feat: update v5.2

NVIDIA · Dec 2, 2022 · 4c014fb · 4c014fb
1 parent b835743
commit 4c014fb
Show file tree

Hide file tree

Showing 841 changed files with 1,892,441 additions and 257,080 deletions.
diff --git a/.gitignore b/.gitignore
@@ -6,3 +6,8 @@ __pycache__/
 .vscode
 ./translation
 .cache
+*.npy
+*.pth
+!tests/data/**/*.npy
+/models
+**/.ipynb_checkpoints/
diff --git a/.gitmodules b/.gitmodules
@@ -11,3 +11,6 @@
 [submodule "examples/pytorch/vit/ViT-quantization/ViT-pytorch"]
 	path = examples/pytorch/vit/ViT-quantization/ViT-pytorch
 	url = https://github.com/jeonsworld/ViT-pytorch
+[submodule "3rdparty/cutlass"]
+	path = 3rdparty/cutlass
+	url = https://github.com/NVIDIA/cutlass.git
diff --git a/3rdparty/trt_fused_multihead_attention/CMakeLists.txt b/3rdparty/trt_fused_multihead_attention/CMakeLists.txt
@@ -20,7 +20,7 @@ set(trt_fused_multi_head_attention_files
     qkvToContext.cu
 )
 
-file(GLOB trt_fused_multi_head_attention_files ${trt_fused_multi_head_attention_files} *.sm*.cpp)
+file(GLOB trt_fused_multi_head_attention_files ${trt_fused_multi_head_attention_files} *sm*.cpp)
 
 add_library(trt_fused_multi_head_attention STATIC ${trt_fused_multi_head_attention_files})
 target_link_libraries(trt_fused_multi_head_attention PUBLIC -lcublas -lcudart)

