<git-pr-chain>
#### Commits in this PR
1. Fix segfault in assertion test.
The issue here is that we were not checking the return values of the
CUDA API
calls we were making. We call one function and then use the data it
returns as
input to another call. Obviously this doesn't work if the first call
returns
an error and doesn't actually return meaningful data.
I don't know why this was passing in CI, but it failed consistently for
me.
#### [PR chain](https://github.com/jlebar/git-pr-chain)
1. 👉#2520👈 **YOU ARE HERE**
</git-pr-chain>
<git-pr-chain>
#### Commits in this PR
1. Make kernel_static_print test work when called twice.
This test is checking that a message is printed when the kernel is
compiled.
But the test had nothing to force the kernel to be compiled every time
you ran
the test. So after you ran it once, the test would fail every time until
you
cleared the cache.
#### [PR chain](https://github.com/jlebar/git-pr-chain)
1. 👉#2518👈 **YOU ARE HERE**
1. #2520
</git-pr-chain>
I noticed that Triton is using the `ptxas` version as part of the
version hash even for non-CUDA targets. This is an attempt at fixing
this. Moving the version calculation to the back-end makes sense to me
from an architectural standpoint, so that's my approach here. I'm not as
confident in the implementation, so please if folks have any feedback
let me know.
Without this change, a constexpr assignment (ie. `A = B & C`, where `B`
and `C` are both constexpr) is getting assigned to a triton tensor,
which becomes an issue when `A` is used as the condition of an If
statement.
Note: I had to add `not isinstance(node.value, ast.Constant)` to the
condition because if we are assigning `x = 0` then the assigned value is
also a constexpr, but in this case we do want to assign a triton tensor
to `x` so that we can do `x.to(tl.int64)` for example, which cannot be
done on a constexpr.
---------
Co-authored-by: Philippe Tillet <phil@openai.com>
By default, ptxas will enable fusion of mul/add to fma instructions. The
backend was also being configured unconditionally to enable this on
conversion from LLVM IR to PTX. This commit adds an option which can be
used to disable the FP fusion behavior in both locations.
In current implementation, warpsPerCTA is always set to [numWarps, 1]
for 2 tt.dot fusion scenario. But, it is not optimal for cases such that
tt.dot doesn't have enough parallelism on row dimension but on column
dimension.
clipping float8e4b15 to +-1.875 is a bad idea because these are
represented as 0x7f and 0xff, which are +- nan on H100 for float8e4nv.
We lose two values but this will make compatibility with float8e4nv way
less painful. (it will just be a matter of adjusting the bias)
Replace a single
mma.sync.aligned.m16n8k32.row.col.satfinite.s32.s8.s8.s32 instruction
that is used on Ampere with 4 x
mma.sync.aligned.m8n8k16.row.col.satfinite.s32.s8.s8.s32 instructions
for Turing
Extracted the Turing-int8, Turing-fp16 and Ampere to separate functions.
Somehow I messed up with my previous PR, so just open a new one.
---------
Co-authored-by: Philippe Tillet <phil@openai.com>
This fixes a few bugs I've encountered
- `atomic_add` with int64/uint64 `Operation .add requires .u32 or .s32
or .u64 [...] for instruction 'atom'`
- `atomic_min/max` with float64 -> `ValueError('Cannot bitcast data-type
of size 64 to data-type of size 32')`
- `atomic_min/max` with float32 returns the old value as int32
1. On the axis, using `getAxisNumWarpsWithUniqueData` instead of getting
the raw number of warps to avoid communication among warps that handle
the same piece of data.
2. When there's a single warp on the axis, using warp Intrinsics for
communication and skip shared memory.
Need a follow up PR for code clean up.
Change the dot to allow taking an initial accumulator and add a flag
that will allow the compiler to accumulate in a lower precision than the
output type.
On Hopper this flag is on by default which allows accumualting with
lower precision.
This only affect Hopper fp8 dot.
…rf on problems that need few blocks.
constrain the number of launched blocks to what it exactely needs for
persistent warp specialized kernel. It's useful when problems need very
few blocks.
e.g. MxNxK=800x800x60000, f16_f16_f32, block size=128x128x64,
non-split-k. Experiments show it can achieve ~16% speedup.
This fixes a few bugs related to scalar tensors:
- `tl.full([], fill_value, dtype)` fails with `TypeError('0d block_type
is forbidden')`
- `scalar[None]` fails with `TypeError("'constexpr' object is not
iterable")`
- `scalar[None, None]` fails with `AttributeError("'dtype' object has no
attribute 'shape'")`
- `scalar.shape` returns `[1]` instead of 0-dim `[]`
- Also related, `tl.zeros_like(scalar)` returns a 1d tensor instead of
another scalar
Prior to this PR, matmul on sm_89 (RTX 4070)
(`test/unit/operators/test_matmul.py::test_op`) would result in test
failure due to too strict atol/rtol.
To avoid having to choose strictness ourselves, and to have better
defaults based on dtype, use the non-deprecated torch testing util.
See: https://github.com/pytorch/pytorch/issues/61844
Replace: https://github.com/openai/triton/pull/2242
* this pr adds a third party backend for triton that works on AMD
* this expose a lot of the work that has been done in our
[fork](https://github.com/ROCmSoftwarePlatform/triton)
* most unit tests on `test_core.py` pass
* it skips some unit tests for various reasons
* we plan to follow up with more prs improving Functionality and
Performance in the future
---------
Co-authored-by: Philippe Tillet <phil@openai.com>
