## Motivation
The staging compiler enables lifetime-safety warnings and we already
worked around a few of them.
This works around a few more instances that came up recently on gfx950
builds.
The initial PR that resolved most issues:
https://github.com/ROCm/composable_kernel/pull/3640
## Technical Details
This follows the pattern to locally ignore the newly added
lifetime-safety warnings that were moved from experimental to production
in upstream LLVM.
As a result, CK turned them on and treats them as errors, which prevents
the staging compiler from building CK.
## Test Plan
## Test Result
## Submission Checklist
- [ ] Look over the contributing guidelines at
https://github.com/ROCm/ROCm/blob/develop/CONTRIBUTING.md#pull-requests.
Co-authored-by: Illia Silin <98187287+illsilin@users.noreply.github.com>
## Proposed changes
add memsetasync for ck moe splitk to fix
## Checklist
Please put an `x` into the boxes that apply. You can also fill these out
after creating the PR. If you're not sure, please don't hesitate to ask.
- [ ] I have added tests relevant to the introduced functionality, and
the unit tests are passing locally
- [ ] I have added the test to REGRESSION_TESTS list defined at the top
of CMakeLists.txt in tests/CMakeLists.txt, **IF** the test takes more
than 30 seconds to run.
- [ ] I have added inline documentation which enables the maintainers
with understanding the motivation
- [ ] I have removed the stale documentation which is no longer relevant
after this pull request
- [ ] (If this change is user-facing) I have added release notes which
provide the end users with a brief summary of the improvement from this
pull request
- [ ] I have run `clang-format` on all changed files
- [ ] Any dependent changes have been merged
## Discussion
If this is a relatively large or complex change, feel free to start a
discussion by explaining why you chose the solution you did and what
alternatives you considered
---
🔁 Imported from
[ROCm/composable_kernel#3630](https://github.com/ROCm/composable_kernel/pull/3630)
🧑💻 Originally authored by @lalala-sh
---------
Co-authored-by: lalala-sh <Jiaxing.Wen@amd.com>
Co-authored-by: systems-assistant[bot] <systems-assistant[bot]@users.noreply.github.com>
Co-authored-by: Illia Silin <98187287+illsilin@users.noreply.github.com>
## Motivation
Recursive template implementations might initially seem attractive to
minimize necessary coding.
Unfortunately, this style is often affects readability and requires
significant resources from the compiler to generate instantiation
chains. In "high-traffic" code (e.g., used in many places + compilation
units), this generally does not scale well and can bloat the overall
compile times to unnecessary lengths.
The aim of this PR is to take some of most high-traffic utility code and
try our best to eliminate recursive templates in favor of fold
expansions and constexpr function helpers.
In local tests with clang build analyzer,
device_grouped_conv2d_fwd_xdl_ngchw_gkcyx_ngkhw_f16_16x16_instance.cpp
showed high hit-rates on slow template instantiations in static_for,
dimensional static_for (static_ford), which are subsequently affected by
implementation of the Sequence class and associated transforms.
Example:
**** Templates that took longest to instantiate:
70111 ms: ck::detail::applier<int, 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11,
12, 1... (372 times, avg 188 ms) // **70 seconds!**
The above is part of the implementation of static_for which uses
Sequence classes..
