[llvm] 0a272d3 - [LV] Use SCEV to analyze second operand for cost query.

[Florian Hahn via llvm-commits]

Author: Florian Hahn
Date: 2024-08-29T12:08:27+01:00
New Revision: 0a272d3a1703415abca42dc2e2cc2b57cb30734e

URL: https://github.com/llvm/llvm-project/commit/0a272d3a1703415abca42dc2e2cc2b57cb30734e
DIFF: https://github.com/llvm/llvm-project/commit/0a272d3a1703415abca42dc2e2cc2b57cb30734e.diff

LOG: [LV] Use SCEV to analyze second operand for cost query.

Improve operand analysis using SCEV for cost purposes. This fixes a
divergence between legacy and VPlan-based cost-modeling after
533e6bbd0d34.

Fixes https://github.com/llvm/llvm-project/issues/106248.

Added: 
    llvm/test/Transforms/LoopVectorize/X86/cost-constant-known-via-scev.ll

Modified: 
    llvm/lib/Transforms/Vectorize/LoopVectorize.cpp

Removed: 
    


################################################################################
diff  --git a/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp b/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp
index 56f51e14a6eba9..4cc75e2e754603 100644
--- a/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp
+++ b/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp
@@ -6529,6 +6529,10 @@ LoopVectorizationCostModel::getInstructionCost(Instruction *I,
     // Certain instructions can be cheaper to vectorize if they have a constant
     // second vector operand. One example of this are shifts on x86.
     Value *Op2 = I->getOperand(1);
+    if (!isa<Constant>(Op2) && PSE.getSE()->isSCEVable(Op2->getType()) &&
+        isa<SCEVConstant>(PSE.getSCEV(Op2))) {
+      Op2 = cast<SCEVConstant>(PSE.getSCEV(Op2))->getValue();
+    }
     auto Op2Info = TTI.getOperandInfo(Op2);
     if (Op2Info.Kind == TargetTransformInfo::OK_AnyValue &&
         Legal->isInvariant(Op2))

