Conflicts:
lib/Conversion/TritonGPUToLLVM/TritonGPUToLLVMPass.cpp
lib/Target/LLVMIR/LLVMIRTranslation.cpp
python/test/unit/language/assert_helper.py
python/triton/third_party/cuda/bin/ptxas
test/Conversion/tritongpu_to_llvm.mlir
It looks like you may be committing a merge.
If this is not correct, please remove the file
.git/MERGE_HEAD
and try again.
Triton firstly prints assert message into stderr stream with the same
(refactored) helper function as `device_print` and then ends the thread
execution.
Note: s_endpgm instruction is used, since s_trap (generated from LLVM::Trap or LLVM::DebugTrap) has some issues on different HW.
Also got back fix in `python/triton/compiler/compiler.py` lost after one
of IFU.
```
at 10:18:def val_multiplier_noinline(val, i):
return val * i
^
Function val_multiplier_noinline is marked noinline, but was called with non-scalar argument val:fp32[constexpr[128]]
```
- Case 1: Return after static control flow is taken. Peel off
instructions after the first `return` for each basic block.
```python
if static_condition:
tl.store(...)
return
return
```
- Case 2: Return exists in both `if` and `else` branches of an inlined
`JITFunction` function
```python
def foo():
if dynamic_condition:
return a
else:
return b
```
- Case 3: Return exists in a `JITFunction` from another module
```python
import module
if cond:
a = module.func()
```
- Case 4: A chain of calls through undefined local variables
```python
import module
if cond:
a = x
a = a.to(tl.int32).to(tl.int32)
```
- Case 5: Call a function `func` without returning variables. `func` is
recognized as an `Expr` first instead of a `Call`.
```python
if cond:
foo()
else:
bar()
```
- Case 6: Call a `noinline` function. We don't need to check if the
function contains any return op.
This is a combination of 7 commits.
use pyt nightly with root
repro with pytorch unit test
hardcode isROCM to true
set is_cuda to False
ignore cc arg
clean up
match triton-mlir branch
Closes https://github.com/openai/triton/issues/1556https://github.com/openai/triton/issues/1512
The current hash used for caching the cubin does not include the
architecture. This leads to the following error when compiling against
one arch and running against another (with no code changes to trigger a
recompilation).
```
RuntimeError: Triton Error [CUDA]: device kernel image is invalid
```
Was not sure what unit tests would be appropriate here (if any)
Co-authored-by: davidma <davidma@speechmatics.com>
# Introducing the `noinline` Parameter for Triton JIT Decorator
We're excited to introduce a new parameter, `noinline`, that can be
added to the `jit` decorator in Triton. This parameter allows developers
to specify that a particular Triton function should not be inlined into
its callers. In this post, we'll dive into the syntax, purpose, and
implementation details of this new feature.
## Syntax
To use the `noinline` parameter, simply add `noinline=True` to the `jit`
decorator for the function that you don't want to be inlined. Here's an
example:
```python
@triton.jit(noinline=True)
def device_fn(x, y, Z):
z = x + y
tl.store(Z, z)
def test_noinline():
@triton.jit
def kernel(X, Y, Z):
x = tl.load(X)
y = tl.load(Y)
device_fn(x, y, Z)
```
In this example, the `device_fn` function is decorated with
`@triton.jit(noinline=True)`, indicating that it should not be inlined
into its caller, `kernel`.
## Purpose
The `noinline` parameter serves several key purposes:
- Reducing code size: By preventing inlining, we can reduce the size of
the compiled code.
- Facilitating debugging: Keeping functions separate can make it easier
to debug the code.
- Avoiding common subexpression elimination (CSE) in certain cases: CSE
can sometimes be avoided by using the `noinline` parameter to reduce
register pressure.
- Enabling dynamic linking: This parameter makes it possible to
dynamically link Triton functions.
## Implementation
The implementation of the `noinline` parameter involves significant
changes to three analysis modules in Triton: *Allocation*, *Membar*, and
*AxisInfo*. Prior to this update, these modules assumed that all Triton
functions had been inlined into the root kernel function. With the
introduction of non-inlined functions, we've had to rework these
assumptions and make corresponding changes to the analyses.
### Call Graph and Limitations
<div style="text-align: center;">
<img
src="https://user-images.githubusercontent.com/2306281/234663904-12864247-3412-4405-987b-6991cdf053bb.png"
alt="figure 1" width="200" height="auto">
</div>
To address the changes, we build a call graph and perform all the
analyses on the call graph instead of a single function. The call graph
is constructed by traversing the call edges and storing them in an edge
map. Roots are extracted by checking nodes with no incoming edges.
The call graph has certain limitations:
- It does not support recursive function calls, although this could be
implemented in the future.