## Technical Details
### Summary of Optimization Techniques
| Technique | Used In | Benefit |
|-----------|---------|---------|
| __Constexpr for-loop computation__ | sequence_reverse_inclusive_scan,
sequence_map_inverse | Moves O(N) work from template instantiation to
constexpr evaluation |
| __Pack expansion with indexing__ | sequence_reverse, Sequence::Modify
| Single template instantiation instead of recursive |
| __Flat iteration + decomposition__ | ford, static_ford | O(1) template
depth instead of O(N^D) |
| __Pre-computed strides__ | index_decomposer | Enables O(1)
linear-to-multi-index conversion |
### Impact on Compile Time
These optimizations reduce template instantiation depth from O(N) or
O(N^D) to O(1), which:
1. Reduces compiler memory usage
2. Reduces compile time exponentially for deep instantiation chains
3. Enables larger iteration spaces without hitting template depth limits
## Test Plan
* Existing tests for Sequence are re-used to affirm correctness
* Unit tests for ford and static_ford are added (dimensional looping)
* 8 new regression tests specifically verify the fixes for the PR
feedback:
- `NonTrivialOrder3D_201` - Tests Orders<2,0,1> for static_ford
- `NonTrivialOrder3D_201_Runtime` - Tests Orders<2,0,1> for ford
- `ConsistencyWithNonTrivialOrder_201` - Verifies static_ford and ford
consistency
- `NonTrivialOrder3D_120` - Tests Orders<1,2,0> for static_ford
- `NonTrivialOrder3D_120_Runtime` - Tests Orders<1,2,0> for ford
- `NonTrivialOrder4D` - Tests 4D with Orders<3,1,0,2> for static_ford
- `NonTrivialOrder4D_Runtime` - Tests 4D with Orders<3,1,0,2> for ford
- `AsymmetricDimensionsWithOrder` - Tests asymmetric dimensions with
non-trivial ordering
## Test Result
### Compile Time Comparison: `8b72bc8` (base) → `477e0686` (optimized)
#### Commits in Range (8 commits)
1. `fd4ca17f48` - Optimize sequence_reverse_inclusive_scan and
sequence_reverse
2. `7a7e3fdeef` - Optimize sequence_map_inverse
3. `92855c9913` - Optimize ford and static_ford calls to eliminate
nested template recursion
4. `88a564032b` - Add unit tests for ford and static_ford
5. `1a0fb22217` - Fix clang-format
6. `8a0d26bddf` - Increase template recursion depth to 1024
7. `dc53bb6e20` - Address copilot feedback and add regression tests
8. `477e06861d` - Increase bracket depth to 1024
#### Build Timing Results
| File | Base (8b72bc8759d9 | HEAD(a0438bd398) | Improvement |
|------|------|------|-------------|
| grouped_conv2d_fwd (f16) -j1 | 313.31s | 272.93s | __12.9% faster__ |
| grouped_conv1d_fwd (bf16) -j1 | 79.33s | 68.61s | __13.5% faster__ |
| grouped_conv1d_bwd_weight (f16) -j1| 15.77s | 14.31s | __9.2% faster__
|
| device_grouped_conv2d_fwd_instance -j64 | s | s | __% faster__ |
#### Key Optimizations
1. __sequence_reverse_inclusive_scan/sequence_reverse__: O(N) → O(1)
template depth
2. __sequence_map_inverse__: O(N) → O(1) template depth
3. __ford/static_ford__: O(N^D) → O(1) template depth using flat
iteration with index decomposition
4. __Copilot feedback fixes__: Corrected New2Old mapping for non-trivial
orderings
## Submission Checklist
- [ ] Look over the contributing guidelines at
https://github.com/ROCm/ROCm/blob/develop/CONTRIBUTING.md#pull-requests.
---------
Co-authored-by: Max Podkorytov <4273004+tenpercent@users.noreply.github.com>
**Supercedes https://github.com/ROCm/rocm-libraries/pull/4281 due to CI
issues on import**
## Proposed changes
Build times can be affected by many different things and is highly
attributed to the way we write and use the code. Two critical areas of
the builds are **frontend parsing** and **backend codegen and
compilation**.
### Frontend Parsing
The length of the code, the include header tree and macro expansions all
affect the front-end parsing time.
This PR seeks to reduce the parsing time of the dtype_vector.hpp
vector_type class by reducing redundant code by generalization.
* Partial specializations of vector_type for native and non-native
datatypes have been generalized to one single class, consolidating all
of the data initialization and AsType casting requirements into one
place.
* The class nnvb_data_t_selector (e.g., Non-native vector base dataT
selector) class has been removed and replaced with scalar_type
instantiations as they have the same purpose. Scalar type class' purpose
is already to map generalized datatypes to native types compatible with
ext_vector_t.
### Backend Codegen
Template instantiation behavior can also affect build times. Recursive
instantiations are very slow versus concrete instantiations. The
compiler must make multiple passes to expand template instantiations so
we need to be careful about how they are used.
* Previous vector_type classes declared a union storage class, which
aliases StaticallyIndexedArray<T,N>.
```
template <typename T>
struct vector_type<T, 4, typename ck::enable_if_t<is_native_type<T>()>>
{
using d1_t = T;
typedef T d2_t __attribute__((ext_vector_type(2)));
typedef T d4_t __attribute__((ext_vector_type(4)));
using type = d4_t;
union
{
d4_t d4_;
StaticallyIndexedArray<d1_t, 4> d1x4_;
StaticallyIndexedArray<d2_t, 2> d2x2_;
StaticallyIndexedArray<d4_t, 1> d4x1_;
} data_;
...
};
```
* Upon further inspection, StaticallyIndexedArray is built on-top of a
recursive Tuple concatenation.
