EnzymeAD · avik-pal · Jan 21, 2026 · Jan 21, 2026 · Jan 21, 2026
diff --git a/Project.toml b/Project.toml
@@ -1,7 +1,10 @@
 name = "Reactant"
 uuid = "3c362404-f566-11ee-1572-e11a4b42c853"
-authors = ["William Moses <wmoses@mit.edu>", "Valentin Churavy <vchuravy@mit.edu>", "Sergio Sánchez Ramírez <sergio.sanchez.ramirez@bsc.es>", "Paul Berg <paul@plutojl.org>", "Avik Pal <avikpal@mit.edu>", "Mosè Giordano <mose@gnu.org>"]
 version = "0.2.203"
+authors = ["William Moses <wmoses@mit.edu>", "Valentin Churavy <vchuravy@mit.edu>", "Sergio Sánchez Ramírez <sergio.sanchez.ramirez@bsc.es>", "Paul Berg <paul@plutojl.org>", "Avik Pal <avikpal@mit.edu>", "Mosè Giordano <mose@gnu.org>"]
+
+[workspace]
+projects = ["docs", "test", "benchmark"]
 
 [deps]
 Adapt = "79e6a3ab-5dfb-504d-930d-738a2a938a0e"
@@ -30,7 +33,6 @@ ReactantCore = "a3311ec8-5e00-46d5-b541-4f83e724a433"
 Reactant_jll = "0192cb87-2b54-54ad-80e0-3be72ad8a3c0"
 ScopedValues = "7e506255-f358-4e82-b7e4-beb19740aa63"
 Scratch = "6c6a2e73-6563-6170-7368-637461726353"
-Setfield = "efcf1570-3423-57d1-acb7-fd33fddbac46"
 Sockets = "6462fe0b-24de-5631-8697-dd941f90decc"
 UUIDs = "cf7118a7-6976-5b1a-9a39-7adc72f591a4"
 p7zip_jll = "3f19e933-33d8-53b3-aaab-bd5110c3b7a0"
@@ -124,7 +126,6 @@ ReactantCore = "0.1.16"
 Reactant_jll = "0.0.305"
 ScopedValues = "1.3.0"
 Scratch = "1.2"
-Setfield = "1.1.2"
 Sockets = "1.10"
 SparseArrays = "1.10"
 SpecialFunctions = "2.4"
@@ -141,6 +142,3 @@ ArrayInterface = "4fba245c-0d91-5ea0-9b3e-6abc04ee57a9"
 CUDA = "052768ef-5323-5732-b1bb-66c8b64840ba"
 NNlib = "872c559c-99b0-510c-b3b7-b6c96a88d5cd"
 PythonCall = "6099a3de-0909-46bc-b1f4-468b9a2dfc0d"
-
-[workspace]
-projects = ["docs", "test", "benchmark"]
diff --git a/src/Compiler.jl b/src/Compiler.jl
@@ -3763,8 +3763,8 @@ function compile_xla(
 
             exec = XLA.compile(
                 client,
-                mod;
-                compile_options=xla_compile_options,
+                mod,
+                xla_compile_options;
                 num_outputs=length(mlir_fn_res.linear_results),
                 num_parameters=length(mlir_fn_res.linear_args),
                 mlir_fn_res.is_sharded,

diff --git a/src/proto/google/protobuf/any_pb.jl b/src/proto/google/protobuf/any_pb.jl
@@ -5,7 +5,7 @@ using ProtoBuf.EnumX: @enumx
 export var"#Any"
 
 
-struct var"#Any"
+mutable struct var"#Any"
     type_url::String
     value::Vector{UInt8}
 end

diff --git a/src/proto/google/protobuf/duration_pb.jl b/src/proto/google/protobuf/duration_pb.jl
@@ -5,7 +5,7 @@ using ProtoBuf.EnumX: @enumx
 export Duration
 
