[RISCV][TTI] Adjust VLS shuffle costing to account for sub-mask reuse

[preames]

If we have a shuffle which can be split via VLA where two or more of the destinations have exactly the same elements, then we only need to account for them once in costing. The duplicate copies are are (at worst) whole register moves.

Note that this change only handles the single source case. Doing the multiple source case seemed a bit more complicated, and I didn't have a motivating test case.

[llvmbot]

@llvm/pr-subscribers-backend-risc-v

@llvm/pr-subscribers-llvm-analysis

Author: Philip Reames (preames)

Changes

If we have a shuffle which can be split via VLA where two or more of the destinations have exactly the same elements, then we only need to account for them once in costing. The duplicate copes are are (at worst) whole register moves.

Note that this change only handles the single source case. Doing the multiple source case seemed a bit more complicated, and I didn't have a motivating test case.

---

Full diff: https://github.com/llvm/llvm-project/pull/129793.diff

2 Files Affected:

• (modified) llvm/lib/Target/RISCV/RISCVTargetTransformInfo.cpp (+3)
• (modified) llvm/test/Analysis/CostModel/RISCV/shuffle-exact-vlen.ll (+4-4)

diff --git a/llvm/lib/Target/RISCV/RISCVTargetTransformInfo.cpp b/llvm/lib/Target/RISCV/RISCVTargetTransformInfo.cpp
index 11a658758a9cb..6871e4b843b31 100644
--- a/llvm/lib/Target/RISCV/RISCVTargetTransformInfo.cpp
+++ b/llvm/lib/Target/RISCV/RISCVTargetTransformInfo.cpp
@@ -430,11 +430,14 @@ costShuffleViaVRegSplitting(RISCVTTIImpl &TTI, MVT LegalVT,
   copy(Mask, NormalizedMask.begin());
   InstructionCost Cost = 0;
   int NumShuffles = 0;
+  DenseSet<std::pair<ArrayRef<int>, unsigned>> ReusedSingleSrcShuffles;
   processShuffleMasks(
       NormalizedMask, NumOfSrcRegs, NumOfDestRegs, NumOfDestRegs, []() {},
       [&](ArrayRef<int> RegMask, unsigned SrcReg, unsigned DestReg) {
         if (ShuffleVectorInst::isIdentityMask(RegMask, RegMask.size()))
           return;
+        if (!ReusedSingleSrcShuffles.insert(std::make_pair(RegMask, SrcReg)).second)
+          return;
         ++NumShuffles;
         Cost += TTI.getShuffleCost(TTI::SK_PermuteSingleSrc, SingleOpTy,
                                    RegMask, CostKind, 0, nullptr);
diff --git a/llvm/test/Analysis/CostModel/RISCV/shuffle-exact-vlen.ll b/llvm/test/Analysis/CostModel/RISCV/shuffle-exact-vlen.ll
index 06c709e4cc879..23d5999237e30 100644
--- a/llvm/test/Analysis/CostModel/RISCV/shuffle-exact-vlen.ll
+++ b/llvm/test/Analysis/CostModel/RISCV/shuffle-exact-vlen.ll
@@ -42,7 +42,7 @@ define void @shuffle() vscale_range(2,2) {
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v11 = shufflevector <2 x i16> poison, <2 x i16> poison, <2 x i32> <i32 1, i32 0>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v12 = shufflevector <4 x i16> poison, <4 x i16> poison, <4 x i32> <i32 1, i32 3, i32 2, i32 0>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v13 = shufflevector <8 x i16> poison, <8 x i16> poison, <8 x i32> <i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1>
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v10b = shufflevector <16 x i16> poison, <16 x i16> poison, <16 x i32> <i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1>
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v10b = shufflevector <16 x i16> poison, <16 x i16> poison, <16 x i32> <i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v14 = shufflevector <2 x i32> poison, <2 x i32> poison, <2 x i32> <i32 1, i32 0>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v15 = shufflevector <4 x i32> poison, <4 x i32> poison, <4 x i32> <i32 1, i32 3, i32 2, i32 0>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v16 = shufflevector <2 x float> poison, <2 x float> poison, <2 x i32> <i32 1, i32 0>
@@ -58,7 +58,7 @@ define void @shuffle() vscale_range(2,2) {
 ; CHECK-SIZE-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v11 = shufflevector <2 x i16> poison, <2 x i16> poison, <2 x i32> <i32 1, i32 0>
 ; CHECK-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v12 = shufflevector <4 x i16> poison, <4 x i16> poison, <4 x i32> <i32 1, i32 3, i32 2, i32 0>
 ; CHECK-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v13 = shufflevector <8 x i16> poison, <8 x i16> poison, <8 x i32> <i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1>
-; CHECK-SIZE-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v10b = shufflevector <16 x i16> poison, <16 x i16> poison, <16 x i32> <i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1>
+; CHECK-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v10b = shufflevector <16 x i16> poison, <16 x i16> poison, <16 x i32> <i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1>