```
template <typename T, index_t N>
struct StaticallyIndexedArrayImpl
{
using type =
typename tuple_concat<typename StaticallyIndexedArrayImpl<T, N / 2>::type,
typename StaticallyIndexedArrayImpl<T, N - N / 2>::type>::type;
};
```
This union storage has been removed from the vector_type storage class.
* Further references to StaticallyIndexedArray have been replaced with
StaticallyIndexedArray_v2, which is a concrete implementation using
C-style arrays.
```
template <typename T, index_t N>
struct StaticallyIndexedArray_v2
{
...
T data_[N];
};
```
### Fixes
* Using bool datatype with vector_type was previously error prone. Bool,
as a native datatype would be stored into bool ext_vector_type(N) for
storage, which is a packed datatype. Meaning that for example,
sizeof(bool ext_vector_type(4)) == 1, which does not equal
sizeof(StaticallyIndexedArray<bool ext_vector_type(1), 4> == 4. The
union of these datatypes has incorrect data slicing, meaning that the
bits location of the packed bool do not match with the
StaticallyIndexedArray member. As such, vector_type will use C-Style
array storage for bool type instead of ext_vector_type.
```
template <typename T, index_t Rank>
using NativeVectorT = T __attribute__((ext_vector_type(Rank)));
sizeof(NativeVectorT<bool, 4>) == 1 (1 byte per 4 bool - packed)
element0 = bit 0 of byte 0
element1 = bit 1 of byte 0
element2 = bit 2 of byte 0
element3 = bit 3 of byte 0
sizeof(StaticallyIndexedArray[NativeVectorT<bool, 1>, 4] == 4 (1 byte per bool)
element0 = bit 0 of byte 0
element1 = bit 0 of byte 1
element1 = bit 0 of byte 2
element1 = bit 0 of byte 3
union{
NativeVectorT<bool, 4> d1_t;
...
StaticallyIndexedArray[NativeVectorT<bool,1>, 4] d4x1;
};
// union size == 4 which means invalid slicing!
```
* Math utilities such as next_power_of_two addressed for invalid cases
of X < 2
* Remove redundant implementation of next_pow2
### Additions
* integer_log2_floor to math.hpp
* is_power_of_two_integer to math.hpp
### Build Time Analysis
Machine: banff-cyxtera-s78-2
Target: gfx942
| Build Target | Threads | Frontend Parse Time (s) | Backend Codegen
Time (s) | TotalTime (s) | commitId |
|---------------|---------|-------------------------|--------------------------|---------------|
---------------|
| device_grouped_conv3d_fwd_bias_bnorm_clamp_instance | 1 | 1452 | 331 |
1783 | 2e08a7e (develop) |
| device_grouped_conv3d_fwd_bias_bnorm_clamp_instance | 1 | 1403 | 332 |
1735 (-2.7%) | fad4235|
## Submission Checklist
- [ ] Look over the contributing guidelines at
https://github.com/ROCm/ROCm/blob/develop/CONTRIBUTING.md#pull-requests.
---------
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
Co-authored-by: systems-assistant[bot] <systems-assistant[bot]@users.noreply.github.com>
Co-authored-by: Illia Silin <98187287+illsilin@users.noreply.github.com>
## Proposed changes
Add parameter to bwd weight V3 type string showing the number of groups
to merge. This is required for MIOpen to be properly tuned since it uses
type strings for performance database entries.
In order to not break existing tuning databases, the parameter is added as a named suffix and only when group merging is enabled.
## Checklist
Please put an `x` into the boxes that apply. You can also fill these out
after creating the PR. If you're not sure, please don't hesitate to ask.
- [ ] I have added tests relevant to the introduced functionality, and
the unit tests are passing locally
- [ ] I have added the test to REGRESSION_TESTS list defined at the top
of CMakeLists.txt in tests/CMakeLists.txt, **IF** the test takes more
than 30 seconds to run.
- [ ] I have added inline documentation which enables the maintainers
with understanding the motivation
- [ ] I have removed the stale documentation which is no longer relevant
after this pull request
- [ ] (If this change is user-facing) I have added release notes which
provide the end users with a brief summary of the improvement from this
pull request
- [ ] I have run `clang-format` on all changed files
- [ ] Any dependent changes have been merged
## Discussion
If this is a relatively large or complex change, feel free to start a
discussion by explaining why you chose the solution you did and what
alternatives you considered
---
🔁 Imported from
[ROCm/composable_kernel#3679](https://github.com/ROCm/composable_kernel/pull/3679)
🧑💻 Originally authored by @johannes-graner
---------
Co-authored-by: Graner, Johannes <Johannes.Graner@amd.com>
Co-authored-by: systems-assistant[bot] <systems-assistant[bot]@users.noreply.github.com>
Co-authored-by: Bartłomiej Kocot <barkocot@amd.com>
I'm collecting information about our current testing (#3664). As part of
this work I a README to the directory to emphasize the GPU-first testing
strategy and our support for type-specific tolerances.
