[LoopUnroll] Consider simplified operands while retrieving TTI instruction cost (#70929)

Get more precise cost of instruction after LoopUnroll considering that
some operands of it can be simplified, e.g. induction variable will be
replaced by constant after full unrolling.

Author: skachkov-sc

llvm/lib/Transforms/Scalar/LoopUnrollPass.cpp

@@ -450,7 +450,15 @@ static std::optional<EstimatedUnrollCost> analyzeLoopUnrollCost(
 
         // First accumulate the cost of this instruction.
         if (!Cost.IsFree) {
-          UnrolledCost += TTI.getInstructionCost(I, CostKind);
+          // Consider simplified operands in instruction cost.
+          SmallVector<Value *, 4> Operands;
+          transform(I->operands(), std::back_inserter(Operands),
+                    [&](Value *Op) {
+                      if (auto Res = SimplifiedValues.lookup(Op))
+                        return Res;
+                      return Op;
+                    });
+          UnrolledCost += TTI.getInstructionCost(I, Operands, CostKind);
           LLVM_DEBUG(dbgs() << "Adding cost of instruction (iteration "
                             << Iteration << "): ");
           LLVM_DEBUG(I->dump());

llvm/test/Transforms/LoopUnroll/AMDGPU/unroll-cost-addrspacecast.ll

@@ -1,4 +1,6 @@
-; RUN: opt -S -mtriple=amdgcn-unknown-amdhsa -mcpu=hawaii -passes=loop-unroll -unroll-threshold=49 -unroll-peel-count=0 -unroll-allow-partial=false -unroll-max-iteration-count-to-analyze=16 < %s | FileCheck %s
+; RUN: opt -S -mtriple=amdgcn-unknown-amdhsa -mcpu=hawaii -passes=loop-unroll -unroll-threshold=57 -unroll-peel-count=0 -unroll-allow-partial=false -unroll-max-iteration-count-to-analyze=16 < %s | FileCheck %s
+
+@indices = external global [16 x i32]
 
 ; CHECK-LABEL: @test_func_addrspacecast_cost_noop(
 ; CHECK-NOT: br i1
@@ -9,8 +11,10 @@ entry:
 for.body:
   %indvars.iv = phi i32 [ %indvars.iv.next, %for.body ], [ 0, %entry ]
   %sum.02 = phi float [ %fmul, %for.body ], [ 0.0, %entry ]
-  %arrayidx.in = getelementptr inbounds float, ptr addrspace(1) %in, i32 %indvars.iv
-  %arrayidx.out = getelementptr inbounds float, ptr addrspace(1) %out, i32 %indvars.iv
+  %idx.ptr = getelementptr inbounds [16 x i32], ptr @indices, i32 0, i32 %indvars.iv
+  %index = load i32, ptr %idx.ptr
+  %arrayidx.in = getelementptr inbounds float, ptr addrspace(1) %in, i32 %index
+  %arrayidx.out = getelementptr inbounds float, ptr addrspace(1) %out, i32 %index
   %cast.in = addrspacecast ptr addrspace(1) %arrayidx.in to ptr
   %cast.out = addrspacecast ptr addrspace(1) %arrayidx.out to ptr
   %load = load float, ptr %cast.in
@@ -34,8 +38,10 @@ entry:
 for.body:
   %indvars.iv = phi i32 [ %indvars.iv.next, %for.body ], [ 0, %entry ]
   %sum.02 = phi float [ %fmul, %for.body ], [ 0.0, %entry ]
-  %arrayidx.in = getelementptr inbounds float, ptr %in, i32 %indvars.iv
-  %arrayidx.out = getelementptr inbounds float, ptr %out, i32 %indvars.iv
+  %idx.ptr = getelementptr inbounds [16 x i32], ptr @indices, i32 0, i32 %indvars.iv
+  %index = load i32, ptr %idx.ptr
+  %arrayidx.in = getelementptr inbounds float, ptr %in, i32 %index
+  %arrayidx.out = getelementptr inbounds float, ptr %out, i32 %index
   %cast.in = addrspacecast ptr %arrayidx.in to ptr addrspace(3)
   %cast.out = addrspacecast ptr %arrayidx.out to ptr addrspace(3)
   %load = load float, ptr addrspace(3) %cast.in
@@ -58,8 +64,10 @@ entry:
 for.body:
   %indvars.iv = phi i32 [ %indvars.iv.next, %for.body ], [ 0, %entry ]
   %sum.02 = phi float [ %fmul, %for.body ], [ 0.0, %entry ]
-  %arrayidx.in = getelementptr inbounds float, ptr addrspace(3) %in, i32 %indvars.iv
-  %arrayidx.out = getelementptr inbounds float, ptr addrspace(3) %out, i32 %indvars.iv
+  %idx.ptr = getelementptr inbounds [16 x i32], ptr @indices, i32 0, i32 %indvars.iv
+  %index = load i32, ptr %idx.ptr
+  %arrayidx.in = getelementptr inbounds float, ptr addrspace(3) %in, i32 %index
+  %arrayidx.out = getelementptr inbounds float, ptr addrspace(3) %out, i32 %index
   %cast.in = addrspacecast ptr addrspace(3) %arrayidx.in to ptr
   %cast.out = addrspacecast ptr addrspace(3) %arrayidx.out to ptr
   %load = load float, ptr %cast.in

llvm/test/Transforms/LoopUnroll/RISCV/unroll-Os.ll

@@ -0,0 +1,35 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 2
+; RUN: opt < %s -S -mtriple=riscv64 -passes=loop-unroll | FileCheck %s
+
+; Function Attrs: optsize
+define void @foo(ptr %array, i32 %x) #0 {
+; CHECK-LABEL: define void @foo
+; CHECK-SAME: (ptr [[ARRAY:%.*]], i32 [[X:%.*]]) #[[ATTR0:[0-9]+]] {
+; CHECK-NEXT:  entry:
+; CHECK-NEXT:    br label [[FOR_BODY:%.*]]
+; CHECK:       for.body:
+; CHECK-NEXT:    store i32 [[X]], ptr [[ARRAY]], align 4
+; CHECK-NEXT:    [[ARRAYIDX_1:%.*]] = getelementptr inbounds i32, ptr [[ARRAY]], i64 1
+; CHECK-NEXT:    store i32 [[X]], ptr [[ARRAYIDX_1]], align 4
+; CHECK-NEXT:    [[ARRAYIDX_2:%.*]] = getelementptr inbounds i32, ptr [[ARRAY]], i64 2
+; CHECK-NEXT:    store i32 [[X]], ptr [[ARRAYIDX_2]], align 4
+; CHECK-NEXT:    [[ARRAYIDX_3:%.*]] = getelementptr inbounds i32, ptr [[ARRAY]], i64 3
+; CHECK-NEXT:    store i32 [[X]], ptr [[ARRAYIDX_3]], align 4
+; CHECK-NEXT:    ret void
+;
+entry:
+  br label %for.body
+
+for.body:
+  %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]
+  %arrayidx = getelementptr inbounds i32, ptr %array, i64 %indvars.iv
+  store i32 %x, ptr %arrayidx, align 4
+  %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
+  %exitcond.not = icmp eq i64 %indvars.iv.next, 4
+  br i1 %exitcond.not, label %for.cond.cleanup, label %for.body
+
+for.cond.cleanup:
+  ret void
+}
+
+attributes #0 = { optsize }