* wmma_op + unit test
* add arch limitation to wmma test
* change arch limitation
* Refactor + Add all type unit test(int4 compile failed)
* Add f32_16x16x16_bf16 unit test
* Remote int4 related
* delete deprecated test
Co-authored-by: Po Yen Chen <PoYen.Chen@amd.com>
Co-authored-by: Chao Liu <chao.liu2@amd.com>
* Re-structure ckProfiler source files
* Rename profiler.cpp to main.cpp
* Modularize ckProfiler operations
* Add description for profiler operations
* Use longer name to avoid name collision
* Use macro to delay expansion
* Use std::move() to avoid object copying
* Prohibit users from calling dtor
* Use macro to eliminate redundant code
* Make friend function hidden
* Add missing include directive <iostream>
* Fix wrong include directives
* Remove int8 from batchnorm-forward instances since it is not needed for forward training and could fail test
Co-authored-by: Qianfeng Zhang <Qianfeng.Zhang@amd.com>
* Refine the device batchnorm-backward base API templates and data type assignments
* Remove duplicated kernel file
* Add batchnorm backward instances and external API
* Add batchnorm-backward profiler and tests
* Add client example which uses batchnorm backward external API
* Merge test/batchnorm_fwd and test/batchnorm_bwd into one directory
* Loose the threshold for batchnorm-backward check_err()
* Update to device_batchnorm_forward base class to include all template parameters for problem description
* Add batchnorm forward instances and external api
* Add batchnorm forward profiler module which uses the external api
* Add some comments in batchnorm_forward example to explain the dimensions in lengths[]
* Replace the reference_batchnorm_forward_nhwc_c by generic reference_batchnorm_forward
* Improvement to the batchnorm infer base API
* Add batchnorm forward client example which shows using the batchnorm forward external API
* Add test for batchnorm forward
* Tuning the batchnorm profiler initialized values and error threshold
* Add support for bhalf_t in instances/external api/tests
* Add support for int8_t in instances/external api/tests
* Add support for double in instances/external api/tests
* Let ScaleDataType and BiasDataType be same as XDataType and YDataType when creating instances
* Checking before running best instance in batchnorm_fwd_nhwc client example
* Add checking for YElementwiseOp in batchnorm_forward external API
* Add more types in batchnorm forward profiler
* Add more test lengths
Co-authored-by: rocking5566 <ChunYu.Lai@amd.com>
* fixed bug in softmax reference & add bf16 examples for batched_gemm_scale_softmax_gemm
* added bf16 tests for batched_gemm_softmax_gemm_permute
* changed format of device_batched_gemm_softmax_gemm_permute_xdl_cshuffle_bf16_bf16_bf16_bf16_gmk_gnk_gno_gmo_instance.cpp
* changed format device_batched_gemm_softmax_gemm_permute_xdl_cshuffle_bf16_bf16_bf16_bf16_gmk_gnk_gno_gmo_instance.cpp
* aligned annotations
* modified CMakeLists for examples
* add common example code of fp16/bf16 version for batched_gemm_scale_softmax_gemm_xdl
* use macro to control the instances
* added macro control into instances
* clang-format some files
* changed error tolerance for bf16
* changed index for 10_elementwise_normalization
* fixed xdlops code bug in amd_xdlops.hpp
Co-authored-by: Po Yen Chen <PoYen.Chen@amd.com>
* Rangify STL algorithms
This commit adapts rangified std::copy(), std::fill() & std::transform()
* Rangify check_err()
By rangifying check_err(), we can not only compare values between
std::vector<>s, but also compare any ranges which have same value
type.
