diff options
Diffstat (limited to 'llvm/test')
4 files changed, 196 insertions, 0 deletions
diff --git a/llvm/test/CodeGen/AArch64/aarch64-interleaved-accesses-extract-user.ll b/llvm/test/CodeGen/AArch64/aarch64-interleaved-accesses-extract-user.ll new file mode 100644 index 00000000000..8628c4288c6 --- /dev/null +++ b/llvm/test/CodeGen/AArch64/aarch64-interleaved-accesses-extract-user.ll @@ -0,0 +1,86 @@ +; RUN: opt < %s -mtriple=aarch64 -interleaved-access -S | FileCheck %s + +; CHECK-LABEL: @extract_user_basic( +; CHECK: %ldN = call { <4 x i32>, <4 x i32> } @llvm.aarch64.neon.ld2.v4i32.p0v4i32 +; CHECK: %[[R:.+]] = extractvalue { <4 x i32>, <4 x i32> } %ldN, 0 +; CHECK: extractelement <4 x i32> %[[R]], i64 1 +define void @extract_user_basic(<8 x i32>* %A, i1 %C) { +entry: + %L = load <8 x i32>, <8 x i32>* %A, align 8 + %S = shufflevector <8 x i32> %L, <8 x i32> undef, <4 x i32> <i32 0, i32 2, i32 4, i32 6> + br i1 %C, label %if.then, label %if.merge + +if.then: + %E = extractelement <8 x i32> %L, i32 2 + br label %if.merge + +if.merge: + ret void +} + +; CHECK-LABEL: @extract_user_multi( +; CHECK: %ldN = call { <4 x i32>, <4 x i32> } @llvm.aarch64.neon.ld2.v4i32.p0v4i32 +; CHECK: %[[R:.+]] = extractvalue { <4 x i32>, <4 x i32> } %ldN, 0 +; CHECK: extractelement <4 x i32> %[[R]], i64 0 +; CHECK: extractelement <4 x i32> %[[R]], i64 1 +define void @extract_user_multi(<8 x i32>* %A, i1 %C) { +entry: + %L = load <8 x i32>, <8 x i32>* %A, align 8 + %S = shufflevector <8 x i32> %L, <8 x i32> undef, <4 x i32> <i32 0, i32 2, i32 4, i32 6> + br i1 %C, label %if.then, label %if.merge + +if.then: + %E1 = extractelement <8 x i32> %L, i32 0 + br label %if.merge + +if.merge: + %E2 = extractelement <8 x i32> %L, i32 2 + ret void +} + +; CHECK-LABEL: @extract_user_multi_no_dom( +; CHECK-NOT: %ldN = call { <4 x i32>, <4 x i32> } @llvm.aarch64.neon.ld2.v4i32.p0v4i32 +define void @extract_user_multi_no_dom(<8 x i32>* %A, i1 %C) { +entry: + %L = load <8 x i32>, <8 x i32>* %A, align 8 + %E1 = extractelement <8 x i32> %L, i32 0 + br i1 %C, label %if.then, label %if.merge + +if.then: + %S = shufflevector <8 x i32> %L, <8 x i32> undef, <4 x i32> <i32 0, i32 2, i32 4, i32 6> + %E2 = extractelement <8 x i32> %L, i32 2 + br label %if.merge + +if.merge: + ret void +} + +; CHECK-LABEL: @extract_user_wrong_const_index( +; CHECK-NOT: %ldN = call { <4 x i32>, <4 x i32> } @llvm.aarch64.neon.ld2.v4i32.p0v4i32 +define void @extract_user_wrong_const_index(<8 x i32>* %A) { +entry: + %L = load <8 x i32>, <8 x i32>* %A, align 8 + %S = shufflevector <8 x i32> %L, <8 x i32> undef, <4 x i32> <i32 0, i32 2, i32 4, i32 6> + %E = extractelement <8 x i32> %L, i32 1 + ret void +} + +; CHECK-LABEL: @extract_user_undef_index( +; CHECK-NOT: %ldN = call { <4 x i32>, <4 x i32> } @llvm.aarch64.neon.ld2.v4i32.p0v4i32 +define void @extract_user_undef_index(<8 x