This readme contains internal code comments for CK developers and does
not need ROCm documentation review.
---
🔁 Imported from
[ROCm/composable_kernel#3665](https://github.com/ROCm/composable_kernel/pull/3665)
🧑💻 Originally authored by @shumway
Co-authored-by: John Shumway <jshumway@amd.com>
Co-authored-by: systems-assistant[bot] <systems-assistant[bot]@users.noreply.github.com>
Co-authored-by: Illia Silin <98187287+illsilin@users.noreply.github.com>
## Motivation
[CK] Fix grouped conv fwd transform for merged groups for 1d and 3d.
## Technical Details
After optimizations for 2d there is a lack of implementation for 1d and
3d
## Test Plan
test_grouped_convnd_fwd
## Test Result
pending CI
## Submission Checklist
- [x] Look over the contributing guidelines at
https://github.com/ROCm/ROCm/blob/develop/CONTRIBUTING.md#pull-requests.
## Motivation
Workaround to fix blockscale wp test failure for pipeline v3
## Technical Details
<!-- Explain the changes along with any relevant GitHub links. -->
## Test Plan
<!-- Explain any relevant testing done to verify this PR. -->
## Test Result
<!-- Briefly summarize test outcomes. -->
## Submission Checklist
- [ ] Look over the contributing guidelines at
https://github.com/ROCm/ROCm/blob/develop/CONTRIBUTING.md#pull-requests.
## Motivation
Fixing some of the CK CI issues
## Technical Details
fixing paths to dockerfiles and scripts;
moving codegen tests to separate stage (collides with main build since
you must call cmake from same folder but different options);
fixing a couple of clang compilation issues with staging compiler;
Replace specific benchmark numbers with qualitative descriptions since
measurements vary across environments and may become outdated.
Co-authored-by: Claude <noreply@anthropic.com>
[ROCm/composable_kernel commit: 3f04d27b68]
* [Compiler] Addressing new compiler warnings
Clang enables new lifetime warnings in production and we see build
errors due to this with the staging compiler.
The attributes added in this PR are suggested by the compiler. However,
I'm not very familiar with the code base, so the changes may be
incorrect.
* Update some more instances
* Adds file-level ignores via clang diagnostic pragma
The number of instances was large, so I decided to use file-level scope
to disable the warning via pragma clang diagnostic ignored.
It also showed this warning coming from the gtest dependency. For that,
I did add the respective command line flag to the CMake variables. I
don't know if this is acceptable or not.
* This adds the remaining instances
For a build on gfx90a.
* fix clang format
* Adding couple more instances from gfx1200 build
* Fixed another few instances
---------
Co-authored-by: Illia Silin <98187287+illsilin@users.noreply.github.com>
Co-authored-by: illsilin_amdeng <Illia.Silin@amd.com>
[ROCm/composable_kernel commit: 069500464d]
* Adding remaining flavors for grouped conv fwd
As titled. Following variants are added:
- grouped_conv2d_fwd_dynamic_op
- grouped_conv3d_fwd_dynamic_op
- grouped_conv3d_fwd_bilinear
- grouped_conv3d_fwd_convscale
- grouped_conv3d_fwd_convinvscale
- grouped_conv3d_fwd_convscale_add
- grouped_conv3d_fwd_convscale_relu
- grouped_conv3d_fwd_scale
- grouped_conv3d_fwd_combconvscale
- grouped_conv3d_fwd_scaleadd_scaleadd_relu
* Fix incomplete parsing of types from source names in add_instance_library() cmakelists function so we don't build f8 on RDNA3.
* Do not build f8 / bf8 only flavor tests on RDNA3
* Make sure we have proper generic instances for all instance lists related to the post-ces extra flavors, with scalarPerVector = 1. Then disable all but one generic instance per instance list to reduce compile time.
* Post rebase fix: Template parameters for Grouped Conv Fwd Device Impl got tweaked upstream.
