51 lines · plain
1; RUN: opt -mtriple=amdgcn-amd-amdhsa -passes=load-store-vectorizer -S -o - %s | FileCheck %s2; RUN: opt -mtriple=amdgcn-amd-amdhsa -aa-pipeline=basic-aa -passes='function(load-store-vectorizer)' -S -o - %s | FileCheck %s3 4declare i64 @_Z12get_local_idj(i32)5 6declare i64 @_Z12get_group_idj(i32)7 8declare double @llvm.fmuladd.f64(double, double, double)9 10; CHECK-LABEL: @factorizedVsNonfactorizedAccess(11; CHECK: load <2 x float>12; CHECK: store <2 x float>13define amdgpu_kernel void @factorizedVsNonfactorizedAccess(ptr addrspace(1) nocapture %c) {14entry:15 %call = tail call i64 @_Z12get_local_idj(i32 0)16 %call1 = tail call i64 @_Z12get_group_idj(i32 0)17 %div = lshr i64 %call, 418 %div2 = lshr i64 %call1, 319 %mul = shl i64 %div2, 720 %rem = shl i64 %call, 321 %mul3 = and i64 %rem, 12022 %add = or i64 %mul, %mul323 %rem4 = shl i64 %call1, 724 %mul5 = and i64 %rem4, 89625 %mul6 = shl nuw nsw i64 %div, 326 %add7 = add nuw i64 %mul5, %mul627 %mul9 = shl i64 %add7, 1028 %add10 = add i64 %mul9, %add29 %arrayidx = getelementptr inbounds float, ptr addrspace(1) %c, i64 %add1030 %load1 = load float, ptr addrspace(1) %arrayidx, align 431 %conv = fpext float %load1 to double32 %mul11 = fmul double %conv, 0x3FEAB481D8F3550633 %conv12 = fptrunc double %mul11 to float34 %conv18 = fpext float %conv12 to double35 %storeval1 = tail call double @llvm.fmuladd.f64(double 0x3FF4FFAFBBEC946A, double 0.000000e+00, double %conv18)36 %cstoreval1 = fptrunc double %storeval1 to float37 store float %cstoreval1, ptr addrspace(1) %arrayidx, align 438 39 %add23 = or disjoint i64 %add10, 140 %arrayidx24 = getelementptr inbounds float, ptr addrspace(1) %c, i64 %add2341 %load2 = load float, ptr addrspace(1) %arrayidx24, align 442 %conv25 = fpext float %load2 to double43 %mul26 = fmul double %conv25, 0x3FEAB481D8F3550644 %conv27 = fptrunc double %mul26 to float45 %conv34 = fpext float %conv27 to double46 %storeval2 = tail call double @llvm.fmuladd.f64(double 0x3FF4FFAFBBEC946A, double 0.000000e+00, double %conv34)47 %cstoreval2 = fptrunc double %storeval2 to float48 store float %cstoreval2, ptr addrspace(1) %arrayidx24, align 449 ret void50}51