This change makes sure that we compute the element size and the byte stride
based on the target representation of the element type.
For example, when REAL*10 is mapped to x86_fp80 each element occupies
16 bytes rather than 10 because of the padding.
Note that the size computation method used here actually returns
the distance between two adjacent element of the *same* type in memory
(which is equivalent to llvm::DataLayout::getTypeAllocSize()).
It does not return the number of bytes that may be overwritten
by storing a value of the specified type (e.g. what can be computed
via llvm::DataLayout::getTypeStoreSize(), but not available in
mlir::DataLayout).
Differential Revision: https://reviews.llvm.org/D133508
Summary:
The AMDGPU and CUDA plugins now relies on the Object and Support
libraries. This patch adds them explicitly rather than hoping that they
share the symbols loaded from the standard `libomptarget`.
The tiling interface generates the order of parallel_insert_slice incorrectly when there are multiple destionation operands. This revision fixes that and adds a test for it. It depends on D132937
Reviewed By: nicolasvasilache
Differential Revision: https://reviews.llvm.org/D133204
Move the broadcasting support from GDBRemoteCommunication
to GDBRemoteClientBase since this is where it is actually used. Remove
GDBRemoteCommunication and subclass constructor arguments left over
after Communication cleanup.
Sponsored by: The FreeBSD Foundation
Differential Revision: https://reviews.llvm.org/D133427
This revision relaxes constraint of tiling when there are multiple destination operands. It also adds a test.
Reviewed By: nicolasvasilache
Differential Revision: https://reviews.llvm.org/D132937
HLSL doesn't have a runtime loader model that supports global
construction by a loader or runtime initializer. To allow us to leverage
global constructors with minimal code generation impact we put calls to
the global constructors inside the generated entry function.
Differential Revision: https://reviews.llvm.org/D132977
Implement acosf function correctly rounded for all rounding modes.
We perform range reduction as follows:
- When `|x| < 2^(-10)`, we use cubic Taylor polynomial:
```
acos(x) = pi/2 - asin(x) ~ pi/2 - x - x^3 / 6.
```
- When `2^(-10) <= |x| <= 0.5`, we use the same approximation that is used for `asinf(x)` when `|x| <= 0.5`:
```
acos(x) = pi/2 - asin(x) ~ pi/2 - x - x^3 * P(x^2).
```
- When `0.5 < x <= 1`, we use the double angle formula: `cos(2y) = 1 - 2 * sin^2 (y)` to reduce to:
```
acos(x) = 2 * asin( sqrt( (1 - x)/2 ) )
```
- When `-1 <= x < -0.5`, we reduce to the positive case above using the formula:
```
acos(x) = pi - acos(-x)
```
Performance benchmark using perf tool from the CORE-MATH project on Ryzen 1700:
```
$ CORE_MATH_PERF_MODE="rdtsc" ./perf.sh acosf
GNU libc version: 2.35
GNU libc release: stable
CORE-MATH reciprocal throughput : 28.613
System LIBC reciprocal throughput : 29.204
LIBC reciprocal throughput : 24.271
$ CORE_MATH_PERF_MODE="rdtsc" ./perf.sh asinf --latency
GNU libc version: 2.35
GNU libc release: stable
CORE-MATH latency : 55.554
System LIBC latency : 76.879
LIBC latency : 62.118
```
Reviewed By: orex, zimmermann6
Differential Revision: https://reviews.llvm.org/D133550
The LLVM performance tips suggest that allocas should be placed at the
beginning of the entry block. So far, llvm doesn’t provide any helper to
find that position.
Add BasicBlock::getFirstNonPHIOrDbgOrAlloca and IRBuilder::SetInsertPointPastAllocas(Function*)
that get an insert position after the (static) allocas at the start of a
function and use it in ShadowStackGCLowering.
Differential Revision: https://reviews.llvm.org/D132554
As discussed here:
https://discourse.llvm.org/t/build-llvm-release-bat-script-options
Add a function to parse command line arguments: `parse_args`.
The format for the arguments is:
Boolean: --option
Value: --option<separator>value
with `<separator>` being: space, colon, semicolon or equal sign
Command line usage example:
my-batch-file.bat --build --type=release --version 123
It will create 3 variables:
`build` with the value `true`
`type` with the value `release`
`version` with the value `123`
Usage:
set "build="
set "type="
set "version="
REM Parse arguments.
call :parse_args %*
if defined build (
...
)
if %type%=='release' (
...
)
if %version%=='123' (
...
)
Uses our existing "error string" extension to provide a better
indication of why the launch failed (the client does not make use of the
error yet).
Also, fix the way we obtain the launch error message (make sure we read
the whole message, and skip trailing garbage), and reduce the size of
TestLldbGdbServer by splitting some tests into a separate file.
Differential Revision: https://reviews.llvm.org/D133352
The Read function could end up blocking if data (or EOF) arrived just as
it was about to start waiting for the events. This was only discovered
now, because we did not have unit tests for this functionality before.
We need to check for data *after* we start listening for incoming
events. There were no changes to the read thread code needed, as we
already use this pattern in SynchronizeWithReadThread, so I just updated
the comments to make it clear that it is used for reading as well.
Differential Revision: https://reviews.llvm.org/D133410
Noticed while investigating BITREVERSE cost numbers with the D103695 script - VPPERM folded loads was using the WriteVarShuffleX defaults and was missing an override like the VPPERM reg-reg variants
It doesn't have a section header string table so add a vector to have
the strings and create name based on the program header type and the
index.