 
-struct Duration
+mutable struct Duration
     seconds::Int64
     nanos::Int32
 end

diff --git a/src/proto/google/protobuf/timestamp_pb.jl b/src/proto/google/protobuf/timestamp_pb.jl
@@ -5,7 +5,7 @@ using ProtoBuf.EnumX: @enumx
 export Timestamp
 
 
-struct Timestamp
+mutable struct Timestamp
     seconds::Int64
     nanos::Int32
 end

diff --git a/src/proto/google/protobuf/wrappers_pb.jl b/src/proto/google/protobuf/wrappers_pb.jl
@@ -6,7 +6,7 @@ export BoolValue, Int64Value, FloatValue, Int32Value, DoubleValue, UInt64Value,
 export BytesValue, StringValue
 
 
-struct BoolValue
+mutable struct BoolValue
     value::Bool
 end
 PB.default_values(::Type{BoolValue}) = (;value = false)
@@ -36,7 +36,7 @@ function PB._encoded_size(x::BoolValue)
     return encoded_size
 end
 
-struct Int64Value
+mutable struct Int64Value
     value::Int64
 end
 PB.default_values(::Type{Int64Value}) = (;value = zero(Int64))
@@ -66,7 +66,7 @@ function PB._encoded_size(x::Int64Value)
     return encoded_size
 end
 
-struct FloatValue
+mutable struct FloatValue
     value::Float32
 end
 PB.default_values(::Type{FloatValue}) = (;value = zero(Float32))
@@ -96,7 +96,7 @@ function PB._encoded_size(x::FloatValue)
     return encoded_size
 end
 
-struct Int32Value
+mutable struct Int32Value
     value::Int32
 end
 PB.default_values(::Type{Int32Value}) = (;value = zero(Int32))
@@ -126,7 +126,7 @@ function PB._encoded_size(x::Int32Value)
     return encoded_size
 end
 
-struct DoubleValue
+mutable struct DoubleValue
     value::Float64
 end
 PB.default_values(::Type{DoubleValue}) = (;value = zero(Float64))
@@ -156,7 +156,7 @@ function PB._encoded_size(x::DoubleValue)
     return encoded_size
 end
 
-struct UInt64Value
+mutable struct UInt64Value
     value::UInt64
 end
 PB.default_values(::Type{UInt64Value}) = (;value = zero(UInt64))
@@ -186,7 +186,7 @@ function PB._encoded_size(x::UInt64Value)
     return encoded_size
 end
 
-struct UInt32Value
+mutable struct UInt32Value
     value::UInt32
 end
 PB.default_values(::Type{UInt32Value}) = (;value = zero(UInt32))
@@ -216,7 +216,7 @@ function PB._encoded_size(x::UInt32Value)
     return encoded_size
 end
 
-struct BytesValue
+mutable struct BytesValue
     value::Vector{UInt8}
 end
 PB.default_values(::Type{BytesValue}) = (;value = UInt8[])
@@ -246,7 +246,7 @@ function PB._encoded_size(x::BytesValue)
     return encoded_size
 end
 
-struct StringValue
+mutable struct StringValue
     value::String
 end
 PB.default_values(::Type{StringValue}) = (;value = "")

diff --git a/src/proto/stream_executor/cuda_compute_capability_pb.jl b/src/proto/stream_executor/cuda_compute_capability_pb.jl
@@ -7,7 +7,7 @@ export var"CudaComputeCapabilityProto.FeatureExtension", CudaComputeCapabilityPr
 
 @enumx var"CudaComputeCapabilityProto.FeatureExtension" UNSPECIFIED=0 NONE=1 ACCELERATED_FEATURES=2 FAMILY_COMPATIBLE_FEATURES=3
 
-struct CudaComputeCapabilityProto
+mutable struct CudaComputeCapabilityProto
     major::Int32
     minor::Int32
     feature_extension::var"CudaComputeCapabilityProto.FeatureExtension".T

diff --git a/src/proto/stream_executor/device_description_pb.jl b/src/proto/stream_executor/device_description_pb.jl
@@ -6,7 +6,7 @@ export RocmComputeCapabilityProto, DnnVersionInfoProto, RuntimeVersionProto
 export GpuDeviceInfoProto, GpuComputeCapabilityProto, GpuTargetConfigProto
 
