The GPU ID operations already implement InferIntRangeInterface, which
gives constant lower and upper bounds on those IDs when appropriate
metadata is prentent on the operations or in the surrounding context.
This commit uses that existing code to implement the
ValueBoundsOpInterface, which is used when analyzing affine operations
(unlike the integer range interface, which is used for arithmetic
optimization).
It also implements the interface for gpu.launch, where we can use it to
express the constraint that block/grid sizes are equal to their value
from outside the launch op and that the corresponding IDs are bounded
above by that size.
As a consequence, the test pass for this inference is updated to work on
a FunctionOpInterface and not a func.func, creating minor churn in other
tests.
The `parseX()` functions that are defined to support `custom<X>` in
`assemblyFormat` should return `ParseResult` rather than
`LogicalResult`. The `ParseResult` type is necessary due to tablegen
generating code that expects this type within an Op `parseX()` function.
This change allows to expose through an interface attributes wrapping
content as external resources, and the usage inside the ModuleToObject
show how we will be able to provide runtime libraries without relying on
the filesystem.
This is a follow-up of #117246.
I thought then it would be easy to edit a DictionaryAttr but it turns
out that these attributes are immutable and need to be passed during the
construction of the gpu.binary Op.
The first commit was using the NVVMTargetAttr to pass the information.
After feedback from @fabianmcg, this PR now passes the information
through a new option of the gpu-module-to-binary pass.
Please add reviewers, as you see fit.
Here is the [merged
MR](https://github.com/llvm/llvm-project/pull/116007) which caused a
failure and [was
reverted](https://github.com/llvm/llvm-project/pull/116811).
Thanks to @joker-eph for the help, I fix it (miss constructing
`ModuleObject` with callback functions in
`mlir/lib/Target/LLVM/NVVM/Target.cpp`) and split unit tests from origin
test which don't need `ptxas` to make the test runs more widely.
Updates the return type of `getNumDynamicDims` and `getNumScalableDims`
from `int64_t` to `size_t`. This is for consistency with other
helpers/methods that return "size" and to reduce the number of
`static_cast`s in various places.
This patch adds the `#gpu.kernel_metadata` and `#gpu.kernel_table`
attributes. The `#gpu.kernel_metadata` attribute allows storing metadata
related to a compiled kernel, for example, the number of scalar
registers used by the kernel. The attribute only has 2 required
parameters, the name and function type. It also has 2 optional
parameters, the arguments attributes and generic dictionary for storing
all other metadata.
The `#gpu.kernel_table` stores a table of `#gpu.kernel_metadata`,
mapping the name of the kernel to the metadata.
Finally, the function `ROCDL::getAMDHSAKernelsELFMetadata` was added to
collect ELF metadata from a binary, and to test the class methods in
both attributes.
Example:
```mlir
gpu.binary @binary [#gpu.object<#rocdl.target<chip = "gfx900">, kernels = #gpu.kernel_table<[
#gpu.kernel_metadata<"kernel0", (i32) -> (), metadata = {sgpr_count = 255}>,
#gpu.kernel_metadata<"kernel1", (i32, f32) -> (), arg_attrs = [{llvm.read_only}, {}]>
]> , bin = "BLOB">]
```
The motivation behind these attributes is to provide useful information
for things like tunning.
---------
Co-authored-by: Mehdi Amini <joker.eph@gmail.com>
This enables performing several reductions in parallel, each smaller
than the size of the subgroup.
One potential application is flash attention with subgroup-wide matrix
multiplication and reduction combined in one kernel. The multiplication
operation requires a 2D matrix to be distributed over the lanes of the
subgroup, which then constrains the shape the following reduction can
have if we want to keep data in registers.
- Replace hand-written parser/printer with auto-generated assembly
format.
- Remove implicit `gpu.module_end` terminator and use the `NoTerminator`
trait instead. (Same as `builtin.module`.)
- Turn the region into a graph region. (Same as `builtin.module`.)
When a type/attribute is defined in TableGen, a type constraint can be
used for parameters, but the type constraint verification was missing.
Example:
```
def TestTypeVerification : Test_Type<"TestTypeVerification"> {
let parameters = (ins AnyTypeOf<[I16, I32]>:$param);
// ...
}
```
No verification code was generated to ensure that `$param` is I16 or
I32.
