Skip to content

[RISCV][NFC] Clarify getRISCVInstructionCost usage #121556

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
Merged
arcbbb merged 2 commits into from
Feb 3, 2025
Merged

[RISCV][NFC] Clarify getRISCVInstructionCost usage #121556

arcbbb merged 2 commits into from
Feb 3, 2025

Conversation

@arcbbb
Copy link
Contributor

@arcbbb arcbbb commented Jan 3, 2025 •
edited
Loading

This patch adds a comment to explicitly state that getRISCVInstructionCost uses vtype associated with widening and narrowing instructions.
For example, vtype = (SEW):
For vfwcvt.f.f.v, the source is (SEW), the destination is (2 * SEW)
For vfncvt.f.f.w, the source is (2 * SEW), the destination is (SEW).
In these cases, the type passed to getRISCVInstructionCost differs

  • The source type is used for vfwcvt.f.f.v.
  • The destination type is used for vfncvt.f.f.w.

@arcbbb arcbbb requested review from lukel97 and topperc January 3, 2025 09:43
@llvmbot llvmbot added backend:RISC-V llvm:analysis Includes value tracking, cost tables and constant folding labels Jan 3, 2025
@llvmbot
Copy link
Member

llvmbot commented Jan 3, 2025 •
edited
Loading

@llvm/pr-subscribers-llvm-analysis

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

Author: Shih-Po Hung (arcbbb)

Changes

This patch addresses

  1. Adds a comment to explicitly state that getRISCVInstructionCost uses vtype associated with widening and narrowing instructions.
    For example, vtype = (SEW):
    For vfwcvt.f.f.v, the source is (SEW), the destination is (2 * SEW)
    For vfncvt.f.f.w, the source is (2 * SEW), the destination is (SEW).
    In these cases, the type passed to getRISCVInstructionCost differs
  • The source type is used for vfwcvt.f.f.v.
  • The destination type is used for vfncvt.f.f.w.
  1. Fixes the MVT handling for ISD::FP_EXTEND.

Patch is 51.32 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/121556.diff

5 Files Affected:

  • (modified) llvm/lib/Target/RISCV/RISCVTargetTransformInfo.cpp (+9-6)
  • (modified) llvm/lib/Target/RISCV/RISCVTargetTransformInfo.h (+5)
  • (modified) llvm/test/Analysis/CostModel/RISCV/arith-fp.ll (+44-44)
  • (modified) llvm/test/Analysis/CostModel/RISCV/cast.ll (+68-68)
  • (modified) llvm/test/Analysis/CostModel/RISCV/reduce-scalable-fp.ll (+6-6)
diff --git a/llvm/lib/Target/RISCV/RISCVTargetTransformInfo.cpp b/llvm/lib/Target/RISCV/RISCVTargetTransformInfo.cpp
index 850d6244affa50..8484ff3696be48 100644
--- a/llvm/lib/Target/RISCV/RISCVTargetTransformInfo.cpp
+++ b/llvm/lib/Target/RISCV/RISCVTargetTransformInfo.cpp
@@ -1301,13 +1301,16 @@ InstructionCost RISCVTTIImpl::getCastInstrCost(unsigned Opcode, Type *Dst,
                                             : RISCV::VFNCVT_F_F_W;
     InstructionCost Cost = 0;
     for (; SrcEltSize != DstEltSize;) {
-      MVT ElementMVT = (ISD == ISD::TRUNCATE)
-                           ? MVT::getIntegerVT(DstEltSize)
+      MVT ElementMVT = (ISD == ISD::TRUNCATE) ? MVT::getIntegerVT(DstEltSize)
+                       : (ISD == ISD::FP_EXTEND)
+                           ? MVT::getFloatingPointVT(SrcEltSize)
                            : MVT::getFloatingPointVT(DstEltSize);
-      MVT DstMVT = DstLT.second.changeVectorElementType(ElementMVT);
-      DstEltSize =
-          (DstEltSize > SrcEltSize) ? DstEltSize >> 1 : DstEltSize << 1;
-      Cost += getRISCVInstructionCost(Op, DstMVT, CostKind);
+      MVT VTypeMVT = DstLT.second.changeVectorElementType(ElementMVT);
+      Cost += getRISCVInstructionCost(Op, VTypeMVT, CostKind);
+      if (DstEltSize > SrcEltSize)
+        SrcEltSize = SrcEltSize << 1;
+      else
+        DstEltSize = DstEltSize << 1;
     }
     return Cost;
   }
diff --git a/llvm/lib/Target/RISCV/RISCVTargetTransformInfo.h b/llvm/lib/Target/RISCV/RISCVTargetTransformInfo.h
index 9b364391f0fa47..3a9073731681ef 100644
--- a/llvm/lib/Target/RISCV/RISCVTargetTransformInfo.h
+++ b/llvm/lib/Target/RISCV/RISCVTargetTransformInfo.h
@@ -48,6 +48,11 @@ class RISCVTTIImpl : public BasicTTIImplBase<RISCVTTIImpl> {
   /// actual target hardware.
   unsigned getEstimatedVLFor(VectorType *Ty);
 
