61 lines · plain
1; REQUIRES: asserts2; RUN: opt -mtriple=aarch64 -mattr=+sve \3; RUN: -force-target-instruction-cost=1 -passes=loop-vectorize -S -debug-only=loop-vectorize --disable-output < %s 2>&1 \4; RUN: | FileCheck %s --check-prefixes=VSCALEFORTUNING15 6; RUN: opt -mtriple=aarch64 -mattr=+sve -mcpu=generic \7; RUN: -force-target-instruction-cost=1 -passes=loop-vectorize -S -debug-only=loop-vectorize --disable-output < %s 2>&1 \8; RUN: | FileCheck %s --check-prefixes=VSCALEFORTUNING19 10; RUN: opt -mtriple=aarch64 -mattr=+sve -mcpu=generic -sve-vscale-for-tuning=2 \11; RUN: -force-target-instruction-cost=1 -passes=loop-vectorize -S -debug-only=loop-vectorize --disable-output < %s 2>&1 \12; RUN: | FileCheck %s --check-prefixes=VSCALEFORTUNING213 14; RUN: opt -mtriple=aarch64 -mcpu=neoverse-v1 \15; RUN: -force-target-instruction-cost=1 -passes=loop-vectorize -S -debug-only=loop-vectorize --disable-output < %s 2>&1 \16; RUN: | FileCheck %s --check-prefixes=VSCALEFORTUNING217 18; RUN: opt -mtriple=aarch64 -mcpu=neoverse-n2 \19; RUN: -force-target-instruction-cost=1 -passes=loop-vectorize -S -debug-only=loop-vectorize --disable-output < %s 2>&1 \20; RUN: | FileCheck %s --check-prefixes=VSCALEFORTUNING121 22; RUN: opt -mtriple=aarch64 -mcpu=neoverse-v2 \23; RUN: -force-target-instruction-cost=1 -passes=loop-vectorize -S -debug-only=loop-vectorize --disable-output < %s 2>&1 \24; RUN: | FileCheck %s --check-prefixes=NEOVERSEV225 26; VSCALEFORTUNING1: Cost for VF vscale x 2: 11 (Estimated cost per lane: 5.27; VSCALEFORTUNING1: Cost for VF vscale x 4: 11 (Estimated cost per lane: 2.28; VSCALEFORTUNING1: LV: Selecting VF: vscale x 1629 30; VSCALEFORTUNING2: Cost for VF vscale x 2: 11 (Estimated cost per lane: 2.31; VSCALEFORTUNING2: Cost for VF vscale x 4: 11 (Estimated cost per lane: 1.32; VSCALEFORTUNING2: LV: Selecting VF: vscale x 1633 34; NEOVERSEV2: Cost for VF vscale x 2: 11 (Estimated cost per lane: 5.35; NEOVERSEV2: Cost for VF vscale x 4: 11 (Estimated cost per lane: 2.36; NEOVERSEV2: LV: Selecting VF: 1637 38; VSCALEFORTUNING1: <vscale x 16 x i8>39; VSCALEFORTUNING2: <vscale x 16 x i8>40; NEOVERSEV2: <16 x i8>41define void @test0(ptr %a, ptr %b, ptr %c) #0 {42entry:43 br label %loop44 45loop:46 %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop ]47 %arrayidx = getelementptr inbounds i8, ptr %c, i64 %iv48 %0 = load i8, ptr %arrayidx, align 449 %arrayidx2 = getelementptr inbounds i8, ptr %b, i64 %iv50 %1 = load i8, ptr %arrayidx2, align 451 %add = add nsw i8 %0, %152 %arrayidx5 = getelementptr inbounds i8, ptr %a, i64 %iv53 store i8 %add, ptr %arrayidx5, align 454 %iv.next = add nuw nsw i64 %iv, 155 %exitcond.not = icmp eq i64 %iv.next, 102456 br i1 %exitcond.not, label %exit, label %loop57 58exit:59 ret void60}61