41 lines · plain
1; RUN: llc -mtriple=hexagon -mcpu=hexagonv5 -enable-pipeliner < %s -pipeliner-experimental-cg=true | FileCheck %s2; RUN: llc -mtriple=hexagon -mcpu=hexagonv5 -O2 < %s -pipeliner-experimental-cg=true | FileCheck %s3; RUN: llc -mtriple=hexagon -mcpu=hexagonv5 -O3 < %s -pipeliner-experimental-cg=true | FileCheck %s4; RUN: llc -mtriple=hexagon -mcpu=hexagonv5 -enable-pipeliner < %s -pipeliner-experimental-cg=true -early-live-intervals -verify-machineinstrs | FileCheck %s5;6; Check that we pipeline a vectorized dot product in a single packet.7;8; CHECK: {9; CHECK: += mpyi10; CHECK: += mpyi11; CHECK: memd12; CHECK: memd13; CHECK: } :endloop014 15@a = common global [5000 x i32] zeroinitializer, align 816@b = common global [5000 x i32] zeroinitializer, align 817 18define i32 @vecMultGlobal() {19entry:20 br label %polly.loop_body21 22polly.loop_after:23 %0 = extractelement <2 x i32> %addp_vec, i32 024 %1 = extractelement <2 x i32> %addp_vec, i32 125 %add_sum = add i32 %0, %126 ret i32 %add_sum27 28polly.loop_body:29 %polly.loopiv13 = phi i32 [ 0, %entry ], [ %polly.next_loopiv, %polly.loop_body ]30 %reduction.012 = phi <2 x i32> [ zeroinitializer, %entry ], [ %addp_vec, %polly.loop_body ]31 %polly.next_loopiv = add nsw i32 %polly.loopiv13, 232 %p_arrayidx1 = getelementptr [5000 x i32], ptr @b, i32 0, i32 %polly.loopiv1333 %p_arrayidx = getelementptr [5000 x i32], ptr @a, i32 0, i32 %polly.loopiv1334 %_p_vec_full = load <2 x i32>, ptr %p_arrayidx1, align 835 %_p_vec_full8 = load <2 x i32>, ptr %p_arrayidx, align 836 %mulp_vec = mul <2 x i32> %_p_vec_full8, %_p_vec_full37 %addp_vec = add <2 x i32> %mulp_vec, %reduction.01238 %2 = icmp slt i32 %polly.next_loopiv, 500039 br i1 %2, label %polly.loop_body, label %polly.loop_after40}41