Differential Revision: https://reviews.llvm.org/D131290
Clang has support of virtual file system for the purpose of testing, but
treatment of config files did not use it. This change enables VFS in it
as well.
Differential Revision: https://reviews.llvm.org/D132867
If there are non-load/store users of the promoted pointer, we
currently abort promotion. However, having such users isn't really
relevant to the transform. We already separately check that a)
there are no instructions that modref the promoted pointer and
b) that a pointer capture disables store promotion.
In the affected @test_captured_in_loop test case we have a readnone
capture of the promoted pointer, which means that load promotion
can be performed (while store promotion cannot).
Differential Revision: https://reviews.llvm.org/D133485
Clang has support of virtual file system for the purpose of testing, but
treatment of config files did not use it. This change enables VFS in it
as well.
Differential Revision: https://reviews.llvm.org/D132867
This reverts commit 053841c5624ca7eacd108a26071d8a1cefe1bebd.
We faced a use-after-free after pushing the D113291, since the
foldSqrt() has a call to eraseFromParent(). The function
should be at the end of the main loop that folds the patterns.
This patch fixes that.
This option can be used to enable Control Flow Guard checks and
generation of address-taken function table. They are equivalent to
`/guard:cf` and `/guard:cf,nochecks` in clang-cl. Passing this flag to
the Clang driver will also pass `--guard-cf` to the MinGW linker.
This feature is disabled by default. The option `-mguard=none` is also
available to explicitly disable this feature.
Reviewed By: rnk
Differential Revision: https://reviews.llvm.org/D132810
These will be LLD-specific options to support Control Flow Guard for the
MinGW target. They are disabled by default, but enabling `--guard-cf`
will also enable `--guard-longjmp` unless `--no-guard-longjmp` is also
specified. These options maps to `-guard:cf,[no]longjmp`.
Note that these features require the `_load_config_used` symbol to
contain the load config directory and be filled with the required
symbols. While current versions of mingw-w64 do not supply this symbol,
the user can provide their own version of it.
Reviewed By: MaskRay, rnk
Differential Revision: https://reviews.llvm.org/D132808
According to the revised description in `LoongArch Reference Manual v1.02`,
frint.[s/d] does not judge whether floating-point inexact exceptions are
allowed indicated by FCSR, i.e. always executes roundToIntegralExact(x).
What's more, the manual also specifically defines that frint.s/d is only
necessary to be defined in LA64. So ISD::FRINT is legal for LA64.
Differential Revision: https://reviews.llvm.org/D133337
This change adds a set of utilities to replace the result of a
`tensor.collapse_shape -> tensor.extract_slice` chain with the
equivalent result formed by aggregating slices of the
`tensor.collapse_shape` source. In general, it is not possible to
commute `extract_slice` and `collapse_shape` if linearized dimensions
are sliced. The i-th dimension of the `tensor.collapse_shape`
result is a "linearized sliced dimension" if:
1) Reassociation indices of tensor.collapse_shape in the i'th position
is greater than size 1 (multiple dimensions of the input are collapsed)
2) The i-th dimension is sliced by `tensor.extract_slice`.
We can work around this by stitching together the result of
`tensor.extract_slice` by iterating over any linearized sliced dimensions.
This is equivalent to "tiling" the linearized-and-sliced dimensions of
the `tensor.collapse_shape` operation in order to manifest the result
tile (the result of the `tensor.extract_slice`). The user of the
utilities must provide the mechanism to create the tiling (e.g. a loop).
In the tests, it is demonstrated how to apply the utilities using either
`scf.for` or `scf.foreach_thread`.
The below example illustrates the pattern using `scf.for`:
```
%0 = linalg.generic ... -> tensor<3x7x11x10xf32>
%1 = tensor.collapse_shape %0 [[0, 1, 2], [3]] : ... to tensor<341x10xf32>
%2 = tensor.extract_slice %1 [13, 0] [10, 10] [2, 1] : .... tensor<10x10xf32>
```
We can construct %2 by generating the following IR:
```
%dest = linalg.init_tensor() : tensor<10x10xf32>
%2 = scf.for %iv = %c0 to %c10 step %c1 iter_args(%arg0) -> tensor<10x10xf32> {
// Step 1: Map this output idx (%iv) to a multi-index for the input (%3):
%linear_index = affine.apply affine_map<(d0)[]->(d0*2 + 11)>(%iv)
%3:3 = arith.delinearize_index %iv into (3, 7, 11)
// Step 2: Extract the slice from the input
%4 = tensor.extract_slice %0 [%3#0, %3#1, %3#2, 0] [1, 1, 1, 10] [1, 1, 1, 1] :
tensor<3x7x11x10xf32> to tensor<1x1x1x10xf32>
%5 = tensor.collapse_shape %4 [[0, 1, 2], [3]] :
tensor<1x1x1x10xf32> into tensor<1x10xf32>
// Step 3: Insert the slice into the destination
%6 = tensor.insert_slice %5 into %arg0 [%iv, 0] [1, 10] [1, 1] :
tensor<1x10xf32> into tensor<10x10xf32>
scf.yield %6 : tensor<10x10xf32>
}
```
The pattern was discussed in the RFC here: https://discourse.llvm.org/t/rfc-tensor-extracting-slices-from-tensor-collapse-shape/64034
Reviewed By: nicolasvasilache
Differential Revision: https://reviews.llvm.org/D129699