- It does not support dynamic function calls, where the function name is
unknown at compilation time.
### Allocation
<div style="text-align: center;">
<img
src="https://user-images.githubusercontent.com/2306281/234665110-bf6a2660-06fb-4648-85dc-16429439e72d.png"
alt="figure 2" width="400" height="auto">
</div>
In Triton, shared memory allocation is achieved through two operations:
`triton_gpu.convert_layout` and `triton_gpu.alloc_tensor`. The
`convert_layout` operation allocates an internal tensor, which we refer
to as a *scratch* buffer, while the `alloc_tensor` operation returns an
allocated tensor and is thus known as an *explicit* buffer.
To accommodate the introduction of function calls, we are introducing a
third type of buffer called a *virtual* buffer. Similar to scratch
buffers, virtual buffers are allocated internally within the scope of a
function call, and the buffers allocated by the called functions remain
invisible to subsequent operations in the calling function. However,
virtual buffers are distinct from scratch buffers in that the call
operation itself does not allocate memory—instead, it specifies the
total amount of memory required by all the child functions being called.
The actual allocation of buffers is performed by individual operations
within these child functions. For example, when invoking edge e1, no
memory is allocated, but the total amount of memory needed by function B
is reserved. Notably, the amount of shared memory used by function B
remains fixed across its call sites due to the consideration of dynamic
control flows within each function.
An additional challenge to address is the calculation of shared memory
offsets for functions within a call graph. While we can assume a shared
memory offset starting at 0 for a single root function, this is not the
case with a call graph, where we must determine each function's starting
offset based on the call path. Although each function has a fixed memory
consumption, the starting offset may vary. For instance, in Figure 2,
the starting offset of function C through edges e1->e2 differs from that
through edges e2->e4. To handle this, we accumulate the starting offset
at each call site and pass it as an argument to the called function.
Additionally, we amend both the function declaration and call sites by
appending an offset variable.
### Membar
<div style="text-align: center;">
<img
src="https://user-images.githubusercontent.com/2306281/234665157-844dd66f-5028-4ef3-bca2-4ca74b8f969d.png"
alt="figure 3" width="300" height="auto">
</div>
The membar pass is dependent on the allocation analysis. Once the offset
and size of each buffer are known, we conduct a post-order traversal of
the call graph and analyze each function on an individual basis. Unlike
previous analyses, we now return buffers that remain unsynchronized at
the end of functions, allowing the calling function to perform
synchronization in cases of overlap.
### AxisInfo
<div style="text-align: center;">
<img
src="https://user-images.githubusercontent.com/2306281/234665183-790a11ac-0ba1-47e1-98b1-e356220405a3.png"
alt="figure 4" width="400" height="auto">
</div>
The AxisInfo analysis operates differently from both membar and
allocation, as it traverses the call graph in topological order. This is
necessary because function arguments may contain axis information that
will be utilized by callee functions. As we do not implement
optimizations like function cloning, each function has a single code
base, and the axis information for an argument is determined as a
conservative result of all axis information passed by the calling
functions.
---------
Co-authored-by: Philippe Tillet <phil@openai.com>
Change the usage of LRU cache decorator from @functools.lru_cache to
@functools.lru_cache().
The former raises an error TypeError('Expected maxsize to be an integer
or None' for Python 3.7 or older.
When running python 3.8, I've found that process creation gets slower
over time (e.g. after creating a CUDA context, it can take 50-300ms per
subprocess.run), and we do one of these calls to `ptxas --version` for
every kernel, so a model with thousands of kernels can end up spending
substantial time just calling ptxas redundantly.
Co-authored-by: Philippe Tillet <phil@openai.com>
MLIR current only supports a custom inlining interface per dialect, so
we cannot change the inlining decision of `func.func`.
https://discourse.llvm.org/t/avoid-inlining-some-functions-using-the-func-dialect/69830/3
Could revert it back once they've designed a better inliner interface.
Inlining attributes will be implemented in the next PR since this PR is
already huge.
The changes here come with a few separate bits:
- Allow replacing the cache manager with an ENV variable to make it
pluggable
- Make the `make_path` api private since its leaking some internal bits
of the cache and allowing file access. Use a get operation instead.
- For the `compile` operation we have a several files part of a single
compile pipeline that are small, this can be not the most performant
with remote caches. Also some operations like
`_triton.get_shared_memory_size` only work when everything is cached or
none(or some key ones aren't). They segfault otherwise. So grouping
these as an entity avoids that.
The purpose of this PR is to remove some circular dependencies and
separate concerns better in the frontend. It's still not perfect --
`triton.compile` still includes a few runtime architecture-specific
component, but at least much better than before.
This PR still assumes that AMD only supports empty kernels right now.
Other PRs will follow to make the frontend supports multiple devices in
a more modular way.
