Make CodeLoopSinking deterministic

Makes code loop sinking pass more deterministic for the
easier comparisons.

Author: stefan-il

IGC/Compiler/CISACodeGen/CodeSinking.hpp

@@ -198,10 +198,10 @@ namespace IGC {
             }
         };
 
-        typedef llvm::SmallVector<std::unique_ptr<Candidate>, 64> CandidateVec;
-        typedef llvm::SmallVector<Candidate*, 64> CandidatePtrVec;
-        typedef llvm::DenseSet<Candidate*> CandidatePtrSet;
-        typedef llvm::DenseMap<Instruction*, Candidate*> InstToCandidateMap;
+        using CandidateVec = llvm::SmallVector<std::unique_ptr<Candidate>, 64>;
+        using CandidatePtrVec = llvm::SmallVector<Candidate *, 64>;
+        using CandidatePtrSet = llvm::DenseSet<Candidate *>;
+        using InstToCandidateMap = llvm::MapVector<Instruction *, Candidate *>;
 
         /// sinking
         bool loopSink(llvm::Function& F);

IGC/Compiler/tests/CodeSinking/LoopSinking/determinism.ll

@@ -0,0 +1,46 @@
+;=========================== begin_copyright_notice ============================
+;
+; Copyright (C) 2025 Intel Corporation
+;
+; SPDX-License-Identifier: MIT
+;
+;============================ end_copyright_notice =============================
+; REQUIRES: llvm-14-plus, regkeys, system-linux
+; RUN: igc_opt --opaque-pointers --regkey ForceLoopSink=1 --regkey LoopSinkMinSave=1 --regkey LoopSinkThresholdDelta=10 --regkey CodeLoopSinkingMinSize=10 --regkey PrepopulateLoadChainLoopSink=1 --igc-wi-analysis --basic-aa --igc-code-loop-sinking -S %s > %t.test1.ll
+; RUN: igc_opt --opaque-pointers --regkey ForceLoopSink=1 --regkey LoopSinkMinSave=1 --regkey LoopSinkThresholdDelta=10 --regkey CodeLoopSinkingMinSize=10 --regkey PrepopulateLoadChainLoopSink=1 --igc-wi-analysis --basic-aa --igc-code-loop-sinking -S %s > %t.test2.ll
+
+; Compares results of two runs of load sinking to ensure that results are deterministic.
+; RUN: diff %t.test1.ll %t.test2.ll
+define spir_kernel void @dsmm23x23x23(i16 %localIdX) {
+
+entry:
+  %conv2 = zext i16 %localIdX to i32
+  %idxprom36 = zext i16 %localIdX to i64
+  %add33.1 = add nuw nsw i32 %conv2, 23
+  %idxprom34.1 = zext i32 %add33.1 to i64
+  %add33.2 = add nuw nsw i32 %conv2, 46
+  %idxprom34.2 = zext i32 %add33.2 to i64
+  %add33.3 = add nuw nsw i32 %conv2, 69
+  %idxprom34.3 = zext i32 %add33.3 to i64
+  %0 = shl nuw nsw i64 %idxprom36, 3
+  %1 = shl nuw nsw i64 %idxprom34.1, 3
+  %2 = shl nuw nsw i64 %idxprom34.2, 3
+  %3 = shl nuw nsw i64 %idxprom34.3, 3
+  br label %for.body
+
+for.body:                                         ; preds = %for.body, %entry
+  %4 = add i64 0, 0
+  %5 = add i64 %4, %0
+  %6 = inttoptr i64 %5 to ptr addrspace(2)
+  %7 = load double, ptr addrspace(2) %6, align 8
+  %8 = add i64 %4, %1
+  %9 = inttoptr i64 %8 to ptr addrspace(2)
+  %10 = load double, ptr addrspace(2) %9, align 8
+  %11 = add i64 %4, %2
+  %12 = inttoptr i64 %11 to ptr addrspace(2)
+  %13 = load double, ptr addrspace(2) %12, align 8
+  %14 = add i64 %4, %3
+  %15 = inttoptr i64 %14 to ptr addrspace(2)
+  %16 = load double, ptr addrspace(2) %15, align 8
+  br label %for.body
+}