100 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 52; RUN: opt -S --passes=complex-deinterleaving %s --mattr=+sve2 | FileCheck %s3 4target datalayout = "e-m:e-i8:8:32-i16:16:32-i64:64-i128:128-n32:64-S128-ni:1-p2:32:8:8:32-ni:2"5target triple = "aarch64-arm-none-linux"6 7; Ensure that a second reduction-like pattern doesn't override the first.8define void @reprocessing_crash() #0 {9; CHECK-LABEL: define void @reprocessing_crash(10; CHECK-SAME: ) #[[ATTR0:[0-9]+]] {11; CHECK-NEXT: [[ENTRY:.*]]:12; CHECK-NEXT: [[TMP0:%.*]] = call <vscale x 4 x double> @llvm.vector.interleave2.nxv4f64(<vscale x 2 x double> zeroinitializer, <vscale x 2 x double> zeroinitializer)13; CHECK-NEXT: br label %[[VECTOR_BODY:.*]]14; CHECK: [[VECTOR_BODY]]:15; CHECK-NEXT: [[TMP1:%.*]] = phi <vscale x 4 x double> [ [[TMP0]], %[[ENTRY]] ], [ [[TMP2:%.*]], %[[VECTOR_BODY]] ]16; CHECK-NEXT: [[TMP2]] = fsub <vscale x 4 x double> [[TMP1]], zeroinitializer17; CHECK-NEXT: br i1 false, label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]]18; CHECK: [[MIDDLE_BLOCK]]:19; CHECK-NEXT: [[TMP3:%.*]] = call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.vector.deinterleave2.nxv4f64(<vscale x 4 x double> [[TMP2]])20; CHECK-NEXT: [[TMP4:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP3]], 021; CHECK-NEXT: [[BIN_RDX:%.*]] = fadd <vscale x 2 x double> [[TMP4]], zeroinitializer22; CHECK-NEXT: [[TMP5:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP3]], 123; CHECK-NEXT: [[BIN_RDX23:%.*]] = fadd <vscale x 2 x double> [[TMP5]], zeroinitializer24; CHECK-NEXT: ret void25;26entry:27 br label %vector.body28 29vector.body: ; preds = %vector.body, %entry30 %vec.phi18 = phi <vscale x 2 x double> [ zeroinitializer, %entry ], [ %2, %vector.body ]31 %vec.phi20 = phi <vscale x 2 x double> [ zeroinitializer, %entry ], [ %3, %vector.body ]32 %strided.vec22 = tail call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.vector.deinterleave2.nxv4f64(<vscale x 4 x double> zeroinitializer)33 %0 = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } %strided.vec22, 034 %1 = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } %strided.vec22, 135 %2 = fsub <vscale x 2 x double> %vec.phi18, %036 %3 = fsub <vscale x 2 x double> %vec.phi20, %137 br i1 false, label %middle.block, label %vector.body38 39middle.block: ; preds = %vector.body40 %bin.rdx = fadd <vscale x 2 x double> %2, zeroinitializer41 %bin.rdx23 = fadd <vscale x 2 x double> %3, zeroinitializer42 ret void43}44 45; Make sure we don't crash on floating point single reductions. For now, they46; should be left as-is.47define double @test_fp_single_reduction(i1 %c) #2 {48; CHECK-LABEL: define double @test_fp_single_reduction(49; CHECK-SAME: i1 [[C:%.*]]) #[[ATTR1:[0-9]+]] {50; CHECK-NEXT: [[ENTRY:.*]]:51; CHECK-NEXT: [[TMP0:%.*]] = call <8 x double> @llvm.vector.interleave2.v8f64(<4 x double> zeroinitializer, <4 x double> zeroinitializer)52; CHECK-NEXT: br label %[[VECTOR_BODY:.