+  /// This function calculates the costs for one or more RVV opcodes based
+  /// on the vtype and the cost kind.
+  /// \param Opcodes A list of opcodes of the RVV instruction to evaluate.
+  /// \param VT The MVT of vtype associated with the RVV instructions.
+  /// \param CostKind The type of cost to compute.
   InstructionCost getRISCVInstructionCost(ArrayRef<unsigned> OpCodes, MVT VT,
                                           TTI::TargetCostKind CostKind);
 
diff --git a/llvm/test/Analysis/CostModel/RISCV/arith-fp.ll b/llvm/test/Analysis/CostModel/RISCV/arith-fp.ll
index d85c9e2a19209c..58b8066a080997 100644
--- a/llvm/test/Analysis/CostModel/RISCV/arith-fp.ll
+++ b/llvm/test/Analysis/CostModel/RISCV/arith-fp.ll
@@ -12,13 +12,13 @@ define void @fadd() {
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %V1BF16 = fadd <1 x bfloat> undef, undef
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %V2BF16 = fadd <2 x bfloat> undef, undef
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %V4BF16 = fadd <4 x bfloat> undef, undef
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %V8BF16 = fadd <8 x bfloat> undef, undef
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %V16BF16 = fadd <16 x bfloat> undef, undef
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %V8BF16 = fadd <8 x bfloat> undef, undef
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V16BF16 = fadd <16 x bfloat> undef, undef
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %NXV1BF16 = fadd <vscale x 1 x bfloat> undef, undef
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %NXV2BF16 = fadd <vscale x 2 x bfloat> undef, undef
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %NXV4BF16 = fadd <vscale x 4 x bfloat> undef, undef
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %NXV8BF16 = fadd <vscale x 8 x bfloat> undef, undef
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %NXV16BF16 = fadd <vscale x 16 x bfloat> undef, undef
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %NXV4BF16 = fadd <vscale x 4 x bfloat> undef, undef
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %NXV8BF16 = fadd <vscale x 8 x bfloat> undef, undef
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %NXV16BF16 = fadd <vscale x 16 x bfloat> undef, undef
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V1F32 = fadd <1 x float> undef, undef
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V2F32 = fadd <2 x float> undef, undef
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V4F32 = fadd <4 x float> undef, undef
@@ -174,14 +174,14 @@ define void @fadd_f16() {
 ; ZVFHMIN-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %V1F16 = fadd <1 x half> undef, undef
 ; ZVFHMIN-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %V2F16 = fadd <2 x half> undef, undef
 ; ZVFHMIN-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %V4F16 = fadd <4 x half> undef, undef
-; ZVFHMIN-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %V8F16 = fadd <8 x half> undef, undef
-; ZVFHMIN-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %V16F16 = fadd <16 x half> undef, undef
-; ZVFHMIN-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %V32F16 = fadd <32 x half> undef, undef
+; ZVFHMIN-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %V8F16 = fadd <8 x half> undef, undef
+; ZVFHMIN-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V16F16 = fadd <16 x half> undef, undef
+; ZVFHMIN-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %V32F16 = fadd <32 x half> undef, undef
 ; ZVFHMIN-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %NXV1F16 = fadd <vscale x 1 x half> undef, undef
 ; ZVFHMIN-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %NXV2F16 = fadd <vscale x 2 x half> undef, undef
-; ZVFHMIN-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %NXV4F16 = fadd <vscale x 4 x half> undef, undef
-; ZVFHMIN-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %NXV8F16 = fadd <vscale x 8 x half> undef, undef
-; ZVFHMIN-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %NXV16F16 = fadd <vscale x 16 x half> undef, undef
+; ZVFHMIN-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %NXV4F16 = fadd <vscale x 4 x half> undef, undef
+; ZVFHMIN-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %NXV8F16 = fadd <vscale x 8 x half> undef, undef
+; ZVFHMIN-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %NXV16F16 = fadd <vscale x 16 x half> undef, undef
 ; ZVFHMIN-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %NXV32F16 = fadd <vscale x 32 x half> undef, undef
 ; ZVFHMIN-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V1F16_VP = call <1 x half> @llvm.vp.fadd.v1f16(<1 x half> undef, <1 x half> undef, <1 x i1> undef, i32 undef)
 ; ZVFHMIN-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F16_VP = call <2 x half> @llvm.vp.fadd.v2f16(<2 x half> undef, <2 x half> undef, <2 x i1> undef, i32 undef)
@@ -234,13 +234,13 @@ define void @fsub() {
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %V1BF16 = fsub <1 x bfloat> undef, undef
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %V2BF16 = fsub <2 x bfloat> undef, undef
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %V4BF16 = fsub <4 x bfloat> undef, undef
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %V8BF16 = fsub <8 x bfloat> undef, undef
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %V16BF16 = fsub <16 x bfloat> undef, undef