* Allow constructing Tensor<> like a HostTensorDescriptor
* Simplify Tensor<> object construction logics
* Remove more unnecessary 'HostTensorDescriptor' objects
* Re-format example code
* Re-write more HostTensorDescriptor ctor call
* Remove redundant CMake setting
* Extract common code from files
* Rename folder 'convnd' to 'conv'
* Use std::array<> to accept compile-time kwnown # of arguments
* Fix compilation error of tuning parameter
* In example, use same setting as unit-test
* Remove no-longer used include directive
* Add interface for grouped conv bwd weight
* Add group support for conv bwd weight
* Add grouped conv bwd weight example
* Use group parameter in example
* Rename example folder
* Remove non-grouped version example source files
* Rename device op template
* Add group support to convolution backward weight
* Remove debug messages
* Use smaller group size in example
* Use named variable as loop terminate condition
* Prettify example output message
* Enlarge used grid size
* Allow real grid size exceeds expected grid size
* Rename interface file
* Add client example for grouped conv2d bwd weight
* Fix wrong include directive
* Rename client example folder
* add fused addition lyernorm
* add fused addition lyernorm
* changed CMakelist
* removed annotates
* modified descriptor of C
* fixed bug in gridwise add layernorm
* format the files
* modified name from add&layernorm into elementwise&layernorm
* created fused elementwise layernorm branch
* change input into tuple type
* add sweep once to reduce load & read of C from global memory
* modified Argument api
* modified way to malloc c in global memory
* changed gamma and beta to m_k_desc
* fixed bug when sweep once and move CDataType when define device level struct
* add src dim for gamma and beta
* implement optimization for coalesced
* delete a annotation line
* fixed some bug to meet the requirements of ck
* add bandwidth computing in example, and fixed the time unit
* move device_elementwise_layernorm_impl.hpp into device/impl
* fixed bug in device_elementwise_layernorm_impl.hpp
* changed name from layernorm into normalization
* clang-format the changed files
* changed the names
* moved immidiate results into lds, it become faster in non-sweeponce cases
* changed naming of C into X to make the defination more clear
* changed naming in example
* add tests for elementwise normalization
* move example_elementwise_layernorm_blockwise into folder 44_elementwise_normalization
* move test_elementwise_layernorm_fp16 into new folder
* move elementwise_normalization_instances into a new folder
* add more tests in test_elementwise_layernorm_fp16.cpp
* added some corner cases in test
* fixed method to compute lds size for matrix X
* changed name of 44_elementwise_normalization into 45_elementwise_normalization
* modified some comments
* modified some other confused comments
* reduce redundant tests in test_elementwise_layernorm_fp16.cpp
* Sync the naming
* Sync the test of layernorm with groupnorm
* Sync the naming
* Minor change for comment and log
* [What] Add saveMean and SaveInvVariance in the interface.
[Why] These can optimize the backward
* Add reduction across all dims cases.
* host softmax: handle all reduce
* Test cases when reduced dim is not innermost axis.
* Fix syntax.
* Test non innermost dim for fp32 and int8
* Group test suites wrt NumReduceDim.
* Additionally test failing cases.
* Throw error when Rank or NumReduceDims doesn't match arguments.
* Check reducedDims has correct values
* Move don't reuse DeviceReduceMultiblock IsSupportedArgument method.
Instead implement own. (in fact just get rid of one check to enable
reduction across inner dimensions).
* Reorganize unit tests to better cover use scenarios.
* Test input validation
* Test reduction of inner dimensions with custom op instances.
* Refactor fp32 and int8 unit tests.
* Fix FP32 instance template parameters.
* Add more instances.
* Instances with InSrcVectorDim=0.
* Do not initialize and copy data when arg not supported.
* ckProfiler Softmax use instance factory.
* Refactor device softmax IsSupported.
* Additionally add non-polymorphic api functions
* Split softmax instances into multiple files.
* Fix profiler.
* Reorganize tests to reuse profiler and cover edge cases.
* Clang-format
* I8 Softmax instances along with UT.
* Reuse type alias definitions from instance factory header.
* Clean included headers
* Fix variable names.
* Add missing checks in Argument constructor.