* adding int8 and fp16 overloads to the elementwise operations
* fixed copilot nits
* Addressing review comments:
- removed unnecessary examples for dynamic op
- removed unnecessary conv specalizations for all the flavors
- removed spurious bilinear and scale source files
* clang-format
* reduced no of tests
---------
Co-authored-by: Wojciech Laskowski <wojciech.laskowski@streamhpc.com>
[ROCm/composable_kernel commit: 2377a62837]
* Add multi AB support to wave transfer
* Improviments to multi ABD examples
* Add instances and use intrawave v1 instead of interwave
* Apply changes to other transfers
* Wave transfer: add support for multiple internal vgpr buffers
* Fix compilation error gfx11
[ROCm/composable_kernel commit: f16d9100e4]
* Enable bwd weight splitk autodeduction with cap
* Fix error threshold calculations
* Add missing logic to wmma multiple d kernel
* Fix threshold calculation
* Update test with new applicability
[ROCm/composable_kernel commit: fabac7e2c3]
* ck-builder: tensor copy function
This function copies one tensor to another, so that the memory
layout can be changed between them.
* ck-builder: fix ck::bhalf literals
These types don't work properly.
* ck-builder: abstract compare_elements in gpu_verification.hpp and make builder use it
This reduces the amount of duplicated code a bit.
* ck-builder: add flat tensor iterator
This "iterator" type pretends to be a pointer, useful for passing
tensors to functions expecting pointer-like types.
* ck-builder: integrate validation with ck gpu verification
By templating the gpu_verify function over iterators, we can use
the new FlatTensorIterator to adapt the function to multi-
dimensional tensors without changing either implementation
too much.
* ck-builder: add check_by_accumulations
This changes the gpu_verification.hpp code to also accept "iterator"
types for the relevant gpu_verify and gpu_reduce_max functions.
* ck: fix test_gpu_verification GenerateRandomData for bhalf
is_integer_it<bhalf_t> yields true, but it is not actually
an integer.
* ck: make gpu_verification kernels be proper persistent kernels
Previously these were using a hardcoded value for the grid size. This
commit changes that so that the grid size is automatically derived
from the kernel's occupancy and the number of multiprocessors on
the GPU.
* ck: clean up gpu_verification.hpp using block_reduce
This implements a small generic block reduce function, and rewrites
the rest of gpu_verification.hpp using that function to clean it up
a bit.
* ck-builder: doc typos
* ck-builder: update testing readme with validation interface.
* ck-builder: rebase fixes + review comments
* ck-builder: fix device integer generation with float types
Passing bfloat here causes a nans due to type_convert performing
a bitcast.
* ck: another bhalf_t bug
CK expects that int-generation with ck::bhalf_t yields bhalf integers,
not unsigned integers. This makes the logic of FillUniformRandInteger
compatible with GeneratorTensor_2<InDataType>, however idiotic that
may be.
[ROCm/composable_kernel commit: 42048bdb7d]
1. Add base class GridwiseGemm_xdl_cshuffle_base for all gridwise_gemm_xdl classes.
- to select correct LDS layout and epilogue behavior , three additional parameters is added.
- ForceNaiveLdsLayout: disable XOR based LDS layout when it is true
- DirectLoad: pipeline only use directload, we need force naive layout and ignore any padding on gfx9
- IsMxGemm: epilogue has two addtional dimensions
2. Move all LDS descriptor layout related fucntion to base class, including
- GetABlockDescriptor_AK0PerBlock_MPerBlock_AK1
- GetBBlockDescriptor_BK0PerBlock_NPerBlock_BK1
- GetCShuffleBlockDescriptor_MBlock_MPerBlock_NBlock_NPerBlock
3. Move several LDS related helper funtions to base class, including
- GetSharedMemoryNumberOfByte
- GetABlockDescriptor_AKB_AK0PerBlock_MPerBlock_AK1
- GetBBlockDescriptor_BKB_BK0PerBlock_NPerBlock_BK1
- GetCBlockDescriptor_MBlock_NXdlPerWave_MWaveMPerXdl_NBlock_NXdlPerWave_NWaveNPerXdl
4. Move all c epilogue related code to base class, and 4 kind of implementation are provided
- RunEpilogueNoShuffle
- RunEpilogue
- RunMultiDEpilogue
- RunMoeEpilogue
[ROCm/composable_kernel commit: 23cefda140]
This document describes techniques for reducing C++ template instantiation
overhead in the Composable Kernel codebase, including:
- Replacing recursive templates with pack expansion (O(N) → O(1) depth)
- Using named functors instead of lambdas to share instantiations
- Replacing template recursion with constexpr loops
- Using fold expressions for accumulation operations
These techniques can significantly reduce build times for template-heavy code.
