onnxruntime
[CPU/CUDA EP] Add DeformConv op support
#27393
Merged
Overview
Commits
58
Changes
View On GitHub
Commits
  • Add deform conv 2d cpu execution provider support
    ShirasawaSama committed 109 days ago
  • Add more tests
    ShirasawaSama committed 109 days ago
  • Add cuda support for deformconv2d
    ShirasawaSama committed 109 days ago
  • Improve deformconv cuda pref
    ShirasawaSama committed 109 days ago
  • Add more test cases
    ShirasawaSama committed 109 days ago
  • Fix copilot suggestions
    ShirasawaSama committed 109 days ago
  • Fix default attrs value of DeformConv
    ShirasawaSama committed 109 days ago
  • Fix schema definition for DeformConv op
    ShirasawaSama committed 109 days ago
  • Refactor DeformConv test cases
    ShirasawaSama committed 109 days ago
  • Fix OrtMemTypeCPUInput issue and add cuda error check
    ShirasawaSama committed 109 days ago
  • Remove GemmEx double specialization
    ShirasawaSama committed 109 days ago
  • Fix potential integer overflow in CUDA DeformableIm2ColKernel
    ShirasawaSama committed 109 days ago
  • Optimize CPU DeformableIm2Col loop order for better cache locality
    ShirasawaSama committed 109 days ago
  • Parallelize CPU DeformConv Im2Col and bias addition
    ShirasawaSama committed 109 days ago
  • Use GPU free memory in DeformConv temp memory heuristic
    ShirasawaSama committed 109 days ago
  • Extract DeformConvAttributes to shared header
    ShirasawaSama committed 109 days ago
  • DeformConv op shared attributes and validation
    ShirasawaSama committed 109 days ago
  • Refactor attributes/validation and optimize CUDA DeformConvIm2Col kernel
    ShirasawaSama committed 109 days ago
  • Add DeformConv OnnxModelTest with reference ONNX model
    ShirasawaSama committed 109 days ago
  • Optimize GetGreatestDivisorBelowBound in CUDA DeformConv
    ShirasawaSama committed 109 days ago
  • Document symmetric-padding-only limitation in deform_conv_test_gen
    ShirasawaSama committed 109 days ago
  • Skip cuda DeformConv op copy kernel when cur_parallel==1
    ShirasawaSama committed 109 days ago
  • Reformat code
    ShirasawaSama committed 109 days ago
  • Fix cuda fp16 test cases
    ShirasawaSama committed 109 days ago
  • Fix int64_t to ptrdiff_t conversion in deform_conv
    ShirasawaSama committed 109 days ago
  • Resolve pipeline failures caused by unit tests
    ShirasawaSama committed 109 days ago
  • Add comments and handle unused variables
    ShirasawaSama committed 109 days ago
  • Address review feedback and align with Conv behavior
    ShirasawaSama committed 109 days ago
  • Optimize DeformConv cpu bias add with Eigen SIMD
    ShirasawaSama committed 109 days ago
  • Document GEMM layout trick in DeformConv cuBLAS path
    ShirasawaSama committed 109 days ago
  • Use int64_t for bilinear interpolation indices
    ShirasawaSama committed 109 days ago
  • refactor(DeformConv CPU): template UseMask and improve im2col performance
    ShirasawaSama committed 109 days ago
  • perf(DeformConv CPU): optimize im2col and BilinearInterpolate
    ShirasawaSama committed 109 days ago
  • Early OOB check for BilinearInterpolate
    ShirasawaSama committed 109 days ago
  • Shrink DeformConv CUDA mutex to UpdateState only
    ShirasawaSama committed 109 days ago
  • Use cublasGemmStridedBatched for gemm_writes_directly path in DeformConv CUDA
    ShirasawaSama committed 109 days ago
  • Fix var name
    ShirasawaSama committed 109 days ago
  • Drop mask==0 branch in im2col to match CPU behavior
    ShirasawaSama committed 109 days ago
  • Add 1x1 im2col kernel specialization dispatch
    ShirasawaSama committed 109 days ago
  • Reformat codes
    ShirasawaSama committed 109 days ago
  • Fix C4244 in deform_conv_op_test by casting rtol/atol to float
    ShirasawaSama committed 109 days ago
  • Add standard MIT license header
    ShirasawaSama committed 109 days ago
  • Register DeformConv BFloat16 only for opset 22
    ShirasawaSama committed 109 days ago
  • Validate DeformConv input ranks and output size before indexing shapes
    ShirasawaSama committed 109 days ago
  • Refactor DeformConv MinimalBilinear tests with shared data/template and add optional bias omitted test
    ShirasawaSama committed 109 days ago
  • Validate optional bias B shape (1D [M]) in DeformConv shared helper for CPU/CUDA
    ShirasawaSama committed 109 days ago
  • Use cached totalGlobalMem for temp budget, remove cudaMemGetInfo and per-kernel state
    ShirasawaSama committed 109 days ago
  • Document int indices in CUDA BilinearInterpolate
    ShirasawaSama committed 109 days ago
  • Document why BFloat16 is not delegated in DeformConv CUDA impl
    ShirasawaSama committed 109 days ago
  • Document prime-batch fallback to single-image chunks in DeformConv GetNParallelImgs
    ShirasawaSama committed 109 days ago
  • Refine deform conv test generator imports and ONNX model save usage
    ShirasawaSama committed 109 days ago
  • Optimize DeformConv CPU bilinear interpolation
    ShirasawaSama committed 109 days ago
  • Optimize DeformConv BilinearInterpolation for performance on CUDA
    ShirasawaSama committed 109 days ago
  • Enforce 2D attribute lengths and validate kernel_shape/pads/overflow-safe shapes
    ShirasawaSama committed 109 days ago
  • Clarify DeformConv OnnxModelTest comment as ORT-reference smoke test
    ShirasawaSama committed 109 days ago
  • DeformConv EmptyBatch test expects failure when batch size N is zero
    ShirasawaSama committed 109 days ago
  • Allow DeformConv empty batch
    ShirasawaSama committed 109 days ago
  • Update docs
    ShirasawaSama committed 109 days ago
FAQ Terms Privacy Refunds Impressum
Loading