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %V8BF16 = fsub <8 x bfloat> undef, undef
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V16BF16 = fsub <16 x bfloat> undef, undef
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %NXV1BF16 = fsub <vscale x 1 x bfloat> undef, undef
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %NXV2BF16 = fsub <vscale x 2 x bfloat> undef, undef
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %NXV4BF16 = fsub <vscale x 4 x bfloat> undef, undef
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %NXV8BF16 = fsub <vscale x 8 x bfloat> undef, undef
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %NXV16BF16 = fsub <vscale x 16 x bfloat> undef, undef
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %NXV4BF16 = fsub <vscale x 4 x bfloat> undef, undef
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %NXV8BF16 = fsub <vscale x 8 x bfloat> undef, undef
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %NXV16BF16 = fsub <vscale x 16 x bfloat> undef, undef
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V1F32 = fsub <1 x float> undef, undef
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V2F32 = fsub <2 x float> undef, undef
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V4F32 = fsub <4 x float> undef, undef
@@ -396,14 +396,14 @@ define void @fsub_f16() {
 ; ZVFHMIN-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %V1F16 = fsub <1 x half> undef, undef
 ; ZVFHMIN-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %V2F16 = fsub <2 x half> undef, undef
 ; ZVFHMIN-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %V4F16 = fsub <4 x half> undef, undef
-; ZVFHMIN-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %V8F16 = fsub <8 x half> undef, undef
-; ZVFHMIN-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %V16F16 = fsub <16 x half> undef, undef
-; ZVFHMIN-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %V32F16 = fsub <32 x half> undef, undef
+; ZVFHMIN-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %V8F16 = fsub <8 x half> undef, undef
+; ZVFHMIN-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V16F16 = fsub <16 x half> undef, undef
+; ZVFHMIN-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %V32F16 = fsub <32 x half> undef, undef
 ; ZVFHMIN-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %NXV1F16 = fsub <vscale x 1 x half> undef, undef
 ; ZVFHMIN-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %NXV2F16 = fsub <vscale x 2 x half> undef, undef
-; ZVFHMIN-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %NXV4F16 = fsub <vscale x 4 x half> undef, undef
-; ZVFHMIN-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %NXV8F16 = fsub <vscale x 8 x half> undef, undef
-; ZVFHMIN-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %NXV16F16 = fsub <vscale x 16 x half> undef, undef
+; ZVFHMIN-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %NXV4F16 = fsub <vscale x 4 x half> undef, undef
+; ZVFHMIN-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %NXV8F16 = fsub <vscale x 8 x half> undef, undef
+; ZVFHMIN-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %NXV16F16 = fsub <vscale x 16 x half> undef, undef
 ; ZVFHMIN-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %NXV32F16 = fsub <vscale x 32 x half> undef, undef
 ; ZVFHMIN-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V1F16_VP = call <1 x half> @llvm.vp.fsub.v1f16(<1 x half> undef, <1 x half> undef, <1 x i1> undef, i32 undef)
 ; ZVFHMIN-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F16_VP = call <2 x half> @llvm.vp.fsub.v2f16(<2 x half> undef, <2 x half> undef, <2 x i1> undef, i32 undef)
@@ -456,13 +456,13 @@ define void @fmul() {
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %V1BF16 = fmul <1 x bfloat> undef, undef
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %V2BF16 = fmul <2 x bfloat> undef, undef
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %V4BF16 = fmul <4 x bfloat> undef, undef
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %V8BF16 = fmul <8 x bfloat> undef, undef
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %V16BF16 = fmul <16 x bfloat> undef, undef
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %V8BF16 = fmul <8 x bfloat> undef, undef
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V16BF16 = fmul <16 x bfloat> undef, undef
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %NXV1BF16 = fmul <vscale x 1 x bfloat> undef, undef
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %NXV2BF16 = fmul <vscale x 2 x bfloat> undef, undef
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %NXV4BF16 = fmul <vscale x 4 x bfloat> undef, undef
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %NXV8BF16 = fmul <vscale x 8 x bfloat> undef, undef
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %NXV16BF16 = fmul <vscale x 16 x bfloat> undef, undef
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %NXV4BF16 = fmul <vscale x 4 x bfloat> undef, undef
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %NXV8BF16 = fmul <vscale x 8 x bfloat> undef, undef
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %NXV16BF16 = fmul <vscale x 16 x bfloat> undef, undef
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V1F32 = fmul <1 x float> undef, undef
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V2F32 = fmul <2 x float> undef, undef
