[llvm] [SLP][TLI] Add cost model/vectorization support for [u|s]cmp (PR #106747)

via llvm-commits llvm-commits at lists.llvm.org
Fri Aug 30 08:27:09 PDT 2024


llvmbot wrote:


<!--LLVM PR SUMMARY COMMENT-->

@llvm/pr-subscribers-backend-x86

Author: Yingwei Zheng (dtcxzyw)

<details>
<summary>Changes</summary>

This patch adds cost model/vectorization support to vectorize [u|s]cmp intrinsic calls.


---

Patch is 554.30 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/106747.diff


8 Files Affected:

- (modified) llvm/include/llvm/CodeGen/BasicTTIImpl.h (+37) 
- (modified) llvm/lib/Analysis/VectorUtils.cpp (+4) 
- (modified) llvm/test/Analysis/CostModel/X86/icmp-codesize.ll (+332-240) 
- (modified) llvm/test/Analysis/CostModel/X86/icmp-latency.ll (+366-320) 
- (modified) llvm/test/Analysis/CostModel/X86/icmp-sizelatency.ll (+332-240) 
- (modified) llvm/test/Analysis/CostModel/X86/icmp.ll (+480-480) 
- (added) llvm/test/Transforms/SLPVectorizer/X86/arith-scmp.ll (+646) 
- (added) llvm/test/Transforms/SLPVectorizer/X86/arith-ucmp.ll (+681) 


``````````diff
diff --git a/llvm/include/llvm/CodeGen/BasicTTIImpl.h b/llvm/include/llvm/CodeGen/BasicTTIImpl.h
index 47323ca067a435..f4bddcfc815ff7 100644
--- a/llvm/include/llvm/CodeGen/BasicTTIImpl.h
+++ b/llvm/include/llvm/CodeGen/BasicTTIImpl.h
@@ -2196,6 +2196,12 @@ class BasicTTIImplBase : public TargetTransformInfoImplCRTPBase<T> {
     case Intrinsic::bitreverse:
       ISD = ISD::BITREVERSE;
       break;
+    case Intrinsic::ucmp:
+      ISD = ISD::UCMP;
+      break;
+    case Intrinsic::scmp:
+      ISD = ISD::SCMP;
+      break;
     }
 
     auto *ST = dyn_cast<StructType>(RetTy);
@@ -2433,6 +2439,37 @@ class BasicTTIImplBase : public TargetTransformInfoImplCRTPBase<T> {
       }
       return Cost;
     }
+    case Intrinsic::ucmp:
+    case Intrinsic::scmp: {
+      InstructionCost Cost = 0;
+      bool IsSigned = IID == Intrinsic::scmp;
+      ICmpInst::Predicate GTPred =
+          IsSigned ? CmpInst::ICMP_SGT : CmpInst::ICMP_UGT;
+      ICmpInst::Predicate LTPred =
+          IsSigned ? CmpInst::ICMP_SLT : CmpInst::ICMP_ULT;
+      Type *CmpTy = Tys[0];
+
+      if (TLI->shouldExpandCmpUsingSelects()) {
+        // x < y ? -1 : (x > y ? 1 : 0)
+        Cost += thisT()->getCmpSelInstrCost(BinaryOperator::ICmp, CmpTy, RetTy,
+                                            GTPred, CostKind);
+        Cost += thisT()->getCmpSelInstrCost(BinaryOperator::ICmp, CmpTy, RetTy,
+                                            LTPred, CostKind);
+      } else {
+        // zext(x > y) - zext(x < y)
+        Type *CondTy = RetTy->getWithNewBitWidth(1);
+        Cost += thisT()->getCmpSelInstrCost(BinaryOperator::ICmp, CmpTy, CondTy,
+                                            GTPred, CostKind);
+        Cost += thisT()->getCmpSelInstrCost(BinaryOperator::ICmp, CmpTy, CondTy,
+                                            LTPred, CostKind);
+        Cost +=
+            2 * thisT()->getCastInstrCost(CastInst::ZExt, RetTy, CondTy,
+                                          TTI::CastContextHint::None, CostKind);
+        Cost += thisT()->getArithmeticInstrCost(BinaryOperator::Sub, RetTy,
+                                                CostKind);
+      }
+      return Cost;
+    }
     default:
       break;
     }
diff --git a/llvm/lib/Analysis/VectorUtils.cpp b/llvm/lib/Analysis/VectorUtils.cpp
index cc742ab35f4498..0c806820a31e17 100644
--- a/llvm/lib/Analysis/VectorUtils.cpp
+++ b/llvm/lib/Analysis/VectorUtils.cpp
@@ -97,6 +97,8 @@ bool llvm::isTriviallyVectorizable(Intrinsic::ID ID) {
   case Intrinsic::fptoui_sat:
   case Intrinsic::lrint:
   case Intrinsic::llrint:
+  case Intrinsic::ucmp:
+  case Intrinsic::scmp:
     return true;
   default:
     return false;
@@ -132,6 +134,8 @@ bool llvm::isVectorIntrinsicWithOverloadTypeAtArg(Intrinsic::ID ID,
   case Intrinsic::fptoui_sat:
   case Intrinsic::lrint:
   case Intrinsic::llrint:
+  case Intrinsic::ucmp:
+  case Intrinsic::scmp:
     return OpdIdx == -1 || OpdIdx == 0;
   case Intrinsic::is_fpclass:
