Skip to content

[ARM][MVE] Add shuffle costs for LDn and STn instructions. #145304

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom
Open
wants to merge 1 commit into
base: main
Choose a base branch
from
Open

[ARM][MVE] Add shuffle costs for LDn and STn instructions. #145304

wants to merge 1 commit into from
+301 −155

Conversation

davemgreen
Copy link
Collaborator

LD2 is represented in IR as deinterleave-shuffle(load), and ST2 as store(interleave-shuffle). Whilst the shuffle would be expensive in general for MVE (it does not have zip/uzp instructions), it should be treated as cheap when part of the LD2/ST2 pattern. This borrows some code from the AArch64 backed to produce lower costs. (Some of which still shows as higher than it should - that just shows how broken the generic shuffle costs are at the moment, they would be lower if getShuffleCost was called directly as opposed to going through getInstructionCost).

@davemgreen
[ARM][MVE] Add shuffle costs for LDn and STn instructions.
f766466 
LD2 is represented in IR as deinterleave-shuffle(load), and ST2 as
store(interleave-shuffle). Whilst the shuffle would be expensive in general for
MVE (it doesn not have zip/uzp instructions), it should be treated as cheap
when part of the LD2/ST2 pattern. This borrows some code from the AArch64
backed to produce lower costs. (Some of which still shows as higher than it
should - that just shows how broken the generic shuffle costs are at the
moment, they would be lower if getShuffleCost was called directly as opposed to
going through getInstructionCost).
@llvmbot llvmbot added backend:ARM llvm:analysis Includes value tracking, cost tables and constant folding labels Jun 23, 2025
@llvmbot
Copy link
Member

llvmbot commented Jun 23, 2025

@llvm/pr-subscribers-backend-arm

Author: David Green (davemgreen)

Changes

LD2 is represented in IR as deinterleave-shuffle(load), and ST2 as store(interleave-shuffle). Whilst the shuffle would be expensive in general for MVE (it does not have zip/uzp instructions), it should be treated as cheap when part of the LD2/ST2 pattern. This borrows some code from the AArch64 backed to produce lower costs. (Some of which still shows as higher than it should - that just shows how broken the generic shuffle costs are at the moment, they would be lower if getShuffleCost was called directly as opposed to going through getInstructionCost).

Patch is 91.21 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/145304.diff

2 Files Affected:

  • (modified) llvm/lib/Target/ARM/ARMTargetTransformInfo.cpp (+31)
  • (modified) llvm/test/Analysis/CostModel/ARM/mve-shuffle-loadstore.ll (+270-155)
diff --git a/llvm/lib/Target/ARM/ARMTargetTransformInfo.cpp b/llvm/lib/Target/ARM/ARMTargetTransformInfo.cpp
index 203fb76d7be86..5244d0e02f750 100644
--- a/llvm/lib/Target/ARM/ARMTargetTransformInfo.cpp
+++ b/llvm/lib/Target/ARM/ARMTargetTransformInfo.cpp
@@ -1336,6 +1336,37 @@ InstructionCost ARMTTIImpl::getShuffleCost(TTI::ShuffleKind Kind,
 
