Use recursion to fix inference failure

wip

.buildkite/pipeline.yml

@@ -693,7 +693,6 @@ steps:
       - label: "Unit: matrix field broadcasting (CPU)"
         key: unit_matrix_field_broadcasting_cpu_scalar_14
         command: "julia --color=yes --check-bounds=yes --project=test test/MatrixFields/matrix_fields_broadcasting/test_scalar_14.jl"
-        soft_fail: true
 
       - label: "Unit: matrix field broadcasting (CPU)"
         key: unit_matrix_field_broadcasting_cpu_scalar_15
@@ -819,7 +818,6 @@ steps:
       - label: "Unit: matrix field broadcasting (GPU)"
         key: unit_matrix_field_broadcasting_gpu_scalar_14
         command: "julia --color=yes --check-bounds=yes --project=test test/MatrixFields/matrix_fields_broadcasting/test_scalar_14.jl"
-        soft_fail: true
         agents:
           slurm_gpus: 1
           slurm_mem: 10GB

src/MatrixFields/matrix_multiplication.jl

@@ -306,6 +306,34 @@ boundary_modified_ud(_, ud, column_space, i) = ud
 boundary_modified_ud(::BottomRightMatrixCorner, ud, column_space, i) =
     min(Operators.right_idx(column_space) - i, ud)
 
+@inline matrix_rows(ld1, ud1, args) =
+    _matrix_rows(ntuple(i -> i + ld1 - 1, ud1 - ld1 + 1), args)
+Base.@propagate_inbounds function _matrix_rows(
+    d::Union{Int, PlusHalf{Int}},
+    args,
+)
+    (
+        bc,
+        boundary_modified_ld1,
+        boundary_modified_ud1,
+        space,
+        matrix2,
+        loc,
+        idx,
+        hidx,
+    ) = args
+    if isnothing(bc) || boundary_modified_ld1 <= d <= boundary_modified_ud1
+        @inbounds Operators.getidx(space, matrix2, loc, idx + d, hidx)
+    else
+        zero(eltype(matrix2)) # This row is outside the matrix.
+    end
+end
+Base.@propagate_inbounds _matrix_rows(tup::Tuple, args) =
+    (_matrix_rows(first(tup), args), _matrix_rows(Base.tail(tup), args)...)
+Base.@propagate_inbounds _matrix_rows(tup::Tuple{<:Any}, args) =
+    (_matrix_rows(first(tup), args),)
+@inline _matrix_rows(tup::Tuple{}, args) = ()
+
 # TODO: Use @propagate_inbounds here, and remove @inbounds from this function.
 # As of Julia 1.8, doing this increases compilation time by more than an order
 # of magnitude, and it also makes type inference fail for some complicated
@@ -341,16 +369,20 @@ function multiply_matrix_at_index(loc, space, idx, hidx, matrix1, arg, bc)
         # of as a map from boundary_modified_ld1 to boundary_modified_ud1. For
         # simplicity, use zero padding for rows that are outside the matrix.
         # Wrap the rows in a BandMatrixRow so that they can be easily indexed.
-        matrix2_rows = map((ld1:ud1...,)) do d
-            # TODO: Use @propagate_inbounds_meta instead of @inline_meta.
-            Base.@_inline_meta
-            if isnothing(bc) ||
-               boundary_modified_ld1 <= d <= boundary_modified_ud1
-                @inbounds Operators.getidx(space, matrix2, loc, idx + d, hidx)
-            else
-                zero(eltype(matrix2)) # This row is outside the matrix.
-            end
-        end
+        matrix2_rows = matrix_rows(
+            ld1,
+            ud1,
+            (
+                bc,
+                boundary_modified_ld1,
+                boundary_modified_ud1,
+                space,
+                matrix2,
+                loc,
+                idx,
+                hidx,
+            ),
+        )
         matrix2_rows_wrapper = BandMatrixRow{ld1}(matrix2_rows...)
 
         # Precompute the zero value to avoid inference issues caused by passing
@@ -443,4 +475,7 @@ if hasfield(Method, :recursion_relation)
     for m in methods(multiply_matrix_at_index)
         m.recursion_relation = dont_limit
     end
+    for m in methods(matrix_rows)
+        m.recursion_relation = dont_limit
+    end
 end