[llvm] aada8f2 - [slp] Tweak debug costing output to include VL
Philip Reames via llvm-commits
llvm-commits at lists.llvm.org
Wed Aug 30 09:16:40 PDT 2023
Author: Philip Reames
Date: 2023-08-30T09:13:19-07:00
New Revision: aada8f2e5496f053909e1e801e431fe051f4f38f
URL: https://github.com/llvm/llvm-project/commit/aada8f2e5496f053909e1e801e431fe051f4f38f
DIFF: https://github.com/llvm/llvm-project/commit/aada8f2e5496f053909e1e801e431fe051f4f38f.diff
LOG: [slp] Tweak debug costing output to include VL
This makes it much easier to understand which vector length is being considered when the same set of nodes are evaluated at multiple vector lengths.
Added:
Modified:
llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
llvm/test/Transforms/SLPVectorizer/SystemZ/SLP-cmp-cost-query.ll
llvm/test/Transforms/SLPVectorizer/X86/reduction_unrolled.ll
Removed:
################################################################################
diff --git a/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp b/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
index 196f1d384e8aa5..b4a392c66803df 100644
--- a/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
+++ b/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
@@ -228,6 +228,17 @@ static bool isVectorLikeInstWithConstOps(Value *V) {
return isConstant(I->getOperand(2));
}
+#if !defined(NDEBUG)
+/// Print a short descriptor of the instruction bundle suitable for debug output.
+static std::string shortBundleName(ArrayRef<Value *> VL) {
+ std::string Result;
+ raw_string_ostream OS(Result);
+ OS << "n=" << VL.size() << " [" << *VL.front() << ", ..]";
+ OS.flush();
+ return Result;
+}
+#endif
+
/// \returns true if all of the instructions in \p VL are in the same block or
/// false otherwise.
static bool allSameBlock(ArrayRef<Value *> VL) {
@@ -7435,8 +7446,8 @@ BoUpSLP::getEntryCost(const TreeEntry *E, ArrayRef<Value *> VectorizedVals,
// node. Cost is 0.
LLVM_DEBUG(
dbgs()
- << "SLP: perfect diamond match for gather bundle that starts with "
- << *VL.front() << ".\n");
+ << "SLP: perfect diamond match for gather bundle "
+ << shortBundleName(VL) << ".\n");
// Restore the mask for previous partially matched values.
for (auto [I, V] : enumerate(E->Scalars)) {
if (isa<PoisonValue>(V)) {
@@ -7462,8 +7473,8 @@ BoUpSLP::getEntryCost(const TreeEntry *E, ArrayRef<Value *> VectorizedVals,
GatheredScalars[I] = PoisonValue::get(ScalarTy);
}
LLVM_DEBUG(dbgs() << "SLP: shuffled " << Entries.size()
- << " entries for bundle that starts with "
- << *VL.front() << ".\n";);
+ << " entries for bundle "
+ << shortBundleName(VL) << ".\n");
if (Entries.size() == 1)
Estimator.add(Entries.front(), Mask);
else
@@ -8575,8 +8586,8 @@ InstructionCost BoUpSLP::getTreeCost(ArrayRef<Value *> VectorizedVals) {
E->isSame(TE.Scalars)) {
// Some gather nodes might be absolutely the same as some vectorizable
// nodes after reordering, need to handle it.
- LLVM_DEBUG(dbgs() << "SLP: Adding cost 0 for bundle that starts with "
- << *TE.Scalars[0] << ".\n"
+ LLVM_DEBUG(dbgs() << "SLP: Adding cost 0 for bundle "
+ << shortBundleName(TE.Scalars) << ".\n"
<< "SLP: Current total cost = " << Cost << "\n");
continue;
}
@@ -8584,9 +8595,8 @@ InstructionCost BoUpSLP::getTreeCost(ArrayRef<Value *> VectorizedVals) {
InstructionCost C = getEntryCost(&TE, VectorizedVals, CheckedExtracts);
Cost += C;
- LLVM_DEBUG(dbgs() << "SLP: Adding cost " << C
- << " for bundle that starts with " << *TE.Scalars[0]
- << ".\n"
+ LLVM_DEBUG(dbgs() << "SLP: Adding cost " << C << " for bundle "
+ << shortBundleName(TE.Scalars) << ".\n"
<< "SLP: Current total cost = " << Cost << "\n");
}
@@ -9959,8 +9969,8 @@ ResTy BoUpSLP::processBuildVector(const TreeEntry *E, Args &...Params) {
// node. Cost is 0.