diff --git a/3rdparty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_0_32_sm75.cubin.cpp b/3rdparty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_0_32_sm75.cubin.cpp
diff --git a/3rdparty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_0_32_sm80.cubin.cpp b/3rdparty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_0_32_sm80.cubin.cpp
diff --git a/3rdparty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_0_32_sm86.cubin.cpp b/3rdparty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_0_32_sm86.cubin.cpp
diff --git a/3rdparty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_0_64_sm70.cubin.cpp b/3rdparty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_0_64_sm70.cubin.cpp
diff --git a/3rdparty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_0_64_sm72.cubin.cpp b/3rdparty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_0_64_sm72.cubin.cpp
diff --git a/3rdparty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_0_64_sm75.cubin.cpp b/3rdparty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_0_64_sm75.cubin.cpp
diff --git a/3rdparty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_0_64_sm80.cubin.cpp b/3rdparty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_0_64_sm80.cubin.cpp
diff --git a/3rdparty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_0_64_sm86.cubin.cpp b/3rdparty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_0_64_sm86.cubin.cpp
diff --git a/...ty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_128_sm70.cubin.cpp b/...ty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_128_sm70.cubin.cpp
diff --git a/...ty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_128_sm72.cubin.cpp b/...ty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_128_sm72.cubin.cpp
diff --git a/...ty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_128_sm75.cubin.cpp b/...ty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_128_sm75.cubin.cpp
diff --git a/...ty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_128_sm80.cubin.cpp b/...ty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_128_sm80.cubin.cpp
diff --git a/...ty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_128_sm86.cubin.cpp b/...ty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_128_sm86.cubin.cpp
diff --git a/...ty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_128_sm89.cubin.cpp b/...ty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_128_sm89.cubin.cpp
diff --git a/...ty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_144_sm70.cubin.cpp b/...ty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_144_sm70.cubin.cpp
diff --git a/...ty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_144_sm72.cubin.cpp b/...ty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_144_sm72.cubin.cpp
diff --git a/...ty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_144_sm75.cubin.cpp b/...ty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_144_sm75.cubin.cpp
diff --git a/...ty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_144_sm80.cubin.cpp b/...ty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_144_sm80.cubin.cpp
diff --git a/...ty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_144_sm86.cubin.cpp b/...ty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_144_sm86.cubin.cpp
diff --git a/...ty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_144_sm89.cubin.cpp b/...ty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_144_sm89.cubin.cpp
diff --git a/...ty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_160_sm70.cubin.cpp b/...ty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_160_sm70.cubin.cpp
diff --git a/...ty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_160_sm72.cubin.cpp b/...ty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_160_sm72.cubin.cpp
diff --git a/...ty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_160_sm75.cubin.cpp b/...ty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_160_sm75.cubin.cpp
diff --git a/...ty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_160_sm80.cubin.cpp b/...ty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_160_sm80.cubin.cpp
diff --git a/...ty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_160_sm86.cubin.cpp b/...ty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_160_sm86.cubin.cpp
diff --git a/...ty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_160_sm89.cubin.cpp b/...ty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_160_sm89.cubin.cpp
diff --git a/...ty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_256_sm70.cubin.cpp b/...ty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_256_sm70.cubin.cpp
diff --git a/...ty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_256_sm72.cubin.cpp b/...ty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_256_sm72.cubin.cpp
diff --git a/...ty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_256_sm75.cubin.cpp b/...ty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_256_sm75.cubin.cpp
diff --git a/...ty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_256_sm80.cubin.cpp b/...ty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_256_sm80.cubin.cpp
diff --git a/...ty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_256_sm86.cubin.cpp b/...ty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_256_sm86.cubin.cpp
diff --git a/...ty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_256_sm89.cubin.cpp b/...ty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_256_sm89.cubin.cpp
diff --git a/...rty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_32_sm70.cubin.cpp b/...rty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_32_sm70.cubin.cpp
diff --git a/...rty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_32_sm72.cubin.cpp b/...rty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_32_sm72.cubin.cpp
diff --git a/...rty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_32_sm75.cubin.cpp b/...rty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_32_sm75.cubin.cpp