     if (!Mask.empty()) {
       std::pair<InstructionCost, MVT> LT = getTypeLegalizationCost(SrcTy);
+      // Check for LD2/LD4 instructions, which are represented in llvm IR as
+      // deinterleaving-shuffle(load). The shuffle cost could potentially be
+      // free, but we model it with a cost of LT.first so that LD2/LD4 have a
+      // higher cost than just the load.
+      if (Args.size() >= 1 && isa<LoadInst>(Args[0]) &&
+          (LT.second.getScalarSizeInBits() == 8 ||
+           LT.second.getScalarSizeInBits() == 16 ||
+           LT.second.getScalarSizeInBits() == 32) &&
+          LT.second.getSizeInBits() == 128 &&
+          (ShuffleVectorInst::isDeInterleaveMaskOfFactor(Mask, 2) ||
+           (TLI->getMaxSupportedInterleaveFactor() == 4 &&
+            ShuffleVectorInst::isDeInterleaveMaskOfFactor(Mask, 4))))
+        return ST->getMVEVectorCostFactor(TTI::TCK_RecipThroughput) *
+               std::max<InstructionCost>(1, LT.first / 4);
+
+      // Check for ST2/ST4 instructions, which are represented in llvm IR as
+      // store(interleaving-shuffle). The shuffle cost could potentially be
+      // free, but we model it with a cost of LT.first so that ST2/ST4 have a
+      // higher cost than just the store.
+      if (CxtI && CxtI->hasOneUse() && isa<StoreInst>(*CxtI->user_begin()) &&
+          (LT.second.getScalarSizeInBits() == 8 ||
+           LT.second.getScalarSizeInBits() == 16 ||
+           LT.second.getScalarSizeInBits() == 32) &&
+          LT.second.getSizeInBits() == 128 &&
+          (ShuffleVectorInst::isInterleaveMask(
+               Mask, 2, SrcTy->getElementCount().getKnownMinValue() * 2) ||
+           (TLI->getMaxSupportedInterleaveFactor() == 4 &&
+            ShuffleVectorInst::isInterleaveMask(
+                Mask, 4, SrcTy->getElementCount().getKnownMinValue() * 2))))
+        return ST->getMVEVectorCostFactor(TTI::TCK_RecipThroughput) * LT.first;
+
       if (LT.second.isVector() &&
           Mask.size() <= LT.second.getVectorNumElements() &&
           (isVREVMask(Mask, LT.second, 16) || isVREVMask(Mask, LT.second, 32) ||
diff --git a/llvm/test/Analysis/CostModel/ARM/mve-shuffle-loadstore.ll b/llvm/test/Analysis/CostModel/ARM/mve-shuffle-loadstore.ll
index 6a327cfed4e4c..ef0b28ea2604a 100644
--- a/llvm/test/Analysis/CostModel/ARM/mve-shuffle-loadstore.ll
+++ b/llvm/test/Analysis/CostModel/ARM/mve-shuffle-loadstore.ll
@@ -7,41 +7,41 @@
 define void @vld2(ptr %p) {
 ; CHECK-LABEL: 'vld2'
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v4i8 = load <4 x i8>, ptr %p, align 4
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 96 for instruction: %v4i8_0 = shufflevector <4 x i8> %v4i8, <4 x i8> undef, <2 x i32> <i32 0, i32 2>
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 96 for instruction: %v4i8_1 = shufflevector <4 x i8> %v4i8, <4 x i8> undef, <2 x i32> <i32 1, i32 3>
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 34 for instruction: %v4i8_0 = shufflevector <4 x i8> %v4i8, <4 x i8> undef, <2 x i32> <i32 0, i32 2>
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 34 for instruction: %v4i8_1 = shufflevector <4 x i8> %v4i8, <4 x i8> undef, <2 x i32> <i32 1, i32 3>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v8i8 = load <8 x i8>, ptr %p, align 8
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 192 for instruction: %v8i8_0 = shufflevector <8 x i8> %v8i8, <8 x i8> undef, <4 x i32> <i32 0, i32 2, i32 4, i32 6>
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 192 for instruction: %v8i8_1 = shufflevector <8 x i8> %v8i8, <8 x i8> undef, <4 x i32> <i32 1, i32 3, i32 5, i32 7>
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 66 for instruction: %v8i8_0 = shufflevector <8 x i8> %v8i8, <8 x i8> undef, <4 x i32> <i32 0, i32 2, i32 4, i32 6>
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 66 for instruction: %v8i8_1 = shufflevector <8 x i8> %v8i8, <8 x i8> undef, <4 x i32> <i32 1, i32 3, i32 5, i32 7>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v16i8 = load <16 x i8>, ptr %p, align 8
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 384 for instruction: %v16i8_0 = shufflevector <16 x i8> %v16i8, <16 x i8> undef, <8 x i32> <i32 0, i32 2, i32 4, i32 6, i32 8, i32 10, i32 12, i32 14>
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 384 for instruction: %v16i8_1 = shufflevector <16 x i8> %v16i8, <16 x i8> undef, <8 x i32> <i32 1, i32 3, i32 5, i32 7, i32 9, i32 11, i32 13, i32 15>
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 130 for instruction: %v16i8_0 = shufflevector <16 x i8> %v16i8, <16 x i8> undef, <8 x i32> <i32 0, i32 2, i32 4, i32 6, i32 8, i32 10, i32 12, i32 14>
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 130 for instruction: %v16i8_1 = shufflevector <16 x i8> %v16i8, <16 x i8> undef, <8 x i32> <i32 1, i32 3, i32 5, i32 7, i32 9, i32 11, i32 13, i32 15>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v32i8 = load <32 x i8>, ptr %p, align 32
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 768 for instruction: %v32i8_0 = shufflevector <32 x i8> %v32i8, <32 x i8> undef, <16 x i32> <i32 0, i32 2, i32 4, i32 6, i32 8, i32 10, i32 12, i32 14, i32 16, i32 18, i32 20, i32 22, i32 24, i32 26, i32 28, i32 30>
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 768 for instruction: %v32i8_1 = shufflevector <32 x i8> %v32i8, <32 x i8> undef, <16 x i32> <i32 1, i32 3, i32 5, i32 7, i32 9, i32 11, i32 13, i32 15, i32 17, i32 19, i32 21, i32 23, i32 25, i32 27, i32 29, i32 31>
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 258 for instruction: %v32i8_0 = shufflevector <32 x i8> %v32i8, <32 x i8> undef, <16 x i32> <i32 0, i32 2, i32 4, i32 6, i32 8, i32 10, i32 12, i32 14, i32 16, i32 18, i32 20, i32 22, i32 24, i32 26, i32 28, i32 30>
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 258 for instruction: %v32i8_1 = shufflevector <32 x i8> %v32i8, <32 x i8> undef, <16 x i32> <i32 1, i32 3, i32 5, i32 7, i32 9, i32 11, i32 13, i32 15, i32 17, i32 19, i32 21, i32 23, i32 25, i32 27, i32 29, i32 31>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v4i16 = load <4 x i16>, ptr %p, align 8
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 96 for instruction: %v4i16_0 = shufflevector <4 x i16> %v4i16, <4 x i16> undef, <2 x i32> <i32 0, i32 2>
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 96 for instruction: %v4i16_1 = shufflevector <4 x i16> %v4i16, <4 x i16> undef, <2 x i32> <i32 1, i32 3>
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 34 for instruction: %v4i16_0 = shufflevector <4 x i16> %v4i16, <4 x i16> undef, <2 x i32> <i32 0, i32 2>
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 34 for instruction: %v4i16_1 = shufflevector <4 x i16> %v4i16, <4 x i16> undef, <2 x i32> <i32 1, i32 3>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v8i16 = load <8 x i16>, ptr %p, align 8
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 192 for instruction: %v8i16_0 = shufflevector <8 x i16> %v8i16, <8 x i16> undef, <4 x i32> <i32 0, i32 2, i32 4, i32 6>
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 192 for instruction: %v8i16_1 = shufflevector <8 x i16> %v8i16, <8 x i16> undef, <4 x i32> <i32 1, i32 3, i32 5, i32 7>
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 66 for instruction: %v8i16_0 = shufflevector <8 x i16> %v8i16, <8 x i16> undef, <4 x i32> <i32 0, i32 2, i32 4, i32 6>
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 66 for instruction: %v8i16_1 = shufflevector <8 x i16> %v8i16, <8 x i16> undef, <4 x i32> <i32 1, i32 3, i32 5, i32 7>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v16i16 = load <16 x i16>, ptr %p, align 32
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 384 for instruction: %v16i16_0 = shufflevector <16 x i16> %v16i16, <16 x i16> undef, <8 x i32> <i32 0, i32 2, i32 4, i32 6, i32 8, i32 10, i32 12, i32 14>
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 384 for instruction: %v16i16_1 = shufflevector <16 x i16> %v16i16, <16 x i16> undef, <8 x i32> <i32 1, i32 3, i32 5, i32 7, i32 9, i32 11, i32 13, i32 15>
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 130 for instruction: %v16i16_0 = shufflevector <16 x i16> %v16i16, <16 x i16> undef, <8 x i32> <i32 0, i32 2, i32 4, i32 6, i32 8, i32 10, i32 12, i32 14>
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 130 for instruction: %v16i16_1 = shufflevector <16 x i16> %v16i16, <16 x i16> undef, <8 x i32> <i32 1, i32 3, i32 5, i32 7, i32 9, i32 11, i32 13, i32 15>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v32i16 = load <32 x i16>, ptr %p, align 64
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 768 for instruction: %v32i16_0 = shufflevector <32 x i16> %v32i16, <32 x i16> undef, <16 x i32> <i32 0, i32 2, i32 4, i32 6, i32 8, i32 10, i32 12, i32 14, i32 16, i32 18, i32 20, i32 22, i32 24, i32 26, i32 28, i32 30>
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 768 for instruction: %v32i16_1 = shufflevector <32 x i16> %v32i16, <32 x i16> undef, <16 x i32> <i32 1, i32 3, i32 5, i32 7, i32 9, i32 11, i32 13, i32 15, i32 17, i32 19, i32 21, i32 23, i32 25, i32 27, i32 29, i32 31>
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 258 for instruction: %v32i16_0 = shufflevector <32 x i16> %v32i16, <32 x i16> undef, <16 x i32> <i32 0, i32 2, i32 4, i32 6, i32 8, i32 10, i32 12, i32 14, i32 16, i32 18, i32 20, i32 22, i32 24, i32 26, i32 28, i32 30>
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 258 for instruction: %v32i16_1 = shufflevector <32 x i16> %v32i16, <32 x i16> undef, <16 x i32> <i32 1, i32 3, i32 5, i32 7, i32 9, i32 11, i32 13, i32 15, i32 17, i32 19, i32 21, i32 23, i32 25, i32 27, i32 29, i32 31>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v4i32 = load <4 x i32>, ptr %p, align 8
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 96 for instruction: %v4i32_0 = shufflevector <4 x i32> %v4i32, <4 x i32> undef, <2 x i32> <i32 0, i32 2>
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 96 for instruction: %v4i32_1 = shufflevector <4 x i32> %v4i32, <4 x i32> undef, <2 x i32> <i32 1, i32 3>
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 34 for instruction: %v4i32_0 = shufflevector <4 x i32> %v4i32, <4 x i32> undef, <2 x i32> <i32 0, i32 2>
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 34 for instruction: %v4i32_1 = shufflevector <4 x i32> %v4i32, <4 x i32> undef, <2 x i32> <i32 1, i32 3>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v8i32 = load <8 x i32>, ptr %p, align 32
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 192 for instruction: %v8i32_0 = shufflevector <8 x i32> %v8i32, <8 x i32> undef, <4 x i32> <i32 0, i32 2, i32 4, i32 6>
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 192 for instruction: %v8i32_1 = shufflevector <8 x i32> %v8i32, <8 x i32> undef, <4 x i32> <i32 1, i32 3, i32 5, i32 7>
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 66 for instruction: %v8i32_0 = shufflevector <8 x i32> %v8i32, <8 x i32> undef, <4 x i32> <i32 0, i32 2, i32 4, i32 6>
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 66 for instruction: %v8i32_1 = shufflevector <8 x i32> %v8i32, <8 x i32> undef, <4 x i32> <i32 1, i32 3, i32 5, i32 7>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v16i32 = load <16 x i32>, ptr %p, align 64
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 384 for instruction: %v16i32_0 = shufflevector <16 x i32> %v16i32, <16 x i32> undef, <8 x i32> <i32 0, i32 2, i32 4, i32 6, i32 8, i32 10, i32 12, i32 14>
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 384 for instruction: %v16i32_1 = shufflevector <16 x i32> %v16i32, <16 x i32> undef, <8 x i32> <i32 1, i32 3, i32 5, i32 7, i32 9, i32 11, i32 13, i32 15>
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 130 for instruction: %v16i32_0 = shufflevector <16 x i32> %v16i32, <16 x i32> undef, <8 x i32> <i32 0, i32 2, i32 4, i32 6, i32 8, i32 10, i32 12, i32 14>
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 130 for instruction: %v16i32_1 = shufflevector <16 x i32> %v16i32, <16 x i32> undef, <8 x i32> <i32 1, i32 3, i32 5, i32 7, i32 9, i32 11, i32 13, i32 15>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v32i32 = load <32 x i32>, ptr %p, align 128
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 768 for instruction: %v32i32_0 = shufflevector <32 x i32> %v32i32, <32 x i32> undef, <16 x i32> <i32 0, i32 2, i32 4, i32 6, i32 8, i32 10, i32 12, i32 14, i32 16, i32 18, i32 20, i32 22, i32 24, i32 26, i32 28, i32 30>
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 768 for instruction: %v32i32_1 = shufflevector <32 x i32> %v32i32, <32 x i32> undef, <16 x i32> <i32 1, i32 3, i32 5, i32 7, i32 9, i32 11, i32 13, i32 15, i32 17, i32 19, i32 21, i32 23, i32 25, i32 27, i32 29, i32 31>
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 260 for instruction: %v32i32_0 = shufflevector <32 x i32> %v32i32, <32 x i32> undef, <16 x i32> <i32 0, i32 2, i32 4, i32 6, i32 8, i32 10, i32 12, i32 14, i32 16, i32 18, i32 20, i32 22, i32 24, i32 26, i32 28, i32 30>
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 260 for instruction: %v32i32_1 = shufflevector <32 x i32> %v32i32, <32 x i32> undef, <16 x i32> <i32 1, i32 3, i32 5, i32 7, i32 9, i32 11, i32 13, i32 15, i32 17, i32 19, i32 21, i32 23, i32 25, i32 27, i32 29, i32 31>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2i64 = load <4 x i64>, ptr %p, align 32
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 192 for instruction: %v2i64_0 = shufflevector <4 x i64> %v2i64, <4 x i64> undef, <2 x i32> <i32 0, i32 2>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 192 for instruction: %v2i64_1 = shufflevector <4 x i64> %v2i64, <4 x i64> undef, <2 x i32> <i32 1, i32 3>
@@ -252,88 +252,171 @@ define void @vld3(ptr %p) {
 }
 
