Skip to content

[Kernel][Build/CI] Bump CUTLASS to 3.8 and add initializers for cutlass epilogues #13797

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 1 commit into from
Feb 26, 2025
Merged

[Kernel][Build/CI] Bump CUTLASS to 3.8 and add initializers for cutlass epilogues #13797

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
8 changes: 4 additions & 4 deletions CMakeLists.txt
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -247,7 +247,7 @@ if(VLLM_GPU_LANG STREQUAL "CUDA")
cutlass
GIT_REPOSITORY https://github.com/nvidia/cutlass.git
# Please keep this in sync with CUTLASS_REVISION line above.
GIT_TAG v3.7.0
GIT_TAG v3.8.0
GIT_PROGRESS TRUE

# Speed up CUTLASS download by retrieving only the specified GIT_TAG instead of the history.
Expand Down Expand Up @@ -302,7 +302,7 @@ if(VLLM_GPU_LANG STREQUAL "CUDA")
# CUDA 12.0 or later (and only work on Hopper, 9.0a for now).
cuda_archs_loose_intersection(SCALED_MM_3X_ARCHS "9.0a" "${CUDA_ARCHS}")
if(${CMAKE_CUDA_COMPILER_VERSION} VERSION_GREATER 12.0 AND SCALED_MM_3X_ARCHS)
set(SRCS
set(SRCS
"csrc/quantization/cutlass_w8a8/scaled_mm_c3x.cu"
"csrc/quantization/cutlass_w8a8/c3x/scaled_mm_sm90_fp8.cu"
"csrc/quantization/cutlass_w8a8/c3x/scaled_mm_sm90_int8.cu"
Expand Down Expand Up @@ -382,7 +382,7 @@ if(VLLM_GPU_LANG STREQUAL "CUDA")
# FP4 Archs and flags
cuda_archs_loose_intersection(FP4_ARCHS "10.0a" "${CUDA_ARCHS}")
if(${CMAKE_CUDA_COMPILER_VERSION} VERSION_GREATER 12.8 AND FP4_ARCHS)
set(SRCS
set(SRCS
"csrc/quantization/fp4/nvfp4_quant_kernels.cu"
"csrc/quantization/fp4/nvfp4_scaled_mm_kernels.cu"
)
Expand Down Expand Up @@ -593,7 +593,7 @@ endif()

if(VLLM_FLASH_ATTN_SRC_DIR)
FetchContent_Declare(
vllm-flash-attn SOURCE_DIR
vllm-flash-attn SOURCE_DIR
${VLLM_FLASH_ATTN_SRC_DIR}
BINARY_DIR ${CMAKE_BINARY_DIR}/vllm-flash-attn
)
Expand Down
26 changes: 14 additions & 12 deletions csrc/cutlass_extensions/epilogue/scaled_mm_epilogues_c2x.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -122,8 +122,8 @@ struct ScaledEpilogue
auto a_args = SUPER::template args_from_tensor<ScaleA, float>(a_scales);
auto b_args = SUPER::template args_from_tensor<ScaleB, float>(b_scales);

typename EVTCompute0::Arguments evt0_args{b_args};
return ArgumentType{a_args, evt0_args};
typename EVTCompute0::Arguments evt0_args{b_args, {}, {}};
return ArgumentType{a_args, evt0_args, {}};
}
};

Expand Down Expand Up @@ -167,8 +167,8 @@ struct ScaledEpilogueBias
auto b_args = SUPER::template args_from_tensor<ScaleB, float>(b_scales);
auto bias_args = SUPER::template args_from_tensor<Bias, ElementD>(bias);

typename EVTCompute0::Arguments evt0_args{b_args};
return ArgumentType{a_args, evt0_args, bias_args};
typename EVTCompute0::Arguments evt0_args{b_args, {}, {}};
return ArgumentType{a_args, evt0_args, bias_args, {}};
}
};

Expand Down Expand Up @@ -230,9 +230,10 @@ struct ScaledEpilogueBiasAzp
auto azp_adj_args =
SUPER::template args_from_tensor<AzpWithAdj, int32_t>(azp_adj);

typename EVTComputeAzp::Arguments evt_azp_args{{}, azp_adj_args};
typename EVTComputeScaleB::Arguments evt_scale_b_args{b_args, evt_azp_args};
return ArgumentType{a_args, evt_scale_b_args, bias_args};
typename EVTComputeAzp::Arguments evt_azp_args{{}, azp_adj_args, {}};
typename EVTComputeScaleB::Arguments evt_scale_b_args{
b_args, evt_azp_args, {}};
return ArgumentType{a_args, evt_scale_b_args, bias_args, {}};
}
};