i32>* %A) { +entry: + %L = load <8 x i32>, <8 x i32>* %A, align 8 + %S = shufflevector <8 x i32> %L, <8 x i32> undef, <4 x i32> <i32 0, i32 2, i32 4, i32 6> + %E = extractelement <8 x i32> %L, i32 undef + ret void +} + +; CHECK-LABEL: @extract_user_var_index( +; CHECK-NOT: %ldN = call { <4 x i32>, <4 x i32> } @llvm.aarch64.neon.ld2.v4i32.p0v4i32 +define void @extract_user_var_index(<8 x i32>* %A, i32 %I) { +entry: + %L = load <8 x i32>, <8 x i32>* %A, align 8 + %S = shufflevector <8 x i32> %L, <8 x i32> undef, <4 x i32> <i32 0, i32 2, i32 4, i32 6> + %E = extractelement <8 x i32> %L, i32 %I + ret void +} diff --git a/llvm/test/CodeGen/AArch64/aarch64-interleaved-accesses.ll b/llvm/test/CodeGen/AArch64/aarch64-interleaved-accesses.ll index 1bc2a3ccb1c..845050156ba 100644 --- a/llvm/test/CodeGen/AArch64/aarch64-interleaved-accesses.ll +++ b/llvm/test/CodeGen/AArch64/aarch64-interleaved-accesses.ll @@ -268,3 +268,15 @@ define void @store_illegal_factor2(<3 x float>* %p, <3 x float> %v) nounwind { store <3 x float> %tmp1, <3 x float>* %p, align 16 ret void } + +; NEON-LABEL: load_factor2_with_extract_user: +; NEON: ld2 { v0.4s, v1.4s }, [x0] +; NEON: mov w0, v0.s[1] +; NONEON-LABEL: load_factor2_with_extract_user: +; NONEON-NOT: ld2 +define i32 @load_factor2_with_extract_user(<8 x i32>* %a) { + %1 = load <8 x i32>, <8 x i32>* %a, align 8 + %2 = shufflevector <8 x i32> %1, <8 x i32> undef, <4 x i32> <i32 0, i32 2, i32 4, i32 6> + %3 = extractelement <8 x i32> %1, i32 2 + ret i32 %3 +} diff --git a/llvm/test/CodeGen/ARM/arm-interleaved-accesses-extract-user.ll b/llvm/test/CodeGen/ARM/arm-interleaved-accesses-extract-user.ll new file mode 100644 index 00000000000..620cb635641 --- /dev/null +++ b/llvm/test/CodeGen/ARM/arm-interleaved-accesses-extract-user.ll @@ -0,0 +1,86 @@ +; RUN: opt < %s -mtriple=arm-eabi -mattr=+neon -interleaved-access -S | FileCheck %s + +; CHECK-LABEL: @extract_user_basic( +; CHECK: %vldN = call { <4 x i32>, <4 x i32> } @llvm.arm.neon.vld2.v4i32.p0i8 +; CHECK: %[[R:.+]] = extractvalue { <4 x i32>, <4 x i32> } %vldN, 0 +; CHECK: extractelement <4 x i32> %[[R]], i64 1 +define void @extract_user_basic(<8 x i32>* %A, i1 %C) { +entry: + %L = load <8 x i32>, <8 x i32>* %A, align 8 + %S = shufflevector <8 x i32> %L, <8 x i32> undef, <4 x i32> <i32 0, i32 2, i32 4, i32 6> + br i1 %C, label %if.then, label %if.merge + +if.then: + %E = extractelement <8 x i32> %L, i32 2 + br label %if.merge + +if.merge: + ret void +} + +; CHECK-LABEL: @extract_user_multi( +; CHECK: %vldN = call { <4 x i32>, <4 x i32> } @llvm.arm.neon.vld2.v4i32.p0i8 +; CHECK: %[[R:.