 define void @vld4(ptr %p) {
-; CHECK-LABEL: 'vld4'
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v2i8 = load <8 x i8>, ptr %p, align 8
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 160 for instruction: %v2i8_0 = shufflevector <8 x i8> %v2i8, <8 x i8> undef, <2 x i32> <i32 0, i32 4>
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 160 for instruction: %v2i8_1 = shufflevector <8 x i8> %v2i8, <8 x i8> undef, <2 x i32> <i32 1, i32 5>
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 160 for instruction: %v2i8_2 = shufflevector <8 x i8> %v2i8, <8 x i8> undef, <2 x i32> <i32 2, i32 6>
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 160 for instruction: %v2i8_3 = shufflevector <8 x i8> %v2i8, <8 x i8> undef, <2 x i32> <i32 3, i32 7>
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v4i8 = load <16 x i8>, ptr %p, align 8
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 320 for instruction: %v4i8_0 = shufflevector <16 x i8> %v4i8, <16 x i8> undef, <4 x i32> <i32 0, i32 4, i32 8, i32 12>
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 320 for instruction: %v4i8_1 = shufflevector <16 x i8> %v4i8, <16 x i8> undef, <4 x i32> <i32 1, i32 5, i32 9, i32 13>
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 320 for instruction: %v4i8_2 = shufflevector <16 x i8> %v4i8, <16 x i8> undef, <4 x i32> <i32 2, i32 6, i32 10, i32 14>
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 320 for instruction: %v4i8_3 = shufflevector <16 x i8> %v4i8, <16 x i8> undef, <4 x i32> <i32 3, i32 7, i32 11, i32 15>
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v8i8 = load <32 x i8>, ptr %p, align 32
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 640 for instruction: %v8i8_0 = shufflevector <32 x i8> %v8i8, <32 x i8> undef, <8 x i32> <i32 0, i32 4, i32 8, i32 12, i32 16, i32 20, i32 24, i32 28>
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 640 for instruction: %v8i8_1 = shufflevector <32 x i8> %v8i8, <32 x i8> undef, <8 x i32> <i32 1, i32 5, i32 9, i32 13, i32 17, i32 21, i32 25, i32 29>
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 640 for instruction: %v8i8_2 = shufflevector <32 x i8> %v8i8, <32 x i8> undef, <8 x i32> <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 640 for instruction: %v8i8_3 = shufflevector <32 x i8> %v8i8, <32 x i8> undef, <8 x i32> <i32 3, i32 7, i32 11, i32 15, i32 19, i32 23, i32 27, i32 31>
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v16i8 = load <64 x i8>, ptr %p, align 64
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 1280 for instruction: %v16i8_0 = shufflevector <64 x i8> %v16i8, <64 x i8> undef, <16 x i32> <i32 0, i32 4, i32 8, i32 12, i32 16, i32 20, i32 24, i32 28, i32 32, i32 36, i32 40, i32 44, i32 48, i32 52, i32 56, i32 60>
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 1280 for instruction: %v16i8_1 = shufflevector <64 x i8> %v16i8, <64 x i8> undef, <16 x i32> <i32 1, i32 5, i32 9, i32 13, i32 17, i32 21, i32 25, i32 29, i32 33, i32 37, i32 41, i32 45, i32 49, i32 53, i32 57, i32 61>
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 1280 for instruction: %v16i8_2 = shufflevector <64 x i8> %v16i8, <64 x i8> undef, <16 x i32> <i32 2, i32 6, i32 10, i32 14, i32 18, i32 22, i32 26, i32 30, i32 34, i32 38, i32 42, i32 46, i32 50, i32 54, i32 58, i32 62>
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 1280 for instruction: %v16i8_3 = shufflevector <64 x i8> %v16i8, <64 x i8> undef, <16 x i32> <i32 3, i32 7, i32 11, i32 15, i32 19, i32 23, i32 27, i32 31, i32 35, i32 39, i32 43, i32 47, i32 51, i32 55, i32 59, i32 63>
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v2i16 = load <8 x i16>, ptr %p, align 8
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 160 for instruction: %v2i16_0 = shufflevector <8 x i16> %v2i16, <8 x i16> undef, <2 x i32> <i32 0, i32 4>
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 160 for instruction: %v2i16_1 = shufflevector <8 x i16> %v2i16, <8 x i16> undef, <2 x i32> <i32 1, i32 5>
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 160 for instruction: %v2i16_2 = shufflevector <8 x i16> %v2i16, <8 x i16> undef, <2 x i32> <i32 2, i32 6>
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 160 for instruction: %v2i16_3 = shufflevector <8 x i16> %v2i16, <8 x i16> undef, <2 x i32> <i32 3, i32 7>
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v4i16 = load <16 x i16>, ptr %p, align 32
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 320 for instruction: %v4i16_0 = shufflevector <16 x i16> %v4i16, <16 x i16> undef, <4 x i32> <i32 0, i32 4, i32 8, i32 12>
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 320 for instruction: %v4i16_1 = shufflevector <16 x i16> %v4i16, <16 x i16> undef, <4 x i32> <i32 1, i32 5, i32 9, i32 13>
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 320 for instruction: %v4i16_2 = shufflevector <16 x i16> %v4i16, <16 x i16> undef, <4 x i32> <i32 2, i32 6, i32 10, i32 14>
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 320 for instruction: %v4i16_3 = shufflevector <16 x i16> %v4i16, <16 x i16> undef, <4 x i32> <i32 3, i32 7, i32 11, i32 15>
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v8i16 = load <32 x i16>, ptr %p, align 64
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 640 for instruction: %v8i16_0 = shufflevector <32 x i16> %v8i16, <32 x i16> undef, <8 x i32> <i32 0, i32 4, i32 8, i32 12, i32 16, i32 20, i32 24, i32 28>
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 640 for instruction: %v8i16_1 = shufflevector <32 x i16> %v8i16, <32 x i16> undef, <8 x i32> <i32 1, i32 5, i32 9, i32 13, i32 17, i32 21, i32 25, i32 29>
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 640 for instruction: %v8i16_2 = shufflevector <32...
[truncated]