Co-authored-by: Adam Osewski <aosewski@amd.com>
Co-authored-by: Anthony Chang <ac.chang@outlook.com>
* reopen masking att instance due to CI is upgraded
* re-enable instances previously failed on 9110
* enable ksize-kpadding pair validity test
* add non-masked attention+permute test; expose masking boolean to attention kernel handles
* disable bench
* fix test
* move files
* bulk rename batched_gemm_masking_scale_softmax_gemm_permute to batched_gemm_softmax_gemm_permute
* format
* amend rename
* disable bench in test
* add mask/no-mask test for non-permute attention kernels
* disable broken kernel instance
* example working
add non-permuted problem statement
evaluating whether overhead comes from permutation or the extra kernel arg
* interface for bias addition without implementing it
* test and profiler running
* tidy
* mask type determined by enum class
* unify example code
* move masking specialization to its own header
* align formats
* extract helper functions
* experiment merging dims for attn w/ permute; shows perf parity with attn wo/ permute
* add tensor specialization to template args
since tensor spec packed shows perf parity when permutation isn't needed
remove redundant template args
comment on 'packed' tensor specialization
* grouped attention with input/output permute example
* format
* clean up
* refactor acc0 tile visitor
Co-authored-by: shaojiewang <wsjmessi@163.com>
Co-authored-by: Chao Liu <chao.liu2@amd.com>
* Fix for lwpck-425, update BlockTransferSrcVectorDim
* Revert "Fix for lwpck-425, update BlockTransferSrcVectorDim"
This reverts commit fd24e280e2.
* Add Batched Gemm int8 test, expect it to fail
* Format
* Re-add the fix
* prototype
4 layouts
fix default stride
all problem sizes
tidy
move file
update build script
restore old file
fix build
* refactor standalone test to use gemm test harness
* simplify gemm test
* update build script
* remove redundant
* early return when cmd arg doesn't match
* tidy
* report failure when result not validated
* tidy
* Apply suggestions from code review
Co-authored-by: Adam Osewski <19374865+aosewski@users.noreply.github.com>
Co-authored-by: Adam Osewski <19374865+aosewski@users.noreply.github.com>
* add fused addition lyernorm
* add fused addition lyernorm
* changed CMakelist
* removed annotates
* modified descriptor of C
* fixed bug in gridwise add layernorm
* format the files
* modified name from add&layernorm into elementwise&layernorm
* created fused elementwise layernorm branch
* change input into tuple type
* add sweep once to reduce load & read of C from global memory
* modified Argument api
* modified way to malloc c in global memory
* changed gamma and beta to m_k_desc
* fixed bug when sweep once and move CDataType when define device level struct
* add src dim for gamma and beta
* implement optimization for coalesced
* delete a annotation line
* fixed some bug to meet the requirements of ck
* add bandwidth computing in example, and fixed the time unit
* move device_elementwise_layernorm_impl.hpp into device/impl
* fixed bug in device_elementwise_layernorm_impl.hpp
* changed name from layernorm into normalization
* clang-format the changed files
* changed the names
* moved immidiate results into lds, it become faster in non-sweeponce cases
* changed naming of C into X to make the defination more clear
* changed naming in example
* add tests for elementwise normalization
* move example_elementwise_layernorm_blockwise into folder 44_elementwise_normalization
* move test_elementwise_layernorm_fp16 into new folder
* move elementwise_normalization_instances into a new folder
* add more tests in test_elementwise_layernorm_fp16.cpp
* added some corner cases in test
* fixed method to compute lds size for matrix X
* changed name of 44_elementwise_normalization into 45_elementwise_normalization
* modified some comments
* modified some other confused comments
* reduce redundant tests in test_elementwise_layernorm_fp16.cpp
* Move kernel implementation files under impl directory.
* Update examples paths.
* Update device kernel impl include paths.
* Update tensor operation instances include paths.
* Update profiler and tests include paths.
* Clang-format
* Update include paths for batched gemm reduce
* Refactor UnitTest ConvNDBwdWeight.
* Refactor fwd and bwd data convND UT.
* Fix used test macro.
* Fix include path.
* Fix include paths.
* Fix include paths in profiler and tests.
* Fix include paths.
Co-authored-by: Adam Osewski <aosewski@amd.com>
* use another instance to check the efficiency
* optimize group layer norm
* 1. coalesce load/store data for gridwise layer norm welford. 2. move a sqrt and divison into a outer static loop
* add more instances to layernorm
* add 2 more test cases
* remove ignore in generating tuple of vector
Co-authored-by: Chao Liu <chao.liu2@amd.com>
* Add groupnorm example by layernorm
1. Reference is not ready
2. shape of gamma and beta need to be fix
* Let shape of gamma and beta can be same as x
* Modify test, instance and client example
* [What] Fix bug of layernorm for greater than 2 dimension.
[Why] We need to get upper length from merge transform instead of embed transform.
