LLVM makes several assumptions about address space 0. However,
alloca is presently constrained to always return this address space.
There's no real way to avoid using alloca, so without this
there is no way to opt out of these assumptions.
The problematic assumptions include:
- That the pointer size used for the stack is the same size as
the code size pointer, which is also the maximum sized pointer.
- That 0 is an invalid, non-dereferencable pointer value.
These are problems for AMDGPU because alloca is used to
implement the private address space, which uses a 32-bit
index as the pointer value. Other pointers are 64-bit
and behave more like LLVM's notion of generic address
space. By changing the address space used for allocas,
we can change our generic pointer type to be LLVM's generic
pointer type which does have similar properties.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@299888 91177308-0d34-0410-b5e6-96231b3b80d8
Summary:
This class is a list of AttributeSetNodes corresponding the function
prototype of a call or function declaration. This class used to be
called ParamAttrListPtr, then AttrListPtr, then AttributeSet. It is
typically accessed by parameter and return value index, so
"AttributeList" seems like a more intuitive name.
Rename AttributeSetImpl to AttributeListImpl to follow suit.
It's useful to rename this class so that we can rename AttributeSetNode
to AttributeSet later. AttributeSet is the set of attributes that apply
to a single function, argument, or return value.
Reviewers: sanjoy, javed.absar, chandlerc, pete
Reviewed By: pete
Subscribers: pete, jholewinski, arsenm, dschuff, mehdi_amini, jfb, nhaehnle, sbc100, void, llvm-commits
Differential Revision: https://reviews.llvm.org/D31102
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@298393 91177308-0d34-0410-b5e6-96231b3b80d8
Make opcode selection code for the load instruction a bit easier
to read and maintain.
This patch also catches number of f16 load/store variants that were
not handled before.
Differential Revision: https://reviews.llvm.org/D30513
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@296785 91177308-0d34-0410-b5e6-96231b3b80d8
Original code only used vector loads/stores for explicit vector arguments.
It could also do more loads/stores than necessary (e.g v5f32 would
touch 8 f32 values). Aggregate types were loaded one element at a time,
even the vectors contained within.
This change attempts to generalize (and simplify) parameter space
loads/stores so that vector loads/stores can be used more broadly.
Functionality of the patch has been verified by compiling thrust
test suite and manually checking the differences between PTX
generated by llvm with and without the patch.
General algorithm:
* ComputePTXValueVTs() flattens input/output argument into a flat list
of scalars to load/store and returns their types and offsets.
* VectorizePTXValueVTs() uses that data to create vectorization plan
which returns an array of flags marking boundaries of vectorized
load/stores. Scalars are represented as 1-element vectors.
* Code that generates loads/stores implements a simple state machine
that constructs a vector according to the plan.
Differential Revision: https://reviews.llvm.org/D30011
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@295784 91177308-0d34-0410-b5e6-96231b3b80d8
x*rsqrt(x) returns NaN for x == 0, whereas 1/rsqrt(x) returns 0, as
desired.
Verified that the particular nvptx approximate instructions here do in
fact return 0 for x = 0.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@293713 91177308-0d34-0410-b5e6-96231b3b80d8
Summary:
This lets us lower to sqrt.approx and rsqrt.approx under more
circumstances.
* Now we emit sqrt.approx and rsqrt.approx for calls to @llvm.sqrt.f32,
when fast-math is enabled. Previously, we only would emit it for
calls to @llvm.nvvm.sqrt.f. (With this patch we no longer emit
sqrt.approx for calls to @llvm.nvvm.sqrt.f; we rely on intcombine to
simplify llvm.nvvm.sqrt.f into llvm.sqrt.f32.)
* Now we emit the ftz version of rsqrt.approx when ftz is enabled.
Previously, we only emitted rsqrt.approx when ftz was disabled.
Reviewers: hfinkel
Subscribers: llvm-commits, tra, jholewinski
Differential Revision: https://reviews.llvm.org/D28508
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@293605 91177308-0d34-0410-b5e6-96231b3b80d8
- Move DEBUG_TYPE below includes
- Change unknown address space constant to be consistent with other
passes
- Grammar fixes in debug output
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@293567 91177308-0d34-0410-b5e6-96231b3b80d8
Different architectures can have different meaning for flags in the
SHF_MASKPROC mask, so we should always check what the architecture use
before checking the flag.