@llvmbot
Copy link
Member

llvmbot commented Jun 23, 2025

@llvm/pr-subscribers-llvm-analysis

Author: David Green (davemgreen)

Changes

LD2 is represented in IR as deinterleave-shuffle(load), and ST2 as store(interleave-shuffle). Whilst the shuffle would be expensive in general for MVE (it does not have zip/uzp instructions), it should be treated as cheap when part of the LD2/ST2 pattern. This borrows some code from the AArch64 backed to produce lower costs. (Some of which still shows as higher than it should - that just shows how broken the generic shuffle costs are at the moment, they would be lower if getShuffleCost was called directly as opposed to going through getInstructionCost).

Patch is 91.21 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/145304.diff

2 Files Affected:

  • (modified) llvm/lib/Target/ARM/ARMTargetTransformInfo.cpp (+31)
  • (modified) llvm/test/Analysis/CostModel/ARM/mve-shuffle-loadstore.ll (+270-155)
diff --git a/llvm/lib/Target/ARM/ARMTargetTransformInfo.cpp b/llvm/lib/Target/ARM/ARMTargetTransformInfo.cpp
index 203fb76d7be86..5244d0e02f750 100644
--- a/llvm/lib/Target/ARM/ARMTargetTransformInfo.cpp
+++ b/llvm/lib/Target/ARM/ARMTargetTransformInfo.cpp
@@ -1336,6 +1336,37 @@ InstructionCost ARMTTIImpl::getShuffleCost(TTI::ShuffleKind Kind,
 
