133 lines · plain
1; __arm_streaming function, no special flags. Should by default not vectorize.2; RUN: sed -e s/REPLACE_PSTATE_MACRO/aarch64_pstate_sm_enabled/ %s | opt -mattr="+sme" -passes=loop-vectorize,slp-vectorizer -S - | FileCheck %s --check-prefix=CHECK3; RUN: sed -e s/REPLACE_PSTATE_MACRO/aarch64_pstate_sm_enabled/ %s | opt -mattr="+sme,+sve" -passes=loop-vectorize,slp-vectorizer -S - | FileCheck %s --check-prefix=CHECK4; RUN: sed -e s/REPLACE_PSTATE_MACRO/aarch64_pstate_sm_body/ %s | opt -mattr="+sme" -passes=loop-vectorize,slp-vectorizer -S - | FileCheck %s --check-prefix=CHECK5; RUN: sed -e s/REPLACE_PSTATE_MACRO/aarch64_pstate_sm_body/ %s | opt -mattr="+sme,+sve" -passes=loop-vectorize,slp-vectorizer -S - | FileCheck %s --check-prefix=CHECK6 7; __arm_streaming function, force use of scalable autovec. Should vectorize.8; RUN: sed -e s/REPLACE_PSTATE_MACRO/aarch64_pstate_sm_enabled/ %s | opt -mattr="+sme" -passes=loop-vectorize,slp-vectorizer -S -enable-scalable-autovec-in-streaming-mode - | FileCheck %s --check-prefix=CHECK-FORCE-SCALABLE9; RUN: sed -e s/REPLACE_PSTATE_MACRO/aarch64_pstate_sm_enabled/ %s | opt -mattr="+sme,+sve" -passes=loop-vectorize,slp-vectorizer -S -enable-scalable-autovec-in-streaming-mode - | FileCheck %s --check-prefix=CHECK-FORCE-SCALABLE10; RUN: sed -e s/REPLACE_PSTATE_MACRO/aarch64_pstate_sm_body/ %s | opt -mattr="+sme" -passes=loop-vectorize,slp-vectorizer -S -enable-scalable-autovec-in-streaming-mode - | FileCheck %s --check-prefix=CHECK-FORCE-SCALABLE11; RUN: sed -e s/REPLACE_PSTATE_MACRO/aarch64_pstate_sm_body/ %s | opt -mattr="+sme,+sve" -passes=loop-vectorize,slp-vectorizer -S -enable-scalable-autovec-in-streaming-mode - | FileCheck %s --check-prefix=CHECK-FORCE-SCALABLE12 13; __arm_streaming function, force use of fixed-width autovec. Should vectorize.14; RUN: sed -e s/REPLACE_PSTATE_MACRO/aarch64_pstate_sm_enabled/ %s | opt -mattr="+sme" -passes=loop-vectorize,slp-vectorizer -S -enable-fixedwidth-autovec-in-streaming-mode - | FileCheck %s --check-prefix=CHECK-FORCE-FIXEDWIDTH15; RUN: sed -e s/REPLACE_PSTATE_MACRO/aarch64_pstate_sm_enabled/ %s | opt -mattr="+sme,+sve" -passes=loop-vectorize,slp-vectorizer -S -enable-fixedwidth-autovec-in-streaming-mode - | FileCheck %s --check-prefix=CHECK-FORCE-FIXEDWIDTH16; RUN: sed -e s/REPLACE_PSTATE_MACRO/aarch64_pstate_sm_body/ %s | opt -mattr="+sme" -passes=loop-vectorize,slp-vectorizer -S -enable-fixedwidth-autovec-in-streaming-mode - | FileCheck %s --check-prefix=CHECK-FORCE-FIXEDWIDTH17; RUN: sed -e s/REPLACE_PSTATE_MACRO/aarch64_pstate_sm_body/ %s | opt -mattr="+sme,+sve" -passes=loop-vectorize,slp-vectorizer -S -enable-fixedwidth-autovec-in-streaming-mode - | FileCheck %s --check-prefix=CHECK-FORCE-FIXEDWIDTH18 