     return OpdIdx == 0;
diff --git a/llvm/test/Analysis/CostModel/X86/icmp-codesize.ll b/llvm/test/Analysis/CostModel/X86/icmp-codesize.ll
index 2dc6737a3d8a07..b8d0e04daccc84 100644
--- a/llvm/test/Analysis/CostModel/X86/icmp-codesize.ll
+++ b/llvm/test/Analysis/CostModel/X86/icmp-codesize.ll
@@ -2090,142 +2090,188 @@ define i32 @cmp_int_ult(i8 %arg8, <16 x i8> %argv16i8, <32 x i8> %argv32i8, <64
 }
 
 define i32 @scmp_int(i8 %arg8, <16 x i8> %argv16i8, <32 x i8> %argv32i8, <64 x i8> %argv64i8, <128 x i8> %argv128i8, i16 %arg16, <8 x i16> %argv8i16, <16 x i16> %argv16i16, <32 x i16> %argv32i16, <64 x i16> %argv64i16, i32 %arg32, <4 x i32> %argv4i32, <8 x i32> %argv8i32, <16 x i32> %argv16i32, <32 x i32> %argv32i32, i64 %arg64, <2 x i64> %argv2i64, <4 x i64> %argv4i64, <8 x i64> %argv8i64, <16 x i64> %argv16i64) {
+; SSE2-LABEL: 'scmp_int'
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %I8 = call i8 @llvm.scmp.i8.i8(i8 %arg8, i8 %arg8)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %V16I8 = call <16 x i8> @llvm.scmp.v16i8.v16i8(<16 x i8> %argv16i8, <16 x i8> %argv16i8)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %V32I8 = call <32 x i8> @llvm.scmp.v32i8.v32i8(<32 x i8> %argv32i8, <32 x i8> %argv32i8)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %V64I8 = call <64 x i8> @llvm.scmp.v64i8.v64i8(<64 x i8> %argv64i8, <64 x i8> %argv64i8)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 19 for instruction: %V128I8 = call <128 x i8> @llvm.scmp.v128i8.v128i8(<128 x i8> %argv128i8, <128 x i8> %argv128i8)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %I16 = call i16 @llvm.scmp.i16.i16(i16 %arg16, i16 %arg16)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %V8I16 = call <8 x i16> @llvm.scmp.v8i16.v8i16(<8 x i16> %argv8i16, <8 x i16> %argv8i16)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %V16I16 = call <16 x i16> @llvm.scmp.v16i16.v16i16(<16 x i16> %argv16i16, <16 x i16> %argv16i16)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 17 for instruction: %V32I16 = call <32 x i16> @llvm.scmp.v32i16.v32i16(<32 x i16> %argv32i16, <32 x i16> %argv32i16)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 33 for instruction: %V64I16 = call <64 x i16> @llvm.scmp.v64i16.v64i16(<64 x i16> %argv64i16, <64 x i16> %argv64i16)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %I32 = call i32 @llvm.scmp.i32.i32(i32 %arg32, i32 %arg32)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %V4I32 = call <4 x i32> @llvm.scmp.v4i32.v4i32(<4 x i32> %argv4i32, <4 x i32> %argv4i32)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %V8I32 = call <8 x i32> @llvm.scmp.v8i32.v8i32(<8 x i32> %argv8i32, <8 x i32> %argv8i32)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 17 for instruction: %V16I32 = call <16 x i32> @llvm.scmp.v16i32.v16i32(<16 x i32> %argv16i32, <16 x i32> %argv16i32)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 33 for instruction: %V32I32 = call <32 x i32> @llvm.scmp.v32i32.v32i32(<32 x i32> %argv32i32, <32 x i32> %argv32i32)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %I64 = call i64 @llvm.scmp.i64.i64(i64 %arg64, i64 %arg64)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 13 for instruction: %V2I64 = call <2 x i64> @llvm.scmp.v2i64.v2i64(<2 x i64> %argv2i64, <2 x i64> %argv2i64)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 26 for instruction: %V4I64 = call <4 x i64> @llvm.scmp.v4i64.v4i64(<4 x i64> %argv4i64, <4 x i64> %argv4i64)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 52 for instruction: %V8I64 = call <8 x i64> @llvm.scmp.v8i64.v8i64(<8 x i64> %argv8i64, <8 x i64> %argv8i64)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 104 for instruction: %V16I64 = call <16 x i64> @llvm.scmp.v16i64.v16i64(<16 x i64> %argv16i64, <16 x i64> %argv16i64)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
+;
+; SSE42-LABEL: 'scmp_int'