     if (!Mask.empty()) {
       std::pair<InstructionCost, MVT> LT = getTypeLegalizationCost(SrcTy);
+      // Check for LD2/LD4 instructions, which are represented in llvm IR as
+      // deinterleaving-shuffle(load). The shuffle cost could potentially be
+      // free, but we model it with a cost of LT.first so that LD2/LD4 have a
+      // higher cost than just the load.
+      if (Args.size() >= 1 && isa<LoadInst>(Args[0]) &&
+          (LT.second.getScalarSizeInBits() == 8 ||
+           LT.second.getScalarSizeInBits() == 16 ||
+           LT.second.getScalarSizeInBits() == 32) &&
+          LT.second.getSizeInBits() == 128 &&
+          (ShuffleVectorInst::isDeInterleaveMaskOfFactor(Mask, 2) ||
+           (TLI->getMaxSupportedInterleaveFactor() == 4 &&
+            ShuffleVectorInst::isDeInterleaveMaskOfFactor(Mask, 4))))
+        return ST->getMVEVectorCostFactor(TTI::TCK_RecipThroughput) *
+               std::max<InstructionCost>(1, LT.first / 4);
+
+      // Check for ST2/ST4 instructions, which are represented in llvm IR as
+      // store(interleaving-shuffle). The shuffle cost could potentially be
+      // free, but we model it with a cost of LT.first so that ST2/ST4 have a
+      // higher cost than just the store.
+      if (CxtI && CxtI->hasOneUse() && isa<StoreInst>(*CxtI->user_begin()) &&
+          (LT.second.getScalarSizeInBits() == 8 ||
+           LT.second.getScalarSizeInBits() == 16 ||
+           LT.second.getScalarSizeInBits() == 32) &&
+          LT.second.getSizeInBits() == 128 &&
+          (ShuffleVectorInst::isInterleaveMask(
+               Mask, 2, SrcTy->getElementCount().getKnownMinValue() * 2) ||
+           (TLI->getMaxSupportedInterleaveFactor() == 4 &&
+            ShuffleVectorInst::isInterleaveMask(
+                Mask, 4, SrcTy->getElementCount().getKnownMinValue() * 2))))
+        return ST->getMVEVectorCostFactor(TTI::TCK_RecipThroughput) * LT.first;
+
       if (LT.second.isVector() &&
           Mask.size() <= LT.second.getVectorNumElements() &&
           (isVREVMask(Mask, LT.second, 16) || isVREVMask(Mask, LT.second, 32) ||
diff --git a/llvm/test/Analysis/CostModel/ARM/mve-shuffle-loadstore.ll b/llvm/test/Analysis/CostModel/ARM/mve-shuffle-loadstore.ll
index 6a327cfed4e4c..ef0b28ea2604a 100644
--- a/llvm/test/Analysis/CostModel/ARM/mve-shuffle-loadstore.ll
+++ b/llvm/test/Analysis/CostModel/ARM/mve-shuffle-loadstore.ll
@@ -7,41 +7,41 @@
 define void @vld2(ptr %p) {
 ; CHECK-LABEL: 'vld2'
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v4i8 = load <4 x i8>, ptr %p, align 4
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 96 for instruction: %v4i8_0 = shufflevector <4 x i8> %v4i8, <4 x i8> undef, <2 x i32> <i32 0, i32 2>
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 96 for instruction: %v4i8_1 = shufflevector <4 x i8> %v4i8, <4 x i8> undef, <2 x i32> <i32 1, i32 3>
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 34 for instruction: %v4i8_0 = shufflevector <4 x i8> %v4i8, <4 x i8> undef, <2 x i32> <i32 0, i32 2>
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 34 for instruction: %v4i8_1 = shufflevector <4 x i8> %v4i8, <4 x i8> undef, <2 x i32> <i32 1, i32 3>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v8i8 = load <8 x i8>, ptr %p, align 8
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 192 for instruction: %v8i8_0 = shufflevector <8 x i8> %v8i8, <8 x i8> undef, <4 x i32> <i32 0, i32 2, i32 4, i32 6>
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 192 for instruction: %v8i8_1 = shufflevector <8 x i8> %v8i8, <8 x i8> undef, <4 x i32> <i32 1, i32 3, i32 5, i32 7>
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 66 for instruction: %v8i8_0 = shufflevector <8 x i8> %v8i8, <8 x i8> undef, <4 x i32> <i32 0, i32 2, i32 4, i32 6>
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 66 for instruction: %v8i8_1 = shufflevector <8 x i8> %v8i8, <8 x i8> undef, <4 x i32> <i32 1, i32 3, i32 5, i32 7>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v16i8 = load <16 x i8>, ptr %p, align 8
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 384 for instruction: %v16i8_0 = shufflevector <16 x i8> %v16i8, <16 x i8> undef, <8 x i32> <i32 0, i32 2, i32 4, i32 6, i32 8, i32 10, i32 12, i32 14>
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 384 for instruction: %v16i8_1 = shufflevector <16 x i8> %v16i8, <16 x i8> undef, <8 x i32> <i32 1, i32 3, i32 5, i32 7, i32 9, i32 11, i32 13, i32 15>
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 130 for instruction: %v16i8_0 = shufflevector <16 x i8> %v16i8, <16 x i8> undef, <8 x i32> <i32 0, i32 2, i32 4, i32 6, i32 8, i32 10, i32 12, i32 14>
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 130 for instruction: %v16i8_1 = shufflevector <16 x i8> %v16i8, <16 x i8> undef, <8 x i32> <i32 1, i32 3, i32 5, i32 7, i32 9, i32 11, i32 13, i32 15>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v32i8 = load <32 x i8>, ptr %p, align 32
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 768 for instruction: %v32i8_0 = shufflevector <32 x i8> %v32i8, <32 x i8> undef, <16 x i32> <i32 0, i32 2, i32 4, i32 6, i32 8, i32 10, i32 12, i32 14, i32 16, i32 18, i32 20, i32 22, i32 24, i32 26, i32 28, i32 30>
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 768 for instruction: %v32i8_1 = shufflevector <32 x i8> %v32i8, <32 x i8> undef, <16 x i32> <i32 1, i32 3, i32 5, i32 7, i32 9, i32 11, i32 13, i32 15, i32 17, i32 19, i32 21, i32 23, i32 25, i32 27, i32 29, i32 31>
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 258 for instruction: %v32i8_0 = shufflevector <32 x i8> %v32i8, <32 x i8> undef, <16 x i32> <i32 0, i32 2, i32 4, i32 6, i32 8, i32 10, i32 12, i32 14, i32 16, i32 18, i32 20, i32 22, i32 24, i32 26, i32 28, i32 30>
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 258 for instruction: %v32i8_1 = shufflevector <32 x i8> %v32i8, <32 x i8> undef, <16 x i32> <i32 1, i32 3, i32 5, i32 7, i32 9, i32 11, i32 13, i32 15, i32 17, i32 19, i32 21, i32 23, i32 25, i32 27, i32 29, i32 31>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v4i16 = load <4 x i16>, ptr %p, align 8
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 96 for instruction: %v4i16_0 = shufflevector <4 x i16> %v4i16, <4 x i16> undef, <2 x i32> <i32 0, i32 2>
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 96 for instruction: %v4i16_1 = shufflevector <4 x i16> %v4i16, <4 x i16> undef, <2 x i32> <i32 1, i32 3>
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 34 for instruction: %v4i16_0 = shufflevector <4 x i16> %v4i16, <4 x i16> undef, <2 x i32> <i32 0, i32 2>
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 34 for instruction: %v4i16_1 = shufflevector <4 x i16> %v4i16, <4 x i16> undef, <2 x i32> <i32 1, i32 3>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v8i16 = load <8 x i16>, ptr %p, align 8
