[RISCV] Fix crash when unrolling loop containing vector instructions

[arcbbb]

When MVT is not a vector type, TCK_CodeSize should return an invalid cost. This patch adds a check in the beginning to make sure all cost kinds return invalid costs consistently.

Before this patch, TCK_CodeSize returns a valid cost on scalar MVT but other cost kinds doesn't.

This fixes the issue #83294 where a loop contains vector instructions and MVT is scalar after type legalization when the vector extension is not enabled,

[llvmbot]

@llvm/pr-subscribers-llvm-analysis

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

Author: Shih-Po Hung (arcbbb)

Changes

When MVT is not a vector type, TCK_CodeSize should return an invalid cost. This patch adds a check in the beginning to make sure all cost kinds return invalid costs consistently.

Before this patch, TCK_CodeSize returns a valid cost on scalar MVT but other cost kinds doesn't.

This fixes the issue #83294 where a loop contains vector instructions and MVT is scalar after type legalization when the vector extension is not enabled,

---

Full diff: https://github.com/llvm/llvm-project/pull/83384.diff

2 Files Affected:

• (modified) llvm/lib/Target/RISCV/RISCVTargetTransformInfo.cpp (+3)
• (added) llvm/test/Analysis/CostModel/RISCV/rvv-invalid-cost.ll (+53)

diff --git a/llvm/lib/Target/RISCV/RISCVTargetTransformInfo.cpp b/llvm/lib/Target/RISCV/RISCVTargetTransformInfo.cpp
index f04968d82e86e2..54c7402e4444cd 100644
--- a/llvm/lib/Target/RISCV/RISCVTargetTransformInfo.cpp
+++ b/llvm/lib/Target/RISCV/RISCVTargetTransformInfo.cpp
@@ -37,6 +37,9 @@ static cl::opt<unsigned> SLPMaxVF(
 InstructionCost
 RISCVTTIImpl::getRISCVInstructionCost(ArrayRef<unsigned> OpCodes, MVT VT,
                                       TTI::TargetCostKind CostKind) {
+  // Check if the type is valid for all CostKind
+  if (!VT.isVector())
+    return InstructionCost::getInvalid();
   size_t NumInstr = OpCodes.size();
   if (CostKind == TTI::TCK_CodeSize)
     return NumInstr;
diff --git a/llvm/test/Analysis/CostModel/RISCV/rvv-invalid-cost.ll b/llvm/test/Analysis/CostModel/RISCV/rvv-invalid-cost.ll
new file mode 100644
index 00000000000000..8df26d88f52898
--- /dev/null
+++ b/llvm/test/Analysis/CostModel/RISCV/rvv-invalid-cost.ll
@@ -0,0 +1,53 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 4
+; RUN: opt < %s -mtriple=riscv64 -mattr=+f,+d --passes=loop-unroll-full -S | FileCheck %s
+
+; Check it doesn' crash when the vector extension is not enabled.
+define void @foo() {
+; CHECK-LABEL: define void @foo(
+; CHECK-SAME: ) #[[ATTR0:[0-9]+]] {
+; CHECK-NEXT:  entry:
+; CHECK-NEXT:    br label [[FOR_BODY:%.*]]
+; CHECK:       for.body:
+; CHECK-NEXT:    [[INDVARS_IV1:%.*]] = phi i64 [ [[INDVARS_IV_NEXT:%.*]], [[FOR_BODY]] ], [ 0, [[ENTRY:%.*]] ]
+; CHECK-NEXT:    [[TMP0:%.*]] = load float, ptr null, align 4
+; CHECK-NEXT:    [[SPLAT_SPLAT_I_I_I:%.*]] = shufflevector <2 x float> zeroinitializer, <2 x float> zeroinitializer, <2 x i32> zeroinitializer
+; CHECK-NEXT:    [[CMP1_I_I_I:%.*]] = fcmp ogt <2 x float> zeroinitializer, zeroinitializer
+; CHECK-NEXT:    [[SPLAT_SPLAT3_I_I_I:%.*]] = shufflevector <2 x i32> zeroinitializer, <2 x i32> zeroinitializer, <2 x i32> zeroinitializer
+; CHECK-NEXT:    [[XOR3_I_I_I_I_I:%.*]] = select <2 x i1> zeroinitializer, <2 x i32> zeroinitializer, <2 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP1:%.*]] = load float, ptr null, align 4
+; CHECK-NEXT:    [[SPLAT_SPLAT8_I_I_I:%.*]] = shufflevector <2 x float> zeroinitializer, <2 x float> zeroinitializer, <2 x i32> zeroinitializer
+; CHECK-NEXT:    [[SUB_I_I_I:%.*]] = fsub <2 x float> zeroinitializer, zeroinitializer
+; CHECK-NEXT:    [[MUL_I_I_I:%.*]] = shl i64 0, 0
+; CHECK-NEXT:    [[TMP2:%.*]] = load float, ptr null, align 4
+; CHECK-NEXT:    [[SPLAT_SPLAT_I_I_I_I:%.*]] = shufflevector <2 x float> zeroinitializer, <2 x float> zeroinitializer, <2 x i32> zeroinitializer
+; CHECK-NEXT:    [[XOR3_I_I_I_V_I_I:%.*]] = select <2 x i1> zeroinitializer, <2 x float> zeroinitializer, <2 x float> zeroinitializer
+; CHECK-NEXT:    [[INDVARS_IV_NEXT]] = add i64 [[INDVARS_IV1]], 1
+; CHECK-NEXT:    [[EXITCOND:%.*]] = icmp ne i64 [[INDVARS_IV1]], 8
+; CHECK-NEXT:    br i1 [[EXITCOND]], label [[FOR_BODY]], label [[EXIT:%.*]]
+; CHECK:       exit:
+; CHECK-NEXT:    ret void
+;
+entry:
+  br label %for.body
+
+for.body:                                         ; preds = %for.body, %entry
+  %indvars.iv1 = phi i64 [ %indvars.iv.next, %for.body ], [ 0, %entry ]
+  %0 = load float, ptr null, align 4
+  %splat.splat.i.i.i = shufflevector <2 x float> zeroinitializer, <2 x float> zeroinitializer, <2 x i32> zeroinitializer
+  %cmp1.i.i.i = fcmp ogt <2 x float> zeroinitializer, zeroinitializer
+  %splat.splat3.i.i.i = shufflevector <2 x i32> zeroinitializer, <2 x i32> zeroinitializer, <2 x i32> zeroinitializer
+  %xor3.i.i.i.i.i = select <2 x i1> zeroinitializer, <2 x i32> zeroinitializer, <2 x i32> zeroinitializer
+  %1 = load float, ptr null, align 4
+  %splat.splat8.i.i.i = shufflevector <2 x float> zeroinitializer, <2 x float> zeroinitializer, <2 x i32> zeroinitializer
+  %sub.i.i.i = fsub <2 x float> zeroinitializer, zeroinitializer
+  %mul.i.i.i = shl i64 0, 0
+  %2 = load float, ptr null, align 4
+  %splat.splat.i.i.i.i = shufflevector <2 x float> zeroinitializer, <2 x float> zeroinitializer, <2 x i32> zeroinitializer
+  %xor3.i.i.i.v.i.i = select <2 x i1> zeroinitializer, <2 x float> zeroinitializer, <2 x float> zeroinitializer
+  %indvars.iv.next = add i64 %indvars.iv1, 1
+  %exitcond = icmp ne i64 %indvars.iv1, 8
+  br i1 %exitcond, label %for.body, label %exit
+
+exit:                                             ; preds = %for.body
+  ret void
+}

