diff --git a/include/triton/Tools/Sys/GetEnv.hpp b/include/triton/Tools/Sys/GetEnv.hpp index 5f5040195..53e421ef2 100644 --- a/include/triton/Tools/Sys/GetEnv.hpp +++ b/include/triton/Tools/Sys/GetEnv.hpp @@ -24,7 +24,9 @@ #include #include +#include #include + namespace triton { const std::set ENV_VARS = { diff --git a/lib/Dialect/TritonGPU/IR/Dialect.cpp b/lib/Dialect/TritonGPU/IR/Dialect.cpp index b4daaee79..60da73a22 100644 --- a/lib/Dialect/TritonGPU/IR/Dialect.cpp +++ b/lib/Dialect/TritonGPU/IR/Dialect.cpp @@ -1460,9 +1460,10 @@ struct TritonGPUInferLayoutInterface return success(); } - LogicalResult inferDotOpEncoding(Attribute operandEncoding, unsigned opIdx, - Attribute retEncoding, - Optional location) const override { + LogicalResult + inferDotOpEncoding(Attribute operandEncoding, unsigned opIdx, + Attribute retEncoding, + std::optional location) const override { auto mmaRetEncoding = retEncoding.dyn_cast(); if (mmaRetEncoding && mmaRetEncoding.isHopper()) { // TODO: support gmma when A/B does not reside in shared memory diff --git a/lib/Dialect/TritonNvidiaGPU/Transforms/WSMaterialization.cpp b/lib/Dialect/TritonNvidiaGPU/Transforms/WSMaterialization.cpp index bd92ba372..9ebc78497 100644 --- a/lib/Dialect/TritonNvidiaGPU/Transforms/WSMaterialization.cpp +++ b/lib/Dialect/TritonNvidiaGPU/Transforms/WSMaterialization.cpp @@ -30,6 +30,8 @@ #include "triton/Dialect/TritonNvidiaGPU/IR/Dialect.h" #include "triton/Dialect/TritonNvidiaGPU/Transforms/Utility.h" +#include + using namespace mlir; namespace ttg = triton::gpu; namespace ttng = triton::nvidia_gpu; diff --git a/python/setup.py b/python/setup.py index 19df06997..18764ec13 100644 --- a/python/setup.py +++ b/python/setup.py @@ -335,6 +335,7 @@ setup( "matplotlib", "pandas", "tabulate", + "torch", ], }, )