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 192 for instruction: %v8i16_0 = shufflevector <8 x i16> %v8i16, <8 x i16> undef, <4 x i32> <i32 0, i32 2, i32 4, i32 6>
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 192 for instruction: %v8i16_1 = shufflevector <8 x i16> %v8i16, <8 x i16> undef, <4 x i32> <i32 1, i32 3, i32 5, i32 7>
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 66 for instruction: %v8i16_0 = shufflevector <8 x i16> %v8i16, <8 x i16> undef, <4 x i32> <i32 0, i32 2, i32 4, i32 6>
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 66 for instruction: %v8i16_1 = shufflevector <8 x i16> %v8i16, <8 x i16> undef, <4 x i32> <i32 1, i32 3, i32 5, i32 7>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v16i16 = load <16 x i16>, ptr %p, align 32
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 384 for instruction: %v16i16_0 = shufflevector <16 x i16> %v16i16, <16 x i16> undef, <8 x i32> <i32 0, i32 2, i32 4, i32 6, i32 8, i32 10, i32 12, i32 14>
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 384 for instruction: %v16i16_1 = shufflevector <16 x i16> %v16i16, <16 x i16> undef, <8 x i32> <i32 1, i32 3, i32 5, i32 7, i32 9, i32 11, i32 13, i32 15>
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 130 for instruction: %v16i16_0 = shufflevector <16 x i16> %v16i16, <16 x i16> undef, <8 x i32> <i32 0, i32 2, i32 4, i32 6, i32 8, i32 10, i32 12, i32 14>
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 130 for instruction: %v16i16_1 = shufflevector <16 x i16> %v16i16, <16 x i16> undef, <8 x i32> <i32 1, i32 3, i32 5, i32 7, i32 9, i32 11, i32 13, i32 15>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v32i16 = load <32 x i16>, ptr %p, align 64
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 768 for instruction: %v32i16_0 = shufflevector <32 x i16> %v32i16, <32 x i16> undef, <16 x i32> <i32 0, i32 2, i32 4, i32 6, i32 8, i32 10, i32 12, i32 14, i32 16, i32 18, i32 20, i32 22, i32 24, i32 26, i32 28, i32 30>
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 768 for instruction: %v32i16_1 = shufflevector <32 x i16> %v32i16, <32 x i16> undef, <16 x i32> <i32 1, i32 3, i32 5, i32 7, i32 9, i32 11, i32 13, i32 15, i32 17, i32 19, i32 21, i32 23, i32 25, i32 27, i32 29, i32 31>
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 258 for instruction: %v32i16_0 = shufflevector <32 x i16> %v32i16, <32 x i16> undef, <16 x i32> <i32 0, i32 2, i32 4, i32 6, i32 8, i32 10, i32 12, i32 14, i32 16, i32 18, i32 20, i32 22, i32 24, i32 26, i32 28, i32 30>
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 258 for instruction: %v32i16_1 = shufflevector <32 x i16> %v32i16, <32 x i16> undef, <16 x i32> <i32 1, i32 3, i32 5, i32 7, i32 9, i32 11, i32 13, i32 15, i32 17, i32 19, i32 21, i32 23, i32 25, i32 27, i32 29, i32 31>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v4i32 = load <4 x i32>, ptr %p, align 8
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 96 for instruction: %v4i32_0 = shufflevector <4 x i32> %v4i32, <4 x i32> undef, <2 x i32> <i32 0, i32 2>
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 96 for instruction: %v4i32_1 = shufflevector <4 x i32> %v4i32, <4 x i32> undef, <2 x i32> <i32 1, i32 3>
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 34 for instruction: %v4i32_0 = shufflevector <4 x i32> %v4i32, <4 x i32> undef, <2 x i32> <i32 0, i32 2>
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 34 for instruction: %v4i32_1 = shufflevector <4 x i32> %v4i32, <4 x i32> undef, <2 x i32> <i32 1, i32 3>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v8i32 = load <8 x i32>, ptr %p, align 32
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 192 for instruction: %v8i32_0 = shufflevector <8 x i32> %v8i32, <8 x i32> undef, <4 x i32> <i32 0, i32 2, i32 4, i32 6>
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 192 for instruction: %v8i32_1 = shufflevector <8 x i32> %v8i32, <8 x i32> undef, <4 x i32> <i32 1, i32 3, i32 5, i32 7>
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 66 for instruction: %v8i32_0 = shufflevector <8 x i32> %v8i32, <8 x i32> undef, <4 x i32> <i32 0, i32 2, i32 4, i32 6>
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 66 for instruction: %v8i32_1 = shufflevector <8 x i32> %v8i32, <8 x i32> undef, <4 x i32> <i32 1, i32 3, i32 5, i32 7>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v16i32 = load <16 x i32>, ptr %p, align 64
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 384 for instruction: %v16i32_0 = shufflevector <16 x i32> %v16i32, <16 x i32> undef, <8 x i32> <i32 0, i32 2, i32 4, i32 6, i32 8, i32 10, i32 12, i32 14>
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 384 for instruction: %v16i32_1 = shufflevector <16 x i32> %v16i32, <16 x i32> undef, <8 x i32> <i32 1, i32 3, i32 5, i32 7, i32 9, i32 11, i32 13, i32 15>
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 130 for instruction: %v16i32_0 = shufflevector <16 x i32> %v16i32, <16 x i32> undef, <8 x i32> <i32 0, i32 2, i32 4, i32 6, i32 8, i32 10, i32 12, i32 14>
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 130 for instruction: %v16i32_1 = shufflevector <16 x i32> %v16i32, <16 x i32> undef, <8 x i32> <i32 1, i32 3, i32 5, i32 7, i32 9, i32 11, i32 13, i32 15>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v32i32 = load <32 x i32>, ptr %p, align 128
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 768 for instruction: %v32i32_0 = shufflevector <32 x i32> %v32i32, <32 x i32> undef, <16 x i32> <i32 0, i32 2, i32 4, i32 6, i32 8, i32 10, i32 12, i32 14, i32 16, i32 18, i32 20, i32 22, i32 24, i32 26, i32 28, i32 30>
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 768 for instruction: %v32i32_1 = shufflevector <32 x i32> %v32i32, <32 x i32> undef, <16 x i32> <i32 1, i32 3, i32 5, i32 7, i32 9, i32 11, i32 13, i32 15, i32 17, i32 19, i32 21, i32 23, i32 25, i32 27, i32 29, i32 31>
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 260 for instruction: %v32i32_0 = shufflevector <32 x i32> %v32i32, <32 x i32> undef, <16 x i32> <i32 0, i32 2, i32 4, i32 6, i32 8, i32 10, i32 12, i32 14, i32 16, i32 18, i32 20, i32 22, i32 24, i32 26, i32 28, i32 30>
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 260 for instruction: %v32i32_1 = shufflevector <32 x i32> %v32i32, <32 x i32> undef, <16 x i32> <i32 1, i32 3, i32 5, i32 7, i32 9, i32 11, i32 13, i32 15, i32 17, i32 19, i32 21, i32 23, i32 25, i32 27, i32 29, i32 31>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2i64 = load <4 x i64>, ptr %p, align 32
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 192 for instruction: %v2i64_0 = shufflevector <4 x i64> %v2i64, <4 x i64> undef, <2 x i32> <i32 0, i32 2>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 192 for instruction: %v2i64_1 = shufflevector <4 x i64> %v2i64, <4 x i64> undef, <2 x i32> <i32 1, i32 3>
@@ -252,88 +252,171 @@ define void @vld3(ptr %p) {
 }
 
