feat: new passes + up jll version (#1902)

Author: avik-pal

Project.toml

@@ -1,7 +1,7 @@
 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 = "0.2.179"
+version = "0.2.180"
 
 [deps]
 Adapt = "79e6a3ab-5dfb-504d-930d-738a2a938a0e"
@@ -105,7 +105,7 @@ PythonCall = "0.9.25"
 Random = "1.10"
 Random123 = "1.7"
 ReactantCore = "0.1.16"
-Reactant_jll = "0.0.264"
+Reactant_jll = "0.0.265"
 ScopedValues = "1.3.0"
 Scratch = "1.2"
 Sockets = "1.10"

src/Compiler.jl

@@ -809,6 +809,7 @@ function optimization_passes(
         "associative_binary_op_reordering<1>",
         "transpose_broadcast_in_dim_to_broadcast_in_dim<16>",
         "replace_neg_add_with_subtract",
+        "replace_subtract_neg_with_add",
         "binop_const_simplify",
         "not_select_simplify",
         "common_compare_expression_rewrite",
@@ -906,6 +907,8 @@ function optimization_passes(
         "enzyme_hlo_unroll($(WHILE_UNROLL_THRESHOLD[]))",
         "dot_general_only_diagonal_access",
         "transpose_symmetric_simplify",
+        "divide_negated_operands_simplify",
+        "multiply_negated_operands_simplify",
     ]
 
     if !compile_options.disable_auto_batching_passes
@@ -923,6 +926,7 @@ function optimization_passes(
                 "concat_insert_dim_sort",
                 "concat_insert_dim_reduce_window",
                 "concat_insert_dim_elementwise",
+                "concat_insert_dim_convolution",
                 "dot_general_slice_to_batch",
                 "gather_slice_to_batch",
                 "iota_slice_to_batch",
@@ -932,6 +936,7 @@ function optimization_passes(
                 "broadcastindim_slice_to_batch",
                 "reducewindow_slice_to_batch",
                 "elementwise_slice_to_batch",
+                "convolution_slice_to_batch",
                 "greedy_while_loop_batch_fission",
             ],
         )
@@ -953,6 +958,7 @@ function optimization_passes(
                 "reduce_licm(0)",
                 "reduce_window_licm(0)",
                 "reverse_licm(0)",
+                "convolution_licm(0)",
             ],
         )
     end