NFC for now, but will allow fixing the value of an xmos flag.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@293484 91177308-0d34-0410-b5e6-96231b3b80d8
Support for barrier synchronization between a subset of threads
in a CTA through one of sixteen explicitly specified barriers.
These intrinsics are not directly exposed in CUDA but are
critical for forthcoming support of OpenMP on NVPTX GPUs.
The intrinsics allow the synchronization of an arbitrary
(multiple of 32) number of threads in a CTA at one of 16
distinct barriers. The two intrinsics added are as follows:
call void @llvm.nvvm.barrier.n(i32 10)
waits for all threads in a CTA to arrive at named barrier #10.
call void @llvm.nvvm.barrier(i32 15, i32 992)
waits for 992 threads in a CTA to arrive at barrier #15.
Detailed description of these intrinsics are available in the PTX manual.
http://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions
Reviewers: hfinkel, jlebar
Differential Revision: https://reviews.llvm.org/D17657
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@293384 91177308-0d34-0410-b5e6-96231b3b80d8
This change introduces adjustPassManager target callback giving a
target an opportunity to tweak PassManagerBuilder before pass
managers are populated.
This generalizes and replaces addEarlyAsPossiblePasses target
callback. In particular that can be used to add custom passes to
extension points other than EP_EarlyAsPossible.
Differential Revision: https://reviews.llvm.org/D28336
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@293189 91177308-0d34-0410-b5e6-96231b3b80d8
Summary:
Specifically, we upgrade llvm.nvvm.:
* brev{32,64}
* clz.{i,ll}
* popc.{i,ll}
* abs.{i,ll}
* {min,max}.{i,ll,u,ull}
* h2f
These either map directly to an existing LLVM target-generic
intrinsic or map to a simple LLVM target-generic idiom.
In all cases, we check that the code we generate is lowered to PTX as we
expect.
These builtins don't need to be backfilled in clang: They're not
accessible to user code from nvcc.
Reviewers: tra
Subscribers: majnemer, cfe-commits, llvm-commits, jholewinski
Differential Revision: https://reviews.llvm.org/D28793
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@292694 91177308-0d34-0410-b5e6-96231b3b80d8
Summary:
DADToDAG has access to TargetLowering, but not vice versa, so this is
the more general location for these functions.
NFC
Reviewers: tra
Subscribers: jholewinski, llvm-commits
Differential Revision: https://reviews.llvm.org/D28795
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@292693 91177308-0d34-0410-b5e6-96231b3b80d8
Some instructions were printed as "foo\tbar", but most are printed as
"foo \bar". Standardize on the latter form.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@292306 91177308-0d34-0410-b5e6-96231b3b80d8
!strconcat is a variadic function; it will concatenate an arbitrary
number of strings. There's no need to nest it.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@292305 91177308-0d34-0410-b5e6-96231b3b80d8
Summary:
This change also lets us use max.{s,u}16. There's a vague warning in a
test about this maybe being less efficient, but I could not come up with
a case where the resulting SASS (sm_35 or sm_60) was different with or
without max.{s,u}16. It's true that nvcc seems to emit only
max.{s,u}32, but even ptxas 7.0 seems to have no problem generating
efficient SASS from max.{s,u}16 (the casts up to i32 and back down to
i16 seem to be implicit and nops, happening via register aliasing).
In the absence of evidence, better to have fewer special cases, emit
more straightforward code, etc. In particular, if a new GPU has 16-bit
min/max instructions, we want to be able to use them.
Reviewers: tra
Subscribers: jholewinski, llvm-commits
Differential Revision: https://reviews.llvm.org/D28732
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@292304 91177308-0d34-0410-b5e6-96231b3b80d8
Summary:
Avoid an unnecessary conversion operation when using the result of
ctpop.i32 or ctpop.i16 as an i32, as in both cases the ptx instruction
we run returns an i32.
(Previously if we used the value as an i32, we'd do an unnecessary
zext+trunc.)
Reviewers: tra
Subscribers: jholewinski, llvm-commits
Differential Revision: https://reviews.llvm.org/D28721
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@292302 91177308-0d34-0410-b5e6-96231b3b80d8
Summary:
* Disable "ctlz speculation", which inserts a branch on every ctlz(x) which
has defined behavior on x == 0 to check whether x is, in fact zero.