 define void @vld4(ptr %p) {
-; CHECK-LABEL: 'vld4'
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v2i8 = load <8 x i8>, ptr %p, align 8
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 160 for instruction: %v2i8_0 = shufflevector <8 x i8> %v2i8, <8 x i8> undef, <2 x i32> <i32 0, i32 4>
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 160 for instruction: %v2i8_1 = shufflevector <8 x i8> %v2i8, <8 x i8> undef, <2 x i32> <i32 1, i32 5>
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 160 for instruction: %v2i8_2 = shufflevector <8 x i8> %v2i8, <8 x i8> undef, <2 x i32> <i32 2, i32 6>
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 160 for instruction: %v2i8_3 = shufflevector <8 x i8> %v2i8, <8 x i8> undef, <2 x i32> <i32 3, i32 7>
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v4i8 = load <16 x i8>, ptr %p, align 8
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 320 for instruction: %v4i8_0 = shufflevector <16 x i8> %v4i8, <16 x i8> undef, <4 x i32> <i32 0, i32 4, i32 8, i32 12>
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 320 for instruction: %v4i8_1 = shufflevector <16 x i8> %v4i8, <16 x i8> undef, <4 x i32> <i32 1, i32 5, i32 9, i32 13>
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 320 for instruction: %v4i8_2 = shufflevector <16 x i8> %v4i8, <16 x i8> undef, <4 x i32> <i32 2, i32 6, i32 10, i32 14>
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 320 for instruction: %v4i8_3 = shufflevector <16 x i8> %v4i8, <16 x i8> undef, <4 x i32> <i32 3, i32 7, i32 11, i32 15>
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v8i8 = load <32 x i8>, ptr %p, align 32
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 640 for instruction: %v8i8_0 = shufflevector <32 x i8> %v8i8, <32 x i8> undef, <8 x i32> <i32 0, i32 4, i32 8, i32 12, i32 16, i32 20, i32 24, i32 28>
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 640 for instruction: %v8i8_1 = shufflevector <32 x i8> %v8i8, <32 x i8> undef, <8 x i32> <i32 1, i32 5, i32 9, i32 13, i32 17, i32 21, i32 25, i32 29>
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 640 for instruction: %v8i8_2 = shufflevector <32 x i8> %v8i8, <32 x i8> undef, <8 x i32> <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 640 for instruction: %v8i8_3 = shufflevector <32 x i8> %v8i8, <32 x i8> undef, <8 x i32> <i32 3, i32 7, i32 11, i32 15, i32 19, i32 23, i32 27, i32 31>
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v16i8 = load <64 x i8>, ptr %p, align 64
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 1280 for instruction: %v16i8_0 = shufflevector <64 x i8> %v16i8, <64 x i8> undef, <16 x i32> <i32 0, i32 4, i32 8, i32 12, i32 16, i32 20, i32 24, i32 28, i32 32, i32 36, i32 40, i32 44, i32 48, i32 52, i32 56, i32 60>
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 1280 for instruction: %v16i8_1 = shufflevector <64 x i8> %v16i8, <64 x i8> undef, <16 x i32> <i32 1, i32 5, i32 9, i32 13, i32 17, i32 21, i32 25, i32 29, i32 33, i32 37, i32 41, i32 45, i32 49, i32 53, i32 57, i32 61>
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 1280 for instruction: %v16i8_2 = shufflevector <64 x i8> %v16i8, <64 x i8> undef, <16 x i32> <i32 2, i32 6, i32 10, i32 14, i32 18, i32 22, i32 26, i32 30, i32 34, i32 38, i32 42, i32 46, i32 50, i32 54, i32 58, i32 62>
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 1280 for instruction: %v16i8_3 = shufflevector <64 x i8> %v16i8, <64 x i8> undef, <16 x i32> <i32 3, i32 7, i32 11, i32 15, i32 19, i32 23, i32 27, i32 31, i32 35, i32 39, i32 43, i32 47, i32 51, i32 55, i32 59, i32 63>
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v2i16 = load <8 x i16>, ptr %p, align 8
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 160 for instruction: %v2i16_0 = shufflevector <8 x i16> %v2i16, <8 x i16> undef, <2 x i32> <i32 0, i32 4>
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 160 for instruction: %v2i16_1 = shufflevector <8 x i16> %v2i16, <8 x i16> undef, <2 x i32> <i32 1, i32 5>
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 160 for instruction: %v2i16_2 = shufflevector <8 x i16> %v2i16, <8 x i16> undef, <2 x i32> <i32 2, i32 6>
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 160 for instruction: %v2i16_3 = shufflevector <8 x i16> %v2i16, <8 x i16> undef, <2 x i32> <i32 3, i32 7>
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v4i16 = load <16 x i16>, ptr %p, align 32
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 320 for instruction: %v4i16_0 = shufflevector <16 x i16> %v4i16, <16 x i16> undef, <4 x i32> <i32 0, i32 4, i32 8, i32 12>
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 320 for instruction: %v4i16_1 = shufflevector <16 x i16> %v4i16, <16 x i16> undef, <4 x i32> <i32 1, i32 5, i32 9, i32 13>
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 320 for instruction: %v4i16_2 = shufflevector <16 x i16> %v4i16, <16 x i16> undef, <4 x i32> <i32 2, i32 6, i32 10, i32 14>
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 320 for instruction: %v4i16_3 = shufflevector <16 x i16> %v4i16, <16 x i16> undef, <4 x i32> <i32 3, i32 7, i32 11, i32 15>
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v8i16 = load <32 x i16>, ptr %p, align 64
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 640 for instruction: %v8i16_0 = shufflevector <32 x i16> %v8i16, <32 x i16> undef, <8 x i32> <i32 0, i32 4, i32 8, i32 12, i32 16, i32 20, i32 24, i32 28>
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 640 for instruction: %v8i16_1 = shufflevector <32 x i16> %v8i16, <32 x i16> undef, <8 x i32> <i32 1, i32 5, i32 9, i32 13, i32 17, i32 21, i32 25, i32 29>
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 640 for instruction: %v8i16_2 = shufflevector <32...
[truncated]