When type constraints a present, a new method will generated for types
and attributes: `verifyInvariantsImpl`. (The naming is similar to op
verifiers.) The user-provided verifier is called `verify` (no change).
There is now a new entry point to type/attribute verification:
`verifyInvariants`. This function calls both `verifyInvariantsImpl` and
`verify`. If neither of those two verifications are present, the
`verifyInvariants` function is not generated.
When a type/attribute is not defined in TableGen, but a verifier is
needed, users can implement the `verifyInvariants` function. (This
function was previously called `verify`.)
Note for LLVM integration: If you have an attribute/type that is not
defined in TableGen (i.e., just C++), you have to rename the
verification function from `verify` to `verifyInvariants`. (Most
attributes/types have no verification, in which case there is nothing to
do.)
Depends on #102657.
This change reworks how range information for GPU dispatch IDs (block
IDs, thread IDs, and so on) is handled.
1. `known_block_size` and `known_grid_size` become inherent attributes
of GPU functions. This makes them less clunky to work with. As a
consequence, the `gpu.func` lowering patterns now only look at the
inherent attributes when setting target-specific attributes on the
`llvm.func` that they lower to.
2. At the same time, `gpu.known_block_size` and `gpu.known_grid_size`
are made official dialect-level discardable attributes which can be
placed on arbitrary functions. This allows for progressive lowerings
(without this, a lowering for `gpu.thread_id` couldn't know about the
bounds if it had already been moved from a `gpu.func` to an `llvm.func`)
and allows for range information to be provided even when
`gpu.*_{id,dim}` are being used outside of a `gpu.func` context.
3. All of these index operations have gained an optional `upper_bound`
attribute, allowing for an alternate mode of operation where the bounds
are specified locally and not inherited from the operation's context.
These also allow handling of cases where the precise launch sizes aren't
known, but can be bounded more precisely than the maximum of what any
platform's API allows. (I'd like to thank @benvanik for pointing out
that this could be useful.)
When inferring bounds (either for range inference or for setting `range`
during lowering) these sources of information are consulted in order of
specificity (`upper_bound` > inherent attribute > discardable attribute,
except that dimension sizes check for `known_*_bounds` to see if they
can be constant-folded before checking their `upper_bound`).
This patch also updates the documentation about the bounds and inference
behavior to clarify what these attributes do when set and the
consequences of setting them up incorrectly.
---------
Co-authored-by: Mehdi Amini <joker.eph@gmail.com>
This patch adds a builder to `gpu.launch_func` allowing it to be created
using `SymbolRefAttr` instead of `GPUFuncOp`. This allows creating
`launch_func` when only a `gpu.binary` is present, instead of the full
`gpu.module {...}`.
Make `sym_name` an inherent attr in GPUModuleOp so that it doesn't show
in the discardable attributes.
The change is safe as the attribute is always expected to be present.
I'm planning to remove StringRef::equals in favor of
StringRef::operator==.
- StringRef::operator==/!= outnumber StringRef::equals by a factor of
10 under mlir/ in terms of their usage.
- The elimination of StringRef::equals brings StringRef closer to
std::string_view, which has operator== but not equals.
- S == "foo" is more readable than S.equals("foo"), especially for
!Long.Expression.equals("str") vs Long.Expression != "str".
This PR adds promised interface declarations for all interfaces declared
in `InitAllDialects.h`.
Promised interfaces allow a dialect to declare that it will have an
implementation of a particular interface, crashing the program if one
isn't provided when the interface is used.
This commit fixes memory leaks that were introduced by #81759. The way
ops and blocks are erased changed slightly.
The leaks were caused by an incorrect implementation of op builders:
blocks must be created with the supplied builder object. Otherwise, they
are not properly tracked by the dialect conversion and can leak during
rollback.
This commit renames 4 pattern rewriter API functions:
* `updateRootInPlace` -> `modifyOpInPlace`
* `startRootUpdate` -> `startOpModification`
* `finalizeRootUpdate` -> `finalizeOpModification`
* `cancelRootUpdate` -> `cancelOpModification`
The term "root" is a misnomer. The root is the op that a rewrite pattern
matches against
(https://mlir.llvm.org/docs/PatternRewriter/#root-operation-name-optional).
A rewriter must be notified of all in-place op modifications, not just
in-place modifications of the root
(https://mlir.llvm.org/docs/PatternRewriter/#pattern-rewriter). The old
function names were confusing and have contributed to various broken
rewrite patterns.