19; __arm_streaming_compatible function, no special flags. Should by default not vectorize.20; RUN: sed -e s/REPLACE_PSTATE_MACRO/aarch64_pstate_sm_compatible/ %s | opt -mattr="+sme" -passes=loop-vectorize,slp-vectorizer -S - | FileCheck %s --check-prefix=CHECK21; RUN: sed -e s/REPLACE_PSTATE_MACRO/aarch64_pstate_sm_compatible/ %s | opt -mattr="+sme,+sve" -passes=loop-vectorize,slp-vectorizer -S - | FileCheck %s --check-prefix=CHECK22 23; __arm_streaming_compatible function, force use of scalable autovec. Can only vectorize if +sve is available.24; RUN: sed -e s/REPLACE_PSTATE_MACRO/aarch64_pstate_sm_compatible/ %s | opt -mattr="+sme" -passes=loop-vectorize,slp-vectorizer -S -enable-scalable-autovec-in-streaming-mode - | FileCheck %s25; RUN: sed -e s/REPLACE_PSTATE_MACRO/aarch64_pstate_sm_compatible/ %s | opt -mattr="+sme,+sve" -passes=loop-vectorize,slp-vectorizer -S -enable-scalable-autovec-in-streaming-mode - | FileCheck %s --check-prefix=CHECK-FORCE-SCALABLE26 27; __arm_streaming_compatible function, force use of fixed-width autovec. Can only vectorize if +sve is available.28; RUN: sed -e s/REPLACE_PSTATE_MACRO/aarch64_pstate_sm_compatible/ %s | opt -mattr="+sme" -passes=loop-vectorize,slp-vectorizer -S -enable-fixedwidth-autovec-in-streaming-mode - | FileCheck %s29; RUN: sed -e s/REPLACE_PSTATE_MACRO/aarch64_pstate_sm_compatible/ %s | opt -mattr="+sme,+sve" -passes=loop-vectorize,slp-vectorizer -S -enable-fixedwidth-autovec-in-streaming-mode - | FileCheck %s --check-prefix=CHECK-FORCE-FIXEDWIDTH30 31target triple = "aarch64-unknown-linux-gnu"32 33attributes #0 = { vscale_range(1,16) "REPLACE_PSTATE_MACRO" }34 35define void @test_fixedwidth_loopvec(ptr noalias %dst, ptr readonly %src, i32 %N) #0 {36; CHECK-LABEL: @test_fixedwidth_loopvec37; CHECK-NOT: <{{[1-9]+}} x i32>38; CHECK-FORCE-FIXEDWIDTH-LABEL: @test_fixedwidth_loopvec39; CHECK-FORCE-FIXEDWIDTH: <{{[1-9]+}} x i32>40entry:41 %cmp6 = icmp sgt i32 %N, 042 br i1 %cmp6, label %for.body.preheader, label %for.cond.cleanup43 44for.body.preheader: ; preds = %entry45 %wide.trip.count = zext i32 %N to i6446 br label %for.body47 48for.cond.cleanup: ; preds = %for.body, %entry49 ret void50 51for.body: ; preds = %for.body.preheader, %for.body52 %indvars.iv = phi i64 [ 0, %for.body.preheader ], [ %indvars.iv.next, %for.body ]53 %arrayidx = getelementptr inbounds i32, ptr %src, i64 %indvars.iv54 %0 = load i32, ptr %arrayidx, align 455 %add = add nsw i32 %0, 4256 %arrayidx2 = getelementptr inbounds i32, ptr %dst, i64 %indvars.iv57 store i32 %add, ptr %arrayidx2, align 458 %indvars.iv.next = add nuw nsw i64 %indvars.iv, 159 %exitcond.not = icmp eq i64 %indvars.iv.next, %wide.trip.count60 br i1 %exitcond.not, label %for.cond.cleanup, label %for.body, !llvm.loop !061}62 63!0 = distinct !