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V4F32 = fmul <4 x float> undef, undef
@@ -618,14 +618,14 @@ define void @fmul_f16() {
 ; ZVFHMIN-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %V1F16 = fmul <1 x half> undef, undef
 ; ZVFHMIN-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %V2F16 = fmul <2 x half> undef, undef
 ; ZVFHMIN-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %V4F16 = fmul <4 x half> undef, undef
-; ZVFHMIN-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %V8F16 = fmul <8 x half> undef, undef
-; ZVFHMIN-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %V16F16 = fmul <16 x half> undef, undef
-; ZVFHMIN-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %V32F16 = fmul <32 x half> undef, undef
+; ZVFHMIN-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %V8F16 = fmul <8 x half> undef, undef
+; ZVFHMIN-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V16F16 = fmul <16 x half> undef, undef
+; ZVFHMIN-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %V32F16 = fmul <32 x half> undef, undef
 ; ZVFHMIN-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %NXV1F16 = fmul <vscale x 1 x half> undef, undef
 ; ZVFHMIN-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %NXV2F16 = fmul <vscale x 2 x half> undef, undef
-; ZVFHMIN-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %NXV4F16 = fmul <vscale x 4 x half> undef, undef
-; ZVFHMIN-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %NXV8F16 = fmul <vscale x 8 x half> undef, undef
-; ZVFHMIN-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %NXV16F16 = fmul <vscale x 16 x half> undef, undef
+; ZVFHMIN-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %NXV4F16 = fmul <vscale x 4 x half> undef, undef
+; ZVFHMIN-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %NXV8F16 = fmul <vscale x 8 x half> undef, undef
+; ZVFHMIN-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %NXV16F16 = fmul <vscale x 16 x half> undef, undef
 ; ZVFHMIN-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %NXV32F16 = fmul <vscale x 32 x half> undef, undef
 ; ZVFHMIN-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V1F16_VP = call <1 x half> @llvm.vp.fmul.v1f16(<1 x half> undef, <1 x half> undef, <1 x i1> undef, i32 undef)
 ; ZVFHMIN-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F16_VP = call <2 x half> @llvm.vp.fmul.v2f16(<2 x half> undef, <2 x half> undef, <2 x i1> undef, i32 undef)
@@ -678,13 +678,13 @@ define void @fdiv() {
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %V1BF16 = fdiv <1 x bfloat> undef, undef
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %V2BF16 = fdiv <2 x bfloat> undef, undef
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %V4BF16 = fdiv <4 x bfloat> undef, undef
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %V8BF16 = fdiv <8 x bfloat> undef, undef
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %V16BF16 = fdiv <16 x bfloat> undef, undef
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %V8BF16 = fdiv <8 x bfloat> undef, undef
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V16BF16 = fdiv <16 x bfloat> undef, undef
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %NXV1BF16 = fdiv <vscale x 1 x bfloat> undef, undef
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %NXV2BF16 = fdiv <vscale x 2 x bfloat> undef, undef
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %NXV4BF16 = fdiv <vscale x 4 x bfloat> undef, undef
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %NXV8BF16 = fdiv <vscale x 8 x bfloat> undef, undef
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %NXV16BF16 = fdiv <vscale x 16 x bfloat> undef, undef
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %NXV4BF16 = fdiv <vscale x 4 x bfloat> undef, undef
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %NXV8BF16 = fdiv <vscale x 8 x bfloat> undef, undef
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %NXV16BF16 = fdiv <vscale x 16 x bfloat> undef, undef
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V1F32 = fdiv <1 x float> undef, undef
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V2F32 = fdiv <2 x float> undef, undef
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V4F32 = fdiv <4 x float> undef, undef
@@ -840,14 +840,14 @@ define void @fdiv_f16() {
 ; ZVFHMIN-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %V1F16 = fdiv <1 x half> undef, undef
 ; ZVFHMIN-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %V2F16 = fdiv <2 x half> undef, undef
 ; ZVFHMIN-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %V4F16 = fdiv <4 x half> undef, undef
-; ZVFHMIN-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %V8F16 = fdiv <8 x half> undef, undef
-; ZVFHMIN-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %V16F16 = fdiv <16 x half> undef, undef
-; ZVFHMIN-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %V32F16 = fdiv <32 x half> undef, undef
+; ZVFHMIN-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %V8F16 = fdiv <8 x half> undef, undef
+; ZVFHMIN-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V16F16 = fdiv <16 x half> undef, undef
+; ZVFHMIN-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %V32F16 = fdiv <32 x half> undef, undef
 ; ZVFHMIN-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %NXV1F16 = fdiv <vscale x 1 x half> undef, undef
 ; ZVFHMIN-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %NXV2F16 =...
[truncated]