*]]53; CHECK: [[VECTOR_BODY]]:54; CHECK-NEXT: [[VEC_PHI218:%.*]] = phi <4 x double> [ zeroinitializer, %[[ENTRY]] ], [ [[TMP2:%.*]], %[[VECTOR_BODY]] ]55; CHECK-NEXT: [[TMP1:%.*]] = phi <8 x double> [ [[TMP0]], %[[ENTRY]] ], [ [[TMP3:%.*]], %[[VECTOR_BODY]] ]56; CHECK-NEXT: [[STRIDED_VEC:%.*]] = shufflevector <8 x double> zeroinitializer, <8 x double> zeroinitializer, <4 x i32> <i32 0, i32 2, i32 4, i32 6>57; CHECK-NEXT: [[TMP2]] = fadd <4 x double> [[VEC_PHI218]], [[STRIDED_VEC]]58; CHECK-NEXT: [[TMP3]] = fadd <8 x double> [[TMP1]], zeroinitializer59; CHECK-NEXT: br i1 [[C]], label %[[EXIT:.*]], label %[[VECTOR_BODY]]60; CHECK: [[EXIT]]:61; CHECK-NEXT: [[TMP4:%.*]] = call { <4 x double>, <4 x double> } @llvm.vector.deinterleave2.v8f64(<8 x double> [[TMP3]])62; CHECK-NEXT: [[TMP5:%.*]] = extractvalue { <4 x double>, <4 x double> } [[TMP4]], 063; CHECK-NEXT: [[TMP6:%.*]] = tail call double @llvm.vector.reduce.fadd.v4f64(double 0.000000e+00, <4 x double> [[TMP5]])64; CHECK-NEXT: [[TMP7:%.*]] = tail call double @llvm.vector.reduce.fadd.v4f64(double 0.000000e+00, <4 x double> [[TMP2]])65; CHECK-NEXT: [[TMP8:%.*]] = extractvalue { <4 x double>, <4 x double> } [[TMP4]], 166; CHECK-NEXT: [[TMP9:%.*]] = tail call double @llvm.vector.reduce.fadd.v4f64(double 0.000000e+00, <4 x double> [[TMP8]])67; CHECK-NEXT: [[ADD_1:%.*]] = fadd double [[TMP6]], [[TMP7]]68; CHECK-NEXT: [[ADD_2:%.*]] = fadd double [[ADD_1]], [[TMP9]]69; CHECK-NEXT: ret double [[ADD_2]]70;71entry:72 br label %vector.body73 74vector.body:75 %vec.phi216 = phi <4 x double> [ zeroinitializer, %entry ], [ %2, %vector.body ]76 %vec.phi218 = phi <4 x double> [ zeroinitializer, %entry ], [ %1, %vector.body ]77 %vec.phi222 = phi <4 x double> [ zeroinitializer, %entry ], [ %3, %vector.body ]78 %strided.vec = shufflevector <8 x double> zeroinitializer, <8 x double> zeroinitializer, <4 x i32> <i32 0, i32 2, i32 4, i32 6>79 %strided.vec223 = shufflevector <8 x double> zeroinitializer, <8 x double> zeroinitializer, <4 x i32> <i32 1, i32 3, i32 5, i32 7>80 %1 = fadd <4 x double> %vec.phi218, %strided.vec81 %2 = fadd <4 x double> %vec.phi216, %strided.vec82 %3 = fadd <4 x double> %vec.phi222, %strided.vec22383 br i1 %c, label %exit, label %vector.body84 85exit:86 %4 = tail call double @llvm.vector.reduce.fadd.v4f64(double 0.000000e+00, <4 x double> %2)87 %5 = tail call double @llvm.vector.reduce.fadd.v4f64(double 0.000000e+00, <4 x double> %1)88 %6 = tail call double @llvm.vector.reduce.fadd.v4f64(double 0.000000e+00, <4 x double> %3)89 %add.1 = fadd double %4, %590 %add.2 = fadd double %add.1, %691 ret double %add.292}93 94; Function Attrs: nocallback nofree nosync nounwind willreturn memory(none)95declare { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.vector.deinterleave2.nxv4f64(<vscale x 4 x double>) #196 97attributes #0 = { "target-cpu"="neoverse-v1" }98attributes #1 = { nocallback nofree nosync nounwind willreturn memory(none) }99attributes #2 = { "target-cpu"="apple-m1" }100