+]] = extractvalue { <4 x i32>, <4 x i32> } %vldN, 0 +; CHECK: extractelement <4 x i32> %[[R]], i64 0 +; CHECK: extractelement <4 x i32> %[[R]], i64 1 +define void @extract_user_multi(<8 x i32>* %A, i1 %C) { +entry: + %L = load <8 x i32>, <8 x i32>* %A, align 8 + %S = shufflevector <8 x i32> %L, <8 x i32> undef, <4 x i32> <i32 0, i32 2, i32 4, i32 6> + br i1 %C, label %if.then, label %if.merge + +if.then: + %E1 = extractelement <8 x i32> %L, i32 0 + br label %if.merge + +if.merge: + %E2 = extractelement <8 x i32> %L, i32 2 + ret void +} + +; CHECK-LABEL: @extract_user_multi_no_dom( +; CHECK-NOT: %vldN = call { <4 x i32>, <4 x i32> } @llvm.arm.neon.vld2.v4i32.p0i8 +define void @extract_user_multi_no_dom(<8 x i32>* %A, i1 %C) { +entry: + %L = load <8 x i32>, <8 x i32>* %A, align 8 + %E1 = extractelement <8 x i32> %L, i32 0 + br i1 %C, label %if.then, label %if.merge + +if.then: + %S = shufflevector <8 x i32> %L, <8 x i32> undef, <4 x i32> <i32 0, i32 2, i32 4, i32 6> + %E2 = extractelement <8 x i32> %L, i32 2 + br label %if.merge + +if.merge: + ret void +} + +; CHECK-LABEL: @extract_user_wrong_const_index( +; CHECK-NOT: %vldN = call { <4 x i32>, <4 x i32> } @llvm.arm.neon.vld2.v4i32.p0i8 +define void @extract_user_wrong_const_index(<8 x i32>* %A) { +entry: + %L = load <8 x i32>, <8 x i32>* %A, align 8 + %S = shufflevector <8 x i32> %L, <8 x i32> undef, <4 x i32> <i32 0, i32 2, i32 4, i32 6> + %E = extractelement <8 x i32> %L, i32 1 + ret void +} + +; CHECK-LABEL: @extract_user_undef_index( +; CHECK-NOT: %vldN = call { <4 x i32>, <4 x i32> } @llvm.arm.neon.vld2.v4i32.p0i8 +define void @extract_user_undef_index(<8 x i32>* %A) { +entry: + %L = load <8 x i32>, <8 x i32>* %A, align 8 + %S = shufflevector <8 x i32> %L, <8 x i32> undef, <4 x i32> <i32 0, i32 2, i32 4, i32 6> + %E = extractelement <8 x i32> %L, i32 undef + ret void +} + +; CHECK-LABEL: @extract_user_var_index( +; CHECK-NOT: %vldN = call { <4 x i32>, <4 x i32> } @llvm.arm.neon.vld2.v4i32.p0i8 +define void @extract_user_var_index(<8 x i32>* %A, i32 %I) { +entry: + %L = load <8 x i32>, <8 x i32>* %A, align 8 + %S = shufflevector <8 x i32> %L, <8 x i32> undef, <4 x i32> <i32 0, i32 2, i32 4, i32 6> + %E = extractelement <8 x i32> %L, i32 %I + ret void +} diff --git a/llvm/test/CodeGen/ARM/arm-interleaved-accesses.ll b/llvm/test/CodeGen/ARM/arm-interleaved-accesses.ll index 002e71f6d9b..6f3d537176c 100644 --- a/llvm/test/CodeGen/ARM/arm-interleaved-accesses.ll +++ b/llvm/test/CodeGen/ARM/arm-interleaved-accesses.ll @@ -304,3 +304,15 @@ define void @store_illegal_factor2(<3 x float>* %p, <3 x float> %v) nounwind { store <3 x float> %tmp1, <3 x float>* %p, align 16 ret void } + +; NEON-LABEL: load_factor2_with_extract_user: +; NEON: vld2.32 {d16, d17, d18, d19}, [r0:64] +; NEON: vmov.32 r0, d16[1] +; NONEON-LABEL: load_factor2_with_extract_user: +; NONEON-NOT: vld2 +define i32 @load_factor2_with_extract_user(<8 x i32>* %a) { + %1 = load <8 x i32>, <8 x i32>* %a, align 8 + %2 = shufflevector <8 x i32> %1, <8 x i32> undef, <4 x i32> <i32 0, i32 2, i32 4, i32 6> + %3 = extractelement <8 x i32> %1, i32 2 + ret i32 %3 +} |