Expand Down Expand Up @@ -309,11 +310,12 @@ struct ScaledEpilogueBiasAzpToken
auto azp_adj_args =
SUPER::template args_from_tensor<AzpAdj, int32_t>(azp_adj);

typename EVTComputeAzp::Arguments evt_azp_args{azp_args, azp_adj_args};
typename EVTComputeAcc::Arguments evt_acc_args{{}, evt_azp_args};
typename EVTComputeScaleB::Arguments evt_scale_b_args{b_args, evt_acc_args};
return ArgumentType{a_args, evt_scale_b_args, bias_args};
typename EVTComputeAzp::Arguments evt_azp_args{azp_args, azp_adj_args, {}};
typename EVTComputeAcc::Arguments evt_acc_args{{}, evt_azp_args, {}};
typename EVTComputeScaleB::Arguments evt_scale_b_args{
b_args, evt_acc_args, {}};
return ArgumentType{a_args, evt_scale_b_args, bias_args, {}};
}
};

}; // namespace vllm::c2x
}; // namespace vllm::c2x
28 changes: 15 additions & 13 deletions csrc/cutlass_extensions/epilogue/scaled_mm_epilogues_c3x.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -146,8 +146,8 @@ struct ScaledEpilogue
auto a_args = SUPER::template args_from_tensor<ScaleA, float>(a_scales);
auto b_args = SUPER::template args_from_tensor<ScaleB, float>(b_scales);

typename EVTCompute0::Arguments evt0_args{b_args};
return ArgumentType{a_args, evt0_args};
typename EVTCompute0::Arguments evt0_args{b_args, {}, {}};
return ArgumentType{a_args, evt0_args, {}};
}
};

Expand Down Expand Up @@ -193,8 +193,8 @@ struct ScaledEpilogueBias
auto b_args = SUPER::template args_from_tensor<ScaleB, float>(b_scales);
auto bias_args = SUPER::template args_from_tensor<Bias, ElementD>(bias);

typename EVTCompute0::Arguments evt0_args{b_args};
return ArgumentType{a_args, evt0_args, bias_args};
typename EVTCompute0::Arguments evt0_args{b_args, {}, {}};
return ArgumentType{a_args, evt0_args, bias_args, {}};
}
};

Expand Down Expand Up @@ -236,8 +236,8 @@ struct ScaledEpilogueColumnBias
auto b_args = SUPER::template args_from_tensor<ScaleB, float>(b_scales);
auto bias_args = SUPER::template args_from_tensor<Bias, ElementD>(bias);

typename EVTCompute0::Arguments evt0_args{b_args};
return ArgumentType{a_args, evt0_args, bias_args};
typename EVTCompute0::Arguments evt0_args{b_args, {}, {}};
return ArgumentType{a_args, evt0_args, bias_args, {}};
}
};

Expand Down Expand Up @@ -297,9 +297,10 @@ struct ScaledEpilogueBiasAzp
auto azp_adj_args =
SUPER::template args_from_tensor<AzpWithAdj, int32_t>(azp_adj);

typename EVTComputeAzp::Arguments evt_azp_args{{}, azp_adj_args};
typename EVTComputeScaleB::Arguments evt_scale_b_args{b_args, evt_azp_args};
return ArgumentType{a_args, evt_scale_b_args, bias_args};
typename EVTComputeAzp::Arguments evt_azp_args{{}, azp_adj_args, {}};
typename EVTComputeScaleB::Arguments evt_scale_b_args{
b_args, evt_azp_args, {}};
return ArgumentType{a_args, evt_scale_b_args, bias_args, {}};
}
};

Expand Down Expand Up @@ -374,10 +375,11 @@ struct ScaledEpilogueBiasAzpToken
auto azp_adj_args =
SUPER::template args_from_tensor<AzpAdj, int32_t>(azp_adj);

typename EVTComputeAzp::Arguments evt_azp_args{azp_args, azp_adj_args};
typename EVTComputeAcc::Arguments evt_acc_args{{}, evt_azp_args};
typename EVTComputeScaleB::Arguments evt_scale_b_args{b_args, evt_acc_args};
return ArgumentType{a_args, evt_scale_b_args, bias_args};
typename EVTComputeAzp::Arguments evt_azp_args{azp_args, azp_adj_args, {}};
typename EVTComputeAcc::Arguments evt_acc_args{{}, evt_azp_args, {}};
typename EVTComputeScaleB::Arguments evt_scale_b_args{
b_args, evt_acc_args, {}};
return ArgumentType{a_args, evt_scale_b_args, bias_args, {}};
}
};

Expand Down