@github-actions GitHub Actions
Copy link

⚠️ undef deprecator found issues in your code. ⚠️

You can test this locally with the following command: 
git diff -U0 --pickaxe-regex -S '([^a-zA-Z0-9#_-]undef[^a-zA-Z0-9_-]|UndefValue::get)' 'HEAD~1' HEAD llvm/lib/Target/ARM/ARMTargetTransformInfo.cpp llvm/test/Analysis/CostModel/ARM/mve-shuffle-loadstore.ll

The following files introduce new uses of undef:

  • llvm/test/Analysis/CostModel/ARM/mve-shuffle-loadstore.ll

Undef is now deprecated and should only be used in the rare cases where no replacement is possible. For example, a load of uninitialized memory yields undef. You should use poison values for placeholders instead.

In tests, avoid using undef and having tests that trigger undefined behavior. If you need an operand with some unimportant value, you can add a new argument to the function and use that instead.

For example, this is considered a bad practice:

define void @fn() {
  ...
  br i1 undef, ...
}

Please use the following instead:

define void @fn(i1 %cond) {
  ...
  br i1 %cond, ...
}

Please refer to the Undefined Behavior Manual for more information.

john-brawn-arm
john-brawn-arm reviewed Jun 23, 2025
View reviewed changes
Copy link
Collaborator

@john-brawn-arm john-brawn-arm left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

There's a lot of duplication between the LD and ST code (I've added comments to the LD code but they also apply to the ST code). I think it makes sense to move at least of the logic here into a separate function.

llvm/lib/Target/ARM/ARMTargetTransformInfo.cpp
LT.second.getScalarSizeInBits() == 16 ||
LT.second.getScalarSizeInBits() == 32) &&
LT.second.getSizeInBits() == 128 &&
(ShuffleVectorInst::isDeInterleaveMaskOfFactor(Mask, 2) ||
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Check MaxSupportedInterleaveFactor() >= 2 here? (So -mve-max-interleave-factor=1 will work as expected)

llvm/lib/Target/ARM/ARMTargetTransformInfo.cpp
LT.second.getScalarSizeInBits() == 32) &&
LT.second.getSizeInBits() == 128 &&
(ShuffleVectorInst::isDeInterleaveMaskOfFactor(Mask, 2) ||
(TLI->getMaxSupportedInterleaveFactor() == 4 &&
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Should be >= 4, I think.

llvm/lib/Target/ARM/ARMTargetTransformInfo.cpp
(ShuffleVectorInst::isDeInterleaveMaskOfFactor(Mask, 2) ||
(TLI->getMaxSupportedInterleaveFactor() == 4 &&
ShuffleVectorInst::isDeInterleaveMaskOfFactor(Mask, 4))))
return ST->getMVEVectorCostFactor(TTI::TCK_RecipThroughput) *
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Should this be using the CostKind passed in as an argument? (Though I notice elsewhere in the function it's ignoring the argument and just using RecipThroughput)

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers

@john-brawn-arm john-brawn-arm john-brawn-arm left review comments

@stuij stuij Awaiting requested review from stuij

@SamTebbs33 SamTebbs33 Awaiting requested review from SamTebbs33

@nasherm nasherm Awaiting requested review from nasherm

Assignees
No one assigned
Labels
backend:ARM llvm:analysis Includes value tracking, cost tables and constant folding
Projects
None yet
Milestone
No milestone
Development

Successfully merging this pull request may close these issues.
3 participants
@davemgreen @llvmbot @john-brawn-arm