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %I8 = call i8 @llvm.scmp.i8.i8(i8 %arg8, i8 %arg8)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %V16I8 = call <16 x i8> @llvm.scmp.v16i8.v16i8(<16 x i8> %argv16i8, <16 x i8> %argv16i8)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %V32I8 = call <32 x i8> @llvm.scmp.v32i8.v32i8(<32 x i8> %argv32i8, <32 x i8> %argv32i8)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %V64I8 = call <64 x i8> @llvm.scmp.v64i8.v64i8(<64 x i8> %argv64i8, <64 x i8> %argv64i8)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 19 for instruction: %V128I8 = call <128 x i8> @llvm.scmp.v128i8.v128i8(<128 x i8> %argv128i8, <128 x i8> %argv128i8)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %I16 = call i16 @llvm.scmp.i16.i16(i16 %arg16, i16 %arg16)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %V8I16 = call <8 x i16> @llvm.scmp.v8i16.v8i16(<8 x i16> %argv8i16, <8 x i16> %argv8i16)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %V16I16 = call <16 x i16> @llvm.scmp.v16i16.v16i16(<16 x i16> %argv16i16, <16 x i16> %argv16i16)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 17 for instruction: %V32I16 = call <32 x i16> @llvm.scmp.v32i16.v32i16(<32 x i16> %argv32i16, <32 x i16> %argv32i16)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 33 for instruction: %V64I16 = call <64 x i16> @llvm.scmp.v64i16.v64i16(<64 x i16> %argv64i16, <64 x i16> %argv64i16)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %I32 = call i32 @llvm.scmp.i32.i32(i32 %arg32, i32 %arg32)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %V4I32 = call <4 x i32> @llvm.scmp.v4i32.v4i32(<4 x i32> %argv4i32, <4 x i32> %argv4i32)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %V8I32 = call <8 x i32> @llvm.scmp.v8i32.v8i32(<8 x i32> %argv8i32, <8 x i32> %argv8i32)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 17 for instruction: %V16I32 = call <16 x i32> @llvm.scmp.v16i32.v16i32(<16 x i32> %argv16i32, <16 x i32> %argv16i32)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 33 for instruction: %V32I32 = call <32 x i32> @llvm.scmp.v32i32.v32i32(<32 x i32> %argv32i32, <32 x i32> %argv32i32)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %I64 = call i64 @llvm.scmp.i64.i64(i64 %arg64, i64 %arg64)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %V2I64 = call <2 x i64> @llvm.scmp.v2i64.v2i64(<2 x i64> %argv2i64, <2 x i64> %argv2i64)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %V4I64 = call <4 x i64> @llvm.scmp.v4i64.v4i64(<4 x i64> %argv4i64, <4 x i64> %argv4i64)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %V8I64 = call <8 x i64> @llvm.scmp.v8i64.v8i64(<8 x i64> %argv8i64, <8 x i64> %argv8i64)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %V16I64 = call <16 x i64> @llvm.scmp.v16i64.v16i64(<16 x i64> %argv16i64, <16 x i64> %argv16i64)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
+;
 ; AVX1-LABEL: 'scmp_int'
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = call i8 @llvm.scmp.i8.i8(i8 %arg8, i8 %arg8)
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 48 for instruction: %V16I8 = call <16 x i8> @llvm.scmp.v16i8.v16i8(<16 x i8> %argv16i8, <16 x i8> %argv16i8)
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 98 for instruction: %V32I8 = call <32 x i8> @llvm.scmp.v32i8.v32i8(<32 x i8> %argv32i8, <32 x i8> %argv32i8)
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 196 for instruction: %V64I8 = call <64 x i8> @llvm.scmp.v64i8.v64i8(<64 x i8> %argv64i8, <64 x i8> %argv64i8)
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 392 for instruction: %V128I8 = call <128 x i8> @llvm.scmp.v128i8.v128i8(<128 x i8> %argv128i8, <128 x i8> %argv128i8)
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = call i16 @llvm.scmp.i16.i16(i16 %arg16, i16 %arg16)
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %V8I16 = call <8 x i16> @llvm.scmp.v8i16.v8i16(<8 x i16> %argv8i16, <8 x i16> %argv8i16)
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 50 for instruction: %V16I16 = call <16 x i16> @llvm.scmp.v16i16.v16i16(<16 x i16> %argv16i16, <16 x i16> %argv16i16)
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 100 for instruction: %V32I16 = call <32 x i16> @llvm.scmp.v32i16.v32i16(<32 x i16> %argv32i16, <32 x i16> %argv32i16)