LLVM_DEBUG(
dbgs()
- << "SLP: perfect diamond match for gather bundle that starts with "
- << *E->Scalars.front() << ".\n");
+ << "SLP: perfect diamond match for gather bundle "
+ << shortBundleName(E->Scalars) << ".\n");
// Restore the mask for previous partially matched values.
if (Entries.front()->ReorderIndices.empty() &&
((Entries.front()->ReuseShuffleIndices.empty() &&
@@ -14082,8 +14092,7 @@ class HorizontalReduction {
bool IsCmpSelMinMax, unsigned ReduxWidth,
FastMathFlags FMF) {
TTI::TargetCostKind CostKind = TTI::TCK_RecipThroughput;
- Value *FirstReducedVal = ReducedVals.front();
- Type *ScalarTy = FirstReducedVal->getType();
+ Type *ScalarTy = ReducedVals.front()->getType();
FixedVectorType *VectorTy = FixedVectorType::get(ScalarTy, ReduxWidth);
InstructionCost VectorCost = 0, ScalarCost;
// If all of the reduced values are constant, the vector cost is 0, since
@@ -14157,7 +14166,7 @@ class HorizontalReduction {
}
LLVM_DEBUG(dbgs() << "SLP: Adding cost " << VectorCost - ScalarCost
- << " for reduction that starts with " << *FirstReducedVal
+ << " for reduction of " << shortBundleName(ReducedVals)
<< " (It is a splitting reduction)\n");
return VectorCost - ScalarCost;
}
diff --git a/llvm/test/Transforms/SLPVectorizer/SystemZ/SLP-cmp-cost-query.ll b/llvm/test/Transforms/SLPVectorizer/SystemZ/SLP-cmp-cost-query.ll
index b03c8788c2b637..7c05355e98d70b 100644
--- a/llvm/test/Transforms/SLPVectorizer/SystemZ/SLP-cmp-cost-query.ll
+++ b/llvm/test/Transforms/SLPVectorizer/SystemZ/SLP-cmp-cost-query.ll
@@ -31,6 +31,6 @@ define void @fun(ptr nocapture, i32 zeroext) local_unnamed_addr #0 {
._crit_edge: ; preds = %.lr.ph
ret void
-; CHECK: SLP: Adding cost -1 for bundle that starts with %4 = icmp ult i32 %2, %1.
+; CHECK: SLP: Adding cost -1 for bundle n=2 [ %4 = icmp ult i32 %2, %1, ..]
}
diff --git a/llvm/test/Transforms/SLPVectorizer/X86/reduction_unrolled.ll b/llvm/test/Transforms/SLPVectorizer/X86/reduction_unrolled.ll
index d20161029303dc..a61fa74e97cf90 100644
--- a/llvm/test/Transforms/SLPVectorizer/X86/reduction_unrolled.ll
+++ b/llvm/test/Transforms/SLPVectorizer/X86/reduction_unrolled.ll
@@ -1,6 +1,6 @@
; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
-; RUN: opt -passes=slp-vectorizer -slp-vectorize-hor -S -mtriple=x86_64-unknown-linux-gnu -mcpu=bdver2 -debug < %s 2>&1 | FileCheck %s --check-prefixes=CHECK,AVX
-; RUN: opt -passes=slp-vectorizer -slp-vectorize-hor -S -mtriple=x86_64-unknown-linux-gnu -mcpu=core2 -debug < %s 2>&1 | FileCheck %s --check-prefixes=CHECK,SSE
+; RUN: opt -passes=slp-vectorizer -slp-vectorize-hor -S -mtriple=x86_64-unknown-linux-gnu -mcpu=bdver2 -debug-only=SLP < %s 2>&1 | FileCheck %s --check-prefixes=CHECK,AVX
+; RUN: opt -passes=slp-vectorizer -slp-vectorize-hor -S -mtriple=x86_64-unknown-linux-gnu -mcpu=core2 -debug-only=SLP < %s 2>&1 | FileCheck %s --check-prefixes=CHECK,SSE
; REQUIRES: asserts
; int test_add(unsigned int *p) {
@@ -11,9 +11,9 @@
; }
; Vector cost is 5, Scalar cost is 7
-; AVX: Adding cost -2 for reduction that starts with %0 = load i32, ptr %p, align 4 (It is a splitting reduction)
+; AVX: Adding cost -2 for reduction of n=8 [ %0 = load i32, ptr %p, align 4, ..] (It is a splitting reduction)
; Vector cost is 6, Scalar cost is 7
-; SSE: Adding cost -1 for reduction that starts with %0 = load i32, ptr %p, align 4 (It is a splitting reduction)
+; SSE: Adding cost -1 for reduction of n=8 [ %0 = load i32, ptr %p, align 4, ..] (It is a splitting reduction)
define i32 @test_add(ptr nocapture readonly %p) {
; CHECK-LABEL: @test_add(
; CHECK-NEXT: entry:
More information about the llvm-commits
mailing list