* Add DAG patterns that avoid re-truncating or re-expanding the result
of the 16- and 64-bit ctz instructions.
Reviewers: tra
Subscribers: llvm-commits, jholewinski
Differential Revision: https://reviews.llvm.org/D28719
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@292299 91177308-0d34-0410-b5e6-96231b3b80d8
Summary:
Previously there were three ways to inform the NVVMReflect pass whether
you wanted to flush denormals to zero:
* An LLVM command-line option
* Parameters to the NVVMReflect constructor
* Metadata on the module itself.
This change removes the first two, leaving only the third.
The motivation for this change, aside from simplifying things, is that
we want LLVM to be aware of whether it's operating in FTZ mode, so other
passes can use this information. Ideally we'd have a target-generic
piece of metadata on the module. This change moves us in that
direction.
Reviewers: tra
Subscribers: jholewinski, llvm-commits
Differential Revision: https://reviews.llvm.org/D28700
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@292068 91177308-0d34-0410-b5e6-96231b3b80d8
Only scalar half-precision operations are supported at the moment.
- Adds general support for 'half' type in NVPTX.
- fp16 math operations are supported on sm_53+ GPUs only
(can be disabled with --nvptx-no-f16-math).
- Type conversions to/from fp16 are supported on all GPU variants.
- On GPU variants that do not have full fp16 support (or if it's disabled),
fp16 operations are promoted to fp32 and results are converted back
to fp16 for storage.
Differential Revision: https://reviews.llvm.org/D28540
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@291956 91177308-0d34-0410-b5e6-96231b3b80d8
Previously we'd always lower @llvm.{sin,cos}.f32 to {sin.cos}.approx.f32
instruction even when unsafe FP math was not allowed.
Clang-generated IR is not affected by this as it uses precise sin/cos
from CUDA's libdevice when unsafe math is disabled.
Differential Revision: https://reviews.llvm.org/D28619
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@291936 91177308-0d34-0410-b5e6-96231b3b80d8
updated instructions:
pmulld, pmullw, pmulhw, mulsd, mulps, mulpd, divss, divps, divsd, divpd, addpd and subpd.
special optimization case which replaces pmulld with pmullw\pmulhw\pshuf seq.
In case if the real operands bitwidth <= 16.
Differential Revision: https://reviews.llvm.org/D28104
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@291657 91177308-0d34-0410-b5e6-96231b3b80d8
The range metadata inserted by NVVMIntrRange is pessimistic, range
metadata already present could be more precise.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@290294 91177308-0d34-0410-b5e6-96231b3b80d8
I've chosen to remove NVPTXInstrInfo::CanTailMerge but not
NVPTXInstrInfo::isLoadInstr and isStoreInstr (which are also dead)
because while the latter two are reasonably useful utilities, the former
cannot be used safely: It relies on successful address space inference
to identify writes to shared memory, but addrspace inference is a
best-effort thing.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@289740 91177308-0d34-0410-b5e6-96231b3b80d8
Summary:
Previously they were defined as a 2D char array in a header file. This
is kind of overkill -- we can let the linker lay out these strings
however it pleases. While we're at it, we might as well just inline
these constants where they're used, as each of them is used only once.
Also move NVPTXUtilities.{h,cpp} into namespace llvm.
Reviewers: tra
Subscribers: jholewinski, mgorny, llvm-commits
Differential Revision: https://reviews.llvm.org/D27636
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@289728 91177308-0d34-0410-b5e6-96231b3b80d8
Summary:
This has been replaced by the NVPTXInferAddressSpaces pass. We've had
the new one as the default with the old one accessible via a flag for
some months now, and we've had no problems.
Reviewers: tra
Subscribers: llvm-commits, jholewinski, jingyue, mgorny
Differential Revision: https://reviews.llvm.org/D26165
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@285642 91177308-0d34-0410-b5e6-96231b3b80d8
Summary:
In isel, transform
Num % Den
into
Num - (Num / Den) * Den
if the result of Num / Den is already available.
Reviewers: tra
Subscribers: hfinkel, llvm-commits, jholewinski
Differential Revision: https://reviews.llvm.org/D26090
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@285461 91177308-0d34-0410-b5e6-96231b3b80d8