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 200 for instruction: %V64I16 = call <64 x i16> @llvm.scmp.v64i16.v64i16(<64 x i16> %argv64i16, <64 x i16> %argv64i16)
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = call i32 @llvm.scmp.i32.i32(i32 %arg32, i32 %arg32)
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V4I32 = call <4 x i32> @llvm.scmp.v4i32.v4i32(<4 x i32> %argv4i32, <4 x i32> %argv4i32)
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 26 for instruction: %V8I32 = call <8 x i32> @llvm.scmp.v8i32.v8i32(<8 x i32> %argv8i32, <8 x i32> %argv8i32)
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 52 for instruction: %V16I32 = call <16 x i32> @llvm.scmp.v16i32.v16i32(<16 x i32> %argv16i32, <16 x i32> %argv16i32)
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 104 for instruction: %V32I32 = call <32 x i32> @llvm.scmp.v32i32.v32i32(<32 x i32> %argv32i32, <32 x i32> %argv32i32)
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = call i64 @llvm.scmp.i64.i64(i64 %arg64, i64 %arg64)
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V2I64 = call <2 x i64> @llvm.scmp.v2i64.v2i64(<2 x i64> %argv2i64, <2 x i64> %argv2i64)
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V4I64 = call <4 x i64> @llvm.scmp.v4i64.v4i64(<4 x i64> %argv4i64, <4 x i64> %argv4i64)
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %V8I64 = call <8 x i64> @llvm.scmp.v8i64.v8i64(<8 x i64> %argv8i64, <8 x i64> %argv8i64)
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 56 for instruction: %V16I64 = call <16 x i64> @llvm.scmp.v16i64.v16i64(<16 x i64> %argv16i64, <16 x i64> %argv16i64)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %I8 = call i8 @llvm.scmp.i8.i8(i8 %arg8, i8 %arg8)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %V16I8 = call <16 x i8> @llvm.scmp.v16i8.v16i8(<16 x i8> %argv16i8, <16 x i8> %argv16i8)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 17 for instruction: %V32I8 = call <32 x i8> @llvm.scmp.v32i8.v32i8(<32 x i8> %argv32i8, <32 x i8> %argv32i8)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V64I8 = call <64 x i8> @llvm.scmp.v64i8.v64i8(<64 x i8> %argv64i8, <64 x i8> %argv64i8)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 62 for instruction: %V128I8 = call <128 x i8> @llvm.scmp.v128i8.v128i8(<128 x i8> %argv128i8, <128 x i8> %argv128i8)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %I16 = call i16 @llvm.scmp.i16.i16(i16 %arg16, i16 %arg16)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %V8I16 = call <8 x i16> @llvm.scmp.v8i16.v8i16(<8 x i16> %argv8i16, <8 x i16> %argv8i16)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 17 for instruction: %V16I16 = call <16 x i16> @llvm.scmp.v16i16.v16i16(<16 x i16> %argv16i16, <16 x i16> %argv16i16)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V32I16 = call <32 x i16> @llvm.scmp.v32i16.v32i16(<32 x i16> %argv32i16, <32 x i16> %argv32i16)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 62 for instruction: %V64I16 = call <64 x i16> @llvm.scmp.v64i16.v64i16(<64 x i16> %argv64i16, <64 x i16> %argv64i16)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %I32 = call i32 @llvm.scmp.i32.i32(i32 %arg32, i32 %arg32)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %V4I32 = call <4 x i32> @llvm.scmp.v4i32.v4i32(<4 x i32> %argv4i32, <4 x i32> %argv4i32)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 17 for instruction: %V8I32 = call <8 x i32> @llvm.scmp.v8i32.v8i32(<8 x i32> %argv8i32, <8 x i32> %argv8i32)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 34 for instruction: %V16I32 = call <16 x i32> @llvm.scmp.v16i32.v16i32(<16 x i32> %argv16i32, <16 x i32> %argv16i32)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 62 for instruction: %V32I32 = call <32 x i32> @llvm.scmp.v32i32.v32i32(<32 x i32> %argv32i32, <32 x i32> %argv32i32)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %I64 = call i64 @llvm.scmp.i64.i64(i64 %arg64, i64 %arg64)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %V2I64 = call <2 x i64> @llvm.scmp.v2i64.v2i64(<2 x i64> %argv2i64, <2 x i64> %argv2i64)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 17 for instruction: %V4I64 = call <4 x i64> @llvm.scmp.v4i64.v4i64(<4 x i64> %argv4i64, <4 x i64> %argv4i64)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 34 for instruction: %V8I64 = call <8 x i64> @llvm.scmp.v8i64.v8i64(<8 x i64> %argv8i64, <8 x i64> %argv8i64)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 68 for instruction: %V16I64 = call <16 x i64> @llvm.scmp.v16i64.v16i64(<16 x i64> %argv16i64, <16 x i64> %argv16i64)