[ROCm/composable_kernel commit: b66597ed96]
* Add padding support with transpose
Also move check before writing storing is_src_valid during reading
* Add/modify instances to use wave transfer for gemm universal
Condition is changed so now the vectorsize of vmem reading and lds
writing must be equal to 8 in order to use the wave transfer
* Fix clang format
* Modify example
* Fix bwd data
* Add restriction for wave transfer with padding and transpose
Add test case which shows this limitation
* Fix validity checks 8 bit types
* Add validity check gemm_bias_add_reduce
* Add validity check grouped gemm tile loop
* Fix validity checks new flavours
* Minor fixes
* Fix clang format
[ROCm/composable_kernel commit: 2e49b6b2f7]
* Added common struct to enable code reduction in gemm gemm and gemm multi_d gemm multi_d wmma implementation
This file includes all shared components. The (shared between the two implementations) kernel, the pointer offset computation struct, the grid descriptor creator and definitions, the invoker struct and the argument struct.
Signed-off-by: Chris Tsiaousis <chris.tsiaousis@streamhpc.com>
* Used the common struct in the batched gemm gemm wmma cshuffle v3 implementation
Signed-off-by: Chris Tsiaousis <chris.tsiaousis@streamhpc.com>
* Used the shared structs in the gemm multiple D gemm multiple D wmma cshuffle v3 implementation
Signed-off-by: Chris Tsiaousis <chris.tsiaousis@streamhpc.com>
* Boy-scout: IWYU paradigm in the gemm gemm and gemm multiple D gemm multiple D wmma cshuffle v3 implementations
Signed-off-by: Chris Tsiaousis <chris.tsiaousis@streamhpc.com>
---------
Signed-off-by: Chris Tsiaousis <chris.tsiaousis@streamhpc.com>
[ROCm/composable_kernel commit: 917f35553a]
This change significantly improves compile-time performance by reducing template
instantiation depth for sequence generation and merging operations:
Optimizations:
- sequence_gen: Reduce instantiation depth from O(log N) to O(1) by using
__make_integer_seq to generate indices in a single step, then applying the
functor via pack expansion
- uniform_sequence_gen: Similarly optimized to O(1) depth using __make_integer_seq
with a helper that applies a constant value via pack expansion
- sequence_merge: Reduce depth from O(N) to O(log N) using binary tree reduction
strategy. Added direct concatenation specializations for 1-4 sequences to
avoid recursion in common cases, falling back to binary tree merging for 5+
sequences
Documentation:
- Added extensive inline comments explaining why sequence_merge cannot achieve
O(1) depth like sequence_gen (requires computing cumulative sequence lengths
from heterogeneous inputs, inherently requiring recursion)
- Documented the binary tree reduction approach and why it's superior to fold
expressions for this use case
Testing:
- Added comprehensive unit tests for uniform_sequence_gen with different values,
sizes, and edge cases
- Added tests for sequence_gen with custom functors (double, square, identity,
constant) to verify the new implementation works with arbitrary functors
- Added tests for sequence_merge with 4, 5, and many sequences to verify both
the direct concatenation path and binary tree reduction path
- Added tests for empty sequence edge cases
[ROCm/composable_kernel commit: de59c0716c]
* Moved device struct for batched gemm wmma to a common file
Signed-off-by: Chris Tsiaousis <chris.tsiaousis@streamhpc.com>
* Use the common device struct in the scaled batched gemm wmma implementation
Signed-off-by: Chris Tsiaousis <chris.tsiaousis@streamhpc.com>
* Boy-scout: Remove unused includes and ambiguous comment
Signed-off-by: Chris Tsiaousis <chris.tsiaousis@streamhpc.com>
* Moved pointer offset calculation and gridwise argument to common struct
This change enables further code reduction by re-using the common structs for the batched gemm and batched gemm b scale wmma implementations.
Signed-off-by: Chris Tsiaousis <chris.tsiaousis@streamhpc.com>
* Moved type string to the common struct of DeviceBatchedGemm_Wmma_CShuffleV3_Common"
Signed-off-by: Chris Tsiaousis <chris.tsiaousis@streamhpc.com>
---------
Signed-off-by: Chris Tsiaousis <chris.tsiaousis@streamhpc.com>
[ROCm/composable_kernel commit: e1c46ff548]
* wip: test suite for batched gemm multiple d gemm multiple d, working on gridwise implenentation
* wip: many fixes in implementation of batched gemm gemm multiple d
* wip: batched gemm gemm multiple d gridwise op compiling, not working yet
* fix: incorrect d0 grid indexing in batched gemm gemm multipled
* feat: add instances for batched gemm add relu gemm add
* chore: configure instance with low vector transfer size for odd sizes
* chore: add some more validation to device batched gemm gemm multiple d, and removed template parameter that didn't really make sense
* fix: upate device_batched_gemm_gemm_wmma to work with new gridwise changes
* fix: disable odd size tests on XDL archs
* chore: removed temporary logging
* chore: update some references to C tensor to E tensor
* Tentative fix for example template params
* Tentative fix for non-multi-D batched gemm gemm device impl.