* Add reference for groupnorm
* Fuse sigmoid after groupnorm
* [What] Rename original layernorm into layernorm2d
[Why] Prepare to add groupnorm using layernorm5d
* clang-format
* Add groupnorm test
* Refine error message
* Add groupnorm ckProfiler
* Test groupnorm kernel from device_instance
* update example
* upadte profiler
* Fix test naming
* Fix argc number
* Move descriptor and sweeponce to argument for quick debugging
Co-authored-by: Chao Liu <chao.liu2@amd.com>
* modify comment
* trim unnecessary check
* add gemm spec in kernel name
* add TNTT gemm_gemm + atten kernel instances
* refactor attention padding to better fit in unit tests
This streamlines usage where "ResetNaNToMinusInf" is now hidden from user facing device op.
Also added compile-time conditionals that load OOB value as NaN only after padding is enabled
* add adhoc padding test for atten
* shrink input value range for attention kernel validation to avoid occasional error by 1e-3
Still unsure whether this kind of deterministic floating point accurary issue is expected
or not. May want to try exact same approach as the GPU kernel in the host reference
GEMM+Softmax+GEMM function to see if the accuracy discrepancy goes away. Until then,
shrink the input value range as it is less likely to produce errors of around ~1e-3.
* attention kernel proper granular padding for all 4 dims
* IsSupportedArgument checks
* test more padded cases
* block PadK specialization in attention kernels
* workaround clang crash for gfx908
(gfx908 only) workaround for compiler crash in fused kernels on mainline #9110; #10738 seems ok
error message was "fatal error: error in backend: Error while trying to spill VGPR0 from class
VGPR_32: Cannot scavenge register without an emergency spill slot!"
this fall back to less ideal way of handle NPadding in fused attention kernel
* comment out kernels giving wrong results on MI100; MI200 doesn't seem affected
* More int4 UT.
* Disable BitwiseRepresentation UT.
* Add UT with static_cast
* Surround cout statements with #if
Co-authored-by: Adam Osewski <aosewski@amd.com>
* GemmPadder and GemmGemmPadder
* proper padding using GemmGemmPadder
* test gemm_gemm padding
* properly check size K in IsSupportedArgument()
* properly check size requirement given SrcScalarPerVector in IsSupportedArgument()
* comment
* format
* Introduce int4 data type.
* Add unit-tests for int4
* Compile int4 UT only when int4 enabled.
* clang-format
Co-authored-by: Adam Osewski <aosewski@amd.com>
* Add threadwise and blockwise welford
* Rename gridwise op, prepare to add welford version
* implement welford and integrate welford into layernorm
* Take care of tail loop
* Fix buf when ThreadSliceK > 1
* Fix bug of merging of two empty set
* Rename clip to clamp
* 1. Fix type of count
2. Remove useless static_assert
* Do not inherit Reduction::Argument
* [What] replace __syncthreads() with block_sync_lds()
[Why] __syncthreads might wait both lgkmcnt(0) and vmcnt(0)
* Add y stride
* Rename.
DeviceLayernorm -> DeviceLayernormImpl
DeviceNormalization2 -> DeviceLayernorm
* Move literal ""_uz & ""_zu into namespace 'literals'
* Move namespace 'literals' as 'ck::literals'
Co-authored-by: Po-Yen, Chen <PoYen.Chen@amd.com>
Co-authored-by: Chao Liu <chao.liu2@amd.com>
* initial stub for gemm_gemm_xdl_cshuffle
* set up example code
* compiles
* prevent integer overflow
* harmonize interface between ref_gemm and ref_batched_gemm
* batched_gemm_gemm
* fix example
* host tensor gen: diagonal pattern in lowest two-dimensions only
* make c descriptors containing only integral constants
* clean up
* add BlockwiseGemmXdlops_v2 while exploring an unified approach
* implement proper interface
* tidy up example
* fix compilation warnings
* coarsely controlled 2nd gemm padding
* remove rocm-cmake's hard requirement for certain revision
* clang-format
* resolve merge conflict
* fix compilation error on gfx10
* adds acc0 elementwise op to interface
* add gemm_gemm instances and tests
* avoid LDS data hazard
* fix build
Co-authored-by: Chao Liu <chao.liu2@amd.com>
* initial stub for gemm_gemm_xdl_cshuffle
* set up example code
* compiles
* prevent integer overflow
* harmonize interface between ref_gemm and ref_batched_gemm
* batched_gemm_gemm
* fix example
