From f3408739da33c6ead2fe0c32efd9d70ed39cb334 Mon Sep 17 00:00:00 2001 From: liqinweng Date: Mon, 9 Jan 2023 11:38:24 +0800 Subject: [PATCH] [RISCV][CostModel] Add cost model for integer abs Reviewed By: craig.topper Differential Revision: https://reviews.llvm.org/D132999 --- llvm/lib/Target/RISCV/RISCVTargetTransformInfo.cpp | 9 ++ llvm/test/Analysis/CostModel/RISCV/abs.ll | 117 +++++++++++++++++++++ 2 files changed, 126 insertions(+) create mode 100644 llvm/test/Analysis/CostModel/RISCV/abs.ll diff --git a/llvm/lib/Target/RISCV/RISCVTargetTransformInfo.cpp b/llvm/lib/Target/RISCV/RISCVTargetTransformInfo.cpp index ed8af259..88f4310 100644 --- a/llvm/lib/Target/RISCV/RISCVTargetTransformInfo.cpp +++ b/llvm/lib/Target/RISCV/RISCVTargetTransformInfo.cpp @@ -880,6 +880,15 @@ RISCVTTIImpl::getIntrinsicInstrCost(const IntrinsicCostAttributes &ICA, return LT.first; break; } + case Intrinsic::abs: { + auto LT = getTypeLegalizationCost(RetTy); + if (ST->hasVInstructions() && LT.second.isVector()) { + // vrsub.vi v10, v8, 0 + // vmax.vv v8, v8, v10 + return LT.first * 2; + } + break; + } // TODO: add more intrinsic case Intrinsic::experimental_stepvector: { unsigned Cost = 1; // vid diff --git a/llvm/test/Analysis/CostModel/RISCV/abs.ll b/llvm/test/Analysis/CostModel/RISCV/abs.ll new file mode 100644 index 0000000..1929a631 --- /dev/null +++ b/llvm/test/Analysis/CostModel/RISCV/abs.ll @@ -0,0 +1,117 @@ +; NOTE: Assertions have been autogenerated by utils/update_analyze_test_checks.py +; RUN: opt < %s -passes="print" 2>&1 -disable-output -S -mtriple=riscv64 -mattr=+v -riscv-v-vector-bits-min=-1 | FileCheck %s +; Check that we don't crash querying costs when vectors are not enabled. +; RUN: opt -passes="print" 2>&1 -disable-output -mtriple=riscv64 + +declare <2 x i64> @llvm.abs.v2i64(<2 x i64>, i1) +declare <4 x i64> @llvm.abs.v4i64(<4 x i64>, i1) +declare <8 x i64> @llvm.abs.v8i64(<8 x i64>, i1) +declare @llvm.abs.nxv2i64(, i1) +declare @llvm.abs.nxv4i64(, i1) +declare @llvm.abs.nxv8i64(, i1) + +declare <2 x i32> @llvm.abs.v2i32(<2 x i32>, i1) +declare <4 x i32> @llvm.abs.v4i32(<4 x i32>, i1) +declare <8 x i32> @llvm.abs.v8i32(<8 x i32>, i1) +declare <16 x i32> @llvm.abs.v16i32(<16 x i32>, i1) +declare @llvm.abs.nxv2i32(, i1) +declare @llvm.abs.nxv4i32(, i1) +declare @llvm.abs.nxv8i32(, i1) +declare @llvm.abs.nxv16i32(, i1) + +declare <2 x i16> @llvm.abs.v2i16(<2 x i16>, i1) +declare <4 x i16> @llvm.abs.v4i16(<4 x i16>, i1) +declare <8 x i16> @llvm.abs.v8i16(<8 x i16>, i1) +declare <16 x i16> @llvm.abs.v16i16(<16 x i16>, i1) +declare <32 x i16> @llvm.abs.v32i16(<32 x i16>, i1) +declare @llvm.abs.nxv2i16(, i1) +declare @llvm.abs.nxv4i16(, i1) +declare @llvm.abs.nxv8i16(, i1) +declare @llvm.abs.nxv16i16(, i1) +declare @llvm.abs.nxv32i16(, i1) + +declare <2 x i8> @llvm.abs.v2i8(<2 x i8>, i1) +declare <4 x i8> @llvm.abs.v4i8(<4 x i8>, i1) +declare <8 x i8> @llvm.abs.v8i8(<8 x i8>, i1) +declare <16 x i8> @llvm.abs.v16i8(<16 x i8>, i1) +declare <32 x i8> @llvm.abs.v32i8(<32 x i8>, i1) +declare <64 x i8> @llvm.abs.v64i8(<64 x i8>, i1) +declare @llvm.abs.nxv8i8(, i1) +declare @llvm.abs.nxv16i8(, i1) +declare @llvm.abs.nxv32i8(, i1) +declare @llvm.abs.nxv64i8(, i1) + +define i32 @abs(i32 %arg) { +; CHECK-LABEL: 'abs' +; CHECK-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %1 = call <2 x i64> @llvm.abs.v2i64(<2 x i64> undef, i1 false) +; CHECK-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %2 = call <4 x i64> @llvm.abs.v4i64(<4 x i64> undef, i1 false) +; CHECK-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %3 = call <8 x i64> @llvm.abs.v8i64(<8 x i64> undef, i1 false) +; CHECK-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %4 = call @llvm.abs.nxv2i64( undef, i1 false) +; CHECK-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %5 = call @llvm.abs.nxv4i64( undef, i1 false) +; CHECK-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %6 = call @llvm.abs.nxv8i64( undef, i1 false) +; CHECK-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %7 = call <2 x i32> @llvm.abs.v2i32(<2 x i32> undef, i1 false) +; CHECK-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %8 = call <4 x i32> @llvm.abs.v4i32(<4 x i32> undef, i1 false) +; CHECK-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %9 = call <8 x i32> @llvm.abs.v8i32(<8 x i32> undef, i1 false) +; CHECK-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %10 = call <16 x i32> @llvm.abs.v16i32(<16 x i32> undef, i1 false) +; CHECK-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %11 = call @llvm.abs.nxv2i32( undef, i1 false) +; CHECK-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %12 = call @llvm.abs.nxv4i32( undef, i1 false) +; CHECK-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %13 = call @llvm.abs.nxv8i32( undef, i1 false) +; CHECK-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %14 = call @llvm.abs.nxv16i32( undef, i1 false) +; CHECK-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %15 = call <2 x i16> @llvm.abs.v2i16(<2 x i16> undef, i1 false) +; CHECK-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %16 = call <4 x i16> @llvm.abs.v4i16(<4 x i16> undef, i1 false) +; CHECK-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %17 = call <8 x i16> @llvm.abs.v8i16(<8 x i16> undef, i1 false) +; CHECK-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %18 = call <16 x i16> @llvm.abs.v16i16(<16 x i16> undef, i1 false) +; CHECK-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %19 = call <32 x i16> @llvm.abs.v32i16(<32 x i16> undef, i1 false) +; CHECK-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %20 = call @llvm.abs.nxv2i16( undef, i1 false) +; CHECK-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %21 = call @llvm.abs.nxv4i16( undef, i1 false) +; CHECK-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %22 = call @llvm.abs.nxv8i16( undef, i1 false) +; CHECK-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %23 = call @llvm.abs.nxv16i16( undef, i1 false) +; CHECK-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %24 = call @llvm.abs.nxv32i16( undef, i1 false) +; CHECK-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %25 = call <8 x i8> @llvm.abs.v8i8(<8 x i8> undef, i1 false) +; CHECK-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %26 = call <16 x i8> @llvm.abs.v16i8(<16 x i8> undef, i1 false) +; CHECK-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %27 = call <32 x i8> @llvm.abs.v32i8(<32 x i8> undef, i1 false) +; CHECK-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %28 = call <64 x i8> @llvm.abs.v64i8(<64 x i8> undef, i1 false) +; CHECK-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %29 = call @llvm.abs.nxv8i8( undef, i1 false) +; CHECK-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %30 = call @llvm.abs.nxv16i8( undef, i1 false) +; CHECK-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %31 = call @llvm.abs.nxv32i8( undef, i1 false) +; CHECK-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %32 = call @llvm.abs.nxv64i8( undef, i1 false) +; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef +; + call <2 x i64> @llvm.abs.v2i64(<2 x i64> undef, i1 false) + call <4 x i64> @llvm.abs.v4i64(<4 x i64> undef, i1 false) + call <8 x i64> @llvm.abs.v8i64(<8 x i64> undef, i1 false) + call @llvm.abs.nxv2i64( undef, i1 false) + call @llvm.abs.nxv4i64( undef, i1 false) + call @llvm.abs.nxv8i64( undef, i1 false) + + call <2 x i32> @llvm.abs.v2i32(<2 x i32> undef, i1 false) + call <4 x i32> @llvm.abs.v4i32(<4 x i32> undef, i1 false) + call <8 x i32> @llvm.abs.v8i32(<8 x i32> undef, i1 false) + call <16 x i32> @llvm.abs.v16i32(<16 x i32> undef, i1 false) + call @llvm.abs.nxv2i32( undef, i1 false) + call @llvm.abs.nxv4i32( undef, i1 false) + call @llvm.abs.nxv8i32( undef, i1 false) + call @llvm.abs.nxv16i32( undef, i1 false) + + call <2 x i16> @llvm.abs.v2i16(<2 x i16> undef, i1 false) + call <4 x i16> @llvm.abs.v4i16(<4 x i16> undef, i1 false) + call <8 x i16> @llvm.abs.v8i16(<8 x i16> undef, i1 false) + call <16 x i16> @llvm.abs.v16i16(<16 x i16> undef, i1 false) + call <32 x i16> @llvm.abs.v32i16(<32 x i16> undef, i1 false) + call @llvm.abs.nxv2i16( undef, i1 false) + call @llvm.abs.nxv4i16( undef, i1 false) + call @llvm.abs.nxv8i16( undef, i1 false) + call @llvm.abs.nxv16i16( undef, i1 false) + call @llvm.abs.nxv32i16( undef, i1 false) + + call <8 x i8> @llvm.abs.v8i8(<8 x i8> undef, i1 false) + call <16 x i8> @llvm.abs.v16i8(<16 x i8> undef, i1 false) + call <32 x i8> @llvm.abs.v32i8(<32 x i8> undef, i1 false) + call <64 x i8> @llvm.abs.v64i8(<64 x i8> undef, i1 false) + call @llvm.abs.nxv8i8( undef, i1 false) + call @llvm.abs.nxv16i8( undef, i1 false) + call @llvm.abs.nxv32i8( undef, i1 false) + call @llvm.abs.nxv64i8( undef, i1 false) + + ret i32 undef +} \ No newline at end of file -- 2.7.4