
X86SpecialNan = true;
}
+ // If this is a truncation of a denormal number, and the target semantics
+ // has larger exponent range than the source semantics (this can happen
+ // when truncating from PowerPC double-double to double format), the
+ // right shift could lose result mantissa bits. Adjust exponent instead
+ // of performing excessive shift.
+ if (shift < 0 && isFiniteNonZero()) {
+ int exponentChange = significandMSB() + 1 - fromSemantics.precision;
+ if (exponent + exponentChange < toSemantics.minExponent)
+ exponentChange = toSemantics.minExponent - exponent;
+ if (exponentChange < shift)
+ exponentChange = shift;
+ if (exponentChange < 0) {
+ shift -= exponentChange;
+ exponent += exponentChange;
+ }
+ }
+
// If this is a truncation, perform the shift before we narrow the storage.
if (shift < 0 && (isFiniteNonZero() || category==fcNaN))
lostFraction = shiftRight(significandParts(), oldPartCount, -shift);
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@186409 91177308-0d34-0410-b5e6-96231b3b80d8
The status function is already using a syscall that returns the file size.
Remember it and implement file_size as a simple wrapper.
No functionally change, but clients that already use status now can avoid
calling file_size.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@186016 91177308-0d34-0410-b5e6-96231b3b80d8
Remove the implementation in include/llvm/Support/YAMLTraits.h.
Added a DenseMap type DITypeHashMap in DebugInfo.h:
DenseMap<std::pair<StringRef, unsigned>, MDNode*>
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@185852 91177308-0d34-0410-b5e6-96231b3b80d8
This function is complementary to createTemporaryFile. It handles the case were
the unique file is *not* temporary: we will rename it in the end. Since we
will rename it, the file has to be in the same filesystem as the final
destination and we don't prepend the system temporary directory.
This has a small semantic difference from unique_file: the default mode is 0666.
This matches the behavior of most unix tools. For example, with this change
lld now produces files with the same permissions as ld. I will add a test
of this change when I port clang over to createUniqueFile (next commit).
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@185726 91177308-0d34-0410-b5e6-96231b3b80d8
This function is inspired by clang's Driver::GetTemporaryPath. It hides the
pattern used for uniquing and requires simple file names that are always
placed in the system temporary directory.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@185716 91177308-0d34-0410-b5e6-96231b3b80d8
Originally if D.firstSigDigit == str.end(), we will have already dereferenced
D.firstSigDigit in the first predicate.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@185437 91177308-0d34-0410-b5e6-96231b3b80d8
Allow a BlockFrequency to be divided by a non-zero BranchProbability
with saturating arithmetic. This will be used to compute the frequency
of a loop header given the probability of leaving the loop.
Our long division algorithm already saturates on overflow, so that was a
freebie.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@185184 91177308-0d34-0410-b5e6-96231b3b80d8
* Don't try to create parent directories in unique_file. It had two problem:
* It violates the contract that it is atomic. If the directory creation
success and the file creation fails, we would return an error but the
file system was modified.
* When creating a temporary file clang would have to first check if the
parent directory existed or not to avoid creating one when it was not
supposed to.
* More efficient implementations of createUniqueDirectory and the unique_file
that produces only the file name. Now all 3 just call into a static
function passing what they want (name, file or directory).
Clang also has to be updated, so tests might fail if a bot picks up this commit
and not the corresponding clang one.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@185126 91177308-0d34-0410-b5e6-96231b3b80d8
This reverts commit r185099.
Looks like both the ppc-64 and mips bots are still failing after I reverted this
change.
Since:
1. The mips bot always performs a clean build,
2. The ppc64-bot failed again after a clean build (I asked the ppc-64
maintainers to clean the bot which they did... Thanks Will!),
I think it is safe to assume that this change was not the cause of the failures
that said builders were seeing. Thus I am recomitting.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@185111 91177308-0d34-0410-b5e6-96231b3b80d8
This reverts commit r185095. This is causing a FileCheck failure on
the 3dnow intrinsics on at least the mips/ppc bots but not on the x86
bots.
Reverting while I figure out what is going on.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@185099 91177308-0d34-0410-b5e6-96231b3b80d8
The category which an APFloat belongs to should be dependent on the
actual value that the APFloat has, not be arbitrarily passed in by the
user. This will prevent inconsistency bugs where the category and the
actual value in APFloat differ.
I also fixed up all of the references to this constructor (which were
only in LLVM).
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@185095 91177308-0d34-0410-b5e6-96231b3b80d8
There are a few valid situation where we care about the structure inside a
directory, but not about the directory itself. A simple example is for unit
testing directory traversal.
PathV1 had a function like this, add one to V2 and port existing users of the
created temp file and delete it hack to using it.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@185059 91177308-0d34-0410-b5e6-96231b3b80d8
Currently inside APFloat fcNormal still implies the old definition of Normal
(i.e. isFiniteNonZero) instead of the proper IEEE-754R definition that the
external method isNormal() uses.
This patch prepares for the internal switch inside APFloat by converting all
references that check if a category is fcNormal directly with an indirect call
via isFiniteNonZero().
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@185036 91177308-0d34-0410-b5e6-96231b3b80d8
This is easier to read than the internal fixed-point representation.
If anybody knows the correct algorithm for converting fixed-point
numbers to base 10, feel free to fix it.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@184881 91177308-0d34-0410-b5e6-96231b3b80d8
This is really ugly, but it is no worse than what we have in clang right now and
it is better to get it working first and clean/optimize it afterwards.
