Skip to content

Commit 0abaf3c

Browse files
Revert "[RISCV][CostModel] VPIntrinsics have same cost as their non-vp counterparts (#67178)"
This reverts commit fc865c2. Triggering assert on X86: ``` iree-compile: /work/third_party/llvm-project/llvm/include/llvm/Support/Casting.h:662: decltype(auto) llvm::dyn_cast(From *) [To = llvm::PointerType, From = llvm::Type]: Assertion `detail::isPresent(Val) && "dyn_cast on a non-existent value"' failed. ``` See PR for comments and full stack trace.
1 parent a4d51e5 commit 0abaf3c

File tree

3 files changed

+5
-429
lines changed

3 files changed

+5
-429
lines changed

llvm/include/llvm/CodeGen/BasicTTIImpl.h

Lines changed: 0 additions & 56 deletions
Original file line numberDiff line numberDiff line change
@@ -1691,62 +1691,6 @@ class BasicTTIImplBase : public TargetTransformInfoImplCRTPBase<T> {
16911691
}
16921692
}
16931693

1694-
// VP Intrinsics should have the same cost as their non-vp counterpart.
1695-
// TODO: Adjust the cost to make the vp intrinsic cheaper than its non-vp
1696-
// counterpart when the vector length argument is smaller than the maximum
1697-
// vector length.
1698-
if (VPIntrinsic::isVPIntrinsic(ICA.getID())) {
1699-
std::optional<unsigned> FOp =
1700-
VPIntrinsic::getFunctionalOpcodeForVP(ICA.getID());
1701-
if (FOp) {
1702-
// TODO: Support other kinds of Intrinsics (i.e. reductions)
1703-
if (ICA.getID() == Intrinsic::vp_load) {
1704-
Align Alignment;
1705-
if (auto *VPI = dyn_cast_or_null<VPIntrinsic>(ICA.getInst()))
1706-
Alignment = VPI->getPointerAlignment().valueOrOne();
1707-
unsigned AS = 0;
1708-
if (ICA.getArgs().size() > 1)
1709-
if (auto *PtrTy =
1710-
dyn_cast<PointerType>(ICA.getArgs()[0]->getType()))
1711-
AS = PtrTy->getAddressSpace();
1712-
return thisT()->getMemoryOpCost(*FOp, ICA.getReturnType(), Alignment,
1713-
AS, CostKind);
1714-
}
1715-
if (ICA.getID() == Intrinsic::vp_store) {
1716-
Align Alignment;
1717-
if (auto *VPI = dyn_cast_or_null<VPIntrinsic>(ICA.getInst()))
1718-
Alignment = VPI->getPointerAlignment().valueOrOne();
1719-
unsigned AS = 0;
1720-
if (ICA.getArgs().size() >= 2)
1721-
if (auto *PtrTy =
1722-
dyn_cast<PointerType>(ICA.getArgs()[1]->getType()))
1723-
AS = PtrTy->getAddressSpace();
1724-
return thisT()->getMemoryOpCost(*FOp, Args[0]->getType(), Alignment,
1725-
AS, CostKind);
1726-
}
1727-
if (VPBinOpIntrinsic::isVPBinOp(ICA.getID())) {
1728-
return thisT()->getArithmeticInstrCost(*FOp, ICA.getReturnType(),
1729-
CostKind);
1730-
}
1731-
}
1732-
1733-
std::optional<Intrinsic::ID> FID =
1734-
VPIntrinsic::getFunctionalIntrinsicIDForVP(ICA.getID());
1735-
if (FID) {
1736-
// Non-vp version will have same Args/Tys except mask and vector length.
1737-
assert(ICA.getArgs().size() >= 2 && ICA.getArgTypes().size() >= 2 &&
1738-
"Expected VPIntrinsic to have Mask and Vector Length args and "
1739-
"types");
1740-
ArrayRef<const Value *> NewArgs = ArrayRef(ICA.getArgs()).drop_back(2);
1741-
ArrayRef<Type *> NewTys = ArrayRef(ICA.getArgTypes()).drop_back(2);
1742-
1743-
IntrinsicCostAttributes NewICA(*FID, ICA.getReturnType(), NewArgs,
1744-
NewTys, ICA.getFlags(), ICA.getInst(),
1745-
ICA.getScalarizationCost());
1746-
return thisT()->getIntrinsicInstrCost(NewICA, CostKind);
1747-
}
1748-
}
1749-
17501694
// Assume that we need to scalarize this intrinsic.
17511695
// Compute the scalarization overhead based on Args for a vector
17521696
// intrinsic.