* Tentative fix for xdl example template params
* Tentative fix for profiler build on gfx90a
* chore: improve device batched gemm gemm multi D comment to include all ops and dimensions
* chore: explicitly call ck::make_tuple to prevent issues when std::make_tuple would apply
* fix: make the gemm1 data types match what happens in the device op
* feat: add d0s/d1s datatypes and layouts to the device op type string
* chore: change element-wise op so addition happens in fp32
* chore: add static asserts for gemm0/gemm1 calculated wave sizes
* chore: also updated other element-wise ops to use fp32 calculations
* chore: log number of supported instances
* chore: update instance comment
* chore: disable kernel timing in example by default
* fix: gemm1 wave size calculation
* fix: make sure batched gemm multiple d gemm multiple d profiler performs correct type conversions
* chore: remove increased tolerance in batched gemm gemm multiple d example
* chore: add comment explaining that verification fails for certain input values
* chore: clarify instance comment
---------
Co-authored-by: kiefer <kiefer.van.teutem@streamhpc.com>
[ROCm/composable_kernel commit: d5ae81b292]
* feat: test setup for batched contraction (aka batched gemm multiple d e permute)
* wip: device struct for WMMA batched contraction multiple d based on new gridwise op
* feat: working batched contraction on RDNA, non-naive tensor descriptors for gridwise_gemm_wmma_cshuffle_v3, test setup for odd cases
* fix: failure to resolve template parameters when calling new function overload
* fix: passing reference type as parameter instead of underlying types
* fix: merge error caused duplicate definitions
* fix: make sure constness of template and parameters types match
* fix: don't compile batched contraction test on unsupported architectures
* feat: add example for new wmma implementation, and consolidate example code between platforms
* style: return inline instead of with branch
* chore: add extra assert on vector memory access sizes
* chore: clean up some unused variables
* fix: correct tail number calculation, added small cases and extra instances to the test
* fix: properly support wave transfer by generating correct grid descriptors dependent on the transfer method
[ROCm/composable_kernel commit: fe40a5d139]
* Factor helpers out of conv_traits.hpp
* Create a non-templated conv_traits struct
* Migrate to new instance-specific instance_to_conv_traits functions
* Clean up reflection concepts
* Clean up ConvTraits helpers
* Update testing for convolution traits
This is a lot of cleanup on tests to have verbose coverage of feature
extraction, explicit tests for each supported device kernel, and
simple, readable test code.
* Address reviewer comments and resolve merge conflict
[ROCm/composable_kernel commit: 5122637215]
* Refactor GPU verification kernel to gather erorr stats on GPU
* Check if result is all zero
* non-negative error count doesn't need custom Atomics
* Remove unnecessary AtomicMaxFloat function
* Simpler warp reduction, remove passed flag
* Move verification header to include
* Fix header path in test
* Fix block reduction loop
[ROCm/composable_kernel commit: f173642087]
- Add support for direct store in epilogue instead of cshuffle
- Add padding support for wave transfer without transpose
- Add wave transfer with interleaved layout to support direct store
- Enable new functionalities on GEMMs
- Add optional new functionality support for grouped convolution fwd
- Add some fast instances for grouped convolution fwd with new functionalities (proper tuning needed)
[ROCm/composable_kernel commit: 693ff3bbb3]
* Add placeholder test.
* Initial conv bwd weight factory.
* Conv builder test refactoring.
* Add missing pieces to bwd weight factory.
* Improve compile time erros message when no matching factory is found.
* Use amcro to ensure automatic macthing between concepts are their string representations.
* Improve compile time diagnostics.
* Small improvements.
* Improve missing member/wrong type compile-time errors.
* Improve compile time diagnostics.
* Concept bug fixes.
* Remove debug assert.
* Update algorithm signature diagnostics.
* Factory bug fixes.
* First functional version of bwd weight conv factory.
* Refactor handing of GEMM-K batch template parameter in conv bwd weight factory.
* Concept improvements.
* Improve concept diagnostics.
* Introduve a common size type for concepts.
* Update compiletime diagnostics to use the size type.
