Bump LLVM to llvm/llvm-project@6b65d79 and deps (2023-12-29) (#16012)

- Bump llvm-projects to
llvm/llvm-project@6b65d79
- Bump torch-mlir to
llvm/torch-mlir@6096fcb

---------

Co-authored-by: Jakub Kuderski <jakub@nod-labs.com>
Co-authored-by: hanhanW <hanhan0912@gmail.com>

compiler/src/iree/compiler/Codegen/Common/ReplaceSlowMinMaxOps.cpp

@@ -41,13 +41,13 @@ struct ReplaceSlowWithFastReductionMinMaxOpPattern final
                                 PatternRewriter &rewriter) const override {
     if (slowReductionOp.getKind() == vector::CombiningKind::MINIMUMF) {
       rewriter.updateRootInPlace(slowReductionOp, [&]() {
-        slowReductionOp.setKind(vector::CombiningKind::MINF);
+        slowReductionOp.setKind(vector::CombiningKind::MINNUMF);
       });
       return success();
     }
     if (slowReductionOp.getKind() == vector::CombiningKind::MAXIMUMF) {
       rewriter.updateRootInPlace(slowReductionOp, [&]() {
-        slowReductionOp.setKind(vector::CombiningKind::MAXF);
+        slowReductionOp.setKind(vector::CombiningKind::MAXNUMF);
       });
       return success();
     }

compiler/src/iree/compiler/Codegen/Common/VectorLayoutAnalysis.cpp

@@ -224,7 +224,7 @@ ChangeResult DistributionLayout::resolveWithPossibleConflict(
   resolvedLayout->subscribeEnforcement(enforcement);
 
   // We can now resolve this resolved value to the required layout.
-  resolvedLayout->resolve(rhs);
+  (void)resolvedLayout->resolve(rhs);
 
   // No change actually needs to be propagated after a conflict resolution.
   // TODO: Ideally, there should be another state in the lattice which says

compiler/src/iree/compiler/Codegen/Common/test/replace_slow_min_max_ops.mlir

@@ -26,7 +26,7 @@ func.func @reduction_min(%arg0: vector<4xf32>) -> f32 {
   return %reduction_min : f32
 }
 
-// CHECK: vector.reduction <minf>
+// CHECK: vector.reduction <minnumf>
 
 // -----
 
@@ -36,7 +36,7 @@ func.func @reduction_max(%arg0: vector<4xf32>) -> f32 {
   return %reduction_max : f32
 }
 
-// CHECK: vector.reduction <maxf>
+// CHECK: vector.reduction <maxnumf>
 
 // -----
 
@@ -46,7 +46,7 @@ func.func @multi_reduction_min(%arg0: vector<4xf32>, %arg1: f32) -> f32 {
   return %multi_reduction_min : f32
 }
 
-// CHECK: vector.multi_reduction <minf>
+// CHECK: vector.multi_reduction <minnumf>
 
 // -----
 
@@ -56,4 +56,4 @@ func.func @multi_reduction_max(%arg0: vector<4xf32>, %arg1: f32) -> f32 {
   return %multi_reduction_max : f32
 }
 
-// CHECK: vector.multi_reduction <maxf>
+// CHECK: vector.multi_reduction <maxnumf>

compiler/src/iree/compiler/Codegen/Utils/GPUUtils.cpp

@@ -451,13 +451,13 @@ static TypedAttr getCombiningKindIdentity(OpBuilder &builder,
   case vector::CombiningKind::XOR:
     return builder.getZeroAttr(type);
   case vector::CombiningKind::MINIMUMF:
-  case vector::CombiningKind::MINF: {
+  case vector::CombiningKind::MINNUMF: {
     auto posInfApFloat = APFloat::getInf(
         llvm::cast<FloatType>(type).getFloatSemantics(), /*Negative=*/false);
     return builder.getFloatAttr(type, posInfApFloat);
   }
   case vector::CombiningKind::MAXIMUMF:
-  case vector::CombiningKind::MAXF: {
+  case vector::CombiningKind::MAXNUMF: {
     auto negInfApFloat = APFloat::getInf(
         llvm::cast<FloatType>(type).getFloatSemantics(), /*Negative=*/true);
     return builder.getFloatAttr(type, negInfApFloat);

compiler/src/iree/compiler/Dialect/Stream/IR/StreamOpFolders.cpp

@@ -1796,7 +1796,8 @@ void AsyncFillOp::getCanonicalizationPatterns(RewritePatternSet &results,
 //===----------------------------------------------------------------------===//
 
 OpFoldResult AsyncUpdateOp::fold(FoldAdaptor operands) {
-  if (getUpdateSize() == getTargetSize()) {
+  if (getUpdateSize() == getTargetSize() &&
+      getUpdate().getType() == getType()) {
     // If updating the entire target then just replace with the update.
     // Note that this breaks copy-on-write semantics but will be fixed up during
     // canonicalization if needed.