 
-struct RocmComputeCapabilityProto
+mutable struct RocmComputeCapabilityProto
     gcn_arch_name::String
 end
 PB.default_values(::Type{RocmComputeCapabilityProto}) = (;gcn_arch_name = "")
@@ -36,7 +36,7 @@ function PB._encoded_size(x::RocmComputeCapabilityProto)
     return encoded_size
 end
 
-struct DnnVersionInfoProto
+mutable struct DnnVersionInfoProto
     major::Int32
     minor::Int32
     patch::Int32
@@ -78,7 +78,7 @@ function PB._encoded_size(x::DnnVersionInfoProto)
     return encoded_size
 end
 
-struct RuntimeVersionProto
+mutable struct RuntimeVersionProto
     major::Int32
     minor::Int32
     patch::Int32
@@ -120,7 +120,7 @@ function PB._encoded_size(x::RuntimeVersionProto)
     return encoded_size
 end
 
-struct GpuDeviceInfoProto
+mutable struct GpuDeviceInfoProto
     threads_per_block_limit::Int32
     threads_per_warp::Int32
     shared_memory_per_block::Int32
@@ -267,7 +267,7 @@ function PB._encoded_size(x::GpuDeviceInfoProto)
     return encoded_size
 end
 
-struct GpuComputeCapabilityProto
+mutable struct GpuComputeCapabilityProto
     compute_capability::Union{Nothing,OneOf{<:Union{CudaComputeCapabilityProto,RocmComputeCapabilityProto}}}
 end
 PB.oneof_field_types(::Type{GpuComputeCapabilityProto}) = (;
@@ -312,7 +312,7 @@ function PB._encoded_size(x::GpuComputeCapabilityProto)
     return encoded_size
 end
 
-struct GpuTargetConfigProto
+mutable struct GpuTargetConfigProto
     gpu_device_info::Union{Nothing,GpuDeviceInfoProto}
     platform_name::String
     dnn_version_info::Union{Nothing,DnnVersionInfoProto}

diff --git a/src/proto/tensorflow/profiler/diagnostics_pb.jl b/src/proto/tensorflow/profiler/diagnostics_pb.jl
@@ -5,7 +5,7 @@ using ProtoBuf.EnumX: @enumx
 export Diagnostics
 
 
-struct Diagnostics
+mutable struct Diagnostics
     info::Vector{String}
     warnings::Vector{String}
     errors::Vector{String}

diff --git a/src/proto/tensorflow/profiler/hardware_types_pb.jl b/src/proto/tensorflow/profiler/hardware_types_pb.jl
@@ -7,7 +7,7 @@ export HardwareType, GPUComputeCapability, DeviceCapabilities
 
 @enumx HardwareType UNKNOWN_HARDWARE=0 CPU_ONLY=1 GPU=2 TPU=3
 
-struct GPUComputeCapability
+mutable struct GPUComputeCapability
     major::UInt32
     minor::UInt32
 end
@@ -43,7 +43,7 @@ function PB._encoded_size(x::GPUComputeCapability)
     return encoded_size
 end
 
-struct DeviceCapabilities
+mutable struct DeviceCapabilities
     clock_rate_in_ghz::Float64
     num_cores::UInt32
     memory_size_in_bytes::UInt64

diff --git a/src/proto/tensorflow/profiler/hlo_stats/hlo_stats_pb.jl b/src/proto/tensorflow/profiler/hlo_stats/hlo_stats_pb.jl
@@ -5,7 +5,7 @@ using ProtoBuf.EnumX: @enumx
 export HloStatsRecord, HloStatsDatabase
 
 
-struct HloStatsRecord
+mutable struct HloStatsRecord
     rank::UInt64
     program_id::UInt64
     hlo_category::String
@@ -240,7 +240,7 @@ function PB._encoded_size(x::HloStatsRecord)
     return encoded_size
 end
 