llvm/test/Analysis/CostModel/RISCV/gep.ll

Lines changed: 4 additions & 4 deletions
Original file line numberDiff line numberDiff line change
@@ -270,7 +270,7 @@ define void @non_foldable_vector_uses(ptr %base, <2 x ptr> %base.vec) {
270270
; RVI-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %4 = getelementptr i8, ptr %base, i32 42
271271
; RVI-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %x4 = call <2 x i8> @llvm.masked.expandload.v2i8(ptr %4, <2 x i1> undef, <2 x i8> undef)
272272
; RVI-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %5 = getelementptr i8, ptr %base, i32 42
273-
; RVI-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %x5 = call <2 x i8> @llvm.vp.load.v2i8.p0(ptr %5, <2 x i1> undef, i32 undef)
273+
; RVI-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %x5 = call <2 x i8> @llvm.vp.load.v2i8.p0(ptr %5, <2 x i1> undef, i32 undef)
274274
; RVI-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %6 = getelementptr i8, ptr %base, i32 42
275275
; RVI-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %x6 = call <2 x i8> @llvm.experimental.vp.strided.load.v2i8.p0.i64(ptr %6, i64 undef, <2 x i1> undef, i32 undef)
276276
; RVI-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %7 = getelementptr i8, ptr %base, i32 42
@@ -282,7 +282,7 @@ define void @non_foldable_vector_uses(ptr %base, <2 x ptr> %base.vec) {
282282
; RVI-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %10 = getelementptr i8, ptr %base, i32 42
283283
; RVI-NEXT: Cost Model: Found an estimated cost of 12 for instruction: call void @llvm.masked.compressstore.v2i8(<2 x i8> undef, ptr %10, <2 x i1> undef)
284284
; RVI-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %11 = getelementptr i8, ptr %base, i32 42
285-
; RVI-NEXT: Cost Model: Found an estimated cost of 1 for instruction: call void @llvm.vp.store.v2i8.p0(<2 x i8> undef, ptr %11, <2 x i1> undef, i32 undef)
285+
; RVI-NEXT: Cost Model: Found an estimated cost of 12 for instruction: call void @llvm.vp.store.v2i8.p0(<2 x i8> undef, ptr %11, <2 x i1> undef, i32 undef)
286286
; RVI-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %12 = getelementptr i8, ptr %base, i32 42
287287
; RVI-NEXT: Cost Model: Found an estimated cost of 12 for instruction: call void @llvm.experimental.vp.strided.store.v2i8.p0.i64(<2 x i8> undef, ptr %12, i64 undef, <2 x i1> undef, i32 undef)
288288
; RVI-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
@@ -340,7 +340,7 @@ define void @foldable_vector_uses(ptr %base, <2 x ptr> %base.vec) {
340340
; RVI-NEXT: Cost Model: Found an estimated cost of 0 for instruction: %4 = getelementptr i8, ptr %base, i32 0
341341
; RVI-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %x4 = call <2 x i8> @llvm.masked.expandload.v2i8(ptr %4, <2 x i1> undef, <2 x i8> undef)
342342
; RVI-NEXT: Cost Model: Found an estimated cost of 0 for instruction: %5 = getelementptr i8, ptr %base, i32 0
343-
; RVI-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %x5 = call <2 x i8> @llvm.vp.load.v2i8.p0(ptr %5, <2 x i1> undef, i32 undef)
343+
; RVI-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %x5 = call <2 x i8> @llvm.vp.load.v2i8.p0(ptr %5, <2 x i1> undef, i32 undef)
344344
; RVI-NEXT: Cost Model: Found an estimated cost of 0 for instruction: %6 = getelementptr i8, ptr %base, i32 0
345345
; RVI-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %x6 = call <2 x i8> @llvm.experimental.vp.strided.load.v2i8.p0.i64(ptr %6, i64 undef, <2 x i1> undef, i32 undef)
346346
; RVI-NEXT: Cost Model: Found an estimated cost of 0 for instruction: %7 = getelementptr i8, ptr %base, i32 0
@@ -352,7 +352,7 @@ define void @foldable_vector_uses(ptr %base, <2 x ptr> %base.vec) {
352352
; RVI-NEXT: Cost Model: Found an estimated cost of 0 for instruction: %10 = getelementptr i8, ptr %base, i32 0
353353
; RVI-NEXT: Cost Model: Found an estimated cost of 12 for instruction: call void @llvm.masked.compressstore.v2i8(<2 x i8> undef, ptr %10, <2 x i1> undef)
354354
; RVI-NEXT: Cost Model: Found an estimated cost of 0 for instruction: %11 = getelementptr i8, ptr %base, i32 0
355-
; RVI-NEXT: Cost Model: Found an estimated cost of 1 for instruction: call void @llvm.vp.store.v2i8.p0(<2 x i8> undef, ptr %11, <2 x i1> undef, i32 undef)
355+
; RVI-NEXT: Cost Model: Found an estimated cost of 12 for instruction: call void @llvm.vp.store.v2i8.p0(<2 x i8> undef, ptr %11, <2 x i1> undef, i32 undef)
356356
; RVI-NEXT: Cost Model: Found an estimated cost of 0 for instruction: %12 = getelementptr i8, ptr %base, i32 0
357357
; RVI-NEXT: Cost Model: Found an estimated cost of 12 for instruction: call void @llvm.experimental.vp.strided.store.v2i8.p0.i64(<2 x i8> undef, ptr %12, i64 undef, <2 x i1> undef, i32 undef)
358358
; RVI-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void

0 commit comments

Comments
 (0)