Tianlei Wu 6ffaaebb60 [CUDA] Attention kernel provider option (#21344) ... ### Description * Add a cuda provider option `sdpa_kernel` to choose which attention kernel to run for testing purpose. * Allow dump which attention kernel is used per node. * Reserve a flag for cudnn flash attention which will be added soon. #### CUDA provider option sdpa_kernel Instead of setting environment variable, we also support setting it in provider option. Note that the setting is global per session. That could help performance testing of each kernel. #### Attention Kernel Debug Info Set an environment variable `ORT_ENABLE_ATTENTION_KERNEL_DEBUG_INFO=1`, and ORT will print sdpa kernel used in each node: For example ``` ORT_ENABLE_ATTENTION_KERNEL_DEBUG_INFO=1 ./onnxruntime_test_all --gtest_filter=MultiHeadAttentionTest* ``` It will show debug information of kernel used in testing: ``` [ RUN ] MultiHeadAttentionTest.SelfAttention_Batch2_HeadSize32_NoBias_NoMask_PackedQKV AttentionKernelOptions: FLASH_ATTENTION=0 EFFICIENT_ATTENTION=0 TRT_FUSED_ATTENTION=1 CUDNN_FLASH_ATTENTION=0 TRT_FLASH_ATTENTION=1 TRT_CROSS_ATTENTION=0 TRT_CAUSAL_ATTENTION=0 MATH=1 Operator=MultiHeadAttention Node=node1 DataType=fp16 TRT_FUSED_ATTENTION=1 AttentionKernelOptions: FLASH_ATTENTION=0 EFFICIENT_ATTENTION=1 TRT_FUSED_ATTENTION=0 CUDNN_FLASH_ATTENTION=0 TRT_FLASH_ATTENTION=0 TRT_CROSS_ATTENTION=0 TRT_CAUSAL_ATTENTION=0 MATH=1 Operator=MultiHeadAttention Node=node1 DataType=fp16 EFFICIENT_ATTENTION=1 ``` In this test case, the debug info shows that one session uses trt fused attention and another session use efficient attention.		2024-07-19 13:58:54 -07:00
..
onnxruntime/core	[CUDA] Attention kernel provider option (#21344)	2024-07-19 13:58:54 -07:00