* host tensor gen: diagonal pattern in lowest two-dimensions only
* make c descriptors containing only integral constants
* clean up
* add BlockwiseGemmXdlops_v2 while exploring an unified approach
* implement proper interface
* tidy up example
* fix compilation warnings
* coarsely controlled 2nd gemm padding
* remove rocm-cmake's hard requirement for certain revision
* clang-format
* resolve merge conflict
* fix compilation error on gfx10
* adds acc0 elementwise op to interface
* attention host validation
* add blockwsie softmax v1
* iteratively update softmax+gemm
* transpose both gemm0 and gemm1 xdl output so as to avoid broadcasting softmax max/sum
* add init method for easier debugging
* do away with manual thread cluster calculation
* generalize blockwise softmax interface
* row-wise softmax sum & max
* format
* rename to DeviceBatchedGemmSoftmaxGemm
* add gemm_softmax_gemm instances and tests
* comment
Co-authored-by: ltqin <letao.qin@amd.com>
Co-authored-by: Chao Liu <chao.liu2@amd.com>
* add verify flag and update scripts
* replace old check_error function with the new check_err
* fix syntax
* remove blank spaces
* remove empty line
* add check_err for tensors
* fix syntax
* replace tensors with vectors in check_err calls
* fix syntax
* remove blank spaces
* fix syntax
* add new line at end of file
* disable conv2d_bwd_weight test, add gpu check
* set check_gpu using export
* check GPU using runShell
* add definition of runShell
* fix script syntax
* reduce the number of threads, add full qa option
* run processing scripts in bash
* fix the branch and host names in performance scripts, add chronos
* replace parameterizedCron with cron
* archive the perf log files
* try to fix git call
* pass branch and host names as arguments into scripts
* fix script arguments
* fix script arguments
* process results on master
* fix pipeline
* add definition of gpu_arch
* run processing scripts in docker
* fix the brackets
* add agent master for the processing stage
* get rid of show_node_info call on master
* try using mici label instead of master, disable MI100 tests for now
* fix syntax
* simplify container for results processing
* remove node(master) from the process_results stage
* put all stages in original order
* change the agent label from master to mici for gfx908
* Implement layernorm kernel and deviceOp
* verify gpu kernel with host code
* 1. Separate gamma aand beta from affine
2. Check if argument is valid
* clean
* Sync the naming
* Support sweep once mode if we can put k dimension data inside one block
* [What] Get length from upper length.
[Why] if we get length directly, we may get length after padding.
* We only use one block in K dimension.
Hence, we can simplify the indexing of global R/W.
* Use 1d descriptor for gamma and beta
* Add accElementwiseOp
* Extract layernorm host code
* Support different YVectorDim in GridwiseLayernorm
* Rename XSrcVectorDim to XYSrcVectorDim. Because we use same parameter in deviceOp
* Gamma and beta can share the VGPR.
* Add test for fp32 and fp16
* Fix bug of concurrency and add test case which may fail orignally
* Propagate NaN for layernorm
Co-authored-by: Chao Liu <chao.liu2@amd.com>
* use 'sweep once' softmax kernel where applicable
* threadwise copy's dst buffer can specify invalid element value
* add int8 in/out float compute softmax support
give a bit of leeway for int absolute tolerance as there's a single data point of all test cases showing off-by-1 error
* format
* softmax inherits DeviceNormalization
* softmax profiler stub
* tighten up reference softmax interface
* example prints tensor dimension
* add fp32 to softmax profiler
* rename header
* hook with ckProfiler
* format
* resolve merge conflict
* resolve merge conflicts
* update normalization profiler help string
* resolve conflict
* typo
* remove residual
* softmax profiler: address feedback
* test for mixed precision input/output
* fully qualify ck::math::isnan
* add comment for device normalization interface
* revise wording
* constness for alpha/beta scaler pointer
* Switch to standard ROCm packaging
* Revert .gitignore changes
* install new rocm-cmake version
* update readme
Co-authored-by: illsilin <Illia.Silin@amd.com>
Co-authored-by: Chao Liu <chao.liu2@amd.com>