[frasercrmck]

LGTM, thanks

[frasercrmck]

Is rvv-invalid-cost the best name for this test? It might make people think there's an invalid cost when using +v.

[arcbbb]

or rvv-cost-without-v.ll?

[frasercrmck]

vector-cost-without-v.ll?

[arcbbb]

Sounds good!

[frasercrmck]

This might be an aside but I find it a little surprising that this function's called getRISCVInstructionCost but it only handles vector instructions. Sounds like it should be getRVVInstructionCost. I might be missing something though.

[arcbbb]

Good point. Your suggested name appears to be clearer.

[lukel97]

Where is this getting called from with a scalar MVT? I would have thought that after scalarization we wouldn't have any vector instructions that would still be calling into this

[arcbbb]

It is from LoopUnrollPass along the way from TTI.getInstructionCost to TTI.getShuffleCost for

%splat.splat.i.i.i = shufflevector <2 x float> zeroinitializer, <2 x float> zeroinitializer, <2 x i32> zeroinitializer

The data type is <2 x float>, but becomes float after getTypeLegalizationCost(Tp); then enters getRISCVInstructionCost().
I'm just not sure how a vector type can be generated without a vector support.

If it is true, we might consider checking if the type is still vector after legalization at the beginning of TTI functions which accept vector types.

[lukel97]

Ah ok, thanks for clarifying. I guess there's nothing stopping a frontend from just emitting vector LLVM IR independently of the autovectorizers.

I'm not strongly opinionated about where we check that the legalized type is vector, I presume both will result in an invalid cost

[frasercrmck]

Yeah, in our case it's because we're compiling OpenCL or SYCL, with vector data types in the source.