diff --git a/...rty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_32_sm80.cubin.cpp b/...rty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_32_sm80.cubin.cpp
diff --git a/...rty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_32_sm86.cubin.cpp b/...rty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_32_sm86.cubin.cpp
diff --git a/...rty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_32_sm89.cubin.cpp b/...rty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_32_sm89.cubin.cpp
diff --git a/...rty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_40_sm70.cubin.cpp b/...rty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_40_sm70.cubin.cpp
diff --git a/...rty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_40_sm72.cubin.cpp b/...rty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_40_sm72.cubin.cpp
diff --git a/...rty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_40_sm75.cubin.cpp b/...rty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_40_sm75.cubin.cpp
diff --git a/...rty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_40_sm80.cubin.cpp b/...rty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_40_sm80.cubin.cpp
diff --git a/...rty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_40_sm86.cubin.cpp b/...rty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_40_sm86.cubin.cpp
diff --git a/...rty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_40_sm89.cubin.cpp b/...rty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_40_sm89.cubin.cpp
diff --git a/...rty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_64_sm70.cubin.cpp b/...rty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_64_sm70.cubin.cpp
diff --git a/...rty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_64_sm72.cubin.cpp b/...rty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_64_sm72.cubin.cpp
diff --git a/...rty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_64_sm75.cubin.cpp b/...rty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_64_sm75.cubin.cpp
diff --git a/...rty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_64_sm80.cubin.cpp b/...rty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_64_sm80.cubin.cpp
diff --git a/...rty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_64_sm86.cubin.cpp b/...rty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_64_sm86.cubin.cpp
diff --git a/...rty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_64_sm89.cubin.cpp b/...rty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_64_sm89.cubin.cpp
diff --git a/...rty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_80_sm70.cubin.cpp b/...rty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_80_sm70.cubin.cpp
diff --git a/...rty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_80_sm72.cubin.cpp b/...rty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_80_sm72.cubin.cpp
diff --git a/...rty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_80_sm75.cubin.cpp b/...rty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_80_sm75.cubin.cpp
diff --git a/...rty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_80_sm80.cubin.cpp b/...rty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_80_sm80.cubin.cpp
diff --git a/...rty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_80_sm86.cubin.cpp b/...rty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_80_sm86.cubin.cpp
diff --git a/...rty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_80_sm89.cubin.cpp b/...rty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_80_sm89.cubin.cpp
diff --git a/3rdparty/trt_fused_multihead_attention/fmha_v2_fp16_Causal_128_32_sm70.cubin.cpp b/3rdparty/trt_fused_multihead_attention/fmha_v2_fp16_Causal_128_32_sm70.cubin.cpp
diff --git a/3rdparty/trt_fused_multihead_attention/fmha_v2_fp16_Causal_128_32_sm72.cubin.cpp b/3rdparty/trt_fused_multihead_attention/fmha_v2_fp16_Causal_128_32_sm72.cubin.cpp
diff --git a/3rdparty/trt_fused_multihead_attention/fmha_v2_fp16_Causal_128_32_sm75.cubin.cpp b/3rdparty/trt_fused_multihead_attention/fmha_v2_fp16_Causal_128_32_sm75.cubin.cpp
diff --git a/3rdparty/trt_fused_multihead_attention/fmha_v2_fp16_Causal_128_32_sm80.cubin.cpp b/3rdparty/trt_fused_multihead_attention/fmha_v2_fp16_Causal_128_32_sm80.cubin.cpp
diff --git a/3rdparty/trt_fused_multihead_attention/fmha_v2_fp16_Causal_128_32_sm86.cubin.cpp b/3rdparty/trt_fused_multihead_attention/fmha_v2_fp16_Causal_128_32_sm86.cubin.cpp
diff --git a/3rdparty/trt_fused_multihead_attention/fmha_v2_fp16_Causal_128_32_sm89.cubin.cpp b/3rdparty/trt_fused_multihead_attention/fmha_v2_fp16_Causal_128_32_sm89.cubin.cpp
diff --git a/3rdparty/trt_fused_multihead_attention/fmha_v2_fp16_Causal_128_40_sm70.cubin.cpp b/3rdparty/trt_fused_multihead_attention/fmha_v2_fp16_Causal_128_40_sm70.cubin.cpp
diff --git a/3rdparty/trt_fused_multihead_attention/fmha_v2_fp16_Causal_128_40_sm72.cubin.cpp b/3rdparty/trt_fused_multihead_attention/fmha_v2_fp16_Causal_128_40_sm72.cubin.cpp
diff --git a/3rdparty/trt_fused_multihead_attention/fmha_v2_fp16_Causal_128_40_sm75.cubin.cpp b/3rdparty/trt_fused_multihead_attention/fmha_v2_fp16_Causal_128_40_sm75.cubin.cpp
diff --git a/3rdparty/trt_fused_multihead_attention/fmha_v2_fp16_Causal_128_40_sm80.cubin.cpp b/3rdparty/trt_fused_multihead_attention/fmha_v2_fp16_Causal_128_40_sm80.cubin.cpp
diff --git a/3rdparty/trt_fused_multihead_attention/fmha_v2_fp16_Causal_128_40_sm86.cubin.cpp b/3rdparty/trt_fused_multihead_attention/fmha_v2_fp16_Causal_128_40_sm86.cubin.cpp
diff --git a/3rdparty/trt_fused_multihead_attention/fmha_v2_fp16_Causal_128_40_sm89.cubin.cpp b/3rdparty/trt_fused_multihead_attention/fmha_v2_fp16_Causal_128_40_sm89.cubin.cpp
diff --git a/3rdparty/trt_fused_multihead_attention/fmha_v2_fp16_Causal_128_64_sm70.cubin.cpp b/3rdparty/trt_fused_multihead_attention/fmha_v2_fp16_Causal_128_64_sm70.cubin.cpp
diff --git a/3rdparty/trt_fused_multihead_attention/fmha_v2_fp16_Causal_128_64_sm72.cubin.cpp b/3rdparty/trt_fused_multihead_attention/fmha_v2_fp16_Causal_128_64_sm72.cubin.cpp