Note: The new function names use the term "modify" instead of "update"
for consistency with the `RewriterBase::Listener` terminology
(`notifyOperationModified`).
This patch adds an optional offloading handler attribute to
the`gpu.module` op. This attribute will be used during
`gpu-module-to-binary` pass to override the offloading handler used in
the `gpu.binary` op.
Each vector element is reduced independently, which is a form of
multi-reduction.
The plan is to allow for gradual lowering of multi-reduction that
results in fewer `gpu.shuffle` ops at the end:
1d `vector.multi_reduction` --> 1d `gpu.subgroup_reduce` --> smaller 1d
`gpu.subgroup_reduce` --> packed `gpu.shuffle` over i32
For example we can perform 2 independent f16 reductions with a series of
`gpu.shuffles` over i32, reducing the final number of `gpu.shuffles` by 2x.
This is to avoid confusion when dealing with reduction/combining kinds.
For example, see a recent PR comment:
https://github.com/llvm/llvm-project/pull/75846#discussion_r1430722175.
Previously, they were picked to mostly mirror the names of the llvm
vector reduction intrinsics:
https://llvm.org/docs/LangRef.html#llvm-vector-reduce-fmin-intrinsic. In
isolation, it was not clear if `<maxf>` has `arith.maxnumf` or
`arith.maximumf` semantics. The new reduction kind names map 1:1 to
arith ops, which makes it easier to tell/look up their semantics.
Because both the vector and the gpu dialect depend on the arith dialect,
it's more natural to align names with those in arith than with the
lowering to llvm intrinsics.
Issue: https://github.com/llvm/llvm-project/issues/72354
The motivation for this change is explained in
https://github.com/llvm/llvm-project/issues/72354.
Before this change, we could not tell between signed/unsigned
minimum/maximum and NaN treatment for floating point values.
The mapping of old reduction operations to the new ones is as follows:
* `min` --> `minsi` for ints, `minf` for floats
* `max` --> `maxsi` for ints, `maxf` for floats
New reduction kinds not represented in the old enum: `minui`, `maxui`,
`minimumf`, `maximumf`.
As a next step, I would like to have a common definition of combining
kinds used by the `vector` and `gpu` dialects. Separately, the GPU to
SPIR-V lowering does not yet properly handle zero and NaN values -- the
behavior of floating point min/max group reductions is not specified by
the SPIR-V spec, see https://github.com/llvm/llvm-project/issues/73459.
Issue: https://github.com/llvm/llvm-project/issues/72354
NVIDIA Hopper architecture introduced the Cooperative Group Array (CGA).
It is a new level of parallelism, allowing clustering of Cooperative
Thread Arrays (CTA) to synchronize and communicate through shared memory
while running concurrently.
This PR enables support for CGA within the `gpu.launch_func` in the GPU
dialect. It extends `gpu.launch_func` to accommodate this functionality.
The GPU dialect remains architecture-agnostic, so we've added CGA
functionality as optional parameters. We want to leverage mechanisms
that we have in the GPU dialects such as outlining and kernel launching,
making it a practical and convenient choice.
An example of this implementation can be seen below:
```
gpu.launch_func @kernel_module::@kernel
clusters in (%1, %0, %0) // <-- Optional
blocks in (%0, %0, %0)
threads in (%0, %0, %0)
```
The PR also introduces index and dimensions Ops specific to clusters,
binding them to NVVM Ops:
```
%cidX = gpu.cluster_id x
%cidY = gpu.cluster_id y
%cidZ = gpu.cluster_id z
%cdimX = gpu.cluster_dim x
%cdimY = gpu.cluster_dim y
%cdimZ = gpu.cluster_dim z
```
We will introduce cluster support in `gpu.launch` Op in an upcoming PR.
See [the
documentation](https://docs.nvidia.com/cuda/parallel-thread-execution/index.html#cluster-of-cooperative-thread-arrays)
provided by NVIDIA for details.
While the `gpu.launch` Op allows setting the size via the
`dynamic_shared_memory_size` argument, accessing the dynamic shared
memory is very convoluted. This PR implements the proposed Op,
`gpu.dynamic_shared_memory` that aims to simplify the utilization of
dynamic shared memory.
