1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 4 2; RUN: opt %s -S -riscv-codegenprepare -mtriple=riscv64 -mattr=+v | FileCheck %s 3 4declare i64 @llvm.vscale.i64() 5declare float @llvm.vector.reduce.fadd.nxv4f32(float, <vscale x 4 x float>) 6 7define float @reduce_fadd(ptr %f) { 8; CHECK-LABEL: define float @reduce_fadd( 9; CHECK-SAME: ptr [[F:%.*]]) #[[ATTR2:[0-9]+]] { 10; CHECK-NEXT: entry: 11; CHECK-NEXT: [[VSCALE:%.*]] = tail call i64 @llvm.vscale.i64() 12; CHECK-NEXT: [[VECSIZE:%.*]] = shl nuw nsw i64 [[VSCALE]], 2 13; CHECK-NEXT: br label [[VECTOR_BODY:%.*]] 14; CHECK: vector.body: 15; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[ENTRY:%.*]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] 16; CHECK-NEXT: [[TMP0:%.*]] = phi <vscale x 4 x float> [ insertelement (<vscale x 4 x float> poison, float 0.000000e+00, i64 0), [[ENTRY]] ], [ [[TMP2:%.*]], [[VECTOR_BODY]] ] 17; CHECK-NEXT: [[GEP:%.*]] = getelementptr inbounds float, ptr [[F]], i64 [[INDEX]] 18; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <vscale x 4 x float>, ptr [[GEP]], align 4 19; CHECK-NEXT: [[TMP1:%.*]] = extractelement <vscale x 4 x float> [[TMP0]], i64 0 20; CHECK-NEXT: [[ACC:%.*]] = tail call float @llvm.vector.reduce.fadd.nxv4f32(float [[TMP1]], <vscale x 4 x float> [[WIDE_LOAD]]) 21; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[VECSIZE]] 22; CHECK-NEXT: [[DONE:%.*]] = icmp eq i64 [[INDEX_NEXT]], 1024 23; CHECK-NEXT: [[TMP2]] = insertelement <vscale x 4 x float> poison, float [[ACC]], i64 0 24; CHECK-NEXT: br i1 [[DONE]], label [[EXIT:%.*]], label [[VECTOR_BODY]] 25; CHECK: exit: 26; CHECK-NEXT: ret float [[ACC]] 27; 28 29entry: 30 %vscale = tail call i64 @llvm.vscale.i64() 31 %vecsize = shl nuw nsw i64 %vscale, 2 32 br label %vector.body 33 34vector.body: 35 %index = phi i64 [ 0, %entry ], [ %index.next, %vector.body ] 36 %vec.phi = phi float [ 0.000000e+00, %entry ], [ %acc, %vector.body ] 37 %gep = getelementptr inbounds float, ptr %f, i64 %index 38 %wide.load = load <vscale x 4 x float>, ptr %gep, align 4 39 %acc = tail call float @llvm.vector.reduce.fadd.nxv4f32(float %vec.phi, <vscale x 4 x float> %wide.load) 40 %index.next = add nuw i64 %index, %vecsize 41 %done = icmp eq i64 %index.next, 1024 42 br i1 %done, label %exit, label %vector.body 43 44exit: 45 ret float %acc 46} 47