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
 ;
 ; AVX2-LABEL: 'scmp_int'
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = call i8 @llvm.scmp.i8.i8(i8 %arg8, i8 %arg8)
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 48 for instruction: %V16I8 = call <16 x i8> @llvm.scmp.v16i8.v16i8(<16 x i8> %argv16i8, <16 x i8> %argv16i8)
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 98 for instruction: %V32I8 = call <32 x i8> @llvm.scmp.v32i8.v32i8(<32 x i8> %argv32i8, <32 x i8> %argv32i8)
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 196 for instruction: %V64I8 = call <64 x i8> @llvm.scmp.v64i8.v64i8(<64 x i8> %argv64i8, <64 x i8> %argv64i8)
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 392 for instruction: %V128I8 = call <128 x i8> @llvm.scmp.v128i8.v128i8(<128 x i8> %argv128i8, <128 x i8> %argv128i8)
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = call i16 @llvm.scmp.i16.i16(i16 %arg16, i16 %arg16)
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %V8I16 = call <8 x i16> @llvm.scmp.v8i16.v8i16(<8 x i16> %argv8i16, <8 x i16> %argv8i16)
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 50 for instruction: %V16I16 = call <16 x i16> @llvm.scmp.v16i16.v16i16(<16 x i16> %argv16i16, <16 x i16> %argv16i16)
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 100 for instruction: %V32I16 = call <32 x i16> @llvm.scmp.v32i16.v32i16(<32 x i16> %argv32i16, <32 x i16> %argv32i16)
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 200 for instruction: %V64I16 = call <64 x i16> @llvm.scmp.v64i16.v64i16(<64 x i16> %argv64i16, <64 x i16> %argv64i16)
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = call i32 @llvm.scmp.i32.i32(i32 %arg32, i32 %arg32)
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V4I32 = call <4 x i32> @llvm.scmp.v4i32.v4i32(<4 x i32> %argv4i32, <4 x i32> %argv4i32)
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 26 for instruction: %V8I32 = call <8 x i32> @llvm.scmp.v8i32.v8i32(<8 x i32> %argv8i32, <8 x i32> %argv8i32)
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 52 for instruction: %V16I32 = call <16 x i32> @llvm.scmp.v16i32.v16i32(<16 x i32> %argv16i32, <16 x i32> %argv16i32)
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 104 for instruction: %V32I32 = call <32 x i32> @llvm.scmp.v32i32.v32i32(<32 x i32> %argv32i32, <32 x i32> %argv32i32)
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = call i64 @llvm.scmp.i64.i64(i64 %arg64, i64 %arg64)
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V2I64 = call <2 x i64> @llvm.scmp.v2i64.v2i64(<2 x i64> %argv2i64, <2 x i64> %argv2i64)
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V4I64 = call <4 x i64> @llvm.scmp.v4i64.v4i64(<4 x i64> %argv4i64, <4 x i64> %argv4i64)
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %V8I64 = call <8 x i64> @llvm.scmp.v8i64.v8i64(<8 x i64> %argv8i64, <8 x i64> %argv8i64)
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 56 for instruction: %V16I64 = call <16 x i64> @llvm.scmp.v16i64.v16i64(<16 x i64> %argv16i64, <16 x i64> %argv16i64)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %I8 = call i8 @llvm.scmp.i8.i8(i8 %arg8, i8 %arg8)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %V16I8 = call <16 x i8> @llvm.scmp.v16i8.v16i8(<16 x i8> %argv16i8, <16 x i8> %argv16i8)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 5 for inst...
[truncated]

``````````

</details>


https://github.com/llvm/llvm-project/pull/106747


More information about the llvm-commits mailing list