RFC:
https://discourse.llvm.org/t/rfc-simplifying-dynamic-shared-memory-access-in-gpu/
**Proposal from RFC**
This PR `gpu.dynamic.shared.memory` Op to use dynamic shared memory
feature efficiently. It is is a powerful feature that enables the
allocation of shared memory at runtime with the kernel launch on the
host. Afterwards, the memory can be accessed directly from the device. I
believe similar story exists for AMDGPU.
**Current way Using Dynamic Shared Memory with MLIR**
Let me illustrate the challenges of using dynamic shared memory in MLIR
with an example below. The process involves several steps:
- memref.global 0-sized array LLVM's NVPTX backend expects
- dynamic_shared_memory_size Set the size of dynamic shared memory
- memref.get_global Access the global symbol
- reinterpret_cast and subview Many OPs for pointer arithmetic
```
// Step 1. Create 0-sized global symbol. Manually set the alignment
memref.global "private" @dynamicShmem : memref<0xf16, 3> { alignment = 16 }
func.func @main() {
// Step 2. Allocate shared memory
gpu.launch blocks(...) threads(...)
dynamic_shared_memory_size %c10000 {
// Step 3. Access the global object
%shmem = memref.get_global @dynamicShmem : memref<0xf16, 3>
// Step 4. A sequence of `memref.reinterpret_cast` and `memref.subview` operations.
%4 = memref.reinterpret_cast %shmem to offset: [0], sizes: [14, 64, 128], strides: [8192,128,1] : memref<0xf16, 3> to memref<14x64x128xf16,3>
%5 = memref.subview %4[7, 0, 0][7, 64, 128][1,1,1] : memref<14x64x128xf16,3> to memref<7x64x128xf16, strided<[8192, 128, 1], offset: 57344>, 3>
%6 = memref.subview %5[2, 0, 0][1, 64, 128][1,1,1] : memref<7x64x128xf16, strided<[8192, 128, 1], offset: 57344>, 3> to memref<64x128xf16, strided<[128, 1], offset: 73728>, 3>
%7 = memref.subview %6[0, 0][64, 64][1,1] : memref<64x128xf16, strided<[128, 1], offset: 73728>, 3> to memref<64x64xf16, strided<[128, 1], offset: 73728>, 3>
%8 = memref.subview %6[32, 0][64, 64][1,1] : memref<64x128xf16, strided<[128, 1], offset: 73728>, 3> to memref<64x64xf16, strided<[128, 1], offset: 77824>, 3>
// Step.5 Use
"test.use.shared.memory"(%7) : (memref<64x64xf16, strided<[128, 1], offset: 73728>, 3>) -> (index)
"test.use.shared.memory"(%8) : (memref<64x64xf16, strided<[128, 1], offset: 77824>, 3>) -> (index)
gpu.terminator
}
```
Let’s write the program above with that:
```
func.func @main() {
gpu.launch blocks(...) threads(...) dynamic_shared_memory_size %c10000 {
%i = arith.constant 18 : index
// Step 1: Obtain shared memory directly
%shmem = gpu.dynamic_shared_memory : memref<?xi8, 3>
%c147456 = arith.constant 147456 : index
%c155648 = arith.constant 155648 : index
%7 = memref.view %shmem[%c147456][] : memref<?xi8, 3> to memref<64x64xf16, 3>
%8 = memref.view %shmem[%c155648][] : memref<?xi8, 3> to memref<64x64xf16, 3>
// Step 2: Utilize the shared memory
"test.use.shared.memory"(%7) : (memref<64x64xf16, 3>) -> (index)
"test.use.shared.memory"(%8) : (memref<64x64xf16, 3>) -> (index)
}
}
```
This PR resolves#72513
This patch adds an NVPTX compilation path that enables JIT compilation
on NVIDIA targets. The following modifications were performed:
1. Adding a format field to the GPU object attribute, allowing the
translation attribute to use the correct runtime function to load the
module. Likewise, a dictionary attribute was added to add any possible
extra options.
2. Adding the `createObject` method to `GPUTargetAttrInterface`; this
method returns a GPU object from a binary string.
3. Adding the function `mgpuModuleLoadJIT`, which is only available for
NVIDIA GPUs, as there is no equivalent for AMD.
4. Adding the CMake flag `MLIR_GPU_COMPILATION_TEST_FORMAT` to specify
the format to use during testing.