 ; CHECK-SIZE-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v14 = shufflevector <2 x i32> poison, <2 x i32> poison, <2 x i32> <i32 1, i32 0>
 ; CHECK-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v15 = shufflevector <4 x i32> poison, <4 x i32> poison, <4 x i32> <i32 1, i32 3, i32 2, i32 0>
 ; CHECK-SIZE-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v16 = shufflevector <2 x float> poison, <2 x float> poison, <2 x i32> <i32 1, i32 0>
@@ -738,7 +738,7 @@ define void @multipart() vscale_range(2,2) {
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %v16c = shufflevector <16 x i16> poison, <16 x i16> poison, <16 x i32> <i32 0, i32 8, i32 1, i32 9, i32 2, i32 10, i32 3, i32 11, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %v16d = shufflevector <16 x i16> poison, <16 x i16> poison, <16 x i32> <i32 0, i32 16, i32 1, i32 17, i32 2, i32 18, i32 3, i32 19, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v32a = shufflevector <4 x i32> poison, <4 x i32> poison, <4 x i32> <i32 2, i32 3, i32 0, i32 1>
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v32a4 = shufflevector <16 x i32> poison, <16 x i32> poison, <16 x i32> <i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1>
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v32a4 = shufflevector <16 x i32> poison, <16 x i32> poison, <16 x i32> <i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %v32idrev = shufflevector <16 x i32> poison, <16 x i32> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 15, i32 14, i32 13, i32 12, i32 16, i32 17, i32 18, i32 19, i32 31, i32 30, i32 29, i32 28>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 47 for instruction: %v32many = shufflevector <16 x i32> poison, <16 x i32> poison, <16 x i32> <i32 0, i32 4, i32 8, i32 12, i32 16, i32 20, i32 24, i32 28, i32 2, i32 6, i32 10, i32 14, i32 18, i32 22, i32 26, i32 30>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 47 for instruction: %v32many2 = shufflevector <16 x i32> poison, <16 x i32> poison, <16 x i32> <i32 1, i32 4, i32 8, i32 12, i32 17, i32 20, i32 24, i32 28, i32 2, i32 6, i32 11, i32 14, i32 18, i32 22, i32 27, i32 30>
@@ -758,7 +758,7 @@ define void @multipart() vscale_range(2,2) {
 ; CHECK-SIZE-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %v16c = shufflevector <16 x i16> poison, <16 x i16> poison, <16 x i32> <i32 0, i32 8, i32 1, i32 9, i32 2, i32 10, i32 3, i32 11, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1>
 ; CHECK-SIZE-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %v16d = shufflevector <16 x i16> poison, <16 x i16> poison, <16 x i32> <i32 0, i32 16, i32 1, i32 17, i32 2, i32 18, i32 3, i32 19, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1>
 ; CHECK-SIZE-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v32a = shufflevector <4 x i32> poison, <4 x i32> poison, <4 x i32> <i32 2, i32 3, i32 0, i32 1>
-; CHECK-SIZE-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v32a4 = shufflevector <16 x i32> poison, <16 x i32> poison, <16 x i32> <i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1>
+; CHECK-SIZE-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v32a4 = shufflevector <16 x i32> poison, <16 x i32> poison, <16 x i32> <i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1>
 ; CHECK-SIZE-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %v32idrev = shufflevector <16 x i32> poison, <16 x i32> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 15, i32 14, i32 13, i32 12, i32 16, i32 17, i32 18, i32 19, i32 31, i32 30, i32 29, i32 28>
 ; CHECK-SIZE-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %v32many = shufflevector <16 x i32> poison, <16 x i32> poison, <16 x i32> <i32 0, i32 4, i32 8, i32 12, i32 16, i32 20, i32 24, i32 28, i32 2, i32 6, i32 10, i32 14, i32 18, i32 22, i32 26, i32 30>
 ; CHECK-SIZE-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %v32many2 = shufflevector <16 x i32> poison, <16 x i32> poison, <16 x i32> <i32 1, i32 4, i32 8, i32 12, i32 17, i32 20, i32 24, i32 28, i32 2, i32 6, i32 11, i32 14, i32 18, i32 22, i32 27, i32 30>

[github-actions]

✅ With the latest revision this PR passed the C/C++ code formatter.

[preames]

For context, x86 TTI has an analogous, but slightly weaker, case which handles duplicates between adjacent destination results (skipping any that require multi build up steps).

This doesn't matter during actual lowering since CSE will cleanup the duplicate sub-trees, it only matters during costing. I considered making the change in processShuffleMasks itself, but wasn't quite sure of how that would interactive with the usage in LegalizeVectorTypes.cpp - the handling of the Input temporaries there confuses me.

[preames]

@alexey-bataev ping

[alexey-bataev]

SmallDenseSet

[preames]

@alexey-bataev or maybe @topperc?

[topperc]

LGTM