Integrate Triton up to [243522bd](https://github.com/openai/triton/commits/243522bd337fb2e3f4245d58452b0247629f0b58)

PiperOrigin-RevId: 661973865

Author: chsigg

third_party/triton/llvm_integration/cl657620552.patch

@@ -8,6 +8,5 @@ LLVM nor MLIR integrator, please do not add any patches to this list.
 """
 
 llvm_patch_list = [
-    "//third_party/triton:llvm_integration/cl657620552.patch",
     # Add new patches just above this line
 ]

third_party/triton/llvm_integration/series.bzl

@@ -14,7 +14,5 @@ those to this list.
 """
 
 temporary_patch_list = [
-    "//third_party/triton:temporary/cuda11-temporary.patch",
-    "//third_party/triton:temporary/undo_tesla_gpu.patch",
     # Add new patches just above this line
 ]

third_party/triton/temporary/cuda11-temporary.patch

@@ -8,8 +8,8 @@ load("//third_party/triton:xla_extensions/series.bzl", "extensions_files_patch_l
 def repo():
     """Imports Triton."""
 
-    TRITON_COMMIT = "cl657175856"
-    TRITON_SHA256 = "316f421a7d7ead2b7e5adc2e8bb68ce1a8f7809db73dbed8abd54c35bd0c1576"
+    TRITON_COMMIT = "cl659604537"
+    TRITON_SHA256 = "4f4699ca0df9d48649efcd838d331aa6bee4ec7fce905cdeed71715f6ca7e033"
     tf_http_archive(
         name = "triton",
         sha256 = TRITON_SHA256,

third_party/triton/temporary/series.bzl

@@ -273,9 +273,9 @@ index d74e0a224..4e45f7c4c 100644
 +  return op->hasTrait<OpTrait::DotLike>() || isa<ttg::SparseDotOp>(op);
 +}
 +
- // Replace the ForOp's yield with a new one with the given operands appended.
- static void appendToYield(scf::ForOp forOp, ArrayRef<Value> newOperands) {
-   // Fix up the yield op.
+ static void createAsyncCopy(scf::ForOp &forOp, tt::LoadOp loadOp, Value alloc,
+                             Value insertIdx, Value extractIdx,
+                             tt::CoarseSchedule &schedule,
 @@ -248,19 +252,28 @@ getSharedEncIfAllUsersAreDotEnc(Value val, bool &incompatible) {
      } else {
        if (!isa<ttg::LocalLoadOp, ttg::ConvertLayoutOp>(user))