This patch adds the option of building an optional symbol table for the
top operation in the `gpu-module-to-binary` pass. The table is not
created by default as most targets don't need it; instead, it is lazily
built. The table is passed through a callback in `TargetOptions`.
This patch is required to integrate #65539 .
This patch employs the updated promise mechanism to enforce Target
Attribute IR constraints. Due to this patch, TargetAttributes
implementations no longer have to be registered before executing
translation to LLVM IR in cases where they are not needed, like when
translating `gpu.binary` operations.
`TargetOptions` includes a field for passing additional command line options to
the GPU compilation process. This field is typically used during the 'gpu-module-to-binary`
pass:
```
--gpu-module-to-binary=opts="-v -c"
```
The problem is that `tokenizeCmdOptions` receives the quoted string, which produces an
incorrect tokenization for `"-v -c"`. This patch removes quotes, fixing this issue.
Reviewed By: mehdi_amini
Differential Revision: https://reviews.llvm.org/D159434
Functions are always callable operations and thus every operation
implementing the `FunctionOpInterface` also implements the
`CallableOpInterface`. The only exception was the FuncOp in the toy
example. To make implementation of the `FunctionOpInterface` easier,
this commit lets `FunctionOpInterface` inherit from
`CallableOpInterface` and merges some of their methods. More precisely,
the `CallableOpInterface` has methods to get the argument and result
attributes and a method to get the result types of the callable region.
These methods are always implemented the same way as their analogues in
`FunctionOpInterface` and thus this commit moves all the argument and
result attribute handling methods to the callable interface as well as
the methods to get the argument and result types. The
`FuntionOpInterface` then does not have to declare them as well, but
just inherits them from the `CallableOpInterface`.
Adding the inheritance relation also required to move the
`FunctionOpInterface` from the IR directory to the Interfaces directory
since IR should not depend on Interfaces.
Reviewed By: jpienaar, springerm
Differential Revision: https://reviews.llvm.org/D157988
**For an explanation of these patches see D154153.**
Commit message:
This patch adds the default offloading handler for GPU binary ops: `#gpu.select_object`,
it selects the object to embed based on an index or a target attribute, embedding
the object as a global string and launches the kernel using the scheme used in the
GPU to LLVM pass.
Depends on D154137
Reviewed By: mehdi_amini
Differential Revision: https://reviews.llvm.org/D154147
**For an explanation of these patches see D154153.**
Commit message:
In order to lower `gpu.launch_func` after running `gpu-to-llvm` it must be
able to handle lowered types -eg. index -> i64. This patch also allows the op
to refer to GPU binaries and not only GPU modules.
Depends on D154132.
Reviewed By: mehdi_amini
Differential Revision: https://reviews.llvm.org/D154137
**For an explanation of these patches see D154153.**
Commit message:
Adds the `#gpu.object` attribute for holding a binary object and the target
attribute used to create it. Also adds the `gpu.binary` operation used to
store GPU objects.
Depends on D154108
Reviewed By: mehdi_amini
Differential Revision: https://reviews.llvm.org/D154132
**For an explanation of these patches see D154153.**
Commit message:
Adds support for Target attributes in GPU modules. This change enables attaching
an optional non empty array of GPU target attributes to the module.
Depends on D154104
Reviewed By: mehdi_amini
Differential Revision: https://reviews.llvm.org/D154113
**For an explanation of these patches see D154153.**
Commit message:
This patch adds the `GPUTargetAttrInterface` attribute interface, this interface
is meant to be used as an opaque interface for serializing GPU modules into
binary strings.
Reviewed By: mehdi_amini, krzysz00
Differential Revision: https://reviews.llvm.org/D154104
This revision significantly simplifies the specification and implementation of mapping loops to GPU ids.
Each type of mapping (block, warpgroup, warp, thread) now comes with 2 mapping modes:
1. a 3-D "grid-like" mode, subject to alignment considerations on threadIdx.x, on which predication
may occur on a per-dimension 3-D sub-rectangle basis.
2. a n-D linearized mode, on which predication may only occur on a linear basis.
In the process, better size and alignment requirement inference are introduced along with improved runtime verification messages.
The `warp_dims` attribute was deemed confusing and is removed from the transform in favor of better size inference.
Differential Revision: https://reviews.llvm.org/D155941
