73 lines · plain
1; REQUIRES: asserts2; RUN: opt -mtriple=armv8m.main-none-none-eabi -mattr=+dsp < %s -arm-parallel-dsp -S -stats 2>&1 | FileCheck %s3;4; A more complicated chain: 4 mul operations, so we expect 2 smlad calls.5;6; CHECK: %mac1{{\.}}054 = phi i32 [ [[V17:%[0-9]+]], %for.body ], [ 0, %for.body.preheader ]7; CHECK: [[V11:%[0-9]+]] = load i32, ptr %arrayidx, align 28; CHECK: [[V16:%[0-9]+]] = load i32, ptr %arrayidx4, align 29; CHECK: [[V9:%[0-9]+]] = load i32, ptr %arrayidx8, align 210; CHECK: [[ACC:%[0-9]+]] = call i32 @llvm.arm.smlad(i32 [[V9]], i32 [[V11]], i32 %mac1{{\.}}054)11; CHECK: [[V14:%[0-9]+]] = load i32, ptr %arrayidx17, align 212; CHECK: [[V12:%[0-9]+]] = call i32 @llvm.arm.smlad(i32 [[V14]], i32 [[V16]], i32 [[ACC]])13;14; And we don't want to see a 3rd smlad:15; CHECK-NOT: call i32 @llvm.arm.smlad16;17; CHECK: 2 arm-parallel-dsp - Number of smlad instructions generated18;19define dso_local i32 @test(i32 %arg, ptr nocapture readnone %arg1, ptr nocapture readonly %arg2, ptr nocapture readonly %arg3) {20entry:21 %cmp52 = icmp sgt i32 %arg, 022 br i1 %cmp52, label %for.body.preheader, label %for.cond.cleanup23 24for.cond.cleanup:25 %mac1.0.lcssa = phi i32 [ 0, %entry ], [ %add28, %for.body ]26 ret i32 %mac1.0.lcssa27 28for.body.preheader:29 br label %for.body30 31for.body:32 %mac1.054 = phi i32 [ %add28, %for.body ], [ 0, %for.body.preheader ]33 %i.053 = phi i32 [ %add29, %for.body ], [ 0, %for.body.preheader ]34 %arrayidx = getelementptr inbounds i16, ptr %arg3, i32 %i.05335 %0 = load i16, ptr %arrayidx, align 236 %add1 = or disjoint i32 %i.053, 137 %arrayidx2 = getelementptr inbounds i16, ptr %arg3, i32 %add138 %1 = load i16, ptr %arrayidx2, align 239 %add3 = or disjoint i32 %i.053, 240 %arrayidx4 = getelementptr inbounds i16, ptr %arg3, i32 %add341 %2 = load i16, ptr %arrayidx4, align 242 %add5 = or disjoint i32 %i.053, 343 %arrayidx6 = getelementptr inbounds i16, ptr %arg3, i32 %add544 %3 = load i16, ptr %arrayidx6, align 245 %arrayidx8 = getelementptr inbounds i16, ptr %arg2, i32 %i.05346 %4 = load i16, ptr %arrayidx8, align 247 %conv = sext i16 %4 to i3248 %conv9 = sext i16 %0 to i3249 %mul = mul nsw i32 %conv, %conv950 %arrayidx11 = getelementptr inbounds i16, ptr %arg2, i32 %add151 %5 = load i16, ptr %arrayidx11, align 252 %conv12 = sext i16 %5 to i3253 %conv13 = sext i16 %1 to i3254 %mul14 = mul nsw i32 %conv12, %conv1355 %arrayidx17 = getelementptr inbounds i16, ptr %arg2, i32 %add356 %6 = load i16, ptr %arrayidx17, align 257 %conv18 = sext i16 %6 to i3258 %conv19 = sext i16 %2 to i3259 %mul20 = mul nsw i32 %conv18, %conv1960 %arrayidx23 = getelementptr inbounds i16, ptr %arg2, i32 %add561 %7 = load i16, ptr %arrayidx23, align 262 %conv24 = sext i16 %7 to i3263 %conv25 = sext i16 %3 to i3264 %mul26 = mul nsw i32 %conv24, %conv2565 %add15 = add i32 %mul, %mac1.05466 %add21 = add i32 %add15, %mul1467 %add27 = add i32 %add21, %mul2068 %add28 = add i32 %add27, %mul2669 %add29 = add nuw nsw i32 %i.053, 470 %cmp = icmp slt i32 %add29, %arg71 br i1 %cmp, label %for.body, label %for.cond.cleanup72}73