* Update conv specialization enum.
* Fix fwd conv builder tests.
* Fix smoke tests.
* Separate bwd weigth and bwd data tests into separate targets.
* Clean-up CK Tile builder tests.
* Add bwd weight XDL CShuffle V3 factory.
* Build conv bwd weigth v3 instances successfully.
* Add instance traits for DeviceGroupedConvBwdWeight_Xdl_CShuffleV3.
* Test fix.
* Add instance traits for bwd weight algorithms.
* Add unit tests for instance strings.
* Build new instance traits unit tests but exclude WMMA for now.
* Added factory for DeviceGroupedConvBwdWeightTwoStage_Xdl_CShuffle.
* Conv bwd weight DL factory.
* Final implementation for bwd weight DL factory.
* Add test for creating DeviceGroupedConvBwdWeightMultipleD_Xdl_CShuffle instance.
* Add factory for DeviceGroupedConvBwdWeightMultipleD_Xdl_CShuffle
* Treat ref algorithm the same way as real algorithms in the dispatcher.
* Refactor large tensor support and WMMA configuration.
* Add factory and tests for DeviceGroupedConvBwdWeight_Wmma_CShuffleV3.
* Update Readme.
* Fix WMMA bwd weight tests.
* Added factory and tests for DeviceGroupedConvBwdWeightTwoStage_Wmma_CShuffleV3.
* Factory and tests for DeviceGroupedConvBwdWeight_Wmma_CShuffle.
* Dispatching for DeviceGroupedConvBwdWeightMultipleD_Wmma_CShuffle.
* Add factory for DeviceGroupedConvBwdWeightMultipleD_Wmma_CShuffleV3
* Fix DeviceGroupedConvBwdWeightMultipleD_Wmma_CShuffleV3 factory and compute types for input and output tensor in bwd weigth convs.
* Fix fwd factories after refactoring.
* clang-format
* Move compile-time diagnostics to a separate branch.
* Fix ref algorithm dispatching.
* Fix smoke tests.
* clang-format
* Fix factory for regular WMMA conv bwd weight.
* Clarify builder Readme.
* Remove obsolete test file.
* Fix test after merge.
* clang-format
* Remove the C++26 extensions.
* Unify conv elementwise ops and layout definitions for fwd and bwd directions.
* Remove old layout and elementwise ops.
* Unify handling of conv tensor types between fwd and bwd directions.
* Unify block transfer for fwd and bwd directions. Rename ThreadSliceDim to ThreadClusterRank.
* Make BlockTransferDescriptor concept parametrized. Introduce a common TileTransferParameters concept for conv algorithms.
* clang-format
---------
Co-authored-by: Ville Pietilä <>
[ROCm/composable_kernel commit: 9908a87c31]
* nt on fp8 blockscale
* some improve and tests needs to be fixed
* update
* fix format
* revert useless change
* revert any change in amd_buffer_coherence
[ROCm/composable_kernel commit: 32408c8bc0]
* Take split_k into account when checking 2GB tensor limit.
* Revert "Take split_k into account when checking 2GB tensor limit."
This reverts commit adf35c91be.
* Optimize grouped conv bwd wei split_k off calc
(cherry picked from commit 2115642ee59050dabd81393c1b8f03b34adc05aa)
* Update gridwise_gemm_xdl_cshuffle_conv_v3.hpp
(cherry picked from commit 900d4d4b466f5730ae1189370d3c96267c35ea69)
* Fix tensor descriptors and stride calculations
* Don't miss half of the elements
* Fix buffer size calculations
* Disable hack if stride not divisible by k_batch
* Clean up comments
* Disallow hack in non-contiguous edge cases
* Index -> Dim
* Fix broken test
* Refactor applicability checks into separate function
* fix missed variable name
* Fix variable name in info print
* update V3 2GB check
* No more regression, use templates instead
* Code deduplication
* Regression fix for cshuffle
* arch-guarded atomic_add implementations for gfx11
* Similar for half(4|8)_t as well
* Only use both offset hacks at the same time
* Revert "arch-guarded atomic_add implementations for gfx11"
This reverts commit 3883fe6935.
This reverts commit 5311ec608d.
* Reapply "arch-guarded atomic_add implementations for gfx11"
This reverts commit 1972adeddc.
* Only remove float4 atomic_add
* Refactor to single flag
* Consolidate template parameters
* Consolidate flag in transformers
---------
Co-authored-by: Bartlomiej Kocot <barkocot@amd.com>
[ROCm/composable_kernel commit: ee2c35b92d]
