89 lines · plain
1; RUN: opt -mtriple=amdgcn-amd-amdhsa -passes=load-store-vectorizer,dce -S -o - %s | FileCheck %s2 3define void @base_case(i1 %cnd, ptr addrspace(1) %a, ptr addrspace(1) %b, ptr addrspace(1) %out) {4; CHECK-LABEL: @base_case5; CHECK: load <3 x i32>6entry:7 %gep1 = getelementptr inbounds i32, ptr addrspace(1) %a, i64 18 %gep2 = getelementptr inbounds i32, ptr addrspace(1) %a, i64 29 %gep4 = getelementptr inbounds i32, ptr addrspace(1) %b, i64 110 %gep5 = getelementptr inbounds i32, ptr addrspace(1) %b, i64 211 %selected = select i1 %cnd, ptr addrspace(1) %a, ptr addrspace(1) %b12 %selected14 = select i1 %cnd, ptr addrspace(1) %gep1, ptr addrspace(1) %gep413 %selected25 = select i1 %cnd, ptr addrspace(1) %gep2, ptr addrspace(1) %gep514 %val0 = load i32, ptr addrspace(1) %selected, align 415 %val1 = load i32, ptr addrspace(1) %selected14, align 416 %val2 = load i32, ptr addrspace(1) %selected25, align 417 %t0 = insertelement <3 x i32> poison, i32 %val0, i32 018 %t1 = insertelement <3 x i32> %t0, i32 %val1, i32 119 %t2 = insertelement <3 x i32> %t1, i32 %val2, i32 220 store <3 x i32> %t2, ptr addrspace(1) %out21 ret void22}23 24define void @scev_targeting_complex_case(i1 %cnd, ptr addrspace(1) %a, ptr addrspace(1) %b, i32 %base, ptr addrspace(1) %out) {25; CHECK-LABEL: @scev_targeting_complex_case26; CHECK: load <2 x i32>27entry:28 %base.x4 = shl i32 %base, 229 %base.x4.p1 = add i32 %base.x4, 130 %base.x4.p2 = add i32 %base.x4, 231 %base.x4.p3 = add i32 %base.x4, 332 %zext.x4 = zext i32 %base.x4 to i6433 %zext.x4.p1 = zext i32 %base.x4.p1 to i6434 %zext.x4.p2 = zext i32 %base.x4.p2 to i6435 %zext.x4.p3 = zext i32 %base.x4.p3 to i6436 %base.x16 = mul i64 %zext.x4, 437 %base.x16.p4 = shl i64 %zext.x4.p1, 238 %base.x16.p8 = shl i64 %zext.x4.p2, 239 %base.x16.p12 = mul i64 %zext.x4.p3, 440 %gep.a.base.x16 = getelementptr inbounds i8, ptr addrspace(1) %a, i64 %base.x1641 %gep.b.base.x16.p4 = getelementptr inbounds i8, ptr addrspace(1) %b, i64 %base.x16.p442 %gep.a.base.x16.p8 = getelementptr inbounds i8, ptr addrspace(1) %a, i64 %base.x16.p843 %gep.b.base.x16.p12 = getelementptr inbounds i8, ptr addrspace(1) %b, i64 %base.x16.p1244 %selected.base.x16.p0.or.4 = select i1 %cnd, ptr addrspace(1) %gep.a.base.x16, ptr addrspace(1) %gep.b.base.x16.p445 %gep.selected.base.x16.p8.or.12 = select i1 %cnd, ptr addrspace(1) %gep.a.base.x16.p8, ptr addrspace(1) %gep.b.base.x16.p1246 %selected.base.x16.p40.or.44 = getelementptr inbounds i32, ptr addrspace(1) %selected.base.x16.p0.or.4, i64 1047 %selected.base.x16.p44.or.48 = getelementptr inbounds i32, ptr addrspace(1) %gep.selected.base.x16.p8.or.12, i64 948 %val0 = load i32, ptr addrspace(1) %selected.base.x16.p40.or.44, align 449 %val1 = load i32, ptr addrspace(1) %selected.base.x16.p44.or.48, align 450 %t0 = insertelement <2 x i32> poison, i32 %val0, i32 051 %t1 = insertelement <2 x i32> %t0, i32 %val1, i32 152 store <2 x i32> %t1, ptr addrspace(1) %out53 ret void54}55 56define void @nested_selects(i1 %cnd0, i1 %cnd1, ptr addrspace(1) %a, ptr addrspace(1) %b, i32 %base, ptr addrspace(1) %out) {57; CHECK-LABEL: @nested_selects58; CHECK: load <2 x i32>59entry:60 %base.p1 = add nsw i32 %base, 161 %base.p2 = add i32 %base, 262 %base.p3 = add nsw i32 %base, 363 %base.x4 = mul i32 %base, 464 %base.x4.p5 = add i32 %base.x4, 565 %base.x4.p6 = add i32 %base.x4, 666 %sext = sext i32 %base to i6467 %sext.p1 = sext i32 %base.p1 to i6468 %sext.p2 = sext i32 %base.p2 to i6469 %sext.p3 = sext i32 %base.p3 to i6470 %sext.x4.p5 = sext i32 %base.x4.p5 to i6471 %sext.x4.p6 = sext i32 %base.x4.p6 to i6472 %gep.a.base = getelementptr inbounds i32, ptr addrspace(1) %a, i64 %sext73 %gep.a.base.p1 = getelementptr inbounds i32, ptr addrspace(1) %a, i64 %sext.p174 %gep.a.base.p2 = getelementptr inbounds i32, ptr addrspace(1) %a, i64 %sext.p275 %gep.a.base.p3 = getelementptr inbounds i32, ptr addrspace(1) %a, i64 %sext.p376 %gep.b.base.x4.p5 = getelementptr inbounds i32, ptr addrspace(1) %a, i64 %sext.x4.p577 %gep.b.base.x4.p6 = getelementptr inbounds i32, ptr addrspace(1) %a, i64 %sext.x4.p678 %selected.1.L = select i1 %cnd1, ptr addrspace(1) %gep.a.base.p2, ptr addrspace(1) %gep.b.base.x4.p579 %selected.1.R = select i1 %cnd1, ptr addrspace(1) %gep.a.base.p3, ptr addrspace(1) %gep.b.base.x4.p680 %selected.0.L = select i1 %cnd0, ptr addrspace(1) %gep.a.base, ptr addrspace(1) %selected.1.L81 %selected.0.R = select i1 %cnd0, ptr addrspace(1) %gep.a.base.p1, ptr addrspace(1) %selected.1.R82 %val0 = load i32, ptr addrspace(1) %selected.0.L, align 483 %val1 = load i32, ptr addrspace(1) %selected.0.R, align 484 %t0 = insertelement <2 x i32> poison, i32 %val0, i32 085 %t1 = insertelement <2 x i32> %t0, i32 %val1, i32 186 store <2 x i32> %t1, ptr addrspace(1) %out87 ret void88}89