diff  --git a/llvm/test/Transforms/LoopVectorize/X86/cost-constant-known-via-scev.ll b/llvm/test/Transforms/LoopVectorize/X86/cost-constant-known-via-scev.ll
new file mode 100644
index 00000000000000..4c10ac3cd519e9
--- /dev/null
+++ b/llvm/test/Transforms/LoopVectorize/X86/cost-constant-known-via-scev.ll
@@ -0,0 +1,66 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 5
+; RUN: opt -p loop-vectorize -S %s | FileCheck %s
+
+target datalayout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128"
+target triple = "x86_64-unknown-linux-gnu"
+
+; Test case for https://github.com/llvm/llvm-project/issues/106248.
+define i64 @test_foldable_live_in_via_scev() {
+; CHECK-LABEL: define i64 @test_foldable_live_in_via_scev() {
+; CHECK-NEXT:  [[ENTRY:.*]]:
+; CHECK-NEXT:    [[CONV:%.*]] = zext i16 -6 to i64
+; CHECK-NEXT:    [[ADD:%.*]] = add nsw i64 [[CONV]], -65528
+; CHECK-NEXT:    br i1 false, label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]]
+; CHECK:       [[VECTOR_PH]]:
+; CHECK-NEXT:    br label %[[VECTOR_BODY:.*]]
+; CHECK:       [[VECTOR_BODY]]:
+; CHECK-NEXT:    [[INDEX:%.*]] = phi i32 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]
+; CHECK-NEXT:    [[VEC_PHI:%.*]] = phi <2 x i64> [ <i64 1, i64 1>, %[[VECTOR_PH]] ], [ [[TMP0:%.*]], %[[VECTOR_BODY]] ]
+; CHECK-NEXT:    [[VEC_PHI1:%.*]] = phi <2 x i64> [ <i64 1, i64 1>, %[[VECTOR_PH]] ], [ [[TMP1:%.*]], %[[VECTOR_BODY]] ]
+; CHECK-NEXT:    [[TMP0]] = mul <2 x i64> [[VEC_PHI]], <i64 2, i64 2>
+; CHECK-NEXT:    [[TMP1]] = mul <2 x i64> [[VEC_PHI1]], <i64 2, i64 2>
+; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 4
+; CHECK-NEXT:    [[TMP2:%.*]] = icmp eq i32 [[INDEX_NEXT]], 96
+; CHECK-NEXT:    br i1 [[TMP2]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
+; CHECK:       [[MIDDLE_BLOCK]]:
+; CHECK-NEXT:    [[BIN_RDX:%.*]] = mul <2 x i64> [[TMP1]], [[TMP0]]
+; CHECK-NEXT:    [[TMP3:%.*]] = call i64 @llvm.vector.reduce.mul.v2i64(<2 x i64> [[BIN_RDX]])
+; CHECK-NEXT:    br i1 false, label %[[EXIT:.*]], label %[[SCALAR_PH]]
+; CHECK:       [[SCALAR_PH]]:
+; CHECK-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i32 [ 97, %[[MIDDLE_BLOCK]] ], [ 1, %[[ENTRY]] ]
+; CHECK-NEXT:    [[BC_MERGE_RDX:%.*]] = phi i64 [ [[TMP3]], %[[MIDDLE_BLOCK]] ], [ 1, %[[ENTRY]] ]
+; CHECK-NEXT:    br label %[[LOOP:.*]]
+; CHECK:       [[LOOP]]:
+; CHECK-NEXT:    [[IV:%.*]] = phi i32 [ [[BC_RESUME_VAL]], %[[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], %[[LOOP]] ]
+; CHECK-NEXT:    [[RED:%.*]] = phi i64 [ [[BC_MERGE_RDX]], %[[SCALAR_PH]] ], [ [[MUL:%.*]], %[[LOOP]] ]
+; CHECK-NEXT:    [[MUL]] = mul nsw i64 [[RED]], [[ADD]]
+; CHECK-NEXT:    [[IV_NEXT]] = add nsw i32 [[IV]], 1
+; CHECK-NEXT:    [[EC:%.*]] = icmp eq i32 [[IV_NEXT]], 100
+; CHECK-NEXT:    br i1 [[EC]], label %[[EXIT]], label %[[LOOP]], !llvm.loop [[LOOP3:![0-9]+]]
+; CHECK:       [[EXIT]]:
+; CHECK-NEXT:    [[RET:%.*]] = phi i64 [ [[MUL]], %[[LOOP]] ], [ [[TMP3]], %[[MIDDLE_BLOCK]] ]
+; CHECK-NEXT:    ret i64 [[RET]]
+;
+entry:
+  %conv = zext i16 -6 to i64
+  %add = add nsw i64 %conv, -65528
+  br label %loop
+
+loop:
+  %iv = phi i32 [ 1, %entry ], [ %iv.next, %loop ]
+  %red = phi i64 [ 1, %entry ], [ %mul, %loop ]
+  %mul = mul nsw i64 %red, %add
+  %iv.next = add nsw i32 %iv, 1
+  %ec = icmp eq i32 %iv.next, 100
+  br i1 %ec, label %exit, label %loop
+
+exit:
+  %ret = phi i64 [ %mul, %loop ]
+  ret i64 %ret
+}
+;.
+; CHECK: [[LOOP0]] = distinct !{[[LOOP0]], [[META1:![0-9]+]], [[META2:![0-9]+]]}
+; CHECK: [[META1]] = !{!"llvm.loop.isvectorized", i32 1}
+; CHECK: [[META2]] = !{!"llvm.loop.unroll.runtime.disable"}
+; CHECK: [[LOOP3]] = distinct !{[[LOOP3]], [[META2]], [[META1]]}
+;.