-struct HloStatsDatabase
+mutable struct HloStatsDatabase
     hlo_stats_record::Vector{HloStatsRecord}
 end
 PB.default_values(::Type{HloStatsDatabase}) = (;hlo_stats_record = Vector{HloStatsRecord}())

diff --git a/src/proto/tensorflow/profiler/input_pipeline_pb.jl b/src/proto/tensorflow/profiler/input_pipeline_pb.jl
@@ -7,7 +7,7 @@ export InputPipelineAnalysisRecommendation, InputOpDetails, StepSummary
 export GenericStepTimeBreakdown, InputPipelineAnalysisResult
 
 
-struct InputTimeBreakdown
+mutable struct InputTimeBreakdown
     demanded_file_read_us::Float64
     advanced_file_read_us::Float64
     preprocessing_us::Float64
@@ -61,7 +61,7 @@ function PB._encoded_size(x::InputTimeBreakdown)
     return encoded_size
 end
 
-struct PerGenericStepDetails
+mutable struct PerGenericStepDetails
     step_number::Int32
     step_name::String
     step_time_ms::Float64
@@ -164,7 +164,7 @@ function PB._encoded_size(x::PerGenericStepDetails)
     return encoded_size
 end
 
-struct BottleneckAnalysis
+mutable struct BottleneckAnalysis
     input_percent::Float64
     output_percent::Float64
     idle_percent::Float64
@@ -260,7 +260,7 @@ function PB._encoded_size(x::BottleneckAnalysis)
     return encoded_size
 end
 
-struct InputPipelineAnalysisRecommendation
+mutable struct InputPipelineAnalysisRecommendation
     details::Vector{String}
     bottleneck_analysis::Union{Nothing,google.protobuf.var"#Any"}
     summary_next_step::String
@@ -302,7 +302,7 @@ function PB._encoded_size(x::InputPipelineAnalysisRecommendation)
     return encoded_size
 end
 
-struct InputOpDetails
+mutable struct InputOpDetails
     op_name::String
     count::UInt64
     time_in_ms::Float64
@@ -368,7 +368,7 @@ function PB._encoded_size(x::InputOpDetails)
     return encoded_size
 end
 
-struct StepSummary
+mutable struct StepSummary
     average::Float64
     standard_deviation::Float64
     minimum::Float64
@@ -416,7 +416,7 @@ function PB._encoded_size(x::StepSummary)
     return encoded_size
 end
 
-struct GenericStepTimeBreakdown
+mutable struct GenericStepTimeBreakdown
     unknown_time_ms_summary::Union{Nothing,StepSummary}
     host_wait_input_ms_summary::Union{Nothing,StepSummary}
     host_to_device_ms_summary::Union{Nothing,StepSummary}
@@ -507,7 +507,7 @@ function PB._encoded_size(x::GenericStepTimeBreakdown)
     return encoded_size
 end
 
-struct InputPipelineAnalysisResult
+mutable struct InputPipelineAnalysisResult
     tag::Bool
     hardware_type::String
     step_time_summary::Union{Nothing,StepSummary}

diff --git a/src/proto/tensorflow/profiler/kernel_stats_pb.jl b/src/proto/tensorflow/profiler/kernel_stats_pb.jl
@@ -5,7 +5,7 @@ using ProtoBuf.EnumX: @enumx
 export KernelReport, KernelStatsDb
 
 
-struct KernelReport
+mutable struct KernelReport
     name::String
     registers_per_thread::UInt32
     static_shmem_bytes::UInt32
@@ -113,7 +113,7 @@ function PB._encoded_size(x::KernelReport)
     return encoded_size
 end
 
-struct KernelStatsDb
+mutable struct KernelStatsDb
     reports::Vector{KernelReport}
 end
 PB.default_values(::Type{KernelStatsDb}) = (;reports = Vector{KernelReport}())