{!0, !1, !2, !3}64!1 = !{!"llvm.loop.mustprogress"}65!2 = !{!"llvm.loop.interleave.count", i32 1}66!3 = !{!"llvm.loop.vectorize.scalable.enable", i1 false}67 68define void @test_scalable_loopvec(ptr noalias %dst, ptr readonly %src, i32 %N) #0 {69; CHECK-LABEL: @test_scalable_loopvec70; CHECK-NOT: <vscale x {{[1-9]+}} x i32>71; CHECK-FORCE-SCALABLE-LABEL: @test_fixedwidth_loopvec72; CHECK-FORCE-SCALABLE-LABEL: <vscale x {{[1-9]+}} x i32>73entry:74 %cmp6 = icmp sgt i32 %N, 075 br i1 %cmp6, label %for.body.preheader, label %for.cond.cleanup76 77for.body.preheader: ; preds = %entry78 %wide.trip.count = zext i32 %N to i6479 br label %for.body80 81for.cond.cleanup: ; preds = %for.body, %entry82 ret void83 84for.body: ; preds = %for.body.preheader, %for.body85 %indvars.iv = phi i64 [ 0, %for.body.preheader ], [ %indvars.iv.next, %for.body ]86 %arrayidx = getelementptr inbounds i32, ptr %src, i64 %indvars.iv87 %0 = load i32, ptr %arrayidx, align 488 %add = add nsw i32 %0, 4289 %arrayidx2 = getelementptr inbounds i32, ptr %dst, i64 %indvars.iv90 store i32 %add, ptr %arrayidx2, align 491 %indvars.iv.next = add nuw nsw i64 %indvars.iv, 192 %exitcond.not = icmp eq i64 %indvars.iv.next, %wide.trip.count93 br i1 %exitcond.not, label %for.cond.cleanup, label %for.body, !llvm.loop !494}95 96!4 = distinct !{!4, !5, !6, !7}97!5 = !{!"llvm.loop.mustprogress"}98!6 = !{!"llvm.loop.interleave.count", i32 1}99!7 = !{!"llvm.loop.vectorize.scalable.enable", i1 true}100 101define void @test_slp(ptr noalias %dst, ptr readonly %src, i32 %N) #0 {102; CHECK-LABEL: @test_slp103; CHECK-NOT: <{{[1-9]+}} x i32>104; CHECK-FORCE-FIXEDWIDTH-LABEL: @test_slp105; CHECK-FORCE-FIXEDWIDTH: <{{[1-9]+}} x i32>106entry:107 %cmp6 = icmp sgt i32 %N, 0108 br i1 %cmp6, label %for.body.preheader, label %for.cond.cleanup109 110for.body.preheader: ; preds = %entry111 %wide.trip.count = zext i32 %N to i64112 br label %for.body113 114for.cond.cleanup: ; preds = %for.body, %entry115 ret void116 117for.body: ; preds = %for.body.preheader, %for.body118 %indvars.iv = phi i64 [ 0, %for.body.preheader ], [ %indvars.iv.next, %for.body ]119 %arrayidx = getelementptr inbounds i32, ptr %src, i64 %indvars.iv120 %0 = load i32, ptr %arrayidx, align 4121 %add = add nsw i32 %0, 42122 %arrayidx2 = getelementptr inbounds i32, ptr %dst, i64 %indvars.iv123 store i32 %add, ptr %arrayidx2, align 4124 %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1125 %exitcond.not = icmp eq i64 %indvars.iv.next, %wide.trip.count126 br i1 %exitcond.not, label %for.cond.cleanup, label %for.body, !llvm.loop !8127}128 129!8 = distinct !{!8, !9, !10, !11}130!9 = !{!"llvm.loop.mustprogress"}131!10 = !{!"llvm.loop.interleave.count", i32 4}132!11 = !{!"llvm.loop.vectorize.width", i32 1}133