arcbbb
arcbbb commented Jan 4, 2025
View reviewed changes
llvm/lib/Target/RISCV/RISCVTargetTransformInfo.cpp Outdated
Comment on lines 1307 to 1313
MVT DstMVT = DstLT.second.changeVectorElementType(ElementMVT);
DstEltSize =
(DstEltSize > SrcEltSize) ? DstEltSize >> 1 : DstEltSize << 1;
Cost += getRISCVInstructionCost(Op, DstMVT, CostKind);
MVT VTypeMVT = DstLT.second.changeVectorElementType(ElementMVT);
Cost += getRISCVInstructionCost(Op, VTypeMVT, CostKind);
if (DstEltSize > SrcEltSize)
SrcEltSize = SrcEltSize << 1;
else
DstEltSize = DstEltSize << 1;
Copy link
Contributor Author

@arcbbb arcbbb Jan 4, 2025

Choose a reason for hiding this comment

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

This doesn't handle lmul8 correctly. I will split it into a separate PR to fix.

@arcbbb arcbbb force-pushed the getcost-with-vtype branch from 1b61788 to 9402fbb Compare January 4, 2025 07:53
@arcbbb arcbbb changed the title [RISCV][TTI] Clarify getRISCVInstructionCost usage and fix MVT for ISD::FP_EXTEND [RISCV][NFC] Clarify getRISCVInstructionCost usage Jan 4, 2025
@lukel97
Copy link
Contributor

lukel97 commented Jan 28, 2025

Is the comment supposed to contain information about widening and narrowing?

@arcbbb
Copy link
Contributor Author

arcbbb commented Feb 3, 2025 •
edited
Loading

Is the comment supposed to contain information about widening and narrowing?

Updated. Thanks!

@arcbbb arcbbb merged commit fc89745 into llvm:main Feb 3, 2025
5 of 8 checks passed
@arcbbb arcbbb deleted the getcost-with-vtype branch February 3, 2025 07:49
Icohedron pushed a commit to Icohedron/llvm-project that referenced this pull request Feb 11, 2025
@arcbbb @Icohedron
[RISCV][NFC] Clarify getRISCVInstructionCost usage (llvm#121556)
ef9e50a 
This patch adds a comment to explicitly state that
getRISCVInstructionCost uses vtype associated with widening and
narrowing instructions.
For example, vtype = (SEW):
For vfwcvt.f.f.v, the source is (SEW), the destination is (2 * SEW)
For vfncvt.f.f.w, the source is (2 * SEW), the destination is (SEW).
In these cases, the type passed to `getRISCVInstructionCost` differs
- The source type is used for `vfwcvt.f.f.v`.
- The destination type is used for `vfncvt.f.f.w`.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

@lukel97 lukel97 lukel97 approved these changes

@topperc topperc Awaiting requested review from topperc

Assignees

No one assigned

Labels

backend:RISC-V 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

@arcbbb @llvmbot @lukel97