Will be tested from clang in the next patch.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@184802 91177308-0d34-0410-b5e6-96231b3b80d8
The method significandParts() is a helper method meant to ease access to
APFloat's significand by allowing the user to not need to be aware of whether or
not the APFloat is using memory allocated in the instance itself or in an
external array.
This assert says that one can only access the significand of FiniteNonZero/NaN
floats. This makes it cumbersome and more importantly dangerous when one wishes
to zero out the significand of a zero/infinity value since one will have to deal
with the aforementioned quandary related to how the memory in APFloat is
allocated.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@184711 91177308-0d34-0410-b5e6-96231b3b80d8
In the context of APFloat, seeing a macro called convolve suggests that APFloat
is using said value in some sort of convolution somewhere in the source code.
This is misleading.
I also added a documentation comment to the macro.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@184710 91177308-0d34-0410-b5e6-96231b3b80d8
exponent_t is only used internally in APFloat and no exponent_t values are
exposed via the APFloat API. In light of such conditions it does not make any
sense to gum up the llvm namespace with said type. Plus it makes it clearer that
exponent_t is associated with APFloat.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@184686 91177308-0d34-0410-b5e6-96231b3b80d8
Zero is used by BlockFrequencyInfo as a special "don't know" value. It also
causes a sink for frequencies as you can't ever get off a zero frequency with
more multiplies.
This recovers a 10% regression on MultiSource/Benchmarks/7zip. A zero frequency
was propagated into an inner loop causing excessive spilling.
PR16402.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@184584 91177308-0d34-0410-b5e6-96231b3b80d8
IR for CUDA should use "nvptx[64]-nvidia-cuda", and IR for NV OpenCL should use "nvptx[64]-nvidia-nvcl"
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@184579 91177308-0d34-0410-b5e6-96231b3b80d8
in the "parent" thread, when we are using CrashRecoveryContext::RunSafelyOnThread.
When using CrashRecoveryContext::RunSafelyOnThread, we would set a CrashRecoveryContextImpl* to a thread-local variable
for the "child" thread, but CrashRecoveryContext would erroneously clear it in the "parent" thread.
The result was that if CrashRecoveryContext::RunSafelyOnThread was called again in the "child" thread it would mess up
crash-recovery for its parent.
A test for this will be added in the clang repository.
rdar://14204560
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@184380 91177308-0d34-0410-b5e6-96231b3b80d8
This ports a missing feature from PathV1.h. I am not sure how to test this
with the regular infrastructure, but an Apple bot should check this when
r183985 is reapplied.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@184119 91177308-0d34-0410-b5e6-96231b3b80d8
This patch also adds a simpler version of sys::fs::remove and a tool_output_file
constructor for when we already have an open file.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@184095 91177308-0d34-0410-b5e6-96231b3b80d8
This broke for example the 'not' utility, if a directory called
'FileCheck' is executable and in the path before the actual 'FileCheck'.
This patch steals the implementation of the "old" PathV1 canExecute
implementation:
- checks for R_OK (file readable): this is necessary for executing
scripts; we should not regress here unless we have good reasons
- checks for S_ISREG; if we want to get rid of this, we'd need to
change all callers who already made the assumption when depending
on Path V1.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@184074 91177308-0d34-0410-b5e6-96231b3b80d8
It looks like clang-tools-extra/unittests/cpp11-migrate/TransformTest.cpp
depends on the behaviour of the old one on Windows. Maybe a difference
between GetCurrentDirectoryA and GetCurrentDirectoryW?
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@184009 91177308-0d34-0410-b5e6-96231b3b80d8
Execute's Data parameter is now optional, so we won't allocate memory
for it on Windows and we'll close the process handle.
The Unix code should probably do something similar to avoid accumulation
of zombie children that haven't been waited on.
Tested on Linux and Windows.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@183906 91177308-0d34-0410-b5e6-96231b3b80d8
It was only used to implement ExecuteAndWait and ExecuteNoWait. Expose just
those two functions and make Execute and Wait implementations details.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@183864 91177308-0d34-0410-b5e6-96231b3b80d8
`typeinfo for llvm:🆑:GenericOptionValue').
Remove an "anchor" method for an abstract class. (This does not
increase the number of vtables.)
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@183830 91177308-0d34-0410-b5e6-96231b3b80d8
Most clients have already been moved from Path V1 to V2. The ones using V1
now include PathV1.h explicitly.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@183801 91177308-0d34-0410-b5e6-96231b3b80d8
For ARM on linux we use /proc/cpuinfo to detect the host CPU's features.
Linux derives these values without ever looking at the vendor of the
specific CPU implementation. Hence, it adds little value, if we parse
the output of /proc/cpuinfo only for certain vendors.
This patch enables us to derive the correct feature flags e.g. for Qualcomm
CPUs.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@183790 91177308-0d34-0410-b5e6-96231b3b80d8
It will be tested in the next commit which moves another user to identify_magic.
Original message:
Fix an out of bounds array access.
We were looking at Magic[5] without checking Length. Since this path would not
return unless Length >= 18 anyway, just move the >= 18 check up.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@183753 91177308-0d34-0410-b5e6-96231b3b80d8
We were looking at Magic[5] without checking Length. Since this path would not
return unless Length >= 18 anyway, just move the >= 18 check up.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@183666 91177308-0d34-0410-b5e6-96231b3b80d8
